Merge pull request #4491 from facebook/cmake_root

[cmake] propose a root wrapper

Merge pull request #4517 from Cyan4973/asyncio_revisit

Remove asyncio from the compression path

syncio interface only enabled when compression is enabled

changed name to syncIO for clarity

minor: use init/destroy pair naming convention

removed asyncio completely for compression path

this does not provide speed benefits,
since most of the leverage happens internally within the library,
and can even become detrimental in certain scenario, due to complex and wasteful memory management.
At a minimum, it makes the logic simpler, easier to debug, at essentially the same performance.

disable asyncio during compression

it's not helpful for performance, and can in some times be detrimental.
the synchronous mode is also much easier to follow and debug.

Merge pull request #4515 from cclauss/patch-1

Fix undefined names in automated_benchmarking.py

Fix undefined names in automated_benchmarking.py

Fix variable name typos in automated_benchmarking.py

% `ruff check`
```
Error: tests/automated_benchmarking.py:237:21: F821 Undefined name `baseline_label`
Error: tests/automated_benchmarking.py:250:21: F821 Undefined name `baseline_label`
Error: tests/automated_benchmarking.py:318:55: F821 Undefined name `frequenc`
Error: Process completed with exit code 1.
```
% [`ruff rule F821`](https://docs.astral.sh/ruff/rules/undefined-name)

Merge pull request #4516 from Cyan4973/mingw_calloc_order

minor: fix mingw warnings: calloc argument order

minor: fix mingw warnings: calloc argument order

Merge pull request #4509 from arpadpanyik-arm/revert_branch_opt

AArch64: Revert previous branch optimization

AArch64: Remove 32-bit code from ZSTD_decodeSequence

Remove the 32-bit code paths from the AArch64 only sections of
ZSTD_decodeSequence.

AArch64: Revert previous branch optimization

Revert a branch optimization that was based on an incorrect
assumption in the AArch64 part of ZSTD_decodeSequence. In extreme
cases the existing implementation could lead to data corruption.

Insert an UNLIKELY hint to guide the compilers toward generating more
efficient machine code.

Merge pull request #4504 from rzikm/asm-conditional

Make ASM language conditional at CMake configure time

Enable ASM language during CMake configure time only when needed

Merge pull request #4502 from zijianli1234/dev

- Modify the GCC version used for CI testing of the RISCV architecture

- Modify the GCC version used for CI testing of the RISCV architecture
- Fix a bug in the ZSTD_row_getRVVMask function
- Improve some performance for ZSTD_copy16()

[cmake] add minimal build test

for CMakeLists.txt at root

[cmake] root wrapper

allow the existence of a `CMakeLists.txt` file at root,
for easier integration with other projects expecting this file at root.

Existing integration point, within `build/cmake/`, still works as expected.

Merge pull request #4500 from facebook/legacy01_init

fix minor initialization warnings

fix minor initialization warnings

Merge pull request #4498 from facebook/contentSize_doc

update documentation of ZSTD_getFrameContentSize()

update documentation of ZSTD_getFrameContentSize()

hopefully answering #4495

Merge pull request #4497 from facebook/legacy_fix

fix minor warning in legacy decoders

fix minor warning in legacy decoders

for mingw + clang CI test

Merge pull request #4490 from facebook/doc_profiling_vs

minor: add profiling notes for Visual Studio

added profiling notes for Visual Studio

Merge pull request #4489 from facebook/doc_install

Added documentation details for Makefile installation and pkg-config.

Added documentation details for Makefile installation and pkg-config.

Merge pull request #4486 from rlefko/fix-pthread-init-memleak

Fix memory leak in pthread init functions on failure

Merge pull request #4487 from neiljohari/adhoc/dictionary-file-counting

make DiB_fileStats skip invalid files (fileSize <= 0) to prevent negative totals and bogus allocation

Merge pull request #4481 from w1m024/support-rvv-getmask

add RVV optimization for ZSTD_row_getMatchMask

Remove debug logging

Fix bug

Add debug logging and simple repro

🔧 Fix memory leak in pthread init functions on failure

When pthread_mutex_init() or pthread_cond_init() fails in the debug
implementation (DEBUGLEVEL >= 1), the previously allocated memory was
not freed, causing a memory leak.

