Skip to content

Instantly share code, notes, and snippets.

@thoughtpolice

thoughtpolice/zdict.h

Created October 8, 2023 19:51
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save thoughtpolice/bb26bee5c02e8986760210b291e43288 to your computer and use it in GitHub Desktop.
Save thoughtpolice/bb26bee5c02e8986760210b291e43288 to your computer and use it in GitHub Desktop.
Download ZIP
zstd single file release: v1.5.5
Raw
zdict.h
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#if defined (__cplusplus)
extern "C" {
#endif
#ifndef ZSTD_ZDICT_H
#define ZSTD_ZDICT_H
/*====== Dependencies ======*/
#include <stddef.h> /* size_t */
/* ===== ZDICTLIB_API : control library symbols visibility ===== */
#ifndef ZDICTLIB_VISIBLE
/* Backwards compatibility with old macro name */
# ifdef ZDICTLIB_VISIBILITY
# define ZDICTLIB_VISIBLE ZDICTLIB_VISIBILITY
# elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
# define ZDICTLIB_VISIBLE __attribute__ ((visibility ("default")))
# else
# define ZDICTLIB_VISIBLE
# endif
#endif
#ifndef ZDICTLIB_HIDDEN
# if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
# define ZDICTLIB_HIDDEN __attribute__ ((visibility ("hidden")))
# else
# define ZDICTLIB_HIDDEN
# endif
#endif
#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
# define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBLE
#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
# define ZDICTLIB_API __declspec(dllimport) ZDICTLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
#else
# define ZDICTLIB_API ZDICTLIB_VISIBLE
#endif
/*******************************************************************************
* Zstd dictionary builder
*
* FAQ
* ===
* Why should I use a dictionary?
* ------------------------------
*
* Zstd can use dictionaries to improve compression ratio of small data.
* Traditionally small files don't compress well because there is very little
* repetition in a single sample, since it is small. But, if you are compressing
* many similar files, like a bunch of JSON records that share the same
* structure, you can train a dictionary on ahead of time on some samples of
* these files. Then, zstd can use the dictionary to find repetitions that are
* present across samples. This can vastly improve compression ratio.
*
* When is a dictionary useful?
* ----------------------------
*
* Dictionaries are useful when compressing many small files that are similar.
* The larger a file is, the less benefit a dictionary will have. Generally,
* we don't expect dictionary compression to be effective past 100KB. And the
* smaller a file is, the more we would expect the dictionary to help.
*
* How do I use a dictionary?
* --------------------------
*
* Simply pass the dictionary to the zstd compressor with
* `ZSTD_CCtx_loadDictionary()`. The same dictionary must then be passed to
* the decompressor, using `ZSTD_DCtx_loadDictionary()`. There are other
* more advanced functions that allow selecting some options, see zstd.h for
* complete documentation.
*
* What is a zstd dictionary?
* --------------------------
*
* A zstd dictionary has two pieces: Its header, and its content. The header
* contains a magic number, the dictionary ID, and entropy tables. These
* entropy tables allow zstd to save on header costs in the compressed file,
* which really matters for small data. The content is just bytes, which are
* repeated content that is common across many samples.
*
* What is a raw content dictionary?
* ---------------------------------
*
* A raw content dictionary is just bytes. It doesn't have a zstd dictionary
* header, a dictionary ID, or entropy tables. Any buffer is a valid raw
* content dictionary.
*
* How do I train a dictionary?
* ----------------------------
*
* Gather samples from your use case. These samples should be similar to each
* other. If you have several use cases, you could try to train one dictionary
* per use case.
*
* Pass those samples to `ZDICT_trainFromBuffer()` and that will train your
* dictionary. There are a few advanced versions of this function, but this
* is a great starting point. If you want to further tune your dictionary
* you could try `ZDICT_optimizeTrainFromBuffer_cover()`. If that is too slow
* you can try `ZDICT_optimizeTrainFromBuffer_fastCover()`.
*
* If the dictionary training function fails, that is likely because you
* either passed too few samples, or a dictionary would not be effective
* for your data. Look at the messages that the dictionary trainer printed,
* if it doesn't say too few samples, then a dictionary would not be effective.
*
* How large should my dictionary be?
* ----------------------------------
*
* A reasonable dictionary size, the `dictBufferCapacity`, is about 100KB.
* The zstd CLI defaults to a 110KB dictionary. You likely don't need a
* dictionary larger than that. But, most use cases can get away with a
* smaller dictionary. The advanced dictionary builders can automatically
* shrink the dictionary for you, and select the smallest size that doesn't
* hurt compression ratio too much. See the `shrinkDict` parameter.
* A smaller dictionary can save memory, and potentially speed up
* compression.
*
* How many samples should I provide to the dictionary builder?
* ------------------------------------------------------------
*
* We generally recommend passing ~100x the size of the dictionary
* in samples. A few thousand should suffice. Having too few samples
* can hurt the dictionaries effectiveness. Having more samples will
* only improve the dictionaries effectiveness. But having too many
* samples can slow down the dictionary builder.
*
* How do I determine if a dictionary will be effective?
* -----------------------------------------------------
*
* Simply train a dictionary and try it out. You can use zstd's built in
* benchmarking tool to test the dictionary effectiveness.
*
* # Benchmark levels 1-3 without a dictionary
* zstd -b1e3 -r /path/to/my/files
* # Benchmark levels 1-3 with a dictionary
* zstd -b1e3 -r /path/to/my/files -D /path/to/my/dictionary
*
* When should I retrain a dictionary?
* -----------------------------------
*
* You should retrain a dictionary when its effectiveness drops. Dictionary
* effectiveness drops as the data you are compressing changes. Generally, we do
* expect dictionaries to "decay" over time, as your data changes, but the rate
* at which they decay depends on your use case. Internally, we regularly
* retrain dictionaries, and if the new dictionary performs significantly
* better than the old dictionary, we will ship the new dictionary.
*
* I have a raw content dictionary, how do I turn it into a zstd dictionary?
* -------------------------------------------------------------------------
*
* If you have a raw content dictionary, e.g. by manually constructing it, or
* using a third-party dictionary builder, you can turn it into a zstd
* dictionary by using `ZDICT_finalizeDictionary()`. You'll also have to
* provide some samples of the data. It will add the zstd header to the
* raw content, which contains a dictionary ID and entropy tables, which
* will improve compression ratio, and allow zstd to write the dictionary ID
* into the frame, if you so choose.
*
* Do I have to use zstd's dictionary builder?
* -------------------------------------------
*
* No! You can construct dictionary content however you please, it is just
* bytes. It will always be valid as a raw content dictionary. If you want
* a zstd dictionary, which can improve compression ratio, use
* `ZDICT_finalizeDictionary()`.
*
* What is the attack surface of a zstd dictionary?
* ------------------------------------------------
*
* Zstd is heavily fuzz tested, including loading fuzzed dictionaries, so
* zstd should never crash, or access out-of-bounds memory no matter what
* the dictionary is. However, if an attacker can control the dictionary
* during decompression, they can cause zstd to generate arbitrary bytes,
* just like if they controlled the compressed data.
*
******************************************************************************/
/*! ZDICT_trainFromBuffer():
* Train a dictionary from an array of samples.
* Redirect towards ZDICT_optimizeTrainFromBuffer_fastCover() single-threaded, with d=8, steps=4,
* f=20, and accel=1.
* Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
* supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
* The resulting dictionary will be saved into `dictBuffer`.
* @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
* or an error code, which can be tested with ZDICT_isError().
* Note: Dictionary training will fail if there are not enough samples to construct a
* dictionary, or if most of the samples are too small (< 8 bytes being the lower limit).
* If dictionary training fails, you should use zstd without a dictionary, as the dictionary
* would've been ineffective anyways. If you believe your samples would benefit from a dictionary
* please open an issue with details, and we can look into it.
* Note: ZDICT_trainFromBuffer()'s memory usage is about 6 MB.
* Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
* It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, though this can vary a lot.
* It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
*/
ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
const void* samplesBuffer,
const size_t* samplesSizes, unsigned nbSamples);
typedef struct {
int compressionLevel; /**< optimize for a specific zstd compression level; 0 means default */
unsigned notificationLevel; /**< Write log to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
unsigned dictID; /**< force dictID value; 0 means auto mode (32-bits random value)
* NOTE: The zstd format reserves some dictionary IDs for future use.
* You may use them in private settings, but be warned that they
* may be used by zstd in a public dictionary registry in the future.
* These dictionary IDs are:
* - low range : <= 32767
* - high range : >= (2^31)
*/
} ZDICT_params_t;
/*! ZDICT_finalizeDictionary():
* Given a custom content as a basis for dictionary, and a set of samples,
* finalize dictionary by adding headers and statistics according to the zstd
* dictionary format.
*
* Samples must be stored concatenated in a flat buffer `samplesBuffer`,
* supplied with an array of sizes `samplesSizes`, providing the size of each
* sample in order. The samples are used to construct the statistics, so they
* should be representative of what you will compress with this dictionary.
*
* The compression level can be set in `parameters`. You should pass the
* compression level you expect to use in production. The statistics for each
* compression level differ, so tuning the dictionary for the compression level
* can help quite a bit.
*
* You can set an explicit dictionary ID in `parameters`, or allow us to pick
* a random dictionary ID for you, but we can't guarantee no collisions.
*
* The dstDictBuffer and the dictContent may overlap, and the content will be
* appended to the end of the header. If the header + the content doesn't fit in
* maxDictSize the beginning of the content is truncated to make room, since it
* is presumed that the most profitable content is at the end of the dictionary,
* since that is the cheapest to reference.
*
* `maxDictSize` must be >= max(dictContentSize, ZSTD_DICTSIZE_MIN).
*
* @return: size of dictionary stored into `dstDictBuffer` (<= `maxDictSize`),
* or an error code, which can be tested by ZDICT_isError().
* Note: ZDICT_finalizeDictionary() will push notifications into stderr if
* instructed to, using notificationLevel>0.
* NOTE: This function currently may fail in several edge cases including:
* * Not enough samples
* * Samples are uncompressible
* * Samples are all exactly the same
*/
ZDICTLIB_API size_t ZDICT_finalizeDictionary(void* dstDictBuffer, size_t maxDictSize,
const void* dictContent, size_t dictContentSize,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_params_t parameters);
/*====== Helper functions ======*/
ZDICTLIB_API unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize); /**< extracts dictID; @return zero if error (not a valid dictionary) */
ZDICTLIB_API size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize); /* returns dict header size; returns a ZSTD error code on failure */
ZDICTLIB_API unsigned ZDICT_isError(size_t errorCode);
ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode);
#endif /* ZSTD_ZDICT_H */
#if defined(ZDICT_STATIC_LINKING_ONLY) && !defined(ZSTD_ZDICT_H_STATIC)
#define ZSTD_ZDICT_H_STATIC
/* This can be overridden externally to hide static symbols. */
#ifndef ZDICTLIB_STATIC_API
# if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
# define ZDICTLIB_STATIC_API __declspec(dllexport) ZDICTLIB_VISIBLE
# elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
# define ZDICTLIB_STATIC_API __declspec(dllimport) ZDICTLIB_VISIBLE
# else
# define ZDICTLIB_STATIC_API ZDICTLIB_VISIBLE
# endif
#endif
/* ====================================================================================
* The definitions in this section are considered experimental.
* They should never be used with a dynamic library, as they may change in the future.
* They are provided for advanced usages.
* Use them only in association with static linking.
* ==================================================================================== */
#define ZDICT_DICTSIZE_MIN 256
/* Deprecated: Remove in v1.6.0 */
#define ZDICT_CONTENTSIZE_MIN 128
/*! ZDICT_cover_params_t:
* k and d are the only required parameters.
* For others, value 0 means default.
*/
typedef struct {
unsigned k; /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
unsigned d; /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
unsigned steps; /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
unsigned nbThreads; /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
double splitPoint; /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (1.0), 1.0 when all samples are used for both training and testing */
unsigned shrinkDict; /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking */
unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */
ZDICT_params_t zParams;
} ZDICT_cover_params_t;
typedef struct {
unsigned k; /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
unsigned d; /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
unsigned f; /* log of size of frequency array : constraint: 0 < f <= 31 : 1 means default(20)*/
unsigned steps; /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */
unsigned nbThreads; /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
double splitPoint; /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (0.75), 1.0 when all samples are used for both training and testing */
unsigned accel; /* Acceleration level: constraint: 0 < accel <= 10, higher means faster and less accurate, 0 means default(1) */
unsigned shrinkDict; /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking */
unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */
ZDICT_params_t zParams;
} ZDICT_fastCover_params_t;
/*! ZDICT_trainFromBuffer_cover():
* Train a dictionary from an array of samples using the COVER algorithm.
* Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
* supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
* The resulting dictionary will be saved into `dictBuffer`.
* @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
* or an error code, which can be tested with ZDICT_isError().
* See ZDICT_trainFromBuffer() for details on failure modes.
* Note: ZDICT_trainFromBuffer_cover() requires about 9 bytes of memory for each input byte.
* Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
* It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, though this can vary a lot.
* It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
*/
ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_cover(
void *dictBuffer, size_t dictBufferCapacity,
const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
ZDICT_cover_params_t parameters);
/*! ZDICT_optimizeTrainFromBuffer_cover():
* The same requirements as above hold for all the parameters except `parameters`.
* This function tries many parameter combinations and picks the best parameters.
* `*parameters` is filled with the best parameters found,
* dictionary constructed with those parameters is stored in `dictBuffer`.
*
* All of the parameters d, k, steps are optional.
* If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
* if steps is zero it defaults to its default value.
* If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
*
* @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
* or an error code, which can be tested with ZDICT_isError().
* On success `*parameters` contains the parameters selected.
* See ZDICT_trainFromBuffer() for details on failure modes.
* Note: ZDICT_optimizeTrainFromBuffer_cover() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread.
*/
ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_cover(
void* dictBuffer, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_cover_params_t* parameters);
/*! ZDICT_trainFromBuffer_fastCover():
* Train a dictionary from an array of samples using a modified version of COVER algorithm.
* Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
* supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
* d and k are required.
* All other parameters are optional, will use default values if not provided
* The resulting dictionary will be saved into `dictBuffer`.
* @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
* or an error code, which can be tested with ZDICT_isError().
* See ZDICT_trainFromBuffer() for details on failure modes.
* Note: ZDICT_trainFromBuffer_fastCover() requires 6 * 2^f bytes of memory.
* Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
* It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, though this can vary a lot.
* It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
*/
ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_fastCover(void *dictBuffer,
size_t dictBufferCapacity, const void *samplesBuffer,
const size_t *samplesSizes, unsigned nbSamples,
ZDICT_fastCover_params_t parameters);
/*! ZDICT_optimizeTrainFromBuffer_fastCover():
* The same requirements as above hold for all the parameters except `parameters`.
* This function tries many parameter combinations (specifically, k and d combinations)
* and picks the best parameters. `*parameters` is filled with the best parameters found,
* dictionary constructed with those parameters is stored in `dictBuffer`.
* All of the parameters d, k, steps, f, and accel are optional.
* If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8}.
* if steps is zero it defaults to its default value.
* If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000].
* If f is zero, default value of 20 is used.
* If accel is zero, default value of 1 is used.
*
* @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
* or an error code, which can be tested with ZDICT_isError().
* On success `*parameters` contains the parameters selected.
* See ZDICT_trainFromBuffer() for details on failure modes.
* Note: ZDICT_optimizeTrainFromBuffer_fastCover() requires about 6 * 2^f bytes of memory for each thread.
*/
ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(void* dictBuffer,
size_t dictBufferCapacity, const void* samplesBuffer,
const size_t* samplesSizes, unsigned nbSamples,
ZDICT_fastCover_params_t* parameters);
typedef struct {
unsigned selectivityLevel; /* 0 means default; larger => select more => larger dictionary */
ZDICT_params_t zParams;
} ZDICT_legacy_params_t;
/*! ZDICT_trainFromBuffer_legacy():
* Train a dictionary from an array of samples.
* Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
* supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
* The resulting dictionary will be saved into `dictBuffer`.
* `parameters` is optional and can be provided with values set to 0 to mean "default".
* @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
* or an error code, which can be tested with ZDICT_isError().
* See ZDICT_trainFromBuffer() for details on failure modes.
* Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
* It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
* In general, it's recommended to provide a few thousands samples, though this can vary a lot.
* It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
* Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0.
*/
ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_legacy(
void* dictBuffer, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_legacy_params_t parameters);
/* Deprecation warnings */
/* It is generally possible to disable deprecation warnings from compiler,
for example with -Wno-deprecated-declarations for gcc
or _CRT_SECURE_NO_WARNINGS in Visual.
Otherwise, it's also possible to manually define ZDICT_DISABLE_DEPRECATE_WARNINGS */
#ifdef ZDICT_DISABLE_DEPRECATE_WARNINGS
# define ZDICT_DEPRECATED(message) /* disable deprecation warnings */
#else
# define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
# if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
# define ZDICT_DEPRECATED(message) [[deprecated(message)]]
# elif defined(__clang__) || (ZDICT_GCC_VERSION >= 405)
# define ZDICT_DEPRECATED(message) __attribute__((deprecated(message)))
# elif (ZDICT_GCC_VERSION >= 301)
# define ZDICT_DEPRECATED(message) __attribute__((deprecated))
# elif defined(_MSC_VER)
# define ZDICT_DEPRECATED(message) __declspec(deprecated(message))
# else
# pragma message("WARNING: You need to implement ZDICT_DEPRECATED for this compiler")
# define ZDICT_DEPRECATED(message)
# endif
#endif /* ZDICT_DISABLE_DEPRECATE_WARNINGS */
ZDICT_DEPRECATED("use ZDICT_finalizeDictionary() instead")
ZDICTLIB_STATIC_API
size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
#endif /* ZSTD_ZDICT_H_STATIC */
#if defined (__cplusplus)
}
#endif
Raw
zstd.c
This file has been truncated, but you can view the full file.
/**
* \file zstd.c
* Single-file Zstandard library.
*
* Generate using:
* \code
* python combine.py -r ../../lib -x legacy/zstd_legacy.h -o zstd.c zstd-in.c
* \endcode
*/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/*
* Settings to bake for the single library file.
*
* Note: It's important that none of these affects 'zstd.h' (only the
* implementation files we're amalgamating).
*
* Note: MEM_MODULE stops xxhash redefining BYTE, U16, etc., which are also
* defined in mem.h (breaking C99 compatibility).
*
* Note: the undefs for xxHash allow Zstd's implementation to coincide with
* standalone xxHash usage (with global defines).
*
* Note: if you enable ZSTD_LEGACY_SUPPORT the combine.py script will need
* re-running without the "-x legacy/zstd_legacy.h" option (it excludes the
* legacy support at the source level).
*
* Note: multithreading is enabled for all platforms apart from Emscripten.
*/
#define DEBUGLEVEL 0
#define MEM_MODULE
#undef XXH_NAMESPACE
#define XXH_NAMESPACE ZSTD_
#undef XXH_PRIVATE_API
#define XXH_PRIVATE_API
#undef XXH_INLINE_ALL
#define XXH_INLINE_ALL
#define ZSTD_LEGACY_SUPPORT 0
#ifndef __EMSCRIPTEN__
#define ZSTD_MULTITHREAD
#endif
#define ZSTD_TRACE 0
/* TODO: Can't amalgamate ASM function */
#define ZSTD_DISABLE_ASM 1
/* Include zstd_deps.h first with all the options we need enabled. */
#define ZSTD_DEPS_NEED_MALLOC
#define ZSTD_DEPS_NEED_MATH64
/**** start inlining common/zstd_deps.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/* This file provides common libc dependencies that zstd requires.
* The purpose is to allow replacing this file with a custom implementation
* to compile zstd without libc support.
*/
/* Need:
* NULL
* INT_MAX
* UINT_MAX
* ZSTD_memcpy()
* ZSTD_memset()
* ZSTD_memmove()
*/
#ifndef ZSTD_DEPS_COMMON
#define ZSTD_DEPS_COMMON
#include <limits.h>
#include <stddef.h>
#include <string.h>
#if defined(__GNUC__) && __GNUC__ >= 4
# define ZSTD_memcpy(d,s,l) __builtin_memcpy((d),(s),(l))
# define ZSTD_memmove(d,s,l) __builtin_memmove((d),(s),(l))
# define ZSTD_memset(p,v,l) __builtin_memset((p),(v),(l))
#else
# define ZSTD_memcpy(d,s,l) memcpy((d),(s),(l))
# define ZSTD_memmove(d,s,l) memmove((d),(s),(l))
# define ZSTD_memset(p,v,l) memset((p),(v),(l))
#endif
#endif /* ZSTD_DEPS_COMMON */
/* Need:
* ZSTD_malloc()
* ZSTD_free()
* ZSTD_calloc()
*/
#ifdef ZSTD_DEPS_NEED_MALLOC
#ifndef ZSTD_DEPS_MALLOC
#define ZSTD_DEPS_MALLOC
#include <stdlib.h>
#define ZSTD_malloc(s) malloc(s)
#define ZSTD_calloc(n,s) calloc((n), (s))
#define ZSTD_free(p) free((p))
#endif /* ZSTD_DEPS_MALLOC */
#endif /* ZSTD_DEPS_NEED_MALLOC */
/*
* Provides 64-bit math support.
* Need:
* U64 ZSTD_div64(U64 dividend, U32 divisor)
*/
#ifdef ZSTD_DEPS_NEED_MATH64
#ifndef ZSTD_DEPS_MATH64
#define ZSTD_DEPS_MATH64
#define ZSTD_div64(dividend, divisor) ((dividend) / (divisor))
#endif /* ZSTD_DEPS_MATH64 */
#endif /* ZSTD_DEPS_NEED_MATH64 */
/* Need:
* assert()
*/
#ifdef ZSTD_DEPS_NEED_ASSERT
#ifndef ZSTD_DEPS_ASSERT
#define ZSTD_DEPS_ASSERT
#include <assert.h>
#endif /* ZSTD_DEPS_ASSERT */
#endif /* ZSTD_DEPS_NEED_ASSERT */
/* Need:
* ZSTD_DEBUG_PRINT()
*/
#ifdef ZSTD_DEPS_NEED_IO
#ifndef ZSTD_DEPS_IO
#define ZSTD_DEPS_IO
#include <stdio.h>
#define ZSTD_DEBUG_PRINT(...) fprintf(stderr, __VA_ARGS__)
#endif /* ZSTD_DEPS_IO */
#endif /* ZSTD_DEPS_NEED_IO */
/* Only requested when <stdint.h> is known to be present.
* Need:
* intptr_t
*/
#ifdef ZSTD_DEPS_NEED_STDINT
#ifndef ZSTD_DEPS_STDINT
#define ZSTD_DEPS_STDINT
#include <stdint.h>
#endif /* ZSTD_DEPS_STDINT */
#endif /* ZSTD_DEPS_NEED_STDINT */
/**** ended inlining common/zstd_deps.h ****/
/**** start inlining common/debug.c ****/
/* ******************************************************************
* debug
* Part of FSE library
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/*
* This module only hosts one global variable
* which can be used to dynamically influence the verbosity of traces,
* such as DEBUGLOG and RAWLOG
*/
/**** start inlining debug.h ****/
/* ******************************************************************
* debug
* Part of FSE library
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/*
* The purpose of this header is to enable debug functions.
* They regroup assert(), DEBUGLOG() and RAWLOG() for run-time,
* and DEBUG_STATIC_ASSERT() for compile-time.
*
* By default, DEBUGLEVEL==0, which means run-time debug is disabled.
*
* Level 1 enables assert() only.
* Starting level 2, traces can be generated and pushed to stderr.
* The higher the level, the more verbose the traces.
*
* It's possible to dynamically adjust level using variable g_debug_level,
* which is only declared if DEBUGLEVEL>=2,
* and is a global variable, not multi-thread protected (use with care)
*/
#ifndef DEBUG_H_12987983217
#define DEBUG_H_12987983217
#if defined (__cplusplus)
extern "C" {
#endif
/* static assert is triggered at compile time, leaving no runtime artefact.
* static assert only works with compile-time constants.
* Also, this variant can only be used inside a function. */
#define DEBUG_STATIC_ASSERT(c) (void)sizeof(char[(c) ? 1 : -1])
/* DEBUGLEVEL is expected to be defined externally,
* typically through compiler command line.
* Value must be a number. */
#ifndef DEBUGLEVEL
# define DEBUGLEVEL 0
#endif
/* recommended values for DEBUGLEVEL :
* 0 : release mode, no debug, all run-time checks disabled
* 1 : enables assert() only, no display
* 2 : reserved, for currently active debug path
* 3 : events once per object lifetime (CCtx, CDict, etc.)
* 4 : events once per frame
* 5 : events once per block
* 6 : events once per sequence (verbose)
* 7+: events at every position (*very* verbose)
*
* It's generally inconvenient to output traces > 5.
* In which case, it's possible to selectively trigger high verbosity levels
* by modifying g_debug_level.
*/
#if (DEBUGLEVEL>=1)
# define ZSTD_DEPS_NEED_ASSERT
/**** skipping file: zstd_deps.h ****/
#else
# ifndef assert /* assert may be already defined, due to prior #include <assert.h> */
# define assert(condition) ((void)0) /* disable assert (default) */
# endif
#endif
#if (DEBUGLEVEL>=2)
# define ZSTD_DEPS_NEED_IO
/**** skipping file: zstd_deps.h ****/
extern int g_debuglevel; /* the variable is only declared,
it actually lives in debug.c,
and is shared by the whole process.
It's not thread-safe.
It's useful when enabling very verbose levels
on selective conditions (such as position in src) */
# define RAWLOG(l, ...) { \
if (l<=g_debuglevel) { \
ZSTD_DEBUG_PRINT(__VA_ARGS__); \
} }
# define DEBUGLOG(l, ...) { \
if (l<=g_debuglevel) { \
ZSTD_DEBUG_PRINT(__FILE__ ": " __VA_ARGS__); \
ZSTD_DEBUG_PRINT(" \n"); \
} }
#else
# define RAWLOG(l, ...) {} /* disabled */
# define DEBUGLOG(l, ...) {} /* disabled */
#endif
#if defined (__cplusplus)
}
#endif
#endif /* DEBUG_H_12987983217 */
/**** ended inlining debug.h ****/
int g_debuglevel = DEBUGLEVEL;
/**** ended inlining common/debug.c ****/
/**** start inlining common/entropy_common.c ****/
/* ******************************************************************
* Common functions of New Generation Entropy library
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
* - Public forum : https://groups.google.com/forum/#!forum/lz4c
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/* *************************************
* Dependencies
***************************************/
/**** start inlining mem.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef MEM_H_MODULE
#define MEM_H_MODULE
#if defined (__cplusplus)
extern "C" {
#endif
/*-****************************************
* Dependencies
******************************************/
#include <stddef.h> /* size_t, ptrdiff_t */
/**** start inlining compiler.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_COMPILER_H
#define ZSTD_COMPILER_H
/**** start inlining portability_macros.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_PORTABILITY_MACROS_H
#define ZSTD_PORTABILITY_MACROS_H
/**
* This header file contains macro definitions to support portability.
* This header is shared between C and ASM code, so it MUST only
* contain macro definitions. It MUST not contain any C code.
*
* This header ONLY defines macros to detect platforms/feature support.
*
*/
/* compat. with non-clang compilers */
#ifndef __has_attribute
#define __has_attribute(x) 0
#endif
/* compat. with non-clang compilers */
#ifndef __has_builtin
# define __has_builtin(x) 0
#endif
/* compat. with non-clang compilers */
#ifndef __has_feature
# define __has_feature(x) 0
#endif
/* detects whether we are being compiled under msan */
#ifndef ZSTD_MEMORY_SANITIZER
# if __has_feature(memory_sanitizer)
# define ZSTD_MEMORY_SANITIZER 1
# else
# define ZSTD_MEMORY_SANITIZER 0
# endif
#endif
/* detects whether we are being compiled under asan */
#ifndef ZSTD_ADDRESS_SANITIZER
# if __has_feature(address_sanitizer)
# define ZSTD_ADDRESS_SANITIZER 1
# elif defined(__SANITIZE_ADDRESS__)
# define ZSTD_ADDRESS_SANITIZER 1
# else
# define ZSTD_ADDRESS_SANITIZER 0
# endif
#endif
/* detects whether we are being compiled under dfsan */
#ifndef ZSTD_DATAFLOW_SANITIZER
# if __has_feature(dataflow_sanitizer)
# define ZSTD_DATAFLOW_SANITIZER 1
# else
# define ZSTD_DATAFLOW_SANITIZER 0
# endif
#endif
/* Mark the internal assembly functions as hidden */
#ifdef __ELF__
# define ZSTD_HIDE_ASM_FUNCTION(func) .hidden func
#elif defined(__APPLE__)
# define ZSTD_HIDE_ASM_FUNCTION(func) .private_extern func
#else
# define ZSTD_HIDE_ASM_FUNCTION(func)
#endif
/* Enable runtime BMI2 dispatch based on the CPU.
* Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
*/
#ifndef DYNAMIC_BMI2
#if ((defined(__clang__) && __has_attribute(__target__)) \
|| (defined(__GNUC__) \
&& (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \
&& (defined(__x86_64__) || defined(_M_X64)) \
&& !defined(__BMI2__)
# define DYNAMIC_BMI2 1
#else
# define DYNAMIC_BMI2 0
#endif
#endif
/**
* Only enable assembly for GNUC compatible compilers,
* because other platforms may not support GAS assembly syntax.
*
* Only enable assembly for Linux / MacOS, other platforms may
* work, but they haven't been tested. This could likely be
* extended to BSD systems.
*
* Disable assembly when MSAN is enabled, because MSAN requires
* 100% of code to be instrumented to work.
*/
#if defined(__GNUC__)
# if defined(__linux__) || defined(__linux) || defined(__APPLE__)
# if ZSTD_MEMORY_SANITIZER
# define ZSTD_ASM_SUPPORTED 0
# elif ZSTD_DATAFLOW_SANITIZER
# define ZSTD_ASM_SUPPORTED 0
# else
# define ZSTD_ASM_SUPPORTED 1
# endif
# else
# define ZSTD_ASM_SUPPORTED 0
# endif
#else
# define ZSTD_ASM_SUPPORTED 0
#endif
/**
* Determines whether we should enable assembly for x86-64
* with BMI2.
*
* Enable if all of the following conditions hold:
* - ASM hasn't been explicitly disabled by defining ZSTD_DISABLE_ASM
* - Assembly is supported
* - We are compiling for x86-64 and either:
* - DYNAMIC_BMI2 is enabled
* - BMI2 is supported at compile time
*/
#if !defined(ZSTD_DISABLE_ASM) && \
ZSTD_ASM_SUPPORTED && \
defined(__x86_64__) && \
(DYNAMIC_BMI2 || defined(__BMI2__))
# define ZSTD_ENABLE_ASM_X86_64_BMI2 1
#else
# define ZSTD_ENABLE_ASM_X86_64_BMI2 0
#endif
/*
* For x86 ELF targets, add .note.gnu.property section for Intel CET in
* assembly sources when CET is enabled.
*
* Additionally, any function that may be called indirectly must begin
* with ZSTD_CET_ENDBRANCH.
*/
#if defined(__ELF__) && (defined(__x86_64__) || defined(__i386__)) \
&& defined(__has_include)
# if __has_include(<cet.h>)
# include <cet.h>
# define ZSTD_CET_ENDBRANCH _CET_ENDBR
# endif
#endif
#ifndef ZSTD_CET_ENDBRANCH
# define ZSTD_CET_ENDBRANCH
#endif
#endif /* ZSTD_PORTABILITY_MACROS_H */
/**** ended inlining portability_macros.h ****/
/*-*******************************************************
* Compiler specifics
*********************************************************/
/* force inlining */
#if !defined(ZSTD_NO_INLINE)
#if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */
# define INLINE_KEYWORD inline
#else
# define INLINE_KEYWORD
#endif
#if defined(__GNUC__) || defined(__ICCARM__)
# define FORCE_INLINE_ATTR __attribute__((always_inline))
#elif defined(_MSC_VER)
# define FORCE_INLINE_ATTR __forceinline
#else
# define FORCE_INLINE_ATTR
#endif
#else
#define INLINE_KEYWORD
#define FORCE_INLINE_ATTR
#endif
/**
On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC).
This explicitly marks such functions as __cdecl so that the code will still compile
if a CC other than __cdecl has been made the default.
*/
#if defined(_MSC_VER)
# define WIN_CDECL __cdecl
#else
# define WIN_CDECL
#endif
/* UNUSED_ATTR tells the compiler it is okay if the function is unused. */
#if defined(__GNUC__)
# define UNUSED_ATTR __attribute__((unused))
#else
# define UNUSED_ATTR
#endif
/**
* FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
* parameters. They must be inlined for the compiler to eliminate the constant
* branches.
*/
#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR UNUSED_ATTR
/**
* HINT_INLINE is used to help the compiler generate better code. It is *not*
* used for "templates", so it can be tweaked based on the compilers
* performance.
*
* gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the
* always_inline attribute.
*
* clang up to 5.0.0 (trunk) benefit tremendously from the always_inline
* attribute.
*/
#if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5
# define HINT_INLINE static INLINE_KEYWORD
#else
# define HINT_INLINE FORCE_INLINE_TEMPLATE
#endif
/* "soft" inline :
* The compiler is free to select if it's a good idea to inline or not.
* The main objective is to silence compiler warnings
* when a defined function in included but not used.
*
* Note : this macro is prefixed `MEM_` because it used to be provided by `mem.h` unit.
* Updating the prefix is probably preferable, but requires a fairly large codemod,
* since this name is used everywhere.
*/
#ifndef MEM_STATIC /* already defined in Linux Kernel mem.h */
#if defined(__GNUC__)
# define MEM_STATIC static __inline UNUSED_ATTR
#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
# define MEM_STATIC static inline
#elif defined(_MSC_VER)
# define MEM_STATIC static __inline
#else
# define MEM_STATIC static /* this version may generate warnings for unused static functions; disable the relevant warning */
#endif
#endif
/* force no inlining */
#ifdef _MSC_VER
# define FORCE_NOINLINE static __declspec(noinline)
#else
# if defined(__GNUC__) || defined(__ICCARM__)
# define FORCE_NOINLINE static __attribute__((__noinline__))
# else
# define FORCE_NOINLINE static
# endif
#endif
/* target attribute */
#if defined(__GNUC__) || defined(__ICCARM__)
# define TARGET_ATTRIBUTE(target) __attribute__((__target__(target)))
#else
# define TARGET_ATTRIBUTE(target)
#endif
/* Target attribute for BMI2 dynamic dispatch.
* Enable lzcnt, bmi, and bmi2.
* We test for bmi1 & bmi2. lzcnt is included in bmi1.
*/
#define BMI2_TARGET_ATTRIBUTE TARGET_ATTRIBUTE("lzcnt,bmi,bmi2")
/* prefetch
* can be disabled, by declaring NO_PREFETCH build macro */
#if defined(NO_PREFETCH)
# define PREFETCH_L1(ptr) (void)(ptr) /* disabled */
# define PREFETCH_L2(ptr) (void)(ptr) /* disabled */
#else
# if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) && !defined(_M_ARM64EC) /* _mm_prefetch() is not defined outside of x86/x64 */
# include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
# define PREFETCH_L1(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
# define PREFETCH_L2(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T1)
# elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
# define PREFETCH_L1(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
# define PREFETCH_L2(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
# elif defined(__aarch64__)
# define PREFETCH_L1(ptr) __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
# define PREFETCH_L2(ptr) __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
# else
# define PREFETCH_L1(ptr) (void)(ptr) /* disabled */
# define PREFETCH_L2(ptr) (void)(ptr) /* disabled */
# endif
#endif /* NO_PREFETCH */
#define CACHELINE_SIZE 64
#define PREFETCH_AREA(p, s) { \
const char* const _ptr = (const char*)(p); \
size_t const _size = (size_t)(s); \
size_t _pos; \
for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) { \
PREFETCH_L2(_ptr + _pos); \
} \
}
/* vectorization
* older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax,
* and some compilers, like Intel ICC and MCST LCC, do not support it at all. */
#if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__) && !defined(__LCC__)
# if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5)
# define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize")))
# else
# define DONT_VECTORIZE _Pragma("GCC optimize(\"no-tree-vectorize\")")
# endif
#else
# define DONT_VECTORIZE
#endif
/* Tell the compiler that a branch is likely or unlikely.
* Only use these macros if it causes the compiler to generate better code.
* If you can remove a LIKELY/UNLIKELY annotation without speed changes in gcc
* and clang, please do.
*/
#if defined(__GNUC__)
#define LIKELY(x) (__builtin_expect((x), 1))
#define UNLIKELY(x) (__builtin_expect((x), 0))
#else
#define LIKELY(x) (x)
#define UNLIKELY(x) (x)
#endif
#if __has_builtin(__builtin_unreachable) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 5)))
# define ZSTD_UNREACHABLE { assert(0), __builtin_unreachable(); }
#else
# define ZSTD_UNREACHABLE { assert(0); }
#endif
/* disable warnings */
#ifdef _MSC_VER /* Visual Studio */
# include <intrin.h> /* For Visual 2005 */
# pragma warning(disable : 4100) /* disable: C4100: unreferenced formal parameter */
# pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
# pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */
# pragma warning(disable : 4214) /* disable: C4214: non-int bitfields */
# pragma warning(disable : 4324) /* disable: C4324: padded structure */
#endif
/*Like DYNAMIC_BMI2 but for compile time determination of BMI2 support*/
#ifndef STATIC_BMI2
# if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))
# ifdef __AVX2__ //MSVC does not have a BMI2 specific flag, but every CPU that supports AVX2 also supports BMI2
# define STATIC_BMI2 1
# endif
# elif defined(__BMI2__) && defined(__x86_64__) && defined(__GNUC__)
# define STATIC_BMI2 1
# endif
#endif
#ifndef STATIC_BMI2
#define STATIC_BMI2 0
#endif
/* compile time determination of SIMD support */
#if !defined(ZSTD_NO_INTRINSICS)
# if defined(__SSE2__) || defined(_M_AMD64) || (defined (_M_IX86) && defined(_M_IX86_FP) && (_M_IX86_FP >= 2))
# define ZSTD_ARCH_X86_SSE2
# endif
# if defined(__ARM_NEON) || defined(_M_ARM64)
# define ZSTD_ARCH_ARM_NEON
# endif
#
# if defined(ZSTD_ARCH_X86_SSE2)
# include <emmintrin.h>
# elif defined(ZSTD_ARCH_ARM_NEON)
# include <arm_neon.h>
# endif
#endif
/* C-language Attributes are added in C23. */
#if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
# define ZSTD_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
#else
# define ZSTD_HAS_C_ATTRIBUTE(x) 0
#endif
/* Only use C++ attributes in C++. Some compilers report support for C++
* attributes when compiling with C.
*/
#if defined(__cplusplus) && defined(__has_cpp_attribute)
# define ZSTD_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
#else
# define ZSTD_HAS_CPP_ATTRIBUTE(x) 0
#endif
/* Define ZSTD_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute.
* - C23: https://en.cppreference.com/w/c/language/attributes/fallthrough
* - CPP17: https://en.cppreference.com/w/cpp/language/attributes/fallthrough
* - Else: __attribute__((__fallthrough__))
*/
#ifndef ZSTD_FALLTHROUGH
# if ZSTD_HAS_C_ATTRIBUTE(fallthrough)
# define ZSTD_FALLTHROUGH [[fallthrough]]
# elif ZSTD_HAS_CPP_ATTRIBUTE(fallthrough)
# define ZSTD_FALLTHROUGH [[fallthrough]]
# elif __has_attribute(__fallthrough__)
/* Leading semicolon is to satisfy gcc-11 with -pedantic. Without the semicolon
* gcc complains about: a label can only be part of a statement and a declaration is not a statement.
*/
# define ZSTD_FALLTHROUGH ; __attribute__((__fallthrough__))
# else
# define ZSTD_FALLTHROUGH
# endif
#endif
/*-**************************************************************
* Alignment check
*****************************************************************/
/* this test was initially positioned in mem.h,
* but this file is removed (or replaced) for linux kernel
* so it's now hosted in compiler.h,
* which remains valid for both user & kernel spaces.
*/
#ifndef ZSTD_ALIGNOF
# if defined(__GNUC__) || defined(_MSC_VER)
/* covers gcc, clang & MSVC */
/* note : this section must come first, before C11,
* due to a limitation in the kernel source generator */
# define ZSTD_ALIGNOF(T) __alignof(T)
# elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)
/* C11 support */
# include <stdalign.h>
# define ZSTD_ALIGNOF(T) alignof(T)
# else
/* No known support for alignof() - imperfect backup */
# define ZSTD_ALIGNOF(T) (sizeof(void*) < sizeof(T) ? sizeof(void*) : sizeof(T))
# endif
#endif /* ZSTD_ALIGNOF */
/*-**************************************************************
* Sanitizer
*****************************************************************/
/* Issue #3240 reports an ASAN failure on an llvm-mingw build. Out of an
* abundance of caution, disable our custom poisoning on mingw. */
#ifdef __MINGW32__
#ifndef ZSTD_ASAN_DONT_POISON_WORKSPACE
#define ZSTD_ASAN_DONT_POISON_WORKSPACE 1
#endif
#ifndef ZSTD_MSAN_DONT_POISON_WORKSPACE
#define ZSTD_MSAN_DONT_POISON_WORKSPACE 1
#endif
#endif
#if ZSTD_MEMORY_SANITIZER && !defined(ZSTD_MSAN_DONT_POISON_WORKSPACE)
/* Not all platforms that support msan provide sanitizers/msan_interface.h.
* We therefore declare the functions we need ourselves, rather than trying to
* include the header file... */
#include <stddef.h> /* size_t */
#define ZSTD_DEPS_NEED_STDINT
/**** skipping file: zstd_deps.h ****/
/* Make memory region fully initialized (without changing its contents). */
void __msan_unpoison(const volatile void *a, size_t size);
/* Make memory region fully uninitialized (without changing its contents).
This is a legacy interface that does not update origin information. Use
__msan_allocated_memory() instead. */
void __msan_poison(const volatile void *a, size_t size);
/* Returns the offset of the first (at least partially) poisoned byte in the
memory range, or -1 if the whole range is good. */
intptr_t __msan_test_shadow(const volatile void *x, size_t size);
/* Print shadow and origin for the memory range to stderr in a human-readable
format. */
void __msan_print_shadow(const volatile void *x, size_t size);
#endif
#if ZSTD_ADDRESS_SANITIZER && !defined(ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* Not all platforms that support asan provide sanitizers/asan_interface.h.
* We therefore declare the functions we need ourselves, rather than trying to
* include the header file... */
#include <stddef.h> /* size_t */
/**
* Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable.
*
* This memory must be previously allocated by your program. Instrumented
* code is forbidden from accessing addresses in this region until it is
* unpoisoned. This function is not guaranteed to poison the entire region -
* it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan
* alignment restrictions.
*
* \note This function is not thread-safe because no two threads can poison or
* unpoison memory in the same memory region simultaneously.
*
* \param addr Start of memory region.
* \param size Size of memory region. */
void __asan_poison_memory_region(void const volatile *addr, size_t size);
/**
* Marks a memory region (<c>[addr, addr+size)</c>) as addressable.
*
* This memory must be previously allocated by your program. Accessing
* addresses in this region is allowed until this region is poisoned again.
* This function could unpoison a super-region of <c>[addr, addr+size)</c> due
* to ASan alignment restrictions.
*
* \note This function is not thread-safe because no two threads can
* poison or unpoison memory in the same memory region simultaneously.
*
* \param addr Start of memory region.
* \param size Size of memory region. */
void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
#endif
#endif /* ZSTD_COMPILER_H */
/**** ended inlining compiler.h ****/
/**** skipping file: debug.h ****/
/**** skipping file: zstd_deps.h ****/
/*-****************************************
* Compiler specifics
******************************************/
#if defined(_MSC_VER) /* Visual Studio */
# include <stdlib.h> /* _byteswap_ulong */
# include <intrin.h> /* _byteswap_* */
#endif
/*-**************************************************************
* Basic Types
*****************************************************************/
#if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
# if defined(_AIX)
# include <inttypes.h>
# else
# include <stdint.h> /* intptr_t */
# endif
typedef uint8_t BYTE;
typedef uint8_t U8;
typedef int8_t S8;
typedef uint16_t U16;
typedef int16_t S16;
typedef uint32_t U32;
typedef int32_t S32;
typedef uint64_t U64;
typedef int64_t S64;
#else
# include <limits.h>
#if CHAR_BIT != 8
# error "this implementation requires char to be exactly 8-bit type"
#endif
typedef unsigned char BYTE;
typedef unsigned char U8;
typedef signed char S8;
#if USHRT_MAX != 65535
# error "this implementation requires short to be exactly 16-bit type"
#endif
typedef unsigned short U16;
typedef signed short S16;
#if UINT_MAX != 4294967295
# error "this implementation requires int to be exactly 32-bit type"
#endif
typedef unsigned int U32;
typedef signed int S32;
/* note : there are no limits defined for long long type in C90.
* limits exist in C99, however, in such case, <stdint.h> is preferred */
typedef unsigned long long U64;
typedef signed long long S64;
#endif
/*-**************************************************************
* Memory I/O API
*****************************************************************/
/*=== Static platform detection ===*/
MEM_STATIC unsigned MEM_32bits(void);
MEM_STATIC unsigned MEM_64bits(void);
MEM_STATIC unsigned MEM_isLittleEndian(void);
/*=== Native unaligned read/write ===*/
MEM_STATIC U16 MEM_read16(const void* memPtr);
MEM_STATIC U32 MEM_read32(const void* memPtr);
MEM_STATIC U64 MEM_read64(const void* memPtr);
MEM_STATIC size_t MEM_readST(const void* memPtr);
MEM_STATIC void MEM_write16(void* memPtr, U16 value);
MEM_STATIC void MEM_write32(void* memPtr, U32 value);
MEM_STATIC void MEM_write64(void* memPtr, U64 value);
/*=== Little endian unaligned read/write ===*/
MEM_STATIC U16 MEM_readLE16(const void* memPtr);
MEM_STATIC U32 MEM_readLE24(const void* memPtr);
MEM_STATIC U32 MEM_readLE32(const void* memPtr);
MEM_STATIC U64 MEM_readLE64(const void* memPtr);
MEM_STATIC size_t MEM_readLEST(const void* memPtr);
MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val);
MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val);
MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32);
MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64);
MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val);
/*=== Big endian unaligned read/write ===*/
MEM_STATIC U32 MEM_readBE32(const void* memPtr);
MEM_STATIC U64 MEM_readBE64(const void* memPtr);
MEM_STATIC size_t MEM_readBEST(const void* memPtr);
MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32);
MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64);
MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val);
/*=== Byteswap ===*/
MEM_STATIC U32 MEM_swap32(U32 in);
MEM_STATIC U64 MEM_swap64(U64 in);
MEM_STATIC size_t MEM_swapST(size_t in);
/*-**************************************************************
* Memory I/O Implementation
*****************************************************************/
/* MEM_FORCE_MEMORY_ACCESS : For accessing unaligned memory:
* Method 0 : always use `memcpy()`. Safe and portable.
* Method 1 : Use compiler extension to set unaligned access.
* Method 2 : direct access. This method is portable but violate C standard.
* It can generate buggy code on targets depending on alignment.
* Default : method 1 if supported, else method 0
*/
#ifndef MEM_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */
# ifdef __GNUC__
# define MEM_FORCE_MEMORY_ACCESS 1
# endif
#endif
MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
MEM_STATIC unsigned MEM_isLittleEndian(void)
{
#if defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
return 1;
#elif defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
return 0;
#elif defined(__clang__) && __LITTLE_ENDIAN__
return 1;
#elif defined(__clang__) && __BIG_ENDIAN__
return 0;
#elif defined(_MSC_VER) && (_M_AMD64 || _M_IX86)
return 1;
#elif defined(__DMC__) && defined(_M_IX86)
return 1;
#else
const union { U32 u; BYTE c[4]; } one = { 1 }; /* don't use static : performance detrimental */
return one.c[0];
#endif
}
#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
/* violates C standard, by lying on structure alignment.
Only use if no other choice to achieve best performance on target platform */
MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
MEM_STATIC size_t MEM_readST(const void* memPtr) { return *(const size_t*) memPtr; }
MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
typedef __attribute__((aligned(1))) U16 unalign16;
typedef __attribute__((aligned(1))) U32 unalign32;
typedef __attribute__((aligned(1))) U64 unalign64;
typedef __attribute__((aligned(1))) size_t unalignArch;
MEM_STATIC U16 MEM_read16(const void* ptr) { return *(const unalign16*)ptr; }
MEM_STATIC U32 MEM_read32(const void* ptr) { return *(const unalign32*)ptr; }
MEM_STATIC U64 MEM_read64(const void* ptr) { return *(const unalign64*)ptr; }
MEM_STATIC size_t MEM_readST(const void* ptr) { return *(const unalignArch*)ptr; }
MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(unalign16*)memPtr = value; }
MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(unalign32*)memPtr = value; }
MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(unalign64*)memPtr = value; }
#else
/* default method, safe and standard.
can sometimes prove slower */
MEM_STATIC U16 MEM_read16(const void* memPtr)
{
U16 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC U32 MEM_read32(const void* memPtr)
{
U32 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC U64 MEM_read64(const void* memPtr)
{
U64 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC size_t MEM_readST(const void* memPtr)
{
size_t val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC void MEM_write16(void* memPtr, U16 value)
{
ZSTD_memcpy(memPtr, &value, sizeof(value));
}
MEM_STATIC void MEM_write32(void* memPtr, U32 value)
{
ZSTD_memcpy(memPtr, &value, sizeof(value));
}
MEM_STATIC void MEM_write64(void* memPtr, U64 value)
{
ZSTD_memcpy(memPtr, &value, sizeof(value));
}
#endif /* MEM_FORCE_MEMORY_ACCESS */
MEM_STATIC U32 MEM_swap32_fallback(U32 in)
{
return ((in << 24) & 0xff000000 ) |
((in << 8) & 0x00ff0000 ) |
((in >> 8) & 0x0000ff00 ) |
((in >> 24) & 0x000000ff );
}
MEM_STATIC U32 MEM_swap32(U32 in)
{
#if defined(_MSC_VER) /* Visual Studio */
return _byteswap_ulong(in);
#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
|| (defined(__clang__) && __has_builtin(__builtin_bswap32))
return __builtin_bswap32(in);
#else
return MEM_swap32_fallback(in);
#endif
}
MEM_STATIC U64 MEM_swap64_fallback(U64 in)
{
return ((in << 56) & 0xff00000000000000ULL) |
((in << 40) & 0x00ff000000000000ULL) |
((in << 24) & 0x0000ff0000000000ULL) |
((in << 8) & 0x000000ff00000000ULL) |
((in >> 8) & 0x00000000ff000000ULL) |
((in >> 24) & 0x0000000000ff0000ULL) |
((in >> 40) & 0x000000000000ff00ULL) |
((in >> 56) & 0x00000000000000ffULL);
}
MEM_STATIC U64 MEM_swap64(U64 in)
{
#if defined(_MSC_VER) /* Visual Studio */
return _byteswap_uint64(in);
#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
|| (defined(__clang__) && __has_builtin(__builtin_bswap64))
return __builtin_bswap64(in);
#else
return MEM_swap64_fallback(in);
#endif
}
MEM_STATIC size_t MEM_swapST(size_t in)
{
if (MEM_32bits())
return (size_t)MEM_swap32((U32)in);
else
return (size_t)MEM_swap64((U64)in);
}
/*=== Little endian r/w ===*/
MEM_STATIC U16 MEM_readLE16(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_read16(memPtr);
else {
const BYTE* p = (const BYTE*)memPtr;
return (U16)(p[0] + (p[1]<<8));
}
}
MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
{
if (MEM_isLittleEndian()) {
MEM_write16(memPtr, val);
} else {
BYTE* p = (BYTE*)memPtr;
p[0] = (BYTE)val;
p[1] = (BYTE)(val>>8);
}
}
MEM_STATIC U32 MEM_readLE24(const void* memPtr)
{
return (U32)MEM_readLE16(memPtr) + ((U32)(((const BYTE*)memPtr)[2]) << 16);
}
MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val)
{
MEM_writeLE16(memPtr, (U16)val);
((BYTE*)memPtr)[2] = (BYTE)(val>>16);
}
MEM_STATIC U32 MEM_readLE32(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_read32(memPtr);
else
return MEM_swap32(MEM_read32(memPtr));
}
MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32)
{
if (MEM_isLittleEndian())
MEM_write32(memPtr, val32);
else
MEM_write32(memPtr, MEM_swap32(val32));
}
MEM_STATIC U64 MEM_readLE64(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_read64(memPtr);
else
return MEM_swap64(MEM_read64(memPtr));
}
MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64)
{
if (MEM_isLittleEndian())
MEM_write64(memPtr, val64);
else
MEM_write64(memPtr, MEM_swap64(val64));
}
MEM_STATIC size_t MEM_readLEST(const void* memPtr)
{
if (MEM_32bits())
return (size_t)MEM_readLE32(memPtr);
else
return (size_t)MEM_readLE64(memPtr);
}
MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val)
{
if (MEM_32bits())
MEM_writeLE32(memPtr, (U32)val);
else
MEM_writeLE64(memPtr, (U64)val);
}
/*=== Big endian r/w ===*/
MEM_STATIC U32 MEM_readBE32(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_swap32(MEM_read32(memPtr));
else
return MEM_read32(memPtr);
}
MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32)
{
if (MEM_isLittleEndian())
MEM_write32(memPtr, MEM_swap32(val32));
else
MEM_write32(memPtr, val32);
}
MEM_STATIC U64 MEM_readBE64(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_swap64(MEM_read64(memPtr));
else
return MEM_read64(memPtr);
}
MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64)
{
if (MEM_isLittleEndian())
MEM_write64(memPtr, MEM_swap64(val64));
else
MEM_write64(memPtr, val64);
}
MEM_STATIC size_t MEM_readBEST(const void* memPtr)
{
if (MEM_32bits())
return (size_t)MEM_readBE32(memPtr);
else
return (size_t)MEM_readBE64(memPtr);
}
MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val)
{
if (MEM_32bits())
MEM_writeBE32(memPtr, (U32)val);
else
MEM_writeBE64(memPtr, (U64)val);
}
/* code only tested on 32 and 64 bits systems */
MEM_STATIC void MEM_check(void) { DEBUG_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
#if defined (__cplusplus)
}
#endif
#endif /* MEM_H_MODULE */
/**** ended inlining mem.h ****/
/**** start inlining error_private.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/* Note : this module is expected to remain private, do not expose it */
#ifndef ERROR_H_MODULE
#define ERROR_H_MODULE
#if defined (__cplusplus)
extern "C" {
#endif
/* ****************************************
* Dependencies
******************************************/
/**** start inlining ../zstd_errors.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_ERRORS_H_398273423
#define ZSTD_ERRORS_H_398273423
#if defined (__cplusplus)
extern "C" {
#endif
/*===== dependency =====*/
#include <stddef.h> /* size_t */
/* ===== ZSTDERRORLIB_API : control library symbols visibility ===== */
#ifndef ZSTDERRORLIB_VISIBLE
/* Backwards compatibility with old macro name */
# ifdef ZSTDERRORLIB_VISIBILITY
# define ZSTDERRORLIB_VISIBLE ZSTDERRORLIB_VISIBILITY
# elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
# define ZSTDERRORLIB_VISIBLE __attribute__ ((visibility ("default")))
# else
# define ZSTDERRORLIB_VISIBLE
# endif
#endif
#ifndef ZSTDERRORLIB_HIDDEN
# if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
# define ZSTDERRORLIB_HIDDEN __attribute__ ((visibility ("hidden")))
# else
# define ZSTDERRORLIB_HIDDEN
# endif
#endif
#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
# define ZSTDERRORLIB_API __declspec(dllexport) ZSTDERRORLIB_VISIBLE
#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
# define ZSTDERRORLIB_API __declspec(dllimport) ZSTDERRORLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
#else
# define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBLE
#endif
/*-*********************************************
* Error codes list
*-*********************************************
* Error codes _values_ are pinned down since v1.3.1 only.
* Therefore, don't rely on values if you may link to any version < v1.3.1.
*
* Only values < 100 are considered stable.
*
* note 1 : this API shall be used with static linking only.
* dynamic linking is not yet officially supported.
* note 2 : Prefer relying on the enum than on its value whenever possible
* This is the only supported way to use the error list < v1.3.1
* note 3 : ZSTD_isError() is always correct, whatever the library version.
**********************************************/
typedef enum {
ZSTD_error_no_error = 0,
ZSTD_error_GENERIC = 1,
ZSTD_error_prefix_unknown = 10,
ZSTD_error_version_unsupported = 12,
ZSTD_error_frameParameter_unsupported = 14,
ZSTD_error_frameParameter_windowTooLarge = 16,
ZSTD_error_corruption_detected = 20,
ZSTD_error_checksum_wrong = 22,
ZSTD_error_literals_headerWrong = 24,
ZSTD_error_dictionary_corrupted = 30,
ZSTD_error_dictionary_wrong = 32,
ZSTD_error_dictionaryCreation_failed = 34,
ZSTD_error_parameter_unsupported = 40,
ZSTD_error_parameter_combination_unsupported = 41,
ZSTD_error_parameter_outOfBound = 42,
ZSTD_error_tableLog_tooLarge = 44,
ZSTD_error_maxSymbolValue_tooLarge = 46,
ZSTD_error_maxSymbolValue_tooSmall = 48,
ZSTD_error_stabilityCondition_notRespected = 50,
ZSTD_error_stage_wrong = 60,
ZSTD_error_init_missing = 62,
ZSTD_error_memory_allocation = 64,
ZSTD_error_workSpace_tooSmall= 66,
ZSTD_error_dstSize_tooSmall = 70,
ZSTD_error_srcSize_wrong = 72,
ZSTD_error_dstBuffer_null = 74,
ZSTD_error_noForwardProgress_destFull = 80,
ZSTD_error_noForwardProgress_inputEmpty = 82,
/* following error codes are __NOT STABLE__, they can be removed or changed in future versions */
ZSTD_error_frameIndex_tooLarge = 100,
ZSTD_error_seekableIO = 102,
ZSTD_error_dstBuffer_wrong = 104,
ZSTD_error_srcBuffer_wrong = 105,
ZSTD_error_sequenceProducer_failed = 106,
ZSTD_error_externalSequences_invalid = 107,
ZSTD_error_maxCode = 120 /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
} ZSTD_ErrorCode;
/*! ZSTD_getErrorCode() :
convert a `size_t` function result into a `ZSTD_ErrorCode` enum type,
which can be used to compare with enum list published above */
ZSTDERRORLIB_API ZSTD_ErrorCode ZSTD_getErrorCode(size_t functionResult);
ZSTDERRORLIB_API const char* ZSTD_getErrorString(ZSTD_ErrorCode code); /**< Same as ZSTD_getErrorName, but using a `ZSTD_ErrorCode` enum argument */
#if defined (__cplusplus)
}
#endif
#endif /* ZSTD_ERRORS_H_398273423 */
/**** ended inlining ../zstd_errors.h ****/
/**** skipping file: compiler.h ****/
/**** skipping file: debug.h ****/
/**** skipping file: zstd_deps.h ****/
/* ****************************************
* Compiler-specific
******************************************/
#if defined(__GNUC__)
# define ERR_STATIC static __attribute__((unused))
#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
# define ERR_STATIC static inline
#elif defined(_MSC_VER)
# define ERR_STATIC static __inline
#else
# define ERR_STATIC static /* this version may generate warnings for unused static functions; disable the relevant warning */
#endif
/*-****************************************
* Customization (error_public.h)
******************************************/
typedef ZSTD_ErrorCode ERR_enum;
#define PREFIX(name) ZSTD_error_##name
/*-****************************************
* Error codes handling
******************************************/
#undef ERROR /* already defined on Visual Studio */
#define ERROR(name) ZSTD_ERROR(name)
#define ZSTD_ERROR(name) ((size_t)-PREFIX(name))
ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
ERR_STATIC ERR_enum ERR_getErrorCode(size_t code) { if (!ERR_isError(code)) return (ERR_enum)0; return (ERR_enum) (0-code); }
/* check and forward error code */
#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
#define CHECK_F(f) { CHECK_V_F(_var_err__, f); }
/*-****************************************
* Error Strings
******************************************/
const char* ERR_getErrorString(ERR_enum code); /* error_private.c */
ERR_STATIC const char* ERR_getErrorName(size_t code)
{
return ERR_getErrorString(ERR_getErrorCode(code));
}
/**
* Ignore: this is an internal helper.
*
* This is a helper function to help force C99-correctness during compilation.
* Under strict compilation modes, variadic macro arguments can't be empty.
* However, variadic function arguments can be. Using a function therefore lets
* us statically check that at least one (string) argument was passed,
* independent of the compilation flags.
*/
static INLINE_KEYWORD UNUSED_ATTR
void _force_has_format_string(const char *format, ...) {
(void)format;
}
/**
* Ignore: this is an internal helper.
*
* We want to force this function invocation to be syntactically correct, but
* we don't want to force runtime evaluation of its arguments.
*/
#define _FORCE_HAS_FORMAT_STRING(...) \
if (0) { \
_force_has_format_string(__VA_ARGS__); \
}
#define ERR_QUOTE(str) #str
/**
* Return the specified error if the condition evaluates to true.
*
* In debug modes, prints additional information.
* In order to do that (particularly, printing the conditional that failed),
* this can't just wrap RETURN_ERROR().
*/
#define RETURN_ERROR_IF(cond, err, ...) \
if (cond) { \
RAWLOG(3, "%s:%d: ERROR!: check %s failed, returning %s", \
__FILE__, __LINE__, ERR_QUOTE(cond), ERR_QUOTE(ERROR(err))); \
_FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
RAWLOG(3, ": " __VA_ARGS__); \
RAWLOG(3, "\n"); \
return ERROR(err); \
}
/**
* Unconditionally return the specified error.
*
* In debug modes, prints additional information.
*/
#define RETURN_ERROR(err, ...) \
do { \
RAWLOG(3, "%s:%d: ERROR!: unconditional check failed, returning %s", \
__FILE__, __LINE__, ERR_QUOTE(ERROR(err))); \
_FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
RAWLOG(3, ": " __VA_ARGS__); \
RAWLOG(3, "\n"); \
return ERROR(err); \
} while(0);
/**
* If the provided expression evaluates to an error code, returns that error code.
*
* In debug modes, prints additional information.
