Import zstd 1.4.8vendor/zstd/1.4.8

89 files changed, 6820 insertions, 3081 deletions

diff --git a/CHANGELOG b/CHANGELOG
index 0ed939a5bbb1..86092563177c 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -1,4 +1,65 @@
-v1.4.5
+v1.4.8  (Dec 18, 2020)
+hotfix: wrong alignment of an internal buffer
+
+v1.4.7  (Dec 16, 2020)
+perf: stronger --long mode at high compression levels, by @senhuang42
+perf: stronger --patch-from at high compression levels, thanks to --long improvements
+perf: faster dictionary compression at medium compression levels, by @felixhandte
+perf: small speed & memory usage improvements for ZSTD_compress2(), by @terrelln
+perf: improved fast compression speeds with Visual Studio, by @animalize
+cli : Set nb of threads with environment variable ZSTD_NBTHREADS, by @senhuang42
+cli : accept decompressing files with *.zstd suffix
+cli : provide a condensed summary by default when processing multiple files
+cli : fix : stdin input no longer confused as user prompt
+cli : improve accuracy of several error messages
+api : new sequence ingestion API, by @senhuang42
+api : shared thread pool: control total nb of threads used by multiple compression jobs, by @marxin
+api : new ZSTD_getDictID_fromCDict(), by @LuAPi
+api : zlibWrapper only uses public API, and is compatible with dynamic library, by @terrelln
+api : fix : multithreaded compression has predictable output even in special cases (see #2327) (issue not accessible from cli)
+api : fix : dictionary compression correctly respects dictionary compression level (see #2303) (issue not accessible from cli)
+build: fix cmake script when using path with spaces, by @terrelln
+build: improved compile-time detection of aarch64/neon platforms, by @bsdimp
+build: Fix building on AIX 5.1, by @likema
+build: compile paramgrill with cmake on Windows, requested by @mirh
+doc : clarify repcode updates in format specification, by @felixhandte
+
+v1.4.6
+fix : Always return dstSize_tooSmall when that is the case
+fix : Fix ZSTD_initCStream_advanced() with static allocation and no dictionary
+perf: Improve small block decompression speed by 20%+, by @terrelln
+perf: Reduce compression stack usage by 1 KB, by @terrelln
+perf: Improve decompression speed by improving ZSTD_wildcopy, by @helloguo (#2252, #2256)
+perf: Improve histogram construction, by @cyan4973 (#2253)
+cli : Add --output-dir-mirror option, by @xxie24 (#2219)
+cli : Warn when (de)compressing multiple files into a single output, by @senhuang42 (#2279)
+cli : Improved progress bar and status summary when (de)compressing multiple files, by @senhuang42 (#2283)
+cli : Call stat less often, by @felixhandte (#2262)
+cli : Allow --patch-from XXX and --filelist XXX in addition to --patch-from=XXX and --filelist=XXX, by @cyan4973 (#2250)
+cli : Allow --patch-from to compress stdin with --stream-size, by @bimbashrestha (#2206)
+api : Do not install zbuff.h, since it has long been deprecated, by @cyan4973 (#2166).
+api : Fix ZSTD_CCtx_setParameter() with ZSTD_c_compressionLevel to make 0 mean default level, by @i-do-cpp (#2291)
+api : Rename ZSTDMT_NBTHREADS_MAX to ZSTDMT_NBWORKERS_MAX, by @marxin (#2228).
+build: Install pkg-config file with CMake and MinGW, by @tonytheodore (#2183)
+build: Install DLL with CMake on Windows, by @BioDataAnalysis (#2221)
+build: Fix DLL install location with CMake, by @xantares and @bimbashrestha (#2186)
+build: Add ZSTD_NO_UNUSED_FUNCTIONS macro to hide unused functions
+build: Add ZSTD_NO_INTRINSICS macro to avoid explicit intrinsics
+build: Add STATIC_BMI2 macro for compile time detection of BMI2 on MSVC, by @Niadb (#2258)
+build: Fix -Wcomma warnings, by @cwoffenden
+build: Remove distutils requirement for meson build, by @neheb (#2197)
+build: Fix cli compilation with uclibc
+build: Fix cli compilation without st_mtime, by @ffontaine (#2246)
+build: Fix shadowing warnings in library
+build: Fix single file library compilation with Enscripten, by @yoshihitoh (#2227)
+misc: Improve single file library and include dictBuilder, by @cwoffenden
+misc: Allow compression dictionaries with missing symbols
+misc: Add freestanding translation script in contrib/freestanding_lib
+misc: Collect all of zstd's libc dependencies into zstd_deps.h
+doc : Add ZSTD_versionString() to manual, by @animalize
+doc : Fix documentation for ZSTD_CCtxParams_setParameter(), by @felixhandte (#2270)
+
+v1.4.5  (May 22, 2020)
 fix : Compression ratio regression on huge files (> 3 GB) using high levels (--ultra) and multithreading, by @terrelln
 perf: Improved decompression speed: x64 : +10% (clang) / +5% (gcc); ARM : from +15% to +50%, depending on SoC, by @terrelln
 perf: Automatically downsizes ZSTD_DCtx when too large for too long (#2069, by @bimbashreshta)
@@ -24,7 +85,7 @@ misc: Edit-distance match finder in contrib/
 doc : Improved beginner CONTRIBUTING.md docs
 doc : New issue templates for zstd
 
-v1.4.4
+v1.4.4  (Nov 6, 2019)
 perf: Improved decompression speed, by > 10%, by @terrelln
 perf: Better compression speed when re-using a context, by @felixhandte
 perf: Fix compression ratio when compressing large files with small dictionary, by @senhuang42
@@ -51,18 +112,18 @@ pack: modified pkgconfig, for better integration into openwrt, requested by @neh
 misc: Improved documentation : ZSTD_CLEVEL, DYNAMIC_BMI2, ZSTD_CDict, function deprecation, zstd format
 misc: fixed educational decoder : accept larger literals section, and removed UNALIGNED() macro
 
-v1.4.3
+v1.4.3  (Aug 20, 2019)
 bug: Fix Dictionary Compression Ratio Regression by @cyan4973 (#1709)
 bug: Fix Buffer Overflow in legacy v0.3 decompression by @felixhandte (#1722)
 build: Add support for IAR C/C++ Compiler for Arm by @joseph0918 (#1705)
 
-v1.4.2
+v1.4.2  (Jul 26, 2019)
 bug: Fix bug in zstd-0.5 decoder by @terrelln (#1696)
 bug: Fix seekable decompression in-memory API by @iburinoc (#1695)
 misc: Validate blocks are smaller than size limit by @vivekmg (#1685)
 misc: Restructure source files by @ephiepark (#1679)
 
-v1.4.1
+v1.4.1  (Jul 20, 2019)
 bug: Fix data corruption in niche use cases by @terrelln (#1659)
 bug: Fuzz legacy modes, fix uncovered bugs by @terrelln (#1593, #1594, #1595)
 bug: Fix out of bounds read by @terrelln (#1590)
@@ -92,7 +153,7 @@ build: Visual Studio: fix linking by @absotively (#1639)
 build: Fix MinGW-W64 build by @myzhang1029 (#1600)
 misc: Expand decodecorpus coverage by @ephiepark (#1664)
 
-v1.4.0
+v1.4.0  (Apr 17, 2019)
 perf: Improve level 1 compression speed in most scenarios by 6% by @gbtucker and @terrelln
 api: Move the advanced API, including all functions in the staging section, to the stable section
 api: Make ZSTD_e_flush and ZSTD_e_end block for maximum forward progress
@@ -129,7 +190,7 @@ misc: Optimize dictionary memory usage in corner cases
 misc: Improve the dictionary builder on small or homogeneous data
 misc: Fix spelling across the repo by @jsoref
 
-v1.3.8
+v1.3.8  (Dec 28, 2018)
 perf: better decompression speed on large files (+7%) and cold dictionaries (+15%)
 perf: slightly better compression ratio at high compression modes
 api : finalized advanced API, last stage before "stable" status
@@ -151,14 +212,14 @@ doc : clarified zstd_compression_format.md, by @ulikunitz
 misc: fixed zstdgrep, returns 1 on failure, by @lzutao
 misc: NEWS renamed as CHANGELOG, in accordance with fboss
 
-v1.3.7
+v1.3.7  (Oct 20, 2018)
 perf: slightly better decompression speed on clang (depending on hardware target)
 fix : performance of dictionary compression for small input < 4 KB at levels 9 and 10
 build: no longer build backtrace by default in release mode; restrict further automatic mode
 build: control backtrace support through build macro BACKTRACE
 misc: added man pages for zstdless and zstdgrep, by @samrussell
 
-v1.3.6
+v1.3.6  (Oct 6, 2018)
 perf: much faster dictionary builder, by @jenniferliu
 perf: faster dictionary compression on small data when using multiple contexts, by @felixhandte
 perf: faster dictionary decompression when using a very large number of dictionaries simultaneously
@@ -172,7 +233,7 @@ build: Read Legacy format is limited to v0.5+ by default. Can be changed at comp
 doc : zstd_compression_format.md updated to match wording in IETF RFC 8478
 misc: tests/paramgrill, a parameter optimizer, by @GeorgeLu97
 
-v1.3.5
+v1.3.5  (Jun 29, 2018)
 perf: much faster dictionary compression, by @felixhandte
 perf: small quality improvement for dictionary generation, by @terrelln
 perf: slightly improved high compression levels (notably level 19)
@@ -187,7 +248,7 @@ build: make and make all are compatible with -j
 doc : clarify zstd_compression_format.md, updated for IETF RFC process
 misc: pzstd compatible with reproducible compilation, by @lamby
 
-v1.3.4
+v1.3.4  (Mar 27, 2018)
 perf: faster speed (especially decoding speed) on recent cpus (haswell+)
 perf: much better performance associating --long with multi-threading, by @terrelln
 perf: better compression at levels 13-15
@@ -205,7 +266,7 @@ build: VS2017 scripts, by @HaydnTrigg
 misc: all /contrib projects fixed
 misc: added /contrib/docker script by @gyscos
 
-v1.3.3
+v1.3.3  (Dec 21, 2017)
 perf: faster zstd_opt strategy (levels 16-19)
 fix : bug #944 : multithreading with shared ditionary and large data, reported by @gsliepen
 cli : fix : content size written in header by default
@@ -217,7 +278,7 @@ api : change : when setting `pledgedSrcSize`, use `ZSTD_CONTENTSIZE_UNKNOWN` mac
 build: fix : compilation under rhel6 and centos6, reported by @pixelb
 build: added `check` target
 
-v1.3.2
+v1.3.2  (Oct 10, 2017)
 new : long range mode, using --long command, by Stella Lau (@stellamplau)
 new : ability to generate and decode magicless frames (#591)
 changed : maximum nb of threads reduced to 200, to avoid address space exhaustion in 32-bits mode
@@ -240,7 +301,7 @@ example : added streaming_memory_usage
 license : changed /examples license to BSD + GPLv2
 license : fix a few header files to reflect new license (#825)
 
-v1.3.1
+v1.3.1  (Aug 21, 2017)
 New license : BSD + GPLv2
 perf: substantially decreased memory usage in Multi-threading mode, thanks to reports by Tino Reichardt (@mcmilk)
 perf: Multi-threading supports up to 256 threads. Cap at 256 when more are requested (#760)
@@ -255,7 +316,7 @@ new : contrib/adaptive-compression, I/O driven compression strength, by Paul Cru
 new : contrib/long_distance_matching, statistics by Stella Lau (@stellamplau)
 updated : contrib/linux-kernel, by Nick Terrell (@terrelln)
 
-v1.3.0
+v1.3.0  (Jul 6, 2017)
 cli : new : `--list` command, by Paul Cruz
 cli : changed : xz/lzma support enabled by default
 cli : changed : `-t *` continue processing list after a decompression error
@@ -270,7 +331,7 @@ tools : decodecorpus can generate random dictionary-compressed samples, by Paul
 new : contrib/seekable_format, demo and API, by Sean Purcell
 changed : contrib/linux-kernel, updated version and license, by Nick Terrell
 
-v1.2.0
+v1.2.0  (May 5, 2017)
 cli : changed : Multithreading enabled by default (use target zstd-nomt or HAVE_THREAD=0 to disable)
 cli : new : command -T0 means "detect and use nb of cores", by Sean Purcell
 cli : new : zstdmt symlink hardwired to `zstd -T0`
@@ -292,7 +353,7 @@ build: enabled Multi-threading support for *BSD, by Baptiste Daroussin
 tools: updated Paramgrill. Command -O# provides best parameters for sample and speed target.
 new : contrib/linux-kernel version, by Nick Terrell
 
-v1.1.4
+v1.1.4  (Mar 18, 2017)
 cli : new : can compress in *.gz format, using --format=gzip command, by Przemyslaw Skibinski
 cli : new : advanced benchmark command --priority=rt
 cli : fix : write on sparse-enabled file systems in 32-bits mode, by @ds77
@@ -308,7 +369,7 @@ build : improved cmake script, by @Majlen
 build : added -Wformat-security flag, as recommended by Padraig Brady
 doc : new : educational decoder, by Sean Purcell
 
-v1.1.3
+v1.1.3  (Feb 7, 2017)
 cli : zstd can decompress .gz files (can be disabled with `make zstd-nogz` or `make HAVE_ZLIB=0`)
 cli : new : experimental target `make zstdmt`, with multi-threading support
 cli : new : improved dictionary builder "cover" (experimental), by Nick Terrell, based on prior work by Giuseppe Ottaviano.
@@ -324,7 +385,7 @@ API : fix : all symbols properly exposed in libzstd, by Nick Terrell
 build : support for Solaris target, by Przemyslaw Skibinski
 doc : clarified specification, by Sean Purcell
 
-v1.1.2
+v1.1.2  (Dec 15, 2016)
 API : streaming : decompression : changed : automatic implicit reset when chain-decoding new frames without init
 API : experimental : added : dictID retrieval functions, and ZSTD_initCStream_srcSize()
 API : zbuff : changed : prototypes now generate deprecation warnings
@@ -341,7 +402,7 @@ zlib_wrapper : added support for gz* functions, by Przemyslaw Skibinski
 install : better compatibility with FreeBSD, by Dimitry Andric
 source tree : changed : zbuff source files moved to lib/deprecated
 
-v1.1.1
+v1.1.1  (Nov 2, 2016)
 New : command -M#, --memory=, --memlimit=, --memlimit-decompress= to limit allowed memory consumption
 New : doc/zstd_manual.html, by Przemyslaw Skibinski
 Improved : slightly better compression ratio at --ultra levels (>= 20)
@@ -352,7 +413,7 @@ Changed : zstd_errors.h is now installed within /include (and replaces errors_pu
 Updated man page
 Fixed : zstd-small, zstd-compress and zstd-decompress compilation targets
 
-v1.1.0
+v1.1.0  (Sep 28, 2016)
 New : contrib/pzstd, parallel version of zstd, by Nick Terrell
 added : NetBSD install target (#338)
 Improved : speed for batches of small files
@@ -366,7 +427,7 @@ Fixed : compatibility with OpenBSD, reported by Juan Francisco Cantero Hurtado (
 Fixed : compatibility with Hurd, by Przemyslaw Skibinski (#365)
 Fixed : zstd-pgo, reported by octoploid (#329)
 
-v1.0.0
+v1.0.0  (Sep 1, 2016)
 Change Licensing, all project is now BSD, Copyright Facebook
 Small decompression speed improvement
 API : Streaming API supports legacy format
@@ -375,7 +436,7 @@ CLI supports legacy formats v0.4+
 Fixed : compression fails on certain huge files, reported by Jesse McGrew
 Enhanced documentation, by Przemyslaw Skibinski
 
-v0.8.1
+v0.8.1  (Aug 18, 2016)
 New streaming API
 Changed : --ultra now enables levels beyond 19
 Changed : -i# now selects benchmark time in second
@@ -384,7 +445,7 @@ Fixed : speed regression on specific patterns (#272)
 Fixed : support for Z_SYNC_FLUSH, by Dmitry Krot (#291)
 Fixed : ICC compilation, by Przemyslaw Skibinski
 
-v0.8.0
+v0.8.0  (Aug 2, 2016)
 Improved : better speed on clang and gcc -O2, thanks to Eric Biggers
 New : Build on FreeBSD and DragonFly, thanks to JrMarino
 Changed : modified API : ZSTD_compressEnd()
@@ -397,17 +458,17 @@ Modified : minor compression level adaptations
 Updated : compression format specification to v0.2.0
 changed : zstd.h moved to /lib directory
 
-v0.7.5
+v0.7.5  (Aug 1, 2016)
 Transition version, supporting decoding of v0.8.x
 
-v0.7.4
+v0.7.4  (Jul 17, 2016)
 Added : homebrew for Mac, by Daniel Cade
 Added : more examples
 Fixed : segfault when using small dictionaries, reported by Felix Handte
 Modified : default compression level for CLI is now 3
 Updated : specification, to v0.1.1
 
-v0.7.3
+v0.7.3  (Jul 9, 2016)
 New : compression format specification
 New : `--` separator, stating that all following arguments are file names. Suggested by Chip Turner.
 New : `ZSTD_getDecompressedSize()`
@@ -419,18 +480,18 @@ fixed : multi-blocks decoding with intermediate uncompressed blocks, reported by
 modified : removed "mem.h" and "error_public.h" dependencies from "zstd.h" (experimental section)
 modified : legacy functions no longer need magic number
 
-v0.7.2
+v0.7.2  (Jul 4, 2016)
 fixed : ZSTD_decompressBlock() using multiple consecutive blocks. Reported by Greg Slazinski.
 fixed : potential segfault on very large files (many gigabytes). Reported by Chip Turner.
 fixed : CLI displays system error message when destination file cannot be created (#231). Reported by Chip Turner.
 
-v0.7.1
+v0.7.1  (Jun 23, 2016)
 fixed : ZBUFF_compressEnd() called multiple times with too small `dst` buffer, reported by Christophe Chevalier
 fixed : dictBuilder fails if first sample is too small, reported by Руслан Ковалёв
 fixed : corruption issue, reported by cj
 modified : checksum enabled by default in command line mode
 
-v0.7.0
+v0.7.0  (Jun 17, 2016)
 New : Support for directory compression, using `-r`, thanks to Przemyslaw Skibinski
 New : Command `--rm`, to remove source file after successful de/compression
 New : Visual build scripts, by Christophe Chevalier
@@ -443,7 +504,7 @@ API : support for custom malloc/free functions
 New : controllable Dictionary ID
 New : Support for skippable frames
 
-v0.6.1
+v0.6.1  (May 13, 2016)
 New : zlib wrapper API, thanks to Przemyslaw Skibinski
 New : Ability to compile compressor / decompressor separately
 Changed : new lib directory structure
@@ -453,103 +514,103 @@ Fixed : null-string roundtrip (#176)
 New : benchmark mode can select directory as input
 Experimental : midipix support, VMS support
 
-v0.6.0
+v0.6.0  (Apr 13, 2016)
 Stronger high compression modes, thanks to Przemyslaw Skibinski
 API : ZSTD_getFrameParams() provides size of decompressed content
 New : highest compression modes require `--ultra` command to fully unleash their capacity
 Fixed : zstd cli return error code > 0 and removes dst file artifact when decompression fails, thanks to Chip Turner
 
-v0.5.1
+v0.5.1  (Feb 18, 2016)
 New : Optimal parsing => Very high compression modes, thanks to Przemyslaw Skibinski
 Changed : Dictionary builder integrated into libzstd and zstd cli
 Changed (!) : zstd cli now uses "multiple input files" as default mode. See `zstd -h`.
 Fix : high compression modes for big-endian platforms
 New : zstd cli : `-t` | `--test` command
 
-v0.5.0
+v0.5.0  (Feb 5, 2016)
 New : dictionary builder utility
 Changed : streaming & dictionary API
 Improved : better compression of small data
 
-v0.4.7
+v0.4.7  (Jan 22, 2016)
 Improved : small compression speed improvement in HC mode
 Changed : `zstd_decompress.c` has ZSTD_LEGACY_SUPPORT to 0 by default
 fix : bt search bug
 
-v0.4.6
+v0.4.6  (Jan 13, 2016)
 fix : fast compression mode on Windows
 New : cmake configuration file, thanks to Artyom Dymchenko
 Improved : high compression mode on repetitive data
 New : block-level API
 New : ZSTD_duplicateCCtx()
 
-v0.4.5
+v0.4.5  (Dec 18, 2015)
 new : -m/--multiple : compress/decompress multiple files
 
-v0.4.4
+v0.4.4  (Dec 14, 2015)
 Fixed : high compression modes for Windows 32 bits
 new : external dictionary API extended to buffered mode and accessible through command line
 new : windows DLL project, thanks to Christophe Chevalier
 
-v0.4.3 :
+v0.4.3  (Dec 7, 2015)
 new : external dictionary API
 new : zstd-frugal
 
-v0.4.2 :
+v0.4.2  (Dec 2, 2015)
 Generic minor improvements for small blocks
 Fixed : big-endian compatibility, by Peter Harris (#85)
 
-v0.4.1
+v0.4.1  (Dec 1, 2015)
 Fixed : ZSTD_LEGACY_SUPPORT=0 build mode (reported by Luben)
 removed `zstd.c`
 
-v0.4.0
+v0.4.0  (Nov 29, 2015)
 Command line utility compatible with high compression levels
 Removed zstdhc => merged into zstd
 Added : ZBUFF API (see zstd_buffered.h)
 Rolling buffer support
 
-v0.3.6
+v0.3.6  (Nov 10, 2015)
 small blocks params
 
-v0.3.5
+v0.3.5  (Nov 9, 2015)
 minor generic compression improvements
 
-v0.3.4
+v0.3.4  (Nov 6, 2015)
 Faster fast cLevels
 
-v0.3.3
+v0.3.3  (Nov 5, 2015)
 Small compression ratio improvement
 
-v0.3.2
+v0.3.2  (Nov 2, 2015)
 Fixed Visual Studio
 
-v0.3.1 :
+v0.3.1  (Nov 2, 2015)
 Small compression ratio improvement
 
-v0.3
+v0.3  (Oct 30, 2015)
 HC mode : compression levels 2-26
 
-v0.2.2
+v0.2.2  (Oct 28, 2015)
 Fix : Visual Studio 2013 & 2015 release compilation, by Christophe Chevalier
 
-v0.2.1
+v0.2.1  (Oct 24, 2015)
 Fix : Read errors, advanced fuzzer tests, by Hanno Böck
 
-v0.2.0
+v0.2.0  (Oct 22, 2015)
 **Breaking format change**
 Faster decompression speed
 Can still decode v0.1 format
 
-v0.1.3
+v0.1.3  (Oct 15, 2015)
 fix uninitialization warning, reported by Evan Nemerson
 
-v0.1.2
+v0.1.2  (Sep 11, 2015)
 frame concatenation support
 
-v0.1.1
+v0.1.1  (Aug 27, 2015)
 fix compression bug
 detects write-flush errors
 
-v0.1.0
+v0.1.0  (Aug 25, 2015)
 first release
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 637e37188550..44f2393a2c15 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -5,7 +5,7 @@ possible.
 ## Our Development Process
 New versions are being developed in the "dev" branch,
 or in their own feature branch.
-When they are deemed ready for a release, they are merged into "master".
+When they are deemed ready for a release, they are merged into "release".
 
