zstandard: vendor python-zstandard 0.9.0
authorGregory Szorc <gregory.szorc@gmail.com>
Mon, 09 Apr 2018 10:13:29 -0700
changeset 37495 b1fb341d8a61
parent 37494 1ce7a55b09d1
child 37496 1765ed63db40
zstandard: vendor python-zstandard 0.9.0 This was just released. It features a number of goodies. More info at https://gregoryszorc.com/blog/2018/04/09/release-of-python-zstandard-0.9/. The clang-format ignore list was updated to reflect the new source of files. The project contains a vendored copy of zstandard 1.3.4. The old version was 1.1.3. One of the changes between those versions is that zstandard is now dual licensed BSD + GPLv2 and the patent rights grant has been removed. Good riddance. The API should be backwards compatible. So no changes in core should be needed. However, there were a number of changes in the library that we'll want to adapt to. Those will be addressed in subsequent commits. Differential Revision: https://phab.mercurial-scm.org/D3198
contrib/clang-format-ignorelist
contrib/python-zstandard/MANIFEST.in
contrib/python-zstandard/NEWS.rst
contrib/python-zstandard/README.rst
contrib/python-zstandard/c-ext/bufferutil.c
contrib/python-zstandard/c-ext/compressiondict.c
contrib/python-zstandard/c-ext/compressionparams.c
contrib/python-zstandard/c-ext/compressionreader.c
contrib/python-zstandard/c-ext/compressionwriter.c
contrib/python-zstandard/c-ext/compressobj.c
contrib/python-zstandard/c-ext/compressor.c
contrib/python-zstandard/c-ext/compressoriterator.c
contrib/python-zstandard/c-ext/constants.c
contrib/python-zstandard/c-ext/decompressionreader.c
contrib/python-zstandard/c-ext/decompressionwriter.c
contrib/python-zstandard/c-ext/decompressobj.c
contrib/python-zstandard/c-ext/decompressor.c
contrib/python-zstandard/c-ext/decompressoriterator.c
contrib/python-zstandard/c-ext/frameparams.c
contrib/python-zstandard/c-ext/python-zstandard.h
contrib/python-zstandard/make_cffi.py
contrib/python-zstandard/setup.py
contrib/python-zstandard/setup_zstd.py
contrib/python-zstandard/tests/common.py
contrib/python-zstandard/tests/test_buffer_util.py
contrib/python-zstandard/tests/test_compressor.py
contrib/python-zstandard/tests/test_compressor_fuzzing.py
contrib/python-zstandard/tests/test_data_structures.py
contrib/python-zstandard/tests/test_data_structures_fuzzing.py
contrib/python-zstandard/tests/test_decompressor.py
contrib/python-zstandard/tests/test_decompressor_fuzzing.py
contrib/python-zstandard/tests/test_estimate_sizes.py
contrib/python-zstandard/tests/test_module_attributes.py
contrib/python-zstandard/tests/test_train_dictionary.py
contrib/python-zstandard/zstandard/__init__.py
contrib/python-zstandard/zstd.c
contrib/python-zstandard/zstd/COPYING
contrib/python-zstandard/zstd/PATENTS
contrib/python-zstandard/zstd/common/bitstream.h
contrib/python-zstandard/zstd/common/compiler.h
contrib/python-zstandard/zstd/common/cpu.h
contrib/python-zstandard/zstd/common/entropy_common.c
contrib/python-zstandard/zstd/common/error_private.c
contrib/python-zstandard/zstd/common/error_private.h
contrib/python-zstandard/zstd/common/fse.h
contrib/python-zstandard/zstd/common/fse_decompress.c
contrib/python-zstandard/zstd/common/huf.h
contrib/python-zstandard/zstd/common/mem.h
contrib/python-zstandard/zstd/common/pool.c
contrib/python-zstandard/zstd/common/pool.h
contrib/python-zstandard/zstd/common/threading.c
contrib/python-zstandard/zstd/common/threading.h
contrib/python-zstandard/zstd/common/xxhash.c
contrib/python-zstandard/zstd/common/xxhash.h
contrib/python-zstandard/zstd/common/zstd_common.c
contrib/python-zstandard/zstd/common/zstd_errors.h
contrib/python-zstandard/zstd/common/zstd_internal.h
contrib/python-zstandard/zstd/compress/fse_compress.c
contrib/python-zstandard/zstd/compress/huf_compress.c
contrib/python-zstandard/zstd/compress/zstd_compress.c
contrib/python-zstandard/zstd/compress/zstd_compress_internal.h
contrib/python-zstandard/zstd/compress/zstd_double_fast.c
contrib/python-zstandard/zstd/compress/zstd_double_fast.h
contrib/python-zstandard/zstd/compress/zstd_fast.c
contrib/python-zstandard/zstd/compress/zstd_fast.h
contrib/python-zstandard/zstd/compress/zstd_lazy.c
contrib/python-zstandard/zstd/compress/zstd_lazy.h
contrib/python-zstandard/zstd/compress/zstd_ldm.c
contrib/python-zstandard/zstd/compress/zstd_ldm.h
contrib/python-zstandard/zstd/compress/zstd_opt.c
contrib/python-zstandard/zstd/compress/zstd_opt.h
contrib/python-zstandard/zstd/compress/zstdmt_compress.c
contrib/python-zstandard/zstd/compress/zstdmt_compress.h
contrib/python-zstandard/zstd/decompress/huf_decompress.c
contrib/python-zstandard/zstd/decompress/zstd_decompress.c
contrib/python-zstandard/zstd/deprecated/zbuff.h
contrib/python-zstandard/zstd/deprecated/zbuff_common.c
contrib/python-zstandard/zstd/deprecated/zbuff_compress.c
contrib/python-zstandard/zstd/deprecated/zbuff_decompress.c
contrib/python-zstandard/zstd/dictBuilder/cover.c
contrib/python-zstandard/zstd/dictBuilder/zdict.c
contrib/python-zstandard/zstd/dictBuilder/zdict.h
contrib/python-zstandard/zstd/zstd.h
contrib/python-zstandard/zstd_cffi.py
--- a/contrib/clang-format-ignorelist	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/clang-format-ignorelist	Mon Apr 09 10:13:29 2018 -0700
@@ -8,11 +8,13 @@
 contrib/python-zstandard/c-ext/bufferutil.c
 contrib/python-zstandard/c-ext/compressiondict.c
 contrib/python-zstandard/c-ext/compressionparams.c
+contrib/python-zstandard/c-ext/compressionreader.c
 contrib/python-zstandard/c-ext/compressionwriter.c
 contrib/python-zstandard/c-ext/compressobj.c
 contrib/python-zstandard/c-ext/compressor.c
 contrib/python-zstandard/c-ext/compressoriterator.c
 contrib/python-zstandard/c-ext/constants.c
+contrib/python-zstandard/c-ext/decompressionreader.c
 contrib/python-zstandard/c-ext/decompressionwriter.c
 contrib/python-zstandard/c-ext/decompressobj.c
 contrib/python-zstandard/c-ext/decompressor.c
@@ -21,11 +23,13 @@
 contrib/python-zstandard/c-ext/python-zstandard.h
 contrib/python-zstandard/zstd.c
 contrib/python-zstandard/zstd/common/bitstream.h
+contrib/python-zstandard/zstd/common/compiler.h
+contrib/python-zstandard/zstd/common/cpu.h
 contrib/python-zstandard/zstd/common/entropy_common.c
 contrib/python-zstandard/zstd/common/error_private.c
 contrib/python-zstandard/zstd/common/error_private.h
+contrib/python-zstandard/zstd/common/fse_decompress.c
 contrib/python-zstandard/zstd/common/fse.h
-contrib/python-zstandard/zstd/common/fse_decompress.c
 contrib/python-zstandard/zstd/common/huf.h
 contrib/python-zstandard/zstd/common/mem.h
 contrib/python-zstandard/zstd/common/pool.c
@@ -40,11 +44,25 @@
 contrib/python-zstandard/zstd/compress/fse_compress.c
 contrib/python-zstandard/zstd/compress/huf_compress.c
 contrib/python-zstandard/zstd/compress/zstd_compress.c
-contrib/python-zstandard/zstd/compress/zstd_opt.h
+contrib/python-zstandard/zstd/compress/zstd_compress_internal.h
+contrib/python-zstandard/zstd/compress/zstd_double_fast.c
+contrib/python-zstandard/zstd/compress/zstd_double_fast.h
+contrib/python-zstandard/zstd/compress/zstd_fast.c
+contrib/python-zstandard/zstd/compress/zstd_fast.h
+contrib/python-zstandard/zstd/compress/zstd_lazy.c
+contrib/python-zstandard/zstd/compress/zstd_lazy.h
+contrib/python-zstandard/zstd/compress/zstd_ldm.c
+contrib/python-zstandard/zstd/compress/zstd_ldm.h
 contrib/python-zstandard/zstd/compress/zstdmt_compress.c
 contrib/python-zstandard/zstd/compress/zstdmt_compress.h
+contrib/python-zstandard/zstd/compress/zstd_opt.c
+contrib/python-zstandard/zstd/compress/zstd_opt.h
 contrib/python-zstandard/zstd/decompress/huf_decompress.c
 contrib/python-zstandard/zstd/decompress/zstd_decompress.c
+contrib/python-zstandard/zstd/deprecated/zbuff_common.c
+contrib/python-zstandard/zstd/deprecated/zbuff_compress.c
+contrib/python-zstandard/zstd/deprecated/zbuff_decompress.c
+contrib/python-zstandard/zstd/deprecated/zbuff.h
 contrib/python-zstandard/zstd/dictBuilder/cover.c
 contrib/python-zstandard/zstd/dictBuilder/divsufsort.c
 contrib/python-zstandard/zstd/dictBuilder/divsufsort.h
--- a/contrib/python-zstandard/MANIFEST.in	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/MANIFEST.in	Mon Apr 09 10:13:29 2018 -0700
@@ -1,5 +1,7 @@
 graft c-ext
 graft zstd
+graft tests
 include make_cffi.py
 include setup_zstd.py
 include zstd.c
+include LICENSE
--- a/contrib/python-zstandard/NEWS.rst	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/NEWS.rst	Mon Apr 09 10:13:29 2018 -0700
@@ -1,13 +1,201 @@
+===============
 Version History
 ===============
 
+1.0.0 (not yet released)
+========================
+
+Actions Blocking Release
+------------------------
+
+* compression and decompression APIs that support ``io.rawIOBase`` interface
+  (#13).
+* Refactor module names so C and CFFI extensions live under ``zstandard``
+  package.
+* Overall API design review.
+* Use Python allocator where possible.
+* Figure out what to do about experimental APIs not implemented by CFFI.
+* APIs for auto adjusting compression parameters based on input size. e.g.
+  clamping the window log so it isn't too large for input.
+* Consider allowing compressor and decompressor instances to be thread safe,
+  support concurrent operations. Or track when an operation is in progress and
+  refuse to let concurrent operations use the same instance.
+* Support for magic-less frames for all decompression operations (``decompress()``
+  doesn't work due to sniffing the content size and the lack of a ZSTD API to
+  sniff magic-less frames - this should be fixed in 1.3.5.).
+* Audit for complete flushing when ending compression streams.
+* Deprecate legacy APIs.
+* Audit for ability to control read/write sizes on all APIs.
+* Detect memory leaks via bench.py.
+* Remove low-level compression parameters from ``ZstdCompressor.__init__`` and
+  require use of ``CompressionParameters``.
+* Expose ``ZSTD_getFrameProgression()`` from more compressor types.
+
+Other Actions Not Blocking Release
+---------------------------------------
+
+* Support for block compression APIs.
+* API for ensuring max memory ceiling isn't exceeded.
+* Move off nose for testing.
+
+0.9.0 (released 2018-04-08)
+===========================
+
+Backwards Compatibility Notes
+-----------------------------
+
+* CFFI 1.11 or newer is now required (previous requirement was 1.8).
+* The primary module is now ``zstandard``. Please change imports of ``zstd``
+  and ``zstd_cffi`` to ``import zstandard``. See the README for more. Support
+  for importing the old names will be dropped in the next release.
+* ``ZstdCompressor.read_from()`` and ``ZstdDecompressor.read_from()`` have
+  been renamed to ``read_to_iter()``. ``read_from()`` is aliased to the new
+  name and will be deleted in a future release.
+* Support for Python 2.6 has been removed.
+* Support for Python 3.3 has been removed.
+* The ``selectivity`` argument to ``train_dictionary()`` has been removed, as
+  the feature disappeared from zstd 1.3.
+* Support for legacy dictionaries has been removed. Cover dictionaries are now
+  the default. ``train_cover_dictionary()`` has effectively been renamed to
+  ``train_dictionary()``.
+* The ``allow_empty`` argument from ``ZstdCompressor.compress()`` has been
+  deleted and the method now allows empty inputs to be compressed by default.
+* ``estimate_compression_context_size()`` has been removed. Use
+  ``CompressionParameters.estimated_compression_context_size()`` instead.
+* ``get_compression_parameters()`` has been removed. Use
+  ``CompressionParameters.from_level()`` instead.
+* The arguments to ``CompressionParameters.__init__()`` have changed. If you
+  were using positional arguments before, the positions now map to different
+  arguments. It is recommended to use keyword arguments to construct
+  ``CompressionParameters`` instances.
+* ``TARGETLENGTH_MAX`` constant has been removed (it disappeared from zstandard
+  1.3.4).
+* ``ZstdCompressor.write_to()`` and ``ZstdDecompressor.write_to()`` have been
+  renamed to ``ZstdCompressor.stream_writer()`` and
+  ``ZstdDecompressor.stream_writer()``, respectively. The old names are still
+  aliased, but will be removed in the next major release.
+* Content sizes are written into frame headers by default
+  (``ZstdCompressor(write_content_size=True)`` is now the default).
+* ``CompressionParameters`` has been renamed to ``ZstdCompressionParameters``
+  for consistency with other types. The old name is an alias and will be removed
+  in the next major release.
+
+Bug Fixes
+---------
+
+* Fixed memory leak in ``ZstdCompressor.copy_stream()`` (#40) (from 0.8.2).
+* Fixed memory leak in ``ZstdDecompressor.copy_stream()`` (#35) (from 0.8.2).
+* Fixed memory leak of ``ZSTD_DDict`` instances in CFFI's ``ZstdDecompressor``.
+
+New Features
+------------
+
+* Bundlded zstandard library upgraded from 1.1.3 to 1.3.4. This delivers various
+  bug fixes and performance improvements. It also gives us access to newer
+  features.
+* Support for negative compression levels.
+* Support for *long distance matching* (facilitates compression ratios that approach
+  LZMA).
+* Supporting for reading empty zstandard frames (with an embedded content size
+  of 0).
+* Support for writing and partial support for reading zstandard frames without a
+  magic header.
+* New ``stream_reader()`` API that exposes the ``io.RawIOBase`` interface (allows
+  you to ``.read()`` from a file-like object).
+* Several minor features, bug fixes, and performance enhancements.
+* Wheels for Linux and macOS are now provided with releases.
+
+Changes
+-------
+
+* Functions accepting bytes data now use the buffer protocol and can accept
+  more types (like ``memoryview`` and ``bytearray``) (#26).
+* Add #includes so compilation on OS X and BSDs works (#20).
+* New ``ZstdDecompressor.stream_reader()`` API to obtain a read-only i/o stream
+  of decompressed data for a source.
+* New ``ZstdCompressor.stream_reader()`` API to obtain a read-only i/o stream of
+  compressed data for a source.
+* Renamed ``ZstdDecompressor.read_from()`` to ``ZstdDecompressor.read_to_iter()``.
+  The old name is still available.
+* Renamed ``ZstdCompressor.read_from()`` to ``ZstdCompressor.read_to_iter()``.
+  ``read_from()`` is still available at its old location.
+* Introduce the ``zstandard`` module to import and re-export the C or CFFI
+  *backend* as appropriate. Behavior can be controlled via the
+  ``PYTHON_ZSTANDARD_IMPORT_POLICY`` environment variable. See README for
+  usage info.
+* Vendored version of zstd upgraded to 1.3.4.
+* Added module constants ``CONTENTSIZE_UNKNOWN`` and ``CONTENTSIZE_ERROR``.
+* Add ``STRATEGY_BTULTRA`` compression strategy constant.
+* Switch from deprecated ``ZSTD_getDecompressedSize()`` to
+  ``ZSTD_getFrameContentSize()`` replacement.
+* ``ZstdCompressor.compress()`` can now compress empty inputs without requiring
+  special handling.
+* ``ZstdCompressor`` and ``ZstdDecompressor`` now have a ``memory_size()``
+  method for determining the current memory utilization of the underlying zstd
+  primitive.
+* ``train_dictionary()`` has new arguments and functionality for trying multiple
+  variations of COVER parameters and selecting the best one.
+* Added module constants ``LDM_MINMATCH_MIN``, ``LDM_MINMATCH_MAX``, and
+  ``LDM_BUCKETSIZELOG_MAX``.
+* Converted all consumers to the zstandard *new advanced API*, which uses
+  ``ZSTD_compress_generic()``
+* ``CompressionParameters.__init__`` now accepts several more arguments,
+  including support for *long distance matching*.
+* ``ZstdCompressionDict.__init__`` now accepts a ``dict_type`` argument that
+  controls how the dictionary should be interpreted. This can be used to
+  force the use of *content-only* dictionaries or to require the presence
+  of the dictionary magic header.
+* ``ZstdCompressionDict.precompute_compress()`` can be used to precompute the
+  compression dictionary so it can efficiently be used with multiple
+  ``ZstdCompressor`` instances.
+* Digested dictionaries are now stored in ``ZstdCompressionDict`` instances,
+  created automatically on first use, and automatically reused by all
+  ``ZstdDecompressor`` instances bound to that dictionary.
+* All meaningful functions now accept keyword arguments.
+* ``ZstdDecompressor.decompressobj()`` now accepts a ``write_size`` argument
+  to control how much work to perform on every decompressor invocation.
+* ``ZstdCompressor.write_to()`` now exposes a ``tell()``, which exposes the
+  total number of bytes written so far.
+* ``ZstdDecompressor.stream_reader()`` now supports ``seek()`` when moving
+  forward in the stream.
+* Removed ``TARGETLENGTH_MAX`` constant.
+* Added ``frame_header_size(data)`` function.
+* Added ``frame_content_size(data)`` function.
+* Consumers of ``ZSTD_decompress*`` have been switched to the new *advanced
+  decompression* API.
+* ``ZstdCompressor`` and ``ZstdCompressionParams`` can now be constructed with
+  negative compression levels.
+* ``ZstdDecompressor`` now accepts a ``max_window_size`` argument to limit the
+  amount of memory required for decompression operations.
+* ``FORMAT_ZSTD1`` and ``FORMAT_ZSTD1_MAGICLESS`` constants to be used with
+  the ``format`` compression parameter to control whether the frame magic
+  header is written.
+* ``ZstdDecompressor`` now accepts a ``format`` argument to control the
+  expected frame format.
+* ``ZstdCompressor`` now has a ``frame_progression()`` method to return
+  information about the current compression operation.
+* Error messages in CFFI no longer have ``b''`` literals.
+* Compiler warnings and underlying overflow issues on 32-bit platforms have been
+  fixed.
+* Builds in CI now build with compiler warnings as errors. This should hopefully
+  fix new compiler warnings from being introduced.
+* Make ``ZstdCompressor(write_content_size=True)`` and
+  ``CompressionParameters(write_content_size=True)`` the default.
+* ``CompressionParameters`` has been renamed to ``ZstdCompressionParameters``.
+
+0.8.2 (released 2018-02-22)
+---------------------------
+
+* Fixed memory leak in ``ZstdCompressor.copy_stream()`` (#40).
+* Fixed memory leak in ``ZstdDecompressor.copy_stream()`` (#35).
+
 0.8.1 (released 2017-04-08)
 ---------------------------
 
 * Add #includes so compilation on OS X and BSDs works (#20).
 
 0.8.0 (released 2017-03-08)
----------------------------
+===========================
 
 * CompressionParameters now has a estimated_compression_context_size() method.
   zstd.estimate_compression_context_size() is now deprecated and slated for
@@ -35,7 +223,7 @@
   DictParameters instance to control dictionary generation.
 
 0.7.0 (released 2017-02-07)
----------------------------
+===========================
 
 * Added zstd.get_frame_parameters() to obtain info about a zstd frame.
 * Added ZstdDecompressor.decompress_content_dict_chain() for efficient
@@ -62,7 +250,7 @@
 * DictParameters instances now expose their values as attributes.
 
 0.6.0 (released 2017-01-14)
----------------------------
+===========================
 
 * Support for legacy zstd protocols (build time opt in feature).
 * Automation improvements to test against Python 3.6, latest versions
@@ -79,17 +267,17 @@
 * Disallow compress(b'') when writing content sizes by default (issue #11).
 
 0.5.2 (released 2016-11-12)
----------------------------
+===========================
 
 * more packaging fixes for source distribution
 
 0.5.1 (released 2016-11-12)
----------------------------
+===========================
 
 * setup_zstd.py is included in the source distribution
 
 0.5.0 (released 2016-11-10)
----------------------------
+===========================
 
 * Vendored version of zstd updated to 1.1.1.
 * Continuous integration for Python 3.6 and 3.7
@@ -114,8 +302,8 @@
 * The monolithic ``zstd.c`` file has been split into a header file defining
   types and separate ``.c`` source files for the implementation.
 
-History of the Project
-======================
+Older History
+=============
 
 2016-08-31 - Zstandard 1.0.0 is released and Gregory starts hacking on a
 Python extension for use by the Mercurial project. A very hacky prototype
--- a/contrib/python-zstandard/README.rst	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/README.rst	Mon Apr 09 10:13:29 2018 -0700
@@ -11,69 +11,18 @@
 performance. This means exposing most of the features and flexibility
 of the C API while not sacrificing usability or safety that Python provides.
 
-The canonical home for this project is
+The canonical home for this project lives in a Mercurial repository run by
+the author. For convenience, that repository is frequently synchronized to
 https://github.com/indygreg/python-zstandard.
 
 |  |ci-status| |win-ci-status|
 
-State of Project
-================
-
-The project is officially in beta state. The author is reasonably satisfied
-that functionality works as advertised. **There will be some backwards
-incompatible changes before 1.0, probably in the 0.9 release.** This may
-involve renaming the main module from *zstd* to *zstandard* and renaming
-various types and methods. Pin the package version to prevent unwanted
-breakage when this change occurs!
-
-This project is vendored and distributed with Mercurial 4.1, where it is
-used in a production capacity.
-
-There is continuous integration for Python versions 2.6, 2.7, and 3.3+
-on Linux x86_x64 and Windows x86 and x86_64. The author is reasonably
-confident the extension is stable and works as advertised on these
-platforms.
-
-The CFFI bindings are mostly feature complete. Where a feature is implemented
-in CFFI, unit tests run against both C extension and CFFI implementation to
-ensure behavior parity.
-
-Expected Changes
-----------------
-
-The author is reasonably confident in the current state of what's
-implemented on the ``ZstdCompressor`` and ``ZstdDecompressor`` types.
-Those APIs likely won't change significantly. Some low-level behavior
-(such as naming and types expected by arguments) may change.
-
-There will likely be arguments added to control the input and output
-buffer sizes (currently, certain operations read and write in chunk
-sizes using zstd's preferred defaults).
-
-There should be an API that accepts an object that conforms to the buffer
-interface and returns an iterator over compressed or decompressed output.
-
-There should be an API that exposes an ``io.RawIOBase`` interface to
-compressor and decompressor streams, like how ``gzip.GzipFile`` from
-the standard library works (issue 13).
-
-The author is on the fence as to whether to support the extremely
-low level compression and decompression APIs. It could be useful to
-support compression without the framing headers. But the author doesn't
-believe it a high priority at this time.
-
-There will likely be a refactoring of the module names. Currently,
-``zstd`` is a C extension and ``zstd_cffi`` is the CFFI interface.
-This means that all code for the C extension must be implemented in
-C. ``zstd`` may be converted to a Python module so code can be reused
-between CFFI and C and so not all code in the C extension has to be C.
-
 Requirements
 ============
 
-This extension is designed to run with Python 2.6, 2.7, 3.3, 3.4, 3.5, and
-3.6 on common platforms (Linux, Windows, and OS X). Only x86_64 is
-currently well-tested as an architecture.
+This extension is designed to run with Python 2.7, 3.4, 3.5, and 3.6
+on common platforms (Linux, Windows, and OS X). x86 and x86_64 are well-tested
+on Windows. Only x86_64 is well-tested on Linux and macOS.
 
 Installing
 ==========
@@ -96,114 +45,82 @@
 Performance
 ===========
 
-Very crude and non-scientific benchmarking (most benchmarks fall in this
-category because proper benchmarking is hard) show that the Python bindings
-perform within 10% of the native C implementation.
-
-The following table compares the performance of compressing and decompressing
-a 1.1 GB tar file comprised of the files in a Firefox source checkout. Values
-obtained with the ``zstd`` program are on the left. The remaining columns detail
-performance of various compression APIs in the Python bindings.
+zstandard is a highly tunable compression algorithm. In its default settings
+(compression level 3), it will be faster at compression and decompression and
+will have better compression ratios than zlib on most data sets. When tuned
+for speed, it approaches lz4's speed and ratios. When tuned for compression
+ratio, it approaches lzma ratios and compression speed, but decompression
+speed is much faster. See the official zstandard documentation for more.
 
-+-------+-----------------+-----------------+-----------------+---------------+
-| Level | Native          | Simple          | Stream In       | Stream Out    |
-|       | Comp / Decomp   | Comp / Decomp   | Comp / Decomp   | Comp          |
-+=======+=================+=================+=================+===============+
-|   1   | 490 / 1338 MB/s | 458 / 1266 MB/s | 407 / 1156 MB/s |  405 MB/s     |
-+-------+-----------------+-----------------+-----------------+---------------+
-|   2   | 412 / 1288 MB/s | 381 / 1203 MB/s | 345 / 1128 MB/s |  349 MB/s     |
-+-------+-----------------+-----------------+-----------------+---------------+
-|   3   | 342 / 1312 MB/s | 319 / 1182 MB/s | 285 / 1165 MB/s |  287 MB/s     |
-+-------+-----------------+-----------------+-----------------+---------------+
-|  11   |  64 / 1506 MB/s |  66 / 1436 MB/s |  56 / 1342 MB/s |   57 MB/s     |
-+-------+-----------------+-----------------+-----------------+---------------+
-
-Again, these are very unscientific. But it shows that Python is capable of
-compressing at several hundred MB/s and decompressing at over 1 GB/s.
-
-Comparison to Other Python Bindings
-===================================
-
-https://pypi.python.org/pypi/zstd is an alternate Python binding to
-Zstandard. At the time this was written, the latest release of that
-package (1.1.2) only exposed the simple APIs for compression and decompression.
-This package exposes much more of the zstd API, including streaming and
-dictionary compression. This package also has CFFI support.
-
-Bundling of Zstandard Source Code
-=================================
-
-The source repository for this project contains a vendored copy of the
-Zstandard source code. This is done for a few reasons.
+zstandard and this library support multi-threaded compression. There is a
+mechanism to compress large inputs using multiple threads.
 
-First, Zstandard is relatively new and not yet widely available as a system
-package. Providing a copy of the source code enables the Python C extension
-to be compiled without requiring the user to obtain the Zstandard source code
-separately.
-
-Second, Zstandard has both a stable *public* API and an *experimental* API.
-The *experimental* API is actually quite useful (contains functionality for
-training dictionaries for example), so it is something we wish to expose to
-Python. However, the *experimental* API is only available via static linking.
-Furthermore, the *experimental* API can change at any time. So, control over
-the exact version of the Zstandard library linked against is important to
-ensure known behavior.
-
-Instructions for Building and Testing
-=====================================
-
-Once you have the source code, the extension can be built via setup.py::
-
-   $ python setup.py build_ext
-
-We recommend testing with ``nose``::
-
-   $ nosetests
+The performance of this library is usually very similar to what the zstandard
+C API can deliver. Overhead in this library is due to general Python overhead
+and can't easily be avoided by *any* zstandard Python binding. This library
+exposes multiple APIs for performing compression and decompression so callers
+can pick an API suitable for their need. Contrast with the compression
+modules in Python's standard library (like ``zlib``), which only offer limited
+mechanisms for performing operations. The API flexibility means consumers can
+choose to use APIs that facilitate zero copying or minimize Python object
+creation and garbage collection overhead.
 
-A Tox configuration is present to test against multiple Python versions::
-
-   $ tox
-
-Tests use the ``hypothesis`` Python package to perform fuzzing. If you
-don't have it, those tests won't run. Since the fuzzing tests take longer
-to execute than normal tests, you'll need to opt in to running them by
-setting the ``ZSTD_SLOW_TESTS`` environment variable. This is set
-automatically when using ``tox``.
-
-The ``cffi`` Python package needs to be installed in order to build the CFFI
-bindings. If it isn't present, the CFFI bindings won't be built.
-
-To create a virtualenv with all development dependencies, do something
-like the following::
-
-  # Python 2
-  $ virtualenv venv
-
-  # Python 3
-  $ python3 -m venv venv
-
-  $ source venv/bin/activate
-  $ pip install cffi hypothesis nose tox
+This library is capable of single-threaded throughputs well over 1 GB/s. For
+exact numbers, measure yourself. The source code repository has a ``bench.py``
+script that can be used to measure things.
 
 API
 ===
 
-The compiled C extension provides a ``zstd`` Python module. The CFFI
-bindings provide a ``zstd_cffi`` module. Both provide an identical API
-interface. The types, functions, and attributes exposed by these modules
+To interface with Zstandard, simply import the ``zstandard`` module::
+
+   import zstandard
+
+It is a popular convention to alias the module as a different name for
+brevity::
+
+   import zstandard as zstd
+
+This module attempts to import and use either the C extension or CFFI
+implementation. On Python platforms known to support C extensions (like
+CPython), it raises an ImportError if the C extension cannot be imported.
+On Python platforms known to not support C extensions (like PyPy), it only
+attempts to import the CFFI implementation and raises ImportError if that
+can't be done. On other platforms, it first tries to import the C extension
+then falls back to CFFI if that fails and raises ImportError if CFFI fails.
+
+To change the module import behavior, a ``PYTHON_ZSTANDARD_IMPORT_POLICY``
+environment variable can be set. The following values are accepted:
+
+default
+   The behavior described above.
+cffi_fallback
+   Always try to import the C extension then fall back to CFFI if that
+   fails.
+cext
+   Only attempt to import the C extension.
+cffi
+   Only attempt to import the CFFI implementation.
+
+In addition, the ``zstandard`` module exports a ``backend`` attribute
+containing the string name of the backend being used. It will be one
+of ``cext`` or ``cffi`` (for *C extension* and *cffi*, respectively).
+
+The types, functions, and attributes exposed by the ``zstandard`` module
 are documented in the sections below.
 
 .. note::
 
    The documentation in this section makes references to various zstd
-   concepts and functionality. The ``Concepts`` section below explains
-   these concepts in more detail.
+   concepts and functionality. The source repository contains a
+   ``docs/concepts.rst`` file explaining these in more detail.
 
 ZstdCompressor
 --------------
 
 The ``ZstdCompressor`` class provides an interface for performing
-compression operations.
+compression operations. Each instance is essentially a wrapper around a
+``ZSTD_CCtx`` from the C API.
 
 Each instance is associated with parameters that control compression
 behavior. These come from the following named arguments (all optional):
@@ -214,21 +131,21 @@
    Compression dictionary to use.
 
    Note: When using dictionary data and ``compress()`` is called multiple
-   times, the ``CompressionParameters`` derived from an integer compression
-   ``level`` and the first compressed data's size will be reused for all
-   subsequent operations. This may not be desirable if source data size
-   varies significantly.
+   times, the ``ZstdCompressionParameters`` derived from an integer
+   compression ``level`` and the first compressed data's size will be reused
+   for all subsequent operations. This may not be desirable if source data
+   size varies significantly.
 compression_params
-   A ``CompressionParameters`` instance (overrides the ``level`` value).
+   A ``ZstdCompressionParameters`` instance defining compression settings.
 write_checksum
    Whether a 4 byte checksum should be written with the compressed data.
    Defaults to False. If True, the decompressor can verify that decompressed
    data matches the original input data.
 write_content_size
    Whether the size of the uncompressed data will be written into the
-   header of compressed data. Defaults to False. The data will only be
+   header of compressed data. Defaults to True. The data will only be
    written if the compressor knows the size of the input data. This is
-   likely not true for streaming compression.
+   often not true for streaming compression.
 write_dict_id
    Whether to write the dictionary ID into the compressed data.
    Defaults to True. The dictionary ID is only written if a dictionary
@@ -242,10 +159,25 @@
    data. APIs that spawn multiple threads for working on multiple pieces of
    data have their own ``threads`` argument.
 
+``compression_params`` is mutually exclusive with ``level``, ``write_checksum``,
+``write_content_size``, ``write_dict_id``, and ``threads``.
+
 Unless specified otherwise, assume that no two methods of ``ZstdCompressor``
 instances can be called from multiple Python threads simultaneously. In other
 words, assume instances are not thread safe unless stated otherwise.
 
+Utility Methods
+^^^^^^^^^^^^^^^
+
+``frame_progression()`` returns a 3-tuple containing the number of bytes
+ingested, consumed, and produced by the current compression operation.
+
+``memory_size()`` obtains the memory utilization of the underlying zstd
+compression context, in bytes.::
+
+    cctx = zstd.ZstdCompressor()
+    memory = cctx.memory_size()
+
 Simple API
 ^^^^^^^^^^
 
@@ -256,40 +188,75 @@
 
 The ``data`` argument can be any object that implements the *buffer protocol*.
 
-Unless ``compression_params`` or ``dict_data`` are passed to the
-``ZstdCompressor``, each invocation of ``compress()`` will calculate the
-optimal compression parameters for the configured compression ``level`` and
-input data size (some parameters are fine-tuned for small input sizes).
+Stream Reader API
+^^^^^^^^^^^^^^^^^
+
+``stream_reader(source)`` can be used to obtain an object conforming to the
+``io.RawIOBase`` interface for reading compressed output as a stream::
+
+   with open(path, 'rb') as fh:
+       cctx = zstd.ZstdCompressor()
+       with cctx.stream_reader(fh) as reader:
+           while True:
+               chunk = reader.read(16384)
+               if not chunk:
+                   break
+
+               # Do something with compressed chunk.
+
+The stream can only be read within a context manager. When the context
+manager exits, the stream is closed and the underlying resource is
+released and future operations against the compression stream stream will fail.
+
+The ``source`` argument to ``stream_reader()`` can be any object with a
+``read(size)`` method or any object implementing the *buffer protocol*.
 
-If a compression dictionary is being used, the compression parameters
-determined from the first input's size will be reused for subsequent
-operations.
+``stream_reader()`` accepts a ``size`` argument specifying how large the input
+stream is. This is used to adjust compression parameters so they are
+tailored to the source size.::
+
+   with open(path, 'rb') as fh:
+       cctx = zstd.ZstdCompressor()
+       with cctx.stream_reader(fh, size=os.stat(path).st_size) as reader:
+           ...
+
+If the ``source`` is a stream, you can specify how large ``read()`` requests
+to that stream should be via the ``read_size`` argument. It defaults to
+``zstandard.COMPRESSION_RECOMMENDED_INPUT_SIZE``.::
 
-There is currently a deficiency in zstd's C APIs that makes it difficult
-to round trip empty inputs when ``write_content_size=True``. Attempting
-this will raise a ``ValueError`` unless ``allow_empty=True`` is passed
-to ``compress()``.
+   with open(path, 'rb') as fh:
+       cctx = zstd.ZstdCompressor()
+       # Will perform fh.read(8192) when obtaining data to feed into the
+       # compressor.
+       with cctx.stream_reader(fh, read_size=8192) as reader:
+           ...
+
+The stream returned by ``stream_reader()`` is neither writable nor seekable
+(even if the underlying source is seekable). ``readline()`` and
+``readlines()`` are not implemented because they don't make sense for
+compressed data. ``tell()`` returns the number of compressed bytes
+emitted so far.
 
 Streaming Input API
 ^^^^^^^^^^^^^^^^^^^
 
-``write_to(fh)`` (which behaves as a context manager) allows you to *stream*
+``stream_writer(fh)`` (which behaves as a context manager) allows you to *stream*
 data into a compressor.::
 
    cctx = zstd.ZstdCompressor(level=10)
-   with cctx.write_to(fh) as compressor:
+   with cctx.stream_writer(fh) as compressor:
        compressor.write(b'chunk 0')
        compressor.write(b'chunk 1')
        ...
 
-The argument to ``write_to()`` must have a ``write(data)`` method. As
+The argument to ``stream_writer()`` must have a ``write(data)`` method. As
 compressed data is available, ``write()`` will be called with the compressed
 data as its argument. Many common Python types implement ``write()``, including
 open file handles and ``io.BytesIO``.
 
-``write_to()`` returns an object representing a streaming compressor instance.
-It **must** be used as a context manager. That object's ``write(data)`` method
-is used to feed data into the compressor.
+``stream_writer()`` returns an object representing a streaming compressor
+instance. It **must** be used as a context manager. That object's
+``write(data)`` method is used to feed data into the compressor.
 
 A ``flush()`` method can be called to evict whatever data remains within the
 compressor's internal state into the output object. This may result in 0 or
@@ -303,7 +270,7 @@
 you can declare it before compression begins::
 
    cctx = zstd.ZstdCompressor()
-   with cctx.write_to(fh, size=data_len) as compressor:
+   with cctx.stream_writer(fh, size=data_len) as compressor:
        compressor.write(chunk0)
        compressor.write(chunk1)
        ...
@@ -315,29 +282,35 @@
 The size of chunks being ``write()`` to the destination can be specified::
 
     cctx = zstd.ZstdCompressor()
-    with cctx.write_to(fh, write_size=32768) as compressor:
+    with cctx.stream_writer(fh, write_size=32768) as compressor:
         ...
 
 To see how much memory is being used by the streaming compressor::
 
     cctx = zstd.ZstdCompressor()
-    with cctx.write_to(fh) as compressor:
+    with cctx.stream_writer(fh) as compressor:
         ...
         byte_size = compressor.memory_size()
 
+Thte total number of bytes written so far are exposed via ``tell()``::
+
+    cctx = zstd.ZstdCompressor()
+    with cctx.stream_writer(fh) as compressor:
+        ...
+        total_written = compressor.tell()
+
 Streaming Output API
 ^^^^^^^^^^^^^^^^^^^^
 
-``read_from(reader)`` provides a mechanism to stream data out of a compressor
-as an iterator of data chunks.::
+``read_to_iter(reader)`` provides a mechanism to stream data out of a
+compressor as an iterator of data chunks.::
 
    cctx = zstd.ZstdCompressor()
-   for chunk in cctx.read_from(fh):
+   for chunk in cctx.read_to_iter(fh):
         # Do something with emitted data.
 
-``read_from()`` accepts an object that has a ``read(size)`` method or conforms
-to the buffer protocol. (``bytes`` and ``memoryview`` are 2 common types that
-provide the buffer protocol.)
+``read_to_iter()`` accepts an object that has a ``read(size)`` method or
+conforms to the buffer protocol.
 
 Uncompressed data is fetched from the source either by calling ``read(size)``
 or by fetching a slice of data from the object directly (in the case where
@@ -348,23 +321,24 @@
 it raises or returns an empty bytes (``b''``). It is perfectly valid for
 the source to deliver fewer bytes than were what requested by ``read(size)``.
 
-Like ``write_to()``, ``read_from()`` also accepts a ``size`` argument
+Like ``stream_writer()``, ``read_to_iter()`` also accepts a ``size`` argument
 declaring the size of the input stream::
 
     cctx = zstd.ZstdCompressor()
-    for chunk in cctx.read_from(fh, size=some_int):
+    for chunk in cctx.read_to_iter(fh, size=some_int):
         pass
 
 You can also control the size that data is ``read()`` from the source and
 the ideal size of output chunks::
 
     cctx = zstd.ZstdCompressor()
-    for chunk in cctx.read_from(fh, read_size=16384, write_size=8192):
+    for chunk in cctx.read_to_iter(fh, read_size=16384, write_size=8192):
         pass
 
-Unlike ``write_to()``, ``read_from()`` does not give direct control over the
-sizes of chunks fed into the compressor. Instead, chunk sizes will be whatever
-the object being read from delivers. These will often be of a uniform size.
+Unlike ``stream_writer()``, ``read_to_iter()`` does not give direct control
+over the sizes of chunks fed into the compressor. Instead, chunk sizes will
+be whatever the object being read from delivers. These will often be of a
+uniform size.
 
 Stream Copying API
 ^^^^^^^^^^^^^^^^^^
@@ -404,7 +378,7 @@
 ``flush()`` methods. Each returns compressed data or an empty bytes.
 
 The purpose of ``compressobj()`` is to provide an API-compatible interface
-with ``zlib.compressobj`` and ``bz2.BZ2Compressor``. This allows callers to
+with ``zlib.compressobj``, ``bz2.BZ2Compressor``, etc. This allows callers to
 swap in different compressor objects while using the same API.
 
 ``flush()`` accepts an optional argument indicating how to end the stream.
@@ -485,13 +459,23 @@
 ----------------
 
 The ``ZstdDecompressor`` class provides an interface for performing
-decompression.
+decompression. It is effectively a wrapper around the ``ZSTD_DCtx`` type from
+the C API.
 
 Each instance is associated with parameters that control decompression. These
 come from the following named arguments (all optional):
 
 dict_data
    Compression dictionary to use.
+max_window_size
+   Sets an uppet limit on the window size for decompression operations in
+   kibibytes. This setting can be used to prevent large memory allocations
+   for inputs using large compression windows.
+format
+   Set the format of data for the decoder. By default, this is
+   ``zstd.FORMAT_ZSTD1``. It can be set to ``zstd.FORMAT_ZSTD1_MAGICLESS`` to
+   allow decoding frames without the 4 byte magic header. Not all decompression
+   APIs support this mode.
 
 The interface of this class is very similar to ``ZstdCompressor`` (by design).
 
@@ -499,6 +483,15 @@
 instances can be called from multiple Python threads simultaneously. In other
 words, assume instances are not thread safe unless stated otherwise.
 
+Utility Methods
+^^^^^^^^^^^^^^^
+
+``memory_size()`` obtains the size of the underlying zstd decompression context,
+in bytes.::
+
+    dctx = zstd.ZstdDecompressor()
+    size = dctx.memory_size()
+
 Simple API
 ^^^^^^^^^^
 
@@ -509,9 +502,10 @@
     decompressed = dctx.decompress(data)
 
 By default, ``decompress(data)`` will only work on data written with the content
-size encoded in its header. This can be achieved by creating a
-``ZstdCompressor`` with ``write_content_size=True``. If compressed data without
-an embedded content size is seen, ``zstd.ZstdError`` will be raised.
+size encoded in its header (this is the default behavior of
+``ZstdCompressor().compress()`` but may not be true for streaming compression). If
+compressed data without an embedded content size is seen, ``zstd.ZstdError`` will
+be raised.
 
 If the compressed data doesn't have its content size embedded within it,
 decompression can be attempted by specifying the ``max_output_size``
@@ -534,17 +528,67 @@
 result in a lot of work for the memory allocator and may result in
 ``MemoryError`` being raised if the allocation fails.
 
-If the exact size of decompressed data is unknown, it is **strongly**
-recommended to use a streaming API.
+.. important::
+
+   If the exact size of decompressed data is unknown (not passed in explicitly
+   and not stored in the zstandard frame), for performance reasons it is
+   encouraged to use a streaming API.
+
+Stream Reader API
+^^^^^^^^^^^^^^^^^
+
+``stream_reader(source)`` can be used to obtain an object conforming to the
+``io.RawIOBase`` interface for reading decompressed output as a stream::
+
+   with open(path, 'rb') as fh:
+       dctx = zstd.ZstdDecompressor()
+       with dctx.stream_reader(fh) as reader:
+           while True:
+               chunk = reader.read(16384)
+               if not chunk:
+                   break
+
+               # Do something with decompressed chunk.
+
+The stream can only be read within a context manager. When the context
+manager exits, the stream is closed and the underlying resource is
+released and future operations against the stream will fail.
+
+The ``source`` argument to ``stream_reader()`` can be any object with a
+``read(size)`` method or any object implementing the *buffer protocol*.
+
+If the ``source`` is a stream, you can specify how large ``read()`` requests
+to that stream should be via the ``read_size`` argument. It defaults to
+``zstandard.DECOMPRESSION_RECOMMENDED_INPUT_SIZE``.::
+
+   with open(path, 'rb') as fh:
+       dctx = zstd.ZstdDecompressor()
+       # Will perform fh.read(8192) when obtaining data for the decompressor.
+       with dctx.stream_reader(fh, read_size=8192) as reader:
+           ...
+
+The stream returned by ``stream_reader()`` is not writable.
+
+The stream returned by ``stream_reader()`` is *partially* seekable.
+Absolute and relative positions (``SEEK_SET`` and ``SEEK_CUR``) forward
+of the current position are allowed. Offsets behind the current read
+position and offsets relative to the end of stream are not allowed and
+will raise ``ValueError`` if attempted.
+
+``tell()`` returns the number of decompressed bytes read so far.
+
+Not all I/O methods are implemented. Notably missing is support for
+``readline()``, ``readlines()``, and linewise iteration support. Support for
+these is planned for a future release.
 
 Streaming Input API
 ^^^^^^^^^^^^^^^^^^^
 
-``write_to(fh)`` can be used to incrementally send compressed data to a
+``stream_writer(fh)`` can be used to incrementally send compressed data to a
 decompressor.::
 
     dctx = zstd.ZstdDecompressor()
-    with dctx.write_to(fh) as decompressor:
+    with dctx.stream_writer(fh) as decompressor:
         decompressor.write(compressed_data)
 
 This behaves similarly to ``zstd.ZstdCompressor``: compressed data is written to
@@ -558,54 +602,56 @@
 The size of chunks being ``write()`` to the destination can be specified::
 
     dctx = zstd.ZstdDecompressor()
-    with dctx.write_to(fh, write_size=16384) as decompressor:
+    with dctx.stream_writer(fh, write_size=16384) as decompressor:
         pass
 
 You can see how much memory is being used by the decompressor::
 
     dctx = zstd.ZstdDecompressor()
-    with dctx.write_to(fh) as decompressor:
+    with dctx.stream_writer(fh) as decompressor:
         byte_size = decompressor.memory_size()
 
 Streaming Output API
 ^^^^^^^^^^^^^^^^^^^^
 
-``read_from(fh)`` provides a mechanism to stream decompressed data out of a
+``read_to_iter(fh)`` provides a mechanism to stream decompressed data out of a
 compressed source as an iterator of data chunks.:: 
 
     dctx = zstd.ZstdDecompressor()
-    for chunk in dctx.read_from(fh):
+    for chunk in dctx.read_to_iter(fh):
         # Do something with original data.
 
-``read_from()`` accepts a) an object with a ``read(size)`` method that will
-return  compressed bytes b) an object conforming to the buffer protocol that
-can expose its data as a contiguous range of bytes. The ``bytes`` and
-``memoryview`` types expose this buffer protocol.
+``read_to_iter()`` accepts an object with a ``read(size)`` method that will
+return  compressed bytes or an object conforming to the buffer protocol that
+can expose its data as a contiguous range of bytes.
 
-``read_from()`` returns an iterator whose elements are chunks of the
+``read_to_iter()`` returns an iterator whose elements are chunks of the
 decompressed data.
 
 The size of requested ``read()`` from the source can be specified::
 
     dctx = zstd.ZstdDecompressor()
-    for chunk in dctx.read_from(fh, read_size=16384):
+    for chunk in dctx.read_to_iter(fh, read_size=16384):
         pass
 
 It is also possible to skip leading bytes in the input data::
 
     dctx = zstd.ZstdDecompressor()
-    for chunk in dctx.read_from(fh, skip_bytes=1):
+    for chunk in dctx.read_to_iter(fh, skip_bytes=1):
         pass
 
-Skipping leading bytes is useful if the source data contains extra
-*header* data but you want to avoid the overhead of making a buffer copy
-or allocating a new ``memoryview`` object in order to decompress the data.
+.. tip::
 
-Similarly to ``ZstdCompressor.read_from()``, the consumer of the iterator
+   Skipping leading bytes is useful if the source data contains extra
+   *header* data. Traditionally, you would need to create a slice or
+   ``memoryview`` of the data you want to decompress. This would create
+   overhead. It is more efficient to pass the offset into this API.
+
+Similarly to ``ZstdCompressor.read_to_iter()``, the consumer of the iterator
 controls when data is decompressed. If the iterator isn't consumed,
 decompression is put on hold.
 
-When ``read_from()`` is passed an object conforming to the buffer protocol,
+When ``read_to_iter()`` is passed an object conforming to the buffer protocol,
 the behavior may seem similar to what occurs when the simple decompression
 API is used. However, this API works when the decompressed size is unknown.
 Furthermore, if feeding large inputs, the decompressor will work in chunks
@@ -636,7 +682,7 @@
 ^^^^^^^^^^^^^^^^
 
 ``decompressobj()`` returns an object that exposes a ``decompress(data)``
-methods. Compressed data chunks are fed into ``decompress(data)`` and
+method. Compressed data chunks are fed into ``decompress(data)`` and
 uncompressed output (or an empty bytes) is returned. Output from subsequent
 calls needs to be concatenated to reassemble the full decompressed byte
 sequence.
@@ -650,11 +696,25 @@
 
 Here is how this API should be used::
 
-   dctx = zstd.ZstdDeompressor()
-   dobj = cctx.decompressobj()
+   dctx = zstd.ZstdDecompressor()
+   dobj = dctx.decompressobj()
    data = dobj.decompress(compressed_chunk_0)
    data = dobj.decompress(compressed_chunk_1)
 
+By default, calls to ``decompress()`` write output data in chunks of size
+``DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE``. These chunks are concatenated
+before being returned to the caller. It is possible to define the size of
+these temporary chunks by passing ``write_size`` to ``decompressobj()``::
+
+   dctx = zstd.ZstdDecompressor()
+   dobj = dctx.decompressobj(write_size=1048576)
+
+.. note::
+
+   Because calls to ``decompress()`` may need to perform multiple
+   memory (re)allocations, this streaming decompression API isn't as
+   efficient as other APIs.
+
 Batch Decompression API
 ^^^^^^^^^^^^^^^^^^^^^^^
 
@@ -671,9 +731,12 @@
 minimal input validation will be done for that type. If calling from
 Python (as opposed to C), constructing one of these instances may add
 overhead cancelling out the performance overhead of validation for list
-inputs.
+inputs.::
 
-The decompressed size of each frame must be discoverable. It can either be
+    dctx = zstd.ZstdDecompressor()
+    results = dctx.multi_decompress_to_buffer([b'...', b'...'])
+
+The decompressed size of each frame MUST be discoverable. It can either be
 embedded within the zstd frame (``write_content_size=True`` argument to
 ``ZstdCompressor``) or passed in via the ``decompressed_sizes`` argument.
 
@@ -681,7 +744,13 @@
 protocol which holds an array of 64-bit unsigned integers in the machine's
 native format defining the decompressed sizes of each frame. If this argument
 is passed, it avoids having to scan each frame for its decompressed size.
-This frame scanning can add noticeable overhead in some scenarios.
+This frame scanning can add noticeable overhead in some scenarios.::
+
+    frames = [...]
+    sizes = struct.pack('=QQQQ', len0, len1, len2, len3)
+
+    dctx = zstd.ZstdDecompressor()
+    results = dctx.multi_decompress_to_buffer(frames, decompressed_sizes=sizes)
 
 The ``threads`` argument controls the number of threads to use to perform
 decompression operations. The default (``0``) or the value ``1`` means to
@@ -701,22 +770,23 @@
 as possible by having as little overhead as possible. Since decompression is
 performed as a single operation and since the decompressed output is stored in
 a single buffer, extra memory allocations, Python objects, and Python function
-calls are avoided. This is ideal for scenarios where callers need to access
-decompressed data for multiple frames.
+calls are avoided. This is ideal for scenarios where callers know up front that
+they need to access data for multiple frames, such as when  *delta chains* are
+being used.
 
 Currently, the implementation always spawns multiple threads when requested,
 even if the amount of work to do is small. In the future, it will be smarter
 about avoiding threads and their associated overhead when the amount of
 work to do is small.
 
-Content-Only Dictionary Chain Decompression
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+Prefix Dictionary Chain Decompression
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 ``decompress_content_dict_chain(frames)`` performs decompression of a list of
-zstd frames produced using chained *content-only* dictionary compression. Such
+zstd frames produced using chained *prefix* dictionary compression. Such
 a list of frames is produced by compressing discrete inputs where each
-non-initial input is compressed with a *content-only* dictionary consisting
-of the content of the previous input.
+non-initial input is compressed with a *prefix* dictionary consisting of the
+content of the previous input.
 
 For example, say you have the following inputs::
 
@@ -725,25 +795,25 @@
 The zstd frame chain consists of:
 
 1. ``b'input 1'`` compressed in standalone/discrete mode
-2. ``b'input 2'`` compressed using ``b'input 1'`` as a *content-only* dictionary
-3. ``b'input 3'`` compressed using ``b'input 2'`` as a *content-only* dictionary
+2. ``b'input 2'`` compressed using ``b'input 1'`` as a *prefix* dictionary
+3. ``b'input 3'`` compressed using ``b'input 2'`` as a *prefix* dictionary
 
 Each zstd frame **must** have the content size written.
 
-The following Python code can be used to produce a *content-only dictionary
-chain*::
+The following Python code can be used to produce a *prefix dictionary chain*::
 
     def make_chain(inputs):
         frames = []
 
         # First frame is compressed in standalone/discrete mode.
-        zctx = zstd.ZstdCompressor(write_content_size=True)
+        zctx = zstd.ZstdCompressor()
         frames.append(zctx.compress(inputs[0]))
 
-        # Subsequent frames use the previous fulltext as a content-only dictionary
+        # Subsequent frames use the previous fulltext as a prefix dictionary
         for i, raw in enumerate(inputs[1:]):
-            dict_data = zstd.ZstdCompressionDict(inputs[i])
-            zctx = zstd.ZstdCompressor(write_content_size=True, dict_data=dict_data)
+            dict_data = zstd.ZstdCompressionDict(
+                inputs[i], dict_type=zstd.DICT_TYPE_RAWCONTENT)
+            zctx = zstd.ZstdCompressor(dict_data=dict_data)
             frames.append(zctx.compress(raw))
 
         return frames
@@ -751,10 +821,13 @@
 ``decompress_content_dict_chain()`` returns the uncompressed data of the last
 element in the input chain.
 
-It is possible to implement *content-only dictionary chain* decompression
-on top of other Python APIs. However, this function will likely be significantly
-faster, especially for long input chains, as it avoids the overhead of
-instantiating and passing around intermediate objects between C and Python.
+
+.. note::
+
+   It is possible to implement *prefix dictionary chain* decompression
+   on top of other APIs. However, this function will likely be faster -
+   especially for long input chains - as it avoids the overhead of instantiating
+   and passing around intermediate objects between C and Python.
 
 Multi-Threaded Compression
 --------------------------
@@ -764,9 +837,15 @@
 into segments and each segment is fed into a worker pool for compression. Once
 a segment is compressed, it is flushed/appended to the output.
 
+.. note::
+
+   These threads are created at the C layer and are not Python threads. So they
+   work outside the GIL. It is therefore possible to CPU saturate multiple cores
+   from Python.
+
 The segment size for multi-threaded compression is chosen from the window size
 of the compressor. This is derived from the ``window_log`` attribute of a
-``CompressionParameters`` instance. By default, segment sizes are in the 1+MB
+``ZstdCompressionParameters`` instance. By default, segment sizes are in the 1+MB
 range.
 
 If multi-threaded compression is requested and the input is smaller than the
@@ -785,31 +864,33 @@
 there is a CPU/wall time versus size trade off that may warrant investigation.
 
 Output from multi-threaded compression does not require any special handling
-on the decompression side. In other words, any zstd decompressor should be able
-to consume data produced with multi-threaded compression.
+on the decompression side. To the decompressor, data generated with single
+threaded compressor looks the same as data generated by a multi-threaded
+compressor and does not require any special handling or additional resource
+requirements.
 
 Dictionary Creation and Management
 ----------------------------------
 
-Compression dictionaries are represented as the ``ZstdCompressionDict`` type.
+Compression dictionaries are represented with the ``ZstdCompressionDict`` type.
 
 Instances can be constructed from bytes::
 
    dict_data = zstd.ZstdCompressionDict(data)
 
-It is possible to construct a dictionary from *any* data. Unless the
-data begins with a magic header, the dictionary will be treated as
-*content-only*. *Content-only* dictionaries allow compression operations
-that follow to reference raw data within the content. For one use of
-*content-only* dictionaries, see
-``ZstdDecompressor.decompress_content_dict_chain()``.
+It is possible to construct a dictionary from *any* data. If the data doesn't
+begin with a magic header, it will be treated as a *prefix* dictionary.
+*Prefix* dictionaries allow compression operations to reference raw data
+within the dictionary.
 
-More interestingly, instances can be created by *training* on sample data::
+It is possible to force the use of *prefix* dictionaries or to require a
+dictionary header:
 
-   dict_data = zstd.train_dictionary(size, samples)
+   dict_data = zstd.ZstdCompressionDict(data,
+                                        dict_type=zstd.DICT_TYPE_RAWCONTENT)
 
-This takes a list of bytes instances and creates and returns a
-``ZstdCompressionDict``.
+   dict_data = zstd.ZstdCompressionDict(data,
+                                        dict_type=zstd.DICT_TYPE_FULLDICT)
 
 You can see how many bytes are in the dictionary by calling ``len()``::
 
@@ -819,7 +900,7 @@
 Once you have a dictionary, you can pass it to the objects performing
 compression and decompression::
 
-   dict_data = zstd.train_dictionary(16384, samples)
+   dict_data = zstd.train_dictionary(131072, samples)
 
    cctx = zstd.ZstdCompressor(dict_data=dict_data)
    for source_data in input_data:
@@ -829,7 +910,7 @@
    dctx = zstd.ZstdDecompressor(dict_data=dict_data)
    for compressed_data in input_data:
        buffer = io.BytesIO()
-       with dctx.write_to(buffer) as decompressor:
+       with dctx.stream_writer(buffer) as decompressor:
            decompressor.write(compressed_data)
        # Do something with raw data in ``buffer``.
 
@@ -843,56 +924,69 @@
    dict_data = zstd.train_dictionary(size, samples)
    raw_data = dict_data.as_bytes()
 
-The following named arguments to ``train_dictionary`` can also be used
-to further control dictionary generation.
+By default, when a ``ZstdCompressionDict`` is *attached* to a
+``ZstdCompressor``, each ``ZstdCompressor`` performs work to prepare the
+dictionary for use. This is fine if only 1 compression operation is being
+performed or if the ``ZstdCompressor`` is being reused for multiple operations.
+But if multiple ``ZstdCompressor`` instances are being used with the dictionary,
+this can add overhead.
 
-selectivity
-   Integer selectivity level. Default is 9. Larger values yield more data in
-   dictionary.
-level
-   Integer compression level. Default is 6.
-dict_id
-   Integer dictionary ID for the produced dictionary. Default is 0, which
-   means to use a random value.
-notifications
-   Controls writing of informational messages to ``stderr``. ``0`` (the
-   default) means to write nothing. ``1`` writes errors. ``2`` writes
-   progression info. ``3`` writes more details. And ``4`` writes all info.
+It is possible to *precompute* the dictionary so it can readily be consumed
+by multiple ``ZstdCompressor`` instances::
+
+    d = zstd.ZstdCompressionDict(data)
 
-Cover Dictionaries
-^^^^^^^^^^^^^^^^^^
+    # Precompute for compression level 3.
+    d.precompute_compress(level=3)
 
-An alternate dictionary training mechanism named *cover* is also available.
-More details about this training mechanism are available in the paper
-*Effective Construction of Relative Lempel-Ziv Dictionaries* (authors:
-Liao, Petri, Moffat, Wirth).
-
-To use this mechanism, use ``zstd.train_cover_dictionary()`` instead of
-``zstd.train_dictionary()``. The function behaves nearly the same except
-its arguments are different and the returned dictionary will contain ``k``
-and ``d`` attributes reflecting the parameters to the cover algorithm.
+    # Precompute with specific compression parameters.
+    params = zstd.ZstdCompressionParameters(...)
+    d.precompute_compress(compression_params=params)
 
 .. note::
 
-   The ``k`` and ``d`` attributes are only populated on dictionary
-   instances created by this function. If a ``ZstdCompressionDict`` is
-   constructed from raw bytes data, the ``k`` and ``d`` attributes will
-   be ``0``.
+   When a dictionary is precomputed, the compression parameters used to
+   precompute the dictionary overwrite some of the compression parameters
+   specified to ``ZstdCompressor.__init__``.
+
+Training Dictionaries
+^^^^^^^^^^^^^^^^^^^^^
+
+Unless using *prefix* dictionaries, dictionary data is produced by *training*
+on existing data::
+
+   dict_data = zstd.train_dictionary(size, samples)
+
+This takes a target dictionary size and list of bytes instances and creates and
+returns a ``ZstdCompressionDict``.
+
+The dictionary training mechanism is known as *cover*. More details about it are
+available in the paper *Effective Construction of Relative Lempel-Ziv
+Dictionaries* (authors: Liao, Petri, Moffat, Wirth).
+
+The cover algorithm takes parameters ``k` and ``d``. These are the
+*segment size* and *dmer size*, respectively. The returned dictionary
+instance created by this function has ``k`` and ``d`` attributes
+containing the values for these parameters. If a ``ZstdCompressionDict``
+is constructed from raw bytes data (a content-only dictionary), the
+``k`` and ``d`` attributes will be ``0``.
 
 The segment and dmer size parameters to the cover algorithm can either be
-specified manually or you can ask ``train_cover_dictionary()`` to try
-multiple values and pick the best one, where *best* means the smallest
-compressed data size.
-
-In manual mode, the ``k`` and ``d`` arguments must be specified or a
-``ZstdError`` will be raised.
+specified manually or ``train_dictionary()`` can try multiple values
+and pick the best one, where *best* means the smallest compressed data size.
+This later mode is called *optimization* mode.
 
-In automatic mode (triggered by specifying ``optimize=True``), ``k``
-and ``d`` are optional. If a value isn't specified, then default values for
-both are tested.  The ``steps`` argument can control the number of steps
-through ``k`` values. The ``level`` argument defines the compression level
-that will be used when testing the compressed size. And ``threads`` can
-specify the number of threads to use for concurrent operation.
+If none of ``k``, ``d``, ``steps``, ``threads``, ``level``, ``notifications``,
+or ``dict_id`` (basically anything from the underlying ``ZDICT_cover_params_t``
+struct) are defined, *optimization* mode is used with default parameter
+values.
+
+If ``steps`` or ``threads`` are defined, then *optimization* mode is engaged
+with explicit control over those parameters. Specifying ``threads=0`` or
+``threads=1`` can be used to engage *optimization* mode if other parameters
+are not defined.
+
+Otherwise, non-*optimization* mode is used with the parameters specified.
 
 This function takes the following arguments:
 
@@ -909,64 +1003,92 @@
 dict_id
    Integer dictionary ID for the produced dictionary. Default is 0, which uses
    a random value.
-optimize
-   When true, test dictionary generation with multiple parameters.
+steps
+   Number of steps through ``k`` values to perform when trying parameter
+   variations.
+threads
+   Number of threads to use when trying parameter variations. Default is 0,
+   which means to use a single thread. A negative value can be specified to
+   use as many threads as there are detected logical CPUs.
 level
-   Integer target compression level when testing compression with
-   ``optimize=True``. Default is 1.
-steps
-   Number of steps through ``k`` values to perform when ``optimize=True``.
-   Default is 32.
-threads
-   Number of threads to use when ``optimize=True``. Default is 0, which means
-   to use a single thread. A negative value can be specified to use as many
-   threads as there are detected logical CPUs.
+   Integer target compression level when trying parameter variations.
 notifications
-   Controls writing of informational messages to ``stderr``. See the
-   documentation for ``train_dictionary()`` for more.
+   Controls writing of informational messages to ``stderr``. ``0`` (the
+   default) means to write nothing. ``1`` writes errors. ``2`` writes
+   progression info. ``3`` writes more details. And ``4`` writes all info.
 
 Explicit Compression Parameters
 -------------------------------
 
-Zstandard's integer compression levels along with the input size and dictionary
-size are converted into a data structure defining multiple parameters to tune
-behavior of the compression algorithm. It is possible to use define this
-data structure explicitly to have lower-level control over compression behavior.
+Zstandard offers a high-level *compression level* that maps to lower-level
+compression parameters. For many consumers, this numeric level is the only
+compression setting you'll need to touch.
+
+But for advanced use cases, it might be desirable to tweak these lower-level
+settings.
 
-The ``zstd.CompressionParameters`` type represents this data structure.
-You can see how Zstandard converts compression levels to this data structure
-by calling ``zstd.get_compression_parameters()``. e.g.::
+The ``ZstdCompressionParameters`` type represents these low-level compression
+settings.
 
-    params = zstd.get_compression_parameters(5)
+Instances of this type can be constructed from a myriad of keyword arguments
+(defined below) for complete low-level control over each adjustable
+compression setting.
+
+From a higher level, one can construct a ``ZstdCompressionParameters`` instance
+given a desired compression level and target input and dictionary size
+using ``ZstdCompressionParameters.from_level()``. e.g.::
 
-This function also accepts the uncompressed data size and dictionary size
-to adjust parameters::
+    # Derive compression settings for compression level 7.
+    params = zstd.ZstdCompressionParameters.from_level(7)
 
-    params = zstd.get_compression_parameters(3, source_size=len(data), dict_size=len(dict_data))
+    # With an input size of 1MB
+    params = zstd.ZstdCompressionParameters.from_level(7, source_size=1048576)
+
+Using ``from_level()``, it is also possible to override individual compression
+parameters or to define additional settings that aren't automatically derived.
+e.g.::
 
-You can also construct compression parameters from their low-level components::
+    params = zstd.ZstdCompressionParameters.from_level(4, window_log=10)
+    params = zstd.ZstdCompressionParameters.from_level(5, threads=4)
+
+Or you can define low-level compression settings directly::
 
-    params = zstd.CompressionParameters(20, 6, 12, 5, 4, 10, zstd.STRATEGY_FAST)
+    params = zstd.ZstdCompressionParameters(window_log=12, enable_ldm=True)
 
-You can then configure a compressor to use the custom parameters::
+Once a ``ZstdCompressionParameters`` instance is obtained, it can be used to
+configure a compressor::
 
     cctx = zstd.ZstdCompressor(compression_params=params)
 
-The members/attributes of ``CompressionParameters`` instances are as follows::
+The named arguments and attributes of ``ZstdCompressionParameters`` are as
+follows:
 
+* format
+* compression_level
 * window_log
+* hash_log
 * chain_log
-* hash_log
 * search_log
-* search_length
+* min_match
 * target_length
-* strategy
+* compression_strategy
+* write_content_size
+* write_checksum
+* write_dict_id
+* job_size
+* overlap_size_log
+* compress_literals
+* force_max_window
+* enable_ldm
+* ldm_hash_log
+* ldm_min_match
+* ldm_bucket_size_log
+* ldm_hash_every_log
+* threads
 
-This is the order the arguments are passed to the constructor if not using
-named arguments.
-
-You'll need to read the Zstandard documentation for what these parameters
-do.
+Some of these are very low-level settings. It may help to consult the official
+zstandard documentation for their behavior. Look for the ``ZSTD_p_*`` constants
+in ``zstd.h`` (https://github.com/facebook/zstd/blob/dev/lib/zstd.h).
 
 Frame Inspection
 ----------------
@@ -1003,15 +1125,17 @@
    Bool indicating whether a 4 byte content checksum is stored at the end
    of the frame.
 
+``zstd.frame_header_size(data)`` returns the size of the zstandard frame
+header.
+
+``zstd.frame_content_size(data)`` returns the content size as parsed from
+the frame header. ``-1`` means the content size is unknown. ``0`` means
+an empty frame. The content size is usually correct. However, it may not
+be accurate.
+
 Misc Functionality
 ------------------
 
-estimate_compression_context_size(CompressionParameters)
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-
-Given a ``CompressionParameters`` struct, estimate the memory size required
-to perform compression.
-
 estimate_decompression_context_size()
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
@@ -1041,6 +1165,11 @@
 MAGIC_NUMBER
     Frame header as an integer
 
+CONTENTSIZE_UNKNOWN
+    Value for content size when the content size is unknown.
+CONTENTSIZE_ERROR
+    Value for content size when content size couldn't be determined.
+
 WINDOWLOG_MIN
     Minimum value for compression parameter
 WINDOWLOG_MAX
@@ -1063,8 +1192,6 @@
     Maximum value for compression parameter
 TARGETLENGTH_MIN
     Minimum value for compression parameter
-TARGETLENGTH_MAX
-    Maximum value for compression parameter
 STRATEGY_FAST
     Compression strategy
 STRATEGY_DFAST
@@ -1079,6 +1206,13 @@
     Compression strategy
 STRATEGY_BTOPT
     Compression strategy
+STRATEGY_BTULTRA
+    Compression strategy
+
+FORMAT_ZSTD1
+    Zstandard frame format
+FORMAT_ZSTD1_MAGICLESS
+    Zstandard frame format without magic header
 
 Performance Considerations
 --------------------------
@@ -1090,7 +1224,7 @@
 operation. The differences are magnified as the size of data decreases. For
 example, the difference between *context* reuse and non-reuse for 100,000
 100 byte inputs will be significant (possiby over 10x faster to reuse contexts)
-whereas 10 1,000,000 byte inputs will be more similar in speed (because the
+whereas 10 100,000,000 byte inputs will be more similar in speed (because the
 time spent doing compression dwarfs time spent creating new *contexts*).
 
 Buffer Types
@@ -1187,9 +1321,8 @@
 because different applications have different needs and the library wants to
 facilitate optimal use in as many use cases as possible.
 
-From a high-level, APIs are divided into *one-shot* and *streaming*. See
-the ``Concepts`` section for a description of how these are different at
-the C layer.
+From a high-level, APIs are divided into *one-shot* and *streaming*: either you
+are operating on all data at once or you operate on it piecemeal.
 
 The *one-shot* APIs are useful for small data, where the input or output
 size is known. (The size can come from a buffer length, file size, or
@@ -1222,145 +1355,39 @@
 underlying stream (say from interacting with a filesystem or network). This
 could add considerable overhead.
 
-Concepts
-========
-
-It is important to have a basic understanding of how Zstandard works in order
-to optimally use this library. In addition, there are some low-level Python
-concepts that are worth explaining to aid understanding. This section aims to
-provide that knowledge.
-
-Zstandard Frames and Compression Format
----------------------------------------
-
-Compressed zstandard data almost always exists within a container called a
-*frame*. (For the technically curious, see the
-`specification <https://github.com/facebook/zstd/blob/3bee41a70eaf343fbcae3637b3f6edbe52f35ed8/doc/zstd_compression_format.md>_.)
-
-The frame contains a header and optional trailer. The header contains a
-magic number to self-identify as a zstd frame and a description of the
-compressed data that follows.
-
-Among other things, the frame *optionally* contains the size of the
-decompressed data the frame represents, a 32-bit checksum of the
-decompressed data (to facilitate verification during decompression),
-and the ID of the dictionary used to compress the data.
-
-Storing the original content size in the frame (``write_content_size=True``
-to ``ZstdCompressor``) is important for performance in some scenarios. Having
-the decompressed size stored there (or storing it elsewhere) allows
-decompression to perform a single memory allocation that is exactly sized to
-the output. This is faster than continuously growing a memory buffer to hold
-output.
+Thread Safety
+=============
 
-Compression and Decompression Contexts
---------------------------------------
-
-In order to perform a compression or decompression operation with the zstd
-C API, you need what's called a *context*. A context essentially holds
-configuration and state for a compression or decompression operation. For
-example, a compression context holds the configured compression level.
-
-Contexts can be reused for multiple operations. Since creating and
-destroying contexts is not free, there are performance advantages to
-reusing contexts.
-
-The ``ZstdCompressor`` and ``ZstdDecompressor`` types are essentially
-wrappers around these contexts in the zstd C API.
+``ZstdCompressor`` and ``ZstdDecompressor`` instances have no guarantees
+about thread safety. Do not operate on the same ``ZstdCompressor`` and
+``ZstdDecompressor`` instance simultaneously from different threads. It is
+fine to have different threads call into a single instance, just not at the
+same time.
 
-One-shot And Streaming Operations
----------------------------------
-
-A compression or decompression operation can either be performed as a
-single *one-shot* operation or as a continuous *streaming* operation.
-
-In one-shot mode (the *simple* APIs provided by the Python interface),
-**all** input is handed to the compressor or decompressor as a single buffer
-and **all** output is returned as a single buffer.
-
-In streaming mode, input is delivered to the compressor or decompressor as
-a series of chunks via multiple function calls. Likewise, output is
-obtained in chunks as well.
-
-Streaming operations require an additional *stream* object to be created
-to track the operation. These are logical extensions of *context*
-instances.
+Some operations require multiple function calls to complete. e.g. streaming
+operations. A single ``ZstdCompressor`` or ``ZstdDecompressor`` cannot be used
+for simultaneously active operations. e.g. you must not start a streaming
+operation when another streaming operation is already active.
 
-There are advantages and disadvantages to each mode of operation. There
-are scenarios where certain modes can't be used. See the
-``Choosing an API`` section for more.
-
-Dictionaries
-------------
-
-A compression *dictionary* is essentially data used to seed the compressor
-state so it can achieve better compression. The idea is that if you are
-compressing a lot of similar pieces of data (e.g. JSON documents or anything
-sharing similar structure), then you can find common patterns across multiple
-objects then leverage those common patterns during compression and
-decompression operations to achieve better compression ratios.
-
-Dictionary compression is generally only useful for small inputs - data no
-larger than a few kilobytes. The upper bound on this range is highly dependent
-on the input data and the dictionary.
-
-Python Buffer Protocol
-----------------------
-
-Many functions in the library operate on objects that implement Python's
-`buffer protocol <https://docs.python.org/3.6/c-api/buffer.html>`_.
-
-The *buffer protocol* is an internal implementation detail of a Python
-type that allows instances of that type (objects) to be exposed as a raw
-pointer (or buffer) in the C API. In other words, it allows objects to be
-exposed as an array of bytes.
+The C extension releases the GIL during non-trivial calls into the zstd C
+API. Non-trivial calls are notably compression and decompression. Trivial
+calls are things like parsing frame parameters. Where the GIL is released
+is considered an implementation detail and can change in any release.
 
-From the perspective of the C API, objects implementing the *buffer protocol*
-all look the same: they are just a pointer to a memory address of a defined
-length. This allows the C API to be largely type agnostic when accessing their
-data. This allows custom types to be passed in without first converting them
-to a specific type.
-
-Many Python types implement the buffer protocol. These include ``bytes``
-(``str`` on Python 2), ``bytearray``, ``array.array``, ``io.BytesIO``,
-``mmap.mmap``, and ``memoryview``.
-
-``python-zstandard`` APIs that accept objects conforming to the buffer
-protocol require that the buffer is *C contiguous* and has a single
-dimension (``ndim==1``). This is usually the case. An example of where it
-is not is a Numpy matrix type.
-
-Requiring Output Sizes for Non-Streaming Decompression APIs
------------------------------------------------------------
-
-Non-streaming decompression APIs require that either the output size is
-explicitly defined (either in the zstd frame header or passed into the
-function) or that a max output size is specified. This restriction is for
-your safety.
-
-The *one-shot* decompression APIs store the decompressed result in a
-single buffer. This means that a buffer needs to be pre-allocated to hold
-the result. If the decompressed size is not known, then there is no universal
-good default size to use. Any default will fail or will be highly sub-optimal
-in some scenarios (it will either be too small or will put stress on the
-memory allocator to allocate a too large block).
-
-A *helpful* API may retry decompression with buffers of increasing size.
-While useful, there are obvious performance disadvantages, namely redoing
-decompression N times until it works. In addition, there is a security
-concern. Say the input came from highly compressible data, like 1 GB of the
-same byte value. The output size could be several magnitudes larger than the
-input size. An input of <100KB could decompress to >1GB. Without a bounds
-restriction on the decompressed size, certain inputs could exhaust all system
-memory. That's not good and is why the maximum output size is limited.
+APIs that accept bytes-like objects don't enforce that the underlying object
+is read-only. However, it is assumed that the passed object is read-only for
+the duration of the function call. It is possible to pass a mutable object
+(like a ``bytearray``) to e.g. ``ZstdCompressor.compress()``, have the GIL
+released, and mutate the object from another thread. Such a race condition
+is a bug in the consumer of python-zstandard. Most Python data types are
+immutable, so unless you are doing something fancy, you don't need to
+worry about this.
 
 Note on Zstandard's *Experimental* API
 ======================================
 
 Many of the Zstandard APIs used by this module are marked as *experimental*
-within the Zstandard project. This includes a large number of useful
-features, such as compression and frame parameters and parts of dictionary
-compression.
+within the Zstandard project.
 
 It is unclear how Zstandard's C API will evolve over time, especially with
 regards to this *experimental* functionality. We will try to maintain
@@ -1371,7 +1398,7 @@
 module and since we compile against it, the behavior of a specific
 version of this module should be constant for all of time. So if you
 pin the version of this module used in your projects (which is a Python
-best practice), you should be buffered from unwanted future changes.
+best practice), you should be shielded from unwanted future changes.
 
 Donate
 ======
--- a/contrib/python-zstandard/c-ext/bufferutil.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/bufferutil.c	Mon Apr 09 10:13:29 2018 -0700
@@ -83,7 +83,7 @@
 	}
 
 	if (segments.len % sizeof(BufferSegment)) {
-		PyErr_Format(PyExc_ValueError, "segments array size is not a multiple of %lu",
+		PyErr_Format(PyExc_ValueError, "segments array size is not a multiple of %zu",
 			sizeof(BufferSegment));
 		goto except;
 	}
@@ -123,7 +123,7 @@
 	PyBuffer_Release(&self->parent);
 	PyBuffer_Release(&segments);
 	return -1;
-};
+}
 
 /**
  * Construct a BufferWithSegments from existing memory and offsets.
@@ -188,6 +188,12 @@
 		return NULL;
 	}
 
+	if (self->segments[i].length > PY_SSIZE_T_MAX) {
+		PyErr_Format(PyExc_ValueError,
+			"item at offset %zd is too large for this platform", i);
+		return NULL;
+	}
+
 	result = (ZstdBufferSegment*)PyObject_CallObject((PyObject*)&ZstdBufferSegmentType, NULL);
 	if (NULL == result) {
 		return NULL;
@@ -197,7 +203,7 @@
 	Py_INCREF(self);
 
 	result->data = (char*)self->data + self->segments[i].offset;
-	result->dataSize = self->segments[i].length;
+	result->dataSize = (Py_ssize_t)self->segments[i].length;
 	result->offset = self->segments[i].offset;
 
 	return result;
@@ -205,7 +211,13 @@
 
 #if PY_MAJOR_VERSION >= 3
 static int BufferWithSegments_getbuffer(ZstdBufferWithSegments* self, Py_buffer* view, int flags) {
-	return PyBuffer_FillInfo(view, (PyObject*)self, self->data, self->dataSize, 1, flags);
+	if (self->dataSize > PY_SSIZE_T_MAX) {
+		view->obj = NULL;
+		PyErr_SetString(PyExc_BufferError, "buffer is too large for this platform");
+		return -1;
+	}
+
+	return PyBuffer_FillInfo(view, (PyObject*)self, self->data, (Py_ssize_t)self->dataSize, 1, flags);
 }
 #else
 static Py_ssize_t BufferWithSegments_getreadbuffer(ZstdBufferWithSegments* self, Py_ssize_t segment, void **ptrptr) {
@@ -214,8 +226,13 @@
 		return -1;
 	}
 
+	if (self->dataSize > PY_SSIZE_T_MAX) {
+		PyErr_SetString(PyExc_ValueError, "buffer is too large for this platform");
+		return -1;
+	}
+
 	*ptrptr = self->data;
-	return self->dataSize;
+	return (Py_ssize_t)self->dataSize;
 }
 
 static Py_ssize_t BufferWithSegments_getsegcount(ZstdBufferWithSegments* self, Py_ssize_t* len) {
@@ -232,7 +249,12 @@
 );
 
 static PyObject* BufferWithSegments_tobytes(ZstdBufferWithSegments* self) {
-	return PyBytes_FromStringAndSize(self->data, self->dataSize);
+	if (self->dataSize > PY_SSIZE_T_MAX) {
+		PyErr_SetString(PyExc_ValueError, "buffer is too large for this platform");
+		return NULL;
+	}
+
+	return PyBytes_FromStringAndSize(self->data, (Py_ssize_t)self->dataSize);
 }
 
 PyDoc_STRVAR(BufferWithSegments_segments__doc__,
--- a/contrib/python-zstandard/c-ext/compressiondict.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/compressiondict.c	Mon Apr 09 10:13:29 2018 -0700
@@ -14,125 +14,11 @@
 	static char* kwlist[] = {
 		"dict_size",
 		"samples",
-		"selectivity",
-		"level",
-		"notifications",
-		"dict_id",
-		NULL
-	};
-	size_t capacity;
-	PyObject* samples;
-	Py_ssize_t samplesLen;
-	unsigned  selectivity = 0;
-	int level = 0;
-	unsigned notifications = 0;
-	unsigned dictID = 0;
-	ZDICT_params_t zparams;
-	Py_ssize_t sampleIndex;
-	Py_ssize_t sampleSize;
-	PyObject* sampleItem;
-	size_t zresult;
-	void* sampleBuffer = NULL;
-	void* sampleOffset;
-	size_t samplesSize = 0;
-	size_t* sampleSizes = NULL;
-	void* dict = NULL;
-	ZstdCompressionDict* result = NULL;
-
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "nO!|IiII:train_dictionary",
-		kwlist,
-		&capacity,
-		&PyList_Type, &samples,
-		&selectivity, &level, &notifications, &dictID)) {
-		return NULL;
-	}
-
-	memset(&zparams, 0, sizeof(zparams));
-
-	zparams.selectivityLevel = selectivity;
-	zparams.compressionLevel = level;
-	zparams.notificationLevel = notifications;
-	zparams.dictID = dictID;
-
-	/* Figure out the size of the raw samples */
-	samplesLen = PyList_Size(samples);
-	for (sampleIndex = 0; sampleIndex < samplesLen; sampleIndex++) {
-		sampleItem = PyList_GetItem(samples, sampleIndex);
-		if (!PyBytes_Check(sampleItem)) {
-			PyErr_SetString(PyExc_ValueError, "samples must be bytes");
-			return NULL;
-		}
-		samplesSize += PyBytes_GET_SIZE(sampleItem);
-	}
-
-	/* Now that we know the total size of the raw simples, we can allocate
-	a buffer for the raw data */
-	sampleBuffer = PyMem_Malloc(samplesSize);
-	if (!sampleBuffer) {
-		PyErr_NoMemory();
-		goto finally;
-	}
-	sampleSizes = PyMem_Malloc(samplesLen * sizeof(size_t));
-	if (!sampleSizes) {
-		PyErr_NoMemory();
-		goto finally;
-	}
-
-	sampleOffset = sampleBuffer;
-	/* Now iterate again and assemble the samples in the buffer */
-	for (sampleIndex = 0; sampleIndex < samplesLen; sampleIndex++) {
-		sampleItem = PyList_GetItem(samples, sampleIndex);
-		sampleSize = PyBytes_GET_SIZE(sampleItem);
-		sampleSizes[sampleIndex] = sampleSize;
-		memcpy(sampleOffset, PyBytes_AS_STRING(sampleItem), sampleSize);
-		sampleOffset = (char*)sampleOffset + sampleSize;
-	}
-
-	dict = PyMem_Malloc(capacity);
-	if (!dict) {
-		PyErr_NoMemory();
-		goto finally;
-	}
-
-	/* TODO consider using dup2() to redirect zstd's stderr writing to a buffer */
-	Py_BEGIN_ALLOW_THREADS
-	zresult = ZDICT_trainFromBuffer_advanced(dict, capacity,
-		sampleBuffer, sampleSizes, (unsigned int)samplesLen,
-		zparams);
-	Py_END_ALLOW_THREADS
-	if (ZDICT_isError(zresult)) {
-		PyErr_Format(ZstdError, "Cannot train dict: %s", ZDICT_getErrorName(zresult));
-		PyMem_Free(dict);
-		goto finally;
-	}
-
-	result = PyObject_New(ZstdCompressionDict, &ZstdCompressionDictType);
-	if (!result) {
-		goto finally;
-	}
-
-	result->dictData = dict;
-	result->dictSize = zresult;
-	result->d = 0;
-	result->k = 0;
-
-finally:
-	PyMem_Free(sampleBuffer);
-	PyMem_Free(sampleSizes);
-
-	return result;
-}
-
-ZstdCompressionDict* train_cover_dictionary(PyObject* self, PyObject* args, PyObject* kwargs) {
-	static char* kwlist[] = {
-		"dict_size",
-		"samples",
 		"k",
 		"d",
 		"notifications",
 		"dict_id",
 		"level",
-		"optimize",
 		"steps",
 		"threads",
 		NULL
@@ -145,10 +31,9 @@
 	unsigned notifications = 0;
 	unsigned dictID = 0;
 	int level = 0;
-	PyObject* optimize = NULL;
 	unsigned steps = 0;
 	int threads = 0;
-	COVER_params_t params;
+	ZDICT_cover_params_t params;
 	Py_ssize_t samplesLen;
 	Py_ssize_t i;
 	size_t samplesSize = 0;
@@ -160,9 +45,9 @@
 	size_t zresult;
 	ZstdCompressionDict* result = NULL;
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "nO!|IIIIiOIi:train_cover_dictionary",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "nO!|IIIIiIi:train_dictionary",
 		kwlist, &capacity, &PyList_Type, &samples,
-		&k, &d, &notifications, &dictID, &level, &optimize, &steps, &threads)) {
+		&k, &d, &notifications, &dictID, &level, &steps, &threads)) {
 		return NULL;
 	}
 
@@ -175,9 +60,9 @@
 	params.d = d;
 	params.steps = steps;
 	params.nbThreads = threads;
-	params.notificationLevel = notifications;
-	params.dictID = dictID;
-	params.compressionLevel = level;
+	params.zParams.notificationLevel = notifications;
+	params.zParams.dictID = dictID;
+	params.zParams.compressionLevel = level;
 
 	/* Figure out total size of input samples. */
 	samplesLen = PyList_Size(samples);
@@ -219,12 +104,21 @@
 	}
 
 	Py_BEGIN_ALLOW_THREADS
-	if (optimize && PyObject_IsTrue(optimize)) {
-		zresult = COVER_optimizeTrainFromBuffer(dict, capacity,
+	/* No parameters uses the default function, which will use default params
+	   and call ZDICT_optimizeTrainFromBuffer_cover under the hood. */
+	if (!params.k && !params.d && !params.zParams.compressionLevel
+		&& !params.zParams.notificationLevel && !params.zParams.dictID) {
+		zresult = ZDICT_trainFromBuffer(dict, capacity, sampleBuffer,
+			sampleSizes, (unsigned)samplesLen);
+	}
+	/* Use optimize mode if user controlled steps or threads explicitly. */
+	else if (params.steps || params.nbThreads) {
+		zresult = ZDICT_optimizeTrainFromBuffer_cover(dict, capacity,
 			sampleBuffer, sampleSizes, (unsigned)samplesLen, &params);
 	}
+	/* Non-optimize mode with explicit control. */
 	else {
-		zresult = COVER_trainFromBuffer(dict, capacity,
+		zresult = ZDICT_trainFromBuffer_cover(dict, capacity,
 			sampleBuffer, sampleSizes, (unsigned)samplesLen, params);
 	}
 	Py_END_ALLOW_THREADS
@@ -243,8 +137,11 @@
 
 	result->dictData = dict;
 	result->dictSize = zresult;
+	result->dictType = ZSTD_dct_fullDict;
 	result->d = params.d;
 	result->k = params.k;
+	result->cdict = NULL;
+	result->ddict = NULL;
 
 finally:
 	PyMem_Free(sampleBuffer);
@@ -253,43 +150,99 @@
 	return result;
 }
 
+int ensure_ddict(ZstdCompressionDict* dict) {
+	if (dict->ddict) {
+		return 0;
+	}
+
+	Py_BEGIN_ALLOW_THREADS
+	dict->ddict = ZSTD_createDDict_advanced(dict->dictData, dict->dictSize,
+		ZSTD_dlm_byRef, dict->dictType, ZSTD_defaultCMem);
+	Py_END_ALLOW_THREADS
+	if (!dict->ddict) {
+		PyErr_SetString(ZstdError, "could not create decompression dict");
+		return 1;
+	}
+
+	return 0;
+}
+
 PyDoc_STRVAR(ZstdCompressionDict__doc__,
 "ZstdCompressionDict(data) - Represents a computed compression dictionary\n"
 "\n"
 "This type holds the results of a computed Zstandard compression dictionary.\n"
-"Instances are obtained by calling ``train_dictionary()`` or by passing bytes\n"
-"obtained from another source into the constructor.\n"
+"Instances are obtained by calling ``train_dictionary()`` or by passing\n"
+"bytes obtained from another source into the constructor.\n"
 );
 
-static int ZstdCompressionDict_init(ZstdCompressionDict* self, PyObject* args) {
-	const char* source;
-	Py_ssize_t sourceSize;
+static int ZstdCompressionDict_init(ZstdCompressionDict* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"data",
+		"dict_type",
+		NULL
+	};
+
+	int result = -1;
+	Py_buffer source;
+	unsigned dictType = ZSTD_dct_auto;
 
 	self->dictData = NULL;
 	self->dictSize = 0;
+	self->cdict = NULL;
+	self->ddict = NULL;
 
 #if PY_MAJOR_VERSION >= 3
-	if (!PyArg_ParseTuple(args, "y#:ZstdCompressionDict",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*|I:ZstdCompressionDict",
 #else
-	if (!PyArg_ParseTuple(args, "s#:ZstdCompressionDict",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*|I:ZstdCompressionDict",
 #endif
-		&source, &sourceSize)) {
+		kwlist, &source, &dictType)) {
 		return -1;
 	}
 
-	self->dictData = PyMem_Malloc(sourceSize);
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
+	}
+
+	if (dictType != ZSTD_dct_auto && dictType != ZSTD_dct_rawContent
+		&& dictType != ZSTD_dct_fullDict) {
+		PyErr_Format(PyExc_ValueError,
+			"invalid dictionary load mode: %d; must use DICT_TYPE_* constants",
+			dictType);
+		goto finally;
+	}
+
+	self->dictType = dictType;
+
+	self->dictData = PyMem_Malloc(source.len);
 	if (!self->dictData) {
 		PyErr_NoMemory();
-		return -1;
+		goto finally;
 	}
 
-	memcpy(self->dictData, source, sourceSize);
-	self->dictSize = sourceSize;
+	memcpy(self->dictData, source.buf, source.len);
+	self->dictSize = source.len;
+
+	result = 0;
 
-	return 0;
+finally:
+	PyBuffer_Release(&source);
+	return result;
+}
+
+static void ZstdCompressionDict_dealloc(ZstdCompressionDict* self) {
+	if (self->cdict) {
+		ZSTD_freeCDict(self->cdict);
+		self->cdict = NULL;
 	}
 
-static void ZstdCompressionDict_dealloc(ZstdCompressionDict* self) {
+	if (self->ddict) {
+		ZSTD_freeDDict(self->ddict);
+		self->ddict = NULL;
+	}
+
 	if (self->dictData) {
 		PyMem_Free(self->dictData);
 		self->dictData = NULL;
@@ -298,6 +251,74 @@
 	PyObject_Del(self);
 }
 
+PyDoc_STRVAR(ZstdCompressionDict_precompute_compress__doc__,
+"Precompute a dictionary so it can be used by multiple compressors.\n"
+);
+
+static PyObject* ZstdCompressionDict_precompute_compress(ZstdCompressionDict* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"level",
+		"compression_params",
+		NULL
+	};
+
+	int level = 0;
+	ZstdCompressionParametersObject* compressionParams = NULL;
+	ZSTD_compressionParameters cParams;
+	size_t zresult;
+
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|iO!:precompute_compress", kwlist,
+		&level, &ZstdCompressionParametersType, &compressionParams)) {
+		return NULL;
+	}
+
+	if (level && compressionParams) {
+		PyErr_SetString(PyExc_ValueError,
+			"must only specify one of level or compression_params");
+		return NULL;
+	}
+
+	if (!level && !compressionParams) {
+		PyErr_SetString(PyExc_ValueError,
+			"must specify one of level or compression_params");
+		return NULL;
+	}
+
+	if (self->cdict) {
+		zresult = ZSTD_freeCDict(self->cdict);
+		self->cdict = NULL;
+		if (ZSTD_isError(zresult)) {
+			PyErr_Format(ZstdError, "unable to free CDict: %s",
+				ZSTD_getErrorName(zresult));
+			return NULL;
+		}
+	}
+
+	if (level) {
+		cParams = ZSTD_getCParams(level, 0, self->dictSize);
+	}
+	else {
+		cParams.chainLog = compressionParams->chainLog;
+		cParams.hashLog = compressionParams->hashLog;
+		cParams.searchLength = compressionParams->minMatch;
+		cParams.searchLog = compressionParams->searchLog;
+		cParams.strategy = compressionParams->compressionStrategy;
+		cParams.targetLength = compressionParams->targetLength;
+		cParams.windowLog = compressionParams->windowLog;
+	}
+
+	assert(!self->cdict);
+	self->cdict = ZSTD_createCDict_advanced(self->dictData, self->dictSize,
+		ZSTD_dlm_byRef, self->dictType, cParams, ZSTD_defaultCMem);
+
+	if (!self->cdict) {
+		PyErr_SetString(ZstdError, "unable to precompute dictionary");
+		return NULL;
+	}
+
+	Py_RETURN_NONE;
+}
+
 static PyObject* ZstdCompressionDict_dict_id(ZstdCompressionDict* self) {
 	unsigned dictID = ZDICT_getDictID(self->dictData, self->dictSize);
 
@@ -313,6 +334,8 @@
 	PyDoc_STR("dict_id() -- obtain the numeric dictionary ID") },
 	{ "as_bytes", (PyCFunction)ZstdCompressionDict_as_bytes, METH_NOARGS,
 	PyDoc_STR("as_bytes() -- obtain the raw bytes constituting the dictionary data") },
+	{ "precompute_compress", (PyCFunction)ZstdCompressionDict_precompute_compress,
+	METH_VARARGS | METH_KEYWORDS, ZstdCompressionDict_precompute_compress__doc__ },
 	{ NULL, NULL }
 };
 
--- a/contrib/python-zstandard/c-ext/compressionparams.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/compressionparams.c	Mon Apr 09 10:13:29 2018 -0700
@@ -8,204 +8,448 @@
 
 #include "python-zstandard.h"
 
-void ztopy_compression_parameters(CompressionParametersObject* params, ZSTD_compressionParameters* zparams) {
-	zparams->windowLog = params->windowLog;
-	zparams->chainLog = params->chainLog;
-	zparams->hashLog = params->hashLog;
-	zparams->searchLog = params->searchLog;
-	zparams->searchLength = params->searchLength;
-	zparams->targetLength = params->targetLength;
-	zparams->strategy = params->strategy;
-}
+extern PyObject* ZstdError;
 
-CompressionParametersObject* get_compression_parameters(PyObject* self, PyObject* args) {
-	int compressionLevel;
-	unsigned PY_LONG_LONG sourceSize = 0;
-	Py_ssize_t dictSize = 0;
-	ZSTD_compressionParameters params;
-	CompressionParametersObject* result;
-
-	if (!PyArg_ParseTuple(args, "i|Kn:get_compression_parameters",
-		&compressionLevel, &sourceSize, &dictSize)) {
-		return NULL;
+int set_parameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, unsigned value) {
+	size_t zresult = ZSTD_CCtxParam_setParameter(params, param, value);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "unable to set compression context parameter: %s",
+			ZSTD_getErrorName(zresult));
+		return 1;
 	}
 
-	params = ZSTD_getCParams(compressionLevel, sourceSize, dictSize);
+	return 0;
+}
+
+#define TRY_SET_PARAMETER(params, param, value) if (set_parameter(params, param, value)) return -1;
 
-	result = PyObject_New(CompressionParametersObject, &CompressionParametersType);
-	if (!result) {
-		return NULL;
+int set_parameters(ZSTD_CCtx_params* params, ZstdCompressionParametersObject* obj) {
+	TRY_SET_PARAMETER(params, ZSTD_p_format, obj->format);
+	TRY_SET_PARAMETER(params, ZSTD_p_compressionLevel, (unsigned)obj->compressionLevel);
+	TRY_SET_PARAMETER(params, ZSTD_p_windowLog, obj->windowLog);
+	TRY_SET_PARAMETER(params, ZSTD_p_hashLog, obj->hashLog);
+	TRY_SET_PARAMETER(params, ZSTD_p_chainLog, obj->chainLog);
+	TRY_SET_PARAMETER(params, ZSTD_p_searchLog, obj->searchLog);
+	TRY_SET_PARAMETER(params, ZSTD_p_minMatch, obj->minMatch);
+	TRY_SET_PARAMETER(params, ZSTD_p_targetLength, obj->targetLength);
+	TRY_SET_PARAMETER(params, ZSTD_p_compressionStrategy, obj->compressionStrategy);
+	TRY_SET_PARAMETER(params, ZSTD_p_contentSizeFlag, obj->contentSizeFlag);
+	TRY_SET_PARAMETER(params, ZSTD_p_checksumFlag, obj->checksumFlag);
+	TRY_SET_PARAMETER(params, ZSTD_p_dictIDFlag, obj->dictIDFlag);
+	TRY_SET_PARAMETER(params, ZSTD_p_nbWorkers, obj->threads);
+	TRY_SET_PARAMETER(params, ZSTD_p_jobSize, obj->jobSize);
+	TRY_SET_PARAMETER(params, ZSTD_p_overlapSizeLog, obj->overlapSizeLog);
+	TRY_SET_PARAMETER(params, ZSTD_p_compressLiterals, obj->compressLiterals);
+	TRY_SET_PARAMETER(params, ZSTD_p_forceMaxWindow, obj->forceMaxWindow);
+	TRY_SET_PARAMETER(params, ZSTD_p_enableLongDistanceMatching, obj->enableLongDistanceMatching);
+	TRY_SET_PARAMETER(params, ZSTD_p_ldmHashLog, obj->ldmHashLog);
+	TRY_SET_PARAMETER(params, ZSTD_p_ldmMinMatch, obj->ldmMinMatch);
+	TRY_SET_PARAMETER(params, ZSTD_p_ldmBucketSizeLog, obj->ldmBucketSizeLog);
+	TRY_SET_PARAMETER(params, ZSTD_p_ldmHashEveryLog, obj->ldmHashEveryLog);
+
+	return 0;
+}
+
+int reset_params(ZstdCompressionParametersObject* params) {
+	if (params->params) {
+		ZSTD_CCtxParams_reset(params->params);
+	}
+	else {
+		params->params = ZSTD_createCCtxParams();
+		if (!params->params) {
+			PyErr_NoMemory();
+			return 1;
+		}
 	}
 
-	result->windowLog = params.windowLog;
-	result->chainLog = params.chainLog;
-	result->hashLog = params.hashLog;
-	result->searchLog = params.searchLog;
-	result->searchLength = params.searchLength;
-	result->targetLength = params.targetLength;
-	result->strategy = params.strategy;
-
-	return result;
+	return set_parameters(params->params, params);
 }
 
-static int CompressionParameters_init(CompressionParametersObject* self, PyObject* args, PyObject* kwargs) {
+static int ZstdCompressionParameters_init(ZstdCompressionParametersObject* self, PyObject* args, PyObject* kwargs) {
 	static char* kwlist[] = {
+		"format",
+		"compression_level",
 		"window_log",
-		"chain_log",
 		"hash_log",
+		"chain_log",
 		"search_log",
-		"search_length",
+		"min_match",
 		"target_length",
-		"strategy",
+		"compression_strategy",
+		"write_content_size",
+		"write_checksum",
+		"write_dict_id",
+		"job_size",
+		"overlap_size_log",
+		"force_max_window",
+		"enable_ldm",
+		"ldm_hash_log",
+		"ldm_min_match",
+		"ldm_bucket_size_log",
+		"ldm_hash_every_log",
+		"threads",
+		"compress_literals",
 		NULL
 	};
 
-	unsigned windowLog;
-	unsigned chainLog;
-	unsigned hashLog;
-	unsigned searchLog;
-	unsigned searchLength;
-	unsigned targetLength;
-	unsigned strategy;
-	ZSTD_compressionParameters params;
-	size_t zresult;
+	unsigned format = 0;
+	int compressionLevel = 0;
+	unsigned windowLog = 0;
+	unsigned hashLog = 0;
+	unsigned chainLog = 0;
+	unsigned searchLog = 0;
+	unsigned minMatch = 0;
+	unsigned targetLength = 0;
+	unsigned compressionStrategy = 0;
+	unsigned contentSizeFlag = 1;
+	unsigned checksumFlag = 0;
+	unsigned dictIDFlag = 0;
+	unsigned jobSize = 0;
+	unsigned overlapSizeLog = 0;
+	unsigned forceMaxWindow = 0;
+	unsigned enableLDM = 0;
+	unsigned ldmHashLog = 0;
+	unsigned ldmMinMatch = 0;
+	unsigned ldmBucketSizeLog = 0;
+	unsigned ldmHashEveryLog = 0;
+	int threads = 0;
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "IIIIIII:CompressionParameters",
-		kwlist, &windowLog, &chainLog, &hashLog, &searchLog, &searchLength,
-		&targetLength, &strategy)) {
-		return -1;
-	}
+	/* Setting value 0 has the effect of disabling. So we use -1 as a default
+	 * to detect whether to set. Then we automatically derive the expected value
+	 * based on the level, just like zstandard does itself. */
+	int compressLiterals = -1;
 
-	if (windowLog < ZSTD_WINDOWLOG_MIN || windowLog > ZSTD_WINDOWLOG_MAX) {
-		PyErr_SetString(PyExc_ValueError, "invalid window log value");
-		return -1;
-	}
-
-	if (chainLog < ZSTD_CHAINLOG_MIN || chainLog > ZSTD_CHAINLOG_MAX) {
-		PyErr_SetString(PyExc_ValueError, "invalid chain log value");
-		return -1;
-	}
-
-	if (hashLog < ZSTD_HASHLOG_MIN || hashLog > ZSTD_HASHLOG_MAX) {
-		PyErr_SetString(PyExc_ValueError, "invalid hash log value");
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs,
+		"|IiIIIIIIIIIIIIIIIIIIii:CompressionParameters",
+		kwlist, &format, &compressionLevel, &windowLog, &hashLog, &chainLog,
+		&searchLog, &minMatch, &targetLength, &compressionStrategy,
+		&contentSizeFlag, &checksumFlag, &dictIDFlag, &jobSize, &overlapSizeLog,
+		&forceMaxWindow, &enableLDM, &ldmHashLog, &ldmMinMatch, &ldmBucketSizeLog,
+		&ldmHashEveryLog, &threads, &compressLiterals)) {
 		return -1;
 	}
 
-	if (searchLog < ZSTD_SEARCHLOG_MIN || searchLog > ZSTD_SEARCHLOG_MAX) {
-		PyErr_SetString(PyExc_ValueError, "invalid search log value");
-		return -1;
+	if (threads < 0) {
+		threads = cpu_count();
 	}
 
-	if (searchLength < ZSTD_SEARCHLENGTH_MIN || searchLength > ZSTD_SEARCHLENGTH_MAX) {
-		PyErr_SetString(PyExc_ValueError, "invalid search length value");
-		return -1;
-	}
-
-	if (targetLength < ZSTD_TARGETLENGTH_MIN || targetLength > ZSTD_TARGETLENGTH_MAX) {
-		PyErr_SetString(PyExc_ValueError, "invalid target length value");
-		return -1;
+	if (compressLiterals < 0) {
+		compressLiterals = compressionLevel >= 0;
 	}
 
-	if (strategy < ZSTD_fast || strategy > ZSTD_btopt) {
-		PyErr_SetString(PyExc_ValueError, "invalid strategy value");
-		return -1;
-	}
-
+	self->format = format;
+	self->compressionLevel = compressionLevel;
 	self->windowLog = windowLog;
+	self->hashLog = hashLog;
 	self->chainLog = chainLog;
-	self->hashLog = hashLog;
 	self->searchLog = searchLog;
-	self->searchLength = searchLength;
+	self->minMatch = minMatch;
 	self->targetLength = targetLength;
-	self->strategy = strategy;
+	self->compressionStrategy = compressionStrategy;
+	self->contentSizeFlag = contentSizeFlag;
+	self->checksumFlag = checksumFlag;
+	self->dictIDFlag = dictIDFlag;
+	self->threads = threads;
+	self->jobSize = jobSize;
+	self->overlapSizeLog = overlapSizeLog;
+	self->compressLiterals = compressLiterals;
+	self->forceMaxWindow = forceMaxWindow;
+	self->enableLongDistanceMatching = enableLDM;
+	self->ldmHashLog = ldmHashLog;
+	self->ldmMinMatch = ldmMinMatch;
+	self->ldmBucketSizeLog = ldmBucketSizeLog;
+	self->ldmHashEveryLog = ldmHashEveryLog;
 
-	ztopy_compression_parameters(self, &params);
-	zresult = ZSTD_checkCParams(params);
-
-	if (ZSTD_isError(zresult)) {
-		PyErr_Format(PyExc_ValueError, "invalid compression parameters: %s",
-			ZSTD_getErrorName(zresult));
+	if (reset_params(self)) {
 		return -1;
 	}
 
 	return 0;
 }
 
-PyDoc_STRVAR(CompressionParameters_estimated_compression_context_size__doc__,
+PyDoc_STRVAR(ZstdCompressionParameters_from_level__doc__,
+"Create a CompressionParameters from a compression level and target sizes\n"
+);
+
+ZstdCompressionParametersObject* CompressionParameters_from_level(PyObject* undef, PyObject* args, PyObject* kwargs) {
+	int managedKwargs = 0;
+	int level;
+	PyObject* sourceSize = NULL;
+	PyObject* dictSize = NULL;
+	unsigned PY_LONG_LONG iSourceSize = 0;
+	Py_ssize_t iDictSize = 0;
+	PyObject* val;
+	ZSTD_compressionParameters params;
+	ZstdCompressionParametersObject* result = NULL;
+	int res;
+
+	if (!PyArg_ParseTuple(args, "i:from_level",
+		&level)) {
+		return NULL;
+	}
+
+	if (!kwargs) {
+		kwargs = PyDict_New();
+		if (!kwargs) {
+			return NULL;
+		}
+		managedKwargs = 1;
+	}
+
+	sourceSize = PyDict_GetItemString(kwargs, "source_size");
+	if (sourceSize) {
+#if PY_MAJOR_VERSION >= 3
+		iSourceSize = PyLong_AsUnsignedLongLong(sourceSize);
+		if (iSourceSize == (unsigned PY_LONG_LONG)(-1)) {
+			goto cleanup;
+		}
+#else
+		iSourceSize = PyInt_AsUnsignedLongLongMask(sourceSize);
+#endif
+
+		PyDict_DelItemString(kwargs, "source_size");
+	}
+
+	dictSize = PyDict_GetItemString(kwargs, "dict_size");
+	if (dictSize) {
+#if PY_MAJOR_VERSION >= 3
+		iDictSize = PyLong_AsSsize_t(dictSize);
+#else
+		iDictSize = PyInt_AsSsize_t(dictSize);
+#endif
+		if (iDictSize == -1) {
+			goto cleanup;
+		}
+
+		PyDict_DelItemString(kwargs, "dict_size");
+	}
+
+
+	params = ZSTD_getCParams(level, iSourceSize, iDictSize);
+
+	/* Values derived from the input level and sizes are passed along to the
+	   constructor. But only if a value doesn't already exist. */
+	val = PyDict_GetItemString(kwargs, "window_log");
+	if (!val) {
+		val = PyLong_FromUnsignedLong(params.windowLog);
+		if (!val) {
+			goto cleanup;
+		}
+		PyDict_SetItemString(kwargs, "window_log", val);
+		Py_DECREF(val);
+	}
+
+	val = PyDict_GetItemString(kwargs, "chain_log");
+	if (!val) {
+		val = PyLong_FromUnsignedLong(params.chainLog);
+		if (!val) {
+			goto cleanup;
+		}
+		PyDict_SetItemString(kwargs, "chain_log", val);
+		Py_DECREF(val);
+	}
+
+	val = PyDict_GetItemString(kwargs, "hash_log");
+	if (!val) {
+		val = PyLong_FromUnsignedLong(params.hashLog);
+		if (!val) {
+			goto cleanup;
+		}
+		PyDict_SetItemString(kwargs, "hash_log", val);
+		Py_DECREF(val);
+	}
+
+	val = PyDict_GetItemString(kwargs, "search_log");
+	if (!val) {
+		val = PyLong_FromUnsignedLong(params.searchLog);
+		if (!val) {
+			goto cleanup;
+		}
+		PyDict_SetItemString(kwargs, "search_log", val);
+		Py_DECREF(val);
+	}
+
+	val = PyDict_GetItemString(kwargs, "min_match");
+	if (!val) {
+		val = PyLong_FromUnsignedLong(params.searchLength);
+		if (!val) {
+			goto cleanup;
+		}
+		PyDict_SetItemString(kwargs, "min_match", val);
+		Py_DECREF(val);
+	}
+
+	val = PyDict_GetItemString(kwargs, "target_length");
+	if (!val) {
+		val = PyLong_FromUnsignedLong(params.targetLength);
+		if (!val) {
+			goto cleanup;
+		}
+		PyDict_SetItemString(kwargs, "target_length", val);
+		Py_DECREF(val);
+	}
+
+	val = PyDict_GetItemString(kwargs, "compression_strategy");
+	if (!val) {
+		val = PyLong_FromUnsignedLong(params.strategy);
+		if (!val) {
+			goto cleanup;
+		}
+		PyDict_SetItemString(kwargs, "compression_strategy", val);
+		Py_DECREF(val);
+	}
+
+	val = PyDict_GetItemString(kwargs, "compress_literals");
+	if (!val) {
+		val = PyLong_FromLong(level >= 0 ? 1 : 0);
+		if (!val) {
+			goto cleanup;
+		}
+		PyDict_SetItemString(kwargs, "compress_literals", val);
+		Py_DECREF(val);
+	}
+
+	result = PyObject_New(ZstdCompressionParametersObject, &ZstdCompressionParametersType);
+	if (!result) {
+		goto cleanup;
+	}
+
+	result->params = NULL;
+
+	val = PyTuple_New(0);
+	if (!val) {
+		Py_CLEAR(result);
+		goto cleanup;
+	}
+
+	res = ZstdCompressionParameters_init(result, val, kwargs);
+	Py_DECREF(val);
+
+	if (res) {
+		Py_CLEAR(result);
+		goto cleanup;
+	}
+
+cleanup:
+	if (managedKwargs) {
+		Py_DECREF(kwargs);
+	}
+
+	return result;
+}
+
+PyDoc_STRVAR(ZstdCompressionParameters_estimated_compression_context_size__doc__,
 "Estimate the size in bytes of a compression context for compression parameters\n"
 );
 
-PyObject* CompressionParameters_estimated_compression_context_size(CompressionParametersObject* self) {
-	ZSTD_compressionParameters params;
-
-	ztopy_compression_parameters(self, &params);
-
-	return PyLong_FromSize_t(ZSTD_estimateCCtxSize(params));
+PyObject* ZstdCompressionParameters_estimated_compression_context_size(ZstdCompressionParametersObject* self) {
+	return PyLong_FromSize_t(ZSTD_estimateCCtxSize_usingCCtxParams(self->params));
 }
 
-PyObject* estimate_compression_context_size(PyObject* self, PyObject* args) {
-	CompressionParametersObject* params;
-	ZSTD_compressionParameters zparams;
-	PyObject* result;
+PyDoc_STRVAR(ZstdCompressionParameters__doc__,
+"ZstdCompressionParameters: low-level control over zstd compression");
 
-	if (!PyArg_ParseTuple(args, "O!:estimate_compression_context_size",
-		&CompressionParametersType, &params)) {
-		return NULL;
+static void ZstdCompressionParameters_dealloc(ZstdCompressionParametersObject* self) {
+	if (self->params) {
+		ZSTD_freeCCtxParams(self->params);
+		self->params = NULL;
 	}
 
-	ztopy_compression_parameters(params, &zparams);
-	result = PyLong_FromSize_t(ZSTD_estimateCCtxSize(zparams));
-	return result;
-}
-
-PyDoc_STRVAR(CompressionParameters__doc__,
-"CompressionParameters: low-level control over zstd compression");
-
-static void CompressionParameters_dealloc(PyObject* self) {
 	PyObject_Del(self);
 }
 
-static PyMethodDef CompressionParameters_methods[] = {
+static PyMethodDef ZstdCompressionParameters_methods[] = {
+	{
+		"from_level",
+		(PyCFunction)CompressionParameters_from_level,
+		METH_VARARGS | METH_KEYWORDS | METH_STATIC,
+		ZstdCompressionParameters_from_level__doc__
+	},
 	{
 		"estimated_compression_context_size",
-		(PyCFunction)CompressionParameters_estimated_compression_context_size,
+		(PyCFunction)ZstdCompressionParameters_estimated_compression_context_size,
 		METH_NOARGS,
-		CompressionParameters_estimated_compression_context_size__doc__
+		ZstdCompressionParameters_estimated_compression_context_size__doc__
 	},
 	{ NULL, NULL }
 };
 
-static PyMemberDef CompressionParameters_members[] = {
+static PyMemberDef ZstdCompressionParameters_members[] = {
+	{ "format", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, format), READONLY,
+	  "compression format" },
+	{ "compression_level", T_INT,
+	  offsetof(ZstdCompressionParametersObject, compressionLevel), READONLY,
+	  "compression level" },
 	{ "window_log", T_UINT,
-	  offsetof(CompressionParametersObject, windowLog), READONLY,
+	  offsetof(ZstdCompressionParametersObject, windowLog), READONLY,
 	  "window log" },
-	{ "chain_log", T_UINT,
-	  offsetof(CompressionParametersObject, chainLog), READONLY,
-	  "chain log" },
 	{ "hash_log", T_UINT,
-	  offsetof(CompressionParametersObject, hashLog), READONLY,
+	  offsetof(ZstdCompressionParametersObject, hashLog), READONLY,
 	  "hash log" },
+	{ "chain_log", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, chainLog), READONLY,
+	  "chain log" },
 	{ "search_log", T_UINT,
-	  offsetof(CompressionParametersObject, searchLog), READONLY,
+	  offsetof(ZstdCompressionParametersObject, searchLog), READONLY,
 	  "search log" },
-	{ "search_length", T_UINT,
-	  offsetof(CompressionParametersObject, searchLength), READONLY,
+	{ "min_match", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, minMatch), READONLY,
 	  "search length" },
 	{ "target_length", T_UINT,
-	  offsetof(CompressionParametersObject, targetLength), READONLY,
+	  offsetof(ZstdCompressionParametersObject, targetLength), READONLY,
 	  "target length" },
-	{ "strategy", T_INT,
-	  offsetof(CompressionParametersObject, strategy), READONLY,
-	  "strategy" },
+	{ "compression_strategy", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, compressionStrategy), READONLY,
+	  "compression strategy" },
+	{ "write_content_size", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, contentSizeFlag), READONLY,
+	  "whether to write content size in frames" },
+	{ "write_checksum", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, checksumFlag), READONLY,
+	  "whether to write checksum in frames" },
+	{ "write_dict_id", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, dictIDFlag), READONLY,
+	  "whether to write dictionary ID in frames" },
+	{ "threads", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, threads), READONLY,
+	  "number of threads to use" },
+	{ "job_size", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, jobSize), READONLY,
+	  "size of compression job when using multiple threads" },
+	{ "overlap_size_log", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, overlapSizeLog), READONLY,
+	  "Size of previous input reloaded at the beginning of each job" },
+	{ "compress_literals", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, compressLiterals), READONLY,
+	  "whether Huffman compression of literals is in use" },
+	{ "force_max_window", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, forceMaxWindow), READONLY,
+	  "force back references to remain smaller than window size" },
+	{ "enable_ldm", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, enableLongDistanceMatching), READONLY,
+	  "whether to enable long distance matching" },
+	{ "ldm_hash_log", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, ldmHashLog), READONLY,
+	  "Size of the table for long distance matching, as a power of 2" },
+	{ "ldm_min_match", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, ldmMinMatch), READONLY,
+	  "minimum size of searched matches for long distance matcher" },
+	{ "ldm_bucket_size_log", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, ldmBucketSizeLog), READONLY,
+	  "log size of each bucket in the LDM hash table for collision resolution" },
+	{ "ldm_hash_every_log", T_UINT,
+	  offsetof(ZstdCompressionParametersObject, ldmHashEveryLog), READONLY,
+	  "frequency of inserting/looking up entries in the LDM hash table" },
 	{ NULL }
 };
 
-PyTypeObject CompressionParametersType = {
+PyTypeObject ZstdCompressionParametersType = {
 	PyVarObject_HEAD_INIT(NULL, 0)
-	"CompressionParameters", /* tp_name */
-	sizeof(CompressionParametersObject), /* tp_basicsize */
+	"ZstdCompressionParameters", /* tp_name */
+	sizeof(ZstdCompressionParametersObject), /* tp_basicsize */
 	0,                         /* tp_itemsize */
-	(destructor)CompressionParameters_dealloc, /* tp_dealloc */
+	(destructor)ZstdCompressionParameters_dealloc, /* tp_dealloc */
 	0,                         /* tp_print */
 	0,                         /* tp_getattr */
 	0,                         /* tp_setattr */
@@ -221,33 +465,38 @@
 	0,                         /* tp_setattro */
 	0,                         /* tp_as_buffer */
 	Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE, /* tp_flags */
-	CompressionParameters__doc__, /* tp_doc */
+	ZstdCompressionParameters__doc__, /* tp_doc */
 	0,                         /* tp_traverse */
 	0,                         /* tp_clear */
 	0,                         /* tp_richcompare */
 	0,                         /* tp_weaklistoffset */
 	0,                         /* tp_iter */
 	0,                         /* tp_iternext */
-	CompressionParameters_methods, /* tp_methods */
-	CompressionParameters_members, /* tp_members */
+	ZstdCompressionParameters_methods, /* tp_methods */
+	ZstdCompressionParameters_members, /* tp_members */
 	0,                         /* tp_getset */
 	0,                         /* tp_base */
 	0,                         /* tp_dict */
 	0,                         /* tp_descr_get */
 	0,                         /* tp_descr_set */
 	0,                         /* tp_dictoffset */
-	(initproc)CompressionParameters_init, /* tp_init */
+	(initproc)ZstdCompressionParameters_init, /* tp_init */
 	0,                         /* tp_alloc */
 	PyType_GenericNew,         /* tp_new */
 };
 
 void compressionparams_module_init(PyObject* mod) {
-	Py_TYPE(&CompressionParametersType) = &PyType_Type;
-	if (PyType_Ready(&CompressionParametersType) < 0) {
+	Py_TYPE(&ZstdCompressionParametersType) = &PyType_Type;
+	if (PyType_Ready(&ZstdCompressionParametersType) < 0) {
 		return;
 	}
 
-	Py_INCREF(&CompressionParametersType);
+	Py_INCREF(&ZstdCompressionParametersType);
+	PyModule_AddObject(mod, "ZstdCompressionParameters",
+		(PyObject*)&ZstdCompressionParametersType);
+
+	/* TODO remove deprecated alias. */
+	Py_INCREF(&ZstdCompressionParametersType);
 	PyModule_AddObject(mod, "CompressionParameters",
-		(PyObject*)&CompressionParametersType);
+		(PyObject*)&ZstdCompressionParametersType);
 }
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/c-ext/compressionreader.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,405 @@
+/**
+* Copyright (c) 2017-present, Gregory Szorc
+* All rights reserved.
+*
+* This software may be modified and distributed under the terms
+* of the BSD license. See the LICENSE file for details.
+*/
+
+#include "python-zstandard.h"
+
+extern PyObject* ZstdError;
+
+static void set_unsupported_operation(void) {
+	PyObject* iomod;
+	PyObject* exc;
+
+	iomod = PyImport_ImportModule("io");
+	if (NULL == iomod) {
+		return;
+	}
+
+	exc = PyObject_GetAttrString(iomod, "UnsupportedOperation");
+	if (NULL == exc) {
+		Py_DECREF(iomod);
+		return;
+	}
+
+	PyErr_SetNone(exc);
+	Py_DECREF(exc);
+	Py_DECREF(iomod);
+}
+
+static void reader_dealloc(ZstdCompressionReader* self) {
+	Py_XDECREF(self->compressor);
+	Py_XDECREF(self->reader);
+
+	if (self->buffer.buf) {
+		PyBuffer_Release(&self->buffer);
+		memset(&self->buffer, 0, sizeof(self->buffer));
+	}
+
+	PyObject_Del(self);
+}
+
+static ZstdCompressionReader* reader_enter(ZstdCompressionReader* self) {
+	size_t zresult;
+
+	if (self->entered) {
+		PyErr_SetString(PyExc_ValueError, "cannot __enter__ multiple times");
+		return NULL;
+	}
+
+	zresult = ZSTD_CCtx_setPledgedSrcSize(self->compressor->cctx, self->sourceSize);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "error setting source size: %s",
+			ZSTD_getErrorName(zresult));
+		return NULL;
+	}
+
+	self->entered = 1;
+
+	Py_INCREF(self);
+	return self;
+}
+
+static PyObject* reader_exit(ZstdCompressionReader* self, PyObject* args) {
+	PyObject* exc_type;
+	PyObject* exc_value;
+	PyObject* exc_tb;
+
+	if (!PyArg_ParseTuple(args, "OOO:__exit__", &exc_type, &exc_value, &exc_tb)) {
+		return NULL;
+	}
+
+	self->entered = 0;
+	self->closed = 1;
+
+	/* Release resources associated with source. */
+	Py_CLEAR(self->reader);
+	if (self->buffer.buf) {
+		PyBuffer_Release(&self->buffer);
+		memset(&self->buffer, 0, sizeof(self->buffer));
+	}
+
+    Py_CLEAR(self->compressor);
+
+	Py_RETURN_FALSE;
+}
+
+static PyObject* reader_readable(ZstdCompressionReader* self) {
+	Py_RETURN_TRUE;
+}
+
+static PyObject* reader_writable(ZstdCompressionReader* self) {
+	Py_RETURN_FALSE;
+}
+
+static PyObject* reader_seekable(ZstdCompressionReader* self) {
+	Py_RETURN_FALSE;
+}
+
+static PyObject* reader_readline(PyObject* self, PyObject* args) {
+	set_unsupported_operation();
+	return NULL;
+}
+
+static PyObject* reader_readlines(PyObject* self, PyObject* args) {
+	set_unsupported_operation();
+	return NULL;
+}
+
+static PyObject* reader_write(PyObject* self, PyObject* args) {
+	PyErr_SetString(PyExc_OSError, "stream is not writable");
+	return NULL;
+}
+
+static PyObject* reader_writelines(PyObject* self, PyObject* args) {
+	PyErr_SetString(PyExc_OSError, "stream is not writable");
+	return NULL;
+}
+
+static PyObject* reader_isatty(PyObject* self) {
+	Py_RETURN_FALSE;
+}
+
+static PyObject* reader_flush(PyObject* self) {
+	Py_RETURN_NONE;
+}
+
+static PyObject* reader_close(ZstdCompressionReader* self) {
+	self->closed = 1;
+	Py_RETURN_NONE;
+}
+
+static PyObject* reader_closed(ZstdCompressionReader* self) {
+	if (self->closed) {
+		Py_RETURN_TRUE;
+	}
+	else {
+		Py_RETURN_FALSE;
+	}
+}
+
+static PyObject* reader_tell(ZstdCompressionReader* self) {
+	/* TODO should this raise OSError since stream isn't seekable? */
+	return PyLong_FromUnsignedLongLong(self->bytesCompressed);
+}
+
+static PyObject* reader_read(ZstdCompressionReader* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"size",
+		NULL
+	};
+
+	Py_ssize_t size = -1;
+	PyObject* result = NULL;
+	char* resultBuffer;
+	Py_ssize_t resultSize;
+	size_t zresult;
+	size_t oldPos;
+
+	if (!self->entered) {
+		PyErr_SetString(ZstdError, "read() must be called from an active context manager");
+		return NULL;
+	}
+
+	if (self->closed) {
+		PyErr_SetString(PyExc_ValueError, "stream is closed");
+		return NULL;
+	}
+
+	if (self->finishedOutput) {
+		return PyBytes_FromStringAndSize("", 0);
+	}
+
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "n", kwlist, &size)) {
+		return NULL;
+	}
+
+	if (size < 1) {
+		PyErr_SetString(PyExc_ValueError, "cannot read negative or size 0 amounts");
+		return NULL;
+	}
+
+	result = PyBytes_FromStringAndSize(NULL, size);
+	if (NULL == result) {
+		return NULL;
+	}
+
+	PyBytes_AsStringAndSize(result, &resultBuffer, &resultSize);
+
+	self->output.dst = resultBuffer;
+	self->output.size = resultSize;
+	self->output.pos = 0;
+
+readinput:
+
+	/* If we have data left over, consume it. */
+	if (self->input.pos < self->input.size) {
+		oldPos = self->output.pos;
+
+		Py_BEGIN_ALLOW_THREADS
+		zresult = ZSTD_compress_generic(self->compressor->cctx,
+			&self->output, &self->input, ZSTD_e_continue);
+
+		Py_END_ALLOW_THREADS
+
+		self->bytesCompressed += self->output.pos - oldPos;
+
+		/* Input exhausted. Clear out state tracking. */
+		if (self->input.pos == self->input.size) {
+			memset(&self->input, 0, sizeof(self->input));
+			Py_CLEAR(self->readResult);
+
+			if (self->buffer.buf) {
+				self->finishedInput = 1;
+			}
+		}
+
+		if (ZSTD_isError(zresult)) {
+			PyErr_Format(ZstdError, "zstd compress error: %s", ZSTD_getErrorName(zresult));
+			return NULL;
+		}
+
+		if (self->output.pos) {
+			/* If no more room in output, emit it. */
+			if (self->output.pos == self->output.size) {
+				memset(&self->output, 0, sizeof(self->output));
+				return result;
+			}
+
+			/*
+			 * There is room in the output. We fall through to below, which will either
+			 * get more input for us or will attempt to end the stream.
+			 */
+		}
+
+		/* Fall through to gather more input. */
+	}
+
+	if (!self->finishedInput) {
+		if (self->reader) {
+			Py_buffer buffer;
+
+			assert(self->readResult == NULL);
+			self->readResult = PyObject_CallMethod(self->reader, "read",
+				"k", self->readSize);
+			if (self->readResult == NULL) {
+				return NULL;
+			}
+
+			memset(&buffer, 0, sizeof(buffer));
+
+			if (0 != PyObject_GetBuffer(self->readResult, &buffer, PyBUF_CONTIG_RO)) {
+				return NULL;
+			}
+
+			/* EOF */
+			if (0 == buffer.len) {
+				self->finishedInput = 1;
+				Py_CLEAR(self->readResult);
+			}
+			else {
+				self->input.src = buffer.buf;
+				self->input.size = buffer.len;
+				self->input.pos = 0;
+			}
+
+			PyBuffer_Release(&buffer);
+		}
+		else {
+			assert(self->buffer.buf);
+
+			self->input.src = self->buffer.buf;
+			self->input.size = self->buffer.len;
+			self->input.pos = 0;
+		}
+	}
+
+	if (self->input.size) {
+		goto readinput;
+	}
+
+	/* Else EOF */
+	oldPos = self->output.pos;
+
+	zresult = ZSTD_compress_generic(self->compressor->cctx, &self->output,
+		&self->input, ZSTD_e_end);
+
+	self->bytesCompressed += self->output.pos - oldPos;
+
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "error ending compression stream: %s",
+			ZSTD_getErrorName(zresult));
+		return NULL;
+	}
+
+	assert(self->output.pos);
+
+	if (0 == zresult) {
+		self->finishedOutput = 1;
+	}
+
+	if (safe_pybytes_resize(&result, self->output.pos)) {
+		Py_XDECREF(result);
+		return NULL;
+	}
+
+	memset(&self->output, 0, sizeof(self->output));
+
+	return result;
+}
+
+static PyObject* reader_readall(PyObject* self) {
+	PyErr_SetNone(PyExc_NotImplementedError);
+	return NULL;
+}
+
+static PyObject* reader_iter(PyObject* self) {
+	set_unsupported_operation();
+	return NULL;
+}
+
+static PyObject* reader_iternext(PyObject* self) {
+	set_unsupported_operation();
+	return NULL;
+}
+
+static PyMethodDef reader_methods[] = {
+	{ "__enter__", (PyCFunction)reader_enter, METH_NOARGS,
+	PyDoc_STR("Enter a compression context") },
+	{ "__exit__", (PyCFunction)reader_exit, METH_VARARGS,
+	PyDoc_STR("Exit a compression context") },
+	{ "close", (PyCFunction)reader_close, METH_NOARGS,
+	PyDoc_STR("Close the stream so it cannot perform any more operations") },
+	{ "closed", (PyCFunction)reader_closed, METH_NOARGS,
+	PyDoc_STR("Whether stream is closed") },
+	{ "flush", (PyCFunction)reader_flush, METH_NOARGS, PyDoc_STR("no-ops") },
+	{ "isatty", (PyCFunction)reader_isatty, METH_NOARGS, PyDoc_STR("Returns False") },
+	{ "readable", (PyCFunction)reader_readable, METH_NOARGS,
+	PyDoc_STR("Returns True") },
+	{ "read", (PyCFunction)reader_read, METH_VARARGS | METH_KEYWORDS, PyDoc_STR("read compressed data") },
+	{ "readall", (PyCFunction)reader_readall, METH_NOARGS, PyDoc_STR("Not implemented") },
+	{ "readline", (PyCFunction)reader_readline, METH_VARARGS, PyDoc_STR("Not implemented") },
+	{ "readlines", (PyCFunction)reader_readlines, METH_VARARGS, PyDoc_STR("Not implemented") },
+	{ "seekable", (PyCFunction)reader_seekable, METH_NOARGS,
+	PyDoc_STR("Returns False") },
+	{ "tell", (PyCFunction)reader_tell, METH_NOARGS,
+	PyDoc_STR("Returns current number of bytes compressed") },
+	{ "writable", (PyCFunction)reader_writable, METH_NOARGS,
+	PyDoc_STR("Returns False") },
+	{ "write", reader_write, METH_VARARGS, PyDoc_STR("Raises OSError") },
+	{ "writelines", reader_writelines, METH_VARARGS, PyDoc_STR("Not implemented") },
+	{ NULL, NULL }
+};
+
+PyTypeObject ZstdCompressionReaderType = {
+	PyVarObject_HEAD_INIT(NULL, 0)
+	"zstd.ZstdCompressionReader", /* tp_name */
+	sizeof(ZstdCompressionReader), /* tp_basicsize */
+	0, /* tp_itemsize */
+	(destructor)reader_dealloc, /* tp_dealloc */
+	0, /* tp_print */
+	0, /* tp_getattr */
+	0, /* tp_setattr */
+	0, /* tp_compare */
+	0, /* tp_repr */
+	0, /* tp_as_number */
+	0, /* tp_as_sequence */
+	0, /* tp_as_mapping */
+	0, /* tp_hash */
+	0, /* tp_call */
+	0, /* tp_str */
+	0, /* tp_getattro */
+	0, /* tp_setattro */
+	0, /* tp_as_buffer */
+	Py_TPFLAGS_DEFAULT, /* tp_flags */
+	0, /* tp_doc */
+	0, /* tp_traverse */
+	0, /* tp_clear */
+	0, /* tp_richcompare */
+	0, /* tp_weaklistoffset */
+	reader_iter, /* tp_iter */
+	reader_iternext, /* tp_iternext */
+	reader_methods, /* tp_methods */
+	0, /* tp_members */
+	0, /* tp_getset */
+	0, /* tp_base */
+	0, /* tp_dict */
+	0, /* tp_descr_get */
+	0, /* tp_descr_set */
+	0, /* tp_dictoffset */
+	0, /* tp_init */
+	0, /* tp_alloc */
+	PyType_GenericNew, /* tp_new */
+};
+
+void compressionreader_module_init(PyObject* mod) {
+	/* TODO make reader a sub-class of io.RawIOBase */
+
+	Py_TYPE(&ZstdCompressionReaderType) = &PyType_Type;
+	if (PyType_Ready(&ZstdCompressionReaderType) < 0) {
+		return;
+	}
+}
--- a/contrib/python-zstandard/c-ext/compressionwriter.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/compressionwriter.c	Mon Apr 09 10:13:29 2018 -0700
@@ -22,20 +22,18 @@
 }
 
 static PyObject* ZstdCompressionWriter_enter(ZstdCompressionWriter* self) {
+	size_t zresult;
+
 	if (self->entered) {
 		PyErr_SetString(ZstdError, "cannot __enter__ multiple times");
 		return NULL;
 	}
 
-	if (self->compressor->mtcctx) {
-		if (init_mtcstream(self->compressor, self->sourceSize)) {
-			return NULL;
-		}
-	}
-	else {
-		if (0 != init_cstream(self->compressor, self->sourceSize)) {
-			return NULL;
-		}
+	zresult = ZSTD_CCtx_setPledgedSrcSize(self->compressor->cctx, self->sourceSize);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "error setting source size: %s",
+			ZSTD_getErrorName(zresult));
+		return NULL;
 	}
 
 	self->entered = 1;
@@ -59,8 +57,12 @@
 
 	self->entered = 0;
 
-	if ((self->compressor->cstream || self->compressor->mtcctx) && exc_type == Py_None
-		&& exc_value == Py_None && exc_tb == Py_None) {
+	if (exc_type == Py_None && exc_value == Py_None && exc_tb == Py_None) {
+		ZSTD_inBuffer inBuffer;
+
+		inBuffer.src = NULL;
+		inBuffer.size = 0;
+		inBuffer.pos = 0;
 
 		output.dst = PyMem_Malloc(self->outSize);
 		if (!output.dst) {
@@ -70,12 +72,7 @@
 		output.pos = 0;
 
 		while (1) {
-			if (self->compressor->mtcctx) {
-				zresult = ZSTDMT_endStream(self->compressor->mtcctx, &output);
-			}
-			else {
-				zresult = ZSTD_endStream(self->compressor->cstream, &output);
-			}
+			zresult = ZSTD_compress_generic(self->compressor->cctx, &output, &inBuffer, ZSTD_e_end);
 			if (ZSTD_isError(zresult)) {
 				PyErr_Format(ZstdError, "error ending compression stream: %s",
 					ZSTD_getErrorName(zresult));
@@ -107,18 +104,17 @@
 }
 
 static PyObject* ZstdCompressionWriter_memory_size(ZstdCompressionWriter* self) {
-	if (!self->compressor->cstream) {
-		PyErr_SetString(ZstdError, "cannot determine size of an inactive compressor; "
-			"call when a context manager is active");
-		return NULL;
-	}
-
-	return PyLong_FromSize_t(ZSTD_sizeof_CStream(self->compressor->cstream));
+	return PyLong_FromSize_t(ZSTD_sizeof_CCtx(self->compressor->cctx));
 }
 
-static PyObject* ZstdCompressionWriter_write(ZstdCompressionWriter* self, PyObject* args) {
-	const char* source;
-	Py_ssize_t sourceSize;
+static PyObject* ZstdCompressionWriter_write(ZstdCompressionWriter* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"data",
+		NULL
+	};
+
+	PyObject* result = NULL;
+	Py_buffer source;
 	size_t zresult;
 	ZSTD_inBuffer input;
 	ZSTD_outBuffer output;
@@ -126,44 +122,46 @@
 	Py_ssize_t totalWrite = 0;
 
 #if PY_MAJOR_VERSION >= 3
-	if (!PyArg_ParseTuple(args, "y#:write", &source, &sourceSize)) {
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*:write",
 #else
-	if (!PyArg_ParseTuple(args, "s#:write", &source, &sourceSize)) {
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*:write",
 #endif
+		kwlist, &source)) {
 		return NULL;
 	}
 
 	if (!self->entered) {
 		PyErr_SetString(ZstdError, "compress must be called from an active context manager");
-		return NULL;
+		goto finally;
+	}
+
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
 	}
 
 	output.dst = PyMem_Malloc(self->outSize);
 	if (!output.dst) {
-		return PyErr_NoMemory();
+		PyErr_NoMemory();
+		goto finally;
 	}
 	output.size = self->outSize;
 	output.pos = 0;
 
-	input.src = source;
-	input.size = sourceSize;
+	input.src = source.buf;
+	input.size = source.len;
 	input.pos = 0;
 
-	while ((ssize_t)input.pos < sourceSize) {
+	while ((ssize_t)input.pos < source.len) {
 		Py_BEGIN_ALLOW_THREADS
-		if (self->compressor->mtcctx) {
-			zresult = ZSTDMT_compressStream(self->compressor->mtcctx,
-				&output, &input);
-		}
-		else {
-			zresult = ZSTD_compressStream(self->compressor->cstream, &output, &input);
-		}
+		zresult = ZSTD_compress_generic(self->compressor->cctx, &output, &input, ZSTD_e_continue);
 		Py_END_ALLOW_THREADS
 
 		if (ZSTD_isError(zresult)) {
 			PyMem_Free(output.dst);
 			PyErr_Format(ZstdError, "zstd compress error: %s", ZSTD_getErrorName(zresult));
-			return NULL;
+			goto finally;
 		}
 
 		/* Copy data from output buffer to writer. */
@@ -176,18 +174,24 @@
 				output.dst, output.pos);
 			Py_XDECREF(res);
 			totalWrite += output.pos;
+			self->bytesCompressed += output.pos;
 		}
 		output.pos = 0;
 	}
 
 	PyMem_Free(output.dst);
 
-	return PyLong_FromSsize_t(totalWrite);
+	result = PyLong_FromSsize_t(totalWrite);
+
+finally:
+	PyBuffer_Release(&source);
+	return result;
 }
 
 static PyObject* ZstdCompressionWriter_flush(ZstdCompressionWriter* self, PyObject* args) {
 	size_t zresult;
 	ZSTD_outBuffer output;
+	ZSTD_inBuffer input;
 	PyObject* res;
 	Py_ssize_t totalWrite = 0;
 
@@ -196,6 +200,10 @@
 		return NULL;
 	}
 
+	input.src = NULL;
+	input.size = 0;
+	input.pos = 0;
+
 	output.dst = PyMem_Malloc(self->outSize);
 	if (!output.dst) {
 		return PyErr_NoMemory();
@@ -205,12 +213,7 @@
 
 	while (1) {
 		Py_BEGIN_ALLOW_THREADS
-		if (self->compressor->mtcctx) {
-			zresult = ZSTDMT_flushStream(self->compressor->mtcctx, &output);
-		}
-		else {
-			zresult = ZSTD_flushStream(self->compressor->cstream, &output);
-		}
+		zresult = ZSTD_compress_generic(self->compressor->cctx, &output, &input, ZSTD_e_flush);
 		Py_END_ALLOW_THREADS
 
 		if (ZSTD_isError(zresult)) {
@@ -233,6 +236,7 @@
 				output.dst, output.pos);
 			Py_XDECREF(res);
 			totalWrite += output.pos;
+			self->bytesCompressed += output.pos;
 		}
 		output.pos = 0;
 	}
@@ -242,6 +246,10 @@
 	return PyLong_FromSsize_t(totalWrite);
 }
 
+static PyObject* ZstdCompressionWriter_tell(ZstdCompressionWriter* self) {
+	return PyLong_FromUnsignedLongLong(self->bytesCompressed);
+}
+
 static PyMethodDef ZstdCompressionWriter_methods[] = {
 	{ "__enter__", (PyCFunction)ZstdCompressionWriter_enter, METH_NOARGS,
 	PyDoc_STR("Enter a compression context.") },
@@ -249,10 +257,12 @@
 	PyDoc_STR("Exit a compression context.") },
 	{ "memory_size", (PyCFunction)ZstdCompressionWriter_memory_size, METH_NOARGS,
 	PyDoc_STR("Obtain the memory size of the underlying compressor") },
-	{ "write", (PyCFunction)ZstdCompressionWriter_write, METH_VARARGS,
+	{ "write", (PyCFunction)ZstdCompressionWriter_write, METH_VARARGS | METH_KEYWORDS,
 	PyDoc_STR("Compress data") },
 	{ "flush", (PyCFunction)ZstdCompressionWriter_flush, METH_NOARGS,
 	PyDoc_STR("Flush data and finish a zstd frame") },
+	{ "tell", (PyCFunction)ZstdCompressionWriter_tell, METH_NOARGS,
+	PyDoc_STR("Returns current number of bytes compressed") },
 	{ NULL, NULL }
 };
 
--- a/contrib/python-zstandard/c-ext/compressobj.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/compressobj.c	Mon Apr 09 10:13:29 2018 -0700
@@ -23,9 +23,13 @@
 	PyObject_Del(self);
 }
 
-static PyObject* ZstdCompressionObj_compress(ZstdCompressionObj* self, PyObject* args) {
-	const char* source;
-	Py_ssize_t sourceSize;
+static PyObject* ZstdCompressionObj_compress(ZstdCompressionObj* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"data",
+		NULL
+	};
+
+	Py_buffer source;
 	ZSTD_inBuffer input;
 	size_t zresult;
 	PyObject* result = NULL;
@@ -37,38 +41,43 @@
 	}
 
 #if PY_MAJOR_VERSION >= 3
-	if (!PyArg_ParseTuple(args, "y#:compress", &source, &sourceSize)) {
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*:compress",
 #else
-	if (!PyArg_ParseTuple(args, "s#:compress", &source, &sourceSize)) {
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*:compress",
 #endif
+		kwlist, &source)) {
 		return NULL;
 	}
 
-	input.src = source;
-	input.size = sourceSize;
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
+	}
+
+	input.src = source.buf;
+	input.size = source.len;
 	input.pos = 0;
 
-	while ((ssize_t)input.pos < sourceSize) {
+	while ((ssize_t)input.pos < source.len) {
 		Py_BEGIN_ALLOW_THREADS
-		if (self->compressor->mtcctx) {
-			zresult = ZSTDMT_compressStream(self->compressor->mtcctx,
-				&self->output, &input);
-		}
-		else {
-			zresult = ZSTD_compressStream(self->compressor->cstream, &self->output, &input);
-		}
+			zresult = ZSTD_compress_generic(self->compressor->cctx, &self->output,
+				&input, ZSTD_e_continue);
 		Py_END_ALLOW_THREADS
 
 		if (ZSTD_isError(zresult)) {
 			PyErr_Format(ZstdError, "zstd compress error: %s", ZSTD_getErrorName(zresult));
-			return NULL;
+			Py_CLEAR(result);
+			goto finally;
 		}
 
 		if (self->output.pos) {
 			if (result) {
 				resultSize = PyBytes_GET_SIZE(result);
-				if (-1 == _PyBytes_Resize(&result, resultSize + self->output.pos)) {
-					return NULL;
+
+				if (safe_pybytes_resize(&result, resultSize + self->output.pos)) {
+					Py_CLEAR(result);
+					goto finally;
 				}
 
 				memcpy(PyBytes_AS_STRING(result) + resultSize,
@@ -77,7 +86,7 @@
 			else {
 				result = PyBytes_FromStringAndSize(self->output.dst, self->output.pos);
 				if (!result) {
-					return NULL;
+					goto finally;
 				}
 			}
 
@@ -85,21 +94,29 @@
 		}
 	}
 
-	if (result) {
-		return result;
+	if (NULL == result) {
+		result = PyBytes_FromString("");
 	}
-	else {
-		return PyBytes_FromString("");
-	}
+
+finally:
+	PyBuffer_Release(&source);
+
+	return result;
 }
 
-static PyObject* ZstdCompressionObj_flush(ZstdCompressionObj* self, PyObject* args) {
+static PyObject* ZstdCompressionObj_flush(ZstdCompressionObj* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"flush_mode",
+		NULL
+	};
+
 	int flushMode = compressorobj_flush_finish;
 	size_t zresult;
 	PyObject* result = NULL;
 	Py_ssize_t resultSize = 0;
+	ZSTD_inBuffer input;
 
-	if (!PyArg_ParseTuple(args, "|i:flush", &flushMode)) {
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|i:flush", kwlist, &flushMode)) {
 		return NULL;
 	}
 
@@ -115,16 +132,16 @@
 
 	assert(self->output.pos == 0);
 
+	input.src = NULL;
+	input.size = 0;
+	input.pos = 0;
+
 	if (flushMode == compressorobj_flush_block) {
 		/* The output buffer is of size ZSTD_CStreamOutSize(), which is 
 		   guaranteed to hold a full block. */
 		Py_BEGIN_ALLOW_THREADS
-		if (self->compressor->mtcctx) {
-			zresult = ZSTDMT_flushStream(self->compressor->mtcctx, &self->output);
-		}
-		else {
-			zresult = ZSTD_flushStream(self->compressor->cstream, &self->output);
-		}
+			zresult = ZSTD_compress_generic(self->compressor->cctx, &self->output,
+				&input, ZSTD_e_flush);
 		Py_END_ALLOW_THREADS
 
 		if (ZSTD_isError(zresult)) {
@@ -156,12 +173,8 @@
 	self->finished = 1;
 
 	while (1) {
-		if (self->compressor->mtcctx) {
-			zresult = ZSTDMT_endStream(self->compressor->mtcctx, &self->output);
-		}
-		else {
-			zresult = ZSTD_endStream(self->compressor->cstream, &self->output);
-		}
+		zresult = ZSTD_compress_generic(self->compressor->cctx, &self->output,
+			&input, ZSTD_e_end);
 		if (ZSTD_isError(zresult)) {
 			PyErr_Format(ZstdError, "error ending compression stream: %s",
 				ZSTD_getErrorName(zresult));
@@ -171,7 +184,9 @@
 		if (self->output.pos) {
 			if (result) {
 				resultSize = PyBytes_GET_SIZE(result);
-				if (-1 == _PyBytes_Resize(&result, resultSize + self->output.pos)) {
+
+				if (safe_pybytes_resize(&result, resultSize + self->output.pos)) {
+					Py_XDECREF(result);
 					return NULL;
 				}
 
@@ -202,9 +217,9 @@
 }
 
 static PyMethodDef ZstdCompressionObj_methods[] = {
-	{ "compress", (PyCFunction)ZstdCompressionObj_compress, METH_VARARGS,
+	{ "compress", (PyCFunction)ZstdCompressionObj_compress, METH_VARARGS | METH_KEYWORDS,
 	PyDoc_STR("compress data") },
-	{ "flush", (PyCFunction)ZstdCompressionObj_flush, METH_VARARGS,
+	{ "flush", (PyCFunction)ZstdCompressionObj_flush, METH_VARARGS | METH_KEYWORDS,
 	PyDoc_STR("finish compression operation") },
 	{ NULL, NULL }
 };
--- a/contrib/python-zstandard/c-ext/compressor.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/compressor.c	Mon Apr 09 10:13:29 2018 -0700
@@ -11,118 +11,78 @@
 
 extern PyObject* ZstdError;
 
-int populate_cdict(ZstdCompressor* compressor, ZSTD_parameters* zparams) {
-	ZSTD_customMem zmem;
+int ensure_cctx(ZstdCompressor* compressor) {
+	size_t zresult;
+
+	assert(compressor);
+	assert(compressor->cctx);
+	assert(compressor->params);
 
-	if (compressor->cdict || !compressor->dict || !compressor->dict->dictData) {
-		return 0;
+	ZSTD_CCtx_reset(compressor->cctx);
+
+	zresult = ZSTD_CCtx_setParametersUsingCCtxParams(compressor->cctx, compressor->params);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "could not set compression parameters: %s",
+			ZSTD_getErrorName(zresult));
+		return 1;
 	}
 
-	Py_BEGIN_ALLOW_THREADS
-	memset(&zmem, 0, sizeof(zmem));
-	compressor->cdict = ZSTD_createCDict_advanced(compressor->dict->dictData,
-		compressor->dict->dictSize, 1, *zparams, zmem);
-	Py_END_ALLOW_THREADS
-
-	if (!compressor->cdict) {
-		PyErr_SetString(ZstdError, "could not create compression dictionary");
-		return 1;
+	if (compressor->dict) {
+		if (compressor->dict->cdict) {
+			zresult = ZSTD_CCtx_refCDict(compressor->cctx, compressor->dict->cdict);
+		}
+		else {
+			zresult = ZSTD_CCtx_loadDictionary_advanced(compressor->cctx,
+				compressor->dict->dictData, compressor->dict->dictSize,
+				ZSTD_dlm_byRef, compressor->dict->dictType);
+		}
+		if (ZSTD_isError(zresult)) {
+			PyErr_Format(ZstdError, "could not load compression dictionary: %s",
+				ZSTD_getErrorName(zresult));
+			return 1;
+		}
 	}
 
 	return 0;
 }
 
-/**
- * Ensure the ZSTD_CStream on a ZstdCompressor instance is initialized.
- *
- * Returns 0 on success. Other value on failure. Will set a Python exception
- * on failure.
- */
-int init_cstream(ZstdCompressor* compressor, unsigned long long sourceSize) {
-	ZSTD_parameters zparams;
-	void* dictData = NULL;
-	size_t dictSize = 0;
-	size_t zresult;
+static PyObject* frame_progression(ZSTD_CCtx* cctx) {
+	PyObject* result = NULL;
+	PyObject* value;
+	ZSTD_frameProgression progression;
 
-	if (compressor->cstream) {
-		zresult = ZSTD_resetCStream(compressor->cstream, sourceSize);
-		if (ZSTD_isError(zresult)) {
-			PyErr_Format(ZstdError, "could not reset CStream: %s",
-				ZSTD_getErrorName(zresult));
-			return -1;
-		}
-
-		return 0;
+	result = PyTuple_New(3);
+	if (!result) {
+		return NULL;
 	}
 
-	compressor->cstream = ZSTD_createCStream();
-	if (!compressor->cstream) {
-		PyErr_SetString(ZstdError, "could not create CStream");
-		return -1;
-	}
+	progression = ZSTD_getFrameProgression(cctx);
 
-	if (compressor->dict) {
-		dictData = compressor->dict->dictData;
-		dictSize = compressor->dict->dictSize;
-	}
-
-	memset(&zparams, 0, sizeof(zparams));
-	if (compressor->cparams) {
-		ztopy_compression_parameters(compressor->cparams, &zparams.cParams);
-		/* Do NOT call ZSTD_adjustCParams() here because the compression params
-		come from the user. */
-	}
-	else {
-		zparams.cParams = ZSTD_getCParams(compressor->compressionLevel, sourceSize, dictSize);
+	value = PyLong_FromUnsignedLongLong(progression.ingested);
+	if (!value) {
+		Py_DECREF(result);
+		return NULL;
 	}
 
-	zparams.fParams = compressor->fparams;
-
-	zresult = ZSTD_initCStream_advanced(compressor->cstream, dictData, dictSize,
-		zparams, sourceSize);
+	PyTuple_SET_ITEM(result, 0, value);
 
-	if (ZSTD_isError(zresult)) {
-		ZSTD_freeCStream(compressor->cstream);
-		compressor->cstream = NULL;
-		PyErr_Format(ZstdError, "cannot init CStream: %s", ZSTD_getErrorName(zresult));
-		return -1;
+	value = PyLong_FromUnsignedLongLong(progression.consumed);
+	if (!value) {
+		Py_DECREF(result);
+		return NULL;
 	}
 
-	return 0;;
-}
+	PyTuple_SET_ITEM(result, 1, value);
 
-int init_mtcstream(ZstdCompressor* compressor, Py_ssize_t sourceSize) {
-	size_t zresult;
-	void* dictData = NULL;
-	size_t dictSize = 0;
-	ZSTD_parameters zparams;
-
-	assert(compressor->mtcctx);
-
-	if (compressor->dict) {
-		dictData = compressor->dict->dictData;
-		dictSize = compressor->dict->dictSize;
+	value = PyLong_FromUnsignedLongLong(progression.produced);
+	if (!value) {
+		Py_DECREF(result);
+		return NULL;
 	}
 
-	memset(&zparams, 0, sizeof(zparams));
-	if (compressor->cparams) {
-		ztopy_compression_parameters(compressor->cparams, &zparams.cParams);
-	}
-	else {
-		zparams.cParams = ZSTD_getCParams(compressor->compressionLevel, sourceSize, dictSize);
-	}
-
-	zparams.fParams = compressor->fparams;
+	PyTuple_SET_ITEM(result, 2, value);
 
-	zresult = ZSTDMT_initCStream_advanced(compressor->mtcctx, dictData, dictSize,
-		zparams, sourceSize);
-
-	if (ZSTD_isError(zresult)) {
-		PyErr_Format(ZstdError, "cannot init CStream: %s", ZSTD_getErrorName(zresult));
-		return -1;
-	}
-
-	return 0;
+	return result;
 }
 
 PyDoc_STRVAR(ZstdCompressor__doc__,
@@ -147,9 +107,9 @@
 "   If True, a 4 byte content checksum will be written with the compressed\n"
 "   data, allowing the decompressor to perform content verification.\n"
 "write_content_size\n"
-"   If True, the decompressed content size will be included in the header of\n"
-"   the compressed data. This data will only be written if the compressor\n"
-"   knows the size of the input data.\n"
+"   If True (the default), the decompressed content size will be included in\n"
+"   the header of the compressed data. This data will only be written if the\n"
+"   compressor knows the size of the input data.\n"
 "write_dict_id\n"
 "   Determines whether the dictionary ID will be written into the compressed\n"
 "   data. Defaults to True. Only adds content to the compressed data if\n"
@@ -175,7 +135,7 @@
 
 	int level = 3;
 	ZstdCompressionDict* dict = NULL;
-	CompressionParametersObject* params = NULL;
+	ZstdCompressionParametersObject* params = NULL;
 	PyObject* writeChecksum = NULL;
 	PyObject* writeContentSize = NULL;
 	PyObject* writeDictID = NULL;
@@ -183,16 +143,11 @@
 
 	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|iO!O!OOOi:ZstdCompressor",
 		kwlist,	&level, &ZstdCompressionDictType, &dict,
-		&CompressionParametersType, &params,
+		&ZstdCompressionParametersType, &params,
 		&writeChecksum, &writeContentSize, &writeDictID, &threads)) {
 		return -1;
 	}
 
-	if (level < 1) {
-		PyErr_SetString(PyExc_ValueError, "level must be greater than 0");
-		return -1;
-	}
-
 	if (level > ZSTD_maxCLevel()) {
 		PyErr_Format(PyExc_ValueError, "level must be less than %d",
 			ZSTD_maxCLevel() + 1);
@@ -203,79 +158,135 @@
 		threads = cpu_count();
 	}
 
-	self->threads = threads;
-
 	/* We create a ZSTD_CCtx for reuse among multiple operations to reduce the
 	   overhead of each compression operation. */
-	if (threads) {
-		self->mtcctx = ZSTDMT_createCCtx(threads);
-		if (!self->mtcctx) {
-			PyErr_NoMemory();
+	self->cctx = ZSTD_createCCtx();
+	if (!self->cctx) {
+		PyErr_NoMemory();
+		return -1;
+	}
+
+	/* TODO stuff the original parameters away somewhere so we can reset later. This
+	   will allow us to do things like automatically adjust cparams based on input
+	   size (assuming zstd isn't doing that internally). */
+
+	self->params = ZSTD_createCCtxParams();
+	if (!self->params) {
+		PyErr_NoMemory();
+		return -1;
+	}
+
+	if (params && writeChecksum) {
+		PyErr_SetString(PyExc_ValueError,
+			"cannot define compression_params and write_checksum");
+		return -1;
+	}
+
+	if (params && writeContentSize) {
+		PyErr_SetString(PyExc_ValueError,
+			"cannot define compression_params and write_content_size");
+		return -1;
+	}
+
+	if (params && writeDictID) {
+		PyErr_SetString(PyExc_ValueError,
+			"cannot define compression_params and write_dict_id");
+		return -1;
+	}
+
+	if (params && threads) {
+		PyErr_SetString(PyExc_ValueError,
+			"cannot define compression_params and threads");
+		return -1;
+	}
+
+	if (params) {
+		if (set_parameters(self->params, params)) {
 			return -1;
 		}
 	}
 	else {
-		self->cctx = ZSTD_createCCtx();
-		if (!self->cctx) {
-			PyErr_NoMemory();
+		if (set_parameter(self->params, ZSTD_p_compressionLevel, level)) {
+			return -1;
+		}
+
+		if (set_parameter(self->params, ZSTD_p_contentSizeFlag,
+			writeContentSize ? PyObject_IsTrue(writeContentSize) : 1)) {
+			return -1;
+		}
+
+		if (set_parameter(self->params, ZSTD_p_checksumFlag,
+			writeChecksum ? PyObject_IsTrue(writeChecksum) : 0)) {
 			return -1;
 		}
-	}
+
+		if (set_parameter(self->params, ZSTD_p_dictIDFlag,
+			writeDictID ? PyObject_IsTrue(writeDictID) : 1)) {
+			return -1;
+		}
 
-	self->compressionLevel = level;
+		if (threads) {
+			if (set_parameter(self->params, ZSTD_p_nbWorkers, threads)) {
+				return -1;
+			}
+		}
+	}
 
 	if (dict) {
 		self->dict = dict;
 		Py_INCREF(dict);
 	}
 
-	if (params) {
-		self->cparams = params;
-		Py_INCREF(params);
-	}
-
-	memset(&self->fparams, 0, sizeof(self->fparams));
-
-	if (writeChecksum && PyObject_IsTrue(writeChecksum)) {
-		self->fparams.checksumFlag = 1;
-	}
-	if (writeContentSize && PyObject_IsTrue(writeContentSize)) {
-		self->fparams.contentSizeFlag = 1;
-	}
-	if (writeDictID && PyObject_Not(writeDictID)) {
-		self->fparams.noDictIDFlag = 1;
+	if (ensure_cctx(self)) {
+		return -1;
 	}
 
 	return 0;
 }
 
 static void ZstdCompressor_dealloc(ZstdCompressor* self) {
-	if (self->cstream) {
-		ZSTD_freeCStream(self->cstream);
-		self->cstream = NULL;
-	}
-
-	Py_XDECREF(self->cparams);
-	Py_XDECREF(self->dict);
-
-	if (self->cdict) {
-		ZSTD_freeCDict(self->cdict);
-		self->cdict = NULL;
-	}
-
 	if (self->cctx) {
 		ZSTD_freeCCtx(self->cctx);
 		self->cctx = NULL;
 	}
 
-	if (self->mtcctx) {
-		ZSTDMT_freeCCtx(self->mtcctx);
-		self->mtcctx = NULL;
+	if (self->params) {
+		ZSTD_freeCCtxParams(self->params);
+		self->params = NULL;
 	}
 
+	Py_XDECREF(self->dict);
 	PyObject_Del(self);
 }
 
+PyDoc_STRVAR(ZstdCompressor_memory_size__doc__,
+"memory_size()\n"
+"\n"
+"Obtain the memory usage of this compressor, in bytes.\n"
+);
+
+static PyObject* ZstdCompressor_memory_size(ZstdCompressor* self) {
+	if (self->cctx) {
+		return PyLong_FromSize_t(ZSTD_sizeof_CCtx(self->cctx));
+	}
+	else {
+		PyErr_SetString(ZstdError, "no compressor context found; this should never happen");
+		return NULL;
+	}
+}
+
+PyDoc_STRVAR(ZstdCompressor_frame_progression__doc__,
+"frame_progression()\n"
+"\n"
+"Return information on how much work the compressor has done.\n"
+"\n"
+"Returns a 3-tuple of (ingested, consumed, produced).\n"
+);
+
+static PyObject* ZstdCompressor_frame_progression(ZstdCompressor* self) {
+	return frame_progression(self->cctx);
+}
+
 PyDoc_STRVAR(ZstdCompressor_copy_stream__doc__,
 "copy_stream(ifh, ofh[, size=0, read_size=default, write_size=default])\n"
 "compress data between streams\n"
@@ -304,7 +315,7 @@
 
 	PyObject* source;
 	PyObject* dest;
-	Py_ssize_t sourceSize = 0;
+	unsigned long long sourceSize = ZSTD_CONTENTSIZE_UNKNOWN;
 	size_t inSize = ZSTD_CStreamInSize();
 	size_t outSize = ZSTD_CStreamOutSize();
 	ZSTD_inBuffer input;
@@ -313,14 +324,14 @@
 	Py_ssize_t totalWrite = 0;
 	char* readBuffer;
 	Py_ssize_t readSize;
-	PyObject* readResult;
+	PyObject* readResult = NULL;
 	PyObject* res = NULL;
 	size_t zresult;
 	PyObject* writeResult;
 	PyObject* totalReadPy;
 	PyObject* totalWritePy;
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "OO|nkk:copy_stream", kwlist,
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "OO|Kkk:copy_stream", kwlist,
 		&source, &dest, &sourceSize, &inSize, &outSize)) {
 		return NULL;
 	}
@@ -335,22 +346,18 @@
 		return NULL;
 	}
 
-	/* Prevent free on uninitialized memory in finally. */
-	output.dst = NULL;
-
-	if (self->mtcctx) {
-		if (init_mtcstream(self, sourceSize)) {
-			res = NULL;
-			goto finally;
-		}
-	}
-	else {
-		if (0 != init_cstream(self, sourceSize)) {
-			res = NULL;
-			goto finally;
-		}
+	if (ensure_cctx(self)) {
+		return NULL;
 	}
 
+	zresult = ZSTD_CCtx_setPledgedSrcSize(self->cctx, sourceSize);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "error setting source size: %s",
+			ZSTD_getErrorName(zresult));
+		return NULL;
+	}
+
+	/* Prevent free on uninitialized memory in finally. */
 	output.dst = PyMem_Malloc(outSize);
 	if (!output.dst) {
 		PyErr_NoMemory();
@@ -360,6 +367,10 @@
 	output.size = outSize;
 	output.pos = 0;
 
+	input.src = NULL;
+	input.size = 0;
+	input.pos = 0;
+
 	while (1) {
 		/* Try to read from source stream. */
 		readResult = PyObject_CallMethod(source, "read", "n", inSize);
@@ -384,12 +395,7 @@
 
 		while (input.pos < input.size) {
 			Py_BEGIN_ALLOW_THREADS
-			if (self->mtcctx) {
-				zresult = ZSTDMT_compressStream(self->mtcctx, &output, &input);
-			}
-			else {
-				zresult = ZSTD_compressStream(self->cstream, &output, &input);
-			}
+			zresult = ZSTD_compress_generic(self->cctx, &output, &input, ZSTD_e_continue);
 			Py_END_ALLOW_THREADS
 
 			if (ZSTD_isError(zresult)) {
@@ -410,16 +416,18 @@
 				output.pos = 0;
 			}
 		}
+
+		Py_CLEAR(readResult);
 	}
 
 	/* We've finished reading. Now flush the compressor stream. */
+	assert(input.pos == input.size);
+
 	while (1) {
-		if (self->mtcctx) {
-			zresult = ZSTDMT_endStream(self->mtcctx, &output);
-		}
-		else {
-			zresult = ZSTD_endStream(self->cstream, &output);
-		}
+		Py_BEGIN_ALLOW_THREADS
+		zresult = ZSTD_compress_generic(self->cctx, &output, &input, ZSTD_e_end);
+		Py_END_ALLOW_THREADS
+
 		if (ZSTD_isError(zresult)) {
 			PyErr_Format(ZstdError, "error ending compression stream: %s",
 				ZSTD_getErrorName(zresult));
@@ -455,11 +463,81 @@
 		PyMem_Free(output.dst);
 	}
 
+	Py_XDECREF(readResult);
+
 	return res;
 }
 
+PyDoc_STRVAR(ZstdCompressor_stream_reader__doc__,
+"stream_reader(source, [size=0])\n"
+"\n"
+"Obtain an object that behaves like an I/O stream.\n"
+"\n"
+"The source object can be any object with a ``read(size)`` method\n"
+"or an object that conforms to the buffer protocol.\n"
+);
+
+static ZstdCompressionReader* ZstdCompressor_stream_reader(ZstdCompressor* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"source",
+		"size",
+		"read_size",
+		NULL
+	};
+
+	PyObject* source;
+	unsigned long long sourceSize = ZSTD_CONTENTSIZE_UNKNOWN;
+	size_t readSize = ZSTD_CStreamInSize();
+	ZstdCompressionReader* result = NULL;
+
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|Kk:stream_reader", kwlist,
+		&source, &sourceSize, &readSize)) {
+		return NULL;
+	}
+
+	result = (ZstdCompressionReader*)PyObject_CallObject((PyObject*)&ZstdCompressionReaderType, NULL);
+	if (!result) {
+		return NULL;
+	}
+
+	if (PyObject_HasAttrString(source, "read")) {
+		result->reader = source;
+		Py_INCREF(source);
+		result->readSize = readSize;
+	}
+	else if (1 == PyObject_CheckBuffer(source)) {
+		if (0 != PyObject_GetBuffer(source, &result->buffer, PyBUF_CONTIG_RO)) {
+			goto except;
+		}
+
+		assert(result->buffer.len >= 0);
+
+		sourceSize = result->buffer.len;
+	}
+	else {
+		PyErr_SetString(PyExc_TypeError,
+			"must pass an object with a read() method or that conforms to the buffer protocol");
+		goto except;
+	}
+
+	if (ensure_cctx(self)) {
+		goto except;
+	}
+
+	result->compressor = self;
+	Py_INCREF(self);
+	result->sourceSize = sourceSize;
+
+	return result;
+
+except:
+	Py_CLEAR(result);
+
+	return NULL;
+}
+
 PyDoc_STRVAR(ZstdCompressor_compress__doc__,
-"compress(data, allow_empty=False)\n"
+"compress(data)\n"
 "\n"
 "Compress data in a single operation.\n"
 "\n"
@@ -473,122 +551,79 @@
 static PyObject* ZstdCompressor_compress(ZstdCompressor* self, PyObject* args, PyObject* kwargs) {
 	static char* kwlist[] = {
 		"data",
-		"allow_empty",
 		NULL
 	};
 
-	const char* source;
-	Py_ssize_t sourceSize;
-	PyObject* allowEmpty = NULL;
+	Py_buffer source;
 	size_t destSize;
-	PyObject* output;
-	char* dest;
-	void* dictData = NULL;
-	size_t dictSize = 0;
+	PyObject* output = NULL;
 	size_t zresult;
-	ZSTD_parameters zparams;
+	ZSTD_outBuffer outBuffer;
+	ZSTD_inBuffer inBuffer;
 
 #if PY_MAJOR_VERSION >= 3
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y#|O:compress",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*|O:compress",
 #else
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s#|O:compress",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*|O:compress",
 #endif
-		kwlist, &source, &sourceSize, &allowEmpty)) {
-		return NULL;
-	}
-
-	if (self->threads && self->dict) {
-		PyErr_SetString(ZstdError,
-			"compress() cannot be used with both dictionaries and multi-threaded compression");
-		return NULL;
-	}
-
-	if (self->threads && self->cparams) {
-		PyErr_SetString(ZstdError,
-			"compress() cannot be used with both compression parameters and multi-threaded compression");
-		return NULL;
-	}
-
-	/* Limitation in zstd C API doesn't let decompression side distinguish
-	   between content size of 0 and unknown content size. This can make round
-	   tripping via Python difficult. Until this is fixed, require a flag
-	   to fire the footgun.
-	   https://github.com/indygreg/python-zstandard/issues/11 */
-	if (0 == sourceSize && self->fparams.contentSizeFlag
-		&& (!allowEmpty || PyObject_Not(allowEmpty))) {
-		PyErr_SetString(PyExc_ValueError, "cannot write empty inputs when writing content sizes");
-		return NULL;
-	}
-
-	destSize = ZSTD_compressBound(sourceSize);
-	output = PyBytes_FromStringAndSize(NULL, destSize);
-	if (!output) {
+		kwlist, &source)) {
 		return NULL;
 	}
 
-	dest = PyBytes_AsString(output);
-
-	if (self->dict) {
-		dictData = self->dict->dictData;
-		dictSize = self->dict->dictSize;
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
 	}
 
-	memset(&zparams, 0, sizeof(zparams));
-	if (!self->cparams) {
-		zparams.cParams = ZSTD_getCParams(self->compressionLevel, sourceSize, dictSize);
+	if (ensure_cctx(self)) {
+		goto finally;
 	}
-	else {
-		ztopy_compression_parameters(self->cparams, &zparams.cParams);
-		/* Do NOT call ZSTD_adjustCParams() here because the compression params
-		come from the user. */
+
+	destSize = ZSTD_compressBound(source.len);
+	output = PyBytes_FromStringAndSize(NULL, destSize);
+	if (!output) {
+		goto finally;
 	}
 
-	zparams.fParams = self->fparams;
-
-	/* The raw dict data has to be processed before it can be used. Since this
-	adds overhead - especially if multiple dictionary compression operations
-	are performed on the same ZstdCompressor instance - we create a
-	ZSTD_CDict once and reuse it for all operations.
+	zresult = ZSTD_CCtx_setPledgedSrcSize(self->cctx, source.len);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "error setting source size: %s",
+			ZSTD_getErrorName(zresult));
+		Py_CLEAR(output);
+		goto finally;
+	}
 
-	Note: the compression parameters used for the first invocation (possibly
-	derived from the source size) will be reused on all subsequent invocations.
-	https://github.com/facebook/zstd/issues/358 contains more info. We could
-	potentially add an argument somewhere to control this behavior.
-	*/
-	if (0 != populate_cdict(self, &zparams)) {
-		Py_DECREF(output);
-		return NULL;
-	}
+	inBuffer.src = source.buf;
+	inBuffer.size = source.len;
+	inBuffer.pos = 0;
+
+	outBuffer.dst = PyBytes_AsString(output);
+	outBuffer.size = destSize;
+	outBuffer.pos = 0;
 
 	Py_BEGIN_ALLOW_THREADS
-	if (self->mtcctx) {
-		zresult = ZSTDMT_compressCCtx(self->mtcctx, dest, destSize,
-			source, sourceSize, self->compressionLevel);
-	}
-	else {
-		/* By avoiding ZSTD_compress(), we don't necessarily write out content
-		   size. This means the argument to ZstdCompressor to control frame
-		   parameters is honored. */
-		if (self->cdict) {
-			zresult = ZSTD_compress_usingCDict(self->cctx, dest, destSize,
-				source, sourceSize, self->cdict);
-		}
-		else {
-			zresult = ZSTD_compress_advanced(self->cctx, dest, destSize,
-				source, sourceSize, dictData, dictSize, zparams);
-		}
-	}
+	/* By avoiding ZSTD_compress(), we don't necessarily write out content
+		size. This means the argument to ZstdCompressor to control frame
+		parameters is honored. */
+	zresult = ZSTD_compress_generic(self->cctx, &outBuffer, &inBuffer, ZSTD_e_end);
 	Py_END_ALLOW_THREADS
 
 	if (ZSTD_isError(zresult)) {
 		PyErr_Format(ZstdError, "cannot compress: %s", ZSTD_getErrorName(zresult));
 		Py_CLEAR(output);
-		return NULL;
+		goto finally;
 	}
-	else {
-		Py_SIZE(output) = zresult;
+	else if (zresult) {
+		PyErr_SetString(ZstdError, "unexpected partial frame flush");
+		Py_CLEAR(output);
+		goto finally;
 	}
 
+	Py_SIZE(output) = outBuffer.pos;
+
+finally:
+	PyBuffer_Release(&source);
 	return output;
 }
 
@@ -608,11 +643,23 @@
 		NULL
 	};
 
-	Py_ssize_t inSize = 0;
+	unsigned long long inSize = ZSTD_CONTENTSIZE_UNKNOWN;
 	size_t outSize = ZSTD_CStreamOutSize();
 	ZstdCompressionObj* result = NULL;
+	size_t zresult;
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|n:compressobj", kwlist, &inSize)) {
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|K:compressobj", kwlist, &inSize)) {
+		return NULL;
+	}
+
+	if (ensure_cctx(self)) {
+		return NULL;
+	}
+
+	zresult = ZSTD_CCtx_setPledgedSrcSize(self->cctx, inSize);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "error setting source size: %s",
+			ZSTD_getErrorName(zresult));
 		return NULL;
 	}
 
@@ -621,19 +668,6 @@
 		return NULL;
 	}
 
-	if (self->mtcctx) {
-		if (init_mtcstream(self, inSize)) {
-			Py_DECREF(result);
-			return NULL;
-		}
-	}
-	else {
-		if (0 != init_cstream(self, inSize)) {
-			Py_DECREF(result);
-			return NULL;
-		}
-	}
-
 	result->output.dst = PyMem_Malloc(outSize);
 	if (!result->output.dst) {
 		PyErr_NoMemory();
@@ -647,9 +681,9 @@
 	return result;
 }
 
-PyDoc_STRVAR(ZstdCompressor_read_from__doc__,
-"read_from(reader, [size=0, read_size=default, write_size=default])\n"
-"Read uncompress data from a reader and return an iterator\n"
+PyDoc_STRVAR(ZstdCompressor_read_to_iter__doc__,
+"read_to_iter(reader, [size=0, read_size=default, write_size=default])\n"
+"Read uncompressed data from a reader and return an iterator\n"
 "\n"
 "Returns an iterator of compressed data produced from reading from ``reader``.\n"
 "\n"
@@ -667,7 +701,7 @@
 "not consume from the reader unless the caller consumes from the iterator.\n"
 );
 
-static ZstdCompressorIterator* ZstdCompressor_read_from(ZstdCompressor* self, PyObject* args, PyObject* kwargs) {
+static ZstdCompressorIterator* ZstdCompressor_read_to_iter(ZstdCompressor* self, PyObject* args, PyObject* kwargs) {
 	static char* kwlist[] = {
 		"reader",
 		"size",
@@ -677,12 +711,13 @@
 	};
 
 	PyObject* reader;
-	Py_ssize_t sourceSize = 0;
+	unsigned long long sourceSize = ZSTD_CONTENTSIZE_UNKNOWN;
 	size_t inSize = ZSTD_CStreamInSize();
 	size_t outSize = ZSTD_CStreamOutSize();
 	ZstdCompressorIterator* result;
+	size_t zresult;
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|nkk:read_from", kwlist,
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|Kkk:read_to_iter", kwlist,
 		&reader, &sourceSize, &inSize, &outSize)) {
 		return NULL;
 	}
@@ -696,18 +731,11 @@
 		Py_INCREF(result->reader);
 	}
 	else if (1 == PyObject_CheckBuffer(reader)) {
-		result->buffer = PyMem_Malloc(sizeof(Py_buffer));
-		if (!result->buffer) {
+		if (0 != PyObject_GetBuffer(reader, &result->buffer, PyBUF_CONTIG_RO)) {
 			goto except;
 		}
 
-		memset(result->buffer, 0, sizeof(Py_buffer));
-
-		if (0 != PyObject_GetBuffer(reader, result->buffer, PyBUF_CONTIG_RO)) {
-			goto except;
-		}
-
-		sourceSize = result->buffer->len;
+		sourceSize = result->buffer.len;
 	}
 	else {
 		PyErr_SetString(PyExc_ValueError,
@@ -715,22 +743,20 @@
 		goto except;
 	}
 
+	if (ensure_cctx(self)) {
+		return NULL;
+	}
+
+	zresult = ZSTD_CCtx_setPledgedSrcSize(self->cctx, sourceSize);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "error setting source size: %s",
+			ZSTD_getErrorName(zresult));
+		return NULL;
+	}
+
 	result->compressor = self;
 	Py_INCREF(result->compressor);
 
-	result->sourceSize = sourceSize;
-
-	if (self->mtcctx) {
-		if (init_mtcstream(self, sourceSize)) {
-			goto except;
-		}
-	}
-	else {
-		if (0 != init_cstream(self, sourceSize)) {
-			goto except;
-		}
-	}
-
 	result->inSize = inSize;
 	result->outSize = outSize;
 
@@ -744,16 +770,13 @@
 	goto finally;
 
 except:
-	Py_XDECREF(result->compressor);
-	Py_XDECREF(result->reader);
-	Py_DECREF(result);
-	result = NULL;
+	Py_CLEAR(result);
 
 finally:
 	return result;
 }
 
-PyDoc_STRVAR(ZstdCompressor_write_to___doc__,
+PyDoc_STRVAR(ZstdCompressor_stream_writer___doc__,
 "Create a context manager to write compressed data to an object.\n"
 "\n"
 "The passed object must have a ``write()`` method.\n"
@@ -771,7 +794,7 @@
 "for a compressor output stream.\n"
 );
 
-static ZstdCompressionWriter* ZstdCompressor_write_to(ZstdCompressor* self, PyObject* args, PyObject* kwargs) {
+static ZstdCompressionWriter* ZstdCompressor_stream_writer(ZstdCompressor* self, PyObject* args, PyObject* kwargs) {
 	static char* kwlist[] = {
 		"writer",
 		"size",
@@ -781,10 +804,10 @@
 
 	PyObject* writer;
 	ZstdCompressionWriter* result;
-	Py_ssize_t sourceSize = 0;
+	unsigned long long sourceSize = ZSTD_CONTENTSIZE_UNKNOWN;
 	size_t outSize = ZSTD_CStreamOutSize();
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|nk:write_to", kwlist,
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|Kk:stream_writer", kwlist,
 		&writer, &sourceSize, &outSize)) {
 		return NULL;
 	}
@@ -794,6 +817,10 @@
 		return NULL;
 	}
 
+	if (ensure_cctx(self)) {
+		return NULL;
+	}
+
 	result = (ZstdCompressionWriter*)PyObject_CallObject((PyObject*)&ZstdCompressionWriterType, NULL);
 	if (!result) {
 		return NULL;
@@ -807,6 +834,7 @@
 
 	result->sourceSize = sourceSize;
 	result->outSize = outSize;
+	result->bytesCompressed = 0;
 
 	return result;
 }
@@ -833,6 +861,7 @@
 	WorkerError_none = 0,
 	WorkerError_zstd = 1,
 	WorkerError_no_memory = 2,
+	WorkerError_nospace = 3,
 } WorkerError;
 
 /**
@@ -841,10 +870,6 @@
 typedef struct {
 	/* Used for compression. */
 	ZSTD_CCtx* cctx;
-	ZSTD_CDict* cdict;
-	int cLevel;
-	CompressionParametersObject* cParams;
-	ZSTD_frameParameters fParams;
 
 	/* What to compress. */
 	DataSource* sources;
@@ -868,7 +893,6 @@
 	Py_ssize_t remainingItems = state->endOffset - state->startOffset + 1;
 	Py_ssize_t currentBufferStartOffset = state->startOffset;
 	size_t zresult;
-	ZSTD_parameters zparams;
 	void* newDest;
 	size_t allocationSize;
 	size_t boundSize;
@@ -879,16 +903,10 @@
 	assert(!state->destBuffers);
 	assert(0 == state->destCount);
 
-	if (state->cParams) {
-		ztopy_compression_parameters(state->cParams, &zparams.cParams);
-	}
-
-	zparams.fParams = state->fParams;
-
 	/*
 	 * The total size of the compressed data is unknown until we actually
 	 * compress data. That means we can't pre-allocate the exact size we need.
-	 * 
+	 *
 	 * There is a cost to every allocation and reallocation. So, it is in our
 	 * interest to minimize the number of allocations.
 	 *
@@ -927,7 +945,8 @@
 
 	destBuffer->segmentsSize = remainingItems;
 
-	allocationSize = roundpow2(state->totalSourceSize >> 4);
+	assert(state->totalSourceSize <= SIZE_MAX);
+	allocationSize = roundpow2((size_t)state->totalSourceSize >> 4);
 
 	/* If the maximum size of the output is larger than that, round up. */
 	boundSize = ZSTD_compressBound(sources[inputOffset].sourceSize);
@@ -949,6 +968,8 @@
 		size_t sourceSize = sources[inputOffset].sourceSize;
 		size_t destAvailable;
 		void* dest;
+		ZSTD_outBuffer opOutBuffer;
+		ZSTD_inBuffer opInBuffer;
 
 		destAvailable = destBuffer->destSize - destOffset;
 		boundSize = ZSTD_compressBound(sourceSize);
@@ -1004,7 +1025,8 @@
 			 * We could dynamically update allocation size based on work done so far.
 			 * For now, keep is simple.
 			 */
-			allocationSize = roundpow2(state->totalSourceSize >> 4);
+			assert(state->totalSourceSize <= SIZE_MAX);
+			allocationSize = roundpow2((size_t)state->totalSourceSize >> 4);
 
 			if (boundSize > allocationSize) {
 				allocationSize = roundpow2(boundSize);
@@ -1032,19 +1054,15 @@
 
 		dest = (char*)destBuffer->dest + destOffset;
 
-		if (state->cdict) {
-			zresult = ZSTD_compress_usingCDict(state->cctx, dest, destAvailable,
-				source, sourceSize, state->cdict);
-		}
-		else {
-			if (!state->cParams) {
-				zparams.cParams = ZSTD_getCParams(state->cLevel, sourceSize, 0);
-			}
+		opInBuffer.src = source;
+		opInBuffer.size = sourceSize;
+		opInBuffer.pos = 0;
 
-			zresult = ZSTD_compress_advanced(state->cctx, dest, destAvailable,
-				source, sourceSize, NULL, 0, zparams);
-		}
+		opOutBuffer.dst = dest;
+		opOutBuffer.size = destAvailable;
+		opOutBuffer.pos = 0;
 
+		zresult = ZSTD_CCtx_setPledgedSrcSize(state->cctx, sourceSize);
 		if (ZSTD_isError(zresult)) {
 			state->error = WorkerError_zstd;
 			state->zresult = zresult;
@@ -1052,10 +1070,23 @@
 			break;
 		}
 
