diff --git a/.github/actions/prepare_vm/action.yaml b/.github/actions/prepare_vm/action.yaml index 2bfa7c95425..f2eadaa9f62 100644 --- a/.github/actions/prepare_vm/action.yaml +++ b/.github/actions/prepare_vm/action.yaml @@ -10,7 +10,7 @@ runs: sudo apt-get -y update sudo apt-get -y install git cmake ninja-build libidn11-dev ragel yasm protobuf-compiler \ protobuf-compiler-grpc libprotobuf-dev libgrpc++-dev libgrpc-dev libgrpc++1 libgrpc10 \ - rapidjson-dev zlib1g-dev libxxhash-dev libzstd-dev + rapidjson-dev zlib1g-dev libxxhash-dev libzstd-dev libsnappy-dev wget https://apt.llvm.org/llvm.sh chmod u+x llvm.sh sudo ./llvm.sh 16 diff --git a/README.md b/README.md index b5a56e6b4f7..9e013d9d876 100644 --- a/README.md +++ b/README.md @@ -26,6 +26,7 @@ If you ok with this warning, then... - xxhash - zlib - zstd +- snappy 1.1.8+ ## Runtime requirements @@ -38,7 +39,7 @@ If you ok with this warning, then... sudo apt-get -y update sudo apt-get -y install git cmake ninja-build libidn11-dev ragel yasm protobuf-compiler \ protobuf-compiler-grpc libprotobuf-dev libgrpc++-dev libgrpc-dev libgrpc++1 libgrpc10 \ - rapidjson-dev zlib1g-dev libxxhash-dev libzstd-dev + rapidjson-dev zlib1g-dev libxxhash-dev libzstd-dev libsnappy-dev wget https://apt.llvm.org/llvm.sh chmod u+x llvm.sh diff --git a/cmake/external_libs.cmake b/cmake/external_libs.cmake index 83b87840362..18f1a522b3f 100644 --- a/cmake/external_libs.cmake +++ b/cmake/external_libs.cmake @@ -9,6 +9,7 @@ find_package(RapidJSON REQUIRED) find_package(ZLIB REQUIRED) find_package(xxHash REQUIRED) find_package(ZSTD REQUIRED) +find_package(Snappy 1.1.8 REQUIRED CONFIG) # rapidjson add_library(ydb-sdk-rapidjson INTERFACE) diff --git a/contrib/libs/CMakeLists.txt b/contrib/libs/CMakeLists.txt index 0165d06b539..29fdf09ae29 100644 --- a/contrib/libs/CMakeLists.txt +++ b/contrib/libs/CMakeLists.txt @@ -8,6 +8,5 @@ add_subdirectory(libc_compat) add_subdirectory(lz4) add_subdirectory(lzmasdk) add_subdirectory(nayuki_md5) -add_subdirectory(snappy) #add_subdirectory(zstd) add_subdirectory(zstd06) diff --git a/contrib/libs/snappy/AUTHORS b/contrib/libs/snappy/AUTHORS deleted file mode 100644 index 4858b377c74..00000000000 --- a/contrib/libs/snappy/AUTHORS +++ /dev/null @@ -1 +0,0 @@ -opensource@google.com diff --git a/contrib/libs/snappy/CMakeLists.darwin-arm64.txt b/contrib/libs/snappy/CMakeLists.darwin-arm64.txt deleted file mode 100644 index 3e55347eabc..00000000000 --- a/contrib/libs/snappy/CMakeLists.darwin-arm64.txt +++ /dev/null @@ -1,26 +0,0 @@ - -# This file was generated by the build system used internally in the Yandex monorepo. -# Only simple modifications are allowed (adding source-files to targets, adding simple properties -# like target_include_directories). These modifications will be ported to original -# ya.make files by maintainers. Any complex modifications which can't be ported back to the -# original buildsystem will not be accepted. - - - -add_library(contrib-libs-snappy) -target_compile_options(contrib-libs-snappy PRIVATE - -DHAVE_CONFIG_H - $,,-Wno-everything> -) -target_include_directories(contrib-libs-snappy PUBLIC - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/include -) -target_link_libraries(contrib-libs-snappy PUBLIC - yutil -) -target_sources(contrib-libs-snappy PRIVATE - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-c.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-sinksource.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-stubs-internal.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy.cc -) diff --git a/contrib/libs/snappy/CMakeLists.darwin-x86_64.txt b/contrib/libs/snappy/CMakeLists.darwin-x86_64.txt deleted file mode 100644 index 3e55347eabc..00000000000 --- a/contrib/libs/snappy/CMakeLists.darwin-x86_64.txt +++ /dev/null @@ -1,26 +0,0 @@ - -# This file was generated by the build system used internally in the Yandex monorepo. -# Only simple modifications are allowed (adding source-files to targets, adding simple properties -# like target_include_directories). These modifications will be ported to original -# ya.make files by maintainers. Any complex modifications which can't be ported back to the -# original buildsystem will not be accepted. - - - -add_library(contrib-libs-snappy) -target_compile_options(contrib-libs-snappy PRIVATE - -DHAVE_CONFIG_H - $,,-Wno-everything> -) -target_include_directories(contrib-libs-snappy PUBLIC - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/include -) -target_link_libraries(contrib-libs-snappy PUBLIC - yutil -) -target_sources(contrib-libs-snappy PRIVATE - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-c.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-sinksource.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-stubs-internal.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy.cc -) diff --git a/contrib/libs/snappy/CMakeLists.linux-aarch64.txt b/contrib/libs/snappy/CMakeLists.linux-aarch64.txt deleted file mode 100644 index 3e55347eabc..00000000000 --- a/contrib/libs/snappy/CMakeLists.linux-aarch64.txt +++ /dev/null @@ -1,26 +0,0 @@ - -# This file was generated by the build system used internally in the Yandex monorepo. -# Only simple modifications are allowed (adding source-files to targets, adding simple properties -# like target_include_directories). These modifications will be ported to original -# ya.make files by maintainers. Any complex modifications which can't be ported back to the -# original buildsystem will not be accepted. - - - -add_library(contrib-libs-snappy) -target_compile_options(contrib-libs-snappy PRIVATE - -DHAVE_CONFIG_H - $,,-Wno-everything> -) -target_include_directories(contrib-libs-snappy PUBLIC - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/include -) -target_link_libraries(contrib-libs-snappy PUBLIC - yutil -) -target_sources(contrib-libs-snappy PRIVATE - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-c.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-sinksource.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-stubs-internal.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy.cc -) diff --git a/contrib/libs/snappy/CMakeLists.linux-x86_64.txt b/contrib/libs/snappy/CMakeLists.linux-x86_64.txt deleted file mode 100644 index 3e55347eabc..00000000000 --- a/contrib/libs/snappy/CMakeLists.linux-x86_64.txt +++ /dev/null @@ -1,26 +0,0 @@ - -# This file was generated by the build system used internally in the Yandex monorepo. -# Only simple modifications are allowed (adding source-files to targets, adding simple properties -# like target_include_directories). These modifications will be ported to original -# ya.make files by maintainers. Any complex modifications which can't be ported back to the -# original buildsystem will not be accepted. - - - -add_library(contrib-libs-snappy) -target_compile_options(contrib-libs-snappy PRIVATE - -DHAVE_CONFIG_H - $,,-Wno-everything> -) -target_include_directories(contrib-libs-snappy PUBLIC - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/include -) -target_link_libraries(contrib-libs-snappy PUBLIC - yutil -) -target_sources(contrib-libs-snappy PRIVATE - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-c.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-sinksource.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-stubs-internal.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy.cc -) diff --git a/contrib/libs/snappy/CMakeLists.txt b/contrib/libs/snappy/CMakeLists.txt deleted file mode 100644 index d863ebd1806..00000000000 --- a/contrib/libs/snappy/CMakeLists.txt +++ /dev/null @@ -1,19 +0,0 @@ - -# This file was generated by the build system used internally in the Yandex monorepo. -# Only simple modifications are allowed (adding source-files to targets, adding simple properties -# like target_include_directories). These modifications will be ported to original -# ya.make files by maintainers. Any complex modifications which can't be ported back to the -# original buildsystem will not be accepted. - - -if (CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" AND NOT HAVE_CUDA) - include(CMakeLists.linux-x86_64.txt) -elseif (CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND NOT HAVE_CUDA) - include(CMakeLists.linux-aarch64.txt) -elseif (CMAKE_SYSTEM_NAME STREQUAL "Darwin" AND CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64") - include(CMakeLists.darwin-x86_64.txt) -elseif (CMAKE_SYSTEM_NAME STREQUAL "Darwin" AND CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64") - include(CMakeLists.darwin-arm64.txt) -elseif (WIN32 AND CMAKE_SYSTEM_PROCESSOR STREQUAL "AMD64" AND NOT HAVE_CUDA) - include(CMakeLists.windows-x86_64.txt) -endif() diff --git a/contrib/libs/snappy/CMakeLists.windows-x86_64.txt b/contrib/libs/snappy/CMakeLists.windows-x86_64.txt deleted file mode 100644 index 3e55347eabc..00000000000 --- a/contrib/libs/snappy/CMakeLists.windows-x86_64.txt +++ /dev/null @@ -1,26 +0,0 @@ - -# This file was generated by the build system used internally in the Yandex monorepo. -# Only simple modifications are allowed (adding source-files to targets, adding simple properties -# like target_include_directories). These modifications will be ported to original -# ya.make files by maintainers. Any complex modifications which can't be ported back to the -# original buildsystem will not be accepted. - - - -add_library(contrib-libs-snappy) -target_compile_options(contrib-libs-snappy PRIVATE - -DHAVE_CONFIG_H - $,,-Wno-everything> -) -target_include_directories(contrib-libs-snappy PUBLIC - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/include -) -target_link_libraries(contrib-libs-snappy PUBLIC - yutil -) -target_sources(contrib-libs-snappy PRIVATE - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-c.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-sinksource.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy-stubs-internal.cc - ${CMAKE_SOURCE_DIR}/contrib/libs/snappy/snappy.cc -) diff --git a/contrib/libs/snappy/CONTRIBUTING.md b/contrib/libs/snappy/CONTRIBUTING.md deleted file mode 100644 index 66a60d5c864..00000000000 --- a/contrib/libs/snappy/CONTRIBUTING.md +++ /dev/null @@ -1,31 +0,0 @@ -# How to Contribute - -We'd love to accept your patches and contributions to this project. There are -just a few small guidelines you need to follow. - -## Contributor License Agreement - -Contributions to this project must be accompanied by a Contributor License -Agreement. You (or your employer) retain the copyright to your contribution; -this simply gives us permission to use and redistribute your contributions as -part of the project. Head over to to see -your current agreements on file or to sign a new one. - -You generally only need to submit a CLA once, so if you've already submitted one -(even if it was for a different project), you probably don't need to do it -again. - -## Code Reviews - -All submissions, including submissions by project members, require review. We -use GitHub pull requests for this purpose. Consult -[GitHub Help](https://help.github.com/articles/about-pull-requests/) for more -information on using pull requests. - -See [the README](README.md#contributing-to-the-snappy-project) for areas -where we are likely to accept external contributions. - -## Community Guidelines - -This project follows [Google's Open Source Community -Guidelines](https://opensource.google/conduct/). diff --git a/contrib/libs/snappy/COPYING b/contrib/libs/snappy/COPYING deleted file mode 100644 index bd0e5971dbf..00000000000 --- a/contrib/libs/snappy/COPYING +++ /dev/null @@ -1,54 +0,0 @@ -Copyright 2011, Google Inc. -All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are -met: - - * Redistributions of source code must retain the above copyright -notice, this list of conditions and the following disclaimer. - * Redistributions in binary form must reproduce the above -copyright notice, this list of conditions and the following disclaimer -in the documentation and/or other materials provided with the -distribution. - * Neither the name of Google Inc. nor the names of its -contributors may be used to endorse or promote products derived from -this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -=== - -Some of the benchmark data in testdata/ is licensed differently: - - - fireworks.jpeg is Copyright 2013 Steinar H. Gunderson, and - is licensed under the Creative Commons Attribution 3.0 license - (CC-BY-3.0). See https://creativecommons.org/licenses/by/3.0/ - for more information. - - - kppkn.gtb is taken from the Gaviota chess tablebase set, and - is licensed under the MIT License. See - https://sites.google.com/site/gaviotachessengine/Home/endgame-tablebases-1 - for more information. - - - paper-100k.pdf is an excerpt (bytes 92160 to 194560) from the paper - “Combinatorial Modeling of Chromatin Features Quantitatively Predicts DNA - Replication Timing in _Drosophila_” by Federico Comoglio and Renato Paro, - which is licensed under the CC-BY license. See - http://www.ploscompbiol.org/static/license for more ifnormation. - - - alice29.txt, asyoulik.txt, plrabn12.txt and lcet10.txt are from Project - Gutenberg. The first three have expired copyrights and are in the public - domain; the latter does not have expired copyright, but is still in the - public domain according to the license information - (http://www.gutenberg.org/ebooks/53). diff --git a/contrib/libs/snappy/NEWS b/contrib/libs/snappy/NEWS deleted file mode 100644 index 792a578001d..00000000000 --- a/contrib/libs/snappy/NEWS +++ /dev/null @@ -1,200 +0,0 @@ -Snappy v1.1.10, Mar 8th 2023: - - * Performance improvements - - * Compilation fixes for various environments - -Snappy v1.1.9, May 4th 2021: - - * Performance improvements. - - * Google Test and Google Benchmark are now bundled in third_party/. - -Snappy v1.1.8, January 15th 2020: - - * Small performance improvements. - - * Removed snappy::string alias for std::string. - - * Improved CMake configuration. - -Snappy v1.1.7, August 24th 2017: - - * Improved CMake build support for 64-bit Linux distributions. - - * MSVC builds now use MSVC-specific intrinsics that map to clzll. - - * ARM64 (AArch64) builds use the code paths optimized for 64-bit processors. - -Snappy v1.1.6, July 12th 2017: - -This is a re-release of v1.1.5 with proper SONAME / SOVERSION values. - -Snappy v1.1.5, June 28th 2017: - -This release has broken SONAME / SOVERSION values. Users of snappy as a shared -library should avoid 1.1.5 and use 1.1.6 instead. SONAME / SOVERSION errors will -manifest as the dynamic library loader complaining that it cannot find snappy's -shared library file (libsnappy.so / libsnappy.dylib), or that the library it -found does not have the required version. 1.1.6 has the same code as 1.1.5, but -carries build configuration fixes for the issues above. - - * Add CMake build support. The autoconf build support is now deprecated, and - will be removed in the next release. - - * Add AppVeyor configuration, for Windows CI coverage. - - * Small performance improvement on little-endian PowerPC. - - * Small performance improvement on LLVM with position-independent executables. - - * Fix a few issues with various build environments. - -Snappy v1.1.4, January 25th 2017: - - * Fix a 1% performance regression when snappy is used in PIE executables. - - * Improve compression performance by 5%. - - * Improve decompression performance by 20%. - -Snappy v1.1.3, July 6th 2015: - -This is the first release to be done from GitHub, which means that -some minor things like the ChangeLog format has changed (git log -format instead of svn log). - - * Add support for Uncompress() from a Source to a Sink. - - * Various minor changes to improve MSVC support; in particular, - the unit tests now compile and run under MSVC. - - -Snappy v1.1.2, February 28th 2014: - -This is a maintenance release with no changes to the actual library -source code. - - * Stop distributing benchmark data files that have unclear - or unsuitable licensing. - - * Add support for padding chunks in the framing format. - - -Snappy v1.1.1, October 15th 2013: - - * Add support for uncompressing to iovecs (scatter I/O). - The bulk of this patch was contributed by Mohit Aron. - - * Speed up decompression by ~2%; much more so (~13-20%) on - a few benchmarks on given compilers and CPUs. - - * Fix a few issues with MSVC compilation. - - * Support truncated test data in the benchmark. - - -Snappy v1.1.0, January 18th 2013: - - * Snappy now uses 64 kB block size instead of 32 kB. On average, - this means it compresses about 3% denser (more so for some - inputs), at the same or better speeds. - - * libsnappy no longer depends on iostream. - - * Some small performance improvements in compression on x86 - (0.5–1%). - - * Various portability fixes for ARM-based platforms, for MSVC, - and for GNU/Hurd. - - -Snappy v1.0.5, February 24th 2012: - - * More speed improvements. Exactly how big will depend on - the architecture: - - - 3–10% faster decompression for the base case (x86-64). - - - ARMv7 and higher can now use unaligned accesses, - and will see about 30% faster decompression and - 20–40% faster compression. - - - 32-bit platforms (ARM and 32-bit x86) will see 2–5% - faster compression. - - These are all cumulative (e.g., ARM gets all three speedups). - - * Fixed an issue where the unit test would crash on system - with less than 256 MB address space available, - e.g. some embedded platforms. - - * Added a framing format description, for use over e.g. HTTP, - or for a command-line compressor. We do not have any - implementations of this at the current point, but there seems - to be enough of a general interest in the topic. - Also make the format description slightly clearer. - - * Remove some compile-time warnings in -Wall - (mostly signed/unsigned comparisons), for easier embedding - into projects that use -Wall -Werror. - - -Snappy v1.0.4, September 15th 2011: - - * Speeded up the decompressor somewhat; typically about 2–8% - for Core i7, in 64-bit mode (comparable for Opteron). - Somewhat more for some tests, almost no gain for others. - - * Make Snappy compile on certain platforms it didn't before - (Solaris with SunPro C++, HP-UX, AIX). - - * Correct some minor errors in the format description. - - -Snappy v1.0.3, June 2nd 2011: - - * Speeded up the decompressor somewhat; about 3-6% for Core 2, - 6-13% for Core i7, and 5-12% for Opteron (all in 64-bit mode). - - * Added compressed format documentation. This text is new, - but an earlier version from Zeev Tarantov was used as reference. - - * Only link snappy_unittest against -lz and other autodetected - libraries, not libsnappy.so (which doesn't need any such dependency). - - * Fixed some display issues in the microbenchmarks, one of which would - frequently make the test crash on GNU/Hurd. - - -Snappy v1.0.2, April 29th 2011: - - * Relicense to a BSD-type license. - - * Added C bindings, contributed by Martin Gieseking. - - * More Win32 fixes, in particular for MSVC. - - * Replace geo.protodata with a newer version. - - * Fix timing inaccuracies in the unit test when comparing Snappy - to other algorithms. - - -Snappy v1.0.1, March 25th 2011: - -This is a maintenance release, mostly containing minor fixes. -There is no new functionality. The most important fixes include: - - * The COPYING file and all licensing headers now correctly state that - Snappy is licensed under the Apache 2.0 license. - - * snappy_unittest should now compile natively under Windows, - as well as on embedded systems with no mmap(). - - * Various autotools nits have been fixed. - - -Snappy v1.0, March 17th 2011: - - * Initial version. diff --git a/contrib/libs/snappy/README.md b/contrib/libs/snappy/README.md deleted file mode 100644 index 398be7d58a6..00000000000 --- a/contrib/libs/snappy/README.md +++ /dev/null @@ -1,165 +0,0 @@ -Snappy, a fast compressor/decompressor. - -[![Build Status](https://github.com/google/snappy/actions/workflows/build.yml/badge.svg)](https://github.com/google/snappy/actions/workflows/build.yml) - -Introduction -============ - -Snappy is a compression/decompression library. It does not aim for maximum -compression, or compatibility with any other compression library; instead, -it aims for very high speeds and reasonable compression. For instance, -compared to the fastest mode of zlib, Snappy is an order of magnitude faster -for most inputs, but the resulting compressed files are anywhere from 20% to -100% bigger. (For more information, see "Performance", below.) - -Snappy has the following properties: - - * Fast: Compression speeds at 250 MB/sec and beyond, with no assembler code. - See "Performance" below. - * Stable: Over the last few years, Snappy has compressed and decompressed - petabytes of data in Google's production environment. The Snappy bitstream - format is stable and will not change between versions. - * Robust: The Snappy decompressor is designed not to crash in the face of - corrupted or malicious input. - * Free