 As a consequences, all contributions must stage first through "dev"
 or their own feature branch.
@@ -126,6 +126,20 @@ just `contrib/largeNbDicts` and nothing else, you can run:
 scan-build make -C contrib/largeNbDicts largeNbDicts
 ```
 
+### Pitfalls of static analysis
+`scan-build` is part of our regular CI suite. Other static analyzers are not.
+
+It can be useful to look at additional static analyzers once in a while (and we do), but it's not a good idea to multiply the nb of analyzers run continuously at each commit and PR. The reasons are :
+
+- Static analyzers are full of false positive. The signal to noise ratio is actually pretty low.
+- A good CI policy is "zero-warning tolerance". That means that all issues must be solved, including false positives. This quickly becomes a tedious workload.
+- Multiple static analyzers will feature multiple kind of false positives, sometimes applying to the same code but in different ways leading to :
+   + torteous code, trying to please multiple constraints, hurting readability and therefore maintenance. Sometimes, such complexity introduce other more subtle bugs, that are just out of scope of the analyzers.
+   + sometimes, these constraints are mutually exclusive : if one try to solve one, the other static analyzer will complain, they can't be both happy at the same time.
+- As if that was not enough, the list of false positives change with each version. It's hard enough to follow one static analyzer, but multiple ones with their own update agenda, this quickly becomes a massive velocity reducer.
+
+This is different from running a static analyzer once in a while, looking at the output, and __cherry picking__ a few warnings that seem helpful, either because they detected a genuine risk of bug, or because it helps expressing the code in a way which is more readable or more difficult to misuse. These kind of reports can be useful, and are accepted.
+
 ## Performance
 Performance is extremely important for zstd and we only merge pull requests whose performance
 landscape and corresponding trade-offs have been adequately analyzed, reproduced, and presented.
@@ -369,7 +383,7 @@ CI tests run every time a pull request (PR) is created or updated. The exact tes
 that get run will depend on the destination branch you specify. Some tests take
 longer to run than others. Currently, our CI is set up to run a short
 series of tests when creating a PR to the dev branch and a longer series of tests
-when creating a PR to the master branch. You can look in the configuration files
+when creating a PR to the release branch. You can look in the configuration files
 of the respective CI platform for more information on what gets run when.
 
 Most people will just want to create a PR with the destination set to their local dev
diff --git a/Makefile b/Makefile
index 2c1d34604fe9..2832fb4752b8 100644
--- a/Makefile
+++ b/Makefile
@@ -8,6 +8,9 @@
 # You may select, at your option, one of the above-listed licenses.
 # ################################################################
 
+# verbose mode (print commands) on V=1 or VERBOSE=1
+Q = $(if $(filter 1,$(V) $(VERBOSE)),,@)
+
 PRGDIR   = programs
 ZSTDDIR  = lib
 BUILDIR  = build
@@ -28,9 +31,9 @@ VOID = /dev/null
 TARGET_SYSTEM ?= $(OS)
 
 ifneq (,$(filter Windows%,$(TARGET_SYSTEM)))
-EXT =.exe
+  EXT =.exe
 else
-EXT =
+  EXT =
 endif
 
 ## default: Build lib-release and zstd-release
@@ -46,8 +49,8 @@ allmost: allzstd zlibwrapper
 # skip zwrapper, can't build that on alternate architectures without the proper zlib installed
 .PHONY: allzstd
 allzstd: lib-all
-	$(MAKE) -C $(PRGDIR) all
-	$(MAKE) -C $(TESTDIR) all
+	$(Q)$(MAKE) -C $(PRGDIR) all
+	$(Q)$(MAKE) -C $(TESTDIR) all
 
 .PHONY: all32
 all32:
@@ -55,18 +58,19 @@ all32:
 	$(MAKE) -C $(TESTDIR) all32
 
 .PHONY: lib lib-release libzstd.a
+lib-all : lib
 lib lib-release lib-all :
-	@$(MAKE) -C $(ZSTDDIR) $@
+	$(Q)$(MAKE) -C $(ZSTDDIR) $@
 
 .PHONY: zstd zstd-release
 zstd zstd-release:
-	@$(MAKE) -C $(PRGDIR) $@
-	cp $(PRGDIR)/zstd$(EXT) .
+	$(Q)$(MAKE) -C $(PRGDIR) $@
+	$(Q)ln -sf $(PRGDIR)/zstd$(EXT) zstd$(EXT)
 
 .PHONY: zstdmt
 zstdmt:
-	@$(MAKE) -C $(PRGDIR) $@
-	cp $(PRGDIR)/zstd$(EXT) ./zstdmt$(EXT)
+	$(Q)$(MAKE) -C $(PRGDIR) $@
+	$(Q)cp $(PRGDIR)/zstd$(EXT) ./zstdmt$(EXT)
 
 .PHONY: zlibwrapper
 zlibwrapper: lib
@@ -75,16 +79,16 @@ zlibwrapper: lib
 ## test: run long-duration tests
 .PHONY: test
 DEBUGLEVEL ?= 1
-test: MOREFLAGS += -g -DDEBUGLEVEL=$(DEBUGLEVEL) -Werror
+test: MOREFLAGS += -g -Werror
 test:
-	MOREFLAGS="$(MOREFLAGS)" $(MAKE) -j -C $(PRGDIR) allVariants
+	DEBUGLEVEL=$(DEBUGLEVEL) MOREFLAGS="$(MOREFLAGS)" $(MAKE) -j -C $(PRGDIR) allVariants
 	$(MAKE) -C $(TESTDIR) $@
-	ZSTD=../../programs/zstd $(MAKE) -C doc/educational_decoder test
+	ZSTD=../../programs/zstd $(MAKE) -C doc/educational_decoder $@
 
 ## shortest: same as `make check`
 .PHONY: shortest
 shortest:
-	$(MAKE) -C $(TESTDIR) $@
+	$(Q)$(MAKE) -C $(TESTDIR) $@
 
 ## check: run basic tests for `zstd` cli
 .PHONY: check
@@ -97,10 +101,10 @@ automated_benchmarking:
 .PHONY: benchmarking
 benchmarking: automated_benchmarking
 
-## examples: build all examples in `/examples` directory
+## examples: build all examples in `examples/` directory
 .PHONY: examples
 examples: lib
-	CPPFLAGS=-I../lib LDFLAGS=-L../lib $(MAKE) -C examples/ all
+	$(MAKE) -C examples all
 
 ## manual: generate API documentation in html format
 .PHONY: manual
@@ -117,6 +121,7 @@ man:
 contrib: lib
 	$(MAKE) -C contrib/pzstd all
 	$(MAKE) -C contrib/seekable_format/examples all
+	$(MAKE) -C contrib/seekable_format/tests test
 	$(MAKE) -C contrib/largeNbDicts all
 	cd contrib/single_file_libs/ ; ./build_decoder_test.sh
 	cd contrib/single_file_libs/ ; ./build_library_test.sh
@@ -127,17 +132,18 @@ cleanTabs:
 
 .PHONY: clean
 clean:
-	@$(MAKE) -C $(ZSTDDIR) $@ > $(VOID)
-	@$(MAKE) -C $(PRGDIR) $@ > $(VOID)
-	@$(MAKE) -C $(TESTDIR) $@ > $(VOID)
-	@$(MAKE) -C $(ZWRAPDIR) $@ > $(VOID)
-	@$(MAKE) -C examples/ $@ > $(VOID)
-	@$(MAKE) -C contrib/gen_html $@ > $(VOID)
-	@$(MAKE) -C contrib/pzstd $@ > $(VOID)
-	@$(MAKE) -C contrib/seekable_format/examples $@ > $(VOID)
-	@$(MAKE) -C contrib/largeNbDicts $@ > $(VOID)
-	@$(RM) zstd$(EXT) zstdmt$(EXT) tmp*
-	@$(RM) -r lz4
+	$(Q)$(MAKE) -C $(ZSTDDIR) $@ > $(VOID)
+	$(Q)$(MAKE) -C $(PRGDIR) $@ > $(VOID)
+	$(Q)$(MAKE) -C $(TESTDIR) $@ > $(VOID)
+	$(Q)$(MAKE) -C $(ZWRAPDIR) $@ > $(VOID)
+	$(Q)$(MAKE) -C examples/ $@ > $(VOID)
+	$(Q)$(MAKE) -C contrib/gen_html $@ > $(VOID)
+	$(Q)$(MAKE) -C contrib/pzstd $@ > $(VOID)
+	$(Q)$(MAKE) -C contrib/seekable_format/examples $@ > $(VOID)
+	$(Q)$(MAKE) -C contrib/seekable_format/tests $@ > $(VOID)
+	$(Q)$(MAKE) -C contrib/largeNbDicts $@ > $(VOID)
+	$(Q)$(RM) zstd$(EXT) zstdmt$(EXT) tmp*
+	$(Q)$(RM) -r lz4
 	@echo Cleaning completed
 
 #------------------------------------------------------------------------------
@@ -161,7 +167,7 @@ EGREP = egrep $(EGREP_OPTIONS)
 ## list: Print all targets and their descriptions (if provided)
 .PHONY: list
 list:
-	@TARGETS=$$($(MAKE) -pRrq -f $(lastword $(MAKEFILE_LIST)) : 2>/dev/null \
+	$(Q)TARGETS=$$($(MAKE) -pRrq -f $(lastword $(MAKEFILE_LIST)) : 2>/dev/null \
 		| awk -v RS= -F: '/^# File/,/^# Finished Make data base/ {if ($$1 !~ "^[#.]") {print $$1}}' \
 		| $(EGREP) -v  -e '^[^[:alnum:]]' | sort); \
 	{ \
@@ -176,13 +182,13 @@ list:
 
 .PHONY: install armtest usan asan uasan
 install:
-	@$(MAKE) -C $(ZSTDDIR) $@
-	@$(MAKE) -C $(PRGDIR) $@
+	$(Q)$(MAKE) -C $(ZSTDDIR) $@
+	$(Q)$(MAKE) -C $(PRGDIR) $@
 
 .PHONY: uninstall
 uninstall:
-	@$(MAKE) -C $(ZSTDDIR) $@
-	@$(MAKE) -C $(PRGDIR) $@
+	$(Q)$(MAKE) -C $(ZSTDDIR) $@
+	$(Q)$(MAKE) -C $(PRGDIR) $@
 
 .PHONY: travis-install
 travis-install:
diff --git a/README.md b/README.md
index 5c300fdc49aa..dcca7662d2ff 100644
--- a/README.md
+++ b/README.md
@@ -176,6 +176,12 @@ Going into `build` directory, you will find additional possibilities:
 You can build the zstd binary via buck by executing: `buck build programs:zstd` from the root of the repo.
 The output binary will be in `buck-out/gen/programs/`.
 
+## Testing
+
+You can run quick local smoke tests by executing the `playTest.sh` script from the `src/tests` directory.
+Two env variables `$ZSTD_BIN` and `$DATAGEN_BIN` are needed for the test script to locate the zstd and datagen binary.
+For information on CI testing, please refer to TESTING.md
+
 ## Status
 
 Zstandard is currently deployed within Facebook. It is used continuously to compress large amounts of data in multiple formats and use cases.
@@ -187,7 +193,7 @@ Zstandard is dual-licensed under [BSD](LICENSE) and [GPLv2](COPYING).
 
 ## Contributing
 
-The "dev" branch is the one where all contributions are merged before reaching "master".
-If you plan to propose a patch, please commit into the "dev" branch, or its own feature branch.
-Direct commit to "master" are not permitted.
+The `dev` branch is the one where all contributions are merged before reaching `release`.
+If you plan to propose a patch, please commit into the `dev` branch, or its own feature branch.
+Direct commit to `release` are not permitted.
 For more information, please read [CONTRIBUTING](CONTRIBUTING.md).
diff --git a/TESTING.md b/TESTING.md
index 7e5305178b97..b851d1c8d71a 100644
--- a/TESTING.md
+++ b/TESTING.md
@@ -27,7 +27,7 @@ They consist of the following tests:
 
 Long Tests
 ----------
-Long tests run on all commits to `master` branch,
+Long tests run on all commits to `release` branch,
 and once a day on the current version of `dev` branch,
 on TravisCI.
 They consist of the following tests:
diff --git a/appveyor.yml b/appveyor.yml
index 5d77b3103481..0e872557525a 100644
--- a/appveyor.yml
+++ b/appveyor.yml
@@ -1,20 +1,20 @@
-# Following tests are run _only_ on master branch
-# To reproduce these tests, it's possible to push into a branch `appveyorTest`
-# or a branch `visual*`, they will intentionnally trigger `master` tests
+# Following tests are run _only_ on `release` branch
+# and on selected feature branch named `appveyorTest` or `visual*`
 
 -
   version: 1.0.{build}
   branches:
     only:
+    - release
     - master
-    - appveyorTest
+    - /appveyor*/
     - /visual*/
   environment:
     matrix:
     - COMPILER: "gcc"
       HOST:     "mingw"
       PLATFORM: "x64"
-      SCRIPT:   "make allzstd MOREFLAGS=-static && make -C tests fullbench-lib"
+      SCRIPT:   "make allzstd MOREFLAGS=-static"
       ARTIFACT: "true"
       BUILD:    "true"
     - COMPILER: "gcc"
@@ -92,9 +92,9 @@
           cd programs\ && 7z a -tzip -mx9 zstd-win-binary-%PLATFORM%.zip zstd.exe &&
           appveyor PushArtifact zstd-win-binary-%PLATFORM%.zip &&
           cp zstd.exe ..\bin\zstd.exe &&
-          git clone --depth 1 --branch master https://github.com/facebook/zstd &&
+          git clone --depth 1 --branch release https://github.com/facebook/zstd &&
           cd zstd &&
-          git archive --format=tar master -o zstd-src.tar &&
+          git archive --format=tar release -o zstd-src.tar &&
           ..\zstd -19 zstd-src.tar &&
           appveyor PushArtifact zstd-src.tar.zst &&
           certUtil -hashfile zstd-src.tar.zst SHA256 > zstd-src.tar.zst.sha256.sig &&
@@ -162,6 +162,8 @@
   - if [%TEST%]==[cmake] (
       mkdir build\cmake\build &&
       cd build\cmake\build &&
+      SET FUZZERTEST=-T2mn &&
+      SET ZSTREAM_TESTTIME=-T2mn &&
       cmake -G "Visual Studio 14 2015 Win64" .. &&
       cd ..\..\.. &&
       make clean
@@ -194,7 +196,7 @@
     - COMPILER: "gcc"
       HOST:     "mingw"
       PLATFORM: "x64"
-      SCRIPT:   "CPPFLAGS=-DDEBUGLEVEL=2 CFLAGS=-Werror make -j allzstd DEBUGLEVEL=2"
+      SCRIPT:   "CFLAGS=-Werror make -j allzstd DEBUGLEVEL=2"
     - COMPILER: "gcc"
       HOST:     "mingw"
       PLATFORM: "x86"
@@ -285,5 +287,6 @@
   - ECHO Testing %COMPILER% %PLATFORM% %CONFIGURATION%
   - if [%HOST%]==[mingw] (
       set "CC=%COMPILER%" &&
+      make clean &&
       make check
     )
diff --git a/doc/zstd_compression_format.md b/doc/zstd_compression_format.md
index fc61726fc98c..0af6bf91a204 100644
--- a/doc/zstd_compression_format.md
+++ b/doc/zstd_compression_format.md
@@ -3,7 +3,7 @@ Zstandard Compression Format
 
 ### Notices
 
-Copyright (c) 2016-present Yann Collet, Facebook, Inc.
+Copyright (c) 2016-2020 Yann Collet, Facebook, Inc.
 
 Permission is granted to copy and distribute this document
 for any purpose and without charge,
@@ -16,7 +16,7 @@ Distribution of this document is unlimited.
 
 ### Version
 
-0.3.5 (13/11/19)
+0.3.7 (2020-12-09)
 
 
 Introduction
@@ -291,21 +291,10 @@ Format is __little-endian__.
 It's allowed to represent a small ID (for example `13`)
 with a large 4-bytes dictionary ID, even if it is less efficient.
 
-_Reserved ranges :_
-Within private environments, any `Dictionary_ID` can be used.
-
-However, for frames and dictionaries distributed in public space,
-`Dictionary_ID` must be attributed carefully.
-Rules for public environment are not yet decided,
-but the following ranges are reserved for some future registrar :
-- low range  : `<= 32767`
-- high range : `>= (1 << 31)`
-
-Outside of these ranges, any value of `Dictionary_ID`
-which is both `>= 32768` and `< (1<<31)` can be used freely,
-even in public environment.
-
-
+A value of `0` has same meaning as no `Dictionary_ID`,
+in which case the frame may or may not need a dictionary to be decoded,
+and the ID of such a dictionary is not specified.
+The decoder must know this information by other means.
 
 #### `Frame_Content_Size`
 
@@ -389,7 +378,7 @@ __`Block_Size`__
 The upper 21 bits of `Block_Header` represent the `Block_Size`.
 
 When `Block_Type` is `Compressed_Block` or `Raw_Block`,
-`Block_Size` is the size of `Block_Content` (hence excluding `Block_Header`).  
+`Block_Size` is the size of `Block_Content` (hence excluding `Block_Header`).
 
 When `Block_Type` is `RLE_Block`, since `Block_Content`’s size is always 1,
 `Block_Size` represents the number of times this byte must be repeated.
@@ -929,38 +918,41 @@ Note that blocks which are not `Compressed_Block` are skipped, they do not contr
 
 ###### Offset updates rules
 
-The newest offset takes the lead in offset history,
-shifting others back by one rank,
-up to the previous rank of the new offset _if it was present in history_.
-
-__Examples__ :
-
-In the common case, when new offset is not part of history :
-`Repeated_Offset3` = `Repeated_Offset2`
-`Repeated_Offset2` = `Repeated_Offset1`
-`Repeated_Offset1` = `NewOffset`
-
-When the new offset _is_ part of history, there may be specific adjustments.
-
-When `NewOffset` == `Repeated_Offset1`, offset history remains actually unmodified.
-
-When `NewOffset` == `Repeated_Offset2`,
-`Repeated_Offset1` and `Repeated_Offset2` ranks are swapped.
-`Repeated_Offset3` is unmodified.
-
-When `NewOffset` == `Repeated_Offset3`,
-there is actually no difference with the common case :
-all offsets are shifted by one rank,
-`NewOffset` (== `Repeated_Offset3`) becomes the new `Repeated_Offset1`.
-
-Also worth mentioning, the specific corner case when `offset_value` == 3,
-and the literal length of the current sequence is zero.
-In which case , `NewOffset` = `Repeated_Offset1` - 1_byte.
-Here also, from an offset history update perspective, it's just a common case :
-`Repeated_Offset3` = `Repeated_Offset2`
-`Repeated_Offset2` = `Repeated_Offset1`
-`Repeated_Offset1` = `NewOffset` ( == `Repeated_Offset1` - 1_byte )
-
+During the execution of the sequences of a `Compressed_Block`, the
+`Repeated_Offsets`' values are kept up to date, so that they always represent
+the three most-recently used offsets. In order to achieve that, they are
+updated after executing each sequence in the following way:
+
+When the sequence's `offset_value` does not refer to one of the
+`Repeated_Offsets`--when it has value greater than 3, or when it has value 3
+and the sequence's `literals_length` is zero--the `Repeated_Offsets`' values
+are shifted back one, and `Repeated_Offset1` takes on the value of the
+just-used offset.
+
+Otherwise, when the sequence's `offset_value` refers to one of the
+`Repeated_Offsets`--when it has value 1 or 2, or when it has value 3 and the
+sequence's `literals_length` is non-zero--the `Repeated_Offsets` are re-ordered
+so that `Repeated_Offset1` takes on the value of the used Repeated_Offset, and
+the existing values are pushed back from the first `Repeated_Offset` through to
+the `Repeated_Offset` selected by the `offset_value`. This effectively performs
+a single-stepped wrapping rotation of the values of these offsets, so that
+their order again reflects the recency of their use.
+
+The following table shows the values of the `Repeated_Offsets` as a series of
+sequences are applied to them:
+
+| `offset_value` | `literals_length` | `Repeated_Offset1` | `Repeated_Offset2` | `Repeated_Offset3` | Comment                 |
+|:--------------:|:-----------------:|:------------------:|:------------------:|:------------------:|:-----------------------:|
+|                |                   |                  1 |                  4 |                  8 | starting values         |
+|           1114 |                11 |               1111 |                  1 |                  4 | non-repeat              |
+|              1 |                22 |               1111 |                  1 |                  4 | repeat 1; no change     |
+|           2225 |                22 |               2222 |               1111 |                  1 | non-repeat              |
+|           1114 |               111 |               1111 |               2222 |               1111 | non-repeat              |
+|           3336 |                33 |               3333 |               1111 |               2222 | non-repeat              |
+|              2 |                22 |               1111 |               3333 |               2222 | repeat 2; swap 1 & 2    |
+|              3 |                33 |               2222 |               1111 |               3333 | repeat 3; rotate 3 to 1 |
+|              3 |                 0 |               2221 |               2222 |               1111 | insert resolved offset  |
+|              1 |                 0 |               2222 |               2221 |               3333 | repeat 2                |
 
 
 Skippable Frames
@@ -1429,13 +1421,17 @@ __`Dictionary_ID`__ : 4 bytes, stored in __little-endian__ format.
               It's used by decoders to check if they use the correct dictionary.
 
 _Reserved ranges :_
-              If the frame is going to be distributed in a private environment,
-              any `Dictionary_ID` can be used.
-              However, for public distribution of compressed frames,
-              the following ranges are reserved and shall not be used :
+If the dictionary is going to be distributed in a public environment,
+the following ranges of `Dictionary_ID` are reserved for some future registrar
+and shall not be used :
+
+    - low range  : <= 32767
+    - high range : >= (2^31)
+
+Outside of these ranges, any value of `Dictionary_ID`
+which is both `>= 32768` and `< (1<<31)` can be used freely,
+even in public environment.
 
-              - low range  : <= 32767
-              - high range : >= (2^31)
 
 __`Entropy_Tables`__ : follow the same format as tables in [compressed blocks].
               See the relevant [FSE](#fse-table-description)
@@ -1447,7 +1443,7 @@ __`Entropy_Tables`__ : follow the same format as tables in [compressed blocks].
               Repeat distribution mode for sequence decoding.
               It's finally followed by 3 offset values, populating recent offsets (instead of using `{1,4,8}`),
               stored in order, 4-bytes __little-endian__ each, for a total of 12 bytes.
-              Each recent offset must have a value < dictionary size.
+              Each recent offset must have a value <= dictionary content size, and cannot equal 0.
 
 __`Content`__ : The rest of the dictionary is its content.
               The content act as a "past" in front of data to compress or decompress,
@@ -1455,7 +1451,7 @@ __`Content`__ : The rest of the dictionary is its content.
               As long as the amount of data decoded from this frame is less than or
               equal to `Window_Size`, sequence commands may specify offsets longer
               than the total length of decoded output so far to reference back to the
-              dictionary, even parts of the dictionary with offsets larger than `Window_Size`.  
+              dictionary, even parts of the dictionary with offsets larger than `Window_Size`.
               After the total output has surpassed `Window_Size` however,
               this is no longer allowed and the dictionary is no longer accessible.
 
@@ -1673,6 +1669,8 @@ or at least provide a meaningful error code explaining for which reason it canno
 
 Version changes
 ---------------
+- 0.3.7 : clarifications for Repeat_Offsets
+- 0.3.6 : clarifications for Dictionary_ID
 - 0.3.5 : clarifications for Block_Maximum_Size
 - 0.3.4 : clarifications for FSE decoding table
 - 0.3.3 : clarifications for field Block_Size
diff --git a/doc/zstd_manual.html b/doc/zstd_manual.html
index fe58f78cb153..315488844ba6 100644
--- a/doc/zstd_manual.html
+++ b/doc/zstd_manual.html
@@ -1,10 +1,10 @@
 <html>
 <head>
 <meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1">
-<title>zstd 1.4.5 Manual</title>
+<title>zstd 1.4.8 Manual</title>
 </head>
 <body>
-<h1>zstd 1.4.5 Manual</h1>
+<h1>zstd 1.4.8 Manual</h1>
 <hr>
 <a name="Contents"></a><h2>Contents</h2>
 <ol>
@@ -27,16 +27,10 @@
 <li><a href="#Chapter17">Advanced compression functions</a></li>
 <li><a href="#Chapter18">Advanced decompression functions</a></li>
 <li><a href="#Chapter19">Advanced streaming functions</a></li>
-<li><a href="#Chapter20">! ZSTD_initCStream_usingDict() :</a></li>
-<li><a href="#Chapter21">! ZSTD_initCStream_advanced() :</a></li>
-<li><a href="#Chapter22">! ZSTD_initCStream_usingCDict() :</a></li>
-<li><a href="#Chapter23">! ZSTD_initCStream_usingCDict_advanced() :</a></li>
-<li><a href="#Chapter24">This function is deprecated, and is equivalent to:</a></li>
-<li><a href="#Chapter25">This function is deprecated, and is equivalent to:</a></li>
-<li><a href="#Chapter26">Buffer-less and synchronous inner streaming functions</a></li>
-<li><a href="#Chapter27">Buffer-less streaming compression (synchronous mode)</a></li>
-<li><a href="#Chapter28">Buffer-less streaming decompression (synchronous mode)</a></li>
-<li><a href="#Chapter29">Block level API</a></li>
+<li><a href="#Chapter20">Buffer-less and synchronous inner streaming functions</a></li>
+<li><a href="#Chapter21">Buffer-less streaming compression (synchronous mode)</a></li>
+<li><a href="#Chapter22">Buffer-less streaming decompression (synchronous mode)</a></li>
+<li><a href="#Chapter23">Block level API</a></li>
 </ol>
 <hr>
 <a name="Chapter1"></a><h2>Introduction</h2><pre>
@@ -72,8 +66,14 @@
 
 <a name="Chapter2"></a><h2>Version</h2><pre></pre>
 
-<pre><b>unsigned ZSTD_versionNumber(void);   </b>/**< to check runtime library version */<b>
-</b></pre><BR>
+<pre><b>unsigned ZSTD_versionNumber(void);
+</b><p>  Return runtime library version, the value is (MAJOR*100*100 + MINOR*100 + RELEASE). 
+</p></pre><BR>
+
+<pre><b>const char* ZSTD_versionString(void);
+</b><p>  Return runtime library version, like "1.4.5". Requires v1.3.0+. 
+</p></pre><BR>
+
 <a name="Chapter3"></a><h2>Simple API</h2><pre></pre>
 
 <pre><b>size_t ZSTD_compress( void* dst, size_t dstCapacity,
@@ -277,7 +277,9 @@ size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
                                      * for large inputs, by finding large matches at long distance.
                                      * It increases memory usage and window size.
                                      * Note: enabling this parameter increases default ZSTD_c_windowLog to 128 MB
-                                     * except when expressly set to a different value. */
+                                     * except when expressly set to a different value.
+                                     * Note: will be enabled by default if ZSTD_c_windowLog >= 128 MB and
+                                     * compression strategy >= ZSTD_btopt (== compression level 16+) */
     ZSTD_c_ldmHashLog=161,   </b>/* Size of the table for long distance matching, as a power of 2.<b>
                               * Larger values increase memory usage and compression ratio,
                               * but decrease compression speed.
@@ -308,16 +310,20 @@ size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
     ZSTD_c_dictIDFlag=202,   </b>/* When applicable, dictionary's ID is written into frame header (default:1) */<b>
 
     </b>/* multi-threading parameters */<b>
-    </b>/* These parameters are only useful if multi-threading is enabled (compiled with build macro ZSTD_MULTITHREAD).<b>
-     * They return an error otherwise. */
+    </b>/* These parameters are only active if multi-threading is enabled (compiled with build macro ZSTD_MULTITHREAD).<b>
+     * Otherwise, trying to set any other value than default (0) will be a no-op and return an error.
+     * In a situation where it's unknown if the linked library supports multi-threading or not,
+     * setting ZSTD_c_nbWorkers to any value >= 1 and consulting the return value provides a quick way to check this property.
+     */
     ZSTD_c_nbWorkers=400,    </b>/* Select how many threads will be spawned to compress in parallel.<b>
-                              * When nbWorkers >= 1, triggers asynchronous mode when used with ZSTD_compressStream*() :
+                              * When nbWorkers >= 1, triggers asynchronous mode when invoking ZSTD_compressStream*() :
                               * ZSTD_compressStream*() consumes input and flush output if possible, but immediately gives back control to caller,
-                              * while compression work is performed in parallel, within worker threads.
+                              * while compression is performed in parallel, within worker thread(s).
                               * (note : a strong exception to this rule is when first invocation of ZSTD_compressStream2() sets ZSTD_e_end :
                               *  in which case, ZSTD_compressStream2() delegates to ZSTD_compress2(), which is always 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 */
+                              * Default value is `0`, aka "single-threaded mode" : no worker is spawned,
+                              * compression is performed inside Caller's thread, and all invocations are blocking */
     ZSTD_c_jobSize=401,      </b>/* Size of a compression job. This value is enforced only when nbWorkers >= 1.<b>
                               * Each compression job is completed in parallel, so this value can indirectly impact the nb of active threads.
                               * 0 means default, which is dynamically determined based on compression parameters.
@@ -346,6 +352,11 @@ size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
      * ZSTD_c_literalCompressionMode
      * ZSTD_c_targetCBlockSize
      * ZSTD_c_srcSizeHint
+     * ZSTD_c_enableDedicatedDictSearch
+     * ZSTD_c_stableInBuffer
+     * ZSTD_c_stableOutBuffer
+     * ZSTD_c_blockDelimiters
+     * ZSTD_c_validateSequences
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly;
      *        also, the enums values themselves are unstable and can still change.
@@ -356,7 +367,12 @@ size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
      ZSTD_c_experimentalParam4=1001,
      ZSTD_c_experimentalParam5=1002,
      ZSTD_c_experimentalParam6=1003,
-     ZSTD_c_experimentalParam7=1004
+     ZSTD_c_experimentalParam7=1004,
+     ZSTD_c_experimentalParam8=1005,
+     ZSTD_c_experimentalParam9=1006,
+     ZSTD_c_experimentalParam10=1007,
+     ZSTD_c_experimentalParam11=1008,
+     ZSTD_c_experimentalParam12=1009
 } ZSTD_cParameter;
 </b></pre><BR>
 <pre><b>typedef struct {
@@ -456,11 +472,13 @@ size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
      * At the time of this writing, they include :
      * ZSTD_d_format
      * ZSTD_d_stableOutBuffer
+     * ZSTD_d_forceIgnoreChecksum
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly
      */
      ZSTD_d_experimentalParam1=1000,
-     ZSTD_d_experimentalParam2=1001
+     ZSTD_d_experimentalParam2=1001,
+     ZSTD_d_experimentalParam3=1002
 
 } ZSTD_dParameter;
 </b></pre><BR>
@@ -591,8 +609,9 @@ size_t ZSTD_freeCStream(ZSTD_CStream* zcs);
   - Compression parameters cannot be changed once compression is started (save a list of exceptions in multi-threading mode)
   - output->pos must be <= dstCapacity, input->pos must be <= srcSize
   - output->pos and input->pos will be updated. They are guaranteed to remain below their respective limit.
+  - endOp must be a valid directive
   - When nbWorkers==0 (default), function is blocking : it completes its job before returning to caller.
-  - When nbWorkers>=1, function is non-blocking : it just acquires a copy of input, and distributes jobs to internal worker threads, flush whatever is available,
+  - When nbWorkers>=1, function is non-blocking : it copies a portion of input, distributes jobs to internal worker threads, flush to output whatever is available,
                                                   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 : if the first call requests a ZSTD_e_end directive and provides enough dstCapacity, the function delegates to ZSTD_compress2() which is always blocking.
@@ -895,21 +914,40 @@ size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 <BR></pre>
 
 <pre><b>typedef struct {
-    unsigned int matchPos; </b>/* Match pos in dst */<b>
-    </b>/* If seqDef.offset > 3, then this is seqDef.offset - 3<b>
-     * If seqDef.offset < 3, then this is the corresponding repeat offset
-     * But if seqDef.offset < 3 and litLength == 0, this is the
-     *   repeat offset before the corresponding repeat offset
-     * And if seqDef.offset == 3 and litLength == 0, this is the
-     *   most recent repeat offset - 1
-     */
-    unsigned int offset;
-    unsigned int litLength; </b>/* Literal length */<b>
-    unsigned int matchLength; </b>/* Match length */<b>
-    </b>/* 0 when seq not rep and seqDef.offset otherwise<b>
-     * when litLength == 0 this will be <= 4, otherwise <= 3 like normal
-     */
-    unsigned int rep;
+    unsigned int offset;      </b>/* The offset of the match. (NOT the same as the offset code)<b>
+                               * If offset == 0 and matchLength == 0, this sequence represents the last
+                               * literals in the block of litLength size.
+                               */
+
+    unsigned int litLength;   </b>/* Literal length of the sequence. */<b>
+    unsigned int matchLength; </b>/* Match length of the sequence. */<b>
+
+                              </b>/* Note: Users of this API may provide a sequence with matchLength == litLength == offset == 0.<b>
+                               * In this case, we will treat the sequence as a marker for a block boundary.
+                               */
+
+    unsigned int rep;         </b>/* Represents which repeat offset is represented by the field 'offset'.<b>
+                               * Ranges from [0, 3].
+                               *
+                               * Repeat offsets are essentially previous offsets from previous sequences sorted in
+                               * recency order. For more detail, see doc/zstd_compression_format.md
+                               *
+                               * If rep == 0, then 'offset' does not contain a repeat offset.
+                               * If rep > 0:
+                               *  If litLength != 0:
+                               *      rep == 1 --> offset == repeat_offset_1
+                               *      rep == 2 --> offset == repeat_offset_2
+                               *      rep == 3 --> offset == repeat_offset_3
+                               *  If litLength == 0:
+                               *      rep == 1 --> offset == repeat_offset_2
+                               *      rep == 2 --> offset == repeat_offset_3
+                               *      rep == 3 --> offset == repeat_offset_1 - 1
+                               *
+                               * Note: This field is optional. ZSTD_generateSequences() will calculate the value of
+                               * 'rep', but repeat offsets do not necessarily need to be calculated from an external
+                               * sequence provider's perspective. For example, ZSTD_compressSequences() does not
+                               * use this 'rep' field at all (as of now).
+                               */
 } ZSTD_Sequence;
 </b></pre><BR>
 <pre><b>typedef struct {
@@ -952,6 +990,12 @@ size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 } ZSTD_format_e;
 </b></pre><BR>
 <pre><b>typedef enum {
+    </b>/* Note: this enum controls ZSTD_d_forceIgnoreChecksum */<b>
+    ZSTD_d_validateChecksum = 0,
+    ZSTD_d_ignoreChecksum = 1
+} ZSTD_forceIgnoreChecksum_e;
+</b></pre><BR>
+<pre><b>typedef enum {
     </b>/* Note: this enum and the behavior it controls are effectively internal<b>
      * implementation details of the compressor. They are expected to continue
      * to evolve and should be considered only in the context of extremely
@@ -1045,12 +1089,69 @@ size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
            or an error code (if srcSize is too small) 
 </p></pre><BR>
 
-<pre><b>size_t ZSTD_getSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
-    size_t outSeqsSize, const void* src, size_t srcSize);
-</b><p> Extract sequences from the sequence store
+<pre><b>typedef enum {
+  ZSTD_sf_noBlockDelimiters = 0,         </b>/* Representation of ZSTD_Sequence has no block delimiters, sequences only */<b>
+  ZSTD_sf_explicitBlockDelimiters = 1    </b>/* Representation of ZSTD_Sequence contains explicit block delimiters */<b>
+} ZSTD_sequenceFormat_e;
+</b></pre><BR>
+<pre><b></b><p> Generate sequences using ZSTD_compress2, given a source buffer.
+
+ Each block will end with a dummy sequence
+ with offset == 0, matchLength == 0, and litLength == length of last literals.
+ litLength may be == 0, and if so, then the sequence of (of: 0 ml: 0 ll: 0)
+ simply acts as a block delimiter.
+
  zc can be used to insert custom compression params.
  This function invokes ZSTD_compress2
- @return : number of sequences extracted
+
+ The output of this function can be fed into ZSTD_compressSequences() with CCtx
+ setting of ZSTD_c_blockDelimiters as ZSTD_sf_explicitBlockDelimiters
+ @return : number of sequences generated
+ 
+</p></pre><BR>
+
+<pre><b>size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize);
+</b><p> Given an array of ZSTD_Sequence, remove all sequences that represent block delimiters/last literals
+ by merging them into into the literals of the next sequence.
+
+ As such, the final generated result has no explicit representation of block boundaries,
+ and the final last literals segment is not represented in the sequences.
+
+ The output of this function can be fed into ZSTD_compressSequences() with CCtx
+ setting of ZSTD_c_blockDelimiters as ZSTD_sf_noBlockDelimiters
+ @return : number of sequences left after merging
+ 
+</p></pre><BR>
+
+<pre><b>size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstSize,
+                      const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                      const void* src, size_t srcSize);
+</b><p> Compress an array of ZSTD_Sequence, generated from the original source buffer, into dst.
+ If a dictionary is included, then the cctx should reference the dict. (see: ZSTD_CCtx_refCDict(), ZSTD_CCtx_loadDictionary(), etc.)
+ The entire source is compressed into a single frame.
+
+ The compression behavior changes based on cctx params. In particular:
+    If ZSTD_c_blockDelimiters == ZSTD_sf_noBlockDelimiters, the array of ZSTD_Sequence is expected to contain
+    no block delimiters (defined in ZSTD_Sequence). Block boundaries are roughly determined based on
+    the block size derived from the cctx, and sequences may be split. This is the default setting.
+
+    If ZSTD_c_blockDelimiters == ZSTD_sf_explicitBlockDelimiters, the array of ZSTD_Sequence is expected to contain
+    block delimiters (defined in ZSTD_Sequence). Behavior is undefined if no block delimiters are provided.
+
+    If ZSTD_c_validateSequences == 0, this function will blindly accept the sequences provided. Invalid sequences cause undefined
+    behavior. If ZSTD_c_validateSequences == 1, then if sequence is invalid (see doc/zstd_compression_format.md for
+    specifics regarding offset/matchlength requirements) then the function will bail out and return an error.
+
+    In addition to the two adjustable experimental params, there are other important cctx params.
+    - ZSTD_c_minMatch MUST be set as less than or equal to the smallest match generated by the match finder. It has a minimum value of ZSTD_MINMATCH_MIN.
+    - ZSTD_c_compressionLevel accordingly adjusts the strength of the entropy coder, as it would in typical compression.
+    - ZSTD_c_windowLog affects offset validation: this function will return an error at higher debug levels if a provided offset
+      is larger than what the spec allows for a given window log and dictionary (if present). See: doc/zstd_compression_format.md
+
+ Note: Repcodes are, as of now, always re-calculated within this function, so ZSTD_Sequence::rep is unused.
+ Note 2: Once we integrate ability to ingest repcodes, the explicit block delims mode must respect those repcodes exactly,
+         and cannot emit an RLE block that disagrees with the repcode history
+ @return : final compressed size or a ZSTD error.
  
 </p></pre><BR>
 
@@ -1141,7 +1242,11 @@ ZSTD_CStream* ZSTD_initStaticCStream(void* workspace, size_t workspaceSize);
 <pre><b>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 };  </b>/**< this constant defers to stdlib's functions */<b>
+static
+#ifdef __GNUC__
+__attribute__((__unused__))
+#endif
+ZSTD_customMem const ZSTD_defaultCMem = { NULL, NULL, NULL };  </b>/**< this constant defers to stdlib's functions */<b>
 </b><p>  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.
@@ -1158,6 +1263,12 @@ static ZSTD_customMem const ZSTD_defaultCMem = { NULL, NULL, NULL };  </b>/**< t
   note: equivalent to ZSTD_createCDict_advanced(), with dictLoadMethod==ZSTD_dlm_byRef 
 </p></pre><BR>
 
+<pre><b>unsigned ZSTD_getDictID_fromCDict(const ZSTD_CDict* cdict);
+</b><p>  Provides the dictID of the dictionary loaded into `cdict`.
+  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. 
+</p></pre><BR>
+
 <pre><b>ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
 </b><p> @return ZSTD_compressionParameters structure for a selected compression level and estimated srcSize.
  `estimatedSrcSize` value is optional, select 0 if not known 
@@ -1264,8 +1375,10 @@ size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);
 <pre><b>size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, int value);
 </b><p>  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().
- @result : 0, or an error code (which can be tested with ZSTD_isError()).
+  Parameters must be applied to a ZSTD_CCtx using
+  ZSTD_CCtx_setParametersUsingCCtxParams().
+ @result : a code representing success or failure (which can be tested with
+           ZSTD_isError()).
  
 </p></pre><BR>
 
@@ -1342,6 +1455,13 @@ size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);
  
 </p></pre><BR>
 
+<pre><b>size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value);
+</b><p>  Get the requested decompression parameter value, selected by enum ZSTD_dParameter,
+  and store it into int* value.
+ @return : 0, or an error code (which can be tested with ZSTD_isError()).
+ 
+</p></pre><BR>
+
 <pre><b>size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
 </b><p>  Instruct the decoder context about what kind of data to decode next.
   This instruction is mandatory to decode data without a fully-formed header,
@@ -1365,24 +1485,29 @@ size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);
   redundant functions will be deprecated, and then at some point removed.
 <BR></pre>
 
-<h3>Advanced Streaming compression functions</h3><pre></pre><b><pre></b>/**! ZSTD_initCStream_srcSize() :<b>
- * This function is deprecated, and equivalent to:
- *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
- *     ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
- *     ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
- *     ZSTD_CCtx_setPledgedSrcSize(zcs, 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.
- * Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
- */
-size_t
+<h3>Advanced Streaming compression functions</h3><pre></pre><b><pre></pre></b><BR>
+<pre><b>size_t
 ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
-                         int compressionLevel,
-                         unsigned long long pledgedSrcSize);
-</pre></b><BR>
-<a name="Chapter20"></a><h2>! ZSTD_initCStream_usingDict() :</h2><pre> This function is deprecated, and is equivalent to:
+             int compressionLevel,
+             unsigned long long pledgedSrcSize);
+</b><p> This function is deprecated, and equivalent to:
+     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
+     ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
+     ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
+     ZSTD_CCtx_setPledgedSrcSize(zcs, 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.
+ Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
+ 
+</p></pre><BR>
+
+<pre><b>size_t
+ZSTD_initCStream_usingDict(ZSTD_CStream* zcs,
+         const void* dict, size_t dictSize,
+               int compressionLevel);
+</b><p> This function is deprecated, and is equivalent to:
      ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
      ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
      ZSTD_CCtx_loadDictionary(zcs, dict, dictSize);
@@ -1393,9 +1518,14 @@ ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
  it begins with ZSTD_MAGIC_DICTIONARY, else as raw content) and ZSTD_dlm_byCopy.
  Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  
-<BR></pre>
+</p></pre><BR>
 
-<a name="Chapter21"></a><h2>! ZSTD_initCStream_advanced() :</h2><pre> This function is deprecated, and is approximately equivalent to:
+<pre><b>size_t
+ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
+        const void* dict, size_t dictSize,
+              ZSTD_parameters params,
+              unsigned long long pledgedSrcSize);
+</b><p> This function is deprecated, and is approximately equivalent to:
      ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
      // Pseudocode: Set each zstd parameter and leave the rest as-is.
      for ((param, value) : params) {
@@ -1409,18 +1539,24 @@ ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
  If srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN.
  Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  
-<BR></pre>
+</p></pre><BR>
 
-<a name="Chapter22"></a><h2>! ZSTD_initCStream_usingCDict() :</h2><pre> This function is deprecated, and equivalent to:
+<pre><b>size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);
+</b><p> This function is deprecated, and equivalent to:
      ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
      ZSTD_CCtx_refCDict(zcs, cdict);
 
  note : cdict will just be referenced, and must outlive compression session
  Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  
-<BR></pre>
+</p></pre><BR>
 
-<a name="Chapter23"></a><h2>! ZSTD_initCStream_usingCDict_advanced() :</h2><pre>   This function is DEPRECATED, and is approximately equivalent to:
+<pre><b>size_t
+ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
+                   const ZSTD_CDict* cdict,
+                         ZSTD_frameParameters fParams,
+                         unsigned long long pledgedSrcSize);
+</b><p>   This function is DEPRECATED, and is approximately equivalent to:
      ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
      // Pseudocode: Set each zstd frame parameter and leave the rest as-is.
      for ((fParam, value) : fParams) {
@@ -1434,7 +1570,7 @@ ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
  value ZSTD_CONTENTSIZE_UNKNOWN.
  Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  
-<BR></pre>
+</p></pre><BR>
 
 <pre><b>size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
 </b><p> This function is deprecated, and is equivalent to:
@@ -1477,42 +1613,44 @@ ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
  
 </p></pre><BR>
 
-<h3>Advanced Streaming decompression functions</h3><pre></pre><b><pre></b>/**<b>
- * This function is deprecated, and is equivalent to:
- *
- *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
- *     ZSTD_DCtx_loadDictionary(zds, dict, dictSize);
- *
- * note: no dictionary will be used if dict == NULL or dictSize < 8
- * Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
- */
-size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
-</pre></b><BR>
-<a name="Chapter24"></a><h2>This function is deprecated, and is equivalent to:</h2><pre>
+<h3>Advanced Streaming decompression functions</h3><pre></pre><b><pre></pre></b><BR>
+<pre><b>size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
+</b><p>
+     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
+     ZSTD_DCtx_loadDictionary(zds, dict, dictSize);
+
+ note: no dictionary will be used if dict == NULL or dictSize < 8
+ Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
+ 
+</p></pre><BR>
+
+<pre><b>size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);
+</b><p>
      ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
      ZSTD_DCtx_refDDict(zds, ddict);
 
  note : ddict is referenced, it must outlive decompression session
  Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  
-<BR></pre>
+</p></pre><BR>
 
-<a name="Chapter25"></a><h2>This function is deprecated, and is equivalent to:</h2><pre>
+<pre><b>size_t ZSTD_resetDStream(ZSTD_DStream* zds);
+</b><p>
      ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
 
  re-use decompression parameters from previous init; saves dictionary loading
  Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  
-<BR></pre>
+</p></pre><BR>
 
-<a name="Chapter26"></a><h2>Buffer-less and synchronous inner streaming functions</h2><pre>
+<a name="Chapter20"></a><h2>Buffer-less and synchronous inner streaming functions</h2><pre>
   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 several restrictions, documented below.
   Prefer normal streaming API for an easier experience.
  
 <BR></pre>
 
-<a name="Chapter27"></a><h2>Buffer-less streaming compression (synchronous mode)</h2><pre>
+<a name="Chapter21"></a><h2>Buffer-less streaming compression (synchronous mode)</h2><pre>
   A ZSTD_CCtx object is required to track streaming operations.
   Use ZSTD_createCCtx() / ZSTD_freeCCtx() to manage resource.
   ZSTD_CCtx object can be re-used multiple times within successive compression operations.
@@ -1548,7 +1686,7 @@ size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
 size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict, ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize);   </b>/* compression parameters are already set within cdict. pledgedSrcSize must be correct. If srcSize is not known, use macro ZSTD_CONTENTSIZE_UNKNOWN */<b>
 size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); </b>/**<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */<b>
 </pre></b><BR>
-<a name="Chapter28"></a><h2>Buffer-less streaming decompression (synchronous mode)</h2><pre>
+<a name="Chapter22"></a><h2>Buffer-less streaming decompression (synchronous mode)</h2><pre>
   A ZSTD_DCtx object is required to track streaming operations.
   Use ZSTD_createDCtx() / ZSTD_freeDCtx() to manage it.
   A ZSTD_DCtx object can be re-used multiple times.
@@ -1644,7 +1782,7 @@ size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long
 
 <pre><b>typedef enum { ZSTDnit_frameHeader, ZSTDnit_blockHeader, ZSTDnit_block, ZSTDnit_lastBlock, ZSTDnit_checksum, ZSTDnit_skippableFrame } ZSTD_nextInputType_e;
 </b></pre><BR>
-<a name="Chapter29"></a><h2>Block level API</h2><pre></pre>
+<a name="Chapter23"></a><h2>Block level API</h2><pre></pre>
 
 <pre><b></b><p>    Frame metadata cost is typically ~12 bytes, which can be non-negligible for very small blocks (< 100 bytes).
     But users will have to take in charge needed metadata to regenerate data, such as compressed and content sizes.
diff --git a/examples/Makefile b/examples/Makefile
index 1ae6bce83d0c..f5e3274b1aab 100644
--- a/examples/Makefile
+++ b/examples/Makefile
@@ -8,15 +8,15 @@
 # You may select, at your option, one of the above-listed licenses.
 # ################################################################
 
-# This Makefile presumes libzstd is installed, using `sudo make install`
+LIBDIR =../lib
+CPPFLAGS += -I$(LIBDIR)
+LIB = $(LIBDIR)/libzstd.a
 
-CPPFLAGS += -I../lib
-LIB = ../lib/libzstd.a
-
-.PHONY: default all clean test
 
+.PHONY: default
 default: all
 
+.PHONY: all
 all: simple_compression simple_decompression \
 	multiple_simple_compression\
 	dictionary_compression dictionary_decompression \
@@ -24,37 +24,39 @@ all: simple_compression simple_decompression \
 	multiple_streaming_compression streaming_memory_usage
 
 $(LIB) :
-	$(MAKE) -C ../lib libzstd.a
+	$(MAKE) -C $(LIBDIR) libzstd.a
+
+simple_compression.o: common.h
+simple_compression : $(LIB)
 
-simple_compression : simple_compression.c common.h $(LIB)
-	$(CC) $(CPPFLAGS) $(CFLAGS) $< $(LIB) $(LDFLAGS) -o $@
+simple_decompression.o: common.h
+simple_decompression : $(LIB)
 
-simple_decompression : simple_decompression.c common.h $(LIB)
-	$(CC) $(CPPFLAGS) $(CFLAGS) $< $(LIB) $(LDFLAGS) -o $@
+multiple_simple_compression.o: common.h
+multiple_simple_compression : $(LIB)
 
-multiple_simple_compression : multiple_simple_compression.c common.h $(LIB)
-	$(CC) $(CPPFLAGS) $(CFLAGS) $< $(LIB) $(LDFLAGS) -o $@
+dictionary_compression.o: common.h
+dictionary_compression : $(LIB)
 
-dictionary_compression : dictionary_compression.c common.h $(LIB)
-	$(CC) $(CPPFLAGS) $(CFLAGS) $< $(LIB) $(LDFLAGS) -o $@
+dictionary_decompression.o: common.h
+dictionary_decompression : $(LIB)
 
-dictionary_decompression : dictionary_decompression.c common.h $(LIB)
-	$(CC) $(CPPFLAGS) $(CFLAGS) $< $(LIB) $(LDFLAGS) -o $@
+streaming_compression.o: common.h
+streaming_compression : $(LIB)
 
-streaming_compression : streaming_compression.c common.h $(LIB)
-	$(CC) $(CPPFLAGS) $(CFLAGS) $< $(LIB) $(LDFLAGS) -o $@
+multiple_streaming_compression.o: common.h
+multiple_streaming_compression : $(LIB)
 
-multiple_streaming_compression : multiple_streaming_compression.c common.h $(LIB)
-	$(CC) $(CPPFLAGS) $(CFLAGS) $< $(LIB) $(LDFLAGS) -o $@
+streaming_decompression.o: common.h
+streaming_decompression : $(LIB)
 
-streaming_decompression : streaming_decompression.c common.h $(LIB)
-	$(CC) $(CPPFLAGS) $(CFLAGS) $< $(LIB) $(LDFLAGS) -o $@
+streaming_memory_usage.o: common.h
+streaming_memory_usage : $(LIB)
 
-streaming_memory_usage : streaming_memory_usage.c $(LIB)
-	$(CC) $(CPPFLAGS) $(CFLAGS) $< $(LIB) $(LDFLAGS) -o $@
 
+.PHONY:clean
 clean:
-	@rm -f core *.o tmp* result* *.zst \
+	@$(RM) core *.o tmp* result* *.zst \
         simple_compression simple_decompression \
         multiple_simple_compression \
         dictionary_compression dictionary_decompression \
@@ -62,6 +64,7 @@ clean:
         multiple_streaming_compression streaming_memory_usage
 	@echo Cleaning completed
 
+.PHONY:test
 test: all
 	cp README.md tmp
 	cp Makefile tmp2
diff --git a/examples/streaming_compression.c b/examples/streaming_compression.c
index f0f1065b1d27..045437f2873d 100644
--- a/examples/streaming_compression.c
+++ b/examples/streaming_compression.c
@@ -39,6 +39,7 @@ static void compressFile_orDie(const char* fname, const char* outName, int cLeve
      */
     CHECK_ZSTD( ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, cLevel) );
     CHECK_ZSTD( ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1) );
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 4);
 
     /* This loop read from the input file, compresses that entire chunk,
      * and writes all output produced to the output file.
diff --git a/examples/streaming_compression_thread_pool.c b/examples/streaming_compression_thread_pool.c
new file mode 100644
index 000000000000..22c3b2efacc9
--- /dev/null
+++ b/examples/streaming_compression_thread_pool.c
@@ -0,0 +1,178 @@
+/*
+ * Copyright (c) 2020, Martin Liska, SUSE, 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 <stdio.h>     // printf
+#include <stdlib.h>    // free
+#include <string.h>    // memset, strcat, strlen
+#include <zstd.h>      // presumes zstd library is installed
+#include "common.h"    // Helper functions, CHECK(), and CHECK_ZSTD()
+#include <pthread.h>
+
+typedef struct compress_args
+{
+  const char *fname;
+  char *outName;
+  int cLevel;
+#if defined(ZSTD_STATIC_LINKING_ONLY)
+  ZSTD_threadPool *pool;
+#endif
+} compress_args_t;
+
+static void *compressFile_orDie(void *data)
+{
+    compress_args_t *args = (compress_args_t *)data;
+    fprintf (stderr, "Starting compression of %s with level %d\n", args->fname, args->cLevel);
+    /* Open the input and output files. */
+    FILE* const fin  = fopen_orDie(args->fname, "rb");
+    FILE* const fout = fopen_orDie(args->outName, "wb");
+    /* Create the input and output buffers.
+     * They may be any size, but we recommend using these functions to size them.
+     * Performance will only suffer significantly for very tiny buffers.
+     */
+    size_t const buffInSize = ZSTD_CStreamInSize();
+    void*  const buffIn  = malloc_orDie(buffInSize);
+    size_t const buffOutSize = ZSTD_CStreamOutSize();
+    void*  const buffOut = malloc_orDie(buffOutSize);
+
+    /* Create the context. */
+    ZSTD_CCtx* const cctx = ZSTD_createCCtx();
+    CHECK(cctx != NULL, "ZSTD_createCCtx() failed!");
+
+#if defined(ZSTD_STATIC_LINKING_ONLY)
+    size_t r = ZSTD_CCtx_refThreadPool(cctx, args->pool);
+    CHECK(r == 0, "ZSTD_CCtx_refThreadPool failed!");
+#endif
+
+    /* Set any parameters you want.
+     * Here we set the compression level, and enable the checksum.
+     */
+    CHECK_ZSTD( ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, args->cLevel) );
+    CHECK_ZSTD( ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1) );
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 16);
+
+    /* This loop read from the input file, compresses that entire chunk,
+     * and writes all output produced to the output file.
+     */
+    size_t const toRead = buffInSize;
+    for (;;) {
+        size_t read = fread_orDie(buffIn, toRead, fin);
+        /* Select the flush mode.
+         * If the read may not be finished (read == toRead) we use
+         * ZSTD_e_continue. If this is the last chunk, we use ZSTD_e_end.
+         * Zstd optimizes the case where the first flush mode is ZSTD_e_end,
+         * since it knows it is compressing the entire source in one pass.
+         */
+        int const lastChunk = (read < toRead);
+        ZSTD_EndDirective const mode = lastChunk ? ZSTD_e_end : ZSTD_e_continue;
+        /* Set the input buffer to what we just read.
+         * We compress until the input buffer is empty, each time flushing the
+         * output.
+         */
+        ZSTD_inBuffer input = { buffIn, read, 0 };
+        int finished;
+        do {
+            /* Compress into the output buffer and write all of the output to
+             * the file so we can reuse the buffer next iteration.
+             */
+            ZSTD_outBuffer output = { buffOut, buffOutSize, 0 };
+            size_t const remaining = ZSTD_compressStream2(cctx, &output , &input, mode);
+            CHECK_ZSTD(remaining);
+            fwrite_orDie(buffOut, output.pos, fout);
+            /* If we're on the last chunk we're finished when zstd returns 0,
+             * which means its consumed all the input AND finished the frame.
+             * Otherwise, we're finished when we've consumed all the input.
+             */
+            finished = lastChunk ? (remaining == 0) : (input.pos == input.size);
+        } while (!finished);
+        CHECK(input.pos == input.size,
+              "Impossible: zstd only returns 0 when the input is completely consumed!");
+
+        if (lastChunk) {
+            break;
+        }
+    }
+
+    fprintf (stderr, "Finishing compression of %s\n", args->outName);
+
+    ZSTD_freeCCtx(cctx);
+    fclose_orDie(fout);
+    fclose_orDie(fin);
+    free(buffIn);
+    free(buffOut);
+    free(args->outName);
+
+    return NULL;
+}
+
+
+static char* createOutFilename_orDie(const char* filename)
+{
+    size_t const inL = strlen(filename);
+    size_t const outL = inL + 5;
+    void* const outSpace = malloc_orDie(outL);
+    memset(outSpace, 0, outL);
+    strcat(outSpace, filename);
+    strcat(outSpace, ".zst");
+    return (char*)outSpace;
+}
+
+int main(int argc, const char** argv)
+{
+    const char* const exeName = argv[0];
+
+    if (argc<=3) {
+        printf("wrong arguments\n");
+        printf("usage:\n");
+        printf("%s POOL_SIZE LEVEL FILES\n", exeName);
+        return 1;
+    }
+
+    int pool_size = atoi (argv[1]);
+    CHECK(pool_size != 0, "can't parse POOL_SIZE!");
+
+    int level = atoi (argv[2]);
+    CHECK(level != 0, "can't parse LEVEL!");
+
+    argc -= 3;
+    argv += 3;
+
+#if defined(ZSTD_STATIC_LINKING_ONLY)
+    ZSTD_threadPool *pool = ZSTD_createThreadPool (pool_size);
+    CHECK(pool != NULL, "ZSTD_createThreadPool() failed!");
+    fprintf (stderr, "Using shared thread pool of size %d\n", pool_size);
+#else
+    fprintf (stderr, "All threads use its own thread pool\n");
+#endif
+
+    pthread_t *threads = malloc_orDie(argc * sizeof(pthread_t));
+    compress_args_t *args = malloc_orDie(argc * sizeof(compress_args_t));
+
+    for (unsigned i = 0; i < argc; i++)
+    {
+      args[i].fname = argv[i];
+      args[i].outName = createOutFilename_orDie(args[i].fname);
+      args[i].cLevel = level;
+#if defined(ZSTD_STATIC_LINKING_ONLY)
+      args[i].pool = pool;
+#endif
+
+      pthread_create (&threads[i], NULL, compressFile_orDie, &args[i]);
+    }
+
+    for (unsigned i = 0; i < argc; i++)
+      pthread_join (threads[i], NULL);
+
+#if defined(ZSTD_STATIC_LINKING_ONLY)
+    ZSTD_freeThreadPool (pool);
+#endif
+
+    return 0;
+}
diff --git a/lib/Makefile b/lib/Makefile
index 7c6dff02468a..869d76630e68 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -8,15 +8,16 @@
 # You may select, at your option, one of the above-listed licenses.
 # ################################################################
 
-Q = $(if $(filter 1,$(V) $(VERBOSE)),,@)
+.PHONY: default
+default: lib-release
 
-# When cross-compiling from linux to windows, you might
-# need to specify this as "Windows." Fedora build fails
-# without it.
-#
-# Note: mingw-w64 build from linux to windows does not
-# fail on other tested distros (ubuntu, debian) even
-# without manually specifying the TARGET_SYSTEM.
+# define silent mode as default (verbose mode with V=1 or VERBOSE=1)
+$(V)$(VERBOSE).SILENT:
+
+# When cross-compiling from linux to windows,
+# one might need to specify TARGET_SYSTEM as "Windows."
+# Building from Fedora fails without it.
+# (but Ubuntu and Debian don't need to set anything)
 TARGET_SYSTEM ?= $(OS)
 
 # Version numbers
@@ -31,24 +32,49 @@ LIBVER := $(shell echo $(LIBVER_SCRIPT))
 VERSION?= $(LIBVER)
 CCVER := $(shell $(CC) --version)
 
-CPPFLAGS+= -DXXH_NAMESPACE=ZSTD_
+# ZSTD_LIB_MINIFY is a helper variable that
+# configures a bunch of other variables to space-optimized defaults.
+ZSTD_LIB_MINIFY ?= 0
+ifneq ($(ZSTD_LIB_MINIFY), 0)
+  HAVE_CC_OZ ?= $(shell echo "" | $(CC) -Oz -x c -c - -o /dev/null 2> /dev/null && echo 1 || echo 0)
+  ZSTD_LEGACY_SUPPORT ?= 0
+  ZSTD_LIB_DEPRECATED ?= 0
+  HUF_FORCE_DECOMPRESS_X1 ?= 1
+  ZSTD_FORCE_DECOMPRESS_SHORT ?= 1
+  ZSTD_NO_INLINE ?= 1
+  ZSTD_STRIP_ERROR_STRINGS ?= 1
+ifneq ($(HAVE_CC_OZ), 0)
+    # Some compilers (clang) support an even more space-optimized setting.
+    CFLAGS += -Oz
+else
+    CFLAGS += -Os
+endif
+  CFLAGS += -fno-stack-protector -fomit-frame-pointer -fno-ident \
+            -DDYNAMIC_BMI2=0 -DNDEBUG
+else
+  CFLAGS += -O3
+endif
+
+DEBUGLEVEL ?= 0
+CPPFLAGS += -DXXH_NAMESPACE=ZSTD_ -DDEBUGLEVEL=$(DEBUGLEVEL)
 ifeq ($(TARGET_SYSTEM),Windows_NT)   # MinGW assumed
-CPPFLAGS   += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
+  CPPFLAGS += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
 endif
 DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
             -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
             -Wstrict-prototypes -Wundef -Wpointer-arith \
             -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
             -Wredundant-decls -Wmissing-prototypes -Wc++-compat
-CFLAGS  += $(DEBUGFLAGS) $(MOREFLAGS)
-FLAGS    = $(CPPFLAGS) $(CFLAGS)
+CFLAGS   += $(DEBUGFLAGS) $(MOREFLAGS)
+FLAGS     = $(CPPFLAGS) $(CFLAGS)
 
 HAVE_COLORNEVER = $(shell echo a | grep --color=never a > /dev/null 2> /dev/null && echo 1 || echo 0)
 GREP_OPTIONS ?=
 ifeq ($HAVE_COLORNEVER, 1)
-GREP_OPTIONS += --color=never
+  GREP_OPTIONS += --color=never
 endif
 GREP = grep $(GREP_OPTIONS)
+SED_ERE_OPT ?= -E
 
 ZSTDCOMMON_FILES := $(sort $(wildcard common/*.c))
 ZSTDCOMP_FILES := $(sort $(wildcard compress/*.c))
@@ -58,30 +84,7 @@ ZDEPR_FILES := $(sort $(wildcard deprecated/*.c))
 ZSTD_FILES := $(ZSTDCOMMON_FILES)
 
 ifeq ($(findstring GCC,$(CCVER)),GCC)
-decompress/zstd_decompress_block.o :	CFLAGS+=-fno-tree-vectorize
-endif
-
-# This is a helper variable that configures a bunch of other variables to new,
-# space-optimized defaults.
-ZSTD_LIB_MINIFY ?= 0
-ifneq ($(ZSTD_LIB_MINIFY), 0)
-	HAVE_CC_OZ ?= $(shell echo "" | $(CC) -Oz -x c -c - -o /dev/null 2> /dev/null && echo 1 || echo 0)
-	ZSTD_LEGACY_SUPPORT ?= 0
-	ZSTD_LIB_DEPRECATED ?= 0
-	HUF_FORCE_DECOMPRESS_X1 ?= 1
-	ZSTD_FORCE_DECOMPRESS_SHORT ?= 1
-	ZSTD_NO_INLINE ?= 1
-	ZSTD_STRIP_ERROR_STRINGS ?= 1
-	ifneq ($(HAVE_CC_OZ), 0)
-		# Some compilers (clang) support an even more space-optimized setting.
-		CFLAGS += -Oz
-	else
-		CFLAGS += -Os
-	endif
-	CFLAGS += -fno-stack-protector -fomit-frame-pointer -fno-ident \
-	          -DDYNAMIC_BMI2=0 -DNDEBUG
-else
-	CFLAGS += -O3
+decompress/zstd_decompress_block.o : CFLAGS+=-fno-tree-vectorize
 endif
 
 # Modules
@@ -103,116 +106,187 @@ ZSTD_NO_INLINE ?= 0
 ZSTD_STRIP_ERROR_STRINGS ?= 0
 
 ifeq ($(ZSTD_LIB_COMPRESSION), 0)
-	ZSTD_LIB_DICTBUILDER = 0
-	ZSTD_LIB_DEPRECATED = 0
+  ZSTD_LIB_DICTBUILDER = 0
+  ZSTD_LIB_DEPRECATED = 0
 endif
 
 ifeq ($(ZSTD_LIB_DECOMPRESSION), 0)
-	ZSTD_LEGACY_SUPPORT = 0
-	ZSTD_LIB_DEPRECATED = 0
+  ZSTD_LEGACY_SUPPORT = 0
+  ZSTD_LIB_DEPRECATED = 0
 endif
 
 ifneq ($(ZSTD_LIB_COMPRESSION), 0)
-	ZSTD_FILES += $(ZSTDCOMP_FILES)
+  ZSTD_FILES += $(ZSTDCOMP_FILES)
 endif
 
 ifneq ($(ZSTD_LIB_DECOMPRESSION), 0)
-	ZSTD_FILES += $(ZSTDDECOMP_FILES)
+  ZSTD_FILES += $(ZSTDDECOMP_FILES)
 endif
 
 ifneq ($(ZSTD_LIB_DEPRECATED), 0)
-	ZSTD_FILES += $(ZDEPR_FILES)
+  ZSTD_FILES += $(ZDEPR_FILES)
 endif
 
 ifneq ($(ZSTD_LIB_DICTBUILDER), 0)
-	ZSTD_FILES += $(ZDICT_FILES)
+  ZSTD_FILES += $(ZDICT_FILES)
 endif
 
 ifneq ($(HUF_FORCE_DECOMPRESS_X1), 0)
-	CFLAGS += -DHUF_FORCE_DECOMPRESS_X1
+  CFLAGS += -DHUF_FORCE_DECOMPRESS_X1
 endif
 
 ifneq ($(HUF_FORCE_DECOMPRESS_X2), 0)
-	CFLAGS += -DHUF_FORCE_DECOMPRESS_X2
+  CFLAGS += -DHUF_FORCE_DECOMPRESS_X2
 endif
 
 ifneq ($(ZSTD_FORCE_DECOMPRESS_SHORT), 0)
-	CFLAGS += -DZSTD_FORCE_DECOMPRESS_SHORT
+  CFLAGS += -DZSTD_FORCE_DECOMPRESS_SHORT
 endif
 
 ifneq ($(ZSTD_FORCE_DECOMPRESS_LONG), 0)
-	CFLAGS += -DZSTD_FORCE_DECOMPRESS_LONG
+  CFLAGS += -DZSTD_FORCE_DECOMPRESS_LONG
 endif
 
 ifneq ($(ZSTD_NO_INLINE), 0)
-	CFLAGS += -DZSTD_NO_INLINE
+  CFLAGS += -DZSTD_NO_INLINE
 endif
 
 ifneq ($(ZSTD_STRIP_ERROR_STRINGS), 0)
-	CFLAGS += -DZSTD_STRIP_ERROR_STRINGS
+  CFLAGS += -DZSTD_STRIP_ERROR_STRINGS
 endif
 
 ifneq ($(ZSTD_LEGACY_MULTITHREADED_API), 0)
-	CFLAGS += -DZSTD_LEGACY_MULTITHREADED_API
+  CFLAGS += -DZSTD_LEGACY_MULTITHREADED_API
 endif
 
 ifneq ($(ZSTD_LEGACY_SUPPORT), 0)
 ifeq ($(shell test $(ZSTD_LEGACY_SUPPORT) -lt 8; echo $$?), 0)
-	ZSTD_FILES += $(shell ls legacy/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
+  ZSTD_FILES += $(shell ls legacy/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
 endif
 endif
 CPPFLAGS  += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
 
-ZSTD_OBJ   := $(patsubst %.c,%.o,$(ZSTD_FILES))
+ZSTD_LOCAL_SRC := $(notdir $(ZSTD_FILES))
+ZSTD_LOCAL_OBJ := $(ZSTD_LOCAL_SRC:.c=.o)
+
+ZSTD_SUBDIR := common compress decompress dictBuilder legacy deprecated
+vpath %.c $(ZSTD_SUBDIR)
+
+UNAME := $(shell uname)
+
+ifndef BUILD_DIR
+ifeq ($(UNAME), Darwin)
+  HASH ?= md5
+else ifeq ($(UNAME), FreeBSD)
+  HASH ?= gmd5sum
+else ifeq ($(UNAME), OpenBSD)
+  HASH ?= md5
+endif
+HASH ?= md5sum
+
+HASH_DIR = conf_$(shell echo $(CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) $(ZSTD_FILES) | $(HASH) | cut -f 1 -d " " )
+HAVE_HASH :=$(shell echo 1 | $(HASH) > /dev/null && echo 1 || echo 0)
+ifeq ($(HAVE_HASH),0)
+  $(info warning : could not find HASH ($(HASH)), needed to differentiate builds using different flags)
+  BUILD_DIR := obj/generic_noconf
+endif
+endif # BUILD_DIR
+
 
 # macOS linker doesn't support -soname, and use different extension
 # see : https://developer.apple.com/library/mac/documentation/DeveloperTools/Conceptual/DynamicLibraries/100-Articles/DynamicLibraryDesignGuidelines.html
-ifeq ($(shell uname), Darwin)
-	SHARED_EXT = dylib
-	SHARED_EXT_MAJOR = $(LIBVER_MAJOR).$(SHARED_EXT)
-	SHARED_EXT_VER = $(LIBVER).$(SHARED_EXT)
-	SONAME_FLAGS = -install_name $(LIBDIR)/libzstd.$(SHARED_EXT_MAJOR) -compatibility_version $(LIBVER_MAJOR) -current_version $(LIBVER)
+ifeq ($(UNAME), Darwin)
+  SHARED_EXT = dylib
+  SHARED_EXT_MAJOR = $(LIBVER_MAJOR).$(SHARED_EXT)
+  SHARED_EXT_VER = $(LIBVER).$(SHARED_EXT)
+  SONAME_FLAGS = -install_name $(LIBDIR)/libzstd.$(SHARED_EXT_MAJOR) -compatibility_version $(LIBVER_MAJOR) -current_version $(LIBVER)
 else
-	SONAME_FLAGS = -Wl,-soname=libzstd.$(SHARED_EXT).$(LIBVER_MAJOR)
-	SHARED_EXT = so
-	SHARED_EXT_MAJOR = $(SHARED_EXT).$(LIBVER_MAJOR)
-	SHARED_EXT_VER = $(SHARED_EXT).$(LIBVER)
+  SONAME_FLAGS = -Wl,-soname=libzstd.$(SHARED_EXT).$(LIBVER_MAJOR)
+  SHARED_EXT = so
+  SHARED_EXT_MAJOR = $(SHARED_EXT).$(LIBVER_MAJOR)
+  SHARED_EXT_VER = $(SHARED_EXT).$(LIBVER)
 endif
 
+SET_CACHE_DIRECTORY = \
+	$(MAKE) --no-print-directory $@ \
+    BUILD_DIR=obj/$(HASH_DIR) \
+    CPPFLAGS="$(CPPFLAGS)" \
+    CFLAGS="$(CFLAGS)" \
+    LDFLAGS="$(LDFLAGS)"
 
-.PHONY: default lib-all all clean install uninstall
 
-default: lib-release
+.PHONY: lib-all all clean install uninstall
 
 # alias
 lib-all: all
 
 all: lib
 
-libzstd.a: ARFLAGS = rcs
-libzstd.a: $(ZSTD_OBJ)
+.PHONY: libzstd.a  # must be run every time
+
+ifndef BUILD_DIR
+# determine BUILD_DIR from compilation flags
+
+libzstd.a:
+	$(SET_CACHE_DIRECTORY)
+
+else
+# BUILD_DIR is defined
+
+ZSTD_STATLIB_DIR := $(BUILD_DIR)/static
+ZSTD_STATLIB := $(ZSTD_STATLIB_DIR)/libzstd.a
+ZSTD_STATLIB_OBJ := $(addprefix $(ZSTD_STATLIB_DIR)/,$(ZSTD_LOCAL_OBJ))
+$(ZSTD_STATLIB): ARFLAGS = rcs
+$(ZSTD_STATLIB): | $(ZSTD_STATLIB_DIR)
+$(ZSTD_STATLIB): $(ZSTD_STATLIB_OBJ)
 	@echo compiling static library
-	$(Q)$(AR) $(ARFLAGS) $@ $^
+	$(AR) $(ARFLAGS) $@ $^
+
+libzstd.a: $(ZSTD_STATLIB)
+	cp -f $< $@
+
+endif
 
 ifneq (,$(filter Windows%,$(TARGET_SYSTEM)))
 
-LIBZSTD = dll\libzstd.dll
+LIBZSTD = dll/libzstd.dll
 $(LIBZSTD): $(ZSTD_FILES)
 	@echo compiling dynamic library $(LIBVER)
-	$(CC) $(FLAGS) -DZSTD_DLL_EXPORT=1 -Wl,--out-implib,dll\libzstd.dll.a -shared $^ -o $@
+	$(CC) $(FLAGS) -DZSTD_DLL_EXPORT=1 -Wl,--out-implib,dll/libzstd.dll.a -shared $^ -o $@
 
-else
+else  # not Windows
 
 LIBZSTD = libzstd.$(SHARED_EXT_VER)
-$(LIBZSTD): LDFLAGS += -shared -fPIC -fvisibility=hidden
-$(LIBZSTD): $(ZSTD_FILES)
+.PHONY: $(LIBZSTD)  # must be run every time
+$(LIBZSTD): CFLAGS += -fPIC
+$(LIBZSTD): LDFLAGS += -shared -fvisibility=hidden
+
+ifndef BUILD_DIR
+# determine BUILD_DIR from compilation flags
+
+$(LIBZSTD):
+	$(SET_CACHE_DIRECTORY)
+
+else
+# BUILD_DIR is defined
+
+ZSTD_DYNLIB_DIR := $(BUILD_DIR)/dynamic
+ZSTD_DYNLIB := $(ZSTD_DYNLIB_DIR)/$(LIBZSTD)
+ZSTD_DYNLIB_OBJ := $(addprefix $(ZSTD_DYNLIB_DIR)/,$(ZSTD_LOCAL_OBJ))
+
+$(ZSTD_DYNLIB): | $(ZSTD_DYNLIB_DIR)
+$(ZSTD_DYNLIB): $(ZSTD_DYNLIB_OBJ)
 	@echo compiling dynamic library $(LIBVER)
-	$(Q)$(CC) $(FLAGS) $^ $(LDFLAGS) $(SONAME_FLAGS) -o $@
+	$(CC) $(FLAGS) $^ $(LDFLAGS) $(SONAME_FLAGS) -o $@
 	@echo creating versioned links
-	$(Q)ln -sf $@ libzstd.$(SHARED_EXT_MAJOR)
-	$(Q)ln -sf $@ libzstd.$(SHARED_EXT)
+	ln -sf $@ libzstd.$(SHARED_EXT_MAJOR)
+	ln -sf $@ libzstd.$(SHARED_EXT)
 
-endif
+$(LIBZSTD): $(ZSTD_DYNLIB)
+	cp -f $< $@
+
+endif  # ifndef BUILD_DIR
+endif  # if windows
 
 .PHONY: libzstd
 libzstd : $(LIBZSTD)
@@ -220,18 +294,42 @@ libzstd : $(LIBZSTD)
 .PHONY: lib
 lib : libzstd.a libzstd
 
-.PHONY: lib-mt
+
+# note : do not define lib-mt or lib-release as .PHONY
+# make does not consider implicit pattern rule for .PHONY target
+
 %-mt : CPPFLAGS += -DZSTD_MULTITHREAD
 %-mt : LDFLAGS  += -pthread
 %-mt : %
 	@echo multi-threading build completed
 
-.PHONY: lib-release
 %-release : DEBUGFLAGS :=
 %-release : %
 	@echo release build completed
 
 
+# Generate .h dependencies automatically
+
+DEPFLAGS = -MT $@ -MMD -MP -MF
+
+$(ZSTD_DYNLIB_DIR)/%.o : %.c $(ZSTD_DYNLIB_DIR)/%.d | $(ZSTD_DYNLIB_DIR)
+	@echo CC $@
+	$(COMPILE.c) $(DEPFLAGS) $(ZSTD_DYNLIB_DIR)/$*.d $(OUTPUT_OPTION) $<
+
+$(ZSTD_STATLIB_DIR)/%.o : %.c $(ZSTD_STATLIB_DIR)/%.d | $(ZSTD_STATLIB_DIR)
+	@echo CC $@
+	$(COMPILE.c) $(DEPFLAGS) $(ZSTD_STATLIB_DIR)/$*.d $(OUTPUT_OPTION) $<
+
+MKDIR ?= mkdir
+$(BUILD_DIR) $(ZSTD_DYNLIB_DIR) $(ZSTD_STATLIB_DIR):
+	$(MKDIR) -p $@
+
+DEPFILES := $(ZSTD_DYNLIB_OBJ:.o=.d) $(ZSTD_STATLIB_OBJ:.o=.d)
+$(DEPFILES):
+
+include $(wildcard $(DEPFILES))
+
+
 # Special case : building library in single-thread mode _and_ without zstdmt_compress.c
 ZSTDMT_FILES = compress/zstdmt_compress.c
 ZSTD_NOMT_FILES = $(filter-out $(ZSTDMT_FILES),$(ZSTD_FILES))
@@ -239,22 +337,24 @@ libzstd-nomt: LDFLAGS += -shared -fPIC -fvisibility=hidden
 libzstd-nomt: $(ZSTD_NOMT_FILES)
 	@echo compiling single-thread dynamic library $(LIBVER)
 	@echo files : $(ZSTD_NOMT_FILES)
-	$(Q)$(CC) $(FLAGS) $^ $(LDFLAGS) $(SONAME_FLAGS) -o $@
+	$(CC) $(FLAGS) $^ $(LDFLAGS) $(SONAME_FLAGS) -o $@
 
 clean:
-	$(Q)$(RM) -r *.dSYM   # macOS-specific
-	$(Q)$(RM) core *.o *.a *.gcda *.$(SHARED_EXT) *.$(SHARED_EXT).* libzstd.pc
-	$(Q)$(RM) dll/libzstd.dll dll/libzstd.lib libzstd-nomt*
-	$(Q)$(RM) common/*.o compress/*.o decompress/*.o dictBuilder/*.o legacy/*.o deprecated/*.o
+	$(RM) -r *.dSYM   # macOS-specific
+	$(RM) core *.o *.a *.gcda *.$(SHARED_EXT) *.$(SHARED_EXT).* libzstd.pc
+	$(RM) dll/libzstd.dll dll/libzstd.lib libzstd-nomt*
+	$(RM) -r obj/*
 	@echo Cleaning library completed
 
 #-----------------------------------------------------------------------------
 # make install is validated only for below listed environments
 #-----------------------------------------------------------------------------
-ifneq (,$(filter $(shell uname),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku))
+ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku))
 
 all: libzstd.pc
 
+HAS_EXPLICIT_EXEC_PREFIX := $(if $(or $(EXEC_PREFIX),$(exec_prefix)),1,)
+
 DESTDIR     ?=
 # directory variables : GNU conventions prefer lowercase
 # see https://www.gnu.org/prep/standards/html_node/Makefile-Conventions.html
@@ -268,35 +368,28 @@ LIBDIR      ?= $(libdir)
 includedir  ?= $(PREFIX)/include
 INCLUDEDIR  ?= $(includedir)
 
-PCLIBDIR ?= $(shell echo "$(LIBDIR)" | sed -n -E -e "s@^$(EXEC_PREFIX)(/|$$)@@p")
-PCINCDIR ?= $(shell echo "$(INCLUDEDIR)" | sed -n -E -e "s@^$(PREFIX)(/|$$)@@p")
+PCINCDIR := $(patsubst $(PREFIX)%,%,$(INCLUDEDIR))
+PCLIBDIR := $(patsubst $(EXEC_PREFIX)%,%,$(LIBDIR))
 
-ifeq (,$(PCLIBDIR))
-# Additional prefix check is required, since the empty string is technically a
-# valid PCLIBDIR
-ifeq (,$(shell echo "$(LIBDIR)" | sed -n -E -e "\\@^$(EXEC_PREFIX)(/|$$)@ p"))
-$(error configured libdir ($(LIBDIR)) is outside of prefix ($(PREFIX)), can't generate pkg-config file)
-endif
-endif
+# If we successfully stripped off a prefix, we'll add a reference to the
+# relevant pc variable.
+PCINCPREFIX := $(if $(findstring $(INCLUDEDIR),$(PCINCDIR)),,$${prefix})
+PCLIBPREFIX := $(if $(findstring $(LIBDIR),$(PCLIBDIR)),,$${exec_prefix})
 
-ifeq (,$(PCINCDIR))
-# Additional prefix check is required, since the empty string is technically a
-# valid PCINCDIR
-ifeq (,$(shell echo "$(INCLUDEDIR)" | sed -n -E -e "\\@^$(PREFIX)(/|$$)@ p"))
-$(error configured includedir ($(INCLUDEDIR)) is outside of exec_prefix ($(EXEC_PREFIX)), can't generate pkg-config file)
-endif
-endif
+# If no explicit EXEC_PREFIX was set by the caller, write it out as a reference
+# to PREFIX, rather than as a resolved value.
+PCEXEC_PREFIX := $(if $(HAS_EXPLICIT_EXEC_PREFIX),$(EXEC_PREFIX),$${prefix})
 
-ifneq (,$(filter $(shell uname),FreeBSD NetBSD DragonFly))
-PKGCONFIGDIR ?= $(PREFIX)/libdata/pkgconfig
+ifneq (,$(filter $(UNAME),FreeBSD NetBSD DragonFly))
+  PKGCONFIGDIR ?= $(PREFIX)/libdata/pkgconfig
 else
-PKGCONFIGDIR ?= $(LIBDIR)/pkgconfig
+  PKGCONFIGDIR ?= $(LIBDIR)/pkgconfig
 endif
 
-ifneq (,$(filter $(shell uname),SunOS))
-INSTALL ?= ginstall
+ifneq (,$(filter $(UNAME),SunOS))
+  INSTALL ?= ginstall
 else
-INSTALL ?= install
+  INSTALL ?= install
 endif
 
 INSTALL_PROGRAM ?= $(INSTALL)
@@ -306,9 +399,11 @@ INSTALL_DATA    ?= $(INSTALL) -m 644
 libzstd.pc:
 libzstd.pc: libzstd.pc.in
 	@echo creating pkgconfig
-	$(Q)@sed -E -e 's|@PREFIX@|$(PREFIX)|' \
-          -e 's|@LIBDIR@|$(PCLIBDIR)|' \
-          -e 's|@INCLUDEDIR@|$(PCINCDIR)|' \
+	@sed $(SED_ERE_OPT) \
+	        -e 's|@PREFIX@|$(PREFIX)|' \
+	        -e 's|@EXEC_PREFIX@|$(PCEXEC_PREFIX)|' \
+          -e 's|@INCLUDEDIR@|$(PCINCPREFIX)$(PCINCDIR)|' \
+          -e 's|@LIBDIR@|$(PCLIBPREFIX)$(PCLIBDIR)|' \
           -e 's|@VERSION@|$(VERSION)|' \
           $< >$@
 
@@ -316,39 +411,41 @@ install: install-pc install-static install-shared install-includes
 	@echo zstd static and shared library installed
 
 install-pc: libzstd.pc
-	$(Q)$(INSTALL) -d -m 755 $(DESTDIR)$(PKGCONFIGDIR)/
-	$(Q)$(INSTALL_DATA) libzstd.pc $(DESTDIR)$(PKGCONFIGDIR)/
+	[ -e $(DESTDIR)$(PKGCONFIGDIR) ] || $(INSTALL) -d -m 755 $(DESTDIR)$(PKGCONFIGDIR)/
+	$(INSTALL_DATA) libzstd.pc $(DESTDIR)$(PKGCONFIGDIR)/
 
-install-static: libzstd.a
+install-static:
+	# only generate libzstd.a if it's not already present
+	[ -e libzstd.a ] || $(MAKE) libzstd.a-release
+	[ -e $(DESTDIR)$(LIBDIR) ] || $(INSTALL) -d -m 755 $(DESTDIR)$(LIBDIR)/
 	@echo Installing static library
-	$(Q)$(INSTALL) -d -m 755 $(DESTDIR)$(LIBDIR)/
-	$(Q)$(INSTALL_DATA) libzstd.a $(DESTDIR)$(LIBDIR)
+	$(INSTALL_DATA) libzstd.a $(DESTDIR)$(LIBDIR)
 
-install-shared: libzstd
+install-shared:
+	# only generate libzstd.so if it's not already present
+	[ -e $(LIBZSTD) ] || $(MAKE) libzstd-release
+	[ -e $(DESTDIR)$(LIBDIR) ] || $(INSTALL) -d -m 755 $(DESTDIR)$(LIBDIR)/
 	@echo Installing shared library
-	$(Q)$(INSTALL) -d -m 755 $(DESTDIR)$(LIBDIR)/
-	$(Q)$(INSTALL_PROGRAM) $(LIBZSTD) $(DESTDIR)$(LIBDIR)
-	$(Q)ln -sf $(LIBZSTD) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT_MAJOR)
-	$(Q)ln -sf $(LIBZSTD) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT)
+	$(INSTALL_PROGRAM) $(LIBZSTD) $(DESTDIR)$(LIBDIR)
+	ln -sf $(LIBZSTD) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT_MAJOR)
+	ln -sf $(LIBZSTD) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT)
 
 install-includes:
+	[ -e $(DESTDIR)$(INCLUDEDIR) ] || $(INSTALL) -d -m 755 $(DESTDIR)$(INCLUDEDIR)/
 	@echo Installing includes
-	$(Q)$(INSTALL) -d -m 755 $(DESTDIR)$(INCLUDEDIR)/
-	$(Q)$(INSTALL_DATA) zstd.h $(DESTDIR)$(INCLUDEDIR)
-	$(Q)$(INSTALL_DATA) common/zstd_errors.h $(DESTDIR)$(INCLUDEDIR)
-	$(Q)$(INSTALL_DATA) deprecated/zbuff.h $(DESTDIR)$(INCLUDEDIR)     # prototypes generate deprecation warnings
-	$(Q)$(INSTALL_DATA) dictBuilder/zdict.h $(DESTDIR)$(INCLUDEDIR)
+	$(INSTALL_DATA) zstd.h $(DESTDIR)$(INCLUDEDIR)
+	$(INSTALL_DATA) common/zstd_errors.h $(DESTDIR)$(INCLUDEDIR)
+	$(INSTALL_DATA) dictBuilder/zdict.h $(DESTDIR)$(INCLUDEDIR)
 
 uninstall:
-	$(Q)$(RM) $(DESTDIR)$(LIBDIR)/libzstd.a
-	$(Q)$(RM) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT)
-	$(Q)$(RM) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT_MAJOR)
-	$(Q)$(RM) $(DESTDIR)$(LIBDIR)/$(LIBZSTD)
-	$(Q)$(RM) $(DESTDIR)$(PKGCONFIGDIR)/libzstd.pc
-	$(Q)$(RM) $(DESTDIR)$(INCLUDEDIR)/zstd.h
-	$(Q)$(RM) $(DESTDIR)$(INCLUDEDIR)/zstd_errors.h
-	$(Q)$(RM) $(DESTDIR)$(INCLUDEDIR)/zbuff.h   # Deprecated streaming functions
-	$(Q)$(RM) $(DESTDIR)$(INCLUDEDIR)/zdict.h
+	$(RM) $(DESTDIR)$(LIBDIR)/libzstd.a
+	$(RM) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT)
+	$(RM) $(DESTDIR)$(LIBDIR)/libzstd.$(SHARED_EXT_MAJOR)
+	$(RM) $(DESTDIR)$(LIBDIR)/$(LIBZSTD)
+	$(RM) $(DESTDIR)$(PKGCONFIGDIR)/libzstd.pc
+	$(RM) $(DESTDIR)$(INCLUDEDIR)/zstd.h
+	$(RM) $(DESTDIR)$(INCLUDEDIR)/zstd_errors.h
+	$(RM) $(DESTDIR)$(INCLUDEDIR)/zdict.h
 	@echo zstd libraries successfully uninstalled
 
 endif
diff --git a/lib/README.md b/lib/README.md
index 6ccffb13868c..db9170a75c42 100644
--- a/lib/README.md
+++ b/lib/README.md
@@ -143,6 +143,14 @@ The file structure is designed to make this selection manually achievable for an
   Setting this macro will either force to generate the BMI2 dispatcher (1)
   or prevent it (0). It overrides automatic detection.
 
+- The build macro `ZSTD_NO_UNUSED_FUNCTIONS` can be defined to hide the definitions of functions
+  that zstd does not use. Not all unused functions are hidden, but they can be if needed.
+  Currently, this macro will hide function definitions in FSE and HUF that use an excessive
+  amount of stack space.
+
+- The build macro `ZSTD_NO_INTRINSICS` can be defined to disable all explicit intrinsics.
+  Compiler builtins are still used.
+
 
 #### Windows : using MinGW+MSYS to create DLL
 
@@ -160,6 +168,26 @@ file it should be linked with `dll\libzstd.dll`. For example:
 The compiled executable will require ZSTD DLL which is available at `dll\libzstd.dll`.
 
 
+#### Advanced Build options
+
+The build system requires a hash function in order to
+separate object files created with different compilation flags.
+By default, it tries to use `md5sum` or equivalent.
+The hash function can be manually switched by setting the `HASH` variable.
+For example : `make HASH=xxhsum`
+The hash function needs to generate at least 64-bit using hexadecimal format.
+When no hash function is found,
+the Makefile just generates all object files into the same default directory,
+irrespective of compilation flags.
+This functionality only matters if `libzstd` is compiled multiple times
+with different build flags.
+
+The build directory, where object files are stored
+can also be manually controlled using variable `BUILD_DIR`,
+for example `make BUILD_DIR=objectDir/v1`.
+In which case, the hash function doesn't matter.
+
+
 #### Deprecated API
 
 Obsolete API on their way out are stored in directory `lib/deprecated`.
diff --git a/lib/common/bitstream.h b/lib/common/bitstream.h
index 37b99c01eed3..d9a2730104da 100644
--- a/lib/common/bitstream.h
+++ b/lib/common/bitstream.h
@@ -17,7 +17,6 @@
 #if defined (__cplusplus)
 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,
@@ -36,10 +35,12 @@ extern "C" {
 /*=========================================
 *  Target specific
 =========================================*/
-#if defined(__BMI__) && defined(__GNUC__)
-#  include <immintrin.h>   /* support for bextr (experimental) */
-#elif defined(__ICCARM__)
-#  include <intrinsics.h>
+#ifndef ZSTD_NO_INTRINSICS
+#  if defined(__BMI__) && defined(__GNUC__)
+#    include <immintrin.h>   /* support for bextr (experimental) */
+#  elif defined(__ICCARM__)
+#    include <intrinsics.h>
+#  endif
 #endif
 
 #define STREAM_ACCUMULATOR_MIN_32  25
@@ -141,8 +142,12 @@ MEM_STATIC unsigned BIT_highbit32 (U32 val)
     assert(val != 0);
     {
 #   if defined(_MSC_VER)   /* Visual */
-        unsigned long r=0;
-        return _BitScanReverse ( &r, val ) ? (unsigned)r : 0;
+#       if STATIC_BMI2 == 1
+		return _lzcnt_u32(val) ^ 31;
+#       else
+		unsigned long r = 0;
+		return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#       endif
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
         return __builtin_clz (val) ^ 31;
 #   elif defined(__ICCARM__)    /* IAR Intrinsic */
@@ -198,7 +203,7 @@ MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
 MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
                             size_t value, unsigned nbBits)
 {
-    MEM_STATIC_ASSERT(BIT_MASK_SIZE == 32);
+    DEBUG_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;
@@ -271,7 +276,7 @@ MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
  */
 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); }
+    if (srcSize < 1) { ZSTD_memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
 
     bitD->start = (const char*)srcBuffer;
     bitD->limitPtr = bitD->start + sizeof(bitD->bitContainer);
@@ -317,12 +322,12 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
     return srcSize;
 }
 
-MEM_STATIC size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
 {
     return bitContainer >> start;
 }
 
-MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
 {
     U32 const regMask = sizeof(bitContainer)*8 - 1;
     /* if start > regMask, bitstream is corrupted, and result is undefined */
@@ -330,10 +335,14 @@ MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 co
     return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
 }
 
-MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
 {
+#if defined(STATIC_BMI2) && STATIC_BMI2 == 1
+	return  _bzhi_u64(bitContainer, nbBits);
+#else
     assert(nbBits < BIT_MASK_SIZE);
     return bitContainer & BIT_mask[nbBits];
+#endif
 }
 
 /*! BIT_lookBits() :
@@ -342,7 +351,7 @@ MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
  *  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)
+MEM_STATIC  FORCE_INLINE_ATTR size_t BIT_lookBits(const BIT_DStream_t*  bitD, U32 nbBits)
 {
     /* arbitrate between double-shift and shift+mask */
 #if 1
@@ -365,7 +374,7 @@ MEM_STATIC size_t BIT_lookBitsFast(const BIT_DStream_t* bitD, U32 nbBits)
     return (bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> (((regMask+1)-nbBits) & regMask);
 }
 
-MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+MEM_STATIC FORCE_INLINE_ATTR void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
 {
     bitD->bitsConsumed += nbBits;
 }
@@ -374,7 +383,7 @@ MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
  *  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. */
-MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
 {
     size_t const value = BIT_lookBits(bitD, nbBits);
     BIT_skipBits(bitD, nbBits);
diff --git a/lib/common/compiler.h b/lib/common/compiler.h
index 95e9483521d4..3e454f38c12b 100644
--- a/lib/common/compiler.h
+++ b/lib/common/compiler.h
@@ -39,6 +39,17 @@
 #endif
 
 /**
+  On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC).
+  This explictly marks such functions as __cdecl so that the code will still compile
+  if a CC other than __cdecl has been made the default.
+*/
+#if  defined(_MSC_VER)
+#  define WIN_CDECL __cdecl
+#else
+#  define WIN_CDECL
+#endif
+
+/**
  * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
  * parameters. They must be inlined for the compiler to eliminate the constant
  * branches.
@@ -114,12 +125,12 @@
 #    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
 #    define PREFETCH_L1(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
 #    define PREFETCH_L2(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T1)
-#    elif defined(__aarch64__)
-#     define PREFETCH_L1(ptr)  __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
-#     define PREFETCH_L2(ptr)  __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
 #  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
 #    define PREFETCH_L1(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
 #    define PREFETCH_L2(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
+#  elif defined(__aarch64__)
+#    define PREFETCH_L1(ptr)  __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
+#    define PREFETCH_L2(ptr)  __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
 #  else
 #    define PREFETCH_L1(ptr) (void)(ptr)  /* disabled */
 #    define PREFETCH_L2(ptr) (void)(ptr)  /* disabled */
@@ -172,4 +183,106 @@
 #  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
 #endif
 
+/*Like DYNAMIC_BMI2 but for compile time determination of BMI2 support*/
+#ifndef STATIC_BMI2
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))
+#    ifdef __AVX2__  //MSVC does not have a BMI2 specific flag, but every CPU that supports AVX2 also supports BMI2
+#       define STATIC_BMI2 1
+#    endif
+#  endif
+#endif
+
+#ifndef STATIC_BMI2
+    #define STATIC_BMI2 0
+#endif
+
+/* compat. with non-clang compilers */
+#ifndef __has_builtin
+#  define __has_builtin(x) 0
+#endif
+
+/* compat. with non-clang compilers */
+#ifndef __has_feature
+#  define __has_feature(x) 0
+#endif
+
+/* detects whether we are being compiled under msan */
+#ifndef ZSTD_MEMORY_SANITIZER
+#  if __has_feature(memory_sanitizer)
+#    define ZSTD_MEMORY_SANITIZER 1
+#  else
+#    define ZSTD_MEMORY_SANITIZER 0
+#  endif
+#endif
+
+#if ZSTD_MEMORY_SANITIZER
+/* Not all platforms that support msan provide sanitizers/msan_interface.h.
+ * We therefore declare the functions we need ourselves, rather than trying to
+ * include the header file... */
+#include <stddef.h>  /* size_t */
+#define ZSTD_DEPS_NEED_STDINT
+#include "zstd_deps.h"  /* intptr_t */
+
+/* Make memory region fully initialized (without changing its contents). */
+void __msan_unpoison(const volatile void *a, size_t size);
+
+/* Make memory region fully uninitialized (without changing its contents).
+   This is a legacy interface that does not update origin information. Use
+   __msan_allocated_memory() instead. */
+void __msan_poison(const volatile void *a, size_t size);
+
+/* Returns the offset of the first (at least partially) poisoned byte in the
+   memory range, or -1 if the whole range is good. */
+intptr_t __msan_test_shadow(const volatile void *x, size_t size);
+#endif
+
+/* detects whether we are being compiled under asan */
+#ifndef ZSTD_ADDRESS_SANITIZER
+#  if __has_feature(address_sanitizer)
+#    define ZSTD_ADDRESS_SANITIZER 1
+#  elif defined(__SANITIZE_ADDRESS__)
+#    define ZSTD_ADDRESS_SANITIZER 1
+#  else
+#    define ZSTD_ADDRESS_SANITIZER 0
+#  endif
+#endif
+
+#if ZSTD_ADDRESS_SANITIZER
+/* Not all platforms that support asan provide sanitizers/asan_interface.h.
+ * We therefore declare the functions we need ourselves, rather than trying to
+ * include the header file... */
+#include <stddef.h>  /* size_t */
+
+/**
+ * Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable.
+ *
+ * This memory must be previously allocated by your program. Instrumented
+ * code is forbidden from accessing addresses in this region until it is
+ * unpoisoned. This function is not guaranteed to poison the entire region -
+ * it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan
+ * alignment restrictions.
+ *
+ * \note This function is not thread-safe because no two threads can poison or
+ * unpoison memory in the same memory region simultaneously.
+ *
+ * \param addr Start of memory region.
+ * \param size Size of memory region. */
+void __asan_poison_memory_region(void const volatile *addr, size_t size);
+
+/**
+ * Marks a memory region (<c>[addr, addr+size)</c>) as addressable.
+ *
+ * This memory must be previously allocated by your program. Accessing
+ * addresses in this region is allowed until this region is poisoned again.
+ * This function could unpoison a super-region of <c>[addr, addr+size)</c> due
+ * to ASan alignment restrictions.
+ *
+ * \note This function is not thread-safe because no two threads can
+ * poison or unpoison memory in the same memory region simultaneously.
+ *
+ * \param addr Start of memory region.
+ * \param size Size of memory region. */
+void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
+#endif
+
 #endif /* ZSTD_COMPILER_H */
diff --git a/lib/common/cpu.h b/lib/common/cpu.h
index 6e8a974f62d7..cb210593ead9 100644
--- a/lib/common/cpu.h
+++ b/lib/common/cpu.h
@@ -16,8 +16,6 @@
  * https://github.com/facebook/folly/blob/master/folly/CpuId.h
  */
 
-#include <string.h>
-
 #include "mem.h"
 
 #ifdef _MSC_VER
diff --git a/lib/common/debug.h b/lib/common/debug.h
index ac6224888d8b..8b5734366ca2 100644
--- a/lib/common/debug.h
+++ b/lib/common/debug.h
@@ -51,15 +51,6 @@ extern "C" {
 #endif
 
 
-/* DEBUGFILE can be defined externally,
- * typically through compiler command line.
- * note : currently useless.
- * Value must be stderr or stdout */
-#ifndef DEBUGFILE
-#  define DEBUGFILE stderr
-#endif
-
-
 /* recommended values for DEBUGLEVEL :
  * 0 : release mode, no debug, all run-time checks disabled
  * 1 : enables assert() only, no display
@@ -76,7 +67,8 @@ extern "C" {
  */
 
 #if (DEBUGLEVEL>=1)
-#  include <assert.h>
+#  define ZSTD_DEPS_NEED_ASSERT
+#  include "zstd_deps.h"
 #else
 #  ifndef assert   /* assert may be already defined, due to prior #include <assert.h> */
 #    define assert(condition) ((void)0)   /* disable assert (default) */
@@ -84,7 +76,8 @@ extern "C" {
 #endif
 
 #if (DEBUGLEVEL>=2)
-#  include <stdio.h>
+#  define ZSTD_DEPS_NEED_IO
+#  include "zstd_deps.h"
 extern int g_debuglevel; /* the variable is only declared,
                             it actually lives in debug.c,
                             and is shared by the whole process.
@@ -92,14 +85,14 @@ extern int g_debuglevel; /* the variable is only declared,
                             It's useful when enabling very verbose levels
                             on selective conditions (such as position in src) */
 
-#  define RAWLOG(l, ...) {                                      \
-                if (l<=g_debuglevel) {                          \
-                    fprintf(stderr, __VA_ARGS__);               \
+#  define RAWLOG(l, ...) {                                       \
+                if (l<=g_debuglevel) {                           \
+                    ZSTD_DEBUG_PRINT(__VA_ARGS__);               \
             }   }
-#  define DEBUGLOG(l, ...) {                                    \
-                if (l<=g_debuglevel) {                          \
-                    fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
-                    fprintf(stderr, " \n");                     \
+#  define DEBUGLOG(l, ...) {                                     \
+                if (l<=g_debuglevel) {                           \
+                    ZSTD_DEBUG_PRINT(__FILE__ ": " __VA_ARGS__); \
+                    ZSTD_DEBUG_PRINT(" \n");                     \
             }   }
 #else
 #  define RAWLOG(l, ...)      {}    /* disabled */
diff --git a/lib/common/entropy_common.c b/lib/common/entropy_common.c
index 9d3e4e8e36ab..f9fcb1acfc36 100644
--- a/lib/common/entropy_common.c
+++ b/lib/common/entropy_common.c
@@ -38,8 +38,31 @@ const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
 /*-**************************************************************
 *  FSE NCount encoding-decoding
 ****************************************************************/
-size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
-                 const void* headerBuffer, size_t hbSize)
+static U32 FSE_ctz(U32 val)
+{
+    assert(val != 0);
+    {
+#   if defined(_MSC_VER)   /* Visual */
+        unsigned long r=0;
+        return _BitScanForward(&r, val) ? (unsigned)r : 0;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* GCC Intrinsic */
+        return __builtin_ctz(val);
+#   elif defined(__ICCARM__)    /* IAR Intrinsic */
+        return __CTZ(val);
+#   else   /* Software version */
+        U32 count = 0;
+        while ((val & 1) == 0) {
+            val >>= 1;
+            ++count;
+        }
+        return count;
+#   endif
+    }
+}
+
+FORCE_INLINE_TEMPLATE
+size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                           const void* headerBuffer, size_t hbSize)
 {
     const BYTE* const istart = (const BYTE*) headerBuffer;
     const BYTE* const iend = istart + hbSize;
@@ -50,23 +73,23 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
     U32 bitStream;
     int bitCount;
     unsigned charnum = 0;
+    unsigned const maxSV1 = *maxSVPtr + 1;
     int previous0 = 0;
 
-    if (hbSize < 4) {
-        /* This function only works when hbSize >= 4 */
-        char buffer[4];
-        memset(buffer, 0, sizeof(buffer));
-        memcpy(buffer, headerBuffer, hbSize);
+    if (hbSize < 8) {
+        /* This function only works when hbSize >= 8 */
+        char buffer[8] = {0};
+        ZSTD_memcpy(buffer, headerBuffer, hbSize);
         {   size_t const countSize = FSE_readNCount(normalizedCounter, maxSVPtr, tableLogPtr,
                                                     buffer, sizeof(buffer));
             if (FSE_isError(countSize)) return countSize;
             if (countSize > hbSize) return ERROR(corruption_detected);
             return countSize;
     }   }
-    assert(hbSize >= 4);
+    assert(hbSize >= 8);
 
     /* init */
-    memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0]));   /* all symbols not present in NCount have a frequency of 0 */
+    ZSTD_memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0]));   /* all symbols not present in NCount have a frequency of 0 */
     bitStream = MEM_readLE32(ip);
     nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
     if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
@@ -77,36 +100,58 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
     threshold = 1<<nbBits;
     nbBits++;
 
-    while ((remaining>1) & (charnum<=*maxSVPtr)) {
+    for (;;) {
         if (previous0) {
-            unsigned n0 = charnum;
-            while ((bitStream & 0xFFFF) == 0xFFFF) {
-                n0 += 24;
-                if (ip < iend-5) {
-                    ip += 2;
-                    bitStream = MEM_readLE32(ip) >> bitCount;
+            /* Count the number of repeats. Each time the
+             * 2-bit repeat code is 0b11 there is another
+             * repeat.
+             * Avoid UB by setting the high bit to 1.
+             */
+            int repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
+            while (repeats >= 12) {
+                charnum += 3 * 12;
+                if (LIKELY(ip <= iend-7)) {
+                    ip += 3;
                 } else {
-                    bitStream >>= 16;
-                    bitCount   += 16;
-            }   }
-            while ((bitStream & 3) == 3) {
-                n0 += 3;
-                bitStream >>= 2;
-                bitCount += 2;
+                    bitCount -= (int)(8 * (iend - 7 - ip));
+                    bitCount &= 31;
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> bitCount;
+                repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
             }
-            n0 += bitStream & 3;
+            charnum += 3 * repeats;
+            bitStream >>= 2 * repeats;
+            bitCount += 2 * repeats;
+
+            /* Add the final repeat which isn't 0b11. */
+            assert((bitStream & 3) < 3);
+            charnum += bitStream & 3;
             bitCount += 2;
-            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
-            while (charnum < n0) normalizedCounter[charnum++] = 0;
-            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+
+            /* This is an error, but break and return an error
+             * at the end, because returning out of a loop makes
+             * it harder for the compiler to optimize.
+             */
+            if (charnum >= maxSV1) break;
+
+            /* We don't need to set the normalized count to 0
+             * because we already memset the whole buffer to 0.
+             */
+
+            if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
                 assert((bitCount >> 3) <= 3); /* For first condition to work */
                 ip += bitCount>>3;
                 bitCount &= 7;
-                bitStream = MEM_readLE32(ip) >> bitCount;
             } else {
-                bitStream >>= 2;
-        }   }
-        {   int const max = (2*threshold-1) - remaining;
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                bitCount &= 31;
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> bitCount;
+        }
+        {
+            int const max = (2*threshold-1) - remaining;
             int count;
 
             if ((bitStream & (threshold-1)) < (U32)max) {
@@ -119,24 +164,43 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
             }
 
             count--;   /* extra accuracy */
-            remaining -= count < 0 ? -count : count;   /* -1 means +1 */
+            /* When it matters (small blocks), this is a
+             * predictable branch, because we don't use -1.
+             */
+            if (count >= 0) {
+                remaining -= count;
+            } else {
+                assert(count == -1);
+                remaining += count;
+            }
             normalizedCounter[charnum++] = (short)count;
             previous0 = !count;
-            while (remaining < threshold) {
-                nbBits--;
-                threshold >>= 1;
+
+            assert(threshold > 1);
+            if (remaining < threshold) {
+                /* This branch can be folded into the
+                 * threshold update condition because we
+                 * know that threshold > 1.
+                 */
+                if (remaining <= 1) break;
+                nbBits = BIT_highbit32(remaining) + 1;
+                threshold = 1 << (nbBits - 1);
             }
+            if (charnum >= maxSV1) break;
 
-            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+            if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
                 ip += bitCount>>3;
                 bitCount &= 7;
             } else {
                 bitCount -= (int)(8 * (iend - 4 - ip));
+                bitCount &= 31;
                 ip = iend - 4;
             }
-            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
-    }   }   /* while ((remaining>1) & (charnum<=*maxSVPtr)) */
+            bitStream = MEM_readLE32(ip) >> bitCount;
+    }   }
     if (remaining != 1) return ERROR(corruption_detected);
+    /* Only possible when there are too many zeros. */
+    if (charnum > maxSV1) return ERROR(maxSymbolValue_tooSmall);
     if (bitCount > 32) return ERROR(corruption_detected);
     *maxSVPtr = charnum-1;
 
@@ -144,6 +208,43 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
     return ip-istart;
 }
 
+/* Avoids the FORCE_INLINE of the _body() function. */
+static size_t FSE_readNCount_body_default(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize)
+{
+    return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+}
+
+#if DYNAMIC_BMI2
+TARGET_ATTRIBUTE("bmi2") static size_t FSE_readNCount_body_bmi2(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize)
+{
+    return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+}
+#endif
+
+size_t FSE_readNCount_bmi2(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return FSE_readNCount_body_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+    }
+#endif
+    (void)bmi2;
+    return FSE_readNCount_body_default(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
+}
+
+size_t FSE_readNCount(
+        short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+        const void* headerBuffer, size_t hbSize)
+{
+    return FSE_readNCount_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize, /* bmi2 */ 0);
+}
+
 
 /*! HUF_readStats() :
     Read compact Huffman tree, saved by HUF_writeCTable().
@@ -156,6 +257,17 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                      U32* nbSymbolsPtr, U32* tableLogPtr,
                      const void* src, size_t srcSize)
 {
+    U32 wksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
+    return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp), /* bmi2 */ 0);
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_readStats_body(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                   U32* nbSymbolsPtr, U32* tableLogPtr,
+                   const void* src, size_t srcSize,
+                   void* workSpace, size_t wkspSize,
+                   int bmi2)
+{
     U32 weightTotal;
     const BYTE* ip = (const BYTE*) src;
     size_t iSize;
@@ -163,7 +275,7 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
 
     if (!srcSize) return ERROR(srcSize_wrong);
     iSize = ip[0];
-    /* memset(huffWeight, 0, hwSize);   *//* is not necessary, even though some analyzer complain ... */
+    /* ZSTD_memset(huffWeight, 0, hwSize);   *//* is not necessary, even though some analyzer complain ... */
 
     if (iSize >= 128) {  /* special header */
         oSize = iSize - 127;
@@ -177,14 +289,14 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                 huffWeight[n+1] = ip[n/2] & 15;
     }   }   }
     else  {   /* header compressed with FSE (normal case) */
-        FSE_DTable fseWorkspace[FSE_DTABLE_SIZE_U32(6)];  /* 6 is max possible tableLog for HUF header (maybe even 5, to be tested) */
         if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
-        oSize = FSE_decompress_wksp(huffWeight, hwSize-1, ip+1, iSize, fseWorkspace, 6);   /* max (hwSize-1) values decoded, as last one is implied */
+        /* max (hwSize-1) values decoded, as last one is implied */
+        oSize = FSE_decompress_wksp_bmi2(huffWeight, hwSize-1, ip+1, iSize, 6, workSpace, wkspSize, bmi2);
         if (FSE_isError(oSize)) return oSize;
     }
 
     /* collect weight stats */
-    memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
+    ZSTD_memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
     weightTotal = 0;
     {   U32 n; for (n=0; n<oSize; n++) {
             if (huffWeight[n] >= HUF_TABLELOG_MAX) return ERROR(corruption_detected);
@@ -214,3 +326,37 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
     *nbSymbolsPtr = (U32)(oSize+1);
     return iSize+1;
 }
+
+/* Avoids the FORCE_INLINE of the _body() function. */
+static size_t HUF_readStats_body_default(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize,
+                     void* workSpace, size_t wkspSize)
+{
+    return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 0);
+}
+
+#if DYNAMIC_BMI2
+static TARGET_ATTRIBUTE("bmi2") size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize,
+                     void* workSpace, size_t wkspSize)
+{
+    return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 1);
+}
+#endif
+
+size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize,
+                     void* workSpace, size_t wkspSize,
+                     int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return HUF_readStats_body_bmi2(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
+    }
+#endif
+    (void)bmi2;
+    return HUF_readStats_body_default(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
+}
diff --git a/lib/common/error_private.c b/lib/common/error_private.c
index cd437529c12b..45bba5305b5a 100644
--- a/lib/common/error_private.c
+++ b/lib/common/error_private.c
@@ -48,6 +48,7 @@ const char* ERR_getErrorString(ERR_enum code)
     case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
     case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
     case PREFIX(dstBuffer_wrong): return "Destination buffer is wrong";
+    case PREFIX(srcBuffer_wrong): return "Source buffer is wrong";
     case PREFIX(maxCode):
     default: return notErrorCode;
     }
diff --git a/lib/common/error_private.h b/lib/common/error_private.h
index 982cf8e9fe6f..71b37b8dfa43 100644
--- a/lib/common/error_private.h
+++ b/lib/common/error_private.h
@@ -21,7 +21,7 @@ extern "C" {
 /* ****************************************
 *  Dependencies
 ******************************************/
-#include <stddef.h>        /* size_t */
+#include "zstd_deps.h"    /* size_t */
 #include "zstd_errors.h"  /* enum list */
 
 
diff --git a/lib/common/fse.h b/lib/common/fse.h
index ff54e70ea75c..dd5fc44e8092 100644
--- a/lib/common/fse.h
+++ b/lib/common/fse.h
@@ -23,7 +23,7 @@ extern "C" {
 /*-*****************************************
 *  Dependencies
 ******************************************/
-#include <stddef.h>    /* size_t, ptrdiff_t */
+#include "zstd_deps.h"    /* size_t, ptrdiff_t */
 
 
 /*-*****************************************
@@ -137,10 +137,16 @@ FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize
 /*! FSE_normalizeCount():
     normalize counts so that sum(count[]) == Power_of_2 (2^tableLog)
     'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
+    useLowProbCount is a boolean parameter which trades off compressed size for
+    faster header decoding. When it is set to 1, the compressed data will be slightly
+    smaller. And when it is set to 0, FSE_readNCount() and FSE_buildDTable() will be
+    faster. If you are compressing a small amount of data (< 2 KB) then useLowProbCount=0
+    is a good default, since header deserialization makes a big speed difference.
+    Otherwise, useLowProbCount=1 is a good default, since the speed difference is small.
     @return : tableLog,
               or an errorCode, which can be tested using FSE_isError() */
 FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog,
-                    const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
+                    const unsigned* count, size_t srcSize, unsigned maxSymbolValue, unsigned useLowProbCount);
 
 /*! FSE_NCountWriteBound():
     Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
@@ -228,6 +234,13 @@ FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter,
                            unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
                            const void* rBuffer, size_t rBuffSize);
 
+/*! FSE_readNCount_bmi2():
+ * Same as FSE_readNCount() but pass bmi2=1 when your CPU supports BMI2 and 0 otherwise.
+ */
+FSE_PUBLIC_API size_t FSE_readNCount_bmi2(short* normalizedCounter,
+                           unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
+                           const void* rBuffer, size_t rBuffSize, int bmi2);
+
 /*! 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* */
@@ -288,12 +301,12 @@ If there is an error, the function will return an error code, which can be teste
 *******************************************/
 /* FSE buffer bounds */
 #define FSE_NCOUNTBOUND 512
-#define FSE_BLOCKBOUND(size) (size + (size>>7) + 4 /* fse states */ + sizeof(size_t) /* bitContainer */)
+#define FSE_BLOCKBOUND(size) ((size) + ((size)>>7) + 4 /* fse states */ + sizeof(size_t) /* bitContainer */)
 #define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
 
 /* It is possible to statically allocate FSE CTable/DTable as a table of FSE_CTable/FSE_DTable using below macros */
-#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
-#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+#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))
@@ -309,9 +322,9 @@ unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsi
 
 /* FSE_compress_wksp() :
  * 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.
+ * FSE_COMPRESS_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) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
+#define FSE_COMPRESS_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);
@@ -322,18 +335,30 @@ size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
 
 /* FSE_buildCTable_wksp() :
  * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
- * `wkspSize` must be >= `(1<<tableLog)`.
+ * `wkspSize` must be >= `FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog)` of `unsigned`.
  */
+#define FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog) (maxSymbolValue + 2 + (1ull << (tableLog - 2)))
+#define FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) (sizeof(unsigned) * FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog))
 size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
 
+#define FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) (sizeof(short) * (maxSymbolValue + 1) + (1ULL << maxTableLog) + 8)
+#define FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) ((FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) + sizeof(unsigned) - 1) / sizeof(unsigned))
+FSE_PUBLIC_API size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
+/**< Same as FSE_buildDTable(), using an externally allocated `workspace` produced with `FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxSymbolValue)` */
+
 size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
 /**< build a fake FSE_DTable, designed to read a flat distribution where each symbol uses nbBits */
 
 size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
 /**< build a fake FSE_DTable, designed to always generate the same symbolValue */
 
-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)` */
+#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue))
+#define FSE_DECOMPRESS_WKSP_SIZE(maxTableLog, maxSymbolValue) (FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(unsigned))
+size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize);
+/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)` */
+
+size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2);
+/**< Same as FSE_decompress_wksp() but with dynamic BMI2 support. Pass 1 if your CPU supports BMI2 or 0 if it doesn't. */
 
 typedef enum {
    FSE_repeat_none,  /**< Cannot use the previous table */
@@ -644,6 +669,9 @@ MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
 #ifndef FSE_DEFAULT_MEMORY_USAGE
 #  define FSE_DEFAULT_MEMORY_USAGE 13
 #endif
+#if (FSE_DEFAULT_MEMORY_USAGE > FSE_MAX_MEMORY_USAGE)
+#  error "FSE_DEFAULT_MEMORY_USAGE must be <= FSE_MAX_MEMORY_USAGE"
+#endif
 
 /*!FSE_MAX_SYMBOL_VALUE :
 *  Maximum symbol value authorized.
@@ -677,7 +705,7 @@ MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
 #  error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
 #endif
 
-#define FSE_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)
+#define FSE_TABLESTEP(tableSize) (((tableSize)>>1) + ((tableSize)>>3) + 3)
 
 
 #endif /* FSE_STATIC_LINKING_ONLY */
diff --git a/lib/common/fse_decompress.c b/lib/common/fse_decompress.c
index bcc2223ccc65..c164430f9905 100644
--- a/lib/common/fse_decompress.c
+++ b/lib/common/fse_decompress.c
@@ -16,13 +16,14 @@
 /* **************************************************************
 *  Includes
 ****************************************************************/
-#include <stdlib.h>     /* malloc, free, qsort */
-#include <string.h>     /* memcpy, memset */
+#include "debug.h"      /* assert */
 #include "bitstream.h"
 #include "compiler.h"
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
 #include "error_private.h"
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"
 
 
 /* **************************************************************
@@ -59,25 +60,27 @@
 FSE_DTable* FSE_createDTable (unsigned tableLog)
 {
     if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
-    return (FSE_DTable*)malloc( FSE_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+    return (FSE_DTable*)ZSTD_malloc( FSE_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
 }
 
 void FSE_freeDTable (FSE_DTable* dt)
 {
-    free(dt);
+    ZSTD_free(dt);
 }
 
-size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
 {
     void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
     FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
-    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U16* symbolNext = (U16*)workSpace;
+    BYTE* spread = (BYTE*)(symbolNext + maxSymbolValue + 1);
 
     U32 const maxSV1 = maxSymbolValue + 1;
     U32 const tableSize = 1 << tableLog;
     U32 highThreshold = tableSize-1;
 
     /* Sanity Checks */
+    if (FSE_BUILD_DTABLE_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(maxSymbolValue_tooLarge);
     if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
 
@@ -95,11 +98,57 @@ size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned
                     if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
                     symbolNext[s] = normalizedCounter[s];
         }   }   }
-        memcpy(dt, &DTableH, sizeof(DTableH));
+        ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
     }
 
     /* Spread symbols */
-    {   U32 const tableMask = tableSize-1;
+    if (highThreshold == tableSize - 1) {
+        size_t const tableMask = tableSize-1;
+        size_t const step = FSE_TABLESTEP(tableSize);
+        /* First lay down the symbols in order.
+         * We use a uint64_t to lay down 8 bytes at a time. This reduces branch
+         * misses since small blocks generally have small table logs, so nearly
+         * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
+         * our buffer to handle the over-write.
+         */
+        {
+            U64 const add = 0x0101010101010101ull;
+            size_t pos = 0;
+            U64 sv = 0;
+            U32 s;
+            for (s=0; s<maxSV1; ++s, sv += add) {
+                int i;
+                int const n = normalizedCounter[s];
+                MEM_write64(spread + pos, sv);
+                for (i = 8; i < n; i += 8) {
+                    MEM_write64(spread + pos + i, sv);
+                }
+                pos += n;
+            }
+        }
+        /* Now we spread those positions across the table.
+         * The benefit of doing it in two stages is that we avoid the the
+         * variable size inner loop, which caused lots of branch misses.
+         * Now we can run through all the positions without any branch misses.
+         * We unroll the loop twice, since that is what emperically worked best.
+         */
+        {
+            size_t position = 0;
+            size_t s;
+            size_t const unroll = 2;
+            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
+            for (s = 0; s < (size_t)tableSize; s += unroll) {
+                size_t u;
+                for (u = 0; u < unroll; ++u) {
+                    size_t const uPosition = (position + (u * step)) & tableMask;
+                    tableDecode[uPosition].symbol = spread[s + u];
+                }
+                position = (position + (unroll * step)) & tableMask;
+            }
+            assert(position == 0);
+        }
+    } else {
+        U32 const tableMask = tableSize-1;
         U32 const step = FSE_TABLESTEP(tableSize);
         U32 s, position = 0;
         for (s=0; s<maxSV1; s++) {
@@ -124,6 +173,11 @@ size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned
     return 0;
 }
 
+size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
+{
+    return FSE_buildDTable_internal(dt, normalizedCounter, maxSymbolValue, tableLog, workSpace, wkspSize);
+}
+
 
 #ifndef FSE_COMMONDEFS_ONLY
 
@@ -251,36 +305,89 @@ size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
 }
 
 
-size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog)
+size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
+{
+    return FSE_decompress_wksp_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, /* bmi2 */ 0);
+}
+
+FORCE_INLINE_TEMPLATE size_t FSE_decompress_wksp_body(
+        void* dst, size_t dstCapacity,
+        const void* cSrc, size_t cSrcSize,
+        unsigned maxLog, void* workSpace, size_t wkspSize,
+        int bmi2)
 {
     const BYTE* const istart = (const BYTE*)cSrc;
     const BYTE* ip = istart;
     short counting[FSE_MAX_SYMBOL_VALUE+1];
     unsigned tableLog;
     unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    FSE_DTable* const dtable = (FSE_DTable*)workSpace;
 
     /* normal FSE decoding mode */
-    size_t const NCountLength = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    size_t const NCountLength = FSE_readNCount_bmi2(counting, &maxSymbolValue, &tableLog, istart, cSrcSize, bmi2);
     if (FSE_isError(NCountLength)) return NCountLength;
-    /* if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong); */  /* too small input size; supposed to be already checked in NCountLength, only remaining case : NCountLength==cSrcSize */
     if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
+    assert(NCountLength <= cSrcSize);
     ip += NCountLength;
     cSrcSize -= NCountLength;
 
-    CHECK_F( FSE_buildDTable (workSpace, counting, maxSymbolValue, tableLog) );
+    if (FSE_DECOMPRESS_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(tableLog_tooLarge);
+    workSpace = dtable + FSE_DTABLE_SIZE_U32(tableLog);
+    wkspSize -= FSE_DTABLE_SIZE(tableLog);
+
+    CHECK_F( FSE_buildDTable_internal(dtable, counting, maxSymbolValue, tableLog, workSpace, wkspSize) );
+
+    {
+        const void* ptr = dtable;
+        const FSE_DTableHeader* DTableH = (const FSE_DTableHeader*)ptr;
+        const U32 fastMode = DTableH->fastMode;
+
+        /* select fast mode (static) */
+        if (fastMode) return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, dtable, 1);
+        return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, dtable, 0);
+    }
+}
+
+/* Avoids the FORCE_INLINE of the _body() function. */
+static size_t FSE_decompress_wksp_body_default(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
+{
+    return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 0);
+}
+
+#if DYNAMIC_BMI2
+TARGET_ATTRIBUTE("bmi2") static size_t FSE_decompress_wksp_body_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
+{
+    return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 1);
+}
+#endif
 
-    return FSE_decompress_usingDTable (dst, dstCapacity, ip, cSrcSize, workSpace);   /* always return, even if it is an error code */
+size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return FSE_decompress_wksp_body_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
+    }
+#endif
+    (void)bmi2;
+    return FSE_decompress_wksp_body_default(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
 }
 
 
 typedef FSE_DTable DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
 
+#ifndef ZSTD_NO_UNUSED_FUNCTIONS
+size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog) {
+    U32 wksp[FSE_BUILD_DTABLE_WKSP_SIZE_U32(FSE_TABLELOG_ABSOLUTE_MAX, FSE_MAX_SYMBOL_VALUE)];
+    return FSE_buildDTable_wksp(dt, normalizedCounter, maxSymbolValue, tableLog, wksp, sizeof(wksp));
+}
+
 size_t FSE_decompress(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize)
 {
-    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
-    return FSE_decompress_wksp(dst, dstCapacity, cSrc, cSrcSize, dt, FSE_MAX_TABLELOG);
+    /* Static analyzer seems unable to understand this table will be properly initialized later */
+    U32 wksp[FSE_DECOMPRESS_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
+    return FSE_decompress_wksp(dst, dstCapacity, cSrc, cSrcSize, FSE_MAX_TABLELOG, wksp, sizeof(wksp));
 }
-
+#endif
 
 
 #endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/lib/common/huf.h b/lib/common/huf.h
index ef432685dac8..1afef90c7ca1 100644
--- a/lib/common/huf.h
+++ b/lib/common/huf.h
@@ -20,7 +20,7 @@ extern "C" {
 #define HUF_H_298734234
 
 /* *** Dependencies *** */
-#include <stddef.h>    /* size_t */
+#include "zstd_deps.h"    /* size_t */
 
 
 /* *** library symbols visibility *** */
@@ -111,6 +111,8 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
 
 /* *** Dependencies *** */
 #include "mem.h"   /* U32 */
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
 
 
 /* *** Constants *** */
@@ -133,12 +135,16 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
 #define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
 
 /* static allocation of HUF's Compression Table */
+/* this is a private definition, just exposed for allocation and strict aliasing purpose. never EVER access its members directly */
+struct HUF_CElt_s {
+  U16  val;
+  BYTE nbBits;
+};   /* typedef'd to HUF_CElt */
+typedef struct HUF_CElt_s HUF_CElt;   /* consider it an incomplete type */
 #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[HUF_CTABLE_SIZE_U32(maxSymbolValue)]; \
-    void* name##hv = &(name##hb); \
-    HUF_CElt* name = (HUF_CElt*)(name##hv)   /* no final ; */
+    HUF_CElt name[HUF_CTABLE_SIZE_U32(maxSymbolValue)] /* no final ; */
 
 /* static allocation of HUF's DTable */
 typedef U32 HUF_DTable;
@@ -184,7 +190,6 @@ size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
  *  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);   /* @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);
@@ -226,6 +231,19 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
                      U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
                      const void* src, size_t srcSize);
 
+/*! HUF_readStats_wksp() :
+ * Same as HUF_readStats() but takes an external workspace which must be
+ * 4-byte aligned and its size must be >= HUF_READ_STATS_WORKSPACE_SIZE.
+ * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
+ */
+#define HUF_READ_STATS_WORKSPACE_SIZE_U32 FSE_DECOMPRESS_WKSP_SIZE_U32(6, HUF_TABLELOG_MAX-1)
+#define HUF_READ_STATS_WORKSPACE_SIZE (HUF_READ_STATS_WORKSPACE_SIZE_U32 * sizeof(unsigned))
+size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
+                          U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
+                          const void* src, size_t srcSize,
+                          void* workspace, size_t wkspSize,
+                          int bmi2);
+
 /** HUF_readCTable() :
  *  Loading a CTable saved with HUF_writeCTable() */
 size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);
@@ -332,6 +350,9 @@ size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstS
 #endif
 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);
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2);
+#endif
 
 #endif /* HUF_STATIC_LINKING_ONLY */
 
diff --git a/lib/common/mem.h b/lib/common/mem.h
index 89c8aea7d221..4728ef767bf0 100644
--- a/lib/common/mem.h
+++ b/lib/common/mem.h
@@ -18,8 +18,10 @@ extern "C" {
 /*-****************************************
 *  Dependencies
 ******************************************/
-#include <stddef.h>     /* size_t, ptrdiff_t */
-#include <string.h>     /* memcpy */
+#include <stddef.h>  /* size_t, ptrdiff_t */
+#include "compiler.h"  /* __has_builtin */
+#include "debug.h"  /* DEBUG_STATIC_ASSERT */
+#include "zstd_deps.h"  /* ZSTD_memcpy */
 
 
 /*-****************************************
@@ -39,93 +41,15 @@ extern "C" {
 #  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
 #endif
 
-#ifndef __has_builtin
-#  define __has_builtin(x) 0  /* compat. with non-clang compilers */
-#endif
-
-/* code only tested on 32 and 64 bits systems */
-#define MEM_STATIC_ASSERT(c)   { enum { MEM_static_assert = 1/(int)(!!(c)) }; }
-MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
-
-/* detects whether we are being compiled under msan */
-#if defined (__has_feature)
-#  if __has_feature(memory_sanitizer)
-#    define MEMORY_SANITIZER 1
-#  endif
-#endif
-
-#if defined (MEMORY_SANITIZER)
-/* Not all platforms that support msan provide sanitizers/msan_interface.h.
- * We therefore declare the functions we need ourselves, rather than trying to
- * include the header file... */
-
-#include <stdint.h> /* intptr_t */
-
-/* Make memory region fully initialized (without changing its contents). */
-void __msan_unpoison(const volatile void *a, size_t size);
-
-/* Make memory region fully uninitialized (without changing its contents).
-   This is a legacy interface that does not update origin information. Use
-   __msan_allocated_memory() instead. */
-void __msan_poison(const volatile void *a, size_t size);
-
-/* Returns the offset of the first (at least partially) poisoned byte in the
-   memory range, or -1 if the whole range is good. */
-intptr_t __msan_test_shadow(const volatile void *x, size_t size);
-#endif
-
-/* detects whether we are being compiled under asan */
-#if defined (__has_feature)
-#  if __has_feature(address_sanitizer)
-#    define ADDRESS_SANITIZER 1
-#  endif
-#elif defined(__SANITIZE_ADDRESS__)
-#  define ADDRESS_SANITIZER 1
-#endif
-
-#if defined (ADDRESS_SANITIZER)
-/* Not all platforms that support asan provide sanitizers/asan_interface.h.
- * We therefore declare the functions we need ourselves, rather than trying to
- * include the header file... */
-
-/**
- * Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable.
- *
- * This memory must be previously allocated by your program. Instrumented
- * code is forbidden from accessing addresses in this region until it is
- * unpoisoned. This function is not guaranteed to poison the entire region -
- * it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan
- * alignment restrictions.
- *
- * \note This function is not thread-safe because no two threads can poison or
- * unpoison memory in the same memory region simultaneously.
- *
- * \param addr Start of memory region.
- * \param size Size of memory region. */
-void __asan_poison_memory_region(void const volatile *addr, size_t size);
-
-/**
- * Marks a memory region (<c>[addr, addr+size)</c>) as addressable.
- *
- * This memory must be previously allocated by your program. Accessing
- * addresses in this region is allowed until this region is poisoned again.
- * This function could unpoison a super-region of <c>[addr, addr+size)</c> due
- * to ASan alignment restrictions.
- *
- * \note This function is not thread-safe because no two threads can
- * poison or unpoison memory in the same memory region simultaneously.
- *
- * \param addr Start of memory region.
- * \param size Size of memory region. */
-void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
-#endif
-
-
 /*-**************************************************************
 *  Basic Types
 *****************************************************************/
 #if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
-# include <stdint.h>
+#  if defined(_AIX)
+#    include <inttypes.h>
+#  else
+#    include <stdint.h> /* intptr_t */
+#  endif
   typedef   uint8_t BYTE;
   typedef  uint16_t U16;
   typedef   int16_t S16;
@@ -157,7 +81,53 @@ void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
 
 
 /*-**************************************************************
-*  Memory I/O
+*  Memory I/O API
+*****************************************************************/
+/*=== Static platform detection ===*/
+MEM_STATIC unsigned MEM_32bits(void);
+MEM_STATIC unsigned MEM_64bits(void);
+MEM_STATIC unsigned MEM_isLittleEndian(void);
+
+/*=== Native unaligned read/write ===*/
+MEM_STATIC U16 MEM_read16(const void* memPtr);
+MEM_STATIC U32 MEM_read32(const void* memPtr);
+MEM_STATIC U64 MEM_read64(const void* memPtr);
+MEM_STATIC size_t MEM_readST(const void* memPtr);
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value);
+MEM_STATIC void MEM_write32(void* memPtr, U32 value);
+MEM_STATIC void MEM_write64(void* memPtr, U64 value);
+
+/*=== Little endian unaligned read/write ===*/
+MEM_STATIC U16 MEM_readLE16(const void* memPtr);
+MEM_STATIC U32 MEM_readLE24(const void* memPtr);
+MEM_STATIC U32 MEM_readLE32(const void* memPtr);
+MEM_STATIC U64 MEM_readLE64(const void* memPtr);
+MEM_STATIC size_t MEM_readLEST(const void* memPtr);
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val);
+MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val);
+MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32);
+MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64);
+MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val);
+
+/*=== Big endian unaligned read/write ===*/
+MEM_STATIC U32 MEM_readBE32(const void* memPtr);
+MEM_STATIC U64 MEM_readBE64(const void* memPtr);
+MEM_STATIC size_t MEM_readBEST(const void* memPtr);
+
+MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32);
+MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64);
+MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val);
+
+/*=== Byteswap ===*/
+MEM_STATIC U32 MEM_swap32(U32 in);
+MEM_STATIC U64 MEM_swap64(U64 in);
+MEM_STATIC size_t MEM_swapST(size_t in);
+
+
+/*-**************************************************************
+*  Memory I/O Implementation
 *****************************************************************/
 /* MEM_FORCE_MEMORY_ACCESS :
  * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
@@ -236,37 +206,37 @@ MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign64*)memPtr)->v =
 
 MEM_STATIC U16 MEM_read16(const void* memPtr)
 {
-    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+    U16 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
 }
 
 MEM_STATIC U32 MEM_read32(const void* memPtr)
 {
-    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+    U32 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
 }
 
 MEM_STATIC U64 MEM_read64(const void* memPtr)
 {
-    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+    U64 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
 }
 
 MEM_STATIC size_t MEM_readST(const void* memPtr)
 {
-    size_t val; memcpy(&val, memPtr, sizeof(val)); return val;
+    size_t val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
 }
 
 MEM_STATIC void MEM_write16(void* memPtr, U16 value)
 {
-    memcpy(memPtr, &value, sizeof(value));
+    ZSTD_memcpy(memPtr, &value, sizeof(value));
 }
 
 MEM_STATIC void MEM_write32(void* memPtr, U32 value)
 {
-    memcpy(memPtr, &value, sizeof(value));
+    ZSTD_memcpy(memPtr, &value, sizeof(value));
 }
 
 MEM_STATIC void MEM_write64(void* memPtr, U64 value)
 {
-    memcpy(memPtr, &value, sizeof(value));
+    ZSTD_memcpy(memPtr, &value, sizeof(value));
 }
 
 #endif /* MEM_FORCE_MEMORY_ACCESS */
@@ -445,6 +415,9 @@ MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val)
         MEM_writeBE64(memPtr, (U64)val);
 }
 
+/* code only tested on 32 and 64 bits systems */
+MEM_STATIC void MEM_check(void) { DEBUG_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
+
 
 #if defined (__cplusplus)
 }
diff --git a/lib/common/pool.c b/lib/common/pool.c
index aa4b4de0d3f6..4c1b83376f4e 100644
--- a/lib/common/pool.c
+++ b/lib/common/pool.c
@@ -10,9 +10,9 @@
 
 
 /* ======   Dependencies   ======= */
-#include <stddef.h>    /* size_t */
+#include "zstd_deps.h" /* size_t */
 #include "debug.h"     /* assert */
-#include "zstd_internal.h"  /* ZSTD_malloc, ZSTD_free */
+#include "zstd_internal.h"  /* ZSTD_customMalloc, ZSTD_customFree */
 #include "pool.h"
 
 /* ======   Compiler specifics   ====== */
@@ -105,6 +105,10 @@ static void* POOL_thread(void* opaque) {
     assert(0);  /* Unreachable */
 }
 
+POOL_ctx* ZSTD_createThreadPool(size_t numThreads) {
+    return POOL_create (numThreads, 0);
+}
+
 POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
     return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
 }
@@ -115,14 +119,14 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
     /* Check parameters */
     if (!numThreads) { return NULL; }
     /* Allocate the context and zero initialize */
-    ctx = (POOL_ctx*)ZSTD_calloc(sizeof(POOL_ctx), customMem);
+    ctx = (POOL_ctx*)ZSTD_customCalloc(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*)ZSTD_malloc(ctx->queueSize * sizeof(POOL_job), customMem);
+    ctx->queue = (POOL_job*)ZSTD_customMalloc(ctx->queueSize * sizeof(POOL_job), customMem);
     ctx->queueHead = 0;
     ctx->queueTail = 0;
     ctx->numThreadsBusy = 0;
@@ -136,7 +140,7 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
     }
     ctx->shutdown = 0;
     /* Allocate space for the thread handles */
-    ctx->threads = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
+    ctx->threads = (ZSTD_pthread_t*)ZSTD_customMalloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
     ctx->threadCapacity = 0;
     ctx->customMem = customMem;
     /* Check for errors */
@@ -179,12 +183,14 @@ void POOL_free(POOL_ctx *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);
+    ZSTD_customFree(ctx->queue, ctx->customMem);
+    ZSTD_customFree(ctx->threads, ctx->customMem);
+    ZSTD_customFree(ctx, ctx->customMem);
 }
 
-
+void ZSTD_freeThreadPool (ZSTD_threadPool* pool) {
+  POOL_free (pool);
+}
 
 size_t POOL_sizeof(POOL_ctx *ctx) {
     if (ctx==NULL) return 0;  /* supports sizeof NULL */
@@ -203,11 +209,11 @@ static int POOL_resize_internal(POOL_ctx* ctx, size_t numThreads)
         return 0;
     }
     /* numThreads > threadCapacity */
-    {   ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
+    {   ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_customMalloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
         if (!threadPool) return 1;
         /* replace existing thread pool */
-        memcpy(threadPool, ctx->threads, ctx->threadCapacity * sizeof(*threadPool));
-        ZSTD_free(ctx->threads, ctx->customMem);
+        ZSTD_memcpy(threadPool, ctx->threads, ctx->threadCapacity * sizeof(*threadPool));
+        ZSTD_customFree(ctx->threads, ctx->customMem);
         ctx->threads = threadPool;
         /* Initialize additional threads */
         {   size_t threadId;
@@ -301,7 +307,7 @@ int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque)
 struct POOL_ctx_s {
     int dummy;
 };
-static POOL_ctx g_ctx;
+static POOL_ctx g_poolCtx;
 
 POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
     return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
@@ -311,11 +317,11 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customM
     (void)numThreads;
     (void)queueSize;
     (void)customMem;
-    return &g_ctx;
+    return &g_poolCtx;
 }
 
 void POOL_free(POOL_ctx* ctx) {
-    assert(!ctx || ctx == &g_ctx);
+    assert(!ctx || ctx == &g_poolCtx);
     (void)ctx;
 }
 
@@ -337,7 +343,7 @@ int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque) {
 
 size_t POOL_sizeof(POOL_ctx* ctx) {
     if (ctx==NULL) return 0;  /* supports sizeof NULL */
-    assert(ctx == &g_ctx);
+    assert(ctx == &g_poolCtx);
     return sizeof(*ctx);
 }
 
diff --git a/lib/common/pool.h b/lib/common/pool.h
index 259bafc97570..63954ca6ca4d 100644
--- a/lib/common/pool.h
+++ b/lib/common/pool.h
@@ -16,7 +16,7 @@ extern "C" {
 #endif
 
 
-#include <stddef.h>   /* size_t */
+#include "zstd_deps.h"
 #define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_customMem */
 #include "../zstd.h"
 
diff --git a/lib/common/threading.c b/lib/common/threading.c
index e2edb313ebff..92cf57c195a5 100644
--- a/lib/common/threading.c
+++ b/lib/common/threading.c
@@ -78,11 +78,12 @@ int ZSTD_pthread_join(ZSTD_pthread_t thread, void **value_ptr)
 
 #if defined(ZSTD_MULTITHREAD) && DEBUGLEVEL >= 1 && !defined(_WIN32)
 
-#include <stdlib.h>
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"
 
 int ZSTD_pthread_mutex_init(ZSTD_pthread_mutex_t* mutex, pthread_mutexattr_t const* attr)
 {
-    *mutex = (pthread_mutex_t*)malloc(sizeof(pthread_mutex_t));
+    *mutex = (pthread_mutex_t*)ZSTD_malloc(sizeof(pthread_mutex_t));
     if (!*mutex)
         return 1;
     return pthread_mutex_init(*mutex, attr);
@@ -94,14 +95,14 @@ int ZSTD_pthread_mutex_destroy(ZSTD_pthread_mutex_t* mutex)
         return 0;
     {
         int const ret = pthread_mutex_destroy(*mutex);
-        free(*mutex);
+        ZSTD_free(*mutex);
         return ret;
     }
 }
 
 int ZSTD_pthread_cond_init(ZSTD_pthread_cond_t* cond, pthread_condattr_t const* attr)
 {
-    *cond = (pthread_cond_t*)malloc(sizeof(pthread_cond_t));
+    *cond = (pthread_cond_t*)ZSTD_malloc(sizeof(pthread_cond_t));
     if (!*cond)
         return 1;
     return pthread_cond_init(*cond, attr);
@@ -113,7 +114,7 @@ int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond)
         return 0;
     {
         int const ret = pthread_cond_destroy(*cond);
-        free(*cond);
+        ZSTD_free(*cond);
         return ret;
     }
 }
diff --git a/lib/common/xxhash.c b/lib/common/xxhash.c
index 597de18fc895..e708df3c3389 100644
--- a/lib/common/xxhash.c
+++ b/lib/common/xxhash.c
@@ -77,14 +77,12 @@
 *  Includes & Memory related functions
 ***************************************/
 /* Modify the local functions below should you wish to use some other memory routines */
-/* for malloc(), free() */
-#include <stdlib.h>
-#include <stddef.h>     /* size_t */
-static void* XXH_malloc(size_t s) { return malloc(s); }
-static void  XXH_free  (void* p)  { free(p); }
-/* for memcpy() */
-#include <string.h>
-static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); }
+/* for ZSTD_malloc(), ZSTD_free() */
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"  /* size_t, ZSTD_malloc, ZSTD_free, ZSTD_memcpy */
+static void* XXH_malloc(size_t s) { return ZSTD_malloc(s); }
+static void  XXH_free  (void* p)  { ZSTD_free(p); }
+static void* XXH_memcpy(void* dest, const void* src, size_t size) { return ZSTD_memcpy(dest,src,size); }
 
 #ifndef XXH_STATIC_LINKING_ONLY
 #  define XXH_STATIC_LINKING_ONLY
@@ -95,49 +93,13 @@ static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcp
 /* *************************************
 *  Compiler Specific Options
 ***************************************/
-#if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
-#  define INLINE_KEYWORD inline
-#else
-#  define INLINE_KEYWORD
-#endif
-
-#if defined(__GNUC__) || defined(__ICCARM__)
-#  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
+#include "compiler.h"
 
 
 /* *************************************
 *  Basic Types
 ***************************************/
-#ifndef MEM_MODULE
-# define MEM_MODULE
-# if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
-#   include <stdint.h>
-    typedef uint8_t  BYTE;
-    typedef uint16_t U16;
-    typedef uint32_t U32;
-    typedef  int32_t S32;
-    typedef uint64_t U64;
-#  else
-    typedef unsigned char      BYTE;
-    typedef unsigned short     U16;
-    typedef unsigned int       U32;
-    typedef   signed int       S32;
-    typedef unsigned long long U64;   /* if your compiler doesn't support unsigned long long, replace by another 64-bit type here. Note that xxhash.h will also need to be updated. */
-#  endif
-#endif
-
+#include "mem.h"  /* BYTE, U32, U64, size_t */
 
 #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
 
@@ -163,14 +125,14 @@ static U64 XXH_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
 static U32 XXH_read32(const void* memPtr)
 {
     U32 val;
-    memcpy(&val, memPtr, sizeof(val));
+    ZSTD_memcpy(&val, memPtr, sizeof(val));
     return val;
 }
 
 static U64 XXH_read64(const void* memPtr)
 {
     U64 val;
-    memcpy(&val, memPtr, sizeof(val));
+    ZSTD_memcpy(&val, memPtr, sizeof(val));
     return val;
 }
 
@@ -307,12 +269,12 @@ XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
 ****************************/
 XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dstState, const XXH32_state_t* restrict srcState)
 {
-    memcpy(dstState, srcState, sizeof(*dstState));
+    ZSTD_memcpy(dstState, srcState, sizeof(*dstState));
 }
 
 XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dstState, const XXH64_state_t* restrict srcState)
 {
-    memcpy(dstState, srcState, sizeof(*dstState));
+    ZSTD_memcpy(dstState, srcState, sizeof(*dstState));
 }
 
 
@@ -554,12 +516,12 @@ XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
 XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed)
 {
     XXH32_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
-    memset(&state, 0, sizeof(state)-4);   /* do not write into reserved, for future removal */
+    ZSTD_memset(&state, 0, sizeof(state)-4);   /* do not write into reserved, for future removal */
     state.v1 = seed + PRIME32_1 + PRIME32_2;
     state.v2 = seed + PRIME32_2;
     state.v3 = seed + 0;
     state.v4 = seed - PRIME32_1;
-    memcpy(statePtr, &state, sizeof(state));
+    ZSTD_memcpy(statePtr, &state, sizeof(state));
     return XXH_OK;
 }
 
@@ -567,12 +529,12 @@ XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int s
 XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed)
 {
     XXH64_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
-    memset(&state, 0, sizeof(state)-8);   /* do not write into reserved, for future removal */
+    ZSTD_memset(&state, 0, sizeof(state)-8);   /* do not write into reserved, for future removal */
     state.v1 = seed + PRIME64_1 + PRIME64_2;
     state.v2 = seed + PRIME64_2;
     state.v3 = seed + 0;
     state.v4 = seed - PRIME64_1;
-    memcpy(statePtr, &state, sizeof(state));
+    ZSTD_memcpy(statePtr, &state, sizeof(state));
     return XXH_OK;
 }
 
@@ -843,14 +805,14 @@ XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t
 {
     XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
     if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
-    memcpy(dst, &hash, sizeof(*dst));
+    ZSTD_memcpy(dst, &hash, sizeof(*dst));
 }
 
 XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
 {
     XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
     if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
-    memcpy(dst, &hash, sizeof(*dst));
+    ZSTD_memcpy(dst, &hash, sizeof(*dst));
 }
 
 XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
diff --git a/lib/common/xxhash.h b/lib/common/xxhash.h
index 4207eba8328f..eceb55d5e071 100644
--- a/lib/common/xxhash.h
+++ b/lib/common/xxhash.h
@@ -55,7 +55,7 @@ extern "C" {
 /* ****************************
 *  Definitions
 ******************************/
-#include <stddef.h>   /* size_t */
+#include "zstd_deps.h"
 typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
 
 
diff --git a/lib/common/zstd_common.c b/lib/common/zstd_common.c
index 91fe3323a5ba..939e9f08fad7 100644
--- a/lib/common/zstd_common.c
+++ b/lib/common/zstd_common.c
@@ -13,8 +13,8 @@
 /*-*************************************
 *  Dependencies
 ***************************************/
-#include <stdlib.h>      /* malloc, calloc, free */
-#include <string.h>      /* memset */
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"   /* ZSTD_malloc, ZSTD_calloc, ZSTD_free, ZSTD_memset */
 #include "error_private.h"
 #include "zstd_internal.h"
 
@@ -53,31 +53,31 @@ const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString
 /*=**************************************************************
 *  Custom allocator
 ****************************************************************/
-void* ZSTD_malloc(size_t size, ZSTD_customMem customMem)
+void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem)
 {
     if (customMem.customAlloc)
         return customMem.customAlloc(customMem.opaque, size);
-    return malloc(size);
+    return ZSTD_malloc(size);
 }
 
-void* ZSTD_calloc(size_t size, ZSTD_customMem customMem)
+void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem)
 {
     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);
+        ZSTD_memset(ptr, 0, size);
         return ptr;
     }
-    return calloc(1, size);
+    return ZSTD_calloc(1, size);
 }
 
-void ZSTD_free(void* ptr, ZSTD_customMem customMem)
+void ZSTD_customFree(void* ptr, ZSTD_customMem customMem)
 {
     if (ptr!=NULL) {
         if (customMem.customFree)
             customMem.customFree(customMem.opaque, ptr);
         else
-            free(ptr);
+            ZSTD_free(ptr);
     }
 }
diff --git a/lib/common/zstd_deps.h b/lib/common/zstd_deps.h
new file mode 100644
index 000000000000..0fb8b7818b8f
--- /dev/null
+++ b/lib/common/zstd_deps.h
@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 2016-2020, 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 file provides common libc dependencies that zstd requires.
+ * The purpose is to allow replacing this file with a custom implementation
+ * to compile zstd without libc support.
+ */
+
+/* Need:
+ * NULL
+ * INT_MAX
+ * UINT_MAX
+ * ZSTD_memcpy()
+ * ZSTD_memset()
+ * ZSTD_memmove()
+ */
+#ifndef ZSTD_DEPS_COMMON
+#define ZSTD_DEPS_COMMON
+
+#include <limits.h>
+#include <stddef.h>
+#include <string.h>
+
+#if defined(__GNUC__) && __GNUC__ >= 4
+# define ZSTD_memcpy(d,s,l) __builtin_memcpy((d),(s),(l))
+# define ZSTD_memmove(d,s,l) __builtin_memmove((d),(s),(l))
+# define ZSTD_memset(p,v,l) __builtin_memset((p),(v),(l))
+#else
+# define ZSTD_memcpy(d,s,l) memcpy((d),(s),(l))
+# define ZSTD_memmove(d,s,l) memmove((d),(s),(l))
+# define ZSTD_memset(p,v,l) memset((p),(v),(l))
+#endif
+
+#endif /* ZSTD_DEPS_COMMON */
+
+/* Need:
+ * ZSTD_malloc()
+ * ZSTD_free()
+ * ZSTD_calloc()
+ */
+#ifdef ZSTD_DEPS_NEED_MALLOC
+#ifndef ZSTD_DEPS_MALLOC
+#define ZSTD_DEPS_MALLOC
+
+#include <stdlib.h>
+
+#define ZSTD_malloc(s) malloc(s)
+#define ZSTD_calloc(n,s) calloc((n), (s))
+#define ZSTD_free(p) free((p))
+
+#endif /* ZSTD_DEPS_MALLOC */
+#endif /* ZSTD_DEPS_NEED_MALLOC */
+
+/*
+ * Provides 64-bit math support.
+ * Need:
+ * U64 ZSTD_div64(U64 dividend, U32 divisor)
+ */
+#ifdef ZSTD_DEPS_NEED_MATH64
+#ifndef ZSTD_DEPS_MATH64
+#define ZSTD_DEPS_MATH64
+
+#define ZSTD_div64(dividend, divisor) ((dividend) / (divisor))
+
+#endif /* ZSTD_DEPS_MATH64 */
+#endif /* ZSTD_DEPS_NEED_MATH64 */
+
+/* Need:
+ * assert()
+ */
+#ifdef ZSTD_DEPS_NEED_ASSERT
+#ifndef ZSTD_DEPS_ASSERT
+#define ZSTD_DEPS_ASSERT
+
+#include <assert.h>
+
+#endif /* ZSTD_DEPS_ASSERT */
+#endif /* ZSTD_DEPS_NEED_ASSERT */
+
+/* Need:
+ * ZSTD_DEBUG_PRINT()
+ */
+#ifdef ZSTD_DEPS_NEED_IO
+#ifndef ZSTD_DEPS_IO
+#define ZSTD_DEPS_IO
+
+#include <stdio.h>
+#define ZSTD_DEBUG_PRINT(...) fprintf(stderr, __VA_ARGS__)
+
+#endif /* ZSTD_DEPS_IO */
+#endif /* ZSTD_DEPS_NEED_IO */
+
+/* Only requested when <stdint.h> is known to be present.
+ * Need:
+ * intptr_t
+ */
+#ifdef ZSTD_DEPS_NEED_STDINT
+#ifndef ZSTD_DEPS_STDINT
+#define ZSTD_DEPS_STDINT
+
+#include <stdint.h>
+
+#endif /* ZSTD_DEPS_STDINT */
+#endif /* ZSTD_DEPS_NEED_STDINT */
diff --git a/lib/common/zstd_errors.h b/lib/common/zstd_errors.h
index 998398e7e57f..6d0d0030043e 100644
--- a/lib/common/zstd_errors.h
+++ b/lib/common/zstd_errors.h
@@ -77,6 +77,7 @@ typedef enum {
   ZSTD_error_frameIndex_tooLarge = 100,
   ZSTD_error_seekableIO          = 102,
   ZSTD_error_dstBuffer_wrong     = 104,
+  ZSTD_error_srcBuffer_wrong     = 105,
   ZSTD_error_maxCode = 120  /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
 } ZSTD_ErrorCode;
 
diff --git a/lib/common/zstd_internal.h b/lib/common/zstd_internal.h
index 3bc7e55a0a97..0991f20a086e 100644
--- a/lib/common/zstd_internal.h
+++ b/lib/common/zstd_internal.h
@@ -19,7 +19,7 @@
 /*-*************************************
 *  Dependencies
 ***************************************/
-#ifdef __aarch64__
+#if !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON)
 #include <arm_neon.h>
 #endif
 #include "compiler.h"
@@ -139,7 +139,7 @@ void _force_has_format_string(const char *format, ...) {
 
 #define ZSTD_REP_NUM      3                 /* number of repcodes */
 #define ZSTD_REP_MOVE     (ZSTD_REP_NUM-1)
-static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
+static UNUSED_ATTR const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
 
 #define KB *(1 <<10)
 #define MB *(1 <<20)
@@ -153,13 +153,13 @@ static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
 #define BIT0   1
 
 #define ZSTD_WINDOWLOG_ABSOLUTEMIN 10
-static const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
-static const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
+static UNUSED_ATTR const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
+static UNUSED_ATTR const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
 
 #define ZSTD_FRAMEIDSIZE 4   /* 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;
+static UNUSED_ATTR const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
 typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
 
 #define ZSTD_FRAMECHECKSUMSIZE 4
@@ -186,61 +186,75 @@ typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingTy
 #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,
-                                     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,
-                                            -1,-1,-1,-1 };
+#define ZSTD_MAX_HUF_HEADER_SIZE 128 /* header + <= 127 byte tree description */
+/* Each table cannot take more than #symbols * FSELog bits */
+#define ZSTD_MAX_FSE_HEADERS_SIZE (((MaxML + 1) * MLFSELog + (MaxLL + 1) * LLFSELog + (MaxOff + 1) * OffFSELog + 7) / 8)
+
+static UNUSED_ATTR 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 UNUSED_ATTR 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,
-                                     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,
-                                            -1,-1,-1,-1,-1 };
+static UNUSED_ATTR const U32 LL_defaultNormLog = LL_DEFAULTNORMLOG;
+
+static UNUSED_ATTR 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 UNUSED_ATTR 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[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 };
+static UNUSED_ATTR const U32 ML_defaultNormLog = ML_DEFAULTNORMLOG;
+
+static UNUSED_ATTR 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;
+static UNUSED_ATTR const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
 
 
 /*-*******************************************
 *  Shared functions to include for inlining
 *********************************************/
 static void ZSTD_copy8(void* dst, const void* src) {
-#ifdef __aarch64__
+#if !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON)
     vst1_u8((uint8_t*)dst, vld1_u8((const uint8_t*)src));
 #else
-    memcpy(dst, src, 8);
+    ZSTD_memcpy(dst, src, 8);
 #endif
 }
 
 #define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
 static void ZSTD_copy16(void* dst, const void* src) {
-#ifdef __aarch64__
+#if !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON)
     vst1q_u8((uint8_t*)dst, vld1q_u8((const uint8_t*)src));
 #else
-    memcpy(dst, src, 16);
+    ZSTD_memcpy(dst, src, 16);
 #endif
 }
 #define COPY16(d,s) { ZSTD_copy16(d,s); d+=16; s+=16; }
@@ -255,13 +269,13 @@ typedef enum {
 } ZSTD_overlap_e;
 
 /*! ZSTD_wildcopy() :
- *  Custom version of memcpy(), can over read/write up to WILDCOPY_OVERLENGTH bytes (if length==0)
+ *  Custom version of ZSTD_memcpy(), can over read/write up to WILDCOPY_OVERLENGTH bytes (if length==0)
  *  @param ovtype controls the overlap detection
  *         - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
  *         - ZSTD_overlap_src_before_dst: The src and dst may overlap, but they MUST be at least 8 bytes apart.
  *           The src buffer must be before the dst buffer.
  */
-MEM_STATIC FORCE_INLINE_ATTR 
+MEM_STATIC FORCE_INLINE_ATTR
 void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e const ovtype)
 {
     ptrdiff_t diff = (BYTE*)dst - (const BYTE*)src;
@@ -284,14 +298,16 @@ void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e
          * one COPY16() in the first call. Then, do two calls per loop since
          * at that point it is more likely to have a high trip count.
          */
-#ifndef __aarch64__
+#ifdef __aarch64__
         do {
             COPY16(op, ip);
         }
         while (op < oend);
 #else
-        COPY16(op, ip);
-        if (op >= oend) return;
+        ZSTD_copy16(op, ip);
+        if (16 >= length) return;
+        op += 16;
+        ip += 16;
         do {
             COPY16(op, ip);
             COPY16(op, ip);
@@ -305,7 +321,7 @@ MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src,
 {
     size_t const length = MIN(dstCapacity, srcSize);
     if (length > 0) {
-        memcpy(dst, src, length);
+        ZSTD_memcpy(dst, src, length);
     }
     return length;
 }
@@ -320,28 +336,39 @@ MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src,
  * In which case, resize it down to free some memory */
 #define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128
 
+/* Controls whether the input/output buffer is buffered or stable. */
+typedef enum {
+    ZSTD_bm_buffered = 0,  /* Buffer the input/output */
+    ZSTD_bm_stable = 1     /* ZSTD_inBuffer/ZSTD_outBuffer is stable */
+} ZSTD_bufferMode_e;
+
 
 /*-*******************************************
 *  Private declarations
 *********************************************/
 typedef struct seqDef_s {
-    U32 offset;
+    U32 offset;         /* Offset code of the sequence */
     U16 litLength;
     U16 matchLength;
 } seqDef;
 
 typedef struct {
     seqDef* sequencesStart;
-    seqDef* sequences;
+    seqDef* sequences;      /* ptr to end of sequences */
     BYTE* litStart;
-    BYTE* lit;
+    BYTE* lit;              /* ptr to end of literals */
     BYTE* llCode;
     BYTE* mlCode;
     BYTE* ofCode;
     size_t maxNbSeq;
     size_t maxNbLit;
-    U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
-    U32   longLengthPos;
+
+    /* longLengthPos and longLengthID to allow us to represent either a single litLength or matchLength
+     * in the seqStore that has a value larger than U16 (if it exists). To do so, we increment
+     * the existing value of the litLength or matchLength by 0x10000. 
+     */
+    U32   longLengthID;   /* 0 == no longLength; 1 == Represent the long literal; 2 == Represent the long match; */
+    U32   longLengthPos;  /* Index of the sequence to apply long length modification to */
 } seqStore_t;
 
 typedef struct {
@@ -384,9 +411,9 @@ const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);   /* compress & dictBu
 void ZSTD_seqToCodes(const seqStore_t* seqStorePtr);   /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
 
 /* custom memory allocation functions */
-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);
+void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem);
+void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem);
+void ZSTD_customFree(void* ptr, ZSTD_customMem customMem);
 
 
 MEM_STATIC U32 ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus */
@@ -394,8 +421,12 @@ MEM_STATIC U32 ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus
     assert(val != 0);
     {
 #   if defined(_MSC_VER)   /* Visual */
-        unsigned long r=0;
-        return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#       if STATIC_BMI2 == 1
+            return _lzcnt_u32(val)^31;
+#       else
+            unsigned long r=0;
+            return _BitScanReverse(&r, val) ? (unsigned)r : 0;
+#       endif
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* GCC Intrinsic */
         return __builtin_clz (val) ^ 31;
 #   elif defined(__ICCARM__)    /* IAR Intrinsic */
diff --git a/lib/compress/fse_compress.c b/lib/compress/fse_compress.c
index a42759814fdd..304a82b3cc47 100644
--- a/lib/compress/fse_compress.c
+++ b/lib/compress/fse_compress.c
@@ -15,8 +15,6 @@
 /* **************************************************************
 *  Includes
 ****************************************************************/
-#include <stdlib.h>     /* malloc, free, qsort */
-#include <string.h>     /* memcpy, memset */
 #include "../common/compiler.h"
 #include "../common/mem.h"        /* U32, U16, etc. */
 #include "../common/debug.h"      /* assert, DEBUGLOG */
@@ -25,6 +23,9 @@
 #define FSE_STATIC_LINKING_ONLY
 #include "../common/fse.h"
 #include "../common/error_private.h"
+#define ZSTD_DEPS_NEED_MALLOC
+#define ZSTD_DEPS_NEED_MATH64
+#include "../common/zstd_deps.h"  /* ZSTD_malloc, ZSTD_free, ZSTD_memcpy, ZSTD_memset */
 
 
 /* **************************************************************
@@ -74,13 +75,15 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
     void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
     FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
     U32 const step = FSE_TABLESTEP(tableSize);
-    U32 cumul[FSE_MAX_SYMBOL_VALUE+2];
 
-    FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)workSpace;
+    U32* cumul = (U32*)workSpace;
+    FSE_FUNCTION_TYPE* tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSymbolValue + 2));
+
     U32 highThreshold = tableSize-1;
 
+    if ((size_t)workSpace & 3) return ERROR(GENERIC); /* Must be 4 byte aligned */
+    if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
     /* CTable header */
-    if (((size_t)1 << tableLog) * sizeof(FSE_FUNCTION_TYPE) > wkspSize) return ERROR(tableLog_tooLarge);
     tableU16[-2] = (U16) tableLog;
     tableU16[-1] = (U16) maxSymbolValue;
     assert(tableLog < 16);   /* required for threshold strategy to work */
@@ -89,7 +92,7 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
      * http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
 
      #ifdef __clang_analyzer__
-     memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
+     ZSTD_memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
      #endif
 
     /* symbol start positions */
@@ -168,12 +171,13 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
     return 0;
 }
 
-
+#ifndef ZSTD_NO_UNUSED_FUNCTIONS
 size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
 {
     FSE_FUNCTION_TYPE tableSymbol[FSE_MAX_TABLESIZE];   /* memset() is not necessary, even if static analyzer complain about it */
     return FSE_buildCTable_wksp(ct, normalizedCounter, maxSymbolValue, tableLog, tableSymbol, sizeof(tableSymbol));
 }
+#endif
 
 
 
@@ -307,10 +311,10 @@ FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
     size_t size;
     if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
     size = FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
-    return (FSE_CTable*)malloc(size);
+    return (FSE_CTable*)ZSTD_malloc(size);
 }
 
-void FSE_freeCTable (FSE_CTable* ct) { free(ct); }
+void FSE_freeCTable (FSE_CTable* ct) { ZSTD_free(ct); }
 
 /* provides the minimum logSize to safely represent a distribution */
 static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
@@ -341,11 +345,10 @@ unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
     return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
 }
 
-
 /* Secondary normalization method.
    To be used when primary method fails. */
 
-static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue)
+static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue, short lowProbCount)
 {
     short const NOT_YET_ASSIGNED = -2;
     U32 s;
@@ -362,7 +365,7 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
             continue;
         }
         if (count[s] <= lowThreshold) {
-            norm[s] = -1;
+            norm[s] = lowProbCount;
             distributed++;
             total -= count[s];
             continue;
@@ -414,7 +417,7 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
 
     {   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 const rStep = ZSTD_div64((((U64)1<<vStepLog) * ToDistribute) + mid, (U32)total);   /* scale on remaining */
         U64 tmpTotal = mid;
         for (s=0; s<=maxSymbolValue; s++) {
             if (norm[s]==NOT_YET_ASSIGNED) {
@@ -431,10 +434,9 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
     return 0;
 }
 
-
 size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
                            const unsigned* count, size_t total,
-                           unsigned maxSymbolValue)
+                           unsigned maxSymbolValue, unsigned useLowProbCount)
 {
     /* Sanity checks */
     if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
@@ -443,8 +445,9 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
     if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC);   /* Too small tableLog, compression potentially impossible */
 
     {   static U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
+        short const lowProbCount = useLowProbCount ? -1 : 1;
         U64 const scale = 62 - tableLog;
-        U64 const step = ((U64)1<<62) / total;   /* <== here, one division ! */
+        U64 const step = ZSTD_div64((U64)1<<62, (U32)total);   /* <== here, one division ! */
         U64 const vStep = 1ULL<<(scale-20);
         int stillToDistribute = 1<<tableLog;
         unsigned s;
@@ -456,7 +459,7 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
             if (count[s] == total) return 0;   /* rle special case */
             if (count[s] == 0) { normalizedCounter[s]=0; continue; }
             if (count[s] <= lowThreshold) {
-                normalizedCounter[s] = -1;
+                normalizedCounter[s] = lowProbCount;
                 stillToDistribute--;
             } else {
                 short proba = (short)((count[s]*step) >> scale);
@@ -470,7 +473,7 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
         }   }
         if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
             /* corner case, need another normalization method */
-            size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue);
+            size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue, lowProbCount);
             if (FSE_isError(errorCode)) return errorCode;
         }
         else normalizedCounter[largest] += (short)stillToDistribute;
@@ -625,6 +628,7 @@ size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
 
 size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
 
+#ifndef ZSTD_NO_UNUSED_FUNCTIONS
 /* FSE_compress_wksp() :
  * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
  * `wkspSize` size must be `(1<<tableLog)`.
@@ -643,7 +647,7 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
     size_t const scratchBufferSize = wkspSize - (CTableSize * sizeof(FSE_CTable));
 
     /* init conditions */
-    if (wkspSize < FSE_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
+    if (wkspSize < FSE_COMPRESS_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
     if (srcSize <= 1) return 0;  /* Not compressible */
     if (!maxSymbolValue) maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
     if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
@@ -656,7 +660,7 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
     }
 
     tableLog = FSE_optimalTableLog(tableLog, srcSize, maxSymbolValue);
-    CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue) );
+    CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue, /* useLowProbCount */ srcSize >= 2048) );
 
     /* Write table description header */
     {   CHECK_V_F(nc_err, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
@@ -678,13 +682,16 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
 
 typedef struct {
     FSE_CTable CTable_max[FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
-    BYTE scratchBuffer[1 << FSE_MAX_TABLELOG];
+    union {
+      U32 hist_wksp[HIST_WKSP_SIZE_U32];
+      BYTE scratchBuffer[1 << FSE_MAX_TABLELOG];
+    } workspace;
 } fseWkspMax_t;
 
 size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
 {
     fseWkspMax_t scratchBuffer;
-    DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
+    DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_COMPRESS_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
     return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
 }
@@ -693,6 +700,6 @@ size_t FSE_compress (void* dst, size_t dstCapacity, const void* src, size_t srcS
 {
     return FSE_compress2(dst, dstCapacity, src, srcSize, FSE_MAX_SYMBOL_VALUE, FSE_DEFAULT_TABLELOG);
 }
-
+#endif
 
 #endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/lib/compress/hist.c b/lib/compress/hist.c
index 61e08c7968be..a9659d11ad0c 100644
--- a/lib/compress/hist.c
+++ b/lib/compress/hist.c
@@ -34,7 +34,7 @@ unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
     unsigned maxSymbolValue = *maxSymbolValuePtr;
     unsigned largestCount=0;
 
-    memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
+    ZSTD_memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
     if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
 
     while (ip<end) {
@@ -60,9 +60,9 @@ typedef enum { trustInput, checkMaxSymbolValue } HIST_checkInput_e;
  * this design makes better use of OoO cpus,
  * and is noticeably faster when some values are heavily repeated.
  * But it needs some additional workspace for intermediate tables.
- * `workSpace` size must be a table of size >= HIST_WKSP_SIZE_U32.
+ * `workSpace` must be a U32 table of size >= HIST_WKSP_SIZE_U32.
  * @return : largest histogram frequency,
- *           or an error code (notably when histogram would be larger than *maxSymbolValuePtr). */
+ *           or an error code (notably when histogram's alphabet is larger than *maxSymbolValuePtr) */
 static size_t HIST_count_parallel_wksp(
                                 unsigned* count, unsigned* maxSymbolValuePtr,
                                 const void* source, size_t sourceSize,
@@ -71,22 +71,21 @@ static size_t HIST_count_parallel_wksp(
 {
     const BYTE* ip = (const BYTE*)source;
     const BYTE* const iend = ip+sourceSize;
-    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    size_t const countSize = (*maxSymbolValuePtr + 1) * sizeof(*count);
     unsigned max=0;
     U32* const Counting1 = workSpace;
     U32* const Counting2 = Counting1 + 256;
     U32* const Counting3 = Counting2 + 256;
     U32* const Counting4 = Counting3 + 256;
 
-    memset(workSpace, 0, 4*256*sizeof(unsigned));
-
     /* safety checks */
+    assert(*maxSymbolValuePtr <= 255);
     if (!sourceSize) {
-        memset(count, 0, maxSymbolValue + 1);
+        ZSTD_memset(count, 0, countSize);
         *maxSymbolValuePtr = 0;
         return 0;
     }
-    if (!maxSymbolValue) maxSymbolValue = 255;            /* 0 == default */
+    ZSTD_memset(workSpace, 0, 4*256*sizeof(unsigned));
 
     /* by stripes of 16 bytes */
     {   U32 cached = MEM_read32(ip); ip += 4;
@@ -118,21 +117,18 @@ static size_t HIST_count_parallel_wksp(
     /* finish last symbols */
     while (ip<iend) Counting1[*ip++]++;
 
-    if (check) {   /* verify stats will fit into destination table */
-        U32 s; for (s=255; s>maxSymbolValue; s--) {
-            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
-            if (Counting1[s]) return ERROR(maxSymbolValue_tooSmall);
-    }   }
-
     {   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];
+        for (s=0; s<256; s++) {
+            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
+            if (Counting1[s] > max) max = Counting1[s];
     }   }
 
-    while (!count[maxSymbolValue]) maxSymbolValue--;
-    *maxSymbolValuePtr = maxSymbolValue;
+    {   unsigned maxSymbolValue = 255;
+        while (!Counting1[maxSymbolValue]) maxSymbolValue--;
+        if (check && maxSymbolValue > *maxSymbolValuePtr) return ERROR(maxSymbolValue_tooSmall);
+        *maxSymbolValuePtr = maxSymbolValue;
+        ZSTD_memmove(count, Counting1, countSize);   /* in case count & Counting1 are overlapping */
+    }
     return (size_t)max;
 }
 
@@ -152,14 +148,6 @@ size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
     return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, trustInput, (U32*)workSpace);
 }
 
-/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
-size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
-                     const void* source, size_t sourceSize)
-{
-    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
-    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters, sizeof(tmpCounters));
-}
-
 /* HIST_count_wksp() :
  * Same as HIST_count(), but using an externally provided scratch buffer.
  * `workSpace` size must be table of >= HIST_WKSP_SIZE_U32 unsigned */
@@ -175,9 +163,19 @@ size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
     return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace, workSpaceSize);
 }
 
+#ifndef ZSTD_NO_UNUSED_FUNCTIONS
+/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
+size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                     const void* source, size_t sourceSize)
+{
+    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
+    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters, sizeof(tmpCounters));
+}
+
 size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
                  const void* src, size_t srcSize)
 {
     unsigned tmpCounters[HIST_WKSP_SIZE_U32];
     return HIST_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters, sizeof(tmpCounters));
 }
+#endif
diff --git a/lib/compress/hist.h b/lib/compress/hist.h
index 77e3ec4fb192..fb9ead6834f6 100644
--- a/lib/compress/hist.h
+++ b/lib/compress/hist.h
@@ -14,7 +14,7 @@
 ****************************************************************** */
 
 /* --- dependencies --- */
-#include <stddef.h>   /* size_t */
+#include "../common/zstd_deps.h"   /* size_t */
 
 
 /* --- simple histogram functions --- */
diff --git a/lib/compress/huf_compress.c b/lib/compress/huf_compress.c
index 546879868a53..302e08864dac 100644
--- a/lib/compress/huf_compress.c
+++ b/lib/compress/huf_compress.c
@@ -23,8 +23,7 @@
 /* **************************************************************
 *  Includes
 ****************************************************************/
-#include <string.h>     /* memcpy, memset */
-#include <stdio.h>      /* printf (debug) */
+#include "../common/zstd_deps.h"     /* ZSTD_memcpy, ZSTD_memset */
 #include "../common/compiler.h"
 #include "../common/bitstream.h"
 #include "hist.h"
@@ -70,7 +69,7 @@ static size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weight
     U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
 
     FSE_CTable CTable[FSE_CTABLE_SIZE_U32(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, HUF_TABLELOG_MAX)];
-    BYTE scratchBuffer[1<<MAX_FSE_TABLELOG_FOR_HUFF_HEADER];
+    U32 scratchBuffer[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(HUF_TABLELOG_MAX, MAX_FSE_TABLELOG_FOR_HUFF_HEADER)];
 
     unsigned count[HUF_TABLELOG_MAX+1];
     S16 norm[HUF_TABLELOG_MAX+1];
@@ -85,7 +84,7 @@ static size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weight
     }
 
     tableLog = FSE_optimalTableLog(tableLog, wtSize, maxSymbolValue);
-    CHECK_F( FSE_normalizeCount(norm, tableLog, count, wtSize, maxSymbolValue) );
+    CHECK_F( FSE_normalizeCount(norm, tableLog, count, wtSize, maxSymbolValue, /* useLowProbCount */ 0) );
 
     /* Write table description header */
     {   CHECK_V_F(hSize, FSE_writeNCount(op, (size_t)(oend-op), norm, maxSymbolValue, tableLog) );
@@ -103,11 +102,6 @@ static size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weight
 }
 
 
-struct HUF_CElt_s {
-  U16  val;
-  BYTE nbBits;
-};   /* typedef'd to HUF_CElt within "huf.h" */
-
 /*! HUF_writeCTable() :
     `CTable` : Huffman tree to save, using huf representation.
     @return : size of saved CTable */
@@ -156,6 +150,7 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
 
     /* get symbol weights */
     CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
+    *hasZeroWeights = (rankVal[0] > 0);
 
     /* check result */
     if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
@@ -164,16 +159,14 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
     /* Prepare base value per rank */
     {   U32 n, nextRankStart = 0;
         for (n=1; n<=tableLog; n++) {
-            U32 current = nextRankStart;
+            U32 curr = nextRankStart;
             nextRankStart += (rankVal[n] << (n-1));
-            rankVal[n] = current;
+            rankVal[n] = curr;
     }   }
 
     /* fill nbBits */
-    *hasZeroWeights = 0;
     {   U32 n; for (n=0; n<nbSymbols; n++) {
             const U32 w = huffWeight[n];
-            *hasZeroWeights |= (w == 0);
             CTable[n].nbBits = (BYTE)(tableLog + 1 - w) & -(w != 0);
     }   }
 
@@ -212,32 +205,63 @@ typedef struct nodeElt_s {
     BYTE nbBits;
 } nodeElt;
 
+/**
+ * HUF_setMaxHeight():
+ * Enforces maxNbBits on the Huffman tree described in huffNode.
+ *
+ * It sets all nodes with nbBits > maxNbBits to be maxNbBits. Then it adjusts
+ * the tree to so that it is a valid canonical Huffman tree.
+ *
+ * @pre               The sum of the ranks of each symbol == 2^largestBits,
+ *                    where largestBits == huffNode[lastNonNull].nbBits.
+ * @post              The sum of the ranks of each symbol == 2^largestBits,
+ *                    where largestBits is the return value <= maxNbBits.
+ *
+ * @param huffNode    The Huffman tree modified in place to enforce maxNbBits.
+ * @param lastNonNull The symbol with the lowest count in the Huffman tree.
+ * @param maxNbBits   The maximum allowed number of bits, which the Huffman tree
+ *                    may not respect. After this function the Huffman tree will
+ *                    respect maxNbBits.
+ * @return            The maximum number of bits of the Huffman tree after adjustment,
+ *                    necessarily no more than maxNbBits.
+ */
 static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
 {
     const U32 largestBits = huffNode[lastNonNull].nbBits;
-    if (largestBits <= maxNbBits) return largestBits;   /* early exit : no elt > maxNbBits */
+    /* early exit : no elt > maxNbBits, so the tree is already valid. */
+    if (largestBits <= maxNbBits) return largestBits;
 
     /* there are several too large elements (at least >= 2) */
     {   int totalCost = 0;
         const U32 baseCost = 1 << (largestBits - maxNbBits);
         int n = (int)lastNonNull;
 
+        /* Adjust any ranks > maxNbBits to maxNbBits.
+         * Compute totalCost, which is how far the sum of the ranks is
+         * we are over 2^largestBits after adjust the offending ranks.
+         */
         while (huffNode[n].nbBits > maxNbBits) {
             totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
             huffNode[n].nbBits = (BYTE)maxNbBits;
-            n --;
-        }  /* n stops at huffNode[n].nbBits <= maxNbBits */
-        while (huffNode[n].nbBits == maxNbBits) n--;   /* n end at index of smallest symbol using < maxNbBits */
+            n--;
+        }
+        /* n stops at huffNode[n].nbBits <= maxNbBits */
+        assert(huffNode[n].nbBits <= maxNbBits);
+        /* n end at index of smallest symbol using < maxNbBits */
+        while (huffNode[n].nbBits == maxNbBits) --n;
 
-        /* renorm totalCost */
-        totalCost >>= (largestBits - maxNbBits);  /* note : totalCost is necessarily a multiple of baseCost */
+        /* renorm totalCost from 2^largestBits to 2^maxNbBits
+         * note : totalCost is necessarily a multiple of baseCost */
+        assert((totalCost & (baseCost - 1)) == 0);
+        totalCost >>= (largestBits - maxNbBits);
+        assert(totalCost > 0);
 
         /* repay normalized cost */
         {   U32 const noSymbol = 0xF0F0F0F0;
             U32 rankLast[HUF_TABLELOG_MAX+2];
 
-            /* Get pos of last (smallest) symbol per rank */
-            memset(rankLast, 0xF0, sizeof(rankLast));
+            /* Get pos of last (smallest = lowest cum. count) symbol per rank */
+            ZSTD_memset(rankLast, 0xF0, sizeof(rankLast));
             {   U32 currentNbBits = maxNbBits;
                 int pos;
                 for (pos=n ; pos >= 0; pos--) {
@@ -247,34 +271,65 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
             }   }
 
             while (totalCost > 0) {
+                /* Try to reduce the next power of 2 above totalCost because we
+                 * gain back half the rank.
+                 */
                 U32 nBitsToDecrease = BIT_highbit32((U32)totalCost) + 1;
                 for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
                     U32 const highPos = rankLast[nBitsToDecrease];
                     U32 const lowPos = rankLast[nBitsToDecrease-1];
                     if (highPos == noSymbol) continue;
+                    /* Decrease highPos if no symbols of lowPos or if it is
+                     * not cheaper to remove 2 lowPos than highPos.
+                     */
                     if (lowPos == noSymbol) break;
                     {   U32 const highTotal = huffNode[highPos].count;
                         U32 const lowTotal = 2 * huffNode[lowPos].count;
                         if (highTotal <= lowTotal) break;
                 }   }
                 /* only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) */
+                assert(rankLast[nBitsToDecrease] != noSymbol || nBitsToDecrease == 1);
                 /* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
                 while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))
-                    nBitsToDecrease ++;
+                    nBitsToDecrease++;
+                assert(rankLast[nBitsToDecrease] != noSymbol);
+                /* Increase the number of bits to gain back half the rank cost. */
                 totalCost -= 1 << (nBitsToDecrease-1);
+                huffNode[rankLast[nBitsToDecrease]].nbBits++;
+
+                /* Fix up the new rank.
+                 * If the new rank was empty, this symbol is now its smallest.
+                 * Otherwise, this symbol will be the largest in the new rank so no adjustment.
+                 */
                 if (rankLast[nBitsToDecrease-1] == noSymbol)
-                    rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease];   /* this rank is no longer empty */
-                huffNode[rankLast[nBitsToDecrease]].nbBits ++;
+                    rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease];
+                /* Fix up the old rank.
+                 * If the symbol was at position 0, meaning it was the highest weight symbol in the tree,
+                 * it must be the only symbol in its rank, so the old rank now has no symbols.
+                 * Otherwise, since the Huffman nodes are sorted by count, the previous position is now
+                 * the smallest node in the rank. If the previous position belongs to a different rank,
+                 * then the rank is now empty.
+                 */
                 if (rankLast[nBitsToDecrease] == 0)    /* special case, reached largest symbol */
                     rankLast[nBitsToDecrease] = noSymbol;
                 else {
                     rankLast[nBitsToDecrease]--;
                     if (huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease)
                         rankLast[nBitsToDecrease] = noSymbol;   /* this rank is now empty */
-            }   }   /* while (totalCost > 0) */
-
+                }
+            }   /* while (totalCost > 0) */
+
+            /* If we've removed too much weight, then we have to add it back.
+             * To avoid overshooting again, we only adjust the smallest rank.
+             * We take the largest nodes from the lowest rank 0 and move them
+             * to rank 1. There's guaranteed to be enough rank 0 symbols because
+             * TODO.
+             */
             while (totalCost < 0) {  /* Sometimes, cost correction overshoot */
-                if (rankLast[1] == noSymbol) {  /* special case : no rank 1 symbol (using maxNbBits-1); let's create one from largest rank 0 (using maxNbBits) */
+                /* special case : no rank 1 symbol (using maxNbBits-1);
+                 * let's create one from largest rank 0 (using maxNbBits).
+                 */
+                if (rankLast[1] == noSymbol) {
                     while (huffNode[n].nbBits == maxNbBits) n--;
                     huffNode[n+1].nbBits--;
                     assert(n >= 0);
@@ -285,14 +340,16 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
                 huffNode[ rankLast[1] + 1 ].nbBits--;
                 rankLast[1]++;
                 totalCost ++;
-    }   }   }   /* there are several too large elements (at least >= 2) */
+            }
+        }   /* repay normalized cost */
+    }   /* there are several too large elements (at least >= 2) */
 
     return maxNbBits;
 }
 
 typedef struct {
     U32 base;
-    U32 current;
+    U32 curr;
 } rankPos;
 
 typedef nodeElt huffNodeTable[HUF_CTABLE_WORKSPACE_SIZE_U32];
@@ -304,21 +361,45 @@ typedef struct {
   rankPos rankPosition[RANK_POSITION_TABLE_SIZE];
 } HUF_buildCTable_wksp_tables;
 
+/**
+ * HUF_sort():
+ * Sorts the symbols [0, maxSymbolValue] by count[symbol] in decreasing order.
+ *
+ * @param[out] huffNode       Sorted symbols by decreasing count. Only members `.count` and `.byte` are filled.
+ *                            Must have (maxSymbolValue + 1) entries.
+ * @param[in]  count          Histogram of the symbols.
+ * @param[in]  maxSymbolValue Maximum symbol value.
+ * @param      rankPosition   This is a scratch workspace. Must have RANK_POSITION_TABLE_SIZE entries.
+ */
 static void HUF_sort(nodeElt* huffNode, const unsigned* count, U32 maxSymbolValue, rankPos* rankPosition)
 {
-    U32 n;
-
-    memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
-    for (n=0; n<=maxSymbolValue; n++) {
-        U32 r = BIT_highbit32(count[n] + 1);
-        rankPosition[r].base ++;
+    int n;
+    int const maxSymbolValue1 = (int)maxSymbolValue + 1;
+
+    /* Compute base and set curr to base.
+     * For symbol s let lowerRank = BIT_highbit32(count[n]+1) and rank = lowerRank + 1.
+     * Then 2^lowerRank <= count[n]+1 <= 2^rank.
+     * We attribute each symbol to lowerRank's base value, because we want to know where
+     * each rank begins in the output, so for rank R we want to count ranks R+1 and above.
+     */
+    ZSTD_memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
+    for (n = 0; n < maxSymbolValue1; ++n) {
+        U32 lowerRank = BIT_highbit32(count[n] + 1);
+        rankPosition[lowerRank].base++;
     }
-    for (n=30; n>0; n--) rankPosition[n-1].base += rankPosition[n].base;
-    for (n=0; n<32; n++) rankPosition[n].current = rankPosition[n].base;
-    for (n=0; n<=maxSymbolValue; n++) {
+    assert(rankPosition[RANK_POSITION_TABLE_SIZE - 1].base == 0);
+    for (n = RANK_POSITION_TABLE_SIZE - 1; n > 0; --n) {
+        rankPosition[n-1].base += rankPosition[n].base;
+        rankPosition[n-1].curr = rankPosition[n-1].base;
+    }
+    /* Sort */
+    for (n = 0; n < maxSymbolValue1; ++n) {
         U32 const c = count[n];
         U32 const r = BIT_highbit32(c+1) + 1;
-        U32 pos = rankPosition[r].current++;
+        U32 pos = rankPosition[r].curr++;
+        /* Insert into the correct position in the rank.
+         * We have at most 256 symbols, so this insertion should be fine.
+         */
         while ((pos > rankPosition[r].base) && (c > huffNode[pos-1].count)) {
             huffNode[pos] = huffNode[pos-1];
             pos--;
@@ -335,28 +416,20 @@ static void HUF_sort(nodeElt* huffNode, const unsigned* count, U32 maxSymbolValu
  */
 #define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
 
-size_t HUF_buildCTable_wksp (HUF_CElt* tree, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
+/* HUF_buildTree():
+ * Takes the huffNode array sorted by HUF_sort() and builds an unlimited-depth Huffman tree.
+ *
+ * @param huffNode        The array sorted by HUF_sort(). Builds the Huffman tree in this array.
+ * @param maxSymbolValue  The maximum symbol value.
+ * @return                The smallest node in the Huffman tree (by count).
+ */
+static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
 {
-    HUF_buildCTable_wksp_tables* const wksp_tables = (HUF_buildCTable_wksp_tables*)workSpace;
-    nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
-    nodeElt* const huffNode = huffNode0+1;
+    nodeElt* const huffNode0 = huffNode - 1;
     int nonNullRank;
     int lowS, lowN;
     int nodeNb = STARTNODE;
     int n, nodeRoot;
-
-    /* safety checks */
-    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
-    if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
-      return ERROR(workSpace_tooSmall);
-    if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
-    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
-      return ERROR(maxSymbolValue_tooLarge);
-    memset(huffNode0, 0, sizeof(huffNodeTable));
-
-    /* sort, decreasing order */
-    HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
-
     /* init for parents */
     nonNullRank = (int)maxSymbolValue;
     while(huffNode[nonNullRank].count == 0) nonNullRank--;
@@ -383,42 +456,72 @@ size_t HUF_buildCTable_wksp (HUF_CElt* tree, const unsigned* count, U32 maxSymbo
     for (n=0; n<=nonNullRank; n++)
         huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
 
+    return nonNullRank;
+}
+
+/**
+ * HUF_buildCTableFromTree():
+ * Build the CTable given the Huffman tree in huffNode.
+ *
+ * @param[out] CTable         The output Huffman CTable.
+ * @param      huffNode       The Huffman tree.
+ * @param      nonNullRank    The last and smallest node in the Huffman tree.
+ * @param      maxSymbolValue The maximum symbol value.
+ * @param      maxNbBits      The exact maximum number of bits used in the Huffman tree.
+ */
+static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, int nonNullRank, U32 maxSymbolValue, U32 maxNbBits)
+{
+    /* fill result into ctable (val, nbBits) */
+    int n;
+    U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
+    U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
+    int const alphabetSize = (int)(maxSymbolValue + 1);
+    for (n=0; n<=nonNullRank; n++)
+        nbPerRank[huffNode[n].nbBits]++;
+    /* determine starting value per rank */
+    {   U16 min = 0;
+        for (n=(int)maxNbBits; n>0; n--) {
+            valPerRank[n] = min;      /* get starting value within each rank */
+            min += nbPerRank[n];
+            min >>= 1;
+    }   }
+    for (n=0; n<alphabetSize; n++)
+        CTable[huffNode[n].byte].nbBits = huffNode[n].nbBits;   /* push nbBits per symbol, symbol order */
+    for (n=0; n<alphabetSize; n++)
+        CTable[n].val = valPerRank[CTable[n].nbBits]++;   /* assign value within rank, symbol order */
+}
+
+size_t HUF_buildCTable_wksp (HUF_CElt* tree, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
+{
+    HUF_buildCTable_wksp_tables* const wksp_tables = (HUF_buildCTable_wksp_tables*)workSpace;
+    nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
+    nodeElt* const huffNode = huffNode0+1;
+    int nonNullRank;
+
+    /* safety checks */
+    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
+      return ERROR(workSpace_tooSmall);
+    if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
+      return ERROR(maxSymbolValue_tooLarge);
+    ZSTD_memset(huffNode0, 0, sizeof(huffNodeTable));
+
+    /* sort, decreasing order */
+    HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
+
+    /* build tree */
+    nonNullRank = HUF_buildTree(huffNode, maxSymbolValue);
+
     /* enforce maxTableLog */
     maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
+    if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC);   /* check fit into table */
 
-    /* fill result into tree (val, nbBits) */
-    {   U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
-        U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
-        int const alphabetSize = (int)(maxSymbolValue + 1);
-        if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC);   /* check fit into table */
-        for (n=0; n<=nonNullRank; n++)
-            nbPerRank[huffNode[n].nbBits]++;
-        /* determine stating value per rank */
-        {   U16 min = 0;
-            for (n=(int)maxNbBits; n>0; n--) {
-                valPerRank[n] = min;      /* get starting value within each rank */
-                min += nbPerRank[n];
-                min >>= 1;
-        }   }
-        for (n=0; n<alphabetSize; n++)
-            tree[huffNode[n].byte].nbBits = huffNode[n].nbBits;   /* push nbBits per symbol, symbol order */
-        for (n=0; n<alphabetSize; n++)
-            tree[n].val = valPerRank[tree[n].nbBits]++;   /* assign value within rank, symbol order */
-    }
+    HUF_buildCTableFromTree(tree, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
 
     return maxNbBits;
 }
 
-/** 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 unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits)
-{
-    HUF_buildCTable_wksp_tables workspace;
-    return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, &workspace, sizeof(workspace));
-}
-
 size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
 {
     size_t nbBits = 0;
@@ -633,25 +736,26 @@ typedef struct {
 } HUF_compress_tables_t;
 
 /* HUF_compress_internal() :
- * `workSpace` must a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
+ * `workSpace_align4` must be aligned on 4-bytes boundaries,
+ * and occupies the same space as a table of 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,
                        HUF_nbStreams_e nbStreams,
-                       void* workSpace, size_t wkspSize,
+                       void* workSpace_align4, 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;
+    HUF_compress_tables_t* const table = (HUF_compress_tables_t*)workSpace_align4;
     BYTE* const ostart = (BYTE*)dst;
     BYTE* const oend = ostart + dstSize;
     BYTE* op = ostart;
 
     HUF_STATIC_ASSERT(sizeof(*table) <= HUF_WORKSPACE_SIZE);
+    assert(((size_t)workSpace_align4 & 3) == 0);   /* must be aligned on 4-bytes boundaries */
 
     /* checks & inits */
-    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
     if (wkspSize < HUF_WORKSPACE_SIZE) return ERROR(workSpace_tooSmall);
     if (!srcSize) return 0;  /* Uncompressed */
     if (!dstSize) return 0;  /* cannot fit anything within dst budget */
@@ -669,7 +773,7 @@ HUF_compress_internal (void* dst, size_t dstSize,
     }
 
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, workSpace, wkspSize) );
+    {   CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, workSpace_align4, wkspSize) );
         if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
         if (largest <= (srcSize >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
     }
@@ -695,7 +799,7 @@ HUF_compress_internal (void* dst, size_t dstSize,
         CHECK_F(maxBits);
         huffLog = (U32)maxBits;
         /* Zero unused symbols in CTable, so we can check it for validity */
-        memset(table->CTable + (maxSymbolValue + 1), 0,
+        ZSTD_memset(table->CTable + (maxSymbolValue + 1), 0,
                sizeof(table->CTable) - ((maxSymbolValue + 1) * sizeof(HUF_CElt)));
     }
 
@@ -716,7 +820,7 @@ HUF_compress_internal (void* dst, size_t dstSize,
         op += hSize;
         if (repeat) { *repeat = HUF_repeat_none; }
         if (oldHufTable)
-            memcpy(oldHufTable, table->CTable, sizeof(table->CTable));  /* Save new table */
+            ZSTD_memcpy(oldHufTable, table->CTable, sizeof(table->CTable));  /* Save new table */
     }
     return HUF_compressCTable_internal(ostart, op, oend,
                                        src, srcSize,
@@ -747,14 +851,6 @@ size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
                                  repeat, preferRepeat, bmi2);
 }
 
-size_t HUF_compress1X (void* dst, size_t dstSize,
-                 const void* src, size_t srcSize,
-                 unsigned maxSymbolValue, unsigned huffLog)
-{
-    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 */
@@ -784,6 +880,25 @@ size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
                                  hufTable, repeat, preferRepeat, bmi2);
 }
 
+#ifndef ZSTD_NO_UNUSED_FUNCTIONS
+/** 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 unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits)
+{
+    HUF_buildCTable_wksp_tables workspace;
+    return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, &workspace, sizeof(workspace));
+}
+
+size_t HUF_compress1X (void* dst, size_t dstSize,
+                 const void* src, size_t srcSize,
+                 unsigned maxSymbolValue, unsigned huffLog)
+{
+    unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
+    return HUF_compress1X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
+}
+
 size_t HUF_compress2 (void* dst, size_t dstSize,
                 const void* src, size_t srcSize,
                 unsigned maxSymbolValue, unsigned huffLog)
@@ -796,3 +911,4 @@ size_t HUF_compress (void* dst, size_t maxDstSize, const void* src, size_t srcSi
 {
     return HUF_compress2(dst, maxDstSize, src, srcSize, 255, HUF_TABLELOG_DEFAULT);
 }
+#endif
diff --git a/lib/compress/zstd_compress.c b/lib/compress/zstd_compress.c
index 3f963b1cfff8..386b051df635 100644
--- a/lib/compress/zstd_compress.c
+++ b/lib/compress/zstd_compress.c
@@ -11,8 +11,7 @@
 /*-*************************************
 *  Dependencies
 ***************************************/
-#include <limits.h>         /* INT_MAX */
-#include <string.h>         /* memset */
+#include "../common/zstd_deps.h"  /* INT_MAX, ZSTD_memset, ZSTD_memcpy */
 #include "../common/cpu.h"
 #include "../common/mem.h"
 #include "hist.h"           /* HIST_countFast_wksp */
@@ -30,6 +29,19 @@
 #include "zstd_ldm.h"
 #include "zstd_compress_superblock.h"
 
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * COMPRESS_HEAPMODE :
+ * Select how default decompression function ZSTD_compress() allocates its context,
+ * on stack (0, default), or into heap (1).
+ * Note that functions with explicit context such as ZSTD_compressCCtx() are unaffected.
+ */
+#ifndef ZSTD_COMPRESS_HEAPMODE
+#  define ZSTD_COMPRESS_HEAPMODE 0
+#endif
+
 
 /*-*************************************
 *  Helper functions
@@ -52,6 +64,7 @@ size_t ZSTD_compressBound(size_t srcSize) {
 struct ZSTD_CDict_s {
     const void* dictContent;
     size_t dictContentSize;
+    ZSTD_dictContentType_e dictContentType; /* The dictContentType the CDict was created with */
     U32* entropyWorkspace; /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
     ZSTD_cwksp workspace;
     ZSTD_matchState_t matchState;
@@ -69,7 +82,7 @@ ZSTD_CCtx* ZSTD_createCCtx(void)
 static void ZSTD_initCCtx(ZSTD_CCtx* cctx, ZSTD_customMem memManager)
 {
     assert(cctx != NULL);
-    memset(cctx, 0, sizeof(*cctx));
+    ZSTD_memset(cctx, 0, sizeof(*cctx));
     cctx->customMem = memManager;
     cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
     {   size_t const err = ZSTD_CCtx_reset(cctx, ZSTD_reset_parameters);
@@ -82,8 +95,8 @@ ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_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_malloc(sizeof(ZSTD_CCtx), customMem);
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+    {   ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_customMalloc(sizeof(ZSTD_CCtx), customMem);
         if (!cctx) return NULL;
         ZSTD_initCCtx(cctx, customMem);
         return cctx;
@@ -96,20 +109,20 @@ ZSTD_CCtx* ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize)
     ZSTD_CCtx* cctx;
     if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL;  /* minimum size */
     if ((size_t)workspace & 7) return NULL;  /* must be 8-aligned */
-    ZSTD_cwksp_init(&ws, workspace, workspaceSize);
+    ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_static_alloc);
 
     cctx = (ZSTD_CCtx*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CCtx));
     if (cctx == NULL) return NULL;
 
-    memset(cctx, 0, sizeof(ZSTD_CCtx));
+    ZSTD_memset(cctx, 0, sizeof(ZSTD_CCtx));
     ZSTD_cwksp_move(&cctx->workspace, &ws);
     cctx->staticSize = workspaceSize;
 
     /* statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
-    if (!ZSTD_cwksp_check_available(&cctx->workspace, HUF_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t))) return NULL;
+    if (!ZSTD_cwksp_check_available(&cctx->workspace, ENTROPY_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t))) return NULL;
     cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
     cctx->blockState.nextCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
-    cctx->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cctx->workspace, HUF_WORKSPACE_SIZE);
+    cctx->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cctx->workspace, ENTROPY_WORKSPACE_SIZE);
     cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
     return cctx;
 }
@@ -119,10 +132,10 @@ ZSTD_CCtx* ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize)
  */
 static void ZSTD_clearAllDicts(ZSTD_CCtx* cctx)
 {
-    ZSTD_free(cctx->localDict.dictBuffer, cctx->customMem);
+    ZSTD_customFree(cctx->localDict.dictBuffer, cctx->customMem);
     ZSTD_freeCDict(cctx->localDict.cdict);
-    memset(&cctx->localDict, 0, sizeof(cctx->localDict));
-    memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));
+    ZSTD_memset(&cctx->localDict, 0, sizeof(cctx->localDict));
+    ZSTD_memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));
     cctx->cdict = NULL;
 }
 
@@ -153,7 +166,7 @@ size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
         int cctxInWorkspace = ZSTD_cwksp_owns_buffer(&cctx->workspace, cctx);
         ZSTD_freeCCtxContent(cctx);
         if (!cctxInWorkspace) {
-            ZSTD_free(cctx, cctx->customMem);
+            ZSTD_customFree(cctx, cctx->customMem);
         }
     }
     return 0;
@@ -189,15 +202,32 @@ size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
 /* private API call, for dictBuilder only */
 const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx) { return &(ctx->seqStore); }
 
+/* Returns 1 if compression parameters are such that we should
+ * enable long distance matching (wlog >= 27, strategy >= btopt).
+ * Returns 0 otherwise.
+ */
+static U32 ZSTD_CParams_shouldEnableLdm(const ZSTD_compressionParameters* const cParams) {
+    return cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27;
+}
+
 static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
         ZSTD_compressionParameters cParams)
 {
     ZSTD_CCtx_params cctxParams;
-    memset(&cctxParams, 0, sizeof(cctxParams));
+    /* should not matter, as all cParams are presumed properly defined */
+    ZSTD_CCtxParams_init(&cctxParams, ZSTD_CLEVEL_DEFAULT);
     cctxParams.cParams = cParams;
-    cctxParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;  /* should not matter, as all cParams are presumed properly defined */
+
+    if (ZSTD_CParams_shouldEnableLdm(&cParams)) {
+        DEBUGLOG(4, "ZSTD_makeCCtxParamsFromCParams(): Including LDM into cctx params");
+        cctxParams.ldmParams.enableLdm = 1;
+        /* LDM is enabled by default for optimal parser and window size >= 128MB */
+        ZSTD_ldm_adjustParameters(&cctxParams.ldmParams, &cParams);
+        assert(cctxParams.ldmParams.hashLog >= cctxParams.ldmParams.bucketSizeLog);
+        assert(cctxParams.ldmParams.hashRateLog < 32);
+    }
+
     assert(!ZSTD_checkCParams(cParams));
-    cctxParams.fParams.contentSizeFlag = 1;
     return cctxParams;
 }
 
@@ -205,13 +235,12 @@ 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(
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
+    params = (ZSTD_CCtx_params*)ZSTD_customCalloc(
             sizeof(ZSTD_CCtx_params), customMem);
     if (!params) { return NULL; }
+    ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
     params->customMem = customMem;
-    params->compressionLevel = ZSTD_CLEVEL_DEFAULT;
-    params->fParams.contentSizeFlag = 1;
     return params;
 }
 
@@ -223,7 +252,7 @@ ZSTD_CCtx_params* ZSTD_createCCtxParams(void)
 size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params)
 {
     if (params == NULL) { return 0; }
-    ZSTD_free(params, params->customMem);
+    ZSTD_customFree(params, params->customMem);
     return 0;
 }
 
@@ -234,7 +263,7 @@ size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params)
 
 size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
     RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
-    memset(cctxParams, 0, sizeof(*cctxParams));
+    ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
     cctxParams->compressionLevel = compressionLevel;
     cctxParams->fParams.contentSizeFlag = 1;
     return 0;
@@ -244,7 +273,7 @@ size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_paramete
 {
     RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
     FORWARD_IF_ERROR( ZSTD_checkCParams(params.cParams) , "");
-    memset(cctxParams, 0, sizeof(*cctxParams));
+    ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
     assert(!ZSTD_checkCParams(params.cParams));
     cctxParams->cParams = params.cParams;
     cctxParams->fParams = params.fParams;
@@ -354,6 +383,11 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
 #endif
         return bounds;
 
+    case ZSTD_c_enableDedicatedDictSearch:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