+		zresult = ZSTD_compress_generic(state->cctx, &opOutBuffer, &opInBuffer, ZSTD_e_end);
+		if (ZSTD_isError(zresult)) {
+			state->error = WorkerError_zstd;
+			state->zresult = zresult;
+			state->errorOffset = inputOffset;
+			break;
+		}
+		else if (zresult) {
+			state->error = WorkerError_nospace;
+			state->errorOffset = inputOffset;
+			break;
+		}
+
 		destBuffer->segments[inputOffset - currentBufferStartOffset].offset = destOffset;
-		destBuffer->segments[inputOffset - currentBufferStartOffset].length = zresult;
+		destBuffer->segments[inputOffset - currentBufferStartOffset].length = opOutBuffer.pos;
 
-		destOffset += zresult;
+		destOffset += opOutBuffer.pos;
 		remainingItems--;
 	}
 
@@ -1072,15 +1103,14 @@
 }
 
 ZstdBufferWithSegmentsCollection* compress_from_datasources(ZstdCompressor* compressor,
-	DataSources* sources, unsigned int threadCount) {
-	ZSTD_parameters zparams;
+	DataSources* sources, Py_ssize_t threadCount) {
 	unsigned long long bytesPerWorker;
 	POOL_ctx* pool = NULL;
 	WorkerState* workerStates = NULL;
 	Py_ssize_t i;
 	unsigned long long workerBytes = 0;
 	Py_ssize_t workerStartOffset = 0;
-	size_t currentThread = 0;
+	Py_ssize_t currentThread = 0;
 	int errored = 0;
 	Py_ssize_t segmentsCount = 0;
 	Py_ssize_t segmentIndex;
@@ -1093,34 +1123,12 @@
 	assert(threadCount >= 1);
 
 	/* More threads than inputs makes no sense. */
-	threadCount = sources->sourcesSize < threadCount ? (unsigned int)sources->sourcesSize
+	threadCount = sources->sourcesSize < threadCount ? sources->sourcesSize
 													 : threadCount;
 
 	/* TODO lower thread count when input size is too small and threads would add
 	overhead. */
 
-	/*
-	 * When dictionaries are used, parameters are derived from the size of the
-	 * first element.
-	 *
-	 * TODO come up with a better mechanism.
-	 */
-	memset(&zparams, 0, sizeof(zparams));
-	if (compressor->cparams) {
-		ztopy_compression_parameters(compressor->cparams, &zparams.cParams);
-	}
-	else {
-		zparams.cParams = ZSTD_getCParams(compressor->compressionLevel,
-			sources->sources[0].sourceSize,
-			compressor->dict ? compressor->dict->dictSize : 0);
-	}
-
-	zparams.fParams = compressor->fparams;
-
-	if (0 != populate_cdict(compressor, &zparams)) {
-		return NULL;
-	}
-
 	workerStates = PyMem_Malloc(threadCount * sizeof(WorkerState));
 	if (NULL == workerStates) {
 		PyErr_NoMemory();
@@ -1140,16 +1148,42 @@
 	bytesPerWorker = sources->totalSourceSize / threadCount;
 
 	for (i = 0; i < threadCount; i++) {
+		size_t zresult;
+
 		workerStates[i].cctx = ZSTD_createCCtx();
 		if (!workerStates[i].cctx) {
 			PyErr_NoMemory();
 			goto finally;
 		}
 
-		workerStates[i].cdict = compressor->cdict;
-		workerStates[i].cLevel = compressor->compressionLevel;
-		workerStates[i].cParams = compressor->cparams;
-		workerStates[i].fParams = compressor->fparams;
+		zresult = ZSTD_CCtx_setParametersUsingCCtxParams(workerStates[i].cctx,
+			compressor->params);
+		if (ZSTD_isError(zresult)) {
+			PyErr_Format(ZstdError, "could not set compression parameters: %s",
+				ZSTD_getErrorName(zresult));
+			goto finally;
+		}
+
+		if (compressor->dict) {
+			if (compressor->dict->cdict) {
+				zresult = ZSTD_CCtx_refCDict(workerStates[i].cctx, compressor->dict->cdict);
+			}
+			else {
+				zresult = ZSTD_CCtx_loadDictionary_advanced(
+					workerStates[i].cctx,
+					compressor->dict->dictData,
+					compressor->dict->dictSize,
+					ZSTD_dlm_byRef,
+					compressor->dict->dictType);
+			}
+
+			if (ZSTD_isError(zresult)) {
+				PyErr_Format(ZstdError, "could not load compression dictionary: %s",
+					ZSTD_getErrorName(zresult));
+				goto finally;
+			}
+
+		}
 
 		workerStates[i].sources = sources->sources;
 		workerStates[i].sourcesSize = sources->sourcesSize;
@@ -1221,6 +1255,13 @@
 				workerStates[i].errorOffset, ZSTD_getErrorName(workerStates[i].zresult));
 			errored = 1;
 			break;
+
+		case WorkerError_nospace:
+			PyErr_Format(ZstdError, "error compressing item %zd: not enough space in output",
+				workerStates[i].errorOffset);
+			errored = 1;
+			break;
+
 		default:
 			;
 		}
@@ -1341,12 +1382,6 @@
 	Py_ssize_t sourceCount = 0;
 	ZstdBufferWithSegmentsCollection* result = NULL;
 
-	if (self->mtcctx) {
-		PyErr_SetString(ZstdError,
-			"function cannot be called on ZstdCompressor configured for multi-threaded compression");
-		return NULL;
-	}
-
 	memset(&sources, 0, sizeof(sources));
 
 	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|i:multi_compress_to_buffer", kwlist,
@@ -1372,8 +1407,14 @@
 		}
 
 		for (i = 0; i < buffer->segmentCount; i++) {
+			if (buffer->segments[i].length > SIZE_MAX) {
+				PyErr_Format(PyExc_ValueError,
+					"buffer segment %zd is too large for this platform", i);
+				goto finally;
+			}
+
 			sources.sources[i].sourceData = (char*)buffer->data + buffer->segments[i].offset;
-			sources.sources[i].sourceSize = buffer->segments[i].length;
+			sources.sources[i].sourceSize = (size_t)buffer->segments[i].length;
 			sources.totalSourceSize += buffer->segments[i].length;
 		}
 
@@ -1397,8 +1438,15 @@
 			buffer = collection->buffers[i];
 
 			for (j = 0; j < buffer->segmentCount; j++) {
+				if (buffer->segments[j].length > SIZE_MAX) {
+					PyErr_Format(PyExc_ValueError,
+						"buffer segment %zd in buffer %zd is too large for this platform",
+						j, i);
+					goto finally;
+				}
+
 				sources.sources[offset].sourceData = (char*)buffer->data + buffer->segments[j].offset;
-				sources.sources[offset].sourceSize = buffer->segments[j].length;
+				sources.sources[offset].sourceSize = (size_t)buffer->segments[j].length;
 				sources.totalSourceSize += buffer->segments[j].length;
 
 				offset++;
@@ -1416,11 +1464,6 @@
 			goto finally;
 		}
 
-		/*
-		 * It isn't clear whether the address referred to by Py_buffer.buf
-		 * is still valid after PyBuffer_Release. We we hold a reference to all
-		 * Py_buffer instances for the duration of the operation.
-		 */
 		dataBuffers = PyMem_Malloc(sourceCount * sizeof(Py_buffer));
 		if (NULL == dataBuffers) {
 			PyErr_NoMemory();
@@ -1459,6 +1502,11 @@
 		goto finally;
 	}
 
+	if (sources.totalSourceSize > SIZE_MAX) {
+		PyErr_SetString(PyExc_ValueError, "sources are too large for this platform");
+		goto finally;
+	}
+
 	result = compress_from_datasources(self, &sources, threads);
 
 finally:
@@ -1482,12 +1530,24 @@
 	METH_VARARGS | METH_KEYWORDS, ZstdCompressionObj__doc__ },
 	{ "copy_stream", (PyCFunction)ZstdCompressor_copy_stream,
 	METH_VARARGS | METH_KEYWORDS, ZstdCompressor_copy_stream__doc__ },
-	{ "read_from", (PyCFunction)ZstdCompressor_read_from,
-	METH_VARARGS | METH_KEYWORDS, ZstdCompressor_read_from__doc__ },
-	{ "write_to", (PyCFunction)ZstdCompressor_write_to,
-	METH_VARARGS | METH_KEYWORDS, ZstdCompressor_write_to___doc__ },
+	{ "stream_reader", (PyCFunction)ZstdCompressor_stream_reader,
+	METH_VARARGS | METH_KEYWORDS, ZstdCompressor_stream_reader__doc__ },
+	{ "stream_writer", (PyCFunction)ZstdCompressor_stream_writer,
+	METH_VARARGS | METH_KEYWORDS, ZstdCompressor_stream_writer___doc__ },
+	{ "read_to_iter", (PyCFunction)ZstdCompressor_read_to_iter,
+	METH_VARARGS | METH_KEYWORDS, ZstdCompressor_read_to_iter__doc__ },
+	/* TODO Remove deprecated API */
+	{ "read_from", (PyCFunction)ZstdCompressor_read_to_iter,
+	METH_VARARGS | METH_KEYWORDS, ZstdCompressor_read_to_iter__doc__ },
+	/* TODO remove deprecated API */
+	{ "write_to", (PyCFunction)ZstdCompressor_stream_writer,
+	METH_VARARGS | METH_KEYWORDS, ZstdCompressor_stream_writer___doc__ },
 	{ "multi_compress_to_buffer", (PyCFunction)ZstdCompressor_multi_compress_to_buffer,
 	METH_VARARGS | METH_KEYWORDS, ZstdCompressor_multi_compress_to_buffer__doc__ },
+	{ "memory_size", (PyCFunction)ZstdCompressor_memory_size,
+	METH_NOARGS, ZstdCompressor_memory_size__doc__ },
+	{ "frame_progression", (PyCFunction)ZstdCompressor_frame_progression,
+	METH_NOARGS, ZstdCompressor_frame_progression__doc__ },
 	{ NULL, NULL }
 };
 
--- a/contrib/python-zstandard/c-ext/compressoriterator.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/compressoriterator.c	Mon Apr 09 10:13:29 2018 -0700
@@ -21,10 +21,9 @@
 	Py_XDECREF(self->compressor);
 	Py_XDECREF(self->reader);
 
-	if (self->buffer) {
-		PyBuffer_Release(self->buffer);
-		PyMem_FREE(self->buffer);
-		self->buffer = NULL;
+	if (self->buffer.buf) {
+		PyBuffer_Release(&self->buffer);
+		memset(&self->buffer, 0, sizeof(self->buffer));
 	}
 
 	if (self->output.dst) {
@@ -58,14 +57,8 @@
 	/* If we have data left in the input, consume it. */
 	if (self->input.pos < self->input.size) {
 		Py_BEGIN_ALLOW_THREADS
-		if (self->compressor->mtcctx) {
-			zresult = ZSTDMT_compressStream(self->compressor->mtcctx,
-				&self->output, &self->input);
-		}
-		else {
-			zresult = ZSTD_compressStream(self->compressor->cstream, &self->output,
-				&self->input);
-		}
+		zresult = ZSTD_compress_generic(self->compressor->cctx, &self->output,
+			&self->input, ZSTD_e_continue);
 		Py_END_ALLOW_THREADS
 
 		/* Release the Python object holding the input buffer. */
@@ -107,14 +100,14 @@
 			PyBytes_AsStringAndSize(readResult, &readBuffer, &readSize);
 		}
 		else {
-			assert(self->buffer && self->buffer->buf);
+			assert(self->buffer.buf);
 
 			/* Only support contiguous C arrays. */
-			assert(self->buffer->strides == NULL && self->buffer->suboffsets == NULL);
-			assert(self->buffer->itemsize == 1);
+			assert(self->buffer.strides == NULL && self->buffer.suboffsets == NULL);
+			assert(self->buffer.itemsize == 1);
 
-			readBuffer = (char*)self->buffer->buf + self->bufferOffset;
-			bufferRemaining = self->buffer->len - self->bufferOffset;
+			readBuffer = (char*)self->buffer.buf + self->bufferOffset;
+			bufferRemaining = self->buffer.len - self->bufferOffset;
 			readSize = min(bufferRemaining, (Py_ssize_t)self->inSize);
 			self->bufferOffset += readSize;
 		}
@@ -130,12 +123,12 @@
 
 	/* EOF */
 	if (0 == readSize) {
-		if (self->compressor->mtcctx) {
-			zresult = ZSTDMT_endStream(self->compressor->mtcctx, &self->output);
-		}
-		else {
-			zresult = ZSTD_endStream(self->compressor->cstream, &self->output);
-		}
+		self->input.src = NULL;
+		self->input.size = 0;
+		self->input.pos = 0;
+
+		zresult = ZSTD_compress_generic(self->compressor->cctx, &self->output,
+			&self->input, ZSTD_e_end);
 		if (ZSTD_isError(zresult)) {
 			PyErr_Format(ZstdError, "error ending compression stream: %s",
 				ZSTD_getErrorName(zresult));
@@ -159,13 +152,8 @@
 	self->input.pos = 0;
 
 	Py_BEGIN_ALLOW_THREADS
-	if (self->compressor->mtcctx) {
-		zresult = ZSTDMT_compressStream(self->compressor->mtcctx, &self->output,
-			&self->input);
-	}
-	else {
-		zresult = ZSTD_compressStream(self->compressor->cstream, &self->output, &self->input);
-	}
+	zresult = ZSTD_compress_generic(self->compressor->cctx, &self->output,
+		&self->input, ZSTD_e_continue);
 	Py_END_ALLOW_THREADS
 
 	/* The input buffer currently points to memory managed by Python
--- a/contrib/python-zstandard/c-ext/constants.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/constants.c	Mon Apr 09 10:13:29 2018 -0700
@@ -52,6 +52,11 @@
 		PyErr_Format(PyExc_ValueError, "could not create frame header object");
 	}
 
+	PyModule_AddObject(mod, "CONTENTSIZE_UNKNOWN",
+		PyLong_FromUnsignedLongLong(ZSTD_CONTENTSIZE_UNKNOWN));
+	PyModule_AddObject(mod, "CONTENTSIZE_ERROR",
+		PyLong_FromUnsignedLongLong(ZSTD_CONTENTSIZE_ERROR));
+
 	PyModule_AddIntConstant(mod, "MAX_COMPRESSION_LEVEL", ZSTD_maxCLevel());
 	PyModule_AddIntConstant(mod, "COMPRESSION_RECOMMENDED_INPUT_SIZE",
 		(long)ZSTD_CStreamInSize());
@@ -75,7 +80,9 @@
 	PyModule_AddIntConstant(mod, "SEARCHLENGTH_MIN", ZSTD_SEARCHLENGTH_MIN);
 	PyModule_AddIntConstant(mod, "SEARCHLENGTH_MAX", ZSTD_SEARCHLENGTH_MAX);
 	PyModule_AddIntConstant(mod, "TARGETLENGTH_MIN", ZSTD_TARGETLENGTH_MIN);
-	PyModule_AddIntConstant(mod, "TARGETLENGTH_MAX", ZSTD_TARGETLENGTH_MAX);
+	PyModule_AddIntConstant(mod, "LDM_MINMATCH_MIN", ZSTD_LDM_MINMATCH_MIN);
+	PyModule_AddIntConstant(mod, "LDM_MINMATCH_MAX", ZSTD_LDM_MINMATCH_MAX);
+	PyModule_AddIntConstant(mod, "LDM_BUCKETSIZELOG_MAX", ZSTD_LDM_BUCKETSIZELOG_MAX);
 
 	PyModule_AddIntConstant(mod, "STRATEGY_FAST", ZSTD_fast);
 	PyModule_AddIntConstant(mod, "STRATEGY_DFAST", ZSTD_dfast);
@@ -84,4 +91,12 @@
 	PyModule_AddIntConstant(mod, "STRATEGY_LAZY2", ZSTD_lazy2);
 	PyModule_AddIntConstant(mod, "STRATEGY_BTLAZY2", ZSTD_btlazy2);
 	PyModule_AddIntConstant(mod, "STRATEGY_BTOPT", ZSTD_btopt);
+	PyModule_AddIntConstant(mod, "STRATEGY_BTULTRA", ZSTD_btultra);
+
+	PyModule_AddIntConstant(mod, "DICT_TYPE_AUTO", ZSTD_dct_auto);
+	PyModule_AddIntConstant(mod, "DICT_TYPE_RAWCONTENT", ZSTD_dct_rawContent);
+	PyModule_AddIntConstant(mod, "DICT_TYPE_FULLDICT", ZSTD_dct_fullDict);
+
+	PyModule_AddIntConstant(mod, "FORMAT_ZSTD1", ZSTD_f_zstd1);
+	PyModule_AddIntConstant(mod, "FORMAT_ZSTD1_MAGICLESS", ZSTD_f_zstd1_magicless);
 }
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/c-ext/decompressionreader.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,459 @@
+/**
+* Copyright (c) 2017-present, Gregory Szorc
+* All rights reserved.
+*
+* This software may be modified and distributed under the terms
+* of the BSD license. See the LICENSE file for details.
+*/
+
+#include "python-zstandard.h"
+
+extern PyObject* ZstdError;
+
+static void set_unsupported_operation(void) {
+	PyObject* iomod;
+	PyObject* exc;
+
+	iomod = PyImport_ImportModule("io");
+	if (NULL == iomod) {
+		return;
+	}
+
+	exc = PyObject_GetAttrString(iomod, "UnsupportedOperation");
+	if (NULL == exc) {
+		Py_DECREF(iomod);
+		return;
+	}
+
+	PyErr_SetNone(exc);
+	Py_DECREF(exc);
+	Py_DECREF(iomod);
+}
+
+static void reader_dealloc(ZstdDecompressionReader* self) {
+	Py_XDECREF(self->decompressor);
+	Py_XDECREF(self->reader);
+
+	if (self->buffer.buf) {
+		PyBuffer_Release(&self->buffer);
+	}
+
+	PyObject_Del(self);
+}
+
+static ZstdDecompressionReader* reader_enter(ZstdDecompressionReader* self) {
+	if (self->entered) {
+		PyErr_SetString(PyExc_ValueError, "cannot __enter__ multiple times");
+		return NULL;
+	}
+
+	if (ensure_dctx(self->decompressor, 1)) {
+		return NULL;
+	}
+
+	self->entered = 1;
+
+	Py_INCREF(self);
+	return self;
+}
+
+static PyObject* reader_exit(ZstdDecompressionReader* self, PyObject* args) {
+	PyObject* exc_type;
+	PyObject* exc_value;
+	PyObject* exc_tb;
+
+	if (!PyArg_ParseTuple(args, "OOO:__exit__", &exc_type, &exc_value, &exc_tb)) {
+		return NULL;
+	}
+
+	self->entered = 0;
+	self->closed = 1;
+
+	/* Release resources. */
+	Py_CLEAR(self->reader);
+	if (self->buffer.buf) {
+		PyBuffer_Release(&self->buffer);
+		memset(&self->buffer, 0, sizeof(self->buffer));
+	}
+
+	Py_CLEAR(self->decompressor);
+
+	Py_RETURN_FALSE;
+}
+
+static PyObject* reader_readable(PyObject* self) {
+	Py_RETURN_TRUE;
+}
+
+static PyObject* reader_writable(PyObject* self) {
+	Py_RETURN_FALSE;
+}
+
+static PyObject* reader_seekable(PyObject* self) {
+	Py_RETURN_TRUE;
+}
+
+static PyObject* reader_close(ZstdDecompressionReader* self) {
+	self->closed = 1;
+	Py_RETURN_NONE;
+}
+
+static PyObject* reader_closed(ZstdDecompressionReader* self) {
+	if (self->closed) {
+		Py_RETURN_TRUE;
+	}
+	else {
+		Py_RETURN_FALSE;
+	}
+}
+
+static PyObject* reader_flush(PyObject* self) {
+	Py_RETURN_NONE;
+}
+
+static PyObject* reader_isatty(PyObject* self) {
+	Py_RETURN_FALSE;
+}
+
+static PyObject* reader_read(ZstdDecompressionReader* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"size",
+		NULL
+	};
+
+	Py_ssize_t size = -1;
+	PyObject* result = NULL;
+	char* resultBuffer;
+	Py_ssize_t resultSize;
+	ZSTD_outBuffer output;
+	size_t zresult;
+
+	if (!self->entered) {
+		PyErr_SetString(ZstdError, "read() must be called from an active context manager");
+		return NULL;
+	}
+
+	if (self->closed) {
+		PyErr_SetString(PyExc_ValueError, "stream is closed");
+		return NULL;
+	}
+
+	if (self->finishedOutput) {
+		return PyBytes_FromStringAndSize("", 0);
+	}
+
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "n", kwlist, &size)) {
+		return NULL;
+	}
+
+	if (size < 1) {
+		PyErr_SetString(PyExc_ValueError, "cannot read negative or size 0 amounts");
+		return NULL;
+	}
+
+	result = PyBytes_FromStringAndSize(NULL, size);
+	if (NULL == result) {
+		return NULL;
+	}
+
+	PyBytes_AsStringAndSize(result, &resultBuffer, &resultSize);
+
+	output.dst = resultBuffer;
+	output.size = resultSize;
+	output.pos = 0;
+
+readinput:
+
+	/* Consume input data left over from last time. */
+	if (self->input.pos < self->input.size) {
+		Py_BEGIN_ALLOW_THREADS
+		zresult = ZSTD_decompress_generic(self->decompressor->dctx,
+			&output, &self->input);
+		Py_END_ALLOW_THREADS
+
+		/* Input exhausted. Clear our state tracking. */
+		if (self->input.pos == self->input.size) {
+			memset(&self->input, 0, sizeof(self->input));
+			Py_CLEAR(self->readResult);
+
+			if (self->buffer.buf) {
+				self->finishedInput = 1;
+			}
+		}
+
+		if (ZSTD_isError(zresult)) {
+			PyErr_Format(ZstdError, "zstd decompress error: %s", ZSTD_getErrorName(zresult));
+			return NULL;
+		}
+		else if (0 == zresult) {
+			self->finishedOutput = 1;
+		}
+
+		/* We fulfilled the full read request. Emit it. */
+		if (output.pos && output.pos == output.size) {
+			self->bytesDecompressed += output.size;
+			return result;
+		}
+
+		/*
+		 * There is more room in the output. Fall through to try to collect
+		 * more data so we can try to fill the output.
+		 */
+	}
+
+	if (!self->finishedInput) {
+		if (self->reader) {
+			Py_buffer buffer;
+
+			assert(self->readResult == NULL);
+			self->readResult = PyObject_CallMethod(self->reader, "read",
+				"k", self->readSize);
+			if (NULL == self->readResult) {
+				return NULL;
+			}
+
+			memset(&buffer, 0, sizeof(buffer));
+
+			if (0 != PyObject_GetBuffer(self->readResult, &buffer, PyBUF_CONTIG_RO)) {
+				return NULL;
+			}
+
+			/* EOF */
+			if (0 == buffer.len) {
+				self->finishedInput = 1;
+				Py_CLEAR(self->readResult);
+			}
+			else {
+				self->input.src = buffer.buf;
+				self->input.size = buffer.len;
+				self->input.pos = 0;
+			}
+
+			PyBuffer_Release(&buffer);
+		}
+		else {
+			assert(self->buffer.buf);
+			/*
+			 * We should only get here once since above block will exhaust
+			 * source buffer until finishedInput is set.
+			 */
+			assert(self->input.src == NULL);
+
+			self->input.src = self->buffer.buf;
+			self->input.size = self->buffer.len;
+			self->input.pos = 0;
+		}
+	}
+
+	if (self->input.size) {
+		goto readinput;
+	}
+
+	/* EOF */
+	self->bytesDecompressed += output.pos;
+
+	if (safe_pybytes_resize(&result, output.pos)) {
+		Py_XDECREF(result);
+		return NULL;
+	}
+
+	return result;
+}
+
+static PyObject* reader_readall(PyObject* self) {
+	PyErr_SetNone(PyExc_NotImplementedError);
+	return NULL;
+}
+
+static PyObject* reader_readline(PyObject* self) {
+	PyErr_SetNone(PyExc_NotImplementedError);
+	return NULL;
+}
+
+static PyObject* reader_readlines(PyObject* self) {
+	PyErr_SetNone(PyExc_NotImplementedError);
+	return NULL;
+}
+
+static PyObject* reader_seek(ZstdDecompressionReader* self, PyObject* args) {
+	Py_ssize_t pos;
+	int whence = 0;
+	unsigned long long readAmount = 0;
+	size_t defaultOutSize = ZSTD_DStreamOutSize();
+
+	if (!self->entered) {
+		PyErr_SetString(ZstdError, "seek() must be called from an active context manager");
+		return NULL;
+	}
+
+	if (self->closed) {
+		PyErr_SetString(PyExc_ValueError, "stream is closed");
+		return NULL;
+	}
+
+	if (!PyArg_ParseTuple(args, "n|i:seek", &pos, &whence)) {
+		return NULL;
+	}
+
+	if (whence == SEEK_SET) {
+		if (pos < 0) {
+			PyErr_SetString(PyExc_ValueError,
+				"cannot seek to negative position with SEEK_SET");
+			return NULL;
+		}
+
+		if ((unsigned long long)pos < self->bytesDecompressed) {
+			PyErr_SetString(PyExc_ValueError,
+				"cannot seek zstd decompression stream backwards");
+			return NULL;
+		}
+
+		readAmount = pos - self->bytesDecompressed;
+	}
+	else if (whence == SEEK_CUR) {
+		if (pos < 0) {
+			PyErr_SetString(PyExc_ValueError,
+				"cannot seek zstd decompression stream backwards");
+			return NULL;
+		}
+
+		readAmount = pos;
+	}
+	else if (whence == SEEK_END) {
+		/* We /could/ support this with pos==0. But let's not do that until someone
+		   needs it. */
+		PyErr_SetString(PyExc_ValueError,
+			"zstd decompression streams cannot be seeked with SEEK_END");
+		return NULL;
+	}
+
+	/* It is a bit inefficient to do this via the Python API. But since there
+	   is a bit of state tracking involved to read from this type, it is the
+	   easiest to implement. */
+	while (readAmount) {
+		Py_ssize_t readSize;
+		PyObject* readResult = PyObject_CallMethod((PyObject*)self, "read", "K",
+			readAmount < defaultOutSize ? readAmount : defaultOutSize);
+
+		if (!readResult) {
+			return NULL;
+		}
+
+		readSize = PyBytes_GET_SIZE(readResult);
+
+		/* Empty read means EOF. */
+		if (!readSize) {
+			break;
+		}
+
+		readAmount -= readSize;
+	}
+
+	return PyLong_FromUnsignedLongLong(self->bytesDecompressed);
+}
+
+static PyObject* reader_tell(ZstdDecompressionReader* self) {
+	/* TODO should this raise OSError since stream isn't seekable? */
+	return PyLong_FromUnsignedLongLong(self->bytesDecompressed);
+}
+
+static PyObject* reader_write(PyObject* self, PyObject* args) {
+	set_unsupported_operation();
+	return NULL;
+}
+
+static PyObject* reader_writelines(PyObject* self, PyObject* args) {
+	set_unsupported_operation();
+	return NULL;
+}
+
+static PyObject* reader_iter(PyObject* self) {
+	PyErr_SetNone(PyExc_NotImplementedError);
+	return NULL;
+}
+
+static PyObject* reader_iternext(PyObject* self) {
+	PyErr_SetNone(PyExc_NotImplementedError);
+	return NULL;
+}
+
+static PyMethodDef reader_methods[] = {
+	{ "__enter__", (PyCFunction)reader_enter, METH_NOARGS,
+	PyDoc_STR("Enter a compression context") },
+	{ "__exit__", (PyCFunction)reader_exit, METH_VARARGS,
+	PyDoc_STR("Exit a compression context") },
+	{ "close", (PyCFunction)reader_close, METH_NOARGS,
+	PyDoc_STR("Close the stream so it cannot perform any more operations") },
+	{ "closed", (PyCFunction)reader_closed, METH_NOARGS,
+	PyDoc_STR("Whether stream is closed") },
+	{ "flush", (PyCFunction)reader_flush, METH_NOARGS, PyDoc_STR("no-ops") },
+	{ "isatty", (PyCFunction)reader_isatty, METH_NOARGS, PyDoc_STR("Returns False") },
+	{ "readable", (PyCFunction)reader_readable, METH_NOARGS,
+	PyDoc_STR("Returns True") },
+	{ "read", (PyCFunction)reader_read, METH_VARARGS | METH_KEYWORDS,
+	PyDoc_STR("read compressed data") },
+	{ "readall", (PyCFunction)reader_readall, METH_NOARGS, PyDoc_STR("Not implemented") },
+	{ "readline", (PyCFunction)reader_readline, METH_NOARGS, PyDoc_STR("Not implemented") },
+	{ "readlines", (PyCFunction)reader_readlines, METH_NOARGS, PyDoc_STR("Not implemented") },
+	{ "seek", (PyCFunction)reader_seek, METH_VARARGS, PyDoc_STR("Seek the stream") },
+	{ "seekable", (PyCFunction)reader_seekable, METH_NOARGS,
+	PyDoc_STR("Returns True") },
+	{ "tell", (PyCFunction)reader_tell, METH_NOARGS,
+	PyDoc_STR("Returns current number of bytes compressed") },
+	{ "writable", (PyCFunction)reader_writable, METH_NOARGS,
+	PyDoc_STR("Returns False") },
+	{ "write", (PyCFunction)reader_write, METH_VARARGS, PyDoc_STR("unsupported operation") },
+	{ "writelines", (PyCFunction)reader_writelines, METH_VARARGS, PyDoc_STR("unsupported operation") },
+	{ NULL, NULL }
+};
+
+PyTypeObject ZstdDecompressionReaderType = {
+	PyVarObject_HEAD_INIT(NULL, 0)
+	"zstd.ZstdDecompressionReader", /* tp_name */
+	sizeof(ZstdDecompressionReader), /* tp_basicsize */
+	0, /* tp_itemsize */
+	(destructor)reader_dealloc, /* tp_dealloc */
+	0, /* tp_print */
+	0, /* tp_getattr */
+	0, /* tp_setattr */
+	0, /* tp_compare */
+	0, /* tp_repr */
+	0, /* tp_as_number */
+	0, /* tp_as_sequence */
+	0, /* tp_as_mapping */
+	0, /* tp_hash */
+	0, /* tp_call */
+	0, /* tp_str */
+	0, /* tp_getattro */
+	0, /* tp_setattro */
+	0, /* tp_as_buffer */
+	Py_TPFLAGS_DEFAULT, /* tp_flags */
+	0, /* tp_doc */
+	0, /* tp_traverse */
+	0, /* tp_clear */
+	0, /* tp_richcompare */
+	0, /* tp_weaklistoffset */
+	reader_iter, /* tp_iter */
+	reader_iternext, /* tp_iternext */
+	reader_methods, /* tp_methods */
+	0, /* tp_members */
+	0, /* tp_getset */
+	0, /* tp_base */
+	0, /* tp_dict */
+	0, /* tp_descr_get */
+	0, /* tp_descr_set */
+	0, /* tp_dictoffset */
+	0, /* tp_init */
+	0, /* tp_alloc */
+	PyType_GenericNew, /* tp_new */
+};
+
+
+void decompressionreader_module_init(PyObject* mod) {
+	/* TODO make reader a sub-class of io.RawIOBase */
+
+	Py_TYPE(&ZstdDecompressionReaderType) = &PyType_Type;
+	if (PyType_Ready(&ZstdDecompressionReaderType) < 0) {
+		return;
+	}
+}
--- a/contrib/python-zstandard/c-ext/decompressionwriter.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/decompressionwriter.c	Mon Apr 09 10:13:29 2018 -0700
@@ -27,7 +27,7 @@
 		return NULL;
 	}
 
-	if (0 != init_dstream(self->decompressor)) {
+	if (ensure_dctx(self->decompressor, 1)) {
 		return NULL;
 	}
 
@@ -44,18 +44,17 @@
 }
 
 static PyObject* ZstdDecompressionWriter_memory_size(ZstdDecompressionWriter* self) {
-	if (!self->decompressor->dstream) {
-		PyErr_SetString(ZstdError, "cannot determine size of inactive decompressor; "
-			"call when context manager is active");
-		return NULL;
-	}
-
-	return PyLong_FromSize_t(ZSTD_sizeof_DStream(self->decompressor->dstream));
+	return PyLong_FromSize_t(ZSTD_sizeof_DCtx(self->decompressor->dctx));
 }
 
-static PyObject* ZstdDecompressionWriter_write(ZstdDecompressionWriter* self, PyObject* args) {
-	const char* source;
-	Py_ssize_t sourceSize;
+static PyObject* ZstdDecompressionWriter_write(ZstdDecompressionWriter* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"data",
+		NULL
+	};
+
+	PyObject* result = NULL;
+	Py_buffer source;
 	size_t zresult = 0;
 	ZSTD_inBuffer input;
 	ZSTD_outBuffer output;
@@ -63,41 +62,47 @@
 	Py_ssize_t totalWrite = 0;
 
 #if PY_MAJOR_VERSION >= 3
-	if (!PyArg_ParseTuple(args, "y#:write", &source, &sourceSize)) {
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*:write",
 #else
-	if (!PyArg_ParseTuple(args, "s#:write", &source, &sourceSize)) {
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*:write",
 #endif
+		kwlist, &source)) {
 		return NULL;
 	}
 
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
+	}
+
 	if (!self->entered) {
 		PyErr_SetString(ZstdError, "write must be called from an active context manager");
-		return NULL;
+		goto finally;
 	}
 
-	assert(self->decompressor->dstream);
-
 	output.dst = PyMem_Malloc(self->outSize);
 	if (!output.dst) {
-		return PyErr_NoMemory();
+		PyErr_NoMemory();
+		goto finally;
 	}
 	output.size = self->outSize;
 	output.pos = 0;
 
-	input.src = source;
-	input.size = sourceSize;
+	input.src = source.buf;
+	input.size = source.len;
 	input.pos = 0;
 
-	while ((ssize_t)input.pos < sourceSize) {
+	while ((ssize_t)input.pos < source.len) {
 		Py_BEGIN_ALLOW_THREADS
-		zresult = ZSTD_decompressStream(self->decompressor->dstream, &output, &input);
+		zresult = ZSTD_decompress_generic(self->decompressor->dctx, &output, &input);
 		Py_END_ALLOW_THREADS
 
 		if (ZSTD_isError(zresult)) {
 			PyMem_Free(output.dst);
 			PyErr_Format(ZstdError, "zstd decompress error: %s",
 				ZSTD_getErrorName(zresult));
-			return NULL;
+			goto finally;
 		}
 
 		if (output.pos) {
@@ -115,7 +120,11 @@
 
 	PyMem_Free(output.dst);
 
-	return PyLong_FromSsize_t(totalWrite);
+	result = PyLong_FromSsize_t(totalWrite);
+
+finally:
+	PyBuffer_Release(&source);
+	return result;
 }
 
 static PyMethodDef ZstdDecompressionWriter_methods[] = {
@@ -125,7 +134,7 @@
 	PyDoc_STR("Exit a decompression context.") },
 	{ "memory_size", (PyCFunction)ZstdDecompressionWriter_memory_size, METH_NOARGS,
 	PyDoc_STR("Obtain the memory size in bytes of the underlying decompressor.") },
-	{ "write", (PyCFunction)ZstdDecompressionWriter_write, METH_VARARGS,
+	{ "write", (PyCFunction)ZstdDecompressionWriter_write, METH_VARARGS | METH_KEYWORDS,
 	PyDoc_STR("Compress data") },
 	{ NULL, NULL }
 };
--- a/contrib/python-zstandard/c-ext/decompressobj.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/decompressobj.c	Mon Apr 09 10:13:29 2018 -0700
@@ -20,56 +20,61 @@
 	PyObject_Del(self);
 }
 
-static PyObject* DecompressionObj_decompress(ZstdDecompressionObj* self, PyObject* args) {
-	const char* source;
-	Py_ssize_t sourceSize;
+static PyObject* DecompressionObj_decompress(ZstdDecompressionObj* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"data",
+		NULL
+	};
+
+	Py_buffer source;
 	size_t zresult;
 	ZSTD_inBuffer input;
 	ZSTD_outBuffer output;
-	size_t outSize = ZSTD_DStreamOutSize();
 	PyObject* result = NULL;
 	Py_ssize_t resultSize = 0;
 
-	/* Constructor should ensure stream is populated. */
-	assert(self->decompressor->dstream);
-
 	if (self->finished) {
 		PyErr_SetString(ZstdError, "cannot use a decompressobj multiple times");
 		return NULL;
 	}
 
 #if PY_MAJOR_VERSION >= 3
-	if (!PyArg_ParseTuple(args, "y#:decompress",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*:decompress",
 #else
-	if (!PyArg_ParseTuple(args, "s#:decompress",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*:decompress",
 #endif
-		&source, &sourceSize)) {
+		kwlist, &source)) {
 		return NULL;
 	}
 
-	input.src = source;
-	input.size = sourceSize;
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
+	}
+
+	input.src = source.buf;
+	input.size = source.len;
 	input.pos = 0;
 
-	output.dst = PyMem_Malloc(outSize);
+	output.dst = PyMem_Malloc(self->outSize);
 	if (!output.dst) {
 		PyErr_NoMemory();
-		return NULL;
+		goto except;
 	}
-	output.size = outSize;
+	output.size = self->outSize;
 	output.pos = 0;
 
 	/* Read input until exhausted. */
 	while (input.pos < input.size) {
 		Py_BEGIN_ALLOW_THREADS
-		zresult = ZSTD_decompressStream(self->decompressor->dstream, &output, &input);
+		zresult = ZSTD_decompress_generic(self->decompressor->dctx, &output, &input);
 		Py_END_ALLOW_THREADS
 
 		if (ZSTD_isError(zresult)) {
 			PyErr_Format(ZstdError, "zstd decompressor error: %s",
 				ZSTD_getErrorName(zresult));
-			result = NULL;
-			goto finally;
+			goto except;
 		}
 
 		if (0 == zresult) {
@@ -79,7 +84,8 @@
 		if (output.pos) {
 			if (result) {
 				resultSize = PyBytes_GET_SIZE(result);
-				if (-1 == _PyBytes_Resize(&result, resultSize + output.pos)) {
+				if (-1 == safe_pybytes_resize(&result, resultSize + output.pos)) {
+					Py_XDECREF(result);
 					goto except;
 				}
 
@@ -108,13 +114,14 @@
 
 finally:
 	PyMem_Free(output.dst);
+	PyBuffer_Release(&source);
 
 	return result;
 }
 
 static PyMethodDef DecompressionObj_methods[] = {
 	{ "decompress", (PyCFunction)DecompressionObj_decompress,
-	  METH_VARARGS, PyDoc_STR("decompress data") },
+	  METH_VARARGS | METH_KEYWORDS, PyDoc_STR("decompress data") },
 	{ NULL, NULL }
 };
 
--- a/contrib/python-zstandard/c-ext/decompressor.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/decompressor.c	Mon Apr 09 10:13:29 2018 -0700
@@ -12,54 +12,40 @@
 extern PyObject* ZstdError;
 
 /**
-  * Ensure the ZSTD_DStream on a ZstdDecompressor is initialized and reset.
-  *
-  * This should be called before starting a decompression operation with a
-  * ZSTD_DStream on a ZstdDecompressor.
-  */
-int init_dstream(ZstdDecompressor* decompressor) {
-	void* dictData = NULL;
-	size_t dictSize = 0;
+ * Ensure the ZSTD_DCtx on a decompressor is initiated and ready for a new operation.
+ */
+int ensure_dctx(ZstdDecompressor* decompressor, int loadDict) {
 	size_t zresult;
 
-	/* Simple case of dstream already exists. Just reset it. */
-	if (decompressor->dstream) {
-		zresult = ZSTD_resetDStream(decompressor->dstream);
+	ZSTD_DCtx_reset(decompressor->dctx);
+
+	if (decompressor->maxWindowSize) {
+		zresult = ZSTD_DCtx_setMaxWindowSize(decompressor->dctx, decompressor->maxWindowSize);
 		if (ZSTD_isError(zresult)) {
-			PyErr_Format(ZstdError, "could not reset DStream: %s",
+			PyErr_Format(ZstdError, "unable to set max window size: %s",
 				ZSTD_getErrorName(zresult));
-			return -1;
+			return 1;
 		}
-
-		return 0;
 	}
 
-	decompressor->dstream = ZSTD_createDStream();
-	if (!decompressor->dstream) {
-		PyErr_SetString(ZstdError, "could not create DStream");
-		return -1;
-	}
-
-	if (decompressor->dict) {
-		dictData = decompressor->dict->dictData;
-		dictSize = decompressor->dict->dictSize;
+	zresult = ZSTD_DCtx_setFormat(decompressor->dctx, decompressor->format);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "unable to set decoding format: %s",
+			ZSTD_getErrorName(zresult));
+		return 1;
 	}
 
-	if (dictData) {
-		zresult = ZSTD_initDStream_usingDict(decompressor->dstream, dictData, dictSize);
-	}
-	else {
-		zresult = ZSTD_initDStream(decompressor->dstream);
-	}
+	if (loadDict && decompressor->dict) {
+		if (ensure_ddict(decompressor->dict)) {
+			return 1;
+		}
 
-	if (ZSTD_isError(zresult)) {
-		/* Don't leave a reference to an invalid object. */
-		ZSTD_freeDStream(decompressor->dstream);
-		decompressor->dstream = NULL;
-
-		PyErr_Format(ZstdError, "could not initialize DStream: %s",
-			ZSTD_getErrorName(zresult));
-		return -1;
+		zresult = ZSTD_DCtx_refDDict(decompressor->dctx, decompressor->dict->ddict);
+		if (ZSTD_isError(zresult)) {
+			PyErr_Format(ZstdError, "unable to reference prepared dictionary: %s",
+				ZSTD_getErrorName(zresult));
+			return 1;
+		}
 	}
 
 	return 0;
@@ -76,36 +62,46 @@
 static int Decompressor_init(ZstdDecompressor* self, PyObject* args, PyObject* kwargs) {
 	static char* kwlist[] = {
 		"dict_data",
+		"max_window_size",
+		"format",
 		NULL
 	};
 
 	ZstdCompressionDict* dict = NULL;
+	size_t maxWindowSize = 0;
+	ZSTD_format_e format = ZSTD_f_zstd1;
 
 	self->dctx = NULL;
 	self->dict = NULL;
-	self->ddict = NULL;
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|O!:ZstdDecompressor", kwlist,
-		&ZstdCompressionDictType, &dict)) {
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|O!II:ZstdDecompressor", kwlist,
+		&ZstdCompressionDictType, &dict, &maxWindowSize, &format)) {
 		return -1;
 	}
 
-	/* TODO lazily initialize the reference ZSTD_DCtx on first use since
-	   not instances of ZstdDecompressor will use a ZSTD_DCtx. */
 	self->dctx = ZSTD_createDCtx();
 	if (!self->dctx) {
 		PyErr_NoMemory();
 		goto except;
 	}
 
+	self->maxWindowSize = maxWindowSize;
+	self->format = format;
+
 	if (dict) {
 		self->dict = dict;
 		Py_INCREF(dict);
 	}
 
+	if (ensure_dctx(self, 1)) {
+		goto except;
+	}
+
 	return 0;
 
 except:
+	Py_CLEAR(self->dict);
+
 	if (self->dctx) {
 		ZSTD_freeDCtx(self->dctx);
 		self->dctx = NULL;
@@ -117,16 +113,6 @@
 static void Decompressor_dealloc(ZstdDecompressor* self) {
 	Py_CLEAR(self->dict);
 
-	if (self->ddict) {
-		ZSTD_freeDDict(self->ddict);
-		self->ddict = NULL;
-	}
-
-	if (self->dstream) {
-		ZSTD_freeDStream(self->dstream);
-		self->dstream = NULL;
-	}
-
 	if (self->dctx) {
 		ZSTD_freeDCtx(self->dctx);
 		self->dctx = NULL;
@@ -135,6 +121,20 @@
 	PyObject_Del(self);
 }
 
+PyDoc_STRVAR(Decompressor_memory_size__doc__,
+"memory_size() -- Size of decompression context, in bytes\n"
+);
+
+static PyObject* Decompressor_memory_size(ZstdDecompressor* self) {
+	if (self->dctx) {
+		return PyLong_FromSize_t(ZSTD_sizeof_DCtx(self->dctx));
+	}
+	else {
+		PyErr_SetString(ZstdError, "no decompressor context found; this should never happen");
+		return NULL;
+	}
+}
+
 PyDoc_STRVAR(Decompressor_copy_stream__doc__,
 	"copy_stream(ifh, ofh[, read_size=default, write_size=default]) -- decompress data between streams\n"
 	"\n"
@@ -166,7 +166,7 @@
 	Py_ssize_t totalWrite = 0;
 	char* readBuffer;
 	Py_ssize_t readSize;
-	PyObject* readResult;
+	PyObject* readResult = NULL;
 	PyObject* res = NULL;
 	size_t zresult = 0;
 	PyObject* writeResult;
@@ -191,7 +191,7 @@
 	/* Prevent free on uninitialized memory in finally. */
 	output.dst = NULL;
 
-	if (0 != init_dstream(self)) {
+	if (ensure_dctx(self, 1)) {
 		res = NULL;
 		goto finally;
 	}
@@ -229,7 +229,7 @@
 
 		while (input.pos < input.size) {
 			Py_BEGIN_ALLOW_THREADS
-			zresult = ZSTD_decompressStream(self->dstream, &output, &input);
+			zresult = ZSTD_decompress_generic(self->dctx, &output, &input);
 			Py_END_ALLOW_THREADS
 
 			if (ZSTD_isError(zresult)) {
@@ -252,6 +252,8 @@
 				output.pos = 0;
 			}
 		}
+
+		Py_CLEAR(readResult);
 	}
 
 	/* Source stream is exhausted. Finish up. */
@@ -267,6 +269,8 @@
 		PyMem_Free(output.dst);
 	}
 
+	Py_XDECREF(readResult);
+
 	return res;
 }
 
@@ -300,98 +304,114 @@
 		NULL
 	};
 
-	const char* source;
-	Py_ssize_t sourceSize;
+	Py_buffer source;
 	Py_ssize_t maxOutputSize = 0;
 	unsigned long long decompressedSize;
 	size_t destCapacity;
 	PyObject* result = NULL;
-	void* dictData = NULL;
-	size_t dictSize = 0;
 	size_t zresult;
+	ZSTD_outBuffer outBuffer;
+	ZSTD_inBuffer inBuffer;
 
 #if PY_MAJOR_VERSION >= 3
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y#|n:decompress",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*|n:decompress",
 #else
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s#|n:decompress",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*|n:decompress",
 #endif
-		kwlist, &source, &sourceSize, &maxOutputSize)) {
+		kwlist, &source, &maxOutputSize)) {
 		return NULL;
 	}
 
-	if (self->dict) {
-		dictData = self->dict->dictData;
-		dictSize = self->dict->dictSize;
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
 	}
 
-	if (dictData && !self->ddict) {
-		Py_BEGIN_ALLOW_THREADS
-		self->ddict = ZSTD_createDDict_byReference(dictData, dictSize);
-		Py_END_ALLOW_THREADS
-
-		if (!self->ddict) {
-			PyErr_SetString(ZstdError, "could not create decompression dict");
-			return NULL;
-		}
+	if (ensure_dctx(self, 1)) {
+		goto finally;
 	}
 
-	decompressedSize = ZSTD_getDecompressedSize(source, sourceSize);
-	/* 0 returned if content size not in the zstd frame header */
-	if (0 == decompressedSize) {
+	decompressedSize = ZSTD_getFrameContentSize(source.buf, source.len);
+
+	if (ZSTD_CONTENTSIZE_ERROR == decompressedSize) {
+		PyErr_SetString(ZstdError, "error determining content size from frame header");
+		goto finally;
+	}
+	/* Special case of empty frame. */
+	else if (0 == decompressedSize) {
+		result = PyBytes_FromStringAndSize("", 0);
+		goto finally;
+	}
+	/* Missing content size in frame header. */
+	if (ZSTD_CONTENTSIZE_UNKNOWN == decompressedSize) {
 		if (0 == maxOutputSize) {
-			PyErr_SetString(ZstdError, "input data invalid or missing content size "
-				"in frame header");
-			return NULL;
+			PyErr_SetString(ZstdError, "could not determine content size in frame header");
+			goto finally;
 		}
-		else {
-			result = PyBytes_FromStringAndSize(NULL, maxOutputSize);
-			destCapacity = maxOutputSize;
+
+		result = PyBytes_FromStringAndSize(NULL, maxOutputSize);
+		destCapacity = maxOutputSize;
+		decompressedSize = 0;
+	}
+	/* Size is recorded in frame header. */
+	else {
+		assert(SIZE_MAX >= PY_SSIZE_T_MAX);
+		if (decompressedSize > PY_SSIZE_T_MAX) {
+			PyErr_SetString(ZstdError, "frame is too large to decompress on this platform");
+			goto finally;
 		}
-	}
-	else {
-		result = PyBytes_FromStringAndSize(NULL, decompressedSize);
-		destCapacity = decompressedSize;
+
+		result = PyBytes_FromStringAndSize(NULL, (Py_ssize_t)decompressedSize);
+		destCapacity = (size_t)decompressedSize;
 	}
 
 	if (!result) {
-		return NULL;
+		goto finally;
 	}
 
+	outBuffer.dst = PyBytes_AsString(result);
+	outBuffer.size = destCapacity;
+	outBuffer.pos = 0;
+
+	inBuffer.src = source.buf;
+	inBuffer.size = source.len;
+	inBuffer.pos = 0;
+
 	Py_BEGIN_ALLOW_THREADS
-	if (self->ddict) {
-		zresult = ZSTD_decompress_usingDDict(self->dctx,
-			PyBytes_AsString(result), destCapacity,
-			source, sourceSize, self->ddict);
-	}
-	else {
-		zresult = ZSTD_decompressDCtx(self->dctx,
-			PyBytes_AsString(result), destCapacity, source, sourceSize);
-	}
+	zresult = ZSTD_decompress_generic(self->dctx, &outBuffer, &inBuffer);
 	Py_END_ALLOW_THREADS
 
 	if (ZSTD_isError(zresult)) {
 		PyErr_Format(ZstdError, "decompression error: %s", ZSTD_getErrorName(zresult));
-		Py_DECREF(result);
-		return NULL;
+		Py_CLEAR(result);
+		goto finally;
 	}
-	else if (decompressedSize && zresult != decompressedSize) {
+	else if (zresult) {
+		PyErr_Format(ZstdError, "decompression error: did not decompress full frame");
+		Py_CLEAR(result);
+		goto finally;
+	}
+	else if (decompressedSize && outBuffer.pos != decompressedSize) {
 		PyErr_Format(ZstdError, "decompression error: decompressed %zu bytes; expected %llu",
 			zresult, decompressedSize);
-		Py_DECREF(result);
-		return NULL;
+		Py_CLEAR(result);
+		goto finally;
 	}
-	else if (zresult < destCapacity) {
-		if (_PyBytes_Resize(&result, zresult)) {
-			Py_DECREF(result);
-			return NULL;
+	else if (outBuffer.pos < destCapacity) {
+		if (safe_pybytes_resize(&result, outBuffer.pos)) {
+			Py_CLEAR(result);
+			goto finally;
 		}
 	}
 
+finally:
+	PyBuffer_Release(&source);
 	return result;
 }
 
 PyDoc_STRVAR(Decompressor_decompressobj__doc__,
-"decompressobj()\n"
+"decompressobj([write_size=default])\n"
 "\n"
 "Incrementally feed data into a decompressor.\n"
 "\n"
@@ -400,25 +420,43 @@
 "callers can swap in the zstd decompressor while using the same API.\n"
 );
 
-static ZstdDecompressionObj* Decompressor_decompressobj(ZstdDecompressor* self) {
-	ZstdDecompressionObj* result = (ZstdDecompressionObj*)PyObject_CallObject((PyObject*)&ZstdDecompressionObjType, NULL);
+static ZstdDecompressionObj* Decompressor_decompressobj(ZstdDecompressor* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"write_size",
+		NULL
+	};
+
+	ZstdDecompressionObj* result = NULL;
+	size_t outSize = ZSTD_DStreamOutSize();
+
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|k:decompressobj", kwlist, &outSize)) {
+		return NULL;
+	}
+
+	if (!outSize) {
+		PyErr_SetString(PyExc_ValueError, "write_size must be positive");
+		return NULL;
+	}
+
+	result = (ZstdDecompressionObj*)PyObject_CallObject((PyObject*)&ZstdDecompressionObjType, NULL);
 	if (!result) {
 		return NULL;
 	}
 
-	if (0 != init_dstream(self)) {
+	if (ensure_dctx(self, 1)) {
 		Py_DECREF(result);
 		return NULL;
 	}
 
 	result->decompressor = self;
 	Py_INCREF(result->decompressor);
+	result->outSize = outSize;
 
 	return result;
 }
 
-PyDoc_STRVAR(Decompressor_read_from__doc__,
-"read_from(reader[, read_size=default, write_size=default, skip_bytes=0])\n"
+PyDoc_STRVAR(Decompressor_read_to_iter__doc__,
+"read_to_iter(reader[, read_size=default, write_size=default, skip_bytes=0])\n"
 "Read compressed data and return an iterator\n"
 "\n"
 "Returns an iterator of decompressed data chunks produced from reading from\n"
@@ -437,7 +475,7 @@
 "the source.\n"
 );
 
-static ZstdDecompressorIterator* Decompressor_read_from(ZstdDecompressor* self, PyObject* args, PyObject* kwargs) {
+static ZstdDecompressorIterator* Decompressor_read_to_iter(ZstdDecompressor* self, PyObject* args, PyObject* kwargs) {
 	static char* kwlist[] = {
 		"reader",
 		"read_size",
@@ -452,7 +490,7 @@
 	ZstdDecompressorIterator* result;
 	size_t skipBytes = 0;
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|kkk:read_from", kwlist,
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|kkk:read_to_iter", kwlist,
 		&reader, &inSize, &outSize, &skipBytes)) {
 		return NULL;
 	}
@@ -474,14 +512,7 @@
 	}
 	else if (1 == PyObject_CheckBuffer(reader)) {
 		/* Object claims it is a buffer. Try to get a handle to it. */
-		result->buffer = PyMem_Malloc(sizeof(Py_buffer));
-		if (!result->buffer) {
-			goto except;
-		}
-
-		memset(result->buffer, 0, sizeof(Py_buffer));
-
-		if (0 != PyObject_GetBuffer(reader, result->buffer, PyBUF_CONTIG_RO)) {
+		if (0 != PyObject_GetBuffer(reader, &result->buffer, PyBUF_CONTIG_RO)) {
 			goto except;
 		}
 	}
@@ -498,7 +529,7 @@
 	result->outSize = outSize;
 	result->skipBytes = skipBytes;
 
-	if (0 != init_dstream(self)) {
+	if (ensure_dctx(self, 1)) {
 		goto except;
 	}
 
@@ -511,13 +542,6 @@
 	goto finally;
 
 except:
-	Py_CLEAR(result->reader);
-
-	if (result->buffer) {
-		PyBuffer_Release(result->buffer);
-		Py_CLEAR(result->buffer);
-	}
-
 	Py_CLEAR(result);
 
 finally:
@@ -525,7 +549,62 @@
 	return result;
 }
 
-PyDoc_STRVAR(Decompressor_write_to__doc__,
+PyDoc_STRVAR(Decompressor_stream_reader__doc__,
+"stream_reader(source, [read_size=default])\n"
+"\n"
+"Obtain an object that behaves like an I/O stream that can be used for\n"
+"reading decompressed output from an object.\n"
+"\n"
+"The source object can be any object with a ``read(size)`` method or that\n"
+"conforms to the buffer protocol.\n"
+);
+
+static ZstdDecompressionReader* Decompressor_stream_reader(ZstdDecompressor* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"source",
+		"read_size",
+		NULL
+	};
+
+	PyObject* source;
+	size_t readSize = ZSTD_DStreamInSize();
+	ZstdDecompressionReader* result;
+
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|k:stream_reader", kwlist,
+		&source, &readSize)) {
+		return NULL;
+	}
+
+	result = (ZstdDecompressionReader*)PyObject_CallObject((PyObject*)&ZstdDecompressionReaderType, NULL);
+	if (NULL == result) {
+		return NULL;
+	}
+
+	if (PyObject_HasAttrString(source, "read")) {
+		result->reader = source;
+		Py_INCREF(source);
+		result->readSize = readSize;
+	}
+	else if (1 == PyObject_CheckBuffer(source)) {
+		if (0 != PyObject_GetBuffer(source, &result->buffer, PyBUF_CONTIG_RO)) {
+			Py_CLEAR(result);
+			return NULL;
+		}
+	}
+	else {
+		PyErr_SetString(PyExc_TypeError,
+			"must pass an object with a read() method or that conforms to the buffer protocol");
+		Py_CLEAR(result);
+		return NULL;
+	}
+
+	result->decompressor = self;
+	Py_INCREF(self);
+
+	return result;
+}
+
+PyDoc_STRVAR(Decompressor_stream_writer__doc__,
 "Create a context manager to write decompressed data to an object.\n"
 "\n"
 "The passed object must have a ``write()`` method.\n"
@@ -538,7 +617,7 @@
 "streaming decompressor.\n"
 );
 
-static ZstdDecompressionWriter* Decompressor_write_to(ZstdDecompressor* self, PyObject* args, PyObject* kwargs) {
+static ZstdDecompressionWriter* Decompressor_stream_writer(ZstdDecompressor* self, PyObject* args, PyObject* kwargs) {
 	static char* kwlist[] = {
 		"writer",
 		"write_size",
@@ -549,7 +628,7 @@
 	size_t outSize = ZSTD_DStreamOutSize();
 	ZstdDecompressionWriter* result;
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|k:write_to", kwlist,
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O|k:stream_writer", kwlist,
 		&writer, &outSize)) {
 		return NULL;
 	}
@@ -579,7 +658,7 @@
 "Decompress a series of chunks using the content dictionary chaining technique\n"
 );
 
-static PyObject* Decompressor_decompress_content_dict_chain(PyObject* self, PyObject* args, PyObject* kwargs) {
+static PyObject* Decompressor_decompress_content_dict_chain(ZstdDecompressor* self, PyObject* args, PyObject* kwargs) {
 	static char* kwlist[] = {
 		"frames",
 		NULL
@@ -592,9 +671,8 @@
 	PyObject* chunk;
 	char* chunkData;
 	Py_ssize_t chunkSize;
-	ZSTD_DCtx* dctx = NULL;
 	size_t zresult;
-	ZSTD_frameParams frameParams;
+	ZSTD_frameHeader frameHeader;
 	void* buffer1 = NULL;
 	size_t buffer1Size = 0;
 	size_t buffer1ContentSize = 0;
@@ -603,6 +681,8 @@
 	size_t buffer2ContentSize = 0;
 	void* destBuffer = NULL;
 	PyObject* result = NULL;
+	ZSTD_outBuffer outBuffer;
+	ZSTD_inBuffer inBuffer;
 
 	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "O!:decompress_content_dict_chain",
 		kwlist, &PyList_Type, &chunks)) {
@@ -624,7 +704,7 @@
 
 	/* We require that all chunks be zstd frames and that they have content size set. */
 	PyBytes_AsStringAndSize(chunk, &chunkData, &chunkSize);
-	zresult = ZSTD_getFrameParams(&frameParams, (void*)chunkData, chunkSize);
+	zresult = ZSTD_getFrameHeader(&frameHeader, (void*)chunkData, chunkSize);
 	if (ZSTD_isError(zresult)) {
 		PyErr_SetString(PyExc_ValueError, "chunk 0 is not a valid zstd frame");
 		return NULL;
@@ -634,32 +714,56 @@
 		return NULL;
 	}
 
-	if (0 == frameParams.frameContentSize) {
+	if (ZSTD_CONTENTSIZE_UNKNOWN == frameHeader.frameContentSize) {
 		PyErr_SetString(PyExc_ValueError, "chunk 0 missing content size in frame");
 		return NULL;
 	}
 
-	dctx = ZSTD_createDCtx();
-	if (!dctx) {
-		PyErr_NoMemory();
+	assert(ZSTD_CONTENTSIZE_ERROR != frameHeader.frameContentSize);
+
+	/* We check against PY_SSIZE_T_MAX here because we ultimately cast the
+	 * result to a Python object and it's length can be no greater than
+	 * Py_ssize_t. In theory, we could have an intermediate frame that is 
+	 * larger. But a) why would this API be used for frames that large b)
+	 * it isn't worth the complexity to support. */
+	assert(SIZE_MAX >= PY_SSIZE_T_MAX);
+	if (frameHeader.frameContentSize > PY_SSIZE_T_MAX) {
+		PyErr_SetString(PyExc_ValueError,
+			"chunk 0 is too large to decompress on this platform");
+		return NULL;
+	}
+
+	if (ensure_dctx(self, 0)) {
 		goto finally;
 	}
 
-	buffer1Size = frameParams.frameContentSize;
+	buffer1Size = (size_t)frameHeader.frameContentSize;
 	buffer1 = PyMem_Malloc(buffer1Size);
 	if (!buffer1) {
 		goto finally;
 	}
 
+	outBuffer.dst = buffer1;
+	outBuffer.size = buffer1Size;
+	outBuffer.pos = 0;
+
+	inBuffer.src = chunkData;
+	inBuffer.size = chunkSize;
+	inBuffer.pos = 0;
+
 	Py_BEGIN_ALLOW_THREADS
-	zresult = ZSTD_decompressDCtx(dctx, buffer1, buffer1Size, chunkData, chunkSize);
+	zresult = ZSTD_decompress_generic(self->dctx, &outBuffer, &inBuffer);
 	Py_END_ALLOW_THREADS
 	if (ZSTD_isError(zresult)) {
 		PyErr_Format(ZstdError, "could not decompress chunk 0: %s", ZSTD_getErrorName(zresult));
 		goto finally;
 	}
+	else if (zresult) {
+		PyErr_Format(ZstdError, "chunk 0 did not decompress full frame");
+		goto finally;
+	}
 
-	buffer1ContentSize = zresult;
+	buffer1ContentSize = outBuffer.pos;
 
 	/* Special case of a simple chain. */
 	if (1 == chunksLen) {
@@ -668,7 +772,7 @@
 	}
 
 	/* This should ideally look at next chunk. But this is slightly simpler. */
-	buffer2Size = frameParams.frameContentSize;
+	buffer2Size = (size_t)frameHeader.frameContentSize;
 	buffer2 = PyMem_Malloc(buffer2Size);
 	if (!buffer2) {
 		goto finally;
@@ -688,7 +792,7 @@
 		}
 
 		PyBytes_AsStringAndSize(chunk, &chunkData, &chunkSize);
-		zresult = ZSTD_getFrameParams(&frameParams, (void*)chunkData, chunkSize);
+		zresult = ZSTD_getFrameHeader(&frameHeader, (void*)chunkData, chunkSize);
 		if (ZSTD_isError(zresult)) {
 			PyErr_Format(PyExc_ValueError, "chunk %zd is not a valid zstd frame", chunkIndex);
 			goto finally;
@@ -698,18 +802,30 @@
 			goto finally;
 		}
 
-		if (0 == frameParams.frameContentSize) {
+		if (ZSTD_CONTENTSIZE_UNKNOWN == frameHeader.frameContentSize) {
 			PyErr_Format(PyExc_ValueError, "chunk %zd missing content size in frame", chunkIndex);
 			goto finally;
 		}
 
+		assert(ZSTD_CONTENTSIZE_ERROR != frameHeader.frameContentSize);
+
+		if (frameHeader.frameContentSize > PY_SSIZE_T_MAX) {
+			PyErr_Format(PyExc_ValueError,
+				"chunk %zd is too large to decompress on this platform", chunkIndex);
+			goto finally;
+		}
+
+		inBuffer.src = chunkData;
+		inBuffer.size = chunkSize;
+		inBuffer.pos = 0;
+
 		parity = chunkIndex % 2;
 
 		/* This could definitely be abstracted to reduce code duplication. */
 		if (parity) {
 			/* Resize destination buffer to hold larger content. */
-			if (buffer2Size < frameParams.frameContentSize) {
-				buffer2Size = frameParams.frameContentSize;
+			if (buffer2Size < frameHeader.frameContentSize) {
+				buffer2Size = (size_t)frameHeader.frameContentSize;
 				destBuffer = PyMem_Realloc(buffer2, buffer2Size);
 				if (!destBuffer) {
 					goto finally;
@@ -718,19 +834,38 @@
 			}
 
 			Py_BEGIN_ALLOW_THREADS
-			zresult = ZSTD_decompress_usingDict(dctx, buffer2, buffer2Size,
-				chunkData, chunkSize, buffer1, buffer1ContentSize);
+			zresult = ZSTD_DCtx_refPrefix_advanced(self->dctx,
+				buffer1, buffer1ContentSize, ZSTD_dct_rawContent);
+			Py_END_ALLOW_THREADS
+			if (ZSTD_isError(zresult)) {
+				PyErr_Format(ZstdError,
+					"failed to load prefix dictionary at chunk %zd", chunkIndex);
+				goto finally;
+			}
+
+			outBuffer.dst = buffer2;
+			outBuffer.size = buffer2Size;
+			outBuffer.pos = 0;
+
+			Py_BEGIN_ALLOW_THREADS
+			zresult = ZSTD_decompress_generic(self->dctx, &outBuffer, &inBuffer);
 			Py_END_ALLOW_THREADS
 			if (ZSTD_isError(zresult)) {
 				PyErr_Format(ZstdError, "could not decompress chunk %zd: %s",
 					chunkIndex, ZSTD_getErrorName(zresult));
 				goto finally;
 			}
-			buffer2ContentSize = zresult;
+			else if (zresult) {
+				PyErr_Format(ZstdError, "chunk %zd did not decompress full frame",
+					chunkIndex);
+				goto finally;
+			}
+
+			buffer2ContentSize = outBuffer.pos;
 		}
 		else {
-			if (buffer1Size < frameParams.frameContentSize) {
-				buffer1Size = frameParams.frameContentSize;
+			if (buffer1Size < frameHeader.frameContentSize) {
+				buffer1Size = (size_t)frameHeader.frameContentSize;
 				destBuffer = PyMem_Realloc(buffer1, buffer1Size);
 				if (!destBuffer) {
 					goto finally;
@@ -739,15 +874,34 @@
 			}
 
 			Py_BEGIN_ALLOW_THREADS
-			zresult = ZSTD_decompress_usingDict(dctx, buffer1, buffer1Size,
-				chunkData, chunkSize, buffer2, buffer2ContentSize);
+			zresult = ZSTD_DCtx_refPrefix_advanced(self->dctx,
+				buffer2, buffer2ContentSize, ZSTD_dct_rawContent);
+			Py_END_ALLOW_THREADS
+			if (ZSTD_isError(zresult)) {
+				PyErr_Format(ZstdError,
+					"failed to load prefix dictionary at chunk %zd", chunkIndex);
+				goto finally;
+			}
+
+			outBuffer.dst = buffer1;
+			outBuffer.size = buffer1Size;
+			outBuffer.pos = 0;
+
+			Py_BEGIN_ALLOW_THREADS
+			zresult = ZSTD_decompress_generic(self->dctx, &outBuffer, &inBuffer);
 			Py_END_ALLOW_THREADS
 			if (ZSTD_isError(zresult)) {
 				PyErr_Format(ZstdError, "could not decompress chunk %zd: %s",
 					chunkIndex, ZSTD_getErrorName(zresult));
 				goto finally;
 			}
-			buffer1ContentSize = zresult;
+			else if (zresult) {
+				PyErr_Format(ZstdError, "chunk %zd did not decompress full frame",
+					chunkIndex);
+				goto finally;
+			}
+
+			buffer1ContentSize = outBuffer.pos;
 		}
 	}
 
@@ -762,17 +916,13 @@
 		PyMem_Free(buffer1);
 	}
 
-	if (dctx) {
-		ZSTD_freeDCtx(dctx);
-	}
-
 	return result;
 }
 
 typedef struct {
 	void* sourceData;
 	size_t sourceSize;
-	unsigned long long destSize;
+	size_t destSize;
 } FramePointer;
 
 typedef struct {
@@ -806,7 +956,6 @@
 
 	/* Compression state and settings. */
 	ZSTD_DCtx* dctx;
-	ZSTD_DDict* ddict;
 	int requireOutputSizes;
 
 	/* Output storage. */
@@ -838,6 +987,14 @@
 	assert(0 == state->destCount);
 	assert(state->endOffset - state->startOffset >= 0);
 
+	/* We could get here due to the way work is allocated. Ideally we wouldn't
+	   get here. But that would require a bit of a refactor in the caller. */
+	if (state->totalSourceSize > SIZE_MAX) {
+		state->error = WorkerError_memory;
+		state->errorOffset = 0;
+		return;
+	}
+
 	/*
 	 * We need to allocate a buffer to hold decompressed data. How we do this
 	 * depends on what we know about the output. The following scenarios are
@@ -853,14 +1010,34 @@
 	/* Resolve ouput segments. */
 	for (frameIndex = state->startOffset; frameIndex <= state->endOffset; frameIndex++) {
 		FramePointer* fp = &framePointers[frameIndex];
+		unsigned long long decompressedSize;
 
 		if (0 == fp->destSize) {
-			fp->destSize = ZSTD_getDecompressedSize(fp->sourceData, fp->sourceSize);
-			if (0 == fp->destSize && state->requireOutputSizes) {
+			decompressedSize = ZSTD_getFrameContentSize(fp->sourceData, fp->sourceSize);
+
+			if (ZSTD_CONTENTSIZE_ERROR == decompressedSize) {
 				state->error = WorkerError_unknownSize;
 				state->errorOffset = frameIndex;
 				return;
 			}
+			else if (ZSTD_CONTENTSIZE_UNKNOWN == decompressedSize) {
+				if (state->requireOutputSizes) {
+					state->error = WorkerError_unknownSize;
+					state->errorOffset = frameIndex;
+					return;
+				}
+
+				/* This will fail the assert for .destSize > 0 below. */
+				decompressedSize = 0;
+			}
+
+			if (decompressedSize > SIZE_MAX) {
+				state->error = WorkerError_memory;
+				state->errorOffset = frameIndex;
+				return;
+			}
+
+			fp->destSize = (size_t)decompressedSize;
 		}
 
 		totalOutputSize += fp->destSize;
@@ -878,7 +1055,7 @@
 
 	assert(framePointers[state->startOffset].destSize > 0); /* For now. */
 
-	allocationSize = roundpow2(state->totalSourceSize);
+	allocationSize = roundpow2((size_t)state->totalSourceSize);
 
 	if (framePointers[state->startOffset].destSize > allocationSize) {
 		allocationSize = roundpow2(framePointers[state->startOffset].destSize);
@@ -902,6 +1079,8 @@
 	destBuffer->segmentsSize = remainingItems;
 
 	for (frameIndex = state->startOffset; frameIndex <= state->endOffset; frameIndex++) {
+		ZSTD_outBuffer outBuffer;
+		ZSTD_inBuffer inBuffer;
 		const void* source = framePointers[frameIndex].sourceData;
 		const size_t sourceSize = framePointers[frameIndex].sourceSize;
 		void* dest;
@@ -956,7 +1135,7 @@
 			/* Don't take any chances will non-NULL pointers. */
 			memset(destBuffer, 0, sizeof(DestBuffer));
 
-			allocationSize = roundpow2(state->totalSourceSize);
+			allocationSize = roundpow2((size_t)state->totalSourceSize);
 
 			if (decompressedSize > allocationSize) {
 				allocationSize = roundpow2(decompressedSize);
@@ -985,31 +1164,31 @@
 
 		dest = (char*)destBuffer->dest + destOffset;
 
-		if (state->ddict) {
-			zresult = ZSTD_decompress_usingDDict(state->dctx, dest, decompressedSize,
-				source, sourceSize, state->ddict);
-		}
-		else {
-			zresult = ZSTD_decompressDCtx(state->dctx, dest, decompressedSize,
-				source, sourceSize);
-		}
+		outBuffer.dst = dest;
+		outBuffer.size = decompressedSize;
+		outBuffer.pos = 0;
 
+		inBuffer.src = source;
+		inBuffer.size = sourceSize;
+		inBuffer.pos = 0;
+
+		zresult = ZSTD_decompress_generic(state->dctx, &outBuffer, &inBuffer);
 		if (ZSTD_isError(zresult)) {
 			state->error = WorkerError_zstd;
 			state->zresult = zresult;
 			state->errorOffset = frameIndex;
 			return;
 		}
-		else if (zresult != decompressedSize) {
+		else if (zresult || outBuffer.pos != decompressedSize) {
 			state->error = WorkerError_sizeMismatch;
-			state->zresult = zresult;
+			state->zresult = outBuffer.pos;
 			state->errorOffset = frameIndex;
 			return;
 		}
 
 		destBuffer->segments[localOffset].offset = destOffset;
-		destBuffer->segments[localOffset].length = decompressedSize;
-		destOffset += zresult;
+		destBuffer->segments[localOffset].length = outBuffer.pos;
+		destOffset += outBuffer.pos;
 		localOffset++;
 		remainingItems--;
 	}
@@ -1027,9 +1206,7 @@
 }
 
 ZstdBufferWithSegmentsCollection* decompress_from_framesources(ZstdDecompressor* decompressor, FrameSources* frames,
-	unsigned int threadCount) {
-	void* dictData = NULL;
-	size_t dictSize = 0;
+	Py_ssize_t threadCount) {
 	Py_ssize_t i = 0;
 	int errored = 0;
 	Py_ssize_t segmentsCount;
@@ -1039,7 +1216,7 @@
 	ZstdBufferWithSegmentsCollection* result = NULL;
 	FramePointer* framePointers = frames->frames;
 	unsigned long long workerBytes = 0;
-	int currentThread = 0;
+	Py_ssize_t currentThread = 0;
 	Py_ssize_t workerStartOffset = 0;
 	POOL_ctx* pool = NULL;
 	WorkerState* workerStates = NULL;
@@ -1049,24 +1226,14 @@
 	assert(threadCount >= 1);
 
 	/* More threads than inputs makes no sense under any conditions. */
-	threadCount = frames->framesSize < threadCount ? (unsigned int)frames->framesSize
+	threadCount = frames->framesSize < threadCount ? frames->framesSize
 												   : threadCount;
 
 	/* TODO lower thread count if input size is too small and threads would just
 	   add overhead. */
 
 	if (decompressor->dict) {
-		dictData = decompressor->dict->dictData;
-		dictSize = decompressor->dict->dictSize;
-	}
-
-	if (dictData && !decompressor->ddict) {
-		Py_BEGIN_ALLOW_THREADS
-		decompressor->ddict = ZSTD_createDDict_byReference(dictData, dictSize);
-		Py_END_ALLOW_THREADS
-
-		if (!decompressor->ddict) {
-			PyErr_SetString(ZstdError, "could not create decompression dict");
+		if (ensure_ddict(decompressor->dict)) {
 			return NULL;
 		}
 	}
@@ -1091,7 +1258,14 @@
 
 	bytesPerWorker = frames->compressedSize / threadCount;
 
+	if (bytesPerWorker > SIZE_MAX) {
+		PyErr_SetString(ZstdError, "too much data per worker for this platform");
+		goto finally;
+	}
+
 	for (i = 0; i < threadCount; i++) {
+		size_t zresult;
+
 		workerStates[i].dctx = ZSTD_createDCtx();
 		if (NULL == workerStates[i].dctx) {
 			PyErr_NoMemory();
@@ -1100,7 +1274,15 @@
 
 		ZSTD_copyDCtx(workerStates[i].dctx, decompressor->dctx);
 
-		workerStates[i].ddict = decompressor->ddict;
+		if (decompressor->dict) {
+			zresult = ZSTD_DCtx_refDDict(workerStates[i].dctx, decompressor->dict->ddict);
+			if (zresult) {
+				PyErr_Format(ZstdError, "unable to reference prepared dictionary: %s",
+					ZSTD_getErrorName(zresult));
+				goto finally;
+			}
+		}
+
 		workerStates[i].framePointers = framePointers;
 		workerStates[i].requireOutputSizes = 1;
 	}
@@ -1178,7 +1360,7 @@
 			break;
 
 		case WorkerError_sizeMismatch:
-			PyErr_Format(ZstdError, "error decompressing item %zd: decompressed %zu bytes; expected %llu",
+			PyErr_Format(ZstdError, "error decompressing item %zd: decompressed %zu bytes; expected %zu",
 				workerStates[i].errorOffset, workerStates[i].zresult,
 				framePointers[workerStates[i].errorOffset].destSize);
 			errored = 1;
@@ -1388,9 +1570,21 @@
 				decompressedSize = frameSizesP[i];
 			}
 
+			if (sourceSize > SIZE_MAX) {
+				PyErr_Format(PyExc_ValueError,
+					"item %zd is too large for this platform", i);
+				goto finally;
+			}
+
+			if (decompressedSize > SIZE_MAX) {
+				PyErr_Format(PyExc_ValueError,
+					"decompressed size of item %zd is too large for this platform", i);
+				goto finally;
+			}
+
 			framePointers[i].sourceData = sourceData;
-			framePointers[i].sourceSize = sourceSize;
-			framePointers[i].destSize = decompressedSize;
+			framePointers[i].sourceSize = (size_t)sourceSize;
+			framePointers[i].destSize = (size_t)decompressedSize;
 		}
 	}
 	else if (PyObject_TypeCheck(frames, &ZstdBufferWithSegmentsCollectionType)) {
@@ -1419,17 +1613,33 @@
 			buffer = collection->buffers[i];
 
 			for (segmentIndex = 0; segmentIndex < buffer->segmentCount; segmentIndex++) {
+				unsigned long long decompressedSize = frameSizesP ? frameSizesP[offset] : 0;
+
 				if (buffer->segments[segmentIndex].offset + buffer->segments[segmentIndex].length > buffer->dataSize) {
 					PyErr_Format(PyExc_ValueError, "item %zd has offset outside memory area",
 						offset);
 					goto finally;
 				}
 
+				if (buffer->segments[segmentIndex].length > SIZE_MAX) {
+					PyErr_Format(PyExc_ValueError,
+						"item %zd in buffer %zd is too large for this platform",
+						segmentIndex, i);
+					goto finally;
+				}
+
+				if (decompressedSize > SIZE_MAX) {
+					PyErr_Format(PyExc_ValueError,
+						"decompressed size of item %zd in buffer %zd is too large for this platform",
+						segmentIndex, i);
+					goto finally;
+				}
+
 				totalInputSize += buffer->segments[segmentIndex].length;
 
 				framePointers[offset].sourceData = (char*)buffer->data + buffer->segments[segmentIndex].offset;
-				framePointers[offset].sourceSize = buffer->segments[segmentIndex].length;
-				framePointers[offset].destSize = frameSizesP ? frameSizesP[offset] : 0;
+				framePointers[offset].sourceSize = (size_t)buffer->segments[segmentIndex].length;
+				framePointers[offset].destSize = (size_t)decompressedSize;
 
 				offset++;
 			}
@@ -1450,11 +1660,6 @@
 			goto finally;
 		}
 
-		/*
-		 * It is not clear whether Py_buffer.buf is still valid after
-		 * PyBuffer_Release. So, we hold a reference to all Py_buffer instances
-		 * for the duration of the operation.
-		 */
 		frameBuffers = PyMem_Malloc(frameCount * sizeof(Py_buffer));
 		if (NULL == frameBuffers) {
 			PyErr_NoMemory();
@@ -1465,6 +1670,8 @@
 
 		/* Do a pass to assemble info about our input buffers and output sizes. */
 		for (i = 0; i < frameCount; i++) {
+			unsigned long long decompressedSize = frameSizesP ? frameSizesP[i] : 0;
+
 			if (0 != PyObject_GetBuffer(PyList_GET_ITEM(frames, i),
 				&frameBuffers[i], PyBUF_CONTIG_RO)) {
 				PyErr_Clear();
@@ -1472,11 +1679,17 @@
 				goto finally;
 			}
 
+			if (decompressedSize > SIZE_MAX) {
+				PyErr_Format(PyExc_ValueError,
+					"decompressed size of item %zd is too large for this platform", i);
+				goto finally;
+			}
+
 			totalInputSize += frameBuffers[i].len;
 
 			framePointers[i].sourceData = frameBuffers[i].buf;
 			framePointers[i].sourceSize = frameBuffers[i].len;
-			framePointers[i].destSize = frameSizesP ? frameSizesP[i] : 0;
+			framePointers[i].destSize = (size_t)decompressedSize;
 		}
 	}
 	else {
@@ -1514,16 +1727,26 @@
 	Decompressor_copy_stream__doc__ },
 	{ "decompress", (PyCFunction)Decompressor_decompress, METH_VARARGS | METH_KEYWORDS,
 	Decompressor_decompress__doc__ },
-	{ "decompressobj", (PyCFunction)Decompressor_decompressobj, METH_NOARGS,
+	{ "decompressobj", (PyCFunction)Decompressor_decompressobj, METH_VARARGS | METH_KEYWORDS,
 	Decompressor_decompressobj__doc__ },
-	{ "read_from", (PyCFunction)Decompressor_read_from, METH_VARARGS | METH_KEYWORDS,
-	Decompressor_read_from__doc__ },
-	{ "write_to", (PyCFunction)Decompressor_write_to, METH_VARARGS | METH_KEYWORDS,
-	Decompressor_write_to__doc__ },
+	{ "read_to_iter", (PyCFunction)Decompressor_read_to_iter, METH_VARARGS | METH_KEYWORDS,
+	Decompressor_read_to_iter__doc__ },
+	/* TODO Remove deprecated API */
+	{ "read_from", (PyCFunction)Decompressor_read_to_iter, METH_VARARGS | METH_KEYWORDS,
+	Decompressor_read_to_iter__doc__ },
+	{ "stream_reader", (PyCFunction)Decompressor_stream_reader,
+	METH_VARARGS | METH_KEYWORDS, Decompressor_stream_reader__doc__ },
+	{ "stream_writer", (PyCFunction)Decompressor_stream_writer, METH_VARARGS | METH_KEYWORDS,
+	Decompressor_stream_writer__doc__ },
+	/* TODO remove deprecated API */
+	{ "write_to", (PyCFunction)Decompressor_stream_writer, METH_VARARGS | METH_KEYWORDS,
+	Decompressor_stream_writer__doc__ },
 	{ "decompress_content_dict_chain", (PyCFunction)Decompressor_decompress_content_dict_chain,
 	  METH_VARARGS | METH_KEYWORDS, Decompressor_decompress_content_dict_chain__doc__ },
 	{ "multi_decompress_to_buffer", (PyCFunction)Decompressor_multi_decompress_to_buffer,
 	  METH_VARARGS | METH_KEYWORDS, Decompressor_multi_decompress_to_buffer__doc__ },
+	{ "memory_size", (PyCFunction)Decompressor_memory_size, METH_NOARGS,
+	Decompressor_memory_size__doc__ },
 	{ NULL, NULL }
 };
 
--- a/contrib/python-zstandard/c-ext/decompressoriterator.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/decompressoriterator.c	Mon Apr 09 10:13:29 2018 -0700
@@ -20,10 +20,9 @@
 	Py_XDECREF(self->decompressor);
 	Py_XDECREF(self->reader);
 
-	if (self->buffer) {
-		PyBuffer_Release(self->buffer);
-		PyMem_FREE(self->buffer);
-		self->buffer = NULL;
+	if (self->buffer.buf) {
+		PyBuffer_Release(&self->buffer);
+		memset(&self->buffer, 0, sizeof(self->buffer));
 	}
 
 	if (self->input.src) {
@@ -45,8 +44,6 @@
 	DecompressorIteratorResult result;
 	size_t oldInputPos = self->input.pos;
 
-	assert(self->decompressor->dstream);
-
 	result.chunk = NULL;
 
 	chunk = PyBytes_FromStringAndSize(NULL, self->outSize);
@@ -60,7 +57,7 @@
 	self->output.pos = 0;
 
 	Py_BEGIN_ALLOW_THREADS
-	zresult = ZSTD_decompressStream(self->decompressor->dstream, &self->output, &self->input);
+	zresult = ZSTD_decompress_generic(self->decompressor->dctx, &self->output, &self->input);
 	Py_END_ALLOW_THREADS
 
 	/* We're done with the pointer. Nullify to prevent anyone from getting a
@@ -86,7 +83,8 @@
 	/* If it produced output data, return it. */
 	if (self->output.pos) {
 		if (self->output.pos < self->outSize) {
-			if (_PyBytes_Resize(&chunk, self->output.pos)) {
+			if (safe_pybytes_resize(&chunk, self->output.pos)) {
+				Py_XDECREF(chunk);
 				result.errored = 1;
 				return result;
 			}
@@ -137,15 +135,15 @@
 			PyBytes_AsStringAndSize(readResult, &readBuffer, &readSize);
 		}
 		else {
-			assert(self->buffer && self->buffer->buf);
+			assert(self->buffer.buf);
 
 			/* Only support contiguous C arrays for now */
-			assert(self->buffer->strides == NULL && self->buffer->suboffsets == NULL);
-			assert(self->buffer->itemsize == 1);
+			assert(self->buffer.strides == NULL && self->buffer.suboffsets == NULL);
+			assert(self->buffer.itemsize == 1);
 
 			/* TODO avoid memcpy() below */
-			readBuffer = (char *)self->buffer->buf + self->bufferOffset;
-			bufferRemaining = self->buffer->len - self->bufferOffset;
+			readBuffer = (char *)self->buffer.buf + self->bufferOffset;
+			bufferRemaining = self->buffer.len - self->bufferOffset;
 			readSize = min(bufferRemaining, (Py_ssize_t)self->inSize);
 			self->bufferOffset += readSize;
 		}
--- a/contrib/python-zstandard/c-ext/frameparams.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/frameparams.c	Mon Apr 09 10:13:29 2018 -0700
@@ -13,50 +13,56 @@
 PyDoc_STRVAR(FrameParameters__doc__,
 	"FrameParameters: information about a zstd frame");
 
-FrameParametersObject* get_frame_parameters(PyObject* self, PyObject* args) {
-	const char* source;
-	Py_ssize_t sourceSize;
-	ZSTD_frameParams params;
+FrameParametersObject* get_frame_parameters(PyObject* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"data",
+		NULL
+	};
+
+	Py_buffer source;
+	ZSTD_frameHeader header;
 	FrameParametersObject* result = NULL;
 	size_t zresult;
 
 #if PY_MAJOR_VERSION >= 3
-	if (!PyArg_ParseTuple(args, "y#:get_frame_parameters",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*:get_frame_parameters",
 #else
-	if (!PyArg_ParseTuple(args, "s#:get_frame_parameters",
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*:get_frame_parameters",
 #endif
-		&source, &sourceSize)) {
+		kwlist, &source)) {
 		return NULL;
 	}
 
-	/* Needed for Python 2 to reject unicode */
-	if (!PyBytes_Check(PyTuple_GET_ITEM(args, 0))) {
-		PyErr_SetString(PyExc_TypeError, "argument must be bytes");
-		return NULL;
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
 	}
 
-	zresult = ZSTD_getFrameParams(&params, (void*)source, sourceSize);
+	zresult = ZSTD_getFrameHeader(&header, source.buf, source.len);
 
 	if (ZSTD_isError(zresult)) {
 		PyErr_Format(ZstdError, "cannot get frame parameters: %s", ZSTD_getErrorName(zresult));
-		return NULL;
+		goto finally;
 	}
 
 	if (zresult) {
 		PyErr_Format(ZstdError, "not enough data for frame parameters; need %zu bytes", zresult);
-		return NULL;
+		goto finally;
 	}
 
 	result = PyObject_New(FrameParametersObject, &FrameParametersType);
 	if (!result) {
-		return NULL;
+		goto finally;
 	}
 
-	result->frameContentSize = params.frameContentSize;
-	result->windowSize = params.windowSize;
-	result->dictID = params.dictID;
-	result->checksumFlag = params.checksumFlag ? 1 : 0;
+	result->frameContentSize = header.frameContentSize;
+	result->windowSize = header.windowSize;
+	result->dictID = header.dictID;
+	result->checksumFlag = header.checksumFlag ? 1 : 0;
 
+finally:
+	PyBuffer_Release(&source);
 	return result;
 }
 
@@ -68,7 +74,7 @@
 	{ "content_size", T_ULONGLONG,
 	  offsetof(FrameParametersObject, frameContentSize), READONLY,
 	  "frame content size" },
-	{ "window_size", T_UINT,
+	{ "window_size", T_ULONGLONG,
 	  offsetof(FrameParametersObject, windowSize), READONLY,
 	  "window size" },
 	{ "dict_id", T_UINT,
--- a/contrib/python-zstandard/c-ext/python-zstandard.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/c-ext/python-zstandard.h	Mon Apr 09 10:13:29 2018 -0700
@@ -12,12 +12,10 @@
 
 #define ZSTD_STATIC_LINKING_ONLY
 #define ZDICT_STATIC_LINKING_ONLY
-#include "mem.h"
-#include "zstd.h"
-#include "zdict.h"
-#include "zstdmt_compress.h"
+#include <zstd.h>
+#include <zdict.h>
 
-#define PYTHON_ZSTANDARD_VERSION "0.8.1"
+#define PYTHON_ZSTANDARD_VERSION "0.9.0"
 
 typedef enum {
 	compressorobj_flush_finish,
@@ -25,22 +23,38 @@
 } CompressorObj_Flush;
 
 /*
-   Represents a CompressionParameters type.
+   Represents a ZstdCompressionParameters type.
 
-   This type is basically a wrapper around ZSTD_compressionParameters.
+   This type holds all the low-level compression parameters that can be set.
 */
 typedef struct {
 	PyObject_HEAD
+	ZSTD_CCtx_params* params;
+	unsigned format;
+	int compressionLevel;
 	unsigned windowLog;
+	unsigned hashLog;
 	unsigned chainLog;
-	unsigned hashLog;
 	unsigned searchLog;
-	unsigned searchLength;
+	unsigned minMatch;
 	unsigned targetLength;
-	ZSTD_strategy strategy;
-} CompressionParametersObject;
+	unsigned compressionStrategy;
+	unsigned contentSizeFlag;
+	unsigned checksumFlag;
+	unsigned dictIDFlag;
+	unsigned threads;
+	unsigned jobSize;
+	unsigned overlapSizeLog;
+	unsigned compressLiterals;
+	unsigned forceMaxWindow;
+	unsigned enableLongDistanceMatching;
+	unsigned ldmHashLog;
+	unsigned ldmMinMatch;
+	unsigned ldmBucketSizeLog;
+	unsigned ldmHashEveryLog;
+} ZstdCompressionParametersObject;
 
-extern PyTypeObject CompressionParametersType;
+extern PyTypeObject ZstdCompressionParametersType;
 
 /*
    Represents a FrameParameters type.
@@ -50,7 +64,7 @@
 typedef struct {
 	PyObject_HEAD
 	unsigned long long frameContentSize;
-	unsigned windowSize;
+	unsigned long long windowSize;
 	unsigned dictID;
 	char checksumFlag;
 } FrameParametersObject;
@@ -69,10 +83,14 @@
 	void* dictData;
 	/* Size of dictionary data. */
 	size_t dictSize;
+	ZSTD_dictContentType_e dictType;
 	/* k parameter for cover dictionaries. Only populated by train_cover_dict(). */
 	unsigned k;
 	/* d parameter for cover dictionaries. Only populated by train_cover_dict(). */
 	unsigned d;
+	/* Digested dictionary, suitable for reuse. */
+	ZSTD_CDict* cdict;
+	ZSTD_DDict* ddict;
 } ZstdCompressionDict;
 
 extern PyTypeObject ZstdCompressionDictType;
@@ -83,29 +101,15 @@
 typedef struct {
 	PyObject_HEAD
 
-	/* Configured compression level. Should be always set. */
-	int compressionLevel;
 	/* Number of threads to use for operations. */
 	unsigned int threads;
 	/* Pointer to compression dictionary to use. NULL if not using dictionary
 	   compression. */
 	ZstdCompressionDict* dict;
-	/* Compression context to use. Populated during object construction. NULL
-	   if using multi-threaded compression. */
+	/* Compression context to use. Populated during object construction. */
 	ZSTD_CCtx* cctx;
-	/* Multi-threaded compression context to use. Populated during object
-	   construction. NULL if not using multi-threaded compression. */
-	ZSTDMT_CCtx* mtcctx;
-	/* Digest compression dictionary. NULL initially. Populated on first use. */
-	ZSTD_CDict* cdict;
-	/* Low-level compression parameter control. NULL unless passed to
-	   constructor. Takes precedence over `compressionLevel` if defined. */
-	CompressionParametersObject* cparams;
-	/* Controls zstd frame options. */
-	ZSTD_frameParameters fparams;
-	/* Holds state for streaming compression. Shared across all invocation.
-	   Populated on first use. */
-	ZSTD_CStream* cstream;
+	/* Compression parameters in use. */
+	ZSTD_CCtx_params* params;
 } ZstdCompressor;
 
 extern PyTypeObject ZstdCompressorType;
@@ -125,9 +129,10 @@
 
 	ZstdCompressor* compressor;
 	PyObject* writer;
-	Py_ssize_t sourceSize;
+	unsigned long long sourceSize;
 	size_t outSize;
 	int entered;
+	unsigned long long bytesCompressed;
 } ZstdCompressionWriter;
 
 extern PyTypeObject ZstdCompressionWriterType;
@@ -137,9 +142,8 @@
 
 	ZstdCompressor* compressor;
 	PyObject* reader;
-	Py_buffer* buffer;
+	Py_buffer buffer;
 	Py_ssize_t bufferOffset;
-	Py_ssize_t sourceSize;
 	size_t inSize;
 	size_t outSize;
 
@@ -155,11 +159,32 @@
 typedef struct {
 	PyObject_HEAD
 
+	ZstdCompressor* compressor;
+	PyObject* reader;
+	Py_buffer buffer;
+	unsigned long long sourceSize;
+	size_t readSize;
+
+	int entered;
+	int closed;
+	unsigned long long bytesCompressed;
+
+	ZSTD_inBuffer input;
+	ZSTD_outBuffer output;
+	int finishedInput;
+	int finishedOutput;
+	PyObject* readResult;
+} ZstdCompressionReader;
+
+extern PyTypeObject ZstdCompressionReaderType;
+
+typedef struct {
+	PyObject_HEAD
+
 	ZSTD_DCtx* dctx;
-
 	ZstdCompressionDict* dict;
-	ZSTD_DDict* ddict;
-	ZSTD_DStream* dstream;
+	size_t maxWindowSize;
+	ZSTD_format_e format;
 } ZstdDecompressor;
 
 extern PyTypeObject ZstdDecompressorType;
@@ -168,6 +193,7 @@
 	PyObject_HEAD
 
 	ZstdDecompressor* decompressor;
+	size_t outSize;
 	int finished;
 } ZstdDecompressionObj;
 
@@ -176,6 +202,40 @@
 typedef struct {
 	PyObject_HEAD
 
+	/* Parent decompressor to which this object is associated. */
+	ZstdDecompressor* decompressor;
+	/* Object to read() from (if reading from a stream). */
+	PyObject* reader;
+	/* Size for read() operations on reader. */
+	size_t readSize;
+	/* Buffer to read from (if reading from a buffer). */
+	Py_buffer buffer;
+
+	/* Whether the context manager is active. */
+	int entered;
+	/* Whether we've closed the stream. */
+	int closed;
+
+	/* Number of bytes decompressed and returned to user. */
+	unsigned long long bytesDecompressed;
+
+	/* Tracks data going into decompressor. */
+	ZSTD_inBuffer input;
+
+	/* Holds output from read() operation on reader. */
+	PyObject* readResult;
+
+	/* Whether all input has been sent to the decompressor. */
+	int finishedInput;
+	/* Whether all output has been flushed from the decompressor. */
+	int finishedOutput;
+} ZstdDecompressionReader;
+
+extern PyTypeObject ZstdDecompressionReaderType;
+
+typedef struct {
+	PyObject_HEAD
+
 	ZstdDecompressor* decompressor;
 	PyObject* writer;
 	size_t outSize;
@@ -189,7 +249,7 @@
 
 	ZstdDecompressor* decompressor;
 	PyObject* reader;
-	Py_buffer* buffer;
+	Py_buffer buffer;
 	Py_ssize_t bufferOffset;
 	size_t inSize;
 	size_t outSize;
@@ -209,6 +269,9 @@
 } DecompressorIteratorResult;
 
 typedef struct {
+	/* The public API is that these are 64-bit unsigned integers. So these can't
+	 * be size_t, even though values larger than SIZE_MAX or PY_SSIZE_T_MAX may 
+	 * be nonsensical for this platform. */
 	unsigned long long offset;
 	unsigned long long length;
 } BufferSegment;
@@ -270,16 +333,14 @@
 
 extern PyTypeObject ZstdBufferWithSegmentsCollectionType;
 
-void ztopy_compression_parameters(CompressionParametersObject* params, ZSTD_compressionParameters* zparams);
-CompressionParametersObject* get_compression_parameters(PyObject* self, PyObject* args);
-FrameParametersObject* get_frame_parameters(PyObject* self, PyObject* args);
-PyObject* estimate_compression_context_size(PyObject* self, PyObject* args);
-int init_cstream(ZstdCompressor* compressor, unsigned long long sourceSize);
-int init_mtcstream(ZstdCompressor* compressor, Py_ssize_t sourceSize);
-int init_dstream(ZstdDecompressor* decompressor);
+int set_parameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, unsigned value);
+int set_parameters(ZSTD_CCtx_params* params, ZstdCompressionParametersObject* obj);
+FrameParametersObject* get_frame_parameters(PyObject* self, PyObject* args, PyObject* kwargs);
+int ensure_ddict(ZstdCompressionDict* dict);
+int ensure_dctx(ZstdDecompressor* decompressor, int loadDict);
 ZstdCompressionDict* train_dictionary(PyObject* self, PyObject* args, PyObject* kwargs);
-ZstdCompressionDict* train_cover_dictionary(PyObject* self, PyObject* args, PyObject* kwargs);
 ZstdBufferWithSegments* BufferWithSegments_FromMemory(void* data, unsigned long long dataSize, BufferSegment* segments, Py_ssize_t segmentsSize);
 Py_ssize_t BufferWithSegmentsCollection_length(ZstdBufferWithSegmentsCollection*);
 int cpu_count(void);
 size_t roundpow2(size_t);
+int safe_pybytes_resize(PyObject** obj, Py_ssize_t size);
--- a/contrib/python-zstandard/make_cffi.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/make_cffi.py	Mon Apr 09 10:13:29 2018 -0700
@@ -27,6 +27,11 @@
     'compress/fse_compress.c',
     'compress/huf_compress.c',
     'compress/zstd_compress.c',
+    'compress/zstd_double_fast.c',
+    'compress/zstd_fast.c',
+    'compress/zstd_lazy.c',
+    'compress/zstd_ldm.c',
+    'compress/zstd_opt.c',
     'compress/zstdmt_compress.c',
     'decompress/huf_decompress.c',
     'decompress/zstd_decompress.c',
@@ -38,7 +43,6 @@
 # Headers whose preprocessed output will be fed into cdef().
 HEADERS = [os.path.join(HERE, 'zstd', *p) for p in (
     ('zstd.h',),
-    ('compress', 'zstdmt_compress.h'),
     ('dictBuilder', 'zdict.h'),
 )]
 
@@ -80,7 +84,9 @@
 def preprocess(path):
     with open(path, 'rb') as fh:
         lines = []
-        for l in fh:
+        it = iter(fh)
+
+        for l in it:
             # zstd.h includes <stddef.h>, which is also included by cffi's
             # boilerplate. This can lead to duplicate declarations. So we strip
             # this include from the preprocessor invocation.
@@ -137,18 +143,21 @@
 
 
 ffi = cffi.FFI()
+# zstd.h uses a possible undefined MIN(). Define it until
+# https://github.com/facebook/zstd/issues/976 is fixed.
 # *_DISABLE_DEPRECATE_WARNINGS prevents the compiler from emitting a warning
 # when cffi uses the function. Since we statically link against zstd, even
 # if we use the deprecated functions it shouldn't be a huge problem.
 ffi.set_source('_zstd_cffi', '''
-#include "mem.h"
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
 #define ZSTD_STATIC_LINKING_ONLY
-#include "zstd.h"
+#include <zstd.h>
 #define ZDICT_STATIC_LINKING_ONLY
 #define ZDICT_DISABLE_DEPRECATE_WARNINGS
-#include "zdict.h"
-#include "zstdmt_compress.h"
-''', sources=SOURCES, include_dirs=INCLUDE_DIRS)
+#include <zdict.h>
+''', sources=SOURCES,
+     include_dirs=INCLUDE_DIRS,
+     extra_compile_args=['-DZSTD_MULTITHREAD'])
 
 DEFINE = re.compile(b'^\\#define ([a-zA-Z0-9_]+) ')
 
--- a/contrib/python-zstandard/setup.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/setup.py	Mon Apr 09 10:13:29 2018 -0700
@@ -5,6 +5,7 @@
 # This software may be modified and distributed under the terms
 # of the BSD license. See the LICENSE file for details.
 
+import os
 import sys
 from setuptools import setup
 
@@ -16,14 +17,32 @@
 import setup_zstd
 
 SUPPORT_LEGACY = False
+SYSTEM_ZSTD = False
+WARNINGS_AS_ERRORS = False
 
-if "--legacy" in sys.argv:
+if os.environ.get('ZSTD_WARNINGS_AS_ERRORS', ''):
+    WARNINGS_AS_ERRORS = True
+
+if '--legacy' in sys.argv:
     SUPPORT_LEGACY = True
-    sys.argv.remove("--legacy")
+    sys.argv.remove('--legacy')
+
+if '--system-zstd' in sys.argv:
+    SYSTEM_ZSTD = True
+    sys.argv.remove('--system-zstd')
+
+if '--warnings-as-errors' in sys.argv:
+    WARNINGS_AS_ERRORS = True
+    sys.argv.remote('--warning-as-errors')
 
 # Code for obtaining the Extension instance is in its own module to
 # facilitate reuse in other projects.
-extensions = [setup_zstd.get_c_extension(SUPPORT_LEGACY, 'zstd')]
+extensions = [
+    setup_zstd.get_c_extension(name='zstd',
+                               support_legacy=SUPPORT_LEGACY,
+                               system_zstd=SYSTEM_ZSTD,
+                               warnings_as_errors=WARNINGS_AS_ERRORS),
+]
 
 install_requires = []
 
@@ -31,8 +50,11 @@
     import make_cffi
     extensions.append(make_cffi.ffi.distutils_extension())
 
-    # Need change in 1.8 for ffi.from_buffer() behavior.
-    install_requires.append('cffi>=1.8')
+    # Need change in 1.10 for ffi.from_buffer() to handle all buffer types
+    # (like memoryview).
+    # Need feature in 1.11 for ffi.gc() to declare size of objects so we avoid
+    # garbage collection pitfalls.
+    install_requires.append('cffi>=1.11')
 
 version = None
 
@@ -62,14 +84,13 @@
         'Intended Audience :: Developers',
         'License :: OSI Approved :: BSD License',
         'Programming Language :: C',
-        'Programming Language :: Python :: 2.6',
         'Programming Language :: Python :: 2.7',
-        'Programming Language :: Python :: 3.3',
         'Programming Language :: Python :: 3.4',
         'Programming Language :: Python :: 3.5',
         'Programming Language :: Python :: 3.6',
     ],
     keywords='zstandard zstd compression',
+    packages=['zstandard'],
     ext_modules=extensions,
     test_suite='tests',
     install_requires=install_requires,
--- a/contrib/python-zstandard/setup_zstd.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/setup_zstd.py	Mon Apr 09 10:13:29 2018 -0700
@@ -4,7 +4,10 @@
 # This software may be modified and distributed under the terms
 # of the BSD license. See the LICENSE file for details.
 
+import distutils.ccompiler
 import os
+import sys
+
 from distutils.extension import Extension
 
 
@@ -19,6 +22,11 @@
     'compress/fse_compress.c',
     'compress/huf_compress.c',
     'compress/zstd_compress.c',
+    'compress/zstd_double_fast.c',
+    'compress/zstd_fast.c',
+    'compress/zstd_lazy.c',
+    'compress/zstd_ldm.c',
+    'compress/zstd_opt.c',
     'compress/zstdmt_compress.c',
     'decompress/huf_decompress.c',
     'decompress/zstd_decompress.c',
@@ -41,7 +49,6 @@
 )]
 
 zstd_includes = [
-    'c-ext',
     'zstd',
     'zstd/common',
     'zstd/compress',
@@ -54,7 +61,14 @@
     'zstd/legacy',
 ]
 
+ext_includes = [
+    'c-ext',
+    'zstd/common',
+]
+
 ext_sources = [
+    'zstd/common/pool.c',
+    'zstd/common/threading.c',
     'zstd.c',
     'c-ext/bufferutil.c',
     'c-ext/compressiondict.c',
@@ -62,11 +76,13 @@
     'c-ext/compressor.c',
     'c-ext/compressoriterator.c',
     'c-ext/compressionparams.c',
+    'c-ext/compressionreader.c',
     'c-ext/compressionwriter.c',
     'c-ext/constants.c',
     'c-ext/decompressobj.c',
     'c-ext/decompressor.c',
     'c-ext/decompressoriterator.c',
+    'c-ext/decompressionreader.c',
     'c-ext/decompressionwriter.c',
     'c-ext/frameparams.c',
 ]
@@ -76,27 +92,67 @@
 ]
 
 
-def get_c_extension(support_legacy=False, name='zstd'):
+def get_c_extension(support_legacy=False, system_zstd=False, name='zstd',
+                    warnings_as_errors=False):
     """Obtain a distutils.extension.Extension for the C extension."""
     root = os.path.abspath(os.path.dirname(__file__))
 
-    sources = [os.path.join(root, p) for p in zstd_sources + ext_sources]
-    if support_legacy:
-        sources.extend([os.path.join(root, p) for p in zstd_sources_legacy])
+    sources = set([os.path.join(root, p) for p in ext_sources])
+    if not system_zstd:
+        sources.update([os.path.join(root, p) for p in zstd_sources])
+        if support_legacy:
+            sources.update([os.path.join(root, p) for p in zstd_sources_legacy])
+    sources = list(sources)
 
-    include_dirs = [os.path.join(root, d) for d in zstd_includes]
-    if support_legacy:
-        include_dirs.extend([os.path.join(root, d) for d in zstd_includes_legacy])
+    include_dirs = set([os.path.join(root, d) for d in ext_includes])
+    if not system_zstd:
+        include_dirs.update([os.path.join(root, d) for d in zstd_includes])
+        if support_legacy:
+            include_dirs.update([os.path.join(root, d) for d in zstd_includes_legacy])
+    include_dirs = list(include_dirs)
 
     depends = [os.path.join(root, p) for p in zstd_depends]
 
+    compiler = distutils.ccompiler.new_compiler()
+
+    # Needed for MSVC.
+    if hasattr(compiler, 'initialize'):
+        compiler.initialize()
+
+    if compiler.compiler_type == 'unix':
+        compiler_type = 'unix'
+    elif compiler.compiler_type == 'msvc':
+        compiler_type = 'msvc'
+    else:
+        raise Exception('unhandled compiler type: %s' %
+                        compiler.compiler_type)
+
     extra_args = ['-DZSTD_MULTITHREAD']
 
-    if support_legacy:
+    if not system_zstd:
+        extra_args.append('-DZSTDLIB_VISIBILITY=')
+        extra_args.append('-DZDICTLIB_VISIBILITY=')
+        extra_args.append('-DZSTDERRORLIB_VISIBILITY=')
+
+        if compiler_type == 'unix':
+            extra_args.append('-fvisibility=hidden')
+
+    if not system_zstd and support_legacy:
         extra_args.append('-DZSTD_LEGACY_SUPPORT=1')
 
+    if warnings_as_errors:
+        if compiler_type == 'unix':
+            extra_args.append('-Werror')
+        elif compiler_type == 'msvc':
+            extra_args.append('/WX')
+        else:
+            assert False
+
+    libraries = ['zstd'] if system_zstd else []
+
     # TODO compile with optimizations.
     return Extension(name, sources,
                      include_dirs=include_dirs,
                      depends=depends,
-                     extra_compile_args=extra_args)
+                     extra_compile_args=extra_args,
+                     libraries=libraries)
--- a/contrib/python-zstandard/tests/common.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/common.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,16 +1,48 @@
+import imp
 import inspect
 import io
 import os
 import types
 
+try:
+    import hypothesis
+except ImportError:
+    hypothesis = None
+
 
 def make_cffi(cls):
     """Decorator to add CFFI versions of each test method."""
 
+    # The module containing this class definition should
+    # `import zstandard as zstd`. Otherwise things may blow up.
+    mod = inspect.getmodule(cls)
+    if not hasattr(mod, 'zstd'):
+        raise Exception('test module does not contain "zstd" symbol')
+
+    if not hasattr(mod.zstd, 'backend'):
+        raise Exception('zstd symbol does not have "backend" attribute; did '
+                        'you `import zstandard as zstd`?')
+
+    # If `import zstandard` already chose the cffi backend, there is nothing
+    # for us to do: we only add the cffi variation if the default backend
+    # is the C extension.
+    if mod.zstd.backend == 'cffi':
+        return cls
+
+    old_env = dict(os.environ)
+    os.environ['PYTHON_ZSTANDARD_IMPORT_POLICY'] = 'cffi'
     try:
-        import zstd_cffi
-    except ImportError:
-        return cls
+        try:
+            mod_info = imp.find_module('zstandard')
+            mod = imp.load_module('zstandard_cffi', *mod_info)
+        except ImportError:
+            return cls
+    finally:
+        os.environ.clear()
+        os.environ.update(old_env)
+
+    if mod.backend != 'cffi':
+        raise Exception('got the zstandard %s backend instead of cffi' % mod.backend)
 
     # If CFFI version is available, dynamically construct test methods
     # that use it.
@@ -29,13 +61,13 @@
         # the function object and install it in a new attribute.
         if isinstance(fn, types.FunctionType):
             globs = dict(fn.__globals__)
-            globs['zstd'] = zstd_cffi
+            globs['zstd'] = mod
             new_fn = types.FunctionType(fn.__code__, globs, name,
                                         fn.__defaults__, fn.__closure__)
             new_method = new_fn
         else:
             globs = dict(fn.__func__.func_globals)
-            globs['zstd'] = zstd_cffi
+            globs['zstd'] = mod
             new_fn = types.FunctionType(fn.__func__.func_code, globs, name,
                                         fn.__func__.func_defaults,
                                         fn.__func__.func_closure)
@@ -86,3 +118,34 @@
                 pass
 
     return _source_files
+
+
+def generate_samples():
+    inputs = [
+        b'foo',
+        b'bar',
+        b'abcdef',
+        b'sometext',
+        b'baz',
+    ]
+
+    samples = []
+
+    for i in range(128):
+        samples.append(inputs[i % 5])
+        samples.append(inputs[i % 5] * (i + 3))
+        samples.append(inputs[-(i % 5)] * (i + 2))
+
+    return samples
+
+
+if hypothesis:
+    default_settings = hypothesis.settings()
+    hypothesis.settings.register_profile('default', default_settings)
+
+    ci_settings = hypothesis.settings(max_examples=2500,
+                                      max_iterations=2500)
+    hypothesis.settings.register_profile('ci', ci_settings)
+
+    hypothesis.settings.load_profile(
+        os.environ.get('HYPOTHESIS_PROFILE', 'default'))
--- a/contrib/python-zstandard/tests/test_buffer_util.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_buffer_util.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,11 +1,7 @@
 import struct
+import unittest
 
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
-
-import zstd
+import zstandard as zstd
 
 ss = struct.Struct('=QQ')
 
--- a/contrib/python-zstandard/tests/test_compressor.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_compressor.py	Mon Apr 09 10:13:29 2018 -0700
@@ -2,13 +2,10 @@
 import io
 import struct
 import sys
+import tarfile
+import unittest
 
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
-
-import zstd
+import zstandard as zstd
 
 from .common import (
     make_cffi,
@@ -23,7 +20,8 @@
 
 
 def multithreaded_chunk_size(level, source_size=0):
-    params = zstd.get_compression_parameters(level, source_size)
+    params = zstd.ZstdCompressionParameters.from_level(level,
+                                                       source_size=source_size)
 
     return 1 << (params.window_log + 2)
 
@@ -32,67 +30,82 @@
 class TestCompressor(unittest.TestCase):
     def test_level_bounds(self):
         with self.assertRaises(ValueError):
-            zstd.ZstdCompressor(level=0)
+            zstd.ZstdCompressor(level=23)
 
-        with self.assertRaises(ValueError):
-            zstd.ZstdCompressor(level=23)
+    def test_memory_size(self):
+        cctx = zstd.ZstdCompressor(level=1)
+        self.assertGreater(cctx.memory_size(), 100)
 
 
 @make_cffi
 class TestCompressor_compress(unittest.TestCase):
-    def test_multithreaded_unsupported(self):
-        samples = []
-        for i in range(128):
-            samples.append(b'foo' * 64)
-            samples.append(b'bar' * 64)
-
-        d = zstd.train_dictionary(8192, samples)
-
-        cctx = zstd.ZstdCompressor(dict_data=d, threads=2)
-
-        with self.assertRaisesRegexp(zstd.ZstdError, 'compress\(\) cannot be used with both dictionaries and multi-threaded compression'):
-            cctx.compress(b'foo')
-
-        params = zstd.get_compression_parameters(3)
-        cctx = zstd.ZstdCompressor(compression_params=params, threads=2)
-        with self.assertRaisesRegexp(zstd.ZstdError, 'compress\(\) cannot be used with both compression parameters and multi-threaded compression'):
-            cctx.compress(b'foo')
-
     def test_compress_empty(self):
-        cctx = zstd.ZstdCompressor(level=1)
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
         result = cctx.compress(b'')
         self.assertEqual(result, b'\x28\xb5\x2f\xfd\x00\x48\x01\x00\x00')
         params = zstd.get_frame_parameters(result)
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 524288)
         self.assertEqual(params.dict_id, 0)
         self.assertFalse(params.has_checksum, 0)
 
-        # TODO should be temporary until https://github.com/facebook/zstd/issues/506
-        # is fixed.
-        cctx = zstd.ZstdCompressor(write_content_size=True)
-        with self.assertRaises(ValueError):
-            cctx.compress(b'')
+        cctx = zstd.ZstdCompressor()
+        result = cctx.compress(b'')
+        self.assertEqual(result, b'\x28\xb5\x2f\xfd\x20\x00\x01\x00\x00')
+        params = zstd.get_frame_parameters(result)
+        self.assertEqual(params.content_size, 0)
+
+    def test_input_types(self):
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
+        expected = b'\x28\xb5\x2f\xfd\x00\x00\x19\x00\x00\x66\x6f\x6f'
 
-        cctx.compress(b'', allow_empty=True)
+        mutable_array = bytearray(3)
+        mutable_array[:] = b'foo'
+
+        sources = [
+            memoryview(b'foo'),
+            bytearray(b'foo'),
+            mutable_array,
+        ]
+
+        for source in sources:
+            self.assertEqual(cctx.compress(source), expected)
 
     def test_compress_large(self):
         chunks = []
         for i in range(255):
             chunks.append(struct.Struct('>B').pack(i) * 16384)
 
-        cctx = zstd.ZstdCompressor(level=3)
+        cctx = zstd.ZstdCompressor(level=3, write_content_size=False)
         result = cctx.compress(b''.join(chunks))
         self.assertEqual(len(result), 999)
         self.assertEqual(result[0:4], b'\x28\xb5\x2f\xfd')
 
-        # This matches the test for read_from() below.
-        cctx = zstd.ZstdCompressor(level=1)
+        # This matches the test for read_to_iter() below.
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
         result = cctx.compress(b'f' * zstd.COMPRESSION_RECOMMENDED_INPUT_SIZE + b'o')
         self.assertEqual(result, b'\x28\xb5\x2f\xfd\x00\x40\x54\x00\x00'
                                  b'\x10\x66\x66\x01\x00\xfb\xff\x39\xc0'
                                  b'\x02\x09\x00\x00\x6f')
 
+    def test_negative_level(self):
+        cctx = zstd.ZstdCompressor(level=-4)
+        result = cctx.compress(b'foo' * 256)
+
+    def test_no_magic(self):
+        params = zstd.ZstdCompressionParameters.from_level(
+            1, format=zstd.FORMAT_ZSTD1)
+        cctx = zstd.ZstdCompressor(compression_params=params)
+        magic = cctx.compress(b'foobar')
+
+        params = zstd.ZstdCompressionParameters.from_level(
+            1, format=zstd.FORMAT_ZSTD1_MAGICLESS)
+        cctx = zstd.ZstdCompressor(compression_params=params)
+        no_magic = cctx.compress(b'foobar')
+
+        self.assertEqual(magic[0:4], b'\x28\xb5\x2f\xfd')
+        self.assertEqual(magic[4:], no_magic)
+
     def test_write_checksum(self):
         cctx = zstd.ZstdCompressor(level=1)
         no_checksum = cctx.compress(b'foobar')
@@ -109,15 +122,15 @@
 
     def test_write_content_size(self):
         cctx = zstd.ZstdCompressor(level=1)
+        with_size = cctx.compress(b'foobar' * 256)
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
         no_size = cctx.compress(b'foobar' * 256)
-        cctx = zstd.ZstdCompressor(level=1, write_content_size=True)
-        with_size = cctx.compress(b'foobar' * 256)
 
         self.assertEqual(len(with_size), len(no_size) + 1)
 
         no_params = zstd.get_frame_parameters(no_size)
         with_params = zstd.get_frame_parameters(with_size)
-        self.assertEqual(no_params.content_size, 0)
+        self.assertEqual(no_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(with_params.content_size, 1536)
 
     def test_no_dict_id(self):
@@ -140,7 +153,7 @@
         no_params = zstd.get_frame_parameters(no_dict_id)
         with_params = zstd.get_frame_parameters(with_dict_id)
         self.assertEqual(no_params.dict_id, 0)
-        self.assertEqual(with_params.dict_id, 1584102229)
+        self.assertEqual(with_params.dict_id, 1387616518)
 
     def test_compress_dict_multiple(self):
         samples = []
@@ -156,6 +169,21 @@
         for i in range(32):
             cctx.compress(b'foo bar foobar foo bar foobar')
 
+    def test_dict_precompute(self):
+        samples = []
+        for i in range(128):
+            samples.append(b'foo' * 64)
+            samples.append(b'bar' * 64)
+            samples.append(b'foobar' * 64)
+
+        d = zstd.train_dictionary(8192, samples)
+        d.precompute_compress(level=1)
+
+        cctx = zstd.ZstdCompressor(level=1, dict_data=d)
+
+        for i in range(32):
+            cctx.compress(b'foo bar foobar foo bar foobar')
+
     def test_multithreaded(self):
         chunk_size = multithreaded_chunk_size(1)
         source = b''.join([b'x' * chunk_size, b'y' * chunk_size])
@@ -171,16 +199,65 @@
         dctx = zstd.ZstdDecompressor()
         self.assertEqual(dctx.decompress(compressed), source)
 
+    def test_multithreaded_dict(self):
+        samples = []
+        for i in range(128):
+            samples.append(b'foo' * 64)
+            samples.append(b'bar' * 64)
+            samples.append(b'foobar' * 64)
+
+        d = zstd.train_dictionary(1024, samples)
+
+        cctx = zstd.ZstdCompressor(dict_data=d, threads=2)
+
+        result = cctx.compress(b'foo')
+        params = zstd.get_frame_parameters(result);
+        self.assertEqual(params.content_size, 3);
+        self.assertEqual(params.dict_id, d.dict_id())
+
+        self.assertEqual(result,
+                         b'\x28\xb5\x2f\xfd\x23\x06\x59\xb5\x52\x03\x19\x00\x00'
+                         b'\x66\x6f\x6f')
+
+    def test_multithreaded_compression_params(self):
+        params = zstd.ZstdCompressionParameters.from_level(0, threads=2)
+        cctx = zstd.ZstdCompressor(compression_params=params)
+
+        result = cctx.compress(b'foo')
+        params = zstd.get_frame_parameters(result);
+        self.assertEqual(params.content_size, 3);
+
+        self.assertEqual(result,
+                         b'\x28\xb5\x2f\xfd\x20\x03\x19\x00\x00\x66\x6f\x6f')
+
 
 @make_cffi
 class TestCompressor_compressobj(unittest.TestCase):
     def test_compressobj_empty(self):
-        cctx = zstd.ZstdCompressor(level=1)
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
         cobj = cctx.compressobj()
         self.assertEqual(cobj.compress(b''), b'')
         self.assertEqual(cobj.flush(),
                          b'\x28\xb5\x2f\xfd\x00\x48\x01\x00\x00')
 
+    def test_input_types(self):
+        expected = b'\x28\xb5\x2f\xfd\x00\x48\x19\x00\x00\x66\x6f\x6f'
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
+
+        mutable_array = bytearray(3)
+        mutable_array[:] = b'foo'
+
+        sources = [
+            memoryview(b'foo'),
+            bytearray(b'foo'),
+            mutable_array,
+        ]
+
+        for source in sources:
+            cobj = cctx.compressobj()
+            self.assertEqual(cobj.compress(source), b'')
+            self.assertEqual(cobj.flush(), expected)
+
     def test_compressobj_large(self):
         chunks = []
         for i in range(255):
@@ -194,7 +271,7 @@
         self.assertEqual(result[0:4], b'\x28\xb5\x2f\xfd')
 
         params = zstd.get_frame_parameters(result)
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 1048576)
         self.assertEqual(params.dict_id, 0)
         self.assertFalse(params.has_checksum)
@@ -209,8 +286,8 @@
 
         no_params = zstd.get_frame_parameters(no_checksum)
         with_params = zstd.get_frame_parameters(with_checksum)
-        self.assertEqual(no_params.content_size, 0)
-        self.assertEqual(with_params.content_size, 0)
+        self.assertEqual(no_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
+        self.assertEqual(with_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(no_params.dict_id, 0)
         self.assertEqual(with_params.dict_id, 0)
         self.assertFalse(no_params.has_checksum)
@@ -221,14 +298,14 @@
     def test_write_content_size(self):
         cctx = zstd.ZstdCompressor(level=1)
         cobj = cctx.compressobj(size=len(b'foobar' * 256))
+        with_size = cobj.compress(b'foobar' * 256) + cobj.flush()
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
+        cobj = cctx.compressobj(size=len(b'foobar' * 256))
         no_size = cobj.compress(b'foobar' * 256) + cobj.flush()
-        cctx = zstd.ZstdCompressor(level=1, write_content_size=True)
-        cobj = cctx.compressobj(size=len(b'foobar' * 256))
-        with_size = cobj.compress(b'foobar' * 256) + cobj.flush()
 
         no_params = zstd.get_frame_parameters(no_size)
         with_params = zstd.get_frame_parameters(with_size)
-        self.assertEqual(no_params.content_size, 0)
+        self.assertEqual(no_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(with_params.content_size, 1536)
         self.assertEqual(no_params.dict_id, 0)
         self.assertEqual(with_params.dict_id, 0)
@@ -300,6 +377,34 @@
 
         self.assertEqual(len(compressed), 295)
 
+    def test_frame_progression(self):
+        cctx = zstd.ZstdCompressor()
+
+        self.assertEqual(cctx.frame_progression(), (0, 0, 0))
+
+        cobj = cctx.compressobj()
+
+        cobj.compress(b'foobar')
+        self.assertEqual(cctx.frame_progression(), (6, 0, 0))
+
+        cobj.flush()
+        self.assertEqual(cctx.frame_progression(), (6, 6, 15))
+
+    def test_bad_size(self):
+        cctx = zstd.ZstdCompressor()
+
+        cobj = cctx.compressobj(size=2)
+        with self.assertRaisesRegexp(zstd.ZstdError, 'Src size is incorrect'):
+            cobj.compress(b'foo')
+
+        # Try another operation on this instance.
+        with self.assertRaisesRegexp(zstd.ZstdError, 'Src size is incorrect'):
+            cobj.compress(b'aa')
+
+        # Try another operation on the compressor.
+        cctx.compressobj(size=4)
+        cctx.compress(b'foobar')
+
 
 @make_cffi
 class TestCompressor_copy_stream(unittest.TestCase):
@@ -323,7 +428,7 @@
         source = io.BytesIO()
         dest = io.BytesIO()
 
-        cctx = zstd.ZstdCompressor(level=1)
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
         r, w = cctx.copy_stream(source, dest)
         self.assertEqual(int(r), 0)
         self.assertEqual(w, 9)
@@ -345,7 +450,7 @@
         self.assertEqual(w, 999)
 
         params = zstd.get_frame_parameters(dest.getvalue())
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 1048576)
         self.assertEqual(params.dict_id, 0)
         self.assertFalse(params.has_checksum)
@@ -367,8 +472,8 @@
 
         no_params = zstd.get_frame_parameters(no_checksum.getvalue())
         with_params = zstd.get_frame_parameters(with_checksum.getvalue())
-        self.assertEqual(no_params.content_size, 0)
-        self.assertEqual(with_params.content_size, 0)
+        self.assertEqual(no_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
+        self.assertEqual(with_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(no_params.dict_id, 0)
         self.assertEqual(with_params.dict_id, 0)
         self.assertFalse(no_params.has_checksum)
@@ -378,12 +483,12 @@
         source = io.BytesIO(b'foobar' * 256)
         no_size = io.BytesIO()
 
-        cctx = zstd.ZstdCompressor(level=1)
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
         cctx.copy_stream(source, no_size)
 
         source.seek(0)
         with_size = io.BytesIO()
-        cctx = zstd.ZstdCompressor(level=1, write_content_size=True)
+        cctx = zstd.ZstdCompressor(level=1)
         cctx.copy_stream(source, with_size)
 
         # Source content size is unknown, so no content size written.
@@ -400,7 +505,7 @@
 
         no_params = zstd.get_frame_parameters(no_size.getvalue())
         with_params = zstd.get_frame_parameters(with_size.getvalue())
-        self.assertEqual(no_params.content_size, 0)
+        self.assertEqual(no_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(with_params.content_size, 1536)
         self.assertEqual(no_params.dict_id, 0)
         self.assertEqual(with_params.dict_id, 0)
@@ -426,19 +531,18 @@
         source.seek(0)
 
         dest = io.BytesIO()
-        cctx = zstd.ZstdCompressor(threads=2)
+        cctx = zstd.ZstdCompressor(threads=2, write_content_size=False)
         r, w = cctx.copy_stream(source, dest)
         self.assertEqual(r, 3145728)
         self.assertEqual(w, 295)
 
         params = zstd.get_frame_parameters(dest.getvalue())
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.dict_id, 0)
         self.assertFalse(params.has_checksum)
 
         # Writing content size and checksum works.
-        cctx = zstd.ZstdCompressor(threads=2, write_content_size=True,
-                                   write_checksum=True)
+        cctx = zstd.ZstdCompressor(threads=2, write_checksum=True)
         dest = io.BytesIO()
         source.seek(0)
         cctx.copy_stream(source, dest, size=len(source.getvalue()))
@@ -448,31 +552,227 @@
         self.assertEqual(params.dict_id, 0)
         self.assertTrue(params.has_checksum)
 
+    def test_bad_size(self):
+        source = io.BytesIO()
+        source.write(b'a' * 32768)
+        source.write(b'b' * 32768)
+        source.seek(0)
 
-def compress(data, level):
-    buffer = io.BytesIO()
-    cctx = zstd.ZstdCompressor(level=level)
-    with cctx.write_to(buffer) as compressor:
-        compressor.write(data)
-    return buffer.getvalue()
+        dest = io.BytesIO()
+
+        cctx = zstd.ZstdCompressor()
+
+        with self.assertRaisesRegexp(zstd.ZstdError, 'Src size is incorrect'):
+            cctx.copy_stream(source, dest, size=42)
+
+        # Try another operation on this compressor.
+        source.seek(0)
+        dest = io.BytesIO()
+        cctx.copy_stream(source, dest)
 
 
 @make_cffi
-class TestCompressor_write_to(unittest.TestCase):
+class TestCompressor_stream_reader(unittest.TestCase):
+    def test_context_manager(self):
+        cctx = zstd.ZstdCompressor()
+
+        reader = cctx.stream_reader(b'foo' * 60)
+        with self.assertRaisesRegexp(zstd.ZstdError, 'read\(\) must be called from an active'):
+            reader.read(10)
+
+        with cctx.stream_reader(b'foo') as reader:
+            with self.assertRaisesRegexp(ValueError, 'cannot __enter__ multiple times'):
+                with reader as reader2:
+                    pass
+
+    def test_not_implemented(self):
+        cctx = zstd.ZstdCompressor()
+
+        with cctx.stream_reader(b'foo' * 60) as reader:
+            with self.assertRaises(io.UnsupportedOperation):
+                reader.readline()
+
+            with self.assertRaises(io.UnsupportedOperation):
+                reader.readlines()
+
+            # This could probably be implemented someday.
+            with self.assertRaises(NotImplementedError):
+                reader.readall()
+
+            with self.assertRaises(io.UnsupportedOperation):
+                iter(reader)
+
+            with self.assertRaises(io.UnsupportedOperation):
+                next(reader)
+
+            with self.assertRaises(OSError):
+                reader.writelines([])
+
+            with self.assertRaises(OSError):
+                reader.write(b'foo')
+
+    def test_constant_methods(self):
+        cctx = zstd.ZstdCompressor()
+
+        with cctx.stream_reader(b'boo') as reader:
+            self.assertTrue(reader.readable())
+            self.assertFalse(reader.writable())
+            self.assertFalse(reader.seekable())
+            self.assertFalse(reader.isatty())
+            self.assertIsNone(reader.flush())
+
+    def test_read_closed(self):
+        cctx = zstd.ZstdCompressor()
+
+        with cctx.stream_reader(b'foo' * 60) as reader:
+            reader.close()
+            with self.assertRaisesRegexp(ValueError, 'stream is closed'):
+                reader.read(10)
+
+    def test_read_bad_size(self):
+        cctx = zstd.ZstdCompressor()
+
+        with cctx.stream_reader(b'foo') as reader:
+            with self.assertRaisesRegexp(ValueError, 'cannot read negative or size 0 amounts'):
+                reader.read(-1)
+
+            with self.assertRaisesRegexp(ValueError, 'cannot read negative or size 0 amounts'):
+                reader.read(0)
+
+    def test_read_buffer(self):
+        cctx = zstd.ZstdCompressor()
+
+        source = b''.join([b'foo' * 60, b'bar' * 60, b'baz' * 60])
+        frame = cctx.compress(source)
+
+        with cctx.stream_reader(source) as reader:
+            self.assertEqual(reader.tell(), 0)
+
+            # We should get entire frame in one read.
+            result = reader.read(8192)
+            self.assertEqual(result, frame)
+            self.assertEqual(reader.tell(), len(result))
+            self.assertEqual(reader.read(), b'')
+            self.assertEqual(reader.tell(), len(result))
+
+    def test_read_buffer_small_chunks(self):
+        cctx = zstd.ZstdCompressor()
+
+        source = b'foo' * 60
+        chunks = []
+
+        with cctx.stream_reader(source) as reader:
+            self.assertEqual(reader.tell(), 0)
+
+            while True:
+                chunk = reader.read(1)
+                if not chunk:
+                    break
+
+                chunks.append(chunk)
+                self.assertEqual(reader.tell(), sum(map(len, chunks)))
+
+        self.assertEqual(b''.join(chunks), cctx.compress(source))
+
+    def test_read_stream(self):
+        cctx = zstd.ZstdCompressor()
+
+        source = b''.join([b'foo' * 60, b'bar' * 60, b'baz' * 60])
+        frame = cctx.compress(source)
+
+        with cctx.stream_reader(io.BytesIO(source), size=len(source)) as reader:
+            self.assertEqual(reader.tell(), 0)
+
+            chunk = reader.read(8192)
+            self.assertEqual(chunk, frame)
+            self.assertEqual(reader.tell(), len(chunk))
+            self.assertEqual(reader.read(), b'')
+            self.assertEqual(reader.tell(), len(chunk))
+
+    def test_read_stream_small_chunks(self):
+        cctx = zstd.ZstdCompressor()
+
+        source = b'foo' * 60
+        chunks = []
+
+        with cctx.stream_reader(io.BytesIO(source), size=len(source)) as reader:
+            self.assertEqual(reader.tell(), 0)
+
+            while True:
+                chunk = reader.read(1)
+                if not chunk:
+                    break
+
+                chunks.append(chunk)
+                self.assertEqual(reader.tell(), sum(map(len, chunks)))
+
+        self.assertEqual(b''.join(chunks), cctx.compress(source))
+
+    def test_read_after_exit(self):
+        cctx = zstd.ZstdCompressor()
+
+        with cctx.stream_reader(b'foo' * 60) as reader:
+            while reader.read(8192):
+                pass
+
+        with self.assertRaisesRegexp(zstd.ZstdError, 'read\(\) must be called from an active'):
+            reader.read(10)
+
+    def test_bad_size(self):
+        cctx = zstd.ZstdCompressor()
+
+        source = io.BytesIO(b'foobar')
+
+        with cctx.stream_reader(source, size=2) as reader:
+            with self.assertRaisesRegexp(zstd.ZstdError, 'Src size is incorrect'):
+                reader.read(10)
+
+        # Try another compression operation.
+        with cctx.stream_reader(source, size=42):
+            pass
+
+
+@make_cffi
+class TestCompressor_stream_writer(unittest.TestCase):
     def test_empty(self):
-        result = compress(b'', 1)
+        buffer = io.BytesIO()
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
+        with cctx.stream_writer(buffer) as compressor:
+            compressor.write(b'')
+
+        result = buffer.getvalue()
         self.assertEqual(result, b'\x28\xb5\x2f\xfd\x00\x48\x01\x00\x00')
 
         params = zstd.get_frame_parameters(result)
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 524288)
         self.assertEqual(params.dict_id, 0)
         self.assertFalse(params.has_checksum)
 
+    def test_input_types(self):
+        expected = b'\x28\xb5\x2f\xfd\x00\x48\x19\x00\x00\x66\x6f\x6f'
+        cctx = zstd.ZstdCompressor(level=1)
+
+        mutable_array = bytearray(3)
+        mutable_array[:] = b'foo'
+
+        sources = [
+            memoryview(b'foo'),
+            bytearray(b'foo'),
+            mutable_array,
+        ]
+
+        for source in sources:
+            buffer = io.BytesIO()
+            with cctx.stream_writer(buffer) as compressor:
+                compressor.write(source)
+
+            self.assertEqual(buffer.getvalue(), expected)
+
     def test_multiple_compress(self):
         buffer = io.BytesIO()
         cctx = zstd.ZstdCompressor(level=5)
-        with cctx.write_to(buffer) as compressor:
+        with cctx.stream_writer(buffer) as compressor:
             self.assertEqual(compressor.write(b'foo'), 0)
             self.assertEqual(compressor.write(b'bar'), 0)
             self.assertEqual(compressor.write(b'x' * 8192), 0)
@@ -491,35 +791,40 @@
 
         d = zstd.train_dictionary(8192, samples)
 
+        h = hashlib.sha1(d.as_bytes()).hexdigest()
+        self.assertEqual(h, '3040faa0ddc37d50e71a4dd28052cb8db5d9d027')
+
         buffer = io.BytesIO()
         cctx = zstd.ZstdCompressor(level=9, dict_data=d)
-        with cctx.write_to(buffer) as compressor:
+        with cctx.stream_writer(buffer) as compressor:
             self.assertEqual(compressor.write(b'foo'), 0)
             self.assertEqual(compressor.write(b'bar'), 0)
-            self.assertEqual(compressor.write(b'foo' * 16384), 634)
+            self.assertEqual(compressor.write(b'foo' * 16384), 0)
 
         compressed = buffer.getvalue()
 
         params = zstd.get_frame_parameters(compressed)
-        self.assertEqual(params.content_size, 0)
-        self.assertEqual(params.window_size, 1024)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
+        self.assertEqual(params.window_size, 2097152)
         self.assertEqual(params.dict_id, d.dict_id())
         self.assertFalse(params.has_checksum)
-
-        self.assertEqual(compressed[0:32],
-                         b'\x28\xb5\x2f\xfd\x03\x00\x55\x7b\x6b\x5e\x54\x00'
-                         b'\x00\x00\x02\xfc\xf4\xa5\xba\x23\x3f\x85\xb3\x54'
-                         b'\x00\x00\x18\x6f\x6f\x66\x01\x00')
-
-        h = hashlib.sha1(compressed).hexdigest()
-        self.assertEqual(h, '1c5bcd25181bcd8c1a73ea8773323e0056129f92')
+        self.assertEqual(compressed,
+                         b'\x28\xb5\x2f\xfd\x03\x58\x06\x59\xb5\x52\x5d\x00'
+                         b'\x00\x00\x02\xfc\x3d\x3f\xd9\xb0\x51\x03\x45\x89')
 
     def test_compression_params(self):
-        params = zstd.CompressionParameters(20, 6, 12, 5, 4, 10, zstd.STRATEGY_FAST)
+        params = zstd.ZstdCompressionParameters(
+            window_log=20,
+            chain_log=6,
+            hash_log=12,
+            min_match=5,
+            search_log=4,
+            target_length=10,
+            compression_strategy=zstd.STRATEGY_FAST)
 
         buffer = io.BytesIO()
         cctx = zstd.ZstdCompressor(compression_params=params)
-        with cctx.write_to(buffer) as compressor:
+        with cctx.stream_writer(buffer) as compressor:
             self.assertEqual(compressor.write(b'foo'), 0)
             self.assertEqual(compressor.write(b'bar'), 0)
             self.assertEqual(compressor.write(b'foobar' * 16384), 0)
@@ -527,29 +832,29 @@
         compressed = buffer.getvalue()
 
         params = zstd.get_frame_parameters(compressed)
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 1048576)
         self.assertEqual(params.dict_id, 0)
         self.assertFalse(params.has_checksum)
 
         h = hashlib.sha1(compressed).hexdigest()
-        self.assertEqual(h, '1ae31f270ed7de14235221a604b31ecd517ebd99')
+        self.assertEqual(h, '2a8111d72eb5004cdcecbdac37da9f26720d30ef')
 
     def test_write_checksum(self):
         no_checksum = io.BytesIO()
         cctx = zstd.ZstdCompressor(level=1)
-        with cctx.write_to(no_checksum) as compressor:
+        with cctx.stream_writer(no_checksum) as compressor:
             self.assertEqual(compressor.write(b'foobar'), 0)
 
         with_checksum = io.BytesIO()
         cctx = zstd.ZstdCompressor(level=1, write_checksum=True)
-        with cctx.write_to(with_checksum) as compressor:
+        with cctx.stream_writer(with_checksum) as compressor:
             self.assertEqual(compressor.write(b'foobar'), 0)
 
         no_params = zstd.get_frame_parameters(no_checksum.getvalue())
         with_params = zstd.get_frame_parameters(with_checksum.getvalue())
-        self.assertEqual(no_params.content_size, 0)
-        self.assertEqual(with_params.content_size, 0)
+        self.assertEqual(no_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
+        self.assertEqual(with_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(no_params.dict_id, 0)
         self.assertEqual(with_params.dict_id, 0)
         self.assertFalse(no_params.has_checksum)
@@ -560,13 +865,13 @@
 
     def test_write_content_size(self):
         no_size = io.BytesIO()
-        cctx = zstd.ZstdCompressor(level=1)
-        with cctx.write_to(no_size) as compressor:
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
+        with cctx.stream_writer(no_size) as compressor:
             self.assertEqual(compressor.write(b'foobar' * 256), 0)
 
         with_size = io.BytesIO()
-        cctx = zstd.ZstdCompressor(level=1, write_content_size=True)
-        with cctx.write_to(with_size) as compressor:
+        cctx = zstd.ZstdCompressor(level=1)
+        with cctx.stream_writer(with_size) as compressor:
             self.assertEqual(compressor.write(b'foobar' * 256), 0)
 
         # Source size is not known in streaming mode, so header not
@@ -576,12 +881,12 @@
 
         # Declaring size will write the header.
         with_size = io.BytesIO()
-        with cctx.write_to(with_size, size=len(b'foobar' * 256)) as compressor:
+        with cctx.stream_writer(with_size, size=len(b'foobar' * 256)) as compressor:
             self.assertEqual(compressor.write(b'foobar' * 256), 0)
 
         no_params = zstd.get_frame_parameters(no_size.getvalue())
         with_params = zstd.get_frame_parameters(with_size.getvalue())
-        self.assertEqual(no_params.content_size, 0)
+        self.assertEqual(no_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(with_params.content_size, 1536)
         self.assertEqual(no_params.dict_id, 0)
         self.assertEqual(with_params.dict_id, 0)
@@ -602,18 +907,22 @@
 
         with_dict_id = io.BytesIO()
         cctx = zstd.ZstdCompressor(level=1, dict_data=d)
-        with cctx.write_to(with_dict_id) as compressor:
+        with cctx.stream_writer(with_dict_id) as compressor:
             self.assertEqual(compressor.write(b'foobarfoobar'), 0)
 
+        self.assertEqual(with_dict_id.getvalue()[4:5], b'\x03')
+
         cctx = zstd.ZstdCompressor(level=1, dict_data=d, write_dict_id=False)
         no_dict_id = io.BytesIO()
-        with cctx.write_to(no_dict_id) as compressor:
+        with cctx.stream_writer(no_dict_id) as compressor:
             self.assertEqual(compressor.write(b'foobarfoobar'), 0)
 
+        self.assertEqual(no_dict_id.getvalue()[4:5], b'\x00')
+
         no_params = zstd.get_frame_parameters(no_dict_id.getvalue())
         with_params = zstd.get_frame_parameters(with_dict_id.getvalue())
-        self.assertEqual(no_params.content_size, 0)
-        self.assertEqual(with_params.content_size, 0)
+        self.assertEqual(no_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
+        self.assertEqual(with_params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(no_params.dict_id, 0)
         self.assertEqual(with_params.dict_id, d.dict_id())
         self.assertFalse(no_params.has_checksum)
@@ -625,7 +934,8 @@
     def test_memory_size(self):
         cctx = zstd.ZstdCompressor(level=3)
         buffer = io.BytesIO()
-        with cctx.write_to(buffer) as compressor:
+        with cctx.stream_writer(buffer) as compressor:
+            compressor.write(b'foo')
             size = compressor.memory_size()
 
         self.assertGreater(size, 100000)
@@ -633,7 +943,7 @@
     def test_write_size(self):
         cctx = zstd.ZstdCompressor(level=3)
         dest = OpCountingBytesIO()
-        with cctx.write_to(dest, write_size=1) as compressor:
+        with cctx.stream_writer(dest, write_size=1) as compressor:
             self.assertEqual(compressor.write(b'foo'), 0)
             self.assertEqual(compressor.write(b'bar'), 0)
             self.assertEqual(compressor.write(b'foobar'), 0)
@@ -643,7 +953,7 @@
     def test_flush_repeated(self):
         cctx = zstd.ZstdCompressor(level=3)
         dest = OpCountingBytesIO()
-        with cctx.write_to(dest) as compressor:
+        with cctx.stream_writer(dest) as compressor:
             self.assertEqual(compressor.write(b'foo'), 0)
             self.assertEqual(dest._write_count, 0)
             self.assertEqual(compressor.flush(), 12)
@@ -659,7 +969,7 @@
     def test_flush_empty_block(self):
         cctx = zstd.ZstdCompressor(level=3, write_checksum=True)
         dest = OpCountingBytesIO()
-        with cctx.write_to(dest) as compressor:
+        with cctx.stream_writer(dest) as compressor:
             self.assertEqual(compressor.write(b'foobar' * 8192), 0)
             count = dest._write_count
             offset = dest.tell()
@@ -680,50 +990,89 @@
     def test_multithreaded(self):
         dest = io.BytesIO()
         cctx = zstd.ZstdCompressor(threads=2)
-        with cctx.write_to(dest) as compressor:
+        with cctx.stream_writer(dest) as compressor:
             compressor.write(b'a' * 1048576)
             compressor.write(b'b' * 1048576)
             compressor.write(b'c' * 1048576)
 
         self.assertEqual(len(dest.getvalue()), 295)
 
+    def test_tell(self):
+        dest = io.BytesIO()
+        cctx = zstd.ZstdCompressor()
+        with cctx.stream_writer(dest) as compressor:
+            self.assertEqual(compressor.tell(), 0)
+
+            for i in range(256):
+                compressor.write(b'foo' * (i + 1))
+                self.assertEqual(compressor.tell(), dest.tell())
+
+    def test_bad_size(self):
+        cctx = zstd.ZstdCompressor()
+
+        dest = io.BytesIO()
+
+        with self.assertRaisesRegexp(zstd.ZstdError, 'Src size is incorrect'):
+            with cctx.stream_writer(dest, size=2) as compressor:
+                compressor.write(b'foo')
+
+        # Test another operation.
+        with cctx.stream_writer(dest, size=42):
+            pass
+
+    def test_tarfile_compat(self):
+        raise unittest.SkipTest('not yet fully working')
+
+        dest = io.BytesIO()
+        cctx = zstd.ZstdCompressor()
+        with cctx.stream_writer(dest) as compressor:
+            with tarfile.open('tf', mode='w', fileobj=compressor) as tf:
+                tf.add(__file__, 'test_compressor.py')
+
+        dest.seek(0)
+
+        dctx = zstd.ZstdDecompressor()
+        with dctx.stream_reader(dest) as reader:
+            with tarfile.open(mode='r:', fileobj=reader) as tf:
+                for member in tf:
+                    self.assertEqual(member.name, 'test_compressor.py')
 
 @make_cffi
-class TestCompressor_read_from(unittest.TestCase):
+class TestCompressor_read_to_iter(unittest.TestCase):
     def test_type_validation(self):
         cctx = zstd.ZstdCompressor()
 
         # Object with read() works.
-        for chunk in cctx.read_from(io.BytesIO()):
+        for chunk in cctx.read_to_iter(io.BytesIO()):
             pass
 
         # Buffer protocol works.
-        for chunk in cctx.read_from(b'foobar'):
+        for chunk in cctx.read_to_iter(b'foobar'):
             pass
 
         with self.assertRaisesRegexp(ValueError, 'must pass an object with a read'):
-            for chunk in cctx.read_from(True):
+            for chunk in cctx.read_to_iter(True):
                 pass
 
     def test_read_empty(self):
-        cctx = zstd.ZstdCompressor(level=1)
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
 
         source = io.BytesIO()
-        it = cctx.read_from(source)
+        it = cctx.read_to_iter(source)
         chunks = list(it)
         self.assertEqual(len(chunks), 1)
         compressed = b''.join(chunks)
         self.assertEqual(compressed, b'\x28\xb5\x2f\xfd\x00\x48\x01\x00\x00')
 
         # And again with the buffer protocol.
-        it = cctx.read_from(b'')
+        it = cctx.read_to_iter(b'')
         chunks = list(it)
         self.assertEqual(len(chunks), 1)
         compressed2 = b''.join(chunks)
         self.assertEqual(compressed2, compressed)
 
     def test_read_large(self):
-        cctx = zstd.ZstdCompressor(level=1)
+        cctx = zstd.ZstdCompressor(level=1, write_content_size=False)
 
         source = io.BytesIO()
         source.write(b'f' * zstd.COMPRESSION_RECOMMENDED_INPUT_SIZE)
@@ -732,7 +1081,7 @@
 
         # Creating an iterator should not perform any compression until
         # first read.
-        it = cctx.read_from(source, size=len(source.getvalue()))
+        it = cctx.read_to_iter(source, size=len(source.getvalue()))
         self.assertEqual(source.tell(), 0)
 
         # We should have exactly 2 output chunks.
@@ -758,21 +1107,28 @@
         self.assertEqual(b''.join(chunks), cctx.compress(source.getvalue()))
 
         params = zstd.get_frame_parameters(b''.join(chunks))
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 262144)
         self.assertEqual(params.dict_id, 0)
         self.assertFalse(params.has_checksum)
 
         # Now check the buffer protocol.
-        it = cctx.read_from(source.getvalue())
+        it = cctx.read_to_iter(source.getvalue())
         chunks = list(it)
         self.assertEqual(len(chunks), 2)
+
+        params = zstd.get_frame_parameters(b''.join(chunks))
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
+        #self.assertEqual(params.window_size, 262144)
+        self.assertEqual(params.dict_id, 0)
+        self.assertFalse(params.has_checksum)
+
         self.assertEqual(b''.join(chunks), cctx.compress(source.getvalue()))
 
     def test_read_write_size(self):
         source = OpCountingBytesIO(b'foobarfoobar')
         cctx = zstd.ZstdCompressor(level=3)
-        for chunk in cctx.read_from(source, read_size=1, write_size=1):
+        for chunk in cctx.read_to_iter(source, read_size=1, write_size=1):
             self.assertEqual(len(chunk), 1)
 
         self.assertEqual(source._read_count, len(source.getvalue()) + 1)
@@ -786,17 +1142,22 @@
 
         cctx = zstd.ZstdCompressor(threads=2)
 
-        compressed = b''.join(cctx.read_from(source))
+        compressed = b''.join(cctx.read_to_iter(source))
         self.assertEqual(len(compressed), 295)
 
+    def test_bad_size(self):
+        cctx = zstd.ZstdCompressor()
+
+        source = io.BytesIO(b'a' * 42)
+
+        with self.assertRaisesRegexp(zstd.ZstdError, 'Src size is incorrect'):
+            b''.join(cctx.read_to_iter(source, size=2))
+
+        # Test another operation on errored compressor.
+        b''.join(cctx.read_to_iter(source))
+
 
 class TestCompressor_multi_compress_to_buffer(unittest.TestCase):
-    def test_multithreaded_unsupported(self):
-        cctx = zstd.ZstdCompressor(threads=2)
-
-        with self.assertRaisesRegexp(zstd.ZstdError, 'function cannot be called on ZstdCompressor configured for multi-threaded compression'):
-            cctx.multi_compress_to_buffer([b'foo'])
-
     def test_invalid_inputs(self):
         cctx = zstd.ZstdCompressor()
 
@@ -819,7 +1180,7 @@
             cctx.multi_compress_to_buffer([b'', b'', b''])
 
     def test_list_input(self):
-        cctx = zstd.ZstdCompressor(write_content_size=True, write_checksum=True)
+        cctx = zstd.ZstdCompressor(write_checksum=True)
 
         original = [b'foo' * 12, b'bar' * 6]
         frames = [cctx.compress(c) for c in original]
@@ -834,7 +1195,7 @@
         self.assertEqual(b[1].tobytes(), frames[1])
 
     def test_buffer_with_segments_input(self):
-        cctx = zstd.ZstdCompressor(write_content_size=True, write_checksum=True)
+        cctx = zstd.ZstdCompressor(write_checksum=True)
 
         original = [b'foo' * 4, b'bar' * 6]
         frames = [cctx.compress(c) for c in original]
@@ -852,7 +1213,7 @@
         self.assertEqual(result[1].tobytes(), frames[1])
 
     def test_buffer_with_segments_collection_input(self):
-        cctx = zstd.ZstdCompressor(write_content_size=True, write_checksum=True)
+        cctx = zstd.ZstdCompressor(write_checksum=True)
 
         original = [
             b'foo1',
@@ -886,10 +1247,10 @@
     def test_multiple_threads(self):
         # threads argument will cause multi-threaded ZSTD APIs to be used, which will
         # make output different.
-        refcctx = zstd.ZstdCompressor(write_content_size=True, write_checksum=True)
+        refcctx = zstd.ZstdCompressor(write_checksum=True)
         reference = [refcctx.compress(b'x' * 64), refcctx.compress(b'y' * 64)]
 
-        cctx = zstd.ZstdCompressor(write_content_size=True, write_checksum=True)
+        cctx = zstd.ZstdCompressor(write_checksum=True)
 
         frames = []
         frames.extend(b'x' * 64 for i in range(256))
--- a/contrib/python-zstandard/tests/test_compressor_fuzzing.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_compressor_fuzzing.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,6 @@
 import io
 import os
-
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
+import unittest
 
 try:
     import hypothesis
@@ -12,7 +8,7 @@
 except ImportError:
     raise unittest.SkipTest('hypothesis not available')
 
-import zstd
+import zstandard as zstd
 
 from . common import (
     make_cffi,
@@ -22,7 +18,57 @@
 
 @unittest.skipUnless('ZSTD_SLOW_TESTS' in os.environ, 'ZSTD_SLOW_TESTS not set')
 @make_cffi
-class TestCompressor_write_to_fuzzing(unittest.TestCase):
+class TestCompressor_stream_reader_fuzzing(unittest.TestCase):
+    @hypothesis.given(original=strategies.sampled_from(random_input_data()),
+                      level=strategies.integers(min_value=1, max_value=5),
+                      source_read_size=strategies.integers(1, 16384),
+                      read_sizes=strategies.data())
+    def test_stream_source_read_variance(self, original, level, source_read_size,
+                                         read_sizes):
+        refctx = zstd.ZstdCompressor(level=level)
+        ref_frame = refctx.compress(original)
+
+        cctx = zstd.ZstdCompressor(level=level)
+        with cctx.stream_reader(io.BytesIO(original), size=len(original),
+                                read_size=source_read_size) as reader:
+            chunks = []
+            while True:
+                read_size = read_sizes.draw(strategies.integers(1, 16384))
+                chunk = reader.read(read_size)
+
+                if not chunk:
+                    break
+                chunks.append(chunk)
+
+        self.assertEqual(b''.join(chunks), ref_frame)
+
+    @hypothesis.given(original=strategies.sampled_from(random_input_data()),
+                      level=strategies.integers(min_value=1, max_value=5),
+                      source_read_size=strategies.integers(1, 16384),
+                      read_sizes=strategies.data())
+    def test_buffer_source_read_variance(self, original, level, source_read_size,
+                                         read_sizes):
+
+        refctx = zstd.ZstdCompressor(level=level)
+        ref_frame = refctx.compress(original)
+
+        cctx = zstd.ZstdCompressor(level=level)
+        with cctx.stream_reader(original, size=len(original),
+                                read_size=source_read_size) as reader:
+            chunks = []
+            while True:
+                read_size = read_sizes.draw(strategies.integers(1, 16384))
+                chunk = reader.read(read_size)
+                if not chunk:
+                    break
+                chunks.append(chunk)
+
+        self.assertEqual(b''.join(chunks), ref_frame)
+
+
+@unittest.skipUnless('ZSTD_SLOW_TESTS' in os.environ, 'ZSTD_SLOW_TESTS not set')
+@make_cffi
+class TestCompressor_stream_writer_fuzzing(unittest.TestCase):
     @hypothesis.given(original=strategies.sampled_from(random_input_data()),
                         level=strategies.integers(min_value=1, max_value=5),
                         write_size=strategies.integers(min_value=1, max_value=1048576))
@@ -32,7 +78,7 @@
 
         cctx = zstd.ZstdCompressor(level=level)
         b = io.BytesIO()
-        with cctx.write_to(b, size=len(original), write_size=write_size) as compressor:
+        with cctx.stream_writer(b, size=len(original), write_size=write_size) as compressor:
             compressor.write(original)
 
         self.assertEqual(b.getvalue(), ref_frame)
@@ -62,13 +108,12 @@
 @unittest.skipUnless('ZSTD_SLOW_TESTS' in os.environ, 'ZSTD_SLOW_TESTS not set')
 @make_cffi
 class TestCompressor_compressobj_fuzzing(unittest.TestCase):
+    @hypothesis.settings(
+        suppress_health_check=[hypothesis.HealthCheck.large_base_example])
     @hypothesis.given(original=strategies.sampled_from(random_input_data()),
                       level=strategies.integers(min_value=1, max_value=5),
-                      chunk_sizes=strategies.streaming(
-                          strategies.integers(min_value=1, max_value=4096)))
+                      chunk_sizes=strategies.data())
     def test_random_input_sizes(self, original, level, chunk_sizes):
-        chunk_sizes = iter(chunk_sizes)
-
         refctx = zstd.ZstdCompressor(level=level)
         ref_frame = refctx.compress(original)
 
@@ -78,7 +123,7 @@
         chunks = []
         i = 0
         while True:
-            chunk_size = next(chunk_sizes)
+            chunk_size = chunk_sizes.draw(strategies.integers(1, 4096))
             source = original[i:i + chunk_size]
             if not source:
                 break
@@ -93,7 +138,7 @@
 
 @unittest.skipUnless('ZSTD_SLOW_TESTS' in os.environ, 'ZSTD_SLOW_TESTS not set')
 @make_cffi
-class TestCompressor_read_from_fuzzing(unittest.TestCase):
+class TestCompressor_read_to_iter_fuzzing(unittest.TestCase):
     @hypothesis.given(original=strategies.sampled_from(random_input_data()),
                       level=strategies.integers(min_value=1, max_value=5),
                       read_size=strategies.integers(min_value=1, max_value=4096),
@@ -105,8 +150,9 @@
         source = io.BytesIO(original)
 
         cctx = zstd.ZstdCompressor(level=level)
-        chunks = list(cctx.read_from(source, size=len(original), read_size=read_size,
-                                     write_size=write_size))
+        chunks = list(cctx.read_to_iter(source, size=len(original),
+                                        read_size=read_size,
+                                        write_size=write_size))
 
         self.assertEqual(b''.join(chunks), ref_frame)
 
@@ -125,7 +171,6 @@
             kwargs['dict_data'] = zstd.ZstdCompressionDict(original[0])
 
         cctx = zstd.ZstdCompressor(level=1,
-                                   write_content_size=True,
                                    write_checksum=True,
                                    **kwargs)
 
--- a/contrib/python-zstandard/tests/test_data_structures.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_data_structures.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,9 +1,7 @@
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
+import sys
+import unittest
 
-import zstd
+import zstandard as zstd
 
 from . common import (
     make_cffi,
@@ -12,52 +10,104 @@
 
 @make_cffi
 class TestCompressionParameters(unittest.TestCase):
-    def test_init_bad_arg_type(self):
-        with self.assertRaises(TypeError):
-            zstd.CompressionParameters()
-
-        with self.assertRaises(TypeError):
-            zstd.CompressionParameters(0, 1)
+    def test_bounds(self):
+        zstd.ZstdCompressionParameters(window_log=zstd.WINDOWLOG_MIN,
+                                       chain_log=zstd.CHAINLOG_MIN,
+                                       hash_log=zstd.HASHLOG_MIN,
+                                       search_log=zstd.SEARCHLOG_MIN,
+                                       min_match=zstd.SEARCHLENGTH_MIN + 1,
+                                       target_length=zstd.TARGETLENGTH_MIN,
+                                       compression_strategy=zstd.STRATEGY_FAST)
 
-    def test_bounds(self):
-        zstd.CompressionParameters(zstd.WINDOWLOG_MIN,
-                                   zstd.CHAINLOG_MIN,
-                                   zstd.HASHLOG_MIN,
-                                   zstd.SEARCHLOG_MIN,
-                                   zstd.SEARCHLENGTH_MIN + 1,
-                                   zstd.TARGETLENGTH_MIN,
-                                   zstd.STRATEGY_FAST)
+        zstd.ZstdCompressionParameters(window_log=zstd.WINDOWLOG_MAX,
+                                       chain_log=zstd.CHAINLOG_MAX,
+                                       hash_log=zstd.HASHLOG_MAX,
+                                       search_log=zstd.SEARCHLOG_MAX,
+                                       min_match=zstd.SEARCHLENGTH_MAX - 1,
+                                       compression_strategy=zstd.STRATEGY_BTULTRA)
 
-        zstd.CompressionParameters(zstd.WINDOWLOG_MAX,
-                                   zstd.CHAINLOG_MAX,
-                                   zstd.HASHLOG_MAX,
-                                   zstd.SEARCHLOG_MAX,
-                                   zstd.SEARCHLENGTH_MAX - 1,
-                                   zstd.TARGETLENGTH_MAX,
-                                   zstd.STRATEGY_BTOPT)
-
-    def test_get_compression_parameters(self):
-        p = zstd.get_compression_parameters(1)
+    def test_from_level(self):
+        p = zstd.ZstdCompressionParameters.from_level(1)
         self.assertIsInstance(p, zstd.CompressionParameters)
 
         self.assertEqual(p.window_log, 19)
 
+        p = zstd.ZstdCompressionParameters.from_level(-4)
+        self.assertEqual(p.window_log, 19)
+        self.assertEqual(p.compress_literals, 0)
+
     def test_members(self):
-        p = zstd.CompressionParameters(10, 6, 7, 4, 5, 8, 1)
+        p = zstd.ZstdCompressionParameters(window_log=10,
+                                           chain_log=6,
+                                           hash_log=7,
+                                           search_log=4,
+                                           min_match=5,
+                                           target_length=8,
+                                           compression_strategy=1)
         self.assertEqual(p.window_log, 10)
         self.assertEqual(p.chain_log, 6)
         self.assertEqual(p.hash_log, 7)
         self.assertEqual(p.search_log, 4)
-        self.assertEqual(p.search_length, 5)
+        self.assertEqual(p.min_match, 5)
         self.assertEqual(p.target_length, 8)
-        self.assertEqual(p.strategy, 1)
+        self.assertEqual(p.compression_strategy, 1)
+
+        p = zstd.ZstdCompressionParameters(compression_level=2)
+        self.assertEqual(p.compression_level, 2)
+
+        p = zstd.ZstdCompressionParameters(threads=4)
+        self.assertEqual(p.threads, 4)
+
+        p = zstd.ZstdCompressionParameters(threads=2, job_size=1048576,
+                                       overlap_size_log=6)
+        self.assertEqual(p.threads, 2)
+        self.assertEqual(p.job_size, 1048576)
+        self.assertEqual(p.overlap_size_log, 6)
+
+        p = zstd.ZstdCompressionParameters(compression_level=2)
+        self.assertEqual(p.compress_literals, 1)
+
+        p = zstd.ZstdCompressionParameters(compress_literals=False)
+        self.assertEqual(p.compress_literals, 0)
+
+        p = zstd.ZstdCompressionParameters(compression_level=-1)
+        self.assertEqual(p.compression_level, -1)
+        self.assertEqual(p.compress_literals, 0)
+
+        p = zstd.ZstdCompressionParameters(compression_level=-2, compress_literals=True)
+        self.assertEqual(p.compression_level, -2)
+        self.assertEqual(p.compress_literals, 1)
+
+        p = zstd.ZstdCompressionParameters(force_max_window=True)
+        self.assertEqual(p.force_max_window, 1)
+
+        p = zstd.ZstdCompressionParameters(enable_ldm=True)
+        self.assertEqual(p.enable_ldm, 1)
+
+        p = zstd.ZstdCompressionParameters(ldm_hash_log=7)
+        self.assertEqual(p.ldm_hash_log, 7)
+
+        p = zstd.ZstdCompressionParameters(ldm_min_match=6)
+        self.assertEqual(p.ldm_min_match, 6)
+
+        p = zstd.ZstdCompressionParameters(ldm_bucket_size_log=7)
+        self.assertEqual(p.ldm_bucket_size_log, 7)
+
+        p = zstd.ZstdCompressionParameters(ldm_hash_every_log=8)
+        self.assertEqual(p.ldm_hash_every_log, 8)
 
     def test_estimated_compression_context_size(self):
-        p = zstd.CompressionParameters(20, 16, 17,  1,  5, 16, zstd.STRATEGY_DFAST)
+        p = zstd.ZstdCompressionParameters(window_log=20,
+                                           chain_log=16,
+                                           hash_log=17,
+                                           search_log=1,
+                                           min_match=5,
+                                           target_length=16,
+                                           compression_strategy=zstd.STRATEGY_DFAST)
 
         # 32-bit has slightly different values from 64-bit.
-        self.assertAlmostEqual(p.estimated_compression_context_size(), 1287076,
-                               delta=110)
+        self.assertAlmostEqual(p.estimated_compression_context_size(), 1294072,
+                               delta=250)
 
 
 @make_cffi
@@ -66,8 +116,18 @@
         with self.assertRaises(TypeError):
             zstd.get_frame_parameters(None)
 
-        with self.assertRaises(TypeError):
-            zstd.get_frame_parameters(u'foobarbaz')
+        # Python 3 doesn't appear to convert unicode to Py_buffer.
+        if sys.version_info[0] >= 3:
+            with self.assertRaises(TypeError):
+                zstd.get_frame_parameters(u'foobarbaz')
+        else:
+            # CPython will convert unicode to Py_buffer. But CFFI won't.
+            if zstd.backend == 'cffi':
+                with self.assertRaises(TypeError):
+                    zstd.get_frame_parameters(u'foobarbaz')
+            else:
+                with self.assertRaises(zstd.ZstdError):
+                    zstd.get_frame_parameters(u'foobarbaz')
 
     def test_invalid_input_sizes(self):
         with self.assertRaisesRegexp(zstd.ZstdError, 'not enough data for frame'):
@@ -82,21 +142,21 @@
 
     def test_attributes(self):
         params = zstd.get_frame_parameters(zstd.FRAME_HEADER + b'\x00\x00')
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 1024)
         self.assertEqual(params.dict_id, 0)
         self.assertFalse(params.has_checksum)
 
         # Lowest 2 bits indicate a dictionary and length. Here, the dict id is 1 byte.
         params = zstd.get_frame_parameters(zstd.FRAME_HEADER + b'\x01\x00\xff')
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 1024)
         self.assertEqual(params.dict_id, 255)
         self.assertFalse(params.has_checksum)
 
         # Lowest 3rd bit indicates if checksum is present.
         params = zstd.get_frame_parameters(zstd.FRAME_HEADER + b'\x04\x00')
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 1024)
         self.assertEqual(params.dict_id, 0)
         self.assertTrue(params.has_checksum)
@@ -110,7 +170,7 @@
 
         # Window descriptor is 2nd byte after frame header.
         params = zstd.get_frame_parameters(zstd.FRAME_HEADER + b'\x00\x40')
-        self.assertEqual(params.content_size, 0)
+        self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
         self.assertEqual(params.window_size, 262144)
         self.assertEqual(params.dict_id, 0)
         self.assertFalse(params.has_checksum)
@@ -121,3 +181,22 @@
         self.assertEqual(params.window_size, 262144)
         self.assertEqual(params.dict_id, 15)
         self.assertTrue(params.has_checksum)
+
+    def test_input_types(self):
+        v = zstd.FRAME_HEADER + b'\x00\x00'
+
+        mutable_array = bytearray(len(v))
+        mutable_array[:] = v
+
+        sources = [
+            memoryview(v),
+            bytearray(v),
+            mutable_array,
+        ]
+
+        for source in sources:
+            params = zstd.get_frame_parameters(source)
+            self.assertEqual(params.content_size, zstd.CONTENTSIZE_UNKNOWN)
+            self.assertEqual(params.window_size, 1024)
+            self.assertEqual(params.dict_id, 0)
+            self.assertFalse(params.has_checksum)
--- a/contrib/python-zstandard/tests/test_data_structures_fuzzing.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_data_structures_fuzzing.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,7 @@
 import io
 import os
-
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
+import sys
+import unittest
 
 try:
     import hypothesis
@@ -12,7 +9,7 @@
 except ImportError:
     raise unittest.SkipTest('hypothesis not available')
 
-import zstd
+import zstandard as zstd
 
 from .common import (
     make_cffi,
@@ -28,16 +25,17 @@
 s_searchlog = strategies.integers(min_value=zstd.SEARCHLOG_MIN,
                                     max_value=zstd.SEARCHLOG_MAX)
 s_searchlength = strategies.integers(min_value=zstd.SEARCHLENGTH_MIN,
-                                        max_value=zstd.SEARCHLENGTH_MAX)
+                                     max_value=zstd.SEARCHLENGTH_MAX)
 s_targetlength = strategies.integers(min_value=zstd.TARGETLENGTH_MIN,
-                                        max_value=zstd.TARGETLENGTH_MAX)
+                                     max_value=2**32)
 s_strategy = strategies.sampled_from((zstd.STRATEGY_FAST,
                                         zstd.STRATEGY_DFAST,
                                         zstd.STRATEGY_GREEDY,
                                         zstd.STRATEGY_LAZY,
                                         zstd.STRATEGY_LAZY2,
                                         zstd.STRATEGY_BTLAZY2,
-                                        zstd.STRATEGY_BTOPT))
+                                        zstd.STRATEGY_BTOPT,
+                                        zstd.STRATEGY_BTULTRA))
 
 
 @make_cffi
@@ -47,24 +45,17 @@
                         s_searchlength, s_targetlength, s_strategy)
     def test_valid_init(self, windowlog, chainlog, hashlog, searchlog,
                         searchlength, targetlength, strategy):
-        # ZSTD_checkCParams moves the goal posts on us from what's advertised
-        # in the constants. So move along with them.
-        if searchlength == zstd.SEARCHLENGTH_MIN and strategy in (zstd.STRATEGY_FAST, zstd.STRATEGY_GREEDY):
-            searchlength += 1
-        elif searchlength == zstd.SEARCHLENGTH_MAX and strategy != zstd.STRATEGY_FAST:
-            searchlength -= 1
-
-        p = zstd.CompressionParameters(windowlog, chainlog, hashlog,
-                                        searchlog, searchlength,
-                                        targetlength, strategy)
-
-        cctx = zstd.ZstdCompressor(compression_params=p)
-        with cctx.write_to(io.BytesIO()):
-            pass
+        zstd.ZstdCompressionParameters(window_log=windowlog,
+                                       chain_log=chainlog,
+                                       hash_log=hashlog,
+                                       search_log=searchlog,
+                                       min_match=searchlength,
+                                       target_length=targetlength,
+                                       compression_strategy=strategy)
 
     @hypothesis.given(s_windowlog, s_chainlog, s_hashlog, s_searchlog,
                         s_searchlength, s_targetlength, s_strategy)
-    def test_estimate_compression_context_size(self, windowlog, chainlog,
+    def test_estimated_compression_context_size(self, windowlog, chainlog,
                                                 hashlog, searchlog,
                                                 searchlength, targetlength,
                                                 strategy):
@@ -73,7 +64,12 @@
         elif searchlength == zstd.SEARCHLENGTH_MAX and strategy != zstd.STRATEGY_FAST:
             searchlength -= 1
 
-        p = zstd.CompressionParameters(windowlog, chainlog, hashlog,
-                            searchlog, searchlength,
-                            targetlength, strategy)
-        size = zstd.estimate_compression_context_size(p)
+        p = zstd.ZstdCompressionParameters(window_log=windowlog,
+                                           chain_log=chainlog,
+                                           hash_log=hashlog,
+                                           search_log=searchlog,
+                                           min_match=searchlength,
+                                           target_length=targetlength,
+                                           compression_strategy=strategy)
+        size = p.estimated_compression_context_size()
+
--- a/contrib/python-zstandard/tests/test_decompressor.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_decompressor.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,16 +1,14 @@
 import io
+import os
 import random
 import struct
 import sys
+import unittest
 
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
-
-import zstd
+import zstandard as zstd
 
 from .common import (
+    generate_samples,
     make_cffi,
     OpCountingBytesIO,
 )
@@ -23,35 +21,124 @@
 
 
 @make_cffi
+class TestFrameHeaderSize(unittest.TestCase):
+    def test_empty(self):
+        with self.assertRaisesRegexp(
+            zstd.ZstdError, 'could not determine frame header size: Src size '
+                            'is incorrect'):
+            zstd.frame_header_size(b'')
+
+    def test_too_small(self):
+        with self.assertRaisesRegexp(
+            zstd.ZstdError, 'could not determine frame header size: Src size '
+                            'is incorrect'):
+            zstd.frame_header_size(b'foob')
+
+    def test_basic(self):
+        # It doesn't matter that it isn't a valid frame.
+        self.assertEqual(zstd.frame_header_size(b'long enough but no magic'), 6)
+
+
+@make_cffi
+class TestFrameContentSize(unittest.TestCase):
+    def test_empty(self):
+        with self.assertRaisesRegexp(zstd.ZstdError,
+                                     'error when determining content size'):
+            zstd.frame_content_size(b'')
+
+    def test_too_small(self):
+        with self.assertRaisesRegexp(zstd.ZstdError,
+                                     'error when determining content size'):
+            zstd.frame_content_size(b'foob')
+
+    def test_bad_frame(self):
+        with self.assertRaisesRegexp(zstd.ZstdError,
+                                     'error when determining content size'):
+            zstd.frame_content_size(b'invalid frame header')
+
+    def test_unknown(self):
+        cctx = zstd.ZstdCompressor(write_content_size=False)
+        frame = cctx.compress(b'foobar')
+
+        self.assertEqual(zstd.frame_content_size(frame), -1)
+
+    def test_empty(self):
+        cctx = zstd.ZstdCompressor()
+        frame = cctx.compress(b'')
+
+        self.assertEqual(zstd.frame_content_size(frame), 0)
+
+    def test_basic(self):
+        cctx = zstd.ZstdCompressor()
+        frame = cctx.compress(b'foobar')
+
+        self.assertEqual(zstd.frame_content_size(frame), 6)
+
+
+@make_cffi
+class TestDecompressor(unittest.TestCase):
+    def test_memory_size(self):
+        dctx = zstd.ZstdDecompressor()
+
+        self.assertGreater(dctx.memory_size(), 100)
+
+
+@make_cffi
 class TestDecompressor_decompress(unittest.TestCase):
     def test_empty_input(self):
         dctx = zstd.ZstdDecompressor()
 
-        with self.assertRaisesRegexp(zstd.ZstdError, 'input data invalid'):
+        with self.assertRaisesRegexp(zstd.ZstdError, 'error determining content size from frame header'):
             dctx.decompress(b'')
 
     def test_invalid_input(self):
         dctx = zstd.ZstdDecompressor()
 
-        with self.assertRaisesRegexp(zstd.ZstdError, 'input data invalid'):
+        with self.assertRaisesRegexp(zstd.ZstdError, 'error determining content size from frame header'):
             dctx.decompress(b'foobar')
 
+    def test_input_types(self):
+        cctx = zstd.ZstdCompressor(level=1)
+        compressed = cctx.compress(b'foo')
+
+        mutable_array = bytearray(len(compressed))
+        mutable_array[:] = compressed
+
+        sources = [
+            memoryview(compressed),
+            bytearray(compressed),
+            mutable_array,
+        ]
+
+        dctx = zstd.ZstdDecompressor()
+        for source in sources:
+            self.assertEqual(dctx.decompress(source), b'foo')
+
     def test_no_content_size_in_frame(self):
         cctx = zstd.ZstdCompressor(write_content_size=False)
         compressed = cctx.compress(b'foobar')
 
         dctx = zstd.ZstdDecompressor()
-        with self.assertRaisesRegexp(zstd.ZstdError, 'input data invalid'):
+        with self.assertRaisesRegexp(zstd.ZstdError, 'could not determine content size in frame header'):
             dctx.decompress(compressed)
 
     def test_content_size_present(self):
-        cctx = zstd.ZstdCompressor(write_content_size=True)
+        cctx = zstd.ZstdCompressor()
         compressed = cctx.compress(b'foobar')
 
         dctx = zstd.ZstdDecompressor()
         decompressed = dctx.decompress(compressed)
         self.assertEqual(decompressed, b'foobar')
 
+    def test_empty_roundtrip(self):
+        cctx = zstd.ZstdCompressor()
+        compressed = cctx.compress(b'')
+
+        dctx = zstd.ZstdDecompressor()
+        decompressed = dctx.decompress(compressed)
+
+        self.assertEqual(decompressed, b'')
+
     def test_max_output_size(self):
         cctx = zstd.ZstdCompressor(write_content_size=False)
         source = b'foobar' * 256
@@ -63,7 +150,8 @@
         self.assertEqual(decompressed, source)
 
         # Input size - 1 fails
-        with self.assertRaisesRegexp(zstd.ZstdError, 'Destination buffer is too small'):
+        with self.assertRaisesRegexp(zstd.ZstdError,
+                'decompression error: did not decompress full frame'):
             dctx.decompress(compressed, max_output_size=len(source) - 1)
 
         # Input size + 1 works
@@ -94,7 +182,7 @@
         d = zstd.train_dictionary(8192, samples)
 
         orig = b'foobar' * 16384
-        cctx = zstd.ZstdCompressor(level=1, dict_data=d, write_content_size=True)
+        cctx = zstd.ZstdCompressor(level=1, dict_data=d)
         compressed = cctx.compress(orig)
 
         dctx = zstd.ZstdDecompressor(dict_data=d)
@@ -113,7 +201,7 @@
 
         sources = (b'foobar' * 8192, b'foo' * 8192, b'bar' * 8192)
         compressed = []
-        cctx = zstd.ZstdCompressor(level=1, dict_data=d, write_content_size=True)
+        cctx = zstd.ZstdCompressor(level=1, dict_data=d)
         for source in sources:
             compressed.append(cctx.compress(source))
 
@@ -122,6 +210,21 @@
             decompressed = dctx.decompress(compressed[i])
             self.assertEqual(decompressed, sources[i])
 
+    def test_max_window_size(self):
+        with open(__file__, 'rb') as fh:
+            source = fh.read()
+
+        # If we write a content size, the decompressor engages single pass
+        # mode and the window size doesn't come into play.
+        cctx = zstd.ZstdCompressor(write_content_size=False)
+        frame = cctx.compress(source)
+
+        dctx = zstd.ZstdDecompressor(max_window_size=1)
+
+        with self.assertRaisesRegexp(
+            zstd.ZstdError, 'decompression error: Frame requires too much memory'):
+            dctx.decompress(frame, max_output_size=len(source))
+
 
 @make_cffi
 class TestDecompressor_copy_stream(unittest.TestCase):
@@ -186,6 +289,211 @@
 
 
 @make_cffi
+class TestDecompressor_stream_reader(unittest.TestCase):
+    def test_context_manager(self):
+        dctx = zstd.ZstdDecompressor()
+
+        reader = dctx.stream_reader(b'foo')
+        with self.assertRaisesRegexp(zstd.ZstdError, 'read\(\) must be called from an active'):
+            reader.read(1)
+
+        with dctx.stream_reader(b'foo') as reader:
+            with self.assertRaisesRegexp(ValueError, 'cannot __enter__ multiple times'):
+                with reader as reader2:
+                    pass
+
+    def test_not_implemented(self):
+        dctx = zstd.ZstdDecompressor()
+
+        with dctx.stream_reader(b'foo') as reader:
+            with self.assertRaises(NotImplementedError):
+                reader.readline()
+
+            with self.assertRaises(NotImplementedError):
+                reader.readlines()
+
+            with self.assertRaises(NotImplementedError):
+                reader.readall()
+
+            with self.assertRaises(NotImplementedError):
+                iter(reader)
+
+            with self.assertRaises(NotImplementedError):
+                next(reader)
+
+            with self.assertRaises(io.UnsupportedOperation):
+                reader.write(b'foo')
+
+            with self.assertRaises(io.UnsupportedOperation):
+                reader.writelines([])
+
+    def test_constant_methods(self):
+        dctx = zstd.ZstdDecompressor()
+
+        with dctx.stream_reader(b'foo') as reader:
+            self.assertTrue(reader.readable())
+            self.assertFalse(reader.writable())
+            self.assertTrue(reader.seekable())
+            self.assertFalse(reader.isatty())
+            self.assertIsNone(reader.flush())
+
+    def test_read_closed(self):
+        dctx = zstd.ZstdDecompressor()
+
+        with dctx.stream_reader(b'foo') as reader:
+            reader.close()
+            with self.assertRaisesRegexp(ValueError, 'stream is closed'):
+                reader.read(1)
+
+    def test_bad_read_size(self):
+        dctx = zstd.ZstdDecompressor()
+
+        with dctx.stream_reader(b'foo') as reader:
+            with self.assertRaisesRegexp(ValueError, 'cannot read negative or size 0 amounts'):
+                reader.read(-1)
+
+            with self.assertRaisesRegexp(ValueError, 'cannot read negative or size 0 amounts'):
+                reader.read(0)
+
+    def test_read_buffer(self):
+        cctx = zstd.ZstdCompressor()
+
+        source = b''.join([b'foo' * 60, b'bar' * 60, b'baz' * 60])
+        frame = cctx.compress(source)
+
+        dctx = zstd.ZstdDecompressor()
+
+        with dctx.stream_reader(frame) as reader:
+            self.assertEqual(reader.tell(), 0)
+
+            # We should get entire frame in one read.
+            result = reader.read(8192)
+            self.assertEqual(result, source)
+            self.assertEqual(reader.tell(), len(source))
+
+            # Read after EOF should return empty bytes.
+            self.assertEqual(reader.read(), b'')
+            self.assertEqual(reader.tell(), len(result))
+
+        self.assertTrue(reader.closed())
+
+    def test_read_buffer_small_chunks(self):
+        cctx = zstd.ZstdCompressor()
+        source = b''.join([b'foo' * 60, b'bar' * 60, b'baz' * 60])
+        frame = cctx.compress(source)
+
+        dctx = zstd.ZstdDecompressor()
+        chunks = []
+
+        with dctx.stream_reader(frame, read_size=1) as reader:
+            while True:
+                chunk = reader.read(1)
+                if not chunk:
+                    break
+
+                chunks.append(chunk)
+                self.assertEqual(reader.tell(), sum(map(len, chunks)))
+
+        self.assertEqual(b''.join(chunks), source)
+
+    def test_read_stream(self):
+        cctx = zstd.ZstdCompressor()
+        source = b''.join([b'foo' * 60, b'bar' * 60, b'baz' * 60])
+        frame = cctx.compress(source)
+
+        dctx = zstd.ZstdDecompressor()
+        with dctx.stream_reader(io.BytesIO(frame)) as reader:
+            self.assertEqual(reader.tell(), 0)
+
+            chunk = reader.read(8192)
+            self.assertEqual(chunk, source)
+            self.assertEqual(reader.tell(), len(source))
+            self.assertEqual(reader.read(), b'')
+            self.assertEqual(reader.tell(), len(source))
+
+    def test_read_stream_small_chunks(self):
+        cctx = zstd.ZstdCompressor()
+        source = b''.join([b'foo' * 60, b'bar' * 60, b'baz' * 60])
+        frame = cctx.compress(source)
+
+        dctx = zstd.ZstdDecompressor()
+        chunks = []
+
+        with dctx.stream_reader(io.BytesIO(frame), read_size=1) as reader:
+            while True:
+                chunk = reader.read(1)
+                if not chunk:
+                    break
+
+                chunks.append(chunk)
+                self.assertEqual(reader.tell(), sum(map(len, chunks)))
+
+        self.assertEqual(b''.join(chunks), source)
+
+    def test_read_after_exit(self):
+        cctx = zstd.ZstdCompressor()
+        frame = cctx.compress(b'foo' * 60)
+
+        dctx = zstd.ZstdDecompressor()
+
+        with dctx.stream_reader(frame) as reader:
+            while reader.read(16):
+                pass
+
+        with self.assertRaisesRegexp(zstd.ZstdError, 'read\(\) must be called from an active'):
+            reader.read(10)
+
+    def test_illegal_seeks(self):
+        cctx = zstd.ZstdCompressor()
+        frame = cctx.compress(b'foo' * 60)
+
+        dctx = zstd.ZstdDecompressor()
+
+        with dctx.stream_reader(frame) as reader:
+            with self.assertRaisesRegexp(ValueError,
+                                         'cannot seek to negative position'):
+                reader.seek(-1, os.SEEK_SET)
+
+            reader.read(1)
+
+            with self.assertRaisesRegexp(
+                ValueError, 'cannot seek zstd decompression stream backwards'):
+                reader.seek(0, os.SEEK_SET)
+
+            with self.assertRaisesRegexp(
+                ValueError, 'cannot seek zstd decompression stream backwards'):
+                reader.seek(-1, os.SEEK_CUR)
+
+            with self.assertRaisesRegexp(
+                ValueError,
+                'zstd decompression streams cannot be seeked with SEEK_END'):
+                reader.seek(0, os.SEEK_END)
+
+            reader.close()
+
+            with self.assertRaisesRegexp(ValueError, 'stream is closed'):
+                reader.seek(4, os.SEEK_SET)
+
+        with self.assertRaisesRegexp(
+            zstd.ZstdError, 'seek\(\) must be called from an active context'):
+            reader.seek(0)
+
+    def test_seek(self):
+        source = b'foobar' * 60
+        cctx = zstd.ZstdCompressor()
+        frame = cctx.compress(source)
+
+        dctx = zstd.ZstdDecompressor()
+
+        with dctx.stream_reader(frame) as reader:
+            reader.seek(3)
+            self.assertEqual(reader.read(3), b'bar')
+
+            reader.seek(4, os.SEEK_CUR)
+            self.assertEqual(reader.read(2), b'ar')
+
+
+@make_cffi
 class TestDecompressor_decompressobj(unittest.TestCase):
     def test_simple(self):
         data = zstd.ZstdCompressor(level=1).compress(b'foobar')
@@ -194,6 +502,24 @@
         dobj = dctx.decompressobj()
         self.assertEqual(dobj.decompress(data), b'foobar')
 