*/
#define FORWARD_IF_ERROR(err, ...) \
do { \
size_t const err_code = (err); \
if (ERR_isError(err_code)) { \
RAWLOG(3, "%s:%d: ERROR!: forwarding error in %s: %s", \
__FILE__, __LINE__, ERR_QUOTE(err), ERR_getErrorName(err_code)); \
_FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
RAWLOG(3, ": " __VA_ARGS__); \
RAWLOG(3, "\n"); \
return err_code; \
} \
} while(0);
#if defined (__cplusplus)
}
#endif
#endif /* ERROR_H_MODULE */
/**** ended inlining error_private.h ****/
#define FSE_STATIC_LINKING_ONLY /* FSE_MIN_TABLELOG */
/**** start inlining fse.h ****/
/* ******************************************************************
* FSE : Finite State Entropy codec
* Public Prototypes declaration
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
#if defined (__cplusplus)
extern "C" {
#endif
#ifndef FSE_H
#define FSE_H
/*-*****************************************
* Dependencies
******************************************/
/**** skipping file: zstd_deps.h ****/
/*-*****************************************
* FSE_PUBLIC_API : control library symbols visibility
******************************************/
#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
# define FSE_PUBLIC_API __attribute__ ((visibility ("default")))
#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) /* Visual expected */
# define FSE_PUBLIC_API __declspec(dllexport)
#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
# define FSE_PUBLIC_API __declspec(dllimport) /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
#else
# define FSE_PUBLIC_API
#endif
/*------ Version ------*/
#define FSE_VERSION_MAJOR 0
#define FSE_VERSION_MINOR 9
#define FSE_VERSION_RELEASE 0
#define FSE_LIB_VERSION FSE_VERSION_MAJOR.FSE_VERSION_MINOR.FSE_VERSION_RELEASE
#define FSE_QUOTE(str) #str
#define FSE_EXPAND_AND_QUOTE(str) FSE_QUOTE(str)
#define FSE_VERSION_STRING FSE_EXPAND_AND_QUOTE(FSE_LIB_VERSION)
#define FSE_VERSION_NUMBER (FSE_VERSION_MAJOR *100*100 + FSE_VERSION_MINOR *100 + FSE_VERSION_RELEASE)
FSE_PUBLIC_API unsigned FSE_versionNumber(void); /**< library version number; to be used when checking dll version */
/*-*****************************************
* Tool functions
******************************************/
FSE_PUBLIC_API size_t FSE_compressBound(size_t size); /* maximum compressed size */
/* Error Management */
FSE_PUBLIC_API unsigned FSE_isError(size_t code); /* tells if a return value is an error code */
FSE_PUBLIC_API const char* FSE_getErrorName(size_t code); /* provides error code string (useful for debugging) */
/*-*****************************************
* FSE detailed API
******************************************/
/*!
FSE_compress() does the following:
1. count symbol occurrence from source[] into table count[] (see hist.h)
2. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)
3. save normalized counters to memory buffer using writeNCount()
4. build encoding table 'CTable' from normalized counters
5. encode the data stream using encoding table 'CTable'
FSE_decompress() does the following:
1. read normalized counters with readNCount()
2. build decoding table 'DTable' from normalized counters
3. decode the data stream using decoding table 'DTable'
The following API allows targeting specific sub-functions for advanced tasks.
For example, it's possible to compress several blocks using the same 'CTable',
or to save and provide normalized distribution using external method.
*/
/* *** COMPRESSION *** */
/*! FSE_optimalTableLog():
dynamically downsize 'tableLog' when conditions are met.
It saves CPU time, by using smaller tables, while preserving or even improving compression ratio.
@return : recommended tableLog (necessarily <= 'maxTableLog') */
FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
/*! FSE_normalizeCount():
normalize counts so that sum(count[]) == Power_of_2 (2^tableLog)
'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
useLowProbCount is a boolean parameter which trades off compressed size for
faster header decoding. When it is set to 1, the compressed data will be slightly
smaller. And when it is set to 0, FSE_readNCount() and FSE_buildDTable() will be
faster. If you are compressing a small amount of data (< 2 KB) then useLowProbCount=0
is a good default, since header deserialization makes a big speed difference.
Otherwise, useLowProbCount=1 is a good default, since the speed difference is small.
@return : tableLog,
or an errorCode, which can be tested using FSE_isError() */
FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog,
const unsigned* count, size_t srcSize, unsigned maxSymbolValue, unsigned useLowProbCount);
/*! FSE_NCountWriteBound():
Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
Typically useful for allocation purpose. */
FSE_PUBLIC_API size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog);
/*! FSE_writeNCount():
Compactly save 'normalizedCounter' into 'buffer'.
@return : size of the compressed table,
or an errorCode, which can be tested using FSE_isError(). */
FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize,
const short* normalizedCounter,
unsigned maxSymbolValue, unsigned tableLog);
/*! Constructor and Destructor of FSE_CTable.
Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
typedef unsigned FSE_CTable; /* don't allocate that. It's only meant to be more restrictive than void* */
/*! FSE_buildCTable():
Builds `ct`, which must be already allocated, using FSE_createCTable().
@return : 0, or an errorCode, which can be tested using FSE_isError() */
FSE_PUBLIC_API size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
/*! FSE_compress_usingCTable():
Compress `src` using `ct` into `dst` which must be already allocated.
@return : size of compressed data (<= `dstCapacity`),
or 0 if compressed data could not fit into `dst`,
or an errorCode, which can be tested using FSE_isError() */
FSE_PUBLIC_API size_t FSE_compress_usingCTable (void* dst, size_t dstCapacity, const void* src, size_t srcSize, const FSE_CTable* ct);
/*!
Tutorial :
----------
The first step is to count all symbols. FSE_count() does this job very fast.
Result will be saved into 'count', a table of unsigned int, which must be already allocated, and have 'maxSymbolValuePtr[0]+1' cells.
'src' is a table of bytes of size 'srcSize'. All values within 'src' MUST be <= maxSymbolValuePtr[0]
maxSymbolValuePtr[0] will be updated, with its real value (necessarily <= original value)
FSE_count() will return the number of occurrence of the most frequent symbol.
This can be used to know if there is a single symbol within 'src', and to quickly evaluate its compressibility.
If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).
The next step is to normalize the frequencies.
FSE_normalizeCount() will ensure that sum of frequencies is == 2 ^'tableLog'.
It also guarantees a minimum of 1 to any Symbol with frequency >= 1.
You can use 'tableLog'==0 to mean "use default tableLog value".
If you are unsure of which tableLog value to use, you can ask FSE_optimalTableLog(),
which will provide the optimal valid tableLog given sourceSize, maxSymbolValue, and a user-defined maximum (0 means "default").
The result of FSE_normalizeCount() will be saved into a table,
called 'normalizedCounter', which is a table of signed short.
'normalizedCounter' must be already allocated, and have at least 'maxSymbolValue+1' cells.
The return value is tableLog if everything proceeded as expected.
It is 0 if there is a single symbol within distribution.
If there is an error (ex: invalid tableLog value), the function will return an ErrorCode (which can be tested using FSE_isError()).
'normalizedCounter' can be saved in a compact manner to a memory area using FSE_writeNCount().
'buffer' must be already allocated.
For guaranteed success, buffer size must be at least FSE_headerBound().
The result of the function is the number of bytes written into 'buffer'.
If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError(); ex : buffer size too small).
'normalizedCounter' can then be used to create the compression table 'CTable'.
The space required by 'CTable' must be already allocated, using FSE_createCTable().
You can then use FSE_buildCTable() to fill 'CTable'.
If there is an error, both functions will return an ErrorCode (which can be tested using FSE_isError()).
'CTable' can then be used to compress 'src', with FSE_compress_usingCTable().
Similar to FSE_count(), the convention is that 'src' is assumed to be a table of char of size 'srcSize'
The function returns the size of compressed data (without header), necessarily <= `dstCapacity`.
If it returns '0', compressed data could not fit into 'dst'.
If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).
*/
/* *** DECOMPRESSION *** */
/*! FSE_readNCount():
Read compactly saved 'normalizedCounter' from 'rBuffer'.
@return : size read from 'rBuffer',
or an errorCode, which can be tested using FSE_isError().
maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter,
unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
const void* rBuffer, size_t rBuffSize);
/*! FSE_readNCount_bmi2():
* Same as FSE_readNCount() but pass bmi2=1 when your CPU supports BMI2 and 0 otherwise.
*/
FSE_PUBLIC_API size_t FSE_readNCount_bmi2(short* normalizedCounter,
unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
const void* rBuffer, size_t rBuffSize, int bmi2);
typedef unsigned FSE_DTable; /* don't allocate that. It's just a way to be more restrictive than void* */
/*!
Tutorial :
----------
(Note : these functions only decompress FSE-compressed blocks.
If block is uncompressed, use memcpy() instead
If block is a single repeated byte, use memset() instead )
The first step is to obtain the normalized frequencies of symbols.
This can be performed by FSE_readNCount() if it was saved using FSE_writeNCount().
'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
or size the table to handle worst case situations (typically 256).
FSE_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
The result of FSE_readNCount() is the number of bytes read from 'rBuffer'.
Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
If there is an error, the function will return an error code, which can be tested using FSE_isError().
The next step is to build the decompression tables 'FSE_DTable' from 'normalizedCounter'.
This is performed by the function FSE_buildDTable().
The space required by 'FSE_DTable' must be already allocated using FSE_createDTable().
If there is an error, the function will return an error code, which can be tested using FSE_isError().
`FSE_DTable` can then be used to decompress `cSrc`, with FSE_decompress_usingDTable().
`cSrcSize` must be strictly correct, otherwise decompression will fail.
FSE_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)
*/
#endif /* FSE_H */
#if defined(FSE_STATIC_LINKING_ONLY) && !defined(FSE_H_FSE_STATIC_LINKING_ONLY)
#define FSE_H_FSE_STATIC_LINKING_ONLY
/* *** Dependency *** */
/**** start inlining bitstream.h ****/
/* ******************************************************************
* bitstream
* Part of FSE library
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
#ifndef BITSTREAM_H_MODULE
#define BITSTREAM_H_MODULE
#if defined (__cplusplus)
extern "C" {
#endif
/*
* This API consists of small unitary functions, which must be inlined for best performance.
* Since link-time-optimization is not available for all compilers,
* these functions are defined into a .h to be included.
*/
/*-****************************************
* Dependencies
******************************************/
/**** skipping file: mem.h ****/
/**** skipping file: compiler.h ****/
/**** skipping file: debug.h ****/
/**** skipping file: error_private.h ****/
/**** start inlining bits.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_BITS_H
#define ZSTD_BITS_H
/**** skipping file: mem.h ****/
MEM_STATIC unsigned ZSTD_countTrailingZeros32_fallback(U32 val)
{
assert(val != 0);
{
static const U32 DeBruijnBytePos[32] = {0, 1, 28, 2, 29, 14, 24, 3,
30, 22, 20, 15, 25, 17, 4, 8,
31, 27, 13, 23, 21, 19, 16, 7,
26, 12, 18, 6, 11, 5, 10, 9};
return DeBruijnBytePos[((U32) ((val & -(S32) val) * 0x077CB531U)) >> 27];
}
}
MEM_STATIC unsigned ZSTD_countTrailingZeros32(U32 val)
{
assert(val != 0);
# if defined(_MSC_VER)
# if STATIC_BMI2 == 1
return (unsigned)_tzcnt_u32(val);
# else
if (val != 0) {
unsigned long r;
_BitScanForward(&r, val);
return (unsigned)r;
} else {
/* Should not reach this code path */
__assume(0);
}
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 4)
return (unsigned)__builtin_ctz(val);
# else
return ZSTD_countTrailingZeros32_fallback(val);
# endif
}
MEM_STATIC unsigned ZSTD_countLeadingZeros32_fallback(U32 val) {
assert(val != 0);
{
static const U32 DeBruijnClz[32] = {0, 9, 1, 10, 13, 21, 2, 29,
11, 14, 16, 18, 22, 25, 3, 30,
8, 12, 20, 28, 15, 17, 24, 7,
19, 27, 23, 6, 26, 5, 4, 31};
val |= val >> 1;
val |= val >> 2;
val |= val >> 4;
val |= val >> 8;
val |= val >> 16;
return 31 - DeBruijnClz[(val * 0x07C4ACDDU) >> 27];
}
}
MEM_STATIC unsigned ZSTD_countLeadingZeros32(U32 val)
{
assert(val != 0);
# if defined(_MSC_VER)
# if STATIC_BMI2 == 1
return (unsigned)_lzcnt_u32(val);
# else
if (val != 0) {
unsigned long r;
_BitScanReverse(&r, val);
return (unsigned)(31 - r);
} else {
/* Should not reach this code path */
__assume(0);
}
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 4)
return (unsigned)__builtin_clz(val);
# else
return ZSTD_countLeadingZeros32_fallback(val);
# endif
}
MEM_STATIC unsigned ZSTD_countTrailingZeros64(U64 val)
{
assert(val != 0);
# if defined(_MSC_VER) && defined(_WIN64)
# if STATIC_BMI2 == 1
return (unsigned)_tzcnt_u64(val);
# else
if (val != 0) {
unsigned long r;
_BitScanForward64(&r, val);
return (unsigned)r;
} else {
/* Should not reach this code path */
__assume(0);
}
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 4) && defined(__LP64__)
return (unsigned)__builtin_ctzll(val);
# else
{
U32 mostSignificantWord = (U32)(val >> 32);
U32 leastSignificantWord = (U32)val;
if (leastSignificantWord == 0) {
return 32 + ZSTD_countTrailingZeros32(mostSignificantWord);
} else {
return ZSTD_countTrailingZeros32(leastSignificantWord);
}
}
# endif
}
MEM_STATIC unsigned ZSTD_countLeadingZeros64(U64 val)
{
assert(val != 0);
# if defined(_MSC_VER) && defined(_WIN64)
# if STATIC_BMI2 == 1
return (unsigned)_lzcnt_u64(val);
# else
if (val != 0) {
unsigned long r;
_BitScanReverse64(&r, val);
return (unsigned)(63 - r);
} else {
/* Should not reach this code path */
__assume(0);
}
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 4)
return (unsigned)(__builtin_clzll(val));
# else
{
U32 mostSignificantWord = (U32)(val >> 32);
U32 leastSignificantWord = (U32)val;
if (mostSignificantWord == 0) {
return 32 + ZSTD_countLeadingZeros32(leastSignificantWord);
} else {
return ZSTD_countLeadingZeros32(mostSignificantWord);
}
}
# endif
}
MEM_STATIC unsigned ZSTD_NbCommonBytes(size_t val)
{
if (MEM_isLittleEndian()) {
if (MEM_64bits()) {
return ZSTD_countTrailingZeros64((U64)val) >> 3;
} else {
return ZSTD_countTrailingZeros32((U32)val) >> 3;
}
} else { /* Big Endian CPU */
if (MEM_64bits()) {
return ZSTD_countLeadingZeros64((U64)val) >> 3;
} else {
return ZSTD_countLeadingZeros32((U32)val) >> 3;
}
}
}
MEM_STATIC unsigned ZSTD_highbit32(U32 val) /* compress, dictBuilder, decodeCorpus */
{
assert(val != 0);
return 31 - ZSTD_countLeadingZeros32(val);
}
/* ZSTD_rotateRight_*():
* Rotates a bitfield to the right by "count" bits.
* https://en.wikipedia.org/w/index.php?title=Circular_shift&oldid=991635599#Implementing_circular_shifts
*/
MEM_STATIC
U64 ZSTD_rotateRight_U64(U64 const value, U32 count) {
assert(count < 64);
count &= 0x3F; /* for fickle pattern recognition */
return (value >> count) | (U64)(value << ((0U - count) & 0x3F));
}
MEM_STATIC
U32 ZSTD_rotateRight_U32(U32 const value, U32 count) {
assert(count < 32);
count &= 0x1F; /* for fickle pattern recognition */
return (value >> count) | (U32)(value << ((0U - count) & 0x1F));
}
MEM_STATIC
U16 ZSTD_rotateRight_U16(U16 const value, U32 count) {
assert(count < 16);
count &= 0x0F; /* for fickle pattern recognition */
return (value >> count) | (U16)(value << ((0U - count) & 0x0F));
}
#endif /* ZSTD_BITS_H */
/**** ended inlining bits.h ****/
/*=========================================
* Target specific
=========================================*/
#ifndef ZSTD_NO_INTRINSICS
# if (defined(__BMI__) || defined(__BMI2__)) && defined(__GNUC__)
# include <immintrin.h> /* support for bextr (experimental)/bzhi */
# elif defined(__ICCARM__)
# include <intrinsics.h>
# endif
#endif
#define STREAM_ACCUMULATOR_MIN_32 25
#define STREAM_ACCUMULATOR_MIN_64 57
#define STREAM_ACCUMULATOR_MIN ((U32)(MEM_32bits() ? STREAM_ACCUMULATOR_MIN_32 : STREAM_ACCUMULATOR_MIN_64))
/*-******************************************
* bitStream encoding API (write forward)
********************************************/
/* bitStream can mix input from multiple sources.
* A critical property of these streams is that they encode and decode in **reverse** direction.
* So the first bit sequence you add will be the last to be read, like a LIFO stack.
*/
typedef struct {
size_t bitContainer;
unsigned bitPos;
char* startPtr;
char* ptr;
char* endPtr;
} BIT_CStream_t;
MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC, void* dstBuffer, size_t dstCapacity);
MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC, size_t value, unsigned nbBits);
MEM_STATIC void BIT_flushBits(BIT_CStream_t* bitC);
MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC);
/* Start with initCStream, providing the size of buffer to write into.
* bitStream will never write outside of this buffer.
* `dstCapacity` must be >= sizeof(bitD->bitContainer), otherwise @return will be an error code.
*
* bits are first added to a local register.
* Local register is size_t, hence 64-bits on 64-bits systems, or 32-bits on 32-bits systems.
* Writing data into memory is an explicit operation, performed by the flushBits function.
* Hence keep track how many bits are potentially stored into local register to avoid register overflow.
* After a flushBits, a maximum of 7 bits might still be stored into local register.
*
* Avoid storing elements of more than 24 bits if you want compatibility with 32-bits bitstream readers.
*
* Last operation is to close the bitStream.
* The function returns the final size of CStream in bytes.
* If data couldn't fit into `dstBuffer`, it will return a 0 ( == not storable)
*/
/*-********************************************
* bitStream decoding API (read backward)
**********************************************/
typedef size_t BitContainerType;
typedef struct {
BitContainerType bitContainer;
unsigned bitsConsumed;
const char* ptr;
const char* start;
const char* limitPtr;
} BIT_DStream_t;
typedef enum { BIT_DStream_unfinished = 0, /* fully refilled */
BIT_DStream_endOfBuffer = 1, /* still some bits left in bitstream */
BIT_DStream_completed = 2, /* bitstream entirely consumed, bit-exact */
BIT_DStream_overflow = 3 /* user requested more bits than present in bitstream */
} BIT_DStream_status; /* result of BIT_reloadDStream() */
MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
/* Start by invoking BIT_initDStream().
* A chunk of the bitStream is then stored into a local register.
* Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (BitContainerType).
* You can then retrieve bitFields stored into the local register, **in reverse order**.
* Local register is explicitly reloaded from memory by the BIT_reloadDStream() method.
* A reload guarantee a minimum of ((8*sizeof(bitD->bitContainer))-7) bits when its result is BIT_DStream_unfinished.
* Otherwise, it can be less than that, so proceed accordingly.
* Checking if DStream has reached its end can be performed with BIT_endOfDStream().
*/
/*-****************************************
* unsafe API
******************************************/
MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC, size_t value, unsigned nbBits);
/* faster, but works only if value is "clean", meaning all high bits above nbBits are 0 */
MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC);
/* unsafe version; does not check buffer overflow */
MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
/* faster, but works only if nbBits >= 1 */
/*===== Local Constants =====*/
static const unsigned BIT_mask[] = {
0, 1, 3, 7, 0xF, 0x1F,
0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF,
0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0x1FFFF,
0x3FFFF, 0x7FFFF, 0xFFFFF, 0x1FFFFF, 0x3FFFFF, 0x7FFFFF,
0xFFFFFF, 0x1FFFFFF, 0x3FFFFFF, 0x7FFFFFF, 0xFFFFFFF, 0x1FFFFFFF,
0x3FFFFFFF, 0x7FFFFFFF}; /* up to 31 bits */
#define BIT_MASK_SIZE (sizeof(BIT_mask) / sizeof(BIT_mask[0]))
/*-**************************************************************
* bitStream encoding
****************************************************************/
/*! BIT_initCStream() :
* `dstCapacity` must be > sizeof(size_t)
* @return : 0 if success,
* otherwise an error code (can be tested using ERR_isError()) */
MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
void* startPtr, size_t dstCapacity)
{
bitC->bitContainer = 0;
bitC->bitPos = 0;
bitC->startPtr = (char*)startPtr;
bitC->ptr = bitC->startPtr;
bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer);
if (dstCapacity <= sizeof(bitC->bitContainer)) return ERROR(dstSize_tooSmall);
return 0;
}
FORCE_INLINE_TEMPLATE size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
{
#if defined(STATIC_BMI2) && STATIC_BMI2 == 1 && !defined(ZSTD_NO_INTRINSICS)
return _bzhi_u64(bitContainer, nbBits);
#else
assert(nbBits < BIT_MASK_SIZE);
return bitContainer & BIT_mask[nbBits];
#endif
}
/*! BIT_addBits() :
* can add up to 31 bits into `bitC`.
* Note : does not check for register overflow ! */
MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
size_t value, unsigned nbBits)
{
DEBUG_STATIC_ASSERT(BIT_MASK_SIZE == 32);
assert(nbBits < BIT_MASK_SIZE);
assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
bitC->bitContainer |= BIT_getLowerBits(value, nbBits) << bitC->bitPos;
bitC->bitPos += nbBits;
}
/*! BIT_addBitsFast() :
* works only if `value` is _clean_,
* meaning all high bits above nbBits are 0 */
MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC,
size_t value, unsigned nbBits)
{
assert((value>>nbBits) == 0);
assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
bitC->bitContainer |= value << bitC->bitPos;
bitC->bitPos += nbBits;
}
/*! BIT_flushBitsFast() :
* assumption : bitContainer has not overflowed
* unsafe version; does not check buffer overflow */
MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC)
{
size_t const nbBytes = bitC->bitPos >> 3;
assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
assert(bitC->ptr <= bitC->endPtr);
MEM_writeLEST(bitC->ptr, bitC->bitContainer);
bitC->ptr += nbBytes;
bitC->bitPos &= 7;
bitC->bitContainer >>= nbBytes*8;
}
/*! BIT_flushBits() :
* assumption : bitContainer has not overflowed
* safe version; check for buffer overflow, and prevents it.
* note : does not signal buffer overflow.
* overflow will be revealed later on using BIT_closeCStream() */
MEM_STATIC void BIT_flushBits(BIT_CStream_t* bitC)
{
size_t const nbBytes = bitC->bitPos >> 3;
assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
assert(bitC->ptr <= bitC->endPtr);
MEM_writeLEST(bitC->ptr, bitC->bitContainer);
bitC->ptr += nbBytes;
if (bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
bitC->bitPos &= 7;
bitC->bitContainer >>= nbBytes*8;
}
/*! BIT_closeCStream() :
* @return : size of CStream, in bytes,
* or 0 if it could not fit into dstBuffer */
MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
{
BIT_addBitsFast(bitC, 1, 1); /* endMark */
BIT_flushBits(bitC);
if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
return (bitC->ptr - bitC->startPtr) + (bitC->bitPos > 0);
}
/*-********************************************************
* bitStream decoding
**********************************************************/
/*! BIT_initDStream() :
* Initialize a BIT_DStream_t.
* `bitD` : a pointer to an already allocated BIT_DStream_t structure.
* `srcSize` must be the *exact* size of the bitStream, in bytes.
* @return : size of stream (== srcSize), or an errorCode if a problem is detected
*/
MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
{
if (srcSize < 1) { ZSTD_memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
bitD->start = (const char*)srcBuffer;
bitD->limitPtr = bitD->start + sizeof(bitD->bitContainer);
if (srcSize >= sizeof(bitD->bitContainer)) { /* normal case */
bitD->ptr = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
bitD->bitContainer = MEM_readLEST(bitD->ptr);
{ BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
bitD->bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0; /* ensures bitsConsumed is always set */
if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
} else {
bitD->ptr = bitD->start;
bitD->bitContainer = *(const BYTE*)(bitD->start);
switch(srcSize)
{
case 7: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
ZSTD_FALLTHROUGH;
case 6: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
ZSTD_FALLTHROUGH;
case 5: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
ZSTD_FALLTHROUGH;
case 4: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[3]) << 24;
ZSTD_FALLTHROUGH;
case 3: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[2]) << 16;
ZSTD_FALLTHROUGH;
case 2: bitD->bitContainer += (BitContainerType)(((const BYTE*)(srcBuffer))[1]) << 8;
ZSTD_FALLTHROUGH;
default: break;
}
{ BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
bitD->bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;
if (lastByte == 0) return ERROR(corruption_detected); /* endMark not present */
}
bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
}
return srcSize;
}
FORCE_INLINE_TEMPLATE size_t BIT_getUpperBits(BitContainerType bitContainer, U32 const start)
{
return bitContainer >> start;
}
FORCE_INLINE_TEMPLATE size_t BIT_getMiddleBits(BitContainerType bitContainer, U32 const start, U32 const nbBits)
{
U32 const regMask = sizeof(bitContainer)*8 - 1;
/* if start > regMask, bitstream is corrupted, and result is undefined */
assert(nbBits < BIT_MASK_SIZE);
/* x86 transform & ((1 << nbBits) - 1) to bzhi instruction, it is better
* than accessing memory. When bmi2 instruction is not present, we consider
* such cpus old (pre-Haswell, 2013) and their performance is not of that
* importance.
*/
#if defined(__x86_64__) || defined(_M_X86)
return (bitContainer >> (start & regMask)) & ((((U64)1) << nbBits) - 1);
#else
return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
#endif
}
/*! BIT_lookBits() :
* Provides next n bits from local register.
* local register is not modified.
* On 32-bits, maxNbBits==24.
* On 64-bits, maxNbBits==56.
* @return : value extracted */
FORCE_INLINE_TEMPLATE size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
{
/* arbitrate between double-shift and shift+mask */
#if 1
/* if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8,
* bitstream is likely corrupted, and result is undefined */
return BIT_getMiddleBits(bitD->bitContainer, (sizeof(bitD->bitContainer)*8) - bitD->bitsConsumed - nbBits, nbBits);
#else
/* this code path is slower on my os-x laptop */
U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
return ((bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> 1) >> ((regMask-nbBits) & regMask);
#endif
}
/*! BIT_lookBitsFast() :
* unsafe version; only works if nbBits >= 1 */
MEM_STATIC size_t BIT_lookBitsFast(const BIT_DStream_t* bitD, U32 nbBits)
{
U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
assert(nbBits >= 1);
return (bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> (((regMask+1)-nbBits) & regMask);
}
FORCE_INLINE_TEMPLATE void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
{
bitD->bitsConsumed += nbBits;
}
/*! BIT_readBits() :
* Read (consume) next n bits from local register and update.
* Pay attention to not read more than nbBits contained into local register.
* @return : extracted value. */
FORCE_INLINE_TEMPLATE size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
{
size_t const value = BIT_lookBits(bitD, nbBits);
BIT_skipBits(bitD, nbBits);
return value;
}
/*! BIT_readBitsFast() :
* unsafe version; only works if nbBits >= 1 */
MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits)
{
size_t const value = BIT_lookBitsFast(bitD, nbBits);
assert(nbBits >= 1);
BIT_skipBits(bitD, nbBits);
return value;
}
/*! BIT_reloadDStream_internal() :
* Simple variant of BIT_reloadDStream(), with two conditions:
* 1. bitstream is valid : bitsConsumed <= sizeof(bitD->bitContainer)*8
* 2. look window is valid after shifted down : bitD->ptr >= bitD->start
*/
MEM_STATIC BIT_DStream_status BIT_reloadDStream_internal(BIT_DStream_t* bitD)
{
assert(bitD->bitsConsumed <= sizeof(bitD->bitContainer)*8);
bitD->ptr -= bitD->bitsConsumed >> 3;
assert(bitD->ptr >= bitD->start);
bitD->bitsConsumed &= 7;
bitD->bitContainer = MEM_readLEST(bitD->ptr);
return BIT_DStream_unfinished;
}
/*! BIT_reloadDStreamFast() :
* Similar to BIT_reloadDStream(), but with two differences:
* 1. bitsConsumed <= sizeof(bitD->bitContainer)*8 must hold!
* 2. Returns BIT_DStream_overflow when bitD->ptr < bitD->limitPtr, at this
* point you must use BIT_reloadDStream() to reload.
*/
MEM_STATIC BIT_DStream_status BIT_reloadDStreamFast(BIT_DStream_t* bitD)
{
if (UNLIKELY(bitD->ptr < bitD->limitPtr))
return BIT_DStream_overflow;
return BIT_reloadDStream_internal(bitD);
}
/*! BIT_reloadDStream() :
* Refill `bitD` from buffer previously set in BIT_initDStream() .
* This function is safe, it guarantees it will not never beyond src buffer.
* @return : status of `BIT_DStream_t` internal register.
* when status == BIT_DStream_unfinished, internal register is filled with at least 25 or 57 bits */
FORCE_INLINE_TEMPLATE BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
{
/* note : once in overflow mode, a bitstream remains in this mode until it's reset */
if (UNLIKELY(bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))) {
static const BitContainerType zeroFilled = 0;
bitD->ptr = (const char*)&zeroFilled; /* aliasing is allowed for char */
/* overflow detected, erroneous scenario or end of stream: no update */
return BIT_DStream_overflow;
}
assert(bitD->ptr >= bitD->start);
if (bitD->ptr >= bitD->limitPtr) {
return BIT_reloadDStream_internal(bitD);
}
if (bitD->ptr == bitD->start) {
/* reached end of bitStream => no update */
if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
return BIT_DStream_completed;
}
/* start < ptr < limitPtr => cautious update */
{ U32 nbBytes = bitD->bitsConsumed >> 3;
BIT_DStream_status result = BIT_DStream_unfinished;
if (bitD->ptr - nbBytes < bitD->start) {
nbBytes = (U32)(bitD->ptr - bitD->start); /* ptr > start */
result = BIT_DStream_endOfBuffer;
}
bitD->ptr -= nbBytes;
bitD->bitsConsumed -= nbBytes*8;
bitD->bitContainer = MEM_readLEST(bitD->ptr); /* reminder : srcSize > sizeof(bitD->bitContainer), otherwise bitD->ptr == bitD->start */
return result;
}
}
/*! BIT_endOfDStream() :
* @return : 1 if DStream has _exactly_ reached its end (all bits consumed).
*/
MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
{
return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
}
#if defined (__cplusplus)
}
#endif
#endif /* BITSTREAM_H_MODULE */
/**** ended inlining bitstream.h ****/
/* *****************************************
* Static allocation
*******************************************/
/* FSE buffer bounds */
#define FSE_NCOUNTBOUND 512
#define FSE_BLOCKBOUND(size) ((size) + ((size)>>7) + 4 /* fse states */ + sizeof(size_t) /* bitContainer */)
#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size)) /* Macro version, useful for static allocation */
/* It is possible to statically allocate FSE CTable/DTable as a table of FSE_CTable/FSE_DTable using below macros */
#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) (1 + (1<<((maxTableLog)-1)) + (((maxSymbolValue)+1)*2))
#define FSE_DTABLE_SIZE_U32(maxTableLog) (1 + (1<<(maxTableLog)))
/* or use the size to malloc() space directly. Pay attention to alignment restrictions though */
#define FSE_CTABLE_SIZE(maxTableLog, maxSymbolValue) (FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(FSE_CTable))
#define FSE_DTABLE_SIZE(maxTableLog) (FSE_DTABLE_SIZE_U32(maxTableLog) * sizeof(FSE_DTable))
/* *****************************************
* FSE advanced API
***************************************** */
unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
/**< same as FSE_optimalTableLog(), which used `minus==2` */
size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
/**< build a fake FSE_CTable, designed to compress always the same symbolValue */
/* FSE_buildCTable_wksp() :
* Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
* `wkspSize` must be >= `FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog)` of `unsigned`.
* See FSE_buildCTable_wksp() for breakdown of workspace usage.
*/
#define FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog) (((maxSymbolValue + 2) + (1ull << (tableLog)))/2 + sizeof(U64)/sizeof(U32) /* additional 8 bytes for potential table overwrite */)
#define FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) (sizeof(unsigned) * FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog))
size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
#define FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) (sizeof(short) * (maxSymbolValue + 1) + (1ULL << maxTableLog) + 8)
#define FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) ((FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) + sizeof(unsigned) - 1) / sizeof(unsigned))
FSE_PUBLIC_API size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
/**< Same as FSE_buildDTable(), using an externally allocated `workspace` produced with `FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxSymbolValue)` */
#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + 1 + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) + (FSE_MAX_SYMBOL_VALUE + 1) / 2 + 1)
#define FSE_DECOMPRESS_WKSP_SIZE(maxTableLog, maxSymbolValue) (FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(unsigned))
size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2);
/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)`.
* Set bmi2 to 1 if your CPU supports BMI2 or 0 if it doesn't */
typedef enum {
FSE_repeat_none, /**< Cannot use the previous table */
FSE_repeat_check, /**< Can use the previous table but it must be checked */
FSE_repeat_valid /**< Can use the previous table and it is assumed to be valid */
} FSE_repeat;
/* *****************************************
* FSE symbol compression API
*******************************************/
/*!
This API consists of small unitary functions, which highly benefit from being inlined.
Hence their body are included in next section.
*/
typedef struct {
ptrdiff_t value;
const void* stateTable;
const void* symbolTT;
unsigned stateLog;
} FSE_CState_t;
static void FSE_initCState(FSE_CState_t* CStatePtr, const FSE_CTable* ct);
static void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* CStatePtr, unsigned symbol);
static void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* CStatePtr);
/**<
These functions are inner components of FSE_compress_usingCTable().
They allow the creation of custom streams, mixing multiple tables and bit sources.
A key property to keep in mind is that encoding and decoding are done **in reverse direction**.
So the first symbol you will encode is the last you will decode, like a LIFO stack.
You will need a few variables to track your CStream. They are :
FSE_CTable ct; // Provided by FSE_buildCTable()
BIT_CStream_t bitStream; // bitStream tracking structure
FSE_CState_t state; // State tracking structure (can have several)
The first thing to do is to init bitStream and state.
size_t errorCode = BIT_initCStream(&bitStream, dstBuffer, maxDstSize);
FSE_initCState(&state, ct);
Note that BIT_initCStream() can produce an error code, so its result should be tested, using FSE_isError();
You can then encode your input data, byte after byte.
FSE_encodeSymbol() outputs a maximum of 'tableLog' bits at a time.
Remember decoding will be done in reverse direction.
FSE_encodeByte(&bitStream, &state, symbol);
At any time, you can also add any bit sequence.
Note : maximum allowed nbBits is 25, for compatibility with 32-bits decoders
BIT_addBits(&bitStream, bitField, nbBits);
The above methods don't commit data to memory, they just store it into local register, for speed.
Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (size_t).
Writing data to memory is a manual operation, performed by the flushBits function.
BIT_flushBits(&bitStream);
Your last FSE encoding operation shall be to flush your last state value(s).
FSE_flushState(&bitStream, &state);
Finally, you must close the bitStream.
The function returns the size of CStream in bytes.
If data couldn't fit into dstBuffer, it will return a 0 ( == not compressible)
If there is an error, it returns an errorCode (which can be tested using FSE_isError()).
size_t size = BIT_closeCStream(&bitStream);
*/
/* *****************************************
* FSE symbol decompression API
*******************************************/
typedef struct {
size_t state;
const void* table; /* precise table may vary, depending on U16 */
} FSE_DState_t;
static void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
/**<
Let's now decompose FSE_decompress_usingDTable() into its unitary components.
You will decode FSE-encoded symbols from the bitStream,
and also any other bitFields you put in, **in reverse order**.
You will need a few variables to track your bitStream. They are :
BIT_DStream_t DStream; // Stream context
FSE_DState_t DState; // State context. Multiple ones are possible
FSE_DTable* DTablePtr; // Decoding table, provided by FSE_buildDTable()
The first thing to do is to init the bitStream.
errorCode = BIT_initDStream(&DStream, srcBuffer, srcSize);
You should then retrieve your initial state(s)
(in reverse flushing order if you have several ones) :
errorCode = FSE_initDState(&DState, &DStream, DTablePtr);
You can then decode your data, symbol after symbol.
For information the maximum number of bits read by FSE_decodeSymbol() is 'tableLog'.
Keep in mind that symbols are decoded in reverse order, like a LIFO stack (last in, first out).
unsigned char symbol = FSE_decodeSymbol(&DState, &DStream);
You can retrieve any bitfield you eventually stored into the bitStream (in reverse order)
Note : maximum allowed nbBits is 25, for 32-bits compatibility
size_t bitField = BIT_readBits(&DStream, nbBits);
All above operations only read from local register (which size depends on size_t).
Refueling the register from memory is manually performed by the reload method.
endSignal = FSE_reloadDStream(&DStream);
BIT_reloadDStream() result tells if there is still some more data to read from DStream.
BIT_DStream_unfinished : there is still some data left into the DStream.
BIT_DStream_endOfBuffer : Dstream reached end of buffer. Its container may no longer be completely filled.
BIT_DStream_completed : Dstream reached its exact end, corresponding in general to decompression completed.
BIT_DStream_tooFar : Dstream went too far. Decompression result is corrupted.
When reaching end of buffer (BIT_DStream_endOfBuffer), progress slowly, notably if you decode multiple symbols per loop,
to properly detect the exact end of stream.
After each decoded symbol, check if DStream is fully consumed using this simple test :
BIT_reloadDStream(&DStream) >= BIT_DStream_completed
When it's done, verify decompression is fully completed, by checking both DStream and the relevant states.
Checking if DStream has reached its end is performed by :
BIT_endOfDStream(&DStream);
Check also the states. There might be some symbols left there, if some high probability ones (>50%) are possible.
FSE_endOfDState(&DState);
*/
/* *****************************************
* FSE unsafe API
*******************************************/
static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
/* *****************************************
* Implementation of inlined functions
*******************************************/
typedef struct {
int deltaFindState;
U32 deltaNbBits;
} FSE_symbolCompressionTransform; /* total 8 bytes */
MEM_STATIC void FSE_initCState(FSE_CState_t* statePtr, const FSE_CTable* ct)
{
const void* ptr = ct;
const U16* u16ptr = (const U16*) ptr;
const U32 tableLog = MEM_read16(ptr);
statePtr->value = (ptrdiff_t)1<<tableLog;
statePtr->stateTable = u16ptr+2;
statePtr->symbolTT = ct + 1 + (tableLog ? (1<<(tableLog-1)) : 1);
statePtr->stateLog = tableLog;
}
/*! FSE_initCState2() :
* Same as FSE_initCState(), but the first symbol to include (which will be the last to be read)
* uses the smallest state value possible, saving the cost of this symbol */
MEM_STATIC void FSE_initCState2(FSE_CState_t* statePtr, const FSE_CTable* ct, U32 symbol)
{
FSE_initCState(statePtr, ct);
{ const FSE_symbolCompressionTransform symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
const U16* stateTable = (const U16*)(statePtr->stateTable);
U32 nbBitsOut = (U32)((symbolTT.deltaNbBits + (1<<15)) >> 16);
statePtr->value = (nbBitsOut << 16) - symbolTT.deltaNbBits;
statePtr->value = stateTable[(statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
}
}
MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, unsigned symbol)
{
FSE_symbolCompressionTransform const symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
const U16* const stateTable = (const U16*)(statePtr->stateTable);
U32 const nbBitsOut = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);
BIT_addBits(bitC, statePtr->value, nbBitsOut);
statePtr->value = stateTable[ (statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
}
MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePtr)
{
BIT_addBits(bitC, statePtr->value, statePtr->stateLog);
BIT_flushBits(bitC);
}
/* FSE_getMaxNbBits() :
* Approximate maximum cost of a symbol, in bits.
* Fractional get rounded up (i.e. a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)
* note 1 : assume symbolValue is valid (<= maxSymbolValue)
* note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
MEM_STATIC U32 FSE_getMaxNbBits(const void* symbolTTPtr, U32 symbolValue)
{
const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
return (symbolTT[symbolValue].deltaNbBits + ((1<<16)-1)) >> 16;
}
/* FSE_bitCost() :
* Approximate symbol cost, as fractional value, using fixed-point format (accuracyLog fractional bits)
* note 1 : assume symbolValue is valid (<= maxSymbolValue)
* note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
MEM_STATIC U32 FSE_bitCost(const void* symbolTTPtr, U32 tableLog, U32 symbolValue, U32 accuracyLog)
{
const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
U32 const minNbBits = symbolTT[symbolValue].deltaNbBits >> 16;
U32 const threshold = (minNbBits+1) << 16;
assert(tableLog < 16);
assert(accuracyLog < 31-tableLog); /* ensure enough room for renormalization double shift */
{ U32 const tableSize = 1 << tableLog;
U32 const deltaFromThreshold = threshold - (symbolTT[symbolValue].deltaNbBits + tableSize);
U32 const normalizedDeltaFromThreshold = (deltaFromThreshold << accuracyLog) >> tableLog; /* linear interpolation (very approximate) */
U32 const bitMultiplier = 1 << accuracyLog;
assert(symbolTT[symbolValue].deltaNbBits + tableSize <= threshold);
assert(normalizedDeltaFromThreshold <= bitMultiplier);
return (minNbBits+1)*bitMultiplier - normalizedDeltaFromThreshold;
}
}
/* ====== Decompression ====== */
typedef struct {
U16 tableLog;
U16 fastMode;
} FSE_DTableHeader; /* sizeof U32 */
typedef struct
{
unsigned short newState;
unsigned char symbol;
unsigned char nbBits;
} FSE_decode_t; /* size == U32 */
MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
{
const void* ptr = dt;
const FSE_DTableHeader* const DTableH = (const FSE_DTableHeader*)ptr;
DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
BIT_reloadDStream(bitD);
DStatePtr->table = dt + 1;
}
MEM_STATIC BYTE FSE_peekSymbol(const FSE_DState_t* DStatePtr)
{
FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
return DInfo.symbol;
}
MEM_STATIC void FSE_updateState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
{
FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
U32 const nbBits = DInfo.nbBits;
size_t const lowBits = BIT_readBits(bitD, nbBits);
DStatePtr->state = DInfo.newState + lowBits;
}
MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
{
FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
U32 const nbBits = DInfo.nbBits;
BYTE const symbol = DInfo.symbol;
size_t const lowBits = BIT_readBits(bitD, nbBits);
DStatePtr->state = DInfo.newState + lowBits;
return symbol;
}
/*! FSE_decodeSymbolFast() :
unsafe, only works if no symbol has a probability > 50% */
MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
{
FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
U32 const nbBits = DInfo.nbBits;
BYTE const symbol = DInfo.symbol;
size_t const lowBits = BIT_readBitsFast(bitD, nbBits);
DStatePtr->state = DInfo.newState + lowBits;
return symbol;
}
MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
{
return DStatePtr->state == 0;
}
#ifndef FSE_COMMONDEFS_ONLY
/* **************************************************************
* Tuning parameters
****************************************************************/
/*!MEMORY_USAGE :
* Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
* Increasing memory usage improves compression ratio
* Reduced memory usage can improve speed, due to cache effect
* Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
#ifndef FSE_MAX_MEMORY_USAGE
# define FSE_MAX_MEMORY_USAGE 14
#endif
#ifndef FSE_DEFAULT_MEMORY_USAGE
# define FSE_DEFAULT_MEMORY_USAGE 13
#endif
#if (FSE_DEFAULT_MEMORY_USAGE > FSE_MAX_MEMORY_USAGE)
# error "FSE_DEFAULT_MEMORY_USAGE must be <= FSE_MAX_MEMORY_USAGE"
#endif
/*!FSE_MAX_SYMBOL_VALUE :
* Maximum symbol value authorized.
* Required for proper stack allocation */
#ifndef FSE_MAX_SYMBOL_VALUE
# define FSE_MAX_SYMBOL_VALUE 255
#endif
/* **************************************************************
* template functions type & suffix
****************************************************************/
#define FSE_FUNCTION_TYPE BYTE
#define FSE_FUNCTION_EXTENSION
#define FSE_DECODE_TYPE FSE_decode_t
#endif /* !FSE_COMMONDEFS_ONLY */
/* ***************************************************************
* Constants
*****************************************************************/
#define FSE_MAX_TABLELOG (FSE_MAX_MEMORY_USAGE-2)
#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
#define FSE_MIN_TABLELOG 5
#define FSE_TABLELOG_ABSOLUTE_MAX 15
#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
# error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
#endif
#define FSE_TABLESTEP(tableSize) (((tableSize)>>1) + ((tableSize)>>3) + 3)
#endif /* FSE_STATIC_LINKING_ONLY */
#if defined (__cplusplus)
}
#endif
/**** ended inlining fse.h ****/
/**** start inlining huf.h ****/
/* ******************************************************************
* huff0 huffman codec,
* part of Finite State Entropy library
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
#if defined (__cplusplus)
extern "C" {
#endif
#ifndef HUF_H_298734234
#define HUF_H_298734234
/* *** Dependencies *** */
/**** skipping file: zstd_deps.h ****/
/**** skipping file: mem.h ****/
#define FSE_STATIC_LINKING_ONLY
/**** skipping file: fse.h ****/
/* *** Tool functions *** */
#define HUF_BLOCKSIZE_MAX (128 * 1024) /**< maximum input size for a single block compressed with HUF_compress */
size_t HUF_compressBound(size_t size); /**< maximum compressed size (worst case) */
/* Error Management */
unsigned HUF_isError(size_t code); /**< tells if a return value is an error code */
const char* HUF_getErrorName(size_t code); /**< provides error code string (useful for debugging) */
#define HUF_WORKSPACE_SIZE ((8 << 10) + 512 /* sorting scratch space */)
#define HUF_WORKSPACE_SIZE_U64 (HUF_WORKSPACE_SIZE / sizeof(U64))
/* *** Constants *** */
#define HUF_TABLELOG_MAX 12 /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_TABLELOG_ABSOLUTEMAX */
#define HUF_TABLELOG_DEFAULT 11 /* default tableLog value when none specified */
#define HUF_SYMBOLVALUE_MAX 255
#define HUF_TABLELOG_ABSOLUTEMAX 12 /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
#if (HUF_TABLELOG_MAX > HUF_TABLELOG_ABSOLUTEMAX)
# error "HUF_TABLELOG_MAX is too large !"
#endif
/* ****************************************
* Static allocation
******************************************/
/* HUF buffer bounds */
#define HUF_CTABLEBOUND 129
#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8) /* only true when incompressible is pre-filtered with fast heuristic */
#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size)) /* Macro version, useful for static allocation */
/* static allocation of HUF's Compression Table */
/* this is a private definition, just exposed for allocation and strict aliasing purpose. never EVER access its members directly */
typedef size_t HUF_CElt; /* consider it an incomplete type */
#define HUF_CTABLE_SIZE_ST(maxSymbolValue) ((maxSymbolValue)+2) /* Use tables of size_t, for proper alignment */
#define HUF_CTABLE_SIZE(maxSymbolValue) (HUF_CTABLE_SIZE_ST(maxSymbolValue) * sizeof(size_t))
#define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
HUF_CElt name[HUF_CTABLE_SIZE_ST(maxSymbolValue)] /* no final ; */
/* static allocation of HUF's DTable */
typedef U32 HUF_DTable;
#define HUF_DTABLE_SIZE(maxTableLog) (1 + (1<<(maxTableLog)))
#define HUF_CREATE_STATIC_DTABLEX1(DTable, maxTableLog) \
HUF_DTable DTable[HUF_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1) * 0x01000001) }
#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
HUF_DTable DTable[HUF_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog) * 0x01000001) }
/* ****************************************
* Advanced decompression functions
******************************************/
/**
* Huffman flags bitset.
* For all flags, 0 is the default value.
*/
typedef enum {
/**
* If compiled with DYNAMIC_BMI2: Set flag only if the CPU supports BMI2 at runtime.
* Otherwise: Ignored.
*/
HUF_flags_bmi2 = (1 << 0),
/**
* If set: Test possible table depths to find the one that produces the smallest header + encoded size.
* If unset: Use heuristic to find the table depth.
*/
HUF_flags_optimalDepth = (1 << 1),
/**
* If set: If the previous table can encode the input, always reuse the previous table.
* If unset: If the previous table can encode the input, reuse the previous table if it results in a smaller output.
*/
HUF_flags_preferRepeat = (1 << 2),
/**
* If set: Sample the input and check if the sample is uncompressible, if it is then don't attempt to compress.
* If unset: Always histogram the entire input.
*/
HUF_flags_suspectUncompressible = (1 << 3),
/**
* If set: Don't use assembly implementations
* If unset: Allow using assembly implementations
*/
HUF_flags_disableAsm = (1 << 4),
/**
* If set: Don't use the fast decoding loop, always use the fallback decoding loop.
* If unset: Use the fast decoding loop when possible.
*/
HUF_flags_disableFast = (1 << 5)
} HUF_flags_e;
/* ****************************************
* HUF detailed API
* ****************************************/
#define HUF_OPTIMAL_DEPTH_THRESHOLD ZSTD_btultra
/*! HUF_compress() does the following:
* 1. count symbol occurrence from source[] into table count[] using FSE_count() (exposed within "fse.h")
* 2. (optional) refine tableLog using HUF_optimalTableLog()
* 3. build Huffman table from count using HUF_buildCTable()
* 4. save Huffman table to memory buffer using HUF_writeCTable()
* 5. encode the data stream using HUF_compress4X_usingCTable()
*
* The following API allows targeting specific sub-functions for advanced tasks.
* For example, it's possible to compress several blocks using the same 'CTable',
* or to save and regenerate 'CTable' using external methods.
*/
unsigned HUF_minTableLog(unsigned symbolCardinality);
unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue);
unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, void* workSpace,
size_t wkspSize, HUF_CElt* table, const unsigned* count, int flags); /* table is used as scratch space for building and testing tables, not a return value */
size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog, void* workspace, size_t workspaceSize);
size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
typedef enum {
HUF_repeat_none, /**< Cannot use the previous table */
HUF_repeat_check, /**< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
HUF_repeat_valid /**< Can use the previous table and it is assumed to be valid */
} HUF_repeat;
/** HUF_compress4X_repeat() :
* Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
* If it uses hufTable it does not modify hufTable or repeat.
* If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
* If preferRepeat then the old table will always be used if valid.
* If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned tableLog,
void* workSpace, size_t wkspSize, /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
/** HUF_buildCTable_wksp() :
* Same as HUF_buildCTable(), but using externally allocated scratch buffer.
* `workSpace` must be aligned on 4-bytes boundaries, and its size must be >= HUF_CTABLE_WORKSPACE_SIZE.
*/
#define HUF_CTABLE_WORKSPACE_SIZE_U32 ((4 * (HUF_SYMBOLVALUE_MAX + 1)) + 192)
#define HUF_CTABLE_WORKSPACE_SIZE (HUF_CTABLE_WORKSPACE_SIZE_U32 * sizeof(unsigned))
size_t HUF_buildCTable_wksp (HUF_CElt* tree,
const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
void* workSpace, size_t wkspSize);
/*! HUF_readStats() :
* Read compact Huffman tree, saved by HUF_writeCTable().
* `huffWeight` is destination buffer.
* @return : size read from `src` , or an error Code .
* Note : Needed by HUF_readCTable() and HUF_readDTableXn() . */
size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize);
/*! HUF_readStats_wksp() :
* Same as HUF_readStats() but takes an external workspace which must be
* 4-byte aligned and its size must be >= HUF_READ_STATS_WORKSPACE_SIZE.
* If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
*/
#define HUF_READ_STATS_WORKSPACE_SIZE_U32 FSE_DECOMPRESS_WKSP_SIZE_U32(6, HUF_TABLELOG_MAX-1)
#define HUF_READ_STATS_WORKSPACE_SIZE (HUF_READ_STATS_WORKSPACE_SIZE_U32 * sizeof(unsigned))
size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workspace, size_t wkspSize,
int flags);
/** HUF_readCTable() :
* Loading a CTable saved with HUF_writeCTable() */
size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);
/** HUF_getNbBitsFromCTable() :
* Read nbBits from CTable symbolTable, for symbol `symbolValue` presumed <= HUF_SYMBOLVALUE_MAX
* Note 1 : If symbolValue > HUF_readCTableHeader(symbolTable).maxSymbolValue, returns 0
* Note 2 : is not inlined, as HUF_CElt definition is private
*/
U32 HUF_getNbBitsFromCTable(const HUF_CElt* symbolTable, U32 symbolValue);
typedef struct {
BYTE tableLog;
BYTE maxSymbolValue;
BYTE unused[sizeof(size_t) - 2];
} HUF_CTableHeader;
/** HUF_readCTableHeader() :
* @returns The header from the CTable specifying the tableLog and the maxSymbolValue.
*/
HUF_CTableHeader HUF_readCTableHeader(HUF_CElt const* ctable);
/*
* HUF_decompress() does the following:
* 1. select the decompression algorithm (X1, X2) based on pre-computed heuristics
* 2. build Huffman table from save, using HUF_readDTableX?()
* 3. decode 1 or 4 segments in parallel using HUF_decompress?X?_usingDTable()
*/
/** HUF_selectDecoder() :
* Tells which decoder is likely to decode faster,
* based on a set of pre-computed metrics.
* @return : 0==HUF_decompress4X1, 1==HUF_decompress4X2 .
* Assumption : 0 < dstSize <= 128 KB */
U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
/**
* The minimum workspace size for the `workSpace` used in
* HUF_readDTableX1_wksp() and HUF_readDTableX2_wksp().
*
* The space used depends on HUF_TABLELOG_MAX, ranging from ~1500 bytes when
* HUF_TABLE_LOG_MAX=12 to ~1850 bytes when HUF_TABLE_LOG_MAX=15.
* Buffer overflow errors may potentially occur if code modifications result in
* a required workspace size greater than that specified in the following
* macro.
*/
#define HUF_DECOMPRESS_WORKSPACE_SIZE ((2 << 10) + (1 << 9))
#define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
/* ====================== */
/* single stream variants */
/* ====================== */
size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
/** HUF_compress1X_repeat() :
* Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
* If it uses hufTable it does not modify hufTable or repeat.
* If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
* If preferRepeat then the old table will always be used if valid.
* If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned tableLog,
void* workSpace, size_t wkspSize, /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
#ifndef HUF_FORCE_DECOMPRESS_X1
size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags); /**< double-symbols decoder */
#endif
/* BMI2 variants.
* If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
*/
size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
#ifndef HUF_FORCE_DECOMPRESS_X2
size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
#endif
size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
#ifndef HUF_FORCE_DECOMPRESS_X2
size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
#endif
#ifndef HUF_FORCE_DECOMPRESS_X1
size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
#endif
#endif /* HUF_H_298734234 */
#if defined (__cplusplus)
}
#endif
/**** ended inlining huf.h ****/
/**** skipping file: bits.h ****/
/*=== Version ===*/
unsigned FSE_versionNumber(void) { return FSE_VERSION_NUMBER; }
/*=== Error Management ===*/
unsigned FSE_isError(size_t code) { return ERR_isError(code); }
const char* FSE_getErrorName(size_t code) { return ERR_getErrorName(code); }
unsigned HUF_isError(size_t code) { return ERR_isError(code); }
const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
/*-**************************************************************
* FSE NCount encoding-decoding
****************************************************************/
FORCE_INLINE_TEMPLATE
size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
{
const BYTE* const istart = (const BYTE*) headerBuffer;
const BYTE* const iend = istart + hbSize;
const BYTE* ip = istart;
int nbBits;
int remaining;
int threshold;
U32 bitStream;
int bitCount;
unsigned charnum = 0;
unsigned const maxSV1 = *maxSVPtr + 1;
int previous0 = 0;
if (hbSize < 8) {
/* This function only works when hbSize >= 8 */
char buffer[8] = {0};
ZSTD_memcpy(buffer, headerBuffer, hbSize);
{ size_t const countSize = FSE_readNCount(normalizedCounter, maxSVPtr, tableLogPtr,
buffer, sizeof(buffer));
if (FSE_isError(countSize)) return countSize;
if (countSize > hbSize) return ERROR(corruption_detected);
return countSize;
} }
assert(hbSize >= 8);
/* init */
ZSTD_memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0])); /* all symbols not present in NCount have a frequency of 0 */
bitStream = MEM_readLE32(ip);
nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG; /* extract tableLog */
if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
bitStream >>= 4;
bitCount = 4;
*tableLogPtr = nbBits;
remaining = (1<<nbBits)+1;
threshold = 1<<nbBits;
nbBits++;
for (;;) {
if (previous0) {
/* Count the number of repeats. Each time the
* 2-bit repeat code is 0b11 there is another
* repeat.
* Avoid UB by setting the high bit to 1.
*/
int repeats = ZSTD_countTrailingZeros32(~bitStream | 0x80000000) >> 1;
while (repeats >= 12) {
charnum += 3 * 12;
if (LIKELY(ip <= iend-7)) {
ip += 3;
} else {
bitCount -= (int)(8 * (iend - 7 - ip));
bitCount &= 31;
ip = iend - 4;
}
bitStream = MEM_readLE32(ip) >> bitCount;
repeats = ZSTD_countTrailingZeros32(~bitStream | 0x80000000) >> 1;
}
charnum += 3 * repeats;
bitStream >>= 2 * repeats;
bitCount += 2 * repeats;
/* Add the final repeat which isn't 0b11. */
assert((bitStream & 3) < 3);
charnum += bitStream & 3;
bitCount += 2;
/* This is an error, but break and return an error
* at the end, because returning out of a loop makes
* it harder for the compiler to optimize.
*/
if (charnum >= maxSV1) break;
/* We don't need to set the normalized count to 0
* because we already memset the whole buffer to 0.
*/
if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
assert((bitCount >> 3) <= 3); /* For first condition to work */
ip += bitCount>>3;
bitCount &= 7;
} else {
bitCount -= (int)(8 * (iend - 4 - ip));
bitCount &= 31;
ip = iend - 4;
}
bitStream = MEM_readLE32(ip) >> bitCount;
}
{
int const max = (2*threshold-1) - remaining;
int count;
if ((bitStream & (threshold-1)) < (U32)max) {
count = bitStream & (threshold-1);
bitCount += nbBits-1;
} else {
count = bitStream & (2*threshold-1);
if (count >= threshold) count -= max;
bitCount += nbBits;
}
count--; /* extra accuracy */
/* When it matters (small blocks), this is a
* predictable branch, because we don't use -1.
*/
if (count >= 0) {
remaining -= count;
} else {
assert(count == -1);
remaining += count;
}
normalizedCounter[charnum++] = (short)count;
previous0 = !count;
assert(threshold > 1);
if (remaining < threshold) {
/* This branch can be folded into the
* threshold update condition because we
* know that threshold > 1.
*/
if (remaining <= 1) break;
nbBits = ZSTD_highbit32(remaining) + 1;
threshold = 1 << (nbBits - 1);
}
if (charnum >= maxSV1) break;
if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
ip += bitCount>>3;
bitCount &= 7;
} else {
bitCount -= (int)(8 * (iend - 4 - ip));
bitCount &= 31;
ip = iend - 4;
}
bitStream = MEM_readLE32(ip) >> bitCount;
} }
if (remaining != 1) return ERROR(corruption_detected);
/* Only possible when there are too many zeros. */
if (charnum > maxSV1) return ERROR(maxSymbolValue_tooSmall);
if (bitCount > 32) return ERROR(corruption_detected);
*maxSVPtr = charnum-1;
ip += (bitCount+7)>>3;
return ip-istart;
}
/* Avoids the FORCE_INLINE of the _body() function. */
static size_t FSE_readNCount_body_default(
short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
{
return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
}
#if DYNAMIC_BMI2
BMI2_TARGET_ATTRIBUTE static size_t FSE_readNCount_body_bmi2(
short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
{
return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
}
#endif
size_t FSE_readNCount_bmi2(
short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize, int bmi2)
{
#if DYNAMIC_BMI2
if (bmi2) {
return FSE_readNCount_body_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
}
#endif
(void)bmi2;
return FSE_readNCount_body_default(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
}
size_t FSE_readNCount(
short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
{
return FSE_readNCount_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize, /* bmi2 */ 0);
}
/*! HUF_readStats() :
Read compact Huffman tree, saved by HUF_writeCTable().
`huffWeight` is destination buffer.
`rankStats` is assumed to be a table of at least HUF_TABLELOG_MAX U32.
@return : size read from `src` , or an error Code .
Note : Needed by HUF_readCTable() and HUF_readDTableX?() .