This fix ensures that allocated memory is properly freed when pthread
initialization functions fail, preventing resource leaks in error
conditions.

The issue affects:
- ZSTD_pthread_mutex_init() at lib/common/threading.c:146
- ZSTD_pthread_cond_init() at lib/common/threading.c:167

This is particularly important for long-running applications or
scenarios with resource constraints where pthread initialization
might fail due to system limits.

Refactor ZSTD_row_getMatchMask for RVV optimization

Performance (vs. SWAR)
- 16-byte data: 5.87x speedup
- 32-byte data: 9.63x speedup
- 64-byte data: 17.98x speedup

Co-authored-by: gong-flying <gongxiaofei24@iscas.ac.cn>

add RVV optimization for ZSTD_row_getMatchMask

Co-authored-by: gong-flying <gongxiaofei24@iscas.ac.cn>

Merge pull request #4480 from facebook/dependabot/github_actions/github/codeql-action-3.30.1

Bump github/codeql-action from 3.29.4 to 3.30.1

Merge pull request #4479 from facebook/dependabot/github_actions/msys2/setup-msys2-2.29.0

Bump msys2/setup-msys2 from 2.28.0 to 2.29.0

Bump github/codeql-action from 3.29.4 to 3.30.1

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.29.4 to 3.30.1.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/4e828ff8d448a8a6e532957b1811f387a63867e8...f1f6e5f6af878fb37288ce1c627459e94dbf7d01)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-version: 3.30.1
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

Bump msys2/setup-msys2 from 2.28.0 to 2.29.0

Bumps [msys2/setup-msys2](https://github.com/msys2/setup-msys2) from 2.28.0 to 2.29.0.
- [Release notes](https://github.com/msys2/setup-msys2/releases)
- [Changelog](https://github.com/msys2/setup-msys2/blob/main/CHANGELOG.md)
- [Commits](https://github.com/msys2/setup-msys2/compare/40677d36a502eb2cf0fb808cc9dec31bf6152638...fb197b72ce45fb24f17bf3f807a388985654d1f2)

---
updated-dependencies:
- dependency-name: msys2/setup-msys2
  dependency-version: 2.29.0
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

Merge pull request #4472 from bgilbert/override_dependency

meson: Call `meson.override_dependency()` if Meson is new enough

Merge pull request #4475 from Cyan4973/default_nbThreads

Default nb threads

fixed minor unused variable warning

in certain compilation modes

benchmark uses 1 thread by default

only display nbThread Msg in nbThreads > 1

specify nb of threads used during benchmarking

used to require `-v` (verbose) modifier

fixed -T# documentation in zstd -H

provide the local value for default nbThreads
which is dynamic and depends on local nb of cores.

Merge pull request #4474 from jlokier/threads-doc-fix

Update manual about the default value of `-T#`/`--threads=#`

Update manual about the default value of `-T#`/`--threads=#`

The section about `ZSTD_NBTHREADS` already explains the default number of
threads, since it changed from 1 (commit 17beeb5).  But the option description
for `-T#`/`--threads=#` incorrectly said the default was still 1.

I noticed this when I found compression slower with `-T1` than without it.

meson: Call meson.override_dependency() if Meson is new enough

This tells Meson that we intend libzstd_dep to be used by a parent project
if the parent looks for a dependency named "libzstd".  Without this, the
mapping from "libzstd" to our variable libzstd_dep must be encoded in the
Meson wrap file or in the parent's meson.build.