+    def test_input_types(self):
+        compressed = zstd.ZstdCompressor(level=1).compress(b'foo')
+
+        dctx = zstd.ZstdDecompressor()
+
+        mutable_array = bytearray(len(compressed))
+        mutable_array[:] = compressed
+
+        sources = [
+            memoryview(compressed),
+            bytearray(compressed),
+            mutable_array,
+        ]
+
+        for source in sources:
+            dobj = dctx.decompressobj()
+            self.assertEqual(dobj.decompress(source), b'foo')
+
     def test_reuse(self):
         data = zstd.ZstdCompressor(level=1).compress(b'foobar')
 
@@ -204,22 +530,58 @@
         with self.assertRaisesRegexp(zstd.ZstdError, 'cannot use a decompressobj'):
             dobj.decompress(data)
 
+    def test_bad_write_size(self):
+        dctx = zstd.ZstdDecompressor()
+
+        with self.assertRaisesRegexp(ValueError, 'write_size must be positive'):
+            dctx.decompressobj(write_size=0)
+
+    def test_write_size(self):
+        source = b'foo' * 64 + b'bar' * 128
+        data = zstd.ZstdCompressor(level=1).compress(source)
+
+        dctx = zstd.ZstdDecompressor()
+
+        for i in range(128):
+            dobj = dctx.decompressobj(write_size=i + 1)
+            self.assertEqual(dobj.decompress(data), source)
 
 def decompress_via_writer(data):
     buffer = io.BytesIO()
     dctx = zstd.ZstdDecompressor()
-    with dctx.write_to(buffer) as decompressor:
+    with dctx.stream_writer(buffer) as decompressor:
         decompressor.write(data)
     return buffer.getvalue()
 
 
 @make_cffi
-class TestDecompressor_write_to(unittest.TestCase):
+class TestDecompressor_stream_writer(unittest.TestCase):
     def test_empty_roundtrip(self):
         cctx = zstd.ZstdCompressor()
         empty = cctx.compress(b'')
         self.assertEqual(decompress_via_writer(empty), b'')
 
+    def test_input_types(self):
+        cctx = zstd.ZstdCompressor(level=1)
+        compressed = cctx.compress(b'foo')
+
+        mutable_array = bytearray(len(compressed))
+        mutable_array[:] = compressed
+
+        sources = [
+            memoryview(compressed),
+            bytearray(compressed),
+            mutable_array,
+        ]
+
+        dctx = zstd.ZstdDecompressor()
+        for source in sources:
+            buffer = io.BytesIO()
+            with dctx.stream_writer(buffer) as decompressor:
+                decompressor.write(source)
+
+            self.assertEqual(buffer.getvalue(), b'foo')
+
     def test_large_roundtrip(self):
         chunks = []
         for i in range(255):
@@ -242,7 +604,7 @@
 
         buffer = io.BytesIO()
         dctx = zstd.ZstdDecompressor()
-        with dctx.write_to(buffer) as decompressor:
+        with dctx.stream_writer(buffer) as decompressor:
             pos = 0
             while pos < len(compressed):
                 pos2 = pos + 8192
@@ -262,14 +624,14 @@
         orig = b'foobar' * 16384
         buffer = io.BytesIO()
         cctx = zstd.ZstdCompressor(dict_data=d)
-        with cctx.write_to(buffer) as compressor:
-            self.assertEqual(compressor.write(orig), 1544)
+        with cctx.stream_writer(buffer) as compressor:
+            self.assertEqual(compressor.write(orig), 0)
 
         compressed = buffer.getvalue()
         buffer = io.BytesIO()
 
         dctx = zstd.ZstdDecompressor(dict_data=d)
-        with dctx.write_to(buffer) as decompressor:
+        with dctx.stream_writer(buffer) as decompressor:
             self.assertEqual(decompressor.write(compressed), len(orig))
 
         self.assertEqual(buffer.getvalue(), orig)
@@ -277,7 +639,7 @@
     def test_memory_size(self):
         dctx = zstd.ZstdDecompressor()
         buffer = io.BytesIO()
-        with dctx.write_to(buffer) as decompressor:
+        with dctx.stream_writer(buffer) as decompressor:
             size = decompressor.memory_size()
 
         self.assertGreater(size, 100000)
@@ -286,7 +648,7 @@
         source = zstd.ZstdCompressor().compress(b'foobarfoobar')
         dest = OpCountingBytesIO()
         dctx = zstd.ZstdDecompressor()
-        with dctx.write_to(dest, write_size=1) as decompressor:
+        with dctx.stream_writer(dest, write_size=1) as decompressor:
             s = struct.Struct('>B')
             for c in source:
                 if not isinstance(c, str):
@@ -298,29 +660,29 @@
 
 
 @make_cffi
-class TestDecompressor_read_from(unittest.TestCase):
+class TestDecompressor_read_to_iter(unittest.TestCase):
     def test_type_validation(self):
         dctx = zstd.ZstdDecompressor()
 
         # Object with read() works.
-        dctx.read_from(io.BytesIO())
+        dctx.read_to_iter(io.BytesIO())
 
         # Buffer protocol works.
-        dctx.read_from(b'foobar')
+        dctx.read_to_iter(b'foobar')
 
         with self.assertRaisesRegexp(ValueError, 'must pass an object with a read'):
-            b''.join(dctx.read_from(True))
+            b''.join(dctx.read_to_iter(True))
 
     def test_empty_input(self):
         dctx = zstd.ZstdDecompressor()
 
         source = io.BytesIO()
-        it = dctx.read_from(source)
+        it = dctx.read_to_iter(source)
         # TODO this is arguably wrong. Should get an error about missing frame foo.
         with self.assertRaises(StopIteration):
             next(it)
 
-        it = dctx.read_from(b'')
+        it = dctx.read_to_iter(b'')
         with self.assertRaises(StopIteration):
             next(it)
 
@@ -328,11 +690,11 @@
         dctx = zstd.ZstdDecompressor()
 
         source = io.BytesIO(b'foobar')
-        it = dctx.read_from(source)
+        it = dctx.read_to_iter(source)
         with self.assertRaisesRegexp(zstd.ZstdError, 'Unknown frame descriptor'):
             next(it)
 
-        it = dctx.read_from(b'foobar')
+        it = dctx.read_to_iter(b'foobar')
         with self.assertRaisesRegexp(zstd.ZstdError, 'Unknown frame descriptor'):
             next(it)
 
@@ -344,7 +706,7 @@
         source.seek(0)
 
         dctx = zstd.ZstdDecompressor()
-        it = dctx.read_from(source)
+        it = dctx.read_to_iter(source)
 
         # No chunks should be emitted since there is no data.
         with self.assertRaises(StopIteration):
@@ -358,17 +720,17 @@
         dctx = zstd.ZstdDecompressor()
 
         with self.assertRaisesRegexp(ValueError, 'skip_bytes must be smaller than read_size'):
-            b''.join(dctx.read_from(b'', skip_bytes=1, read_size=1))
+            b''.join(dctx.read_to_iter(b'', skip_bytes=1, read_size=1))
 
         with self.assertRaisesRegexp(ValueError, 'skip_bytes larger than first input chunk'):
-            b''.join(dctx.read_from(b'foobar', skip_bytes=10))
+            b''.join(dctx.read_to_iter(b'foobar', skip_bytes=10))
 
     def test_skip_bytes(self):
         cctx = zstd.ZstdCompressor(write_content_size=False)
         compressed = cctx.compress(b'foobar')
 
         dctx = zstd.ZstdDecompressor()
-        output = b''.join(dctx.read_from(b'hdr' + compressed, skip_bytes=3))
+        output = b''.join(dctx.read_to_iter(b'hdr' + compressed, skip_bytes=3))
         self.assertEqual(output, b'foobar')
 
     def test_large_output(self):
@@ -382,7 +744,7 @@
         compressed.seek(0)
 
         dctx = zstd.ZstdDecompressor()
-        it = dctx.read_from(compressed)
+        it = dctx.read_to_iter(compressed)
 
         chunks = []
         chunks.append(next(it))
@@ -395,7 +757,7 @@
         self.assertEqual(decompressed, source.getvalue())
 
         # And again with buffer protocol.
-        it = dctx.read_from(compressed.getvalue())
+        it = dctx.read_to_iter(compressed.getvalue())
         chunks = []
         chunks.append(next(it))
         chunks.append(next(it))
@@ -406,12 +768,13 @@
         decompressed = b''.join(chunks)
         self.assertEqual(decompressed, source.getvalue())
 
+    @unittest.skipUnless('ZSTD_SLOW_TESTS' in os.environ, 'ZSTD_SLOW_TESTS not set')
     def test_large_input(self):
         bytes = list(struct.Struct('>B').pack(i) for i in range(256))
         compressed = io.BytesIO()
         input_size = 0
         cctx = zstd.ZstdCompressor(level=1)
-        with cctx.write_to(compressed) as compressor:
+        with cctx.stream_writer(compressed) as compressor:
             while True:
                 compressor.write(random.choice(bytes))
                 input_size += 1
@@ -426,7 +789,7 @@
                            zstd.DECOMPRESSION_RECOMMENDED_INPUT_SIZE)
 
         dctx = zstd.ZstdDecompressor()
-        it = dctx.read_from(compressed)
+        it = dctx.read_to_iter(compressed)
 
         chunks = []
         chunks.append(next(it))
@@ -440,7 +803,7 @@
         self.assertEqual(len(decompressed), input_size)
 
         # And again with buffer protocol.
-        it = dctx.read_from(compressed.getvalue())
+        it = dctx.read_to_iter(compressed.getvalue())
 
         chunks = []
         chunks.append(next(it))
@@ -460,7 +823,7 @@
         source = io.BytesIO()
 
         compressed = io.BytesIO()
-        with cctx.write_to(compressed) as compressor:
+        with cctx.stream_writer(compressed) as compressor:
             for i in range(256):
                 chunk = b'\0' * 1024
                 compressor.write(chunk)
@@ -473,17 +836,34 @@
         self.assertEqual(simple, source.getvalue())
 
         compressed.seek(0)
-        streamed = b''.join(dctx.read_from(compressed))
+        streamed = b''.join(dctx.read_to_iter(compressed))
         self.assertEqual(streamed, source.getvalue())
 
     def test_read_write_size(self):
         source = OpCountingBytesIO(zstd.ZstdCompressor().compress(b'foobarfoobar'))
         dctx = zstd.ZstdDecompressor()
-        for chunk in dctx.read_from(source, read_size=1, write_size=1):
+        for chunk in dctx.read_to_iter(source, read_size=1, write_size=1):
             self.assertEqual(len(chunk), 1)
 
         self.assertEqual(source._read_count, len(source.getvalue()))
 
+    def test_magic_less(self):
+        params = zstd.CompressionParameters.from_level(
+            1, format=zstd.FORMAT_ZSTD1_MAGICLESS)
+        cctx = zstd.ZstdCompressor(compression_params=params)
+        frame = cctx.compress(b'foobar')
+
+        self.assertNotEqual(frame[0:4], b'\x28\xb5\x2f\xfd')
+
+        dctx = zstd.ZstdDecompressor()
+        with self.assertRaisesRegexp(
+            zstd.ZstdError, 'error determining content size from frame header'):
+            dctx.decompress(frame)
+
+        dctx = zstd.ZstdDecompressor(format=zstd.FORMAT_ZSTD1_MAGICLESS)
+        res = b''.join(dctx.read_to_iter(frame))
+        self.assertEqual(res, b'foobar')
+
 
 @make_cffi
 class TestDecompressor_content_dict_chain(unittest.TestCase):
@@ -511,19 +891,20 @@
         with self.assertRaisesRegexp(ValueError, 'chunk 0 is not a valid zstd frame'):
             dctx.decompress_content_dict_chain([b'foo' * 8])
 
-        no_size = zstd.ZstdCompressor().compress(b'foo' * 64)
+        no_size = zstd.ZstdCompressor(write_content_size=False).compress(b'foo' * 64)
 
         with self.assertRaisesRegexp(ValueError, 'chunk 0 missing content size in frame'):
             dctx.decompress_content_dict_chain([no_size])
 
         # Corrupt first frame.
-        frame = zstd.ZstdCompressor(write_content_size=True).compress(b'foo' * 64)
+        frame = zstd.ZstdCompressor().compress(b'foo' * 64)
         frame = frame[0:12] + frame[15:]
-        with self.assertRaisesRegexp(zstd.ZstdError, 'could not decompress chunk 0'):
+        with self.assertRaisesRegexp(zstd.ZstdError,
+                                     'chunk 0 did not decompress full frame'):
             dctx.decompress_content_dict_chain([frame])
 
     def test_bad_subsequent_input(self):
-        initial = zstd.ZstdCompressor(write_content_size=True).compress(b'foo' * 64)
+        initial = zstd.ZstdCompressor().compress(b'foo' * 64)
 
         dctx = zstd.ZstdDecompressor()
 
@@ -539,17 +920,17 @@
         with self.assertRaisesRegexp(ValueError, 'chunk 1 is not a valid zstd frame'):
             dctx.decompress_content_dict_chain([initial, b'foo' * 8])
 
-        no_size = zstd.ZstdCompressor().compress(b'foo' * 64)
+        no_size = zstd.ZstdCompressor(write_content_size=False).compress(b'foo' * 64)
 
         with self.assertRaisesRegexp(ValueError, 'chunk 1 missing content size in frame'):
             dctx.decompress_content_dict_chain([initial, no_size])
 
         # Corrupt second frame.
-        cctx = zstd.ZstdCompressor(write_content_size=True, dict_data=zstd.ZstdCompressionDict(b'foo' * 64))
+        cctx = zstd.ZstdCompressor(dict_data=zstd.ZstdCompressionDict(b'foo' * 64))
         frame = cctx.compress(b'bar' * 64)
         frame = frame[0:12] + frame[15:]
 
-        with self.assertRaisesRegexp(zstd.ZstdError, 'could not decompress chunk 1'):
+        with self.assertRaisesRegexp(zstd.ZstdError, 'chunk 1 did not decompress full frame'):
             dctx.decompress_content_dict_chain([initial, frame])
 
     def test_simple(self):
@@ -562,10 +943,10 @@
         ]
 
         chunks = []
-        chunks.append(zstd.ZstdCompressor(write_content_size=True).compress(original[0]))
+        chunks.append(zstd.ZstdCompressor().compress(original[0]))
         for i, chunk in enumerate(original[1:]):
             d = zstd.ZstdCompressionDict(original[i])
-            cctx = zstd.ZstdCompressor(dict_data=d, write_content_size=True)
+            cctx = zstd.ZstdCompressor(dict_data=d)
             chunks.append(cctx.compress(chunk))
 
         for i in range(1, len(original)):
@@ -594,7 +975,7 @@
             dctx.multi_decompress_to_buffer([b'foobarbaz'])
 
     def test_list_input(self):
-        cctx = zstd.ZstdCompressor(write_content_size=True)
+        cctx = zstd.ZstdCompressor()
 
         original = [b'foo' * 4, b'bar' * 6]
         frames = [cctx.compress(d) for d in original]
@@ -614,7 +995,7 @@
         self.assertEqual(len(result[1]), 18)
 
     def test_list_input_frame_sizes(self):
-        cctx = zstd.ZstdCompressor(write_content_size=False)
+        cctx = zstd.ZstdCompressor()
 
         original = [b'foo' * 4, b'bar' * 6, b'baz' * 8]
         frames = [cctx.compress(d) for d in original]
@@ -630,7 +1011,7 @@
             self.assertEqual(result[i].tobytes(), data)
 
     def test_buffer_with_segments_input(self):
-        cctx = zstd.ZstdCompressor(write_content_size=True)
+        cctx = zstd.ZstdCompressor()
 
         original = [b'foo' * 4, b'bar' * 6]
         frames = [cctx.compress(d) for d in original]
@@ -669,7 +1050,7 @@
             self.assertEqual(result[i].tobytes(), data)
 
     def test_buffer_with_segments_collection_input(self):
-        cctx = zstd.ZstdCompressor(write_content_size=True)
+        cctx = zstd.ZstdCompressor()
 
         original = [
             b'foo0' * 2,
@@ -711,8 +1092,18 @@
         for i in range(5):
             self.assertEqual(decompressed[i].tobytes(), original[i])
 
+    def test_dict(self):
+        d = zstd.train_dictionary(16384, generate_samples(), k=64, d=16)
+
+        cctx = zstd.ZstdCompressor(dict_data=d, level=1)
+        frames = [cctx.compress(s) for s in generate_samples()]
+
+        dctx = zstd.ZstdDecompressor(dict_data=d)
+        result = dctx.multi_decompress_to_buffer(frames)
+        self.assertEqual([o.tobytes() for o in result], generate_samples())
+
     def test_multiple_threads(self):
-        cctx = zstd.ZstdCompressor(write_content_size=True)
+        cctx = zstd.ZstdCompressor()
 
         frames = []
         frames.extend(cctx.compress(b'x' * 64) for i in range(256))
@@ -727,15 +1118,22 @@
         self.assertEqual(result[256].tobytes(), b'y' * 64)
 
     def test_item_failure(self):
-        cctx = zstd.ZstdCompressor(write_content_size=True)
+        cctx = zstd.ZstdCompressor()
         frames = [cctx.compress(b'x' * 128), cctx.compress(b'y' * 128)]
 
-        frames[1] = frames[1] + b'extra'
+        frames[1] = frames[1][0:15] + b'extra' + frames[1][15:]
 
         dctx = zstd.ZstdDecompressor()
 
-        with self.assertRaisesRegexp(zstd.ZstdError, 'error decompressing item 1: Src size incorrect'):
+        with self.assertRaisesRegexp(zstd.ZstdError,
+                                     'error decompressing item 1: ('
+                                     'Corrupted block|'
+                                     'Destination buffer is too small)'):
             dctx.multi_decompress_to_buffer(frames)
 
-        with self.assertRaisesRegexp(zstd.ZstdError, 'error decompressing item 1: Src size incorrect'):
+        with self.assertRaisesRegexp(zstd.ZstdError,
+                            'error decompressing item 1: ('
+                            'Corrupted block|'
+                            'Destination buffer is too small)'):
             dctx.multi_decompress_to_buffer(frames, threads=2)
+
--- a/contrib/python-zstandard/tests/test_decompressor_fuzzing.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_decompressor_fuzzing.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,6 @@
 import io
 import os
-
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
+import unittest
 
 try:
     import hypothesis
@@ -12,7 +8,7 @@
 except ImportError:
     raise unittest.SkipTest('hypothesis not available')
 
-import zstd
+import zstandard as zstd
 
 from . common import (
     make_cffi,
@@ -22,15 +18,96 @@
 
 @unittest.skipUnless('ZSTD_SLOW_TESTS' in os.environ, 'ZSTD_SLOW_TESTS not set')
 @make_cffi
-class TestDecompressor_write_to_fuzzing(unittest.TestCase):
+class TestDecompressor_stream_reader_fuzzing(unittest.TestCase):
+    @hypothesis.settings(
+        suppress_health_check=[hypothesis.HealthCheck.large_base_example])
+    @hypothesis.given(original=strategies.sampled_from(random_input_data()),
+                      level=strategies.integers(min_value=1, max_value=5),
+                      source_read_size=strategies.integers(1, 16384),
+                      read_sizes=strategies.data())
+    def test_stream_source_read_variance(self, original, level, source_read_size,
+                                         read_sizes):
+        cctx = zstd.ZstdCompressor(level=level)
+        frame = cctx.compress(original)
+
+        dctx = zstd.ZstdDecompressor()
+        source = io.BytesIO(frame)
+
+        chunks = []
+        with dctx.stream_reader(source, read_size=source_read_size) as reader:
+            while True:
+                read_size = read_sizes.draw(strategies.integers(1, 16384))
+                chunk = reader.read(read_size)
+                if not chunk:
+                    break
+
+                chunks.append(chunk)
+
+        self.assertEqual(b''.join(chunks), original)
+
+    @hypothesis.settings(
+        suppress_health_check=[hypothesis.HealthCheck.large_base_example])
+    @hypothesis.given(original=strategies.sampled_from(random_input_data()),
+                      level=strategies.integers(min_value=1, max_value=5),
+                      source_read_size=strategies.integers(1, 16384),
+                      read_sizes=strategies.data())
+    def test_buffer_source_read_variance(self, original, level, source_read_size,
+                                         read_sizes):
+        cctx = zstd.ZstdCompressor(level=level)
+        frame = cctx.compress(original)
+
+        dctx = zstd.ZstdDecompressor()
+        chunks = []
+
+        with dctx.stream_reader(frame, read_size=source_read_size) as reader:
+            while True:
+                read_size = read_sizes.draw(strategies.integers(1, 16384))
+                chunk = reader.read(read_size)
+                if not chunk:
+                    break
+
+                chunks.append(chunk)
+
+        self.assertEqual(b''.join(chunks), original)
+
+    @hypothesis.settings(
+        suppress_health_check=[hypothesis.HealthCheck.large_base_example])
+    @hypothesis.given(
+        original=strategies.sampled_from(random_input_data()),
+        level=strategies.integers(min_value=1, max_value=5),
+        source_read_size=strategies.integers(1, 16384),
+        seek_amounts=strategies.data(),
+        read_sizes=strategies.data())
+    def test_relative_seeks(self, original, level, source_read_size, seek_amounts,
+                            read_sizes):
+        cctx = zstd.ZstdCompressor(level=level)
+        frame = cctx.compress(original)
+
+        dctx = zstd.ZstdDecompressor()
+
+        with dctx.stream_reader(frame, read_size=source_read_size) as reader:
+            while True:
+                amount = seek_amounts.draw(strategies.integers(0, 16384))
+                reader.seek(amount, os.SEEK_CUR)
+
+                offset = reader.tell()
+                read_amount = read_sizes.draw(strategies.integers(1, 16384))
+                chunk = reader.read(read_amount)
+
+                if not chunk:
+                    break
+
+                self.assertEqual(original[offset:offset + len(chunk)], chunk)
+
+
+@unittest.skipUnless('ZSTD_SLOW_TESTS' in os.environ, 'ZSTD_SLOW_TESTS not set')
+@make_cffi
+class TestDecompressor_stream_writer_fuzzing(unittest.TestCase):
     @hypothesis.given(original=strategies.sampled_from(random_input_data()),
                       level=strategies.integers(min_value=1, max_value=5),
                       write_size=strategies.integers(min_value=1, max_value=8192),
-                      input_sizes=strategies.streaming(
-                          strategies.integers(min_value=1, max_value=4096)))
+                      input_sizes=strategies.data())
     def test_write_size_variance(self, original, level, write_size, input_sizes):
-        input_sizes = iter(input_sizes)
-
         cctx = zstd.ZstdCompressor(level=level)
         frame = cctx.compress(original)
 
@@ -38,9 +115,10 @@
         source = io.BytesIO(frame)
         dest = io.BytesIO()
 
-        with dctx.write_to(dest, write_size=write_size) as decompressor:
+        with dctx.stream_writer(dest, write_size=write_size) as decompressor:
             while True:
-                chunk = source.read(next(input_sizes))
+                input_size = input_sizes.draw(strategies.integers(1, 4096))
+                chunk = source.read(input_size)
                 if not chunk:
                     break
 
@@ -74,11 +152,8 @@
 class TestDecompressor_decompressobj_fuzzing(unittest.TestCase):
     @hypothesis.given(original=strategies.sampled_from(random_input_data()),
                       level=strategies.integers(min_value=1, max_value=5),
-                      chunk_sizes=strategies.streaming(
-                          strategies.integers(min_value=1, max_value=4096)))
+                      chunk_sizes=strategies.data())
     def test_random_input_sizes(self, original, level, chunk_sizes):
-        chunk_sizes = iter(chunk_sizes)
-
         cctx = zstd.ZstdCompressor(level=level)
         frame = cctx.compress(original)
 
@@ -89,7 +164,33 @@
 
         chunks = []
         while True:
-            chunk = source.read(next(chunk_sizes))
+            chunk_size = chunk_sizes.draw(strategies.integers(1, 4096))
+            chunk = source.read(chunk_size)
+            if not chunk:
+                break
+
+            chunks.append(dobj.decompress(chunk))
+
+        self.assertEqual(b''.join(chunks), original)
+
+    @hypothesis.given(original=strategies.sampled_from(random_input_data()),
+                      level=strategies.integers(min_value=1, max_value=5),
+                      write_size=strategies.integers(min_value=1,
+                                                     max_value=4 * zstd.DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE),
+                      chunk_sizes=strategies.data())
+    def test_random_output_sizes(self, original, level, write_size, chunk_sizes):
+        cctx = zstd.ZstdCompressor(level=level)
+        frame = cctx.compress(original)
+
+        source = io.BytesIO(frame)
+
+        dctx = zstd.ZstdDecompressor()
+        dobj = dctx.decompressobj(write_size=write_size)
+
+        chunks = []
+        while True:
+            chunk_size = chunk_sizes.draw(strategies.integers(1, 4096))
+            chunk = source.read(chunk_size)
             if not chunk:
                 break
 
@@ -100,7 +201,7 @@
 
 @unittest.skipUnless('ZSTD_SLOW_TESTS' in os.environ, 'ZSTD_SLOW_TESTS not set')
 @make_cffi
-class TestDecompressor_read_from_fuzzing(unittest.TestCase):
+class TestDecompressor_read_to_iter_fuzzing(unittest.TestCase):
     @hypothesis.given(original=strategies.sampled_from(random_input_data()),
                       level=strategies.integers(min_value=1, max_value=5),
                       read_size=strategies.integers(min_value=1, max_value=4096),
@@ -112,7 +213,7 @@
         source = io.BytesIO(frame)
 
         dctx = zstd.ZstdDecompressor()
-        chunks = list(dctx.read_from(source, read_size=read_size, write_size=write_size))
+        chunks = list(dctx.read_to_iter(source, read_size=read_size, write_size=write_size))
 
         self.assertEqual(b''.join(chunks), original)
 
--- a/contrib/python-zstandard/tests/test_estimate_sizes.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_estimate_sizes.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,9 +1,6 @@
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
+import unittest
 
-import zstd
+import zstandard as zstd
 
 from . common import (
     make_cffi,
@@ -16,7 +13,3 @@
         size = zstd.estimate_decompression_context_size()
         self.assertGreater(size, 100000)
 
-    def test_compression_size(self):
-        params = zstd.get_compression_parameters(3)
-        size = zstd.estimate_compression_context_size(params)
-        self.assertGreater(size, 100000)
--- a/contrib/python-zstandard/tests/test_module_attributes.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_module_attributes.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,11 +1,8 @@
 from __future__ import unicode_literals
 
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
+import unittest
 
-import zstd
+import zstandard as zstd
 
 from . common import (
     make_cffi,
@@ -15,7 +12,7 @@
 @make_cffi
 class TestModuleAttributes(unittest.TestCase):
     def test_version(self):
-        self.assertEqual(zstd.ZSTD_VERSION, (1, 1, 3))
+        self.assertEqual(zstd.ZSTD_VERSION, (1, 3, 4))
 
     def test_constants(self):
         self.assertEqual(zstd.MAX_COMPRESSION_LEVEL, 22)
@@ -23,6 +20,8 @@
 
     def test_hasattr(self):
         attrs = (
+            'CONTENTSIZE_UNKNOWN',
+            'CONTENTSIZE_ERROR',
             'COMPRESSION_RECOMMENDED_INPUT_SIZE',
             'COMPRESSION_RECOMMENDED_OUTPUT_SIZE',
             'DECOMPRESSION_RECOMMENDED_INPUT_SIZE',
@@ -40,7 +39,9 @@
             'SEARCHLENGTH_MIN',
             'SEARCHLENGTH_MAX',
             'TARGETLENGTH_MIN',
-            'TARGETLENGTH_MAX',
+            'LDM_MINMATCH_MIN',
+            'LDM_MINMATCH_MAX',
+            'LDM_BUCKETSIZELOG_MAX',
             'STRATEGY_FAST',
             'STRATEGY_DFAST',
             'STRATEGY_GREEDY',
@@ -48,6 +49,10 @@
             'STRATEGY_LAZY2',
             'STRATEGY_BTLAZY2',
             'STRATEGY_BTOPT',
+            'STRATEGY_BTULTRA',
+            'DICT_TYPE_AUTO',
+            'DICT_TYPE_RAWCONTENT',
+            'DICT_TYPE_FULLDICT',
         )
 
         for a in attrs:
--- a/contrib/python-zstandard/tests/test_train_dictionary.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/tests/test_train_dictionary.py	Mon Apr 09 10:13:29 2018 -0700
@@ -1,13 +1,11 @@
+import struct
 import sys
+import unittest
 
-try:
-    import unittest2 as unittest
-except ImportError:
-    import unittest
-
-import zstd
+import zstandard as zstd
 
 from . common import (
+    generate_samples,
     make_cffi,
 )
 
@@ -30,55 +28,18 @@
         with self.assertRaises(ValueError):
             zstd.train_dictionary(8192, [u'foo'])
 
-    def test_basic(self):
-        samples = []
-        for i in range(128):
-            samples.append(b'foo' * 64)
-            samples.append(b'bar' * 64)
-            samples.append(b'foobar' * 64)
-            samples.append(b'baz' * 64)
-            samples.append(b'foobaz' * 64)
-            samples.append(b'bazfoo' * 64)
+    def test_no_params(self):
+        d = zstd.train_dictionary(8192, generate_samples())
+        self.assertIsInstance(d.dict_id(), int_type)
 
-        d = zstd.train_dictionary(8192, samples)
-        self.assertLessEqual(len(d), 8192)
-
-        dict_id = d.dict_id()
-        self.assertIsInstance(dict_id, int_type)
+        # The dictionary ID may be different across platforms.
+        expected = b'\x37\xa4\x30\xec' + struct.pack('<I', d.dict_id())
 
         data = d.as_bytes()
-        self.assertEqual(data[0:4], b'\x37\xa4\x30\xec')
-
-    def test_set_dict_id(self):
-        samples = []
-        for i in range(128):
-            samples.append(b'foo' * 64)
-            samples.append(b'foobar' * 64)
-
-        d = zstd.train_dictionary(8192, samples, dict_id=42)
-        self.assertEqual(d.dict_id(), 42)
-
-
-@make_cffi
-class TestTrainCoverDictionary(unittest.TestCase):
-    def test_no_args(self):
-        with self.assertRaises(TypeError):
-            zstd.train_cover_dictionary()
-
-    def test_bad_args(self):
-        with self.assertRaises(TypeError):
-            zstd.train_cover_dictionary(8192, u'foo')
-
-        with self.assertRaises(ValueError):
-            zstd.train_cover_dictionary(8192, [u'foo'])
+        self.assertEqual(data[0:8], expected)
 
     def test_basic(self):
-        samples = []
-        for i in range(128):
-            samples.append(b'foo' * 64)
-            samples.append(b'foobar' * 64)
-
-        d = zstd.train_cover_dictionary(8192, samples, k=64, d=16)
+        d = zstd.train_dictionary(8192, generate_samples(), k=64, d=16)
         self.assertIsInstance(d.dict_id(), int_type)
 
         data = d.as_bytes()
@@ -88,23 +49,39 @@
         self.assertEqual(d.d, 16)
 
     def test_set_dict_id(self):
-        samples = []
-        for i in range(128):
-            samples.append(b'foo' * 64)
-            samples.append(b'foobar' * 64)
-
-        d = zstd.train_cover_dictionary(8192, samples, k=64, d=16,
-                                        dict_id=42)
+        d = zstd.train_dictionary(8192, generate_samples(), k=64, d=16,
+                                  dict_id=42)
         self.assertEqual(d.dict_id(), 42)
 
     def test_optimize(self):
-        samples = []
-        for i in range(128):
-            samples.append(b'foo' * 64)
-            samples.append(b'foobar' * 64)
+        d = zstd.train_dictionary(8192, generate_samples(), threads=-1, steps=1,
+                                  d=16)
+
+        self.assertEqual(d.k, 50)
+        self.assertEqual(d.d, 16)
+
+@make_cffi
+class TestCompressionDict(unittest.TestCase):
+    def test_bad_mode(self):
+        with self.assertRaisesRegexp(ValueError, 'invalid dictionary load mode'):
+            zstd.ZstdCompressionDict(b'foo', dict_type=42)
+
+    def test_bad_precompute_compress(self):
+        d = zstd.train_dictionary(8192, generate_samples(), k=64, d=16)
 
-        d = zstd.train_cover_dictionary(8192, samples, optimize=True,
-                                        threads=-1, steps=1, d=16)
+        with self.assertRaisesRegexp(ValueError, 'must specify one of level or '):
+            d.precompute_compress()
+
+        with self.assertRaisesRegexp(ValueError, 'must only specify one of level or '):
+            d.precompute_compress(level=3,
+                                  compression_params=zstd.CompressionParameters())
 
-        self.assertEqual(d.k, 16)
-        self.assertEqual(d.d, 16)
+    def test_precompute_compress_rawcontent(self):
+        d = zstd.ZstdCompressionDict(b'dictcontent' * 64,
+                                     dict_type=zstd.DICT_TYPE_RAWCONTENT)
+        d.precompute_compress(level=1)
+
+        d = zstd.ZstdCompressionDict(b'dictcontent' * 64,
+                                     dict_type=zstd.DICT_TYPE_FULLDICT)
+        with self.assertRaisesRegexp(zstd.ZstdError, 'unable to precompute dictionary'):
+            d.precompute_compress(level=1)
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstandard/__init__.py	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,62 @@
+# Copyright (c) 2017-present, Gregory Szorc
+# All rights reserved.
+#
+# This software may be modified and distributed under the terms
+# of the BSD license. See the LICENSE file for details.
+
+"""Python interface to the Zstandard (zstd) compression library."""
+
+from __future__ import absolute_import, unicode_literals
+
+# This module serves 2 roles:
+#
+# 1) Export the C or CFFI "backend" through a central module.
+# 2) Implement additional functionality built on top of C or CFFI backend.
+
+import os
+import platform
+
+# Some Python implementations don't support C extensions. That's why we have
+# a CFFI implementation in the first place. The code here import one of our
+# "backends" then re-exports the symbols from this module. For convenience,
+# we support falling back to the CFFI backend if the C extension can't be
+# imported. But for performance reasons, we only do this on unknown Python
+# implementation. Notably, for CPython we require the C extension by default.
+# Because someone will inevitably want special behavior, the behavior is
+# configurable via an environment variable. A potentially better way to handle
+# this is to import a special ``__importpolicy__`` module or something
+# defining a variable and `setup.py` could write the file with whatever
+# policy was specified at build time. Until someone needs it, we go with
+# the hacky but simple environment variable approach.
+_module_policy = os.environ.get('PYTHON_ZSTANDARD_IMPORT_POLICY', 'default')
+
+if _module_policy == 'default':
+    if platform.python_implementation() in ('CPython',):
+        from zstd import *
+        backend = 'cext'
+    elif platform.python_implementation() in ('PyPy',):
+        from zstd_cffi import *
+        backend = 'cffi'
+    else:
+        try:
+            from zstd import *
+            backend = 'cext'
+        except ImportError:
+            from zstd_cffi import *
+            backend = 'cffi'
+elif _module_policy == 'cffi_fallback':
+    try:
+        from zstd import *
+        backend = 'cext'
+    except ImportError:
+        from zstd_cffi import *
+        backend = 'cffi'
+elif _module_policy == 'cext':
+    from zstd import *
+    backend = 'cext'
+elif _module_policy == 'cffi':
+    from zstd_cffi import *
+    backend = 'cffi'
+else:
+    raise ImportError('unknown module import policy: %s; use default, cffi_fallback, '
+                      'cext, or cffi' % _module_policy)
--- a/contrib/python-zstandard/zstd.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd.c	Mon Apr 09 10:13:29 2018 -0700
@@ -20,12 +20,6 @@
 
 PyObject *ZstdError;
 
-PyDoc_STRVAR(estimate_compression_context_size__doc__,
-"estimate_compression_context_size(compression_parameters)\n"
-"\n"
-"Give the amount of memory allocated for a compression context given a\n"
-"CompressionParameters instance");
-
 PyDoc_STRVAR(estimate_decompression_context_size__doc__,
 "estimate_decompression_context_size()\n"
 "\n"
@@ -36,11 +30,101 @@
 	return PyLong_FromSize_t(ZSTD_estimateDCtxSize());
 }
 
-PyDoc_STRVAR(get_compression_parameters__doc__,
-"get_compression_parameters(compression_level[, source_size[, dict_size]])\n"
+PyDoc_STRVAR(frame_content_size__doc__,
+"frame_content_size(data)\n"
 "\n"
-"Obtains a ``CompressionParameters`` instance from a compression level and\n"
-"optional input size and dictionary size");
+"Obtain the decompressed size of a frame."
+);
+
+static PyObject* frame_content_size(PyObject* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"source",
+		NULL
+	};
+
+	Py_buffer source;
+	PyObject* result = NULL;
+	unsigned long long size;
+
+#if PY_MAJOR_VERSION >= 3
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*:frame_content_size",
+#else
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*:frame_content_size",
+#endif
+		kwlist, &source)) {
+		return NULL;
+	}
+
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
+	}
+
+	size = ZSTD_getFrameContentSize(source.buf, source.len);
+
+	if (size == ZSTD_CONTENTSIZE_ERROR) {
+		PyErr_SetString(ZstdError, "error when determining content size");
+	}
+	else if (size == ZSTD_CONTENTSIZE_UNKNOWN) {
+		result = PyLong_FromLong(-1);
+	}
+	else {
+		result = PyLong_FromUnsignedLongLong(size);
+	}
+
+finally:
+	PyBuffer_Release(&source);
+
+	return result;
+}
+
+PyDoc_STRVAR(frame_header_size__doc__,
+"frame_header_size(data)\n"
+"\n"
+"Obtain the size of a frame header.\n"
+);
+
+static PyObject* frame_header_size(PyObject* self, PyObject* args, PyObject* kwargs) {
+	static char* kwlist[] = {
+		"source",
+		NULL
+	};
+
+	Py_buffer source;
+	PyObject* result = NULL;
+	size_t zresult;
+
+#if PY_MAJOR_VERSION >= 3
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "y*:frame_header_size",
+#else
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "s*:frame_header_size",
+#endif
+		kwlist, &source)) {
+		return NULL;
+	}
+
+	if (!PyBuffer_IsContiguous(&source, 'C') || source.ndim > 1) {
+		PyErr_SetString(PyExc_ValueError,
+			"data buffer should be contiguous and have at most one dimension");
+		goto finally;
+	}
+
+	zresult = ZSTD_frameHeaderSize(source.buf, source.len);
+	if (ZSTD_isError(zresult)) {
+		PyErr_Format(ZstdError, "could not determine frame header size: %s",
+			ZSTD_getErrorName(zresult));
+	}
+	else {
+		result = PyLong_FromSize_t(zresult);
+	}
+
+finally:
+
+	PyBuffer_Release(&source);
+
+	return result;
+}
 
 PyDoc_STRVAR(get_frame_parameters__doc__,
 "get_frame_parameters(data)\n"
@@ -48,43 +132,48 @@
 "Obtains a ``FrameParameters`` instance by parsing data.\n");
 
 PyDoc_STRVAR(train_dictionary__doc__,
-"train_dictionary(dict_size, samples)\n"
-"\n"
-"Train a dictionary from sample data.\n"
-"\n"
-"A compression dictionary of size ``dict_size`` will be created from the\n"
-"iterable of samples provided by ``samples``.\n"
-"\n"
-"The raw dictionary content will be returned\n");
-
-PyDoc_STRVAR(train_cover_dictionary__doc__,
-"train_cover_dictionary(dict_size, samples, k=None, d=None, notifications=0, dict_id=0, level=0)\n"
+"train_dictionary(dict_size, samples, k=None, d=None, steps=None,\n"
+"                 threads=None,notifications=0, dict_id=0, level=0)\n"
 "\n"
 "Train a dictionary from sample data using the COVER algorithm.\n"
 "\n"
-"This behaves like ``train_dictionary()`` except a different algorithm is\n"
-"used to create the dictionary. The algorithm has 2 parameters: ``k`` and\n"
-"``d``. These control the *segment size* and *dmer size*. A reasonable range\n"
-"for ``k`` is ``[16, 2048+]``. A reasonable range for ``d`` is ``[6, 16]``.\n"
+"A compression dictionary of size ``dict_size`` will be created from the\n"
+"iterable of ``samples``. The raw dictionary bytes will be returned.\n"
+"\n"
+"The COVER algorithm has 2 parameters: ``k`` and ``d``. These control the\n"
+"*segment size* and *dmer size*. A reasonable range for ``k`` is\n"
+"``[16, 2048+]``. A reasonable range for ``d`` is ``[6, 16]``.\n"
 "``d`` must be less than or equal to ``k``.\n"
+"\n"
+"``steps`` can be specified to control the number of steps through potential\n"
+"values of ``k`` and ``d`` to try. ``k`` and ``d`` will only be varied if\n"
+"those arguments are not defined. i.e. if ``d`` is ``8``, then only ``k``\n"
+"will be varied in this mode.\n"
+"\n"
+"``threads`` can specify how many threads to use to test various ``k`` and\n"
+"``d`` values. ``-1`` will use as many threads as available CPUs. By default,\n"
+"a single thread is used.\n"
+"\n"
+"When ``k`` and ``d`` are not defined, default values are used and the\n"
+"algorithm will perform multiple iterations - or steps - to try to find\n"
+"ideal parameters. If both ``k`` and ``d`` are specified, then those values\n"
+"will be used. ``steps`` or ``threads`` triggers optimization mode to test\n"
+"multiple ``k`` and ``d`` variations.\n"
 );
 
 static char zstd_doc[] = "Interface to zstandard";
 
 static PyMethodDef zstd_methods[] = {
-	/* TODO remove since it is a method on CompressionParameters. */
-	{ "estimate_compression_context_size", (PyCFunction)estimate_compression_context_size,
-	METH_VARARGS, estimate_compression_context_size__doc__ },
 	{ "estimate_decompression_context_size", (PyCFunction)estimate_decompression_context_size,
 	METH_NOARGS, estimate_decompression_context_size__doc__ },
-	{ "get_compression_parameters", (PyCFunction)get_compression_parameters,
-	METH_VARARGS, get_compression_parameters__doc__ },
+	{ "frame_content_size", (PyCFunction)frame_content_size,
+	METH_VARARGS | METH_KEYWORDS, frame_content_size__doc__ },
+	{ "frame_header_size", (PyCFunction)frame_header_size,
+	METH_VARARGS | METH_KEYWORDS, frame_header_size__doc__ },
 	{ "get_frame_parameters", (PyCFunction)get_frame_parameters,
-	METH_VARARGS, get_frame_parameters__doc__ },
+	METH_VARARGS | METH_KEYWORDS, get_frame_parameters__doc__ },
 	{ "train_dictionary", (PyCFunction)train_dictionary,
 	METH_VARARGS | METH_KEYWORDS, train_dictionary__doc__ },
-	{ "train_cover_dictionary", (PyCFunction)train_cover_dictionary,
-	METH_VARARGS | METH_KEYWORDS, train_cover_dictionary__doc__ },
 	{ NULL, NULL }
 };
 
@@ -94,10 +183,12 @@
 void compressionparams_module_init(PyObject* mod);
 void constants_module_init(PyObject* mod);
 void compressiondict_module_init(PyObject* mod);
+void compressionreader_module_init(PyObject* mod);
 void compressionwriter_module_init(PyObject* mod);
 void compressoriterator_module_init(PyObject* mod);
 void decompressor_module_init(PyObject* mod);
 void decompressobj_module_init(PyObject* mod);
+void decompressionreader_module_init(PyObject *mod);
 void decompressionwriter_module_init(PyObject* mod);
 void decompressoriterator_module_init(PyObject* mod);
 void frameparams_module_init(PyObject* mod);
@@ -118,7 +209,7 @@
 	   We detect this mismatch here and refuse to load the module if this
 	   scenario is detected.
 	*/
-	if (ZSTD_VERSION_NUMBER != 10103 || ZSTD_versionNumber() != 10103) {
+	if (ZSTD_VERSION_NUMBER != 10304 || ZSTD_versionNumber() != 10304) {
 		PyErr_SetString(PyExc_ImportError, "zstd C API mismatch; Python bindings not compiled against expected zstd version");
 		return;
 	}
@@ -128,16 +219,24 @@
 	compressiondict_module_init(m);
 	compressobj_module_init(m);
 	compressor_module_init(m);
+	compressionreader_module_init(m);
 	compressionwriter_module_init(m);
 	compressoriterator_module_init(m);
 	constants_module_init(m);
 	decompressor_module_init(m);
 	decompressobj_module_init(m);
+	decompressionreader_module_init(m);
 	decompressionwriter_module_init(m);
 	decompressoriterator_module_init(m);
 	frameparams_module_init(m);
 }
 
+#if defined(__GNUC__) && (__GNUC__ >= 4)
+#  define PYTHON_ZSTD_VISIBILITY __attribute__ ((visibility ("default")))
+#else
+#  define PYTHON_ZSTD_VISIBILITY
+#endif
+
 #if PY_MAJOR_VERSION >= 3
 static struct PyModuleDef zstd_module = {
 	PyModuleDef_HEAD_INIT,
@@ -147,7 +246,7 @@
 	zstd_methods
 };
 
-PyMODINIT_FUNC PyInit_zstd(void) {
+PYTHON_ZSTD_VISIBILITY PyMODINIT_FUNC PyInit_zstd(void) {
 	PyObject *m = PyModule_Create(&zstd_module);
 	if (m) {
 		zstd_module_init(m);
@@ -159,7 +258,7 @@
 	return m;
 }
 #else
-PyMODINIT_FUNC initzstd(void) {
+PYTHON_ZSTD_VISIBILITY PyMODINIT_FUNC initzstd(void) {
 	PyObject *m = Py_InitModule3("zstd", zstd_methods, zstd_doc);
 	if (m) {
 		zstd_module_init(m);
@@ -211,3 +310,33 @@
 
 	return i;
 }
+
+/* Safer version of _PyBytes_Resize().
+ *
+ * _PyBytes_Resize() only works if the refcount is 1. In some scenarios,
+ * we can get an object with a refcount > 1, even if it was just created
+ * with PyBytes_FromStringAndSize()! That's because (at least) CPython
+ * pre-allocates PyBytes instances of size 1 for every possible byte value.
+ *
+ * If non-0 is returned, obj may or may not be NULL.
+ */
+int safe_pybytes_resize(PyObject** obj, Py_ssize_t size) {
+	PyObject* tmp;
+
+	if ((*obj)->ob_refcnt == 1) {
+		return _PyBytes_Resize(obj, size);
+	}
+
+	tmp = PyBytes_FromStringAndSize(NULL, size);
+	if (!tmp) {
+		return -1;
+	}
+
+	memcpy(PyBytes_AS_STRING(tmp), PyBytes_AS_STRING(*obj),
+		PyBytes_GET_SIZE(*obj));
+
+	Py_DECREF(*obj);
+	*obj = tmp;
+
+	return 0;
+}
\ No newline at end of file
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/COPYING	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,339 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 2, June 1991
+
+ Copyright (C) 1989, 1991 Free Software Foundation, Inc.,
+ 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The licenses for most software are designed to take away your
+freedom to share and change it.  By contrast, the GNU General Public
+License is intended to guarantee your freedom to share and change free
+software--to make sure the software is free for all its users.  This
+General Public License applies to most of the Free Software
+Foundation's software and to any other program whose authors commit to
+using it.  (Some other Free Software Foundation software is covered by
+the GNU Lesser General Public License instead.)  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+this service if you wish), that you receive source code or can get it
+if you want it, that you can change the software or use pieces of it
+in new free programs; and that you know you can do these things.
+
+  To protect your rights, we need to make restrictions that forbid
+anyone to deny you these rights or to ask you to surrender the rights.
+These restrictions translate to certain responsibilities for you if you
+distribute copies of the software, or if you modify it.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must give the recipients all the rights that
+you have.  You must make sure that they, too, receive or can get the
+source code.  And you must show them these terms so they know their
+rights.
+
+  We protect your rights with two steps: (1) copyright the software, and
+(2) offer you this license which gives you legal permission to copy,
+distribute and/or modify the software.
+
+  Also, for each author's protection and ours, we want to make certain
+that everyone understands that there is no warranty for this free
+software.  If the software is modified by someone else and passed on, we
+want its recipients to know that what they have is not the original, so
+that any problems introduced by others will not reflect on the original
+authors' reputations.
+
+  Finally, any free program is threatened constantly by software
+patents.  We wish to avoid the danger that redistributors of a free
+program will individually obtain patent licenses, in effect making the
+program proprietary.  To prevent this, we have made it clear that any
+patent must be licensed for everyone's free use or not licensed at all.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                    GNU GENERAL PUBLIC LICENSE
+   TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
+
+  0. This License applies to any program or other work which contains
+a notice placed by the copyright holder saying it may be distributed
+under the terms of this General Public License.  The "Program", below,
+refers to any such program or work, and a "work based on the Program"
+means either the Program or any derivative work under copyright law:
+that is to say, a work containing the Program or a portion of it,
+either verbatim or with modifications and/or translated into another
+language.  (Hereinafter, translation is included without limitation in
+the term "modification".)  Each licensee is addressed as "you".
+
+Activities other than copying, distribution and modification are not
+covered by this License; they are outside its scope.  The act of
+running the Program is not restricted, and the output from the Program
+is covered only if its contents constitute a work based on the
+Program (independent of having been made by running the Program).
+Whether that is true depends on what the Program does.
+
+  1. You may copy and distribute verbatim copies of the Program's
+source code as you receive it, in any medium, provided that you
+conspicuously and appropriately publish on each copy an appropriate
+copyright notice and disclaimer of warranty; keep intact all the
+notices that refer to this License and to the absence of any warranty;
+and give any other recipients of the Program a copy of this License
+along with the Program.
+
+You may charge a fee for the physical act of transferring a copy, and
+you may at your option offer warranty protection in exchange for a fee.
+
+  2. You may modify your copy or copies of the Program or any portion
+of it, thus forming a work based on the Program, and copy and
+distribute such modifications or work under the terms of Section 1
+above, provided that you also meet all of these conditions:
+
+    a) You must cause the modified files to carry prominent notices
+    stating that you changed the files and the date of any change.
+
+    b) You must cause any work that you distribute or publish, that in
+    whole or in part contains or is derived from the Program or any
+    part thereof, to be licensed as a whole at no charge to all third
+    parties under the terms of this License.
+
+    c) If the modified program normally reads commands interactively
+    when run, you must cause it, when started running for such
+    interactive use in the most ordinary way, to print or display an
+    announcement including an appropriate copyright notice and a
+    notice that there is no warranty (or else, saying that you provide
+    a warranty) and that users may redistribute the program under
+    these conditions, and telling the user how to view a copy of this
+    License.  (Exception: if the Program itself is interactive but
+    does not normally print such an announcement, your work based on
+    the Program is not required to print an announcement.)
+
+These requirements apply to the modified work as a whole.  If
+identifiable sections of that work are not derived from the Program,
+and can be reasonably considered independent and separate works in
+themselves, then this License, and its terms, do not apply to those
+sections when you distribute them as separate works.  But when you
+distribute the same sections as part of a whole which is a work based
+on the Program, the distribution of the whole must be on the terms of
+this License, whose permissions for other licensees extend to the
+entire whole, and thus to each and every part regardless of who wrote it.
+
+Thus, it is not the intent of this section to claim rights or contest
+your rights to work written entirely by you; rather, the intent is to
+exercise the right to control the distribution of derivative or
+collective works based on the Program.
+
+In addition, mere aggregation of another work not based on the Program
+with the Program (or with a work based on the Program) on a volume of
+a storage or distribution medium does not bring the other work under
+the scope of this License.
+
+  3. You may copy and distribute the Program (or a work based on it,
+under Section 2) in object code or executable form under the terms of
+Sections 1 and 2 above provided that you also do one of the following:
+
+    a) Accompany it with the complete corresponding machine-readable
+    source code, which must be distributed under the terms of Sections
+    1 and 2 above on a medium customarily used for software interchange; or,
+
+    b) Accompany it with a written offer, valid for at least three
+    years, to give any third party, for a charge no more than your
+    cost of physically performing source distribution, a complete
+    machine-readable copy of the corresponding source code, to be
+    distributed under the terms of Sections 1 and 2 above on a medium
+    customarily used for software interchange; or,
+
+    c) Accompany it with the information you received as to the offer
+    to distribute corresponding source code.  (This alternative is
+    allowed only for noncommercial distribution and only if you
+    received the program in object code or executable form with such
+    an offer, in accord with Subsection b above.)
+
+The source code for a work means the preferred form of the work for
+making modifications to it.  For an executable work, complete source
+code means all the source code for all modules it contains, plus any
+associated interface definition files, plus the scripts used to
+control compilation and installation of the executable.  However, as a
+special exception, the source code distributed need not include
+anything that is normally distributed (in either source or binary
+form) with the major components (compiler, kernel, and so on) of the
+operating system on which the executable runs, unless that component
+itself accompanies the executable.
+
+If distribution of executable or object code is made by offering
+access to copy from a designated place, then offering equivalent
+access to copy the source code from the same place counts as
+distribution of the source code, even though third parties are not
+compelled to copy the source along with the object code.
+
+  4. You may not copy, modify, sublicense, or distribute the Program
+except as expressly provided under this License.  Any attempt
+otherwise to copy, modify, sublicense or distribute the Program is
+void, and will automatically terminate your rights under this License.
+However, parties who have received copies, or rights, from you under
+this License will not have their licenses terminated so long as such
+parties remain in full compliance.
+
+  5. You are not required to accept this License, since you have not
+signed it.  However, nothing else grants you permission to modify or
+distribute the Program or its derivative works.  These actions are
+prohibited by law if you do not accept this License.  Therefore, by
+modifying or distributing the Program (or any work based on the
+Program), you indicate your acceptance of this License to do so, and
+all its terms and conditions for copying, distributing or modifying
+the Program or works based on it.
+
+  6. Each time you redistribute the Program (or any work based on the
+Program), the recipient automatically receives a license from the
+original licensor to copy, distribute or modify the Program subject to
+these terms and conditions.  You may not impose any further
+restrictions on the recipients' exercise of the rights granted herein.
+You are not responsible for enforcing compliance by third parties to
+this License.
+
+  7. If, as a consequence of a court judgment or allegation of patent
+infringement or for any other reason (not limited to patent issues),
+conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot
+distribute so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you
+may not distribute the Program at all.  For example, if a patent
+license would not permit royalty-free redistribution of the Program by
+all those who receive copies directly or indirectly through you, then
+the only way you could satisfy both it and this License would be to
+refrain entirely from distribution of the Program.
+
+If any portion of this section is held invalid or unenforceable under
+any particular circumstance, the balance of the section is intended to
+apply and the section as a whole is intended to apply in other
+circumstances.
+
+It is not the purpose of this section to induce you to infringe any
+patents or other property right claims or to contest validity of any
+such claims; this section has the sole purpose of protecting the
+integrity of the free software distribution system, which is
+implemented by public license practices.  Many people have made
+generous contributions to the wide range of software distributed
+through that system in reliance on consistent application of that
+system; it is up to the author/donor to decide if he or she is willing
+to distribute software through any other system and a licensee cannot
+impose that choice.
+
+This section is intended to make thoroughly clear what is believed to
+be a consequence of the rest of this License.
+
+  8. If the distribution and/or use of the Program is restricted in
+certain countries either by patents or by copyrighted interfaces, the
+original copyright holder who places the Program under this License
+may add an explicit geographical distribution limitation excluding
+those countries, so that distribution is permitted only in or among
+countries not thus excluded.  In such case, this License incorporates
+the limitation as if written in the body of this License.
+
+  9. The Free Software Foundation may publish revised and/or new versions
+of the General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+Each version is given a distinguishing version number.  If the Program
+specifies a version number of this License which applies to it and "any
+later version", you have the option of following the terms and conditions
+either of that version or of any later version published by the Free
+Software Foundation.  If the Program does not specify a version number of
+this License, you may choose any version ever published by the Free Software
+Foundation.
+
+  10. If you wish to incorporate parts of the Program into other free
+programs whose distribution conditions are different, write to the author
+to ask for permission.  For software which is copyrighted by the Free
+Software Foundation, write to the Free Software Foundation; we sometimes
+make exceptions for this.  Our decision will be guided by the two goals
+of preserving the free status of all derivatives of our free software and
+of promoting the sharing and reuse of software generally.
+
+                            NO WARRANTY
+
+  11. BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY
+FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW.  EXCEPT WHEN
+OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES
+PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED
+OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.  THE ENTIRE RISK AS
+TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU.  SHOULD THE
+PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING,
+REPAIR OR CORRECTION.
+
+  12. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR
+REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES,
+INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING
+OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED
+TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY
+YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER
+PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGES.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+convey the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+Also add information on how to contact you by electronic and paper mail.
+
+If the program is interactive, make it output a short notice like this
+when it starts in an interactive mode:
+
+    Gnomovision version 69, Copyright (C) year name of author
+    Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, the commands you use may
+be called something other than `show w' and `show c'; they could even be
+mouse-clicks or menu items--whatever suits your program.
+
+You should also get your employer (if you work as a programmer) or your
+school, if any, to sign a "copyright disclaimer" for the program, if
+necessary.  Here is a sample; alter the names:
+
+  Yoyodyne, Inc., hereby disclaims all copyright interest in the program
+  `Gnomovision' (which makes passes at compilers) written by James Hacker.
+
+  <signature of Ty Coon>, 1 April 1989
+  Ty Coon, President of Vice
+
+This General Public License does not permit incorporating your program into
+proprietary programs.  If your program is a subroutine library, you may
+consider it more useful to permit linking proprietary applications with the
+library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.
\ No newline at end of file
--- a/contrib/python-zstandard/zstd/PATENTS	Sun Apr 08 01:08:43 2018 +0200
+++ /dev/null	Thu Jan 01 00:00:00 1970 +0000
@@ -1,33 +0,0 @@
-Additional Grant of Patent Rights Version 2
-
-"Software" means the Zstandard software distributed by Facebook, Inc.
-
-Facebook, Inc. ("Facebook") hereby grants to each recipient of the Software
-("you") a perpetual, worldwide, royalty-free, non-exclusive, irrevocable
-(subject to the termination provision below) license under any Necessary
-Claims, to make, have made, use, sell, offer to sell, import, and otherwise
-transfer the Software. For avoidance of doubt, no license is granted under
-Facebook’s rights in any patent claims that are infringed by (i) modifications
-to the Software made by you or any third party or (ii) the Software in
-combination with any software or other technology.
-
-The license granted hereunder will terminate, automatically and without notice,
-if you (or any of your subsidiaries, corporate affiliates or agents) initiate
-directly or indirectly, or take a direct financial interest in, any Patent
-Assertion: (i) against Facebook or any of its subsidiaries or corporate
-affiliates, (ii) against any party if such Patent Assertion arises in whole or
-in part from any software, technology, product or service of Facebook or any of
-its subsidiaries or corporate affiliates, or (iii) against any party relating
-to the Software. Notwithstanding the foregoing, if Facebook or any of its
-subsidiaries or corporate affiliates files a lawsuit alleging patent
-infringement against you in the first instance, and you respond by filing a
-patent infringement counterclaim in that lawsuit against that party that is
-unrelated to the Software, the license granted hereunder will not terminate
-under section (i) of this paragraph due to such counterclaim.
-
-A "Necessary Claim" is a claim of a patent owned by Facebook that is
-necessarily infringed by the Software standing alone.
-
-A "Patent Assertion" is any lawsuit or other action alleging direct, indirect,
-or contributory infringement or inducement to infringe any patent, including a
-cross-claim or counterclaim.
--- a/contrib/python-zstandard/zstd/common/bitstream.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/bitstream.h	Mon Apr 09 10:13:29 2018 -0700
@@ -2,7 +2,7 @@
    bitstream
    Part of FSE library
    header file (to include)
-   Copyright (C) 2013-2016, Yann Collet.
+   Copyright (C) 2013-2017, Yann Collet.
 
    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
 
@@ -39,7 +39,6 @@
 extern "C" {
 #endif
 
-
 /*
 *  This API consists of small unitary functions, which must be inlined for best performance.
 *  Since link-time-optimization is not available for all compilers,
@@ -53,6 +52,18 @@
 #include "error_private.h"  /* error codes and messages */
 
 
+/*-*************************************
+*  Debug
+***************************************/
+#if defined(BIT_DEBUG) && (BIT_DEBUG>=1)
+#  include <assert.h>
+#else
+#  ifndef assert
+#    define assert(condition) ((void)0)
+#  endif
+#endif
+
+
 /*=========================================
 *  Target specific
 =========================================*/
@@ -60,18 +71,22 @@
 #  include <immintrin.h>   /* support for bextr (experimental) */
 #endif
 
+#define STREAM_ACCUMULATOR_MIN_32  25
+#define STREAM_ACCUMULATOR_MIN_64  57
+#define STREAM_ACCUMULATOR_MIN    ((U32)(MEM_32bits() ? STREAM_ACCUMULATOR_MIN_32 : STREAM_ACCUMULATOR_MIN_64))
+
 
 /*-******************************************
 *  bitStream encoding API (write forward)
 ********************************************/
 /* bitStream can mix input from multiple sources.
-*  A critical property of these streams is that they encode and decode in **reverse** direction.
-*  So the first bit sequence you add will be the last to be read, like a LIFO stack.
-*/
+ * A critical property of these streams is that they encode and decode in **reverse** direction.
+ * So the first bit sequence you add will be the last to be read, like a LIFO stack.
+ */
 typedef struct
 {
     size_t bitContainer;
-    int    bitPos;
+    unsigned bitPos;
     char*  startPtr;
     char*  ptr;
     char*  endPtr;
@@ -109,6 +124,7 @@
     unsigned bitsConsumed;
     const char* ptr;
     const char* start;
+    const char* limitPtr;
 } BIT_DStream_t;
 
 typedef enum { BIT_DStream_unfinished = 0,
@@ -151,140 +167,178 @@
 /*-**************************************************************
 *  Internal functions
 ****************************************************************/
-MEM_STATIC unsigned BIT_highbit32 (register U32 val)
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
 {
+    assert(val != 0);
+    {
 #   if defined(_MSC_VER)   /* Visual */
-    unsigned long r=0;
-    _BitScanReverse ( &r, val );
-    return (unsigned) r;
+        unsigned long r=0;
+        _BitScanReverse ( &r, val );
+        return (unsigned) r;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
-    return 31 - __builtin_clz (val);
+        return 31 - __builtin_clz (val);
 #   else   /* Software version */
-    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
-    U32 v = val;
-    v |= v >> 1;
-    v |= v >> 2;
-    v |= v >> 4;
-    v |= v >> 8;
-    v |= v >> 16;
-    return DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+        static const unsigned DeBruijnClz[32] = { 0,  9,  1, 10, 13, 21,  2, 29,
+                                                 11, 14, 16, 18, 22, 25,  3, 30,
+                                                  8, 12, 20, 28, 15, 17, 24,  7,
+                                                 19, 27, 23,  6, 26,  5,  4, 31 };
+        U32 v = val;
+        v |= v >> 1;
+        v |= v >> 2;
+        v |= v >> 4;
+        v |= v >> 8;
+        v |= v >> 16;
+        return DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
 #   endif
+    }
 }
 
 /*=====    Local Constants   =====*/
-static const unsigned BIT_mask[] = { 0, 1, 3, 7, 0xF, 0x1F, 0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF, 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0x1FFFF, 0x3FFFF, 0x7FFFF, 0xFFFFF, 0x1FFFFF, 0x3FFFFF, 0x7FFFFF,  0xFFFFFF, 0x1FFFFFF, 0x3FFFFFF };   /* up to 26 bits */
-
+static const unsigned BIT_mask[] = {
+    0,          1,         3,         7,         0xF,       0x1F,
+    0x3F,       0x7F,      0xFF,      0x1FF,     0x3FF,     0x7FF,
+    0xFFF,      0x1FFF,    0x3FFF,    0x7FFF,    0xFFFF,    0x1FFFF,
+    0x3FFFF,    0x7FFFF,   0xFFFFF,   0x1FFFFF,  0x3FFFFF,  0x7FFFFF,
+    0xFFFFFF,   0x1FFFFFF, 0x3FFFFFF, 0x7FFFFFF, 0xFFFFFFF, 0x1FFFFFFF,
+    0x3FFFFFFF, 0x7FFFFFFF}; /* up to 31 bits */
+#define BIT_MASK_SIZE (sizeof(BIT_mask) / sizeof(BIT_mask[0]))
 
 /*-**************************************************************
 *  bitStream encoding
 ****************************************************************/
 /*! BIT_initCStream() :
- *  `dstCapacity` must be > sizeof(void*)
+ *  `dstCapacity` must be > sizeof(size_t)
  *  @return : 0 if success,
-              otherwise an error code (can be tested using ERR_isError() ) */
-MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC, void* startPtr, size_t dstCapacity)
+ *            otherwise an error code (can be tested using ERR_isError()) */
+MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
+                                  void* startPtr, size_t dstCapacity)
 {
     bitC->bitContainer = 0;
     bitC->bitPos = 0;
     bitC->startPtr = (char*)startPtr;
     bitC->ptr = bitC->startPtr;
-    bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->ptr);
-    if (dstCapacity <= sizeof(bitC->ptr)) return ERROR(dstSize_tooSmall);
+    bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer);
+    if (dstCapacity <= sizeof(bitC->bitContainer)) return ERROR(dstSize_tooSmall);
     return 0;
 }
 
 /*! BIT_addBits() :
-    can add up to 26 bits into `bitC`.
-    Does not check for register overflow ! */
-MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC, size_t value, unsigned nbBits)
+ *  can add up to 31 bits into `bitC`.
+ *  Note : does not check for register overflow ! */
+MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
+                            size_t value, unsigned nbBits)
 {
+    MEM_STATIC_ASSERT(BIT_MASK_SIZE == 32);
+    assert(nbBits < BIT_MASK_SIZE);
+    assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
     bitC->bitContainer |= (value & BIT_mask[nbBits]) << bitC->bitPos;
     bitC->bitPos += nbBits;
 }
 
 /*! BIT_addBitsFast() :
  *  works only if `value` is _clean_, meaning all high bits above nbBits are 0 */
-MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC, size_t value, unsigned nbBits)
+MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC,
+                                size_t value, unsigned nbBits)
 {
+    assert((value>>nbBits) == 0);
+    assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
     bitC->bitContainer |= value << bitC->bitPos;
     bitC->bitPos += nbBits;
 }
 
 /*! BIT_flushBitsFast() :
+ *  assumption : bitContainer has not overflowed
  *  unsafe version; does not check buffer overflow */
 MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC)
 {
     size_t const nbBytes = bitC->bitPos >> 3;
+    assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
     MEM_writeLEST(bitC->ptr, bitC->bitContainer);
     bitC->ptr += nbBytes;
+    assert(bitC->ptr <= bitC->endPtr);
     bitC->bitPos &= 7;
-    bitC->bitContainer >>= nbBytes*8;   /* if bitPos >= sizeof(bitContainer)*8 --> undefined behavior */
+    bitC->bitContainer >>= nbBytes*8;
 }
 
 /*! BIT_flushBits() :
+ *  assumption : bitContainer has not overflowed
  *  safe version; check for buffer overflow, and prevents it.
- *  note : does not signal buffer overflow. This will be revealed later on using BIT_closeCStream() */
+ *  note : does not signal buffer overflow.
+ *  overflow will be revealed later on using BIT_closeCStream() */
 MEM_STATIC void BIT_flushBits(BIT_CStream_t* bitC)
 {
     size_t const nbBytes = bitC->bitPos >> 3;
+    assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
     MEM_writeLEST(bitC->ptr, bitC->bitContainer);
     bitC->ptr += nbBytes;
     if (bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
     bitC->bitPos &= 7;
-    bitC->bitContainer >>= nbBytes*8;   /* if bitPos >= sizeof(bitContainer)*8 --> undefined behavior */
+    bitC->bitContainer >>= nbBytes*8;
 }
 
 /*! BIT_closeCStream() :
  *  @return : size of CStream, in bytes,
-              or 0 if it could not fit into dstBuffer */
+ *            or 0 if it could not fit into dstBuffer */
 MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
 {
     BIT_addBitsFast(bitC, 1, 1);   /* endMark */
     BIT_flushBits(bitC);
-
-    if (bitC->ptr >= bitC->endPtr) return 0; /* doesn't fit within authorized budget : cancel */
-
+    if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
     return (bitC->ptr - bitC->startPtr) + (bitC->bitPos > 0);
 }
 
 
 /*-********************************************************
-* bitStream decoding
+*  bitStream decoding
 **********************************************************/
 /*! BIT_initDStream() :
-*   Initialize a BIT_DStream_t.
-*   `bitD` : a pointer to an already allocated BIT_DStream_t structure.
-*   `srcSize` must be the *exact* size of the bitStream, in bytes.
-*   @return : size of stream (== srcSize) or an errorCode if a problem is detected
-*/
+ *  Initialize a BIT_DStream_t.
+ * `bitD` : a pointer to an already allocated BIT_DStream_t structure.
+ * `srcSize` must be the *exact* size of the bitStream, in bytes.
+ * @return : size of stream (== srcSize), or an errorCode if a problem is detected
+ */
 MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
 {
     if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
 
+    bitD->start = (const char*)srcBuffer;
+    bitD->limitPtr = bitD->start + sizeof(bitD->bitContainer);
+
     if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
-        bitD->start = (const char*)srcBuffer;
         bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
         bitD->bitContainer = MEM_readLEST(bitD->ptr);
         { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
           bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;  /* ensures bitsConsumed is always set */
           if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
     } else {
-        bitD->start = (const char*)srcBuffer;
         bitD->ptr   = bitD->start;
         bitD->bitContainer = *(const BYTE*)(bitD->start);
         switch(srcSize)
         {
-            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
-            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
-            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
-            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24;
-            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16;
-            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8;
-            default:;
+        case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
+                /* fall-through */
+
+        case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
+                /* fall-through */
+
+        case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
+                /* fall-through */
+
+        case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24;
+                /* fall-through */
+
+        case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16;
+                /* fall-through */
+
+        case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8;
+                /* fall-through */
+
+        default: break;
         }
-        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
-          bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
-          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+        {   BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+            bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
+            if (lastByte == 0) return ERROR(corruption_detected);  /* endMark not present */
+        }
         bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
     }
 
@@ -306,12 +360,14 @@
 #  endif
         return _bextr_u32(bitContainer, start, nbBits);
 #else
+    assert(nbBits < BIT_MASK_SIZE);
     return (bitContainer >> start) & BIT_mask[nbBits];
 #endif
 }
 
 MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
 {
+    assert(nbBits < BIT_MASK_SIZE);
     return bitContainer & BIT_mask[nbBits];
 }
 
@@ -320,24 +376,24 @@
  *  local register is not modified.
  *  On 32-bits, maxNbBits==24.
  *  On 64-bits, maxNbBits==56.
- *  @return : value extracted
- */
- MEM_STATIC size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
+ * @return : value extracted */
+MEM_STATIC size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
 {
 #if defined(__BMI__) && defined(__GNUC__)   /* experimental; fails if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8 */
     return BIT_getMiddleBits(bitD->bitContainer, (sizeof(bitD->bitContainer)*8) - bitD->bitsConsumed - nbBits, nbBits);
 #else
-    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
-    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+    U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> 1) >> ((regMask-nbBits) & regMask);
 #endif
 }
 
 /*! BIT_lookBitsFast() :
-*   unsafe version; only works only if nbBits >= 1 */
+ *  unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_lookBitsFast(const BIT_DStream_t* bitD, U32 nbBits)
 {
-    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
-    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+    U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
+    assert(nbBits >= 1);
+    return (bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> (((regMask+1)-nbBits) & regMask);
 }
 
 MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
@@ -348,8 +404,7 @@
 /*! BIT_readBits() :
  *  Read (consume) next n bits from local register and update.
  *  Pay attention to not read more than nbBits contained into local register.
- *  @return : extracted value.
- */
+ * @return : extracted value. */
 MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
 {
     size_t const value = BIT_lookBits(bitD, nbBits);
@@ -358,25 +413,26 @@
 }
 
 /*! BIT_readBitsFast() :
-*   unsafe version; only works only if nbBits >= 1 */
+ *  unsafe version; only works only if nbBits >= 1 */
 MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
 {
     size_t const value = BIT_lookBitsFast(bitD, nbBits);
+    assert(nbBits >= 1);
     BIT_skipBits(bitD, nbBits);
     return value;
 }
 
 /*! BIT_reloadDStream() :
-*   Refill `bitD` from buffer previously set in BIT_initDStream() .
-*   This function is safe, it guarantees it will not read beyond src buffer.
-*   @return : status of `BIT_DStream_t` internal register.
-              if status == BIT_DStream_unfinished, internal register is filled with >= (sizeof(bitD->bitContainer)*8 - 7) bits */
+ *  Refill `bitD` from buffer previously set in BIT_initDStream() .
+ *  This function is safe, it guarantees it will not read beyond src buffer.
+ * @return : status of `BIT_DStream_t` internal register.
+ *           when status == BIT_DStream_unfinished, internal register is filled with at least 25 or 57 bits */
 MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
 {
-	if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should not happen => corruption detected */
-		return BIT_DStream_overflow;
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* overflow detected, like end of stream */
+        return BIT_DStream_overflow;
 
-    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+    if (bitD->ptr >= bitD->limitPtr) {
         bitD->ptr -= bitD->bitsConsumed >> 3;
         bitD->bitsConsumed &= 7;
         bitD->bitContainer = MEM_readLEST(bitD->ptr);
@@ -386,6 +442,7 @@
         if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
         return BIT_DStream_completed;
     }
+    /* start < ptr < limitPtr */
     {   U32 nbBytes = bitD->bitsConsumed >> 3;
         BIT_DStream_status result = BIT_DStream_unfinished;
         if (bitD->ptr - nbBytes < bitD->start) {
@@ -394,14 +451,14 @@
         }
         bitD->ptr -= nbBytes;
         bitD->bitsConsumed -= nbBytes*8;
-        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD->bitContainer), otherwise bitD->ptr == bitD->start */
         return result;
     }
 }
 
 /*! BIT_endOfDStream() :
-*   @return Tells if DStream has exactly reached its end (all bits consumed).
-*/
+ * @return : 1 if DStream has _exactly_ reached its end (all bits consumed).
+ */
 MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
 {
     return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/common/compiler.h	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPILER_H
+#define ZSTD_COMPILER_H
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+/* force inlining */
+#if defined (__GNUC__) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#  define INLINE_KEYWORD inline
+#else
+#  define INLINE_KEYWORD
+#endif
+
+#if defined(__GNUC__)
+#  define FORCE_INLINE_ATTR __attribute__((always_inline))
+#elif defined(_MSC_VER)
+#  define FORCE_INLINE_ATTR __forceinline
+#else
+#  define FORCE_INLINE_ATTR
+#endif
+
+/**
+ * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
+ * parameters. They must be inlined for the compiler to elimininate the constant
+ * branches.
+ */
+#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
+/**
+ * HINT_INLINE is used to help the compiler generate better code. It is *not*
+ * used for "templates", so it can be tweaked based on the compilers
+ * performance.
+ *
+ * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the
+ * always_inline attribute.
+ *
+ * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline
+ * attribute.
+ */
+#if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5
+#  define HINT_INLINE static INLINE_KEYWORD
+#else
+#  define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR
+#endif
+
+/* force no inlining */
+#ifdef _MSC_VER
+#  define FORCE_NOINLINE static __declspec(noinline)
+#else
+#  ifdef __GNUC__
+#    define FORCE_NOINLINE static __attribute__((__noinline__))
+#  else
+#    define FORCE_NOINLINE static
+#  endif
+#endif
+
+/* target attribute */
+#ifndef __has_attribute
+  #define __has_attribute(x) 0  /* Compatibility with non-clang compilers. */
+#endif
+#if defined(__GNUC__)
+#  define TARGET_ATTRIBUTE(target) __attribute__((__target__(target)))
+#else
+#  define TARGET_ATTRIBUTE(target)
+#endif
+
+/* Enable runtime BMI2 dispatch based on the CPU.
+ * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
+ */
+#ifndef DYNAMIC_BMI2
+  #if (defined(__clang__) && __has_attribute(__target__)) \
+      || (defined(__GNUC__) \
+          && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8))) \
+      && (defined(__x86_64__) || defined(_M_X86)) \
+      && !defined(__BMI2__)
+  #  define DYNAMIC_BMI2 1
+  #else
+  #  define DYNAMIC_BMI2 0
+  #endif
+#endif
+
+/* prefetch */
+#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))  /* _mm_prefetch() is not defined outside of x86/x64 */
+#  include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#  define PREFETCH(ptr)   _mm_prefetch((const char*)ptr, _MM_HINT_T0)
+#elif defined(__GNUC__)
+#  define PREFETCH(ptr)   __builtin_prefetch(ptr, 0, 0)
+#else
+#  define PREFETCH(ptr)   /* disabled */
+#endif
+
+/* disable warnings */
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4100)        /* disable: C4100: unreferenced formal parameter */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4204)        /* disable: C4204: non-constant aggregate initializer */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+#endif /* ZSTD_COMPILER_H */
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/common/cpu.h	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,216 @@
+/*
+ * Copyright (c) 2018-present, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMMON_CPU_H
+#define ZSTD_COMMON_CPU_H
+
+/**
+ * Implementation taken from folly/CpuId.h
+ * https://github.com/facebook/folly/blob/master/folly/CpuId.h
+ */
+
+#include <string.h>
+
+#include "mem.h"
+
+#ifdef _MSC_VER
+#include <intrin.h>
+#endif
+
+typedef struct {
+    U32 f1c;
+    U32 f1d;
+    U32 f7b;
+    U32 f7c;
+} ZSTD_cpuid_t;
+
+MEM_STATIC ZSTD_cpuid_t ZSTD_cpuid(void) {
+    U32 f1c = 0;
+    U32 f1d = 0;
+    U32 f7b = 0;
+    U32 f7c = 0;
+#ifdef _MSC_VER
+    int reg[4];
+    __cpuid((int*)reg, 0);
+    {
+        int const n = reg[0];
+        if (n >= 1) {
+            __cpuid((int*)reg, 1);
+            f1c = (U32)reg[2];
+            f1d = (U32)reg[3];
+        }
+        if (n >= 7) {
+            __cpuidex((int*)reg, 7, 0);
+            f7b = (U32)reg[1];
+            f7c = (U32)reg[2];
+        }
+    }
+#elif defined(__i386__) && defined(__PIC__) && !defined(__clang__) && defined(__GNUC__)
+    /* The following block like the normal cpuid branch below, but gcc
+     * reserves ebx for use of its pic register so we must specially
+     * handle the save and restore to avoid clobbering the register
+     */
+    U32 n;
+    __asm__(
+        "pushl %%ebx\n\t"
+        "cpuid\n\t"
+        "popl %%ebx\n\t"
+        : "=a"(n)
+        : "a"(0)
+        : "ecx", "edx");
+    if (n >= 1) {
+      U32 f1a;
+      __asm__(
+          "pushl %%ebx\n\t"
+          "cpuid\n\t"
+          "popl %%ebx\n\t"
+          : "=a"(f1a), "=c"(f1c), "=d"(f1d)
+          : "a"(1)
+          :);
+    }
+    if (n >= 7) {
+      __asm__(
+          "pushl %%ebx\n\t"
+          "cpuid\n\t"
+          "movl %%ebx, %%eax\n\r"
+          "popl %%ebx"
+          : "=a"(f7b), "=c"(f7c)
+          : "a"(7), "c"(0)
+          : "edx");
+    }
+#elif defined(__x86_64__) || defined(_M_X64) || defined(__i386__)
+    U32 n;
+    __asm__("cpuid" : "=a"(n) : "a"(0) : "ebx", "ecx", "edx");
+    if (n >= 1) {
+      U32 f1a;
+      __asm__("cpuid" : "=a"(f1a), "=c"(f1c), "=d"(f1d) : "a"(1) : "ebx");
+    }
+    if (n >= 7) {
+      U32 f7a;
+      __asm__("cpuid"
+              : "=a"(f7a), "=b"(f7b), "=c"(f7c)
+              : "a"(7), "c"(0)
+              : "edx");
+    }
+#endif
+    {
+        ZSTD_cpuid_t cpuid;
+        cpuid.f1c = f1c;
+        cpuid.f1d = f1d;
+        cpuid.f7b = f7b;
+        cpuid.f7c = f7c;
+        return cpuid;
+    }
+}
+
+#define X(name, r, bit)                                                        \
+  MEM_STATIC int ZSTD_cpuid_##name(ZSTD_cpuid_t const cpuid) {                 \
+    return ((cpuid.r) & (1U << bit)) != 0;                                     \
+  }
+
+/* cpuid(1): Processor Info and Feature Bits. */
+#define C(name, bit) X(name, f1c, bit)
+  C(sse3, 0)
+  C(pclmuldq, 1)
+  C(dtes64, 2)
+  C(monitor, 3)
+  C(dscpl, 4)
+  C(vmx, 5)
+  C(smx, 6)
+  C(eist, 7)
+  C(tm2, 8)
+  C(ssse3, 9)
+  C(cnxtid, 10)
+  C(fma, 12)
+  C(cx16, 13)
+  C(xtpr, 14)
+  C(pdcm, 15)
+  C(pcid, 17)
+  C(dca, 18)
+  C(sse41, 19)
+  C(sse42, 20)
+  C(x2apic, 21)
+  C(movbe, 22)
+  C(popcnt, 23)
+  C(tscdeadline, 24)
+  C(aes, 25)
+  C(xsave, 26)
+  C(osxsave, 27)
+  C(avx, 28)
+  C(f16c, 29)
+  C(rdrand, 30)
+#undef C
+#define D(name, bit) X(name, f1d, bit)
+  D(fpu, 0)
+  D(vme, 1)
+  D(de, 2)
+  D(pse, 3)
+  D(tsc, 4)
+  D(msr, 5)
+  D(pae, 6)
+  D(mce, 7)
+  D(cx8, 8)
+  D(apic, 9)
+  D(sep, 11)
+  D(mtrr, 12)
+  D(pge, 13)
+  D(mca, 14)
+  D(cmov, 15)
+  D(pat, 16)
+  D(pse36, 17)
+  D(psn, 18)
+  D(clfsh, 19)
+  D(ds, 21)
+  D(acpi, 22)
+  D(mmx, 23)
+  D(fxsr, 24)
+  D(sse, 25)
+  D(sse2, 26)
+  D(ss, 27)
+  D(htt, 28)
+  D(tm, 29)
+  D(pbe, 31)
+#undef D
+
+/* cpuid(7): Extended Features. */
+#define B(name, bit) X(name, f7b, bit)
+  B(bmi1, 3)
+  B(hle, 4)
+  B(avx2, 5)
+  B(smep, 7)
+  B(bmi2, 8)
+  B(erms, 9)
+  B(invpcid, 10)
+  B(rtm, 11)
+  B(mpx, 14)
+  B(avx512f, 16)
+  B(avx512dq, 17)
+  B(rdseed, 18)
+  B(adx, 19)
+  B(smap, 20)
+  B(avx512ifma, 21)
+  B(pcommit, 22)
+  B(clflushopt, 23)
+  B(clwb, 24)
+  B(avx512pf, 26)
+  B(avx512er, 27)
+  B(avx512cd, 28)
+  B(sha, 29)
+  B(avx512bw, 30)
+  B(avx512vl, 31)
+#undef B
+#define C(name, bit) X(name, f7c, bit)
+  C(prefetchwt1, 0)
+  C(avx512vbmi, 1)
+#undef C
+
+#undef X
+
+#endif /* ZSTD_COMMON_CPU_H */
--- a/contrib/python-zstandard/zstd/common/entropy_common.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/entropy_common.c	Mon Apr 09 10:13:29 2018 -0700
@@ -43,27 +43,21 @@
 #include "huf.h"
 
 
-/*-****************************************
-*  FSE Error Management
-******************************************/
+/*===   Version   ===*/
+unsigned FSE_versionNumber(void) { return FSE_VERSION_NUMBER; }
+
+
+/*===   Error Management   ===*/
 unsigned FSE_isError(size_t code) { return ERR_isError(code); }
-
 const char* FSE_getErrorName(size_t code) { return ERR_getErrorName(code); }
 
-
-/* **************************************************************
-*  HUF Error Management
-****************************************************************/
 unsigned HUF_isError(size_t code) { return ERR_isError(code); }
-
 const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
 
 
 /*-**************************************************************
 *  FSE NCount encoding-decoding
 ****************************************************************/
-static short FSE_abs(short a) { return (short)(a<0 ? -a : a); }
-
 size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
                  const void* headerBuffer, size_t hbSize)
 {
@@ -117,21 +111,21 @@
             } else {
                 bitStream >>= 2;
         }   }
-        {   short const max = (short)((2*threshold-1)-remaining);
-            short count;
+        {   int const max = (2*threshold-1) - remaining;
+            int count;
 
             if ((bitStream & (threshold-1)) < (U32)max) {
-                count = (short)(bitStream & (threshold-1));
-                bitCount   += nbBits-1;
+                count = bitStream & (threshold-1);
+                bitCount += nbBits-1;
             } else {
-                count = (short)(bitStream & (2*threshold-1));
+                count = bitStream & (2*threshold-1);
                 if (count >= threshold) count -= max;
-                bitCount   += nbBits;
+                bitCount += nbBits;
             }
 
             count--;   /* extra accuracy */
-            remaining -= FSE_abs(count);
-            normalizedCounter[charnum++] = count;
+            remaining -= count < 0 ? -count : count;   /* -1 means +1 */
+            normalizedCounter[charnum++] = (short)count;
             previous0 = !count;
             while (remaining < threshold) {
                 nbBits--;
--- a/contrib/python-zstandard/zstd/common/error_private.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/error_private.c	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,11 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 /* The purpose of this file is to have a single list of error strings embedded in binary */
@@ -20,23 +21,27 @@
     case PREFIX(GENERIC):  return "Error (generic)";
     case PREFIX(prefix_unknown): return "Unknown frame descriptor";
     case PREFIX(version_unsupported): return "Version not supported";
-    case PREFIX(parameter_unknown): return "Unknown parameter type";
     case PREFIX(frameParameter_unsupported): return "Unsupported frame parameter";
-    case PREFIX(frameParameter_unsupportedBy32bits): return "Frame parameter unsupported in 32-bits mode";
     case PREFIX(frameParameter_windowTooLarge): return "Frame requires too much memory for decoding";
-    case PREFIX(compressionParameter_unsupported): return "Compression parameter is out of bound";
+    case PREFIX(corruption_detected): return "Corrupted block detected";
+    case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
+    case PREFIX(parameter_unsupported): return "Unsupported parameter";
+    case PREFIX(parameter_outOfBound): return "Parameter is out of bound";
     case PREFIX(init_missing): return "Context should be init first";
     case PREFIX(memory_allocation): return "Allocation error : not enough memory";
+    case PREFIX(workSpace_tooSmall): return "workSpace buffer is not large enough";
     case PREFIX(stage_wrong): return "Operation not authorized at current processing stage";
-    case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
-    case PREFIX(srcSize_wrong): return "Src size incorrect";
-    case PREFIX(corruption_detected): return "Corrupted block detected";
-    case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
     case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
     case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
     case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";
     case PREFIX(dictionary_corrupted): return "Dictionary is corrupted";
     case PREFIX(dictionary_wrong): return "Dictionary mismatch";
+    case PREFIX(dictionaryCreation_failed): return "Cannot create Dictionary from provided samples";
+    case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
+    case PREFIX(srcSize_wrong): return "Src size is incorrect";
+        /* following error codes are not stable and may be removed or changed in a future version */
+    case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
+    case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
     case PREFIX(maxCode):
     default: return notErrorCode;
     }
--- a/contrib/python-zstandard/zstd/common/error_private.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/error_private.h	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,11 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 /* Note : this module is expected to remain private, do not expose it */
@@ -48,10 +49,9 @@
 /*-****************************************
 *  Error codes handling
 ******************************************/
-#ifdef ERROR
-#  undef ERROR   /* reported already defined on VS 2015 (Rich Geldreich) */
-#endif
-#define ERROR(name) ((size_t)-PREFIX(name))
+#undef ERROR   /* reported already defined on VS 2015 (Rich Geldreich) */
+#define ERROR(name) ZSTD_ERROR(name)
+#define ZSTD_ERROR(name) ((size_t)-PREFIX(name))
 
 ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
 
--- a/contrib/python-zstandard/zstd/common/fse.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/fse.h	Mon Apr 09 10:13:29 2018 -0700
@@ -31,13 +31,14 @@
    You can contact the author at :
    - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
 ****************************************************************** */
-#ifndef FSE_H
-#define FSE_H
 
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
+#ifndef FSE_H
+#define FSE_H
+
 
 /*-*****************************************
 *  Dependencies
@@ -45,6 +46,32 @@
 #include <stddef.h>    /* size_t, ptrdiff_t */
 
 
+/*-*****************************************
+*  FSE_PUBLIC_API : control library symbols visibility
+******************************************/
+#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
+#  define FSE_PUBLIC_API __attribute__ ((visibility ("default")))
+#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
+#  define FSE_PUBLIC_API __declspec(dllexport)
+#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
+#  define FSE_PUBLIC_API __declspec(dllimport) /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define FSE_PUBLIC_API
+#endif
+
+/*------   Version   ------*/
+#define FSE_VERSION_MAJOR    0
+#define FSE_VERSION_MINOR    9
+#define FSE_VERSION_RELEASE  0
+
+#define FSE_LIB_VERSION FSE_VERSION_MAJOR.FSE_VERSION_MINOR.FSE_VERSION_RELEASE
+#define FSE_QUOTE(str) #str
+#define FSE_EXPAND_AND_QUOTE(str) FSE_QUOTE(str)
+#define FSE_VERSION_STRING FSE_EXPAND_AND_QUOTE(FSE_LIB_VERSION)
+
+#define FSE_VERSION_NUMBER  (FSE_VERSION_MAJOR *100*100 + FSE_VERSION_MINOR *100 + FSE_VERSION_RELEASE)
+FSE_PUBLIC_API unsigned FSE_versionNumber(void);   /**< library version number; to be used when checking dll version */
+
 /*-****************************************
 *  FSE simple functions
 ******************************************/
@@ -56,8 +83,8 @@
                      if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression instead.
                      if FSE_isError(return), compression failed (more details using FSE_getErrorName())
 */
-size_t FSE_compress(void* dst, size_t dstCapacity,
-              const void* src, size_t srcSize);
+FSE_PUBLIC_API size_t FSE_compress(void* dst, size_t dstCapacity,
+                             const void* src, size_t srcSize);
 
 /*! FSE_decompress():
     Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
@@ -69,18 +96,18 @@
     Why ? : making this distinction requires a header.
     Header management is intentionally delegated to the user layer, which can better manage special cases.
 */
-size_t FSE_decompress(void* dst,  size_t dstCapacity,
-                const void* cSrc, size_t cSrcSize);
+FSE_PUBLIC_API size_t FSE_decompress(void* dst,  size_t dstCapacity,
+                               const void* cSrc, size_t cSrcSize);
 
 
 /*-*****************************************
 *  Tool functions
 ******************************************/
-size_t FSE_compressBound(size_t size);       /* maximum compressed size */
+FSE_PUBLIC_API size_t FSE_compressBound(size_t size);       /* maximum compressed size */
 
 /* Error Management */
-unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
-const char* FSE_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+FSE_PUBLIC_API unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
+FSE_PUBLIC_API const char* FSE_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
 
 
 /*-*****************************************
@@ -94,7 +121,7 @@
                      if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression.
                      if FSE_isError(return), it's an error code.
 */
-size_t FSE_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
 
 
 /*-*****************************************
@@ -127,50 +154,50 @@
     @return : the count of the most frequent symbol (which is not identified).
               if return == srcSize, there is only one symbol.
               Can also return an error code, which can be tested with FSE_isError(). */
-size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+FSE_PUBLIC_API size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
 
 /*! FSE_optimalTableLog():
     dynamically downsize 'tableLog' when conditions are met.
     It saves CPU time, by using smaller tables, while preserving or even improving compression ratio.
     @return : recommended tableLog (necessarily <= 'maxTableLog') */
-unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
+FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
 
 /*! FSE_normalizeCount():
     normalize counts so that sum(count[]) == Power_of_2 (2^tableLog)
     'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
     @return : tableLog,
               or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog, const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
+FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog, const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
 
 /*! FSE_NCountWriteBound():
     Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
     Typically useful for allocation purpose. */
-size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog);
 
 /*! FSE_writeNCount():
     Compactly save 'normalizedCounter' into 'buffer'.
     @return : size of the compressed table,
               or an errorCode, which can be tested using FSE_isError(). */
-size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
 
 
 /*! Constructor and Destructor of FSE_CTable.
     Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
 typedef unsigned FSE_CTable;   /* don't allocate that. It's only meant to be more restrictive than void* */
-FSE_CTable* FSE_createCTable (unsigned tableLog, unsigned maxSymbolValue);
-void        FSE_freeCTable (FSE_CTable* ct);
+FSE_PUBLIC_API FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API void        FSE_freeCTable (FSE_CTable* ct);
 
 /*! FSE_buildCTable():
     Builds `ct`, which must be already allocated, using FSE_createCTable().
     @return : 0, or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
 
 /*! FSE_compress_usingCTable():
     Compress `src` using `ct` into `dst` which must be already allocated.
     @return : size of compressed data (<= `dstCapacity`),
               or 0 if compressed data could not fit into `dst`,
               or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_compress_usingCTable (void* dst, size_t dstCapacity, const void* src, size_t srcSize, const FSE_CTable* ct);
+FSE_PUBLIC_API size_t FSE_compress_usingCTable (void* dst, size_t dstCapacity, const void* src, size_t srcSize, const FSE_CTable* ct);
 
 /*!
 Tutorial :
@@ -223,25 +250,25 @@
     @return : size read from 'rBuffer',
               or an errorCode, which can be tested using FSE_isError().
               maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
-size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
 
 /*! Constructor and Destructor of FSE_DTable.
     Note that its size depends on 'tableLog' */
 typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
-FSE_DTable* FSE_createDTable(unsigned tableLog);
-void        FSE_freeDTable(FSE_DTable* dt);
+FSE_PUBLIC_API FSE_DTable* FSE_createDTable(unsigned tableLog);
+FSE_PUBLIC_API void        FSE_freeDTable(FSE_DTable* dt);
 
 /*! FSE_buildDTable():
     Builds 'dt', which must be already allocated, using FSE_createDTable().
     return : 0, or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_buildDTable (FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_buildDTable (FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
 
 /*! FSE_decompress_usingDTable():
     Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
     into `dst` which must be already allocated.
     @return : size of regenerated data (necessarily <= `dstCapacity`),
               or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
+FSE_PUBLIC_API size_t FSE_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
 
 /*!
 Tutorial :
@@ -271,8 +298,10 @@
 If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)
 */
 
+#endif  /* FSE_H */
 
-#ifdef FSE_STATIC_LINKING_ONLY
+#if defined(FSE_STATIC_LINKING_ONLY) && !defined(FSE_H_FSE_STATIC_LINKING_ONLY)
+#define FSE_H_FSE_STATIC_LINKING_ONLY
 
 /* *** Dependency *** */
 #include "bitstream.h"
@@ -290,6 +319,10 @@
 #define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
 #define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
 
+/* or use the size to malloc() space directly. Pay attention to alignment restrictions though */
+#define FSE_CTABLE_SIZE(maxTableLog, maxSymbolValue)   (FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(FSE_CTable))
+#define FSE_DTABLE_SIZE(maxTableLog)                   (FSE_DTABLE_SIZE_U32(maxTableLog) * sizeof(FSE_DTable))
+
 
 /* *****************************************
 *  FSE advanced API
@@ -312,7 +345,7 @@
  */
 size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* workSpace);
 
-/*! FSE_count_simple
+/*! FSE_count_simple() :
  * Same as FSE_countFast(), but does not use any additional memory (not even on stack).
  * This function is unsafe, and will segfault if any value within `src` is `> *maxSymbolValuePtr` (presuming it's also the size of `count`).
 */
@@ -327,7 +360,7 @@
  * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
  * FSE_WKSP_SIZE_U32() provides the minimum size required for `workSpace` as a table of FSE_CTable.
  */
-#define FSE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue)   ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + (1<<((maxTableLog>2)?(maxTableLog-2):0)) )
+#define FSE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue)   ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
 size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
 
 size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits);
@@ -351,6 +384,11 @@
 size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog);
 /**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DTABLE_SIZE_U32(maxLog)` */
 
+typedef enum {
+   FSE_repeat_none,  /**< Cannot use the previous table */
+   FSE_repeat_check, /**< Can use the previous table but it must be checked */
+   FSE_repeat_valid  /**< Can use the previous table and it is asumed to be valid */
+ } FSE_repeat;
 
 /* *****************************************
 *  FSE symbol compression API
@@ -524,9 +562,9 @@
 
 MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, U32 symbol)
 {
-    const FSE_symbolCompressionTransform symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
+    FSE_symbolCompressionTransform const symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
     const U16* const stateTable = (const U16*)(statePtr->stateTable);
-    U32 nbBitsOut  = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);
+    U32 const nbBitsOut  = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);
     BIT_addBits(bitC, statePtr->value, nbBitsOut);
     statePtr->value = stateTable[ (statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
 }
@@ -664,5 +702,3 @@
 #if defined (__cplusplus)
 }
 #endif
-
-#endif  /* FSE_H */
--- a/contrib/python-zstandard/zstd/common/fse_decompress.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/fse_decompress.c	Mon Apr 09 10:13:29 2018 -0700
@@ -34,35 +34,15 @@
 
 
 /* **************************************************************
-*  Compiler specifics
-****************************************************************/
-#ifdef _MSC_VER    /* Visual Studio */
-#  define FORCE_INLINE static __forceinline
-#  include <intrin.h>                    /* For Visual 2005 */
-#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
-#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
-#else
-#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
-#    ifdef __GNUC__
-#      define FORCE_INLINE static inline __attribute__((always_inline))
-#    else
-#      define FORCE_INLINE static inline
-#    endif
-#  else
-#    define FORCE_INLINE static
-#  endif /* __STDC_VERSION__ */
-#endif
-
-
-/* **************************************************************
 *  Includes
 ****************************************************************/
 #include <stdlib.h>     /* malloc, free, qsort */
 #include <string.h>     /* memcpy, memset */
-#include <stdio.h>      /* printf (debug) */
 #include "bitstream.h"
+#include "compiler.h"
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
+#include "error_private.h"
 
 
 /* **************************************************************
@@ -159,8 +139,8 @@
     {   U32 u;
         for (u=0; u<tableSize; u++) {
             FSE_FUNCTION_TYPE const symbol = (FSE_FUNCTION_TYPE)(tableDecode[u].symbol);
-            U16 nextState = symbolNext[symbol]++;
-            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            U32 const nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
             tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
     }   }
 
@@ -217,7 +197,7 @@
     return 0;
 }
 
-FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+FORCE_INLINE_TEMPLATE size_t FSE_decompress_usingDTable_generic(
           void* dst, size_t maxDstSize,
     const void* cSrc, size_t cSrcSize,
     const FSE_DTable* dt, const unsigned fast)
--- a/contrib/python-zstandard/zstd/common/huf.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/huf.h	Mon Apr 09 10:13:29 2018 -0700
@@ -31,81 +31,114 @@
    You can contact the author at :
    - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
 ****************************************************************** */
-#ifndef HUF_H_298734234
-#define HUF_H_298734234
 
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
+#ifndef HUF_H_298734234
+#define HUF_H_298734234
 
 /* *** Dependencies *** */
 #include <stddef.h>    /* size_t */
 
 
-/* *** simple functions *** */
-/**
-HUF_compress() :
-    Compress content from buffer 'src', of size 'srcSize', into buffer 'dst'.
-    'dst' buffer must be already allocated.
-    Compression runs faster if `dstCapacity` >= HUF_compressBound(srcSize).
-    `srcSize` must be <= `HUF_BLOCKSIZE_MAX` == 128 KB.
-    @return : size of compressed data (<= `dstCapacity`).
-    Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
-                     if return == 1, srcData is a single repeated byte symbol (RLE compression).
-                     if HUF_isError(return), compression failed (more details using HUF_getErrorName())
-*/
-size_t HUF_compress(void* dst, size_t dstCapacity,
-              const void* src, size_t srcSize);
+/* *** library symbols visibility *** */
+/* Note : when linking with -fvisibility=hidden on gcc, or by default on Visual,
+ *        HUF symbols remain "private" (internal symbols for library only).
+ *        Set macro FSE_DLL_EXPORT to 1 if you want HUF symbols visible on DLL interface */
+#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
+#  define HUF_PUBLIC_API __attribute__ ((visibility ("default")))
+#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
+#  define HUF_PUBLIC_API __declspec(dllexport)
+#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
+#  define HUF_PUBLIC_API __declspec(dllimport)  /* not required, just to generate faster code (saves a function pointer load from IAT and an indirect jump) */
+#else
+#  define HUF_PUBLIC_API
+#endif
+
+
+/* ========================== */
+/* ***  simple functions  *** */
+/* ========================== */
 
-/**
-HUF_decompress() :
-    Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
-    into already allocated buffer 'dst', of minimum size 'dstSize'.
-    `originalSize` : **must** be the ***exact*** size of original (uncompressed) data.
-    Note : in contrast with FSE, HUF_decompress can regenerate
-           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
-           because it knows size to regenerate.
-    @return : size of regenerated data (== originalSize),
-              or an error code, which can be tested using HUF_isError()
-*/
-size_t HUF_decompress(void* dst,  size_t originalSize,
-                const void* cSrc, size_t cSrcSize);
+/** HUF_compress() :
+ *  Compress content from buffer 'src', of size 'srcSize', into buffer 'dst'.
+ * 'dst' buffer must be already allocated.
+ *  Compression runs faster if `dstCapacity` >= HUF_compressBound(srcSize).
+ * `srcSize` must be <= `HUF_BLOCKSIZE_MAX` == 128 KB.
+ * @return : size of compressed data (<= `dstCapacity`).
+ *  Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
+ *                   if HUF_isError(return), compression failed (more details using HUF_getErrorName())
+ */
+HUF_PUBLIC_API size_t HUF_compress(void* dst, size_t dstCapacity,
+                             const void* src, size_t srcSize);
+
+/** HUF_decompress() :
+ *  Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
+ *  into already allocated buffer 'dst', of minimum size 'dstSize'.
+ * `originalSize` : **must** be the ***exact*** size of original (uncompressed) data.
+ *  Note : in contrast with FSE, HUF_decompress can regenerate
+ *         RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+ *         because it knows size to regenerate (originalSize).
+ * @return : size of regenerated data (== originalSize),
+ *           or an error code, which can be tested using HUF_isError()
+ */
+HUF_PUBLIC_API size_t HUF_decompress(void* dst,  size_t originalSize,
+                               const void* cSrc, size_t cSrcSize);
 
 
 /* ***   Tool functions *** */
-#define HUF_BLOCKSIZE_MAX (128 * 1024)       /**< maximum input size for a single block compressed with HUF_compress */
-size_t HUF_compressBound(size_t size);       /**< maximum compressed size (worst case) */
+#define HUF_BLOCKSIZE_MAX (128 * 1024)                  /**< maximum input size for a single block compressed with HUF_compress */
+HUF_PUBLIC_API size_t HUF_compressBound(size_t size);   /**< maximum compressed size (worst case) */
 
 /* Error Management */
-unsigned    HUF_isError(size_t code);        /**< tells if a return value is an error code */
-const char* HUF_getErrorName(size_t code);   /**< provides error code string (useful for debugging) */
+HUF_PUBLIC_API unsigned    HUF_isError(size_t code);       /**< tells if a return value is an error code */
+HUF_PUBLIC_API const char* HUF_getErrorName(size_t code);  /**< provides error code string (useful for debugging) */
 
 
 /* ***   Advanced function   *** */
 
 /** HUF_compress2() :
- *   Same as HUF_compress(), but offers direct control over `maxSymbolValue` and `tableLog` .
- *   `tableLog` must be `<= HUF_TABLELOG_MAX` . */
-size_t HUF_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
+ *  Same as HUF_compress(), but offers control over `maxSymbolValue` and `tableLog`.
+ * `maxSymbolValue` must be <= HUF_SYMBOLVALUE_MAX .
+ * `tableLog` must be `<= HUF_TABLELOG_MAX` . */
+HUF_PUBLIC_API size_t HUF_compress2 (void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                               unsigned maxSymbolValue, unsigned tableLog);
 
 /** HUF_compress4X_wksp() :
-*   Same as HUF_compress2(), but uses externally allocated `workSpace`, which must be a table of >= 1024 unsigned */
-size_t HUF_compress4X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);  /**< `workSpace` must be a table of at least 1024 unsigned */
+ *  Same as HUF_compress2(), but uses externally allocated `workSpace`.
+ * `workspace` must have minimum alignment of 4, and be at least as large as HUF_WORKSPACE_SIZE */
+#define HUF_WORKSPACE_SIZE (6 << 10)
+#define HUF_WORKSPACE_SIZE_U32 (HUF_WORKSPACE_SIZE / sizeof(U32))
+HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     unsigned maxSymbolValue, unsigned tableLog,
+                                     void* workSpace, size_t wkspSize);
 
-
+#endif   /* HUF_H_298734234 */
 
-#ifdef HUF_STATIC_LINKING_ONLY
+/* ******************************************************************
+ *  WARNING !!
+ *  The following section contains advanced and experimental definitions
+ *  which shall never be used in the context of a dynamic library,
+ *  because they are not guaranteed to remain stable in the future.
+ *  Only consider them in association with static linking.
+ * *****************************************************************/
+#if defined(HUF_STATIC_LINKING_ONLY) && !defined(HUF_H_HUF_STATIC_LINKING_ONLY)
+#define HUF_H_HUF_STATIC_LINKING_ONLY
 
 /* *** Dependencies *** */
 #include "mem.h"   /* U32 */
 
 
 /* *** Constants *** */
-#define HUF_TABLELOG_ABSOLUTEMAX  15   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
-#define HUF_TABLELOG_MAX  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
-#define HUF_TABLELOG_DEFAULT  11       /* tableLog by default, when not specified */
-#define HUF_SYMBOLVALUE_MAX 255
+#define HUF_TABLELOG_MAX      12      /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_TABLELOG_DEFAULT  11      /* default tableLog value when none specified */
+#define HUF_SYMBOLVALUE_MAX  255
+
+#define HUF_TABLELOG_ABSOLUTEMAX  15  /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
 #if (HUF_TABLELOG_MAX > HUF_TABLELOG_ABSOLUTEMAX)
 #  error "HUF_TABLELOG_MAX is too large !"
 #endif
@@ -116,12 +149,14 @@
 ******************************************/
 /* HUF buffer bounds */
 #define HUF_CTABLEBOUND 129
-#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true when incompressible is pre-filtered with fast heuristic */
 #define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
 
 /* static allocation of HUF's Compression Table */
+#define HUF_CTABLE_SIZE_U32(maxSymbolValue)   ((maxSymbolValue)+1)   /* Use tables of U32, for proper alignment */
+#define HUF_CTABLE_SIZE(maxSymbolValue)       (HUF_CTABLE_SIZE_U32(maxSymbolValue) * sizeof(U32))
 #define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
-    U32 name##hb[maxSymbolValue+1]; \
+    U32 name##hb[HUF_CTABLE_SIZE_U32(maxSymbolValue)]; \
     void* name##hv = &(name##hb); \
     HUF_CElt* name = (HUF_CElt*)(name##hv)   /* no final ; */
 
@@ -142,97 +177,151 @@
 
 size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< decodes RLE and uncompressed */
 size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /**< considers RLE and uncompressed as errors */
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize); /**< considers RLE and uncompressed as errors */
 size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
 size_t HUF_decompress4X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress4X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
 
 
 /* ****************************************
-*  HUF detailed API
-******************************************/
-/*!
-HUF_compress() does the following:
-1. count symbol occurrence from source[] into table count[] using FSE_count()
-2. (optional) refine tableLog using HUF_optimalTableLog()
-3. build Huffman table from count using HUF_buildCTable()
-4. save Huffman table to memory buffer using HUF_writeCTable()
-5. encode the data stream using HUF_compress4X_usingCTable()
+ *  HUF detailed API
+ * ****************************************/
 
-The following API allows targeting specific sub-functions for advanced tasks.
-For example, it's possible to compress several blocks using the same 'CTable',
-or to save and regenerate 'CTable' using external methods.
-*/
-/* FSE_count() : find it within "fse.h" */
+/*! HUF_compress() does the following:
+ *  1. count symbol occurrence from source[] into table count[] using FSE_count() (exposed within "fse.h")
+ *  2. (optional) refine tableLog using HUF_optimalTableLog()
+ *  3. build Huffman table from count using HUF_buildCTable()
+ *  4. save Huffman table to memory buffer using HUF_writeCTable()
+ *  5. encode the data stream using HUF_compress4X_usingCTable()
+ *
+ *  The following API allows targeting specific sub-functions for advanced tasks.
+ *  For example, it's possible to compress several blocks using the same 'CTable',
+ *  or to save and regenerate 'CTable' using external methods.
+ */
 unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
 typedef struct HUF_CElt_s HUF_CElt;   /* incomplete type */
-size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits);
+size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits);   /* @return : maxNbBits; CTable and count can overlap. In which case, CTable will overwrite count content */
 size_t HUF_writeCTable (void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog);
 size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
 
+typedef enum {
+   HUF_repeat_none,  /**< Cannot use the previous table */
+   HUF_repeat_check, /**< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
+   HUF_repeat_valid  /**< Can use the previous table and it is asumed to be valid */
+ } HUF_repeat;
+/** HUF_compress4X_repeat() :
+ *  Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+ *  If it uses hufTable it does not modify hufTable or repeat.
+ *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+ *  If preferRepeat then the old table will always be used if valid. */
+size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
+                       const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned tableLog,
+                       void* workSpace, size_t wkspSize,    /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
 
 /** HUF_buildCTable_wksp() :
  *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
- *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as a table of 1024 unsigned.
+ * `workSpace` must be aligned on 4-bytes boundaries, and its size must be >= HUF_CTABLE_WORKSPACE_SIZE.
  */
+#define HUF_CTABLE_WORKSPACE_SIZE_U32 (2*HUF_SYMBOLVALUE_MAX +1 +1)
+#define HUF_CTABLE_WORKSPACE_SIZE (HUF_CTABLE_WORKSPACE_SIZE_U32 * sizeof(unsigned))
 size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize);
 
 /*! HUF_readStats() :
-    Read compact Huffman tree, saved by HUF_writeCTable().
-    `huffWeight` is destination buffer.
-    @return : size read from `src` , or an error Code .
-    Note : Needed by HUF_readCTable() and HUF_readDTableXn() . */
-size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
-                     U32* nbSymbolsPtr, U32* tableLogPtr,
+ *  Read compact Huffman tree, saved by HUF_writeCTable().
+ * `huffWeight` is destination buffer.
+ * @return : size read from `src` , or an error Code .
+ *  Note : Needed by HUF_readCTable() and HUF_readDTableXn() . */
+size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
+                     U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
                      const void* src, size_t srcSize);
 
 /** HUF_readCTable() :
-*   Loading a CTable saved with HUF_writeCTable() */
-size_t HUF_readCTable (HUF_CElt* CTable, unsigned maxSymbolValue, const void* src, size_t srcSize);
+ *  Loading a CTable saved with HUF_writeCTable() */
+size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
 
 
 /*
-HUF_decompress() does the following:
-1. select the decompression algorithm (X2, X4) based on pre-computed heuristics
-2. build Huffman table from save, using HUF_readDTableXn()
-3. decode 1 or 4 segments in parallel using HUF_decompressSXn_usingDTable
-*/
+ * HUF_decompress() does the following:
+ * 1. select the decompression algorithm (X2, X4) based on pre-computed heuristics
+ * 2. build Huffman table from save, using HUF_readDTableX?()
+ * 3. decode 1 or 4 segments in parallel using HUF_decompress?X?_usingDTable()
+ */
 
 /** HUF_selectDecoder() :
-*   Tells which decoder is likely to decode faster,
-*   based on a set of pre-determined metrics.
-*   @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
-*   Assumption : 0 < cSrcSize < dstSize <= 128 KB */
+ *  Tells which decoder is likely to decode faster,
+ *  based on a set of pre-computed metrics.
+ * @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
+ *  Assumption : 0 < dstSize <= 128 KB */
 U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
 
+/**
+ *  The minimum workspace size for the `workSpace` used in
+ *  HUF_readDTableX2_wksp() and HUF_readDTableX4_wksp().
+ *
+ *  The space used depends on HUF_TABLELOG_MAX, ranging from ~1500 bytes when
+ *  HUF_TABLE_LOG_MAX=12 to ~1850 bytes when HUF_TABLE_LOG_MAX=15.
+ *  Buffer overflow errors may potentially occur if code modifications result in
+ *  a required workspace size greater than that specified in the following
+ *  macro.
+ */
+#define HUF_DECOMPRESS_WORKSPACE_SIZE (2 << 10)
+#define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
+
 size_t HUF_readDTableX2 (HUF_DTable* DTable, const void* src, size_t srcSize);
+size_t HUF_readDTableX2_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
 size_t HUF_readDTableX4 (HUF_DTable* DTable, const void* src, size_t srcSize);
+size_t HUF_readDTableX4_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
 
 size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 size_t HUF_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 
 
+/* ====================== */
 /* single stream variants */
+/* ====================== */
 
 size_t HUF_compress1X (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
-size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);  /**< `workSpace` must be a table of at least 1024 unsigned */
+size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);  /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
 size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
+/** HUF_compress1X_repeat() :
+ *  Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+ *  If it uses hufTable it does not modify hufTable or repeat.
+ *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+ *  If preferRepeat then the old table will always be used if valid. */
+size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
+                       const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned tableLog,
+                       void* workSpace, size_t wkspSize,   /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
 
 size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
 size_t HUF_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
 
 size_t HUF_decompress1X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+size_t HUF_decompress1X_DCtx_wksp (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);
 size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
 size_t HUF_decompress1X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress1X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
 
 size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);   /**< automatic selection of sing or double symbol decoder, based on DTable */
 size_t HUF_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 size_t HUF_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 
+/* BMI2 variants.
+ * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
+ */
+size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
+size_t HUF_decompress1X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
+size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+
 #endif /* HUF_STATIC_LINKING_ONLY */
 
-
 #if defined (__cplusplus)
 }
 #endif
-
-#endif   /* HUF_H_298734234 */
--- a/contrib/python-zstandard/zstd/common/mem.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/mem.h	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,11 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 #ifndef MEM_H_MODULE
@@ -48,14 +49,13 @@
 *****************************************************************/
 #if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
 # include <stdint.h>
-  typedef  uint8_t BYTE;
-  typedef uint16_t U16;
-  typedef  int16_t S16;
-  typedef uint32_t U32;
-  typedef  int32_t S32;
-  typedef uint64_t U64;
-  typedef  int64_t S64;
-  typedef intptr_t iPtrDiff;
+  typedef   uint8_t BYTE;
+  typedef  uint16_t U16;
+  typedef   int16_t S16;
+  typedef  uint32_t U32;
+  typedef   int32_t S32;
+  typedef  uint64_t U64;
+  typedef   int64_t S64;
 #else
   typedef unsigned char      BYTE;
   typedef unsigned short      U16;
@@ -64,7 +64,6 @@
   typedef   signed int        S32;
   typedef unsigned long long  U64;
   typedef   signed long long  S64;
-  typedef ptrdiff_t      iPtrDiff;
 #endif
 
 
@@ -76,19 +75,18 @@
  * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
  * The below switch allow to select different access method for improved performance.
  * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ * Method 1 : `__packed` statement. It depends on compiler extension (i.e., not portable).
  *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
  * Method 2 : direct access. This method is portable but violate C standard.
  *            It can generate buggy code on targets depending on alignment.
- *            In some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ *            In some circumstances, it's the only known way to get the most performance (i.e. GCC + ARMv6)
  * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
  * Prefer these methods in priority order (0 > 1 > 2)
  */
 #ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
 #  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
 #    define MEM_FORCE_MEMORY_ACCESS 2
-#  elif defined(__INTEL_COMPILER) /*|| defined(_MSC_VER)*/ || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#  elif defined(__INTEL_COMPILER) || defined(__GNUC__)
 #    define MEM_FORCE_MEMORY_ACCESS 1
 #  endif
 #endif
@@ -109,7 +107,7 @@
 MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
 MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
 MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
-MEM_STATIC U64 MEM_readST(const void* memPtr) { return *(const size_t*) memPtr; }
+MEM_STATIC size_t MEM_readST(const void* memPtr) { return *(const size_t*) memPtr; }
 
 MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
 MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
@@ -120,21 +118,27 @@
 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
 /* currently only defined for gcc and icc */
 #if defined(_MSC_VER) || (defined(__INTEL_COMPILER) && defined(WIN32))
-	__pragma( pack(push, 1) )
-    typedef union { U16 u16; U32 u32; U64 u64; size_t st; } unalign;
+    __pragma( pack(push, 1) )
+    typedef struct { U16 v; } unalign16;
+    typedef struct { U32 v; } unalign32;
+    typedef struct { U64 v; } unalign64;
+    typedef struct { size_t v; } unalignArch;
     __pragma( pack(pop) )
 #else
-    typedef union { U16 u16; U32 u32; U64 u64; size_t st; } __attribute__((packed)) unalign;
+    typedef struct { U16 v; } __attribute__((packed)) unalign16;
+    typedef struct { U32 v; } __attribute__((packed)) unalign32;
+    typedef struct { U64 v; } __attribute__((packed)) unalign64;
+    typedef struct { size_t v; } __attribute__((packed)) unalignArch;
 #endif
 
-MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
-MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-MEM_STATIC U64 MEM_readST(const void* ptr) { return ((const unalign*)ptr)->st; }
+MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign16*)ptr)->v; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign32*)ptr)->v; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign64*)ptr)->v; }
+MEM_STATIC size_t MEM_readST(const void* ptr) { return ((const unalignArch*)ptr)->v; }
 
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
-MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign*)memPtr)->u32 = value; }
-MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign*)memPtr)->u64 = value; }
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign16*)memPtr)->v = value; }
+MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign32*)memPtr)->v = value; }
+MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign64*)memPtr)->v = value; }
 
 #else
 
@@ -182,7 +186,7 @@
 {
 #if defined(_MSC_VER)     /* Visual Studio */
     return _byteswap_ulong(in);
-#elif defined (__GNUC__)
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
     return __builtin_bswap32(in);
 #else
     return  ((in << 24) & 0xff000000 ) |
@@ -196,7 +200,7 @@
 {
 #if defined(_MSC_VER)     /* Visual Studio */
     return _byteswap_uint64(in);
-#elif defined (__GNUC__)
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
     return __builtin_bswap64(in);
 #else
     return  ((in << 56) & 0xff00000000000000ULL) |
@@ -351,20 +355,6 @@
 }
 
 
-/* function safe only for comparisons */
-MEM_STATIC U32 MEM_readMINMATCH(const void* memPtr, U32 length)
-{
-    switch (length)
-    {
-    default :
-    case 4 : return MEM_read32(memPtr);
-    case 3 : if (MEM_isLittleEndian())
-                return MEM_read32(memPtr)<<8;
-             else
-                return MEM_read32(memPtr)>>8;
-    }
-}
-
 #if defined (__cplusplus)
 }
 #endif
--- a/contrib/python-zstandard/zstd/common/pool.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/pool.c	Mon Apr 09 10:13:29 2018 -0700
@@ -1,17 +1,18 @@
-/**
- * Copyright (c) 2016-present, Facebook, Inc.
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 
 /* ======   Dependencies   ======= */
 #include <stddef.h>  /* size_t */
-#include <stdlib.h>  /* malloc, calloc, free */
 #include "pool.h"
+#include "zstd_internal.h"  /* ZSTD_malloc, ZSTD_free */
 
 /* ======   Compiler specifics   ====== */
 #if defined(_MSC_VER)
@@ -25,13 +26,14 @@
 
 /* A job is a function and an opaque argument */
 typedef struct POOL_job_s {
-  POOL_function function;
-  void *opaque;
+    POOL_function function;
+    void *opaque;
 } POOL_job;
 
 struct POOL_ctx_s {
+    ZSTD_customMem customMem;
     /* Keep track of the threads */
-    pthread_t *threads;
+    ZSTD_pthread_t *threads;
     size_t numThreads;
 
     /* The queue is a circular buffer */
@@ -39,12 +41,18 @@
     size_t queueHead;
     size_t queueTail;
     size_t queueSize;
+
+    /* The number of threads working on jobs */
+    size_t numThreadsBusy;
+    /* Indicates if the queue is empty */
+    int queueEmpty;
+
     /* The mutex protects the queue */
-    pthread_mutex_t queueMutex;
+    ZSTD_pthread_mutex_t queueMutex;
     /* Condition variable for pushers to wait on when the queue is full */
-    pthread_cond_t queuePushCond;
+    ZSTD_pthread_cond_t queuePushCond;
     /* Condition variables for poppers to wait on when the queue is empty */
-    pthread_cond_t queuePopCond;
+    ZSTD_pthread_cond_t queuePopCond;
     /* Indicates if the queue is shutting down */
     int shutdown;
 };
@@ -59,55 +67,73 @@
     if (!ctx) { return NULL; }
     for (;;) {
         /* Lock the mutex and wait for a non-empty queue or until shutdown */
-        pthread_mutex_lock(&ctx->queueMutex);
-        while (ctx->queueHead == ctx->queueTail && !ctx->shutdown) {
-            pthread_cond_wait(&ctx->queuePopCond, &ctx->queueMutex);
+        ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+
+        while (ctx->queueEmpty && !ctx->shutdown) {
+            ZSTD_pthread_cond_wait(&ctx->queuePopCond, &ctx->queueMutex);
         }
         /* empty => shutting down: so stop */
-        if (ctx->queueHead == ctx->queueTail) {
-            pthread_mutex_unlock(&ctx->queueMutex);
+        if (ctx->queueEmpty) {
+            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
             return opaque;
         }
         /* Pop a job off the queue */
         {   POOL_job const job = ctx->queue[ctx->queueHead];
             ctx->queueHead = (ctx->queueHead + 1) % ctx->queueSize;
+            ctx->numThreadsBusy++;
+            ctx->queueEmpty = ctx->queueHead == ctx->queueTail;
             /* Unlock the mutex, signal a pusher, and run the job */
-            pthread_mutex_unlock(&ctx->queueMutex);
-            pthread_cond_signal(&ctx->queuePushCond);
+            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+            ZSTD_pthread_cond_signal(&ctx->queuePushCond);
+
             job.function(job.opaque);
-        }
-    }
+
+            /* If the intended queue size was 0, signal after finishing job */
+            if (ctx->queueSize == 1) {
+                ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+                ctx->numThreadsBusy--;
+                ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+                ZSTD_pthread_cond_signal(&ctx->queuePushCond);
+        }   }
+    }  /* for (;;) */
     /* Unreachable */
 }
 
-POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) {
-    POOL_ctx *ctx;
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
+    return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
+}
+
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem) {
+    POOL_ctx* ctx;
     /* Check the parameters */
-    if (!numThreads || !queueSize) { return NULL; }
+    if (!numThreads) { return NULL; }
     /* Allocate the context and zero initialize */
-    ctx = (POOL_ctx *)calloc(1, sizeof(POOL_ctx));
+    ctx = (POOL_ctx*)ZSTD_calloc(sizeof(POOL_ctx), customMem);
     if (!ctx) { return NULL; }
     /* Initialize the job queue.
      * It needs one extra space since one space is wasted to differentiate empty
      * and full queues.
      */
     ctx->queueSize = queueSize + 1;
-    ctx->queue = (POOL_job *)malloc(ctx->queueSize * sizeof(POOL_job));
+    ctx->queue = (POOL_job*)ZSTD_malloc(ctx->queueSize * sizeof(POOL_job), customMem);
     ctx->queueHead = 0;
     ctx->queueTail = 0;
-    pthread_mutex_init(&ctx->queueMutex, NULL);
-    pthread_cond_init(&ctx->queuePushCond, NULL);
-    pthread_cond_init(&ctx->queuePopCond, NULL);
+    ctx->numThreadsBusy = 0;
+    ctx->queueEmpty = 1;
+    (void)ZSTD_pthread_mutex_init(&ctx->queueMutex, NULL);
+    (void)ZSTD_pthread_cond_init(&ctx->queuePushCond, NULL);
+    (void)ZSTD_pthread_cond_init(&ctx->queuePopCond, NULL);
     ctx->shutdown = 0;
     /* Allocate space for the thread handles */
-    ctx->threads = (pthread_t *)malloc(numThreads * sizeof(pthread_t));
+    ctx->threads = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
     ctx->numThreads = 0;
+    ctx->customMem = customMem;
     /* Check for errors */
     if (!ctx->threads || !ctx->queue) { POOL_free(ctx); return NULL; }
     /* Initialize the threads */
     {   size_t i;
         for (i = 0; i < numThreads; ++i) {
-            if (pthread_create(&ctx->threads[i], NULL, &POOL_thread, ctx)) {
+            if (ZSTD_pthread_create(&ctx->threads[i], NULL, &POOL_thread, ctx)) {
                 ctx->numThreads = i;
                 POOL_free(ctx);
                 return NULL;
@@ -120,75 +146,138 @@
 /*! POOL_join() :
     Shutdown the queue, wake any sleeping threads, and join all of the threads.
 */
-static void POOL_join(POOL_ctx *ctx) {
+static void POOL_join(POOL_ctx* ctx) {
     /* Shut down the queue */
-    pthread_mutex_lock(&ctx->queueMutex);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
     ctx->shutdown = 1;
-    pthread_mutex_unlock(&ctx->queueMutex);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
     /* Wake up sleeping threads */
-    pthread_cond_broadcast(&ctx->queuePushCond);
-    pthread_cond_broadcast(&ctx->queuePopCond);
+    ZSTD_pthread_cond_broadcast(&ctx->queuePushCond);
+    ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
     /* Join all of the threads */
     {   size_t i;
         for (i = 0; i < ctx->numThreads; ++i) {
-            pthread_join(ctx->threads[i], NULL);
+            ZSTD_pthread_join(ctx->threads[i], NULL);
     }   }
 }
 
 void POOL_free(POOL_ctx *ctx) {
     if (!ctx) { return; }
     POOL_join(ctx);
-    pthread_mutex_destroy(&ctx->queueMutex);
-    pthread_cond_destroy(&ctx->queuePushCond);
-    pthread_cond_destroy(&ctx->queuePopCond);
-    if (ctx->queue) free(ctx->queue);
-    if (ctx->threads) free(ctx->threads);
-    free(ctx);
+    ZSTD_pthread_mutex_destroy(&ctx->queueMutex);
+    ZSTD_pthread_cond_destroy(&ctx->queuePushCond);
+    ZSTD_pthread_cond_destroy(&ctx->queuePopCond);
+    ZSTD_free(ctx->queue, ctx->customMem);
+    ZSTD_free(ctx->threads, ctx->customMem);
+    ZSTD_free(ctx, ctx->customMem);
+}
+
+size_t POOL_sizeof(POOL_ctx *ctx) {
+    if (ctx==NULL) return 0;  /* supports sizeof NULL */
+    return sizeof(*ctx)
+        + ctx->queueSize * sizeof(POOL_job)
+        + ctx->numThreads * sizeof(ZSTD_pthread_t);
+}
+
+/**
+ * Returns 1 if the queue is full and 0 otherwise.
+ *
+ * If the queueSize is 1 (the pool was created with an intended queueSize of 0),
+ * then a queue is empty if there is a thread free and no job is waiting.
+ */
+static int isQueueFull(POOL_ctx const* ctx) {
+    if (ctx->queueSize > 1) {
+        return ctx->queueHead == ((ctx->queueTail + 1) % ctx->queueSize);
+    } else {
+        return ctx->numThreadsBusy == ctx->numThreads ||
+               !ctx->queueEmpty;
+    }
+}
+
+
+static void POOL_add_internal(POOL_ctx* ctx, POOL_function function, void *opaque)
+{
+    POOL_job const job = {function, opaque};
+    assert(ctx != NULL);
+    if (ctx->shutdown) return;
+
+    ctx->queueEmpty = 0;
+    ctx->queue[ctx->queueTail] = job;
+    ctx->queueTail = (ctx->queueTail + 1) % ctx->queueSize;
+    ZSTD_pthread_cond_signal(&ctx->queuePopCond);
+}
+
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque)
+{
+    assert(ctx != NULL);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    /* Wait until there is space in the queue for the new job */
+    while (isQueueFull(ctx) && (!ctx->shutdown)) {
+        ZSTD_pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
+    }
+    POOL_add_internal(ctx, function, opaque);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
 }
 
-void POOL_add(void *ctxVoid, POOL_function function, void *opaque) {
-    POOL_ctx *ctx = (POOL_ctx *)ctxVoid;
-    if (!ctx) { return; }
 
-    pthread_mutex_lock(&ctx->queueMutex);
-    {   POOL_job const job = {function, opaque};
-        /* Wait until there is space in the queue for the new job */
-        size_t newTail = (ctx->queueTail + 1) % ctx->queueSize;
-        while (ctx->queueHead == newTail && !ctx->shutdown) {
-          pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
-          newTail = (ctx->queueTail + 1) % ctx->queueSize;
-        }
-        /* The queue is still going => there is space */
-        if (!ctx->shutdown) {
-            ctx->queue[ctx->queueTail] = job;
-            ctx->queueTail = newTail;
-        }
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque)
+{
+    assert(ctx != NULL);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    if (isQueueFull(ctx)) {
+        ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+        return 0;
     }
-    pthread_mutex_unlock(&ctx->queueMutex);
-    pthread_cond_signal(&ctx->queuePopCond);
+    POOL_add_internal(ctx, function, opaque);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+    return 1;
 }
 
+
 #else  /* ZSTD_MULTITHREAD  not defined */
+
+/* ========================== */
 /* No multi-threading support */
+/* ========================== */
 
-/* We don't need any data, but if it is empty malloc() might return NULL. */
+
+/* We don't need any data, but if it is empty, malloc() might return NULL. */
 struct POOL_ctx_s {
-  int data;
+    int dummy;
 };
+static POOL_ctx g_ctx;
 
-POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) {
-  (void)numThreads;
-  (void)queueSize;
-  return (POOL_ctx *)malloc(sizeof(POOL_ctx));
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
+    return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
 }
 
-void POOL_free(POOL_ctx *ctx) {
-  if (ctx) free(ctx);
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem) {
+    (void)numThreads;
+    (void)queueSize;
+    (void)customMem;
+    return &g_ctx;
+}
+
+void POOL_free(POOL_ctx* ctx) {
+    assert(!ctx || ctx == &g_ctx);
+    (void)ctx;
 }
 
-void POOL_add(void *ctx, POOL_function function, void *opaque) {
-  (void)ctx;
-  function(opaque);
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque) {
+    (void)ctx;
+    function(opaque);
+}
+
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque) {
+    (void)ctx;
+    function(opaque);
+    return 1;
+}
+
+size_t POOL_sizeof(POOL_ctx* ctx) {
+    if (ctx==NULL) return 0;  /* supports sizeof NULL */
+    assert(ctx == &g_ctx);
+    return sizeof(*ctx);
 }
 
 #endif  /* ZSTD_MULTITHREAD */
--- a/contrib/python-zstandard/zstd/common/pool.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/pool.h	Mon Apr 09 10:13:29 2018 -0700
@@ -1,11 +1,13 @@
-/**
- * Copyright (c) 2016-present, Facebook, Inc.
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
+
 #ifndef POOL_H
 #define POOL_H
 
@@ -15,38 +17,54 @@
 
 
 #include <stddef.h>   /* size_t */
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_customMem */
+#include "zstd.h"
 
 typedef struct POOL_ctx_s POOL_ctx;
 
 /*! POOL_create() :
-    Create a thread pool with at most `numThreads` threads.
-    `numThreads` must be at least 1.
-    The maximum number of queued jobs before blocking is `queueSize`.
-    `queueSize` must be at least 1.
-    @return : The POOL_ctx pointer on success else NULL.
+ *  Create a thread pool with at most `numThreads` threads.
+ * `numThreads` must be at least 1.
+ *  The maximum number of queued jobs before blocking is `queueSize`.
+ * @return : POOL_ctx pointer on success, else NULL.
 */
-POOL_ctx *POOL_create(size_t numThreads, size_t queueSize);
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize);
+
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem);
 
 /*! POOL_free() :
     Free a thread pool returned by POOL_create().
 */
-void POOL_free(POOL_ctx *ctx);
+void POOL_free(POOL_ctx* ctx);
+
+/*! POOL_sizeof() :
+    return memory usage of pool returned by POOL_create().
+*/
+size_t POOL_sizeof(POOL_ctx* ctx);
 
 /*! POOL_function :
     The function type that can be added to a thread pool.
 */
-typedef void (*POOL_function)(void *);
+typedef void (*POOL_function)(void*);
 /*! POOL_add_function :
     The function type for a generic thread pool add function.
 */
-typedef void (*POOL_add_function)(void *, POOL_function, void *);
+typedef void (*POOL_add_function)(void*, POOL_function, void*);
 
 /*! POOL_add() :
-    Add the job `function(opaque)` to the thread pool.
+    Add the job `function(opaque)` to the thread pool. `ctx` must be valid.
     Possibly blocks until there is room in the queue.
     Note : The function may be executed asynchronously, so `opaque` must live until the function has been completed.
 */
-void POOL_add(void *ctx, POOL_function function, void *opaque);
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque);
+
+
+/*! POOL_tryAdd() :
+    Add the job `function(opaque)` to the thread pool if a worker is available.
+    return immediately otherwise.
+   @return : 1 if successful, 0 if not.
+*/
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque);
 
 
 #if defined (__cplusplus)
--- a/contrib/python-zstandard/zstd/common/threading.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/threading.c	Mon Apr 09 10:13:29 2018 -0700
@@ -1,11 +1,10 @@
-
 /**
  * Copyright (c) 2016 Tino Reichardt
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
  *
  * You can contact the author at:
  * - zstdmt source repository: https://github.com/mcmilk/zstdmt
@@ -15,11 +14,8 @@
  * This file will hold wrapper for systems, which do not support pthreads
  */
 
-/* ======   Compiler specifics   ====== */
-#if defined(_MSC_VER)
-#  pragma warning(disable : 4206)        /* disable: C4206: translation unit is empty (when ZSTD_MULTITHREAD is not defined) */
-#endif
-
+/* create fake symbol to avoid empty trnaslation unit warning */
+int g_ZSTD_threading_useles_symbol;
 
 #if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
 
@@ -39,12 +35,12 @@
 
 static unsigned __stdcall worker(void *arg)
 {
-    pthread_t* const thread = (pthread_t*) arg;
+    ZSTD_pthread_t* const thread = (ZSTD_pthread_t*) arg;
     thread->arg = thread->start_routine(thread->arg);
     return 0;
 }
 
-int pthread_create(pthread_t* thread, const void* unused,
+int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
             void* (*start_routine) (void*), void* arg)
 {
     (void)unused;
@@ -58,16 +54,16 @@
         return 0;
 }
 
-int _pthread_join(pthread_t * thread, void **value_ptr)
+int ZSTD_pthread_join(ZSTD_pthread_t thread, void **value_ptr)
 {
     DWORD result;
 
-    if (!thread->handle) return 0;
+    if (!thread.handle) return 0;
 
-    result = WaitForSingleObject(thread->handle, INFINITE);
+    result = WaitForSingleObject(thread.handle, INFINITE);
     switch (result) {
     case WAIT_OBJECT_0:
-        if (value_ptr) *value_ptr = thread->arg;
+        if (value_ptr) *value_ptr = thread.arg;
         return 0;
     case WAIT_ABANDONED:
         return EINVAL;
--- a/contrib/python-zstandard/zstd/common/threading.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/threading.h	Mon Apr 09 10:13:29 2018 -0700
@@ -1,11 +1,10 @@
-
 /**
  * Copyright (c) 2016 Tino Reichardt
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
  *
  * You can contact the author at:
  * - zstdmt source repository: https://github.com/mcmilk/zstdmt
@@ -38,62 +37,82 @@
 #  define WIN32_LEAN_AND_MEAN
 #endif
 
+#undef ERROR   /* reported already defined on VS 2015 (Rich Geldreich) */
 #include <windows.h>
+#undef ERROR
+#define ERROR(name) ZSTD_ERROR(name)
+
 
 /* mutex */
-#define pthread_mutex_t           CRITICAL_SECTION
-#define pthread_mutex_init(a,b)   InitializeCriticalSection((a))
-#define pthread_mutex_destroy(a)  DeleteCriticalSection((a))
-#define pthread_mutex_lock(a)     EnterCriticalSection((a))
-#define pthread_mutex_unlock(a)   LeaveCriticalSection((a))
+#define ZSTD_pthread_mutex_t           CRITICAL_SECTION
+#define ZSTD_pthread_mutex_init(a, b)  ((void)(b), InitializeCriticalSection((a)), 0)
+#define ZSTD_pthread_mutex_destroy(a)  DeleteCriticalSection((a))
+#define ZSTD_pthread_mutex_lock(a)     EnterCriticalSection((a))
+#define ZSTD_pthread_mutex_unlock(a)   LeaveCriticalSection((a))
 
 /* condition variable */
-#define pthread_cond_t             CONDITION_VARIABLE
-#define pthread_cond_init(a, b)    InitializeConditionVariable((a))
-#define pthread_cond_destroy(a)    /* No delete */
-#define pthread_cond_wait(a, b)    SleepConditionVariableCS((a), (b), INFINITE)
-#define pthread_cond_signal(a)     WakeConditionVariable((a))
-#define pthread_cond_broadcast(a)  WakeAllConditionVariable((a))
+#define ZSTD_pthread_cond_t             CONDITION_VARIABLE
+#define ZSTD_pthread_cond_init(a, b)    ((void)(b), InitializeConditionVariable((a)), 0)
+#define ZSTD_pthread_cond_destroy(a)    ((void)(a))
+#define ZSTD_pthread_cond_wait(a, b)    SleepConditionVariableCS((a), (b), INFINITE)
+#define ZSTD_pthread_cond_signal(a)     WakeConditionVariable((a))
+#define ZSTD_pthread_cond_broadcast(a)  WakeAllConditionVariable((a))
 
-/* pthread_create() and pthread_join() */
+/* ZSTD_pthread_create() and ZSTD_pthread_join() */
 typedef struct {
     HANDLE handle;
     void* (*start_routine)(void*);
     void* arg;
-} pthread_t;
+} ZSTD_pthread_t;
 
-int pthread_create(pthread_t* thread, const void* unused,
+int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
                    void* (*start_routine) (void*), void* arg);
 
-#define pthread_join(a, b) _pthread_join(&(a), (b))
-int _pthread_join(pthread_t* thread, void** value_ptr);
+int ZSTD_pthread_join(ZSTD_pthread_t thread, void** value_ptr);
 
 /**
  * add here more wrappers as required
  */
 
 
-#elif defined(ZSTD_MULTITHREAD)   /* posix assumed ; need a better detection mathod */
+#elif defined(ZSTD_MULTITHREAD)   /* posix assumed ; need a better detection method */
 /* ===   POSIX Systems   === */
 #  include <pthread.h>
 
+#define ZSTD_pthread_mutex_t            pthread_mutex_t
+#define ZSTD_pthread_mutex_init(a, b)   pthread_mutex_init((a), (b))
+#define ZSTD_pthread_mutex_destroy(a)   pthread_mutex_destroy((a))
+#define ZSTD_pthread_mutex_lock(a)      pthread_mutex_lock((a))
+#define ZSTD_pthread_mutex_unlock(a)    pthread_mutex_unlock((a))
+
+#define ZSTD_pthread_cond_t             pthread_cond_t
+#define ZSTD_pthread_cond_init(a, b)    pthread_cond_init((a), (b))
+#define ZSTD_pthread_cond_destroy(a)    pthread_cond_destroy((a))
+#define ZSTD_pthread_cond_wait(a, b)    pthread_cond_wait((a), (b))
+#define ZSTD_pthread_cond_signal(a)     pthread_cond_signal((a))
+#define ZSTD_pthread_cond_broadcast(a)  pthread_cond_broadcast((a))
+
+#define ZSTD_pthread_t                  pthread_t
+#define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
+#define ZSTD_pthread_join(a, b)         pthread_join((a),(b))
+
 #else  /* ZSTD_MULTITHREAD not defined */
 /* No multithreading support */
 
-#define pthread_mutex_t int   /* #define rather than typedef, as sometimes pthread support is implicit, resulting in duplicated symbols */
-#define pthread_mutex_init(a,b)
-#define pthread_mutex_destroy(a)
-#define pthread_mutex_lock(a)
-#define pthread_mutex_unlock(a)
+typedef int ZSTD_pthread_mutex_t;
+#define ZSTD_pthread_mutex_init(a, b)   ((void)(a), (void)(b), 0)
+#define ZSTD_pthread_mutex_destroy(a)   ((void)(a))
+#define ZSTD_pthread_mutex_lock(a)      ((void)(a))
+#define ZSTD_pthread_mutex_unlock(a)    ((void)(a))
 
-#define pthread_cond_t int
-#define pthread_cond_init(a,b)
-#define pthread_cond_destroy(a)
-#define pthread_cond_wait(a,b)
-#define pthread_cond_signal(a)
-#define pthread_cond_broadcast(a)
+typedef int ZSTD_pthread_cond_t;
+#define ZSTD_pthread_cond_init(a, b)    ((void)(a), (void)(b), 0)
+#define ZSTD_pthread_cond_destroy(a)    ((void)(a))
+#define ZSTD_pthread_cond_wait(a, b)    ((void)(a), (void)(b))
+#define ZSTD_pthread_cond_signal(a)     ((void)(a))
+#define ZSTD_pthread_cond_broadcast(a)  ((void)(a))
 
-/* do not use pthread_t */
+/* do not use ZSTD_pthread_t */
 
 #endif /* ZSTD_MULTITHREAD */
 
--- a/contrib/python-zstandard/zstd/common/xxhash.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/xxhash.c	Mon Apr 09 10:13:29 2018 -0700
@@ -104,26 +104,34 @@
 #include <string.h>
 static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); }
 
-#define XXH_STATIC_LINKING_ONLY
+#ifndef XXH_STATIC_LINKING_ONLY
+#  define XXH_STATIC_LINKING_ONLY
+#endif
 #include "xxhash.h"
 
 
 /* *************************************
 *  Compiler Specific Options
 ***************************************/
-#ifdef _MSC_VER    /* Visual Studio */
-#  pragma warning(disable : 4127)      /* disable: C4127: conditional expression is constant */
-#  define FORCE_INLINE static __forceinline
+#if defined (__GNUC__) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#  define INLINE_KEYWORD inline
 #else
-#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
-#    ifdef __GNUC__
-#      define FORCE_INLINE static inline __attribute__((always_inline))
-#    else
-#      define FORCE_INLINE static inline
-#    endif
-#  else
-#    define FORCE_INLINE static
-#  endif /* __STDC_VERSION__ */
+#  define INLINE_KEYWORD
+#endif
+
+#if defined(__GNUC__)
+#  define FORCE_INLINE_ATTR __attribute__((always_inline))
+#elif defined(_MSC_VER)
+#  define FORCE_INLINE_ATTR __forceinline
+#else
+#  define FORCE_INLINE_ATTR
+#endif
+
+#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
+
+
+#ifdef _MSC_VER
+#  pragma warning(disable : 4127)      /* disable: C4127: conditional expression is constant */
 #endif
 
 
@@ -246,7 +254,7 @@
 *****************************/
 typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment;
 
-FORCE_INLINE U32 XXH_readLE32_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
+FORCE_INLINE_TEMPLATE U32 XXH_readLE32_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
 {
     if (align==XXH_unaligned)
         return endian==XXH_littleEndian ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
@@ -254,7 +262,7 @@
         return endian==XXH_littleEndian ? *(const U32*)ptr : XXH_swap32(*(const U32*)ptr);
 }
 
-FORCE_INLINE U32 XXH_readLE32(const void* ptr, XXH_endianess endian)
+FORCE_INLINE_TEMPLATE U32 XXH_readLE32(const void* ptr, XXH_endianess endian)
 {
     return XXH_readLE32_align(ptr, endian, XXH_unaligned);
 }
@@ -264,7 +272,7 @@
     return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
 }
 
-FORCE_INLINE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
+FORCE_INLINE_TEMPLATE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
 {
     if (align==XXH_unaligned)
         return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
@@ -272,7 +280,7 @@
         return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr);
 }
 
-FORCE_INLINE U64 XXH_readLE64(const void* ptr, XXH_endianess endian)
+FORCE_INLINE_TEMPLATE U64 XXH_readLE64(const void* ptr, XXH_endianess endian)
 {
     return XXH_readLE64_align(ptr, endian, XXH_unaligned);
 }
@@ -333,7 +341,7 @@
     return seed;
 }
 
-FORCE_INLINE U32 XXH32_endian_align(const void* input, size_t len, U32 seed, XXH_endianess endian, XXH_alignment align)
+FORCE_INLINE_TEMPLATE U32 XXH32_endian_align(const void* input, size_t len, U32 seed, XXH_endianess endian, XXH_alignment align)
 {
     const BYTE* p = (const BYTE*)input;
     const BYTE* bEnd = p + len;
@@ -433,7 +441,7 @@
     return acc;
 }
 
-FORCE_INLINE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align)
+FORCE_INLINE_TEMPLATE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align)
 {
     const BYTE* p = (const BYTE*)input;
     const BYTE* const bEnd = p + len;
@@ -582,7 +590,7 @@
 }
 
 
-FORCE_INLINE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian)
+FORCE_INLINE_TEMPLATE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian)
 {
     const BYTE* p = (const BYTE*)input;
     const BYTE* const bEnd = p + len;
@@ -652,7 +660,7 @@
 
 
 
-FORCE_INLINE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian)
+FORCE_INLINE_TEMPLATE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian)
 {
     const BYTE * p = (const BYTE*)state->mem32;
     const BYTE* const bEnd = (const BYTE*)(state->mem32) + state->memsize;
@@ -702,7 +710,7 @@
 
 /* **** XXH64 **** */
 
-FORCE_INLINE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian)
+FORCE_INLINE_TEMPLATE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian)
 {
     const BYTE* p = (const BYTE*)input;
     const BYTE* const bEnd = p + len;
@@ -769,7 +777,7 @@
 
 
 
-FORCE_INLINE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian)
+FORCE_INLINE_TEMPLATE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian)
 {
     const BYTE * p = (const BYTE*)state->mem64;
     const BYTE* const bEnd = (const BYTE*)state->mem64 + state->memsize;
--- a/contrib/python-zstandard/zstd/common/xxhash.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/xxhash.h	Mon Apr 09 10:13:29 2018 -0700
@@ -64,16 +64,12 @@
 XXH32        6.8 GB/s            6.0 GB/s
 */
 
-#ifndef XXHASH_H_5627135585666179
-#define XXHASH_H_5627135585666179 1
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
-#ifndef XXH_NAMESPACE
-#  define XXH_NAMESPACE ZSTD_  /* Zstandard specific */
-#endif
+#ifndef XXHASH_H_5627135585666179
+#define XXHASH_H_5627135585666179 1
 
 
 /* ****************************
@@ -242,6 +238,11 @@
 /* **************************
 *  Canonical representation
 ****************************/
+/* Default result type for XXH functions are primitive unsigned 32 and 64 bits.
+*  The canonical representation uses human-readable write convention, aka big-endian (large digits first).
+*  These functions allow transformation of hash result into and from its canonical format.
+*  This way, hash values can be written into a file / memory, and remain comparable on different systems and programs.
+*/
 typedef struct { unsigned char digest[4]; } XXH32_canonical_t;
 typedef struct { unsigned char digest[8]; } XXH64_canonical_t;
 
@@ -251,14 +252,9 @@
 XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
 XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src);
 
-/* Default result type for XXH functions are primitive unsigned 32 and 64 bits.
-*  The canonical representation uses human-readable write convention, aka big-endian (large digits first).
-*  These functions allow transformation of hash result into and from its canonical format.
-*  This way, hash values can be written into a file / memory, and remain comparable on different systems and programs.
-*/
+#endif /* XXHASH_H_5627135585666179 */
 
 
-#ifdef XXH_STATIC_LINKING_ONLY
 
 /* ================================================================================================
    This section contains definitions which are not guaranteed to remain stable.
@@ -266,6 +262,8 @@
    They shall only be used with static linking.
    Never use these definitions in association with dynamic linking !
 =================================================================================================== */
+#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXH_STATIC_H_3543687687345)
+#define XXH_STATIC_H_3543687687345
 
 /* These definitions are only meant to allow allocation of XXH state
    statically, on stack, or in a struct for example.
@@ -299,11 +297,9 @@
 #    include "xxhash.c"   /* include xxhash functions as `static`, for inlining */
 #  endif
 
-#endif /* XXH_STATIC_LINKING_ONLY */
+#endif /* XXH_STATIC_LINKING_ONLY && XXH_STATIC_H_3543687687345 */
 
 
 #if defined (__cplusplus)
 }
 #endif
-
-#endif /* XXHASH_H_5627135585666179 */
--- a/contrib/python-zstandard/zstd/common/zstd_common.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/zstd_common.c	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,11 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 
@@ -12,62 +13,74 @@
 /*-*************************************
 *  Dependencies
 ***************************************/
-#include <stdlib.h>         /* malloc */
+#include <stdlib.h>      /* malloc, calloc, free */
+#include <string.h>      /* memset */
 #include "error_private.h"
-#define ZSTD_STATIC_LINKING_ONLY
-#include "zstd.h"           /* declaration of ZSTD_isError, ZSTD_getErrorName, ZSTD_getErrorCode, ZSTD_getErrorString, ZSTD_versionNumber */
+#include "zstd_internal.h"
 
 
 /*-****************************************
 *  Version
 ******************************************/
-unsigned ZSTD_versionNumber (void) { return ZSTD_VERSION_NUMBER; }
+unsigned ZSTD_versionNumber(void) { return ZSTD_VERSION_NUMBER; }
+
+const char* ZSTD_versionString(void) { return ZSTD_VERSION_STRING; }
 
 
 /*-****************************************
 *  ZSTD Error Management
 ******************************************/
 /*! ZSTD_isError() :
-*   tells if a return value is an error code */
+ *  tells if a return value is an error code */
 unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
 
 /*! ZSTD_getErrorName() :
-*   provides error code string from function result (useful for debugging) */
+ *  provides error code string from function result (useful for debugging) */
 const char* ZSTD_getErrorName(size_t code) { return ERR_getErrorName(code); }
 
 /*! ZSTD_getError() :
-*   convert a `size_t` function result into a proper ZSTD_errorCode enum */
+ *  convert a `size_t` function result into a proper ZSTD_errorCode enum */
 ZSTD_ErrorCode ZSTD_getErrorCode(size_t code) { return ERR_getErrorCode(code); }
 
 /*! ZSTD_getErrorString() :
-*   provides error code string from enum */
-const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorName(code); }
+ *  provides error code string from enum */
+const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString(code); }
+
+/*! g_debuglog_enable :
+ *  turn on/off debug traces (global switch) */
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG >= 2)
+int g_debuglog_enable = 1;
+#endif
 
 
 /*=**************************************************************
 *  Custom allocator
 ****************************************************************/
-/* default uses stdlib */
-void* ZSTD_defaultAllocFunction(void* opaque, size_t size)
+void* ZSTD_malloc(size_t size, ZSTD_customMem customMem)
 {
-    void* address = malloc(size);
-    (void)opaque;
-    return address;
+    if (customMem.customAlloc)
+        return customMem.customAlloc(customMem.opaque, size);
+    return malloc(size);
 }
 
-void ZSTD_defaultFreeFunction(void* opaque, void* address)
+void* ZSTD_calloc(size_t size, ZSTD_customMem customMem)
 {
-    (void)opaque;
-    free(address);
-}
-
-void* ZSTD_malloc(size_t size, ZSTD_customMem customMem)
-{
-    return customMem.customAlloc(customMem.opaque, size);
+    if (customMem.customAlloc) {
+        /* calloc implemented as malloc+memset;
+         * not as efficient as calloc, but next best guess for custom malloc */
+        void* const ptr = customMem.customAlloc(customMem.opaque, size);
+        memset(ptr, 0, size);
+        return ptr;
+    }
+    return calloc(1, size);
 }
 
 void ZSTD_free(void* ptr, ZSTD_customMem customMem)
 {
-    if (ptr!=NULL)
-        customMem.customFree(customMem.opaque, ptr);
+    if (ptr!=NULL) {
+        if (customMem.customFree)
+            customMem.customFree(customMem.opaque, ptr);
+        else
+            free(ptr);
+    }
 }
--- a/contrib/python-zstandard/zstd/common/zstd_errors.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/zstd_errors.h	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,11 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 #ifndef ZSTD_ERRORS_H_398273423
@@ -19,10 +20,12 @@
 
 
 /* =====   ZSTDERRORLIB_API : control library symbols visibility   ===== */
-#if defined(__GNUC__) && (__GNUC__ >= 4)
-#  define ZSTDERRORLIB_VISIBILITY __attribute__ ((visibility ("default")))
-#else
-#  define ZSTDERRORLIB_VISIBILITY
+#ifndef ZSTDERRORLIB_VISIBILITY
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define ZSTDERRORLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#  else
+#    define ZSTDERRORLIB_VISIBILITY
+#  endif
 #endif
 #if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
 #  define ZSTDERRORLIB_API __declspec(dllexport) ZSTDERRORLIB_VISIBILITY
@@ -32,39 +35,54 @@
 #  define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBILITY
 #endif
 
-/*-****************************************
-*  error codes list
-******************************************/
+/*-*********************************************
+ *  Error codes list
+ *-*********************************************
+ *  Error codes _values_ are pinned down since v1.3.1 only.
+ *  Therefore, don't rely on values if you may link to any version < v1.3.1.
+ *
+ *  Only values < 100 are considered stable.
+ *
+ *  note 1 : this API shall be used with static linking only.
+ *           dynamic linking is not yet officially supported.
+ *  note 2 : Prefer relying on the enum than on its value whenever possible
+ *           This is the only supported way to use the error list < v1.3.1
+ *  note 3 : ZSTD_isError() is always correct, whatever the library version.
+ **********************************************/
 typedef enum {
-  ZSTD_error_no_error,
-  ZSTD_error_GENERIC,
-  ZSTD_error_prefix_unknown,
-  ZSTD_error_version_unsupported,
-  ZSTD_error_parameter_unknown,
-  ZSTD_error_frameParameter_unsupported,
-  ZSTD_error_frameParameter_unsupportedBy32bits,
-  ZSTD_error_frameParameter_windowTooLarge,
-  ZSTD_error_compressionParameter_unsupported,
-  ZSTD_error_init_missing,
-  ZSTD_error_memory_allocation,
-  ZSTD_error_stage_wrong,
-  ZSTD_error_dstSize_tooSmall,
-  ZSTD_error_srcSize_wrong,
-  ZSTD_error_corruption_detected,
-  ZSTD_error_checksum_wrong,
-  ZSTD_error_tableLog_tooLarge,
-  ZSTD_error_maxSymbolValue_tooLarge,
-  ZSTD_error_maxSymbolValue_tooSmall,
-  ZSTD_error_dictionary_corrupted,
-  ZSTD_error_dictionary_wrong,
-  ZSTD_error_maxCode
+  ZSTD_error_no_error = 0,
+  ZSTD_error_GENERIC  = 1,
+  ZSTD_error_prefix_unknown                = 10,
+  ZSTD_error_version_unsupported           = 12,
+  ZSTD_error_frameParameter_unsupported    = 14,
+  ZSTD_error_frameParameter_windowTooLarge = 16,
+  ZSTD_error_corruption_detected = 20,
+  ZSTD_error_checksum_wrong      = 22,
+  ZSTD_error_dictionary_corrupted      = 30,
+  ZSTD_error_dictionary_wrong          = 32,
+  ZSTD_error_dictionaryCreation_failed = 34,
+  ZSTD_error_parameter_unsupported   = 40,
+  ZSTD_error_parameter_outOfBound    = 42,
+  ZSTD_error_tableLog_tooLarge       = 44,
+  ZSTD_error_maxSymbolValue_tooLarge = 46,
+  ZSTD_error_maxSymbolValue_tooSmall = 48,
+  ZSTD_error_stage_wrong       = 60,
+  ZSTD_error_init_missing      = 62,
+  ZSTD_error_memory_allocation = 64,
+  ZSTD_error_workSpace_tooSmall= 66,
+  ZSTD_error_dstSize_tooSmall = 70,
+  ZSTD_error_srcSize_wrong    = 72,
+  /* following error codes are __NOT STABLE__, they can be removed or changed in future versions */
+  ZSTD_error_frameIndex_tooLarge = 100,
+  ZSTD_error_seekableIO          = 102,
+  ZSTD_error_maxCode = 120  /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
 } ZSTD_ErrorCode;
 
 /*! ZSTD_getErrorCode() :
     convert a `size_t` function result into a `ZSTD_ErrorCode` enum type,
-    which can be used to compare directly with enum list published into "error_public.h" */
+    which can be used to compare with enum list published above */
 ZSTDERRORLIB_API ZSTD_ErrorCode ZSTD_getErrorCode(size_t functionResult);
-ZSTDERRORLIB_API const char* ZSTD_getErrorString(ZSTD_ErrorCode code);
+ZSTDERRORLIB_API const char* ZSTD_getErrorString(ZSTD_ErrorCode code);   /**< Same as ZSTD_getErrorName, but using a `ZSTD_ErrorCode` enum argument */
 
 
 #if defined (__cplusplus)
--- a/contrib/python-zstandard/zstd/common/zstd_internal.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/common/zstd_internal.h	Mon Apr 09 10:13:29 2018 -0700
@@ -1,59 +1,87 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 #ifndef ZSTD_CCOMMON_H_MODULE
 #define ZSTD_CCOMMON_H_MODULE
 
-/*-*******************************************************
-*  Compiler specifics
-*********************************************************/
-#ifdef _MSC_VER    /* Visual Studio */
-#  define FORCE_INLINE static __forceinline
-#  include <intrin.h>                    /* For Visual 2005 */
-#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
-#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
-#  pragma warning(disable : 4100)        /* disable: C4100: unreferenced formal parameter */
-#else
-#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
-#    ifdef __GNUC__
-#      define FORCE_INLINE static inline __attribute__((always_inline))
-#    else
-#      define FORCE_INLINE static inline
-#    endif
-#  else
-#    define FORCE_INLINE static
-#  endif /* __STDC_VERSION__ */
+/* this module contains definitions which must be identical
+ * across compression, decompression and dictBuilder.
+ * It also contains a few functions useful to at least 2 of them
+ * and which benefit from being inlined */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "compiler.h"
+#include "mem.h"
+#include "error_private.h"
+#define ZSTD_STATIC_LINKING_ONLY
+#include "zstd.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+#ifndef XXH_STATIC_LINKING_ONLY
+#  define XXH_STATIC_LINKING_ONLY  /* XXH64_state_t */
 #endif
+#include "xxhash.h"                /* XXH_reset, update, digest */
 
-#ifdef _MSC_VER
-#  define FORCE_NOINLINE static __declspec(noinline)
-#else
-#  ifdef __GNUC__
-#    define FORCE_NOINLINE static __attribute__((__noinline__))
-#  else
-#    define FORCE_NOINLINE static
-#  endif
+
+#if defined (__cplusplus)
+extern "C" {
 #endif
 
 
 /*-*************************************
-*  Dependencies
+*  Debug
 ***************************************/
-#include "mem.h"
-#include "error_private.h"
-#define ZSTD_STATIC_LINKING_ONLY
-#include "zstd.h"
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
+#  include <assert.h>
+#else
+#  ifndef assert
+#    define assert(condition) ((void)0)
+#  endif
+#endif
+
+#define ZSTD_STATIC_ASSERT(c) { enum { ZSTD_static_assert = 1/(int)(!!(c)) }; }
+
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
+#  include <stdio.h>
+extern int g_debuglog_enable;
+/* recommended values for ZSTD_DEBUG display levels :
+ * 1 : no display, enables assert() only
+ * 2 : reserved for currently active debug path
+ * 3 : events once per object lifetime (CCtx, CDict, etc.)
+ * 4 : events once per frame
+ * 5 : events once per block
+ * 6 : events once per sequence (*very* verbose) */
+#  define RAWLOG(l, ...) {                                      \
+                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
+                    fprintf(stderr, __VA_ARGS__);               \
+            }   }
+#  define DEBUGLOG(l, ...) {                                    \
+                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
+                    fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
+                    fprintf(stderr, " \n");                     \
+            }   }
+#else
+#  define RAWLOG(l, ...)      {}    /* disabled */
+#  define DEBUGLOG(l, ...)    {}    /* disabled */
+#endif
 
 
 /*-*************************************
 *  shared macros
 ***************************************/
+#undef MIN
+#undef MAX
 #define MIN(a,b) ((a)<(b) ? (a) : (b))
 #define MAX(a,b) ((a)>(b) ? (a) : (b))
 #define CHECK_F(f) { size_t const errcod = f; if (ERR_isError(errcod)) return errcod; }  /* check and Forward error code */
@@ -64,12 +92,9 @@
 *  Common constants
 ***************************************/
 #define ZSTD_OPT_NUM    (1<<12)
-#define ZSTD_DICT_MAGIC  0xEC30A437   /* v0.7+ */
 
 #define ZSTD_REP_NUM      3                 /* number of repcodes */
-#define ZSTD_REP_CHECK    (ZSTD_REP_NUM)    /* number of repcodes to check by the optimal parser */
 #define ZSTD_REP_MOVE     (ZSTD_REP_NUM-1)
-#define ZSTD_REP_MOVE_OPT (ZSTD_REP_NUM)
 static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
 
 #define KB *(1 <<10)
@@ -84,9 +109,13 @@
 #define BIT0   1
 
 #define ZSTD_WINDOWLOG_ABSOLUTEMIN 10
+#define ZSTD_WINDOWLOG_DEFAULTMAX 27 /* Default maximum allowed window log */
 static const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
 static const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
 
+#define ZSTD_FRAMEIDSIZE 4
+static const size_t ZSTD_frameIdSize = ZSTD_FRAMEIDSIZE;  /* magic number size */
+
 #define ZSTD_BLOCKHEADERSIZE 3   /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
 static const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
 typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
@@ -100,40 +129,53 @@
 #define LONGNBSEQ 0x7F00
 
 #define MINMATCH 3
-#define EQUAL_READ32 4
 
 #define Litbits  8
 #define MaxLit ((1<<Litbits) - 1)
-#define MaxML  52
-#define MaxLL  35
-#define MaxOff 28
+#define MaxML   52
+#define MaxLL   35
+#define DefaultMaxOff 28
+#define MaxOff  31
 #define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
 #define MLFSELog    9
 #define LLFSELog    9
 #define OffFSELog   8
+#define MaxFSELog  MAX(MAX(MLFSELog, LLFSELog), OffFSELog)
 
-static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9,10,11,12,
+static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3,
+                                      4, 6, 7, 8, 9,10,11,12,
                                      13,14,15,16 };
-static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1,
-                                             2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1,
+static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2,
+                                             2, 2, 2, 2, 2, 1, 1, 1,
+                                             2, 2, 2, 2, 2, 2, 2, 2,
+                                             2, 3, 2, 1, 1, 1, 1, 1,
                                             -1,-1,-1,-1 };
 #define LL_DEFAULTNORMLOG 6  /* for static allocation */
 static const U32 LL_defaultNormLog = LL_DEFAULTNORMLOG;
 
-static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9,10,11,
+static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3,
+                                      4, 4, 5, 7, 8, 9,10,11,
                                      12,13,14,15,16 };
-static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
-                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
-                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,-1,-1,
+static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2,
+                                             2, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1,-1,-1,
                                             -1,-1,-1,-1,-1 };
 #define ML_DEFAULTNORMLOG 6  /* for static allocation */
 static const U32 ML_defaultNormLog = ML_DEFAULTNORMLOG;
 
-static const S16 OF_defaultNorm[MaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
-                                              1, 1, 1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1 };
+static const S16 OF_defaultNorm[DefaultMaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2,
+                                                     2, 1, 1, 1, 1, 1, 1, 1,
+                                                     1, 1, 1, 1, 1, 1, 1, 1,
+                                                    -1,-1,-1,-1,-1 };
 #define OF_DEFAULTNORMLOG 5  /* for static allocation */
 static const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
 
@@ -145,7 +187,7 @@
 #define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
 
 /*! ZSTD_wildcopy() :
-*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+ *  custom version of memcpy(), can overwrite up to WILDCOPY_OVERLENGTH bytes (if length==0) */
 #define WILDCOPY_OVERLENGTH 8
 MEM_STATIC void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
 {
@@ -169,31 +211,14 @@
 
 
 /*-*******************************************
-*  Private interfaces
+*  Private declarations
 *********************************************/
-typedef struct ZSTD_stats_s ZSTD_stats_t;
-
-typedef struct {
-    U32 off;
-    U32 len;
-} ZSTD_match_t;
-
-typedef struct {
-    U32 price;
-    U32 off;
-    U32 mlen;
-    U32 litlen;
-    U32 rep[ZSTD_REP_NUM];
-} ZSTD_optimal_t;
-
-
 typedef struct seqDef_s {
     U32 offset;
     U16 litLength;
     U16 matchLength;
 } seqDef;
 
-
 typedef struct {
     seqDef* sequencesStart;
     seqDef* sequences;
@@ -204,76 +229,62 @@
     BYTE* ofCode;
     U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
     U32   longLengthPos;
-    /* opt */
-    ZSTD_optimal_t* priceTable;
-    ZSTD_match_t* matchTable;
-    U32* matchLengthFreq;
-    U32* litLengthFreq;
-    U32* litFreq;
-    U32* offCodeFreq;
-    U32  matchLengthSum;
-    U32  matchSum;
-    U32  litLengthSum;
-    U32  litSum;
-    U32  offCodeSum;
-    U32  log2matchLengthSum;
-    U32  log2matchSum;
-    U32  log2litLengthSum;
-    U32  log2litSum;
-    U32  log2offCodeSum;
-    U32  factor;
-    U32  staticPrices;
-    U32  cachedPrice;
-    U32  cachedLitLength;
-    const BYTE* cachedLiterals;
 } seqStore_t;
 
-const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);
-void ZSTD_seqToCodes(const seqStore_t* seqStorePtr);
-int ZSTD_isSkipFrame(ZSTD_DCtx* dctx);
+const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);   /* compress & dictBuilder */
+void ZSTD_seqToCodes(const seqStore_t* seqStorePtr);   /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
 
 /* custom memory allocation functions */
-void* ZSTD_defaultAllocFunction(void* opaque, size_t size);
-void ZSTD_defaultFreeFunction(void* opaque, void* address);
-#ifndef ZSTD_DLL_IMPORT
-static const ZSTD_customMem defaultCustomMem = { ZSTD_defaultAllocFunction, ZSTD_defaultFreeFunction, NULL };
-#endif
 void* ZSTD_malloc(size_t size, ZSTD_customMem customMem);
+void* ZSTD_calloc(size_t size, ZSTD_customMem customMem);
 void ZSTD_free(void* ptr, ZSTD_customMem customMem);
 
 
-/*======  common function  ======*/
-
-MEM_STATIC U32 ZSTD_highbit32(U32 val)
+MEM_STATIC U32 ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus */
 {
+    assert(val != 0);
+    {
 #   if defined(_MSC_VER)   /* Visual */
-    unsigned long r=0;
-    _BitScanReverse(&r, val);
-    return (unsigned)r;
+        unsigned long r=0;
+        _BitScanReverse(&r, val);
+        return (unsigned)r;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* GCC Intrinsic */
-    return 31 - __builtin_clz(val);
+        return 31 - __builtin_clz(val);
 #   else   /* Software version */
-    static const int DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
-    U32 v = val;
-    int r;
-    v |= v >> 1;
-    v |= v >> 2;
-    v |= v >> 4;
-    v |= v >> 8;
-    v |= v >> 16;
-    r = DeBruijnClz[(U32)(v * 0x07C4ACDDU) >> 27];
-    return r;
+        static const U32 DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+        U32 v = val;
+        v |= v >> 1;
+        v |= v >> 2;
+        v |= v >> 4;
+        v |= v >> 8;
+        v |= v >> 16;
+        return DeBruijnClz[(v * 0x07C4ACDDU) >> 27];
 #   endif
+    }
 }
 
 
-/* hidden functions */
-
 /* ZSTD_invalidateRepCodes() :
  * ensures next compression will not use repcodes from previous block.
  * Note : only works with regular variant;
  *        do not use with extDict variant ! */
-void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx);
+void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx);   /* zstdmt, adaptive_compression (shouldn't get this definition from here) */
 
 
+typedef struct {
+    blockType_e blockType;
+    U32 lastBlock;
+    U32 origSize;
+} blockProperties_t;
+
+/*! ZSTD_getcBlockSize() :
+ *  Provides the size of compressed block from block header `src` */
+/* Used by: decompress, fullbench (does not get its definition from here) */
+size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
+                          blockProperties_t* bpPtr);
+
+#if defined (__cplusplus)
+}
+#endif
+
 #endif   /* ZSTD_CCOMMON_H_MODULE */
--- a/contrib/python-zstandard/zstd/compress/fse_compress.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/compress/fse_compress.c	Mon Apr 09 10:13:29 2018 -0700
@@ -33,40 +33,22 @@
 ****************************************************************** */
 
 /* **************************************************************
-*  Compiler specifics
-****************************************************************/
-#ifdef _MSC_VER    /* Visual Studio */
-#  define FORCE_INLINE static __forceinline
-#  include <intrin.h>                    /* For Visual 2005 */
-#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
-#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
-#else
-#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
-#    ifdef __GNUC__
-#      define FORCE_INLINE static inline __attribute__((always_inline))
-#    else
-#      define FORCE_INLINE static inline
-#    endif
-#  else
-#    define FORCE_INLINE static
-#  endif /* __STDC_VERSION__ */
-#endif
-
-
-/* **************************************************************
 *  Includes
 ****************************************************************/
 #include <stdlib.h>     /* malloc, free, qsort */
 #include <string.h>     /* memcpy, memset */
 #include <stdio.h>      /* printf (debug) */
 #include "bitstream.h"
+#include "compiler.h"
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
+#include "error_private.h"
 
 
 /* **************************************************************
 *  Error Management
 ****************************************************************/
+#define FSE_isError ERR_isError
 #define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
 
 
@@ -201,8 +183,6 @@
     return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
 }
 
-static short FSE_abs(short a) { return (short)(a<0 ? -a : a); }
-
 static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
                                        const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
                                        unsigned writeIsSafe)
@@ -258,17 +238,17 @@
                 bitStream >>= 16;
                 bitCount -= 16;
         }   }
-        {   short count = normalizedCounter[charnum++];
-            const short max = (short)((2*threshold-1)-remaining);
-            remaining -= FSE_abs(count);
-            if (remaining<1) return ERROR(GENERIC);
+        {   int count = normalizedCounter[charnum++];
+            int const max = (2*threshold-1)-remaining;
+            remaining -= count < 0 ? -count : count;
             count++;   /* +1 for extra accuracy */
             if (count>=threshold) count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
             bitStream += count << bitCount;
             bitCount  += nbBits;
             bitCount  -= (count<max);
             previous0  = (count==1);
-            while (remaining<threshold) nbBits--, threshold>>=1;
+            if (remaining<1) return ERROR(GENERIC);
+            while (remaining<threshold) { nbBits--; threshold>>=1; }
         }
         if (bitCount>16) {
             if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
@@ -293,7 +273,7 @@
 
 size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
 {
-    if (tableLog > FSE_MAX_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
     if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
 
     if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
@@ -312,7 +292,7 @@
     It doesn't use any additional memory.
     But this function is unsafe : it doesn't check that all values within `src` can fit into `count`.
     For this reason, prefer using a table `count` with 256 elements.
-    @return : count of most numerous element
+    @return : count of most numerous element.
 */
 size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
                         const void* src, size_t srcSize)
@@ -325,7 +305,10 @@
     memset(count, 0, (maxSymbolValue+1)*sizeof(*count));
     if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
 
-    while (ip<end) count[*ip++]++;
+    while (ip<end) {
+        assert(*ip <= maxSymbolValue);
+        count[*ip++]++;
+    }
 
     while (!count[maxSymbolValue]) maxSymbolValue--;
     *maxSymbolValuePtr = maxSymbolValue;
@@ -338,7 +321,8 @@
 
 /* FSE_count_parallel_wksp() :
  * Same as FSE_count_parallel(), but using an externally provided scratch buffer.
- * `workSpace` size must be a minimum of `1024 * sizeof(unsigned)`` */
+ * `workSpace` size must be a minimum of `1024 * sizeof(unsigned)`.
+ * @return : largest histogram frequency, or an error code (notably when histogram would be larger than *maxSymbolValuePtr). */
 static size_t FSE_count_parallel_wksp(
                                 unsigned* count, unsigned* maxSymbolValuePtr,
                                 const void* source, size_t sourceSize,
@@ -353,7 +337,7 @@
     U32* const Counting3 = Counting2 + 256;
     U32* const Counting4 = Counting3 + 256;
 
-    memset(Counting1, 0, 4*256*sizeof(unsigned));
+    memset(workSpace, 0, 4*256*sizeof(unsigned));
 
     /* safety checks */
     if (!sourceSize) {
@@ -399,7 +383,9 @@
             if (Counting1[s]) return ERROR(maxSymbolValue_tooSmall);
     }   }
 
-    {   U32 s; for (s=0; s<=maxSymbolValue; s++) {
+    {   U32 s;
+        if (maxSymbolValue > 255) maxSymbolValue = 255;
+        for (s=0; s<=maxSymbolValue; s++) {
             count[s] = Counting1[s] + Counting2[s] + Counting3[s] + Counting4[s];
             if (count[s] > max) max = count[s];
     }   }
@@ -413,9 +399,11 @@
  * Same as FSE_countFast(), but using an externally provided scratch buffer.
  * `workSpace` size must be table of >= `1024` unsigned */
 size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
-                     const void* source, size_t sourceSize, unsigned* workSpace)
+                          const void* source, size_t sourceSize,
+                          unsigned* workSpace)
 {
-    if (sourceSize < 1500) return FSE_count_simple(count, maxSymbolValuePtr, source, sourceSize);
+    if (sourceSize < 1500) /* heuristic threshold */
+        return FSE_count_simple(count, maxSymbolValuePtr, source, sourceSize);
     return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 0, workSpace);
 }
 
@@ -478,20 +466,22 @@
 /* provides the minimum logSize to safely represent a distribution */
 static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
 {
-	U32 minBitsSrc = BIT_highbit32((U32)(srcSize - 1)) + 1;
-	U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
-	U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
-	return minBits;
+    U32 minBitsSrc = BIT_highbit32((U32)(srcSize - 1)) + 1;
+    U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
+    U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    return minBits;
 }
 
 unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
 {
-	U32 maxBitsSrc = BIT_highbit32((U32)(srcSize - 1)) - minus;
+    U32 maxBitsSrc = BIT_highbit32((U32)(srcSize - 1)) - minus;
     U32 tableLog = maxTableLog;
-	U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
+    U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
     if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
-	if (maxBitsSrc < tableLog) tableLog = maxBitsSrc;   /* Accuracy can be reduced */
-	if (minBits > tableLog) tableLog = minBits;   /* Need a minimum to safely represent all symbol values */
+    if (maxBitsSrc < tableLog) tableLog = maxBitsSrc;   /* Accuracy can be reduced */
+    if (minBits > tableLog) tableLog = minBits;   /* Need a minimum to safely represent all symbol values */
     if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
     if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
     return tableLog;
@@ -508,6 +498,7 @@
 
 static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue)
 {
+    short const NOT_YET_ASSIGNED = -2;
     U32 s;
     U32 distributed = 0;
     U32 ToDistribute;
@@ -533,7 +524,8 @@
             total -= count[s];
             continue;
         }
-        norm[s]=-2;
+
+        norm[s]=NOT_YET_ASSIGNED;
     }
     ToDistribute = (1 << tableLog) - distributed;
 
@@ -541,7 +533,7 @@
         /* risk of rounding to zero */
         lowOne = (U32)((total * 3) / (ToDistribute * 2));
         for (s=0; s<=maxSymbolValue; s++) {
-            if ((norm[s] == -2) && (count[s] <= lowOne)) {
+            if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
                 norm[s] = 1;
                 distributed++;
                 total -= count[s];
@@ -556,17 +548,24 @@
            find max, then give all remaining points to max */
         U32 maxV = 0, maxC = 0;
         for (s=0; s<=maxSymbolValue; s++)
-            if (count[s] > maxC) maxV=s, maxC=count[s];
+            if (count[s] > maxC) { maxV=s; maxC=count[s]; }
         norm[maxV] += (short)ToDistribute;
         return 0;
     }
 
+    if (total == 0) {
+        /* all of the symbols were low enough for the lowOne or lowThreshold */
+        for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
+            if (norm[s] > 0) { ToDistribute--; norm[s]++; }
+        return 0;
+    }
+
     {   U64 const vStepLog = 62 - tableLog;
         U64 const mid = (1ULL << (vStepLog-1)) - 1;
         U64 const rStep = ((((U64)1<<vStepLog) * ToDistribute) + mid) / total;   /* scale on remaining */
         U64 tmpTotal = mid;
         for (s=0; s<=maxSymbolValue; s++) {
-            if (norm[s]==-2) {
+            if (norm[s]==NOT_YET_ASSIGNED) {
                 U64 const end = tmpTotal + (count[s] * rStep);
                 U32 const sStart = (U32)(tmpTotal >> vStepLog);
                 U32 const sEnd = (U32)(end >> vStepLog);
@@ -591,7 +590,7 @@
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported size */
     if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC);   /* Too small tableLog, compression potentially impossible */
 
-    {   U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
+    {   static U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
         U64 const scale = 62 - tableLog;
         U64 const step = ((U64)1<<62) / total;   /* <== here, one division ! */
         U64 const vStep = 1ULL<<(scale-20);
@@ -613,7 +612,7 @@
                     U64 restToBeat = vStep * rtbTable[proba];
                     proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
                 }
-                if (proba > largestP) largestP=proba, largest=s;
+                if (proba > largestP) { largestP=proba; largest=s; }
                 normalizedCounter[s] = proba;
                 stillToDistribute -= proba;
         }   }
@@ -774,7 +773,7 @@
 
 size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
 
-#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return f
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
 #define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
 
 /* FSE_compress_wksp() :
@@ -801,7 +800,7 @@
     if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
 