*/
size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize)
{
U32 wksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp), /* flags */ 0);
}
FORCE_INLINE_TEMPLATE size_t
HUF_readStats_body(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize,
int bmi2)
{
U32 weightTotal;
const BYTE* ip = (const BYTE*) src;
size_t iSize;
size_t oSize;
if (!srcSize) return ERROR(srcSize_wrong);
iSize = ip[0];
/* ZSTD_memset(huffWeight, 0, hwSize); *//* is not necessary, even though some analyzer complain ... */
if (iSize >= 128) { /* special header */
oSize = iSize - 127;
iSize = ((oSize+1)/2);
if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
if (oSize >= hwSize) return ERROR(corruption_detected);
ip += 1;
{ U32 n;
for (n=0; n<oSize; n+=2) {
huffWeight[n] = ip[n/2] >> 4;
huffWeight[n+1] = ip[n/2] & 15;
} } }
else { /* header compressed with FSE (normal case) */
if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
/* max (hwSize-1) values decoded, as last one is implied */
oSize = FSE_decompress_wksp_bmi2(huffWeight, hwSize-1, ip+1, iSize, 6, workSpace, wkspSize, bmi2);
if (FSE_isError(oSize)) return oSize;
}
/* collect weight stats */
ZSTD_memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
weightTotal = 0;
{ U32 n; for (n=0; n<oSize; n++) {
if (huffWeight[n] > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
rankStats[huffWeight[n]]++;
weightTotal += (1 << huffWeight[n]) >> 1;
} }
if (weightTotal == 0) return ERROR(corruption_detected);
/* get last non-null symbol weight (implied, total must be 2^n) */
{ U32 const tableLog = ZSTD_highbit32(weightTotal) + 1;
if (tableLog > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
*tableLogPtr = tableLog;
/* determine last weight */
{ U32 const total = 1 << tableLog;
U32 const rest = total - weightTotal;
U32 const verif = 1 << ZSTD_highbit32(rest);
U32 const lastWeight = ZSTD_highbit32(rest) + 1;
if (verif != rest) return ERROR(corruption_detected); /* last value must be a clean power of 2 */
huffWeight[oSize] = (BYTE)lastWeight;
rankStats[lastWeight]++;
} }
/* check tree construction validity */
if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected); /* by construction : at least 2 elts of rank 1, must be even */
/* results */
*nbSymbolsPtr = (U32)(oSize+1);
return iSize+1;
}
/* Avoids the FORCE_INLINE of the _body() function. */
static size_t HUF_readStats_body_default(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize)
{
return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 0);
}
#if DYNAMIC_BMI2
static BMI2_TARGET_ATTRIBUTE size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize)
{
return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 1);
}
#endif
size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize,
int flags)
{
#if DYNAMIC_BMI2
if (flags & HUF_flags_bmi2) {
return HUF_readStats_body_bmi2(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
}
#endif
(void)flags;
return HUF_readStats_body_default(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
}
/**** ended inlining common/entropy_common.c ****/
/**** start inlining common/error_private.c ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/* The purpose of this file is to have a single list of error strings embedded in binary */
/**** skipping file: error_private.h ****/
const char* ERR_getErrorString(ERR_enum code)
{
#ifdef ZSTD_STRIP_ERROR_STRINGS
(void)code;
return "Error strings stripped";
#else
static const char* const notErrorCode = "Unspecified error code";
switch( code )
{
case PREFIX(no_error): return "No error detected";
case PREFIX(GENERIC): return "Error (generic)";
case PREFIX(prefix_unknown): return "Unknown frame descriptor";
case PREFIX(version_unsupported): return "Version not supported";
case PREFIX(frameParameter_unsupported): return "Unsupported frame parameter";
case PREFIX(frameParameter_windowTooLarge): return "Frame requires too much memory for decoding";
case PREFIX(corruption_detected): return "Data corruption detected";
case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
case PREFIX(literals_headerWrong): return "Header of Literals' block doesn't respect format specification";
case PREFIX(parameter_unsupported): return "Unsupported parameter";
case PREFIX(parameter_combination_unsupported): return "Unsupported combination of parameters";
case PREFIX(parameter_outOfBound): return "Parameter is out of bound";
case PREFIX(init_missing): return "Context should be init first";
case PREFIX(memory_allocation): return "Allocation error : not enough memory";
case PREFIX(workSpace_tooSmall): return "workSpace buffer is not large enough";
case PREFIX(stage_wrong): return "Operation not authorized at current processing stage";
case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";
case PREFIX(stabilityCondition_notRespected): return "pledged buffer stability condition is not respected";
case PREFIX(dictionary_corrupted): return "Dictionary is corrupted";
case PREFIX(dictionary_wrong): return "Dictionary mismatch";
case PREFIX(dictionaryCreation_failed): return "Cannot create Dictionary from provided samples";
case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
case PREFIX(srcSize_wrong): return "Src size is incorrect";
case PREFIX(dstBuffer_null): return "Operation on NULL destination buffer";
case PREFIX(noForwardProgress_destFull): return "Operation made no progress over multiple calls, due to output buffer being full";
case PREFIX(noForwardProgress_inputEmpty): return "Operation made no progress over multiple calls, due to input being empty";
/* following error codes are not stable and may be removed or changed in a future version */
case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
case PREFIX(dstBuffer_wrong): return "Destination buffer is wrong";
case PREFIX(srcBuffer_wrong): return "Source buffer is wrong";
case PREFIX(sequenceProducer_failed): return "Block-level external sequence producer returned an error code";
case PREFIX(externalSequences_invalid): return "External sequences are not valid";
case PREFIX(maxCode):
default: return notErrorCode;
}
#endif
}
/**** ended inlining common/error_private.c ****/
/**** start inlining common/fse_decompress.c ****/
/* ******************************************************************
* FSE : Finite State Entropy decoder
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
* - Public forum : https://groups.google.com/forum/#!forum/lz4c
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/* **************************************************************
* Includes
****************************************************************/
/**** skipping file: debug.h ****/
/**** skipping file: bitstream.h ****/
/**** skipping file: compiler.h ****/
#define FSE_STATIC_LINKING_ONLY
/**** skipping file: fse.h ****/
/**** skipping file: error_private.h ****/
#define ZSTD_DEPS_NEED_MALLOC
/**** skipping file: zstd_deps.h ****/
/**** skipping file: bits.h ****/
/* **************************************************************
* Error Management
****************************************************************/
#define FSE_isError ERR_isError
#define FSE_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c) /* use only *after* variable declarations */
/* **************************************************************
* Templates
****************************************************************/
/*
designed to be included
for type-specific functions (template emulation in C)
Objective is to write these functions only once, for improved maintenance
*/
/* safety checks */
#ifndef FSE_FUNCTION_EXTENSION
# error "FSE_FUNCTION_EXTENSION must be defined"
#endif
#ifndef FSE_FUNCTION_TYPE
# error "FSE_FUNCTION_TYPE must be defined"
#endif
/* Function names */
#define FSE_CAT(X,Y) X##Y
#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
{
void* const tdPtr = dt+1; /* because *dt is unsigned, 32-bits aligned on 32-bits */
FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
U16* symbolNext = (U16*)workSpace;
BYTE* spread = (BYTE*)(symbolNext + maxSymbolValue + 1);
U32 const maxSV1 = maxSymbolValue + 1;
U32 const tableSize = 1 << tableLog;
U32 highThreshold = tableSize-1;
/* Sanity Checks */
if (FSE_BUILD_DTABLE_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(maxSymbolValue_tooLarge);
if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
/* Init, lay down lowprob symbols */
{ FSE_DTableHeader DTableH;
DTableH.tableLog = (U16)tableLog;
DTableH.fastMode = 1;
{ S16 const largeLimit= (S16)(1 << (tableLog-1));
U32 s;
for (s=0; s<maxSV1; s++) {
if (normalizedCounter[s]==-1) {
tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
symbolNext[s] = 1;
} else {
if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
symbolNext[s] = normalizedCounter[s];
} } }
ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
}
/* Spread symbols */
if (highThreshold == tableSize - 1) {
size_t const tableMask = tableSize-1;
size_t const step = FSE_TABLESTEP(tableSize);
/* First lay down the symbols in order.
* We use a uint64_t to lay down 8 bytes at a time. This reduces branch
* misses since small blocks generally have small table logs, so nearly
* all symbols have counts <= 8. We ensure we have 8 bytes at the end of
* our buffer to handle the over-write.
*/
{
U64 const add = 0x0101010101010101ull;
size_t pos = 0;
U64 sv = 0;
U32 s;
for (s=0; s<maxSV1; ++s, sv += add) {
int i;
int const n = normalizedCounter[s];
MEM_write64(spread + pos, sv);
for (i = 8; i < n; i += 8) {
MEM_write64(spread + pos + i, sv);
}
pos += n;
}
}
/* Now we spread those positions across the table.
* The benefit of doing it in two stages is that we avoid the
* variable size inner loop, which caused lots of branch misses.
* Now we can run through all the positions without any branch misses.
* We unroll the loop twice, since that is what empirically worked best.
*/
{
size_t position = 0;
size_t s;
size_t const unroll = 2;
assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
for (s = 0; s < (size_t)tableSize; s += unroll) {
size_t u;
for (u = 0; u < unroll; ++u) {
size_t const uPosition = (position + (u * step)) & tableMask;
tableDecode[uPosition].symbol = spread[s + u];
}
position = (position + (unroll * step)) & tableMask;
}
assert(position == 0);
}
} else {
U32 const tableMask = tableSize-1;
U32 const step = FSE_TABLESTEP(tableSize);
U32 s, position = 0;
for (s=0; s<maxSV1; s++) {
int i;
for (i=0; i<normalizedCounter[s]; i++) {
tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
position = (position + step) & tableMask;
while (position > highThreshold) position = (position + step) & tableMask; /* lowprob area */
} }
if (position!=0) return ERROR(GENERIC); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
}
/* Build Decoding table */
{ U32 u;
for (u=0; u<tableSize; u++) {
FSE_FUNCTION_TYPE const symbol = (FSE_FUNCTION_TYPE)(tableDecode[u].symbol);
U32 const nextState = symbolNext[symbol]++;
tableDecode[u].nbBits = (BYTE) (tableLog - ZSTD_highbit32(nextState) );
tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
} }
return 0;
}
size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
{
return FSE_buildDTable_internal(dt, normalizedCounter, maxSymbolValue, tableLog, workSpace, wkspSize);
}
#ifndef FSE_COMMONDEFS_ONLY
/*-*******************************************************
* Decompression (Byte symbols)
*********************************************************/
FORCE_INLINE_TEMPLATE size_t FSE_decompress_usingDTable_generic(
void* dst, size_t maxDstSize,
const void* cSrc, size_t cSrcSize,
const FSE_DTable* dt, const unsigned fast)
{
BYTE* const ostart = (BYTE*) dst;
BYTE* op = ostart;
BYTE* const omax = op + maxDstSize;
BYTE* const olimit = omax-3;
BIT_DStream_t bitD;
FSE_DState_t state1;
FSE_DState_t state2;
/* Init */
CHECK_F(BIT_initDStream(&bitD, cSrc, cSrcSize));
FSE_initDState(&state1, &bitD, dt);
FSE_initDState(&state2, &bitD, dt);
#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
/* 4 symbols per loop */
for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) & (op<olimit) ; op+=4) {
op[0] = FSE_GETSYMBOL(&state1);
if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8) /* This test must be static */
BIT_reloadDStream(&bitD);
op[1] = FSE_GETSYMBOL(&state2);
if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8) /* This test must be static */
{ if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
op[2] = FSE_GETSYMBOL(&state1);
if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8) /* This test must be static */
BIT_reloadDStream(&bitD);
op[3] = FSE_GETSYMBOL(&state2);
}
/* tail */
/* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
while (1) {
if (op>(omax-2)) return ERROR(dstSize_tooSmall);
*op++ = FSE_GETSYMBOL(&state1);
if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
*op++ = FSE_GETSYMBOL(&state2);
break;
}
if (op>(omax-2)) return ERROR(dstSize_tooSmall);
*op++ = FSE_GETSYMBOL(&state2);
if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
*op++ = FSE_GETSYMBOL(&state1);
break;
} }
return op-ostart;
}
typedef struct {
short ncount[FSE_MAX_SYMBOL_VALUE + 1];
FSE_DTable dtable[1]; /* Dynamically sized */
} FSE_DecompressWksp;
FORCE_INLINE_TEMPLATE size_t FSE_decompress_wksp_body(
void* dst, size_t dstCapacity,
const void* cSrc, size_t cSrcSize,
unsigned maxLog, void* workSpace, size_t wkspSize,
int bmi2)
{
const BYTE* const istart = (const BYTE*)cSrc;
const BYTE* ip = istart;
unsigned tableLog;
unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
FSE_DecompressWksp* const wksp = (FSE_DecompressWksp*)workSpace;
DEBUG_STATIC_ASSERT((FSE_MAX_SYMBOL_VALUE + 1) % 2 == 0);
if (wkspSize < sizeof(*wksp)) return ERROR(GENERIC);
/* normal FSE decoding mode */
{
size_t const NCountLength = FSE_readNCount_bmi2(wksp->ncount, &maxSymbolValue, &tableLog, istart, cSrcSize, bmi2);
if (FSE_isError(NCountLength)) return NCountLength;
if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
assert(NCountLength <= cSrcSize);
ip += NCountLength;
cSrcSize -= NCountLength;
}
if (FSE_DECOMPRESS_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(tableLog_tooLarge);
assert(sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog) <= wkspSize);
workSpace = (BYTE*)workSpace + sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
wkspSize -= sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
CHECK_F( FSE_buildDTable_internal(wksp->dtable, wksp->ncount, maxSymbolValue, tableLog, workSpace, wkspSize) );
{
const void* ptr = wksp->dtable;
const FSE_DTableHeader* DTableH = (const FSE_DTableHeader*)ptr;
const U32 fastMode = DTableH->fastMode;
/* select fast mode (static) */
if (fastMode) return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 1);
return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 0);
}
}
/* Avoids the FORCE_INLINE of the _body() function. */
static size_t FSE_decompress_wksp_body_default(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
{
return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 0);
}
#if DYNAMIC_BMI2
BMI2_TARGET_ATTRIBUTE static size_t FSE_decompress_wksp_body_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
{
return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 1);
}
#endif
size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2)
{
#if DYNAMIC_BMI2
if (bmi2) {
return FSE_decompress_wksp_body_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
}
#endif
(void)bmi2;
return FSE_decompress_wksp_body_default(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
}
#endif /* FSE_COMMONDEFS_ONLY */
/**** ended inlining common/fse_decompress.c ****/
/**** start inlining common/threading.c ****/
/**
* Copyright (c) 2016 Tino Reichardt
* All rights reserved.
*
* You can contact the author at:
* - zstdmt source repository: https://github.com/mcmilk/zstdmt
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/**
* This file will hold wrapper for systems, which do not support pthreads
*/
/**** start inlining threading.h ****/
/**
* Copyright (c) 2016 Tino Reichardt
* All rights reserved.
*
* You can contact the author at:
* - zstdmt source repository: https://github.com/mcmilk/zstdmt
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef THREADING_H_938743
#define THREADING_H_938743
/**** skipping file: debug.h ****/
#if defined (__cplusplus)
extern "C" {
#endif
#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
/**
* Windows minimalist Pthread Wrapper
*/
#ifdef WINVER
# undef WINVER
#endif
#define WINVER 0x0600
#ifdef _WIN32_WINNT
# undef _WIN32_WINNT
#endif
#define _WIN32_WINNT 0x0600
#ifndef WIN32_LEAN_AND_MEAN
# define WIN32_LEAN_AND_MEAN
#endif
#undef ERROR /* reported already defined on VS 2015 (Rich Geldreich) */
#include <windows.h>
#undef ERROR
#define ERROR(name) ZSTD_ERROR(name)
/* mutex */
#define ZSTD_pthread_mutex_t CRITICAL_SECTION
#define ZSTD_pthread_mutex_init(a, b) ((void)(b), InitializeCriticalSection((a)), 0)
#define ZSTD_pthread_mutex_destroy(a) DeleteCriticalSection((a))
#define ZSTD_pthread_mutex_lock(a) EnterCriticalSection((a))
#define ZSTD_pthread_mutex_unlock(a) LeaveCriticalSection((a))
/* condition variable */
#define ZSTD_pthread_cond_t CONDITION_VARIABLE
#define ZSTD_pthread_cond_init(a, b) ((void)(b), InitializeConditionVariable((a)), 0)
#define ZSTD_pthread_cond_destroy(a) ((void)(a))
#define ZSTD_pthread_cond_wait(a, b) SleepConditionVariableCS((a), (b), INFINITE)
#define ZSTD_pthread_cond_signal(a) WakeConditionVariable((a))
#define ZSTD_pthread_cond_broadcast(a) WakeAllConditionVariable((a))
/* ZSTD_pthread_create() and ZSTD_pthread_join() */
typedef HANDLE ZSTD_pthread_t;
int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
void* (*start_routine) (void*), void* arg);
int ZSTD_pthread_join(ZSTD_pthread_t thread);
/**
* add here more wrappers as required
*/
#elif defined(ZSTD_MULTITHREAD) /* posix assumed ; need a better detection method */
/* === POSIX Systems === */
# include <pthread.h>
#if DEBUGLEVEL < 1
#define ZSTD_pthread_mutex_t pthread_mutex_t
#define ZSTD_pthread_mutex_init(a, b) pthread_mutex_init((a), (b))
#define ZSTD_pthread_mutex_destroy(a) pthread_mutex_destroy((a))
#define ZSTD_pthread_mutex_lock(a) pthread_mutex_lock((a))
#define ZSTD_pthread_mutex_unlock(a) pthread_mutex_unlock((a))
#define ZSTD_pthread_cond_t pthread_cond_t
#define ZSTD_pthread_cond_init(a, b) pthread_cond_init((a), (b))
#define ZSTD_pthread_cond_destroy(a) pthread_cond_destroy((a))
#define ZSTD_pthread_cond_wait(a, b) pthread_cond_wait((a), (b))
#define ZSTD_pthread_cond_signal(a) pthread_cond_signal((a))
#define ZSTD_pthread_cond_broadcast(a) pthread_cond_broadcast((a))
#define ZSTD_pthread_t pthread_t
#define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
#define ZSTD_pthread_join(a) pthread_join((a),NULL)
#else /* DEBUGLEVEL >= 1 */
/* Debug implementation of threading.
* In this implementation we use pointers for mutexes and condition variables.
* This way, if we forget to init/destroy them the program will crash or ASAN
* will report leaks.
*/
#define ZSTD_pthread_mutex_t pthread_mutex_t*
int ZSTD_pthread_mutex_init(ZSTD_pthread_mutex_t* mutex, pthread_mutexattr_t const* attr);
int ZSTD_pthread_mutex_destroy(ZSTD_pthread_mutex_t* mutex);
#define ZSTD_pthread_mutex_lock(a) pthread_mutex_lock(*(a))
#define ZSTD_pthread_mutex_unlock(a) pthread_mutex_unlock(*(a))
#define ZSTD_pthread_cond_t pthread_cond_t*
int ZSTD_pthread_cond_init(ZSTD_pthread_cond_t* cond, pthread_condattr_t const* attr);
int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond);
#define ZSTD_pthread_cond_wait(a, b) pthread_cond_wait(*(a), *(b))
#define ZSTD_pthread_cond_signal(a) pthread_cond_signal(*(a))
#define ZSTD_pthread_cond_broadcast(a) pthread_cond_broadcast(*(a))
#define ZSTD_pthread_t pthread_t
#define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
#define ZSTD_pthread_join(a) pthread_join((a),NULL)
#endif
#else /* ZSTD_MULTITHREAD not defined */
/* No multithreading support */
typedef int ZSTD_pthread_mutex_t;
#define ZSTD_pthread_mutex_init(a, b) ((void)(a), (void)(b), 0)
#define ZSTD_pthread_mutex_destroy(a) ((void)(a))
#define ZSTD_pthread_mutex_lock(a) ((void)(a))
#define ZSTD_pthread_mutex_unlock(a) ((void)(a))
typedef int ZSTD_pthread_cond_t;
#define ZSTD_pthread_cond_init(a, b) ((void)(a), (void)(b), 0)
#define ZSTD_pthread_cond_destroy(a) ((void)(a))
#define ZSTD_pthread_cond_wait(a, b) ((void)(a), (void)(b))
#define ZSTD_pthread_cond_signal(a) ((void)(a))
#define ZSTD_pthread_cond_broadcast(a) ((void)(a))
/* do not use ZSTD_pthread_t */
#endif /* ZSTD_MULTITHREAD */
#if defined (__cplusplus)
}
#endif
#endif /* THREADING_H_938743 */
/**** ended inlining threading.h ****/
/* create fake symbol to avoid empty translation unit warning */
int g_ZSTD_threading_useless_symbol;
#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
/**
* Windows minimalist Pthread Wrapper
*/
/* === Dependencies === */
#include <process.h>
#include <errno.h>
/* === Implementation === */
typedef struct {
void* (*start_routine)(void*);
void* arg;
int initialized;
ZSTD_pthread_cond_t initialized_cond;
ZSTD_pthread_mutex_t initialized_mutex;
} ZSTD_thread_params_t;
static unsigned __stdcall worker(void *arg)
{
void* (*start_routine)(void*);
void* thread_arg;
/* Initialized thread_arg and start_routine and signal main thread that we don't need it
* to wait any longer.
*/
{
ZSTD_thread_params_t* thread_param = (ZSTD_thread_params_t*)arg;
thread_arg = thread_param->arg;
start_routine = thread_param->start_routine;
/* Signal main thread that we are running and do not depend on its memory anymore */
ZSTD_pthread_mutex_lock(&thread_param->initialized_mutex);
thread_param->initialized = 1;
ZSTD_pthread_cond_signal(&thread_param->initialized_cond);
ZSTD_pthread_mutex_unlock(&thread_param->initialized_mutex);
}
start_routine(thread_arg);
return 0;
}
int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
void* (*start_routine) (void*), void* arg)
{
ZSTD_thread_params_t thread_param;
(void)unused;
if (thread==NULL) return -1;
*thread = NULL;
thread_param.start_routine = start_routine;
thread_param.arg = arg;
thread_param.initialized = 0;
/* Setup thread initialization synchronization */
if(ZSTD_pthread_cond_init(&thread_param.initialized_cond, NULL)) {
/* Should never happen on Windows */
return -1;
}
if(ZSTD_pthread_mutex_init(&thread_param.initialized_mutex, NULL)) {
/* Should never happen on Windows */
ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
return -1;
}
/* Spawn thread */
*thread = (HANDLE)_beginthreadex(NULL, 0, worker, &thread_param, 0, NULL);
if (*thread==NULL) {
ZSTD_pthread_mutex_destroy(&thread_param.initialized_mutex);
ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
return errno;
}
/* Wait for thread to be initialized */
ZSTD_pthread_mutex_lock(&thread_param.initialized_mutex);
while(!thread_param.initialized) {
ZSTD_pthread_cond_wait(&thread_param.initialized_cond, &thread_param.initialized_mutex);
}
ZSTD_pthread_mutex_unlock(&thread_param.initialized_mutex);
ZSTD_pthread_mutex_destroy(&thread_param.initialized_mutex);
ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
return 0;
}
int ZSTD_pthread_join(ZSTD_pthread_t thread)
{
DWORD result;
if (!thread) return 0;
result = WaitForSingleObject(thread, INFINITE);
CloseHandle(thread);
switch (result) {
case WAIT_OBJECT_0:
return 0;
case WAIT_ABANDONED:
return EINVAL;
default:
return GetLastError();
}
}
#endif /* ZSTD_MULTITHREAD */
#if defined(ZSTD_MULTITHREAD) && DEBUGLEVEL >= 1 && !defined(_WIN32)
#define ZSTD_DEPS_NEED_MALLOC
/**** skipping file: zstd_deps.h ****/
int ZSTD_pthread_mutex_init(ZSTD_pthread_mutex_t* mutex, pthread_mutexattr_t const* attr)
{
assert(mutex != NULL);
*mutex = (pthread_mutex_t*)ZSTD_malloc(sizeof(pthread_mutex_t));
if (!*mutex)
return 1;
return pthread_mutex_init(*mutex, attr);
}
int ZSTD_pthread_mutex_destroy(ZSTD_pthread_mutex_t* mutex)
{
assert(mutex != NULL);
if (!*mutex)
return 0;
{
int const ret = pthread_mutex_destroy(*mutex);
ZSTD_free(*mutex);
return ret;
}
}
int ZSTD_pthread_cond_init(ZSTD_pthread_cond_t* cond, pthread_condattr_t const* attr)
{
assert(cond != NULL);
*cond = (pthread_cond_t*)ZSTD_malloc(sizeof(pthread_cond_t));
if (!*cond)
return 1;
return pthread_cond_init(*cond, attr);
}
int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond)
{
assert(cond != NULL);
if (!*cond)
return 0;
{
int const ret = pthread_cond_destroy(*cond);
ZSTD_free(*cond);
return ret;
}
}
#endif
/**** ended inlining common/threading.c ****/
/**** start inlining common/pool.c ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/* ====== Dependencies ======= */
/**** start inlining ../common/allocations.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/* This file provides custom allocation primitives
*/
#define ZSTD_DEPS_NEED_MALLOC
/**** skipping file: zstd_deps.h ****/
/**** skipping file: compiler.h ****/
#define ZSTD_STATIC_LINKING_ONLY
/**** start inlining ../zstd.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#if defined (__cplusplus)
extern "C" {
#endif
#ifndef ZSTD_H_235446
#define ZSTD_H_235446
/* ====== Dependencies ======*/
#include <limits.h> /* INT_MAX */
#include <stddef.h> /* size_t */
/* ===== ZSTDLIB_API : control library symbols visibility ===== */
#ifndef ZSTDLIB_VISIBLE
/* Backwards compatibility with old macro name */
# ifdef ZSTDLIB_VISIBILITY
# define ZSTDLIB_VISIBLE ZSTDLIB_VISIBILITY
# elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
# define ZSTDLIB_VISIBLE __attribute__ ((visibility ("default")))
# else
# define ZSTDLIB_VISIBLE
# endif
#endif
#ifndef ZSTDLIB_HIDDEN
# if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
# define ZSTDLIB_HIDDEN __attribute__ ((visibility ("hidden")))
# else
# define ZSTDLIB_HIDDEN
# endif
#endif
#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
# define ZSTDLIB_API __declspec(dllexport) ZSTDLIB_VISIBLE
#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
# define ZSTDLIB_API __declspec(dllimport) ZSTDLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
#else
# define ZSTDLIB_API ZSTDLIB_VISIBLE
#endif
/* Deprecation warnings :
* Should these warnings be a problem, it is generally possible to disable them,
* typically with -Wno-deprecated-declarations for gcc or _CRT_SECURE_NO_WARNINGS in Visual.
* Otherwise, it's also possible to define ZSTD_DISABLE_DEPRECATE_WARNINGS.
*/
#ifdef ZSTD_DISABLE_DEPRECATE_WARNINGS
# define ZSTD_DEPRECATED(message) /* disable deprecation warnings */
#else
# if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
# define ZSTD_DEPRECATED(message) [[deprecated(message)]]
# elif (defined(GNUC) && (GNUC > 4 || (GNUC == 4 && GNUC_MINOR >= 5))) || defined(__clang__)
# define ZSTD_DEPRECATED(message) __attribute__((deprecated(message)))
# elif defined(__GNUC__) && (__GNUC__ >= 3)
# define ZSTD_DEPRECATED(message) __attribute__((deprecated))
# elif defined(_MSC_VER)
# define ZSTD_DEPRECATED(message) __declspec(deprecated(message))
# else
# pragma message("WARNING: You need to implement ZSTD_DEPRECATED for this compiler")
# define ZSTD_DEPRECATED(message)
# endif
#endif /* ZSTD_DISABLE_DEPRECATE_WARNINGS */
/*******************************************************************************
Introduction
zstd, short for Zstandard, is a fast lossless compression algorithm, targeting
real-time compression scenarios at zlib-level and better compression ratios.
The zstd compression library provides in-memory compression and decompression
functions.
The library supports regular compression levels from 1 up to ZSTD_maxCLevel(),
which is currently 22. Levels >= 20, labeled `--ultra`, should be used with
caution, as they require more memory. The library also offers negative
compression levels, which extend the range of speed vs. ratio preferences.
The lower the level, the faster the speed (at the cost of compression).
Compression can be done in:
- a single step (described as Simple API)
- a single step, reusing a context (described as Explicit context)
- unbounded multiple steps (described as Streaming compression)
The compression ratio achievable on small data can be highly improved using
a dictionary. Dictionary compression can be performed in:
- a single step (described as Simple dictionary API)
- a single step, reusing a dictionary (described as Bulk-processing
dictionary API)
Advanced experimental functions can be accessed using
`#define ZSTD_STATIC_LINKING_ONLY` before including zstd.h.
Advanced experimental APIs should never be used with a dynamically-linked
library. They are not "stable"; their definitions or signatures may change in
the future. Only static linking is allowed.
*******************************************************************************/
/*------ Version ------*/
#define ZSTD_VERSION_MAJOR 1
#define ZSTD_VERSION_MINOR 5
#define ZSTD_VERSION_RELEASE 5
#define ZSTD_VERSION_NUMBER (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
/*! ZSTD_versionNumber() :
* Return runtime library version, the value is (MAJOR*100*100 + MINOR*100 + RELEASE). */
ZSTDLIB_API unsigned ZSTD_versionNumber(void);
#define ZSTD_LIB_VERSION ZSTD_VERSION_MAJOR.ZSTD_VERSION_MINOR.ZSTD_VERSION_RELEASE
#define ZSTD_QUOTE(str) #str
#define ZSTD_EXPAND_AND_QUOTE(str) ZSTD_QUOTE(str)
#define ZSTD_VERSION_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_LIB_VERSION)
/*! ZSTD_versionString() :
* Return runtime library version, like "1.4.5". Requires v1.3.0+. */
ZSTDLIB_API const char* ZSTD_versionString(void);
/* *************************************
* Default constant
***************************************/
#ifndef ZSTD_CLEVEL_DEFAULT
# define ZSTD_CLEVEL_DEFAULT 3
#endif
/* *************************************
* Constants
***************************************/
/* All magic numbers are supposed read/written to/from files/memory using little-endian convention */
#define ZSTD_MAGICNUMBER 0xFD2FB528 /* valid since v0.8.0 */
#define ZSTD_MAGIC_DICTIONARY 0xEC30A437 /* valid since v0.7.0 */
#define ZSTD_MAGIC_SKIPPABLE_START 0x184D2A50 /* all 16 values, from 0x184D2A50 to 0x184D2A5F, signal the beginning of a skippable frame */
#define ZSTD_MAGIC_SKIPPABLE_MASK 0xFFFFFFF0
#define ZSTD_BLOCKSIZELOG_MAX 17
#define ZSTD_BLOCKSIZE_MAX (1<<ZSTD_BLOCKSIZELOG_MAX)
/***************************************
* Simple API
***************************************/
/*! ZSTD_compress() :
* Compresses `src` content as a single zstd compressed frame into already allocated `dst`.
* NOTE: Providing `dstCapacity >= ZSTD_compressBound(srcSize)` guarantees that zstd will have
* enough space to successfully compress the data.
* @return : compressed size written into `dst` (<= `dstCapacity),
* or an error code if it fails (which can be tested using ZSTD_isError()). */
ZSTDLIB_API size_t ZSTD_compress( void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
int compressionLevel);
/*! ZSTD_decompress() :
* `compressedSize` : must be the _exact_ size of some number of compressed and/or skippable frames.
* `dstCapacity` is an upper bound of originalSize to regenerate.
* If user cannot imply a maximum upper bound, it's better to use streaming mode to decompress data.
* @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
* or an errorCode if it fails (which can be tested using ZSTD_isError()). */
ZSTDLIB_API size_t ZSTD_decompress( void* dst, size_t dstCapacity,
const void* src, size_t compressedSize);
/*! ZSTD_getFrameContentSize() : requires v1.3.0+
* `src` should point to the start of a ZSTD encoded frame.
* `srcSize` must be at least as large as the frame header.
* hint : any size >= `ZSTD_frameHeaderSize_max` is large enough.
* @return : - decompressed size of `src` frame content, if known
* - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
* - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small)
* note 1 : a 0 return value means the frame is valid but "empty".
* note 2 : decompressed size is an optional field, it may not be present, typically in streaming mode.
* When `return==ZSTD_CONTENTSIZE_UNKNOWN`, data to decompress could be any size.
* In which case, it's necessary to use streaming mode to decompress data.
* Optionally, application can rely on some implicit limit,
* as ZSTD_decompress() only needs an upper bound of decompressed size.
* (For example, data could be necessarily cut into blocks <= 16 KB).
* note 3 : decompressed size is always present when compression is completed using single-pass functions,
* such as ZSTD_compress(), ZSTD_compressCCtx() ZSTD_compress_usingDict() or ZSTD_compress_usingCDict().
* note 4 : decompressed size can be very large (64-bits value),
* potentially larger than what local system can handle as a single memory segment.
* In which case, it's necessary to use streaming mode to decompress data.
* note 5 : If source is untrusted, decompressed size could be wrong or intentionally modified.
* Always ensure return value fits within application's authorized limits.
* Each application can set its own limits.
* note 6 : This function replaces ZSTD_getDecompressedSize() */
#define ZSTD_CONTENTSIZE_UNKNOWN (0ULL - 1)
#define ZSTD_CONTENTSIZE_ERROR (0ULL - 2)
ZSTDLIB_API unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize);
/*! ZSTD_getDecompressedSize() :
* NOTE: This function is now obsolete, in favor of ZSTD_getFrameContentSize().
* Both functions work the same way, but ZSTD_getDecompressedSize() blends
* "empty", "unknown" and "error" results to the same return value (0),
* while ZSTD_getFrameContentSize() gives them separate return values.
* @return : decompressed size of `src` frame content _if known and not empty_, 0 otherwise. */
ZSTD_DEPRECATED("Replaced by ZSTD_getFrameContentSize")
ZSTDLIB_API
unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
/*! ZSTD_findFrameCompressedSize() : Requires v1.4.0+
* `src` should point to the start of a ZSTD frame or skippable frame.
* `srcSize` must be >= first frame size
* @return : the compressed size of the first frame starting at `src`,
* suitable to pass as `srcSize` to `ZSTD_decompress` or similar,
* or an error code if input is invalid */
ZSTDLIB_API size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize);
/*====== Helper functions ======*/
/* ZSTD_compressBound() :
* maximum compressed size in worst case single-pass scenario.
* When invoking `ZSTD_compress()` or any other one-pass compression function,
* it's recommended to provide @dstCapacity >= ZSTD_compressBound(srcSize)
* as it eliminates one potential failure scenario,
* aka not enough room in dst buffer to write the compressed frame.
* Note : ZSTD_compressBound() itself can fail, if @srcSize > ZSTD_MAX_INPUT_SIZE .
* In which case, ZSTD_compressBound() will return an error code
* which can be tested using ZSTD_isError().
*
* ZSTD_COMPRESSBOUND() :
* same as ZSTD_compressBound(), but as a macro.
* It can be used to produce constants, which can be useful for static allocation,
* for example to size a static array on stack.
* Will produce constant value 0 if srcSize too large.
*/
#define ZSTD_MAX_INPUT_SIZE ((sizeof(size_t)==8) ? 0xFF00FF00FF00FF00ULL : 0xFF00FF00U)
#define ZSTD_COMPRESSBOUND(srcSize) (((size_t)(srcSize) >= ZSTD_MAX_INPUT_SIZE) ? 0 : (srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0)) /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
ZSTDLIB_API size_t ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
/* ZSTD_isError() :
* Most ZSTD_* functions returning a size_t value can be tested for error,
* using ZSTD_isError().
* @return 1 if error, 0 otherwise
*/
ZSTDLIB_API unsigned ZSTD_isError(size_t code); /*!< tells if a `size_t` function result is an error code */
ZSTDLIB_API const char* ZSTD_getErrorName(size_t code); /*!< provides readable string from an error code */
ZSTDLIB_API int ZSTD_minCLevel(void); /*!< minimum negative compression level allowed, requires v1.4.0+ */
ZSTDLIB_API int ZSTD_maxCLevel(void); /*!< maximum compression level available */
ZSTDLIB_API int ZSTD_defaultCLevel(void); /*!< default compression level, specified by ZSTD_CLEVEL_DEFAULT, requires v1.5.0+ */
/***************************************
* Explicit context
***************************************/
/*= Compression context
* When compressing many times,
* it is recommended to allocate a context just once,
* and re-use it for each successive compression operation.
* This will make workload friendlier for system's memory.
* Note : re-using context is just a speed / resource optimization.
* It doesn't change the compression ratio, which remains identical.
* Note 2 : In multi-threaded environments,
* use one different context per thread for parallel execution.
*/
typedef struct ZSTD_CCtx_s ZSTD_CCtx;
ZSTDLIB_API ZSTD_CCtx* ZSTD_createCCtx(void);
ZSTDLIB_API size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx); /* accept NULL pointer */
/*! ZSTD_compressCCtx() :
* Same as ZSTD_compress(), using an explicit ZSTD_CCtx.
* Important : in order to behave similarly to `ZSTD_compress()`,
* this function compresses at requested compression level,
* __ignoring any other parameter__ .
* If any advanced parameter was set using the advanced API,
* they will all be reset. Only `compressionLevel` remains.
*/
ZSTDLIB_API size_t ZSTD_compressCCtx(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
int compressionLevel);
/*= Decompression context
* When decompressing many times,
* it is recommended to allocate a context only once,
* and re-use it for each successive compression operation.
* This will make workload friendlier for system's memory.
* Use one context per thread for parallel execution. */
typedef struct ZSTD_DCtx_s ZSTD_DCtx;
ZSTDLIB_API ZSTD_DCtx* ZSTD_createDCtx(void);
ZSTDLIB_API size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx); /* accept NULL pointer */
/*! ZSTD_decompressDCtx() :
* Same as ZSTD_decompress(),
* requires an allocated ZSTD_DCtx.
* Compatible with sticky parameters.
*/
ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize);
/*********************************************
* Advanced compression API (Requires v1.4.0+)
**********************************************/
/* API design :
* Parameters are pushed one by one into an existing context,
* using ZSTD_CCtx_set*() functions.
* Pushed parameters are sticky : they are valid for next compressed frame, and any subsequent frame.
* "sticky" parameters are applicable to `ZSTD_compress2()` and `ZSTD_compressStream*()` !
* __They do not apply to "simple" one-shot variants such as ZSTD_compressCCtx()__ .
*
* It's possible to reset all parameters to "default" using ZSTD_CCtx_reset().
*
* This API supersedes all other "advanced" API entry points in the experimental section.
* In the future, we expect to remove from experimental API entry points which are redundant with this API.
*/
/* Compression strategies, listed from fastest to strongest */
typedef enum { ZSTD_fast=1,
ZSTD_dfast=2,
ZSTD_greedy=3,
ZSTD_lazy=4,
ZSTD_lazy2=5,
ZSTD_btlazy2=6,
ZSTD_btopt=7,
ZSTD_btultra=8,
ZSTD_btultra2=9
/* note : new strategies _might_ be added in the future.
Only the order (from fast to strong) is guaranteed */
} ZSTD_strategy;
typedef enum {
/* compression parameters
* Note: When compressing with a ZSTD_CDict these parameters are superseded
* by the parameters used to construct the ZSTD_CDict.
* See ZSTD_CCtx_refCDict() for more info (superseded-by-cdict). */
ZSTD_c_compressionLevel=100, /* Set compression parameters according to pre-defined cLevel table.
* Note that exact compression parameters are dynamically determined,
* depending on both compression level and srcSize (when known).
* Default level is ZSTD_CLEVEL_DEFAULT==3.
* Special: value 0 means default, which is controlled by ZSTD_CLEVEL_DEFAULT.
* Note 1 : it's possible to pass a negative compression level.
* Note 2 : setting a level does not automatically set all other compression parameters
* to default. Setting this will however eventually dynamically impact the compression
* parameters which have not been manually set. The manually set
* ones will 'stick'. */
/* Advanced compression parameters :
* It's possible to pin down compression parameters to some specific values.
* In which case, these values are no longer dynamically selected by the compressor */
ZSTD_c_windowLog=101, /* Maximum allowed back-reference distance, expressed as power of 2.
* This will set a memory budget for streaming decompression,
* with larger values requiring more memory
* and typically compressing more.
* Must be clamped between ZSTD_WINDOWLOG_MIN and ZSTD_WINDOWLOG_MAX.
* Special: value 0 means "use default windowLog".
* Note: Using a windowLog greater than ZSTD_WINDOWLOG_LIMIT_DEFAULT
* requires explicitly allowing such size at streaming decompression stage. */
ZSTD_c_hashLog=102, /* Size of the initial probe table, as a power of 2.
* Resulting memory usage is (1 << (hashLog+2)).
* Must be clamped between ZSTD_HASHLOG_MIN and ZSTD_HASHLOG_MAX.
* Larger tables improve compression ratio of strategies <= dFast,
* and improve speed of strategies > dFast.
* Special: value 0 means "use default hashLog". */
ZSTD_c_chainLog=103, /* Size of the multi-probe search table, as a power of 2.
* Resulting memory usage is (1 << (chainLog+2)).
* Must be clamped between ZSTD_CHAINLOG_MIN and ZSTD_CHAINLOG_MAX.
* Larger tables result in better and slower compression.
* This parameter is useless for "fast" strategy.
* It's still useful when using "dfast" strategy,
* in which case it defines a secondary probe table.
* Special: value 0 means "use default chainLog". */
ZSTD_c_searchLog=104, /* Number of search attempts, as a power of 2.
* More attempts result in better and slower compression.
* This parameter is useless for "fast" and "dFast" strategies.
* Special: value 0 means "use default searchLog". */
ZSTD_c_minMatch=105, /* Minimum size of searched matches.
* Note that Zstandard can still find matches of smaller size,
* it just tweaks its search algorithm to look for this size and larger.
* Larger values increase compression and decompression speed, but decrease ratio.
* Must be clamped between ZSTD_MINMATCH_MIN and ZSTD_MINMATCH_MAX.
* Note that currently, for all strategies < btopt, effective minimum is 4.
* , for all strategies > fast, effective maximum is 6.
* Special: value 0 means "use default minMatchLength". */
ZSTD_c_targetLength=106, /* Impact of this field depends on strategy.
* For strategies btopt, btultra & btultra2:
* Length of Match considered "good enough" to stop search.
* Larger values make compression stronger, and slower.
* For strategy fast:
* Distance between match sampling.
* Larger values make compression faster, and weaker.
* Special: value 0 means "use default targetLength". */
ZSTD_c_strategy=107, /* See ZSTD_strategy enum definition.
* The higher the value of selected strategy, the more complex it is,
* resulting in stronger and slower compression.
* Special: value 0 means "use default strategy". */
/* LDM mode parameters */
ZSTD_c_enableLongDistanceMatching=160, /* Enable long distance matching.
* This parameter is designed to improve compression ratio
* for large inputs, by finding large matches at long distance.
* It increases memory usage and window size.
* Note: enabling this parameter increases default ZSTD_c_windowLog to 128 MB
* except when expressly set to a different value.
* Note: will be enabled by default if ZSTD_c_windowLog >= 128 MB and
* compression strategy >= ZSTD_btopt (== compression level 16+) */
ZSTD_c_ldmHashLog=161, /* Size of the table for long distance matching, as a power of 2.
* Larger values increase memory usage and compression ratio,
* but decrease compression speed.
* Must be clamped between ZSTD_HASHLOG_MIN and ZSTD_HASHLOG_MAX
* default: windowlog - 7.
* Special: value 0 means "automatically determine hashlog". */
ZSTD_c_ldmMinMatch=162, /* Minimum match size for long distance matcher.
* Larger/too small values usually decrease compression ratio.
* Must be clamped between ZSTD_LDM_MINMATCH_MIN and ZSTD_LDM_MINMATCH_MAX.
* Special: value 0 means "use default value" (default: 64). */
ZSTD_c_ldmBucketSizeLog=163, /* Log size of each bucket in the LDM hash table for collision resolution.
* Larger values improve collision resolution but decrease compression speed.
* The maximum value is ZSTD_LDM_BUCKETSIZELOG_MAX.
* Special: value 0 means "use default value" (default: 3). */
ZSTD_c_ldmHashRateLog=164, /* Frequency of inserting/looking up entries into the LDM hash table.
* Must be clamped between 0 and (ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN).
* Default is MAX(0, (windowLog - ldmHashLog)), optimizing hash table usage.
* Larger values improve compression speed.
* Deviating far from default value will likely result in a compression ratio decrease.
* Special: value 0 means "automatically determine hashRateLog". */
/* frame parameters */
ZSTD_c_contentSizeFlag=200, /* Content size will be written into frame header _whenever known_ (default:1)
* Content size must be known at the beginning of compression.
* This is automatically the case when using ZSTD_compress2(),
* For streaming scenarios, content size must be provided with ZSTD_CCtx_setPledgedSrcSize() */
ZSTD_c_checksumFlag=201, /* A 32-bits checksum of content is written at end of frame (default:0) */
ZSTD_c_dictIDFlag=202, /* When applicable, dictionary's ID is written into frame header (default:1) */
/* multi-threading parameters */
/* These parameters are only active if multi-threading is enabled (compiled with build macro ZSTD_MULTITHREAD).
* Otherwise, trying to set any other value than default (0) will be a no-op and return an error.
* In a situation where it's unknown if the linked library supports multi-threading or not,
* setting ZSTD_c_nbWorkers to any value >= 1 and consulting the return value provides a quick way to check this property.
*/
ZSTD_c_nbWorkers=400, /* Select how many threads will be spawned to compress in parallel.
* When nbWorkers >= 1, triggers asynchronous mode when invoking ZSTD_compressStream*() :
* ZSTD_compressStream*() consumes input and flush output if possible, but immediately gives back control to caller,
* while compression is performed in parallel, within worker thread(s).
* (note : a strong exception to this rule is when first invocation of ZSTD_compressStream2() sets ZSTD_e_end :
* in which case, ZSTD_compressStream2() delegates to ZSTD_compress2(), which is always a blocking call).
* More workers improve speed, but also increase memory usage.
* Default value is `0`, aka "single-threaded mode" : no worker is spawned,
* compression is performed inside Caller's thread, and all invocations are blocking */
ZSTD_c_jobSize=401, /* Size of a compression job. This value is enforced only when nbWorkers >= 1.
* Each compression job is completed in parallel, so this value can indirectly impact the nb of active threads.
* 0 means default, which is dynamically determined based on compression parameters.
* Job size must be a minimum of overlap size, or ZSTDMT_JOBSIZE_MIN (= 512 KB), whichever is largest.
* The minimum size is automatically and transparently enforced. */
ZSTD_c_overlapLog=402, /* Control the overlap size, as a fraction of window size.
* The overlap size is an amount of data reloaded from previous job at the beginning of a new job.
* It helps preserve compression ratio, while each job is compressed in parallel.
* This value is enforced only when nbWorkers >= 1.
* Larger values increase compression ratio, but decrease speed.
* Possible values range from 0 to 9 :
* - 0 means "default" : value will be determined by the library, depending on strategy
* - 1 means "no overlap"
* - 9 means "full overlap", using a full window size.
* Each intermediate rank increases/decreases load size by a factor 2 :
* 9: full window; 8: w/2; 7: w/4; 6: w/8; 5:w/16; 4: w/32; 3:w/64; 2:w/128; 1:no overlap; 0:default
* default value varies between 6 and 9, depending on strategy */
/* note : additional experimental parameters are also available
* within the experimental section of the API.
* At the time of this writing, they include :
* ZSTD_c_rsyncable
* ZSTD_c_format
* ZSTD_c_forceMaxWindow
* ZSTD_c_forceAttachDict
* ZSTD_c_literalCompressionMode
* ZSTD_c_targetCBlockSize
* ZSTD_c_srcSizeHint
* ZSTD_c_enableDedicatedDictSearch
* ZSTD_c_stableInBuffer
* ZSTD_c_stableOutBuffer
* ZSTD_c_blockDelimiters
* ZSTD_c_validateSequences
* ZSTD_c_useBlockSplitter
* ZSTD_c_useRowMatchFinder
* ZSTD_c_prefetchCDictTables
* ZSTD_c_enableSeqProducerFallback
* ZSTD_c_maxBlockSize
* Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
* note : never ever use experimentalParam? names directly;
* also, the enums values themselves are unstable and can still change.
*/
ZSTD_c_experimentalParam1=500,
ZSTD_c_experimentalParam2=10,
ZSTD_c_experimentalParam3=1000,
ZSTD_c_experimentalParam4=1001,
ZSTD_c_experimentalParam5=1002,
ZSTD_c_experimentalParam6=1003,
ZSTD_c_experimentalParam7=1004,
ZSTD_c_experimentalParam8=1005,
ZSTD_c_experimentalParam9=1006,
ZSTD_c_experimentalParam10=1007,
ZSTD_c_experimentalParam11=1008,
ZSTD_c_experimentalParam12=1009,
ZSTD_c_experimentalParam13=1010,
ZSTD_c_experimentalParam14=1011,
ZSTD_c_experimentalParam15=1012,
ZSTD_c_experimentalParam16=1013,
ZSTD_c_experimentalParam17=1014,
ZSTD_c_experimentalParam18=1015,
ZSTD_c_experimentalParam19=1016
} ZSTD_cParameter;
typedef struct {
size_t error;
int lowerBound;
int upperBound;
} ZSTD_bounds;
/*! ZSTD_cParam_getBounds() :
* All parameters must belong to an interval with lower and upper bounds,
* otherwise they will either trigger an error or be automatically clamped.
* @return : a structure, ZSTD_bounds, which contains
* - an error status field, which must be tested using ZSTD_isError()
* - lower and upper bounds, both inclusive
*/
ZSTDLIB_API ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter cParam);
/*! ZSTD_CCtx_setParameter() :
* Set one compression parameter, selected by enum ZSTD_cParameter.
* All parameters have valid bounds. Bounds can be queried using ZSTD_cParam_getBounds().
* Providing a value beyond bound will either clamp it, or trigger an error (depending on parameter).
* Setting a parameter is generally only possible during frame initialization (before starting compression).
* Exception : when using multi-threading mode (nbWorkers >= 1),
* the following parameters can be updated _during_ compression (within same frame):
* => compressionLevel, hashLog, chainLog, searchLog, minMatch, targetLength and strategy.
* new parameters will be active for next job only (after a flush()).
* @return : an error code (which can be tested using ZSTD_isError()).
*/
ZSTDLIB_API size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value);
/*! ZSTD_CCtx_setPledgedSrcSize() :
* Total input data size to be compressed as a single frame.
* Value will be written in frame header, unless if explicitly forbidden using ZSTD_c_contentSizeFlag.
* This value will also be controlled at end of frame, and trigger an error if not respected.
* @result : 0, or an error code (which can be tested with ZSTD_isError()).
* Note 1 : pledgedSrcSize==0 actually means zero, aka an empty frame.
* In order to mean "unknown content size", pass constant ZSTD_CONTENTSIZE_UNKNOWN.
* ZSTD_CONTENTSIZE_UNKNOWN is default value for any new frame.
* Note 2 : pledgedSrcSize is only valid once, for the next frame.
* It's discarded at the end of the frame, and replaced by ZSTD_CONTENTSIZE_UNKNOWN.
* Note 3 : Whenever all input data is provided and consumed in a single round,
* for example with ZSTD_compress2(),
* or invoking immediately ZSTD_compressStream2(,,,ZSTD_e_end),
* this value is automatically overridden by srcSize instead.
*/
ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize);
typedef enum {
ZSTD_reset_session_only = 1,
ZSTD_reset_parameters = 2,
ZSTD_reset_session_and_parameters = 3
} ZSTD_ResetDirective;
/*! ZSTD_CCtx_reset() :
* There are 2 different things that can be reset, independently or jointly :
* - The session : will stop compressing current frame, and make CCtx ready to start a new one.
* Useful after an error, or to interrupt any ongoing compression.
* Any internal data not yet flushed is cancelled.
* Compression parameters and dictionary remain unchanged.
* They will be used to compress next frame.
* Resetting session never fails.
* - The parameters : changes all parameters back to "default".
* This also removes any reference to any dictionary or external sequence producer.
* Parameters can only be changed between 2 sessions (i.e. no compression is currently ongoing)
* otherwise the reset fails, and function returns an error value (which can be tested using ZSTD_isError())
* - Both : similar to resetting the session, followed by resetting parameters.
*/
ZSTDLIB_API size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset);
/*! ZSTD_compress2() :
* Behave the same as ZSTD_compressCCtx(), but compression parameters are set using the advanced API.
* ZSTD_compress2() always starts a new frame.
* Should cctx hold data from a previously unfinished frame, everything about it is forgotten.
* - Compression parameters are pushed into CCtx before starting compression, using ZSTD_CCtx_set*()
* - The function is always blocking, returns when compression is completed.
* NOTE: Providing `dstCapacity >= ZSTD_compressBound(srcSize)` guarantees that zstd will have
* enough space to successfully compress the data, though it is possible it fails for other reasons.
* @return : compressed size written into `dst` (<= `dstCapacity),
* or an error code if it fails (which can be tested using ZSTD_isError()).
*/
ZSTDLIB_API size_t ZSTD_compress2( ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize);
/***********************************************
* Advanced decompression API (Requires v1.4.0+)
************************************************/
/* The advanced API pushes parameters one by one into an existing DCtx context.
* Parameters are sticky, and remain valid for all following frames
* using the same DCtx context.
* It's possible to reset parameters to default values using ZSTD_DCtx_reset().
* Note : This API is compatible with existing ZSTD_decompressDCtx() and ZSTD_decompressStream().
* Therefore, no new decompression function is necessary.
*/
typedef enum {
ZSTD_d_windowLogMax=100, /* Select a size limit (in power of 2) beyond which
* the streaming API will refuse to allocate memory buffer
* in order to protect the host from unreasonable memory requirements.
* This parameter is only useful in streaming mode, since no internal buffer is allocated in single-pass mode.
* By default, a decompression context accepts window sizes <= (1 << ZSTD_WINDOWLOG_LIMIT_DEFAULT).
* Special: value 0 means "use default maximum windowLog". */
/* note : additional experimental parameters are also available
* within the experimental section of the API.
* At the time of this writing, they include :
* ZSTD_d_format
* ZSTD_d_stableOutBuffer
* ZSTD_d_forceIgnoreChecksum
* ZSTD_d_refMultipleDDicts
* ZSTD_d_disableHuffmanAssembly
* ZSTD_d_maxBlockSize
* Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
* note : never ever use experimentalParam? names directly
*/
ZSTD_d_experimentalParam1=1000,
ZSTD_d_experimentalParam2=1001,
ZSTD_d_experimentalParam3=1002,
ZSTD_d_experimentalParam4=1003,
ZSTD_d_experimentalParam5=1004,
ZSTD_d_experimentalParam6=1005
} ZSTD_dParameter;
/*! ZSTD_dParam_getBounds() :
* All parameters must belong to an interval with lower and upper bounds,
* otherwise they will either trigger an error or be automatically clamped.
* @return : a structure, ZSTD_bounds, which contains
* - an error status field, which must be tested using ZSTD_isError()
* - both lower and upper bounds, inclusive
*/
ZSTDLIB_API ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam);
/*! ZSTD_DCtx_setParameter() :
* Set one compression parameter, selected by enum ZSTD_dParameter.
* All parameters have valid bounds. Bounds can be queried using ZSTD_dParam_getBounds().
* Providing a value beyond bound will either clamp it, or trigger an error (depending on parameter).
* Setting a parameter is only possible during frame initialization (before starting decompression).
* @return : 0, or an error code (which can be tested using ZSTD_isError()).
*/
ZSTDLIB_API size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int value);
/*! ZSTD_DCtx_reset() :
* Return a DCtx to clean state.
* Session and parameters can be reset jointly or separately.
* Parameters can only be reset when no active frame is being decompressed.
* @return : 0, or an error code, which can be tested with ZSTD_isError()
*/
ZSTDLIB_API size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset);
/****************************
* Streaming
****************************/
typedef struct ZSTD_inBuffer_s {
const void* src; /**< start of input buffer */
size_t size; /**< size of input buffer */
size_t pos; /**< position where reading stopped. Will be updated. Necessarily 0 <= pos <= size */
} ZSTD_inBuffer;
typedef struct ZSTD_outBuffer_s {
void* dst; /**< start of output buffer */
size_t size; /**< size of output buffer */
size_t pos; /**< position where writing stopped. Will be updated. Necessarily 0 <= pos <= size */
} ZSTD_outBuffer;
/*-***********************************************************************
* Streaming compression - HowTo
*
* A ZSTD_CStream object is required to track streaming operation.
* Use ZSTD_createCStream() and ZSTD_freeCStream() to create/release resources.
* ZSTD_CStream objects can be reused multiple times on consecutive compression operations.
* It is recommended to re-use ZSTD_CStream since it will play nicer with system's memory, by re-using already allocated memory.
*
* For parallel execution, use one separate ZSTD_CStream per thread.
*
* note : since v1.3.0, ZSTD_CStream and ZSTD_CCtx are the same thing.
*
* Parameters are sticky : when starting a new compression on the same context,
* it will re-use the same sticky parameters as previous compression session.
* When in doubt, it's recommended to fully initialize the context before usage.
* Use ZSTD_CCtx_reset() to reset the context and ZSTD_CCtx_setParameter(),
* ZSTD_CCtx_setPledgedSrcSize(), or ZSTD_CCtx_loadDictionary() and friends to
* set more specific parameters, the pledged source size, or load a dictionary.
*
* Use ZSTD_compressStream2() with ZSTD_e_continue as many times as necessary to
* consume input stream. The function will automatically update both `pos`
* fields within `input` and `output`.
* Note that the function may not consume the entire input, for example, because
* the output buffer is already full, in which case `input.pos < input.size`.
* The caller must check if input has been entirely consumed.
* If not, the caller must make some room to receive more compressed data,
* and then present again remaining input data.
* note: ZSTD_e_continue is guaranteed to make some forward progress when called,
* but doesn't guarantee maximal forward progress. This is especially relevant
* when compressing with multiple threads. The call won't block if it can
* consume some input, but if it can't it will wait for some, but not all,
* output to be flushed.
* @return : provides a minimum amount of data remaining to be flushed from internal buffers
* or an error code, which can be tested using ZSTD_isError().
*
* At any moment, it's possible to flush whatever data might remain stuck within internal buffer,
* using ZSTD_compressStream2() with ZSTD_e_flush. `output->pos` will be updated.
* Note that, if `output->size` is too small, a single invocation with ZSTD_e_flush might not be enough (return code > 0).
* In which case, make some room to receive more compressed data, and call again ZSTD_compressStream2() with ZSTD_e_flush.
* You must continue calling ZSTD_compressStream2() with ZSTD_e_flush until it returns 0, at which point you can change the
* operation.
* note: ZSTD_e_flush will flush as much output as possible, meaning when compressing with multiple threads, it will
* block until the flush is complete or the output buffer is full.
* @return : 0 if internal buffers are entirely flushed,
* >0 if some data still present within internal buffer (the value is minimal estimation of remaining size),
* or an error code, which can be tested using ZSTD_isError().
*
* Calling ZSTD_compressStream2() with ZSTD_e_end instructs to finish a frame.
* It will perform a flush and write frame epilogue.
* The epilogue is required for decoders to consider a frame completed.
* flush operation is the same, and follows same rules as calling ZSTD_compressStream2() with ZSTD_e_flush.
* You must continue calling ZSTD_compressStream2() with ZSTD_e_end until it returns 0, at which point you are free to
* start a new frame.
* note: ZSTD_e_end will flush as much output as possible, meaning when compressing with multiple threads, it will
* block until the flush is complete or the output buffer is full.
* @return : 0 if frame fully completed and fully flushed,
* >0 if some data still present within internal buffer (the value is minimal estimation of remaining size),
* or an error code, which can be tested using ZSTD_isError().
*
* *******************************************************************/
typedef ZSTD_CCtx ZSTD_CStream; /**< CCtx and CStream are now effectively same object (>= v1.3.0) */
/* Continue to distinguish them for compatibility with older versions <= v1.2.0 */
/*===== ZSTD_CStream management functions =====*/
ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream(void);
ZSTDLIB_API size_t ZSTD_freeCStream(ZSTD_CStream* zcs); /* accept NULL pointer */
/*===== Streaming compression functions =====*/
typedef enum {
ZSTD_e_continue=0, /* collect more data, encoder decides when to output compressed result, for optimal compression ratio */
ZSTD_e_flush=1, /* flush any data provided so far,
* it creates (at least) one new block, that can be decoded immediately on reception;
* frame will continue: any future data can still reference previously compressed data, improving compression.
* note : multithreaded compression will block to flush as much output as possible. */
ZSTD_e_end=2 /* flush any remaining data _and_ close current frame.
* note that frame is only closed after compressed data is fully flushed (return value == 0).
* After that point, any additional data starts a new frame.
* note : each frame is independent (does not reference any content from previous frame).
: note : multithreaded compression will block to flush as much output as possible. */
} ZSTD_EndDirective;
/*! ZSTD_compressStream2() : Requires v1.4.0+
* Behaves about the same as ZSTD_compressStream, with additional control on end directive.
* - Compression parameters are pushed into CCtx before starting compression, using ZSTD_CCtx_set*()
* - Compression parameters cannot be changed once compression is started (save a list of exceptions in multi-threading mode)
* - output->pos must be <= dstCapacity, input->pos must be <= srcSize
* - output->pos and input->pos will be updated. They are guaranteed to remain below their respective limit.
* - endOp must be a valid directive
* - When nbWorkers==0 (default), function is blocking : it completes its job before returning to caller.
* - When nbWorkers>=1, function is non-blocking : it copies a portion of input, distributes jobs to internal worker threads, flush to output whatever is available,
* and then immediately returns, just indicating that there is some data remaining to be flushed.
* The function nonetheless guarantees forward progress : it will return only after it reads or write at least 1+ byte.
* - Exception : if the first call requests a ZSTD_e_end directive and provides enough dstCapacity, the function delegates to ZSTD_compress2() which is always blocking.
* - @return provides a minimum amount of data remaining to be flushed from internal buffers
* or an error code, which can be tested using ZSTD_isError().
* if @return != 0, flush is not fully completed, there is still some data left within internal buffers.
* This is useful for ZSTD_e_flush, since in this case more flushes are necessary to empty all buffers.
* For ZSTD_e_end, @return == 0 when internal buffers are fully flushed and frame is completed.
* - after a ZSTD_e_end directive, if internal buffer is not fully flushed (@return != 0),
* only ZSTD_e_end or ZSTD_e_flush operations are allowed.
* Before starting a new compression job, or changing compression parameters,
* it is required to fully flush internal buffers.
*/
ZSTDLIB_API size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
ZSTD_outBuffer* output,
ZSTD_inBuffer* input,
ZSTD_EndDirective endOp);
/* These buffer sizes are softly recommended.
* They are not required : ZSTD_compressStream*() happily accepts any buffer size, for both input and output.
* Respecting the recommended size just makes it a bit easier for ZSTD_compressStream*(),
* reducing the amount of memory shuffling and buffering, resulting in minor performance savings.
*
* However, note that these recommendations are from the perspective of a C caller program.
* If the streaming interface is invoked from some other language,
* especially managed ones such as Java or Go, through a foreign function interface such as jni or cgo,
* a major performance rule is to reduce crossing such interface to an absolute minimum.
* It's not rare that performance ends being spent more into the interface, rather than compression itself.
* In which cases, prefer using large buffers, as large as practical,
* for both input and output, to reduce the nb of roundtrips.
*/
ZSTDLIB_API size_t ZSTD_CStreamInSize(void); /**< recommended size for input buffer */
ZSTDLIB_API size_t ZSTD_CStreamOutSize(void); /**< recommended size for output buffer. Guarantee to successfully flush at least one complete compressed block. */
/* *****************************************************************************
* This following is a legacy streaming API, available since v1.0+ .
* It can be replaced by ZSTD_CCtx_reset() and ZSTD_compressStream2().
* It is redundant, but remains fully supported.
******************************************************************************/
/*!
* Equivalent to:
*
* ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
* ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
* ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
*
* Note that ZSTD_initCStream() clears any previously set dictionary. Use the new API
* to compress with a dictionary.
*/
ZSTDLIB_API size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel);
/*!
* Alternative for ZSTD_compressStream2(zcs, output, input, ZSTD_e_continue).
* NOTE: The return value is different. ZSTD_compressStream() returns a hint for
* the next read size (if non-zero and not an error). ZSTD_compressStream2()
* returns the minimum nb of bytes left to flush (if non-zero and not an error).
*/
ZSTDLIB_API size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
/*! Equivalent to ZSTD_compressStream2(zcs, output, &emptyInput, ZSTD_e_flush). */
ZSTDLIB_API size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
/*! Equivalent to ZSTD_compressStream2(zcs, output, &emptyInput, ZSTD_e_end). */
ZSTDLIB_API size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
/*-***************************************************************************
* Streaming decompression - HowTo
*
* A ZSTD_DStream object is required to track streaming operations.