and open source software: Snappy is licensed under a BSD-type license. - For more information, see the included COPYING file. - -Snappy has previously been called "Zippy" in some Google presentations -and the like. - - -Performance -=========== - -Snappy is intended to be fast. On a single core of a Core i7 processor -in 64-bit mode, it compresses at about 250 MB/sec or more and decompresses at -about 500 MB/sec or more. (These numbers are for the slowest inputs in our -benchmark suite; others are much faster.) In our tests, Snappy usually -is faster than algorithms in the same class (e.g. LZO, LZF, QuickLZ, -etc.) while achieving comparable compression ratios. - -Typical compression ratios (based on the benchmark suite) are about 1.5-1.7x -for plain text, about 2-4x for HTML, and of course 1.0x for JPEGs, PNGs and -other already-compressed data. Similar numbers for zlib in its fastest mode -are 2.6-2.8x, 3-7x and 1.0x, respectively. More sophisticated algorithms are -capable of achieving yet higher compression rates, although usually at the -expense of speed. Of course, compression ratio will vary significantly with -the input. - -Although Snappy should be fairly portable, it is primarily optimized -for 64-bit x86-compatible processors, and may run slower in other environments. -In particular: - - - Snappy uses 64-bit operations in several places to process more data at - once than would otherwise be possible. - - Snappy assumes unaligned 32 and 64-bit loads and stores are cheap. - On some platforms, these must be emulated with single-byte loads - and stores, which is much slower. - - Snappy assumes little-endian throughout, and needs to byte-swap data in - several places if running on a big-endian platform. - -Experience has shown that even heavily tuned code can be improved. -Performance optimizations, whether for 64-bit x86 or other platforms, -are of course most welcome; see "Contact", below. - - -Building -======== - -You need the CMake version specified in [CMakeLists.txt](./CMakeLists.txt) -or later to build: - -```bash -git submodule update --init -mkdir build -cd build && cmake ../ && make -``` - -Usage -===== - -Note that Snappy, both the implementation and the main interface, -is written in C++. However, several third-party bindings to other languages -are available; see the [home page](docs/README.md) for more information. -Also, if you want to use Snappy from C code, you can use the included C -bindings in snappy-c.h. - -To use Snappy from your own C++ program, include the file "snappy.h" from -your calling file, and link against the compiled library. - -There are many ways to call Snappy, but the simplest possible is - -```c++ -snappy::Compress(input.data(), input.size(), &output); -``` - -and similarly - -```c++ -snappy::Uncompress(input.data(), input.size(), &output); -``` - -where "input" and "output" are both instances of std::string. - -There are other interfaces that are more flexible in various ways, including -support for custom (non-array) input sources. See the header file for more -information. - - -Tests and benchmarks -==================== - -When you compile Snappy, the following binaries are compiled in addition to the -library itself. You do not need them to use the compressor from your own -library, but they are useful for Snappy development. - -* `snappy_benchmark` contains microbenchmarks used to tune compression and - decompression performance. -* `snappy_unittests` contains unit tests, verifying correctness on your machine - in various scenarios. -* `snappy_test_tool` can benchmark Snappy against a few other compression - libraries (zlib, LZO, LZF, and QuickLZ), if they were detected at configure - time. To benchmark using a given file, give the compression algorithm you want - to test Snappy against (e.g. --zlib) and then a list of one or more file names - on the command line. - -If you want to change or optimize Snappy, please run the tests and benchmarks to -verify you have not broken anything. - -The testdata/ directory contains the files used by the microbenchmarks, which -should provide a reasonably balanced starting point for benchmarking. (Note that -baddata[1-3].snappy are not intended as benchmarks; they are used to verify -correctness in the presence of corrupted data in the unit test.) - -Contributing to the Snappy Project -================================== - -In addition to the aims listed at the top of the [README](README.md) Snappy -explicitly supports the following: - -1. C++11 -2. Clang (gcc and MSVC are best-effort). -3. Low level optimizations (e.g. assembly or equivalent intrinsics) for: - 1. [x86](https://en.wikipedia.org/wiki/X86) - 2. [x86-64](https://en.wikipedia.org/wiki/X86-64) - 3. ARMv7 (32-bit) - 4. ARMv8 (AArch64) -4. Supports only the Snappy compression scheme as described in - [format_description.txt](format_description.txt). -5. CMake for building - -Changes adding features or dependencies outside of the core area of focus listed -above might not be accepted. If in doubt post a message to the -[Snappy discussion mailing list](https://groups.google.com/g/snappy-compression). - -We are unlikely to accept contributions to the build configuration files, such -as `CMakeLists.txt`. We are focused on maintaining a build configuration that -allows us to test that the project works in a few supported configurations -inside Google. We are not currently interested in supporting other requirements, -such as different operating systems, compilers, or build systems. - -Contact -======= - -Snappy is distributed through GitHub. For the latest version and other -information, see https://github.com/google/snappy. diff --git a/contrib/libs/snappy/config-linux.h b/contrib/libs/snappy/config-linux.h deleted file mode 100644 index d7997787fa8..00000000000 --- a/contrib/libs/snappy/config-linux.h +++ /dev/null @@ -1,66 +0,0 @@ -#ifndef THIRD_PARTY_SNAPPY_OPENSOURCE_CMAKE_CONFIG_H_ -#define THIRD_PARTY_SNAPPY_OPENSOURCE_CMAKE_CONFIG_H_ - -/* Define to 1 if the compiler supports __attribute__((always_inline)). */ -#define HAVE_ATTRIBUTE_ALWAYS_INLINE 0 - -/* Define to 1 if the compiler supports __builtin_ctz and friends. */ -#define HAVE_BUILTIN_CTZ 1 - -/* Define to 1 if the compiler supports __builtin_expect. */ -#define HAVE_BUILTIN_EXPECT 1 - -/* Define to 1 if you have a definition for mmap() in . */ -#define HAVE_FUNC_MMAP 1 - -/* Define to 1 if you have a definition for sysconf() in . */ -#define HAVE_FUNC_SYSCONF 1 - -/* Define to 1 if you have the `lzo2' library (-llzo2). */ -#define HAVE_LIBLZO2 0 - -/* Define to 1 if you have the `z' library (-lz). */ -#define HAVE_LIBZ 0 - -/* Define to 1 if you have the `lz4' library (-llz4). */ -#define HAVE_LIBLZ4 0 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_MMAN_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_RESOURCE_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_TIME_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_UIO_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_UNISTD_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_WINDOWS_H 0 - -/* Define to 1 if you target processors with SSSE3+ and have . */ -#define SNAPPY_HAVE_SSSE3 0 - -/* Define to 1 if you target processors with SSE4.2 and have . */ -#define SNAPPY_HAVE_X86_CRC32 0 - -/* Define to 1 if you target processors with BMI2+ and have . */ -#define SNAPPY_HAVE_BMI2 0 - -/* Define to 1 if you target processors with NEON and have . */ -#define SNAPPY_HAVE_NEON 0 - -/* Define to 1 if you have and and want to optimize - compression speed by using __crc32cw from . */ -#define SNAPPY_HAVE_NEON_CRC32 0 - -/* Define to 1 if your processor stores words with the most significant byte - first (like Motorola and SPARC, unlike Intel and VAX). */ -#define SNAPPY_IS_BIG_ENDIAN 0 - -#endif // THIRD_PARTY_SNAPPY_OPENSOURCE_CMAKE_CONFIG_H_ diff --git a/contrib/libs/snappy/config-win.h b/contrib/libs/snappy/config-win.h deleted file mode 100644 index 58b8be48392..00000000000 --- a/contrib/libs/snappy/config-win.h +++ /dev/null @@ -1,9 +0,0 @@ -#pragma once - -#include "config-linux.h" - -#undef HAVE_SYS_UIO_H -#undef HAVE_SYS_MMAN_H -#undef HAVE_UNISTD_H -#undef HAVE_BUILTIN_EXPECT -#undef HAVE_BUILTIN_CTZ diff --git a/contrib/libs/snappy/config.h b/contrib/libs/snappy/config.h deleted file mode 100644 index 5623f311fad..00000000000 --- a/contrib/libs/snappy/config.h +++ /dev/null @@ -1,7 +0,0 @@ -#pragma once - -#if defined(_MSC_VER) -# include "config-win.h" -#else -# include "config-linux.h" -#endif diff --git a/contrib/libs/snappy/include/snappy.h b/contrib/libs/snappy/include/snappy.h deleted file mode 100644 index a3dc8633d62..00000000000 --- a/contrib/libs/snappy/include/snappy.h +++ /dev/null @@ -1 +0,0 @@ -#include "../snappy.h" /* inclink generated by yamaker */ diff --git a/contrib/libs/snappy/snappy-c.cc b/contrib/libs/snappy/snappy-c.cc deleted file mode 100644 index 473a0b09786..00000000000 --- a/contrib/libs/snappy/snappy-c.cc +++ /dev/null @@ -1,90 +0,0 @@ -// Copyright 2011 Martin Gieseking . -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -#include "snappy.h" -#include "snappy-c.h" - -extern "C" { - -snappy_status snappy_compress(const char* input, - size_t input_length, - char* compressed, - size_t *compressed_length) { - if (*compressed_length < snappy_max_compressed_length(input_length)) { - return SNAPPY_BUFFER_TOO_SMALL; - } - snappy::RawCompress(input, input_length, compressed, compressed_length); - return SNAPPY_OK; -} - -snappy_status snappy_uncompress(const char* compressed, - size_t compressed_length, - char* uncompressed, - size_t* uncompressed_length) { - size_t real_uncompressed_length; - if (!snappy::GetUncompressedLength(compressed, - compressed_length, - &real_uncompressed_length)) { - return SNAPPY_INVALID_INPUT; - } - if (*uncompressed_length < real_uncompressed_length) { - return SNAPPY_BUFFER_TOO_SMALL; - } - if (!snappy::RawUncompress(compressed, compressed_length, uncompressed)) { - return SNAPPY_INVALID_INPUT; - } - *uncompressed_length = real_uncompressed_length; - return SNAPPY_OK; -} - -size_t snappy_max_compressed_length(size_t source_length) { - return snappy::MaxCompressedLength(source_length); -} - -snappy_status snappy_uncompressed_length(const char *compressed, - size_t compressed_length, - size_t *result) { - if (snappy::GetUncompressedLength(compressed, - compressed_length, - result)) { - return SNAPPY_OK; - } else { - return SNAPPY_INVALID_INPUT; - } -} - -snappy_status snappy_validate_compressed_buffer(const char *compressed, - size_t compressed_length) { - if (snappy::IsValidCompressedBuffer(compressed, compressed_length)) { - return SNAPPY_OK; - } else { - return SNAPPY_INVALID_INPUT; - } -} - -} // extern "C" diff --git a/contrib/libs/snappy/snappy-c.h b/contrib/libs/snappy/snappy-c.h deleted file mode 100644 index 32aa0c6b8b5..00000000000 --- a/contrib/libs/snappy/snappy-c.h +++ /dev/null @@ -1,138 +0,0 @@ -/* - * Copyright 2011 Martin Gieseking . - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * Plain C interface (a wrapper around the C++ implementation). - */ - -#ifndef THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_C_H_ -#define THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_C_H_ - -#ifdef __cplusplus -extern "C" { -#endif - -#include - -/* - * Return values; see the documentation for each function to know - * what each can return. - */ -typedef enum { - SNAPPY_OK = 0, - SNAPPY_INVALID_INPUT = 1, - SNAPPY_BUFFER_TOO_SMALL = 2 -} snappy_status; - -/* - * Takes the data stored in "input[0..input_length-1]" and stores - * it in the array pointed to by "compressed". - * - * signals the space available in "compressed". - * If it is not at least equal to "snappy_max_compressed_length(input_length)", - * SNAPPY_BUFFER_TOO_SMALL is returned. After successful compression, - * contains the true length of the compressed output, - * and SNAPPY_OK is returned. - * - * Example: - * size_t output_length = snappy_max_compressed_length(input_length); - * char* output = (char*)malloc(output_length); - * if (snappy_compress(input, input_length, output, &output_length) - * == SNAPPY_OK) { - * ... Process(output, output_length) ... - * } - * free(output); - */ -snappy_status snappy_compress(const char* input, - size_t input_length, - char* compressed, - size_t* compressed_length); - -/* - * Given data in "compressed[0..compressed_length-1]" generated by - * calling the snappy_compress routine, this routine stores - * the uncompressed data to - * uncompressed[0..uncompressed_length-1]. - * Returns failure (a value not equal to SNAPPY_OK) if the message - * is corrupted and could not be decrypted. - * - * signals the space available in "uncompressed". - * If it is not at least equal to the value returned by - * snappy_uncompressed_length for this stream, SNAPPY_BUFFER_TOO_SMALL - * is returned. After successful decompression, - * contains the true length of the decompressed output. - * - * Example: - * size_t output_length; - * if (snappy_uncompressed_length(input, input_length, &output_length) - * != SNAPPY_OK) { - * ... fail ... - * } - * char* output = (char*)malloc(output_length); - * if (snappy_uncompress(input, input_length, output, &output_length) - * == SNAPPY_OK) { - * ... Process(output, output_length) ... - * } - * free(output); - */ -snappy_status snappy_uncompress(const char* compressed, - size_t compressed_length, - char* uncompressed, - size_t* uncompressed_length); - -/* - * Returns the maximal size of the compressed representation of - * input data that is "source_length" bytes in length. - */ -size_t snappy_max_compressed_length(size_t source_length); - -/* - * REQUIRES: "compressed[]" was produced by snappy_compress() - * Returns SNAPPY_OK and stores the length of the uncompressed data in - * *result normally. Returns SNAPPY_INVALID_INPUT on parsing error. - * This operation takes O(1) time. - */ -snappy_status snappy_uncompressed_length(const char* compressed, - size_t compressed_length, - size_t* result); - -/* - * Check if the contents of "compressed[]" can be uncompressed successfully. - * Does not return the uncompressed data; if so, returns SNAPPY_OK, - * or if not, returns SNAPPY_INVALID_INPUT. - * Takes time proportional to compressed_length, but is usually at least a - * factor of four faster than actual decompression. - */ -snappy_status snappy_validate_compressed_buffer(const char* compressed, - size_t compressed_length); - -#ifdef __cplusplus -} // extern "C" -#endif - -#endif /* THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_C_H_ */ diff --git a/contrib/libs/snappy/snappy-internal.h b/contrib/libs/snappy/snappy-internal.h deleted file mode 100644 index 0923f399a38..00000000000 --- a/contrib/libs/snappy/snappy-internal.h +++ /dev/null @@ -1,395 +0,0 @@ -// Copyright 2008 Google Inc. All Rights Reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// -// Internals shared between the Snappy implementation and its unittest. - -#ifndef THIRD_PARTY_SNAPPY_SNAPPY_INTERNAL_H_ -#define THIRD_PARTY_SNAPPY_SNAPPY_INTERNAL_H_ - -#include "snappy-stubs-internal.h" - -#if SNAPPY_HAVE_SSSE3 -// Please do not replace with or with headers that assume more -// advanced SSE versions without checking with all the OWNERS. -#include -#include -#endif - -#if SNAPPY_HAVE_NEON -#include -#endif - -#if SNAPPY_HAVE_SSSE3 || SNAPPY_HAVE_NEON -#define SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE 1 -#else -#define SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE 0 -#endif - -namespace snappy { -namespace internal { - -#if SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE -#if SNAPPY_HAVE_SSSE3 -using V128 = __m128i; -#elif SNAPPY_HAVE_NEON -using V128 = uint8x16_t; -#endif - -// Load 128 bits of integer data. `src` must be 16-byte aligned. -inline V128 V128_Load(const V128* src); - -// Load 128 bits of integer data. `src` does not need to be aligned. -inline V128 V128_LoadU(const V128* src); - -// Store 128 bits of integer data. `dst` does not need to be aligned. -inline void V128_StoreU(V128* dst, V128 val); - -// Shuffle packed 8-bit integers using a shuffle mask. -// Each packed integer in the shuffle mask must be in [0,16). -inline V128 V128_Shuffle(V128 input, V128 shuffle_mask); - -// Constructs V128 with 16 chars |c|. -inline V128 V128_DupChar(char c); - -#if SNAPPY_HAVE_SSSE3 -inline V128 V128_Load(const V128* src) { return _mm_load_si128(src); } - -inline V128 V128_LoadU(const V128* src) { return _mm_loadu_si128(src); } - -inline void V128_StoreU(V128* dst, V128 val) { _mm_storeu_si128(dst, val); } - -inline V128 V128_Shuffle(V128 input, V128 shuffle_mask) { - return _mm_shuffle_epi8(input, shuffle_mask); -} - -inline V128 V128_DupChar(char c) { return _mm_set1_epi8(c); } - -#elif SNAPPY_HAVE_NEON -inline V128 V128_Load(const V128* src) { - return vld1q_u8(reinterpret_cast(src)); -} - -inline V128 V128_LoadU(const V128* src) { - return vld1q_u8(reinterpret_cast(src)); -} - -inline void V128_StoreU(V128* dst, V128 val) { - vst1q_u8(reinterpret_cast(dst), val); -} - -inline V128 V128_Shuffle(V128 input, V128 shuffle_mask) { - assert(vminvq_u8(shuffle_mask) >= 0 && vmaxvq_u8(shuffle_mask) <= 15); - return vqtbl1q_u8(input, shuffle_mask); -} - -inline V128 V128_DupChar(char c) { return vdupq_n_u8(c); } -#endif -#endif // SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE - -// Working memory performs a single allocation to hold all scratch space -// required for compression. -class WorkingMemory { - public: - explicit WorkingMemory(size_t input_size); - ~WorkingMemory(); - - // Allocates and clears a hash table using memory in "*this", - // stores the number of buckets in "*table_size" and returns a pointer to - // the base of the hash table. - uint16_t* GetHashTable(size_t fragment_size, int* table_size) const; - char* GetScratchInput() const { return input_; } - char* GetScratchOutput() const { return output_; } - - private: - char* mem_; // the allocated memory, never nullptr - size_t size_; // the size of the allocated memory, never 0 - uint16_t* table_; // the pointer to the hashtable - char* input_; // the pointer to the input scratch buffer - char* output_; // the pointer to the output scratch buffer - - // No copying - WorkingMemory(const WorkingMemory&); - void operator=(const WorkingMemory&); -}; - -// Flat array compression that does not emit the "uncompressed length" -// prefix. Compresses "input" string to the "*op" buffer. -// -// REQUIRES: "input_length <= kBlockSize" -// REQUIRES: "op" points to an array of memory that is at least -// "MaxCompressedLength(input_length)" in size. -// REQUIRES: All elements in "table[0..table_size-1]" are initialized to zero. -// REQUIRES: "table_size" is a power of two -// -// Returns an "end" pointer into "op" buffer. -// "end - op" is the compressed size of "input". -char* CompressFragment(const char* input, - size_t input_length, - char* op, - uint16_t* table, - const int table_size); - -// Find the largest n such that -// -// s1[0,n-1] == s2[0,n-1] -// and n <= (s2_limit - s2). -// -// Return make_pair(n, n < 8). -// Does not read *s2_limit or beyond. -// Does not read *(s1 + (s2_limit - s2)) or beyond. -// Requires that s2_limit >= s2. -// -// In addition populate *data with the next 5 bytes from the end of the match. -// This is only done if 8 bytes are available (s2_limit - s2 >= 8). The point is -// that on some arch's this can be done faster in this routine than subsequent -// loading from s2 + n. -// -// Separate implementation for 64-bit, little-endian cpus. -#if !SNAPPY_IS_BIG_ENDIAN && \ - (defined(__x86_64__) || defined(_M_X64) || defined(ARCH_PPC) || \ - defined(ARCH_ARM)) -static inline std::pair FindMatchLength(const char* s1, - const char* s2, - const char* s2_limit, - uint64_t* data) { - assert(s2_limit >= s2); - size_t matched = 0; - - // This block isn't necessary for correctness; we could just start looping - // immediately. As an optimization though, it is useful. It creates some not - // uncommon code paths that determine, without extra effort, whether the match - // length is less than 8. In short, we are hoping to avoid a conditional - // branch, and perhaps get better code layout from the C++ compiler. - if (SNAPPY_PREDICT_TRUE(s2 <= s2_limit - 16)) { - uint64_t a1 = UNALIGNED_LOAD64(s1); - uint64_t a2 = UNALIGNED_LOAD64(s2); - if (SNAPPY_PREDICT_TRUE(a1 != a2)) { - // This code is critical for performance. The reason is that it determines - // how much to advance `ip` (s2). This obviously depends on both the loads - // from the `candidate` (s1) and `ip`. Furthermore the next `candidate` - // depends on the advanced `ip` calculated here through a load, hash and - // new candidate hash lookup (a lot of cycles). This makes s1 (ie. - // `candidate`) the variable that limits throughput. This is the reason we - // go through hoops to have this function update `data` for the next iter. - // The straightforward code would use *data, given