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(maxCount, FSE_count(count, &maxSymbolValue, src, srcSize) );
+    {   CHECK_V_F(maxCount, FSE_count_wksp(count, &maxSymbolValue, src, srcSize, (unsigned*)scratchBuffer) );
         if (maxCount == srcSize) return 1;   /* only a single symbol in src : rle */
         if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
         if (maxCount < (srcSize >> 7)) return 0;   /* Heuristic : not compressible enough */
--- a/contrib/python-zstandard/zstd/compress/huf_compress.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/compress/huf_compress.c	Mon Apr 09 10:13:29 2018 -0700
@@ -46,17 +46,20 @@
 #include <string.h>     /* memcpy, memset */
 #include <stdio.h>      /* printf (debug) */
 #include "bitstream.h"
+#include "compiler.h"
 #define FSE_STATIC_LINKING_ONLY   /* FSE_optimalTableLog_internal */
 #include "fse.h"        /* header compression */
 #define HUF_STATIC_LINKING_ONLY
 #include "huf.h"
+#include "error_private.h"
 
 
 /* **************************************************************
 *  Error Management
 ****************************************************************/
+#define HUF_isError ERR_isError
 #define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
-#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return f
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
 #define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
 
 
@@ -127,7 +130,7 @@
 };   /* typedef'd to HUF_CElt within "huf.h" */
 
 /*! HUF_writeCTable() :
-    `CTable` : huffman tree to save, using huf representation.
+    `CTable` : Huffman tree to save, using huf representation.
     @return : size of saved CTable */
 size_t HUF_writeCTable (void* dst, size_t maxDstSize,
                         const HUF_CElt* CTable, U32 maxSymbolValue, U32 huffLog)
@@ -165,7 +168,7 @@
 }
 
 
-size_t HUF_readCTable (HUF_CElt* CTable, U32 maxSymbolValue, const void* src, size_t srcSize)
+size_t HUF_readCTable (HUF_CElt* CTable, U32* maxSymbolValuePtr, const void* src, size_t srcSize)
 {
     BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];   /* init not required, even though some static analyzer may complain */
     U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
@@ -177,7 +180,7 @@
 
     /* check result */
     if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
-    if (nbSymbols > maxSymbolValue+1) return ERROR(maxSymbolValue_tooSmall);
+    if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
 
     /* Prepare base value per rank */
     {   U32 n, nextRankStart = 0;
@@ -206,9 +209,10 @@
                 min >>= 1;
         }   }
         /* assign value within rank, symbol order */
-        { U32 n; for (n=0; n<=maxSymbolValue; n++) CTable[n].val = valPerRank[CTable[n].nbBits]++; }
+        { U32 n; for (n=0; n<nbSymbols; n++) CTable[n].val = valPerRank[CTable[n].nbBits]++; }
     }
 
+    *maxSymbolValuePtr = nbSymbols - 1;
     return readSize;
 }
 
@@ -266,7 +270,8 @@
                         if (highTotal <= lowTotal) break;
                 }   }
                 /* only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) */
-                while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))  /* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
+                /* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
+                while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))
                     nBitsToDecrease ++;
                 totalCost -= 1 << (nBitsToDecrease-1);
                 if (rankLast[nBitsToDecrease-1] == noSymbol)
@@ -318,7 +323,10 @@
         U32 const c = count[n];
         U32 const r = BIT_highbit32(c+1) + 1;
         U32 pos = rank[r].current++;
-        while ((pos > rank[r].base) && (c > huffNode[pos-1].count)) huffNode[pos]=huffNode[pos-1], pos--;
+        while ((pos > rank[r].base) && (c > huffNode[pos-1].count)) {
+            huffNode[pos] = huffNode[pos-1];
+            pos--;
+        }
         huffNode[pos].count = c;
         huffNode[pos].byte  = (BYTE)n;
     }
@@ -327,10 +335,10 @@
 
 /** HUF_buildCTable_wksp() :
  *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
- *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as a table of 1024 unsigned.
+ *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as a table of HUF_CTABLE_WORKSPACE_SIZE_U32 unsigned.
  */
 #define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
-typedef nodeElt huffNodeTable[2*HUF_SYMBOLVALUE_MAX+1 +1];
+typedef nodeElt huffNodeTable[HUF_CTABLE_WORKSPACE_SIZE_U32];
 size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
 {
     nodeElt* const huffNode0 = (nodeElt*)workSpace;
@@ -341,9 +349,10 @@
     U32 nodeRoot;
 
     /* safety checks */
-    if (wkspSize < sizeof(huffNodeTable)) return ERROR(GENERIC);   /* workSpace is not large enough */
+    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (wkspSize < sizeof(huffNodeTable)) return ERROR(workSpace_tooSmall);
     if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
-    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(GENERIC);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
     memset(huffNode0, 0, sizeof(huffNodeTable));
 
     /* sort, decreasing order */
@@ -401,6 +410,7 @@
 }
 
 /** HUF_buildCTable() :
+ * @return : maxNbBits
  *  Note : count is used before tree is written, so they can safely overlap
  */
 size_t HUF_buildCTable (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits)
@@ -409,14 +419,34 @@
     return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, nodeTable, sizeof(nodeTable));
 }
 
-static void HUF_encodeSymbol(BIT_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable)
+static size_t HUF_estimateCompressedSize(HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
+{
+    size_t nbBits = 0;
+    int s;
+    for (s = 0; s <= (int)maxSymbolValue; ++s) {
+        nbBits += CTable[s].nbBits * count[s];
+    }
+    return nbBits >> 3;
+}
+
+static int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
+  int bad = 0;
+  int s;
+  for (s = 0; s <= (int)maxSymbolValue; ++s) {
+    bad |= (count[s] != 0) & (CTable[s].nbBits == 0);
+  }
+  return !bad;
+}
+
+size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
+
+FORCE_INLINE_TEMPLATE void
+HUF_encodeSymbol(BIT_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable)
 {
     BIT_addBitsFast(bitCPtr, CTable[symbol].val, CTable[symbol].nbBits);
 }
 
-size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
-
-#define HUF_FLUSHBITS(s)  (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
+#define HUF_FLUSHBITS(s)  BIT_flushBits(s)
 
 #define HUF_FLUSHBITS_1(stream) \
     if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*2+7) HUF_FLUSHBITS(stream)
@@ -424,14 +454,16 @@
 #define HUF_FLUSHBITS_2(stream) \
     if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*4+7) HUF_FLUSHBITS(stream)
 
-size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+FORCE_INLINE_TEMPLATE size_t
+HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
 {
     const BYTE* ip = (const BYTE*) src;
     BYTE* const ostart = (BYTE*)dst;
     BYTE* const oend = ostart + dstSize;
     BYTE* op = ostart;
     size_t n;
-    const unsigned fast = (dstSize >= HUF_BLOCKBOUND(srcSize));
     BIT_CStream_t bitC;
 
     /* init */
@@ -444,12 +476,15 @@
     {
         case 3 : HUF_encodeSymbol(&bitC, ip[n+ 2], CTable);
                  HUF_FLUSHBITS_2(&bitC);
+		 /* fall-through */
         case 2 : HUF_encodeSymbol(&bitC, ip[n+ 1], CTable);
                  HUF_FLUSHBITS_1(&bitC);
+		 /* fall-through */
         case 1 : HUF_encodeSymbol(&bitC, ip[n+ 0], CTable);
                  HUF_FLUSHBITS(&bitC);
-        case 0 :
-        default: ;
+		 /* fall-through */
+        case 0 : /* fall-through */
+        default: break;
     }
 
     for (; n>0; n-=4) {  /* note : n&3==0 at this stage */
@@ -466,8 +501,58 @@
     return BIT_closeCStream(&bitC);
 }
 
+#if DYNAMIC_BMI2
 
-size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+static TARGET_ATTRIBUTE("bmi2") size_t
+HUF_compress1X_usingCTable_internal_bmi2(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal_default(void* dst, size_t dstSize,
+                                      const void* src, size_t srcSize,
+                                      const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int bmi2)
+{
+    if (bmi2) {
+        return HUF_compress1X_usingCTable_internal_bmi2(dst, dstSize, src, srcSize, CTable);
+    }
+    return HUF_compress1X_usingCTable_internal_default(dst, dstSize, src, srcSize, CTable);
+}
+
+#else
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int bmi2)
+{
+    (void)bmi2;
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+#endif
+
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
+}
+
+
+static size_t
+HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, int bmi2)
 {
     size_t const segmentSize = (srcSize+3)/4;   /* first 3 segments */
     const BYTE* ip = (const BYTE*) src;
@@ -480,28 +565,31 @@
     if (srcSize < 12) return 0;   /* no saving possible : too small input */
     op += 6;   /* jumpTable */
 
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable) );
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, oend-op, ip, segmentSize, CTable, bmi2) );
         if (cSize==0) return 0;
+        assert(cSize <= 65535);
         MEM_writeLE16(ostart, (U16)cSize);
         op += cSize;
     }
 
     ip += segmentSize;
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable) );
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, oend-op, ip, segmentSize, CTable, bmi2) );
         if (cSize==0) return 0;
+        assert(cSize <= 65535);
         MEM_writeLE16(ostart+2, (U16)cSize);
         op += cSize;
     }
 
     ip += segmentSize;
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable) );
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, oend-op, ip, segmentSize, CTable, bmi2) );
         if (cSize==0) return 0;
+        assert(cSize <= 65535);
         MEM_writeLE16(ostart+4, (U16)cSize);
         op += cSize;
     }
 
     ip += segmentSize;
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, iend-ip, CTable) );
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, oend-op, ip, iend-ip, CTable, bmi2) );
         if (cSize==0) return 0;
         op += cSize;
     }
@@ -509,65 +597,120 @@
     return op-ostart;
 }
 
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+{
+    return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
+}
 
-/* `workSpace` must a table of at least 1024 unsigned */
+
+static size_t HUF_compressCTable_internal(
+                BYTE* const ostart, BYTE* op, BYTE* const oend,
+                const void* src, size_t srcSize,
+                unsigned singleStream, const HUF_CElt* CTable, const int bmi2)
+{
+    size_t const cSize = singleStream ?
+                         HUF_compress1X_usingCTable_internal(op, oend - op, src, srcSize, CTable, bmi2) :
+                         HUF_compress4X_usingCTable_internal(op, oend - op, src, srcSize, CTable, bmi2);
+    if (HUF_isError(cSize)) { return cSize; }
+    if (cSize==0) { return 0; }   /* uncompressible */
+    op += cSize;
+    /* check compressibility */
+    if ((size_t)(op-ostart) >= srcSize-1) { return 0; }
+    return op-ostart;
+}
+
+typedef struct {
+    U32 count[HUF_SYMBOLVALUE_MAX + 1];
+    HUF_CElt CTable[HUF_SYMBOLVALUE_MAX + 1];
+    huffNodeTable nodeTable;
+} HUF_compress_tables_t;
+
+/* HUF_compress_internal() :
+ * `workSpace` must a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
 static size_t HUF_compress_internal (
                 void* dst, size_t dstSize,
                 const void* src, size_t srcSize,
                 unsigned maxSymbolValue, unsigned huffLog,
                 unsigned singleStream,
-                void* workSpace, size_t wkspSize)
+                void* workSpace, size_t wkspSize,
+                HUF_CElt* oldHufTable, HUF_repeat* repeat, int preferRepeat,
+                const int bmi2)
 {
+    HUF_compress_tables_t* const table = (HUF_compress_tables_t*)workSpace;
     BYTE* const ostart = (BYTE*)dst;
     BYTE* const oend = ostart + dstSize;
     BYTE* op = ostart;
 
-    union {
-        U32 count[HUF_SYMBOLVALUE_MAX+1];
-        HUF_CElt CTable[HUF_SYMBOLVALUE_MAX+1];
-    } table;   /* `count` can overlap with `CTable`; saves 1 KB */
-
     /* checks & inits */
-    if (wkspSize < sizeof(huffNodeTable)) return ERROR(GENERIC);
-    if (!srcSize) return 0;  /* Uncompressed (note : 1 means rle, so first byte must be correct) */
-    if (!dstSize) return 0;  /* cannot fit within dst budget */
+    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (wkspSize < sizeof(*table)) return ERROR(workSpace_tooSmall);
+    if (!srcSize) return 0;  /* Uncompressed */
+    if (!dstSize) return 0;  /* cannot fit anything within dst budget */
     if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);   /* current block size limit */
     if (huffLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
     if (!maxSymbolValue) maxSymbolValue = HUF_SYMBOLVALUE_MAX;
     if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
 
+    /* Heuristic : If old table is valid, use it for small inputs */
+    if (preferRepeat && repeat && *repeat == HUF_repeat_valid) {
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           singleStream, oldHufTable, bmi2);
+    }
+
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(largest, FSE_count_wksp (table.count, &maxSymbolValue, (const BYTE*)src, srcSize, (U32*)workSpace) );
+    {   CHECK_V_F(largest, FSE_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->count) );
         if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
-        if (largest <= (srcSize >> 7)+1) return 0;   /* Fast heuristic : not compressible enough */
+        if (largest <= (srcSize >> 7)+1) return 0;   /* heuristic : probably not compressible enough */
+    }
+
+    /* Check validity of previous table */
+    if ( repeat
+      && *repeat == HUF_repeat_check
+      && !HUF_validateCTable(oldHufTable, table->count, maxSymbolValue)) {
+        *repeat = HUF_repeat_none;
+    }
+    /* Heuristic : use existing table for small inputs */
+    if (preferRepeat && repeat && *repeat != HUF_repeat_none) {
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           singleStream, oldHufTable, bmi2);
     }
 
     /* Build Huffman Tree */
     huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
-    {   CHECK_V_F(maxBits, HUF_buildCTable_wksp (table.CTable, table.count, maxSymbolValue, huffLog, workSpace, wkspSize) );
+    {   CHECK_V_F(maxBits, HUF_buildCTable_wksp(table->CTable, table->count,
+                                                maxSymbolValue, huffLog,
+                                                table->nodeTable, sizeof(table->nodeTable)) );
         huffLog = (U32)maxBits;
+        /* Zero unused symbols in CTable, so we can check it for validity */
+        memset(table->CTable + (maxSymbolValue + 1), 0,
+               sizeof(table->CTable) - ((maxSymbolValue + 1) * sizeof(HUF_CElt)));
     }
 
     /* Write table description header */
-    {   CHECK_V_F(hSize, HUF_writeCTable (op, dstSize, table.CTable, maxSymbolValue, huffLog) );
-        if (hSize + 12 >= srcSize) return 0;   /* not useful to try compression */
-        op += hSize;
-    }
+    {   CHECK_V_F(hSize, HUF_writeCTable (op, dstSize, table->CTable, maxSymbolValue, huffLog) );
+        /* Check if using previous huffman table is beneficial */
+        if (repeat && *repeat != HUF_repeat_none) {
+            size_t const oldSize = HUF_estimateCompressedSize(oldHufTable, table->count, maxSymbolValue);
+            size_t const newSize = HUF_estimateCompressedSize(table->CTable, table->count, maxSymbolValue);
+            if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
+                return HUF_compressCTable_internal(ostart, op, oend,
+                                                   src, srcSize,
+                                                   singleStream, oldHufTable, bmi2);
+        }   }
 
-    /* Compress */
-    {   size_t const cSize = (singleStream) ?
-                            HUF_compress1X_usingCTable(op, oend - op, src, srcSize, table.CTable) :   /* single segment */
-                            HUF_compress4X_usingCTable(op, oend - op, src, srcSize, table.CTable);
-        if (HUF_isError(cSize)) return cSize;
-        if (cSize==0) return 0;   /* uncompressible */
-        op += cSize;
+        /* Use the new huffman table */
+        if (hSize + 12ul >= srcSize) { return 0; }
+        op += hSize;
+        if (repeat) { *repeat = HUF_repeat_none; }
+        if (oldHufTable)
+            memcpy(oldHufTable, table->CTable, sizeof(table->CTable));  /* Save new table */
     }
-
-    /* check compressibility */
-    if ((size_t)(op-ostart) >= srcSize-1)
-        return 0;
-
-    return op-ostart;
+    return HUF_compressCTable_internal(ostart, op, oend,
+                                       src, srcSize,
+                                       singleStream, table->CTable, bmi2);
 }
 
 
@@ -576,34 +719,70 @@
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize)
 {
-    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 1 /* single stream */, workSpace, wkspSize);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, 1 /*single stream*/,
+                                 workSpace, wkspSize,
+                                 NULL, NULL, 0, 0 /*bmi2*/);
+}
+
+size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, 1 /*single stream*/,
+                                 workSpace, wkspSize, hufTable,
+                                 repeat, preferRepeat, bmi2);
 }
 
 size_t HUF_compress1X (void* dst, size_t dstSize,
                  const void* src, size_t srcSize,
                  unsigned maxSymbolValue, unsigned huffLog)
 {
-    unsigned workSpace[1024];
+    unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
     return HUF_compress1X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
 }
 
+/* HUF_compress4X_repeat():
+ * compress input using 4 streams.
+ * provide workspace to generate compression tables */
 size_t HUF_compress4X_wksp (void* dst, size_t dstSize,
                       const void* src, size_t srcSize,
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize)
 {
-    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 0 /* 4 streams */, workSpace, wkspSize);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, 0 /*4 streams*/,
+                                 workSpace, wkspSize,
+                                 NULL, NULL, 0, 0 /*bmi2*/);
+}
+
+/* HUF_compress4X_repeat():
+ * compress input using 4 streams.
+ * re-use an existing huffman compression table */
+size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, 0 /* 4 streams */,
+                                 workSpace, wkspSize,
+                                 hufTable, repeat, preferRepeat, bmi2);
 }
 
 size_t HUF_compress2 (void* dst, size_t dstSize,
                 const void* src, size_t srcSize,
                 unsigned maxSymbolValue, unsigned huffLog)
 {
-    unsigned workSpace[1024];
+    unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
     return HUF_compress4X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
 }
 
 size_t HUF_compress (void* dst, size_t maxDstSize, const void* src, size_t srcSize)
 {
-    return HUF_compress2(dst, maxDstSize, src, (U32)srcSize, 255, HUF_TABLELOG_DEFAULT);
+    return HUF_compress2(dst, maxDstSize, src, srcSize, 255, HUF_TABLELOG_DEFAULT);
 }
--- a/contrib/python-zstandard/zstd/compress/zstd_compress.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/compress/zstd_compress.c	Mon Apr 09 10:13:29 2018 -0700
@@ -1,162 +1,619 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 
 /*-*************************************
+*  Tuning parameters
+***************************************/
+#ifndef ZSTD_CLEVEL_DEFAULT
+#  define ZSTD_CLEVEL_DEFAULT 3
+#endif
+
+
+/*-*************************************
 *  Dependencies
 ***************************************/
 #include <string.h>         /* memset */
+#include "cpu.h"
 #include "mem.h"
-#define XXH_STATIC_LINKING_ONLY   /* XXH64_state_t */
-#include "xxhash.h"               /* XXH_reset, update, digest */
 #define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
 #include "fse.h"
 #define HUF_STATIC_LINKING_ONLY
 #include "huf.h"
-#include "zstd_internal.h"  /* includes zstd.h */
-
-
-/*-*************************************
-*  Constants
-***************************************/
-static const U32 g_searchStrength = 8;   /* control skip over incompressible data */
-#define HASH_READ_SIZE 8
-typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
+#include "zstd_compress_internal.h"
+#include "zstd_fast.h"
+#include "zstd_double_fast.h"
+#include "zstd_lazy.h"
+#include "zstd_opt.h"
+#include "zstd_ldm.h"
 
 
 /*-*************************************
 *  Helper functions
 ***************************************/
-#define ZSTD_STATIC_ASSERT(c) { enum { ZSTD_static_assert = 1/(int)(!!(c)) }; }
-size_t ZSTD_compressBound(size_t srcSize) { return FSE_compressBound(srcSize) + 12; }
-
-
-/*-*************************************
-*  Sequence storage
-***************************************/
-static void ZSTD_resetSeqStore(seqStore_t* ssPtr)
-{
-    ssPtr->lit = ssPtr->litStart;
-    ssPtr->sequences = ssPtr->sequencesStart;
-    ssPtr->longLengthID = 0;
+size_t ZSTD_compressBound(size_t srcSize) {
+    return ZSTD_COMPRESSBOUND(srcSize);
 }
 
 
 /*-*************************************
 *  Context memory management
 ***************************************/
-struct ZSTD_CCtx_s {
-    const BYTE* nextSrc;    /* next block here to continue on current prefix */
-    const BYTE* base;       /* All regular indexes relative to this position */
-    const BYTE* dictBase;   /* extDict indexes relative to this position */
-    U32   dictLimit;        /* below that point, need extDict */
-    U32   lowLimit;         /* below that point, no more data */
-    U32   nextToUpdate;     /* index from which to continue dictionary update */
-    U32   nextToUpdate3;    /* index from which to continue dictionary update */
-    U32   hashLog3;         /* dispatch table : larger == faster, more memory */
-    U32   loadedDictEnd;    /* index of end of dictionary */
-    U32   forceWindow;      /* force back-references to respect limit of 1<<wLog, even for dictionary */
-    ZSTD_compressionStage_e stage;
-    U32   rep[ZSTD_REP_NUM];
-    U32   repToConfirm[ZSTD_REP_NUM];
-    U32   dictID;
-    ZSTD_parameters params;
-    void* workSpace;
-    size_t workSpaceSize;
-    size_t blockSize;
-    U64 frameContentSize;
-    XXH64_state_t xxhState;
+struct ZSTD_CDict_s {
+    void* dictBuffer;
+    const void* dictContent;
+    size_t dictContentSize;
+    void* workspace;
+    size_t workspaceSize;
+    ZSTD_matchState_t matchState;
+    ZSTD_compressedBlockState_t cBlockState;
+    ZSTD_compressionParameters cParams;
     ZSTD_customMem customMem;
-
-    seqStore_t seqStore;    /* sequences storage ptrs */
-    U32* hashTable;
-    U32* hashTable3;
-    U32* chainTable;
-    HUF_CElt* hufTable;
-    U32 flagStaticTables;
-    FSE_CTable offcodeCTable  [FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
-    FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
-    FSE_CTable litlengthCTable  [FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
-    unsigned tmpCounters[1024];
-};
+    U32 dictID;
+};  /* typedef'd to ZSTD_CDict within "zstd.h" */
 
 ZSTD_CCtx* ZSTD_createCCtx(void)
 {
-    return ZSTD_createCCtx_advanced(defaultCustomMem);
+    return ZSTD_createCCtx_advanced(ZSTD_defaultCMem);
 }
 
 ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
 {
-    ZSTD_CCtx* cctx;
-
-    if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
-    if (!customMem.customAlloc || !customMem.customFree) return NULL;
-
-    cctx = (ZSTD_CCtx*) ZSTD_malloc(sizeof(ZSTD_CCtx), customMem);
-    if (!cctx) return NULL;
-    memset(cctx, 0, sizeof(ZSTD_CCtx));
-    cctx->customMem = customMem;
+    ZSTD_STATIC_ASSERT(zcss_init==0);
+    ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN==(0ULL - 1));
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    {   ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_calloc(sizeof(ZSTD_CCtx), customMem);
+        if (!cctx) return NULL;
+        cctx->customMem = customMem;
+        cctx->requestedParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+        cctx->requestedParams.fParams.contentSizeFlag = 1;
+        cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+        return cctx;
+    }
+}
+
+ZSTD_CCtx* ZSTD_initStaticCCtx(void *workspace, size_t workspaceSize)
+{
+    ZSTD_CCtx* const cctx = (ZSTD_CCtx*) workspace;
+    if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL;  /* minimum size */
+    if ((size_t)workspace & 7) return NULL;  /* must be 8-aligned */
+    memset(workspace, 0, workspaceSize);   /* may be a bit generous, could memset be smaller ? */
+    cctx->staticSize = workspaceSize;
+    cctx->workSpace = (void*)(cctx+1);
+    cctx->workSpaceSize = workspaceSize - sizeof(ZSTD_CCtx);
+
+    /* statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
+    if (cctx->workSpaceSize < HUF_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t)) return NULL;
+    assert(((size_t)cctx->workSpace & (sizeof(void*)-1)) == 0);   /* ensure correct alignment */
+    cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)cctx->workSpace;
+    cctx->blockState.nextCBlock = cctx->blockState.prevCBlock + 1;
+    {
+        void* const ptr = cctx->blockState.nextCBlock + 1;
+        cctx->entropyWorkspace = (U32*)ptr;
+    }
+    cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
     return cctx;
 }
 
 size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
 {
     if (cctx==NULL) return 0;   /* support free on NULL */
-    ZSTD_free(cctx->workSpace, cctx->customMem);
+    if (cctx->staticSize) return ERROR(memory_allocation);   /* not compatible with static CCtx */
+    ZSTD_free(cctx->workSpace, cctx->customMem); cctx->workSpace = NULL;
+    ZSTD_freeCDict(cctx->cdictLocal); cctx->cdictLocal = NULL;
+#ifdef ZSTD_MULTITHREAD
+    ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL;
+#endif
     ZSTD_free(cctx, cctx->customMem);
     return 0;   /* reserved as a potential error code in the future */
 }
 
+
+static size_t ZSTD_sizeof_mtctx(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    return ZSTDMT_sizeof_CCtx(cctx->mtctx);
+#else
+    (void) cctx;
+    return 0;
+#endif
+}
+
+
 size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx)
 {
     if (cctx==NULL) return 0;   /* support sizeof on NULL */
-    return sizeof(*cctx) + cctx->workSpaceSize;
+    return sizeof(*cctx) + cctx->workSpaceSize
+           + ZSTD_sizeof_CDict(cctx->cdictLocal)
+           + ZSTD_sizeof_mtctx(cctx);
+}
+
+size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
+{
+    return ZSTD_sizeof_CCtx(zcs);  /* same object */
+}
+
+/* private API call, for dictBuilder only */
+const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx) { return &(ctx->seqStore); }
+
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams(CCtxParams->compressionLevel, srcSizeHint, dictSize);
+    if (CCtxParams->ldmParams.enableLdm) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
+    if (CCtxParams->cParams.windowLog) cParams.windowLog = CCtxParams->cParams.windowLog;
+    if (CCtxParams->cParams.hashLog) cParams.hashLog = CCtxParams->cParams.hashLog;
+    if (CCtxParams->cParams.chainLog) cParams.chainLog = CCtxParams->cParams.chainLog;
+    if (CCtxParams->cParams.searchLog) cParams.searchLog = CCtxParams->cParams.searchLog;
+    if (CCtxParams->cParams.searchLength) cParams.searchLength = CCtxParams->cParams.searchLength;
+    if (CCtxParams->cParams.targetLength) cParams.targetLength = CCtxParams->cParams.targetLength;
+    if (CCtxParams->cParams.strategy) cParams.strategy = CCtxParams->cParams.strategy;
+    return cParams;
+}
+
+static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
+        ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params cctxParams;
+    memset(&cctxParams, 0, sizeof(cctxParams));
+    cctxParams.cParams = cParams;
+    cctxParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;  /* should not matter, as all cParams are presumed properly defined */
+    assert(!ZSTD_checkCParams(cParams));
+    cctxParams.fParams.contentSizeFlag = 1;
+    return cctxParams;
+}
+
+static ZSTD_CCtx_params* ZSTD_createCCtxParams_advanced(
+        ZSTD_customMem customMem)
+{
+    ZSTD_CCtx_params* params;
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    params = (ZSTD_CCtx_params*)ZSTD_calloc(
+            sizeof(ZSTD_CCtx_params), customMem);
+    if (!params) { return NULL; }
+    params->customMem = customMem;
+    params->compressionLevel = ZSTD_CLEVEL_DEFAULT;
+    params->fParams.contentSizeFlag = 1;
+    return params;
 }
 
-size_t ZSTD_setCCtxParameter(ZSTD_CCtx* cctx, ZSTD_CCtxParameter param, unsigned value)
+ZSTD_CCtx_params* ZSTD_createCCtxParams(void)
+{
+    return ZSTD_createCCtxParams_advanced(ZSTD_defaultCMem);
+}
+
+size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params)
+{
+    if (params == NULL) { return 0; }
+    ZSTD_free(params, params->customMem);
+    return 0;
+}
+
+size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params)
+{
+    return ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
+}
+
+size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
+    if (!cctxParams) { return ERROR(GENERIC); }
+    memset(cctxParams, 0, sizeof(*cctxParams));
+    cctxParams->compressionLevel = compressionLevel;
+    cctxParams->fParams.contentSizeFlag = 1;
+    return 0;
+}
+
+size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
+{
+    if (!cctxParams) { return ERROR(GENERIC); }
+    CHECK_F( ZSTD_checkCParams(params.cParams) );
+    memset(cctxParams, 0, sizeof(*cctxParams));
+    cctxParams->cParams = params.cParams;
+    cctxParams->fParams = params.fParams;
+    cctxParams->compressionLevel = ZSTD_CLEVEL_DEFAULT;   /* should not matter, as all cParams are presumed properly defined */
+    assert(!ZSTD_checkCParams(params.cParams));
+    return 0;
+}
+
+/* ZSTD_assignParamsToCCtxParams() :
+ * params is presumed valid at this stage */
+static ZSTD_CCtx_params ZSTD_assignParamsToCCtxParams(
+        ZSTD_CCtx_params cctxParams, ZSTD_parameters params)
+{
+    ZSTD_CCtx_params ret = cctxParams;
+    ret.cParams = params.cParams;
+    ret.fParams = params.fParams;
+    ret.compressionLevel = ZSTD_CLEVEL_DEFAULT;   /* should not matter, as all cParams are presumed properly defined */
+    assert(!ZSTD_checkCParams(params.cParams));
+    return ret;
+}
+
+#define CLAMPCHECK(val,min,max) {            \
+    if (((val)<(min)) | ((val)>(max))) {     \
+        return ERROR(parameter_outOfBound);  \
+}   }
+
+
+static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
 {
     switch(param)
     {
-    case ZSTD_p_forceWindow : cctx->forceWindow = value>0; cctx->loadedDictEnd = 0; return 0;
-    default: return ERROR(parameter_unknown);
+    case ZSTD_p_compressionLevel:
+    case ZSTD_p_hashLog:
+    case ZSTD_p_chainLog:
+    case ZSTD_p_searchLog:
+    case ZSTD_p_minMatch:
+    case ZSTD_p_targetLength:
+    case ZSTD_p_compressionStrategy:
+    case ZSTD_p_compressLiterals:
+        return 1;
+
+    case ZSTD_p_format:
+    case ZSTD_p_windowLog:
+    case ZSTD_p_contentSizeFlag:
+    case ZSTD_p_checksumFlag:
+    case ZSTD_p_dictIDFlag:
+    case ZSTD_p_forceMaxWindow :
+    case ZSTD_p_nbWorkers:
+    case ZSTD_p_jobSize:
+    case ZSTD_p_overlapSizeLog:
+    case ZSTD_p_enableLongDistanceMatching:
+    case ZSTD_p_ldmHashLog:
+    case ZSTD_p_ldmMinMatch:
+    case ZSTD_p_ldmBucketSizeLog:
+    case ZSTD_p_ldmHashEveryLog:
+    default:
+        return 0;
+    }
+}
+
+size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned value)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParameter (%u, %u)", (U32)param, value);
+    if (cctx->streamStage != zcss_init) {
+        if (ZSTD_isUpdateAuthorized(param)) {
+            cctx->cParamsChanged = 1;
+        } else {
+            return ERROR(stage_wrong);
+    }   }
+
+    switch(param)
+    {
+    case ZSTD_p_format :
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_compressionLevel:
+        if (cctx->cdict) return ERROR(stage_wrong);
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_windowLog:
+    case ZSTD_p_hashLog:
+    case ZSTD_p_chainLog:
+    case ZSTD_p_searchLog:
+    case ZSTD_p_minMatch:
+    case ZSTD_p_targetLength:
+    case ZSTD_p_compressionStrategy:
+        if (cctx->cdict) return ERROR(stage_wrong);
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_compressLiterals:
+    case ZSTD_p_contentSizeFlag:
+    case ZSTD_p_checksumFlag:
+    case ZSTD_p_dictIDFlag:
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_forceMaxWindow :  /* Force back-references to remain < windowSize,
+                                   * even when referencing into Dictionary content.
+                                   * default : 0 when using a CDict, 1 when using a Prefix */
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_nbWorkers:
+        if ((value>0) && cctx->staticSize) {
+            return ERROR(parameter_unsupported);  /* MT not compatible with static alloc */
+        }
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_jobSize:
+    case ZSTD_p_overlapSizeLog:
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_enableLongDistanceMatching:
+    case ZSTD_p_ldmHashLog:
+    case ZSTD_p_ldmMinMatch:
+    case ZSTD_p_ldmBucketSizeLog:
+    case ZSTD_p_ldmHashEveryLog:
+        if (cctx->cdict) return ERROR(stage_wrong);
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    default: return ERROR(parameter_unsupported);
     }
 }
 
-const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx)   /* hidden interface */
-{
-    return &(ctx->seqStore);
-}
-
-static ZSTD_parameters ZSTD_getParamsFromCCtx(const ZSTD_CCtx* cctx)
+size_t ZSTD_CCtxParam_setParameter(
+        ZSTD_CCtx_params* CCtxParams, ZSTD_cParameter param, unsigned value)
 {
-    return cctx->params;
+    DEBUGLOG(4, "ZSTD_CCtxParam_setParameter (%u, %u)", (U32)param, value);
+    switch(param)
+    {
+    case ZSTD_p_format :
+        if (value > (unsigned)ZSTD_f_zstd1_magicless)
+            return ERROR(parameter_unsupported);
+        CCtxParams->format = (ZSTD_format_e)value;
+        return (size_t)CCtxParams->format;
+
+    case ZSTD_p_compressionLevel : {
+        int cLevel = (int)value;  /* cast expected to restore negative sign */
+        if (cLevel > ZSTD_maxCLevel()) cLevel = ZSTD_maxCLevel();
+        if (cLevel) {  /* 0 : does not change current level */
+            CCtxParams->disableLiteralCompression = (cLevel<0);  /* negative levels disable huffman */
+            CCtxParams->compressionLevel = cLevel;
+        }
+        if (CCtxParams->compressionLevel >= 0) return CCtxParams->compressionLevel;
+        return 0;  /* return type (size_t) cannot represent negative values */
+    }
+
+    case ZSTD_p_windowLog :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
+        CCtxParams->cParams.windowLog = value;
+        return CCtxParams->cParams.windowLog;
+
+    case ZSTD_p_hashLog :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
+        CCtxParams->cParams.hashLog = value;
+        return CCtxParams->cParams.hashLog;
+
+    case ZSTD_p_chainLog :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
+        CCtxParams->cParams.chainLog = value;
+        return CCtxParams->cParams.chainLog;
+
+    case ZSTD_p_searchLog :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
+        CCtxParams->cParams.searchLog = value;
+        return value;
+
+    case ZSTD_p_minMatch :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
+        CCtxParams->cParams.searchLength = value;
+        return CCtxParams->cParams.searchLength;
+
+    case ZSTD_p_targetLength :
+        /* all values are valid. 0 => use default */
+        CCtxParams->cParams.targetLength = value;
+        return CCtxParams->cParams.targetLength;
+
+    case ZSTD_p_compressionStrategy :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, (unsigned)ZSTD_fast, (unsigned)ZSTD_btultra);
+        CCtxParams->cParams.strategy = (ZSTD_strategy)value;
+        return (size_t)CCtxParams->cParams.strategy;
+
+    case ZSTD_p_compressLiterals:
+        CCtxParams->disableLiteralCompression = !value;
+        return !CCtxParams->disableLiteralCompression;
+
+    case ZSTD_p_contentSizeFlag :
+        /* Content size written in frame header _when known_ (default:1) */
+        DEBUGLOG(4, "set content size flag = %u", (value>0));
+        CCtxParams->fParams.contentSizeFlag = value > 0;
+        return CCtxParams->fParams.contentSizeFlag;
+
+    case ZSTD_p_checksumFlag :
+        /* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
+        CCtxParams->fParams.checksumFlag = value > 0;
+        return CCtxParams->fParams.checksumFlag;
+
+    case ZSTD_p_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
+        DEBUGLOG(4, "set dictIDFlag = %u", (value>0));
+        CCtxParams->fParams.noDictIDFlag = !value;
+        return !CCtxParams->fParams.noDictIDFlag;
+
+    case ZSTD_p_forceMaxWindow :
+        CCtxParams->forceWindow = (value > 0);
+        return CCtxParams->forceWindow;
+
+    case ZSTD_p_nbWorkers :
+#ifndef ZSTD_MULTITHREAD
+        if (value>0) return ERROR(parameter_unsupported);
+        return 0;
+#else
+        return ZSTDMT_CCtxParam_setNbWorkers(CCtxParams, value);
+#endif
+
+    case ZSTD_p_jobSize :
+#ifndef ZSTD_MULTITHREAD
+        return ERROR(parameter_unsupported);
+#else
+        return ZSTDMT_CCtxParam_setMTCtxParameter(CCtxParams, ZSTDMT_p_jobSize, value);
+#endif
+
+    case ZSTD_p_overlapSizeLog :
+#ifndef ZSTD_MULTITHREAD
+        return ERROR(parameter_unsupported);
+#else
+        return ZSTDMT_CCtxParam_setMTCtxParameter(CCtxParams, ZSTDMT_p_overlapSectionLog, value);
+#endif
+
+    case ZSTD_p_enableLongDistanceMatching :
+        CCtxParams->ldmParams.enableLdm = (value>0);
+        return CCtxParams->ldmParams.enableLdm;
+
+    case ZSTD_p_ldmHashLog :
+        if (value>0)   /* 0 ==> auto */
+            CLAMPCHECK(value, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
+        CCtxParams->ldmParams.hashLog = value;
+        return CCtxParams->ldmParams.hashLog;
+
+    case ZSTD_p_ldmMinMatch :
+        if (value>0)   /* 0 ==> default */
+            CLAMPCHECK(value, ZSTD_LDM_MINMATCH_MIN, ZSTD_LDM_MINMATCH_MAX);
+        CCtxParams->ldmParams.minMatchLength = value;
+        return CCtxParams->ldmParams.minMatchLength;
+
+    case ZSTD_p_ldmBucketSizeLog :
+        if (value > ZSTD_LDM_BUCKETSIZELOG_MAX)
+            return ERROR(parameter_outOfBound);
+        CCtxParams->ldmParams.bucketSizeLog = value;
+        return CCtxParams->ldmParams.bucketSizeLog;
+
+    case ZSTD_p_ldmHashEveryLog :
+        if (value > ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN)
+            return ERROR(parameter_outOfBound);
+        CCtxParams->ldmParams.hashEveryLog = value;
+        return CCtxParams->ldmParams.hashEveryLog;
+
+    default: return ERROR(parameter_unsupported);
+    }
 }
 
-
-/** ZSTD_checkParams() :
-    ensure param values remain within authorized range.
+/** ZSTD_CCtx_setParametersUsingCCtxParams() :
+ *  just applies `params` into `cctx`
+ *  no action is performed, parameters are merely stored.
+ *  If ZSTDMT is enabled, parameters are pushed to cctx->mtctx.
+ *    This is possible even if a compression is ongoing.
+ *    In which case, new parameters will be applied on the fly, starting with next compression job.
+ */
+size_t ZSTD_CCtx_setParametersUsingCCtxParams(
+        ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params)
+{
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    if (cctx->cdict) return ERROR(stage_wrong);
+
+    cctx->requestedParams = *params;
+    return 0;
+}
+
+ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %u bytes", (U32)pledgedSrcSize);
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+    return 0;
+}
+
+size_t ZSTD_CCtx_loadDictionary_advanced(
+        ZSTD_CCtx* cctx, const void* dict, size_t dictSize,
+        ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType)
+{
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    if (cctx->staticSize) return ERROR(memory_allocation);  /* no malloc for static CCtx */
+    DEBUGLOG(4, "ZSTD_CCtx_loadDictionary_advanced (size: %u)", (U32)dictSize);
+    ZSTD_freeCDict(cctx->cdictLocal);  /* in case one already exists */
+    if (dict==NULL || dictSize==0) {   /* no dictionary mode */
+        cctx->cdictLocal = NULL;
+        cctx->cdict = NULL;
+    } else {
+        ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(&cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1, dictSize);
+        cctx->cdictLocal = ZSTD_createCDict_advanced(
+                                dict, dictSize,
+                                dictLoadMethod, dictContentType,
+                                cParams, cctx->customMem);
+        cctx->cdict = cctx->cdictLocal;
+        if (cctx->cdictLocal == NULL)
+            return ERROR(memory_allocation);
+    }
+    return 0;
+}
+
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_byReference(
+      ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_CCtx_loadDictionary_advanced(
+            cctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
+}
+
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_CCtx_loadDictionary_advanced(
+            cctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
+}
+
+
+size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    cctx->cdict = cdict;
+    memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));  /* exclusive */
+    return 0;
+}
+
+size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize)
+{
+    return ZSTD_CCtx_refPrefix_advanced(cctx, prefix, prefixSize, ZSTD_dct_rawContent);
+}
+
+size_t ZSTD_CCtx_refPrefix_advanced(
+        ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
+{
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    cctx->cdict = NULL;   /* prefix discards any prior cdict */
+    cctx->prefixDict.dict = prefix;
+    cctx->prefixDict.dictSize = prefixSize;
+    cctx->prefixDict.dictContentType = dictContentType;
+    return 0;
+}
+
+static void ZSTD_startNewCompression(ZSTD_CCtx* cctx)
+{
+    cctx->streamStage = zcss_init;
+    cctx->pledgedSrcSizePlusOne = 0;
+}
+
+/*! ZSTD_CCtx_reset() :
+ *  Also dumps dictionary */
+void ZSTD_CCtx_reset(ZSTD_CCtx* cctx)
+{
+    ZSTD_startNewCompression(cctx);
+    cctx->cdict = NULL;
+}
+
+/** ZSTD_checkCParams() :
+    control CParam values remain within authorized range.
     @return : 0, or an error code if one value is beyond authorized range */
 size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
 {
-#   define CLAMPCHECK(val,min,max) { if ((val<min) | (val>max)) return ERROR(compressionParameter_unsupported); }
     CLAMPCHECK(cParams.windowLog, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
     CLAMPCHECK(cParams.chainLog, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
     CLAMPCHECK(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
     CLAMPCHECK(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
-    { U32 const searchLengthMin = ((cParams.strategy == ZSTD_fast) | (cParams.strategy == ZSTD_greedy)) ? ZSTD_SEARCHLENGTH_MIN+1 : ZSTD_SEARCHLENGTH_MIN;
-      U32 const searchLengthMax = (cParams.strategy == ZSTD_fast) ? ZSTD_SEARCHLENGTH_MAX : ZSTD_SEARCHLENGTH_MAX-1;
-      CLAMPCHECK(cParams.searchLength, searchLengthMin, searchLengthMax); }
-    CLAMPCHECK(cParams.targetLength, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
-    if ((U32)(cParams.strategy) > (U32)ZSTD_btopt2) return ERROR(compressionParameter_unsupported);
+    CLAMPCHECK(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
+    if ((U32)(cParams.targetLength) < ZSTD_TARGETLENGTH_MIN)
+        return ERROR(parameter_unsupported);
+    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra)
+        return ERROR(parameter_unsupported);
     return 0;
 }
 
+/** ZSTD_clampCParams() :
+ *  make CParam values within valid range.
+ *  @return : valid CParams */
+static ZSTD_compressionParameters ZSTD_clampCParams(ZSTD_compressionParameters cParams)
+{
+#   define CLAMP(val,min,max) {      \
+        if (val<min) val=min;        \
+        else if (val>max) val=max;   \
+    }
+    CLAMP(cParams.windowLog, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
+    CLAMP(cParams.chainLog, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
+    CLAMP(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
+    CLAMP(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
+    CLAMP(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
+    if ((U32)(cParams.targetLength) < ZSTD_TARGETLENGTH_MIN) cParams.targetLength = ZSTD_TARGETLENGTH_MIN;
+    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra) cParams.strategy = ZSTD_btultra;
+    return cParams;
+}
 
 /** ZSTD_cycleLog() :
  *  condition for correct operation : hashLog > 1 */
@@ -166,162 +623,466 @@
     return hashLog - btScale;
 }
 
-/** ZSTD_adjustCParams() :
+/** ZSTD_adjustCParams_internal() :
     optimize `cPar` for a given input (`srcSize` and `dictSize`).
-    mostly downsizing to reduce memory consumption and initialization.
-    Both `srcSize` and `dictSize` are optional (use 0 if unknown),
-    but if both are 0, no optimization can be done.
-    Note : cPar is considered validated at this stage. Use ZSTD_checkParams() to ensure that. */
-ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
+    mostly downsizing to reduce memory consumption and initialization latency.
+    Both `srcSize` and `dictSize` are optional (use 0 if unknown).
+    Note : cPar is considered validated at this stage. Use ZSTD_checkCParams() to ensure that condition. */
+ZSTD_compressionParameters ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
 {
-    if (srcSize+dictSize == 0) return cPar;   /* no size information available : no adjustment */
-
-    /* resize params, to use less memory when necessary */
-    {   U32 const minSrcSize = (srcSize==0) ? 500 : 0;
-        U64 const rSize = srcSize + dictSize + minSrcSize;
-        if (rSize < ((U64)1<<ZSTD_WINDOWLOG_MAX)) {
-            U32 const srcLog = MAX(ZSTD_HASHLOG_MIN, ZSTD_highbit32((U32)(rSize)-1) + 1);
-            if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
-    }   }
+    static const U64 minSrcSize = 513; /* (1<<9) + 1 */
+    static const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
+    assert(ZSTD_checkCParams(cPar)==0);
+
+    if (dictSize && (srcSize+1<2) /* srcSize unknown */ )
+        srcSize = minSrcSize;  /* presumed small when there is a dictionary */
+    else if (srcSize == 0)
+        srcSize = ZSTD_CONTENTSIZE_UNKNOWN;  /* 0 == unknown : presumed large */
+
+    /* resize windowLog if input is small enough, to use less memory */
+    if ( (srcSize < maxWindowResize)
+      && (dictSize < maxWindowResize) )  {
+        U32 const tSize = (U32)(srcSize + dictSize);
+        static U32 const hashSizeMin = 1 << ZSTD_HASHLOG_MIN;
+        U32 const srcLog = (tSize < hashSizeMin) ? ZSTD_HASHLOG_MIN :
+                            ZSTD_highbit32(tSize-1) + 1;
+        if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
+    }
     if (cPar.hashLog > cPar.windowLog) cPar.hashLog = cPar.windowLog;
     {   U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
-        if (cycleLog > cPar.windowLog) cPar.chainLog -= (cycleLog - cPar.windowLog);
+        if (cycleLog > cPar.windowLog)
+            cPar.chainLog -= (cycleLog - cPar.windowLog);
     }
 
-    if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN) cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* required for frame header */
+    if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN)
+        cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* required for frame header */
 
     return cPar;
 }
 
-
-size_t ZSTD_estimateCCtxSize(ZSTD_compressionParameters cParams)
+ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
 {
-    size_t const blockSize = MIN(ZSTD_BLOCKSIZE_ABSOLUTEMAX, (size_t)1 << cParams.windowLog);
-    U32    const divider = (cParams.searchLength==3) ? 3 : 4;
-    size_t const maxNbSeq = blockSize / divider;
-    size_t const tokenSpace = blockSize + 11*maxNbSeq;
-
-    size_t const chainSize = (cParams.strategy == ZSTD_fast) ? 0 : (1 << cParams.chainLog);
-    size_t const hSize = ((size_t)1) << cParams.hashLog;
-    U32    const hashLog3 = (cParams.searchLength>3) ? 0 : MIN(ZSTD_HASHLOG3_MAX, cParams.windowLog);
+    cPar = ZSTD_clampCParams(cPar);
+    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize);
+}
+
+static size_t ZSTD_sizeof_matchState(ZSTD_compressionParameters const* cParams, const U32 forCCtx)
+{
+    size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = (forCCtx && cParams->searchLength==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
     size_t const h3Size = ((size_t)1) << hashLog3;
     size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
-
-    size_t const optSpace = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits))*sizeof(U32)
-                          + (ZSTD_OPT_NUM+1)*(sizeof(ZSTD_match_t) + sizeof(ZSTD_optimal_t));
-    size_t const neededSpace = tableSpace + (256*sizeof(U32)) /* huffTable */ + tokenSpace
-                             + (((cParams.strategy == ZSTD_btopt) || (cParams.strategy == ZSTD_btopt2)) ? optSpace : 0);
-
-    return sizeof(ZSTD_CCtx) + neededSpace;
+    size_t const optPotentialSpace = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits)) * sizeof(U32)
+                          + (ZSTD_OPT_NUM+1) * (sizeof(ZSTD_match_t)+sizeof(ZSTD_optimal_t));
+    size_t const optSpace = (forCCtx && ((cParams->strategy == ZSTD_btopt) ||
+                                         (cParams->strategy == ZSTD_btultra)))
+                                ? optPotentialSpace
+                                : 0;
+    DEBUGLOG(4, "chainSize: %u - hSize: %u - h3Size: %u",
+                (U32)chainSize, (U32)hSize, (U32)h3Size);
+    return tableSpace + optSpace;
+}
+
+size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    /* Estimate CCtx size is supported for single-threaded compression only. */
+    if (params->nbWorkers > 0) { return ERROR(GENERIC); }
+    {   ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(params, 0, 0);
+        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
+        U32    const divider = (cParams.searchLength==3) ? 3 : 4;
+        size_t const maxNbSeq = blockSize / divider;
+        size_t const tokenSpace = blockSize + 11*maxNbSeq;
+        size_t const entropySpace = HUF_WORKSPACE_SIZE;
+        size_t const blockStateSpace = 2 * sizeof(ZSTD_compressedBlockState_t);
+        size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 1);
+
+        size_t const ldmSpace = ZSTD_ldm_getTableSize(params->ldmParams);
+        size_t const ldmSeqSpace = ZSTD_ldm_getMaxNbSeq(params->ldmParams, blockSize) * sizeof(rawSeq);
+
+        size_t const neededSpace = entropySpace + blockStateSpace + tokenSpace +
+                                   matchStateSize + ldmSpace + ldmSeqSpace;
+
+        DEBUGLOG(5, "sizeof(ZSTD_CCtx) : %u", (U32)sizeof(ZSTD_CCtx));
+        DEBUGLOG(5, "estimate workSpace : %u", (U32)neededSpace);
+        return sizeof(ZSTD_CCtx) + neededSpace;
+    }
+}
+
+size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params const params = ZSTD_makeCCtxParamsFromCParams(cParams);
+    return ZSTD_estimateCCtxSize_usingCCtxParams(&params);
+}
+
+static size_t ZSTD_estimateCCtxSize_internal(int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, 0);
+    return ZSTD_estimateCCtxSize_usingCParams(cParams);
+}
+
+size_t ZSTD_estimateCCtxSize(int compressionLevel)
+{
+    int level;
+    size_t memBudget = 0;
+    for (level=1; level<=compressionLevel; level++) {
+        size_t const newMB = ZSTD_estimateCCtxSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
+}
+
+size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    if (params->nbWorkers > 0) { return ERROR(GENERIC); }
+    {   size_t const CCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(params);
+        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << params->cParams.windowLog);
+        size_t const inBuffSize = ((size_t)1 << params->cParams.windowLog) + blockSize;
+        size_t const outBuffSize = ZSTD_compressBound(blockSize) + 1;
+        size_t const streamingSize = inBuffSize + outBuffSize;
+
+        return CCtxSize + streamingSize;
+    }
+}
+
+size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params const params = ZSTD_makeCCtxParamsFromCParams(cParams);
+    return ZSTD_estimateCStreamSize_usingCCtxParams(&params);
+}
+
+static size_t ZSTD_estimateCStreamSize_internal(int compressionLevel) {
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, 0);
+    return ZSTD_estimateCStreamSize_usingCParams(cParams);
+}
+
+size_t ZSTD_estimateCStreamSize(int compressionLevel) {
+    int level;
+    size_t memBudget = 0;
+    for (level=1; level<=compressionLevel; level++) {
+        size_t const newMB = ZSTD_estimateCStreamSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
 }
 
-
-static U32 ZSTD_equivalentParams(ZSTD_parameters param1, ZSTD_parameters param2)
+/* ZSTD_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads (non-blocking mode).
+ */
+ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        return ZSTDMT_getFrameProgression(cctx->mtctx);
+    }
+#endif
+    {   ZSTD_frameProgression fp;
+        size_t const buffered = (cctx->inBuff == NULL) ? 0 :
+                                cctx->inBuffPos - cctx->inToCompress;
+        if (buffered) assert(cctx->inBuffPos >= cctx->inToCompress);
+        assert(buffered <= ZSTD_BLOCKSIZE_MAX);
+        fp.ingested = cctx->consumedSrcSize + buffered;
+        fp.consumed = cctx->consumedSrcSize;
+        fp.produced = cctx->producedCSize;
+        return fp;
+}   }
+
+
+static U32 ZSTD_equivalentCParams(ZSTD_compressionParameters cParams1,
+                                  ZSTD_compressionParameters cParams2)
+{
+    return (cParams1.hashLog  == cParams2.hashLog)
+         & (cParams1.chainLog == cParams2.chainLog)
+         & (cParams1.strategy == cParams2.strategy)   /* opt parser space */
+         & ((cParams1.searchLength==3) == (cParams2.searchLength==3));  /* hashlog3 space */
+}
+
+/** The parameters are equivalent if ldm is not enabled in both sets or
+ *  all the parameters are equivalent. */
+static U32 ZSTD_equivalentLdmParams(ldmParams_t ldmParams1,
+                                    ldmParams_t ldmParams2)
 {
-    return (param1.cParams.hashLog  == param2.cParams.hashLog)
-         & (param1.cParams.chainLog == param2.cParams.chainLog)
-         & (param1.cParams.strategy == param2.cParams.strategy)
-         & ((param1.cParams.searchLength==3) == (param2.cParams.searchLength==3));
+    return (!ldmParams1.enableLdm && !ldmParams2.enableLdm) ||
+           (ldmParams1.enableLdm == ldmParams2.enableLdm &&
+            ldmParams1.hashLog == ldmParams2.hashLog &&
+            ldmParams1.bucketSizeLog == ldmParams2.bucketSizeLog &&
+            ldmParams1.minMatchLength == ldmParams2.minMatchLength &&
+            ldmParams1.hashEveryLog == ldmParams2.hashEveryLog);
+}
+
+typedef enum { ZSTDb_not_buffered, ZSTDb_buffered } ZSTD_buffered_policy_e;
+
+/* ZSTD_sufficientBuff() :
+ * check internal buffers exist for streaming if buffPol == ZSTDb_buffered .
+ * Note : they are assumed to be correctly sized if ZSTD_equivalentCParams()==1 */
+static U32 ZSTD_sufficientBuff(size_t bufferSize1, size_t blockSize1,
+                            ZSTD_buffered_policy_e buffPol2,
+                            ZSTD_compressionParameters cParams2,
+                            U64 pledgedSrcSize)
+{
+    size_t const windowSize2 = MAX(1, (size_t)MIN(((U64)1 << cParams2.windowLog), pledgedSrcSize));
+    size_t const blockSize2 = MIN(ZSTD_BLOCKSIZE_MAX, windowSize2);
+    size_t const neededBufferSize2 = (buffPol2==ZSTDb_buffered) ? windowSize2 + blockSize2 : 0;
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is windowSize2=%u <= wlog1=%u",
+                (U32)windowSize2, cParams2.windowLog);
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is blockSize2=%u <= blockSize1=%u",
+                (U32)blockSize2, (U32)blockSize1);
+    return (blockSize2 <= blockSize1) /* seqStore space depends on blockSize */
+         & (neededBufferSize2 <= bufferSize1);
+}
+
+/** Equivalence for resetCCtx purposes */
+static U32 ZSTD_equivalentParams(ZSTD_CCtx_params params1,
+                                 ZSTD_CCtx_params params2,
+                                 size_t buffSize1, size_t blockSize1,
+                                 ZSTD_buffered_policy_e buffPol2,
+                                 U64 pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_equivalentParams: pledgedSrcSize=%u", (U32)pledgedSrcSize);
+    return ZSTD_equivalentCParams(params1.cParams, params2.cParams) &&
+           ZSTD_equivalentLdmParams(params1.ldmParams, params2.ldmParams) &&
+           ZSTD_sufficientBuff(buffSize1, blockSize1, buffPol2, params2.cParams, pledgedSrcSize);
+}
+
+static void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
+{
+    int i;
+    for (i = 0; i < ZSTD_REP_NUM; ++i)
+        bs->rep[i] = repStartValue[i];
+    bs->entropy.hufCTable_repeatMode = HUF_repeat_none;
+    bs->entropy.offcode_repeatMode = FSE_repeat_none;
+    bs->entropy.matchlength_repeatMode = FSE_repeat_none;
+    bs->entropy.litlength_repeatMode = FSE_repeat_none;
+}
+
+/*! ZSTD_invalidateMatchState()
+ * Invalidate all the matches in the match finder tables.
+ * Requires nextSrc and base to be set (can be NULL).
+ */
+static void ZSTD_invalidateMatchState(ZSTD_matchState_t* ms)
+{
+    ZSTD_window_clear(&ms->window);
+
+    ms->nextToUpdate = ms->window.dictLimit + 1;
+    ms->loadedDictEnd = 0;
+    ms->opt.litLengthSum = 0;  /* force reset of btopt stats */
 }
 
 /*! ZSTD_continueCCtx() :
-    reuse CCtx without reset (note : requires no dictionary) */
-static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_parameters params, U64 frameContentSize)
+ *  reuse CCtx without reset (note : requires no dictionary) */
+static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_CCtx_params params, U64 pledgedSrcSize)
 {
-    U32 const end = (U32)(cctx->nextSrc - cctx->base);
-    cctx->params = params;
-    cctx->frameContentSize = frameContentSize;
-    cctx->lowLimit = end;
-    cctx->dictLimit = end;
-    cctx->nextToUpdate = end+1;
+    size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params.cParams.windowLog), pledgedSrcSize));
+    size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
+    DEBUGLOG(4, "ZSTD_continueCCtx: re-use context in place");
+
+    cctx->blockSize = blockSize;   /* previous block size could be different even for same windowLog, due to pledgedSrcSize */
+    cctx->appliedParams = params;
+    cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+    cctx->consumedSrcSize = 0;
+    cctx->producedCSize = 0;
+    if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
+        cctx->appliedParams.fParams.contentSizeFlag = 0;
+    DEBUGLOG(4, "pledged content size : %u ; flag : %u",
+        (U32)pledgedSrcSize, cctx->appliedParams.fParams.contentSizeFlag);
     cctx->stage = ZSTDcs_init;
     cctx->dictID = 0;
-    cctx->loadedDictEnd = 0;
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) cctx->rep[i] = repStartValue[i]; }
-    cctx->seqStore.litLengthSum = 0;  /* force reset of btopt stats */
+    if (params.ldmParams.enableLdm)
+        ZSTD_window_clear(&cctx->ldmState.window);
+    ZSTD_referenceExternalSequences(cctx, NULL, 0);
+    ZSTD_invalidateMatchState(&cctx->blockState.matchState);
+    ZSTD_reset_compressedBlockState(cctx->blockState.prevCBlock);
     XXH64_reset(&cctx->xxhState, 0);
     return 0;
 }
 
-typedef enum { ZSTDcrp_continue, ZSTDcrp_noMemset, ZSTDcrp_fullReset } ZSTD_compResetPolicy_e;
-
-/*! ZSTD_resetCCtx_advanced() :
-    note : 'params' must be validated */
-static size_t ZSTD_resetCCtx_advanced (ZSTD_CCtx* zc,
-                                       ZSTD_parameters params, U64 frameContentSize,
-                                       ZSTD_compResetPolicy_e const crp)
+typedef enum { ZSTDcrp_continue, ZSTDcrp_noMemset } ZSTD_compResetPolicy_e;
+
+static void* ZSTD_reset_matchState(ZSTD_matchState_t* ms, void* ptr, ZSTD_compressionParameters const* cParams, ZSTD_compResetPolicy_e const crp, U32 const forCCtx)
 {
-    if (crp == ZSTDcrp_continue)
-        if (ZSTD_equivalentParams(params, zc->params))
-            return ZSTD_continueCCtx(zc, params, frameContentSize);
-
-    {   size_t const blockSize = MIN(ZSTD_BLOCKSIZE_ABSOLUTEMAX, (size_t)1 << params.cParams.windowLog);
+    size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = (forCCtx && cParams->searchLength==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    size_t const h3Size = ((size_t)1) << hashLog3;
+    size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+
+    assert(((size_t)ptr & 3) == 0);
+
+    ms->hashLog3 = hashLog3;
+    memset(&ms->window, 0, sizeof(ms->window));
+    ZSTD_invalidateMatchState(ms);
+
+    /* opt parser space */
+    if (forCCtx && ((cParams->strategy == ZSTD_btopt) | (cParams->strategy == ZSTD_btultra))) {
+        DEBUGLOG(4, "reserving optimal parser space");
+        ms->opt.litFreq = (U32*)ptr;
+        ms->opt.litLengthFreq = ms->opt.litFreq + (1<<Litbits);
+        ms->opt.matchLengthFreq = ms->opt.litLengthFreq + (MaxLL+1);
+        ms->opt.offCodeFreq = ms->opt.matchLengthFreq + (MaxML+1);
+        ptr = ms->opt.offCodeFreq + (MaxOff+1);
+        ms->opt.matchTable = (ZSTD_match_t*)ptr;
+        ptr = ms->opt.matchTable + ZSTD_OPT_NUM+1;
+        ms->opt.priceTable = (ZSTD_optimal_t*)ptr;
+        ptr = ms->opt.priceTable + ZSTD_OPT_NUM+1;
+    }
+
+    /* table Space */
+    DEBUGLOG(4, "reset table : %u", crp!=ZSTDcrp_noMemset);
+    assert(((size_t)ptr & 3) == 0);  /* ensure ptr is properly aligned */
+    if (crp!=ZSTDcrp_noMemset) memset(ptr, 0, tableSpace);   /* reset tables only */
+    ms->hashTable = (U32*)(ptr);
+    ms->chainTable = ms->hashTable + hSize;
+    ms->hashTable3 = ms->chainTable + chainSize;
+    ptr = ms->hashTable3 + h3Size;
+
+    assert(((size_t)ptr & 3) == 0);
+    return ptr;
+}
+
+/*! ZSTD_resetCCtx_internal() :
+    note : `params` are assumed fully validated at this stage */
+static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
+                                      ZSTD_CCtx_params params, U64 pledgedSrcSize,
+                                      ZSTD_compResetPolicy_e const crp,
+                                      ZSTD_buffered_policy_e const zbuff)
+{
+    DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u",
+                (U32)pledgedSrcSize, params.cParams.windowLog);
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+
+    if (crp == ZSTDcrp_continue) {
+        if (ZSTD_equivalentParams(zc->appliedParams, params,
+                                zc->inBuffSize, zc->blockSize,
+                                zbuff, pledgedSrcSize)) {
+            DEBUGLOG(4, "ZSTD_equivalentParams()==1 -> continue mode (wLog1=%u, blockSize1=%u)",
+                        zc->appliedParams.cParams.windowLog, (U32)zc->blockSize);
+            return ZSTD_continueCCtx(zc, params, pledgedSrcSize);
+    }   }
+    DEBUGLOG(4, "ZSTD_equivalentParams()==0 -> reset CCtx");
+
+    if (params.ldmParams.enableLdm) {
+        /* Adjust long distance matching parameters */
+        params.ldmParams.windowLog = params.cParams.windowLog;
+        ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
+        assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
+        assert(params.ldmParams.hashEveryLog < 32);
+        zc->ldmState.hashPower =
+                ZSTD_ldm_getHashPower(params.ldmParams.minMatchLength);
+    }
+
+    {   size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params.cParams.windowLog), pledgedSrcSize));
+        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
         U32    const divider = (params.cParams.searchLength==3) ? 3 : 4;
         size_t const maxNbSeq = blockSize / divider;
         size_t const tokenSpace = blockSize + 11*maxNbSeq;
-        size_t const chainSize = (params.cParams.strategy == ZSTD_fast) ? 0 : (1 << params.cParams.chainLog);
-        size_t const hSize = ((size_t)1) << params.cParams.hashLog;
-        U32    const hashLog3 = (params.cParams.searchLength>3) ? 0 : MIN(ZSTD_HASHLOG3_MAX, params.cParams.windowLog);
-        size_t const h3Size = ((size_t)1) << hashLog3;
-        size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+        size_t const buffOutSize = (zbuff==ZSTDb_buffered) ? ZSTD_compressBound(blockSize)+1 : 0;
+        size_t const buffInSize = (zbuff==ZSTDb_buffered) ? windowSize + blockSize : 0;
+        size_t const matchStateSize = ZSTD_sizeof_matchState(&params.cParams, /* forCCtx */ 1);
+        size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(params.ldmParams, blockSize);
         void* ptr;
 
         /* Check if workSpace is large enough, alloc a new one if needed */
-        {   size_t const optSpace = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits))*sizeof(U32)
-                                  + (ZSTD_OPT_NUM+1)*(sizeof(ZSTD_match_t) + sizeof(ZSTD_optimal_t));
-            size_t const neededSpace = tableSpace + (256*sizeof(U32)) /* huffTable */ + tokenSpace
-                                  + (((params.cParams.strategy == ZSTD_btopt) || (params.cParams.strategy == ZSTD_btopt2)) ? optSpace : 0);
-            if (zc->workSpaceSize < neededSpace) {
+        {   size_t const entropySpace = HUF_WORKSPACE_SIZE;
+            size_t const blockStateSpace = 2 * sizeof(ZSTD_compressedBlockState_t);
+            size_t const bufferSpace = buffInSize + buffOutSize;
+            size_t const ldmSpace = ZSTD_ldm_getTableSize(params.ldmParams);
+            size_t const ldmSeqSpace = maxNbLdmSeq * sizeof(rawSeq);
+
+            size_t const neededSpace = entropySpace + blockStateSpace + ldmSpace +
+                                       ldmSeqSpace + matchStateSize + tokenSpace +
+                                       bufferSpace;
+            DEBUGLOG(4, "Need %uKB workspace, including %uKB for match state, and %uKB for buffers",
+                        (U32)(neededSpace>>10), (U32)(matchStateSize>>10), (U32)(bufferSpace>>10));
+            DEBUGLOG(4, "windowSize: %u - blockSize: %u", (U32)windowSize, (U32)blockSize);
+
+            if (zc->workSpaceSize < neededSpace) {  /* too small : resize */
+                DEBUGLOG(4, "Need to update workSpaceSize from %uK to %uK",
+                            (unsigned)(zc->workSpaceSize>>10),
+                            (unsigned)(neededSpace>>10));
+                /* static cctx : no resize, error out */
+                if (zc->staticSize) return ERROR(memory_allocation);
+
+                zc->workSpaceSize = 0;
                 ZSTD_free(zc->workSpace, zc->customMem);
                 zc->workSpace = ZSTD_malloc(neededSpace, zc->customMem);
                 if (zc->workSpace == NULL) return ERROR(memory_allocation);
                 zc->workSpaceSize = neededSpace;
+                ptr = zc->workSpace;
+
+                /* Statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
+                assert(((size_t)zc->workSpace & 3) == 0);   /* ensure correct alignment */
+                assert(zc->workSpaceSize >= 2 * sizeof(ZSTD_compressedBlockState_t));
+                zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)zc->workSpace;
+                zc->blockState.nextCBlock = zc->blockState.prevCBlock + 1;
+                ptr = zc->blockState.nextCBlock + 1;
+                zc->entropyWorkspace = (U32*)ptr;
         }   }
 
-        if (crp!=ZSTDcrp_noMemset) memset(zc->workSpace, 0, tableSpace);   /* reset tables only */
+        /* init params */
+        zc->appliedParams = params;
+        zc->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+        zc->consumedSrcSize = 0;
+        zc->producedCSize = 0;
+        if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
+            zc->appliedParams.fParams.contentSizeFlag = 0;
+        DEBUGLOG(4, "pledged content size : %u ; flag : %u",
+            (U32)pledgedSrcSize, zc->appliedParams.fParams.contentSizeFlag);
+        zc->blockSize = blockSize;
+
         XXH64_reset(&zc->xxhState, 0);
-        zc->hashLog3 = hashLog3;
-        zc->hashTable = (U32*)(zc->workSpace);
-        zc->chainTable = zc->hashTable + hSize;
-        zc->hashTable3 = zc->chainTable + chainSize;
-        ptr = zc->hashTable3 + h3Size;
-        zc->hufTable = (HUF_CElt*)ptr;
-        zc->flagStaticTables = 0;
-        ptr = ((U32*)ptr) + 256;  /* note : HUF_CElt* is incomplete type, size is simulated using U32 */
-
-        zc->nextToUpdate = 1;
-        zc->nextSrc = NULL;
-        zc->base = NULL;
-        zc->dictBase = NULL;
-        zc->dictLimit = 0;
-        zc->lowLimit = 0;
-        zc->params = params;
-        zc->blockSize = blockSize;
-        zc->frameContentSize = frameContentSize;
-        { int i; for (i=0; i<ZSTD_REP_NUM; i++) zc->rep[i] = repStartValue[i]; }
-
-        if ((params.cParams.strategy == ZSTD_btopt) || (params.cParams.strategy == ZSTD_btopt2)) {
-            zc->seqStore.litFreq = (U32*)ptr;
-            zc->seqStore.litLengthFreq = zc->seqStore.litFreq + (1<<Litbits);
-            zc->seqStore.matchLengthFreq = zc->seqStore.litLengthFreq + (MaxLL+1);
-            zc->seqStore.offCodeFreq = zc->seqStore.matchLengthFreq + (MaxML+1);
-            ptr = zc->seqStore.offCodeFreq + (MaxOff+1);
-            zc->seqStore.matchTable = (ZSTD_match_t*)ptr;
-            ptr = zc->seqStore.matchTable + ZSTD_OPT_NUM+1;
-            zc->seqStore.priceTable = (ZSTD_optimal_t*)ptr;
-            ptr = zc->seqStore.priceTable + ZSTD_OPT_NUM+1;
-            zc->seqStore.litLengthSum = 0;
+        zc->stage = ZSTDcs_init;
+        zc->dictID = 0;
+
+        ZSTD_reset_compressedBlockState(zc->blockState.prevCBlock);
+
+        ptr = zc->entropyWorkspace + HUF_WORKSPACE_SIZE_U32;
+
+        /* ldm hash table */
+        /* initialize bucketOffsets table later for pointer alignment */
+        if (params.ldmParams.enableLdm) {
+            size_t const ldmHSize = ((size_t)1) << params.ldmParams.hashLog;
+            memset(ptr, 0, ldmHSize * sizeof(ldmEntry_t));
+            assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */
+            zc->ldmState.hashTable = (ldmEntry_t*)ptr;
+            ptr = zc->ldmState.hashTable + ldmHSize;
+            zc->ldmSequences = (rawSeq*)ptr;
+            ptr = zc->ldmSequences + maxNbLdmSeq;
+            zc->maxNbLdmSequences = maxNbLdmSeq;
+
+            memset(&zc->ldmState.window, 0, sizeof(zc->ldmState.window));
         }
+        assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */
+
+        ptr = ZSTD_reset_matchState(&zc->blockState.matchState, ptr, &params.cParams, crp, /* forCCtx */ 1);
+
+        /* sequences storage */
         zc->seqStore.sequencesStart = (seqDef*)ptr;
         ptr = zc->seqStore.sequencesStart + maxNbSeq;
         zc->seqStore.llCode = (BYTE*) ptr;
         zc->seqStore.mlCode = zc->seqStore.llCode + maxNbSeq;
         zc->seqStore.ofCode = zc->seqStore.mlCode + maxNbSeq;
         zc->seqStore.litStart = zc->seqStore.ofCode + maxNbSeq;
-
-        zc->stage = ZSTDcs_init;
-        zc->dictID = 0;
-        zc->loadedDictEnd = 0;
+        ptr = zc->seqStore.litStart + blockSize;
+
+        /* ldm bucketOffsets table */
+        if (params.ldmParams.enableLdm) {
+            size_t const ldmBucketSize =
+                  ((size_t)1) << (params.ldmParams.hashLog -
+                                  params.ldmParams.bucketSizeLog);
+            memset(ptr, 0, ldmBucketSize);
+            zc->ldmState.bucketOffsets = (BYTE*)ptr;
+            ptr = zc->ldmState.bucketOffsets + ldmBucketSize;
+            ZSTD_window_clear(&zc->ldmState.window);
+        }
+        ZSTD_referenceExternalSequences(zc, NULL, 0);
+
+        /* buffers */
+        zc->inBuffSize = buffInSize;
+        zc->inBuff = (char*)ptr;
+        zc->outBuffSize = buffOutSize;
+        zc->outBuff = zc->inBuff + buffInSize;
 
         return 0;
     }
@@ -333,75 +1094,197 @@
  *        do not use with extDict variant ! */
 void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx) {
     int i;
-    for (i=0; i<ZSTD_REP_NUM; i++) cctx->rep[i] = 0;
+    for (i=0; i<ZSTD_REP_NUM; i++) cctx->blockState.prevCBlock->rep[i] = 0;
+    assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
 }
 
-/*! ZSTD_copyCCtx() :
-*   Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
-*   Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
-*   @return : 0, or an error code */
-size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx, unsigned long long pledgedSrcSize)
+static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
+                            const ZSTD_CDict* cdict,
+                            unsigned windowLog,
+                            ZSTD_frameParameters fParams,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
 {
-    if (srcCCtx->stage!=ZSTDcs_init) return ERROR(stage_wrong);
-
-    memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
-    ZSTD_resetCCtx_advanced(dstCCtx, srcCCtx->params, pledgedSrcSize, ZSTDcrp_noMemset);
+    {   ZSTD_CCtx_params params = cctx->requestedParams;
+        /* Copy only compression parameters related to tables. */
+        params.cParams = cdict->cParams;
+        if (windowLog) params.cParams.windowLog = windowLog;
+        params.fParams = fParams;
+        ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
+                                ZSTDcrp_noMemset, zbuff);
+        assert(cctx->appliedParams.cParams.strategy == cdict->cParams.strategy);
+        assert(cctx->appliedParams.cParams.hashLog == cdict->cParams.hashLog);
+        assert(cctx->appliedParams.cParams.chainLog == cdict->cParams.chainLog);
+    }
 
     /* copy tables */
-    {   size_t const chainSize = (srcCCtx->params.cParams.strategy == ZSTD_fast) ? 0 : (1 << srcCCtx->params.cParams.chainLog);
-        size_t const hSize = ((size_t)1) << srcCCtx->params.cParams.hashLog;
-        size_t const h3Size = (size_t)1 << srcCCtx->hashLog3;
-        size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
-        memcpy(dstCCtx->workSpace, srcCCtx->workSpace, tableSpace);
+    {   size_t const chainSize = (cdict->cParams.strategy == ZSTD_fast) ? 0 : ((size_t)1 << cdict->cParams.chainLog);
+        size_t const hSize =  (size_t)1 << cdict->cParams.hashLog;
+        size_t const tableSpace = (chainSize + hSize) * sizeof(U32);
+        assert((U32*)cctx->blockState.matchState.chainTable == (U32*)cctx->blockState.matchState.hashTable + hSize);  /* chainTable must follow hashTable */
+        assert((U32*)cctx->blockState.matchState.hashTable3 == (U32*)cctx->blockState.matchState.chainTable + chainSize);
+        assert((U32*)cdict->matchState.chainTable == (U32*)cdict->matchState.hashTable + hSize);  /* chainTable must follow hashTable */
+        assert((U32*)cdict->matchState.hashTable3 == (U32*)cdict->matchState.chainTable + chainSize);
+        memcpy(cctx->blockState.matchState.hashTable, cdict->matchState.hashTable, tableSpace);   /* presumes all tables follow each other */
+    }
+    /* Zero the hashTable3, since the cdict never fills it */
+    {   size_t const h3Size = (size_t)1 << cctx->blockState.matchState.hashLog3;
+        assert(cdict->matchState.hashLog3 == 0);
+        memset(cctx->blockState.matchState.hashTable3, 0, h3Size * sizeof(U32));
     }
 
     /* copy dictionary offsets */
-    dstCCtx->nextToUpdate = srcCCtx->nextToUpdate;
-    dstCCtx->nextToUpdate3= srcCCtx->nextToUpdate3;
-    dstCCtx->nextSrc      = srcCCtx->nextSrc;
-    dstCCtx->base         = srcCCtx->base;
-    dstCCtx->dictBase     = srcCCtx->dictBase;
-    dstCCtx->dictLimit    = srcCCtx->dictLimit;
-    dstCCtx->lowLimit     = srcCCtx->lowLimit;
-    dstCCtx->loadedDictEnd= srcCCtx->loadedDictEnd;
-    dstCCtx->dictID       = srcCCtx->dictID;
-
-    /* copy entropy tables */
-    dstCCtx->flagStaticTables = srcCCtx->flagStaticTables;
-    if (srcCCtx->flagStaticTables) {
-        memcpy(dstCCtx->hufTable, srcCCtx->hufTable, 256*4);
-        memcpy(dstCCtx->litlengthCTable, srcCCtx->litlengthCTable, sizeof(dstCCtx->litlengthCTable));
-        memcpy(dstCCtx->matchlengthCTable, srcCCtx->matchlengthCTable, sizeof(dstCCtx->matchlengthCTable));
-        memcpy(dstCCtx->offcodeCTable, srcCCtx->offcodeCTable, sizeof(dstCCtx->offcodeCTable));
+    {
+        ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
+        ZSTD_matchState_t* dstMatchState = &cctx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->nextToUpdate3= srcMatchState->nextToUpdate3;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
     }
+    cctx->dictID = cdict->dictID;
+
+    /* copy block state */
+    memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
 
     return 0;
 }
 
-
-/*! ZSTD_reduceTable() :
-*   reduce table indexes by `reducerValue` */
-static void ZSTD_reduceTable (U32* const table, U32 const size, U32 const reducerValue)
+/*! ZSTD_copyCCtx_internal() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  The "context", in this case, refers to the hash and chain tables,
+ *  entropy tables, and dictionary references.
+ * `windowLog` value is enforced if != 0, otherwise value is copied from srcCCtx.
+ * @return : 0, or an error code */
+static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
+                            const ZSTD_CCtx* srcCCtx,
+                            ZSTD_frameParameters fParams,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
 {
-    U32 u;
-    for (u=0 ; u < size ; u++) {
-        if (table[u] < reducerValue) table[u] = 0;
-        else table[u] -= reducerValue;
+    DEBUGLOG(5, "ZSTD_copyCCtx_internal");
+    if (srcCCtx->stage!=ZSTDcs_init) return ERROR(stage_wrong);
+
+    memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
+    {   ZSTD_CCtx_params params = dstCCtx->requestedParams;
+        /* Copy only compression parameters related to tables. */
+        params.cParams = srcCCtx->appliedParams.cParams;
+        params.fParams = fParams;
+        ZSTD_resetCCtx_internal(dstCCtx, params, pledgedSrcSize,
+                                ZSTDcrp_noMemset, zbuff);
+        assert(dstCCtx->appliedParams.cParams.windowLog == srcCCtx->appliedParams.cParams.windowLog);
+        assert(dstCCtx->appliedParams.cParams.strategy == srcCCtx->appliedParams.cParams.strategy);
+        assert(dstCCtx->appliedParams.cParams.hashLog == srcCCtx->appliedParams.cParams.hashLog);
+        assert(dstCCtx->appliedParams.cParams.chainLog == srcCCtx->appliedParams.cParams.chainLog);
+        assert(dstCCtx->blockState.matchState.hashLog3 == srcCCtx->blockState.matchState.hashLog3);
+    }
+
+    /* copy tables */
+    {   size_t const chainSize = (srcCCtx->appliedParams.cParams.strategy == ZSTD_fast) ? 0 : ((size_t)1 << srcCCtx->appliedParams.cParams.chainLog);
+        size_t const hSize =  (size_t)1 << srcCCtx->appliedParams.cParams.hashLog;
+        size_t const h3Size = (size_t)1 << srcCCtx->blockState.matchState.hashLog3;
+        size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+        assert((U32*)dstCCtx->blockState.matchState.chainTable == (U32*)dstCCtx->blockState.matchState.hashTable + hSize);  /* chainTable must follow hashTable */
+        assert((U32*)dstCCtx->blockState.matchState.hashTable3 == (U32*)dstCCtx->blockState.matchState.chainTable + chainSize);
+        memcpy(dstCCtx->blockState.matchState.hashTable, srcCCtx->blockState.matchState.hashTable, tableSpace);   /* presumes all tables follow each other */
+    }
+
+    /* copy dictionary offsets */
+    {
+        ZSTD_matchState_t const* srcMatchState = &srcCCtx->blockState.matchState;
+        ZSTD_matchState_t* dstMatchState = &dstCCtx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->nextToUpdate3= srcMatchState->nextToUpdate3;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
     }
+    dstCCtx->dictID = srcCCtx->dictID;
+
+    /* copy block state */
+    memcpy(dstCCtx->blockState.prevCBlock, srcCCtx->blockState.prevCBlock, sizeof(*srcCCtx->blockState.prevCBlock));
+
+    return 0;
+}
+
+/*! ZSTD_copyCCtx() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  pledgedSrcSize==0 means "unknown".
+*   @return : 0, or an error code */
+size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx, unsigned long long pledgedSrcSize)
+{
+    ZSTD_frameParameters fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    ZSTD_buffered_policy_e const zbuff = (ZSTD_buffered_policy_e)(srcCCtx->inBuffSize>0);
+    ZSTD_STATIC_ASSERT((U32)ZSTDb_buffered==1);
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    fParams.contentSizeFlag = (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN);
+
+    return ZSTD_copyCCtx_internal(dstCCtx, srcCCtx,
+                                fParams, pledgedSrcSize,
+                                zbuff);
+}
+
+
+#define ZSTD_ROWSIZE 16
+/*! ZSTD_reduceTable() :
+ *  reduce table indexes by `reducerValue`, or squash to zero.
+ *  PreserveMark preserves "unsorted mark" for btlazy2 strategy.
+ *  It must be set to a clear 0/1 value, to remove branch during inlining.
+ *  Presume table size is a multiple of ZSTD_ROWSIZE
+ *  to help auto-vectorization */
+FORCE_INLINE_TEMPLATE void
+ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerValue, int const preserveMark)
+{
+    int const nbRows = (int)size / ZSTD_ROWSIZE;
+    int cellNb = 0;
+    int rowNb;
+    assert((size & (ZSTD_ROWSIZE-1)) == 0);  /* multiple of ZSTD_ROWSIZE */
+    assert(size < (1U<<31));   /* can be casted to int */
+    for (rowNb=0 ; rowNb < nbRows ; rowNb++) {
+        int column;
+        for (column=0; column<ZSTD_ROWSIZE; column++) {
+            if (preserveMark) {
+                U32 const adder = (table[cellNb] == ZSTD_DUBT_UNSORTED_MARK) ? reducerValue : 0;
+                table[cellNb] += adder;
+            }
+            if (table[cellNb] < reducerValue) table[cellNb] = 0;
+            else table[cellNb] -= reducerValue;
+            cellNb++;
+    }   }
+}
+
+static void ZSTD_reduceTable(U32* const table, U32 const size, U32 const reducerValue)
+{
+    ZSTD_reduceTable_internal(table, size, reducerValue, 0);
+}
+
+static void ZSTD_reduceTable_btlazy2(U32* const table, U32 const size, U32 const reducerValue)
+{
+    ZSTD_reduceTable_internal(table, size, reducerValue, 1);
 }
 
 /*! ZSTD_reduceIndex() :
 *   rescale all indexes to avoid future overflow (indexes are U32) */
 static void ZSTD_reduceIndex (ZSTD_CCtx* zc, const U32 reducerValue)
 {
-    { U32 const hSize = 1 << zc->params.cParams.hashLog;
-      ZSTD_reduceTable(zc->hashTable, hSize, reducerValue); }
-
-    { U32 const chainSize = (zc->params.cParams.strategy == ZSTD_fast) ? 0 : (1 << zc->params.cParams.chainLog);
-      ZSTD_reduceTable(zc->chainTable, chainSize, reducerValue); }
-
-    { U32 const h3Size = (zc->hashLog3) ? 1 << zc->hashLog3 : 0;
-      ZSTD_reduceTable(zc->hashTable3, h3Size, reducerValue); }
+    ZSTD_matchState_t* const ms = &zc->blockState.matchState;
+    {   U32 const hSize = (U32)1 << zc->appliedParams.cParams.hashLog;
+        ZSTD_reduceTable(ms->hashTable, hSize, reducerValue);
+    }
+
+    if (zc->appliedParams.cParams.strategy != ZSTD_fast) {
+        U32 const chainSize = (U32)1 << zc->appliedParams.cParams.chainLog;
+        if (zc->appliedParams.cParams.strategy == ZSTD_btlazy2)
+            ZSTD_reduceTable_btlazy2(ms->chainTable, chainSize, reducerValue);
+        else
+            ZSTD_reduceTable(ms->chainTable, chainSize, reducerValue);
+    }
+
+    if (ms->hashLog3) {
+        U32 const h3Size = (U32)1 << ms->hashLog3;
+        ZSTD_reduceTable(ms->hashTable3, h3Size, reducerValue);
+    }
 }
 
 
@@ -435,10 +1318,11 @@
         case 2: /* 2 - 2 - 12 */
             MEM_writeLE16(ostart, (U16)((U32)set_basic + (1<<2) + (srcSize<<4)));
             break;
-        default:   /*note : should not be necessary : flSize is within {1,2,3} */
         case 3: /* 2 - 2 - 20 */
             MEM_writeLE32(ostart, (U32)((U32)set_basic + (3<<2) + (srcSize<<4)));
             break;
+        default:   /* not necessary : flSize is {1,2,3} */
+            assert(0);
     }
 
     memcpy(ostart + flSize, src, srcSize);
@@ -460,10 +1344,11 @@
         case 2: /* 2 - 2 - 12 */
             MEM_writeLE16(ostart, (U16)((U32)set_rle + (1<<2) + (srcSize<<4)));
             break;
-        default:   /*note : should not be necessary : flSize is necessarily within {1,2,3} */
         case 3: /* 2 - 2 - 20 */
             MEM_writeLE32(ostart, (U32)((U32)set_rle + (3<<2) + (srcSize<<4)));
             break;
+        default:   /* not necessary : flSize is {1,2,3} */
+            assert(0);
     }
 
     ostart[flSize] = *(const BYTE*)src;
@@ -473,9 +1358,12 @@
 
 static size_t ZSTD_minGain(size_t srcSize) { return (srcSize >> 6) + 2; }
 
-static size_t ZSTD_compressLiterals (ZSTD_CCtx* zc,
+static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t const* prevEntropy,
+                                     ZSTD_entropyCTables_t* nextEntropy,
+                                     ZSTD_strategy strategy, int disableLiteralCompression,
                                      void* dst, size_t dstCapacity,
-                               const void* src, size_t srcSize)
+                               const void* src, size_t srcSize,
+                                     U32* workspace, const int bmi2)
 {
     size_t const minGain = ZSTD_minGain(srcSize);
     size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
@@ -484,27 +1372,50 @@
     symbolEncodingType_e hType = set_compressed;
     size_t cLitSize;
 
+    DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i)",
+                disableLiteralCompression);
+
+    /* Prepare nextEntropy assuming reusing the existing table */
+    nextEntropy->hufCTable_repeatMode = prevEntropy->hufCTable_repeatMode;
+    memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable,
+           sizeof(prevEntropy->hufCTable));
+
+    if (disableLiteralCompression)
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
 
     /* small ? don't even attempt compression (speed opt) */
-#   define LITERAL_NOENTROPY 63
-    {   size_t const minLitSize = zc->flagStaticTables ? 6 : LITERAL_NOENTROPY;
+#   define COMPRESS_LITERALS_SIZE_MIN 63
+    {   size_t const minLitSize = (prevEntropy->hufCTable_repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
         if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
     }
 
     if (dstCapacity < lhSize+1) return ERROR(dstSize_tooSmall);   /* not enough space for compression */
-    if (zc->flagStaticTables && (lhSize==3)) {
-        hType = set_repeat;
-        singleStream = 1;
-        cLitSize = HUF_compress1X_usingCTable(ostart+lhSize, dstCapacity-lhSize, src, srcSize, zc->hufTable);
-    } else {
-        cLitSize = singleStream ? HUF_compress1X_wksp(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11, zc->tmpCounters, sizeof(zc->tmpCounters))
-                                : HUF_compress4X_wksp(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11, zc->tmpCounters, sizeof(zc->tmpCounters));
+    {   HUF_repeat repeat = prevEntropy->hufCTable_repeatMode;
+        int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
+        if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
+        cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
+                                      workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat, bmi2)
+                                : HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
+                                      workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat, bmi2);
+        if (repeat != HUF_repeat_none) {
+            /* reused the existing table */
+            hType = set_repeat;
+        }
     }
 
-    if ((cLitSize==0) | (cLitSize >= srcSize - minGain))
+    if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
+        memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable));
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
-    if (cLitSize==1)
+    }
+    if (cLitSize==1) {
+        memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable));
         return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
+    }
+
+    if (hType == set_compressed) {
+        /* using a newly constructed table */
+        nextEntropy->hufCTable_repeatMode = HUF_repeat_check;
+    }
 
     /* Build header */
     switch(lhSize)
@@ -519,40 +1430,21 @@
             MEM_writeLE32(ostart, lhc);
             break;
         }
-    default:   /* should not be necessary, lhSize is only {3,4,5} */
     case 5: /* 2 - 2 - 18 - 18 */
         {   U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
             MEM_writeLE32(ostart, lhc);
             ostart[4] = (BYTE)(cLitSize >> 10);
             break;
         }
+    default:  /* not possible : lhSize is {3,4,5} */
+        assert(0);
     }
     return lhSize+cLitSize;
 }
 
-static const BYTE LL_Code[64] = {  0,  1,  2,  3,  4,  5,  6,  7,
-                                   8,  9, 10, 11, 12, 13, 14, 15,
-                                  16, 16, 17, 17, 18, 18, 19, 19,
-                                  20, 20, 20, 20, 21, 21, 21, 21,
-                                  22, 22, 22, 22, 22, 22, 22, 22,
-                                  23, 23, 23, 23, 23, 23, 23, 23,
-                                  24, 24, 24, 24, 24, 24, 24, 24,
-                                  24, 24, 24, 24, 24, 24, 24, 24 };
-
-static const BYTE ML_Code[128] = { 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15,
-                                  16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
-                                  32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37,
-                                  38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39,
-                                  40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
-                                  41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41,
-                                  42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42,
-                                  42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42 };
-
 
 void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
 {
-    BYTE const LL_deltaCode = 19;
-    BYTE const ML_deltaCode = 36;
     const seqDef* const sequences = seqStorePtr->sequencesStart;
     BYTE* const llCodeTable = seqStorePtr->llCode;
     BYTE* const ofCodeTable = seqStorePtr->ofCode;
@@ -562,9 +1454,9 @@
     for (u=0; u<nbSeq; u++) {
         U32 const llv = sequences[u].litLength;
         U32 const mlv = sequences[u].matchLength;
-        llCodeTable[u] = (llv> 63) ? (BYTE)ZSTD_highbit32(llv) + LL_deltaCode : LL_Code[llv];
+        llCodeTable[u] = (BYTE)ZSTD_LLcode(llv);
         ofCodeTable[u] = (BYTE)ZSTD_highbit32(sequences[u].offset);
-        mlCodeTable[u] = (mlv>127) ? (BYTE)ZSTD_highbit32(mlv) + ML_deltaCode : ML_Code[mlv];
+        mlCodeTable[u] = (BYTE)ZSTD_MLcode(mlv);
     }
     if (seqStorePtr->longLengthID==1)
         llCodeTable[seqStorePtr->longLengthPos] = MaxLL;
@@ -572,17 +1464,252 @@
         mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
 }
 
-
-size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
-                              void* dst, size_t dstCapacity,
-                              size_t srcSize)
+typedef enum {
+    ZSTD_defaultDisallowed = 0,
+    ZSTD_defaultAllowed = 1
+} ZSTD_defaultPolicy_e;
+
+MEM_STATIC
+symbolEncodingType_e ZSTD_selectEncodingType(
+        FSE_repeat* repeatMode, size_t const mostFrequent, size_t nbSeq,
+        U32 defaultNormLog, ZSTD_defaultPolicy_e const isDefaultAllowed)
+{
+#define MIN_SEQ_FOR_DYNAMIC_FSE   64
+#define MAX_SEQ_FOR_STATIC_FSE  1000
+    ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
+    if ((mostFrequent == nbSeq) && (!isDefaultAllowed || nbSeq > 2)) {
+        DEBUGLOG(5, "Selected set_rle");
+        /* Prefer set_basic over set_rle when there are 2 or less symbols,
+         * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
+         * If basic encoding isn't possible, always choose RLE.
+         */
+        *repeatMode = FSE_repeat_check;
+        return set_rle;
+    }
+    if ( isDefaultAllowed
+      && (*repeatMode == FSE_repeat_valid) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
+        DEBUGLOG(5, "Selected set_repeat");
+        return set_repeat;
+    }
+    if ( isDefaultAllowed
+      && ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (defaultNormLog-1)))) ) {
+        DEBUGLOG(5, "Selected set_basic");
+        /* The format allows default tables to be repeated, but it isn't useful.
+         * When using simple heuristics to select encoding type, we don't want
+         * to confuse these tables with dictionaries. When running more careful
+         * analysis, we don't need to waste time checking both repeating tables
+         * and default tables.
+         */
+        *repeatMode = FSE_repeat_none;
+        return set_basic;
+    }
+    DEBUGLOG(5, "Selected set_compressed");
+    *repeatMode = FSE_repeat_check;
+    return set_compressed;
+}
+
+MEM_STATIC
+size_t ZSTD_buildCTable(void* dst, size_t dstCapacity,
+        FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
+        U32* count, U32 max,
+        BYTE const* codeTable, size_t nbSeq,
+        S16 const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+        FSE_CTable const* prevCTable, size_t prevCTableSize,
+        void* workspace, size_t workspaceSize)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE const* const oend = op + dstCapacity;
+
+    switch (type) {
+    case set_rle:
+        *op = codeTable[0];
+        CHECK_F(FSE_buildCTable_rle(nextCTable, (BYTE)max));
+        return 1;
+    case set_repeat:
+        memcpy(nextCTable, prevCTable, prevCTableSize);
+        return 0;
+    case set_basic:
+        CHECK_F(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, workspace, workspaceSize));  /* note : could be pre-calculated */
+        return 0;
+    case set_compressed: {
+        S16 norm[MaxSeq + 1];
+        size_t nbSeq_1 = nbSeq;
+        const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
+        if (count[codeTable[nbSeq-1]] > 1) {
+            count[codeTable[nbSeq-1]]--;
+            nbSeq_1--;
+        }
+        assert(nbSeq_1 > 1);
+        CHECK_F(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max));
+        {   size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog);   /* overflow protected */
+            if (FSE_isError(NCountSize)) return NCountSize;
+            CHECK_F(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, workspace, workspaceSize));
+            return NCountSize;
+        }
+    }
+    default: return assert(0), ERROR(GENERIC);
+    }
+}
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_encodeSequences_body(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
 {
-    const seqStore_t* seqStorePtr = &(zc->seqStore);
+    BIT_CStream_t blockStream;
+    FSE_CState_t  stateMatchLength;
+    FSE_CState_t  stateOffsetBits;
+    FSE_CState_t  stateLitLength;
+
+    CHECK_E(BIT_initCStream(&blockStream, dst, dstCapacity), dstSize_tooSmall); /* not enough space remaining */
+
+    /* first symbols */
+    FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateOffsetBits,  CTable_OffsetBits,  ofCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    if (longOffsets) {
+        U32 const ofBits = ofCodeTable[nbSeq-1];
+        int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+        if (extraBits) {
+            BIT_addBits(&blockStream, sequences[nbSeq-1].offset, extraBits);
+            BIT_flushBits(&blockStream);
+        }
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offset >> extraBits,
+                    ofBits - extraBits);
+    } else {
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
+    }
+    BIT_flushBits(&blockStream);
+
+    {   size_t n;
+        for (n=nbSeq-2 ; n<nbSeq ; n--) {      /* intentional underflow */
+            BYTE const llCode = llCodeTable[n];
+            BYTE const ofCode = ofCodeTable[n];
+            BYTE const mlCode = mlCodeTable[n];
+            U32  const llBits = LL_bits[llCode];
+            U32  const ofBits = ofCode;
+            U32  const mlBits = ML_bits[mlCode];
+            DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
+                        sequences[n].litLength,
+                        sequences[n].matchLength + MINMATCH,
+                        sequences[n].offset);
+                                                                            /* 32b*/  /* 64b*/
+                                                                            /* (7)*/  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
+            FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode);      /* 24 */  /* 24 */
+            if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateLitLength, llCode);        /* 16 */  /* 33 */
+            if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
+                BIT_flushBits(&blockStream);                                /* (7)*/
+            BIT_addBits(&blockStream, sequences[n].litLength, llBits);
+            if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
+            BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
+            if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
+            if (longOffsets) {
+                int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+                if (extraBits) {
+                    BIT_addBits(&blockStream, sequences[n].offset, extraBits);
+                    BIT_flushBits(&blockStream);                            /* (7)*/
+                }
+                BIT_addBits(&blockStream, sequences[n].offset >> extraBits,
+                            ofBits - extraBits);                            /* 31 */
+            } else {
+                BIT_addBits(&blockStream, sequences[n].offset, ofBits);     /* 31 */
+            }
+            BIT_flushBits(&blockStream);                                    /* (7)*/
+    }   }
+
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing ML state with %u bits", stateMatchLength.stateLog);
+    FSE_flushCState(&blockStream, &stateMatchLength);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing Off state with %u bits", stateOffsetBits.stateLog);
+    FSE_flushCState(&blockStream, &stateOffsetBits);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing LL state with %u bits", stateLitLength.stateLog);
+    FSE_flushCState(&blockStream, &stateLitLength);
+
+    {   size_t const streamSize = BIT_closeCStream(&blockStream);
+        if (streamSize==0) return ERROR(dstSize_tooSmall);   /* not enough space */
+        return streamSize;
+    }
+}
+
+static size_t
+ZSTD_encodeSequences_default(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+
+#if DYNAMIC_BMI2
+
+static TARGET_ATTRIBUTE("bmi2") size_t
+ZSTD_encodeSequences_bmi2(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+#endif
+
+size_t ZSTD_encodeSequences(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return ZSTD_encodeSequences_bmi2(dst, dstCapacity,
+                                         CTable_MatchLength, mlCodeTable,
+                                         CTable_OffsetBits, ofCodeTable,
+                                         CTable_LitLength, llCodeTable,
+                                         sequences, nbSeq, longOffsets);
+    }
+#endif
+    (void)bmi2;
+    return ZSTD_encodeSequences_default(dst, dstCapacity,
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq, longOffsets);
+}
+
+MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
+                              ZSTD_entropyCTables_t const* prevEntropy,
+                              ZSTD_entropyCTables_t* nextEntropy,
+                              ZSTD_CCtx_params const* cctxParams,
+                              void* dst, size_t dstCapacity, U32* workspace,
+                              const int bmi2)
+{
+    const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
     U32 count[MaxSeq+1];
-    S16 norm[MaxSeq+1];
-    FSE_CTable* CTable_LitLength = zc->litlengthCTable;
-    FSE_CTable* CTable_OffsetBits = zc->offcodeCTable;
-    FSE_CTable* CTable_MatchLength = zc->matchlengthCTable;
+    FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
+    FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
     U32 LLtype, Offtype, MLtype;   /* compressed, raw or rle */
     const seqDef* const sequences = seqStorePtr->sequencesStart;
     const BYTE* const ofCodeTable = seqStorePtr->ofCode;
@@ -593,1699 +1720,261 @@
     BYTE* op = ostart;
     size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
     BYTE* seqHead;
-    BYTE scratchBuffer[1<<MAX(MLFSELog,LLFSELog)];
+
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
 
     /* Compress literals */
     {   const BYTE* const literals = seqStorePtr->litStart;
         size_t const litSize = seqStorePtr->lit - literals;
-        size_t const cSize = ZSTD_compressLiterals(zc, op, dstCapacity, literals, litSize);
-        if (ZSTD_isError(cSize)) return cSize;
+        size_t const cSize = ZSTD_compressLiterals(
+                                    prevEntropy, nextEntropy,
+                                    cctxParams->cParams.strategy, cctxParams->disableLiteralCompression,
+                                    op, dstCapacity,
+                                    literals, litSize,
+                                    workspace, bmi2);
+        if (ZSTD_isError(cSize))
+          return cSize;
+        assert(cSize <= dstCapacity);
         op += cSize;
     }
 
     /* Sequences Header */
-    if ((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead */) return ERROR(dstSize_tooSmall);
-    if (nbSeq < 0x7F) *op++ = (BYTE)nbSeq;
-    else if (nbSeq < LONGNBSEQ) op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
-    else op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
-    if (nbSeq==0) goto _check_compressibility;
+    if ((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/) return ERROR(dstSize_tooSmall);
+    if (nbSeq < 0x7F)
+        *op++ = (BYTE)nbSeq;
+    else if (nbSeq < LONGNBSEQ)
+        op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
+    else
+        op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
+    if (nbSeq==0) {
+      memcpy(nextEntropy->litlengthCTable, prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable));
+      nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
+      memcpy(nextEntropy->offcodeCTable, prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable));
+      nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
+      memcpy(nextEntropy->matchlengthCTable, prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable));
+      nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
+      return op - ostart;
+    }
 
     /* seqHead : flags for FSE encoding type */
     seqHead = op++;
 
-#define MIN_SEQ_FOR_DYNAMIC_FSE   64
-#define MAX_SEQ_FOR_STATIC_FSE  1000
-
     /* convert length/distances into codes */
     ZSTD_seqToCodes(seqStorePtr);
-
-    /* CTable for Literal Lengths */
+    /* build CTable for Literal Lengths */
     {   U32 max = MaxLL;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, zc->tmpCounters);
-        if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
-            *op++ = llCodeTable[0];
-            FSE_buildCTable_rle(CTable_LitLength, (BYTE)max);
-            LLtype = set_rle;
-        } else if ((zc->flagStaticTables) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
-            LLtype = set_repeat;
-        } else if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (LL_defaultNormLog-1)))) {
-            FSE_buildCTable_wksp(CTable_LitLength, LL_defaultNorm, MaxLL, LL_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
-            LLtype = set_basic;
-        } else {
-            size_t nbSeq_1 = nbSeq;
-            const U32 tableLog = FSE_optimalTableLog(LLFSELog, nbSeq, max);
-            if (count[llCodeTable[nbSeq-1]]>1) { count[llCodeTable[nbSeq-1]]--; nbSeq_1--; }
-            FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max);
-            { size_t const NCountSize = FSE_writeNCount(op, oend-op, norm, max, tableLog);   /* overflow protected */
-              if (FSE_isError(NCountSize)) return ERROR(GENERIC);
-              op += NCountSize; }
-            FSE_buildCTable_wksp(CTable_LitLength, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
-            LLtype = set_compressed;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, workspace);
+        DEBUGLOG(5, "Building LL table");
+        nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
+        LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode, mostFrequent, nbSeq, LL_defaultNormLog, ZSTD_defaultAllowed);
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
+                    count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                    prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable),
+                    workspace, HUF_WORKSPACE_SIZE);
+            if (ZSTD_isError(countSize)) return countSize;
+            op += countSize;
     }   }
-
-    /* CTable for Offsets */
+    /* build CTable for Offsets */
     {   U32 max = MaxOff;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, zc->tmpCounters);
-        if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
-            *op++ = ofCodeTable[0];
-            FSE_buildCTable_rle(CTable_OffsetBits, (BYTE)max);
-            Offtype = set_rle;
-        } else if ((zc->flagStaticTables) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
-            Offtype = set_repeat;
-        } else if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (OF_defaultNormLog-1)))) {
-            FSE_buildCTable_wksp(CTable_OffsetBits, OF_defaultNorm, MaxOff, OF_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
-            Offtype = set_basic;
-        } else {
-            size_t nbSeq_1 = nbSeq;
-            const U32 tableLog = FSE_optimalTableLog(OffFSELog, nbSeq, max);
-            if (count[ofCodeTable[nbSeq-1]]>1) { count[ofCodeTable[nbSeq-1]]--; nbSeq_1--; }
-            FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max);
-            { size_t const NCountSize = FSE_writeNCount(op, oend-op, norm, max, tableLog);   /* overflow protected */
-              if (FSE_isError(NCountSize)) return ERROR(GENERIC);
-              op += NCountSize; }
-            FSE_buildCTable_wksp(CTable_OffsetBits, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
-            Offtype = set_compressed;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, workspace);
+        /* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
+        ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
+        DEBUGLOG(5, "Building OF table");
+        nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
+        Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode, mostFrequent, nbSeq, OF_defaultNormLog, defaultPolicy);
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
+                    count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                    prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable),
+                    workspace, HUF_WORKSPACE_SIZE);
+            if (ZSTD_isError(countSize)) return countSize;
+            op += countSize;
     }   }
-
-    /* CTable for MatchLengths */
+    /* build CTable for MatchLengths */
     {   U32 max = MaxML;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, zc->tmpCounters);
-        if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
-            *op++ = *mlCodeTable;
-            FSE_buildCTable_rle(CTable_MatchLength, (BYTE)max);
-            MLtype = set_rle;
-        } else if ((zc->flagStaticTables) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
-            MLtype = set_repeat;
-        } else if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (ML_defaultNormLog-1)))) {
-            FSE_buildCTable_wksp(CTable_MatchLength, ML_defaultNorm, MaxML, ML_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
-            MLtype = set_basic;
-        } else {
-            size_t nbSeq_1 = nbSeq;
-            const U32 tableLog = FSE_optimalTableLog(MLFSELog, nbSeq, max);
-            if (count[mlCodeTable[nbSeq-1]]>1) { count[mlCodeTable[nbSeq-1]]--; nbSeq_1--; }
-            FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max);
-            { size_t const NCountSize = FSE_writeNCount(op, oend-op, norm, max, tableLog);   /* overflow protected */
-              if (FSE_isError(NCountSize)) return ERROR(GENERIC);
-              op += NCountSize; }
-            FSE_buildCTable_wksp(CTable_MatchLength, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
-            MLtype = set_compressed;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, workspace);
+        DEBUGLOG(5, "Building ML table");
+        nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
+        MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode, mostFrequent, nbSeq, ML_defaultNormLog, ZSTD_defaultAllowed);
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
+                    count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
+                    prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable),
+                    workspace, HUF_WORKSPACE_SIZE);
+            if (ZSTD_isError(countSize)) return countSize;
+            op += countSize;
     }   }
 
     *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
-    zc->flagStaticTables = 0;
-
-    /* Encoding Sequences */
-    {   BIT_CStream_t blockStream;
-        FSE_CState_t  stateMatchLength;
-        FSE_CState_t  stateOffsetBits;
-        FSE_CState_t  stateLitLength;
-
-        CHECK_E(BIT_initCStream(&blockStream, op, oend-op), dstSize_tooSmall); /* not enough space remaining */
-
-        /* first symbols */
-        FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
-        FSE_initCState2(&stateOffsetBits,  CTable_OffsetBits,  ofCodeTable[nbSeq-1]);
-        FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
-        BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
-        if (MEM_32bits()) BIT_flushBits(&blockStream);
-        BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
-        if (MEM_32bits()) BIT_flushBits(&blockStream);
-        BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
-        BIT_flushBits(&blockStream);
-
-        {   size_t n;
-            for (n=nbSeq-2 ; n<nbSeq ; n--) {      /* intentional underflow */
-                BYTE const llCode = llCodeTable[n];
-                BYTE const ofCode = ofCodeTable[n];
-                BYTE const mlCode = mlCodeTable[n];
-                U32  const llBits = LL_bits[llCode];
-                U32  const ofBits = ofCode;                                     /* 32b*/  /* 64b*/
-                U32  const mlBits = ML_bits[mlCode];
-                                                                                /* (7)*/  /* (7)*/
-                FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
-                FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode);      /* 24 */  /* 24 */
-                if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
-                FSE_encodeSymbol(&blockStream, &stateLitLength, llCode);        /* 16 */  /* 33 */
-                if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
-                    BIT_flushBits(&blockStream);                                /* (7)*/
-                BIT_addBits(&blockStream, sequences[n].litLength, llBits);
-                if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
-                BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
-                if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
-                BIT_addBits(&blockStream, sequences[n].offset, ofBits);         /* 31 */
-                BIT_flushBits(&blockStream);                                    /* (7)*/
-        }   }
-
-        FSE_flushCState(&blockStream, &stateMatchLength);
-        FSE_flushCState(&blockStream, &stateOffsetBits);
-        FSE_flushCState(&blockStream, &stateLitLength);
-
-        {   size_t const streamSize = BIT_closeCStream(&blockStream);
-            if (streamSize==0) return ERROR(dstSize_tooSmall);   /* not enough space */
-            op += streamSize;
-    }   }
-
-    /* check compressibility */
-_check_compressibility:
-    { size_t const minGain = ZSTD_minGain(srcSize);
-      size_t const maxCSize = srcSize - minGain;
-      if ((size_t)(op-ostart) >= maxCSize) return 0; }
-
-    /* confirm repcodes */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) zc->rep[i] = zc->repToConfirm[i]; }
+
+    {   size_t const bitstreamSize = ZSTD_encodeSequences(
+                                        op, oend - op,
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq,
+                                        longOffsets, bmi2);
+        if (ZSTD_isError(bitstreamSize)) return bitstreamSize;
+        op += bitstreamSize;
+    }
 
     return op - ostart;
 }
 
-
-#if 0 /* for debug */
-#  define STORESEQ_DEBUG
-#include <stdio.h>   /* fprintf */
-U32 g_startDebug = 0;
-const BYTE* g_start = NULL;
-#endif
-
-/*! ZSTD_storeSeq() :
-    Store a sequence (literal length, literals, offset code and match length code) into seqStore_t.
-    `offsetCode` : distance to match, or 0 == repCode.
-    `matchCode` : matchLength - MINMATCH
-*/
-MEM_STATIC void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const void* literals, U32 offsetCode, size_t matchCode)
-{
-#ifdef STORESEQ_DEBUG
-    if (g_startDebug) {
-        const U32 pos = (U32)((const BYTE*)literals - g_start);
-        if (g_start==NULL) g_start = (const BYTE*)literals;
-        if ((pos > 1895000) && (pos < 1895300))
-            fprintf(stderr, "Cpos %6u :%5u literals & match %3u bytes at distance %6u \n",
-                   pos, (U32)litLength, (U32)matchCode+MINMATCH, (U32)offsetCode);
-    }
-#endif
-    /* copy Literals */
-    ZSTD_wildcopy(seqStorePtr->lit, literals, litLength);
-    seqStorePtr->lit += litLength;
-
-    /* literal Length */
-    if (litLength>0xFFFF) { seqStorePtr->longLengthID = 1; seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart); }
-    seqStorePtr->sequences[0].litLength = (U16)litLength;
-
-    /* match offset */
-    seqStorePtr->sequences[0].offset = offsetCode + 1;
-
-    /* match Length */
-    if (matchCode>0xFFFF) { seqStorePtr->longLengthID = 2; seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart); }
-    seqStorePtr->sequences[0].matchLength = (U16)matchCode;
-
-    seqStorePtr->sequences++;
-}
-
-
-/*-*************************************
-*  Match length counter
-***************************************/
-static unsigned ZSTD_NbCommonBytes (register size_t val)
+MEM_STATIC size_t ZSTD_compressSequences(seqStore_t* seqStorePtr,
+                              ZSTD_entropyCTables_t const* prevEntropy,
+                              ZSTD_entropyCTables_t* nextEntropy,
+                              ZSTD_CCtx_params const* cctxParams,
+                              void* dst, size_t dstCapacity,
+                              size_t srcSize, U32* workspace, int bmi2)
 {
-    if (MEM_isLittleEndian()) {
-        if (MEM_64bits()) {
-#       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            _BitScanForward64( &r, (U64)val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_ctzll((U64)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2, 0, 3, 1, 3, 1, 4, 2, 7, 0, 2, 3, 6, 1, 5, 3, 5, 1, 3, 4, 4, 2, 5, 6, 7, 7, 0, 1, 2, 3, 3, 4, 6, 2, 6, 5, 5, 3, 4, 5, 6, 7, 1, 2, 4, 6, 4, 4, 5, 7, 2, 6, 5, 7, 6, 7, 7 };
-            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
-#       endif
-        } else { /* 32 bits */
-#       if defined(_MSC_VER)
-            unsigned long r=0;
-            _BitScanForward( &r, (U32)val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_ctz((U32)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0, 3, 2, 2, 1, 3, 2, 0, 1, 3, 3, 1, 2, 2, 2, 2, 0, 3, 1, 2, 0, 1, 0, 1, 1 };
-            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
-#       endif
-        }
-    } else {  /* Big Endian CPU */
-        if (MEM_64bits()) {
-#       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            _BitScanReverse64( &r, val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_clzll(val) >> 3);
-#       else
-            unsigned r;
-            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
-            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
-            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-        } else { /* 32 bits */
-#       if defined(_MSC_VER)
-            unsigned long r = 0;
-            _BitScanReverse( &r, (unsigned long)val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_clz((U32)val) >> 3);
-#       else
-            unsigned r;
-            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-    }   }
-}
-
-
-static size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
-{
-    const BYTE* const pStart = pIn;
-    const BYTE* const pInLoopLimit = pInLimit - (sizeof(size_t)-1);
-
-    while (pIn < pInLoopLimit) {
-        size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
-        if (!diff) { pIn+=sizeof(size_t); pMatch+=sizeof(size_t); continue; }
-        pIn += ZSTD_NbCommonBytes(diff);
-        return (size_t)(pIn - pStart);
+    size_t const cSize = ZSTD_compressSequences_internal(
+            seqStorePtr, prevEntropy, nextEntropy, cctxParams, dst, dstCapacity,
+            workspace, bmi2);
+    /* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block.
+     * Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block.
+     */
+    if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity))
+        return 0;  /* block not compressed */
+    if (ZSTD_isError(cSize)) return cSize;
+
+    /* Check compressibility */
+    {   size_t const maxCSize = srcSize - ZSTD_minGain(srcSize);  /* note : fixed formula, maybe should depend on compression level, or strategy */
+        if (cSize >= maxCSize) return 0;  /* block not compressed */
     }
-    if (MEM_64bits()) if ((pIn<(pInLimit-3)) && (MEM_read32(pMatch) == MEM_read32(pIn))) { pIn+=4; pMatch+=4; }
-    if ((pIn<(pInLimit-1)) && (MEM_read16(pMatch) == MEM_read16(pIn))) { pIn+=2; pMatch+=2; }
-    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
-    return (size_t)(pIn - pStart);
-}
-
-/** ZSTD_count_2segments() :
-*   can count match length with `ip` & `match` in 2 different segments.
-*   convention : on reaching mEnd, match count continue starting from iStart
-*/
-static size_t ZSTD_count_2segments(const BYTE* ip, const BYTE* match, const BYTE* iEnd, const BYTE* mEnd, const BYTE* iStart)
-{
-    const BYTE* const vEnd = MIN( ip + (mEnd - match), iEnd);
-    size_t const matchLength = ZSTD_count(ip, match, vEnd);
-    if (match + matchLength != mEnd) return matchLength;
-    return matchLength + ZSTD_count(ip+matchLength, iStart, iEnd);
+
+    /* We check that dictionaries have offset codes available for the first
+     * block. After the first block, the offcode table might not have large
+     * enough codes to represent the offsets in the data.
+     */
+    if (nextEntropy->offcode_repeatMode == FSE_repeat_valid)
+        nextEntropy->offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
 }
 
-
-/*-*************************************
-*  Hashes
-***************************************/
-static const U32 prime3bytes = 506832829U;
-static U32    ZSTD_hash3(U32 u, U32 h) { return ((u << (32-24)) * prime3bytes)  >> (32-h) ; }
-MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h); }   /* only in zstd_opt.h */
-
-static const U32 prime4bytes = 2654435761U;
-static U32    ZSTD_hash4(U32 u, U32 h) { return (u * prime4bytes) >> (32-h) ; }
-static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_read32(ptr), h); }
-
-static const U64 prime5bytes = 889523592379ULL;
-static size_t ZSTD_hash5(U64 u, U32 h) { return (size_t)(((u  << (64-40)) * prime5bytes) >> (64-h)) ; }
-static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h); }
-
-static const U64 prime6bytes = 227718039650203ULL;
-static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
-static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
-
-static const U64 prime7bytes = 58295818150454627ULL;
-static size_t ZSTD_hash7(U64 u, U32 h) { return (size_t)(((u  << (64-56)) * prime7bytes) >> (64-h)) ; }
-static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h); }
-
-static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
-static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
-static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
-
-static size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
-{
-    switch(mls)
-    {
-    default:
-    case 4: return ZSTD_hash4Ptr(p, hBits);
-    case 5: return ZSTD_hash5Ptr(p, hBits);
-    case 6: return ZSTD_hash6Ptr(p, hBits);
-    case 7: return ZSTD_hash7Ptr(p, hBits);
-    case 8: return ZSTD_hash8Ptr(p, hBits);
-    }
-}
-
-
-/*-*************************************
-*  Fast Scan
-***************************************/
-static void ZSTD_fillHashTable (ZSTD_CCtx* zc, const void* end, const U32 mls)
-{
-    U32* const hashTable = zc->hashTable;
-    U32  const hBits = zc->params.cParams.hashLog;
-    const BYTE* const base = zc->base;
-    const BYTE* ip = base + zc->nextToUpdate;
-    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
-    const size_t fastHashFillStep = 3;
-
-    while(ip <= iend) {
-        hashTable[ZSTD_hashPtr(ip, hBits, mls)] = (U32)(ip - base);
-        ip += fastHashFillStep;
-    }
-}
-
-
-FORCE_INLINE
-void ZSTD_compressBlock_fast_generic(ZSTD_CCtx* cctx,
-                               const void* src, size_t srcSize,
-                               const U32 mls)
+/* ZSTD_selectBlockCompressor() :
+ * Not static, but internal use only (used by long distance matcher)
+ * assumption : strat is a valid strategy */
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict)
 {
-    U32* const hashTable = cctx->hashTable;
-    U32  const hBits = cctx->params.cParams.hashLog;
-    seqStore_t* seqStorePtr = &(cctx->seqStore);
-    const BYTE* const base = cctx->base;
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const U32   lowestIndex = cctx->dictLimit;
-    const BYTE* const lowest = base + lowestIndex;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - HASH_READ_SIZE;
-    U32 offset_1=cctx->rep[0], offset_2=cctx->rep[1];
-    U32 offsetSaved = 0;
-
-    /* init */
-    ip += (ip==lowest);
-    {   U32 const maxRep = (U32)(ip-lowest);
-        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
-        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
-    }
-
-    /* Main Search Loop */
-    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
-        size_t mLength;
-        size_t const h = ZSTD_hashPtr(ip, hBits, mls);
-        U32 const current = (U32)(ip-base);
-        U32 const matchIndex = hashTable[h];
-        const BYTE* match = base + matchIndex;
-        hashTable[h] = current;   /* update hash table */
-
-        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
-            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
-            ip++;
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
-        } else {
-            U32 offset;
-            if ( (matchIndex <= lowestIndex) || (MEM_read32(match) != MEM_read32(ip)) ) {
-                ip += ((ip-anchor) >> g_searchStrength) + 1;
-                continue;
-            }
-            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
-            offset = (U32)(ip-match);
-            while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
-            offset_2 = offset_1;
-            offset_1 = offset;
-
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
-        }
-
-        /* match found */
-        ip += mLength;
-        anchor = ip;
-
-        if (ip <= ilimit) {
-            /* Fill Table */
-            hashTable[ZSTD_hashPtr(base+current+2, hBits, mls)] = current+2;  /* here because current+2 could be > iend-8 */
-            hashTable[ZSTD_hashPtr(ip-2, hBits, mls)] = (U32)(ip-2-base);
-            /* check immediate repcode */
-            while ( (ip <= ilimit)
-                 && ( (offset_2>0)
-                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
-                /* store sequence */
-                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; }  /* swap offset_2 <=> offset_1 */
-                hashTable[ZSTD_hashPtr(ip, hBits, mls)] = (U32)(ip-base);
-                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, rLength-MINMATCH);
-                ip += rLength;
-                anchor = ip;
-                continue;   /* faster when present ... (?) */
-    }   }   }
-
-    /* save reps for next block */
-    cctx->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
-    cctx->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
-
-    /* Last Literals */
-    {   size_t const lastLLSize = iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
-    }
+    static const ZSTD_blockCompressor blockCompressor[2][(unsigned)ZSTD_btultra+1] = {
+        { ZSTD_compressBlock_fast  /* default for 0 */,
+          ZSTD_compressBlock_fast, ZSTD_compressBlock_doubleFast, ZSTD_compressBlock_greedy,
+          ZSTD_compressBlock_lazy, ZSTD_compressBlock_lazy2, ZSTD_compressBlock_btlazy2,
+          ZSTD_compressBlock_btopt, ZSTD_compressBlock_btultra },
+        { ZSTD_compressBlock_fast_extDict  /* default for 0 */,
+          ZSTD_compressBlock_fast_extDict, ZSTD_compressBlock_doubleFast_extDict, ZSTD_compressBlock_greedy_extDict,
+          ZSTD_compressBlock_lazy_extDict,ZSTD_compressBlock_lazy2_extDict, ZSTD_compressBlock_btlazy2_extDict,
+          ZSTD_compressBlock_btopt_extDict, ZSTD_compressBlock_btultra_extDict }
+    };
+    ZSTD_STATIC_ASSERT((unsigned)ZSTD_fast == 1);
+
+    assert((U32)strat >= (U32)ZSTD_fast);
+    assert((U32)strat <= (U32)ZSTD_btultra);
+    return blockCompressor[extDict!=0][(U32)strat];
 }
 
-
-static void ZSTD_compressBlock_fast(ZSTD_CCtx* ctx,
-                       const void* src, size_t srcSize)
+static void ZSTD_storeLastLiterals(seqStore_t* seqStorePtr,
+                                   const BYTE* anchor, size_t lastLLSize)
+{
+    memcpy(seqStorePtr->lit, anchor, lastLLSize);
+    seqStorePtr->lit += lastLLSize;
+}
+
+static void ZSTD_resetSeqStore(seqStore_t* ssPtr)
 {
-    const U32 mls = ctx->params.cParams.searchLength;
-    switch(mls)
-    {
-    default:
-    case 4 :
-        ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 4); return;
-    case 5 :
-        ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 5); return;
-    case 6 :
-        ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 6); return;
-    case 7 :
-        ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 7); return;
-    }
+    ssPtr->lit = ssPtr->litStart;
+    ssPtr->sequences = ssPtr->sequencesStart;
+    ssPtr->longLengthID = 0;
 }
 
-
-static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
-                                 const void* src, size_t srcSize,
-                                 const U32 mls)
+static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
+                                        void* dst, size_t dstCapacity,
+                                        const void* src, size_t srcSize)
 {
-    U32* hashTable = ctx->hashTable;
-    const U32 hBits = ctx->params.cParams.hashLog;
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    const BYTE* const base = ctx->base;
-    const BYTE* const dictBase = ctx->dictBase;
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const U32   lowestIndex = ctx->lowLimit;
-    const BYTE* const dictStart = dictBase + lowestIndex;
-    const U32   dictLimit = ctx->dictLimit;
-    const BYTE* const lowPrefixPtr = base + dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - 8;
-    U32 offset_1=ctx->rep[0], offset_2=ctx->rep[1];
-
-    /* Search Loop */
-    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
-        const size_t h = ZSTD_hashPtr(ip, hBits, mls);
-        const U32 matchIndex = hashTable[h];
-        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
-        const BYTE* match = matchBase + matchIndex;
-        const U32 current = (U32)(ip-base);
-        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
-        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
-        const BYTE* repMatch = repBase + repIndex;
-        size_t mLength;
-        hashTable[h] = current;   /* update hash table */
-
-        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
-           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
-            mLength = ZSTD_count_2segments(ip+1+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repMatchEnd, lowPrefixPtr) + EQUAL_READ32;
-            ip++;
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
-        } else {
-            if ( (matchIndex < lowestIndex) ||
-                 (MEM_read32(match) != MEM_read32(ip)) ) {
-                ip += ((ip-anchor) >> g_searchStrength) + 1;
-                continue;
-            }
-            {   const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
-                const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
-                U32 offset;
-                mLength = ZSTD_count_2segments(ip+EQUAL_READ32, match+EQUAL_READ32, iend, matchEnd, lowPrefixPtr) + EQUAL_READ32;
-                while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
-                offset = current - matchIndex;
-                offset_2 = offset_1;
-                offset_1 = offset;
-                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
-        }   }
-
-        /* found a match : store it */
-        ip += mLength;
-        anchor = ip;
-
-        if (ip <= ilimit) {
-            /* Fill Table */
-            hashTable[ZSTD_hashPtr(base+current+2, hBits, mls)] = current+2;
-            hashTable[ZSTD_hashPtr(ip-2, hBits, mls)] = (U32)(ip-2-base);
-            /* check immediate repcode */
-            while (ip <= ilimit) {
-                U32 const current2 = (U32)(ip-base);
-                U32 const repIndex2 = current2 - offset_2;
-                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
-                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
-                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
-                    size_t repLength2 = ZSTD_count_2segments(ip+EQUAL_READ32, repMatch2+EQUAL_READ32, iend, repEnd2, lowPrefixPtr) + EQUAL_READ32;
-                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, repLength2-MINMATCH);
-                    hashTable[ZSTD_hashPtr(ip, hBits, mls)] = current2;
-                    ip += repLength2;
-                    anchor = ip;
-                    continue;
-                }
-                break;
-    }   }   }
-
-    /* save reps for next block */
-    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
-
-    /* Last Literals */
-    {   size_t const lastLLSize = iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
+    ZSTD_matchState_t* const ms = &zc->blockState.matchState;
+    DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+                (U32)dstCapacity, ms->window.dictLimit, ms->nextToUpdate);
+    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
+        ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.searchLength);
+        return 0;   /* don't even attempt compression below a certain srcSize */
     }
-}
-
-
-static void ZSTD_compressBlock_fast_extDict(ZSTD_CCtx* ctx,
-                         const void* src, size_t srcSize)
-{
-    U32 const mls = ctx->params.cParams.searchLength;
-    switch(mls)
-    {
-    default:
-    case 4 :
-        ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 4); return;
-    case 5 :
-        ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 5); return;
-    case 6 :
-        ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 6); return;
-    case 7 :
-        ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 7); return;
-    }
-}
-
-
-/*-*************************************
-*  Double Fast
-***************************************/
-static void ZSTD_fillDoubleHashTable (ZSTD_CCtx* cctx, const void* end, const U32 mls)
-{
-    U32* const hashLarge = cctx->hashTable;
-    U32  const hBitsL = cctx->params.cParams.hashLog;
-    U32* const hashSmall = cctx->chainTable;
-    U32  const hBitsS = cctx->params.cParams.chainLog;
-    const BYTE* const base = cctx->base;
-    const BYTE* ip = base + cctx->nextToUpdate;
-    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
-    const size_t fastHashFillStep = 3;
-
-    while(ip <= iend) {
-        hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip - base);
-        hashLarge[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip - base);
-        ip += fastHashFillStep;
-    }
-}
-
-
-FORCE_INLINE
-void ZSTD_compressBlock_doubleFast_generic(ZSTD_CCtx* cctx,
-                                 const void* src, size_t srcSize,
-                                 const U32 mls)
-{
-    U32* const hashLong = cctx->hashTable;
-    const U32 hBitsL = cctx->params.cParams.hashLog;
-    U32* const hashSmall = cctx->chainTable;
-    const U32 hBitsS = cctx->params.cParams.chainLog;
-    seqStore_t* seqStorePtr = &(cctx->seqStore);
-    const BYTE* const base = cctx->base;
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const U32 lowestIndex = cctx->dictLimit;
-    const BYTE* const lowest = base + lowestIndex;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - HASH_READ_SIZE;
-    U32 offset_1=cctx->rep[0], offset_2=cctx->rep[1];
-    U32 offsetSaved = 0;
-
-    /* init */
-    ip += (ip==lowest);
-    {   U32 const maxRep = (U32)(ip-lowest);
-        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
-        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
+    ZSTD_resetSeqStore(&(zc->seqStore));
+
+    /* limited update after a very long match */
+    {   const BYTE* const base = ms->window.base;
+        const BYTE* const istart = (const BYTE*)src;
+        const U32 current = (U32)(istart-base);
+        if (current > ms->nextToUpdate + 384)
+            ms->nextToUpdate = current - MIN(192, (U32)(current - ms->nextToUpdate - 384));
     }
 
-    /* Main Search Loop */
-    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
-        size_t mLength;
-        size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
-        size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
-        U32 const current = (U32)(ip-base);
-        U32 const matchIndexL = hashLong[h2];
-        U32 const matchIndexS = hashSmall[h];
-        const BYTE* matchLong = base + matchIndexL;
-        const BYTE* match = base + matchIndexS;
-        hashLong[h2] = hashSmall[h] = current;   /* update hash tables */
-
-        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) { /* note : by construction, offset_1 <= current */
-            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
-            ip++;
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
-        } else {
-            U32 offset;
-            if ( (matchIndexL > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip)) ) {
-                mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
-                offset = (U32)(ip-matchLong);
-                while (((ip>anchor) & (matchLong>lowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
-            } else if ( (matchIndexS > lowestIndex) && (MEM_read32(match) == MEM_read32(ip)) ) {
-                size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
-                U32 const matchIndex3 = hashLong[h3];
-                const BYTE* match3 = base + matchIndex3;
-                hashLong[h3] = current + 1;
-                if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
-                    mLength = ZSTD_count(ip+9, match3+8, iend) + 8;
-                    ip++;
-                    offset = (U32)(ip-match3);
-                    while (((ip>anchor) & (match3>lowest)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
-                } else {
-                    mLength = ZSTD_count(ip+4, match+4, iend) + 4;
-                    offset = (U32)(ip-match);
-                    while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
-                }
-            } else {
-                ip += ((ip-anchor) >> g_searchStrength) + 1;
-                continue;
-            }
-
-            offset_2 = offset_1;
-            offset_1 = offset;
-
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+    /* select and store sequences */
+    {   U32 const extDict = ZSTD_window_hasExtDict(ms->window);
+        size_t lastLLSize;
+        {   int i;
+            for (i = 0; i < ZSTD_REP_NUM; ++i)
+                zc->blockState.nextCBlock->rep[i] = zc->blockState.prevCBlock->rep[i];
         }
-
-        /* match found */
-        ip += mLength;
-        anchor = ip;
-
-        if (ip <= ilimit) {
-            /* Fill Table */
-            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] =
-                hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;  /* here because current+2 could be > iend-8 */
-            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] =
-                hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
-
-            /* check immediate repcode */
-            while ( (ip <= ilimit)
-                 && ( (offset_2>0)
-                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
-                /* store sequence */
-                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
-                hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
-                hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
-                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, rLength-MINMATCH);
-                ip += rLength;
-                anchor = ip;
-                continue;   /* faster when present ... (?) */
-    }   }   }
-
-    /* save reps for next block */
-    cctx->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
-    cctx->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
-
-    /* Last Literals */
-    {   size_t const lastLLSize = iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
-    }
-}
-
-
-static void ZSTD_compressBlock_doubleFast(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-    const U32 mls = ctx->params.cParams.searchLength;
-    switch(mls)
-    {
-    default:
-    case 4 :
-        ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 4); return;
-    case 5 :
-        ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 5); return;
-    case 6 :
-        ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 6); return;
-    case 7 :
-        ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 7); return;
-    }
-}
-
-
-static void ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx,
-                                 const void* src, size_t srcSize,
-                                 const U32 mls)
-{
-    U32* const hashLong = ctx->hashTable;
-    U32  const hBitsL = ctx->params.cParams.hashLog;
-    U32* const hashSmall = ctx->chainTable;
-    U32  const hBitsS = ctx->params.cParams.chainLog;
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    const BYTE* const base = ctx->base;
-    const BYTE* const dictBase = ctx->dictBase;
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const U32   lowestIndex = ctx->lowLimit;
-    const BYTE* const dictStart = dictBase + lowestIndex;
-    const U32   dictLimit = ctx->dictLimit;
-    const BYTE* const lowPrefixPtr = base + dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - 8;
-    U32 offset_1=ctx->rep[0], offset_2=ctx->rep[1];
-
-    /* Search Loop */
-    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
-        const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
-        const U32 matchIndex = hashSmall[hSmall];
-        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
-        const BYTE* match = matchBase + matchIndex;
-
-        const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
-        const U32 matchLongIndex = hashLong[hLong];
-        const BYTE* matchLongBase = matchLongIndex < dictLimit ? dictBase : base;
-        const BYTE* matchLong = matchLongBase + matchLongIndex;
-
-        const U32 current = (U32)(ip-base);
-        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
-        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
-        const BYTE* repMatch = repBase + repIndex;
-        size_t mLength;
-        hashSmall[hSmall] = hashLong[hLong] = current;   /* update hash table */
-
-        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
-           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
-            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
-            ip++;
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
-        } else {
-            if ((matchLongIndex > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
-                const BYTE* matchEnd = matchLongIndex < dictLimit ? dictEnd : iend;
-                const BYTE* lowMatchPtr = matchLongIndex < dictLimit ? dictStart : lowPrefixPtr;
-                U32 offset;
-                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, lowPrefixPtr) + 8;
-                offset = current - matchLongIndex;
-                while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
-                offset_2 = offset_1;
-                offset_1 = offset;
-                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
-
-            } else if ((matchIndex > lowestIndex) && (MEM_read32(match) == MEM_read32(ip))) {
-                size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
-                U32 const matchIndex3 = hashLong[h3];
-                const BYTE* const match3Base = matchIndex3 < dictLimit ? dictBase : base;
-                const BYTE* match3 = match3Base + matchIndex3;
-                U32 offset;
-                hashLong[h3] = current + 1;
-                if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
-                    const BYTE* matchEnd = matchIndex3 < dictLimit ? dictEnd : iend;
-                    const BYTE* lowMatchPtr = matchIndex3 < dictLimit ? dictStart : lowPrefixPtr;
-                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, lowPrefixPtr) + 8;
-                    ip++;
-                    offset = current+1 - matchIndex3;
-                    while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
-                } else {
-                    const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
-                    const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
-                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
-                    offset = current - matchIndex;
-                    while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
-                }
-                offset_2 = offset_1;
-                offset_1 = offset;
-                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
-
-            } else {
-                ip += ((ip-anchor) >> g_searchStrength) + 1;
-                continue;
-        }   }
-
-        /* found a match : store it */
-        ip += mLength;
-        anchor = ip;
-
-        if (ip <= ilimit) {
-            /* Fill Table */
-			hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;
-			hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = current+2;
-            hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
-            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
-            /* check immediate repcode */
-            while (ip <= ilimit) {
-                U32 const current2 = (U32)(ip-base);
-                U32 const repIndex2 = current2 - offset_2;
-                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
-                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
-                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
-                    size_t const repLength2 = ZSTD_count_2segments(ip+EQUAL_READ32, repMatch2+EQUAL_READ32, iend, repEnd2, lowPrefixPtr) + EQUAL_READ32;
-                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, repLength2-MINMATCH);
-                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
-                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
-                    ip += repLength2;
-                    anchor = ip;
-                    continue;
-                }
-                break;
-    }   }   }
-
-    /* save reps for next block */
-    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
-
-    /* Last Literals */
-    {   size_t const lastLLSize = iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
-    }
-}
-
-
-static void ZSTD_compressBlock_doubleFast_extDict(ZSTD_CCtx* ctx,
-                         const void* src, size_t srcSize)
-{
-    U32 const mls = ctx->params.cParams.searchLength;
-    switch(mls)
-    {
-    default:
-    case 4 :
-        ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 4); return;
-    case 5 :
-        ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 5); return;
-    case 6 :
-        ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 6); return;
-    case 7 :
-        ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 7); return;
+        if (zc->externSeqStore.pos < zc->externSeqStore.size) {
+            assert(!zc->appliedParams.ldmParams.enableLdm);
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&zc->externSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       &zc->appliedParams.cParams,
+                                       src, srcSize, extDict);
+            assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
+        } else if (zc->appliedParams.ldmParams.enableLdm) {
+            rawSeqStore_t ldmSeqStore = {NULL, 0, 0, 0};
+
+            ldmSeqStore.seq = zc->ldmSequences;
+            ldmSeqStore.capacity = zc->maxNbLdmSequences;
+            /* Updates ldmSeqStore.size */
+            CHECK_F(ZSTD_ldm_generateSequences(&zc->ldmState, &ldmSeqStore,
+                                               &zc->appliedParams.ldmParams,
+                                               src, srcSize));
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&ldmSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       &zc->appliedParams.cParams,
+                                       src, srcSize, extDict);
+            assert(ldmSeqStore.pos == ldmSeqStore.size);
+        } else {   /* not long range mode */
+            ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy, extDict);
+            lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, &zc->appliedParams.cParams, src, srcSize);
+        }
+        {   const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
+            ZSTD_storeLastLiterals(&zc->seqStore, lastLiterals, lastLLSize);
+    }   }
+
+    /* encode sequences and literals */
+    {   size_t const cSize = ZSTD_compressSequences(&zc->seqStore,
+                                &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
+                                &zc->appliedParams,
+                                dst, dstCapacity,
+                                srcSize, zc->entropyWorkspace, zc->bmi2);
+        if (ZSTD_isError(cSize) || cSize == 0) return cSize;
+        /* confirm repcodes and entropy tables */
+        {   ZSTD_compressedBlockState_t* const tmp = zc->blockState.prevCBlock;
+            zc->blockState.prevCBlock = zc->blockState.nextCBlock;
+            zc->blockState.nextCBlock = tmp;
+        }
+        return cSize;
     }
 }
 
 
-/*-*************************************
-*  Binary Tree search
-***************************************/
-/** ZSTD_insertBt1() : add one or multiple positions to tree.
-*   ip : assumed <= iend-8 .
-*   @return : nb of positions added */
-static U32 ZSTD_insertBt1(ZSTD_CCtx* zc, const BYTE* const ip, const U32 mls, const BYTE* const iend, U32 nbCompares,
-                          U32 extDict)
-{
-    U32*   const hashTable = zc->hashTable;
-    U32    const hashLog = zc->params.cParams.hashLog;
-    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
-    U32*   const bt = zc->chainTable;
-    U32    const btLog  = zc->params.cParams.chainLog - 1;
-    U32    const btMask = (1 << btLog) - 1;
-    U32 matchIndex = hashTable[h];
-    size_t commonLengthSmaller=0, commonLengthLarger=0;
-    const BYTE* const base = zc->base;
-    const BYTE* const dictBase = zc->dictBase;
-    const U32 dictLimit = zc->dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
-    const BYTE* const prefixStart = base + dictLimit;
-    const BYTE* match;
-    const U32 current = (U32)(ip-base);
-    const U32 btLow = btMask >= current ? 0 : current - btMask;
-    U32* smallerPtr = bt + 2*(current&btMask);
-    U32* largerPtr  = smallerPtr + 1;
-    U32 dummy32;   /* to be nullified at the end */
-    U32 const windowLow = zc->lowLimit;
-    U32 matchEndIdx = current+8;
-    size_t bestLength = 8;
-#ifdef ZSTD_C_PREDICT
-    U32 predictedSmall = *(bt + 2*((current-1)&btMask) + 0);
-    U32 predictedLarge = *(bt + 2*((current-1)&btMask) + 1);
-    predictedSmall += (predictedSmall>0);
-    predictedLarge += (predictedLarge>0);
-#endif /* ZSTD_C_PREDICT */
-
-    hashTable[h] = current;   /* Update Hash Table */
-
-    while (nbCompares-- && (matchIndex > windowLow)) {
-        U32* const nextPtr = bt + 2*(matchIndex & btMask);
-        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
-
-#ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */
-        const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */
-        if (matchIndex == predictedSmall) {
-            /* no need to check length, result known */
-            *smallerPtr = matchIndex;
-            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
-            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
-            predictedSmall = predictPtr[1] + (predictPtr[1]>0);
-            continue;
-        }
-        if (matchIndex == predictedLarge) {
-            *largerPtr = matchIndex;
-            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            largerPtr = nextPtr;
-            matchIndex = nextPtr[0];
-            predictedLarge = predictPtr[0] + (predictPtr[0]>0);
-            continue;
-        }
-#endif
-        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
-            match = base + matchIndex;
-            if (match[matchLength] == ip[matchLength])
-                matchLength += ZSTD_count(ip+matchLength+1, match+matchLength+1, iend) +1;
-        } else {
-            match = dictBase + matchIndex;
-            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
-            if (matchIndex+matchLength >= dictLimit)
-				match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
-        }
-
-        if (matchLength > bestLength) {
-            bestLength = matchLength;
-            if (matchLength > matchEndIdx - matchIndex)
-                matchEndIdx = matchIndex + (U32)matchLength;
-        }
-
-        if (ip+matchLength == iend)   /* equal : no way to know if inf or sup */
-            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt the tree */
-
-        if (match[matchLength] < ip[matchLength]) {  /* necessarily within correct buffer */
-            /* match is smaller than current */
-            *smallerPtr = matchIndex;             /* update smaller idx */
-            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
-            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
-            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
-        } else {
-            /* match is larger than current */
-            *largerPtr = matchIndex;
-            commonLengthLarger = matchLength;
-            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            largerPtr = nextPtr;
-            matchIndex = nextPtr[0];
-    }   }
-
-    *smallerPtr = *largerPtr = 0;
-    if (bestLength > 384) return MIN(192, (U32)(bestLength - 384));   /* speed optimization */
-    if (matchEndIdx > current + 8) return matchEndIdx - current - 8;
-    return 1;
-}
-
-
-static size_t ZSTD_insertBtAndFindBestMatch (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iend,
-                        size_t* offsetPtr,
-                        U32 nbCompares, const U32 mls,
-                        U32 extDict)
-{
-    U32*   const hashTable = zc->hashTable;
-    U32    const hashLog = zc->params.cParams.hashLog;
-    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
-    U32*   const bt = zc->chainTable;
-    U32    const btLog  = zc->params.cParams.chainLog - 1;
-    U32    const btMask = (1 << btLog) - 1;
-    U32 matchIndex  = hashTable[h];
-    size_t commonLengthSmaller=0, commonLengthLarger=0;
-    const BYTE* const base = zc->base;
-    const BYTE* const dictBase = zc->dictBase;
-    const U32 dictLimit = zc->dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
-    const BYTE* const prefixStart = base + dictLimit;
-    const U32 current = (U32)(ip-base);
-    const U32 btLow = btMask >= current ? 0 : current - btMask;
-    const U32 windowLow = zc->lowLimit;
-    U32* smallerPtr = bt + 2*(current&btMask);
-    U32* largerPtr  = bt + 2*(current&btMask) + 1;
-    U32 matchEndIdx = current+8;
-    U32 dummy32;   /* to be nullified at the end */
-    size_t bestLength = 0;
-
-    hashTable[h] = current;   /* Update Hash Table */
-
-    while (nbCompares-- && (matchIndex > windowLow)) {
-        U32* const nextPtr = bt + 2*(matchIndex & btMask);
-        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
-        const BYTE* match;
-
-        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
-            match = base + matchIndex;
-            if (match[matchLength] == ip[matchLength])
-                matchLength += ZSTD_count(ip+matchLength+1, match+matchLength+1, iend) +1;
-        } else {
-            match = dictBase + matchIndex;
-            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
-            if (matchIndex+matchLength >= dictLimit)
-				match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
-        }
-
-        if (matchLength > bestLength) {
-            if (matchLength > matchEndIdx - matchIndex)
-                matchEndIdx = matchIndex + (U32)matchLength;
-            if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
-                bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + current - matchIndex;
-            if (ip+matchLength == iend)   /* equal : no way to know if inf or sup */
-                break;   /* drop, to guarantee consistency (miss a little bit of compression) */
-        }
-
-        if (match[matchLength] < ip[matchLength]) {
-            /* match is smaller than current */
-            *smallerPtr = matchIndex;             /* update smaller idx */
-            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
-            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
-            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
-        } else {
-            /* match is larger than current */
-            *largerPtr = matchIndex;
-            commonLengthLarger = matchLength;
-            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            largerPtr = nextPtr;
-            matchIndex = nextPtr[0];
-    }   }
-
-    *smallerPtr = *largerPtr = 0;
-
-    zc->nextToUpdate = (matchEndIdx > current + 8) ? matchEndIdx - 8 : current+1;
-    return bestLength;
-}
-
-
-static void ZSTD_updateTree(ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iend, const U32 nbCompares, const U32 mls)
-{
-    const BYTE* const base = zc->base;
-    const U32 target = (U32)(ip - base);
-    U32 idx = zc->nextToUpdate;
-
-    while(idx < target)
-        idx += ZSTD_insertBt1(zc, base+idx, mls, iend, nbCompares, 0);
-}
-
-/** ZSTD_BtFindBestMatch() : Tree updater, providing best match */
-static size_t ZSTD_BtFindBestMatch (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 mls)
-{
-    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree(zc, ip, iLimit, maxNbAttempts, mls);
-    return ZSTD_insertBtAndFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, mls, 0);
-}
-
-
-static size_t ZSTD_BtFindBestMatch_selectMLS (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch)
-{
-    switch(matchLengthSearch)
-    {
-    default :
-    case 4 : return ZSTD_BtFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4);
-    case 5 : return ZSTD_BtFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5);
-    case 6 : return ZSTD_BtFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6);
-    }
-}
-
-
-static void ZSTD_updateTree_extDict(ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iend, const U32 nbCompares, const U32 mls)
-{
-    const BYTE* const base = zc->base;
-    const U32 target = (U32)(ip - base);
-    U32 idx = zc->nextToUpdate;
-
-    while (idx < target) idx += ZSTD_insertBt1(zc, base+idx, mls, iend, nbCompares, 1);
-}
-
-
-/** Tree updater, providing best match */
-static size_t ZSTD_BtFindBestMatch_extDict (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 mls)
-{
-    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree_extDict(zc, ip, iLimit, maxNbAttempts, mls);
-    return ZSTD_insertBtAndFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, mls, 1);
-}
-
-
-static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch)
-{
-    switch(matchLengthSearch)
-    {
-    default :
-    case 4 : return ZSTD_BtFindBestMatch_extDict(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4);
-    case 5 : return ZSTD_BtFindBestMatch_extDict(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5);
-    case 6 : return ZSTD_BtFindBestMatch_extDict(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6);
-    }
-}
-
-
-
-/* *********************************
-*  Hash Chain
-***********************************/
-#define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & mask]
-
-/* Update chains up to ip (excluded)
-   Assumption : always within prefix (ie. not within extDict) */
-FORCE_INLINE
-U32 ZSTD_insertAndFindFirstIndex (ZSTD_CCtx* zc, const BYTE* ip, U32 mls)
-{
-    U32* const hashTable  = zc->hashTable;
-    const U32 hashLog = zc->params.cParams.hashLog;
-    U32* const chainTable = zc->chainTable;
-    const U32 chainMask = (1 << zc->params.cParams.chainLog) - 1;
-    const BYTE* const base = zc->base;
-    const U32 target = (U32)(ip - base);
-    U32 idx = zc->nextToUpdate;
-
-    while(idx < target) { /* catch up */
-        size_t const h = ZSTD_hashPtr(base+idx, hashLog, mls);
-        NEXT_IN_CHAIN(idx, chainMask) = hashTable[h];
-        hashTable[h] = idx;
-        idx++;
-    }
-
-    zc->nextToUpdate = target;
-    return hashTable[ZSTD_hashPtr(ip, hashLog, mls)];
-}
-
-
-
-FORCE_INLINE /* inlining is important to hardwire a hot branch (template emulation) */
-size_t ZSTD_HcFindBestMatch_generic (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 mls, const U32 extDict)
-{
-    U32* const chainTable = zc->chainTable;
-    const U32 chainSize = (1 << zc->params.cParams.chainLog);
-    const U32 chainMask = chainSize-1;
-    const BYTE* const base = zc->base;
-    const BYTE* const dictBase = zc->dictBase;
-    const U32 dictLimit = zc->dictLimit;
-    const BYTE* const prefixStart = base + dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
-    const U32 lowLimit = zc->lowLimit;
-    const U32 current = (U32)(ip-base);
-    const U32 minChain = current > chainSize ? current - chainSize : 0;
-    int nbAttempts=maxNbAttempts;
-    size_t ml=EQUAL_READ32-1;
-
-    /* HC4 match finder */
-    U32 matchIndex = ZSTD_insertAndFindFirstIndex (zc, ip, mls);
-
-    for ( ; (matchIndex>lowLimit) & (nbAttempts>0) ; nbAttempts--) {
-        const BYTE* match;
-        size_t currentMl=0;
-        if ((!extDict) || matchIndex >= dictLimit) {
-            match = base + matchIndex;
-            if (match[ml] == ip[ml])   /* potentially better */
-                currentMl = ZSTD_count(ip, match, iLimit);
-        } else {
-            match = dictBase + matchIndex;
-            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
-                currentMl = ZSTD_count_2segments(ip+EQUAL_READ32, match+EQUAL_READ32, iLimit, dictEnd, prefixStart) + EQUAL_READ32;
-        }
-
-        /* save best solution */
-        if (currentMl > ml) { ml = currentMl; *offsetPtr = current - matchIndex + ZSTD_REP_MOVE; if (ip+currentMl == iLimit) break; /* best possible, and avoid read overflow*/ }
-
-        if (matchIndex <= minChain) break;
-        matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
-    }
-
-    return ml;
-}
-
-
-FORCE_INLINE size_t ZSTD_HcFindBestMatch_selectMLS (
-                        ZSTD_CCtx* zc,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch)
-{
-    switch(matchLengthSearch)
-    {
-    default :
-    case 4 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4, 0);
-    case 5 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5, 0);
-    case 6 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6, 0);
-    }
-}
-
-
-FORCE_INLINE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
-                        ZSTD_CCtx* zc,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch)
-{
-    switch(matchLengthSearch)
-    {
-    default :
-    case 4 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4, 1);
-    case 5 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5, 1);
-    case 6 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6, 1);
-    }
-}
-
-
-/* *******************************
-*  Common parser - lazy strategy
-*********************************/
-FORCE_INLINE
-void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
-                                     const void* src, size_t srcSize,
-                                     const U32 searchMethod, const U32 depth)
-{
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - 8;
-    const BYTE* const base = ctx->base + ctx->dictLimit;
-
-    U32 const maxSearches = 1 << ctx->params.cParams.searchLog;
-    U32 const mls = ctx->params.cParams.searchLength;
-
-    typedef size_t (*searchMax_f)(ZSTD_CCtx* zc, const BYTE* ip, const BYTE* iLimit,
-                        size_t* offsetPtr,
-                        U32 maxNbAttempts, U32 matchLengthSearch);
-    searchMax_f const searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS : ZSTD_HcFindBestMatch_selectMLS;
-    U32 offset_1 = ctx->rep[0], offset_2 = ctx->rep[1], savedOffset=0;
-
-    /* init */
-    ip += (ip==base);
-    ctx->nextToUpdate3 = ctx->nextToUpdate;
-    {   U32 const maxRep = (U32)(ip-base);
-        if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
-        if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
-    }
-
-    /* Match Loop */
-    while (ip < ilimit) {
-        size_t matchLength=0;
-        size_t offset=0;
-        const BYTE* start=ip+1;
-
-        /* check repCode */
-        if ((offset_1>0) & (MEM_read32(ip+1) == MEM_read32(ip+1 - offset_1))) {
-            /* repcode : we take it */
-            matchLength = ZSTD_count(ip+1+EQUAL_READ32, ip+1+EQUAL_READ32-offset_1, iend) + EQUAL_READ32;
-            if (depth==0) goto _storeSequence;
-        }
-
-        /* first search (depth 0) */
-        {   size_t offsetFound = 99999999;
-            size_t const ml2 = searchMax(ctx, ip, iend, &offsetFound, maxSearches, mls);
-            if (ml2 > matchLength)
-                matchLength = ml2, start = ip, offset=offsetFound;
-        }
-
-        if (matchLength < EQUAL_READ32) {
-            ip += ((ip-anchor) >> g_searchStrength) + 1;   /* jump faster over incompressible sections */
-            continue;
-        }
-
-        /* let's try to find a better solution */
-        if (depth>=1)
-        while (ip<ilimit) {
-            ip ++;
-            if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
-                size_t const mlRep = ZSTD_count(ip+EQUAL_READ32, ip+EQUAL_READ32-offset_1, iend) + EQUAL_READ32;
-                int const gain2 = (int)(mlRep * 3);
-                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
-                if ((mlRep >= EQUAL_READ32) && (gain2 > gain1))
-                    matchLength = mlRep, offset = 0, start = ip;
-            }
-            {   size_t offset2=99999999;
-                size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
-                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
-                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
-                if ((ml2 >= EQUAL_READ32) && (gain2 > gain1)) {
-                    matchLength = ml2, offset = offset2, start = ip;
-                    continue;   /* search a better one */
-            }   }
-
-            /* let's find an even better one */
-            if ((depth==2) && (ip<ilimit)) {
-                ip ++;
-                if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
-                    size_t const ml2 = ZSTD_count(ip+EQUAL_READ32, ip+EQUAL_READ32-offset_1, iend) + EQUAL_READ32;
-                    int const gain2 = (int)(ml2 * 4);
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
-                    if ((ml2 >= EQUAL_READ32) && (gain2 > gain1))
-                        matchLength = ml2, offset = 0, start = ip;
-                }
-                {   size_t offset2=99999999;
-                    size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
-                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
-                    if ((ml2 >= EQUAL_READ32) && (gain2 > gain1)) {
-                        matchLength = ml2, offset = offset2, start = ip;
-                        continue;
-            }   }   }
-            break;  /* nothing found : store previous solution */
-        }
-
-        /* catch up */
-        if (offset) {
-            while ((start>anchor) && (start>base+offset-ZSTD_REP_MOVE) && (start[-1] == start[-1-offset+ZSTD_REP_MOVE]))   /* only search for offset within prefix */
-                { start--; matchLength++; }
-            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
-        }
-
-        /* store sequence */
-_storeSequence:
-        {   size_t const litLength = start - anchor;
-            ZSTD_storeSeq(seqStorePtr, litLength, anchor, (U32)offset, matchLength-MINMATCH);
-            anchor = ip = start + matchLength;
-        }
-
-        /* check immediate repcode */
-        while ( (ip <= ilimit)
-             && ((offset_2>0)
-             & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
-            /* store sequence */
-            matchLength = ZSTD_count(ip+EQUAL_READ32, ip+EQUAL_READ32-offset_2, iend) + EQUAL_READ32;
-            offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset; /* swap repcodes */
-            ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, matchLength-MINMATCH);
-            ip += matchLength;
-            anchor = ip;
-            continue;   /* faster when present ... (?) */
-    }   }
-
-    /* Save reps for next block */
-    ctx->repToConfirm[0] = offset_1 ? offset_1 : savedOffset;
-    ctx->repToConfirm[1] = offset_2 ? offset_2 : savedOffset;
-
-    /* Last Literals */
-    {   size_t const lastLLSize = iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
-    }
-}
-
-
-static void ZSTD_compressBlock_btlazy2(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-    ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 1, 2);
-}
-
-static void ZSTD_compressBlock_lazy2(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-    ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 0, 2);
-}
-
-static void ZSTD_compressBlock_lazy(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-    ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 0, 1);
-}
-
-static void ZSTD_compressBlock_greedy(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-    ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 0, 0);
-}
-
-
-FORCE_INLINE
-void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
-                                     const void* src, size_t srcSize,
-                                     const U32 searchMethod, const U32 depth)
-{
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - 8;
-    const BYTE* const base = ctx->base;
-    const U32 dictLimit = ctx->dictLimit;
-    const U32 lowestIndex = ctx->lowLimit;
-    const BYTE* const prefixStart = base + dictLimit;
-    const BYTE* const dictBase = ctx->dictBase;
-    const BYTE* const dictEnd  = dictBase + dictLimit;
-    const BYTE* const dictStart  = dictBase + ctx->lowLimit;
-
-    const U32 maxSearches = 1 << ctx->params.cParams.searchLog;
-    const U32 mls = ctx->params.cParams.searchLength;
-
-    typedef size_t (*searchMax_f)(ZSTD_CCtx* zc, const BYTE* ip, const BYTE* iLimit,
-                        size_t* offsetPtr,
-                        U32 maxNbAttempts, U32 matchLengthSearch);
-    searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS_extDict : ZSTD_HcFindBestMatch_extDict_selectMLS;
-
-    U32 offset_1 = ctx->rep[0], offset_2 = ctx->rep[1];
-
-    /* init */
-    ctx->nextToUpdate3 = ctx->nextToUpdate;
-    ip += (ip == prefixStart);
-
-    /* Match Loop */
-    while (ip < ilimit) {
-        size_t matchLength=0;
-        size_t offset=0;
-        const BYTE* start=ip+1;
-        U32 current = (U32)(ip-base);
-
-        /* check repCode */
-        {   const U32 repIndex = (U32)(current+1 - offset_1);
-            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
-            const BYTE* const repMatch = repBase + repIndex;
-            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))   /* intentional overflow */
-            if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
-                /* repcode detected we should take it */
-                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
-                matchLength = ZSTD_count_2segments(ip+1+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repEnd, prefixStart) + EQUAL_READ32;
-                if (depth==0) goto _storeSequence;
-        }   }
-
-        /* first search (depth 0) */
-        {   size_t offsetFound = 99999999;
-            size_t const ml2 = searchMax(ctx, ip, iend, &offsetFound, maxSearches, mls);
-            if (ml2 > matchLength)
-                matchLength = ml2, start = ip, offset=offsetFound;
-        }
-
-         if (matchLength < EQUAL_READ32) {
-            ip += ((ip-anchor) >> g_searchStrength) + 1;   /* jump faster over incompressible sections */
-            continue;
-        }
-
-        /* let's try to find a better solution */
-        if (depth>=1)
-        while (ip<ilimit) {
-            ip ++;
-            current++;
-            /* check repCode */
-            if (offset) {
-                const U32 repIndex = (U32)(current - offset_1);
-                const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
-                const BYTE* const repMatch = repBase + repIndex;
-                if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
-                if (MEM_read32(ip) == MEM_read32(repMatch)) {
-                    /* repcode detected */
-                    const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
-                    size_t const repLength = ZSTD_count_2segments(ip+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repEnd, prefixStart) + EQUAL_READ32;
-                    int const gain2 = (int)(repLength * 3);
-                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
-                    if ((repLength >= EQUAL_READ32) && (gain2 > gain1))
-                        matchLength = repLength, offset = 0, start = ip;
-            }   }
-
-            /* search match, depth 1 */
-            {   size_t offset2=99999999;
-                size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
-                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
-                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
-                if ((ml2 >= EQUAL_READ32) && (gain2 > gain1)) {
-                    matchLength = ml2, offset = offset2, start = ip;
-                    continue;   /* search a better one */
-            }   }
-
-            /* let's find an even better one */
-            if ((depth==2) && (ip<ilimit)) {
-                ip ++;
-                current++;
-                /* check repCode */
-                if (offset) {
-                    const U32 repIndex = (U32)(current - offset_1);
-                    const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
-                    const BYTE* const repMatch = repBase + repIndex;
-                    if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
-                    if (MEM_read32(ip) == MEM_read32(repMatch)) {
-                        /* repcode detected */
-                        const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
-                        size_t repLength = ZSTD_count_2segments(ip+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repEnd, prefixStart) + EQUAL_READ32;
-                        int gain2 = (int)(repLength * 4);
-                        int gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
-                        if ((repLength >= EQUAL_READ32) && (gain2 > gain1))
-                            matchLength = repLength, offset = 0, start = ip;
-                }   }
-
-                /* search match, depth 2 */
-                {   size_t offset2=99999999;
-                    size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
-                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
-                    if ((ml2 >= EQUAL_READ32) && (gain2 > gain1)) {
-                        matchLength = ml2, offset = offset2, start = ip;
-                        continue;
-            }   }   }
-            break;  /* nothing found : store previous solution */
-        }
-
-        /* catch up */
-        if (offset) {
-            U32 const matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
-            const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex;
-            const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart;
-            while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
-            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
-        }
-
-        /* store sequence */
-_storeSequence:
-        {   size_t const litLength = start - anchor;
-            ZSTD_storeSeq(seqStorePtr, litLength, anchor, (U32)offset, matchLength-MINMATCH);
-            anchor = ip = start + matchLength;
-        }
-
-        /* check immediate repcode */
-        while (ip <= ilimit) {
-            const U32 repIndex = (U32)((ip-base) - offset_2);
-            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
-            const BYTE* const repMatch = repBase + repIndex;
-            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
-            if (MEM_read32(ip) == MEM_read32(repMatch)) {
-                /* repcode detected we should take it */
-                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
-                matchLength = ZSTD_count_2segments(ip+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repEnd, prefixStart) + EQUAL_READ32;
-                offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset;   /* swap offset history */
-                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, matchLength-MINMATCH);
-                ip += matchLength;
-                anchor = ip;
-                continue;   /* faster when present ... (?) */
-            }
-            break;
-    }   }
-
-    /* Save reps for next block */
-    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
-
-    /* Last Literals */
-    {   size_t const lastLLSize = iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
-    }
-}
-
-
-void ZSTD_compressBlock_greedy_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-    ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 0, 0);
-}
-
-static void ZSTD_compressBlock_lazy_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-    ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 0, 1);
-}
-
-static void ZSTD_compressBlock_lazy2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-    ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 0, 2);
-}
-
-static void ZSTD_compressBlock_btlazy2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-    ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 1, 2);
-}
-
-
-/* The optimal parser */
-#include "zstd_opt.h"
-
-static void ZSTD_compressBlock_btopt(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-#ifdef ZSTD_OPT_H_91842398743
-    ZSTD_compressBlock_opt_generic(ctx, src, srcSize, 0);
-#else
-    (void)ctx; (void)src; (void)srcSize;
-    return;
-#endif
-}
-
-static void ZSTD_compressBlock_btopt2(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-#ifdef ZSTD_OPT_H_91842398743
-    ZSTD_compressBlock_opt_generic(ctx, src, srcSize, 1);
-#else
-    (void)ctx; (void)src; (void)srcSize;
-    return;
-#endif
-}
-
-static void ZSTD_compressBlock_btopt_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-#ifdef ZSTD_OPT_H_91842398743
-    ZSTD_compressBlock_opt_extDict_generic(ctx, src, srcSize, 0);
-#else
-    (void)ctx; (void)src; (void)srcSize;
-    return;
-#endif
-}
-
-static void ZSTD_compressBlock_btopt2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
-{
-#ifdef ZSTD_OPT_H_91842398743
-    ZSTD_compressBlock_opt_extDict_generic(ctx, src, srcSize, 1);
-#else
-    (void)ctx; (void)src; (void)srcSize;
-    return;
-#endif
-}
-
-
-typedef void (*ZSTD_blockCompressor) (ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-
-static ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict)
-{
-    static const ZSTD_blockCompressor blockCompressor[2][8] = {
-        { ZSTD_compressBlock_fast, ZSTD_compressBlock_doubleFast, ZSTD_compressBlock_greedy, ZSTD_compressBlock_lazy, ZSTD_compressBlock_lazy2, ZSTD_compressBlock_btlazy2, ZSTD_compressBlock_btopt, ZSTD_compressBlock_btopt2 },
-        { ZSTD_compressBlock_fast_extDict, ZSTD_compressBlock_doubleFast_extDict, ZSTD_compressBlock_greedy_extDict, ZSTD_compressBlock_lazy_extDict,ZSTD_compressBlock_lazy2_extDict, ZSTD_compressBlock_btlazy2_extDict, ZSTD_compressBlock_btopt_extDict, ZSTD_compressBlock_btopt2_extDict }
-    };
-
-    return blockCompressor[extDict][(U32)strat];
-}
-
-
-static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
-{
-    ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->params.cParams.strategy, zc->lowLimit < zc->dictLimit);
-    const BYTE* const base = zc->base;
-    const BYTE* const istart = (const BYTE*)src;
-    const U32 current = (U32)(istart-base);
-    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) return 0;   /* don't even attempt compression below a certain srcSize */
-    ZSTD_resetSeqStore(&(zc->seqStore));
-    if (current > zc->nextToUpdate + 384)
-        zc->nextToUpdate = current - MIN(192, (U32)(current - zc->nextToUpdate - 384));   /* update tree not updated after finding very long rep matches */
-    blockCompressor(zc, src, srcSize);
-    return ZSTD_compressSequences(zc, dst, dstCapacity, srcSize);
-}
-
-
-/*! ZSTD_compress_generic() :
+/*! ZSTD_compress_frameChunk() :
 *   Compress a chunk of data into one or multiple blocks.
 *   All blocks will be terminated, all input will be consumed.
 *   Function will issue an error if there is not enough `dstCapacity` to hold the compressed content.
 *   Frame is supposed already started (header already produced)
 *   @return : compressed size, or an error code
 */
-static size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
+static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
                                      void* dst, size_t dstCapacity,
                                const void* src, size_t srcSize,
                                      U32 lastFrameChunk)
@@ -2295,61 +1984,62 @@
     const BYTE* ip = (const BYTE*)src;
     BYTE* const ostart = (BYTE*)dst;
     BYTE* op = ostart;
-    U32 const maxDist = 1 << cctx->params.cParams.windowLog;
-
-    if (cctx->params.fParams.checksumFlag && srcSize)
+    U32 const maxDist = (U32)1 << cctx->appliedParams.cParams.windowLog;
+    assert(cctx->appliedParams.cParams.windowLog <= 31);
+
+    DEBUGLOG(5, "ZSTD_compress_frameChunk (blockSize=%u)", (U32)blockSize);
+    if (cctx->appliedParams.fParams.checksumFlag && srcSize)
         XXH64_update(&cctx->xxhState, src, srcSize);
 
     while (remaining) {
+        ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
         U32 const lastBlock = lastFrameChunk & (blockSize >= remaining);
-        size_t cSize;
-
-        if (dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE) return ERROR(dstSize_tooSmall);   /* not enough space to store compressed block */
+
+        if (dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE)
+            return ERROR(dstSize_tooSmall);   /* not enough space to store compressed block */
         if (remaining < blockSize) blockSize = remaining;
 
-        /* preemptive overflow correction */
-        if (cctx->lowLimit > (2U<<30)) {
-            U32 const cycleMask = (1 << ZSTD_cycleLog(cctx->params.cParams.hashLog, cctx->params.cParams.strategy)) - 1;
-            U32 const current = (U32)(ip - cctx->base);
-            U32 const newCurrent = (current & cycleMask) + (1 << cctx->params.cParams.windowLog);
-            U32 const correction = current - newCurrent;
-            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_64 <= 30);
+        if (ZSTD_window_needOverflowCorrection(ms->window, ip + blockSize)) {
+            U32 const cycleLog = ZSTD_cycleLog(cctx->appliedParams.cParams.chainLog, cctx->appliedParams.cParams.strategy);
+            U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, maxDist, ip);
+            ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
+            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
+            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+
             ZSTD_reduceIndex(cctx, correction);
-            cctx->base += correction;
-            cctx->dictBase += correction;
-            cctx->lowLimit -= correction;
-            cctx->dictLimit -= correction;
-            if (cctx->nextToUpdate < correction) cctx->nextToUpdate = 0;
-            else cctx->nextToUpdate -= correction;
+            if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
+            else ms->nextToUpdate -= correction;
+            ms->loadedDictEnd = 0;
         }
-
-        if ((U32)(ip+blockSize - cctx->base) > cctx->loadedDictEnd + maxDist) {
-            /* enforce maxDist */
-            U32 const newLowLimit = (U32)(ip+blockSize - cctx->base) - maxDist;
-            if (cctx->lowLimit < newLowLimit) cctx->lowLimit = newLowLimit;
-            if (cctx->dictLimit < cctx->lowLimit) cctx->dictLimit = cctx->lowLimit;
-        }
-
-        cSize = ZSTD_compressBlock_internal(cctx, op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize, ip, blockSize);
-        if (ZSTD_isError(cSize)) return cSize;
-
-        if (cSize == 0) {  /* block is not compressible */
-            U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(blockSize << 3);
-            if (blockSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
-            MEM_writeLE32(op, cBlockHeader24);   /* no pb, 4th byte will be overwritten */
-            memcpy(op + ZSTD_blockHeaderSize, ip, blockSize);
-            cSize = ZSTD_blockHeaderSize+blockSize;
-        } else {
-            U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
-            MEM_writeLE24(op, cBlockHeader24);
-            cSize += ZSTD_blockHeaderSize;
-        }
-
-        remaining -= blockSize;
-        dstCapacity -= cSize;
-        ip += blockSize;
-        op += cSize;
-    }
+        ZSTD_window_enforceMaxDist(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd);
+        if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
+
+        {   size_t cSize = ZSTD_compressBlock_internal(cctx,
+                                op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize,
+                                ip, blockSize);
+            if (ZSTD_isError(cSize)) return cSize;
+
+            if (cSize == 0) {  /* block is not compressible */
+                U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(blockSize << 3);
+                if (blockSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
+                MEM_writeLE32(op, cBlockHeader24);   /* 4th byte will be overwritten */
+                memcpy(op + ZSTD_blockHeaderSize, ip, blockSize);
+                cSize = ZSTD_blockHeaderSize + blockSize;
+            } else {
+                U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+                MEM_writeLE24(op, cBlockHeader24);
+                cSize += ZSTD_blockHeaderSize;
+            }
+
+            ip += blockSize;
+            assert(remaining >= blockSize);
+            remaining -= blockSize;
+            op += cSize;
+            assert(dstCapacity >= cSize);
+            dstCapacity -= cSize;
+            DEBUGLOG(5, "ZSTD_compress_frameChunk: adding a block of size %u",
+                        (U32)cSize);
+    }   }
 
     if (lastFrameChunk && (op>ostart)) cctx->stage = ZSTDcs_ending;
     return op-ostart;
@@ -2357,27 +2047,32 @@
 
 
 static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
-                                    ZSTD_parameters params, U64 pledgedSrcSize, U32 dictID)
+                                    ZSTD_CCtx_params params, U64 pledgedSrcSize, U32 dictID)
 {   BYTE* const op = (BYTE*)dst;
-    U32   const dictIDSizeCode = (dictID>0) + (dictID>=256) + (dictID>=65536);   /* 0-3 */
+    U32   const dictIDSizeCodeLength = (dictID>0) + (dictID>=256) + (dictID>=65536);   /* 0-3 */
+    U32   const dictIDSizeCode = params.fParams.noDictIDFlag ? 0 : dictIDSizeCodeLength;   /* 0-3 */
     U32   const checksumFlag = params.fParams.checksumFlag>0;
-    U32   const windowSize = 1U << params.cParams.windowLog;
-    U32   const singleSegment = params.fParams.contentSizeFlag && (windowSize > (pledgedSrcSize-1));
+    U32   const windowSize = (U32)1 << params.cParams.windowLog;
+    U32   const singleSegment = params.fParams.contentSizeFlag && (windowSize >= pledgedSrcSize);
     BYTE  const windowLogByte = (BYTE)((params.cParams.windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN) << 3);
     U32   const fcsCode = params.fParams.contentSizeFlag ?
-                     (pledgedSrcSize>=256) + (pledgedSrcSize>=65536+256) + (pledgedSrcSize>=0xFFFFFFFFU) :   /* 0-3 */
-                      0;
+                     (pledgedSrcSize>=256) + (pledgedSrcSize>=65536+256) + (pledgedSrcSize>=0xFFFFFFFFU) : 0;  /* 0-3 */
     BYTE  const frameHeaderDecriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
-    size_t pos;
+    size_t pos=0;
 
     if (dstCapacity < ZSTD_frameHeaderSize_max) return ERROR(dstSize_tooSmall);
-
-    MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
-    op[4] = frameHeaderDecriptionByte; pos=5;
+    DEBUGLOG(4, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
+                !params.fParams.noDictIDFlag, dictID,  dictIDSizeCode);
+
+    if (params.format == ZSTD_f_zstd1) {
+        MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
+        pos = 4;
+    }
+    op[pos++] = frameHeaderDecriptionByte;
     if (!singleSegment) op[pos++] = windowLogByte;
     switch(dictIDSizeCode)
     {
-        default:   /* impossible */
+        default:  assert(0); /* impossible */
         case 0 : break;
         case 1 : op[pos] = (BYTE)(dictID); pos++; break;
         case 2 : MEM_writeLE16(op+pos, (U16)dictID); pos+=2; break;
@@ -2385,7 +2080,7 @@
     }
     switch(fcsCode)
     {
-        default:   /* impossible */
+        default:  assert(0); /* impossible */
         case 0 : if (singleSegment) op[pos++] = (BYTE)(pledgedSrcSize); break;
         case 1 : MEM_writeLE16(op+pos, (U16)(pledgedSrcSize-256)); pos+=2; break;
         case 2 : MEM_writeLE32(op+pos, (U32)(pledgedSrcSize)); pos+=4; break;
@@ -2394,121 +2089,146 @@
     return pos;
 }
 
+/* ZSTD_writeLastEmptyBlock() :
+ * output an empty Block with end-of-frame mark to complete a frame
+ * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
+ *           or an error code if `dstCapcity` is too small (<ZSTD_blockHeaderSize)
+ */
+size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity)
+{
+    if (dstCapacity < ZSTD_blockHeaderSize) return ERROR(dstSize_tooSmall);
+    {   U32 const cBlockHeader24 = 1 /*lastBlock*/ + (((U32)bt_raw)<<1);  /* 0 size */
+        MEM_writeLE24(dst, cBlockHeader24);
+        return ZSTD_blockHeaderSize;
+    }
+}
+
+size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq)
+{
+    if (cctx->stage != ZSTDcs_init)
+        return ERROR(stage_wrong);
+    if (cctx->appliedParams.ldmParams.enableLdm)
+        return ERROR(parameter_unsupported);
+    cctx->externSeqStore.seq = seq;
+    cctx->externSeqStore.size = nbSeq;
+    cctx->externSeqStore.capacity = nbSeq;
+    cctx->externSeqStore.pos = 0;
+    return 0;
+}
+
 
 static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
                               void* dst, size_t dstCapacity,
                         const void* src, size_t srcSize,
                                U32 frame, U32 lastFrameChunk)
 {
-    const BYTE* const ip = (const BYTE*) src;
+    ZSTD_matchState_t* ms = &cctx->blockState.matchState;
     size_t fhSize = 0;
 
+    DEBUGLOG(5, "ZSTD_compressContinue_internal, stage: %u, srcSize: %u",
+                cctx->stage, (U32)srcSize);
     if (cctx->stage==ZSTDcs_created) return ERROR(stage_wrong);   /* missing init (ZSTD_compressBegin) */
 
     if (frame && (cctx->stage==ZSTDcs_init)) {
-        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->params, cctx->frameContentSize, cctx->dictID);
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->appliedParams,
+                                       cctx->pledgedSrcSizePlusOne-1, cctx->dictID);
         if (ZSTD_isError(fhSize)) return fhSize;
         dstCapacity -= fhSize;
         dst = (char*)dst + fhSize;
         cctx->stage = ZSTDcs_ongoing;
     }
 
-    /* Check if blocks follow each other */
-    if (src != cctx->nextSrc) {
-        /* not contiguous */
-        ptrdiff_t const delta = cctx->nextSrc - ip;
-        cctx->lowLimit = cctx->dictLimit;
-        cctx->dictLimit = (U32)(cctx->nextSrc - cctx->base);
-        cctx->dictBase = cctx->base;
-        cctx->base -= delta;
-        cctx->nextToUpdate = cctx->dictLimit;
-        if (cctx->dictLimit - cctx->lowLimit < HASH_READ_SIZE) cctx->lowLimit = cctx->dictLimit;   /* too small extDict */
+    if (!srcSize) return fhSize;  /* do not generate an empty block if no input */
+
+    if (!ZSTD_window_update(&ms->window, src, srcSize)) {
+        ms->nextToUpdate = ms->window.dictLimit;
     }
-
-    /* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
-    if ((ip+srcSize > cctx->dictBase + cctx->lowLimit) & (ip < cctx->dictBase + cctx->dictLimit)) {
-        ptrdiff_t const highInputIdx = (ip + srcSize) - cctx->dictBase;
-        U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)cctx->dictLimit) ? cctx->dictLimit : (U32)highInputIdx;
-        cctx->lowLimit = lowLimitMax;
-    }
-
-    cctx->nextSrc = ip + srcSize;
-
-    if (srcSize) {
-        size_t const cSize = frame ?
-                             ZSTD_compress_generic (cctx, dst, dstCapacity, src, srcSize, lastFrameChunk) :
+    if (cctx->appliedParams.ldmParams.enableLdm)
+        ZSTD_window_update(&cctx->ldmState.window, src, srcSize);
+
+    DEBUGLOG(5, "ZSTD_compressContinue_internal (blockSize=%u)", (U32)cctx->blockSize);
+    {   size_t const cSize = frame ?
+                             ZSTD_compress_frameChunk (cctx, dst, dstCapacity, src, srcSize, lastFrameChunk) :
                              ZSTD_compressBlock_internal (cctx, dst, dstCapacity, src, srcSize);
         if (ZSTD_isError(cSize)) return cSize;
+        cctx->consumedSrcSize += srcSize;
+        cctx->producedCSize += (cSize + fhSize);
+        if (cctx->appliedParams.fParams.contentSizeFlag) {  /* control src size */
+            if (cctx->consumedSrcSize+1 > cctx->pledgedSrcSizePlusOne) {
+                DEBUGLOG(4, "error : pledgedSrcSize = %u, while realSrcSize >= %u",
+                    (U32)cctx->pledgedSrcSizePlusOne-1, (U32)cctx->consumedSrcSize);
+                return ERROR(srcSize_wrong);
+            }
+        }
         return cSize + fhSize;
-    } else
-        return fhSize;
+    }
 }
 
-
 size_t ZSTD_compressContinue (ZSTD_CCtx* cctx,
                               void* dst, size_t dstCapacity,
                         const void* src, size_t srcSize)
 {
-    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1, 0);
+    DEBUGLOG(5, "ZSTD_compressContinue (srcSize=%u)", (U32)srcSize);
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 0 /* last chunk */);
 }
 
 
-size_t ZSTD_getBlockSizeMax(ZSTD_CCtx* cctx)
+size_t ZSTD_getBlockSize(const ZSTD_CCtx* cctx)
 {
-    return MIN (ZSTD_BLOCKSIZE_ABSOLUTEMAX, 1 << cctx->params.cParams.windowLog);
+    ZSTD_compressionParameters const cParams = cctx->appliedParams.cParams;
+    assert(!ZSTD_checkCParams(cParams));
+    return MIN (ZSTD_BLOCKSIZE_MAX, (U32)1 << cParams.windowLog);
 }
 
 size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
-    size_t const blockSizeMax = ZSTD_getBlockSizeMax(cctx);
+    size_t const blockSizeMax = ZSTD_getBlockSize(cctx);
     if (srcSize > blockSizeMax) return ERROR(srcSize_wrong);
-    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0, 0);
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0 /* frame mode */, 0 /* last chunk */);
 }
 
-
-static size_t ZSTD_loadDictionaryContent(ZSTD_CCtx* zc, const void* src, size_t srcSize)
+/*! ZSTD_loadDictionaryContent() :
+ *  @return : 0, or an error code
+ */
+static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const void* src, size_t srcSize)
 {
     const BYTE* const ip = (const BYTE*) src;
     const BYTE* const iend = ip + srcSize;
-
-    /* input becomes current prefix */
-    zc->lowLimit = zc->dictLimit;
-    zc->dictLimit = (U32)(zc->nextSrc - zc->base);
-    zc->dictBase = zc->base;
-    zc->base += ip - zc->nextSrc;
-    zc->nextToUpdate = zc->dictLimit;
-    zc->loadedDictEnd = zc->forceWindow ? 0 : (U32)(iend - zc->base);
-
-    zc->nextSrc = iend;
+    ZSTD_compressionParameters const* cParams = &params->cParams;
+
+    ZSTD_window_update(&ms->window, src, srcSize);
+    ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->window.base);
+
     if (srcSize <= HASH_READ_SIZE) return 0;
 
-    switch(zc->params.cParams.strategy)
+    switch(params->cParams.strategy)
     {
     case ZSTD_fast:
-        ZSTD_fillHashTable (zc, iend, zc->params.cParams.searchLength);
+        ZSTD_fillHashTable(ms, cParams, iend);
         break;
-
     case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable (zc, iend, zc->params.cParams.searchLength);
+        ZSTD_fillDoubleHashTable(ms, cParams, iend);
         break;
 
     case ZSTD_greedy:
     case ZSTD_lazy:
     case ZSTD_lazy2:
-        ZSTD_insertAndFindFirstIndex (zc, iend-HASH_READ_SIZE, zc->params.cParams.searchLength);
+        if (srcSize >= HASH_READ_SIZE)
+            ZSTD_insertAndFindFirstIndex(ms, cParams, iend-HASH_READ_SIZE);
         break;
 
-    case ZSTD_btlazy2:
+    case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
     case ZSTD_btopt:
-    case ZSTD_btopt2:
-        ZSTD_updateTree(zc, iend-HASH_READ_SIZE, iend, 1 << zc->params.cParams.searchLog, zc->params.cParams.searchLength);
+    case ZSTD_btultra:
+        if (srcSize >= HASH_READ_SIZE)
+            ZSTD_updateTree(ms, cParams, iend-HASH_READ_SIZE, iend);
         break;
 
     default:
-        return ERROR(GENERIC);   /* strategy doesn't exist; impossible */
+        assert(0);  /* not possible : not a valid strategy id */
     }
 
-    zc->nextToUpdate = zc->loadedDictEnd;
+    ms->nextToUpdate = (U32)(iend - ms->window.base);
     return 0;
 }
 