* Use ZSTD_createDStream() and ZSTD_freeDStream() to create/release resources.
* ZSTD_DStream objects can be re-used multiple times.
*
* Use ZSTD_initDStream() to start a new decompression operation.
* @return : recommended first input size
* Alternatively, use advanced API to set specific properties.
*
* Use ZSTD_decompressStream() repetitively to consume your input.
* The function will update both `pos` fields.
* If `input.pos < input.size`, some input has not been consumed.
* It's up to the caller to present again remaining data.
* The function tries to flush all data decoded immediately, respecting output buffer size.
* If `output.pos < output.size`, decoder has flushed everything it could.
* But if `output.pos == output.size`, there might be some data left within internal buffers.,
* In which case, call ZSTD_decompressStream() again to flush whatever remains in the buffer.
* Note : with no additional input provided, amount of data flushed is necessarily <= ZSTD_BLOCKSIZE_MAX.
* @return : 0 when a frame is completely decoded and fully flushed,
* or an error code, which can be tested using ZSTD_isError(),
* or any other value > 0, which means there is still some decoding or flushing to do to complete current frame :
* the return value is a suggested next input size (just a hint for better latency)
* that will never request more than the remaining frame size.
* *******************************************************************************/
typedef ZSTD_DCtx ZSTD_DStream; /**< DCtx and DStream are now effectively same object (>= v1.3.0) */
/* For compatibility with versions <= v1.2.0, prefer differentiating them. */
/*===== ZSTD_DStream management functions =====*/
ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream(void);
ZSTDLIB_API size_t ZSTD_freeDStream(ZSTD_DStream* zds); /* accept NULL pointer */
/*===== Streaming decompression functions =====*/
/*! ZSTD_initDStream() :
* Initialize/reset DStream state for new decompression operation.
* Call before new decompression operation using same DStream.
*
* Note : This function is redundant with the advanced API and equivalent to:
* ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
* ZSTD_DCtx_refDDict(zds, NULL);
*/
ZSTDLIB_API size_t ZSTD_initDStream(ZSTD_DStream* zds);
/*! ZSTD_decompressStream() :
* Streaming decompression function.
* Call repetitively to consume full input updating it as necessary.
* Function will update both input and output `pos` fields exposing current state via these fields:
* - `input.pos < input.size`, some input remaining and caller should provide remaining input
* on the next call.
* - `output.pos < output.size`, decoder finished and flushed all remaining buffers.
* - `output.pos == output.size`, potentially uncflushed data present in the internal buffers,
* call ZSTD_decompressStream() again to flush remaining data to output.
* Note : with no additional input, amount of data flushed <= ZSTD_BLOCKSIZE_MAX.
*
* @return : 0 when a frame is completely decoded and fully flushed,
* or an error code, which can be tested using ZSTD_isError(),
* or any other value > 0, which means there is some decoding or flushing to do to complete current frame.
*/
ZSTDLIB_API size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
ZSTDLIB_API size_t ZSTD_DStreamInSize(void); /*!< recommended size for input buffer */
ZSTDLIB_API size_t ZSTD_DStreamOutSize(void); /*!< recommended size for output buffer. Guarantee to successfully flush at least one complete block in all circumstances. */
/**************************
* Simple dictionary API
***************************/
/*! ZSTD_compress_usingDict() :
* Compression at an explicit compression level using a Dictionary.
* A dictionary can be any arbitrary data segment (also called a prefix),
* or a buffer with specified information (see zdict.h).
* Note : This function loads the dictionary, resulting in significant startup delay.
* It's intended for a dictionary used only once.
* Note 2 : When `dict == NULL || dictSize < 8` no dictionary is used. */
ZSTDLIB_API size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const void* dict,size_t dictSize,
int compressionLevel);
/*! ZSTD_decompress_usingDict() :
* Decompression using a known Dictionary.
* Dictionary must be identical to the one used during compression.
* Note : This function loads the dictionary, resulting in significant startup delay.
* It's intended for a dictionary used only once.
* Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
ZSTDLIB_API size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const void* dict,size_t dictSize);
/***********************************
* Bulk processing dictionary API
**********************************/
typedef struct ZSTD_CDict_s ZSTD_CDict;
/*! ZSTD_createCDict() :
* When compressing multiple messages or blocks using the same dictionary,
* it's recommended to digest the dictionary only once, since it's a costly operation.
* ZSTD_createCDict() will create a state from digesting a dictionary.
* The resulting state can be used for future compression operations with very limited startup cost.
* ZSTD_CDict can be created once and shared by multiple threads concurrently, since its usage is read-only.
* @dictBuffer can be released after ZSTD_CDict creation, because its content is copied within CDict.
* Note 1 : Consider experimental function `ZSTD_createCDict_byReference()` if you prefer to not duplicate @dictBuffer content.
* Note 2 : A ZSTD_CDict can be created from an empty @dictBuffer,
* in which case the only thing that it transports is the @compressionLevel.
* This can be useful in a pipeline featuring ZSTD_compress_usingCDict() exclusively,
* expecting a ZSTD_CDict parameter with any data, including those without a known dictionary. */
ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict(const void* dictBuffer, size_t dictSize,
int compressionLevel);
/*! ZSTD_freeCDict() :
* Function frees memory allocated by ZSTD_createCDict().
* If a NULL pointer is passed, no operation is performed. */
ZSTDLIB_API size_t ZSTD_freeCDict(ZSTD_CDict* CDict);
/*! ZSTD_compress_usingCDict() :
* Compression using a digested Dictionary.
* Recommended when same dictionary is used multiple times.
* Note : compression level is _decided at dictionary creation time_,
* and frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no) */
ZSTDLIB_API size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const ZSTD_CDict* cdict);
typedef struct ZSTD_DDict_s ZSTD_DDict;
/*! ZSTD_createDDict() :
* Create a digested dictionary, ready to start decompression operation without startup delay.
* dictBuffer can be released after DDict creation, as its content is copied inside DDict. */
ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict(const void* dictBuffer, size_t dictSize);
/*! ZSTD_freeDDict() :
* Function frees memory allocated with ZSTD_createDDict()
* If a NULL pointer is passed, no operation is performed. */
ZSTDLIB_API size_t ZSTD_freeDDict(ZSTD_DDict* ddict);
/*! ZSTD_decompress_usingDDict() :
* Decompression using a digested Dictionary.
* Recommended when same dictionary is used multiple times. */
ZSTDLIB_API size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const ZSTD_DDict* ddict);
/********************************
* Dictionary helper functions
*******************************/
/*! ZSTD_getDictID_fromDict() : Requires v1.4.0+
* Provides the dictID stored within dictionary.
* if @return == 0, the dictionary is not conformant with Zstandard specification.
* It can still be loaded, but as a content-only dictionary. */
ZSTDLIB_API unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize);
/*! ZSTD_getDictID_fromCDict() : Requires v1.5.0+
* Provides the dictID of the dictionary loaded into `cdict`.
* If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
* Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
ZSTDLIB_API unsigned ZSTD_getDictID_fromCDict(const ZSTD_CDict* cdict);
/*! ZSTD_getDictID_fromDDict() : Requires v1.4.0+
* Provides the dictID of the dictionary loaded into `ddict`.
* If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
* Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
ZSTDLIB_API unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict);
/*! ZSTD_getDictID_fromFrame() : Requires v1.4.0+
* Provides the dictID required to decompressed the frame stored within `src`.
* If @return == 0, the dictID could not be decoded.
* This could for one of the following reasons :
* - The frame does not require a dictionary to be decoded (most common case).
* - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden piece of information.
* Note : this use case also happens when using a non-conformant dictionary.
* - `srcSize` is too small, and as a result, the frame header could not be decoded (only possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`).
* - This is not a Zstandard frame.
* When identifying the exact failure cause, it's possible to use ZSTD_getFrameHeader(), which will provide a more precise error code. */
ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
/*******************************************************************************
* Advanced dictionary and prefix API (Requires v1.4.0+)
*
* This API allows dictionaries to be used with ZSTD_compress2(),
* ZSTD_compressStream2(), and ZSTD_decompressDCtx().
* Dictionaries are sticky, they remain valid when same context is re-used,
* they only reset when the context is reset
* with ZSTD_reset_parameters or ZSTD_reset_session_and_parameters.
* In contrast, Prefixes are single-use.
******************************************************************************/
/*! ZSTD_CCtx_loadDictionary() : Requires v1.4.0+
* Create an internal CDict from `dict` buffer.
* Decompression will have to use same dictionary.
* @result : 0, or an error code (which can be tested with ZSTD_isError()).
* Special: Loading a NULL (or 0-size) dictionary invalidates previous dictionary,
* meaning "return to no-dictionary mode".
* Note 1 : Dictionary is sticky, it will be used for all future compressed frames,
* until parameters are reset, a new dictionary is loaded, or the dictionary
* is explicitly invalidated by loading a NULL dictionary.
* Note 2 : Loading a dictionary involves building tables.
* It's also a CPU consuming operation, with non-negligible impact on latency.
* Tables are dependent on compression parameters, and for this reason,
* compression parameters can no longer be changed after loading a dictionary.
* Note 3 :`dict` content will be copied internally.
* Use experimental ZSTD_CCtx_loadDictionary_byReference() to reference content instead.
* In such a case, dictionary buffer must outlive its users.
* Note 4 : Use ZSTD_CCtx_loadDictionary_advanced()
* to precisely select how dictionary content must be interpreted.
* Note 5 : This method does not benefit from LDM (long distance mode).
* If you want to employ LDM on some large dictionary content,
* prefer employing ZSTD_CCtx_refPrefix() described below.
*/
ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
/*! ZSTD_CCtx_refCDict() : Requires v1.4.0+
* Reference a prepared dictionary, to be used for all future compressed frames.
* Note that compression parameters are enforced from within CDict,
* and supersede any compression parameter previously set within CCtx.
* The parameters ignored are labelled as "superseded-by-cdict" in the ZSTD_cParameter enum docs.
* The ignored parameters will be used again if the CCtx is returned to no-dictionary mode.
* The dictionary will remain valid for future compressed frames using same CCtx.
* @result : 0, or an error code (which can be tested with ZSTD_isError()).
* Special : Referencing a NULL CDict means "return to no-dictionary mode".
* Note 1 : Currently, only one dictionary can be managed.
* Referencing a new dictionary effectively "discards" any previous one.
* Note 2 : CDict is just referenced, its lifetime must outlive its usage within CCtx. */
ZSTDLIB_API size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
/*! ZSTD_CCtx_refPrefix() : Requires v1.4.0+
* Reference a prefix (single-usage dictionary) for next compressed frame.
* A prefix is **only used once**. Tables are discarded at end of frame (ZSTD_e_end).
* Decompression will need same prefix to properly regenerate data.
* Compressing with a prefix is similar in outcome as performing a diff and compressing it,
* but performs much faster, especially during decompression (compression speed is tunable with compression level).
* This method is compatible with LDM (long distance mode).
* @result : 0, or an error code (which can be tested with ZSTD_isError()).
* Special: Adding any prefix (including NULL) invalidates any previous prefix or dictionary
* Note 1 : Prefix buffer is referenced. It **must** outlive compression.
* Its content must remain unmodified during compression.
* Note 2 : If the intention is to diff some large src data blob with some prior version of itself,
* ensure that the window size is large enough to contain the entire source.
* See ZSTD_c_windowLog.
* Note 3 : Referencing a prefix involves building tables, which are dependent on compression parameters.
* It's a CPU consuming operation, with non-negligible impact on latency.
* If there is a need to use the same prefix multiple times, consider loadDictionary instead.
* Note 4 : By default, the prefix is interpreted as raw content (ZSTD_dct_rawContent).
* Use experimental ZSTD_CCtx_refPrefix_advanced() to alter dictionary interpretation. */
ZSTDLIB_API size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx,
const void* prefix, size_t prefixSize);
/*! ZSTD_DCtx_loadDictionary() : Requires v1.4.0+
* Create an internal DDict from dict buffer, to be used to decompress all future frames.
* The dictionary remains valid for all future frames, until explicitly invalidated, or
* a new dictionary is loaded.
* @result : 0, or an error code (which can be tested with ZSTD_isError()).
* Special : Adding a NULL (or 0-size) dictionary invalidates any previous dictionary,
* meaning "return to no-dictionary mode".
* Note 1 : Loading a dictionary involves building tables,
* which has a non-negligible impact on CPU usage and latency.
* It's recommended to "load once, use many times", to amortize the cost
* Note 2 :`dict` content will be copied internally, so `dict` can be released after loading.
* Use ZSTD_DCtx_loadDictionary_byReference() to reference dictionary content instead.
* Note 3 : Use ZSTD_DCtx_loadDictionary_advanced() to take control of
* how dictionary content is loaded and interpreted.
*/
ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
/*! ZSTD_DCtx_refDDict() : Requires v1.4.0+
* Reference a prepared dictionary, to be used to decompress next frames.
* The dictionary remains active for decompression of future frames using same DCtx.
*
* If called with ZSTD_d_refMultipleDDicts enabled, repeated calls of this function
* will store the DDict references in a table, and the DDict used for decompression
* will be determined at decompression time, as per the dict ID in the frame.
* The memory for the table is allocated on the first call to refDDict, and can be
* freed with ZSTD_freeDCtx().
*
* If called with ZSTD_d_refMultipleDDicts disabled (the default), only one dictionary
* will be managed, and referencing a dictionary effectively "discards" any previous one.
*
* @result : 0, or an error code (which can be tested with ZSTD_isError()).
* Special: referencing a NULL DDict means "return to no-dictionary mode".
* Note 2 : DDict is just referenced, its lifetime must outlive its usage from DCtx.
*/
ZSTDLIB_API size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
/*! ZSTD_DCtx_refPrefix() : Requires v1.4.0+
* Reference a prefix (single-usage dictionary) to decompress next frame.
* This is the reverse operation of ZSTD_CCtx_refPrefix(),
* and must use the same prefix as the one used during compression.
* Prefix is **only used once**. Reference is discarded at end of frame.
* End of frame is reached when ZSTD_decompressStream() returns 0.
* @result : 0, or an error code (which can be tested with ZSTD_isError()).
* Note 1 : Adding any prefix (including NULL) invalidates any previously set prefix or dictionary
* Note 2 : Prefix buffer is referenced. It **must** outlive decompression.
* Prefix buffer must remain unmodified up to the end of frame,
* reached when ZSTD_decompressStream() returns 0.
* Note 3 : By default, the prefix is treated as raw content (ZSTD_dct_rawContent).
* Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode (Experimental section)
* Note 4 : Referencing a raw content prefix has almost no cpu nor memory cost.
* A full dictionary is more costly, as it requires building tables.
*/
ZSTDLIB_API size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx,
const void* prefix, size_t prefixSize);
/* === Memory management === */
/*! ZSTD_sizeof_*() : Requires v1.4.0+
* These functions give the _current_ memory usage of selected object.
* Note that object memory usage can evolve (increase or decrease) over time. */
ZSTDLIB_API size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx);
ZSTDLIB_API size_t ZSTD_sizeof_DCtx(const ZSTD_DCtx* dctx);
ZSTDLIB_API size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs);
ZSTDLIB_API size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds);
ZSTDLIB_API size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict);
ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
#endif /* ZSTD_H_235446 */
/* **************************************************************************************
* ADVANCED AND EXPERIMENTAL FUNCTIONS
****************************************************************************************
* The definitions in the following section are considered experimental.
* They are provided for advanced scenarios.
* They should never be used with a dynamic library, as prototypes may change in the future.
* Use them only in association with static linking.
* ***************************************************************************************/
#if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
#define ZSTD_H_ZSTD_STATIC_LINKING_ONLY
/* This can be overridden externally to hide static symbols. */
#ifndef ZSTDLIB_STATIC_API
# if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
# define ZSTDLIB_STATIC_API __declspec(dllexport) ZSTDLIB_VISIBLE
# elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
# define ZSTDLIB_STATIC_API __declspec(dllimport) ZSTDLIB_VISIBLE
# else
# define ZSTDLIB_STATIC_API ZSTDLIB_VISIBLE
# endif
#endif
/****************************************************************************************
* experimental API (static linking only)
****************************************************************************************
* The following symbols and constants
* are not planned to join "stable API" status in the near future.
* They can still change in future versions.
* Some of them are planned to remain in the static_only section indefinitely.
* Some of them might be removed in the future (especially when redundant with existing stable functions)
* ***************************************************************************************/
#define ZSTD_FRAMEHEADERSIZE_PREFIX(format) ((format) == ZSTD_f_zstd1 ? 5 : 1) /* minimum input size required to query frame header size */
#define ZSTD_FRAMEHEADERSIZE_MIN(format) ((format) == ZSTD_f_zstd1 ? 6 : 2)
#define ZSTD_FRAMEHEADERSIZE_MAX 18 /* can be useful for static allocation */
#define ZSTD_SKIPPABLEHEADERSIZE 8
/* compression parameter bounds */
#define ZSTD_WINDOWLOG_MAX_32 30
#define ZSTD_WINDOWLOG_MAX_64 31
#define ZSTD_WINDOWLOG_MAX ((int)(sizeof(size_t) == 4 ? ZSTD_WINDOWLOG_MAX_32 : ZSTD_WINDOWLOG_MAX_64))
#define ZSTD_WINDOWLOG_MIN 10
#define ZSTD_HASHLOG_MAX ((ZSTD_WINDOWLOG_MAX < 30) ? ZSTD_WINDOWLOG_MAX : 30)
#define ZSTD_HASHLOG_MIN 6
#define ZSTD_CHAINLOG_MAX_32 29
#define ZSTD_CHAINLOG_MAX_64 30
#define ZSTD_CHAINLOG_MAX ((int)(sizeof(size_t) == 4 ? ZSTD_CHAINLOG_MAX_32 : ZSTD_CHAINLOG_MAX_64))
#define ZSTD_CHAINLOG_MIN ZSTD_HASHLOG_MIN
#define ZSTD_SEARCHLOG_MAX (ZSTD_WINDOWLOG_MAX-1)
#define ZSTD_SEARCHLOG_MIN 1
#define ZSTD_MINMATCH_MAX 7 /* only for ZSTD_fast, other strategies are limited to 6 */
#define ZSTD_MINMATCH_MIN 3 /* only for ZSTD_btopt+, faster strategies are limited to 4 */
#define ZSTD_TARGETLENGTH_MAX ZSTD_BLOCKSIZE_MAX
#define ZSTD_TARGETLENGTH_MIN 0 /* note : comparing this constant to an unsigned results in a tautological test */
#define ZSTD_STRATEGY_MIN ZSTD_fast
#define ZSTD_STRATEGY_MAX ZSTD_btultra2
#define ZSTD_BLOCKSIZE_MAX_MIN (1 << 10) /* The minimum valid max blocksize. Maximum blocksizes smaller than this make compressBound() inaccurate. */
#define ZSTD_OVERLAPLOG_MIN 0
#define ZSTD_OVERLAPLOG_MAX 9
#define ZSTD_WINDOWLOG_LIMIT_DEFAULT 27 /* by default, the streaming decoder will refuse any frame
* requiring larger than (1<<ZSTD_WINDOWLOG_LIMIT_DEFAULT) window size,
* to preserve host's memory from unreasonable requirements.
* This limit can be overridden using ZSTD_DCtx_setParameter(,ZSTD_d_windowLogMax,).
* The limit does not apply for one-pass decoders (such as ZSTD_decompress()), since no additional memory is allocated */
/* LDM parameter bounds */
#define ZSTD_LDM_HASHLOG_MIN ZSTD_HASHLOG_MIN
#define ZSTD_LDM_HASHLOG_MAX ZSTD_HASHLOG_MAX
#define ZSTD_LDM_MINMATCH_MIN 4
#define ZSTD_LDM_MINMATCH_MAX 4096
#define ZSTD_LDM_BUCKETSIZELOG_MIN 1
#define ZSTD_LDM_BUCKETSIZELOG_MAX 8
#define ZSTD_LDM_HASHRATELOG_MIN 0
#define ZSTD_LDM_HASHRATELOG_MAX (ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN)
/* Advanced parameter bounds */
#define ZSTD_TARGETCBLOCKSIZE_MIN 64
#define ZSTD_TARGETCBLOCKSIZE_MAX ZSTD_BLOCKSIZE_MAX
#define ZSTD_SRCSIZEHINT_MIN 0
#define ZSTD_SRCSIZEHINT_MAX INT_MAX
/* --- Advanced types --- */
typedef struct ZSTD_CCtx_params_s ZSTD_CCtx_params;
typedef struct {
unsigned int offset; /* The offset of the match. (NOT the same as the offset code)
* If offset == 0 and matchLength == 0, this sequence represents the last
* literals in the block of litLength size.
*/
unsigned int litLength; /* Literal length of the sequence. */
unsigned int matchLength; /* Match length of the sequence. */
/* Note: Users of this API may provide a sequence with matchLength == litLength == offset == 0.
* In this case, we will treat the sequence as a marker for a block boundary.
*/
unsigned int rep; /* Represents which repeat offset is represented by the field 'offset'.
* Ranges from [0, 3].
*
* Repeat offsets are essentially previous offsets from previous sequences sorted in
* recency order. For more detail, see doc/zstd_compression_format.md
*
* If rep == 0, then 'offset' does not contain a repeat offset.
* If rep > 0:
* If litLength != 0:
* rep == 1 --> offset == repeat_offset_1
* rep == 2 --> offset == repeat_offset_2
* rep == 3 --> offset == repeat_offset_3
* If litLength == 0:
* rep == 1 --> offset == repeat_offset_2
* rep == 2 --> offset == repeat_offset_3
* rep == 3 --> offset == repeat_offset_1 - 1
*
* Note: This field is optional. ZSTD_generateSequences() will calculate the value of
* 'rep', but repeat offsets do not necessarily need to be calculated from an external
* sequence provider's perspective. For example, ZSTD_compressSequences() does not
* use this 'rep' field at all (as of now).
*/
} ZSTD_Sequence;
typedef struct {
unsigned windowLog; /**< largest match distance : larger == more compression, more memory needed during decompression */
unsigned chainLog; /**< fully searched segment : larger == more compression, slower, more memory (useless for fast) */
unsigned hashLog; /**< dispatch table : larger == faster, more memory */
unsigned searchLog; /**< nb of searches : larger == more compression, slower */
unsigned minMatch; /**< match length searched : larger == faster decompression, sometimes less compression */
unsigned targetLength; /**< acceptable match size for optimal parser (only) : larger == more compression, slower */
ZSTD_strategy strategy; /**< see ZSTD_strategy definition above */
} ZSTD_compressionParameters;
typedef struct {
int contentSizeFlag; /**< 1: content size will be in frame header (when known) */
int checksumFlag; /**< 1: generate a 32-bits checksum using XXH64 algorithm at end of frame, for error detection */
int noDictIDFlag; /**< 1: no dictID will be saved into frame header (dictID is only useful for dictionary compression) */
} ZSTD_frameParameters;
typedef struct {
ZSTD_compressionParameters cParams;
ZSTD_frameParameters fParams;
} ZSTD_parameters;
typedef enum {
ZSTD_dct_auto = 0, /* dictionary is "full" when starting with ZSTD_MAGIC_DICTIONARY, otherwise it is "rawContent" */
ZSTD_dct_rawContent = 1, /* ensures dictionary is always loaded as rawContent, even if it starts with ZSTD_MAGIC_DICTIONARY */
ZSTD_dct_fullDict = 2 /* refuses to load a dictionary if it does not respect Zstandard's specification, starting with ZSTD_MAGIC_DICTIONARY */
} ZSTD_dictContentType_e;
typedef enum {
ZSTD_dlm_byCopy = 0, /**< Copy dictionary content internally */
ZSTD_dlm_byRef = 1 /**< Reference dictionary content -- the dictionary buffer must outlive its users. */
} ZSTD_dictLoadMethod_e;
typedef enum {
ZSTD_f_zstd1 = 0, /* zstd frame format, specified in zstd_compression_format.md (default) */
ZSTD_f_zstd1_magicless = 1 /* Variant of zstd frame format, without initial 4-bytes magic number.
* Useful to save 4 bytes per generated frame.
* Decoder cannot recognise automatically this format, requiring this instruction. */
} ZSTD_format_e;
typedef enum {
/* Note: this enum controls ZSTD_d_forceIgnoreChecksum */
ZSTD_d_validateChecksum = 0,
ZSTD_d_ignoreChecksum = 1
} ZSTD_forceIgnoreChecksum_e;
typedef enum {
/* Note: this enum controls ZSTD_d_refMultipleDDicts */
ZSTD_rmd_refSingleDDict = 0,
ZSTD_rmd_refMultipleDDicts = 1
} ZSTD_refMultipleDDicts_e;
typedef enum {
/* Note: this enum and the behavior it controls are effectively internal
* implementation details of the compressor. They are expected to continue
* to evolve and should be considered only in the context of extremely
* advanced performance tuning.
*
* Zstd currently supports the use of a CDict in three ways:
*
* - The contents of the CDict can be copied into the working context. This
* means that the compression can search both the dictionary and input
* while operating on a single set of internal tables. This makes
* the compression faster per-byte of input. However, the initial copy of
* the CDict's tables incurs a fixed cost at the beginning of the
* compression. For small compressions (< 8 KB), that copy can dominate
* the cost of the compression.
*
* - The CDict's tables can be used in-place. In this model, compression is
* slower per input byte, because the compressor has to search two sets of
* tables. However, this model incurs no start-up cost (as long as the
* working context's tables can be reused). For small inputs, this can be
* faster than copying the CDict's tables.
*
* - The CDict's tables are not used at all, and instead we use the working
* context alone to reload the dictionary and use params based on the source
* size. See ZSTD_compress_insertDictionary() and ZSTD_compress_usingDict().
* This method is effective when the dictionary sizes are very small relative
* to the input size, and the input size is fairly large to begin with.
*
* Zstd has a simple internal heuristic that selects which strategy to use
* at the beginning of a compression. However, if experimentation shows that
* Zstd is making poor choices, it is possible to override that choice with
* this enum.
*/
ZSTD_dictDefaultAttach = 0, /* Use the default heuristic. */
ZSTD_dictForceAttach = 1, /* Never copy the dictionary. */
ZSTD_dictForceCopy = 2, /* Always copy the dictionary. */
ZSTD_dictForceLoad = 3 /* Always reload the dictionary */
} ZSTD_dictAttachPref_e;
typedef enum {
ZSTD_lcm_auto = 0, /**< Automatically determine the compression mode based on the compression level.
* Negative compression levels will be uncompressed, and positive compression
* levels will be compressed. */
ZSTD_lcm_huffman = 1, /**< Always attempt Huffman compression. Uncompressed literals will still be
* emitted if Huffman compression is not profitable. */
ZSTD_lcm_uncompressed = 2 /**< Always emit uncompressed literals. */
} ZSTD_literalCompressionMode_e;
typedef enum {
/* Note: This enum controls features which are conditionally beneficial. Zstd typically will make a final
* decision on whether or not to enable the feature (ZSTD_ps_auto), but setting the switch to ZSTD_ps_enable
* or ZSTD_ps_disable allow for a force enable/disable the feature.
*/
ZSTD_ps_auto = 0, /* Let the library automatically determine whether the feature shall be enabled */
ZSTD_ps_enable = 1, /* Force-enable the feature */
ZSTD_ps_disable = 2 /* Do not use the feature */
} ZSTD_paramSwitch_e;
/***************************************
* Frame header and size functions
***************************************/
/*! ZSTD_findDecompressedSize() :
* `src` should point to the start of a series of ZSTD encoded and/or skippable frames
* `srcSize` must be the _exact_ size of this series
* (i.e. there should be a frame boundary at `src + srcSize`)
* @return : - decompressed size of all data in all successive frames
* - if the decompressed size cannot be determined: ZSTD_CONTENTSIZE_UNKNOWN
* - if an error occurred: ZSTD_CONTENTSIZE_ERROR
*
* note 1 : decompressed size is an optional field, that may not be present, especially in streaming mode.
* When `return==ZSTD_CONTENTSIZE_UNKNOWN`, data to decompress could be any size.
* In which case, it's necessary to use streaming mode to decompress data.
* note 2 : decompressed size is always present when compression is done with ZSTD_compress()
* note 3 : decompressed size can be very large (64-bits value),
* potentially larger than what local system can handle as a single memory segment.
* In which case, it's necessary to use streaming mode to decompress data.
* note 4 : If source is untrusted, decompressed size could be wrong or intentionally modified.
* Always ensure result fits within application's authorized limits.
* Each application can set its own limits.
* note 5 : ZSTD_findDecompressedSize handles multiple frames, and so it must traverse the input to
* read each contained frame header. This is fast as most of the data is skipped,
* however it does mean that all frame data must be present and valid. */
ZSTDLIB_STATIC_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
/*! ZSTD_decompressBound() :
* `src` should point to the start of a series of ZSTD encoded and/or skippable frames
* `srcSize` must be the _exact_ size of this series
* (i.e. there should be a frame boundary at `src + srcSize`)
* @return : - upper-bound for the decompressed size of all data in all successive frames
* - if an error occurred: ZSTD_CONTENTSIZE_ERROR
*
* note 1 : an error can occur if `src` contains an invalid or incorrectly formatted frame.
* note 2 : the upper-bound is exact when the decompressed size field is available in every ZSTD encoded frame of `src`.
* in this case, `ZSTD_findDecompressedSize` and `ZSTD_decompressBound` return the same value.
* note 3 : when the decompressed size field isn't available, the upper-bound for that frame is calculated by:
* upper-bound = # blocks * min(128 KB, Window_Size)
*/
ZSTDLIB_STATIC_API unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize);
/*! ZSTD_frameHeaderSize() :
* srcSize must be >= ZSTD_FRAMEHEADERSIZE_PREFIX.
* @return : size of the Frame Header,
* or an error code (if srcSize is too small) */
ZSTDLIB_STATIC_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
typedef enum { ZSTD_frame, ZSTD_skippableFrame } ZSTD_frameType_e;
typedef struct {
unsigned long long frameContentSize; /* if == ZSTD_CONTENTSIZE_UNKNOWN, it means this field is not available. 0 means "empty" */
unsigned long long windowSize; /* can be very large, up to <= frameContentSize */
unsigned blockSizeMax;
ZSTD_frameType_e frameType; /* if == ZSTD_skippableFrame, frameContentSize is the size of skippable content */
unsigned headerSize;
unsigned dictID;
unsigned checksumFlag;
unsigned _reserved1;
unsigned _reserved2;
} ZSTD_frameHeader;
/*! ZSTD_getFrameHeader() :
* decode Frame Header, or requires larger `srcSize`.
* @return : 0, `zfhPtr` is correctly filled,
* >0, `srcSize` is too small, value is wanted `srcSize` amount,
* or an error code, which can be tested using ZSTD_isError() */
ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize); /**< doesn't consume input */
/*! ZSTD_getFrameHeader_advanced() :
* same as ZSTD_getFrameHeader(),
* with added capability to select a format (like ZSTD_f_zstd1_magicless) */
ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format);
/*! ZSTD_decompressionMargin() :
* Zstd supports in-place decompression, where the input and output buffers overlap.
* In this case, the output buffer must be at least (Margin + Output_Size) bytes large,
* and the input buffer must be at the end of the output buffer.
*
* _______________________ Output Buffer ________________________
* | |
* | ____ Input Buffer ____|
* | | |
* v v v
* |---------------------------------------|-----------|----------|
* ^ ^ ^
* |___________________ Output_Size ___________________|_ Margin _|
*
* NOTE: See also ZSTD_DECOMPRESSION_MARGIN().
* NOTE: This applies only to single-pass decompression through ZSTD_decompress() or
* ZSTD_decompressDCtx().
* NOTE: This function supports multi-frame input.
*
* @param src The compressed frame(s)
* @param srcSize The size of the compressed frame(s)
* @returns The decompression margin or an error that can be checked with ZSTD_isError().
*/
ZSTDLIB_STATIC_API size_t ZSTD_decompressionMargin(const void* src, size_t srcSize);
/*! ZSTD_DECOMPRESS_MARGIN() :
* Similar to ZSTD_decompressionMargin(), but instead of computing the margin from
* the compressed frame, compute it from the original size and the blockSizeLog.
* See ZSTD_decompressionMargin() for details.
*
* WARNING: This macro does not support multi-frame input, the input must be a single
* zstd frame. If you need that support use the function, or implement it yourself.
*
* @param originalSize The original uncompressed size of the data.
* @param blockSize The block size == MIN(windowSize, ZSTD_BLOCKSIZE_MAX).
* Unless you explicitly set the windowLog smaller than
* ZSTD_BLOCKSIZELOG_MAX you can just use ZSTD_BLOCKSIZE_MAX.
*/
#define ZSTD_DECOMPRESSION_MARGIN(originalSize, blockSize) ((size_t)( \
ZSTD_FRAMEHEADERSIZE_MAX /* Frame header */ + \
4 /* checksum */ + \
((originalSize) == 0 ? 0 : 3 * (((originalSize) + (blockSize) - 1) / blockSize)) /* 3 bytes per block */ + \
(blockSize) /* One block of margin */ \
))
typedef enum {
ZSTD_sf_noBlockDelimiters = 0, /* Representation of ZSTD_Sequence has no block delimiters, sequences only */
ZSTD_sf_explicitBlockDelimiters = 1 /* Representation of ZSTD_Sequence contains explicit block delimiters */
} ZSTD_sequenceFormat_e;
/*! ZSTD_sequenceBound() :
* `srcSize` : size of the input buffer
* @return : upper-bound for the number of sequences that can be generated
* from a buffer of srcSize bytes
*
* note : returns number of sequences - to get bytes, multiply by sizeof(ZSTD_Sequence).
*/
ZSTDLIB_STATIC_API size_t ZSTD_sequenceBound(size_t srcSize);
/*! ZSTD_generateSequences() :
* Generate sequences using ZSTD_compress2(), given a source buffer.
*
* Each block will end with a dummy sequence
* with offset == 0, matchLength == 0, and litLength == length of last literals.
* litLength may be == 0, and if so, then the sequence of (of: 0 ml: 0 ll: 0)
* simply acts as a block delimiter.
*
* @zc can be used to insert custom compression params.
* This function invokes ZSTD_compress2().
*
* The output of this function can be fed into ZSTD_compressSequences() with CCtx
* setting of ZSTD_c_blockDelimiters as ZSTD_sf_explicitBlockDelimiters
* @return : number of sequences generated
*/
ZSTDLIB_STATIC_API size_t
ZSTD_generateSequences( ZSTD_CCtx* zc,
ZSTD_Sequence* outSeqs, size_t outSeqsSize,
const void* src, size_t srcSize);
/*! ZSTD_mergeBlockDelimiters() :
* Given an array of ZSTD_Sequence, remove all sequences that represent block delimiters/last literals
* by merging them into the literals of the next sequence.
*
* As such, the final generated result has no explicit representation of block boundaries,
* and the final last literals segment is not represented in the sequences.
*
* The output of this function can be fed into ZSTD_compressSequences() with CCtx
* setting of ZSTD_c_blockDelimiters as ZSTD_sf_noBlockDelimiters
* @return : number of sequences left after merging
*/
ZSTDLIB_STATIC_API size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize);
/*! ZSTD_compressSequences() :
* Compress an array of ZSTD_Sequence, associated with @src buffer, into dst.
* @src contains the entire input (not just the literals).
* If @srcSize > sum(sequence.length), the remaining bytes are considered all literals
* If a dictionary is included, then the cctx should reference the dict. (see: ZSTD_CCtx_refCDict(), ZSTD_CCtx_loadDictionary(), etc.)
* The entire source is compressed into a single frame.
*
* The compression behavior changes based on cctx params. In particular:
* If ZSTD_c_blockDelimiters == ZSTD_sf_noBlockDelimiters, the array of ZSTD_Sequence is expected to contain
* no block delimiters (defined in ZSTD_Sequence). Block boundaries are roughly determined based on
* the block size derived from the cctx, and sequences may be split. This is the default setting.
*
* If ZSTD_c_blockDelimiters == ZSTD_sf_explicitBlockDelimiters, the array of ZSTD_Sequence is expected to contain
* block delimiters (defined in ZSTD_Sequence). Behavior is undefined if no block delimiters are provided.
*
* If ZSTD_c_validateSequences == 0, this function will blindly accept the sequences provided. Invalid sequences cause undefined
* behavior. If ZSTD_c_validateSequences == 1, then if sequence is invalid (see doc/zstd_compression_format.md for
* specifics regarding offset/matchlength requirements) then the function will bail out and return an error.
*
* In addition to the two adjustable experimental params, there are other important cctx params.
* - ZSTD_c_minMatch MUST be set as less than or equal to the smallest match generated by the match finder. It has a minimum value of ZSTD_MINMATCH_MIN.
* - ZSTD_c_compressionLevel accordingly adjusts the strength of the entropy coder, as it would in typical compression.
* - ZSTD_c_windowLog affects offset validation: this function will return an error at higher debug levels if a provided offset
* is larger than what the spec allows for a given window log and dictionary (if present). See: doc/zstd_compression_format.md
*
* Note: Repcodes are, as of now, always re-calculated within this function, so ZSTD_Sequence::rep is unused.
* Note 2: Once we integrate ability to ingest repcodes, the explicit block delims mode must respect those repcodes exactly,
* and cannot emit an RLE block that disagrees with the repcode history
* @return : final compressed size, or a ZSTD error code.
*/
ZSTDLIB_STATIC_API size_t
ZSTD_compressSequences( ZSTD_CCtx* cctx, void* dst, size_t dstSize,
const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
const void* src, size_t srcSize);
/*! ZSTD_writeSkippableFrame() :
* Generates a zstd skippable frame containing data given by src, and writes it to dst buffer.
*
* Skippable frames begin with a 4-byte magic number. There are 16 possible choices of magic number,
* ranging from ZSTD_MAGIC_SKIPPABLE_START to ZSTD_MAGIC_SKIPPABLE_START+15.
* As such, the parameter magicVariant controls the exact skippable frame magic number variant used, so
* the magic number used will be ZSTD_MAGIC_SKIPPABLE_START + magicVariant.
*
* Returns an error if destination buffer is not large enough, if the source size is not representable
* with a 4-byte unsigned int, or if the parameter magicVariant is greater than 15 (and therefore invalid).
*
* @return : number of bytes written or a ZSTD error.
*/
ZSTDLIB_STATIC_API size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
const void* src, size_t srcSize, unsigned magicVariant);
/*! ZSTD_readSkippableFrame() :
* Retrieves a zstd skippable frame containing data given by src, and writes it to dst buffer.
*
* The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
* i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START. This can be NULL if the caller is not interested
* in the magicVariant.
*
* Returns an error if destination buffer is not large enough, or if the frame is not skippable.
*
* @return : number of bytes written or a ZSTD error.
*/
ZSTDLIB_API size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity, unsigned* magicVariant,
const void* src, size_t srcSize);
/*! ZSTD_isSkippableFrame() :
* Tells if the content of `buffer` starts with a valid Frame Identifier for a skippable frame.
*/
ZSTDLIB_API unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size);
/***************************************
* Memory management
***************************************/
/*! ZSTD_estimate*() :
* These functions make it possible to estimate memory usage
* of a future {D,C}Ctx, before its creation.
* This is useful in combination with ZSTD_initStatic(),
* which makes it possible to employ a static buffer for ZSTD_CCtx* state.
*
* ZSTD_estimateCCtxSize() will provide a memory budget large enough
* to compress data of any size using one-shot compression ZSTD_compressCCtx() or ZSTD_compress2()
* associated with any compression level up to max specified one.
* The estimate will assume the input may be arbitrarily large,
* which is the worst case.
*
* Note that the size estimation is specific for one-shot compression,
* it is not valid for streaming (see ZSTD_estimateCStreamSize*())
* nor other potential ways of using a ZSTD_CCtx* state.
*
* When srcSize can be bound by a known and rather "small" value,
* this knowledge can be used to provide a tighter budget estimation
* because the ZSTD_CCtx* state will need less memory for small inputs.
* This tighter estimation can be provided by employing more advanced functions
* ZSTD_estimateCCtxSize_usingCParams(), which can be used in tandem with ZSTD_getCParams(),
* and ZSTD_estimateCCtxSize_usingCCtxParams(), which can be used in tandem with ZSTD_CCtxParams_setParameter().
* Both can be used to estimate memory using custom compression parameters and arbitrary srcSize limits.
*
* Note : only single-threaded compression is supported.
* ZSTD_estimateCCtxSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
*
* Note 2 : ZSTD_estimateCCtxSize* functions are not compatible with the Block-Level Sequence Producer API at this time.
* Size estimates assume that no external sequence producer is registered.
*/
ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize(int maxCompressionLevel);
ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams);
ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params);
ZSTDLIB_STATIC_API size_t ZSTD_estimateDCtxSize(void);
/*! ZSTD_estimateCStreamSize() :
* ZSTD_estimateCStreamSize() will provide a memory budget large enough for streaming compression
* using any compression level up to the max specified one.
* It will also consider src size to be arbitrarily "large", which is a worst case scenario.
* If srcSize is known to always be small, ZSTD_estimateCStreamSize_usingCParams() can provide a tighter estimation.
* ZSTD_estimateCStreamSize_usingCParams() can be used in tandem with ZSTD_getCParams() to create cParams from compressionLevel.
* ZSTD_estimateCStreamSize_usingCCtxParams() can be used in tandem with ZSTD_CCtxParams_setParameter(). Only single-threaded compression is supported. This function will return an error code if ZSTD_c_nbWorkers is >= 1.
* Note : CStream size estimation is only correct for single-threaded compression.
* ZSTD_estimateCStreamSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
* Note 2 : ZSTD_estimateCStreamSize* functions are not compatible with the Block-Level Sequence Producer API at this time.
* Size estimates assume that no external sequence producer is registered.
*
* ZSTD_DStream memory budget depends on frame's window Size.
* This information can be passed manually, using ZSTD_estimateDStreamSize,
* or deducted from a valid frame Header, using ZSTD_estimateDStreamSize_fromFrame();
* Any frame requesting a window size larger than max specified one will be rejected.
* Note : if streaming is init with function ZSTD_init?Stream_usingDict(),
* an internal ?Dict will be created, which additional size is not estimated here.
* In this case, get total size by adding ZSTD_estimate?DictSize
*/
ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize(int maxCompressionLevel);
ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams);
ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params);
ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize(size_t maxWindowSize);
ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize);
/*! ZSTD_estimate?DictSize() :
* ZSTD_estimateCDictSize() will bet that src size is relatively "small", and content is copied, like ZSTD_createCDict().
* ZSTD_estimateCDictSize_advanced() makes it possible to control compression parameters precisely, like ZSTD_createCDict_advanced().
* Note : dictionaries created by reference (`ZSTD_dlm_byRef`) are logically smaller.
*/
ZSTDLIB_STATIC_API size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel);
ZSTDLIB_STATIC_API size_t ZSTD_estimateCDictSize_advanced(size_t dictSize, ZSTD_compressionParameters cParams, ZSTD_dictLoadMethod_e dictLoadMethod);
ZSTDLIB_STATIC_API size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod);
/*! ZSTD_initStatic*() :
* Initialize an object using a pre-allocated fixed-size buffer.
* workspace: The memory area to emplace the object into.
* Provided pointer *must be 8-bytes aligned*.
* Buffer must outlive object.
* workspaceSize: Use ZSTD_estimate*Size() to determine
* how large workspace must be to support target scenario.
* @return : pointer to object (same address as workspace, just different type),
* or NULL if error (size too small, incorrect alignment, etc.)
* Note : zstd will never resize nor malloc() when using a static buffer.
* If the object requires more memory than available,
* zstd will just error out (typically ZSTD_error_memory_allocation).
* Note 2 : there is no corresponding "free" function.
* Since workspace is allocated externally, it must be freed externally too.
* Note 3 : cParams : use ZSTD_getCParams() to convert a compression level
* into its associated cParams.
* Limitation 1 : currently not compatible with internal dictionary creation, triggered by
* ZSTD_CCtx_loadDictionary(), ZSTD_initCStream_usingDict() or ZSTD_initDStream_usingDict().
* Limitation 2 : static cctx currently not compatible with multi-threading.
* Limitation 3 : static dctx is incompatible with legacy support.
*/
ZSTDLIB_STATIC_API ZSTD_CCtx* ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize);
ZSTDLIB_STATIC_API ZSTD_CStream* ZSTD_initStaticCStream(void* workspace, size_t workspaceSize); /**< same as ZSTD_initStaticCCtx() */
ZSTDLIB_STATIC_API ZSTD_DCtx* ZSTD_initStaticDCtx(void* workspace, size_t workspaceSize);
ZSTDLIB_STATIC_API ZSTD_DStream* ZSTD_initStaticDStream(void* workspace, size_t workspaceSize); /**< same as ZSTD_initStaticDCtx() */
ZSTDLIB_STATIC_API const ZSTD_CDict* ZSTD_initStaticCDict(
void* workspace, size_t workspaceSize,
const void* dict, size_t dictSize,
ZSTD_dictLoadMethod_e dictLoadMethod,
ZSTD_dictContentType_e dictContentType,
ZSTD_compressionParameters cParams);
ZSTDLIB_STATIC_API const ZSTD_DDict* ZSTD_initStaticDDict(
void* workspace, size_t workspaceSize,
const void* dict, size_t dictSize,
ZSTD_dictLoadMethod_e dictLoadMethod,
ZSTD_dictContentType_e dictContentType);
/*! Custom memory allocation :
* These prototypes make it possible to pass your own allocation/free functions.
* ZSTD_customMem is provided at creation time, using ZSTD_create*_advanced() variants listed below.
* All allocation/free operations will be completed using these custom variants instead of regular <stdlib.h> ones.
*/
typedef void* (*ZSTD_allocFunction) (void* opaque, size_t size);
typedef void (*ZSTD_freeFunction) (void* opaque, void* address);
typedef struct { ZSTD_allocFunction customAlloc; ZSTD_freeFunction customFree; void* opaque; } ZSTD_customMem;
static
#ifdef __GNUC__
__attribute__((__unused__))
#endif
ZSTD_customMem const ZSTD_defaultCMem = { NULL, NULL, NULL }; /**< this constant defers to stdlib's functions */
ZSTDLIB_STATIC_API ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem);
ZSTDLIB_STATIC_API ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem);
ZSTDLIB_STATIC_API ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem);
ZSTDLIB_STATIC_API ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem);
ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize,
ZSTD_dictLoadMethod_e dictLoadMethod,
ZSTD_dictContentType_e dictContentType,
ZSTD_compressionParameters cParams,
ZSTD_customMem customMem);
/*! Thread pool :
* These prototypes make it possible to share a thread pool among multiple compression contexts.
* This can limit resources for applications with multiple threads where each one uses
* a threaded compression mode (via ZSTD_c_nbWorkers parameter).
* ZSTD_createThreadPool creates a new thread pool with a given number of threads.
* Note that the lifetime of such pool must exist while being used.
* ZSTD_CCtx_refThreadPool assigns a thread pool to a context (use NULL argument value
* to use an internal thread pool).
* ZSTD_freeThreadPool frees a thread pool, accepts NULL pointer.
*/
typedef struct POOL_ctx_s ZSTD_threadPool;
ZSTDLIB_STATIC_API ZSTD_threadPool* ZSTD_createThreadPool(size_t numThreads);
ZSTDLIB_STATIC_API void ZSTD_freeThreadPool (ZSTD_threadPool* pool); /* accept NULL pointer */
ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refThreadPool(ZSTD_CCtx* cctx, ZSTD_threadPool* pool);
/*
* This API is temporary and is expected to change or disappear in the future!
*/
ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_advanced2(
const void* dict, size_t dictSize,
ZSTD_dictLoadMethod_e dictLoadMethod,
ZSTD_dictContentType_e dictContentType,
const ZSTD_CCtx_params* cctxParams,
ZSTD_customMem customMem);
ZSTDLIB_STATIC_API ZSTD_DDict* ZSTD_createDDict_advanced(
const void* dict, size_t dictSize,
ZSTD_dictLoadMethod_e dictLoadMethod,
ZSTD_dictContentType_e dictContentType,
ZSTD_customMem customMem);
/***************************************
* Advanced compression functions
***************************************/
/*! ZSTD_createCDict_byReference() :
* Create a digested dictionary for compression
* Dictionary content is just referenced, not duplicated.
* As a consequence, `dictBuffer` **must** outlive CDict,
* and its content must remain unmodified throughout the lifetime of CDict.
* note: equivalent to ZSTD_createCDict_advanced(), with dictLoadMethod==ZSTD_dlm_byRef */
ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
/*! ZSTD_getCParams() :
* @return ZSTD_compressionParameters structure for a selected compression level and estimated srcSize.
* `estimatedSrcSize` value is optional, select 0 if not known */
ZSTDLIB_STATIC_API ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
/*! ZSTD_getParams() :
* same as ZSTD_getCParams(), but @return a full `ZSTD_parameters` object instead of sub-component `ZSTD_compressionParameters`.
* All fields of `ZSTD_frameParameters` are set to default : contentSize=1, checksum=0, noDictID=0 */
ZSTDLIB_STATIC_API ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
/*! ZSTD_checkCParams() :
* Ensure param values remain within authorized range.
* @return 0 on success, or an error code (can be checked with ZSTD_isError()) */
ZSTDLIB_STATIC_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
/*! ZSTD_adjustCParams() :
* optimize params for a given `srcSize` and `dictSize`.
* `srcSize` can be unknown, in which case use ZSTD_CONTENTSIZE_UNKNOWN.
* `dictSize` must be `0` when there is no dictionary.
* cPar can be invalid : all parameters will be clamped within valid range in the @return struct.
* This function never fails (wide contract) */
ZSTDLIB_STATIC_API ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
/*! ZSTD_CCtx_setCParams() :
* Set all parameters provided within @p cparams into the working @p cctx.
* Note : if modifying parameters during compression (MT mode only),
* note that changes to the .windowLog parameter will be ignored.
* @return 0 on success, or an error code (can be checked with ZSTD_isError()).
* On failure, no parameters are updated.
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setCParams(ZSTD_CCtx* cctx, ZSTD_compressionParameters cparams);
/*! ZSTD_CCtx_setFParams() :
* Set all parameters provided within @p fparams into the working @p cctx.
* @return 0 on success, or an error code (can be checked with ZSTD_isError()).
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setFParams(ZSTD_CCtx* cctx, ZSTD_frameParameters fparams);
/*! ZSTD_CCtx_setParams() :
* Set all parameters provided within @p params into the working @p cctx.
* @return 0 on success, or an error code (can be checked with ZSTD_isError()).
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setParams(ZSTD_CCtx* cctx, ZSTD_parameters params);
/*! ZSTD_compress_advanced() :
* Note : this function is now DEPRECATED.
* It can be replaced by ZSTD_compress2(), in combination with ZSTD_CCtx_setParameter() and other parameter setters.
* This prototype will generate compilation warnings. */
ZSTD_DEPRECATED("use ZSTD_compress2")
ZSTDLIB_STATIC_API
size_t ZSTD_compress_advanced(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const void* dict,size_t dictSize,
ZSTD_parameters params);
/*! ZSTD_compress_usingCDict_advanced() :
* Note : this function is now DEPRECATED.
* It can be replaced by ZSTD_compress2(), in combination with ZSTD_CCtx_loadDictionary() and other parameter setters.
* This prototype will generate compilation warnings. */
ZSTD_DEPRECATED("use ZSTD_compress2 with ZSTD_CCtx_loadDictionary")
ZSTDLIB_STATIC_API
size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const ZSTD_CDict* cdict,
ZSTD_frameParameters fParams);
/*! ZSTD_CCtx_loadDictionary_byReference() :
* Same as ZSTD_CCtx_loadDictionary(), but dictionary content is referenced, instead of being copied into CCtx.
* It saves some memory, but also requires that `dict` outlives its usage within `cctx` */
ZSTDLIB_STATIC_API size_t ZSTD_CCtx_loadDictionary_byReference(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
/*! ZSTD_CCtx_loadDictionary_advanced() :
* Same as ZSTD_CCtx_loadDictionary(), but gives finer control over
* how to load the dictionary (by copy ? by reference ?)
* and how to interpret it (automatic ? force raw mode ? full mode only ?) */
ZSTDLIB_STATIC_API size_t ZSTD_CCtx_loadDictionary_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
/*! ZSTD_CCtx_refPrefix_advanced() :
* Same as ZSTD_CCtx_refPrefix(), but gives finer control over
* how to interpret prefix content (automatic ? force raw mode (default) ? full mode only ?) */
ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
/* === experimental parameters === */
/* these parameters can be used with ZSTD_setParameter()
* they are not guaranteed to remain supported in the future */
/* Enables rsyncable mode,
* which makes compressed files more rsync friendly
* by adding periodic synchronization points to the compressed data.
* The target average block size is ZSTD_c_jobSize / 2.
* It's possible to modify the job size to increase or decrease
* the granularity of the synchronization point.
* Once the jobSize is smaller than the window size,
* it will result in compression ratio degradation.
* NOTE 1: rsyncable mode only works when multithreading is enabled.
* NOTE 2: rsyncable performs poorly in combination with long range mode,
* since it will decrease the effectiveness of synchronization points,
* though mileage may vary.
* NOTE 3: Rsyncable mode limits maximum compression speed to ~400 MB/s.
* If the selected compression level is already running significantly slower,
* the overall speed won't be significantly impacted.
*/
#define ZSTD_c_rsyncable ZSTD_c_experimentalParam1
/* Select a compression format.
* The value must be of type ZSTD_format_e.
* See ZSTD_format_e enum definition for details */
#define ZSTD_c_format ZSTD_c_experimentalParam2
/* Force back-reference distances to remain < windowSize,
* even when referencing into Dictionary content (default:0) */
#define ZSTD_c_forceMaxWindow ZSTD_c_experimentalParam3
/* Controls whether the contents of a CDict
* are used in place, or copied into the working context.
* Accepts values from the ZSTD_dictAttachPref_e enum.
* See the comments on that enum for an explanation of the feature. */
#define ZSTD_c_forceAttachDict ZSTD_c_experimentalParam4
/* Controlled with ZSTD_paramSwitch_e enum.
* Default is ZSTD_ps_auto.
* Set to ZSTD_ps_disable to never compress literals.
* Set to ZSTD_ps_enable to always compress literals. (Note: uncompressed literals
* may still be emitted if huffman is not beneficial to use.)
*
* By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
* literals compression based on the compression parameters - specifically,
* negative compression levels do not use literal compression.
*/
#define ZSTD_c_literalCompressionMode ZSTD_c_experimentalParam5
/* Tries to fit compressed block size to be around targetCBlockSize.
* No target when targetCBlockSize == 0.
* There is no guarantee on compressed block size (default:0) */
#define ZSTD_c_targetCBlockSize ZSTD_c_experimentalParam6
/* User's best guess of source size.
* Hint is not valid when srcSizeHint == 0.
* There is no guarantee that hint is close to actual source size,
* but compression ratio may regress significantly if guess considerably underestimates */
#define ZSTD_c_srcSizeHint ZSTD_c_experimentalParam7
/* Controls whether the new and experimental "dedicated dictionary search
* structure" can be used. This feature is still rough around the edges, be
* prepared for surprising behavior!
*
* How to use it:
*
* When using a CDict, whether to use this feature or not is controlled at
* CDict creation, and it must be set in a CCtxParams set passed into that
* construction (via ZSTD_createCDict_advanced2()). A compression will then
* use the feature or not based on how the CDict was constructed; the value of
* this param, set in the CCtx, will have no effect.
*
* However, when a dictionary buffer is passed into a CCtx, such as via
* ZSTD_CCtx_loadDictionary(), this param can be set on the CCtx to control
* whether the CDict that is created internally can use the feature or not.
*
* What it does:
*
* Normally, the internal data structures of the CDict are analogous to what
* would be stored in a CCtx after compressing the contents of a dictionary.
* To an approximation, a compression using a dictionary can then use those
* data structures to simply continue what is effectively a streaming
* compression where the simulated compression of the dictionary left off.
* Which is to say, the search structures in the CDict are normally the same
* format as in the CCtx.
*
* It is possible to do better, since the CDict is not like a CCtx: the search
* structures are written once during CDict creation, and then are only read
* after that, while the search structures in the CCtx are both read and
* written as the compression goes along. This means we can choose a search
* structure for the dictionary that is read-optimized.
*
* This feature enables the use of that different structure.
*
* Note that some of the members of the ZSTD_compressionParameters struct have
* different semantics and constraints in the dedicated search structure. It is
* highly recommended that you simply set a compression level in the CCtxParams
* you pass into the CDict creation call, and avoid messing with the cParams
* directly.
*
* Effects:
*
* This will only have any effect when the selected ZSTD_strategy
* implementation supports this feature. Currently, that's limited to
* ZSTD_greedy, ZSTD_lazy, and ZSTD_lazy2.
*
* Note that this means that the CDict tables can no longer be copied into the
* CCtx, so the dict attachment mode ZSTD_dictForceCopy will no longer be
* usable. The dictionary can only be attached or reloaded.
*
* In general, you should expect compression to be faster--sometimes very much
* so--and CDict creation to be slightly slower. Eventually, we will probably
* make this mode the default.
*/
#define ZSTD_c_enableDedicatedDictSearch ZSTD_c_experimentalParam8
/* ZSTD_c_stableInBuffer
* Experimental parameter.
* Default is 0 == disabled. Set to 1 to enable.
*
* Tells the compressor that input data presented with ZSTD_inBuffer
* will ALWAYS be the same between calls.
* Technically, the @src pointer must never be changed,
* and the @pos field can only be updated by zstd.
* However, it's possible to increase the @size field,
* allowing scenarios where more data can be appended after compressions starts.
* These conditions are checked by the compressor,
* and compression will fail if they are not respected.
* Also, data in the ZSTD_inBuffer within the range [src, src + pos)
* MUST not be modified during compression or it will result in data corruption.
*
* When this flag is enabled zstd won't allocate an input window buffer,
* because the user guarantees it can reference the ZSTD_inBuffer until
* the frame is complete. But, it will still allocate an output buffer
* large enough to fit a block (see ZSTD_c_stableOutBuffer). This will also
* avoid the memcpy() from the input buffer to the input window buffer.
*
* NOTE: So long as the ZSTD_inBuffer always points to valid memory, using
* this flag is ALWAYS memory safe, and will never access out-of-bounds
* memory. However, compression WILL fail if conditions are not respected.
*
* WARNING: The data in the ZSTD_inBuffer in the range [src, src + pos) MUST
* not be modified during compression or it will result in data corruption.
* This is because zstd needs to reference data in the ZSTD_inBuffer to find
* matches. Normally zstd maintains its own window buffer for this purpose,
* but passing this flag tells zstd to rely on user provided buffer instead.
*/
#define ZSTD_c_stableInBuffer ZSTD_c_experimentalParam9
/* ZSTD_c_stableOutBuffer
* Experimental parameter.
* Default is 0 == disabled. Set to 1 to enable.
*
* Tells he compressor that the ZSTD_outBuffer will not be resized between
* calls. Specifically: (out.size - out.pos) will never grow. This gives the
* compressor the freedom to say: If the compressed data doesn't fit in the
* output buffer then return ZSTD_error_dstSizeTooSmall. This allows us to
* always decompress directly into the output buffer, instead of decompressing
* into an internal buffer and copying to the output buffer.
*
* When this flag is enabled zstd won't allocate an output buffer, because
* it can write directly to the ZSTD_outBuffer. It will still allocate the
* input window buffer (see ZSTD_c_stableInBuffer).
*
* Zstd will check that (out.size - out.pos) never grows and return an error
* if it does. While not strictly necessary, this should prevent surprises.
*/
#define ZSTD_c_stableOutBuffer ZSTD_c_experimentalParam10
/* ZSTD_c_blockDelimiters
* Default is 0 == ZSTD_sf_noBlockDelimiters.
*
* For use with sequence compression API: ZSTD_compressSequences().
*
* Designates whether or not the given array of ZSTD_Sequence contains block delimiters
* and last literals, which are defined as sequences with offset == 0 and matchLength == 0.
* See the definition of ZSTD_Sequence for more specifics.
*/
#define ZSTD_c_blockDelimiters ZSTD_c_experimentalParam11
/* ZSTD_c_validateSequences
* Default is 0 == disabled. Set to 1 to enable sequence validation.
*
* For use with sequence compression API: ZSTD_compressSequences().
* Designates whether or not we validate sequences provided to ZSTD_compressSequences()
* during function execution.
*
* Without validation, providing a sequence that does not conform to the zstd spec will cause
* undefined behavior, and may produce a corrupted block.
*
* With validation enabled, if sequence is invalid (see doc/zstd_compression_format.md for
* specifics regarding offset/matchlength requirements) then the function will bail out and
* return an error.
*
*/
#define ZSTD_c_validateSequences ZSTD_c_experimentalParam12
/* ZSTD_c_useBlockSplitter
* Controlled with ZSTD_paramSwitch_e enum.
* Default is ZSTD_ps_auto.
* Set to ZSTD_ps_disable to never use block splitter.
* Set to ZSTD_ps_enable to always use block splitter.
*
* By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
* block splitting based on the compression parameters.
*/
#define ZSTD_c_useBlockSplitter ZSTD_c_experimentalParam13
/* ZSTD_c_useRowMatchFinder
* Controlled with ZSTD_paramSwitch_e enum.
* Default is ZSTD_ps_auto.
* Set to ZSTD_ps_disable to never use row-based matchfinder.
* Set to ZSTD_ps_enable to force usage of row-based matchfinder.
*
* By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
* the row-based matchfinder based on support for SIMD instructions and the window log.
* Note that this only pertains to compression strategies: greedy, lazy, and lazy2
*/
#define ZSTD_c_useRowMatchFinder ZSTD_c_experimentalParam14
/* ZSTD_c_deterministicRefPrefix
* Default is 0 == disabled. Set to 1 to enable.
*
* Zstd produces different results for prefix compression when the prefix is
* directly adjacent to the data about to be compressed vs. when it isn't.
* This is because zstd detects that the two buffers are contiguous and it can
* use a more efficient match finding algorithm. However, this produces different
* results than when the two buffers are non-contiguous. This flag forces zstd
* to always load the prefix in non-contiguous mode, even if it happens to be
* adjacent to the data, to guarantee determinism.
*
* If you really care about determinism when using a dictionary or prefix,
* like when doing delta compression, you should select this option. It comes
* at a speed penalty of about ~2.5% if the dictionary and data happened to be
* contiguous, and is free if they weren't contiguous. We don't expect that
* intentionally making the dictionary and data contiguous will be worth the
* cost to memcpy() the data.
*/
#define ZSTD_c_deterministicRefPrefix ZSTD_c_experimentalParam15
/* ZSTD_c_prefetchCDictTables
* Controlled with ZSTD_paramSwitch_e enum. Default is ZSTD_ps_auto.
*
* In some situations, zstd uses CDict tables in-place rather than copying them
* into the working context. (See docs on ZSTD_dictAttachPref_e above for details).