Merge pull request #4469 from facebook/dependabot/github_actions/actions/setup-java-5.0.0

Bump actions/setup-java from 4.7.1 to 5.0.0

Bump actions/setup-java from 4.7.1 to 5.0.0

Bumps [actions/setup-java](https://github.com/actions/setup-java) from 4.7.1 to 5.0.0.
- [Release notes](https://github.com/actions/setup-java/releases)
- [Commits](https://github.com/actions/setup-java/compare/c5195efecf7bdfc987ee8bae7a71cb8b11521c00...dded0888837ed1f317902acf8a20df0ad188d165)

---
updated-dependencies:
- dependency-name: actions/setup-java
  dependency-version: 5.0.0
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>

Merge pull request #4440 from arpadpanyik-arm/convert_seq_sve2

AArch64: Add SVE2 path for convertSequences_noRepcodes

AArch64: Add SVE2 path for convertSequences_noRepcodes

Add an 8-way vector length agnostic (VLA) SVE2 code path for
convertSequences_noRepcodes. It works with any SVE vector length.

Relative performance to GCC-13 using: `./fullbench -b18 -l5 enwik5`

               Neon      SVE2
Neoverse-V2   before     after    uplift
GCC-13:      100.000%  103.209%   1.032x
GCC-14:      100.309%  134.872%   1.344x
GCC-15:      100.355%  134.827%   1.343x
Clang-18:    123.614%  128.565%   1.040x
Clang-19:    123.587%  132.984%   1.076x
Clang-20:    123.629%  133.023%   1.075x

               Neon      SVE2
Cortex-A720   before     after    uplift
GCC-13:      100.000%  116.032%   1.160x
GCC-14:       99.700%  116.648%   1.169x
GCC-15:      100.354%  117.047%   1.166x
Clang-18:    100.447%  116.762%   1.162x
Clang-19:    100.454%  116.627%   1.160x
Clang-20:    100.452%  116.649%   1.161x

Merge pull request #4463 from brad0/gnu_source_qsort

Check for build environment instead of just _GNU_SOURCE

Merge pull request #4465 from thiru-mcw/arm64_support

WOA_support:: Add CI setup for packaging Windows on ARM artifacts

CI: Enable MSVC ARM64 job using Github WOA runner

- Reintroduce the MSVC ARM64 build configuration with "Visual Studio 17 2022"
- Update runner to `windows-11-arm` (GitHub-hosted Windows on ARM)

CI: Add CI setup for packaging Win-ARM64 artifacts

CI: Update build_package.bat for CMake builds

Merge pull request #4464 from facebook/cli_traces_div0

fixed a potential division by 0 in the cli trace unit

fixed a potential division by 0 in the cli trace unit

Check for build environment instead of just _GNU_SOURCE

Fixes the build on OpenBSD and NetBSD. It is too easy for _GNU_SOURCE
to be defined even on non-Linux systems. Found via py-zstandard with
the embedded copy of zstandard and Python defines _GNU_SOURCE.

Also simplify the Linux checking, there is no need to check the rest
of the symbol names.

Merge pull request #4419 from AZero13/patch-1

Check for job before releasing resources

Merge pull request #4462 from facebook/dependabot/github_actions/actions/checkout-5

Bump actions/checkout from 4 to 5

Bump actions/checkout from 4 to 5

Bumps [actions/checkout](https://github.com/actions/checkout) from 4 to 5.
- [Release notes](https://github.com/actions/checkout/releases)
- [Commits](https://github.com/actions/checkout/compare/v4...v5)

---
updated-dependencies:
- dependency-name: actions/checkout
  dependency-version: '5'
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>

Merge pull request #4459 from Margen67/premake

Remove need for trailing forward slash in dir

Remove need for trailing forward slash in dir

Merge pull request #4448 from Cyan4973/install_oses

regroup list of OSes for install inside common variable

regroup list of OSes for install inside common variable

within lib/install_oses.mk.

fixes #4445

Merge pull request #4450 from facebook/dependabot/github_actions/github/codeql-action-3.29.4

Bump github/codeql-action from 3.28.9 to 3.29.4

Bump github/codeql-action from 3.28.9 to 3.29.4

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.28.9 to 3.29.4.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/9e8d0789d4a0fa9ceb6b1738f7e269594bdd67f0...4e828ff8d448a8a6e532957b1811f387a63867e8)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-version: 3.29.4
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