@@ -2528,27 +2248,32 @@
 
 
 /* Dictionary format :
-    Magic == ZSTD_DICT_MAGIC (4 bytes)
-    HUF_writeCTable(256)
-    FSE_writeNCount(off)
-    FSE_writeNCount(ml)
-    FSE_writeNCount(ll)
-    RepOffsets
-    Dictionary content
-*/
-/*! ZSTD_loadDictEntropyStats() :
-    @return : size read from dictionary
-    note : magic number supposed already checked */
-static size_t ZSTD_loadDictEntropyStats(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+ * See :
+ * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format
+ */
+/*! ZSTD_loadZstdDictionary() :
+ * @return : dictID, or an error code
+ *  assumptions : magic number supposed already checked
+ *                dictSize supposed > 8
+ */
+static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const void* dict, size_t dictSize, void* workspace)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
     short offcodeNCount[MaxOff+1];
     unsigned offcodeMaxValue = MaxOff;
-    BYTE scratchBuffer[1<<MAX(MLFSELog,LLFSELog)];
-
-    {   size_t const hufHeaderSize = HUF_readCTable(cctx->hufTable, 255, dict, dictSize);
+    size_t dictID;
+
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+
+    dictPtr += 4;   /* skip magic number */
+    dictID = params->fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr);
+    dictPtr += 4;
+
+    {   unsigned maxSymbolValue = 255;
+        size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.hufCTable, &maxSymbolValue, dictPtr, dictEnd-dictPtr);
         if (HUF_isError(hufHeaderSize)) return ERROR(dictionary_corrupted);
+        if (maxSymbolValue < 255) return ERROR(dictionary_corrupted);
         dictPtr += hufHeaderSize;
     }
 
@@ -2557,7 +2282,8 @@
         if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
         if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
         /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
-        CHECK_E (FSE_buildCTable_wksp(cctx->offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog, scratchBuffer, sizeof(scratchBuffer)), dictionary_corrupted);
+        CHECK_E( FSE_buildCTable_wksp(bs->entropy.offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog, workspace, HUF_WORKSPACE_SIZE),
+                 dictionary_corrupted);
         dictPtr += offcodeHeaderSize;
     }
 
@@ -2567,8 +2293,9 @@
         if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
         if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
         /* Every match length code must have non-zero probability */
-        CHECK_F (ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
-        CHECK_E (FSE_buildCTable_wksp(cctx->matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, scratchBuffer, sizeof(scratchBuffer)), dictionary_corrupted);
+        CHECK_F( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
+        CHECK_E( FSE_buildCTable_wksp(bs->entropy.matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, workspace, HUF_WORKSPACE_SIZE),
+                 dictionary_corrupted);
         dictPtr += matchlengthHeaderSize;
     }
 
@@ -2578,60 +2305,122 @@
         if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
         if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
         /* Every literal length code must have non-zero probability */
-        CHECK_F (ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
-        CHECK_E(FSE_buildCTable_wksp(cctx->litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, scratchBuffer, sizeof(scratchBuffer)), dictionary_corrupted);
+        CHECK_F( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
+        CHECK_E( FSE_buildCTable_wksp(bs->entropy.litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, workspace, HUF_WORKSPACE_SIZE),
+                 dictionary_corrupted);
         dictPtr += litlengthHeaderSize;
     }
 
     if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
-    cctx->rep[0] = MEM_readLE32(dictPtr+0); if (cctx->rep[0] == 0 || cctx->rep[0] >= dictSize) return ERROR(dictionary_corrupted);
-    cctx->rep[1] = MEM_readLE32(dictPtr+4); if (cctx->rep[1] == 0 || cctx->rep[1] >= dictSize) return ERROR(dictionary_corrupted);
-    cctx->rep[2] = MEM_readLE32(dictPtr+8); if (cctx->rep[2] == 0 || cctx->rep[2] >= dictSize) return ERROR(dictionary_corrupted);
+    bs->rep[0] = MEM_readLE32(dictPtr+0);
+    bs->rep[1] = MEM_readLE32(dictPtr+4);
+    bs->rep[2] = MEM_readLE32(dictPtr+8);
     dictPtr += 12;
 
-    {   U32 offcodeMax = MaxOff;
-        if ((size_t)(dictEnd - dictPtr) <= ((U32)-1) - 128 KB) {
-            U32 const maxOffset = (U32)(dictEnd - dictPtr) + 128 KB; /* The maximum offset that must be supported */
-            /* Calculate minimum offset code required to represent maxOffset */
-            offcodeMax = ZSTD_highbit32(maxOffset);
+    {   size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
+        U32 offcodeMax = MaxOff;
+        if (dictContentSize <= ((U32)-1) - 128 KB) {
+            U32 const maxOffset = (U32)dictContentSize + 128 KB; /* The maximum offset that must be supported */
+            offcodeMax = ZSTD_highbit32(maxOffset); /* Calculate minimum offset code required to represent maxOffset */
         }
-        /* Every possible supported offset <= dictContentSize + 128 KB must be representable */
+        /* All offset values <= dictContentSize + 128 KB must be representable */
         CHECK_F (ZSTD_checkDictNCount(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff)));
+        /* All repCodes must be <= dictContentSize and != 0*/
+        {   U32 u;
+            for (u=0; u<3; u++) {
+                if (bs->rep[u] == 0) return ERROR(dictionary_corrupted);
+                if (bs->rep[u] > dictContentSize) return ERROR(dictionary_corrupted);
+        }   }
+
+        bs->entropy.hufCTable_repeatMode = HUF_repeat_valid;
+        bs->entropy.offcode_repeatMode = FSE_repeat_valid;
+        bs->entropy.matchlength_repeatMode = FSE_repeat_valid;
+        bs->entropy.litlength_repeatMode = FSE_repeat_valid;
+        CHECK_F(ZSTD_loadDictionaryContent(ms, params, dictPtr, dictContentSize));
+        return dictID;
     }
-
-    cctx->flagStaticTables = 1;
-    return dictPtr - (const BYTE*)dict;
 }
 
 /** ZSTD_compress_insertDictionary() :
-*   @return : 0, or an error code */
-static size_t ZSTD_compress_insertDictionary(ZSTD_CCtx* zc, const void* dict, size_t dictSize)
+*   @return : dictID, or an error code */
+static size_t ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matchState_t* ms,
+                                             ZSTD_CCtx_params const* params,
+                                       const void* dict, size_t dictSize,
+                                             ZSTD_dictContentType_e dictContentType,
+                                             void* workspace)
 {
+    DEBUGLOG(4, "ZSTD_compress_insertDictionary (dictSize=%u)", (U32)dictSize);
     if ((dict==NULL) || (dictSize<=8)) return 0;
 
-    /* default : dict is pure content */
-    if (MEM_readLE32(dict) != ZSTD_DICT_MAGIC) return ZSTD_loadDictionaryContent(zc, dict, dictSize);
-    zc->dictID = zc->params.fParams.noDictIDFlag ? 0 :  MEM_readLE32((const char*)dict+4);
-
-    /* known magic number : dict is parsed for entropy stats and content */
-    {   size_t const loadError = ZSTD_loadDictEntropyStats(zc, (const char*)dict+8 /* skip dictHeader */, dictSize-8);
-        size_t const eSize = loadError + 8;
-        if (ZSTD_isError(loadError)) return loadError;
-        return ZSTD_loadDictionaryContent(zc, (const char*)dict+eSize, dictSize-eSize);
+    ZSTD_reset_compressedBlockState(bs);
+
+    /* dict restricted modes */
+    if (dictContentType == ZSTD_dct_rawContent)
+        return ZSTD_loadDictionaryContent(ms, params, dict, dictSize);
+
+    if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
+        if (dictContentType == ZSTD_dct_auto) {
+            DEBUGLOG(4, "raw content dictionary detected");
+            return ZSTD_loadDictionaryContent(ms, params, dict, dictSize);
+        }
+        if (dictContentType == ZSTD_dct_fullDict)
+            return ERROR(dictionary_wrong);
+        assert(0);   /* impossible */
     }
+
+    /* dict as full zstd dictionary */
+    return ZSTD_loadZstdDictionary(bs, ms, params, dict, dictSize, workspace);
 }
 
 /*! ZSTD_compressBegin_internal() :
-*   @return : 0, or an error code */
-static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
+ * @return : 0, or an error code */
+size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
                              const void* dict, size_t dictSize,
-                                   ZSTD_parameters params, U64 pledgedSrcSize)
+                             ZSTD_dictContentType_e dictContentType,
+                             const ZSTD_CDict* cdict,
+                             ZSTD_CCtx_params params, U64 pledgedSrcSize,
+                             ZSTD_buffered_policy_e zbuff)
 {
-    ZSTD_compResetPolicy_e const crp = dictSize ? ZSTDcrp_fullReset : ZSTDcrp_continue;
-    CHECK_F(ZSTD_resetCCtx_advanced(cctx, params, pledgedSrcSize, crp));
-    return ZSTD_compress_insertDictionary(cctx, dict, dictSize);
+    DEBUGLOG(4, "ZSTD_compressBegin_internal: wlog=%u", params.cParams.windowLog);
+    /* params are supposed to be fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+
+    if (cdict && cdict->dictContentSize>0) {
+        cctx->requestedParams = params;
+        return ZSTD_resetCCtx_usingCDict(cctx, cdict, params.cParams.windowLog,
+                                         params.fParams, pledgedSrcSize, zbuff);
+    }
+
+    CHECK_F( ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
+                                     ZSTDcrp_continue, zbuff) );
+    {
+        size_t const dictID = ZSTD_compress_insertDictionary(
+                cctx->blockState.prevCBlock, &cctx->blockState.matchState,
+                &params, dict, dictSize, dictContentType, cctx->entropyWorkspace);
+        if (ZSTD_isError(dictID)) return dictID;
+        assert(dictID <= (size_t)(U32)-1);
+        cctx->dictID = (U32)dictID;
+    }
+    return 0;
 }
 
+size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    const ZSTD_CDict* cdict,
+                                    ZSTD_CCtx_params params,
+                                    unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_compressBegin_advanced_internal: wlog=%u", params.cParams.windowLog);
+    /* compression parameters verification and optimization */
+    CHECK_F( ZSTD_checkCParams(params.cParams) );
+    return ZSTD_compressBegin_internal(cctx,
+                                       dict, dictSize, dictContentType,
+                                       cdict,
+                                       params, pledgedSrcSize,
+                                       ZSTDb_not_buffered);
+}
 
 /*! ZSTD_compressBegin_advanced() :
 *   @return : 0, or an error code */
@@ -2639,19 +2428,24 @@
                              const void* dict, size_t dictSize,
                                    ZSTD_parameters params, unsigned long long pledgedSrcSize)
 {
-    /* compression parameters verification and optimization */
-    CHECK_F(ZSTD_checkCParams(params.cParams));
-    return ZSTD_compressBegin_internal(cctx, dict, dictSize, params, pledgedSrcSize);
+    ZSTD_CCtx_params const cctxParams =
+            ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
+    return ZSTD_compressBegin_advanced_internal(cctx,
+                                            dict, dictSize, ZSTD_dct_auto,
+                                            NULL /*cdict*/,
+                                            cctxParams, pledgedSrcSize);
 }
 
-
 size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
-    return ZSTD_compressBegin_internal(cctx, dict, dictSize, params, 0);
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
+    ZSTD_CCtx_params const cctxParams =
+            ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
+    DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (U32)dictSize);
+    return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, NULL,
+                                       cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, ZSTDb_not_buffered);
 }
 
-
 size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel)
 {
     return ZSTD_compressBegin_usingDict(cctx, NULL, 0, compressionLevel);
@@ -2667,11 +2461,12 @@
     BYTE* op = ostart;
     size_t fhSize = 0;
 
+    DEBUGLOG(4, "ZSTD_writeEpilogue");
     if (cctx->stage == ZSTDcs_created) return ERROR(stage_wrong);  /* init missing */
 
     /* special case : empty frame */
     if (cctx->stage == ZSTDcs_init) {
-        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->params, 0, 0);
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->appliedParams, 0, 0);
         if (ZSTD_isError(fhSize)) return fhSize;
         dstCapacity -= fhSize;
         op += fhSize;
@@ -2687,9 +2482,10 @@
         dstCapacity -= ZSTD_blockHeaderSize;
     }
 
-    if (cctx->params.fParams.checksumFlag) {
+    if (cctx->appliedParams.fParams.checksumFlag) {
         U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
         if (dstCapacity<4) return ERROR(dstSize_tooSmall);
+        DEBUGLOG(4, "ZSTD_writeEpilogue: write checksum : %08X", checksum);
         MEM_writeLE32(op, checksum);
         op += 4;
     }
@@ -2698,16 +2494,24 @@
     return op-ostart;
 }
 
-
 size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
                          void* dst, size_t dstCapacity,
                    const void* src, size_t srcSize)
 {
     size_t endResult;
-    size_t const cSize = ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1, 1);
+    size_t const cSize = ZSTD_compressContinue_internal(cctx,
+                                dst, dstCapacity, src, srcSize,
+                                1 /* frame mode */, 1 /* last chunk */);
     if (ZSTD_isError(cSize)) return cSize;
     endResult = ZSTD_writeEpilogue(cctx, (char*)dst + cSize, dstCapacity-cSize);
     if (ZSTD_isError(endResult)) return endResult;
+    if (cctx->appliedParams.fParams.contentSizeFlag) {  /* control src size */
+        DEBUGLOG(4, "end of frame : controlling src size");
+        if (cctx->pledgedSrcSizePlusOne != cctx->consumedSrcSize+1) {
+            DEBUGLOG(4, "error : pledgedSrcSize = %u, while realSrcSize = %u",
+                (U32)cctx->pledgedSrcSizePlusOne-1, (U32)cctx->consumedSrcSize);
+            return ERROR(srcSize_wrong);
+    }   }
     return cSize + endResult;
 }
 
@@ -2718,8 +2522,14 @@
                          const void* dict,size_t dictSize,
                                ZSTD_parameters params)
 {
-    CHECK_F(ZSTD_compressBegin_internal(cctx, dict, dictSize, params, srcSize));
-    return ZSTD_compressEnd(cctx, dst,  dstCapacity, src, srcSize);
+    ZSTD_CCtx_params const cctxParams =
+            ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
+    DEBUGLOG(4, "ZSTD_compress_internal");
+    return ZSTD_compress_advanced_internal(cctx,
+                                          dst, dstCapacity,
+                                          src, srcSize,
+                                          dict, dictSize,
+                                          cctxParams);
 }
 
 size_t ZSTD_compress_advanced (ZSTD_CCtx* ctx,
@@ -2728,20 +2538,40 @@
                          const void* dict,size_t dictSize,
                                ZSTD_parameters params)
 {
+    DEBUGLOG(4, "ZSTD_compress_advanced");
     CHECK_F(ZSTD_checkCParams(params.cParams));
     return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
 }
 
-size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, const void* dict, size_t dictSize, int compressionLevel)
+/* Internal */
+size_t ZSTD_compress_advanced_internal(
+        ZSTD_CCtx* cctx,
+        void* dst, size_t dstCapacity,
+        const void* src, size_t srcSize,
+        const void* dict,size_t dictSize,
+        ZSTD_CCtx_params params)
 {
-    ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, dict ? dictSize : 0);
-    params.fParams.contentSizeFlag = 1;
-    return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
+    DEBUGLOG(4, "ZSTD_compress_advanced_internal (srcSize:%u)",
+                (U32)srcSize);
+    CHECK_F( ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, NULL,
+                                         params, srcSize, ZSTDb_not_buffered) );
+    return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
 }
 
-size_t ZSTD_compressCCtx (ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize,
+                               const void* dict, size_t dictSize, int compressionLevel)
 {
-    return ZSTD_compress_usingDict(ctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, srcSize ? srcSize : 1, dict ? dictSize : 0);
+    ZSTD_CCtx_params cctxParams = ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
+    assert(params.fParams.contentSizeFlag == 1);
+    ZSTD_CCtxParam_setParameter(&cctxParams, ZSTD_p_compressLiterals, compressionLevel>=0);
+    return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, dict, dictSize, cctxParams);
+}
+
+size_t ZSTD_compressCCtx (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+{
+    DEBUGLOG(4, "ZSTD_compressCCtx (srcSize=%u)", (U32)srcSize);
+    return ZSTD_compress_usingDict(cctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
 }
 
 size_t ZSTD_compress(void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
@@ -2749,123 +2579,264 @@
     size_t result;
     ZSTD_CCtx ctxBody;
     memset(&ctxBody, 0, sizeof(ctxBody));
-    memcpy(&ctxBody.customMem, &defaultCustomMem, sizeof(ZSTD_customMem));
+    ctxBody.customMem = ZSTD_defaultCMem;
     result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
-    ZSTD_free(ctxBody.workSpace, defaultCustomMem);  /* can't free ctxBody itself, as it's on stack; free only heap content */
+    ZSTD_free(ctxBody.workSpace, ZSTD_defaultCMem);  /* can't free ctxBody itself, as it's on stack; free only heap content */
     return result;
 }
 
 
 /* =====  Dictionary API  ===== */
 
-struct ZSTD_CDict_s {
-    void* dictBuffer;
-    const void* dictContent;
-    size_t dictContentSize;
-    ZSTD_CCtx* refContext;
-};  /* typedef'd tp ZSTD_CDict within "zstd.h" */
+/*! ZSTD_estimateCDictSize_advanced() :
+ *  Estimate amount of memory that will be needed to create a dictionary with following arguments */
+size_t ZSTD_estimateCDictSize_advanced(
+        size_t dictSize, ZSTD_compressionParameters cParams,
+        ZSTD_dictLoadMethod_e dictLoadMethod)
+{
+    DEBUGLOG(5, "sizeof(ZSTD_CDict) : %u", (U32)sizeof(ZSTD_CDict));
+    return sizeof(ZSTD_CDict) + HUF_WORKSPACE_SIZE + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0)
+           + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+}
+
+size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    return ZSTD_estimateCDictSize_advanced(dictSize, cParams, ZSTD_dlm_byCopy);
+}
 
 size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict)
 {
     if (cdict==NULL) return 0;   /* support sizeof on NULL */
-    return ZSTD_sizeof_CCtx(cdict->refContext) + (cdict->dictBuffer ? cdict->dictContentSize : 0) + sizeof(*cdict);
+    DEBUGLOG(5, "sizeof(*cdict) : %u", (U32)sizeof(*cdict));
+    return cdict->workspaceSize + (cdict->dictBuffer ? cdict->dictContentSize : 0) + sizeof(*cdict);
 }
 
-ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize, unsigned byReference,
-                                      ZSTD_parameters params, ZSTD_customMem customMem)
+static size_t ZSTD_initCDict_internal(
+                    ZSTD_CDict* cdict,
+              const void* dictBuffer, size_t dictSize,
+                    ZSTD_dictLoadMethod_e dictLoadMethod,
+                    ZSTD_dictContentType_e dictContentType,
+                    ZSTD_compressionParameters cParams)
 {
-    if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
-    if (!customMem.customAlloc || !customMem.customFree) return NULL;
-
-    {   ZSTD_CDict* const cdict = (ZSTD_CDict*) ZSTD_malloc(sizeof(ZSTD_CDict), customMem);
-        ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(customMem);
-
-        if (!cdict || !cctx) {
+    DEBUGLOG(3, "ZSTD_initCDict_internal, dictContentType %u", (U32)dictContentType);
+    assert(!ZSTD_checkCParams(cParams));
+    cdict->cParams = cParams;
+    if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dictBuffer) || (!dictSize)) {
+        cdict->dictBuffer = NULL;
+        cdict->dictContent = dictBuffer;
+    } else {
+        void* const internalBuffer = ZSTD_malloc(dictSize, cdict->customMem);
+        cdict->dictBuffer = internalBuffer;
+        cdict->dictContent = internalBuffer;
+        if (!internalBuffer) return ERROR(memory_allocation);
+        memcpy(internalBuffer, dictBuffer, dictSize);
+    }
+    cdict->dictContentSize = dictSize;
+
+    /* Reset the state to no dictionary */
+    ZSTD_reset_compressedBlockState(&cdict->cBlockState);
+    {   void* const end = ZSTD_reset_matchState(
+                &cdict->matchState,
+                (U32*)cdict->workspace + HUF_WORKSPACE_SIZE_U32,
+                &cParams, ZSTDcrp_continue, /* forCCtx */ 0);
+        assert(end == (char*)cdict->workspace + cdict->workspaceSize);
+        (void)end;
+    }
+    /* (Maybe) load the dictionary
+     * Skips loading the dictionary if it is <= 8 bytes.
+     */
+    {   ZSTD_CCtx_params params;
+        memset(&params, 0, sizeof(params));
+        params.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+        params.fParams.contentSizeFlag = 1;
+        params.cParams = cParams;
+        {   size_t const dictID = ZSTD_compress_insertDictionary(
+                    &cdict->cBlockState, &cdict->matchState, &params,
+                    cdict->dictContent, cdict->dictContentSize,
+                    dictContentType, cdict->workspace);
+            if (ZSTD_isError(dictID)) return dictID;
+            assert(dictID <= (size_t)(U32)-1);
+            cdict->dictID = (U32)dictID;
+        }
+    }
+
+    return 0;
+}
+
+ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType,
+                                      ZSTD_compressionParameters cParams, ZSTD_customMem customMem)
+{
+    DEBUGLOG(3, "ZSTD_createCDict_advanced, mode %u", (U32)dictContentType);
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+
+    {   ZSTD_CDict* const cdict = (ZSTD_CDict*)ZSTD_malloc(sizeof(ZSTD_CDict), customMem);
+        size_t const workspaceSize = HUF_WORKSPACE_SIZE + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0);
+        void* const workspace = ZSTD_malloc(workspaceSize, customMem);
+
+        if (!cdict || !workspace) {
             ZSTD_free(cdict, customMem);
-            ZSTD_free(cctx, customMem);
+            ZSTD_free(workspace, customMem);
             return NULL;
         }
-
-        if ((byReference) || (!dictBuffer) || (!dictSize)) {
-            cdict->dictBuffer = NULL;
-            cdict->dictContent = dictBuffer;
-        } else {
-            void* const internalBuffer = ZSTD_malloc(dictSize, customMem);
-            if (!internalBuffer) { ZSTD_free(cctx, customMem); ZSTD_free(cdict, customMem); return NULL; }
-            memcpy(internalBuffer, dictBuffer, dictSize);
-            cdict->dictBuffer = internalBuffer;
-            cdict->dictContent = internalBuffer;
+        cdict->customMem = customMem;
+        cdict->workspace = workspace;
+        cdict->workspaceSize = workspaceSize;
+        if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
+                                        dictBuffer, dictSize,
+                                        dictLoadMethod, dictContentType,
+                                        cParams) )) {
+            ZSTD_freeCDict(cdict);
+            return NULL;
         }
 
-        {   size_t const errorCode = ZSTD_compressBegin_advanced(cctx, cdict->dictContent, dictSize, params, 0);
-            if (ZSTD_isError(errorCode)) {
-                ZSTD_free(cdict->dictBuffer, customMem);
-                ZSTD_free(cctx, customMem);
-                ZSTD_free(cdict, customMem);
-                return NULL;
-        }   }
-
-        cdict->refContext = cctx;
-        cdict->dictContentSize = dictSize;
         return cdict;
     }
 }
 
 ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_customMem const allocator = { NULL, NULL, NULL };
-    ZSTD_parameters params = ZSTD_getParams(compressionLevel, 0, dictSize);
-    params.fParams.contentSizeFlag = 1;
-    return ZSTD_createCDict_advanced(dict, dictSize, 0, params, allocator);
+    ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    return ZSTD_createCDict_advanced(dict, dictSize,
+                                     ZSTD_dlm_byCopy, ZSTD_dct_auto,
+                                     cParams, ZSTD_defaultCMem);
 }
 
 ZSTD_CDict* ZSTD_createCDict_byReference(const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_customMem const allocator = { NULL, NULL, NULL };
-    ZSTD_parameters params = ZSTD_getParams(compressionLevel, 0, dictSize);
-    params.fParams.contentSizeFlag = 1;
-    return ZSTD_createCDict_advanced(dict, dictSize, 1, params, allocator);
+    ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    return ZSTD_createCDict_advanced(dict, dictSize,
+                                     ZSTD_dlm_byRef, ZSTD_dct_auto,
+                                     cParams, ZSTD_defaultCMem);
 }
 
 size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
 {
     if (cdict==NULL) return 0;   /* support free on NULL */
-    {   ZSTD_customMem const cMem = cdict->refContext->customMem;
-        ZSTD_freeCCtx(cdict->refContext);
+    {   ZSTD_customMem const cMem = cdict->customMem;
+        ZSTD_free(cdict->workspace, cMem);
         ZSTD_free(cdict->dictBuffer, cMem);
         ZSTD_free(cdict, cMem);
         return 0;
     }
 }
 
-static ZSTD_parameters ZSTD_getParamsFromCDict(const ZSTD_CDict* cdict) {
-    return ZSTD_getParamsFromCCtx(cdict->refContext);
+/*! ZSTD_initStaticCDict_advanced() :
+ *  Generate a digested dictionary in provided memory area.
+ *  workspace: The memory area to emplace the dictionary into.
+ *             Provided pointer must 8-bytes aligned.
+ *             It must outlive dictionary usage.
+ *  workspaceSize: Use ZSTD_estimateCDictSize()
+ *                 to determine how large workspace must be.
+ *  cParams : use ZSTD_getCParams() to transform a compression level
+ *            into its relevants cParams.
+ * @return : pointer to ZSTD_CDict*, or NULL if error (size too small)
+ *  Note : there is no corresponding "free" function.
+ *         Since workspace was allocated externally, it must be freed externally.
+ */
+const ZSTD_CDict* ZSTD_initStaticCDict(
+                                 void* workspace, size_t workspaceSize,
+                           const void* dict, size_t dictSize,
+                                 ZSTD_dictLoadMethod_e dictLoadMethod,
+                                 ZSTD_dictContentType_e dictContentType,
+                                 ZSTD_compressionParameters cParams)
+{
+    size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0);
+    size_t const neededSize = sizeof(ZSTD_CDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize)
+                            + HUF_WORKSPACE_SIZE + matchStateSize;
+    ZSTD_CDict* const cdict = (ZSTD_CDict*) workspace;
+    void* ptr;
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+    DEBUGLOG(4, "(workspaceSize < neededSize) : (%u < %u) => %u",
+        (U32)workspaceSize, (U32)neededSize, (U32)(workspaceSize < neededSize));
+    if (workspaceSize < neededSize) return NULL;
+
+    if (dictLoadMethod == ZSTD_dlm_byCopy) {
+        memcpy(cdict+1, dict, dictSize);
+        dict = cdict+1;
+        ptr = (char*)workspace + sizeof(ZSTD_CDict) + dictSize;
+    } else {
+        ptr = cdict+1;
+    }
+    cdict->workspace = ptr;
+    cdict->workspaceSize = HUF_WORKSPACE_SIZE + matchStateSize;
+
+    if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
+                                              dict, dictSize,
+                                              ZSTD_dlm_byRef, dictContentType,
+                                              cParams) ))
+        return NULL;
+
+    return cdict;
 }
 
-size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict, unsigned long long pledgedSrcSize)
+ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict)
+{
+    assert(cdict != NULL);
+    return cdict->cParams;
+}
+
+/* ZSTD_compressBegin_usingCDict_advanced() :
+ * cdict must be != NULL */
+size_t ZSTD_compressBegin_usingCDict_advanced(
+    ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict,
+    ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize)
 {
-    if (cdict->dictContentSize) CHECK_F(ZSTD_copyCCtx(cctx, cdict->refContext, pledgedSrcSize))
-    else CHECK_F(ZSTD_compressBegin_advanced(cctx, NULL, 0, cdict->refContext->params, pledgedSrcSize));
-    return 0;
+    DEBUGLOG(4, "ZSTD_compressBegin_usingCDict_advanced");
+    if (cdict==NULL) return ERROR(dictionary_wrong);
+    {   ZSTD_CCtx_params params = cctx->requestedParams;
+        params.cParams = ZSTD_getCParamsFromCDict(cdict);
+        /* Increase window log to fit the entire dictionary and source if the
+         * source size is known. Limit the increase to 19, which is the
+         * window log for compression level 1 with the largest source size.
+         */
+        if (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+            U32 const limitedSrcSize = (U32)MIN(pledgedSrcSize, 1U << 19);
+            U32 const limitedSrcLog = limitedSrcSize > 1 ? ZSTD_highbit32(limitedSrcSize - 1) + 1 : 1;
+            params.cParams.windowLog = MAX(params.cParams.windowLog, limitedSrcLog);
+        }
+        params.fParams = fParams;
+        return ZSTD_compressBegin_internal(cctx,
+                                           NULL, 0, ZSTD_dct_auto,
+                                           cdict,
+                                           params, pledgedSrcSize,
+                                           ZSTDb_not_buffered);
+    }
+}
+
+/* ZSTD_compressBegin_usingCDict() :
+ * pledgedSrcSize=0 means "unknown"
+ * if pledgedSrcSize>0, it will enable contentSizeFlag */
+size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 0 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    DEBUGLOG(4, "ZSTD_compressBegin_usingCDict : dictIDFlag == %u", !fParams.noDictIDFlag);
+    return ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, 0);
+}
+
+size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
+{
+    CHECK_F (ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, srcSize));   /* will check if cdict != NULL */
+    return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
 }
 
 /*! ZSTD_compress_usingCDict() :
-*   Compression using a digested Dictionary.
-*   Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
-*   Note that compression level is decided during dictionary creation */
+ *  Compression using a digested Dictionary.
+ *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
+ *  Note that compression parameters are decided at CDict creation time
+ *  while frame parameters are hardcoded */
 size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
                                 void* dst, size_t dstCapacity,
                                 const void* src, size_t srcSize,
                                 const ZSTD_CDict* cdict)
 {
-    CHECK_F(ZSTD_compressBegin_usingCDict(cctx, cdict, srcSize));
-
-    if (cdict->refContext->params.fParams.contentSizeFlag==1) {
-        cctx->params.fParams.contentSizeFlag = 1;
-        cctx->frameContentSize = srcSize;
-    }
-
-    return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
+    ZSTD_frameParameters const fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, fParams);
 }
 
 
@@ -2874,243 +2845,329 @@
 *  Streaming
 ********************************************************************/
 
-typedef enum { zcss_init, zcss_load, zcss_flush, zcss_final } ZSTD_cStreamStage;
-
-struct ZSTD_CStream_s {
-    ZSTD_CCtx* cctx;
-    ZSTD_CDict* cdictLocal;
-    const ZSTD_CDict* cdict;
-    char*  inBuff;
-    size_t inBuffSize;
-    size_t inToCompress;
-    size_t inBuffPos;
-    size_t inBuffTarget;
-    size_t blockSize;
-    char*  outBuff;
-    size_t outBuffSize;
-    size_t outBuffContentSize;
-    size_t outBuffFlushedSize;
-    ZSTD_cStreamStage stage;
-    U32    checksum;
-    U32    frameEnded;
-    U64    pledgedSrcSize;
-    U64    inputProcessed;
-    ZSTD_parameters params;
-    ZSTD_customMem customMem;
-};   /* typedef'd to ZSTD_CStream within "zstd.h" */
-
 ZSTD_CStream* ZSTD_createCStream(void)
 {
-    return ZSTD_createCStream_advanced(defaultCustomMem);
+    DEBUGLOG(3, "ZSTD_createCStream");
+    return ZSTD_createCStream_advanced(ZSTD_defaultCMem);
+}
+
+ZSTD_CStream* ZSTD_initStaticCStream(void *workspace, size_t workspaceSize)
+{
+    return ZSTD_initStaticCCtx(workspace, workspaceSize);
 }
 
 ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem)
-{
-    ZSTD_CStream* zcs;
-
-    if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
-    if (!customMem.customAlloc || !customMem.customFree) return NULL;
-
-    zcs = (ZSTD_CStream*)ZSTD_malloc(sizeof(ZSTD_CStream), customMem);
-    if (zcs==NULL) return NULL;
-    memset(zcs, 0, sizeof(ZSTD_CStream));
-    memcpy(&zcs->customMem, &customMem, sizeof(ZSTD_customMem));
-    zcs->cctx = ZSTD_createCCtx_advanced(customMem);
-    if (zcs->cctx == NULL) { ZSTD_freeCStream(zcs); return NULL; }
-    return zcs;
+{   /* CStream and CCtx are now same object */
+    return ZSTD_createCCtx_advanced(customMem);
 }
 
 size_t ZSTD_freeCStream(ZSTD_CStream* zcs)
 {
-    if (zcs==NULL) return 0;   /* support free on NULL */
-    {   ZSTD_customMem const cMem = zcs->customMem;
-        ZSTD_freeCCtx(zcs->cctx);
-        ZSTD_freeCDict(zcs->cdictLocal);
-        ZSTD_free(zcs->inBuff, cMem);
-        ZSTD_free(zcs->outBuff, cMem);
-        ZSTD_free(zcs, cMem);
-        return 0;
-    }
+    return ZSTD_freeCCtx(zcs);   /* same object */
 }
 
 
+
 /*======   Initialization   ======*/
 
-size_t ZSTD_CStreamInSize(void)  { return ZSTD_BLOCKSIZE_ABSOLUTEMAX; }
-size_t ZSTD_CStreamOutSize(void) { return ZSTD_compressBound(ZSTD_BLOCKSIZE_ABSOLUTEMAX) + ZSTD_blockHeaderSize + 4 /* 32-bits hash */ ; }
-
-size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize)
+size_t ZSTD_CStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX; }
+
+size_t ZSTD_CStreamOutSize(void)
+{
+    return ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + 4 /* 32-bits hash */ ;
+}
+
+static size_t ZSTD_resetCStream_internal(ZSTD_CStream* cctx,
+                    const void* const dict, size_t const dictSize, ZSTD_dictContentType_e const dictContentType,
+                    const ZSTD_CDict* const cdict,
+                    ZSTD_CCtx_params const params, unsigned long long const pledgedSrcSize)
 {
-    if (zcs->inBuffSize==0) return ERROR(stage_wrong);   /* zcs has not been init at least once => can't reset */
-
-    if (zcs->cdict) CHECK_F(ZSTD_compressBegin_usingCDict(zcs->cctx, zcs->cdict, pledgedSrcSize))
-    else CHECK_F(ZSTD_compressBegin_advanced(zcs->cctx, NULL, 0, zcs->params, pledgedSrcSize));
-
-    zcs->inToCompress = 0;
-    zcs->inBuffPos = 0;
-    zcs->inBuffTarget = zcs->blockSize;
-    zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
-    zcs->stage = zcss_load;
-    zcs->frameEnded = 0;
-    zcs->pledgedSrcSize = pledgedSrcSize;
-    zcs->inputProcessed = 0;
+    DEBUGLOG(4, "ZSTD_resetCStream_internal (disableLiteralCompression=%i)",
+                params.disableLiteralCompression);
+    /* params are supposed to be fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+
+    CHECK_F( ZSTD_compressBegin_internal(cctx,
+                                         dict, dictSize, dictContentType,
+                                         cdict,
+                                         params, pledgedSrcSize,
+                                         ZSTDb_buffered) );
+
+    cctx->inToCompress = 0;
+    cctx->inBuffPos = 0;
+    cctx->inBuffTarget = cctx->blockSize
+                      + (cctx->blockSize == pledgedSrcSize);   /* for small input: avoid automatic flush on reaching end of block, since it would require to add a 3-bytes null block to end frame */
+    cctx->outBuffContentSize = cctx->outBuffFlushedSize = 0;
+    cctx->streamStage = zcss_load;
+    cctx->frameEnded = 0;
     return 0;   /* ready to go */
 }
 
-size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
-                                 const void* dict, size_t dictSize,
-                                 ZSTD_parameters params, unsigned long long pledgedSrcSize)
+/* ZSTD_resetCStream():
+ * pledgedSrcSize == 0 means "unknown" */
+size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize)
 {
-    /* allocate buffers */
-    {   size_t const neededInBuffSize = (size_t)1 << params.cParams.windowLog;
-        if (zcs->inBuffSize < neededInBuffSize) {
-            zcs->inBuffSize = neededInBuffSize;
-            ZSTD_free(zcs->inBuff, zcs->customMem);
-            zcs->inBuff = (char*) ZSTD_malloc(neededInBuffSize, zcs->customMem);
-            if (zcs->inBuff == NULL) return ERROR(memory_allocation);
-        }
-        zcs->blockSize = MIN(ZSTD_BLOCKSIZE_ABSOLUTEMAX, neededInBuffSize);
-    }
-    if (zcs->outBuffSize < ZSTD_compressBound(zcs->blockSize)+1) {
-        zcs->outBuffSize = ZSTD_compressBound(zcs->blockSize)+1;
-        ZSTD_free(zcs->outBuff, zcs->customMem);
-        zcs->outBuff = (char*) ZSTD_malloc(zcs->outBuffSize, zcs->customMem);
-        if (zcs->outBuff == NULL) return ERROR(memory_allocation);
-    }
+    ZSTD_CCtx_params params = zcs->requestedParams;
+    DEBUGLOG(4, "ZSTD_resetCStream: pledgedSrcSize = %u", (U32)pledgedSrcSize);
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    params.fParams.contentSizeFlag = 1;
+    params.cParams = ZSTD_getCParamsFromCCtxParams(&params, pledgedSrcSize, 0);
+    return ZSTD_resetCStream_internal(zcs, NULL, 0, ZSTD_dct_auto, zcs->cdict, params, pledgedSrcSize);
+}
+
+/*! ZSTD_initCStream_internal() :
+ *  Note : for lib/compress only. Used by zstdmt_compress.c.
+ *  Assumption 1 : params are valid
+ *  Assumption 2 : either dict, or cdict, is defined, not both */
+size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
+                    const void* dict, size_t dictSize, const ZSTD_CDict* cdict,
+                    ZSTD_CCtx_params params, unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_internal");
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
 
     if (dict && dictSize >= 8) {
+        DEBUGLOG(4, "loading dictionary of size %u", (U32)dictSize);
+        if (zcs->staticSize) {   /* static CCtx : never uses malloc */
+            /* incompatible with internal cdict creation */
+            return ERROR(memory_allocation);
+        }
         ZSTD_freeCDict(zcs->cdictLocal);
-        zcs->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize, 0, params, zcs->customMem);
-        if (zcs->cdictLocal == NULL) return ERROR(memory_allocation);
+        zcs->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
+                                            ZSTD_dlm_byCopy, ZSTD_dct_auto,
+                                            params.cParams, zcs->customMem);
         zcs->cdict = zcs->cdictLocal;
-    } else zcs->cdict = NULL;
-
-    zcs->checksum = params.fParams.checksumFlag > 0;
-    zcs->params = params;
-
-    return ZSTD_resetCStream(zcs, pledgedSrcSize);
+        if (zcs->cdictLocal == NULL) return ERROR(memory_allocation);
+    } else {
+        if (cdict) {
+            params.cParams = ZSTD_getCParamsFromCDict(cdict);  /* cParams are enforced from cdict; it includes windowLog */
+        }
+        ZSTD_freeCDict(zcs->cdictLocal);
+        zcs->cdictLocal = NULL;
+        zcs->cdict = cdict;
+    }
+
+    return ZSTD_resetCStream_internal(zcs, NULL, 0, ZSTD_dct_auto, zcs->cdict, params, pledgedSrcSize);
+}
+
+/* ZSTD_initCStream_usingCDict_advanced() :
+ * same as ZSTD_initCStream_usingCDict(), with control over frame parameters */
+size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
+                                            const ZSTD_CDict* cdict,
+                                            ZSTD_frameParameters fParams,
+                                            unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict_advanced");
+    if (!cdict) return ERROR(dictionary_wrong); /* cannot handle NULL cdict (does not know what to do) */
+    {   ZSTD_CCtx_params params = zcs->requestedParams;
+        params.cParams = ZSTD_getCParamsFromCDict(cdict);
+        params.fParams = fParams;
+        return ZSTD_initCStream_internal(zcs,
+                                NULL, 0, cdict,
+                                params, pledgedSrcSize);
+    }
 }
 
 /* note : cdict must outlive compression session */
 size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict)
 {
-    ZSTD_parameters const params = ZSTD_getParamsFromCDict(cdict);
-    size_t const initError =  ZSTD_initCStream_advanced(zcs, NULL, 0, params, 0);
-    zcs->cdict = cdict;
-    zcs->cctx->dictID = params.fParams.noDictIDFlag ? 0 : cdict->refContext->dictID;
-    return initError;
+    ZSTD_frameParameters const fParams = { 0 /* contentSizeFlag */, 0 /* checksum */, 0 /* hideDictID */ };
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict");
+    return ZSTD_initCStream_usingCDict_advanced(zcs, cdict, fParams, ZSTD_CONTENTSIZE_UNKNOWN);  /* note : will check that cdict != NULL */
+}
+
+
+/* ZSTD_initCStream_advanced() :
+ * pledgedSrcSize must be exact.
+ * if srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN.
+ * dict is loaded with default parameters ZSTD_dm_auto and ZSTD_dlm_byCopy. */
+size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
+                                 const void* dict, size_t dictSize,
+                                 ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_advanced: pledgedSrcSize=%u, flag=%u",
+                (U32)pledgedSrcSize, params.fParams.contentSizeFlag);
+    CHECK_F( ZSTD_checkCParams(params.cParams) );
+    if ((pledgedSrcSize==0) && (params.fParams.contentSizeFlag==0)) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;  /* for compatibility with older programs relying on this behavior. Users should now specify ZSTD_CONTENTSIZE_UNKNOWN. This line will be removed in the future. */
+    {   ZSTD_CCtx_params const cctxParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
+        return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL /*cdict*/, cctxParams, pledgedSrcSize);
+    }
 }
 
 size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel)
 {
     ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
-    return ZSTD_initCStream_advanced(zcs, dict, dictSize, params, 0);
+    ZSTD_CCtx_params const cctxParams =
+            ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
+    return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL, cctxParams, ZSTD_CONTENTSIZE_UNKNOWN);
 }
 
-size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pledgedSrcSize)
+size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pss)
 {
-    ZSTD_parameters params = ZSTD_getParams(compressionLevel, pledgedSrcSize, 0);
-    if (pledgedSrcSize) params.fParams.contentSizeFlag = 1;
-    return ZSTD_initCStream_advanced(zcs, NULL, 0, params, pledgedSrcSize);
+    U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;  /* temporary : 0 interpreted as "unknown" during transition period. Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN. `0` will be interpreted as "empty" in the future */
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, pledgedSrcSize, 0);
+    ZSTD_CCtx_params const cctxParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
+    return ZSTD_initCStream_internal(zcs, NULL, 0, NULL, cctxParams, pledgedSrcSize);
 }
 
 size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
 {
-    return ZSTD_initCStream_usingDict(zcs, NULL, 0, compressionLevel);
-}
-
-size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
-{
-    if (zcs==NULL) return 0;   /* support sizeof on NULL */
-    return sizeof(zcs) + ZSTD_sizeof_CCtx(zcs->cctx) + ZSTD_sizeof_CDict(zcs->cdictLocal) + zcs->outBuffSize + zcs->inBuffSize;
+    DEBUGLOG(4, "ZSTD_initCStream");
+    return ZSTD_initCStream_srcSize(zcs, compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN);
 }
 
 /*======   Compression   ======*/
 
-typedef enum { zsf_gather, zsf_flush, zsf_end } ZSTD_flush_e;
-
-MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity,
+                           const void* src, size_t srcSize)
 {
     size_t const length = MIN(dstCapacity, srcSize);
-    memcpy(dst, src, length);
+    if (length) memcpy(dst, src, length);
     return length;
 }
 
-static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
-                              void* dst, size_t* dstCapacityPtr,
-                        const void* src, size_t* srcSizePtr,
-                              ZSTD_flush_e const flush)
+/** ZSTD_compressStream_generic():
+ *  internal function for all *compressStream*() variants and *compress_generic()
+ *  non-static, because can be called from zstdmt_compress.c
+ * @return : hint size for next input */
+size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
+                                   ZSTD_outBuffer* output,
+                                   ZSTD_inBuffer* input,
+                                   ZSTD_EndDirective const flushMode)
 {
+    const char* const istart = (const char*)input->src;
+    const char* const iend = istart + input->size;
+    const char* ip = istart + input->pos;
+    char* const ostart = (char*)output->dst;
+    char* const oend = ostart + output->size;
+    char* op = ostart + output->pos;
     U32 someMoreWork = 1;
-    const char* const istart = (const char*)src;
-    const char* const iend = istart + *srcSizePtr;
-    const char* ip = istart;
-    char* const ostart = (char*)dst;
-    char* const oend = ostart + *dstCapacityPtr;
-    char* op = ostart;
+
+    /* check expectations */
+    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%u", (U32)flushMode);
+    assert(zcs->inBuff != NULL);
+    assert(zcs->inBuffSize > 0);
+    assert(zcs->outBuff !=  NULL);
+    assert(zcs->outBuffSize > 0);
+    assert(output->pos <= output->size);
+    assert(input->pos <= input->size);
 
     while (someMoreWork) {
-        switch(zcs->stage)
+        switch(zcs->streamStage)
         {
-        case zcss_init: return ERROR(init_missing);   /* call ZBUFF_compressInit() first ! */
+        case zcss_init:
+            /* call ZSTD_initCStream() first ! */
+            return ERROR(init_missing);
 
         case zcss_load:
-            /* complete inBuffer */
+            if ( (flushMode == ZSTD_e_end)
+              && ((size_t)(oend-op) >= ZSTD_compressBound(iend-ip))  /* enough dstCapacity */
+              && (zcs->inBuffPos == 0) ) {
+                /* shortcut to compression pass directly into output buffer */
+                size_t const cSize = ZSTD_compressEnd(zcs,
+                                                op, oend-op, ip, iend-ip);
+                DEBUGLOG(4, "ZSTD_compressEnd : %u", (U32)cSize);
+                if (ZSTD_isError(cSize)) return cSize;
+                ip = iend;
+                op += cSize;
+                zcs->frameEnded = 1;
+                ZSTD_startNewCompression(zcs);
+                someMoreWork = 0; break;
+            }
+            /* complete loading into inBuffer */
             {   size_t const toLoad = zcs->inBuffTarget - zcs->inBuffPos;
-                size_t const loaded = ZSTD_limitCopy(zcs->inBuff + zcs->inBuffPos, toLoad, ip, iend-ip);
+                size_t const loaded = ZSTD_limitCopy(
+                                        zcs->inBuff + zcs->inBuffPos, toLoad,
+                                        ip, iend-ip);
                 zcs->inBuffPos += loaded;
                 ip += loaded;
-                if ( (zcs->inBuffPos==zcs->inToCompress) || (!flush && (toLoad != loaded)) ) {
-                    someMoreWork = 0; break;  /* not enough input to get a full block : stop there, wait for more */
-            }   }
+                if ( (flushMode == ZSTD_e_continue)
+                  && (zcs->inBuffPos < zcs->inBuffTarget) ) {
+                    /* not enough input to fill full block : stop here */
+                    someMoreWork = 0; break;
+                }
+                if ( (flushMode == ZSTD_e_flush)
+                  && (zcs->inBuffPos == zcs->inToCompress) ) {
+                    /* empty */
+                    someMoreWork = 0; break;
+                }
+            }
             /* compress current block (note : this stage cannot be stopped in the middle) */
+            DEBUGLOG(5, "stream compression stage (flushMode==%u)", flushMode);
             {   void* cDst;
                 size_t cSize;
                 size_t const iSize = zcs->inBuffPos - zcs->inToCompress;
                 size_t oSize = oend-op;
+                unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip==iend);
                 if (oSize >= ZSTD_compressBound(iSize))
-                    cDst = op;   /* compress directly into output buffer (avoid flush stage) */
+                    cDst = op;   /* compress into output buffer, to skip flush stage */
                 else
                     cDst = zcs->outBuff, oSize = zcs->outBuffSize;
-                cSize = (flush == zsf_end) ?
-                        ZSTD_compressEnd(zcs->cctx, cDst, oSize, zcs->inBuff + zcs->inToCompress, iSize) :
-                        ZSTD_compressContinue(zcs->cctx, cDst, oSize, zcs->inBuff + zcs->inToCompress, iSize);
+                cSize = lastBlock ?
+                        ZSTD_compressEnd(zcs, cDst, oSize,
+                                    zcs->inBuff + zcs->inToCompress, iSize) :
+                        ZSTD_compressContinue(zcs, cDst, oSize,
+                                    zcs->inBuff + zcs->inToCompress, iSize);
                 if (ZSTD_isError(cSize)) return cSize;
-                if (flush == zsf_end) zcs->frameEnded = 1;
+                zcs->frameEnded = lastBlock;
                 /* prepare next block */
                 zcs->inBuffTarget = zcs->inBuffPos + zcs->blockSize;
                 if (zcs->inBuffTarget > zcs->inBuffSize)
-                    zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;   /* note : inBuffSize >= blockSize */
+                    zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;
+                DEBUGLOG(5, "inBuffTarget:%u / inBuffSize:%u",
+                         (U32)zcs->inBuffTarget, (U32)zcs->inBuffSize);
+                if (!lastBlock)
+                    assert(zcs->inBuffTarget <= zcs->inBuffSize);
                 zcs->inToCompress = zcs->inBuffPos;
-                if (cDst == op) { op += cSize; break; }   /* no need to flush */
+                if (cDst == op) {  /* no need to flush */
+                    op += cSize;
+                    if (zcs->frameEnded) {
+                        DEBUGLOG(5, "Frame completed directly in outBuffer");
+                        someMoreWork = 0;
+                        ZSTD_startNewCompression(zcs);
+                    }
+                    break;
+                }
                 zcs->outBuffContentSize = cSize;
                 zcs->outBuffFlushedSize = 0;
-                zcs->stage = zcss_flush;   /* pass-through to flush stage */
+                zcs->streamStage = zcss_flush; /* pass-through to flush stage */
             }
-
+	    /* fall-through */
         case zcss_flush:
+            DEBUGLOG(5, "flush stage");
             {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
-                size_t const flushed = ZSTD_limitCopy(op, oend-op, zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
+                size_t const flushed = ZSTD_limitCopy(op, oend-op,
+                            zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
+                DEBUGLOG(5, "toFlush: %u into %u ==> flushed: %u",
+                            (U32)toFlush, (U32)(oend-op), (U32)flushed);
                 op += flushed;
                 zcs->outBuffFlushedSize += flushed;
-                if (toFlush!=flushed) { someMoreWork = 0; break; }  /* dst too small to store flushed data : stop there */
+                if (toFlush!=flushed) {
+                    /* flush not fully completed, presumably because dst is too small */
+                    assert(op==oend);
+                    someMoreWork = 0;
+                    break;
+                }
                 zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
-                zcs->stage = zcss_load;
+                if (zcs->frameEnded) {
+                    DEBUGLOG(5, "Frame completed on flush");
+                    someMoreWork = 0;
+                    ZSTD_startNewCompression(zcs);
+                    break;
+                }
+                zcs->streamStage = zcss_load;
                 break;
             }
 
-        case zcss_final:
-            someMoreWork = 0;   /* do nothing */
-            break;
-
-        default:
-            return ERROR(GENERIC);   /* impossible */
+        default: /* impossible */
+            assert(0);
         }
     }
 
-    *srcSizePtr = ip - istart;
-    *dstCapacityPtr = op - ostart;
-    zcs->inputProcessed += *srcSizePtr;
+    input->pos = ip - istart;
+    output->pos = op - ostart;
     if (zcs->frameEnded) return 0;
     {   size_t hintInSize = zcs->inBuffTarget - zcs->inBuffPos;
         if (hintInSize==0) hintInSize = zcs->blockSize;
@@ -3120,111 +3177,169 @@
 
 size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
 {
-    size_t sizeRead = input->size - input->pos;
-    size_t sizeWritten = output->size - output->pos;
-    size_t const result = ZSTD_compressStream_generic(zcs,
-                                                      (char*)(output->dst) + output->pos, &sizeWritten,
-                                                      (const char*)(input->src) + input->pos, &sizeRead, zsf_gather);
-    input->pos += sizeRead;
-    output->pos += sizeWritten;
-    return result;
+    /* check conditions */
+    if (output->pos > output->size) return ERROR(GENERIC);
+    if (input->pos  > input->size)  return ERROR(GENERIC);
+
+    return ZSTD_compressStream_generic(zcs, output, input, ZSTD_e_continue);
+}
+
+
+size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
+                              ZSTD_outBuffer* output,
+                              ZSTD_inBuffer* input,
+                              ZSTD_EndDirective endOp)
+{
+    DEBUGLOG(5, "ZSTD_compress_generic, endOp=%u ", (U32)endOp);
+    /* check conditions */
+    if (output->pos > output->size) return ERROR(GENERIC);
+    if (input->pos  > input->size)  return ERROR(GENERIC);
+    assert(cctx!=NULL);
+
+    /* transparent initialization stage */
+    if (cctx->streamStage == zcss_init) {
+        ZSTD_CCtx_params params = cctx->requestedParams;
+        ZSTD_prefixDict const prefixDict = cctx->prefixDict;
+        memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));  /* single usage */
+        assert(prefixDict.dict==NULL || cctx->cdict==NULL);   /* only one can be set */
+        DEBUGLOG(4, "ZSTD_compress_generic : transparent init stage");
+        if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = input->size + 1;  /* auto-fix pledgedSrcSize */
+        params.cParams = ZSTD_getCParamsFromCCtxParams(
+                &cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1, 0 /*dictSize*/);
+
+#ifdef ZSTD_MULTITHREAD
+        if ((cctx->pledgedSrcSizePlusOne-1) <= ZSTDMT_JOBSIZE_MIN) {
+            params.nbWorkers = 0; /* do not invoke multi-threading when src size is too small */
+        }
+        if (params.nbWorkers > 0) {
+            /* mt context creation */
+            if (cctx->mtctx == NULL || (params.nbWorkers != ZSTDMT_getNbWorkers(cctx->mtctx))) {
+                DEBUGLOG(4, "ZSTD_compress_generic: creating new mtctx for nbWorkers=%u",
+                            params.nbWorkers);
+                if (cctx->mtctx != NULL)
+                    DEBUGLOG(4, "ZSTD_compress_generic: previous nbWorkers was %u",
+                                ZSTDMT_getNbWorkers(cctx->mtctx));
+                ZSTDMT_freeCCtx(cctx->mtctx);
+                cctx->mtctx = ZSTDMT_createCCtx_advanced(params.nbWorkers, cctx->customMem);
+                if (cctx->mtctx == NULL) return ERROR(memory_allocation);
+            }
+            /* mt compression */
+            DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbWorkers=%u", params.nbWorkers);
+            CHECK_F( ZSTDMT_initCStream_internal(
+                        cctx->mtctx,
+                        prefixDict.dict, prefixDict.dictSize, ZSTD_dct_rawContent,
+                        cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) );
+            cctx->streamStage = zcss_load;
+            cctx->appliedParams.nbWorkers = params.nbWorkers;
+        } else
+#endif
+        {   CHECK_F( ZSTD_resetCStream_internal(cctx,
+                            prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType,
+                            cctx->cdict,
+                            params, cctx->pledgedSrcSizePlusOne-1) );
+            assert(cctx->streamStage == zcss_load);
+            assert(cctx->appliedParams.nbWorkers == 0);
+    }   }
+
+    /* compression stage */
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        if (cctx->cParamsChanged) {
+            ZSTDMT_updateCParams_whileCompressing(cctx->mtctx, &cctx->requestedParams);
+            cctx->cParamsChanged = 0;
+        }
+        {   size_t const flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
+            if ( ZSTD_isError(flushMin)
+              || (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
+                ZSTD_startNewCompression(cctx);
+            }
+            return flushMin;
+    }   }
+#endif
+    CHECK_F( ZSTD_compressStream_generic(cctx, output, input, endOp) );
+    DEBUGLOG(5, "completed ZSTD_compress_generic");
+    return cctx->outBuffContentSize - cctx->outBuffFlushedSize; /* remaining to flush */
+}
+
+size_t ZSTD_compress_generic_simpleArgs (
+                            ZSTD_CCtx* cctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos,
+                            ZSTD_EndDirective endOp)
+{
+    ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
+    ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
+    /* ZSTD_compress_generic() will check validity of dstPos and srcPos */
+    size_t const cErr = ZSTD_compress_generic(cctx, &output, &input, endOp);
+    *dstPos = output.pos;
+    *srcPos = input.pos;
+    return cErr;
 }
 
 
 /*======   Finalize   ======*/
 
 /*! ZSTD_flushStream() :
-*   @return : amount of data remaining to flush */
+ * @return : amount of data remaining to flush */
 size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 {
-    size_t srcSize = 0;
-    size_t sizeWritten = output->size - output->pos;
-    size_t const result = ZSTD_compressStream_generic(zcs,
-                                                     (char*)(output->dst) + output->pos, &sizeWritten,
-                                                     &srcSize, &srcSize, /* use a valid src address instead of NULL */
-                                                      zsf_flush);
-    output->pos += sizeWritten;
-    if (ZSTD_isError(result)) return result;
-    return zcs->outBuffContentSize - zcs->outBuffFlushedSize;   /* remaining to flush */
+    ZSTD_inBuffer input = { NULL, 0, 0 };
+    if (output->pos > output->size) return ERROR(GENERIC);
+    CHECK_F( ZSTD_compressStream_generic(zcs, output, &input, ZSTD_e_flush) );
+    return zcs->outBuffContentSize - zcs->outBuffFlushedSize;  /* remaining to flush */
 }
 
 
 size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 {
-    BYTE* const ostart = (BYTE*)(output->dst) + output->pos;
-    BYTE* const oend = (BYTE*)(output->dst) + output->size;
-    BYTE* op = ostart;
-
-    if ((zcs->pledgedSrcSize) && (zcs->inputProcessed != zcs->pledgedSrcSize))
-        return ERROR(srcSize_wrong);   /* pledgedSrcSize not respected */
-
-    if (zcs->stage != zcss_final) {
-        /* flush whatever remains */
-        size_t srcSize = 0;
-        size_t sizeWritten = output->size - output->pos;
-        size_t const notEnded = ZSTD_compressStream_generic(zcs, ostart, &sizeWritten, &srcSize, &srcSize, zsf_end);  /* use a valid src address instead of NULL */
-        size_t const remainingToFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
-        op += sizeWritten;
-        if (remainingToFlush) {
-            output->pos += sizeWritten;
-            return remainingToFlush + ZSTD_BLOCKHEADERSIZE /* final empty block */ + (zcs->checksum * 4);
-        }
-        /* create epilogue */
-        zcs->stage = zcss_final;
-        zcs->outBuffContentSize = !notEnded ? 0 :
-            ZSTD_compressEnd(zcs->cctx, zcs->outBuff, zcs->outBuffSize, NULL, 0);  /* write epilogue, including final empty block, into outBuff */
-    }
-
-    /* flush epilogue */
-    {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
-        size_t const flushed = ZSTD_limitCopy(op, oend-op, zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
-        op += flushed;
-        zcs->outBuffFlushedSize += flushed;
-        output->pos += op-ostart;
-        if (toFlush==flushed) zcs->stage = zcss_init;  /* end reached */
-        return toFlush - flushed;
+    ZSTD_inBuffer input = { NULL, 0, 0 };
+    if (output->pos > output->size) return ERROR(GENERIC);
+    CHECK_F( ZSTD_compressStream_generic(zcs, output, &input, ZSTD_e_end) );
+    {   size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
+        size_t const checksumSize = zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4;
+        size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize + lastBlockSize + checksumSize;
+        DEBUGLOG(4, "ZSTD_endStream : remaining to flush : %u", (U32)toFlush);
+        return toFlush;
     }
 }
 
 
-
 /*-=====  Pre-defined compression levels  =====-*/
 
-#define ZSTD_DEFAULT_CLEVEL 1
 #define ZSTD_MAX_CLEVEL     22
 int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
 
 static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
-{   /* "default" */
+{   /* "default" - guarantees a monotonically increasing memory budget */
     /* W,  C,  H,  S,  L, TL, strat */
-    { 18, 12, 12,  1,  7, 16, ZSTD_fast    },  /* level  0 - never used */
-    { 19, 13, 14,  1,  7, 16, ZSTD_fast    },  /* level  1 */
-    { 19, 15, 16,  1,  6, 16, ZSTD_fast    },  /* level  2 */
-    { 20, 16, 17,  1,  5, 16, ZSTD_dfast   },  /* level  3.*/
-    { 20, 18, 18,  1,  5, 16, ZSTD_dfast   },  /* level  4.*/
-    { 20, 15, 18,  3,  5, 16, ZSTD_greedy  },  /* level  5 */
-    { 21, 16, 19,  2,  5, 16, ZSTD_lazy    },  /* level  6 */
-    { 21, 17, 20,  3,  5, 16, ZSTD_lazy    },  /* level  7 */
+    { 19, 12, 13,  1,  6,  1, ZSTD_fast    },  /* base for negative levels */
+    { 19, 13, 14,  1,  7,  1, ZSTD_fast    },  /* level  1 */
+    { 19, 15, 16,  1,  6,  1, ZSTD_fast    },  /* level  2 */
+    { 20, 16, 17,  1,  5,  8, ZSTD_dfast   },  /* level  3 */
+    { 20, 17, 18,  1,  5,  8, ZSTD_dfast   },  /* level  4 */
+    { 20, 17, 18,  2,  5, 16, ZSTD_greedy  },  /* level  5 */
+    { 21, 17, 19,  2,  5, 16, ZSTD_lazy    },  /* level  6 */
+    { 21, 18, 19,  3,  5, 16, ZSTD_lazy    },  /* level  7 */
     { 21, 18, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  8 */
-    { 21, 20, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  9 */
+    { 21, 19, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  9 */
     { 21, 19, 21,  4,  5, 16, ZSTD_lazy2   },  /* level 10 */
     { 22, 20, 22,  4,  5, 16, ZSTD_lazy2   },  /* level 11 */
     { 22, 20, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 12 */
-    { 22, 21, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 13 */
-    { 22, 21, 22,  6,  5, 16, ZSTD_lazy2   },  /* level 14 */
-    { 22, 21, 21,  5,  5, 16, ZSTD_btlazy2 },  /* level 15 */
-    { 23, 22, 22,  5,  5, 16, ZSTD_btlazy2 },  /* level 16 */
-    { 23, 21, 22,  4,  5, 24, ZSTD_btopt   },  /* level 17 */
-    { 23, 23, 22,  6,  5, 32, ZSTD_btopt   },  /* level 18 */
-    { 23, 23, 22,  6,  3, 48, ZSTD_btopt   },  /* level 19 */
-    { 25, 25, 23,  7,  3, 64, ZSTD_btopt2  },  /* level 20 */
-    { 26, 26, 23,  7,  3,256, ZSTD_btopt2  },  /* level 21 */
-    { 27, 27, 25,  9,  3,512, ZSTD_btopt2  },  /* level 22 */
+    { 22, 21, 22,  4,  5, 32, ZSTD_btlazy2 },  /* level 13 */
+    { 22, 21, 22,  5,  5, 32, ZSTD_btlazy2 },  /* level 14 */
+    { 22, 22, 22,  6,  5, 32, ZSTD_btlazy2 },  /* level 15 */
+    { 22, 21, 22,  4,  5, 48, ZSTD_btopt   },  /* level 16 */
+    { 23, 22, 22,  4,  4, 48, ZSTD_btopt   },  /* level 17 */
+    { 23, 22, 22,  5,  3, 64, ZSTD_btopt   },  /* level 18 */
+    { 23, 23, 22,  7,  3,128, ZSTD_btopt   },  /* level 19 */
+    { 25, 25, 23,  7,  3,128, ZSTD_btultra },  /* level 20 */
+    { 26, 26, 24,  7,  3,256, ZSTD_btultra },  /* level 21 */
+    { 27, 27, 25,  9,  3,512, ZSTD_btultra },  /* level 22 */
 },
 {   /* for srcSize <= 256 KB */
     /* W,  C,  H,  S,  L,  T, strat */
-    {  0,  0,  0,  0,  0,  0, ZSTD_fast    },  /* level  0 - not used */
-    { 18, 13, 14,  1,  6,  8, ZSTD_fast    },  /* level  1 */
+    { 18, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 18, 13, 14,  1,  6,  1, ZSTD_fast    },  /* level  1 */
     { 18, 14, 13,  1,  5,  8, ZSTD_dfast   },  /* level  2 */
     { 18, 16, 15,  1,  5,  8, ZSTD_dfast   },  /* level  3 */
     { 18, 15, 17,  1,  5,  8, ZSTD_greedy  },  /* level  4.*/
@@ -3235,23 +3350,23 @@
     { 18, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
     { 18, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
     { 18, 18, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 11.*/
-    { 18, 18, 17,  7,  4,  8, ZSTD_lazy2   },  /* level 12.*/
-    { 18, 19, 17,  6,  4,  8, ZSTD_btlazy2 },  /* level 13 */
+    { 18, 18, 17,  5,  4,  8, ZSTD_btlazy2 },  /* level 12.*/
+    { 18, 19, 17,  7,  4,  8, ZSTD_btlazy2 },  /* level 13 */
     { 18, 18, 18,  4,  4, 16, ZSTD_btopt   },  /* level 14.*/
     { 18, 18, 18,  4,  3, 16, ZSTD_btopt   },  /* level 15.*/
     { 18, 19, 18,  6,  3, 32, ZSTD_btopt   },  /* level 16.*/
     { 18, 19, 18,  8,  3, 64, ZSTD_btopt   },  /* level 17.*/
     { 18, 19, 18,  9,  3,128, ZSTD_btopt   },  /* level 18.*/
     { 18, 19, 18, 10,  3,256, ZSTD_btopt   },  /* level 19.*/
-    { 18, 19, 18, 11,  3,512, ZSTD_btopt2  },  /* level 20.*/
-    { 18, 19, 18, 12,  3,512, ZSTD_btopt2  },  /* level 21.*/
-    { 18, 19, 18, 13,  3,512, ZSTD_btopt2  },  /* level 22.*/
+    { 18, 19, 18, 11,  3,512, ZSTD_btultra },  /* level 20.*/
+    { 18, 19, 18, 12,  3,512, ZSTD_btultra },  /* level 21.*/
+    { 18, 19, 18, 13,  3,512, ZSTD_btultra },  /* level 22.*/
 },
 {   /* for srcSize <= 128 KB */
     /* W,  C,  H,  S,  L,  T, strat */
-    { 17, 12, 12,  1,  7,  8, ZSTD_fast    },  /* level  0 - not used */
-    { 17, 12, 13,  1,  6,  8, ZSTD_fast    },  /* level  1 */
-    { 17, 13, 16,  1,  5,  8, ZSTD_fast    },  /* level  2 */
+    { 17, 12, 12,  1,  5,  1, ZSTD_fast    },  /* level  0 - not used */
+    { 17, 12, 13,  1,  6,  1, ZSTD_fast    },  /* level  1 */
+    { 17, 13, 16,  1,  5,  1, ZSTD_fast    },  /* level  2 */
     { 17, 16, 16,  2,  5,  8, ZSTD_dfast   },  /* level  3 */
     { 17, 13, 15,  3,  4,  8, ZSTD_greedy  },  /* level  4 */
     { 17, 15, 17,  4,  4,  8, ZSTD_greedy  },  /* level  5 */
@@ -3269,15 +3384,15 @@
     { 17, 18, 17,  7,  3, 64, ZSTD_btopt   },  /* level 17.*/
     { 17, 18, 17,  7,  3,256, ZSTD_btopt   },  /* level 18.*/
     { 17, 18, 17,  8,  3,256, ZSTD_btopt   },  /* level 19.*/
-    { 17, 18, 17,  9,  3,256, ZSTD_btopt2  },  /* level 20.*/
-    { 17, 18, 17, 10,  3,256, ZSTD_btopt2  },  /* level 21.*/
-    { 17, 18, 17, 11,  3,512, ZSTD_btopt2  },  /* level 22.*/
+    { 17, 18, 17,  9,  3,256, ZSTD_btultra },  /* level 20.*/
+    { 17, 18, 17, 10,  3,256, ZSTD_btultra },  /* level 21.*/
+    { 17, 18, 17, 11,  3,512, ZSTD_btultra },  /* level 22.*/
 },
 {   /* for srcSize <= 16 KB */
     /* W,  C,  H,  S,  L,  T, strat */
-    { 14, 12, 12,  1,  7,  6, ZSTD_fast    },  /* level  0 - not used */
-    { 14, 14, 14,  1,  6,  6, ZSTD_fast    },  /* level  1 */
-    { 14, 14, 14,  1,  4,  6, ZSTD_fast    },  /* level  2 */
+    { 14, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 14, 14, 14,  1,  6,  1, ZSTD_fast    },  /* level  1 */
+    { 14, 14, 14,  1,  4,  1, ZSTD_fast    },  /* level  2 */
     { 14, 14, 14,  1,  4,  6, ZSTD_dfast   },  /* level  3.*/
     { 14, 14, 14,  4,  4,  6, ZSTD_greedy  },  /* level  4.*/
     { 14, 14, 14,  3,  4,  6, ZSTD_lazy    },  /* level  5.*/
@@ -3295,40 +3410,40 @@
     { 14, 15, 15,  6,  3,128, ZSTD_btopt   },  /* level 17.*/
     { 14, 15, 15,  6,  3,256, ZSTD_btopt   },  /* level 18.*/
     { 14, 15, 15,  7,  3,256, ZSTD_btopt   },  /* level 19.*/
-    { 14, 15, 15,  8,  3,256, ZSTD_btopt2  },  /* level 20.*/
-    { 14, 15, 15,  9,  3,256, ZSTD_btopt2  },  /* level 21.*/
-    { 14, 15, 15, 10,  3,256, ZSTD_btopt2  },  /* level 22.*/
+    { 14, 15, 15,  8,  3,256, ZSTD_btultra },  /* level 20.*/
+    { 14, 15, 15,  9,  3,256, ZSTD_btultra },  /* level 21.*/
+    { 14, 15, 15, 10,  3,256, ZSTD_btultra },  /* level 22.*/
 },
 };
 
 /*! ZSTD_getCParams() :
-*   @return ZSTD_compressionParameters structure for a selected compression level, `srcSize` and `dictSize`.
+*  @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
 *   Size values are optional, provide 0 if not known or unused */
-ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSize, size_t dictSize)
+ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
 {
-    ZSTD_compressionParameters cp;
-    size_t const addedSize = srcSize ? 0 : 500;
-    U64 const rSize = srcSize+dictSize ? srcSize+dictSize+addedSize : (U64)-1;
+    size_t const addedSize = srcSizeHint ? 0 : 500;
+    U64 const rSize = srcSizeHint+dictSize ? srcSizeHint+dictSize+addedSize : (U64)-1;
     U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB);   /* intentional underflow for srcSizeHint == 0 */
-    if (compressionLevel <= 0) compressionLevel = ZSTD_DEFAULT_CLEVEL;   /* 0 == default; no negative compressionLevel yet */
-    if (compressionLevel > ZSTD_MAX_CLEVEL) compressionLevel = ZSTD_MAX_CLEVEL;
-    cp = ZSTD_defaultCParameters[tableID][compressionLevel];
-    if (MEM_32bits()) {   /* auto-correction, for 32-bits mode */
-        if (cp.windowLog > ZSTD_WINDOWLOG_MAX) cp.windowLog = ZSTD_WINDOWLOG_MAX;
-        if (cp.chainLog > ZSTD_CHAINLOG_MAX) cp.chainLog = ZSTD_CHAINLOG_MAX;
-        if (cp.hashLog > ZSTD_HASHLOG_MAX) cp.hashLog = ZSTD_HASHLOG_MAX;
-    }
-    cp = ZSTD_adjustCParams(cp, srcSize, dictSize);
-    return cp;
+    int row = compressionLevel;
+    DEBUGLOG(5, "ZSTD_getCParams (cLevel=%i)", compressionLevel);
+    if (compressionLevel == 0) row = ZSTD_CLEVEL_DEFAULT;   /* 0 == default */
+    if (compressionLevel < 0) row = 0;   /* entry 0 is baseline for fast mode */
+    if (compressionLevel > ZSTD_MAX_CLEVEL) row = ZSTD_MAX_CLEVEL;
+    {   ZSTD_compressionParameters cp = ZSTD_defaultCParameters[tableID][row];
+        if (compressionLevel < 0) cp.targetLength = (unsigned)(-compressionLevel);   /* acceleration factor */
+        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize); }
+
 }
 
 /*! ZSTD_getParams() :
 *   same as ZSTD_getCParams(), but @return a `ZSTD_parameters` object (instead of `ZSTD_compressionParameters`).
 *   All fields of `ZSTD_frameParameters` are set to default (0) */
-ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSize, size_t dictSize) {
+ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize) {
     ZSTD_parameters params;
-    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, srcSize, dictSize);
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, srcSizeHint, dictSize);
+    DEBUGLOG(5, "ZSTD_getParams (cLevel=%i)", compressionLevel);
     memset(&params, 0, sizeof(params));
     params.cParams = cParams;
+    params.fParams.contentSizeFlag = 1;
     return params;
 }
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_compress_internal.h	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,709 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* This header contains definitions
+ * that shall **only** be used by modules within lib/compress.
+ */
+
+#ifndef ZSTD_COMPRESS_H
+#define ZSTD_COMPRESS_H
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "zstd_internal.h"
+#ifdef ZSTD_MULTITHREAD
+#  include "zstdmt_compress.h"
+#endif
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-*************************************
+*  Constants
+***************************************/
+#define kSearchStrength      8
+#define HASH_READ_SIZE       8
+#define ZSTD_DUBT_UNSORTED_MARK 1   /* For btlazy2 strategy, index 1 now means "unsorted".
+                                       It could be confused for a real successor at index "1", if sorted as larger than its predecessor.
+                                       It's not a big deal though : candidate will just be sorted again.
+                                       Additionnally, candidate position 1 will be lost.
+                                       But candidate 1 cannot hide a large tree of candidates, so it's a minimal loss.
+                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be misdhandled after table re-use with a different strategy */
+
+
+/*-*************************************
+*  Context memory management
+***************************************/
+typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
+typedef enum { zcss_init=0, zcss_load, zcss_flush } ZSTD_cStreamStage;
+
+typedef struct ZSTD_prefixDict_s {
+    const void* dict;
+    size_t dictSize;
+    ZSTD_dictContentType_e dictContentType;
+} ZSTD_prefixDict;
+
+typedef struct {
+    U32 hufCTable[HUF_CTABLE_SIZE_U32(255)];
+    FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
+    FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
+    FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
+    HUF_repeat hufCTable_repeatMode;
+    FSE_repeat offcode_repeatMode;
+    FSE_repeat matchlength_repeatMode;
+    FSE_repeat litlength_repeatMode;
+} ZSTD_entropyCTables_t;
+
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTD_match_t;
+
+typedef struct {
+    int price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTD_REP_NUM];
+} ZSTD_optimal_t;
+
+typedef struct {
+    /* All tables are allocated inside cctx->workspace by ZSTD_resetCCtx_internal() */
+    U32* litFreq;               /* table of literals statistics, of size 256 */
+    U32* litLengthFreq;         /* table of litLength statistics, of size (MaxLL+1) */
+    U32* matchLengthFreq;       /* table of matchLength statistics, of size (MaxML+1) */
+    U32* offCodeFreq;           /* table of offCode statistics, of size (MaxOff+1) */
+    ZSTD_match_t* matchTable;   /* list of found matches, of size ZSTD_OPT_NUM+1 */
+    ZSTD_optimal_t* priceTable; /* All positions tracked by optimal parser, of size ZSTD_OPT_NUM+1 */
+
+    U32  litSum;                 /* nb of literals */
+    U32  litLengthSum;           /* nb of litLength codes */
+    U32  matchLengthSum;         /* nb of matchLength codes */
+    U32  offCodeSum;             /* nb of offset codes */
+    /* begin updated by ZSTD_setLog2Prices */
+    U32  log2litSum;             /* pow2 to compare log2(litfreq) to */
+    U32  log2litLengthSum;       /* pow2 to compare log2(llfreq) to */
+    U32  log2matchLengthSum;     /* pow2 to compare log2(mlfreq) to */
+    U32  log2offCodeSum;         /* pow2 to compare log2(offreq) to */
+    /* end : updated by ZSTD_setLog2Prices */
+    U32  staticPrices;           /* prices follow a pre-defined cost structure, statistics are irrelevant */
+} optState_t;
+
+typedef struct {
+  ZSTD_entropyCTables_t entropy;
+  U32 rep[ZSTD_REP_NUM];
+} ZSTD_compressedBlockState_t;
+
+typedef struct {
+    BYTE const* nextSrc;    /* next block here to continue on current prefix */
+    BYTE const* base;       /* All regular indexes relative to this position */
+    BYTE const* dictBase;   /* extDict indexes relative to this position */
+    U32 dictLimit;          /* below that point, need extDict */
+    U32 lowLimit;           /* below that point, no more data */
+} ZSTD_window_t;
+
+typedef struct {
+    ZSTD_window_t window;      /* State for window round buffer management */
+    U32 loadedDictEnd;         /* index of end of dictionary */
+    U32 nextToUpdate;          /* index from which to continue table update */
+    U32 nextToUpdate3;         /* index from which to continue table update */
+    U32 hashLog3;              /* dispatch table : larger == faster, more memory */
+    U32* hashTable;
+    U32* hashTable3;
+    U32* chainTable;
+    optState_t opt;         /* optimal parser state */
+} ZSTD_matchState_t;
+
+typedef struct {
+    ZSTD_compressedBlockState_t* prevCBlock;
+    ZSTD_compressedBlockState_t* nextCBlock;
+    ZSTD_matchState_t matchState;
+} ZSTD_blockState_t;
+
+typedef struct {
+    U32 offset;
+    U32 checksum;
+} ldmEntry_t;
+
+typedef struct {
+    ZSTD_window_t window;   /* State for the window round buffer management */
+    ldmEntry_t* hashTable;
+    BYTE* bucketOffsets;    /* Next position in bucket to insert entry */
+    U64 hashPower;          /* Used to compute the rolling hash.
+                             * Depends on ldmParams.minMatchLength */
+} ldmState_t;
+
+typedef struct {
+    U32 enableLdm;          /* 1 if enable long distance matching */
+    U32 hashLog;            /* Log size of hashTable */
+    U32 bucketSizeLog;      /* Log bucket size for collision resolution, at most 8 */
+    U32 minMatchLength;     /* Minimum match length */
+    U32 hashEveryLog;       /* Log number of entries to skip */
+    U32 windowLog;          /* Window log for the LDM */
+} ldmParams_t;
+
+typedef struct {
+    U32 offset;
+    U32 litLength;
+    U32 matchLength;
+} rawSeq;
+
+typedef struct {
+  rawSeq* seq;     /* The start of the sequences */
+  size_t pos;      /* The position where reading stopped. <= size. */
+  size_t size;     /* The number of sequences. <= capacity. */
+  size_t capacity; /* The capacity of the `seq` pointer */
+} rawSeqStore_t;
+
+struct ZSTD_CCtx_params_s {
+    ZSTD_format_e format;
+    ZSTD_compressionParameters cParams;
+    ZSTD_frameParameters fParams;
+
+    int compressionLevel;
+    int disableLiteralCompression;
+    int forceWindow;           /* force back-references to respect limit of
+                                * 1<<wLog, even for dictionary */
+
+    /* Multithreading: used to pass parameters to mtctx */
+    unsigned nbWorkers;
+    unsigned jobSize;
+    unsigned overlapSizeLog;
+
+    /* Long distance matching parameters */
+    ldmParams_t ldmParams;
+
+    /* Internal use, for createCCtxParams() and freeCCtxParams() only */
+    ZSTD_customMem customMem;
+};  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
+
+struct ZSTD_CCtx_s {
+    ZSTD_compressionStage_e stage;
+    int cParamsChanged;                  /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
+    int bmi2;                            /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
+    ZSTD_CCtx_params requestedParams;
+    ZSTD_CCtx_params appliedParams;
+    U32   dictID;
+    void* workSpace;
+    size_t workSpaceSize;
+    size_t blockSize;
+    unsigned long long pledgedSrcSizePlusOne;  /* this way, 0 (default) == unknown */
+    unsigned long long consumedSrcSize;
+    unsigned long long producedCSize;
+    XXH64_state_t xxhState;
+    ZSTD_customMem customMem;
+    size_t staticSize;
+
+    seqStore_t seqStore;      /* sequences storage ptrs */
+    ldmState_t ldmState;      /* long distance matching state */
+    rawSeq* ldmSequences;     /* Storage for the ldm output sequences */
+    size_t maxNbLdmSequences;
+    rawSeqStore_t externSeqStore; /* Mutable reference to external sequences */
+    ZSTD_blockState_t blockState;
+    U32* entropyWorkspace;  /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
+
+    /* streaming */
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inToCompress;
+    size_t inBuffPos;
+    size_t inBuffTarget;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outBuffContentSize;
+    size_t outBuffFlushedSize;
+    ZSTD_cStreamStage streamStage;
+    U32    frameEnded;
+
+    /* Dictionary */
+    ZSTD_CDict* cdictLocal;
+    const ZSTD_CDict* cdict;
+    ZSTD_prefixDict prefixDict;   /* single-usage dictionary */
+
+    /* Multi-threading */
+#ifdef ZSTD_MULTITHREAD
+    ZSTDMT_CCtx* mtctx;
+#endif
+};
+
+
+typedef size_t (*ZSTD_blockCompressor) (
+        ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict);
+
+
+MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
+{
+    static const BYTE LL_Code[64] = {  0,  1,  2,  3,  4,  5,  6,  7,
+                                       8,  9, 10, 11, 12, 13, 14, 15,
+                                      16, 16, 17, 17, 18, 18, 19, 19,
+                                      20, 20, 20, 20, 21, 21, 21, 21,
+                                      22, 22, 22, 22, 22, 22, 22, 22,
+                                      23, 23, 23, 23, 23, 23, 23, 23,
+                                      24, 24, 24, 24, 24, 24, 24, 24,
+                                      24, 24, 24, 24, 24, 24, 24, 24 };
+    static const U32 LL_deltaCode = 19;
+    return (litLength > 63) ? ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
+}
+
+/* ZSTD_MLcode() :
+ * note : mlBase = matchLength - MINMATCH;
+ *        because it's the format it's stored in seqStore->sequences */
+MEM_STATIC U32 ZSTD_MLcode(U32 mlBase)
+{
+    static const BYTE ML_Code[128] = { 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15,
+                                      16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
+                                      32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37,
+                                      38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39,
+                                      40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
+                                      41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41,
+                                      42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42,
+                                      42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42 };
+    static const U32 ML_deltaCode = 36;
+    return (mlBase > 127) ? ZSTD_highbit32(mlBase) + ML_deltaCode : ML_Code[mlBase];
+}
+
+/*! ZSTD_storeSeq() :
+ *  Store a sequence (literal length, literals, offset code and match length code) into seqStore_t.
+ *  `offsetCode` : distance to match + 3 (values 1-3 are repCodes).
+ *  `mlBase` : matchLength - MINMATCH
+*/
+MEM_STATIC void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const void* literals, U32 offsetCode, size_t mlBase)
+{
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG >= 6)
+    static const BYTE* g_start = NULL;
+    if (g_start==NULL) g_start = (const BYTE*)literals;  /* note : index only works for compression within a single segment */
+    {   U32 const pos = (U32)((const BYTE*)literals - g_start);
+        DEBUGLOG(6, "Cpos%7u :%3u literals, match%3u bytes at dist.code%7u",
+               pos, (U32)litLength, (U32)mlBase+MINMATCH, (U32)offsetCode);
+    }
+#endif
+    /* copy Literals */
+    assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + 128 KB);
+    ZSTD_wildcopy(seqStorePtr->lit, literals, litLength);
+    seqStorePtr->lit += litLength;
+
+    /* literal Length */
+    if (litLength>0xFFFF) {
+        assert(seqStorePtr->longLengthID == 0); /* there can only be a single long length */
+        seqStorePtr->longLengthID = 1;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
+    seqStorePtr->sequences[0].litLength = (U16)litLength;
+
+    /* match offset */
+    seqStorePtr->sequences[0].offset = offsetCode + 1;
+
+    /* match Length */
+    if (mlBase>0xFFFF) {
+        assert(seqStorePtr->longLengthID == 0); /* there can only be a single long length */
+        seqStorePtr->longLengthID = 2;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
+    seqStorePtr->sequences[0].matchLength = (U16)mlBase;
+
+    seqStorePtr->sequences++;
+}
+
+
+/*-*************************************
+*  Match length counter
+***************************************/
+static unsigned ZSTD_NbCommonBytes (size_t val)
+{
+    if (MEM_isLittleEndian()) {
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            _BitScanForward64( &r, (U64)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 4)
+            return (__builtin_ctzll((U64)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2,
+                                                     0, 3, 1, 3, 1, 4, 2, 7,
+                                                     0, 2, 3, 6, 1, 5, 3, 5,
+                                                     1, 3, 4, 4, 2, 5, 6, 7,
+                                                     7, 0, 1, 2, 3, 3, 4, 6,
+                                                     2, 6, 5, 5, 3, 4, 5, 6,
+                                                     7, 1, 2, 4, 6, 4, 4, 5,
+                                                     7, 2, 6, 5, 7, 6, 7, 7 };
+            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r=0;
+            _BitScanForward( &r, (U32)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_ctz((U32)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0,
+                                                     3, 2, 2, 1, 3, 2, 0, 1,
+                                                     3, 3, 1, 2, 2, 2, 2, 0,
+                                                     3, 1, 2, 0, 1, 0, 1, 1 };
+            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
+#       endif
+        }
+    } else {  /* Big Endian CPU */
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            _BitScanReverse64( &r, val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 4)
+            return (__builtin_clzll(val) >> 3);
+#       else
+            unsigned r;
+            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
+            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
+            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r = 0;
+            _BitScanReverse( &r, (unsigned long)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_clz((U32)val) >> 3);
+#       else
+            unsigned r;
+            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+    }   }
+}
+
+
+MEM_STATIC size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
+{
+    const BYTE* const pStart = pIn;
+    const BYTE* const pInLoopLimit = pInLimit - (sizeof(size_t)-1);
+
+    if (pIn < pInLoopLimit) {
+        { size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+          if (diff) return ZSTD_NbCommonBytes(diff); }
+        pIn+=sizeof(size_t); pMatch+=sizeof(size_t);
+        while (pIn < pInLoopLimit) {
+            size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+            if (!diff) { pIn+=sizeof(size_t); pMatch+=sizeof(size_t); continue; }
+            pIn += ZSTD_NbCommonBytes(diff);
+            return (size_t)(pIn - pStart);
+    }   }
+    if (MEM_64bits() && (pIn<(pInLimit-3)) && (MEM_read32(pMatch) == MEM_read32(pIn))) { pIn+=4; pMatch+=4; }
+    if ((pIn<(pInLimit-1)) && (MEM_read16(pMatch) == MEM_read16(pIn))) { pIn+=2; pMatch+=2; }
+    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
+    return (size_t)(pIn - pStart);
+}
+
+/** ZSTD_count_2segments() :
+ *  can count match length with `ip` & `match` in 2 different segments.
+ *  convention : on reaching mEnd, match count continue starting from iStart
+ */
+MEM_STATIC size_t
+ZSTD_count_2segments(const BYTE* ip, const BYTE* match,
+                     const BYTE* iEnd, const BYTE* mEnd, const BYTE* iStart)
+{
+    const BYTE* const vEnd = MIN( ip + (mEnd - match), iEnd);
+    size_t const matchLength = ZSTD_count(ip, match, vEnd);
+    if (match + matchLength != mEnd) return matchLength;
+    return matchLength + ZSTD_count(ip+matchLength, iStart, iEnd);
+}
+
+
+/*-*************************************
+ *  Hashes
+ ***************************************/
+static const U32 prime3bytes = 506832829U;
+static U32    ZSTD_hash3(U32 u, U32 h) { return ((u << (32-24)) * prime3bytes)  >> (32-h) ; }
+MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h); } /* only in zstd_opt.h */
+
+static const U32 prime4bytes = 2654435761U;
+static U32    ZSTD_hash4(U32 u, U32 h) { return (u * prime4bytes) >> (32-h) ; }
+static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_read32(ptr), h); }
+
+static const U64 prime5bytes = 889523592379ULL;
+static size_t ZSTD_hash5(U64 u, U32 h) { return (size_t)(((u  << (64-40)) * prime5bytes) >> (64-h)) ; }
+static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h); }
+
+static const U64 prime6bytes = 227718039650203ULL;
+static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
+static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
+
+static const U64 prime7bytes = 58295818150454627ULL;
+static size_t ZSTD_hash7(U64 u, U32 h) { return (size_t)(((u  << (64-56)) * prime7bytes) >> (64-h)) ; }
+static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h); }
+
+static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
+static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
+static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
+
+MEM_STATIC size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
+{
+    switch(mls)
+    {
+    default:
+    case 4: return ZSTD_hash4Ptr(p, hBits);
+    case 5: return ZSTD_hash5Ptr(p, hBits);
+    case 6: return ZSTD_hash6Ptr(p, hBits);
+    case 7: return ZSTD_hash7Ptr(p, hBits);
+    case 8: return ZSTD_hash8Ptr(p, hBits);
+    }
+}
+
+/*-*************************************
+*  Round buffer management
+***************************************/
+/* Max current allowed */
+#define ZSTD_CURRENT_MAX ((3U << 29) + (1U << ZSTD_WINDOWLOG_MAX))
+/* Maximum chunk size before overflow correction needs to be called again */
+#define ZSTD_CHUNKSIZE_MAX                                                     \
+    ( ((U32)-1)                  /* Maximum ending current index */            \
+    - ZSTD_CURRENT_MAX)          /* Maximum beginning lowLimit */
+
+/**
+ * ZSTD_window_clear():
+ * Clears the window containing the history by simply setting it to empty.
+ */
+MEM_STATIC void ZSTD_window_clear(ZSTD_window_t* window)
+{
+    size_t const endT = (size_t)(window->nextSrc - window->base);
+    U32 const end = (U32)endT;
+
+    window->lowLimit = end;
+    window->dictLimit = end;
+}
+
+/**
+ * ZSTD_window_hasExtDict():
+ * Returns non-zero if the window has a non-empty extDict.
+ */
+MEM_STATIC U32 ZSTD_window_hasExtDict(ZSTD_window_t const window)
+{
+    return window.lowLimit < window.dictLimit;
+}
+
+/**
+ * ZSTD_window_needOverflowCorrection():
+ * Returns non-zero if the indices are getting too large and need overflow
+ * protection.
+ */
+MEM_STATIC U32 ZSTD_window_needOverflowCorrection(ZSTD_window_t const window,
+                                                  void const* srcEnd)
+{
+    U32 const current = (U32)((BYTE const*)srcEnd - window.base);
+    return current > ZSTD_CURRENT_MAX;
+}
+
+/**
+ * ZSTD_window_correctOverflow():
+ * Reduces the indices to protect from index overflow.
+ * Returns the correction made to the indices, which must be applied to every
+ * stored index.
+ *
+ * The least significant cycleLog bits of the indices must remain the same,
+ * which may be 0. Every index up to maxDist in the past must be valid.
+ * NOTE: (maxDist & cycleMask) must be zero.
+ */
+MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
+                                           U32 maxDist, void const* src)
+{
+    /* preemptive overflow correction:
+     * 1. correction is large enough:
+     *    lowLimit > (3<<29) ==> current > 3<<29 + 1<<windowLog
+     *    1<<windowLog <= newCurrent < 1<<chainLog + 1<<windowLog
+     *
+     *    current - newCurrent
+     *    > (3<<29 + 1<<windowLog) - (1<<windowLog + 1<<chainLog)
+     *    > (3<<29) - (1<<chainLog)
+     *    > (3<<29) - (1<<30)             (NOTE: chainLog <= 30)
+     *    > 1<<29
+     *
+     * 2. (ip+ZSTD_CHUNKSIZE_MAX - cctx->base) doesn't overflow:
+     *    After correction, current is less than (1<<chainLog + 1<<windowLog).
+     *    In 64-bit mode we are safe, because we have 64-bit ptrdiff_t.
+     *    In 32-bit mode we are safe, because (chainLog <= 29), so
+     *    ip+ZSTD_CHUNKSIZE_MAX - cctx->base < 1<<32.
+     * 3. (cctx->lowLimit + 1<<windowLog) < 1<<32:
+     *    windowLog <= 31 ==> 3<<29 + 1<<windowLog < 7<<29 < 1<<32.
+     */
+    U32 const cycleMask = (1U << cycleLog) - 1;
+    U32 const current = (U32)((BYTE const*)src - window->base);
+    U32 const newCurrent = (current & cycleMask) + maxDist;
+    U32 const correction = current - newCurrent;
+    assert((maxDist & cycleMask) == 0);
+    assert(current > newCurrent);
+    /* Loose bound, should be around 1<<29 (see above) */
+    assert(correction > 1<<28);
+
+    window->base += correction;
+    window->dictBase += correction;
+    window->lowLimit -= correction;
+    window->dictLimit -= correction;
+
+    DEBUGLOG(4, "Correction of 0x%x bytes to lowLimit=0x%x", correction,
+             window->lowLimit);
+    return correction;
+}
+
+/**
+ * ZSTD_window_enforceMaxDist():
+ * Updates lowLimit so that:
+ *    (srcEnd - base) - lowLimit == maxDist + loadedDictEnd
+ * This allows a simple check that index >= lowLimit to see if index is valid.
+ * This must be called before a block compression call, with srcEnd as the block
+ * source end.
+ * If loadedDictEndPtr is not NULL, we set it to zero once we update lowLimit.
+ * This is because dictionaries are allowed to be referenced as long as the last
+ * byte of the dictionary is in the window, but once they are out of range,
+ * they cannot be referenced. If loadedDictEndPtr is NULL, we use
+ * loadedDictEnd == 0.
+ */
+MEM_STATIC void ZSTD_window_enforceMaxDist(ZSTD_window_t* window,
+                                           void const* srcEnd, U32 maxDist,
+                                           U32* loadedDictEndPtr)
+{
+    U32 const current = (U32)((BYTE const*)srcEnd - window->base);
+    U32 loadedDictEnd = loadedDictEndPtr != NULL ? *loadedDictEndPtr : 0;
+    if (current > maxDist + loadedDictEnd) {
+        U32 const newLowLimit = current - maxDist;
+        if (window->lowLimit < newLowLimit) window->lowLimit = newLowLimit;
+        if (window->dictLimit < window->lowLimit) {
+            DEBUGLOG(5, "Update dictLimit from %u to %u", window->dictLimit,
+                     window->lowLimit);
+            window->dictLimit = window->lowLimit;
+        }
+        if (loadedDictEndPtr)
+            *loadedDictEndPtr = 0;
+    }
+}
+
+/**
+ * ZSTD_window_update():
+ * Updates the window by appending [src, src + srcSize) to the window.
+ * If it is not contiguous, the current prefix becomes the extDict, and we
+ * forget about the extDict. Handles overlap of the prefix and extDict.
+ * Returns non-zero if the segment is contiguous.
+ */
+MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
+                                  void const* src, size_t srcSize)
+{
+    BYTE const* const ip = (BYTE const*)src;
+    U32 contiguous = 1;
+    /* Check if blocks follow each other */
+    if (src != window->nextSrc) {
+        /* not contiguous */
+        size_t const distanceFromBase = (size_t)(window->nextSrc - window->base);
+        DEBUGLOG(5, "Non contiguous blocks, new segment starts at %u",
+                 window->dictLimit);
+        window->lowLimit = window->dictLimit;
+        assert(distanceFromBase == (size_t)(U32)distanceFromBase);  /* should never overflow */
+        window->dictLimit = (U32)distanceFromBase;
+        window->dictBase = window->base;
+        window->base = ip - distanceFromBase;
+        // ms->nextToUpdate = window->dictLimit;
+        if (window->dictLimit - window->lowLimit < HASH_READ_SIZE) window->lowLimit = window->dictLimit;   /* too small extDict */
+        contiguous = 0;
+    }
+    window->nextSrc = ip + srcSize;
+    /* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
+    if ( (ip+srcSize > window->dictBase + window->lowLimit)
+       & (ip < window->dictBase + window->dictLimit)) {
+        ptrdiff_t const highInputIdx = (ip + srcSize) - window->dictBase;
+        U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)window->dictLimit) ? window->dictLimit : (U32)highInputIdx;
+        window->lowLimit = lowLimitMax;
+    }
+    return contiguous;
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+
+/* ==============================================================
+ * Private declarations
+ * These prototypes shall only be called from within lib/compress
+ * ============================================================== */
+
+/* ZSTD_getCParamsFromCCtxParams() :
+ * cParams are built depending on compressionLevel, src size hints, 
+ * LDM and manually set compression parameters.
+ */
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize);
+
+/*! ZSTD_initCStream_internal() :
+ *  Private use only. Init streaming operation.
+ *  expects params to be valid.
+ *  must receive dict, or cdict, or none, but not both.
+ *  @return : 0, or an error code */
+size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
+                     const void* dict, size_t dictSize,
+                     const ZSTD_CDict* cdict,
+                     ZSTD_CCtx_params  params, unsigned long long pledgedSrcSize);
+
+/*! ZSTD_compressStream_generic() :
+ *  Private use only. To be called from zstdmt_compress.c in single-thread mode. */
+size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
+                                   ZSTD_outBuffer* output,
+                                   ZSTD_inBuffer* input,
+                                   ZSTD_EndDirective const flushMode);
+
+/*! ZSTD_getCParamsFromCDict() :
+ *  as the name implies */
+ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict);
+
+/* ZSTD_compressBegin_advanced_internal() :
+ * Private use only. To be called from zstdmt_compress.c. */
+size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    const ZSTD_CDict* cdict,
+                                    ZSTD_CCtx_params params,
+                                    unsigned long long pledgedSrcSize);
+
+/* ZSTD_compress_advanced_internal() :
+ * Private use only. To be called from zstdmt_compress.c. */
+size_t ZSTD_compress_advanced_internal(ZSTD_CCtx* cctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict,size_t dictSize,
+                                 ZSTD_CCtx_params params);
+
+
+/* ZSTD_writeLastEmptyBlock() :
+ * output an empty Block with end-of-frame mark to complete a frame
+ * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
+ *           or an error code if `dstCapcity` is too small (<ZSTD_blockHeaderSize)
+ */
+size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity);
+
+
+/* ZSTD_referenceExternalSequences() :
+ * Must be called before starting a compression operation.
+ * seqs must parse a prefix of the source.
+ * This cannot be used when long range matching is enabled.
+ * Zstd will use these sequences, and pass the literals to a secondary block
+ * compressor.
+ * @return : An error code on failure.
+ * NOTE: seqs are not verified! Invalid sequences can cause out-of-bounds memory
+ * access and data corruption.
+ */
+size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq);
+
+
+#endif /* ZSTD_COMPRESS_H */
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_double_fast.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,327 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_double_fast.h"
+
+
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                              ZSTD_compressionParameters const* cParams,
+                              void const* end)
+{
+    U32* const hashLarge = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32  const mls = cParams->searchLength;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash tables.
+     * Insert the other positions into the large hash table if their entry
+     * is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const current = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
+            size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
+            if (i == 0)
+                hashSmall[smHash] = current + i;
+            if (i == 0 || hashLarge[lgHash] == 0)
+                hashLarge[lgHash] = current + i;
+        }
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_doubleFast_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
+        U32 const mls /* template */)
+{
+    U32* const hashLong = ms->hashTable;
+    const U32 hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    const U32 hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32 lowestIndex = ms->window.dictLimit;
+    const BYTE* const lowest = base + lowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved = 0;
+
+    /* init */
+    ip += (ip==lowest);
+    {   U32 const maxRep = (U32)(ip-lowest);
+        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
+    }
+
+    /* Main Search Loop */
+    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+        size_t mLength;
+        size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
+        size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
+        U32 const current = (U32)(ip-base);
+        U32 const matchIndexL = hashLong[h2];
+        U32 const matchIndexS = hashSmall[h];
+        const BYTE* matchLong = base + matchIndexL;
+        const BYTE* match = base + matchIndexS;
+        hashLong[h2] = hashSmall[h] = current;   /* update hash tables */
+
+        assert(offset_1 <= current);   /* supposed guaranteed by construction */
+        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+            /* favor repcode */
+            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            U32 offset;
+            if ( (matchIndexL > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip)) ) {
+                mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
+                offset = (U32)(ip-matchLong);
+                while (((ip>anchor) & (matchLong>lowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
+            } else if ( (matchIndexS > lowestIndex) && (MEM_read32(match) == MEM_read32(ip)) ) {
+                size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+                U32 const matchIndexL3 = hashLong[hl3];
+                const BYTE* matchL3 = base + matchIndexL3;
+                hashLong[hl3] = current + 1;
+                if ( (matchIndexL3 > lowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1)) ) {
+                    mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
+                    ip++;
+                    offset = (U32)(ip-matchL3);
+                    while (((ip>anchor) & (matchL3>lowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
+                } else {
+                    mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+                    offset = (U32)(ip-match);
+                    while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+                }
+            } else {
+                ip += ((ip-anchor) >> kSearchStrength) + 1;
+                continue;
+            }
+
+            offset_2 = offset_1;
+            offset_1 = offset;
+
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }
+
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] =
+                hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;  /* here because current+2 could be > iend-8 */
+            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] =
+                hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
+
+            /* check immediate repcode */
+            while ( (ip <= ilimit)
+                 && ( (offset_2>0)
+                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
+                hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
+                hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
+                ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH);
+                ip += rLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved;
+    rep[1] = offset_2 ? offset_2 : offsetSaved;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_doubleFast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    const U32 mls = cParams->searchLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 4);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 5);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 6);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 7);
+    }
+}
+
+
+static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
+        U32 const mls /* template */)
+{
+    U32* const hashLong = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   lowestIndex = ms->window.lowLimit;
+    const BYTE* const dictStart = dictBase + lowestIndex;
+    const U32   dictLimit = ms->window.dictLimit;
+    const BYTE* const lowPrefixPtr = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    /* Search Loop */
+    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
+        const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
+        const U32 matchIndex = hashSmall[hSmall];
+        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
+        const BYTE* match = matchBase + matchIndex;
+
+        const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
+        const U32 matchLongIndex = hashLong[hLong];
+        const BYTE* matchLongBase = matchLongIndex < dictLimit ? dictBase : base;
+        const BYTE* matchLong = matchLongBase + matchLongIndex;
+
+        const U32 current = (U32)(ip-base);
+        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
+        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
+        const BYTE* repMatch = repBase + repIndex;
+        size_t mLength;
+        hashSmall[hSmall] = hashLong[hLong] = current;   /* update hash table */
+
+        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
+           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            if ((matchLongIndex > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
+                const BYTE* matchEnd = matchLongIndex < dictLimit ? dictEnd : iend;
+                const BYTE* lowMatchPtr = matchLongIndex < dictLimit ? dictStart : lowPrefixPtr;
+                U32 offset;
+                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, lowPrefixPtr) + 8;
+                offset = current - matchLongIndex;
+                while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+
+            } else if ((matchIndex > lowestIndex) && (MEM_read32(match) == MEM_read32(ip))) {
+                size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+                U32 const matchIndex3 = hashLong[h3];
+                const BYTE* const match3Base = matchIndex3 < dictLimit ? dictBase : base;
+                const BYTE* match3 = match3Base + matchIndex3;
+                U32 offset;
+                hashLong[h3] = current + 1;
+                if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
+                    const BYTE* matchEnd = matchIndex3 < dictLimit ? dictEnd : iend;
+                    const BYTE* lowMatchPtr = matchIndex3 < dictLimit ? dictStart : lowPrefixPtr;
+                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, lowPrefixPtr) + 8;
+                    ip++;
+                    offset = current+1 - matchIndex3;
+                    while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
+                } else {
+                    const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
+                    const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
+                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+                    offset = current - matchIndex;
+                    while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
+                }
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+
+            } else {
+                ip += ((ip-anchor) >> kSearchStrength) + 1;
+                continue;
+        }   }
+
+        /* found a match : store it */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;
+            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = current+2;
+            hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
+            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH);
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_doubleFast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    U32 const mls = cParams->searchLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 4);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 5);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 6);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 7);
+    }
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_double_fast.h	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,36 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_DOUBLE_FAST_H
+#define ZSTD_DOUBLE_FAST_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "mem.h"      /* U32 */
+#include "zstd_compress_internal.h"     /* ZSTD_CCtx, size_t */
+
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                              ZSTD_compressionParameters const* cParams,
+                              void const* end);
+size_t ZSTD_compressBlock_doubleFast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_doubleFast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_DOUBLE_FAST_H */
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_fast.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,259 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_fast.h"
+
+
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        ZSTD_compressionParameters const* cParams,
+                        void const* end)
+{
+    U32* const hashTable = ms->hashTable;
+    U32  const hBits = cParams->hashLog;
+    U32  const mls = cParams->searchLength;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash table.
+     * Insert the other positions if their hash entry is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const current = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const hash = ZSTD_hashPtr(ip + i, hBits, mls);
+            if (i == 0 || hashTable[hash] == 0)
+                hashTable[hash] = current + i;
+        }
+    }
+}
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_fast_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const hlog, U32 const stepSize, U32 const mls)
+{
+    U32* const hashTable = ms->hashTable;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   lowestIndex = ms->window.dictLimit;
+    const BYTE* const lowest = base + lowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved = 0;
+
+    /* init */
+    ip += (ip==lowest);
+    {   U32 const maxRep = (U32)(ip-lowest);
+        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
+    }
+
+    /* Main Search Loop */
+    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+        size_t mLength;
+        size_t const h = ZSTD_hashPtr(ip, hlog, mls);
+        U32 const current = (U32)(ip-base);
+        U32 const matchIndex = hashTable[h];
+        const BYTE* match = base + matchIndex;
+        hashTable[h] = current;   /* update hash table */
+
+        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            if ( (matchIndex <= lowestIndex)
+              || (MEM_read32(match) != MEM_read32(ip)) ) {
+                assert(stepSize >= 1);
+                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
+                continue;
+            }
+            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+            {   U32 const offset = (U32)(ip-match);
+                while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }   }
+
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;  /* here because current+2 could be > iend-8 */
+            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
+            /* check immediate repcode */
+            while ( (ip <= ilimit)
+                 && ( (offset_2>0)
+                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; }  /* swap offset_2 <=> offset_1 */
+                hashTable[ZSTD_hashPtr(ip, hlog, mls)] = (U32)(ip-base);
+                ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH);
+                ip += rLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved;
+    rep[1] = offset_2 ? offset_2 : offsetSaved;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_fast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    U32 const hlog = cParams->hashLog;
+    U32 const mls = cParams->searchLength;
+    U32 const stepSize = cParams->targetLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 4);
+    case 5 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 5);
+    case 6 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 6);
+    case 7 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 7);
+    }
+}
+
+
+static size_t ZSTD_compressBlock_fast_extDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const hlog, U32 const stepSize, U32 const mls)
+{
+    U32* hashTable = ms->hashTable;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   lowestIndex = ms->window.lowLimit;
+    const BYTE* const dictStart = dictBase + lowestIndex;
+    const U32   dictLimit = ms->window.dictLimit;
+    const BYTE* const lowPrefixPtr = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    /* Search Loop */
+    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
+        const size_t h = ZSTD_hashPtr(ip, hlog, mls);
+        const U32 matchIndex = hashTable[h];
+        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
+        const BYTE* match = matchBase + matchIndex;
+        const U32 current = (U32)(ip-base);
+        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
+        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
+        const BYTE* repMatch = repBase + repIndex;
+        size_t mLength;
+        hashTable[h] = current;   /* update hash table */
+
+        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
+           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            if ( (matchIndex < lowestIndex) ||
+                 (MEM_read32(match) != MEM_read32(ip)) ) {
+                assert(stepSize >= 1);
+                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
+                continue;
+            }
+            {   const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
+                const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
+                U32 offset;
+                mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+                while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
+                offset = current - matchIndex;
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }   }
+
+        /* found a match : store it */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;
+            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH);
+                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_fast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    U32 const hlog = cParams->hashLog;
+    U32 const mls = cParams->searchLength;
+    U32 const stepSize = cParams->targetLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 4);
+    case 5 :
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 5);
+    case 6 :
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 6);
+    case 7 :
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 7);
+    }
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_fast.h	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_FAST_H
+#define ZSTD_FAST_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "mem.h"      /* U32 */
+#include "zstd_compress_internal.h"
+
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        ZSTD_compressionParameters const* cParams,
+                        void const* end);
+size_t ZSTD_compressBlock_fast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_fast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_FAST_H */
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_lazy.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,824 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_lazy.h"
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+
+void ZSTD_updateDUBT(
+                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                const BYTE* ip, const BYTE* iend,
+                U32 mls)
+{
+    U32* const hashTable = ms->hashTable;
+    U32  const hashLog = cParams->hashLog;
+
+    U32* const bt = ms->chainTable;
+    U32  const btLog  = cParams->chainLog - 1;
+    U32  const btMask = (1 << btLog) - 1;
+
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+
+    if (idx != target)
+        DEBUGLOG(7, "ZSTD_updateDUBT, from %u to %u (dictLimit:%u)",
+                    idx, target, ms->window.dictLimit);
+    assert(ip + 8 <= iend);   /* condition for ZSTD_hashPtr */
+    (void)iend;
+
+    assert(idx >= ms->window.dictLimit);   /* condition for valid base+idx */
+    for ( ; idx < target ; idx++) {
+        size_t const h  = ZSTD_hashPtr(base + idx, hashLog, mls);   /* assumption : ip + 8 <= iend */
+        U32    const matchIndex = hashTable[h];
+
+        U32*   const nextCandidatePtr = bt + 2*(idx&btMask);
+        U32*   const sortMarkPtr  = nextCandidatePtr + 1;
+
+        DEBUGLOG(8, "ZSTD_updateDUBT: insert %u", idx);
+        hashTable[h] = idx;   /* Update Hash Table */
+        *nextCandidatePtr = matchIndex;   /* update BT like a chain */
+        *sortMarkPtr = ZSTD_DUBT_UNSORTED_MARK;
+    }
+    ms->nextToUpdate = target;
+}
+
+
+/** ZSTD_insertDUBT1() :
+ *  sort one already inserted but unsorted position
+ *  assumption : current >= btlow == (current - btmask)
+ *  doesn't fail */
+static void ZSTD_insertDUBT1(
+                 ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                 U32 current, const BYTE* inputEnd,
+                 U32 nbCompares, U32 btLow, int extDict)
+{
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const ip = (current>=dictLimit) ? base + current : dictBase + current;
+    const BYTE* const iend = (current>=dictLimit) ? inputEnd : dictBase + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* match;
+    U32* smallerPtr = bt + 2*(current&btMask);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 matchIndex = *smallerPtr;
+    U32 dummy32;   /* to be nullified at the end */
+    U32 const windowLow = ms->window.lowLimit;
+
+    DEBUGLOG(8, "ZSTD_insertDUBT1(%u) (dictLimit=%u, lowLimit=%u)",
+                current, dictLimit, windowLow);
+    assert(current >= btLow);
+    assert(ip < iend);   /* condition for ZSTD_count */
+
+    while (nbCompares-- && (matchIndex > windowLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(matchIndex < current);
+
+        if ( (!extDict)
+          || (matchIndex+matchLength >= dictLimit)  /* both in current segment*/
+          || (current < dictLimit) /* both in extDict */) {
+            const BYTE* const mBase = !extDict || ((matchIndex+matchLength) >= dictLimit) ? base : dictBase;
+            assert( (matchIndex+matchLength >= dictLimit)   /* might be wrong if extDict is incorrectly set to 0 */
+                 || (current < dictLimit) );
+            match = mBase + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        DEBUGLOG(8, "ZSTD_insertDUBT1: comparing %u with %u : found %u common bytes ",
+                    current, matchIndex, (U32)matchLength);
+
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is smaller : next => %u",
+                        matchIndex, btLow, nextPtr[1]);
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is larger => %u",
+                        matchIndex, btLow, nextPtr[0]);
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+}
+
+
+static size_t ZSTD_DUBT_findBestMatch (
+                            ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                            const BYTE* const ip, const BYTE* const iend,
+                            size_t* offsetPtr,
+                            U32 const mls,
+                            U32 const extDict)
+{
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32          matchIndex  = hashTable[h];
+
+    const BYTE* const base = ms->window.base;
+    U32    const current = (U32)(ip-base);
+    U32    const windowLow = ms->window.lowLimit;
+
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32    const btLow = (btMask >= current) ? 0 : current - btMask;
+    U32    const unsortLimit = MAX(btLow, windowLow);
+
+    U32*         nextCandidate = bt + 2*(matchIndex&btMask);
+    U32*         unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+    U32          nbCompares = 1U << cParams->searchLog;
+    U32          nbCandidates = nbCompares;
+    U32          previousCandidate = 0;
+
+    DEBUGLOG(7, "ZSTD_DUBT_findBestMatch (%u) ", current);
+    assert(ip <= iend-8);   /* required for h calculation */
+
+    /* reach end of unsorted candidates list */
+    while ( (matchIndex > unsortLimit)
+         && (*unsortedMark == ZSTD_DUBT_UNSORTED_MARK)
+         && (nbCandidates > 1) ) {
+        DEBUGLOG(8, "ZSTD_DUBT_findBestMatch: candidate %u is unsorted",
+                    matchIndex);
+        *unsortedMark = previousCandidate;
+        previousCandidate = matchIndex;
+        matchIndex = *nextCandidate;
+        nextCandidate = bt + 2*(matchIndex&btMask);
+        unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+        nbCandidates --;
+    }
+
+    if ( (matchIndex > unsortLimit)
+      && (*unsortedMark==ZSTD_DUBT_UNSORTED_MARK) ) {
+        DEBUGLOG(7, "ZSTD_DUBT_findBestMatch: nullify last unsorted candidate %u",
+                    matchIndex);
+        *nextCandidate = *unsortedMark = 0;   /* nullify next candidate if it's still unsorted (note : simplification, detrimental to compression ratio, beneficial for speed) */
+    }
+
+    /* batch sort stacked candidates */
+    matchIndex = previousCandidate;
+    while (matchIndex) {  /* will end on matchIndex == 0 */
+        U32* const nextCandidateIdxPtr = bt + 2*(matchIndex&btMask) + 1;
+        U32 const nextCandidateIdx = *nextCandidateIdxPtr;
+        ZSTD_insertDUBT1(ms, cParams, matchIndex, iend,
+                         nbCandidates, unsortLimit, extDict);
+        matchIndex = nextCandidateIdx;
+        nbCandidates++;
+    }
+
+    /* find longest match */
+    {   size_t commonLengthSmaller=0, commonLengthLarger=0;
+        const BYTE* const dictBase = ms->window.dictBase;
+        const U32 dictLimit = ms->window.dictLimit;
+        const BYTE* const dictEnd = dictBase + dictLimit;
+        const BYTE* const prefixStart = base + dictLimit;
+        U32* smallerPtr = bt + 2*(current&btMask);
+        U32* largerPtr  = bt + 2*(current&btMask) + 1;
+        U32 matchEndIdx = current+8+1;
+        U32 dummy32;   /* to be nullified at the end */
+        size_t bestLength = 0;
+
+        matchIndex  = hashTable[h];
+        hashTable[h] = current;   /* Update Hash Table */
+
+        while (nbCompares-- && (matchIndex > windowLow)) {
+            U32* const nextPtr = bt + 2*(matchIndex & btMask);
+            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+            const BYTE* match;
+
+            if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+                match = base + matchIndex;
+                matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+            } else {
+                match = dictBase + matchIndex;
+                matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+                if (matchIndex+matchLength >= dictLimit)
+                    match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+            }
+
+            if (matchLength > bestLength) {
+                if (matchLength > matchEndIdx - matchIndex)
+                    matchEndIdx = matchIndex + (U32)matchLength;
+                if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
+                    bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + current - matchIndex;
+                if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+                }
+            }
+
+            if (match[matchLength] < ip[matchLength]) {
+                /* match is smaller than current */
+                *smallerPtr = matchIndex;             /* update smaller idx */
+                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+                if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+                matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            } else {
+                /* match is larger than current */
+                *largerPtr = matchIndex;
+                commonLengthLarger = matchLength;
+                if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                largerPtr = nextPtr;
+                matchIndex = nextPtr[0];
+        }   }
+
+        *smallerPtr = *largerPtr = 0;
+
+        assert(matchEndIdx > current+8); /* ensure nextToUpdate is increased */
+        ms->nextToUpdate = matchEndIdx - 8;   /* skip repetitive patterns */
+        if (bestLength >= MINMATCH) {
+            U32 const mIndex = current - ((U32)*offsetPtr - ZSTD_REP_MOVE); (void)mIndex;
+            DEBUGLOG(8, "ZSTD_DUBT_findBestMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
+                        current, (U32)bestLength, (U32)*offsetPtr, mIndex);
+        }
+        return bestLength;
+    }
+}
+
+
+/** ZSTD_BtFindBestMatch() : Tree updater, providing best match */
+static size_t ZSTD_BtFindBestMatch (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls /* template */)
+{
+    DEBUGLOG(7, "ZSTD_BtFindBestMatch");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateDUBT(ms, cParams, ip, iLimit, mls);
+    return ZSTD_DUBT_findBestMatch(ms, cParams, ip, iLimit, offsetPtr, mls, 0);
+}
+
+
+static size_t ZSTD_BtFindBestMatch_selectMLS (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
+{
+    switch(cParams->searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_BtFindBestMatch(ms, cParams, ip, iLimit, offsetPtr, 4);
+    case 5 : return ZSTD_BtFindBestMatch(ms, cParams, ip, iLimit, offsetPtr, 5);
+    case 7 :
+    case 6 : return ZSTD_BtFindBestMatch(ms, cParams, ip, iLimit, offsetPtr, 6);
+    }
+}
+
+
+/** Tree updater, providing best match */
+static size_t ZSTD_BtFindBestMatch_extDict (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls)
+{
+    DEBUGLOG(7, "ZSTD_BtFindBestMatch_extDict");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateDUBT(ms, cParams, ip, iLimit, mls);
+    return ZSTD_DUBT_findBestMatch(ms, cParams, ip, iLimit, offsetPtr, mls, 1);
+}
+
+
+static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
+{
+    switch(cParams->searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_BtFindBestMatch_extDict(ms, cParams, ip, iLimit, offsetPtr, 4);
+    case 5 : return ZSTD_BtFindBestMatch_extDict(ms, cParams, ip, iLimit, offsetPtr, 5);
+    case 7 :
+    case 6 : return ZSTD_BtFindBestMatch_extDict(ms, cParams, ip, iLimit, offsetPtr, 6);
+    }
+}
+
+
+
+/* *********************************
+*  Hash Chain
+***********************************/
+#define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & mask]
+
+/* Update chains up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+static U32 ZSTD_insertAndFindFirstIndex_internal(
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, U32 const mls)
+{
+    U32* const hashTable  = ms->hashTable;
+    const U32 hashLog = cParams->hashLog;
+    U32* const chainTable = ms->chainTable;
+    const U32 chainMask = (1 << cParams->chainLog) - 1;
+    const BYTE* const base = ms->window.base;
+    const U32 target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+
+    while(idx < target) { /* catch up */
+        size_t const h = ZSTD_hashPtr(base+idx, hashLog, mls);
+        NEXT_IN_CHAIN(idx, chainMask) = hashTable[h];
+        hashTable[h] = idx;
+        idx++;
+    }
+
+    ms->nextToUpdate = target;
+    return hashTable[ZSTD_hashPtr(ip, hashLog, mls)];
+}
+
+U32 ZSTD_insertAndFindFirstIndex(
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip)
+{
+    return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, cParams->searchLength);
+}
+
+
+/* inlining is important to hardwire a hot branch (template emulation) */
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_HcFindBestMatch_generic (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls, const U32 extDict)
+{
+    U32* const chainTable = ms->chainTable;
+    const U32 chainSize = (1 << cParams->chainLog);
+    const U32 chainMask = chainSize-1;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const U32 lowLimit = ms->window.lowLimit;
+    const U32 current = (U32)(ip-base);
+    const U32 minChain = current > chainSize ? current - chainSize : 0;
+    U32 nbAttempts = 1U << cParams->searchLog;
+    size_t ml=4-1;
+
+    /* HC4 match finder */
+    U32 matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls);
+
+    for ( ; (matchIndex>lowLimit) & (nbAttempts>0) ; nbAttempts--) {
+        size_t currentMl=0;
+        if ((!extDict) || matchIndex >= dictLimit) {
+            const BYTE* const match = base + matchIndex;
+            if (match[ml] == ip[ml])   /* potentially better */
+                currentMl = ZSTD_count(ip, match, iLimit);
+        } else {
+            const BYTE* const match = dictBase + matchIndex;
+            assert(match+4 <= dictEnd);
+            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
+        }
+
+        /* save best solution */
+        if (currentMl > ml) {
+            ml = currentMl;
+            *offsetPtr = current - matchIndex + ZSTD_REP_MOVE;
+            if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+        }
+
+        if (matchIndex <= minChain) break;
+        matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
+    }
+
+    return ml;
+}
+
+
+FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_selectMLS (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
+{
+    switch(cParams->searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 4, 0);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 5, 0);
+    case 7 :
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 6, 0);
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* const offsetPtr)
+{
+    switch(cParams->searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 4, 1);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 5, 1);
+    case 7 :
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 6, 1);
+    }
+}
+
+
+/* *******************************
+*  Common parser - lazy strategy
+*********************************/
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_lazy_generic(
+                        ZSTD_matchState_t* ms, seqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        ZSTD_compressionParameters const* cParams,
+                        const void* src, size_t srcSize,
+                        const U32 searchMethod, const U32 depth)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base + ms->window.dictLimit;
+
+    typedef size_t (*searchMax_f)(
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
+    searchMax_f const searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS : ZSTD_HcFindBestMatch_selectMLS;
+    U32 offset_1 = rep[0], offset_2 = rep[1], savedOffset=0;
+
+    /* init */
+    ip += (ip==base);
+    ms->nextToUpdate3 = ms->nextToUpdate;
+    {   U32 const maxRep = (U32)(ip-base);
+        if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
+    }
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offset=0;
+        const BYTE* start=ip+1;
+
+        /* check repCode */
+        if ((offset_1>0) & (MEM_read32(ip+1) == MEM_read32(ip+1 - offset_1))) {
+            /* repcode : we take it */
+            matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            if (depth==0) goto _storeSequence;
+        }
+
+        /* first search (depth 0) */
+        {   size_t offsetFound = 99999999;
+            size_t const ml2 = searchMax(ms, cParams, ip, iend, &offsetFound);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offset=offsetFound;
+        }
+
+        if (matchLength < 4) {
+            ip += ((ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            ip ++;
+            if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                int const gain2 = (int)(mlRep * 3);
+                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                if ((mlRep >= 4) && (gain2 > gain1))
+                    matchLength = mlRep, offset = 0, start = ip;
+            }
+            {   size_t offset2=99999999;
+                size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offset = offset2, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                ip ++;
+                if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                    size_t const ml2 = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                    int const gain2 = (int)(ml2 * 4);
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                    if ((ml2 >= 4) && (gain2 > gain1))
+                        matchLength = ml2, offset = 0, start = ip;
+                }
+                {   size_t offset2=99999999;
+                    size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offset = offset2, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* NOTE:
+         * start[-offset+ZSTD_REP_MOVE-1] is undefined behavior.
+         * (-offset+ZSTD_REP_MOVE-1) is unsigned, and is added to start, which
+         * overflows the pointer, which is undefined behavior.
+         */
+        /* catch up */
+        if (offset) {
+            while ( ((start > anchor) & (start - (offset-ZSTD_REP_MOVE) > base))
+                 && (start[-1] == (start-(offset-ZSTD_REP_MOVE))[-1]) )  /* only search for offset within prefix */
+                { start--; matchLength++; }
+            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
+        }
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = start - anchor;
+            ZSTD_storeSeq(seqStore, litLength, anchor, (U32)offset, matchLength-MINMATCH);
+            anchor = ip = start + matchLength;
+        }
+
+        /* check immediate repcode */
+        while ( ((ip <= ilimit) & (offset_2>0))
+             && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
+            /* store sequence */
+            matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+            offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset; /* swap repcodes */
+            ZSTD_storeSeq(seqStore, 0, anchor, 0, matchLength-MINMATCH);
+            ip += matchLength;
+            anchor = ip;
+            continue;   /* faster when present ... (?) */
+    }   }
+
+    /* Save reps for next block */
+    rep[0] = offset_1 ? offset_1 : savedOffset;
+    rep[1] = offset_2 ? offset_2 : savedOffset;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_btlazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 1, 2);
+}
+
+size_t ZSTD_compressBlock_lazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 2);
+}
+
+size_t ZSTD_compressBlock_lazy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 1);
+}
+
+size_t ZSTD_compressBlock_greedy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 0);
+}
+
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_lazy_extDict_generic(
+                        ZSTD_matchState_t* ms, seqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        ZSTD_compressionParameters const* cParams,
+                        const void* src, size_t srcSize,
+                        const U32 searchMethod, const U32 depth)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base;
+    const U32 dictLimit = ms->window.dictLimit;
+    const U32 lowestIndex = ms->window.lowLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const dictEnd  = dictBase + dictLimit;
+    const BYTE* const dictStart  = dictBase + lowestIndex;
+
+    typedef size_t (*searchMax_f)(
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
+    searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS_extDict : ZSTD_HcFindBestMatch_extDict_selectMLS;
+
+    U32 offset_1 = rep[0], offset_2 = rep[1];
+
+    /* init */
+    ms->nextToUpdate3 = ms->nextToUpdate;
+    ip += (ip == prefixStart);
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offset=0;
+        const BYTE* start=ip+1;
+        U32 current = (U32)(ip-base);
+
+        /* check repCode */
+        {   const U32 repIndex = (U32)(current+1 - offset_1);
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))   /* intentional overflow */
+            if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                if (depth==0) goto _storeSequence;
+        }   }
+
+        /* first search (depth 0) */
+        {   size_t offsetFound = 99999999;
+            size_t const ml2 = searchMax(ms, cParams, ip, iend, &offsetFound);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offset=offsetFound;
+        }
+
+         if (matchLength < 4) {
+            ip += ((ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            ip ++;
+            current++;
+            /* check repCode */
+            if (offset) {
+                const U32 repIndex = (U32)(current - offset_1);
+                const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                const BYTE* const repMatch = repBase + repIndex;
+                if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                    /* repcode detected */
+                    const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                    size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                    int const gain2 = (int)(repLength * 3);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                    if ((repLength >= 4) && (gain2 > gain1))
+                        matchLength = repLength, offset = 0, start = ip;
+            }   }
+
+            /* search match, depth 1 */
+            {   size_t offset2=99999999;
+                size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offset = offset2, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                ip ++;
+                current++;
+                /* check repCode */
+                if (offset) {
+                    const U32 repIndex = (U32)(current - offset_1);
+                    const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                    const BYTE* const repMatch = repBase + repIndex;
+                    if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                    if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                        /* repcode detected */
+                        const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                        size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                        int const gain2 = (int)(repLength * 4);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                        if ((repLength >= 4) && (gain2 > gain1))
+                            matchLength = repLength, offset = 0, start = ip;
+                }   }
+
+                /* search match, depth 2 */
+                {   size_t offset2=99999999;
+                    size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offset = offset2, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* catch up */
+        if (offset) {
+            U32 const matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
+            const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex;
+            const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart;
+            while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
+            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
+        }
+
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = start - anchor;
+            ZSTD_storeSeq(seqStore, litLength, anchor, (U32)offset, matchLength-MINMATCH);
+            anchor = ip = start + matchLength;
+        }
+
+        /* check immediate repcode */
+        while (ip <= ilimit) {
+            const U32 repIndex = (U32)((ip-base) - offset_2);
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+            if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset;   /* swap offset history */
+                ZSTD_storeSeq(seqStore, 0, anchor, 0, matchLength-MINMATCH);
+                ip += matchLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+            }
+            break;
+    }   }
+
+    /* Save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_greedy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 0);
+}
+
+size_t ZSTD_compressBlock_lazy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 1);
+}
+
+size_t ZSTD_compressBlock_lazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 2);
+}
+
+size_t ZSTD_compressBlock_btlazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 1, 2);
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_lazy.h	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,56 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_LAZY_H
+#define ZSTD_LAZY_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "zstd_compress_internal.h"
+
+U32 ZSTD_insertAndFindFirstIndex(
+        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+        const BYTE* ip);
+
+void ZSTD_preserveUnsortedMark (U32* const table, U32 const size, U32 const reducerValue);  /*! used in ZSTD_reduceIndex(). pre-emptively increase value of ZSTD_DUBT_UNSORTED_MARK */
+
+size_t ZSTD_compressBlock_btlazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_greedy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btlazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_LAZY_H */
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_ldm.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,653 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ */
+
+#include "zstd_ldm.h"
+
+#include "zstd_fast.h"          /* ZSTD_fillHashTable() */
+#include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */
+
+#define LDM_BUCKET_SIZE_LOG 3
+#define LDM_MIN_MATCH_LENGTH 64
+#define LDM_HASH_RLOG 7
+#define LDM_HASH_CHAR_OFFSET 10
+
+void ZSTD_ldm_adjustParameters(ldmParams_t* params,
+                               ZSTD_compressionParameters const* cParams)
+{
+    U32 const windowLog = cParams->windowLog;
+    ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
+    DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
+    if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
+    if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
+    if (cParams->strategy >= ZSTD_btopt) {
+      /* Get out of the way of the optimal parser */
+      U32 const minMatch = MAX(cParams->targetLength, params->minMatchLength);
+      assert(minMatch >= ZSTD_LDM_MINMATCH_MIN);
+      assert(minMatch <= ZSTD_LDM_MINMATCH_MAX);
+      params->minMatchLength = minMatch;
+    }
+    if (params->hashLog == 0) {
+        params->hashLog = MAX(ZSTD_HASHLOG_MIN, windowLog - LDM_HASH_RLOG);
+        assert(params->hashLog <= ZSTD_HASHLOG_MAX);
+    }
+    if (params->hashEveryLog == 0) {
+        params->hashEveryLog =
+                windowLog < params->hashLog ? 0 : windowLog - params->hashLog;
+    }
+    params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
+}
+
+size_t ZSTD_ldm_getTableSize(ldmParams_t params)
+{
+    size_t const ldmHSize = ((size_t)1) << params.hashLog;
+    size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
+    size_t const ldmBucketSize =
+        ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
+    size_t const totalSize = ldmBucketSize + ldmHSize * sizeof(ldmEntry_t);
+    return params.enableLdm ? totalSize : 0;
+}
+
+size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
+{
+    return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
+}
+
+/** ZSTD_ldm_getSmallHash() :
+ *  numBits should be <= 32
+ *  If numBits==0, returns 0.
+ *  @return : the most significant numBits of value. */
+static U32 ZSTD_ldm_getSmallHash(U64 value, U32 numBits)
+{
+    assert(numBits <= 32);
+    return numBits == 0 ? 0 : (U32)(value >> (64 - numBits));
+}
+
+/** ZSTD_ldm_getChecksum() :
+ *  numBitsToDiscard should be <= 32
+ *  @return : the next most significant 32 bits after numBitsToDiscard */
+static U32 ZSTD_ldm_getChecksum(U64 hash, U32 numBitsToDiscard)
+{
+    assert(numBitsToDiscard <= 32);
+    return (hash >> (64 - 32 - numBitsToDiscard)) & 0xFFFFFFFF;
+}
+
+/** ZSTD_ldm_getTag() ;
+ *  Given the hash, returns the most significant numTagBits bits
+ *  after (32 + hbits) bits.
+ *
+ *  If there are not enough bits remaining, return the last
+ *  numTagBits bits. */
+static U32 ZSTD_ldm_getTag(U64 hash, U32 hbits, U32 numTagBits)
+{
+    assert(numTagBits < 32 && hbits <= 32);
+    if (32 - hbits < numTagBits) {
+        return hash & (((U32)1 << numTagBits) - 1);
+    } else {
+        return (hash >> (32 - hbits - numTagBits)) & (((U32)1 << numTagBits) - 1);
+    }
+}
+
+/** ZSTD_ldm_getBucket() :
+ *  Returns a pointer to the start of the bucket associated with hash. */
+static ldmEntry_t* ZSTD_ldm_getBucket(
+        ldmState_t* ldmState, size_t hash, ldmParams_t const ldmParams)
+{
+    return ldmState->hashTable + (hash << ldmParams.bucketSizeLog);
+}
+
+/** ZSTD_ldm_insertEntry() :
+ *  Insert the entry with corresponding hash into the hash table */
+static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
+                                 size_t const hash, const ldmEntry_t entry,
+                                 ldmParams_t const ldmParams)
+{
+    BYTE* const bucketOffsets = ldmState->bucketOffsets;
+    *(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + bucketOffsets[hash]) = entry;
+    bucketOffsets[hash]++;
+    bucketOffsets[hash] &= ((U32)1 << ldmParams.bucketSizeLog) - 1;
+}
+
+/** ZSTD_ldm_makeEntryAndInsertByTag() :
+ *
+ *  Gets the small hash, checksum, and tag from the rollingHash.
+ *
+ *  If the tag matches (1 << ldmParams.hashEveryLog)-1, then
+ *  creates an ldmEntry from the offset, and inserts it into the hash table.
+ *
+ *  hBits is the length of the small hash, which is the most significant hBits
+ *  of rollingHash. The checksum is the next 32 most significant bits, followed
+ *  by ldmParams.hashEveryLog bits that make up the tag. */
+static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
+                                             U64 const rollingHash,
+                                             U32 const hBits,
+                                             U32 const offset,
+                                             ldmParams_t const ldmParams)
+{
+    U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashEveryLog);
+    U32 const tagMask = ((U32)1 << ldmParams.hashEveryLog) - 1;
+    if (tag == tagMask) {
+        U32 const hash = ZSTD_ldm_getSmallHash(rollingHash, hBits);
+        U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
+        ldmEntry_t entry;
+        entry.offset = offset;
+        entry.checksum = checksum;
+        ZSTD_ldm_insertEntry(ldmState, hash, entry, ldmParams);
+    }
+}
+
+/** ZSTD_ldm_getRollingHash() :
+ *  Get a 64-bit hash using the first len bytes from buf.
+ *
+ *  Giving bytes s = s_1, s_2, ... s_k, the hash is defined to be
+ *  H(s) = s_1*(a^(k-1)) + s_2*(a^(k-2)) + ... + s_k*(a^0)
+ *
+ *  where the constant a is defined to be prime8bytes.
+ *
+ *  The implementation adds an offset to each byte, so
+ *  H(s) = (s_1 + HASH_CHAR_OFFSET)*(a^(k-1)) + ... */
+static U64 ZSTD_ldm_getRollingHash(const BYTE* buf, U32 len)
+{
+    U64 ret = 0;
+    U32 i;
+    for (i = 0; i < len; i++) {
+        ret *= prime8bytes;
+        ret += buf[i] + LDM_HASH_CHAR_OFFSET;
+    }
+    return ret;
+}
+
+/** ZSTD_ldm_ipow() :
+ *  Return base^exp. */
+static U64 ZSTD_ldm_ipow(U64 base, U64 exp)
+{
+    U64 ret = 1;
+    while (exp) {
+        if (exp & 1) { ret *= base; }
+        exp >>= 1;
+        base *= base;
+    }
+    return ret;
+}
+
+U64 ZSTD_ldm_getHashPower(U32 minMatchLength) {
+    DEBUGLOG(4, "ZSTD_ldm_getHashPower: mml=%u", minMatchLength);
+    assert(minMatchLength >= ZSTD_LDM_MINMATCH_MIN);
+    return ZSTD_ldm_ipow(prime8bytes, minMatchLength - 1);
+}
+
+/** ZSTD_ldm_updateHash() :
+ *  Updates hash by removing toRemove and adding toAdd. */
+static U64 ZSTD_ldm_updateHash(U64 hash, BYTE toRemove, BYTE toAdd, U64 hashPower)
+{
+    hash -= ((toRemove + LDM_HASH_CHAR_OFFSET) * hashPower);
+    hash *= prime8bytes;
+    hash += toAdd + LDM_HASH_CHAR_OFFSET;
+    return hash;
+}
+
+/** ZSTD_ldm_countBackwardsMatch() :
+ *  Returns the number of bytes that match backwards before pIn and pMatch.
+ *
+ *  We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
+static size_t ZSTD_ldm_countBackwardsMatch(
+            const BYTE* pIn, const BYTE* pAnchor,
+            const BYTE* pMatch, const BYTE* pBase)
+{
+    size_t matchLength = 0;
+    while (pIn > pAnchor && pMatch > pBase && pIn[-1] == pMatch[-1]) {
+        pIn--;
+        pMatch--;
+        matchLength++;
+    }
+    return matchLength;
+}
+
+/** ZSTD_ldm_fillFastTables() :
+ *
+ *  Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
+ *  This is similar to ZSTD_loadDictionaryContent.
+ *
+ *  The tables for the other strategies are filled within their
+ *  block compressors. */
+static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
+                                      ZSTD_compressionParameters const* cParams,
+                                      void const* end)
+{
+    const BYTE* const iend = (const BYTE*)end;
+
+    switch(cParams->strategy)
+    {
+    case ZSTD_fast:
+        ZSTD_fillHashTable(ms, cParams, iend);
+        ms->nextToUpdate = (U32)(iend - ms->window.base);
+        break;
+
+    case ZSTD_dfast:
+        ZSTD_fillDoubleHashTable(ms, cParams, iend);
+        ms->nextToUpdate = (U32)(iend - ms->window.base);
+        break;
+
+    case ZSTD_greedy:
+    case ZSTD_lazy:
+    case ZSTD_lazy2:
+    case ZSTD_btlazy2:
+    case ZSTD_btopt:
+    case ZSTD_btultra:
+        break;
+    default:
+        assert(0);  /* not possible : not a valid strategy id */
+    }
+
+    return 0;
+}
+
+/** ZSTD_ldm_fillLdmHashTable() :
+ *
+ *  Fills hashTable from (lastHashed + 1) to iend (non-inclusive).
+ *  lastHash is the rolling hash that corresponds to lastHashed.
+ *
+ *  Returns the rolling hash corresponding to position iend-1. */
+static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
+                                     U64 lastHash, const BYTE* lastHashed,
+                                     const BYTE* iend, const BYTE* base,
+                                     U32 hBits, ldmParams_t const ldmParams)
+{
+    U64 rollingHash = lastHash;
+    const BYTE* cur = lastHashed + 1;
+
+    while (cur < iend) {
+        rollingHash = ZSTD_ldm_updateHash(rollingHash, cur[-1],
+                                          cur[ldmParams.minMatchLength-1],
+                                          state->hashPower);
+        ZSTD_ldm_makeEntryAndInsertByTag(state,
+                                         rollingHash, hBits,
+                                         (U32)(cur - base), ldmParams);
+        ++cur;
+    }
+    return rollingHash;
+}
+
+
+/** ZSTD_ldm_limitTableUpdate() :
+ *
+ *  Sets cctx->nextToUpdate to a position corresponding closer to anchor
+ *  if it is far way
+ *  (after a long match, only update tables a limited amount). */
+static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
+{
+    U32 const current = (U32)(anchor - ms->window.base);
+    if (current > ms->nextToUpdate + 1024) {
+        ms->nextToUpdate =
+            current - MIN(512, current - ms->nextToUpdate - 1024);
+    }
+}
+
+static size_t ZSTD_ldm_generateSequences_internal(
+        ldmState_t* ldmState, rawSeqStore_t* rawSeqStore,
+        ldmParams_t const* params, void const* src, size_t srcSize)
+{
+    /* LDM parameters */
+    int const extDict = ZSTD_window_hasExtDict(ldmState->window);
+    U32 const minMatchLength = params->minMatchLength;
+    U64 const hashPower = ldmState->hashPower;
+    U32 const hBits = params->hashLog - params->bucketSizeLog;
+    U32 const ldmBucketSize = 1U << params->bucketSizeLog;
+    U32 const hashEveryLog = params->hashEveryLog;
+    U32 const ldmTagMask = (1U << params->hashEveryLog) - 1;
+    /* Prefix and extDict parameters */
+    U32 const dictLimit = ldmState->window.dictLimit;
+    U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
+    BYTE const* const base = ldmState->window.base;
+    BYTE const* const dictBase = extDict ? ldmState->window.dictBase : NULL;
+    BYTE const* const dictStart = extDict ? dictBase + lowestIndex : NULL;
+    BYTE const* const dictEnd = extDict ? dictBase + dictLimit : NULL;
+    BYTE const* const lowPrefixPtr = base + dictLimit;
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    BYTE const* const ilimit = iend - MAX(minMatchLength, HASH_READ_SIZE);
+    /* Input positions */
+    BYTE const* anchor = istart;
+    BYTE const* ip = istart;
+    /* Rolling hash */
+    BYTE const* lastHashed = NULL;
+    U64 rollingHash = 0;
+
+    while (ip <= ilimit) {
+        size_t mLength;
+        U32 const current = (U32)(ip - base);
+        size_t forwardMatchLength = 0, backwardMatchLength = 0;
+        ldmEntry_t* bestEntry = NULL;
+        if (ip != istart) {
+            rollingHash = ZSTD_ldm_updateHash(rollingHash, lastHashed[0],
+                                              lastHashed[minMatchLength],
+                                              hashPower);
+        } else {
+            rollingHash = ZSTD_ldm_getRollingHash(ip, minMatchLength);
+        }
+        lastHashed = ip;
+
+        /* Do not insert and do not look for a match */
+        if (ZSTD_ldm_getTag(rollingHash, hBits, hashEveryLog) != ldmTagMask) {
+           ip++;
+           continue;
+        }
+
+        /* Get the best entry and compute the match lengths */
+        {
+            ldmEntry_t* const bucket =
+                ZSTD_ldm_getBucket(ldmState,
+                                   ZSTD_ldm_getSmallHash(rollingHash, hBits),
+                                   *params);
+            ldmEntry_t* cur;
+            size_t bestMatchLength = 0;
+            U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
+
+            for (cur = bucket; cur < bucket + ldmBucketSize; ++cur) {
+                size_t curForwardMatchLength, curBackwardMatchLength,
+                       curTotalMatchLength;
+                if (cur->checksum != checksum || cur->offset <= lowestIndex) {
+                    continue;
+                }
+                if (extDict) {
+                    BYTE const* const curMatchBase =
+                        cur->offset < dictLimit ? dictBase : base;
+                    BYTE const* const pMatch = curMatchBase + cur->offset;
+                    BYTE const* const matchEnd =
+                        cur->offset < dictLimit ? dictEnd : iend;
+                    BYTE const* const lowMatchPtr =
+                        cur->offset < dictLimit ? dictStart : lowPrefixPtr;
+
+                    curForwardMatchLength = ZSTD_count_2segments(
+                                                ip, pMatch, iend,
+                                                matchEnd, lowPrefixPtr);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength =
+                        ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
+                                                     lowMatchPtr);
+                    curTotalMatchLength = curForwardMatchLength +
+                                          curBackwardMatchLength;
+                } else { /* !extDict */
+                    BYTE const* const pMatch = base + cur->offset;
+                    curForwardMatchLength = ZSTD_count(ip, pMatch, iend);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength =
+                        ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
+                                                     lowPrefixPtr);
+                    curTotalMatchLength = curForwardMatchLength +
+                                          curBackwardMatchLength;
+                }
+
+                if (curTotalMatchLength > bestMatchLength) {
+                    bestMatchLength = curTotalMatchLength;
+                    forwardMatchLength = curForwardMatchLength;
+                    backwardMatchLength = curBackwardMatchLength;
+                    bestEntry = cur;
+                }
+            }
+        }
+
+        /* No match found -- continue searching */
+        if (bestEntry == NULL) {
+            ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash,
+                                             hBits, current,
+                                             *params);
+            ip++;
+            continue;
+        }
+
+        /* Match found */
+        mLength = forwardMatchLength + backwardMatchLength;
+        ip -= backwardMatchLength;
+
+        {
+            /* Store the sequence:
+             * ip = current - backwardMatchLength
+             * The match is at (bestEntry->offset - backwardMatchLength)
+             */
+            U32 const matchIndex = bestEntry->offset;
+            U32 const offset = current - matchIndex;
+            rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
+
+            /* Out of sequence storage */
+            if (rawSeqStore->size == rawSeqStore->capacity)
+                return ERROR(dstSize_tooSmall);
+            seq->litLength = (U32)(ip - anchor);
+            seq->matchLength = (U32)mLength;
+            seq->offset = offset;
+            rawSeqStore->size++;
+        }
+
+        /* Insert the current entry into the hash table */
+        ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
+                                         (U32)(lastHashed - base),
+                                         *params);
+
+        assert(ip + backwardMatchLength == lastHashed);
+
+        /* Fill the hash table from lastHashed+1 to ip+mLength*/
+        /* Heuristic: don't need to fill the entire table at end of block */
+        if (ip + mLength <= ilimit) {
+            rollingHash = ZSTD_ldm_fillLdmHashTable(
+                              ldmState, rollingHash, lastHashed,
+                              ip + mLength, base, hBits, *params);
+            lastHashed = ip + mLength - 1;
+        }
+        ip += mLength;
+        anchor = ip;
+    }
+    return iend - anchor;
+}
+
+/*! ZSTD_ldm_reduceTable() :
+ *  reduce table indexes by `reducerValue` */
+static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
+                                 U32 const reducerValue)
+{
+    U32 u;
+    for (u = 0; u < size; u++) {
+        if (table[u].offset < reducerValue) table[u].offset = 0;
+        else table[u].offset -= reducerValue;
+    }
+}
+
+size_t ZSTD_ldm_generateSequences(
+        ldmState_t* ldmState, rawSeqStore_t* sequences,
+        ldmParams_t const* params, void const* src, size_t srcSize)
+{
+    U32 const maxDist = 1U << params->windowLog;
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    size_t const kMaxChunkSize = 1 << 20;
+    size_t const nbChunks = (srcSize / kMaxChunkSize) + ((srcSize % kMaxChunkSize) != 0);
+    size_t chunk;
+    size_t leftoverSize = 0;
+
+    assert(ZSTD_CHUNKSIZE_MAX >= kMaxChunkSize);
+    /* Check that ZSTD_window_update() has been called for this chunk prior
+     * to passing it to this function.
+     */
+    assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
+    /* The input could be very large (in zstdmt), so it must be broken up into
+     * chunks to enforce the maximmum distance and handle overflow correction.
+     */
+    assert(sequences->pos <= sequences->size);
+    assert(sequences->size <= sequences->capacity);
+    for (chunk = 0; chunk < nbChunks && sequences->size < sequences->capacity; ++chunk) {
+        BYTE const* const chunkStart = istart + chunk * kMaxChunkSize;
+        size_t const remaining = (size_t)(iend - chunkStart);
+        BYTE const *const chunkEnd =
+            (remaining < kMaxChunkSize) ? iend : chunkStart + kMaxChunkSize;
+        size_t const chunkSize = chunkEnd - chunkStart;
+        size_t newLeftoverSize;
+        size_t const prevSize = sequences->size;
+
+        assert(chunkStart < iend);
+        /* 1. Perform overflow correction if necessary. */
+        if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
+            U32 const ldmHSize = 1U << params->hashLog;
+            U32 const correction = ZSTD_window_correctOverflow(
+                &ldmState->window, /* cycleLog */ 0, maxDist, src);
+            ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
+        }
+        /* 2. We enforce the maximum offset allowed.
+         *
+         * kMaxChunkSize should be small enough that we don't lose too much of
+         * the window through early invalidation.
+         * TODO: * Test the chunk size.
+         *       * Try invalidation after the sequence generation and test the
+         *         the offset against maxDist directly.
+         */
+        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, NULL);
+        /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
+        newLeftoverSize = ZSTD_ldm_generateSequences_internal(
+            ldmState, sequences, params, chunkStart, chunkSize);
+        if (ZSTD_isError(newLeftoverSize))
+            return newLeftoverSize;
+        /* 4. We add the leftover literals from previous iterations to the first
+         *    newly generated sequence, or add the `newLeftoverSize` if none are
+         *    generated.
+         */
+        /* Prepend the leftover literals from the last call */
+        if (prevSize < sequences->size) {
+            sequences->seq[prevSize].litLength += (U32)leftoverSize;
+            leftoverSize = newLeftoverSize;
+        } else {
+            assert(newLeftoverSize == chunkSize);
+            leftoverSize += chunkSize;
+        }
+    }
+    return 0;
+}
+
+void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch) {
+    while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
+        if (srcSize <= seq->litLength) {
+            /* Skip past srcSize literals */
+            seq->litLength -= (U32)srcSize;
+            return;
+        }
+        srcSize -= seq->litLength;
+        seq->litLength = 0;
+        if (srcSize < seq->matchLength) {
+            /* Skip past the first srcSize of the match */
+            seq->matchLength -= (U32)srcSize;
+            if (seq->matchLength < minMatch) {
+                /* The match is too short, omit it */
+                if (rawSeqStore->pos + 1 < rawSeqStore->size) {
+                    seq[1].litLength += seq[0].matchLength;
+                }
+                rawSeqStore->pos++;
+            }
+            return;
+        }
+        srcSize -= seq->matchLength;
+        seq->matchLength = 0;
+        rawSeqStore->pos++;
+    }
+}
+
+/**
+ * If the sequence length is longer than remaining then the sequence is split
+ * between this block and the next.
+ *
+ * Returns the current sequence to handle, or if the rest of the block should
+ * be literals, it returns a sequence with offset == 0.
+ */
+static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
+                                 U32 const remaining, U32 const minMatch)
+{
+    rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
+    assert(sequence.offset > 0);
+    /* Likely: No partial sequence */
+    if (remaining >= sequence.litLength + sequence.matchLength) {
+        rawSeqStore->pos++;
+        return sequence;
+    }
+    /* Cut the sequence short (offset == 0 ==> rest is literals). */
+    if (remaining <= sequence.litLength) {
+        sequence.offset = 0;
+    } else if (remaining < sequence.litLength + sequence.matchLength) {
+        sequence.matchLength = remaining - sequence.litLength;
+        if (sequence.matchLength < minMatch) {
+            sequence.offset = 0;
+        }
+    }
+    /* Skip past `remaining` bytes for the future sequences. */
+    ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
+    return sequence;
+}
+
+size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
+    ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+    ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
+    int const extDict)
+{
+    unsigned const minMatch = cParams->searchLength;
+    ZSTD_blockCompressor const blockCompressor =
+        ZSTD_selectBlockCompressor(cParams->strategy, extDict);
+    BYTE const* const base = ms->window.base;
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    /* Input positions */
+    BYTE const* ip = istart;
+
+    assert(rawSeqStore->pos <= rawSeqStore->size);
+    assert(rawSeqStore->size <= rawSeqStore->capacity);
+    /* Loop through each sequence and apply the block compressor to the lits */
+    while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
+        /* maybeSplitSequence updates rawSeqStore->pos */
+        rawSeq const sequence = maybeSplitSequence(rawSeqStore,
+                                                   (U32)(iend - ip), minMatch);
+        int i;
+        /* End signal */
+        if (sequence.offset == 0)
+            break;
+
+        assert(sequence.offset <= (1U << cParams->windowLog));
+        assert(ip + sequence.litLength + sequence.matchLength <= iend);
+
+        /* Fill tables for block compressor */
+        ZSTD_ldm_limitTableUpdate(ms, ip);
+        ZSTD_ldm_fillFastTables(ms, cParams, ip);
+        /* Run the block compressor */
+        {
+            size_t const newLitLength =
+                blockCompressor(ms, seqStore, rep, cParams, ip,
+                                sequence.litLength);
+            ip += sequence.litLength;
+            ms->nextToUpdate = (U32)(ip - base);
+            /* Update the repcodes */
+            for (i = ZSTD_REP_NUM - 1; i > 0; i--)
+                rep[i] = rep[i-1];
+            rep[0] = sequence.offset;
+            /* Store the sequence */
+            ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength,
+                          sequence.offset + ZSTD_REP_MOVE,
+                          sequence.matchLength - MINMATCH);
+            ip += sequence.matchLength;
+        }
+    }
+    /* Fill the tables for the block compressor */
+    ZSTD_ldm_limitTableUpdate(ms, ip);
+    ZSTD_ldm_fillFastTables(ms, cParams, ip);
+    /* Compress the last literals */
+    {
+        size_t const lastLiterals = blockCompressor(ms, seqStore, rep, cParams,
+                                                    ip, iend - ip);
+        ms->nextToUpdate = (U32)(iend - base);
+        return lastLiterals;
+    }
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_ldm.h	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ */
+
+#ifndef ZSTD_LDM_H
+#define ZSTD_LDM_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "zstd_compress_internal.h"   /* ldmParams_t, U32 */
+#include "zstd.h"   /* ZSTD_CCtx, size_t */
+
+/*-*************************************
+*  Long distance matching
+***************************************/
+
+#define ZSTD_LDM_DEFAULT_WINDOW_LOG ZSTD_WINDOWLOG_DEFAULTMAX
+
+/**
+ * ZSTD_ldm_generateSequences():
+ *
+ * Generates the sequences using the long distance match finder.
+ * Generates long range matching sequences in `sequences`, which parse a prefix
+ * of the source. `sequences` must be large enough to store every sequence,
+ * which can be checked with `ZSTD_ldm_getMaxNbSeq()`.
+ * @returns 0 or an error code.
+ *
+ * NOTE: The user must have called ZSTD_window_update() for all of the input
+ * they have, even if they pass it to ZSTD_ldm_generateSequences() in chunks.
+ * NOTE: This function returns an error if it runs out of space to store
+ *       sequences.
+ */
+size_t ZSTD_ldm_generateSequences(
+            ldmState_t* ldms, rawSeqStore_t* sequences,
+            ldmParams_t const* params, void const* src, size_t srcSize);
+
+/**
+ * ZSTD_ldm_blockCompress():
+ *
+ * Compresses a block using the predefined sequences, along with a secondary
+ * block compressor. The literals section of every sequence is passed to the
+ * secondary block compressor, and those sequences are interspersed with the
+ * predefined sequences. Returns the length of the last literals.
+ * Updates `rawSeqStore.pos` to indicate how many sequences have been consumed.
+ * `rawSeqStore.seq` may also be updated to split the last sequence between two
+ * blocks.
+ * @return The length of the last literals.
+ *
+ * NOTE: The source must be at most the maximum block size, but the predefined
+ * sequences can be any size, and may be longer than the block. In the case that
+ * they are longer than the block, the last sequences may need to be split into
+ * two. We handle that case correctly, and update `rawSeqStore` appropriately.
+ * NOTE: This function does not return any errors.
+ */
+size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
+            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+            ZSTD_compressionParameters const* cParams,
+            void const* src, size_t srcSize,
+            int const extDict);
+
+/**
+ * ZSTD_ldm_skipSequences():
+ *
+ * Skip past `srcSize` bytes worth of sequences in `rawSeqStore`.
+ * Avoids emitting matches less than `minMatch` bytes.
+ * Must be called for data with is not passed to ZSTD_ldm_blockCompress().
+ */
+void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize,
+    U32 const minMatch);
+
+
+/** ZSTD_ldm_getTableSize() :
+ *  Estimate the space needed for long distance matching tables or 0 if LDM is
+ *  disabled.
+ */
+size_t ZSTD_ldm_getTableSize(ldmParams_t params);
+
+/** ZSTD_ldm_getSeqSpace() :
+ *  Return an upper bound on the number of sequences that can be produced by
+ *  the long distance matcher, or 0 if LDM is disabled.
+ */
+size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize);
+
+/** ZSTD_ldm_getTableSize() :
+ *  Return prime8bytes^(minMatchLength-1) */
+U64 ZSTD_ldm_getHashPower(U32 minMatchLength);
+
+/** ZSTD_ldm_adjustParameters() :
+ *  If the params->hashEveryLog is not set, set it to its default value based on
+ *  windowLog and params->hashLog.
+ *
+ *  Ensures that params->bucketSizeLog is <= params->hashLog (setting it to
+ *  params->hashLog if it is not).
+ *
+ *  Ensures that the minMatchLength >= targetLength during optimal parsing.
+ */
+void ZSTD_ldm_adjustParameters(ldmParams_t* params,
+                               ZSTD_compressionParameters const* cParams);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_FAST_H */
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/compress/zstd_opt.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,923 @@
+/*
+ * Copyright (c) 2016-present, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_opt.h"
+
+
+#define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats. Also used for matchSum (?) */
+#define ZSTD_FREQ_DIV       4   /* log factor when using previous stats to init next stats */
+#define ZSTD_MAX_PRICE     (1<<30)
+
+
+/*-*************************************
+*  Price functions for optimal parser
+***************************************/
+static void ZSTD_setLog2Prices(optState_t* optPtr)
+{
+    optPtr->log2litSum = ZSTD_highbit32(optPtr->litSum+1);
+    optPtr->log2litLengthSum = ZSTD_highbit32(optPtr->litLengthSum+1);
+    optPtr->log2matchLengthSum = ZSTD_highbit32(optPtr->matchLengthSum+1);
+    optPtr->log2offCodeSum = ZSTD_highbit32(optPtr->offCodeSum+1);
+}
+
+
+static void ZSTD_rescaleFreqs(optState_t* const optPtr,
+                              const BYTE* const src, size_t const srcSize)
+{
+    optPtr->staticPrices = 0;
+
+    if (optPtr->litLengthSum == 0) {  /* first init */
+        unsigned u;
+        if (srcSize <= 1024) optPtr->staticPrices = 1;
+
+        assert(optPtr->litFreq!=NULL);
+        for (u=0; u<=MaxLit; u++)
+            optPtr->litFreq[u] = 0;
+        for (u=0; u<srcSize; u++)
+            optPtr->litFreq[src[u]]++;
+        optPtr->litSum = 0;
+        for (u=0; u<=MaxLit; u++) {
+            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u] >> ZSTD_FREQ_DIV);
+            optPtr->litSum += optPtr->litFreq[u];
+        }
+
+        for (u=0; u<=MaxLL; u++)
+            optPtr->litLengthFreq[u] = 1;
+        optPtr->litLengthSum = MaxLL+1;
+        for (u=0; u<=MaxML; u++)
+            optPtr->matchLengthFreq[u] = 1;
+        optPtr->matchLengthSum = MaxML+1;
+        for (u=0; u<=MaxOff; u++)
+            optPtr->offCodeFreq[u] = 1;
+        optPtr->offCodeSum = (MaxOff+1);
+
+    } else {
+        unsigned u;
+
+        optPtr->litSum = 0;
+        for (u=0; u<=MaxLit; u++) {
+            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u] >> (ZSTD_FREQ_DIV+1));
+            optPtr->litSum += optPtr->litFreq[u];
+        }
+        optPtr->litLengthSum = 0;
+        for (u=0; u<=MaxLL; u++) {
+            optPtr->litLengthFreq[u] = 1 + (optPtr->litLengthFreq[u]>>(ZSTD_FREQ_DIV+1));
+            optPtr->litLengthSum += optPtr->litLengthFreq[u];
+        }
+        optPtr->matchLengthSum = 0;
+        for (u=0; u<=MaxML; u++) {
+            optPtr->matchLengthFreq[u] = 1 + (optPtr->matchLengthFreq[u]>>ZSTD_FREQ_DIV);
+            optPtr->matchLengthSum += optPtr->matchLengthFreq[u];
+        }
+        optPtr->offCodeSum = 0;
+        for (u=0; u<=MaxOff; u++) {
+            optPtr->offCodeFreq[u] = 1 + (optPtr->offCodeFreq[u]>>ZSTD_FREQ_DIV);
+            optPtr->offCodeSum += optPtr->offCodeFreq[u];
+        }
+    }
+
+    ZSTD_setLog2Prices(optPtr);
+}
+
+
+/* ZSTD_rawLiteralsCost() :
+ * cost of literals (only) in given segment (which length can be null)
+ * does not include cost of literalLength symbol */
+static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
+                                const optState_t* const optPtr)
+{
+    if (optPtr->staticPrices) return (litLength*6);  /* 6 bit per literal - no statistic used */
+    if (litLength == 0) return 0;
+
+    /* literals */
+    {   U32 u;
+        U32 cost = litLength * optPtr->log2litSum;
+        for (u=0; u < litLength; u++)
+            cost -= ZSTD_highbit32(optPtr->litFreq[literals[u]]+1);
+        return cost;
+    }
+}
+
+/* ZSTD_litLengthPrice() :
+ * cost of literalLength symbol */
+static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr)
+{
+    if (optPtr->staticPrices) return ZSTD_highbit32((U32)litLength+1);
+
+    /* literal Length */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        U32 const price = LL_bits[llCode] + optPtr->log2litLengthSum - ZSTD_highbit32(optPtr->litLengthFreq[llCode]+1);
+        return price;
+    }
+}
+
+/* ZSTD_litLengthPrice() :
+ * cost of the literal part of a sequence,
+ * including literals themselves, and literalLength symbol */
+static U32 ZSTD_fullLiteralsCost(const BYTE* const literals, U32 const litLength,
+                                 const optState_t* const optPtr)
+{
+    return ZSTD_rawLiteralsCost(literals, litLength, optPtr)
+         + ZSTD_litLengthPrice(litLength, optPtr);
+}
+
+/* ZSTD_litLengthContribution() :
+ * @return ( cost(litlength) - cost(0) )
+ * this value can then be added to rawLiteralsCost()
+ * to provide a cost which is directly comparable to a match ending at same position */
+static int ZSTD_litLengthContribution(U32 const litLength, const optState_t* const optPtr)
+{
+    if (optPtr->staticPrices) return ZSTD_highbit32(litLength+1);
+
+    /* literal Length */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        int const contribution = LL_bits[llCode]
+                        + ZSTD_highbit32(optPtr->litLengthFreq[0]+1)
+                        - ZSTD_highbit32(optPtr->litLengthFreq[llCode]+1);
+#if 1
+        return contribution;
+#else
+        return MAX(0, contribution); /* sometimes better, sometimes not ... */
+#endif
+    }
+}
+
+/* ZSTD_literalsContribution() :
+ * creates a fake cost for the literals part of a sequence
+ * which can be compared to the ending cost of a match
+ * should a new match start at this position */
+static int ZSTD_literalsContribution(const BYTE* const literals, U32 const litLength,
+                                     const optState_t* const optPtr)
+{
+    int const contribution = ZSTD_rawLiteralsCost(literals, litLength, optPtr)
+                           + ZSTD_litLengthContribution(litLength, optPtr);
+    return contribution;
+}
+
+/* ZSTD_getMatchPrice() :
+ * Provides the cost of the match part (offset + matchLength) of a sequence
+ * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
+ * optLevel: when <2, favors small offset for decompression speed (improved cache efficiency) */
+FORCE_INLINE_TEMPLATE U32 ZSTD_getMatchPrice(
+                                    U32 const offset, U32 const matchLength,
+                                    const optState_t* const optPtr,
+                                    int const optLevel)
+{
+    U32 price;
+    U32 const offCode = ZSTD_highbit32(offset+1);
+    U32 const mlBase = matchLength - MINMATCH;
+    assert(matchLength >= MINMATCH);
+
+    if (optPtr->staticPrices)  /* fixed scheme, do not use statistics */
+        return ZSTD_highbit32((U32)mlBase+1) + 16 + offCode;
+
+    price = offCode + optPtr->log2offCodeSum - ZSTD_highbit32(optPtr->offCodeFreq[offCode]+1);
+    if ((optLevel<2) /*static*/ && offCode >= 20) price += (offCode-19)*2; /* handicap for long distance offsets, favor decompression speed */
+
+    /* match Length */
+    {   U32 const mlCode = ZSTD_MLcode(mlBase);
+        price += ML_bits[mlCode] + optPtr->log2matchLengthSum - ZSTD_highbit32(optPtr->matchLengthFreq[mlCode]+1);
+    }
+
+    DEBUGLOG(8, "ZSTD_getMatchPrice(ml:%u) = %u", matchLength, price);
+    return price;
+}
+
+static void ZSTD_updateStats(optState_t* const optPtr,
+                             U32 litLength, const BYTE* literals,
+                             U32 offsetCode, U32 matchLength)
+{
+    /* literals */
+    {   U32 u;
+        for (u=0; u < litLength; u++)
+            optPtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
+        optPtr->litSum += litLength*ZSTD_LITFREQ_ADD;
+    }
+
+    /* literal Length */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        optPtr->litLengthFreq[llCode]++;
+        optPtr->litLengthSum++;
+    }
+
+    /* match offset code (0-2=>repCode; 3+=>offset+2) */
+    {   U32 const offCode = ZSTD_highbit32(offsetCode+1);
+        assert(offCode <= MaxOff);
+        optPtr->offCodeFreq[offCode]++;
+        optPtr->offCodeSum++;
+    }
+
+    /* match Length */
+    {   U32 const mlBase = matchLength - MINMATCH;
+        U32 const mlCode = ZSTD_MLcode(mlBase);
+        optPtr->matchLengthFreq[mlCode]++;
+        optPtr->matchLengthSum++;
+    }
+}
+
+
+/* ZSTD_readMINMATCH() :
+ * function safe only for comparisons
+ * assumption : memPtr must be at least 4 bytes before end of buffer */
+MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
+{
+    switch (length)
+    {
+    default :
+    case 4 : return MEM_read32(memPtr);
+    case 3 : if (MEM_isLittleEndian())
+                return MEM_read32(memPtr)<<8;
+             else
+                return MEM_read32(memPtr)>>8;
+    }
+}
+
+
+/* Update hashTable3 up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms, const BYTE* const ip)
+{
+    U32* const hashTable3 = ms->hashTable3;
+    U32 const hashLog3 = ms->hashLog3;
+    const BYTE* const base = ms->window.base;
+    U32 idx = ms->nextToUpdate3;
+    U32 const target = ms->nextToUpdate3 = (U32)(ip - base);
+    size_t const hash3 = ZSTD_hash3Ptr(ip, hashLog3);
+    assert(hashLog3 > 0);
+
+    while(idx < target) {
+        hashTable3[ZSTD_hash3Ptr(base+idx, hashLog3)] = idx;
+        idx++;
+    }
+
+    return hashTable3[hash3];
+}
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+/** ZSTD_insertBt1() : add one or multiple positions to tree.
+ *  ip : assumed <= iend-8 .
+ * @return : nb of positions added */
+static U32 ZSTD_insertBt1(
+                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                const BYTE* const ip, const BYTE* const iend,
+                U32 const mls, U32 const extDict)
+{
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32 matchIndex = hashTable[h];
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* match;
+    const U32 current = (U32)(ip-base);
+    const U32 btLow = btMask >= current ? 0 : current - btMask;
+    U32* smallerPtr = bt + 2*(current&btMask);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 dummy32;   /* to be nullified at the end */
+    U32 const windowLow = ms->window.lowLimit;
+    U32 matchEndIdx = current+8+1;
+    size_t bestLength = 8;
+    U32 nbCompares = 1U << cParams->searchLog;
+#ifdef ZSTD_C_PREDICT
+    U32 predictedSmall = *(bt + 2*((current-1)&btMask) + 0);
+    U32 predictedLarge = *(bt + 2*((current-1)&btMask) + 1);
+    predictedSmall += (predictedSmall>0);
+    predictedLarge += (predictedLarge>0);
+#endif /* ZSTD_C_PREDICT */
+
+    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", current);
+
+    assert(ip <= iend-8);   /* required for h calculation */
+    hashTable[h] = current;   /* Update Hash Table */
+
+    while (nbCompares-- && (matchIndex > windowLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(matchIndex < current);
+
+#ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */
+        const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */
+        if (matchIndex == predictedSmall) {
+            /* no need to check length, result known */
+            *smallerPtr = matchIndex;
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            predictedSmall = predictPtr[1] + (predictPtr[1]>0);
+            continue;
+        }
+        if (matchIndex == predictedLarge) {
+            *largerPtr = matchIndex;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+            predictedLarge = predictPtr[0] + (predictPtr[0]>0);
+            continue;
+        }
+#endif
+
+        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);   /* might be wrong if extDict is incorrectly set to 0 */
+            match = base + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            bestLength = matchLength;
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+        }
+
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+    if (bestLength > 384) return MIN(192, (U32)(bestLength - 384));   /* speed optimization */
+    assert(matchEndIdx > current + 8);
+    return matchEndIdx - (current + 8);
+}
+
+FORCE_INLINE_TEMPLATE
+void ZSTD_updateTree_internal(
+                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                const BYTE* const ip, const BYTE* const iend,
+                const U32 mls, const U32 extDict)
+{
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+    DEBUGLOG(7, "ZSTD_updateTree_internal, from %u to %u  (extDict:%u)",
+                idx, target, extDict);
+
+    while(idx < target)
+        idx += ZSTD_insertBt1(ms, cParams, base+idx, iend, mls, extDict);
+    ms->nextToUpdate = target;
+}
+
+void ZSTD_updateTree(
+                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                const BYTE* ip, const BYTE* iend)
+{
+    ZSTD_updateTree_internal(ms, cParams, ip, iend, cParams->searchLength, 0 /*extDict*/);
+}
+
+FORCE_INLINE_TEMPLATE
+U32 ZSTD_insertBtAndGetAllMatches (
+                    ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                    const BYTE* const ip, const BYTE* const iLimit, int const extDict,
+                    U32 rep[ZSTD_REP_NUM], U32 const ll0,
+                    ZSTD_match_t* matches, const U32 lengthToBeat, U32 const mls /* template */)
+{
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    const BYTE* const base = ms->window.base;
+    U32 const current = (U32)(ip-base);
+    U32 const hashLog = cParams->hashLog;
+    U32 const minMatch = (mls==3) ? 3 : 4;
+    U32* const hashTable = ms->hashTable;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32 matchIndex  = hashTable[h];
+    U32* const bt   = ms->chainTable;
+    U32 const btLog = cParams->chainLog - 1;
+    U32 const btMask= (1U << btLog) - 1;
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const dictBase = ms->window.dictBase;
+    U32 const dictLimit = ms->window.dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    U32 const btLow = btMask >= current ? 0 : current - btMask;
+    U32 const windowLow = ms->window.lowLimit;
+    U32* smallerPtr = bt + 2*(current&btMask);
+    U32* largerPtr  = bt + 2*(current&btMask) + 1;
+    U32 matchEndIdx = current+8+1;   /* farthest referenced position of any match => detects repetitive patterns */
+    U32 dummy32;   /* to be nullified at the end */
+    U32 mnum = 0;
+    U32 nbCompares = 1U << cParams->searchLog;
+
+    size_t bestLength = lengthToBeat-1;
+    DEBUGLOG(7, "ZSTD_insertBtAndGetAllMatches");
+
+    /* check repCode */
+    {   U32 const lastR = ZSTD_REP_NUM + ll0;
+        U32 repCode;
+        for (repCode = ll0; repCode < lastR; repCode++) {
+            U32 const repOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            U32 const repIndex = current - repOffset;
+            U32 repLen = 0;
+            assert(current >= dictLimit);
+            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < current-dictLimit) {  /* equivalent to `current > repIndex >= dictLimit` */
+                if (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repOffset, minMatch)) {
+                    repLen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repOffset, iLimit) + minMatch;
+                }
+            } else {  /* repIndex < dictLimit || repIndex >= current */
+                const BYTE* const repMatch = dictBase + repIndex;
+                assert(current >= windowLow);
+                if ( extDict /* this case only valid in extDict mode */
+                  && ( ((repOffset-1) /*intentional overflow*/ < current - windowLow)  /* equivalent to `current > repIndex >= windowLow` */
+                     & (((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */)
+                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
+                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dictEnd, prefixStart) + minMatch;
+            }   }
+            /* save longer solution */
+            if (repLen > bestLength) {
+                DEBUGLOG(8, "found rep-match %u of length %u",
+                            repCode - ll0, (U32)repLen);
+                bestLength = repLen;
+                matches[mnum].off = repCode - ll0;
+                matches[mnum].len = (U32)repLen;
+                mnum++;
+                if ( (repLen > sufficient_len)
+                   | (ip+repLen == iLimit) ) {  /* best possible */
+                    return mnum;
+    }   }   }   }
+
+    /* HC3 match finder */
+    if ((mls == 3) /*static*/ && (bestLength < mls)) {
+        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, ip);
+        if ((matchIndex3 > windowLow)
+          & (current - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
+            size_t mlen;
+            if ((!extDict) /*static*/ || (matchIndex3 >= dictLimit)) {
+                const BYTE* const match = base + matchIndex3;
+                mlen = ZSTD_count(ip, match, iLimit);
+            } else {
+                const BYTE* const match = dictBase + matchIndex3;
+                mlen = ZSTD_count_2segments(ip, match, iLimit, dictEnd, prefixStart);
+            }
+
+            /* save best solution */
+            if (mlen >= mls /* == 3 > bestLength */) {
+                DEBUGLOG(8, "found small match with hlog3, of length %u",
+                            (U32)mlen);
+                bestLength = mlen;
+                assert(current > matchIndex3);
+                assert(mnum==0);  /* no prior solution */
+                matches[0].off = (current - matchIndex3) + ZSTD_REP_MOVE;
+                matches[0].len = (U32)mlen;
+                mnum = 1;
+                if ( (mlen > sufficient_len) |
+                     (ip+mlen == iLimit) ) {  /* best possible length */
+                    ms->nextToUpdate = current+1;  /* skip insertion */
+                    return 1;
+    }   }   }   }
+
+    hashTable[h] = current;   /* Update Hash Table */
+
+    while (nbCompares-- && (matchIndex > windowLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        const BYTE* match;
+        assert(current > matchIndex);
+
+        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
+            match = base + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iLimit);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* prepare for match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            DEBUGLOG(8, "found match of length %u at distance %u",
+                    (U32)matchLength, current - matchIndex);
+            assert(matchEndIdx > matchIndex);
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+            bestLength = matchLength;
+            matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
+            matches[mnum].len = (U32)matchLength;
+            mnum++;
+            if (matchLength > ZSTD_OPT_NUM) break;
+            if (ip+matchLength == iLimit) {  /* equal : no way to know if inf or sup */
+                break;   /* drop, to preserve bt consistency (miss a little bit of compression) */
+            }
+        }
+
+        if (match[matchLength] < ip[matchLength]) {
+            /* match smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new candidate => larger than match, which was smaller than current */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous, closer to current */
+        } else {
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+
+    assert(matchEndIdx > current+8);
+    ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
+    return mnum;
+}
+
+
+FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iHighLimit, int const extDict,
+                        U32 rep[ZSTD_REP_NUM], U32 const ll0,
+                        ZSTD_match_t* matches, U32 const lengthToBeat)
+{
+    U32 const matchLengthSearch = cParams->searchLength;
+    DEBUGLOG(7, "ZSTD_BtGetAllMatches");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateTree_internal(ms, cParams, ip, iHighLimit, matchLengthSearch, extDict);
+    switch(matchLengthSearch)
+    {
+    case 3 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 3);
+    default :
+    case 4 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 4);
+    case 5 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 5);
+    case 7 :
+    case 6 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 6);
+    }
+}
+
+
+/*-*******************************
+*  Optimal parser
+*********************************/
+typedef struct repcodes_s {
+    U32 rep[3];
+} repcodes_t;
+
+repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 const ll0)
+{
+    repcodes_t newReps;
+    if (offset >= ZSTD_REP_NUM) {  /* full offset */
+        newReps.rep[2] = rep[1];
+        newReps.rep[1] = rep[0];
+        newReps.rep[0] = offset - ZSTD_REP_MOVE;
+    } else {   /* repcode */
+        U32 const repCode = offset + ll0;
+        if (repCode > 0) {  /* note : if repCode==0, no change */
+            U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            newReps.rep[2] = (repCode >= 2) ? rep[1] : rep[2];
+            newReps.rep[1] = rep[0];
+            newReps.rep[0] = currentOffset;
+        } else {   /* repCode == 0 */
+            memcpy(&newReps, rep, sizeof(newReps));
+        }
+    }
+    return newReps;
+}
+
+
+typedef struct {
+    const BYTE* anchor;
+    U32 litlen;
+    U32 rawLitCost;
+} cachedLiteralPrice_t;
+
+static U32 ZSTD_rawLiteralsCost_cached(
+                            cachedLiteralPrice_t* const cachedLitPrice,
+                            const BYTE* const anchor, U32 const litlen,
+                            const optState_t* const optStatePtr)
+{
+    U32 startCost;
+    U32 remainingLength;
+    const BYTE* startPosition;
+
+    if (anchor == cachedLitPrice->anchor) {
+        startCost = cachedLitPrice->rawLitCost;
+        startPosition = anchor + cachedLitPrice->litlen;
+        assert(litlen >= cachedLitPrice->litlen);
+        remainingLength = litlen - cachedLitPrice->litlen;
+    } else {
+        startCost = 0;
+        startPosition = anchor;
+        remainingLength = litlen;
+    }
+
+    {   U32 const rawLitCost = startCost + ZSTD_rawLiteralsCost(startPosition, remainingLength, optStatePtr);
+        cachedLitPrice->anchor = anchor;
+        cachedLitPrice->litlen = litlen;
+        cachedLitPrice->rawLitCost = rawLitCost;
+        return rawLitCost;
+    }
+}
+
+static U32 ZSTD_fullLiteralsCost_cached(
+                            cachedLiteralPrice_t* const cachedLitPrice,
+                            const BYTE* const anchor, U32 const litlen,
+                            const optState_t* const optStatePtr)
+{
+    return ZSTD_rawLiteralsCost_cached(cachedLitPrice, anchor, litlen, optStatePtr)
+         + ZSTD_litLengthPrice(litlen, optStatePtr);
+}
+
+static int ZSTD_literalsContribution_cached(
+                            cachedLiteralPrice_t* const cachedLitPrice,
+                            const BYTE* const anchor, U32 const litlen,
+                            const optState_t* const optStatePtr)
+{
+    int const contribution = ZSTD_rawLiteralsCost_cached(cachedLitPrice, anchor, litlen, optStatePtr)
+                           + ZSTD_litLengthContribution(litlen, optStatePtr);
+    return contribution;
+}
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,seqStore_t* seqStore,
+                                      U32 rep[ZSTD_REP_NUM],
+                                      ZSTD_compressionParameters const* cParams,
+                                      const void* src, size_t srcSize,
+                                      const int optLevel, const int extDict)
+{
+    optState_t* const optStatePtr = &ms->opt;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + ms->window.dictLimit;
+
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    U32 const minMatch = (cParams->searchLength == 3) ? 3 : 4;
+
+    ZSTD_optimal_t* const opt = optStatePtr->priceTable;
+    ZSTD_match_t* const matches = optStatePtr->matchTable;
+    cachedLiteralPrice_t cachedLitPrice;
+
+    /* init */
+    DEBUGLOG(5, "ZSTD_compressBlock_opt_generic");
+    ms->nextToUpdate3 = ms->nextToUpdate;
+    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize);
+    ip += (ip==prefixStart);
+    memset(&cachedLitPrice, 0, sizeof(cachedLitPrice));
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        U32 cur, last_pos = 0;
+        U32 best_mlen, best_off;
+
+        /* find first match */
+        {   U32 const litlen = (U32)(ip - anchor);
+            U32 const ll0 = !litlen;
+            U32 const nbMatches = ZSTD_BtGetAllMatches(ms, cParams, ip, iend, extDict, rep, ll0, matches, minMatch);
+            if (!nbMatches) { ip++; continue; }
+
+            /* initialize opt[0] */
+            { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
+            opt[0].mlen = 1;
+            opt[0].litlen = litlen;
+
+            /* large match -> immediate encoding */
+            {   U32 const maxML = matches[nbMatches-1].len;
+                DEBUGLOG(7, "found %u matches of maxLength=%u and offset=%u at cPos=%u => start new serie",
+                            nbMatches, maxML, matches[nbMatches-1].off, (U32)(ip-prefixStart));
+
+                if (maxML > sufficient_len) {
+                    best_mlen = maxML;
+                    best_off = matches[nbMatches-1].off;
+                    DEBUGLOG(7, "large match (%u>%u), immediate encoding",
+                                best_mlen, sufficient_len);
+                    cur = 0;
+                    last_pos = 1;
+                    goto _shortestPath;
+            }   }
+
+            /* set prices for first matches starting position == 0 */
+            {   U32 const literalsPrice = ZSTD_fullLiteralsCost_cached(&cachedLitPrice, anchor, litlen, optStatePtr);
+                U32 pos;
+                U32 matchNb;
+                for (pos = 0; pos < minMatch; pos++) {
+                    opt[pos].mlen = 1;
+                    opt[pos].price = ZSTD_MAX_PRICE;
+                }
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offset = matches[matchNb].off;
+                    U32 const end = matches[matchNb].len;
+                    repcodes_t const repHistory = ZSTD_updateRep(rep, offset, ll0);
+                    for ( ; pos <= end ; pos++ ) {
+                        U32 const matchPrice = literalsPrice + ZSTD_getMatchPrice(offset, pos, optStatePtr, optLevel);
+                        DEBUGLOG(7, "rPos:%u => set initial price : %u",
+                                    pos, matchPrice);
+                        opt[pos].mlen = pos;
+                        opt[pos].off = offset;
+                        opt[pos].litlen = litlen;
+                        opt[pos].price = matchPrice;
+                        memcpy(opt[pos].rep, &repHistory, sizeof(repHistory));
+                }   }
+                last_pos = pos-1;
+            }
+        }
+
+        /* check further positions */
+        for (cur = 1; cur <= last_pos; cur++) {
+            const BYTE* const inr = ip + cur;
+            assert(cur < ZSTD_OPT_NUM);
+
+            /* Fix current position with one literal if cheaper */
+            {   U32 const litlen = (opt[cur-1].mlen == 1) ? opt[cur-1].litlen + 1 : 1;
+                int price;  /* note : contribution can be negative */
+                if (cur > litlen) {
+                    price = opt[cur - litlen].price + ZSTD_literalsContribution(inr-litlen, litlen, optStatePtr);
+                } else {
+                    price = ZSTD_literalsContribution_cached(&cachedLitPrice, anchor, litlen, optStatePtr);
+                }
+                assert(price < 1000000000); /* overflow check */
+                if (price <= opt[cur].price) {
+                    DEBUGLOG(7, "rPos:%u : better price (%u<%u) using literal",
+                                cur, price, opt[cur].price);
+                    opt[cur].mlen = 1;
+                    opt[cur].off = 0;
+                    opt[cur].litlen = litlen;
+                    opt[cur].price = price;
+                    memcpy(opt[cur].rep, opt[cur-1].rep, sizeof(opt[cur].rep));
+            }   }
+
+            /* last match must start at a minimum distance of 8 from oend */
+            if (inr > ilimit) continue;
+
+            if (cur == last_pos) break;
+
+             if ( (optLevel==0) /*static*/
+               && (opt[cur+1].price <= opt[cur].price) )
+                continue;  /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
+
+            {   U32 const ll0 = (opt[cur].mlen != 1);
+                U32 const litlen = (opt[cur].mlen == 1) ? opt[cur].litlen : 0;
+                U32 const previousPrice = (cur > litlen) ? opt[cur-litlen].price : 0;
+                U32 const basePrice = previousPrice + ZSTD_fullLiteralsCost(inr-litlen, litlen, optStatePtr);
+                U32 const nbMatches = ZSTD_BtGetAllMatches(ms, cParams, inr, iend, extDict, opt[cur].rep, ll0, matches, minMatch);
+                U32 matchNb;
+                if (!nbMatches) continue;
+
+                {   U32 const maxML = matches[nbMatches-1].len;
+                    DEBUGLOG(7, "rPos:%u, found %u matches, of maxLength=%u",
+                                cur, nbMatches, maxML);
+
+                    if ( (maxML > sufficient_len)
+                       | (cur + maxML >= ZSTD_OPT_NUM) ) {
+                        best_mlen = maxML;
+                        best_off = matches[nbMatches-1].off;
+                        last_pos = cur + 1;
+                        goto _shortestPath;
+                    }
+                }
+
+                /* set prices using matches found at position == cur */
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offset = matches[matchNb].off;
+                    repcodes_t const repHistory = ZSTD_updateRep(opt[cur].rep, offset, ll0);
+                    U32 const lastML = matches[matchNb].len;
+                    U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
+                    U32 mlen;
+
+                    DEBUGLOG(7, "testing match %u => offCode=%u, mlen=%u, llen=%u",
+                                matchNb, matches[matchNb].off, lastML, litlen);
+
+                    for (mlen = lastML; mlen >= startML; mlen--) {
+                        U32 const pos = cur + mlen;
+                        int const price = basePrice + ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
+
+                        if ((pos > last_pos) || (price < opt[pos].price)) {
+                            DEBUGLOG(7, "rPos:%u => new better price (%u<%u)",
+                                        pos, price, opt[pos].price);
+                            while (last_pos < pos) { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; }
+                            opt[pos].mlen = mlen;
+                            opt[pos].off = offset;
+                            opt[pos].litlen = litlen;
+                            opt[pos].price = price;
+                            memcpy(opt[pos].rep, &repHistory, sizeof(repHistory));
+                        } else {
+                            if (optLevel==0) break;  /* gets ~+10% speed for about -0.01 ratio loss */
+                        }
+            }   }   }
+        }  /* for (cur = 1; cur <= last_pos; cur++) */
+
+        best_mlen = opt[last_pos].mlen;
+        best_off = opt[last_pos].off;
+        cur = last_pos - best_mlen;
+
+_shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
+        assert(opt[0].mlen == 1);
+
+        /* reverse traversal */
+        DEBUGLOG(7, "start reverse traversal (last_pos:%u, cur:%u)",
+                    last_pos, cur);
+        {   U32 selectedMatchLength = best_mlen;
+            U32 selectedOffset = best_off;
+            U32 pos = cur;
+            while (1) {
+                U32 const mlen = opt[pos].mlen;
+                U32 const off = opt[pos].off;
+                opt[pos].mlen = selectedMatchLength;
+                opt[pos].off = selectedOffset;
+                selectedMatchLength = mlen;
+                selectedOffset = off;
+                if (mlen > pos) break;
+                pos -= mlen;
+        }   }
+
+        /* save sequences */
+        {   U32 pos;
+            for (pos=0; pos < last_pos; ) {
+                U32 const llen = (U32)(ip - anchor);
+                U32 const mlen = opt[pos].mlen;
+                U32 const offset = opt[pos].off;
+                if (mlen == 1) { ip++; pos++; continue; }  /* literal position => move on */
+                pos += mlen; ip += mlen;
+
+                /* repcodes update : like ZSTD_updateRep(), but update in place */
+                if (offset >= ZSTD_REP_NUM) {  /* full offset */
+                    rep[2] = rep[1];
+                    rep[1] = rep[0];
+                    rep[0] = offset - ZSTD_REP_MOVE;
+                } else {   /* repcode */
+                    U32 const repCode = offset + (llen==0);
+                    if (repCode) {  /* note : if repCode==0, no change */
+                        U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+                        if (repCode >= 2) rep[2] = rep[1];
+                        rep[1] = rep[0];
+                        rep[0] = currentOffset;
+                    }
+                }
+
+                ZSTD_updateStats(optStatePtr, llen, anchor, offset, mlen);
+                ZSTD_storeSeq(seqStore, llen, anchor, offset, mlen-MINMATCH);
+                anchor = ip;
+        }   }
+        ZSTD_setLog2Prices(optStatePtr);
+    }   /* while (ip < ilimit) */
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_btopt(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock_btopt");
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 0 /*optLevel*/, 0 /*extDict*/);
+}
+
+size_t ZSTD_compressBlock_btultra(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 2 /*optLevel*/, 0 /*extDict*/);
+}
+
+size_t ZSTD_compressBlock_btopt_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 0 /*optLevel*/, 1 /*extDict*/);
+}
+
+size_t ZSTD_compressBlock_btultra_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 2 /*optLevel*/, 1 /*extDict*/);
+}
--- a/contrib/python-zstandard/zstd/compress/zstd_opt.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/compress/zstd_opt.h	Mon Apr 09 10:13:29 2018 -0700
@@ -1,919 +1,42 @@
-/**
- * Copyright (c) 2016-present, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
-
-/* Note : this file is intended to be included within zstd_compress.c */
-
-
-#ifndef ZSTD_OPT_H_91842398743
-#define ZSTD_OPT_H_91842398743
-
-
-#define ZSTD_LITFREQ_ADD    2
-#define ZSTD_FREQ_DIV       4
-#define ZSTD_MAX_PRICE      (1<<30)
-
-/*-*************************************
-*  Price functions for optimal parser
-***************************************/
-FORCE_INLINE void ZSTD_setLog2Prices(seqStore_t* ssPtr)
-{
-    ssPtr->log2matchLengthSum = ZSTD_highbit32(ssPtr->matchLengthSum+1);
-    ssPtr->log2litLengthSum = ZSTD_highbit32(ssPtr->litLengthSum+1);
-    ssPtr->log2litSum = ZSTD_highbit32(ssPtr->litSum+1);
-    ssPtr->log2offCodeSum = ZSTD_highbit32(ssPtr->offCodeSum+1);
-    ssPtr->factor = 1 + ((ssPtr->litSum>>5) / ssPtr->litLengthSum) + ((ssPtr->litSum<<1) / (ssPtr->litSum + ssPtr->matchSum));
-}
-
-
-MEM_STATIC void ZSTD_rescaleFreqs(seqStore_t* ssPtr, const BYTE* src, size_t srcSize)
-{
-    unsigned u;
-
-    ssPtr->cachedLiterals = NULL;
-    ssPtr->cachedPrice = ssPtr->cachedLitLength = 0;
-    ssPtr->staticPrices = 0;
-
-    if (ssPtr->litLengthSum == 0) {
-        if (srcSize <= 1024) ssPtr->staticPrices = 1;
-
-        for (u=0; u<=MaxLit; u++)
-            ssPtr->litFreq[u] = 0;
-        for (u=0; u<srcSize; u++)
-            ssPtr->litFreq[src[u]]++;
-
-        ssPtr->litSum = 0;
-        ssPtr->litLengthSum = MaxLL+1;
-        ssPtr->matchLengthSum = MaxML+1;
-        ssPtr->offCodeSum = (MaxOff+1);
-        ssPtr->matchSum = (ZSTD_LITFREQ_ADD<<Litbits);
-
-        for (u=0; u<=MaxLit; u++) {
-            ssPtr->litFreq[u] = 1 + (ssPtr->litFreq[u]>>ZSTD_FREQ_DIV);
-            ssPtr->litSum += ssPtr->litFreq[u];
-        }
-        for (u=0; u<=MaxLL; u++)
-            ssPtr->litLengthFreq[u] = 1;
-        for (u=0; u<=MaxML; u++)
-            ssPtr->matchLengthFreq[u] = 1;
-        for (u=0; u<=MaxOff; u++)
-            ssPtr->offCodeFreq[u] = 1;
-    } else {
-        ssPtr->matchLengthSum = 0;
-        ssPtr->litLengthSum = 0;
-        ssPtr->offCodeSum = 0;
-        ssPtr->matchSum = 0;
-        ssPtr->litSum = 0;
-
-        for (u=0; u<=MaxLit; u++) {
-            ssPtr->litFreq[u] = 1 + (ssPtr->litFreq[u]>>(ZSTD_FREQ_DIV+1));
-            ssPtr->litSum += ssPtr->litFreq[u];
-        }
-        for (u=0; u<=MaxLL; u++) {
-            ssPtr->litLengthFreq[u] = 1 + (ssPtr->litLengthFreq[u]>>(ZSTD_FREQ_DIV+1));
-            ssPtr->litLengthSum += ssPtr->litLengthFreq[u];
-        }
-        for (u=0; u<=MaxML; u++) {
-            ssPtr->matchLengthFreq[u] = 1 + (ssPtr->matchLengthFreq[u]>>ZSTD_FREQ_DIV);
-            ssPtr->matchLengthSum += ssPtr->matchLengthFreq[u];
-            ssPtr->matchSum += ssPtr->matchLengthFreq[u] * (u + 3);
-        }
-        ssPtr->matchSum *= ZSTD_LITFREQ_ADD;
-        for (u=0; u<=MaxOff; u++) {
-            ssPtr->offCodeFreq[u] = 1 + (ssPtr->offCodeFreq[u]>>ZSTD_FREQ_DIV);
-            ssPtr->offCodeSum += ssPtr->offCodeFreq[u];
-        }
-    }
-
-    ZSTD_setLog2Prices(ssPtr);
-}
-
-
-FORCE_INLINE U32 ZSTD_getLiteralPrice(seqStore_t* ssPtr, U32 litLength, const BYTE* literals)
-{
-    U32 price, u;
-
-    if (ssPtr->staticPrices)
-        return ZSTD_highbit32((U32)litLength+1) + (litLength*6);
-
-    if (litLength == 0)
-        return ssPtr->log2litLengthSum - ZSTD_highbit32(ssPtr->litLengthFreq[0]+1);
-
-    /* literals */
-    if (ssPtr->cachedLiterals == literals) {
-        U32 const additional = litLength - ssPtr->cachedLitLength;
-        const BYTE* literals2 = ssPtr->cachedLiterals + ssPtr->cachedLitLength;
-        price = ssPtr->cachedPrice + additional * ssPtr->log2litSum;
-        for (u=0; u < additional; u++)
-            price -= ZSTD_highbit32(ssPtr->litFreq[literals2[u]]+1);
-        ssPtr->cachedPrice = price;
-        ssPtr->cachedLitLength = litLength;
-    } else {
-        price = litLength * ssPtr->log2litSum;
-        for (u=0; u < litLength; u++)
-            price -= ZSTD_highbit32(ssPtr->litFreq[literals[u]]+1);
-
-        if (litLength >= 12) {
-            ssPtr->cachedLiterals = literals;
-            ssPtr->cachedPrice = price;
-            ssPtr->cachedLitLength = litLength;
-        }
-    }
-
-    /* literal Length */
-    {   const BYTE LL_deltaCode = 19;
-        const BYTE llCode = (litLength>63) ? (BYTE)ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
-        price += LL_bits[llCode] + ssPtr->log2litLengthSum - ZSTD_highbit32(ssPtr->litLengthFreq[llCode]+1);
-    }
-
-    return price;
-}
-
-
-FORCE_INLINE U32 ZSTD_getPrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength, const int ultra)
-{
-    /* offset */
-    U32 price;
-    BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
-
-    if (seqStorePtr->staticPrices)
-        return ZSTD_getLiteralPrice(seqStorePtr, litLength, literals) + ZSTD_highbit32((U32)matchLength+1) + 16 + offCode;
-
-    price = offCode + seqStorePtr->log2offCodeSum - ZSTD_highbit32(seqStorePtr->offCodeFreq[offCode]+1);
-    if (!ultra && offCode >= 20) price += (offCode-19)*2;
-
-    /* match Length */
-    {   const BYTE ML_deltaCode = 36;
-        const BYTE mlCode = (matchLength>127) ? (BYTE)ZSTD_highbit32(matchLength) + ML_deltaCode : ML_Code[matchLength];
-        price += ML_bits[mlCode] + seqStorePtr->log2matchLengthSum - ZSTD_highbit32(seqStorePtr->matchLengthFreq[mlCode]+1);
-    }
-
-    return price + ZSTD_getLiteralPrice(seqStorePtr, litLength, literals) + seqStorePtr->factor;
-}
-
-
-MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength)
-{
-    U32 u;
-
-    /* literals */
-    seqStorePtr->litSum += litLength*ZSTD_LITFREQ_ADD;
-    for (u=0; u < litLength; u++)
-        seqStorePtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
-
-    /* literal Length */
-    {   const BYTE LL_deltaCode = 19;
-        const BYTE llCode = (litLength>63) ? (BYTE)ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
-        seqStorePtr->litLengthFreq[llCode]++;
-        seqStorePtr->litLengthSum++;
-    }
-
-    /* match offset */
-	{   BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
-		seqStorePtr->offCodeSum++;
-		seqStorePtr->offCodeFreq[offCode]++;
-	}
-
-    /* match Length */
-    {   const BYTE ML_deltaCode = 36;
-        const BYTE mlCode = (matchLength>127) ? (BYTE)ZSTD_highbit32(matchLength) + ML_deltaCode : ML_Code[matchLength];
-        seqStorePtr->matchLengthFreq[mlCode]++;
-        seqStorePtr->matchLengthSum++;
-    }
-
-    ZSTD_setLog2Prices(seqStorePtr);
-}
-
-
-#define SET_PRICE(pos, mlen_, offset_, litlen_, price_)   \
-    {                                                 \
-        while (last_pos < pos)  { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; } \
-        opt[pos].mlen = mlen_;                         \
-        opt[pos].off = offset_;                        \
-        opt[pos].litlen = litlen_;                     \
-        opt[pos].price = price_;                       \
-    }
-
-
-
-/* Update hashTable3 up to ip (excluded)
-   Assumption : always within prefix (ie. not within extDict) */
-FORCE_INLINE
-U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_CCtx* zc, const BYTE* ip)
-{
-    U32* const hashTable3  = zc->hashTable3;
-    U32 const hashLog3  = zc->hashLog3;
-    const BYTE* const base = zc->base;
-    U32 idx = zc->nextToUpdate3;
-    const U32 target = zc->nextToUpdate3 = (U32)(ip - base);
-    const size_t hash3 = ZSTD_hash3Ptr(ip, hashLog3);
-
-    while(idx < target) {
-        hashTable3[ZSTD_hash3Ptr(base+idx, hashLog3)] = idx;
-        idx++;
-    }
-
-    return hashTable3[hash3];
-}
-
+#ifndef ZSTD_OPT_H
+#define ZSTD_OPT_H
 