* In such situations, compression speed is seriously impacted when CDict tables are
* "cold" (outside CPU cache). This parameter instructs zstd to prefetch CDict tables
* when they are used in-place.
*
* For sufficiently small inputs, the cost of the prefetch will outweigh the benefit.
* For sufficiently large inputs, zstd will by default memcpy() CDict tables
* into the working context, so there is no need to prefetch. This parameter is
* targeted at a middle range of input sizes, where a prefetch is cheap enough to be
* useful but memcpy() is too expensive. The exact range of input sizes where this
* makes sense is best determined by careful experimentation.
*
* Note: for this parameter, ZSTD_ps_auto is currently equivalent to ZSTD_ps_disable,
* but in the future zstd may conditionally enable this feature via an auto-detection
* heuristic for cold CDicts.
* Use ZSTD_ps_disable to opt out of prefetching under any circumstances.
*/
#define ZSTD_c_prefetchCDictTables ZSTD_c_experimentalParam16
/* ZSTD_c_enableSeqProducerFallback
* Allowed values are 0 (disable) and 1 (enable). The default setting is 0.
*
* Controls whether zstd will fall back to an internal sequence producer if an
* external sequence producer is registered and returns an error code. This fallback
* is block-by-block: the internal sequence producer will only be called for blocks
* where the external sequence producer returns an error code. Fallback parsing will
* follow any other cParam settings, such as compression level, the same as in a
* normal (fully-internal) compression operation.
*
* The user is strongly encouraged to read the full Block-Level Sequence Producer API
* documentation (below) before setting this parameter. */
#define ZSTD_c_enableSeqProducerFallback ZSTD_c_experimentalParam17
/* ZSTD_c_maxBlockSize
* Allowed values are between 1KB and ZSTD_BLOCKSIZE_MAX (128KB).
* The default is ZSTD_BLOCKSIZE_MAX, and setting to 0 will set to the default.
*
* This parameter can be used to set an upper bound on the blocksize
* that overrides the default ZSTD_BLOCKSIZE_MAX. It cannot be used to set upper
* bounds greater than ZSTD_BLOCKSIZE_MAX or bounds lower than 1KB (will make
* compressBound() inaccurate). Only currently meant to be used for testing.
*
*/
#define ZSTD_c_maxBlockSize ZSTD_c_experimentalParam18
/* ZSTD_c_searchForExternalRepcodes
* This parameter affects how zstd parses external sequences, such as sequences
* provided through the compressSequences() API or from an external block-level
* sequence producer.
*
* If set to ZSTD_ps_enable, the library will check for repeated offsets in
* external sequences, even if those repcodes are not explicitly indicated in
* the "rep" field. Note that this is the only way to exploit repcode matches
* while using compressSequences() or an external sequence producer, since zstd
* currently ignores the "rep" field of external sequences.
*
* If set to ZSTD_ps_disable, the library will not exploit repeated offsets in
* external sequences, regardless of whether the "rep" field has been set. This
* reduces sequence compression overhead by about 25% while sacrificing some
* compression ratio.
*
* The default value is ZSTD_ps_auto, for which the library will enable/disable
* based on compression level.
*
* Note: for now, this param only has an effect if ZSTD_c_blockDelimiters is
* set to ZSTD_sf_explicitBlockDelimiters. That may change in the future.
*/
#define ZSTD_c_searchForExternalRepcodes ZSTD_c_experimentalParam19
/*! ZSTD_CCtx_getParameter() :
* Get the requested compression parameter value, selected by enum ZSTD_cParameter,
* and store it into int* value.
* @return : 0, or an error code (which can be tested with ZSTD_isError()).
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtx_getParameter(const ZSTD_CCtx* cctx, ZSTD_cParameter param, int* value);
/*! ZSTD_CCtx_params :
* Quick howto :
* - ZSTD_createCCtxParams() : Create a ZSTD_CCtx_params structure
* - ZSTD_CCtxParams_setParameter() : Push parameters one by one into
* an existing ZSTD_CCtx_params structure.
* This is similar to
* ZSTD_CCtx_setParameter().
* - ZSTD_CCtx_setParametersUsingCCtxParams() : Apply parameters to
* an existing CCtx.
* These parameters will be applied to
* all subsequent frames.
* - ZSTD_compressStream2() : Do compression using the CCtx.
* - ZSTD_freeCCtxParams() : Free the memory, accept NULL pointer.
*
* This can be used with ZSTD_estimateCCtxSize_advanced_usingCCtxParams()
* for static allocation of CCtx for single-threaded compression.
*/
ZSTDLIB_STATIC_API ZSTD_CCtx_params* ZSTD_createCCtxParams(void);
ZSTDLIB_STATIC_API size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params); /* accept NULL pointer */
/*! ZSTD_CCtxParams_reset() :
* Reset params to default values.
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params);
/*! ZSTD_CCtxParams_init() :
* Initializes the compression parameters of cctxParams according to
* compression level. All other parameters are reset to their default values.
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel);
/*! ZSTD_CCtxParams_init_advanced() :
* Initializes the compression and frame parameters of cctxParams according to
* params. All other parameters are reset to their default values.
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params);
/*! ZSTD_CCtxParams_setParameter() : Requires v1.4.0+
* Similar to ZSTD_CCtx_setParameter.
* Set one compression parameter, selected by enum ZSTD_cParameter.
* Parameters must be applied to a ZSTD_CCtx using
* ZSTD_CCtx_setParametersUsingCCtxParams().
* @result : a code representing success or failure (which can be tested with
* ZSTD_isError()).
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, int value);
/*! ZSTD_CCtxParams_getParameter() :
* Similar to ZSTD_CCtx_getParameter.
* Get the requested value of one compression parameter, selected by enum ZSTD_cParameter.
* @result : 0, or an error code (which can be tested with ZSTD_isError()).
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_getParameter(const ZSTD_CCtx_params* params, ZSTD_cParameter param, int* value);
/*! ZSTD_CCtx_setParametersUsingCCtxParams() :
* Apply a set of ZSTD_CCtx_params to the compression context.
* This can be done even after compression is started,
* if nbWorkers==0, this will have no impact until a new compression is started.
* if nbWorkers>=1, new parameters will be picked up at next job,
* with a few restrictions (windowLog, pledgedSrcSize, nbWorkers, jobSize, and overlapLog are not updated).
*/
ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setParametersUsingCCtxParams(
ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params);
/*! ZSTD_compressStream2_simpleArgs() :
* Same as ZSTD_compressStream2(),
* but using only integral types as arguments.
* This variant might be helpful for binders from dynamic languages
* which have troubles handling structures containing memory pointers.
*/
ZSTDLIB_STATIC_API size_t ZSTD_compressStream2_simpleArgs (
ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity, size_t* dstPos,
const void* src, size_t srcSize, size_t* srcPos,
ZSTD_EndDirective endOp);
/***************************************
* Advanced decompression functions
***************************************/
/*! ZSTD_isFrame() :
* Tells if the content of `buffer` starts with a valid Frame Identifier.
* Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
* Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
* Note 3 : Skippable Frame Identifiers are considered valid. */
ZSTDLIB_STATIC_API unsigned ZSTD_isFrame(const void* buffer, size_t size);
/*! ZSTD_createDDict_byReference() :
* Create a digested dictionary, ready to start decompression operation without startup delay.
* Dictionary content is referenced, and therefore stays in dictBuffer.
* It is important that dictBuffer outlives DDict,
* it must remain read accessible throughout the lifetime of DDict */
ZSTDLIB_STATIC_API ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize);
/*! ZSTD_DCtx_loadDictionary_byReference() :
* Same as ZSTD_DCtx_loadDictionary(),
* but references `dict` content instead of copying it into `dctx`.
* This saves memory if `dict` remains around.,
* However, it's imperative that `dict` remains accessible (and unmodified) while being used, so it must outlive decompression. */
ZSTDLIB_STATIC_API size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
/*! ZSTD_DCtx_loadDictionary_advanced() :
* Same as ZSTD_DCtx_loadDictionary(),
* but gives direct control over
* how to load the dictionary (by copy ? by reference ?)
* and how to interpret it (automatic ? force raw mode ? full mode only ?). */
ZSTDLIB_STATIC_API size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
/*! ZSTD_DCtx_refPrefix_advanced() :
* Same as ZSTD_DCtx_refPrefix(), but gives finer control over
* how to interpret prefix content (automatic ? force raw mode (default) ? full mode only ?) */
ZSTDLIB_STATIC_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
/*! ZSTD_DCtx_setMaxWindowSize() :
* Refuses allocating internal buffers for frames requiring a window size larger than provided limit.
* This protects a decoder context from reserving too much memory for itself (potential attack scenario).
* This parameter is only useful in streaming mode, since no internal buffer is allocated in single-pass mode.
* By default, a decompression context accepts all window sizes <= (1 << ZSTD_WINDOWLOG_LIMIT_DEFAULT)
* @return : 0, or an error code (which can be tested using ZSTD_isError()).
*/
ZSTDLIB_STATIC_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize);
/*! ZSTD_DCtx_getParameter() :
* Get the requested decompression parameter value, selected by enum ZSTD_dParameter,
* and store it into int* value.
* @return : 0, or an error code (which can be tested with ZSTD_isError()).
*/
ZSTDLIB_STATIC_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value);
/* ZSTD_d_format
* experimental parameter,
* allowing selection between ZSTD_format_e input compression formats
*/
#define ZSTD_d_format ZSTD_d_experimentalParam1
/* ZSTD_d_stableOutBuffer
* Experimental parameter.
* Default is 0 == disabled. Set to 1 to enable.
*
* Tells the decompressor that the ZSTD_outBuffer will ALWAYS be the same
* between calls, except for the modifications that zstd makes to pos (the
* caller must not modify pos). This is checked by the decompressor, and
* decompression will fail if it ever changes. Therefore the ZSTD_outBuffer
* MUST be large enough to fit the entire decompressed frame. This will be
* checked when the frame content size is known. The data in the ZSTD_outBuffer
* in the range [dst, dst + pos) MUST not be modified during decompression
* or you will get data corruption.
*
* When this flag is enabled zstd won't allocate an output buffer, because
* it can write directly to the ZSTD_outBuffer, but it will still allocate
* an input buffer large enough to fit any compressed block. This will also
* avoid the memcpy() from the internal output buffer to the ZSTD_outBuffer.
* If you need to avoid the input buffer allocation use the buffer-less
* streaming API.
*
* NOTE: So long as the ZSTD_outBuffer always points to valid memory, using
* this flag is ALWAYS memory safe, and will never access out-of-bounds
* memory. However, decompression WILL fail if you violate the preconditions.
*
* WARNING: The data in the ZSTD_outBuffer in the range [dst, dst + pos) MUST
* not be modified during decompression or you will get data corruption. This
* is because zstd needs to reference data in the ZSTD_outBuffer to regenerate
* matches. Normally zstd maintains its own buffer for this purpose, but passing
* this flag tells zstd to use the user provided buffer.
*/
#define ZSTD_d_stableOutBuffer ZSTD_d_experimentalParam2
/* ZSTD_d_forceIgnoreChecksum
* Experimental parameter.
* Default is 0 == disabled. Set to 1 to enable
*
* Tells the decompressor to skip checksum validation during decompression, regardless
* of whether checksumming was specified during compression. This offers some
* slight performance benefits, and may be useful for debugging.
* Param has values of type ZSTD_forceIgnoreChecksum_e
*/
#define ZSTD_d_forceIgnoreChecksum ZSTD_d_experimentalParam3
/* ZSTD_d_refMultipleDDicts
* Experimental parameter.
* Default is 0 == disabled. Set to 1 to enable
*
* If enabled and dctx is allocated on the heap, then additional memory will be allocated
* to store references to multiple ZSTD_DDict. That is, multiple calls of ZSTD_refDDict()
* using a given ZSTD_DCtx, rather than overwriting the previous DDict reference, will instead
* store all references. At decompression time, the appropriate dictID is selected
* from the set of DDicts based on the dictID in the frame.
*
* Usage is simply calling ZSTD_refDDict() on multiple dict buffers.
*
* Param has values of byte ZSTD_refMultipleDDicts_e
*
* WARNING: Enabling this parameter and calling ZSTD_DCtx_refDDict(), will trigger memory
* allocation for the hash table. ZSTD_freeDCtx() also frees this memory.
* Memory is allocated as per ZSTD_DCtx::customMem.
*
* Although this function allocates memory for the table, the user is still responsible for
* memory management of the underlying ZSTD_DDict* themselves.
*/
#define ZSTD_d_refMultipleDDicts ZSTD_d_experimentalParam4
/* ZSTD_d_disableHuffmanAssembly
* Set to 1 to disable the Huffman assembly implementation.
* The default value is 0, which allows zstd to use the Huffman assembly
* implementation if available.
*
* This parameter can be used to disable Huffman assembly at runtime.
* If you want to disable it at compile time you can define the macro
* ZSTD_DISABLE_ASM.
*/
#define ZSTD_d_disableHuffmanAssembly ZSTD_d_experimentalParam5
/* ZSTD_d_maxBlockSize
* Allowed values are between 1KB and ZSTD_BLOCKSIZE_MAX (128KB).
* The default is ZSTD_BLOCKSIZE_MAX, and setting to 0 will set to the default.
*
* Forces the decompressor to reject blocks whose content size is
* larger than the configured maxBlockSize. When maxBlockSize is
* larger than the windowSize, the windowSize is used instead.
* This saves memory on the decoder when you know all blocks are small.
*
* This option is typically used in conjunction with ZSTD_c_maxBlockSize.
*
* WARNING: This causes the decoder to reject otherwise valid frames
* that have block sizes larger than the configured maxBlockSize.
*/
#define ZSTD_d_maxBlockSize ZSTD_d_experimentalParam6
/*! ZSTD_DCtx_setFormat() :
* This function is REDUNDANT. Prefer ZSTD_DCtx_setParameter().
* Instruct the decoder context about what kind of data to decode next.
* This instruction is mandatory to decode data without a fully-formed header,
* such ZSTD_f_zstd1_magicless for example.
* @return : 0, or an error code (which can be tested using ZSTD_isError()). */
ZSTD_DEPRECATED("use ZSTD_DCtx_setParameter() instead")
ZSTDLIB_STATIC_API
size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
/*! ZSTD_decompressStream_simpleArgs() :
* Same as ZSTD_decompressStream(),
* but using only integral types as arguments.
* This can be helpful for binders from dynamic languages
* which have troubles handling structures containing memory pointers.
*/
ZSTDLIB_STATIC_API size_t ZSTD_decompressStream_simpleArgs (
ZSTD_DCtx* dctx,
void* dst, size_t dstCapacity, size_t* dstPos,
const void* src, size_t srcSize, size_t* srcPos);
/********************************************************************
* Advanced streaming functions
* Warning : most of these functions are now redundant with the Advanced API.
* Once Advanced API reaches "stable" status,
* redundant functions will be deprecated, and then at some point removed.
********************************************************************/
/*===== Advanced Streaming compression functions =====*/
/*! ZSTD_initCStream_srcSize() :
* This function is DEPRECATED, and equivalent to:
* ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
* ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
* ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
* ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
*
* pledgedSrcSize must be correct. If it is not known at init time, use
* ZSTD_CONTENTSIZE_UNKNOWN. Note that, for compatibility with older programs,
* "0" also disables frame content size field. It may be enabled in the future.
* This prototype will generate compilation warnings.
*/
ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
ZSTDLIB_STATIC_API
size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
int compressionLevel,
unsigned long long pledgedSrcSize);
/*! ZSTD_initCStream_usingDict() :
* This function is DEPRECATED, and is equivalent to:
* ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
* ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
* ZSTD_CCtx_loadDictionary(zcs, dict, dictSize);
*
* Creates of an internal CDict (incompatible with static CCtx), except if
* dict == NULL or dictSize < 8, in which case no dict is used.
* Note: dict is loaded with ZSTD_dct_auto (treated as a full zstd dictionary if
* it begins with ZSTD_MAGIC_DICTIONARY, else as raw content) and ZSTD_dlm_byCopy.
* This prototype will generate compilation warnings.
*/
ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
ZSTDLIB_STATIC_API
size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs,
const void* dict, size_t dictSize,
int compressionLevel);
/*! ZSTD_initCStream_advanced() :
* This function is DEPRECATED, and is equivalent to:
* ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
* ZSTD_CCtx_setParams(zcs, params);
* ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
* ZSTD_CCtx_loadDictionary(zcs, dict, dictSize);
*
* dict is loaded with ZSTD_dct_auto and ZSTD_dlm_byCopy.
* pledgedSrcSize must be correct.
* If srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN.
* This prototype will generate compilation warnings.
*/
ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
ZSTDLIB_STATIC_API
size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
const void* dict, size_t dictSize,
ZSTD_parameters params,
unsigned long long pledgedSrcSize);
/*! ZSTD_initCStream_usingCDict() :
* This function is DEPRECATED, and equivalent to:
* ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
* ZSTD_CCtx_refCDict(zcs, cdict);
*
* note : cdict will just be referenced, and must outlive compression session
* This prototype will generate compilation warnings.
*/
ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
ZSTDLIB_STATIC_API
size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);
/*! ZSTD_initCStream_usingCDict_advanced() :
* This function is DEPRECATED, and is equivalent to:
* ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
* ZSTD_CCtx_setFParams(zcs, fParams);
* ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
* ZSTD_CCtx_refCDict(zcs, cdict);
*
* same as ZSTD_initCStream_usingCDict(), with control over frame parameters.
* pledgedSrcSize must be correct. If srcSize is not known at init time, use
* value ZSTD_CONTENTSIZE_UNKNOWN.
* This prototype will generate compilation warnings.
*/
ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
ZSTDLIB_STATIC_API
size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
const ZSTD_CDict* cdict,
ZSTD_frameParameters fParams,
unsigned long long pledgedSrcSize);
/*! ZSTD_resetCStream() :
* This function is DEPRECATED, and is equivalent to:
* ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
* ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize);
* Note: ZSTD_resetCStream() interprets pledgedSrcSize == 0 as ZSTD_CONTENTSIZE_UNKNOWN, but
* ZSTD_CCtx_setPledgedSrcSize() does not do the same, so ZSTD_CONTENTSIZE_UNKNOWN must be
* explicitly specified.
*
* start a new frame, using same parameters from previous frame.
* This is typically useful to skip dictionary loading stage, since it will re-use it in-place.
* Note that zcs must be init at least once before using ZSTD_resetCStream().
* If pledgedSrcSize is not known at reset time, use macro ZSTD_CONTENTSIZE_UNKNOWN.
* If pledgedSrcSize > 0, its value must be correct, as it will be written in header, and controlled at the end.
* For the time being, pledgedSrcSize==0 is interpreted as "srcSize unknown" for compatibility with older programs,
* but it will change to mean "empty" in future version, so use macro ZSTD_CONTENTSIZE_UNKNOWN instead.
* @return : 0, or an error code (which can be tested using ZSTD_isError())
* This prototype will generate compilation warnings.
*/
ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
ZSTDLIB_STATIC_API
size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
typedef struct {
unsigned long long ingested; /* nb input bytes read and buffered */
unsigned long long consumed; /* nb input bytes actually compressed */
unsigned long long produced; /* nb of compressed bytes generated and buffered */
unsigned long long flushed; /* nb of compressed bytes flushed : not provided; can be tracked from caller side */
unsigned currentJobID; /* MT only : latest started job nb */
unsigned nbActiveWorkers; /* MT only : nb of workers actively compressing at probe time */
} ZSTD_frameProgression;
/* ZSTD_getFrameProgression() :
* tells how much data has been ingested (read from input)
* consumed (input actually compressed) and produced (output) for current frame.
* Note : (ingested - consumed) is amount of input data buffered internally, not yet compressed.
* Aggregates progression inside active worker threads.
*/
ZSTDLIB_STATIC_API ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx);
/*! ZSTD_toFlushNow() :
* Tell how many bytes are ready to be flushed immediately.
* Useful for multithreading scenarios (nbWorkers >= 1).
* Probe the oldest active job, defined as oldest job not yet entirely flushed,
* and check its output buffer.
* @return : amount of data stored in oldest job and ready to be flushed immediately.
* if @return == 0, it means either :
* + there is no active job (could be checked with ZSTD_frameProgression()), or
* + oldest job is still actively compressing data,
* but everything it has produced has also been flushed so far,
* therefore flush speed is limited by production speed of oldest job
* irrespective of the speed of concurrent (and newer) jobs.
*/
ZSTDLIB_STATIC_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
/*===== Advanced Streaming decompression functions =====*/
/*!
* This function is deprecated, and is equivalent to:
*
* ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
* ZSTD_DCtx_loadDictionary(zds, dict, dictSize);
*
* note: no dictionary will be used if dict == NULL or dictSize < 8
*/
ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_loadDictionary, see zstd.h for detailed instructions")
ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
/*!
* This function is deprecated, and is equivalent to:
*
* ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
* ZSTD_DCtx_refDDict(zds, ddict);
*
* note : ddict is referenced, it must outlive decompression session
*/
ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_refDDict, see zstd.h for detailed instructions")
ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);
/*!
* This function is deprecated, and is equivalent to:
*
* ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
*
* re-use decompression parameters from previous init; saves dictionary loading
*/
ZSTD_DEPRECATED("use ZSTD_DCtx_reset, see zstd.h for detailed instructions")
ZSTDLIB_STATIC_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
/* ********************* BLOCK-LEVEL SEQUENCE PRODUCER API *********************
*
* *** OVERVIEW ***
* The Block-Level Sequence Producer API allows users to provide their own custom
* sequence producer which libzstd invokes to process each block. The produced list
* of sequences (literals and matches) is then post-processed by libzstd to produce
* valid compressed blocks.
*
* This block-level offload API is a more granular complement of the existing
* frame-level offload API compressSequences() (introduced in v1.5.1). It offers
* an easier migration story for applications already integrated with libzstd: the
* user application continues to invoke the same compression functions
* ZSTD_compress2() or ZSTD_compressStream2() as usual, and transparently benefits
* from the specific advantages of the external sequence producer. For example,
* the sequence producer could be tuned to take advantage of known characteristics
* of the input, to offer better speed / ratio, or could leverage hardware
* acceleration not available within libzstd itself.
*
* See contrib/externalSequenceProducer for an example program employing the
* Block-Level Sequence Producer API.
*
* *** USAGE ***
* The user is responsible for implementing a function of type
* ZSTD_sequenceProducer_F. For each block, zstd will pass the following
* arguments to the user-provided function:
*
* - sequenceProducerState: a pointer to a user-managed state for the sequence
* producer.
*
* - outSeqs, outSeqsCapacity: an output buffer for the sequence producer.
* outSeqsCapacity is guaranteed >= ZSTD_sequenceBound(srcSize). The memory
* backing outSeqs is managed by the CCtx.
*
* - src, srcSize: an input buffer for the sequence producer to parse.
* srcSize is guaranteed to be <= ZSTD_BLOCKSIZE_MAX.
*
* - dict, dictSize: a history buffer, which may be empty, which the sequence
* producer may reference as it parses the src buffer. Currently, zstd will
* always pass dictSize == 0 into external sequence producers, but this will
* change in the future.
*
* - compressionLevel: a signed integer representing the zstd compression level
* set by the user for the current operation. The sequence producer may choose
* to use this information to change its compression strategy and speed/ratio
* tradeoff. Note: the compression level does not reflect zstd parameters set
* through the advanced API.
*
* - windowSize: a size_t representing the maximum allowed offset for external
* sequences. Note that sequence offsets are sometimes allowed to exceed the
* windowSize if a dictionary is present, see doc/zstd_compression_format.md
* for details.
*
* The user-provided function shall return a size_t representing the number of
* sequences written to outSeqs. This return value will be treated as an error
* code if it is greater than outSeqsCapacity. The return value must be non-zero
* if srcSize is non-zero. The ZSTD_SEQUENCE_PRODUCER_ERROR macro is provided
* for convenience, but any value greater than outSeqsCapacity will be treated as
* an error code.
*
* If the user-provided function does not return an error code, the sequences
* written to outSeqs must be a valid parse of the src buffer. Data corruption may
* occur if the parse is not valid. A parse is defined to be valid if the
* following conditions hold:
* - The sum of matchLengths and literalLengths must equal srcSize.
* - All sequences in the parse, except for the final sequence, must have
* matchLength >= ZSTD_MINMATCH_MIN. The final sequence must have
* matchLength >= ZSTD_MINMATCH_MIN or matchLength == 0.
* - All offsets must respect the windowSize parameter as specified in
* doc/zstd_compression_format.md.
* - If the final sequence has matchLength == 0, it must also have offset == 0.
*
* zstd will only validate these conditions (and fail compression if they do not
* hold) if the ZSTD_c_validateSequences cParam is enabled. Note that sequence
* validation has a performance cost.
*
* If the user-provided function returns an error, zstd will either fall back
* to an internal sequence producer or fail the compression operation. The user can
* choose between the two behaviors by setting the ZSTD_c_enableSeqProducerFallback
* cParam. Fallback compression will follow any other cParam settings, such as
* compression level, the same as in a normal compression operation.
*
* The user shall instruct zstd to use a particular ZSTD_sequenceProducer_F
* function by calling
* ZSTD_registerSequenceProducer(cctx,
* sequenceProducerState,
* sequenceProducer)
* This setting will persist until the next parameter reset of the CCtx.
*
* The sequenceProducerState must be initialized by the user before calling
* ZSTD_registerSequenceProducer(). The user is responsible for destroying the
* sequenceProducerState.
*
* *** LIMITATIONS ***
* This API is compatible with all zstd compression APIs which respect advanced parameters.
* However, there are three limitations:
*
* First, the ZSTD_c_enableLongDistanceMatching cParam is not currently supported.
* COMPRESSION WILL FAIL if it is enabled and the user tries to compress with a block-level
* external sequence producer.
* - Note that ZSTD_c_enableLongDistanceMatching is auto-enabled by default in some
* cases (see its documentation for details). Users must explicitly set
* ZSTD_c_enableLongDistanceMatching to ZSTD_ps_disable in such cases if an external
* sequence producer is registered.
* - As of this writing, ZSTD_c_enableLongDistanceMatching is disabled by default
* whenever ZSTD_c_windowLog < 128MB, but that's subject to change. Users should
* check the docs on ZSTD_c_enableLongDistanceMatching whenever the Block-Level Sequence
* Producer API is used in conjunction with advanced settings (like ZSTD_c_windowLog).
*
* Second, history buffers are not currently supported. Concretely, zstd will always pass
* dictSize == 0 to the external sequence producer (for now). This has two implications:
* - Dictionaries are not currently supported. Compression will *not* fail if the user
* references a dictionary, but the dictionary won't have any effect.
* - Stream history is not currently supported. All advanced compression APIs, including
* streaming APIs, work with external sequence producers, but each block is treated as
* an independent chunk without history from previous blocks.
*
* Third, multi-threading within a single compression is not currently supported. In other words,
* COMPRESSION WILL FAIL if ZSTD_c_nbWorkers > 0 and an external sequence producer is registered.
* Multi-threading across compressions is fine: simply create one CCtx per thread.
*
* Long-term, we plan to overcome all three limitations. There is no technical blocker to
* overcoming them. It is purely a question of engineering effort.
*/
#define ZSTD_SEQUENCE_PRODUCER_ERROR ((size_t)(-1))
typedef size_t ZSTD_sequenceProducer_F (
void* sequenceProducerState,
ZSTD_Sequence* outSeqs, size_t outSeqsCapacity,
const void* src, size_t srcSize,
const void* dict, size_t dictSize,
int compressionLevel,
size_t windowSize
);
/*! ZSTD_registerSequenceProducer() :
* Instruct zstd to use a block-level external sequence producer function.
*
* The sequenceProducerState must be initialized by the caller, and the caller is
* responsible for managing its lifetime. This parameter is sticky across
* compressions. It will remain set until the user explicitly resets compression
* parameters.
*
* Sequence producer registration is considered to be an "advanced parameter",
* part of the "advanced API". This means it will only have an effect on compression
* APIs which respect advanced parameters, such as compress2() and compressStream2().
* Older compression APIs such as compressCCtx(), which predate the introduction of
* "advanced parameters", will ignore any external sequence producer setting.
*
* The sequence producer can be "cleared" by registering a NULL function pointer. This
* removes all limitations described above in the "LIMITATIONS" section of the API docs.
*
* The user is strongly encouraged to read the full API documentation (above) before
* calling this function. */
ZSTDLIB_STATIC_API void
ZSTD_registerSequenceProducer(
ZSTD_CCtx* cctx,
void* sequenceProducerState,
ZSTD_sequenceProducer_F* sequenceProducer
);
/*********************************************************************
* Buffer-less and synchronous inner streaming functions (DEPRECATED)
*
* This API is deprecated, and will be removed in a future version.
* It allows streaming (de)compression with user allocated buffers.
* However, it is hard to use, and not as well tested as the rest of
* our API.
*
* Please use the normal streaming API instead: ZSTD_compressStream2,
* and ZSTD_decompressStream.
* If there is functionality that you need, but it doesn't provide,
* please open an issue on our GitHub.
********************************************************************* */
/**
Buffer-less streaming compression (synchronous mode)
A ZSTD_CCtx object is required to track streaming operations.
Use ZSTD_createCCtx() / ZSTD_freeCCtx() to manage resource.
ZSTD_CCtx object can be re-used multiple times within successive compression operations.
Start by initializing a context.
Use ZSTD_compressBegin(), or ZSTD_compressBegin_usingDict() for dictionary compression.
Then, consume your input using ZSTD_compressContinue().
There are some important considerations to keep in mind when using this advanced function :
- ZSTD_compressContinue() has no internal buffer. It uses externally provided buffers only.
- Interface is synchronous : input is consumed entirely and produces 1+ compressed blocks.
- Caller must ensure there is enough space in `dst` to store compressed data under worst case scenario.
Worst case evaluation is provided by ZSTD_compressBound().
ZSTD_compressContinue() doesn't guarantee recover after a failed compression.
- ZSTD_compressContinue() presumes prior input ***is still accessible and unmodified*** (up to maximum distance size, see WindowLog).
It remembers all previous contiguous blocks, plus one separated memory segment (which can itself consists of multiple contiguous blocks)
- ZSTD_compressContinue() detects that prior input has been overwritten when `src` buffer overlaps.
In which case, it will "discard" the relevant memory section from its history.
Finish a frame with ZSTD_compressEnd(), which will write the last block(s) and optional checksum.
It's possible to use srcSize==0, in which case, it will write a final empty block to end the frame.
Without last block mark, frames are considered unfinished (hence corrupted) by compliant decoders.
`ZSTD_CCtx` object can be re-used (ZSTD_compressBegin()) to compress again.
*/
/*===== Buffer-less streaming compression functions =====*/
ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
ZSTDLIB_STATIC_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /**< note: fails if cdict==NULL */
ZSTD_DEPRECATED("This function will likely be removed in a future release. It is misleading and has very limited utility.")
ZSTDLIB_STATIC_API
size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /**< note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
ZSTDLIB_STATIC_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
ZSTD_DEPRECATED("The buffer-less API is deprecated in favor of the normal streaming API. See docs.")
ZSTDLIB_STATIC_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
/* The ZSTD_compressBegin_advanced() and ZSTD_compressBegin_usingCDict_advanced() are now DEPRECATED and will generate a compiler warning */
ZSTD_DEPRECATED("use advanced API to access custom parameters")
ZSTDLIB_STATIC_API
size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize); /**< pledgedSrcSize : If srcSize is not known at init time, use ZSTD_CONTENTSIZE_UNKNOWN */
ZSTD_DEPRECATED("use advanced API to access custom parameters")
ZSTDLIB_STATIC_API
size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict, ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize); /* compression parameters are already set within cdict. pledgedSrcSize must be correct. If srcSize is not known, use macro ZSTD_CONTENTSIZE_UNKNOWN */
/**
Buffer-less streaming decompression (synchronous mode)
A ZSTD_DCtx object is required to track streaming operations.
Use ZSTD_createDCtx() / ZSTD_freeDCtx() to manage it.
A ZSTD_DCtx object can be re-used multiple times.
First typical operation is to retrieve frame parameters, using ZSTD_getFrameHeader().
Frame header is extracted from the beginning of compressed frame, so providing only the frame's beginning is enough.
Data fragment must be large enough to ensure successful decoding.
`ZSTD_frameHeaderSize_max` bytes is guaranteed to always be large enough.
result : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
>0 : `srcSize` is too small, please provide at least result bytes on next attempt.
errorCode, which can be tested using ZSTD_isError().
It fills a ZSTD_frameHeader structure with important information to correctly decode the frame,
such as the dictionary ID, content size, or maximum back-reference distance (`windowSize`).
Note that these values could be wrong, either because of data corruption, or because a 3rd party deliberately spoofs false information.
As a consequence, check that values remain within valid application range.
For example, do not allocate memory blindly, check that `windowSize` is within expectation.
Each application can set its own limits, depending on local restrictions.
For extended interoperability, it is recommended to support `windowSize` of at least 8 MB.
ZSTD_decompressContinue() needs previous data blocks during decompression, up to `windowSize` bytes.
ZSTD_decompressContinue() is very sensitive to contiguity,
if 2 blocks don't follow each other, make sure that either the compressor breaks contiguity at the same place,
or that previous contiguous segment is large enough to properly handle maximum back-reference distance.
There are multiple ways to guarantee this condition.
The most memory efficient way is to use a round buffer of sufficient size.
Sufficient size is determined by invoking ZSTD_decodingBufferSize_min(),
which can return an error code if required value is too large for current system (in 32-bits mode).
In a round buffer methodology, ZSTD_decompressContinue() decompresses each block next to previous one,
up to the moment there is not enough room left in the buffer to guarantee decoding another full block,
which maximum size is provided in `ZSTD_frameHeader` structure, field `blockSizeMax`.
At which point, decoding can resume from the beginning of the buffer.
Note that already decoded data stored in the buffer should be flushed before being overwritten.
There are alternatives possible, for example using two or more buffers of size `windowSize` each, though they consume more memory.
Finally, if you control the compression process, you can also ignore all buffer size rules,
as long as the encoder and decoder progress in "lock-step",
aka use exactly the same buffer sizes, break contiguity at the same place, etc.
Once buffers are setup, start decompression, with ZSTD_decompressBegin().
If decompression requires a dictionary, use ZSTD_decompressBegin_usingDict() or ZSTD_decompressBegin_usingDDict().
Then use ZSTD_nextSrcSizeToDecompress() and ZSTD_decompressContinue() alternatively.
ZSTD_nextSrcSizeToDecompress() tells how many bytes to provide as 'srcSize' to ZSTD_decompressContinue().
ZSTD_decompressContinue() requires this _exact_ amount of bytes, or it will fail.
result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
It can be zero : it just means ZSTD_decompressContinue() has decoded some metadata item.
It can also be an error code, which can be tested with ZSTD_isError().
A frame is fully decoded when ZSTD_nextSrcSizeToDecompress() returns zero.
Context can then be reset to start a new decompression.
Note : it's possible to know if next input to present is a header or a block, using ZSTD_nextInputType().
This information is not required to properly decode a frame.
== Special case : skippable frames ==
Skippable frames allow integration of user-defined data into a flow of concatenated frames.
Skippable frames will be ignored (skipped) by decompressor.
The format of skippable frames is as follows :
a) Skippable frame ID - 4 Bytes, Little endian format, any value from 0x184D2A50 to 0x184D2A5F
b) Frame Size - 4 Bytes, Little endian format, unsigned 32-bits
c) Frame Content - any content (User Data) of length equal to Frame Size
For skippable frames ZSTD_getFrameHeader() returns zfhPtr->frameType==ZSTD_skippableFrame.
For skippable frames ZSTD_decompressContinue() always returns 0 : it only skips the content.
*/
/*===== Buffer-less streaming decompression functions =====*/
ZSTDLIB_STATIC_API size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize); /**< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx);
ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
ZSTDLIB_STATIC_API size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
ZSTDLIB_STATIC_API size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
/* misc */
ZSTD_DEPRECATED("This function will likely be removed in the next minor release. It is misleading and has very limited utility.")
ZSTDLIB_STATIC_API void ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
typedef enum { ZSTDnit_frameHeader, ZSTDnit_blockHeader, ZSTDnit_block, ZSTDnit_lastBlock, ZSTDnit_checksum, ZSTDnit_skippableFrame } ZSTD_nextInputType_e;
ZSTDLIB_STATIC_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
/* ========================================= */
/** Block level API (DEPRECATED) */
/* ========================================= */
/*!
This API is deprecated in favor of the regular compression API.
You can get the frame header down to 2 bytes by setting:
- ZSTD_c_format = ZSTD_f_zstd1_magicless
- ZSTD_c_contentSizeFlag = 0
- ZSTD_c_checksumFlag = 0
- ZSTD_c_dictIDFlag = 0
This API is not as well tested as our normal API, so we recommend not using it.
We will be removing it in a future version. If the normal API doesn't provide
the functionality you need, please open a GitHub issue.
Block functions produce and decode raw zstd blocks, without frame metadata.
Frame metadata cost is typically ~12 bytes, which can be non-negligible for very small blocks (< 100 bytes).
But users will have to take in charge needed metadata to regenerate data, such as compressed and content sizes.
A few rules to respect :
- Compressing and decompressing require a context structure
+ Use ZSTD_createCCtx() and ZSTD_createDCtx()
- It is necessary to init context before starting
+ compression : any ZSTD_compressBegin*() variant, including with dictionary
+ decompression : any ZSTD_decompressBegin*() variant, including with dictionary
- Block size is limited, it must be <= ZSTD_getBlockSize() <= ZSTD_BLOCKSIZE_MAX == 128 KB
+ If input is larger than a block size, it's necessary to split input data into multiple blocks
+ For inputs larger than a single block, consider using regular ZSTD_compress() instead.
Frame metadata is not that costly, and quickly becomes negligible as source size grows larger than a block.
- When a block is considered not compressible enough, ZSTD_compressBlock() result will be 0 (zero) !
===> In which case, nothing is produced into `dst` !
+ User __must__ test for such outcome and deal directly with uncompressed data
+ A block cannot be declared incompressible if ZSTD_compressBlock() return value was != 0.
Doing so would mess up with statistics history, leading to potential data corruption.
+ ZSTD_decompressBlock() _doesn't accept uncompressed data as input_ !!
+ In case of multiple successive blocks, should some of them be uncompressed,
decoder must be informed of their existence in order to follow proper history.
Use ZSTD_insertBlock() for such a case.
*/
/*===== Raw zstd block functions =====*/
ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
ZSTDLIB_STATIC_API size_t ZSTD_getBlockSize (const ZSTD_CCtx* cctx);
ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
ZSTDLIB_STATIC_API size_t ZSTD_compressBlock (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
ZSTDLIB_STATIC_API size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
ZSTD_DEPRECATED("The block API is deprecated in favor of the normal compression API. See docs.")
ZSTDLIB_STATIC_API size_t ZSTD_insertBlock (ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize); /**< insert uncompressed block into `dctx` history. Useful for multi-blocks decompression. */
#endif /* ZSTD_H_ZSTD_STATIC_LINKING_ONLY */
#if defined (__cplusplus)
}
#endif
/**** ended inlining ../zstd.h ****/
#ifndef ZSTD_ALLOCATIONS_H
#define ZSTD_ALLOCATIONS_H
/* custom memory allocation functions */
MEM_STATIC void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem)
{
if (customMem.customAlloc)
return customMem.customAlloc(customMem.opaque, size);
return ZSTD_malloc(size);
}
MEM_STATIC void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem)
{
if (customMem.customAlloc) {
/* calloc implemented as malloc+memset;
* not as efficient as calloc, but next best guess for custom malloc */
void* const ptr = customMem.customAlloc(customMem.opaque, size);
ZSTD_memset(ptr, 0, size);
return ptr;
}
return ZSTD_calloc(1, size);
}
MEM_STATIC void ZSTD_customFree(void* ptr, ZSTD_customMem customMem)
{
if (ptr!=NULL) {
if (customMem.customFree)
customMem.customFree(customMem.opaque, ptr);
else
ZSTD_free(ptr);
}
}
#endif /* ZSTD_ALLOCATIONS_H */
/**** ended inlining ../common/allocations.h ****/
/**** skipping file: zstd_deps.h ****/
/**** skipping file: debug.h ****/
/**** start inlining pool.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef POOL_H
#define POOL_H
#if defined (__cplusplus)
extern "C" {
#endif
/**** skipping file: zstd_deps.h ****/
#define ZSTD_STATIC_LINKING_ONLY /* ZSTD_customMem */
/**** skipping file: ../zstd.h ****/
typedef struct POOL_ctx_s POOL_ctx;
/*! POOL_create() :
* Create a thread pool with at most `numThreads` threads.
* `numThreads` must be at least 1.
* The maximum number of queued jobs before blocking is `queueSize`.
* @return : POOL_ctx pointer on success, else NULL.
*/
POOL_ctx* POOL_create(size_t numThreads, size_t queueSize);
POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
ZSTD_customMem customMem);
/*! POOL_free() :
* Free a thread pool returned by POOL_create().
*/
void POOL_free(POOL_ctx* ctx);
/*! POOL_joinJobs() :
* Waits for all queued jobs to finish executing.
*/
void POOL_joinJobs(POOL_ctx* ctx);
/*! POOL_resize() :
* Expands or shrinks pool's number of threads.
* This is more efficient than releasing + creating a new context,
* since it tries to preserve and re-use existing threads.
* `numThreads` must be at least 1.
* @return : 0 when resize was successful,
* !0 (typically 1) if there is an error.
* note : only numThreads can be resized, queueSize remains unchanged.
*/
int POOL_resize(POOL_ctx* ctx, size_t numThreads);
/*! POOL_sizeof() :
* @return threadpool memory usage
* note : compatible with NULL (returns 0 in this case)
*/
size_t POOL_sizeof(const POOL_ctx* ctx);
/*! POOL_function :
* The function type that can be added to a thread pool.
*/
typedef void (*POOL_function)(void*);
/*! POOL_add() :
* Add the job `function(opaque)` to the thread pool. `ctx` must be valid.
* Possibly blocks until there is room in the queue.
* Note : The function may be executed asynchronously,
* therefore, `opaque` must live until function has been completed.
*/
void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque);
/*! POOL_tryAdd() :
* Add the job `function(opaque)` to thread pool _if_ a queue slot is available.
* Returns immediately even if not (does not block).
* @return : 1 if successful, 0 if not.
*/
int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque);
#if defined (__cplusplus)
}
#endif
#endif
/**** ended inlining pool.h ****/
/* ====== Compiler specifics ====== */
#if defined(_MSC_VER)
# pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */
#endif
#ifdef ZSTD_MULTITHREAD
/**** skipping file: threading.h ****/
/* A job is a function and an opaque argument */
typedef struct POOL_job_s {
POOL_function function;
void *opaque;
} POOL_job;
struct POOL_ctx_s {
ZSTD_customMem customMem;
/* Keep track of the threads */
ZSTD_pthread_t* threads;
size_t threadCapacity;
size_t threadLimit;
/* The queue is a circular buffer */
POOL_job *queue;
size_t queueHead;
size_t queueTail;
size_t queueSize;
/* The number of threads working on jobs */
size_t numThreadsBusy;
/* Indicates if the queue is empty */
int queueEmpty;
/* The mutex protects the queue */
ZSTD_pthread_mutex_t queueMutex;
/* Condition variable for pushers to wait on when the queue is full */
ZSTD_pthread_cond_t queuePushCond;
/* Condition variables for poppers to wait on when the queue is empty */
ZSTD_pthread_cond_t queuePopCond;
/* Indicates if the queue is shutting down */
int shutdown;
};
/* POOL_thread() :
* Work thread for the thread pool.
* Waits for jobs and executes them.
* @returns : NULL on failure else non-null.
*/
static void* POOL_thread(void* opaque) {
POOL_ctx* const ctx = (POOL_ctx*)opaque;
if (!ctx) { return NULL; }
for (;;) {
/* Lock the mutex and wait for a non-empty queue or until shutdown */
ZSTD_pthread_mutex_lock(&ctx->queueMutex);
while ( ctx->queueEmpty
|| (ctx->numThreadsBusy >= ctx->threadLimit) ) {
if (ctx->shutdown) {
/* even if !queueEmpty, (possible if numThreadsBusy >= threadLimit),
* a few threads will be shutdown while !queueEmpty,
* but enough threads will remain active to finish the queue */
ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
return opaque;
}
ZSTD_pthread_cond_wait(&ctx->queuePopCond, &ctx->queueMutex);
}
/* Pop a job off the queue */
{ POOL_job const job = ctx->queue[ctx->queueHead];
ctx->queueHead = (ctx->queueHead + 1) % ctx->queueSize;
ctx->numThreadsBusy++;
ctx->queueEmpty = (ctx->queueHead == ctx->queueTail);
/* Unlock the mutex, signal a pusher, and run the job */
ZSTD_pthread_cond_signal(&ctx->queuePushCond);
ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
job.function(job.opaque);
/* If the intended queue size was 0, signal after finishing job */
ZSTD_pthread_mutex_lock(&ctx->queueMutex);
ctx->numThreadsBusy--;
ZSTD_pthread_cond_signal(&ctx->queuePushCond);
ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
}
} /* for (;;) */
assert(0); /* Unreachable */
}
/* ZSTD_createThreadPool() : public access point */
POOL_ctx* ZSTD_createThreadPool(size_t numThreads) {
return POOL_create (numThreads, 0);
}
POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
}
POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
ZSTD_customMem customMem)
{
POOL_ctx* ctx;
/* Check parameters */
if (!numThreads) { return NULL; }
/* Allocate the context and zero initialize */
ctx = (POOL_ctx*)ZSTD_customCalloc(sizeof(POOL_ctx), customMem);
if (!ctx) { return NULL; }
/* Initialize the job queue.
* It needs one extra space since one space is wasted to differentiate
* empty and full queues.
*/
ctx->queueSize = queueSize + 1;
ctx->queue = (POOL_job*)ZSTD_customCalloc(ctx->queueSize * sizeof(POOL_job), customMem);
ctx->queueHead = 0;
ctx->queueTail = 0;
ctx->numThreadsBusy = 0;
ctx->queueEmpty = 1;
{
int error = 0;
error |= ZSTD_pthread_mutex_init(&ctx->queueMutex, NULL);
error |= ZSTD_pthread_cond_init(&ctx->queuePushCond, NULL);
error |= ZSTD_pthread_cond_init(&ctx->queuePopCond, NULL);
if (error) { POOL_free(ctx); return NULL; }
}
ctx->shutdown = 0;
/* Allocate space for the thread handles */
ctx->threads = (ZSTD_pthread_t*)ZSTD_customCalloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
ctx->threadCapacity = 0;
ctx->customMem = customMem;
/* Check for errors */
if (!ctx->threads || !ctx->queue) { POOL_free(ctx); return NULL; }
/* Initialize the threads */
{ size_t i;
for (i = 0; i < numThreads; ++i) {
if (ZSTD_pthread_create(&ctx->threads[i], NULL, &POOL_thread, ctx)) {
ctx->threadCapacity = i;
POOL_free(ctx);
return NULL;
} }
ctx->threadCapacity = numThreads;
ctx->threadLimit = numThreads;
}
return ctx;
}
/*! POOL_join() :
Shutdown the queue, wake any sleeping threads, and join all of the threads.
*/
static void POOL_join(POOL_ctx* ctx) {
/* Shut down the queue */
ZSTD_pthread_mutex_lock(&ctx->queueMutex);
ctx->shutdown = 1;
ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
/* Wake up sleeping threads */
ZSTD_pthread_cond_broadcast(&ctx->queuePushCond);
ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
/* Join all of the threads */
{ size_t i;
for (i = 0; i < ctx->threadCapacity; ++i) {
ZSTD_pthread_join(ctx->threads[i]); /* note : could fail */
} }
}
void POOL_free(POOL_ctx *ctx) {
if (!ctx) { return; }
POOL_join(ctx);
ZSTD_pthread_mutex_destroy(&ctx->queueMutex);
ZSTD_pthread_cond_destroy(&ctx->queuePushCond);
ZSTD_pthread_cond_destroy(&ctx->queuePopCond);
ZSTD_customFree(ctx->queue, ctx->customMem);
ZSTD_customFree(ctx->threads, ctx->customMem);
ZSTD_customFree(ctx, ctx->customMem);
}
/*! POOL_joinJobs() :
* Waits for all queued jobs to finish executing.
*/
void POOL_joinJobs(POOL_ctx* ctx) {
ZSTD_pthread_mutex_lock(&ctx->queueMutex);
while(!ctx->queueEmpty || ctx->numThreadsBusy > 0) {
ZSTD_pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
}
ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
}
void ZSTD_freeThreadPool (ZSTD_threadPool* pool) {
POOL_free (pool);
}
size_t POOL_sizeof(const POOL_ctx* ctx) {
if (ctx==NULL) return 0; /* supports sizeof NULL */
return sizeof(*ctx)
+ ctx->queueSize * sizeof(POOL_job)
+ ctx->threadCapacity * sizeof(ZSTD_pthread_t);
}
/* @return : 0 on success, 1 on error */
static int POOL_resize_internal(POOL_ctx* ctx, size_t numThreads)
{
if (numThreads <= ctx->threadCapacity) {
if (!numThreads) return 1;
ctx->threadLimit = numThreads;
return 0;
}
/* numThreads > threadCapacity */
{ ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_customCalloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
if (!threadPool) return 1;
/* replace existing thread pool */
ZSTD_memcpy(threadPool, ctx->threads, ctx->threadCapacity * sizeof(*threadPool));
ZSTD_customFree(ctx->threads, ctx->customMem);
ctx->threads = threadPool;
/* Initialize additional threads */
{ size_t threadId;
for (threadId = ctx->threadCapacity; threadId < numThreads; ++threadId) {
if (ZSTD_pthread_create(&threadPool[threadId], NULL, &POOL_thread, ctx)) {
ctx->threadCapacity = threadId;
return 1;
} }
} }
/* successfully expanded */
ctx->threadCapacity = numThreads;
ctx->threadLimit = numThreads;
return 0;
}
/* @return : 0 on success, 1 on error */
int POOL_resize(POOL_ctx* ctx, size_t numThreads)
{
int result;
if (ctx==NULL) return 1;
ZSTD_pthread_mutex_lock(&ctx->queueMutex);
result = POOL_resize_internal(ctx, numThreads);
ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
return result;
}
/**
* Returns 1 if the queue is full and 0 otherwise.
*
* When queueSize is 1 (pool was created with an intended queueSize of 0),
* then a queue is empty if there is a thread free _and_ no job is waiting.
*/
static int isQueueFull(POOL_ctx const* ctx) {
if (ctx->queueSize > 1) {
return ctx->queueHead == ((ctx->queueTail + 1) % ctx->queueSize);
} else {
return (ctx->numThreadsBusy == ctx->threadLimit) ||
!ctx->queueEmpty;
}
}
static void
POOL_add_internal(POOL_ctx* ctx, POOL_function function, void *opaque)
{
POOL_job job;
job.function = function;
job.opaque = opaque;
assert(ctx != NULL);
if (ctx->shutdown) return;
ctx->queueEmpty = 0;
ctx->queue[ctx->queueTail] = job;
ctx->queueTail = (ctx->queueTail + 1) % ctx->queueSize;
ZSTD_pthread_cond_signal(&ctx->queuePopCond);
}
void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque)
{
assert(ctx != NULL);
ZSTD_pthread_mutex_lock(&ctx->queueMutex);
/* Wait until there is space in the queue for the new job */
while (isQueueFull(ctx) && (!ctx->shutdown)) {
ZSTD_pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
}
POOL_add_internal(ctx, function, opaque);
ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
}
int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque)
{
assert(ctx != NULL);
ZSTD_pthread_mutex_lock(&ctx->queueMutex);
if (isQueueFull(ctx)) {
ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
return 0;
}
POOL_add_internal(ctx, function, opaque);
ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
return 1;
}
#else /* ZSTD_MULTITHREAD not defined */
/* ========================== */
/* No multi-threading support */
/* ========================== */
/* We don't need any data, but if it is empty, malloc() might return NULL. */
struct POOL_ctx_s {
int dummy;
};
static POOL_ctx g_poolCtx;
POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
}
POOL_ctx*
POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem)
{
(void)numThreads;
(void)queueSize;
(void)customMem;
return &g_poolCtx;
}
void POOL_free(POOL_ctx* ctx) {
assert(!ctx || ctx == &g_poolCtx);
(void)ctx;
}
void POOL_joinJobs(POOL_ctx* ctx){
assert(!ctx || ctx == &g_poolCtx);
(void)ctx;
}
int POOL_resize(POOL_ctx* ctx, size_t numThreads) {
(void)ctx; (void)numThreads;
return 0;
}
void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque) {
(void)ctx;
function(opaque);
}
int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque) {
(void)ctx;
function(opaque);
return 1;
}
size_t POOL_sizeof(const POOL_ctx* ctx) {
if (ctx==NULL) return 0; /* supports sizeof NULL */
assert(ctx == &g_poolCtx);
return sizeof(*ctx);
}
#endif /* ZSTD_MULTITHREAD */
/**** ended inlining common/pool.c ****/
/**** start inlining common/zstd_common.c ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/*-*************************************
* Dependencies
***************************************/
#define ZSTD_DEPS_NEED_MALLOC
/**** skipping file: error_private.h ****/
/**** start inlining zstd_internal.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_CCOMMON_H_MODULE
#define ZSTD_CCOMMON_H_MODULE
/* this module contains definitions which must be identical
* across compression, decompression and dictBuilder.
* It also contains a few functions useful to at least 2 of them
* and which benefit from being inlined */
/*-*************************************
* Dependencies
***************************************/
/**** skipping file: compiler.h ****/
/**** start inlining cpu.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_COMMON_CPU_H
#define ZSTD_COMMON_CPU_H
/**
* Implementation taken from folly/CpuId.h
* https://github.com/facebook/folly/blob/master/folly/CpuId.h
*/
/**** skipping file: mem.h ****/
#ifdef _MSC_VER
#include <intrin.h>
#endif
typedef struct {
U32 f1c;
U32 f1d;
U32 f7b;
U32 f7c;
} ZSTD_cpuid_t;
MEM_STATIC ZSTD_cpuid_t ZSTD_cpuid(void) {
U32 f1c = 0;
U32 f1d = 0;
U32 f7b = 0;
U32 f7c = 0;
#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
int reg[4];
__cpuid((int*)reg, 0);
{
int const n = reg[0];
if (n >= 1) {
__cpuid((int*)reg, 1);
f1c = (U32)reg[2];
f1d = (U32)reg[3];
}
if (n >= 7) {
__cpuidex((int*)reg, 7, 0);
f7b = (U32)reg[1];
f7c = (U32)reg[2];
}
}
#elif defined(__i386__) && defined(__PIC__) && !defined(__clang__) && defined(__GNUC__)
/* The following block like the normal cpuid branch below, but gcc
* reserves ebx for use of its pic register so we must specially
* handle the save and restore to avoid clobbering the register
*/
U32 n;
__asm__(
"pushl %%ebx\n\t"
"cpuid\n\t"
"popl %%ebx\n\t"
: "=a"(n)
: "a"(0)
: "ecx", "edx");
if (n >= 1) {
U32 f1a;
__asm__(
"pushl %%ebx\n\t"
"cpuid\n\t"
"popl %%ebx\n\t"
: "=a"(f1a), "=c"(f1c), "=d"(f1d)
: "a"(1));
}
if (n >= 7) {
__asm__(
"pushl %%ebx\n\t"
"cpuid\n\t"
"movl %%ebx, %%eax\n\t"
"popl %%ebx"
: "=a"(f7b), "=c"(f7c)
: "a"(7), "c"(0)
: "edx");
}
#elif defined(__x86_64__) || defined(_M_X64) || defined(__i386__)
U32 n;
__asm__("cpuid" : "=a"(n) : "a"(0) : "ebx", "ecx", "edx");
if (n >= 1) {
U32 f1a;
__asm__("cpuid" : "=a"(f1a), "=c"(f1c), "=d"(f1d) : "a"(1) : "ebx");
}
if (n >= 7) {
U32 f7a;
__asm__("cpuid"
: "=a"(f7a), "=b"(f7b), "=c"(f7c)
: "a"(7), "c"(0)
: "edx");
}
#endif
{
ZSTD_cpuid_t cpuid;
cpuid.f1c = f1c;
cpuid.f1d = f1d;
cpuid.f7b = f7b;
cpuid.f7c = f7c;
return cpuid;
}
}
#define X(name, r, bit) \
MEM_STATIC int ZSTD_cpuid_##name(ZSTD_cpuid_t const cpuid) { \
return ((cpuid.r) & (1U << bit)) != 0; \
}
/* cpuid(1): Processor Info and Feature Bits. */
#define C(name, bit) X(name, f1c, bit)
C(sse3, 0)
C(pclmuldq, 1)
C(dtes64, 2)
C(monitor, 3)
C(dscpl, 4)
C(vmx, 5)
C(smx, 6)
C(eist, 7)
C(tm2, 8)
C(ssse3, 9)
C(cnxtid, 10)
C(fma, 12)
C(cx16, 13)
C(xtpr, 14)
C(pdcm, 15)
C(pcid, 17)
C(dca, 18)
C(sse41, 19)
C(sse42, 20)
C(x2apic, 21)
C(movbe, 22)
C(popcnt, 23)
C(tscdeadline, 24)
C(aes, 25)
C(xsave, 26)
C(osxsave, 27)
C(avx, 28)
C(f16c, 29)
C(rdrand, 30)
#undef C
#define D(name, bit) X(name, f1d, bit)
D(fpu, 0)
D(vme, 1)
D(de, 2)
D(pse, 3)
D(tsc, 4)
D(msr, 5)
D(pae, 6)
D(mce, 7)
D(cx8, 8)
D(apic, 9)
D(sep, 11)
D(mtrr, 12)
D(pge, 13)
D(mca, 14)
D(cmov, 15)
D(pat, 16)
D(pse36, 17)
D(psn, 18)
D(clfsh, 19)
D(ds, 21)
D(acpi, 22)
D(mmx, 23)
D(fxsr, 24)
D(sse, 25)
D(sse2, 26)
D(ss, 27)
D(htt, 28)
D(tm, 29)
D(pbe, 31)
#undef D
/* cpuid(7): Extended Features. */
#define B(name, bit) X(name, f7b, bit)
B(bmi1, 3)
B(hle, 4)
B(avx2, 5)
B(smep, 7)
B(bmi2, 8)
B(erms, 9)
B(invpcid, 10)
B(rtm, 11)
B(mpx, 14)
B(avx512f, 16)
B(avx512dq, 17)
B(rdseed, 18)
B(adx, 19)
B(smap, 20)
B(avx512ifma, 21)
B(pcommit, 22)
B(clflushopt, 23)
B(clwb, 24)
B(avx512pf, 26)
B(avx512er, 27)
B(avx512cd, 28)
B(sha, 29)
B(avx512bw, 30)
B(avx512vl, 31)
#undef B
#define C(name, bit) X(name, f7c, bit)
C(prefetchwt1, 0)
C(avx512vbmi, 1)
#undef C
#undef X
#endif /* ZSTD_COMMON_CPU_H */
/**** ended inlining cpu.h ****/
/**** skipping file: mem.h ****/
/**** skipping file: debug.h ****/
/**** skipping file: error_private.h ****/
#define ZSTD_STATIC_LINKING_ONLY
/**** skipping file: ../zstd.h ****/
#define FSE_STATIC_LINKING_ONLY
/**** skipping file: fse.h ****/
/**** skipping file: huf.h ****/
#ifndef XXH_STATIC_LINKING_ONLY
# define XXH_STATIC_LINKING_ONLY /* XXH64_state_t */
#endif
/**** start inlining xxhash.h ****/
/*
* xxHash - Fast Hash algorithm
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - xxHash homepage: https://cyan4973.github.io/xxHash/
* - xxHash source repository : https://github.com/Cyan4973/xxHash
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef XXH_NO_XXH3
# define XXH_NO_XXH3
#endif
#ifndef XXH_NAMESPACE
# define XXH_NAMESPACE ZSTD_
#endif
/*!
* @mainpage xxHash
*
* @file xxhash.h
* xxHash prototypes and implementation
*/
/* TODO: update */
/* Notice extracted from xxHash homepage:
xxHash is an extremely fast hash algorithm, running at RAM speed limits.
It also successfully passes all tests from the SMHasher suite.
Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo @3GHz)
Name Speed Q.Score Author
xxHash 5.4 GB/s 10
CrapWow 3.2 GB/s 2 Andrew
MurmurHash 3a 2.7 GB/s 10 Austin Appleby
SpookyHash 2.0 GB/s 10 Bob Jenkins
SBox 1.4 GB/s 9 Bret Mulvey
Lookup3 1.2 GB/s 9 Bob Jenkins
SuperFastHash 1.2 GB/s 1 Paul Hsieh
CityHash64 1.05 GB/s 10 Pike & Alakuijala
FNV 0.55 GB/s 5 Fowler, Noll, Vo
CRC32 0.43 GB/s 9
MD5-32 0.33 GB/s 10 Ronald L. Rivest
SHA1-32 0.28 GB/s 10
Q.Score is a measure of quality of the hash function.
It depends on successfully passing SMHasher test set.
10 is a perfect score.
Note: SMHasher's CRC32 implementation is not the fastest one.
Other speed-oriented implementations can be faster,
especially in combination with PCLMUL instruction:
https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html?showComment=1552696407071#c3490092340461170735
A 64-bit version, named XXH64, is available since r35.
It offers much better speed, but for 64-bit applications only.
Name Speed on 64 bits Speed on 32 bits
XXH64 13.8 GB/s 1.9 GB/s
XXH32 6.8 GB/s 6.0 GB/s
*/
#if defined (__cplusplus)
extern "C" {
#endif
/* ****************************
* INLINE mode
******************************/
/*!
* XXH_INLINE_ALL (and XXH_PRIVATE_API)
* Use these build macros to inline xxhash into the target unit.
* Inlining improves performance on small inputs, especially when the length is
* expressed as a compile-time constant:
*
* https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
*
* It also keeps xxHash symbols private to the unit, so they are not exported.
*
* Usage:
* #define XXH_INLINE_ALL
* #include "xxhash.h"
*
* Do not compile and link xxhash.o as a separate object, as it is not useful.
*/
#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \
&& !defined(XXH_INLINE_ALL_31684351384)
/* this section should be traversed only once */
# define XXH_INLINE_ALL_31684351384
/* give access to the advanced API, required to compile implementations */
# undef XXH_STATIC_LINKING_ONLY /* avoid macro redef */
# define XXH_STATIC_LINKING_ONLY
/* make all functions private */
# undef XXH_PUBLIC_API
# if defined(__GNUC__)
# define XXH_PUBLIC_API static __inline __attribute__((unused))
# elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
# define XXH_PUBLIC_API static inline
# elif defined(_MSC_VER)
# define XXH_PUBLIC_API static __inline
# else
/* note: this version may generate warnings for unused static functions */
# define XXH_PUBLIC_API static
# endif
/*
* This part deals with the special case where a unit wants to inline xxHash,
* but "xxhash.h" has previously been included without XXH_INLINE_ALL,
* such as part of some previously included *.h header file.