Merge pull request #4447 from facebook/android-cmake

added android cmake build

Merge pull request #4413 from arpadpanyik-arm/huf_decode2x

AArch64: Enhance struct access in Huffman decode 2X

Merge pull request #4443 from facebook/opt_simplify_4442

simplify sequence resolution in zstd_opt

added android cmake build

is expecte to fail, due to #4444

simplify sequence resolution in zstd_opt

initially hinted by @pitaj in #4442

Merge pull request #4394 from AZero13/zstd

Remove redundant setting of allJobsCompleted to 1

Merge pull request #4418 from arpadpanyik-arm/decode_seq_opt

AArch64: Improve ZSTD_decodeSequence performance

Merge pull request #4435 from zijianli1234/dev

add riscv  ci

Merge pull request #4429 from arpadpanyik-arm/convertSequences_Neon

Improve speed of ZSTD_compressSequencesAndLiterals using Neon

Merge pull request #4436 from facebook/dependabot/github_actions/cygwin/cygwin-install-action-6

Bump cygwin/cygwin-install-action from 5 to 6

Bump cygwin/cygwin-install-action from 5 to 6

Bumps [cygwin/cygwin-install-action](https://github.com/cygwin/cygwin-install-action) from 5 to 6.
- [Release notes](https://github.com/cygwin/cygwin-install-action/releases)
- [Commits](https://github.com/cygwin/cygwin-install-action/compare/f61179d72284ceddc397ed07ddb444d82bf9e559...f2009323764960f80959895c7bc3bb30210afe4d)

---
updated-dependencies:
- dependency-name: cygwin/cygwin-install-action
  dependency-version: '6'
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>

Merge pull request #4433 from facebook/vs2025

removed VS2019 runners

add -DMEM_FORCE_MEMORY_ACCESS=0 in CI RVV test

removed VS2019 runners

replaced by one vs2025 runner,
which is badly named since it still running MSVC 2022,
but it's a good test that  shows that the matrix is able to handle multiple MSVC versions.

AArch64: Enable optimized QEMU CI builds

Add missing `-O3` flag to the compilation of AArch64 SVE2 builds
executed by QEMU. This can decrease the CI job runtime considerably.

AArch64: Add Neon path for convertSequences_noRepcodes

Add a 4-way Neon implementation for the convertSequences_noRepcodes
function. Remove 'static' keywords from all of its implementations to
be able to add unit tests.

Relative performance to Clang-18 using: `./fullbench -b18 -l5 enwik5`

Neoverse-V2   before     after
Clang-18:    100.000%  311.703%
Clang-19:    100.191%  311.714%
Clang-20:    100.181%  311.723%
GCC-13:      107.520%  252.309%
GCC-14:      107.652%  253.158%
GCC-15:      107.674%  253.168%

Cortex-A720   before     after
Clang-18:    100.000%  204.512%
Clang-19:    102.825%  204.600%
Clang-20:    102.807%  204.558%
GCC-13:      110.668%  203.594%
GCC-14:      110.684%  203.978%
GCC-15:      102.864%  204.299%

Co-authored by, Thomas Daubney <Thomas.Daubney@arm.com>

Improve ZSTD_get1BlockSummary

Add a faster scalar implementation of ZSTD_get1BlockSummary which
removes the data dependency of the accumulators in the hot loop to
leverage the superscalar potential of recent out-of-order CPUs.
The new algorithm leverages SWAR (SIMD Within A Register) methodology
to exploit the capabilities of 64-bit architectures. It achieves this
by packing two 32-bit data elements into a single 64-bit register,
enabling parallel operations on these subcomponents while ensuring
that the 32-bit boundaries prevent overflow, thereby optimizing
computational efficiency.

Corresponding unit tests are included.