-/*-*************************************
-*  Binary Tree search
-***************************************/
-static U32 ZSTD_insertBtAndGetAllMatches (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        U32 nbCompares, const U32 mls,
-                        U32 extDict, ZSTD_match_t* matches, const U32 minMatchLen)
-{
-    const BYTE* const base = zc->base;
-    const U32 current = (U32)(ip-base);
-    const U32 hashLog = zc->params.cParams.hashLog;
-    const size_t h  = ZSTD_hashPtr(ip, hashLog, mls);
-    U32* const hashTable = zc->hashTable;
-    U32 matchIndex  = hashTable[h];
-    U32* const bt   = zc->chainTable;
-    const U32 btLog = zc->params.cParams.chainLog - 1;
-    const U32 btMask= (1U << btLog) - 1;
-    size_t commonLengthSmaller=0, commonLengthLarger=0;
-    const BYTE* const dictBase = zc->dictBase;
-    const U32 dictLimit = zc->dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
-    const BYTE* const prefixStart = base + dictLimit;
-    const U32 btLow = btMask >= current ? 0 : current - btMask;
-    const U32 windowLow = zc->lowLimit;
-    U32* smallerPtr = bt + 2*(current&btMask);
-    U32* largerPtr  = bt + 2*(current&btMask) + 1;
-    U32 matchEndIdx = current+8;
-    U32 dummy32;   /* to be nullified at the end */
-    U32 mnum = 0;
-
-    const U32 minMatch = (mls == 3) ? 3 : 4;
-    size_t bestLength = minMatchLen-1;
-
-    if (minMatch == 3) { /* HC3 match finder */
-        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3 (zc, ip);
-        if (matchIndex3>windowLow && (current - matchIndex3 < (1<<18))) {
-            const BYTE* match;
-            size_t currentMl=0;
-            if ((!extDict) || matchIndex3 >= dictLimit) {
-                match = base + matchIndex3;
-                if (match[bestLength] == ip[bestLength]) currentMl = ZSTD_count(ip, match, iLimit);
-            } else {
-                match = dictBase + matchIndex3;
-                if (MEM_readMINMATCH(match, MINMATCH) == MEM_readMINMATCH(ip, MINMATCH))    /* assumption : matchIndex3 <= dictLimit-4 (by table construction) */
-                    currentMl = ZSTD_count_2segments(ip+MINMATCH, match+MINMATCH, iLimit, dictEnd, prefixStart) + MINMATCH;
-            }
-
-            /* save best solution */
-            if (currentMl > bestLength) {
-                bestLength = currentMl;
-                matches[mnum].off = ZSTD_REP_MOVE_OPT + current - matchIndex3;
-                matches[mnum].len = (U32)currentMl;
-                mnum++;
-                if (currentMl > ZSTD_OPT_NUM) goto update;
-                if (ip+currentMl == iLimit) goto update; /* best possible, and avoid read overflow*/
-            }
-        }
-    }
-
-    hashTable[h] = current;   /* Update Hash Table */
-
-    while (nbCompares-- && (matchIndex > windowLow)) {
-        U32* nextPtr = bt + 2*(matchIndex & btMask);
-        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
-        const BYTE* match;
-
-        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
-            match = base + matchIndex;
-            if (match[matchLength] == ip[matchLength]) {
-                matchLength += ZSTD_count(ip+matchLength+1, match+matchLength+1, iLimit) +1;
-            }
-        } else {
-            match = dictBase + matchIndex;
-            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
-            if (matchIndex+matchLength >= dictLimit)
-                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
-        }
-
-        if (matchLength > bestLength) {
-            if (matchLength > matchEndIdx - matchIndex) matchEndIdx = matchIndex + (U32)matchLength;
-            bestLength = matchLength;
-            matches[mnum].off = ZSTD_REP_MOVE_OPT + current - matchIndex;
-            matches[mnum].len = (U32)matchLength;
-            mnum++;
-            if (matchLength > ZSTD_OPT_NUM) break;
-            if (ip+matchLength == iLimit)   /* equal : no way to know if inf or sup */
-                break;   /* drop, to guarantee consistency (miss a little bit of compression) */
-        }
-
-        if (match[matchLength] < ip[matchLength]) {
-            /* match is smaller than current */
-            *smallerPtr = matchIndex;             /* update smaller idx */
-            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
-            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
-            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
-        } else {
-            /* match is larger than current */
-            *largerPtr = matchIndex;
-            commonLengthLarger = matchLength;
-            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            largerPtr = nextPtr;
-            matchIndex = nextPtr[0];
-    }   }
-
-    *smallerPtr = *largerPtr = 0;
-
-update:
-    zc->nextToUpdate = (matchEndIdx > current + 8) ? matchEndIdx - 8 : current+1;
-    return mnum;
-}
-
+#if defined (__cplusplus)
+extern "C" {
+#endif
 
-/** Tree updater, providing best match */
-static U32 ZSTD_BtGetAllMatches (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        const U32 maxNbAttempts, const U32 mls, ZSTD_match_t* matches, const U32 minMatchLen)
-{
-    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree(zc, ip, iLimit, maxNbAttempts, mls);
-    return ZSTD_insertBtAndGetAllMatches(zc, ip, iLimit, maxNbAttempts, mls, 0, matches, minMatchLen);
-}
-
-
-static U32 ZSTD_BtGetAllMatches_selectMLS (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
-                        const BYTE* ip, const BYTE* const iHighLimit,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, const U32 minMatchLen)
-{
-    switch(matchLengthSearch)
-    {
-    case 3 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 3, matches, minMatchLen);
-    default :
-    case 4 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 4, matches, minMatchLen);
-    case 5 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 5, matches, minMatchLen);
-    case 6 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 6, matches, minMatchLen);
-    }
-}
-
-/** Tree updater, providing best match */
-static U32 ZSTD_BtGetAllMatches_extDict (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        const U32 maxNbAttempts, const U32 mls, ZSTD_match_t* matches, const U32 minMatchLen)
-{
-    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree_extDict(zc, ip, iLimit, maxNbAttempts, mls);
-    return ZSTD_insertBtAndGetAllMatches(zc, ip, iLimit, maxNbAttempts, mls, 1, matches, minMatchLen);
-}
-
-
-static U32 ZSTD_BtGetAllMatches_selectMLS_extDict (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
-                        const BYTE* ip, const BYTE* const iHighLimit,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, const U32 minMatchLen)
-{
-    switch(matchLengthSearch)
-    {
-    case 3 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 3, matches, minMatchLen);
-    default :
-    case 4 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 4, matches, minMatchLen);
-    case 5 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 5, matches, minMatchLen);
-    case 6 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 6, matches, minMatchLen);
-    }
-}
-
+#include "zstd_compress_internal.h"
 
-/*-*******************************
-*  Optimal parser
-*********************************/
-FORCE_INLINE
-void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
-                                    const void* src, size_t srcSize, const int ultra)
-{
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - 8;
-    const BYTE* const base = ctx->base;
-    const BYTE* const prefixStart = base + ctx->dictLimit;
-
-    const U32 maxSearches = 1U << ctx->params.cParams.searchLog;
-    const U32 sufficient_len = ctx->params.cParams.targetLength;
-    const U32 mls = ctx->params.cParams.searchLength;
-    const U32 minMatch = (ctx->params.cParams.searchLength == 3) ? 3 : 4;
-
-    ZSTD_optimal_t* opt = seqStorePtr->priceTable;
-    ZSTD_match_t* matches = seqStorePtr->matchTable;
-    const BYTE* inr;
-    U32 offset, rep[ZSTD_REP_NUM];
-
-    /* init */
-    ctx->nextToUpdate3 = ctx->nextToUpdate;
-    ZSTD_rescaleFreqs(seqStorePtr, (const BYTE*)src, srcSize);
-    ip += (ip==prefixStart);
-    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=ctx->rep[i]; }
-
-    /* Match Loop */
-    while (ip < ilimit) {
-        U32 cur, match_num, last_pos, litlen, price;
-        U32 u, mlen, best_mlen, best_off, litLength;
-        memset(opt, 0, sizeof(ZSTD_optimal_t));
-        last_pos = 0;
-        litlen = (U32)(ip - anchor);
-
-        /* check repCode */
-        {   U32 i, last_i = ZSTD_REP_CHECK + (ip==anchor);
-            for (i=(ip == anchor); i<last_i; i++) {
-                const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : rep[i];
-                if ( (repCur > 0) && (repCur < (S32)(ip-prefixStart))
-                    && (MEM_readMINMATCH(ip, minMatch) == MEM_readMINMATCH(ip - repCur, minMatch))) {
-                    mlen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repCur, iend) + minMatch;
-                    if (mlen > sufficient_len || mlen >= ZSTD_OPT_NUM) {
-                        best_mlen = mlen; best_off = i; cur = 0; last_pos = 1;
-                        goto _storeSequence;
-                    }
-                    best_off = i - (ip == anchor);
-                    do {
-                        price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
-                        if (mlen > last_pos || price < opt[mlen].price)
-                            SET_PRICE(mlen, mlen, i, litlen, price);   /* note : macro modifies last_pos */
-                        mlen--;
-                    } while (mlen >= minMatch);
-        }   }   }
-
-        match_num = ZSTD_BtGetAllMatches_selectMLS(ctx, ip, iend, maxSearches, mls, matches, minMatch);
-
-        if (!last_pos && !match_num) { ip++; continue; }
+void ZSTD_updateTree(
+        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+        const BYTE* ip, const BYTE* iend);  /* used in ZSTD_loadDictionaryContent() */
 
-        if (match_num && (matches[match_num-1].len > sufficient_len || matches[match_num-1].len >= ZSTD_OPT_NUM)) {
-            best_mlen = matches[match_num-1].len;
-            best_off = matches[match_num-1].off;
-            cur = 0;
-            last_pos = 1;
-            goto _storeSequence;
-        }
-
-        /* set prices using matches at position = 0 */
-        best_mlen = (last_pos) ? last_pos : minMatch;
-        for (u = 0; u < match_num; u++) {
-            mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
-            best_mlen = matches[u].len;
-            while (mlen <= best_mlen) {
-                price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
-                if (mlen > last_pos || price < opt[mlen].price)
-                    SET_PRICE(mlen, mlen, matches[u].off, litlen, price);   /* note : macro modifies last_pos */
-                mlen++;
-        }   }
-
-        if (last_pos < minMatch) { ip++; continue; }
-
-        /* initialize opt[0] */
-        { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
-        opt[0].mlen = 1;
-        opt[0].litlen = litlen;
-
-         /* check further positions */
-        for (cur = 1; cur <= last_pos; cur++) {
-           inr = ip + cur;
-
-           if (opt[cur-1].mlen == 1) {
-                litlen = opt[cur-1].litlen + 1;
-                if (cur > litlen) {
-                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-litlen);
-                } else
-                    price = ZSTD_getLiteralPrice(seqStorePtr, litlen, anchor);
-           } else {
-                litlen = 1;
-                price = opt[cur - 1].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-1);
-           }
-
-           if (cur > last_pos || price <= opt[cur].price)
-                SET_PRICE(cur, 1, 0, litlen, price);
-
-           if (cur == last_pos) break;
-
-           if (inr > ilimit)  /* last match must start at a minimum distance of 8 from oend */
-               continue;
-
-           mlen = opt[cur].mlen;
-           if (opt[cur].off > ZSTD_REP_MOVE_OPT) {
-                opt[cur].rep[2] = opt[cur-mlen].rep[1];
-                opt[cur].rep[1] = opt[cur-mlen].rep[0];
-                opt[cur].rep[0] = opt[cur].off - ZSTD_REP_MOVE_OPT;
-           } else {
-                opt[cur].rep[2] = (opt[cur].off > 1) ? opt[cur-mlen].rep[1] : opt[cur-mlen].rep[2];
-                opt[cur].rep[1] = (opt[cur].off > 0) ? opt[cur-mlen].rep[0] : opt[cur-mlen].rep[1];
-                opt[cur].rep[0] = ((opt[cur].off==ZSTD_REP_MOVE_OPT) && (mlen != 1)) ? (opt[cur-mlen].rep[0] - 1) : (opt[cur-mlen].rep[opt[cur].off]);
-           }
-
-            best_mlen = minMatch;
-            {   U32 i, last_i = ZSTD_REP_CHECK + (mlen != 1);
-                for (i=(opt[cur].mlen != 1); i<last_i; i++) {  /* check rep */
-                    const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (opt[cur].rep[0] - 1) : opt[cur].rep[i];
-                    if ( (repCur > 0) && (repCur < (S32)(inr-prefixStart))
-                       && (MEM_readMINMATCH(inr, minMatch) == MEM_readMINMATCH(inr - repCur, minMatch))) {
-                       mlen = (U32)ZSTD_count(inr+minMatch, inr+minMatch - repCur, iend) + minMatch;
-
-                       if (mlen > sufficient_len || cur + mlen >= ZSTD_OPT_NUM) {
-                            best_mlen = mlen; best_off = i; last_pos = cur + 1;
-                            goto _storeSequence;
-                       }
-
-                       best_off = i - (opt[cur].mlen != 1);
-                       if (mlen > best_mlen) best_mlen = mlen;
-
-                       do {
-                           if (opt[cur].mlen == 1) {
-                                litlen = opt[cur].litlen;
-                                if (cur > litlen) {
-                                    price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
-                                } else
-                                    price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
-                            } else {
-                                litlen = 0;
-                                price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
-                            }
-
-                            if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
-                                SET_PRICE(cur + mlen, mlen, i, litlen, price);
-                            mlen--;
-                        } while (mlen >= minMatch);
-            }   }   }
-
-            match_num = ZSTD_BtGetAllMatches_selectMLS(ctx, inr, iend, maxSearches, mls, matches, best_mlen);
-
-            if (match_num > 0 && (matches[match_num-1].len > sufficient_len || cur + matches[match_num-1].len >= ZSTD_OPT_NUM)) {
-                best_mlen = matches[match_num-1].len;
-                best_off = matches[match_num-1].off;
-                last_pos = cur + 1;
-                goto _storeSequence;
-            }
-
-            /* set prices using matches at position = cur */
-            for (u = 0; u < match_num; u++) {
-                mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
-                best_mlen = matches[u].len;
-
-                while (mlen <= best_mlen) {
-                    if (opt[cur].mlen == 1) {
-                        litlen = opt[cur].litlen;
-                        if (cur > litlen)
-                            price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
-                        else
-                            price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
-                    } else {
-                        litlen = 0;
-                        price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
-                    }
-
-                    if (cur + mlen > last_pos || (price < opt[cur + mlen].price))
-                        SET_PRICE(cur + mlen, mlen, matches[u].off, litlen, price);
-
-                    mlen++;
-        }   }   }
-
-        best_mlen = opt[last_pos].mlen;
-        best_off = opt[last_pos].off;
-        cur = last_pos - best_mlen;
-
-        /* store sequence */
-_storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
-        opt[0].mlen = 1;
-
-        while (1) {
-            mlen = opt[cur].mlen;
-            offset = opt[cur].off;
-            opt[cur].mlen = best_mlen;
-            opt[cur].off = best_off;
-            best_mlen = mlen;
-            best_off = offset;
-            if (mlen > cur) break;
-            cur -= mlen;
-        }
-
-        for (u = 0; u <= last_pos;) {
-            u += opt[u].mlen;
-        }
-
-        for (cur=0; cur < last_pos; ) {
-            mlen = opt[cur].mlen;
-            if (mlen == 1) { ip++; cur++; continue; }
-            offset = opt[cur].off;
-            cur += mlen;
-            litLength = (U32)(ip - anchor);
-
-            if (offset > ZSTD_REP_MOVE_OPT) {
-                rep[2] = rep[1];
-                rep[1] = rep[0];
-                rep[0] = offset - ZSTD_REP_MOVE_OPT;
-                offset--;
-            } else {
-                if (offset != 0) {
-                    best_off = (offset==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : (rep[offset]);
-                    if (offset != 1) rep[2] = rep[1];
-                    rep[1] = rep[0];
-                    rep[0] = best_off;
-                }
-                if (litLength==0) offset--;
-            }
-
-            ZSTD_updatePrice(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
-            ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
-            anchor = ip = ip + mlen;
-    }    }   /* for (cur=0; cur < last_pos; ) */
-
-    /* Save reps for next block */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->repToConfirm[i] = rep[i]; }
-
-    /* Last Literals */
-    {   size_t const lastLLSize = iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
-    }
-}
-
-
-FORCE_INLINE
-void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
-                                     const void* src, size_t srcSize, const int ultra)
-{
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - 8;
-    const BYTE* const base = ctx->base;
-    const U32 lowestIndex = ctx->lowLimit;
-    const U32 dictLimit = ctx->dictLimit;
-    const BYTE* const prefixStart = base + dictLimit;
-    const BYTE* const dictBase = ctx->dictBase;
-    const BYTE* const dictEnd  = dictBase + dictLimit;
-
-    const U32 maxSearches = 1U << ctx->params.cParams.searchLog;
-    const U32 sufficient_len = ctx->params.cParams.targetLength;
-    const U32 mls = ctx->params.cParams.searchLength;
-    const U32 minMatch = (ctx->params.cParams.searchLength == 3) ? 3 : 4;
-
-    ZSTD_optimal_t* opt = seqStorePtr->priceTable;
-    ZSTD_match_t* matches = seqStorePtr->matchTable;
-    const BYTE* inr;
-
-    /* init */
-    U32 offset, rep[ZSTD_REP_NUM];
-    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=ctx->rep[i]; }
-
-    ctx->nextToUpdate3 = ctx->nextToUpdate;
-    ZSTD_rescaleFreqs(seqStorePtr, (const BYTE*)src, srcSize);
-    ip += (ip==prefixStart);
-
-    /* Match Loop */
-    while (ip < ilimit) {
-        U32 cur, match_num, last_pos, litlen, price;
-        U32 u, mlen, best_mlen, best_off, litLength;
-        U32 current = (U32)(ip-base);
-        memset(opt, 0, sizeof(ZSTD_optimal_t));
-        last_pos = 0;
-        opt[0].litlen = (U32)(ip - anchor);
+size_t ZSTD_compressBlock_btopt(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
 
-        /* check repCode */
-        {   U32 i, last_i = ZSTD_REP_CHECK + (ip==anchor);
-            for (i = (ip==anchor); i<last_i; i++) {
-                const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : rep[i];
-                const U32 repIndex = (U32)(current - repCur);
-                const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
-                const BYTE* const repMatch = repBase + repIndex;
-                if ( (repCur > 0 && repCur <= (S32)current)
-                   && (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex>lowestIndex))  /* intentional overflow */
-                   && (MEM_readMINMATCH(ip, minMatch) == MEM_readMINMATCH(repMatch, minMatch)) ) {
-                    /* repcode detected we should take it */
-                    const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
-                    mlen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iend, repEnd, prefixStart) + minMatch;
-
-                    if (mlen > sufficient_len || mlen >= ZSTD_OPT_NUM) {
-                        best_mlen = mlen; best_off = i; cur = 0; last_pos = 1;
-                        goto _storeSequence;
-                    }
-
-                    best_off = i - (ip==anchor);
-                    litlen = opt[0].litlen;
-                    do {
-                        price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
-                        if (mlen > last_pos || price < opt[mlen].price)
-                            SET_PRICE(mlen, mlen, i, litlen, price);   /* note : macro modifies last_pos */
-                        mlen--;
-                    } while (mlen >= minMatch);
-        }   }   }
-
-        match_num = ZSTD_BtGetAllMatches_selectMLS_extDict(ctx, ip, iend, maxSearches, mls, matches, minMatch);  /* first search (depth 0) */
-
-        if (!last_pos && !match_num) { ip++; continue; }
-
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
-        opt[0].mlen = 1;
-
-        if (match_num && (matches[match_num-1].len > sufficient_len || matches[match_num-1].len >= ZSTD_OPT_NUM)) {
-            best_mlen = matches[match_num-1].len;
-            best_off = matches[match_num-1].off;
-            cur = 0;
-            last_pos = 1;
-            goto _storeSequence;
-        }
-
-        best_mlen = (last_pos) ? last_pos : minMatch;
-
-        /* set prices using matches at position = 0 */
-        for (u = 0; u < match_num; u++) {
-            mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
-            best_mlen = matches[u].len;
-            litlen = opt[0].litlen;
-            while (mlen <= best_mlen) {
-                price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
-                if (mlen > last_pos || price < opt[mlen].price)
-                    SET_PRICE(mlen, mlen, matches[u].off, litlen, price);
-                mlen++;
-        }   }
-
-        if (last_pos < minMatch) {
-            ip++; continue;
-        }
-
-        /* check further positions */
-        for (cur = 1; cur <= last_pos; cur++) {
-            inr = ip + cur;
-
-            if (opt[cur-1].mlen == 1) {
-                litlen = opt[cur-1].litlen + 1;
-                if (cur > litlen) {
-                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-litlen);
-                } else
-                    price = ZSTD_getLiteralPrice(seqStorePtr, litlen, anchor);
-            } else {
-                litlen = 1;
-                price = opt[cur - 1].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-1);
-            }
-
-            if (cur > last_pos || price <= opt[cur].price)
-                SET_PRICE(cur, 1, 0, litlen, price);
-
-            if (cur == last_pos) break;
-
-            if (inr > ilimit)  /* last match must start at a minimum distance of 8 from oend */
-                continue;
-
-            mlen = opt[cur].mlen;
-            if (opt[cur].off > ZSTD_REP_MOVE_OPT) {
-                opt[cur].rep[2] = opt[cur-mlen].rep[1];
-                opt[cur].rep[1] = opt[cur-mlen].rep[0];
-                opt[cur].rep[0] = opt[cur].off - ZSTD_REP_MOVE_OPT;
-            } else {
-                opt[cur].rep[2] = (opt[cur].off > 1) ? opt[cur-mlen].rep[1] : opt[cur-mlen].rep[2];
-                opt[cur].rep[1] = (opt[cur].off > 0) ? opt[cur-mlen].rep[0] : opt[cur-mlen].rep[1];
-                opt[cur].rep[0] = ((opt[cur].off==ZSTD_REP_MOVE_OPT) && (mlen != 1)) ? (opt[cur-mlen].rep[0] - 1) : (opt[cur-mlen].rep[opt[cur].off]);
-            }
-
-            best_mlen = minMatch;
-            {   U32 i, last_i = ZSTD_REP_CHECK + (mlen != 1);
-                for (i = (mlen != 1); i<last_i; i++) {
-                    const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (opt[cur].rep[0] - 1) : opt[cur].rep[i];
-                    const U32 repIndex = (U32)(current+cur - repCur);
-                    const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
-                    const BYTE* const repMatch = repBase + repIndex;
-                    if ( (repCur > 0 && repCur <= (S32)(current+cur))
-                      && (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex>lowestIndex))  /* intentional overflow */
-                      && (MEM_readMINMATCH(inr, minMatch) == MEM_readMINMATCH(repMatch, minMatch)) ) {
-                        /* repcode detected */
-                        const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
-                        mlen = (U32)ZSTD_count_2segments(inr+minMatch, repMatch+minMatch, iend, repEnd, prefixStart) + minMatch;
-
-                        if (mlen > sufficient_len || cur + mlen >= ZSTD_OPT_NUM) {
-                            best_mlen = mlen; best_off = i; last_pos = cur + 1;
-                            goto _storeSequence;
-                        }
+size_t ZSTD_compressBlock_btopt_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
 
-                        best_off = i - (opt[cur].mlen != 1);
-                        if (mlen > best_mlen) best_mlen = mlen;
-
-                        do {
-                            if (opt[cur].mlen == 1) {
-                                litlen = opt[cur].litlen;
-                                if (cur > litlen) {
-                                    price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
-                                } else
-                                    price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
-                            } else {
-                                litlen = 0;
-                                price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
-                            }
-
-                            if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
-                                SET_PRICE(cur + mlen, mlen, i, litlen, price);
-                            mlen--;
-                        } while (mlen >= minMatch);
-            }   }   }
-
-            match_num = ZSTD_BtGetAllMatches_selectMLS_extDict(ctx, inr, iend, maxSearches, mls, matches, minMatch);
-
-            if (match_num > 0 && (matches[match_num-1].len > sufficient_len || cur + matches[match_num-1].len >= ZSTD_OPT_NUM)) {
-                best_mlen = matches[match_num-1].len;
-                best_off = matches[match_num-1].off;
-                last_pos = cur + 1;
-                goto _storeSequence;
-            }
-
-            /* set prices using matches at position = cur */
-            for (u = 0; u < match_num; u++) {
-                mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
-                best_mlen = matches[u].len;
-
-                while (mlen <= best_mlen) {
-                    if (opt[cur].mlen == 1) {
-                        litlen = opt[cur].litlen;
-                        if (cur > litlen)
-                            price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
-                        else
-                            price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
-                    } else {
-                        litlen = 0;
-                        price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
-                    }
-
-                    if (cur + mlen > last_pos || (price < opt[cur + mlen].price))
-                        SET_PRICE(cur + mlen, mlen, matches[u].off, litlen, price);
-
-                    mlen++;
-        }   }   }   /* for (cur = 1; cur <= last_pos; cur++) */
-
-        best_mlen = opt[last_pos].mlen;
-        best_off = opt[last_pos].off;
-        cur = last_pos - best_mlen;
+#if defined (__cplusplus)
+}
+#endif
 
-        /* store sequence */
-_storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
-        opt[0].mlen = 1;
-
-        while (1) {
-            mlen = opt[cur].mlen;
-            offset = opt[cur].off;
-            opt[cur].mlen = best_mlen;
-            opt[cur].off = best_off;
-            best_mlen = mlen;
-            best_off = offset;
-            if (mlen > cur) break;
-            cur -= mlen;
-        }
-
-        for (u = 0; u <= last_pos; ) {
-            u += opt[u].mlen;
-        }
-
-        for (cur=0; cur < last_pos; ) {
-            mlen = opt[cur].mlen;
-            if (mlen == 1) { ip++; cur++; continue; }
-            offset = opt[cur].off;
-            cur += mlen;
-            litLength = (U32)(ip - anchor);
-
-            if (offset > ZSTD_REP_MOVE_OPT) {
-                rep[2] = rep[1];
-                rep[1] = rep[0];
-                rep[0] = offset - ZSTD_REP_MOVE_OPT;
-                offset--;
-            } else {
-                if (offset != 0) {
-                    best_off = (offset==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : (rep[offset]);
-                    if (offset != 1) rep[2] = rep[1];
-                    rep[1] = rep[0];
-                    rep[0] = best_off;
-                }
-
-                if (litLength==0) offset--;
-            }
-
-            ZSTD_updatePrice(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
-            ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
-            anchor = ip = ip + mlen;
-    }    }   /* for (cur=0; cur < last_pos; ) */
-
-    /* Save reps for next block */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->repToConfirm[i] = rep[i]; }
-
-    /* Last Literals */
-    {   size_t lastLLSize = iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
-    }
-}
-
-#endif  /* ZSTD_OPT_H_91842398743 */
+#endif /* ZSTD_OPT_H */
--- a/contrib/python-zstandard/zstd/compress/zstdmt_compress.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/compress/zstdmt_compress.c	Mon Apr 09 10:13:29 2018 -0700
@@ -1,264 +1,712 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 
 /* ======   Tuning parameters   ====== */
-#define ZSTDMT_NBTHREADS_MAX 128
+#define ZSTDMT_NBWORKERS_MAX 200
+#define ZSTDMT_JOBSIZE_MAX  (MEM_32bits() ? (512 MB) : (2 GB))  /* note : limited by `jobSize` type, which is `unsigned` */
+#define ZSTDMT_OVERLAPLOG_DEFAULT 6
 
 
 /* ======   Compiler specifics   ====== */
 #if defined(_MSC_VER)
-#  pragma warning(disable : 4204)        /* disable: C4204: non-constant aggregate initializer */
+#  pragma warning(disable : 4204)   /* disable: C4204: non-constant aggregate initializer */
 #endif
 
 
 /* ======   Dependencies   ====== */
-#include <stdlib.h>   /* malloc */
-#include <string.h>   /* memcpy */
-#include "pool.h"     /* threadpool */
-#include "threading.h"  /* mutex */
-#include "zstd_internal.h"   /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */
+#include <string.h>      /* memcpy, memset */
+#include <limits.h>      /* INT_MAX */
+#include "pool.h"        /* threadpool */
+#include "threading.h"   /* mutex */
+#include "zstd_compress_internal.h"  /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */
+#include "zstd_ldm.h"
 #include "zstdmt_compress.h"
-#define XXH_STATIC_LINKING_ONLY   /* XXH64_state_t */
-#include "xxhash.h"
 
+/* Guards code to support resizing the SeqPool.
+ * We will want to resize the SeqPool to save memory in the future.
+ * Until then, comment the code out since it is unused.
+ */
+#define ZSTD_RESIZE_SEQPOOL 0
 
 /* ======   Debug   ====== */
-#if 0
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
 
 #  include <stdio.h>
 #  include <unistd.h>
 #  include <sys/times.h>
-   static unsigned g_debugLevel = 3;
-#  define DEBUGLOGRAW(l, ...) if (l<=g_debugLevel) { fprintf(stderr, __VA_ARGS__); }
-#  define DEBUGLOG(l, ...) if (l<=g_debugLevel) { fprintf(stderr, __FILE__ ": "); fprintf(stderr, __VA_ARGS__); fprintf(stderr, " \n"); }
+#  define DEBUGLOGRAW(l, ...) if (l<=ZSTD_DEBUG) { fprintf(stderr, __VA_ARGS__); }
 
-#  define DEBUG_PRINTHEX(l,p,n) { \
-    unsigned debug_u;                   \
-    for (debug_u=0; debug_u<(n); debug_u++)           \
+#  define DEBUG_PRINTHEX(l,p,n) {            \
+    unsigned debug_u;                        \
+    for (debug_u=0; debug_u<(n); debug_u++)  \
         DEBUGLOGRAW(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \
-    DEBUGLOGRAW(l, " \n");       \
+    DEBUGLOGRAW(l, " \n");                   \
 }
 
-static unsigned long long GetCurrentClockTimeMicroseconds()
+static unsigned long long GetCurrentClockTimeMicroseconds(void)
 {
    static clock_t _ticksPerSecond = 0;
    if (_ticksPerSecond <= 0) _ticksPerSecond = sysconf(_SC_CLK_TCK);
 
-   struct tms junk; clock_t newTicks = (clock_t) times(&junk);
-   return ((((unsigned long long)newTicks)*(1000000))/_ticksPerSecond);
+   { struct tms junk; clock_t newTicks = (clock_t) times(&junk);
+     return ((((unsigned long long)newTicks)*(1000000))/_ticksPerSecond); }
 }
 
-#define MUTEX_WAIT_TIME_DLEVEL 5
-#define PTHREAD_MUTEX_LOCK(mutex) \
-if (g_debugLevel>=MUTEX_WAIT_TIME_DLEVEL) { \
-   unsigned long long beforeTime = GetCurrentClockTimeMicroseconds(); \
-   pthread_mutex_lock(mutex); \
-   unsigned long long afterTime = GetCurrentClockTimeMicroseconds(); \
-   unsigned long long elapsedTime = (afterTime-beforeTime); \
-   if (elapsedTime > 1000) {  /* or whatever threshold you like; I'm using 1 millisecond here */ \
-      DEBUGLOG(MUTEX_WAIT_TIME_DLEVEL, "Thread took %llu microseconds to acquire mutex %s \n", \
-               elapsedTime, #mutex); \
-  } \
-} else pthread_mutex_lock(mutex);
+#define MUTEX_WAIT_TIME_DLEVEL 6
+#define ZSTD_PTHREAD_MUTEX_LOCK(mutex) {          \
+    if (ZSTD_DEBUG >= MUTEX_WAIT_TIME_DLEVEL) {   \
+        unsigned long long const beforeTime = GetCurrentClockTimeMicroseconds(); \
+        ZSTD_pthread_mutex_lock(mutex);           \
+        {   unsigned long long const afterTime = GetCurrentClockTimeMicroseconds(); \
+            unsigned long long const elapsedTime = (afterTime-beforeTime); \
+            if (elapsedTime > 1000) {  /* or whatever threshold you like; I'm using 1 millisecond here */ \
+                DEBUGLOG(MUTEX_WAIT_TIME_DLEVEL, "Thread took %llu microseconds to acquire mutex %s \n", \
+                   elapsedTime, #mutex);          \
+        }   }                                     \
+    } else {                                      \
+        ZSTD_pthread_mutex_lock(mutex);           \
+    }                                             \
+}
 
 #else
 
-#  define DEBUGLOG(l, ...)      {}    /* disabled */
-#  define PTHREAD_MUTEX_LOCK(m) pthread_mutex_lock(m)
+#  define ZSTD_PTHREAD_MUTEX_LOCK(m) ZSTD_pthread_mutex_lock(m)
 #  define DEBUG_PRINTHEX(l,p,n) {}
 
 #endif
 
 
 /* =====   Buffer Pool   ===== */
+/* a single Buffer Pool can be invoked from multiple threads in parallel */
 
 typedef struct buffer_s {
     void* start;
-    size_t size;
+    size_t capacity;
 } buffer_t;
 
 static const buffer_t g_nullBuffer = { NULL, 0 };
 
 typedef struct ZSTDMT_bufferPool_s {
+    ZSTD_pthread_mutex_t poolMutex;
+    size_t bufferSize;
     unsigned totalBuffers;
     unsigned nbBuffers;
+    ZSTD_customMem cMem;
     buffer_t bTable[1];   /* variable size */
 } ZSTDMT_bufferPool;
 
-static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbThreads)
+static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbWorkers, ZSTD_customMem cMem)
 {
-    unsigned const maxNbBuffers = 2*nbThreads + 2;
-    ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)calloc(1, sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t));
+    unsigned const maxNbBuffers = 2*nbWorkers + 3;
+    ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_calloc(
+        sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem);
     if (bufPool==NULL) return NULL;
+    if (ZSTD_pthread_mutex_init(&bufPool->poolMutex, NULL)) {
+        ZSTD_free(bufPool, cMem);
+        return NULL;
+    }
+    bufPool->bufferSize = 64 KB;
     bufPool->totalBuffers = maxNbBuffers;
     bufPool->nbBuffers = 0;
+    bufPool->cMem = cMem;
     return bufPool;
 }
 
 static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
 {
     unsigned u;
+    DEBUGLOG(3, "ZSTDMT_freeBufferPool (address:%08X)", (U32)(size_t)bufPool);
     if (!bufPool) return;   /* compatibility with free on NULL */
+    for (u=0; u<bufPool->totalBuffers; u++) {
+        DEBUGLOG(4, "free buffer %2u (address:%08X)", u, (U32)(size_t)bufPool->bTable[u].start);
+        ZSTD_free(bufPool->bTable[u].start, bufPool->cMem);
+    }
+    ZSTD_pthread_mutex_destroy(&bufPool->poolMutex);
+    ZSTD_free(bufPool, bufPool->cMem);
+}
+
+/* only works at initialization, not during compression */
+static size_t ZSTDMT_sizeof_bufferPool(ZSTDMT_bufferPool* bufPool)
+{
+    size_t const poolSize = sizeof(*bufPool)
+                          + (bufPool->totalBuffers - 1) * sizeof(buffer_t);
+    unsigned u;
+    size_t totalBufferSize = 0;
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
     for (u=0; u<bufPool->totalBuffers; u++)
-        free(bufPool->bTable[u].start);
-    free(bufPool);
+        totalBufferSize += bufPool->bTable[u].capacity;
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+
+    return poolSize + totalBufferSize;
+}
+
+/* ZSTDMT_setBufferSize() :
+ * all future buffers provided by this buffer pool will have _at least_ this size
+ * note : it's better for all buffers to have same size,
+ * as they become freely interchangeable, reducing malloc/free usages and memory fragmentation */
+static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* const bufPool, size_t const bSize)
+{
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    DEBUGLOG(4, "ZSTDMT_setBufferSize: bSize = %u", (U32)bSize);
+    bufPool->bufferSize = bSize;
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
 }
 
-/* assumption : invocation from main thread only ! */
-static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* pool, size_t bSize)
+/** ZSTDMT_getBuffer() :
+ *  assumption : bufPool must be valid
+ * @return : a buffer, with start pointer and size
+ *  note: allocation may fail, in this case, start==NULL and size==0 */
+static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
 {
-    if (pool->nbBuffers) {   /* try to use an existing buffer */
-        buffer_t const buf = pool->bTable[--(pool->nbBuffers)];
-        size_t const availBufferSize = buf.size;
-        if ((availBufferSize >= bSize) & (availBufferSize <= 10*bSize))   /* large enough, but not too much */
+    size_t const bSize = bufPool->bufferSize;
+    DEBUGLOG(5, "ZSTDMT_getBuffer: bSize = %u", (U32)bufPool->bufferSize);
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers) {   /* try to use an existing buffer */
+        buffer_t const buf = bufPool->bTable[--(bufPool->nbBuffers)];
+        size_t const availBufferSize = buf.capacity;
+        bufPool->bTable[bufPool->nbBuffers] = g_nullBuffer;
+        if ((availBufferSize >= bSize) & ((availBufferSize>>3) <= bSize)) {
+            /* large enough, but not too much */
+            DEBUGLOG(5, "ZSTDMT_getBuffer: provide buffer %u of size %u",
+                        bufPool->nbBuffers, (U32)buf.capacity);
+            ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
             return buf;
-        free(buf.start);   /* size conditions not respected : scratch this buffer and create a new one */
+        }
+        /* size conditions not respected : scratch this buffer, create new one */
+        DEBUGLOG(5, "ZSTDMT_getBuffer: existing buffer does not meet size conditions => freeing");
+        ZSTD_free(buf.start, bufPool->cMem);
     }
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
     /* create new buffer */
+    DEBUGLOG(5, "ZSTDMT_getBuffer: create a new buffer");
     {   buffer_t buffer;
-        void* const start = malloc(bSize);
-        if (start==NULL) bSize = 0;
+        void* const start = ZSTD_malloc(bSize, bufPool->cMem);
         buffer.start = start;   /* note : start can be NULL if malloc fails ! */
-        buffer.size = bSize;
+        buffer.capacity = (start==NULL) ? 0 : bSize;
+        if (start==NULL) {
+            DEBUGLOG(5, "ZSTDMT_getBuffer: buffer allocation failure !!");
+        } else {
+            DEBUGLOG(5, "ZSTDMT_getBuffer: created buffer of size %u", (U32)bSize);
+        }
         return buffer;
     }
 }
 
-/* store buffer for later re-use, up to pool capacity */
-static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* pool, buffer_t buf)
+#if ZSTD_RESIZE_SEQPOOL
+/** ZSTDMT_resizeBuffer() :
+ * assumption : bufPool must be valid
+ * @return : a buffer that is at least the buffer pool buffer size.
+ *           If a reallocation happens, the data in the input buffer is copied.
+ */
+static buffer_t ZSTDMT_resizeBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buffer)
 {
-    if (buf.start == NULL) return;   /* release on NULL */
-    if (pool->nbBuffers < pool->totalBuffers) {
-        pool->bTable[pool->nbBuffers++] = buf;   /* store for later re-use */
+    size_t const bSize = bufPool->bufferSize;
+    if (buffer.capacity < bSize) {
+        void* const start = ZSTD_malloc(bSize, bufPool->cMem);
+        buffer_t newBuffer;
+        newBuffer.start = start;
+        newBuffer.capacity = start == NULL ? 0 : bSize;
+        if (start != NULL) {
+            assert(newBuffer.capacity >= buffer.capacity);
+            memcpy(newBuffer.start, buffer.start, buffer.capacity);
+            DEBUGLOG(5, "ZSTDMT_resizeBuffer: created buffer of size %u", (U32)bSize);
+            return newBuffer;
+        }
+        DEBUGLOG(5, "ZSTDMT_resizeBuffer: buffer allocation failure !!");
+    }
+    return buffer;
+}
+#endif
+
+/* store buffer for later re-use, up to pool capacity */
+static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
+{
+    if (buf.start == NULL) return;   /* compatible with release on NULL */
+    DEBUGLOG(5, "ZSTDMT_releaseBuffer");
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers < bufPool->totalBuffers) {
+        bufPool->bTable[bufPool->nbBuffers++] = buf;  /* stored for later use */
+        DEBUGLOG(5, "ZSTDMT_releaseBuffer: stored buffer of size %u in slot %u",
+                    (U32)buf.capacity, (U32)(bufPool->nbBuffers-1));
+        ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
         return;
     }
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
     /* Reached bufferPool capacity (should not happen) */
-    free(buf.start);
+    DEBUGLOG(5, "ZSTDMT_releaseBuffer: pool capacity reached => freeing ");
+    ZSTD_free(buf.start, bufPool->cMem);
 }
 
 
+/* =====   Seq Pool Wrapper   ====== */
+
+static rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0};
+
+typedef ZSTDMT_bufferPool ZSTDMT_seqPool;
+
+static size_t ZSTDMT_sizeof_seqPool(ZSTDMT_seqPool* seqPool)
+{
+    return ZSTDMT_sizeof_bufferPool(seqPool);
+}
+
+static rawSeqStore_t bufferToSeq(buffer_t buffer)
+{
+    rawSeqStore_t seq = {NULL, 0, 0, 0};
+    seq.seq = (rawSeq*)buffer.start;
+    seq.capacity = buffer.capacity / sizeof(rawSeq);
+    return seq;
+}
+
+static buffer_t seqToBuffer(rawSeqStore_t seq)
+{
+    buffer_t buffer;
+    buffer.start = seq.seq;
+    buffer.capacity = seq.capacity * sizeof(rawSeq);
+    return buffer;
+}
+
+static rawSeqStore_t ZSTDMT_getSeq(ZSTDMT_seqPool* seqPool)
+{
+    if (seqPool->bufferSize == 0) {
+        return kNullRawSeqStore;
+    }
+    return bufferToSeq(ZSTDMT_getBuffer(seqPool));
+}
+
+#if ZSTD_RESIZE_SEQPOOL
+static rawSeqStore_t ZSTDMT_resizeSeq(ZSTDMT_seqPool* seqPool, rawSeqStore_t seq)
+{
+  return bufferToSeq(ZSTDMT_resizeBuffer(seqPool, seqToBuffer(seq)));
+}
+#endif
+
+static void ZSTDMT_releaseSeq(ZSTDMT_seqPool* seqPool, rawSeqStore_t seq)
+{
+  ZSTDMT_releaseBuffer(seqPool, seqToBuffer(seq));
+}
+
+static void ZSTDMT_setNbSeq(ZSTDMT_seqPool* const seqPool, size_t const nbSeq)
+{
+  ZSTDMT_setBufferSize(seqPool, nbSeq * sizeof(rawSeq));
+}
+
+static ZSTDMT_seqPool* ZSTDMT_createSeqPool(unsigned nbWorkers, ZSTD_customMem cMem)
+{
+    ZSTDMT_seqPool* seqPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    ZSTDMT_setNbSeq(seqPool, 0);
+    return seqPool;
+}
+
+static void ZSTDMT_freeSeqPool(ZSTDMT_seqPool* seqPool)
+{
+    ZSTDMT_freeBufferPool(seqPool);
+}
+
+
+
 /* =====   CCtx Pool   ===== */
+/* a single CCtx Pool can be invoked from multiple threads in parallel */
 
 typedef struct {
+    ZSTD_pthread_mutex_t poolMutex;
     unsigned totalCCtx;
     unsigned availCCtx;
+    ZSTD_customMem cMem;
     ZSTD_CCtx* cctx[1];   /* variable size */
 } ZSTDMT_CCtxPool;
 
-/* assumption : CCtxPool invocation only from main thread */
-
 /* note : all CCtx borrowed from the pool should be released back to the pool _before_ freeing the pool */
 static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
 {
     unsigned u;
     for (u=0; u<pool->totalCCtx; u++)
         ZSTD_freeCCtx(pool->cctx[u]);  /* note : compatible with free on NULL */
-    free(pool);
+    ZSTD_pthread_mutex_destroy(&pool->poolMutex);
+    ZSTD_free(pool, pool->cMem);
 }
 
 /* ZSTDMT_createCCtxPool() :
- * implies nbThreads >= 1 , checked by caller ZSTDMT_createCCtx() */
-static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbThreads)
+ * implies nbWorkers >= 1 , checked by caller ZSTDMT_createCCtx() */
+static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbWorkers,
+                                              ZSTD_customMem cMem)
 {
-    ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) calloc(1, sizeof(ZSTDMT_CCtxPool) + (nbThreads-1)*sizeof(ZSTD_CCtx*));
+    ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_calloc(
+        sizeof(ZSTDMT_CCtxPool) + (nbWorkers-1)*sizeof(ZSTD_CCtx*), cMem);
+    assert(nbWorkers > 0);
     if (!cctxPool) return NULL;
-    cctxPool->totalCCtx = nbThreads;
+    if (ZSTD_pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
+        ZSTD_free(cctxPool, cMem);
+        return NULL;
+    }
+    cctxPool->cMem = cMem;
+    cctxPool->totalCCtx = nbWorkers;
     cctxPool->availCCtx = 1;   /* at least one cctx for single-thread mode */
-    cctxPool->cctx[0] = ZSTD_createCCtx();
+    cctxPool->cctx[0] = ZSTD_createCCtx_advanced(cMem);
     if (!cctxPool->cctx[0]) { ZSTDMT_freeCCtxPool(cctxPool); return NULL; }
-    DEBUGLOG(1, "cctxPool created, with %u threads", nbThreads);
+    DEBUGLOG(3, "cctxPool created, with %u workers", nbWorkers);
     return cctxPool;
 }
 
-static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* pool)
+/* only works during initialization phase, not during compression */
+static size_t ZSTDMT_sizeof_CCtxPool(ZSTDMT_CCtxPool* cctxPool)
 {
-    if (pool->availCCtx) {
-        pool->availCCtx--;
-        return pool->cctx[pool->availCCtx];
+    ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
+    {   unsigned const nbWorkers = cctxPool->totalCCtx;
+        size_t const poolSize = sizeof(*cctxPool)
+                                + (nbWorkers-1) * sizeof(ZSTD_CCtx*);
+        unsigned u;
+        size_t totalCCtxSize = 0;
+        for (u=0; u<nbWorkers; u++) {
+            totalCCtxSize += ZSTD_sizeof_CCtx(cctxPool->cctx[u]);
+        }
+        ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+        assert(nbWorkers > 0);
+        return poolSize + totalCCtxSize;
     }
-    return ZSTD_createCCtx();   /* note : can be NULL, when creation fails ! */
+}
+
+static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* cctxPool)
+{
+    DEBUGLOG(5, "ZSTDMT_getCCtx");
+    ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
+    if (cctxPool->availCCtx) {
+        cctxPool->availCCtx--;
+        {   ZSTD_CCtx* const cctx = cctxPool->cctx[cctxPool->availCCtx];
+            ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+            return cctx;
+    }   }
+    ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+    DEBUGLOG(5, "create one more CCtx");
+    return ZSTD_createCCtx_advanced(cctxPool->cMem);   /* note : can be NULL, when creation fails ! */
 }
 
 static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx)
 {
     if (cctx==NULL) return;   /* compatibility with release on NULL */
+    ZSTD_pthread_mutex_lock(&pool->poolMutex);
     if (pool->availCCtx < pool->totalCCtx)
         pool->cctx[pool->availCCtx++] = cctx;
-    else
-        /* pool overflow : should not happen, since totalCCtx==nbThreads */
+    else {
+        /* pool overflow : should not happen, since totalCCtx==nbWorkers */
+        DEBUGLOG(4, "CCtx pool overflow : free cctx");
         ZSTD_freeCCtx(cctx);
+    }
+    ZSTD_pthread_mutex_unlock(&pool->poolMutex);
+}
+
+/* ====   Serial State   ==== */
+
+typedef struct {
+    void const* start;
+    size_t size;
+} range_t;
+
+typedef struct {
+    /* All variables in the struct are protected by mutex. */
+    ZSTD_pthread_mutex_t mutex;
+    ZSTD_pthread_cond_t cond;
+    ZSTD_CCtx_params params;
+    ldmState_t ldmState;
+    XXH64_state_t xxhState;
+    unsigned nextJobID;
+    /* Protects ldmWindow.
+     * Must be acquired after the main mutex when acquiring both.
+     */
+    ZSTD_pthread_mutex_t ldmWindowMutex;
+    ZSTD_pthread_cond_t ldmWindowCond;  /* Signaled when ldmWindow is udpated */
+    ZSTD_window_t ldmWindow;  /* A thread-safe copy of ldmState.window */
+} serialState_t;
+
+static int ZSTDMT_serialState_reset(serialState_t* serialState, ZSTDMT_seqPool* seqPool, ZSTD_CCtx_params params)
+{
+    /* Adjust parameters */
+    if (params.ldmParams.enableLdm) {
+        DEBUGLOG(4, "LDM window size = %u KB", (1U << params.cParams.windowLog) >> 10);
+        params.ldmParams.windowLog = params.cParams.windowLog;
+        ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
+        assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
+        assert(params.ldmParams.hashEveryLog < 32);
+        serialState->ldmState.hashPower =
+                ZSTD_ldm_getHashPower(params.ldmParams.minMatchLength);
+    } else {
+        memset(&params.ldmParams, 0, sizeof(params.ldmParams));
+    }
+    serialState->nextJobID = 0;
+    if (params.fParams.checksumFlag)
+        XXH64_reset(&serialState->xxhState, 0);
+    if (params.ldmParams.enableLdm) {
+        ZSTD_customMem cMem = params.customMem;
+        unsigned const hashLog = params.ldmParams.hashLog;
+        size_t const hashSize = ((size_t)1 << hashLog) * sizeof(ldmEntry_t);
+        unsigned const bucketLog =
+            params.ldmParams.hashLog - params.ldmParams.bucketSizeLog;
+        size_t const bucketSize = (size_t)1 << bucketLog;
+        unsigned const prevBucketLog =
+            serialState->params.ldmParams.hashLog -
+            serialState->params.ldmParams.bucketSizeLog;
+        /* Size the seq pool tables */
+        ZSTDMT_setNbSeq(seqPool, ZSTD_ldm_getMaxNbSeq(params.ldmParams, params.jobSize));
+        /* Reset the window */
+        ZSTD_window_clear(&serialState->ldmState.window);
+        serialState->ldmWindow = serialState->ldmState.window;
+        /* Resize tables and output space if necessary. */
+        if (serialState->ldmState.hashTable == NULL || serialState->params.ldmParams.hashLog < hashLog) {
+            ZSTD_free(serialState->ldmState.hashTable, cMem);
+            serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_malloc(hashSize, cMem);
+        }
+        if (serialState->ldmState.bucketOffsets == NULL || prevBucketLog < bucketLog) {
+            ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
+            serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_malloc(bucketSize, cMem);
+        }
+        if (!serialState->ldmState.hashTable || !serialState->ldmState.bucketOffsets)
+            return 1;
+        /* Zero the tables */
+        memset(serialState->ldmState.hashTable, 0, hashSize);
+        memset(serialState->ldmState.bucketOffsets, 0, bucketSize);
+    }
+    serialState->params = params;
+    return 0;
+}
+
+static int ZSTDMT_serialState_init(serialState_t* serialState)
+{
+    int initError = 0;
+    memset(serialState, 0, sizeof(*serialState));
+    initError |= ZSTD_pthread_mutex_init(&serialState->mutex, NULL);
+    initError |= ZSTD_pthread_cond_init(&serialState->cond, NULL);
+    initError |= ZSTD_pthread_mutex_init(&serialState->ldmWindowMutex, NULL);
+    initError |= ZSTD_pthread_cond_init(&serialState->ldmWindowCond, NULL);
+    return initError;
+}
+
+static void ZSTDMT_serialState_free(serialState_t* serialState)
+{
+    ZSTD_customMem cMem = serialState->params.customMem;
+    ZSTD_pthread_mutex_destroy(&serialState->mutex);
+    ZSTD_pthread_cond_destroy(&serialState->cond);
+    ZSTD_pthread_mutex_destroy(&serialState->ldmWindowMutex);
+    ZSTD_pthread_cond_destroy(&serialState->ldmWindowCond);
+    ZSTD_free(serialState->ldmState.hashTable, cMem);
+    ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
+}
+
+static void ZSTDMT_serialState_update(serialState_t* serialState,
+                                      ZSTD_CCtx* jobCCtx, rawSeqStore_t seqStore,
+                                      range_t src, unsigned jobID)
+{
+    /* Wait for our turn */
+    ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
+    while (serialState->nextJobID < jobID) {
+        ZSTD_pthread_cond_wait(&serialState->cond, &serialState->mutex);
+    }
+    /* A future job may error and skip our job */
+    if (serialState->nextJobID == jobID) {
+        /* It is now our turn, do any processing necessary */
+        if (serialState->params.ldmParams.enableLdm) {
+            size_t error;
+            assert(seqStore.seq != NULL && seqStore.pos == 0 &&
+                   seqStore.size == 0 && seqStore.capacity > 0);
+            ZSTD_window_update(&serialState->ldmState.window, src.start, src.size);
+            error = ZSTD_ldm_generateSequences(
+                &serialState->ldmState, &seqStore,
+                &serialState->params.ldmParams, src.start, src.size);
+            /* We provide a large enough buffer to never fail. */
+            assert(!ZSTD_isError(error)); (void)error;
+            /* Update ldmWindow to match the ldmState.window and signal the main
+             * thread if it is waiting for a buffer.
+             */
+            ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+            serialState->ldmWindow = serialState->ldmState.window;
+            ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+            ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
+        }
+        if (serialState->params.fParams.checksumFlag && src.size > 0)
+            XXH64_update(&serialState->xxhState, src.start, src.size);
+    }
+    /* Now it is the next jobs turn */
+    serialState->nextJobID++;
+    ZSTD_pthread_cond_broadcast(&serialState->cond);
+    ZSTD_pthread_mutex_unlock(&serialState->mutex);
+
+    if (seqStore.size > 0) {
+        size_t const err = ZSTD_referenceExternalSequences(
+            jobCCtx, seqStore.seq, seqStore.size);
+        assert(serialState->params.ldmParams.enableLdm);
+        assert(!ZSTD_isError(err));
+        (void)err;
+    }
+}
+
+static void ZSTDMT_serialState_ensureFinished(serialState_t* serialState,
+                                              unsigned jobID, size_t cSize)
+{
+    ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
+    if (serialState->nextJobID <= jobID) {
+        assert(ZSTD_isError(cSize)); (void)cSize;
+        DEBUGLOG(5, "Skipping past job %u because of error", jobID);
+        serialState->nextJobID = jobID + 1;
+        ZSTD_pthread_cond_broadcast(&serialState->cond);
+
+        ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+        ZSTD_window_clear(&serialState->ldmWindow);
+        ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+        ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
+    }
+    ZSTD_pthread_mutex_unlock(&serialState->mutex);
+
 }
 
 
-/* =====   Thread worker   ===== */
+/* ------------------------------------------ */
+/* =====          Worker thread         ===== */
+/* ------------------------------------------ */
 
-typedef struct {
-    buffer_t buffer;
-    size_t filled;
-} inBuff_t;
+static const range_t kNullRange = { NULL, 0 };
 
 typedef struct {
-    ZSTD_CCtx* cctx;
-    buffer_t src;
-    const void* srcStart;
-    size_t   srcSize;
-    size_t   dictSize;
-    buffer_t dstBuff;
-    size_t   cSize;
-    size_t   dstFlushed;
-    unsigned firstChunk;
-    unsigned lastChunk;
-    unsigned jobCompleted;
-    unsigned jobScanned;
-    pthread_mutex_t* jobCompleted_mutex;
-    pthread_cond_t* jobCompleted_cond;
-    ZSTD_parameters params;
-    ZSTD_CDict* cdict;
-    unsigned long long fullFrameSize;
+    size_t   consumed;                   /* SHARED - set0 by mtctx, then modified by worker AND read by mtctx */
+    size_t   cSize;                      /* SHARED - set0 by mtctx, then modified by worker AND read by mtctx, then set0 by mtctx */
+    ZSTD_pthread_mutex_t job_mutex;      /* Thread-safe - used by mtctx and worker */
+    ZSTD_pthread_cond_t job_cond;        /* Thread-safe - used by mtctx and worker */
+    ZSTDMT_CCtxPool* cctxPool;           /* Thread-safe - used by mtctx and (all) workers */
+    ZSTDMT_bufferPool* bufPool;          /* Thread-safe - used by mtctx and (all) workers */
+    ZSTDMT_seqPool* seqPool;             /* Thread-safe - used by mtctx and (all) workers */
+    serialState_t* serial;               /* Thread-safe - used by mtctx and (all) workers */
+    buffer_t dstBuff;                    /* set by worker (or mtctx), then read by worker & mtctx, then modified by mtctx => no barrier */
+    range_t prefix;                      /* set by mtctx, then read by worker & mtctx => no barrier */
+    range_t src;                         /* set by mtctx, then read by worker & mtctx => no barrier */
+    unsigned jobID;                      /* set by mtctx, then read by worker => no barrier */
+    unsigned firstJob;                   /* set by mtctx, then read by worker => no barrier */
+    unsigned lastJob;                    /* set by mtctx, then read by worker => no barrier */
+    ZSTD_CCtx_params params;             /* set by mtctx, then read by worker => no barrier */
+    const ZSTD_CDict* cdict;             /* set by mtctx, then read by worker => no barrier */
+    unsigned long long fullFrameSize;    /* set by mtctx, then read by worker => no barrier */
+    size_t   dstFlushed;                 /* used only by mtctx */
+    unsigned frameChecksumNeeded;        /* used only by mtctx */
 } ZSTDMT_jobDescription;
 
-/* ZSTDMT_compressChunk() : POOL_function type */
-void ZSTDMT_compressChunk(void* jobDescription)
+/* ZSTDMT_compressionJob() is a POOL_function type */
+void ZSTDMT_compressionJob(void* jobDescription)
 {
     ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
-    const void* const src = (const char*)job->srcStart + job->dictSize;
-    buffer_t const dstBuff = job->dstBuff;
-    DEBUGLOG(3, "job (first:%u) (last:%u) : dictSize %u, srcSize %u", job->firstChunk, job->lastChunk, (U32)job->dictSize, (U32)job->srcSize);
-    if (job->cdict) {
-        size_t const initError = ZSTD_compressBegin_usingCDict(job->cctx, job->cdict, job->fullFrameSize);
-        if (job->cdict) DEBUGLOG(3, "using CDict ");
-        if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }
-    } else {
-        size_t const initError = ZSTD_compressBegin_advanced(job->cctx, job->srcStart, job->dictSize, job->params, job->fullFrameSize);
-        if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }
-        ZSTD_setCCtxParameter(job->cctx, ZSTD_p_forceWindow, 1);
+    ZSTD_CCtx_params jobParams = job->params;   /* do not modify job->params ! copy it, modify the copy */
+    ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(job->cctxPool);
+    rawSeqStore_t rawSeqStore = ZSTDMT_getSeq(job->seqPool);
+    buffer_t dstBuff = job->dstBuff;
+
+    /* Don't compute the checksum for chunks, since we compute it externally,
+     * but write it in the header.
+     */
+    if (job->jobID != 0) jobParams.fParams.checksumFlag = 0;
+    /* Don't run LDM for the chunks, since we handle it externally */
+    jobParams.ldmParams.enableLdm = 0;
+
+    /* ressources */
+    if (cctx==NULL) {
+        job->cSize = ERROR(memory_allocation);
+        goto _endJob;
     }
-    if (!job->firstChunk) {  /* flush frame header */
-        size_t const hSize = ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, 0);
-        if (ZSTD_isError(hSize)) { job->cSize = hSize; goto _endJob; }
-        ZSTD_invalidateRepCodes(job->cctx);
+    if (dstBuff.start == NULL) {   /* streaming job : doesn't provide a dstBuffer */
+        dstBuff = ZSTDMT_getBuffer(job->bufPool);
+        if (dstBuff.start==NULL) {
+            job->cSize = ERROR(memory_allocation);
+            goto _endJob;
+        }
+        job->dstBuff = dstBuff;   /* this value can be read in ZSTDMT_flush, when it copies the whole job */
     }
 
-    DEBUGLOG(4, "Compressing : ");
-    DEBUG_PRINTHEX(4, job->srcStart, 12);
-    job->cSize = (job->lastChunk) ?   /* last chunk signal */
-                 ZSTD_compressEnd     (job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize) :
-                 ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize);
-    DEBUGLOG(3, "compressed %u bytes into %u bytes   (first:%u) (last:%u)", (unsigned)job->srcSize, (unsigned)job->cSize, job->firstChunk, job->lastChunk);
+    /* init */
+    if (job->cdict) {
+        size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, NULL, 0, ZSTD_dct_auto, job->cdict, jobParams, job->fullFrameSize);
+        assert(job->firstJob);  /* only allowed for first job */
+        if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }
+    } else {  /* srcStart points at reloaded section */
+        U64 const pledgedSrcSize = job->firstJob ? job->fullFrameSize : job->src.size;
+        {   size_t const forceWindowError = ZSTD_CCtxParam_setParameter(&jobParams, ZSTD_p_forceMaxWindow, !job->firstJob);
+            if (ZSTD_isError(forceWindowError)) {
+                job->cSize = forceWindowError;
+                goto _endJob;
+        }   }
+        {   size_t const initError = ZSTD_compressBegin_advanced_internal(cctx,
+                                        job->prefix.start, job->prefix.size, ZSTD_dct_rawContent, /* load dictionary in "content-only" mode (no header analysis) */
+                                        NULL, /*cdict*/
+                                        jobParams, pledgedSrcSize);
+            if (ZSTD_isError(initError)) {
+                job->cSize = initError;
+                goto _endJob;
+    }   }   }
+
+    /* Perform serial step as early as possible, but after CCtx initialization */
+    ZSTDMT_serialState_update(job->serial, cctx, rawSeqStore, job->src, job->jobID);
+
+    if (!job->firstJob) {  /* flush and overwrite frame header when it's not first job */
+        size_t const hSize = ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.capacity, job->src.start, 0);
+        if (ZSTD_isError(hSize)) { job->cSize = hSize; /* save error code */ goto _endJob; }
+        DEBUGLOG(5, "ZSTDMT_compressionJob: flush and overwrite %u bytes of frame header (not first job)", (U32)hSize);
+        ZSTD_invalidateRepCodes(cctx);
+    }
+
+    /* compress */
+    {   size_t const chunkSize = 4*ZSTD_BLOCKSIZE_MAX;
+        int const nbChunks = (int)((job->src.size + (chunkSize-1)) / chunkSize);
+        const BYTE* ip = (const BYTE*) job->src.start;
+        BYTE* const ostart = (BYTE*)dstBuff.start;
+        BYTE* op = ostart;
+        BYTE* oend = op + dstBuff.capacity;
+        int chunkNb;
+        if (sizeof(size_t) > sizeof(int)) assert(job->src.size < ((size_t)INT_MAX) * chunkSize);   /* check overflow */
+        DEBUGLOG(5, "ZSTDMT_compressionJob: compress %u bytes in %i blocks", (U32)job->src.size, nbChunks);
+        assert(job->cSize == 0);
+        for (chunkNb = 1; chunkNb < nbChunks; chunkNb++) {
+            size_t const cSize = ZSTD_compressContinue(cctx, op, oend-op, ip, chunkSize);
+            if (ZSTD_isError(cSize)) { job->cSize = cSize; goto _endJob; }
+            ip += chunkSize;
+            op += cSize; assert(op < oend);
+            /* stats */
+            ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+            job->cSize += cSize;
+            job->consumed = chunkSize * chunkNb;
+            DEBUGLOG(5, "ZSTDMT_compressionJob: compress new block : cSize==%u bytes (total: %u)",
+                        (U32)cSize, (U32)job->cSize);
+            ZSTD_pthread_cond_signal(&job->job_cond);   /* warns some more data is ready to be flushed */
+            ZSTD_pthread_mutex_unlock(&job->job_mutex);
+        }
+        /* last block */
+        assert(chunkSize > 0); assert((chunkSize & (chunkSize - 1)) == 0);  /* chunkSize must be power of 2 for mask==(chunkSize-1) to work */
+        if ((nbChunks > 0) | job->lastJob /*must output a "last block" flag*/ ) {
+            size_t const lastBlockSize1 = job->src.size & (chunkSize-1);
+            size_t const lastBlockSize = ((lastBlockSize1==0) & (job->src.size>=chunkSize)) ? chunkSize : lastBlockSize1;
+            size_t const cSize = (job->lastJob) ?
+                 ZSTD_compressEnd     (cctx, op, oend-op, ip, lastBlockSize) :
+                 ZSTD_compressContinue(cctx, op, oend-op, ip, lastBlockSize);
+            if (ZSTD_isError(cSize)) { job->cSize = cSize; goto _endJob; }
+            /* stats */
+            ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+            job->cSize += cSize;
+            ZSTD_pthread_mutex_unlock(&job->job_mutex);
+    }   }
 
 _endJob:
-    PTHREAD_MUTEX_LOCK(job->jobCompleted_mutex);
-    job->jobCompleted = 1;
-    job->jobScanned = 0;
-    pthread_cond_signal(job->jobCompleted_cond);
-    pthread_mutex_unlock(job->jobCompleted_mutex);
+    ZSTDMT_serialState_ensureFinished(job->serial, job->jobID, job->cSize);
+    if (job->prefix.size > 0)
+        DEBUGLOG(5, "Finished with prefix: %zx", (size_t)job->prefix.start);
+    DEBUGLOG(5, "Finished with source: %zx", (size_t)job->src.start);
+    /* release resources */
+    ZSTDMT_releaseSeq(job->seqPool, rawSeqStore);
+    ZSTDMT_releaseCCtx(job->cctxPool, cctx);
+    /* report */
+    ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+    job->consumed = job->src.size;
+    ZSTD_pthread_cond_signal(&job->job_cond);
+    ZSTD_pthread_mutex_unlock(&job->job_mutex);
 }
 
 
@@ -266,213 +714,500 @@
 /* =====   Multi-threaded compression   ===== */
 /* ------------------------------------------ */
 
+typedef struct {
+    range_t prefix;         /* read-only non-owned prefix buffer */
+    buffer_t buffer;
+    size_t filled;
+} inBuff_t;
+
+typedef struct {
+  BYTE* buffer;     /* The round input buffer. All jobs get references
+                     * to pieces of the buffer. ZSTDMT_tryGetInputRange()
+                     * handles handing out job input buffers, and makes
+                     * sure it doesn't overlap with any pieces still in use.
+                     */
+  size_t capacity;  /* The capacity of buffer. */
+  size_t pos;       /* The position of the current inBuff in the round
+                     * buffer. Updated past the end if the inBuff once
+                     * the inBuff is sent to the worker thread.
+                     * pos <= capacity.
+                     */
+} roundBuff_t;
+
+static const roundBuff_t kNullRoundBuff = {NULL, 0, 0};
+
 struct ZSTDMT_CCtx_s {
     POOL_ctx* factory;
-    ZSTDMT_bufferPool* buffPool;
+    ZSTDMT_jobDescription* jobs;
+    ZSTDMT_bufferPool* bufPool;
     ZSTDMT_CCtxPool* cctxPool;
-    pthread_mutex_t jobCompleted_mutex;
-    pthread_cond_t jobCompleted_cond;
+    ZSTDMT_seqPool* seqPool;
+    ZSTD_CCtx_params params;
     size_t targetSectionSize;
-    size_t marginSize;
-    size_t inBuffSize;
-    size_t dictSize;
-    size_t targetDictSize;
+    size_t targetPrefixSize;
+    roundBuff_t roundBuff;
     inBuff_t inBuff;
-    ZSTD_parameters params;
-    XXH64_state_t xxhState;
-    unsigned nbThreads;
+    int jobReady;        /* 1 => one job is already prepared, but pool has shortage of workers. Don't create another one. */
+    serialState_t serial;
+    unsigned singleBlockingThread;
     unsigned jobIDMask;
     unsigned doneJobID;
     unsigned nextJobID;
     unsigned frameEnded;
     unsigned allJobsCompleted;
-    unsigned overlapRLog;
     unsigned long long frameContentSize;
-    size_t sectionSize;
-    ZSTD_CDict* cdict;
-    ZSTD_CStream* cstream;
-    ZSTDMT_jobDescription jobs[1];   /* variable size (must lies at the end) */
+    unsigned long long consumed;
+    unsigned long long produced;
+    ZSTD_customMem cMem;
+    ZSTD_CDict* cdictLocal;
+    const ZSTD_CDict* cdict;
 };
 
-ZSTDMT_CCtx *ZSTDMT_createCCtx(unsigned nbThreads)
+static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZSTD_customMem cMem)
 {
-    ZSTDMT_CCtx* cctx;
-    U32 const minNbJobs = nbThreads + 2;
-    U32 const nbJobsLog2 = ZSTD_highbit32(minNbJobs) + 1;
+    U32 jobNb;
+    if (jobTable == NULL) return;
+    for (jobNb=0; jobNb<nbJobs; jobNb++) {
+        ZSTD_pthread_mutex_destroy(&jobTable[jobNb].job_mutex);
+        ZSTD_pthread_cond_destroy(&jobTable[jobNb].job_cond);
+    }
+    ZSTD_free(jobTable, cMem);
+}
+
+/* ZSTDMT_allocJobsTable()
+ * allocate and init a job table.
+ * update *nbJobsPtr to next power of 2 value, as size of table */
+static ZSTDMT_jobDescription* ZSTDMT_createJobsTable(U32* nbJobsPtr, ZSTD_customMem cMem)
+{
+    U32 const nbJobsLog2 = ZSTD_highbit32(*nbJobsPtr) + 1;
     U32 const nbJobs = 1 << nbJobsLog2;
-    DEBUGLOG(5, "nbThreads : %u  ; minNbJobs : %u ;  nbJobsLog2 : %u ;  nbJobs : %u  \n",
-            nbThreads, minNbJobs, nbJobsLog2, nbJobs);
-    if ((nbThreads < 1) | (nbThreads > ZSTDMT_NBTHREADS_MAX)) return NULL;
-    cctx = (ZSTDMT_CCtx*) calloc(1, sizeof(ZSTDMT_CCtx) + nbJobs*sizeof(ZSTDMT_jobDescription));
-    if (!cctx) return NULL;
-    cctx->nbThreads = nbThreads;
-    cctx->jobIDMask = nbJobs - 1;
-    cctx->allJobsCompleted = 1;
-    cctx->sectionSize = 0;
-    cctx->overlapRLog = 3;
-    cctx->factory = POOL_create(nbThreads, 1);
-    cctx->buffPool = ZSTDMT_createBufferPool(nbThreads);
-    cctx->cctxPool = ZSTDMT_createCCtxPool(nbThreads);
-    if (!cctx->factory | !cctx->buffPool | !cctx->cctxPool) {  /* one object was not created */
-        ZSTDMT_freeCCtx(cctx);
+    U32 jobNb;
+    ZSTDMT_jobDescription* const jobTable = (ZSTDMT_jobDescription*)
+                ZSTD_calloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
+    int initError = 0;
+    if (jobTable==NULL) return NULL;
+    *nbJobsPtr = nbJobs;
+    for (jobNb=0; jobNb<nbJobs; jobNb++) {
+        initError |= ZSTD_pthread_mutex_init(&jobTable[jobNb].job_mutex, NULL);
+        initError |= ZSTD_pthread_cond_init(&jobTable[jobNb].job_cond, NULL);
+    }
+    if (initError != 0) {
+        ZSTDMT_freeJobsTable(jobTable, nbJobs, cMem);
         return NULL;
     }
-    if (nbThreads==1) {
-        cctx->cstream = ZSTD_createCStream();
-        if (!cctx->cstream) {
-            ZSTDMT_freeCCtx(cctx); return NULL;
-    }   }
-    pthread_mutex_init(&cctx->jobCompleted_mutex, NULL);   /* Todo : check init function return */
-    pthread_cond_init(&cctx->jobCompleted_cond, NULL);
-    DEBUGLOG(4, "mt_cctx created, for %u threads \n", nbThreads);
-    return cctx;
+    return jobTable;
+}
+
+/* ZSTDMT_CCtxParam_setNbWorkers():
+ * Internal use only */
+size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
+{
+    if (nbWorkers > ZSTDMT_NBWORKERS_MAX) nbWorkers = ZSTDMT_NBWORKERS_MAX;
+    params->nbWorkers = nbWorkers;
+    params->overlapSizeLog = ZSTDMT_OVERLAPLOG_DEFAULT;
+    params->jobSize = 0;
+    return nbWorkers;
 }
 
+ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem)
+{
+    ZSTDMT_CCtx* mtctx;
+    U32 nbJobs = nbWorkers + 2;
+    int initError;
+    DEBUGLOG(3, "ZSTDMT_createCCtx_advanced (nbWorkers = %u)", nbWorkers);
+
+    if (nbWorkers < 1) return NULL;
+    nbWorkers = MIN(nbWorkers , ZSTDMT_NBWORKERS_MAX);
+    if ((cMem.customAlloc!=NULL) ^ (cMem.customFree!=NULL))
+        /* invalid custom allocator */
+        return NULL;
+
+    mtctx = (ZSTDMT_CCtx*) ZSTD_calloc(sizeof(ZSTDMT_CCtx), cMem);
+    if (!mtctx) return NULL;
+    ZSTDMT_CCtxParam_setNbWorkers(&mtctx->params, nbWorkers);
+    mtctx->cMem = cMem;
+    mtctx->allJobsCompleted = 1;
+    mtctx->factory = POOL_create_advanced(nbWorkers, 0, cMem);
+    mtctx->jobs = ZSTDMT_createJobsTable(&nbJobs, cMem);
+    assert(nbJobs > 0); assert((nbJobs & (nbJobs - 1)) == 0);  /* ensure nbJobs is a power of 2 */
+    mtctx->jobIDMask = nbJobs - 1;
+    mtctx->bufPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    mtctx->cctxPool = ZSTDMT_createCCtxPool(nbWorkers, cMem);
+    mtctx->seqPool = ZSTDMT_createSeqPool(nbWorkers, cMem);
+    initError = ZSTDMT_serialState_init(&mtctx->serial);
+    mtctx->roundBuff = kNullRoundBuff;
+    if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool | !mtctx->seqPool | initError) {
+        ZSTDMT_freeCCtx(mtctx);
+        return NULL;
+    }
+    DEBUGLOG(3, "mt_cctx created, for %u threads", nbWorkers);
+    return mtctx;
+}
+
+ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbWorkers)
+{
+    return ZSTDMT_createCCtx_advanced(nbWorkers, ZSTD_defaultCMem);
+}
+
+
 /* ZSTDMT_releaseAllJobResources() :
- * Ensure all workers are killed first. */
+ * note : ensure all workers are killed first ! */
 static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
 {
     unsigned jobID;
+    DEBUGLOG(3, "ZSTDMT_releaseAllJobResources");
     for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {
-        ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].dstBuff);
+        DEBUGLOG(4, "job%02u: release dst address %08X", jobID, (U32)(size_t)mtctx->jobs[jobID].dstBuff.start);
+        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
         mtctx->jobs[jobID].dstBuff = g_nullBuffer;
-        ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].src);
-        mtctx->jobs[jobID].src = g_nullBuffer;
-        ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[jobID].cctx);
-        mtctx->jobs[jobID].cctx = NULL;
+        mtctx->jobs[jobID].cSize = 0;
     }
     memset(mtctx->jobs, 0, (mtctx->jobIDMask+1)*sizeof(ZSTDMT_jobDescription));
-    ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->inBuff.buffer);
     mtctx->inBuff.buffer = g_nullBuffer;
+    mtctx->inBuff.filled = 0;
     mtctx->allJobsCompleted = 1;
 }
 
+static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* mtctx)
+{
+    DEBUGLOG(4, "ZSTDMT_waitForAllJobsCompleted");
+    while (mtctx->doneJobID < mtctx->nextJobID) {
+        unsigned const jobID = mtctx->doneJobID & mtctx->jobIDMask;
+        ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[jobID].job_mutex);
+        while (mtctx->jobs[jobID].consumed < mtctx->jobs[jobID].src.size) {
+            DEBUGLOG(5, "waiting for jobCompleted signal from job %u", mtctx->doneJobID);   /* we want to block when waiting for data to flush */
+            ZSTD_pthread_cond_wait(&mtctx->jobs[jobID].job_cond, &mtctx->jobs[jobID].job_mutex);
+        }
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[jobID].job_mutex);
+        mtctx->doneJobID++;
+    }
+}
+
 size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
 {
     if (mtctx==NULL) return 0;   /* compatible with free on NULL */
-    POOL_free(mtctx->factory);
-    if (!mtctx->allJobsCompleted) ZSTDMT_releaseAllJobResources(mtctx); /* stop workers first */
-    ZSTDMT_freeBufferPool(mtctx->buffPool);  /* release job resources into pools first */
+    POOL_free(mtctx->factory);   /* stop and free worker threads */
+    ZSTDMT_releaseAllJobResources(mtctx);  /* release job resources into pools first */
+    ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
+    ZSTDMT_freeBufferPool(mtctx->bufPool);
     ZSTDMT_freeCCtxPool(mtctx->cctxPool);
-    ZSTD_freeCDict(mtctx->cdict);
-    ZSTD_freeCStream(mtctx->cstream);
-    pthread_mutex_destroy(&mtctx->jobCompleted_mutex);
-    pthread_cond_destroy(&mtctx->jobCompleted_cond);
-    free(mtctx);
+    ZSTDMT_freeSeqPool(mtctx->seqPool);
+    ZSTDMT_serialState_free(&mtctx->serial);
+    ZSTD_freeCDict(mtctx->cdictLocal);
+    if (mtctx->roundBuff.buffer)
+        ZSTD_free(mtctx->roundBuff.buffer, mtctx->cMem);
+    ZSTD_free(mtctx, mtctx->cMem);
     return 0;
 }
 
-size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSDTMT_parameter parameter, unsigned value)
+size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
 {
+    if (mtctx == NULL) return 0;   /* supports sizeof NULL */
+    return sizeof(*mtctx)
+            + POOL_sizeof(mtctx->factory)
+            + ZSTDMT_sizeof_bufferPool(mtctx->bufPool)
+            + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription)
+            + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool)
+            + ZSTDMT_sizeof_seqPool(mtctx->seqPool)
+            + ZSTD_sizeof_CDict(mtctx->cdictLocal)
+            + mtctx->roundBuff.capacity;
+}
+
+/* Internal only */
+size_t ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params,
+                                ZSTDMT_parameter parameter, unsigned value) {
+    DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter");
+    switch(parameter)
+    {
+    case ZSTDMT_p_jobSize :
+        DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter : set jobSize to %u", value);
+        if ( (value > 0)  /* value==0 => automatic job size */
+           & (value < ZSTDMT_JOBSIZE_MIN) )
+            value = ZSTDMT_JOBSIZE_MIN;
+        params->jobSize = value;
+        return value;
+    case ZSTDMT_p_overlapSectionLog :
+        if (value > 9) value = 9;
+        DEBUGLOG(4, "ZSTDMT_p_overlapSectionLog : %u", value);
+        params->overlapSizeLog = (value >= 9) ? 9 : value;
+        return value;
+    default :
+        return ERROR(parameter_unsupported);
+    }
+}
+
+size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned value)
+{
+    DEBUGLOG(4, "ZSTDMT_setMTCtxParameter");
     switch(parameter)
     {
-    case ZSTDMT_p_sectionSize :
-        mtctx->sectionSize = value;
-        return 0;
+    case ZSTDMT_p_jobSize :
+        return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
     case ZSTDMT_p_overlapSectionLog :
-    DEBUGLOG(4, "ZSTDMT_p_overlapSectionLog : %u", value);
-        mtctx->overlapRLog = (value >= 9) ? 0 : 9 - value;
-        return 0;
+        return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
     default :
-        return ERROR(compressionParameter_unsupported);
+        return ERROR(parameter_unsupported);
+    }
+}
+
+/* Sets parameters relevant to the compression job,
+ * initializing others to default values. */
+static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(ZSTD_CCtx_params const params)
+{
+    ZSTD_CCtx_params jobParams;
+    memset(&jobParams, 0, sizeof(jobParams));
+
+    jobParams.cParams = params.cParams;
+    jobParams.fParams = params.fParams;
+    jobParams.compressionLevel = params.compressionLevel;
+    jobParams.disableLiteralCompression = params.disableLiteralCompression;
+
+    return jobParams;
+}
+
+/*! ZSTDMT_updateCParams_whileCompressing() :
+ *  Updates only a selected set of compression parameters, to remain compatible with current frame.
+ *  New parameters will be applied to next compression job. */
+void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams)
+{
+    U32 const saved_wlog = mtctx->params.cParams.windowLog;   /* Do not modify windowLog while compressing */
+    int const compressionLevel = cctxParams->compressionLevel;
+    DEBUGLOG(5, "ZSTDMT_updateCParams_whileCompressing (level:%i)",
+                compressionLevel);
+    mtctx->params.compressionLevel = compressionLevel;
+    {   ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, 0, 0);
+        cParams.windowLog = saved_wlog;
+        mtctx->params.cParams = cParams;
     }
 }
 
+/* ZSTDMT_getNbWorkers():
+ * @return nb threads currently active in mtctx.
+ * mtctx must be valid */
+unsigned ZSTDMT_getNbWorkers(const ZSTDMT_CCtx* mtctx)
+{
+    assert(mtctx != NULL);
+    return mtctx->params.nbWorkers;
+}
+
+/* ZSTDMT_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads.
+ * Note : mutex will be acquired during statistics collection. */
+ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx)
+{
+    ZSTD_frameProgression fps;
+    DEBUGLOG(6, "ZSTDMT_getFrameProgression");
+    fps.consumed = mtctx->consumed;
+    fps.produced = mtctx->produced;
+    fps.ingested = mtctx->consumed + mtctx->inBuff.filled;
+    {   unsigned jobNb;
+        unsigned lastJobNb = mtctx->nextJobID + mtctx->jobReady; assert(mtctx->jobReady <= 1);
+        DEBUGLOG(6, "ZSTDMT_getFrameProgression: jobs: from %u to <%u (jobReady:%u)",
+                    mtctx->doneJobID, lastJobNb, mtctx->jobReady)
+        for (jobNb = mtctx->doneJobID ; jobNb < lastJobNb ; jobNb++) {
+            unsigned const wJobID = jobNb & mtctx->jobIDMask;
+            ZSTD_pthread_mutex_lock(&mtctx->jobs[wJobID].job_mutex);
+            {   size_t const cResult = mtctx->jobs[wJobID].cSize;
+                size_t const produced = ZSTD_isError(cResult) ? 0 : cResult;
+                fps.consumed += mtctx->jobs[wJobID].consumed;
+                fps.ingested += mtctx->jobs[wJobID].src.size;
+                fps.produced += produced;
+            }
+            ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+        }
+    }
+    return fps;
+}
+
 
 /* ------------------------------------------ */
 /* =====   Multi-threaded compression   ===== */
 /* ------------------------------------------ */
 
+static size_t ZSTDMT_computeTargetJobLog(ZSTD_CCtx_params const params)
+{
+    if (params.ldmParams.enableLdm)
+        return MAX(21, params.cParams.chainLog + 4);
+    return MAX(20, params.cParams.windowLog + 2);
+}
+
+static size_t ZSTDMT_computeOverlapLog(ZSTD_CCtx_params const params)
+{
+    unsigned const overlapRLog = (params.overlapSizeLog>9) ? 0 : 9-params.overlapSizeLog;
+    if (params.ldmParams.enableLdm)
+        return (MIN(params.cParams.windowLog, ZSTDMT_computeTargetJobLog(params) - 2) - overlapRLog);
+    return overlapRLog >= 9 ? 0 : (params.cParams.windowLog - overlapRLog);
+}
+
+static unsigned ZSTDMT_computeNbJobs(ZSTD_CCtx_params params, size_t srcSize, unsigned nbWorkers) {
+    assert(nbWorkers>0);
+    {   size_t const jobSizeTarget = (size_t)1 << ZSTDMT_computeTargetJobLog(params);
+        size_t const jobMaxSize = jobSizeTarget << 2;
+        size_t const passSizeMax = jobMaxSize * nbWorkers;
+        unsigned const multiplier = (unsigned)(srcSize / passSizeMax) + 1;
+        unsigned const nbJobsLarge = multiplier * nbWorkers;
+        unsigned const nbJobsMax = (unsigned)(srcSize / jobSizeTarget) + 1;
+        unsigned const nbJobsSmall = MIN(nbJobsMax, nbWorkers);
+        return (multiplier>1) ? nbJobsLarge : nbJobsSmall;
+}   }
+
+/* ZSTDMT_compress_advanced_internal() :
+ * This is a blocking function : it will only give back control to caller after finishing its compression job.
+ */
+static size_t ZSTDMT_compress_advanced_internal(
+                ZSTDMT_CCtx* mtctx,
+                void* dst, size_t dstCapacity,
+          const void* src, size_t srcSize,
+          const ZSTD_CDict* cdict,
+                ZSTD_CCtx_params params)
+{
+    ZSTD_CCtx_params const jobParams = ZSTDMT_initJobCCtxParams(params);
+    size_t const overlapSize = (size_t)1 << ZSTDMT_computeOverlapLog(params);
+    unsigned const nbJobs = ZSTDMT_computeNbJobs(params, srcSize, params.nbWorkers);
+    size_t const proposedJobSize = (srcSize + (nbJobs-1)) / nbJobs;
+    size_t const avgJobSize = (((proposedJobSize-1) & 0x1FFFF) < 0x7FFF) ? proposedJobSize + 0xFFFF : proposedJobSize;   /* avoid too small last block */
+    const char* const srcStart = (const char*)src;
+    size_t remainingSrcSize = srcSize;
+    unsigned const compressWithinDst = (dstCapacity >= ZSTD_compressBound(srcSize)) ? nbJobs : (unsigned)(dstCapacity / ZSTD_compressBound(avgJobSize));  /* presumes avgJobSize >= 256 KB, which should be the case */
+    size_t frameStartPos = 0, dstBufferPos = 0;
+    assert(jobParams.nbWorkers == 0);
+    assert(mtctx->cctxPool->totalCCtx == params.nbWorkers);
+
+    params.jobSize = (U32)avgJobSize;
+    DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: nbJobs=%2u (rawSize=%u bytes; fixedSize=%u) ",
+                nbJobs, (U32)proposedJobSize, (U32)avgJobSize);
+
+    if ((nbJobs==1) | (params.nbWorkers<=1)) {   /* fallback to single-thread mode : this is a blocking invocation anyway */
+        ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];
+        DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: fallback to single-thread mode");
+        if (cdict) return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, jobParams.fParams);
+        return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, NULL, 0, jobParams);
+    }
+
+    assert(avgJobSize >= 256 KB);  /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), required to compress directly into Dst (no additional buffer) */
+    ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgJobSize) );
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params))
+        return ERROR(memory_allocation);
+
+    if (nbJobs > mtctx->jobIDMask+1) {  /* enlarge job table */
+        U32 jobsTableSize = nbJobs;
+        ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
+        mtctx->jobIDMask = 0;
+        mtctx->jobs = ZSTDMT_createJobsTable(&jobsTableSize, mtctx->cMem);
+        if (mtctx->jobs==NULL) return ERROR(memory_allocation);
+        assert((jobsTableSize != 0) && ((jobsTableSize & (jobsTableSize - 1)) == 0));  /* ensure jobsTableSize is a power of 2 */
+        mtctx->jobIDMask = jobsTableSize - 1;
+    }
+
+    {   unsigned u;
+        for (u=0; u<nbJobs; u++) {
+            size_t const jobSize = MIN(remainingSrcSize, avgJobSize);
+            size_t const dstBufferCapacity = ZSTD_compressBound(jobSize);
+            buffer_t const dstAsBuffer = { (char*)dst + dstBufferPos, dstBufferCapacity };
+            buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : g_nullBuffer;
+            size_t dictSize = u ? overlapSize : 0;
+
+            mtctx->jobs[u].prefix.start = srcStart + frameStartPos - dictSize;
+            mtctx->jobs[u].prefix.size = dictSize;
+            mtctx->jobs[u].src.start = srcStart + frameStartPos;
+            mtctx->jobs[u].src.size = jobSize; assert(jobSize > 0);  /* avoid job.src.size == 0 */
+            mtctx->jobs[u].consumed = 0;
+            mtctx->jobs[u].cSize = 0;
+            mtctx->jobs[u].cdict = (u==0) ? cdict : NULL;
+            mtctx->jobs[u].fullFrameSize = srcSize;
+            mtctx->jobs[u].params = jobParams;
+            /* do not calculate checksum within sections, but write it in header for first section */
+            mtctx->jobs[u].dstBuff = dstBuffer;
+            mtctx->jobs[u].cctxPool = mtctx->cctxPool;
+            mtctx->jobs[u].bufPool = mtctx->bufPool;
+            mtctx->jobs[u].seqPool = mtctx->seqPool;
+            mtctx->jobs[u].serial = &mtctx->serial;
+            mtctx->jobs[u].jobID = u;
+            mtctx->jobs[u].firstJob = (u==0);
+            mtctx->jobs[u].lastJob = (u==nbJobs-1);
+
+            DEBUGLOG(5, "ZSTDMT_compress_advanced_internal: posting job %u  (%u bytes)", u, (U32)jobSize);
+            DEBUG_PRINTHEX(6, mtctx->jobs[u].prefix.start, 12);
+            POOL_add(mtctx->factory, ZSTDMT_compressionJob, &mtctx->jobs[u]);
+
+            frameStartPos += jobSize;
+            dstBufferPos += dstBufferCapacity;
+            remainingSrcSize -= jobSize;
+    }   }
+
+    /* collect result */
+    {   size_t error = 0, dstPos = 0;
+        unsigned jobID;
+        for (jobID=0; jobID<nbJobs; jobID++) {
+            DEBUGLOG(5, "waiting for job %u ", jobID);
+            ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[jobID].job_mutex);
+            while (mtctx->jobs[jobID].consumed < mtctx->jobs[jobID].src.size) {
+                DEBUGLOG(5, "waiting for jobCompleted signal from job %u", jobID);
+                ZSTD_pthread_cond_wait(&mtctx->jobs[jobID].job_cond, &mtctx->jobs[jobID].job_mutex);
+            }
+            ZSTD_pthread_mutex_unlock(&mtctx->jobs[jobID].job_mutex);
+            DEBUGLOG(5, "ready to write job %u ", jobID);
+
+            {   size_t const cSize = mtctx->jobs[jobID].cSize;
+                if (ZSTD_isError(cSize)) error = cSize;
+                if ((!error) && (dstPos + cSize > dstCapacity)) error = ERROR(dstSize_tooSmall);
+                if (jobID) {   /* note : job 0 is written directly at dst, which is correct position */
+                    if (!error)
+                        memmove((char*)dst + dstPos, mtctx->jobs[jobID].dstBuff.start, cSize);  /* may overlap when job compressed within dst */
+                    if (jobID >= compressWithinDst) {  /* job compressed into its own buffer, which must be released */
+                        DEBUGLOG(5, "releasing buffer %u>=%u", jobID, compressWithinDst);
+                        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
+                }   }
+                mtctx->jobs[jobID].dstBuff = g_nullBuffer;
+                mtctx->jobs[jobID].cSize = 0;
+                dstPos += cSize ;
+            }
+        }  /* for (jobID=0; jobID<nbJobs; jobID++) */
+
+        DEBUGLOG(4, "checksumFlag : %u ", params.fParams.checksumFlag);
+        if (params.fParams.checksumFlag) {
+            U32 const checksum = (U32)XXH64_digest(&mtctx->serial.xxhState);
+            if (dstPos + 4 > dstCapacity) {
+                error = ERROR(dstSize_tooSmall);
+            } else {
+                DEBUGLOG(4, "writing checksum : %08X \n", checksum);
+                MEM_writeLE32((char*)dst + dstPos, checksum);
+                dstPos += 4;
+        }   }
+
+        if (!error) DEBUGLOG(4, "compressed size : %u  ", (U32)dstPos);
+        return error ? error : dstPos;
+    }
+}
+
+size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const ZSTD_CDict* cdict,
+                               ZSTD_parameters params,
+                               unsigned overlapLog)
+{
+    ZSTD_CCtx_params cctxParams = mtctx->params;
+    cctxParams.cParams = params.cParams;
+    cctxParams.fParams = params.fParams;
+    cctxParams.overlapSizeLog = overlapLog;
+    return ZSTDMT_compress_advanced_internal(mtctx,
+                                             dst, dstCapacity,
+                                             src, srcSize,
+                                             cdict, cctxParams);
+}
+
+
 size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
                            void* dst, size_t dstCapacity,
                      const void* src, size_t srcSize,
                            int compressionLevel)
 {
+    U32 const overlapLog = (compressionLevel >= ZSTD_maxCLevel()) ? 9 : ZSTDMT_OVERLAPLOG_DEFAULT;
     ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);
-    size_t const chunkTargetSize = (size_t)1 << (params.cParams.windowLog + 2);
-    unsigned const nbChunksMax = (unsigned)(srcSize / chunkTargetSize) + (srcSize < chunkTargetSize) /* min 1 */;
-    unsigned nbChunks = MIN(nbChunksMax, mtctx->nbThreads);
-    size_t const proposedChunkSize = (srcSize + (nbChunks-1)) / nbChunks;
-    size_t const avgChunkSize = ((proposedChunkSize & 0x1FFFF) < 0xFFFF) ? proposedChunkSize + 0xFFFF : proposedChunkSize;   /* avoid too small last block */
-    size_t remainingSrcSize = srcSize;
-    const char* const srcStart = (const char*)src;
-    size_t frameStartPos = 0;
-
-    DEBUGLOG(3, "windowLog : %2u => chunkTargetSize : %u bytes  ", params.cParams.windowLog, (U32)chunkTargetSize);
-    DEBUGLOG(2, "nbChunks  : %2u   (chunkSize : %u bytes)   ", nbChunks, (U32)avgChunkSize);
     params.fParams.contentSizeFlag = 1;
-
-    if (nbChunks==1) {   /* fallback to single-thread mode */
-        ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];
-        return ZSTD_compressCCtx(cctx, dst, dstCapacity, src, srcSize, compressionLevel);
-    }
-
-    {   unsigned u;
-        for (u=0; u<nbChunks; u++) {
-            size_t const chunkSize = MIN(remainingSrcSize, avgChunkSize);
-            size_t const dstBufferCapacity = u ? ZSTD_compressBound(chunkSize) : dstCapacity;
-            buffer_t const dstAsBuffer = { dst, dstCapacity };
-            buffer_t const dstBuffer = u ? ZSTDMT_getBuffer(mtctx->buffPool, dstBufferCapacity) : dstAsBuffer;
-            ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(mtctx->cctxPool);
-
-            if ((cctx==NULL) || (dstBuffer.start==NULL)) {
-                mtctx->jobs[u].cSize = ERROR(memory_allocation);   /* job result */
-                mtctx->jobs[u].jobCompleted = 1;
-                nbChunks = u+1;
-                break;   /* let's wait for previous jobs to complete, but don't start new ones */
-            }
-
-            mtctx->jobs[u].srcStart = srcStart + frameStartPos;
-            mtctx->jobs[u].srcSize = chunkSize;
-            mtctx->jobs[u].fullFrameSize = srcSize;
-            mtctx->jobs[u].params = params;
-            mtctx->jobs[u].dstBuff = dstBuffer;
-            mtctx->jobs[u].cctx = cctx;
-            mtctx->jobs[u].firstChunk = (u==0);
-            mtctx->jobs[u].lastChunk = (u==nbChunks-1);
-            mtctx->jobs[u].jobCompleted = 0;
-            mtctx->jobs[u].jobCompleted_mutex = &mtctx->jobCompleted_mutex;
-            mtctx->jobs[u].jobCompleted_cond = &mtctx->jobCompleted_cond;
-
-            DEBUGLOG(3, "posting job %u   (%u bytes)", u, (U32)chunkSize);
-            DEBUG_PRINTHEX(3, mtctx->jobs[u].srcStart, 12);
-            POOL_add(mtctx->factory, ZSTDMT_compressChunk, &mtctx->jobs[u]);
-
-            frameStartPos += chunkSize;
-            remainingSrcSize -= chunkSize;
-    }   }
-    /* note : since nbChunks <= nbThreads, all jobs should be running immediately in parallel */
-
-    {   unsigned chunkID;
-        size_t error = 0, dstPos = 0;
-        for (chunkID=0; chunkID<nbChunks; chunkID++) {
-            DEBUGLOG(3, "waiting for chunk %u ", chunkID);
-            PTHREAD_MUTEX_LOCK(&mtctx->jobCompleted_mutex);
-            while (mtctx->jobs[chunkID].jobCompleted==0) {
-                DEBUGLOG(4, "waiting for jobCompleted signal from chunk %u", chunkID);
-                pthread_cond_wait(&mtctx->jobCompleted_cond, &mtctx->jobCompleted_mutex);
-            }
-            pthread_mutex_unlock(&mtctx->jobCompleted_mutex);
-            DEBUGLOG(3, "ready to write chunk %u ", chunkID);
-
-            ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[chunkID].cctx);
-            mtctx->jobs[chunkID].cctx = NULL;
-            mtctx->jobs[chunkID].srcStart = NULL;
-            {   size_t const cSize = mtctx->jobs[chunkID].cSize;
-                if (ZSTD_isError(cSize)) error = cSize;
-                if ((!error) && (dstPos + cSize > dstCapacity)) error = ERROR(dstSize_tooSmall);
-                if (chunkID) {   /* note : chunk 0 is already written directly into dst */
-                    if (!error) memcpy((char*)dst + dstPos, mtctx->jobs[chunkID].dstBuff.start, cSize);
-                    ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[chunkID].dstBuff);
-                    mtctx->jobs[chunkID].dstBuff = g_nullBuffer;
-                }
-                dstPos += cSize ;
-            }
-        }
-        if (!error) DEBUGLOG(3, "compressed size : %u  ", (U32)dstPos);
-        return error ? error : dstPos;
-    }
-
+    return ZSTDMT_compress_advanced(mtctx, dst, dstCapacity, src, srcSize, NULL, params, overlapLog);
 }
 
 
@@ -480,261 +1215,617 @@
 /* =======      Streaming API     ======= */
 /* ====================================== */
 
-static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* zcs) {
-    while (zcs->doneJobID < zcs->nextJobID) {
-        unsigned const jobID = zcs->doneJobID & zcs->jobIDMask;
-        PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);
-        while (zcs->jobs[jobID].jobCompleted==0) {
-            DEBUGLOG(4, "waiting for jobCompleted signal from chunk %u", zcs->doneJobID);   /* we want to block when waiting for data to flush */
-            pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex);
+size_t ZSTDMT_initCStream_internal(
+        ZSTDMT_CCtx* mtctx,
+        const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
+        const ZSTD_CDict* cdict, ZSTD_CCtx_params params,
+        unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTDMT_initCStream_internal (pledgedSrcSize=%u, nbWorkers=%u, cctxPool=%u, disableLiteralCompression=%i)",
+                (U32)pledgedSrcSize, params.nbWorkers, mtctx->cctxPool->totalCCtx, params.disableLiteralCompression);
+    /* params are supposed to be fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+    assert(mtctx->cctxPool->totalCCtx == params.nbWorkers);
+
+    /* init */
+    if (params.jobSize == 0) {
+        params.jobSize = 1U << ZSTDMT_computeTargetJobLog(params);
+    }
+    if (params.jobSize > ZSTDMT_JOBSIZE_MAX) params.jobSize = ZSTDMT_JOBSIZE_MAX;
+
+    mtctx->singleBlockingThread = (pledgedSrcSize <= ZSTDMT_JOBSIZE_MIN);  /* do not trigger multi-threading when srcSize is too small */
+    if (mtctx->singleBlockingThread) {
+        ZSTD_CCtx_params const singleThreadParams = ZSTDMT_initJobCCtxParams(params);
+        DEBUGLOG(5, "ZSTDMT_initCStream_internal: switch to single blocking thread mode");
+        assert(singleThreadParams.nbWorkers == 0);
+        return ZSTD_initCStream_internal(mtctx->cctxPool->cctx[0],
+                                         dict, dictSize, cdict,
+                                         singleThreadParams, pledgedSrcSize);
+    }
+
+    DEBUGLOG(4, "ZSTDMT_initCStream_internal: %u workers", params.nbWorkers);
+
+    if (mtctx->allJobsCompleted == 0) {   /* previous compression not correctly finished */
+        ZSTDMT_waitForAllJobsCompleted(mtctx);
+        ZSTDMT_releaseAllJobResources(mtctx);
+        mtctx->allJobsCompleted = 1;
+    }
+
+    mtctx->params = params;
+    mtctx->frameContentSize = pledgedSrcSize;
+    if (dict) {
+        ZSTD_freeCDict(mtctx->cdictLocal);
+        mtctx->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
+                                                    ZSTD_dlm_byCopy, dictContentType, /* note : a loadPrefix becomes an internal CDict */
+                                                    params.cParams, mtctx->cMem);
+        mtctx->cdict = mtctx->cdictLocal;
+        if (mtctx->cdictLocal == NULL) return ERROR(memory_allocation);
+    } else {
+        ZSTD_freeCDict(mtctx->cdictLocal);
+        mtctx->cdictLocal = NULL;
+        mtctx->cdict = cdict;
+    }
+
+    mtctx->targetPrefixSize = (size_t)1 << ZSTDMT_computeOverlapLog(params);
+    DEBUGLOG(4, "overlapLog=%u => %u KB", params.overlapSizeLog, (U32)(mtctx->targetPrefixSize>>10));
+    mtctx->targetSectionSize = params.jobSize;
+    if (mtctx->targetSectionSize < ZSTDMT_JOBSIZE_MIN) mtctx->targetSectionSize = ZSTDMT_JOBSIZE_MIN;
+    if (mtctx->targetSectionSize < mtctx->targetPrefixSize) mtctx->targetSectionSize = mtctx->targetPrefixSize;  /* job size must be >= overlap size */
+    DEBUGLOG(4, "Job Size : %u KB (note : set to %u)", (U32)(mtctx->targetSectionSize>>10), params.jobSize);
+    DEBUGLOG(4, "inBuff Size : %u KB", (U32)(mtctx->targetSectionSize>>10));
+    ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(mtctx->targetSectionSize));
+    {
+        /* If ldm is enabled we need windowSize space. */
+        size_t const windowSize = mtctx->params.ldmParams.enableLdm ? (1U << mtctx->params.cParams.windowLog) : 0;
+        /* Two buffers of slack, plus extra space for the overlap
+         * This is the minimum slack that LDM works with. One extra because
+         * flush might waste up to targetSectionSize-1 bytes. Another extra
+         * for the overlap (if > 0), then one to fill which doesn't overlap
+         * with the LDM window.
+         */
+        size_t const nbSlackBuffers = 2 + (mtctx->targetPrefixSize > 0);
+        size_t const slackSize = mtctx->targetSectionSize * nbSlackBuffers;
+        /* Compute the total size, and always have enough slack */
+        size_t const nbWorkers = MAX(mtctx->params.nbWorkers, 1);
+        size_t const sectionsSize = mtctx->targetSectionSize * nbWorkers;
+        size_t const capacity = MAX(windowSize, sectionsSize) + slackSize;
+        if (mtctx->roundBuff.capacity < capacity) {
+            if (mtctx->roundBuff.buffer)
+                ZSTD_free(mtctx->roundBuff.buffer, mtctx->cMem);
+            mtctx->roundBuff.buffer = (BYTE*)ZSTD_malloc(capacity, mtctx->cMem);
+            if (mtctx->roundBuff.buffer == NULL) {
+                mtctx->roundBuff.capacity = 0;
+                return ERROR(memory_allocation);
+            }
+            mtctx->roundBuff.capacity = capacity;
+        }
+    }
+    DEBUGLOG(4, "roundBuff capacity : %u KB", (U32)(mtctx->roundBuff.capacity>>10));
+    mtctx->roundBuff.pos = 0;
+    mtctx->inBuff.buffer = g_nullBuffer;
+    mtctx->inBuff.filled = 0;
+    mtctx->inBuff.prefix = kNullRange;
+    mtctx->doneJobID = 0;
+    mtctx->nextJobID = 0;
+    mtctx->frameEnded = 0;
+    mtctx->allJobsCompleted = 0;
+    mtctx->consumed = 0;
+    mtctx->produced = 0;
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params))
+        return ERROR(memory_allocation);
+    return 0;
+}
+
+size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
+                             const void* dict, size_t dictSize,
+                                   ZSTD_parameters params,
+                                   unsigned long long pledgedSrcSize)
+{
+    ZSTD_CCtx_params cctxParams = mtctx->params;  /* retrieve sticky params */
+    DEBUGLOG(4, "ZSTDMT_initCStream_advanced (pledgedSrcSize=%u)", (U32)pledgedSrcSize);
+    cctxParams.cParams = params.cParams;
+    cctxParams.fParams = params.fParams;
+    return ZSTDMT_initCStream_internal(mtctx, dict, dictSize, ZSTD_dct_auto, NULL,
+                                       cctxParams, pledgedSrcSize);
+}
+
+size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
+                               const ZSTD_CDict* cdict,
+                                     ZSTD_frameParameters fParams,
+                                     unsigned long long pledgedSrcSize)
+{
+    ZSTD_CCtx_params cctxParams = mtctx->params;
+    if (cdict==NULL) return ERROR(dictionary_wrong);   /* method incompatible with NULL cdict */
+    cctxParams.cParams = ZSTD_getCParamsFromCDict(cdict);
+    cctxParams.fParams = fParams;
+    return ZSTDMT_initCStream_internal(mtctx, NULL, 0 /*dictSize*/, ZSTD_dct_auto, cdict,
+                                       cctxParams, pledgedSrcSize);
+}
+
+
+/* ZSTDMT_resetCStream() :
+ * pledgedSrcSize can be zero == unknown (for the time being)
+ * prefer using ZSTD_CONTENTSIZE_UNKNOWN,
+ * as `0` might mean "empty" in the future */
+size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize)
+{
+    if (!pledgedSrcSize) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTDMT_initCStream_internal(mtctx, NULL, 0, ZSTD_dct_auto, 0, mtctx->params,
+                                       pledgedSrcSize);
+}
+
+size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel) {
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0);
+    ZSTD_CCtx_params cctxParams = mtctx->params;   /* retrieve sticky params */
+    DEBUGLOG(4, "ZSTDMT_initCStream (cLevel=%i)", compressionLevel);
+    cctxParams.cParams = params.cParams;
+    cctxParams.fParams = params.fParams;
+    return ZSTDMT_initCStream_internal(mtctx, NULL, 0, ZSTD_dct_auto, NULL, cctxParams, ZSTD_CONTENTSIZE_UNKNOWN);
+}
+
+
+/* ZSTDMT_writeLastEmptyBlock()
+ * Write a single empty block with an end-of-frame to finish a frame.
+ * Job must be created from streaming variant.
+ * This function is always successfull if expected conditions are fulfilled.
+ */
+static void ZSTDMT_writeLastEmptyBlock(ZSTDMT_jobDescription* job)
+{
+    assert(job->lastJob == 1);
+    assert(job->src.size == 0);   /* last job is empty -> will be simplified into a last empty block */
+    assert(job->firstJob == 0);   /* cannot be first job, as it also needs to create frame header */
+    assert(job->dstBuff.start == NULL);   /* invoked from streaming variant only (otherwise, dstBuff might be user's output) */
+    job->dstBuff = ZSTDMT_getBuffer(job->bufPool);
+    if (job->dstBuff.start == NULL) {
+      job->cSize = ERROR(memory_allocation);
+      return;
+    }
+    assert(job->dstBuff.capacity >= ZSTD_blockHeaderSize);   /* no buffer should ever be that small */
+    job->src = kNullRange;
+    job->cSize = ZSTD_writeLastEmptyBlock(job->dstBuff.start, job->dstBuff.capacity);
+    assert(!ZSTD_isError(job->cSize));
+    assert(job->consumed == 0);
+}
+
+static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* mtctx, size_t srcSize, ZSTD_EndDirective endOp)
+{
+    unsigned const jobID = mtctx->nextJobID & mtctx->jobIDMask;
+    int const endFrame = (endOp == ZSTD_e_end);
+
+    if (mtctx->nextJobID > mtctx->doneJobID + mtctx->jobIDMask) {
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: will not create new job : table is full");
+        assert((mtctx->nextJobID & mtctx->jobIDMask) == (mtctx->doneJobID & mtctx->jobIDMask));
+        return 0;
+    }
+
+    if (!mtctx->jobReady) {
+        BYTE const* src = (BYTE const*)mtctx->inBuff.buffer.start;
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: preparing job %u to compress %u bytes with %u preload ",
+                    mtctx->nextJobID, (U32)srcSize, (U32)mtctx->inBuff.prefix.size);
+        mtctx->jobs[jobID].src.start = src;
+        mtctx->jobs[jobID].src.size = srcSize;
+        assert(mtctx->inBuff.filled >= srcSize);
+        mtctx->jobs[jobID].prefix = mtctx->inBuff.prefix;
+        mtctx->jobs[jobID].consumed = 0;
+        mtctx->jobs[jobID].cSize = 0;
+        mtctx->jobs[jobID].params = mtctx->params;
+        mtctx->jobs[jobID].cdict = mtctx->nextJobID==0 ? mtctx->cdict : NULL;
+        mtctx->jobs[jobID].fullFrameSize = mtctx->frameContentSize;
+        mtctx->jobs[jobID].dstBuff = g_nullBuffer;
+        mtctx->jobs[jobID].cctxPool = mtctx->cctxPool;
+        mtctx->jobs[jobID].bufPool = mtctx->bufPool;
+        mtctx->jobs[jobID].seqPool = mtctx->seqPool;
+        mtctx->jobs[jobID].serial = &mtctx->serial;
+        mtctx->jobs[jobID].jobID = mtctx->nextJobID;
+        mtctx->jobs[jobID].firstJob = (mtctx->nextJobID==0);
+        mtctx->jobs[jobID].lastJob = endFrame;
+        mtctx->jobs[jobID].frameChecksumNeeded = endFrame && (mtctx->nextJobID>0) && mtctx->params.fParams.checksumFlag;
+        mtctx->jobs[jobID].dstFlushed = 0;
+
+        /* Update the round buffer pos and clear the input buffer to be reset */
+        mtctx->roundBuff.pos += srcSize;
+        mtctx->inBuff.buffer = g_nullBuffer;
+        mtctx->inBuff.filled = 0;
+        /* Set the prefix */
+        if (!endFrame) {
+            size_t const newPrefixSize = MIN(srcSize, mtctx->targetPrefixSize);
+            mtctx->inBuff.prefix.start = src + srcSize - newPrefixSize;
+            mtctx->inBuff.prefix.size = newPrefixSize;
+        } else {   /* endFrame==1 => no need for another input buffer */
+            mtctx->inBuff.prefix = kNullRange;
+            mtctx->frameEnded = endFrame;
+            if (mtctx->nextJobID == 0) {
+                /* single job exception : checksum is already calculated directly within worker thread */
+                mtctx->params.fParams.checksumFlag = 0;
+        }   }
+
+        if ( (srcSize == 0)
+          && (mtctx->nextJobID>0)/*single job must also write frame header*/ ) {
+            DEBUGLOG(5, "ZSTDMT_createCompressionJob: creating a last empty block to end frame");
+            assert(endOp == ZSTD_e_end);  /* only possible case : need to end the frame with an empty last block */
+            ZSTDMT_writeLastEmptyBlock(mtctx->jobs + jobID);
+            mtctx->nextJobID++;
+            return 0;
         }
-        pthread_mutex_unlock(&zcs->jobCompleted_mutex);
-        zcs->doneJobID++;
+    }
+
+    DEBUGLOG(5, "ZSTDMT_createCompressionJob: posting job %u : %u bytes  (end:%u, jobNb == %u (mod:%u))",
+                mtctx->nextJobID,
+                (U32)mtctx->jobs[jobID].src.size,
+                mtctx->jobs[jobID].lastJob,
+                mtctx->nextJobID,
+                jobID);
+    if (POOL_tryAdd(mtctx->factory, ZSTDMT_compressionJob, &mtctx->jobs[jobID])) {
+        mtctx->nextJobID++;
+        mtctx->jobReady = 0;
+    } else {
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: no worker available for job %u", mtctx->nextJobID);
+        mtctx->jobReady = 1;
+    }
+    return 0;
+}
+
+
+/*! ZSTDMT_flushProduced() :
+ * `output` : `pos` will be updated with amount of data flushed .
+ * `blockToFlush` : if >0, the function will block and wait if there is no data available to flush .
+ * @return : amount of data remaining within internal buffer, 0 if no more, 1 if unknown but > 0, or an error code */
+static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, unsigned blockToFlush, ZSTD_EndDirective end)
+{
+    unsigned const wJobID = mtctx->doneJobID & mtctx->jobIDMask;
+    DEBUGLOG(5, "ZSTDMT_flushProduced (blocking:%u , job %u <= %u)",
+                blockToFlush, mtctx->doneJobID, mtctx->nextJobID);
+    assert(output->size >= output->pos);
+
+    ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[wJobID].job_mutex);
+    if (  blockToFlush
+      && (mtctx->doneJobID < mtctx->nextJobID) ) {
+        assert(mtctx->jobs[wJobID].dstFlushed <= mtctx->jobs[wJobID].cSize);
+        while (mtctx->jobs[wJobID].dstFlushed == mtctx->jobs[wJobID].cSize) {  /* nothing to flush */
+            if (mtctx->jobs[wJobID].consumed == mtctx->jobs[wJobID].src.size) {
+                DEBUGLOG(5, "job %u is completely consumed (%u == %u) => don't wait for cond, there will be none",
+                            mtctx->doneJobID, (U32)mtctx->jobs[wJobID].consumed, (U32)mtctx->jobs[wJobID].src.size);
+                break;
+            }
+            DEBUGLOG(5, "waiting for something to flush from job %u (currently flushed: %u bytes)",
+                        mtctx->doneJobID, (U32)mtctx->jobs[wJobID].dstFlushed);
+            ZSTD_pthread_cond_wait(&mtctx->jobs[wJobID].job_cond, &mtctx->jobs[wJobID].job_mutex);  /* block when nothing to flush but some to come */
+    }   }
+
+    /* try to flush something */
+    {   size_t cSize = mtctx->jobs[wJobID].cSize;                  /* shared */
+        size_t const srcConsumed = mtctx->jobs[wJobID].consumed;   /* shared */
+        size_t const srcSize = mtctx->jobs[wJobID].src.size;        /* read-only, could be done after mutex lock, but no-declaration-after-statement */
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+        if (ZSTD_isError(cSize)) {
+            DEBUGLOG(5, "ZSTDMT_flushProduced: job %u : compression error detected : %s",
+                        mtctx->doneJobID, ZSTD_getErrorName(cSize));
+            ZSTDMT_waitForAllJobsCompleted(mtctx);
+            ZSTDMT_releaseAllJobResources(mtctx);
+            return cSize;
+        }
+        /* add frame checksum if necessary (can only happen once) */
+        assert(srcConsumed <= srcSize);
+        if ( (srcConsumed == srcSize)   /* job completed -> worker no longer active */
+          && mtctx->jobs[wJobID].frameChecksumNeeded ) {
+            U32 const checksum = (U32)XXH64_digest(&mtctx->serial.xxhState);
+            DEBUGLOG(4, "ZSTDMT_flushProduced: writing checksum : %08X \n", checksum);
+            MEM_writeLE32((char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].cSize, checksum);
+            cSize += 4;
+            mtctx->jobs[wJobID].cSize += 4;  /* can write this shared value, as worker is no longer active */
+            mtctx->jobs[wJobID].frameChecksumNeeded = 0;
+        }
+        if (cSize > 0) {   /* compression is ongoing or completed */
+            size_t const toFlush = MIN(cSize - mtctx->jobs[wJobID].dstFlushed, output->size - output->pos);
+            DEBUGLOG(5, "ZSTDMT_flushProduced: Flushing %u bytes from job %u (completion:%u/%u, generated:%u)",
+                        (U32)toFlush, mtctx->doneJobID, (U32)srcConsumed, (U32)srcSize, (U32)cSize);
+            assert(mtctx->doneJobID < mtctx->nextJobID);
+            assert(cSize >= mtctx->jobs[wJobID].dstFlushed);
+            assert(mtctx->jobs[wJobID].dstBuff.start != NULL);
+            memcpy((char*)output->dst + output->pos,
+                   (const char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].dstFlushed,
+                   toFlush);
+            output->pos += toFlush;
+            mtctx->jobs[wJobID].dstFlushed += toFlush;  /* can write : this value is only used by mtctx */
+
+            if ( (srcConsumed == srcSize)    /* job completed */
+              && (mtctx->jobs[wJobID].dstFlushed == cSize) ) {   /* output buffer fully flushed => free this job position */
+                DEBUGLOG(5, "Job %u completed (%u bytes), moving to next one",
+                        mtctx->doneJobID, (U32)mtctx->jobs[wJobID].dstFlushed);
+                ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[wJobID].dstBuff);
+                mtctx->jobs[wJobID].dstBuff = g_nullBuffer;
+                mtctx->jobs[wJobID].cSize = 0;   /* ensure this job slot is considered "not started" in future check */
+                mtctx->consumed += srcSize;
+                mtctx->produced += cSize;
+                mtctx->doneJobID++;
+        }   }
+
+        /* return value : how many bytes left in buffer ; fake it to 1 when unknown but >0 */
+        if (cSize > mtctx->jobs[wJobID].dstFlushed) return (cSize - mtctx->jobs[wJobID].dstFlushed);
+        if (srcSize > srcConsumed) return 1;   /* current job not completely compressed */
+    }
+    if (mtctx->doneJobID < mtctx->nextJobID) return 1;   /* some more jobs ongoing */
+    if (mtctx->jobReady) return 1;      /* one job is ready to push, just not yet in the list */
+    if (mtctx->inBuff.filled > 0) return 1;   /* input is not empty, and still needs to be converted into a job */
+    mtctx->allJobsCompleted = mtctx->frameEnded;   /* all jobs are entirely flushed => if this one is last one, frame is completed */
+    if (end == ZSTD_e_end) return !mtctx->frameEnded;  /* for ZSTD_e_end, question becomes : is frame completed ? instead of : are internal buffers fully flushed ? */
+    return 0;   /* internal buffers fully flushed */
+}
+
+/**
+ * Returns the range of data used by the earliest job that is not yet complete.
+ * If the data of the first job is broken up into two segments, we cover both
+ * sections.
+ */
+static range_t ZSTDMT_getInputDataInUse(ZSTDMT_CCtx* mtctx)
+{
+    unsigned const firstJobID = mtctx->doneJobID;
+    unsigned const lastJobID = mtctx->nextJobID;
+    unsigned jobID;
+
+    for (jobID = firstJobID; jobID < lastJobID; ++jobID) {
+        unsigned const wJobID = jobID & mtctx->jobIDMask;
+        size_t consumed;
+
+        ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[wJobID].job_mutex);
+        consumed = mtctx->jobs[wJobID].consumed;
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+
+        if (consumed < mtctx->jobs[wJobID].src.size) {
+            range_t range = mtctx->jobs[wJobID].prefix;
+            if (range.size == 0) {
+                /* Empty prefix */
+                range = mtctx->jobs[wJobID].src;
+            }
+            /* Job source in multiple segments not supported yet */
+            assert(range.start <= mtctx->jobs[wJobID].src.start);
+            return range;
+        }
+    }
+    return kNullRange;
+}
+
+/**
+ * Returns non-zero iff buffer and range overlap.
+ */
+static int ZSTDMT_isOverlapped(buffer_t buffer, range_t range)
+{
+    BYTE const* const bufferStart = (BYTE const*)buffer.start;
+    BYTE const* const bufferEnd = bufferStart + buffer.capacity;
+    BYTE const* const rangeStart = (BYTE const*)range.start;
+    BYTE const* const rangeEnd = rangeStart + range.size;
+
+    if (rangeStart == NULL || bufferStart == NULL)
+        return 0;
+    /* Empty ranges cannot overlap */
+    if (bufferStart == bufferEnd || rangeStart == rangeEnd)
+        return 0;
+
+    return bufferStart < rangeEnd && rangeStart < bufferEnd;
+}
+
+static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
+{
+    range_t extDict;
+    range_t prefix;
+
+    extDict.start = window.dictBase + window.lowLimit;
+    extDict.size = window.dictLimit - window.lowLimit;
+
+    prefix.start = window.base + window.dictLimit;
+    prefix.size = window.nextSrc - (window.base + window.dictLimit);
+    DEBUGLOG(5, "extDict [0x%zx, 0x%zx)",
+                (size_t)extDict.start,
+                (size_t)extDict.start + extDict.size);
+    DEBUGLOG(5, "prefix  [0x%zx, 0x%zx)",
+                (size_t)prefix.start,
+                (size_t)prefix.start + prefix.size);
+
+    return ZSTDMT_isOverlapped(buffer, extDict)
+        || ZSTDMT_isOverlapped(buffer, prefix);
+}
+
+static void ZSTDMT_waitForLdmComplete(ZSTDMT_CCtx* mtctx, buffer_t buffer)
+{
+    if (mtctx->params.ldmParams.enableLdm) {
+        ZSTD_pthread_mutex_t* mutex = &mtctx->serial.ldmWindowMutex;
+        DEBUGLOG(5, "source  [0x%zx, 0x%zx)",
+                    (size_t)buffer.start,
+                    (size_t)buffer.start + buffer.capacity);
+        ZSTD_PTHREAD_MUTEX_LOCK(mutex);
+        while (ZSTDMT_doesOverlapWindow(buffer, mtctx->serial.ldmWindow)) {
+            DEBUGLOG(6, "Waiting for LDM to finish...");
+            ZSTD_pthread_cond_wait(&mtctx->serial.ldmWindowCond, mutex);
+        }
+        DEBUGLOG(6, "Done waiting for LDM to finish");
+        ZSTD_pthread_mutex_unlock(mutex);
+    }
+}
+
+/**
+ * Attempts to set the inBuff to the next section to fill.
+ * If any part of the new section is still in use we give up.
+ * Returns non-zero if the buffer is filled.
+ */
+static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
+{
+    range_t const inUse = ZSTDMT_getInputDataInUse(mtctx);
+    size_t const spaceLeft = mtctx->roundBuff.capacity - mtctx->roundBuff.pos;
+    size_t const target = mtctx->targetSectionSize;
+    buffer_t buffer;
+
+    assert(mtctx->inBuff.buffer.start == NULL);
+    assert(mtctx->roundBuff.capacity >= target);
+
+    if (spaceLeft < target) {
+        /* ZSTD_invalidateRepCodes() doesn't work for extDict variants.
+         * Simply copy the prefix to the beginning in that case.
+         */
+        BYTE* const start = (BYTE*)mtctx->roundBuff.buffer;
+        size_t const prefixSize = mtctx->inBuff.prefix.size;
+
+        buffer.start = start;
+        buffer.capacity = prefixSize;
+        if (ZSTDMT_isOverlapped(buffer, inUse)) {
+            DEBUGLOG(6, "Waiting for buffer...");
+            return 0;
+        }
+        ZSTDMT_waitForLdmComplete(mtctx, buffer);
+        memmove(start, mtctx->inBuff.prefix.start, prefixSize);
+        mtctx->inBuff.prefix.start = start;
+        mtctx->roundBuff.pos = prefixSize;
+    }
+    buffer.start = mtctx->roundBuff.buffer + mtctx->roundBuff.pos;
+    buffer.capacity = target;
+
+    if (ZSTDMT_isOverlapped(buffer, inUse)) {
+        DEBUGLOG(6, "Waiting for buffer...");
+        return 0;
+    }
+    assert(!ZSTDMT_isOverlapped(buffer, mtctx->inBuff.prefix));
+
+    ZSTDMT_waitForLdmComplete(mtctx, buffer);
+
+    DEBUGLOG(5, "Using prefix range [%zx, %zx)",
+                (size_t)mtctx->inBuff.prefix.start,
+                (size_t)mtctx->inBuff.prefix.start + mtctx->inBuff.prefix.size);
+    DEBUGLOG(5, "Using source range [%zx, %zx)",
+                (size_t)buffer.start,
+                (size_t)buffer.start + buffer.capacity);
+
+
+    mtctx->inBuff.buffer = buffer;
+    mtctx->inBuff.filled = 0;
+    assert(mtctx->roundBuff.pos + buffer.capacity <= mtctx->roundBuff.capacity);
+    return 1;
+}
+
+
+/** ZSTDMT_compressStream_generic() :
+ *  internal use only - exposed to be invoked from zstd_compress.c
+ *  assumption : output and input are valid (pos <= size)
+ * @return : minimum amount of data remaining to flush, 0 if none */
+size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
+                                     ZSTD_outBuffer* output,
+                                     ZSTD_inBuffer* input,
+                                     ZSTD_EndDirective endOp)
+{
+    unsigned forwardInputProgress = 0;
+    DEBUGLOG(5, "ZSTDMT_compressStream_generic (endOp=%u, srcSize=%u)",
+                (U32)endOp, (U32)(input->size - input->pos));
+    assert(output->pos <= output->size);
+    assert(input->pos  <= input->size);
+
+    if (mtctx->singleBlockingThread) {  /* delegate to single-thread (synchronous) */
+        return ZSTD_compressStream_generic(mtctx->cctxPool->cctx[0], output, input, endOp);
+    }
+
+    if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) {
+        /* current frame being ended. Only flush/end are allowed */
+        return ERROR(stage_wrong);
+    }
+
+    /* single-pass shortcut (note : synchronous-mode) */
+    if ( (mtctx->nextJobID == 0)      /* just started */
+      && (mtctx->inBuff.filled == 0)  /* nothing buffered */
+      && (!mtctx->jobReady)           /* no job already created */
+      && (endOp == ZSTD_e_end)        /* end order */
+      && (output->size - output->pos >= ZSTD_compressBound(input->size - input->pos)) ) { /* enough space in dst */
+        size_t const cSize = ZSTDMT_compress_advanced_internal(mtctx,
+                (char*)output->dst + output->pos, output->size - output->pos,
+                (const char*)input->src + input->pos, input->size - input->pos,
+                mtctx->cdict, mtctx->params);
+        if (ZSTD_isError(cSize)) return cSize;
+        input->pos = input->size;
+        output->pos += cSize;
+        mtctx->allJobsCompleted = 1;
+        mtctx->frameEnded = 1;
+        return 0;
+    }
+
+    /* fill input buffer */
+    if ( (!mtctx->jobReady)
+      && (input->size > input->pos) ) {   /* support NULL input */
+        if (mtctx->inBuff.buffer.start == NULL) {
+            assert(mtctx->inBuff.filled == 0); /* Can't fill an empty buffer */
+            if (!ZSTDMT_tryGetInputRange(mtctx)) {
+                /* It is only possible for this operation to fail if there are
+                 * still compression jobs ongoing.
+                 */
+                assert(mtctx->doneJobID != mtctx->nextJobID);
+            }
+        }
+        if (mtctx->inBuff.buffer.start != NULL) {
+            size_t const toLoad = MIN(input->size - input->pos, mtctx->targetSectionSize - mtctx->inBuff.filled);
+            assert(mtctx->inBuff.buffer.capacity >= mtctx->targetSectionSize);
+            DEBUGLOG(5, "ZSTDMT_compressStream_generic: adding %u bytes on top of %u to buffer of size %u",
+                        (U32)toLoad, (U32)mtctx->inBuff.filled, (U32)mtctx->targetSectionSize);
+            memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, toLoad);
+            input->pos += toLoad;
+            mtctx->inBuff.filled += toLoad;
+            forwardInputProgress = toLoad>0;
+        }
+        if ((input->pos < input->size) && (endOp == ZSTD_e_end))
+            endOp = ZSTD_e_flush;   /* can't end now : not all input consumed */
+    }
+
+    if ( (mtctx->jobReady)
+      || (mtctx->inBuff.filled >= mtctx->targetSectionSize)  /* filled enough : let's compress */
+      || ((endOp != ZSTD_e_continue) && (mtctx->inBuff.filled > 0))  /* something to flush : let's go */
+      || ((endOp == ZSTD_e_end) && (!mtctx->frameEnded)) ) {   /* must finish the frame with a zero-size block */
+        size_t const jobSize = mtctx->inBuff.filled;
+        assert(mtctx->inBuff.filled <= mtctx->targetSectionSize);
+        CHECK_F( ZSTDMT_createCompressionJob(mtctx, jobSize, endOp) );
+    }
+
+    /* check for potential compressed data ready to be flushed */
+    {   size_t const remainingToFlush = ZSTDMT_flushProduced(mtctx, output, !forwardInputProgress, endOp); /* block if there was no forward input progress */
+        if (input->pos < input->size) return MAX(remainingToFlush, 1);  /* input not consumed : do not end flush yet */
+        return remainingToFlush;
     }
 }
 
 
-static size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
-                                    const void* dict, size_t dictSize, unsigned updateDict,
-                                    ZSTD_parameters params, unsigned long long pledgedSrcSize)
+size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
 {
-    ZSTD_customMem const cmem = { NULL, NULL, NULL };
-    DEBUGLOG(3, "Started new compression, with windowLog : %u", params.cParams.windowLog);
-    if (zcs->nbThreads==1) return ZSTD_initCStream_advanced(zcs->cstream, dict, dictSize, params, pledgedSrcSize);
-    if (zcs->allJobsCompleted == 0) {   /* previous job not correctly finished */
-        ZSTDMT_waitForAllJobsCompleted(zcs);
-        ZSTDMT_releaseAllJobResources(zcs);
-        zcs->allJobsCompleted = 1;
-    }
-    zcs->params = params;
-    if (updateDict) {
-        ZSTD_freeCDict(zcs->cdict); zcs->cdict = NULL;
-        if (dict && dictSize) {
-            zcs->cdict = ZSTD_createCDict_advanced(dict, dictSize, 0, params, cmem);
-            if (zcs->cdict == NULL) return ERROR(memory_allocation);
-    }   }
-    zcs->frameContentSize = pledgedSrcSize;
-    zcs->targetDictSize = (zcs->overlapRLog>=9) ? 0 : (size_t)1 << (zcs->params.cParams.windowLog - zcs->overlapRLog);
-    DEBUGLOG(4, "overlapRLog : %u ", zcs->overlapRLog);
-    DEBUGLOG(3, "overlap Size : %u KB", (U32)(zcs->targetDictSize>>10));
-    zcs->targetSectionSize = zcs->sectionSize ? zcs->sectionSize : (size_t)1 << (zcs->params.cParams.windowLog + 2);
-    zcs->targetSectionSize = MAX(ZSTDMT_SECTION_SIZE_MIN, zcs->targetSectionSize);
-    zcs->targetSectionSize = MAX(zcs->targetDictSize, zcs->targetSectionSize);
-    DEBUGLOG(3, "Section Size : %u KB", (U32)(zcs->targetSectionSize>>10));
-    zcs->marginSize = zcs->targetSectionSize >> 2;
-    zcs->inBuffSize = zcs->targetDictSize + zcs->targetSectionSize + zcs->marginSize;
-    zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize);
-    if (zcs->inBuff.buffer.start == NULL) return ERROR(memory_allocation);
-    zcs->inBuff.filled = 0;
-    zcs->dictSize = 0;
-    zcs->doneJobID = 0;
-    zcs->nextJobID = 0;
-    zcs->frameEnded = 0;
-    zcs->allJobsCompleted = 0;
-    if (params.fParams.checksumFlag) XXH64_reset(&zcs->xxhState, 0);
-    return 0;
-}
+    CHECK_F( ZSTDMT_compressStream_generic(mtctx, output, input, ZSTD_e_continue) );
 
-size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* zcs,
-                                const void* dict, size_t dictSize,
-                                ZSTD_parameters params, unsigned long long pledgedSrcSize)
-{
-    return ZSTDMT_initCStream_internal(zcs, dict, dictSize, 1, params, pledgedSrcSize);
-}
-
-/* ZSTDMT_resetCStream() :
- * pledgedSrcSize is optional and can be zero == unknown */
-size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* zcs, unsigned long long pledgedSrcSize)
-{
-    if (zcs->nbThreads==1) return ZSTD_resetCStream(zcs->cstream, pledgedSrcSize);
-    return ZSTDMT_initCStream_internal(zcs, NULL, 0, 0, zcs->params, pledgedSrcSize);
-}
-
-size_t ZSTDMT_initCStream(ZSTDMT_CCtx* zcs, int compressionLevel) {
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, 0);
-    return ZSTDMT_initCStream_internal(zcs, NULL, 0, 1, params, 0);
+    /* recommended next input size : fill current input buffer */
+    return mtctx->targetSectionSize - mtctx->inBuff.filled;   /* note : could be zero when input buffer is fully filled and no more availability to create new job */
 }
 
 
-static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsigned endFrame)
+static size_t ZSTDMT_flushStream_internal(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_EndDirective endFrame)
 {
-    size_t const dstBufferCapacity = ZSTD_compressBound(srcSize);
-    buffer_t const dstBuffer = ZSTDMT_getBuffer(zcs->buffPool, dstBufferCapacity);
-    ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(zcs->cctxPool);
-    unsigned const jobID = zcs->nextJobID & zcs->jobIDMask;
+    size_t const srcSize = mtctx->inBuff.filled;
+    DEBUGLOG(5, "ZSTDMT_flushStream_internal");
 
-    if ((cctx==NULL) || (dstBuffer.start==NULL)) {
-        zcs->jobs[jobID].jobCompleted = 1;
-        zcs->nextJobID++;
-        ZSTDMT_waitForAllJobsCompleted(zcs);
-        ZSTDMT_releaseAllJobResources(zcs);
-        return ERROR(memory_allocation);
+    if ( mtctx->jobReady     /* one job ready for a worker to pick up */
+      || (srcSize > 0)       /* still some data within input buffer */
+      || ((endFrame==ZSTD_e_end) && !mtctx->frameEnded)) {  /* need a last 0-size block to end frame */
+           DEBUGLOG(5, "ZSTDMT_flushStream_internal : create a new job (%u bytes, end:%u)",
+                        (U32)srcSize, (U32)endFrame);
+        CHECK_F( ZSTDMT_createCompressionJob(mtctx, srcSize, endFrame) );
     }
 
-    DEBUGLOG(4, "preparing job %u to compress %u bytes with %u preload ", zcs->nextJobID, (U32)srcSize, (U32)zcs->dictSize);
-    zcs->jobs[jobID].src = zcs->inBuff.buffer;
-    zcs->jobs[jobID].srcStart = zcs->inBuff.buffer.start;
-    zcs->jobs[jobID].srcSize = srcSize;
-    zcs->jobs[jobID].dictSize = zcs->dictSize;   /* note : zcs->inBuff.filled is presumed >= srcSize + dictSize */
-    zcs->jobs[jobID].params = zcs->params;
-    if (zcs->nextJobID) zcs->jobs[jobID].params.fParams.checksumFlag = 0;  /* do not calculate checksum within sections, just keep it in header for first section */
-    zcs->jobs[jobID].cdict = zcs->nextJobID==0 ? zcs->cdict : NULL;
-    zcs->jobs[jobID].fullFrameSize = zcs->frameContentSize;
-    zcs->jobs[jobID].dstBuff = dstBuffer;
-    zcs->jobs[jobID].cctx = cctx;
-    zcs->jobs[jobID].firstChunk = (zcs->nextJobID==0);
-    zcs->jobs[jobID].lastChunk = endFrame;
-    zcs->jobs[jobID].jobCompleted = 0;
-    zcs->jobs[jobID].dstFlushed = 0;
-    zcs->jobs[jobID].jobCompleted_mutex = &zcs->jobCompleted_mutex;
-    zcs->jobs[jobID].jobCompleted_cond = &zcs->jobCompleted_cond;
-
-    /* get a new buffer for next input */
-    if (!endFrame) {
-        size_t const newDictSize = MIN(srcSize + zcs->dictSize, zcs->targetDictSize);
-        zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize);
-        if (zcs->inBuff.buffer.start == NULL) {   /* not enough memory to allocate next input buffer */
-            zcs->jobs[jobID].jobCompleted = 1;
-            zcs->nextJobID++;
-            ZSTDMT_waitForAllJobsCompleted(zcs);
-            ZSTDMT_releaseAllJobResources(zcs);
-            return ERROR(memory_allocation);
-        }
-        DEBUGLOG(5, "inBuff filled to %u", (U32)zcs->inBuff.filled);
-        zcs->inBuff.filled -= srcSize + zcs->dictSize - newDictSize;
-        DEBUGLOG(5, "new job : filled to %u, with %u dict and %u src", (U32)zcs->inBuff.filled, (U32)newDictSize, (U32)(zcs->inBuff.filled - newDictSize));
-        memmove(zcs->inBuff.buffer.start, (const char*)zcs->jobs[jobID].srcStart + zcs->dictSize + srcSize - newDictSize, zcs->inBuff.filled);
-        DEBUGLOG(5, "new inBuff pre-filled");
-        zcs->dictSize = newDictSize;
-    } else {
-        zcs->inBuff.buffer = g_nullBuffer;
-        zcs->inBuff.filled = 0;
-        zcs->dictSize = 0;
-        zcs->frameEnded = 1;
-        if (zcs->nextJobID == 0)
-            zcs->params.fParams.checksumFlag = 0;   /* single chunk : checksum is calculated directly within worker thread */
-    }
-
-    DEBUGLOG(3, "posting job %u : %u bytes  (end:%u) (note : doneJob = %u=>%u)", zcs->nextJobID, (U32)zcs->jobs[jobID].srcSize, zcs->jobs[jobID].lastChunk, zcs->doneJobID, zcs->doneJobID & zcs->jobIDMask);
-    POOL_add(zcs->factory, ZSTDMT_compressChunk, &zcs->jobs[jobID]);   /* this call is blocking when thread worker pool is exhausted */
-    zcs->nextJobID++;
-    return 0;
+    /* check if there is any data available to flush */
+    return ZSTDMT_flushProduced(mtctx, output, 1 /* blockToFlush */, endFrame);
 }
 
 
-/* ZSTDMT_flushNextJob() :
- * output : will be updated with amount of data flushed .
- * blockToFlush : if >0, the function will block and wait if there is no data available to flush .
- * @return : amount of data remaining within internal buffer, 1 if unknown but > 0, 0 if no more, or an error code */
-static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned blockToFlush)
+size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
 {
-    unsigned const wJobID = zcs->doneJobID & zcs->jobIDMask;
-    if (zcs->doneJobID == zcs->nextJobID) return 0;   /* all flushed ! */
-    PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);
-    while (zcs->jobs[wJobID].jobCompleted==0) {
-        DEBUGLOG(5, "waiting for jobCompleted signal from job %u", zcs->doneJobID);
-        if (!blockToFlush) { pthread_mutex_unlock(&zcs->jobCompleted_mutex); return 0; }  /* nothing ready to be flushed => skip */
-        pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex);  /* block when nothing available to flush */
-    }
-    pthread_mutex_unlock(&zcs->jobCompleted_mutex);
-    /* compression job completed : output can be flushed */
-    {   ZSTDMT_jobDescription job = zcs->jobs[wJobID];
-        if (!job.jobScanned) {
-            if (ZSTD_isError(job.cSize)) {
-                DEBUGLOG(5, "compression error detected ");
-                ZSTDMT_waitForAllJobsCompleted(zcs);
-                ZSTDMT_releaseAllJobResources(zcs);
-                return job.cSize;
-            }
-            ZSTDMT_releaseCCtx(zcs->cctxPool, job.cctx);
-            zcs->jobs[wJobID].cctx = NULL;
-            DEBUGLOG(5, "zcs->params.fParams.checksumFlag : %u ", zcs->params.fParams.checksumFlag);
-            if (zcs->params.fParams.checksumFlag) {
-                XXH64_update(&zcs->xxhState, (const char*)job.srcStart + job.dictSize, job.srcSize);
-                if (zcs->frameEnded && (zcs->doneJobID+1 == zcs->nextJobID)) {  /* write checksum at end of last section */
-                    U32 const checksum = (U32)XXH64_digest(&zcs->xxhState);
-                    DEBUGLOG(4, "writing checksum : %08X \n", checksum);
-                    MEM_writeLE32((char*)job.dstBuff.start + job.cSize, checksum);
-                    job.cSize += 4;
-                    zcs->jobs[wJobID].cSize += 4;
-            }   }
-            ZSTDMT_releaseBuffer(zcs->buffPool, job.src);
-            zcs->jobs[wJobID].srcStart = NULL;
-            zcs->jobs[wJobID].src = g_nullBuffer;
-            zcs->jobs[wJobID].jobScanned = 1;
-        }
-        {   size_t const toWrite = MIN(job.cSize - job.dstFlushed, output->size - output->pos);
-            DEBUGLOG(4, "Flushing %u bytes from job %u ", (U32)toWrite, zcs->doneJobID);
-            memcpy((char*)output->dst + output->pos, (const char*)job.dstBuff.start + job.dstFlushed, toWrite);
-            output->pos += toWrite;
-            job.dstFlushed += toWrite;
-        }
-        if (job.dstFlushed == job.cSize) {   /* output buffer fully flushed => move to next one */
-            ZSTDMT_releaseBuffer(zcs->buffPool, job.dstBuff);
-            zcs->jobs[wJobID].dstBuff = g_nullBuffer;
-            zcs->jobs[wJobID].jobCompleted = 0;
-            zcs->doneJobID++;
-        } else {
-            zcs->jobs[wJobID].dstFlushed = job.dstFlushed;
-        }
-        /* return value : how many bytes left in buffer ; fake it to 1 if unknown but >0 */
-        if (job.cSize > job.dstFlushed) return (job.cSize - job.dstFlushed);
-        if (zcs->doneJobID < zcs->nextJobID) return 1;   /* still some buffer to flush */
-        zcs->allJobsCompleted = zcs->frameEnded;   /* frame completed and entirely flushed */
-        return 0;   /* everything flushed */
-}   }
-
-
-size_t ZSTDMT_compressStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
-{
-    size_t const newJobThreshold = zcs->dictSize + zcs->targetSectionSize + zcs->marginSize;
-    if (zcs->frameEnded) return ERROR(stage_wrong);   /* current frame being ended. Only flush is allowed. Restart with init */
-    if (zcs->nbThreads==1) return ZSTD_compressStream(zcs->cstream, output, input);
-
-    /* fill input buffer */
-    {   size_t const toLoad = MIN(input->size - input->pos, zcs->inBuffSize - zcs->inBuff.filled);
-        memcpy((char*)zcs->inBuff.buffer.start + zcs->inBuff.filled, input->src, toLoad);
-        input->pos += toLoad;
-        zcs->inBuff.filled += toLoad;
-    }
-
-    if ( (zcs->inBuff.filled >= newJobThreshold)  /* filled enough : let's compress */
-        && (zcs->nextJobID <= zcs->doneJobID + zcs->jobIDMask) ) {   /* avoid overwriting job round buffer */
-        CHECK_F( ZSTDMT_createCompressionJob(zcs, zcs->targetSectionSize, 0) );
-    }
-
-    /* check for data to flush */
-    CHECK_F( ZSTDMT_flushNextJob(zcs, output, (zcs->inBuff.filled == zcs->inBuffSize)) ); /* block if it wasn't possible to create new job due to saturation */
-
-    /* recommended next input size : fill current input buffer */
-    return zcs->inBuffSize - zcs->inBuff.filled;   /* note : could be zero when input buffer is fully filled and no more availability to create new job */
+    DEBUGLOG(5, "ZSTDMT_flushStream");
+    if (mtctx->singleBlockingThread)
+        return ZSTD_flushStream(mtctx->cctxPool->cctx[0], output);
+    return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_flush);
 }
 
-
-static size_t ZSTDMT_flushStream_internal(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned endFrame)
+size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
 {
-    size_t const srcSize = zcs->inBuff.filled - zcs->dictSize;
-
-    if (srcSize) DEBUGLOG(4, "flushing : %u bytes left to compress", (U32)srcSize);
-    if ( ((srcSize > 0) || (endFrame && !zcs->frameEnded))
-       && (zcs->nextJobID <= zcs->doneJobID + zcs->jobIDMask) ) {
-        CHECK_F( ZSTDMT_createCompressionJob(zcs, srcSize, endFrame) );
-    }
-
-    /* check if there is any data available to flush */
-    DEBUGLOG(5, "zcs->doneJobID : %u  ; zcs->nextJobID : %u ", zcs->doneJobID, zcs->nextJobID);
-    return ZSTDMT_flushNextJob(zcs, output, 1);
+    DEBUGLOG(4, "ZSTDMT_endStream");
+    if (mtctx->singleBlockingThread)
+        return ZSTD_endStream(mtctx->cctxPool->cctx[0], output);
+    return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_end);
 }
-
-
-size_t ZSTDMT_flushStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)
-{
-    if (zcs->nbThreads==1) return ZSTD_flushStream(zcs->cstream, output);
-    return ZSTDMT_flushStream_internal(zcs, output, 0);
-}
-
-size_t ZSTDMT_endStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)
-{
-    if (zcs->nbThreads==1) return ZSTD_endStream(zcs->cstream, output);
-    return ZSTDMT_flushStream_internal(zcs, output, 1);
-}
--- a/contrib/python-zstandard/zstd/compress/zstdmt_compress.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/compress/zstdmt_compress.h	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,11 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
  #ifndef ZSTDMT_COMPRESS_H
@@ -15,31 +16,41 @@
  #endif
 
 
-/* Note : All prototypes defined in this file shall be considered experimental.
- *        There is no guarantee of API continuity (yet) on any of these prototypes */
+/* Note : This is an internal API.
+ *        Some methods are still exposed (ZSTDLIB_API),
+ *        because it used to be the only way to invoke MT compression.
+ *        Now, it's recommended to use ZSTD_compress_generic() instead.
+ *        These methods will stop being exposed in a future version */
 
 /* ===   Dependencies   === */
-#include <stddef.h>   /* size_t */
+#include <stddef.h>                /* size_t */
 #define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_parameters */
-#include "zstd.h"     /* ZSTD_inBuffer, ZSTD_outBuffer, ZSTDLIB_API */
+#include "zstd.h"            /* ZSTD_inBuffer, ZSTD_outBuffer, ZSTDLIB_API */
 
 
-/* ===   Simple one-pass functions   === */
-
+/* ===   Memory management   === */
 typedef struct ZSTDMT_CCtx_s ZSTDMT_CCtx;
-ZSTDLIB_API ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbThreads);
-ZSTDLIB_API size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* cctx);
+ZSTDLIB_API ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbWorkers);
+ZSTDLIB_API ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers,
+                                                    ZSTD_customMem cMem);
+ZSTDLIB_API size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx);
+
+ZSTDLIB_API size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx);
+
 
-ZSTDLIB_API size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* cctx,
-                           void* dst, size_t dstCapacity,
-                     const void* src, size_t srcSize,
-                           int compressionLevel);
+/* ===   Simple one-pass compression function   === */
+
+ZSTDLIB_API size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                       int compressionLevel);
+
 
 
 /* ===   Streaming functions   === */
 
 ZSTDLIB_API size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel);
-ZSTDLIB_API size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize);    /**< pledgedSrcSize is optional and can be zero == unknown */
+ZSTDLIB_API size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize);  /**< if srcSize is not known at reset time, use ZSTD_CONTENTSIZE_UNKNOWN. Note: for compatibility with older programs, 0 means the same as ZSTD_CONTENTSIZE_UNKNOWN, but it will change in the future to mean "empty" */
 
 ZSTDLIB_API size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
 
@@ -49,26 +60,93 @@
 
 /* ===   Advanced functions and parameters  === */
 
-#ifndef ZSTDMT_SECTION_SIZE_MIN
-#  define ZSTDMT_SECTION_SIZE_MIN (1U << 20)   /* 1 MB - Minimum size of each compression job */
+#ifndef ZSTDMT_JOBSIZE_MIN
+#  define ZSTDMT_JOBSIZE_MIN (1U << 20)   /* 1 MB - Minimum size of each compression job */
 #endif
 
-ZSTDLIB_API size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx, const void* dict, size_t dictSize,  /**< dict can be released after init, a local copy is preserved within zcs */
-                                          ZSTD_parameters params, unsigned long long pledgedSrcSize);  /**< pledgedSrcSize is optional and can be zero == unknown */
+ZSTDLIB_API size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     const ZSTD_CDict* cdict,
+                                           ZSTD_parameters params,
+                                           unsigned overlapLog);
 
-/* ZSDTMT_parameter :
+ZSTDLIB_API size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
+                                        const void* dict, size_t dictSize,   /* dict can be released after init, a local copy is preserved within zcs */
+                                        ZSTD_parameters params,
+                                        unsigned long long pledgedSrcSize);  /* pledgedSrcSize is optional and can be zero == unknown */
+
+ZSTDLIB_API size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
+                                        const ZSTD_CDict* cdict,
+                                        ZSTD_frameParameters fparams,
+                                        unsigned long long pledgedSrcSize);  /* note : zero means empty */
+
+/* ZSTDMT_parameter :
  * List of parameters that can be set using ZSTDMT_setMTCtxParameter() */
 typedef enum {
-    ZSTDMT_p_sectionSize,        /* size of input "section". Each section is compressed in parallel. 0 means default, which is dynamically determined within compression functions */
-    ZSTDMT_p_overlapSectionLog   /* Log of overlapped section; 0 == no overlap, 6(default) == use 1/8th of window, >=9 == use full window */
-} ZSDTMT_parameter;
+    ZSTDMT_p_jobSize,           /* Each job is compressed in parallel. By default, this value is dynamically determined depending on compression parameters. Can be set explicitly here. */
+    ZSTDMT_p_overlapSectionLog  /* Each job may reload a part of previous job to enhance compressionr ratio; 0 == no overlap, 6(default) == use 1/8th of window, >=9 == use full window. This is a "sticky" parameter : its value will be re-used on next compression job */
+} ZSTDMT_parameter;
 
 /* ZSTDMT_setMTCtxParameter() :
  * allow setting individual parameters, one at a time, among a list of enums defined in ZSTDMT_parameter.
- * The function must be called typically after ZSTD_createCCtx().
+ * The function must be called typically after ZSTD_createCCtx() but __before ZSTDMT_init*() !__
  * Parameters not explicitly reset by ZSTDMT_init*() remain the same in consecutive compression sessions.
  * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
-ZSTDLIB_API size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSDTMT_parameter parameter, unsigned value);
+ZSTDLIB_API size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned value);
+
+
+/*! ZSTDMT_compressStream_generic() :
+ *  Combines ZSTDMT_compressStream() with optional ZSTDMT_flushStream() or ZSTDMT_endStream()
+ *  depending on flush directive.
+ * @return : minimum amount of data still to be flushed
+ *           0 if fully flushed
+ *           or an error code
+ *  note : needs to be init using any ZSTD_initCStream*() variant */
+ZSTDLIB_API size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
+                                                ZSTD_outBuffer* output,
+                                                ZSTD_inBuffer* input,
+                                                ZSTD_EndDirective endOp);
+
+
+/* ========================================================
+ * ===  Private interface, for use by ZSTD_compress.c   ===
+ * ===  Not exposed in libzstd. Never invoke directly   ===
+ * ======================================================== */
+
+size_t ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params, ZSTDMT_parameter parameter, unsigned value);
+
+/* ZSTDMT_CCtxParam_setNbWorkers()
+ * Set nbWorkers, and clamp it.
+ * Also reset jobSize and overlapLog */
+size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers);
+
+/*! ZSTDMT_updateCParams_whileCompressing() :
+ *  Updates only a selected set of compression parameters, to remain compatible with current frame.
+ *  New parameters will be applied to next compression job. */
+void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams);
+
+/* ZSTDMT_getNbWorkers():
+ * @return nb threads currently active in mtctx.
+ * mtctx must be valid */
+unsigned ZSTDMT_getNbWorkers(const ZSTDMT_CCtx* mtctx);
+
+/* ZSTDMT_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads.
+ */
+ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx);
+
+
+/*! ZSTDMT_initCStream_internal() :
+ *  Private use only. Init streaming operation.
+ *  expects params to be valid.
+ *  must receive dict, or cdict, or none, but not both.
+ *  @return : 0, or an error code */
+size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
+                    const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
+                    const ZSTD_CDict* cdict,
+                    ZSTD_CCtx_params params, unsigned long long pledgedSrcSize);
 
 
 #if defined (__cplusplus)
--- a/contrib/python-zstandard/zstd/decompress/huf_decompress.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/decompress/huf_decompress.c	Mon Apr 09 10:13:29 2018 -0700
@@ -33,41 +33,35 @@
 ****************************************************************** */
 
 /* **************************************************************
-*  Compiler specifics
-****************************************************************/
-#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
-/* inline is defined */
-#elif defined(_MSC_VER) || defined(__GNUC__)
-#  define inline __inline
-#else
-#  define inline /* disable inline */
-#endif
-
-#ifdef _MSC_VER    /* Visual Studio */
-#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
-#endif
-
-
-/* **************************************************************
 *  Dependencies
 ****************************************************************/
 #include <string.h>     /* memcpy, memset */
 #include "bitstream.h"  /* BIT_* */
+#include "compiler.h"
 #include "fse.h"        /* header compression */
 #define HUF_STATIC_LINKING_ONLY
 #include "huf.h"
+#include "error_private.h"
 
 
 /* **************************************************************
 *  Error Management
 ****************************************************************/
+#define HUF_isError ERR_isError
 #define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+#define CHECK_F(f) { size_t const err_ = (f); if (HUF_isError(err_)) return err_; }
+
+
+/* **************************************************************
+*  Byte alignment for workSpace management
+****************************************************************/
+#define HUF_ALIGN(x, a)         HUF_ALIGN_MASK((x), (a) - 1)
+#define HUF_ALIGN_MASK(x, mask) (((x) + (mask)) & ~(mask))
 
 
 /*-***************************/
 /*  generic DTableDesc       */
 /*-***************************/
-
 typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved; } DTableDesc;
 
 static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
@@ -81,19 +75,27 @@
 /*-***************************/
 /*  single-symbol decoding   */
 /*-***************************/
-
 typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
 
-size_t HUF_readDTableX2 (HUF_DTable* DTable, const void* src, size_t srcSize)
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
 {
-    BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];
-    U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
     U32 tableLog = 0;
     U32 nbSymbols = 0;
     size_t iSize;
     void* const dtPtr = DTable + 1;
     HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
 
+    U32* rankVal;
+    BYTE* huffWeight;
+    size_t spaceUsed32 = 0;
+
+    rankVal = (U32 *)workSpace + spaceUsed32;
+    spaceUsed32 += HUF_TABLELOG_ABSOLUTEMAX + 1;
+    huffWeight = (BYTE *)((U32 *)workSpace + spaceUsed32);
+    spaceUsed32 += HUF_ALIGN(HUF_SYMBOLVALUE_MAX + 1, sizeof(U32)) >> 2;
+
+    if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
+
     HUF_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
     /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
 
@@ -102,16 +104,16 @@
 
     /* Table header */
     {   DTableDesc dtd = HUF_getDTableDesc(DTable);
-        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, huffman tree cannot fit in */
+        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, Huffman tree cannot fit in */
         dtd.tableType = 0;
         dtd.tableLog = (BYTE)tableLog;
         memcpy(DTable, &dtd, sizeof(dtd));
     }
 
-    /* Prepare ranks */
+    /* Calculate starting value for each rank */
     {   U32 n, nextRankStart = 0;
         for (n=1; n<tableLog+1; n++) {
-            U32 current = nextRankStart;
+            U32 const current = nextRankStart;
             nextRankStart += (rankVal[n] << (n-1));
             rankVal[n] = current;
     }   }
@@ -121,19 +123,28 @@
         for (n=0; n<nbSymbols; n++) {
             U32 const w = huffWeight[n];
             U32 const length = (1 << w) >> 1;
-            U32 i;
+            U32 u;
             HUF_DEltX2 D;
             D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
-            for (i = rankVal[w]; i < rankVal[w] + length; i++)
-                dt[i] = D;
+            for (u = rankVal[w]; u < rankVal[w] + length; u++)
+                dt[u] = D;
             rankVal[w] += length;
     }   }
 
     return iSize;
 }
 
+size_t HUF_readDTableX2(HUF_DTable* DTable, const void* src, size_t srcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_readDTableX2_wksp(DTable, src, srcSize,
+                                 workSpace, sizeof(workSpace));
+}
 
-static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+FORCE_INLINE_TEMPLATE BYTE
+HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
 {
     size_t const val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
     BYTE const c = dt[val].byte;
@@ -144,7 +155,7 @@
 #define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
     *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
 
-#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr)  \
     if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
         HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
 
@@ -152,30 +163,33 @@
     if (MEM_64bits()) \
         HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
 
-static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+HINT_INLINE size_t
+HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
 {
     BYTE* const pStart = p;
 
     /* up to 4 symbols at a time */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4)) {
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
         HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
         HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
         HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
         HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
     }
 
-    /* closer to the end */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
-        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    /* [0-3] symbols remaining */
+    if (MEM_32bits())
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd))
+            HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
 
-    /* no more data to retrieve from bitstream, hence no need to reload */
+    /* no more data to retrieve from bitstream, no need to reload */
     while (p < pEnd)
         HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
 
     return pEnd-pStart;
 }
 
-static size_t HUF_decompress1X2_usingDTable_internal(
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X2_usingDTable_internal_body(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
@@ -188,47 +202,17 @@
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
     U32 const dtLog = dtd.tableLog;
 
-    { size_t const errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);
-      if (HUF_isError(errorCode)) return errorCode; }
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
 
     HUF_decodeStreamX2(op, &bitD, oend, dt, dtLog);
 
-    /* check */
     if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
 
     return dstSize;
 }
 
-size_t HUF_decompress1X2_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
-}
-
-size_t HUF_decompress1X2_DCtx (HUF_DTable* DCtx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    const BYTE* ip = (const BYTE*) cSrc;
-
-    size_t const hSize = HUF_readDTableX2 (DCtx, cSrc, cSrcSize);
-    if (HUF_isError(hSize)) return hSize;
-    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
-    ip += hSize; cSrcSize -= hSize;
-
-    return HUF_decompress1X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
-}
-
-size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress1X2_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
-}
-
-
-static size_t HUF_decompress4X2_usingDTable_internal(
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X2_usingDTable_internal_body(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
@@ -263,23 +247,19 @@
         BYTE* op2 = opStart2;
         BYTE* op3 = opStart3;
         BYTE* op4 = opStart4;
-        U32 endSignal;
+        U32 endSignal = BIT_DStream_unfinished;
         DTableDesc const dtd = HUF_getDTableDesc(DTable);
         U32 const dtLog = dtd.tableLog;
 
         if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
-        { size_t const errorCode = BIT_initDStream(&bitD1, istart1, length1);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD2, istart2, length2);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD3, istart3, length3);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD4, istart4, length4);
-          if (HUF_isError(errorCode)) return errorCode; }
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
 
-        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        /* up to 16 symbols per loop (4 symbols per stream) in 64-bit mode */
         endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
-        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; ) {
+        while ( (endSignal==BIT_DStream_unfinished) && (op4<(oend-3)) ) {
             HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
             HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
             HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
@@ -296,10 +276,15 @@
             HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
             HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
             HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
-            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+            BIT_reloadDStream(&bitD1);
+            BIT_reloadDStream(&bitD2);
+            BIT_reloadDStream(&bitD3);
+            BIT_reloadDStream(&bitD4);
         }
 
         /* check corruption */
+        /* note : should not be necessary : op# advance in lock step, and we control op4.
+         *        but curiously, binary generated by gcc 7.2 & 7.3 with -mbmi2 runs faster when >=1 test is present */
         if (op1 > opStart2) return ERROR(corruption_detected);
         if (op2 > opStart3) return ERROR(corruption_detected);
         if (op3 > opStart4) return ERROR(corruption_detected);
@@ -312,8 +297,8 @@
         HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
 
         /* check */
-        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
-        if (!endSignal) return ERROR(corruption_detected);
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
 
         /* decoded size */
         return dstSize;
@@ -321,6 +306,279 @@
 }
 
 
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);
+    }   }
+    return 1;
+}
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+HINT_INLINE size_t
+HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
+                const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to end : up to 2 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X4_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    BIT_DStream_t bitD;
+
+    /* Init */
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
+
+    /* decode */
+    {   BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
+        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        HUF_decodeStreamX4(ostart, &bitD, oend, dt, dtd.tableLog);
+    }
+
+    /* check */
+    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X4_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;
+        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) & (op4<(oend-(sizeof(bitD4.bitContainer)-1))) ; ) {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
+                                               const void *cSrc,
+                                               size_t cSrcSize,
+                                               const HUF_DTable *DTable);
+#if DYNAMIC_BMI2
+
+#define X(fn)                                                               \
+                                                                            \
+    static size_t fn##_default(                                             \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static TARGET_ATTRIBUTE("bmi2") size_t fn##_bmi2(                       \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+    {                                                                       \
+        if (bmi2) {                                                         \
+            return fn##_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);         \
+        }                                                                   \
+        return fn##_default(dst, dstSize, cSrc, cSrcSize, DTable);          \
+    }
+
+#else
+
+#define X(fn)                                                               \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+    {                                                                       \
+        (void)bmi2;                                                         \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }
+
+#endif
+
+X(HUF_decompress1X2_usingDTable_internal)
+X(HUF_decompress4X2_usingDTable_internal)
+X(HUF_decompress1X4_usingDTable_internal)
+X(HUF_decompress4X4_usingDTable_internal)
+
+#undef X
+
+
+size_t HUF_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize, workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+}
+
+
+size_t HUF_decompress1X2_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress1X2_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
+}
+
+size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress1X2_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
 size_t HUF_decompress4X2_usingDTable(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
@@ -328,22 +586,38 @@
 {
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+static size_t HUF_decompress4X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int bmi2)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX2_wksp (dctx, cSrc, cSrcSize,
+                                                workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+}
+
+size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
+{
+    return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, 0);
 }
 
 
 size_t HUF_decompress4X2_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
-    const BYTE* ip = (const BYTE*) cSrc;
-
-    size_t const hSize = HUF_readDTableX2 (dctx, cSrc, cSrcSize);
-    if (HUF_isError(hSize)) return hSize;
-    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
-    ip += hSize; cSrcSize -= hSize;
-
-    return HUF_decompress4X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, dctx);
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
 }
-
 size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
     HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
@@ -354,8 +628,6 @@
 /* *************************/
 /* double-symbols decoding */
 /* *************************/
-typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
-
 typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
 
 /* HUF_fillDTableX4Level2() :
@@ -400,7 +672,8 @@
     }   }
 }
 
-typedef U32 rankVal_t[HUF_TABLELOG_MAX][HUF_TABLELOG_MAX + 1];
+typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
+typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
 
 static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
                            const sortedSymbol_t* sortedList, const U32 sortedListSize,
@@ -444,22 +717,42 @@
     }
 }
 
-size_t HUF_readDTableX4 (HUF_DTable* DTable, const void* src, size_t srcSize)
+size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
+                             size_t srcSize, void* workSpace,
+                             size_t wkspSize)
 {
-    BYTE weightList[HUF_SYMBOLVALUE_MAX + 1];
-    sortedSymbol_t sortedSymbol[HUF_SYMBOLVALUE_MAX + 1];
-    U32 rankStats[HUF_TABLELOG_MAX + 1] = { 0 };
-    U32 rankStart0[HUF_TABLELOG_MAX + 2] = { 0 };
-    U32* const rankStart = rankStart0+1;
-    rankVal_t rankVal;
     U32 tableLog, maxW, sizeOfSort, nbSymbols;
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     U32 const maxTableLog = dtd.maxTableLog;
     size_t iSize;
     void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
     HUF_DEltX4* const dt = (HUF_DEltX4*)dtPtr;
+    U32 *rankStart;
 
-    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(HUF_DTable));   /* if compilation fails here, assertion is false */
+    rankValCol_t* rankVal;
+    U32* rankStats;
+    U32* rankStart0;
+    sortedSymbol_t* sortedSymbol;
+    BYTE* weightList;
+    size_t spaceUsed32 = 0;
+
+    rankVal = (rankValCol_t *)((U32 *)workSpace + spaceUsed32);
+    spaceUsed32 += (sizeof(rankValCol_t) * HUF_TABLELOG_MAX) >> 2;
+    rankStats = (U32 *)workSpace + spaceUsed32;
+    spaceUsed32 += HUF_TABLELOG_MAX + 1;
+    rankStart0 = (U32 *)workSpace + spaceUsed32;
+    spaceUsed32 += HUF_TABLELOG_MAX + 2;
+    sortedSymbol = (sortedSymbol_t *)workSpace + (spaceUsed32 * sizeof(U32)) / sizeof(sortedSymbol_t);
+    spaceUsed32 += HUF_ALIGN(sizeof(sortedSymbol_t) * (HUF_SYMBOLVALUE_MAX + 1), sizeof(U32)) >> 2;
+    weightList = (BYTE *)((U32 *)workSpace + spaceUsed32);
+    spaceUsed32 += HUF_ALIGN(HUF_SYMBOLVALUE_MAX + 1, sizeof(U32)) >> 2;
+
+    if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
+
+    rankStart = rankStart0 + 1;
+    memset(rankStats, 0, sizeof(U32) * (2 * HUF_TABLELOG_MAX + 2 + 1));
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(HUF_DTable));   /* if compiler fails here, assertion is wrong */
     if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
     /* memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
 
@@ -524,93 +817,11 @@
     return iSize;
 }
 
-
-static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
-{
-    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
-    memcpy(op, dt+val, 2);
-    BIT_skipBits(DStream, dt[val].nbBits);
-    return dt[val].length;
-}
-
-static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
-{
-    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
-    memcpy(op, dt+val, 1);
-    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
-    else {
-        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
-            BIT_skipBits(DStream, dt[val].nbBits);
-            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
-                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
-    }   }
-    return 1;
-}
-
-
-#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
-    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
-
-#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
-    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
-        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
-
-#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
-    if (MEM_64bits()) \
-        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
-
-static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+size_t HUF_readDTableX4(HUF_DTable* DTable, const void* src, size_t srcSize)
 {
-    BYTE* const pStart = p;
-
-    /* up to 8 symbols at a time */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
-        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
-        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
-    }
-
-    /* closer to end : up to 2 symbols at a time */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
-        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
-
-    while (p <= pEnd-2)
-        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
-
-    if (p < pEnd)
-        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
-
-    return p-pStart;
-}
-
-
-static size_t HUF_decompress1X4_usingDTable_internal(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    BIT_DStream_t bitD;
-
-    /* Init */
-    {   size_t const errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);
-        if (HUF_isError(errorCode)) return errorCode;
-    }
-
-    /* decode */
-    {   BYTE* const ostart = (BYTE*) dst;
-        BYTE* const oend = ostart + dstSize;
-        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
-        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
-        DTableDesc const dtd = HUF_getDTableDesc(DTable);
-        HUF_decodeStreamX4(ostart, &bitD, oend, dt, dtd.tableLog);
-    }
-
-    /* check */
-    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
-
-    /* decoded size */
-    return dstSize;
+  U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+  return HUF_readDTableX4_wksp(DTable, src, srcSize,
+                               workSpace, sizeof(workSpace));
 }
 
 size_t HUF_decompress1X4_usingDTable(
@@ -620,19 +831,31 @@
 {
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress1X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+    return HUF_decompress1X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
-size_t HUF_decompress1X4_DCtx (HUF_DTable* DCtx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress1X4_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX4 (DCtx, cSrc, cSrcSize);
+    size_t const hSize = HUF_readDTableX4_wksp(DCtx, cSrc, cSrcSize,
+                                               workSpace, wkspSize);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress1X4_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
+    return HUF_decompress1X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+}
+
+
+size_t HUF_decompress1X4_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress1X4_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
 }
 
 size_t HUF_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
@@ -641,99 +864,6 @@
     return HUF_decompress1X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
 }
 