* Without further action, the new include would just be ignored,
* and functions would effectively _not_ be inlined (silent failure).
* The following macros solve this situation by prefixing all inlined names,
* avoiding naming collision with previous inclusions.
*/
/* Before that, we unconditionally #undef all symbols,
* in case they were already defined with XXH_NAMESPACE.
* They will then be redefined for XXH_INLINE_ALL
*/
# undef XXH_versionNumber
/* XXH32 */
# undef XXH32
# undef XXH32_createState
# undef XXH32_freeState
# undef XXH32_reset
# undef XXH32_update
# undef XXH32_digest
# undef XXH32_copyState
# undef XXH32_canonicalFromHash
# undef XXH32_hashFromCanonical
/* XXH64 */
# undef XXH64
# undef XXH64_createState
# undef XXH64_freeState
# undef XXH64_reset
# undef XXH64_update
# undef XXH64_digest
# undef XXH64_copyState
# undef XXH64_canonicalFromHash
# undef XXH64_hashFromCanonical
/* XXH3_64bits */
# undef XXH3_64bits
# undef XXH3_64bits_withSecret
# undef XXH3_64bits_withSeed
# undef XXH3_64bits_withSecretandSeed
# undef XXH3_createState
# undef XXH3_freeState
# undef XXH3_copyState
# undef XXH3_64bits_reset
# undef XXH3_64bits_reset_withSeed
# undef XXH3_64bits_reset_withSecret
# undef XXH3_64bits_update
# undef XXH3_64bits_digest
# undef XXH3_generateSecret
/* XXH3_128bits */
# undef XXH128
# undef XXH3_128bits
# undef XXH3_128bits_withSeed
# undef XXH3_128bits_withSecret
# undef XXH3_128bits_reset
# undef XXH3_128bits_reset_withSeed
# undef XXH3_128bits_reset_withSecret
# undef XXH3_128bits_reset_withSecretandSeed
# undef XXH3_128bits_update
# undef XXH3_128bits_digest
# undef XXH128_isEqual
# undef XXH128_cmp
# undef XXH128_canonicalFromHash
# undef XXH128_hashFromCanonical
/* Finally, free the namespace itself */
# undef XXH_NAMESPACE
/* employ the namespace for XXH_INLINE_ALL */
# define XXH_NAMESPACE XXH_INLINE_
/*
* Some identifiers (enums, type names) are not symbols,
* but they must nonetheless be renamed to avoid redeclaration.
* Alternative solution: do not redeclare them.
* However, this requires some #ifdefs, and has a more dispersed impact.
* Meanwhile, renaming can be achieved in a single place.
*/
# define XXH_IPREF(Id) XXH_NAMESPACE ## Id
# define XXH_OK XXH_IPREF(XXH_OK)
# define XXH_ERROR XXH_IPREF(XXH_ERROR)
# define XXH_errorcode XXH_IPREF(XXH_errorcode)
# define XXH32_canonical_t XXH_IPREF(XXH32_canonical_t)
# define XXH64_canonical_t XXH_IPREF(XXH64_canonical_t)
# define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t)
# define XXH32_state_s XXH_IPREF(XXH32_state_s)
# define XXH32_state_t XXH_IPREF(XXH32_state_t)
# define XXH64_state_s XXH_IPREF(XXH64_state_s)
# define XXH64_state_t XXH_IPREF(XXH64_state_t)
# define XXH3_state_s XXH_IPREF(XXH3_state_s)
# define XXH3_state_t XXH_IPREF(XXH3_state_t)
# define XXH128_hash_t XXH_IPREF(XXH128_hash_t)
/* Ensure the header is parsed again, even if it was previously included */
# undef XXHASH_H_5627135585666179
# undef XXHASH_H_STATIC_13879238742
#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */
/* ****************************************************************
* Stable API
*****************************************************************/
#ifndef XXHASH_H_5627135585666179
#define XXHASH_H_5627135585666179 1
/*!
* @defgroup public Public API
* Contains details on the public xxHash functions.
* @{
*/
/* specific declaration modes for Windows */
#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
# if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
# ifdef XXH_EXPORT
# define XXH_PUBLIC_API __declspec(dllexport)
# elif XXH_IMPORT
# define XXH_PUBLIC_API __declspec(dllimport)
# endif
# else
# define XXH_PUBLIC_API /* do nothing */
# endif
#endif
#ifdef XXH_DOXYGEN
/*!
* @brief Emulate a namespace by transparently prefixing all symbols.
*
* If you want to include _and expose_ xxHash functions from within your own
* library, but also want to avoid symbol collisions with other libraries which
* may also include xxHash, you can use XXH_NAMESPACE to automatically prefix
* any public symbol from xxhash library with the value of XXH_NAMESPACE
* (therefore, avoid empty or numeric values).
*
* Note that no change is required within the calling program as long as it
* includes `xxhash.h`: Regular symbol names will be automatically translated
* by this header.
*/
# define XXH_NAMESPACE /* YOUR NAME HERE */
# undef XXH_NAMESPACE
#endif
#ifdef XXH_NAMESPACE
# define XXH_CAT(A,B) A##B
# define XXH_NAME2(A,B) XXH_CAT(A,B)
# define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
/* XXH32 */
# define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
# define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
# define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
# define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
# define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
# define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
# define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
# define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
# define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
/* XXH64 */
# define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
# define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
# define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
# define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
# define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
# define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
# define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
# define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
# define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
/* XXH3_64bits */
# define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits)
# define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret)
# define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed)
# define XXH3_64bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed)
# define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState)
# define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState)
# define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState)
# define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset)
# define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed)
# define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret)
# define XXH3_64bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed)
# define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update)
# define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest)
# define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret)
# define XXH3_generateSecret_fromSeed XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed)
/* XXH3_128bits */
# define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128)
# define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits)
# define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed)
# define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret)
# define XXH3_128bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed)
# define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset)
# define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed)
# define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret)
# define XXH3_128bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed)
# define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update)
# define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest)
# define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual)
# define XXH128_cmp XXH_NAME2(XXH_NAMESPACE, XXH128_cmp)
# define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash)
# define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical)
#endif
/* *************************************
* Version
***************************************/
#define XXH_VERSION_MAJOR 0
#define XXH_VERSION_MINOR 8
#define XXH_VERSION_RELEASE 1
#define XXH_VERSION_NUMBER (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE)
/*!
* @brief Obtains the xxHash version.
*
* This is mostly useful when xxHash is compiled as a shared library,
* since the returned value comes from the library, as opposed to header file.
*
* @return `XXH_VERSION_NUMBER` of the invoked library.
*/
XXH_PUBLIC_API unsigned XXH_versionNumber (void);
/* ****************************
* Common basic types
******************************/
#include <stddef.h> /* size_t */
typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
/*-**********************************************************************
* 32-bit hash
************************************************************************/
#if defined(XXH_DOXYGEN) /* Don't show <stdint.h> include */
/*!
* @brief An unsigned 32-bit integer.
*
* Not necessarily defined to `uint32_t` but functionally equivalent.
*/
typedef uint32_t XXH32_hash_t;
#elif !defined (__VMS) \
&& (defined (__cplusplus) \
|| (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
# include <stdint.h>
typedef uint32_t XXH32_hash_t;
#else
# include <limits.h>
# if UINT_MAX == 0xFFFFFFFFUL
typedef unsigned int XXH32_hash_t;
# else
# if ULONG_MAX == 0xFFFFFFFFUL
typedef unsigned long XXH32_hash_t;
# else
# error "unsupported platform: need a 32-bit type"
# endif
# endif
#endif
/*!
* @}
*
* @defgroup xxh32_family XXH32 family
* @ingroup public
* Contains functions used in the classic 32-bit xxHash algorithm.
*
* @note
* XXH32 is useful for older platforms, with no or poor 64-bit performance.
* Note that @ref xxh3_family provides competitive speed
* for both 32-bit and 64-bit systems, and offers true 64/128 bit hash results.
*
* @see @ref xxh64_family, @ref xxh3_family : Other xxHash families
* @see @ref xxh32_impl for implementation details
* @{
*/
/*!
* @brief Calculates the 32-bit hash of @p input using xxHash32.
*
* Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark): 5.4 GB/s
*
* @param input The block of data to be hashed, at least @p length bytes in size.
* @param length The length of @p input, in bytes.
* @param seed The 32-bit seed to alter the hash's output predictably.
*
* @pre
* The memory between @p input and @p input + @p length must be valid,
* readable, contiguous memory. However, if @p length is `0`, @p input may be
* `NULL`. In C++, this also must be *TriviallyCopyable*.
*
* @return The calculated 32-bit hash value.
*
* @see
* XXH64(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
* Direct equivalents for the other variants of xxHash.
* @see
* XXH32_createState(), XXH32_update(), XXH32_digest(): Streaming version.
*/
XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed);
/*!
* Streaming functions generate the xxHash value from an incremental input.
* This method is slower than single-call functions, due to state management.
* For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized.
*
* An XXH state must first be allocated using `XXH*_createState()`.
*
* Start a new hash by initializing the state with a seed using `XXH*_reset()`.
*
* Then, feed the hash state by calling `XXH*_update()` as many times as necessary.
*
* The function returns an error code, with 0 meaning OK, and any other value
* meaning there is an error.
*
* Finally, a hash value can be produced anytime, by using `XXH*_digest()`.
* This function returns the nn-bits hash as an int or long long.
*
* It's still possible to continue inserting input into the hash state after a
* digest, and generate new hash values later on by invoking `XXH*_digest()`.
*
* When done, release the state using `XXH*_freeState()`.
*
* Example code for incrementally hashing a file:
* @code{.c}
* #include <stdio.h>
* #include <xxhash.h>
* #define BUFFER_SIZE 256
*
* // Note: XXH64 and XXH3 use the same interface.
* XXH32_hash_t
* hashFile(FILE* stream)
* {
* XXH32_state_t* state;
* unsigned char buf[BUFFER_SIZE];
* size_t amt;
* XXH32_hash_t hash;
*
* state = XXH32_createState(); // Create a state
* assert(state != NULL); // Error check here
* XXH32_reset(state, 0xbaad5eed); // Reset state with our seed
* while ((amt = fread(buf, 1, sizeof(buf), stream)) != 0) {
* XXH32_update(state, buf, amt); // Hash the file in chunks
* }
* hash = XXH32_digest(state); // Finalize the hash
* XXH32_freeState(state); // Clean up
* return hash;
* }
* @endcode
*/
/*!
* @typedef struct XXH32_state_s XXH32_state_t
* @brief The opaque state struct for the XXH32 streaming API.
*
* @see XXH32_state_s for details.
*/
typedef struct XXH32_state_s XXH32_state_t;
/*!
* @brief Allocates an @ref XXH32_state_t.
*
* Must be freed with XXH32_freeState().
* @return An allocated XXH32_state_t on success, `NULL` on failure.
*/
XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void);
/*!
* @brief Frees an @ref XXH32_state_t.
*
* Must be allocated with XXH32_createState().
* @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref XXH32_createState().
* @return XXH_OK.
*/
XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr);
/*!
* @brief Copies one @ref XXH32_state_t to another.
*
* @param dst_state The state to copy to.
* @param src_state The state to copy from.
* @pre
* @p dst_state and @p src_state must not be `NULL` and must not overlap.
*/
XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state);
/*!
* @brief Resets an @ref XXH32_state_t to begin a new hash.
*
* This function resets and seeds a state. Call it before @ref XXH32_update().
*
* @param statePtr The state struct to reset.
* @param seed The 32-bit seed to alter the hash result predictably.
*
* @pre
* @p statePtr must not be `NULL`.
*
* @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
*/
XXH_PUBLIC_API XXH_errorcode XXH32_reset (XXH32_state_t* statePtr, XXH32_hash_t seed);
/*!
* @brief Consumes a block of @p input to an @ref XXH32_state_t.
*
* Call this to incrementally consume blocks of data.
*
* @param statePtr The state struct to update.
* @param input The block of data to be hashed, at least @p length bytes in size.
* @param length The length of @p input, in bytes.
*
* @pre
* @p statePtr must not be `NULL`.
* @pre
* The memory between @p input and @p input + @p length must be valid,
* readable, contiguous memory. However, if @p length is `0`, @p input may be
* `NULL`. In C++, this also must be *TriviallyCopyable*.
*
* @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
*/
XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);
/*!
* @brief Returns the calculated hash value from an @ref XXH32_state_t.
*
* @note
* Calling XXH32_digest() will not affect @p statePtr, so you can update,
* digest, and update again.
*
* @param statePtr The state struct to calculate the hash from.
*
* @pre
* @p statePtr must not be `NULL`.
*
* @return The calculated xxHash32 value from that state.
*/
XXH_PUBLIC_API XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr);
/******* Canonical representation *******/
/*
* The default return values from XXH functions are unsigned 32 and 64 bit
* integers.
* This the simplest and fastest format for further post-processing.
*
* However, this leaves open the question of what is the order on the byte level,
* since little and big endian conventions will store the same number differently.
*
* The canonical representation settles this issue by mandating big-endian
* convention, the same convention as human-readable numbers (large digits first).
*
* When writing hash values to storage, sending them over a network, or printing
* them, it's highly recommended to use the canonical representation to ensure
* portability across a wider range of systems, present and future.
*
* The following functions allow transformation of hash values to and from
* canonical format.
*/
/*!
* @brief Canonical (big endian) representation of @ref XXH32_hash_t.
*/
typedef struct {
unsigned char digest[4]; /*!< Hash bytes, big endian */
} XXH32_canonical_t;
/*!
* @brief Converts an @ref XXH32_hash_t to a big endian @ref XXH32_canonical_t.
*
* @param dst The @ref XXH32_canonical_t pointer to be stored to.
* @param hash The @ref XXH32_hash_t to be converted.
*
* @pre
* @p dst must not be `NULL`.
*/
XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);
/*!
* @brief Converts an @ref XXH32_canonical_t to a native @ref XXH32_hash_t.
*
* @param src The @ref XXH32_canonical_t to convert.
*
* @pre
* @p src must not be `NULL`.
*
* @return The converted hash.
*/
XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
#ifdef __has_attribute
# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
#else
# define XXH_HAS_ATTRIBUTE(x) 0
#endif
/* C-language Attributes are added in C23. */
#if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
#else
# define XXH_HAS_C_ATTRIBUTE(x) 0
#endif
#if defined(__cplusplus) && defined(__has_cpp_attribute)
# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
#else
# define XXH_HAS_CPP_ATTRIBUTE(x) 0
#endif
/*
Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute
introduced in CPP17 and C23.
CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough
C23 : https://en.cppreference.com/w/c/language/attributes/fallthrough
*/
#if XXH_HAS_C_ATTRIBUTE(x)
# define XXH_FALLTHROUGH [[fallthrough]]
#elif XXH_HAS_CPP_ATTRIBUTE(x)
# define XXH_FALLTHROUGH [[fallthrough]]
#elif XXH_HAS_ATTRIBUTE(__fallthrough__)
# define XXH_FALLTHROUGH __attribute__ ((fallthrough))
#else
# define XXH_FALLTHROUGH
#endif
/*!
* @}
* @ingroup public
* @{
*/
#ifndef XXH_NO_LONG_LONG
/*-**********************************************************************
* 64-bit hash
************************************************************************/
#if defined(XXH_DOXYGEN) /* don't include <stdint.h> */
/*!
* @brief An unsigned 64-bit integer.
*
* Not necessarily defined to `uint64_t` but functionally equivalent.
*/
typedef uint64_t XXH64_hash_t;
#elif !defined (__VMS) \
&& (defined (__cplusplus) \
|| (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
# include <stdint.h>
typedef uint64_t XXH64_hash_t;
#else
# include <limits.h>
# if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL
/* LP64 ABI says uint64_t is unsigned long */
typedef unsigned long XXH64_hash_t;
# else
/* the following type must have a width of 64-bit */
typedef unsigned long long XXH64_hash_t;
# endif
#endif
/*!
* @}
*
* @defgroup xxh64_family XXH64 family
* @ingroup public
* @{
* Contains functions used in the classic 64-bit xxHash algorithm.
*
* @note
* XXH3 provides competitive speed for both 32-bit and 64-bit systems,
* and offers true 64/128 bit hash results.
* It provides better speed for systems with vector processing capabilities.
*/
/*!
* @brief Calculates the 64-bit hash of @p input using xxHash64.
*
* This function usually runs faster on 64-bit systems, but slower on 32-bit
* systems (see benchmark).
*
* @param input The block of data to be hashed, at least @p length bytes in size.
* @param length The length of @p input, in bytes.
* @param seed The 64-bit seed to alter the hash's output predictably.
*
* @pre
* The memory between @p input and @p input + @p length must be valid,
* readable, contiguous memory. However, if @p length is `0`, @p input may be
* `NULL`. In C++, this also must be *TriviallyCopyable*.
*
* @return The calculated 64-bit hash.
*
* @see
* XXH32(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
* Direct equivalents for the other variants of xxHash.
* @see
* XXH64_createState(), XXH64_update(), XXH64_digest(): Streaming version.
*/
XXH_PUBLIC_API XXH64_hash_t XXH64(const void* input, size_t length, XXH64_hash_t seed);
/******* Streaming *******/
/*!
* @brief The opaque state struct for the XXH64 streaming API.
*
* @see XXH64_state_s for details.
*/
typedef struct XXH64_state_s XXH64_state_t; /* incomplete type */
XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void);
XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr);
XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dst_state, const XXH64_state_t* src_state);
XXH_PUBLIC_API XXH_errorcode XXH64_reset (XXH64_state_t* statePtr, XXH64_hash_t seed);
XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
XXH_PUBLIC_API XXH64_hash_t XXH64_digest (const XXH64_state_t* statePtr);
/******* Canonical representation *******/
typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t;
XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash);
XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src);
#ifndef XXH_NO_XXH3
/*!
* @}
* ************************************************************************
* @defgroup xxh3_family XXH3 family
* @ingroup public
* @{
*
* XXH3 is a more recent hash algorithm featuring:
* - Improved speed for both small and large inputs
* - True 64-bit and 128-bit outputs
* - SIMD acceleration
* - Improved 32-bit viability
*
* Speed analysis methodology is explained here:
*
* https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html
*
* Compared to XXH64, expect XXH3 to run approximately
* ~2x faster on large inputs and >3x faster on small ones,
* exact differences vary depending on platform.
*
* XXH3's speed benefits greatly from SIMD and 64-bit arithmetic,
* but does not require it.
* Any 32-bit and 64-bit targets that can run XXH32 smoothly
* can run XXH3 at competitive speeds, even without vector support.
* Further details are explained in the implementation.
*
* Optimized implementations are provided for AVX512, AVX2, SSE2, NEON, POWER8,
* ZVector and scalar targets. This can be controlled via the XXH_VECTOR macro.
*
* XXH3 implementation is portable:
* it has a generic C90 formulation that can be compiled on any platform,
* all implementations generage exactly the same hash value on all platforms.
* Starting from v0.8.0, it's also labelled "stable", meaning that
* any future version will also generate the same hash value.
*
* XXH3 offers 2 variants, _64bits and _128bits.
*
* When only 64 bits are needed, prefer invoking the _64bits variant, as it
* reduces the amount of mixing, resulting in faster speed on small inputs.
* It's also generally simpler to manipulate a scalar return type than a struct.
*
* The API supports one-shot hashing, streaming mode, and custom secrets.
*/
/*-**********************************************************************
* XXH3 64-bit variant
************************************************************************/
/* XXH3_64bits():
* default 64-bit variant, using default secret and default seed of 0.
* It's the fastest variant. */
XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* data, size_t len);
/*
* XXH3_64bits_withSeed():
* This variant generates a custom secret on the fly
* based on default secret altered using the `seed` value.
* While this operation is decently fast, note that it's not completely free.
* Note: seed==0 produces the same results as XXH3_64bits().
*/
XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
/*!
* The bare minimum size for a custom secret.
*
* @see
* XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(),
* XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret().
*/
#define XXH3_SECRET_SIZE_MIN 136
/*
* XXH3_64bits_withSecret():
* It's possible to provide any blob of bytes as a "secret" to generate the hash.
* This makes it more difficult for an external actor to prepare an intentional collision.
* The main condition is that secretSize *must* be large enough (>= XXH3_SECRET_SIZE_MIN).
* However, the quality of the secret impacts the dispersion of the hash algorithm.
* Therefore, the secret _must_ look like a bunch of random bytes.
* Avoid "trivial" or structured data such as repeated sequences or a text document.
* Whenever in doubt about the "randomness" of the blob of bytes,
* consider employing "XXH3_generateSecret()" instead (see below).
* It will generate a proper high entropy secret derived from the blob of bytes.
* Another advantage of using XXH3_generateSecret() is that
* it guarantees that all bits within the initial blob of bytes
* will impact every bit of the output.
* This is not necessarily the case when using the blob of bytes directly
* because, when hashing _small_ inputs, only a portion of the secret is employed.
*/
XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
/******* Streaming *******/
/*
* Streaming requires state maintenance.
* This operation costs memory and CPU.
* As a consequence, streaming is slower than one-shot hashing.
* For better performance, prefer one-shot functions whenever applicable.
*/
/*!
* @brief The state struct for the XXH3 streaming API.
*
* @see XXH3_state_s for details.
*/
typedef struct XXH3_state_s XXH3_state_t;
XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void);
XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);
XXH_PUBLIC_API void XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state);
/*
* XXH3_64bits_reset():
* Initialize with default parameters.
* digest will be equivalent to `XXH3_64bits()`.
*/
XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH3_state_t* statePtr);
/*
* XXH3_64bits_reset_withSeed():
* Generate a custom secret from `seed`, and store it into `statePtr`.
* digest will be equivalent to `XXH3_64bits_withSeed()`.
*/
XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
/*
* XXH3_64bits_reset_withSecret():
* `secret` is referenced, it _must outlive_ the hash streaming session.
* Similar to one-shot API, `secretSize` must be >= `XXH3_SECRET_SIZE_MIN`,
* and the quality of produced hash values depends on secret's entropy
* (secret's content should look like a bunch of random bytes).
* When in doubt about the randomness of a candidate `secret`,
* consider employing `XXH3_generateSecret()` instead (see below).
*/
XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* statePtr);
/* note : canonical representation of XXH3 is the same as XXH64
* since they both produce XXH64_hash_t values */
/*-**********************************************************************
* XXH3 128-bit variant
************************************************************************/
/*!
* @brief The return value from 128-bit hashes.
*
* Stored in little endian order, although the fields themselves are in native
* endianness.
*/
typedef struct {
XXH64_hash_t low64; /*!< `value & 0xFFFFFFFFFFFFFFFF` */
XXH64_hash_t high64; /*!< `value >> 64` */
} XXH128_hash_t;
XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* data, size_t len);
XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
/******* Streaming *******/
/*
* Streaming requires state maintenance.
* This operation costs memory and CPU.
* As a consequence, streaming is slower than one-shot hashing.
* For better performance, prefer one-shot functions whenever applicable.
*
* XXH3_128bits uses the same XXH3_state_t as XXH3_64bits().
* Use already declared XXH3_createState() and XXH3_freeState().
*
* All reset and streaming functions have same meaning as their 64-bit counterpart.
*/
XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH3_state_t* statePtr);
XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* statePtr);
/* Following helper functions make it possible to compare XXH128_hast_t values.
* Since XXH128_hash_t is a structure, this capability is not offered by the language.
* Note: For better performance, these functions can be inlined using XXH_INLINE_ALL */
/*!
* XXH128_isEqual():
* Return: 1 if `h1` and `h2` are equal, 0 if they are not.
*/
XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);
/*!
* XXH128_cmp():
*
* This comparator is compatible with stdlib's `qsort()`/`bsearch()`.
*
* return: >0 if *h128_1 > *h128_2
* =0 if *h128_1 == *h128_2
* <0 if *h128_1 < *h128_2
*/
XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2);
/******* Canonical representation *******/
typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t;
XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash);
XXH_PUBLIC_API XXH128_hash_t XXH128_hashFromCanonical(const XXH128_canonical_t* src);
#endif /* !XXH_NO_XXH3 */
#endif /* XXH_NO_LONG_LONG */
/*!
* @}
*/
#endif /* XXHASH_H_5627135585666179 */
#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742)
#define XXHASH_H_STATIC_13879238742
/* ****************************************************************************
* This section contains declarations which are not guaranteed to remain stable.
* They may change in future versions, becoming incompatible with a different
* version of the library.
* These declarations should only be used with static linking.
* Never use them in association with dynamic linking!
***************************************************************************** */
/*
* These definitions are only present to allow static allocation
* of XXH states, on stack or in a struct, for example.
* Never **ever** access their members directly.
*/
/*!
* @internal
* @brief Structure for XXH32 streaming API.
*
* @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
* @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
* an opaque type. This allows fields to safely be changed.
*
* Typedef'd to @ref XXH32_state_t.
* Do not access the members of this struct directly.
* @see XXH64_state_s, XXH3_state_s
*/
struct XXH32_state_s {
XXH32_hash_t total_len_32; /*!< Total length hashed, modulo 2^32 */
XXH32_hash_t large_len; /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */
XXH32_hash_t v[4]; /*!< Accumulator lanes */
XXH32_hash_t mem32[4]; /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
XXH32_hash_t memsize; /*!< Amount of data in @ref mem32 */
XXH32_hash_t reserved; /*!< Reserved field. Do not read nor write to it. */
}; /* typedef'd to XXH32_state_t */
#ifndef XXH_NO_LONG_LONG /* defined when there is no 64-bit support */
/*!
* @internal
* @brief Structure for XXH64 streaming API.
*
* @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
* @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
* an opaque type. This allows fields to safely be changed.
*
* Typedef'd to @ref XXH64_state_t.
* Do not access the members of this struct directly.
* @see XXH32_state_s, XXH3_state_s
*/
struct XXH64_state_s {
XXH64_hash_t total_len; /*!< Total length hashed. This is always 64-bit. */
XXH64_hash_t v[4]; /*!< Accumulator lanes */
XXH64_hash_t mem64[4]; /*!< Internal buffer for partial reads. Treated as unsigned char[32]. */
XXH32_hash_t memsize; /*!< Amount of data in @ref mem64 */
XXH32_hash_t reserved32; /*!< Reserved field, needed for padding anyways*/
XXH64_hash_t reserved64; /*!< Reserved field. Do not read or write to it. */
}; /* typedef'd to XXH64_state_t */
#ifndef XXH_NO_XXH3
#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */
# include <stdalign.h>
# define XXH_ALIGN(n) alignas(n)
#elif defined(__cplusplus) && (__cplusplus >= 201103L) /* >= C++11 */
/* In C++ alignas() is a keyword */
# define XXH_ALIGN(n) alignas(n)
#elif defined(__GNUC__)
# define XXH_ALIGN(n) __attribute__ ((aligned(n)))
#elif defined(_MSC_VER)
# define XXH_ALIGN(n) __declspec(align(n))
#else
# define XXH_ALIGN(n) /* disabled */
#endif
/* Old GCC versions only accept the attribute after the type in structures. */
#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)) /* C11+ */ \
&& ! (defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \
&& defined(__GNUC__)
# define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align)
#else
# define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type
#endif
/*!
* @brief The size of the internal XXH3 buffer.
*
* This is the optimal update size for incremental hashing.
*
* @see XXH3_64b_update(), XXH3_128b_update().
*/
#define XXH3_INTERNALBUFFER_SIZE 256
/*!
* @brief Default size of the secret buffer (and @ref XXH3_kSecret).
*
* This is the size used in @ref XXH3_kSecret and the seeded functions.
*
* Not to be confused with @ref XXH3_SECRET_SIZE_MIN.
*/
#define XXH3_SECRET_DEFAULT_SIZE 192
/*!
* @internal
* @brief Structure for XXH3 streaming API.
*
* @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
* @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined.
* Otherwise it is an opaque type.
* Never use this definition in combination with dynamic library.
* This allows fields to safely be changed in the future.
*
* @note ** This structure has a strict alignment requirement of 64 bytes!! **
* Do not allocate this with `malloc()` or `new`,
* it will not be sufficiently aligned.
* Use @ref XXH3_createState() and @ref XXH3_freeState(), or stack allocation.
*
* Typedef'd to @ref XXH3_state_t.
* Do never access the members of this struct directly.
*
* @see XXH3_INITSTATE() for stack initialization.
* @see XXH3_createState(), XXH3_freeState().
* @see XXH32_state_s, XXH64_state_s
*/
struct XXH3_state_s {
XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
/*!< The 8 accumulators. Similar to `vN` in @ref XXH32_state_s::v1 and @ref XXH64_state_s */
XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
/*!< Used to store a custom secret generated from a seed. */
XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
/*!< The internal buffer. @see XXH32_state_s::mem32 */
XXH32_hash_t bufferedSize;
/*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */
XXH32_hash_t useSeed;
/*!< Reserved field. Needed for padding on 64-bit. */
size_t nbStripesSoFar;
/*!< Number or stripes processed. */
XXH64_hash_t totalLen;
/*!< Total length hashed. 64-bit even on 32-bit targets. */
size_t nbStripesPerBlock;
/*!< Number of stripes per block. */
size_t secretLimit;
/*!< Size of @ref customSecret or @ref extSecret */
XXH64_hash_t seed;
/*!< Seed for _withSeed variants. Must be zero otherwise, @see XXH3_INITSTATE() */
XXH64_hash_t reserved64;
/*!< Reserved field. */
const unsigned char* extSecret;
/*!< Reference to an external secret for the _withSecret variants, NULL
* for other variants. */
/* note: there may be some padding at the end due to alignment on 64 bytes */
}; /* typedef'd to XXH3_state_t */
#undef XXH_ALIGN_MEMBER
/*!
* @brief Initializes a stack-allocated `XXH3_state_s`.
*
* When the @ref XXH3_state_t structure is merely emplaced on stack,
* it should be initialized with XXH3_INITSTATE() or a memset()
* in case its first reset uses XXH3_NNbits_reset_withSeed().
* This init can be omitted if the first reset uses default or _withSecret mode.
* This operation isn't necessary when the state is created with XXH3_createState().
* Note that this doesn't prepare the state for a streaming operation,
* it's still necessary to use XXH3_NNbits_reset*() afterwards.
*/
#define XXH3_INITSTATE(XXH3_state_ptr) { (XXH3_state_ptr)->seed = 0; }
/* XXH128() :
* simple alias to pre-selected XXH3_128bits variant
*/
XXH_PUBLIC_API XXH128_hash_t XXH128(const void* data, size_t len, XXH64_hash_t seed);
/* === Experimental API === */
/* Symbols defined below must be considered tied to a specific library version. */
/*
* XXH3_generateSecret():
*
* Derive a high-entropy secret from any user-defined content, named customSeed.
* The generated secret can be used in combination with `*_withSecret()` functions.
* The `_withSecret()` variants are useful to provide a higher level of protection than 64-bit seed,
* as it becomes much more difficult for an external actor to guess how to impact the calculation logic.
*
* The function accepts as input a custom seed of any length and any content,
* and derives from it a high-entropy secret of length @secretSize
* into an already allocated buffer @secretBuffer.
* @secretSize must be >= XXH3_SECRET_SIZE_MIN
*
* The generated secret can then be used with any `*_withSecret()` variant.
* Functions `XXH3_128bits_withSecret()`, `XXH3_64bits_withSecret()`,
* `XXH3_128bits_reset_withSecret()` and `XXH3_64bits_reset_withSecret()`
* are part of this list. They all accept a `secret` parameter
* which must be large enough for implementation reasons (>= XXH3_SECRET_SIZE_MIN)
* _and_ feature very high entropy (consist of random-looking bytes).
* These conditions can be a high bar to meet, so
* XXH3_generateSecret() can be employed to ensure proper quality.
*
* customSeed can be anything. It can have any size, even small ones,
* and its content can be anything, even "poor entropy" sources such as a bunch of zeroes.
* The resulting `secret` will nonetheless provide all required qualities.
*
* When customSeedSize > 0, supplying NULL as customSeed is undefined behavior.
*/
XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize);
/*
* XXH3_generateSecret_fromSeed():
*
* Generate the same secret as the _withSeed() variants.
*
* The resulting secret has a length of XXH3_SECRET_DEFAULT_SIZE (necessarily).
* @secretBuffer must be already allocated, of size at least XXH3_SECRET_DEFAULT_SIZE bytes.
*
* The generated secret can be used in combination with
*`*_withSecret()` and `_withSecretandSeed()` variants.
* This generator is notably useful in combination with `_withSecretandSeed()`,
* as a way to emulate a faster `_withSeed()` variant.
*/
XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed);
/*
* *_withSecretandSeed() :
* These variants generate hash values using either
* @seed for "short" keys (< XXH3_MIDSIZE_MAX = 240 bytes)
* or @secret for "large" keys (>= XXH3_MIDSIZE_MAX).
*
* This generally benefits speed, compared to `_withSeed()` or `_withSecret()`.
* `_withSeed()` has to generate the secret on the fly for "large" keys.
* It's fast, but can be perceptible for "not so large" keys (< 1 KB).
* `_withSecret()` has to generate the masks on the fly for "small" keys,
* which requires more instructions than _withSeed() variants.
* Therefore, _withSecretandSeed variant combines the best of both worlds.
*
* When @secret has been generated by XXH3_generateSecret_fromSeed(),
* this variant produces *exactly* the same results as `_withSeed()` variant,
* hence offering only a pure speed benefit on "large" input,
* by skipping the need to regenerate the secret for every large input.
*
* Another usage scenario is to hash the secret to a 64-bit hash value,
* for example with XXH3_64bits(), which then becomes the seed,
* and then employ both the seed and the secret in _withSecretandSeed().
* On top of speed, an added benefit is that each bit in the secret
* has a 50% chance to swap each bit in the output,
* via its impact to the seed.
* This is not guaranteed when using the secret directly in "small data" scenarios,
* because only portions of the secret are employed for small data.
*/
XXH_PUBLIC_API XXH64_hash_t
XXH3_64bits_withSecretandSeed(const void* data, size_t len,
const void* secret, size_t secretSize,
XXH64_hash_t seed);
XXH_PUBLIC_API XXH128_hash_t
XXH3_128bits_withSecretandSeed(const void* data, size_t len,
const void* secret, size_t secretSize,
XXH64_hash_t seed64);
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
const void* secret, size_t secretSize,
XXH64_hash_t seed64);
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
const void* secret, size_t secretSize,
XXH64_hash_t seed64);
#endif /* XXH_NO_XXH3 */
#endif /* XXH_NO_LONG_LONG */
#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
# define XXH_IMPLEMENTATION
#endif
#endif /* defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) */
/* ======================================================================== */
/* ======================================================================== */
/* ======================================================================== */
/*-**********************************************************************
* xxHash implementation
*-**********************************************************************
* xxHash's implementation used to be hosted inside xxhash.c.
*
* However, inlining requires implementation to be visible to the compiler,
* hence be included alongside the header.
* Previously, implementation was hosted inside xxhash.c,
* which was then #included when inlining was activated.
* This construction created issues with a few build and install systems,
* as it required xxhash.c to be stored in /include directory.
*
* xxHash implementation is now directly integrated within xxhash.h.
* As a consequence, xxhash.c is no longer needed in /include.
*
* xxhash.c is still available and is still useful.
* In a "normal" setup, when xxhash is not inlined,
* xxhash.h only exposes the prototypes and public symbols,
* while xxhash.c can be built into an object file xxhash.o
* which can then be linked into the final binary.
************************************************************************/
#if ( defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) \
|| defined(XXH_IMPLEMENTATION) ) && !defined(XXH_IMPLEM_13a8737387)
# define XXH_IMPLEM_13a8737387
/* *************************************
* Tuning parameters
***************************************/
/*!
* @defgroup tuning Tuning parameters
* @{
*
* Various macros to control xxHash's behavior.
*/
#ifdef XXH_DOXYGEN
/*!
* @brief Define this to disable 64-bit code.
*
* Useful if only using the @ref xxh32_family and you have a strict C90 compiler.
*/
# define XXH_NO_LONG_LONG
# undef XXH_NO_LONG_LONG /* don't actually */
/*!
* @brief Controls how unaligned memory is accessed.
*
* By default, access to unaligned memory is controlled by `memcpy()`, which is
* safe and portable.
*
* Unfortunately, on some target/compiler combinations, the generated assembly
* is sub-optimal.
*
* The below switch allow selection of a different access method
* in the search for improved performance.
*
* @par Possible options:
*
* - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy`
* @par
* Use `memcpy()`. Safe and portable. Note that most modern compilers will
* eliminate the function call and treat it as an unaligned access.
*
* - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((packed))`
* @par
* Depends on compiler extensions and is therefore not portable.
* This method is safe _if_ your compiler supports it,
* and *generally* as fast or faster than `memcpy`.
*
* - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast
* @par
* Casts directly and dereferences. This method doesn't depend on the
* compiler, but it violates the C standard as it directly dereferences an
* unaligned pointer. It can generate buggy code on targets which do not
* support unaligned memory accesses, but in some circumstances, it's the
* only known way to get the most performance.
*
* - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift
* @par
* Also portable. This can generate the best code on old compilers which don't
* inline small `memcpy()` calls, and it might also be faster on big-endian
* systems which lack a native byteswap instruction. However, some compilers
* will emit literal byteshifts even if the target supports unaligned access.
* .
*
* @warning
* Methods 1 and 2 rely on implementation-defined behavior. Use these with
* care, as what works on one compiler/platform/optimization level may cause
* another to read garbage data or even crash.
*
* See https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details.
*
* Prefer these methods in priority order (0 > 3 > 1 > 2)
*/
# define XXH_FORCE_MEMORY_ACCESS 0
/*!
* @def XXH_FORCE_ALIGN_CHECK
* @brief If defined to non-zero, adds a special path for aligned inputs (XXH32()
* and XXH64() only).
*
* This is an important performance trick for architectures without decent
* unaligned memory access performance.
*
* It checks for input alignment, and when conditions are met, uses a "fast
* path" employing direct 32-bit/64-bit reads, resulting in _dramatically
* faster_ read speed.
*
* The check costs one initial branch per hash, which is generally negligible,
* but not zero.
*
* Moreover, it's not useful to generate an additional code path if memory
* access uses the same instruction for both aligned and unaligned
* addresses (e.g. x86 and aarch64).
*
* In these cases, the alignment check can be removed by setting this macro to 0.
* Then the code will always use unaligned memory access.
* Align check is automatically disabled on x86, x64 & arm64,
* which are platforms known to offer good unaligned memory accesses performance.
*
* This option does not affect XXH3 (only XXH32 and XXH64).
*/
# define XXH_FORCE_ALIGN_CHECK 0
/*!
* @def XXH_NO_INLINE_HINTS
* @brief When non-zero, sets all functions to `static`.
*
* By default, xxHash tries to force the compiler to inline almost all internal
* functions.
*
* This can usually improve performance due to reduced jumping and improved
* constant folding, but significantly increases the size of the binary which
* might not be favorable.
*
* Additionally, sometimes the forced inlining can be detrimental to performance,
* depending on the architecture.
*
* XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
* compiler full control on whether to inline or not.
*
* When not optimizing (-O0), optimizing for size (-Os, -Oz), or using
* -fno-inline with GCC or Clang, this will automatically be defined.
*/
# define XXH_NO_INLINE_HINTS 0
/*!
* @def XXH32_ENDJMP
* @brief Whether to use a jump for `XXH32_finalize`.
*
* For performance, `XXH32_finalize` uses multiple branches in the finalizer.
* This is generally preferable for performance,
* but depending on exact architecture, a jmp may be preferable.
*
* This setting is only possibly making a difference for very small inputs.
*/
# define XXH32_ENDJMP 0
/*!
* @internal
* @brief Redefines old internal names.
*
* For compatibility with code that uses xxHash's internals before the names
* were changed to improve namespacing. There is no other reason to use this.
*/
# define XXH_OLD_NAMES
# undef XXH_OLD_NAMES /* don't actually use, it is ugly. */
#endif /* XXH_DOXYGEN */
/*!
* @}
*/
#ifndef XXH_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */
/* prefer __packed__ structures (method 1) for gcc on armv7+ and mips */
# if !defined(__clang__) && \
( \
(defined(__INTEL_COMPILER) && !defined(_WIN32)) || \
( \
defined(__GNUC__) && ( \
(defined(__ARM_ARCH) && __ARM_ARCH >= 7) || \
( \
defined(__mips__) && \
(__mips <= 5 || __mips_isa_rev < 6) && \
(!defined(__mips16) || defined(__mips_mips16e2)) \
) \
) \
) \
)
# define XXH_FORCE_MEMORY_ACCESS 1
# endif
#endif
#ifndef XXH_FORCE_ALIGN_CHECK /* can be defined externally */
# if defined(__i386) || defined(__x86_64__) || defined(__aarch64__) \
|| defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM64) /* visual */
# define XXH_FORCE_ALIGN_CHECK 0
# else
# define XXH_FORCE_ALIGN_CHECK 1
# endif
#endif
#ifndef XXH_NO_INLINE_HINTS
# if defined(__OPTIMIZE_SIZE__) /* -Os, -Oz */ \
|| defined(__NO_INLINE__) /* -O0, -fno-inline */
# define XXH_NO_INLINE_HINTS 1
# else
# define XXH_NO_INLINE_HINTS 0
# endif
#endif
#ifndef XXH32_ENDJMP
/* generally preferable for performance */
# define XXH32_ENDJMP 0
#endif
/*!
* @defgroup impl Implementation
* @{
*/
/* *************************************
* Includes & Memory related functions
***************************************/
/* Modify the local functions below should you wish to use some other memory routines */
/* for ZSTD_malloc(), ZSTD_free() */
#define ZSTD_DEPS_NEED_MALLOC
/**** skipping file: zstd_deps.h ****/
static void* XXH_malloc(size_t s) { return ZSTD_malloc(s); }
static void XXH_free (void* p) { ZSTD_free(p); }
static void* XXH_memcpy(void* dest, const void* src, size_t size) { return ZSTD_memcpy(dest,src,size); }
/* *************************************
* Compiler Specific Options
***************************************/
#ifdef _MSC_VER /* Visual Studio warning fix */
# pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
#endif
#if XXH_NO_INLINE_HINTS /* disable inlining hints */
# if defined(__GNUC__) || defined(__clang__)
# define XXH_FORCE_INLINE static __attribute__((unused))
# else
# define XXH_FORCE_INLINE static
# endif
# define XXH_NO_INLINE static
/* enable inlining hints */
#elif defined(__GNUC__) || defined(__clang__)
# define XXH_FORCE_INLINE static __inline__ __attribute__((always_inline, unused))
# define XXH_NO_INLINE static __attribute__((noinline))
#elif defined(_MSC_VER) /* Visual Studio */
# define XXH_FORCE_INLINE static __forceinline
# define XXH_NO_INLINE static __declspec(noinline)
#elif defined (__cplusplus) \
|| (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)) /* C99 */
# define XXH_FORCE_INLINE static inline
# define XXH_NO_INLINE static
#else
# define XXH_FORCE_INLINE static
# define XXH_NO_INLINE static
#endif
/* *************************************
* Debug
***************************************/
/*!
* @ingroup tuning
* @def XXH_DEBUGLEVEL
* @brief Sets the debugging level.
*
* XXH_DEBUGLEVEL is expected to be defined externally, typically via the
* compiler's command line options. The value must be a number.
*/
#ifndef XXH_DEBUGLEVEL
# ifdef DEBUGLEVEL /* backwards compat */
# define XXH_DEBUGLEVEL DEBUGLEVEL
# else
# define XXH_DEBUGLEVEL 0
# endif
#endif
#if (XXH_DEBUGLEVEL>=1)
# include <assert.h> /* note: can still be disabled with NDEBUG */
# define XXH_ASSERT(c) assert(c)
#else
# define XXH_ASSERT(c) ((void)0)
#endif
/* note: use after variable declarations */
#ifndef XXH_STATIC_ASSERT
# if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* C11 */
# include <assert.h>
# define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
# elif defined(__cplusplus) && (__cplusplus >= 201103L) /* C++11 */
# define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
# else
# define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { struct xxh_sa { char x[(c) ? 1 : -1]; }; } while(0)
# endif
# define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c),#c)
#endif
/*!
* @internal
* @def XXH_COMPILER_GUARD(var)
* @brief Used to prevent unwanted optimizations for @p var.
*
* It uses an empty GCC inline assembly statement with a register constraint
* which forces @p var into a general purpose register (e.g. eax, ebx, ecx
* on x86) and marks it as modified.
*
* This is used in a few places to avoid unwanted autovectorization (e.g.
* XXH32_round()). All vectorization we want is explicit via intrinsics,
* and _usually_ isn't wanted elsewhere.
*
* We also use it to prevent unwanted constant folding for AArch64 in
* XXH3_initCustomSecret_scalar().
*/
#if defined(__GNUC__) || defined(__clang__)
# define XXH_COMPILER_GUARD(var) __asm__ __volatile__("" : "+r" (var))
#else
# define XXH_COMPILER_GUARD(var) ((void)0)
#endif
/* *************************************
* Basic Types
***************************************/
#if !defined (__VMS) \
&& (defined (__cplusplus) \
|| (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
# include <stdint.h>
typedef uint8_t xxh_u8;
#else
typedef unsigned char xxh_u8;
#endif
typedef XXH32_hash_t xxh_u32;
#ifdef XXH_OLD_NAMES
# define BYTE xxh_u8
# define U8 xxh_u8
# define U32 xxh_u32
#endif
/* *** Memory access *** */
/*!
* @internal
* @fn xxh_u32 XXH_read32(const void* ptr)
* @brief Reads an unaligned 32-bit integer from @p ptr in native endianness.
*
* Affected by @ref XXH_FORCE_MEMORY_ACCESS.
*
* @param ptr The pointer to read from.
* @return The 32-bit native endian integer from the bytes at @p ptr.
*/
/*!
* @internal
* @fn xxh_u32 XXH_readLE32(const void* ptr)
* @brief Reads an unaligned 32-bit little endian integer from @p ptr.
*
* Affected by @ref XXH_FORCE_MEMORY_ACCESS.
*
* @param ptr The pointer to read from.
* @return The 32-bit little endian integer from the bytes at @p ptr.
*/
/*!
* @internal
* @fn xxh_u32 XXH_readBE32(const void* ptr)
* @brief Reads an unaligned 32-bit big endian integer from @p ptr.
*
* Affected by @ref XXH_FORCE_MEMORY_ACCESS.
*
* @param ptr The pointer to read from.
* @return The 32-bit big endian integer from the bytes at @p ptr.
*/
/*!
* @internal
* @fn xxh_u32 XXH_readLE32_align(const void* ptr, XXH_alignment align)
* @brief Like @ref XXH_readLE32(), but has an option for aligned reads.
*
* Affected by @ref XXH_FORCE_MEMORY_ACCESS.
* Note that when @ref XXH_FORCE_ALIGN_CHECK == 0, the @p align parameter is
* always @ref XXH_alignment::XXH_unaligned.
*
* @param ptr The pointer to read from.
* @param align Whether @p ptr is aligned.
* @pre
* If @p align == @ref XXH_alignment::XXH_aligned, @p ptr must be 4 byte
* aligned.
* @return The 32-bit little endian integer from the bytes at @p ptr.
*/
#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
/*
* Manual byteshift. Best for old compilers which don't inline memcpy.
* We actually directly use XXH_readLE32 and XXH_readBE32.
*/
#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
/*
* Force direct memory access. Only works on CPU which support unaligned memory
* access in hardware.
*/
static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; }
#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
/*
* __pack instructions are safer but compiler specific, hence potentially
* problematic for some compilers.
*
* Currently only defined for GCC and ICC.
*/
#ifdef XXH_OLD_NAMES
typedef union { xxh_u32 u32; } __attribute__((packed)) unalign;
#endif
static xxh_u32 XXH_read32(const void* ptr)
{
typedef union { xxh_u32 u32; } __attribute__((packed)) xxh_unalign;
return ((const xxh_unalign*)ptr)->u32;
}
#else
/*
* Portable and safe solution. Generally efficient.
* see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
*/
static xxh_u32 XXH_read32(const void* memPtr)
{
xxh_u32 val;
XXH_memcpy(&val, memPtr, sizeof(val));
return val;
}
#endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
/* *** Endianness *** */
/*!
* @ingroup tuning
* @def XXH_CPU_LITTLE_ENDIAN
* @brief Whether the target is little endian.
*
* Defined to 1 if the target is little endian, or 0 if it is big endian.
* It can be defined externally, for example on the compiler command line.
*
* If it is not defined,
* a runtime check (which is usually constant folded) is used instead.
*
* @note
* This is not necessarily defined to an integer constant.
*
* @see XXH_isLittleEndian() for the runtime check.
*/
#ifndef XXH_CPU_LITTLE_ENDIAN
/*
* Try to detect endianness automatically, to avoid the nonstandard behavior
* in `XXH_isLittleEndian()`
*/
# if defined(_WIN32) /* Windows is always little endian */ \
|| defined(__LITTLE_ENDIAN__) \
|| (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
# define XXH_CPU_LITTLE_ENDIAN 1
# elif defined(__BIG_ENDIAN__) \
|| (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
# define XXH_CPU_LITTLE_ENDIAN 0
# else
/*!
* @internal
* @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN.
*
* Most compilers will constant fold this.
*/
static int XXH_isLittleEndian(void)
{
/*
* Portable and well-defined behavior.
* Don't use static: it is detrimental to performance.
*/
const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 };
return one.c[0];
}
# define XXH_CPU_LITTLE_ENDIAN XXH_isLittleEndian()
# endif
#endif
/* ****************************************
* Compiler-specific Functions and Macros
******************************************/
#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
#ifdef __has_builtin
# define XXH_HAS_BUILTIN(x) __has_builtin(x)
#else
# define XXH_HAS_BUILTIN(x) 0
#endif
/*!
* @internal
* @def XXH_rotl32(x,r)
* @brief 32-bit rotate left.
*
* @param x The 32-bit integer to be rotated.
* @param r The number of bits to rotate.
* @pre
* @p r > 0 && @p r < 32
* @note
* @p x and @p r may be evaluated multiple times.
* @return The rotated result.
*/
#if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) \
&& XXH_HAS_BUILTIN(__builtin_rotateleft64)
# define XXH_rotl32 __builtin_rotateleft32
# define XXH_rotl64 __builtin_rotateleft64
/* Note: although _rotl exists for minGW (GCC under windows), performance seems poor */
#elif defined(_MSC_VER)
# define XXH_rotl32(x,r) _rotl(x,r)
# define XXH_rotl64(x,r) _rotl64(x,r)
#else
# define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
# define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r))))
#endif
/*!
* @internal
* @fn xxh_u32 XXH_swap32(xxh_u32 x)
* @brief A 32-bit byteswap.
*
* @param x The 32-bit integer to byteswap.
* @return @p x, byteswapped.
*/
#if defined(_MSC_VER) /* Visual Studio */
# define XXH_swap32 _byteswap_ulong
#elif XXH_GCC_VERSION >= 403
# define XXH_swap32 __builtin_bswap32
#else
static xxh_u32 XXH_swap32 (xxh_u32 x)
{
return ((x << 24) & 0xff000000 ) |
((x << 8) & 0x00ff0000 ) |
((x >> 8) & 0x0000ff00 ) |
((x >> 24) & 0x000000ff );
}
#endif
/* ***************************
* Memory reads
*****************************/
/*!
* @internal
* @brief Enum to indicate whether a pointer is aligned.
*/
typedef enum {
XXH_aligned, /*!< Aligned */
XXH_unaligned /*!< Possibly unaligned */
} XXH_alignment;
/*
* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load.
*
* This is ideal for older compilers which don't inline memcpy.
*/
#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* memPtr)
{
const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
return bytePtr[0]
| ((xxh_u32)bytePtr[1] << 8)
| ((xxh_u32)bytePtr[2] << 16)
| ((xxh_u32)bytePtr[3] << 24);
}
XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void* memPtr)
{
const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
return bytePtr[3]
| ((xxh_u32)bytePtr[2] << 8)
| ((xxh_u32)bytePtr[1] << 16)
| ((xxh_u32)bytePtr[0] << 24);
}
#else
XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr)
{
return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
}
static xxh_u32 XXH_readBE32(const void* ptr)
{
return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
}
#endif
XXH_FORCE_INLINE xxh_u32
XXH_readLE32_align(const void* ptr, XXH_alignment align)
{
if (align==XXH_unaligned) {
return XXH_readLE32(ptr);
} else {
return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr);
}
}
/* *************************************
* Misc
***************************************/
/*! @ingroup public */
XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
/* *******************************************************************
* 32-bit hash functions
*********************************************************************/
/*!
* @}
* @defgroup xxh32_impl XXH32 implementation
* @ingroup impl
* @{
*/
/* #define instead of static const, to be used as initializers */
#define XXH_PRIME32_1 0x9E3779B1U /*!< 0b10011110001101110111100110110001 */
#define XXH_PRIME32_2 0x85EBCA77U /*!< 0b10000101111010111100101001110111 */
#define XXH_PRIME32_3 0xC2B2AE3DU /*!< 0b11000010101100101010111000111101 */
#define XXH_PRIME32_4 0x27D4EB2FU /*!< 0b00100111110101001110101100101111 */
#define XXH_PRIME32_5 0x165667B1U /*!< 0b00010110010101100110011110110001 */
#ifdef XXH_OLD_NAMES
# define PRIME32_1 XXH_PRIME32_1
# define PRIME32_2 XXH_PRIME32_2
# define PRIME32_3 XXH_PRIME32_3
# define PRIME32_4 XXH_PRIME32_4
# define PRIME32_5 XXH_PRIME32_5
#endif
/*!
* @internal
* @brief Normal stripe processing routine.
*
* This shuffles the bits so that any bit from @p input impacts several bits in
* @p acc.
*
* @param acc The accumulator lane.
* @param input The stripe of input to mix.
* @return The mixed accumulator lane.
*/
static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
{
acc += input * XXH_PRIME32_2;
acc = XXH_rotl32(acc, 13);
acc *= XXH_PRIME32_1;
#if (defined(__SSE4_1__) || defined(__aarch64__)) && !defined(XXH_ENABLE_AUTOVECTORIZE)
/*
* UGLY HACK:
* A compiler fence is the only thing that prevents GCC and Clang from
* autovectorizing the XXH32 loop (pragmas and attributes don't work for some
* reason) without globally disabling SSE4.1.
*
* The reason we want to avoid vectorization is because despite working on
* 4 integers at a time, there are multiple factors slowing XXH32 down on
* SSE4:
* - There's a ridiculous amount of lag from pmulld (10 cycles of latency on
* newer chips!) making it slightly slower to multiply four integers at
* once compared to four integers independently. Even when pmulld was
* fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE
* just to multiply unless doing a long operation.
*
* - Four instructions are required to rotate,
* movqda tmp, v // not required with VEX encoding
* pslld tmp, 13 // tmp <<= 13
* psrld v, 19 // x >>= 19
* por v, tmp // x |= tmp
* compared to one for scalar:
* roll v, 13 // reliably fast across the board
* shldl v, v, 13 // Sandy Bridge and later prefer this for some reason
*
* - Instruction level parallelism is actually more beneficial here because
* the SIMD actually serializes this operation: While v1 is rotating, v2
* can load data, while v3 can multiply. SSE forces them to operate
* together.
*
* This is also enabled on AArch64, as Clang autovectorizes it incorrectly
* and it is pointless writing a NEON implementation that is basically the
* same speed as scalar for XXH32.
*/
XXH_COMPILER_GUARD(acc);
#endif
return acc;
}
/*!
* @internal
* @brief Mixes all bits to finalize the hash.
*
* The final mix ensures that all input bits have a chance to impact any bit in
* the output digest, resulting in an unbiased distribution.
*
* @param h32 The hash to avalanche.
* @return The avalanched hash.
*/
static xxh_u32 XXH32_avalanche(xxh_u32 h32)
{
h32 ^= h32 >> 15;
h32 *= XXH_PRIME32_2;
h32 ^= h32 >> 13;
h32 *= XXH_PRIME32_3;
h32 ^= h32 >> 16;
return(h32);
}
#define XXH_get32bits(p) XXH_readLE32_align(p, align)
/*!
* @internal
* @brief Processes the last 0-15 bytes of @p ptr.
*
* There may be up to 15 bytes remaining to consume from the input.
* This final stage will digest them to ensure that all input bytes are present
* in the final mix.
*
* @param h32 The hash to finalize.
* @param ptr The pointer to the remaining input.
* @param len The remaining length, modulo 16.
* @param align Whether @p ptr is aligned.
* @return The finalized hash.
*/
static xxh_u32
XXH32_finalize(xxh_u32 h32, const xxh_u8* ptr, size_t len, XXH_alignment align)
{
#define XXH_PROCESS1 do { \
h32 += (*ptr++) * XXH_PRIME32_5; \
h32 = XXH_rotl32(h32, 11) * XXH_PRIME32_1; \
} while (0)
#define XXH_PROCESS4 do { \
h32 += XXH_get32bits(ptr) * XXH_PRIME32_3; \
ptr += 4; \
h32 = XXH_rotl32(h32, 17) * XXH_PRIME32_4; \
} while (0)
if (ptr==NULL) XXH_ASSERT(len == 0);
/* Compact rerolled version; generally faster */
if (!XXH32_ENDJMP) {
len &= 15;
while (len >= 4) {
XXH_PROCESS4;
len -= 4;
}
while (len > 0) {
XXH_PROCESS1;
--len;
}
return XXH32_avalanche(h32);
} else {
switch(len&15) /* or switch(bEnd - p) */ {
case 12: XXH_PROCESS4;
XXH_FALLTHROUGH;
case 8: XXH_PROCESS4;
XXH_FALLTHROUGH;
case 4: XXH_PROCESS4;
return XXH32_avalanche(h32);
case 13: XXH_PROCESS4;
XXH_FALLTHROUGH;
case 9: XXH_PROCESS4;
XXH_FALLTHROUGH;
case 5: XXH_PROCESS4;
XXH_PROCESS1;
return XXH32_avalanche(h32);
case 14: XXH_PROCESS4;
XXH_FALLTHROUGH;
case 10: XXH_PROCESS4;
XXH_FALLTHROUGH;
case 6: XXH_PROCESS4;
XXH_PROCESS1;
XXH_PROCESS1;
return XXH32_avalanche(h32);
case 15: XXH_PROCESS4;
XXH_FALLTHROUGH;
case 11: XXH_PROCESS4;
XXH_FALLTHROUGH;
case 7: XXH_PROCESS4;
XXH_FALLTHROUGH;
case 3: XXH_PROCESS1;
XXH_FALLTHROUGH;
case 2: XXH_PROCESS1;
XXH_FALLTHROUGH;
case 1: XXH_PROCESS1;
XXH_FALLTHROUGH;
case 0: return XXH32_avalanche(h32);
}
XXH_ASSERT(0);
return h32; /* reaching this point is deemed impossible */
}
}
#ifdef XXH_OLD_NAMES
# define PROCESS1 XXH_PROCESS1
# define PROCESS4 XXH_PROCESS4
#else
# undef XXH_PROCESS1
# undef XXH_PROCESS4
#endif
/*!
* @internal
* @brief The implementation for @ref XXH32().
*
* @param input , len , seed Directly passed from @ref XXH32().
* @param align Whether @p input is aligned.
* @return The calculated hash.
*/
XXH_FORCE_INLINE xxh_u32
XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align)
{
xxh_u32 h32;
if (input==NULL) XXH_ASSERT(len == 0);
if (len>=16) {
const xxh_u8* const bEnd = input + len;
const xxh_u8* const limit = bEnd - 15;
xxh_u32 v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
xxh_u32 v2 = seed + XXH_PRIME32_2;
xxh_u32 v3 = seed + 0;
xxh_u32 v4 = seed - XXH_PRIME32_1;
do {
v1 = XXH32_round(v1, XXH_get32bits(input)); input += 4;
v2 = XXH32_round(v2, XXH_get32bits(input)); input += 4;
v3 = XXH32_round(v3, XXH_get32bits(input)); input += 4;
v4 = XXH32_round(v4, XXH_get32bits(input)); input += 4;
} while (input < limit);
h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7)
+ XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
} else {
h32 = seed + XXH_PRIME32_5;
}
h32 += (xxh_u32)len;
return XXH32_finalize(h32, input, len&15, align);
}
/*! @ingroup xxh32_family */
XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed)
{
#if 0
/* Simple version, good for code maintenance, but unfortunately slow for small inputs */
XXH32_state_t state;
XXH32_reset(&state, seed);
XXH32_update(&state, (const xxh_u8*)input, len);
return XXH32_digest(&state);
#else
if (XXH_FORCE_ALIGN_CHECK) {
if ((((size_t)input) & 3) == 0) { /* Input is 4-bytes aligned, leverage the speed benefit */
return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
} }
return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
#endif
}
/******* Hash streaming *******/
/*!
* @ingroup xxh32_family
*/
XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
{
return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
}
/*! @ingroup xxh32_family */
XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
{
XXH_free(statePtr);
return XXH_OK;
}
/*! @ingroup xxh32_family */
XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
{
XXH_memcpy(dstState, srcState, sizeof(*dstState));
}
/*! @ingroup xxh32_family */
XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed)
{
XXH_ASSERT(statePtr != NULL);
memset(statePtr, 0, sizeof(*statePtr));
statePtr->v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
statePtr->v[1] = seed + XXH_PRIME32_2;
statePtr->v[2] = seed + 0;
statePtr->v[3] = seed - XXH_PRIME32_1;
return XXH_OK;
}
/*! @ingroup xxh32_family */
XXH_PUBLIC_API XXH_errorcode
XXH32_update(XXH32_state_t* state, const void* input, size_t len)
{
if (input==NULL) {
XXH_ASSERT(len == 0);
return XXH_OK;
}
{ const xxh_u8* p = (const xxh_u8*)input;
const xxh_u8* const bEnd = p + len;
state->total_len_32 += (XXH32_hash_t)len;
state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16));
if (state->memsize + len < 16) { /* fill in tmp buffer */
XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, len);
state->memsize += (XXH32_hash_t)len;
return XXH_OK;
}
if (state->memsize) { /* some data left from previous update */
XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, 16-state->memsize);
{ const xxh_u32* p32 = state->mem32;
state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p32)); p32++;
state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p32)); p32++;
state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p32)); p32++;
state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p32));
}
p += 16-state->memsize;
state->memsize = 0;
}
if (p <= bEnd-16) {
const xxh_u8* const limit = bEnd - 16;
do {
state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p)); p+=4;
state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p)); p+=4;
state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p)); p+=4;
state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p)); p+=4;
} while (p<=limit);
}
if (p < bEnd) {
XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
state->memsize = (unsigned)(bEnd-p);
}
}
return XXH_OK;
}
/*! @ingroup xxh32_family */
XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state)
{
xxh_u32 h32;
if (state->large_len) {
h32 = XXH_rotl32(state->v[0], 1)
+ XXH_rotl32(state->v[1], 7)
+ XXH_rotl32(state->v[2], 12)
+ XXH_rotl32(state->v[3], 18);
} else {
h32 = state->v[2] /* == seed */ + XXH_PRIME32_5;
}
h32 += state->total_len_32;
return XXH32_finalize(h32, (const xxh_u8*)state->mem32, state->memsize, XXH_aligned);
}
/******* Canonical representation *******/
/*!