by - // - // *data = UNALIGNED_LOAD64(s2 + matched_bytes) (Latency of 5 cycles), - // - // as input for the hash table lookup to find next candidate. However - // this forces the load on the data dependency chain of s1, because - // matched_bytes directly depends on s1. However matched_bytes is 0..7, so - // we can also calculate *data by - // - // *data = AlignRight(UNALIGNED_LOAD64(s2), UNALIGNED_LOAD64(s2 + 8), - // matched_bytes); - // - // The loads do not depend on s1 anymore and are thus off the bottleneck. - // The straightforward implementation on x86_64 would be to use - // - // shrd rax, rdx, cl (cl being matched_bytes * 8) - // - // unfortunately shrd with a variable shift has a 4 cycle latency. So this - // only wins 1 cycle. The BMI2 shrx instruction is a 1 cycle variable - // shift instruction but can only shift 64 bits. If we focus on just - // obtaining the least significant 4 bytes, we can obtain this by - // - // *data = ConditionalMove(matched_bytes < 4, UNALIGNED_LOAD64(s2), - // UNALIGNED_LOAD64(s2 + 4) >> ((matched_bytes & 3) * 8); - // - // Writen like above this is not a big win, the conditional move would be - // a cmp followed by a cmov (2 cycles) followed by a shift (1 cycle). - // However matched_bytes < 4 is equal to - // static_cast(xorval) != 0. Writen that way, the conditional - // move (2 cycles) can execute in parallel with FindLSBSetNonZero64 - // (tzcnt), which takes 3 cycles. - uint64_t xorval = a1 ^ a2; - int shift = Bits::FindLSBSetNonZero64(xorval); - size_t matched_bytes = shift >> 3; - uint64_t a3 = UNALIGNED_LOAD64(s2 + 4); -#ifndef __x86_64__ - a2 = static_cast(xorval) == 0 ? a3 : a2; -#else - // Ideally this would just be - // - // a2 = static_cast(xorval) == 0 ? a3 : a2; - // - // However clang correctly infers that the above statement participates on - // a critical data dependency chain and thus, unfortunately, refuses to - // use a conditional move (it's tuned to cut data dependencies). In this - // case there is a longer parallel chain anyway AND this will be fairly - // unpredictable. - asm("testl %k2, %k2\n\t" - "cmovzq %1, %0\n\t" - : "+r"(a2) - : "r"(a3), "r"(xorval) - : "cc"); -#endif - *data = a2 >> (shift & (3 * 8)); - return std::pair(matched_bytes, true); - } else { - matched = 8; - s2 += 8; - } - } - - // Find out how long the match is. We loop over the data 64 bits at a - // time until we find a 64-bit block that doesn't match; then we find - // the first non-matching bit and use that to calculate the total - // length of the match. - while (SNAPPY_PREDICT_TRUE(s2 <= s2_limit - 16)) { - uint64_t a1 = UNALIGNED_LOAD64(s1 + matched); - uint64_t a2 = UNALIGNED_LOAD64(s2); - if (a1 == a2) { - s2 += 8; - matched += 8; - } else { - uint64_t xorval = a1 ^ a2; - int shift = Bits::FindLSBSetNonZero64(xorval); - size_t matched_bytes = shift >> 3; - uint64_t a3 = UNALIGNED_LOAD64(s2 + 4); -#ifndef __x86_64__ - a2 = static_cast(xorval) == 0 ? a3 : a2; -#else - asm("testl %k2, %k2\n\t" - "cmovzq %1, %0\n\t" - : "+r"(a2) - : "r"(a3), "r"(xorval) - : "cc"); -#endif - *data = a2 >> (shift & (3 * 8)); - matched += matched_bytes; - assert(matched >= 8); - return std::pair(matched, false); - } - } - while (SNAPPY_PREDICT_TRUE(s2 < s2_limit)) { - if (s1[matched] == *s2) { - ++s2; - ++matched; - } else { - if (s2 <= s2_limit - 8) { - *data = UNALIGNED_LOAD64(s2); - } - return std::pair(matched, matched < 8); - } - } - return std::pair(matched, matched < 8); -} -#else -static inline std::pair FindMatchLength(const char* s1, - const char* s2, - const char* s2_limit, - uint64_t* data) { - // Implementation based on the x86-64 version, above. - assert(s2_limit >= s2); - int matched = 0; - - while (s2 <= s2_limit - 4 && - UNALIGNED_LOAD32(s2) == UNALIGNED_LOAD32(s1 + matched)) { - s2 += 4; - matched += 4; - } - if (LittleEndian::IsLittleEndian() && s2 <= s2_limit - 4) { - uint32_t x = UNALIGNED_LOAD32(s2) ^ UNALIGNED_LOAD32(s1 + matched); - int matching_bits = Bits::FindLSBSetNonZero(x); - matched += matching_bits >> 3; - s2 += matching_bits >> 3; - } else { - while ((s2 < s2_limit) && (s1[matched] == *s2)) { - ++s2; - ++matched; - } - } - if (s2 <= s2_limit - 8) *data = LittleEndian::Load64(s2); - return std::pair(matched, matched < 8); -} -#endif - -// Lookup tables for decompression code. Give --snappy_dump_decompression_table -// to the unit test to recompute char_table. - -enum { - LITERAL = 0, - COPY_1_BYTE_OFFSET = 1, // 3 bit length + 3 bits of offset in opcode - COPY_2_BYTE_OFFSET = 2, - COPY_4_BYTE_OFFSET = 3 -}; -static const int kMaximumTagLength = 5; // COPY_4_BYTE_OFFSET plus the actual offset. - -// Data stored per entry in lookup table: -// Range Bits-used Description -// ------------------------------------ -// 1..64 0..7 Literal/copy length encoded in opcode byte -// 0..7 8..10 Copy offset encoded in opcode byte / 256 -// 0..4 11..13 Extra bytes after opcode -// -// We use eight bits for the length even though 7 would have sufficed -// because of efficiency reasons: -// (1) Extracting a byte is faster than a bit-field -// (2) It properly aligns copy offset so we do not need a <<8 -static constexpr uint16_t char_table[256] = { - // clang-format off - 0x0001, 0x0804, 0x1001, 0x2001, 0x0002, 0x0805, 0x1002, 0x2002, - 0x0003, 0x0806, 0x1003, 0x2003, 0x0004, 0x0807, 0x1004, 0x2004, - 0x0005, 0x0808, 0x1005, 0x2005, 0x0006, 0x0809, 0x1006, 0x2006, - 0x0007, 0x080a, 0x1007, 0x2007, 0x0008, 0x080b, 0x1008, 0x2008, - 0x0009, 0x0904, 0x1009, 0x2009, 0x000a, 0x0905, 0x100a, 0x200a, - 0x000b, 0x0906, 0x100b, 0x200b, 0x000c, 0x0907, 0x100c, 0x200c, - 0x000d, 0x0908, 0x100d, 0x200d, 0x000e, 0x0909, 0x100e, 0x200e, - 0x000f, 0x090a, 0x100f, 0x200f, 0x0010, 0x090b, 0x1010, 0x2010, - 0x0011, 0x0a04, 0x1011, 0x2011, 0x0012, 0x0a05, 0x1012, 0x2012, - 0x0013, 0x0a06, 0x1013, 0x2013, 0x0014, 0x0a07, 0x1014, 0x2014, - 0x0015, 0x0a08, 0x1015, 0x2015, 0x0016, 0x0a09, 0x1016, 0x2016, - 0x0017, 0x0a0a, 0x1017, 0x2017, 0x0018, 0x0a0b, 0x1018, 0x2018, - 0x0019, 0x0b04, 0x1019, 0x2019, 0x001a, 0x0b05, 0x101a, 0x201a, - 0x001b, 0x0b06, 0x101b, 0x201b, 0x001c, 0x0b07, 0x101c, 0x201c, - 0x001d, 0x0b08, 0x101d, 0x201d, 0x001e, 0x0b09, 0x101e, 0x201e, - 0x001f, 0x0b0a, 0x101f, 0x201f, 0x0020, 0x0b0b, 0x1020, 0x2020, - 0x0021, 0x0c04, 0x1021, 0x2021, 0x0022, 0x0c05, 0x1022, 0x2022, - 0x0023, 0x0c06, 0x1023, 0x2023, 0x0024, 0x0c07, 0x1024, 0x2024, - 0x0025, 0x0c08, 0x1025, 0x2025, 0x0026, 0x0c09, 0x1026, 0x2026, - 0x0027, 0x0c0a, 0x1027, 0x2027, 0x0028, 0x0c0b, 0x1028, 0x2028, - 0x0029, 0x0d04, 0x1029, 0x2029, 0x002a, 0x0d05, 0x102a, 0x202a, - 0x002b, 0x0d06, 0x102b, 0x202b, 0x002c, 0x0d07, 0x102c, 0x202c, - 0x002d, 0x0d08, 0x102d, 0x202d, 0x002e, 0x0d09, 0x102e, 0x202e, - 0x002f, 0x0d0a, 0x102f, 0x202f, 0x0030, 0x0d0b, 0x1030, 0x2030, - 0x0031, 0x0e04, 0x1031, 0x2031, 0x0032, 0x0e05, 0x1032, 0x2032, - 0x0033, 0x0e06, 0x1033, 0x2033, 0x0034, 0x0e07, 0x1034, 0x2034, - 0x0035, 0x0e08, 0x1035, 0x2035, 0x0036, 0x0e09, 0x1036, 0x2036, - 0x0037, 0x0e0a, 0x1037, 0x2037, 0x0038, 0x0e0b, 0x1038, 0x2038, - 0x0039, 0x0f04, 0x1039, 0x2039, 0x003a, 0x0f05, 0x103a, 0x203a, - 0x003b, 0x0f06, 0x103b, 0x203b, 0x003c, 0x0f07, 0x103c, 0x203c, - 0x0801, 0x0f08, 0x103d, 0x203d, 0x1001, 0x0f09, 0x103e, 0x203e, - 0x1801, 0x0f0a, 0x103f, 0x203f, 0x2001, 0x0f0b, 0x1040, 0x2040, - // clang-format on -}; - -} // end namespace internal -} // end namespace snappy - -#endif // THIRD_PARTY_SNAPPY_SNAPPY_INTERNAL_H_ diff --git a/contrib/libs/snappy/snappy-sinksource.cc b/contrib/libs/snappy/snappy-sinksource.cc deleted file mode 100644 index 8214964a7ec..00000000000 --- a/contrib/libs/snappy/snappy-sinksource.cc +++ /dev/null @@ -1,121 +0,0 @@ -// Copyright 2011 Google Inc. All Rights Reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -#include -#include - -#include "snappy-sinksource.h" - -namespace snappy { - -Source::~Source() = default; - -Sink::~Sink() = default; - -char* Sink::GetAppendBuffer(size_t length, char* scratch) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)length; - - return scratch; -} - -char* Sink::GetAppendBufferVariable( - size_t min_size, size_t desired_size_hint, char* scratch, - size_t scratch_size, size_t* allocated_size) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)min_size; - (void)desired_size_hint; - - *allocated_size = scratch_size; - return scratch; -} - -void Sink::AppendAndTakeOwnership( - char* bytes, size_t n, - void (*deleter)(void*, const char*, size_t), - void *deleter_arg) { - Append(bytes, n); - (*deleter)(deleter_arg, bytes, n); -} - -ByteArraySource::~ByteArraySource() = default; - -size_t ByteArraySource::Available() const { return left_; } - -const char* ByteArraySource::Peek(size_t* len) { - *len = left_; - return ptr_; -} - -void ByteArraySource::Skip(size_t n) { - left_ -= n; - ptr_ += n; -} - -UncheckedByteArraySink::~UncheckedByteArraySink() { } - -void UncheckedByteArraySink::Append(const char* data, size_t n) { - // Do no copying if the caller filled in the result of GetAppendBuffer() - if (data != dest_) { - std::memcpy(dest_, data, n); - } - dest_ += n; -} - -char* UncheckedByteArraySink::GetAppendBuffer(size_t len, char* scratch) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)len; - (void)scratch; - - return dest_; -} - -void UncheckedByteArraySink::AppendAndTakeOwnership( - char* bytes, size_t n, - void (*deleter)(void*, const char*, size_t), - void *deleter_arg) { - if (bytes != dest_) { - std::memcpy(dest_, bytes, n); - (*deleter)(deleter_arg, bytes, n); - } - dest_ += n; -} - -char* UncheckedByteArraySink::GetAppendBufferVariable( - size_t min_size, size_t desired_size_hint, char* scratch, - size_t scratch_size, size_t* allocated_size) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)min_size; - (void)scratch; - (void)scratch_size; - - *allocated_size = desired_size_hint; - return dest_; -} - -} // namespace snappy diff --git a/contrib/libs/snappy/snappy-sinksource.h b/contrib/libs/snappy/snappy-sinksource.h deleted file mode 100644 index 3c74e1bb6e9..00000000000 --- a/contrib/libs/snappy/snappy-sinksource.h +++ /dev/null @@ -1,182 +0,0 @@ -// Copyright 2011 Google Inc. All Rights Reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -#ifndef THIRD_PARTY_SNAPPY_SNAPPY_SINKSOURCE_H_ -#define THIRD_PARTY_SNAPPY_SNAPPY_SINKSOURCE_H_ - -#include - -namespace snappy { - -// A Sink is an interface that consumes a sequence of bytes. -class Sink { - public: - Sink() { } - virtual ~Sink(); - - // Append "bytes[0,n-1]" to this. - virtual void Append(const char* bytes, size_t n) = 0; - - // Returns a writable buffer of the specified length for appending. - // May return a pointer to the caller-owned scratch buffer which - // must have at least the indicated length. The returned buffer is - // only valid until the next operation on this Sink. - // - // After writing at most "length" bytes, call Append() with the - // pointer returned from this function and the number of bytes - // written. Many Append() implementations will avoid copying - // bytes if this function returned an internal buffer. - // - // If a non-scratch buffer is returned, the caller may only pass a - // prefix of it to Append(). That is, it is not correct to pass an - // interior pointer of the returned array to Append(). - // - // The default implementation always returns the scratch buffer. - virtual char* GetAppendBuffer(size_t length, char* scratch); - - // For higher performance, Sink implementations can provide custom - // AppendAndTakeOwnership() and GetAppendBufferVariable() methods. - // These methods can reduce the number of copies done during - // compression/decompression. - - // Append "bytes[0,n-1] to the sink. Takes ownership of "bytes" - // and calls the deleter function as (*deleter)(deleter_arg, bytes, n) - // to free the buffer. deleter function must be non NULL. - // - // The default implementation just calls Append and frees "bytes". - // Other implementations may avoid a copy while appending the buffer. - virtual void AppendAndTakeOwnership( - char* bytes, size_t n, void (*deleter)(void*, const char*, size_t), - void *deleter_arg); - - // Returns a writable buffer for appending and writes the buffer's capacity to - // *allocated_size. Guarantees *allocated_size >= min_size. - // May return a pointer to the caller-owned scratch buffer which must have - // scratch_size >= min_size. - // - // The returned buffer is only valid until the next operation - // on this ByteSink. - // - // After writing at most *allocated_size bytes, call Append() with the - // pointer returned from this function and the number of bytes written. - // Many Append() implementations will avoid copying bytes if this function - // returned an internal buffer. - // - // If the sink implementation allocates or reallocates an internal buffer, - // it should use the desired_size_hint if appropriate. If a caller cannot - // provide a reasonable guess at the desired capacity, it should set - // desired_size_hint = 0. - // - // If a non-scratch buffer is returned, the caller may only pass - // a prefix to it to Append(). That is, it is not correct to pass an - // interior pointer to Append(). - // - // The default implementation always returns the scratch buffer. - virtual char* GetAppendBufferVariable( - size_t min_size, size_t desired_size_hint, char* scratch, - size_t scratch_size, size_t* allocated_size); - - private: - // No copying - Sink(const Sink&); - void operator=(const Sink&); -}; - -// A Source is an interface that yields a sequence of bytes -class Source { - public: - Source() { } - virtual ~Source(); - - // Return the number of bytes left to read from the source - virtual size_t Available() const = 0; - - // Peek at the next flat region of the source. Does not reposition - // the source. The returned region is empty iff Available()==0. - // - // Returns a pointer to the beginning of the region and store its - // length in *len. - // - // The returned region is valid until the next call to Skip() or - // until this object is destroyed, whichever occurs first. - // - // The returned region may be larger than Available() (for example - // if this ByteSource is a view on a substring of a larger source). - // The caller is responsible for ensuring that it only reads the - // Available() bytes. - virtual const char* Peek(size_t* len) = 0; - - // Skip the next n bytes. Invalidates any buffer returned by - // a previous call to Peek(). - // REQUIRES: Available() >= n - virtual void Skip(size_t n) = 0; - - private: - // No copying - Source(const Source&); - void operator=(const Source&); -}; - -// A Source implementation that yields the contents of a flat array -class ByteArraySource : public Source { - public: - ByteArraySource(const char* p, size_t n) : ptr_(p), left_(n) { } - ~ByteArraySource() override; - size_t Available() const override; - const char* Peek(size_t* len) override; - void Skip(size_t n) override; - private: - const char* ptr_; - size_t left_; -}; - -// A Sink implementation that writes to a flat array without any bound checks. -class UncheckedByteArraySink : public Sink { - public: - explicit UncheckedByteArraySink(char* dest) : dest_(dest) { } - ~UncheckedByteArraySink() override; - void Append(const char* data, size_t n) override; - char* GetAppendBuffer(size_t len, char* scratch) override; - char* GetAppendBufferVariable( - size_t min_size, size_t desired_size_hint, char* scratch, - size_t scratch_size, size_t* allocated_size) override; - void AppendAndTakeOwnership( - char* bytes, size_t n, void (*deleter)(void*, const char*, size_t), - void *deleter_arg) override; - - // Return the current output pointer so that a caller can see how - // many bytes were produced. - // Note: this is not a Sink method. - char* CurrentDestination() const { return dest_; } - private: - char* dest_; -}; - -} // namespace snappy - -#endif // THIRD_PARTY_SNAPPY_SNAPPY_SINKSOURCE_H_ diff --git a/contrib/libs/snappy/snappy-stubs-internal.cc b/contrib/libs/snappy/snappy-stubs-internal.cc deleted file mode 100644 index 0bc8c2d3449..00000000000 --- a/contrib/libs/snappy/snappy-stubs-internal.cc +++ /dev/null @@ -1,42 +0,0 @@ -// Copyright 2011 Google Inc. All Rights Reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -#include -#include - -#include "snappy-stubs-internal.h" - -namespace snappy { - -void Varint::Append32(std::string* s, uint32_t value) { - char buf[Varint::kMax32]; - const char* p = Varint::Encode32(buf, value); - s->append(buf, p - buf); -} - -} // namespace snappy diff --git a/contrib/libs/snappy/snappy-stubs-internal.h b/contrib/libs/snappy/snappy-stubs-internal.h deleted file mode 100644 index 1548ed7ac70..00000000000 --- a/contrib/libs/snappy/snappy-stubs-internal.h +++ /dev/null @@ -1,525 +0,0 @@ -// Copyright 2011 Google Inc. All Rights Reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// -// Various stubs for the open-source version of Snappy. - -#ifndef THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_ -#define THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_ - -#if HAVE_CONFIG_H -#include "config.h" -#endif - -#include - -#include -#include -#include -#include -#include - -#if HAVE_SYS_MMAN_H -#include -#endif - -#if HAVE_UNISTD_H -#include -#endif - -#if defined(_MSC_VER) -#include -#endif // defined(_MSC_VER) - -#ifndef __has_feature -#define __has_feature(x) 0 -#endif - -#if __has_feature(memory_sanitizer) -#include -#define SNAPPY_ANNOTATE_MEMORY_IS_INITIALIZED(address, size) \ - __msan_unpoison((address), (size)) -#else -#define SNAPPY_ANNOTATE_MEMORY_IS_INITIALIZED(address, size) /* empty */ -#endif // __has_feature(memory_sanitizer) - -#include "snappy-stubs-public.h" - -// Used to enable 64-bit optimized versions of some routines. -#if defined(__PPC64__) || defined(__powerpc64__) -#define ARCH_PPC 1 -#elif defined(__aarch64__) || defined(_M_ARM64) -#define ARCH_ARM 1 -#endif - -// Needed by OS X, among others. -#ifndef MAP_ANONYMOUS -#define MAP_ANONYMOUS MAP_ANON -#endif - -// The size of an array, if known at compile-time. -// Will give unexpected results if used on a pointer. -// We undefine it first, since some compilers already have a definition. -#ifdef ARRAYSIZE -#undef ARRAYSIZE -#endif -#define ARRAYSIZE(a) int{sizeof(a) / sizeof(*(a))} - -// Static prediction hints. -#if HAVE_BUILTIN_EXPECT -#define SNAPPY_PREDICT_FALSE(x) (__builtin_expect(x, 0)) -#define SNAPPY_PREDICT_TRUE(x) (__builtin_expect(!!