-static size_t HUF_decompress4X4_usingDTable_internal(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
-
-    {   const BYTE* const istart = (const BYTE*) cSrc;
-        BYTE* const ostart = (BYTE*) dst;
-        BYTE* const oend = ostart + dstSize;
-        const void* const dtPtr = DTable+1;
-        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
-
-        /* Init */
-        BIT_DStream_t bitD1;
-        BIT_DStream_t bitD2;
-        BIT_DStream_t bitD3;
-        BIT_DStream_t bitD4;
-        size_t const length1 = MEM_readLE16(istart);
-        size_t const length2 = MEM_readLE16(istart+2);
-        size_t const length3 = MEM_readLE16(istart+4);
-        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
-        const BYTE* const istart1 = istart + 6;  /* jumpTable */
-        const BYTE* const istart2 = istart1 + length1;
-        const BYTE* const istart3 = istart2 + length2;
-        const BYTE* const istart4 = istart3 + length3;
-        size_t const segmentSize = (dstSize+3) / 4;
-        BYTE* const opStart2 = ostart + segmentSize;
-        BYTE* const opStart3 = opStart2 + segmentSize;
-        BYTE* const opStart4 = opStart3 + segmentSize;
-        BYTE* op1 = ostart;
-        BYTE* op2 = opStart2;
-        BYTE* op3 = opStart3;
-        BYTE* op4 = opStart4;
-        U32 endSignal;
-        DTableDesc const dtd = HUF_getDTableDesc(DTable);
-        U32 const dtLog = dtd.tableLog;
-
-        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
-        { size_t const errorCode = BIT_initDStream(&bitD1, istart1, length1);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD2, istart2, length2);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD3, istart3, length3);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD4, istart4, length4);
-          if (HUF_isError(errorCode)) return errorCode; }
-
-        /* 16-32 symbols per loop (4-8 symbols per stream) */
-        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
-        for ( ; (endSignal==BIT_DStream_unfinished) & (op4<(oend-(sizeof(bitD4.bitContainer)-1))) ; ) {
-            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
-            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
-
-            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
-        }
-
-        /* check corruption */
-        if (op1 > opStart2) return ERROR(corruption_detected);
-        if (op2 > opStart3) return ERROR(corruption_detected);
-        if (op3 > opStart4) return ERROR(corruption_detected);
-        /* note : op4 already verified within main loop */
-
-        /* finish bitStreams one by one */
-        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
-        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
-        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
-        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
-
-        /* check */
-        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
-          if (!endCheck) return ERROR(corruption_detected); }
-
-        /* decoded size */
-        return dstSize;
-    }
-}
-
-
 size_t HUF_decompress4X4_usingDTable(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
@@ -741,20 +871,38 @@
 {
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
-
-size_t HUF_decompress4X4_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+static size_t HUF_decompress4X4_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int bmi2)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t hSize = HUF_readDTableX4 (dctx, cSrc, cSrcSize);
+    size_t hSize = HUF_readDTableX4_wksp(dctx, cSrc, cSrcSize,
+                                         workSpace, wkspSize);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx);
+    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+}
+
+size_t HUF_decompress4X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
+{
+    return HUF_decompress4X4_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, /* bmi2 */ 0);
+}
+
+
+size_t HUF_decompress4X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress4X4_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
 }
 
 size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
@@ -773,8 +921,8 @@
                                     const HUF_DTable* DTable)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
-    return dtd.tableType ? HUF_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
-                           HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+    return dtd.tableType ? HUF_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
+                           HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
 size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize,
@@ -782,8 +930,8 @@
                                     const HUF_DTable* DTable)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
-    return dtd.tableType ? HUF_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
-                           HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+    return dtd.tableType ? HUF_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
+                           HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
 
@@ -810,21 +958,22 @@
 };
 
 /** HUF_selectDecoder() :
-*   Tells which decoder is likely to decode faster,
-*   based on a set of pre-determined metrics.
-*   @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
-*   Assumption : 0 < cSrcSize < dstSize <= 128 KB */
+ *  Tells which decoder is likely to decode faster,
+ *  based on a set of pre-computed metrics.
+ * @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
+ *  Assumption : 0 < dstSize <= 128 KB */
 U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
 {
+    assert(dstSize > 0);
+    assert(dstSize <= 128 KB);
     /* decoder timing evaluation */
-    U32 const Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
-    U32 const D256 = (U32)(dstSize >> 8);
-    U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
-    U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
-    DTime1 += DTime1 >> 3;  /* advantage to algorithm using less memory, for cache eviction */
-
-    return DTime1 < DTime0;
-}
+    {   U32 const Q = (cSrcSize >= dstSize) ? 15 : (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 */
+        U32 const D256 = (U32)(dstSize >> 8);
+        U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
+        U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
+        DTime1 += DTime1 >> 3;  /* advantage to algorithm using less memory, to reduce cache eviction */
+        return DTime1 < DTime0;
+}   }
 
 
 typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
@@ -858,19 +1007,32 @@
     }
 }
 
-size_t HUF_decompress4X_hufOnly (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress4X_hufOnly_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+                                         workSpace, sizeof(workSpace));
+}
+
+
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst,
+                                     size_t dstSize, const void* cSrc,
+                                     size_t cSrcSize, void* workSpace,
+                                     size_t wkspSize)
 {
     /* validation checks */
     if (dstSize == 0) return ERROR(dstSize_tooSmall);
-    if ((cSrcSize >= dstSize) || (cSrcSize <= 1)) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == 0) return ERROR(corruption_detected);
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-        return algoNb ? HUF_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
-                        HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+        return algoNb ? HUF_decompress4X4_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize):
+                        HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
     }
 }
 
-size_t HUF_decompress1X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                  const void* cSrc, size_t cSrcSize,
+                                  void* workSpace, size_t wkspSize)
 {
     /* validation checks */
     if (dstSize == 0) return ERROR(dstSize_tooSmall);
@@ -879,7 +1041,56 @@
     if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-        return algoNb ? HUF_decompress1X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
-                        HUF_decompress1X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+        return algoNb ? HUF_decompress1X4_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize):
+                        HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize);
     }
 }
+
+size_t HUF_decompress1X_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
+                             const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress1X_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+                                      workSpace, sizeof(workSpace));
+}
+
+
+size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+    return dtd.tableType ? HUF_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
+                           HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+}
+
+size_t HUF_decompress1X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX2_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+}
+
+size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+    return dtd.tableType ? HUF_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
+                           HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+}
+
+size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize == 0) return ERROR(corruption_detected);
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUF_decompress4X4_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2) :
+                        HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+    }
+}
--- a/contrib/python-zstandard/zstd/decompress/zstd_decompress.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/decompress/zstd_decompress.c	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,11 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 
@@ -13,8 +14,9 @@
 *****************************************************************/
 /*!
  * HEAPMODE :
- * Select how default decompression function ZSTD_decompress() will allocate memory,
- * in memory stack (0), or in memory heap (1, requires malloc())
+ * Select how default decompression function ZSTD_decompress() allocates its context,
+ * on stack (0), or into heap (1, default; requires malloc()).
+ * Note that functions with explicit context such as ZSTD_decompressDCtx() are unaffected.
  */
 #ifndef ZSTD_HEAPMODE
 #  define ZSTD_HEAPMODE 1
@@ -22,19 +24,20 @@
 
 /*!
 *  LEGACY_SUPPORT :
-*  if set to 1, ZSTD_decompress() can decode older formats (v0.1+)
+*  if set to 1+, ZSTD_decompress() can decode older formats (v0.1+)
 */
 #ifndef ZSTD_LEGACY_SUPPORT
 #  define ZSTD_LEGACY_SUPPORT 0
 #endif
 
 /*!
-*  MAXWINDOWSIZE_DEFAULT :
-*  maximum window size accepted by DStream, by default.
-*  Frames requiring more memory will be rejected.
-*/
+ *  MAXWINDOWSIZE_DEFAULT :
+ *  maximum window size accepted by DStream __by default__.
+ *  Frames requiring more memory will be rejected.
+ *  It's possible to set a different limit using ZSTD_DCtx_setMaxWindowSize().
+ */
 #ifndef ZSTD_MAXWINDOWSIZE_DEFAULT
-#  define ZSTD_MAXWINDOWSIZE_DEFAULT ((1 << ZSTD_WINDOWLOG_MAX) + 1)   /* defined within zstd.h */
+#  define ZSTD_MAXWINDOWSIZE_DEFAULT (((U32)1 << ZSTD_WINDOWLOG_DEFAULTMAX) + 1)
 #endif
 
 
@@ -42,9 +45,8 @@
 *  Dependencies
 *********************************************************/
 #include <string.h>      /* memcpy, memmove, memset */
+#include "cpu.h"
 #include "mem.h"         /* low level memory routines */
-#define XXH_STATIC_LINKING_ONLY   /* XXH64_state_t */
-#include "xxhash.h"      /* XXH64_* */
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
 #define HUF_STATIC_LINKING_ONLY
@@ -56,17 +58,8 @@
 #endif
 
 
-#if defined(_MSC_VER)
-#  include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
-#  define ZSTD_PREFETCH(ptr)   _mm_prefetch((const char*)ptr, _MM_HINT_T0)
-#elif defined(__GNUC__)
-#  define ZSTD_PREFETCH(ptr)   __builtin_prefetch(ptr, 0, 0)
-#else
-#  define ZSTD_PREFETCH(ptr)   /* disabled */
-#endif
-
 /*-*************************************
-*  Macros
+*  Errors
 ***************************************/
 #define ZSTD_isError ERR_isError   /* for inlining */
 #define FSE_isError  ERR_isError
@@ -87,119 +80,183 @@
                ZSTDds_decompressLastBlock, ZSTDds_checkChecksum,
                ZSTDds_decodeSkippableHeader, ZSTDds_skipFrame } ZSTD_dStage;
 
+typedef enum { zdss_init=0, zdss_loadHeader,
+               zdss_read, zdss_load, zdss_flush } ZSTD_dStreamStage;
+
+
+typedef struct {
+    U32 fastMode;
+    U32 tableLog;
+} ZSTD_seqSymbol_header;
+
+typedef struct {
+    U16  nextState;
+    BYTE nbAdditionalBits;
+    BYTE nbBits;
+    U32  baseValue;
+} ZSTD_seqSymbol;
+
+#define SEQSYMBOL_TABLE_SIZE(log)   (1 + (1 << (log)))
+
+typedef struct {
+    ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];
+    ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];
+    ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];
+    HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
+    U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    U32 rep[ZSTD_REP_NUM];
+} ZSTD_entropyDTables_t;
+
 struct ZSTD_DCtx_s
 {
-    const FSE_DTable* LLTptr;
-    const FSE_DTable* MLTptr;
-    const FSE_DTable* OFTptr;
+    const ZSTD_seqSymbol* LLTptr;
+    const ZSTD_seqSymbol* MLTptr;
+    const ZSTD_seqSymbol* OFTptr;
     const HUF_DTable* HUFptr;
-    FSE_DTable LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
-    FSE_DTable OFTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
-    FSE_DTable MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
-    HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
-    const void* previousDstEnd;
-    const void* base;
-    const void* vBase;
-    const void* dictEnd;
+    ZSTD_entropyDTables_t entropy;
+    const void* previousDstEnd;   /* detect continuity */
+    const void* base;             /* start of current segment */
+    const void* vBase;            /* virtual start of previous segment if it was just before current one */
+    const void* dictEnd;          /* end of previous segment */
     size_t expected;
-    U32 rep[ZSTD_REP_NUM];
-    ZSTD_frameParams fParams;
-    blockType_e bType;   /* used in ZSTD_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTD_frameHeader fParams;
+    U64 decodedSize;
+    blockType_e bType;            /* used in ZSTD_decompressContinue(), store blockType between block header decoding and block decompression stages */
     ZSTD_dStage stage;
     U32 litEntropy;
     U32 fseEntropy;
     XXH64_state_t xxhState;
     size_t headerSize;
     U32 dictID;
+    ZSTD_format_e format;
     const BYTE* litPtr;
     ZSTD_customMem customMem;
     size_t litSize;
     size_t rleSize;
-    BYTE litBuffer[ZSTD_BLOCKSIZE_ABSOLUTEMAX + WILDCOPY_OVERLENGTH];
+    size_t staticSize;
+    int bmi2;                     /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
+
+    /* streaming */
+    ZSTD_DDict* ddictLocal;
+    const ZSTD_DDict* ddict;
+    ZSTD_dStreamStage streamStage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    size_t maxWindowSize;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t lhSize;
+    void* legacyContext;
+    U32 previousLegacyVersion;
+    U32 legacyVersion;
+    U32 hostageByte;
+
+    /* workspace */
+    BYTE litBuffer[ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
     BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
 };  /* typedef'd to ZSTD_DCtx within "zstd.h" */
 
-size_t ZSTD_sizeof_DCtx (const ZSTD_DCtx* dctx) { return (dctx==NULL) ? 0 : sizeof(ZSTD_DCtx); }
+size_t ZSTD_sizeof_DCtx (const ZSTD_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support sizeof NULL */
+    return sizeof(*dctx)
+           + ZSTD_sizeof_DDict(dctx->ddictLocal)
+           + dctx->inBuffSize + dctx->outBuffSize;
+}
 
 size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
 
-size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
+
+static size_t ZSTD_startingInputLength(ZSTD_format_e format)
+{
+    size_t const startingInputLength = (format==ZSTD_f_zstd1_magicless) ?
+                    ZSTD_frameHeaderSize_prefix - ZSTD_frameIdSize :
+                    ZSTD_frameHeaderSize_prefix;
+    ZSTD_STATIC_ASSERT(ZSTD_FRAMEHEADERSIZE_PREFIX >= ZSTD_FRAMEIDSIZE);
+    /* only supports formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless */
+    assert( (format == ZSTD_f_zstd1) || (format == ZSTD_f_zstd1_magicless) );
+    return startingInputLength;
+}
+
+static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
 {
-    dctx->expected = ZSTD_frameHeaderSize_prefix;
-    dctx->stage = ZSTDds_getFrameHeaderSize;
-    dctx->previousDstEnd = NULL;
-    dctx->base = NULL;
-    dctx->vBase = NULL;
-    dctx->dictEnd = NULL;
-    dctx->hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
-    dctx->litEntropy = dctx->fseEntropy = 0;
-    dctx->dictID = 0;
-    MEM_STATIC_ASSERT(sizeof(dctx->rep) == sizeof(repStartValue));
-    memcpy(dctx->rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
-    dctx->LLTptr = dctx->LLTable;
-    dctx->MLTptr = dctx->MLTable;
-    dctx->OFTptr = dctx->OFTable;
-    dctx->HUFptr = dctx->hufTable;
-    return 0;
+    dctx->format = ZSTD_f_zstd1;  /* ZSTD_decompressBegin() invokes ZSTD_startingInputLength() with argument dctx->format */
+    dctx->staticSize  = 0;
+    dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+    dctx->ddict       = NULL;
+    dctx->ddictLocal  = NULL;
+    dctx->inBuff      = NULL;
+    dctx->inBuffSize  = 0;
+    dctx->outBuffSize = 0;
+    dctx->streamStage = zdss_init;
+    dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+}
+
+ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
+{
+    ZSTD_DCtx* const dctx = (ZSTD_DCtx*) workspace;
+
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+    if (workspaceSize < sizeof(ZSTD_DCtx)) return NULL;  /* minimum size */
+
+    ZSTD_initDCtx_internal(dctx);
+    dctx->staticSize = workspaceSize;
+    dctx->inBuff = (char*)(dctx+1);
+    return dctx;
 }
 
 ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
 {
-    ZSTD_DCtx* dctx;
-
-    if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
-    if (!customMem.customAlloc || !customMem.customFree) return NULL;
-
-    dctx = (ZSTD_DCtx*)ZSTD_malloc(sizeof(ZSTD_DCtx), customMem);
-    if (!dctx) return NULL;
-    memcpy(&dctx->customMem, &customMem, sizeof(customMem));
-    ZSTD_decompressBegin(dctx);
-    return dctx;
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+
+    {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_malloc(sizeof(*dctx), customMem);
+        if (!dctx) return NULL;
+        dctx->customMem = customMem;
+        dctx->legacyContext = NULL;
+        dctx->previousLegacyVersion = 0;
+        ZSTD_initDCtx_internal(dctx);
+        return dctx;
+    }
 }
 
 ZSTD_DCtx* ZSTD_createDCtx(void)
 {
-    return ZSTD_createDCtx_advanced(defaultCustomMem);
+    DEBUGLOG(3, "ZSTD_createDCtx");
+    return ZSTD_createDCtx_advanced(ZSTD_defaultCMem);
 }
 
 size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
 {
     if (dctx==NULL) return 0;   /* support free on NULL */
-    ZSTD_free(dctx, dctx->customMem);
-    return 0;   /* reserved as a potential error code in the future */
+    if (dctx->staticSize) return ERROR(memory_allocation);   /* not compatible with static DCtx */
+    {   ZSTD_customMem const cMem = dctx->customMem;
+        ZSTD_freeDDict(dctx->ddictLocal);
+        dctx->ddictLocal = NULL;
+        ZSTD_free(dctx->inBuff, cMem);
+        dctx->inBuff = NULL;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+        if (dctx->legacyContext)
+            ZSTD_freeLegacyStreamContext(dctx->legacyContext, dctx->previousLegacyVersion);
+#endif
+        ZSTD_free(dctx, cMem);
+        return 0;
+    }
 }
 
+/* no longer useful */
 void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
 {
-    size_t const workSpaceSize = (ZSTD_BLOCKSIZE_ABSOLUTEMAX+WILDCOPY_OVERLENGTH) + ZSTD_frameHeaderSize_max;
-    memcpy(dstDCtx, srcDCtx, sizeof(ZSTD_DCtx) - workSpaceSize);  /* no need to copy workspace */
-}
-
-static void ZSTD_refDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
-{
-    ZSTD_decompressBegin(dstDCtx);  /* init */
-    if (srcDCtx) {   /* support refDCtx on NULL */
-        dstDCtx->dictEnd = srcDCtx->dictEnd;
-        dstDCtx->vBase = srcDCtx->vBase;
-        dstDCtx->base = srcDCtx->base;
-        dstDCtx->previousDstEnd = srcDCtx->previousDstEnd;
-        dstDCtx->dictID = srcDCtx->dictID;
-        dstDCtx->litEntropy = srcDCtx->litEntropy;
-        dstDCtx->fseEntropy = srcDCtx->fseEntropy;
-        dstDCtx->LLTptr = srcDCtx->LLTable;
-        dstDCtx->MLTptr = srcDCtx->MLTable;
-        dstDCtx->OFTptr = srcDCtx->OFTable;
-        dstDCtx->HUFptr = srcDCtx->hufTable;
-        dstDCtx->rep[0] = srcDCtx->rep[0];
-        dstDCtx->rep[1] = srcDCtx->rep[1];
-        dstDCtx->rep[2] = srcDCtx->rep[2];
-    }
+    size_t const toCopy = (size_t)((char*)(&dstDCtx->inBuff) - (char*)dstDCtx);
+    memcpy(dstDCtx, srcDCtx, toCopy);  /* no need to copy workspace */
 }
 
 
 /*-*************************************************************
-*   Decompression section
-***************************************************************/
+ *   Frame header decoding
+ ***************************************************************/
 
 /*! ZSTD_isFrame() :
  *  Tells if the content of `buffer` starts with a valid Frame Identifier.
@@ -208,7 +265,7 @@
  *  Note 3 : Skippable Frame Identifiers are considered valid. */
 unsigned ZSTD_isFrame(const void* buffer, size_t size)
 {
-    if (size < 4) return 0;
+    if (size < ZSTD_frameIdSize) return 0;
     {   U32 const magic = MEM_readLE32(buffer);
         if (magic == ZSTD_MAGICNUMBER) return 1;
         if ((magic & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
@@ -219,70 +276,91 @@
     return 0;
 }
 
-
-/** ZSTD_frameHeaderSize() :
-*   srcSize must be >= ZSTD_frameHeaderSize_prefix.
-*   @return : size of the Frame Header */
-static size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
+/** ZSTD_frameHeaderSize_internal() :
+ *  srcSize must be large enough to reach header size fields.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless.
+ * @return : size of the Frame Header
+ *           or an error code, which can be tested with ZSTD_isError() */
+static size_t ZSTD_frameHeaderSize_internal(const void* src, size_t srcSize, ZSTD_format_e format)
 {
-    if (srcSize < ZSTD_frameHeaderSize_prefix) return ERROR(srcSize_wrong);
-    {   BYTE const fhd = ((const BYTE*)src)[4];
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+    if (srcSize < minInputSize) return ERROR(srcSize_wrong);
+
+    {   BYTE const fhd = ((const BYTE*)src)[minInputSize-1];
         U32 const dictID= fhd & 3;
         U32 const singleSegment = (fhd >> 5) & 1;
         U32 const fcsId = fhd >> 6;
-        return ZSTD_frameHeaderSize_prefix + !singleSegment + ZSTD_did_fieldSize[dictID] + ZSTD_fcs_fieldSize[fcsId]
-                + (singleSegment && !fcsId);
+        return minInputSize + !singleSegment
+             + ZSTD_did_fieldSize[dictID] + ZSTD_fcs_fieldSize[fcsId]
+             + (singleSegment && !fcsId);
     }
 }
 
-
-/** ZSTD_getFrameParams() :
-*   decode Frame Header, or require larger `srcSize`.
-*   @return : 0, `fparamsPtr` is correctly filled,
-*            >0, `srcSize` is too small, result is expected `srcSize`,
-*             or an error code, which can be tested using ZSTD_isError() */
-size_t ZSTD_getFrameParams(ZSTD_frameParams* fparamsPtr, const void* src, size_t srcSize)
+/** ZSTD_frameHeaderSize() :
+ *  srcSize must be >= ZSTD_frameHeaderSize_prefix.
+ * @return : size of the Frame Header */
+size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
+{
+    return ZSTD_frameHeaderSize_internal(src, srcSize, ZSTD_f_zstd1);
+}
+
+
+/** ZSTD_getFrameHeader_internal() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+static size_t ZSTD_getFrameHeader_internal(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
 {
     const BYTE* ip = (const BYTE*)src;
-
-    if (srcSize < ZSTD_frameHeaderSize_prefix) return ZSTD_frameHeaderSize_prefix;
-    if (MEM_readLE32(src) != ZSTD_MAGICNUMBER) {
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+
+    if (srcSize < minInputSize) return minInputSize;
+
+    if ( (format != ZSTD_f_zstd1_magicless)
+      && (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
         if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
-            if (srcSize < ZSTD_skippableHeaderSize) return ZSTD_skippableHeaderSize; /* magic number + skippable frame length */
-            memset(fparamsPtr, 0, sizeof(*fparamsPtr));
-            fparamsPtr->frameContentSize = MEM_readLE32((const char *)src + 4);
-            fparamsPtr->windowSize = 0; /* windowSize==0 means a frame is skippable */
+            /* skippable frame */
+            if (srcSize < ZSTD_skippableHeaderSize)
+                return ZSTD_skippableHeaderSize; /* magic number + frame length */
+            memset(zfhPtr, 0, sizeof(*zfhPtr));
+            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_frameIdSize);
+            zfhPtr->frameType = ZSTD_skippableFrame;
             return 0;
         }
         return ERROR(prefix_unknown);
     }
 
     /* ensure there is enough `srcSize` to fully read/decode frame header */
-    { size_t const fhsize = ZSTD_frameHeaderSize(src, srcSize);
-      if (srcSize < fhsize) return fhsize; }
-
-    {   BYTE const fhdByte = ip[4];
-        size_t pos = 5;
+    {   size_t const fhsize = ZSTD_frameHeaderSize_internal(src, srcSize, format);
+        if (srcSize < fhsize) return fhsize;
+        zfhPtr->headerSize = (U32)fhsize;
+    }
+
+    {   BYTE const fhdByte = ip[minInputSize-1];
+        size_t pos = minInputSize;
         U32 const dictIDSizeCode = fhdByte&3;
         U32 const checksumFlag = (fhdByte>>2)&1;
         U32 const singleSegment = (fhdByte>>5)&1;
         U32 const fcsID = fhdByte>>6;
-        U32 const windowSizeMax = 1U << ZSTD_WINDOWLOG_MAX;
-        U32 windowSize = 0;
+        U64 windowSize = 0;
         U32 dictID = 0;
-        U64 frameContentSize = 0;
-        if ((fhdByte & 0x08) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits, which must be zero */
+        U64 frameContentSize = ZSTD_CONTENTSIZE_UNKNOWN;
+        if ((fhdByte & 0x08) != 0)
+            return ERROR(frameParameter_unsupported); /* reserved bits, must be zero */
+
         if (!singleSegment) {
             BYTE const wlByte = ip[pos++];
             U32 const windowLog = (wlByte >> 3) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
-            if (windowLog > ZSTD_WINDOWLOG_MAX) return ERROR(frameParameter_windowTooLarge);  /* avoids issue with 1 << windowLog */
-            windowSize = (1U << windowLog);
+            if (windowLog > ZSTD_WINDOWLOG_MAX)
+                return ERROR(frameParameter_windowTooLarge);
+            windowSize = (1ULL << windowLog);
             windowSize += (windowSize >> 3) * (wlByte&7);
         }
-
         switch(dictIDSizeCode)
         {
-            default:   /* impossible */
+            default: assert(0);  /* impossible */
             case 0 : break;
             case 1 : dictID = ip[pos]; pos++; break;
             case 2 : dictID = MEM_readLE16(ip+pos); pos+=2; break;
@@ -290,40 +368,121 @@
         }
         switch(fcsID)
         {
-            default:   /* impossible */
+            default: assert(0);  /* impossible */
             case 0 : if (singleSegment) frameContentSize = ip[pos]; break;
             case 1 : frameContentSize = MEM_readLE16(ip+pos)+256; break;
             case 2 : frameContentSize = MEM_readLE32(ip+pos); break;
             case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
         }
-        if (!windowSize) windowSize = (U32)frameContentSize;
-        if (windowSize > windowSizeMax) return ERROR(frameParameter_windowTooLarge);
-        fparamsPtr->frameContentSize = frameContentSize;
-        fparamsPtr->windowSize = windowSize;
-        fparamsPtr->dictID = dictID;
-        fparamsPtr->checksumFlag = checksumFlag;
+        if (singleSegment) windowSize = frameContentSize;
+
+        zfhPtr->frameType = ZSTD_frame;
+        zfhPtr->frameContentSize = frameContentSize;
+        zfhPtr->windowSize = windowSize;
+        zfhPtr->blockSizeMax = (unsigned) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+        zfhPtr->dictID = dictID;
+        zfhPtr->checksumFlag = checksumFlag;
     }
     return 0;
 }
 
+/** ZSTD_getFrameHeader() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : this function does not consume input, it only reads it.
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize)
+{
+    return ZSTD_getFrameHeader_internal(zfhPtr, src, srcSize, ZSTD_f_zstd1);
+}
+
+
+/** ZSTD_getFrameContentSize() :
+ *  compatible with legacy mode
+ * @return : decompressed size of the single frame pointed to be `src` if known, otherwise
+ *         - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
+ *         - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small) */
+unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize)
+{
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize)) {
+        unsigned long long const ret = ZSTD_getDecompressedSize_legacy(src, srcSize);
+        return ret == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : ret;
+    }
+#endif
+    {   ZSTD_frameHeader zfh;
+        if (ZSTD_getFrameHeader(&zfh, src, srcSize) != 0)
+            return ZSTD_CONTENTSIZE_ERROR;
+        if (zfh.frameType == ZSTD_skippableFrame) {
+            return 0;
+        } else {
+            return zfh.frameContentSize;
+    }   }
+}
+
+/** ZSTD_findDecompressedSize() :
+ *  compatible with legacy mode
+ *  `srcSize` must be the exact length of some number of ZSTD compressed and/or
+ *      skippable frames
+ *  @return : decompressed size of the frames contained */
+unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
+{
+    unsigned long long totalDstSize = 0;
+
+    while (srcSize >= ZSTD_frameHeaderSize_prefix) {
+        U32 const magicNumber = MEM_readLE32(src);
+
+        if ((magicNumber & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+            size_t skippableSize;
+            if (srcSize < ZSTD_skippableHeaderSize)
+                return ERROR(srcSize_wrong);
+            skippableSize = MEM_readLE32((const BYTE *)src + ZSTD_frameIdSize)
+                          + ZSTD_skippableHeaderSize;
+            if (srcSize < skippableSize) {
+                return ZSTD_CONTENTSIZE_ERROR;
+            }
+
+            src = (const BYTE *)src + skippableSize;
+            srcSize -= skippableSize;
+            continue;
+        }
+
+        {   unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
+            if (ret >= ZSTD_CONTENTSIZE_ERROR) return ret;
+
+            /* check for overflow */
+            if (totalDstSize + ret < totalDstSize) return ZSTD_CONTENTSIZE_ERROR;
+            totalDstSize += ret;
+        }
+        {   size_t const frameSrcSize = ZSTD_findFrameCompressedSize(src, srcSize);
+            if (ZSTD_isError(frameSrcSize)) {
+                return ZSTD_CONTENTSIZE_ERROR;
+            }
+
+            src = (const BYTE *)src + frameSrcSize;
+            srcSize -= frameSrcSize;
+        }
+    }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
+
+    if (srcSize) return ZSTD_CONTENTSIZE_ERROR;
+
+    return totalDstSize;
+}
 
 /** ZSTD_getDecompressedSize() :
 *   compatible with legacy mode
 *   @return : decompressed size if known, 0 otherwise
               note : 0 can mean any of the following :
-                   - decompressed size is not present within frame header
+                   - frame content is empty
+                   - decompressed size field is not present in frame header
                    - frame header unknown / not supported
                    - frame header not complete (`srcSize` too small) */
 unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
 {
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT==1)
-    if (ZSTD_isLegacy(src, srcSize)) return ZSTD_getDecompressedSize_legacy(src, srcSize);
-#endif
-    {   ZSTD_frameParams fparams;
-        size_t const frResult = ZSTD_getFrameParams(&fparams, src, srcSize);
-        if (frResult!=0) return 0;
-        return fparams.frameContentSize;
-    }
+    unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
+    ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_ERROR < ZSTD_CONTENTSIZE_UNKNOWN);
+    return (ret >= ZSTD_CONTENTSIZE_ERROR) ? 0 : ret;
 }
 
 
@@ -332,25 +491,24 @@
 *   @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
 static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
 {
-    size_t const result = ZSTD_getFrameParams(&(dctx->fParams), src, headerSize);
-    if (ZSTD_isError(result)) return result;  /* invalid header */
-    if (result>0) return ERROR(srcSize_wrong);   /* headerSize too small */
-    if (dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID)) return ERROR(dictionary_wrong);
+    size_t const result = ZSTD_getFrameHeader_internal(&(dctx->fParams), src, headerSize, dctx->format);
+    if (ZSTD_isError(result)) return result;    /* invalid header */
+    if (result>0) return ERROR(srcSize_wrong);  /* headerSize too small */
+    if (dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID))
+        return ERROR(dictionary_wrong);
     if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
     return 0;
 }
 
 
-typedef struct
-{
-    blockType_e blockType;
-    U32 lastBlock;
-    U32 origSize;
-} blockProperties_t;
+/*-*************************************************************
+ *   Block decoding
+ ***************************************************************/
 
 /*! ZSTD_getcBlockSize() :
 *   Provides the size of compressed block from block header `src` */
-size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
+                          blockProperties_t* bpPtr)
 {
     if (srcSize < ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
     {   U32 const cBlockHeader = MEM_readLE24(src);
@@ -365,7 +523,8 @@
 }
 
 
-static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
+                          const void* src, size_t srcSize)
 {
     if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
     memcpy(dst, src, srcSize);
@@ -373,7 +532,9 @@
 }
 
 
-static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize, size_t regenSize)
+static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                               size_t regenSize)
 {
     if (srcSize != 1) return ERROR(srcSize_wrong);
     if (regenSize > dstCapacity) return ERROR(dstSize_tooSmall);
@@ -382,7 +543,8 @@
 }
 
 /*! ZSTD_decodeLiteralsBlock() :
-    @return : nb of bytes read from src (< srcSize ) */
+ * @return : nb of bytes read from src (< srcSize )
+ *  note : symbol not declared but exposed for fullbench */
 size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                           const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
 {
@@ -424,22 +586,24 @@
                     litCSize = (lhc >> 22) + (istart[4] << 10);
                     break;
                 }
-                if (litSize > ZSTD_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
+                if (litSize > ZSTD_BLOCKSIZE_MAX) return ERROR(corruption_detected);
                 if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
 
                 if (HUF_isError((litEncType==set_repeat) ?
                                     ( singleStream ?
-                                        HUF_decompress1X_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr) :
-                                        HUF_decompress4X_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr) ) :
+                                        HUF_decompress1X_usingDTable_bmi2(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2) :
+                                        HUF_decompress4X_usingDTable_bmi2(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2) ) :
                                     ( singleStream ?
-                                        HUF_decompress1X2_DCtx(dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize) :
-                                        HUF_decompress4X_hufOnly (dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize)) ))
+                                        HUF_decompress1X2_DCtx_wksp_bmi2(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                                                                         dctx->entropy.workspace, sizeof(dctx->entropy.workspace), dctx->bmi2) :
+                                        HUF_decompress4X_hufOnly_wksp_bmi2(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                                                                           dctx->entropy.workspace, sizeof(dctx->entropy.workspace), dctx->bmi2))))
                     return ERROR(corruption_detected);
 
                 dctx->litPtr = dctx->litBuffer;
                 dctx->litSize = litSize;
                 dctx->litEntropy = 1;
-                if (litEncType==set_compressed) dctx->HUFptr = dctx->hufTable;
+                if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
                 memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
                 return litCSize + lhSize;
             }
@@ -496,7 +660,7 @@
                     if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
                     break;
                 }
-                if (litSize > ZSTD_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
+                if (litSize > ZSTD_BLOCKSIZE_MAX) return ERROR(corruption_detected);
                 memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
                 dctx->litPtr = dctx->litBuffer;
                 dctx->litSize = litSize;
@@ -508,227 +672,275 @@
     }
 }
 
-
-typedef union {
-    FSE_decode_t realData;
-    U32 alignedBy4;
-} FSE_decode_t4;
-
-static const FSE_decode_t4 LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = {
-    { { LL_DEFAULTNORMLOG, 1, 1 } }, /* header : tableLog, fastMode, fastMode */
-    { {  0,  0,  4 } },              /* 0 : base, symbol, bits */
-    { { 16,  0,  4 } },
-    { { 32,  1,  5 } },
-    { {  0,  3,  5 } },
-    { {  0,  4,  5 } },
-    { {  0,  6,  5 } },
-    { {  0,  7,  5 } },
-    { {  0,  9,  5 } },
-    { {  0, 10,  5 } },
-    { {  0, 12,  5 } },
-    { {  0, 14,  6 } },
-    { {  0, 16,  5 } },
-    { {  0, 18,  5 } },
-    { {  0, 19,  5 } },
-    { {  0, 21,  5 } },
-    { {  0, 22,  5 } },
-    { {  0, 24,  5 } },
-    { { 32, 25,  5 } },
-    { {  0, 26,  5 } },
-    { {  0, 27,  6 } },
-    { {  0, 29,  6 } },
-    { {  0, 31,  6 } },
-    { { 32,  0,  4 } },
-    { {  0,  1,  4 } },
-    { {  0,  2,  5 } },
-    { { 32,  4,  5 } },
-    { {  0,  5,  5 } },
-    { { 32,  7,  5 } },
-    { {  0,  8,  5 } },
-    { { 32, 10,  5 } },
-    { {  0, 11,  5 } },
-    { {  0, 13,  6 } },
-    { { 32, 16,  5 } },
-    { {  0, 17,  5 } },
-    { { 32, 19,  5 } },
-    { {  0, 20,  5 } },
-    { { 32, 22,  5 } },
-    { {  0, 23,  5 } },
-    { {  0, 25,  4 } },
-    { { 16, 25,  4 } },
-    { { 32, 26,  5 } },
-    { {  0, 28,  6 } },
-    { {  0, 30,  6 } },
-    { { 48,  0,  4 } },
-    { { 16,  1,  4 } },
-    { { 32,  2,  5 } },
-    { { 32,  3,  5 } },
-    { { 32,  5,  5 } },
-    { { 32,  6,  5 } },
-    { { 32,  8,  5 } },
-    { { 32,  9,  5 } },
-    { { 32, 11,  5 } },
-    { { 32, 12,  5 } },
-    { {  0, 15,  6 } },
-    { { 32, 17,  5 } },
-    { { 32, 18,  5 } },
-    { { 32, 20,  5 } },
-    { { 32, 21,  5 } },
-    { { 32, 23,  5 } },
-    { { 32, 24,  5 } },
-    { {  0, 35,  6 } },
-    { {  0, 34,  6 } },
-    { {  0, 33,  6 } },
-    { {  0, 32,  6 } },
+/* Default FSE distribution tables.
+ * These are pre-calculated FSE decoding tables using default distributions as defined in specification :
+ * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#default-distributions
+ * They were generated programmatically with following method :
+ * - start from default distributions, present in /lib/common/zstd_internal.h
+ * - generate tables normally, using ZSTD_buildFSETable()
+ * - printout the content of tables
+ * - pretify output, report below, test with fuzzer to ensure it's correct */
+
+/* Default FSE distribution table for Literal Lengths */
+static const ZSTD_seqSymbol LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = {
+     {  1,  1,  1, LL_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+     /* nextState, nbAddBits, nbBits, baseVal */
+     {  0,  0,  4,    0},  { 16,  0,  4,    0},
+     { 32,  0,  5,    1},  {  0,  0,  5,    3},
+     {  0,  0,  5,    4},  {  0,  0,  5,    6},
+     {  0,  0,  5,    7},  {  0,  0,  5,    9},
+     {  0,  0,  5,   10},  {  0,  0,  5,   12},
+     {  0,  0,  6,   14},  {  0,  1,  5,   16},
+     {  0,  1,  5,   20},  {  0,  1,  5,   22},
+     {  0,  2,  5,   28},  {  0,  3,  5,   32},
+     {  0,  4,  5,   48},  { 32,  6,  5,   64},
+     {  0,  7,  5,  128},  {  0,  8,  6,  256},
+     {  0, 10,  6, 1024},  {  0, 12,  6, 4096},
+     { 32,  0,  4,    0},  {  0,  0,  4,    1},
+     {  0,  0,  5,    2},  { 32,  0,  5,    4},
+     {  0,  0,  5,    5},  { 32,  0,  5,    7},
+     {  0,  0,  5,    8},  { 32,  0,  5,   10},
+     {  0,  0,  5,   11},  {  0,  0,  6,   13},
+     { 32,  1,  5,   16},  {  0,  1,  5,   18},
+     { 32,  1,  5,   22},  {  0,  2,  5,   24},
+     { 32,  3,  5,   32},  {  0,  3,  5,   40},
+     {  0,  6,  4,   64},  { 16,  6,  4,   64},
+     { 32,  7,  5,  128},  {  0,  9,  6,  512},
+     {  0, 11,  6, 2048},  { 48,  0,  4,    0},
+     { 16,  0,  4,    1},  { 32,  0,  5,    2},
+     { 32,  0,  5,    3},  { 32,  0,  5,    5},
+     { 32,  0,  5,    6},  { 32,  0,  5,    8},
+     { 32,  0,  5,    9},  { 32,  0,  5,   11},
+     { 32,  0,  5,   12},  {  0,  0,  6,   15},
+     { 32,  1,  5,   18},  { 32,  1,  5,   20},
+     { 32,  2,  5,   24},  { 32,  2,  5,   28},
+     { 32,  3,  5,   40},  { 32,  4,  5,   48},
+     {  0, 16,  6,65536},  {  0, 15,  6,32768},
+     {  0, 14,  6,16384},  {  0, 13,  6, 8192},
 };   /* LL_defaultDTable */
 
-static const FSE_decode_t4 ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = {
-    { { ML_DEFAULTNORMLOG, 1, 1 } }, /* header : tableLog, fastMode, fastMode */
-    { {  0,  0,  6 } },              /* 0 : base, symbol, bits */
-    { {  0,  1,  4 } },
-    { { 32,  2,  5 } },
-    { {  0,  3,  5 } },
-    { {  0,  5,  5 } },
-    { {  0,  6,  5 } },
-    { {  0,  8,  5 } },
-    { {  0, 10,  6 } },
-    { {  0, 13,  6 } },
-    { {  0, 16,  6 } },
-    { {  0, 19,  6 } },
-    { {  0, 22,  6 } },
-    { {  0, 25,  6 } },
-    { {  0, 28,  6 } },
-    { {  0, 31,  6 } },
-    { {  0, 33,  6 } },
-    { {  0, 35,  6 } },
-    { {  0, 37,  6 } },
-    { {  0, 39,  6 } },
-    { {  0, 41,  6 } },
-    { {  0, 43,  6 } },
-    { {  0, 45,  6 } },
-    { { 16,  1,  4 } },
-    { {  0,  2,  4 } },
-    { { 32,  3,  5 } },
-    { {  0,  4,  5 } },
-    { { 32,  6,  5 } },
-    { {  0,  7,  5 } },
-    { {  0,  9,  6 } },
-    { {  0, 12,  6 } },
-    { {  0, 15,  6 } },
-    { {  0, 18,  6 } },
-    { {  0, 21,  6 } },
-    { {  0, 24,  6 } },
-    { {  0, 27,  6 } },
-    { {  0, 30,  6 } },
-    { {  0, 32,  6 } },
-    { {  0, 34,  6 } },
-    { {  0, 36,  6 } },
-    { {  0, 38,  6 } },
-    { {  0, 40,  6 } },
-    { {  0, 42,  6 } },
-    { {  0, 44,  6 } },
-    { { 32,  1,  4 } },
-    { { 48,  1,  4 } },
-    { { 16,  2,  4 } },
-    { { 32,  4,  5 } },
-    { { 32,  5,  5 } },
-    { { 32,  7,  5 } },
-    { { 32,  8,  5 } },
-    { {  0, 11,  6 } },
-    { {  0, 14,  6 } },
-    { {  0, 17,  6 } },
-    { {  0, 20,  6 } },
-    { {  0, 23,  6 } },
-    { {  0, 26,  6 } },
-    { {  0, 29,  6 } },
-    { {  0, 52,  6 } },
-    { {  0, 51,  6 } },
-    { {  0, 50,  6 } },
-    { {  0, 49,  6 } },
-    { {  0, 48,  6 } },
-    { {  0, 47,  6 } },
-    { {  0, 46,  6 } },
+/* Default FSE distribution table for Offset Codes */
+static const ZSTD_seqSymbol OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, OF_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  5,    0},     {  0,  6,  4,   61},
+    {  0,  9,  5,  509},     {  0, 15,  5,32765},
+    {  0, 21,  5,2097149},   {  0,  3,  5,    5},
+    {  0,  7,  4,  125},     {  0, 12,  5, 4093},
+    {  0, 18,  5,262141},    {  0, 23,  5,8388605},
+    {  0,  5,  5,   29},     {  0,  8,  4,  253},
+    {  0, 14,  5,16381},     {  0, 20,  5,1048573},
+    {  0,  2,  5,    1},     { 16,  7,  4,  125},
+    {  0, 11,  5, 2045},     {  0, 17,  5,131069},
+    {  0, 22,  5,4194301},   {  0,  4,  5,   13},
+    { 16,  8,  4,  253},     {  0, 13,  5, 8189},
+    {  0, 19,  5,524285},    {  0,  1,  5,    1},
+    { 16,  6,  4,   61},     {  0, 10,  5, 1021},
+    {  0, 16,  5,65533},     {  0, 28,  5,268435453},
+    {  0, 27,  5,134217725}, {  0, 26,  5,67108861},
+    {  0, 25,  5,33554429},  {  0, 24,  5,16777213},
+};   /* OF_defaultDTable */
+
+
+/* Default FSE distribution table for Match Lengths */
+static const ZSTD_seqSymbol ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, ML_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  6,    3},  {  0,  0,  4,    4},
+    { 32,  0,  5,    5},  {  0,  0,  5,    6},
+    {  0,  0,  5,    8},  {  0,  0,  5,    9},
+    {  0,  0,  5,   11},  {  0,  0,  6,   13},
+    {  0,  0,  6,   16},  {  0,  0,  6,   19},
+    {  0,  0,  6,   22},  {  0,  0,  6,   25},
+    {  0,  0,  6,   28},  {  0,  0,  6,   31},
+    {  0,  0,  6,   34},  {  0,  1,  6,   37},
+    {  0,  1,  6,   41},  {  0,  2,  6,   47},
+    {  0,  3,  6,   59},  {  0,  4,  6,   83},
+    {  0,  7,  6,  131},  {  0,  9,  6,  515},
+    { 16,  0,  4,    4},  {  0,  0,  4,    5},
+    { 32,  0,  5,    6},  {  0,  0,  5,    7},
+    { 32,  0,  5,    9},  {  0,  0,  5,   10},
+    {  0,  0,  6,   12},  {  0,  0,  6,   15},
+    {  0,  0,  6,   18},  {  0,  0,  6,   21},
+    {  0,  0,  6,   24},  {  0,  0,  6,   27},
+    {  0,  0,  6,   30},  {  0,  0,  6,   33},
+    {  0,  1,  6,   35},  {  0,  1,  6,   39},
+    {  0,  2,  6,   43},  {  0,  3,  6,   51},
+    {  0,  4,  6,   67},  {  0,  5,  6,   99},
+    {  0,  8,  6,  259},  { 32,  0,  4,    4},
+    { 48,  0,  4,    4},  { 16,  0,  4,    5},
+    { 32,  0,  5,    7},  { 32,  0,  5,    8},
+    { 32,  0,  5,   10},  { 32,  0,  5,   11},
+    {  0,  0,  6,   14},  {  0,  0,  6,   17},
+    {  0,  0,  6,   20},  {  0,  0,  6,   23},
+    {  0,  0,  6,   26},  {  0,  0,  6,   29},
+    {  0,  0,  6,   32},  {  0, 16,  6,65539},
+    {  0, 15,  6,32771},  {  0, 14,  6,16387},
+    {  0, 13,  6, 8195},  {  0, 12,  6, 4099},
+    {  0, 11,  6, 2051},  {  0, 10,  6, 1027},
 };   /* ML_defaultDTable */
 
-static const FSE_decode_t4 OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = {
-    { { OF_DEFAULTNORMLOG, 1, 1 } }, /* header : tableLog, fastMode, fastMode */
-    { {  0,  0,  5 } },              /* 0 : base, symbol, bits */
-    { {  0,  6,  4 } },
-    { {  0,  9,  5 } },
-    { {  0, 15,  5 } },
-    { {  0, 21,  5 } },
-    { {  0,  3,  5 } },
-    { {  0,  7,  4 } },
-    { {  0, 12,  5 } },
-    { {  0, 18,  5 } },
-    { {  0, 23,  5 } },
-    { {  0,  5,  5 } },
-    { {  0,  8,  4 } },
-    { {  0, 14,  5 } },
-    { {  0, 20,  5 } },
-    { {  0,  2,  5 } },
-    { { 16,  7,  4 } },
-    { {  0, 11,  5 } },
-    { {  0, 17,  5 } },
-    { {  0, 22,  5 } },
-    { {  0,  4,  5 } },
-    { { 16,  8,  4 } },
-    { {  0, 13,  5 } },
-    { {  0, 19,  5 } },
-    { {  0,  1,  5 } },
-    { { 16,  6,  4 } },
-    { {  0, 10,  5 } },
-    { {  0, 16,  5 } },
-    { {  0, 28,  5 } },
-    { {  0, 27,  5 } },
-    { {  0, 26,  5 } },
-    { {  0, 25,  5 } },
-    { {  0, 24,  5 } },
-};   /* OF_defaultDTable */
+
+static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U32 nbAddBits)
+{
+    void* ptr = dt;
+    ZSTD_seqSymbol_header* const DTableH = (ZSTD_seqSymbol_header*)ptr;
+    ZSTD_seqSymbol* const cell = dt + 1;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->nbBits = 0;
+    cell->nextState = 0;
+    assert(nbAddBits < 255);
+    cell->nbAdditionalBits = (BYTE)nbAddBits;
+    cell->baseValue = baseValue;
+}
+
+
+/* ZSTD_buildFSETable() :
+ * generate FSE decoding table for one symbol (ll, ml or off) */
+static void
+ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
+    const short* normalizedCounter, unsigned maxSymbolValue,
+    const U32* baseValue, const U32* nbAdditionalBits,
+    unsigned tableLog)
+{
+    ZSTD_seqSymbol* const tableDecode = dt+1;
+    U16 symbolNext[MaxSeq+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    assert(maxSymbolValue <= MaxSeq);
+    assert(tableLog <= MaxFSELog);
+
+    /* Init, lay down lowprob symbols */
+    {   ZSTD_seqSymbol_header DTableH;
+        DTableH.tableLog = tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].baseValue = s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSE_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].baseValue = s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+        assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            U32 const symbol = tableDecode[u].baseValue;
+            U32 const nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
+            tableDecode[u].nextState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+            assert(nbAdditionalBits[symbol] < 255);
+            tableDecode[u].nbAdditionalBits = (BYTE)nbAdditionalBits[symbol];
+            tableDecode[u].baseValue = baseValue[symbol];
+    }   }
+}
+
 
 /*! ZSTD_buildSeqTable() :
-    @return : nb bytes read from src,
-              or an error code if it fails, testable with ZSTD_isError()
-*/
-static size_t ZSTD_buildSeqTable(FSE_DTable* DTableSpace, const FSE_DTable** DTablePtr,
+ * @return : nb bytes read from src,
+ *           or an error code if it fails */
+static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymbol** DTablePtr,
                                  symbolEncodingType_e type, U32 max, U32 maxLog,
                                  const void* src, size_t srcSize,
-                                 const FSE_decode_t4* defaultTable, U32 flagRepeatTable)
+                                 const U32* baseValue, const U32* nbAdditionalBits,
+                                 const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable)
 {
-    const void* const tmpPtr = defaultTable;   /* bypass strict aliasing */
     switch(type)
     {
     case set_rle :
         if (!srcSize) return ERROR(srcSize_wrong);
         if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
-        FSE_buildDTable_rle(DTableSpace, *(const BYTE*)src);
+        {   U32 const symbol = *(const BYTE*)src;
+            U32 const baseline = baseValue[symbol];
+            U32 const nbBits = nbAdditionalBits[symbol];
+            ZSTD_buildSeqTable_rle(DTableSpace, baseline, nbBits);
+        }
         *DTablePtr = DTableSpace;
         return 1;
     case set_basic :
-        *DTablePtr = (const FSE_DTable*)tmpPtr;
+        *DTablePtr = defaultTable;
         return 0;
     case set_repeat:
         if (!flagRepeatTable) return ERROR(corruption_detected);
         return 0;
-    default :   /* impossible */
     case set_compressed :
         {   U32 tableLog;
             S16 norm[MaxSeq+1];
             size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
             if (FSE_isError(headerSize)) return ERROR(corruption_detected);
             if (tableLog > maxLog) return ERROR(corruption_detected);
-            FSE_buildDTable(DTableSpace, norm, max, tableLog);
+            ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog);
             *DTablePtr = DTableSpace;
             return headerSize;
-    }   }
+        }
+    default :   /* impossible */
+        assert(0);
+        return ERROR(GENERIC);
+    }
 }
 
+static const U32 LL_base[MaxLL+1] = {
+                 0,    1,    2,     3,     4,     5,     6,      7,
+                 8,    9,   10,    11,    12,    13,    14,     15,
+                16,   18,   20,    22,    24,    28,    32,     40,
+                48,   64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                0x2000, 0x4000, 0x8000, 0x10000 };
+
+static const U32 OF_base[MaxOff+1] = {
+                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD, 0x1FFFFFFD, 0x3FFFFFFD, 0x7FFFFFFD };
+
+static const U32 OF_bits[MaxOff+1] = {
+                     0,  1,  2,  3,  4,  5,  6,  7,
+                     8,  9, 10, 11, 12, 13, 14, 15,
+                    16, 17, 18, 19, 20, 21, 22, 23,
+                    24, 25, 26, 27, 28, 29, 30, 31 };
+
+static const U32 ML_base[MaxML+1] = {
+                     3,  4,  5,    6,     7,     8,     9,    10,
+                    11, 12, 13,   14,    15,    16,    17,    18,
+                    19, 20, 21,   22,    23,    24,    25,    26,
+                    27, 28, 29,   30,    31,    32,    33,    34,
+                    35, 37, 39,   41,    43,    47,    51,    59,
+                    67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
+                    0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
+
+
 size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                              const void* src, size_t srcSize)
 {
     const BYTE* const istart = (const BYTE* const)src;
     const BYTE* const iend = istart + srcSize;
     const BYTE* ip = istart;
+    DEBUGLOG(5, "ZSTD_decodeSeqHeaders");
 
     /* check */
     if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
@@ -756,21 +968,29 @@
         ip++;
 
         /* Build DTables */
-        {   size_t const llhSize = ZSTD_buildSeqTable(dctx->LLTable, &dctx->LLTptr,
+        {   size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr,
                                                       LLtype, MaxLL, LLFSELog,
-                                                      ip, iend-ip, LL_defaultDTable, dctx->fseEntropy);
+                                                      ip, iend-ip,
+                                                      LL_base, LL_bits,
+                                                      LL_defaultDTable, dctx->fseEntropy);
             if (ZSTD_isError(llhSize)) return ERROR(corruption_detected);
             ip += llhSize;
         }
-        {   size_t const ofhSize = ZSTD_buildSeqTable(dctx->OFTable, &dctx->OFTptr,
+
+        {   size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr,
                                                       OFtype, MaxOff, OffFSELog,
-                                                      ip, iend-ip, OF_defaultDTable, dctx->fseEntropy);
+                                                      ip, iend-ip,
+                                                      OF_base, OF_bits,
+                                                      OF_defaultDTable, dctx->fseEntropy);
             if (ZSTD_isError(ofhSize)) return ERROR(corruption_detected);
             ip += ofhSize;
         }
-        {   size_t const mlhSize = ZSTD_buildSeqTable(dctx->MLTable, &dctx->MLTptr,
+
+        {   size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr,
                                                       MLtype, MaxML, MLFSELog,
-                                                      ip, iend-ip, ML_defaultDTable, dctx->fseEntropy);
+                                                      ip, iend-ip,
+                                                      ML_base, ML_bits,
+                                                      ML_defaultDTable, dctx->fseEntropy);
             if (ZSTD_isError(mlhSize)) return ERROR(corruption_detected);
             ip += mlhSize;
         }
@@ -788,14 +1008,19 @@
 } seq_t;
 
 typedef struct {
+    size_t state;
+    const ZSTD_seqSymbol* table;
+} ZSTD_fseState;
+
+typedef struct {
     BIT_DStream_t DStream;
-    FSE_DState_t stateLL;
-    FSE_DState_t stateOffb;
-    FSE_DState_t stateML;
+    ZSTD_fseState stateLL;
+    ZSTD_fseState stateOffb;
+    ZSTD_fseState stateML;
     size_t prevOffset[ZSTD_REP_NUM];
-    const BYTE* base;
+    const BYTE* prefixStart;
+    const BYTE* dictEnd;
     size_t pos;
-    iPtrDiff gotoDict;
 } seqState_t;
 
 
@@ -846,88 +1071,11 @@
 }
 
 
-
-
-static seq_t ZSTD_decodeSequence(seqState_t* seqState)
-{
-    seq_t seq;
-
-    U32 const llCode = FSE_peekSymbol(&seqState->stateLL);
-    U32 const mlCode = FSE_peekSymbol(&seqState->stateML);
-    U32 const ofCode = FSE_peekSymbol(&seqState->stateOffb);   /* <= maxOff, by table construction */
-
-    U32 const llBits = LL_bits[llCode];
-    U32 const mlBits = ML_bits[mlCode];
-    U32 const ofBits = ofCode;
-    U32 const totalBits = llBits+mlBits+ofBits;
-
-    static const U32 LL_base[MaxLL+1] = {
-                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9,   10,    11,    12,    13,    14,     15,
-                            16, 18, 20, 22, 24, 28, 32, 40, 48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
-                            0x2000, 0x4000, 0x8000, 0x10000 };
-
-    static const U32 ML_base[MaxML+1] = {
-                             3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13,   14,    15,    16,    17,    18,
-                            19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29,   30,    31,    32,    33,    34,
-                            35, 37, 39, 41, 43, 47, 51, 59, 67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
-                            0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
-
-    static const U32 OF_base[MaxOff+1] = {
-                             0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
-                             0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
-                             0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
-                             0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
-
-    /* sequence */
-    {   size_t offset;
-        if (!ofCode)
-            offset = 0;
-        else {
-            offset = OF_base[ofCode] + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
-            if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
-        }
-
-        if (ofCode <= 1) {
-            offset += (llCode==0);
-            if (offset) {
-                size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
-                temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
-                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
-                seqState->prevOffset[1] = seqState->prevOffset[0];
-                seqState->prevOffset[0] = offset = temp;
-            } else {
-                offset = seqState->prevOffset[0];
-            }
-        } else {
-            seqState->prevOffset[2] = seqState->prevOffset[1];
-            seqState->prevOffset[1] = seqState->prevOffset[0];
-            seqState->prevOffset[0] = offset;
-        }
-        seq.offset = offset;
-    }
-
-    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
-    if (MEM_32bits() && (mlBits+llBits>24)) BIT_reloadDStream(&seqState->DStream);
-
-    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
-    if (MEM_32bits() ||
-       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BIT_reloadDStream(&seqState->DStream);
-
-    /* ANS state update */
-    FSE_updateState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
-    FSE_updateState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
-    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
-    FSE_updateState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
-
-    return seq;
-}
-
-
-FORCE_INLINE
+HINT_INLINE
 size_t ZSTD_execSequence(BYTE* op,
-                                BYTE* const oend, seq_t sequence,
-                                const BYTE** litPtr, const BYTE* const litLimit,
-                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+                         BYTE* const oend, seq_t sequence,
+                         const BYTE** litPtr, const BYTE* const litLimit,
+                         const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
 {
     BYTE* const oLitEnd = op + sequence.litLength;
     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
@@ -950,9 +1098,10 @@
 
     /* copy Match */
     if (sequence.offset > (size_t)(oLitEnd - base)) {
-        /* offset beyond prefix */
-        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
-        match += (dictEnd-base);
+        /* offset beyond prefix -> go into extDict */
+        if (sequence.offset > (size_t)(oLitEnd - vBase))
+            return ERROR(corruption_detected);
+        match = dictEnd + (match - base);
         if (match + sequence.matchLength <= dictEnd) {
             memmove(oLitEnd, match, sequence.matchLength);
             return sequenceLength;
@@ -975,7 +1124,87 @@
     if (sequence.offset < 8) {
         /* close range match, overlap */
         static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
-        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* substracted */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTD_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTD_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_w) {
+            ZSTD_wildcopy(op, match, oend_w - op);
+            match += oend_w - op;
+            op = oend_w;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+HINT_INLINE
+size_t ZSTD_execSequenceLong(BYTE* op,
+                             BYTE* const oend, seq_t sequence,
+                             const BYTE** litPtr, const BYTE* const litLimit,
+                             const BYTE* const prefixStart, const BYTE* const dictStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = sequence.match;
+
+    /* check */
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    if (oLitEnd > oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, dictStart, dictEnd);
+
+    /* copy Literals */
+    ZSTD_copy8(op, *litPtr);  /* note : op <= oLitEnd <= oend_w == oend - 8 */
+    if (sequence.litLength > 8)
+        ZSTD_wildcopy(op+8, (*litPtr)+8, sequence.litLength - 8);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - dictStart)) return ERROR(corruption_detected);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = prefixStart;
+            if (op > oend_w || sequence.matchLength < MINMATCH) {
+              U32 i;
+              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
+              return sequenceLength;
+            }
+    }   }
+    assert(op <= oend_w);
+    assert(sequence.matchLength >= MINMATCH);
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
         int const sub2 = dec64table[sequence.offset];
         op[0] = match[0];
         op[1] = match[1];
@@ -1002,11 +1231,121 @@
     return sequenceLength;
 }
 
-
-static size_t ZSTD_decompressSequences(
-                               ZSTD_DCtx* dctx,
+static void
+ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt)
+{
+    const void* ptr = dt;
+    const ZSTD_seqSymbol_header* const DTableH = (const ZSTD_seqSymbol_header*)ptr;
+    DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
+    DEBUGLOG(6, "ZSTD_initFseState : val=%u using %u bits",
+                (U32)DStatePtr->state, DTableH->tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+FORCE_INLINE_TEMPLATE void
+ZSTD_updateFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD)
+{
+    ZSTD_seqSymbol const DInfo = DStatePtr->table[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.nextState + lowBits;
+}
+
+/* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
+ * offset bits. But we can only read at most (STREAM_ACCUMULATOR_MIN_32 - 1)
+ * bits before reloading. This value is the maximum number of bytes we read
+ * after reloading when we are decoding long offets.
+ */
+#define LONG_OFFSETS_MAX_EXTRA_BITS_32                       \
+    (ZSTD_WINDOWLOG_MAX_32 > STREAM_ACCUMULATOR_MIN_32       \
+        ? ZSTD_WINDOWLOG_MAX_32 - STREAM_ACCUMULATOR_MIN_32  \
+        : 0)
+
+typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e;
+
+FORCE_INLINE_TEMPLATE seq_t
+ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
+{
+    seq_t seq;
+    U32 const llBits = seqState->stateLL.table[seqState->stateLL.state].nbAdditionalBits;
+    U32 const mlBits = seqState->stateML.table[seqState->stateML.state].nbAdditionalBits;
+    U32 const ofBits = seqState->stateOffb.table[seqState->stateOffb.state].nbAdditionalBits;
+    U32 const totalBits = llBits+mlBits+ofBits;
+    U32 const llBase = seqState->stateLL.table[seqState->stateLL.state].baseValue;
+    U32 const mlBase = seqState->stateML.table[seqState->stateML.state].baseValue;
+    U32 const ofBase = seqState->stateOffb.table[seqState->stateOffb.state].baseValue;
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofBits)
+            offset = 0;
+        else {
+            ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
+            ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
+            assert(ofBits <= MaxOff);
+            if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
+                U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed);
+                offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
+                BIT_reloadDStream(&seqState->DStream);
+                if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
+                assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32);   /* to avoid another reload */
+            } else {
+                offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+                if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+            }
+        }
+
+        if (ofBits <= 1) {
+            offset += (llBase==0);
+            if (offset) {
+                size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
+                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+            } else {  /* offset == 0 */
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq.offset = offset;
+    }
+
+    seq.matchLength = mlBase
+                    + ((mlBits>0) ? BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/) : 0);  /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
+        BIT_reloadDStream(&seqState->DStream);
+    if (MEM_64bits() && (totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
+        BIT_reloadDStream(&seqState->DStream);
+    /* Ensure there are enough bits to read the rest of data in 64-bit mode. */
+    ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
+
+    seq.litLength = llBase
+                  + ((llBits>0) ? BIT_readBitsFast(&seqState->DStream, llBits/*>0*/) : 0);    /* <=  16 bits */
+    if (MEM_32bits())
+        BIT_reloadDStream(&seqState->DStream);
+
+    DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
+                (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+
+    /* ANS state update */
+    ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
+    ZSTD_updateFseState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
+    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+    ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
+
+    return seq;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
                                void* dst, size_t maxDstSize,
-                         const void* seqStart, size_t seqSize)
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset)
 {
     const BYTE* ip = (const BYTE*)seqStart;
     const BYTE* const iend = ip + seqSize;
@@ -1018,36 +1357,32 @@
     const BYTE* const base = (const BYTE*) (dctx->base);
     const BYTE* const vBase = (const BYTE*) (dctx->vBase);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    int nbSeq;
-
-    /* Build Decoding Tables */
-    {   size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, seqSize);
-        if (ZSTD_isError(seqHSize)) return seqHSize;
-        ip += seqHSize;
-    }
+    DEBUGLOG(5, "ZSTD_decompressSequences");
 
     /* Regen sequences */
     if (nbSeq) {
         seqState_t seqState;
         dctx->fseEntropy = 1;
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->rep[i]; }
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
         CHECK_E(BIT_initDStream(&seqState.DStream, ip, iend-ip), corruption_detected);
-        FSE_initDState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
-        FSE_initDState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
-        FSE_initDState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
 
         for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; ) {
             nbSeq--;
-            {   seq_t const sequence = ZSTD_decodeSequence(&seqState);
+            {   seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
                 size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+                DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
                 if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
                 op += oneSeqSize;
         }   }
 
         /* check if reached exact end */
+        DEBUGLOG(5, "ZSTD_decompressSequences: after decode loop, remaining nbSeq : %i", nbSeq);
         if (nbSeq) return ERROR(corruption_detected);
         /* save reps for next block */
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->rep[i] = (U32)(seqState.prevOffset[i]); }
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
     }
 
     /* last literal segment */
@@ -1060,48 +1395,50 @@
     return op-ostart;
 }
 
-
-static seq_t ZSTD_decodeSequenceLong(seqState_t* seqState)
+static size_t
+ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+
+
+FORCE_INLINE_TEMPLATE seq_t
+ZSTD_decodeSequenceLong(seqState_t* seqState, ZSTD_longOffset_e const longOffsets)
 {
     seq_t seq;
-
-    U32 const llCode = FSE_peekSymbol(&seqState->stateLL);
-    U32 const mlCode = FSE_peekSymbol(&seqState->stateML);
-    U32 const ofCode = FSE_peekSymbol(&seqState->stateOffb);   /* <= maxOff, by table construction */
-
-    U32 const llBits = LL_bits[llCode];
-    U32 const mlBits = ML_bits[mlCode];
-    U32 const ofBits = ofCode;
+    U32 const llBits = seqState->stateLL.table[seqState->stateLL.state].nbAdditionalBits;
+    U32 const mlBits = seqState->stateML.table[seqState->stateML.state].nbAdditionalBits;
+    U32 const ofBits = seqState->stateOffb.table[seqState->stateOffb.state].nbAdditionalBits;
     U32 const totalBits = llBits+mlBits+ofBits;
-
-    static const U32 LL_base[MaxLL+1] = {
-                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9,   10,    11,    12,    13,    14,     15,
-                            16, 18, 20, 22, 24, 28, 32, 40, 48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
-                            0x2000, 0x4000, 0x8000, 0x10000 };
-
-    static const U32 ML_base[MaxML+1] = {
-                             3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13,   14,    15,    16,    17,    18,
-                            19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29,   30,    31,    32,    33,    34,
-                            35, 37, 39, 41, 43, 47, 51, 59, 67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
-                            0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
-
-    static const U32 OF_base[MaxOff+1] = {
-                             0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
-                             0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
-                             0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
-                             0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
+    U32 const llBase = seqState->stateLL.table[seqState->stateLL.state].baseValue;
+    U32 const mlBase = seqState->stateML.table[seqState->stateML.state].baseValue;
+    U32 const ofBase = seqState->stateOffb.table[seqState->stateOffb.state].baseValue;
 
     /* sequence */
     {   size_t offset;
-        if (!ofCode)
+        if (!ofBits)
             offset = 0;
         else {
-            offset = OF_base[ofCode] + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
-            if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+            ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
+            ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
+            assert(ofBits <= MaxOff);
+            if (MEM_32bits() && longOffsets) {
+                U32 const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN_32-1);
+                offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
+                if (MEM_32bits() || extraBits) BIT_reloadDStream(&seqState->DStream);
+                if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
+            } else {
+                offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+                if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+            }
         }
 
-        if (ofCode <= 1) {
-            offset += (llCode==0);
+        if (ofBits <= 1) {
+            offset += (llBase==0);
             if (offset) {
                 size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
                 temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
@@ -1119,114 +1456,40 @@
         seq.offset = offset;
     }
 
-    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
-    if (MEM_32bits() && (mlBits+llBits>24)) BIT_reloadDStream(&seqState->DStream);
-
-    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
-    if (MEM_32bits() ||
-       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BIT_reloadDStream(&seqState->DStream);
+    seq.matchLength = mlBase + ((mlBits>0) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
+        BIT_reloadDStream(&seqState->DStream);
+    if (MEM_64bits() && (totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
+        BIT_reloadDStream(&seqState->DStream);
+    /* Verify that there is enough bits to read the rest of the data in 64-bit mode. */
+    ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
+
+    seq.litLength = llBase + ((llBits>0) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
+    if (MEM_32bits())
+        BIT_reloadDStream(&seqState->DStream);
 
     {   size_t const pos = seqState->pos + seq.litLength;
-        seq.match = seqState->base + pos - seq.offset;    /* single memory segment */
-        if (seq.offset > pos) seq.match += seqState->gotoDict;   /* separate memory segment */
+        const BYTE* const matchBase = (seq.offset > pos) ? seqState->dictEnd : seqState->prefixStart;
+        seq.match = matchBase + pos - seq.offset;  /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
+                                                    * No consequence though : no memory access will occur, overly large offset will be detected in ZSTD_execSequenceLong() */
         seqState->pos = pos + seq.matchLength;
     }
 
     /* ANS state update */
-    FSE_updateState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
-    FSE_updateState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
+    ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
+    ZSTD_updateFseState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
     if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
-    FSE_updateState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
+    ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
 
     return seq;
 }
 
-FORCE_INLINE
-size_t ZSTD_execSequenceLong(BYTE* op,
-                                BYTE* const oend, seq_t sequence,
-                                const BYTE** litPtr, const BYTE* const litLimit,
-                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
-{
-    BYTE* const oLitEnd = op + sequence.litLength;
-    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
-    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
-    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
-    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
-    const BYTE* match = sequence.match;
-
-    /* check */
-#if 1
-    if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
-    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
-    if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, base, vBase, dictEnd);
-#endif
-
-    /* copy Literals */
-    ZSTD_copy8(op, *litPtr);
-    if (sequence.litLength > 8)
-        ZSTD_wildcopy(op+8, (*litPtr)+8, sequence.litLength - 8);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
-    op = oLitEnd;
-    *litPtr = iLitEnd;   /* update for next sequence */
-
-    /* copy Match */
-#if 1
-    if (sequence.offset > (size_t)(oLitEnd - base)) {
-        /* offset beyond prefix */
-        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
-        if (match + sequence.matchLength <= dictEnd) {
-            memmove(oLitEnd, match, sequence.matchLength);
-            return sequenceLength;
-        }
-        /* span extDict & currentPrefixSegment */
-        {   size_t const length1 = dictEnd - match;
-            memmove(oLitEnd, match, length1);
-            op = oLitEnd + length1;
-            sequence.matchLength -= length1;
-            match = base;
-            if (op > oend_w || sequence.matchLength < MINMATCH) {
-              U32 i;
-              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
-              return sequenceLength;
-            }
-    }   }
-    /* Requirement: op <= oend_w && sequence.matchLength >= MINMATCH */
-#endif
-
-    /* match within prefix */
-    if (sequence.offset < 8) {
-        /* close range match, overlap */
-        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
-        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* substracted */
-        int const sub2 = dec64table[sequence.offset];
-        op[0] = match[0];
-        op[1] = match[1];
-        op[2] = match[2];
-        op[3] = match[3];
-        match += dec32table[sequence.offset];
-        ZSTD_copy4(op+4, match);
-        match -= sub2;
-    } else {
-        ZSTD_copy8(op, match);
-    }
-    op += 8; match += 8;
-
-    if (oMatchEnd > oend-(16-MINMATCH)) {
-        if (op < oend_w) {
-            ZSTD_wildcopy(op, match, oend_w - op);
-            match += oend_w - op;
-            op = oend_w;
-        }
-        while (op < oMatchEnd) *op++ = *match++;
-    } else {
-        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
-    }
-    return sequenceLength;
-}
-
-static size_t ZSTD_decompressSequencesLong(
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_decompressSequencesLong_body(
                                ZSTD_DCtx* dctx,
                                void* dst, size_t maxDstSize,
-                         const void* seqStart, size_t seqSize)
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset)
 {
     const BYTE* ip = (const BYTE*)seqStart;
     const BYTE* const iend = ip + seqSize;
@@ -1235,16 +1498,9 @@
     BYTE* op = ostart;
     const BYTE* litPtr = dctx->litPtr;
     const BYTE* const litEnd = litPtr + dctx->litSize;
-    const BYTE* const base = (const BYTE*) (dctx->base);
-    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const prefixStart = (const BYTE*) (dctx->base);
+    const BYTE* const dictStart = (const BYTE*) (dctx->vBase);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    int nbSeq;
-
-    /* Build Decoding Tables */
-    {   size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, seqSize);
-        if (ZSTD_isError(seqHSize)) return seqHSize;
-        ip += seqHSize;
-    }
 
     /* Regen sequences */
     if (nbSeq) {
@@ -1256,27 +1512,27 @@
         seqState_t seqState;
         int seqNb;
         dctx->fseEntropy = 1;
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->rep[i]; }
-        seqState.base = base;
-        seqState.pos = (size_t)(op-base);
-        seqState.gotoDict = (iPtrDiff)(dictEnd - base);
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        seqState.prefixStart = prefixStart;
+        seqState.pos = (size_t)(op-prefixStart);
+        seqState.dictEnd = dictEnd;
         CHECK_E(BIT_initDStream(&seqState.DStream, ip, iend-ip), corruption_detected);
-        FSE_initDState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
-        FSE_initDState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
-        FSE_initDState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
 
         /* prepare in advance */
-        for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && seqNb<seqAdvance; seqNb++) {
-            sequences[seqNb] = ZSTD_decodeSequenceLong(&seqState);
+        for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && (seqNb<seqAdvance); seqNb++) {
+            sequences[seqNb] = ZSTD_decodeSequenceLong(&seqState, isLongOffset);
         }
         if (seqNb<seqAdvance) return ERROR(corruption_detected);
 
         /* decode and decompress */
-        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && seqNb<nbSeq ; seqNb++) {
-            seq_t const sequence = ZSTD_decodeSequenceLong(&seqState);
-            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STOSEQ_MASK], &litPtr, litEnd, base, vBase, dictEnd);
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb<nbSeq) ; seqNb++) {
+            seq_t const sequence = ZSTD_decodeSequenceLong(&seqState, isLongOffset);
+            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STOSEQ_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
             if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
-            ZSTD_PREFETCH(sequence.match);
+            PREFETCH(sequence.match);  /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
             sequences[seqNb&STOSEQ_MASK] = sequence;
             op += oneSeqSize;
         }
@@ -1285,13 +1541,16 @@
         /* finish queue */
         seqNb -= seqAdvance;
         for ( ; seqNb<nbSeq ; seqNb++) {
-            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[seqNb&STOSEQ_MASK], &litPtr, litEnd, base, vBase, dictEnd);
+            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[seqNb&STOSEQ_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
             if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
             op += oneSeqSize;
         }
 
         /* save reps for next block */
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->rep[i] = (U32)(seqState.prevOffset[i]); }
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+#undef STORED_SEQS
+#undef STOSEQ_MASK
+#undef ADVANCED_SEQS
     }
 
     /* last literal segment */
@@ -1304,23 +1563,137 @@
     return op-ostart;
 }
 
+static size_t
+ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+
+
+#if DYNAMIC_BMI2
+
+static TARGET_ATTRIBUTE("bmi2") size_t
+ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+static TARGET_ATTRIBUTE("bmi2") size_t
+ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+#endif
+
+typedef size_t (*ZSTD_decompressSequences_t)(
+    ZSTD_DCtx *dctx, void *dst, size_t maxDstSize,
+    const void *seqStart, size_t seqSize, int nbSeq,
+    const ZSTD_longOffset_e isLongOffset);
+
+static size_t ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
+                                const void* seqStart, size_t seqSize, int nbSeq,
+                                const ZSTD_longOffset_e isLongOffset)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequences");
+#if DYNAMIC_BMI2
+    if (dctx->bmi2) {
+        return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+    }
+#endif
+  return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+static size_t ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
+                                void* dst, size_t maxDstSize,
+                                const void* seqStart, size_t seqSize, int nbSeq,
+                                const ZSTD_longOffset_e isLongOffset)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequencesLong");
+#if DYNAMIC_BMI2
+    if (dctx->bmi2) {
+        return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+    }
+#endif
+  return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+/* ZSTD_getLongOffsetsShare() :
+ * condition : offTable must be valid
+ * @return : "share" of long offsets (arbitrarily defined as > (1<<23))
+ *           compared to maximum possible of (1<<OffFSELog) */
+static unsigned
+ZSTD_getLongOffsetsShare(const ZSTD_seqSymbol* offTable)
+{
+    const void* ptr = offTable;
+    U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
+    const ZSTD_seqSymbol* table = offTable + 1;
+    U32 const max = 1 << tableLog;
+    U32 u, total = 0;
+    DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
+
+    assert(max <= (1 << OffFSELog));  /* max not too large */
+    for (u=0; u<max; u++) {
+        if (table[u].nbAdditionalBits > 22) total += 1;
+    }
+
+    assert(tableLog <= OffFSELog);
+    total <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
+
+    return total;
+}
+
 
 static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
                             void* dst, size_t dstCapacity,
-                      const void* src, size_t srcSize)
+                      const void* src, size_t srcSize, const int frame)
 {   /* blockType == blockCompressed */
     const BYTE* ip = (const BYTE*)src;
-
-    if (srcSize >= ZSTD_BLOCKSIZE_ABSOLUTEMAX) return ERROR(srcSize_wrong);
-
-    /* Decode literals sub-block */
+    /* isLongOffset must be true if there are long offsets.
+     * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
+     * We don't expect that to be the case in 64-bit mode.
+     * In block mode, window size is not known, so we have to be conservative. (note: but it could be evaluated from current-lowLimit)
+     */
+    ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN)));
+    DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
+
+    if (srcSize >= ZSTD_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);
+
+    /* Decode literals section */
     {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
+        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize);
         if (ZSTD_isError(litCSize)) return litCSize;
         ip += litCSize;
         srcSize -= litCSize;
     }
-    if (dctx->fParams.windowSize > (1<<23)) return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize);
-    return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+
+    /* Build Decoding Tables */
+    {   int nbSeq;
+        size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
+        if (ZSTD_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+        srcSize -= seqHSize;
+
+        if ( (!frame || dctx->fParams.windowSize > (1<<24))
+          && (nbSeq>0) ) {  /* could probably use a larger nbSeq limit */
+            U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr);
+            U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
+            if (shareLongOffsets >= minShare)
+                return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
+        }
+
+        return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
+    }
 }
 
 
@@ -1340,7 +1713,7 @@
 {
     size_t dSize;
     ZSTD_checkContinuity(dctx, dst);
-    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0);
     dctx->previousDstEnd = (char*)dst + dSize;
     return dSize;
 }
@@ -1356,34 +1729,89 @@
 }
 
 
-size_t ZSTD_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
+static size_t ZSTD_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
 {
     if (length > dstCapacity) return ERROR(dstSize_tooSmall);
     memset(dst, byte, length);
     return length;
 }
 
+/** ZSTD_findFrameCompressedSize() :
+ *  compatible with legacy mode
+ *  `src` must point to the start of a ZSTD frame, ZSTD legacy frame, or skippable frame
+ *  `srcSize` must be at least as large as the frame contained
+ *  @return : the compressed size of the frame starting at `src` */
+size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
+{
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize))
+        return ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
+#endif
+    if ( (srcSize >= ZSTD_skippableHeaderSize)
+      && (MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START ) {
+        return ZSTD_skippableHeaderSize + MEM_readLE32((const BYTE*)src + ZSTD_frameIdSize);
+    } else {
+        const BYTE* ip = (const BYTE*)src;
+        const BYTE* const ipstart = ip;
+        size_t remainingSize = srcSize;
+        ZSTD_frameHeader zfh;
+
+        /* Extract Frame Header */
+        {   size_t const ret = ZSTD_getFrameHeader(&zfh, src, srcSize);
+            if (ZSTD_isError(ret)) return ret;
+            if (ret > 0) return ERROR(srcSize_wrong);
+        }
+
+        ip += zfh.headerSize;
+        remainingSize -= zfh.headerSize;
+
+        /* Loop on each block */
+        while (1) {
+            blockProperties_t blockProperties;
+            size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+            if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+            if (ZSTD_blockHeaderSize + cBlockSize > remainingSize)
+                return ERROR(srcSize_wrong);
+
+            ip += ZSTD_blockHeaderSize + cBlockSize;
+            remainingSize -= ZSTD_blockHeaderSize + cBlockSize;
+
+            if (blockProperties.lastBlock) break;
+        }
+
+        if (zfh.checksumFlag) {   /* Final frame content checksum */
+            if (remainingSize < 4) return ERROR(srcSize_wrong);
+            ip += 4;
+            remainingSize -= 4;
+        }
+
+        return ip - ipstart;
+    }
+}
 
 /*! ZSTD_decompressFrame() :
-*   `dctx` must be properly initialized */
+*   @dctx must be properly initialized */
 static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
-                                 void* dst, size_t dstCapacity,
-                                 const void* src, size_t srcSize)
+                                   void* dst, size_t dstCapacity,
+                             const void** srcPtr, size_t *srcSizePtr)
 {
-    const BYTE* ip = (const BYTE*)src;
+    const BYTE* ip = (const BYTE*)(*srcPtr);
     BYTE* const ostart = (BYTE* const)dst;
     BYTE* const oend = ostart + dstCapacity;
     BYTE* op = ostart;
-    size_t remainingSize = srcSize;
+    size_t remainingSize = *srcSizePtr;
 
     /* check */
-    if (srcSize < ZSTD_frameHeaderSize_min+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    if (remainingSize < ZSTD_frameHeaderSize_min+ZSTD_blockHeaderSize)
+        return ERROR(srcSize_wrong);
 
     /* Frame Header */
-    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize(src, ZSTD_frameHeaderSize_prefix);
+    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize(ip, ZSTD_frameHeaderSize_prefix);
         if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
-        if (srcSize < frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
-        CHECK_F(ZSTD_decodeFrameHeader(dctx, src, frameHeaderSize));
+        if (remainingSize < frameHeaderSize+ZSTD_blockHeaderSize)
+            return ERROR(srcSize_wrong);
+        CHECK_F( ZSTD_decodeFrameHeader(dctx, ip, frameHeaderSize) );
         ip += frameHeaderSize; remainingSize -= frameHeaderSize;
     }
 
@@ -1401,7 +1829,7 @@
         switch(blockProperties.blockType)
         {
         case bt_compressed:
-            decodedSize = ZSTD_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            decodedSize = ZSTD_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize, /* frame */ 1);
             break;
         case bt_raw :
             decodedSize = ZSTD_copyRawBlock(op, oend-op, ip, cBlockSize);
@@ -1415,38 +1843,123 @@
         }
 
         if (ZSTD_isError(decodedSize)) return decodedSize;
-        if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, op, decodedSize);
+        if (dctx->fParams.checksumFlag)
+            XXH64_update(&dctx->xxhState, op, decodedSize);
         op += decodedSize;
         ip += cBlockSize;
         remainingSize -= cBlockSize;
         if (blockProperties.lastBlock) break;
     }
 
-    if (dctx->fParams.checksumFlag) {   /* Frame content checksum verification */
+    if (dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+        if ((U64)(op-ostart) != dctx->fParams.frameContentSize) {
+            return ERROR(corruption_detected);
+    }   }
+    if (dctx->fParams.checksumFlag) { /* Frame content checksum verification */
         U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
         U32 checkRead;
         if (remainingSize<4) return ERROR(checksum_wrong);
         checkRead = MEM_readLE32(ip);
         if (checkRead != checkCalc) return ERROR(checksum_wrong);
+        ip += 4;
         remainingSize -= 4;
     }
 
-    if (remainingSize) return ERROR(srcSize_wrong);
+    /* Allow caller to get size read */
+    *srcPtr = ip;
+    *srcSizePtr = remainingSize;
     return op-ostart;
 }
 
+static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict);
+static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict);
+
+static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
+                                        void* dst, size_t dstCapacity,
+                                  const void* src, size_t srcSize,
+                                  const void* dict, size_t dictSize,
+                                  const ZSTD_DDict* ddict)
+{
+    void* const dststart = dst;
+    assert(dict==NULL || ddict==NULL);  /* either dict or ddict set, not both */
+
+    if (ddict) {
+        dict = ZSTD_DDictDictContent(ddict);
+        dictSize = ZSTD_DDictDictSize(ddict);
+    }
+
+    while (srcSize >= ZSTD_frameHeaderSize_prefix) {
+        U32 magicNumber;
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+        if (ZSTD_isLegacy(src, srcSize)) {
+            size_t decodedSize;
+            size_t const frameSize = ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
+            if (ZSTD_isError(frameSize)) return frameSize;
+            /* legacy support is not compatible with static dctx */
+            if (dctx->staticSize) return ERROR(memory_allocation);
+
+            decodedSize = ZSTD_decompressLegacy(dst, dstCapacity, src, frameSize, dict, dictSize);
+
+            dst = (BYTE*)dst + decodedSize;
+            dstCapacity -= decodedSize;
+
+            src = (const BYTE*)src + frameSize;
+            srcSize -= frameSize;
+
+            continue;
+        }
+#endif
+
+        magicNumber = MEM_readLE32(src);
+        DEBUGLOG(4, "reading magic number %08X (expecting %08X)",
+                    (U32)magicNumber, (U32)ZSTD_MAGICNUMBER);
+        if (magicNumber != ZSTD_MAGICNUMBER) {
+            if ((magicNumber & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+                size_t skippableSize;
+                if (srcSize < ZSTD_skippableHeaderSize)
+                    return ERROR(srcSize_wrong);
+                skippableSize = MEM_readLE32((const BYTE*)src + ZSTD_frameIdSize)
+                              + ZSTD_skippableHeaderSize;
+                if (srcSize < skippableSize) return ERROR(srcSize_wrong);
+
+                src = (const BYTE *)src + skippableSize;
+                srcSize -= skippableSize;
+                continue;
+            }
+            return ERROR(prefix_unknown);
+        }
+
+        if (ddict) {
+            /* we were called from ZSTD_decompress_usingDDict */
+            CHECK_F(ZSTD_decompressBegin_usingDDict(dctx, ddict));
+        } else {
+            /* this will initialize correctly with no dict if dict == NULL, so
+             * use this in all cases but ddict */
+            CHECK_F(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize));
+        }
+        ZSTD_checkContinuity(dctx, dst);
+
+        {   const size_t res = ZSTD_decompressFrame(dctx, dst, dstCapacity,
+                                                    &src, &srcSize);
+            if (ZSTD_isError(res)) return res;
+            /* no need to bound check, ZSTD_decompressFrame already has */
+            dst = (BYTE*)dst + res;
+            dstCapacity -= res;
+        }
+    }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
+
+    if (srcSize) return ERROR(srcSize_wrong); /* input not entirely consumed */
+
+    return (BYTE*)dst - (BYTE*)dststart;
+}
 
 size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
                                  void* dst, size_t dstCapacity,
                            const void* src, size_t srcSize,
                            const void* dict, size_t dictSize)
 {
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT==1)
-    if (ZSTD_isLegacy(src, srcSize)) return ZSTD_decompressLegacy(dst, dstCapacity, src, srcSize, dict, dictSize);
-#endif
-    CHECK_F(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize));
-    ZSTD_checkContinuity(dctx, dst);
-    return ZSTD_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize, dict, dictSize, NULL);
 }
 
 
@@ -1458,7 +1971,7 @@
 
 size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
-#if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE==1)
+#if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE>=1)
     size_t regenSize;
     ZSTD_DCtx* const dctx = ZSTD_createDCtx();
     if (dctx==NULL) return ERROR(memory_allocation);
@@ -1482,6 +1995,7 @@
     switch(dctx->stage)
     {
     default:   /* should not happen */
+        assert(0);
     case ZSTDds_getFrameHeaderSize:
     case ZSTDds_decodeFrameHeader:
         return ZSTDnit_frameHeader;
@@ -1499,39 +2013,41 @@
     }
 }
 
-int ZSTD_isSkipFrame(ZSTD_DCtx* dctx) { return dctx->stage == ZSTDds_skipFrame; }   /* for zbuff */
+static int ZSTD_isSkipFrame(ZSTD_DCtx* dctx) { return dctx->stage == ZSTDds_skipFrame; }
 
 /** ZSTD_decompressContinue() :
-*   @return : nb of bytes generated into `dst` (necessarily <= `dstCapacity)
-*             or an error code, which can be tested using ZSTD_isError() */
+ *  srcSize : must be the exact nb of bytes expected (see ZSTD_nextSrcSizeToDecompress())
+ *  @return : nb of bytes generated into `dst` (necessarily <= `dstCapacity)
+ *            or an error code, which can be tested using ZSTD_isError() */
 size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
+    DEBUGLOG(5, "ZSTD_decompressContinue (srcSize:%u)", (U32)srcSize);
     /* Sanity check */
-    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);  /* not allowed */
     if (dstCapacity) ZSTD_checkContinuity(dctx, dst);
 
     switch (dctx->stage)
     {
     case ZSTDds_getFrameHeaderSize :
-        if (srcSize != ZSTD_frameHeaderSize_prefix) return ERROR(srcSize_wrong);      /* impossible */
-        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
-            memcpy(dctx->headerBuffer, src, ZSTD_frameHeaderSize_prefix);
-            dctx->expected = ZSTD_skippableHeaderSize - ZSTD_frameHeaderSize_prefix;  /* magic number + skippable frame length */
-            dctx->stage = ZSTDds_decodeSkippableHeader;
-            return 0;
-        }
-        dctx->headerSize = ZSTD_frameHeaderSize(src, ZSTD_frameHeaderSize_prefix);
+        assert(src != NULL);
+        if (dctx->format == ZSTD_f_zstd1) {  /* allows header */
+            assert(srcSize >= ZSTD_frameIdSize);  /* to read skippable magic number */
+            if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
+                memcpy(dctx->headerBuffer, src, srcSize);
+                dctx->expected = ZSTD_skippableHeaderSize - srcSize;  /* remaining to load to get full skippable frame header */
+                dctx->stage = ZSTDds_decodeSkippableHeader;
+                return 0;
+        }   }
+        dctx->headerSize = ZSTD_frameHeaderSize_internal(src, srcSize, dctx->format);
         if (ZSTD_isError(dctx->headerSize)) return dctx->headerSize;
-        memcpy(dctx->headerBuffer, src, ZSTD_frameHeaderSize_prefix);
-        if (dctx->headerSize > ZSTD_frameHeaderSize_prefix) {
-            dctx->expected = dctx->headerSize - ZSTD_frameHeaderSize_prefix;
-            dctx->stage = ZSTDds_decodeFrameHeader;
-            return 0;
-        }
-        dctx->expected = 0;   /* not necessary to copy more */
+        memcpy(dctx->headerBuffer, src, srcSize);
+        dctx->expected = dctx->headerSize - srcSize;
+        dctx->stage = ZSTDds_decodeFrameHeader;
+        return 0;
 
     case ZSTDds_decodeFrameHeader:
-        memcpy(dctx->headerBuffer + ZSTD_frameHeaderSize_prefix, src, dctx->expected);
+        assert(src != NULL);
+        memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
         CHECK_F(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize));
         dctx->expected = ZSTD_blockHeaderSize;
         dctx->stage = ZSTDds_decodeBlockHeader;
@@ -1558,18 +2074,21 @@
                     dctx->stage = ZSTDds_getFrameHeaderSize;
                 }
             } else {
-                dctx->expected = 3;  /* go directly to next header */
+                dctx->expected = ZSTD_blockHeaderSize;  /* jump to next header */
                 dctx->stage = ZSTDds_decodeBlockHeader;
             }
             return 0;
         }
+
     case ZSTDds_decompressLastBlock:
     case ZSTDds_decompressBlock:
+        DEBUGLOG(5, "ZSTD_decompressContinue: case ZSTDds_decompressBlock");
         {   size_t rSize;
             switch(dctx->bType)
             {
             case bt_compressed:
-                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+                DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
+                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1);
                 break;
             case bt_raw :
                 rSize = ZSTD_copyRawBlock(dst, dstCapacity, src, srcSize);
@@ -1582,9 +2101,16 @@
                 return ERROR(corruption_detected);
             }
             if (ZSTD_isError(rSize)) return rSize;
+            DEBUGLOG(5, "ZSTD_decompressContinue: decoded size from block : %u", (U32)rSize);
+            dctx->decodedSize += rSize;
             if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, dst, rSize);
 
             if (dctx->stage == ZSTDds_decompressLastBlock) {   /* end of frame */
+                DEBUGLOG(4, "ZSTD_decompressContinue: decoded size from frame : %u", (U32)dctx->decodedSize);
+                if (dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+                    if (dctx->decodedSize != dctx->fParams.frameContentSize) {
+                        return ERROR(corruption_detected);
+                }   }
                 if (dctx->fParams.checksumFlag) {  /* another round for frame checksum */
                     dctx->expected = 4;
                     dctx->stage = ZSTDds_checkChecksum;
@@ -1599,25 +2125,31 @@
             }
             return rSize;
         }
+
     case ZSTDds_checkChecksum:
+        assert(srcSize == 4);  /* guaranteed by dctx->expected */
         {   U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
-            U32 const check32 = MEM_readLE32(src);   /* srcSize == 4, guaranteed by dctx->expected */
+            U32 const check32 = MEM_readLE32(src);
+            DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", h32, check32);
             if (check32 != h32) return ERROR(checksum_wrong);
             dctx->expected = 0;
             dctx->stage = ZSTDds_getFrameHeaderSize;
             return 0;
         }
+
     case ZSTDds_decodeSkippableHeader:
-        {   memcpy(dctx->headerBuffer + ZSTD_frameHeaderSize_prefix, src, dctx->expected);
-            dctx->expected = MEM_readLE32(dctx->headerBuffer + 4);
-            dctx->stage = ZSTDds_skipFrame;
-            return 0;
-        }
+        assert(src != NULL);
+        assert(srcSize <= ZSTD_skippableHeaderSize);
+        memcpy(dctx->headerBuffer + (ZSTD_skippableHeaderSize - srcSize), src, srcSize);   /* complete skippable header */
+        dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_frameIdSize);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
+        dctx->stage = ZSTDds_skipFrame;
+        return 0;
+
     case ZSTDds_skipFrame:
-        {   dctx->expected = 0;
-            dctx->stage = ZSTDds_getFrameHeaderSize;
-            return 0;
-        }
+        dctx->expected = 0;
+        dctx->stage = ZSTDds_getFrameHeaderSize;
+        return 0;
+
     default:
         return ERROR(GENERIC);   /* impossible */
     }
@@ -1633,22 +2165,35 @@
     return 0;
 }
 
-static size_t ZSTD_loadEntropy(ZSTD_DCtx* dctx, const void* const dict, size_t const dictSize)
+/* ZSTD_loadEntropy() :
+ * dict : must point at beginning of a valid zstd dictionary
+ * @return : size of entropy tables read */
+static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const dict, size_t const dictSize)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
 
-    {   size_t const hSize = HUF_readDTableX4(dctx->hufTable, dict, dictSize);
+    if (dictSize <= 8) return ERROR(dictionary_corrupted);
+    dictPtr += 8;   /* skip header = magic + dictID */
+
+
+    {   size_t const hSize = HUF_readDTableX4_wksp(
+            entropy->hufTable, dictPtr, dictEnd - dictPtr,
+            entropy->workspace, sizeof(entropy->workspace));
         if (HUF_isError(hSize)) return ERROR(dictionary_corrupted);
         dictPtr += hSize;
     }
 
     {   short offcodeNCount[MaxOff+1];
-        U32 offcodeMaxValue=MaxOff, offcodeLog;
+        U32 offcodeMaxValue = MaxOff, offcodeLog;
         size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
         if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeMaxValue > MaxOff) return ERROR(dictionary_corrupted);
         if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
-        CHECK_E(FSE_buildDTable(dctx->OFTable, offcodeNCount, offcodeMaxValue, offcodeLog), dictionary_corrupted);
+        ZSTD_buildFSETable(entropy->OFTable,
+                            offcodeNCount, offcodeMaxValue,
+                            OF_base, OF_bits,
+                            offcodeLog);
         dictPtr += offcodeHeaderSize;
     }
 
@@ -1656,8 +2201,12 @@
         unsigned matchlengthMaxValue = MaxML, matchlengthLog;
         size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
         if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthMaxValue > MaxML) return ERROR(dictionary_corrupted);
         if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
-        CHECK_E(FSE_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog), dictionary_corrupted);
+        ZSTD_buildFSETable(entropy->MLTable,
+                            matchlengthNCount, matchlengthMaxValue,
+                            ML_base, ML_bits,
+                            matchlengthLog);
         dictPtr += matchlengthHeaderSize;
     }
 
@@ -1665,18 +2214,24 @@
         unsigned litlengthMaxValue = MaxLL, litlengthLog;
         size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
         if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthMaxValue > MaxLL) return ERROR(dictionary_corrupted);
         if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
-        CHECK_E(FSE_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog), dictionary_corrupted);
+        ZSTD_buildFSETable(entropy->LLTable,
+                            litlengthNCount, litlengthMaxValue,
+                            LL_base, LL_bits,
+                            litlengthLog);
         dictPtr += litlengthHeaderSize;
     }
 
     if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
-    dctx->rep[0] = MEM_readLE32(dictPtr+0); if (dctx->rep[0] == 0 || dctx->rep[0] >= dictSize) return ERROR(dictionary_corrupted);
-    dctx->rep[1] = MEM_readLE32(dictPtr+4); if (dctx->rep[1] == 0 || dctx->rep[1] >= dictSize) return ERROR(dictionary_corrupted);
-    dctx->rep[2] = MEM_readLE32(dictPtr+8); if (dctx->rep[2] == 0 || dctx->rep[2] >= dictSize) return ERROR(dictionary_corrupted);
-    dictPtr += 12;
-
-    dctx->litEntropy = dctx->fseEntropy = 1;
+    {   int i;
+        size_t const dictContentSize = (size_t)(dictEnd - (dictPtr+12));
+        for (i=0; i<3; i++) {
+            U32 const rep = MEM_readLE32(dictPtr); dictPtr += 4;
+            if (rep==0 || rep >= dictContentSize) return ERROR(dictionary_corrupted);
+            entropy->rep[i] = rep;
+    }   }
+
     return dictPtr - (const BYTE*)dict;
 }
 
@@ -1684,28 +2239,51 @@
 {
     if (dictSize < 8) return ZSTD_refDictContent(dctx, dict, dictSize);
     {   U32 const magic = MEM_readLE32(dict);
-        if (magic != ZSTD_DICT_MAGIC) {
+        if (magic != ZSTD_MAGIC_DICTIONARY) {
             return ZSTD_refDictContent(dctx, dict, dictSize);   /* pure content mode */
     }   }
-    dctx->dictID = MEM_readLE32((const char*)dict + 4);
+    dctx->dictID = MEM_readLE32((const char*)dict + ZSTD_frameIdSize);
 
     /* load entropy tables */
-    dict = (const char*)dict + 8;
-    dictSize -= 8;
-    {   size_t const eSize = ZSTD_loadEntropy(dctx, dict, dictSize);
+    {   size_t const eSize = ZSTD_loadEntropy(&dctx->entropy, dict, dictSize);
         if (ZSTD_isError(eSize)) return ERROR(dictionary_corrupted);
         dict = (const char*)dict + eSize;
         dictSize -= eSize;
     }
+    dctx->litEntropy = dctx->fseEntropy = 1;
 
     /* reference dictionary content */
     return ZSTD_refDictContent(dctx, dict, dictSize);
 }
 
+/* Note : this function cannot fail */
+size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
+{
+    assert(dctx != NULL);
+    dctx->expected = ZSTD_startingInputLength(dctx->format);  /* dctx->format must be properly set */
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->decodedSize = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+    dctx->litEntropy = dctx->fseEntropy = 0;
+    dctx->dictID = 0;
+    ZSTD_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
+    memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
+    dctx->LLTptr = dctx->entropy.LLTable;
+    dctx->MLTptr = dctx->entropy.MLTable;
+    dctx->OFTptr = dctx->entropy.OFTable;
+    dctx->HUFptr = dctx->entropy.hufTable;
+    return 0;
+}
+
 size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
 {
-    CHECK_F(ZSTD_decompressBegin(dctx));
-    if (dict && dictSize) CHECK_E(ZSTD_decompress_insertDictionary(dctx, dict, dictSize), dictionary_corrupted);
+    CHECK_F( ZSTD_decompressBegin(dctx) );
+    if (dict && dictSize)
+        CHECK_E(ZSTD_decompress_insertDictionary(dctx, dict, dictSize), dictionary_corrupted);
     return 0;
 }
 
@@ -1716,83 +2294,186 @@
     void* dictBuffer;
     const void* dictContent;
     size_t dictSize;
-    ZSTD_DCtx* refContext;
+    ZSTD_entropyDTables_t entropy;
+    U32 dictID;
+    U32 entropyPresent;
+    ZSTD_customMem cMem;
 };  /* typedef'd to ZSTD_DDict within "zstd.h" */
 
-ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize, unsigned byReference, ZSTD_customMem customMem)
+static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict)
+{
+    return ddict->dictContent;
+}
+
+static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict)
+{
+    return ddict->dictSize;
+}
+
+size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dstDCtx, const ZSTD_DDict* ddict)
+{
+    CHECK_F( ZSTD_decompressBegin(dstDCtx) );
+    if (ddict) {   /* support begin on NULL */
+        dstDCtx->dictID = ddict->dictID;
+        dstDCtx->base = ddict->dictContent;
+        dstDCtx->vBase = ddict->dictContent;
+        dstDCtx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
+        dstDCtx->previousDstEnd = dstDCtx->dictEnd;
+        if (ddict->entropyPresent) {
+            dstDCtx->litEntropy = 1;
+            dstDCtx->fseEntropy = 1;
+            dstDCtx->LLTptr = ddict->entropy.LLTable;
+            dstDCtx->MLTptr = ddict->entropy.MLTable;
+            dstDCtx->OFTptr = ddict->entropy.OFTable;
+            dstDCtx->HUFptr = ddict->entropy.hufTable;
+            dstDCtx->entropy.rep[0] = ddict->entropy.rep[0];
+            dstDCtx->entropy.rep[1] = ddict->entropy.rep[1];
+            dstDCtx->entropy.rep[2] = ddict->entropy.rep[2];
+        } else {
+            dstDCtx->litEntropy = 0;
+            dstDCtx->fseEntropy = 0;
+        }
+    }
+    return 0;
+}
+
+static size_t ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict, ZSTD_dictContentType_e dictContentType)
 {
-    if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
-    if (!customMem.customAlloc || !customMem.customFree) return NULL;
+    ddict->dictID = 0;
+    ddict->entropyPresent = 0;
+    if (dictContentType == ZSTD_dct_rawContent) return 0;
+
+    if (ddict->dictSize < 8) {
+        if (dictContentType == ZSTD_dct_fullDict)
+            return ERROR(dictionary_corrupted);   /* only accept specified dictionaries */
+        return 0;   /* pure content mode */
+    }
+    {   U32 const magic = MEM_readLE32(ddict->dictContent);
+        if (magic != ZSTD_MAGIC_DICTIONARY) {
+            if (dictContentType == ZSTD_dct_fullDict)
+                return ERROR(dictionary_corrupted);   /* only accept specified dictionaries */
+            return 0;   /* pure content mode */
+        }
+    }
+    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + ZSTD_frameIdSize);
+
+    /* load entropy tables */
+    CHECK_E( ZSTD_loadEntropy(&ddict->entropy, ddict->dictContent, ddict->dictSize), dictionary_corrupted );
+    ddict->entropyPresent = 1;
+    return 0;
+}
+
+
+static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
+                                      const void* dict, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType)
+{
+    if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dict) || (!dictSize)) {
+        ddict->dictBuffer = NULL;
+        ddict->dictContent = dict;
+    } else {
+        void* const internalBuffer = ZSTD_malloc(dictSize, ddict->cMem);
+        ddict->dictBuffer = internalBuffer;
+        ddict->dictContent = internalBuffer;
+        if (!internalBuffer) return ERROR(memory_allocation);
+        memcpy(internalBuffer, dict, dictSize);
+    }
+    ddict->dictSize = dictSize;
+    ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+
+    /* parse dictionary content */
+    CHECK_F( ZSTD_loadEntropy_inDDict(ddict, dictContentType) );
+
+    return 0;
+}
+
+ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType,
+                                      ZSTD_customMem customMem)
+{
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
 
     {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
-        ZSTD_DCtx* const dctx = ZSTD_createDCtx_advanced(customMem);
-
-        if (!ddict || !dctx) {
-            ZSTD_free(ddict, customMem);
-            ZSTD_free(dctx, customMem);
+        if (!ddict) return NULL;
+        ddict->cMem = customMem;
+
+        if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, dictLoadMethod, dictContentType) )) {
+            ZSTD_freeDDict(ddict);
             return NULL;
         }
 
-        if ((byReference) || (!dict) || (!dictSize)) {
-            ddict->dictBuffer = NULL;
-            ddict->dictContent = dict;
-        } else {
-            void* const internalBuffer = ZSTD_malloc(dictSize, customMem);
-            if (!internalBuffer) { ZSTD_free(dctx, customMem); ZSTD_free(ddict, customMem); return NULL; }
-            memcpy(internalBuffer, dict, dictSize);
-            ddict->dictBuffer = internalBuffer;
-            ddict->dictContent = internalBuffer;
-        }
-        {   size_t const errorCode = ZSTD_decompressBegin_usingDict(dctx, ddict->dictContent, dictSize);
-            if (ZSTD_isError(errorCode)) {
-                ZSTD_free(ddict->dictBuffer, customMem);
-                ZSTD_free(ddict, customMem);
-                ZSTD_free(dctx, customMem);
-                return NULL;
-        }   }
-
-        ddict->dictSize = dictSize;
-        ddict->refContext = dctx;
         return ddict;
     }
 }
 
 /*! ZSTD_createDDict() :
-*   Create a digested dictionary, ready to start decompression without startup delay.
-*   `dict` can be released after `ZSTD_DDict` creation */
+*   Create a digested dictionary, to start decompression without startup delay.
+*   `dict` content is copied inside DDict.
+*   Consequently, `dict` can be released after `ZSTD_DDict` creation */
 ZSTD_DDict* ZSTD_createDDict(const void* dict, size_t dictSize)
 {
     ZSTD_customMem const allocator = { NULL, NULL, NULL };
-    return ZSTD_createDDict_advanced(dict, dictSize, 0, allocator);
+    return ZSTD_createDDict_advanced(dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto, allocator);
 }
 
-
 /*! ZSTD_createDDict_byReference() :
- *  Create a digested dictionary, ready to start decompression operation without startup delay.
- *  Dictionary content is simply referenced, and therefore stays in dictBuffer.
- *  It is important that dictBuffer outlives DDict, it must remain read accessible throughout the lifetime of DDict */
+ *  Create a digested dictionary, to start decompression without startup delay.
+ *  Dictionary content is simply referenced, it will be accessed during decompression.
+ *  Warning : dictBuffer must outlive DDict (DDict must be freed before dictBuffer) */
 ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize)
 {
     ZSTD_customMem const allocator = { NULL, NULL, NULL };
-    return ZSTD_createDDict_advanced(dictBuffer, dictSize, 1, allocator);
+    return ZSTD_createDDict_advanced(dictBuffer, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto, allocator);
+}
+
+
+const ZSTD_DDict* ZSTD_initStaticDDict(
+                                void* workspace, size_t workspaceSize,
+                                const void* dict, size_t dictSize,
+                                ZSTD_dictLoadMethod_e dictLoadMethod,
+                                ZSTD_dictContentType_e dictContentType)
+{
+    size_t const neededSpace =
+            sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+    ZSTD_DDict* const ddict = (ZSTD_DDict*)workspace;
+    assert(workspace != NULL);
+    assert(dict != NULL);
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+    if (workspaceSize < neededSpace) return NULL;
+    if (dictLoadMethod == ZSTD_dlm_byCopy) {
+        memcpy(ddict+1, dict, dictSize);  /* local copy */
+        dict = ddict+1;
+    }
+    if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, ZSTD_dlm_byRef, dictContentType) ))
+        return NULL;
+    return ddict;
 }
 
 
 size_t ZSTD_freeDDict(ZSTD_DDict* ddict)
 {
     if (ddict==NULL) return 0;   /* support free on NULL */
-    {   ZSTD_customMem const cMem = ddict->refContext->customMem;
-        ZSTD_freeDCtx(ddict->refContext);
+    {   ZSTD_customMem const cMem = ddict->cMem;
         ZSTD_free(ddict->dictBuffer, cMem);
         ZSTD_free(ddict, cMem);
         return 0;
     }
 }
 
+/*! ZSTD_estimateDDictSize() :
+ *  Estimate amount of memory that will be needed to create a dictionary for decompression.
+ *  Note : dictionary created by reference using ZSTD_dlm_byRef are smaller */
+size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod)
+{
+    return sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+}
+
 size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict)
 {
     if (ddict==NULL) return 0;   /* support sizeof on NULL */
-    return sizeof(*ddict) + ZSTD_sizeof_DCtx(ddict->refContext) + (ddict->dictBuffer ? ddict->dictSize : 0) ;
+    return sizeof(*ddict) + (ddict->dictBuffer ? ddict->dictSize : 0) ;
 }
 
 /*! ZSTD_getDictID_fromDict() :
@@ -1802,8 +2483,8 @@
 unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
 {
     if (dictSize < 8) return 0;
-    if (MEM_readLE32(dict) != ZSTD_DICT_MAGIC) return 0;
-    return MEM_readLE32((const char*)dict + 4);
+    if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) return 0;
+    return MEM_readLE32((const char*)dict + ZSTD_frameIdSize);
 }
 
 /*! ZSTD_getDictID_fromDDict() :
@@ -1817,19 +2498,22 @@
 }
 
 /*! ZSTD_getDictID_fromFrame() :
- *  Provides the dictID required to decompressed the frame stored within `src`.
+ *  Provides the dictID required to decompresse frame stored within `src`.
  *  If @return == 0, the dictID could not be decoded.
  *  This could for one of the following reasons :
- *  - The frame does not require a dictionary to be decoded (most common case).
- *  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden information.
+ *  - The frame does not require a dictionary (most common case).
+ *  - The frame was built with dictID intentionally removed.
+ *    Needed dictionary is a hidden information.
  *    Note : this use case also happens when using a non-conformant dictionary.
- *  - `srcSize` is too small, and as a result, the frame header could not be decoded (only possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`).
+ *  - `srcSize` is too small, and as a result, frame header could not be decoded.
+ *    Note : possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`.
  *  - This is not a Zstandard frame.
- *  When identifying the exact failure cause, it's possible to used ZSTD_getFrameParams(), which will provide a more precise error code. */
+ *  When identifying the exact failure cause, it's possible to use
+ *  ZSTD_getFrameHeader(), which will provide a more precise error code. */
 unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize)
 {
-    ZSTD_frameParams zfp = { 0 , 0 , 0 , 0 };
-    size_t const hError = ZSTD_getFrameParams(&zfp, src, srcSize);
+    ZSTD_frameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0 };
+    size_t const hError = ZSTD_getFrameHeader(&zfp, src, srcSize);
     if (ZSTD_isError(hError)) return 0;
     return zfp.dictID;
 }
@@ -1843,12 +2527,10 @@
                             const void* src, size_t srcSize,
                             const ZSTD_DDict* ddict)
 {
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT==1)
-    if (ZSTD_isLegacy(src, srcSize)) return ZSTD_decompressLegacy(dst, dstCapacity, src, srcSize, ddict->dictContent, ddict->dictSize);
-#endif
-    ZSTD_refDCtx(dctx, ddict->refContext);
-    ZSTD_checkContinuity(dctx, dst);
-    return ZSTD_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+    /* pass content and size in case legacy frames are encountered */
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize,
+                                     NULL, 0,
+                                     ddict);
 }
 
 
@@ -1856,133 +2538,180 @@
 *   Streaming decompression
 *====================================*/
 
-typedef enum { zdss_init, zdss_loadHeader,
-               zdss_read, zdss_load, zdss_flush } ZSTD_dStreamStage;
-
-/* *** Resource management *** */
-struct ZSTD_DStream_s {
-    ZSTD_DCtx* dctx;
-    ZSTD_DDict* ddictLocal;
-    const ZSTD_DDict* ddict;
-    ZSTD_frameParams fParams;
-    ZSTD_dStreamStage stage;
-    char*  inBuff;
-    size_t inBuffSize;
-    size_t inPos;
-    size_t maxWindowSize;
-    char*  outBuff;
-    size_t outBuffSize;
-    size_t outStart;
-    size_t outEnd;
-    size_t blockSize;
-    BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];   /* tmp buffer to store frame header */
-    size_t lhSize;
-    ZSTD_customMem customMem;
-    void* legacyContext;
-    U32 previousLegacyVersion;
-    U32 legacyVersion;
-    U32 hostageByte;
-};   /* typedef'd to ZSTD_DStream within "zstd.h" */
-
-
 ZSTD_DStream* ZSTD_createDStream(void)
 {
-    return ZSTD_createDStream_advanced(defaultCustomMem);
+    DEBUGLOG(3, "ZSTD_createDStream");
+    return ZSTD_createDStream_advanced(ZSTD_defaultCMem);
+}
+
+ZSTD_DStream* ZSTD_initStaticDStream(void *workspace, size_t workspaceSize)
+{
+    return ZSTD_initStaticDCtx(workspace, workspaceSize);
 }
 
 ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem)
 {
-    ZSTD_DStream* zds;
-
-    if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
-    if (!customMem.customAlloc || !customMem.customFree) return NULL;
-
-    zds = (ZSTD_DStream*) ZSTD_malloc(sizeof(ZSTD_DStream), customMem);
-    if (zds==NULL) return NULL;
-    memset(zds, 0, sizeof(ZSTD_DStream));
-    memcpy(&zds->customMem, &customMem, sizeof(ZSTD_customMem));
-    zds->dctx = ZSTD_createDCtx_advanced(customMem);
-    if (zds->dctx == NULL) { ZSTD_freeDStream(zds); return NULL; }
-    zds->stage = zdss_init;
-    zds->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
-    return zds;
+    return ZSTD_createDCtx_advanced(customMem);
 }
 
 size_t ZSTD_freeDStream(ZSTD_DStream* zds)
 {
-    if (zds==NULL) return 0;   /* support free on null */
-    {   ZSTD_customMem const cMem = zds->customMem;
-        ZSTD_freeDCtx(zds->dctx);
-        ZSTD_freeDDict(zds->ddictLocal);
-        ZSTD_free(zds->inBuff, cMem);
-        ZSTD_free(zds->outBuff, cMem);
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
-        if (zds->legacyContext)
-            ZSTD_freeLegacyStreamContext(zds->legacyContext, zds->previousLegacyVersion);
-#endif
-        ZSTD_free(zds, cMem);
-        return 0;
-    }
+    return ZSTD_freeDCtx(zds);
 }
 
 
 /* *** Initialization *** */
 
-size_t ZSTD_DStreamInSize(void)  { return ZSTD_BLOCKSIZE_ABSOLUTEMAX + ZSTD_blockHeaderSize; }
-size_t ZSTD_DStreamOutSize(void) { return ZSTD_BLOCKSIZE_ABSOLUTEMAX; }
-
+size_t ZSTD_DStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize; }
+size_t ZSTD_DStreamOutSize(void) { return ZSTD_BLOCKSIZE_MAX; }
+
+size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType)
+{
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    ZSTD_freeDDict(dctx->ddictLocal);
+    if (dict && dictSize >= 8) {
+        dctx->ddictLocal = ZSTD_createDDict_advanced(dict, dictSize, dictLoadMethod, dictContentType, dctx->customMem);
+        if (dctx->ddictLocal == NULL) return ERROR(memory_allocation);
+    } else {
+        dctx->ddictLocal = NULL;
+    }
+    dctx->ddict = dctx->ddictLocal;
+    return 0;
+}
+
+size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, prefix, prefixSize, ZSTD_dlm_byRef, dictContentType);
+}
+
+size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize)
+{
+    return ZSTD_DCtx_refPrefix_advanced(dctx, prefix, prefixSize, ZSTD_dct_rawContent);
+}
+
+
+/* ZSTD_initDStream_usingDict() :
+ * return : expected size, aka ZSTD_frameHeaderSize_prefix.
+ * this function cannot fail */
 size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize)
 {
-    zds->stage = zdss_loadHeader;
-    zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
-    ZSTD_freeDDict(zds->ddictLocal);
-    if (dict && dictSize >= 8) {
-        zds->ddictLocal = ZSTD_createDDict(dict, dictSize);
-        if (zds->ddictLocal == NULL) return ERROR(memory_allocation);
-    } else zds->ddictLocal = NULL;
-    zds->ddict = zds->ddictLocal;
-    zds->legacyVersion = 0;
-    zds->hostageByte = 0;
+    DEBUGLOG(4, "ZSTD_initDStream_usingDict");
+    zds->streamStage = zdss_init;
+    CHECK_F( ZSTD_DCtx_loadDictionary(zds, dict, dictSize) );
     return ZSTD_frameHeaderSize_prefix;
 }
 
+/* note : this variant can't fail */
 size_t ZSTD_initDStream(ZSTD_DStream* zds)
 {
+    DEBUGLOG(4, "ZSTD_initDStream");
     return ZSTD_initDStream_usingDict(zds, NULL, 0);
 }
 
-size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict)  /**< note : ddict will just be referenced, and must outlive decompression session */
+size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
 {
-    size_t const initResult = ZSTD_initDStream(zds);
-    zds->ddict = ddict;
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    dctx->ddict = ddict;
+    return 0;
+}
+
+/* ZSTD_initDStream_usingDDict() :
+ * ddict will just be referenced, and must outlive decompression session
+ * this function cannot fail */
+size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
+{
+    size_t const initResult = ZSTD_initDStream(dctx);
+    dctx->ddict = ddict;
     return initResult;
 }
 
-size_t ZSTD_resetDStream(ZSTD_DStream* zds)
+/* ZSTD_resetDStream() :
+ * return : expected size, aka ZSTD_frameHeaderSize_prefix.
+ * this function cannot fail */
+size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
 {
-    zds->stage = zdss_loadHeader;
-    zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
-    zds->legacyVersion = 0;
-    zds->hostageByte = 0;
+    DEBUGLOG(4, "ZSTD_resetDStream");
+    dctx->streamStage = zdss_loadHeader;
+    dctx->lhSize = dctx->inPos = dctx->outStart = dctx->outEnd = 0;
+    dctx->legacyVersion = 0;
+    dctx->hostageByte = 0;
     return ZSTD_frameHeaderSize_prefix;
 }
 
-size_t ZSTD_setDStreamParameter(ZSTD_DStream* zds,
+size_t ZSTD_setDStreamParameter(ZSTD_DStream* dctx,
                                 ZSTD_DStreamParameter_e paramType, unsigned paramValue)
 {
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
     switch(paramType)
     {
-        default : return ERROR(parameter_unknown);
-        case DStream_p_maxWindowSize : zds->maxWindowSize = paramValue ? paramValue : (U32)(-1); break;
+        default : return ERROR(parameter_unsupported);
+        case DStream_p_maxWindowSize :
+            DEBUGLOG(4, "setting maxWindowSize = %u KB", paramValue >> 10);
+            dctx->maxWindowSize = paramValue ? paramValue : (U32)(-1);
+            break;
     }
     return 0;
 }
 
-
-size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds)
+size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
+{
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    dctx->maxWindowSize = maxWindowSize;
+    return 0;
+}
+
+size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format)
+{
+    DEBUGLOG(4, "ZSTD_DCtx_setFormat : %u", (unsigned)format);
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    dctx->format = format;
+    return 0;
+}
+
+
+size_t ZSTD_sizeof_DStream(const ZSTD_DStream* dctx)
+{
+    return ZSTD_sizeof_DCtx(dctx);
+}
+
+size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
 {
-    if (zds==NULL) return 0;   /* support sizeof on NULL */
-    return sizeof(*zds) + ZSTD_sizeof_DCtx(zds->dctx) + ZSTD_sizeof_DDict(zds->ddictLocal) + zds->inBuffSize + zds->outBuffSize;
+    size_t const blockSize = (size_t) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+    unsigned long long const neededRBSize = windowSize + blockSize + (WILDCOPY_OVERLENGTH * 2);
+    unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
+    size_t const minRBSize = (size_t) neededSize;
+    if ((unsigned long long)minRBSize != neededSize) return ERROR(frameParameter_windowTooLarge);
+    return minRBSize;
+}
+
+size_t ZSTD_estimateDStreamSize(size_t windowSize)
+{
+    size_t const blockSize = MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+    size_t const inBuffSize = blockSize;  /* no block can be larger */
+    size_t const outBuffSize = ZSTD_decodingBufferSize_min(windowSize, ZSTD_CONTENTSIZE_UNKNOWN);
+    return ZSTD_estimateDCtxSize() + inBuffSize + outBuffSize;
+}
+
+size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize)
+{
+    U32 const windowSizeMax = 1U << ZSTD_WINDOWLOG_MAX;   /* note : should be user-selectable */
+    ZSTD_frameHeader zfh;
+    size_t const err = ZSTD_getFrameHeader(&zfh, src, srcSize);
+    if (ZSTD_isError(err)) return err;
+    if (err>0) return ERROR(srcSize_wrong);
+    if (zfh.windowSize > windowSizeMax)
+        return ERROR(frameParameter_windowTooLarge);
+    return ZSTD_estimateDStreamSize((size_t)zfh.windowSize);
 }
 
 
@@ -2006,126 +2735,193 @@
     char* op = ostart;
     U32 someMoreWork = 1;
 
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
-    if (zds->legacyVersion)
-        return ZSTD_decompressLegacyStream(zds->legacyContext, zds->legacyVersion, output, input);
-#endif
+    DEBUGLOG(5, "ZSTD_decompressStream");
+    if (input->pos > input->size) {  /* forbidden */
+        DEBUGLOG(5, "in: pos: %u   vs size: %u",
+                    (U32)input->pos, (U32)input->size);
+        return ERROR(srcSize_wrong);
+    }
+    if (output->pos > output->size) {  /* forbidden */
+        DEBUGLOG(5, "out: pos: %u   vs size: %u",
+                    (U32)output->pos, (U32)output->size);
+        return ERROR(dstSize_tooSmall);
+    }
+    DEBUGLOG(5, "input size : %u", (U32)(input->size - input->pos));
 
     while (someMoreWork) {
-        switch(zds->stage)
+        switch(zds->streamStage)
         {
         case zdss_init :
+            DEBUGLOG(5, "stage zdss_init => transparent reset ");
             ZSTD_resetDStream(zds);   /* transparent reset on starting decoding a new frame */
             /* fall-through */
 
         case zdss_loadHeader :
-            {   size_t const hSize = ZSTD_getFrameParams(&zds->fParams, zds->headerBuffer, zds->lhSize);
-                if (ZSTD_isError(hSize))
+            DEBUGLOG(5, "stage zdss_loadHeader (srcSize : %u)", (U32)(iend - ip));
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
-                {   U32 const legacyVersion = ZSTD_isLegacy(istart, iend-istart);
+            if (zds->legacyVersion) {
+                /* legacy support is incompatible with static dctx */
+                if (zds->staticSize) return ERROR(memory_allocation);
+                {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, zds->legacyVersion, output, input);
+                    if (hint==0) zds->streamStage = zdss_init;
+                    return hint;
+            }   }
+#endif
+            {   size_t const hSize = ZSTD_getFrameHeader_internal(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
+                DEBUGLOG(5, "header size : %u", (U32)hSize);
+                if (ZSTD_isError(hSize)) {
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+                    U32 const legacyVersion = ZSTD_isLegacy(istart, iend-istart);
                     if (legacyVersion) {
                         const void* const dict = zds->ddict ? zds->ddict->dictContent : NULL;
                         size_t const dictSize = zds->ddict ? zds->ddict->dictSize : 0;
-                        CHECK_F(ZSTD_initLegacyStream(&zds->legacyContext, zds->previousLegacyVersion, legacyVersion,
-                                                       dict, dictSize));
+                        DEBUGLOG(5, "ZSTD_decompressStream: detected legacy version v0.%u", legacyVersion);
+                        /* legacy support is incompatible with static dctx */
+                        if (zds->staticSize) return ERROR(memory_allocation);
+                        CHECK_F(ZSTD_initLegacyStream(&zds->legacyContext,
+                                    zds->previousLegacyVersion, legacyVersion,
+                                    dict, dictSize));
                         zds->legacyVersion = zds->previousLegacyVersion = legacyVersion;
-                        return ZSTD_decompressLegacyStream(zds->legacyContext, zds->legacyVersion, output, input);
-                    } else {
-                        return hSize; /* error */
-                }   }
-#else
-                return hSize;
+                        {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, legacyVersion, output, input);
+                            if (hint==0) zds->streamStage = zdss_init;   /* or stay in stage zdss_loadHeader */
+                            return hint;
+                    }   }
 #endif
+                    return hSize;   /* error */
+                }
                 if (hSize != 0) {   /* need more input */
                     size_t const toLoad = hSize - zds->lhSize;   /* if hSize!=0, hSize > zds->lhSize */
-                    if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
-                        memcpy(zds->headerBuffer + zds->lhSize, ip, iend-ip);
-                        zds->lhSize += iend-ip;
+                    size_t const remainingInput = (size_t)(iend-ip);
+                    assert(iend >= ip);
+                    if (toLoad > remainingInput) {   /* not enough input to load full header */
+                        if (remainingInput > 0) {
+                            memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
+                            zds->lhSize += remainingInput;
+                        }
                         input->pos = input->size;
                         return (MAX(ZSTD_frameHeaderSize_min, hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
                     }
+                    assert(ip != NULL);
                     memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
                     break;
             }   }
 
-            /* Consume header */
-            {   const ZSTD_DCtx* refContext = zds->ddict ? zds->ddict->refContext : NULL;
-                ZSTD_refDCtx(zds->dctx, refContext);
+            /* check for single-pass mode opportunity */
+            if (zds->fParams.frameContentSize && zds->fParams.windowSize /* skippable frame if == 0 */
+                && (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
+                size_t const cSize = ZSTD_findFrameCompressedSize(istart, iend-istart);
+                if (cSize <= (size_t)(iend-istart)) {
+                    /* shortcut : using single-pass mode */
+                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, oend-op, istart, cSize, zds->ddict);
+                    if (ZSTD_isError(decompressedSize)) return decompressedSize;
+                    DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()")
+                    ip = istart + cSize;
+                    op += decompressedSize;
+                    zds->expected = 0;
+                    zds->streamStage = zdss_init;
+                    someMoreWork = 0;
+                    break;
+            }   }
+
+            /* Consume header (see ZSTDds_decodeFrameHeader) */
+            DEBUGLOG(4, "Consume header");
+            CHECK_F(ZSTD_decompressBegin_usingDDict(zds, zds->ddict));
+
+            if ((MEM_readLE32(zds->headerBuffer) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
+                zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_frameIdSize);
+                zds->stage = ZSTDds_skipFrame;
+            } else {
+                CHECK_F(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize));
+                zds->expected = ZSTD_blockHeaderSize;
+                zds->stage = ZSTDds_decodeBlockHeader;
             }
-            {   size_t const h1Size = ZSTD_nextSrcSizeToDecompress(zds->dctx);  /* == ZSTD_frameHeaderSize_prefix */
-                CHECK_F(ZSTD_decompressContinue(zds->dctx, NULL, 0, zds->headerBuffer, h1Size));
-                {   size_t const h2Size = ZSTD_nextSrcSizeToDecompress(zds->dctx);
-                    CHECK_F(ZSTD_decompressContinue(zds->dctx, NULL, 0, zds->headerBuffer+h1Size, h2Size));
-            }   }
-
+
+            /* control buffer memory usage */
+            DEBUGLOG(4, "Control max memory usage (%u KB <= max %u KB)",
+                        (U32)(zds->fParams.windowSize >>10),
+                        (U32)(zds->maxWindowSize >> 10) );
             zds->fParams.windowSize = MAX(zds->fParams.windowSize, 1U << ZSTD_WINDOWLOG_ABSOLUTEMIN);
             if (zds->fParams.windowSize > zds->maxWindowSize) return ERROR(frameParameter_windowTooLarge);
 
             /* Adapt buffer sizes to frame header instructions */
-            {   size_t const blockSize = MIN(zds->fParams.windowSize, ZSTD_BLOCKSIZE_ABSOLUTEMAX);
-                size_t const neededOutSize = zds->fParams.windowSize + blockSize;
-                zds->blockSize = blockSize;
-                if (zds->inBuffSize < blockSize) {
-                    ZSTD_free(zds->inBuff, zds->customMem);
-                    zds->inBuffSize = blockSize;
-                    zds->inBuff = (char*)ZSTD_malloc(blockSize, zds->customMem);
-                    if (zds->inBuff == NULL) return ERROR(memory_allocation);
-                }
-                if (zds->outBuffSize < neededOutSize) {
-                    ZSTD_free(zds->outBuff, zds->customMem);
-                    zds->outBuffSize = neededOutSize;
-                    zds->outBuff = (char*)ZSTD_malloc(neededOutSize, zds->customMem);
-                    if (zds->outBuff == NULL) return ERROR(memory_allocation);
+            {   size_t const neededInBuffSize = MAX(zds->fParams.blockSizeMax, 4 /* frame checksum */);
+                size_t const neededOutBuffSize = ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize);
+                if ((zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize)) {
+                    size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
+                    DEBUGLOG(4, "inBuff  : from %u to %u",
+                                (U32)zds->inBuffSize, (U32)neededInBuffSize);
+                    DEBUGLOG(4, "outBuff : from %u to %u",
+                                (U32)zds->outBuffSize, (U32)neededOutBuffSize);
+                    if (zds->staticSize) {  /* static DCtx */
+                        DEBUGLOG(4, "staticSize : %u", (U32)zds->staticSize);
+                        assert(zds->staticSize >= sizeof(ZSTD_DCtx));  /* controlled at init */
+                        if (bufferSize > zds->staticSize - sizeof(ZSTD_DCtx))
+                            return ERROR(memory_allocation);
+                    } else {
+                        ZSTD_free(zds->inBuff, zds->customMem);
+                        zds->inBuffSize = 0;
+                        zds->outBuffSize = 0;
+                        zds->inBuff = (char*)ZSTD_malloc(bufferSize, zds->customMem);
+                        if (zds->inBuff == NULL) return ERROR(memory_allocation);
+                    }
+                    zds->inBuffSize = neededInBuffSize;
+                    zds->outBuff = zds->inBuff + zds->inBuffSize;
+                    zds->outBuffSize = neededOutBuffSize;
             }   }
-            zds->stage = zdss_read;
-            /* pass-through */
+            zds->streamStage = zdss_read;
+            /* fall-through */
 
         case zdss_read:
-            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds->dctx);
+            DEBUGLOG(5, "stage zdss_read");
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
+                DEBUGLOG(5, "neededInSize = %u", (U32)neededInSize);
                 if (neededInSize==0) {  /* end of frame */
-                    zds->stage = zdss_init;
+                    zds->streamStage = zdss_init;
                     someMoreWork = 0;
                     break;
                 }
                 if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
-                    const int isSkipFrame = ZSTD_isSkipFrame(zds->dctx);
-                    size_t const decodedSize = ZSTD_decompressContinue(zds->dctx,
+                    int const isSkipFrame = ZSTD_isSkipFrame(zds);
+                    size_t const decodedSize = ZSTD_decompressContinue(zds,
                         zds->outBuff + zds->outStart, (isSkipFrame ? 0 : zds->outBuffSize - zds->outStart),
                         ip, neededInSize);
                     if (ZSTD_isError(decodedSize)) return decodedSize;
                     ip += neededInSize;
                     if (!decodedSize && !isSkipFrame) break;   /* this was just a header */
                     zds->outEnd = zds->outStart + decodedSize;
-                    zds->stage = zdss_flush;
+                    zds->streamStage = zdss_flush;
                     break;
-                }
-                if (ip==iend) { someMoreWork = 0; break; }   /* no more input */
-                zds->stage = zdss_load;
-                /* pass-through */
-            }
+            }   }
+            if (ip==iend) { someMoreWork = 0; break; }   /* no more input */
+            zds->streamStage = zdss_load;
+            /* fall-through */
 
         case zdss_load:
-            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds->dctx);
-                size_t const toLoad = neededInSize - zds->inPos;   /* should always be <= remaining space within inBuff */
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
+                size_t const toLoad = neededInSize - zds->inPos;
+                int const isSkipFrame = ZSTD_isSkipFrame(zds);
                 size_t loadedSize;
-                if (toLoad > zds->inBuffSize - zds->inPos) return ERROR(corruption_detected);   /* should never happen */
-                loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, iend-ip);
+                if (isSkipFrame) {
+                    loadedSize = MIN(toLoad, (size_t)(iend-ip));
+                } else {
+                    if (toLoad > zds->inBuffSize - zds->inPos) return ERROR(corruption_detected);   /* should never happen */
+                    loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, iend-ip);
+                }
                 ip += loadedSize;
                 zds->inPos += loadedSize;
                 if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
 
                 /* decode loaded input */
-                {  const int isSkipFrame = ZSTD_isSkipFrame(zds->dctx);
-                   size_t const decodedSize = ZSTD_decompressContinue(zds->dctx,
+                {   size_t const decodedSize = ZSTD_decompressContinue(zds,
                         zds->outBuff + zds->outStart, zds->outBuffSize - zds->outStart,
                         zds->inBuff, neededInSize);
                     if (ZSTD_isError(decodedSize)) return decodedSize;
                     zds->inPos = 0;   /* input is consumed */
-                    if (!decodedSize && !isSkipFrame) { zds->stage = zdss_read; break; }   /* this was just a header */
+                    if (!decodedSize && !isSkipFrame) { zds->streamStage = zdss_read; break; }   /* this was just a header */
                     zds->outEnd = zds->outStart +  decodedSize;
-                    zds->stage = zdss_flush;
-                    /* pass-through */
             }   }
+            zds->streamStage = zdss_flush;
+            /* fall-through */
 
         case zdss_flush:
             {   size_t const toFlushSize = zds->outEnd - zds->outStart;
@@ -2133,39 +2929,75 @@
                 op += flushedSize;
                 zds->outStart += flushedSize;
                 if (flushedSize == toFlushSize) {  /* flush completed */
-                    zds->stage = zdss_read;
-                    if (zds->outStart + zds->blockSize > zds->outBuffSize)
+                    zds->streamStage = zdss_read;
+                    if ( (zds->outBuffSize < zds->fParams.frameContentSize)
+                      && (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
+                        DEBUGLOG(5, "restart filling outBuff from beginning (left:%i, needed:%u)",
+                                (int)(zds->outBuffSize - zds->outStart),
+                                (U32)zds->fParams.blockSizeMax);
                         zds->outStart = zds->outEnd = 0;
+                    }
                     break;
-                }
-                /* cannot complete flush */
-                someMoreWork = 0;
-                break;
-            }
+            }   }
+            /* cannot complete flush */
+            someMoreWork = 0;
+            break;
+
         default: return ERROR(GENERIC);   /* impossible */
     }   }
 
     /* result */
     input->pos += (size_t)(ip-istart);
     output->pos += (size_t)(op-ostart);
-    {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zds->dctx);
+    {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zds);
         if (!nextSrcSizeHint) {   /* frame fully decoded */
             if (zds->outEnd == zds->outStart) {  /* output fully flushed */
                 if (zds->hostageByte) {
-                    if (input->pos >= input->size) { zds->stage = zdss_read; return 1; }  /* can't release hostage (not present) */
+                    if (input->pos >= input->size) {
+                        /* can't release hostage (not present) */
+                        zds->streamStage = zdss_read;
+                        return 1;
+                    }
                     input->pos++;  /* release hostage */
-                }
+                }   /* zds->hostageByte */
                 return 0;
-            }
+            }  /* zds->outEnd == zds->outStart */
             if (!zds->hostageByte) { /* output not fully flushed; keep last byte as hostage; will be released when all output is flushed */
                 input->pos--;   /* note : pos > 0, otherwise, impossible to finish reading last block */
                 zds->hostageByte=1;
             }
             return 1;
-        }
-        nextSrcSizeHint += ZSTD_blockHeaderSize * (ZSTD_nextInputType(zds->dctx) == ZSTDnit_block);   /* preload header of next block */
-        if (zds->inPos > nextSrcSizeHint) return ERROR(GENERIC);   /* should never happen */
-        nextSrcSizeHint -= zds->inPos;   /* already loaded*/
+        }  /* nextSrcSizeHint==0 */
+        nextSrcSizeHint += ZSTD_blockHeaderSize * (ZSTD_nextInputType(zds) == ZSTDnit_block);   /* preload header of next block */
+        assert(zds->inPos <= nextSrcSizeHint);
+        nextSrcSizeHint -= zds->inPos;   /* part already loaded*/
         return nextSrcSizeHint;
     }
 }
+
+
+size_t ZSTD_decompress_generic(ZSTD_DCtx* dctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    return ZSTD_decompressStream(dctx, output, input);
+}
+
+size_t ZSTD_decompress_generic_simpleArgs (
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos)
+{
+    ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
+    ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
+    /* ZSTD_compress_generic() will check validity of dstPos and srcPos */
+    size_t const cErr = ZSTD_decompress_generic(dctx, &output, &input);
+    *dstPos = output.pos;
+    *srcPos = input.pos;
+    return cErr;
+}
+
+void ZSTD_DCtx_reset(ZSTD_DCtx* dctx)
+{
+    (void)ZSTD_initDStream(dctx);
+    dctx->format = ZSTD_f_zstd1;
+    dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/deprecated/zbuff.h	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,213 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* ***************************************************************
+*  NOTES/WARNINGS
+******************************************************************/
+/* The streaming API defined here is deprecated.
+ * Consider migrating towards ZSTD_compressStream() API in `zstd.h`
+ * See 'lib/README.md'.
+ *****************************************************************/
+
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef ZSTD_BUFFERED_H_23987
+#define ZSTD_BUFFERED_H_23987
+
+/* *************************************
+*  Dependencies
+***************************************/
+#include <stddef.h>      /* size_t */
+#include "zstd.h"        /* ZSTD_CStream, ZSTD_DStream, ZSTDLIB_API */
+
+
+/* ***************************************************************
+*  Compiler specifics
+*****************************************************************/
+/* Deprecation warnings */
+/* Should these warnings be a problem,
+   it is generally possible to disable them,
+   typically with -Wno-deprecated-declarations for gcc
+   or _CRT_SECURE_NO_WARNINGS in Visual.
+   Otherwise, it's also possible to define ZBUFF_DISABLE_DEPRECATE_WARNINGS */
+#ifdef ZBUFF_DISABLE_DEPRECATE_WARNINGS
+#  define ZBUFF_DEPRECATED(message) ZSTDLIB_API  /* disable deprecation warnings */
+#else
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define ZBUFF_DEPRECATED(message) [[deprecated(message)]] ZSTDLIB_API
+#  elif (defined(__GNUC__) && (__GNUC__ >= 5)) || defined(__clang__)
+#    define ZBUFF_DEPRECATED(message) ZSTDLIB_API __attribute__((deprecated(message)))
+#  elif defined(__GNUC__) && (__GNUC__ >= 3)
+#    define ZBUFF_DEPRECATED(message) ZSTDLIB_API __attribute__((deprecated))
+#  elif defined(_MSC_VER)
+#    define ZBUFF_DEPRECATED(message) ZSTDLIB_API __declspec(deprecated(message))
+#  else
+#    pragma message("WARNING: You need to implement ZBUFF_DEPRECATED for this compiler")
+#    define ZBUFF_DEPRECATED(message) ZSTDLIB_API
+#  endif
+#endif /* ZBUFF_DISABLE_DEPRECATE_WARNINGS */
+
+
+/* *************************************
+*  Streaming functions
+***************************************/
+/* This is the easier "buffered" streaming API,
+*  using an internal buffer to lift all restrictions on user-provided buffers
+*  which can be any size, any place, for both input and output.
+*  ZBUFF and ZSTD are 100% interoperable,
+*  frames created by one can be decoded by the other one */
+
+typedef ZSTD_CStream ZBUFF_CCtx;
+ZBUFF_DEPRECATED("use ZSTD_createCStream") ZBUFF_CCtx* ZBUFF_createCCtx(void);
+ZBUFF_DEPRECATED("use ZSTD_freeCStream")   size_t      ZBUFF_freeCCtx(ZBUFF_CCtx* cctx);
+
+ZBUFF_DEPRECATED("use ZSTD_initCStream")           size_t ZBUFF_compressInit(ZBUFF_CCtx* cctx, int compressionLevel);
+ZBUFF_DEPRECATED("use ZSTD_initCStream_usingDict") size_t ZBUFF_compressInitDictionary(ZBUFF_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
+
+ZBUFF_DEPRECATED("use ZSTD_compressStream") size_t ZBUFF_compressContinue(ZBUFF_CCtx* cctx, void* dst, size_t* dstCapacityPtr, const void* src, size_t* srcSizePtr);
+ZBUFF_DEPRECATED("use ZSTD_flushStream")    size_t ZBUFF_compressFlush(ZBUFF_CCtx* cctx, void* dst, size_t* dstCapacityPtr);
+ZBUFF_DEPRECATED("use ZSTD_endStream")      size_t ZBUFF_compressEnd(ZBUFF_CCtx* cctx, void* dst, size_t* dstCapacityPtr);
+
+/*-*************************************************
+*  Streaming compression - howto
+*
+*  A ZBUFF_CCtx object is required to track streaming operation.
+*  Use ZBUFF_createCCtx() and ZBUFF_freeCCtx() to create/release resources.
+*  ZBUFF_CCtx objects can be reused multiple times.
+*
+*  Start by initializing ZBUF_CCtx.
+*  Use ZBUFF_compressInit() to start a new compression operation.
+*  Use ZBUFF_compressInitDictionary() for a compression which requires a dictionary.
+*
+*  Use ZBUFF_compressContinue() repetitively to consume input stream.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written within *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present again remaining data.
+*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each call, so save its content if it matters or change @dst .
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's just a hint, to improve latency)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  At any moment, it's possible to flush whatever data remains within buffer, using ZBUFF_compressFlush().
+*  The nb of bytes written into `dst` will be reported into *dstCapacityPtr.
+*  Note that the function cannot output more than *dstCapacityPtr,
+*  therefore, some content might still be left into internal buffer if *dstCapacityPtr is too small.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  ZBUFF_compressEnd() instructs to finish a frame.
+*  It will perform a flush and write frame epilogue.
+*  The epilogue is required for decoders to consider a frame completed.
+*  Similar to ZBUFF_compressFlush(), it may not be able to output the entire internal buffer content if *dstCapacityPtr is too small.
+*  In which case, call again ZBUFF_compressFlush() to complete the flush.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : _recommended buffer_ sizes (not compulsory) : ZBUFF_recommendedCInSize() / ZBUFF_recommendedCOutSize()
+*  input : ZBUFF_recommendedCInSize==128 KB block size is the internal unit, use this value to reduce intermediate stages (better latency)
+*  output : ZBUFF_recommendedCOutSize==ZSTD_compressBound(128 KB) + 3 + 3 : ensures it's always possible to write/flush/end a full block. Skip some buffering.
+*  By using both, it ensures that input will be entirely consumed, and output will always contain the result, reducing intermediate buffering.
+* **************************************************/
+
+
+typedef ZSTD_DStream ZBUFF_DCtx;
+ZBUFF_DEPRECATED("use ZSTD_createDStream") ZBUFF_DCtx* ZBUFF_createDCtx(void);
+ZBUFF_DEPRECATED("use ZSTD_freeDStream")   size_t      ZBUFF_freeDCtx(ZBUFF_DCtx* dctx);
+
+ZBUFF_DEPRECATED("use ZSTD_initDStream")           size_t ZBUFF_decompressInit(ZBUFF_DCtx* dctx);
+ZBUFF_DEPRECATED("use ZSTD_initDStream_usingDict") size_t ZBUFF_decompressInitDictionary(ZBUFF_DCtx* dctx, const void* dict, size_t dictSize);
+
+ZBUFF_DEPRECATED("use ZSTD_decompressStream") size_t ZBUFF_decompressContinue(ZBUFF_DCtx* dctx,
+                                            void* dst, size_t* dstCapacityPtr,
+                                      const void* src, size_t* srcSizePtr);
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFF_DCtx object is required to track streaming operations.
+*  Use ZBUFF_createDCtx() and ZBUFF_freeDCtx() to create/release resources.
+*  Use ZBUFF_decompressInit() to start a new decompression operation,
+*   or ZBUFF_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFF_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFF_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change `dst`.
+*  @return : 0 when a frame is completely decoded and fully flushed,
+*            1 when there is still some data left within internal buffer to flush,
+*            >1 when more data is expected, with value being a suggested next input size (it's just a hint, which helps latency),
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFF_recommendedDInSize() and ZBUFF_recommendedDOutSize()
+*  output : ZBUFF_recommendedDOutSize== 128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFF_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFF_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+ZBUFF_DEPRECATED("use ZSTD_isError")      unsigned ZBUFF_isError(size_t errorCode);
+ZBUFF_DEPRECATED("use ZSTD_getErrorName") const char* ZBUFF_getErrorName(size_t errorCode);
+
+/** Functions below provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are just hints, they tend to offer better latency */
+ZBUFF_DEPRECATED("use ZSTD_CStreamInSize")  size_t ZBUFF_recommendedCInSize(void);
+ZBUFF_DEPRECATED("use ZSTD_CStreamOutSize") size_t ZBUFF_recommendedCOutSize(void);
+ZBUFF_DEPRECATED("use ZSTD_DStreamInSize")  size_t ZBUFF_recommendedDInSize(void);
+ZBUFF_DEPRECATED("use ZSTD_DStreamOutSize") size_t ZBUFF_recommendedDOutSize(void);
+
+#endif  /* ZSTD_BUFFERED_H_23987 */
+
+
+#ifdef ZBUFF_STATIC_LINKING_ONLY
+#ifndef ZBUFF_STATIC_H_30298098432
+#define ZBUFF_STATIC_H_30298098432
+
+/* ====================================================================================
+ * The definitions in this section are considered experimental.
+ * They should never be used in association with a dynamic library, as they may change in the future.
+ * They are provided for advanced usages.
+ * Use them only in association with static linking.
+ * ==================================================================================== */
+
+/*--- Dependency ---*/
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_parameters, ZSTD_customMem */
+#include "zstd.h"
+
+
+/*--- Custom memory allocator ---*/
+/*! ZBUFF_createCCtx_advanced() :
+ *  Create a ZBUFF compression context using external alloc and free functions */
+ZBUFF_DEPRECATED("use ZSTD_createCStream_advanced") ZBUFF_CCtx* ZBUFF_createCCtx_advanced(ZSTD_customMem customMem);
+
+/*! ZBUFF_createDCtx_advanced() :
+ *  Create a ZBUFF decompression context using external alloc and free functions */
+ZBUFF_DEPRECATED("use ZSTD_createDStream_advanced") ZBUFF_DCtx* ZBUFF_createDCtx_advanced(ZSTD_customMem customMem);
+
+
+/*--- Advanced Streaming Initialization ---*/
+ZBUFF_DEPRECATED("use ZSTD_initDStream_usingDict") size_t ZBUFF_compressInit_advanced(ZBUFF_CCtx* zbc,
+                                               const void* dict, size_t dictSize,
+                                               ZSTD_parameters params, unsigned long long pledgedSrcSize);
+
+
+#endif    /* ZBUFF_STATIC_H_30298098432 */
+#endif    /* ZBUFF_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/deprecated/zbuff_common.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,26 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "error_private.h"
+#include "zbuff.h"
+
+/*-****************************************
+*  ZBUFF Error Management  (deprecated)
+******************************************/
+
+/*! ZBUFF_isError() :
+*   tells if a return value is an error code */
+unsigned ZBUFF_isError(size_t errorCode) { return ERR_isError(errorCode); }
+/*! ZBUFF_getErrorName() :
+*   provides error code string from function result (useful for debugging) */
+const char* ZBUFF_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/deprecated/zbuff_compress.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,147 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+
+/* *************************************
+*  Dependencies
+***************************************/
+#define ZBUFF_STATIC_LINKING_ONLY
+#include "zbuff.h"
+
+
+/*-***********************************************************
+*  Streaming compression
+*
+*  A ZBUFF_CCtx object is required to track streaming operation.
+*  Use ZBUFF_createCCtx() and ZBUFF_freeCCtx() to create/release resources.
+*  Use ZBUFF_compressInit() to start a new compression operation.
+*  ZBUFF_CCtx objects can be reused multiple times.
+*
+*  Use ZBUFF_compressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
+*  The content of dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters or change dst .
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  ZBUFF_compressFlush() can be used to instruct ZBUFF to compress and output whatever remains within its buffer.
+*  Note that it will not output more than *dstCapacityPtr.
+*  Therefore, some content might still be left into its internal buffer if dst buffer is too small.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  ZBUFF_compressEnd() instructs to finish a frame.
+*  It will perform a flush and write frame epilogue.
+*  Similar to ZBUFF_compressFlush(), it may not be able to output the entire internal buffer content if *dstCapacityPtr is too small.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory)
+*  input : ZSTD_BLOCKSIZE_MAX (128 KB), internal unit size, it improves latency to use this value.
+*  output : ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + ZBUFF_endFrameSize : ensures it's always possible to write/flush/end a full block at best speed.
+* ***********************************************************/
+
+ZBUFF_CCtx* ZBUFF_createCCtx(void)
+{
+    return ZSTD_createCStream();
+}
+
+ZBUFF_CCtx* ZBUFF_createCCtx_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createCStream_advanced(customMem);
+}
+
+size_t ZBUFF_freeCCtx(ZBUFF_CCtx* zbc)
+{
+    return ZSTD_freeCStream(zbc);
+}
+
+
+/* ======   Initialization   ====== */
+
+size_t ZBUFF_compressInit_advanced(ZBUFF_CCtx* zbc,
+                                   const void* dict, size_t dictSize,
+                                   ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;  /* preserve "0 == unknown" behavior */
+    return ZSTD_initCStream_advanced(zbc, dict, dictSize, params, pledgedSrcSize);
+}
+
+
+size_t ZBUFF_compressInitDictionary(ZBUFF_CCtx* zbc, const void* dict, size_t dictSize, int compressionLevel)
+{
+    return ZSTD_initCStream_usingDict(zbc, dict, dictSize, compressionLevel);
+}
+
+size_t ZBUFF_compressInit(ZBUFF_CCtx* zbc, int compressionLevel)
+{
+    return ZSTD_initCStream(zbc, compressionLevel);
+}
+
+/* ======   Compression   ====== */
+
+
+size_t ZBUFF_compressContinue(ZBUFF_CCtx* zbc,
+                              void* dst, size_t* dstCapacityPtr,
+                        const void* src, size_t* srcSizePtr)
+{
+    size_t result;
+    ZSTD_outBuffer outBuff;
+    ZSTD_inBuffer inBuff;
+    outBuff.dst = dst;
+    outBuff.pos = 0;
+    outBuff.size = *dstCapacityPtr;
+    inBuff.src = src;
+    inBuff.pos = 0;
+    inBuff.size = *srcSizePtr;
+    result = ZSTD_compressStream(zbc, &outBuff, &inBuff);
+    *dstCapacityPtr = outBuff.pos;
+    *srcSizePtr = inBuff.pos;
+    return result;
+}
+
+
+
+/* ======   Finalize   ====== */
+
+size_t ZBUFF_compressFlush(ZBUFF_CCtx* zbc, void* dst, size_t* dstCapacityPtr)
+{
+    size_t result;
+    ZSTD_outBuffer outBuff;
+    outBuff.dst = dst;
+    outBuff.pos = 0;
+    outBuff.size = *dstCapacityPtr;
+    result = ZSTD_flushStream(zbc, &outBuff);
+    *dstCapacityPtr = outBuff.pos;
+    return result;
+}
+
+
+size_t ZBUFF_compressEnd(ZBUFF_CCtx* zbc, void* dst, size_t* dstCapacityPtr)
+{
+    size_t result;
+    ZSTD_outBuffer outBuff;
+    outBuff.dst = dst;
+    outBuff.pos = 0;
+    outBuff.size = *dstCapacityPtr;
+    result = ZSTD_endStream(zbc, &outBuff);
+    *dstCapacityPtr = outBuff.pos;
+    return result;
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFF_recommendedCInSize(void)  { return ZSTD_CStreamInSize(); }
+size_t ZBUFF_recommendedCOutSize(void) { return ZSTD_CStreamOutSize(); }
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/contrib/python-zstandard/zstd/deprecated/zbuff_decompress.c	Mon Apr 09 10:13:29 2018 -0700
@@ -0,0 +1,75 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+
+/* *************************************
+*  Dependencies
+***************************************/
+#define ZBUFF_STATIC_LINKING_ONLY
+#include "zbuff.h"
+
+
+ZBUFF_DCtx* ZBUFF_createDCtx(void)
+{
+    return ZSTD_createDStream();
+}
+
+ZBUFF_DCtx* ZBUFF_createDCtx_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createDStream_advanced(customMem);
+}
+
+size_t ZBUFF_freeDCtx(ZBUFF_DCtx* zbd)
+{
+    return ZSTD_freeDStream(zbd);
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFF_decompressInitDictionary(ZBUFF_DCtx* zbd, const void* dict, size_t dictSize)
+{
+    return ZSTD_initDStream_usingDict(zbd, dict, dictSize);
+}
+
+size_t ZBUFF_decompressInit(ZBUFF_DCtx* zbd)
+{
+    return ZSTD_initDStream(zbd);
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFF_decompressContinue(ZBUFF_DCtx* zbd,
+                                void* dst, size_t* dstCapacityPtr,
+                          const void* src, size_t* srcSizePtr)
+{
+    ZSTD_outBuffer outBuff;
+    ZSTD_inBuffer inBuff;
+    size_t result;
+    outBuff.dst  = dst;
+    outBuff.pos  = 0;
+    outBuff.size = *dstCapacityPtr;
+    inBuff.src  = src;
+    inBuff.pos  = 0;
+    inBuff.size = *srcSizePtr;
+    result = ZSTD_decompressStream(zbd, &outBuff, &inBuff);
+    *dstCapacityPtr = outBuff.pos;
+    *srcSizePtr = inBuff.pos;
+    return result;
+}
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFF_recommendedDInSize(void)  { return ZSTD_DStreamInSize(); }
+size_t ZBUFF_recommendedDOutSize(void) { return ZSTD_DStreamOutSize(); }
--- a/contrib/python-zstandard/zstd/dictBuilder/cover.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/dictBuilder/cover.c	Mon Apr 09 10:13:29 2018 -0700
@@ -1,12 +1,23 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
+/* *****************************************************************************
+ * Constructs a dictionary using a heuristic based on the following paper:
+ *
+ * Liao, Petri, Moffat, Wirth
+ * Effective Construction of Relative Lempel-Ziv Dictionaries
+ * Published in WWW 2016.
+ *
+ * Adapted from code originally written by @ot (Giuseppe Ottaviano).
+ ******************************************************************************/
+
 /*-*************************************
 *  Dependencies
 ***************************************/
@@ -49,8 +60,6 @@
     if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
       g_time = clock();                                                        \
       DISPLAY(__VA_ARGS__);                                                    \
-      if (displayLevel >= 4)                                                   \
-        fflush(stdout);                                                        \
     }                                                                          \
   }
 #define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
@@ -226,10 +235,22 @@
  * Returns 1 if the dmer at lp is greater than the dmer at rp.
  */
 static int COVER_cmp(COVER_ctx_t *ctx, const void *lp, const void *rp) {
-  const U32 lhs = *(const U32 *)lp;
-  const U32 rhs = *(const U32 *)rp;
+  U32 const lhs = *(U32 const *)lp;
+  U32 const rhs = *(U32 const *)rp;
   return memcmp(ctx->samples + lhs, ctx->samples + rhs, ctx->d);
 }
+/**
+ * Faster version for d <= 8.
+ */
+static int COVER_cmp8(COVER_ctx_t *ctx, const void *lp, const void *rp) {
+  U64 const mask = (ctx->d == 8) ? (U64)-1 : (((U64)1 << (8 * ctx->d)) - 1);
+  U64 const lhs = MEM_readLE64(ctx->samples + *(U32 const *)lp) & mask;
+  U64 const rhs = MEM_readLE64(ctx->samples + *(U32 const *)rp) & mask;
+  if (lhs < rhs) {
+    return -1;
+  }
+  return (lhs > rhs);
+}
 
 /**
  * Same as COVER_cmp() except ties are broken by pointer value
@@ -243,6 +264,16 @@
   }
   return result;
 }
+/**
+ * Faster version for d <= 8.
+ */
+static int COVER_strict_cmp8(const void *lp, const void *rp) {
+  int result = COVER_cmp8(g_ctx, lp, rp);
+  if (result == 0) {
+    result = lp < rp ? -1 : 1;
+  }
+  return result;
+}
 
 /**
  * Returns the first pointer in [first, last) whose element does not compare
@@ -352,7 +383,7 @@
 typedef struct {
   U32 begin;
   U32 end;
-  double score;
+  U32 score;
 } COVER_segment_t;
 
 /**
@@ -368,7 +399,8 @@
  */
 static COVER_segment_t COVER_selectSegment(const COVER_ctx_t *ctx, U32 *freqs,
                                            COVER_map_t *activeDmers, U32 begin,
-                                           U32 end, COVER_params_t parameters) {
+                                           U32 end,
+                                           ZDICT_cover_params_t parameters) {
   /* Constants */
   const U32 k = parameters.k;
   const U32 d = parameters.d;
@@ -448,11 +480,16 @@
  * Check the validity of the parameters.
  * Returns non-zero if the parameters are valid and 0 otherwise.
  */
-static int COVER_checkParameters(COVER_params_t parameters) {
+static int COVER_checkParameters(ZDICT_cover_params_t parameters,
+                                 size_t maxDictSize) {
   /* k and d are required parameters */
   if (parameters.d == 0 || parameters.k == 0) {
     return 0;
   }
+  /* k <= maxDictSize */
+  if (parameters.k > maxDictSize) {
+    return 0;
+  }
   /* d <= k */
   if (parameters.d > parameters.k) {
     return 0;
@@ -498,10 +535,10 @@
   const BYTE *const samples = (const BYTE *)samplesBuffer;
   const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
   /* Checks */
-  if (totalSamplesSize < d ||
+  if (totalSamplesSize < MAX(d, sizeof(U64)) ||
       totalSamplesSize >= (size_t)COVER_MAX_SAMPLES_SIZE) {
-    DISPLAYLEVEL(1, "Total samples size is too large, maximum size is %u MB\n",
-                 (COVER_MAX_SAMPLES_SIZE >> 20));
+    DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
+                 (U32)(totalSamplesSize>>20), (COVER_MAX_SAMPLES_SIZE >> 20));
     return 0;
   }
   /* Zero the context */
@@ -512,7 +549,7 @@
   ctx->samplesSizes = samplesSizes;
   ctx->nbSamples = nbSamples;
   /* Partial suffix array */
-  ctx->suffixSize = totalSamplesSize - d + 1;
+  ctx->suffixSize = totalSamplesSize - MAX(d, sizeof(U64)) + 1;
   ctx->suffix = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
   /* Maps index to the dmerID */
   ctx->dmerAt = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
@@ -546,7 +583,8 @@
     }
     /* qsort doesn't take an opaque pointer, so pass as a global */
     g_ctx = ctx;
-    qsort(ctx->suffix, ctx->suffixSize, sizeof(U32), &COVER_strict_cmp);
+    qsort(ctx->suffix, ctx->suffixSize, sizeof(U32),
+          (ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
   }
   DISPLAYLEVEL(2, "Computing frequencies\n");
   /* For each dmer group (group of positions with the same first d bytes):
@@ -556,8 +594,8 @@
    * 2. We calculate how many samples the dmer occurs in and save it in
    *    freqs[dmerId].
    */
-  COVER_groupBy(ctx->suffix, ctx->suffixSize, sizeof(U32), ctx, &COVER_cmp,
-                &COVER_group);
+  COVER_groupBy(ctx->suffix, ctx->suffixSize, sizeof(U32), ctx,
+                (ctx->d <= 8 ? &COVER_cmp8 : &COVER_cmp), &COVER_group);
   ctx->freqs = ctx->suffix;
   ctx->suffix = NULL;
   return 1;
@@ -569,7 +607,7 @@
 static size_t COVER_buildDictionary(const COVER_ctx_t *ctx, U32 *freqs,
                                     COVER_map_t *activeDmers, void *dictBuffer,
                                     size_t dictBufferCapacity,
-                                    COVER_params_t parameters) {
+                                    ZDICT_cover_params_t parameters) {
   BYTE *const dict = (BYTE *)dictBuffer;
   size_t tail = dictBufferCapacity;
   /* Divide the data up into epochs of equal size.
@@ -590,9 +628,13 @@
     /* Select a segment */
     COVER_segment_t segment = COVER_selectSegment(
         ctx, freqs, activeDmers, epochBegin, epochEnd, parameters);
-    /* Trim the segment if necessary and if it is empty then we are done */
+    /* If the segment covers no dmers, then we are out of content */
+    if (segment.score == 0) {
+      break;
+    }
+    /* Trim the segment if necessary and if it is too small then we are done */
     segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
-    if (segmentSize == 0) {
+    if (segmentSize < parameters.d) {
       break;
     }
     /* We fill the dictionary from the back to allow the best segments to be
@@ -608,34 +650,19 @@
   return tail;
 }
 
-/**
- * Translate from COVER_params_t to ZDICT_params_t required for finalizing the
- * dictionary.
- */
-static ZDICT_params_t COVER_translateParams(COVER_params_t parameters) {
-  ZDICT_params_t zdictParams;
-  memset(&zdictParams, 0, sizeof(zdictParams));
-  zdictParams.notificationLevel = 1;
-  zdictParams.dictID = parameters.dictID;
-  zdictParams.compressionLevel = parameters.compressionLevel;
-  return zdictParams;
-}
-
-/**
- * Constructs a dictionary using a heuristic based on the following paper:
- *
- * Liao, Petri, Moffat, Wirth
- * Effective Construction of Relative Lempel-Ziv Dictionaries
- * Published in WWW 2016.
- */
-ZDICTLIB_API size_t COVER_trainFromBuffer(
-    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
-    const size_t *samplesSizes, unsigned nbSamples, COVER_params_t parameters) {
-  BYTE *const dict = (BYTE *)dictBuffer;
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
+    void *dictBuffer, size_t dictBufferCapacity,
+    const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
+    ZDICT_cover_params_t parameters)
+{
+  BYTE* const dict = (BYTE*)dictBuffer;
   COVER_ctx_t ctx;
   COVER_map_t activeDmers;
+
+  /* Initialize global data */
+  g_displayLevel = parameters.zParams.notificationLevel;
   /* Checks */
-  if (!COVER_checkParameters(parameters)) {
+  if (!COVER_checkParameters(parameters, dictBufferCapacity)) {
     DISPLAYLEVEL(1, "Cover parameters incorrect\n");
     return ERROR(GENERIC);
   }
@@ -648,8 +675,6 @@
                  ZDICT_DICTSIZE_MIN);
     return ERROR(dstSize_tooSmall);
   }
-  /* Initialize global data */
-  g_displayLevel = parameters.notificationLevel;
   /* Initialize context and activeDmers */
   if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
                       parameters.d)) {
@@ -666,10 +691,9 @@
     const size_t tail =
         COVER_buildDictionary(&ctx, ctx.freqs, &activeDmers, dictBuffer,
                               dictBufferCapacity, parameters);
-    ZDICT_params_t zdictParams = COVER_translateParams(parameters);
     const size_t dictionarySize = ZDICT_finalizeDictionary(
         dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
-        samplesBuffer, samplesSizes, nbSamples, zdictParams);
+        samplesBuffer, samplesSizes, nbSamples, parameters.zParams);
     if (!ZSTD_isError(dictionarySize)) {
       DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
                    (U32)dictionarySize);
@@ -689,12 +713,12 @@
  * compiled with multithreaded support.
  */
 typedef struct COVER_best_s {
-  pthread_mutex_t mutex;
-  pthread_cond_t cond;
+  ZSTD_pthread_mutex_t mutex;
+  ZSTD_pthread_cond_t cond;
   size_t liveJobs;
   void *dict;
   size_t dictSize;
-  COVER_params_t parameters;
+  ZDICT_cover_params_t parameters;
   size_t compressedSize;
 } COVER_best_t;
 
@@ -702,11 +726,9 @@
  * Initialize the `COVER_best_t`.
  */
 static void COVER_best_init(COVER_best_t *best) {
-  if (!best) {
-    return;
-  }
-  pthread_mutex_init(&best->mutex, NULL);
-  pthread_cond_init(&best->cond, NULL);
+  if (best==NULL) return; /* compatible with init on NULL */
+  (void)ZSTD_pthread_mutex_init(&best->mutex, NULL);
+  (void)ZSTD_pthread_cond_init(&best->cond, NULL);
   best->liveJobs = 0;
   best->dict = NULL;
   best->dictSize = 0;
@@ -721,11 +743,11 @@
   if (!best) {
     return;
   }
-  pthread_mutex_lock(&best->mutex);
+  ZSTD_pthread_mutex_lock(&best->mutex);
   while (best->liveJobs != 0) {
-    pthread_cond_wait(&best->cond, &best->mutex);
+    ZSTD_pthread_cond_wait(&best->cond, &best->mutex);
   }
-  pthread_mutex_unlock(&best->mutex);
+  ZSTD_pthread_mutex_unlock(&best->mutex);
 }
 
 /**
@@ -739,8 +761,8 @@
   if (best->dict) {
     free(best->dict);
   }
-  pthread_mutex_destroy(&best->mutex);
-  pthread_cond_destroy(&best->cond);
+  ZSTD_pthread_mutex_destroy(&best->mutex);
+  ZSTD_pthread_cond_destroy(&best->cond);
 }
 
 /**
@@ -751,9 +773,9 @@
   if (!best) {
     return;
   }
-  pthread_mutex_lock(&best->mutex);
+  ZSTD_pthread_mutex_lock(&best->mutex);
   ++best->liveJobs;
-  pthread_mutex_unlock(&best->mutex);
+  ZSTD_pthread_mutex_unlock(&best->mutex);
 }
 
 /**
@@ -762,14 +784,14 @@
  * If this dictionary is the best so far save it and its parameters.
  */
 static void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
-                              COVER_params_t parameters, void *dict,
+                              ZDICT_cover_params_t parameters, void *dict,
                               size_t dictSize) {
   if (!best) {
     return;
   }
   {
     size_t liveJobs;
-    pthread_mutex_lock(&best->mutex);
+    ZSTD_pthread_mutex_lock(&best->mutex);
     --best->liveJobs;
     liveJobs = best->liveJobs;
     /* If the new dictionary is better */
@@ -792,9 +814,9 @@
       best->parameters = parameters;
       best->compressedSize = compressedSize;
     }
-    pthread_mutex_unlock(&best->mutex);
+    ZSTD_pthread_mutex_unlock(&best->mutex);
     if (liveJobs == 0) {
-      pthread_cond_broadcast(&best->cond);
+      ZSTD_pthread_cond_broadcast(&best->cond);
     }
   }
 }
@@ -806,7 +828,7 @@
   const COVER_ctx_t *ctx;
   COVER_best_t *best;
   size_t dictBufferCapacity;
-  COVER_params_t parameters;
+  ZDICT_cover_params_t parameters;
 } COVER_tryParameters_data_t;
 
 /**
@@ -818,7 +840,7 @@
   /* Save parameters as local variables */
   COVER_tryParameters_data_t *const data = (COVER_tryParameters_data_t *)opaque;
   const COVER_ctx_t *const ctx = data->ctx;
-  const COVER_params_t parameters = data->parameters;
+  const ZDICT_cover_params_t parameters = data->parameters;
   size_t dictBufferCapacity = data->dictBufferCapacity;
   size_t totalCompressedSize = ERROR(GENERIC);
   /* Allocate space for hash table, dict, and freqs */
@@ -839,10 +861,10 @@
   {
     const size_t tail = COVER_buildDictionary(ctx, freqs, &activeDmers, dict,
                                               dictBufferCapacity, parameters);
-    const ZDICT_params_t zdictParams = COVER_translateParams(parameters);
     dictBufferCapacity = ZDICT_finalizeDictionary(
         dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
-        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbSamples, zdictParams);
+        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbSamples,
+        parameters.zParams);
     if (ZDICT_isError(dictBufferCapacity)) {
       DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
       goto _cleanup;
@@ -868,13 +890,13 @@
     }
     /* Create the cctx and cdict */
     cctx = ZSTD_createCCtx();
-    cdict =
-        ZSTD_createCDict(dict, dictBufferCapacity, parameters.compressionLevel);
+    cdict = ZSTD_createCDict(dict, dictBufferCapacity,
+                             parameters.zParams.compressionLevel);
     if (!dst || !cctx || !cdict) {
       goto _compressCleanup;
     }
     /* Compress each sample and sum their sizes (or error) */
-    totalCompressedSize = 0;
+    totalCompressedSize = dictBufferCapacity;
     for (i = 0; i < ctx->nbSamples; ++i) {
       const size_t size = ZSTD_compress_usingCDict(
           cctx, dst, dstCapacity, ctx->samples + ctx->offsets[i],
@@ -906,29 +928,28 @@
   }
 }
 
-ZDICTLIB_API size_t COVER_optimizeTrainFromBuffer(void *dictBuffer,
-                                                  size_t dictBufferCapacity,
-                                                  const void *samplesBuffer,
-                                                  const size_t *samplesSizes,
-                                                  unsigned nbSamples,
-                                                  COVER_params_t *parameters) {
+ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
+    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
+    const size_t *samplesSizes, unsigned nbSamples,
+    ZDICT_cover_params_t *parameters) {
   /* constants */
   const unsigned nbThreads = parameters->nbThreads;
   const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
-  const unsigned kMaxD = parameters->d == 0 ? 16 : parameters->d;
-  const unsigned kMinK = parameters->k == 0 ? kMaxD : parameters->k;
-  const unsigned kMaxK = parameters->k == 0 ? 2048 : parameters->k;
-  const unsigned kSteps = parameters->steps == 0 ? 32 : parameters->steps;
+  const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
+  const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
+  const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
+  const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
   const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
   const unsigned kIterations =
       (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
   /* Local variables */
-  const int displayLevel = parameters->notificationLevel;
+  const int displayLevel = parameters->zParams.notificationLevel;
   unsigned iteration = 1;
   unsigned d;
   unsigned k;
   COVER_best_t best;
   POOL_ctx *pool = NULL;
+
   /* Checks */
   if (kMinK < kMaxD || kMaxK < kMinK) {
     LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
@@ -952,7 +973,7 @@
   /* Initialization */
   COVER_best_init(&best);
   /* Turn down global display level to clean up display at level 2 and below */
-  g_displayLevel = parameters->notificationLevel - 1;
+  g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
   /* Loop through d first because each new value needs a new context */
   LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
                     kIterations);
@@ -963,6 +984,7 @@
     if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d)) {
       LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
       COVER_best_destroy(&best);
+      POOL_free(pool);
       return ERROR(GENERIC);
     }
     /* Loop through k reusing the same context */
@@ -975,6 +997,7 @@
         LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
         COVER_best_destroy(&best);
         COVER_ctx_destroy(&ctx);
+        POOL_free(pool);
         return ERROR(GENERIC);
       }
       data->ctx = &ctx;
@@ -984,9 +1007,11 @@
       data->parameters.k = k;
       data->parameters.d = d;
       data->parameters.steps = kSteps;
+      data->parameters.zParams.notificationLevel = g_displayLevel;
       /* Check the parameters */
-      if (!COVER_checkParameters(data->parameters)) {
+      if (!COVER_checkParameters(data->parameters, dictBufferCapacity)) {
         DISPLAYLEVEL(1, "Cover parameters incorrect\n");
+        free(data);
         continue;
       }
       /* Call the function and pass ownership of data to it */
@@ -1009,8 +1034,10 @@
   {
     const size_t dictSize = best.dictSize;
     if (ZSTD_isError(best.compressedSize)) {
+      const size_t compressedSize = best.compressedSize;
       COVER_best_destroy(&best);
-      return best.compressedSize;
+      POOL_free(pool);
+      return compressedSize;
     }
     *parameters = best.parameters;
     memcpy(dictBuffer, best.dict, dictSize);
--- a/contrib/python-zstandard/zstd/dictBuilder/zdict.c	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/dictBuilder/zdict.c	Mon Apr 09 10:13:29 2018 -0700
@@ -1,18 +1,20 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 
 /*-**************************************
 *  Tuning parameters
 ****************************************/
+#define MINRATIO 4   /* minimum nb of apparition to be selected in dictionary */
 #define ZDICT_MAX_SAMPLES_SIZE (2000U << 20)
-#define ZDICT_MIN_SAMPLES_SIZE 512
+#define ZDICT_MIN_SAMPLES_SIZE (ZDICT_CONTENTSIZE_MIN * MINRATIO)
 
 
 /*-**************************************
@@ -59,11 +61,8 @@
 
 #define NOISELENGTH 32
 
-#define MINRATIO 4
-static const int g_compressionLevel_default = 6;
+static const int g_compressionLevel_default = 3;
 static const U32 g_selectivity_default = 9;
-static const size_t g_provision_entropySize = 200;
-static const size_t g_min_fast_dictContent = 192;
 
 
 /*-*************************************
@@ -96,7 +95,7 @@
 unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize)
 {
     if (dictSize < 8) return 0;
-    if (MEM_readLE32(dictBuffer) != ZSTD_DICT_MAGIC) return 0;
+    if (MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return 0;
     return MEM_readLE32((const char*)dictBuffer + 4);
 }
 
@@ -104,7 +103,7 @@
 /*-********************************************************
 *  Dictionary training functions
 **********************************************************/
-static unsigned ZDICT_NbCommonBytes (register size_t val)
+static unsigned ZDICT_NbCommonBytes (size_t val)
 {
     if (MEM_isLittleEndian()) {
         if (MEM_64bits()) {
@@ -208,7 +207,6 @@
     U32 cumulLength[LLIMIT] = {0};
     U32 savings[LLIMIT] = {0};
     const BYTE* b = (const BYTE*)buffer;
-    size_t length;
     size_t maxLength = LLIMIT;
     size_t pos = suffix[start];
     U32 end = start;
@@ -223,26 +221,30 @@
        ||(MEM_read16(b+pos+1) == MEM_read16(b+pos+3))
        ||(MEM_read16(b+pos+2) == MEM_read16(b+pos+4)) ) {
         /* skip and mark segment */
-        U16 u16 = MEM_read16(b+pos+4);
-        U32 u, e = 6;
-        while (MEM_read16(b+pos+e) == u16) e+=2 ;
-        if (b[pos+e] == b[pos+e-1]) e++;
-        for (u=1; u<e; u++)
+        U16 const pattern16 = MEM_read16(b+pos+4);
+        U32 u, patternEnd = 6;
+        while (MEM_read16(b+pos+patternEnd) == pattern16) patternEnd+=2 ;
+        if (b[pos+patternEnd] == b[pos+patternEnd-1]) patternEnd++;
+        for (u=1; u<patternEnd; u++)
             doneMarks[pos+u] = 1;
         return solution;
     }
 
     /* look forward */
-    do {
-        end++;
-        length = ZDICT_count(b + pos, b + suffix[end]);
-    } while (length >=MINMATCHLENGTH);
+    {   size_t length;
+        do {
+            end++;
+            length = ZDICT_count(b + pos, b + suffix[end]);
+        } while (length >= MINMATCHLENGTH);
+    }
 
     /* look backward */
-    do {
-        length = ZDICT_count(b + pos, b + *(suffix+start-1));
-        if (length >=MINMATCHLENGTH) start--;
-    } while(length >= MINMATCHLENGTH);
+    {   size_t length;
+        do {
+            length = ZDICT_count(b + pos, b + *(suffix+start-1));
+            if (length >=MINMATCHLENGTH) start--;
+        } while(length >= MINMATCHLENGTH);
+    }
 
     /* exit if not found a minimum nb of repetitions */
     if (end-start < minRatio) {
@@ -269,7 +271,7 @@
             U32 selectedCount = 0;
             U32 selectedID = currentID;
             for (id =refinedStart; id < refinedEnd; id++) {
-                if (b[ suffix[id] + searchLength] != currentChar) {
+                if (b[suffix[id] + searchLength] != currentChar) {
                     if (currentCount > selectedCount) {
                         selectedCount = currentCount;
                         selectedID = currentID;
@@ -298,20 +300,23 @@
         memset(lengthList, 0, sizeof(lengthList));
 
         /* look forward */
-        do {
-            end++;
-            length = ZDICT_count(b + pos, b + suffix[end]);
-            if (length >= LLIMIT) length = LLIMIT-1;
-            lengthList[length]++;
-        } while (length >=MINMATCHLENGTH);
+        {   size_t length;
+            do {
+                end++;
+                length = ZDICT_count(b + pos, b + suffix[end]);
+                if (length >= LLIMIT) length = LLIMIT-1;
+                lengthList[length]++;
+            } while (length >=MINMATCHLENGTH);
+        }
 
         /* look backward */
-        length = MINMATCHLENGTH;
-        while ((length >= MINMATCHLENGTH) & (start > 0)) {
-        	length = ZDICT_count(b + pos, b + suffix[start - 1]);
-        	if (length >= LLIMIT) length = LLIMIT - 1;
-        	lengthList[length]++;
-        	if (length >= MINMATCHLENGTH) start--;
+        {   size_t length = MINMATCHLENGTH;
+            while ((length >= MINMATCHLENGTH) & (start > 0)) {
+                length = ZDICT_count(b + pos, b + suffix[start - 1]);
+                if (length >= LLIMIT) length = LLIMIT - 1;
+                lengthList[length]++;
+                if (length >= MINMATCHLENGTH) start--;
+            }
         }
 
         /* largest useful length */
@@ -346,12 +351,12 @@
         /* mark positions done */
         {   U32 id;
             for (id=start; id<end; id++) {
-                U32 p, pEnd;
+                U32 p, pEnd, length;
                 U32 const testedPos = suffix[id];
                 if (testedPos == pos)
                     length = solution.length;
                 else {
-                    length = ZDICT_count(b+pos, b+testedPos);
+                    length = (U32)ZDICT_count(b+pos, b+testedPos);
                     if (length > solution.length) length = solution.length;
                 }
                 pEnd = (U32)(testedPos + length);
@@ -363,21 +368,35 @@
 }
 
 
-/*! ZDICT_checkMerge
+static int isIncluded(const void* in, const void* container, size_t length)
+{
+    const char* const ip = (const char*) in;
+    const char* const into = (const char*) container;
+    size_t u;
+
+    for (u=0; u<length; u++) {  /* works because end of buffer is a noisy guard band */
+        if (ip[u] != into[u]) break;
+    }
+
+    return u==length;
+}
+
+/*! ZDICT_tryMerge() :
     check if dictItem can be merged, do it if possible
     @return : id of destination elt, 0 if not merged
 */
-static U32 ZDICT_checkMerge(dictItem* table, dictItem elt, U32 eltNbToSkip)
+static U32 ZDICT_tryMerge(dictItem* table, dictItem elt, U32 eltNbToSkip, const void* buffer)
 {
     const U32 tableSize = table->pos;
     const U32 eltEnd = elt.pos + elt.length;
+    const char* const buf = (const char*) buffer;
 
     /* tail overlap */
     U32 u; for (u=1; u<tableSize; u++) {
         if (u==eltNbToSkip) continue;
         if ((table[u].pos > elt.pos) && (table[u].pos <= eltEnd)) {  /* overlap, existing > new */
             /* append */
-            U32 addedLength = table[u].pos - elt.pos;
+            U32 const addedLength = table[u].pos - elt.pos;
             table[u].length += addedLength;
             table[u].pos = elt.pos;
             table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
@@ -393,9 +412,10 @@
     /* front overlap */
     for (u=1; u<tableSize; u++) {
         if (u==eltNbToSkip) continue;
+
         if ((table[u].pos + table[u].length >= elt.pos) && (table[u].pos < elt.pos)) {  /* overlap, existing < new */
             /* append */
-            int addedLength = (int)eltEnd - (table[u].pos + table[u].length);
+            int const addedLength = (int)eltEnd - (table[u].pos + table[u].length);
             table[u].savings += elt.length / 8;    /* rough approx bonus */
             if (addedLength > 0) {   /* otherwise, elt fully included into existing */
                 table[u].length += addedLength;
@@ -407,7 +427,18 @@
                 table[u] = table[u-1], u--;
             table[u] = elt;
             return u;
-    }   }
+        }
+
+        if (MEM_read64(buf + table[u].pos) == MEM_read64(buf + elt.pos + 1)) {
+            if (isIncluded(buf + table[u].pos, buf + elt.pos + 1, table[u].length)) {
+                size_t const addedLength = MAX( (int)elt.length - (int)table[u].length , 1 );
+                table[u].pos = elt.pos;
+                table[u].savings += (U32)(elt.savings * addedLength / elt.length);
+                table[u].length = MIN(elt.length, table[u].length + 1);
+                return u;
+            }
+        }
+    }
 
     return 0;
 }
@@ -415,8 +446,8 @@
 
 static void ZDICT_removeDictItem(dictItem* table, U32 id)
 {
-    /* convention : first element is nb of elts */
-    U32 const max = table->pos;
+    /* convention : table[0].pos stores nb of elts */
+    U32 const max = table[0].pos;
     U32 u;
     if (!id) return;   /* protection, should never happen */
     for (u=id; u<max-1; u++)
@@ -425,14 +456,14 @@
 }
 