* @ingroup xxh32_family
* The default return values from XXH functions are unsigned 32 and 64 bit
* integers.
*
* The canonical representation uses big endian convention, the same convention
* as human-readable numbers (large digits first).
*
* This way, hash values can be written into a file or buffer, remaining
* comparable across different systems.
*
* The following functions allow transformation of hash values to and from their
* canonical format.
*/
XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
{
/* XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t)); */
if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
XXH_memcpy(dst, &hash, sizeof(*dst));
}
/*! @ingroup xxh32_family */
XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
{
return XXH_readBE32(src);
}
#ifndef XXH_NO_LONG_LONG
/* *******************************************************************
* 64-bit hash functions
*********************************************************************/
/*!
* @}
* @ingroup impl
* @{
*/
/******* Memory access *******/
typedef XXH64_hash_t xxh_u64;
#ifdef XXH_OLD_NAMES
# define U64 xxh_u64
#endif
#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
/*
* Manual byteshift. Best for old compilers which don't inline memcpy.
* We actually directly use XXH_readLE64 and XXH_readBE64.
*/
#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
static xxh_u64 XXH_read64(const void* memPtr)
{
return *(const xxh_u64*) memPtr;
}
#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
/*
* __pack instructions are safer, but compiler specific, hence potentially
* problematic for some compilers.
*
* Currently only defined for GCC and ICC.
*/
#ifdef XXH_OLD_NAMES
typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) unalign64;
#endif
static xxh_u64 XXH_read64(const void* ptr)
{
typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) xxh_unalign64;
return ((const xxh_unalign64*)ptr)->u64;
}
#else
/*
* Portable and safe solution. Generally efficient.
* see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
*/
static xxh_u64 XXH_read64(const void* memPtr)
{
xxh_u64 val;
XXH_memcpy(&val, memPtr, sizeof(val));
return val;
}
#endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
#if defined(_MSC_VER) /* Visual Studio */
# define XXH_swap64 _byteswap_uint64
#elif XXH_GCC_VERSION >= 403
# define XXH_swap64 __builtin_bswap64
#else
static xxh_u64 XXH_swap64(xxh_u64 x)
{
return ((x << 56) & 0xff00000000000000ULL) |
((x << 40) & 0x00ff000000000000ULL) |
((x << 24) & 0x0000ff0000000000ULL) |
((x << 8) & 0x000000ff00000000ULL) |
((x >> 8) & 0x00000000ff000000ULL) |
((x >> 24) & 0x0000000000ff0000ULL) |
((x >> 40) & 0x000000000000ff00ULL) |
((x >> 56) & 0x00000000000000ffULL);
}
#endif
/* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */
#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* memPtr)
{
const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
return bytePtr[0]
| ((xxh_u64)bytePtr[1] << 8)
| ((xxh_u64)bytePtr[2] << 16)
| ((xxh_u64)bytePtr[3] << 24)
| ((xxh_u64)bytePtr[4] << 32)
| ((xxh_u64)bytePtr[5] << 40)
| ((xxh_u64)bytePtr[6] << 48)
| ((xxh_u64)bytePtr[7] << 56);
}
XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void* memPtr)
{
const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
return bytePtr[7]
| ((xxh_u64)bytePtr[6] << 8)
| ((xxh_u64)bytePtr[5] << 16)
| ((xxh_u64)bytePtr[4] << 24)
| ((xxh_u64)bytePtr[3] << 32)
| ((xxh_u64)bytePtr[2] << 40)
| ((xxh_u64)bytePtr[1] << 48)
| ((xxh_u64)bytePtr[0] << 56);
}
#else
XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr)
{
return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
}
static xxh_u64 XXH_readBE64(const void* ptr)
{
return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
}
#endif
XXH_FORCE_INLINE xxh_u64
XXH_readLE64_align(const void* ptr, XXH_alignment align)
{
if (align==XXH_unaligned)
return XXH_readLE64(ptr);
else
return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr);
}
/******* xxh64 *******/
/*!
* @}
* @defgroup xxh64_impl XXH64 implementation
* @ingroup impl
* @{
*/
/* #define rather that static const, to be used as initializers */
#define XXH_PRIME64_1 0x9E3779B185EBCA87ULL /*!< 0b1001111000110111011110011011000110000101111010111100101010000111 */
#define XXH_PRIME64_2 0xC2B2AE3D27D4EB4FULL /*!< 0b1100001010110010101011100011110100100111110101001110101101001111 */
#define XXH_PRIME64_3 0x165667B19E3779F9ULL /*!< 0b0001011001010110011001111011000110011110001101110111100111111001 */
#define XXH_PRIME64_4 0x85EBCA77C2B2AE63ULL /*!< 0b1000010111101011110010100111011111000010101100101010111001100011 */
#define XXH_PRIME64_5 0x27D4EB2F165667C5ULL /*!< 0b0010011111010100111010110010111100010110010101100110011111000101 */
#ifdef XXH_OLD_NAMES
# define PRIME64_1 XXH_PRIME64_1
# define PRIME64_2 XXH_PRIME64_2
# define PRIME64_3 XXH_PRIME64_3
# define PRIME64_4 XXH_PRIME64_4
# define PRIME64_5 XXH_PRIME64_5
#endif
static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input)
{
acc += input * XXH_PRIME64_2;
acc = XXH_rotl64(acc, 31);
acc *= XXH_PRIME64_1;
return acc;
}
static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val)
{
val = XXH64_round(0, val);
acc ^= val;
acc = acc * XXH_PRIME64_1 + XXH_PRIME64_4;
return acc;
}
static xxh_u64 XXH64_avalanche(xxh_u64 h64)
{
h64 ^= h64 >> 33;
h64 *= XXH_PRIME64_2;
h64 ^= h64 >> 29;
h64 *= XXH_PRIME64_3;
h64 ^= h64 >> 32;
return h64;
}
#define XXH_get64bits(p) XXH_readLE64_align(p, align)
static xxh_u64
XXH64_finalize(xxh_u64 h64, const xxh_u8* ptr, size_t len, XXH_alignment align)
{
if (ptr==NULL) XXH_ASSERT(len == 0);
len &= 31;
while (len >= 8) {
xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
ptr += 8;
h64 ^= k1;
h64 = XXH_rotl64(h64,27) * XXH_PRIME64_1 + XXH_PRIME64_4;
len -= 8;
}
if (len >= 4) {
h64 ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
ptr += 4;
h64 = XXH_rotl64(h64, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
len -= 4;
}
while (len > 0) {
h64 ^= (*ptr++) * XXH_PRIME64_5;
h64 = XXH_rotl64(h64, 11) * XXH_PRIME64_1;
--len;
}
return XXH64_avalanche(h64);
}
#ifdef XXH_OLD_NAMES
# define PROCESS1_64 XXH_PROCESS1_64
# define PROCESS4_64 XXH_PROCESS4_64
# define PROCESS8_64 XXH_PROCESS8_64
#else
# undef XXH_PROCESS1_64
# undef XXH_PROCESS4_64
# undef XXH_PROCESS8_64
#endif
XXH_FORCE_INLINE xxh_u64
XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align)
{
xxh_u64 h64;
if (input==NULL) XXH_ASSERT(len == 0);
if (len>=32) {
const xxh_u8* const bEnd = input + len;
const xxh_u8* const limit = bEnd - 31;
xxh_u64 v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
xxh_u64 v2 = seed + XXH_PRIME64_2;
xxh_u64 v3 = seed + 0;
xxh_u64 v4 = seed - XXH_PRIME64_1;
do {
v1 = XXH64_round(v1, XXH_get64bits(input)); input+=8;
v2 = XXH64_round(v2, XXH_get64bits(input)); input+=8;
v3 = XXH64_round(v3, XXH_get64bits(input)); input+=8;
v4 = XXH64_round(v4, XXH_get64bits(input)); input+=8;
} while (input<limit);
h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
h64 = XXH64_mergeRound(h64, v1);
h64 = XXH64_mergeRound(h64, v2);
h64 = XXH64_mergeRound(h64, v3);
h64 = XXH64_mergeRound(h64, v4);
} else {
h64 = seed + XXH_PRIME64_5;
}
h64 += (xxh_u64) len;
return XXH64_finalize(h64, input, len, align);
}
/*! @ingroup xxh64_family */
XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t len, XXH64_hash_t seed)
{
#if 0
/* Simple version, good for code maintenance, but unfortunately slow for small inputs */
XXH64_state_t state;
XXH64_reset(&state, seed);
XXH64_update(&state, (const xxh_u8*)input, len);
return XXH64_digest(&state);
#else
if (XXH_FORCE_ALIGN_CHECK) {
if ((((size_t)input) & 7)==0) { /* Input is aligned, let's leverage the speed advantage */
return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
} }
return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
#endif
}
/******* Hash Streaming *******/
/*! @ingroup xxh64_family*/
XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
{
return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
}
/*! @ingroup xxh64_family */
XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
{
XXH_free(statePtr);
return XXH_OK;
}
/*! @ingroup xxh64_family */
XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dstState, const XXH64_state_t* srcState)
{
XXH_memcpy(dstState, srcState, sizeof(*dstState));
}
/*! @ingroup xxh64_family */
XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, XXH64_hash_t seed)
{
XXH_ASSERT(statePtr != NULL);
memset(statePtr, 0, sizeof(*statePtr));
statePtr->v[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
statePtr->v[1] = seed + XXH_PRIME64_2;
statePtr->v[2] = seed + 0;
statePtr->v[3] = seed - XXH_PRIME64_1;
return XXH_OK;
}
/*! @ingroup xxh64_family */
XXH_PUBLIC_API XXH_errorcode
XXH64_update (XXH64_state_t* state, const void* input, size_t len)
{
if (input==NULL) {
XXH_ASSERT(len == 0);
return XXH_OK;
}
{ const xxh_u8* p = (const xxh_u8*)input;
const xxh_u8* const bEnd = p + len;
state->total_len += len;
if (state->memsize + len < 32) { /* fill in tmp buffer */
XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, len);
state->memsize += (xxh_u32)len;
return XXH_OK;
}
if (state->memsize) { /* tmp buffer is full */
XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, 32-state->memsize);
state->v[0] = XXH64_round(state->v[0], XXH_readLE64(state->mem64+0));
state->v[1] = XXH64_round(state->v[1], XXH_readLE64(state->mem64+1));
state->v[2] = XXH64_round(state->v[2], XXH_readLE64(state->mem64+2));
state->v[3] = XXH64_round(state->v[3], XXH_readLE64(state->mem64+3));
p += 32 - state->memsize;
state->memsize = 0;
}
if (p+32 <= bEnd) {
const xxh_u8* const limit = bEnd - 32;
do {
state->v[0] = XXH64_round(state->v[0], XXH_readLE64(p)); p+=8;
state->v[1] = XXH64_round(state->v[1], XXH_readLE64(p)); p+=8;
state->v[2] = XXH64_round(state->v[2], XXH_readLE64(p)); p+=8;
state->v[3] = XXH64_round(state->v[3], XXH_readLE64(p)); p+=8;
} while (p<=limit);
}
if (p < bEnd) {
XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
state->memsize = (unsigned)(bEnd-p);
}
}
return XXH_OK;
}
/*! @ingroup xxh64_family */
XXH_PUBLIC_API XXH64_hash_t XXH64_digest(const XXH64_state_t* state)
{
xxh_u64 h64;
if (state->total_len >= 32) {
h64 = XXH_rotl64(state->v[0], 1) + XXH_rotl64(state->v[1], 7) + XXH_rotl64(state->v[2], 12) + XXH_rotl64(state->v[3], 18);
h64 = XXH64_mergeRound(h64, state->v[0]);
h64 = XXH64_mergeRound(h64, state->v[1]);
h64 = XXH64_mergeRound(h64, state->v[2]);
h64 = XXH64_mergeRound(h64, state->v[3]);
} else {
h64 = state->v[2] /*seed*/ + XXH_PRIME64_5;
}
h64 += (xxh_u64) state->total_len;
return XXH64_finalize(h64, (const xxh_u8*)state->mem64, (size_t)state->total_len, XXH_aligned);
}
/******* Canonical representation *******/
/*! @ingroup xxh64_family */
XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
{
/* XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t)); */
if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
XXH_memcpy(dst, &hash, sizeof(*dst));
}
/*! @ingroup xxh64_family */
XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
{
return XXH_readBE64(src);
}
#ifndef XXH_NO_XXH3
/* *********************************************************************
* XXH3
* New generation hash designed for speed on small keys and vectorization
************************************************************************ */
/*!
* @}
* @defgroup xxh3_impl XXH3 implementation
* @ingroup impl
* @{
*/
/* === Compiler specifics === */
#if ((defined(sun) || defined(__sun)) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */
# define XXH_RESTRICT /* disable */
#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* >= C99 */
# define XXH_RESTRICT restrict
#else
/* Note: it might be useful to define __restrict or __restrict__ for some C++ compilers */
# define XXH_RESTRICT /* disable */
#endif
#if (defined(__GNUC__) && (__GNUC__ >= 3)) \
|| (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) \
|| defined(__clang__)
# define XXH_likely(x) __builtin_expect(x, 1)
# define XXH_unlikely(x) __builtin_expect(x, 0)
#else
# define XXH_likely(x) (x)
# define XXH_unlikely(x) (x)
#endif
#if defined(__GNUC__) || defined(__clang__)
# if defined(__ARM_NEON__) || defined(__ARM_NEON) \
|| defined(__aarch64__) || defined(_M_ARM) \
|| defined(_M_ARM64) || defined(_M_ARM64EC)
# define inline __inline__ /* circumvent a clang bug */
# include <arm_neon.h>
# undef inline
# elif defined(__AVX2__)
# include <immintrin.h>
# elif defined(__SSE2__)
# include <emmintrin.h>
# endif
#endif
#if defined(_MSC_VER)
# include <intrin.h>
#endif
/*
* One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while
* remaining a true 64-bit/128-bit hash function.
*
* This is done by prioritizing a subset of 64-bit operations that can be
* emulated without too many steps on the average 32-bit machine.
*
* For example, these two lines seem similar, and run equally fast on 64-bit:
*
* xxh_u64 x;
* x ^= (x >> 47); // good
* x ^= (x >> 13); // bad
*
* However, to a 32-bit machine, there is a major difference.
*
* x ^= (x >> 47) looks like this:
*
* x.lo ^= (x.hi >> (47 - 32));
*
* while x ^= (x >> 13) looks like this:
*
* // note: funnel shifts are not usually cheap.
* x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13));
* x.hi ^= (x.hi >> 13);
*
* The first one is significantly faster than the second, simply because the
* shift is larger than 32. This means:
* - All the bits we need are in the upper 32 bits, so we can ignore the lower
* 32 bits in the shift.
* - The shift result will always fit in the lower 32 bits, and therefore,
* we can ignore the upper 32 bits in the xor.
*
* Thanks to this optimization, XXH3 only requires these features to be efficient:
*
* - Usable unaligned access
* - A 32-bit or 64-bit ALU
* - If 32-bit, a decent ADC instruction
* - A 32 or 64-bit multiply with a 64-bit result
* - For the 128-bit variant, a decent byteswap helps short inputs.
*
* The first two are already required by XXH32, and almost all 32-bit and 64-bit
* platforms which can run XXH32 can run XXH3 efficiently.
*
* Thumb-1, the classic 16-bit only subset of ARM's instruction set, is one
* notable exception.
*
* First of all, Thumb-1 lacks support for the UMULL instruction which
* performs the important long multiply. This means numerous __aeabi_lmul
* calls.
*
* Second of all, the 8 functional registers are just not enough.
* Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic need
* Lo registers, and this shuffling results in thousands more MOVs than A32.
*
* A32 and T32 don't have this limitation. They can access all 14 registers,
* do a 32->64 multiply with UMULL, and the flexible operand allowing free
* shifts is helpful, too.
*
* Therefore, we do a quick sanity check.
*
* If compiling Thumb-1 for a target which supports ARM instructions, we will
* emit a warning, as it is not a "sane" platform to compile for.
*
* Usually, if this happens, it is because of an accident and you probably need
* to specify -march, as you likely meant to compile for a newer architecture.
*
* Credit: large sections of the vectorial and asm source code paths
* have been contributed by @easyaspi314
*/
#if defined(__thumb__) && !defined(__thumb2__) && defined(__ARM_ARCH_ISA_ARM)
# warning "XXH3 is highly inefficient without ARM or Thumb-2."
#endif
/* ==========================================
* Vectorization detection
* ========================================== */
#ifdef XXH_DOXYGEN
/*!
* @ingroup tuning
* @brief Overrides the vectorization implementation chosen for XXH3.
*
* Can be defined to 0 to disable SIMD or any of the values mentioned in
* @ref XXH_VECTOR_TYPE.
*
* If this is not defined, it uses predefined macros to determine the best
* implementation.
*/
# define XXH_VECTOR XXH_SCALAR
/*!
* @ingroup tuning
* @brief Possible values for @ref XXH_VECTOR.
*
* Note that these are actually implemented as macros.
*
* If this is not defined, it is detected automatically.
* @ref XXH_X86DISPATCH overrides this.
*/
enum XXH_VECTOR_TYPE /* fake enum */ {
XXH_SCALAR = 0, /*!< Portable scalar version */
XXH_SSE2 = 1, /*!<
* SSE2 for Pentium 4, Opteron, all x86_64.
*
* @note SSE2 is also guaranteed on Windows 10, macOS, and
* Android x86.
*/
XXH_AVX2 = 2, /*!< AVX2 for Haswell and Bulldozer */
XXH_AVX512 = 3, /*!< AVX512 for Skylake and Icelake */
XXH_NEON = 4, /*!< NEON for most ARMv7-A and all AArch64 */
XXH_VSX = 5, /*!< VSX and ZVector for POWER8/z13 (64-bit) */
};
/*!
* @ingroup tuning
* @brief Selects the minimum alignment for XXH3's accumulators.
*
* When using SIMD, this should match the alignment required for said vector
* type, so, for example, 32 for AVX2.
*
* Default: Auto detected.
*/
# define XXH_ACC_ALIGN 8
#endif
/* Actual definition */
#ifndef XXH_DOXYGEN
# define XXH_SCALAR 0
# define XXH_SSE2 1
# define XXH_AVX2 2
# define XXH_AVX512 3
# define XXH_NEON 4
# define XXH_VSX 5
#endif
#ifndef XXH_VECTOR /* can be defined on command line */
# if ( \
defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \
|| defined(_M_ARM) || defined(_M_ARM64) || defined(_M_ARM64EC) /* msvc */ \
) && ( \
defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \
|| (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
)
# define XXH_VECTOR XXH_NEON
# elif defined(__AVX512F__)
# define XXH_VECTOR XXH_AVX512
# elif defined(__AVX2__)
# define XXH_VECTOR XXH_AVX2
# elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
# define XXH_VECTOR XXH_SSE2
# elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \
|| (defined(__s390x__) && defined(__VEC__)) \
&& defined(__GNUC__) /* TODO: IBM XL */
# define XXH_VECTOR XXH_VSX
# else
# define XXH_VECTOR XXH_SCALAR
# endif
#endif
/*
* Controls the alignment of the accumulator,
* for compatibility with aligned vector loads, which are usually faster.
*/
#ifndef XXH_ACC_ALIGN
# if defined(XXH_X86DISPATCH)
# define XXH_ACC_ALIGN 64 /* for compatibility with avx512 */
# elif XXH_VECTOR == XXH_SCALAR /* scalar */
# define XXH_ACC_ALIGN 8
# elif XXH_VECTOR == XXH_SSE2 /* sse2 */
# define XXH_ACC_ALIGN 16
# elif XXH_VECTOR == XXH_AVX2 /* avx2 */
# define XXH_ACC_ALIGN 32
# elif XXH_VECTOR == XXH_NEON /* neon */
# define XXH_ACC_ALIGN 16
# elif XXH_VECTOR == XXH_VSX /* vsx */
# define XXH_ACC_ALIGN 16
# elif XXH_VECTOR == XXH_AVX512 /* avx512 */
# define XXH_ACC_ALIGN 64
# endif
#endif
#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \
|| XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
# define XXH_SEC_ALIGN XXH_ACC_ALIGN
#else
# define XXH_SEC_ALIGN 8
#endif
/*
* UGLY HACK:
* GCC usually generates the best code with -O3 for xxHash.
*
* However, when targeting AVX2, it is overzealous in its unrolling resulting
* in code roughly 3/4 the speed of Clang.
*
* There are other issues, such as GCC splitting _mm256_loadu_si256 into
* _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization which
* only applies to Sandy and Ivy Bridge... which don't even support AVX2.
*
* That is why when compiling the AVX2 version, it is recommended to use either
* -O2 -mavx2 -march=haswell
* or
* -O2 -mavx2 -mno-avx256-split-unaligned-load
* for decent performance, or to use Clang instead.
*
* Fortunately, we can control the first one with a pragma that forces GCC into
* -O2, but the other one we can't control without "failed to inline always
* inline function due to target mismatch" warnings.
*/
#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
&& defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
&& defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) /* respect -O0 and -Os */
# pragma GCC push_options
# pragma GCC optimize("-O2")
#endif
#if XXH_VECTOR == XXH_NEON
/*
* NEON's setup for vmlal_u32 is a little more complicated than it is on
* SSE2, AVX2, and VSX.
*
* While PMULUDQ and VMULEUW both perform a mask, VMLAL.U32 performs an upcast.
*
* To do the same operation, the 128-bit 'Q' register needs to be split into
* two 64-bit 'D' registers, performing this operation::
*
* [ a | b ]
* | '---------. .--------' |
* | x |
* | .---------' '--------. |
* [ a & 0xFFFFFFFF | b & 0xFFFFFFFF ],[ a >> 32 | b >> 32 ]
*
* Due to significant changes in aarch64, the fastest method for aarch64 is
* completely different than the fastest method for ARMv7-A.
*
* ARMv7-A treats D registers as unions overlaying Q registers, so modifying
* D11 will modify the high half of Q5. This is similar to how modifying AH
* will only affect bits 8-15 of AX on x86.
*
* VZIP takes two registers, and puts even lanes in one register and odd lanes
* in the other.
*
* On ARMv7-A, this strangely modifies both parameters in place instead of
* taking the usual 3-operand form.
*
* Therefore, if we want to do this, we can simply use a D-form VZIP.32 on the
* lower and upper halves of the Q register to end up with the high and low
* halves where we want - all in one instruction.
*
* vzip.32 d10, d11 @ d10 = { d10[0], d11[0] }; d11 = { d10[1], d11[1] }
*
* Unfortunately we need inline assembly for this: Instructions modifying two
* registers at once is not possible in GCC or Clang's IR, and they have to
* create a copy.
*
* aarch64 requires a different approach.
*
* In order to make it easier to write a decent compiler for aarch64, many
* quirks were removed, such as conditional execution.
*
* NEON was also affected by this.
*
* aarch64 cannot access the high bits of a Q-form register, and writes to a
* D-form register zero the high bits, similar to how writes to W-form scalar
* registers (or DWORD registers on x86_64) work.
*
* The formerly free vget_high intrinsics now require a vext (with a few
* exceptions)
*
* Additionally, VZIP was replaced by ZIP1 and ZIP2, which are the equivalent
* of PUNPCKL* and PUNPCKH* in SSE, respectively, in order to only modify one
* operand.
*
* The equivalent of the VZIP.32 on the lower and upper halves would be this
* mess:
*
* ext v2.4s, v0.4s, v0.4s, #2 // v2 = { v0[2], v0[3], v0[0], v0[1] }
* zip1 v1.2s, v0.2s, v2.2s // v1 = { v0[0], v2[0] }
* zip2 v0.2s, v0.2s, v1.2s // v0 = { v0[1], v2[1] }
*
* Instead, we use a literal downcast, vmovn_u64 (XTN), and vshrn_n_u64 (SHRN):
*
* shrn v1.2s, v0.2d, #32 // v1 = (uint32x2_t)(v0 >> 32);
* xtn v0.2s, v0.2d // v0 = (uint32x2_t)(v0 & 0xFFFFFFFF);
*
* This is available on ARMv7-A, but is less efficient than a single VZIP.32.
*/
/*!
* Function-like macro:
* void XXH_SPLIT_IN_PLACE(uint64x2_t &in, uint32x2_t &outLo, uint32x2_t &outHi)
* {
* outLo = (uint32x2_t)(in & 0xFFFFFFFF);
* outHi = (uint32x2_t)(in >> 32);
* in = UNDEFINED;
* }
*/
# if !defined(XXH_NO_VZIP_HACK) /* define to disable */ \
&& (defined(__GNUC__) || defined(__clang__)) \
&& (defined(__arm__) || defined(__thumb__) || defined(_M_ARM))
# define XXH_SPLIT_IN_PLACE(in, outLo, outHi) \
do { \
/* Undocumented GCC/Clang operand modifier: %e0 = lower D half, %f0 = upper D half */ \
/* https://github.com/gcc-mirror/gcc/blob/38cf91e5/gcc/config/arm/arm.c#L22486 */ \
/* https://github.com/llvm-mirror/llvm/blob/2c4ca683/lib/Target/ARM/ARMAsmPrinter.cpp#L399 */ \
__asm__("vzip.32 %e0, %f0" : "+w" (in)); \
(outLo) = vget_low_u32 (vreinterpretq_u32_u64(in)); \
(outHi) = vget_high_u32(vreinterpretq_u32_u64(in)); \
} while (0)
# else
# define XXH_SPLIT_IN_PLACE(in, outLo, outHi) \
do { \
(outLo) = vmovn_u64 (in); \
(outHi) = vshrn_n_u64 ((in), 32); \
} while (0)
# endif
/*!
* @ingroup tuning
* @brief Controls the NEON to scalar ratio for XXH3
*
* On AArch64 when not optimizing for size, XXH3 will run 6 lanes using NEON and
* 2 lanes on scalar by default.
*
* This can be set to 2, 4, 6, or 8. ARMv7 will default to all 8 NEON lanes, as the
* emulated 64-bit arithmetic is too slow.
*
* Modern ARM CPUs are _very_ sensitive to how their pipelines are used.
*
* For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but it can't
* have more than 2 NEON (F0/F1) micro-ops. If you are only using NEON instructions,
* you are only using 2/3 of the CPU bandwidth.
*
* This is even more noticeable on the more advanced cores like the A76 which
* can dispatch 8 micro-ops per cycle, but still only 2 NEON micro-ops at once.
*
* Therefore, @ref XXH3_NEON_LANES lanes will be processed using NEON, and the
* remaining lanes will use scalar instructions. This improves the bandwidth
* and also gives the integer pipelines something to do besides twiddling loop
* counters and pointers.
*
* This change benefits CPUs with large micro-op buffers without negatively affecting
* other CPUs:
*
* | Chipset | Dispatch type | NEON only | 6:2 hybrid | Diff. |
* |:----------------------|:--------------------|----------:|-----------:|------:|
* | Snapdragon 730 (A76) | 2 NEON/8 micro-ops | 8.8 GB/s | 10.1 GB/s | ~16% |
* | Snapdragon 835 (A73) | 2 NEON/3 micro-ops | 5.1 GB/s | 5.3 GB/s | ~5% |
* | Marvell PXA1928 (A53) | In-order dual-issue | 1.9 GB/s | 1.9 GB/s | 0% |
*
* It also seems to fix some bad codegen on GCC, making it almost as fast as clang.
*
* @see XXH3_accumulate_512_neon()
*/
# ifndef XXH3_NEON_LANES
# if (defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) || defined(_M_ARM64EC)) \
&& !defined(__OPTIMIZE_SIZE__)
# define XXH3_NEON_LANES 6
# else
# define XXH3_NEON_LANES XXH_ACC_NB
# endif
# endif
#endif /* XXH_VECTOR == XXH_NEON */
/*
* VSX and Z Vector helpers.
*
* This is very messy, and any pull requests to clean this up are welcome.
*
* There are a lot of problems with supporting VSX and s390x, due to
* inconsistent intrinsics, spotty coverage, and multiple endiannesses.
*/
#if XXH_VECTOR == XXH_VSX
# if defined(__s390x__)
# include <s390intrin.h>
# else
/* gcc's altivec.h can have the unwanted consequence to unconditionally
* #define bool, vector, and pixel keywords,
* with bad consequences for programs already using these keywords for other purposes.
* The paragraph defining these macros is skipped when __APPLE_ALTIVEC__ is defined.
* __APPLE_ALTIVEC__ is _generally_ defined automatically by the compiler,
* but it seems that, in some cases, it isn't.
* Force the build macro to be defined, so that keywords are not altered.
*/
# if defined(__GNUC__) && !defined(__APPLE_ALTIVEC__)
# define __APPLE_ALTIVEC__
# endif
# include <altivec.h>
# endif
typedef __vector unsigned long long xxh_u64x2;
typedef __vector unsigned char xxh_u8x16;
typedef __vector unsigned xxh_u32x4;
# ifndef XXH_VSX_BE
# if defined(__BIG_ENDIAN__) \
|| (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
# define XXH_VSX_BE 1
# elif defined(__VEC_ELEMENT_REG_ORDER__) && __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__
# warning "-maltivec=be is not recommended. Please use native endianness."
# define XXH_VSX_BE 1
# else
# define XXH_VSX_BE 0
# endif
# endif /* !defined(XXH_VSX_BE) */
# if XXH_VSX_BE
# if defined(__POWER9_VECTOR__) || (defined(__clang__) && defined(__s390x__))
# define XXH_vec_revb vec_revb
# else
/*!
* A polyfill for POWER9's vec_revb().
*/
XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val)
{
xxh_u8x16 const vByteSwap = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00,
0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
return vec_perm(val, val, vByteSwap);
}
# endif
# endif /* XXH_VSX_BE */
/*!
* Performs an unaligned vector load and byte swaps it on big endian.
*/
XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr)
{
xxh_u64x2 ret;
XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2));
# if XXH_VSX_BE
ret = XXH_vec_revb(ret);
# endif
return ret;
}
/*
* vec_mulo and vec_mule are very problematic intrinsics on PowerPC
*
* These intrinsics weren't added until GCC 8, despite existing for a while,
* and they are endian dependent. Also, their meaning swap depending on version.
* */
# if defined(__s390x__)
/* s390x is always big endian, no issue on this platform */
# define XXH_vec_mulo vec_mulo
# define XXH_vec_mule vec_mule
# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw)
/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */
# define XXH_vec_mulo __builtin_altivec_vmulouw
# define XXH_vec_mule __builtin_altivec_vmuleuw
# else
/* gcc needs inline assembly */
/* Adapted from https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */
XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b)
{
xxh_u64x2 result;
__asm__("vmulouw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
return result;
}
XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b)
{
xxh_u64x2 result;
__asm__("vmuleuw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
return result;
}
# endif /* XXH_vec_mulo, XXH_vec_mule */
#endif /* XXH_VECTOR == XXH_VSX */
/* prefetch
* can be disabled, by declaring XXH_NO_PREFETCH build macro */
#if defined(XXH_NO_PREFETCH)
# define XXH_PREFETCH(ptr) (void)(ptr) /* disabled */
#else
# if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)) && !defined(_M_ARM64EC) /* _mm_prefetch() not defined outside of x86/x64 */
# include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
# define XXH_PREFETCH(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
# elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
# define XXH_PREFETCH(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
# else
# define XXH_PREFETCH(ptr) (void)(ptr) /* disabled */
# endif
#endif /* XXH_NO_PREFETCH */
/* ==========================================
* XXH3 default settings
* ========================================== */
#define XXH_SECRET_DEFAULT_SIZE 192 /* minimum XXH3_SECRET_SIZE_MIN */
#if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN)
# error "default keyset is not large enough"
#endif
/*! Pseudorandom secret taken directly from FARSH. */
XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c,
0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f,
0xcb, 0x79, 0xe6, 0x4e, 0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21,
0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6, 0x81, 0x3a, 0x26, 0x4c,
0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb, 0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3,
0x71, 0x64, 0x48, 0x97, 0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8,
0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7, 0xc7, 0x0b, 0x4f, 0x1d,
0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31, 0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64,
0xea, 0xc5, 0xac, 0x83, 0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb,
0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26, 0x29, 0xd4, 0x68, 0x9e,
0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc, 0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce,
0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
};
#ifdef XXH_OLD_NAMES
# define kSecret XXH3_kSecret
#endif
#ifdef XXH_DOXYGEN
/*!
* @brief Calculates a 32-bit to 64-bit long multiply.
*
* Implemented as a macro.
*
* Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it doesn't
* need to (but it shouldn't need to anyways, it is about 7 instructions to do
* a 64x64 multiply...). Since we know that this will _always_ emit `MULL`, we
* use that instead of the normal method.
*
* If you are compiling for platforms like Thumb-1 and don't have a better option,
* you may also want to write your own long multiply routine here.
*
* @param x, y Numbers to be multiplied
* @return 64-bit product of the low 32 bits of @p x and @p y.
*/
XXH_FORCE_INLINE xxh_u64
XXH_mult32to64(xxh_u64 x, xxh_u64 y)
{
return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
}
#elif defined(_MSC_VER) && defined(_M_IX86)
# define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
#else
/*
* Downcast + upcast is usually better than masking on older compilers like
* GCC 4.2 (especially 32-bit ones), all without affecting newer compilers.
*
* The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both operands
* and perform a full 64x64 multiply -- entirely redundant on 32-bit.
*/
# define XXH_mult32to64(x, y) ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
#endif
/*!
* @brief Calculates a 64->128-bit long multiply.
*
* Uses `__uint128_t` and `_umul128` if available, otherwise uses a scalar
* version.
*
* @param lhs , rhs The 64-bit integers to be multiplied
* @return The 128-bit result represented in an @ref XXH128_hash_t.
*/
static XXH128_hash_t
XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
{
/*
* GCC/Clang __uint128_t method.
*
* On most 64-bit targets, GCC and Clang define a __uint128_t type.
* This is usually the best way as it usually uses a native long 64-bit
* multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64.
*
* Usually.
*
* Despite being a 32-bit platform, Clang (and emscripten) define this type
* despite not having the arithmetic for it. This results in a laggy
* compiler builtin call which calculates a full 128-bit multiply.
* In that case it is best to use the portable one.
* https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
*/
#if (defined(__GNUC__) || defined(__clang__)) && !defined(__wasm__) \
&& defined(__SIZEOF_INT128__) \
|| (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
__uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs;
XXH128_hash_t r128;
r128.low64 = (xxh_u64)(product);
r128.high64 = (xxh_u64)(product >> 64);
return r128;
/*
* MSVC for x64's _umul128 method.
*
* xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64 *HighProduct);
*
* This compiles to single operand MUL on x64.
*/
#elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC)
#ifndef _MSC_VER
# pragma intrinsic(_umul128)
#endif
xxh_u64 product_high;
xxh_u64 const product_low = _umul128(lhs, rhs, &product_high);
XXH128_hash_t r128;
r128.low64 = product_low;
r128.high64 = product_high;
return r128;
/*
* MSVC for ARM64's __umulh method.
*
* This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method.
*/
#elif defined(_M_ARM64) || defined(_M_ARM64EC)
#ifndef _MSC_VER
# pragma intrinsic(__umulh)
#endif
XXH128_hash_t r128;
r128.low64 = lhs * rhs;
r128.high64 = __umulh(lhs, rhs);
return r128;
#else
/*
* Portable scalar method. Optimized for 32-bit and 64-bit ALUs.
*
* This is a fast and simple grade school multiply, which is shown below
* with base 10 arithmetic instead of base 0x100000000.
*
* 9 3 // D2 lhs = 93
* x 7 5 // D2 rhs = 75
* ----------
* 1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15
* 4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45
* 2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21
* + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63
* ---------
* 2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27
* + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67
* ---------
* 6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975
*
* The reasons for adding the products like this are:
* 1. It avoids manual carry tracking. Just like how
* (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX.
* This avoids a lot of complexity.
*
* 2. It hints for, and on Clang, compiles to, the powerful UMAAL
* instruction available in ARM's Digital Signal Processing extension
* in 32-bit ARMv6 and later, which is shown below:
*
* void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm)
* {
* xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm;
* *RdLo = (xxh_u32)(product & 0xFFFFFFFF);
* *RdHi = (xxh_u32)(product >> 32);
* }
*
* This instruction was designed for efficient long multiplication, and
* allows this to be calculated in only 4 instructions at speeds
* comparable to some 64-bit ALUs.
*
* 3. It isn't terrible on other platforms. Usually this will be a couple
* of 32-bit ADD/ADCs.
*/
/* First calculate all of the cross products. */
xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF);
xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32, rhs & 0xFFFFFFFF);
xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32);
xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32, rhs >> 32);
/* Now add the products together. These will never overflow. */
xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi;
xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32) + hi_hi;
xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF);
XXH128_hash_t r128;
r128.low64 = lower;
r128.high64 = upper;
return r128;
#endif
}
/*!
* @brief Calculates a 64-bit to 128-bit multiply, then XOR folds it.
*
* The reason for the separate function is to prevent passing too many structs
* around by value. This will hopefully inline the multiply, but we don't force it.
*
* @param lhs , rhs The 64-bit integers to multiply
* @return The low 64 bits of the product XOR'd by the high 64 bits.
* @see XXH_mult64to128()
*/
static xxh_u64
XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs)
{
XXH128_hash_t product = XXH_mult64to128(lhs, rhs);
return product.low64 ^ product.high64;
}
/*! Seems to produce slightly better code on GCC for some reason. */
XXH_FORCE_INLINE xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
{
XXH_ASSERT(0 <= shift && shift < 64);
return v64 ^ (v64 >> shift);
}
/*
* This is a fast avalanche stage,
* suitable when input bits are already partially mixed
*/
static XXH64_hash_t XXH3_avalanche(xxh_u64 h64)
{
h64 = XXH_xorshift64(h64, 37);
h64 *= 0x165667919E3779F9ULL;
h64 = XXH_xorshift64(h64, 32);
return h64;
}
/*
* This is a stronger avalanche,
* inspired by Pelle Evensen's rrmxmx
* preferable when input has not been previously mixed
*/
static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
{
/* this mix is inspired by Pelle Evensen's rrmxmx */
h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
h64 *= 0x9FB21C651E98DF25ULL;
h64 ^= (h64 >> 35) + len ;
h64 *= 0x9FB21C651E98DF25ULL;
return XXH_xorshift64(h64, 28);
}
/* ==========================================
* Short keys
* ==========================================
* One of the shortcomings of XXH32 and XXH64 was that their performance was
* sub-optimal on short lengths. It used an iterative algorithm which strongly
* favored lengths that were a multiple of 4 or 8.
*
* Instead of iterating over individual inputs, we use a set of single shot
* functions which piece together a range of lengths and operate in constant time.
*
* Additionally, the number of multiplies has been significantly reduced. This
* reduces latency, especially when emulating 64-bit multiplies on 32-bit.
*
* Depending on the platform, this may or may not be faster than XXH32, but it
* is almost guaranteed to be faster than XXH64.
*/
/*
* At very short lengths, there isn't enough input to fully hide secrets, or use
* the entire secret.
*
* There is also only a limited amount of mixing we can do before significantly
* impacting performance.
*
* Therefore, we use different sections of the secret and always mix two secret
* samples with an XOR. This should have no effect on performance on the
* seedless or withSeed variants because everything _should_ be constant folded
* by modern compilers.
*
* The XOR mixing hides individual parts of the secret and increases entropy.
*
* This adds an extra layer of strength for custom secrets.
*/
XXH_FORCE_INLINE XXH64_hash_t
XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
XXH_ASSERT(1 <= len && len <= 3);
XXH_ASSERT(secret != NULL);
/*
* len = 1: combined = { input[0], 0x01, input[0], input[0] }
* len = 2: combined = { input[1], 0x02, input[0], input[1] }
* len = 3: combined = { input[2], 0x03, input[0], input[1] }
*/
{ xxh_u8 const c1 = input[0];
xxh_u8 const c2 = input[len >> 1];
xxh_u8 const c3 = input[len - 1];
xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2 << 24)
| ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
xxh_u64 const bitflip = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
xxh_u64 const keyed = (xxh_u64)combined ^ bitflip;
return XXH64_avalanche(keyed);
}
}
XXH_FORCE_INLINE XXH64_hash_t
XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
XXH_ASSERT(secret != NULL);
XXH_ASSERT(4 <= len && len <= 8);
seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
{ xxh_u32 const input1 = XXH_readLE32(input);
xxh_u32 const input2 = XXH_readLE32(input + len - 4);
xxh_u64 const bitflip = (XXH_readLE64(secret+8) ^ XXH_readLE64(secret+16)) - seed;
xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32);
xxh_u64 const keyed = input64 ^ bitflip;
return XXH3_rrmxmx(keyed, len);
}
}
XXH_FORCE_INLINE XXH64_hash_t
XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
XXH_ASSERT(secret != NULL);
XXH_ASSERT(9 <= len && len <= 16);
{ xxh_u64 const bitflip1 = (XXH_readLE64(secret+24) ^ XXH_readLE64(secret+32)) + seed;
xxh_u64 const bitflip2 = (XXH_readLE64(secret+40) ^ XXH_readLE64(secret+48)) - seed;
xxh_u64 const input_lo = XXH_readLE64(input) ^ bitflip1;
xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2;
xxh_u64 const acc = len
+ XXH_swap64(input_lo) + input_hi
+ XXH3_mul128_fold64(input_lo, input_hi);
return XXH3_avalanche(acc);
}
}
XXH_FORCE_INLINE XXH64_hash_t
XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(len <= 16);
{ if (XXH_likely(len > 8)) return XXH3_len_9to16_64b(input, len, secret, seed);
if (XXH_likely(len >= 4)) return XXH3_len_4to8_64b(input, len, secret, seed);
if (len) return XXH3_len_1to3_64b(input, len, secret, seed);
return XXH64_avalanche(seed ^ (XXH_readLE64(secret+56) ^ XXH_readLE64(secret+64)));
}
}
/*
* DISCLAIMER: There are known *seed-dependent* multicollisions here due to
* multiplication by zero, affecting hashes of lengths 17 to 240.
*
* However, they are very unlikely.
*
* Keep this in mind when using the unseeded XXH3_64bits() variant: As with all
* unseeded non-cryptographic hashes, it does not attempt to defend itself
* against specially crafted inputs, only random inputs.
*
* Compared to classic UMAC where a 1 in 2^31 chance of 4 consecutive bytes
* cancelling out the secret is taken an arbitrary number of times (addressed
* in XXH3_accumulate_512), this collision is very unlikely with random inputs
* and/or proper seeding:
*
* This only has a 1 in 2^63 chance of 8 consecutive bytes cancelling out, in a
* function that is only called up to 16 times per hash with up to 240 bytes of
* input.
*
* This is not too bad for a non-cryptographic hash function, especially with
* only 64 bit outputs.
*
* The 128-bit variant (which trades some speed for strength) is NOT affected
* by this, although it is always a good idea to use a proper seed if you care
* about strength.
*/
XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input,
const xxh_u8* XXH_RESTRICT secret, xxh_u64 seed64)
{
#if defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
&& defined(__i386__) && defined(__SSE2__) /* x86 + SSE2 */ \
&& !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable like XXH32 hack */
/*
* UGLY HACK:
* GCC for x86 tends to autovectorize the 128-bit multiply, resulting in
* slower code.
*
* By forcing seed64 into a register, we disrupt the cost model and
* cause it to scalarize. See `XXH32_round()`
*
* FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600,
* XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on
* GCC 9.2, despite both emitting scalar code.
*
* GCC generates much better scalar code than Clang for the rest of XXH3,
* which is why finding a more optimal codepath is an interest.
*/
XXH_COMPILER_GUARD(seed64);
#endif
{ xxh_u64 const input_lo = XXH_readLE64(input);
xxh_u64 const input_hi = XXH_readLE64(input+8);
return XXH3_mul128_fold64(
input_lo ^ (XXH_readLE64(secret) + seed64),
input_hi ^ (XXH_readLE64(secret+8) - seed64)
);
}
}
/* For mid range keys, XXH3 uses a Mum-hash variant. */
XXH_FORCE_INLINE XXH64_hash_t
XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH64_hash_t seed)
{
XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
XXH_ASSERT(16 < len && len <= 128);
{ xxh_u64 acc = len * XXH_PRIME64_1;
if (len > 32) {
if (len > 64) {
if (len > 96) {
acc += XXH3_mix16B(input+48, secret+96, seed);
acc += XXH3_mix16B(input+len-64, secret+112, seed);
}
acc += XXH3_mix16B(input+32, secret+64, seed);
acc += XXH3_mix16B(input+len-48, secret+80, seed);
}
acc += XXH3_mix16B(input+16, secret+32, seed);
acc += XXH3_mix16B(input+len-32, secret+48, seed);
}
acc += XXH3_mix16B(input+0, secret+0, seed);
acc += XXH3_mix16B(input+len-16, secret+16, seed);
return XXH3_avalanche(acc);
}
}
#define XXH3_MIDSIZE_MAX 240
XXH_NO_INLINE XXH64_hash_t
XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH64_hash_t seed)
{
XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
#define XXH3_MIDSIZE_STARTOFFSET 3
#define XXH3_MIDSIZE_LASTOFFSET 17
{ xxh_u64 acc = len * XXH_PRIME64_1;
int const nbRounds = (int)len / 16;
int i;
for (i=0; i<8; i++) {
acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed);
}
acc = XXH3_avalanche(acc);
XXH_ASSERT(nbRounds >= 8);
#if defined(__clang__) /* Clang */ \
&& (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \
&& !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable */
/*
* UGLY HACK:
* Clang for ARMv7-A tries to vectorize this loop, similar to GCC x86.
* In everywhere else, it uses scalar code.
*
* For 64->128-bit multiplies, even if the NEON was 100% optimal, it
* would still be slower than UMAAL (see XXH_mult64to128).
*
* Unfortunately, Clang doesn't handle the long multiplies properly and
* converts them to the nonexistent "vmulq_u64" intrinsic, which is then
* scalarized into an ugly mess of VMOV.32 instructions.
*
* This mess is difficult to avoid without turning autovectorization
* off completely, but they are usually relatively minor and/or not
* worth it to fix.
*
* This loop is the easiest to fix, as unlike XXH32, this pragma
* _actually works_ because it is a loop vectorization instead of an
* SLP vectorization.
*/
#pragma clang loop vectorize(disable)
#endif
for (i=8 ; i < nbRounds; i++) {
acc += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
}
/* last bytes */
acc += XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
return XXH3_avalanche(acc);
}
}
/* ======= Long Keys ======= */
#define XXH_STRIPE_LEN 64
#define XXH_SECRET_CONSUME_RATE 8 /* nb of secret bytes consumed at each accumulation */
#define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64))
#ifdef XXH_OLD_NAMES
# define STRIPE_LEN XXH_STRIPE_LEN
# define ACC_NB XXH_ACC_NB
#endif
XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64)
{
if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
XXH_memcpy(dst, &v64, sizeof(v64));
}
/* Several intrinsic functions below are supposed to accept __int64 as argument,
* as documented in https://software.intel.com/sites/landingpage/IntrinsicsGuide/ .
* However, several environments do not define __int64 type,
* requiring a workaround.
*/
#if !defined (__VMS) \
&& (defined (__cplusplus) \
|| (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
typedef int64_t xxh_i64;
#else
/* the following type must have a width of 64-bit */
typedef long long xxh_i64;
#endif
/*
* XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized.
*
* It is a hardened version of UMAC, based off of FARSH's implementation.
*
* This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD
* implementations, and it is ridiculously fast.
*
* We harden it by mixing the original input to the accumulators as well as the product.
*
* This means that in the (relatively likely) case of a multiply by zero, the
* original input is preserved.
*
* On 128-bit inputs, we swap 64-bit pairs when we add the input to improve
* cross-pollination, as otherwise the upper and lower halves would be
* essentially independent.
*
* This doesn't matter on 64-bit hashes since they all get merged together in
* the end, so we skip the extra step.
*
* Both XXH3_64bits and XXH3_128bits use this subroutine.
*/
#if (XXH_VECTOR == XXH_AVX512) \
|| (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0)
#ifndef XXH_TARGET_AVX512
# define XXH_TARGET_AVX512 /* disable attribute target */
#endif
XXH_FORCE_INLINE XXH_TARGET_AVX512 void
XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc,
const void* XXH_RESTRICT input,
const void* XXH_RESTRICT secret)
{
__m512i* const xacc = (__m512i *) acc;
XXH_ASSERT((((size_t)acc) & 63) == 0);
XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
{
/* data_vec = input[0]; */
__m512i const data_vec = _mm512_loadu_si512 (input);
/* key_vec = secret[0]; */
__m512i const key_vec = _mm512_loadu_si512 (secret);
/* data_key = data_vec ^ key_vec; */
__m512i const data_key = _mm512_xor_si512 (data_vec, key_vec);
/* data_key_lo = data_key >> 32; */
__m512i const data_key_lo = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
/* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
__m512i const product = _mm512_mul_epu32 (data_key, data_key_lo);
/* xacc[0] += swap(data_vec); */
__m512i const data_swap = _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2));
__m512i const sum = _mm512_add_epi64(*xacc, data_swap);
/* xacc[0] += product; */
*xacc = _mm512_add_epi64(product, sum);
}
}
/*
* XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
*
* Multiplication isn't perfect, as explained by Google in HighwayHash:
*
* // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to
* // varying degrees. In descending order of goodness, bytes
* // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32.
* // As expected, the upper and lower bytes are much worse.
*
* Source: https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291
*
* Since our algorithm uses a pseudorandom secret to add some variance into the
* mix, we don't need to (or want to) mix as often or as much as HighwayHash does.
*
* This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid
* extraction.
*
* Both XXH3_64bits and XXH3_128bits use this subroutine.
*/
XXH_FORCE_INLINE XXH_TARGET_AVX512 void
XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 63) == 0);
XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
{ __m512i* const xacc = (__m512i*) acc;
const __m512i prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1);
/* xacc[0] ^= (xacc[0] >> 47) */
__m512i const acc_vec = *xacc;
__m512i const shifted = _mm512_srli_epi64 (acc_vec, 47);
__m512i const data_vec = _mm512_xor_si512 (acc_vec, shifted);
/* xacc[0] ^= secret; */
__m512i const key_vec = _mm512_loadu_si512 (secret);
__m512i const data_key = _mm512_xor_si512 (data_vec, key_vec);
/* xacc[0] *= XXH_PRIME32_1; */
__m512i const data_key_hi = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
__m512i const prod_lo = _mm512_mul_epu32 (data_key, prime32);
__m512i const prod_hi = _mm512_mul_epu32 (data_key_hi, prime32);
*xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
}
}
XXH_FORCE_INLINE XXH_TARGET_AVX512 void
XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
{
XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0);
XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64);
XXH_ASSERT(((size_t)customSecret & 63) == 0);
(void)(&XXH_writeLE64);
{ int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
__m512i const seed = _mm512_mask_set1_epi64(_mm512_set1_epi64((xxh_i64)seed64), 0xAA, (xxh_i64)(0U - seed64));
const __m512i* const src = (const __m512i*) ((const void*) XXH3_kSecret);
__m512i* const dest = ( __m512i*) customSecret;
int i;
XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */
XXH_ASSERT(((size_t)dest & 63) == 0);
for (i=0; i < nbRounds; ++i) {
/* GCC has a bug, _mm512_stream_load_si512 accepts 'void*', not 'void const*',
* this will warn "discards 'const' qualifier". */
union {
const __m512i* cp;
void* p;
} remote_const_void;
remote_const_void.cp = src + i;
dest[i] = _mm512_add_epi64(_mm512_stream_load_si512(remote_const_void.p), seed);
} }
}
#endif
#if (XXH_VECTOR == XXH_AVX2) \
|| (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0)
#ifndef XXH_TARGET_AVX2
# define XXH_TARGET_AVX2 /* disable attribute target */
#endif
XXH_FORCE_INLINE XXH_TARGET_AVX2 void
XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc,
const void* XXH_RESTRICT input,
const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 31) == 0);
{ __m256i* const xacc = (__m256i *) acc;
/* Unaligned. This is mainly for pointer arithmetic, and because
* _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
const __m256i* const xinput = (const __m256i *) input;
/* Unaligned. This is mainly for pointer arithmetic, and because
* _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
const __m256i* const xsecret = (const __m256i *) secret;
size_t i;
for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
/* data_vec = xinput[i]; */
__m256i const data_vec = _mm256_loadu_si256 (xinput+i);
/* key_vec = xsecret[i]; */
__m256i const key_vec = _mm256_loadu_si256 (xsecret+i);
/* data_key = data_vec ^ key_vec; */
__m256i const data_key = _mm256_xor_si256 (data_vec, key_vec);
/* data_key_lo = data_key >> 32; */
__m256i const data_key_lo = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
/* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
__m256i const product = _mm256_mul_epu32 (data_key, data_key_lo);
/* xacc[i] += swap(data_vec); */
__m256i const data_swap = _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
__m256i const sum = _mm256_add_epi64(xacc[i], data_swap);
/* xacc[i] += product; */
xacc[i] = _mm256_add_epi64(product, sum);
} }
}
XXH_FORCE_INLINE XXH_TARGET_AVX2 void
XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 31) == 0);
{ __m256i* const xacc = (__m256i*) acc;
/* Unaligned. This is mainly for pointer arithmetic, and because
* _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
const __m256i* const xsecret = (const __m256i *) secret;
const __m256i prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1);
size_t i;
for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
/* xacc[i] ^= (xacc[i] >> 47) */
__m256i const acc_vec = xacc[i];
__m256i const shifted = _mm256_srli_epi64 (acc_vec, 47);
__m256i const data_vec = _mm256_xor_si256 (acc_vec, shifted);
/* xacc[i] ^= xsecret; */
__m256i const key_vec = _mm256_loadu_si256 (xsecret+i);
__m256i const data_key = _mm256_xor_si256 (data_vec, key_vec);
/* xacc[i] *= XXH_PRIME32_1; */
__m256i const data_key_hi = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
__m256i const prod_lo = _mm256_mul_epu32 (data_key, prime32);
__m256i const prod_hi = _mm256_mul_epu32 (data_key_hi, prime32);
xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
}
}
}
XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
{
XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0);
XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6);
XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64);
(void)(&XXH_writeLE64);
XXH_PREFETCH(customSecret);
{ __m256i const seed = _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64, (xxh_i64)(0U - seed64), (xxh_i64)seed64);
const __m256i* const src = (const __m256i*) ((const void*) XXH3_kSecret);
__m256i* dest = ( __m256i*) customSecret;
# if defined(__GNUC__) || defined(__clang__)
/*
* On GCC & Clang, marking 'dest' as modified will cause the compiler:
* - do not extract the secret from sse registers in the internal loop
* - use less common registers, and avoid pushing these reg into stack
*/
XXH_COMPILER_GUARD(dest);
# endif
XXH_ASSERT(((size_t)src & 31) == 0); /* control alignment */
XXH_ASSERT(((size_t)dest & 31) == 0);
/* GCC -O2 need unroll loop manually */
dest[0] = _mm256_add_epi64(_mm256_stream_load_si256(src+0), seed);
dest[1] = _mm256_add_epi64(_mm256_stream_load_si256(src+1), seed);
dest[2] = _mm256_add_epi64(_mm256_stream_load_si256(src+2), seed);
dest[3] = _mm256_add_epi64(_mm256_stream_load_si256(src+3), seed);
dest[4] = _mm256_add_epi64(_mm256_stream_load_si256(src+4), seed);
dest[5] = _mm256_add_epi64(_mm256_stream_load_si256(src+5), seed);
}
}
#endif
/* x86dispatch always generates SSE2 */
#if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH)
#ifndef XXH_TARGET_SSE2
# define XXH_TARGET_SSE2 /* disable attribute target */
#endif
XXH_FORCE_INLINE XXH_TARGET_SSE2 void
XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc,
const void* XXH_RESTRICT input,
const void* XXH_RESTRICT secret)
{
/* SSE2 is just a half-scale version of the AVX2 version. */
XXH_ASSERT((((size_t)acc) & 15) == 0);
{ __m128i* const xacc = (__m128i *) acc;
/* Unaligned. This is mainly for pointer arithmetic, and because
* _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
const __m128i* const xinput = (const __m128i *) input;
/* Unaligned. This is mainly for pointer arithmetic, and because
* _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
const __m128i* const xsecret = (const __m128i *) secret;
size_t i;
for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
/* data_vec = xinput[i]; */
__m128i const data_vec = _mm_loadu_si128 (xinput+i);
/* key_vec = xsecret[i]; */
__m128i const key_vec = _mm_loadu_si128 (xsecret+i);
/* data_key = data_vec ^ key_vec; */
__m128i const data_key = _mm_xor_si128 (data_vec, key_vec);
/* data_key_lo = data_key >> 32; */
__m128i const data_key_lo = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
/* product = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
__m128i const product = _mm_mul_epu32 (data_key, data_key_lo);
/* xacc[i] += swap(data_vec); */
__m128i const data_swap = _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1,0,3,2));
__m128i const sum = _mm_add_epi64(xacc[i], data_swap);
/* xacc[i] += product; */
xacc[i] = _mm_add_epi64(product, sum);
} }
}
XXH_FORCE_INLINE XXH_TARGET_SSE2 void
XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 15) == 0);
{ __m128i* const xacc = (__m128i*) acc;
/* Unaligned. This is mainly for pointer arithmetic, and because
* _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
const __m128i* const xsecret = (const __m128i *) secret;
const __m128i prime32 = _mm_set1_epi32((int)XXH_PRIME32_1);
size_t i;
for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
/* xacc[i] ^= (xacc[i] >> 47) */
__m128i const acc_vec = xacc[i];
__m128i const shifted = _mm_srli_epi64 (acc_vec, 47);
__m128i const data_vec = _mm_xor_si128 (acc_vec, shifted);
/* xacc[i] ^= xsecret[i]; */
__m128i const key_vec = _mm_loadu_si128 (xsecret+i);
__m128i const data_key = _mm_xor_si128 (data_vec, key_vec);
/* xacc[i] *= XXH_PRIME32_1; */
__m128i const data_key_hi = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
__m128i const prod_lo = _mm_mul_epu32 (data_key, prime32);
__m128i const prod_hi = _mm_mul_epu32 (data_key_hi, prime32);
xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32));
}
}
}
XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
{
XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
(void)(&XXH_writeLE64);
{ int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i);
# if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
/* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */
XXH_ALIGN(16) const xxh_i64 seed64x2[2] = { (xxh_i64)seed64, (xxh_i64)(0U - seed64) };
__m128i const seed = _mm_load_si128((__m128i const*)seed64x2);
# else
__m128i const seed = _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64);
# endif
int i;
const void* const src16 = XXH3_kSecret;
__m128i* dst16 = (__m128i*) customSecret;
# if defined(__GNUC__) || defined(__clang__)
/*
* On GCC & Clang, marking 'dest' as modified will cause the compiler:
* - do not extract the secret from sse registers in the internal loop
* - use less common registers, and avoid pushing these reg into stack
*/
XXH_COMPILER_GUARD(dst16);
# endif
XXH_ASSERT(((size_t)src16 & 15) == 0); /* control alignment */
XXH_ASSERT(((size_t)dst16 & 15) == 0);
for (i=0; i < nbRounds; ++i) {
dst16[i] = _mm_add_epi64(_mm_load_si128((const __m128i *)src16+i), seed);
} }
}
#endif
#if (XXH_VECTOR == XXH_NEON)
/* forward declarations for the scalar routines */
XXH_FORCE_INLINE void
XXH3_scalarRound(void* XXH_RESTRICT acc, void const* XXH_RESTRICT input,
void const* XXH_RESTRICT secret, size_t lane);
XXH_FORCE_INLINE void
XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
void const* XXH_RESTRICT secret, size_t lane);
/*!
* @internal
* @brief The bulk processing loop for NEON.
*
* The NEON code path is actually partially scalar when running on AArch64. This
* is to optimize the pipelining and can have up to 15% speedup depending on the
* CPU, and it also mitigates some GCC codegen issues.
*
* @see XXH3_NEON_LANES for configuring this and details about this optimization.
*/
XXH_FORCE_INLINE void
XXH3_accumulate_512_neon( void* XXH_RESTRICT acc,
const void* XXH_RESTRICT input,
const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 15) == 0);
XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB && XXH3_NEON_LANES % 2 == 0);
{
uint64x2_t* const xacc = (uint64x2_t *) acc;
/* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */
uint8_t const* const xinput = (const uint8_t *) input;
uint8_t const* const xsecret = (const uint8_t *) secret;
size_t i;
/* NEON for the first few lanes (these loops are normally interleaved) */
for (i=0; i < XXH3_NEON_LANES / 2; i++) {
/* data_vec = xinput[i]; */
uint8x16_t data_vec = vld1q_u8(xinput + (i * 16));
/* key_vec = xsecret[i]; */
uint8x16_t key_vec = vld1q_u8(xsecret + (i * 16));
uint64x2_t data_key;
uint32x2_t data_key_lo, data_key_hi;
/* xacc[i] += swap(data_vec); */
uint64x2_t const data64 = vreinterpretq_u64_u8(data_vec);
uint64x2_t const swapped = vextq_u64(data64, data64, 1);
xacc[i] = vaddq_u64 (xacc[i], swapped);
/* data_key = data_vec ^ key_vec; */
data_key = vreinterpretq_u64_u8(veorq_u8(data_vec, key_vec));
/* data_key_lo = (uint32x2_t) (data_key & 0xFFFFFFFF);
* data_key_hi = (uint32x2_t) (data_key >> 32);
* data_key = UNDEFINED; */
XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
/* xacc[i] += (uint64x2_t) data_key_lo * (uint64x2_t) data_key_hi; */
xacc[i] = vmlal_u32 (xacc[i], data_key_lo, data_key_hi);
}
/* Scalar for the remainder. This may be a zero iteration loop. */
for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
XXH3_scalarRound(acc, input, secret, i);
}
}
}
XXH_FORCE_INLINE void
XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 15) == 0);
{ uint64x2_t* xacc = (uint64x2_t*) acc;
uint8_t const* xsecret = (uint8_t const*) secret;
uint32x2_t prime = vdup_n_u32 (XXH_PRIME32_1);
size_t i;
/* NEON for the first few lanes (these loops are normally interleaved) */
for (i=0; i < XXH3_NEON_LANES / 2; i++) {
/* xacc[i] ^= (xacc[i] >> 47); */
uint64x2_t acc_vec = xacc[i];
uint64x2_t shifted = vshrq_n_u64 (acc_vec, 47);
uint64x2_t data_vec = veorq_u64 (acc_vec, shifted);
/* xacc[i] ^= xsecret[i]; */
uint8x16_t key_vec = vld1q_u8 (xsecret + (i * 16));
uint64x2_t data_key = veorq_u64 (data_vec, vreinterpretq_u64_u8(key_vec));
/* xacc[i] *= XXH_PRIME32_1 */
uint32x2_t data_key_lo, data_key_hi;
/* data_key_lo = (uint32x2_t) (xacc[i] & 0xFFFFFFFF);
* data_key_hi = (uint32x2_t) (xacc[i] >> 32);
* xacc[i] = UNDEFINED; */
XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
{ /*
* prod_hi = (data_key >> 32) * XXH_PRIME32_1;
*
* Avoid vmul_u32 + vshll_n_u32 since Clang 6 and 7 will
* incorrectly "optimize" this:
* tmp = vmul_u32(vmovn_u64(a), vmovn_u64(b));
* shifted = vshll_n_u32(tmp, 32);
* to this:
* tmp = "vmulq_u64"(a, b); // no such thing!