(x), 1)) -#else -#define SNAPPY_PREDICT_FALSE(x) x -#define SNAPPY_PREDICT_TRUE(x) x -#endif // HAVE_BUILTIN_EXPECT - -// Inlining hints. -#if HAVE_ATTRIBUTE_ALWAYS_INLINE -#define SNAPPY_ATTRIBUTE_ALWAYS_INLINE __attribute__((always_inline)) -#else -#define SNAPPY_ATTRIBUTE_ALWAYS_INLINE -#endif // HAVE_ATTRIBUTE_ALWAYS_INLINE - -// Stubbed version of ABSL_FLAG. -// -// In the open source version, flags can only be changed at compile time. -#define SNAPPY_FLAG(flag_type, flag_name, default_value, help) \ - flag_type FLAGS_ ## flag_name = default_value - -namespace snappy { - -// Stubbed version of absl::GetFlag(). -template -inline T GetFlag(T flag) { return flag; } - -static const uint32_t kuint32max = std::numeric_limits::max(); -static const int64_t kint64max = std::numeric_limits::max(); - -// Potentially unaligned loads and stores. - -inline uint16_t UNALIGNED_LOAD16(const void *p) { - // Compiles to a single movzx/ldrh on clang/gcc/msvc. - uint16_t v; - std::memcpy(&v, p, sizeof(v)); - return v; -} - -inline uint32_t UNALIGNED_LOAD32(const void *p) { - // Compiles to a single mov/ldr on clang/gcc/msvc. - uint32_t v; - std::memcpy(&v, p, sizeof(v)); - return v; -} - -inline uint64_t UNALIGNED_LOAD64(const void *p) { - // Compiles to a single mov/ldr on clang/gcc/msvc. - uint64_t v; - std::memcpy(&v, p, sizeof(v)); - return v; -} - -inline void UNALIGNED_STORE16(void *p, uint16_t v) { - // Compiles to a single mov/strh on clang/gcc/msvc. - std::memcpy(p, &v, sizeof(v)); -} - -inline void UNALIGNED_STORE32(void *p, uint32_t v) { - // Compiles to a single mov/str on clang/gcc/msvc. - std::memcpy(p, &v, sizeof(v)); -} - -inline void UNALIGNED_STORE64(void *p, uint64_t v) { - // Compiles to a single mov/str on clang/gcc/msvc. - std::memcpy(p, &v, sizeof(v)); -} - -// Convert to little-endian storage, opposite of network format. -// Convert x from host to little endian: x = LittleEndian.FromHost(x); -// convert x from little endian to host: x = LittleEndian.ToHost(x); -// -// Store values into unaligned memory converting to little endian order: -// LittleEndian.Store16(p, x); -// -// Load unaligned values stored in little endian converting to host order: -// x = LittleEndian.Load16(p); -class LittleEndian { - public: - // Functions to do unaligned loads and stores in little-endian order. - static inline uint16_t Load16(const void *ptr) { - // Compiles to a single mov/str on recent clang and gcc. -#if SNAPPY_IS_BIG_ENDIAN - const uint8_t* const buffer = reinterpret_cast(ptr); - return (static_cast(buffer[0])) | - (static_cast(buffer[1]) << 8); -#else - // memcpy() turns into a single instruction early in the optimization - // pipeline (relatively to a series of byte accesses). So, using memcpy - // instead of byte accesses may lead to better decisions in more stages of - // the optimization pipeline. - uint16_t value; - std::memcpy(&value, ptr, 2); - return value; -#endif - } - - static inline uint32_t Load32(const void *ptr) { - // Compiles to a single mov/str on recent clang and gcc. -#if SNAPPY_IS_BIG_ENDIAN - const uint8_t* const buffer = reinterpret_cast(ptr); - return (static_cast(buffer[0])) | - (static_cast(buffer[1]) << 8) | - (static_cast(buffer[2]) << 16) | - (static_cast(buffer[3]) << 24); -#else - // See Load16() for the rationale of using memcpy(). - uint32_t value; - std::memcpy(&value, ptr, 4); - return value; -#endif - } - - static inline uint64_t Load64(const void *ptr) { - // Compiles to a single mov/str on recent clang and gcc. -#if SNAPPY_IS_BIG_ENDIAN - const uint8_t* const buffer = reinterpret_cast(ptr); - return (static_cast(buffer[0])) | - (static_cast(buffer[1]) << 8) | - (static_cast(buffer[2]) << 16) | - (static_cast(buffer[3]) << 24) | - (static_cast(buffer[4]) << 32) | - (static_cast(buffer[5]) << 40) | - (static_cast(buffer[6]) << 48) | - (static_cast(buffer[7]) << 56); -#else - // See Load16() for the rationale of using memcpy(). - uint64_t value; - std::memcpy(&value, ptr, 8); - return value; -#endif - } - - static inline void Store16(void *dst, uint16_t value) { - // Compiles to a single mov/str on recent clang and gcc. -#if SNAPPY_IS_BIG_ENDIAN - uint8_t* const buffer = reinterpret_cast(dst); - buffer[0] = static_cast(value); - buffer[1] = static_cast(value >> 8); -#else - // See Load16() for the rationale of using memcpy(). - std::memcpy(dst, &value, 2); -#endif - } - - static void Store32(void *dst, uint32_t value) { - // Compiles to a single mov/str on recent clang and gcc. -#if SNAPPY_IS_BIG_ENDIAN - uint8_t* const buffer = reinterpret_cast(dst); - buffer[0] = static_cast(value); - buffer[1] = static_cast(value >> 8); - buffer[2] = static_cast(value >> 16); - buffer[3] = static_cast(value >> 24); -#else - // See Load16() for the rationale of using memcpy(). - std::memcpy(dst, &value, 4); -#endif - } - - static void Store64(void* dst, uint64_t value) { - // Compiles to a single mov/str on recent clang and gcc. -#if SNAPPY_IS_BIG_ENDIAN - uint8_t* const buffer = reinterpret_cast(dst); - buffer[0] = static_cast(value); - buffer[1] = static_cast(value >> 8); - buffer[2] = static_cast(value >> 16); - buffer[3] = static_cast(value >> 24); - buffer[4] = static_cast(value >> 32); - buffer[5] = static_cast(value >> 40); - buffer[6] = static_cast(value >> 48); - buffer[7] = static_cast(value >> 56); -#else - // See Load16() for the rationale of using memcpy(). - std::memcpy(dst, &value, 8); -#endif - } - - static inline constexpr bool IsLittleEndian() { -#if SNAPPY_IS_BIG_ENDIAN - return false; -#else - return true; -#endif // SNAPPY_IS_BIG_ENDIAN - } -}; - -// Some bit-manipulation functions. -class Bits { - public: - // Return floor(log2(n)) for positive integer n. - static int Log2FloorNonZero(uint32_t n); - - // Return floor(log2(n)) for positive integer n. Returns -1 iff n == 0. - static int Log2Floor(uint32_t n); - - // Return the first set least / most significant bit, 0-indexed. Returns an - // undefined value if n == 0. FindLSBSetNonZero() is similar to ffs() except - // that it's 0-indexed. - static int FindLSBSetNonZero(uint32_t n); - - static int FindLSBSetNonZero64(uint64_t n); - - private: - // No copying - Bits(const Bits&); - void operator=(const Bits&); -}; - -#if HAVE_BUILTIN_CTZ - -inline int Bits::Log2FloorNonZero(uint32_t n) { - assert(n != 0); - // (31 ^ x) is equivalent to (31 - x) for x in [0, 31]. An easy proof - // represents subtraction in base 2 and observes that there's no carry. - // - // GCC and Clang represent __builtin_clz on x86 as 31 ^ _bit_scan_reverse(x). - // Using "31 ^" here instead of "31 -" allows the optimizer to strip the - // function body down to _bit_scan_reverse(x). - return 31 ^ __builtin_clz(n); -} - -inline int Bits::Log2Floor(uint32_t n) { - return (n == 0) ? -1 : Bits::Log2FloorNonZero(n); -} - -inline int Bits::FindLSBSetNonZero(uint32_t n) { - assert(n != 0); - return __builtin_ctz(n); -} - -#elif defined(_MSC_VER) - -inline int Bits::Log2FloorNonZero(uint32_t n) { - assert(n != 0); - // NOLINTNEXTLINE(runtime/int): The MSVC intrinsic demands unsigned long. - unsigned long where; - _BitScanReverse(&where, n); - return static_cast(where); -} - -inline int Bits::Log2Floor(uint32_t n) { - // NOLINTNEXTLINE(runtime/int): The MSVC intrinsic demands unsigned long. - unsigned long where; - if (_BitScanReverse(&where, n)) - return static_cast(where); - return -1; -} - -inline int Bits::FindLSBSetNonZero(uint32_t n) { - assert(n != 0); - // NOLINTNEXTLINE(runtime/int): The MSVC intrinsic demands unsigned long. - unsigned long where; - if (_BitScanForward(&where, n)) - return static_cast(where); - return 32; -} - -#else // Portable versions. - -inline int Bits::Log2FloorNonZero(uint32_t n) { - assert(n != 0); - - int log = 0; - uint32_t value = n; - for (int i = 4; i >= 0; --i) { - int shift = (1 << i); - uint32_t x = value >> shift; - if (x != 0) { - value = x; - log += shift; - } - } - assert(value == 1); - return log; -} - -inline int Bits::Log2Floor(uint32_t n) { - return (n == 0) ? -1 : Bits::Log2FloorNonZero(n); -} - -inline int Bits::FindLSBSetNonZero(uint32_t n) { - assert(n != 0); - - int rc = 31; - for (int i = 4, shift = 1 << 4; i >= 0; --i) { - const uint32_t x = n << shift; - if (x != 0) { - n = x; - rc -= shift; - } - shift >>= 1; - } - return rc; -} - -#endif // End portable versions. - -#if HAVE_BUILTIN_CTZ - -inline int Bits::FindLSBSetNonZero64(uint64_t n) { - assert(n != 0); - return __builtin_ctzll(n); -} - -#elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_ARM64)) -// _BitScanForward64() is only available on x64 and ARM64. - -inline int Bits::FindLSBSetNonZero64(uint64_t n) { - assert(n != 0); - // NOLINTNEXTLINE(runtime/int): The MSVC intrinsic demands unsigned long. - unsigned long where; - if (_BitScanForward64(&where, n)) - return static_cast(where); - return 64; -} - -#else // Portable version. - -// FindLSBSetNonZero64() is defined in terms of FindLSBSetNonZero(). -inline int Bits::FindLSBSetNonZero64(uint64_t n) { - assert(n != 0); - - const uint32_t bottombits = static_cast(n); - if (bottombits == 0) { - // Bottom bits are zero, so scan the top bits. - return 32 + FindLSBSetNonZero(static_cast(n >> 32)); - } else { - return FindLSBSetNonZero(bottombits); - } -} - -#endif // HAVE_BUILTIN_CTZ - -// Variable-length integer encoding. -class Varint { - public: - // Maximum lengths of varint encoding of uint32_t. - static const int kMax32 = 5; - - // Attempts to parse a varint32 from a prefix of the bytes in [ptr,limit-1]. - // Never reads a character at or beyond limit. If a valid/terminated varint32 - // was found in the range, stores it in *OUTPUT and returns a pointer just - // past the last byte of the varint32. Else returns NULL. On success, - // "result <= limit". - static const char* Parse32WithLimit(const char* ptr, const char* limit, - uint32_t* OUTPUT); - - // REQUIRES "ptr" points to a buffer of length sufficient to hold "v". - // EFFECTS Encodes "v" into "ptr" and returns a pointer to the - // byte just past the last encoded byte. - static char* Encode32(char* ptr, uint32_t v); - - // EFFECTS Appends the varint representation of "value" to "*s". - static void Append32(std::string* s, uint32_t value); -}; - -inline const char* Varint::Parse32WithLimit(const char* p, - const char* l, - uint32_t* OUTPUT) { - const unsigned char* ptr = reinterpret_cast(p); - const unsigned char* limit = reinterpret_cast(l); - uint32_t b, result; - if (ptr >= limit) return NULL; - b = *(ptr++); result = b & 127; if (b < 128) goto done; - if (ptr >= limit) return NULL; - b = *(ptr++); result |= (b & 127) << 7; if (b < 128) goto done; - if (ptr >= limit) return NULL; - b = *(ptr++); result |= (b & 127) << 14; if (b < 128) goto done; - if (ptr >= limit) return NULL; - b = *(ptr++); result |= (b & 127) << 21; if (b < 128) goto done; - if (ptr >= limit) return NULL; - b = *(ptr++); result |= (b & 127) << 28; if (b < 16) goto done; - return NULL; // Value is too long to be a varint32 - done: - *OUTPUT = result; - return reinterpret_cast(ptr); -} - -inline char* Varint::Encode32(char* sptr, uint32_t v) { - // Operate on characters as unsigneds - uint8_t* ptr = reinterpret_cast(sptr); - static const uint8_t B = 128; - if (v < (1 << 7)) { - *(ptr++) = static_cast(v); - } else if (v < (1 << 14)) { - *(ptr++) = static_cast(v | B); - *(ptr++) = static_cast(v >> 7); - } else if (v < (1 << 21)) { - *(ptr++) = static_cast(v | B); - *(ptr++) = static_cast((v >> 7) | B); - *(ptr++) = static_cast(v >> 14); - } else if (v < (1 << 28)) { - *(ptr++) = static_cast(v | B); - *(ptr++) = static_cast((v >> 7) | B); - *(ptr++) = static_cast((v >> 14) | B); - *(ptr++) = static_cast(v >> 21); - } else { - *(ptr++) = static_cast(v | B); - *(ptr++) = static_cast((v>>7) | B); - *(ptr++) = static_cast((v>>14) | B); - *(ptr++) = static_cast((v>>21) | B); - *(ptr++) = static_cast(v >> 28); - } - return reinterpret_cast(ptr); -} - -// If you know the internal layout of the std::string in use, you can -// replace this function with one that resizes the string without -// filling the new space with zeros (if applicable) -- -// it will be non-portable but faster. -inline void STLStringResizeUninitialized(std::string* s, size_t new_size) { - s->resize(new_size); -} - -// Return a mutable char* pointing to a string's internal buffer, -// which may not be null-terminated. Writing through this pointer will -// modify the string. -// -// string_as_array(&str)[i] is valid for 0 <= i < str.size() until the -// next call to a string method that invalidates iterators. -// -// As of 2006-04, there is no standard-blessed way of getting a -// mutable reference to a string's internal buffer. However, issue 530 -// (http://www.open-std.org/JTC1/SC22/WG21/docs/lwg-defects.html#530) -// proposes this as the method. It will officially be part of the standard -// for C++0x. This should already work on all current implementations. -inline char* string_as_array(std::string* str) { - return str->empty() ? NULL : &*str->begin(); -} - -} // namespace snappy - -#endif // THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_ diff --git a/contrib/libs/snappy/snappy-stubs-public.h b/contrib/libs/snappy/snappy-stubs-public.h deleted file mode 100644 index 13a00c67dd8..00000000000 --- a/contrib/libs/snappy/snappy-stubs-public.h +++ /dev/null @@ -1,64 +0,0 @@ -// Copyright 2011 Google Inc. All Rights Reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// -// Various type stubs for the open-source version of Snappy. -// -// This file cannot include config.h, as it is included from snappy.h, -// which is a public header. Instead, snappy-stubs-public.h is generated by -// from snappy-stubs-public.h.in at configure time. - -#ifndef THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_STUBS_PUBLIC_H_ -#define THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_STUBS_PUBLIC_H_ - -#include -#include "config.h" - -#if defined(HAVE_SYS_UIO_H) -#include -#endif // HAVE_SYS_UIO_H - -#define SNAPPY_MAJOR 1 -#define SNAPPY_MINOR 1 -#define SNAPPY_PATCHLEVEL 10 -#define SNAPPY_VERSION \ - ((SNAPPY_MAJOR << 16) | (SNAPPY_MINOR << 8) | SNAPPY_PATCHLEVEL) - -namespace snappy { - -#if !defined(HAVE_SYS_UIO_H) -// Windows does not have an iovec type, yet the concept is universally useful. -// It is simple to define it ourselves, so we put it inside our own namespace. -struct iovec { - void* iov_base; - size_t iov_len; -}; -#endif // !HAVE_SYS_UIO_H - -} // namespace snappy - -#endif // THIRD_PARTY_SNAPPY_OPENSOURCE_SNAPPY_STUBS_PUBLIC_H_ diff --git a/contrib/libs/snappy/snappy.cc b/contrib/libs/snappy/snappy.cc deleted file mode 100644 index 3467c26d2b7..00000000000 --- a/contrib/libs/snappy/snappy.cc +++ /dev/null @@ -1,2428 +0,0 @@ -// Copyright 2005 Google Inc. All Rights Reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -#include "snappy-internal.h" -#include "snappy-sinksource.h" -#include "snappy.h" -#if !defined(SNAPPY_HAVE_BMI2) -// __BMI2__ is defined by GCC and Clang. Visual Studio doesn't target BMI2 -// specifically, but it does define __AVX2__ when AVX2 support is available. -// Fortunately, AVX2 was introduced in Haswell, just like BMI2. -// -// BMI2 is not defined as a subset of AVX2 (unlike SSSE3 and AVX above). So, -// GCC and Clang can build code with AVX2 enabled but BMI2 disabled, in which -// case issuing BMI2 instructions results in a compiler error. -#if defined(__BMI2__) || (defined(_MSC_VER) && defined(__AVX2__)) -#define SNAPPY_HAVE_BMI2 1 -#else -#define SNAPPY_HAVE_BMI2 0 -#endif -#endif // !defined(SNAPPY_HAVE_BMI2) - -#if !defined(SNAPPY_HAVE_X86_CRC32) -#if defined(__SSE4_2__) -#define SNAPPY_HAVE_X86_CRC32 1 -#else -#define SNAPPY_HAVE_X86_CRC32 0 -#endif -#endif // !defined(SNAPPY_HAVE_X86_CRC32) - -#if !defined(SNAPPY_HAVE_NEON_CRC32) -#if SNAPPY_HAVE_NEON && defined(__ARM_FEATURE_CRC32) -#define SNAPPY_HAVE_NEON_CRC32 1 -#else -#define SNAPPY_HAVE_NEON_CRC32 0 -#endif -#endif // !defined(SNAPPY_HAVE_NEON_CRC32) - -#if SNAPPY_HAVE_BMI2 || SNAPPY_HAVE_X86_CRC32 || (defined(__x86_64__) && defined(__AVX__)) -// Please do not replace with . or with headers that assume more -// advanced SSE versions without checking with all the OWNERS. -#include -#elif SNAPPY_HAVE_NEON_CRC32 -#include -#endif - -#if defined(__GNUC__) -#define SNAPPY_PREFETCH(ptr) __builtin_prefetch(ptr, 0, 3) -#else -#define SNAPPY_PREFETCH(ptr) (void)(ptr) -#endif - -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -namespace snappy { - -namespace { - -// The amount of slop bytes writers are using for unconditional copies. -constexpr int kSlopBytes = 64; - -using internal::char_table; -using internal::COPY_1_BYTE_OFFSET; -using internal::COPY_2_BYTE_OFFSET; -using internal::COPY_4_BYTE_OFFSET; -using internal::kMaximumTagLength; -using internal::LITERAL; -#if SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE -using internal::V128; -using internal::V128_Load; -using internal::V128_LoadU; -using internal::V128_Shuffle; -using internal::V128_StoreU; -using internal::V128_DupChar; -#endif - -// We translate the information encoded in a tag through a lookup table to a -// format that requires fewer instructions to decode. Effectively we store -// the length minus the tag part of the offset. The lowest significant byte -// thus stores the length. While total length - offset is given by -// entry - ExtractOffset(type). The nice thing is that the subtraction -// immediately sets the flags for the necessary check that offset >= length. -// This folds the cmp with sub. We engineer the long literals and copy-4 to -// always fail this check, so their presence doesn't affect the fast path. -// To prevent literals from triggering the guard against offset < length (offset -// does not apply to literals) the table is giving them a spurious offset of -// 256. -inline constexpr int16_t MakeEntry(int16_t len, int16_t offset) { - return len - (offset << 8); -} - -inline constexpr int16_t LengthMinusOffset(int data, int type) { - return type == 3 ? 0xFF // copy-4 (or type == 3) - : type == 2 ? MakeEntry(data + 1, 0) // copy-2 - : type == 1 ? MakeEntry((data & 7) + 4, data >> 3) // copy-1 - : data < 60 ? MakeEntry(data + 1, 1) // note spurious offset. - : 0xFF; // long literal -} - -inline constexpr int16_t LengthMinusOffset(uint8_t tag) { - return LengthMinusOffset(tag >> 2, tag & 3); -} - -template -struct index_sequence {}; - -template -struct make_index_sequence : make_index_sequence {}; - -template -struct make_index_sequence<0, Is...> : index_sequence {}; - -template -constexpr std::array MakeTable(index_sequence) { - return std::array{LengthMinusOffset(seq)...}; -} - -alignas(64) const std::array kLengthMinusOffset = - MakeTable(make_index_sequence<256>{}); - -// Given a table of uint16_t whose size is mask / 2 + 1, return a pointer to the -// relevant entry, if any, for the given bytes. Any hash function will do, -// but a good hash function reduces the number of collisions and thus yields -// better compression for compressible input. -// -// REQUIRES: mask is 2 * (table_size - 1), and table_size is a power of two. -inline uint16_t* TableEntry(uint16_t* table, uint32_t bytes, uint32_t mask) { - // Our choice is quicker-and-dirtier than the typical hash function; - // empirically, that seems beneficial. The upper bits of kMagic * bytes are a - // higher-quality hash than the lower bits, so when using kMagic * bytes we - // also shift right to get a higher-quality end result. There's no similar - // issue with a CRC because all of the output bits of a CRC are equally good - // "hashes." So, a CPU instruction for CRC, if available, tends to be a good - // choice. -#if SNAPPY_HAVE_NEON_CRC32 - // We use mask as the second arg to the CRC function, as it's about to - // be used anyway; it'd be equally correct to use 0 or some constant. - // Mathematically, _mm_crc32_u32 (or similar) is a function of the - // xor of its arguments. - const uint32_t hash = __crc32cw(bytes, mask); -#elif SNAPPY_HAVE_X86_CRC32 - const uint32_t hash = _mm_crc32_u32(bytes, mask); -#else - constexpr uint32_t kMagic = 0x1e35a7bd; - const uint32_t hash = (kMagic * bytes) >> (31 - kMaxHashTableBits); -#endif - return reinterpret_cast(reinterpret_cast(table) + - (hash & mask)); -} - -} // namespace - -size_t MaxCompressedLength(size_t source_bytes) { - // Compressed data can be defined as: - // compressed := item* literal* - // item := literal* copy - // - // The trailing literal sequence has a space blowup of at most 62/60 - // since a literal of length 60 needs one tag byte + one extra byte - // for length information. - // - // Item blowup is trickier to measure. Suppose the "copy" op copies - // 4 bytes of data. Because of a special check in the encoding code, - // we produce a 4-byte copy only if the offset is < 65536. Therefore - // the copy op takes 3 bytes to encode, and this type of item leads - // to at most the 62/60 blowup for representing literals. - // - // Suppose the "copy" op copies 5 bytes of data. If the offset is big - // enough, it will take 5 bytes to encode the copy op. Therefore the - // worst case here is a one-byte literal followed by a five-byte copy. - // I.e., 6 bytes of input turn into 7 bytes of "compressed" data. - // - // This last factor dominates the blowup, so the final estimate is: - return 32 + source_bytes + source_bytes / 6; -} - -namespace { - -void UnalignedCopy64(const void* src, void* dst) { - char tmp[8]; - std::memcpy(tmp, src, 8); - std::memcpy(dst, tmp, 8); -} - -void UnalignedCopy128(const void* src, void* dst) { - // std::memcpy() gets vectorized when the appropriate compiler options are - // used. For example, x86 compilers targeting SSE2+ will optimize to an SSE2 - // load and store. - char tmp[16]; - std::memcpy(tmp, src, 16); - std::memcpy(dst, tmp, 16); -} - -template -inline void ConditionalUnalignedCopy128(const char* src, char* dst) { - if (use_16bytes_chunk) { - UnalignedCopy128(src, dst); - } else { - UnalignedCopy64(src, dst); - UnalignedCopy64(src + 8, dst + 8); - } -} - -// Copy [src, src+(op_limit-op)) to [op, (op_limit-op)) a byte at a time. Used -// for handling COPY operations where the input and output regions may overlap. -// For example, suppose: -// src == "ab" -// op == src + 2 -// op_limit == op + 20 -// After IncrementalCopySlow(src, op, op_limit), the result will have eleven -// copies of "ab" -// ababababababababababab -// Note that this does not match the semantics of either std::memcpy() or -// std::memmove(). -inline char* IncrementalCopySlow(const char* src, char* op, - char* const op_limit) { - // TODO: Remove pragma when LLVM is aware this - // function is only called in cold regions and when cold regions don't get - // vectorized or unrolled. -#ifdef __clang__ -#pragma clang loop unroll(disable) -#endif - while (op < op_limit) { - *op++ = *src++; - } - return op_limit; -} - -#if SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE - -// Computes the bytes for shuffle control mask (please read comments on -// 'pattern_generation_masks' as well) for the given index_offset and -// pattern_size. For example, when the 'offset' is 6, it will generate a -// repeating pattern of size 6. So, the first 16 byte indexes will correspond to -// the pattern-bytes {0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3} and the -// next 16 byte indexes will correspond to the pattern-bytes {4, 5, 0, 1, 2, 3, -// 4, 5, 0, 1, 2, 3, 4, 5, 0, 1}. These byte index sequences are generated by -// calling MakePatternMaskBytes(0, 6, index_sequence<16>()) and -// MakePatternMaskBytes(16, 6, index_sequence<16>()) respectively. -template -inline constexpr std::array MakePatternMaskBytes( - int index_offset, int pattern_size, index_sequence) { - return {static_cast((index_offset + indexes) % pattern_size)...