Relative performance to GCC-13 using: `./fullbench -b19 -l5 enwik5`

Neoverse-V2   before     after
GCC-13:      100.000%  290.527%
GCC-14:      100.000%  291.714%
GCC-15:       99.914%  291.495%
Clang-18:    148.072%  264.524%
Clang-19:    148.075%  264.512%
Clang-20:    148.062%  264.490%

Cortex-A720   before     after
GCC-13:      100.000%  235.261%
GCC-14:      101.064%  234.903%
GCC-15:      112.977%  218.547%
Clang-18:    127.135%  180.359%
Clang-19:    127.149%  180.297%
Clang-20:    127.154%  180.260%

Co-authored by, Thomas Daubney <Thomas.Daubney@arm.com>

add compiler version check.

fix dereferencing type-punned pointer error

add riscv rvv ci

Merge pull request #4414 from arpadpanyik-arm/copy8

AArch64: Use better block COPY8

Check for job before releasing

ZSTDMT_freeCCtx calls ZSTDMT_releaseAllJobResources, but ZSTDMT_releaseAllJobResources may be called when ZSTDMT_freeCCtx is called when initialization fails, resulting in a NULL pointer dereference.

Remove redundant setting of allJobsCompleted to 1

This will do it automatically.

AArch64: Improve ZSTD_decodeSequence performance

LLVM's alias-analysis sometimes fails to see that a static-array member
of a struct cannot alias other members. This patch:

- Reduces array accesses via struct indirection to aid load/store alias
  analysis under Clang.
- Converts dynamic array indexing into conditional-move arithmetic,
  eliminating branches and extra loads/stores on out-of-order CPUs.
- Reloads the bitstream only when match-length bits are consumed
  (assuming each reload only needs to happen once per match-length
  read), improving branch-prediction rates.
- Removes the UNLIKELY() hint, which recent compilers already handle
  well without cost.

Decompression uplifts on a Neoverse V2 system, using Zstd-1.5.8
compiled with "-O3 -march=armv8.2-a+sve2":

                 Clang-19  Clang-20   Clang-*    GCC-14    GCC-15
 1#silesia.tar:  +11.556%  +16.203%   +0.240%   +2.216%   +7.891%
 2#silesia.tar:  +15.493%  +21.140%   -0.041%   +2.850%   +9.926%
 3#silesia.tar:  +16.887%  +22.570%   -0.183%   +3.056%  +10.660%
 4#silesia.tar:  +17.785%  +23.315%   -0.262%   +3.343%  +11.187%
 5#silesia.tar:  +18.125%  +24.175%   -0.466%   +3.350%  +11.228%
 6#silesia.tar:  +17.607%  +23.339%   -0.591%   +3.175%  +10.851%
 7#silesia.tar:  +17.463%  +22.837%   -0.486%   +3.292%  +10.868%

* Requires Clang-21 support from LLVM commit hash
  `a53003fe23cb6c871e72d70ff2d3a075a7490da2`
   (Clang-21 hasn’t been released as of this writing)

Co-authored by:
 David Sherwood, David.Sherwood@arm.com
 Ola Liljedahl, Ola.Liljedahl@arm.com

AArch64: Enhance struct access in Huffman decode 2X

In the multi-stream multi-symbol Huffman decoder GCC generates
suboptimal code - emitting more loads for HUF_DEltX2 struct member
accesses. Forcing it to use 32-bit loads and bit arithmetic to extract
the necessary parts (UBFX) improves the overall decode speed.

Also avoid integer type conversions in the symbol decodes, which
leads to better instruction selection in table lookup accesses.

On AArch64 the decoder no longer runs into register-pressure limits,
so we can simplify the hot path and improve throughput

Decompression uplifts on a Neoverse V2 system, using Zstd-1.5.8
compiled with "-O3 -march=armv8.2-a+sve2":

                 Clang-20   Clang-*    GCC-13    GCC-14    GCC-15
 1#silesia.tar:   +0.820%   +1.365%   +2.480%   +1.348%   +0.987%
 2#silesia.tar:   +0.426%   +0.784%   +1.218%   +0.665%   +0.554%
 3#silesia.tar:   +0.112%   +0.389%   +0.508%   +0.188%   +0.261%

* Requires Clang-21 support from LLVM commit hash
  `a53003fe23cb6c871e72d70ff2d3a075a7490da2`
  (Clang-21 hasn’t been released as of this writing)

Merge pull request #4417 from facebook/dependabot/github_actions/msys2/setup-msys2-2.28.0

Bump msys2/setup-msys2 from 2.27.0 to 2.28.0