     case ZSTD_c_enableLongDistanceMatching:
         bounds.lowerBound = 0;
         bounds.upperBound = 1;
@@ -397,7 +431,7 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         return bounds;
 
     case ZSTD_c_forceAttachDict:
-        ZSTD_STATIC_ASSERT(ZSTD_dictDefaultAttach < ZSTD_dictForceCopy);
+        ZSTD_STATIC_ASSERT(ZSTD_dictDefaultAttach < ZSTD_dictForceLoad);
         bounds.lowerBound = ZSTD_dictDefaultAttach;
         bounds.upperBound = ZSTD_dictForceLoad;       /* note : how to ensure at compile time that this is the highest value enum ? */
         return bounds;
@@ -418,6 +452,22 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         bounds.upperBound = ZSTD_SRCSIZEHINT_MAX;
         return bounds;
 
+    case ZSTD_c_stableInBuffer:
+    case ZSTD_c_stableOutBuffer:
+        bounds.lowerBound = (int)ZSTD_bm_buffered;
+        bounds.upperBound = (int)ZSTD_bm_stable;
+        return bounds;
+
+    case ZSTD_c_blockDelimiters:
+        bounds.lowerBound = (int)ZSTD_sf_noBlockDelimiters;
+        bounds.upperBound = (int)ZSTD_sf_explicitBlockDelimiters;
+        return bounds;
+
+    case ZSTD_c_validateSequences:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
     default:
         bounds.error = ERROR(parameter_unsupported);
         return bounds;
@@ -465,6 +515,7 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
     case ZSTD_c_jobSize:
     case ZSTD_c_overlapLog:
     case ZSTD_c_rsyncable:
+    case ZSTD_c_enableDedicatedDictSearch:
     case ZSTD_c_enableLongDistanceMatching:
     case ZSTD_c_ldmHashLog:
     case ZSTD_c_ldmMinMatch:
@@ -474,6 +525,10 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
     case ZSTD_c_literalCompressionMode:
     case ZSTD_c_targetCBlockSize:
     case ZSTD_c_srcSizeHint:
+    case ZSTD_c_stableInBuffer:
+    case ZSTD_c_stableOutBuffer:
+    case ZSTD_c_blockDelimiters:
+    case ZSTD_c_validateSequences:
     default:
         return 0;
     }
@@ -515,12 +570,17 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
     case ZSTD_c_jobSize:
     case ZSTD_c_overlapLog:
     case ZSTD_c_rsyncable:
+    case ZSTD_c_enableDedicatedDictSearch:
     case ZSTD_c_enableLongDistanceMatching:
     case ZSTD_c_ldmHashLog:
     case ZSTD_c_ldmMinMatch:
     case ZSTD_c_ldmBucketSizeLog:
     case ZSTD_c_targetCBlockSize:
     case ZSTD_c_srcSizeHint:
+    case ZSTD_c_stableInBuffer:
+    case ZSTD_c_stableOutBuffer:
+    case ZSTD_c_blockDelimiters:
+    case ZSTD_c_validateSequences:
         break;
 
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
@@ -541,9 +601,10 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
 
     case ZSTD_c_compressionLevel : {
         FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
-        if (value) {  /* 0 : does not change current level */
+        if (value == 0)
+            CCtxParams->compressionLevel = ZSTD_CLEVEL_DEFAULT; /* 0 == default */
+        else
             CCtxParams->compressionLevel = value;
-        }
         if (CCtxParams->compressionLevel >= 0) return (size_t)CCtxParams->compressionLevel;
         return 0;  /* return type (size_t) cannot represent negative values */
     }
@@ -667,6 +728,10 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
         return CCtxParams->rsyncable;
 #endif
 
+    case ZSTD_c_enableDedicatedDictSearch :
+        CCtxParams->enableDedicatedDictSearch = (value!=0);
+        return CCtxParams->enableDedicatedDictSearch;
+
     case ZSTD_c_enableLongDistanceMatching :
         CCtxParams->ldmParams.enableLdm = (value!=0);
         return CCtxParams->ldmParams.enableLdm;
@@ -707,6 +772,26 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
         CCtxParams->srcSizeHint = value;
         return CCtxParams->srcSizeHint;
 
+    case ZSTD_c_stableInBuffer:
+        BOUNDCHECK(ZSTD_c_stableInBuffer, value);
+        CCtxParams->inBufferMode = (ZSTD_bufferMode_e)value;
+        return CCtxParams->inBufferMode;
+
+    case ZSTD_c_stableOutBuffer:
+        BOUNDCHECK(ZSTD_c_stableOutBuffer, value);
+        CCtxParams->outBufferMode = (ZSTD_bufferMode_e)value;
+        return CCtxParams->outBufferMode;
+
+    case ZSTD_c_blockDelimiters:
+        BOUNDCHECK(ZSTD_c_blockDelimiters, value);
+        CCtxParams->blockDelimiters = (ZSTD_sequenceFormat_e)value;
+        return CCtxParams->blockDelimiters;
+
+    case ZSTD_c_validateSequences:
+        BOUNDCHECK(ZSTD_c_validateSequences, value);
+        CCtxParams->validateSequences = value;
+        return CCtxParams->validateSequences;
+
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
 }
@@ -794,6 +879,9 @@ size_t ZSTD_CCtxParams_getParameter(
         *value = CCtxParams->rsyncable;
         break;
 #endif
+    case ZSTD_c_enableDedicatedDictSearch :
+        *value = CCtxParams->enableDedicatedDictSearch;
+        break;
     case ZSTD_c_enableLongDistanceMatching :
         *value = CCtxParams->ldmParams.enableLdm;
         break;
@@ -815,6 +903,18 @@ size_t ZSTD_CCtxParams_getParameter(
     case ZSTD_c_srcSizeHint :
         *value = (int)CCtxParams->srcSizeHint;
         break;
+    case ZSTD_c_stableInBuffer :
+        *value = (int)CCtxParams->inBufferMode;
+        break;
+    case ZSTD_c_stableOutBuffer :
+        *value = (int)CCtxParams->outBufferMode;
+        break;
+    case ZSTD_c_blockDelimiters :
+        *value = (int)CCtxParams->blockDelimiters;
+        break;
+    case ZSTD_c_validateSequences :
+        *value = (int)CCtxParams->validateSequences;
+        break;
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
     return 0;
@@ -850,6 +950,14 @@ ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long lo
     return 0;
 }
 
+static ZSTD_compressionParameters ZSTD_dedicatedDictSearch_getCParams(
+        int const compressionLevel,
+        size_t const dictSize);
+static int ZSTD_dedicatedDictSearch_isSupported(
+        const ZSTD_compressionParameters* cParams);
+static void ZSTD_dedicatedDictSearch_revertCParams(
+        ZSTD_compressionParameters* cParams);
+
 /**
  * Initializes the local dict using the requested parameters.
  * NOTE: This does not use the pledged src size, because it may be used for more
@@ -858,8 +966,6 @@ ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long lo
 static size_t ZSTD_initLocalDict(ZSTD_CCtx* cctx)
 {
     ZSTD_localDict* const dl = &cctx->localDict;
-    ZSTD_compressionParameters const cParams = ZSTD_getCParamsFromCCtxParams(
-            &cctx->requestedParams, ZSTD_CONTENTSIZE_UNKNOWN, dl->dictSize);
     if (dl->dict == NULL) {
         /* No local dictionary. */
         assert(dl->dictBuffer == NULL);
@@ -876,12 +982,12 @@ static size_t ZSTD_initLocalDict(ZSTD_CCtx* cctx)
     assert(cctx->cdict == NULL);
     assert(cctx->prefixDict.dict == NULL);
 
-    dl->cdict = ZSTD_createCDict_advanced(
+    dl->cdict = ZSTD_createCDict_advanced2(
             dl->dict,
             dl->dictSize,
             ZSTD_dlm_byRef,
             dl->dictContentType,
-            cParams,
+            &cctx->requestedParams,
             cctx->customMem);
     RETURN_ERROR_IF(!dl->cdict, memory_allocation, "ZSTD_createCDict_advanced failed");
     cctx->cdict = dl->cdict;
@@ -894,8 +1000,6 @@ size_t ZSTD_CCtx_loadDictionary_advanced(
 {
     RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
                     "Can't load a dictionary when ctx is not in init stage.");
-    RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
-                    "no malloc for static CCtx");
     DEBUGLOG(4, "ZSTD_CCtx_loadDictionary_advanced (size: %u)", (U32)dictSize);
     ZSTD_clearAllDicts(cctx);  /* in case one already exists */
     if (dict == NULL || dictSize == 0)  /* no dictionary mode */
@@ -903,9 +1007,12 @@ size_t ZSTD_CCtx_loadDictionary_advanced(
     if (dictLoadMethod == ZSTD_dlm_byRef) {
         cctx->localDict.dict = dict;
     } else {
-        void* dictBuffer = ZSTD_malloc(dictSize, cctx->customMem);
+        void* dictBuffer;
+        RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
+                        "no malloc for static CCtx");
+        dictBuffer = ZSTD_customMalloc(dictSize, cctx->customMem);
         RETURN_ERROR_IF(!dictBuffer, memory_allocation, "NULL pointer!");
-        memcpy(dictBuffer, dict, dictSize);
+        ZSTD_memcpy(dictBuffer, dict, dictSize);
         cctx->localDict.dictBuffer = dictBuffer;
         cctx->localDict.dict = dictBuffer;
     }
@@ -938,6 +1045,14 @@ size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
     return 0;
 }
 
+size_t ZSTD_CCtx_refThreadPool(ZSTD_CCtx* cctx, ZSTD_threadPool* pool)
+{
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't ref a pool when ctx not in init stage.");
+    cctx->pool = pool;
+    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);
@@ -1022,24 +1137,73 @@ U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
     return hashLog - btScale;
 }
 
+/** ZSTD_dictAndWindowLog() :
+ * Returns an adjusted window log that is large enough to fit the source and the dictionary.
+ * The zstd format says that the entire dictionary is valid if one byte of the dictionary
+ * is within the window. So the hashLog and chainLog should be large enough to reference both
+ * the dictionary and the window. So we must use this adjusted dictAndWindowLog when downsizing
+ * the hashLog and windowLog.
+ * NOTE: srcSize must not be ZSTD_CONTENTSIZE_UNKNOWN.
+ */
+static U32 ZSTD_dictAndWindowLog(U32 windowLog, U64 srcSize, U64 dictSize)
+{
+    const U64 maxWindowSize = 1ULL << ZSTD_WINDOWLOG_MAX;
+    /* No dictionary ==> No change */
+    if (dictSize == 0) {
+        return windowLog;
+    }
+    assert(windowLog <= ZSTD_WINDOWLOG_MAX);
+    assert(srcSize != ZSTD_CONTENTSIZE_UNKNOWN); /* Handled in ZSTD_adjustCParams_internal() */
+    {
+        U64 const windowSize = 1ULL << windowLog;
+        U64 const dictAndWindowSize = dictSize + windowSize;
+        /* If the window size is already large enough to fit both the source and the dictionary
+         * then just use the window size. Otherwise adjust so that it fits the dictionary and
+         * the window.
+         */
+        if (windowSize >= dictSize + srcSize) {
+            return windowLog; /* Window size large enough already */
+        } else if (dictAndWindowSize >= maxWindowSize) {
+            return ZSTD_WINDOWLOG_MAX; /* Larger than max window log */
+        } else  {
+            return ZSTD_highbit32((U32)dictAndWindowSize - 1) + 1;
+        }
+    }
+}
+
 /** ZSTD_adjustCParams_internal() :
  *  optimize `cPar` for a specified input (`srcSize` and `dictSize`).
  *  mostly downsize to reduce memory consumption and initialization latency.
  * `srcSize` can be ZSTD_CONTENTSIZE_UNKNOWN when not known.
+ * `mode` is the mode for parameter adjustment. See docs for `ZSTD_cParamMode_e`.
  *  note : `srcSize==0` means 0!
  *  condition : cPar is presumed validated (can be checked using ZSTD_checkCParams()). */
 static ZSTD_compressionParameters
 ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
                             unsigned long long srcSize,
-                            size_t dictSize)
+                            size_t dictSize,
+                            ZSTD_cParamMode_e mode)
 {
-    static const U64 minSrcSize = 513; /* (1<<9) + 1 */
-    static const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
+    const U64 minSrcSize = 513; /* (1<<9) + 1 */
+    const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
     assert(ZSTD_checkCParams(cPar)==0);
 
     if (dictSize && srcSize == ZSTD_CONTENTSIZE_UNKNOWN)
         srcSize = minSrcSize;
 
+    switch (mode) {
+    case ZSTD_cpm_noAttachDict:
+    case ZSTD_cpm_unknown:
+    case ZSTD_cpm_createCDict:
+        break;
+    case ZSTD_cpm_attachDict:
+        dictSize = 0;
+        break;
+    default:
+        assert(0);
+        break;
+    }
+
     /* resize windowLog if input is small enough, to use less memory */
     if ( (srcSize < maxWindowResize)
       && (dictSize < maxWindowResize) )  {
@@ -1049,10 +1213,11 @@ ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
                             ZSTD_highbit32(tSize-1) + 1;
         if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
     }
-    if (cPar.hashLog > cPar.windowLog+1) cPar.hashLog = cPar.windowLog+1;
-    {   U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
-        if (cycleLog > cPar.windowLog)
-            cPar.chainLog -= (cycleLog - cPar.windowLog);
+    {   U32 const dictAndWindowLog = ZSTD_dictAndWindowLog(cPar.windowLog, (U64)srcSize, (U64)dictSize);
+        U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
+        if (cPar.hashLog > dictAndWindowLog+1) cPar.hashLog = dictAndWindowLog+1;
+        if (cycleLog > dictAndWindowLog)
+            cPar.chainLog -= (cycleLog - dictAndWindowLog);
     }
 
     if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN)
@@ -1068,31 +1233,38 @@ ZSTD_adjustCParams(ZSTD_compressionParameters cPar,
 {
     cPar = ZSTD_clampCParams(cPar);   /* resulting cPar is necessarily valid (all parameters within range) */
     if (srcSize == 0) srcSize = ZSTD_CONTENTSIZE_UNKNOWN;
-    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize);
+    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize, ZSTD_cpm_unknown);
 }
 
-static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize);
-static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize);
+static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
+static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
+
+static void ZSTD_overrideCParams(
+              ZSTD_compressionParameters* cParams,
+        const ZSTD_compressionParameters* overrides)
+{
+    if (overrides->windowLog)    cParams->windowLog    = overrides->windowLog;
+    if (overrides->hashLog)      cParams->hashLog      = overrides->hashLog;
+    if (overrides->chainLog)     cParams->chainLog     = overrides->chainLog;
+    if (overrides->searchLog)    cParams->searchLog    = overrides->searchLog;
+    if (overrides->minMatch)     cParams->minMatch     = overrides->minMatch;
+    if (overrides->targetLength) cParams->targetLength = overrides->targetLength;
+    if (overrides->strategy)     cParams->strategy     = overrides->strategy;
+}
 
 ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
-        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize)
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode)
 {
     ZSTD_compressionParameters cParams;
     if (srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN && CCtxParams->srcSizeHint > 0) {
       srcSizeHint = CCtxParams->srcSizeHint;
     }
-    cParams = ZSTD_getCParams_internal(CCtxParams->compressionLevel, srcSizeHint, dictSize);
+    cParams = ZSTD_getCParams_internal(CCtxParams->compressionLevel, srcSizeHint, dictSize, mode);
     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.minMatch) cParams.minMatch = CCtxParams->cParams.minMatch;
-    if (CCtxParams->cParams.targetLength) cParams.targetLength = CCtxParams->cParams.targetLength;
-    if (CCtxParams->cParams.strategy) cParams.strategy = CCtxParams->cParams.strategy;
+    ZSTD_overrideCParams(&cParams, &CCtxParams->cParams);
     assert(!ZSTD_checkCParams(cParams));
     /* srcSizeHint == 0 means 0 */
-    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize);
+    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize, mode);
 }
 
 static size_t
@@ -1123,45 +1295,61 @@ ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
     return tableSpace + optSpace;
 }
 
-size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+        const ZSTD_compressionParameters* cParams,
+        const ldmParams_t* ldmParams,
+        const int isStatic,
+        const size_t buffInSize,
+        const size_t buffOutSize,
+        const U64 pledgedSrcSize)
 {
-    RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
-    {   ZSTD_compressionParameters const cParams =
-                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0);
-        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
-        U32    const divider = (cParams.minMatch==3) ? 3 : 4;
-        size_t const maxNbSeq = blockSize / divider;
-        size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize)
-                                + ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(seqDef))
-                                + 3 * ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(BYTE));
-        size_t const entropySpace = ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE);
-        size_t const blockStateSpace = 2 * ZSTD_cwksp_alloc_size(sizeof(ZSTD_compressedBlockState_t));
-        size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 1);
+    size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << cParams->windowLog), pledgedSrcSize));
+    size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
+    U32    const divider = (cParams->minMatch==3) ? 3 : 4;
+    size_t const maxNbSeq = blockSize / divider;
+    size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize)
+                            + ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(seqDef))
+                            + 3 * ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(BYTE));
+    size_t const entropySpace = ZSTD_cwksp_alloc_size(ENTROPY_WORKSPACE_SIZE);
+    size_t const blockStateSpace = 2 * ZSTD_cwksp_alloc_size(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_cwksp_alloc_size(ZSTD_ldm_getMaxNbSeq(params->ldmParams, blockSize) * sizeof(rawSeq));
+    size_t const ldmSpace = ZSTD_ldm_getTableSize(*ldmParams);
+    size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(*ldmParams, blockSize);
+    size_t const ldmSeqSpace = ldmParams->enableLdm ?
+        ZSTD_cwksp_alloc_size(maxNbLdmSeq * sizeof(rawSeq)) : 0;
 
-        /* estimateCCtxSize is for one-shot compression. So no buffers should
-         * be needed. However, we still allocate two 0-sized buffers, which can
-         * take space under ASAN. */
-        size_t const bufferSpace = ZSTD_cwksp_alloc_size(0)
-                                 + ZSTD_cwksp_alloc_size(0);
 
-        size_t const cctxSpace = ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx));
+    size_t const bufferSpace = ZSTD_cwksp_alloc_size(buffInSize)
+                             + ZSTD_cwksp_alloc_size(buffOutSize);
 
-        size_t const neededSpace =
-            cctxSpace +
-            entropySpace +
-            blockStateSpace +
-            ldmSpace +
-            ldmSeqSpace +
-            matchStateSize +
-            tokenSpace +
-            bufferSpace;
+    size_t const cctxSpace = isStatic ? ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx)) : 0;
 
-        DEBUGLOG(5, "estimate workspace : %u", (U32)neededSpace);
-        return neededSpace;
-    }
+    size_t const neededSpace =
+        cctxSpace +
+        entropySpace +
+        blockStateSpace +
+        ldmSpace +
+        ldmSeqSpace +
+        matchStateSize +
+        tokenSpace +
+        bufferSpace;
+
+    DEBUGLOG(5, "estimate workspace : %u", (U32)neededSpace);
+    return neededSpace;
+}
+
+size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
+
+    RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
+    /* estimateCCtxSize is for one-shot compression. So no buffers should
+     * be needed. However, we still allocate two 0-sized buffers, which can
+     * take space under ASAN. */
+    return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+        &cParams, &params->ldmParams, 1, 0, 0, ZSTD_CONTENTSIZE_UNKNOWN);
 }
 
 size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
@@ -1172,7 +1360,7 @@ size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
 
 static size_t ZSTD_estimateCCtxSize_internal(int compressionLevel)
 {
-    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0);
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
     return ZSTD_estimateCCtxSize_usingCParams(cParams);
 }
 
@@ -1191,15 +1379,18 @@ size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
 {
     RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
     {   ZSTD_compressionParameters const cParams =
-                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0);
-        size_t const CCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(params);
+                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
         size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
-        size_t const inBuffSize = ((size_t)1 << cParams.windowLog) + blockSize;
-        size_t const outBuffSize = ZSTD_compressBound(blockSize) + 1;
-        size_t const streamingSize = ZSTD_cwksp_alloc_size(inBuffSize)
-                                   + ZSTD_cwksp_alloc_size(outBuffSize);
-
-        return CCtxSize + streamingSize;
+        size_t const inBuffSize = (params->inBufferMode == ZSTD_bm_buffered)
+                ? ((size_t)1 << cParams.windowLog) + blockSize
+                : 0;
+        size_t const outBuffSize = (params->outBufferMode == ZSTD_bm_buffered)
+                ? ZSTD_compressBound(blockSize) + 1
+                : 0;
+
+        return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+            &cParams, &params->ldmParams, 1, inBuffSize, outBuffSize,
+            ZSTD_CONTENTSIZE_UNKNOWN);
     }
 }
 
@@ -1211,7 +1402,7 @@ size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
 
 static size_t ZSTD_estimateCStreamSize_internal(int compressionLevel)
 {
-    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0);
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
     return ZSTD_estimateCStreamSize_usingCParams(cParams);
 }
 
@@ -1305,16 +1496,6 @@ static void ZSTD_invalidateMatchState(ZSTD_matchState_t* ms)
 }
 
 /**
- * Indicates whether this compression proceeds directly from user-provided
- * source buffer to user-provided destination buffer (ZSTDb_not_buffered), or
- * whether the context needs to buffer the input/output (ZSTDb_buffered).
- */
-typedef enum {
-    ZSTDb_not_buffered,
-    ZSTDb_buffered
-} ZSTD_buffered_policy_e;
-
-/**
  * Controls, for this matchState reset, whether the tables need to be cleared /
  * prepared for the coming compression (ZSTDcrp_makeClean), or whether the
  * tables can be left unclean (ZSTDcrp_leaveDirty), because we know that a
@@ -1441,45 +1622,32 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
         U32    const divider = (params.cParams.minMatch==3) ? 3 : 4;
         size_t const maxNbSeq = blockSize / divider;
-        size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize)
-                                + ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(seqDef))
-                                + 3 * ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(BYTE));
-        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 buffOutSize = (zbuff == ZSTDb_buffered && params.outBufferMode == ZSTD_bm_buffered)
+                ? ZSTD_compressBound(blockSize) + 1
+                : 0;
+        size_t const buffInSize = (zbuff == ZSTDb_buffered && params.inBufferMode == ZSTD_bm_buffered)
+                ? windowSize + blockSize
+                : 0;
         size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(params.ldmParams, blockSize);
 
-        ZSTD_indexResetPolicy_e needsIndexReset = zc->initialized ? ZSTDirp_continue : ZSTDirp_reset;
+        int const indexTooClose = ZSTD_indexTooCloseToMax(zc->blockState.matchState.window);
+        ZSTD_indexResetPolicy_e needsIndexReset =
+            (!indexTooClose && zc->initialized) ? ZSTDirp_continue : ZSTDirp_reset;
 
-        if (ZSTD_indexTooCloseToMax(zc->blockState.matchState.window)) {
-            needsIndexReset = ZSTDirp_reset;
-        }
+        size_t const neededSpace =
+            ZSTD_estimateCCtxSize_usingCCtxParams_internal(
+                &params.cParams, &params.ldmParams, zc->staticSize != 0,
+                buffInSize, buffOutSize, pledgedSrcSize);
+        FORWARD_IF_ERROR(neededSpace, "cctx size estimate failed!");
 
         if (!zc->staticSize) ZSTD_cwksp_bump_oversized_duration(ws, 0);
 
         /* Check if workspace is large enough, alloc a new one if needed */
-        {   size_t const cctxSpace = zc->staticSize ? ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx)) : 0;
-            size_t const entropySpace = ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE);
-            size_t const blockStateSpace = 2 * ZSTD_cwksp_alloc_size(sizeof(ZSTD_compressedBlockState_t));
-            size_t const bufferSpace = ZSTD_cwksp_alloc_size(buffInSize) + ZSTD_cwksp_alloc_size(buffOutSize);
-            size_t const ldmSpace = ZSTD_ldm_getTableSize(params.ldmParams);
-            size_t const ldmSeqSpace = ZSTD_cwksp_alloc_size(maxNbLdmSeq * sizeof(rawSeq));
-
-            size_t const neededSpace =
-                cctxSpace +
-                entropySpace +
-                blockStateSpace +
-                ldmSpace +
-                ldmSeqSpace +
-                matchStateSize +
-                tokenSpace +
-                bufferSpace;
-
+        {
             int const workspaceTooSmall = ZSTD_cwksp_sizeof(ws) < neededSpace;
             int const workspaceWasteful = ZSTD_cwksp_check_wasteful(ws, neededSpace);
 
-            DEBUGLOG(4, "Need %zuKB workspace, including %zuKB for match state, and %zuKB for buffers",
-                        neededSpace>>10, matchStateSize>>10, bufferSpace>>10);
+            DEBUGLOG(4, "Need %zu B workspace", neededSpace);
             DEBUGLOG(4, "windowSize: %zu - blockSize: %zu", windowSize, blockSize);
 
             if (workspaceTooSmall || workspaceWasteful) {
@@ -1503,7 +1671,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
                 RETURN_ERROR_IF(zc->blockState.prevCBlock == NULL, memory_allocation, "couldn't allocate prevCBlock");
                 zc->blockState.nextCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
                 RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate nextCBlock");
-                zc->entropyWorkspace = (U32*) ZSTD_cwksp_reserve_object(ws, HUF_WORKSPACE_SIZE);
+                zc->entropyWorkspace = (U32*) ZSTD_cwksp_reserve_object(ws, ENTROPY_WORKSPACE_SIZE);
                 RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate entropyWorkspace");
         }   }
 
@@ -1534,6 +1702,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         zc->seqStore.maxNbLit = blockSize;
 
         /* buffers */
+        zc->bufferedPolicy = zbuff;
         zc->inBuffSize = buffInSize;
         zc->inBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffInSize);
         zc->outBuffSize = buffOutSize;
@@ -1546,7 +1715,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
                   ((size_t)1) << (params.ldmParams.hashLog -
                                   params.ldmParams.bucketSizeLog);
             zc->ldmState.bucketOffsets = ZSTD_cwksp_reserve_buffer(ws, ldmBucketSize);
-            memset(zc->ldmState.bucketOffsets, 0, ldmBucketSize);
+            ZSTD_memset(zc->ldmState.bucketOffsets, 0, ldmBucketSize);
         }
 
         /* sequences storage */
@@ -1570,7 +1739,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
             /* TODO: avoid memset? */
             size_t const ldmHSize = ((size_t)1) << params.ldmParams.hashLog;
             zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned(ws, ldmHSize * sizeof(ldmEntry_t));
-            memset(zc->ldmState.hashTable, 0, ldmHSize * sizeof(ldmEntry_t));
+            ZSTD_memset(zc->ldmState.hashTable, 0, ldmHSize * sizeof(ldmEntry_t));
             zc->ldmSequences = (rawSeq*)ZSTD_cwksp_reserve_aligned(ws, maxNbLdmSeq * sizeof(rawSeq));
             zc->maxNbLdmSequences = maxNbLdmSeq;
 
@@ -1579,6 +1748,12 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
             zc->ldmState.loadedDictEnd = 0;
         }
 
+        /* Due to alignment, when reusing a workspace, we can actually consume
+         * up to 3 extra bytes for alignment. See the comments in zstd_cwksp.h
+         */
+        assert(ZSTD_cwksp_used(ws) >= neededSpace &&
+               ZSTD_cwksp_used(ws) <= neededSpace + 3);
+
         DEBUGLOG(3, "wksp: finished allocating, %zd bytes remain available", ZSTD_cwksp_available_space(ws));
         zc->initialized = 1;
 
@@ -1618,12 +1793,14 @@ static int ZSTD_shouldAttachDict(const ZSTD_CDict* cdict,
                                  U64 pledgedSrcSize)
 {
     size_t cutoff = attachDictSizeCutoffs[cdict->matchState.cParams.strategy];
-    return ( pledgedSrcSize <= cutoff
-          || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
-          || params->attachDictPref == ZSTD_dictForceAttach )
-        && params->attachDictPref != ZSTD_dictForceCopy
-        && !params->forceWindow; /* dictMatchState isn't correctly
-                                 * handled in _enforceMaxDist */
+    int const dedicatedDictSearch = cdict->matchState.dedicatedDictSearch;
+    return dedicatedDictSearch
+        || ( ( pledgedSrcSize <= cutoff
+            || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+            || params->attachDictPref == ZSTD_dictForceAttach )
+          && params->attachDictPref != ZSTD_dictForceCopy
+          && !params->forceWindow ); /* dictMatchState isn't correctly
+                                      * handled in _enforceMaxDist */
 }
 
 static size_t
@@ -1633,17 +1810,24 @@ ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
                         U64 pledgedSrcSize,
                         ZSTD_buffered_policy_e zbuff)
 {
-    {   const ZSTD_compressionParameters* const cdict_cParams = &cdict->matchState.cParams;
+    {
+        ZSTD_compressionParameters adjusted_cdict_cParams = cdict->matchState.cParams;
         unsigned const windowLog = params.cParams.windowLog;
         assert(windowLog != 0);
         /* Resize working context table params for input only, since the dict
          * has its own tables. */
-        /* pledgeSrcSize == 0 means 0! */
-        params.cParams = ZSTD_adjustCParams_internal(*cdict_cParams, pledgedSrcSize, 0);
+        /* pledgedSrcSize == 0 means 0! */
+
+        if (cdict->matchState.dedicatedDictSearch) {
+            ZSTD_dedicatedDictSearch_revertCParams(&adjusted_cdict_cParams);
+        }
+
+        params.cParams = ZSTD_adjustCParams_internal(adjusted_cdict_cParams, pledgedSrcSize,
+                                                     cdict->dictContentSize, ZSTD_cpm_attachDict);
         params.cParams.windowLog = windowLog;
         FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
                                                  ZSTDcrp_makeClean, zbuff), "");
-        assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy);
+        assert(cctx->appliedParams.cParams.strategy == adjusted_cdict_cParams.strategy);
     }
 
     {   const U32 cdictEnd = (U32)( cdict->matchState.window.nextSrc
@@ -1670,7 +1854,7 @@ ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
     cctx->dictID = cdict->dictID;
 
     /* copy block state */
-    memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+    ZSTD_memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
 
     return 0;
 }
@@ -1683,6 +1867,8 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
 {
     const ZSTD_compressionParameters *cdict_cParams = &cdict->matchState.cParams;
 
+    assert(!cdict->matchState.dedicatedDictSearch);
+
     DEBUGLOG(4, "copying dictionary into context");
 
     {   unsigned const windowLog = params.cParams.windowLog;
@@ -1703,10 +1889,10 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
     {   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;
 
-        memcpy(cctx->blockState.matchState.hashTable,
+        ZSTD_memcpy(cctx->blockState.matchState.hashTable,
                cdict->matchState.hashTable,
                hSize * sizeof(U32));
-        memcpy(cctx->blockState.matchState.chainTable,
+        ZSTD_memcpy(cctx->blockState.matchState.chainTable,
                cdict->matchState.chainTable,
                chainSize * sizeof(U32));
     }
@@ -1715,7 +1901,7 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
     {   int const h3log = cctx->blockState.matchState.hashLog3;
         size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
         assert(cdict->matchState.hashLog3 == 0);
-        memset(cctx->blockState.matchState.hashTable3, 0, h3Size * sizeof(U32));
+        ZSTD_memset(cctx->blockState.matchState.hashTable3, 0, h3Size * sizeof(U32));
     }
 
     ZSTD_cwksp_mark_tables_clean(&cctx->workspace);
@@ -1731,7 +1917,7 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
     cctx->dictID = cdict->dictID;
 
     /* copy block state */
-    memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+    ZSTD_memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
 
     return 0;
 }
@@ -1775,7 +1961,7 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
     RETURN_ERROR_IF(srcCCtx->stage!=ZSTDcs_init, stage_wrong,
                     "Can't copy a ctx that's not in init stage.");
 
-    memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
+    ZSTD_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;
@@ -1797,13 +1983,13 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
         int const h3log = srcCCtx->blockState.matchState.hashLog3;
         size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
 
-        memcpy(dstCCtx->blockState.matchState.hashTable,
+        ZSTD_memcpy(dstCCtx->blockState.matchState.hashTable,
                srcCCtx->blockState.matchState.hashTable,
                hSize * sizeof(U32));
-        memcpy(dstCCtx->blockState.matchState.chainTable,
+        ZSTD_memcpy(dstCCtx->blockState.matchState.chainTable,
                srcCCtx->blockState.matchState.chainTable,
                chainSize * sizeof(U32));
-        memcpy(dstCCtx->blockState.matchState.hashTable3,
+        ZSTD_memcpy(dstCCtx->blockState.matchState.hashTable3,
                srcCCtx->blockState.matchState.hashTable3,
                h3Size * sizeof(U32));
     }
@@ -1821,7 +2007,7 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
     dstCCtx->dictID = srcCCtx->dictID;
 
     /* copy block state */
-    memcpy(dstCCtx->blockState.prevCBlock, srcCCtx->blockState.prevCBlock, sizeof(*srcCCtx->blockState.prevCBlock));
+    ZSTD_memcpy(dstCCtx->blockState.prevCBlock, srcCCtx->blockState.prevCBlock, sizeof(*srcCCtx->blockState.prevCBlock));
 
     return 0;
 }
@@ -1834,7 +2020,7 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
 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_buffered_policy_e const zbuff = srcCCtx->bufferedPolicy;
     ZSTD_STATIC_ASSERT((U32)ZSTDb_buffered==1);
     if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
     fParams.contentSizeFlag = (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN);
@@ -1861,7 +2047,7 @@ ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerVa
     assert((size & (ZSTD_ROWSIZE-1)) == 0);  /* multiple of ZSTD_ROWSIZE */
     assert(size < (1U<<31));   /* can be casted to int */
 
-#if defined (MEMORY_SANITIZER) && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
     /* To validate that the table re-use logic is sound, and that we don't
      * access table space that we haven't cleaned, we re-"poison" the table
      * space every time we mark it dirty.
@@ -1958,10 +2144,10 @@ static int ZSTD_useTargetCBlockSize(const ZSTD_CCtx_params* cctxParams)
     return (cctxParams->targetCBlockSize != 0);
 }
 
-/* ZSTD_compressSequences_internal():
+/* ZSTD_entropyCompressSequences_internal():
  * actually compresses both literals and sequences */
 MEM_STATIC size_t
-ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
+ZSTD_entropyCompressSequences_internal(seqStore_t* seqStorePtr,
                           const ZSTD_entropyCTables_t* prevEntropy,
                                 ZSTD_entropyCTables_t* nextEntropy,
                           const ZSTD_CCtx_params* cctxParams,
@@ -1971,7 +2157,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
 {
     const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
     ZSTD_strategy const strategy = cctxParams->cParams.strategy;
-    unsigned count[MaxSeq+1];
+    unsigned* count = (unsigned*)entropyWorkspace;
     FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
     FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
     FSE_CTable* CTable_MatchLength = nextEntropy->fse.matchlengthCTable;
@@ -1987,8 +2173,12 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
     BYTE* seqHead;
     BYTE* lastNCount = NULL;
 
-    DEBUGLOG(5, "ZSTD_compressSequences_internal (nbSeq=%zu)", nbSeq);
+    entropyWorkspace = count + (MaxSeq + 1);
+    entropyWkspSize -= (MaxSeq + 1) * sizeof(*count);
+
+    DEBUGLOG(4, "ZSTD_entropyCompressSequences_internal (nbSeq=%zu)", nbSeq);
     ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+    assert(entropyWkspSize >= HUF_WORKSPACE_SIZE);
 
     /* Compress literals */
     {   const BYTE* const literals = seqStorePtr->litStart;
@@ -2023,7 +2213,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
     assert(op <= oend);
     if (nbSeq==0) {
         /* Copy the old tables over as if we repeated them */
-        memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
+        ZSTD_memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
         return (size_t)(op - ostart);
     }
 
@@ -2148,7 +2338,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
 }
 
 MEM_STATIC size_t
-ZSTD_compressSequences(seqStore_t* seqStorePtr,
+ZSTD_entropyCompressSequences(seqStore_t* seqStorePtr,
                        const ZSTD_entropyCTables_t* prevEntropy,
                              ZSTD_entropyCTables_t* nextEntropy,
                        const ZSTD_CCtx_params* cctxParams,
@@ -2157,7 +2347,7 @@ ZSTD_compressSequences(seqStore_t* seqStorePtr,
                              void* entropyWorkspace, size_t entropyWkspSize,
                              int bmi2)
 {
-    size_t const cSize = ZSTD_compressSequences_internal(
+    size_t const cSize = ZSTD_entropyCompressSequences_internal(
                             seqStorePtr, prevEntropy, nextEntropy, cctxParams,
                             dst, dstCapacity,
                             entropyWorkspace, entropyWkspSize, bmi2);
@@ -2167,13 +2357,13 @@ ZSTD_compressSequences(seqStore_t* seqStorePtr,
      */
     if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity))
         return 0;  /* block not compressed */
-    FORWARD_IF_ERROR(cSize, "ZSTD_compressSequences_internal failed");
+    FORWARD_IF_ERROR(cSize, "ZSTD_entropyCompressSequences_internal failed");
 
     /* Check compressibility */
     {   size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, cctxParams->cParams.strategy);
         if (cSize >= maxCSize) return 0;  /* block not compressed */
     }
-
+    DEBUGLOG(4, "ZSTD_entropyCompressSequences() cSize: %zu\n", cSize);
     return cSize;
 }
 
@@ -2182,7 +2372,7 @@ ZSTD_compressSequences(seqStore_t* seqStorePtr,
  * assumption : strat is a valid strategy */
 ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_dictMode_e dictMode)
 {
-    static const ZSTD_blockCompressor blockCompressor[3][ZSTD_STRATEGY_MAX+1] = {
+    static const ZSTD_blockCompressor blockCompressor[4][ZSTD_STRATEGY_MAX+1] = {
         { ZSTD_compressBlock_fast  /* default for 0 */,
           ZSTD_compressBlock_fast,
           ZSTD_compressBlock_doubleFast,
@@ -2212,7 +2402,17 @@ ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_dictMo
           ZSTD_compressBlock_btlazy2_dictMatchState,
           ZSTD_compressBlock_btopt_dictMatchState,
           ZSTD_compressBlock_btultra_dictMatchState,
-          ZSTD_compressBlock_btultra_dictMatchState }
+          ZSTD_compressBlock_btultra_dictMatchState },
+        { NULL  /* default for 0 */,
+          NULL,
+          NULL,
+          ZSTD_compressBlock_greedy_dedicatedDictSearch,
+          ZSTD_compressBlock_lazy_dedicatedDictSearch,
+          ZSTD_compressBlock_lazy2_dedicatedDictSearch,
+          NULL,
+          NULL,
+          NULL,
+          NULL }
     };
     ZSTD_blockCompressor selectedCompressor;
     ZSTD_STATIC_ASSERT((unsigned)ZSTD_fast == 1);
@@ -2226,7 +2426,7 @@ ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_dictMo
 static void ZSTD_storeLastLiterals(seqStore_t* seqStorePtr,
                                    const BYTE* anchor, size_t lastLLSize)
 {
-    memcpy(seqStorePtr->lit, anchor, lastLLSize);
+    ZSTD_memcpy(seqStorePtr->lit, anchor, lastLLSize);
     seqStorePtr->lit += lastLLSize;
 }
 
@@ -2247,7 +2447,11 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
     /* Assert that we have correctly flushed the ctx params into the ms's copy */
     ZSTD_assertEqualCParams(zc->appliedParams.cParams, ms->cParams);
     if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
-        ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.minMatch);
+        if (zc->appliedParams.cParams.strategy >= ZSTD_btopt) {
+            ZSTD_ldm_skipRawSeqStoreBytes(&zc->externSeqStore, srcSize);
+        } else {
+            ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.minMatch);
+        }
         return ZSTDbss_noCompress; /* don't even attempt compression below a certain srcSize */
     }
     ZSTD_resetSeqStore(&(zc->seqStore));
@@ -2263,10 +2467,10 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
     /* 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);
+        const U32 curr = (U32)(istart-base);
         if (sizeof(ptrdiff_t)==8) assert(istart - base < (ptrdiff_t)(U32)(-1));   /* ensure no overflow */
-        if (current > ms->nextToUpdate + 384)
-            ms->nextToUpdate = current - MIN(192, (U32)(current - ms->nextToUpdate - 384));
+        if (curr > ms->nextToUpdate + 384)
+            ms->nextToUpdate = curr - MIN(192, (U32)(curr - ms->nextToUpdate - 384));
     }
 
     /* select and store sequences */
@@ -2286,7 +2490,7 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
                                        src, srcSize);
             assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
         } else if (zc->appliedParams.ldmParams.enableLdm) {
-            rawSeqStore_t ldmSeqStore = {NULL, 0, 0, 0};
+            rawSeqStore_t ldmSeqStore = kNullRawSeqStore;
 
             ldmSeqStore.seq = zc->ldmSequences;
             ldmSeqStore.capacity = zc->maxNbLdmSequences;
@@ -2303,6 +2507,7 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
             assert(ldmSeqStore.pos == ldmSeqStore.size);
         } else {   /* not long range mode */
             ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy, dictMode);
+            ms->ldmSeqStore = NULL;
             lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
         }
         {   const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
@@ -2314,17 +2519,25 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
 static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
 {
     const seqStore_t* seqStore = ZSTD_getSeqStore(zc);
-    const seqDef* seqs = seqStore->sequencesStart;
-    size_t seqsSize = seqStore->sequences - seqs;
+    const seqDef* seqStoreSeqs = seqStore->sequencesStart;
+    size_t seqStoreSeqSize = seqStore->sequences - seqStoreSeqs;
+    size_t seqStoreLiteralsSize = (size_t)(seqStore->lit - seqStore->litStart);
+    size_t literalsRead = 0;
+    size_t lastLLSize;
 
     ZSTD_Sequence* outSeqs = &zc->seqCollector.seqStart[zc->seqCollector.seqIndex];
-    size_t i; size_t position; int repIdx;
+    size_t i;
+    repcodes_t updatedRepcodes;
 
     assert(zc->seqCollector.seqIndex + 1 < zc->seqCollector.maxSequences);
-    for (i = 0, position = 0; i < seqsSize; ++i) {
-        outSeqs[i].offset = seqs[i].offset;
-        outSeqs[i].litLength = seqs[i].litLength;
-        outSeqs[i].matchLength = seqs[i].matchLength + MINMATCH;
+    /* Ensure we have enough space for last literals "sequence" */
+    assert(zc->seqCollector.maxSequences >= seqStoreSeqSize + 1);
+    ZSTD_memcpy(updatedRepcodes.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
+    for (i = 0; i < seqStoreSeqSize; ++i) {
+        U32 rawOffset = seqStoreSeqs[i].offset - ZSTD_REP_NUM;
+        outSeqs[i].litLength = seqStoreSeqs[i].litLength;
+        outSeqs[i].matchLength = seqStoreSeqs[i].matchLength + MINMATCH;
+        outSeqs[i].rep = 0;
 
         if (i == seqStore->longLengthPos) {
             if (seqStore->longLengthID == 1) {
@@ -2334,39 +2547,44 @@ static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
             }
         }
 
-        if (outSeqs[i].offset <= ZSTD_REP_NUM) {
-            outSeqs[i].rep = outSeqs[i].offset;
-            repIdx = (unsigned int)i - outSeqs[i].offset;
-
-            if (outSeqs[i].litLength == 0) {
-                if (outSeqs[i].offset < 3) {
-                    --repIdx;
+        if (seqStoreSeqs[i].offset <= ZSTD_REP_NUM) {
+            /* Derive the correct offset corresponding to a repcode */
+            outSeqs[i].rep = seqStoreSeqs[i].offset;
+            if (outSeqs[i].litLength != 0) {
+                rawOffset = updatedRepcodes.rep[outSeqs[i].rep - 1];
+            } else {
+                if (outSeqs[i].rep == 3) {
+                    rawOffset = updatedRepcodes.rep[0] - 1;
                 } else {
-                    repIdx = (unsigned int)i - 1;
+                    rawOffset = updatedRepcodes.rep[outSeqs[i].rep];
                 }
-                ++outSeqs[i].rep;
-            }
-            assert(repIdx >= -3);
-            outSeqs[i].offset = repIdx >= 0 ? outSeqs[repIdx].offset : repStartValue[-repIdx - 1];
-            if (outSeqs[i].rep == 4) {
-                --outSeqs[i].offset;
             }
-        } else {
-            outSeqs[i].offset -= ZSTD_REP_NUM;
         }
-
-        position += outSeqs[i].litLength;
-        outSeqs[i].matchPos = (unsigned int)position;
-        position += outSeqs[i].matchLength;
+        outSeqs[i].offset = rawOffset;
+        /* seqStoreSeqs[i].offset == offCode+1, and ZSTD_updateRep() expects offCode
+           so we provide seqStoreSeqs[i].offset - 1 */
+        updatedRepcodes = ZSTD_updateRep(updatedRepcodes.rep,
+                                         seqStoreSeqs[i].offset - 1,
+                                         seqStoreSeqs[i].litLength == 0);
+        literalsRead += outSeqs[i].litLength;
     }
-    zc->seqCollector.seqIndex += seqsSize;
+    /* Insert last literals (if any exist) in the block as a sequence with ml == off == 0.
+     * If there are no last literals, then we'll emit (of: 0, ml: 0, ll: 0), which is a marker
+     * for the block boundary, according to the API.
+     */
+    assert(seqStoreLiteralsSize >= literalsRead);
+    lastLLSize = seqStoreLiteralsSize - literalsRead;
+    outSeqs[i].litLength = (U32)lastLLSize;
+    outSeqs[i].matchLength = outSeqs[i].offset = outSeqs[i].rep = 0;
+    seqStoreSeqSize++;
+    zc->seqCollector.seqIndex += seqStoreSeqSize;
 }
 
-size_t ZSTD_getSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
-    size_t outSeqsSize, const void* src, size_t srcSize)
+size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
+                              size_t outSeqsSize, const void* src, size_t srcSize)
 {
     const size_t dstCapacity = ZSTD_compressBound(srcSize);
-    void* dst = ZSTD_malloc(dstCapacity, ZSTD_defaultCMem);
+    void* dst = ZSTD_customMalloc(dstCapacity, ZSTD_defaultCMem);
     SeqCollector seqCollector;
 
     RETURN_ERROR_IF(dst == NULL, memory_allocation, "NULL pointer!");
@@ -2378,16 +2596,47 @@ size_t ZSTD_getSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
     zc->seqCollector = seqCollector;
 
     ZSTD_compress2(zc, dst, dstCapacity, src, srcSize);
-    ZSTD_free(dst, ZSTD_defaultCMem);
+    ZSTD_customFree(dst, ZSTD_defaultCMem);
     return zc->seqCollector.seqIndex;
 }
 
-/* Returns true if the given block is a RLE block */
-static int ZSTD_isRLE(const BYTE *ip, size_t length) {
+size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize) {
+    size_t in = 0;
+    size_t out = 0;
+    for (; in < seqsSize; ++in) {
+        if (sequences[in].offset == 0 && sequences[in].matchLength == 0) {
+            if (in != seqsSize - 1) {
+                sequences[in+1].litLength += sequences[in].litLength;
+            }
+        } else {
+            sequences[out] = sequences[in];
+            ++out;
+        }
+    }
+    return out;
+}
+
+/* Unrolled loop to read four size_ts of input at a time. Returns 1 if is RLE, 0 if not. */
+static int ZSTD_isRLE(const BYTE* src, size_t length) {
+    const BYTE* ip = src;
+    const BYTE value = ip[0];
+    const size_t valueST = (size_t)((U64)value * 0x0101010101010101ULL);
+    const size_t unrollSize = sizeof(size_t) * 4;
+    const size_t unrollMask = unrollSize - 1;
+    const size_t prefixLength = length & unrollMask;
     size_t i;
-    if (length < 2) return 1;
-    for (i = 1; i < length; ++i) {
-        if (ip[0] != ip[i]) return 0;
+    size_t u;
+    if (length == 1) return 1;
+    /* Check if prefix is RLE first before using unrolled loop */
+    if (prefixLength && ZSTD_count(ip+1, ip, ip+prefixLength) != prefixLength-1) {
+        return 0;
+    }
+    for (i = prefixLength; i != length; i += unrollSize) {
+        for (u = 0; u < unrollSize; u += sizeof(size_t)) {
+            if (MEM_readST(ip + i + u) != valueST) {
+                return 0;
+            }
+        }
     }
     return 1;
 }
@@ -2434,18 +2683,25 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
 
     if (zc->seqCollector.collectSequences) {
         ZSTD_copyBlockSequences(zc);
+        ZSTD_confirmRepcodesAndEntropyTables(zc);
         return 0;
     }
 
     /* encode sequences and literals */
-    cSize = ZSTD_compressSequences(&zc->seqStore,
+    cSize = ZSTD_entropyCompressSequences(&zc->seqStore,
             &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
             &zc->appliedParams,
             dst, dstCapacity,
             srcSize,
-            zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
+            zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
             zc->bmi2);
 
+    if (zc->seqCollector.collectSequences) {
+        ZSTD_copyBlockSequences(zc);
+        return 0;
+    }
+
+
     if (frame &&
         /* We don't want to emit our first block as a RLE even if it qualifies because
          * doing so will cause the decoder (cli only) to throw a "should consume all input error."
@@ -2593,7 +2849,7 @@ static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
 
     assert(cctx->appliedParams.cParams.windowLog <= ZSTD_WINDOWLOG_MAX);
 
-    DEBUGLOG(5, "ZSTD_compress_frameChunk (blockSize=%u)", (unsigned)blockSize);
+    DEBUGLOG(4, "ZSTD_compress_frameChunk (blockSize=%u)", (unsigned)blockSize);
     if (cctx->appliedParams.fParams.checksumFlag && srcSize)
         XXH64_update(&cctx->xxhState, src, srcSize);
 
@@ -2673,7 +2929,6 @@ static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
                     "dst buf is too small to fit worst-case frame header size.");
     DEBUGLOG(4, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
                 !params->fParams.noDictIDFlag, (unsigned)dictID, (unsigned)dictIDSizeCode);
-
     if (params->format == ZSTD_f_zstd1) {
         MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
         pos = 4;
@@ -2725,6 +2980,7 @@ size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSe
     cctx->externSeqStore.size = nbSeq;
     cctx->externSeqStore.capacity = nbSeq;
     cctx->externSeqStore.pos = 0;
+    cctx->externSeqStore.posInSequence = 0;
     return 0;
 }
 
@@ -2862,8 +3118,12 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
         case ZSTD_greedy:
         case ZSTD_lazy:
         case ZSTD_lazy2:
-            if (chunk >= HASH_READ_SIZE)
+            if (chunk >= HASH_READ_SIZE && ms->dedicatedDictSearch) {
+                assert(chunk == remaining); /* must load everything in one go */
+                ZSTD_dedicatedDictSearch_lazy_loadDictionary(ms, ichunk-HASH_READ_SIZE);
+            } else if (chunk >= HASH_READ_SIZE) {
                 ZSTD_insertAndFindFirstIndex(ms, ichunk-HASH_READ_SIZE);
+            }
             break;
 
         case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
@@ -2887,22 +3147,28 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
 
 
 /* Dictionaries that assign zero probability to symbols that show up causes problems
-   when FSE encoding.  Refuse dictionaries that assign zero probability to symbols
-   that we may encounter during compression.
-   NOTE: This behavior is not standard and could be improved in the future. */
-static size_t ZSTD_checkDictNCount(short* normalizedCounter, unsigned dictMaxSymbolValue, unsigned maxSymbolValue) {
+ * when FSE encoding. Mark dictionaries with zero probability symbols as FSE_repeat_check
+ * and only dictionaries with 100% valid symbols can be assumed valid.
+ */
+static FSE_repeat ZSTD_dictNCountRepeat(short* normalizedCounter, unsigned dictMaxSymbolValue, unsigned maxSymbolValue)
+{
     U32 s;
-    RETURN_ERROR_IF(dictMaxSymbolValue < maxSymbolValue, dictionary_corrupted, "dict fse tables don't have all symbols");
+    if (dictMaxSymbolValue < maxSymbolValue) {
+        return FSE_repeat_check;
+    }
     for (s = 0; s <= maxSymbolValue; ++s) {
-        RETURN_ERROR_IF(normalizedCounter[s] == 0, dictionary_corrupted, "dict fse tables don't have all symbols");
+        if (normalizedCounter[s] == 0) {
+            return FSE_repeat_check;
+        }
     }
-    return 0;
+    return FSE_repeat_valid;
 }
 
 size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
-                         short* offcodeNCount, unsigned* offcodeMaxValue,
                          const void* const dict, size_t dictSize)
 {
+    short offcodeNCount[MaxOff+1];
+    unsigned offcodeMaxValue = MaxOff;
     const BYTE* dictPtr = (const BYTE*)dict;    /* skip magic num and dict ID */
     const BYTE* const dictEnd = dictPtr + dictSize;
     dictPtr += 8;
@@ -2924,16 +3190,16 @@ size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
     }
 
     {   unsigned offcodeLog;
-        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
         RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
         RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
-        /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
         /* fill all offset symbols to avoid garbage at end of table */
         RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
                 bs->entropy.fse.offcodeCTable,
                 offcodeNCount, MaxOff, offcodeLog,
                 workspace, HUF_WORKSPACE_SIZE)),
             dictionary_corrupted, "");
+        /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
         dictPtr += offcodeHeaderSize;
     }
 
@@ -2942,13 +3208,12 @@ size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
         size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
         RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
         RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
-        /* Every match length code must have non-zero probability */
-        FORWARD_IF_ERROR( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML), "");
         RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
                 bs->entropy.fse.matchlengthCTable,
                 matchlengthNCount, matchlengthMaxValue, matchlengthLog,
                 workspace, HUF_WORKSPACE_SIZE)),
             dictionary_corrupted, "");
+        bs->entropy.fse.matchlength_repeatMode = ZSTD_dictNCountRepeat(matchlengthNCount, matchlengthMaxValue, MaxML);
         dictPtr += matchlengthHeaderSize;
     }
 
@@ -2957,13 +3222,12 @@ size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
         size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
         RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
         RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
-        /* Every literal length code must have non-zero probability */
-        FORWARD_IF_ERROR( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL), "");
         RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
                 bs->entropy.fse.litlengthCTable,
                 litlengthNCount, litlengthMaxValue, litlengthLog,
                 workspace, HUF_WORKSPACE_SIZE)),
             dictionary_corrupted, "");
+        bs->entropy.fse.litlength_repeatMode = ZSTD_dictNCountRepeat(litlengthNCount, litlengthMaxValue, MaxLL);
         dictPtr += litlengthHeaderSize;
     }
 
@@ -2973,12 +3237,28 @@ size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
     bs->rep[2] = MEM_readLE32(dictPtr+8);
     dictPtr += 12;
 
+    {   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 */
+        }
+        /* All offset values <= dictContentSize + 128 KB must be representable for a valid table */
+        bs->entropy.fse.offcode_repeatMode = ZSTD_dictNCountRepeat(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff));
+
+        /* All repCodes must be <= dictContentSize and != 0 */
+        {   U32 u;
+            for (u=0; u<3; u++) {
+                RETURN_ERROR_IF(bs->rep[u] == 0, dictionary_corrupted, "");
+                RETURN_ERROR_IF(bs->rep[u] > dictContentSize, dictionary_corrupted, "");
+    }   }   }
+
     return dictPtr - (const BYTE*)dict;
 }
 
 /* Dictionary format :
  * See :
- * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format
+ * https://github.com/facebook/zstd/blob/release/doc/zstd_compression_format.md#dictionary-format
  */
 /*! ZSTD_loadZstdDictionary() :
  * @return : dictID, or an error code
@@ -2995,8 +3275,6 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
 {
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
-    short offcodeNCount[MaxOff+1];
-    unsigned offcodeMaxValue = MaxOff;
     size_t dictID;
     size_t eSize;
 
@@ -3005,32 +3283,16 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
     assert(MEM_readLE32(dictPtr) == ZSTD_MAGIC_DICTIONARY);
 
     dictID = params->fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr + 4 /* skip magic number */ );
-    eSize = ZSTD_loadCEntropy(bs, workspace, offcodeNCount, &offcodeMaxValue, dict, dictSize);
+    eSize = ZSTD_loadCEntropy(bs, workspace, dict, dictSize);
     FORWARD_IF_ERROR(eSize, "ZSTD_loadCEntropy failed");
     dictPtr += eSize;
 
-    {   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 */
-        }
-        /* All offset values <= dictContentSize + 128 KB must be representable */
-        FORWARD_IF_ERROR(ZSTD_checkDictNCount(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff)), "");
-        /* All repCodes must be <= dictContentSize and != 0*/
-        {   U32 u;
-            for (u=0; u<3; u++) {
-                RETURN_ERROR_IF(bs->rep[u] == 0, dictionary_corrupted, "");
-                RETURN_ERROR_IF(bs->rep[u] > dictContentSize, dictionary_corrupted, "");
-        }   }
-
-        bs->entropy.fse.offcode_repeatMode = FSE_repeat_valid;
-        bs->entropy.fse.matchlength_repeatMode = FSE_repeat_valid;
-        bs->entropy.fse.litlength_repeatMode = FSE_repeat_valid;
+    {
+        size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
         FORWARD_IF_ERROR(ZSTD_loadDictionaryContent(
             ms, NULL, ws, params, dictPtr, dictContentSize, dtlm), "");
-        return dictID;
     }
+    return dictID;
 }
 
 /** ZSTD_compress_insertDictionary() :
@@ -3074,7 +3336,7 @@ ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
 }
 
 #define ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF (128 KB)
-#define ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER (6)
+#define ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER (6ULL)
 
 /*! ZSTD_compressBegin_internal() :
  * @return : 0, or an error code */
@@ -3106,7 +3368,7 @@ static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
                 ZSTD_compress_insertDictionary(
                         cctx->blockState.prevCBlock, &cctx->blockState.matchState,
                         &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, cdict->dictContent,
-                        cdict->dictContentSize, dictContentType, dtlm,
+                        cdict->dictContentSize, cdict->dictContentType, dtlm,
                         cctx->entropyWorkspace)
               : ZSTD_compress_insertDictionary(
                         cctx->blockState.prevCBlock, &cctx->blockState.matchState,
@@ -3153,7 +3415,7 @@ size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
 
 size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
+    ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_noAttachDict);
     ZSTD_CCtx_params const cctxParams =
             ZSTD_assignParamsToCCtxParams(&cctx->requestedParams, &params);
     DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (unsigned)dictSize);
@@ -3234,7 +3496,6 @@ size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
     return cSize + endResult;
 }
 
-
 static size_t ZSTD_compress_internal (ZSTD_CCtx* cctx,
                                       void* dst, size_t dstCapacity,
                                 const void* src, size_t srcSize,
@@ -3287,7 +3548,7 @@ size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx,
                          const void* dict, size_t dictSize,
                                int compressionLevel)
 {
-    ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, srcSize, dict ? dictSize : 0);
+    ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, srcSize, dict ? dictSize : 0, ZSTD_cpm_noAttachDict);
     ZSTD_CCtx_params cctxParams = ZSTD_assignParamsToCCtxParams(&cctx->requestedParams, &params);
     DEBUGLOG(4, "ZSTD_compress_usingDict (srcSize=%u)", (unsigned)srcSize);
     assert(params.fParams.contentSizeFlag == 1);
@@ -3309,10 +3570,17 @@ size_t ZSTD_compress(void* dst, size_t dstCapacity,
                      int compressionLevel)
 {
     size_t result;
+#if ZSTD_COMPRESS_HEAPMODE
+    ZSTD_CCtx* cctx = ZSTD_createCCtx();
+    RETURN_ERROR_IF(!cctx, memory_allocation, "ZSTD_createCCtx failed");
+    result = ZSTD_compressCCtx(cctx, dst, dstCapacity, src, srcSize, compressionLevel);
+    ZSTD_freeCCtx(cctx);
+#else
     ZSTD_CCtx ctxBody;
     ZSTD_initCCtx(&ctxBody, ZSTD_defaultCMem);
     result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
     ZSTD_freeCCtxContent(&ctxBody);   /* can't free ctxBody itself, as it's on stack; free only heap content */
+#endif
     return result;
 }
 
@@ -3335,7 +3603,7 @@ size_t ZSTD_estimateCDictSize_advanced(
 
 size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel)
 {
-    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
     return ZSTD_estimateCDictSize_advanced(dictSize, cParams, ZSTD_dlm_byCopy);
 }
 
@@ -3353,20 +3621,25 @@ static size_t ZSTD_initCDict_internal(
               const void* dictBuffer, size_t dictSize,
                     ZSTD_dictLoadMethod_e dictLoadMethod,
                     ZSTD_dictContentType_e dictContentType,
-                    ZSTD_compressionParameters cParams)
+                    ZSTD_CCtx_params params)
 {
     DEBUGLOG(3, "ZSTD_initCDict_internal (dictContentType:%u)", (unsigned)dictContentType);
-    assert(!ZSTD_checkCParams(cParams));
-    cdict->matchState.cParams = cParams;
+    assert(!ZSTD_checkCParams(params.cParams));
+    cdict->matchState.cParams = params.cParams;
+    cdict->matchState.dedicatedDictSearch = params.enableDedicatedDictSearch;
+    if (cdict->matchState.dedicatedDictSearch && dictSize > ZSTD_CHUNKSIZE_MAX) {
+        cdict->matchState.dedicatedDictSearch = 0;
+    }
     if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dictBuffer) || (!dictSize)) {
         cdict->dictContent = dictBuffer;
     } else {
          void *internalBuffer = ZSTD_cwksp_reserve_object(&cdict->workspace, ZSTD_cwksp_align(dictSize, sizeof(void*)));
         RETURN_ERROR_IF(!internalBuffer, memory_allocation, "NULL pointer!");
         cdict->dictContent = internalBuffer;
-        memcpy(internalBuffer, dictBuffer, dictSize);
+        ZSTD_memcpy(internalBuffer, dictBuffer, dictSize);
     }
     cdict->dictContentSize = dictSize;
+    cdict->dictContentType = dictContentType;
 
     cdict->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cdict->workspace, HUF_WORKSPACE_SIZE);
 
@@ -3376,18 +3649,15 @@ static size_t ZSTD_initCDict_internal(
     FORWARD_IF_ERROR(ZSTD_reset_matchState(
         &cdict->matchState,
         &cdict->workspace,
-        &cParams,
+        &params.cParams,
         ZSTDcrp_makeClean,
         ZSTDirp_reset,
         ZSTD_resetTarget_CDict), "");
     /* (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.compressionLevel = ZSTD_CLEVEL_DEFAULT;
         params.fParams.contentSizeFlag = 1;
-        params.cParams = cParams;
         {   size_t const dictID = ZSTD_compress_insertDictionary(
                     &cdict->cBlockState, &cdict->matchState, NULL, &cdict->workspace,
                     &params, cdict->dictContent, cdict->dictContentSize,
@@ -3401,13 +3671,11 @@ static size_t ZSTD_initCDict_internal(
     return 0;
 }
 
-ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
+static ZSTD_CDict* ZSTD_createCDict_advanced_internal(size_t dictSize,
                                       ZSTD_dictLoadMethod_e dictLoadMethod,
-                                      ZSTD_dictContentType_e dictContentType,
                                       ZSTD_compressionParameters cParams, ZSTD_customMem customMem)
 {
-    DEBUGLOG(3, "ZSTD_createCDict_advanced, mode %u", (unsigned)dictContentType);
-    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
 
     {   size_t const workspaceSize =
             ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict)) +
@@ -3415,16 +3683,16 @@ ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
             ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0) +
             (dictLoadMethod == ZSTD_dlm_byRef ? 0
              : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void*))));
-        void* const workspace = ZSTD_malloc(workspaceSize, customMem);
+        void* const workspace = ZSTD_customMalloc(workspaceSize, customMem);
         ZSTD_cwksp ws;
         ZSTD_CDict* cdict;
 
         if (!workspace) {
-            ZSTD_free(workspace, customMem);
+            ZSTD_customFree(workspace, customMem);
             return NULL;
         }
 
-        ZSTD_cwksp_init(&ws, workspace, workspaceSize);
+        ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_dynamic_alloc);
 
         cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
         assert(cdict != NULL);
@@ -3432,35 +3700,94 @@ ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
         cdict->customMem = customMem;
         cdict->compressionLevel = 0; /* signals advanced API usage */
 
-        if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
-                                        dictBuffer, dictSize,
-                                        dictLoadMethod, dictContentType,
-                                        cParams) )) {
-            ZSTD_freeCDict(cdict);
-            return NULL;
-        }
-
         return cdict;
     }
 }
 
+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)
+{
+    ZSTD_CCtx_params cctxParams;
+    ZSTD_memset(&cctxParams, 0, sizeof(cctxParams));
+    ZSTD_CCtxParams_init(&cctxParams, 0);
+    cctxParams.cParams = cParams;
+    cctxParams.customMem = customMem;
+    return ZSTD_createCDict_advanced2(
+        dictBuffer, dictSize,
+        dictLoadMethod, dictContentType,
+        &cctxParams, customMem);
+}
+
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced2(
+        const void* dict, size_t dictSize,
+        ZSTD_dictLoadMethod_e dictLoadMethod,
+        ZSTD_dictContentType_e dictContentType,
+        const ZSTD_CCtx_params* originalCctxParams,
+        ZSTD_customMem customMem)
+{
+    ZSTD_CCtx_params cctxParams = *originalCctxParams;
+    ZSTD_compressionParameters cParams;
+    ZSTD_CDict* cdict;
+
+    DEBUGLOG(3, "ZSTD_createCDict_advanced2, mode %u", (unsigned)dictContentType);
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+
+    if (cctxParams.enableDedicatedDictSearch) {
+        cParams = ZSTD_dedicatedDictSearch_getCParams(
+            cctxParams.compressionLevel, dictSize);
+        ZSTD_overrideCParams(&cParams, &cctxParams.cParams);
+    } else {
+        cParams = ZSTD_getCParamsFromCCtxParams(
+            &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    }
+
+    if (!ZSTD_dedicatedDictSearch_isSupported(&cParams)) {
+        /* Fall back to non-DDSS params */
+        cctxParams.enableDedicatedDictSearch = 0;
+        cParams = ZSTD_getCParamsFromCCtxParams(
+            &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    }
+
+    cctxParams.cParams = cParams;
+
+    cdict = ZSTD_createCDict_advanced_internal(dictSize,
+                        dictLoadMethod, cctxParams.cParams,
+                        customMem);
+
+    if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
+                                    dict, dictSize,
+                                    dictLoadMethod, dictContentType,
+                                    cctxParams) )) {
+        ZSTD_freeCDict(cdict);
+        return NULL;
+    }
+
+    return cdict;
+}
+
 ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
-    ZSTD_CDict* cdict = ZSTD_createCDict_advanced(dict, dictSize,
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dict, dictSize,
                                                   ZSTD_dlm_byCopy, ZSTD_dct_auto,
                                                   cParams, ZSTD_defaultCMem);
     if (cdict)
-        cdict->compressionLevel = compressionLevel == 0 ? ZSTD_CLEVEL_DEFAULT : compressionLevel;
+        cdict->compressionLevel = (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel;
     return cdict;
 }
 
 ZSTD_CDict* ZSTD_createCDict_byReference(const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
-    return ZSTD_createCDict_advanced(dict, dictSize,
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_createCDict);
+    ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dict, dictSize,
                                      ZSTD_dlm_byRef, ZSTD_dct_auto,
                                      cParams, ZSTD_defaultCMem);
+    if (cdict)
+        cdict->compressionLevel = (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel;
+    return cdict;
 }
 
 size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
@@ -3470,7 +3797,7 @@ size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
         int cdictInWorkspace = ZSTD_cwksp_owns_buffer(&cdict->workspace, cdict);
         ZSTD_cwksp_free(&cdict->workspace, cMem);
         if (!cdictInWorkspace) {
-            ZSTD_free(cdict, cMem);
+            ZSTD_customFree(cdict, cMem);
         }
         return 0;
     }
@@ -3503,12 +3830,13 @@ const ZSTD_CDict* ZSTD_initStaticCDict(
                             + ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
                             + matchStateSize;
     ZSTD_CDict* cdict;
+    ZSTD_CCtx_params params;
 
     if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
 
     {
         ZSTD_cwksp ws;
-        ZSTD_cwksp_init(&ws, workspace, workspaceSize);
+        ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_static_alloc);
         cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
         if (cdict == NULL) return NULL;
         ZSTD_cwksp_move(&cdict->workspace, &ws);
@@ -3518,10 +3846,13 @@ const ZSTD_CDict* ZSTD_initStaticCDict(
         (unsigned)workspaceSize, (unsigned)neededSize, (unsigned)(workspaceSize < neededSize));
     if (workspaceSize < neededSize) return NULL;
 
+    ZSTD_CCtxParams_init(&params, 0);
+    params.cParams = cParams;
+
     if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
                                               dict, dictSize,
                                               dictLoadMethod, dictContentType,
-                                              cParams) ))
+                                              params) ))
         return NULL;
 
     return cdict;
@@ -3533,6 +3864,17 @@ ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict)
     return cdict->matchState.cParams;
 }
 
+/*! ZSTD_getDictID_fromCDict() :
+ *  Provides the dictID of the dictionary loaded into `cdict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+unsigned ZSTD_getDictID_fromCDict(const ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;
+    return cdict->dictID;
+}
+
+
 /* ZSTD_compressBegin_usingCDict_advanced() :
  * cdict must be != NULL */
 size_t ZSTD_compressBegin_usingCDict_advanced(
@@ -3640,32 +3982,12 @@ 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 params, unsigned long long const pledgedSrcSize)
+static ZSTD_cParamMode_e ZSTD_getCParamMode(ZSTD_CDict const* cdict, ZSTD_CCtx_params const* params, U64 pledgedSrcSize)
 {
-    DEBUGLOG(4, "ZSTD_resetCStream_internal");
-    /* Finalize the compression parameters */
-    params.cParams = ZSTD_getCParamsFromCCtxParams(&params, pledgedSrcSize, dictSize);
-    /* 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 */
-
-    FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
-                                         dict, dictSize, dictContentType, ZSTD_dtlm_fast,
-                                         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 */
+    if (cdict != NULL && ZSTD_shouldAttachDict(cdict, params, pledgedSrcSize))
+        return ZSTD_cpm_attachDict;
+    else
+        return ZSTD_cpm_noAttachDict;
 }
 
 /* ZSTD_resetCStream():
@@ -3815,12 +4137,17 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
 
     /* check expectations */
     DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%u", (unsigned)flushMode);
-    assert(zcs->inBuff != NULL);
-    assert(zcs->inBuffSize > 0);
-    assert(zcs->outBuff !=  NULL);
-    assert(zcs->outBuffSize > 0);
+    if (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered) {
+        assert(zcs->inBuff != NULL);
+        assert(zcs->inBuffSize > 0);
+    }
+    if (zcs->appliedParams.outBufferMode == ZSTD_bm_buffered) {
+        assert(zcs->outBuff !=  NULL);
+        assert(zcs->outBuffSize > 0);
+    }
     assert(output->pos <= output->size);
     assert(input->pos <= input->size);
+    assert((U32)flushMode <= (U32)ZSTD_e_end);
 
     while (someMoreWork) {
         switch(zcs->streamStage)
@@ -3830,7 +4157,8 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
 
         case zcss_load:
             if ( (flushMode == ZSTD_e_end)
-              && ((size_t)(oend-op) >= ZSTD_compressBound(iend-ip))  /* enough dstCapacity */
+              && ( (size_t)(oend-op) >= ZSTD_compressBound(iend-ip)     /* Enough output space */
+                || zcs->appliedParams.outBufferMode == ZSTD_bm_stable)  /* OR we are allowed to return dstSizeTooSmall */
               && (zcs->inBuffPos == 0) ) {
                 /* shortcut to compression pass directly into output buffer */
                 size_t const cSize = ZSTD_compressEnd(zcs,
@@ -3843,8 +4171,9 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
                 someMoreWork = 0; break;
             }
-            /* complete loading into inBuffer */
-            {   size_t const toLoad = zcs->inBuffTarget - zcs->inBuffPos;
+            /* complete loading into inBuffer in buffered mode */
+            if (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered) {
+                size_t const toLoad = zcs->inBuffTarget - zcs->inBuffPos;
                 size_t const loaded = ZSTD_limitCopy(
                                         zcs->inBuff + zcs->inBuffPos, toLoad,
                                         ip, iend-ip);
@@ -3864,31 +4193,49 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
             }
             /* compress current block (note : this stage cannot be stopped in the middle) */
             DEBUGLOG(5, "stream compression stage (flushMode==%u)", flushMode);
-            {   void* cDst;
+            {   int const inputBuffered = (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered);
+                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))
+                size_t const iSize = inputBuffered
+                    ? zcs->inBuffPos - zcs->inToCompress
+                    : MIN((size_t)(iend - ip), zcs->blockSize);
+                if (oSize >= ZSTD_compressBound(iSize) || zcs->appliedParams.outBufferMode == ZSTD_bm_stable)
                     cDst = op;   /* compress into output buffer, to skip flush stage */
                 else
                     cDst = zcs->outBuff, oSize = zcs->outBuffSize;
-                cSize = lastBlock ?
-                        ZSTD_compressEnd(zcs, cDst, oSize,
-                                    zcs->inBuff + zcs->inToCompress, iSize) :
-                        ZSTD_compressContinue(zcs, cDst, oSize,
-                                    zcs->inBuff + zcs->inToCompress, iSize);
-                FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
-                zcs->frameEnded = lastBlock;
-                /* prepare next block */
-                zcs->inBuffTarget = zcs->inBuffPos + zcs->blockSize;
-                if (zcs->inBuffTarget > zcs->inBuffSize)
-                    zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;
-                DEBUGLOG(5, "inBuffTarget:%u / inBuffSize:%u",
-                         (unsigned)zcs->inBuffTarget, (unsigned)zcs->inBuffSize);
-                if (!lastBlock)
-                    assert(zcs->inBuffTarget <= zcs->inBuffSize);
-                zcs->inToCompress = zcs->inBuffPos;
+                if (inputBuffered) {
+                    unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip==iend);
+                    cSize = lastBlock ?
+                            ZSTD_compressEnd(zcs, cDst, oSize,
+                                        zcs->inBuff + zcs->inToCompress, iSize) :
+                            ZSTD_compressContinue(zcs, cDst, oSize,
+                                        zcs->inBuff + zcs->inToCompress, iSize);
+                    FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
+                    zcs->frameEnded = lastBlock;
+                    /* prepare next block */
+                    zcs->inBuffTarget = zcs->inBuffPos + zcs->blockSize;
+                    if (zcs->inBuffTarget > zcs->inBuffSize)
+                        zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;
+                    DEBUGLOG(5, "inBuffTarget:%u / inBuffSize:%u",
+                            (unsigned)zcs->inBuffTarget, (unsigned)zcs->inBuffSize);
+                    if (!lastBlock)
+                        assert(zcs->inBuffTarget <= zcs->inBuffSize);
+                    zcs->inToCompress = zcs->inBuffPos;
+                } else {
+                    unsigned const lastBlock = (ip + iSize == iend);
+                    assert(flushMode == ZSTD_e_end /* Already validated */);
+                    cSize = lastBlock ?
+                            ZSTD_compressEnd(zcs, cDst, oSize, ip, iSize) :
+                            ZSTD_compressContinue(zcs, cDst, oSize, ip, iSize);
+                    /* Consume the input prior to error checking to mirror buffered mode. */
+                    if (iSize > 0)
+                        ip += iSize;
+                    FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
+                    zcs->frameEnded = lastBlock;
+                    if (lastBlock)
+                        assert(ip == iend);
+                }
                 if (cDst == op) {  /* no need to flush */
                     op += cSize;
                     if (zcs->frameEnded) {
@@ -3905,6 +4252,7 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
 	    /* fall-through */
         case zcss_flush:
             DEBUGLOG(5, "flush stage");
+            assert(zcs->appliedParams.outBufferMode == ZSTD_bm_buffered);
             {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
                 size_t const flushed = ZSTD_limitCopy(op, (size_t)(oend-op),
                             zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
@@ -3959,6 +4307,116 @@ size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuf
     return ZSTD_nextInputSizeHint_MTorST(zcs);
 }
 
+/* After a compression call set the expected input/output buffer.
+ * This is validated at the start of the next compression call.
+ */
+static void ZSTD_setBufferExpectations(ZSTD_CCtx* cctx, ZSTD_outBuffer const* output, ZSTD_inBuffer const* input)
+{
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        cctx->expectedInBuffer = *input;
+    }
+    if (cctx->appliedParams.outBufferMode == ZSTD_bm_stable) {
+        cctx->expectedOutBufferSize = output->size - output->pos;
+    }
+}
+
+/* Validate that the input/output buffers match the expectations set by
+ * ZSTD_setBufferExpectations.
+ */
+static size_t ZSTD_checkBufferStability(ZSTD_CCtx const* cctx,
+                                        ZSTD_outBuffer const* output,
+                                        ZSTD_inBuffer const* input,
+                                        ZSTD_EndDirective endOp)
+{
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        ZSTD_inBuffer const expect = cctx->expectedInBuffer;
+        if (expect.src != input->src || expect.pos != input->pos || expect.size != input->size)
+            RETURN_ERROR(srcBuffer_wrong, "ZSTD_c_stableInBuffer enabled but input differs!");
+        if (endOp != ZSTD_e_end)
+            RETURN_ERROR(srcBuffer_wrong, "ZSTD_c_stableInBuffer can only be used with ZSTD_e_end!");
+    }
+    if (cctx->appliedParams.outBufferMode == ZSTD_bm_stable) {
+        size_t const outBufferSize = output->size - output->pos;
+        if (cctx->expectedOutBufferSize != outBufferSize)
+            RETURN_ERROR(dstBuffer_wrong, "ZSTD_c_stableOutBuffer enabled but output size differs!");
+    }
+    return 0;
+}
+
+static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
+                                             ZSTD_EndDirective endOp,
+                                             size_t inSize) {
+    ZSTD_CCtx_params params = cctx->requestedParams;
+    ZSTD_prefixDict const prefixDict = cctx->prefixDict;
+    FORWARD_IF_ERROR( ZSTD_initLocalDict(cctx) , ""); /* Init the local dict if present. */
+    ZSTD_memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));   /* single usage */
+    assert(prefixDict.dict==NULL || cctx->cdict==NULL);    /* only one can be set */
+    if (cctx->cdict)
+        params.compressionLevel = cctx->cdict->compressionLevel; /* let cdict take priority in terms of compression level */
+    DEBUGLOG(4, "ZSTD_compressStream2 : transparent init stage");
+    if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = inSize + 1;  /* auto-fix pledgedSrcSize */
+    {
+        size_t const dictSize = prefixDict.dict
+                ? prefixDict.dictSize
+                : (cctx->cdict ? cctx->cdict->dictContentSize : 0);
+        ZSTD_cParamMode_e const mode = ZSTD_getCParamMode(cctx->cdict, &params, cctx->pledgedSrcSizePlusOne - 1);
+        params.cParams = ZSTD_getCParamsFromCCtxParams(
+                &params, cctx->pledgedSrcSizePlusOne-1,
+                dictSize, mode);
+    }
+
+    if (ZSTD_CParams_shouldEnableLdm(&params.cParams)) {
+        /* Enable LDM by default for optimal parser and window size >= 128MB */
+        DEBUGLOG(4, "LDM enabled by default (window size >= 128MB, strategy >= btopt)");
+        params.ldmParams.enableLdm = 1;
+    }
+
+#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) {
+            DEBUGLOG(4, "ZSTD_compressStream2: creating new mtctx for nbWorkers=%u",
+                        params.nbWorkers);
+            cctx->mtctx = ZSTDMT_createCCtx_advanced((U32)params.nbWorkers, cctx->customMem, cctx->pool);
+            RETURN_ERROR_IF(cctx->mtctx == NULL, memory_allocation, "NULL pointer!");
+        }
+        /* mt compression */
+        DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbWorkers=%u", params.nbWorkers);
+        FORWARD_IF_ERROR( ZSTDMT_initCStream_internal(
+                    cctx->mtctx,
+                    prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType,
+                    cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) , "");
+        cctx->streamStage = zcss_load;
+        cctx->appliedParams = params;
+    } else
+#endif
+    {   U64 const pledgedSrcSize = cctx->pledgedSrcSizePlusOne - 1;
+        assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+        FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
+                prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType, ZSTD_dtlm_fast,
+                cctx->cdict,
+                &params, pledgedSrcSize,
+                ZSTDb_buffered) , "");
+        assert(cctx->appliedParams.nbWorkers == 0);
+        cctx->inToCompress = 0;
+        cctx->inBuffPos = 0;
+        if (cctx->appliedParams.inBufferMode == ZSTD_bm_buffered) {
+            /* 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->inBuffTarget = cctx->blockSize + (cctx->blockSize == pledgedSrcSize);
+        } else {
+            cctx->inBuffTarget = 0;
+        }
+        cctx->outBuffContentSize = cctx->outBuffFlushedSize = 0;
+        cctx->streamStage = zcss_load;
+        cctx->frameEnded = 0;
+    }
+    return 0;
+}
 
 size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
                              ZSTD_outBuffer* output,
@@ -3967,82 +4425,65 @@ size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
 {
     DEBUGLOG(5, "ZSTD_compressStream2, endOp=%u ", (unsigned)endOp);
     /* check conditions */
-    RETURN_ERROR_IF(output->pos > output->size, GENERIC, "invalid buffer");
-    RETURN_ERROR_IF(input->pos  > input->size, GENERIC, "invalid buffer");
-    assert(cctx!=NULL);
+    RETURN_ERROR_IF(output->pos > output->size, dstSize_tooSmall, "invalid output buffer");
+    RETURN_ERROR_IF(input->pos  > input->size, srcSize_wrong, "invalid input buffer");
+    RETURN_ERROR_IF((U32)endOp > (U32)ZSTD_e_end, parameter_outOfBound, "invalid endDirective");
+    assert(cctx != NULL);
 
     /* transparent initialization stage */
     if (cctx->streamStage == zcss_init) {
-        ZSTD_CCtx_params params = cctx->requestedParams;
-        ZSTD_prefixDict const prefixDict = cctx->prefixDict;
-        FORWARD_IF_ERROR( ZSTD_initLocalDict(cctx) , ""); /* Init the local dict if present. */
-        memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));   /* single usage */
-        assert(prefixDict.dict==NULL || cctx->cdict==NULL);    /* only one can be set */
-        DEBUGLOG(4, "ZSTD_compressStream2 : 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) {
-                DEBUGLOG(4, "ZSTD_compressStream2: creating new mtctx for nbWorkers=%u",
-                            params.nbWorkers);
-                cctx->mtctx = ZSTDMT_createCCtx_advanced((U32)params.nbWorkers, cctx->customMem);
-                RETURN_ERROR_IF(cctx->mtctx == NULL, memory_allocation, "NULL pointer!");
-            }
-            /* mt compression */
-            DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbWorkers=%u", params.nbWorkers);
-            FORWARD_IF_ERROR( ZSTDMT_initCStream_internal(
-                        cctx->mtctx,
-                        prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType,
-                        cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) , "");
-            cctx->streamStage = zcss_load;
-            cctx->appliedParams.nbWorkers = params.nbWorkers;
-        } else
-#endif
-        {   FORWARD_IF_ERROR( 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);
-    }   }
+        FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, endOp, input->size), "CompressStream2 initialization failed");
+        ZSTD_setBufferExpectations(cctx, output, input);    /* Set initial buffer expectations now that we've initialized */
+    }
     /* end of transparent initialization stage */
 
+    FORWARD_IF_ERROR(ZSTD_checkBufferStability(cctx, output, input, endOp), "invalid buffers");
     /* compression stage */
 #ifdef ZSTD_MULTITHREAD
     if (cctx->appliedParams.nbWorkers > 0) {
-        int const forceMaxProgress = (endOp == ZSTD_e_flush || endOp == ZSTD_e_end);
         size_t flushMin;
-        assert(forceMaxProgress || endOp == ZSTD_e_continue /* Protection for a new flush type */);
         if (cctx->cParamsChanged) {
             ZSTDMT_updateCParams_whileCompressing(cctx->mtctx, &cctx->requestedParams);
             cctx->cParamsChanged = 0;
         }
-        do {
+        for (;;) {
+            size_t const ipos = input->pos;
+            size_t const opos = output->pos;
             flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
             if ( ZSTD_isError(flushMin)
               || (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
                 ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
             }
             FORWARD_IF_ERROR(flushMin, "ZSTDMT_compressStream_generic failed");
-        } while (forceMaxProgress && flushMin != 0 && output->pos < output->size);
+
+            if (endOp == ZSTD_e_continue) {
+                /* We only require some progress with ZSTD_e_continue, not maximal progress.
+                 * We're done if we've consumed or produced any bytes, or either buffer is
+                 * full.
+                 */
+                if (input->pos != ipos || output->pos != opos || input->pos == input->size || output->pos == output->size)
+                    break;
+            } else {
+                assert(endOp == ZSTD_e_flush || endOp == ZSTD_e_end);
+                /* We require maximal progress. We're done when the flush is complete or the
+                 * output buffer is full.
+                 */
+                if (flushMin == 0 || output->pos == output->size)
+                    break;
+            }
+        }
         DEBUGLOG(5, "completed ZSTD_compressStream2 delegating to ZSTDMT_compressStream_generic");
         /* Either we don't require maximum forward progress, we've finished the
          * flush, or we are out of output space.
          */
-        assert(!forceMaxProgress || flushMin == 0 || output->pos == output->size);
+        assert(endOp == ZSTD_e_continue || flushMin == 0 || output->pos == output->size);
+        ZSTD_setBufferExpectations(cctx, output, input);
         return flushMin;
     }
 #endif
     FORWARD_IF_ERROR( ZSTD_compressStream_generic(cctx, output, input, endOp) , "");
     DEBUGLOG(5, "completed ZSTD_compressStream2");
+    ZSTD_setBufferExpectations(cctx, output, input);
     return cctx->outBuffContentSize - cctx->outBuffFlushedSize; /* remaining to flush */
 }
 
@@ -4065,14 +4506,22 @@ size_t ZSTD_compress2(ZSTD_CCtx* cctx,
                       void* dst, size_t dstCapacity,
                       const void* src, size_t srcSize)
 {
+    ZSTD_bufferMode_e const originalInBufferMode = cctx->requestedParams.inBufferMode;
+    ZSTD_bufferMode_e const originalOutBufferMode = cctx->requestedParams.outBufferMode;
     DEBUGLOG(4, "ZSTD_compress2 (srcSize=%u)", (unsigned)srcSize);
     ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
+    /* Enable stable input/output buffers. */
+    cctx->requestedParams.inBufferMode = ZSTD_bm_stable;
+    cctx->requestedParams.outBufferMode = ZSTD_bm_stable;
     {   size_t oPos = 0;
         size_t iPos = 0;
         size_t const result = ZSTD_compressStream2_simpleArgs(cctx,
                                         dst, dstCapacity, &oPos,
                                         src, srcSize, &iPos,
                                         ZSTD_e_end);
+        /* Reset to the original values. */
+        cctx->requestedParams.inBufferMode = originalInBufferMode;
+        cctx->requestedParams.outBufferMode = originalOutBufferMode;
         FORWARD_IF_ERROR(result, "ZSTD_compressStream2_simpleArgs failed");
         if (result != 0) {  /* compression not completed, due to lack of output space */
             assert(oPos == dstCapacity);
@@ -4083,6 +4532,409 @@ size_t ZSTD_compress2(ZSTD_CCtx* cctx,
     }
 }
 
+typedef struct {
+    U32 idx;             /* Index in array of ZSTD_Sequence */
+    U32 posInSequence;   /* Position within sequence at idx */
+    size_t posInSrc;        /* Number of bytes given by sequences provided so far */
+} ZSTD_sequencePosition;
+
+/* Returns a ZSTD error code if sequence is not valid */
+static size_t ZSTD_validateSequence(U32 offCode, U32 matchLength,
+                                    size_t posInSrc, U32 windowLog, size_t dictSize, U32 minMatch) {
+    size_t offsetBound;
+    U32 windowSize = 1 << windowLog;
+    /* posInSrc represents the amount of data the the decoder would decode up to this point.
+     * As long as the amount of data decoded is less than or equal to window size, offsets may be
+     * larger than the total length of output decoded in order to reference the dict, even larger than
+     * window size. After output surpasses windowSize, we're limited to windowSize offsets again.
+     */
+    offsetBound = posInSrc > windowSize ? (size_t)windowSize : posInSrc + (size_t)dictSize;
+    RETURN_ERROR_IF(offCode > offsetBound + ZSTD_REP_MOVE, corruption_detected, "Offset too large!");
+    RETURN_ERROR_IF(matchLength < minMatch, corruption_detected, "Matchlength too small");
+    return 0;
+}
+
+/* Returns an offset code, given a sequence's raw offset, the ongoing repcode array, and whether litLength == 0 */
+static U32 ZSTD_finalizeOffCode(U32 rawOffset, const U32 rep[ZSTD_REP_NUM], U32 ll0) {
+    U32 offCode = rawOffset + ZSTD_REP_MOVE;
+    U32 repCode = 0;
+
+    if (!ll0 && rawOffset == rep[0]) {
+        repCode = 1;
+    } else if (rawOffset == rep[1]) {
+        repCode = 2 - ll0;
+    } else if (rawOffset == rep[2]) {
+        repCode = 3 - ll0;
+    } else if (ll0 && rawOffset == rep[0] - 1) {
+        repCode = 3;
+    }
+    if (repCode) {
+        /* ZSTD_storeSeq expects a number in the range [0, 2] to represent a repcode */
+        offCode = repCode - 1;
+    }
+    return offCode;
+}
+
+/* Returns 0 on success, and a ZSTD_error otherwise. This function scans through an array of
+ * ZSTD_Sequence, storing the sequences it finds, until it reaches a block delimiter.
+ */
+static size_t ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
+                                                             const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                                             const void* src, size_t blockSize) {
+    U32 idx = seqPos->idx;
+    BYTE const* ip = (BYTE const*)(src);
+    const BYTE* const iend = ip + blockSize;
+    repcodes_t updatedRepcodes;
+    U32 dictSize;
+    U32 litLength;
+    U32 matchLength;
+    U32 ll0;
+    U32 offCode;
+
+    if (cctx->cdict) {
+        dictSize = (U32)cctx->cdict->dictContentSize;
+    } else if (cctx->prefixDict.dict) {
+        dictSize = (U32)cctx->prefixDict.dictSize;
+    } else {
+        dictSize = 0;
+    }
+    ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(repcodes_t));
+    for (; (inSeqs[idx].matchLength != 0 || inSeqs[idx].offset != 0) && idx < inSeqsSize; ++idx) {
+        litLength = inSeqs[idx].litLength;
+        matchLength = inSeqs[idx].matchLength;
+        ll0 = litLength == 0;
+        offCode = ZSTD_finalizeOffCode(inSeqs[idx].offset, updatedRepcodes.rep, ll0);
+        updatedRepcodes = ZSTD_updateRep(updatedRepcodes.rep, offCode, ll0);
+
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offCode, matchLength, litLength);
+        if (cctx->appliedParams.validateSequences) {
+            seqPos->posInSrc += litLength + matchLength;
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offCode, matchLength, seqPos->posInSrc,
+                                                cctx->appliedParams.cParams.windowLog, dictSize,
+                                                cctx->appliedParams.cParams.minMatch),
+                                                "Sequence validation failed");
+        }
+        RETURN_ERROR_IF(idx - seqPos->idx > cctx->seqStore.maxNbSeq, memory_allocation,
+                        "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offCode, matchLength - MINMATCH);
+        ip += matchLength + litLength;
+    }
+    ZSTD_memcpy(cctx->blockState.nextCBlock->rep, updatedRepcodes.rep, sizeof(repcodes_t));
+
+    if (inSeqs[idx].litLength) {
+        DEBUGLOG(6, "Storing last literals of size: %u", inSeqs[idx].litLength);
+        ZSTD_storeLastLiterals(&cctx->seqStore, ip, inSeqs[idx].litLength);
+        ip += inSeqs[idx].litLength;
+        seqPos->posInSrc += inSeqs[idx].litLength;
+    }
+    RETURN_ERROR_IF(ip != iend, corruption_detected, "Blocksize doesn't agree with block delimiter!");
+    seqPos->idx = idx+1;
+    return 0;
+}
+
+/* Returns the number of bytes to move the current read position back by. Only non-zero
+ * if we ended up splitting a sequence. Otherwise, it may return a ZSTD error if something
+ * went wrong.
+ *
+ * This function will attempt to scan through blockSize bytes represented by the sequences
+ * in inSeqs, storing any (partial) sequences.
+ *
+ * Occasionally, we may want to change the actual number of bytes we consumed from inSeqs to
+ * avoid splitting a match, or to avoid splitting a match such that it would produce a match
+ * smaller than MINMATCH. In this case, we return the number of bytes that we didn't read from this block.
+ */
+static size_t ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
+                                                       const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                                       const void* src, size_t blockSize) {
+    U32 idx = seqPos->idx;
+    U32 startPosInSequence = seqPos->posInSequence;
+    U32 endPosInSequence = seqPos->posInSequence + (U32)blockSize;
+    size_t dictSize;
+    BYTE const* ip = (BYTE const*)(src);
+    BYTE const* iend = ip + blockSize;  /* May be adjusted if we decide to process fewer than blockSize bytes */
+    repcodes_t updatedRepcodes;
+    U32 bytesAdjustment = 0;
+    U32 finalMatchSplit = 0;
+    U32 litLength;
+    U32 matchLength;
+    U32 rawOffset;
+    U32 offCode;
+
+    if (cctx->cdict) {
+        dictSize = cctx->cdict->dictContentSize;
+    } else if (cctx->prefixDict.dict) {
+        dictSize = cctx->prefixDict.dictSize;
+    } else {
+        dictSize = 0;
+    }
+    DEBUGLOG(5, "ZSTD_copySequencesToSeqStore: idx: %u PIS: %u blockSize: %zu", idx, startPosInSequence, blockSize);
+    DEBUGLOG(5, "Start seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
+    ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(repcodes_t));
+    while (endPosInSequence && idx < inSeqsSize && !finalMatchSplit) {
+        const ZSTD_Sequence currSeq = inSeqs[idx];
+        litLength = currSeq.litLength;
+        matchLength = currSeq.matchLength;
+        rawOffset = currSeq.offset;
+
+        /* Modify the sequence depending on where endPosInSequence lies */
+        if (endPosInSequence >= currSeq.litLength + currSeq.matchLength) {
+            if (startPosInSequence >= litLength) {
+                startPosInSequence -= litLength;
+                litLength = 0;
+                matchLength -= startPosInSequence;
+            } else {
+                litLength -= startPosInSequence;
+            }
+            /* Move to the next sequence */
+            endPosInSequence -= currSeq.litLength + currSeq.matchLength;
+            startPosInSequence = 0;
+            idx++;
+        } else {
+            /* This is the final (partial) sequence we're adding from inSeqs, and endPosInSequence
+               does not reach the end of the match. So, we have to split the sequence */
+            DEBUGLOG(6, "Require a split: diff: %u, idx: %u PIS: %u",
+                     currSeq.litLength + currSeq.matchLength - endPosInSequence, idx, endPosInSequence);
+            if (endPosInSequence > litLength) {
+                U32 firstHalfMatchLength;
+                litLength = startPosInSequence >= litLength ? 0 : litLength - startPosInSequence;
+                firstHalfMatchLength = endPosInSequence - startPosInSequence - litLength;
+                if (matchLength > blockSize && firstHalfMatchLength >= cctx->appliedParams.cParams.minMatch) {
+                    /* Only ever split the match if it is larger than the block size */
+                    U32 secondHalfMatchLength = currSeq.matchLength + currSeq.litLength - endPosInSequence;
+                    if (secondHalfMatchLength < cctx->appliedParams.cParams.minMatch) {
+                        /* Move the endPosInSequence backward so that it creates match of minMatch length */
+                        endPosInSequence -= cctx->appliedParams.cParams.minMatch - secondHalfMatchLength;
+                        bytesAdjustment = cctx->appliedParams.cParams.minMatch - secondHalfMatchLength;
+                        firstHalfMatchLength -= bytesAdjustment;
+                    }
+                    matchLength = firstHalfMatchLength;
+                    /* Flag that we split the last match - after storing the sequence, exit the loop,
+                       but keep the value of endPosInSequence */
+                    finalMatchSplit = 1;
+                } else {
+                    /* Move the position in sequence backwards so that we don't split match, and break to store
+                     * the last literals. We use the original currSeq.litLength as a marker for where endPosInSequence
+                     * should go. We prefer to do this whenever it is not necessary to split the match, or if doing so
+                     * would cause the first half of the match to be too small
+                     */
+                    bytesAdjustment = endPosInSequence - currSeq.litLength;
+                    endPosInSequence = currSeq.litLength;
+                    break;
+                }
+            } else {
+                /* This sequence ends inside the literals, break to store the last literals */
+                break;
+            }
+        }
+        /* Check if this offset can be represented with a repcode */
+        {   U32 ll0 = (litLength == 0);
+            offCode = ZSTD_finalizeOffCode(rawOffset, updatedRepcodes.rep, ll0);
+            updatedRepcodes = ZSTD_updateRep(updatedRepcodes.rep, offCode, ll0);
+        }
+
+        if (cctx->appliedParams.validateSequences) {
+            seqPos->posInSrc += litLength + matchLength;
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offCode, matchLength, seqPos->posInSrc,
+                                                   cctx->appliedParams.cParams.windowLog, dictSize,
+                                                   cctx->appliedParams.cParams.minMatch),
+                                                   "Sequence validation failed");
+        }
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offCode, matchLength, litLength);
+        RETURN_ERROR_IF(idx - seqPos->idx > cctx->seqStore.maxNbSeq, memory_allocation,
+                        "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offCode, matchLength - MINMATCH);
+        ip += matchLength + litLength;
+    }
+    DEBUGLOG(5, "Ending seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
+    assert(idx == inSeqsSize || endPosInSequence <= inSeqs[idx].litLength + inSeqs[idx].matchLength);
+    seqPos->idx = idx;
+    seqPos->posInSequence = endPosInSequence;
+    ZSTD_memcpy(cctx->blockState.nextCBlock->rep, updatedRepcodes.rep, sizeof(repcodes_t));
+
+    iend -= bytesAdjustment;
+    if (ip != iend) {
+        /* Store any last literals */
+        U32 lastLLSize = (U32)(iend - ip);
+        assert(ip <= iend);
+        DEBUGLOG(6, "Storing last literals of size: %u", lastLLSize);
+        ZSTD_storeLastLiterals(&cctx->seqStore, ip, lastLLSize);
+        seqPos->posInSrc += lastLLSize;
+    }
+
+    return bytesAdjustment;
+}
+
+typedef size_t (*ZSTD_sequenceCopier) (ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
+                                       const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                       const void* src, size_t blockSize);
+static ZSTD_sequenceCopier ZSTD_selectSequenceCopier(ZSTD_sequenceFormat_e mode) {
+    ZSTD_sequenceCopier sequenceCopier = NULL;
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_blockDelimiters, mode));
+    if (mode == ZSTD_sf_explicitBlockDelimiters) {
+        return ZSTD_copySequencesToSeqStoreExplicitBlockDelim;
+    } else if (mode == ZSTD_sf_noBlockDelimiters) {
+        return ZSTD_copySequencesToSeqStoreNoBlockDelim;
+    }
+    assert(sequenceCopier != NULL);
+    return sequenceCopier;
+}
+
+/* Compress, block-by-block, all of the sequences given.
+ *
+ * Returns the cumulative size of all compressed blocks (including their headers), otherwise a ZSTD error.
+ */
+static size_t ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
+                                              void* dst, size_t dstCapacity,
+                                              const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                                              const void* src, size_t srcSize) {
+    size_t cSize = 0;
+    U32 lastBlock;
+    size_t blockSize;
+    size_t compressedSeqsSize;
+    size_t remaining = srcSize;
+    ZSTD_sequencePosition seqPos = {0, 0, 0};
+
+    BYTE const* ip = (BYTE const*)src;
+    BYTE* op = (BYTE*)dst;
+    ZSTD_sequenceCopier sequenceCopier = ZSTD_selectSequenceCopier(cctx->appliedParams.blockDelimiters);
+
+    DEBUGLOG(4, "ZSTD_compressSequences_internal srcSize: %zu, inSeqsSize: %zu", srcSize, inSeqsSize);
+    /* Special case: empty frame */
+    if (remaining == 0) {
+        U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "No room for empty frame block header");
+        MEM_writeLE32(op, cBlockHeader24);
+        op += ZSTD_blockHeaderSize;
+        dstCapacity -= ZSTD_blockHeaderSize;
+        cSize += ZSTD_blockHeaderSize;
+    }
+
+    while (remaining) {
+        size_t cBlockSize;
+        size_t additionalByteAdjustment;
+        lastBlock = remaining <= cctx->blockSize;
+        blockSize = lastBlock ? (U32)remaining : (U32)cctx->blockSize;
+        ZSTD_resetSeqStore(&cctx->seqStore);
+        DEBUGLOG(4, "Working on new block. Blocksize: %zu", blockSize);
+
+        additionalByteAdjustment = sequenceCopier(cctx, &seqPos, inSeqs, inSeqsSize, ip, blockSize);
+        FORWARD_IF_ERROR(additionalByteAdjustment, "Bad sequence copy");
+        blockSize -= additionalByteAdjustment;
+
+        /* If blocks are too small, emit as a nocompress block */
+        if (blockSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
+            cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+            FORWARD_IF_ERROR(cBlockSize, "Nocompress block failed");
+            DEBUGLOG(4, "Block too small, writing out nocompress block: cSize: %zu", cBlockSize);
+            cSize += cBlockSize;
+            ip += blockSize;
+            op += cBlockSize;
+            remaining -= blockSize;
+            dstCapacity -= cBlockSize;
+            continue;
+        }
+
+        compressedSeqsSize = ZSTD_entropyCompressSequences(&cctx->seqStore,
+                                &cctx->blockState.prevCBlock->entropy, &cctx->blockState.nextCBlock->entropy,
+                                &cctx->appliedParams,
+                                op + ZSTD_blockHeaderSize /* Leave space for block header */, dstCapacity - ZSTD_blockHeaderSize,
+                                blockSize,
+                                cctx->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
+                                cctx->bmi2);
+        FORWARD_IF_ERROR(compressedSeqsSize, "Compressing sequences of block failed");
+        DEBUGLOG(4, "Compressed sequences size: %zu", compressedSeqsSize);
+
+        if (!cctx->isFirstBlock &&
+            ZSTD_maybeRLE(&cctx->seqStore) &&
+            ZSTD_isRLE((BYTE const*)src, srcSize)) {
+            /* We don't want to emit our first block as a RLE even if it qualifies because
+            * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+            * This is only an issue for zstd <= v1.4.3
+            */
+            compressedSeqsSize = 1;
+        }
+
+        if (compressedSeqsSize == 0) {
+            /* ZSTD_noCompressBlock writes the block header as well */
+            cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+            FORWARD_IF_ERROR(cBlockSize, "Nocompress block failed");
+            DEBUGLOG(4, "Writing out nocompress block, size: %zu", cBlockSize);
+        } else if (compressedSeqsSize == 1) {
+            cBlockSize = ZSTD_rleCompressBlock(op, dstCapacity, *ip, blockSize, lastBlock);
+            FORWARD_IF_ERROR(cBlockSize, "RLE compress block failed");
+            DEBUGLOG(4, "Writing out RLE block, size: %zu", cBlockSize);
+        } else {
+            U32 cBlockHeader;
+            /* Error checking and repcodes update */
+            ZSTD_confirmRepcodesAndEntropyTables(cctx);
+            if (cctx->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+                cctx->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+            /* Write block header into beginning of block*/
+            cBlockHeader = lastBlock + (((U32)bt_compressed)<<1) + (U32)(compressedSeqsSize << 3);
+            MEM_writeLE24(op, cBlockHeader);
+            cBlockSize = ZSTD_blockHeaderSize + compressedSeqsSize;
+            DEBUGLOG(4, "Writing out compressed block, size: %zu", cBlockSize);
+        }
+
+        cSize += cBlockSize;
+        DEBUGLOG(4, "cSize running total: %zu", cSize);
+
+        if (lastBlock) {
+            break;
+        } else {
+            ip += blockSize;
+            op += cBlockSize;
+            remaining -= blockSize;
+            dstCapacity -= cBlockSize;
+            cctx->isFirstBlock = 0;
+        }
+    }
+
+    return cSize;
+}
+
+size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstCapacity,
+                              const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                              const void* src, size_t srcSize) {
+    BYTE* op = (BYTE*)dst;
+    size_t cSize = 0;
+    size_t compressedBlocksSize = 0;
+    size_t frameHeaderSize = 0;
+
+    /* Transparent initialization stage, same as compressStream2() */
+    DEBUGLOG(3, "ZSTD_compressSequences()");
+    assert(cctx != NULL);
+    FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, ZSTD_e_end, srcSize), "CCtx initialization failed");
+    /* Begin writing output, starting with frame header */
+    frameHeaderSize = ZSTD_writeFrameHeader(op, dstCapacity, &cctx->appliedParams, srcSize, cctx->dictID);
+    op += frameHeaderSize;
+    dstCapacity -= frameHeaderSize;
+    cSize += frameHeaderSize;
+    if (cctx->appliedParams.fParams.checksumFlag && srcSize) {
+        XXH64_update(&cctx->xxhState, src, srcSize);
+    }
+    /* cSize includes block header size and compressed sequences size */
+    compressedBlocksSize = ZSTD_compressSequences_internal(cctx,
+                                                           op, dstCapacity,
+                                                           inSeqs, inSeqsSize,
+                                                           src, srcSize);
+    FORWARD_IF_ERROR(compressedBlocksSize, "Compressing blocks failed!");
+    cSize += compressedBlocksSize;
+    dstCapacity -= compressedBlocksSize;
+
+    if (cctx->appliedParams.fParams.checksumFlag) {
+        U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "no room for checksum");
+        DEBUGLOG(4, "Write checksum : %08X", (unsigned)checksum);
+        MEM_writeLE32((char*)dst + cSize, checksum);
+        cSize += 4;
+    }
+
+    DEBUGLOG(3, "Final compressed size: %zu", cSize);
+    return cSize;
+}
+
 /*======   Finalize   ======*/
 
 /*! ZSTD_flushStream() :
@@ -4223,25 +5075,103 @@ static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEV
 },
 };
 
+static ZSTD_compressionParameters ZSTD_dedicatedDictSearch_getCParams(int const compressionLevel, size_t const dictSize)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, 0, dictSize, ZSTD_cpm_createCDict);
+    switch (cParams.strategy) {
+        case ZSTD_fast:
+        case ZSTD_dfast:
+            break;
+        case ZSTD_greedy:
+        case ZSTD_lazy:
+        case ZSTD_lazy2:
+            cParams.hashLog += ZSTD_LAZY_DDSS_BUCKET_LOG;
+            break;
+        case ZSTD_btlazy2:
+        case ZSTD_btopt:
+        case ZSTD_btultra:
+        case ZSTD_btultra2:
+            break;
+    }
+    return cParams;
+}
+
+static int ZSTD_dedicatedDictSearch_isSupported(
+        ZSTD_compressionParameters const* cParams)
+{
+    return (cParams->strategy >= ZSTD_greedy) && (cParams->strategy <= ZSTD_lazy2);
+}
+
+/**
+ * Reverses the adjustment applied to cparams when enabling dedicated dict
+ * search. This is used to recover the params set to be used in the working
+ * context. (Otherwise, those tables would also grow.)
+ */
+static void ZSTD_dedicatedDictSearch_revertCParams(
+        ZSTD_compressionParameters* cParams) {
+    switch (cParams->strategy) {
+        case ZSTD_fast:
+        case ZSTD_dfast:
+            break;
+        case ZSTD_greedy:
+        case ZSTD_lazy:
+        case ZSTD_lazy2:
+            cParams->hashLog -= ZSTD_LAZY_DDSS_BUCKET_LOG;
+            break;
+        case ZSTD_btlazy2:
+        case ZSTD_btopt:
+        case ZSTD_btultra:
+        case ZSTD_btultra2:
+            break;
+    }
+}
+
+static U64 ZSTD_getCParamRowSize(U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode)
+{
+    switch (mode) {
+    case ZSTD_cpm_unknown:
+    case ZSTD_cpm_noAttachDict:
+    case ZSTD_cpm_createCDict:
+        break;
+    case ZSTD_cpm_attachDict:
+        dictSize = 0;
+        break;
+    default:
+        assert(0);
+        break;
+    }
+    {   int const unknown = srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN;
+        size_t const addedSize = unknown && dictSize > 0 ? 500 : 0;
+        return unknown && dictSize == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : srcSizeHint+dictSize+addedSize;
+    }
+}
+
 /*! ZSTD_getCParams_internal() :
  * @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
  *  Note: srcSizeHint 0 means 0, use ZSTD_CONTENTSIZE_UNKNOWN for unknown.
- *        Use dictSize == 0 for unknown or unused. */
-static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
+ *        Use dictSize == 0 for unknown or unused.
+ *  Note: `mode` controls how we treat the `dictSize`. See docs for `ZSTD_cParamMode_e`. */
+static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode)
 {
-    int const unknown = srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN;
-    size_t const addedSize = unknown && dictSize > 0 ? 500 : 0;
-    U64 const rSize = unknown && dictSize == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : srcSizeHint+dictSize+addedSize;
+    U64 const rSize = ZSTD_getCParamRowSize(srcSizeHint, dictSize, mode);
     U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB);
-    int row = compressionLevel;
+    int row;
     DEBUGLOG(5, "ZSTD_getCParams_internal (cLevel=%i)", compressionLevel);
+
+    /* row */
     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;
+    else if (compressionLevel < 0) row = 0;   /* entry 0 is baseline for fast mode */
+    else if (compressionLevel > ZSTD_MAX_CLEVEL) row = ZSTD_MAX_CLEVEL;
+    else row = compressionLevel;
+
     {   ZSTD_compressionParameters cp = ZSTD_defaultCParameters[tableID][row];
-        if (compressionLevel < 0) cp.targetLength = (unsigned)(-compressionLevel);   /* acceleration factor */
+        /* acceleration factor */
+        if (compressionLevel < 0) {
+            int const clampedCompressionLevel = MAX(ZSTD_minCLevel(), compressionLevel);
+            cp.targetLength = (unsigned)(-clampedCompressionLevel);
+        }
         /* refine parameters based on srcSize & dictSize */
-        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize);
+        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize, mode);
     }
 }
 
@@ -4251,18 +5181,18 @@ static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel,
 ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
 {
     if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
-    return ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize);
+    return ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize, ZSTD_cpm_unknown);
 }
 
 /*! ZSTD_getParams() :
  *  same idea as ZSTD_getCParams()
  * @return a `ZSTD_parameters` structure (instead of `ZSTD_compressionParameters`).
  *  Fields of `ZSTD_frameParameters` are set to default values */
-static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize) {
+static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode) {
     ZSTD_parameters params;
-    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize);
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize, mode);
     DEBUGLOG(5, "ZSTD_getParams (cLevel=%i)", compressionLevel);
-    memset(&params, 0, sizeof(params));
+    ZSTD_memset(&params, 0, sizeof(params));
     params.cParams = cParams;
     params.fParams.contentSizeFlag = 1;
     return params;
@@ -4274,5 +5204,5 @@ static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned lo
  *  Fields of `ZSTD_frameParameters` are set to default values */
 ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize) {
     if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
-    return ZSTD_getParams_internal(compressionLevel, srcSizeHint, dictSize);
+    return ZSTD_getParams_internal(compressionLevel, srcSizeHint, dictSize, ZSTD_cpm_unknown);
 }
diff --git a/lib/compress/zstd_compress_internal.h b/lib/compress/zstd_compress_internal.h
index db73f6ce21f2..c04998b8b18f 100644
--- a/lib/compress/zstd_compress_internal.h
+++ b/lib/compress/zstd_compress_internal.h
@@ -28,7 +28,6 @@
 extern "C" {
 #endif
 
-
 /*-*************************************
 *  Constants
 ***************************************/
@@ -64,7 +63,7 @@ typedef struct {
 } ZSTD_localDict;
 
 typedef struct {
-    U32 CTable[HUF_CTABLE_SIZE_U32(255)];
+    HUF_CElt CTable[HUF_CTABLE_SIZE_U32(255)];
     HUF_repeat repeatMode;
 } ZSTD_hufCTables_t;
 
@@ -83,11 +82,28 @@ typedef struct {
 } ZSTD_entropyCTables_t;
 
 typedef struct {
-    U32 off;
-    U32 len;
+    U32 off;            /* Offset code (offset + ZSTD_REP_MOVE) for the match */
+    U32 len;            /* Raw length of match */
 } ZSTD_match_t;
 
 typedef struct {
+    U32 offset;         /* Offset of sequence */
+    U32 litLength;      /* Length of literals prior to match */
+    U32 matchLength;    /* Raw length of match */
+} rawSeq;
+
+typedef struct {
+  rawSeq* seq;          /* The start of the sequences */
+  size_t pos;           /* The index in seq where reading stopped. pos <= size. */
+  size_t posInSequence; /* The position within the sequence at seq[pos] where reading
+                           stopped. posInSequence <= seq[pos].litLength + seq[pos].matchLength */
+  size_t size;          /* The number of sequences. <= capacity. */
+  size_t capacity;      /* The capacity starting from `seq` pointer */
+} rawSeqStore_t;
+
+UNUSED_ATTR static const rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0, 0};
+
+typedef struct {
     int price;
     U32 off;
     U32 mlen;
@@ -147,9 +163,13 @@ struct ZSTD_matchState_t {
     U32* hashTable;
     U32* hashTable3;
     U32* chainTable;
+    int dedicatedDictSearch;  /* Indicates whether this matchState is using the
+                               * dedicated dictionary search structure.
+                               */
     optState_t opt;         /* optimal parser state */
     const ZSTD_matchState_t* dictMatchState;
     ZSTD_compressionParameters cParams;
+    const rawSeqStore_t* ldmSeqStore;
 };
 
 typedef struct {
@@ -182,19 +202,6 @@ typedef struct {
 } 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 starting from `seq` pointer */
-} rawSeqStore_t;
-
-typedef struct {
     int collectSequences;
     ZSTD_Sequence* seqStart;
     size_t seqIndex;
@@ -228,10 +235,34 @@ struct ZSTD_CCtx_params_s {
     /* Long distance matching parameters */
     ldmParams_t ldmParams;
 
+    /* Dedicated dict search algorithm trigger */
+    int enableDedicatedDictSearch;
+
+    /* Input/output buffer modes */
+    ZSTD_bufferMode_e inBufferMode;
+    ZSTD_bufferMode_e outBufferMode;
+
+    /* Sequence compression API */
+    ZSTD_sequenceFormat_e blockDelimiters;
+    int validateSequences;
+
     /* Internal use, for createCCtxParams() and freeCCtxParams() only */
     ZSTD_customMem customMem;
 };  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
 
+#define COMPRESS_SEQUENCES_WORKSPACE_SIZE (sizeof(unsigned) * (MaxSeq + 2))
+#define ENTROPY_WORKSPACE_SIZE (HUF_WORKSPACE_SIZE + COMPRESS_SEQUENCES_WORKSPACE_SIZE)
+
+/**
+ * Indicates whether this compression proceeds directly from user-provided
+ * source buffer to user-provided destination buffer (ZSTDb_not_buffered), or
+ * whether the context needs to buffer the input/output (ZSTDb_buffered).
+ */
+typedef enum {
+    ZSTDb_not_buffered,
+    ZSTDb_buffered
+} ZSTD_buffered_policy_e;
+
 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. */
@@ -247,6 +278,7 @@ struct ZSTD_CCtx_s {
     unsigned long long producedCSize;
     XXH64_state_t xxhState;
     ZSTD_customMem customMem;
+    ZSTD_threadPool* pool;
     size_t staticSize;
     SeqCollector seqCollector;
     int isFirstBlock;
@@ -258,7 +290,10 @@ struct ZSTD_CCtx_s {
     size_t maxNbLdmSequences;
     rawSeqStore_t externSeqStore; /* Mutable reference to external sequences */
     ZSTD_blockState_t blockState;
-    U32* entropyWorkspace;  /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
+    U32* entropyWorkspace;  /* entropy workspace of ENTROPY_WORKSPACE_SIZE bytes */
+
+    /* Wether we are streaming or not */
+    ZSTD_buffered_policy_e bufferedPolicy;
 
     /* streaming */
     char*  inBuff;
@@ -273,6 +308,10 @@ struct ZSTD_CCtx_s {
     ZSTD_cStreamStage streamStage;
     U32    frameEnded;
 
+    /* Stable in/out buffer verification */
+    ZSTD_inBuffer expectedInBuffer;
+    size_t expectedOutBufferSize;
+
     /* Dictionary */
     ZSTD_localDict localDict;
     const ZSTD_CDict* cdict;
@@ -286,8 +325,32 @@ struct ZSTD_CCtx_s {
 
 typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
 
-typedef enum { ZSTD_noDict = 0, ZSTD_extDict = 1, ZSTD_dictMatchState = 2 } ZSTD_dictMode_e;
-
+typedef enum {
+    ZSTD_noDict = 0,
+    ZSTD_extDict = 1,
+    ZSTD_dictMatchState = 2,
+    ZSTD_dedicatedDictSearch = 3
+} ZSTD_dictMode_e;
+
+typedef enum {
+    ZSTD_cpm_noAttachDict = 0,  /* Compression with ZSTD_noDict or ZSTD_extDict.
+                                 * In this mode we use both the srcSize and the dictSize
+                                 * when selecting and adjusting parameters.
+                                 */
+    ZSTD_cpm_attachDict = 1,    /* Compression with ZSTD_dictMatchState or ZSTD_dedicatedDictSearch.
+                                 * In this mode we only take the srcSize into account when selecting
+                                 * and adjusting parameters.
+                                 */
+    ZSTD_cpm_createCDict = 2,   /* Creating a CDict.
+                                 * In this mode we take both the source size and the dictionary size
+                                 * into account when selecting and adjusting the parameters.
+                                 */
+    ZSTD_cpm_unknown = 3,       /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
+                                 * We don't know what these parameters are for. We default to the legacy
+                                 * behavior of taking both the source size and the dict size into account
+                                 * when selecting and adjusting parameters.
+                                 */
+} ZSTD_cParamMode_e;
 
 typedef size_t (*ZSTD_blockCompressor) (
         ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -345,7 +408,7 @@ MEM_STATIC repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 con
             newReps.rep[1] = rep[0];
             newReps.rep[0] = currentOffset;
         } else {   /* repCode == 0 */
-            memcpy(&newReps, rep, sizeof(newReps));
+            ZSTD_memcpy(&newReps, rep, sizeof(newReps));
         }
     }
     return newReps;
@@ -372,7 +435,7 @@ MEM_STATIC size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const voi
     RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
                     dstSize_tooSmall, "dst buf too small for uncompressed block");
     MEM_writeLE24(dst, cBlockHeader24);
-    memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
+    ZSTD_memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
     return ZSTD_blockHeaderSize + srcSize;
 }
 
@@ -498,8 +561,12 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
     if (MEM_isLittleEndian()) {
         if (MEM_64bits()) {
 #       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            return _BitScanForward64( &r, (U64)val ) ? (unsigned)(r >> 3) : 0;
+#           if STATIC_BMI2
+                return _tzcnt_u64(val) >> 3;
+#           else
+                unsigned long r = 0;
+                return _BitScanForward64( &r, (U64)val ) ? (unsigned)(r >> 3) : 0;
+#           endif
 #       elif defined(__GNUC__) && (__GNUC__ >= 4)
             return (__builtin_ctzll((U64)val) >> 3);
 #       else
@@ -530,8 +597,12 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
     } else {  /* Big Endian CPU */
         if (MEM_64bits()) {
 #       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            return _BitScanReverse64( &r, val ) ? (unsigned)(r >> 3) : 0;
+#           if STATIC_BMI2
+			    return _lzcnt_u64(val) >> 3;
+#           else
+			    unsigned long r = 0;
+			    return _BitScanReverse64(&r, (U64)val) ? (unsigned)(r >> 3) : 0;
+#           endif
 #       elif defined(__GNUC__) && (__GNUC__ >= 4)
             return (__builtin_clzll(val) >> 3);
 #       else
@@ -626,7 +697,8 @@ 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)
+MEM_STATIC FORCE_INLINE_ATTR
+size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
 {
     switch(mls)
     {
@@ -742,7 +814,7 @@ MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
     return ZSTD_window_hasExtDict(ms->window) ?
         ZSTD_extDict :
         ms->dictMatchState != NULL ?
-            ZSTD_dictMatchState :
+            (ms->dictMatchState->dedicatedDictSearch ? ZSTD_dedicatedDictSearch : ZSTD_dictMatchState) :
             ZSTD_noDict;
 }
 
@@ -754,8 +826,8 @@ MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
 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;
+    U32 const curr = (U32)((BYTE const*)srcEnd - window.base);
+    return curr > ZSTD_CURRENT_MAX;
 }
 
 /**
@@ -791,14 +863,14 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
      *    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 currentCycle0 = current & cycleMask;
+    U32 const curr = (U32)((BYTE const*)src - window->base);
+    U32 const currentCycle0 = curr & cycleMask;
     /* Exclude zero so that newCurrent - maxDist >= 1. */
     U32 const currentCycle1 = currentCycle0 == 0 ? (1U << cycleLog) : currentCycle0;
     U32 const newCurrent = currentCycle1 + maxDist;
-    U32 const correction = current - newCurrent;
+    U32 const correction = curr - newCurrent;
     assert((maxDist & cycleMask) == 0);
-    assert(current > newCurrent);
+    assert(curr > newCurrent);
     /* Loose bound, should be around 1<<29 (see above) */
     assert(correction > 1<<28);
 
@@ -919,7 +991,7 @@ ZSTD_checkDictValidity(const ZSTD_window_t* window,
 }
 
 MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
-    memset(window, 0, sizeof(*window));
+    ZSTD_memset(window, 0, sizeof(*window));
     window->base = (BYTE const*)"";
     window->dictBase = (BYTE const*)"";
     window->dictLimit = 1;    /* start from 1, so that 1st position is valid */
@@ -973,12 +1045,16 @@ MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
 /**
  * Returns the lowest allowed match index. It may either be in the ext-dict or the prefix.
  */
-MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 current, unsigned windowLog)
+MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
 {
     U32    const maxDistance = 1U << windowLog;
     U32    const lowestValid = ms->window.lowLimit;
-    U32    const withinWindow = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
+    U32    const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
     U32    const isDictionary = (ms->loadedDictEnd != 0);
+    /* When using a dictionary the entire dictionary is valid if a single byte of the dictionary
+     * is within the window. We invalidate the dictionary (and set loadedDictEnd to 0) when it isn't
+     * valid for the entire block. So this check is sufficient to find the lowest valid match index.
+     */
     U32    const matchLowest = isDictionary ? lowestValid : withinWindow;
     return matchLowest;
 }
@@ -986,12 +1062,15 @@ MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 current
 /**
  * Returns the lowest allowed match index in the prefix.
  */
-MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 current, unsigned windowLog)
+MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
 {
     U32    const maxDistance = 1U << windowLog;
     U32    const lowestValid = ms->window.dictLimit;
-    U32    const withinWindow = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
+    U32    const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
     U32    const isDictionary = (ms->loadedDictEnd != 0);
+    /* When computing the lowest prefix index we need to take the dictionary into account to handle
+     * the edge case where the dictionary and the source are contiguous in memory.
+     */
     U32    const matchLowest = isDictionary ? lowestValid : withinWindow;
     return matchLowest;
 }
@@ -1045,7 +1124,6 @@ MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
  * assumptions : magic number supposed already checked
  *               and dictSize >= 8 */
 size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
-                         short* offcodeNCount, unsigned* offcodeMaxValue,
                          const void* const dict, size_t dictSize);
 
 void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
@@ -1061,7 +1139,7 @@ void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
  * Note: srcSizeHint == 0 means 0!
  */
 ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
-        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize);
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
 
 /*! ZSTD_initCStream_internal() :
  *  Private use only. Init streaming operation.
diff --git a/lib/compress/zstd_compress_literals.c b/lib/compress/zstd_compress_literals.c
index 17e7168d8936..6dd1c1447a93 100644
--- a/lib/compress/zstd_compress_literals.c
+++ b/lib/compress/zstd_compress_literals.c
@@ -35,7 +35,7 @@ size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src,
             assert(0);
     }
 
-    memcpy(ostart + flSize, src, srcSize);
+    ZSTD_memcpy(ostart + flSize, src, srcSize);
     DEBUGLOG(5, "Raw literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
     return srcSize + flSize;
 }
@@ -86,7 +86,7 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
                 disableLiteralCompression, (U32)srcSize);
 
     /* Prepare nextEntropy assuming reusing the existing table */
-    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
 
     if (disableLiteralCompression)
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
@@ -118,11 +118,11 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
     }
 
     if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
-        memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+        ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
     }
     if (cLitSize==1) {
-        memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+        ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
         return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
     }
 
diff --git a/lib/compress/zstd_compress_sequences.c b/lib/compress/zstd_compress_sequences.c
index f9f8097c839b..be30c08c6b8b 100644
--- a/lib/compress/zstd_compress_sequences.c
+++ b/lib/compress/zstd_compress_sequences.c
@@ -51,6 +51,19 @@ static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
 }
 
 /**
+ * Returns true if we should use ncount=-1 else we should
+ * use ncount=1 for low probability symbols instead.
+ */
+static unsigned ZSTD_useLowProbCount(size_t const nbSeq)
+{
+    /* Heuristic: This should cover most blocks <= 16K and
+     * start to fade out after 16K to about 32K depending on
+     * comprssibility.
+     */
+    return nbSeq >= 2048;
+}
+
+/**
  * Returns the cost in bytes of encoding the normalized count header.
  * Returns an error if any of the helper functions return an error.
  */
@@ -60,7 +73,7 @@ static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
     BYTE wksp[FSE_NCOUNTBOUND];
     S16 norm[MaxSeq + 1];
     const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
-    FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max), "");
+    FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max, ZSTD_useLowProbCount(nbSeq)), "");
     return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
 }
 
@@ -239,7 +252,7 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
         *op = codeTable[0];
         return 1;
     case set_repeat:
-        memcpy(nextCTable, prevCTable, prevCTableSize);
+        ZSTD_memcpy(nextCTable, prevCTable, prevCTableSize);
         return 0;
     case set_basic:
         FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize), "");  /* note : could be pre-calculated */
@@ -253,7 +266,8 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
             nbSeq_1--;
         }
         assert(nbSeq_1 > 1);
-        FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max), "");
+        assert(entropyWorkspaceSize >= FSE_BUILD_CTABLE_WORKSPACE_SIZE(MaxSeq, MaxFSELog));
+        FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "");
         {   size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog);   /* overflow protected */
             FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
             FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, entropyWorkspace, entropyWorkspaceSize), "");
diff --git a/lib/compress/zstd_compress_superblock.c b/lib/compress/zstd_compress_superblock.c
index b693866c0ac1..e23e619eef14 100644
--- a/lib/compress/zstd_compress_superblock.c
+++ b/lib/compress/zstd_compress_superblock.c
@@ -29,7 +29,7 @@
  *  This metadata is populated in ZSTD_buildSuperBlockEntropy_literal() */
 typedef struct {
     symbolEncodingType_e hType;
-    BYTE hufDesBuffer[500]; /* TODO give name to this value */
+    BYTE hufDesBuffer[ZSTD_MAX_HUF_HEADER_SIZE];
     size_t hufDesSize;
 } ZSTD_hufCTablesMetadata_t;
 
@@ -42,7 +42,7 @@ typedef struct {
     symbolEncodingType_e llType;
     symbolEncodingType_e ofType;
     symbolEncodingType_e mlType;
-    BYTE fseTablesBuffer[500]; /* TODO give name to this value */
+    BYTE fseTablesBuffer[ZSTD_MAX_FSE_HEADERS_SIZE];
     size_t fseTablesSize;
     size_t lastCountSize; /* This is to account for bug in 1.3.4. More detail in ZSTD_compressSubBlock_sequences() */
 } ZSTD_fseCTablesMetadata_t;
@@ -79,7 +79,7 @@ static size_t ZSTD_buildSuperBlockEntropy_literal(void* const src, size_t srcSiz
     DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_literal (srcSize=%zu)", srcSize);
 
     /* Prepare nextEntropy assuming reusing the existing table */
-    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
 
     if (disableLiteralsCompression) {
         DEBUGLOG(5, "set_basic - disabled");
@@ -118,7 +118,7 @@ static size_t ZSTD_buildSuperBlockEntropy_literal(void* const src, size_t srcSiz
     }
 
     /* Build Huffman Tree */
-    memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
+    ZSTD_memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
     huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
     {   size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
                                                     maxSymbolValue, huffLog,
@@ -137,14 +137,14 @@ static size_t ZSTD_buildSuperBlockEntropy_literal(void* const src, size_t srcSiz
                         (HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
                 if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
                     DEBUGLOG(5, "set_repeat - smaller");
-                    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+                    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
                     hufMetadata->hType = set_repeat;
                     return 0;
                 }
             }
             if (newCSize + hSize >= srcSize) {
                 DEBUGLOG(5, "set_basic - no gains");
-                memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+                ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
                 hufMetadata->hType = set_basic;
                 return 0;
             }
@@ -188,7 +188,7 @@ static size_t ZSTD_buildSuperBlockEntropy_sequences(seqStore_t* seqStorePtr,
 
     assert(cTableWkspSize >= (1 << MaxFSELog) * sizeof(FSE_FUNCTION_TYPE));
     DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_sequences (nbSeq=%zu)", nbSeq);
-    memset(workspace, 0, wkspSize);
+    ZSTD_memset(workspace, 0, wkspSize);
 
     fseMetadata->lastCountSize = 0;
     /* convert length/distances into codes */
@@ -348,7 +348,7 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
     assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);
 
     if (writeEntropy && hufMetadata->hType == set_compressed) {
-        memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
+        ZSTD_memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
         op += hufMetadata->hufDesSize;
         cLitSize += hufMetadata->hufDesSize;
         DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
@@ -474,7 +474,7 @@ static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables
         const U32 MLtype = fseMetadata->mlType;
         DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
         *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
-        memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
+        ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
         op += fseMetadata->fseTablesSize;
     } else {
         const U32 repeat = set_repeat;
@@ -603,7 +603,7 @@ static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
                         const BYTE* codeTable, unsigned maxCode,
                         size_t nbSeq, const FSE_CTable* fseCTable,
                         const U32* additionalBits,
-                        short const* defaultNorm, U32 defaultNormLog,
+                        short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
                         void* workspace, size_t wkspSize)
 {
     unsigned* const countWksp = (unsigned*)workspace;
@@ -615,7 +615,11 @@ static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
 
     HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize);  /* can't fail */
     if (type == set_basic) {
-        cSymbolTypeSizeEstimateInBits = ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max);
+        /* We selected this encoding type, so it must be valid. */
+        assert(max <= defaultMax);
+        cSymbolTypeSizeEstimateInBits = max <= defaultMax
+                ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
+                : ERROR(GENERIC);
     } else if (type == set_rle) {
         cSymbolTypeSizeEstimateInBits = 0;
     } else if (type == set_compressed || type == set_repeat) {
@@ -643,15 +647,15 @@ static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
     size_t cSeqSizeEstimate = 0;
     cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
                                          nbSeq, fseTables->offcodeCTable, NULL,
-                                         OF_defaultNorm, OF_defaultNormLog,
+                                         OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
                                          workspace, wkspSize);
     cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
                                          nbSeq, fseTables->litlengthCTable, LL_bits,
-                                         LL_defaultNorm, LL_defaultNormLog,
+                                         LL_defaultNorm, LL_defaultNormLog, MaxLL,
                                          workspace, wkspSize);
     cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
                                          nbSeq, fseTables->matchlengthCTable, ML_bits,
-                                         ML_defaultNorm, ML_defaultNormLog,
+                                         ML_defaultNorm, ML_defaultNormLog, MaxML,
                                          workspace, wkspSize);
     if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
     return cSeqSizeEstimate + sequencesSectionHeaderSize;
@@ -790,7 +794,7 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
     } while (!lastSequence);
     if (writeLitEntropy) {
         DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
-        memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
+        ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
     }
     if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
         /* If we haven't written our entropy tables, then we've violated our contract and
@@ -809,11 +813,11 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
         if (sp < send) {
             seqDef const* seq;
             repcodes_t rep;
-            memcpy(&rep, prevCBlock->rep, sizeof(rep)); 
+            ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
             for (seq = sstart; seq < sp; ++seq) {
                 rep = ZSTD_updateRep(rep.rep, seq->offset - 1, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
             }
-            memcpy(nextCBlock->rep, &rep, sizeof(rep));
+            ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
         }
     }
     DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
@@ -831,7 +835,7 @@ size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
           &zc->blockState.nextCBlock->entropy,
           &zc->appliedParams,
           &entropyMetadata,
-          zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
+          zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
 
     return ZSTD_compressSubBlock_multi(&zc->seqStore,
             zc->blockState.prevCBlock,
@@ -841,5 +845,5 @@ size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
             dst, dstCapacity,
             src, srcSize,
             zc->bmi2, lastBlock,
-            zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
+            zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
 }
diff --git a/lib/compress/zstd_cwksp.h b/lib/compress/zstd_cwksp.h
index a25c9263b7d7..d65170b39cad 100644
--- a/lib/compress/zstd_cwksp.h
+++ b/lib/compress/zstd_cwksp.h
@@ -45,6 +45,16 @@ typedef enum {
 } ZSTD_cwksp_alloc_phase_e;
 
 /**
+ * Used to describe whether the workspace is statically allocated (and will not
+ * necessarily ever be freed), or if it's dynamically allocated and we can
+ * expect a well-formed caller to free this.
+ */
+typedef enum {
+    ZSTD_cwksp_dynamic_alloc,
+    ZSTD_cwksp_static_alloc
+} ZSTD_cwksp_static_alloc_e;
+
+/**
  * Zstd fits all its internal datastructures into a single continuous buffer,
  * so that it only needs to perform a single OS allocation (or so that a buffer
  * can be provided to it and it can perform no allocations at all). This buffer
@@ -92,7 +102,7 @@ typedef enum {
  *
  * - Static objects: this is optionally the enclosing ZSTD_CCtx or ZSTD_CDict,
  *   so that literally everything fits in a single buffer. Note: if present,
- *   this must be the first object in the workspace, since ZSTD_free{CCtx,
+ *   this must be the first object in the workspace, since ZSTD_customFree{CCtx,
  *   CDict}() rely on a pointer comparison to see whether one or two frees are
  *   required.
  *
@@ -137,9 +147,10 @@ typedef struct {
     void* tableValidEnd;
     void* allocStart;
 
-    int allocFailed;
+    BYTE allocFailed;
     int workspaceOversizedDuration;
     ZSTD_cwksp_alloc_phase_e phase;
+    ZSTD_cwksp_static_alloc_e isStatic;
 } ZSTD_cwksp;
 
 /*-*************************************
@@ -178,7 +189,9 @@ MEM_STATIC size_t ZSTD_cwksp_align(size_t size, size_t const align) {
  * else is though.
  */
 MEM_STATIC size_t ZSTD_cwksp_alloc_size(size_t size) {
-#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    if (size == 0)
+        return 0;
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
     return size + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
 #else
     return size;
@@ -228,7 +241,10 @@ MEM_STATIC void* ZSTD_cwksp_reserve_internal(
     ZSTD_cwksp_internal_advance_phase(ws, phase);
     alloc = (BYTE *)ws->allocStart - bytes;
 
-#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    if (bytes == 0)
+        return NULL;
+
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
     /* over-reserve space */
     alloc = (BYTE *)alloc - 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
 #endif
@@ -247,11 +263,13 @@ MEM_STATIC void* ZSTD_cwksp_reserve_internal(
     }
     ws->allocStart = alloc;
 
-#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
     /* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
      * either size. */
     alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
-    __asan_unpoison_memory_region(alloc, bytes);
+    if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
+        __asan_unpoison_memory_region(alloc, bytes);
+    }
 #endif
 
     return alloc;
@@ -296,8 +314,10 @@ MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes) {
     }
     ws->tableEnd = end;
 
-#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
-    __asan_unpoison_memory_region(alloc, bytes);
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
+        __asan_unpoison_memory_region(alloc, bytes);
+    }
 #endif
 
     return alloc;
@@ -311,7 +331,7 @@ MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
     void* alloc = ws->objectEnd;
     void* end = (BYTE*)alloc + roundedBytes;
 
-#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
     /* over-reserve space */
     end = (BYTE *)end + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
 #endif
@@ -332,11 +352,13 @@ MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
     ws->tableEnd = end;
     ws->tableValidEnd = end;
 
-#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
     /* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
      * either size. */
     alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
-    __asan_unpoison_memory_region(alloc, bytes);
+    if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
+        __asan_unpoison_memory_region(alloc, bytes);
+    }
 #endif
 
     return alloc;
@@ -345,7 +367,7 @@ MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
 MEM_STATIC void ZSTD_cwksp_mark_tables_dirty(ZSTD_cwksp* ws) {
     DEBUGLOG(4, "cwksp: ZSTD_cwksp_mark_tables_dirty");
 
-#if defined (MEMORY_SANITIZER) && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
     /* To validate that the table re-use logic is sound, and that we don't
      * access table space that we haven't cleaned, we re-"poison" the table
      * space every time we mark it dirty. */
@@ -380,7 +402,7 @@ MEM_STATIC void ZSTD_cwksp_clean_tables(ZSTD_cwksp* ws) {
     assert(ws->tableValidEnd >= ws->objectEnd);
     assert(ws->tableValidEnd <= ws->allocStart);
     if (ws->tableValidEnd < ws->tableEnd) {
-        memset(ws->tableValidEnd, 0, (BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd);
+        ZSTD_memset(ws->tableValidEnd, 0, (BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd);
     }
     ZSTD_cwksp_mark_tables_clean(ws);
 }
@@ -392,8 +414,12 @@ MEM_STATIC void ZSTD_cwksp_clean_tables(ZSTD_cwksp* ws) {
 MEM_STATIC void ZSTD_cwksp_clear_tables(ZSTD_cwksp* ws) {
     DEBUGLOG(4, "cwksp: clearing tables!");
 
-#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
-    {
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* We don't do this when the workspace is statically allocated, because
+     * when that is the case, we have no capability to hook into the end of the
+     * workspace's lifecycle to unpoison the memory.
+     */
+    if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
         size_t size = (BYTE*)ws->tableValidEnd - (BYTE*)ws->objectEnd;
         __asan_poison_memory_region(ws->objectEnd, size);
     }
@@ -410,7 +436,7 @@ MEM_STATIC void ZSTD_cwksp_clear_tables(ZSTD_cwksp* ws) {
 MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
     DEBUGLOG(4, "cwksp: clearing!");
 
-#if defined (MEMORY_SANITIZER) && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
     /* To validate that the context re-use logic is sound, and that we don't
      * access stuff that this compression hasn't initialized, we re-"poison"
      * the workspace (or at least the non-static, non-table parts of it)
@@ -421,8 +447,12 @@ MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
     }
 #endif
 
-#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
-    {
+#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* We don't do this when the workspace is statically allocated, because
+     * when that is the case, we have no capability to hook into the end of the
+     * workspace's lifecycle to unpoison the memory.
+     */
+    if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
         size_t size = (BYTE*)ws->workspaceEnd - (BYTE*)ws->objectEnd;
         __asan_poison_memory_region(ws->objectEnd, size);
     }
@@ -442,7 +472,7 @@ MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
  * Any existing values in the workspace are ignored (the previously managed
  * buffer, if present, must be separately freed).
  */
-MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size) {
+MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size, ZSTD_cwksp_static_alloc_e isStatic) {
     DEBUGLOG(4, "cwksp: init'ing workspace with %zd bytes", size);
     assert(((size_t)start & (sizeof(void*)-1)) == 0); /* ensure correct alignment */
     ws->workspace = start;
@@ -450,24 +480,25 @@ MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size) {
     ws->objectEnd = ws->workspace;
     ws->tableValidEnd = ws->objectEnd;
     ws->phase = ZSTD_cwksp_alloc_objects;
+    ws->isStatic = isStatic;
     ZSTD_cwksp_clear(ws);
     ws->workspaceOversizedDuration = 0;
     ZSTD_cwksp_assert_internal_consistency(ws);
 }
 
 MEM_STATIC size_t ZSTD_cwksp_create(ZSTD_cwksp* ws, size_t size, ZSTD_customMem customMem) {
-    void* workspace = ZSTD_malloc(size, customMem);
+    void* workspace = ZSTD_customMalloc(size, customMem);
     DEBUGLOG(4, "cwksp: creating new workspace with %zd bytes", size);
     RETURN_ERROR_IF(workspace == NULL, memory_allocation, "NULL pointer!");
-    ZSTD_cwksp_init(ws, workspace, size);
+    ZSTD_cwksp_init(ws, workspace, size, ZSTD_cwksp_dynamic_alloc);
     return 0;
 }
 
 MEM_STATIC void ZSTD_cwksp_free(ZSTD_cwksp* ws, ZSTD_customMem customMem) {
     void *ptr = ws->workspace;
     DEBUGLOG(4, "cwksp: freeing workspace");
-    memset(ws, 0, sizeof(ZSTD_cwksp));
-    ZSTD_free(ptr, customMem);
+    ZSTD_memset(ws, 0, sizeof(ZSTD_cwksp));
+    ZSTD_customFree(ptr, customMem);
 }
 
 /**
@@ -476,13 +507,18 @@ MEM_STATIC void ZSTD_cwksp_free(ZSTD_cwksp* ws, ZSTD_customMem customMem) {
  */
 MEM_STATIC void ZSTD_cwksp_move(ZSTD_cwksp* dst, ZSTD_cwksp* src) {
     *dst = *src;
-    memset(src, 0, sizeof(ZSTD_cwksp));
+    ZSTD_memset(src, 0, sizeof(ZSTD_cwksp));
 }
 
 MEM_STATIC size_t ZSTD_cwksp_sizeof(const ZSTD_cwksp* ws) {
     return (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->workspace);
 }
 
+MEM_STATIC size_t ZSTD_cwksp_used(const ZSTD_cwksp* ws) {
+    return (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->workspace)
+         + (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->allocStart);
+}
+
 MEM_STATIC int ZSTD_cwksp_reserve_failed(const ZSTD_cwksp* ws) {
     return ws->allocFailed;
 }
diff --git a/lib/compress/zstd_double_fast.c b/lib/compress/zstd_double_fast.c
index 27eed66cfedd..ef12a524f7c6 100644
--- a/lib/compress/zstd_double_fast.c
+++ b/lib/compress/zstd_double_fast.c
@@ -31,15 +31,15 @@ void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
      * is empty.
      */
     for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
-        U32 const current = (U32)(ip - base);
+        U32 const curr = (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;
+                hashSmall[smHash] = curr + i;
             if (i == 0 || hashLarge[lgHash] == 0)
-                hashLarge[lgHash] = current + i;
+                hashLarge[lgHash] = curr + i;
             /* Only load extra positions for ZSTD_dtlm_full */
             if (dtlm == ZSTD_dtlm_fast)
                 break;
@@ -108,9 +108,9 @@ size_t ZSTD_compressBlock_doubleFast_generic(
     /* init */
     ip += (dictAndPrefixLength == 0);
     if (dictMode == ZSTD_noDict) {
-        U32 const current = (U32)(ip - base);
-        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
-        U32 const maxRep = current - windowLow;
+        U32 const curr = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
+        U32 const maxRep = curr - windowLow;
         if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
     }
@@ -129,17 +129,17 @@ size_t ZSTD_compressBlock_doubleFast_generic(
         size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
         size_t const dictHL = ZSTD_hashPtr(ip, dictHBitsL, 8);
         size_t const dictHS = ZSTD_hashPtr(ip, dictHBitsS, mls);
-        U32 const current = (U32)(ip-base);
+        U32 const curr = (U32)(ip-base);
         U32 const matchIndexL = hashLong[h2];
         U32 matchIndexS = hashSmall[h];
         const BYTE* matchLong = base + matchIndexL;
         const BYTE* match = base + matchIndexS;
-        const U32 repIndex = current + 1 - offset_1;
+        const U32 repIndex = curr + 1 - offset_1;
         const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
                             && repIndex < prefixLowestIndex) ?
                                dictBase + (repIndex - dictIndexDelta) :
                                base + repIndex;
-        hashLong[h2] = hashSmall[h] = current;   /* update hash tables */
+        hashLong[h2] = hashSmall[h] = curr;   /* update hash tables */
 
         /* check dictMatchState repcode */
         if (dictMode == ZSTD_dictMatchState
@@ -177,7 +177,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
 
             if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
                 mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
-                offset = (U32)(current - dictMatchIndexL - dictIndexDelta);
+                offset = (U32)(curr - dictMatchIndexL - dictIndexDelta);
                 while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
                 goto _match_found;
         }   }
@@ -209,7 +209,7 @@ _search_next_long:
             size_t const dictHLNext = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
             U32 const matchIndexL3 = hashLong[hl3];
             const BYTE* matchL3 = base + matchIndexL3;
-            hashLong[hl3] = current + 1;
+            hashLong[hl3] = curr + 1;
 
             /* check prefix long +1 match */
             if (matchIndexL3 > prefixLowestIndex) {
@@ -228,7 +228,7 @@ _search_next_long:
                 if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
                     mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
                     ip++;
-                    offset = (U32)(current + 1 - dictMatchIndexL3 - dictIndexDelta);
+                    offset = (U32)(curr + 1 - dictMatchIndexL3 - dictIndexDelta);
                     while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
                     goto _match_found;
         }   }   }
@@ -236,7 +236,7 @@ _search_next_long:
         /* if no long +1 match, explore the short match we found */
         if (dictMode == ZSTD_dictMatchState && matchIndexS < prefixLowestIndex) {
             mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
-            offset = (U32)(current - matchIndexS);
+            offset = (U32)(curr - matchIndexS);
             while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
         } else {
             mLength = ZSTD_count(ip+4, match+4, iend) + 4;
@@ -260,7 +260,7 @@ _match_stored:
         if (ip <= ilimit) {
             /* Complementary insertion */
             /* done after iLimit test, as candidates could be > iend-8 */
-            {   U32 const indexToInsert = current+2;
+            {   U32 const indexToInsert = curr+2;
                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
@@ -401,12 +401,12 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
         const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? 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 U32 curr = (U32)(ip-base);
+        const U32 repIndex = curr + 1 - offset_1;   /* offset_1 expected <= curr +1 */
         const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
         const BYTE* const repMatch = repBase + repIndex;
         size_t mLength;
-        hashSmall[hSmall] = hashLong[hLong] = current;   /* update hash table */
+        hashSmall[hSmall] = hashLong[hLong] = curr;   /* update hash table */
 
         if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
             & (repIndex > dictStartIndex))
@@ -421,7 +421,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                 const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
                 U32 offset;
                 mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
-                offset = current - matchLongIndex;
+                offset = curr - matchLongIndex;
                 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
                 offset_2 = offset_1;
                 offset_1 = offset;
@@ -433,19 +433,19 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                 const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
                 const BYTE* match3 = match3Base + matchIndex3;
                 U32 offset;
-                hashLong[h3] = current + 1;
+                hashLong[h3] = curr + 1;
                 if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
                     const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
                     const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
                     mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
                     ip++;
-                    offset = current+1 - matchIndex3;
+                    offset = curr+1 - matchIndex3;
                     while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
                 } else {
                     const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
                     const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
                     mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
-                    offset = current - matchIndex;
+                    offset = curr - matchIndex;
                     while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
                 }
                 offset_2 = offset_1;
@@ -464,7 +464,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
         if (ip <= ilimit) {
             /* Complementary insertion */
             /* done after iLimit test, as candidates could be > iend-8 */
-            {   U32 const indexToInsert = current+2;
+            {   U32 const indexToInsert = curr+2;
                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
diff --git a/lib/compress/zstd_fast.c b/lib/compress/zstd_fast.c
index 85a3a7a91e49..db7ce83d0ad7 100644
--- a/lib/compress/zstd_fast.c
+++ b/lib/compress/zstd_fast.c
@@ -29,16 +29,16 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
      * Insert the other positions if their hash entry is empty.
      */
     for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
-        U32 const current = (U32)(ip - base);
+        U32 const curr = (U32)(ip - base);
         size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls);
-        hashTable[hash0] = current;
+        hashTable[hash0] = curr;
         if (dtlm == ZSTD_dtlm_fast) continue;
         /* Only load extra positions for ZSTD_dtlm_full */
         {   U32 p;
             for (p = 1; p < fastHashFillStep; ++p) {
                 size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls);
                 if (hashTable[hash] == 0) {  /* not yet filled */
-                    hashTable[hash] = current + p;
+                    hashTable[hash] = curr + p;
     }   }   }   }
 }
 
@@ -72,9 +72,9 @@ ZSTD_compressBlock_fast_generic(
     DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
     ip0 += (ip0 == prefixStart);
     ip1 = ip0 + 1;
-    {   U32 const current = (U32)(ip0 - base);
-        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
-        U32 const maxRep = current - windowLow;
+    {   U32 const curr = (U32)(ip0 - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
+        U32 const maxRep = curr - windowLow;
         if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
     }
@@ -258,14 +258,14 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
     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 curr = (U32)(ip-base);
         U32 const matchIndex = hashTable[h];
         const BYTE* match = base + matchIndex;
-        const U32 repIndex = current + 1 - offset_1;
+        const U32 repIndex = curr + 1 - offset_1;
         const BYTE* repMatch = (repIndex < prefixStartIndex) ?
                                dictBase + (repIndex - dictIndexDelta) :
                                base + repIndex;
-        hashTable[h] = current;   /* update hash table */
+        hashTable[h] = curr;   /* update hash table */
 
         if ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
@@ -284,7 +284,7 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
                 continue;
             } else {
                 /* found a dict match */
-                U32 const offset = (U32)(current-dictMatchIndex-dictIndexDelta);
+                U32 const offset = (U32)(curr-dictMatchIndex-dictIndexDelta);
                 mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4;
                 while (((ip>anchor) & (dictMatch>dictStart))
                      && (ip[-1] == dictMatch[-1])) {
@@ -316,8 +316,8 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
 
         if (ip <= ilimit) {
             /* Fill Table */
-            assert(base+current+2 > istart);  /* check base overflow */
-            hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;  /* here because current+2 could be > iend-8 */
+            assert(base+curr+2 > istart);  /* check base overflow */
+            hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;  /* here because curr+2 could be > iend-8 */
             hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
 
             /* check immediate repcode */
@@ -410,13 +410,13 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
         const U32    matchIndex = hashTable[h];
         const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
         const BYTE*  match = matchBase + matchIndex;
-        const U32    current = (U32)(ip-base);
-        const U32    repIndex = current + 1 - offset_1;
+        const U32    curr = (U32)(ip-base);
+        const U32    repIndex = curr + 1 - offset_1;
         const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
         const BYTE* const repMatch = repBase + repIndex;
-        hashTable[h] = current;   /* update hash table */
-        DEBUGLOG(7, "offset_1 = %u , current = %u", offset_1, current);
-        assert(offset_1 <= current +1);   /* check repIndex */
+        hashTable[h] = curr;   /* update hash table */
+        DEBUGLOG(7, "offset_1 = %u , curr = %u", offset_1, curr);
+        assert(offset_1 <= curr +1);   /* check repIndex */
 
         if ( (((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > dictStartIndex))
            && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
@@ -435,7 +435,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
             }
             {   const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
                 const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
-                U32 const offset = current - matchIndex;
+                U32 const offset = curr - matchIndex;
                 size_t mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
                 while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
                 offset_2 = offset_1; offset_1 = offset;  /* update offset history */
@@ -446,7 +446,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
 
         if (ip <= ilimit) {
             /* Fill Table */
-            hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;
+            hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;
             hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
             /* check immediate repcode */
             while (ip <= ilimit) {
diff --git a/lib/compress/zstd_lazy.c b/lib/compress/zstd_lazy.c
index 4cf5c88b5325..49ec1b09eff7 100644
--- a/lib/compress/zstd_lazy.c
+++ b/lib/compress/zstd_lazy.c
@@ -58,11 +58,11 @@ ZSTD_updateDUBT(ZSTD_matchState_t* ms,
 
 /** ZSTD_insertDUBT1() :
  *  sort one already inserted but unsorted position
- *  assumption : current >= btlow == (current - btmask)
+ *  assumption : curr >= btlow == (curr - btmask)
  *  doesn't fail */
 static void
 ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
-                 U32 current, const BYTE* inputEnd,
+                 U32 curr, const BYTE* inputEnd,
                  U32 nbCompares, U32 btLow,
                  const ZSTD_dictMode_e dictMode)
 {
@@ -74,41 +74,41 @@ ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
     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 ip = (curr>=dictLimit) ? base + curr : dictBase + curr;
+    const BYTE* const iend = (curr>=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* smallerPtr = bt + 2*(curr&btMask);
     U32* largerPtr  = smallerPtr + 1;
     U32 matchIndex = *smallerPtr;   /* this candidate is unsorted : next sorted candidate is reached through *smallerPtr, while *largerPtr contains previous unsorted candidate (which is already saved and can be overwritten) */
     U32 dummy32;   /* to be nullified at the end */
     U32 const windowValid = ms->window.lowLimit;
     U32 const maxDistance = 1U << cParams->windowLog;
-    U32 const windowLow = (current - windowValid > maxDistance) ? current - maxDistance : windowValid;
+    U32 const windowLow = (curr - windowValid > maxDistance) ? curr - maxDistance : windowValid;
 
 
     DEBUGLOG(8, "ZSTD_insertDUBT1(%u) (dictLimit=%u, lowLimit=%u)",
-                current, dictLimit, windowLow);
-    assert(current >= btLow);
+                curr, dictLimit, windowLow);
+    assert(curr >= 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);
+        assert(matchIndex < curr);
         /* note : all candidates are now supposed sorted,
          * but it's still possible to have nextPtr[1] == ZSTD_DUBT_UNSORTED_MARK
          * when a real index has the same value as ZSTD_DUBT_UNSORTED_MARK */
 
         if ( (dictMode != ZSTD_extDict)
           || (matchIndex+matchLength >= dictLimit)  /* both in current segment*/
-          || (current < dictLimit) /* both in extDict */) {
+          || (curr < dictLimit) /* both in extDict */) {
             const BYTE* const mBase = ( (dictMode != ZSTD_extDict)
                                      || (matchIndex+matchLength >= dictLimit)) ?
                                         base : dictBase;
             assert( (matchIndex+matchLength >= dictLimit)   /* might be wrong if extDict is incorrectly set to 0 */
-                 || (current < dictLimit) );
+                 || (curr < dictLimit) );
             match = mBase + matchIndex;
             matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
         } else {
@@ -119,7 +119,7 @@ ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
         }
 
         DEBUGLOG(8, "ZSTD_insertDUBT1: comparing %u with %u : found %u common bytes ",
-                    current, matchIndex, (U32)matchLength);
+                    curr, 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 */
@@ -168,7 +168,7 @@ ZSTD_DUBT_findBetterDictMatch (
 
     const BYTE* const base = ms->window.base;
     const BYTE* const prefixStart = base + ms->window.dictLimit;
-    U32         const current = (U32)(ip-base);
+    U32         const curr = (U32)(ip-base);
     const BYTE* const dictBase = dms->window.base;
     const BYTE* const dictEnd = dms->window.nextSrc;
     U32         const dictHighLimit = (U32)(dms->window.nextSrc - dms->window.base);
@@ -195,10 +195,10 @@ ZSTD_DUBT_findBetterDictMatch (
 
         if (matchLength > bestLength) {
             U32 matchIndex = dictMatchIndex + dictIndexDelta;
-            if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) {
+            if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) {
                 DEBUGLOG(9, "ZSTD_DUBT_findBetterDictMatch(%u) : found better match length %u -> %u and offsetCode %u -> %u (dictMatchIndex %u, matchIndex %u)",
-                    current, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, ZSTD_REP_MOVE + current - matchIndex, dictMatchIndex, matchIndex);
-                bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + current - matchIndex;
+                    curr, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, ZSTD_REP_MOVE + curr - matchIndex, dictMatchIndex, matchIndex);
+                bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + curr - matchIndex;
             }
             if (ip+matchLength == iend) {   /* reached end of input : ip[matchLength] is not valid, no way to know if it's larger or smaller than match */
                 break;   /* drop, to guarantee consistency (miss a little bit of compression) */
@@ -218,9 +218,9 @@ ZSTD_DUBT_findBetterDictMatch (
     }
 
     if (bestLength >= MINMATCH) {
-        U32 const mIndex = current - ((U32)*offsetPtr - ZSTD_REP_MOVE); (void)mIndex;
+        U32 const mIndex = curr - ((U32)*offsetPtr - ZSTD_REP_MOVE); (void)mIndex;
         DEBUGLOG(8, "ZSTD_DUBT_findBetterDictMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
-                    current, (U32)bestLength, (U32)*offsetPtr, mIndex);
+                    curr, (U32)bestLength, (U32)*offsetPtr, mIndex);
     }
     return bestLength;
 
@@ -241,13 +241,13 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
     U32          matchIndex  = hashTable[h];
 
     const BYTE* const base = ms->window.base;
-    U32    const current = (U32)(ip-base);
-    U32    const windowLow = ZSTD_getLowestMatchIndex(ms, current, cParams->windowLog);
+    U32    const curr = (U32)(ip-base);
+    U32    const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
 
     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 btLow = (btMask >= curr) ? 0 : curr - btMask;
     U32    const unsortLimit = MAX(btLow, windowLow);
 
     U32*         nextCandidate = bt + 2*(matchIndex&btMask);
@@ -256,8 +256,9 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
     U32          nbCandidates = nbCompares;
     U32          previousCandidate = 0;
 
-    DEBUGLOG(7, "ZSTD_DUBT_findBestMatch (%u) ", current);
+    DEBUGLOG(7, "ZSTD_DUBT_findBestMatch (%u) ", curr);
     assert(ip <= iend-8);   /* required for h calculation */
+    assert(dictMode != ZSTD_dedicatedDictSearch);
 
     /* reach end of unsorted candidates list */
     while ( (matchIndex > unsortLimit)
@@ -299,14 +300,14 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
         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* smallerPtr = bt + 2*(curr&btMask);
+        U32* largerPtr  = bt + 2*(curr&btMask) + 1;
+        U32 matchEndIdx = curr + 8 + 1;
         U32 dummy32;   /* to be nullified at the end */
         size_t bestLength = 0;
 
         matchIndex  = hashTable[h];
-        hashTable[h] = current;   /* Update Hash Table */
+        hashTable[h] = curr;   /* Update Hash Table */
 
         while (nbCompares-- && (matchIndex > windowLow)) {
             U32* const nextPtr = bt + 2*(matchIndex & btMask);
@@ -326,8 +327,8 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
             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 ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
+                    bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + curr - matchIndex;
                 if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
                     if (dictMode == ZSTD_dictMatchState) {
                         nbCompares = 0; /* in addition to avoiding checking any
@@ -363,12 +364,12 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
                     mls, dictMode);
         }
 
-        assert(matchEndIdx > current+8); /* ensure nextToUpdate is increased */
+        assert(matchEndIdx > curr+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;
+            U32 const mIndex = curr - ((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);
+                        curr, (U32)bestLength, (U32)*offsetPtr, mIndex);
         }
         return bestLength;
     }
@@ -446,7 +447,7 @@ static size_t ZSTD_BtFindBestMatch_extDict_selectMLS (
 
 /* Update chains up to ip (excluded)
    Assumption : always within prefix (i.e. not within extDict) */
-static U32 ZSTD_insertAndFindFirstIndex_internal(
+FORCE_INLINE_TEMPLATE U32 ZSTD_insertAndFindFirstIndex_internal(
                         ZSTD_matchState_t* ms,
                         const ZSTD_compressionParameters* const cParams,
                         const BYTE* ip, U32 const mls)
@@ -475,6 +476,121 @@ U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip) {
     return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, ms->cParams.minMatch);
 }
 
+void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_matchState_t* ms, const BYTE* const ip)
+{
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32* const hashTable = ms->hashTable;
+    U32* const chainTable = ms->chainTable;
+    U32 const chainSize = 1 << ms->cParams.chainLog;
+    U32 idx = ms->nextToUpdate;
+    U32 const minChain = chainSize < target ? target - chainSize : idx;
+    U32 const bucketSize = 1 << ZSTD_LAZY_DDSS_BUCKET_LOG;
+    U32 const cacheSize = bucketSize - 1;
+    U32 const chainAttempts = (1 << ms->cParams.searchLog) - cacheSize;
+    U32 const chainLimit = chainAttempts > 255 ? 255 : chainAttempts;
+
+    /* We know the hashtable is oversized by a factor of `bucketSize`.
+     * We are going to temporarily pretend `bucketSize == 1`, keeping only a
+     * single entry. We will use the rest of the space to construct a temporary
+     * chaintable.
+     */
+    U32 const hashLog = ms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
+    U32* const tmpHashTable = hashTable;
+    U32* const tmpChainTable = hashTable + ((size_t)1 << hashLog);
+    U32 const tmpChainSize = ((1 << ZSTD_LAZY_DDSS_BUCKET_LOG) - 1) << hashLog;
+    U32 const tmpMinChain = tmpChainSize < target ? target - tmpChainSize : idx;
+
+    U32 hashIdx;
+
+    assert(ms->cParams.chainLog <= 24);
+    assert(ms->cParams.hashLog >= ms->cParams.chainLog);
+    assert(idx != 0);
+    assert(tmpMinChain <= minChain);
+
+    /* fill conventional hash table and conventional chain table */
+    for ( ; idx < target; idx++) {
+        U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch);
+        if (idx >= tmpMinChain) {
+            tmpChainTable[idx - tmpMinChain] = hashTable[h];
+        }
+        tmpHashTable[h] = idx;
+    }
+
+    /* sort chains into ddss chain table */
+    {
+        U32 chainPos = 0;
+        for (hashIdx = 0; hashIdx < (1U << hashLog); hashIdx++) {
+            U32 count;
+            U32 countBeyondMinChain = 0;
+            U32 i = tmpHashTable[hashIdx];
+            for (count = 0; i >= tmpMinChain && count < cacheSize; count++) {
+                /* skip through the chain to the first position that won't be
+                 * in the hash cache bucket */
+                if (i < minChain) {
+                    countBeyondMinChain++;
+                }
+                i = tmpChainTable[i - tmpMinChain];
+            }
+            if (count == cacheSize) {
+                for (count = 0; count < chainLimit;) {
+                    if (i < minChain) {
+                        if (!i || countBeyondMinChain++ > cacheSize) {
+                            /* only allow pulling `cacheSize` number of entries
+                             * into the cache or chainTable beyond `minChain`,
+                             * to replace the entries pulled out of the
+                             * chainTable into the cache. This lets us reach
+                             * back further without increasing the total number
+                             * of entries in the chainTable, guaranteeing the
+                             * DDSS chain table will fit into the space
+                             * allocated for the regular one. */
+                            break;
+                        }
+                    }
+                    chainTable[chainPos++] = i;
+                    count++;
+                    if (i < tmpMinChain) {
+                        break;
+                    }
+                    i = tmpChainTable[i - tmpMinChain];
+                }
+            } else {
+                count = 0;
+            }
+            if (count) {
+                tmpHashTable[hashIdx] = ((chainPos - count) << 8) + count;
+            } else {
+                tmpHashTable[hashIdx] = 0;
+            }
+        }
+        assert(chainPos <= chainSize); /* I believe this is guaranteed... */
+    }
+
+    /* move chain pointers into the last entry of each hash bucket */
+    for (hashIdx = (1 << hashLog); hashIdx; ) {
+        U32 const bucketIdx = --hashIdx << ZSTD_LAZY_DDSS_BUCKET_LOG;
+        U32 const chainPackedPointer = tmpHashTable[hashIdx];
+        U32 i;
+        for (i = 0; i < cacheSize; i++) {
+            hashTable[bucketIdx + i] = 0;
+        }
+        hashTable[bucketIdx + bucketSize - 1] = chainPackedPointer;
+    }
+
+    /* fill the buckets of the hash table */
+    for (idx = ms->nextToUpdate; idx < target; idx++) {
+        U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch)
+                   << ZSTD_LAZY_DDSS_BUCKET_LOG;
+        U32 i;
+        /* Shift hash cache down 1. */
+        for (i = cacheSize - 1; i; i--)
+            hashTable[h + i] = hashTable[h + i - 1];
+        hashTable[h] = idx;
+    }
+
+    ms->nextToUpdate = target;
+}
+
 
 /* inlining is important to hardwire a hot branch (template emulation) */
 FORCE_INLINE_TEMPLATE
@@ -493,20 +609,33 @@ size_t ZSTD_HcFindBestMatch_generic (
     const U32 dictLimit = ms->window.dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
     const BYTE* const dictEnd = dictBase + dictLimit;
-    const U32 current = (U32)(ip-base);
+    const U32 curr = (U32)(ip-base);
     const U32 maxDistance = 1U << cParams->windowLog;
     const U32 lowestValid = ms->window.lowLimit;
-    const U32 withinMaxDistance = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
+    const U32 withinMaxDistance = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
     const U32 isDictionary = (ms->loadedDictEnd != 0);
     const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance;
-    const U32 minChain = current > chainSize ? current - chainSize : 0;
+    const U32 minChain = curr > chainSize ? curr - chainSize : 0;
     U32 nbAttempts = 1U << cParams->searchLog;
     size_t ml=4-1;
 
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const U32 ddsHashLog = dictMode == ZSTD_dedicatedDictSearch
+                         ? dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG : 0;
+    const size_t ddsIdx = dictMode == ZSTD_dedicatedDictSearch
+                        ? ZSTD_hashPtr(ip, ddsHashLog, mls) << ZSTD_LAZY_DDSS_BUCKET_LOG : 0;
+
+    U32 matchIndex;
+
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        const U32* entry = &dms->hashTable[ddsIdx];
+        PREFETCH_L1(entry);
+    }
+
     /* HC4 match finder */
-    U32 matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls);
+    matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls);
 
-    for ( ; (matchIndex>lowLimit) & (nbAttempts>0) ; nbAttempts--) {
+    for ( ; (matchIndex>=lowLimit) & (nbAttempts>0) ; nbAttempts--) {
         size_t currentMl=0;
         if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
             const BYTE* const match = base + matchIndex;
@@ -523,7 +652,7 @@ size_t ZSTD_HcFindBestMatch_generic (
         /* save best solution */
         if (currentMl > ml) {
             ml = currentMl;
-            *offsetPtr = current - matchIndex + ZSTD_REP_MOVE;
+            *offsetPtr = curr - matchIndex + ZSTD_REP_MOVE;
             if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
         }
 
@@ -531,8 +660,92 @@ size_t ZSTD_HcFindBestMatch_generic (
         matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
     }
 
-    if (dictMode == ZSTD_dictMatchState) {
-        const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    if (dictMode == ZSTD_dedicatedDictSearch) {
+        const U32 ddsLowestIndex  = dms->window.dictLimit;
+        const BYTE* const ddsBase = dms->window.base;
+        const BYTE* const ddsEnd  = dms->window.nextSrc;
+        const U32 ddsSize         = (U32)(ddsEnd - ddsBase);
+        const U32 ddsIndexDelta   = dictLimit - ddsSize;
+        const U32 bucketSize      = (1 << ZSTD_LAZY_DDSS_BUCKET_LOG);
+        const U32 bucketLimit     = nbAttempts < bucketSize - 1 ? nbAttempts : bucketSize - 1;
+        U32 ddsAttempt;
+
+        for (ddsAttempt = 0; ddsAttempt < bucketSize - 1; ddsAttempt++) {
+            PREFETCH_L1(ddsBase + dms->hashTable[ddsIdx + ddsAttempt]);
+        }
+
+        {
+            U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1];
+            U32 const chainIndex = chainPackedPointer >> 8;
+
+            PREFETCH_L1(&dms->chainTable[chainIndex]);
+        }
+
+        for (ddsAttempt = 0; ddsAttempt < bucketLimit; ddsAttempt++) {
+            size_t currentMl=0;
+            const BYTE* match;
+            matchIndex = dms->hashTable[ddsIdx + ddsAttempt];
+            match = ddsBase + matchIndex;
+
+            if (!matchIndex) {
+                return ml;
+            }
+
+            /* guaranteed by table construction */
+            (void)ddsLowestIndex;
+            assert(matchIndex >= ddsLowestIndex);
+            assert(match+4 <= ddsEnd);
+            if (MEM_read32(match) == MEM_read32(ip)) {
+                /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4;
+            }
+
+            /* save best solution */
+            if (currentMl > ml) {
+                ml = currentMl;
+                *offsetPtr = curr - (matchIndex + ddsIndexDelta) + ZSTD_REP_MOVE;
+                if (ip+currentMl == iLimit) {
+                    /* best possible, avoids read overflow on next attempt */
+                    return ml;
+                }
+            }
+        }
+
+        {
+            U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1];
+            U32 chainIndex = chainPackedPointer >> 8;
+            U32 const chainLength = chainPackedPointer & 0xFF;
+            U32 const chainAttempts = nbAttempts - ddsAttempt;
+            U32 const chainLimit = chainAttempts > chainLength ? chainLength : chainAttempts;
+            U32 chainAttempt;
+
+            for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++) {
+                PREFETCH_L1(ddsBase + dms->chainTable[chainIndex + chainAttempt]);
+            }
+
+            for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++, chainIndex++) {
+                size_t currentMl=0;
+                const BYTE* match;
+                matchIndex = dms->chainTable[chainIndex];
+                match = ddsBase + matchIndex;
+
+                /* guaranteed by table construction */
+                assert(matchIndex >= ddsLowestIndex);
+                assert(match+4 <= ddsEnd);
+                if (MEM_read32(match) == MEM_read32(ip)) {
+                    /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                    currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4;
+                }
+
+                /* save best solution */
+                if (currentMl > ml) {
+                    ml = currentMl;
+                    *offsetPtr = curr - (matchIndex + ddsIndexDelta) + ZSTD_REP_MOVE;
+                    if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+                }
+            }
+        }
+    } else if (dictMode == ZSTD_dictMatchState) {
         const U32* const dmsChainTable = dms->chainTable;
         const U32 dmsChainSize         = (1 << dms->cParams.chainLog);
         const U32 dmsChainMask         = dmsChainSize - 1;
@@ -545,7 +758,7 @@ size_t ZSTD_HcFindBestMatch_generic (
 
         matchIndex = dms->hashTable[ZSTD_hashPtr(ip, dms->cParams.hashLog, mls)];
 
-        for ( ; (matchIndex>dmsLowestIndex) & (nbAttempts>0) ; nbAttempts--) {
+        for ( ; (matchIndex>=dmsLowestIndex) & (nbAttempts>0) ; nbAttempts--) {
             size_t currentMl=0;
             const BYTE* const match = dmsBase + matchIndex;
             assert(match+4 <= dmsEnd);
@@ -555,11 +768,12 @@ size_t ZSTD_HcFindBestMatch_generic (
             /* save best solution */
             if (currentMl > ml) {
                 ml = currentMl;
-                *offsetPtr = current - (matchIndex + dmsIndexDelta) + ZSTD_REP_MOVE;
+                *offsetPtr = curr - (matchIndex + dmsIndexDelta) + ZSTD_REP_MOVE;
                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
             }
 
             if (matchIndex <= dmsMinChain) break;
+
             matchIndex = dmsChainTable[matchIndex & dmsChainMask];
         }
     }
@@ -600,6 +814,22 @@ static size_t ZSTD_HcFindBestMatch_dictMatchState_selectMLS (
 }
 
 
+static size_t ZSTD_HcFindBestMatch_dedicatedDictSearch_selectMLS (
+                        ZSTD_matchState_t* ms,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
+{
+    switch(ms->cParams.minMatch)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_dedicatedDictSearch);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_dedicatedDictSearch);
+    case 7 :
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_dedicatedDictSearch);
+    }
+}
+
+
 FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
                         ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* const iLimit,
@@ -641,39 +871,62 @@ ZSTD_compressBlock_lazy_generic(
     typedef size_t (*searchMax_f)(
                         ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
-    searchMax_f const searchMax = dictMode == ZSTD_dictMatchState ?
-        (searchMethod==search_binaryTree ? ZSTD_BtFindBestMatch_dictMatchState_selectMLS
-                                         : ZSTD_HcFindBestMatch_dictMatchState_selectMLS) :
-        (searchMethod==search_binaryTree ? ZSTD_BtFindBestMatch_selectMLS
-                                         : ZSTD_HcFindBestMatch_selectMLS);
+
+    /**
+     * This table is indexed first by the four ZSTD_dictMode_e values, and then
+     * by the two searchMethod_e values. NULLs are placed for configurations
+     * that should never occur (extDict modes go to the other implementation
+     * below and there is no DDSS for binary tree search yet).
+     */
+    const searchMax_f searchFuncs[4][2] = {
+        {
+            ZSTD_HcFindBestMatch_selectMLS,
+            ZSTD_BtFindBestMatch_selectMLS
+        },
+        {
+            NULL,
+            NULL
+        },
+        {
+            ZSTD_HcFindBestMatch_dictMatchState_selectMLS,
+            ZSTD_BtFindBestMatch_dictMatchState_selectMLS
+        },
+        {
+            ZSTD_HcFindBestMatch_dedicatedDictSearch_selectMLS,
+            NULL
+        }
+    };
+
+    searchMax_f const searchMax = searchFuncs[dictMode][searchMethod == search_binaryTree];
     U32 offset_1 = rep[0], offset_2 = rep[1], savedOffset=0;
 
+    const int isDMS = dictMode == ZSTD_dictMatchState;
+    const int isDDS = dictMode == ZSTD_dedicatedDictSearch;
+    const int isDxS = isDMS || isDDS;
     const ZSTD_matchState_t* const dms = ms->dictMatchState;
-    const U32 dictLowestIndex      = dictMode == ZSTD_dictMatchState ?
-                                     dms->window.dictLimit : 0;
-    const BYTE* const dictBase     = dictMode == ZSTD_dictMatchState ?
-                                     dms->window.base : NULL;
-    const BYTE* const dictLowest   = dictMode == ZSTD_dictMatchState ?
-                                     dictBase + dictLowestIndex : NULL;
-    const BYTE* const dictEnd      = dictMode == ZSTD_dictMatchState ?
-                                     dms->window.nextSrc : NULL;
-    const U32 dictIndexDelta       = dictMode == ZSTD_dictMatchState ?
+    const U32 dictLowestIndex      = isDxS ? dms->window.dictLimit : 0;
+    const BYTE* const dictBase     = isDxS ? dms->window.base : NULL;
+    const BYTE* const dictLowest   = isDxS ? dictBase + dictLowestIndex : NULL;
+    const BYTE* const dictEnd      = isDxS ? dms->window.nextSrc : NULL;
+    const U32 dictIndexDelta       = isDxS ?
                                      prefixLowestIndex - (U32)(dictEnd - dictBase) :
                                      0;
     const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictLowest));
 
+    assert(searchMax != NULL);
+
     DEBUGLOG(5, "ZSTD_compressBlock_lazy_generic (dictMode=%u)", (U32)dictMode);
 
     /* init */
     ip += (dictAndPrefixLength == 0);
     if (dictMode == ZSTD_noDict) {
-        U32 const current = (U32)(ip - base);
-        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, ms->cParams.windowLog);
-        U32 const maxRep = current - windowLow;
+        U32 const curr = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, ms->cParams.windowLog);
+        U32 const maxRep = curr - windowLow;
         if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
     }
-    if (dictMode == ZSTD_dictMatchState) {
+    if (isDxS) {
         /* dictMatchState repCode checks don't currently handle repCode == 0
          * disabling. */
         assert(offset_1 <= dictAndPrefixLength);
@@ -693,9 +946,9 @@ ZSTD_compressBlock_lazy_generic(
         const BYTE* start=ip+1;
 
         /* check repCode */
-        if (dictMode == ZSTD_dictMatchState) {
+        if (isDxS) {
             const U32 repIndex = (U32)(ip - base) + 1 - offset_1;
-            const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
+            const BYTE* repMatch = ((dictMode == ZSTD_dictMatchState || dictMode == ZSTD_dedicatedDictSearch)
                                 && repIndex < prefixLowestIndex) ?
                                    dictBase + (repIndex - dictIndexDelta) :
                                    base + repIndex;
@@ -736,7 +989,7 @@ ZSTD_compressBlock_lazy_generic(
                 if ((mlRep >= 4) && (gain2 > gain1))
                     matchLength = mlRep, offset = 0, start = ip;
             }
-            if (dictMode == ZSTD_dictMatchState) {
+            if (isDxS) {
                 const U32 repIndex = (U32)(ip - base) - offset_1;
                 const BYTE* repMatch = repIndex < prefixLowestIndex ?
                                dictBase + (repIndex - dictIndexDelta) :
@@ -771,7 +1024,7 @@ ZSTD_compressBlock_lazy_generic(
                     if ((mlRep >= 4) && (gain2 > gain1))
                         matchLength = mlRep, offset = 0, start = ip;
                 }
-                if (dictMode == ZSTD_dictMatchState) {
+                if (isDxS) {
                     const U32 repIndex = (U32)(ip - base) - offset_1;
                     const BYTE* repMatch = repIndex < prefixLowestIndex ?
                                    dictBase + (repIndex - dictIndexDelta) :
@@ -809,7 +1062,7 @@ ZSTD_compressBlock_lazy_generic(
                      && (start[-1] == (start-(offset-ZSTD_REP_MOVE))[-1]) )  /* only search for offset within prefix */
                     { start--; matchLength++; }
             }
-            if (dictMode == ZSTD_dictMatchState) {
+            if (isDxS) {
                 U32 const matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
                 const BYTE* match = (matchIndex < prefixLowestIndex) ? dictBase + matchIndex - dictIndexDelta : base + matchIndex;
                 const BYTE* const mStart = (matchIndex < prefixLowestIndex) ? dictLowest : prefixLowest;
@@ -825,12 +1078,11 @@ _storeSequence:
         }
 
         /* check immediate repcode */
-        if (dictMode == ZSTD_dictMatchState) {
+        if (isDxS) {
             while (ip <= ilimit) {
                 U32 const current2 = (U32)(ip-base);
                 U32 const repIndex = current2 - offset_2;
-                const BYTE* repMatch = dictMode == ZSTD_dictMatchState
-                    && repIndex < prefixLowestIndex ?
+                const BYTE* repMatch = repIndex < prefixLowestIndex ?
                         dictBase - dictIndexDelta + repIndex :
                         base + repIndex;
                 if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex) >= 3 /* intentional overflow */)
@@ -925,6 +1177,28 @@ size_t ZSTD_compressBlock_greedy_dictMatchState(
 }
 
 
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dedicatedDictSearch);
+}
+
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dedicatedDictSearch);
+}
+
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dedicatedDictSearch);
+}
+
+
 FORCE_INLINE_TEMPLATE
 size_t ZSTD_compressBlock_lazy_extDict_generic(
                         ZSTD_matchState_t* ms, seqStore_t* seqStore,
@@ -968,11 +1242,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
         size_t matchLength=0;
         size_t offset=0;
         const BYTE* start=ip+1;
-        U32 current = (U32)(ip-base);
+        U32 curr = (U32)(ip-base);
 
         /* check repCode */
-        {   const U32 windowLow = ZSTD_getLowestMatchIndex(ms, current+1, windowLog);
-            const U32 repIndex = (U32)(current+1 - offset_1);
+        {   const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr+1, windowLog);
+            const U32 repIndex = (U32)(curr+1 - offset_1);
             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
             const BYTE* const repMatch = repBase + repIndex;
             if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))   /* intentional overflow */
@@ -999,11 +1273,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
         if (depth>=1)
         while (ip<ilimit) {
             ip ++;
-            current++;
+            curr++;
             /* check repCode */
             if (offset) {
-                const U32 windowLow = ZSTD_getLowestMatchIndex(ms, current, windowLog);
-                const U32 repIndex = (U32)(current - offset_1);
+                const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
+                const U32 repIndex = (U32)(curr - offset_1);
                 const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                 const BYTE* const repMatch = repBase + repIndex;
                 if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))  /* intentional overflow */
@@ -1030,11 +1304,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
             /* let's find an even better one */
             if ((depth==2) && (ip<ilimit)) {
                 ip ++;
-                current++;
+                curr++;
                 /* check repCode */
                 if (offset) {
-                    const U32 windowLow = ZSTD_getLowestMatchIndex(ms, current, windowLog);
-                    const U32 repIndex = (U32)(current - offset_1);
+                    const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
+                    const U32 repIndex = (U32)(curr - offset_1);
                     const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                     const BYTE* const repMatch = repBase + repIndex;
                     if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))  /* intentional overflow */
diff --git a/lib/compress/zstd_lazy.h b/lib/compress/zstd_lazy.h
index 581936f03bd4..d0214d5e7396 100644
--- a/lib/compress/zstd_lazy.h
+++ b/lib/compress/zstd_lazy.h
@@ -17,8 +17,18 @@ extern "C" {
 
 #include "zstd_compress_internal.h"
 
+/**
+ * Dedicated Dictionary Search Structure bucket log. In the
+ * ZSTD_dedicatedDictSearch mode, the hashTable has
+ * 2 ** ZSTD_LAZY_DDSS_BUCKET_LOG entries in each bucket, rather than just
+ * one.
+ */
+#define ZSTD_LAZY_DDSS_BUCKET_LOG 2
+
 U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip);
 
+void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_matchState_t* ms, const BYTE* const ip);
+
 void ZSTD_preserveUnsortedMark (U32* const table, U32 const size, U32 const reducerValue);  /*! used in ZSTD_reduceIndex(). preemptively increase value of ZSTD_DUBT_UNSORTED_MARK */
 
 size_t ZSTD_compressBlock_btlazy2(
@@ -47,6 +57,16 @@ size_t ZSTD_compressBlock_greedy_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
 
+size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
 size_t ZSTD_compressBlock_greedy_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
diff --git a/lib/compress/zstd_ldm.c b/lib/compress/zstd_ldm.c
index 8c479483581c..3f3d7c46ab02 100644
--- a/lib/compress/zstd_ldm.c
+++ b/lib/compress/zstd_ldm.c
@@ -27,13 +27,6 @@ void ZSTD_ldm_adjustParameters(ldmParams_t* params,
     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, params->windowLog - LDM_HASH_RLOG);
         assert(params->hashLog <= ZSTD_HASHLOG_MAX);
@@ -150,10 +143,10 @@ static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
  *  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)
+            const BYTE* pMatch, const BYTE* pMatchBase)
 {
     size_t matchLength = 0;
-    while (pIn > pAnchor && pMatch > pBase && pIn[-1] == pMatch[-1]) {
+    while (pIn > pAnchor && pMatch > pMatchBase && pIn[-1] == pMatch[-1]) {
         pIn--;
         pMatch--;
         matchLength++;
@@ -161,6 +154,27 @@ static size_t ZSTD_ldm_countBackwardsMatch(
     return matchLength;
 }
 
+/** ZSTD_ldm_countBackwardsMatch_2segments() :
+ *  Returns the number of bytes that match backwards from pMatch,
+ *  even with the backwards match spanning 2 different segments.
+ *
+ *  On reaching `pMatchBase`, start counting from mEnd */
+static size_t ZSTD_ldm_countBackwardsMatch_2segments(
+                    const BYTE* pIn, const BYTE* pAnchor,
+                    const BYTE* pMatch, const BYTE* pMatchBase,
+                    const BYTE* pExtDictStart, const BYTE* pExtDictEnd)
+{
+    size_t matchLength = ZSTD_ldm_countBackwardsMatch(pIn, pAnchor, pMatch, pMatchBase);
+    if (pMatch - matchLength != pMatchBase || pMatchBase == pExtDictStart) {
+        /* If backwards match is entirely in the extDict or prefix, immediately return */
+        return matchLength;
+    }
+    DEBUGLOG(7, "ZSTD_ldm_countBackwardsMatch_2segments: found 2-parts backwards match (length in prefix==%zu)", matchLength);
+    matchLength += ZSTD_ldm_countBackwardsMatch(pIn - matchLength, pAnchor, pExtDictEnd, pExtDictStart);
+    DEBUGLOG(7, "final backwards match length = %zu", matchLength);
+    return matchLength;
+}
+
 /** ZSTD_ldm_fillFastTables() :
  *