}; -} - -// Computes the shuffle control mask bytes array for given pattern-sizes and -// returns an array. -template -inline constexpr std::array, - sizeof...(pattern_sizes_minus_one)> -MakePatternMaskBytesTable(int index_offset, - index_sequence) { - return { - MakePatternMaskBytes(index_offset, pattern_sizes_minus_one + 1, - make_index_sequence())...}; -} - -// This is an array of shuffle control masks that can be used as the source -// operand for PSHUFB to permute the contents of the destination XMM register -// into a repeating byte pattern. -alignas(16) constexpr std::array, - 16> pattern_generation_masks = - MakePatternMaskBytesTable( - /*index_offset=*/0, - /*pattern_sizes_minus_one=*/make_index_sequence<16>()); - -// Similar to 'pattern_generation_masks', this table is used to "rotate" the -// pattern so that we can copy the *next 16 bytes* consistent with the pattern. -// Basically, pattern_reshuffle_masks is a continuation of -// pattern_generation_masks. It follows that, pattern_reshuffle_masks is same as -// pattern_generation_masks for offsets 1, 2, 4, 8 and 16. -alignas(16) constexpr std::array, - 16> pattern_reshuffle_masks = - MakePatternMaskBytesTable( - /*index_offset=*/16, - /*pattern_sizes_minus_one=*/make_index_sequence<16>()); - -SNAPPY_ATTRIBUTE_ALWAYS_INLINE -static inline V128 LoadPattern(const char* src, const size_t pattern_size) { - V128 generation_mask = V128_Load(reinterpret_cast( - pattern_generation_masks[pattern_size - 1].data())); - // Uninitialized bytes are masked out by the shuffle mask. - // TODO: remove annotation and macro defs once MSan is fixed. - SNAPPY_ANNOTATE_MEMORY_IS_INITIALIZED(src + pattern_size, 16 - pattern_size); - return V128_Shuffle(V128_LoadU(reinterpret_cast(src)), - generation_mask); -} - -SNAPPY_ATTRIBUTE_ALWAYS_INLINE -static inline std::pair -LoadPatternAndReshuffleMask(const char* src, const size_t pattern_size) { - V128 pattern = LoadPattern(src, pattern_size); - - // This mask will generate the next 16 bytes in-place. Doing so enables us to - // write data by at most 4 V128_StoreU. - // - // For example, suppose pattern is: abcdefabcdefabcd - // Shuffling with this mask will generate: efabcdefabcdefab - // Shuffling again will generate: cdefabcdefabcdef - V128 reshuffle_mask = V128_Load(reinterpret_cast( - pattern_reshuffle_masks[pattern_size - 1].data())); - return {pattern, reshuffle_mask}; -} - -#endif // SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE - -// Fallback for when we need to copy while extending the pattern, for example -// copying 10 bytes from 3 positions back abc -> abcabcabcabca. -// -// REQUIRES: [dst - offset, dst + 64) is a valid address range. -SNAPPY_ATTRIBUTE_ALWAYS_INLINE -static inline bool Copy64BytesWithPatternExtension(char* dst, size_t offset) { -#if SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE - if (SNAPPY_PREDICT_TRUE(offset <= 16)) { - switch (offset) { - case 0: - return false; - case 1: { - // TODO: Ideally we should memset, move back once the - // codegen issues are fixed. - V128 pattern = V128_DupChar(dst[-1]); - for (int i = 0; i < 4; i++) { - V128_StoreU(reinterpret_cast(dst + 16 * i), pattern); - } - return true; - } - case 2: - case 4: - case 8: - case 16: { - V128 pattern = LoadPattern(dst - offset, offset); - for (int i = 0; i < 4; i++) { - V128_StoreU(reinterpret_cast(dst + 16 * i), pattern); - } - return true; - } - default: { - auto pattern_and_reshuffle_mask = - LoadPatternAndReshuffleMask(dst - offset, offset); - V128 pattern = pattern_and_reshuffle_mask.first; - V128 reshuffle_mask = pattern_and_reshuffle_mask.second; - for (int i = 0; i < 4; i++) { - V128_StoreU(reinterpret_cast(dst + 16 * i), pattern); - pattern = V128_Shuffle(pattern, reshuffle_mask); - } - return true; - } - } - } -#else - if (SNAPPY_PREDICT_TRUE(offset < 16)) { - if (SNAPPY_PREDICT_FALSE(offset == 0)) return false; - // Extend the pattern to the first 16 bytes. - // The simpler formulation of `dst[i - offset]` induces undefined behavior. - for (int i = 0; i < 16; i++) dst[i] = (dst - offset)[i]; - // Find a multiple of pattern >= 16. - static std::array pattern_sizes = []() { - std::array res; - for (int i = 1; i < 16; i++) res[i] = (16 / i + 1) * i; - return res; - }(); - offset = pattern_sizes[offset]; - for (int i = 1; i < 4; i++) { - std::memcpy(dst + i * 16, dst + i * 16 - offset, 16); - } - return true; - } -#endif // SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE - - // Very rare. - for (int i = 0; i < 4; i++) { - std::memcpy(dst + i * 16, dst + i * 16 - offset, 16); - } - return true; -} - -// Copy [src, src+(op_limit-op)) to [op, op_limit) but faster than -// IncrementalCopySlow. buf_limit is the address past the end of the writable -// region of the buffer. -inline char* IncrementalCopy(const char* src, char* op, char* const op_limit, - char* const buf_limit) { -#if SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE - constexpr int big_pattern_size_lower_bound = 16; -#else - constexpr int big_pattern_size_lower_bound = 8; -#endif - - // Terminology: - // - // slop = buf_limit - op - // pat = op - src - // len = op_limit - op - assert(src < op); - assert(op < op_limit); - assert(op_limit <= buf_limit); - // NOTE: The copy tags use 3 or 6 bits to store the copy length, so len <= 64. - assert(op_limit - op <= 64); - // NOTE: In practice the compressor always emits len >= 4, so it is ok to - // assume that to optimize this function, but this is not guaranteed by the - // compression format, so we have to also handle len < 4 in case the input - // does not satisfy these conditions. - - size_t pattern_size = op - src; - // The cases are split into different branches to allow the branch predictor, - // FDO, and static prediction hints to work better. For each input we list the - // ratio of invocations that match each condition. - // - // input slop < 16 pat < 8 len > 16 - // ------------------------------------------ - // html|html4|cp 0% 1.01% 27.73% - // urls 0% 0.88% 14.79% - // jpg 0% 64.29% 7.14% - // pdf 0% 2.56% 58.06% - // txt[1-4] 0% 0.23% 0.97% - // pb 0% 0.96% 13.88% - // bin 0.01% 22.27% 41.17% - // - // It is very rare that we don't have enough slop for doing block copies. It - // is also rare that we need to expand a pattern. Small patterns are common - // for incompressible formats and for those we are plenty fast already. - // Lengths are normally not greater than 16 but they vary depending on the - // input. In general if we always predict len <= 16 it would be an ok - // prediction. - // - // In order to be fast we want a pattern >= 16 bytes (or 8 bytes in non-SSE) - // and an unrolled loop copying 1x 16 bytes (or 2x 8 bytes in non-SSE) at a - // time. - - // Handle the uncommon case where pattern is less than 16 (or 8 in non-SSE) - // bytes. - if (pattern_size < big_pattern_size_lower_bound) { -#if SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE - // Load the first eight bytes into an 128-bit XMM register, then use PSHUFB - // to permute the register's contents in-place into a repeating sequence of - // the first "pattern_size" bytes. - // For example, suppose: - // src == "abc" - // op == op + 3 - // After V128_Shuffle(), "pattern" will have five copies of "abc" - // followed by one byte of slop: abcabcabcabcabca. - // - // The non-SSE fallback implementation suffers from store-forwarding stalls - // because its loads and stores partly overlap. By expanding the pattern - // in-place, we avoid the penalty. - - // Typically, the op_limit is the gating factor so try to simplify the loop - // based on that. - if (SNAPPY_PREDICT_TRUE(op_limit <= buf_limit - 15)) { - auto pattern_and_reshuffle_mask = - LoadPatternAndReshuffleMask(src, pattern_size); - V128 pattern = pattern_and_reshuffle_mask.first; - V128 reshuffle_mask = pattern_and_reshuffle_mask.second; - - // There is at least one, and at most four 16-byte blocks. Writing four - // conditionals instead of a loop allows FDO to layout the code with - // respect to the actual probabilities of each length. - // TODO: Replace with loop with trip count hint. - V128_StoreU(reinterpret_cast(op), pattern); - - if (op + 16 < op_limit) { - pattern = V128_Shuffle(pattern, reshuffle_mask); - V128_StoreU(reinterpret_cast(op + 16), pattern); - } - if (op + 32 < op_limit) { - pattern = V128_Shuffle(pattern, reshuffle_mask); - V128_StoreU(reinterpret_cast(op + 32), pattern); - } - if (op + 48 < op_limit) { - pattern = V128_Shuffle(pattern, reshuffle_mask); - V128_StoreU(reinterpret_cast(op + 48), pattern); - } - return op_limit; - } - char* const op_end = buf_limit - 15; - if (SNAPPY_PREDICT_TRUE(op < op_end)) { - auto pattern_and_reshuffle_mask = - LoadPatternAndReshuffleMask(src, pattern_size); - V128 pattern = pattern_and_reshuffle_mask.first; - V128 reshuffle_mask = pattern_and_reshuffle_mask.second; - - // This code path is relatively cold however so we save code size - // by avoiding unrolling and vectorizing. - // - // TODO: Remove pragma when when cold regions don't get - // vectorized or unrolled. -#ifdef __clang__ -#pragma clang loop unroll(disable) -#endif - do { - V128_StoreU(reinterpret_cast(op), pattern); - pattern = V128_Shuffle(pattern, reshuffle_mask); - op += 16; - } while (SNAPPY_PREDICT_TRUE(op < op_end)); - } - return IncrementalCopySlow(op - pattern_size, op, op_limit); -#else // !SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE - // If plenty of buffer space remains, expand the pattern to at least 8 - // bytes. The way the following loop is written, we need 8 bytes of buffer - // space if pattern_size >= 4, 11 bytes if pattern_size is 1 or 3, and 10 - // bytes if pattern_size is 2. Precisely encoding that is probably not - // worthwhile; instead, invoke the slow path if we cannot write 11 bytes - // (because 11 are required in the worst case). - if (SNAPPY_PREDICT_TRUE(op <= buf_limit - 11)) { - while (pattern_size < 8) { - UnalignedCopy64(src, op); - op += pattern_size; - pattern_size *= 2; - } - if (SNAPPY_PREDICT_TRUE(op >= op_limit)) return op_limit; - } else { - return IncrementalCopySlow(src, op, op_limit); - } -#endif // SNAPPY_HAVE_VECTOR_BYTE_SHUFFLE - } - assert(pattern_size >= big_pattern_size_lower_bound); - constexpr bool use_16bytes_chunk = big_pattern_size_lower_bound == 16; - - // Copy 1x 16 bytes (or 2x 8 bytes in non-SSE) at a time. Because op - src can - // be < 16 in non-SSE, a single UnalignedCopy128 might overwrite data in op. - // UnalignedCopy64 is safe because expanding the pattern to at least 8 bytes - // guarantees that op - src >= 8. - // - // Typically, the op_limit is the gating factor so try to simplify the loop - // based on that. - if (SNAPPY_PREDICT_TRUE(op_limit <= buf_limit - 15)) { - // There is at least one, and at most four 16-byte blocks. Writing four - // conditionals instead of a loop allows FDO to layout the code with respect - // to the actual probabilities of each length. - // TODO: Replace with loop with trip count hint. - ConditionalUnalignedCopy128(src, op); - if (op + 16 < op_limit) { - ConditionalUnalignedCopy128(src + 16, op + 16); - } - if (op + 32 < op_limit) { - ConditionalUnalignedCopy128(src + 32, op + 32); - } - if (op + 48 < op_limit) { - ConditionalUnalignedCopy128(src + 48, op + 48); - } - return op_limit; - } - - // Fall back to doing as much as we can with the available slop in the - // buffer. This code path is relatively cold however so we save code size by - // avoiding unrolling and vectorizing. - // - // TODO: Remove pragma when when cold regions don't get vectorized - // or unrolled. -#ifdef __clang__ -#pragma clang loop unroll(disable) -#endif - for (char* op_end = buf_limit - 16; op < op_end; op += 16, src += 16) { - ConditionalUnalignedCopy128(src, op); - } - if (op >= op_limit) return op_limit; - - // We only take this branch if we didn't have enough slop and we can do a - // single 8 byte copy. - if (SNAPPY_PREDICT_FALSE(op <= buf_limit - 8)) { - UnalignedCopy64(src, op); - src += 8; - op += 8; - } - return IncrementalCopySlow(src, op, op_limit); -} - -} // namespace - -template -static inline char* EmitLiteral(char* op, const char* literal, int len) { - // The vast majority of copies are below 16 bytes, for which a - // call to std::memcpy() is overkill. This fast path can sometimes - // copy up to 15 bytes too much, but that is okay in the - // main loop, since we have a bit to go on for both sides: - // - // - The input will always have kInputMarginBytes = 15 extra - // available bytes, as long as we're in the main loop, and - // if not, allow_fast_path = false. - // - The output will always have 32 spare bytes (see - // MaxCompressedLength). - assert(len > 0); // Zero-length literals are disallowed - int n = len - 1; - if (allow_fast_path && len <= 16) { - // Fits in tag byte - *op++ = LITERAL | (n << 2); - - UnalignedCopy128(literal, op); - return op + len; - } - - if (n < 60) { - // Fits in tag byte - *op++ = LITERAL | (n << 2); - } else { - int count = (Bits::Log2Floor(n) >> 3) + 1; - assert(count >= 1); - assert(count <= 4); - *op++ = LITERAL | ((59 + count) << 2); - // Encode in upcoming bytes. - // Write 4 bytes, though we may care about only 1 of them. The output buffer - // is guaranteed to have at least 3 more spaces left as 'len >= 61' holds - // here and there is a std::memcpy() of size 'len' below. - LittleEndian::Store32(op, n); - op += count; - } - // When allow_fast_path is true, we can overwrite up to 16 bytes. - if (allow_fast_path) { - char* destination = op; - const char* source = literal; - const char* end = destination + len; - do { - std::memcpy(destination, source, 16); - destination += 16; - source += 16; - } while (destination < end); - } else { - std::memcpy(op, literal, len); - } - return op + len; -} - -template -static inline char* EmitCopyAtMost64(char* op, size_t offset, size_t len) { - assert(len <= 64); - assert(len >= 4); - assert(offset < 65536); - assert(len_less_than_12 == (len < 12)); - - if (len_less_than_12) { - uint32_t u = (len << 2) + (offset << 8); - uint32_t copy1 = COPY_1_BYTE_OFFSET - (4 << 2) + ((offset >> 3) & 0xe0); - uint32_t copy2 = COPY_2_BYTE_OFFSET - (1 << 2); - // It turns out that offset < 2048 is a difficult to predict branch. - // `perf record` shows this is the highest percentage of branch misses in - // benchmarks. This code produces branch free code, the data dependency - // chain that bottlenecks the throughput is so long that a few extra - // instructions are completely free (IPC << 6 because of data deps). - u += offset < 2048 ? copy1 : copy2; - LittleEndian::Store32(op, u); - op += offset < 2048 ? 2 : 3; - } else { - // Write 4 bytes, though we only care about 3 of them. The output buffer - // is required to have some slack, so the extra byte won't overrun it. - uint32_t u = COPY_2_BYTE_OFFSET + ((len - 1) << 2) + (offset << 8); - LittleEndian::Store32(op, u); - op += 3; - } - return op; -} - -template -static inline char* EmitCopy(char* op, size_t offset, size_t len) { - assert(len_less_than_12 == (len < 12)); - if (len_less_than_12) { - return EmitCopyAtMost64(op, offset, len); - } else { - // A special case for len <= 64 might help, but so far measurements suggest - // it's in the noise. - - // Emit 64 byte copies but make sure to keep at least four bytes reserved. - while (SNAPPY_PREDICT_FALSE(len >= 68)) { - op = EmitCopyAtMost64(op, offset, 64); - len -= 64; - } - - // One or two copies will now finish the job. - if (len > 64) { - op = EmitCopyAtMost64(op, offset, 60); - len -= 60; - } - - // Emit remainder. - if (len < 12) { - op = EmitCopyAtMost64(op, offset, len); - } else { - op = EmitCopyAtMost64(op, offset, len); - } - return op; - } -} - -bool GetUncompressedLength(const char* start, size_t n, size_t* result) { - uint32_t v = 0; - const char* limit = start + n; - if (Varint::Parse32WithLimit(start, limit, &v) != NULL) { - *result = v; - return true; - } else { - return false; - } -} - -namespace { -uint32_t CalculateTableSize(uint32_t input_size) { - static_assert( - kMaxHashTableSize >= kMinHashTableSize, - "kMaxHashTableSize should be greater or equal to kMinHashTableSize."); - if (input_size > kMaxHashTableSize) { - return kMaxHashTableSize; - } - if (input_size < kMinHashTableSize) { - return kMinHashTableSize; - } - // This is equivalent to Log2Ceiling(input_size), assuming input_size > 1. - // 2 << Log2Floor(x - 1) is equivalent to 1 << (1 + Log2Floor(x - 1)). - return 2u << Bits::Log2Floor(input_size - 1); -} -} // namespace - -namespace internal { -WorkingMemory::WorkingMemory(size_t input_size) { - const size_t max_fragment_size = std::min(input_size, kBlockSize); - const size_t table_size = CalculateTableSize(max_fragment_size); - size_ = table_size * sizeof(*table_) + max_fragment_size + - MaxCompressedLength(max_fragment_size); - mem_ = std::allocator().allocate(size_); - table_ = reinterpret_cast(mem_); - input_ = mem_ + table_size * sizeof(*table_); - output_ = input_ + max_fragment_size; -} - -WorkingMemory::~WorkingMemory() { - std::allocator().deallocate(mem_, size_); -} - -uint16_t* WorkingMemory::GetHashTable(size_t fragment_size, - int* table_size) const { - const size_t htsize = CalculateTableSize(fragment_size); - memset(table_, 0, htsize * sizeof(*table_)); - *table_size = htsize; - return table_; -} -} // end namespace internal - -// Flat array compression that does not emit the "uncompressed length" -// prefix. Compresses "input" string to the "*op" buffer. -// -// REQUIRES: "input" is at most "kBlockSize" bytes long. -// REQUIRES: "op" points to an array of memory that is at least -// "MaxCompressedLength(input.size())" in size. -// REQUIRES: All elements in "table[0..table_size-1]" are initialized to zero. -// REQUIRES: "table_size" is a power of two -// -// Returns an "end" pointer into "op" buffer. -// "end - op" is the compressed size of "input". -namespace internal { -char* CompressFragment(const char* input, size_t input_size, char* op, - uint16_t* table, const int table_size) { - // "ip" is the input pointer, and "op" is the output pointer. - const char* ip = input; - assert(input_size <= kBlockSize); - assert((table_size & (table_size - 1)) == 0); // table must be power of two - const uint32_t mask = 2 * (table_size - 1); - const char* ip_end = input + input_size; - const char* base_ip = ip; - - const size_t kInputMarginBytes = 15; - if (SNAPPY_PREDICT_TRUE(input_size >= kInputMarginBytes)) { - const char* ip_limit = input + input_size - kInputMarginBytes; - - for (uint32_t preload = LittleEndian::Load32(ip + 1);;) { - // Bytes in [next_emit, ip) will be emitted as literal bytes. Or - // [next_emit, ip_end) after the main loop. - const char* next_emit = ip++; - uint64_t data = LittleEndian::Load64(ip); - // The body of this loop calls EmitLiteral once and then EmitCopy one or - // more times. (The exception is that when we're close to exhausting - // the input we goto emit_remainder.) - // - // In the first iteration of this loop we're just starting, so - // there's nothing to copy, so calling EmitLiteral once is - // necessary. And we only start a new iteration when the - // current iteration has determined that a call to EmitLiteral will - // precede the next call to EmitCopy (if any). - // - // Step 1: Scan forward in the input looking for a 4-byte-long match. - // If we get close to exhausting the input then goto emit_remainder. - // - // Heuristic match skipping: If 32 bytes are scanned with no matches - // found, start looking only at every other byte. If 32 more bytes are - // scanned (or skipped), look at every third byte, etc.. When a match is - // found, immediately go back to looking at every byte. This is a small - // loss (~5% performance, ~0.1% density) for compressible data due to more - // bookkeeping, but for non-compressible data (such as JPEG) it's a huge - // win since the compressor quickly "realizes" the data is incompressible - // and doesn't bother looking for matches everywhere. - // - // The "skip" variable keeps track of how many bytes there are since the - // last match; dividing it by 32 (ie. right-shifting by five) gives the - // number of bytes to move ahead for each iteration. - uint32_t skip = 32; - - const char* candidate; - if (ip_limit - ip >= 16) { - auto delta = ip - base_ip; - for (int j = 0; j < 4; ++j) { - for (int k = 0; k < 4; ++k) { - int i = 4 * j + k; - // These for-loops are meant to be unrolled. So we can freely - // special case the first iteration to use the value already - // loaded in preload. - uint32_t dword = i == 0 ? preload : static_cast(data); - assert(dword == LittleEndian::Load32(ip + i)); - uint16_t* table_entry = TableEntry(table, dword, mask); - candidate = base_ip + *table_entry; - assert(candidate >= base_ip); - assert(candidate < ip + i); - *table_entry = delta + i; - if (SNAPPY_PREDICT_FALSE(LittleEndian::Load32(candidate) == dword)) { - *op = LITERAL | (i << 2); - UnalignedCopy128(next_emit, op + 1); - ip += i; - op = op + i + 2; - goto emit_match; - } - data >>= 8; - } - data = LittleEndian::Load64(ip + 4 * j + 4); - } - ip += 16; - skip += 16; - } - while (true) { - assert(static_cast(data) == LittleEndian::Load32(ip)); - uint16_t* table_entry = TableEntry(table, data, mask); - uint32_t bytes_between_hash_lookups = skip >> 5; - skip += bytes_between_hash_lookups; - const char* next_ip = ip + bytes_between_hash_lookups; - if (SNAPPY_PREDICT_FALSE(next_ip > ip_limit)) { - ip = next_emit; - goto emit_remainder; - } - candidate = base_ip + *table_entry; - assert(candidate >= base_ip); - assert(candidate < ip); - - *table_entry = ip - base_ip; - if (SNAPPY_PREDICT_FALSE(static_cast(data) == - LittleEndian::Load32(candidate))) { - break; - } - data = LittleEndian::Load32(next_ip); - ip = next_ip; - } - - // Step 2: A 4-byte match has been found. We'll later see if more - // than 4 bytes match. But, prior to the match, input - // bytes [next_emit, ip) are unmatched. Emit them as "literal bytes." - assert(next_emit + 16 <= ip_end); - op = EmitLiteral(op, next_emit, ip - next_emit); - - // Step 3: Call EmitCopy, and then see if another EmitCopy could - // be our next move. Repeat until we find no match for the - // input immediately after what was consumed by the last EmitCopy call. - // - // If we exit this loop normally then we need to call EmitLiteral next, - // though we don't yet know how big the literal will be. We handle that - // by proceeding to the next iteration of the main loop. We also can exit - // this loop via goto if we get close to exhausting the input. - emit_match: - do { - // We have a 4-byte match at ip, and no need to emit any - // "literal bytes" prior to ip. - const char* base = ip; - std::pair p = - FindMatchLength(candidate + 4, ip + 4, ip_end, &data); - size_t matched = 4 + p.first; - ip += matched; - size_t offset = base - candidate; - assert(0 == memcmp(base, candidate, matched)); - if (p.second) { - op = EmitCopy(op, offset, matched); - } else { - op = EmitCopy(op, offset, matched); - } - if (SNAPPY_PREDICT_FALSE(ip >= ip_limit)) { - goto emit_remainder; - } - // Expect 5 bytes to match - assert((data & 0xFFFFFFFFFF) == - (LittleEndian::Load64(ip) & 0xFFFFFFFFFF)); - // We are now looking for a 4-byte match again. We read - // table[Hash(ip, mask)] for that. To improve compression, - // we also update table[Hash(ip - 1, mask)] and table[Hash(ip, mask)]. - *TableEntry(table, LittleEndian::Load32(ip - 1), mask) = - ip - base_ip - 1; - uint16_t* table_entry = TableEntry(table, data, mask); - candidate = base_ip + *table_entry; - *table_entry = ip - base_ip; - // Measurements on the benchmarks have shown the following probabilities - // for the loop to exit (ie. avg. number of iterations is reciprocal). - // BM_Flat/6 txt1 p = 0.3-0.4 - // BM_Flat/7 txt2 p = 0.35 - // BM_Flat/8 txt3 p = 0.3-0.4 - // BM_Flat/9 txt3 p = 0.34-0.4 - // BM_Flat/10 pb p = 0.4 - // BM_Flat/11 gaviota p = 0.1 - // BM_Flat/12 cp p = 0.5 - // BM_Flat/13 c p = 0.3 - } while (static_cast(data) == LittleEndian::Load32(candidate)); - // Because the least significant 5 bytes matched, we can utilize data - // for the next iteration. - preload = data >> 8; - } - } - -emit_remainder: - // Emit the remaining bytes as a literal - if (ip < ip_end) { - op = EmitLiteral(op, ip, ip_end - ip); - } - - return op; -} -} // end namespace internal - -// Called back at avery compression call to trace parameters and sizes. -static inline void Report(const char *algorithm, size_t compressed_size, - size_t uncompressed_size) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)algorithm; - (void)compressed_size; - (void)uncompressed_size; -} - -// Signature of output types needed by decompression code. -// The decompression code is templatized on a type that obeys this -// signature so that we do not pay virtual function call overhead in -// the middle of a tight decompression loop. -// -// class DecompressionWriter { -// public: -// // Called before decompression -// void SetExpectedLength(size_t length); -// -// // For performance a writer may choose to donate the cursor variable to the -// // decompression function. The decompression will inject it in all its -// // function calls to the writer. Keeping the important output cursor as a -// // function local stack variable allows the compiler to keep it in -// // register, which greatly aids performance by avoiding loads and stores of -// // this variable in the fast path loop iterations. -// T GetOutputPtr() const; -// -// // At end of decompression the loop donates the ownership of the cursor -// // variable back to the writer by calling this function. -// void SetOutputPtr(T op); -// -// // Called after decompression -// bool CheckLength() const; -// -// // Called repeatedly during decompression -// // Each function get a pointer to the op (output pointer), that the writer -// // can use and update. Note it's important that these functions get fully -// // inlined so that no actual address of the local variable needs to be -// // taken. -// bool Append(const char* ip, size_t length, T* op); -// bool AppendFromSelf(uint32_t offset, size_t length, T* op); -// -// // The rules for how TryFastAppend differs from Append are somewhat -// // convoluted: -// // -// // - TryFastAppend is allowed to decline (return false) at any -// // time, for any reason -- just "return false" would be -// // a perfectly legal implementation of TryFastAppend. -// // The intention is for TryFastAppend to allow a fast path -// // in the common case of a small append. -// // - TryFastAppend is allowed to read up to bytes -// // from the input buffer, whereas Append is allowed to read -// // . However, if it returns true, it must leave -// // at least five (kMaximumTagLength) bytes in the input buffer -// // afterwards, so that there is always enough space to read the -// // next tag without checking for a refill. -// // - TryFastAppend must always return decline (return false) -// // if is 61 or more, as in this case the literal length is not -// // decoded fully. In practice, this should not be a big problem, -// // as it is unlikely that one would implement a fast path accepting -// // this much data. -// // -// bool TryFastAppend(const char* ip, size_t available, size_t length, T* op); -// }; - -static inline uint32_t ExtractLowBytes(const uint32_t& v, int n) { - assert(n >= 0); - assert(n <= 4); -#if SNAPPY_HAVE_BMI2 - return _bzhi_u32(v, 8 * n); -#else - // This needs to be wider than uint32_t otherwise `mask << 32` will be - // undefined. - uint64_t mask = 0xffffffff; - return v & ~(mask << (8 * n)); -#endif -} - -static inline bool LeftShiftOverflows(uint8_t value, uint32_t shift) { - assert(shift < 32); - static const uint8_t masks[] = { - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // - 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // - 0x00, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe}; - return (value & masks[shift]) != 0; -} - -inline bool Copy64BytesWithPatternExtension(ptrdiff_t dst, size_t offset) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)dst; - return offset != 0; -} - -// Copies between size bytes and 64 bytes from src to dest. size cannot exceed -// 64. More than size bytes, but never exceeding 64, might be copied if doing -// so gives better performance. [src, src + size) must not overlap with -// [dst, dst + size), but [src, src + 64) may overlap with [dst, dst + 64). -void MemCopy64(char* dst, const void* src, size_t size) { - // Always copy this many bytes. If that's below size then copy the full 64. - constexpr int kShortMemCopy = 32; - - assert(size <= 64); - assert(std::less_equal()(static_cast(src) + size, - dst) || - std::less_equal()(dst + size, src)); - - // We know that src and dst are at least size bytes apart. However, because we - // might copy more than size bytes the copy still might overlap past size. - // E.g. if src and dst appear consecutively in memory (src + size >= dst). - // TODO: Investigate wider copies on other platforms. -#if defined(__x86_64__) && defined(__AVX__) - assert(kShortMemCopy <= 32); - __m256i data = _mm256_lddqu_si256(static_cast(src)); - _mm256_storeu_si256(reinterpret_cast<__m256i *>(dst), data); - // Profiling shows that nearly all copies are short. - if (SNAPPY_PREDICT_FALSE(size > kShortMemCopy)) { - data = _mm256_lddqu_si256(static_cast(src) + 1); - _mm256_storeu_si256(reinterpret_cast<__m256i *>(dst) + 1, data); - } -#else - std::memmove(dst, src, kShortMemCopy); - // Profiling shows that nearly all copies are short. - if (SNAPPY_PREDICT_FALSE(size > kShortMemCopy)) { - std::memmove(dst + kShortMemCopy, - static_cast(src) + kShortMemCopy, - 64 - kShortMemCopy); - } -#endif -} - -void MemCopy64(ptrdiff_t dst, const void* src, size_t size) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)dst; - (void)src; - (void)size; -} - -void ClearDeferred(const void** deferred_src, size_t* deferred_length, - uint8_t* safe_source) { - *deferred_src = safe_source; - *deferred_length = 0; -} - -void DeferMemCopy(const void** deferred_src, size_t* deferred_length, - const void* src, size_t length) { - *deferred_src = src; - *deferred_length = length; -} - -SNAPPY_ATTRIBUTE_ALWAYS_INLINE -inline size_t AdvanceToNextTagARMOptimized(const uint8_t** ip_p, size_t* tag) { - const uint8_t*& ip = *ip_p; - // This section is crucial for the throughput of the decompression loop. - // The latency of an iteration is fundamentally constrained by the - // following data chain on ip. - // ip -> c = Load(ip) -> delta1 = (c & 3) -> ip += delta1 or delta2 - // delta2 = ((c >> 2) + 1) ip++ - // This is different from X86 optimizations because ARM has conditional add - // instruction (csinc) and it removes several register moves. - const size_t tag_type = *tag & 3; - const bool is_literal = (tag_type == 0); - if (is_literal) { - size_t next_literal_tag = (*tag >> 2) + 1; - *tag = ip[next_literal_tag]; - ip += next_literal_tag + 1; - } else { - *tag = ip[tag_type]; - ip += tag_type + 1; - } - return tag_type; -} - -SNAPPY_ATTRIBUTE_ALWAYS_INLINE -inline size_t AdvanceToNextTagX86Optimized(const uint8_t** ip_p, size_t* tag) { - const uint8_t*& ip = *ip_p; - // This section is crucial for the throughput of the decompression loop. - // The latency of an iteration is fundamentally constrained by the - // following data chain on ip. - // ip -> c = Load(ip) -> ip1 = ip + 1 + (c & 3) -> ip = ip1 or ip2 - // ip2 = ip + 2 + (c >> 2) - // This amounts to 8 cycles. - // 5 (load) + 1 (c & 3) + 1 (lea ip1, [ip + (c & 3) + 1]) + 1 (cmov) - size_t literal_len = *tag >> 2; - size_t tag_type = *tag; - bool is_literal; -#if defined(__GCC_ASM_FLAG_OUTPUTS__) && defined(__x86_64__) - // TODO clang misses the fact that the (c & 3) already correctly - // sets the zero flag. - asm("and $3, %k[tag_type]\n\t" - : [tag_type] "+r"(tag_type), "=@ccz"(is_literal) - :: "cc"); -#else - tag_type &= 3; - is_literal = (tag_type == 0); -#endif - // TODO - // This is code is subtle. Loading the values first and then cmov has less - // latency then cmov ip and then load. However clang would move the loads - // in an optimization phase, volatile prevents this transformation. - // Note that we have enough slop bytes (64) that the loads are always valid. - size_t tag_literal = - static_cast(ip)[1 + literal_len]; - size_t tag_copy = static_cast(ip)[tag_type]; - *tag = is_literal ? tag_literal : tag_copy; - const uint8_t* ip_copy = ip + 1 + tag_type; - const uint8_t* ip_literal = ip + 2 + literal_len; - ip = is_literal ? ip_literal : ip_copy; -#if defined(__GNUC__) && defined(__x86_64__) - // TODO Clang is "optimizing" zero-extension (a totally free - // operation) this means that after the cmov of tag, it emits another movzb - // tag, byte(tag). It really matters as it's on the core chain. This dummy - // asm, persuades clang to do the zero-extension at the load (it's automatic) - // removing the expensive movzb. - asm("" ::"r"(tag_copy)); -#endif - return tag_type; -} - -// Extract the offset for copy-1 and copy-2 returns 0 for literals or copy-4. -inline uint32_t ExtractOffset(uint32_t val, size_t tag_type) { - // For x86 non-static storage works better. For ARM static storage is better. - // TODO: Once the array is recognized as a register, improve the - // readability for x86. -#if defined(__x86_64__) - constexpr uint64_t kExtractMasksCombined = 0x0000FFFF00FF0000ull; - uint16_t result; - memcpy(&result, - reinterpret_cast(&kExtractMasksCombined) + 2 * tag_type, - sizeof(result)); - return val & result; -#elif defined(__aarch64__) - constexpr uint64_t kExtractMasksCombined = 0x0000FFFF00FF0000ull; - return val & static_cast( - (kExtractMasksCombined >> (tag_type * 16)) & 0xFFFF); -#else - static constexpr uint32_t kExtractMasks[4] = {0, 0xFF, 0xFFFF, 0}; - return val & kExtractMasks[tag_type]; -#endif -}; - -// Core decompression loop, when there is enough data available. -// Decompresses the input buffer [ip, ip_limit) into the output buffer -// [op, op_limit_min_slop). Returning when either we are too close to the end -// of the input buffer, or we exceed op_limit_min_slop or when a exceptional -// tag is encountered (literal of length > 60) or a copy-4. -// Returns {ip, op} at the points it stopped decoding. -// TODO This function probably does not need to be inlined, as it -// should decode large chunks at a time. This allows runtime dispatch to -// implementations based on CPU capability (BMI2 / perhaps 32 / 64 byte memcpy). -template -std::pair DecompressBranchless( - const uint8_t* ip, const uint8_t* ip_limit, ptrdiff_t op, T op_base, - ptrdiff_t op_limit_min_slop) { - // If deferred_src is invalid point it here. - uint8_t safe_source[64]; - const void* deferred_src; - size_t deferred_length; - ClearDeferred(&deferred_src, &deferred_length, safe_source); - - // We unroll the inner loop twice so we need twice the spare room. - op_limit_min_slop -= kSlopBytes; - if (2 * (kSlopBytes + 1) < ip_limit - ip && op < op_limit_min_slop) { - const uint8_t* const ip_limit_min_slop = ip_limit - 2 * kSlopBytes - 1; - ip++; - // ip points just past the tag and we are touching at maximum kSlopBytes - // in an iteration. - size_t tag = ip[-1]; -#if defined(__clang__) && defined(__aarch64__) - // Workaround for https://bugs.llvm.org/show_bug.cgi?id=51317 - // when loading 1 byte, clang for aarch64 doesn't realize that it(ldrb) - // comes with free zero-extension, so clang generates another - // 'and xn, xm, 0xff' before it use that as the offset. This 'and' is - // redundant and can be removed by adding this dummy asm, which gives - // clang a hint that we're doing the zero-extension at the load. - asm("" ::"r"(tag)); -#endif - do { - // The throughput is limited by instructions, unrolling the inner loop - // twice reduces the amount of instructions checking limits and also - // leads to reduced mov's. - - SNAPPY_PREFETCH(ip + 128); - for (int i = 0; i < 2; i++) { - const uint8_t* old_ip = ip; - assert(tag == ip[-1]); - // For literals tag_type = 0, hence we will always obtain 0 from - // ExtractLowBytes. For literals offset will thus be kLiteralOffset. - ptrdiff_t len_min_offset = kLengthMinusOffset[tag]; -#if defined(__aarch64__) - size_t tag_type = AdvanceToNextTagARMOptimized(&ip, &tag); -#else - size_t tag_type = AdvanceToNextTagX86Optimized(&ip, &tag); -#endif - uint32_t next = LittleEndian::Load32(old_ip); - size_t len = len_min_offset & 0xFF; - len_min_offset -= ExtractOffset(next, tag_type); - if (SNAPPY_PREDICT_FALSE(len_min_offset > 0)) { - if (SNAPPY_PREDICT_FALSE(len & 0x80)) { - // Exceptional case (long literal or copy 4). - // Actually doing the copy here is negatively impacting the main - // loop due to compiler incorrectly allocating a register for - // this fallback. Hence we just break. - break_loop: - ip = old_ip; - goto exit; - } - // Only copy-1 or copy-2 tags can get here. - assert(tag_type == 1 || tag_type == 2); - std::ptrdiff_t delta = (op + deferred_length) + len_min_offset - len; - // Guard against copies before the buffer start. - // Execute any deferred MemCopy since we write to dst here. - MemCopy64(op_base + op, deferred_src, deferred_length); - op += deferred_length; - ClearDeferred(&deferred_src, &deferred_length, safe_source); - if (SNAPPY_PREDICT_FALSE(delta < 0 || - !Copy64BytesWithPatternExtension( - op_base + op, len - len_min_offset))) { - goto break_loop; - } - // We aren't deferring this copy so add length right away. - op += len; - continue; - } - std::ptrdiff_t delta = (op + deferred_length) + len_min_offset - len; - if (SNAPPY_PREDICT_FALSE(delta < 0)) { - // Due to the spurious offset in literals have this will trigger - // at the start of a block when op is still smaller than 256. - if (tag_type != 0) goto break_loop; - MemCopy64(op_base + op, deferred_src, deferred_length); - op += deferred_length; - DeferMemCopy(&deferred_src, &deferred_length, old_ip, len); - continue; - } - - // For copies we need to copy from op_base + delta, for literals - // we need to copy from ip instead of from the stream. - const void* from = - tag_type ? reinterpret_cast(op_base + delta) : old_ip; - MemCopy64(op_base + op, deferred_src, deferred_length); - op += deferred_length; - DeferMemCopy(&deferred_src, &deferred_length, from, len); - } - } while (ip < ip_limit_min_slop && - (op + deferred_length) < op_limit_min_slop); - exit: - ip--; - assert(ip <= ip_limit); - } - // If we deferred a copy then we can perform. If we are up to date then we - // might not have enough slop bytes and could run past the end. - if (deferred_length) { - MemCopy64(op_base + op, deferred_src, deferred_length); - op += deferred_length; - ClearDeferred(&deferred_src, &deferred_length, safe_source); - } - return {ip, op}; -} - -// Helper class for decompression -class SnappyDecompressor { - private: - Source* reader_; // Underlying source of bytes to decompress - const char* ip_; // Points to next buffered byte - const char* ip_limit_; // Points just past buffered bytes - // If ip < ip_limit_min_maxtaglen_ it's safe to read kMaxTagLength from - // buffer. - const char* ip_limit_min_maxtaglen_; - uint32_t peeked_; // Bytes peeked from reader (need to skip) - bool eof_; // Hit end of input without an error? - char scratch_[kMaximumTagLength]; // See RefillTag(). - - // Ensure that all of the tag metadata for the next tag is available - // in [ip_..ip_limit_-1]. Also ensures that [ip,ip+4] is readable even - // if (ip_limit_ - ip_ < 5). - // - // Returns true on success, false on error or end of input. - bool RefillTag(); - - void ResetLimit(const char* ip) { - ip_limit_min_maxtaglen_ = - ip_limit_ - std::min(ip_limit_ - ip, kMaximumTagLength - 1); - } - - public: - explicit SnappyDecompressor(Source* reader) - : reader_(reader), ip_(NULL), ip_limit_(NULL), peeked_(0), eof_(false) {} - - ~SnappyDecompressor() { - // Advance past any bytes we peeked at from the reader - reader_->Skip(peeked_); - } - - // Returns true iff we have hit the end of the input without an error. - bool eof() const { return eof_; } - - // Read the uncompressed length stored at the start of the compressed data. - // On success, stores the length in *result and returns true. - // On failure, returns false. - bool ReadUncompressedLength(uint32_t* result) { - assert(ip_ == NULL); // Must not have read anything yet - // Length is encoded in 1..5 bytes - *result = 0; - uint32_t shift = 0; - while (true) { - if (shift >= 32) return false; - size_t n; - const char* ip = reader_->Peek(&n); - if (n == 0) return false; - const unsigned char c = *(reinterpret_cast(ip)); - reader_->Skip(1); - uint32_t val = c & 0x7f; - if (LeftShiftOverflows(static_cast(val), shift)) return false; - *result |= val << shift; - if (c < 128) { - break; - } - shift += 7; - } - return true; - } - - // Process the next item found in the input. - // Returns true if successful, false on error or end of input. - template -#if defined(__GNUC__) && defined(__x86_64__) - __attribute__((aligned(32))) -#endif - void - DecompressAllTags(Writer* writer) { - const char* ip = ip_; - ResetLimit(ip); - auto op = writer->GetOutputPtr(); - // We could have put this refill fragment only at the beginning of the loop. - // However, duplicating it at the end of each branch gives the compiler more - // scope to optimize the expression based on the local - // context, which overall increases speed. -#define MAYBE_REFILL() \ - if (SNAPPY_PREDICT_FALSE(ip >= ip_limit_min_maxtaglen_)) { \ - ip_ = ip; \ - if (SNAPPY_PREDICT_FALSE(!RefillTag())) goto exit; \ - ip = ip_; \ - ResetLimit(ip); \ - } \ - preload = static_cast(*ip) - - // At the start of the for loop below the least significant byte of preload - // contains the tag. - uint32_t preload; - MAYBE_REFILL(); - for (;;) { - { - ptrdiff_t op_limit_min_slop; - auto op_base = writer->GetBase(&op_limit_min_slop); - if (op_base) { - auto res = - DecompressBranchless(reinterpret_cast(ip), - reinterpret_cast(ip_limit_), - op - op_base, op_base, op_limit_min_slop); - ip = reinterpret_cast(res.first); - op = op_base + res.second; - MAYBE_REFILL(); - } - } - const uint8_t c = static_cast(preload); - ip++; - - // Ratio of iterations that have LITERAL vs non-LITERAL for different - // inputs. - // - // input LITERAL NON_LITERAL - // ----------------------------------- - // html|html4|cp 23% 77% - // urls 36% 64% - // jpg 47% 53% - // pdf 19% 81% - // txt[1-4] 25% 75% - // pb 24% 76% - // bin 24% 76% - if (SNAPPY_PREDICT_FALSE((c & 0x3) == LITERAL)) { - size_t literal_length = (c >> 2) + 1u; - if (writer->TryFastAppend(ip, ip_limit_ - ip, literal_length, &op)) { - assert(literal_length < 61); - ip += literal_length; - // NOTE: There is no MAYBE_REFILL() here, as TryFastAppend() - // will not return true unless there's already at least five spare - // bytes in addition to the literal. - preload = static_cast(*ip); - continue; - } - if (SNAPPY_PREDICT_FALSE(literal_length >= 61)) { - // Long literal. - const size_t literal_length_length = literal_length - 60; - literal_length = - ExtractLowBytes(LittleEndian::Load32(ip), literal_length_length) + - 1; - ip += literal_length_length; - } - - size_t avail = ip_limit_ - ip; - while (avail < literal_length) { - if (!writer->Append(ip, avail, &op)) goto exit; - literal_length -= avail; - reader_->Skip(peeked_); - size_t n; - ip = reader_->Peek(&n); - avail = n; - peeked_ = avail; - if (avail == 0) goto exit; - ip_limit_ = ip + avail; - ResetLimit(ip); - } - if (!writer->Append(ip, literal_length, &op)) goto exit; - ip += literal_length; - MAYBE_REFILL(); - } else { - if (SNAPPY_PREDICT_FALSE((c & 3) == COPY_4_BYTE_OFFSET)) { - const size_t copy_offset = LittleEndian::Load32(ip); - const size_t length = (c >> 2) + 1; - ip += 4; - - if (!writer->AppendFromSelf(copy_offset, length, &op)) goto exit; - } else { - const ptrdiff_t entry = kLengthMinusOffset[c]; - preload = LittleEndian::Load32(ip); - const uint32_t trailer = ExtractLowBytes(preload, c & 3); - const uint32_t length = entry & 0xff; - assert(length > 0); - - // copy_offset/256 is encoded in bits 8..10. By just fetching - // those bits, we get copy_offset (since the bit-field starts at - // bit 8). - const uint32_t copy_offset = trailer - entry + length; - if (!writer->AppendFromSelf(copy_offset, length, &op)) goto exit; - - ip += (c & 3); - // By using the result of the previous load we reduce the critical - // dependency chain of ip to 4 cycles. - preload >>= (c & 3) * 8; - if (ip < ip_limit_min_maxtaglen_) continue; - } - MAYBE_REFILL(); - } - } -#undef MAYBE_REFILL - exit: - writer->SetOutputPtr(op); - } -}; - -constexpr uint32_t CalculateNeeded(uint8_t tag) { - return ((tag & 3) == 0 && tag >= (60 * 4)) - ? (tag >> 2) - 58 - : (0x05030201 >> ((tag * 8) & 31)) & 0xFF; -} - -#if __cplusplus >= 201402L -constexpr bool VerifyCalculateNeeded() { - for (int i = 0; i < 1; i++) { - if (CalculateNeeded(i) != (char_table[i] >> 11) + 1) return false; - } - return true; -} - -// Make sure CalculateNeeded is correct by verifying it against the established -// table encoding the number of added bytes needed. -static_assert(VerifyCalculateNeeded(), ""); -#endif // c++14 - -bool SnappyDecompressor::RefillTag() { - const char* ip = ip_; - if (ip == ip_limit_) { - // Fetch a new fragment from the reader - reader_->Skip(peeked_); // All peeked bytes are used up - size_t n; - ip = reader_->Peek(&n); - peeked_ = n; - eof_ = (n == 0); - if (eof_) return false; - ip_limit_ = ip + n; - } - - // Read the tag character - assert(ip < ip_limit_); - const unsigned char c = *(reinterpret_cast(ip)); - // At this point make sure that the data for the next tag is consecutive. - // For copy 1 this means the next 2 bytes (tag and 1 byte offset) - // For copy 2 the next 3 bytes (tag and 2 byte offset) - // For copy 4 the next 5 bytes (tag and 4 byte offset) - // For all small literals we only need 1 byte buf for literals 60...63 the - // length is encoded in 1...4 extra bytes. - const uint32_t needed = CalculateNeeded(c); - assert(needed <= sizeof(scratch_)); - - // Read more bytes from reader if needed - uint32_t nbuf = ip_limit_ - ip; - if (nbuf < needed) { - // Stitch together bytes from ip and reader to form the word - // contents. We store the needed bytes in "scratch_". They - // will be consumed immediately by the caller since we do not - // read more than we need. - std::memmove(scratch_, ip, nbuf); - reader_->Skip(peeked_); // All peeked bytes are used up - peeked_ = 0; - while (nbuf < needed) { - size_t length; - const char* src = reader_->Peek(&length); - if (length == 0) return false; - uint32_t to_add = std::min(needed - nbuf, length); - std::memcpy(scratch_ + nbuf, src, to_add); - nbuf += to_add; - reader_->Skip(to_add); - } - assert(nbuf == needed); - ip_ = scratch_; - ip_limit_ = scratch_ + needed; - } else if (nbuf < kMaximumTagLength) { - // Have enough bytes, but move into scratch_ so that we do not - // read past end of input - std::memmove(scratch_, ip, nbuf); - reader_->Skip(peeked_); // All peeked bytes are used up - peeked_ = 0; - ip_ = scratch_; - ip_limit_ = scratch_ + nbuf; - } else { - // Pass pointer to buffer returned by reader_. - ip_ = ip; - } - return true; -} - -template -static bool InternalUncompress(Source* r, Writer* writer) { - // Read the uncompressed length from the front of the compressed input - SnappyDecompressor decompressor(r); - uint32_t uncompressed_len = 0; - if (!decompressor.ReadUncompressedLength(&uncompressed_len)) return false; - - return InternalUncompressAllTags(&decompressor, writer, r->Available(), - uncompressed_len); -} - -template -static bool InternalUncompressAllTags(SnappyDecompressor* decompressor, - Writer* writer, uint32_t compressed_len, - uint32_t uncompressed_len) { - Report("snappy_uncompress", compressed_len, uncompressed_len); - - writer->SetExpectedLength(uncompressed_len); - - // Process the entire input - decompressor->DecompressAllTags(writer); - writer->Flush(); - return (decompressor->eof() && writer->CheckLength()); -} - -bool GetUncompressedLength(Source* source, uint32_t* result) { - SnappyDecompressor decompressor(source); - return decompressor.ReadUncompressedLength(result); -} - -size_t Compress(Source* reader, Sink* writer) { - size_t written = 0; - size_t N = reader->Available(); - const size_t uncompressed_size = N; - char ulength[Varint::kMax32]; - char* p = Varint::Encode32(ulength, N); - writer->Append(ulength, p - ulength); - written += (p - ulength); - - internal::WorkingMemory wmem(N); - - while (N > 0) { - // Get next block to compress (without copying if possible) - size_t fragment_size; - const char* fragment = reader->Peek(&fragment_size); - assert(fragment_size != 0); // premature end of input - const size_t num_to_read = std::min(N, kBlockSize); - size_t bytes_read = fragment_size; - - size_t pending_advance = 0; - if (bytes_read >= num_to_read) { - // Buffer returned by reader is large enough - pending_advance = num_to_read; - fragment_size = num_to_read; - } else { - char* scratch = wmem.GetScratchInput(); - std::memcpy(scratch, fragment, bytes_read); - reader->Skip(bytes_read); - - while (bytes_read < num_to_read) { - fragment = reader->Peek(&fragment_size); - size_t n = std::min(fragment_size, num_to_read - bytes_read); - std::memcpy(scratch + bytes_read, fragment, n); - bytes_read += n; - reader->Skip(n); - } - assert(bytes_read == num_to_read); - fragment = scratch; - fragment_size = num_to_read; - } - assert(fragment_size == num_to_read); - - // Get encoding table for compression - int table_size; - uint16_t* table = wmem.GetHashTable(num_to_read, &table_size); - - // Compress input_fragment and append to dest - const int max_output = MaxCompressedLength(num_to_read); - - // Need a scratch buffer for the output, in case the byte sink doesn't - // have room for us directly. - - // Since we encode kBlockSize regions followed by a region - // which is <= kBlockSize in length, a previously allocated - // scratch_output[] region is big enough for this iteration. - char* dest = writer->GetAppendBuffer(max_output, wmem.GetScratchOutput()); - char* end = internal::CompressFragment(fragment, fragment_size, dest, table, - table_size); - writer->Append(dest, end - dest); - written += (end - dest); - - N -= num_to_read; - reader->Skip(pending_advance); - } - - Report("snappy_compress", written, uncompressed_size); - - return written; -} - -// ----------------------------------------------------------------------- -// IOVec interfaces -// ----------------------------------------------------------------------- - -// A `Source` implementation that yields the contents of an `iovec` array. Note -// that `total_size` is the total number of bytes to be read from the elements -// of `iov` (_not_ the total number of elements in `iov`). -class SnappyIOVecReader : public Source { - public: - SnappyIOVecReader(const struct iovec* iov, size_t total_size) - : curr_iov_(iov), - curr_pos_(total_size > 0 ? reinterpret_cast(iov->iov_base) - : nullptr), - curr_size_remaining_(total_size > 0 ? iov->iov_len : 0), - total_size_remaining_(total_size) { - // Skip empty leading `iovec`s. - if (total_size > 0 && curr_size_remaining_ == 0) Advance(); - } - - ~SnappyIOVecReader() = default; - - size_t Available() const { return total_size_remaining_; } - - const char* Peek(size_t* len) { - *len = curr_size_remaining_; - return curr_pos_; - } - - void Skip(size_t n) { - while (n >= curr_size_remaining_ && n > 0) { - n -= curr_size_remaining_; - Advance(); - } - curr_size_remaining_ -= n; - total_size_remaining_ -= n; - curr_pos_ += n; - } - - private: - // Advances to the next nonempty `iovec` and updates related variables. - void Advance() { - do { - assert(total_size_remaining_ >= curr_size_remaining_); - total_size_remaining_ -= curr_size_remaining_; - if (total_size_remaining_ == 0) { - curr_pos_ = nullptr; - curr_size_remaining_ = 0; - return; - } - ++curr_iov_; - curr_pos_ = reinterpret_cast(curr_iov_->iov_base); - curr_size_remaining_ = curr_iov_->iov_len; - } while (curr_size_remaining_ == 0); - } - - // The `iovec` currently being read. - const struct iovec* curr_iov_; - // The location in `curr_iov_` currently being read. - const char* curr_pos_; - // The amount of unread data in `curr_iov_`. - size_t curr_size_remaining_; - // The amount of unread data in the entire input array. - size_t total_size_remaining_; -}; - -// A type that writes to an iovec. -// Note that this is not a "ByteSink", but a type that matches the -// Writer template argument to SnappyDecompressor::DecompressAllTags(). -class SnappyIOVecWriter { - private: - // output_iov_end_ is set to iov + count and used to determine when - // the end of the iovs is reached. - const struct iovec* output_iov_end_; - -#if !defined(NDEBUG) - const struct iovec* output_iov_; -#endif // !defined(NDEBUG) - - // Current iov that is being written into. - const struct iovec* curr_iov_; - - // Pointer to current iov's write location. - char* curr_iov_output_; - - // Remaining bytes to write into curr_iov_output. - size_t curr_iov_remaining_; - - // Total bytes decompressed into output_iov_ so far. - size_t total_written_; - - // Maximum number of bytes that will be decompressed into output_iov_. - size_t output_limit_; - - static inline char* GetIOVecPointer(const struct iovec* iov, size_t offset) { - return reinterpret_cast(iov->iov_base) + offset; - } - - public: - // Does not take ownership of iov. iov must be valid during the - // entire lifetime of the SnappyIOVecWriter. - inline SnappyIOVecWriter(const struct iovec* iov, size_t iov_count) - : output_iov_end_(iov + iov_count), -#if !defined(NDEBUG) - output_iov_(iov), -#endif // !defined(NDEBUG) - curr_iov_(iov), - curr_iov_output_(iov_count ? reinterpret_cast(iov->iov_base) - : nullptr), - curr_iov_remaining_(iov_count ? iov->iov_len : 0), - total_written_(0), - output_limit_(-1) { - } - - inline void SetExpectedLength(size_t len) { output_limit_ = len; } - - inline bool CheckLength() const { return total_written_ == output_limit_; } - - inline bool Append(const char* ip, size_t len, char**) { - if (total_written_ + len > output_limit_) { - return false; - } - - return AppendNoCheck(ip, len); - } - - char* GetOutputPtr() { return nullptr; } - char* GetBase(ptrdiff_t*) { return nullptr; } - void SetOutputPtr(char* op) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)op; - } - - inline bool AppendNoCheck(const char* ip, size_t len) { - while (len > 0) { - if (curr_iov_remaining_ == 0) { - // This iovec is full. Go to the next one. - if (curr_iov_ + 1 >= output_iov_end_) { - return false; - } - ++curr_iov_; - curr_iov_output_ = reinterpret_cast(curr_iov_->iov_base); - curr_iov_remaining_ = curr_iov_->iov_len; - } - - const size_t to_write = std::min(len, curr_iov_remaining_); - std::memcpy(curr_iov_output_, ip, to_write); - curr_iov_output_ += to_write; - curr_iov_remaining_ -= to_write; - total_written_ += to_write; - ip += to_write; - len -= to_write; - } - - return true; - } - - inline bool TryFastAppend(const char* ip, size_t available, size_t len, - char**) { - const size_t space_left = output_limit_ - total_written_; - if (len <= 16 && available >= 16 + kMaximumTagLength && space_left >= 16 && - curr_iov_remaining_ >= 16) { - // Fast path, used for the majority (about 95%) of invocations. - UnalignedCopy128(ip, curr_iov_output_); - curr_iov_output_ += len; - curr_iov_remaining_ -= len; - total_written_ += len; - return true; - } - - return false; - } - - inline bool AppendFromSelf(size_t offset, size_t len, char**) { - // See SnappyArrayWriter::AppendFromSelf for an explanation of - // the "offset - 1u" trick. - if (offset - 1u >= total_written_) { - return false; - } - const size_t space_left = output_limit_ - total_written_; - if (len > space_left) { - return false; - } - - // Locate the iovec from which we need to start the copy. - const iovec* from_iov = curr_iov_; - size_t from_iov_offset = curr_iov_->iov_len - curr_iov_remaining_; - while (offset > 0) { - if (from_iov_offset >= offset) { - from_iov_offset -= offset; - break; - } - - offset -= from_iov_offset; - --from_iov; -#if !defined(NDEBUG) - assert(from_iov >= output_iov_); -#endif // !defined(NDEBUG) - from_iov_offset = from_iov->iov_len; - } - - // Copy bytes starting from the iovec pointed to by from_iov_index to - // the current iovec. - while (len > 0) { - assert(from_iov <= curr_iov_); - if (from_iov != curr_iov_) { - const size_t to_copy = - std::min(from_iov->iov_len - from_iov_offset, len); - AppendNoCheck(GetIOVecPointer(from_iov, from_iov_offset), to_copy); - len -= to_copy; - if (len > 0) { - ++from_iov; - from_iov_offset = 0; - } - } else { - size_t to_copy = curr_iov_remaining_; - if (to_copy == 0) { - // This iovec is full. Go to the next one. - if (curr_iov_ + 1 >= output_iov_end_) { - return false; - } - ++curr_iov_; - curr_iov_output_ = reinterpret_cast(curr_iov_->iov_base); - curr_iov_remaining_ = curr_iov_->iov_len; - continue; - } - if (to_copy > len) { - to_copy = len; - } - assert(to_copy > 0); - - IncrementalCopy(GetIOVecPointer(from_iov, from_iov_offset), - curr_iov_output_, curr_iov_output_ + to_copy, - curr_iov_output_ + curr_iov_remaining_); - curr_iov_output_ += to_copy; - curr_iov_remaining_ -= to_copy; - from_iov_offset += to_copy; - total_written_ += to_copy; - len -= to_copy; - } - } - - return true; - } - - inline void Flush() {} -}; - -bool RawUncompressToIOVec(const char* compressed, size_t compressed_length, - const struct iovec* iov, size_t iov_cnt) { - ByteArraySource reader(compressed, compressed_length); - return RawUncompressToIOVec(&reader, iov, iov_cnt); -} - -bool RawUncompressToIOVec(Source* compressed, const struct iovec* iov, - size_t iov_cnt) { - SnappyIOVecWriter output(iov, iov_cnt); - return InternalUncompress(compressed, &output); -} - -// ----------------------------------------------------------------------- -// Flat array interfaces -// ----------------------------------------------------------------------- - -// A type that writes to a flat array. -// Note that this is not a "ByteSink", but a type that matches the -// Writer template argument to SnappyDecompressor::DecompressAllTags(). -class SnappyArrayWriter { - private: - char* base_; - char* op_; - char* op_limit_; - // If op < op_limit_min_slop_ then it's safe to unconditionally write - // kSlopBytes starting at op. - char* op_limit_min_slop_; - - public: - inline explicit SnappyArrayWriter(char* dst) - : base_(dst), - op_(dst), - op_limit_(dst), - op_limit_min_slop_(dst) {} // Safe default see invariant. - - inline void SetExpectedLength(size_t len) { - op_limit_ = op_ + len; - // Prevent pointer from being past the buffer. - op_limit_min_slop_ = op_limit_ - std::min(kSlopBytes - 1, len); - } - - inline bool CheckLength() const { return op_ == op_limit_; } - - char* GetOutputPtr() { return op_; } - char* GetBase(ptrdiff_t* op_limit_min_slop) { - *op_limit_min_slop = op_limit_min_slop_ - base_; - return base_; - } - void SetOutputPtr(char* op) { op_ = op; } - - inline bool Append(const char* ip, size_t len, char** op_p) { - char* op = *op_p; - const size_t space_left = op_limit_ - op; - if (space_left < len) return false; - std::memcpy(op, ip, len); - *op_p = op + len; - return true; - } - - inline bool TryFastAppend(const char* ip, size_t available, size_t len, - char** op_p) { - char* op = *op_p; - const size_t space_left = op_limit_ - op; - if (len <= 16 && available >= 16 + kMaximumTagLength && space_left >= 16) { - // Fast path, used for the majority (about 95%) of invocations. - UnalignedCopy128(ip, op); - *op_p = op + len; - return true; - } else { - return false; - } - } - - SNAPPY_ATTRIBUTE_ALWAYS_INLINE - inline bool AppendFromSelf(size_t offset, size_t len, char** op_p) { - assert(len > 0); - char* const op = *op_p; - assert(op >= base_); - char* const op_end = op + len; - - // Check if we try to append from before the start of the buffer. - if (SNAPPY_PREDICT_FALSE(static_cast(op - base_) < offset)) - return false; - - if (SNAPPY_PREDICT_FALSE((kSlopBytes < 64 && len > kSlopBytes) || - op >= op_limit_min_slop_ || offset < len)) { - if (op_end > op_limit_ || offset == 0) return false; - *op_p = IncrementalCopy(op - offset, op, op_end, op_limit_); - return true; - } - std::memmove(op, op - offset, kSlopBytes); - *op_p = op_end; - return true; - } - inline size_t Produced() const { - assert(op_ >= base_); - return op_ - base_; - } - inline void Flush() {} -}; - -bool RawUncompress(const char* compressed, size_t compressed_length, - char* uncompressed) { - ByteArraySource reader(compressed, compressed_length); - return RawUncompress(&reader, uncompressed); -} - -bool RawUncompress(Source* compressed, char* uncompressed) { - SnappyArrayWriter output(uncompressed); - return InternalUncompress(compressed, &output); -} - -bool Uncompress(const char* compressed, size_t compressed_length, - std::string* uncompressed) { - size_t ulength; - if (!GetUncompressedLength(compressed, compressed_length, &ulength)) { - return false; - } - // On 32-bit builds: max_size() < kuint32max. Check for that instead - // of crashing (e.g., consider externally specified compressed data). - if (ulength > uncompressed->max_size()) { - return false; - } - STLStringResizeUninitialized(uncompressed, ulength); - return RawUncompress(compressed, compressed_length, - string_as_array(uncompressed)); -} - -// A Writer that drops everything on the floor and just does validation -class SnappyDecompressionValidator { - private: - size_t expected_; - size_t produced_; - - public: - inline SnappyDecompressionValidator() : expected_(0), produced_(0) {} - inline void SetExpectedLength(size_t len) { expected_ = len; } - size_t GetOutputPtr() { return produced_; } - size_t GetBase(ptrdiff_t* op_limit_min_slop) { - *op_limit_min_slop = std::numeric_limits::max() - kSlopBytes + 1; - return 1; - } - void SetOutputPtr(size_t op) { produced_ = op; } - inline bool CheckLength() const { return expected_ == produced_; } - inline bool Append(const char* ip, size_t len, size_t* produced) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)ip; - - *produced += len; - return *produced <= expected_; - } - inline bool TryFastAppend(const char* ip, size_t available, size_t length, - size_t* produced) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)ip; - (void)available; - (void)length; - (void)produced; - - return false; - } - inline bool AppendFromSelf(size_t offset, size_t len, size_t* produced) { - // See SnappyArrayWriter::AppendFromSelf for an explanation of - // the "offset - 1u" trick. - if (*produced <= offset - 1u) return false; - *produced += len; - return *produced <= expected_; - } - inline void Flush() {} -}; - -bool