 
-static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt)
+static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt, const void* buffer)
 {
     /* merge if possible */
-    U32 mergeId = ZDICT_checkMerge(table, elt, 0);
+    U32 mergeId = ZDICT_tryMerge(table, elt, 0, buffer);
     if (mergeId) {
         U32 newMerge = 1;
         while (newMerge) {
-            newMerge = ZDICT_checkMerge(table, table[mergeId], mergeId);
+            newMerge = ZDICT_tryMerge(table, table[mergeId], mergeId, buffer);
             if (newMerge) ZDICT_removeDictItem(table, mergeId);
             mergeId = newMerge;
         }
@@ -463,7 +494,7 @@
 }
 
 
-static size_t ZDICT_trainBuffer(dictItem* dictList, U32 dictListSize,
+static size_t ZDICT_trainBuffer_legacy(dictItem* dictList, U32 dictListSize,
                             const void* const buffer, size_t bufferSize,   /* buffer must end with noisy guard band */
                             const size_t* fileSizes, unsigned nbFiles,
                             U32 minRatio, U32 notificationLevel)
@@ -480,7 +511,7 @@
 #   define DISPLAYUPDATE(l, ...) if (notificationLevel>=l) { \
             if (ZDICT_clockSpan(displayClock) > refreshRate)  \
             { displayClock = clock(); DISPLAY(__VA_ARGS__); \
-            if (notificationLevel>=4) fflush(stdout); } }
+            if (notificationLevel>=4) fflush(stderr); } }
 
     /* init */
     DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
@@ -521,7 +552,7 @@
             if (doneMarks[cursor]) { cursor++; continue; }
             solution = ZDICT_analyzePos(doneMarks, suffix, reverseSuffix[cursor], buffer, minRatio, notificationLevel);
             if (solution.length==0) { cursor++; continue; }
-            ZDICT_insertDictItem(dictList, dictListSize, solution);
+            ZDICT_insertDictItem(dictList, dictListSize, solution, buffer);
             cursor += solution.length;
             DISPLAYUPDATE(2, "\r%4.2f %% \r", (double)cursor / bufferSize * 100);
     }   }
@@ -550,29 +581,30 @@
 
 typedef struct
 {
-    ZSTD_CCtx* ref;
-    ZSTD_CCtx* zc;
-    void* workPlace;   /* must be ZSTD_BLOCKSIZE_ABSOLUTEMAX allocated */
+    ZSTD_CCtx* ref;    /* contains reference to dictionary */
+    ZSTD_CCtx* zc;     /* working context */
+    void* workPlace;   /* must be ZSTD_BLOCKSIZE_MAX allocated */
 } EStats_ress_t;
 
 #define MAXREPOFFSET 1024
 
 static void ZDICT_countEStats(EStats_ress_t esr, ZSTD_parameters params,
-                            U32* countLit, U32* offsetcodeCount, U32* matchlengthCount, U32* litlengthCount, U32* repOffsets,
-                            const void* src, size_t srcSize, U32 notificationLevel)
+                              U32* countLit, U32* offsetcodeCount, U32* matchlengthCount, U32* litlengthCount, U32* repOffsets,
+                              const void* src, size_t srcSize,
+                              U32 notificationLevel)
 {
-    size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_ABSOLUTEMAX, 1 << params.cParams.windowLog);
+    size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_MAX, 1 << params.cParams.windowLog);
     size_t cSize;
 
     if (srcSize > blockSizeMax) srcSize = blockSizeMax;   /* protection vs large samples */
-    {  size_t const errorCode = ZSTD_copyCCtx(esr.zc, esr.ref, 0);
-            if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_copyCCtx failed \n"); return; }
+    {   size_t const errorCode = ZSTD_copyCCtx(esr.zc, esr.ref, 0);
+        if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_copyCCtx failed \n"); return; }
     }
-    cSize = ZSTD_compressBlock(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_ABSOLUTEMAX, src, srcSize);
+    cSize = ZSTD_compressBlock(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_MAX, src, srcSize);
     if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (U32)srcSize); return; }
 
     if (cSize) {  /* if == 0; block is not compressible */
-        const seqStore_t* seqStorePtr = ZSTD_getSeqStore(esr.zc);
+        const seqStore_t* const seqStorePtr = ZSTD_getSeqStore(esr.zc);
 
         /* literals stats */
         {   const BYTE* bytePtr;
@@ -610,17 +642,6 @@
     }   }   }
 }
 
-/*
-static size_t ZDICT_maxSampleSize(const size_t* fileSizes, unsigned nbFiles)
-{
-    unsigned u;
-    size_t max=0;
-    for (u=0; u<nbFiles; u++)
-        if (max < fileSizes[u]) max = fileSizes[u];
-    return max;
-}
-*/
-
 static size_t ZDICT_totalSampleSize(const size_t* fileSizes, unsigned nbFiles)
 {
     size_t total=0;
@@ -645,6 +666,18 @@
     }
 }
 
+/* ZDICT_flatLit() :
+ * rewrite `countLit` to contain a mostly flat but still compressible distribution of literals.
+ * necessary to avoid generating a non-compressible distribution that HUF_writeCTable() cannot encode.
+ */
+static void ZDICT_flatLit(U32* countLit)
+{
+    int u;
+    for (u=1; u<256; u++) countLit[u] = 2;
+    countLit[0]   = 4;
+    countLit[253] = 1;
+    countLit[254] = 1;
+}
 
 #define OFFCODE_MAX 30  /* only applicable to first block */
 static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
@@ -674,32 +707,33 @@
     BYTE* dstPtr = (BYTE*)dstBuffer;
 
     /* init */
+    DEBUGLOG(4, "ZDICT_analyzeEntropy");
     esr.ref = ZSTD_createCCtx();
     esr.zc = ZSTD_createCCtx();
-    esr.workPlace = malloc(ZSTD_BLOCKSIZE_ABSOLUTEMAX);
+    esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX);
     if (!esr.ref || !esr.zc || !esr.workPlace) {
         eSize = ERROR(memory_allocation);
         DISPLAYLEVEL(1, "Not enough memory \n");
         goto _cleanup;
     }
-    if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionary_wrong); goto _cleanup; }   /* too large dictionary */
-    for (u=0; u<256; u++) countLit[u]=1;   /* any character must be described */
-    for (u=0; u<=offcodeMax; u++) offcodeCount[u]=1;
-    for (u=0; u<=MaxML; u++) matchLengthCount[u]=1;
-    for (u=0; u<=MaxLL; u++) litLengthCount[u]=1;
+    if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionaryCreation_failed); goto _cleanup; }   /* too large dictionary */
+    for (u=0; u<256; u++) countLit[u] = 1;   /* any character must be described */
+    for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1;
+    for (u=0; u<=MaxML; u++) matchLengthCount[u] = 1;
+    for (u=0; u<=MaxLL; u++) litLengthCount[u] = 1;
     memset(repOffset, 0, sizeof(repOffset));
     repOffset[1] = repOffset[4] = repOffset[8] = 1;
     memset(bestRepOffset, 0, sizeof(bestRepOffset));
-    if (compressionLevel==0) compressionLevel=g_compressionLevel_default;
+    if (compressionLevel<=0) compressionLevel = g_compressionLevel_default;
     params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
     {   size_t const beginResult = ZSTD_compressBegin_advanced(esr.ref, dictBuffer, dictBufferSize, params, 0);
-            if (ZSTD_isError(beginResult)) {
+        if (ZSTD_isError(beginResult)) {
+            DISPLAYLEVEL(1, "error : ZSTD_compressBegin_advanced() failed : %s \n", ZSTD_getErrorName(beginResult));
             eSize = ERROR(GENERIC);
-            DISPLAYLEVEL(1, "error : ZSTD_compressBegin_advanced failed \n");
             goto _cleanup;
     }   }
 
-    /* collect stats on all files */
+    /* collect stats on all samples */
     for (u=0; u<nbFiles; u++) {
         ZDICT_countEStats(esr, params,
                           countLit, offcodeCount, matchLengthCount, litLengthCount, repOffset,
@@ -708,14 +742,21 @@
         pos += fileSizes[u];
     }
 
-    /* analyze */
-    errorCode = HUF_buildCTable (hufTable, countLit, 255, huffLog);
-    if (HUF_isError(errorCode)) {
-        eSize = ERROR(GENERIC);
-        DISPLAYLEVEL(1, "HUF_buildCTable error \n");
-        goto _cleanup;
+    /* analyze, build stats, starting with literals */
+    {   size_t maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
+        if (HUF_isError(maxNbBits)) {
+            eSize = ERROR(GENERIC);
+            DISPLAYLEVEL(1, " HUF_buildCTable error \n");
+            goto _cleanup;
+        }
+        if (maxNbBits==8) {  /* not compressible : will fail on HUF_writeCTable() */
+            DISPLAYLEVEL(2, "warning : pathological dataset : literals are not compressible : samples are noisy or too regular \n");
+            ZDICT_flatLit(countLit);  /* replace distribution by a fake "mostly flat but still compressible" distribution, that HUF_writeCTable() can encode */
+            maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
+            assert(maxNbBits==9);
+        }
+        huffLog = (U32)maxNbBits;
     }
-    huffLog = (U32)errorCode;
 
     /* looking for most common first offsets */
     {   U32 offset;
@@ -812,7 +853,6 @@
     MEM_writeLE32(dstPtr+4, repStartValue[1]);
     MEM_writeLE32(dstPtr+8, repStartValue[2]);
 #endif
-    //dstPtr += 12;
     eSize += 12;
 
 _cleanup:
@@ -831,18 +871,19 @@
                           ZDICT_params_t params)
 {
     size_t hSize;
-#define HBUFFSIZE 256
+#define HBUFFSIZE 256   /* should prove large enough for all entropy headers */
     BYTE header[HBUFFSIZE];
     int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
     U32 const notificationLevel = params.notificationLevel;
 
     /* check conditions */
+    DEBUGLOG(4, "ZDICT_finalizeDictionary");
     if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
     if (dictContentSize < ZDICT_CONTENTSIZE_MIN) return ERROR(srcSize_wrong);
     if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
 
     /* dictionary header */
-    MEM_writeLE32(header, ZSTD_DICT_MAGIC);
+    MEM_writeLE32(header, ZSTD_MAGIC_DICTIONARY);
     {   U64 const randomID = XXH64(customDictContent, dictContentSize, 0);
         U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
         U32 const dictID = params.dictID ? params.dictID : compliantID;
@@ -877,20 +918,11 @@
                                                  const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
                                                  ZDICT_params_t params)
 {
-    size_t hSize;
     int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
     U32 const notificationLevel = params.notificationLevel;
+    size_t hSize = 8;
 
-    /* dictionary header */
-    MEM_writeLE32(dictBuffer, ZSTD_DICT_MAGIC);
-    {   U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
-        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
-        U32 const dictID = params.dictID ? params.dictID : compliantID;
-        MEM_writeLE32((char*)dictBuffer+4, dictID);
-    }
-    hSize = 8;
-
-    /* entropy tables */
+    /* calculate entropy tables */
     DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
     DISPLAYLEVEL(2, "statistics ... \n");
     {   size_t const eSize = ZDICT_analyzeEntropy((char*)dictBuffer+hSize, dictBufferCapacity-hSize,
@@ -902,6 +934,13 @@
         hSize += eSize;
     }
 
+    /* add dictionary header (after entropy tables) */
+    MEM_writeLE32(dictBuffer, ZSTD_MAGIC_DICTIONARY);
+    {   U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
+        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
+        U32 const dictID = params.dictID ? params.dictID : compliantID;
+        MEM_writeLE32((char*)dictBuffer+4, dictID);
+    }
 
     if (hSize + dictContentSize < dictBufferCapacity)
         memmove((char*)dictBuffer + hSize, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
@@ -909,14 +948,14 @@
 }
 
 
-/*! ZDICT_trainFromBuffer_unsafe() :
+/*! ZDICT_trainFromBuffer_unsafe_legacy() :
 *   Warning : `samplesBuffer` must be followed by noisy guard band.
 *   @return : size of dictionary, or an error code which can be tested with ZDICT_isError()
 */
-size_t ZDICT_trainFromBuffer_unsafe(
+size_t ZDICT_trainFromBuffer_unsafe_legacy(
                             void* dictBuffer, size_t maxDictSize,
                             const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                            ZDICT_params_t params)
+                            ZDICT_legacy_params_t params)
 {
     U32 const dictListSize = MAX(MAX(DICTLISTSIZE_DEFAULT, nbSamples), (U32)(maxDictSize/16));
     dictItem* const dictList = (dictItem*)malloc(dictListSize * sizeof(*dictList));
@@ -925,24 +964,24 @@
     size_t const targetDictSize = maxDictSize;
     size_t const samplesBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
     size_t dictSize = 0;
-    U32 const notificationLevel = params.notificationLevel;
+    U32 const notificationLevel = params.zParams.notificationLevel;
 
     /* checks */
     if (!dictList) return ERROR(memory_allocation);
-    if (maxDictSize <= g_provision_entropySize + g_min_fast_dictContent) { free(dictList); return ERROR(dstSize_tooSmall); }
-    if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return 0; }   /* not enough source to create dictionary */
+    if (maxDictSize < ZDICT_DICTSIZE_MIN) { free(dictList); return ERROR(dstSize_tooSmall); }   /* requested dictionary size is too small */
+    if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* not enough source to create dictionary */
 
     /* init */
     ZDICT_initDictItem(dictList);
 
     /* build dictionary */
-    ZDICT_trainBuffer(dictList, dictListSize,
-                    samplesBuffer, samplesBuffSize,
-                    samplesSizes, nbSamples,
-                    minRep, notificationLevel);
+    ZDICT_trainBuffer_legacy(dictList, dictListSize,
+                       samplesBuffer, samplesBuffSize,
+                       samplesSizes, nbSamples,
+                       minRep, notificationLevel);
 
     /* display best matches */
-    if (params.notificationLevel>= 3) {
+    if (params.zParams.notificationLevel>= 3) {
         U32 const nb = MIN(25, dictList[0].pos);
         U32 const dictContentSize = ZDICT_dictSize(dictList);
         U32 u;
@@ -963,14 +1002,15 @@
 
     /* create dictionary */
     {   U32 dictContentSize = ZDICT_dictSize(dictList);
-        if (dictContentSize < targetDictSize/3) {
+        if (dictContentSize < ZDICT_CONTENTSIZE_MIN) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* dictionary content too small */
+        if (dictContentSize < targetDictSize/4) {
             DISPLAYLEVEL(2, "!  warning : selected content significantly smaller than requested (%u < %u) \n", dictContentSize, (U32)maxDictSize);
+            if (samplesBuffSize < 10 * targetDictSize)
+                DISPLAYLEVEL(2, "!  consider increasing the number of samples (total size : %u MB)\n", (U32)(samplesBuffSize>>20));
             if (minRep > MINRATIO) {
                 DISPLAYLEVEL(2, "!  consider increasing selectivity to produce larger dictionary (-s%u) \n", selectivity+1);
                 DISPLAYLEVEL(2, "!  note : larger dictionaries are not necessarily better, test its efficiency on samples \n");
             }
-            if (samplesBuffSize < 10 * targetDictSize)
-                DISPLAYLEVEL(2, "!  consider increasing the number of samples (total size : %u MB)\n", (U32)(samplesBuffSize>>20));
         }
 
         if ((dictContentSize > targetDictSize*3) && (nbSamples > 2*MINRATIO) && (selectivity>1)) {
@@ -978,7 +1018,7 @@
             while ((nbSamples >> proposedSelectivity) <= MINRATIO) { proposedSelectivity--; }
             DISPLAYLEVEL(2, "!  note : calculated dictionary significantly larger than requested (%u > %u) \n", dictContentSize, (U32)maxDictSize);
             DISPLAYLEVEL(2, "!  consider increasing dictionary size, or produce denser dictionary (-s%u) \n", proposedSelectivity);
-            DISPLAYLEVEL(2, "!  always test dictionary efficiency on samples \n");
+            DISPLAYLEVEL(2, "!  always test dictionary efficiency on real samples \n");
         }
 
         /* limit dictionary size */
@@ -1004,7 +1044,7 @@
 
         dictSize = ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, maxDictSize,
                                                              samplesBuffer, samplesSizes, nbSamples,
-                                                             params);
+                                                             params.zParams);
     }
 
     /* clean up */
@@ -1013,11 +1053,12 @@
 }
 
 
-/* issue : samplesBuffer need to be followed by a noisy guard band.
-*  work around : duplicate the buffer, and add the noise */
-size_t ZDICT_trainFromBuffer_advanced(void* dictBuffer, size_t dictBufferCapacity,
-                                      const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                                      ZDICT_params_t params)
+/* ZDICT_trainFromBuffer_legacy() :
+ * issue : samplesBuffer need to be followed by a noisy guard band.
+ * work around : duplicate the buffer, and add the noise */
+size_t ZDICT_trainFromBuffer_legacy(void* dictBuffer, size_t dictBufferCapacity,
+                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                              ZDICT_legacy_params_t params)
 {
     size_t result;
     void* newBuff;
@@ -1030,10 +1071,9 @@
     memcpy(newBuff, samplesBuffer, sBuffSize);
     ZDICT_fillNoise((char*)newBuff + sBuffSize, NOISELENGTH);   /* guard band, for end of buffer condition */
 
-    result = ZDICT_trainFromBuffer_unsafe(
-                                        dictBuffer, dictBufferCapacity,
-                                        newBuff, samplesSizes, nbSamples,
-                                        params);
+    result =
+        ZDICT_trainFromBuffer_unsafe_legacy(dictBuffer, dictBufferCapacity, newBuff,
+                                            samplesSizes, nbSamples, params);
     free(newBuff);
     return result;
 }
@@ -1042,15 +1082,23 @@
 size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
 {
-    ZDICT_params_t params;
+    ZDICT_cover_params_t params;
+    DEBUGLOG(3, "ZDICT_trainFromBuffer");
     memset(&params, 0, sizeof(params));
-    return ZDICT_trainFromBuffer_advanced(dictBuffer, dictBufferCapacity,
-                                          samplesBuffer, samplesSizes, nbSamples,
-                                          params);
+    params.d = 8;
+    params.steps = 4;
+    /* Default to level 6 since no compression level information is available */
+    params.zParams.compressionLevel = 6;
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
+    params.zParams.notificationLevel = ZSTD_DEBUG;
+#endif
+    return ZDICT_optimizeTrainFromBuffer_cover(dictBuffer, dictBufferCapacity,
+                                               samplesBuffer, samplesSizes, nbSamples,
+                                               &params);
 }
 
 size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
-                                        const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
+                                  const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
 {
     ZDICT_params_t params;
     memset(&params, 0, sizeof(params));
--- a/contrib/python-zstandard/zstd/dictBuilder/zdict.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/dictBuilder/zdict.h	Mon Apr 09 10:13:29 2018 -0700
@@ -1,10 +1,11 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 #ifndef DICTBUILDER_H_001
@@ -20,10 +21,12 @@
 
 
 /* =====   ZDICTLIB_API : control library symbols visibility   ===== */
-#if defined(__GNUC__) && (__GNUC__ >= 4)
-#  define ZDICTLIB_VISIBILITY __attribute__ ((visibility ("default")))
-#else
-#  define ZDICTLIB_VISIBILITY
+#ifndef ZDICTLIB_VISIBILITY
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define ZDICTLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#  else
+#    define ZDICTLIB_VISIBILITY
+#  endif
 #endif
 #if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
 #  define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBILITY
@@ -34,20 +37,22 @@
 #endif
 
 
-/*! ZDICT_trainFromBuffer() :
-    Train a dictionary from an array of samples.
-    Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
-    supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
-    The resulting dictionary will be saved into `dictBuffer`.
-    @return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
-              or an error code, which can be tested with ZDICT_isError().
-    Tips : In general, a reasonable dictionary has a size of ~ 100 KB.
-           It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
-           In general, it's recommended to provide a few thousands samples, but this can vary a lot.
-           It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
-*/
+/*! ZDICT_trainFromBuffer():
+ *  Train a dictionary from an array of samples.
+ *  Redirect towards ZDICT_optimizeTrainFromBuffer_cover() single-threaded, with d=8 and steps=4.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *  Note: ZDICT_trainFromBuffer() requires about 9 bytes of memory for each input byte.
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ */
 ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
-                       const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
+                                    const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
 
 
 /*======   Helper functions   ======*/
@@ -67,102 +72,108 @@
  * ==================================================================================== */
 
 typedef struct {
-    unsigned selectivityLevel;   /* 0 means default; larger => select more => larger dictionary */
-    int      compressionLevel;   /* 0 means default; target a specific zstd compression level */
-    unsigned notificationLevel;  /* Write to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
-    unsigned dictID;             /* 0 means auto mode (32-bits random value); other : force dictID value */
-    unsigned reserved[2];        /* reserved space for future parameters */
+    int      compressionLevel;   /* optimize for a specific zstd compression level; 0 means default */
+    unsigned notificationLevel;  /* Write log to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
+    unsigned dictID;             /* force dictID value; 0 means auto mode (32-bits random value) */
 } ZDICT_params_t;
 
-
-/*! ZDICT_trainFromBuffer_advanced() :
-    Same as ZDICT_trainFromBuffer() with control over more parameters.
-    `parameters` is optional and can be provided with values set to 0 to mean "default".
-    @return : size of dictionary stored into `dictBuffer` (<= `dictBufferSize`),
-              or an error code, which can be tested by ZDICT_isError().
-    note : ZDICT_trainFromBuffer_advanced() will send notifications into stderr if instructed to, using notificationLevel>0.
-*/
-ZDICTLIB_API size_t ZDICT_trainFromBuffer_advanced(void* dictBuffer, size_t dictBufferCapacity,
-                                const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                                ZDICT_params_t parameters);
-
-/*! COVER_params_t :
-    For all values 0 means default.
-    kMin and d are the only required parameters.
-*/
+/*! ZDICT_cover_params_t:
+ *  k and d are the only required parameters.
+ *  For others, value 0 means default.
+ */
 typedef struct {
     unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
     unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
     unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (32) : Higher means more parameters checked */
-
     unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
-    unsigned notificationLevel;  /* Write to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
-    unsigned dictID;             /* 0 means auto mode (32-bits random value); other : force dictID value */
-    int      compressionLevel;   /* 0 means default; target a specific zstd compression level */
-} COVER_params_t;
+    ZDICT_params_t zParams;
+} ZDICT_cover_params_t;
 
 
-/*! COVER_trainFromBuffer() :
-    Train a dictionary from an array of samples using the COVER algorithm.
-    Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
-    supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
-    The resulting dictionary will be saved into `dictBuffer`.
-    @return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
-              or an error code, which can be tested with ZDICT_isError().
-    Note : COVER_trainFromBuffer() requires about 9 bytes of memory for each input byte.
-    Tips : In general, a reasonable dictionary has a size of ~ 100 KB.
-           It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
-           In general, it's recommended to provide a few thousands samples, but this can vary a lot.
-           It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
-*/
-ZDICTLIB_API size_t COVER_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
-                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                              COVER_params_t parameters);
-
-/*! COVER_optimizeTrainFromBuffer() :
-    The same requirements as above hold for all the parameters except `parameters`.
-    This function tries many parameter combinations and picks the best parameters.
-    `*parameters` is filled with the best parameters found, and the dictionary
-    constructed with those parameters is stored in `dictBuffer`.
-
-    All of the parameters d, k, steps are optional.
-    If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8, 10, 12, 14, 16}.
-    if steps is zero it defaults to its default value.
-    If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [16, 2048].
+/*! ZDICT_trainFromBuffer_cover():
+ *  Train a dictionary from an array of samples using the COVER algorithm.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *  Note: ZDICT_trainFromBuffer_cover() requires about 9 bytes of memory for each input byte.
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ */
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
+          void *dictBuffer, size_t dictBufferCapacity,
+    const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
+          ZDICT_cover_params_t parameters);
 
-    @return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
-              or an error code, which can be tested with ZDICT_isError().
-              On success `*parameters` contains the parameters selected.
-    Note : COVER_optimizeTrainFromBuffer() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread.
-*/
-ZDICTLIB_API size_t COVER_optimizeTrainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
-                                     const void* samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
-                                     COVER_params_t *parameters);
-
-/*! ZDICT_finalizeDictionary() :
-
-    Given a custom content as a basis for dictionary, and a set of samples,
-    finalize dictionary by adding headers and statistics.
+/*! ZDICT_optimizeTrainFromBuffer_cover():
+ * The same requirements as above hold for all the parameters except `parameters`.
+ * This function tries many parameter combinations and picks the best parameters.
+ * `*parameters` is filled with the best parameters found,
+ * dictionary constructed with those parameters is stored in `dictBuffer`.
+ *
+ * All of the parameters d, k, steps are optional.
+ * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8, 10, 12, 14, 16}.
+ * if steps is zero it defaults to its default value.
+ * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [16, 2048].
+ *
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *           or an error code, which can be tested with ZDICT_isError().
+ *           On success `*parameters` contains the parameters selected.
+ * Note: ZDICT_optimizeTrainFromBuffer_cover() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread.
+ */
+ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
+          void* dictBuffer, size_t dictBufferCapacity,
+    const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+          ZDICT_cover_params_t* parameters);
 
-    Samples must be stored concatenated in a flat buffer `samplesBuffer`,
-    supplied with an array of sizes `samplesSizes`, providing the size of each sample in order.
-
-    dictContentSize must be > ZDICT_CONTENTSIZE_MIN bytes.
-    maxDictSize must be >= dictContentSize, and must be > ZDICT_DICTSIZE_MIN bytes.
-
-    @return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`),
-              or an error code, which can be tested by ZDICT_isError().
-    note : ZDICT_finalizeDictionary() will push notifications into stderr if instructed to, using notificationLevel>0.
-    note 2 : dictBuffer and customDictContent can overlap
-*/
-#define ZDICT_CONTENTSIZE_MIN 256
-#define ZDICT_DICTSIZE_MIN    512
+/*! ZDICT_finalizeDictionary():
+ * Given a custom content as a basis for dictionary, and a set of samples,
+ * finalize dictionary by adding headers and statistics.
+ *
+ * Samples must be stored concatenated in a flat buffer `samplesBuffer`,
+ * supplied with an array of sizes `samplesSizes`, providing the size of each sample in order.
+ *
+ * dictContentSize must be >= ZDICT_CONTENTSIZE_MIN bytes.
+ * maxDictSize must be >= dictContentSize, and must be >= ZDICT_DICTSIZE_MIN bytes.
+ *
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`),
+ *           or an error code, which can be tested by ZDICT_isError().
+ * Note: ZDICT_finalizeDictionary() will push notifications into stderr if instructed to, using notificationLevel>0.
+ * Note 2: dictBuffer and dictContent can overlap
+ */
+#define ZDICT_CONTENTSIZE_MIN 128
+#define ZDICT_DICTSIZE_MIN    256
 ZDICTLIB_API size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
-                                const void* customDictContent, size_t dictContentSize,
+                                const void* dictContent, size_t dictContentSize,
                                 const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
                                 ZDICT_params_t parameters);
 
+typedef struct {
+    unsigned selectivityLevel;   /* 0 means default; larger => select more => larger dictionary */
+    ZDICT_params_t zParams;
+} ZDICT_legacy_params_t;
 
+/*! ZDICT_trainFromBuffer_legacy():
+ *  Train a dictionary from an array of samples.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * `parameters` is optional and can be provided with values set to 0 to mean "default".
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ *  Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0.
+ */
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_legacy(
+    void *dictBuffer, size_t dictBufferCapacity,
+    const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
+    ZDICT_legacy_params_t parameters);
 
 /* Deprecation warnings */
 /* It is generally possible to disable deprecation warnings from compiler,
@@ -174,7 +185,7 @@
 #else
 #  define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
 #  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
-#    define ZDICT_DEPRECATED(message) ZDICTLIB_API [[deprecated(message)]]
+#    define ZDICT_DEPRECATED(message) [[deprecated(message)]] ZDICTLIB_API
 #  elif (ZDICT_GCC_VERSION >= 405) || defined(__clang__)
 #    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated(message)))
 #  elif (ZDICT_GCC_VERSION >= 301)
--- a/contrib/python-zstandard/zstd/zstd.h	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd/zstd.h	Mon Apr 09 10:13:29 2018 -0700
@@ -2,11 +2,11 @@
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
- * This source code is licensed under the BSD-style license found in the
- * LICENSE file in the root directory of this source tree. An additional grant
- * of patent rights can be found in the PATENTS file in the same directory.
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
@@ -19,10 +19,12 @@
 
 
 /* =====   ZSTDLIB_API : control library symbols visibility   ===== */
-#if defined(__GNUC__) && (__GNUC__ >= 4)
-#  define ZSTDLIB_VISIBILITY __attribute__ ((visibility ("default")))
-#else
-#  define ZSTDLIB_VISIBILITY
+#ifndef ZSTDLIB_VISIBILITY
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define ZSTDLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#  else
+#    define ZSTDLIB_VISIBILITY
+#  endif
 #endif
 #if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
 #  define ZSTDLIB_API __declspec(dllexport) ZSTDLIB_VISIBILITY
@@ -36,117 +38,147 @@
 /*******************************************************************************************************
   Introduction
 
-  zstd, short for Zstandard, is a fast lossless compression algorithm, targeting real-time compression scenarios
-  at zlib-level and better compression ratios. The zstd compression library provides in-memory compression and
-  decompression functions. The library supports compression levels from 1 up to ZSTD_maxCLevel() which is 22.
-  Levels >= 20, labelled `--ultra`, should be used with caution, as they require more memory.
+  zstd, short for Zstandard, is a fast lossless compression algorithm,
+  targeting real-time compression scenarios at zlib-level and better compression ratios.
+  The zstd compression library provides in-memory compression and decompression functions.
+  The library supports compression levels from 1 up to ZSTD_maxCLevel() which is currently 22.
+  Levels >= 20, labeled `--ultra`, should be used with caution, as they require more memory.
   Compression can be done in:
     - a single step (described as Simple API)
-    - a single step, reusing a context (described as Explicit memory management)
+    - a single step, reusing a context (described as Explicit context)
     - unbounded multiple steps (described as Streaming compression)
-  The compression ratio achievable on small data can be highly improved using compression with a dictionary in:
+  The compression ratio achievable on small data can be highly improved using a dictionary in:
     - a single step (described as Simple dictionary API)
-    - a single step, reusing a dictionary (described as Fast dictionary API)
+    - a single step, reusing a dictionary (described as Bulk-processing dictionary API)
 
   Advanced experimental functions can be accessed using #define ZSTD_STATIC_LINKING_ONLY before including zstd.h.
-  These APIs shall never be used with a dynamic library.
+  Advanced experimental APIs shall never be used with a dynamic library.
   They are not "stable", their definition may change in the future. Only static linking is allowed.
 *********************************************************************************************************/
 
 /*------   Version   ------*/
 #define ZSTD_VERSION_MAJOR    1
-#define ZSTD_VERSION_MINOR    1
-#define ZSTD_VERSION_RELEASE  3
+#define ZSTD_VERSION_MINOR    3
+#define ZSTD_VERSION_RELEASE  4
+
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+ZSTDLIB_API unsigned ZSTD_versionNumber(void);   /**< useful to check dll version */
 
 #define ZSTD_LIB_VERSION ZSTD_VERSION_MAJOR.ZSTD_VERSION_MINOR.ZSTD_VERSION_RELEASE
 #define ZSTD_QUOTE(str) #str
 #define ZSTD_EXPAND_AND_QUOTE(str) ZSTD_QUOTE(str)
 #define ZSTD_VERSION_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_LIB_VERSION)
-
-#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
-ZSTDLIB_API unsigned ZSTD_versionNumber(void);   /**< library version number; to be used when checking dll version */
+ZSTDLIB_API const char* ZSTD_versionString(void);   /* added in v1.3.0 */
 
 
 /***************************************
 *  Simple API
 ***************************************/
 /*! ZSTD_compress() :
-    Compresses `src` content as a single zstd compressed frame into already allocated `dst`.
-    Hint : compression runs faster if `dstCapacity` >=  `ZSTD_compressBound(srcSize)`.
-    @return : compressed size written into `dst` (<= `dstCapacity),
-              or an error code if it fails (which can be tested using ZSTD_isError()). */
+ *  Compresses `src` content as a single zstd compressed frame into already allocated `dst`.
+ *  Hint : compression runs faster if `dstCapacity` >=  `ZSTD_compressBound(srcSize)`.
+ *  @return : compressed size written into `dst` (<= `dstCapacity),
+ *            or an error code if it fails (which can be tested using ZSTD_isError()). */
 ZSTDLIB_API size_t ZSTD_compress( void* dst, size_t dstCapacity,
                             const void* src, size_t srcSize,
                                   int compressionLevel);
 
 /*! ZSTD_decompress() :
-    `compressedSize` : must be the _exact_ size of a single compressed frame.
-    `dstCapacity` is an upper bound of originalSize.
-    If user cannot imply a maximum upper bound, it's better to use streaming mode to decompress data.
-    @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
-              or an errorCode if it fails (which can be tested using ZSTD_isError()). */
+ *  `compressedSize` : must be the _exact_ size of some number of compressed and/or skippable frames.
+ *  `dstCapacity` is an upper bound of originalSize to regenerate.
+ *  If user cannot imply a maximum upper bound, it's better to use streaming mode to decompress data.
+ *  @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+ *            or an errorCode if it fails (which can be tested using ZSTD_isError()). */
 ZSTDLIB_API size_t ZSTD_decompress( void* dst, size_t dstCapacity,
                               const void* src, size_t compressedSize);
 
+/*! ZSTD_getFrameContentSize() : added in v1.3.0
+ *  `src` should point to the start of a ZSTD encoded frame.
+ *  `srcSize` must be at least as large as the frame header.
+ *            hint : any size >= `ZSTD_frameHeaderSize_max` is large enough.
+ *  @return : - decompressed size of the frame in `src`, if known
+ *            - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
+ *            - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small)
+ *   note 1 : a 0 return value means the frame is valid but "empty".
+ *   note 2 : decompressed size is an optional field, it may not be present, typically in streaming mode.
+ *            When `return==ZSTD_CONTENTSIZE_UNKNOWN`, data to decompress could be any size.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *            Optionally, application can rely on some implicit limit,
+ *            as ZSTD_decompress() only needs an upper bound of decompressed size.
+ *            (For example, data could be necessarily cut into blocks <= 16 KB).
+ *   note 3 : decompressed size is always present when compression is done with ZSTD_compress()
+ *   note 4 : decompressed size can be very large (64-bits value),
+ *            potentially larger than what local system can handle as a single memory segment.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 5 : If source is untrusted, decompressed size could be wrong or intentionally modified.
+ *            Always ensure return value fits within application's authorized limits.
+ *            Each application can set its own limits.
+ *   note 6 : This function replaces ZSTD_getDecompressedSize() */
+#define ZSTD_CONTENTSIZE_UNKNOWN (0ULL - 1)
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+ZSTDLIB_API unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize);
+
 /*! ZSTD_getDecompressedSize() :
-*   'src' is the start of a zstd compressed frame.
-*   @return : content size to be decompressed, as a 64-bits value _if known_, 0 otherwise.
-*    note 1 : decompressed size is an optional field, that may not be present, especially in streaming mode.
-*             When `return==0`, data to decompress could be any size.
-*             In which case, it's necessary to use streaming mode to decompress data.
-*             Optionally, application can still use ZSTD_decompress() while relying on implied limits.
-*             (For example, data may be necessarily cut into blocks <= 16 KB).
-*    note 2 : decompressed size is always present when compression is done with ZSTD_compress()
-*    note 3 : decompressed size can be very large (64-bits value),
-*             potentially larger than what local system can handle as a single memory segment.
-*             In which case, it's necessary to use streaming mode to decompress data.
-*    note 4 : If source is untrusted, decompressed size could be wrong or intentionally modified.
-*             Always ensure result fits within application's authorized limits.
-*             Each application can set its own limits.
-*    note 5 : when `return==0`, if precise failure cause is needed, use ZSTD_getFrameParams() to know more. */
+ *  NOTE: This function is now obsolete, in favor of ZSTD_getFrameContentSize().
+ *  Both functions work the same way, but ZSTD_getDecompressedSize() blends
+ *  "empty", "unknown" and "error" results to the same return value (0),
+ *  while ZSTD_getFrameContentSize() gives them separate return values.
+ * `src` is the start of a zstd compressed frame.
+ * @return : content size to be decompressed, as a 64-bits value _if known and not empty_, 0 otherwise. */
 ZSTDLIB_API unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
 
 
 /*======  Helper functions  ======*/
-ZSTDLIB_API int         ZSTD_maxCLevel(void);               /*!< maximum compression level available */
-ZSTDLIB_API size_t      ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case scenario */
+#define ZSTD_COMPRESSBOUND(srcSize)   ((srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0))  /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
+ZSTDLIB_API size_t      ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
 ZSTDLIB_API unsigned    ZSTD_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
 ZSTDLIB_API const char* ZSTD_getErrorName(size_t code);     /*!< provides readable string from an error code */
+ZSTDLIB_API int         ZSTD_maxCLevel(void);               /*!< maximum compression level available */
 
 
 /***************************************
-*  Explicit memory management
+*  Explicit context
 ***************************************/
 /*= Compression context
-*   When compressing many times,
-*   it is recommended to allocate a context just once, and re-use it for each successive compression operation.
-*   This will make workload friendlier for system's memory.
-*   Use one context per thread for parallel execution in multi-threaded environments. */
+ *  When compressing many times,
+ *  it is recommended to allocate a context just once, and re-use it for each successive compression operation.
+ *  This will make workload friendlier for system's memory.
+ *  Use one context per thread for parallel execution in multi-threaded environments. */
 typedef struct ZSTD_CCtx_s ZSTD_CCtx;
 ZSTDLIB_API ZSTD_CCtx* ZSTD_createCCtx(void);
 ZSTDLIB_API size_t     ZSTD_freeCCtx(ZSTD_CCtx* cctx);
 
 /*! ZSTD_compressCCtx() :
-    Same as ZSTD_compress(), requires an allocated ZSTD_CCtx (see ZSTD_createCCtx()). */
-ZSTDLIB_API size_t ZSTD_compressCCtx(ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel);
+ *  Same as ZSTD_compress(), requires an allocated ZSTD_CCtx (see ZSTD_createCCtx()). */
+ZSTDLIB_API size_t ZSTD_compressCCtx(ZSTD_CCtx* ctx,
+                                     void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                                     int compressionLevel);
 
-/*= Decompression context */
+/*= Decompression context
+ *  When decompressing many times,
+ *  it is recommended to allocate a context only once,
+ *  and re-use it for each successive compression operation.
+ *  This will make workload friendlier for system's memory.
+ *  Use one context per thread for parallel execution. */
 typedef struct ZSTD_DCtx_s ZSTD_DCtx;
 ZSTDLIB_API ZSTD_DCtx* ZSTD_createDCtx(void);
 ZSTDLIB_API size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
 
 /*! ZSTD_decompressDCtx() :
-*   Same as ZSTD_decompress(), requires an allocated ZSTD_DCtx (see ZSTD_createDCtx()). */
-ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ *  Same as ZSTD_decompress(), requires an allocated ZSTD_DCtx (see ZSTD_createDCtx()) */
+ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* ctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize);
 
 
 /**************************
 *  Simple dictionary API
 ***************************/
 /*! ZSTD_compress_usingDict() :
-*   Compression using a predefined Dictionary (see dictBuilder/zdict.h).
-*   Note : This function loads the dictionary, resulting in significant startup delay.
-*   Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
+ *  Compression using a predefined Dictionary (see dictBuilder/zdict.h).
+ *  Note : This function loads the dictionary, resulting in significant startup delay.
+ *  Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
 ZSTDLIB_API size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx,
                                            void* dst, size_t dstCapacity,
                                      const void* src, size_t srcSize,
@@ -154,36 +186,38 @@
                                            int compressionLevel);
 
 /*! ZSTD_decompress_usingDict() :
-*   Decompression using a predefined Dictionary (see dictBuilder/zdict.h).
-*   Dictionary must be identical to the one used during compression.
-*   Note : This function loads the dictionary, resulting in significant startup delay.
-*   Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
+ *  Decompression using a predefined Dictionary (see dictBuilder/zdict.h).
+ *  Dictionary must be identical to the one used during compression.
+ *  Note : This function loads the dictionary, resulting in significant startup delay.
+ *  Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
 ZSTDLIB_API size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
                                              void* dst, size_t dstCapacity,
                                        const void* src, size_t srcSize,
                                        const void* dict,size_t dictSize);
 
 
-/****************************
-*  Fast dictionary API
-****************************/
+/**********************************
+ *  Bulk processing dictionary API
+ *********************************/
 typedef struct ZSTD_CDict_s ZSTD_CDict;
 
 /*! ZSTD_createCDict() :
-*   When compressing multiple messages / blocks with the same dictionary, it's recommended to load it just once.
-*   ZSTD_createCDict() will create a digested dictionary, ready to start future compression operations without startup delay.
-*   ZSTD_CDict can be created once and used by multiple threads concurrently, as its usage is read-only.
-*   `dictBuffer` can be released after ZSTD_CDict creation, as its content is copied within CDict */
-ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict(const void* dictBuffer, size_t dictSize, int compressionLevel);
+ *  When compressing multiple messages / blocks with the same dictionary, it's recommended to load it just once.
+ *  ZSTD_createCDict() will create a digested dictionary, ready to start future compression operations without startup delay.
+ *  ZSTD_CDict can be created once and shared by multiple threads concurrently, since its usage is read-only.
+ *  `dictBuffer` can be released after ZSTD_CDict creation, since its content is copied within CDict */
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict(const void* dictBuffer, size_t dictSize,
+                                         int compressionLevel);
 
 /*! ZSTD_freeCDict() :
-*   Function frees memory allocated by ZSTD_createCDict(). */
+ *  Function frees memory allocated by ZSTD_createCDict(). */
 ZSTDLIB_API size_t      ZSTD_freeCDict(ZSTD_CDict* CDict);
 
 /*! ZSTD_compress_usingCDict() :
-*   Compression using a digested Dictionary.
-*   Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
-*   Note that compression level is decided during dictionary creation. */
+ *  Compression using a digested Dictionary.
+ *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
+ *  Note that compression level is decided during dictionary creation.
+ *  Frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no) */
 ZSTDLIB_API size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
                                             void* dst, size_t dstCapacity,
                                       const void* src, size_t srcSize,
@@ -193,17 +227,17 @@
 typedef struct ZSTD_DDict_s ZSTD_DDict;
 
 /*! ZSTD_createDDict() :
-*   Create a digested dictionary, ready to start decompression operation without startup delay.
-*   dictBuffer can be released after DDict creation, as its content is copied inside DDict */
+ *  Create a digested dictionary, ready to start decompression operation without startup delay.
+ *  dictBuffer can be released after DDict creation, as its content is copied inside DDict */
 ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict(const void* dictBuffer, size_t dictSize);
 
 /*! ZSTD_freeDDict() :
-*   Function frees memory allocated with ZSTD_createDDict() */
+ *  Function frees memory allocated with ZSTD_createDDict() */
 ZSTDLIB_API size_t      ZSTD_freeDDict(ZSTD_DDict* ddict);
 
 /*! ZSTD_decompress_usingDDict() :
-*   Decompression using a digested Dictionary.
-*   Faster startup than ZSTD_decompress_usingDict(), recommended when same dictionary is used multiple times. */
+ *  Decompression using a digested Dictionary.
+ *  Faster startup than ZSTD_decompress_usingDict(), recommended when same dictionary is used multiple times. */
 ZSTDLIB_API size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
                                               void* dst, size_t dstCapacity,
                                         const void* src, size_t srcSize,
@@ -260,17 +294,22 @@
 *  ZSTD_endStream() instructs to finish a frame.
 *  It will perform a flush and write frame epilogue.
 *  The epilogue is required for decoders to consider a frame completed.
-*  Similar to ZSTD_flushStream(), it may not be able to flush the full content if `output->size` is too small.
+*  ZSTD_endStream() may not be able to flush full data if `output->size` is too small.
 *  In which case, call again ZSTD_endStream() to complete the flush.
-*  @return : nb of bytes still present within internal buffer (0 if it's empty, hence compression completed)
+*  @return : 0 if frame fully completed and fully flushed,
+             or >0 if some data is still present within internal buffer
+                  (value is minimum size estimation for remaining data to flush, but it could be more)
 *            or an error code, which can be tested using ZSTD_isError().
 *
 * *******************************************************************/
 
-typedef struct ZSTD_CStream_s ZSTD_CStream;
+typedef ZSTD_CCtx ZSTD_CStream;  /**< CCtx and CStream are now effectively same object (>= v1.3.0) */
+                                 /* Continue to distinguish them for compatibility with versions <= v1.2.0 */
+/*===== ZSTD_CStream management functions =====*/
 ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream(void);
 ZSTDLIB_API size_t ZSTD_freeCStream(ZSTD_CStream* zcs);
 
+/*===== Streaming compression functions =====*/
 ZSTDLIB_API size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel);
 ZSTDLIB_API size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
 ZSTDLIB_API size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
@@ -303,10 +342,13 @@
 *            The return value is a suggested next input size (a hint to improve latency) that will never load more than the current frame.
 * *******************************************************************************/
 
-typedef struct ZSTD_DStream_s ZSTD_DStream;
+typedef ZSTD_DCtx ZSTD_DStream;  /**< DCtx and DStream are now effectively same object (>= v1.3.0) */
+                                 /* For compatibility with versions <= v1.2.0, continue to consider them separated. */
+/*===== ZSTD_DStream management functions =====*/
 ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream(void);
 ZSTDLIB_API size_t ZSTD_freeDStream(ZSTD_DStream* zds);
 
+/*===== Streaming decompression functions =====*/
 ZSTDLIB_API size_t ZSTD_initDStream(ZSTD_DStream* zds);
 ZSTDLIB_API size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
 
@@ -316,47 +358,55 @@
 #endif  /* ZSTD_H_235446 */
 
 
-#if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
-#define ZSTD_H_ZSTD_STATIC_LINKING_ONLY
 
 /****************************************************************************************
  * START OF ADVANCED AND EXPERIMENTAL FUNCTIONS
  * The definitions in this section are considered experimental.
- * They should never be used with a dynamic library, as they may change in the future.
- * They are provided for advanced usages.
+ * They should never be used with a dynamic library, as prototypes may change in the future.
+ * They are provided for advanced scenarios.
  * Use them only in association with static linking.
  * ***************************************************************************************/
 
+#if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
+#define ZSTD_H_ZSTD_STATIC_LINKING_ONLY
+
 /* --- Constants ---*/
 #define ZSTD_MAGICNUMBER            0xFD2FB528   /* >= v0.8.0 */
 #define ZSTD_MAGIC_SKIPPABLE_START  0x184D2A50U
+#define ZSTD_MAGIC_DICTIONARY       0xEC30A437   /* >= v0.7.0 */
 
-#define ZSTD_WINDOWLOG_MAX_32  25
-#define ZSTD_WINDOWLOG_MAX_64  27
-#define ZSTD_WINDOWLOG_MAX    ((U32)(MEM_32bits() ? ZSTD_WINDOWLOG_MAX_32 : ZSTD_WINDOWLOG_MAX_64))
-#define ZSTD_WINDOWLOG_MIN     10
-#define ZSTD_HASHLOG_MAX       ZSTD_WINDOWLOG_MAX
-#define ZSTD_HASHLOG_MIN        6
-#define ZSTD_CHAINLOG_MAX     (ZSTD_WINDOWLOG_MAX+1)
-#define ZSTD_CHAINLOG_MIN      ZSTD_HASHLOG_MIN
-#define ZSTD_HASHLOG3_MAX      17
-#define ZSTD_SEARCHLOG_MAX    (ZSTD_WINDOWLOG_MAX-1)
-#define ZSTD_SEARCHLOG_MIN      1
-#define ZSTD_SEARCHLENGTH_MAX   7   /* only for ZSTD_fast, other strategies are limited to 6 */
-#define ZSTD_SEARCHLENGTH_MIN   3   /* only for ZSTD_btopt, other strategies are limited to 4 */
-#define ZSTD_TARGETLENGTH_MIN   4
-#define ZSTD_TARGETLENGTH_MAX 999
+#define ZSTD_WINDOWLOG_MAX_32   30
+#define ZSTD_WINDOWLOG_MAX_64   31
+#define ZSTD_WINDOWLOG_MAX    ((unsigned)(sizeof(size_t) == 4 ? ZSTD_WINDOWLOG_MAX_32 : ZSTD_WINDOWLOG_MAX_64))
+#define ZSTD_WINDOWLOG_MIN      10
+#define ZSTD_HASHLOG_MAX      ((ZSTD_WINDOWLOG_MAX < 30) ? ZSTD_WINDOWLOG_MAX : 30)
+#define ZSTD_HASHLOG_MIN         6
+#define ZSTD_CHAINLOG_MAX_32    29
+#define ZSTD_CHAINLOG_MAX_64    30
+#define ZSTD_CHAINLOG_MAX     ((unsigned)(sizeof(size_t) == 4 ? ZSTD_CHAINLOG_MAX_32 : ZSTD_CHAINLOG_MAX_64))
+#define ZSTD_CHAINLOG_MIN       ZSTD_HASHLOG_MIN
+#define ZSTD_HASHLOG3_MAX       17
+#define ZSTD_SEARCHLOG_MAX     (ZSTD_WINDOWLOG_MAX-1)
+#define ZSTD_SEARCHLOG_MIN       1
+#define ZSTD_SEARCHLENGTH_MAX    7   /* only for ZSTD_fast, other strategies are limited to 6 */
+#define ZSTD_SEARCHLENGTH_MIN    3   /* only for ZSTD_btopt, other strategies are limited to 4 */
+#define ZSTD_TARGETLENGTH_MIN    1   /* only used by btopt, btultra and btfast */
+#define ZSTD_LDM_MINMATCH_MIN    4
+#define ZSTD_LDM_MINMATCH_MAX 4096
+#define ZSTD_LDM_BUCKETSIZELOG_MAX 8
 
-#define ZSTD_FRAMEHEADERSIZE_MAX 18    /* for static allocation */
-#define ZSTD_FRAMEHEADERSIZE_MIN  6
-static const size_t ZSTD_frameHeaderSize_prefix = 5;
+#define ZSTD_FRAMEHEADERSIZE_PREFIX 5   /* minimum input size to know frame header size */
+#define ZSTD_FRAMEHEADERSIZE_MIN    6
+#define ZSTD_FRAMEHEADERSIZE_MAX   18   /* for static allocation */
+static const size_t ZSTD_frameHeaderSize_prefix = ZSTD_FRAMEHEADERSIZE_PREFIX;
 static const size_t ZSTD_frameHeaderSize_min = ZSTD_FRAMEHEADERSIZE_MIN;
 static const size_t ZSTD_frameHeaderSize_max = ZSTD_FRAMEHEADERSIZE_MAX;
 static const size_t ZSTD_skippableHeaderSize = 8;  /* magic number + skippable frame length */
 
 
 /*--- Advanced types ---*/
-typedef enum { ZSTD_fast, ZSTD_dfast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2, ZSTD_btlazy2, ZSTD_btopt, ZSTD_btopt2 } ZSTD_strategy;   /* from faster to stronger */
+typedef enum { ZSTD_fast=1, ZSTD_dfast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2,
+               ZSTD_btlazy2, ZSTD_btopt, ZSTD_btultra } ZSTD_strategy;   /* from faster to stronger */
 
 typedef struct {
     unsigned windowLog;      /**< largest match distance : larger == more compression, more memory needed during decompression */
@@ -379,35 +429,190 @@
     ZSTD_frameParameters fParams;
 } ZSTD_parameters;
 
-/*= Custom memory allocation functions */
+typedef struct ZSTD_CCtx_params_s ZSTD_CCtx_params;
+
+typedef enum {
+    ZSTD_dct_auto=0,      /* dictionary is "full" when starting with ZSTD_MAGIC_DICTIONARY, otherwise it is "rawContent" */
+    ZSTD_dct_rawContent,  /* ensures dictionary is always loaded as rawContent, even if it starts with ZSTD_MAGIC_DICTIONARY */
+    ZSTD_dct_fullDict     /* refuses to load a dictionary if it does not respect Zstandard's specification */
+} ZSTD_dictContentType_e;
+
+typedef enum {
+    ZSTD_dlm_byCopy = 0, /**< Copy dictionary content internally */
+    ZSTD_dlm_byRef,      /**< Reference dictionary content -- the dictionary buffer must outlive its users. */
+} ZSTD_dictLoadMethod_e;
+
+
+
+/***************************************
+*  Frame size functions
+***************************************/
+
+/*! ZSTD_findFrameCompressedSize() :
+ *  `src` should point to the start of a ZSTD encoded frame or skippable frame
+ *  `srcSize` must be >= first frame size
+ *  @return : the compressed size of the first frame starting at `src`,
+ *            suitable to pass to `ZSTD_decompress` or similar,
+ *            or an error code if input is invalid */
+ZSTDLIB_API size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize);
+
+/*! ZSTD_findDecompressedSize() :
+ *  `src` should point the start of a series of ZSTD encoded and/or skippable frames
+ *  `srcSize` must be the _exact_ size of this series
+ *       (i.e. there should be a frame boundary exactly at `srcSize` bytes after `src`)
+ *  @return : - decompressed size of all data in all successive frames
+ *            - if the decompressed size cannot be determined: ZSTD_CONTENTSIZE_UNKNOWN
+ *            - if an error occurred: ZSTD_CONTENTSIZE_ERROR
+ *
+ *   note 1 : decompressed size is an optional field, that may not be present, especially in streaming mode.
+ *            When `return==ZSTD_CONTENTSIZE_UNKNOWN`, data to decompress could be any size.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 2 : decompressed size is always present when compression is done with ZSTD_compress()
+ *   note 3 : decompressed size can be very large (64-bits value),
+ *            potentially larger than what local system can handle as a single memory segment.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 4 : If source is untrusted, decompressed size could be wrong or intentionally modified.
+ *            Always ensure result fits within application's authorized limits.
+ *            Each application can set its own limits.
+ *   note 5 : ZSTD_findDecompressedSize handles multiple frames, and so it must traverse the input to
+ *            read each contained frame header.  This is fast as most of the data is skipped,
+ *            however it does mean that all frame data must be present and valid. */
+ZSTDLIB_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
+
+/*! ZSTD_frameHeaderSize() :
+*   `src` should point to the start of a ZSTD frame
+*   `srcSize` must be >= ZSTD_frameHeaderSize_prefix.
+*   @return : size of the Frame Header */
+ZSTDLIB_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
+
+
+/***************************************
+*  Memory management
+***************************************/
+
+/*! ZSTD_sizeof_*() :
+ *  These functions give the current memory usage of selected object.
+ *  Object memory usage can evolve when re-used. */
+ZSTDLIB_API size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx);
+ZSTDLIB_API size_t ZSTD_sizeof_DCtx(const ZSTD_DCtx* dctx);
+ZSTDLIB_API size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs);
+ZSTDLIB_API size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds);
+ZSTDLIB_API size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict);
+ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
+
+/*! ZSTD_estimate*() :
+ *  These functions make it possible to estimate memory usage
+ *  of a future {D,C}Ctx, before its creation.
+ *  ZSTD_estimateCCtxSize() will provide a budget large enough for any compression level up to selected one.
+ *  It will also consider src size to be arbitrarily "large", which is worst case.
+ *  If srcSize is known to always be small, ZSTD_estimateCCtxSize_usingCParams() can provide a tighter estimation.
+ *  ZSTD_estimateCCtxSize_usingCParams() can be used in tandem with ZSTD_getCParams() to create cParams from compressionLevel.
+ *  ZSTD_estimateCCtxSize_usingCCtxParams() can be used in tandem with ZSTD_CCtxParam_setParameter(). Only single-threaded compression is supported. This function will return an error code if ZSTD_p_nbWorkers is >= 1.
+ *  Note : CCtx size estimation is only correct for single-threaded compression. */
+ZSTDLIB_API size_t ZSTD_estimateCCtxSize(int compressionLevel);
+ZSTDLIB_API size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_API size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_API size_t ZSTD_estimateDCtxSize(void);
+
+/*! ZSTD_estimateCStreamSize() :
+ *  ZSTD_estimateCStreamSize() will provide a budget large enough for any compression level up to selected one.
+ *  It will also consider src size to be arbitrarily "large", which is worst case.
+ *  If srcSize is known to always be small, ZSTD_estimateCStreamSize_usingCParams() can provide a tighter estimation.
+ *  ZSTD_estimateCStreamSize_usingCParams() can be used in tandem with ZSTD_getCParams() to create cParams from compressionLevel.
+ *  ZSTD_estimateCStreamSize_usingCCtxParams() can be used in tandem with ZSTD_CCtxParam_setParameter(). Only single-threaded compression is supported. This function will return an error code if ZSTD_p_nbWorkers is >= 1.
+ *  Note : CStream size estimation is only correct for single-threaded compression.
+ *  ZSTD_DStream memory budget depends on window Size.
+ *  This information can be passed manually, using ZSTD_estimateDStreamSize,
+ *  or deducted from a valid frame Header, using ZSTD_estimateDStreamSize_fromFrame();
+ *  Note : if streaming is init with function ZSTD_init?Stream_usingDict(),
+ *         an internal ?Dict will be created, which additional size is not estimated here.
+ *         In this case, get total size by adding ZSTD_estimate?DictSize */
+ZSTDLIB_API size_t ZSTD_estimateCStreamSize(int compressionLevel);
+ZSTDLIB_API size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_API size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_API size_t ZSTD_estimateDStreamSize(size_t windowSize);
+ZSTDLIB_API size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize);
+
+/*! ZSTD_estimate?DictSize() :
+ *  ZSTD_estimateCDictSize() will bet that src size is relatively "small", and content is copied, like ZSTD_createCDict().
+ *  ZSTD_estimateCDictSize_advanced() makes it possible to control compression parameters precisely, like ZSTD_createCDict_advanced().
+ *  Note : dictionaries created by reference (`ZSTD_dlm_byRef`) are logically smaller.
+ */
+ZSTDLIB_API size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel);
+ZSTDLIB_API size_t ZSTD_estimateCDictSize_advanced(size_t dictSize, ZSTD_compressionParameters cParams, ZSTD_dictLoadMethod_e dictLoadMethod);
+ZSTDLIB_API size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod);
+
+/*! ZSTD_initStatic*() :
+ *  Initialize an object using a pre-allocated fixed-size buffer.
+ *  workspace: The memory area to emplace the object into.
+ *             Provided pointer *must be 8-bytes aligned*.
+ *             Buffer must outlive object.
+ *  workspaceSize: Use ZSTD_estimate*Size() to determine
+ *                 how large workspace must be to support target scenario.
+ * @return : pointer to object (same address as workspace, just different type),
+ *           or NULL if error (size too small, incorrect alignment, etc.)
+ *  Note : zstd will never resize nor malloc() when using a static buffer.
+ *         If the object requires more memory than available,
+ *         zstd will just error out (typically ZSTD_error_memory_allocation).
+ *  Note 2 : there is no corresponding "free" function.
+ *           Since workspace is allocated externally, it must be freed externally too.
+ *  Note 3 : cParams : use ZSTD_getCParams() to convert a compression level
+ *           into its associated cParams.
+ *  Limitation 1 : currently not compatible with internal dictionary creation, triggered by
+ *                 ZSTD_CCtx_loadDictionary(), ZSTD_initCStream_usingDict() or ZSTD_initDStream_usingDict().
+ *  Limitation 2 : static cctx currently not compatible with multi-threading.
+ *  Limitation 3 : static dctx is incompatible with legacy support.
+ */
+ZSTDLIB_API ZSTD_CCtx*    ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize);
+ZSTDLIB_API ZSTD_CStream* ZSTD_initStaticCStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticCCtx() */
+
+ZSTDLIB_API ZSTD_DCtx*    ZSTD_initStaticDCtx(void* workspace, size_t workspaceSize);
+ZSTDLIB_API ZSTD_DStream* ZSTD_initStaticDStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticDCtx() */
+
+ZSTDLIB_API const ZSTD_CDict* ZSTD_initStaticCDict(
+                                        void* workspace, size_t workspaceSize,
+                                        const void* dict, size_t dictSize,
+                                        ZSTD_dictLoadMethod_e dictLoadMethod,
+                                        ZSTD_dictContentType_e dictContentType,
+                                        ZSTD_compressionParameters cParams);
+
+ZSTDLIB_API const ZSTD_DDict* ZSTD_initStaticDDict(
+                                        void* workspace, size_t workspaceSize,
+                                        const void* dict, size_t dictSize,
+                                        ZSTD_dictLoadMethod_e dictLoadMethod,
+                                        ZSTD_dictContentType_e dictContentType);
+
+/*! Custom memory allocation :
+ *  These prototypes make it possible to pass your own allocation/free functions.
+ *  ZSTD_customMem is provided at creation time, using ZSTD_create*_advanced() variants listed below.
+ *  All allocation/free operations will be completed using these custom variants instead of regular <stdlib.h> ones.
+ */
 typedef void* (*ZSTD_allocFunction) (void* opaque, size_t size);
 typedef void  (*ZSTD_freeFunction) (void* opaque, void* address);
 typedef struct { ZSTD_allocFunction customAlloc; ZSTD_freeFunction customFree; void* opaque; } ZSTD_customMem;
+static ZSTD_customMem const ZSTD_defaultCMem = { NULL, NULL, NULL };  /**< this constant defers to stdlib's functions */
+
+ZSTDLIB_API ZSTD_CCtx*    ZSTD_createCCtx_advanced(ZSTD_customMem customMem);
+ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem);
+ZSTDLIB_API ZSTD_DCtx*    ZSTD_createDCtx_advanced(ZSTD_customMem customMem);
+ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem);
+
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize,
+                                                  ZSTD_dictLoadMethod_e dictLoadMethod,
+                                                  ZSTD_dictContentType_e dictContentType,
+                                                  ZSTD_compressionParameters cParams,
+                                                  ZSTD_customMem customMem);
+
+ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
+                                                  ZSTD_dictLoadMethod_e dictLoadMethod,
+                                                  ZSTD_dictContentType_e dictContentType,
+                                                  ZSTD_customMem customMem);
+
 
 
 /***************************************
 *  Advanced compression functions
 ***************************************/
-/*! ZSTD_estimateCCtxSize() :
- *  Gives the amount of memory allocated for a ZSTD_CCtx given a set of compression parameters.
- *  `frameContentSize` is an optional parameter, provide `0` if unknown */
-ZSTDLIB_API size_t ZSTD_estimateCCtxSize(ZSTD_compressionParameters cParams);
-
-/*! ZSTD_createCCtx_advanced() :
- *  Create a ZSTD compression context using external alloc and free functions */
-ZSTDLIB_API ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem);
-
-/*! ZSTD_sizeofCCtx() :
- *  Gives the amount of memory used by a given ZSTD_CCtx */
-ZSTDLIB_API size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx);
-
-typedef enum {
-    ZSTD_p_forceWindow   /* Force back-references to remain < windowSize, even when referencing Dictionary content (default:0)*/
-} ZSTD_CCtxParameter;
-/*! ZSTD_setCCtxParameter() :
- *  Set advanced parameters, selected through enum ZSTD_CCtxParameter
- *  @result : 0, or an error code (which can be tested with ZSTD_isError()) */
-ZSTDLIB_API size_t ZSTD_setCCtxParameter(ZSTD_CCtx* cctx, ZSTD_CCtxParameter param, unsigned value);
 
 /*! ZSTD_createCDict_byReference() :
  *  Create a digested dictionary for compression
@@ -415,15 +620,6 @@
  *  It is important that dictBuffer outlives CDict, it must remain read accessible throughout the lifetime of CDict */
 ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
 
-/*! ZSTD_createCDict_advanced() :
- *  Create a ZSTD_CDict using external alloc and free, and customized compression parameters */
-ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize, unsigned byReference,
-                                                  ZSTD_parameters params, ZSTD_customMem customMem);
-
-/*! ZSTD_sizeof_CDict() :
- *  Gives the amount of memory used by a given ZSTD_sizeof_CDict */
-ZSTDLIB_API size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict);
-
 /*! ZSTD_getCParams() :
 *   @return ZSTD_compressionParameters structure for a selected compression level and estimated srcSize.
 *   `estimatedSrcSize` value is optional, select 0 if not known */
@@ -431,7 +627,7 @@
 
 /*! ZSTD_getParams() :
 *   same as ZSTD_getCParams(), but @return a full `ZSTD_parameters` object instead of sub-component `ZSTD_compressionParameters`.
-*   All fields of `ZSTD_frameParameters` are set to default (0) */
+*   All fields of `ZSTD_frameParameters` are set to default : contentSize=1, checksum=0, noDictID=0 */
 ZSTDLIB_API ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
 
 /*! ZSTD_checkCParams() :
@@ -439,17 +635,24 @@
 ZSTDLIB_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
 
 /*! ZSTD_adjustCParams() :
-*   optimize params for a given `srcSize` and `dictSize`.
-*   both values are optional, select `0` if unknown. */
+ *  optimize params for a given `srcSize` and `dictSize`.
+ *  both values are optional, select `0` if unknown. */
 ZSTDLIB_API ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
 
 /*! ZSTD_compress_advanced() :
-*   Same as ZSTD_compress_usingDict(), with fine-tune control of each compression parameter */
-ZSTDLIB_API size_t ZSTD_compress_advanced (ZSTD_CCtx* ctx,
-                                           void* dst, size_t dstCapacity,
-                                     const void* src, size_t srcSize,
-                                     const void* dict,size_t dictSize,
-                                           ZSTD_parameters params);
+*   Same as ZSTD_compress_usingDict(), with fine-tune control over each compression parameter */
+ZSTDLIB_API size_t ZSTD_compress_advanced (ZSTD_CCtx* cctx,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const void* dict,size_t dictSize,
+                                  ZSTD_parameters params);
+
+/*! ZSTD_compress_usingCDict_advanced() :
+*   Same as ZSTD_compress_usingCDict(), with fine-tune control over frame parameters */
+ZSTDLIB_API size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const ZSTD_CDict* cdict, ZSTD_frameParameters fParams);
 
 
 /*--- Advanced decompression functions ---*/
@@ -461,30 +664,13 @@
  *  Note 3 : Skippable Frame Identifiers are considered valid. */
 ZSTDLIB_API unsigned ZSTD_isFrame(const void* buffer, size_t size);
 
-/*! ZSTD_estimateDCtxSize() :
- *  Gives the potential amount of memory allocated to create a ZSTD_DCtx */
-ZSTDLIB_API size_t ZSTD_estimateDCtxSize(void);
-
-/*! ZSTD_createDCtx_advanced() :
- *  Create a ZSTD decompression context using external alloc and free functions */
-ZSTDLIB_API ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem);
-
-/*! ZSTD_sizeof_DCtx() :
- *  Gives the amount of memory used by a given ZSTD_DCtx */
-ZSTDLIB_API size_t ZSTD_sizeof_DCtx(const ZSTD_DCtx* dctx);
-
 /*! ZSTD_createDDict_byReference() :
  *  Create a digested dictionary, ready to start decompression operation without startup delay.
- *  Dictionary content is simply referenced, and therefore stays in dictBuffer.
- *  It is important that dictBuffer outlives DDict, it must remain read accessible throughout the lifetime of DDict */
+ *  Dictionary content is referenced, and therefore stays in dictBuffer.
+ *  It is important that dictBuffer outlives DDict,
+ *  it must remain read accessible throughout the lifetime of DDict */
 ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize);
 
-ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
-                                                  unsigned byReference, ZSTD_customMem customMem);
-
-/*! ZSTD_sizeof_DDict() :
- *  Gives the amount of memory used by a given ZSTD_DDict */
-ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 
 /*! ZSTD_getDictID_fromDict() :
  *  Provides the dictID stored within dictionary.
@@ -507,7 +693,7 @@
  *    Note : this use case also happens when using a non-conformant dictionary.
  *  - `srcSize` is too small, and as a result, the frame header could not be decoded (only possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`).
  *  - This is not a Zstandard frame.
- *  When identifying the exact failure cause, it's possible to used ZSTD_getFrameParams(), which will provide a more precise error code. */
+ *  When identifying the exact failure cause, it's possible to use ZSTD_getFrameHeader(), which will provide a more precise error code. */
 ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
 
 
@@ -516,32 +702,55 @@
 ********************************************************************/
 
 /*=====   Advanced Streaming compression functions  =====*/
-ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem);
-ZSTDLIB_API size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pledgedSrcSize);   /**< pledgedSrcSize must be correct */
-ZSTDLIB_API size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel); /**< note: a dict will not be used if dict == NULL or dictSize < 8 */
+ZSTDLIB_API size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pledgedSrcSize);   /**< pledgedSrcSize must be correct. If it is not known at init time, use ZSTD_CONTENTSIZE_UNKNOWN. Note that, for compatibility with older programs, "0" also disables frame content size field. It may be enabled in the future. */
+ZSTDLIB_API size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel); /**< creates of an internal CDict (incompatible with static CCtx), except if dict == NULL or dictSize < 8, in which case no dict is used. Note: dict is loaded with ZSTD_dm_auto (treated as a full zstd dictionary if it begins with ZSTD_MAGIC_DICTIONARY, else as raw content) and ZSTD_dlm_byCopy.*/
 ZSTDLIB_API size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs, const void* dict, size_t dictSize,
-                                             ZSTD_parameters params, unsigned long long pledgedSrcSize);  /**< pledgedSrcSize is optional and can be zero == unknown */
+                                             ZSTD_parameters params, unsigned long long pledgedSrcSize);  /**< pledgedSrcSize must be correct. If srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN. dict is loaded with ZSTD_dm_auto and ZSTD_dlm_byCopy. */
 ZSTDLIB_API size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);  /**< note : cdict will just be referenced, and must outlive compression session */
-ZSTDLIB_API size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);  /**< re-use compression parameters from previous init; skip dictionary loading stage; zcs must be init at least once before */
-ZSTDLIB_API size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs);
+ZSTDLIB_API size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs, const ZSTD_CDict* cdict, ZSTD_frameParameters fParams, unsigned long long pledgedSrcSize);  /**< same as ZSTD_initCStream_usingCDict(), with control over frame parameters. pledgedSrcSize must be correct. If srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN. */
+
+/*! ZSTD_resetCStream() :
+ *  start a new compression job, using same parameters from previous job.
+ *  This is typically useful to skip dictionary loading stage, since it will re-use it in-place..
+ *  Note that zcs must be init at least once before using ZSTD_resetCStream().
+ *  If pledgedSrcSize is not known at reset time, use macro ZSTD_CONTENTSIZE_UNKNOWN.
+ *  If pledgedSrcSize > 0, its value must be correct, as it will be written in header, and controlled at the end.
+ *  For the time being, pledgedSrcSize==0 is interpreted as "srcSize unknown" for compatibility with older programs,
+ *  but it will change to mean "empty" in future version, so use macro ZSTD_CONTENTSIZE_UNKNOWN instead.
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
+ZSTDLIB_API size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
+
+
+typedef struct {
+    unsigned long long ingested;
+    unsigned long long consumed;
+    unsigned long long produced;
+} ZSTD_frameProgression;
+
+/* ZSTD_getFrameProgression():
+ * tells how much data has been ingested (read from input)
+ * consumed (input actually compressed) and produced (output) for current frame.
+ * Therefore, (ingested - consumed) is amount of input data buffered internally, not yet compressed.
+ * Can report progression inside worker threads (multi-threading and non-blocking mode).
+ */
+ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx);
+
 
 
 /*=====   Advanced Streaming decompression functions  =====*/
 typedef enum { DStream_p_maxWindowSize } ZSTD_DStreamParameter_e;
-ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem);
-ZSTDLIB_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize); /**< note: a dict will not be used if dict == NULL or dictSize < 8 */
-ZSTDLIB_API size_t ZSTD_setDStreamParameter(ZSTD_DStream* zds, ZSTD_DStreamParameter_e paramType, unsigned paramValue);
-ZSTDLIB_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);  /**< note : ddict will just be referenced, and must outlive decompression session */
+ZSTDLIB_API size_t ZSTD_setDStreamParameter(ZSTD_DStream* zds, ZSTD_DStreamParameter_e paramType, unsigned paramValue);   /* obsolete : this API will be removed in a future version */
+ZSTDLIB_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize); /**< note: no dictionary will be used if dict == NULL or dictSize < 8 */
+ZSTDLIB_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);  /**< note : ddict is referenced, it must outlive decompression session */
 ZSTDLIB_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);  /**< re-use decompression parameters from previous init; saves dictionary loading */
-ZSTDLIB_API size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds);
 
 
 /*********************************************************************
 *  Buffer-less and synchronous inner streaming functions
 *
 *  This is an advanced API, giving full control over buffer management, for users which need direct control over memory.
-*  But it's also a complex one, with many restrictions (documented below).
-*  Prefer using normal streaming API for an easier experience
+*  But it's also a complex one, with several restrictions, documented below.
+*  Prefer normal streaming API for an easier experience.
 ********************************************************************* */
 
 /**
@@ -558,8 +767,8 @@
 
   Then, consume your input using ZSTD_compressContinue().
   There are some important considerations to keep in mind when using this advanced function :
-  - ZSTD_compressContinue() has no internal buffer. It uses externally provided buffer only.
-  - Interface is synchronous : input is consumed entirely and produce 1+ (or more) compressed blocks.
+  - ZSTD_compressContinue() has no internal buffer. It uses externally provided buffers only.
+  - Interface is synchronous : input is consumed entirely and produces 1+ compressed blocks.
   - Caller must ensure there is enough space in `dst` to store compressed data under worst case scenario.
     Worst case evaluation is provided by ZSTD_compressBound().
     ZSTD_compressContinue() doesn't guarantee recover after a failed compression.
@@ -570,22 +779,23 @@
 
   Finish a frame with ZSTD_compressEnd(), which will write the last block(s) and optional checksum.
   It's possible to use srcSize==0, in which case, it will write a final empty block to end the frame.
-  Without last block mark, frames will be considered unfinished (corrupted) by decoders.
+  Without last block mark, frames are considered unfinished (hence corrupted) by compliant decoders.
 
-  `ZSTD_CCtx` object can be re-used (ZSTD_compressBegin()) to compress some new frame.
+  `ZSTD_CCtx` object can be re-used (ZSTD_compressBegin()) to compress again.
 */
 
 /*=====   Buffer-less streaming compression functions  =====*/
 ZSTDLIB_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
 ZSTDLIB_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
-ZSTDLIB_API size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize);
-ZSTDLIB_API size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize);
-ZSTDLIB_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict, unsigned long long pledgedSrcSize);
+ZSTDLIB_API size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize); /**< pledgedSrcSize : If srcSize is not known at init time, use ZSTD_CONTENTSIZE_UNKNOWN */
+ZSTDLIB_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /**< note: fails if cdict==NULL */
+ZSTDLIB_API size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict, ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize);   /* compression parameters are already set within cdict. pledgedSrcSize must be correct. If srcSize is not known, use macro ZSTD_CONTENTSIZE_UNKNOWN */
+ZSTDLIB_API size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /**<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
+
 ZSTDLIB_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 ZSTDLIB_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
 
-
 /*-
   Buffer-less streaming decompression (synchronous mode)
 
@@ -593,38 +803,54 @@
   Use ZSTD_createDCtx() / ZSTD_freeDCtx() to manage it.
   A ZSTD_DCtx object can be re-used multiple times.
 
-  First typical operation is to retrieve frame parameters, using ZSTD_getFrameParams().
-  It fills a ZSTD_frameParams structure which provide important information to correctly decode the frame,
-  such as the minimum rolling buffer size to allocate to decompress data (`windowSize`),
-  and the dictionary ID used.
-  (Note : content size is optional, it may not be present. 0 means : content size unknown).
-  Note that these values could be wrong, either because of data malformation, or because an attacker is spoofing deliberate false information.
-  As a consequence, check that values remain within valid application range, especially `windowSize`, before allocation.
-  Each application can set its own limit, depending on local restrictions. For extended interoperability, it is recommended to support at least 8 MB.
-  Frame parameters are extracted from the beginning of the compressed frame.
-  Data fragment must be large enough to ensure successful decoding, typically `ZSTD_frameHeaderSize_max` bytes.
-  @result : 0 : successful decoding, the `ZSTD_frameParams` structure is correctly filled.
+  First typical operation is to retrieve frame parameters, using ZSTD_getFrameHeader().
+  Frame header is extracted from the beginning of compressed frame, so providing only the frame's beginning is enough.
+  Data fragment must be large enough to ensure successful decoding.
+ `ZSTD_frameHeaderSize_max` bytes is guaranteed to always be large enough.
+  @result : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
            >0 : `srcSize` is too small, please provide at least @result bytes on next attempt.
            errorCode, which can be tested using ZSTD_isError().
 
-  Start decompression, with ZSTD_decompressBegin() or ZSTD_decompressBegin_usingDict().
-  Alternatively, you can copy a prepared context, using ZSTD_copyDCtx().
+  It fills a ZSTD_frameHeader structure with important information to correctly decode the frame,
+  such as the dictionary ID, content size, or maximum back-reference distance (`windowSize`).
+  Note that these values could be wrong, either because of data corruption, or because a 3rd party deliberately spoofs false information.
+  As a consequence, check that values remain within valid application range.
+  For example, do not allocate memory blindly, check that `windowSize` is within expectation.
+  Each application can set its own limits, depending on local restrictions.
+  For extended interoperability, it is recommended to support `windowSize` of at least 8 MB.
+
+  ZSTD_decompressContinue() needs previous data blocks during decompression, up to `windowSize` bytes.
+  ZSTD_decompressContinue() is very sensitive to contiguity,
+  if 2 blocks don't follow each other, make sure that either the compressor breaks contiguity at the same place,
+  or that previous contiguous segment is large enough to properly handle maximum back-reference distance.
+  There are multiple ways to guarantee this condition.
+
+  The most memory efficient way is to use a round buffer of sufficient size.
+  Sufficient size is determined by invoking ZSTD_decodingBufferSize_min(),
+  which can @return an error code if required value is too large for current system (in 32-bits mode).
+  In a round buffer methodology, ZSTD_decompressContinue() decompresses each block next to previous one,
+  up to the moment there is not enough room left in the buffer to guarantee decoding another full block,
+  which maximum size is provided in `ZSTD_frameHeader` structure, field `blockSizeMax`.
+  At which point, decoding can resume from the beginning of the buffer.
+  Note that already decoded data stored in the buffer should be flushed before being overwritten.
+
+  There are alternatives possible, for example using two or more buffers of size `windowSize` each, though they consume more memory.
+
+  Finally, if you control the compression process, you can also ignore all buffer size rules,
+  as long as the encoder and decoder progress in "lock-step",
+  aka use exactly the same buffer sizes, break contiguity at the same place, etc.
+
+  Once buffers are setup, start decompression, with ZSTD_decompressBegin().
+  If decompression requires a dictionary, use ZSTD_decompressBegin_usingDict() or ZSTD_decompressBegin_usingDDict().
 
   Then use ZSTD_nextSrcSizeToDecompress() and ZSTD_decompressContinue() alternatively.
   ZSTD_nextSrcSizeToDecompress() tells how many bytes to provide as 'srcSize' to ZSTD_decompressContinue().
   ZSTD_decompressContinue() requires this _exact_ amount of bytes, or it will fail.
 
-  @result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
-  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some metadata item.
+ @result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
+  It can be zero : it just means ZSTD_decompressContinue() has decoded some metadata item.
   It can also be an error code, which can be tested with ZSTD_isError().
 
-  ZSTD_decompressContinue() needs previous data blocks during decompression, up to `windowSize`.
-  They should preferably be located contiguously, prior to current block.
-  Alternatively, a round buffer of sufficient size is also possible. Sufficient size is determined by frame parameters.
-  ZSTD_decompressContinue() is very sensitive to contiguity,
-  if 2 blocks don't follow each other, make sure that either the compressor breaks contiguity at the same place,
-  or that previous contiguous segment is large enough to properly handle maximum back-reference.
-
   A frame is fully decoded when ZSTD_nextSrcSizeToDecompress() returns zero.
   Context can then be reset to start a new decompression.
 
@@ -634,35 +860,505 @@
   == Special case : skippable frames ==
 
   Skippable frames allow integration of user-defined data into a flow of concatenated frames.
-  Skippable frames will be ignored (skipped) by a decompressor. The format of skippable frames is as follows :
+  Skippable frames will be ignored (skipped) by decompressor.
+  The format of skippable frames is as follows :
   a) Skippable frame ID - 4 Bytes, Little endian format, any value from 0x184D2A50 to 0x184D2A5F
   b) Frame Size - 4 Bytes, Little endian format, unsigned 32-bits
   c) Frame Content - any content (User Data) of length equal to Frame Size
-  For skippable frames ZSTD_decompressContinue() always returns 0.
-  For skippable frames ZSTD_getFrameParams() returns fparamsPtr->windowLog==0 what means that a frame is skippable.
-  It also returns Frame Size as fparamsPtr->frameContentSize.
+  For skippable frames ZSTD_getFrameHeader() returns zfhPtr->frameType==ZSTD_skippableFrame.
+  For skippable frames ZSTD_decompressContinue() always returns 0 : it only skips the content.
 */
 
+/*=====   Buffer-less streaming decompression functions  =====*/
+typedef enum { ZSTD_frame, ZSTD_skippableFrame } ZSTD_frameType_e;
 typedef struct {
-    unsigned long long frameContentSize;
-    unsigned windowSize;
+    unsigned long long frameContentSize; /* if == ZSTD_CONTENTSIZE_UNKNOWN, it means this field is not available. 0 means "empty" */
+    unsigned long long windowSize;       /* can be very large, up to <= frameContentSize */
+    unsigned blockSizeMax;
+    ZSTD_frameType_e frameType;          /* if == ZSTD_skippableFrame, frameContentSize is the size of skippable content */
+    unsigned headerSize;
     unsigned dictID;
     unsigned checksumFlag;
-} ZSTD_frameParams;
+} ZSTD_frameHeader;
+ZSTDLIB_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
+ZSTDLIB_API size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize);  /**< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
 
-/*=====   Buffer-less streaming decompression functions  =====*/
-ZSTDLIB_API size_t ZSTD_getFrameParams(ZSTD_frameParams* fparamsPtr, const void* src, size_t srcSize);   /**< doesn't consume input, see details below */
 ZSTDLIB_API size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx);
 ZSTDLIB_API size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
-ZSTDLIB_API void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
+ZSTDLIB_API size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
+
 ZSTDLIB_API size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
 ZSTDLIB_API size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+/* misc */
+ZSTDLIB_API void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
 typedef enum { ZSTDnit_frameHeader, ZSTDnit_blockHeader, ZSTDnit_block, ZSTDnit_lastBlock, ZSTDnit_checksum, ZSTDnit_skippableFrame } ZSTD_nextInputType_e;
 ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
 
-/**
-    Block functions
+
+
+/* ============================================ */
+/**       New advanced API (experimental)       */
+/* ============================================ */
+
+/* notes on API design :
+ *   In this proposal, parameters are pushed one by one into an existing context,
+ *   and then applied on all subsequent compression jobs.
+ *   When no parameter is ever provided, CCtx is created with compression level ZSTD_CLEVEL_DEFAULT.
+ *
+ *   This API is intended to replace all others advanced / experimental API entry points.
+ *   But it stands a reasonable chance to become "stable", after a reasonable testing period.
+ */
+
+/* note on naming convention :
+ *   Initially, the API favored names like ZSTD_setCCtxParameter() .
+ *   In this proposal, convention is changed towards ZSTD_CCtx_setParameter() .
+ *   The main driver is that it identifies more clearly the target object type.
+ *   It feels clearer when considering multiple targets :
+ *   ZSTD_CDict_setParameter() (rather than ZSTD_setCDictParameter())
+ *   ZSTD_CCtxParams_setParameter()  (rather than ZSTD_setCCtxParamsParameter() )
+ *   etc...
+ */
+
+/* note on enum design :
+ * All enum will be pinned to explicit values before reaching "stable API" status */
+
+typedef enum {
+    /* Opened question : should we have a format ZSTD_f_auto ?
+     * Today, it would mean exactly the same as ZSTD_f_zstd1.
+     * But, in the future, should several formats become supported,
+     * on the compression side, it would mean "default format".
+     * On the decompression side, it would mean "automatic format detection",
+     * so that ZSTD_f_zstd1 would mean "accept *only* zstd frames".
+     * Since meaning is a little different, another option could be to define different enums for compression and decompression.
+     * This question could be kept for later, when there are actually multiple formats to support,
+     * but there is also the question of pinning enum values, and pinning value `0` is especially important */
+    ZSTD_f_zstd1 = 0,        /* zstd frame format, specified in zstd_compression_format.md (default) */
+    ZSTD_f_zstd1_magicless,  /* Variant of zstd frame format, without initial 4-bytes magic number.
+                              * Useful to save 4 bytes per generated frame.
+                              * Decoder cannot recognise automatically this format, requiring instructions. */
+} ZSTD_format_e;
+
+typedef enum {
+    /* compression format */
+    ZSTD_p_format = 10,      /* See ZSTD_format_e enum definition.
+                              * Cast selected format as unsigned for ZSTD_CCtx_setParameter() compatibility. */
+
+    /* compression parameters */
+    ZSTD_p_compressionLevel=100, /* Update all compression parameters according to pre-defined cLevel table
+                              * Default level is ZSTD_CLEVEL_DEFAULT==3.
+                              * Special: value 0 means "do not change cLevel".
+                              * Note 1 : it's possible to pass a negative compression level by casting it to unsigned type.
+                              * Note 2 : setting a level sets all default values of other compression parameters.
+                              * Note 3 : setting compressionLevel automatically updates ZSTD_p_compressLiterals. */
+    ZSTD_p_windowLog,        /* Maximum allowed back-reference distance, expressed as power of 2.
+                              * Must be clamped between ZSTD_WINDOWLOG_MIN and ZSTD_WINDOWLOG_MAX.
+                              * Special: value 0 means "use default windowLog".
+                              * Note: Using a window size greater than ZSTD_MAXWINDOWSIZE_DEFAULT (default: 2^27)
+                              *       requires explicitly allowing such window size during decompression stage. */
+    ZSTD_p_hashLog,          /* Size of the probe table, as a power of 2.
+                              * Resulting table size is (1 << (hashLog+2)).
+                              * Must be clamped between ZSTD_HASHLOG_MIN and ZSTD_HASHLOG_MAX.
+                              * Larger tables improve compression ratio of strategies <= dFast,
+                              * and improve speed of strategies > dFast.
+                              * Special: value 0 means "use default hashLog". */
+    ZSTD_p_chainLog,         /* Size of the full-search table, as a power of 2.
+                              * Resulting table size is (1 << (chainLog+2)).
+                              * Larger tables result in better and slower compression.
+                              * This parameter is useless when using "fast" strategy.
+                              * Special: value 0 means "use default chainLog". */
+    ZSTD_p_searchLog,        /* Number of search attempts, as a power of 2.
+                              * More attempts result in better and slower compression.
+                              * This parameter is useless when using "fast" and "dFast" strategies.
+                              * Special: value 0 means "use default searchLog". */
+    ZSTD_p_minMatch,         /* Minimum size of searched matches (note : repCode matches can be smaller).
+                              * Larger values make faster compression and decompression, but decrease ratio.
+                              * Must be clamped between ZSTD_SEARCHLENGTH_MIN and ZSTD_SEARCHLENGTH_MAX.
+                              * Note that currently, for all strategies < btopt, effective minimum is 4.
+                              *                    , for all strategies > fast, effective maximum is 6.
+                              * Special: value 0 means "use default minMatchLength". */
+    ZSTD_p_targetLength,     /* Impact of this field depends on strategy.
+                              * For strategies btopt & btultra:
+                              *     Length of Match considered "good enough" to stop search.
+                              *     Larger values make compression stronger, and slower.
+                              * For strategy fast:
+                              *     Distance between match sampling.
+                              *     Larger values make compression faster, and weaker.
+                              * Special: value 0 means "use default targetLength". */
+    ZSTD_p_compressionStrategy, /* See ZSTD_strategy enum definition.
+                              * Cast selected strategy as unsigned for ZSTD_CCtx_setParameter() compatibility.
+                              * The higher the value of selected strategy, the more complex it is,
+                              * resulting in stronger and slower compression.
+                              * Special: value 0 means "use default strategy". */
+
+    ZSTD_p_enableLongDistanceMatching=160, /* Enable long distance matching.
+                                         * This parameter is designed to improve compression ratio
+                                         * for large inputs, by finding large matches at long distance.
+                                         * It increases memory usage and window size.
+                                         * Note: enabling this parameter increases ZSTD_p_windowLog to 128 MB
+                                         * except when expressly set to a different value. */
+    ZSTD_p_ldmHashLog,       /* Size of the table for long distance matching, as a power of 2.
+                              * Larger values increase memory usage and compression ratio,
+                              * but decrease compression speed.
+                              * Must be clamped between ZSTD_HASHLOG_MIN and ZSTD_HASHLOG_MAX
+                              * default: windowlog - 7.
+                              * Special: value 0 means "automatically determine hashlog". */
+    ZSTD_p_ldmMinMatch,      /* Minimum match size for long distance matcher.
+                              * Larger/too small values usually decrease compression ratio.
+                              * Must be clamped between ZSTD_LDM_MINMATCH_MIN and ZSTD_LDM_MINMATCH_MAX.
+                              * Special: value 0 means "use default value" (default: 64). */
+    ZSTD_p_ldmBucketSizeLog, /* Log size of each bucket in the LDM hash table for collision resolution.
+                              * Larger values improve collision resolution but decrease compression speed.
+                              * The maximum value is ZSTD_LDM_BUCKETSIZELOG_MAX .
+                              * Special: value 0 means "use default value" (default: 3). */
+    ZSTD_p_ldmHashEveryLog,  /* Frequency of inserting/looking up entries in the LDM hash table.
+                              * Must be clamped between 0 and (ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN).
+                              * Default is MAX(0, (windowLog - ldmHashLog)), optimizing hash table usage.
+                              * Larger values improve compression speed.
+                              * Deviating far from default value will likely result in a compression ratio decrease.
+                              * Special: value 0 means "automatically determine hashEveryLog". */
+
+    /* frame parameters */
+    ZSTD_p_contentSizeFlag=200, /* Content size will be written into frame header _whenever known_ (default:1)
+                              * Content size must be known at the beginning of compression,
+                              * it is provided using ZSTD_CCtx_setPledgedSrcSize() */
+    ZSTD_p_checksumFlag,     /* A 32-bits checksum of content is written at end of frame (default:0) */
+    ZSTD_p_dictIDFlag,       /* When applicable, dictionary's ID is written into frame header (default:1) */
+
+    /* multi-threading parameters */
+    /* These parameters are only useful if multi-threading is enabled (ZSTD_MULTITHREAD).
+     * They return an error otherwise. */
+    ZSTD_p_nbWorkers=400,    /* Select how many threads will be spawned to compress in parallel.
+                              * When nbWorkers >= 1, triggers asynchronous mode :
+                              * ZSTD_compress_generic() consumes some input, flush some output if possible, and immediately gives back control to caller,
+                              * while compression work is performed in parallel, within worker threads.
+                              * (note : a strong exception to this rule is when first invocation sets ZSTD_e_end : it becomes a blocking call).
+                              * More workers improve speed, but also increase memory usage.
+                              * Default value is `0`, aka "single-threaded mode" : no worker is spawned, compression is performed inside Caller's thread, all invocations are blocking */
+    ZSTD_p_jobSize,          /* Size of a compression job. This value is enforced only in non-blocking mode.
+                              * Each compression job is completed in parallel, so this value indirectly controls the nb of active threads.
+                              * 0 means default, which is dynamically determined based on compression parameters.
+                              * Job size must be a minimum of overlapSize, or 1 MB, whichever is largest.
+                              * The minimum size is automatically and transparently enforced */
+    ZSTD_p_overlapSizeLog,   /* Size of previous input reloaded at the beginning of each job.
+                              * 0 => no overlap, 6(default) => use 1/8th of windowSize, >=9 => use full windowSize */
+
+    /* =================================================================== */
+    /* experimental parameters - no stability guaranteed                   */
+    /* =================================================================== */
+
+    ZSTD_p_compressLiterals=1000, /* control huffman compression of literals (enabled) by default.
+                              * disabling it improves speed and decreases compression ratio by a large amount.
+                              * note : this setting is automatically updated when changing compression level.
+                              *        positive compression levels set ZSTD_p_compressLiterals to 1.
+                              *        negative compression levels set ZSTD_p_compressLiterals to 0. */
+
+    ZSTD_p_forceMaxWindow=1100, /* Force back-reference distances to remain < windowSize,
+                              * even when referencing into Dictionary content (default:0) */
+
+} ZSTD_cParameter;
+
+
+/*! ZSTD_CCtx_setParameter() :
+ *  Set one compression parameter, selected by enum ZSTD_cParameter.
+ *  Setting a parameter is generally only possible during frame initialization (before starting compression),
+ *  except for a few exceptions which can be updated during compression: compressionLevel, hashLog, chainLog, searchLog, minMatch, targetLength and strategy.
+ *  Note : when `value` is an enum, cast it to unsigned for proper type checking.
+ *  @result : informational value (typically, value being set clamped correctly),
+ *            or an error code (which can be tested with ZSTD_isError()). */
+ZSTDLIB_API size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned value);
+
+/*! ZSTD_CCtx_setPledgedSrcSize() :
+ *  Total input data size to be compressed as a single frame.
+ *  This value will be controlled at the end, and result in error if not respected.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Note 1 : 0 means zero, empty.
+ *           In order to mean "unknown content size", pass constant ZSTD_CONTENTSIZE_UNKNOWN.
+ *           ZSTD_CONTENTSIZE_UNKNOWN is default value for any new compression job.
+ *  Note 2 : If all data is provided and consumed in a single round,
+ *           this value is overriden by srcSize instead. */
+ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize);
+
+/*! ZSTD_CCtx_loadDictionary() :
+ *  Create an internal CDict from `dict` buffer.
+ *  Decompression will have to use same dictionary.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special: Adding a NULL (or 0-size) dictionary invalidates previous dictionary,
+ *           meaning "return to no-dictionary mode".
+ *  Note 1 : Dictionary will be used for all future compression jobs.
+ *           To return to "no-dictionary" situation, load a NULL dictionary
+ *  Note 2 : Loading a dictionary involves building tables, which are dependent on compression parameters.
+ *           For this reason, compression parameters cannot be changed anymore after loading a dictionary.
+ *           It's also a CPU consuming operation, with non-negligible impact on latency.
+ *  Note 3 :`dict` content will be copied internally.
+ *           Use ZSTD_CCtx_loadDictionary_byReference() to reference dictionary content instead.
+ *           In such a case, dictionary buffer must outlive its users.
+ *  Note 4 : Use ZSTD_CCtx_loadDictionary_advanced()
+ *           to precisely select how dictionary content must be interpreted. */
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_byReference(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+
+
+/*! ZSTD_CCtx_refCDict() :
+ *  Reference a prepared dictionary, to be used for all next compression jobs.
+ *  Note that compression parameters are enforced from within CDict,
+ *  and supercede any compression parameter previously set within CCtx.
+ *  The dictionary will remain valid for future compression jobs using same CCtx.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special : adding a NULL CDict means "return to no-dictionary mode".
+ *  Note 1 : Currently, only one dictionary can be managed.
+ *           Adding a new dictionary effectively "discards" any previous one.
+ *  Note 2 : CDict is just referenced, its lifetime must outlive CCtx. */
+ZSTDLIB_API size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
 
+/*! ZSTD_CCtx_refPrefix() :
+ *  Reference a prefix (single-usage dictionary) for next compression job.
+ *  Decompression need same prefix to properly regenerate data.
+ *  Prefix is **only used once**. Tables are discarded at end of compression job.
+ *  Subsequent compression jobs will be done without prefix (if none is explicitly referenced).
+ *  If there is a need to use same prefix multiple times, consider embedding it into a ZSTD_CDict instead.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special: Adding any prefix (including NULL) invalidates any previous prefix or dictionary
+ *  Note 1 : Prefix buffer is referenced. It must outlive compression job.
+ *  Note 2 : Referencing a prefix involves building tables, which are dependent on compression parameters.
+ *           It's a CPU consuming operation, with non-negligible impact on latency.
+ *  Note 3 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
+ *           Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode. */
+ZSTDLIB_API size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize);
+ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+
+/*! ZSTD_CCtx_reset() :
+ *  Return a CCtx to clean state.
+ *  Useful after an error, or to interrupt an ongoing compression job and start a new one.
+ *  Any internal data not yet flushed is cancelled.
+ *  Dictionary (if any) is dropped.
+ *  All parameters are back to default values.
+ *  It's possible to modify compression parameters after a reset.
+ */
+ZSTDLIB_API void ZSTD_CCtx_reset(ZSTD_CCtx* cctx);
+
+
+
+typedef enum {
+    ZSTD_e_continue=0, /* collect more data, encoder decides when to output compressed result, for optimal conditions */
+    ZSTD_e_flush,      /* flush any data provided so far - frame will continue, future data can still reference previous data for better compression */
+    ZSTD_e_end         /* flush any remaining data and close current frame. Any additional data starts a new frame. */
+} ZSTD_EndDirective;
+
+/*! ZSTD_compress_generic() :
+ *  Behave about the same as ZSTD_compressStream. To note :
+ *  - Compression parameters are pushed into CCtx before starting compression, using ZSTD_CCtx_setParameter()
+ *  - Compression parameters cannot be changed once compression is started.
+ *  - outpot->pos must be <= dstCapacity, input->pos must be <= srcSize
+ *  - outpot->pos and input->pos will be updated. They are guaranteed to remain below their respective limit.
+ *  - In single-thread mode (default), function is blocking : it completed its job before returning to caller.
+ *  - In multi-thread mode, function is non-blocking : it just acquires a copy of input, and distribute job to internal worker threads,
+ *                                                     and then immediately returns, just indicating that there is some data remaining to be flushed.
+ *                                                     The function nonetheless guarantees forward progress : it will return only after it reads or write at least 1+ byte.
+ *  - Exception : in multi-threading mode, if the first call requests a ZSTD_e_end directive, it is blocking : it will complete compression before giving back control to caller.
+ *  - @return provides a minimum amount of data remaining to be flushed from internal buffers
+ *            or an error code, which can be tested using ZSTD_isError().
+ *            if @return != 0, flush is not fully completed, there is still some data left within internal buffers.
+ *            This is useful for ZSTD_e_flush, since in this case more flushes are necessary to empty all buffers.
+ *            For ZSTD_e_end, @return == 0 when internal buffers are fully flushed and frame is completed.
+ *  - after a ZSTD_e_end directive, if internal buffer is not fully flushed (@return != 0),
+ *            only ZSTD_e_end or ZSTD_e_flush operations are allowed.
+ *            Before starting a new compression job, or changing compression parameters,
+ *            it is required to fully flush internal buffers.
+ */
+ZSTDLIB_API size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
+                                          ZSTD_outBuffer* output,
+                                          ZSTD_inBuffer* input,
+                                          ZSTD_EndDirective endOp);
+
+
+/*! ZSTD_compress_generic_simpleArgs() :
+ *  Same as ZSTD_compress_generic(),
+ *  but using only integral types as arguments.
+ *  Argument list is larger than ZSTD_{in,out}Buffer,
+ *  but can be helpful for binders from dynamic languages
+ *  which have troubles handling structures containing memory pointers.
+ */
+ZSTDLIB_API size_t ZSTD_compress_generic_simpleArgs (
+                            ZSTD_CCtx* cctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos,
+                            ZSTD_EndDirective endOp);
+
+
+/*! ZSTD_CCtx_params :
+ *  Quick howto :
+ *  - ZSTD_createCCtxParams() : Create a ZSTD_CCtx_params structure
+ *  - ZSTD_CCtxParam_setParameter() : Push parameters one by one into
+ *                                    an existing ZSTD_CCtx_params structure.
+ *                                    This is similar to
+ *                                    ZSTD_CCtx_setParameter().
+ *  - ZSTD_CCtx_setParametersUsingCCtxParams() : Apply parameters to
+ *                                    an existing CCtx.
+ *                                    These parameters will be applied to
+ *                                    all subsequent compression jobs.
+ *  - ZSTD_compress_generic() : Do compression using the CCtx.
+ *  - ZSTD_freeCCtxParams() : Free the memory.
+ *
+ *  This can be used with ZSTD_estimateCCtxSize_advanced_usingCCtxParams()
+ *  for static allocation for single-threaded compression.
+ */
+ZSTDLIB_API ZSTD_CCtx_params* ZSTD_createCCtxParams(void);
+ZSTDLIB_API size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);
+
+
+/*! ZSTD_CCtxParams_reset() :
+ *  Reset params to default values.
+ */
+ZSTDLIB_API size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params);
+
+/*! ZSTD_CCtxParams_init() :
+ *  Initializes the compression parameters of cctxParams according to
+ *  compression level. All other parameters are reset to their default values.
+ */
+ZSTDLIB_API size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel);
+
+/*! ZSTD_CCtxParams_init_advanced() :
+ *  Initializes the compression and frame parameters of cctxParams according to
+ *  params. All other parameters are reset to their default values.
+ */
+ZSTDLIB_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params);
+
+
+/*! ZSTD_CCtxParam_setParameter() :
+ *  Similar to ZSTD_CCtx_setParameter.
+ *  Set one compression parameter, selected by enum ZSTD_cParameter.
+ *  Parameters must be applied to a ZSTD_CCtx using ZSTD_CCtx_setParametersUsingCCtxParams().
+ *  Note : when `value` is an enum, cast it to unsigned for proper type checking.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_CCtxParam_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, unsigned value);
+
+/*! ZSTD_CCtx_setParametersUsingCCtxParams() :
+ *  Apply a set of ZSTD_CCtx_params to the compression context.
+ *  This can be done even after compression is started,
+ *    if nbWorkers==0, this will have no impact until a new compression is started.
+ *    if nbWorkers>=1, new parameters will be picked up at next job,
+ *       with a few restrictions (windowLog, pledgedSrcSize, nbWorkers, jobSize, and overlapLog are not updated).
+ */
+ZSTDLIB_API size_t ZSTD_CCtx_setParametersUsingCCtxParams(
+        ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params);
+
+
+/*===   Advanced parameters for decompression API  ===*/
+
+/* The following parameters must be set after creating a ZSTD_DCtx* (or ZSTD_DStream*) object,
+ * but before starting decompression of a frame.
+ */
+
+/*! ZSTD_DCtx_loadDictionary() :
+ *  Create an internal DDict from dict buffer,
+ *  to be used to decompress next frames.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special : Adding a NULL (or 0-size) dictionary invalidates any previous dictionary,
+ *            meaning "return to no-dictionary mode".
+ *  Note 1 : `dict` content will be copied internally.
+ *            Use ZSTD_DCtx_loadDictionary_byReference()
+ *            to reference dictionary content instead.
+ *            In which case, the dictionary buffer must outlive its users.
+ *  Note 2 : Loading a dictionary involves building tables,
+ *           which has a non-negligible impact on CPU usage and latency.
+ *  Note 3 : Use ZSTD_DCtx_loadDictionary_advanced() to select
+ *           how dictionary content will be interpreted and loaded.
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+
+
+/*! ZSTD_DCtx_refDDict() :
+ *  Reference a prepared dictionary, to be used to decompress next frames.
+ *  The dictionary remains active for decompression of future frames using same DCtx.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Note 1 : Currently, only one dictionary can be managed.
+ *           Referencing a new dictionary effectively "discards" any previous one.
+ *  Special : adding a NULL DDict means "return to no-dictionary mode".
+ *  Note 2 : DDict is just referenced, its lifetime must outlive its usage from DCtx.
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
+
+
+/*! ZSTD_DCtx_refPrefix() :
+ *  Reference a prefix (single-usage dictionary) for next compression job.
+ *  Prefix is **only used once**. It must be explicitly referenced before each frame.
+ *  If there is a need to use same prefix multiple times, consider embedding it into a ZSTD_DDict instead.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Note 1 : Adding any prefix (including NULL) invalidates any previously set prefix or dictionary
+ *  Note 2 : Prefix buffer is referenced. It must outlive compression job.
+ *  Note 3 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
+ *           Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode.
+ *  Note 4 : Referencing a raw content prefix has almost no cpu nor memory cost.
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize);
+ZSTDLIB_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+
+
+/*! ZSTD_DCtx_setMaxWindowSize() :
+ *  Refuses allocating internal buffers for frames requiring a window size larger than provided limit.
+ *  This is useful to prevent a decoder context from reserving too much memory for itself (potential attack scenario).
+ *  This parameter is only useful in streaming mode, since no internal buffer is allocated in direct mode.
+ *  By default, a decompression context accepts all window sizes <= (1 << ZSTD_WINDOWLOG_MAX)
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize);
+
+
+/*! ZSTD_DCtx_setFormat() :
+ *  Instruct the decoder context about what kind of data to decode next.
+ *  This instruction is mandatory to decode data without a fully-formed header,
+ *  such ZSTD_f_zstd1_magicless for example.
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
+
+
+/*! ZSTD_decompress_generic() :
+ *  Behave the same as ZSTD_decompressStream.
+ *  Decompression parameters cannot be changed once decompression is started.
+ * @return : an error code, which can be tested using ZSTD_isError()
+ *           if >0, a hint, nb of expected input bytes for next invocation.
+ *           `0` means : a frame has just been fully decoded and flushed.
+ */
+ZSTDLIB_API size_t ZSTD_decompress_generic(ZSTD_DCtx* dctx,
+                                           ZSTD_outBuffer* output,
+                                           ZSTD_inBuffer* input);
+
+
+/*! ZSTD_decompress_generic_simpleArgs() :
+ *  Same as ZSTD_decompress_generic(),
+ *  but using only integral types as arguments.
+ *  Argument list is larger than ZSTD_{in,out}Buffer,
+ *  but can be helpful for binders from dynamic languages
+ *  which have troubles handling structures containing memory pointers.
+ */
+ZSTDLIB_API size_t ZSTD_decompress_generic_simpleArgs (
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos);
+
+
+/*! ZSTD_DCtx_reset() :
+ *  Return a DCtx to clean state.
+ *  If a decompression was ongoing, any internal data not yet flushed is cancelled.
+ *  All parameters are back to default values, including sticky ones.
+ *  Dictionary (if any) is dropped.
+ *  Parameters can be modified again after a reset.
+ */
+ZSTDLIB_API void ZSTD_DCtx_reset(ZSTD_DCtx* dctx);
+
+
+
+/* ============================ */
+/**       Block level API       */
+/* ============================ */
+
+/*!
     Block functions produce and decode raw zstd blocks, without frame metadata.
     Frame metadata cost is typically ~18 bytes, which can be non-negligible for very small blocks (< 100 bytes).
     User will have to take in charge required information to regenerate data, such as compressed and content sizes.
@@ -671,27 +1367,29 @@
     - Compressing and decompressing require a context structure
       + Use ZSTD_createCCtx() and ZSTD_createDCtx()
     - It is necessary to init context before starting
-      + compression : ZSTD_compressBegin()
-      + decompression : ZSTD_decompressBegin()
-      + variants _usingDict() are also allowed
-      + copyCCtx() and copyDCtx() work too
-    - Block size is limited, it must be <= ZSTD_getBlockSizeMax()
-      + If you need to compress more, cut data into multiple blocks
-      + Consider using the regular ZSTD_compress() instead, as frame metadata costs become negligible when source size is large.
+      + compression : any ZSTD_compressBegin*() variant, including with dictionary
+      + decompression : any ZSTD_decompressBegin*() variant, including with dictionary
+      + copyCCtx() and copyDCtx() can be used too
+    - Block size is limited, it must be <= ZSTD_getBlockSize() <= ZSTD_BLOCKSIZE_MAX == 128 KB
+      + If input is larger than a block size, it's necessary to split input data into multiple blocks
+      + For inputs larger than a single block size, consider using the regular ZSTD_compress() instead.
+        Frame metadata is not that costly, and quickly becomes negligible as source size grows larger.
     - When a block is considered not compressible enough, ZSTD_compressBlock() result will be zero.
       In which case, nothing is produced into `dst`.
       + User must test for such outcome and deal directly with uncompressed data
       + ZSTD_decompressBlock() doesn't accept uncompressed data as input !!!
-      + In case of multiple successive blocks, decoder must be informed of uncompressed block existence to follow proper history.
-        Use ZSTD_insertBlock() in such a case.
+      + In case of multiple successive blocks, should some of them be uncompressed,
+        decoder must be informed of their existence in order to follow proper history.
+        Use ZSTD_insertBlock() for such a case.
 */
 
-#define ZSTD_BLOCKSIZE_ABSOLUTEMAX (128 * 1024)   /* define, for static allocation */
+#define ZSTD_BLOCKSIZELOG_MAX 17
+#define ZSTD_BLOCKSIZE_MAX   (1<<ZSTD_BLOCKSIZELOG_MAX)   /* define, for static allocation */
 /*=====   Raw zstd block functions  =====*/
-ZSTDLIB_API size_t ZSTD_getBlockSizeMax(ZSTD_CCtx* cctx);
+ZSTDLIB_API size_t ZSTD_getBlockSize   (const ZSTD_CCtx* cctx);
 ZSTDLIB_API size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 ZSTDLIB_API size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
-ZSTDLIB_API size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert block into `dctx` history. Useful for uncompressed blocks */
+ZSTDLIB_API size_t ZSTD_insertBlock    (ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert uncompressed block into `dctx` history. Useful for multi-blocks decompression. */
 
 
 #endif   /* ZSTD_H_ZSTD_STATIC_LINKING_ONLY */
--- a/contrib/python-zstandard/zstd_cffi.py	Sun Apr 08 01:08:43 2018 +0200
+++ b/contrib/python-zstandard/zstd_cffi.py	Mon Apr 09 10:13:29 2018 -0700
@@ -8,6 +8,69 @@
 
 from __future__ import absolute_import, unicode_literals
 
+# This should match what the C extension exports.
+__all__ = [
+    #'BufferSegment',
+    #'BufferSegments',
+    #'BufferWithSegments',
+    #'BufferWithSegmentsCollection',
+    'CompressionParameters',
+    'ZstdCompressionDict',
+    'ZstdCompressionParameters',
+    'ZstdCompressor',
+    'ZstdError',
+    'ZstdDecompressor',
+    'FrameParameters',
+    'estimate_decompression_context_size',
+    'frame_content_size',
+    'frame_header_size',
+    'get_frame_parameters',
+    'train_dictionary',
+
+    # Constants.
+    'COMPRESSOBJ_FLUSH_FINISH',
+    'COMPRESSOBJ_FLUSH_BLOCK',
+    'ZSTD_VERSION',
+    'FRAME_HEADER',
+    'CONTENTSIZE_UNKNOWN',
+    'CONTENTSIZE_ERROR',
+    'MAX_COMPRESSION_LEVEL',
+    'COMPRESSION_RECOMMENDED_INPUT_SIZE',
+    'COMPRESSION_RECOMMENDED_OUTPUT_SIZE',
+    'DECOMPRESSION_RECOMMENDED_INPUT_SIZE',
+    'DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE',
+    'MAGIC_NUMBER',
+    'WINDOWLOG_MIN',
+    'WINDOWLOG_MAX',
+    'CHAINLOG_MIN',
+    'CHAINLOG_MAX',
+    'HASHLOG_MIN',
+    'HASHLOG_MAX',
+    'HASHLOG3_MAX',
+    'SEARCHLOG_MIN',
+    'SEARCHLOG_MAX',
+    'SEARCHLENGTH_MIN',
+    'SEARCHLENGTH_MAX',
+    'TARGETLENGTH_MIN',
+    'LDM_MINMATCH_MIN',
+    'LDM_MINMATCH_MAX',
+    'LDM_BUCKETSIZELOG_MAX',
+    'STRATEGY_FAST',
+    'STRATEGY_DFAST',
+    'STRATEGY_GREEDY',
+    'STRATEGY_LAZY',
+    'STRATEGY_LAZY2',
+    'STRATEGY_BTLAZY2',
+    'STRATEGY_BTOPT',
+    'STRATEGY_BTULTRA',
+    'DICT_TYPE_AUTO',
+    'DICT_TYPE_RAWCONTENT',
+    'DICT_TYPE_FULLDICT',
+    'FORMAT_ZSTD1',
+    'FORMAT_ZSTD1_MAGICLESS',
+]
+
+import io
 import os
 import sys
 
@@ -35,6 +98,8 @@
 MAX_COMPRESSION_LEVEL = lib.ZSTD_maxCLevel()
 MAGIC_NUMBER = lib.ZSTD_MAGICNUMBER
 FRAME_HEADER = b'\x28\xb5\x2f\xfd'
+CONTENTSIZE_UNKNOWN = lib.ZSTD_CONTENTSIZE_UNKNOWN
+CONTENTSIZE_ERROR = lib.ZSTD_CONTENTSIZE_ERROR
 ZSTD_VERSION = (lib.ZSTD_VERSION_MAJOR, lib.ZSTD_VERSION_MINOR, lib.ZSTD_VERSION_RELEASE)
 
 WINDOWLOG_MIN = lib.ZSTD_WINDOWLOG_MIN
@@ -49,7 +114,9 @@
 SEARCHLENGTH_MIN = lib.ZSTD_SEARCHLENGTH_MIN
 SEARCHLENGTH_MAX = lib.ZSTD_SEARCHLENGTH_MAX
 TARGETLENGTH_MIN = lib.ZSTD_TARGETLENGTH_MIN
-TARGETLENGTH_MAX = lib.ZSTD_TARGETLENGTH_MAX
+LDM_MINMATCH_MIN = lib.ZSTD_LDM_MINMATCH_MIN
+LDM_MINMATCH_MAX = lib.ZSTD_LDM_MINMATCH_MAX
+LDM_BUCKETSIZELOG_MAX = lib.ZSTD_LDM_BUCKETSIZELOG_MAX
 
 STRATEGY_FAST = lib.ZSTD_fast
 STRATEGY_DFAST = lib.ZSTD_dfast
@@ -58,6 +125,14 @@
 STRATEGY_LAZY2 = lib.ZSTD_lazy2
 STRATEGY_BTLAZY2 = lib.ZSTD_btlazy2
 STRATEGY_BTOPT = lib.ZSTD_btopt
+STRATEGY_BTULTRA = lib.ZSTD_btultra
+
+DICT_TYPE_AUTO = lib.ZSTD_dct_auto
+DICT_TYPE_RAWCONTENT = lib.ZSTD_dct_rawContent
+DICT_TYPE_FULLDICT = lib.ZSTD_dct_fullDict
+
+FORMAT_ZSTD1 = lib.ZSTD_f_zstd1
+FORMAT_ZSTD1_MAGICLESS = lib.ZSTD_f_zstd1_magicless
 
 COMPRESSOBJ_FLUSH_FINISH = 0
 COMPRESSOBJ_FLUSH_BLOCK = 1
@@ -87,81 +162,128 @@
     pass
 
 
-class CompressionParameters(object):
-    def __init__(self, window_log, chain_log, hash_log, search_log,
-                 search_length, target_length, strategy):
-        if window_log < WINDOWLOG_MIN or window_log > WINDOWLOG_MAX:
-            raise ValueError('invalid window log value')
+def _zstd_error(zresult):
+    # Resolves to bytes on Python 2 and 3. We use the string for formatting
+    # into error messages, which will be literal unicode. So convert it to
+    # unicode.
+    return ffi.string(lib.ZSTD_getErrorName(zresult)).decode('utf-8')
 
-        if chain_log < CHAINLOG_MIN or chain_log > CHAINLOG_MAX:
-            raise ValueError('invalid chain log value')
+def _make_cctx_params(params):
+    res = lib.ZSTD_createCCtxParams()
+    if res == ffi.NULL:
+        raise MemoryError()
+
+    res = ffi.gc(res, lib.ZSTD_freeCCtxParams)
 
-        if hash_log < HASHLOG_MIN or hash_log > HASHLOG_MAX:
-            raise ValueError('invalid hash log value')
+    attrs = [
+        (lib.ZSTD_p_format, params.format),
+        (lib.ZSTD_p_compressionLevel, params.compression_level),
+        (lib.ZSTD_p_windowLog, params.window_log),
+        (lib.ZSTD_p_hashLog, params.hash_log),
+        (lib.ZSTD_p_chainLog, params.chain_log),
+        (lib.ZSTD_p_searchLog, params.search_log),
+        (lib.ZSTD_p_minMatch, params.min_match),
+        (lib.ZSTD_p_targetLength, params.target_length),
+        (lib.ZSTD_p_compressionStrategy, params.compression_strategy),
+        (lib.ZSTD_p_contentSizeFlag, params.write_content_size),
+        (lib.ZSTD_p_checksumFlag, params.write_checksum),
+        (lib.ZSTD_p_dictIDFlag, params.write_dict_id),
+        (lib.ZSTD_p_nbWorkers, params.threads),
+        (lib.ZSTD_p_jobSize, params.job_size),
+        (lib.ZSTD_p_overlapSizeLog, params.overlap_size_log),
+        (lib.ZSTD_p_compressLiterals, params.compress_literals),
+        (lib.ZSTD_p_forceMaxWindow, params.force_max_window),
+        (lib.ZSTD_p_enableLongDistanceMatching, params.enable_ldm),
+        (lib.ZSTD_p_ldmHashLog, params.ldm_hash_log),
+        (lib.ZSTD_p_ldmMinMatch, params.ldm_min_match),
+        (lib.ZSTD_p_ldmBucketSizeLog, params.ldm_bucket_size_log),
+        (lib.ZSTD_p_ldmHashEveryLog, params.ldm_hash_every_log),
+    ]
 
-        if search_log < SEARCHLOG_MIN or search_log > SEARCHLOG_MAX:
-            raise ValueError('invalid search log value')
+    for param, value in attrs:
+        _set_compression_parameter(res, param, value)
+
+    return res
 
-        if search_length < SEARCHLENGTH_MIN or search_length > SEARCHLENGTH_MAX:
-            raise ValueError('invalid search length value')
+class ZstdCompressionParameters(object):
+    @staticmethod
+    def from_level(level, source_size=0, dict_size=0, **kwargs):
+        params = lib.ZSTD_getCParams(level, source_size, dict_size)
 
-        if target_length < TARGETLENGTH_MIN or target_length > TARGETLENGTH_MAX:
-            raise ValueError('invalid target length value')
+        args = {
+            'window_log': 'windowLog',
+            'chain_log': 'chainLog',
+            'hash_log': 'hashLog',
+            'search_log': 'searchLog',
+            'min_match': 'searchLength',
+            'target_length': 'targetLength',
+            'compression_strategy': 'strategy',
+        }
+
+        for arg, attr in args.items():
+            if arg not in kwargs:
+                kwargs[arg] = getattr(params, attr)
+
+        if 'compress_literals' not in kwargs:
+            kwargs['compress_literals'] = 1 if level >= 0 else 0
 
-        if strategy < STRATEGY_FAST or strategy > STRATEGY_BTOPT:
-            raise ValueError('invalid strategy value')
+        return ZstdCompressionParameters(**kwargs)
+
+    def __init__(self, format=0, compression_level=0, window_log=0, hash_log=0,
+                 chain_log=0, search_log=0, min_match=0, target_length=0,
+                 compression_strategy=0, write_content_size=1, write_checksum=0,
+                 write_dict_id=0, job_size=0, overlap_size_log=0,
+                 force_max_window=0, enable_ldm=0, ldm_hash_log=0,
+                 ldm_min_match=0, ldm_bucket_size_log=0, ldm_hash_every_log=0,
+                 threads=0, compress_literals=None):
 
+        if threads < 0:
+            threads = _cpu_count()
+
+        if compress_literals is None:
+            compress_literals = compression_level >= 0
+
+        self.format = format
+        self.compression_level = compression_level
         self.window_log = window_log
-        self.chain_log = chain_log
         self.hash_log = hash_log
+        self.chain_log = chain_log
         self.search_log = search_log
-        self.search_length = search_length
+        self.min_match = min_match
         self.target_length = target_length
-        self.strategy = strategy
+        self.compression_strategy = compression_strategy
+        self.write_content_size = write_content_size
+        self.write_checksum = write_checksum
+        self.write_dict_id = write_dict_id
+        self.job_size = job_size
+        self.overlap_size_log = overlap_size_log
+        self.compress_literals = compress_literals
+        self.force_max_window = force_max_window
+        self.enable_ldm = enable_ldm
+        self.ldm_hash_log = ldm_hash_log
+        self.ldm_min_match = ldm_min_match
+        self.ldm_bucket_size_log = ldm_bucket_size_log
+        self.ldm_hash_every_log = ldm_hash_every_log
+        self.threads = threads
 
-        zresult = lib.ZSTD_checkCParams(self.as_compression_parameters())
-        if lib.ZSTD_isError(zresult):
-            raise ValueError('invalid compression parameters: %s',
-                             ffi.string(lib.ZSTD_getErrorName(zresult)))
+        self.params = _make_cctx_params(self)
 
     def estimated_compression_context_size(self):
-        return lib.ZSTD_estimateCCtxSize(self.as_compression_parameters())
-
-    def as_compression_parameters(self):
-        p = ffi.new('ZSTD_compressionParameters *')[0]
-        p.windowLog = self.window_log
-        p.chainLog = self.chain_log
-        p.hashLog = self.hash_log
-        p.searchLog = self.search_log
-        p.searchLength = self.search_length
-        p.targetLength = self.target_length
-        p.strategy = self.strategy
-
-        return p
+        return lib.ZSTD_estimateCCtxSize_usingCCtxParams(self.params)
 
-def get_compression_parameters(level, source_size=0, dict_size=0):
-    params = lib.ZSTD_getCParams(level, source_size, dict_size)
-    return CompressionParameters(window_log=params.windowLog,
-                                 chain_log=params.chainLog,
-                                 hash_log=params.hashLog,
-                                 search_log=params.searchLog,
-                                 search_length=params.searchLength,
-                                 target_length=params.targetLength,
-                                 strategy=params.strategy)
-
-
-def estimate_compression_context_size(params):
-    if not isinstance(params, CompressionParameters):
-        raise ValueError('argument must be a CompressionParameters')
-
-    cparams = params.as_compression_parameters()
-    return lib.ZSTD_estimateCCtxSize(cparams)
-
+CompressionParameters = ZstdCompressionParameters
 
 def estimate_decompression_context_size():
     return lib.ZSTD_estimateDCtxSize()
 
 
+def _set_compression_parameter(params, param, value):
+    zresult = lib.ZSTD_CCtxParam_setParameter(params, param,
+                                              ffi.cast('unsigned', value))
+    if lib.ZSTD_isError(zresult):
+        raise ZstdError('unable to set compression context parameter: %s' %
+                        _zstd_error(zresult))
+
 class ZstdCompressionWriter(object):
     def __init__(self, compressor, writer, source_size, write_size):
         self._compressor = compressor
@@ -169,16 +291,18 @@
         self._source_size = source_size
         self._write_size = write_size
         self._entered = False
-        self._mtcctx = compressor._cctx if compressor._multithreaded else None
+        self._bytes_compressed = 0
 
     def __enter__(self):
         if self._entered:
             raise ZstdError('cannot __enter__ multiple times')
 
-        if self._mtcctx:
-            self._compressor._init_mtcstream(self._source_size)
-        else:
-            self._compressor._ensure_cstream(self._source_size)
+        zresult = lib.ZSTD_CCtx_setPledgedSrcSize(self._compressor._cctx,
+                                                  self._source_size)
+        if lib.ZSTD_isError(zresult):
+            raise ZstdError('error setting source size: %s' %
+                            _zstd_error(zresult))
+
         self._entered = True
         return self
 
@@ -186,20 +310,27 @@
         self._entered = False
 
         if not exc_type and not exc_value and not exc_tb:
+            dst_buffer = ffi.new('char[]', self._write_size)
+
             out_buffer = ffi.new('ZSTD_outBuffer *')
-            dst_buffer = ffi.new('char[]', self._write_size)
+            in_buffer = ffi.new('ZSTD_inBuffer *')
+
             out_buffer.dst = dst_buffer
-            out_buffer.size = self._write_size
+            out_buffer.size = len(dst_buffer)
             out_buffer.pos = 0
 
+            in_buffer.src = ffi.NULL
+            in_buffer.size = 0
+            in_buffer.pos = 0
+
             while True:
-                if self._mtcctx:
-                    zresult = lib.ZSTDMT_endStream(self._mtcctx, out_buffer)
-                else:
-                    zresult = lib.ZSTD_endStream(self._compressor._cstream, out_buffer)
+                zresult = lib.ZSTD_compress_generic(self._compressor._cctx,
+                                                    out_buffer, in_buffer,
+                                                    lib.ZSTD_e_end)
+
                 if lib.ZSTD_isError(zresult):
                     raise ZstdError('error ending compression stream: %s' %
-                                    ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                    _zstd_error(zresult))
 
                 if out_buffer.pos:
                     self._writer.write(ffi.buffer(out_buffer.dst, out_buffer.pos)[:])
@@ -217,7 +348,7 @@
             raise ZstdError('cannot determine size of an inactive compressor; '
                             'call when a context manager is active')
 
-        return lib.ZSTD_sizeof_CStream(self._compressor._cstream)
+        return lib.ZSTD_sizeof_CCtx(self._compressor._cctx)
 
     def write(self, data):
         if not self._entered:
@@ -240,19 +371,17 @@
         out_buffer.pos = 0
 
         while in_buffer.pos < in_buffer.size:
-            if self._mtcctx:
-                zresult = lib.ZSTDMT_compressStream(self._mtcctx, out_buffer,
-                                                    in_buffer)
-            else:
-                zresult = lib.ZSTD_compressStream(self._compressor._cstream, out_buffer,
-                                                  in_buffer)
+            zresult = lib.ZSTD_compress_generic(self._compressor._cctx,
+                                                out_buffer, in_buffer,
+                                                lib.ZSTD_e_continue)
             if lib.ZSTD_isError(zresult):
                 raise ZstdError('zstd compress error: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                _zstd_error(zresult))
 
             if out_buffer.pos:
                 self._writer.write(ffi.buffer(out_buffer.dst, out_buffer.pos)[:])
                 total_write += out_buffer.pos
+                self._bytes_compressed += out_buffer.pos
                 out_buffer.pos = 0
 
         return total_write
@@ -269,24 +398,32 @@
         out_buffer.size = self._write_size
         out_buffer.pos = 0
 
+        in_buffer = ffi.new('ZSTD_inBuffer *')
+        in_buffer.src = ffi.NULL
+        in_buffer.size = 0
+        in_buffer.pos = 0
+
         while True:
-            if self._mtcctx:
-                zresult = lib.ZSTDMT_flushStream(self._mtcctx, out_buffer)
-            else:
-                zresult = lib.ZSTD_flushStream(self._compressor._cstream, out_buffer)
+            zresult = lib.ZSTD_compress_generic(self._compressor._cctx,
+                                                out_buffer, in_buffer,
+                                                lib.ZSTD_e_flush)
             if lib.ZSTD_isError(zresult):
                 raise ZstdError('zstd compress error: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                _zstd_error(zresult))
 
             if not out_buffer.pos:
                 break
 
             self._writer.write(ffi.buffer(out_buffer.dst, out_buffer.pos)[:])
             total_write += out_buffer.pos
+            self._bytes_compressed += out_buffer.pos
             out_buffer.pos = 0
 
         return total_write
 
+    def tell(self):
+        return self._bytes_compressed
+
 
 class ZstdCompressionObj(object):
     def compress(self, data):
@@ -302,15 +439,13 @@
         chunks = []
 
         while source.pos < len(data):
-            if self._mtcctx:
-                zresult = lib.ZSTDMT_compressStream(self._mtcctx,
-                                                    self._out, source)
-            else:
-                zresult = lib.ZSTD_compressStream(self._compressor._cstream, self._out,
-                                                  source)
+            zresult = lib.ZSTD_compress_generic(self._compressor._cctx,
+                                                self._out,
+                                                source,
+                                                lib.ZSTD_e_continue)
             if lib.ZSTD_isError(zresult):
                 raise ZstdError('zstd compress error: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                _zstd_error(zresult))
 
             if self._out.pos:
                 chunks.append(ffi.buffer(self._out.dst, self._out.pos)[:])
@@ -327,14 +462,19 @@
 
         assert self._out.pos == 0
 
+        in_buffer = ffi.new('ZSTD_inBuffer *')
+        in_buffer.src = ffi.NULL
+        in_buffer.size = 0
+        in_buffer.pos = 0
+
         if flush_mode == COMPRESSOBJ_FLUSH_BLOCK:
-            if self._mtcctx:
-                zresult = lib.ZSTDMT_flushStream(self._mtcctx, self._out)
-            else:
-                zresult = lib.ZSTD_flushStream(self._compressor._cstream, self._out)
+            zresult = lib.ZSTD_compress_generic(self._compressor._cctx,
+                                                self._out,
+                                                in_buffer,
+                                                lib.ZSTD_e_flush)
             if lib.ZSTD_isError(zresult):
                 raise ZstdError('zstd compress error: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                _zstd_error(zresult))
 
             # Output buffer is guaranteed to hold full block.
             assert zresult == 0
@@ -352,13 +492,13 @@
         chunks = []
 
         while True:
-            if self._mtcctx:
-                zresult = lib.ZSTDMT_endStream(self._mtcctx, self._out)
-            else:
-                zresult = lib.ZSTD_endStream(self._compressor._cstream, self._out)
+            zresult = lib.ZSTD_compress_generic(self._compressor._cctx,
+                                                self._out,
+                                                in_buffer,
+                                                lib.ZSTD_e_end)
             if lib.ZSTD_isError(zresult):
                 raise ZstdError('error ending compression stream: %s' %
-                                ffi.string(lib.ZSTD_getErroName(zresult)))
+                                _zstd_error(zresult))
 
             if self._out.pos:
                 chunks.append(ffi.buffer(self._out.dst, self._out.pos)[:])
@@ -370,95 +510,335 @@
         return b''.join(chunks)
 
 
+class CompressionReader(object):
+    def __init__(self, compressor, source, size, read_size):
+        self._compressor = compressor
+        self._source = source
+        self._source_size = size
+        self._read_size = read_size
+        self._entered = False
+        self._closed = False
+        self._bytes_compressed = 0
+        self._finished_input = False
+        self._finished_output = False
+
+        self._in_buffer = ffi.new('ZSTD_inBuffer *')
+        # Holds a ref so backing bytes in self._in_buffer stay alive.
+        self._source_buffer = None
+
+    def __enter__(self):
+        if self._entered:
+            raise ValueError('cannot __enter__ multiple times')
+
+        zresult = lib.ZSTD_CCtx_setPledgedSrcSize(self._compressor._cctx,
+                                                  self._source_size)
+        if lib.ZSTD_isError(zresult):
+            raise ZstdError('error setting source size: %s' %
+                            _zstd_error(zresult))
+
+        self._entered = True
+        return self
+
+    def __exit__(self, exc_type, exc_value, exc_tb):
+        self._entered = False
+        self._closed = True
+        self._source = None
+        self._compressor = None
+
+        return False
+
+    def readable(self):
+        return True
+
+    def writable(self):
+        return False
+
+    def seekable(self):
+        return False
+
+    def readline(self):
+        raise io.UnsupportedOperation()
+
+    def readlines(self):
+        raise io.UnsupportedOperation()
+
+    def write(self, data):
+        raise OSError('stream is not writable')
+
+    def writelines(self, ignored):
+        raise OSError('stream is not writable')
+
+    def isatty(self):
+        return False
+
+    def flush(self):
+        return None
+
+    def close(self):
+        self._closed = True
+        return None
+
+    def closed(self):
+        return self._closed
+
+    def tell(self):
+        return self._bytes_compressed
+
+    def readall(self):
+        raise NotImplementedError()
+
+    def __iter__(self):
+        raise io.UnsupportedOperation()
+
+    def __next__(self):
+        raise io.UnsupportedOperation()
+
+    next = __next__
+
+    def read(self, size=-1):
+        if not self._entered:
+            raise ZstdError('read() must be called from an active context manager')
+
+        if self._closed:
+            raise ValueError('stream is closed')
+
+        if self._finished_output:
+            return b''
+
+        if size < 1:
+            raise ValueError('cannot read negative or size 0 amounts')
+
+        # Need a dedicated ref to dest buffer otherwise it gets collected.
+        dst_buffer = ffi.new('char[]', size)
+        out_buffer = ffi.new('ZSTD_outBuffer *')
+        out_buffer.dst = dst_buffer
+        out_buffer.size = size
+        out_buffer.pos = 0
+
+        def compress_input():
+            if self._in_buffer.pos >= self._in_buffer.size:
+                return
+
+            old_pos = out_buffer.pos
+
+            zresult = lib.ZSTD_compress_generic(self._compressor._cctx,
+                                                out_buffer, self._in_buffer,
+                                                lib.ZSTD_e_continue)
+
+            self._bytes_compressed += out_buffer.pos - old_pos
+
+            if self._in_buffer.pos == self._in_buffer.size:
+                self._in_buffer.src = ffi.NULL
+                self._in_buffer.pos = 0
+                self._in_buffer.size = 0
+                self._source_buffer = None
+
+                if not hasattr(self._source, 'read'):
+                    self._finished_input = True
+
+            if lib.ZSTD_isError(zresult):
+                raise ZstdError('zstd compress error: %s',
+                                _zstd_error(zresult))
+
+            if out_buffer.pos and out_buffer.pos == out_buffer.size:
+                return ffi.buffer(out_buffer.dst, out_buffer.pos)[:]
+
+        def get_input():
+            if self._finished_input:
+                return
+
+            if hasattr(self._source, 'read'):
+                data = self._source.read(self._read_size)
+
+                if not data:
+                    self._finished_input = True
+                    return
+
+                self._source_buffer = ffi.from_buffer(data)
+                self._in_buffer.src = self._source_buffer
+                self._in_buffer.size = len(self._source_buffer)
+                self._in_buffer.pos = 0
+            else:
+                self._source_buffer = ffi.from_buffer(self._source)
+                self._in_buffer.src = self._source_buffer
+                self._in_buffer.size = len(self._source_buffer)
+                self._in_buffer.pos = 0
+
+        result = compress_input()
+        if result:
+            return result
+
+        while not self._finished_input:
+            get_input()
+            result = compress_input()
+            if result:
+                return result
+
+        # EOF
+        old_pos = out_buffer.pos
+
+        zresult = lib.ZSTD_compress_generic(self._compressor._cctx,
+                                            out_buffer, self._in_buffer,
+                                            lib.ZSTD_e_end)
+
+        self._bytes_compressed += out_buffer.pos - old_pos
+
+        if lib.ZSTD_isError(zresult):
+            raise ZstdError('error ending compression stream: %s',
+                            _zstd_error(zresult))
+
+        if zresult == 0:
+            self._finished_output = True
+
+        return ffi.buffer(out_buffer.dst, out_buffer.pos)[:]
+
 class ZstdCompressor(object):
     def __init__(self, level=3, dict_data=None, compression_params=None,
-                 write_checksum=False, write_content_size=False,
-                 write_dict_id=True, threads=0):
-        if level < 1:
-            raise ValueError('level must be greater than 0')
-        elif level > lib.ZSTD_maxCLevel():
+                 write_checksum=None, write_content_size=None,
+                 write_dict_id=None, threads=0):
+        if level > lib.ZSTD_maxCLevel():
             raise ValueError('level must be less than %d' % lib.ZSTD_maxCLevel())
 
         if threads < 0:
             threads = _cpu_count()
 
-        self._compression_level = level
-        self._dict_data = dict_data
-        self._cparams = compression_params
-        self._fparams = ffi.new('ZSTD_frameParameters *')[0]
-        self._fparams.checksumFlag = write_checksum
-        self._fparams.contentSizeFlag = write_content_size
-        self._fparams.noDictIDFlag = not write_dict_id
+        if compression_params and write_checksum is not None:
+            raise ValueError('cannot define compression_params and '
+                             'write_checksum')
+
+        if compression_params and write_content_size is not None:
+            raise ValueError('cannot define compression_params and '
+                             'write_content_size')
+
+        if compression_params and write_dict_id is not None:
+            raise ValueError('cannot define compression_params and '
+                             'write_dict_id')
 
-        if threads:
-            cctx = lib.ZSTDMT_createCCtx(threads)
-            if cctx == ffi.NULL:
-                raise MemoryError()
+        if compression_params and threads:
+            raise ValueError('cannot define compression_params and threads')
 
-            self._cctx = ffi.gc(cctx, lib.ZSTDMT_freeCCtx)
-            self._multithreaded = True
+        if compression_params:
+            self._params = _make_cctx_params(compression_params)
         else:
-            cctx = lib.ZSTD_createCCtx()
-            if cctx == ffi.NULL:
+            if write_dict_id is None:
+                write_dict_id = True
+
+            params = lib.ZSTD_createCCtxParams()
+            if params == ffi.NULL:
                 raise MemoryError()
 
-            self._cctx = ffi.gc(cctx, lib.ZSTD_freeCCtx)
-            self._multithreaded = False
+            self._params = ffi.gc(params, lib.ZSTD_freeCCtxParams)
+
+            _set_compression_parameter(self._params,
+                                       lib.ZSTD_p_compressionLevel,
+                                       level)
 
-        self._cstream = None
+            _set_compression_parameter(
+                self._params,
+                lib.ZSTD_p_contentSizeFlag,
+                write_content_size if write_content_size is not None else 1)
+
+            _set_compression_parameter(self._params,
+                                       lib.ZSTD_p_checksumFlag,
+                                       1 if write_checksum else 0)
 
-    def compress(self, data, allow_empty=False):
-        if len(data) == 0 and self._fparams.contentSizeFlag and not allow_empty:
-            raise ValueError('cannot write empty inputs when writing content sizes')
+            _set_compression_parameter(self._params,
+                                       lib.ZSTD_p_dictIDFlag,
+                                       1 if write_dict_id else 0)
 
-        if self._multithreaded and self._dict_data:
-            raise ZstdError('compress() cannot be used with both dictionaries and multi-threaded compression')
+            if threads:
+                _set_compression_parameter(self._params,
+                                           lib.ZSTD_p_nbWorkers,
+                                           threads)
 
-        if self._multithreaded and self._cparams:
-            raise ZstdError('compress() cannot be used with both compression parameters and multi-threaded compression')
+        cctx = lib.ZSTD_createCCtx()
+        if cctx == ffi.NULL:
+            raise MemoryError()
+
+        self._cctx = cctx
+        self._dict_data = dict_data
 
-        # TODO use a CDict for performance.
-        dict_data = ffi.NULL
-        dict_size = 0
+        # We defer setting up garbage collection until after calling
+        # _ensure_cctx() to ensure the memory size estimate is more accurate.
+        try:
+            self._ensure_cctx()
+        finally:
+            self._cctx = ffi.gc(cctx, lib.ZSTD_freeCCtx,
+                                size=lib.ZSTD_sizeof_CCtx(cctx))
 
-        if self._dict_data:
-            dict_data = self._dict_data.as_bytes()
-            dict_size = len(self._dict_data)
+    def _ensure_cctx(self):
+        lib.ZSTD_CCtx_reset(self._cctx)
+
+        zresult = lib.ZSTD_CCtx_setParametersUsingCCtxParams(self._cctx,
+                                                             self._params)
+        if lib.ZSTD_isError(zresult):
+            raise ZstdError('could not set compression parameters: %s' %
+                            _zstd_error(zresult))
+
+        dict_data = self._dict_data
 
-        params = ffi.new('ZSTD_parameters *')[0]
-        if self._cparams:
-            params.cParams = self._cparams.as_compression_parameters()
-        else:
-            params.cParams = lib.ZSTD_getCParams(self._compression_level, len(data),
-                                                 dict_size)
-        params.fParams = self._fparams
+        if dict_data:
+            if dict_data._cdict:
+                zresult = lib.ZSTD_CCtx_refCDict(self._cctx, dict_data._cdict)
+            else:
+                zresult = lib.ZSTD_CCtx_loadDictionary_advanced(
+                    self._cctx, dict_data.as_bytes(), len(dict_data),
+                    lib.ZSTD_dlm_byRef, dict_data._dict_type)
 
-        dest_size = lib.ZSTD_compressBound(len(data))
+            if lib.ZSTD_isError(zresult):
+                raise ZstdError('could not load compression dictionary: %s' %
+                                _zstd_error(zresult))
+
+    def memory_size(self):
+        return lib.ZSTD_sizeof_CCtx(self._cctx)
+
+    def compress(self, data):
+        self._ensure_cctx()
+
+        data_buffer = ffi.from_buffer(data)
+
+        dest_size = lib.ZSTD_compressBound(len(data_buffer))
         out = new_nonzero('char[]', dest_size)
 
-        if self._multithreaded:
-            zresult = lib.ZSTDMT_compressCCtx(self._cctx,
-                                              ffi.addressof(out), dest_size,
-                                              data, len(data),
-                                              self._compression_level)
-        else:
-            zresult = lib.ZSTD_compress_advanced(self._cctx,
-                                                 ffi.addressof(out), dest_size,
-                                                 data, len(data),
-                                                 dict_data, dict_size,
-                                                 params)
+        zresult = lib.ZSTD_CCtx_setPledgedSrcSize(self._cctx, len(data_buffer))
+        if lib.ZSTD_isError(zresult):
+            raise ZstdError('error setting source size: %s' %
+                            _zstd_error(zresult))
+
+        out_buffer = ffi.new('ZSTD_outBuffer *')
+        in_buffer = ffi.new('ZSTD_inBuffer *')
+
+        out_buffer.dst = out
+        out_buffer.size = dest_size
+        out_buffer.pos = 0
+
+        in_buffer.src = data_buffer
+        in_buffer.size = len(data_buffer)
+        in_buffer.pos = 0
+
+        zresult = lib.ZSTD_compress_generic(self._cctx,
+                                            out_buffer,
+                                            in_buffer,
+                                            lib.ZSTD_e_end)
 
         if lib.ZSTD_isError(zresult):
             raise ZstdError('cannot compress: %s' %
-                            ffi.string(lib.ZSTD_getErrorName(zresult)))
+                            _zstd_error(zresult))
+        elif zresult:
+            raise ZstdError('unexpected partial frame flush')
 
-        return ffi.buffer(out, zresult)[:]
+        return ffi.buffer(out, out_buffer.pos)[:]
 
-    def compressobj(self, size=0):
-        if self._multithreaded:
-            self._init_mtcstream(size)
-        else:
-            self._ensure_cstream(size)
+    def compressobj(self, size=-1):
+        self._ensure_cctx()
+
+        if size < 0:
+            size = lib.ZSTD_CONTENTSIZE_UNKNOWN
+
+        zresult = lib.ZSTD_CCtx_setPledgedSrcSize(self._cctx, size)
+        if lib.ZSTD_isError(zresult):
+            raise ZstdError('error setting source size: %s' %
+                            _zstd_error(zresult))
 
         cobj = ZstdCompressionObj()
         cobj._out = ffi.new('ZSTD_outBuffer *')
@@ -469,14 +849,9 @@
         cobj._compressor = self
         cobj._finished = False
 
-        if self._multithreaded:
-            cobj._mtcctx = self._cctx
-        else:
-            cobj._mtcctx = None
-
         return cobj
 
-    def copy_stream(self, ifh, ofh, size=0,
+    def copy_stream(self, ifh, ofh, size=-1,
                     read_size=COMPRESSION_RECOMMENDED_INPUT_SIZE,
                     write_size=COMPRESSION_RECOMMENDED_OUTPUT_SIZE):
 
@@ -485,11 +860,15 @@
         if not hasattr(ofh, 'write'):
             raise ValueError('second argument must have a write() method')
 
-        mt = self._multithreaded
-        if mt:
-            self._init_mtcstream(size)
-        else:
-            self._ensure_cstream(size)
+        self._ensure_cctx()
+
+        if size < 0:
+            size = lib.ZSTD_CONTENTSIZE_UNKNOWN
+
+        zresult = lib.ZSTD_CCtx_setPledgedSrcSize(self._cctx, size)
+        if lib.ZSTD_isError(zresult):
+            raise ZstdError('error setting source size: %s' %
+                            _zstd_error(zresult))
 
         in_buffer = ffi.new('ZSTD_inBuffer *')
         out_buffer = ffi.new('ZSTD_outBuffer *')
@@ -513,14 +892,13 @@
             in_buffer.pos = 0
 
             while in_buffer.pos < in_buffer.size:
-                if mt:
-                    zresult = lib.ZSTDMT_compressStream(self._cctx, out_buffer, in_buffer)
-                else:
-                    zresult = lib.ZSTD_compressStream(self._cstream,
-                                                      out_buffer, in_buffer)
+                zresult = lib.ZSTD_compress_generic(self._cctx,
+                                                    out_buffer,
+                                                    in_buffer,
+                                                    lib.ZSTD_e_continue)
                 if lib.ZSTD_isError(zresult):
                     raise ZstdError('zstd compress error: %s' %
-                                    ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                    _zstd_error(zresult))
 
                 if out_buffer.pos:
                     ofh.write(ffi.buffer(out_buffer.dst, out_buffer.pos))
@@ -529,13 +907,13 @@
 
         # We've finished reading. Flush the compressor.
         while True:
-            if mt:
-                zresult = lib.ZSTDMT_endStream(self._cctx, out_buffer)
-            else:
-                zresult = lib.ZSTD_endStream(self._cstream, out_buffer)
+            zresult = lib.ZSTD_compress_generic(self._cctx,
+                                                out_buffer,
+                                                in_buffer,
+                                                lib.ZSTD_e_end)
             if lib.ZSTD_isError(zresult):
                 raise ZstdError('error ending compression stream: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                _zstd_error(zresult))
 
             if out_buffer.pos:
                 ofh.write(ffi.buffer(out_buffer.dst, out_buffer.pos))
@@ -547,17 +925,38 @@
 
         return total_read, total_write
 
-    def write_to(self, writer, size=0,
+    def stream_reader(self, source, size=-1,
+                      read_size=COMPRESSION_RECOMMENDED_INPUT_SIZE):
+        self._ensure_cctx()
+
+        try:
+            size = len(source)
+        except Exception:
+            pass
+
+        if size < 0:
+            size = lib.ZSTD_CONTENTSIZE_UNKNOWN
+
+        return CompressionReader(self, source, size, read_size)
+
+    def stream_writer(self, writer, size=-1,
                  write_size=COMPRESSION_RECOMMENDED_OUTPUT_SIZE):
 
         if not hasattr(writer, 'write'):
             raise ValueError('must pass an object with a write() method')
 
+        self._ensure_cctx()
+
+        if size < 0:
+            size = lib.ZSTD_CONTENTSIZE_UNKNOWN
+
         return ZstdCompressionWriter(self, writer, size, write_size)
 
-    def read_from(self, reader, size=0,
-                  read_size=COMPRESSION_RECOMMENDED_INPUT_SIZE,
-                  write_size=COMPRESSION_RECOMMENDED_OUTPUT_SIZE):
+    write_to = stream_writer
+
+    def read_to_iter(self, reader, size=-1,
+                     read_size=COMPRESSION_RECOMMENDED_INPUT_SIZE,
+                     write_size=COMPRESSION_RECOMMENDED_OUTPUT_SIZE):
         if hasattr(reader, 'read'):
             have_read = True
         elif hasattr(reader, '__getitem__'):
@@ -568,10 +967,15 @@
             raise ValueError('must pass an object with a read() method or '
                              'conforms to buffer protocol')
 
-        if self._multithreaded:
-            self._init_mtcstream(size)
-        else:
-            self._ensure_cstream(size)
+        self._ensure_cctx()
+
+        if size < 0:
+            size = lib.ZSTD_CONTENTSIZE_UNKNOWN
+
+        zresult = lib.ZSTD_CCtx_setPledgedSrcSize(self._cctx, size)
+        if lib.ZSTD_isError(zresult):
+            raise ZstdError('error setting source size: %s' %
+                            _zstd_error(zresult))
 
         in_buffer = ffi.new('ZSTD_inBuffer *')
         out_buffer = ffi.new('ZSTD_outBuffer *')
@@ -611,13 +1015,11 @@
             in_buffer.pos = 0
 
             while in_buffer.pos < in_buffer.size:
-                if self._multithreaded:
-                    zresult = lib.ZSTDMT_compressStream(self._cctx, out_buffer, in_buffer)
-                else:
-                    zresult = lib.ZSTD_compressStream(self._cstream, out_buffer, in_buffer)
+                zresult = lib.ZSTD_compress_generic(self._cctx, out_buffer, in_buffer,
+                                                    lib.ZSTD_e_continue)
                 if lib.ZSTD_isError(zresult):
                     raise ZstdError('zstd compress error: %s' %
-                                    ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                    _zstd_error(zresult))
 
                 if out_buffer.pos:
                     data = ffi.buffer(out_buffer.dst, out_buffer.pos)[:]
@@ -633,13 +1035,13 @@
         # remains.
         while True:
             assert out_buffer.pos == 0
-            if self._multithreaded:
-                zresult = lib.ZSTDMT_endStream(self._cctx, out_buffer)
-            else:
-                zresult = lib.ZSTD_endStream(self._cstream, out_buffer)
+            zresult = lib.ZSTD_compress_generic(self._cctx,
+                                                out_buffer,
+                                                in_buffer,
+                                                lib.ZSTD_e_end)
             if lib.ZSTD_isError(zresult):
                 raise ZstdError('error ending compression stream: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                _zstd_error(zresult))
 
             if out_buffer.pos:
                 data = ffi.buffer(out_buffer.dst, out_buffer.pos)[:]
@@ -649,67 +1051,12 @@
             if zresult == 0:
                 break
 
-    def _ensure_cstream(self, size):
-        if self._cstream:
-            zresult = lib.ZSTD_resetCStream(self._cstream, size)
-            if lib.ZSTD_isError(zresult):
-                raise ZstdError('could not reset CStream: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
-
-            return
-
-        cstream = lib.ZSTD_createCStream()
-        if cstream == ffi.NULL:
-            raise MemoryError()
-
-        cstream = ffi.gc(cstream, lib.ZSTD_freeCStream)
-
-        dict_data = ffi.NULL
-        dict_size = 0
-        if self._dict_data:
-            dict_data = self._dict_data.as_bytes()
-            dict_size = len(self._dict_data)
-
-        zparams = ffi.new('ZSTD_parameters *')[0]
-        if self._cparams:
-            zparams.cParams = self._cparams.as_compression_parameters()
-        else:
-            zparams.cParams = lib.ZSTD_getCParams(self._compression_level,
-                                                  size, dict_size)
-        zparams.fParams = self._fparams
+    read_from = read_to_iter
 
-        zresult = lib.ZSTD_initCStream_advanced(cstream, dict_data, dict_size,
-                                                zparams, size)
-        if lib.ZSTD_isError(zresult):
-            raise Exception('cannot init CStream: %s' %
-                            ffi.string(lib.ZSTD_getErrorName(zresult)))
-
-        self._cstream = cstream
-
-    def _init_mtcstream(self, size):
-        assert self._multithreaded
+    def frame_progression(self):
+        progression = lib.ZSTD_getFrameProgression(self._cctx)
 
-        dict_data = ffi.NULL
-        dict_size = 0
-        if self._dict_data:
-            dict_data = self._dict_data.as_bytes()
-            dict_size = len(self._dict_data)
-
-        zparams = ffi.new('ZSTD_parameters *')[0]
-        if self._cparams:
-            zparams.cParams = self._cparams.as_compression_parameters()
-        else:
-            zparams.cParams = lib.ZSTD_getCParams(self._compression_level,
-                                                  size, dict_size)
-
-        zparams.fParams = self._fparams
-
-        zresult = lib.ZSTDMT_initCStream_advanced(self._cctx, dict_data, dict_size,
-                                                  zparams, size)
-
-        if lib.ZSTD_isError(zresult):
-            raise ZstdError('cannot init CStream: %s' %
-                            ffi.string(lib.ZSTD_getErrorName(zresult)))
+        return progression.ingested, progression.consumed, progression.produced
 
 
 class FrameParameters(object):
@@ -720,16 +1067,38 @@
         self.has_checksum = bool(fparams.checksumFlag)
 
 
-def get_frame_parameters(data):
-    if not isinstance(data, bytes_type):
-        raise TypeError('argument must be bytes')
+def frame_content_size(data):
+    data_buffer = ffi.from_buffer(data)
+
+    size = lib.ZSTD_getFrameContentSize(data_buffer, len(data_buffer))
+
+    if size == lib.ZSTD_CONTENTSIZE_ERROR:
+        raise ZstdError('error when determining content size')
+    elif size == lib.ZSTD_CONTENTSIZE_UNKNOWN:
+        return -1
+    else:
+        return size
+
 
-    params = ffi.new('ZSTD_frameParams *')
+def frame_header_size(data):
+    data_buffer = ffi.from_buffer(data)
+
+    zresult = lib.ZSTD_frameHeaderSize(data_buffer, len(data_buffer))
+    if lib.ZSTD_isError(zresult):
+        raise ZstdError('could not determine frame header size: %s' %
+                        _zstd_error(zresult))
 
-    zresult = lib.ZSTD_getFrameParams(params, data, len(data))
+    return zresult
+
+
+def get_frame_parameters(data):
+    params = ffi.new('ZSTD_frameHeader *')
+
+    data_buffer = ffi.from_buffer(data)
+    zresult = lib.ZSTD_getFrameHeader(params, data_buffer, len(data_buffer))
     if lib.ZSTD_isError(zresult):
         raise ZstdError('cannot get frame parameters: %s' %
-                        ffi.string(lib.ZSTD_getErrorName(zresult)))
+                        _zstd_error(zresult))
 
     if zresult:
         raise ZstdError('not enough data for frame parameters; need %d bytes' %
@@ -739,12 +1108,20 @@
 
 
 class ZstdCompressionDict(object):
-    def __init__(self, data, k=0, d=0):
+    def __init__(self, data, dict_type=DICT_TYPE_AUTO, k=0, d=0):
         assert isinstance(data, bytes_type)
         self._data = data
         self.k = k
         self.d = d
 
+        if dict_type not in (DICT_TYPE_AUTO, DICT_TYPE_RAWCONTENT,
+                             DICT_TYPE_FULLDICT):
+            raise ValueError('invalid dictionary load mode: %d; must use '
+                             'DICT_TYPE_* constants')
+
+        self._dict_type = dict_type
+        self._cdict = None
+
     def __len__(self):
         return len(self._data)
 
@@ -754,51 +1131,55 @@
     def as_bytes(self):
         return self._data
 
+    def precompute_compress(self, level=0, compression_params=None):
+        if level and compression_params:
+            raise ValueError('must only specify one of level or '
+                             'compression_params')
 
-def train_dictionary(dict_size, samples, selectivity=0, level=0,
-                     notifications=0, dict_id=0):
-    if not isinstance(samples, list):
-        raise TypeError('samples must be a list')
-
-    total_size = sum(map(len, samples))
-
-    samples_buffer = new_nonzero('char[]', total_size)
-    sample_sizes = new_nonzero('size_t[]', len(samples))
+        if not level and not compression_params:
+            raise ValueError('must specify one of level or compression_params')
 
-    offset = 0
-    for i, sample in enumerate(samples):
-        if not isinstance(sample, bytes_type):
-            raise ValueError('samples must be bytes')
-
-        l = len(sample)
-        ffi.memmove(samples_buffer + offset, sample, l)
-        offset += l
-        sample_sizes[i] = l
-
-    dict_data = new_nonzero('char[]', dict_size)
+        if level:
+            cparams = lib.ZSTD_getCParams(level, 0, len(self._data))
+        else:
+            cparams = ffi.new('ZSTD_compressionParameters')
+            cparams.chainLog = compression_params.chain_log
+            cparams.hashLog = compression_params.hash_log
+            cparams.searchLength = compression_params.min_match
+            cparams.searchLog = compression_params.search_log
+            cparams.strategy = compression_params.compression_strategy
+            cparams.targetLength = compression_params.target_length
+            cparams.windowLog = compression_params.window_log
 
-    dparams = ffi.new('ZDICT_params_t *')[0]
-    dparams.selectivityLevel = selectivity
-    dparams.compressionLevel = level
-    dparams.notificationLevel = notifications
-    dparams.dictID = dict_id
+        cdict = lib.ZSTD_createCDict_advanced(self._data, len(self._data),
+                                              lib.ZSTD_dlm_byRef,
+                                              self._dict_type,
+                                              cparams,
+                                              lib.ZSTD_defaultCMem)
+        if cdict == ffi.NULL:
+            raise ZstdError('unable to precompute dictionary')
+
+        self._cdict = ffi.gc(cdict, lib.ZSTD_freeCDict,
+                             size=lib.ZSTD_sizeof_CDict(cdict))
 
-    zresult = lib.ZDICT_trainFromBuffer_advanced(
-        ffi.addressof(dict_data), dict_size,
-        ffi.addressof(samples_buffer),
-        ffi.addressof(sample_sizes, 0), len(samples),
-        dparams)
+    @property
+    def _ddict(self):
+        ddict = lib.ZSTD_createDDict_advanced(self._data, len(self._data),
+                                              lib.ZSTD_dlm_byRef,
+                                              self._dict_type,
+                                              lib.ZSTD_defaultCMem)
 
-    if lib.ZDICT_isError(zresult):
-        raise ZstdError('Cannot train dict: %s' %
-                        ffi.string(lib.ZDICT_getErrorName(zresult)))
+        if ddict == ffi.NULL:
+            raise ZstdError('could not create decompression dict')
 
-    return ZstdCompressionDict(ffi.buffer(dict_data, zresult)[:])
-
+        ddict = ffi.gc(ddict, lib.ZSTD_freeDDict,
+                       size=lib.ZSTD_sizeof_DDict(ddict))
+        self.__dict__['_ddict'] = ddict
 
-def train_cover_dictionary(dict_size, samples, k=0, d=0,
-                           notifications=0, dict_id=0, level=0, optimize=False,
-                           steps=0, threads=0):
+        return ddict
+
+def train_dictionary(dict_size, samples, k=0, d=0, notifications=0, dict_id=0,
+                     level=0, steps=0, threads=0):
     if not isinstance(samples, list):
         raise TypeError('samples must be a list')
 
@@ -822,47 +1203,55 @@
 
     dict_data = new_nonzero('char[]', dict_size)
 
-    dparams = ffi.new('COVER_params_t *')[0]
+    dparams = ffi.new('ZDICT_cover_params_t *')[0]
     dparams.k = k
     dparams.d = d
     dparams.steps = steps
     dparams.nbThreads = threads
-    dparams.notificationLevel = notifications
-    dparams.dictID = dict_id
-    dparams.compressionLevel = level
+    dparams.zParams.notificationLevel = notifications
+    dparams.zParams.dictID = dict_id
+    dparams.zParams.compressionLevel = level
 
-    if optimize:
-        zresult = lib.COVER_optimizeTrainFromBuffer(
+    if (not dparams.k and not dparams.d and not dparams.steps
+        and not dparams.nbThreads and not dparams.zParams.notificationLevel
+        and not dparams.zParams.dictID
+        and not dparams.zParams.compressionLevel):
+        zresult = lib.ZDICT_trainFromBuffer(
+            ffi.addressof(dict_data), dict_size,
+            ffi.addressof(samples_buffer),
+            ffi.addressof(sample_sizes, 0), len(samples))
+    elif dparams.steps or dparams.nbThreads:
+        zresult = lib.ZDICT_optimizeTrainFromBuffer_cover(
             ffi.addressof(dict_data), dict_size,
             ffi.addressof(samples_buffer),
             ffi.addressof(sample_sizes, 0), len(samples),
             ffi.addressof(dparams))
     else:
-        zresult = lib.COVER_trainFromBuffer(
+        zresult = lib.ZDICT_trainFromBuffer_cover(
             ffi.addressof(dict_data), dict_size,
             ffi.addressof(samples_buffer),
             ffi.addressof(sample_sizes, 0), len(samples),
             dparams)
 
     if lib.ZDICT_isError(zresult):
-        raise ZstdError('cannot train dict: %s' %
-                        ffi.string(lib.ZDICT_getErrorName(zresult)))
+        msg = ffi.string(lib.ZDICT_getErrorName(zresult)).decode('utf-8')
+        raise ZstdError('cannot train dict: %s' % msg)
 
     return ZstdCompressionDict(ffi.buffer(dict_data, zresult)[:],
+                               dict_type=DICT_TYPE_FULLDICT,
                                k=dparams.k, d=dparams.d)
 
 
 class ZstdDecompressionObj(object):
-    def __init__(self, decompressor):
+    def __init__(self, decompressor, write_size):
         self._decompressor = decompressor
+        self._write_size = write_size
         self._finished = False
 
     def decompress(self, data):
         if self._finished:
             raise ZstdError('cannot use a decompressobj multiple times')
 
-        assert(self._decompressor._dstream)
-
         in_buffer = ffi.new('ZSTD_inBuffer *')
         out_buffer = ffi.new('ZSTD_outBuffer *')
 
@@ -871,7 +1260,7 @@
         in_buffer.size = len(data_buffer)
         in_buffer.pos = 0
 
-        dst_buffer = ffi.new('char[]', DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE)
+        dst_buffer = ffi.new('char[]', self._write_size)
         out_buffer.dst = dst_buffer
         out_buffer.size = len(dst_buffer)
         out_buffer.pos = 0
@@ -879,11 +1268,11 @@
         chunks = []
 
         while in_buffer.pos < in_buffer.size:
-            zresult = lib.ZSTD_decompressStream(self._decompressor._dstream,
-                                                out_buffer, in_buffer)
+            zresult = lib.ZSTD_decompress_generic(self._decompressor._dctx,
+                                                  out_buffer, in_buffer)
             if lib.ZSTD_isError(zresult):
                 raise ZstdError('zstd decompressor error: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                _zstd_error(zresult))
 
             if zresult == 0:
                 self._finished = True
@@ -896,6 +1285,203 @@
         return b''.join(chunks)
 
 
+class DecompressionReader(object):
+    def __init__(self, decompressor, source, read_size):
+        self._decompressor = decompressor
+        self._source = source
+        self._read_size = read_size
+        self._entered = False
+        self._closed = False
+        self._bytes_decompressed = 0
+        self._finished_input = False
+        self._finished_output = False
+        self._in_buffer = ffi.new('ZSTD_inBuffer *')
+        # Holds a ref to self._in_buffer.src.
+        self._source_buffer = None
+
+    def __enter__(self):
+        if self._entered:
+            raise ValueError('cannot __enter__ multiple times')
+
+        self._decompressor._ensure_dctx()
+
+        self._entered = True
+        return self
+
+    def __exit__(self, exc_type, exc_value, exc_tb):
+        self._entered = False
+        self._closed = True
+        self._source = None
+        self._decompressor = None
+
+        return False
+
+    def readable(self):
+        return True
+
+    def writable(self):
+        return False
+
+    def seekable(self):
+        return True
+
+    def readline(self):
+        raise NotImplementedError()
+
+    def readlines(self):
+        raise NotImplementedError()
+
+    def write(self, data):
+        raise io.UnsupportedOperation()
+
+    def writelines(self, lines):
+        raise io.UnsupportedOperation()
+
+    def isatty(self):
+        return False
+
+    def flush(self):
+        return None
+
+    def close(self):
+        self._closed = True
+        return None
+
+    def closed(self):
+        return self._closed
+
+    def tell(self):
+        return self._bytes_decompressed
+
+    def readall(self):
+        raise NotImplementedError()
+
+    def __iter__(self):
+        raise NotImplementedError()
+
+    def __next__(self):
+        raise NotImplementedError()
+
+    next = __next__
+
+    def read(self, size=-1):
+        if not self._entered:
+            raise ZstdError('read() must be called from an active context manager')
+
+        if self._closed:
+            raise ValueError('stream is closed')
+
+        if self._finished_output:
+            return b''
+
+        if size < 1:
+            raise ValueError('cannot read negative or size 0 amounts')
+
+        dst_buffer = ffi.new('char[]', size)
+        out_buffer = ffi.new('ZSTD_outBuffer *')
+        out_buffer.dst = dst_buffer
+        out_buffer.size = size
+        out_buffer.pos = 0
+
+        def decompress():
+            zresult = lib.ZSTD_decompress_generic(self._decompressor._dctx,
+                                                  out_buffer, self._in_buffer)
+
+            if self._in_buffer.pos == self._in_buffer.size:
+                self._in_buffer.src = ffi.NULL
+                self._in_buffer.pos = 0
+                self._in_buffer.size = 0
+                self._source_buffer = None
+
+                if not hasattr(self._source, 'read'):
+                    self._finished_input = True
+
+            if lib.ZSTD_isError(zresult):
+                raise ZstdError('zstd decompress error: %s',
+                                _zstd_error(zresult))
+            elif zresult == 0:
+                self._finished_output = True
+
+            if out_buffer.pos and out_buffer.pos == out_buffer.size:
+                self._bytes_decompressed += out_buffer.size
+                return ffi.buffer(out_buffer.dst, out_buffer.pos)[:]
+
+        def get_input():
+            if self._finished_input:
+                return
+
+            if hasattr(self._source, 'read'):
+                data = self._source.read(self._read_size)
+
+                if not data:
+                    self._finished_input = True
+                    return
+
+                self._source_buffer = ffi.from_buffer(data)
+                self._in_buffer.src = self._source_buffer
+                self._in_buffer.size = len(self._source_buffer)
+                self._in_buffer.pos = 0
+            else:
+                self._source_buffer = ffi.from_buffer(self._source)
+                self._in_buffer.src = self._source_buffer
+                self._in_buffer.size = len(self._source_buffer)
+                self._in_buffer.pos = 0
+
+        get_input()
+        result = decompress()
+        if result:
+            return result
+
+        while not self._finished_input:
+            get_input()
+            result = decompress()
+            if result:
+                return result
+
+        self._bytes_decompressed += out_buffer.pos
+        return ffi.buffer(out_buffer.dst, out_buffer.pos)[:]
+
+    def seek(self, pos, whence=os.SEEK_SET):
+        if not self._entered:
+            raise ZstdError('seek() must be called from an active context '
+                            'manager')
+
+        if self._closed:
+            raise ValueError('stream is closed')
+
+        read_amount = 0
+
+        if whence == os.SEEK_SET:
+            if pos < 0:
+                raise ValueError('cannot seek to negative position with SEEK_SET')
+
+            if pos < self._bytes_decompressed:
+                raise ValueError('cannot seek zstd decompression stream '
+                                 'backwards')
+
+            read_amount = pos - self._bytes_decompressed
+
+        elif whence == os.SEEK_CUR:
+            if pos < 0:
+                raise ValueError('cannot seek zstd decompression stream '
+                                 'backwards')
+
+            read_amount = pos
+        elif whence == os.SEEK_END:
+            raise ValueError('zstd decompression streams cannot be seeked '
+                             'with SEEK_END')
+
+        while read_amount:
+            result = self.read(min(read_amount,
+                                   DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE))
+
+            if not result:
+                break
+
+            read_amount -= len(result)
+
+        return self._bytes_decompressed
+
 class ZstdDecompressionWriter(object):
     def __init__(self, decompressor, writer, write_size):
         self._decompressor = decompressor
@@ -907,7 +1493,7 @@
         if self._entered:
             raise ZstdError('cannot __enter__ multiple times')
 
-        self._decompressor._ensure_dstream()
+        self._decompressor._ensure_dctx()
         self._entered = True
 
         return self
@@ -916,11 +1502,11 @@
         self._entered = False
 
     def memory_size(self):
-        if not self._decompressor._dstream:
+        if not self._decompressor._dctx:
             raise ZstdError('cannot determine size of inactive decompressor '
                             'call when context manager is active')
 
-        return lib.ZSTD_sizeof_DStream(self._decompressor._dstream)
+        return lib.ZSTD_sizeof_DCtx(self._decompressor._dctx)
 
     def write(self, data):
         if not self._entered:
@@ -941,13 +1527,13 @@
         out_buffer.size = len(dst_buffer)
         out_buffer.pos = 0
 
-        dstream = self._decompressor._dstream
+        dctx = self._decompressor._dctx
 
         while in_buffer.pos < in_buffer.size:
-            zresult = lib.ZSTD_decompressStream(dstream, out_buffer, in_buffer)
+            zresult = lib.ZSTD_decompress_generic(dctx, out_buffer, in_buffer)
             if lib.ZSTD_isError(zresult):
                 raise ZstdError('zstd decompress error: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                _zstd_error(zresult))
 
             if out_buffer.pos:
                 self._writer.write(ffi.buffer(out_buffer.dst, out_buffer.pos)[:])
@@ -958,77 +1544,86 @@
 
 
 class ZstdDecompressor(object):
-    def __init__(self, dict_data=None):
+    def __init__(self, dict_data=None, max_window_size=0, format=FORMAT_ZSTD1):
         self._dict_data = dict_data
+        self._max_window_size = max_window_size
+        self._format = format
 
         dctx = lib.ZSTD_createDCtx()
         if dctx == ffi.NULL:
             raise MemoryError()
 
-        self._refdctx = ffi.gc(dctx, lib.ZSTD_freeDCtx)
-        self._dstream = None
-
-    @property
-    def _ddict(self):
-        if self._dict_data:
-            dict_data = self._dict_data.as_bytes()
-            dict_size = len(self._dict_data)
+        self._dctx = dctx
 
-            ddict = lib.ZSTD_createDDict(dict_data, dict_size)
-            if ddict == ffi.NULL:
-                raise ZstdError('could not create decompression dict')
-        else:
-            ddict = None
+        # Defer setting up garbage collection until full state is loaded so
+        # the memory size is more accurate.
+        try:
+            self._ensure_dctx()
+        finally:
+            self._dctx = ffi.gc(dctx, lib.ZSTD_freeDCtx,
+                                size=lib.ZSTD_sizeof_DCtx(dctx))
 
-        self.__dict__['_ddict'] = ddict
-        return ddict
+    def memory_size(self):
+        return lib.ZSTD_sizeof_DCtx(self._dctx)
 
     def decompress(self, data, max_output_size=0):
+        self._ensure_dctx()
+
         data_buffer = ffi.from_buffer(data)
 
-        orig_dctx = new_nonzero('char[]', lib.ZSTD_sizeof_DCtx(self._refdctx))
-        dctx = ffi.cast('ZSTD_DCtx *', orig_dctx)
-        lib.ZSTD_copyDCtx(dctx, self._refdctx)
-
-        ddict = self._ddict
+        output_size = lib.ZSTD_getFrameContentSize(data_buffer, len(data_buffer))
 
-        output_size = lib.ZSTD_getDecompressedSize(data_buffer, len(data_buffer))
-        if output_size:
-            result_buffer = ffi.new('char[]', output_size)
-            result_size = output_size
-        else:
+        if output_size == lib.ZSTD_CONTENTSIZE_ERROR:
+            raise ZstdError('error determining content size from frame header')
+        elif output_size == 0:
+            return b''
+        elif output_size == lib.ZSTD_CONTENTSIZE_UNKNOWN:
             if not max_output_size:
-                raise ZstdError('input data invalid or missing content size '
-                                'in frame header')
+                raise ZstdError('could not determine content size in frame header')
 
             result_buffer = ffi.new('char[]', max_output_size)
             result_size = max_output_size
+            output_size = 0
+        else:
+            result_buffer = ffi.new('char[]', output_size)
+            result_size = output_size
 
-        if ddict:
-            zresult = lib.ZSTD_decompress_usingDDict(dctx,
-                                                     result_buffer, result_size,
-                                                     data_buffer, len(data_buffer),
-                                                     ddict)
-        else:
-            zresult = lib.ZSTD_decompressDCtx(dctx,
-                                              result_buffer, result_size,
-                                              data_buffer, len(data_buffer))
+        out_buffer = ffi.new('ZSTD_outBuffer *')
+        out_buffer.dst = result_buffer
+        out_buffer.size = result_size
+        out_buffer.pos = 0
+
+        in_buffer = ffi.new('ZSTD_inBuffer *')
+        in_buffer.src = data_buffer
+        in_buffer.size = len(data_buffer)
+        in_buffer.pos = 0
+
+        zresult = lib.ZSTD_decompress_generic(self._dctx, out_buffer, in_buffer)
         if lib.ZSTD_isError(zresult):
             raise ZstdError('decompression error: %s' %
-                            ffi.string(lib.ZSTD_getErrorName(zresult)))
-        elif output_size and zresult != output_size:
+                            _zstd_error(zresult))
+        elif zresult:
+            raise ZstdError('decompression error: did not decompress full frame')
+        elif output_size and out_buffer.pos != output_size:
             raise ZstdError('decompression error: decompressed %d bytes; expected %d' %
                             (zresult, output_size))
 
-        return ffi.buffer(result_buffer, zresult)[:]
+        return ffi.buffer(result_buffer, out_buffer.pos)[:]
+
+    def stream_reader(self, source, read_size=DECOMPRESSION_RECOMMENDED_INPUT_SIZE):
+        self._ensure_dctx()
+        return DecompressionReader(self, source, read_size)
 
-    def decompressobj(self):
-        self._ensure_dstream()
-        return ZstdDecompressionObj(self)
+    def decompressobj(self, write_size=DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE):
+        if write_size < 1:
+            raise ValueError('write_size must be positive')
 
-    def read_from(self, reader, read_size=DECOMPRESSION_RECOMMENDED_INPUT_SIZE,
-                  write_size=DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE,
-                  skip_bytes=0):
+        self._ensure_dctx()
+        return ZstdDecompressionObj(self, write_size=write_size)
+
+    def read_to_iter(self, reader, read_size=DECOMPRESSION_RECOMMENDED_INPUT_SIZE,
+                     write_size=DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE,
+                     skip_bytes=0):
         if skip_bytes >= read_size:
             raise ValueError('skip_bytes must be smaller than read_size')
 
@@ -1051,7 +1646,7 @@
 
                 buffer_offset = skip_bytes
 
-        self._ensure_dstream()
+        self._ensure_dctx()
 
         in_buffer = ffi.new('ZSTD_inBuffer *')
         out_buffer = ffi.new('ZSTD_outBuffer *')
@@ -1086,10 +1681,10 @@
             while in_buffer.pos < in_buffer.size:
                 assert out_buffer.pos == 0
 
-                zresult = lib.ZSTD_decompressStream(self._dstream, out_buffer, in_buffer)
+                zresult = lib.ZSTD_decompress_generic(self._dctx, out_buffer, in_buffer)
                 if lib.ZSTD_isError(zresult):
                     raise ZstdError('zstd decompress error: %s' %
-                                    ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                    _zstd_error(zresult))
 
                 if out_buffer.pos:
                     data = ffi.buffer(out_buffer.dst, out_buffer.pos)[:]
@@ -1104,12 +1699,16 @@
 
         # If we get here, input is exhausted.
 
-    def write_to(self, writer, write_size=DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE):
+    read_from = read_to_iter
+
+    def stream_writer(self, writer, write_size=DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE):
         if not hasattr(writer, 'write'):
             raise ValueError('must pass an object with a write() method')
 
         return ZstdDecompressionWriter(self, writer, write_size)
 
+    write_to = stream_writer
+
     def copy_stream(self, ifh, ofh,
                     read_size=DECOMPRESSION_RECOMMENDED_INPUT_SIZE,
                     write_size=DECOMPRESSION_RECOMMENDED_OUTPUT_SIZE):
@@ -1118,7 +1717,7 @@
         if not hasattr(ofh, 'write'):
             raise ValueError('second argument must have a write() method')
 
-        self._ensure_dstream()
+        self._ensure_dctx()
 
         in_buffer = ffi.new('ZSTD_inBuffer *')
         out_buffer = ffi.new('ZSTD_outBuffer *')
@@ -1144,10 +1743,10 @@
 
             # Flush all read data to output.
             while in_buffer.pos < in_buffer.size:
-                zresult = lib.ZSTD_decompressStream(self._dstream, out_buffer, in_buffer)
+                zresult = lib.ZSTD_decompress_generic(self._dctx, out_buffer, in_buffer)
                 if lib.ZSTD_isError(zresult):
                     raise ZstdError('zstd decompressor error: %s' %
-                                    ffi.string(lib.ZSTD_getErrorName(zresult)))
+                                    _zstd_error(zresult))
 
                 if out_buffer.pos:
                     ofh.write(ffi.buffer(out_buffer.dst, out_buffer.pos))
@@ -1172,29 +1771,36 @@
 
         # All chunks should be zstd frames and should have content size set.
         chunk_buffer = ffi.from_buffer(chunk)
-        params = ffi.new('ZSTD_frameParams *')
-        zresult = lib.ZSTD_getFrameParams(params, chunk_buffer, len(chunk_buffer))
+        params = ffi.new('ZSTD_frameHeader *')
+        zresult = lib.ZSTD_getFrameHeader(params, chunk_buffer, len(chunk_buffer))
         if lib.ZSTD_isError(zresult):
             raise ValueError('chunk 0 is not a valid zstd frame')
         elif zresult:
             raise ValueError('chunk 0 is too small to contain a zstd frame')
 
-        if not params.frameContentSize:
+        if params.frameContentSize == lib.ZSTD_CONTENTSIZE_UNKNOWN:
             raise ValueError('chunk 0 missing content size in frame')
 
-        dctx = lib.ZSTD_createDCtx()
-        if dctx == ffi.NULL:
-            raise MemoryError()
-
-        dctx = ffi.gc(dctx, lib.ZSTD_freeDCtx)
+        self._ensure_dctx(load_dict=False)
 
         last_buffer = ffi.new('char[]', params.frameContentSize)
 
-        zresult = lib.ZSTD_decompressDCtx(dctx, last_buffer, len(last_buffer),
-                                          chunk_buffer, len(chunk_buffer))
+        out_buffer = ffi.new('ZSTD_outBuffer *')
+        out_buffer.dst = last_buffer
+        out_buffer.size = len(last_buffer)
+        out_buffer.pos = 0
+
+        in_buffer = ffi.new('ZSTD_inBuffer *')
+        in_buffer.src = chunk_buffer
+        in_buffer.size = len(chunk_buffer)
+        in_buffer.pos = 0
+
+        zresult = lib.ZSTD_decompress_generic(self._dctx, out_buffer, in_buffer)
         if lib.ZSTD_isError(zresult):
             raise ZstdError('could not decompress chunk 0: %s' %
-                            ffi.string(lib.ZSTD_getErrorName(zresult)))
+                            _zstd_error(zresult))
+        elif zresult:
+            raise ZstdError('chunk 0 did not decompress full frame')
 
         # Special case of chain length of 1
         if len(frames) == 1:
@@ -1207,51 +1813,54 @@
                 raise ValueError('chunk %d must be bytes' % i)
 
             chunk_buffer = ffi.from_buffer(chunk)
-            zresult = lib.ZSTD_getFrameParams(params, chunk_buffer, len(chunk_buffer))
+            zresult = lib.ZSTD_getFrameHeader(params, chunk_buffer, len(chunk_buffer))
             if lib.ZSTD_isError(zresult):
                 raise ValueError('chunk %d is not a valid zstd frame' % i)
             elif zresult:
                 raise ValueError('chunk %d is too small to contain a zstd frame' % i)
 
-            if not params.frameContentSize:
+            if params.frameContentSize == lib.ZSTD_CONTENTSIZE_UNKNOWN:
                 raise ValueError('chunk %d missing content size in frame' % i)
 
             dest_buffer = ffi.new('char[]', params.frameContentSize)
 
-            zresult = lib.ZSTD_decompress_usingDict(dctx, dest_buffer, len(dest_buffer),
-                                                    chunk_buffer, len(chunk_buffer),
-                                                    last_buffer, len(last_buffer))
+            out_buffer.dst = dest_buffer
+            out_buffer.size = len(dest_buffer)
+            out_buffer.pos = 0
+
+            in_buffer.src = chunk_buffer
+            in_buffer.size = len(chunk_buffer)
+            in_buffer.pos = 0
+
+            zresult = lib.ZSTD_decompress_generic(self._dctx, out_buffer, in_buffer)
             if lib.ZSTD_isError(zresult):
-                raise ZstdError('could not decompress chunk %d' % i)
+                raise ZstdError('could not decompress chunk %d: %s' %
+                                _zstd_error(zresult))
+            elif zresult:
+                raise ZstdError('chunk %d did not decompress full frame' % i)
 
             last_buffer = dest_buffer
             i += 1
 
         return ffi.buffer(last_buffer, len(last_buffer))[:]
 
-    def _ensure_dstream(self):
-        if self._dstream:
-            zresult = lib.ZSTD_resetDStream(self._dstream)
+    def _ensure_dctx(self, load_dict=True):
+        lib.ZSTD_DCtx_reset(self._dctx)
+
+        if self._max_window_size:
+            zresult = lib.ZSTD_DCtx_setMaxWindowSize(self._dctx,
+                                                     self._max_window_size)
             if lib.ZSTD_isError(zresult):
-                raise ZstdError('could not reset DStream: %s' %
-                                ffi.string(lib.ZSTD_getErrorName(zresult)))
-
-            return
-
-        self._dstream = lib.ZSTD_createDStream()
-        if self._dstream == ffi.NULL:
-            raise MemoryError()
+                raise ZstdError('unable to set max window size: %s' %
+                                _zstd_error(zresult))
 
-        self._dstream = ffi.gc(self._dstream, lib.ZSTD_freeDStream)
+        zresult = lib.ZSTD_DCtx_setFormat(self._dctx, self._format)
+        if lib.ZSTD_isError(zresult):
+            raise ZstdError('unable to set decoding format: %s' %
+                            _zstd_error(zresult))
 
-        if self._dict_data:
-            zresult = lib.ZSTD_initDStream_usingDict(self._dstream,
-                                                     self._dict_data.as_bytes(),
-                                                     len(self._dict_data))
-        else:
-            zresult = lib.ZSTD_initDStream(self._dstream)
-
-        if lib.ZSTD_isError(zresult):
-            self._dstream = None
-            raise ZstdError('could not initialize DStream: %s' %
-                            ffi.string(lib.ZSTD_getErrorName(zresult)))
+        if self._dict_data and load_dict:
+            zresult = lib.ZSTD_DCtx_refDDict(self._dctx, self._dict_data._ddict)
+            if lib.ZSTD_isError(zresult):
+                raise ZstdError('unable to reference prepared dictionary: %s' %
+                                _zstd_error(zresult))