  *  Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
@@ -246,10 +260,10 @@ void ZSTD_ldm_fillHashTable(
  *  (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) {
+    U32 const curr = (U32)(anchor - ms->window.base);
+    if (curr > ms->nextToUpdate + 1024) {
         ms->nextToUpdate =
-            current - MIN(512, current - ms->nextToUpdate - 1024);
+            curr - MIN(512, curr - ms->nextToUpdate - 1024);
     }
 }
 
@@ -286,7 +300,7 @@ static size_t ZSTD_ldm_generateSequences_internal(
 
     while (ip <= ilimit) {
         size_t mLength;
-        U32 const current = (U32)(ip - base);
+        U32 const curr = (U32)(ip - base);
         size_t forwardMatchLength = 0, backwardMatchLength = 0;
         ldmEntry_t* bestEntry = NULL;
         if (ip != istart) {
@@ -336,8 +350,9 @@ static size_t ZSTD_ldm_generateSequences_internal(
                         continue;
                     }
                     curBackwardMatchLength =
-                        ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
-                                                     lowMatchPtr);
+                        ZSTD_ldm_countBackwardsMatch_2segments(ip, anchor,
+                                                               pMatch, lowMatchPtr,
+                                                               dictStart, dictEnd);
                     curTotalMatchLength = curForwardMatchLength +
                                           curBackwardMatchLength;
                 } else { /* !extDict */
@@ -365,7 +380,7 @@ static size_t ZSTD_ldm_generateSequences_internal(
         /* No match found -- continue searching */
         if (bestEntry == NULL) {
             ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash,
-                                             hBits, current,
+                                             hBits, curr,
                                              *params);
             ip++;
             continue;
@@ -377,11 +392,11 @@ static size_t ZSTD_ldm_generateSequences_internal(
 
         {
             /* Store the sequence:
-             * ip = current - backwardMatchLength
+             * ip = curr - backwardMatchLength
              * The match is at (bestEntry->offset - backwardMatchLength)
              */
             U32 const matchIndex = bestEntry->offset;
-            U32 const offset = current - matchIndex;
+            U32 const offset = curr - matchIndex;
             rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
 
             /* Out of sequence storage */
@@ -562,6 +577,23 @@ static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
     return sequence;
 }
 
+void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
+    U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
+    while (currPos && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
+        if (currPos >= currSeq.litLength + currSeq.matchLength) {
+            currPos -= currSeq.litLength + currSeq.matchLength;
+            rawSeqStore->pos++;
+        } else {
+            rawSeqStore->posInSequence = currPos;
+            break;
+        }
+    }
+    if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
+        rawSeqStore->posInSequence = 0;
+    }
+}
+
 size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
     ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
     void const* src, size_t srcSize)
@@ -577,6 +609,15 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
     BYTE const* ip = istart;
 
     DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
+    /* If using opt parser, use LDMs only as candidates rather than always accepting them */
+    if (cParams->strategy >= ZSTD_btopt) {
+        size_t lastLLSize;
+        ms->ldmSeqStore = rawSeqStore;
+        lastLLSize = blockCompressor(ms, seqStore, rep, src, srcSize);
+        ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore, srcSize);
+        return lastLLSize;
+    }
+
     assert(rawSeqStore->pos <= rawSeqStore->size);
     assert(rawSeqStore->size <= rawSeqStore->capacity);
     /* Loop through each sequence and apply the block compressor to the lits */
diff --git a/lib/compress/zstd_ldm.h b/lib/compress/zstd_ldm.h
index 229ea05a9e1e..6561024e4c97 100644
--- a/lib/compress/zstd_ldm.h
+++ b/lib/compress/zstd_ldm.h
@@ -78,6 +78,12 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
 void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize,
     U32 const minMatch);
 
+/* ZSTD_ldm_skipRawSeqStoreBytes():
+ * Moves forward in rawSeqStore by nbBytes, updating fields 'pos' and 'posInSequence'.
+ * Not to be used in conjunction with ZSTD_ldm_skipSequences().
+ * Must be called for data with is not passed to ZSTD_ldm_blockCompress().
+ */
+void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes);
 
 /** ZSTD_ldm_getTableSize() :
  *  Estimate the space needed for long distance matching tables or 0 if LDM is
diff --git a/lib/compress/zstd_opt.c b/lib/compress/zstd_opt.c
index 36fff050cf5a..e55c459debd1 100644
--- a/lib/compress/zstd_opt.c
+++ b/lib/compress/zstd_opt.c
@@ -386,32 +386,32 @@ static U32 ZSTD_insertBt1(
     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);
+    const U32 curr = (U32)(ip-base);
+    const U32 btLow = btMask >= curr ? 0 : curr - btMask;
+    U32* smallerPtr = bt + 2*(curr&btMask);
     U32* largerPtr  = smallerPtr + 1;
     U32 dummy32;   /* to be nullified at the end */
     U32 const windowLow = ms->window.lowLimit;
-    U32 matchEndIdx = current+8+1;
+    U32 matchEndIdx = curr+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);
+    U32 predictedSmall = *(bt + 2*((curr-1)&btMask) + 0);
+    U32 predictedLarge = *(bt + 2*((curr-1)&btMask) + 1);
     predictedSmall += (predictedSmall>0);
     predictedLarge += (predictedLarge>0);
 #endif /* ZSTD_C_PREDICT */
 
-    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", current);
+    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", curr);
 
     assert(ip <= iend-8);   /* required for h calculation */
-    hashTable[h] = current;   /* Update Hash Table */
+    hashTable[h] = curr;   /* Update Hash Table */
 
     assert(windowLow > 0);
     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);
+        assert(matchIndex < curr);
 
 #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 */
@@ -474,8 +474,8 @@ static U32 ZSTD_insertBt1(
     *smallerPtr = *largerPtr = 0;
     {   U32 positions = 0;
         if (bestLength > 384) positions = MIN(192, (U32)(bestLength - 384));   /* speed optimization */
-        assert(matchEndIdx > current + 8);
-        return MAX(positions, matchEndIdx - (current + 8));
+        assert(matchEndIdx > curr + 8);
+        return MAX(positions, matchEndIdx - (curr + 8));
     }
 }
 
@@ -519,7 +519,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     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 curr = (U32)(ip-base);
     U32 const hashLog = cParams->hashLog;
     U32 const minMatch = (mls==3) ? 3 : 4;
     U32* const hashTable = ms->hashTable;
@@ -533,12 +533,12 @@ U32 ZSTD_insertBtAndGetAllMatches (
     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 = ZSTD_getLowestMatchIndex(ms, current, cParams->windowLog);
+    U32 const btLow = (btMask >= curr) ? 0 : curr - btMask;
+    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
     U32 const matchLow = windowLow ? windowLow : 1;
-    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* smallerPtr = bt + 2*(curr&btMask);
+    U32* largerPtr  = bt + 2*(curr&btMask) + 1;
+    U32 matchEndIdx = curr+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;
@@ -557,7 +557,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
     U32         const dmsBtLow      = dictMode == ZSTD_dictMatchState && dmsBtMask < dmsHighLimit - dmsLowLimit ? dmsHighLimit - dmsBtMask : dmsLowLimit;
 
     size_t bestLength = lengthToBeat-1;
-    DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", current);
+    DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", curr);
 
     /* check repCode */
     assert(ll0 <= 1);   /* necessarily 1 or 0 */
@@ -565,29 +565,29 @@ U32 ZSTD_insertBtAndGetAllMatches (
         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 const repIndex = curr - repOffset;
             U32 repLen = 0;
-            assert(current >= dictLimit);
-            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < current-dictLimit) {  /* equivalent to `current > repIndex >= dictLimit` */
+            assert(curr >= dictLimit);
+            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < curr-dictLimit) {  /* equivalent to `curr > repIndex >= dictLimit` */
                 /* We must validate the repcode offset because when we're using a dictionary the
                  * valid offset range shrinks when the dictionary goes out of bounds.
                  */
                 if ((repIndex >= windowLow) & (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 */
+            } else {  /* repIndex < dictLimit || repIndex >= curr */
                 const BYTE* const repMatch = dictMode == ZSTD_dictMatchState ?
                                              dmsBase + repIndex - dmsIndexDelta :
                                              dictBase + repIndex;
-                assert(current >= windowLow);
+                assert(curr >= windowLow);
                 if ( dictMode == ZSTD_extDict
-                  && ( ((repOffset-1) /*intentional overflow*/ < current - windowLow)  /* equivalent to `current > repIndex >= windowLow` */
+                  && ( ((repOffset-1) /*intentional overflow*/ < curr - windowLow)  /* equivalent to `curr > 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;
                 }
                 if (dictMode == ZSTD_dictMatchState
-                  && ( ((repOffset-1) /*intentional overflow*/ < current - (dmsLowLimit + dmsIndexDelta))  /* equivalent to `current > repIndex >= dmsLowLimit` */
+                  && ( ((repOffset-1) /*intentional overflow*/ < curr - (dmsLowLimit + dmsIndexDelta))  /* equivalent to `curr > repIndex >= dmsLowLimit` */
                      & ((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, dmsEnd, prefixStart) + minMatch;
@@ -609,7 +609,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
     if ((mls == 3) /*static*/ && (bestLength < mls)) {
         U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, nextToUpdate3, ip);
         if ((matchIndex3 >= matchLow)
-          & (current - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
+          & (curr - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
             size_t mlen;
             if ((dictMode == ZSTD_noDict) /*static*/ || (dictMode == ZSTD_dictMatchState) /*static*/ || (matchIndex3 >= dictLimit)) {
                 const BYTE* const match = base + matchIndex3;
@@ -624,26 +624,26 @@ U32 ZSTD_insertBtAndGetAllMatches (
                 DEBUGLOG(8, "found small match with hlog3, of length %u",
                             (U32)mlen);
                 bestLength = mlen;
-                assert(current > matchIndex3);
+                assert(curr > matchIndex3);
                 assert(mnum==0);  /* no prior solution */
-                matches[0].off = (current - matchIndex3) + ZSTD_REP_MOVE;
+                matches[0].off = (curr - 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 */
+                    ms->nextToUpdate = curr+1;  /* skip insertion */
                     return 1;
         }   }   }
         /* no dictMatchState lookup: dicts don't have a populated HC3 table */
     }
 
-    hashTable[h] = current;   /* Update Hash Table */
+    hashTable[h] = curr;   /* Update Hash Table */
 
     while (nbCompares-- && (matchIndex >= matchLow)) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
         const BYTE* match;
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
-        assert(current > matchIndex);
+        assert(curr > matchIndex);
 
         if ((dictMode == ZSTD_noDict) || (dictMode == ZSTD_dictMatchState) || (matchIndex+matchLength >= dictLimit)) {
             assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
@@ -660,12 +660,12 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
         if (matchLength > bestLength) {
             DEBUGLOG(8, "found match of length %u at distance %u (offCode=%u)",
-                    (U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
+                    (U32)matchLength, curr - matchIndex, curr - matchIndex + ZSTD_REP_MOVE);
             assert(matchEndIdx > matchIndex);
             if (matchLength > matchEndIdx - matchIndex)
                 matchEndIdx = matchIndex + (U32)matchLength;
             bestLength = matchLength;
-            matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
+            matches[mnum].off = (curr - matchIndex) + ZSTD_REP_MOVE;
             matches[mnum].len = (U32)matchLength;
             mnum++;
             if ( (matchLength > ZSTD_OPT_NUM)
@@ -708,11 +708,11 @@ U32 ZSTD_insertBtAndGetAllMatches (
             if (matchLength > bestLength) {
                 matchIndex = dictMatchIndex + dmsIndexDelta;
                 DEBUGLOG(8, "found dms match of length %u at distance %u (offCode=%u)",
-                        (U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
+                        (U32)matchLength, curr - matchIndex, curr - matchIndex + ZSTD_REP_MOVE);
                 if (matchLength > matchEndIdx - matchIndex)
                     matchEndIdx = matchIndex + (U32)matchLength;
                 bestLength = matchLength;
-                matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
+                matches[mnum].off = (curr - matchIndex) + ZSTD_REP_MOVE;
                 matches[mnum].len = (U32)matchLength;
                 mnum++;
                 if ( (matchLength > ZSTD_OPT_NUM)
@@ -733,7 +733,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
         }
     }
 
-    assert(matchEndIdx > current+8);
+    assert(matchEndIdx > curr+8);
     ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
     return mnum;
 }
@@ -764,6 +764,140 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
     }
 }
 
+/*************************
+*  LDM helper functions  *
+*************************/
+
+/* Struct containing info needed to make decision about ldm inclusion */
+typedef struct {
+    rawSeqStore_t seqStore;         /* External match candidates store for this block */
+    U32 startPosInBlock;            /* Start position of the current match candidate */
+    U32 endPosInBlock;              /* End position of the current match candidate */
+    U32 offset;                     /* Offset of the match candidate */
+} ZSTD_optLdm_t;
+
+/* ZSTD_optLdm_skipRawSeqStoreBytes():
+ * Moves forward in rawSeqStore by nbBytes, which will update the fields 'pos' and 'posInSequence'.
+ */
+static void ZSTD_optLdm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
+    U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
+    while (currPos && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
+        if (currPos >= currSeq.litLength + currSeq.matchLength) {
+            currPos -= currSeq.litLength + currSeq.matchLength;
+            rawSeqStore->pos++;
+        } else {
+            rawSeqStore->posInSequence = currPos;
+            break;
+        }
+    }
+    if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
+        rawSeqStore->posInSequence = 0;
+    }
+}
+
+/* ZSTD_opt_getNextMatchAndUpdateSeqStore():
+ * Calculates the beginning and end of the next match in the current block.
+ * Updates 'pos' and 'posInSequence' of the ldmSeqStore.
+ */
+static void ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
+                                                   U32 blockBytesRemaining) {
+    rawSeq currSeq;
+    U32 currBlockEndPos;
+    U32 literalsBytesRemaining;
+    U32 matchBytesRemaining;
+
+    /* Setting match end position to MAX to ensure we never use an LDM during this block */
+    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
+        optLdm->startPosInBlock = UINT_MAX;
+        optLdm->endPosInBlock = UINT_MAX;
+        return;
+    }
+    /* Calculate appropriate bytes left in matchLength and litLength after adjusting
+       based on ldmSeqStore->posInSequence */
+    currSeq = optLdm->seqStore.seq[optLdm->seqStore.pos];
+    assert(optLdm->seqStore.posInSequence <= currSeq.litLength + currSeq.matchLength);
+    currBlockEndPos = currPosInBlock + blockBytesRemaining;
+    literalsBytesRemaining = (optLdm->seqStore.posInSequence < currSeq.litLength) ?
+            currSeq.litLength - (U32)optLdm->seqStore.posInSequence :
+            0;
+    matchBytesRemaining = (literalsBytesRemaining == 0) ?
+            currSeq.matchLength - ((U32)optLdm->seqStore.posInSequence - currSeq.litLength) :
+            currSeq.matchLength;
+
+    /* If there are more literal bytes than bytes remaining in block, no ldm is possible */
+    if (literalsBytesRemaining >= blockBytesRemaining) {
+        optLdm->startPosInBlock = UINT_MAX;
+        optLdm->endPosInBlock = UINT_MAX;
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, blockBytesRemaining);
+        return;
+    }
+
+    /* Matches may be < MINMATCH by this process. In that case, we will reject them
+       when we are deciding whether or not to add the ldm */
+    optLdm->startPosInBlock = currPosInBlock + literalsBytesRemaining;
+    optLdm->endPosInBlock = optLdm->startPosInBlock + matchBytesRemaining;
+    optLdm->offset = currSeq.offset;
+
+    if (optLdm->endPosInBlock > currBlockEndPos) {
+        /* Match ends after the block ends, we can't use the whole match */
+        optLdm->endPosInBlock = currBlockEndPos;
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, currBlockEndPos - currPosInBlock);
+    } else {
+        /* Consume nb of bytes equal to size of sequence left */
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, literalsBytesRemaining + matchBytesRemaining);
+    }
+}
+
+/* ZSTD_optLdm_maybeAddMatch():
+ * Adds a match if it's long enough, based on it's 'matchStartPosInBlock'
+ * and 'matchEndPosInBlock', into 'matches'. Maintains the correct ordering of 'matches'
+ */
+static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
+                                      ZSTD_optLdm_t* optLdm, U32 currPosInBlock) {
+    U32 posDiff = currPosInBlock - optLdm->startPosInBlock;
+    /* Note: ZSTD_match_t actually contains offCode and matchLength (before subtracting MINMATCH) */
+    U32 candidateMatchLength = optLdm->endPosInBlock - optLdm->startPosInBlock - posDiff;
+    U32 candidateOffCode = optLdm->offset + ZSTD_REP_MOVE;
+
+    /* Ensure that current block position is not outside of the match */
+    if (currPosInBlock < optLdm->startPosInBlock
+      || currPosInBlock >= optLdm->endPosInBlock
+      || candidateMatchLength < MINMATCH) {
+        return;
+    }
+
+    if (*nbMatches == 0 || ((candidateMatchLength > matches[*nbMatches-1].len) && *nbMatches < ZSTD_OPT_NUM)) {
+        DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offCode: %u matchLength %u) at block position=%u",
+                 candidateOffCode, candidateMatchLength, currPosInBlock);
+        matches[*nbMatches].len = candidateMatchLength;
+        matches[*nbMatches].off = candidateOffCode;
+        (*nbMatches)++;
+    }
+}
+
+/* ZSTD_optLdm_processMatchCandidate():
+ * Wrapper function to update ldm seq store and call ldm functions as necessary.
+ */
+static void ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm, ZSTD_match_t* matches, U32* nbMatches,
+                                              U32 currPosInBlock, U32 remainingBytes) {
+    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
+        return;
+    }
+
+    if (currPosInBlock >= optLdm->endPosInBlock) {
+        if (currPosInBlock > optLdm->endPosInBlock) {
+            /* The position at which ZSTD_optLdm_processMatchCandidate() is called is not necessarily
+             * at the end of a match from the ldm seq store, and will often be some bytes
+             * over beyond matchEndPosInBlock. As such, we need to correct for these "overshoots"
+             */
+            U32 posOvershoot = currPosInBlock - optLdm->endPosInBlock;
+            ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, posOvershoot);
+        } 
+        ZSTD_opt_getNextMatchAndUpdateSeqStore(optLdm, currPosInBlock, remainingBytes);
+    }
+    ZSTD_optLdm_maybeAddMatch(matches, nbMatches, optLdm, currPosInBlock);
+}
 
 /*-*******************************
 *  Optimal parser
@@ -817,6 +951,11 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
     ZSTD_optimal_t* const opt = optStatePtr->priceTable;
     ZSTD_match_t* const matches = optStatePtr->matchTable;
     ZSTD_optimal_t lastSequence;
+    ZSTD_optLdm_t optLdm;
+
+    optLdm.seqStore = ms->ldmSeqStore ? *ms->ldmSeqStore : kNullRawSeqStore;
+    optLdm.endPosInBlock = optLdm.startPosInBlock = optLdm.offset = 0;
+    ZSTD_opt_getNextMatchAndUpdateSeqStore(&optLdm, (U32)(ip-istart), (U32)(iend-ip));
 
     /* init */
     DEBUGLOG(5, "ZSTD_compressBlock_opt_generic: current=%u, prefix=%u, nextToUpdate=%u",
@@ -832,7 +971,9 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
         /* find first match */
         {   U32 const litlen = (U32)(ip - anchor);
             U32 const ll0 = !litlen;
-            U32 const nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, ip, iend, dictMode, rep, ll0, minMatch);
+            U32 nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, ip, iend, dictMode, rep, ll0, minMatch);
+            ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
+                                              (U32)(ip-istart), (U32)(iend - ip));
             if (!nbMatches) { ip++; continue; }
 
             /* initialize opt[0] */
@@ -925,9 +1066,9 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
             if (opt[cur].mlen != 0) {
                 U32 const prev = cur - opt[cur].mlen;
                 repcodes_t newReps = ZSTD_updateRep(opt[prev].rep, opt[cur].off, opt[cur].litlen==0);
-                memcpy(opt[cur].rep, &newReps, sizeof(repcodes_t));
+                ZSTD_memcpy(opt[cur].rep, &newReps, sizeof(repcodes_t));
             } else {
-                memcpy(opt[cur].rep, opt[cur - 1].rep, sizeof(repcodes_t));
+                ZSTD_memcpy(opt[cur].rep, opt[cur - 1].rep, sizeof(repcodes_t));
             }
 
             /* last match must start at a minimum distance of 8 from oend */
@@ -945,8 +1086,12 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                 U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
                 U32 const previousPrice = opt[cur].price;
                 U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
-                U32 const nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, inr, iend, dictMode, opt[cur].rep, ll0, minMatch);
+                U32 nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, inr, iend, dictMode, opt[cur].rep, ll0, minMatch);
                 U32 matchNb;
+
+                ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
+                                                  (U32)(inr-istart), (U32)(iend-inr));
+
                 if (!nbMatches) {
                     DEBUGLOG(7, "rPos:%u : no match found", cur);
                     continue;
@@ -1010,9 +1155,9 @@ _shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
          */
         if (lastSequence.mlen != 0) {
             repcodes_t reps = ZSTD_updateRep(opt[cur].rep, lastSequence.off, lastSequence.litlen==0);
-            memcpy(rep, &reps, sizeof(reps));
+            ZSTD_memcpy(rep, &reps, sizeof(reps));
         } else {
-            memcpy(rep, opt[cur].rep, sizeof(repcodes_t));
+            ZSTD_memcpy(rep, opt[cur].rep, sizeof(repcodes_t));
         }
 
         {   U32 const storeEnd = cur + 1;
@@ -1110,7 +1255,7 @@ ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
                const void* src, size_t srcSize)
 {
     U32 tmpRep[ZSTD_REP_NUM];  /* updated rep codes will sink here */
-    memcpy(tmpRep, rep, sizeof(tmpRep));
+    ZSTD_memcpy(tmpRep, rep, sizeof(tmpRep));
 
     DEBUGLOG(4, "ZSTD_initStats_ultra (srcSize=%zu)", srcSize);
     assert(ms->opt.litLengthSum == 0);    /* first block */
@@ -1143,7 +1288,7 @@ size_t ZSTD_compressBlock_btultra2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
 {
-    U32 const current = (U32)((const BYTE*)src - ms->window.base);
+    U32 const curr = (U32)((const BYTE*)src - ms->window.base);
     DEBUGLOG(5, "ZSTD_compressBlock_btultra2 (srcSize=%zu)", srcSize);
 
     /* 2-pass strategy:
@@ -1158,7 +1303,7 @@ size_t ZSTD_compressBlock_btultra2(
     if ( (ms->opt.litLengthSum==0)   /* first block */
       && (seqStore->sequences == seqStore->sequencesStart)  /* no ldm */
       && (ms->window.dictLimit == ms->window.lowLimit)   /* no dictionary */
-      && (current == ms->window.dictLimit)   /* start of frame, nothing already loaded nor skipped */
+      && (curr == ms->window.dictLimit)   /* start of frame, nothing already loaded nor skipped */
       && (srcSize > ZSTD_PREDEF_THRESHOLD)
       ) {
         ZSTD_initStats_ultra(ms, seqStore, rep, src, srcSize);
diff --git a/lib/compress/zstdmt_compress.c b/lib/compress/zstdmt_compress.c
index 1e3c8fdbee28..50454a50b9b7 100644
--- a/lib/compress/zstdmt_compress.c
+++ b/lib/compress/zstdmt_compress.c
@@ -20,8 +20,7 @@
 
 
 /* ======   Dependencies   ====== */
-#include <string.h>      /* memcpy, memset */
-#include <limits.h>      /* INT_MAX, UINT_MAX */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memset, INT_MAX, UINT_MAX */
 #include "../common/mem.h"         /* MEM_STATIC */
 #include "../common/pool.h"        /* threadpool */
 #include "../common/threading.h"   /* mutex */
@@ -106,11 +105,11 @@ typedef struct ZSTDMT_bufferPool_s {
 static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbWorkers, ZSTD_customMem cMem)
 {
     unsigned const maxNbBuffers = 2*nbWorkers + 3;
-    ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_calloc(
+    ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_customCalloc(
         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);
+        ZSTD_customFree(bufPool, cMem);
         return NULL;
     }
     bufPool->bufferSize = 64 KB;
@@ -127,10 +126,10 @@ static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* 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_customFree(bufPool->bTable[u].start, bufPool->cMem);
     }
     ZSTD_pthread_mutex_destroy(&bufPool->poolMutex);
-    ZSTD_free(bufPool, bufPool->cMem);
+    ZSTD_customFree(bufPool, bufPool->cMem);
 }
 
 /* only works at initialization, not during compression */
@@ -201,13 +200,13 @@ static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
         }
         /* 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_customFree(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 = ZSTD_malloc(bSize, bufPool->cMem);
+        void* const start = ZSTD_customMalloc(bSize, bufPool->cMem);
         buffer.start = start;   /* note : start can be NULL if malloc fails ! */
         buffer.capacity = (start==NULL) ? 0 : bSize;
         if (start==NULL) {
@@ -229,13 +228,13 @@ static buffer_t ZSTDMT_resizeBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buffer)
 {
     size_t const bSize = bufPool->bufferSize;
     if (buffer.capacity < bSize) {
-        void* const start = ZSTD_malloc(bSize, bufPool->cMem);
+        void* const start = ZSTD_customMalloc(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);
+            ZSTD_memcpy(newBuffer.start, buffer.start, buffer.capacity);
             DEBUGLOG(5, "ZSTDMT_resizeBuffer: created buffer of size %u", (U32)bSize);
             return newBuffer;
         }
@@ -261,14 +260,12 @@ static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
     ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
     /* Reached bufferPool capacity (should not happen) */
     DEBUGLOG(5, "ZSTDMT_releaseBuffer: pool capacity reached => freeing ");
-    ZSTD_free(buf.start, bufPool->cMem);
+    ZSTD_customFree(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)
@@ -278,7 +275,7 @@ static size_t ZSTDMT_sizeof_seqPool(ZSTDMT_seqPool* seqPool)
 
 static rawSeqStore_t bufferToSeq(buffer_t buffer)
 {
-    rawSeqStore_t seq = {NULL, 0, 0, 0};
+    rawSeqStore_t seq = kNullRawSeqStore;
     seq.seq = (rawSeq*)buffer.start;
     seq.capacity = buffer.capacity / sizeof(rawSeq);
     return seq;
@@ -354,7 +351,7 @@ static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
     for (cid=0; cid<pool->totalCCtx; cid++)
         ZSTD_freeCCtx(pool->cctx[cid]);  /* note : compatible with free on NULL */
     ZSTD_pthread_mutex_destroy(&pool->poolMutex);
-    ZSTD_free(pool, pool->cMem);
+    ZSTD_customFree(pool, pool->cMem);
 }
 
 /* ZSTDMT_createCCtxPool() :
@@ -362,12 +359,12 @@ static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
 static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(int nbWorkers,
                                               ZSTD_customMem cMem)
 {
-    ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_calloc(
+    ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_customCalloc(
         sizeof(ZSTDMT_CCtxPool) + (nbWorkers-1)*sizeof(ZSTD_CCtx*), cMem);
     assert(nbWorkers > 0);
     if (!cctxPool) return NULL;
     if (ZSTD_pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
-        ZSTD_free(cctxPool, cMem);
+        ZSTD_customFree(cctxPool, cMem);
         return NULL;
     }
     cctxPool->cMem = cMem;
@@ -478,7 +475,7 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
         serialState->ldmState.hashPower =
                 ZSTD_rollingHash_primePower(params.ldmParams.minMatchLength);
     } else {
-        memset(&params.ldmParams, 0, sizeof(params.ldmParams));
+        ZSTD_memset(&params.ldmParams, 0, sizeof(params.ldmParams));
     }
     serialState->nextJobID = 0;
     if (params.fParams.checksumFlag)
@@ -499,18 +496,18 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
         ZSTD_window_init(&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);
+            ZSTD_customFree(serialState->ldmState.hashTable, cMem);
+            serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_customMalloc(hashSize, cMem);
         }
         if (serialState->ldmState.bucketOffsets == NULL || prevBucketLog < bucketLog) {
-            ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
-            serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_malloc(bucketSize, cMem);
+            ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
+            serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_customMalloc(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);
+        ZSTD_memset(serialState->ldmState.hashTable, 0, hashSize);
+        ZSTD_memset(serialState->ldmState.bucketOffsets, 0, bucketSize);
 
         /* Update window state and fill hash table with dict */
         serialState->ldmState.loadedDictEnd = 0;
@@ -537,7 +534,7 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
 static int ZSTDMT_serialState_init(serialState_t* serialState)
 {
     int initError = 0;
-    memset(serialState, 0, sizeof(*serialState));
+    ZSTD_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);
@@ -552,8 +549,8 @@ static void ZSTDMT_serialState_free(serialState_t* serialState)
     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);
+    ZSTD_customFree(serialState->ldmState.hashTable, cMem);
+    ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
 }
 
 static void ZSTDMT_serialState_update(serialState_t* serialState,
@@ -820,7 +817,6 @@ struct ZSTDMT_CCtx_s {
     roundBuff_t roundBuff;
     serialState_t serial;
     rsyncState_t rsync;
-    unsigned singleBlockingThread;
     unsigned jobIDMask;
     unsigned doneJobID;
     unsigned nextJobID;
@@ -832,6 +828,7 @@ struct ZSTDMT_CCtx_s {
     ZSTD_customMem cMem;
     ZSTD_CDict* cdictLocal;
     const ZSTD_CDict* cdict;
+    unsigned providedFactory: 1;
 };
 
 static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZSTD_customMem cMem)
@@ -842,7 +839,7 @@ static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZS
         ZSTD_pthread_mutex_destroy(&jobTable[jobNb].job_mutex);
         ZSTD_pthread_cond_destroy(&jobTable[jobNb].job_cond);
     }
-    ZSTD_free(jobTable, cMem);
+    ZSTD_customFree(jobTable, cMem);
 }
 
 /* ZSTDMT_allocJobsTable()
@@ -854,7 +851,7 @@ static ZSTDMT_jobDescription* ZSTDMT_createJobsTable(U32* nbJobsPtr, ZSTD_custom
     U32 const nbJobs = 1 << nbJobsLog2;
     U32 jobNb;
     ZSTDMT_jobDescription* const jobTable = (ZSTDMT_jobDescription*)
-                ZSTD_calloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
+                ZSTD_customCalloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
     int initError = 0;
     if (jobTable==NULL) return NULL;
     *nbJobsPtr = nbJobs;
@@ -885,12 +882,12 @@ static size_t ZSTDMT_expandJobsTable (ZSTDMT_CCtx* mtctx, U32 nbWorkers) {
 
 /* ZSTDMT_CCtxParam_setNbWorkers():
  * Internal use only */
-size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
+static size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
 {
     return ZSTD_CCtxParams_setParameter(params, ZSTD_c_nbWorkers, (int)nbWorkers);
 }
 
-MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers, ZSTD_customMem cMem)
+MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
 {
     ZSTDMT_CCtx* mtctx;
     U32 nbJobs = nbWorkers + 2;
@@ -903,12 +900,19 @@ MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers,
         /* invalid custom allocator */
         return NULL;
 
-    mtctx = (ZSTDMT_CCtx*) ZSTD_calloc(sizeof(ZSTDMT_CCtx), cMem);
+    mtctx = (ZSTDMT_CCtx*) ZSTD_customCalloc(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);
+    if (pool != NULL) {
+      mtctx->factory = pool;
+      mtctx->providedFactory = 1;
+    }
+    else {
+      mtctx->factory = POOL_create_advanced(nbWorkers, 0, cMem);
+      mtctx->providedFactory = 0;
+    }
     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;
@@ -925,22 +929,18 @@ MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers,
     return mtctx;
 }
 
-ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem)
+ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
 {
 #ifdef ZSTD_MULTITHREAD
-    return ZSTDMT_createCCtx_advanced_internal(nbWorkers, cMem);
+    return ZSTDMT_createCCtx_advanced_internal(nbWorkers, cMem, pool);
 #else
     (void)nbWorkers;
     (void)cMem;
+    (void)pool;
     return NULL;
 #endif
 }
 
-ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbWorkers)
-{
-    return ZSTDMT_createCCtx_advanced(nbWorkers, ZSTD_defaultCMem);
-}
-
 
 /* ZSTDMT_releaseAllJobResources() :
  * note : ensure all workers are killed first ! */
@@ -957,7 +957,7 @@ static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
         ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
 
         /* Clear the job description, but keep the mutex/cond */
-        memset(&mtctx->jobs[jobID], 0, sizeof(mtctx->jobs[jobID]));
+        ZSTD_memset(&mtctx->jobs[jobID], 0, sizeof(mtctx->jobs[jobID]));
         mtctx->jobs[jobID].job_mutex = mutex;
         mtctx->jobs[jobID].job_cond = cond;
     }
@@ -984,7 +984,8 @@ static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* mtctx)
 size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
 {
     if (mtctx==NULL) return 0;   /* compatible with free on NULL */
-    POOL_free(mtctx->factory);   /* stop and free worker threads */
+    if (!mtctx->providedFactory)
+        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);
@@ -993,8 +994,8 @@ size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
     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);
+        ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
+    ZSTD_customFree(mtctx, mtctx->cMem);
     return 0;
 }
 
@@ -1011,65 +1012,6 @@ size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
             + mtctx->roundBuff.capacity;
 }
 
-/* Internal only */
-size_t
-ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params,
-                                   ZSTDMT_parameter parameter,
-                                   int value)
-{
-    DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter");
-    switch(parameter)
-    {
-    case ZSTDMT_p_jobSize :
-        DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter : set jobSize to %i", value);
-        return ZSTD_CCtxParams_setParameter(params, ZSTD_c_jobSize, value);
-    case ZSTDMT_p_overlapLog :
-        DEBUGLOG(4, "ZSTDMT_p_overlapLog : %i", value);
-        return ZSTD_CCtxParams_setParameter(params, ZSTD_c_overlapLog, value);
-    case ZSTDMT_p_rsyncable :
-        DEBUGLOG(4, "ZSTD_p_rsyncable : %i", value);
-        return ZSTD_CCtxParams_setParameter(params, ZSTD_c_rsyncable, value);
-    default :
-        return ERROR(parameter_unsupported);
-    }
-}
-
-size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int value)
-{
-    DEBUGLOG(4, "ZSTDMT_setMTCtxParameter");
-    return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
-}
-
-size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int* value)
-{
-    switch (parameter) {
-    case ZSTDMT_p_jobSize:
-        return ZSTD_CCtxParams_getParameter(&mtctx->params, ZSTD_c_jobSize, value);
-    case ZSTDMT_p_overlapLog:
-        return ZSTD_CCtxParams_getParameter(&mtctx->params, ZSTD_c_overlapLog, value);
-    case ZSTDMT_p_rsyncable:
-        return ZSTD_CCtxParams_getParameter(&mtctx->params, ZSTD_c_rsyncable, value);
-    default:
-        return ERROR(parameter_unsupported);
-    }
-}
-
-/* Sets parameters relevant to the compression job,
- * initializing others to default values. */
-static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(const ZSTD_CCtx_params* params)
-{
-    ZSTD_CCtx_params jobParams = *params;
-    /* Clear parameters related to multithreading */
-    jobParams.forceWindow = 0;
-    jobParams.nbWorkers = 0;
-    jobParams.jobSize = 0;
-    jobParams.overlapLog = 0;
-    jobParams.rsyncable = 0;
-    memset(&jobParams.ldmParams, 0, sizeof(ldmParams_t));
-    memset(&jobParams.customMem, 0, sizeof(ZSTD_customMem));
-    return jobParams;
-}
-
 
 /* ZSTDMT_resize() :
  * @return : error code if fails, 0 on success */
@@ -1098,7 +1040,7 @@ void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_p
     DEBUGLOG(5, "ZSTDMT_updateCParams_whileCompressing (level:%i)",
                 compressionLevel);
     mtctx->params.compressionLevel = compressionLevel;
-    {   ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, 0);
+    {   ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
         cParams.windowLog = saved_wlog;
         mtctx->params.cParams = cParams;
     }
@@ -1185,8 +1127,8 @@ static unsigned ZSTDMT_computeTargetJobLog(const ZSTD_CCtx_params* params)
     if (params->ldmParams.enableLdm) {
         /* In Long Range Mode, the windowLog is typically oversized.
          * In which case, it's preferable to determine the jobSize
-         * based on chainLog instead. */
-        jobLog = MAX(21, params->cParams.chainLog + 4);
+         * based on cycleLog instead. */
+        jobLog = MAX(21, ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy) + 3);
     } else {
         jobLog = MAX(20, params->cParams.windowLog + 2);
     }
@@ -1240,174 +1182,6 @@ static size_t ZSTDMT_computeOverlapSize(const ZSTD_CCtx_params* params)
     return (ovLog==0) ? 0 : (size_t)1 << ovLog;
 }
 
-static unsigned
-ZSTDMT_computeNbJobs(const 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 = ZSTDMT_computeOverlapSize(&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) );
-    /* LDM doesn't even try to load the dictionary in single-ingestion mode */
-    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params, avgJobSize, NULL, 0, ZSTD_dct_auto))
-        return ERROR(memory_allocation);
-
-    FORWARD_IF_ERROR( ZSTDMT_expandJobsTable(mtctx, nbJobs) , "");  /* only expands if necessary */
-
-    {   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,
-                                int overlapLog)
-{
-    ZSTD_CCtx_params cctxParams = mtctx->params;
-    cctxParams.cParams = params.cParams;
-    cctxParams.fParams = params.fParams;
-    assert(ZSTD_OVERLAPLOG_MIN <= overlapLog && overlapLog <= ZSTD_OVERLAPLOG_MAX);
-    cctxParams.overlapLog = 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)
-{
-    ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);
-    int const overlapLog = ZSTDMT_overlapLog_default(params.cParams.strategy);
-    params.fParams.contentSizeFlag = 1;
-    return ZSTDMT_compress_advanced(mtctx, dst, dstCapacity, src, srcSize, NULL, params, overlapLog);
-}
-
-
 /* ====================================== */
 /* =======      Streaming API     ======= */
 /* ====================================== */
@@ -1432,16 +1206,6 @@ size_t ZSTDMT_initCStream_internal(
     if (params.jobSize != 0 && params.jobSize < ZSTDMT_JOBSIZE_MIN) params.jobSize = ZSTDMT_JOBSIZE_MIN;
     if (params.jobSize > (size_t)ZSTDMT_JOBSIZE_MAX) params.jobSize = (size_t)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 */
@@ -1504,8 +1268,8 @@ size_t ZSTDMT_initCStream_internal(
         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);
+                ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
+            mtctx->roundBuff.buffer = (BYTE*)ZSTD_customMalloc(capacity, mtctx->cMem);
             if (mtctx->roundBuff.buffer == NULL) {
                 mtctx->roundBuff.capacity = 0;
                 return ERROR(memory_allocation);
@@ -1530,53 +1294,6 @@ size_t ZSTDMT_initCStream_internal(
     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.
@@ -1740,7 +1457,7 @@ static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, u
             assert(cSize >= mtctx->jobs[wJobID].dstFlushed);
             assert(mtctx->jobs[wJobID].dstBuff.start != NULL);
             if (toFlush > 0) {
-                memcpy((char*)output->dst + output->pos,
+                ZSTD_memcpy((char*)output->dst + output->pos,
                     (const char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].dstFlushed,
                     toFlush);
             }
@@ -1894,7 +1611,7 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
             return 0;
         }
         ZSTDMT_waitForLdmComplete(mtctx, buffer);
-        memmove(start, mtctx->inBuff.prefix.start, prefixSize);
+        ZSTD_memmove(start, mtctx->inBuff.prefix.start, prefixSize);
         mtctx->inBuff.prefix.start = start;
         mtctx->roundBuff.pos = prefixSize;
     }
@@ -1968,6 +1685,16 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
         pos = 0;
         prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
         hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
+        if ((hash & hitMask) == hitMask) {
+            /* We're already at a sync point so don't load any more until
+             * we're able to flush this sync point.
+             * This likely happened because the job table was full so we
+             * couldn't add our job.
+             */
+            syncPoint.toLoad = 0;
+            syncPoint.flush = 1;
+            return syncPoint;
+        }
     } else {
         /* We don't have enough bytes buffered to initialize the hash, but
          * we know we have at least RSYNC_LENGTH bytes total.
@@ -2022,34 +1749,11 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
     assert(output->pos <= output->size);
     assert(input->pos  <= input->size);
 
-    if (mtctx->singleBlockingThread) {  /* delegate to single-thread (synchronous) */
-        return ZSTD_compressStream2(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->params.rsyncable)   /* rsyncable mode is disabled */
-      && (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 */
@@ -2072,13 +1776,21 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
             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)syncPoint.toLoad, (U32)mtctx->inBuff.filled, (U32)mtctx->targetSectionSize);
-            memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, syncPoint.toLoad);
+            ZSTD_memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, syncPoint.toLoad);
             input->pos += syncPoint.toLoad;
             mtctx->inBuff.filled += syncPoint.toLoad;
             forwardInputProgress = syncPoint.toLoad>0;
         }
-        if ((input->pos < input->size) && (endOp == ZSTD_e_end))
-            endOp = ZSTD_e_flush;   /* can't end now : not all input consumed */
+    }
+    if ((input->pos < input->size) && (endOp == ZSTD_e_end)) {
+        /* Can't end yet because the input is not fully consumed.
+            * We are in one of these cases:
+            * - mtctx->inBuff is NULL & empty: we couldn't get an input buffer so don't create a new job.
+            * - We filled the input buffer: flush this job but don't end the frame.
+            * - We hit a synchronization point: flush this job but don't end the frame.
+            */
+        assert(mtctx->inBuff.filled == 0 || mtctx->inBuff.filled == mtctx->targetSectionSize || mtctx->params.rsyncable);
+        endOp = ZSTD_e_flush;
     }
 
     if ( (mtctx->jobReady)
@@ -2097,47 +1809,3 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
         return remainingToFlush;
     }
 }
-
-
-size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
-{
-    FORWARD_IF_ERROR( ZSTDMT_compressStream_generic(mtctx, output, input, ZSTD_e_continue) , "");
-
-    /* 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_flushStream_internal(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_EndDirective endFrame)
-{
-    size_t const srcSize = mtctx->inBuff.filled;
-    DEBUGLOG(5, "ZSTDMT_flushStream_internal");
-
-    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);
-        FORWARD_IF_ERROR( ZSTDMT_createCompressionJob(mtctx, srcSize, endFrame) , "");
-    }
-
-    /* check if there is any data available to flush */
-    return ZSTDMT_flushProduced(mtctx, output, 1 /* blockToFlush */, endFrame);
-}
-
-
-size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
-{
-    DEBUGLOG(5, "ZSTDMT_flushStream");
-    if (mtctx->singleBlockingThread)
-        return ZSTD_flushStream(mtctx->cctxPool->cctx[0], output);
-    return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_flush);
-}
-
-size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
-{
-    DEBUGLOG(4, "ZSTDMT_endStream");
-    if (mtctx->singleBlockingThread)
-        return ZSTD_endStream(mtctx->cctxPool->cctx[0], output);
-    return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_end);
-}
diff --git a/lib/compress/zstdmt_compress.h b/lib/compress/zstdmt_compress.h
index 89914eb7f883..0a9e551c99b6 100644
--- a/lib/compress/zstdmt_compress.h
+++ b/lib/compress/zstdmt_compress.h
@@ -19,26 +19,14 @@
 /* Note : This is an internal API.
  *        These APIs used to be exposed with ZSTDLIB_API,
  *        because it used to be the only way to invoke MT compression.
- *        Now, it's recommended to use ZSTD_compress2 and ZSTD_compressStream2()
- *        instead.
- *
- *        If you depend on these APIs and can't switch, then define
- *        ZSTD_LEGACY_MULTITHREADED_API when making the dynamic library.
- *        However, we may completely remove these functions in a future
- *        release, so please switch soon.
+ *        Now, you must use ZSTD_compress2 and ZSTD_compressStream2() instead.
  *
  *        This API requires ZSTD_MULTITHREAD to be defined during compilation,
  *        otherwise ZSTDMT_createCCtx*() will fail.
  */
 
-#ifdef ZSTD_LEGACY_MULTITHREADED_API
-#  define ZSTDMT_API ZSTDLIB_API
-#else
-#  define ZSTDMT_API
-#endif
-
 /* ===   Dependencies   === */
-#include <stddef.h>                /* size_t */
+#include "../common/zstd_deps.h"   /* size_t */
 #define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_parameters */
 #include "../zstd.h"            /* ZSTD_inBuffer, ZSTD_outBuffer, ZSTDLIB_API */
 
@@ -54,78 +42,34 @@
 #define ZSTDMT_JOBSIZE_MAX  (MEM_32bits() ? (512 MB) : (1024 MB))
 
 
+/* ========================================================
+ * ===  Private interface, for use by ZSTD_compress.c   ===
+ * ===  Not exposed in libzstd. Never invoke directly   ===
+ * ======================================================== */
+
 /* ===   Memory management   === */
 typedef struct ZSTDMT_CCtx_s ZSTDMT_CCtx;
 /* Requires ZSTD_MULTITHREAD to be defined during compilation, otherwise it will return NULL. */
-ZSTDMT_API ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbWorkers);
-/* Requires ZSTD_MULTITHREAD to be defined during compilation, otherwise it will return NULL. */
-ZSTDMT_API ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers,
-                                                    ZSTD_customMem cMem);
-ZSTDMT_API size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx);
-
-ZSTDMT_API size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx);
-
-
-/* ===   Simple one-pass compression function   === */
-
-ZSTDMT_API size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
-                                       void* dst, size_t dstCapacity,
-                                 const void* src, size_t srcSize,
-                                       int compressionLevel);
-
+ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers,
+                                        ZSTD_customMem cMem,
+					ZSTD_threadPool *pool);
+size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx);
 
+size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx);
 
 /* ===   Streaming functions   === */
 
-ZSTDMT_API size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel);
-ZSTDMT_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" */
-
-ZSTDMT_API size_t ZSTDMT_nextInputSizeHint(const ZSTDMT_CCtx* mtctx);
-ZSTDMT_API size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
-
-ZSTDMT_API size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output);   /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
-ZSTDMT_API size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output);     /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
-
-
-/* ===   Advanced functions and parameters  === */
-
-ZSTDMT_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,
-                                          int overlapLog);
-
-ZSTDMT_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 */
-
-ZSTDMT_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_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_overlapLog,  /* Each job may reload a part of previous job to enhance compression 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_p_rsyncable    /* Enables rsyncable mode. */
-} 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() 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()) */
-ZSTDMT_API size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int value);
-
-/* ZSTDMT_getMTCtxParameter() :
- * Query the ZSTDMT_CCtx for a parameter value.
- * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
-ZSTDMT_API size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int* value);
+size_t ZSTDMT_nextInputSizeHint(const 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);
 
 /*! ZSTDMT_compressStream_generic() :
  *  Combines ZSTDMT_compressStream() with optional ZSTDMT_flushStream() or ZSTDMT_endStream()
@@ -134,16 +78,10 @@ ZSTDMT_API size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter
  *           0 if fully flushed
  *           or an error code
  *  note : needs to be init using any ZSTD_initCStream*() variant */
-ZSTDMT_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_compressStream_generic(ZSTDMT_CCtx* mtctx,
+                                     ZSTD_outBuffer* output,
+                                     ZSTD_inBuffer* input,
+                                     ZSTD_EndDirective endOp);
 
  /*! ZSTDMT_toFlushNow()
   *  Tell how many bytes are ready to be flushed immediately.
@@ -153,15 +91,6 @@ ZSTDMT_API size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
   *  therefore flushing is limited by speed of oldest job. */
 size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx);
 
-/*! ZSTDMT_CCtxParam_setMTCtxParameter()
- *  like ZSTDMT_setMTCtxParameter(), but into a ZSTD_CCtx_Params */
-size_t ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params, ZSTDMT_parameter parameter, int 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. */
@@ -174,17 +103,6 @@ void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_p
 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)
 }
 #endif
diff --git a/lib/decompress/huf_decompress.c b/lib/decompress/huf_decompress.c
index 68293a13096a..141820671852 100644
--- a/lib/decompress/huf_decompress.c
+++ b/lib/decompress/huf_decompress.c
@@ -15,7 +15,7 @@
 /* **************************************************************
 *  Dependencies
 ****************************************************************/
-#include <string.h>     /* memcpy, memset */
+#include "../common/zstd_deps.h"  /* ZSTD_memcpy, ZSTD_memset */
 #include "../common/compiler.h"
 #include "../common/bitstream.h"  /* BIT_* */
 #include "../common/fse.h"        /* to compress headers */
@@ -103,7 +103,7 @@ typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved;
 static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
 {
     DTableDesc dtd;
-    memcpy(&dtd, table, sizeof(dtd));
+    ZSTD_memcpy(&dtd, table, sizeof(dtd));
     return dtd;
 }
 
@@ -115,29 +115,51 @@ static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
 /*-***************************/
 typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX1;   /* single-symbol decoding */
 
+/**
+ * Packs 4 HUF_DEltX1 structs into a U64. This is used to lay down 4 entries at
+ * a time.
+ */
+static U64 HUF_DEltX1_set4(BYTE symbol, BYTE nbBits) {
+    U64 D4;
+    if (MEM_isLittleEndian()) {
+        D4 = symbol + (nbBits << 8);
+    } else {
+        D4 = (symbol << 8) + nbBits;
+    }
+    D4 *= 0x0001000100010001ULL;
+    return D4;
+}
+
+typedef struct {
+        U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];
+        U32 rankStart[HUF_TABLELOG_ABSOLUTEMAX + 1];
+        U32 statsWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
+        BYTE symbols[HUF_SYMBOLVALUE_MAX + 1];
+        BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];
+} HUF_ReadDTableX1_Workspace;
+
+
 size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
 {
+    return HUF_readDTableX1_wksp_bmi2(DTable, src, srcSize, workSpace, wkspSize, /* bmi2 */ 0);
+}
+
+size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2)
+{
     U32 tableLog = 0;
     U32 nbSymbols = 0;
     size_t iSize;
     void* const dtPtr = DTable + 1;
     HUF_DEltX1* const dt = (HUF_DEltX1*)dtPtr;
+    HUF_ReadDTableX1_Workspace* wksp = (HUF_ReadDTableX1_Workspace*)workSpace;
 
-    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);
+    DEBUG_STATIC_ASSERT(HUF_DECOMPRESS_WORKSPACE_SIZE >= sizeof(*wksp));
+    if (sizeof(*wksp) > wkspSize) return ERROR(tableLog_tooLarge);
 
     DEBUG_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
-    /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+    /* ZSTD_memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
 
-    iSize = HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp), bmi2);
     if (HUF_isError(iSize)) return iSize;
 
     /* Table header */
@@ -145,52 +167,117 @@ size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize
         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));
+        ZSTD_memcpy(DTable, &dtd, sizeof(dtd));
     }
 
-    /* Calculate starting value for each rank */
-    {   U32 n, nextRankStart = 0;
-        for (n=1; n<tableLog+1; n++) {
-            U32 const current = nextRankStart;
-            nextRankStart += (rankVal[n] << (n-1));
-            rankVal[n] = current;
-    }   }
+    /* Compute symbols and rankStart given rankVal:
+     *
+     * rankVal already contains the number of values of each weight.
+     *
+     * symbols contains the symbols ordered by weight. First are the rankVal[0]
+     * weight 0 symbols, followed by the rankVal[1] weight 1 symbols, and so on.
+     * symbols[0] is filled (but unused) to avoid a branch.
+     *
+     * rankStart contains the offset where each rank belongs in the DTable.
+     * rankStart[0] is not filled because there are no entries in the table for
+     * weight 0.
+     */
+    {
+        int n;
+        int nextRankStart = 0;
+        int const unroll = 4;
+        int const nLimit = (int)nbSymbols - unroll + 1;
+        for (n=0; n<(int)tableLog+1; n++) {
+            U32 const curr = nextRankStart;
+            nextRankStart += wksp->rankVal[n];
+            wksp->rankStart[n] = curr;
+        }
+        for (n=0; n < nLimit; n += unroll) {
+            int u;
+            for (u=0; u < unroll; ++u) {
+                size_t const w = wksp->huffWeight[n+u];
+                wksp->symbols[wksp->rankStart[w]++] = (BYTE)(n+u);
+            }
+        }
+        for (; n < (int)nbSymbols; ++n) {
+            size_t const w = wksp->huffWeight[n];
+            wksp->symbols[wksp->rankStart[w]++] = (BYTE)n;
+        }
+    }
 
-    /* fill DTable */
-    {   U32 n;
-        size_t const nEnd = nbSymbols;
-        for (n=0; n<nEnd; n++) {
-            size_t const w = huffWeight[n];
-            size_t const length = (1 << w) >> 1;
-            size_t const uStart = rankVal[w];
-            size_t const uEnd = uStart + length;
-            size_t u;
-            HUF_DEltX1 D;
-            D.byte = (BYTE)n;
-            D.nbBits = (BYTE)(tableLog + 1 - w);
-            rankVal[w] = (U32)uEnd;
-            if (length < 4) {
-                /* Use length in the loop bound so the compiler knows it is short. */
-                for (u = 0; u < length; ++u)
-                    dt[uStart + u] = D;
-            } else {
-                /* Unroll the loop 4 times, we know it is a power of 2. */
-                for (u = uStart; u < uEnd; u += 4) {
-                    dt[u + 0] = D;
-                    dt[u + 1] = D;
-                    dt[u + 2] = D;
-                    dt[u + 3] = D;
-    }   }   }   }
+    /* fill DTable
+     * We fill all entries of each weight in order.
+     * That way length is a constant for each iteration of the outter loop.
+     * We can switch based on the length to a different inner loop which is
+     * optimized for that particular case.
+     */
+    {
+        U32 w;
+        int symbol=wksp->rankVal[0];
+        int rankStart=0;
+        for (w=1; w<tableLog+1; ++w) {
+            int const symbolCount = wksp->rankVal[w];
+            int const length = (1 << w) >> 1;
+            int uStart = rankStart;
+            BYTE const nbBits = (BYTE)(tableLog + 1 - w);
+            int s;
+            int u;
+            switch (length) {
+            case 1:
+                for (s=0; s<symbolCount; ++s) {
+                    HUF_DEltX1 D;
+                    D.byte = wksp->symbols[symbol + s];
+                    D.nbBits = nbBits;
+                    dt[uStart] = D;
+                    uStart += 1;
+                }
+                break;
+            case 2:
+                for (s=0; s<symbolCount; ++s) {
+                    HUF_DEltX1 D;
+                    D.byte = wksp->symbols[symbol + s];
+                    D.nbBits = nbBits;
+                    dt[uStart+0] = D;
+                    dt[uStart+1] = D;
+                    uStart += 2;
+                }
+                break;
+            case 4:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    MEM_write64(dt + uStart, D4);
+                    uStart += 4;
+                }
+                break;
+            case 8:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    MEM_write64(dt + uStart, D4);
+                    MEM_write64(dt + uStart + 4, D4);
+                    uStart += 8;
+                }
+                break;
+            default:
+                for (s=0; s<symbolCount; ++s) {
+                    U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
+                    for (u=0; u < length; u += 16) {
+                        MEM_write64(dt + uStart + u + 0, D4);
+                        MEM_write64(dt + uStart + u + 4, D4);
+                        MEM_write64(dt + uStart + u + 8, D4);
+                        MEM_write64(dt + uStart + u + 12, D4);
+                    }
+                    assert(u == length);
+                    uStart += length;
+                }
+                break;
+            }
+            symbol += symbolCount;
+            rankStart += symbolCount * length;
+        }
+    }
     return iSize;
 }
 
-size_t HUF_readDTableX1(HUF_DTable* DTable, const void* src, size_t srcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_readDTableX1_wksp(DTable, src, srcSize,
-                                 workSpace, sizeof(workSpace));
-}
-
 FORCE_INLINE_TEMPLATE BYTE
 HUF_decodeSymbolX1(BIT_DStream_t* Dstream, const HUF_DEltX1* dt, const U32 dtLog)
 {
@@ -389,20 +476,6 @@ size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
 }
 
 
-size_t HUF_decompress1X1_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
-                              const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress1X1_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
-                                       workSpace, sizeof(workSpace));
-}
-
-size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress1X1_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
-}
-
 size_t HUF_decompress4X1_usingDTable(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
@@ -419,8 +492,7 @@ static size_t HUF_decompress4X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX1_wksp (dctx, cSrc, cSrcSize,
-                                                workSpace, wkspSize);
+    size_t const hSize = HUF_readDTableX1_wksp_bmi2(dctx, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
@@ -436,18 +508,6 @@ size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
 }
 
 
-size_t HUF_decompress4X1_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
-                                       workSpace, sizeof(workSpace));
-}
-size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress4X1_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
-}
-
 #endif /* HUF_FORCE_DECOMPRESS_X2 */
 
 
@@ -474,7 +534,7 @@ static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 sizeLog, const U32 co
     U32 rankVal[HUF_TABLELOG_MAX + 1];
 
     /* get pre-calculated rankVal */
-    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+    ZSTD_memcpy(rankVal, rankValOrigin, sizeof(rankVal));
 
     /* fill skipped values */
     if (minWeight>1) {
@@ -516,7 +576,7 @@ static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
     const U32 minBits  = nbBitsBaseline - maxWeight;
     U32 s;
 
-    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+    ZSTD_memcpy(rankVal, rankValOrigin, sizeof(rankVal));
 
     /* fill DTable */
     for (s=0; s<sortedListSize; s++) {
@@ -581,11 +641,11 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
     if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
 
     rankStart = rankStart0 + 1;
-    memset(rankStats, 0, sizeof(U32) * (2 * HUF_TABLELOG_MAX + 2 + 1));
+    ZSTD_memset(rankStats, 0, sizeof(U32) * (2 * HUF_TABLELOG_MAX + 2 + 1));
 
     DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == 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 ... */
+    /* ZSTD_memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
 
     iSize = HUF_readStats(weightList, HUF_SYMBOLVALUE_MAX + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
     if (HUF_isError(iSize)) return iSize;
@@ -599,9 +659,9 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
     /* Get start index of each weight */
     {   U32 w, nextRankStart = 0;
         for (w=1; w<maxW+1; w++) {
-            U32 current = nextRankStart;
+            U32 curr = nextRankStart;
             nextRankStart += rankStats[w];
-            rankStart[w] = current;
+            rankStart[w] = curr;
         }
         rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
         sizeOfSort = nextRankStart;
@@ -624,9 +684,9 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
             U32 nextRankVal = 0;
             U32 w;
             for (w=1; w<maxW+1; w++) {
-                U32 current = nextRankVal;
+                U32 curr = nextRankVal;
                 nextRankVal += rankStats[w] << (w+rescale);
-                rankVal0[w] = current;
+                rankVal0[w] = curr;
         }   }
         {   U32 const minBits = tableLog+1 - maxW;
             U32 consumed;
@@ -644,23 +704,16 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
 
     dtd.tableLog = (BYTE)maxTableLog;
     dtd.tableType = 1;
-    memcpy(DTable, &dtd, sizeof(dtd));
+    ZSTD_memcpy(DTable, &dtd, sizeof(dtd));
     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));
-}
-
 
 FORCE_INLINE_TEMPLATE U32
 HUF_decodeSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
 {
     size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
-    memcpy(op, dt+val, 2);
+    ZSTD_memcpy(op, dt+val, 2);
     BIT_skipBits(DStream, dt[val].nbBits);
     return dt[val].length;
 }
@@ -669,7 +722,7 @@ FORCE_INLINE_TEMPLATE U32
 HUF_decodeLastSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
 {
     size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
-    memcpy(op, dt+val, 1);
+    ZSTD_memcpy(op, dt+val, 1);
     if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
     else {
         if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
@@ -890,20 +943,6 @@ size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
 }
 
 
-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,
@@ -937,20 +976,6 @@ size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
 }
 
 
-size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
-                              const void* cSrc, size_t cSrcSize)
-{
-    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);
-    return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
-}
-
 #endif /* HUF_FORCE_DECOMPRESS_X1 */
 
 
@@ -1051,67 +1076,6 @@ U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
 }
 
 
-typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
-
-size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
-    static const decompressionAlgo decompress[2] = { HUF_decompress4X1, HUF_decompress4X2 };
-#endif
-
-    /* validation checks */
-    if (dstSize == 0) return ERROR(dstSize_tooSmall);
-    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
-    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
-    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
-
-    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-        (void)algoNb;
-        assert(algoNb == 0);
-        return HUF_decompress4X1(dst, dstSize, cSrc, cSrcSize);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-        (void)algoNb;
-        assert(algoNb == 1);
-        return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);
-#else
-        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
-#endif
-    }
-}
-
-size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    /* validation checks */
-    if (dstSize == 0) return ERROR(dstSize_tooSmall);
-    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
-    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
-    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
-
-    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-        (void)algoNb;
-        assert(algoNb == 0);
-        return HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-        (void)algoNb;
-        assert(algoNb == 1);
-        return HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
-#else
-        return algoNb ? HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
-                        HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
-#endif
-    }
-}
-
-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,
@@ -1145,8 +1109,8 @@ size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
     /* validation checks */
     if (dstSize == 0) return ERROR(dstSize_tooSmall);
     if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
-    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
-    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+    if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
 #if defined(HUF_FORCE_DECOMPRESS_X1)
@@ -1168,14 +1132,6 @@ size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
     }
 }
 
-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)
 {
@@ -1199,7 +1155,7 @@ size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstS
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize);
+    size_t const hSize = HUF_readDTableX1_wksp_bmi2(dctx, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
@@ -1246,3 +1202,149 @@ size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t ds
 #endif
     }
 }
+
+#ifndef ZSTD_NO_UNUSED_FUNCTIONS
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_readDTableX1(HUF_DTable* DTable, const void* src, size_t srcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_readDTableX1_wksp(DTable, src, srcSize,
+                                 workSpace, sizeof(workSpace));
+}
+
+size_t HUF_decompress1X1_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress1X1_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
+}
+
+size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress1X1_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
+}
+#endif 
+
+#ifndef HUF_FORCE_DECOMPRESS_X1
+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));
+}
+
+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);
+}
+#endif
+
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_decompress4X1_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
+}
+size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress4X1_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+#endif
+
+#ifndef HUF_FORCE_DECOMPRESS_X1
+size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    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);
+    return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+#endif
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
+    static const decompressionAlgo decompress[2] = { HUF_decompress4X1, HUF_decompress4X2 };
+#endif
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress4X1(dst, dstSize, cSrc, cSrcSize);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);
+#else
+        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+#endif
+    }
+}
+
+size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
+#else
+        return algoNb ? HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+#endif
+    }
+}
+
+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_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));
+}
+#endif
diff --git a/lib/decompress/zstd_ddict.c b/lib/decompress/zstd_ddict.c
index c8cb8ecc9524..f5cc23b387c4 100644
--- a/lib/decompress/zstd_ddict.c
+++ b/lib/decompress/zstd_ddict.c
@@ -14,7 +14,7 @@
 /*-*******************************************************
 *  Dependencies
 *********************************************************/
-#include <string.h>      /* memcpy, memmove, memset */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
 #include "../common/cpu.h"         /* bmi2 */
 #include "../common/mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
@@ -127,11 +127,11 @@ static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
         ddict->dictContent = dict;
         if (!dict) dictSize = 0;
     } else {
-        void* const internalBuffer = ZSTD_malloc(dictSize, ddict->cMem);
+        void* const internalBuffer = ZSTD_customMalloc(dictSize, ddict->cMem);
         ddict->dictBuffer = internalBuffer;
         ddict->dictContent = internalBuffer;
         if (!internalBuffer) return ERROR(memory_allocation);
-        memcpy(internalBuffer, dict, dictSize);
+        ZSTD_memcpy(internalBuffer, dict, dictSize);
     }
     ddict->dictSize = dictSize;
     ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
@@ -147,9 +147,9 @@ ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
                                       ZSTD_dictContentType_e dictContentType,
                                       ZSTD_customMem customMem)
 {
-    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
 
-    {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
+    {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_customMalloc(sizeof(ZSTD_DDict), customMem);
         if (ddict == NULL) return NULL;
         ddict->cMem = customMem;
         {   size_t const initResult = ZSTD_initDDict_internal(ddict,
@@ -198,7 +198,7 @@ const ZSTD_DDict* ZSTD_initStaticDDict(
     if ((size_t)sBuffer & 7) return NULL;   /* 8-aligned */
     if (sBufferSize < neededSpace) return NULL;
     if (dictLoadMethod == ZSTD_dlm_byCopy) {
-        memcpy(ddict+1, dict, dictSize);  /* local copy */
+        ZSTD_memcpy(ddict+1, dict, dictSize);  /* local copy */
         dict = ddict+1;
     }
     if (ZSTD_isError( ZSTD_initDDict_internal(ddict,
@@ -213,8 +213,8 @@ size_t ZSTD_freeDDict(ZSTD_DDict* ddict)
 {
     if (ddict==NULL) return 0;   /* support free on NULL */
     {   ZSTD_customMem const cMem = ddict->cMem;
-        ZSTD_free(ddict->dictBuffer, cMem);
-        ZSTD_free(ddict, cMem);
+        ZSTD_customFree(ddict->dictBuffer, cMem);
+        ZSTD_customFree(ddict, cMem);
         return 0;
     }
 }
diff --git a/lib/decompress/zstd_ddict.h b/lib/decompress/zstd_ddict.h
index af307efd3d76..8906a71c9404 100644
--- a/lib/decompress/zstd_ddict.h
+++ b/lib/decompress/zstd_ddict.h
@@ -15,7 +15,7 @@
 /*-*******************************************************
  *  Dependencies
  *********************************************************/
-#include <stddef.h>   /* size_t */
+#include "../common/zstd_deps.h"   /* size_t */
 #include "../zstd.h"     /* ZSTD_DDict, and several public functions */
 
 
diff --git a/lib/decompress/zstd_decompress.c b/lib/decompress/zstd_decompress.c
index be5c7cfc3347..21f846bc77e7 100644
--- a/lib/decompress/zstd_decompress.c
+++ b/lib/decompress/zstd_decompress.c
@@ -55,7 +55,7 @@
 /*-*******************************************************
 *  Dependencies
 *********************************************************/
-#include <string.h>      /* memcpy, memmove, memset */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
 #include "../common/cpu.h"         /* bmi2 */
 #include "../common/mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
@@ -94,11 +94,18 @@ static size_t ZSTD_startingInputLength(ZSTD_format_e format)
     return startingInputLength;
 }
 
+static void ZSTD_DCtx_resetParameters(ZSTD_DCtx* dctx)
+{
+    assert(dctx->streamStage == zdss_init);
+    dctx->format = ZSTD_f_zstd1;
+    dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+    dctx->outBufferMode = ZSTD_bm_buffered;
+    dctx->forceIgnoreChecksum = ZSTD_d_validateChecksum;
+}
+
 static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
 {
-    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->dictEnd     = NULL;
@@ -113,7 +120,8 @@ static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
     dctx->noForwardProgress = 0;
     dctx->oversizedDuration = 0;
     dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
-    dctx->outBufferMode = ZSTD_obm_buffered;
+    ZSTD_DCtx_resetParameters(dctx);
+    dctx->validateChecksum = 1;
 #ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
     dctx->dictContentEndForFuzzing = NULL;
 #endif
@@ -134,9 +142,9 @@ ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
 
 ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
 {
-    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
 
-    {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_malloc(sizeof(*dctx), customMem);
+    {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_customMalloc(sizeof(*dctx), customMem);
         if (!dctx) return NULL;
         dctx->customMem = customMem;
         ZSTD_initDCtx_internal(dctx);
@@ -164,13 +172,13 @@ size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
     RETURN_ERROR_IF(dctx->staticSize, memory_allocation, "not compatible with static DCtx");
     {   ZSTD_customMem const cMem = dctx->customMem;
         ZSTD_clearDict(dctx);
-        ZSTD_free(dctx->inBuff, cMem);
+        ZSTD_customFree(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);
+        ZSTD_customFree(dctx, cMem);
         return 0;
     }
 }
@@ -179,7 +187,7 @@ size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
 void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
 {
     size_t const toCopy = (size_t)((char*)(&dstDCtx->inBuff) - (char*)dstDCtx);
-    memcpy(dstDCtx, srcDCtx, toCopy);  /* no need to copy workspace */
+    ZSTD_memcpy(dstDCtx, srcDCtx, toCopy);  /* no need to copy workspace */
 }
 
 
@@ -246,7 +254,7 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
     const BYTE* ip = (const BYTE*)src;
     size_t const minInputSize = ZSTD_startingInputLength(format);
 
-    memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
+    ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
     if (srcSize < minInputSize) return minInputSize;
     RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter");
 
@@ -256,7 +264,7 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
             /* skippable frame */
             if (srcSize < ZSTD_SKIPPABLEHEADERSIZE)
                 return ZSTD_SKIPPABLEHEADERSIZE; /* magic number + frame length */
-            memset(zfhPtr, 0, sizeof(*zfhPtr));
+            ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));
             zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_FRAMEIDSIZE);
             zfhPtr->frameType = ZSTD_skippableFrame;
             return 0;
@@ -446,7 +454,8 @@ static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t he
     RETURN_ERROR_IF(dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID),
                     dictionary_wrong, "");
 #endif
-    if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
+    dctx->validateChecksum = (dctx->fParams.checksumFlag && !dctx->forceIgnoreChecksum) ? 1 : 0;
+    if (dctx->validateChecksum) XXH64_reset(&dctx->xxhState, 0);
     return 0;
 }
 
@@ -461,7 +470,7 @@ static ZSTD_frameSizeInfo ZSTD_errorFrameSizeInfo(size_t ret)
 static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize)
 {
     ZSTD_frameSizeInfo frameSizeInfo;
-    memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
+    ZSTD_memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
     if (ZSTD_isLegacy(src, srcSize))
@@ -516,7 +525,7 @@ static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize
             ip += 4;
         }
 
-        frameSizeInfo.compressedSize = ip - ipstart;
+        frameSizeInfo.compressedSize = (size_t)(ip - ipstart);
         frameSizeInfo.decompressedBound = (zfh.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN)
                                         ? zfh.frameContentSize
                                         : nbBlocks * zfh.blockSizeMax;
@@ -579,12 +588,12 @@ static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
                           const void* src, size_t srcSize)
 {
     DEBUGLOG(5, "ZSTD_copyRawBlock");
+    RETURN_ERROR_IF(srcSize > dstCapacity, dstSize_tooSmall, "");
     if (dst == NULL) {
         if (srcSize == 0) return 0;
         RETURN_ERROR(dstBuffer_null, "");
     }
-    RETURN_ERROR_IF(srcSize > dstCapacity, dstSize_tooSmall, "");
-    memcpy(dst, src, srcSize);
+    ZSTD_memcpy(dst, src, srcSize);
     return srcSize;
 }
 
@@ -592,12 +601,12 @@ static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
                                BYTE b,
                                size_t regenSize)
 {
+    RETURN_ERROR_IF(regenSize > dstCapacity, dstSize_tooSmall, "");
     if (dst == NULL) {
         if (regenSize == 0) return 0;
         RETURN_ERROR(dstBuffer_null, "");
     }
-    RETURN_ERROR_IF(regenSize > dstCapacity, dstSize_tooSmall, "");
-    memset(dst, b, regenSize);
+    ZSTD_memset(dst, b, regenSize);
     return regenSize;
 }
 
@@ -647,13 +656,13 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
         switch(blockProperties.blockType)
         {
         case bt_compressed:
-            decodedSize = ZSTD_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize, /* frame */ 1);
+            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oend-op), ip, cBlockSize, /* frame */ 1);
             break;
         case bt_raw :
-            decodedSize = ZSTD_copyRawBlock(op, oend-op, ip, cBlockSize);
+            decodedSize = ZSTD_copyRawBlock(op, (size_t)(oend-op), ip, cBlockSize);
             break;
         case bt_rle :
-            decodedSize = ZSTD_setRleBlock(op, oend-op, *ip, blockProperties.origSize);
+            decodedSize = ZSTD_setRleBlock(op, (size_t)(oend-op), *ip, blockProperties.origSize);
             break;
         case bt_reserved :
         default:
@@ -661,7 +670,7 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
         }
 
         if (ZSTD_isError(decodedSize)) return decodedSize;
-        if (dctx->fParams.checksumFlag)
+        if (dctx->validateChecksum)
             XXH64_update(&dctx->xxhState, op, decodedSize);
         if (decodedSize != 0)
             op += decodedSize;
@@ -676,11 +685,13 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
                         corruption_detected, "");
     }
     if (dctx->fParams.checksumFlag) { /* Frame content checksum verification */
-        U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
-        U32 checkRead;
         RETURN_ERROR_IF(remainingSrcSize<4, checksum_wrong, "");
-        checkRead = MEM_readLE32(ip);
-        RETURN_ERROR_IF(checkRead != checkCalc, checksum_wrong, "");
+        if (!dctx->forceIgnoreChecksum) {
+            U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
+            U32 checkRead;
+            checkRead = MEM_readLE32(ip);
+            RETURN_ERROR_IF(checkRead != checkCalc, checksum_wrong, "");
+        }
         ip += 4;
         remainingSrcSize -= 4;
     }
@@ -688,7 +699,7 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
     /* Allow caller to get size read */
     *srcPtr = ip;
     *srcSizePtr = remainingSrcSize;
-    return op-ostart;
+    return (size_t)(op-ostart);
 }
 
 static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
@@ -721,7 +732,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
             decodedSize = ZSTD_decompressLegacy(dst, dstCapacity, src, frameSize, dict, dictSize);
             if (ZSTD_isError(decodedSize)) return decodedSize;
 
-            assert(decodedSize <=- dstCapacity);
+            assert(decodedSize <= dstCapacity);
             dst = (BYTE*)dst + decodedSize;
             dstCapacity -= decodedSize;
 
@@ -761,15 +772,13 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                 (ZSTD_getErrorCode(res) == ZSTD_error_prefix_unknown)
              && (moreThan1Frame==1),
                 srcSize_wrong,
-                "at least one frame successfully completed, but following "
-                "bytes are garbage: it's more likely to be a srcSize error, "
-                "specifying more bytes than compressed size of frame(s). This "
-                "error message replaces ERROR(prefix_unknown), which would be "
-                "confusing, as the first header is actually correct. Note that "
-                "one could be unlucky, it might be a corruption error instead, "
-                "happening right at the place where we expect zstd magic "
-                "bytes. But this is _much_ less likely than a srcSize field "
-                "error.");
+                "At least one frame successfully completed, "
+                "but following bytes are garbage: "
+                "it's more likely to be a srcSize error, "
+                "specifying more input bytes than size of frame(s). "
+                "Note: one could be unlucky, it might be a corruption error instead, "
+                "happening right at the place where we expect zstd magic bytes. "
+                "But this is _much_ less likely than a srcSize field error.");
             if (ZSTD_isError(res)) return res;
             assert(res <= dstCapacity);
             if (res != 0)
@@ -781,7 +790,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
 
     RETURN_ERROR_IF(srcSize, srcSize_wrong, "input not entirely consumed");
 
-    return (BYTE*)dst - (BYTE*)dststart;
+    return (size_t)((BYTE*)dst - (BYTE*)dststart);
 }
 
 size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
@@ -899,21 +908,21 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
         if (dctx->format == ZSTD_f_zstd1) {  /* allows header */
             assert(srcSize >= ZSTD_FRAMEIDSIZE);  /* to read skippable magic number */
             if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
-                memcpy(dctx->headerBuffer, src, srcSize);
+                ZSTD_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, srcSize);
+        ZSTD_memcpy(dctx->headerBuffer, src, srcSize);
         dctx->expected = dctx->headerSize - srcSize;
         dctx->stage = ZSTDds_decodeFrameHeader;
         return 0;
 
     case ZSTDds_decodeFrameHeader:
         assert(src != NULL);
-        memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
+        ZSTD_memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
         FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize), "");
         dctx->expected = ZSTD_blockHeaderSize;
         dctx->stage = ZSTDds_decodeBlockHeader;
@@ -977,7 +986,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
             RETURN_ERROR_IF(rSize > dctx->fParams.blockSizeMax, corruption_detected, "Decompressed Block Size Exceeds Maximum");
             DEBUGLOG(5, "ZSTD_decompressContinue: decoded size from block : %u", (unsigned)rSize);
             dctx->decodedSize += rSize;
-            if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, dst, rSize);
+            if (dctx->validateChecksum) XXH64_update(&dctx->xxhState, dst, rSize);
             dctx->previousDstEnd = (char*)dst + rSize;
 
             /* Stay on the same stage until we are finished streaming the block. */
@@ -1007,10 +1016,13 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
 
     case ZSTDds_checkChecksum:
         assert(srcSize == 4);  /* guaranteed by dctx->expected */
-        {   U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
-            U32 const check32 = MEM_readLE32(src);
-            DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", (unsigned)h32, (unsigned)check32);
-            RETURN_ERROR_IF(check32 != h32, checksum_wrong, "");
+        {
+            if (dctx->validateChecksum) {
+                U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
+                U32 const check32 = MEM_readLE32(src);
+                DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", (unsigned)h32, (unsigned)check32);
+                RETURN_ERROR_IF(check32 != h32, checksum_wrong, "");
+            }
             dctx->expected = 0;
             dctx->stage = ZSTDds_getFrameHeaderSize;
             return 0;
@@ -1019,7 +1031,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
     case ZSTDds_decodeSkippableHeader:
         assert(src != NULL);
         assert(srcSize <= ZSTD_SKIPPABLEHEADERSIZE);
-        memcpy(dctx->headerBuffer + (ZSTD_SKIPPABLEHEADERSIZE - srcSize), src, srcSize);   /* complete skippable header */
+        ZSTD_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;
@@ -1075,7 +1087,7 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
                                                 workspace, workspaceSize);
 #else
         size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
-                                                dictPtr, dictEnd - dictPtr,
+                                                dictPtr, (size_t)(dictEnd - dictPtr),
                                                 workspace, workspaceSize);
 #endif
         RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted, "");
@@ -1084,40 +1096,46 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
 
     {   short offcodeNCount[MaxOff+1];
         unsigned offcodeMaxValue = MaxOff, offcodeLog;
-        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, (size_t)(dictEnd-dictPtr));
         RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
         RETURN_ERROR_IF(offcodeMaxValue > MaxOff, dictionary_corrupted, "");
         RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
         ZSTD_buildFSETable( entropy->OFTable,
                             offcodeNCount, offcodeMaxValue,
                             OF_base, OF_bits,
-                            offcodeLog);
+                            offcodeLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */0);
         dictPtr += offcodeHeaderSize;
     }
 
     {   short matchlengthNCount[MaxML+1];
         unsigned matchlengthMaxValue = MaxML, matchlengthLog;
-        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
+        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
         RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
         RETURN_ERROR_IF(matchlengthMaxValue > MaxML, dictionary_corrupted, "");
         RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
         ZSTD_buildFSETable( entropy->MLTable,
                             matchlengthNCount, matchlengthMaxValue,
                             ML_base, ML_bits,
-                            matchlengthLog);
+                            matchlengthLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */ 0);
         dictPtr += matchlengthHeaderSize;
     }
 
     {   short litlengthNCount[MaxLL+1];
         unsigned litlengthMaxValue = MaxLL, litlengthLog;
-        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
+        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
         RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
         RETURN_ERROR_IF(litlengthMaxValue > MaxLL, dictionary_corrupted, "");
         RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
         ZSTD_buildFSETable( entropy->LLTable,
                             litlengthNCount, litlengthMaxValue,
                             LL_base, LL_bits,
-                            litlengthLog);
+                            litlengthLog,
+                            entropy->workspace, sizeof(entropy->workspace),
+                            /* bmi2 */ 0);
         dictPtr += litlengthHeaderSize;
     }
 
@@ -1131,7 +1149,7 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
             entropy->rep[i] = rep;
     }   }
 
-    return dictPtr - (const BYTE*)dict;
+    return (size_t)(dictPtr - (const BYTE*)dict);
 }
 
 static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
@@ -1170,7 +1188,7 @@ size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
     dctx->dictID = 0;
     dctx->bType = bt_reserved;
     ZSTD_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
-    memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
+    ZSTD_memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
     dctx->LLTptr = dctx->entropy.LLTable;
     dctx->MLTptr = dctx->entropy.MLTable;
     dctx->OFTptr = dctx->entropy.OFTable;
@@ -1394,7 +1412,7 @@ size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
 
 size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format)
 {
-    return ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, format);
+    return ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, (int)format);
 }
 
 ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
@@ -1411,8 +1429,12 @@ ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
             ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
             return bounds;
         case ZSTD_d_stableOutBuffer:
-            bounds.lowerBound = (int)ZSTD_obm_buffered;
-            bounds.upperBound = (int)ZSTD_obm_stable;
+            bounds.lowerBound = (int)ZSTD_bm_buffered;
+            bounds.upperBound = (int)ZSTD_bm_stable;
+            return bounds;
+        case ZSTD_d_forceIgnoreChecksum:
+            bounds.lowerBound = (int)ZSTD_d_validateChecksum;
+            bounds.upperBound = (int)ZSTD_d_ignoreChecksum;
             return bounds;
         default:;
     }
@@ -1436,6 +1458,26 @@ static int ZSTD_dParam_withinBounds(ZSTD_dParameter dParam, int value)
     RETURN_ERROR_IF(!ZSTD_dParam_withinBounds(p, v), parameter_outOfBound, ""); \
 }
 
+size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value)
+{
+    switch (param) {
+        case ZSTD_d_windowLogMax:
+            *value = (int)ZSTD_highbit32((U32)dctx->maxWindowSize);
+            return 0;
+        case ZSTD_d_format:
+            *value = (int)dctx->format;
+            return 0;
+        case ZSTD_d_stableOutBuffer:
+            *value = (int)dctx->outBufferMode;
+            return 0;
+        case ZSTD_d_forceIgnoreChecksum:
+            *value = (int)dctx->forceIgnoreChecksum;
+            return 0;
+        default:;
+    }
+    RETURN_ERROR(parameter_unsupported, "");
+}
+
 size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value)
 {
     RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
@@ -1451,7 +1493,11 @@ size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value
             return 0;
         case ZSTD_d_stableOutBuffer:
             CHECK_DBOUNDS(ZSTD_d_stableOutBuffer, value);
-            dctx->outBufferMode = (ZSTD_outBufferMode_e)value;
+            dctx->outBufferMode = (ZSTD_bufferMode_e)value;
+            return 0;
+        case ZSTD_d_forceIgnoreChecksum:
+            CHECK_DBOUNDS(ZSTD_d_forceIgnoreChecksum, value);
+            dctx->forceIgnoreChecksum = (ZSTD_forceIgnoreChecksum_e)value;
             return 0;
         default:;
     }
@@ -1469,8 +1515,7 @@ size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset)
       || (reset == ZSTD_reset_session_and_parameters) ) {
         RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
         ZSTD_clearDict(dctx);
-        dctx->format = ZSTD_f_zstd1;
-        dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+        ZSTD_DCtx_resetParameters(dctx);
     }
     return 0;
 }
@@ -1524,7 +1569,7 @@ static void ZSTD_DCtx_updateOversizedDuration(ZSTD_DStream* zds, size_t const ne
 {
     if (ZSTD_DCtx_isOverflow(zds, neededInBuffSize, neededOutBuffSize))
         zds->oversizedDuration++;
-    else 
+    else
         zds->oversizedDuration = 0;
 }
 
@@ -1538,7 +1583,7 @@ static size_t ZSTD_checkOutBuffer(ZSTD_DStream const* zds, ZSTD_outBuffer const*
 {
     ZSTD_outBuffer const expect = zds->expectedOutBuffer;
     /* No requirement when ZSTD_obm_stable is not enabled. */
-    if (zds->outBufferMode != ZSTD_obm_stable)
+    if (zds->outBufferMode != ZSTD_bm_stable)
         return 0;
     /* Any buffer is allowed in zdss_init, this must be the same for every other call until
      * the context is reset.
@@ -1548,7 +1593,7 @@ static size_t ZSTD_checkOutBuffer(ZSTD_DStream const* zds, ZSTD_outBuffer const*
     /* The buffer must match our expectation exactly. */
     if (expect.dst == output->dst && expect.pos == output->pos && expect.size == output->size)
         return 0;
-    RETURN_ERROR(dstBuffer_wrong, "ZSTD_obm_stable enabled but output differs!");
+    RETURN_ERROR(dstBuffer_wrong, "ZSTD_d_stableOutBuffer enabled but output differs!");
 }
 
 /* Calls ZSTD_decompressContinue() with the right parameters for ZSTD_decompressStream()
@@ -1560,7 +1605,7 @@ static size_t ZSTD_decompressContinueStream(
             ZSTD_DStream* zds, char** op, char* oend,
             void const* src, size_t srcSize) {
     int const isSkipFrame = ZSTD_isSkipFrame(zds);
-    if (zds->outBufferMode == ZSTD_obm_buffered) {
+    if (zds->outBufferMode == ZSTD_bm_buffered) {
         size_t const dstSize = isSkipFrame ? 0 : zds->outBuffSize - zds->outStart;
         size_t const decodedSize = ZSTD_decompressContinue(zds,
                 zds->outBuff + zds->outStart, dstSize, src, srcSize);
@@ -1573,14 +1618,14 @@ static size_t ZSTD_decompressContinueStream(
         }
     } else {
         /* Write directly into the output buffer */
-        size_t const dstSize = isSkipFrame ? 0 : oend - *op;
+        size_t const dstSize = isSkipFrame ? 0 : (size_t)(oend - *op);
         size_t const decodedSize = ZSTD_decompressContinue(zds, *op, dstSize, src, srcSize);
         FORWARD_IF_ERROR(decodedSize, "");
         *op += decodedSize;
         /* Flushing is not needed. */
         zds->streamStage = zdss_read;
         assert(*op <= oend);
-        assert(zds->outBufferMode == ZSTD_obm_stable);
+        assert(zds->outBufferMode == ZSTD_bm_stable);
     }
     return 0;
 }
@@ -1663,14 +1708,14 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                     assert(iend >= ip);
                     if (toLoad > remainingInput) {   /* not enough input to load full header */
                         if (remainingInput > 0) {
-                            memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
+                            ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
                             zds->lhSize += remainingInput;
                         }
                         input->pos = input->size;
                         return (MAX((size_t)ZSTD_FRAMEHEADERSIZE_MIN(zds->format), 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;
+                    ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
                     break;
             }   }
 
@@ -1678,10 +1723,10 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             if (zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
                 && zds->fParams.frameType != ZSTD_skippableFrame
                 && (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
-                size_t const cSize = ZSTD_findFrameCompressedSize(istart, iend-istart);
+                size_t const cSize = ZSTD_findFrameCompressedSize(istart, (size_t)(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, ZSTD_getDDict(zds));
+                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, (size_t)(oend-op), istart, cSize, ZSTD_getDDict(zds));
                     if (ZSTD_isError(decompressedSize)) return decompressedSize;
                     DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()")
                     ip = istart + cSize;
@@ -1693,7 +1738,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             }   }
 
             /* Check output buffer is large enough for ZSTD_odm_stable. */
-            if (zds->outBufferMode == ZSTD_obm_stable
+            if (zds->outBufferMode == ZSTD_bm_stable
                 && zds->fParams.frameType != ZSTD_skippableFrame
                 && zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
                 && (U64)(size_t)(oend-op) < zds->fParams.frameContentSize) {
@@ -1723,7 +1768,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
 
             /* Adapt buffer sizes to frame header instructions */
             {   size_t const neededInBuffSize = MAX(zds->fParams.blockSizeMax, 4 /* frame checksum */);
-                size_t const neededOutBuffSize = zds->outBufferMode == ZSTD_obm_buffered
+                size_t const neededOutBuffSize = zds->outBufferMode == ZSTD_bm_buffered
                         ? ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize)
                         : 0;
 
@@ -1731,7 +1776,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
 
                 {   int const tooSmall = (zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize);
                     int const tooLarge = ZSTD_DCtx_isOversizedTooLong(zds);
-                    
+
                     if (tooSmall || tooLarge) {
                         size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
                         DEBUGLOG(4, "inBuff  : from %u to %u",
@@ -1745,10 +1790,10 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                                 bufferSize > zds->staticSize - sizeof(ZSTD_DCtx),
                                 memory_allocation, "");
                         } else {
-                            ZSTD_free(zds->inBuff, zds->customMem);
+                            ZSTD_customFree(zds->inBuff, zds->customMem);
                             zds->inBuffSize = 0;
                             zds->outBuffSize = 0;
-                            zds->inBuff = (char*)ZSTD_malloc(bufferSize, zds->customMem);
+                            zds->inBuff = (char*)ZSTD_customMalloc(bufferSize, zds->customMem);
                             RETURN_ERROR_IF(zds->inBuff == NULL, memory_allocation, "");
                         }
                         zds->inBuffSize = neededInBuffSize;
@@ -1760,7 +1805,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
 
         case zdss_read:
             DEBUGLOG(5, "stage zdss_read");
-            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompressWithInputSize(zds, iend - ip);
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip));
                 DEBUGLOG(5, "neededInSize = %u", (U32)neededInSize);
                 if (neededInSize==0) {  /* end of frame */
                     zds->streamStage = zdss_init;
@@ -1790,7 +1835,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                     RETURN_ERROR_IF(toLoad > zds->inBuffSize - zds->inPos,
                                     corruption_detected,
                                     "should never happen");
-                    loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, iend-ip);
+                    loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, (size_t)(iend-ip));
                 }
                 ip += loadedSize;
                 zds->inPos += loadedSize;
@@ -1804,7 +1849,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             }
         case zdss_flush:
             {   size_t const toFlushSize = zds->outEnd - zds->outStart;
-                size_t const flushedSize = ZSTD_limitCopy(op, oend-op, zds->outBuff + zds->outStart, toFlushSize);
+                size_t const flushedSize = ZSTD_limitCopy(op, (size_t)(oend-op), zds->outBuff + zds->outStart, toFlushSize);
                 op += flushedSize;
                 zds->outStart += flushedSize;
                 if (flushedSize == toFlushSize) {  /* flush completed */
diff --git a/lib/decompress/zstd_decompress_block.c b/lib/decompress/zstd_decompress_block.c
index ad3b3d8dbd0a..19cbdc5c16eb 100644
--- a/lib/decompress/zstd_decompress_block.c
+++ b/lib/decompress/zstd_decompress_block.c
@@ -14,7 +14,7 @@
 /*-*******************************************************
 *  Dependencies
 *********************************************************/
-#include <string.h>      /* memcpy, memmove, memset */
+#include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
 #include "../common/compiler.h"    /* prefetch */
 #include "../common/cpu.h"         /* bmi2 */
 #include "../common/mem.h"         /* low level memory routines */
@@ -44,7 +44,7 @@
 /*_*******************************************************
 *  Memory operations
 **********************************************************/
-static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+static void ZSTD_copy4(void* dst, const void* src) { ZSTD_memcpy(dst, src, 4); }
 
 
 /*-*************************************************************
@@ -166,7 +166,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                 dctx->litSize = litSize;
                 dctx->litEntropy = 1;
                 if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
-                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                ZSTD_memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
                 return litCSize + lhSize;
             }
 
@@ -191,10 +191,10 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
 
                 if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
                     RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected, "");
-                    memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                    ZSTD_memcpy(dctx->litBuffer, istart+lhSize, litSize);
                     dctx->litPtr = dctx->litBuffer;
                     dctx->litSize = litSize;
-                    memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                    ZSTD_memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
                     return lhSize+litSize;
                 }
                 /* direct reference into compressed stream */
@@ -223,7 +223,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     break;
                 }
                 RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
-                memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+                ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
                 dctx->litPtr = dctx->litBuffer;
                 dctx->litSize = litSize;
                 return lhSize+1;
@@ -236,7 +236,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
 
 /* 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
+ * https://github.com/facebook/zstd/blob/release/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()
@@ -364,23 +364,26 @@ static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U32 nbAddB
  * generate FSE decoding table for one symbol (ll, ml or off)
  * cannot fail if input is valid =>
  * all inputs are presumed validated at this stage */
-void
-ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
+FORCE_INLINE_TEMPLATE
+void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
             const short* normalizedCounter, unsigned maxSymbolValue,
             const U32* baseValue, const U32* nbAdditionalBits,
-            unsigned tableLog)
+            unsigned tableLog, void* wksp, size_t wkspSize)
 {
     ZSTD_seqSymbol* const tableDecode = dt+1;
-    U16 symbolNext[MaxSeq+1];
-
     U32 const maxSV1 = maxSymbolValue + 1;
     U32 const tableSize = 1 << tableLog;
-    U32 highThreshold = tableSize-1;
+
+    U16* symbolNext = (U16*)wksp;
+    BYTE* spread = (BYTE*)(symbolNext + MaxSeq + 1);
+    U32 highThreshold = tableSize - 1;
+
 
     /* Sanity Checks */
     assert(maxSymbolValue <= MaxSeq);
     assert(tableLog <= MaxFSELog);
-
+    assert(wkspSize >= ZSTD_BUILD_FSE_TABLE_WKSP_SIZE);
+    (void)wkspSize;
     /* Init, lay down lowprob symbols */
     {   ZSTD_seqSymbol_header DTableH;
         DTableH.tableLog = tableLog;
@@ -396,16 +399,69 @@ ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
                     assert(normalizedCounter[s]>=0);
                     symbolNext[s] = (U16)normalizedCounter[s];
         }   }   }
-        memcpy(dt, &DTableH, sizeof(DTableH));
+        ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
     }
 
     /* Spread symbols */
-    {   U32 const tableMask = tableSize-1;
+    assert(tableSize <= 512);
+    /* Specialized symbol spreading for the case when there are
+     * no low probability (-1 count) symbols. When compressing
+     * small blocks we avoid low probability symbols to hit this
+     * case, since header decoding speed matters more.
+     */
+    if (highThreshold == tableSize - 1) {
+        size_t const tableMask = tableSize-1;
+        size_t const step = FSE_TABLESTEP(tableSize);
+        /* First lay down the symbols in order.
+         * We use a uint64_t to lay down 8 bytes at a time. This reduces branch
+         * misses since small blocks generally have small table logs, so nearly
+         * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
+         * our buffer to handle the over-write.
+         */
+        {
+            U64 const add = 0x0101010101010101ull;
+            size_t pos = 0;
+            U64 sv = 0;
+            U32 s;
+            for (s=0; s<maxSV1; ++s, sv += add) {
+                int i;
+                int const n = normalizedCounter[s];
+                MEM_write64(spread + pos, sv);
+                for (i = 8; i < n; i += 8) {
+                    MEM_write64(spread + pos + i, sv);
+                }
+                pos += n;
+            }
+        }
+        /* Now we spread those positions across the table.
+         * The benefit of doing it in two stages is that we avoid the the
+         * variable size inner loop, which caused lots of branch misses.
+         * Now we can run through all the positions without any branch misses.
+         * We unroll the loop twice, since that is what emperically worked best.
+         */
+        {
+            size_t position = 0;
+            size_t s;
+            size_t const unroll = 2;
+            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
+            for (s = 0; s < (size_t)tableSize; s += unroll) {
+                size_t u;
+                for (u = 0; u < unroll; ++u) {
+                    size_t const uPosition = (position + (u * step)) & tableMask;
+                    tableDecode[uPosition].baseValue = spread[s + u];
+                }
+                position = (position + (unroll * step)) & tableMask;
+            }
+            assert(position == 0);
+        }
+    } else {
+        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++) {
+            int const n = normalizedCounter[s];
+            for (i=0; i<n; i++) {
                 tableDecode[position].baseValue = s;
                 position = (position + step) & tableMask;
                 while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
@@ -414,7 +470,8 @@ ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
     }
 
     /* Build Decoding table */
-    {   U32 u;
+    {
+        U32 u;
         for (u=0; u<tableSize; u++) {
             U32 const symbol = tableDecode[u].baseValue;
             U32 const nextState = symbolNext[symbol]++;
@@ -423,7 +480,46 @@ ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
             assert(nbAdditionalBits[symbol] < 255);
             tableDecode[u].nbAdditionalBits = (BYTE)nbAdditionalBits[symbol];
             tableDecode[u].baseValue = baseValue[symbol];
-    }   }
+        }
+    }
+}
+
+/* Avoids the FORCE_INLINE of the _body() function. */
+static void ZSTD_buildFSETable_body_default(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U32* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize)
+{
+    ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
+            baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+}
+
+#if DYNAMIC_BMI2
+TARGET_ATTRIBUTE("bmi2") static void ZSTD_buildFSETable_body_bmi2(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U32* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize)
+{
+    ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
+            baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+}
+#endif
+
+void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U32* nbAdditionalBits,
+            unsigned tableLog, void* wksp, size_t wkspSize, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        ZSTD_buildFSETable_body_bmi2(dt, normalizedCounter, maxSymbolValue,
+                baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
+        return;
+    }
+#endif
+    (void)bmi2;
+    ZSTD_buildFSETable_body_default(dt, normalizedCounter, maxSymbolValue,
+            baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
 }
 
 
@@ -435,7 +531,8 @@ static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymb
                                  const void* src, size_t srcSize,
                                  const U32* baseValue, const U32* nbAdditionalBits,
                                  const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable,
-                                 int ddictIsCold, int nbSeq)
+                                 int ddictIsCold, int nbSeq, U32* wksp, size_t wkspSize,
+                                 int bmi2)
 {
     switch(type)
     {
@@ -467,7 +564,7 @@ static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymb
             size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
             RETURN_ERROR_IF(FSE_isError(headerSize), corruption_detected, "");
             RETURN_ERROR_IF(tableLog > maxLog, corruption_detected, "");
-            ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog);
+            ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog, wksp, wkspSize, bmi2);
             *DTablePtr = DTableSpace;
             return headerSize;
         }
@@ -499,7 +596,8 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
     if (nbSeq > 0x7F) {
         if (nbSeq == 0xFF) {
             RETURN_ERROR_IF(ip+2 > iend, srcSize_wrong, "");
-            nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+            nbSeq = MEM_readLE16(ip) + LONGNBSEQ;
+            ip+=2;
         } else {
             RETURN_ERROR_IF(ip >= iend, srcSize_wrong, "");
             nbSeq = ((nbSeq-0x80)<<8) + *ip++;
@@ -520,7 +618,9 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       ip, iend-ip,
                                                       LL_base, LL_bits,
                                                       LL_defaultDTable, dctx->fseEntropy,
-                                                      dctx->ddictIsCold, nbSeq);
+                                                      dctx->ddictIsCold, nbSeq,
+                                                      dctx->workspace, sizeof(dctx->workspace),
+                                                      dctx->bmi2);
             RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected, "ZSTD_buildSeqTable failed");
             ip += llhSize;
         }
@@ -530,7 +630,9 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       ip, iend-ip,
                                                       OF_base, OF_bits,
                                                       OF_defaultDTable, dctx->fseEntropy,
-                                                      dctx->ddictIsCold, nbSeq);
+                                                      dctx->ddictIsCold, nbSeq,
+                                                      dctx->workspace, sizeof(dctx->workspace),
+                                                      dctx->bmi2);
             RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected, "ZSTD_buildSeqTable failed");
             ip += ofhSize;
         }
@@ -540,7 +642,9 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       ip, iend-ip,
                                                       ML_base, ML_bits,
                                                       ML_defaultDTable, dctx->fseEntropy,
-                                                      dctx->ddictIsCold, nbSeq);
+                                                      dctx->ddictIsCold, nbSeq,
+                                                      dctx->workspace, sizeof(dctx->workspace),
+                                                      dctx->bmi2);
             RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected, "ZSTD_buildSeqTable failed");
             ip += mlhSize;
         }
@@ -686,12 +790,12 @@ size_t ZSTD_execSequenceEnd(BYTE* op,
         RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
         match = dictEnd - (prefixStart-match);
         if (match + sequence.matchLength <= dictEnd) {
-            memmove(oLitEnd, match, sequence.matchLength);
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
             return sequenceLength;
         }
         /* span extDict & currentPrefixSegment */
         {   size_t const length1 = dictEnd - match;
-            memmove(oLitEnd, match, length1);
+            ZSTD_memmove(oLitEnd, match, length1);
             op = oLitEnd + length1;
             sequence.matchLength -= length1;
             match = prefixStart;
@@ -752,12 +856,12 @@ size_t ZSTD_execSequence(BYTE* op,
         RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
         match = dictEnd + (match - prefixStart);
         if (match + sequence.matchLength <= dictEnd) {
-            memmove(oLitEnd, match, sequence.matchLength);
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
             return sequenceLength;
         }
         /* span extDict & currentPrefixSegment */
         {   size_t const length1 = dictEnd - match;
-            memmove(oLitEnd, match, length1);
+            ZSTD_memmove(oLitEnd, match, length1);
             op = oLitEnd + length1;
             sequence.matchLength -= length1;
             match = prefixStart;
@@ -948,7 +1052,7 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets, c
 }
 
 #ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
-static int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
+MEM_STATIC int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
 {
     size_t const windowSize = dctx->fParams.windowSize;
     /* No dictionary used. */
@@ -969,6 +1073,7 @@ MEM_STATIC void ZSTD_assertValidSequence(
         seq_t const seq,
         BYTE const* prefixStart, BYTE const* virtualStart)
 {
+#if DEBUGLEVEL >= 1
     size_t const windowSize = dctx->fParams.windowSize;
     size_t const sequenceSize = seq.litLength + seq.matchLength;
     BYTE const* const oLitEnd = op + seq.litLength;
@@ -986,6 +1091,9 @@ MEM_STATIC void ZSTD_assertValidSequence(
         /* Offset must be within our window. */
         assert(seq.offset <= windowSize);
     }
+#else
+    (void)dctx, (void)op, (void)oend, (void)seq, (void)prefixStart, (void)virtualStart;
+#endif
 }
 #endif
 
@@ -1080,14 +1188,14 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
 #endif
             DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
             BIT_reloadDStream(&(seqState.DStream));
+            op += oneSeqSize;
             /* gcc and clang both don't like early returns in this loop.
-             * gcc doesn't like early breaks either.
-             * Instead save an error and report it at the end.
-             * When there is an error, don't increment op, so we don't
-             * overwrite.
+             * Instead break and check for an error at the end of the loop.
              */
-            if (UNLIKELY(ZSTD_isError(oneSeqSize))) error = oneSeqSize;
-            else op += oneSeqSize;
+            if (UNLIKELY(ZSTD_isError(oneSeqSize))) {
+                error = oneSeqSize;
+                break;
+            }
             if (UNLIKELY(!--nbSeq)) break;
         }
 
@@ -1104,7 +1212,7 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
     {   size_t const lastLLSize = litEnd - litPtr;
         RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
         if (op != NULL) {
-            memcpy(op, litPtr, lastLLSize);
+            ZSTD_memcpy(op, litPtr, lastLLSize);
             op += lastLLSize;
         }
     }
@@ -1209,7 +1317,7 @@ ZSTD_decompressSequencesLong_body(
     {   size_t const lastLLSize = litEnd - litPtr;
         RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
         if (op != NULL) {
-            memcpy(op, litPtr, lastLLSize);
+            ZSTD_memcpy(op, litPtr, lastLLSize);
             op += lastLLSize;
         }
     }
diff --git a/lib/decompress/zstd_decompress_block.h b/lib/decompress/zstd_decompress_block.h
index bf39b7350cf9..b5715c168e2b 100644
--- a/lib/decompress/zstd_decompress_block.h
+++ b/lib/decompress/zstd_decompress_block.h
@@ -15,7 +15,7 @@
 /*-*******************************************************
  *  Dependencies
  *********************************************************/
-#include <stddef.h>   /* size_t */
+#include "../common/zstd_deps.h"   /* size_t */
 #include "../zstd.h"    /* DCtx, and some public functions */
 #include "../common/zstd_internal.h"  /* blockProperties_t, and some public functions */
 #include "zstd_decompress_internal.h"  /* ZSTD_seqSymbol */
@@ -48,12 +48,15 @@ size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
  * this function must be called with valid parameters only
  * (dt is large enough, normalizedCounter distribution total is a power of 2, max is within range, etc.)
  * in which case it cannot fail.
+ * The workspace must be 4-byte aligned and at least ZSTD_BUILD_FSE_TABLE_WKSP_SIZE bytes, which is
+ * defined in zstd_decompress_internal.h.
  * Internal use only.
  */
 void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
              const short* normalizedCounter, unsigned maxSymbolValue,
              const U32* baseValue, const U32* nbAdditionalBits,
-                   unsigned tableLog);
+                   unsigned tableLog, void* wksp, size_t wkspSize,
+                   int bmi2);
 
 
 #endif /* ZSTD_DEC_BLOCK_H */
diff --git a/lib/decompress/zstd_decompress_internal.h b/lib/decompress/zstd_decompress_internal.h
index 9ad96c554885..f80b471e9940 100644
--- a/lib/decompress/zstd_decompress_internal.h
+++ b/lib/decompress/zstd_decompress_internal.h
@@ -27,26 +27,26 @@
 /*-*******************************************************
  *  Constants
  *********************************************************/
-static const U32 LL_base[MaxLL+1] = {
+static UNUSED_ATTR 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] = {
+static UNUSED_ATTR 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] = {
+static UNUSED_ATTR 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] = {
+static UNUSED_ATTR 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,
@@ -73,12 +73,16 @@ static const U32 ML_base[MaxML+1] = {
 
  #define SEQSYMBOL_TABLE_SIZE(log)   (1 + (1 << (log)))
 
+#define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE (sizeof(S16) * (MaxSeq + 1) + (1u << MaxFSELog) + sizeof(U64))
+#define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32 ((ZSTD_BUILD_FSE_TABLE_WKSP_SIZE + sizeof(U32) - 1) / sizeof(U32))
+
 typedef struct {
     ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];    /* Note : Space reserved for FSE Tables */
     ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];   /* is also used as temporary workspace while building hufTable during DDict creation */
     ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];    /* and therefore must be at least HUF_DECOMPRESS_WORKSPACE_SIZE large */
     HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
     U32 rep[ZSTD_REP_NUM];
+    U32 workspace[ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32];
 } ZSTD_entropyDTables_t;
 
 typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
@@ -95,11 +99,6 @@ typedef enum {
     ZSTD_use_once = 1            /* Use the dictionary once and set to ZSTD_dont_use */
 } ZSTD_dictUses_e;
 
-typedef enum {
-    ZSTD_obm_buffered = 0,  /* Buffer the output */
-    ZSTD_obm_stable = 1     /* ZSTD_outBuffer is stable */
-} ZSTD_outBufferMode_e;
-
 struct ZSTD_DCtx_s
 {
     const ZSTD_seqSymbol* LLTptr;
@@ -122,6 +121,8 @@ struct ZSTD_DCtx_s
     XXH64_state_t xxhState;
     size_t headerSize;
     ZSTD_format_e format;
+    ZSTD_forceIgnoreChecksum_e forceIgnoreChecksum;   /* User specified: if == 1, will ignore checksums in compressed frame. Default == 0 */
+    U32 validateChecksum;         /* if == 1, will validate checksum. Is == 1 if (fParams.checksumFlag == 1) and (forceIgnoreChecksum == 0). */
     const BYTE* litPtr;
     ZSTD_customMem customMem;
     size_t litSize;
@@ -152,7 +153,7 @@ struct ZSTD_DCtx_s
     U32 legacyVersion;
     U32 hostageByte;
     int noForwardProgress;
-    ZSTD_outBufferMode_e outBufferMode;
+    ZSTD_bufferMode_e outBufferMode;
     ZSTD_outBuffer expectedOutBuffer;
 
     /* workspace */
diff --git a/lib/dictBuilder/cover.c b/lib/dictBuilder/cover.c
index da54ef15f2b8..c78af133a9d5 100644
--- a/lib/dictBuilder/cover.c
+++ b/lib/dictBuilder/cover.c
@@ -40,33 +40,42 @@
 *  Constants
 ***************************************/
 #define COVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
-#define DEFAULT_SPLITPOINT 1.0
+#define COVER_DEFAULT_SPLITPOINT 1.0
 
 /*-*************************************
 *  Console display
 ***************************************/
+#ifndef LOCALDISPLAYLEVEL
 static int g_displayLevel = 2;
+#endif
+#undef  DISPLAY
 #define DISPLAY(...)                                                           \
   {                                                                            \
     fprintf(stderr, __VA_ARGS__);                                              \
     fflush(stderr);                                                            \
   }
+#undef  LOCALDISPLAYLEVEL
 #define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
   if (displayLevel >= l) {                                                     \
     DISPLAY(__VA_ARGS__);                                                      \
   } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+#undef  DISPLAYLEVEL
 #define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
 
+#ifndef LOCALDISPLAYUPDATE
+static const clock_t g_refreshRate = CLOCKS_PER_SEC * 15 / 100;
+static clock_t g_time = 0;
+#endif
+#undef  LOCALDISPLAYUPDATE
 #define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
   if (displayLevel >= l) {                                                     \
-    if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
+    if ((clock() - g_time > g_refreshRate) || (displayLevel >= 4)) {             \
       g_time = clock();                                                        \
       DISPLAY(__VA_ARGS__);                                                    \
     }                                                                          \
   }
+#undef  DISPLAYUPDATE
 #define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
-static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
-static clock_t g_time = 0;
 
 /*-*************************************
 * Hash table
@@ -120,9 +129,9 @@ static int COVER_map_init(COVER_map_t *map, U32 size) {
 /**
  * Internal hash function
  */
-static const U32 prime4bytes = 2654435761U;
+static const U32 COVER_prime4bytes = 2654435761U;
 static U32 COVER_map_hash(COVER_map_t *map, U32 key) {
-  return (key * prime4bytes) >> (32 - map->sizeLog);
+  return (key * COVER_prime4bytes) >> (32 - map->sizeLog);
 }
 
 /**
@@ -215,7 +224,7 @@ typedef struct {
 } COVER_ctx_t;
 
 /* We need a global context for qsort... */
-static COVER_ctx_t *g_ctx = NULL;
+static COVER_ctx_t *g_coverCtx = NULL;
 
 /*-*************************************
 *  Helper functions
@@ -258,11 +267,11 @@ static int COVER_cmp8(COVER_ctx_t *ctx, const void *lp, const void *rp) {
 
 /**
  * Same as COVER_cmp() except ties are broken by pointer value
- * NOTE: g_ctx must be set to call this function.  A global is required because
+ * NOTE: g_coverCtx must be set to call this function.  A global is required because
  * qsort doesn't take an opaque pointer.
  */
-static int COVER_strict_cmp(const void *lp, const void *rp) {
-  int result = COVER_cmp(g_ctx, lp, rp);
+static int WIN_CDECL COVER_strict_cmp(const void *lp, const void *rp) {
+  int result = COVER_cmp(g_coverCtx, lp, rp);
   if (result == 0) {
     result = lp < rp ? -1 : 1;
   }
@@ -271,8 +280,8 @@ static int COVER_strict_cmp(const void *lp, const void *rp) {
 /**
  * Faster version for d <= 8.
  */
-static int COVER_strict_cmp8(const void *lp, const void *rp) {
-  int result = COVER_cmp8(g_ctx, lp, rp);
+static int WIN_CDECL COVER_strict_cmp8(const void *lp, const void *rp) {
+  int result = COVER_cmp8(g_coverCtx, lp, rp);
   if (result == 0) {
     result = lp < rp ? -1 : 1;
   }
@@ -603,7 +612,7 @@ static size_t COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
     /* qsort doesn't take an opaque pointer, so pass as a global.
      * On OpenBSD qsort() is not guaranteed to be stable, their mergesort() is.
      */
-    g_ctx = ctx;
+    g_coverCtx = ctx;
 #if defined(__OpenBSD__)
     mergesort(ctx->suffix, ctx->suffixSize, sizeof(U32),
           (ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
@@ -946,7 +955,7 @@ void COVER_dictSelectionFree(COVER_dictSelection_t selection){
   free(selection.dictContent);
 }
 
-COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
+COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent, size_t dictBufferCapacity,
         size_t dictContentSize, const BYTE* samplesBuffer, const size_t* samplesSizes, unsigned nbFinalizeSamples,
         size_t nbCheckSamples, size_t nbSamples, ZDICT_cover_params_t params, size_t* offsets, size_t totalCompressedSize) {
 
@@ -954,8 +963,8 @@ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
   size_t largestCompressed = 0;
   BYTE* customDictContentEnd = customDictContent + dictContentSize;
 
-  BYTE * largestDictbuffer = (BYTE *)malloc(dictContentSize);
-  BYTE * candidateDictBuffer = (BYTE *)malloc(dictContentSize);
+  BYTE * largestDictbuffer = (BYTE *)malloc(dictBufferCapacity);
+  BYTE * candidateDictBuffer = (BYTE *)malloc(dictBufferCapacity);
   double regressionTolerance = ((double)params.shrinkDictMaxRegression / 100.0) + 1.00;
 
   if (!largestDictbuffer || !candidateDictBuffer) {
@@ -967,7 +976,7 @@ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
   /* Initial dictionary size and compressed size */
   memcpy(largestDictbuffer, customDictContent, dictContentSize);
   dictContentSize = ZDICT_finalizeDictionary(
-    largestDictbuffer, dictContentSize, customDictContent, dictContentSize,
+    largestDictbuffer, dictBufferCapacity, customDictContent, dictContentSize,
     samplesBuffer, samplesSizes, nbFinalizeSamples, params.zParams);
 
   if (ZDICT_isError(dictContentSize)) {
@@ -1001,7 +1010,7 @@ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
   while (dictContentSize < largestDict) {
     memcpy(candidateDictBuffer, largestDictbuffer, largestDict);
     dictContentSize = ZDICT_finalizeDictionary(
-      candidateDictBuffer, dictContentSize, customDictContentEnd - dictContentSize, dictContentSize,
+      candidateDictBuffer, dictBufferCapacity, customDictContentEnd - dictContentSize, dictContentSize,
       samplesBuffer, samplesSizes, nbFinalizeSamples, params.zParams);
 
     if (ZDICT_isError(dictContentSize)) {
@@ -1079,7 +1088,7 @@ static void COVER_tryParameters(void *opaque) {
   {
     const size_t tail = COVER_buildDictionary(ctx, freqs, &activeDmers, dict,
                                               dictBufferCapacity, parameters);
-    selection = COVER_selectDict(dict + tail, dictBufferCapacity - tail,
+    selection = COVER_selectDict(dict + tail, dictBufferCapacity, dictBufferCapacity - tail,
         ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbTrainSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
         totalCompressedSize);
 
@@ -1106,7 +1115,7 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
   /* constants */
   const unsigned nbThreads = parameters->nbThreads;
   const double splitPoint =
-      parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
+      parameters->splitPoint <= 0.0 ? COVER_DEFAULT_SPLITPOINT : parameters->splitPoint;
   const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
   const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
   const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
diff --git a/lib/dictBuilder/cover.h b/lib/dictBuilder/cover.h
index f2aa0e35e3aa..9f1cb5fb9551 100644
--- a/lib/dictBuilder/cover.h
+++ b/lib/dictBuilder/cover.h
@@ -152,6 +152,6 @@ void COVER_dictSelectionFree(COVER_dictSelection_t selection);
  * smallest dictionary within a specified regression of the compressed size
  * from the largest dictionary.
  */
- COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
+ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent, size_t dictBufferCapacity,
                        size_t dictContentSize, const BYTE* samplesBuffer, const size_t* samplesSizes, unsigned nbFinalizeSamples,
                        size_t nbCheckSamples, size_t nbSamples, ZDICT_cover_params_t params, size_t* offsets, size_t totalCompressedSize);
diff --git a/lib/dictBuilder/fastcover.c b/lib/dictBuilder/fastcover.c
index 485c333b54a8..5e60f24c5879 100644
--- a/lib/dictBuilder/fastcover.c
+++ b/lib/dictBuilder/fastcover.c
@@ -21,6 +21,7 @@
 #include "../common/threading.h"
 #include "cover.h"
 #include "../common/zstd_internal.h" /* includes zstd.h */
+#include "../compress/zstd_compress_internal.h" /* ZSTD_hash*() */
 #ifndef ZDICT_STATIC_LINKING_ONLY
 #define ZDICT_STATIC_LINKING_ONLY
 #endif
@@ -33,7 +34,7 @@
 #define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
 #define FASTCOVER_MAX_F 31
 #define FASTCOVER_MAX_ACCEL 10
-#define DEFAULT_SPLITPOINT 0.75
+#define FASTCOVER_DEFAULT_SPLITPOINT 0.75
 #define DEFAULT_F 20
 #define DEFAULT_ACCEL 1
 
@@ -41,50 +42,50 @@
 /*-*************************************
 *  Console display
 ***************************************/
+#ifndef LOCALDISPLAYLEVEL
 static int g_displayLevel = 2;
+#endif
+#undef  DISPLAY
 #define DISPLAY(...)                                                           \
   {                                                                            \
     fprintf(stderr, __VA_ARGS__);                                              \
     fflush(stderr);                                                            \
   }
+#undef  LOCALDISPLAYLEVEL
 #define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
   if (displayLevel >= l) {                                                     \
     DISPLAY(__VA_ARGS__);                                                      \
   } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+#undef  DISPLAYLEVEL
 #define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
 
+#ifndef LOCALDISPLAYUPDATE
+static const clock_t g_refreshRate = CLOCKS_PER_SEC * 15 / 100;
+static clock_t g_time = 0;
+#endif
+#undef  LOCALDISPLAYUPDATE
 #define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
   if (displayLevel >= l) {                                                     \
-    if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
+    if ((clock() - g_time > g_refreshRate) || (displayLevel >= 4)) {             \
       g_time = clock();                                                        \
       DISPLAY(__VA_ARGS__);                                                    \
     }                                                                          \
   }
+#undef  DISPLAYUPDATE
 #define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
-static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
-static clock_t g_time = 0;
 
 
 /*-*************************************
 * Hash Functions
 ***************************************/
-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 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); }
-
-
 /**
- * Hash the d-byte value pointed to by p and mod 2^f
+ * Hash the d-byte value pointed to by p and mod 2^f into the frequency vector
  */
-static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 h, unsigned d) {
+static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 f, unsigned d) {
   if (d == 6) {
-    return ZSTD_hash6Ptr(p, h) & ((1 << h) - 1);
+    return ZSTD_hash6Ptr(p, f);
   }
-  return ZSTD_hash8Ptr(p, h) & ((1 << h) - 1);
+  return ZSTD_hash8Ptr(p, f);
 }
 
 
@@ -486,7 +487,7 @@ static void FASTCOVER_tryParameters(void *opaque)
                                                     parameters, segmentFreqs);
 
     const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
-    selection = COVER_selectDict(dict + tail, dictBufferCapacity - tail,
+    selection = COVER_selectDict(dict + tail, dictBufferCapacity, dictBufferCapacity - tail,
          ctx->samples, ctx->samplesSizes, nbFinalizeSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
          totalCompressedSize);
 
@@ -617,7 +618,7 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
     /* constants */
     const unsigned nbThreads = parameters->nbThreads;
     const double splitPoint =
-        parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
+        parameters->splitPoint <= 0.0 ? FASTCOVER_DEFAULT_SPLITPOINT : parameters->splitPoint;
     const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
     const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
     const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
diff --git a/lib/dictBuilder/zdict.c b/lib/dictBuilder/zdict.c
index 6d0b04231bb5..79c522ef3679 100644
--- a/lib/dictBuilder/zdict.c
+++ b/lib/dictBuilder/zdict.c
@@ -62,14 +62,15 @@
 
 #define NOISELENGTH 32
 
-static const int g_compressionLevel_default = 3;
 static const U32 g_selectivity_default = 9;
 
 
 /*-*************************************
 *  Console display
 ***************************************/
+#undef  DISPLAY
 #define DISPLAY(...)         { fprintf(stderr, __VA_ARGS__); fflush( stderr ); }
+#undef  DISPLAYLEVEL
 #define DISPLAYLEVEL(l, ...) if (notificationLevel>=l) { DISPLAY(__VA_ARGS__); }    /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
 
 static clock_t ZDICT_clockSpan(clock_t nPrevious) { return clock() - nPrevious; }
@@ -105,20 +106,17 @@ size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize)
     size_t headerSize;
     if (dictSize <= 8 || MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return ERROR(dictionary_corrupted);
 
-    {   unsigned offcodeMaxValue = MaxOff;
-        ZSTD_compressedBlockState_t* bs = (ZSTD_compressedBlockState_t*)malloc(sizeof(ZSTD_compressedBlockState_t));
+    {   ZSTD_compressedBlockState_t* bs = (ZSTD_compressedBlockState_t*)malloc(sizeof(ZSTD_compressedBlockState_t));
         U32* wksp = (U32*)malloc(HUF_WORKSPACE_SIZE);
-        short* offcodeNCount = (short*)malloc((MaxOff+1)*sizeof(short));
-        if (!bs || !wksp || !offcodeNCount) {
+        if (!bs || !wksp) {
             headerSize = ERROR(memory_allocation);
         } else {
             ZSTD_reset_compressedBlockState(bs);
-            headerSize = ZSTD_loadCEntropy(bs, wksp, offcodeNCount, &offcodeMaxValue, dictBuffer, dictSize);
+            headerSize = ZSTD_loadCEntropy(bs, wksp, dictBuffer, dictSize);
         }
 
         free(bs);
         free(wksp);
-        free(offcodeNCount);
     }
 
     return headerSize;
@@ -532,6 +530,7 @@ static size_t ZDICT_trainBuffer_legacy(dictItem* dictList, U32 dictListSize,
     clock_t displayClock = 0;
     clock_t const refreshRate = CLOCKS_PER_SEC * 3 / 10;
 
+#   undef  DISPLAYUPDATE
 #   define DISPLAYUPDATE(l, ...) if (notificationLevel>=l) { \
             if (ZDICT_clockSpan(displayClock) > refreshRate)  \
             { displayClock = clock(); DISPLAY(__VA_ARGS__); \
@@ -706,7 +705,7 @@ static void ZDICT_flatLit(unsigned* countLit)
 
 #define OFFCODE_MAX 30  /* only applicable to first block */
 static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
-                                   unsigned compressionLevel,
+                                   int compressionLevel,
                              const void*  srcBuffer, const size_t* fileSizes, unsigned nbFiles,
                              const void* dictBuffer, size_t  dictBufferSize,
                                    unsigned notificationLevel)
@@ -741,7 +740,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     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 = ZSTD_CLEVEL_DEFAULT;
     params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
 
     esr.dict = ZSTD_createCDict_advanced(dictBuffer, dictBufferSize, ZSTD_dlm_byRef, ZSTD_dct_rawContent, params.cParams, ZSTD_defaultCMem);
@@ -786,7 +785,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     /* note : the result of this phase should be used to better appreciate the impact on statistics */
 
     total=0; for (u=0; u<=offcodeMax; u++) total+=offcodeCount[u];
-    errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax);
+    errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax, /* useLowProbCount */ 1);
     if (FSE_isError(errorCode)) {
         eSize = errorCode;
         DISPLAYLEVEL(1, "FSE_normalizeCount error with offcodeCount \n");
@@ -795,7 +794,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     Offlog = (U32)errorCode;
 
     total=0; for (u=0; u<=MaxML; u++) total+=matchLengthCount[u];
-    errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML);
+    errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML, /* useLowProbCount */ 1);
     if (FSE_isError(errorCode)) {
         eSize = errorCode;
         DISPLAYLEVEL(1, "FSE_normalizeCount error with matchLengthCount \n");
@@ -804,7 +803,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     mlLog = (U32)errorCode;
 
     total=0; for (u=0; u<=MaxLL; u++) total+=litLengthCount[u];
-    errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL);
+    errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL, /* useLowProbCount */ 1);
     if (FSE_isError(errorCode)) {
         eSize = errorCode;
         DISPLAYLEVEL(1, "FSE_normalizeCount error with litLengthCount \n");
@@ -893,7 +892,7 @@ size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
     size_t hSize;
 #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;
+    int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
     U32 const notificationLevel = params.notificationLevel;
 
     /* check conditions */
@@ -939,7 +938,7 @@ static size_t ZDICT_addEntropyTablesFromBuffer_advanced(
         const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
         ZDICT_params_t params)
 {
-    int const compressionLevel = (params.compressionLevel == 0) ? g_compressionLevel_default : params.compressionLevel;
+    int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
     U32 const notificationLevel = params.notificationLevel;
     size_t hSize = 8;
 
@@ -1114,8 +1113,8 @@ size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
     memset(&params, 0, sizeof(params));
     params.d = 8;
     params.steps = 4;
-    /* Default to level 6 since no compression level information is available */
-    params.zParams.compressionLevel = 3;
+    /* Use default level since no compression level information is available */
+    params.zParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
 #if defined(DEBUGLEVEL) && (DEBUGLEVEL>=1)
     params.zParams.notificationLevel = DEBUGLEVEL;
 #endif
diff --git a/lib/dictBuilder/zdict.h b/lib/dictBuilder/zdict.h
index ff2e77faf773..b782993f9d2e 100644
--- a/lib/dictBuilder/zdict.h
+++ b/lib/dictBuilder/zdict.h
@@ -279,7 +279,7 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_legacy(
 #  define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
 #  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
 #    define ZDICT_DEPRECATED(message) [[deprecated(message)]] ZDICTLIB_API
-#  elif (ZDICT_GCC_VERSION >= 405) || defined(__clang__)
+#  elif defined(__clang__) || (ZDICT_GCC_VERSION >= 405)
 #    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated(message)))
 #  elif (ZDICT_GCC_VERSION >= 301)
 #    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated))
diff --git a/lib/legacy/zstd_v01.c b/lib/legacy/zstd_v01.c
index eb23628330e7..13115bec5f81 100644
--- a/lib/legacy/zstd_v01.c
+++ b/lib/legacy/zstd_v01.c
@@ -1280,7 +1280,11 @@ static size_t HUF_decompress (void* dst, size_t maxDstSize, const void* cSrc, si
 *  Basic Types
 *********************************************************/
 #if defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
-# include <stdint.h>
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
 typedef  uint8_t BYTE;
 typedef uint16_t U16;
 typedef  int16_t S16;
diff --git a/lib/legacy/zstd_v02.c b/lib/legacy/zstd_v02.c
index 32d45a6d4bd3..9abb6d03390f 100644
--- a/lib/legacy/zstd_v02.c
+++ b/lib/legacy/zstd_v02.c
@@ -89,7 +89,11 @@ extern "C" {
 *  Basic Types
 *****************************************************************/
 #if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
-# include <stdint.h>
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
   typedef  uint8_t BYTE;
   typedef uint16_t U16;
   typedef  int16_t S16;
diff --git a/lib/legacy/zstd_v03.c b/lib/legacy/zstd_v03.c
index b541eae2abb7..a19cb205a56c 100644
--- a/lib/legacy/zstd_v03.c
+++ b/lib/legacy/zstd_v03.c
@@ -90,7 +90,11 @@ extern "C" {
 *  Basic Types
 *****************************************************************/
 #if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
-# include <stdint.h>
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
   typedef  uint8_t BYTE;
   typedef uint16_t U16;
   typedef  int16_t S16;
diff --git a/lib/legacy/zstd_v04.c b/lib/legacy/zstd_v04.c
index 56bf45222498..77d52555b313 100644
--- a/lib/legacy/zstd_v04.c
+++ b/lib/legacy/zstd_v04.c
@@ -52,7 +52,11 @@ extern "C" {
 *  Basic Types
 *****************************************************************/
 #if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
-# include <stdint.h>
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
   typedef  uint8_t BYTE;
   typedef uint16_t U16;
   typedef  int16_t S16;
@@ -74,7 +78,7 @@ extern "C" {
 /*-*************************************
 *  Debug
 ***************************************/
-#include "debug.h"
+#include "../common/debug.h"
 #ifndef assert
 #  define assert(condition) ((void)0)
 #endif
diff --git a/lib/legacy/zstd_v05.c b/lib/legacy/zstd_v05.c
index 243d222562e3..ca8d5c9bf03c 100644
--- a/lib/legacy/zstd_v05.c
+++ b/lib/legacy/zstd_v05.c
@@ -80,7 +80,11 @@ extern "C" {
 *  Basic Types
 *****************************************************************/
 #if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
-# include <stdint.h>
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
   typedef  uint8_t BYTE;
   typedef uint16_t U16;
   typedef  int16_t S16;
diff --git a/lib/legacy/zstd_v06.c b/lib/legacy/zstd_v06.c
index c56f582772f8..c4ac7dba8eb3 100644
--- a/lib/legacy/zstd_v06.c
+++ b/lib/legacy/zstd_v06.c
@@ -82,7 +82,11 @@ extern "C" {
 *  Basic Types
 *****************************************************************/
 #if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
-# include <stdint.h>
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
   typedef  uint8_t BYTE;
   typedef uint16_t U16;
   typedef  int16_t S16;
diff --git a/lib/legacy/zstd_v07.c b/lib/legacy/zstd_v07.c
index 9f3a597f1254..049ba4749409 100644
--- a/lib/legacy/zstd_v07.c
+++ b/lib/legacy/zstd_v07.c
@@ -242,7 +242,11 @@ extern "C" {
 *  Basic Types
 *****************************************************************/
 #if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
-# include <stdint.h>
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
   typedef  uint8_t BYTE;
   typedef uint16_t U16;
   typedef  int16_t S16;
diff --git a/lib/libzstd.pc.in b/lib/libzstd.pc.in
index 8ec0235ad9b8..8465c9770681 100644
--- a/lib/libzstd.pc.in
+++ b/lib/libzstd.pc.in
@@ -3,9 +3,9 @@
 #   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
 
 prefix=@PREFIX@
-exec_prefix=${prefix}
-includedir=${prefix}/@INCLUDEDIR@
-libdir=${exec_prefix}/@LIBDIR@
+exec_prefix=@EXEC_PREFIX@
+includedir=@INCLUDEDIR@
+libdir=@LIBDIR@
 
 Name: zstd
 Description: fast lossless compression algorithm library
diff --git a/lib/zstd.h b/lib/zstd.h
index 8c6fc6ae90e6..b0ecdf55385c 100644
--- a/lib/zstd.h
+++ b/lib/zstd.h
@@ -72,16 +72,21 @@ extern "C" {
 /*------   Version   ------*/
 #define ZSTD_VERSION_MAJOR    1
 #define ZSTD_VERSION_MINOR    4
-#define ZSTD_VERSION_RELEASE  5
-
+#define ZSTD_VERSION_RELEASE  8
 #define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
-ZSTDLIB_API unsigned ZSTD_versionNumber(void);   /**< to check runtime library version */
+
+/*! ZSTD_versionNumber() :
+ *  Return runtime library version, the value is (MAJOR*100*100 + MINOR*100 + RELEASE). */
+ZSTDLIB_API unsigned ZSTD_versionNumber(void);
 
 #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)
-ZSTDLIB_API const char* ZSTD_versionString(void);   /* requires v1.3.0+ */
+
+/*! ZSTD_versionString() :
+ *  Return runtime library version, like "1.4.5". Requires v1.3.0+. */
+ZSTDLIB_API const char* ZSTD_versionString(void);
 
 /* *************************************
  *  Default constant
@@ -334,7 +339,9 @@ typedef enum {
                                      * for large inputs, by finding large matches at long distance.
                                      * It increases memory usage and window size.
                                      * Note: enabling this parameter increases default ZSTD_c_windowLog to 128 MB
-                                     * except when expressly set to a different value. */
+                                     * except when expressly set to a different value.
+                                     * Note: will be enabled by default if ZSTD_c_windowLog >= 128 MB and
+                                     * compression strategy >= ZSTD_btopt (== compression level 16+) */
     ZSTD_c_ldmHashLog=161,   /* 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.
@@ -365,16 +372,20 @@ typedef enum {
     ZSTD_c_dictIDFlag=202,   /* 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 (compiled with build macro ZSTD_MULTITHREAD).
-     * They return an error otherwise. */
+    /* These parameters are only active if multi-threading is enabled (compiled with build macro ZSTD_MULTITHREAD).
+     * Otherwise, trying to set any other value than default (0) will be a no-op and return an error.
+     * In a situation where it's unknown if the linked library supports multi-threading or not,
+     * setting ZSTD_c_nbWorkers to any value >= 1 and consulting the return value provides a quick way to check this property.
+     */
     ZSTD_c_nbWorkers=400,    /* Select how many threads will be spawned to compress in parallel.
-                              * When nbWorkers >= 1, triggers asynchronous mode when used with ZSTD_compressStream*() :
+                              * When nbWorkers >= 1, triggers asynchronous mode when invoking ZSTD_compressStream*() :
                               * ZSTD_compressStream*() consumes input and flush output if possible, but immediately gives back control to caller,
-                              * while compression work is performed in parallel, within worker threads.
+                              * while compression is performed in parallel, within worker thread(s).
                               * (note : a strong exception to this rule is when first invocation of ZSTD_compressStream2() sets ZSTD_e_end :
                               *  in which case, ZSTD_compressStream2() delegates to ZSTD_compress2(), which is always 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 */
+                              * Default value is `0`, aka "single-threaded mode" : no worker is spawned,
+                              * compression is performed inside Caller's thread, and all invocations are blocking */
     ZSTD_c_jobSize=401,      /* Size of a compression job. This value is enforced only when nbWorkers >= 1.
                               * Each compression job is completed in parallel, so this value can indirectly impact the nb of active threads.
                               * 0 means default, which is dynamically determined based on compression parameters.
@@ -403,6 +414,11 @@ typedef enum {
      * ZSTD_c_literalCompressionMode
      * ZSTD_c_targetCBlockSize
      * ZSTD_c_srcSizeHint
+     * ZSTD_c_enableDedicatedDictSearch
+     * ZSTD_c_stableInBuffer
+     * ZSTD_c_stableOutBuffer
+     * ZSTD_c_blockDelimiters
+     * ZSTD_c_validateSequences
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly;
      *        also, the enums values themselves are unstable and can still change.
@@ -413,7 +429,12 @@ typedef enum {
      ZSTD_c_experimentalParam4=1001,
      ZSTD_c_experimentalParam5=1002,
      ZSTD_c_experimentalParam6=1003,
-     ZSTD_c_experimentalParam7=1004
+     ZSTD_c_experimentalParam7=1004,
+     ZSTD_c_experimentalParam8=1005,
+     ZSTD_c_experimentalParam9=1006,
+     ZSTD_c_experimentalParam10=1007,
+     ZSTD_c_experimentalParam11=1008,
+     ZSTD_c_experimentalParam12=1009
 } ZSTD_cParameter;
 
 typedef struct {
@@ -524,11 +545,13 @@ typedef enum {
      * At the time of this writing, they include :
      * ZSTD_d_format
      * ZSTD_d_stableOutBuffer
+     * ZSTD_d_forceIgnoreChecksum
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly
      */
      ZSTD_d_experimentalParam1=1000,
-     ZSTD_d_experimentalParam2=1001
+     ZSTD_d_experimentalParam2=1001,
+     ZSTD_d_experimentalParam3=1002
 
 } ZSTD_dParameter;
 
@@ -664,8 +687,9 @@ typedef enum {
  *  - Compression parameters cannot be changed once compression is started (save a list of exceptions in multi-threading mode)
  *  - output->pos must be <= dstCapacity, input->pos must be <= srcSize
  *  - output->pos and input->pos will be updated. They are guaranteed to remain below their respective limit.
+ *  - endOp must be a valid directive
  *  - When nbWorkers==0 (default), function is blocking : it completes its job before returning to caller.
- *  - When nbWorkers>=1, function is non-blocking : it just acquires a copy of input, and distributes jobs to internal worker threads, flush whatever is available,
+ *  - When nbWorkers>=1, function is non-blocking : it copies a portion of input, distributes jobs to internal worker threads, flush to output whatever is available,
  *                                                  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 : if the first call requests a ZSTD_e_end directive and provides enough dstCapacity, the function delegates to ZSTD_compress2() which is always blocking.
@@ -1100,21 +1124,40 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 typedef struct ZSTD_CCtx_params_s ZSTD_CCtx_params;
 
 typedef struct {
-    unsigned int matchPos; /* Match pos in dst */
-    /* If seqDef.offset > 3, then this is seqDef.offset - 3
-     * If seqDef.offset < 3, then this is the corresponding repeat offset
-     * But if seqDef.offset < 3 and litLength == 0, this is the
-     *   repeat offset before the corresponding repeat offset
-     * And if seqDef.offset == 3 and litLength == 0, this is the
-     *   most recent repeat offset - 1
-     */
-    unsigned int offset;
-    unsigned int litLength; /* Literal length */
-    unsigned int matchLength; /* Match length */
-    /* 0 when seq not rep and seqDef.offset otherwise
-     * when litLength == 0 this will be <= 4, otherwise <= 3 like normal
-     */
-    unsigned int rep;
+    unsigned int offset;      /* The offset of the match. (NOT the same as the offset code)
+                               * If offset == 0 and matchLength == 0, this sequence represents the last
+                               * literals in the block of litLength size.
+                               */
+
+    unsigned int litLength;   /* Literal length of the sequence. */
+    unsigned int matchLength; /* Match length of the sequence. */
+
+                              /* Note: Users of this API may provide a sequence with matchLength == litLength == offset == 0.
+                               * In this case, we will treat the sequence as a marker for a block boundary.
+                               */
+
+    unsigned int rep;         /* Represents which repeat offset is represented by the field 'offset'.
+                               * Ranges from [0, 3].
+                               *
+                               * Repeat offsets are essentially previous offsets from previous sequences sorted in
+                               * recency order. For more detail, see doc/zstd_compression_format.md
+                               *
+                               * If rep == 0, then 'offset' does not contain a repeat offset.
+                               * If rep > 0:
+                               *  If litLength != 0:
+                               *      rep == 1 --> offset == repeat_offset_1
+                               *      rep == 2 --> offset == repeat_offset_2
+                               *      rep == 3 --> offset == repeat_offset_3
+                               *  If litLength == 0:
+                               *      rep == 1 --> offset == repeat_offset_2
+                               *      rep == 2 --> offset == repeat_offset_3
+                               *      rep == 3 --> offset == repeat_offset_1 - 1
+                               *
+                               * Note: This field is optional. ZSTD_generateSequences() will calculate the value of
+                               * 'rep', but repeat offsets do not necessarily need to be calculated from an external
+                               * sequence provider's perspective. For example, ZSTD_compressSequences() does not
+                               * use this 'rep' field at all (as of now).
+                               */
 } ZSTD_Sequence;
 
 typedef struct {
@@ -1157,6 +1200,12 @@ typedef enum {
 } ZSTD_format_e;
 
 typedef enum {
+    /* Note: this enum controls ZSTD_d_forceIgnoreChecksum */
+    ZSTD_d_validateChecksum = 0,
+    ZSTD_d_ignoreChecksum = 1
+} ZSTD_forceIgnoreChecksum_e;
+
+typedef enum {
     /* Note: this enum and the behavior it controls are effectively internal
      * implementation details of the compressor. They are expected to continue
      * to evolve and should be considered only in the context of extremely
@@ -1253,14 +1302,74 @@ ZSTDLIB_API unsigned long long ZSTD_decompressBound(const void* src, size_t srcS
  *           or an error code (if srcSize is too small) */
 ZSTDLIB_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
 
-/*! ZSTD_getSequences() :
- * Extract sequences from the sequence store
+typedef enum {
+  ZSTD_sf_noBlockDelimiters = 0,         /* Representation of ZSTD_Sequence has no block delimiters, sequences only */
+  ZSTD_sf_explicitBlockDelimiters = 1    /* Representation of ZSTD_Sequence contains explicit block delimiters */
+} ZSTD_sequenceFormat_e;
+
+/*! ZSTD_generateSequences() :
+ * Generate sequences using ZSTD_compress2, given a source buffer.
+ *
+ * Each block will end with a dummy sequence
+ * with offset == 0, matchLength == 0, and litLength == length of last literals.
+ * litLength may be == 0, and if so, then the sequence of (of: 0 ml: 0 ll: 0)
+ * simply acts as a block delimiter.
+ *
  * zc can be used to insert custom compression params.
  * This function invokes ZSTD_compress2
- * @return : number of sequences extracted
+ *
+ * The output of this function can be fed into ZSTD_compressSequences() with CCtx
+ * setting of ZSTD_c_blockDelimiters as ZSTD_sf_explicitBlockDelimiters
+ * @return : number of sequences generated
+ */
+
+ZSTDLIB_API size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
+                                          size_t outSeqsSize, const void* src, size_t srcSize);
+
+/*! ZSTD_mergeBlockDelimiters() :
+ * Given an array of ZSTD_Sequence, remove all sequences that represent block delimiters/last literals
+ * by merging them into into the literals of the next sequence.
+ *
+ * As such, the final generated result has no explicit representation of block boundaries,
+ * and the final last literals segment is not represented in the sequences.
+ *
+ * The output of this function can be fed into ZSTD_compressSequences() with CCtx
+ * setting of ZSTD_c_blockDelimiters as ZSTD_sf_noBlockDelimiters
+ * @return : number of sequences left after merging
+ */
+ZSTDLIB_API size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize);
+
+/*! ZSTD_compressSequences() :
+ * Compress an array of ZSTD_Sequence, generated from the original source buffer, into dst.
+ * If a dictionary is included, then the cctx should reference the dict. (see: ZSTD_CCtx_refCDict(), ZSTD_CCtx_loadDictionary(), etc.)
+ * The entire source is compressed into a single frame.
+ *
+ * The compression behavior changes based on cctx params. In particular:
+ *    If ZSTD_c_blockDelimiters == ZSTD_sf_noBlockDelimiters, the array of ZSTD_Sequence is expected to contain
+ *    no block delimiters (defined in ZSTD_Sequence). Block boundaries are roughly determined based on
+ *    the block size derived from the cctx, and sequences may be split. This is the default setting.
+ *
+ *    If ZSTD_c_blockDelimiters == ZSTD_sf_explicitBlockDelimiters, the array of ZSTD_Sequence is expected to contain
+ *    block delimiters (defined in ZSTD_Sequence). Behavior is undefined if no block delimiters are provided.
+ *
+ *    If ZSTD_c_validateSequences == 0, this function will blindly accept the sequences provided. Invalid sequences cause undefined
+ *    behavior. If ZSTD_c_validateSequences == 1, then if sequence is invalid (see doc/zstd_compression_format.md for
+ *    specifics regarding offset/matchlength requirements) then the function will bail out and return an error.
+ *
+ *    In addition to the two adjustable experimental params, there are other important cctx params.
+ *    - ZSTD_c_minMatch MUST be set as less than or equal to the smallest match generated by the match finder. It has a minimum value of ZSTD_MINMATCH_MIN.
+ *    - ZSTD_c_compressionLevel accordingly adjusts the strength of the entropy coder, as it would in typical compression.
+ *    - ZSTD_c_windowLog affects offset validation: this function will return an error at higher debug levels if a provided offset
+ *      is larger than what the spec allows for a given window log and dictionary (if present). See: doc/zstd_compression_format.md
+ *
+ * Note: Repcodes are, as of now, always re-calculated within this function, so ZSTD_Sequence::rep is unused.
+ * Note 2: Once we integrate ability to ingest repcodes, the explicit block delims mode must respect those repcodes exactly,
+ *         and cannot emit an RLE block that disagrees with the repcode history
+ * @return : final compressed size or a ZSTD error.
  */
-ZSTDLIB_API size_t ZSTD_getSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
-    size_t outSeqsSize, const void* src, size_t srcSize);
+ZSTDLIB_API size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstSize,
+                                  const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                                  const void* src, size_t srcSize);
 
 
 /***************************************
@@ -1372,7 +1481,11 @@ ZSTDLIB_API const ZSTD_DDict* ZSTD_initStaticDDict(
 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 */
+static
+#ifdef __GNUC__
+__attribute__((__unused__))
+#endif
+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);
@@ -1385,13 +1498,36 @@ ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictS
                                                   ZSTD_compressionParameters cParams,
                                                   ZSTD_customMem customMem);
 
+/* ! Thread pool :
+ * These prototypes make it possible to share a thread pool among multiple compression contexts.
+ * This can limit resources for applications with multiple threads where each one uses
+ * a threaded compression mode (via ZSTD_c_nbWorkers parameter).
+ * ZSTD_createThreadPool creates a new thread pool with a given number of threads.
+ * Note that the lifetime of such pool must exist while being used.
+ * ZSTD_CCtx_refThreadPool assigns a thread pool to a context (use NULL argument value
+ * to use an internal thread pool).
+ * ZSTD_freeThreadPool frees a thread pool.
+ */
+typedef struct POOL_ctx_s ZSTD_threadPool;
+ZSTDLIB_API ZSTD_threadPool* ZSTD_createThreadPool(size_t numThreads);
+ZSTDLIB_API void ZSTD_freeThreadPool (ZSTD_threadPool* pool);
+ZSTDLIB_API size_t ZSTD_CCtx_refThreadPool(ZSTD_CCtx* cctx, ZSTD_threadPool* pool);
+
+/*
+ * This API is temporary and is expected to change or disappear in the future!
+ */
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced2(
+    const void* dict, size_t dictSize,
+    ZSTD_dictLoadMethod_e dictLoadMethod,
+    ZSTD_dictContentType_e dictContentType,
+    const ZSTD_CCtx_params* cctxParams,
+    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
 ***************************************/
@@ -1404,6 +1540,12 @@ ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictS
  *  note: equivalent to ZSTD_createCDict_advanced(), with dictLoadMethod==ZSTD_dlm_byRef */
 ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
 
+/*! ZSTD_getDictID_fromCDict() :
+ *  Provides the dictID of the dictionary loaded into `cdict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromCDict(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 */
@@ -1518,6 +1660,143 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
  * but compression ratio may regress significantly if guess considerably underestimates */
 #define ZSTD_c_srcSizeHint ZSTD_c_experimentalParam7
 
+/* Controls whether the new and experimental "dedicated dictionary search
+ * structure" can be used. This feature is still rough around the edges, be
+ * prepared for surprising behavior!
+ *
+ * How to use it:
+ *
+ * When using a CDict, whether to use this feature or not is controlled at
+ * CDict creation, and it must be set in a CCtxParams set passed into that
+ * construction (via ZSTD_createCDict_advanced2()). A compression will then
+ * use the feature or not based on how the CDict was constructed; the value of
+ * this param, set in the CCtx, will have no effect.
+ *
+ * However, when a dictionary buffer is passed into a CCtx, such as via
+ * ZSTD_CCtx_loadDictionary(), this param can be set on the CCtx to control
+ * whether the CDict that is created internally can use the feature or not.
+ *
+ * What it does:
+ *
+ * Normally, the internal data structures of the CDict are analogous to what
+ * would be stored in a CCtx after compressing the contents of a dictionary.
+ * To an approximation, a compression using a dictionary can then use those
+ * data structures to simply continue what is effectively a streaming
+ * compression where the simulated compression of the dictionary left off.
+ * Which is to say, the search structures in the CDict are normally the same
+ * format as in the CCtx.
+ *
+ * It is possible to do better, since the CDict is not like a CCtx: the search
+ * structures are written once during CDict creation, and then are only read
+ * after that, while the search structures in the CCtx are both read and
+ * written as the compression goes along. This means we can choose a search
+ * structure for the dictionary that is read-optimized.
+ *
+ * This feature enables the use of that different structure.
+ *
+ * Note that some of the members of the ZSTD_compressionParameters struct have
+ * different semantics and constraints in the dedicated search structure. It is
+ * highly recommended that you simply set a compression level in the CCtxParams
+ * you pass into the CDict creation call, and avoid messing with the cParams
+ * directly.
+ *
+ * Effects:
+ *
+ * This will only have any effect when the selected ZSTD_strategy
+ * implementation supports this feature. Currently, that's limited to
+ * ZSTD_greedy, ZSTD_lazy, and ZSTD_lazy2.
+ *
+ * Note that this means that the CDict tables can no longer be copied into the
+ * CCtx, so the dict attachment mode ZSTD_dictForceCopy will no longer be
+ * useable. The dictionary can only be attached or reloaded.
+ *
+ * In general, you should expect compression to be faster--sometimes very much
+ * so--and CDict creation to be slightly slower. Eventually, we will probably
+ * make this mode the default.
+ */
+#define ZSTD_c_enableDedicatedDictSearch ZSTD_c_experimentalParam8
+
+/* ZSTD_c_stableInBuffer
+ * Experimental parameter.
+ * Default is 0 == disabled. Set to 1 to enable.
+ *
+ * Tells the compressor that the ZSTD_inBuffer will ALWAYS be the same
+ * between calls, except for the modifications that zstd makes to pos (the
+ * caller must not modify pos). This is checked by the compressor, and
+ * compression will fail if it ever changes. This means the only flush
+ * mode that makes sense is ZSTD_e_end, so zstd will error if ZSTD_e_end
+ * is not used. The data in the ZSTD_inBuffer in the range [src, src + pos)
+ * MUST not be modified during compression or you will get data corruption.
+ *
+ * When this flag is enabled zstd won't allocate an input window buffer,
+ * because the user guarantees it can reference the ZSTD_inBuffer until
+ * the frame is complete. But, it will still allocate an output buffer
+ * large enough to fit a block (see ZSTD_c_stableOutBuffer). This will also
+ * avoid the memcpy() from the input buffer to the input window buffer.
+ *
+ * NOTE: ZSTD_compressStream2() will error if ZSTD_e_end is not used.
+ * That means this flag cannot be used with ZSTD_compressStream().
+ *
+ * NOTE: So long as the ZSTD_inBuffer always points to valid memory, using
+ * this flag is ALWAYS memory safe, and will never access out-of-bounds
+ * memory. However, compression WILL fail if you violate the preconditions.
+ *
+ * WARNING: The data in the ZSTD_inBuffer in the range [dst, dst + pos) MUST
+ * not be modified during compression or you will get data corruption. This
+ * is because zstd needs to reference data in the ZSTD_inBuffer to find
+ * matches. Normally zstd maintains its own window buffer for this purpose,
+ * but passing this flag tells zstd to use the user provided buffer.
+ */
+#define ZSTD_c_stableInBuffer ZSTD_c_experimentalParam9
+
+/* ZSTD_c_stableOutBuffer
+ * Experimental parameter.
+ * Default is 0 == disabled. Set to 1 to enable.
+ *
+ * Tells he compressor that the ZSTD_outBuffer will not be resized between
+ * calls. Specifically: (out.size - out.pos) will never grow. This gives the
+ * compressor the freedom to say: If the compressed data doesn't fit in the
+ * output buffer then return ZSTD_error_dstSizeTooSmall. This allows us to
+ * always decompress directly into the output buffer, instead of decompressing
+ * into an internal buffer and copying to the output buffer.
+ *
+ * When this flag is enabled zstd won't allocate an output buffer, because
+ * it can write directly to the ZSTD_outBuffer. It will still allocate the
+ * input window buffer (see ZSTD_c_stableInBuffer).
+ *
+ * Zstd will check that (out.size - out.pos) never grows and return an error
+ * if it does. While not strictly necessary, this should prevent surprises.
+ */
+#define ZSTD_c_stableOutBuffer ZSTD_c_experimentalParam10
+
+/* ZSTD_c_blockDelimiters
+ * Default is 0 == ZSTD_sf_noBlockDelimiters.
+ *
+ * For use with sequence compression API: ZSTD_compressSequences().
+ *
+ * Designates whether or not the given array of ZSTD_Sequence contains block delimiters
+ * and last literals, which are defined as sequences with offset == 0 and matchLength == 0.
+ * See the definition of ZSTD_Sequence for more specifics.
+ */
+#define ZSTD_c_blockDelimiters ZSTD_c_experimentalParam11
+
+/* ZSTD_c_validateSequences
+ * Default is 0 == disabled. Set to 1 to enable sequence validation.
+ *
+ * For use with sequence compression API: ZSTD_compressSequences().
+ * Designates whether or not we validate sequences provided to ZSTD_compressSequences()
+ * during function execution.
+ *
+ * Without validation, providing a sequence that does not conform to the zstd spec will cause
+ * undefined behavior, and may produce a corrupted block.
+ *
+ * With validation enabled, a if sequence is invalid (see doc/zstd_compression_format.md for
+ * specifics regarding offset/matchlength requirements) then the function will bail out and
+ * return an error.
+ *
+ */
+#define ZSTD_c_validateSequences ZSTD_c_experimentalParam12
+
 /*! ZSTD_CCtx_getParameter() :
  *  Get the requested compression parameter value, selected by enum ZSTD_cParameter,
  *  and store it into int* value.
@@ -1566,8 +1845,10 @@ ZSTDLIB_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, Z
 /*! ZSTD_CCtxParams_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().
- * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Parameters must be applied to a ZSTD_CCtx using
+ *  ZSTD_CCtx_setParametersUsingCCtxParams().
+ * @result : a code representing success or failure (which can be tested with
+ *           ZSTD_isError()).
  */
 ZSTDLIB_API size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, int value);
 
@@ -1647,6 +1928,13 @@ ZSTDLIB_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* pre
  */
 ZSTDLIB_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize);
 
+/*! ZSTD_DCtx_getParameter() :
+ *  Get the requested decompression parameter value, selected by enum ZSTD_dParameter,
+ *  and store it into int* value.
+ * @return : 0, or an error code (which can be tested with ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value);
+
 /* ZSTD_d_format
  * experimental parameter,
  * allowing selection between ZSTD_format_e input compression formats
@@ -1684,6 +1972,17 @@ ZSTDLIB_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowS
  */
 #define ZSTD_d_stableOutBuffer ZSTD_d_experimentalParam2
 
+/* ZSTD_d_forceIgnoreChecksum
+ * Experimental parameter.
+ * Default is 0 == disabled. Set to 1 to enable
+ *
+ * Tells the decompressor to skip checksum validation during decompression, regardless
+ * of whether checksumming was specified during compression. This offers some
+ * slight performance benefits, and may be useful for debugging.
+ * Param has values of type ZSTD_forceIgnoreChecksum_e
+ */
+#define ZSTD_d_forceIgnoreChecksum ZSTD_d_experimentalParam3
+
 /*! 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,
@@ -1711,7 +2010,8 @@ ZSTDLIB_API size_t ZSTD_decompressStream_simpleArgs (
 ********************************************************************/
 
 /*=====   Advanced Streaming compression functions  =====*/
-/**! ZSTD_initCStream_srcSize() :
+
+/*! ZSTD_initCStream_srcSize() :
  * This function is deprecated, and equivalent to:
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
  *     ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
@@ -1728,7 +2028,7 @@ ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
                          int compressionLevel,
                          unsigned long long pledgedSrcSize);
 
-/**! ZSTD_initCStream_usingDict() :
+/*! ZSTD_initCStream_usingDict() :
  * This function is deprecated, and is equivalent to:
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
  *     ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
@@ -1745,7 +2045,7 @@ ZSTD_initCStream_usingDict(ZSTD_CStream* zcs,
                      const void* dict, size_t dictSize,
                            int compressionLevel);
 
-/**! ZSTD_initCStream_advanced() :
+/*! ZSTD_initCStream_advanced() :
  * This function is deprecated, and is approximately equivalent to:
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
  *     // Pseudocode: Set each zstd parameter and leave the rest as-is.
@@ -1766,7 +2066,7 @@ ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
                           ZSTD_parameters params,
                           unsigned long long pledgedSrcSize);
 
-/**! ZSTD_initCStream_usingCDict() :
+/*! ZSTD_initCStream_usingCDict() :
  * This function is deprecated, and equivalent to:
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
  *     ZSTD_CCtx_refCDict(zcs, cdict);
@@ -1776,7 +2076,7 @@ ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
  */
 ZSTDLIB_API size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);
 
-/**! ZSTD_initCStream_usingCDict_advanced() :
+/*! ZSTD_initCStream_usingCDict_advanced() :
  *   This function is DEPRECATED, and is approximately equivalent to:
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
  *     // Pseudocode: Set each zstd frame parameter and leave the rest as-is.
@@ -1849,7 +2149,8 @@ ZSTDLIB_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
 
 
 /*=====   Advanced Streaming decompression functions  =====*/
-/**
+
+/*!
  * This function is deprecated, and is equivalent to:
  *
  *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
@@ -1860,7 +2161,7 @@ ZSTDLIB_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
  */
 ZSTDLIB_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
 
-/**
+/*!
  * This function is deprecated, and is equivalent to:
  *
  *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
@@ -1871,7 +2172,7 @@ ZSTDLIB_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dic
  */
 ZSTDLIB_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);
 
-/**
+/*!
  * This function is deprecated, and is equivalent to:
  *
  *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
@@ -1933,7 +2234,7 @@ ZSTDLIB_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstC
 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)
 
   A ZSTD_DCtx object is required to track streaming operations.
diff --git a/programs/Makefile b/programs/Makefile
index 418ad4e6348e..8641d0ee48a0 100644
--- a/programs/Makefile
+++ b/programs/Makefile
@@ -15,7 +15,14 @@
 # zstd-decompress : decompressor-only version of zstd
 # ##########################################################################
 
-ZSTDDIR = ../lib
+.PHONY: default
+default: zstd-release
+
+# silent mode by default; verbose can be triggered by V=1 or VERBOSE=1
+$(V)$(VERBOSE).SILENT:
+
+
+ZSTDDIR := ../lib
 
 # Version numbers
 LIBVER_SRC := $(ZSTDDIR)/zstd.h
@@ -33,64 +40,94 @@ ZSTD_VERSION = $(LIBVER)
 HAVE_COLORNEVER = $(shell echo a | grep --color=never a > /dev/null 2> /dev/null && echo 1 || echo 0)
 GREP_OPTIONS ?=
 ifeq ($HAVE_COLORNEVER, 1)
-GREP_OPTIONS += --color=never
+  GREP_OPTIONS += --color=never
 endif
 GREP = grep $(GREP_OPTIONS)
 
 ifeq ($(shell $(CC) -v 2>&1 | $(GREP) -c "gcc version "), 1)
-ALIGN_LOOP = -falign-loops=32
+  ALIGN_LOOP = -falign-loops=32
 else
-ALIGN_LOOP =
+  ALIGN_LOOP =
 endif
 
-CPPFLAGS+= -DXXH_NAMESPACE=ZSTD_
+DEBUGLEVEL ?= 0
+CPPFLAGS += -DXXH_NAMESPACE=ZSTD_ -DDEBUGLEVEL=$(DEBUGLEVEL)
 ifeq ($(OS),Windows_NT)   # MinGW assumed
-CPPFLAGS   += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
+  CPPFLAGS += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
 endif
-CFLAGS  ?= -O3
+CFLAGS   ?= -O3
 DEBUGFLAGS+=-Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
             -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
             -Wstrict-prototypes -Wundef -Wpointer-arith \
             -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
             -Wredundant-decls -Wmissing-prototypes -Wc++-compat
-CFLAGS  += $(DEBUGFLAGS) $(MOREFLAGS)
-FLAGS    = $(CPPFLAGS) $(CFLAGS) $(LDFLAGS)
+CFLAGS   += $(DEBUGFLAGS) $(MOREFLAGS)
+FLAGS     = $(CPPFLAGS) $(CFLAGS) $(LDFLAGS)
 
+ZSTDLIB_COMMON := $(ZSTDDIR)/common
+ZSTDLIB_COMPRESS := $(ZSTDDIR)/compress
+ZSTDLIB_DECOMPRESS := $(ZSTDDIR)/decompress
+ZDICT_DIR := $(ZSTDDIR)/dictBuilder
+ZSTDLEGACY_DIR := $(ZSTDDIR)/legacy
 
-ZSTDCOMMON_FILES := $(ZSTDDIR)/common/*.c
-ZSTDCOMP_FILES := $(ZSTDDIR)/compress/*.c
-ZSTDDECOMP_FILES := $(ZSTDDIR)/decompress/*.c
-ZSTD_FILES := $(ZSTDDECOMP_FILES) $(ZSTDCOMMON_FILES) $(ZSTDCOMP_FILES)
-ZDICT_FILES := $(ZSTDDIR)/dictBuilder/*.c
-ZSTDDECOMP_O = $(ZSTDDIR)/decompress/zstd_decompress.o
+vpath %.c $(ZSTDLIB_COMMON) $(ZSTDLIB_COMPRESS) $(ZSTDLIB_DECOMPRESS) $(ZDICT_DIR) $(ZSTDLEGACY_DIR)
+
+ZSTDLIB_COMMON_C := $(wildcard $(ZSTDLIB_COMMON)/*.c)
+ZSTDLIB_COMPRESS_C := $(wildcard $(ZSTDLIB_COMPRESS)/*.c)
+ZSTDLIB_DECOMPRESS_C := $(wildcard $(ZSTDLIB_DECOMPRESS)/*.c)
+ZSTDLIB_CORE_SRC := $(ZSTDLIB_DECOMPRESS_C) $(ZSTDLIB_COMMON_C) $(ZSTDLIB_COMPRESS_C)
+ZDICT_SRC := $(wildcard $(ZDICT_DIR)/*.c)
 
 ZSTD_LEGACY_SUPPORT ?= 5
-ZSTDLEGACY_FILES :=
+ZSTDLEGACY_SRC :=
 ifneq ($(ZSTD_LEGACY_SUPPORT), 0)
 ifeq ($(shell test $(ZSTD_LEGACY_SUPPORT) -lt 8; echo $$?), 0)
-	ZSTDLEGACY_FILES += $(shell ls $(ZSTDDIR)/legacy/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
+  ZSTDLEGACY_SRC += $(shell ls $(ZSTDLEGACY_DIR)/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
 endif
-else
 endif
 
 # Sort files in alphabetical order for reproducible builds
-ZSTDLIB_FILES := $(sort $(wildcard $(ZSTD_FILES)) $(wildcard $(ZSTDLEGACY_FILES)) $(wildcard $(ZDICT_FILES)))
-
-ZSTD_CLI_FILES := $(wildcard *.c)
-ZSTD_CLI_OBJ := $(patsubst %.c,%.o,$(ZSTD_CLI_FILES))
+ZSTDLIB_FULL_SRC = $(sort $(ZSTDLIB_CORE_SRC) $(ZSTDLEGACY_SRC) $(ZDICT_SRC))
+ZSTDLIB_LOCAL_SRC := $(notdir $(ZSTDLIB_FULL_SRC))
+ZSTDLIB_LOCAL_OBJ := $(ZSTDLIB_LOCAL_SRC:.c=.o)
+
+ZSTD_CLI_SRC := $(wildcard *.c)
+ZSTD_CLI_OBJ := $(ZSTD_CLI_SRC:.c=.o)
+
+ZSTD_ALL_SRC := $(ZSTDLIB_LOCAL_SRC) $(ZSTD_CLI_SRC)
+ZSTD_ALL_OBJ := $(ZSTD_ALL_SRC:.c=.o)
+
+UNAME := $(shell uname)
+ifeq ($(UNAME), Darwin)
+  HASH ?= md5
+else ifeq ($(UNAME), FreeBSD)
+  HASH ?= gmd5sum
+else ifeq ($(UNAME), OpenBSD)
+  HASH ?= md5
+endif
+HASH ?= md5sum
+HAVE_HASH :=$(shell echo 1 | $(HASH) > /dev/null && echo 1 || echo 0)
+
+ifndef BUILD_DIR
+HASH_DIR = conf_$(shell echo $(CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) $(ZSTD_FILES) | $(HASH) | cut -f 1 -d " ")
+ifeq ($(HAVE_HASH),0)
+  $(info warning : could not find HASH ($(HASH)), needed to differentiate builds using different flags)
+  BUILD_DIR := obj/generic_noconf
+endif
+endif # BUILD_DIR
 
 # Define *.exe as extension for Windows systems
 ifneq (,$(filter Windows%,$(OS)))
-EXT =.exe
-RES64_FILE = windres/zstd64.res
-RES32_FILE = windres/zstd32.res
+  EXT =.exe
+  RES64_FILE = windres/zstd64.res
+  RES32_FILE = windres/zstd32.res
 ifneq (,$(filter x86_64%,$(shell $(CC) -dumpmachine)))
     RES_FILE = $(RES64_FILE)
 else
     RES_FILE = $(RES32_FILE)
 endif
 else
-EXT =
+  EXT =
 endif
 
 VOID = /dev/null
@@ -103,60 +140,64 @@ NO_THREAD_MSG := ==> no threads, building without multithreading support
 HAVE_PTHREAD := $(shell printf '$(NUM_SYMBOL)include <pthread.h>\nint main(void) { return 0; }' > have_pthread.c && $(CC) $(FLAGS) -o have_pthread$(EXT) have_pthread.c -pthread 2> $(VOID) && rm have_pthread$(EXT) && echo 1 || echo 0; rm have_pthread.c)
 HAVE_THREAD := $(shell [ "$(HAVE_PTHREAD)" -eq "1" -o -n "$(filter Windows%,$(OS))" ] && echo 1 || echo 0)
 ifeq ($(HAVE_THREAD), 1)
-THREAD_MSG := ==> building with threading support
-THREAD_CPP := -DZSTD_MULTITHREAD
-THREAD_LD := -pthread
+  THREAD_MSG := ==> building with threading support
+  THREAD_CPP := -DZSTD_MULTITHREAD
+  THREAD_LD := -pthread
 else
-THREAD_MSG := $(NO_THREAD_MSG)
+  THREAD_MSG := $(NO_THREAD_MSG)
 endif
 
 # zlib detection
 NO_ZLIB_MSG := ==> no zlib, building zstd without .gz support
 HAVE_ZLIB := $(shell printf '$(NUM_SYMBOL)include <zlib.h>\nint main(void) { return 0; }' > have_zlib.c && $(CC) $(FLAGS) -o have_zlib$(EXT) have_zlib.c -lz 2> $(VOID) && rm have_zlib$(EXT) && echo 1 || echo 0; rm have_zlib.c)
 ifeq ($(HAVE_ZLIB), 1)
-ZLIB_MSG := ==> building zstd with .gz compression support
-ZLIBCPP = -DZSTD_GZCOMPRESS -DZSTD_GZDECOMPRESS
-ZLIBLD = -lz
+  ZLIB_MSG := ==> building zstd with .gz compression support
+  ZLIBCPP = -DZSTD_GZCOMPRESS -DZSTD_GZDECOMPRESS
+  ZLIBLD = -lz
 else
-ZLIB_MSG := $(NO_ZLIB_MSG)
+  ZLIB_MSG := $(NO_ZLIB_MSG)
 endif
 
 # lzma detection
 NO_LZMA_MSG := ==> no liblzma, building zstd without .xz/.lzma support
 HAVE_LZMA := $(shell printf '$(NUM_SYMBOL)include <lzma.h>\nint main(void) { return 0; }' > have_lzma.c && $(CC) $(FLAGS) -o have_lzma$(EXT) have_lzma.c -llzma 2> $(VOID) && rm have_lzma$(EXT) && echo 1 || echo 0; rm have_lzma.c)
 ifeq ($(HAVE_LZMA), 1)
-LZMA_MSG := ==> building zstd with .xz/.lzma compression support
-LZMACPP = -DZSTD_LZMACOMPRESS -DZSTD_LZMADECOMPRESS
-LZMALD = -llzma
+  LZMA_MSG := ==> building zstd with .xz/.lzma compression support
+  LZMACPP = -DZSTD_LZMACOMPRESS -DZSTD_LZMADECOMPRESS
+  LZMALD = -llzma
 else
-LZMA_MSG := $(NO_LZMA_MSG)
+  LZMA_MSG := $(NO_LZMA_MSG)
 endif
 
 # lz4 detection
 NO_LZ4_MSG := ==> no liblz4, building zstd without .lz4 support
 HAVE_LZ4 := $(shell printf '$(NUM_SYMBOL)include <lz4frame.h>\n$(NUM_SYMBOL)include <lz4.h>\nint main(void) { return 0; }' > have_lz4.c && $(CC) $(FLAGS) -o have_lz4$(EXT) have_lz4.c -llz4 2> $(VOID) && rm have_lz4$(EXT) && echo 1 || echo 0; rm have_lz4.c)
 ifeq ($(HAVE_LZ4), 1)
-LZ4_MSG := ==> building zstd with .lz4 compression support
-LZ4CPP = -DZSTD_LZ4COMPRESS -DZSTD_LZ4DECOMPRESS
-LZ4LD = -llz4
+  LZ4_MSG := ==> building zstd with .lz4 compression support
+  LZ4CPP = -DZSTD_LZ4COMPRESS -DZSTD_LZ4DECOMPRESS
+  LZ4LD = -llz4
 else
-LZ4_MSG := $(NO_LZ4_MSG)
+  LZ4_MSG := $(NO_LZ4_MSG)
 endif
 
 # explicit backtrace enable/disable for Linux & Darwin
 ifeq ($(BACKTRACE), 0)
-DEBUGFLAGS += -DBACKTRACE_ENABLE=0
+  DEBUGFLAGS += -DBACKTRACE_ENABLE=0
 endif
 ifeq (,$(filter Windows%, $(OS)))
 ifeq ($(BACKTRACE), 1)
-DEBUGFLAGS += -DBACKTRACE_ENABLE=1
-DEBUGFLAGS_LD += -rdynamic
+  DEBUGFLAGS += -DBACKTRACE_ENABLE=1
+  DEBUGFLAGS_LD += -rdynamic
 endif
 endif
 
+SET_CACHE_DIRECTORY = \
+	$(MAKE) --no-print-directory $@ \
+    BUILD_DIR=obj/$(HASH_DIR) \
+    CPPFLAGS="$(CPPFLAGS)" \
+    CFLAGS="$(CFLAGS)" \
+    LDFLAGS="$(LDFLAGS)"
 
-.PHONY: default
-default: zstd-release
 
 .PHONY: all
 all: zstd
@@ -164,21 +205,51 @@ all: zstd
 .PHONY: allVariants
 allVariants: zstd zstd-compress zstd-decompress zstd-small zstd-nolegacy zstd-dictBuilder
 
-$(ZSTDDECOMP_O): CFLAGS += $(ALIGN_LOOP)
-
+.PHONY: zstd  # must always be run
 zstd : CPPFLAGS += $(THREAD_CPP) $(ZLIBCPP) $(LZMACPP) $(LZ4CPP)
 zstd : LDFLAGS += $(THREAD_LD) $(ZLIBLD) $(LZMALD) $(LZ4LD) $(DEBUGFLAGS_LD)
 zstd : CPPFLAGS += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
 ifneq (,$(filter Windows%,$(OS)))
 zstd : $(RES_FILE)
 endif
-zstd : $(ZSTDLIB_FILES) $(ZSTD_CLI_OBJ)
+
+ifndef BUILD_DIR
+# generate BUILD_DIR from flags
+
+zstd:
+	$(SET_CACHE_DIRECTORY)
+
+else
+# BUILD_DIR is defined
+
+ZSTD_OBJ := $(addprefix $(BUILD_DIR)/, $(ZSTD_ALL_OBJ))
+$(BUILD_DIR)/zstd : $(ZSTD_OBJ)
 	@echo "$(THREAD_MSG)"
 	@echo "$(ZLIB_MSG)"
 	@echo "$(LZMA_MSG)"
 	@echo "$(LZ4_MSG)"
+	@echo LINK $@
 	$(CC) $(FLAGS) $^ -o $@$(EXT) $(LDFLAGS)
 
+ifeq ($(HAVE_HASH),1)
+SRCBIN_HASH = $(shell cat $(BUILD_DIR)/zstd 2> $(VOID) | $(HASH) | cut -f 1 -d " ")
+DSTBIN_HASH = $(shell cat zstd 2> $(VOID) | $(HASH) | cut -f 1 -d " ")
+BIN_ISDIFFERENT = $(if $(filter $(SRCBIN_HASH),$(DSTBIN_HASH)),0,1)
+else
+BIN_ISDIFFERENT = 1
+endif
+
+zstd : $(BUILD_DIR)/zstd
+	if [ $(BIN_ISDIFFERENT) -eq 1 ]; then \
+		cp -f $< $@; \
+		echo zstd build completed; \
+	else \
+		echo zstd already built; \
+	fi
+
+endif  # BUILD_DIR
+
+
 .PHONY: zstd-release
 zstd-release: DEBUGFLAGS := -DBACKTRACE_ENABLE=0
 zstd-release: DEBUGFLAGS_LD :=
@@ -190,12 +261,12 @@ zstd32 : CPPFLAGS += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
 ifneq (,$(filter Windows%,$(OS)))
 zstd32 : $(RES32_FILE)
 endif
-zstd32 : $(ZSTDLIB_FILES) $(ZSTD_CLI_FILES)
+zstd32 : $(ZSTDLIB_FULL_SRC) $(ZSTD_CLI_SRC)
 	$(CC) -m32 $(FLAGS) $^ -o $@$(EXT)
 
 ## zstd-nolegacy: same scope as zstd, with just support of legacy formats removed
 zstd-nolegacy : LDFLAGS += $(THREAD_LD) $(ZLIBLD) $(LZMALD) $(LZ4LD) $(DEBUGFLAGS_LD)
-zstd-nolegacy : $(ZSTD_FILES) $(ZDICT_FILES) $(ZSTD_CLI_OBJ)
+zstd-nolegacy : $(ZSTDLIB_CORE_SRC) $(ZDICT_SRC) $(ZSTD_CLI_OBJ)
 	$(CC) $(FLAGS) $^ -o $@$(EXT) $(LDFLAGS)
 
 zstd-nomt : THREAD_CPP :=
@@ -222,12 +293,12 @@ zstd-noxz : zstd
 #        It's unclear at this stage if this is a scenario that must be supported
 .PHONY: zstd-dll
 zstd-dll : LDFLAGS+= -L$(ZSTDDIR) -lzstd
-zstd-dll : ZSTDLIB_FILES =
+zstd-dll : ZSTDLIB_FULL_SRC =
 zstd-dll : $(ZSTD_CLI_OBJ)
 	$(CC) $(FLAGS) $^ -o $@$(EXT) $(LDFLAGS)
 
 
-## zstd-pgo: zstd executable optimized with pgo. `gcc` only.
+## zstd-pgo: zstd executable optimized with PGO.
 zstd-pgo :
 	$(MAKE) clean
 	$(MAKE) zstd MOREFLAGS=-fprofile-generate
@@ -237,24 +308,24 @@ zstd-pgo :
 	./zstd -b $(PROFILE_WITH)
 	./zstd -b7i2 $(PROFILE_WITH)
 	./zstd -b5 $(PROFILE_WITH)
-	$(RM) zstd *.o $(ZSTDDECOMP_O) $(ZSTDDIR)/compress/*.o
+	$(RM) zstd *.o
 	case $(CC) in *clang*) if ! [ -e default.profdata ]; then llvm-profdata merge -output=default.profdata default*.profraw; fi ;; esac
 	$(MAKE) zstd MOREFLAGS=-fprofile-use
 
 ## zstd-small: minimal target, supporting only zstd compression and decompression. no bench. no legacy. no other format.
 zstd-small: CFLAGS = -Os -s
-zstd-frugal zstd-small: $(ZSTD_FILES) zstdcli.c util.c timefn.c fileio.c
+zstd-frugal zstd-small: $(ZSTDLIB_CORE_SRC) zstdcli.c util.c timefn.c fileio.c
 	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODICT $^ -o $@$(EXT)
 
-zstd-decompress: $(ZSTDCOMMON_FILES) $(ZSTDDECOMP_FILES) zstdcli.c util.c timefn.c fileio.c
+zstd-decompress: $(ZSTDLIB_COMMON_C) $(ZSTDLIB_DECOMPRESS_C) zstdcli.c util.c timefn.c fileio.c
 	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODICT -DZSTD_NOCOMPRESS $^ -o $@$(EXT)
 
-zstd-compress: $(ZSTDCOMMON_FILES) $(ZSTDCOMP_FILES) zstdcli.c util.c timefn.c fileio.c
+zstd-compress: $(ZSTDLIB_COMMON_C) $(ZSTDLIB_COMPRESS_C) zstdcli.c util.c timefn.c fileio.c
 	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODICT -DZSTD_NODECOMPRESS $^ -o $@$(EXT)
 
 ## zstd-dictBuilder: executable supporting dictionary creation and compression (only)
 zstd-dictBuilder: CPPFLAGS += -DZSTD_NOBENCH -DZSTD_NODECOMPRESS
-zstd-dictBuilder: $(ZSTDCOMMON_FILES) $(ZSTDCOMP_FILES) $(ZDICT_FILES) zstdcli.c util.c timefn.c fileio.c dibio.c
+zstd-dictBuilder: $(ZSTDLIB_COMMON_C) $(ZSTDLIB_COMPRESS_C) $(ZDICT_SRC) zstdcli.c util.c timefn.c fileio.c dibio.c
 	$(CC) $(FLAGS) $^ -o $@$(EXT)
 
 zstdmt: zstd
@@ -274,12 +345,11 @@ endif
 
 .PHONY: clean
 clean:
-	$(MAKE) -C $(ZSTDDIR) clean
-	@$(RM) $(ZSTDDIR)/decompress/*.o $(ZSTDDIR)/decompress/zstd_decompress.gcda
-	@$(RM) core *.o tmp* result* *.gcda dictionary *.zst \
+	$(RM) core *.o tmp* result* *.gcda dictionary *.zst \
         zstd$(EXT) zstd32$(EXT) zstd-compress$(EXT) zstd-decompress$(EXT) \
         zstd-small$(EXT) zstd-frugal$(EXT) zstd-nolegacy$(EXT) zstd4$(EXT) \
         zstd-dictBuilder$(EXT) *.gcda default*.profraw default.profdata have_zlib$(EXT)
+	$(RM) -r obj/*
 	@echo Cleaning completed
 
 MD2ROFF = ronn
@@ -309,15 +379,34 @@ preview-man: clean-man man
 	man ./zstdgrep.1
 	man ./zstdless.1
 
+
+# Generate .h dependencies automatically
+
+DEPFLAGS = -MT $@ -MMD -MP -MF
+
+$(BUILD_DIR)/%.o : %.c $(BUILD_DIR)/%.d | $(BUILD_DIR)
+	@echo CC $@
+	$(COMPILE.c) $(DEPFLAGS) $(BUILD_DIR)/$*.d $(OUTPUT_OPTION) $<
+
+MKDIR ?= mkdir
+$(BUILD_DIR): ; $(MKDIR) -p $@
+
+DEPFILES := $(ZSTD_OBJ:.o=.d)
+$(DEPFILES):
+
+include $(wildcard $(DEPFILES))
+
+
+
 #-----------------------------------------------------------------------------
 # make install is validated only for Linux, macOS, BSD, Hurd and Solaris targets
 #-----------------------------------------------------------------------------
-ifneq (,$(filter $(shell uname),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku))
+ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku))
 
 HAVE_COLORNEVER = $(shell echo a | egrep --color=never a > /dev/null 2> /dev/null && echo 1 || echo 0)
 EGREP_OPTIONS ?=
 ifeq ($HAVE_COLORNEVER, 1)
-EGREP_OPTIONS += --color=never
+  EGREP_OPTIONS += --color=never
 endif
 EGREP = egrep $(EGREP_OPTIONS)
 AWK = awk
@@ -328,7 +417,7 @@ AWK = awk
 ## list: Print all targets and their descriptions (if provided)
 .PHONY: list
 list:
-	@TARGETS=$$($(MAKE) -pRrq -f $(lastword $(MAKEFILE_LIST)) : 2>/dev/null \
+	TARGETS=$$($(MAKE) -pRrq -f $(lastword $(MAKEFILE_LIST)) : 2>/dev/null \
 		| $(AWK) -v RS= -F: '/^# File/,/^# Finished Make data base/ {if ($$1 !~ "^[#.]") {print $$1}}' \
 		| $(EGREP) -v  -e '^[^[:alnum:]]' | sort); \
 	{ \
@@ -355,17 +444,17 @@ datarootdir ?= $(PREFIX)/share
 mandir      ?= $(datarootdir)/man
 man1dir     ?= $(mandir)/man1
 
-ifneq (,$(filter $(shell uname),OpenBSD FreeBSD NetBSD DragonFly SunOS))
-MANDIR  ?= $(PREFIX)/man
-MAN1DIR ?= $(MANDIR)/man1
+ifneq (,$(filter $(UNAME),OpenBSD FreeBSD NetBSD DragonFly SunOS))
+  MANDIR  ?= $(PREFIX)/man
+  MAN1DIR ?= $(MANDIR)/man1
 else
-MAN1DIR ?= $(man1dir)
+  MAN1DIR ?= $(man1dir)
 endif
 
-ifneq (,$(filter $(shell uname),SunOS))
-INSTALL ?= ginstall
+ifneq (,$(filter $(UNAME),SunOS))
+  INSTALL ?= ginstall
 else
-INSTALL ?= install
+  INSTALL ?= install
 endif
 
 INSTALL_PROGRAM ?= $(INSTALL)
@@ -374,36 +463,39 @@ INSTALL_DATA    ?= $(INSTALL) -m 644
 INSTALL_MAN     ?= $(INSTALL_DATA)
 
 .PHONY: install
-install: zstd
+install:
+	# generate zstd only if not already present
+	[ -e zstd ] || $(MAKE) zstd-release
+	[ -e $(DESTDIR)$(BINDIR) ] || $(INSTALL) -d -m 755 $(DESTDIR)$(BINDIR)/
+	[ -e $(DESTDIR)$(MAN1DIR) ] || $(INSTALL) -d -m 755 $(DESTDIR)$(MAN1DIR)/
 	@echo Installing binaries
-	@$(INSTALL) -d -m 755 $(DESTDIR)$(BINDIR)/ $(DESTDIR)$(MAN1DIR)/
-	@$(INSTALL_PROGRAM) zstd$(EXT) $(DESTDIR)$(BINDIR)/zstd$(EXT)
-	@ln -sf zstd$(EXT) $(DESTDIR)$(BINDIR)/zstdcat$(EXT)
-	@ln -sf zstd$(EXT) $(DESTDIR)$(BINDIR)/unzstd$(EXT)
-	@ln -sf zstd$(EXT) $(DESTDIR)$(BINDIR)/zstdmt$(EXT)
-	@$(INSTALL_SCRIPT) zstdless $(DESTDIR)$(BINDIR)/zstdless
-	@$(INSTALL_SCRIPT) zstdgrep $(DESTDIR)$(BINDIR)/zstdgrep
+	$(INSTALL_PROGRAM) zstd$(EXT) $(DESTDIR)$(BINDIR)/zstd$(EXT)
+	ln -sf zstd$(EXT) $(DESTDIR)$(BINDIR)/zstdcat$(EXT)
+	ln -sf zstd$(EXT) $(DESTDIR)$(BINDIR)/unzstd$(EXT)
+	ln -sf zstd$(EXT) $(DESTDIR)$(BINDIR)/zstdmt$(EXT)
+	$(INSTALL_SCRIPT) zstdless $(DESTDIR)$(BINDIR)/zstdless
+	$(INSTALL_SCRIPT) zstdgrep $(DESTDIR)$(BINDIR)/zstdgrep
 	@echo Installing man pages
-	@$(INSTALL_MAN) zstd.1 $(DESTDIR)$(MAN1DIR)/zstd.1
-	@ln -sf zstd.1 $(DESTDIR)$(MAN1DIR)/zstdcat.1
-	@ln -sf zstd.1 $(DESTDIR)$(MAN1DIR)/unzstd.1
-	@$(INSTALL_MAN) zstdgrep.1 $(DESTDIR)$(MAN1DIR)/zstdgrep.1
-	@$(INSTALL_MAN) zstdless.1 $(DESTDIR)$(MAN1DIR)/zstdless.1
+	$(INSTALL_MAN) zstd.1 $(DESTDIR)$(MAN1DIR)/zstd.1
+	ln -sf zstd.1 $(DESTDIR)$(MAN1DIR)/zstdcat.1
+	ln -sf zstd.1 $(DESTDIR)$(MAN1DIR)/unzstd.1
+	$(INSTALL_MAN) zstdgrep.1 $(DESTDIR)$(MAN1DIR)/zstdgrep.1
+	$(INSTALL_MAN) zstdless.1 $(DESTDIR)$(MAN1DIR)/zstdless.1
 	@echo zstd installation completed
 
 .PHONY: uninstall
 uninstall:
-	@$(RM) $(DESTDIR)$(BINDIR)/zstdgrep
-	@$(RM) $(DESTDIR)$(BINDIR)/zstdless
-	@$(RM) $(DESTDIR)$(BINDIR)/zstdcat
-	@$(RM) $(DESTDIR)$(BINDIR)/unzstd
-	@$(RM) $(DESTDIR)$(BINDIR)/zstdmt
-	@$(RM) $(DESTDIR)$(BINDIR)/zstd
-	@$(RM) $(DESTDIR)$(MAN1DIR)/zstdless.1
-	@$(RM) $(DESTDIR)$(MAN1DIR)/zstdgrep.1
-	@$(RM) $(DESTDIR)$(MAN1DIR)/zstdcat.1
-	@$(RM) $(DESTDIR)$(MAN1DIR)/unzstd.1
-	@$(RM) $(DESTDIR)$(MAN1DIR)/zstd.1
+	$(RM) $(DESTDIR)$(BINDIR)/zstdgrep
+	$(RM) $(DESTDIR)$(BINDIR)/zstdless
+	$(RM) $(DESTDIR)$(BINDIR)/zstdcat
+	$(RM) $(DESTDIR)$(BINDIR)/unzstd
+	$(RM) $(DESTDIR)$(BINDIR)/zstdmt
+	$(RM) $(DESTDIR)$(BINDIR)/zstd
+	$(RM) $(DESTDIR)$(MAN1DIR)/zstdless.1
+	$(RM) $(DESTDIR)$(MAN1DIR)/zstdgrep.1
+	$(RM) $(DESTDIR)$(MAN1DIR)/zstdcat.1
+	$(RM) $(DESTDIR)$(MAN1DIR)/unzstd.1
+	$(RM) $(DESTDIR)$(MAN1DIR)/zstd.1
 	@echo zstd programs successfully uninstalled
 
 endif
diff --git a/programs/README.md b/programs/README.md
index 53706de727b7..cf7f5ba46d12 100644
--- a/programs/README.md
+++ b/programs/README.md
@@ -3,7 +3,7 @@ Command Line Interface for Zstandard library
 
 Command Line Interface (CLI) can be created using the `make` command without any additional parameters.
 There are however other Makefile targets that create different variations of CLI:
-- `zstd` : default CLI supporting gzip-like arguments; includes dictionary builder, benchmark, and support for decompression of legacy zstd formats
+- `zstd` : default CLI supporting gzip-like arguments; includes dictionary builder, benchmark, and supports decompression of legacy zstd formats
 - `zstd_nolegacy` : Same as `zstd` but without support for legacy zstd formats
 - `zstd-small` : CLI optimized for minimal size; no dictionary builder, no benchmark, and no support for legacy zstd formats
 - `zstd-compress` : version of CLI which can only compress into zstd format
@@ -147,71 +147,91 @@ FILE    : a filename
 Arguments :
  -#     : # compression level (1-19, default: 3)
  -d     : decompression
- -D file: use `file` as Dictionary
- -o file: result stored into `file` (only if 1 input file)
- -f     : overwrite output without prompting and (de)compress links
+ -D DICT: use DICT as Dictionary for compression or decompression
+ -o file: result stored into `file` (only 1 output file)
+ -f     : overwrite output without prompting, also (de)compress links
 --rm    : remove source file(s) after successful de/compression
  -k     : preserve source file(s) (default)
  -h/-H  : display help/long help and exit
 
 Advanced arguments :
  -V     : display Version number and exit
+ -c     : force write to standard output, even if it is the console
  -v     : verbose mode; specify multiple times to increase verbosity
  -q     : suppress warnings; specify twice to suppress errors too
- -c     : force write to standard output, even if it is the console
- -l     : print information about zstd compressed files
---exclude-compressed:  only compress files that are not previously compressed
+--no-progress : do not display the progress counter
+ -r     : operate recursively on directories
+--filelist FILE : read list of files to operate upon from FILE
+--output-dir-flat DIR : processed files are stored into DIR
+--output-dir-mirror DIR : processed files are stored into DIR respecting original directory structure
+--[no-]check : during compression, add XXH64 integrity checksum to frame (default: enabled). If specified with -d, decompressor will ignore/validate checksums in compressed frame (default: validate).
+--      : All arguments after "--" are treated as files
+
+Advanced compression arguments :
 --ultra : enable levels beyond 19, up to 22 (requires more memory)
 --long[=#]: enable long distance matching with given window log (default: 27)
 --fast[=#]: switch to very fast compression levels (default: 1)
 --adapt : dynamically adapt compression level to I/O conditions
---stream-size=# : optimize compression parameters for streaming input of given number of bytes
---size-hint=# optimize compression parameters for streaming input of approximately this size
---target-compressed-block-size=# : make compressed block near targeted size
  -T#    : spawns # compression threads (default: 1, 0==# cores)
  -B#    : select size of each job (default: 0==automatic)
+--single-thread : use a single thread for both I/O and compression (result slightly different than -T1)
 --rsyncable : compress using a rsync-friendly method (-B sets block size)
---no-dictID : don't write dictID into header (dictionary compression)
---[no-]check : integrity check (default: enabled)
+--exclude-compressed: only compress files that are not already compressed
+--stream-size=# : specify size of streaming input from `stdin`
+--size-hint=# optimize compression parameters for streaming input of approximately this size
+--target-compressed-block-size=# : generate compressed block of approximately targeted size
+--no-dictID : don't write dictID into header (dictionary compression only)
 --[no-]compress-literals : force (un)compressed literals
- -r     : operate recursively on directories
---output-dir-flat[=directory]: all resulting files stored into `directory`.
 --format=zstd : compress files to the .zst format (default)
 --format=gzip : compress files to the .gz format
+--format=xz : compress files to the .xz format
+--format=lzma : compress files to the .lzma format
+--format=lz4 : compress files to the .lz4 format
+
+Advanced decompression arguments :
+ -l     : print information about zstd compressed files
 --test  : test compressed file integrity
---[no-]sparse : sparse mode (default: disabled)
  -M#    : Set a memory usage limit for decompression
---no-progress : do not display the progress bar
---      : All arguments after "--" are treated as files
+--[no-]sparse : sparse mode (default: disabled)
 
 Dictionary builder :
 --train ## : create a dictionary from a training set of files
 --train-cover[=k=#,d=#,steps=#,split=#,shrink[=#]] : use the cover algorithm with optional args
 --train-fastcover[=k=#,d=#,f=#,steps=#,split=#,accel=#,shrink[=#]] : use the fast cover algorithm with optional args
 --train-legacy[=s=#] : use the legacy algorithm with selectivity (default: 9)
- -o file : `file` is dictionary name (default: dictionary)
+ -o DICT : DICT is dictionary name (default: dictionary)
 --maxdict=# : limit dictionary to specified size (default: 112640)
 --dictID=# : force dictionary ID to specified value (default: random)
 
 Benchmark arguments :
  -b#    : benchmark file(s), using # compression level (default: 3)
- -e#    : test all compression levels from -bX to # (default: 1)
+ -e#    : test all compression levels successively from -b# to -e# (default: 1)
  -i#    : minimum evaluation time in seconds (default: 3s)
  -B#    : cut file into independent blocks of size # (default: no block)
+ -S     : output one benchmark result per input file (default: consolidated result)
 --priority=rt : set process priority to real-time
 ```
 