IsValidCompressedBuffer(const char* compressed, size_t compressed_length) { - ByteArraySource reader(compressed, compressed_length); - SnappyDecompressionValidator writer; - return InternalUncompress(&reader, &writer); -} - -bool IsValidCompressed(Source* compressed) { - SnappyDecompressionValidator writer; - return InternalUncompress(compressed, &writer); -} - -void RawCompress(const char* input, size_t input_length, char* compressed, - size_t* compressed_length) { - ByteArraySource reader(input, input_length); - UncheckedByteArraySink writer(compressed); - Compress(&reader, &writer); - - // Compute how many bytes were added - *compressed_length = (writer.CurrentDestination() - compressed); -} - -void RawCompressFromIOVec(const struct iovec* iov, size_t uncompressed_length, - char* compressed, size_t* compressed_length) { - SnappyIOVecReader reader(iov, uncompressed_length); - UncheckedByteArraySink writer(compressed); - Compress(&reader, &writer); - - // Compute how many bytes were added. - *compressed_length = writer.CurrentDestination() - compressed; -} - -size_t Compress(const char* input, size_t input_length, - std::string* compressed) { - // Pre-grow the buffer to the max length of the compressed output - STLStringResizeUninitialized(compressed, MaxCompressedLength(input_length)); - - size_t compressed_length; - RawCompress(input, input_length, string_as_array(compressed), - &compressed_length); - compressed->erase(compressed_length); - return compressed_length; -} - -size_t CompressFromIOVec(const struct iovec* iov, size_t iov_cnt, - std::string* compressed) { - // Compute the number of bytes to be compressed. - size_t uncompressed_length = 0; - for (size_t i = 0; i < iov_cnt; ++i) { - uncompressed_length += iov[i].iov_len; - } - - // Pre-grow the buffer to the max length of the compressed output. - STLStringResizeUninitialized(compressed, MaxCompressedLength( - uncompressed_length)); - - size_t compressed_length; - RawCompressFromIOVec(iov, uncompressed_length, string_as_array(compressed), - &compressed_length); - compressed->erase(compressed_length); - return compressed_length; -} - -// ----------------------------------------------------------------------- -// Sink interface -// ----------------------------------------------------------------------- - -// A type that decompresses into a Sink. The template parameter -// Allocator must export one method "char* Allocate(int size);", which -// allocates a buffer of "size" and appends that to the destination. -template -class SnappyScatteredWriter { - Allocator allocator_; - - // We need random access into the data generated so far. Therefore - // we keep track of all of the generated data as an array of blocks. - // All of the blocks except the last have length kBlockSize. - std::vector blocks_; - size_t expected_; - - // Total size of all fully generated blocks so far - size_t full_size_; - - // Pointer into current output block - char* op_base_; // Base of output block - char* op_ptr_; // Pointer to next unfilled byte in block - char* op_limit_; // Pointer just past block - // If op < op_limit_min_slop_ then it's safe to unconditionally write - // kSlopBytes starting at op. - char* op_limit_min_slop_; - - inline size_t Size() const { return full_size_ + (op_ptr_ - op_base_); } - - bool SlowAppend(const char* ip, size_t len); - bool SlowAppendFromSelf(size_t offset, size_t len); - - public: - inline explicit SnappyScatteredWriter(const Allocator& allocator) - : allocator_(allocator), - full_size_(0), - op_base_(NULL), - op_ptr_(NULL), - op_limit_(NULL), - op_limit_min_slop_(NULL) {} - char* GetOutputPtr() { return op_ptr_; } - char* GetBase(ptrdiff_t* op_limit_min_slop) { - *op_limit_min_slop = op_limit_min_slop_ - op_base_; - return op_base_; - } - void SetOutputPtr(char* op) { op_ptr_ = op; } - - inline void SetExpectedLength(size_t len) { - assert(blocks_.empty()); - expected_ = len; - } - - inline bool CheckLength() const { return Size() == expected_; } - - // Return the number of bytes actually uncompressed so far - inline size_t Produced() const { return Size(); } - - inline bool Append(const char* ip, size_t len, char** op_p) { - char* op = *op_p; - size_t avail = op_limit_ - op; - if (len <= avail) { - // Fast path - std::memcpy(op, ip, len); - *op_p = op + len; - return true; - } else { - op_ptr_ = op; - bool res = SlowAppend(ip, len); - *op_p = op_ptr_; - return res; - } - } - - inline bool TryFastAppend(const char* ip, size_t available, size_t length, - char** op_p) { - char* op = *op_p; - const int space_left = op_limit_ - op; - if (length <= 16 && available >= 16 + kMaximumTagLength && - space_left >= 16) { - // Fast path, used for the majority (about 95%) of invocations. - UnalignedCopy128(ip, op); - *op_p = op + length; - return true; - } else { - return false; - } - } - - inline bool AppendFromSelf(size_t offset, size_t len, char** op_p) { - char* op = *op_p; - assert(op >= op_base_); - // Check if we try to append from before the start of the buffer. - if (SNAPPY_PREDICT_FALSE((kSlopBytes < 64 && len > kSlopBytes) || - static_cast(op - op_base_) < offset || - op >= op_limit_min_slop_ || offset < len)) { - if (offset == 0) return false; - if (SNAPPY_PREDICT_FALSE(static_cast(op - op_base_) < offset || - op + len > op_limit_)) { - op_ptr_ = op; - bool res = SlowAppendFromSelf(offset, len); - *op_p = op_ptr_; - return res; - } - *op_p = IncrementalCopy(op - offset, op, op + len, op_limit_); - return true; - } - // Fast path - char* const op_end = op + len; - std::memmove(op, op - offset, kSlopBytes); - *op_p = op_end; - return true; - } - - // Called at the end of the decompress. We ask the allocator - // write all blocks to the sink. - inline void Flush() { allocator_.Flush(Produced()); } -}; - -template -bool SnappyScatteredWriter::SlowAppend(const char* ip, size_t len) { - size_t avail = op_limit_ - op_ptr_; - while (len > avail) { - // Completely fill this block - std::memcpy(op_ptr_, ip, avail); - op_ptr_ += avail; - assert(op_limit_ - op_ptr_ == 0); - full_size_ += (op_ptr_ - op_base_); - len -= avail; - ip += avail; - - // Bounds check - if (full_size_ + len > expected_) return false; - - // Make new block - size_t bsize = std::min(kBlockSize, expected_ - full_size_); - op_base_ = allocator_.Allocate(bsize); - op_ptr_ = op_base_; - op_limit_ = op_base_ + bsize; - op_limit_min_slop_ = op_limit_ - std::min(kSlopBytes - 1, bsize); - - blocks_.push_back(op_base_); - avail = bsize; - } - - std::memcpy(op_ptr_, ip, len); - op_ptr_ += len; - return true; -} - -template -bool SnappyScatteredWriter::SlowAppendFromSelf(size_t offset, - size_t len) { - // Overflow check - // See SnappyArrayWriter::AppendFromSelf for an explanation of - // the "offset - 1u" trick. - const size_t cur = Size(); - if (offset - 1u >= cur) return false; - if (expected_ - cur < len) return false; - - // Currently we shouldn't ever hit this path because Compress() chops the - // input into blocks and does not create cross-block copies. However, it is - // nice if we do not rely on that, since we can get better compression if we - // allow cross-block copies and thus might want to change the compressor in - // the future. - // TODO Replace this with a properly optimized path. This is not - // triggered right now. But this is so super slow, that it would regress - // performance unacceptably if triggered. - size_t src = cur - offset; - char* op = op_ptr_; - while (len-- > 0) { - char c = blocks_[src >> kBlockLog][src & (kBlockSize - 1)]; - if (!Append(&c, 1, &op)) { - op_ptr_ = op; - return false; - } - src++; - } - op_ptr_ = op; - return true; -} - -class SnappySinkAllocator { - public: - explicit SnappySinkAllocator(Sink* dest) : dest_(dest) {} - ~SnappySinkAllocator() {} - - char* Allocate(int size) { - Datablock block(new char[size], size); - blocks_.push_back(block); - return block.data; - } - - // We flush only at the end, because the writer wants - // random access to the blocks and once we hand the - // block over to the sink, we can't access it anymore. - // Also we don't write more than has been actually written - // to the blocks. - void Flush(size_t size) { - size_t size_written = 0; - for (Datablock& block : blocks_) { - size_t block_size = std::min(block.size, size - size_written); - dest_->AppendAndTakeOwnership(block.data, block_size, - &SnappySinkAllocator::Deleter, NULL); - size_written += block_size; - } - blocks_.clear(); - } - - private: - struct Datablock { - char* data; - size_t size; - Datablock(char* p, size_t s) : data(p), size(s) {} - }; - - static void Deleter(void* arg, const char* bytes, size_t size) { - // TODO: Switch to [[maybe_unused]] when we can assume C++17. - (void)arg; - (void)size; - - delete[] bytes; - } - - Sink* dest_; - std::vector blocks_; - - // Note: copying this object is allowed -}; - -size_t UncompressAsMuchAsPossible(Source* compressed, Sink* uncompressed) { - SnappySinkAllocator allocator(uncompressed); - SnappyScatteredWriter writer(allocator); - InternalUncompress(compressed, &writer); - return writer.Produced(); -} - -bool Uncompress(Source* compressed, Sink* uncompressed) { - // Read the uncompressed length from the front of the compressed input - SnappyDecompressor decompressor(compressed); - uint32_t uncompressed_len = 0; - if (!decompressor.ReadUncompressedLength(&uncompressed_len)) { - return false; - } - - char c; - size_t allocated_size; - char* buf = uncompressed->GetAppendBufferVariable(1, uncompressed_len, &c, 1, - &allocated_size); - - const size_t compressed_len = compressed->Available(); - // If we can get a flat buffer, then use it, otherwise do block by block - // uncompression - if (allocated_size >= uncompressed_len) { - SnappyArrayWriter writer(buf); - bool result = InternalUncompressAllTags(&decompressor, &writer, - compressed_len, uncompressed_len); - uncompressed->Append(buf, writer.Produced()); - return result; - } else { - SnappySinkAllocator allocator(uncompressed); - SnappyScatteredWriter writer(allocator); - return InternalUncompressAllTags(&decompressor, &writer, compressed_len, - uncompressed_len); - } -} - -} // namespace snappy diff --git a/contrib/libs/snappy/snappy.h b/contrib/libs/snappy/snappy.h deleted file mode 100644 index c7412fb26f2..00000000000 --- a/contrib/libs/snappy/snappy.h +++ /dev/null @@ -1,223 +0,0 @@ -// Copyright 2005 and onwards Google Inc. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// -// A light-weight compression algorithm. It is designed for speed of -// compression and decompression, rather than for the utmost in space -// savings. -// -// For getting better compression ratios when you are compressing data -// with long repeated sequences or compressing data that is similar to -// other data, while still compressing fast, you might look at first -// using BMDiff and then compressing the output of BMDiff with -// Snappy. - -#ifndef THIRD_PARTY_SNAPPY_SNAPPY_H__ -#define THIRD_PARTY_SNAPPY_SNAPPY_H__ - -#include -#include - -#include -#include - -#include "snappy-stubs-public.h" - -namespace snappy { - class Source; - class Sink; - - // ------------------------------------------------------------------------ - // Generic compression/decompression routines. - // ------------------------------------------------------------------------ - - // Compress the bytes read from "*source" and append to "*sink". Return the - // number of bytes written. - size_t Compress(Source* source, Sink* sink); - - // Find the uncompressed length of the given stream, as given by the header. - // Note that the true length could deviate from this; the stream could e.g. - // be truncated. - // - // Also note that this leaves "*source" in a state that is unsuitable for - // further operations, such as RawUncompress(). You will need to rewind - // or recreate the source yourself before attempting any further calls. - bool GetUncompressedLength(Source* source, uint32_t* result); - - // ------------------------------------------------------------------------ - // Higher-level string based routines (should be sufficient for most users) - // ------------------------------------------------------------------------ - - // Sets "*compressed" to the compressed version of "input[0..input_length-1]". - // Original contents of *compressed are lost. - // - // REQUIRES: "input[]" is not an alias of "*compressed". - size_t Compress(const char* input, size_t input_length, - std::string* compressed); - - // Same as `Compress` above but taking an `iovec` array as input. Note that - // this function preprocesses the inputs to compute the sum of - // `iov[0..iov_cnt-1].iov_len` before reading. To avoid this, use - // `RawCompressFromIOVec` below. - size_t CompressFromIOVec(const struct iovec* iov, size_t iov_cnt, - std::string* compressed); - - // Decompresses "compressed[0..compressed_length-1]" to "*uncompressed". - // Original contents of "*uncompressed" are lost. - // - // REQUIRES: "compressed[]" is not an alias of "*uncompressed". - // - // returns false if the message is corrupted and could not be decompressed - bool Uncompress(const char* compressed, size_t compressed_length, - std::string* uncompressed); - - // Decompresses "compressed" to "*uncompressed". - // - // returns false if the message is corrupted and could not be decompressed - bool Uncompress(Source* compressed, Sink* uncompressed); - - // This routine uncompresses as much of the "compressed" as possible - // into sink. It returns the number of valid bytes added to sink - // (extra invalid bytes may have been added due to errors; the caller - // should ignore those). The emitted data typically has length - // GetUncompressedLength(), but may be shorter if an error is - // encountered. - size_t UncompressAsMuchAsPossible(Source* compressed, Sink* uncompressed); - - // ------------------------------------------------------------------------ - // Lower-level character array based routines. May be useful for - // efficiency reasons in certain circumstances. - // ------------------------------------------------------------------------ - - // REQUIRES: "compressed" must point to an area of memory that is at - // least "MaxCompressedLength(input_length)" bytes in length. - // - // Takes the data stored in "input[0..input_length]" and stores - // it in the array pointed to by "compressed". - // - // "*compressed_length" is set to the length of the compressed output. - // - // Example: - // char* output = new char[snappy::MaxCompressedLength(input_length)]; - // size_t output_length; - // RawCompress(input, input_length, output, &output_length); - // ... Process(output, output_length) ... - // delete [] output; - void RawCompress(const char* input, - size_t input_length, - char* compressed, - size_t* compressed_length); - - // Same as `RawCompress` above but taking an `iovec` array as input. Note that - // `uncompressed_length` is the total number of bytes to be read from the - // elements of `iov` (_not_ the number of elements in `iov`). - void RawCompressFromIOVec(const struct iovec* iov, size_t uncompressed_length, - char* compressed, size_t* compressed_length); - - // Given data in "compressed[0..compressed_length-1]" generated by - // calling the Snappy::Compress routine, this routine - // stores the uncompressed data to - // uncompressed[0..GetUncompressedLength(compressed)-1] - // returns false if the message is corrupted and could not be decrypted - bool RawUncompress(const char* compressed, size_t compressed_length, - char* uncompressed); - - // Given data from the byte source 'compressed' generated by calling - // the Snappy::Compress routine, this routine stores the uncompressed - // data to - // uncompressed[0..GetUncompressedLength(compressed,compressed_length)-1] - // returns false if the message is corrupted and could not be decrypted - bool RawUncompress(Source* compressed, char* uncompressed); - - // Given data in "compressed[0..compressed_length-1]" generated by - // calling the Snappy::Compress routine, this routine - // stores the uncompressed data to the iovec "iov". The number of physical - // buffers in "iov" is given by iov_cnt and their cumulative size - // must be at least GetUncompressedLength(compressed). The individual buffers - // in "iov" must not overlap with each other. - // - // returns false if the message is corrupted and could not be decrypted - bool RawUncompressToIOVec(const char* compressed, size_t compressed_length, - const struct iovec* iov, size_t iov_cnt); - - // Given data from the byte source 'compressed' generated by calling - // the Snappy::Compress routine, this routine stores the uncompressed - // data to the iovec "iov". The number of physical - // buffers in "iov" is given by iov_cnt and their cumulative size - // must be at least GetUncompressedLength(compressed). The individual buffers - // in "iov" must not overlap with each other. - // - // returns false if the message is corrupted and could not be decrypted - bool RawUncompressToIOVec(Source* compressed, const struct iovec* iov, - size_t iov_cnt); - - // Returns the maximal size of the compressed representation of - // input data that is "source_bytes" bytes in length; - size_t MaxCompressedLength(size_t source_bytes); - - // REQUIRES: "compressed[]" was produced by RawCompress() or Compress() - // Returns true and stores the length of the uncompressed data in - // *result normally. Returns false on parsing error. - // This operation takes O(1) time. - bool GetUncompressedLength(const char* compressed, size_t compressed_length, - size_t* result); - - // Returns true iff the contents of "compressed[]" can be uncompressed - // successfully. Does not return the uncompressed data. Takes - // time proportional to compressed_length, but is usually at least - // a factor of four faster than actual decompression. - bool IsValidCompressedBuffer(const char* compressed, - size_t compressed_length); - - // Returns true iff the contents of "compressed" can be uncompressed - // successfully. Does not return the uncompressed data. Takes - // time proportional to *compressed length, but is usually at least - // a factor of four faster than actual decompression. - // On success, consumes all of *compressed. On failure, consumes an - // unspecified prefix of *compressed. - bool IsValidCompressed(Source* compressed); - - // The size of a compression block. Note that many parts of the compression - // code assumes that kBlockSize <= 65536; in particular, the hash table - // can only store 16-bit offsets, and EmitCopy() also assumes the offset - // is 65535 bytes or less. Note also that if you change this, it will - // affect the framing format (see framing_format.txt). - // - // Note that there might be older data around that is compressed with larger - // block sizes, so the decompression code should not rely on the - // non-existence of long backreferences. - static constexpr int kBlockLog = 16; - static constexpr size_t kBlockSize = 1 << kBlockLog; - - static constexpr int kMinHashTableBits = 8; - static constexpr size_t kMinHashTableSize = 1 << kMinHashTableBits; - - static constexpr int kMaxHashTableBits = 14; - static constexpr size_t kMaxHashTableSize = 1 << kMaxHashTableBits; -} // end namespace snappy - -#endif // THIRD_PARTY_SNAPPY_SNAPPY_H__ diff --git a/contrib/libs/snappy/ya.make b/contrib/libs/snappy/ya.make deleted file mode 100644 index f567bd1e9b8..00000000000 --- a/contrib/libs/snappy/ya.make +++ /dev/null @@ -1,34 +0,0 @@ -# Generated by devtools/yamaker from nixpkgs 22.11. - -LIBRARY() - -LICENSE(BSD-3-Clause) - -LICENSE_TEXTS(.yandex_meta/licenses.list.txt) - -VERSION(1.1.10) - -ORIGINAL_SOURCE(https://github.com/google/snappy/archive/1.1.10.tar.gz) - -PEERDIR( - library/cpp/sanitizer/include -) - -ADDINCL( - GLOBAL contrib/libs/snappy/include -) - -NO_COMPILER_WARNINGS() - -CFLAGS( - -DHAVE_CONFIG_H -) - -SRCS( - snappy-c.cc - snappy-sinksource.cc - snappy-stubs-internal.cc - snappy.cc -) - -END() diff --git a/library/cpp/blockcodecs/codecs/snappy/CMakeLists.txt b/library/cpp/blockcodecs/codecs/snappy/CMakeLists.txt index ccaa2bd5bad..1a26a18196b 100644 --- a/library/cpp/blockcodecs/codecs/snappy/CMakeLists.txt +++ b/library/cpp/blockcodecs/codecs/snappy/CMakeLists.txt @@ -2,7 +2,7 @@ add_library(blockcodecs-codecs-snappy INTERFACE) target_link_libraries(blockcodecs-codecs-snappy INTERFACE yutil - contrib-libs-snappy + Snappy::snappy cpp-blockcodecs-core ) @@ -10,7 +10,7 @@ add_global_library_for(blockcodecs-codecs-snappy.global blockcodecs-codecs-snapp target_link_libraries(blockcodecs-codecs-snappy.global PUBLIC yutil - contrib-libs-snappy + Snappy::snappy cpp-blockcodecs-core ) diff --git a/library/cpp/blockcodecs/codecs/snappy/snappy.cpp b/library/cpp/blockcodecs/codecs/snappy/snappy.cpp index 5e79b667c9f..927339ae3bf 100644 --- a/library/cpp/blockcodecs/codecs/snappy/snappy.cpp +++ b/library/cpp/blockcodecs/codecs/snappy/snappy.cpp @@ -2,7 +2,7 @@ #include #include -#include +#include using namespace NBlockCodecs;