 ### Passing parameters through Environment Variables
+There is no "generic" way to pass "any kind of parameter" to `zstd` in a pass-through manner.
+Using environment variables for this purpose has security implications.
+Therefore, this avenue is intentionally restricted and only supports `ZSTD_CLEVEL` and `ZSTD_NBTHREADS`.
+
 `ZSTD_CLEVEL` can be used to modify the default compression level of `zstd`
 (usually set to `3`) to another value between 1 and 19 (the "normal" range).
-This can be useful when `zstd` CLI is invoked in a way that doesn't allow passing arguments.
+
+`ZSTD_NBTHREADS` can be used to specify a number of threads
+that `zstd` will use for compression, which by default is `1`.
+This functionality only exists when `zstd` is compiled with multithread support.
+`0` means "use as many threads as detected cpu cores on local system".
+The max # of threads is capped at: `ZSTDMT_NBWORKERS_MAX==200`.
+
+This functionality can be useful when `zstd` CLI is invoked in a way that doesn't allow passing arguments.
 One such scenario is `tar --zstd`.
-As `ZSTD_CLEVEL` only replaces the default compression level,
-it can then be overridden by corresponding command line arguments.
+As `ZSTD_CLEVEL` and `ZSTD_NBTHREADS` only replace the default compression level
+and number of threads respectively, they can both be overridden by corresponding command line arguments:
+`-#` for compression level and `-T#` for number of threads.
 
-There is no "generic" way to pass "any kind of parameter" to `zstd` in a pass-through manner.
-Using environment variables for this purpose has security implications.
-Therefore, this avenue is intentionally restricted and only supports `ZSTD_CLEVEL`.
 
 ### Long distance matching mode
 The long distance matching mode, enabled with `--long`, is designed to improve
diff --git a/programs/dibio.c b/programs/dibio.c
index 463095a8e813..cb3829e3e59b 100644
--- a/programs/dibio.c
+++ b/programs/dibio.c
@@ -301,7 +301,7 @@ int DiB_trainFromFiles(const char* dictFileName, unsigned maxDictSize,
         DISPLAYLEVEL(2, "!  Alternatively, split files into fixed-size blocks representative of samples, with -B# \n");
         EXM_THROW(14, "nb of samples too low");   /* we now clearly forbid this case */
     }
-    if (fs.totalSizeToLoad < (unsigned long long)(8 * maxDictSize)) {
+    if (fs.totalSizeToLoad < (unsigned long long)maxDictSize * 8) {
         DISPLAYLEVEL(2, "!  Warning : data size of samples too small for target dictionary size \n");
         DISPLAYLEVEL(2, "!  Samples should be about 100x larger than target dictionary size \n");
     }
diff --git a/programs/fileio.c b/programs/fileio.c
index d72879d64eae..65f2d531a81d 100644
--- a/programs/fileio.c
+++ b/programs/fileio.c
@@ -321,6 +321,25 @@ struct FIO_prefs_s {
     int contentSize;
 };
 
+/*-*************************************
+*  Parameters: FIO_ctx_t
+***************************************/
+
+/* typedef'd to FIO_ctx_t within fileio.h */
+struct FIO_ctx_s {
+
+    /* file i/o info */
+    int nbFilesTotal;
+    int hasStdinInput;
+    int hasStdoutOutput;
+
+    /* file i/o state */
+    int currFileIdx;
+    int nbFilesProcessed;
+    size_t totalBytesInput;
+    size_t totalBytesOutput;
+};
+
 
 /*-*************************************
 *  Parameters: Initialization
@@ -363,11 +382,31 @@ FIO_prefs_t* FIO_createPreferences(void)
     return ret;
 }
 
+FIO_ctx_t* FIO_createContext(void)
+{
+    FIO_ctx_t* const ret = (FIO_ctx_t*)malloc(sizeof(FIO_ctx_t));
+    if (!ret) EXM_THROW(21, "Allocation error : not enough memory");
+
+    ret->currFileIdx = 0;
+    ret->hasStdinInput = 0;
+    ret->hasStdoutOutput = 0;
+    ret->nbFilesTotal = 1;
+    ret->nbFilesProcessed = 0;
+    ret->totalBytesInput = 0;
+    ret->totalBytesOutput = 0;
+    return ret;
+}
+
 void FIO_freePreferences(FIO_prefs_t* const prefs)
 {
     free(prefs);
 }
 
+void FIO_freeContext(FIO_ctx_t* const fCtx)
+{
+    free(fCtx);
+}
+
 
 /*-*************************************
 *  Parameters: Display Options
@@ -382,6 +421,8 @@ void FIO_setNoProgress(unsigned noProgress) { g_display_prefs.noProgress = noPro
 *  Parameters: Setters
 ***************************************/
 
+/* FIO_prefs_t functions */
+
 void FIO_setCompressionType(FIO_prefs_t* const prefs, FIO_compressionType_t compressionType) { prefs->compressionType = compressionType; }
 
 void FIO_overwriteMode(FIO_prefs_t* const prefs) { prefs->overwrite = 1; }
@@ -495,21 +536,49 @@ void FIO_setContentSize(FIO_prefs_t* const prefs, int value)
     prefs->contentSize = value != 0;
 }
 
+/* FIO_ctx_t functions */
+
+void FIO_setHasStdoutOutput(FIO_ctx_t* const fCtx, int value) {
+    fCtx->hasStdoutOutput = value;
+}
+
+void FIO_setNbFilesTotal(FIO_ctx_t* const fCtx, int value)
+{
+    fCtx->nbFilesTotal = value;
+}
+
+void FIO_determineHasStdinInput(FIO_ctx_t* const fCtx, const FileNamesTable* const filenames) {
+    size_t i = 0;
+    for ( ; i < filenames->tableSize; ++i) {
+        if (!strcmp(stdinmark, filenames->fileNames[i])) {
+            fCtx->hasStdinInput = 1;
+            return;
+        }
+    }
+}
+
 /*-*************************************
 *  Functions
 ***************************************/
-/** FIO_remove() :
+/** FIO_removeFile() :
  * @result : Unlink `fileName`, even if it's read-only */
-static int FIO_remove(const char* path)
+static int FIO_removeFile(const char* path)
 {
-    if (!UTIL_isRegularFile(path)) {
-        DISPLAYLEVEL(2, "zstd: Refusing to remove non-regular file %s \n", path);
+    stat_t statbuf;
+    if (!UTIL_stat(path, &statbuf)) {
+        DISPLAYLEVEL(2, "zstd: Failed to stat %s while trying to remove it\n", path);
+        return 0;
+    }
+    if (!UTIL_isRegularFileStat(&statbuf)) {
+        DISPLAYLEVEL(2, "zstd: Refusing to remove non-regular file %s\n", path);
         return 0;
     }
 #if defined(_WIN32) || defined(WIN32)
     /* windows doesn't allow remove read-only files,
      * so try to make it writable first */
-    UTIL_chmod(path, _S_IWRITE);
+    if (!(statbuf.st_mode & _S_IWRITE)) {
+        UTIL_chmod(path, &statbuf, _S_IWRITE);
+    }
 #endif
     return remove(path);
 }
@@ -519,6 +588,7 @@ static int FIO_remove(const char* path)
  * @result : FILE* to `srcFileName`, or NULL if it fails */
 static FILE* FIO_openSrcFile(const char* srcFileName)
 {
+    stat_t statbuf;
     assert(srcFileName != NULL);
     if (!strcmp (srcFileName, stdinmark)) {
         DISPLAYLEVEL(4,"Using stdin for input \n");
@@ -526,14 +596,14 @@ static FILE* FIO_openSrcFile(const char* srcFileName)
         return stdin;
     }
 
-    if (!UTIL_fileExist(srcFileName)) {
+    if (!UTIL_stat(srcFileName, &statbuf)) {
         DISPLAYLEVEL(1, "zstd: can't stat %s : %s -- ignored \n",
                         srcFileName, strerror(errno));
         return NULL;
     }
 
-    if (!UTIL_isRegularFile(srcFileName)
-     && !UTIL_isFIFO(srcFileName)
+    if (!UTIL_isRegularFileStat(&statbuf)
+     && !UTIL_isFIFOStat(&statbuf)
     ) {
         DISPLAYLEVEL(1, "zstd: %s is not a regular file -- ignored \n",
                         srcFileName);
@@ -551,8 +621,8 @@ static FILE* FIO_openSrcFile(const char* srcFileName)
  *  condition : `dstFileName` must be non-NULL.
  * @result : FILE* to `dstFileName`, or NULL if it fails */
 static FILE*
-FIO_openDstFile(FIO_prefs_t* const prefs,
-          const char* srcFileName, const char* dstFileName)
+FIO_openDstFile(FIO_ctx_t* fCtx, FIO_prefs_t* const prefs,
+                const char* srcFileName, const char* dstFileName)
 {
     if (prefs->testMode) return NULL;  /* do not open file in test mode */
 
@@ -597,18 +667,12 @@ FIO_openDstFile(FIO_prefs_t* const prefs,
                             dstFileName);
                     return NULL;
                 }
-                DISPLAY("zstd: %s already exists; overwrite (y/N) ? ",
-                        dstFileName);
-                {   int ch = getchar();
-                    if ((ch!='Y') && (ch!='y')) {
-                        DISPLAY("    not overwritten  \n");
-                        return NULL;
-                    }
-                    /* flush rest of input line */
-                    while ((ch!=EOF) && (ch!='\n')) ch = getchar();
-            }   }
+                DISPLAY("zstd: %s already exists; ", dstFileName);
+                if (UTIL_requireUserConfirmation("overwrite (y/n) ? ", "Not overwritten  \n", "yY", fCtx->hasStdinInput))
+                    return NULL;
+            }
             /* need to unlink */
-            FIO_remove(dstFileName);
+            FIO_removeFile(dstFileName);
     }   }
 
     {   FILE* const f = fopen( dstFileName, "wb" );
@@ -618,13 +682,12 @@ FIO_openDstFile(FIO_prefs_t* const prefs,
                && strcmp (srcFileName, stdinmark)
                && strcmp(dstFileName, nulmark) ) {
             /* reduce rights on newly created dst file while compression is ongoing */
-            UTIL_chmod(dstFileName, 00600);
+            UTIL_chmod(dstFileName, NULL, 00600);
         }
         return f;
     }
 }
 
-
 /*! FIO_createDictBuffer() :
  *  creates a buffer, pointed by `*bufferPtr`,
  *  loads `filename` content into it, up to DICTSIZE_MAX bytes.
@@ -669,15 +732,9 @@ static size_t FIO_createDictBuffer(void** bufferPtr, const char* fileName, FIO_p
  * Checks for and warns if there are any files that would have the same output path
  */
 int FIO_checkFilenameCollisions(const char** filenameTable, unsigned nbFiles) {
-    const char **filenameTableSorted, *c, *prevElem, *filename;
+    const char **filenameTableSorted, *prevElem, *filename;
     unsigned u;
 
-    #if defined(_MSC_VER) || defined(__MINGW32__) || defined (__MSVCRT__) /* windows support */
-    c = "\\";
-    #else
-    c = "/";
-    #endif
-
     filenameTableSorted = (const char**) malloc(sizeof(char*) * nbFiles);
     if (!filenameTableSorted) {
         DISPLAY("Unable to malloc new str array, not checking for name collisions\n");
@@ -685,7 +742,7 @@ int FIO_checkFilenameCollisions(const char** filenameTable, unsigned nbFiles) {
     }
 
     for (u = 0; u < nbFiles; ++u) {
-        filename = strrchr(filenameTable[u], c[0]);
+        filename = strrchr(filenameTable[u], PATH_SEP);
         if (filename == NULL) {
             filenameTableSorted[u] = filenameTable[u];
         } else {
@@ -771,12 +828,57 @@ static void FIO_adjustMemLimitForPatchFromMode(FIO_prefs_t* const prefs,
                                     unsigned long long const maxSrcFileSize)
 {
     unsigned long long maxSize = MAX(prefs->memLimit, MAX(dictSize, maxSrcFileSize));
+    unsigned const maxWindowSize = (1U << ZSTD_WINDOWLOG_MAX);
+    if (maxSize == UTIL_FILESIZE_UNKNOWN)
+        EXM_THROW(42, "Using --patch-from with stdin requires --stream-size");
     assert(maxSize != UTIL_FILESIZE_UNKNOWN);
-    if (maxSize > UINT_MAX)
-        EXM_THROW(42, "Can't handle files larger than %u GB\n", UINT_MAX/(1 GB) + 1);
+    if (maxSize > maxWindowSize)
+        EXM_THROW(42, "Can't handle files larger than %u GB\n", maxWindowSize/(1 GB));
     FIO_setMemLimit(prefs, (unsigned)maxSize);
 }
 
+/* FIO_removeMultiFilesWarning() :
+ * Returns 1 if the console should abort, 0 if console should proceed.
+ * This function handles logic when processing multiple files with -o, displaying the appropriate warnings/prompts.
+ * 
+ * If -f is specified, or there is just 1 file, zstd will always proceed as usual.
+ * If --rm is specified, there will be a prompt asking for user confirmation.
+ *         If -f is specified with --rm, zstd will proceed as usual
+ *         If -q is specified with --rm, zstd will abort pre-emptively
+ *         If neither flag is specified, zstd will prompt the user for confirmation to proceed.
+ * If --rm is not specified, then zstd will print a warning to the user (which can be silenced with -q).
+ * However, if the output is stdout, we will always abort rather than displaying the warning prompt.
+ */
+static int FIO_removeMultiFilesWarning(FIO_ctx_t* const fCtx, const FIO_prefs_t* const prefs, const char* outFileName, int displayLevelCutoff)
+{
+    int error = 0;
+    if (fCtx->nbFilesTotal > 1 && !prefs->overwrite) {
+        if (g_display_prefs.displayLevel <= displayLevelCutoff) {
+            if (prefs->removeSrcFile) {
+                DISPLAYLEVEL(1, "zstd: Aborting... not deleting files and processing into dst: %s", outFileName);
+                error =  1;
+            }
+        } else {
+            if (!strcmp(outFileName, stdoutmark)) {
+                DISPLAYLEVEL(2, "zstd: WARNING: all input files will be processed and concatenated into stdout. ");
+            } else {
+                DISPLAYLEVEL(2, "zstd: WARNING: all input files will be processed and concatenated into a single output file: %s ", outFileName);
+            }
+            DISPLAYLEVEL(2, "\nThe concatenated output CANNOT regenerate the original directory tree. ")
+            if (prefs->removeSrcFile) {
+                if (fCtx->hasStdoutOutput) {
+                    DISPLAYLEVEL(1, "\nAborting. Use -f if you really want to delete the files and output to stdout");
+                    error = 1;
+                } else {
+                    error = g_display_prefs.displayLevel > displayLevelCutoff && UTIL_requireUserConfirmation("This is a destructive operation. Proceed? (y/n): ", "Aborting...", "yY", fCtx->hasStdinInput);
+                }
+            }
+        }
+        DISPLAY("\n");
+    }
+    return error;
+}
+
 #ifndef ZSTD_NOCOMPRESS
 
 /* **********************************************************************
@@ -807,9 +909,9 @@ static void FIO_adjustParamsForPatchFromMode(FIO_prefs_t* const prefs,
     if (fileWindowLog > ZSTD_WINDOWLOG_MAX)
         DISPLAYLEVEL(1, "Max window log exceeded by file (compression ratio will suffer)\n");
     comprParams->windowLog = MIN(ZSTD_WINDOWLOG_MAX, fileWindowLog);
-    if (fileWindowLog > ZSTD_cycleLog(cParams.hashLog, cParams.strategy)) {
+    if (fileWindowLog > ZSTD_cycleLog(cParams.chainLog, cParams.strategy)) {
         if (!prefs->ldmFlag)
-            DISPLAYLEVEL(1, "long mode automaticaly triggered\n");
+            DISPLAYLEVEL(1, "long mode automatically triggered\n");
         FIO_setLdmFlag(prefs, 1);
     }
     if (cParams.strategy >= ZSTD_btopt) {
@@ -838,8 +940,10 @@ static cRess_t FIO_createCResources(FIO_prefs_t* const prefs,
 
     /* need to update memLimit before calling createDictBuffer
      * because of memLimit check inside it */
-    if (prefs->patchFromMode)
-        FIO_adjustParamsForPatchFromMode(prefs, &comprParams, UTIL_getFileSize(dictFileName), maxSrcFileSize, cLevel);
+    if (prefs->patchFromMode) {
+        unsigned long long const ssSize = (unsigned long long)prefs->streamSrcSize;
+        FIO_adjustParamsForPatchFromMode(prefs, &comprParams, UTIL_getFileSize(dictFileName), ssSize > 0 ? ssSize : maxSrcFileSize, cLevel);
+    }
     ress.dstBuffer = malloc(ress.dstBufferSize);
     ress.dictBufferSize = FIO_createDictBuffer(&ress.dictBuffer, dictFileName, prefs);   /* works with dictFileName==NULL */
     if (!ress.srcBuffer || !ress.dstBuffer)
@@ -881,6 +985,7 @@ static cRess_t FIO_createCResources(FIO_prefs_t* const prefs,
     CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_targetLength, (int)comprParams.targetLength) );
     CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_strategy, comprParams.strategy) );
     CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_literalCompressionMode, (int)prefs->literalCompressionMode) );
+    CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_enableDedicatedDictSearch, 1) );
     /* multi-threading */
 #ifdef ZSTD_MULTITHREAD
     DISPLAYLEVEL(5,"set nb workers = %u \n", prefs->nbWorkers);
@@ -902,12 +1007,12 @@ static cRess_t FIO_createCResources(FIO_prefs_t* const prefs,
     return ress;
 }
 
-static void FIO_freeCResources(cRess_t ress)
+static void FIO_freeCResources(const cRess_t* const ress)
 {
-    free(ress.srcBuffer);
-    free(ress.dstBuffer);
-    free(ress.dictBuffer);
-    ZSTD_freeCStream(ress.cctx);   /* never fails */
+    free(ress->srcBuffer);
+    free(ress->dstBuffer);
+    free(ress->dictBuffer);
+    ZSTD_freeCStream(ress->cctx);   /* never fails */
 }
 
 
@@ -1172,7 +1277,8 @@ FIO_compressLz4Frame(cRess_t* ress,
 
 
 static unsigned long long
-FIO_compressZstdFrame(FIO_prefs_t* const prefs,
+FIO_compressZstdFrame(FIO_ctx_t* const fCtx,
+                      FIO_prefs_t* const prefs,
                       const cRess_t* ressPtr,
                       const char* srcFileName, U64 fileSize,
                       int compressionLevel, U64* readsize)
@@ -1254,11 +1360,24 @@ FIO_compressZstdFrame(FIO_prefs_t* const prefs,
                                 (unsigned)(zfp.consumed >> 20),
                                 (unsigned)(zfp.produced >> 20),
                                 cShare );
-                } else {   /* summarized notifications if == 2; */
-                    DISPLAYLEVEL(2, "\rRead : %u ", (unsigned)(zfp.consumed >> 20));
+                } else {   /* summarized notifications if == 2 */
+                    DISPLAYLEVEL(2, "\r%79s\r", "");    /* Clear out the current displayed line */
+                    if (fCtx->nbFilesTotal > 1) {
+                        size_t srcFileNameSize = strlen(srcFileName);
+                        /* Ensure that the string we print is roughly the same size each time */
+                        if (srcFileNameSize > 18) {
+                            const char* truncatedSrcFileName = srcFileName + srcFileNameSize - 15;
+                            DISPLAYLEVEL(2, "Compress: %u/%u files. Current: ...%s ",
+                                         fCtx->currFileIdx+1, fCtx->nbFilesTotal, truncatedSrcFileName);
+                        } else {
+                            DISPLAYLEVEL(2, "Compress: %u/%u files. Current: %*s ",
+                                         fCtx->currFileIdx+1, fCtx->nbFilesTotal, (int)(18-srcFileNameSize), srcFileName);
+                        }
+                    }
+                    DISPLAYLEVEL(2, "Read : %2u ", (unsigned)(zfp.consumed >> 20));
                     if (fileSize != UTIL_FILESIZE_UNKNOWN)
-                        DISPLAYLEVEL(2, "/ %u ", (unsigned)(fileSize >> 20));
-                    DISPLAYLEVEL(2, "MB ==> %2.f%% ", cShare);
+                        DISPLAYLEVEL(2, "/ %2u ", (unsigned)(fileSize >> 20));
+                    DISPLAYLEVEL(2, "MB ==> %2.f%%", cShare);
                     DELAY_NEXT_UPDATE();
                 }
 
@@ -1369,7 +1488,8 @@ FIO_compressZstdFrame(FIO_prefs_t* const prefs,
  *            1 : missing or pb opening srcFileName
  */
 static int
-FIO_compressFilename_internal(FIO_prefs_t* const prefs,
+FIO_compressFilename_internal(FIO_ctx_t* const fCtx,
+                              FIO_prefs_t* const prefs,
                               cRess_t ress,
                               const char* dstFileName, const char* srcFileName,
                               int compressionLevel)
@@ -1385,7 +1505,7 @@ FIO_compressFilename_internal(FIO_prefs_t* const prefs,
     switch (prefs->compressionType) {
         default:
         case FIO_zstdCompression:
-            compressedfilesize = FIO_compressZstdFrame(prefs, &ress, srcFileName, fileSize, compressionLevel, &readsize);
+            compressedfilesize = FIO_compressZstdFrame(fCtx, prefs, &ress, srcFileName, fileSize, compressionLevel, &readsize);
             break;
 
         case FIO_gzipCompression:
@@ -1421,18 +1541,24 @@ FIO_compressFilename_internal(FIO_prefs_t* const prefs,
     }
 
     /* Status */
+    fCtx->totalBytesInput += (size_t)readsize;
+    fCtx->totalBytesOutput += (size_t)compressedfilesize;
     DISPLAYLEVEL(2, "\r%79s\r", "");
-    if (readsize == 0) {
-        DISPLAYLEVEL(2,"%-20s :  (%6llu => %6llu bytes, %s) \n",
-            srcFileName,
-            (unsigned long long)readsize, (unsigned long long) compressedfilesize,
-            dstFileName);
-    } else {
-        DISPLAYLEVEL(2,"%-20s :%6.2f%%   (%6llu => %6llu bytes, %s) \n",
-            srcFileName,
-            (double)compressedfilesize / readsize * 100,
-            (unsigned long long)readsize, (unsigned long long) compressedfilesize,
-            dstFileName);
+    if (g_display_prefs.displayLevel >= 2 &&
+        !fCtx->hasStdoutOutput && 
+        (g_display_prefs.displayLevel >= 3 || fCtx->nbFilesTotal <= 1)) {
+        if (readsize == 0) {
+            DISPLAYLEVEL(2,"%-20s :  (%6llu => %6llu bytes, %s) \n",
+                srcFileName,
+                (unsigned long long)readsize, (unsigned long long) compressedfilesize,
+                dstFileName);
+        } else {
+            DISPLAYLEVEL(2,"%-20s :%6.2f%%   (%6llu => %6llu bytes, %s) \n",
+                srcFileName,
+                (double)compressedfilesize / readsize * 100,
+                (unsigned long long)readsize, (unsigned long long) compressedfilesize,
+                dstFileName);
+        }
     }
 
     /* Elapsed Time and CPU Load */
@@ -1457,7 +1583,8 @@ FIO_compressFilename_internal(FIO_prefs_t* const prefs,
  *  @return : 0 : compression completed correctly,
  *            1 : pb
  */
-static int FIO_compressFilename_dstFile(FIO_prefs_t* const prefs,
+static int FIO_compressFilename_dstFile(FIO_ctx_t* const fCtx,
+                                        FIO_prefs_t* const prefs,
                                         cRess_t ress,
                                         const char* dstFileName,
                                         const char* srcFileName,
@@ -1471,7 +1598,7 @@ static int FIO_compressFilename_dstFile(FIO_prefs_t* const prefs,
     if (ress.dstFile == NULL) {
         closeDstFile = 1;
         DISPLAYLEVEL(6, "FIO_compressFilename_dstFile: opening dst: %s \n", dstFileName);
-        ress.dstFile = FIO_openDstFile(prefs, srcFileName, dstFileName);
+        ress.dstFile = FIO_openDstFile(fCtx, prefs, srcFileName, dstFileName);
         if (ress.dstFile==NULL) return 1;  /* could not open dstFileName */
         /* Must only be added after FIO_openDstFile() succeeds.
          * Otherwise we may delete the destination file if it already exists,
@@ -1480,11 +1607,12 @@ static int FIO_compressFilename_dstFile(FIO_prefs_t* const prefs,
         addHandler(dstFileName);
 
         if ( strcmp (srcFileName, stdinmark)
-          && UTIL_getFileStat(srcFileName, &statbuf))
+          && UTIL_stat(srcFileName, &statbuf)
+          && UTIL_isRegularFileStat(&statbuf) )
             transfer_permissions = 1;
     }
 
-    result = FIO_compressFilename_internal(prefs, ress, dstFileName, srcFileName, compressionLevel);
+    result = FIO_compressFilename_internal(fCtx, prefs, ress, dstFileName, srcFileName, compressionLevel);
 
     if (closeDstFile) {
         FILE* const dstFile = ress.dstFile;
@@ -1498,14 +1626,11 @@ static int FIO_compressFilename_dstFile(FIO_prefs_t* const prefs,
             result=1;
         }
         if ( (result != 0)  /* operation failure */
-          && strcmp(dstFileName, nulmark)     /* special case : don't remove() /dev/null */
           && strcmp(dstFileName, stdoutmark)  /* special case : don't remove() stdout */
           ) {
-            FIO_remove(dstFileName); /* remove compression artefact; note don't do anything special if remove() fails */
-        } else if ( strcmp(dstFileName, stdoutmark)
-                 && strcmp(dstFileName, nulmark)
-                 && transfer_permissions) {
-            DISPLAYLEVEL(6, "FIO_compressFilename_dstFile: transfering permissions into dst: %s \n", dstFileName);
+            FIO_removeFile(dstFileName); /* remove compression artefact; note don't do anything special if remove() fails */
+        } else if (transfer_permissions) {
+            DISPLAYLEVEL(6, "FIO_compressFilename_dstFile: transferring permissions into dst: %s \n", dstFileName);
             UTIL_setFileStat(dstFileName, &statbuf);
         } else {
             DISPLAYLEVEL(6, "FIO_compressFilename_dstFile: do not transfer permissions into dst: %s \n", dstFileName);
@@ -1536,7 +1661,8 @@ static const char *compressedFileExtensions[] = {
  *            1 : missing or pb opening srcFileName
  */
 static int
-FIO_compressFilename_srcFile(FIO_prefs_t* const prefs,
+FIO_compressFilename_srcFile(FIO_ctx_t* const fCtx,
+                             FIO_prefs_t* const prefs,
                              cRess_t ress,
                              const char* dstFileName,
                              const char* srcFileName,
@@ -1569,7 +1695,7 @@ FIO_compressFilename_srcFile(FIO_prefs_t* const prefs,
     ress.srcFile = FIO_openSrcFile(srcFileName);
     if (ress.srcFile == NULL) return 1;   /* srcFile could not be opened */
 
-    result = FIO_compressFilename_dstFile(prefs, ress, dstFileName, srcFileName, compressionLevel);
+    result = FIO_compressFilename_dstFile(fCtx, prefs, ress, dstFileName, srcFileName, compressionLevel);
 
     fclose(ress.srcFile);
     ress.srcFile = NULL;
@@ -1581,21 +1707,22 @@ FIO_compressFilename_srcFile(FIO_prefs_t* const prefs,
          * delete both the source and destination files.
          */
         clearHandler();
-        if (FIO_remove(srcFileName))
+        if (FIO_removeFile(srcFileName))
             EXM_THROW(1, "zstd: %s: %s", srcFileName, strerror(errno));
     }
     return result;
 }
 
-int FIO_compressFilename(FIO_prefs_t* const prefs, const char* dstFileName,
+int FIO_compressFilename(FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs, const char* dstFileName,
                          const char* srcFileName, const char* dictFileName,
                          int compressionLevel, ZSTD_compressionParameters comprParams)
 {
     cRess_t const ress = FIO_createCResources(prefs, dictFileName, UTIL_getFileSize(srcFileName), compressionLevel, comprParams);
-    int const result = FIO_compressFilename_srcFile(prefs, ress, dstFileName, srcFileName, compressionLevel);
+    int const result = FIO_compressFilename_srcFile(fCtx, prefs, ress, dstFileName, srcFileName, compressionLevel);
 
+#define DISPLAY_LEVEL_DEFAULT 2
 
-    FIO_freeCResources(ress);
+    FIO_freeCResources(&ress);
     return result;
 }
 
@@ -1656,45 +1783,79 @@ static unsigned long long FIO_getLargestFileSize(const char** inFileNames, unsig
  * or into one file each (outFileName == NULL, but suffix != NULL),
  * or into a destination folder (specified with -O)
  */
-int FIO_compressMultipleFilenames(FIO_prefs_t* const prefs,
-                                  const char** inFileNamesTable, unsigned nbFiles,
+int FIO_compressMultipleFilenames(FIO_ctx_t* const fCtx,
+                                  FIO_prefs_t* const prefs,
+                                  const char** inFileNamesTable,
+                                  const char* outMirroredRootDirName,
                                   const char* outDirName,
                                   const char* outFileName, const char* suffix,
                                   const char* dictFileName, int compressionLevel,
                                   ZSTD_compressionParameters comprParams)
 {
+    int status;
     int error = 0;
     cRess_t ress = FIO_createCResources(prefs, dictFileName,
-        FIO_getLargestFileSize(inFileNamesTable, nbFiles),
+        FIO_getLargestFileSize(inFileNamesTable, fCtx->nbFilesTotal),
         compressionLevel, comprParams);
 
     /* init */
     assert(outFileName != NULL || suffix != NULL);
     if (outFileName != NULL) {   /* output into a single destination (stdout typically) */
-        ress.dstFile = FIO_openDstFile(prefs, NULL, outFileName);
+        if (FIO_removeMultiFilesWarning(fCtx, prefs, outFileName, 1 /* displayLevelCutoff */)) {
+            FIO_freeCResources(&ress);
+            return 1;
+        }
+        ress.dstFile = FIO_openDstFile(fCtx, prefs, NULL, outFileName);
         if (ress.dstFile == NULL) {  /* could not open outFileName */
             error = 1;
         } else {
-            unsigned u;
-            for (u=0; u<nbFiles; u++)
-                error |= FIO_compressFilename_srcFile(prefs, ress, outFileName, inFileNamesTable[u], compressionLevel);
+            for (; fCtx->currFileIdx < fCtx->nbFilesTotal; ++fCtx->currFileIdx) {
+                status = FIO_compressFilename_srcFile(fCtx, prefs, ress, outFileName, inFileNamesTable[fCtx->currFileIdx], compressionLevel);
+                if (!status) fCtx->nbFilesProcessed++;
+                error |= status;
+            }
             if (fclose(ress.dstFile))
                 EXM_THROW(29, "Write error (%s) : cannot properly close %s",
                             strerror(errno), outFileName);
             ress.dstFile = NULL;
         }
     } else {
-        unsigned u;
-        for (u=0; u<nbFiles; u++) {
-            const char* const srcFileName = inFileNamesTable[u];
-            const char* const dstFileName = FIO_determineCompressedName(srcFileName, outDirName, suffix);  /* cannot fail */
-            error |= FIO_compressFilename_srcFile(prefs, ress, dstFileName, srcFileName, compressionLevel);
+        if (outMirroredRootDirName)
+            UTIL_mirrorSourceFilesDirectories(inFileNamesTable, fCtx->nbFilesTotal, outMirroredRootDirName);
+
+        for (; fCtx->currFileIdx < fCtx->nbFilesTotal; ++fCtx->currFileIdx) {
+            const char* const srcFileName = inFileNamesTable[fCtx->currFileIdx];
+            const char* dstFileName = NULL;
+            if (outMirroredRootDirName) {
+                char* validMirroredDirName = UTIL_createMirroredDestDirName(srcFileName, outMirroredRootDirName);
+                if (validMirroredDirName) {
+                    dstFileName = FIO_determineCompressedName(srcFileName, validMirroredDirName, suffix);
+                    free(validMirroredDirName);
+                } else {
+                    DISPLAYLEVEL(2, "zstd: --output-dir-mirror cannot compress '%s' into '%s' \n", srcFileName, outMirroredRootDirName);
+                    error=1;
+                    continue;
+                }
+            } else {
+                dstFileName = FIO_determineCompressedName(srcFileName, outDirName, suffix);  /* cannot fail */
+            }
+            status = FIO_compressFilename_srcFile(fCtx, prefs, ress, dstFileName, srcFileName, compressionLevel);
+            if (!status) fCtx->nbFilesProcessed++;
+            error |= status;
         }
+
         if (outDirName)
-            FIO_checkFilenameCollisions(inFileNamesTable ,nbFiles);
+            FIO_checkFilenameCollisions(inFileNamesTable , fCtx->nbFilesTotal);
     }
 
-    FIO_freeCResources(ress);
+    if (fCtx->nbFilesProcessed >= 1 && fCtx->nbFilesTotal > 1 && fCtx->totalBytesInput != 0) {
+        DISPLAYLEVEL(2, "\r%79s\r", "");
+        DISPLAYLEVEL(2, "%d files compressed : %.2f%%  (%6zu => %6zu bytes)\n", fCtx->nbFilesProcessed,
+                        (double)fCtx->totalBytesOutput/((double)fCtx->totalBytesInput)*100,
+                        fCtx->totalBytesInput, fCtx->totalBytesOutput);
+    }
+
+    FIO_freeCResources(&ress);
     return error;
 }
 
@@ -1730,6 +1891,8 @@ static dRess_t FIO_createDResources(FIO_prefs_t* const prefs, const char* dictFi
     if (ress.dctx==NULL)
         EXM_THROW(60, "Error: %s : can't create ZSTD_DStream", strerror(errno));
     CHECK( ZSTD_DCtx_setMaxWindowSize(ress.dctx, prefs->memLimit) );
+    CHECK( ZSTD_DCtx_setParameter(ress.dctx, ZSTD_d_forceIgnoreChecksum, !prefs->checksumFlag));
+    
     ress.srcBufferSize = ZSTD_DStreamInSize();
     ress.srcBuffer = malloc(ress.srcBufferSize);
     ress.dstBufferSize = ZSTD_DStreamOutSize();
@@ -1923,7 +2086,7 @@ FIO_zstdErrorHelp(const FIO_prefs_t* const prefs,
  */
 #define FIO_ERROR_FRAME_DECODING   ((unsigned long long)(-2))
 static unsigned long long
-FIO_decompressZstdFrame(dRess_t* ress, FILE* finput,
+FIO_decompressZstdFrame(FIO_ctx_t* const fCtx, dRess_t* ress, FILE* finput,
                         const FIO_prefs_t* const prefs,
                         const char* srcFileName,
                         U64 alreadyDecoded)  /* for multi-frames streams */
@@ -1961,8 +2124,22 @@ FIO_decompressZstdFrame(dRess_t* ress, FILE* finput,
         /* Write block */
         storedSkips = FIO_fwriteSparse(ress->dstFile, ress->dstBuffer, outBuff.pos, prefs, storedSkips);
         frameSize += outBuff.pos;
-        DISPLAYUPDATE(2, "\r%-20.20s : %u MB...     ",
-                         srcFileName, (unsigned)((alreadyDecoded+frameSize)>>20) );
+        if (!fCtx->hasStdoutOutput) {
+            if (fCtx->nbFilesTotal > 1) {
+                size_t srcFileNameSize = strlen(srcFileName);
+                if (srcFileNameSize > 18) {
+                    const char* truncatedSrcFileName = srcFileName + srcFileNameSize - 15;
+                    DISPLAYUPDATE(2, "\rDecompress: %2u/%2u files. Current: ...%s : %u MB...    ",
+                                    fCtx->currFileIdx+1, fCtx->nbFilesTotal, truncatedSrcFileName, (unsigned)((alreadyDecoded+frameSize)>>20) );
+                } else {
+                    DISPLAYUPDATE(2, "\rDecompress: %2u/%2u files. Current: %s : %u MB...    ",
+                                fCtx->currFileIdx+1, fCtx->nbFilesTotal, srcFileName, (unsigned)((alreadyDecoded+frameSize)>>20) );
+                }
+            } else {
+                DISPLAYUPDATE(2, "\r%-20.20s : %u MB...     ",
+                                srcFileName, (unsigned)((alreadyDecoded+frameSize)>>20) );
+            }
+        }
 
         if (inBuff.pos > 0) {
             memmove(ress->srcBuffer, (char*)ress->srcBuffer + inBuff.pos, inBuff.size - inBuff.pos);
@@ -2220,7 +2397,8 @@ FIO_decompressLz4Frame(dRess_t* ress, FILE* srcFile,
  * @return : 0 : OK
  *           1 : error
  */
-static int FIO_decompressFrames(dRess_t ress, FILE* srcFile,
+static int FIO_decompressFrames(FIO_ctx_t* const fCtx,
+                          dRess_t ress, FILE* srcFile,
                           const FIO_prefs_t* const prefs,
                           const char* dstFileName, const char* srcFileName)
 {
@@ -2249,7 +2427,7 @@ static int FIO_decompressFrames(dRess_t ress, FILE* srcFile,
             return 1;
         }
         if (ZSTD_isFrame(buf, ress.srcBufferLoaded)) {
-            unsigned long long const frameSize = FIO_decompressZstdFrame(&ress, srcFile, prefs, srcFileName, filesize);
+            unsigned long long const frameSize = FIO_decompressZstdFrame(fCtx, &ress, srcFile, prefs, srcFileName, filesize);
             if (frameSize == FIO_ERROR_FRAME_DECODING) return 1;
             filesize += frameSize;
         } else if (buf[0] == 31 && buf[1] == 139) { /* gz magic number */
@@ -2291,8 +2469,14 @@ static int FIO_decompressFrames(dRess_t ress, FILE* srcFile,
     }   }  /* for each frame */
 
     /* Final Status */
+    fCtx->totalBytesOutput += (size_t)filesize;
     DISPLAYLEVEL(2, "\r%79s\r", "");
-    DISPLAYLEVEL(2, "%-20s: %llu bytes \n", srcFileName, filesize);
+    /* No status message in pipe mode (stdin - stdout) or multi-files mode */
+    if (g_display_prefs.displayLevel >= 2) {
+        if (fCtx->nbFilesTotal <= 1 || g_display_prefs.displayLevel >= 3) {
+            DISPLAYLEVEL(2, "%-20s: %llu bytes \n", srcFileName, filesize);
+        }
+    }
 
     return 0;
 }
@@ -2304,7 +2488,8 @@ static int FIO_decompressFrames(dRess_t ress, FILE* srcFile,
     @return : 0 : OK
               1 : operation aborted
 */
-static int FIO_decompressDstFile(FIO_prefs_t* const prefs,
+static int FIO_decompressDstFile(FIO_ctx_t* const fCtx,
+                                 FIO_prefs_t* const prefs,
                                  dRess_t ress, FILE* srcFile,
                                  const char* dstFileName, const char* srcFileName)
 {
@@ -2316,7 +2501,7 @@ static int FIO_decompressDstFile(FIO_prefs_t* const prefs,
     if ((ress.dstFile == NULL) && (prefs->testMode==0)) {
         releaseDstFile = 1;
 
-        ress.dstFile = FIO_openDstFile(prefs, srcFileName, dstFileName);
+        ress.dstFile = FIO_openDstFile(fCtx, prefs, srcFileName, dstFileName);
         if (ress.dstFile==NULL) return 1;
 
         /* Must only be added after FIO_openDstFile() succeeds.
@@ -2326,11 +2511,12 @@ static int FIO_decompressDstFile(FIO_prefs_t* const prefs,
         addHandler(dstFileName);
 
         if ( strcmp(srcFileName, stdinmark)   /* special case : don't transfer permissions from stdin */
-          && UTIL_getFileStat(srcFileName, &statbuf) )
+          && UTIL_stat(srcFileName, &statbuf)
+          && UTIL_isRegularFileStat(&statbuf) )
             transfer_permissions = 1;
     }
 
-    result = FIO_decompressFrames(ress, srcFile, prefs, dstFileName, srcFileName);
+    result = FIO_decompressFrames(fCtx, ress, srcFile, prefs, dstFileName, srcFileName);
 
     if (releaseDstFile) {
         FILE* const dstFile = ress.dstFile;
@@ -2342,15 +2528,11 @@ static int FIO_decompressDstFile(FIO_prefs_t* const prefs,
         }
 
         if ( (result != 0)  /* operation failure */
-          && strcmp(dstFileName, nulmark)     /* special case : don't remove() /dev/null (#316) */
           && strcmp(dstFileName, stdoutmark)  /* special case : don't remove() stdout */
           ) {
-            FIO_remove(dstFileName);  /* remove decompression artefact; note: don't do anything special if remove() fails */
-        } else {  /* operation success */
-            if ( strcmp(dstFileName, stdoutmark) /* special case : don't chmod stdout */
-              && strcmp(dstFileName, nulmark)    /* special case : don't chmod /dev/null */
-              && transfer_permissions )          /* file permissions correctly extracted from src */
-                UTIL_setFileStat(dstFileName, &statbuf);  /* transfer file permissions from src into dst */
+            FIO_removeFile(dstFileName);  /* remove decompression artefact; note: don't do anything special if remove() fails */
+        } else if ( transfer_permissions /* file permissions correctly extracted from src */ ) {
+            UTIL_setFileStat(dstFileName, &statbuf);  /* transfer file permissions from src into dst */
         }
     }
 
@@ -2363,7 +2545,7 @@ static int FIO_decompressDstFile(FIO_prefs_t* const prefs,
     @return : 0 : OK
               1 : error
 */
-static int FIO_decompressSrcFile(FIO_prefs_t* const prefs, dRess_t ress, const char* dstFileName, const char* srcFileName)
+static int FIO_decompressSrcFile(FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs, dRess_t ress, const char* dstFileName, const char* srcFileName)
 {
     FILE* srcFile;
     int result;
@@ -2377,7 +2559,7 @@ static int FIO_decompressSrcFile(FIO_prefs_t* const prefs, dRess_t ress, const c
     if (srcFile==NULL) return 1;
     ress.srcBufferLoaded = 0;
 
-    result = FIO_decompressDstFile(prefs, ress, srcFile, dstFileName, srcFileName);
+    result = FIO_decompressDstFile(fCtx, prefs, ress, srcFile, dstFileName, srcFileName);
 
     /* Close file */
     if (fclose(srcFile)) {
@@ -2391,7 +2573,7 @@ static int FIO_decompressSrcFile(FIO_prefs_t* const prefs, dRess_t ress, const c
          * delete both the source and destination files.
          */
         clearHandler();
-        if (FIO_remove(srcFileName)) {
+        if (FIO_removeFile(srcFileName)) {
             /* failed to remove src file */
             DISPLAYLEVEL(1, "zstd: %s: %s \n", srcFileName, strerror(errno));
             return 1;
@@ -2401,13 +2583,13 @@ static int FIO_decompressSrcFile(FIO_prefs_t* const prefs, dRess_t ress, const c
 
 
 
-int FIO_decompressFilename(FIO_prefs_t* const prefs,
+int FIO_decompressFilename(FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs,
                            const char* dstFileName, const char* srcFileName,
                            const char* dictFileName)
 {
     dRess_t const ress = FIO_createDResources(prefs, dictFileName);
 
-    int const decodingError = FIO_decompressSrcFile(prefs, ress, dstFileName, srcFileName);
+    int const decodingError = FIO_decompressSrcFile(fCtx, prefs, ress, dstFileName, srcFileName);
 
     FIO_freeDResources(ress);
     return decodingError;
@@ -2416,6 +2598,9 @@ int FIO_decompressFilename(FIO_prefs_t* const prefs,
 static const char *suffixList[] = {
     ZSTD_EXTENSION,
     TZSTD_EXTENSION,
+#ifndef ZSTD_NODECOMPRESS
+    ZSTD_ALT_EXTENSION,
+#endif
 #ifdef ZSTD_GZDECOMPRESS
     GZ_EXTENSION,
     TGZ_EXTENSION,
@@ -2531,39 +2716,64 @@ FIO_determineDstName(const char* srcFileName, const char* outDirName)
     /* note : dstFileNameBuffer memory is not going to be free */
 }
 
-
 int
-FIO_decompressMultipleFilenames(FIO_prefs_t* const prefs,
-                                const char** srcNamesTable, unsigned nbFiles,
+FIO_decompressMultipleFilenames(FIO_ctx_t* const fCtx,
+                                FIO_prefs_t* const prefs,
+                                const char** srcNamesTable,
+                                const char* outMirroredRootDirName,
                                 const char* outDirName, const char* outFileName,
                                 const char* dictFileName)
 {
+    int status;
     int error = 0;
     dRess_t ress = FIO_createDResources(prefs, dictFileName);
 
     if (outFileName) {
-        unsigned u;
+        if (FIO_removeMultiFilesWarning(fCtx, prefs, outFileName, 1 /* displayLevelCutoff */)) {
+            FIO_freeDResources(ress);
+            return 1;
+        }
         if (!prefs->testMode) {
-            ress.dstFile = FIO_openDstFile(prefs, NULL, outFileName);
+            ress.dstFile = FIO_openDstFile(fCtx, prefs, NULL, outFileName);
             if (ress.dstFile == 0) EXM_THROW(19, "cannot open %s", outFileName);
         }
-        for (u=0; u<nbFiles; u++)
-            error |= FIO_decompressSrcFile(prefs, ress, outFileName, srcNamesTable[u]);
+        for (; fCtx->currFileIdx < fCtx->nbFilesTotal; fCtx->currFileIdx++) {
+            status = FIO_decompressSrcFile(fCtx, prefs, ress, outFileName, srcNamesTable[fCtx->currFileIdx]);
+            if (!status) fCtx->nbFilesProcessed++;
+            error |= status;
+        }
         if ((!prefs->testMode) && (fclose(ress.dstFile)))
             EXM_THROW(72, "Write error : %s : cannot properly close output file",
                         strerror(errno));
     } else {
-        unsigned u;
-        for (u=0; u<nbFiles; u++) {   /* create dstFileName */
-            const char* const srcFileName = srcNamesTable[u];
-            const char* const dstFileName = FIO_determineDstName(srcFileName, outDirName);
+        if (outMirroredRootDirName)
+            UTIL_mirrorSourceFilesDirectories(srcNamesTable, fCtx->nbFilesTotal, outMirroredRootDirName);
+
+        for (; fCtx->currFileIdx < fCtx->nbFilesTotal; fCtx->currFileIdx++) {   /* create dstFileName */
+            const char* const srcFileName = srcNamesTable[fCtx->currFileIdx];
+            const char* dstFileName = NULL;
+            if (outMirroredRootDirName) {
+                char* validMirroredDirName = UTIL_createMirroredDestDirName(srcFileName, outMirroredRootDirName);
+                if (validMirroredDirName) {
+                    dstFileName = FIO_determineDstName(srcFileName, validMirroredDirName);
+                    free(validMirroredDirName);
+                } else {
+                    DISPLAYLEVEL(2, "zstd: --output-dir-mirror cannot decompress '%s' into '%s'\n", srcFileName, outMirroredRootDirName);
+                }
+            } else {
+                dstFileName = FIO_determineDstName(srcFileName, outDirName);
+            }
             if (dstFileName == NULL) { error=1; continue; }
-
-            error |= FIO_decompressSrcFile(prefs, ress, dstFileName, srcFileName);
+            status = FIO_decompressSrcFile(fCtx, prefs, ress, dstFileName, srcFileName);
+            if (!status) fCtx->nbFilesProcessed++;
+            error |= status;
         }
         if (outDirName)
-            FIO_checkFilenameCollisions(srcNamesTable ,nbFiles);
+            FIO_checkFilenameCollisions(srcNamesTable , fCtx->nbFilesTotal);
     }
+    
+    if (fCtx->nbFilesProcessed >= 1  && fCtx->nbFilesTotal > 1 && fCtx->totalBytesOutput != 0)
+        DISPLAYLEVEL(2, "%d files decompressed : %6zu bytes total \n", fCtx->nbFilesProcessed, fCtx->totalBytesOutput);
 
     FIO_freeDResources(ress);
     return error;
diff --git a/programs/fileio.h b/programs/fileio.h
index 2fbf01f82824..05e6d06815f7 100644
--- a/programs/fileio.h
+++ b/programs/fileio.h
@@ -44,6 +44,7 @@ extern "C" {
 
 #define ZSTD_EXTENSION  ".zst"
 #define TZSTD_EXTENSION ".tzst"
+#define ZSTD_ALT_EXTENSION  ".zstd" /* allow decompression of .zstd files */
 
 #define LZ4_EXTENSION   ".lz4"
 #define TLZ4_EXTENSION  ".tlz4"
@@ -59,11 +60,18 @@ typedef struct FIO_prefs_s FIO_prefs_t;
 FIO_prefs_t* FIO_createPreferences(void);
 void FIO_freePreferences(FIO_prefs_t* const prefs);
 
+/* Mutable struct containing relevant context and state regarding (de)compression with respect to file I/O */
+typedef struct FIO_ctx_s FIO_ctx_t;
+
+FIO_ctx_t* FIO_createContext(void);
+void FIO_freeContext(FIO_ctx_t* const fCtx);
+
 typedef struct FIO_display_prefs_s FIO_display_prefs_t;
 
 /*-*************************************
 *  Parameters
 ***************************************/
+/* FIO_prefs_t functions */
 void FIO_setCompressionType(FIO_prefs_t* const prefs, FIO_compressionType_t compressionType);
 void FIO_overwriteMode(FIO_prefs_t* const prefs);
 void FIO_setAdaptiveMode(FIO_prefs_t* const prefs, unsigned adapt);
@@ -97,19 +105,24 @@ void FIO_setExcludeCompressedFile(FIO_prefs_t* const prefs, int excludeCompresse
 void FIO_setPatchFromMode(FIO_prefs_t* const prefs, int value);
 void FIO_setContentSize(FIO_prefs_t* const prefs, int value);
 
+/* FIO_ctx_t functions */
+void FIO_setNbFilesTotal(FIO_ctx_t* const fCtx, int value);
+void FIO_setHasStdoutOutput(FIO_ctx_t* const fCtx, int value);
+void FIO_determineHasStdinInput(FIO_ctx_t* const fCtx, const FileNamesTable* const filenames);
+
 /*-*************************************
 *  Single File functions
 ***************************************/
 /** FIO_compressFilename() :
  * @return : 0 == ok;  1 == pb with src file. */
-int FIO_compressFilename (FIO_prefs_t* const prefs,
+int FIO_compressFilename (FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs,
                           const char* outfilename, const char* infilename,
                           const char* dictFileName, int compressionLevel,
                           ZSTD_compressionParameters comprParams);
 
 /** FIO_decompressFilename() :
  * @return : 0 == ok;  1 == pb with src file. */
-int FIO_decompressFilename (FIO_prefs_t* const prefs,
+int FIO_decompressFilename (FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs,
                             const char* outfilename, const char* infilename, const char* dictFileName);
 
 int FIO_listMultipleFiles(unsigned numFiles, const char** filenameTable, int displayLevel);
@@ -120,8 +133,10 @@ int FIO_listMultipleFiles(unsigned numFiles, const char** filenameTable, int dis
 ***************************************/
 /** FIO_compressMultipleFilenames() :
  * @return : nb of missing files */
-int FIO_compressMultipleFilenames(FIO_prefs_t* const prefs,
-                                  const char** inFileNamesTable, unsigned nbFiles,
+int FIO_compressMultipleFilenames(FIO_ctx_t* const fCtx,
+                                  FIO_prefs_t* const prefs,
+                                  const char** inFileNamesTable,
+                                  const char* outMirroredDirName,
                                   const char* outDirName,
                                   const char* outFileName, const char* suffix,
                                   const char* dictFileName, int compressionLevel,
@@ -129,8 +144,10 @@ int FIO_compressMultipleFilenames(FIO_prefs_t* const prefs,
 
 /** FIO_decompressMultipleFilenames() :
  * @return : nb of missing or skipped files */
-int FIO_decompressMultipleFilenames(FIO_prefs_t* const prefs,
-                                    const char** srcNamesTable, unsigned nbFiles,
+int FIO_decompressMultipleFilenames(FIO_ctx_t* const fCtx,
+                                    FIO_prefs_t* const prefs,
+                                    const char** srcNamesTable,
+                                    const char* outMirroredDirName,
                                     const char* outDirName,
                                     const char* outFileName,
                                     const char* dictFileName);
diff --git a/programs/platform.h b/programs/platform.h
index 2b4b9f2d8677..68be70bb333b 100644
--- a/programs/platform.h
+++ b/programs/platform.h
@@ -102,6 +102,12 @@ extern "C" {
 #      define PLATFORM_POSIX_VERSION 1
 #    endif
 
+#    ifdef __UCLIBC__
+#     ifndef __USE_MISC
+#      define __USE_MISC /* enable st_mtim on uclibc */
+#     endif
+#    endif
+
 #  else  /* non-unix target platform (like Windows) */
 #    define PLATFORM_POSIX_VERSION 0
 #  endif
diff --git a/programs/timefn.h b/programs/timefn.h
index eb3c130934eb..5d2818e8a1b7 100644
--- a/programs/timefn.h
+++ b/programs/timefn.h
@@ -28,7 +28,11 @@ extern "C" {
 ******************************************/
 
 #if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
-# include <stdint.h>
+# if defined(_AIX)
+#  include <inttypes.h>
+# else
+#  include <stdint.h> /* intptr_t */
+# endif
   typedef uint64_t           PTime;  /* Precise Time */
 #else
   typedef unsigned long long PTime;  /* does not support compilers without long long support */
diff --git a/programs/util.c b/programs/util.c
index ab1abd3b1862..5386d005c26c 100644
--- a/programs/util.c
+++ b/programs/util.c
@@ -28,7 +28,7 @@ extern "C" {
 #  include <io.h>         /* _chmod */
 #else
 #  include <unistd.h>     /* chown, stat */
-#  if PLATFORM_POSIX_VERSION < 200809L
+#  if PLATFORM_POSIX_VERSION < 200809L || !defined(st_mtime)
 #    include <utime.h>    /* utime */
 #  else
 #    include <fcntl.h>    /* AT_FDCWD */
@@ -45,7 +45,6 @@ extern "C" {
 #  include <string.h>       /* strerror, memcpy */
 #endif /* #ifdef _WIN32 */
 
-
 /*-****************************************
 *  Internal Macros
 ******************************************/
@@ -88,6 +87,28 @@ UTIL_STATIC void* UTIL_realloc(void *ptr, size_t size)
 ******************************************/
 int g_utilDisplayLevel;
 
+int UTIL_requireUserConfirmation(const char* prompt, const char* abortMsg,
+                                 const char* acceptableLetters, int hasStdinInput) {
+    int ch, result;
+
+    if (hasStdinInput) {
+        UTIL_DISPLAY("stdin is an input - not proceeding.\n");
+        return 1;
+    }
+
+    UTIL_DISPLAY("%s", prompt);
+    ch = getchar();
+    result = 0;
+    if (strchr(acceptableLetters, ch) == NULL) {
+        UTIL_DISPLAY("%s", abortMsg);
+        result = 1;
+    }
+    /* flush the rest */
+    while ((ch!=EOF) && (ch!='\n'))
+        ch = getchar();
+    return result;
+}
+
 
 /*-*************************************
 *  Constants
@@ -100,48 +121,50 @@ int g_utilDisplayLevel;
 *  Functions
 ***************************************/
 
-int UTIL_fileExist(const char* filename)
+int UTIL_stat(const char* filename, stat_t* statbuf)
 {
-    stat_t statbuf;
 #if defined(_MSC_VER)
-    int const stat_error = _stat64(filename, &statbuf);
+    return !_stat64(filename, statbuf);
+#elif defined(__MINGW32__) && defined (__MSVCRT__)
+    return !_stati64(filename, statbuf);
 #else
-    int const stat_error = stat(filename, &statbuf);
+    return !stat(filename, statbuf);
 #endif
-    return !stat_error;
 }
 
 int UTIL_isRegularFile(const char* infilename)
 {
     stat_t statbuf;
-    return UTIL_getFileStat(infilename, &statbuf); /* Only need to know whether it is a regular file */
+    return UTIL_stat(infilename, &statbuf) && UTIL_isRegularFileStat(&statbuf);
 }
 
-int UTIL_getFileStat(const char* infilename, stat_t *statbuf)
+int UTIL_isRegularFileStat(const stat_t* statbuf)
 {
-    int r;
 #if defined(_MSC_VER)
-    r = _stat64(infilename, statbuf);
-    if (r || !(statbuf->st_mode & S_IFREG)) return 0;   /* No good... */
+    return (statbuf->st_mode & S_IFREG) != 0;
 #else
-    r = stat(infilename, statbuf);
-    if (r || !S_ISREG(statbuf->st_mode)) return 0;   /* No good... */
+    return S_ISREG(statbuf->st_mode) != 0;
 #endif
-    return 1;
 }
 
 /* like chmod, but avoid changing permission of /dev/null */
-int UTIL_chmod(char const* filename, mode_t permissions)
+int UTIL_chmod(char const* filename, const stat_t* statbuf, mode_t permissions)
 {
-    if (!strcmp(filename, "/dev/null")) return 0;   /* pretend success, but don't change anything */
+    stat_t localStatBuf;
+    if (statbuf == NULL) {
+        if (!UTIL_stat(filename, &localStatBuf)) return 0;
+        statbuf = &localStatBuf;
+    }
+    if (!UTIL_isRegularFileStat(statbuf)) return 0; /* pretend success, but don't change anything */
     return chmod(filename, permissions);
 }
 
-int UTIL_setFileStat(const char *filename, stat_t *statbuf)
+int UTIL_setFileStat(const char *filename, const stat_t *statbuf)
 {
     int res = 0;
 
-    if (!UTIL_isRegularFile(filename))
+    stat_t curStatBuf;
+    if (!UTIL_stat(filename, &curStatBuf) || !UTIL_isRegularFileStat(&curStatBuf))
         return -1;
 
     /* set access and modification times */
@@ -169,7 +192,7 @@ int UTIL_setFileStat(const char *filename, stat_t *statbuf)
     res += chown(filename, statbuf->st_uid, statbuf->st_gid);  /* Copy ownership */
 #endif
 
-    res += UTIL_chmod(filename, statbuf->st_mode & 07777);  /* Copy file permissions */
+    res += UTIL_chmod(filename, &curStatBuf, statbuf->st_mode & 07777);  /* Copy file permissions */
 
     errno = 0;
     return -res; /* number of errors is returned */
@@ -178,14 +201,16 @@ int UTIL_setFileStat(const char *filename, stat_t *statbuf)
 int UTIL_isDirectory(const char* infilename)
 {
     stat_t statbuf;
+    return UTIL_stat(infilename, &statbuf) && UTIL_isDirectoryStat(&statbuf);
+}
+
+int UTIL_isDirectoryStat(const stat_t* statbuf)
+{
 #if defined(_MSC_VER)
-    int const r = _stat64(infilename, &statbuf);
-    if (!r && (statbuf.st_mode & _S_IFDIR)) return 1;
+    return (statbuf->st_mode & _S_IFDIR) != 0;
 #else
-    int const r = stat(infilename, &statbuf);
-    if (!r && S_ISDIR(statbuf.st_mode)) return 1;
+    return S_ISDIR(statbuf->st_mode) != 0;
 #endif
-    return 0;
 }
 
 int UTIL_compareStr(const void *p1, const void *p2) {
@@ -204,8 +229,8 @@ int UTIL_isSameFile(const char* fName1, const char* fName2)
 #else
     {   stat_t file1Stat;
         stat_t file2Stat;
-        return UTIL_getFileStat(fName1, &file1Stat)
-            && UTIL_getFileStat(fName2, &file2Stat)
+        return UTIL_stat(fName1, &file1Stat)
+            && UTIL_stat(fName2, &file2Stat)
             && (file1Stat.st_dev == file2Stat.st_dev)
             && (file1Stat.st_ino == file2Stat.st_ino);
     }
@@ -218,13 +243,23 @@ int UTIL_isFIFO(const char* infilename)
 /* macro guards, as defined in : https://linux.die.net/man/2/lstat */
 #if PLATFORM_POSIX_VERSION >= 200112L
     stat_t statbuf;
-    int const r = UTIL_getFileStat(infilename, &statbuf);
-    if (!r && S_ISFIFO(statbuf.st_mode)) return 1;
+    if (UTIL_stat(infilename, &statbuf) && UTIL_isFIFOStat(&statbuf)) return 1;
 #endif
     (void)infilename;
     return 0;
 }
 
+/* UTIL_isFIFO : distinguish named pipes */
+int UTIL_isFIFOStat(const stat_t* statbuf)
+{
+/* macro guards, as defined in : https://linux.die.net/man/2/lstat */
+#if PLATFORM_POSIX_VERSION >= 200112L
+    if (S_ISFIFO(statbuf->st_mode)) return 1;
+#endif
+    (void)statbuf;
+    return 0;
+}
+
 int UTIL_isLink(const char* infilename)
 {
 /* macro guards, as defined in : https://linux.die.net/man/2/lstat */
@@ -239,23 +274,22 @@ int UTIL_isLink(const char* infilename)
 
 U64 UTIL_getFileSize(const char* infilename)
 {
-    if (!UTIL_isRegularFile(infilename)) return UTIL_FILESIZE_UNKNOWN;
-    {   int r;
+    stat_t statbuf;
+    if (!UTIL_stat(infilename, &statbuf)) return UTIL_FILESIZE_UNKNOWN;
+    return UTIL_getFileSizeStat(&statbuf);
+}
+
+U64 UTIL_getFileSizeStat(const stat_t* statbuf)
+{
+    if (!UTIL_isRegularFileStat(statbuf)) return UTIL_FILESIZE_UNKNOWN;
 #if defined(_MSC_VER)
-        struct __stat64 statbuf;
-        r = _stat64(infilename, &statbuf);
-        if (r || !(statbuf.st_mode & S_IFREG)) return UTIL_FILESIZE_UNKNOWN;
+    if (!(statbuf->st_mode & S_IFREG)) return UTIL_FILESIZE_UNKNOWN;
 #elif defined(__MINGW32__) && defined (__MSVCRT__)
-        struct _stati64 statbuf;
-        r = _stati64(infilename, &statbuf);
-        if (r || !(statbuf.st_mode & S_IFREG)) return UTIL_FILESIZE_UNKNOWN;
+    if (!(statbuf->st_mode & S_IFREG)) return UTIL_FILESIZE_UNKNOWN;
 #else
-        struct stat statbuf;
-        r = stat(infilename, &statbuf);
-        if (r || !S_ISREG(statbuf.st_mode)) return UTIL_FILESIZE_UNKNOWN;
+    if (!S_ISREG(statbuf->st_mode)) return UTIL_FILESIZE_UNKNOWN;
 #endif
-        return (U64)statbuf.st_size;
-    }
+    return (U64)statbuf->st_size;
 }
 
 
@@ -332,11 +366,12 @@ UTIL_createFileNamesTable_fromFileName(const char* inputFileName)
     char* buf;
     size_t bufSize;
     size_t pos = 0;
+    stat_t statbuf;
 
-    if (!UTIL_fileExist(inputFileName) || !UTIL_isRegularFile(inputFileName))
+    if (!UTIL_stat(inputFileName, &statbuf) || !UTIL_isRegularFileStat(&statbuf))
         return NULL;
 
-    {   U64 const inputFileSize = UTIL_getFileSize(inputFileName);
+    {   U64 const inputFileSize = UTIL_getFileSizeStat(&statbuf);
         if(inputFileSize > MAX_FILE_OF_FILE_NAMES_SIZE)
             return NULL;
         bufSize = (size_t)(inputFileSize + 1); /* (+1) to add '\0' at the end of last filename */
@@ -633,6 +668,290 @@ const char* UTIL_getFileExtension(const char* infilename)
    return extension;
 }
 
+static int pathnameHas2Dots(const char *pathname)
+{
+    return NULL != strstr(pathname, "..");
+}
+
+static int isFileNameValidForMirroredOutput(const char *filename)
+{
+    return !pathnameHas2Dots(filename);
+}
+
+
+#define DIR_DEFAULT_MODE 0755
+static mode_t getDirMode(const char *dirName)
+{
+    stat_t st;
+    if (!UTIL_stat(dirName, &st)) {
+        UTIL_DISPLAY("zstd: failed to get DIR stats %s: %s\n", dirName, strerror(errno));
+        return DIR_DEFAULT_MODE;
+    }
+    if (!UTIL_isDirectoryStat(&st)) {
+        UTIL_DISPLAY("zstd: expected directory: %s\n", dirName);
+        return DIR_DEFAULT_MODE;
+    }
+    return st.st_mode;
+}
+
+static int makeDir(const char *dir, mode_t mode)
+{
+#if defined(_MSC_VER) || defined(__MINGW32__) || defined (__MSVCRT__)
+    int ret = _mkdir(dir);
+    (void) mode;
+#else
+    int ret = mkdir(dir, mode);
+#endif
+    if (ret != 0) {
+        if (errno == EEXIST)
+            return 0;
+        UTIL_DISPLAY("zstd: failed to create DIR %s: %s\n", dir, strerror(errno));
+    }
+    return ret;
+}
+
+/* this function requires a mutable input string */
+static void convertPathnameToDirName(char *pathname)
+{
+    size_t len = 0;
+    char* pos = NULL;
+    /* get dir name from pathname similar to 'dirname()' */
+    assert(pathname != NULL);
+
+    /* remove trailing '/' chars */
+    len = strlen(pathname);
+    assert(len > 0);
+    while (pathname[len] == PATH_SEP) {
+        pathname[len] = '\0';
+        len--;
+    }
+    if (len == 0) return;
+
+    /* if input is a single file, return '.' instead. i.e.
+     * "xyz/abc/file.txt" => "xyz/abc"
+       "./file.txt"       => "."
+       "file.txt"         => "."
+     */
+    pos = strrchr(pathname, PATH_SEP);
+    if (pos == NULL) {
+        pathname[0] = '.';
+        pathname[1] = '\0';
+    } else {
+        *pos = '\0';
+    }
+}
+
+/* pathname must be valid */
+static const char* trimLeadingRootChar(const char *pathname)
+{
+    assert(pathname != NULL);
+    if (pathname[0] == PATH_SEP)
+        return pathname + 1;
+    return pathname;
+}
+
+/* pathname must be valid */
+static const char* trimLeadingCurrentDirConst(const char *pathname)
+{
+    assert(pathname != NULL);
+    if ((pathname[0] == '.') && (pathname[1] == PATH_SEP))
+        return pathname + 2;
+    return pathname;
+}
+
+static char*
+trimLeadingCurrentDir(char *pathname)
+{
+    /* 'union charunion' can do const-cast without compiler warning */
+    union charunion {
+        char *chr;
+        const char* cchr;
+    } ptr;
+    ptr.cchr = trimLeadingCurrentDirConst(pathname);
+    return ptr.chr;
+}
+
+/* remove leading './' or '/' chars here */
+static const char * trimPath(const char *pathname)
+{
+    return trimLeadingRootChar(
+            trimLeadingCurrentDirConst(pathname));
+}
+
+static char* mallocAndJoin2Dir(const char *dir1, const char *dir2)
+{
+    const size_t dir1Size = strlen(dir1);
+    const size_t dir2Size = strlen(dir2);
+    char *outDirBuffer, *buffer, trailingChar;
+
+    assert(dir1 != NULL && dir2 != NULL);
+    outDirBuffer = (char *) malloc(dir1Size + dir2Size + 2);
+    CONTROL(outDirBuffer != NULL);
+
+    memcpy(outDirBuffer, dir1, dir1Size);
+    outDirBuffer[dir1Size] = '\0';
+
+    if (dir2[0] == '.')
+        return outDirBuffer;
+
+    buffer = outDirBuffer + dir1Size;
+    trailingChar = *(buffer - 1);
+    if (trailingChar != PATH_SEP) {
+        *buffer = PATH_SEP;
+        buffer++;
+    }
+    memcpy(buffer, dir2, dir2Size);
+    buffer[dir2Size] = '\0';
+
+    return outDirBuffer;
+}
+
+/* this function will return NULL if input srcFileName is not valid name for mirrored output path */
+char* UTIL_createMirroredDestDirName(const char* srcFileName, const char* outDirRootName)
+{
+    char* pathname = NULL;
+    if (!isFileNameValidForMirroredOutput(srcFileName))
+        return NULL;
+
+    pathname = mallocAndJoin2Dir(outDirRootName, trimPath(srcFileName));
+
+    convertPathnameToDirName(pathname);
+    return pathname;
+}
+
+static int
+mirrorSrcDir(char* srcDirName, const char* outDirName)
+{
+    mode_t srcMode;
+    int status = 0;
+    char* newDir = mallocAndJoin2Dir(outDirName, trimPath(srcDirName));
+    if (!newDir)
+        return -ENOMEM;
+
+    srcMode = getDirMode(srcDirName);
+    status = makeDir(newDir, srcMode);
+    free(newDir);
+    return status;
+}
+
+static int
+mirrorSrcDirRecursive(char* srcDirName, const char* outDirName)
+{
+    int status = 0;
+    char* pp = trimLeadingCurrentDir(srcDirName);
+    char* sp = NULL;
+
+    while ((sp = strchr(pp, PATH_SEP)) != NULL) {
+        if (sp != pp) {
+            *sp = '\0';
+            status = mirrorSrcDir(srcDirName, outDirName);
+            if (status != 0)
+                return status;
+            *sp = PATH_SEP;
+        }
+        pp = sp + 1;
+    }
+    status = mirrorSrcDir(srcDirName, outDirName);
+    return status;
+}
+
+static void
+makeMirroredDestDirsWithSameSrcDirMode(char** srcDirNames, unsigned nbFile, const char* outDirName)
+{
+    unsigned int i = 0;
+    for (i = 0; i < nbFile; i++)
+        mirrorSrcDirRecursive(srcDirNames[i], outDirName);
+}
+
+static int
+firstIsParentOrSameDirOfSecond(const char* firstDir, const char* secondDir)
+{
+    size_t firstDirLen  = strlen(firstDir),
+           secondDirLen = strlen(secondDir);
+    return firstDirLen <= secondDirLen &&
+           (secondDir[firstDirLen] == PATH_SEP || secondDir[firstDirLen] == '\0') &&
+           0 == strncmp(firstDir, secondDir, firstDirLen);
+}
+
+static int compareDir(const void* pathname1, const void* pathname2) {
+    /* sort it after remove the leading '/'  or './'*/
+    const char* s1 = trimPath(*(char * const *) pathname1);
+    const char* s2 = trimPath(*(char * const *) pathname2);
+    return strcmp(s1, s2);
+}
+
+static void
+makeUniqueMirroredDestDirs(char** srcDirNames, unsigned nbFile, const char* outDirName)
+{
+    unsigned int i = 0, uniqueDirNr = 0;
+    char** uniqueDirNames = NULL;
+
+    if (nbFile == 0)
+        return;
+
+    uniqueDirNames = (char** ) malloc(nbFile * sizeof (char *));
+    CONTROL(uniqueDirNames != NULL);
+
+    /* if dirs is "a/b/c" and "a/b/c/d", we only need call:
+     * we just need "a/b/c/d" */
+    qsort((void *)srcDirNames, nbFile, sizeof(char*), compareDir);
+
+    uniqueDirNr = 1;
+    uniqueDirNames[uniqueDirNr - 1] = srcDirNames[0];
+    for (i = 1; i < nbFile; i++) {
+        char* prevDirName = srcDirNames[i - 1];
+        char* currDirName = srcDirNames[i];
+
+        /* note: we alwasy compare trimmed path, i.e.:
+         * src dir of "./foo" and "/foo" will be both saved into:
+         * "outDirName/foo/" */
+        if (!firstIsParentOrSameDirOfSecond(trimPath(prevDirName),
+                                            trimPath(currDirName)))
+            uniqueDirNr++;
+
+        /* we need maintain original src dir name instead of trimmed
+         * dir, so we can retrive the original src dir's mode_t */
+        uniqueDirNames[uniqueDirNr - 1] = currDirName;
+    }
+
+    makeMirroredDestDirsWithSameSrcDirMode(uniqueDirNames, uniqueDirNr, outDirName);
+
+    free(uniqueDirNames);
+}
+
+static void
+makeMirroredDestDirs(char** srcFileNames, unsigned nbFile, const char* outDirName)
+{
+    unsigned int i = 0;
+    for (i = 0; i < nbFile; ++i)
+        convertPathnameToDirName(srcFileNames[i]);
+    makeUniqueMirroredDestDirs(srcFileNames, nbFile, outDirName);
+}
+
+void UTIL_mirrorSourceFilesDirectories(const char** inFileNames, unsigned int nbFile, const char* outDirName)
+{
+    unsigned int i = 0, validFilenamesNr = 0;
+    char** srcFileNames = (char **) malloc(nbFile * sizeof (char *));
+    CONTROL(srcFileNames != NULL);
+
+    /* check input filenames is valid */
+    for (i = 0; i < nbFile; ++i) {
+        if (isFileNameValidForMirroredOutput(inFileNames[i])) {
+            char* fname = STRDUP(inFileNames[i]);
+            CONTROL(fname != NULL);
+            srcFileNames[validFilenamesNr++] = fname;
+        }
+    }
+
+    if (validFilenamesNr > 0) {
+        makeDir(outDirName, DIR_DEFAULT_MODE);
+        makeMirroredDestDirs(srcFileNames, validFilenamesNr, outDirName);
+    }
+
+    for (i = 0; i < validFilenamesNr; i++)
+        free(srcFileNames[i]);
+    free(srcFileNames);
+}
 
 FileNamesTable*
 UTIL_createExpandedFNT(const char** inputNames, size_t nbIfns, int followLinks)
diff --git a/programs/util.h b/programs/util.h
index 8e187e4f2999..25fa3f53aab4 100644
--- a/programs/util.h
+++ b/programs/util.h
@@ -93,6 +93,13 @@ extern "C" {
 ******************************************/
 extern int g_utilDisplayLevel;
 
+/**
+ * Displays a message prompt and returns success (0) if first character from stdin
+ * matches any from acceptableLetters. Otherwise, returns failure (1) and displays abortMsg.
+ * If any of the inputs are stdin itself, then automatically return failure (1).
+ */
+int UTIL_requireUserConfirmation(const char* prompt, const char* abortMsg, const char* acceptableLetters, int hasStdinInput);
+
 
 /*-****************************************
 *  File functions
@@ -100,12 +107,57 @@ extern int g_utilDisplayLevel;
 #if defined(_MSC_VER)
     typedef struct __stat64 stat_t;
     typedef int mode_t;
+#elif defined(__MINGW32__) && defined (__MSVCRT__)
+    typedef struct _stati64 stat_t;
 #else
     typedef struct stat stat_t;
 #endif
 
+#if defined(_MSC_VER) || defined(__MINGW32__) || defined (__MSVCRT__) /* windows support */
+#define PATH_SEP '\\'
+#define STRDUP(s) _strdup(s)
+#else
+#define PATH_SEP '/'
+#include <libgen.h>
+#define STRDUP(s) strdup(s)
+#endif
+
+/**
+ * Calls platform's equivalent of stat() on filename and writes info to statbuf.
+ * Returns success (1) or failure (0).
+ */
+int UTIL_stat(const char* filename, stat_t* statbuf);
+
+/**
+ * Instead of getting a file's stats, this updates them with the info in the
+ * provided stat_t. Currently sets owner, group, atime, and mtime. Will only
+ * update this info for regular files.
+ */
+int UTIL_setFileStat(const char* filename, const stat_t* statbuf);
+
+/*
+ * These helpers operate on a pre-populated stat_t, i.e., the result of
+ * calling one of the above functions.
+ */
+
+int UTIL_isRegularFileStat(const stat_t* statbuf);
+int UTIL_isDirectoryStat(const stat_t* statbuf);
+int UTIL_isFIFOStat(const stat_t* statbuf);
+U64 UTIL_getFileSizeStat(const stat_t* statbuf);
+
+/**
+ * Like chmod(), but only modifies regular files. Provided statbuf may be NULL,
+ * in which case this function will stat() the file internally, in order to
+ * check whether it should be modified.
+ */
+int UTIL_chmod(char const* filename, const stat_t* statbuf, mode_t permissions);
+
+/*
+ * In the absence of a pre-existing stat result on the file in question, these
+ * functions will do a stat() call internally and then use that result to
+ * compute the needed information.
+ */
 
-int UTIL_fileExist(const char* filename);
 int UTIL_isRegularFile(const char* infilename);
 int UTIL_isDirectory(const char* infilename);
 int UTIL_isSameFile(const char* file1, const char* file2);
@@ -116,11 +168,12 @@ int UTIL_isFIFO(const char* infilename);
 #define UTIL_FILESIZE_UNKNOWN  ((U64)(-1))
 U64 UTIL_getFileSize(const char* infilename);
 U64 UTIL_getTotalFileSize(const char* const * fileNamesTable, unsigned nbFiles);
-int UTIL_getFileStat(const char* infilename, stat_t* statbuf);
-int UTIL_setFileStat(const char* filename, stat_t* statbuf);
-int UTIL_chmod(char const* filename, mode_t permissions);   /*< like chmod, but avoid changing permission of /dev/null */
+
 int UTIL_compareStr(const void *p1, const void *p2);
 const char* UTIL_getFileExtension(const char* infilename);
+void  UTIL_mirrorSourceFilesDirectories(const char** fileNamesTable, unsigned int nbFiles, const char *outDirName);
+char* UTIL_createMirroredDestDirName(const char* srcFileName, const char* outDirRootName);
+
 
 
 /*-****************************************
@@ -207,6 +260,7 @@ void UTIL_refFilename(FileNamesTable* fnt, const char* filename);
 #  define UTIL_HAS_CREATEFILELIST
 #elif defined(__linux__) || (PLATFORM_POSIX_VERSION >= 200112L)  /* opendir, readdir require POSIX.1-2001 */
 #  define UTIL_HAS_CREATEFILELIST
+#  define UTIL_HAS_MIRRORFILELIST
 #else
    /* do not define UTIL_HAS_CREATEFILELIST */
 #endif
diff --git a/programs/zstd.1 b/programs/zstd.1
index 9ba0b4fa4162..a8fc277672b6 100644
--- a/programs/zstd.1
+++ b/programs/zstd.1
@@ -1,5 +1,5 @@
 .
-.TH "ZSTD" "1" "May 2020" "zstd 1.4.5" "User Commands"
+.TH "ZSTD" "1" "December 2020" "zstd 1.4.8" "User Commands"
 .
 .SH "NAME"
 \fBzstd\fR \- zstd, zstdmt, unzstd, zstdcat \- Compress or decompress \.zst files
@@ -99,10 +99,19 @@ Display information related to a zstd compressed file, such as size, ratio, and
 \fB\-#\fR: \fB#\fR compression level [1\-19] (default: 3)
 .
 .IP "\(bu" 4
+\fB\-\-ultra\fR: unlocks high compression levels 20+ (maximum 22), using a lot more memory\. Note that decompression will also require more memory when using these levels\.
+.
+.IP "\(bu" 4
 \fB\-\-fast[=#]\fR: switch to ultra\-fast compression levels\. If \fB=#\fR is not present, it defaults to \fB1\fR\. The higher the value, the faster the compression speed, at the cost of some compression ratio\. This setting overwrites compression level if one was set previously\. Similarly, if a compression level is set after \fB\-\-fast\fR, it overrides it\.
 .
 .IP "\(bu" 4
-\fB\-\-ultra\fR: unlocks high compression levels 20+ (maximum 22), using a lot more memory\. Note that decompression will also require more memory when using these levels\.
+\fB\-T#\fR, \fB\-\-threads=#\fR: Compress using \fB#\fR working threads (default: 1)\. If \fB#\fR is 0, attempt to detect and use the number of physical CPU cores\. In all cases, the nb of threads is capped to ZSTDMT_NBWORKERS_MAX==200\. This modifier does nothing if \fBzstd\fR is compiled without multithread support\.
+.
+.IP "\(bu" 4
+\fB\-\-single\-thread\fR: Does not spawn a thread for compression, use a single thread for both I/O and compression\. In this mode, compression is serialized with I/O, which is slightly slower\. (This is different from \fB\-T1\fR, which spawns 1 compression thread in parallel of I/O)\. This mode is the only one available when multithread support is disabled\. Single\-thread mode features lower memory usage\. Final compressed result is slightly different from \fB\-T1\fR\.
+.
+.IP "\(bu" 4
+\fB\-\-adapt[=min=#,max=#]\fR : \fBzstd\fR will dynamically adapt compression level to perceived I/O conditions\. Compression level adaptation can be observed live by using command \fB\-v\fR\. Adaptation can be constrained between supplied \fBmin\fR and \fBmax\fR levels\. The feature works when combined with multi\-threading and \fB\-\-long\fR mode\. It does not work with \fB\-\-single\-thread\fR\. It sets window size to 8 MB by default (can be changed manually, see \fBwlog\fR)\. Due to the chaotic nature of dynamic adaptation, compressed result is not reproducible\. \fInote\fR : at the time of this writing, \fB\-\-adapt\fR can remain stuck at low speed when combined with multiple worker threads (>=2)\.
 .
 .IP "\(bu" 4
 \fB\-\-long[=#]\fR: enables long distance matching with \fB#\fR \fBwindowLog\fR, if not \fB#\fR is not present it defaults to \fB27\fR\. This increases the window size (\fBwindowLog\fR) and memory usage for both the compressor and decompressor\. This setting is designed to improve the compression ratio for files with long matches at a large distance\.
@@ -111,43 +120,40 @@ Display information related to a zstd compressed file, such as size, ratio, and
 Note: If \fBwindowLog\fR is set to larger than 27, \fB\-\-long=windowLog\fR or \fB\-\-memory=windowSize\fR needs to be passed to the decompressor\.
 .
 .IP "\(bu" 4
-\fB\-\-patch\-from=FILE\fR: Specify the file to be used as a reference point for zstd\'s diff engine\. This is effectively dictionary compression with some convenient parameter selection, namely that windowSize > srcSize\.
-.
-.IP
-Note: cannot use both this and \-D together Note: \fB\-\-long\fR mode will be automatically activated if chainLog < fileLog (fileLog being the windowLog requried to cover the whole file)\. You can also manually force it\. Node: for all levels, you can use \-\-patch\-from in \-\-single\-thread mode to improve compression ratio at the cost of speed Note: for level 19, you can get increased compression ratio at the cost of speed by specifying \fB\-\-zstd=targetLength=\fR to be something large (i\.e 4096), and by setting a large \fB\-\-zstd=chainLog=\fR
+\fB\-D DICT\fR: use \fBDICT\fR as Dictionary to compress or decompress FILE(s)
 .
 .IP "\(bu" 4
-\fB\-M#\fR, \fB\-\-memory=#\fR: Set a memory usage limit\. By default, Zstandard uses 128 MB for decompression as the maximum amount of memory the decompressor is allowed to use, but you can override this manually if need be in either direction (ie\. you can increase or decrease it)\.
+\fB\-\-patch\-from FILE\fR: Specify the file to be used as a reference point for zstd\'s diff engine\. This is effectively dictionary compression with some convenient parameter selection, namely that windowSize > srcSize\.
 .
 .IP
-This is also used during compression when using with \-\-patch\-from=\. In this case, this parameter overrides that maximum size allowed for a dictionary\. (128 MB)\.
+Note: cannot use both this and \-D together Note: \fB\-\-long\fR mode will be automatically activated if chainLog < fileLog (fileLog being the windowLog required to cover the whole file)\. You can also manually force it\. Node: for all levels, you can use \-\-patch\-from in \-\-single\-thread mode to improve compression ratio at the cost of speed Note: for level 19, you can get increased compression ratio at the cost of speed by specifying \fB\-\-zstd=targetLength=\fR to be something large (i\.e 4096), and by setting a large \fB\-\-zstd=chainLog=\fR
 .
 .IP "\(bu" 4
-\fB\-T#\fR, \fB\-\-threads=#\fR: Compress using \fB#\fR working threads (default: 1)\. If \fB#\fR is 0, attempt to detect and use the number of physical CPU cores\. In all cases, the nb of threads is capped to ZSTDMT_NBTHREADS_MAX==200\. This modifier does nothing if \fBzstd\fR is compiled without multithread support\.
+\fB\-\-rsyncable\fR : \fBzstd\fR will periodically synchronize the compression state to make the compressed file more rsync\-friendly\. There is a negligible impact to compression ratio, and the faster compression levels will see a small compression speed hit\. This feature does not work with \fB\-\-single\-thread\fR\. You probably don\'t want to use it with long range mode, since it will decrease the effectiveness of the synchronization points, but your milage may vary\.
 .
 .IP "\(bu" 4
-\fB\-\-single\-thread\fR: Does not spawn a thread for compression, use a single thread for both I/O and compression\. In this mode, compression is serialized with I/O, which is slightly slower\. (This is different from \fB\-T1\fR, which spawns 1 compression thread in parallel of I/O)\. This mode is the only one available when multithread support is disabled\. Single\-thread mode features lower memory usage\. Final compressed result is slightly different from \fB\-T1\fR\.
+\fB\-C\fR, \fB\-\-[no\-]check\fR: add integrity check computed from uncompressed data (default: enabled)
 .
 .IP "\(bu" 4
-\fB\-\-adapt[=min=#,max=#]\fR : \fBzstd\fR will dynamically adapt compression level to perceived I/O conditions\. Compression level adaptation can be observed live by using command \fB\-v\fR\. Adaptation can be constrained between supplied \fBmin\fR and \fBmax\fR levels\. The feature works when combined with multi\-threading and \fB\-\-long\fR mode\. It does not work with \fB\-\-single\-thread\fR\. It sets window size to 8 MB by default (can be changed manually, see \fBwlog\fR)\. Due to the chaotic nature of dynamic adaptation, compressed result is not reproducible\. \fInote\fR : at the time of this writing, \fB\-\-adapt\fR can remain stuck at low speed when combined with multiple worker threads (>=2)\.
+\fB\-\-[no\-]content\-size\fR: enable / disable whether or not the original size of the file is placed in the header of the compressed file\. The default option is \-\-content\-size (meaning that the original size will be placed in the header)\.
 .
 .IP "\(bu" 4
-\fB\-\-stream\-size=#\fR : Sets the pledged source size of input coming from a stream\. This value must be exact, as it will be included in the produced frame header\. Incorrect stream sizes will cause an error\. This information will be used to better optimize compression parameters, resulting in better and potentially faster compression, especially for smaller source sizes\.
+\fB\-\-no\-dictID\fR: do not store dictionary ID within frame header (dictionary compression)\. The decoder will have to rely on implicit knowledge about which dictionary to use, it won\'t be able to check if it\'s correct\.
 .
 .IP "\(bu" 4
-\fB\-\-size\-hint=#\fR: When handling input from a stream, \fBzstd\fR must guess how large the source size will be when optimizing compression parameters\. If the stream size is relatively small, this guess may be a poor one, resulting in a higher compression ratio than expected\. This feature allows for controlling the guess when needed\. Exact guesses result in better compression ratios\. Overestimates result in slightly degraded compression ratios, while underestimates may result in significant degradation\.
+\fB\-M#\fR, \fB\-\-memory=#\fR: Set a memory usage limit\. By default, Zstandard uses 128 MB for decompression as the maximum amount of memory the decompressor is allowed to use, but you can override this manually if need be in either direction (ie\. you can increase or decrease it)\.
 .
-.IP "\(bu" 4
-\fB\-\-rsyncable\fR : \fBzstd\fR will periodically synchronize the compression state to make the compressed file more rsync\-friendly\. There is a negligible impact to compression ratio, and the faster compression levels will see a small compression speed hit\. This feature does not work with \fB\-\-single\-thread\fR\. You probably don\'t want to use it with long range mode, since it will decrease the effectiveness of the synchronization points, but your milage may vary\.
+.IP
+This is also used during compression when using with \-\-patch\-from=\. In this case, this parameter overrides that maximum size allowed for a dictionary\. (128 MB)\.
 .
 .IP "\(bu" 4
-\fB\-D file\fR: use \fBfile\fR as Dictionary to compress or decompress FILE(s)
+\fB\-\-stream\-size=#\fR : Sets the pledged source size of input coming from a stream\. This value must be exact, as it will be included in the produced frame header\. Incorrect stream sizes will cause an error\. This information will be used to better optimize compression parameters, resulting in better and potentially faster compression, especially for smaller source sizes\.
 .
 .IP "\(bu" 4
-\fB\-\-no\-dictID\fR: do not store dictionary ID within frame header (dictionary compression)\. The decoder will have to rely on implicit knowledge about which dictionary to use, it won\'t be able to check if it\'s correct\.
+\fB\-\-size\-hint=#\fR: When handling input from a stream, \fBzstd\fR must guess how large the source size will be when optimizing compression parameters\. If the stream size is relatively small, this guess may be a poor one, resulting in a higher compression ratio than expected\. This feature allows for controlling the guess when needed\. Exact guesses result in better compression ratios\. Overestimates result in slightly degraded compression ratios, while underestimates may result in significant degradation\.
 .
 .IP "\(bu" 4
-\fB\-o file\fR: save result into \fBfile\fR (only possible with a single \fIINPUT\-FILE\fR)
+\fB\-o FILE\fR: save result into \fBFILE\fR
 .
 .IP "\(bu" 4
 \fB\-f\fR, \fB\-\-force\fR: overwrite output without prompting, and (de)compress symbolic links
@@ -159,10 +165,7 @@ This is also used during compression when using with \-\-patch\-from=\. In this
 \fB\-\-[no\-]sparse\fR: enable / disable sparse FS support, to make files with many zeroes smaller on disk\. Creating sparse files may save disk space and speed up decompression by reducing the amount of disk I/O\. default: enabled when output is into a file, and disabled when output is stdout\. This setting overrides default and can force sparse mode over stdout\.
 .
 .IP "\(bu" 4
-\fB\-\-[no\-]content\-size\fR: enable / disable whether or not the original size of the file is placed in the header of the compressed file\. The default option is \-\-content\-size (meaning that the original size will be placed in the header)\.
-.
-.IP "\(bu" 4
-\fB\-\-rm\fR: remove source file(s) after successful compression or decompression
+\fB\-\-rm\fR: remove source file(s) after successful compression or decompression\. If used in combination with \-o, will trigger a confirmation prompt (which can be silenced with \-f), as this is a destructive operation\.
 .
 .IP "\(bu" 4
 \fB\-k\fR, \fB\-\-keep\fR: keep source file(s) after successful compression or decompression\. This is the default behavior\.
@@ -171,25 +174,28 @@ This is also used during compression when using with \-\-patch\-from=\. In this
 \fB\-r\fR: operate recursively on directories
 .
 .IP "\(bu" 4
-\fB\-\-filelist=FILE\fR read a list of files to process as content from \fBFILE\fR\. Format is compatible with \fBls\fR output, with one file per line\.
+\fB\-\-filelist FILE\fR read a list of files to process as content from \fBFILE\fR\. Format is compatible with \fBls\fR output, with one file per line\.
 .
 .IP "\(bu" 4
-\fB\-\-output\-dir\-flat[=dir]\fR: resulting files are stored into target \fBdir\fR directory, instead of same directory as origin file\. Be aware that this command can introduce name collision issues, if multiple files, from different directories, end up having the same name\. Collision resolution ensures first file with a given name will be present in \fBdir\fR, while in combination with \fB\-f\fR, the last file will be present instead\.
+\fB\-\-output\-dir\-flat DIR\fR: resulting files are stored into target \fBDIR\fR directory, instead of same directory as origin file\. Be aware that this command can introduce name collision issues, if multiple files, from different directories, end up having the same name\. Collision resolution ensures first file with a given name will be present in \fBDIR\fR, while in combination with \fB\-f\fR, the last file will be present instead\.
 .
 .IP "\(bu" 4
-\fB\-\-format=FORMAT\fR: compress and decompress in other formats\. If compiled with support, zstd can compress to or decompress from other compression algorithm formats\. Possibly available options are \fBzstd\fR, \fBgzip\fR, \fBxz\fR, \fBlzma\fR, and \fBlz4\fR\. If no such format is provided, \fBzstd\fR is the default\.
+\fB\-\-output\-dir\-mirror DIR\fR: similar to \fB\-\-output\-dir\-flat\fR, the output files are stored underneath target \fBDIR\fR directory, but this option will replicate input directory hierarchy into output \fBDIR\fR\.
+.
+.IP
+If input directory contains "\.\.", the files in this directory will be ignored\. If input directory is an absolute directory (i\.e\. "/var/tmp/abc"), it will be stored into the "output\-dir/var/tmp/abc"\. If there are multiple input files or directories, name collision resolution will follow the same rules as \fB\-\-output\-dir\-flat\fR\.
 .
 .IP "\(bu" 4
-\fB\-h\fR/\fB\-H\fR, \fB\-\-help\fR: display help/long help and exit
+\fB\-\-format=FORMAT\fR: compress and decompress in other formats\. If compiled with support, zstd can compress to or decompress from other compression algorithm formats\. Possibly available options are \fBzstd\fR, \fBgzip\fR, \fBxz\fR, \fBlzma\fR, and \fBlz4\fR\. If no such format is provided, \fBzstd\fR is the default\.
 .
 .IP "\(bu" 4
-\fB\-V\fR, \fB\-\-version\fR: display version number and exit\. Advanced : \fB\-vV\fR also displays supported formats\. \fB\-vvV\fR also displays POSIX support\.
+\fB\-h\fR/\fB\-H\fR, \fB\-\-help\fR: display help/long help and exit
 .
 .IP "\(bu" 4
-\fB\-v\fR, \fB\-\-verbose\fR: verbose mode
+\fB\-V\fR, \fB\-\-version\fR: display version number and exit\. Advanced : \fB\-vV\fR also displays supported formats\. \fB\-vvV\fR also displays POSIX support\. \fB\-q\fR will only display the version number, suitable for machine reading\.
 .
 .IP "\(bu" 4
-\fB\-\-show\-default\-cparams\fR: Shows the default compresssion parameters that will be used for a particular src file\. If the provided src file is not a regular file (eg\. named pipe), the cli will just output the default paramters\. That is, the parameters that are used when the src size is unknown\.
+\fB\-v\fR, \fB\-\-verbose\fR: verbose mode, display more information
 .
 .IP "\(bu" 4
 \fB\-q\fR, \fB\-\-quiet\fR: suppress warnings, interactivity, and notifications\. specify twice to suppress errors too\.
@@ -198,7 +204,7 @@ This is also used during compression when using with \-\-patch\-from=\. In this
 \fB\-\-no\-progress\fR: do not display the progress bar, but keep all other messages\.
 .
 .IP "\(bu" 4
-\fB\-C\fR, \fB\-\-[no\-]check\fR: add integrity check computed from uncompressed data (default: enabled)
+\fB\-\-show\-default\-cparams\fR: Shows the default compression parameters that will be used for a particular src file\. If the provided src file is not a regular file (eg\. named pipe), the cli will just output the default parameters\. That is, the parameters that are used when the src size is unknown\.
 .
 .IP "\(bu" 4
 \fB\-\-\fR: All arguments after \fB\-\-\fR are treated as files
@@ -206,7 +212,16 @@ This is also used during compression when using with \-\-patch\-from=\. In this
 .IP "" 0
 .
 .SS "Restricted usage of Environment Variables"
-Using environment variables to set parameters has security implications\. Therefore, this avenue is intentionally restricted\. Only \fBZSTD_CLEVEL\fR is supported currently, for setting compression level\. \fBZSTD_CLEVEL\fR can be used to set the level between 1 and 19 (the "normal" range)\. If the value of \fBZSTD_CLEVEL\fR is not a valid integer, it will be ignored with a warning message\. \fBZSTD_CLEVEL\fR just replaces the default compression level (\fB3\fR)\. It can be overridden by corresponding command line arguments\.
+Using environment variables to set parameters has security implications\. Therefore, this avenue is intentionally restricted\. Only \fBZSTD_CLEVEL\fR and \fBZSTD_NBTHREADS\fR are currently supported\. They set the compression level and number of threads to use during compression, respectively\.
+.
+.P
+\fBZSTD_CLEVEL\fR can be used to set the level between 1 and 19 (the "normal" range)\. If the value of \fBZSTD_CLEVEL\fR is not a valid integer, it will be ignored with a warning message\. \fBZSTD_CLEVEL\fR just replaces the default compression level (\fB3\fR)\.
+.
+.P
+\fBZSTD_NBTHREADS\fR can be used to set the number of threads \fBzstd\fR will attempt to use during compression\. If the value of \fBZSTD_NBTHREADS\fR is not a valid unsigned integer, it will be ignored with a warning message\. \'ZSTD_NBTHREADS\fBhas a default value of (\fR1\fB), and is capped at ZSTDMT_NBWORKERS_MAX==200\.\fRzstd` must be compiled with multithread support for this to have any effect\.
+.
+.P
+They can both be overridden by corresponding command line arguments: \fB\-#\fR for compression level and \fB\-T#\fR for number of compression threads\.
 .
 .SH "DICTIONARY BUILDER"
 \fBzstd\fR offers \fIdictionary\fR compression, which greatly improves efficiency on small files and messages\. It\'s possible to train \fBzstd\fR with a set of samples, the result of which is saved into a file called a \fBdictionary\fR\. Then during compression and decompression, reference the same dictionary, using command \fB\-D dictionaryFileName\fR\. Compression of small files similar to the sample set will be greatly improved\.
diff --git a/programs/zstd.1.md b/programs/zstd.1.md
index 550e2e53fe5f..73670daf6dc2 100644
--- a/programs/zstd.1.md
+++ b/programs/zstd.1.md
@@ -102,6 +102,9 @@ the last one takes effect.
 
 * `-#`:
     `#` compression level \[1-19] (default: 3)
+* `--ultra`:
+    unlocks high compression levels 20+ (maximum 22), using a lot more memory.
+    Note that decompression will also require more memory when using these levels.
 * `--fast[=#]`:
     switch to ultra-fast compression levels.
     If `=#` is not present, it defaults to `1`.
@@ -109,9 +112,28 @@ the last one takes effect.
     at the cost of some compression ratio.
     This setting overwrites compression level if one was set previously.
     Similarly, if a compression level is set after `--fast`, it overrides it.
-* `--ultra`:
-    unlocks high compression levels 20+ (maximum 22), using a lot more memory.
-    Note that decompression will also require more memory when using these levels.
+* `-T#`, `--threads=#`:
+    Compress using `#` working threads (default: 1).
+    If `#` is 0, attempt to detect and use the number of physical CPU cores.
+    In all cases, the nb of threads is capped to ZSTDMT_NBWORKERS_MAX==200.
+    This modifier does nothing if `zstd` is compiled without multithread support.
+* `--single-thread`:
+    Does not spawn a thread for compression, use a single thread for both I/O and compression.
+    In this mode, compression is serialized with I/O, which is slightly slower.
+    (This is different from `-T1`, which spawns 1 compression thread in parallel of I/O).
+    This mode is the only one available when multithread support is disabled.
+    Single-thread mode features lower memory usage.
+    Final compressed result is slightly different from `-T1`.
+* `--adapt[=min=#,max=#]` :
+    `zstd` will dynamically adapt compression level to perceived I/O conditions.
+    Compression level adaptation can be observed live by using command `-v`.
+    Adaptation can be constrained between supplied `min` and `max` levels.
+    The feature works when combined with multi-threading and `--long` mode.
+    It does not work with `--single-thread`.
+    It sets window size to 8 MB by default (can be changed manually, see `wlog`).
+    Due to the chaotic nature of dynamic adaptation, compressed result is not reproducible.
+    _note_ : at the time of this writing, `--adapt` can remain stuck at low speed
+    when combined with multiple worker threads (>=2).
 * `--long[=#]`:
     enables long distance matching with `#` `windowLog`, if not `#` is not
     present it defaults to `27`.
@@ -122,20 +144,40 @@ the last one takes effect.
 
     Note: If `windowLog` is set to larger than 27, `--long=windowLog` or
     `--memory=windowSize` needs to be passed to the decompressor.
-* `--patch-from=FILE`:
+* `-D DICT`:
+    use `DICT` as Dictionary to compress or decompress FILE(s)
+* `--patch-from FILE`:
     Specify the file to be used as a reference point for zstd's diff engine.
     This is effectively dictionary compression with some convenient parameter
     selection, namely that windowSize > srcSize.
 
     Note: cannot use both this and -D together
     Note: `--long` mode will be automatically activated if chainLog < fileLog
-        (fileLog being the windowLog requried to cover the whole file). You 
+        (fileLog being the windowLog required to cover the whole file). You
         can also manually force it.
 	Node: for all levels, you can use --patch-from in --single-thread mode
 		to improve compression ratio at the cost of speed
-    Note: for level 19, you can get increased compression ratio at the cost 
-        of speed by specifying `--zstd=targetLength=` to be something large 
+    Note: for level 19, you can get increased compression ratio at the cost
+        of speed by specifying `--zstd=targetLength=` to be something large
         (i.e 4096), and by setting a large `--zstd=chainLog=`
+* `--rsyncable` :
+    `zstd` will periodically synchronize the compression state to make the
+    compressed file more rsync-friendly. There is a negligible impact to
+    compression ratio, and the faster compression levels will see a small
+    compression speed hit.
+    This feature does not work with `--single-thread`. You probably don't want
+    to use it with long range mode, since it will decrease the effectiveness of
+    the synchronization points, but your milage may vary.
+* `-C`, `--[no-]check`:
+    add integrity check computed from uncompressed data (default: enabled)
+* `--[no-]content-size`:
+    enable / disable whether or not the original size of the file is placed in
+    the header of the compressed file. The default option is
+    --content-size (meaning that the original size will be placed in the header).
+* `--no-dictID`:
+    do not store dictionary ID within frame header (dictionary compression).
+    The decoder will have to rely on implicit knowledge about which dictionary to use,
+    it won't be able to check if it's correct.
 * `-M#`, `--memory=#`:
     Set a memory usage limit. By default, Zstandard uses 128 MB for decompression
     as the maximum amount of memory the decompressor is allowed to use, but you can
@@ -144,28 +186,6 @@ the last one takes effect.
 
     This is also used during compression when using with --patch-from=. In this case,
     this parameter overrides that maximum size allowed for a dictionary. (128 MB).
-* `-T#`, `--threads=#`:
-    Compress using `#` working threads (default: 1).
-    If `#` is 0, attempt to detect and use the number of physical CPU cores.
-    In all cases, the nb of threads is capped to ZSTDMT_NBTHREADS_MAX==200.
-    This modifier does nothing if `zstd` is compiled without multithread support.
-* `--single-thread`:
-    Does not spawn a thread for compression, use a single thread for both I/O and compression.
-    In this mode, compression is serialized with I/O, which is slightly slower.
-    (This is different from `-T1`, which spawns 1 compression thread in parallel of I/O).
-    This mode is the only one available when multithread support is disabled.
-    Single-thread mode features lower memory usage.
-    Final compressed result is slightly different from `-T1`.
-* `--adapt[=min=#,max=#]` :
-    `zstd` will dynamically adapt compression level to perceived I/O conditions.
-    Compression level adaptation can be observed live by using command `-v`.
-    Adaptation can be constrained between supplied `min` and `max` levels.
-    The feature works when combined with multi-threading and `--long` mode.
-    It does not work with `--single-thread`.
-    It sets window size to 8 MB by default (can be changed manually, see `wlog`).
-    Due to the chaotic nature of dynamic adaptation, compressed result is not reproducible.
-    _note_ : at the time of this writing, `--adapt` can remain stuck at low speed
-    when combined with multiple worker threads (>=2).
 * `--stream-size=#` :
     Sets the pledged source size of input coming from a stream. This value must be exact, as it
     will be included in the produced frame header. Incorrect stream sizes will cause an error.
@@ -178,22 +198,8 @@ the last one takes effect.
     expected. This feature allows for controlling the guess when needed.
     Exact guesses result in better compression ratios. Overestimates result in slightly
     degraded compression ratios, while underestimates may result in significant degradation.
-* `--rsyncable` :
-    `zstd` will periodically synchronize the compression state to make the
-    compressed file more rsync-friendly. There is a negligible impact to
-    compression ratio, and the faster compression levels will see a small
-    compression speed hit.
-    This feature does not work with `--single-thread`. You probably don't want
-    to use it with long range mode, since it will decrease the effectiveness of
-    the synchronization points, but your milage may vary.
-* `-D file`:
-    use `file` as Dictionary to compress or decompress FILE(s)
-* `--no-dictID`:
-    do not store dictionary ID within frame header (dictionary compression).
-    The decoder will have to rely on implicit knowledge about which dictionary to use,
-    it won't be able to check if it's correct.
-* `-o file`:
-    save result into `file` (only possible with a single _INPUT-FILE_)
+* `-o FILE`:
+    save result into `FILE`
 * `-f`, `--force`:
     overwrite output without prompting, and (de)compress symbolic links
 * `-c`, `--stdout`:
@@ -206,27 +212,34 @@ the last one takes effect.
     default: enabled when output is into a file,
     and disabled when output is stdout.
     This setting overrides default and can force sparse mode over stdout.
-* `--[no-]content-size`:
-    enable / disable whether or not the original size of the file is placed in
-    the header of the compressed file. The default option is 
-    --content-size (meaning that the original size will be placed in the header).
 * `--rm`:
-    remove source file(s) after successful compression or decompression
+    remove source file(s) after successful compression or decompression. If used in combination with
+    -o, will trigger a confirmation prompt (which can be silenced with -f), as this is a destructive operation. 
 * `-k`, `--keep`:
     keep source file(s) after successful compression or decompression.
     This is the default behavior.
 * `-r`:
     operate recursively on directories
-* `--filelist=FILE`
+* `--filelist FILE`
     read a list of files to process as content from `FILE`.
     Format is compatible with `ls` output, with one file per line.
-* `--output-dir-flat[=dir]`:
-    resulting files are stored into target `dir` directory,
+* `--output-dir-flat DIR`:
+    resulting files are stored into target `DIR` directory,
     instead of same directory as origin file.
     Be aware that this command can introduce name collision issues,
     if multiple files, from different directories, end up having the same name.
-    Collision resolution ensures first file with a given name will be present in `dir`,
+    Collision resolution ensures first file with a given name will be present in `DIR`,
     while in combination with `-f`, the last file will be present instead.
+* `--output-dir-mirror DIR`:
+    similar to `--output-dir-flat`,
+    the output files are stored underneath target `DIR` directory,
+    but this option will replicate input directory hierarchy into output `DIR`.
+
+    If input directory contains "..", the files in this directory will be ignored.
+    If input directory is an absolute directory (i.e. "/var/tmp/abc"),
+    it will be stored into the "output-dir/var/tmp/abc".
+    If there are multiple input files or directories,
+    name collision resolution will follow the same rules as `--output-dir-flat`.
 * `--format=FORMAT`:
     compress and decompress in other formats. If compiled with
     support, zstd can compress to or decompress from other compression algorithm
@@ -238,21 +251,19 @@ the last one takes effect.
     display version number and exit.
     Advanced : `-vV` also displays supported formats.
     `-vvV` also displays POSIX support.
+    `-q` will only display the version number, suitable for machine reading.
 * `-v`, `--verbose`:
-    verbose mode
-* `--show-default-cparams`:
-    Shows the default compresssion parameters that will be used for a
-    particular src file. If the provided src file is not a regular file
-    (eg. named pipe), the cli will just output the default paramters.
-    That is, the parameters that are used when the src size is
-    unknown.
+    verbose mode, display more information
 * `-q`, `--quiet`:
     suppress warnings, interactivity, and notifications.
     specify twice to suppress errors too.
 * `--no-progress`:
     do not display the progress bar, but keep all other messages.
-* `-C`, `--[no-]check`:
-    add integrity check computed from uncompressed data (default: enabled)
+* `--show-default-cparams`:
+    Shows the default compression parameters that will be used for a
+    particular src file. If the provided src file is not a regular file
+    (eg. named pipe), the cli will just output the default parameters.
+    That is, the parameters that are used when the src size is unknown.
 * `--`:
     All arguments after `--` are treated as files
 
@@ -260,11 +271,20 @@ the last one takes effect.
 
 Using environment variables to set parameters has security implications.
 Therefore, this avenue is intentionally restricted.
-Only `ZSTD_CLEVEL` is supported currently, for setting compression level.
+Only `ZSTD_CLEVEL` and `ZSTD_NBTHREADS` are currently supported.
+They set the compression level and number of threads to use during compression, respectively.
+
 `ZSTD_CLEVEL` can be used to set the level between 1 and 19 (the "normal" range).
 If the value of `ZSTD_CLEVEL` is not a valid integer, it will be ignored with a warning message.
 `ZSTD_CLEVEL` just replaces the default compression level (`3`).
-It can be overridden by corresponding command line arguments.
+
+`ZSTD_NBTHREADS` can be used to set the number of threads `zstd` will attempt to use during compression.
+If the value of `ZSTD_NBTHREADS` is not a valid unsigned integer, it will be ignored with a warning message.
+'ZSTD_NBTHREADS` has a default value of (`1`), and is capped at ZSTDMT_NBWORKERS_MAX==200. `zstd` must be
+compiled with multithread support for this to have any effect.
+
+They can both be overridden by corresponding command line arguments:
+`-#` for compression level and `-T#` for number of compression threads. 
 
 
 DICTIONARY BUILDER
diff --git a/programs/zstdcli.c b/programs/zstdcli.c
index 70e2b70666bf..9b6f91533462 100644
--- a/programs/zstdcli.c
+++ b/programs/zstdcli.c
@@ -20,7 +20,9 @@
 #  define ZSTDCLI_CLEVEL_MAX 19   /* without using --ultra */
 #endif
 
-
+#ifndef ZSTDCLI_NBTHREADS_DEFAULT
+#  define ZSTDCLI_NBTHREADS_DEFAULT 1
+#endif
 
 /*-************************************
 *  Dependencies
@@ -146,9 +148,26 @@ static void usage_advanced(const char* programName)
 
 #ifdef UTIL_HAS_CREATEFILELIST
     DISPLAYOUT( " -r     : operate recursively on directories \n");
-    DISPLAYOUT( "--filelist=FILE : read list of files to operate upon from FILE \n");
-    DISPLAYOUT( "--output-dir-flat=DIR : all resulting files are stored into DIR \n");
+    DISPLAYOUT( "--filelist FILE : read list of files to operate upon from FILE \n");
+    DISPLAYOUT( "--output-dir-flat DIR : processed files are stored into DIR \n");
+#endif
+
+#ifdef UTIL_HAS_MIRRORFILELIST
+    DISPLAYOUT( "--output-dir-mirror DIR : processed files are stored into DIR respecting original directory structure \n");
+#endif
+
+
+#ifndef ZSTD_NOCOMPRESS
+    DISPLAYOUT( "--[no-]check : during compression, add XXH64 integrity checksum to frame (default: enabled)");
+#ifndef ZSTD_NODECOMPRESS
+    DISPLAYOUT( ". If specified with -d, decompressor will ignore/validate checksums in compressed frame (default: validate).");
 #endif
+#else
+#ifdef ZSTD_NOCOMPRESS
+    DISPLAYOUT( "--[no-]check : during decompression, ignore/validate checksums in compressed frame (default: validate).");
+#endif
+#endif /* ZSTD_NOCOMPRESS */
+    DISPLAYOUT( "\n");
 
     DISPLAYOUT( "--      : All arguments after \"--\" are treated as files \n");
 
@@ -170,7 +189,6 @@ static void usage_advanced(const char* programName)
     DISPLAYOUT( "--size-hint=# optimize compression parameters for streaming input of approximately this size \n");
     DISPLAYOUT( "--target-compressed-block-size=# : generate compressed block of approximately targeted size \n");
     DISPLAYOUT( "--no-dictID : don't write dictID into header (dictionary compression only) \n");
-    DISPLAYOUT( "--[no-]check : add XXH64 integrity checksum to frame (default: enabled) \n");
     DISPLAYOUT( "--[no-]compress-literals : force (un)compressed literals \n");
 
     DISPLAYOUT( "--format=zstd : compress files to the .zst format (default) \n");
@@ -544,6 +562,11 @@ static unsigned parseCompressionParameters(const char* stringPtr, ZSTD_compressi
 
 static void printVersion(void)
 {
+    if (g_displayLevel < DISPLAY_LEVEL_DEFAULT) {
+        DISPLAYOUT("%s\n", ZSTD_VERSION_STRING);
+        return;
+    }
+
     DISPLAYOUT(WELCOME_MESSAGE);
     if (g_displayLevel >= 3) {
     /* format support */
@@ -577,6 +600,7 @@ static void printVersion(void)
 
 /* Environment variables for parameter setting */
 #define ENV_CLEVEL "ZSTD_CLEVEL"
+#define ENV_NBTHREADS "ZSTD_NBTHREADS"    /* takes lower precedence than directly specifying -T# in the CLI */
 
 /* pick up environment variable */
 static int init_cLevel(void) {
@@ -606,8 +630,51 @@ static int init_cLevel(void) {
     return ZSTDCLI_CLEVEL_DEFAULT;
 }
 
-#define ZSTD_NB_STRATEGIES 9
+#ifdef ZSTD_MULTITHREAD
+static unsigned init_nbThreads(void) {
+    const char* const env = getenv(ENV_NBTHREADS);
+    if (env != NULL) {
+        const char* ptr = env;
+        if ((*ptr>='0') && (*ptr<='9')) {
+            unsigned nbThreads;
+            if (readU32FromCharChecked(&ptr, &nbThreads)) {
+                DISPLAYLEVEL(2, "Ignore environment variable setting %s=%s: numeric value too large \n", ENV_NBTHREADS, env);
+                return ZSTDCLI_NBTHREADS_DEFAULT;
+            } else if (*ptr == 0) {
+                return nbThreads;
+            }
+        }
+        DISPLAYLEVEL(2, "Ignore environment variable setting %s=%s: not a valid unsigned value \n", ENV_NBTHREADS, env);
+    }
 
+    return ZSTDCLI_NBTHREADS_DEFAULT;
+}
+#endif
+
+#define NEXT_FIELD(ptr) {         \
+    if (*argument == '=') {       \
+        ptr = ++argument;         \
+        argument += strlen(ptr);  \
+    } else {                      \
+        argNb++;                  \
+        if (argNb >= argCount) {  \
+            DISPLAY("error: missing command argument \n"); \
+            CLEAN_RETURN(1);      \
+        }                         \
+        ptr = argv[argNb];        \
+        assert(ptr != NULL);      \
+        if (ptr[0]=='-') {        \
+            DISPLAY("error: command cannot be separated from its argument by another command \n"); \
+            CLEAN_RETURN(1);      \
+}   }   }
+
+#define NEXT_UINT32(val32) {      \
+    const char* __nb;             \
+    NEXT_FIELD(__nb);             \
+    val32 = readU32FromChar(&__nb); \
+}
+
+#define ZSTD_NB_STRATEGIES 9
 static const char* ZSTD_strategyMap[ZSTD_NB_STRATEGIES + 1] = { "", "ZSTD_fast",
                 "ZSTD_dfast", "ZSTD_greedy", "ZSTD_lazy", "ZSTD_lazy2", "ZSTD_btlazy2",
                 "ZSTD_btopt", "ZSTD_btultra", "ZSTD_btultra2"};
@@ -630,7 +697,7 @@ int main(int const argCount, const char* argv[])
     int argNb,
         followLinks = 0,
         forceStdout = 0,
-        lastCommand = 0,
+        hasStdout = 0,
         ldmFlag = 0,
         main_pause = 0,
         nbWorkers = 0,
@@ -638,12 +705,7 @@ int main(int const argCount, const char* argv[])
         adaptMin = MINCLEVEL,
         adaptMax = MAXCLEVEL,
         rsyncable = 0,
-        nextArgumentIsOutFileName = 0,
-        nextArgumentIsOutDirName = 0,
-        nextArgumentIsMaxDict = 0,
-        nextArgumentIsDictID = 0,
         nextArgumentsAreFiles = 0,
-        nextEntryIsDictionary = 0,
         operationResult = 0,
         separateFiles = 0,
         setRealTimePrio = 0,
@@ -656,6 +718,7 @@ int main(int const argCount, const char* argv[])
     size_t blockSize = 0;
 
     FIO_prefs_t* const prefs = FIO_createPreferences();
+    FIO_ctx_t* const fCtx = FIO_createContext();
     zstd_operation_mode operation = zom_compress;
     ZSTD_compressionParameters compressionParams;
     int cLevel = init_cLevel();
@@ -667,6 +730,7 @@ int main(int const argCount, const char* argv[])
     const char* programName = argv[0];
     const char* outFileName = NULL;
     const char* outDirName = NULL;
+    const char* outMirroredDirName = NULL;
     const char* dictFileName = NULL;
     const char* patchFromDictFileName = NULL;
     const char* suffix = ZSTD_EXTENSION;
@@ -695,7 +759,7 @@ int main(int const argCount, const char* argv[])
     if ((filenames==NULL) || (file_of_names==NULL)) { DISPLAY("zstd: allocation error \n"); exit(1); }
     programName = lastNameFromPath(programName);
 #ifdef ZSTD_MULTITHREAD
-    nbWorkers = 1;
+    nbWorkers = init_nbThreads();
 #endif
 
     /* preset behaviors */
@@ -756,13 +820,10 @@ int main(int const argCount, const char* argv[])
                 if (!strcmp(argument, "--no-sparse")) { FIO_setSparseWrite(prefs, 0); continue; }
                 if (!strcmp(argument, "--test")) { operation=zom_test; continue; }
                 if (!strcmp(argument, "--train")) { operation=zom_train; if (outFileName==NULL) outFileName=g_defaultDictName; continue; }
-                if (!strcmp(argument, "--maxdict")) { nextArgumentIsMaxDict=1; lastCommand=1; continue; }  /* kept available for compatibility with old syntax ; will be removed one day */
-                if (!strcmp(argument, "--dictID")) { nextArgumentIsDictID=1; lastCommand=1; continue; }  /* kept available for compatibility with old syntax ; will be removed one day */
                 if (!strcmp(argument, "--no-dictID")) { FIO_setDictIDFlag(prefs, 0); continue; }
                 if (!strcmp(argument, "--keep")) { FIO_setRemoveSrcFile(prefs, 0); continue; }
                 if (!strcmp(argument, "--rm")) { FIO_setRemoveSrcFile(prefs, 1); continue; }
                 if (!strcmp(argument, "--priority=rt")) { setRealTimePrio = 1; continue; }
-                if (!strcmp(argument, "--output-dir-flat")) {nextArgumentIsOutDirName=1; lastCommand=1; continue; }
                 if (!strcmp(argument, "--show-default-cparams")) { showDefaultCParams = 1; continue; }
                 if (!strcmp(argument, "--content-size")) { contentSize = 1; continue; }
                 if (!strcmp(argument, "--no-content-size")) { contentSize = 0; continue; }
@@ -785,6 +846,7 @@ int main(int const argCount, const char* argv[])
                 if (!strcmp(argument, "--no-compress-literals")) { literalCompressionMode = ZSTD_lcm_uncompressed; continue; }
                 if (!strcmp(argument, "--no-progress")) { FIO_setNoProgress(1); continue; }
                 if (!strcmp(argument, "--exclude-compressed")) { FIO_setExcludeCompressedFile(prefs, 1); continue; }
+
                 /* long commands with arguments */
 #ifndef ZSTD_NODICT
                 if (longCommandWArg(&argument, "--train-cover")) {
@@ -821,19 +883,22 @@ int main(int const argCount, const char* argv[])
                   continue;
                 }
 #endif
-                if (longCommandWArg(&argument, "--threads=")) { nbWorkers = (int)readU32FromChar(&argument); continue; }
-                if (longCommandWArg(&argument, "--memlimit=")) { memLimit = readU32FromChar(&argument); continue; }
-                if (longCommandWArg(&argument, "--memory=")) { memLimit = readU32FromChar(&argument); continue; }
-                if (longCommandWArg(&argument, "--memlimit-decompress=")) { memLimit = readU32FromChar(&argument); continue; }
+                if (longCommandWArg(&argument, "--threads")) { NEXT_UINT32(nbWorkers); continue; }
+                if (longCommandWArg(&argument, "--memlimit")) { NEXT_UINT32(memLimit); continue; }
+                if (longCommandWArg(&argument, "--memory")) { NEXT_UINT32(memLimit); continue; }
+                if (longCommandWArg(&argument, "--memlimit-decompress")) { NEXT_UINT32(memLimit); continue; }
                 if (longCommandWArg(&argument, "--block-size=")) { blockSize = readSizeTFromChar(&argument); continue; }
-                if (longCommandWArg(&argument, "--maxdict=")) { maxDictSize = readU32FromChar(&argument); continue; }
-                if (longCommandWArg(&argument, "--dictID=")) { dictID = readU32FromChar(&argument); continue; }
+                if (longCommandWArg(&argument, "--maxdict")) { NEXT_UINT32(maxDictSize); continue; }
+                if (longCommandWArg(&argument, "--dictID")) { NEXT_UINT32(dictID); continue; }
                 if (longCommandWArg(&argument, "--zstd=")) { if (!parseCompressionParameters(argument, &compressionParams)) { badusage(programName); CLEAN_RETURN(1); } continue; }
                 if (longCommandWArg(&argument, "--stream-size=")) { streamSrcSize = readSizeTFromChar(&argument); continue; }
                 if (longCommandWArg(&argument, "--target-compressed-block-size=")) { targetCBlockSize = readSizeTFromChar(&argument); continue; }
                 if (longCommandWArg(&argument, "--size-hint=")) { srcSizeHint = readSizeTFromChar(&argument); continue; }
-                if (longCommandWArg(&argument, "--output-dir-flat=")) { outDirName = argument; continue; }
-                if (longCommandWArg(&argument, "--patch-from=")) { patchFromDictFileName = argument; continue; }
+                if (longCommandWArg(&argument, "--output-dir-flat")) { NEXT_FIELD(outDirName); continue; }
+#ifdef UTIL_HAS_MIRRORFILELIST
+                if (longCommandWArg(&argument, "--output-dir-mirror")) { NEXT_FIELD(outMirroredDirName); continue; }
+#endif
+                if (longCommandWArg(&argument, "--patch-from")) { NEXT_FIELD(patchFromDictFileName); continue; }
                 if (longCommandWArg(&argument, "--long")) {
                     unsigned ldmWindowLog = 0;
                     ldmFlag = 1;
@@ -877,8 +942,10 @@ int main(int const argCount, const char* argv[])
                 }
 #endif
 
-                if (longCommandWArg(&argument, "--filelist=")) {
-                    UTIL_refFilename(file_of_names, argument);
+                if (longCommandWArg(&argument, "--filelist")) {
+                    const char* listName;
+                    NEXT_FIELD(listName);
+                    UTIL_refFilename(file_of_names, listName);
                     continue;
                 }
 
@@ -887,10 +954,7 @@ int main(int const argCount, const char* argv[])
 
             argument++;
             while (argument[0]!=0) {
-                if (lastCommand) {
-                    DISPLAY("error : command must be followed by argument \n");
-                    CLEAN_RETURN(1);
-                }
+
 #ifndef ZSTD_NOCOMPRESS
                 /* compression Level */
                 if ((*argument>='0') && (*argument<='9')) {
@@ -921,7 +985,7 @@ int main(int const argCount, const char* argv[])
                 case 'c': forceStdout=1; outFileName=stdoutmark; argument++; break;
 
                     /* Use file content as dictionary */
-                case 'D': nextEntryIsDictionary = 1; lastCommand = 1; argument++; break;
+                case 'D': argument++; NEXT_FIELD(dictFileName); break;
 
                     /* Overwrite */
                 case 'f': FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; argument++; break;
@@ -942,7 +1006,7 @@ int main(int const argCount, const char* argv[])
                 case 't': operation=zom_test; argument++; break;
 
                     /* destination file name */
-                case 'o': nextArgumentIsOutFileName=1; lastCommand=1; argument++; break;
+                case 'o': argument++; NEXT_FIELD(outFileName); break;
 
                     /* limit memory */
                 case 'M':
@@ -1013,10 +1077,9 @@ int main(int const argCount, const char* argv[])
 
                     /* Select compressibility of synthetic sample */
                 case 'P':
-                {   argument++;
+                    argument++;
                     compressibility = (double)readU32FromChar(&argument) / 100;
-                }
-                break;
+                    break;
 
                     /* unknown command */
                 default : badusage(programName); CLEAN_RETURN(1);
@@ -1025,51 +1088,10 @@ int main(int const argCount, const char* argv[])
             continue;
         }   /* if (argument[0]=='-') */
 
-        if (nextArgumentIsMaxDict) {  /* kept available for compatibility with old syntax ; will be removed one day */
-            nextArgumentIsMaxDict = 0;
-            lastCommand = 0;
-            maxDictSize = readU32FromChar(&argument);
-            continue;
-        }
-
-        if (nextArgumentIsDictID) {  /* kept available for compatibility with old syntax ; will be removed one day */
-            nextArgumentIsDictID = 0;
-            lastCommand = 0;
-            dictID = readU32FromChar(&argument);
-            continue;
-        }
-
-        if (nextEntryIsDictionary) {
-            nextEntryIsDictionary = 0;
-            lastCommand = 0;
-            dictFileName = argument;
-            continue;
-        }
-
-        if (nextArgumentIsOutFileName) {
-            nextArgumentIsOutFileName = 0;
-            lastCommand = 0;
-            outFileName = argument;
-            if (!strcmp(outFileName, "-")) outFileName = stdoutmark;
-            continue;
-        }
-
-        if (nextArgumentIsOutDirName) {
-            nextArgumentIsOutDirName = 0;
-            lastCommand = 0;
-            outDirName = argument;
-            continue;
-        }
-
         /* none of the above : add filename to list */
         UTIL_refFilename(filenames, argument);
     }
 
-    if (lastCommand) { /* forgotten argument */
-        DISPLAY("error : command must be followed by argument \n");
-        CLEAN_RETURN(1);
-    }
-
     /* Welcome message (if verbose) */
     DISPLAYLEVEL(3, WELCOME_MESSAGE);
 
@@ -1222,7 +1244,7 @@ int main(int const argCount, const char* argv[])
 
     /* Check if input/output defined as console; trigger an error in this case */
     if (!strcmp(filenames->fileNames[0], stdinmark) && IS_CONSOLE(stdin) ) {
-        badusage(programName);
+        DISPLAYLEVEL(1, "stdin is a console, aborting\n");
         CLEAN_RETURN(1);
     }
     if ( outFileName && !strcmp(outFileName, stdoutmark)
@@ -1230,7 +1252,7 @@ int main(int const argCount, const char* argv[])
       && !strcmp(filenames->fileNames[0], stdinmark)
       && !forceStdout
       && operation!=zom_decompress ) {
-        badusage(programName);
+        DISPLAYLEVEL(1, "stdout is a console, aborting\n");
         CLEAN_RETURN(1);
     }
 
@@ -1259,12 +1281,16 @@ int main(int const argCount, const char* argv[])
         DISPLAY("error : can't use --patch-from=# on multiple files \n");
         CLEAN_RETURN(1);
     }
+    
+    /* No status message in pipe mode (stdin - stdout) */	
+    hasStdout = outFileName && !strcmp(outFileName,stdoutmark);
 
-    /* No status message in pipe mode (stdin - stdout) or multi-files mode */
-    if (!strcmp(filenames->fileNames[0], stdinmark) && outFileName && !strcmp(outFileName,stdoutmark) && (g_displayLevel==2)) g_displayLevel=1;
-    if ((filenames->tableSize > 1) & (g_displayLevel==2)) g_displayLevel=1;
+    if (hasStdout && (g_displayLevel==2)) g_displayLevel=1;
 
     /* IO Stream/File */
+    FIO_setHasStdoutOutput(fCtx, hasStdout);
+    FIO_setNbFilesTotal(fCtx, (int)filenames->tableSize); 
+    FIO_determineHasStdinInput(fCtx, filenames);
     FIO_setNotificationLevel(g_displayLevel);
     FIO_setPatchFromMode(prefs, patchFromDictFileName != NULL);
     if (memLimit == 0) {
@@ -1323,9 +1349,9 @@ int main(int const argCount, const char* argv[])
         }
 
         if ((filenames->tableSize==1) && outFileName)
-          operationResult = FIO_compressFilename(prefs, outFileName, filenames->fileNames[0], dictFileName, cLevel, compressionParams);
+            operationResult = FIO_compressFilename(fCtx, prefs, outFileName, filenames->fileNames[0], dictFileName, cLevel, compressionParams);
         else
-          operationResult = FIO_compressMultipleFilenames(prefs, filenames->fileNames, (unsigned)filenames->tableSize, outDirName, outFileName, suffix, dictFileName, cLevel, compressionParams);
+            operationResult = FIO_compressMultipleFilenames(fCtx, prefs, filenames->fileNames, outMirroredDirName, outDirName, outFileName, suffix, dictFileName, cLevel, compressionParams);
 #else
         (void)contentSize; (void)suffix; (void)adapt; (void)rsyncable; (void)ultra; (void)cLevel; (void)ldmFlag; (void)literalCompressionMode; (void)targetCBlockSize; (void)streamSrcSize; (void)srcSizeHint; (void)ZSTD_strategyMap; /* not used when ZSTD_NOCOMPRESS set */
         DISPLAY("Compression not supported \n");
@@ -1333,9 +1359,9 @@ int main(int const argCount, const char* argv[])
     } else {  /* decompression or test */
 #ifndef ZSTD_NODECOMPRESS
         if (filenames->tableSize == 1 && outFileName) {
-            operationResult = FIO_decompressFilename(prefs, outFileName, filenames->fileNames[0], dictFileName);
+            operationResult = FIO_decompressFilename(fCtx, prefs, outFileName, filenames->fileNames[0], dictFileName);
         } else {
-            operationResult = FIO_decompressMultipleFilenames(prefs, filenames->fileNames, (unsigned)filenames->tableSize, outDirName, outFileName, dictFileName);
+            operationResult = FIO_decompressMultipleFilenames(fCtx, prefs, filenames->fileNames, outMirroredDirName, outDirName, outFileName, dictFileName);
         }
 #else
         DISPLAY("Decompression not supported \n");
@@ -1344,6 +1370,7 @@ int main(int const argCount, const char* argv[])
 
 _end:
     FIO_freePreferences(prefs);
+    FIO_freeContext(fCtx);
     if (main_pause) waitEnter();
     UTIL_freeFileNamesTable(filenames);
     UTIL_freeFileNamesTable(file_of_names);
diff --git a/programs/zstdgrep.1 b/programs/zstdgrep.1
index fe1d29bbb062..4d143b598735 100644
--- a/programs/zstdgrep.1
+++ b/programs/zstdgrep.1
@@ -1,5 +1,5 @@
 .
-.TH "ZSTDGREP" "1" "May 2020" "zstd 1.4.5" "User Commands"
+.TH "ZSTDGREP" "1" "December 2020" "zstd 1.4.8" "User Commands"
 .
 .SH "NAME"
 \fBzstdgrep\fR \- print lines matching a pattern in zstandard\-compressed files
diff --git a/programs/zstdless.1 b/programs/zstdless.1
index d54c6400cee2..43f235453ee6 100644
--- a/programs/zstdless.1
+++ b/programs/zstdless.1
@@ -1,5 +1,5 @@
 .
-.TH "ZSTDLESS" "1" "May 2020" "zstd 1.4.5" "User Commands"
+.TH "ZSTDLESS" "1" "December 2020" "zstd 1.4.8" "User Commands"
 .
 .SH "NAME"
 \fBzstdless\fR \- view zstandard\-compressed files
diff --git a/zlibWrapper/Makefile b/zlibWrapper/Makefile
index feed5b84a34f..d74c41bc4b84 100644
--- a/zlibWrapper/Makefile
+++ b/zlibWrapper/Makefile
@@ -18,6 +18,8 @@ EXAMPLE_PATH = examples
 PROGRAMS_PATH = ../programs
 TEST_FILE = ../doc/zstd_compression_format.md
 
+VPATH = $(PROGRAMS_PATH)
+
 CPPFLAGS += -DXXH_NAMESPACE=ZSTD_ -I$(ZLIB_PATH) -I$(PROGRAMS_PATH)       \
             -I$(ZSTDLIBDIR) -I$(ZSTDLIBDIR)/common -I$(ZLIBWRAPPER_PATH)
 STDFLAGS  = -std=c89 -pedantic -Wno-long-long -Wno-variadic-macros -Wc++-compat \
@@ -95,7 +97,7 @@ fitblk: $(EXAMPLE_PATH)/fitblk.o zstd_zlibwrapper.o $(ZSTDLIBRARY)
 fitblk_zstd: $(EXAMPLE_PATH)/fitblk.o zstdTurnedOn_zlibwrapper.o $(ZSTDLIBRARY)
 	$(CC) $(CFLAGS) $(CPPFLAGS) $(LDFLAGS) $^ $(ZLIB_LIBRARY) -o $@
 
-zwrapbench: $(EXAMPLE_PATH)/zwrapbench.o zstd_zlibwrapper.o $(PROGRAMS_PATH)/util.o $(PROGRAMS_PATH)/timefn.o $(PROGRAMS_PATH)/datagen.o $(ZSTDLIBRARY)
+zwrapbench: $(EXAMPLE_PATH)/zwrapbench.o zstd_zlibwrapper.o util.o timefn.o datagen.o $(ZSTDLIBRARY)
 	$(CC) $(CFLAGS) $(CPPFLAGS) $(LDFLAGS) $^ $(ZLIB_LIBRARY) -o $@
 
 
diff --git a/zlibWrapper/zstd_zlibwrapper.c b/zlibWrapper/zstd_zlibwrapper.c
index 4d2c7be5b44c..0ae5012a184c 100644
--- a/zlibWrapper/zstd_zlibwrapper.c
+++ b/zlibWrapper/zstd_zlibwrapper.c
@@ -19,13 +19,13 @@
 #include <stdlib.h>
 #include <stdio.h>                 /* vsprintf */
 #include <stdarg.h>                /* va_list, for z_gzprintf */
+#include <string.h>
 #define NO_DUMMY_DECL
 #define ZLIB_CONST
 #include <zlib.h>                  /* without #define Z_PREFIX */
 #include "zstd_zlibwrapper.h"
-#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_isFrame, ZSTD_MAGICNUMBER */
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_isFrame, ZSTD_MAGICNUMBER, ZSTD_customMem */
 #include "zstd.h"
-#include "zstd_internal.h"         /* ZSTD_malloc, ZSTD_free */
 
 
 /* ===   Constants   === */
@@ -42,6 +42,45 @@
 #define FINISH_WITH_GZ_ERR(msg) { (void)msg; return Z_STREAM_ERROR; }
 #define FINISH_WITH_NULL_ERR(msg) { (void)msg; return NULL; }
 
+/* ===   Utility   === */
+
+#define MIN(x,y) ((x) < (y) ? (x) : (y))
+
+static unsigned ZWRAP_isLittleEndian(void)
+{
+    const union { unsigned u; char c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+#ifndef __has_builtin
+# define __has_builtin(x) 0
+#endif
+
+static unsigned ZWRAP_swap32(unsigned in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_ulong(in);
+#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
+  || (defined(__clang__) && __has_builtin(__builtin_bswap32))
+    return __builtin_bswap32(in);
+#else
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+#endif
+}
+
+static unsigned ZWRAP_readLE32(const void* ptr)
+{
+    unsigned value;
+    memcpy(&value, ptr, sizeof(value));
+    if (ZWRAP_isLittleEndian())
+        return value;
+    else
+        return ZWRAP_swap32(value);
+}
+
 
 /* ===   Wrapper   === */
 static int g_ZWRAP_useZSTDcompression = ZWRAP_USE_ZSTD; /* 0 = don't use ZSTD */
@@ -64,8 +103,6 @@ const char * zstdVersion(void) { return ZSTD_VERSION_STRING; }
 
 ZEXTERN const char * ZEXPORT z_zlibVersion OF((void)) { return zlibVersion();  }
 
-
-
 static void* ZWRAP_allocFunction(void* opaque, size_t size)
 {
     z_streamp strm = (z_streamp) opaque;
@@ -81,6 +118,35 @@ static void ZWRAP_freeFunction(void* opaque, void* address)
    /* if (address) LOG_WRAPPERC("ZWRAP free %p \n", address); */
 }
 
+static void* ZWRAP_customMalloc(size_t size, ZSTD_customMem customMem)
+{
+    if (customMem.customAlloc)
+        return customMem.customAlloc(customMem.opaque, size);
+    return malloc(size);
+}
+
+static void* ZWRAP_customCalloc(size_t size, ZSTD_customMem customMem)
+{
+    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);
+}
+
+static void ZWRAP_customFree(void* ptr, ZSTD_customMem customMem)
+{
+    if (ptr!=NULL) {
+        if (customMem.customFree)
+            customMem.customFree(customMem.opaque, ptr);
+        else
+            free(ptr);
+    }
+}
+
 
 
 /* ===   Compression   === */
@@ -107,7 +173,7 @@ static size_t ZWRAP_freeCCtx(ZWRAP_CCtx* zwc)
 {
     if (zwc==NULL) return 0;   /* support free on NULL */
     ZSTD_freeCStream(zwc->zbc);
-    ZSTD_free(zwc, zwc->customMem);
+    ZWRAP_customFree(zwc, zwc->customMem);
     return 0;
 }
 
@@ -115,20 +181,19 @@ static size_t ZWRAP_freeCCtx(ZWRAP_CCtx* zwc)
 static ZWRAP_CCtx* ZWRAP_createCCtx(z_streamp strm)
 {
     ZWRAP_CCtx* zwc;
+    ZSTD_customMem customMem = { NULL, NULL, NULL };
 
     if (strm->zalloc && strm->zfree) {
-        zwc = (ZWRAP_CCtx*)strm->zalloc(strm->opaque, 1, sizeof(ZWRAP_CCtx));
-        if (zwc==NULL) return NULL;
-        memset(zwc, 0, sizeof(ZWRAP_CCtx));
-        memcpy(&zwc->allocFunc, strm, sizeof(z_stream));
-        {   ZSTD_customMem ZWRAP_customMem = { ZWRAP_allocFunction, ZWRAP_freeFunction, NULL };
-            ZWRAP_customMem.opaque = &zwc->allocFunc;
-            zwc->customMem = ZWRAP_customMem;
-        }
-    } else {
-        zwc = (ZWRAP_CCtx*)calloc(1, sizeof(*zwc));
-        if (zwc==NULL) return NULL;
+        customMem.customAlloc = ZWRAP_allocFunction;
+        customMem.customFree = ZWRAP_freeFunction;
     }
+    customMem.opaque = strm;
+
+    zwc = (ZWRAP_CCtx*)ZWRAP_customCalloc(sizeof(ZWRAP_CCtx), customMem);
+    if (zwc == NULL) return NULL;
+    zwc->allocFunc = *strm;
+    customMem.opaque = &zwc->allocFunc;
+    zwc->customMem = customMem;
 
     return zwc;
 }
@@ -181,6 +246,10 @@ int ZWRAP_setPledgedSrcSize(z_streamp strm, unsigned long long pledgedSrcSize)
     return Z_OK;
 }
 
+static struct internal_state* convert_into_sis(void* ptr)
+{
+    return (struct internal_state*) ptr;
+}
 
 ZEXTERN int ZEXPORT z_deflateInit_ OF((z_streamp strm, int level,
                                      const char *version, int stream_size))
@@ -201,7 +270,7 @@ ZEXTERN int ZEXPORT z_deflateInit_ OF((z_streamp strm, int level,
     zwc->streamEnd = 0;
     zwc->totalInBytes = 0;
     zwc->compressionLevel = level;
-    strm->state = (struct internal_state*) zwc; /* use state which in not used by user */
+    strm->state = convert_into_sis(zwc); /* use state which in not used by user */
     strm->total_in = 0;
     strm->total_out = 0;
     strm->adler = 0;
@@ -457,21 +526,19 @@ static void ZWRAP_initDCtx(ZWRAP_DCtx* zwd)
 static ZWRAP_DCtx* ZWRAP_createDCtx(z_streamp strm)
 {
     ZWRAP_DCtx* zwd;
-    MEM_STATIC_ASSERT(sizeof(zwd->headerBuf) >= ZSTD_HEADERSIZE);   /* check static buffer size condition */
+    ZSTD_customMem customMem = { NULL, NULL, NULL };
 
     if (strm->zalloc && strm->zfree) {
-        zwd = (ZWRAP_DCtx*)strm->zalloc(strm->opaque, 1, sizeof(ZWRAP_DCtx));
-        if (zwd==NULL) return NULL;
-        memset(zwd, 0, sizeof(ZWRAP_DCtx));
-        zwd->allocFunc = *strm;  /* just to copy zalloc, zfree & opaque */
-        {   ZSTD_customMem ZWRAP_customMem = { ZWRAP_allocFunction, ZWRAP_freeFunction, NULL };
-            ZWRAP_customMem.opaque = &zwd->allocFunc;
-            zwd->customMem = ZWRAP_customMem;
-        }
-    } else {
-        zwd = (ZWRAP_DCtx*)calloc(1, sizeof(*zwd));
-        if (zwd==NULL) return NULL;
+        customMem.customAlloc = ZWRAP_allocFunction;
+        customMem.customFree = ZWRAP_freeFunction;
     }
+    customMem.opaque = strm;
+
+    zwd = (ZWRAP_DCtx*)ZWRAP_customCalloc(sizeof(ZWRAP_DCtx), customMem);
+    if (zwd == NULL) return NULL;
+    zwd->allocFunc = *strm;
+    customMem.opaque = &zwd->allocFunc;
+    zwd->customMem = customMem;
 
     ZWRAP_initDCtx(zwd);
     return zwd;
@@ -481,8 +548,8 @@ static size_t ZWRAP_freeDCtx(ZWRAP_DCtx* zwd)
 {
     if (zwd==NULL) return 0;   /* support free on null */
     ZSTD_freeDStream(zwd->zbd);
-    ZSTD_free(zwd->version, zwd->customMem);
-    ZSTD_free(zwd, zwd->customMem);
+    ZWRAP_customFree(zwd->version, zwd->customMem);
+    ZWRAP_customFree(zwd, zwd->customMem);
     return 0;
 }
 
@@ -524,13 +591,13 @@ ZEXTERN int ZEXPORT z_inflateInit_ OF((z_streamp strm,
         LOG_WRAPPERD("- inflateInit\n");
         if (zwd == NULL) return ZWRAPD_finishWithError(zwd, strm, 0);
 
-        zwd->version = (char*)ZSTD_malloc(strlen(version)+1, zwd->customMem);
+        zwd->version = (char*)ZWRAP_customMalloc(strlen(version)+1, zwd->customMem);
         if (zwd->version == NULL) return ZWRAPD_finishWithError(zwd, strm, 0);
         strcpy(zwd->version, version);
 
         zwd->stream_size = stream_size;
         zwd->totalInBytes = 0;
-        strm->state = (struct internal_state*) zwd;
+        strm->state = convert_into_sis(zwd);
         strm->total_in = 0;
         strm->total_out = 0;
         strm->reserved = ZWRAP_UNKNOWN_STREAM;
@@ -670,7 +737,7 @@ ZEXTERN int ZEXPORT z_inflate OF((z_streamp strm, int flush))
 
     if (zwd->totalInBytes < ZLIB_HEADERSIZE) {
         if (zwd->totalInBytes == 0 && strm->avail_in >= ZLIB_HEADERSIZE) {
-            if (MEM_readLE32(strm->next_in) != ZSTD_MAGICNUMBER) {
+            if (ZWRAP_readLE32(strm->next_in) != ZSTD_MAGICNUMBER) {
                 {   int const initErr = (zwd->windowBits) ?
                                 inflateInit2_(strm, zwd->windowBits, zwd->version, zwd->stream_size) :
                                 inflateInit_(strm, zwd->version, zwd->stream_size);
@@ -698,7 +765,7 @@ ZEXTERN int ZEXPORT z_inflate OF((z_streamp strm, int flush))
             strm->avail_in -= srcSize;
             if (zwd->totalInBytes < ZLIB_HEADERSIZE) return Z_OK;
 
-            if (MEM_readLE32(zwd->headerBuf) != ZSTD_MAGICNUMBER) {
+            if (ZWRAP_readLE32(zwd->headerBuf) != ZSTD_MAGICNUMBER) {
                 z_stream strm2;
                 strm2.next_in = strm->next_in;
                 strm2.avail_in = strm->avail_in;