* shifted = vshlq_n_u64(tmp, 32);
*
* However, unlike SSE, Clang lacks a 64-bit multiply routine
* for NEON, and it scalarizes two 64-bit multiplies instead.
*
* vmull_u32 has the same timing as vmul_u32, and it avoids
* this bug completely.
* See https://bugs.llvm.org/show_bug.cgi?id=39967
*/
uint64x2_t prod_hi = vmull_u32 (data_key_hi, prime);
/* xacc[i] = prod_hi << 32; */
xacc[i] = vshlq_n_u64(prod_hi, 32);
/* xacc[i] += (prod_hi & 0xFFFFFFFF) * XXH_PRIME32_1; */
xacc[i] = vmlal_u32(xacc[i], data_key_lo, prime);
}
}
/* Scalar for the remainder. This may be a zero iteration loop. */
for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
XXH3_scalarScrambleRound(acc, secret, i);
}
}
}
#endif
#if (XXH_VECTOR == XXH_VSX)
XXH_FORCE_INLINE void
XXH3_accumulate_512_vsx( void* XXH_RESTRICT acc,
const void* XXH_RESTRICT input,
const void* XXH_RESTRICT secret)
{
/* presumed aligned */
unsigned int* const xacc = (unsigned int*) acc;
xxh_u64x2 const* const xinput = (xxh_u64x2 const*) input; /* no alignment restriction */
xxh_u64x2 const* const xsecret = (xxh_u64x2 const*) secret; /* no alignment restriction */
xxh_u64x2 const v32 = { 32, 32 };
size_t i;
for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
/* data_vec = xinput[i]; */
xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + i);
/* key_vec = xsecret[i]; */
xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + i);
xxh_u64x2 const data_key = data_vec ^ key_vec;
/* shuffled = (data_key << 32) | (data_key >> 32); */
xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
/* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */
xxh_u64x2 const product = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
/* acc_vec = xacc[i]; */
xxh_u64x2 acc_vec = (xxh_u64x2)vec_xl(0, xacc + 4 * i);
acc_vec += product;
/* swap high and low halves */
#ifdef __s390x__
acc_vec += vec_permi(data_vec, data_vec, 2);
#else
acc_vec += vec_xxpermdi(data_vec, data_vec, 2);
#endif
/* xacc[i] = acc_vec; */
vec_xst((xxh_u32x4)acc_vec, 0, xacc + 4 * i);
}
}
XXH_FORCE_INLINE void
XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
XXH_ASSERT((((size_t)acc) & 15) == 0);
{ xxh_u64x2* const xacc = (xxh_u64x2*) acc;
const xxh_u64x2* const xsecret = (const xxh_u64x2*) secret;
/* constants */
xxh_u64x2 const v32 = { 32, 32 };
xxh_u64x2 const v47 = { 47, 47 };
xxh_u32x4 const prime = { XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1 };
size_t i;
for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
/* xacc[i] ^= (xacc[i] >> 47); */
xxh_u64x2 const acc_vec = xacc[i];
xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);
/* xacc[i] ^= xsecret[i]; */
xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + i);
xxh_u64x2 const data_key = data_vec ^ key_vec;
/* xacc[i] *= XXH_PRIME32_1 */
/* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime & 0xFFFFFFFF); */
xxh_u64x2 const prod_even = XXH_vec_mule((xxh_u32x4)data_key, prime);
/* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32); */
xxh_u64x2 const prod_odd = XXH_vec_mulo((xxh_u32x4)data_key, prime);
xacc[i] = prod_odd + (prod_even << v32);
} }
}
#endif
/* scalar variants - universal */
/*!
* @internal
* @brief Scalar round for @ref XXH3_accumulate_512_scalar().
*
* This is extracted to its own function because the NEON path uses a combination
* of NEON and scalar.
*/
XXH_FORCE_INLINE void
XXH3_scalarRound(void* XXH_RESTRICT acc,
void const* XXH_RESTRICT input,
void const* XXH_RESTRICT secret,
size_t lane)
{
xxh_u64* xacc = (xxh_u64*) acc;
xxh_u8 const* xinput = (xxh_u8 const*) input;
xxh_u8 const* xsecret = (xxh_u8 const*) secret;
XXH_ASSERT(lane < XXH_ACC_NB);
XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0);
{
xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8);
xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8);
xacc[lane ^ 1] += data_val; /* swap adjacent lanes */
xacc[lane] += XXH_mult32to64(data_key & 0xFFFFFFFF, data_key >> 32);
}
}
/*!
* @internal
* @brief Processes a 64 byte block of data using the scalar path.
*/
XXH_FORCE_INLINE void
XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc,
const void* XXH_RESTRICT input,
const void* XXH_RESTRICT secret)
{
size_t i;
for (i=0; i < XXH_ACC_NB; i++) {
XXH3_scalarRound(acc, input, secret, i);
}
}
/*!
* @internal
* @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar().
*
* This is extracted to its own function because the NEON path uses a combination
* of NEON and scalar.
*/
XXH_FORCE_INLINE void
XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
void const* XXH_RESTRICT secret,
size_t lane)
{
xxh_u64* const xacc = (xxh_u64*) acc; /* presumed aligned */
const xxh_u8* const xsecret = (const xxh_u8*) secret; /* no alignment restriction */
XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0);
XXH_ASSERT(lane < XXH_ACC_NB);
{
xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8);
xxh_u64 acc64 = xacc[lane];
acc64 = XXH_xorshift64(acc64, 47);
acc64 ^= key64;
acc64 *= XXH_PRIME32_1;
xacc[lane] = acc64;
}
}
/*!
* @internal
* @brief Scrambles the accumulators after a large chunk has been read
*/
XXH_FORCE_INLINE void
XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
{
size_t i;
for (i=0; i < XXH_ACC_NB; i++) {
XXH3_scalarScrambleRound(acc, secret, i);
}
}
XXH_FORCE_INLINE void
XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
{
/*
* We need a separate pointer for the hack below,
* which requires a non-const pointer.
* Any decent compiler will optimize this out otherwise.
*/
const xxh_u8* kSecretPtr = XXH3_kSecret;
XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
#if defined(__clang__) && defined(__aarch64__)
/*
* UGLY HACK:
* Clang generates a bunch of MOV/MOVK pairs for aarch64, and they are
* placed sequentially, in order, at the top of the unrolled loop.
*
* While MOVK is great for generating constants (2 cycles for a 64-bit
* constant compared to 4 cycles for LDR), it fights for bandwidth with
* the arithmetic instructions.
*
* I L S
* MOVK
* MOVK
* MOVK
* MOVK
* ADD
* SUB STR
* STR
* By forcing loads from memory (as the asm line causes Clang to assume
* that XXH3_kSecretPtr has been changed), the pipelines are used more
* efficiently:
* I L S
* LDR
* ADD LDR
* SUB STR
* STR
*
* See XXH3_NEON_LANES for details on the pipsline.
*
* XXH3_64bits_withSeed, len == 256, Snapdragon 835
* without hack: 2654.4 MB/s
* with hack: 3202.9 MB/s
*/
XXH_COMPILER_GUARD(kSecretPtr);
#endif
/*
* Note: in debug mode, this overrides the asm optimization
* and Clang will emit MOVK chains again.
*/
XXH_ASSERT(kSecretPtr == XXH3_kSecret);
{ int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
int i;
for (i=0; i < nbRounds; i++) {
/*
* The asm hack causes Clang to assume that kSecretPtr aliases with
* customSecret, and on aarch64, this prevented LDP from merging two
* loads together for free. Putting the loads together before the stores
* properly generates LDP.
*/
xxh_u64 lo = XXH_readLE64(kSecretPtr + 16*i) + seed64;
xxh_u64 hi = XXH_readLE64(kSecretPtr + 16*i + 8) - seed64;
XXH_writeLE64((xxh_u8*)customSecret + 16*i, lo);
XXH_writeLE64((xxh_u8*)customSecret + 16*i + 8, hi);
} }
}
typedef void (*XXH3_f_accumulate_512)(void* XXH_RESTRICT, const void*, const void*);
typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*);
typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64);
#if (XXH_VECTOR == XXH_AVX512)
#define XXH3_accumulate_512 XXH3_accumulate_512_avx512
#define XXH3_scrambleAcc XXH3_scrambleAcc_avx512
#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512
#elif (XXH_VECTOR == XXH_AVX2)
#define XXH3_accumulate_512 XXH3_accumulate_512_avx2
#define XXH3_scrambleAcc XXH3_scrambleAcc_avx2
#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2
#elif (XXH_VECTOR == XXH_SSE2)
#define XXH3_accumulate_512 XXH3_accumulate_512_sse2
#define XXH3_scrambleAcc XXH3_scrambleAcc_sse2
#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2
#elif (XXH_VECTOR == XXH_NEON)
#define XXH3_accumulate_512 XXH3_accumulate_512_neon
#define XXH3_scrambleAcc XXH3_scrambleAcc_neon
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
#elif (XXH_VECTOR == XXH_VSX)
#define XXH3_accumulate_512 XXH3_accumulate_512_vsx
#define XXH3_scrambleAcc XXH3_scrambleAcc_vsx
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
#else /* scalar */
#define XXH3_accumulate_512 XXH3_accumulate_512_scalar
#define XXH3_scrambleAcc XXH3_scrambleAcc_scalar
#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
#endif
#ifndef XXH_PREFETCH_DIST
# ifdef __clang__
# define XXH_PREFETCH_DIST 320
# else
# if (XXH_VECTOR == XXH_AVX512)
# define XXH_PREFETCH_DIST 512
# else
# define XXH_PREFETCH_DIST 384
# endif
# endif /* __clang__ */
#endif /* XXH_PREFETCH_DIST */
/*
* XXH3_accumulate()
* Loops over XXH3_accumulate_512().
* Assumption: nbStripes will not overflow the secret size
*/
XXH_FORCE_INLINE void
XXH3_accumulate( xxh_u64* XXH_RESTRICT acc,
const xxh_u8* XXH_RESTRICT input,
const xxh_u8* XXH_RESTRICT secret,
size_t nbStripes,
XXH3_f_accumulate_512 f_acc512)
{
size_t n;
for (n = 0; n < nbStripes; n++ ) {
const xxh_u8* const in = input + n*XXH_STRIPE_LEN;
XXH_PREFETCH(in + XXH_PREFETCH_DIST);
f_acc512(acc,
in,
secret + n*XXH_SECRET_CONSUME_RATE);
}
}
XXH_FORCE_INLINE void
XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH3_f_accumulate_512 f_acc512,
XXH3_f_scrambleAcc f_scramble)
{
size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock;
size_t const nb_blocks = (len - 1) / block_len;
size_t n;
XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
for (n = 0; n < nb_blocks; n++) {
XXH3_accumulate(acc, input + n*block_len, secret, nbStripesPerBlock, f_acc512);
f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
}
/* last partial block */
XXH_ASSERT(len > XXH_STRIPE_LEN);
{ size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
XXH3_accumulate(acc, input + nb_blocks*block_len, secret, nbStripes, f_acc512);
/* last stripe */
{ const xxh_u8* const p = input + len - XXH_STRIPE_LEN;
#define XXH_SECRET_LASTACC_START 7 /* not aligned on 8, last secret is different from acc & scrambler */
f_acc512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
} }
}
XXH_FORCE_INLINE xxh_u64
XXH3_mix2Accs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret)
{
return XXH3_mul128_fold64(
acc[0] ^ XXH_readLE64(secret),
acc[1] ^ XXH_readLE64(secret+8) );
}
static XXH64_hash_t
XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 start)
{
xxh_u64 result64 = start;
size_t i = 0;
for (i = 0; i < 4; i++) {
result64 += XXH3_mix2Accs(acc+2*i, secret + 16*i);
#if defined(__clang__) /* Clang */ \
&& (defined(__arm__) || defined(__thumb__)) /* ARMv7 */ \
&& (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \
&& !defined(XXH_ENABLE_AUTOVECTORIZE) /* Define to disable */
/*
* UGLY HACK:
* Prevent autovectorization on Clang ARMv7-a. Exact same problem as
* the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b.
* XXH3_64bits, len == 256, Snapdragon 835:
* without hack: 2063.7 MB/s
* with hack: 2560.7 MB/s
*/
XXH_COMPILER_GUARD(result64);
#endif
}
return XXH3_avalanche(result64);
}
#define XXH3_INIT_ACC { XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3, \
XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 }
XXH_FORCE_INLINE XXH64_hash_t
XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len,
const void* XXH_RESTRICT secret, size_t secretSize,
XXH3_f_accumulate_512 f_acc512,
XXH3_f_scrambleAcc f_scramble)
{
XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc512, f_scramble);
/* converge into final hash */
XXH_STATIC_ASSERT(sizeof(acc) == 64);
/* do not align on 8, so that the secret is different from the accumulator */
#define XXH_SECRET_MERGEACCS_START 11
XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
return XXH3_mergeAccs(acc, (const xxh_u8*)secret + XXH_SECRET_MERGEACCS_START, (xxh_u64)len * XXH_PRIME64_1);
}
/*
* It's important for performance to transmit secret's size (when it's static)
* so that the compiler can properly optimize the vectorized loop.
* This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set.
*/
XXH_FORCE_INLINE XXH64_hash_t
XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
{
(void)seed64;
return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate_512, XXH3_scrambleAcc);
}
/*
* It's preferable for performance that XXH3_hashLong is not inlined,
* as it results in a smaller function for small data, easier to the instruction cache.
* Note that inside this no_inline function, we do inline the internal loop,
* and provide a statically defined secret size to allow optimization of vector loop.
*/
XXH_NO_INLINE XXH64_hash_t
XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
{
(void)seed64; (void)secret; (void)secretLen;
return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate_512, XXH3_scrambleAcc);
}
/*
* XXH3_hashLong_64b_withSeed():
* Generate a custom key based on alteration of default XXH3_kSecret with the seed,
* and then use this key for long mode hashing.
*
* This operation is decently fast but nonetheless costs a little bit of time.
* Try to avoid it whenever possible (typically when seed==0).
*
* It's important for performance that XXH3_hashLong is not inlined. Not sure
* why (uop cache maybe?), but the difference is large and easily measurable.
*/
XXH_FORCE_INLINE XXH64_hash_t
XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len,
XXH64_hash_t seed,
XXH3_f_accumulate_512 f_acc512,
XXH3_f_scrambleAcc f_scramble,
XXH3_f_initCustomSecret f_initSec)
{
if (seed == 0)
return XXH3_hashLong_64b_internal(input, len,
XXH3_kSecret, sizeof(XXH3_kSecret),
f_acc512, f_scramble);
{ XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
f_initSec(secret, seed);
return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret),
f_acc512, f_scramble);
}
}
/*
* It's important for performance that XXH3_hashLong is not inlined.
*/
XXH_NO_INLINE XXH64_hash_t
XXH3_hashLong_64b_withSeed(const void* input, size_t len,
XXH64_hash_t seed, const xxh_u8* secret, size_t secretLen)
{
(void)secret; (void)secretLen;
return XXH3_hashLong_64b_withSeed_internal(input, len, seed,
XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
}
typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void* XXH_RESTRICT, size_t,
XXH64_hash_t, const xxh_u8* XXH_RESTRICT, size_t);
XXH_FORCE_INLINE XXH64_hash_t
XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
XXH3_hashLong64_f f_hashLong)
{
XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
/*
* If an action is to be taken if `secretLen` condition is not respected,
* it should be done here.
* For now, it's a contract pre-condition.
* Adding a check and a branch here would cost performance at every hash.
* Also, note that function signature doesn't offer room to return an error.
*/
if (len <= 16)
return XXH3_len_0to16_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
if (len <= 128)
return XXH3_len_17to128_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
if (len <= XXH3_MIDSIZE_MAX)
return XXH3_len_129to240_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
return f_hashLong(input, len, seed64, (const xxh_u8*)secret, secretLen);
}
/* === Public entry point === */
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* input, size_t len)
{
return XXH3_64bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH64_hash_t
XXH3_64bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
{
return XXH3_64bits_internal(input, len, 0, secret, secretSize, XXH3_hashLong_64b_withSecret);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH64_hash_t
XXH3_64bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
{
return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
}
XXH_PUBLIC_API XXH64_hash_t
XXH3_64bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
{
if (len <= XXH3_MIDSIZE_MAX)
return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
return XXH3_hashLong_64b_withSecret(input, len, seed, (const xxh_u8*)secret, secretSize);
}
/* === XXH3 streaming === */
/*
* Malloc's a pointer that is always aligned to align.
*
* This must be freed with `XXH_alignedFree()`.
*
* malloc typically guarantees 16 byte alignment on 64-bit systems and 8 byte
* alignment on 32-bit. This isn't enough for the 32 byte aligned loads in AVX2
* or on 32-bit, the 16 byte aligned loads in SSE2 and NEON.
*
* This underalignment previously caused a rather obvious crash which went
* completely unnoticed due to XXH3_createState() not actually being tested.
* Credit to RedSpah for noticing this bug.
*
* The alignment is done manually: Functions like posix_memalign or _mm_malloc
* are avoided: To maintain portability, we would have to write a fallback
* like this anyways, and besides, testing for the existence of library
* functions without relying on external build tools is impossible.
*
* The method is simple: Overallocate, manually align, and store the offset
* to the original behind the returned pointer.
*
* Align must be a power of 2 and 8 <= align <= 128.
*/
static void* XXH_alignedMalloc(size_t s, size_t align)
{
XXH_ASSERT(align <= 128 && align >= 8); /* range check */
XXH_ASSERT((align & (align-1)) == 0); /* power of 2 */
XXH_ASSERT(s != 0 && s < (s + align)); /* empty/overflow */
{ /* Overallocate to make room for manual realignment and an offset byte */
xxh_u8* base = (xxh_u8*)XXH_malloc(s + align);
if (base != NULL) {
/*
* Get the offset needed to align this pointer.
*
* Even if the returned pointer is aligned, there will always be
* at least one byte to store the offset to the original pointer.
*/
size_t offset = align - ((size_t)base & (align - 1)); /* base % align */
/* Add the offset for the now-aligned pointer */
xxh_u8* ptr = base + offset;
XXH_ASSERT((size_t)ptr % align == 0);
/* Store the offset immediately before the returned pointer. */
ptr[-1] = (xxh_u8)offset;
return ptr;
}
return NULL;
}
}
/*
* Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass
* normal malloc'd pointers, XXH_alignedMalloc has a specific data layout.
*/
static void XXH_alignedFree(void* p)
{
if (p != NULL) {
xxh_u8* ptr = (xxh_u8*)p;
/* Get the offset byte we added in XXH_malloc. */
xxh_u8 offset = ptr[-1];
/* Free the original malloc'd pointer */
xxh_u8* base = ptr - offset;
XXH_free(base);
}
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
{
XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
if (state==NULL) return NULL;
XXH3_INITSTATE(state);
return state;
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr)
{
XXH_alignedFree(statePtr);
return XXH_OK;
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API void
XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state)
{
XXH_memcpy(dst_state, src_state, sizeof(*dst_state));
}
static void
XXH3_reset_internal(XXH3_state_t* statePtr,
XXH64_hash_t seed,
const void* secret, size_t secretSize)
{
size_t const initStart = offsetof(XXH3_state_t, bufferedSize);
size_t const initLength = offsetof(XXH3_state_t, nbStripesPerBlock) - initStart;
XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart);
XXH_ASSERT(statePtr != NULL);
/* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */
memset((char*)statePtr + initStart, 0, initLength);
statePtr->acc[0] = XXH_PRIME32_3;
statePtr->acc[1] = XXH_PRIME64_1;
statePtr->acc[2] = XXH_PRIME64_2;
statePtr->acc[3] = XXH_PRIME64_3;
statePtr->acc[4] = XXH_PRIME64_4;
statePtr->acc[5] = XXH_PRIME32_2;
statePtr->acc[6] = XXH_PRIME64_5;
statePtr->acc[7] = XXH_PRIME32_1;
statePtr->seed = seed;
statePtr->useSeed = (seed != 0);
statePtr->extSecret = (const unsigned char*)secret;
XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
statePtr->secretLimit = secretSize - XXH_STRIPE_LEN;
statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset(XXH3_state_t* statePtr)
{
if (statePtr == NULL) return XXH_ERROR;
XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
return XXH_OK;
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
{
if (statePtr == NULL) return XXH_ERROR;
XXH3_reset_internal(statePtr, 0, secret, secretSize);
if (secret == NULL) return XXH_ERROR;
if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
return XXH_OK;
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
{
if (statePtr == NULL) return XXH_ERROR;
if (seed==0) return XXH3_64bits_reset(statePtr);
if ((seed != statePtr->seed) || (statePtr->extSecret != NULL))
XXH3_initCustomSecret(statePtr->customSecret, seed);
XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
return XXH_OK;
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed64)
{
if (statePtr == NULL) return XXH_ERROR;
if (secret == NULL) return XXH_ERROR;
if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
XXH3_reset_internal(statePtr, seed64, secret, secretSize);
statePtr->useSeed = 1; /* always, even if seed64==0 */
return XXH_OK;
}
/* Note : when XXH3_consumeStripes() is invoked,
* there must be a guarantee that at least one more byte must be consumed from input
* so that the function can blindly consume all stripes using the "normal" secret segment */
XXH_FORCE_INLINE void
XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc,
size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock,
const xxh_u8* XXH_RESTRICT input, size_t nbStripes,
const xxh_u8* XXH_RESTRICT secret, size_t secretLimit,
XXH3_f_accumulate_512 f_acc512,
XXH3_f_scrambleAcc f_scramble)
{
XXH_ASSERT(nbStripes <= nbStripesPerBlock); /* can handle max 1 scramble per invocation */
XXH_ASSERT(*nbStripesSoFarPtr < nbStripesPerBlock);
if (nbStripesPerBlock - *nbStripesSoFarPtr <= nbStripes) {
/* need a scrambling operation */
size_t const nbStripesToEndofBlock = nbStripesPerBlock - *nbStripesSoFarPtr;
size_t const nbStripesAfterBlock = nbStripes - nbStripesToEndofBlock;
XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripesToEndofBlock, f_acc512);
f_scramble(acc, secret + secretLimit);
XXH3_accumulate(acc, input + nbStripesToEndofBlock * XXH_STRIPE_LEN, secret, nbStripesAfterBlock, f_acc512);
*nbStripesSoFarPtr = nbStripesAfterBlock;
} else {
XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripes, f_acc512);
*nbStripesSoFarPtr += nbStripes;
}
}
#ifndef XXH3_STREAM_USE_STACK
# ifndef __clang__ /* clang doesn't need additional stack space */
# define XXH3_STREAM_USE_STACK 1
# endif
#endif
/*
* Both XXH3_64bits_update and XXH3_128bits_update use this routine.
*/
XXH_FORCE_INLINE XXH_errorcode
XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
const xxh_u8* XXH_RESTRICT input, size_t len,
XXH3_f_accumulate_512 f_acc512,
XXH3_f_scrambleAcc f_scramble)
{
if (input==NULL) {
XXH_ASSERT(len == 0);
return XXH_OK;
}
XXH_ASSERT(state != NULL);
{ const xxh_u8* const bEnd = input + len;
const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
/* For some reason, gcc and MSVC seem to suffer greatly
* when operating accumulators directly into state.
* Operating into stack space seems to enable proper optimization.
* clang, on the other hand, doesn't seem to need this trick */
XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8]; memcpy(acc, state->acc, sizeof(acc));
#else
xxh_u64* XXH_RESTRICT const acc = state->acc;
#endif
state->totalLen += len;
XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);
/* small input : just fill in tmp buffer */
if (state->bufferedSize + len <= XXH3_INTERNALBUFFER_SIZE) {
XXH_memcpy(state->buffer + state->bufferedSize, input, len);
state->bufferedSize += (XXH32_hash_t)len;
return XXH_OK;
}
/* total input is now > XXH3_INTERNALBUFFER_SIZE */
#define XXH3_INTERNALBUFFER_STRIPES (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN)
XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN == 0); /* clean multiple */
/*
* Internal buffer is partially filled (always, except at beginning)
* Complete it, then consume it.
*/
if (state->bufferedSize) {
size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize;
XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize);
input += loadSize;
XXH3_consumeStripes(acc,
&state->nbStripesSoFar, state->nbStripesPerBlock,
state->buffer, XXH3_INTERNALBUFFER_STRIPES,
secret, state->secretLimit,
f_acc512, f_scramble);
state->bufferedSize = 0;
}
XXH_ASSERT(input < bEnd);
/* large input to consume : ingest per full block */
if ((size_t)(bEnd - input) > state->nbStripesPerBlock * XXH_STRIPE_LEN) {
size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN;
XXH_ASSERT(state->nbStripesPerBlock >= state->nbStripesSoFar);
/* join to current block's end */
{ size_t const nbStripesToEnd = state->nbStripesPerBlock - state->nbStripesSoFar;
XXH_ASSERT(nbStripesToEnd <= nbStripes);
XXH3_accumulate(acc, input, secret + state->nbStripesSoFar * XXH_SECRET_CONSUME_RATE, nbStripesToEnd, f_acc512);
f_scramble(acc, secret + state->secretLimit);
state->nbStripesSoFar = 0;
input += nbStripesToEnd * XXH_STRIPE_LEN;
nbStripes -= nbStripesToEnd;
}
/* consume per entire blocks */
while(nbStripes >= state->nbStripesPerBlock) {
XXH3_accumulate(acc, input, secret, state->nbStripesPerBlock, f_acc512);
f_scramble(acc, secret + state->secretLimit);
input += state->nbStripesPerBlock * XXH_STRIPE_LEN;
nbStripes -= state->nbStripesPerBlock;
}
/* consume last partial block */
XXH3_accumulate(acc, input, secret, nbStripes, f_acc512);
input += nbStripes * XXH_STRIPE_LEN;
XXH_ASSERT(input < bEnd); /* at least some bytes left */
state->nbStripesSoFar = nbStripes;
/* buffer predecessor of last partial stripe */
XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
XXH_ASSERT(bEnd - input <= XXH_STRIPE_LEN);
} else {
/* content to consume <= block size */
/* Consume input by a multiple of internal buffer size */
if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
const xxh_u8* const limit = bEnd - XXH3_INTERNALBUFFER_SIZE;
do {
XXH3_consumeStripes(acc,
&state->nbStripesSoFar, state->nbStripesPerBlock,
input, XXH3_INTERNALBUFFER_STRIPES,
secret, state->secretLimit,
f_acc512, f_scramble);
input += XXH3_INTERNALBUFFER_SIZE;
} while (input<limit);
/* buffer predecessor of last partial stripe */
XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
}
}
/* Some remaining input (always) : buffer it */
XXH_ASSERT(input < bEnd);
XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE);
XXH_ASSERT(state->bufferedSize == 0);
XXH_memcpy(state->buffer, input, (size_t)(bEnd-input));
state->bufferedSize = (XXH32_hash_t)(bEnd-input);
#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
/* save stack accumulators into state */
memcpy(state->acc, acc, sizeof(acc));
#endif
}
return XXH_OK;
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_64bits_update(XXH3_state_t* state, const void* input, size_t len)
{
return XXH3_update(state, (const xxh_u8*)input, len,
XXH3_accumulate_512, XXH3_scrambleAcc);
}
XXH_FORCE_INLINE void
XXH3_digest_long (XXH64_hash_t* acc,
const XXH3_state_t* state,
const unsigned char* secret)
{
/*
* Digest on a local copy. This way, the state remains unaltered, and it can
* continue ingesting more input afterwards.
*/
XXH_memcpy(acc, state->acc, sizeof(state->acc));
if (state->bufferedSize >= XXH_STRIPE_LEN) {
size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
size_t nbStripesSoFar = state->nbStripesSoFar;
XXH3_consumeStripes(acc,
&nbStripesSoFar, state->nbStripesPerBlock,
state->buffer, nbStripes,
secret, state->secretLimit,
XXH3_accumulate_512, XXH3_scrambleAcc);
/* last stripe */
XXH3_accumulate_512(acc,
state->buffer + state->bufferedSize - XXH_STRIPE_LEN,
secret + state->secretLimit - XXH_SECRET_LASTACC_START);
} else { /* bufferedSize < XXH_STRIPE_LEN */
xxh_u8 lastStripe[XXH_STRIPE_LEN];
size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
XXH_ASSERT(state->bufferedSize > 0); /* there is always some input buffered */
XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize);
XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
XXH3_accumulate_512(acc,
lastStripe,
secret + state->secretLimit - XXH_SECRET_LASTACC_START);
}
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* state)
{
const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
if (state->totalLen > XXH3_MIDSIZE_MAX) {
XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
XXH3_digest_long(acc, state, secret);
return XXH3_mergeAccs(acc,
secret + XXH_SECRET_MERGEACCS_START,
(xxh_u64)state->totalLen * XXH_PRIME64_1);
}
/* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */
if (state->useSeed)
return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
secret, state->secretLimit + XXH_STRIPE_LEN);
}
/* ==========================================
* XXH3 128 bits (a.k.a XXH128)
* ==========================================
* XXH3's 128-bit variant has better mixing and strength than the 64-bit variant,
* even without counting the significantly larger output size.
*
* For example, extra steps are taken to avoid the seed-dependent collisions
* in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B).
*
* This strength naturally comes at the cost of some speed, especially on short
* lengths. Note that longer hashes are about as fast as the 64-bit version
* due to it using only a slight modification of the 64-bit loop.
*
* XXH128 is also more oriented towards 64-bit machines. It is still extremely
* fast for a _128-bit_ hash on 32-bit (it usually clears XXH64).
*/
XXH_FORCE_INLINE XXH128_hash_t
XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
/* A doubled version of 1to3_64b with different constants. */
XXH_ASSERT(input != NULL);
XXH_ASSERT(1 <= len && len <= 3);
XXH_ASSERT(secret != NULL);
/*
* len = 1: combinedl = { input[0], 0x01, input[0], input[0] }
* len = 2: combinedl = { input[1], 0x02, input[0], input[1] }
* len = 3: combinedl = { input[2], 0x03, input[0], input[1] }
*/
{ xxh_u8 const c1 = input[0];
xxh_u8 const c2 = input[len >> 1];
xxh_u8 const c3 = input[len - 1];
xxh_u32 const combinedl = ((xxh_u32)c1 <<16) | ((xxh_u32)c2 << 24)
| ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13);
xxh_u64 const bitflipl = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
xxh_u64 const bitfliph = (XXH_readLE32(secret+8) ^ XXH_readLE32(secret+12)) - seed;
xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl;
xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph;
XXH128_hash_t h128;
h128.low64 = XXH64_avalanche(keyed_lo);
h128.high64 = XXH64_avalanche(keyed_hi);
return h128;
}
}
XXH_FORCE_INLINE XXH128_hash_t
XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
XXH_ASSERT(secret != NULL);
XXH_ASSERT(4 <= len && len <= 8);
seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
{ xxh_u32 const input_lo = XXH_readLE32(input);
xxh_u32 const input_hi = XXH_readLE32(input + len - 4);
xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32);
xxh_u64 const bitflip = (XXH_readLE64(secret+16) ^ XXH_readLE64(secret+24)) + seed;
xxh_u64 const keyed = input_64 ^ bitflip;
/* Shift len to the left to ensure it is even, this avoids even multiplies. */
XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2));
m128.high64 += (m128.low64 << 1);
m128.low64 ^= (m128.high64 >> 3);
m128.low64 = XXH_xorshift64(m128.low64, 35);
m128.low64 *= 0x9FB21C651E98DF25ULL;
m128.low64 = XXH_xorshift64(m128.low64, 28);
m128.high64 = XXH3_avalanche(m128.high64);
return m128;
}
}
XXH_FORCE_INLINE XXH128_hash_t
XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(input != NULL);
XXH_ASSERT(secret != NULL);
XXH_ASSERT(9 <= len && len <= 16);
{ xxh_u64 const bitflipl = (XXH_readLE64(secret+32) ^ XXH_readLE64(secret+40)) - seed;
xxh_u64 const bitfliph = (XXH_readLE64(secret+48) ^ XXH_readLE64(secret+56)) + seed;
xxh_u64 const input_lo = XXH_readLE64(input);
xxh_u64 input_hi = XXH_readLE64(input + len - 8);
XXH128_hash_t m128 = XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1);
/*
* Put len in the middle of m128 to ensure that the length gets mixed to
* both the low and high bits in the 128x64 multiply below.
*/
m128.low64 += (xxh_u64)(len - 1) << 54;
input_hi ^= bitfliph;
/*
* Add the high 32 bits of input_hi to the high 32 bits of m128, then
* add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to
* the high 64 bits of m128.
*
* The best approach to this operation is different on 32-bit and 64-bit.
*/
if (sizeof(void *) < sizeof(xxh_u64)) { /* 32-bit */
/*
* 32-bit optimized version, which is more readable.
*
* On 32-bit, it removes an ADC and delays a dependency between the two
* halves of m128.high64, but it generates an extra mask on 64-bit.
*/
m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2);
} else {
/*
* 64-bit optimized (albeit more confusing) version.
*
* Uses some properties of addition and multiplication to remove the mask:
*
* Let:
* a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF)
* b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000)
* c = XXH_PRIME32_2
*
* a + (b * c)
* Inverse Property: x + y - x == y
* a + (b * (1 + c - 1))
* Distributive Property: x * (y + z) == (x * y) + (x * z)
* a + (b * 1) + (b * (c - 1))
* Identity Property: x * 1 == x
* a + b + (b * (c - 1))
*
* Substitute a, b, and c:
* input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
*
* Since input_hi.hi + input_hi.lo == input_hi, we get this:
* input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
*/
m128.high64 += input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1);
}
/* m128 ^= XXH_swap64(m128 >> 64); */
m128.low64 ^= XXH_swap64(m128.high64);
{ /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */
XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2);
h128.high64 += m128.high64 * XXH_PRIME64_2;
h128.low64 = XXH3_avalanche(h128.low64);
h128.high64 = XXH3_avalanche(h128.high64);
return h128;
} }
}
/*
* Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN
*/
XXH_FORCE_INLINE XXH128_hash_t
XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
{
XXH_ASSERT(len <= 16);
{ if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed);
if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed);
if (len) return XXH3_len_1to3_128b(input, len, secret, seed);
{ XXH128_hash_t h128;
xxh_u64 const bitflipl = XXH_readLE64(secret+64) ^ XXH_readLE64(secret+72);
xxh_u64 const bitfliph = XXH_readLE64(secret+80) ^ XXH_readLE64(secret+88);
h128.low64 = XXH64_avalanche(seed ^ bitflipl);
h128.high64 = XXH64_avalanche( seed ^ bitfliph);
return h128;
} }
}
/*
* A bit slower than XXH3_mix16B, but handles multiply by zero better.
*/
XXH_FORCE_INLINE XXH128_hash_t
XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2,
const xxh_u8* secret, XXH64_hash_t seed)
{
acc.low64 += XXH3_mix16B (input_1, secret+0, seed);
acc.low64 ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8);
acc.high64 += XXH3_mix16B (input_2, secret+16, seed);
acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8);
return acc;
}
XXH_FORCE_INLINE XXH128_hash_t
XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH64_hash_t seed)
{
XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
XXH_ASSERT(16 < len && len <= 128);
{ XXH128_hash_t acc;
acc.low64 = len * XXH_PRIME64_1;
acc.high64 = 0;
if (len > 32) {
if (len > 64) {
if (len > 96) {
acc = XXH128_mix32B(acc, input+48, input+len-64, secret+96, seed);
}
acc = XXH128_mix32B(acc, input+32, input+len-48, secret+64, seed);
}
acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed);
}
acc = XXH128_mix32B(acc, input, input+len-16, secret, seed);
{ XXH128_hash_t h128;
h128.low64 = acc.low64 + acc.high64;
h128.high64 = (acc.low64 * XXH_PRIME64_1)
+ (acc.high64 * XXH_PRIME64_4)
+ ((len - seed) * XXH_PRIME64_2);
h128.low64 = XXH3_avalanche(h128.low64);
h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
return h128;
}
}
}
XXH_NO_INLINE XXH128_hash_t
XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH64_hash_t seed)
{
XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
{ XXH128_hash_t acc;
int const nbRounds = (int)len / 32;
int i;
acc.low64 = len * XXH_PRIME64_1;
acc.high64 = 0;
for (i=0; i<4; i++) {
acc = XXH128_mix32B(acc,
input + (32 * i),
input + (32 * i) + 16,
secret + (32 * i),
seed);
}
acc.low64 = XXH3_avalanche(acc.low64);
acc.high64 = XXH3_avalanche(acc.high64);
XXH_ASSERT(nbRounds >= 4);
for (i=4 ; i < nbRounds; i++) {
acc = XXH128_mix32B(acc,
input + (32 * i),
input + (32 * i) + 16,
secret + XXH3_MIDSIZE_STARTOFFSET + (32 * (i - 4)),
seed);
}
/* last bytes */
acc = XXH128_mix32B(acc,
input + len - 16,
input + len - 32,
secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
0ULL - seed);
{ XXH128_hash_t h128;
h128.low64 = acc.low64 + acc.high64;
h128.high64 = (acc.low64 * XXH_PRIME64_1)
+ (acc.high64 * XXH_PRIME64_4)
+ ((len - seed) * XXH_PRIME64_2);
h128.low64 = XXH3_avalanche(h128.low64);
h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
return h128;
}
}
}
XXH_FORCE_INLINE XXH128_hash_t
XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len,
const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
XXH3_f_accumulate_512 f_acc512,
XXH3_f_scrambleAcc f_scramble)
{
XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc512, f_scramble);
/* converge into final hash */
XXH_STATIC_ASSERT(sizeof(acc) == 64);
XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
{ XXH128_hash_t h128;
h128.low64 = XXH3_mergeAccs(acc,
secret + XXH_SECRET_MERGEACCS_START,
(xxh_u64)len * XXH_PRIME64_1);
h128.high64 = XXH3_mergeAccs(acc,
secret + secretSize
- sizeof(acc) - XXH_SECRET_MERGEACCS_START,
~((xxh_u64)len * XXH_PRIME64_2));
return h128;
}
}
/*
* It's important for performance that XXH3_hashLong is not inlined.
*/
XXH_NO_INLINE XXH128_hash_t
XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64,
const void* XXH_RESTRICT secret, size_t secretLen)
{
(void)seed64; (void)secret; (void)secretLen;
return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret),
XXH3_accumulate_512, XXH3_scrambleAcc);
}
/*
* It's important for performance to pass @secretLen (when it's static)
* to the compiler, so that it can properly optimize the vectorized loop.
*/
XXH_FORCE_INLINE XXH128_hash_t
XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64,
const void* XXH_RESTRICT secret, size_t secretLen)
{
(void)seed64;
return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen,
XXH3_accumulate_512, XXH3_scrambleAcc);
}
XXH_FORCE_INLINE XXH128_hash_t
XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len,
XXH64_hash_t seed64,
XXH3_f_accumulate_512 f_acc512,
XXH3_f_scrambleAcc f_scramble,
XXH3_f_initCustomSecret f_initSec)
{
if (seed64 == 0)
return XXH3_hashLong_128b_internal(input, len,
XXH3_kSecret, sizeof(XXH3_kSecret),
f_acc512, f_scramble);
{ XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
f_initSec(secret, seed64);
return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret),
f_acc512, f_scramble);
}
}
/*
* It's important for performance that XXH3_hashLong is not inlined.
*/
XXH_NO_INLINE XXH128_hash_t
XXH3_hashLong_128b_withSeed(const void* input, size_t len,
XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen)
{
(void)secret; (void)secretLen;
return XXH3_hashLong_128b_withSeed_internal(input, len, seed64,
XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
}
typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t,
XXH64_hash_t, const void* XXH_RESTRICT, size_t);
XXH_FORCE_INLINE XXH128_hash_t
XXH3_128bits_internal(const void* input, size_t len,
XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
XXH3_hashLong128_f f_hl128)
{
XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
/*
* If an action is to be taken if `secret` conditions are not respected,
* it should be done here.
* For now, it's a contract pre-condition.
* Adding a check and a branch here would cost performance at every hash.
*/
if (len <= 16)
return XXH3_len_0to16_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
if (len <= 128)
return XXH3_len_17to128_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
if (len <= XXH3_MIDSIZE_MAX)
return XXH3_len_129to240_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
return f_hl128(input, len, seed64, secret, secretLen);
}
/* === Public XXH128 API === */
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* input, size_t len)
{
return XXH3_128bits_internal(input, len, 0,
XXH3_kSecret, sizeof(XXH3_kSecret),
XXH3_hashLong_128b_default);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH3_128bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
{
return XXH3_128bits_internal(input, len, 0,
(const xxh_u8*)secret, secretSize,
XXH3_hashLong_128b_withSecret);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH3_128bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
{
return XXH3_128bits_internal(input, len, seed,
XXH3_kSecret, sizeof(XXH3_kSecret),
XXH3_hashLong_128b_withSeed);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH3_128bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
{
if (len <= XXH3_MIDSIZE_MAX)
return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH128(const void* input, size_t len, XXH64_hash_t seed)
{
return XXH3_128bits_withSeed(input, len, seed);
}
/* === XXH3 128-bit streaming === */
/*
* All initialization and update functions are identical to 64-bit streaming variant.
* The only difference is the finalization routine.
*/
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset(XXH3_state_t* statePtr)
{
return XXH3_64bits_reset(statePtr);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
{
return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
{
return XXH3_64bits_reset_withSeed(statePtr, seed);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed)
{
return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_128bits_update(XXH3_state_t* state, const void* input, size_t len)
{
return XXH3_update(state, (const xxh_u8*)input, len,
XXH3_accumulate_512, XXH3_scrambleAcc);
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* state)
{
const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
if (state->totalLen > XXH3_MIDSIZE_MAX) {
XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
XXH3_digest_long(acc, state, secret);
XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
{ XXH128_hash_t h128;
h128.low64 = XXH3_mergeAccs(acc,
secret + XXH_SECRET_MERGEACCS_START,
(xxh_u64)state->totalLen * XXH_PRIME64_1);
h128.high64 = XXH3_mergeAccs(acc,
secret + state->secretLimit + XXH_STRIPE_LEN
- sizeof(acc) - XXH_SECRET_MERGEACCS_START,
~((xxh_u64)state->totalLen * XXH_PRIME64_2));
return h128;
}
}
/* len <= XXH3_MIDSIZE_MAX : short code */
if (state->seed)
return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
secret, state->secretLimit + XXH_STRIPE_LEN);
}
/* 128-bit utility functions */
#include <string.h> /* memcmp, memcpy */
/* return : 1 is equal, 0 if different */
/*! @ingroup xxh3_family */
XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
{
/* note : XXH128_hash_t is compact, it has no padding byte */
return !(memcmp(&h1, &h2, sizeof(h1)));
}
/* This prototype is compatible with stdlib's qsort().
* return : >0 if *h128_1 > *h128_2
* <0 if *h128_1 < *h128_2
* =0 if *h128_1 == *h128_2 */
/*! @ingroup xxh3_family */
XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2)
{
XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1;
XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2;
int const hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64);
/* note : bets that, in most cases, hash values are different */
if (hcmp) return hcmp;
return (h1.low64 > h2.low64) - (h2.low64 > h1.low64);
}
/*====== Canonical representation ======*/
/*! @ingroup xxh3_family */
XXH_PUBLIC_API void
XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash)
{
XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
if (XXH_CPU_LITTLE_ENDIAN) {
hash.high64 = XXH_swap64(hash.high64);
hash.low64 = XXH_swap64(hash.low64);
}
XXH_memcpy(dst, &hash.high64, sizeof(hash.high64));
XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64));
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH128_hash_t
XXH128_hashFromCanonical(const XXH128_canonical_t* src)
{
XXH128_hash_t h;
h.high64 = XXH_readBE64(src);
h.low64 = XXH_readBE64(src->digest + 8);
return h;
}
/* ==========================================
* Secret generators
* ==========================================
*/
#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))
XXH_FORCE_INLINE void XXH3_combine16(void* dst, XXH128_hash_t h128)
{
XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 );
XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 );
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API XXH_errorcode
XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize)
{
#if (XXH_DEBUGLEVEL >= 1)
XXH_ASSERT(secretBuffer != NULL);
XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
#else
/* production mode, assert() are disabled */
if (secretBuffer == NULL) return XXH_ERROR;
if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
#endif
if (customSeedSize == 0) {
customSeed = XXH3_kSecret;
customSeedSize = XXH_SECRET_DEFAULT_SIZE;
}
#if (XXH_DEBUGLEVEL >= 1)
XXH_ASSERT(customSeed != NULL);
#else
if (customSeed == NULL) return XXH_ERROR;
#endif
/* Fill secretBuffer with a copy of customSeed - repeat as needed */
{ size_t pos = 0;
while (pos < secretSize) {
size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize);
memcpy((char*)secretBuffer + pos, customSeed, toCopy);
pos += toCopy;
} }
{ size_t const nbSeg16 = secretSize / 16;
size_t n;
XXH128_canonical_t scrambler;
XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0));
for (n=0; n<nbSeg16; n++) {
XXH128_hash_t const h128 = XXH128(&scrambler, sizeof(scrambler), n);
XXH3_combine16((char*)secretBuffer + n*16, h128);
}
/* last segment */
XXH3_combine16((char*)secretBuffer + secretSize - 16, XXH128_hashFromCanonical(&scrambler));
}
return XXH_OK;
}
/*! @ingroup xxh3_family */
XXH_PUBLIC_API void
XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed)
{
XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
XXH3_initCustomSecret(secret, seed);
XXH_ASSERT(secretBuffer != NULL);
memcpy(secretBuffer, secret, XXH_SECRET_DEFAULT_SIZE);
}
/* Pop our optimization override from above */
#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
&& defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
&& defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) /* respect -O0 and -Os */
# pragma GCC pop_options
#endif
#endif /* XXH_NO_LONG_LONG */
#endif /* XXH_NO_XXH3 */
/*!
* @}
*/
#endif /* XXH_IMPLEMENTATION */
#if defined (__cplusplus)
}
#endif
/**** ended inlining xxhash.h ****/
#ifndef ZSTD_NO_TRACE
/**** start inlining zstd_trace.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_TRACE_H
#define ZSTD_TRACE_H
#if defined (__cplusplus)
extern "C" {
#endif
#include <stddef.h>
/* weak symbol support
* For now, enable conservatively:
* - Only GNUC
* - Only ELF
* - Only x86-64, i386 and aarch64
* Also, explicitly disable on platforms known not to work so they aren't
* forgotten in the future.
*/
#if !defined(ZSTD_HAVE_WEAK_SYMBOLS) && \
defined(__GNUC__) && defined(__ELF__) && \
(defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || defined(_M_IX86) || defined(__aarch64__)) && \
!defined(__APPLE__) && !defined(_WIN32) && !defined(__MINGW32__) && \
!defined(__CYGWIN__) && !defined(_AIX)
# define ZSTD_HAVE_WEAK_SYMBOLS 1
#else
# define ZSTD_HAVE_WEAK_SYMBOLS 0
#endif
#if ZSTD_HAVE_WEAK_SYMBOLS
# define ZSTD_WEAK_ATTR __attribute__((__weak__))
#else
# define ZSTD_WEAK_ATTR
#endif
/* Only enable tracing when weak symbols are available. */
#ifndef ZSTD_TRACE
# define ZSTD_TRACE ZSTD_HAVE_WEAK_SYMBOLS
#endif
#if ZSTD_TRACE
struct ZSTD_CCtx_s;
struct ZSTD_DCtx_s;
struct ZSTD_CCtx_params_s;
typedef struct {
/**
* ZSTD_VERSION_NUMBER
*
* This is guaranteed to be the first member of ZSTD_trace.
* Otherwise, this struct is not stable between versions. If
* the version number does not match your expectation, you
* should not interpret the rest of the struct.
*/
unsigned version;
/**
* Non-zero if streaming (de)compression is used.
*/
unsigned streaming;
/**
* The dictionary ID.
*/
unsigned dictionaryID;
/**
* Is the dictionary cold?
* Only set on decompression.
*/
unsigned dictionaryIsCold;
/**
* The dictionary size or zero if no dictionary.
*/
size_t dictionarySize;
/**
* The uncompressed size of the data.
*/
size_t uncompressedSize;
/**
* The compressed size of the data.
*/
size_t compressedSize;
/**
* The fully resolved CCtx parameters (NULL on decompression).
*/
struct ZSTD_CCtx_params_s const* params;
/**
* The ZSTD_CCtx pointer (NULL on decompression).
*/
struct ZSTD_CCtx_s const* cctx;
/**
* The ZSTD_DCtx pointer (NULL on compression).
*/
struct ZSTD_DCtx_s const* dctx;
} ZSTD_Trace;
/**
* A tracing context. It must be 0 when tracing is disabled.
* Otherwise, any non-zero value returned by a tracing begin()
* function is presented to any subsequent calls to end().
*
* Any non-zero value is treated as tracing is enabled and not
* interpreted by the library.
*
* Two possible uses are:
* * A timestamp for when the begin() function was called.
* * A unique key identifying the (de)compression, like the
* address of the [dc]ctx pointer if you need to track
* more information than just a timestamp.
*/
typedef unsigned long long ZSTD_TraceCtx;
/**
* Trace the beginning of a compression call.
* @param cctx The dctx pointer for the compression.
* It can be used as a key to map begin() to end().
* @returns Non-zero if tracing is enabled. The return value is
* passed to ZSTD_trace_compress_end().
*/
ZSTD_WEAK_ATTR ZSTD_TraceCtx ZSTD_trace_compress_begin(
struct ZSTD_CCtx_s const* cctx);
/**
* Trace the end of a compression call.
* @param ctx The return value of ZSTD_trace_compress_begin().
* @param trace The zstd tracing info.
*/
ZSTD_WEAK_ATTR void ZSTD_trace_compress_end(
ZSTD_TraceCtx ctx,
ZSTD_Trace const* trace);
/**
* Trace the beginning of a decompression call.
* @param dctx The dctx pointer for the decompression.
* It can be used as a key to map begin() to end().
* @returns Non-zero if tracing is enabled. The return value is
* passed to ZSTD_trace_compress_end().
*/
ZSTD_WEAK_ATTR ZSTD_TraceCtx ZSTD_trace_decompress_begin(
struct ZSTD_DCtx_s const* dctx);
/**
* Trace the end of a decompression call.
* @param ctx The return value of ZSTD_trace_decompress_begin().
* @param trace The zstd tracing info.
*/
ZSTD_WEAK_ATTR void ZSTD_trace_decompress_end(
ZSTD_TraceCtx ctx,
ZSTD_Trace const* trace);
#endif /* ZSTD_TRACE */
#if defined (__cplusplus)
}
#endif
#endif /* ZSTD_TRACE_H */
/**** ended inlining zstd_trace.h ****/
#else
# define ZSTD_TRACE 0
#endif
#if defined (__cplusplus)
extern "C" {
#endif
/* ---- static assert (debug) --- */
#define ZSTD_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)
#define ZSTD_isError ERR_isError /* for inlining */
#define FSE_isError ERR_isError
#define HUF_isError ERR_isError
/*-*************************************
* shared macros
***************************************/
#undef MIN
#undef MAX
#define MIN(a,b) ((a)<(b) ? (a) : (b))
#define MAX(a,b) ((a)>(b) ? (a) : (b))
#define BOUNDED(min,val,max) (MAX(min,MIN(val,max)))
/*-*************************************
* Common constants
***************************************/
#define ZSTD_OPT_NUM (1<<12)
#define ZSTD_REP_NUM 3 /* number of repcodes */
static UNUSED_ATTR const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
#define KB *(1 <<10)
#define MB *(1 <<20)
#define GB *(1U<<30)
#define BIT7 128
#define BIT6 64
#define BIT5 32
#define BIT4 16
#define BIT1 2
#define BIT0 1
#define ZSTD_WINDOWLOG_ABSOLUTEMIN 10
static UNUSED_ATTR const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
static UNUSED_ATTR const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
#define ZSTD_FRAMEIDSIZE 4 /* magic number size */
#define ZSTD_BLOCKHEADERSIZE 3 /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
static UNUSED_ATTR const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
#define ZSTD_FRAMECHECKSUMSIZE 4
#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */) /* for a non-null block */
#define MIN_LITERALS_FOR_4_STREAMS 6
typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingType_e;
#define LONGNBSEQ 0x7F00
#define MINMATCH 3
#define Litbits 8
#define LitHufLog 11
#define MaxLit ((1<<Litbits) - 1)
#define MaxML 52
#define MaxLL 35
#define DefaultMaxOff 28
#define MaxOff 31
#define MaxSeq MAX(MaxLL, MaxML) /* Assumption : MaxOff < MaxLL,MaxML */
#define MLFSELog 9
#define LLFSELog 9
#define OffFSELog 8
#define MaxFSELog MAX(MAX(MLFSELog, LLFSELog), OffFSELog)
#define MaxMLBits 16
#define MaxLLBits 16
#define ZSTD_MAX_HUF_HEADER_SIZE 128 /* header + <= 127 byte tree description */
/* Each table cannot take more than #symbols * FSELog bits */
#define ZSTD_MAX_FSE_HEADERS_SIZE (((MaxML + 1) * MLFSELog + (MaxLL + 1) * LLFSELog + (MaxOff + 1) * OffFSELog + 7) / 8)
static UNUSED_ATTR const U8 LL_bits[MaxLL+1] = {
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
1, 1, 1, 1, 2, 2, 3, 3,
4, 6, 7, 8, 9,10,11,12,
13,14,15,16
};
static UNUSED_ATTR const S16 LL_defaultNorm[MaxLL+1] = {
4, 3, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 1, 1, 1,
2, 2, 2, 2, 2, 2, 2, 2,
2, 3, 2, 1, 1, 1, 1, 1,
-1,-1,-1,-1
};
#define LL_DEFAULTNORMLOG 6 /* for static allocation */
static UNUSED_ATTR const U32 LL_defaultNormLog = LL_DEFAULTNORMLOG;
static UNUSED_ATTR const U8 ML_bits[MaxML+1] = {
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
1, 1, 1, 1, 2, 2, 3, 3,
4, 4, 5, 7, 8, 9,10,11,
12,13,14,15,16
};
static UNUSED_ATTR const S16 ML_defaultNorm[MaxML+1] = {
1, 4, 3, 2, 2, 2, 2, 2,
2, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1,-1,-1,
-1,-1,-1,-1,-1
};
#define ML_DEFAULTNORMLOG 6 /* for static allocation */
static UNUSED_ATTR const U32 ML_defaultNormLog = ML_DEFAULTNORMLOG;
static UNUSED_ATTR const S16 OF_defaultNorm[DefaultMaxOff+1] = {
1, 1, 1, 1, 1, 1, 2, 2,
2, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
-1,-1,-1,-1,-1
};
#define OF_DEFAULTNORMLOG 5 /* for static allocation */
static UNUSED_ATTR const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
/*-*******************************************
* Shared functions to include for inlining
*********************************************/
static void ZSTD_copy8(void* dst, const void* src) {
#if defined(ZSTD_ARCH_ARM_NEON)
vst1_u8((uint8_t*)dst, vld1_u8((const uint8_t*)src));
#else
ZSTD_memcpy(dst, src, 8);
#endif
}
#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
/* Need to use memmove here since the literal buffer can now be located within
the dst buffer. In circumstances where the op "catches up" to where the
literal buffer is, there can be partial overlaps in this call on the final
copy if the literal is being shifted by less than 16 bytes. */
static void ZSTD_copy16(void* dst, const void* src) {
#if defined(ZSTD_ARCH_ARM_NEON)
vst1q_u8((uint8_t*)dst, vld1q_u8((const uint8_t*)src));
#elif defined(ZSTD_ARCH_X86_SSE2)
_mm_storeu_si128((__m128i*)dst, _mm_loadu_si128((const __m128i*)src));
#elif defined(__clang__)
ZSTD_memmove(dst, src, 16);
#else
/* ZSTD_memmove is not inlined properly by gcc */
BYTE copy16_buf[16];
ZSTD_memcpy(copy16_buf, src, 16);
ZSTD_memcpy(dst, copy16_buf, 16);
#endif
}
#define COPY16(d,s) { ZSTD_copy16(d,s); d+=16; s+=16; }
#define WILDCOPY_OVERLENGTH 32
#define WILDCOPY_VECLEN 16
typedef enum {
ZSTD_no_overlap,
ZSTD_overlap_src_before_dst
/* ZSTD_overlap_dst_before_src, */
} ZSTD_overlap_e;
/*! ZSTD_wildcopy() :
* Custom version of ZSTD_memcpy(), can over read/write up to WILDCOPY_OVERLENGTH bytes (if length==0)
* @param ovtype controls the overlap detection
* - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
* - ZSTD_overlap_src_before_dst: The src and dst may overlap, but they MUST be at least 8 bytes apart.
* The src buffer must be before the dst buffer.
*/
MEM_STATIC FORCE_INLINE_ATTR
void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e const ovtype)
{
ptrdiff_t diff = (BYTE*)dst - (const BYTE*)src;
const BYTE* ip = (const BYTE*)src;
BYTE* op = (BYTE*)dst;
BYTE* const oend = op + length;
if (ovtype == ZSTD_overlap_src_before_dst && diff < WILDCOPY_VECLEN) {
/* Handle short offset copies. */
do {
COPY8(op, ip)
} while (op < oend);
} else {
assert(diff >= WILDCOPY_VECLEN || diff <= -WILDCOPY_VECLEN);
/* Separate out the first COPY16() call because the copy length is
* almost certain to be short, so the branches have different
* probabilities. Since it is almost certain to be short, only do
* one COPY16() in the first call. Then, do two calls per loop since
* at that point it is more likely to have a high trip count.
*/
ZSTD_copy16(op, ip);
if (16 >= length) return;
op += 16;
ip += 16;
do {
COPY16(op, ip);
COPY16(op, ip);
}
while (op < oend);
}
}
MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
{
size_t const length = MIN(dstCapacity, srcSize);
if (length > 0) {
ZSTD_memcpy(dst, src, length);
}
return length;
}
/* define "workspace is too large" as this number of times larger than needed */
#define ZSTD_WORKSPACETOOLARGE_FACTOR 3
/* when workspace is continuously too large
* during at least this number of times,
* context's memory usage is considered wasteful,
* because it's sized to handle a worst case scenario which rarely happens.
* In which case, resize it down to free some memory */
#define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128
/* Controls whether the input/output buffer is buffered or stable. */
typedef enum {
ZSTD_bm_buffered = 0, /* Buffer the input/output */
ZSTD_bm_stable = 1 /* ZSTD_inBuffer/ZSTD_outBuffer is stable */
} ZSTD_bufferMode_e;
/*-*******************************************
* Private declarations
*********************************************/
typedef struct seqDef_s {
U32 offBase; /* offBase == Offset + ZSTD_REP_NUM, or repcode 1,2,3 */
U16 litLength;
U16 mlBase; /* mlBase == matchLength - MINMATCH */
} seqDef;
/* Controls whether seqStore has a single "long" litLength or matchLength. See seqStore_t. */
typedef enum {
ZSTD_llt_none = 0, /* no longLengthType */
ZSTD_llt_literalLength = 1, /* represents a long literal */
ZSTD_llt_matchLength = 2 /* represents a long match */
} ZSTD_longLengthType_e;
typedef struct {
seqDef* sequencesStart;
seqDef* sequences; /* ptr to end of sequences */
BYTE* litStart;
BYTE* lit; /* ptr to end of literals */
BYTE* llCode;
BYTE* mlCode;
BYTE* ofCode;
size_t maxNbSeq;
size_t maxNbLit;
/* longLengthPos and longLengthType to allow us to represent either a single litLength or matchLength
* in the seqStore that has a value larger than U16 (if it exists). To do so, we increment
* the existing value of the litLength or matchLength by 0x10000.
*/
ZSTD_longLengthType_e longLengthType;
U32 longLengthPos; /* Index of the sequence to apply long length modification to */
} seqStore_t;
typedef struct {
U32 litLength;
U32 matchLength;
} ZSTD_sequenceLength;
/**
* Returns the ZSTD_sequenceLength for the given sequences. It handles the decoding of long sequences
* indicated by longLengthPos and longLengthType, and adds MINMATCH back to matchLength.
*/
MEM_STATIC ZSTD_sequenceLength ZSTD_getSequenceLength(seqStore_t const* seqStore, seqDef const* seq)
{
ZSTD_sequenceLength seqLen;
seqLen.litLength = seq->litLength;
seqLen.matchLength = seq->mlBase + MINMATCH;
if (seqStore->longLengthPos == (U32)(seq - seqStore->sequencesStart)) {
if (seqStore->longLengthType == ZSTD_llt_literalLength) {
seqLen.litLength += 0x10000;
}
if (seqStore->longLengthType == ZSTD_llt_matchLength) {
seqLen.matchLength += 0x10000;
}
}
return seqLen;
}
/**
* Contains the compressed frame size and an upper-bound for the decompressed frame size.
* Note: before using `compressedSize`, check for errors using ZSTD_isError().
* similarly, before using `decompressedBound`, check for errors using:
* `decompressedBound != ZSTD_CONTENTSIZE_ERROR`
*/
typedef struct {
size_t nbBlocks;
size_t compressedSize;
unsigned long long decompressedBound;
} ZSTD_frameSizeInfo; /* decompress & legacy */
const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx); /* compress & dictBuilder */
int ZSTD_seqToCodes(const seqStore_t* seqStorePtr); /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
/* ZSTD_invalidateRepCodes() :
* ensures next compression will not use repcodes from previous block.
* Note : only works with regular variant;
* do not use with extDict variant ! */
void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx); /* zstdmt, adaptive_compression (shouldn't get this definition from here) */
typedef struct {
blockType_e blockType;
U32 lastBlock;
U32 origSize;
} blockProperties_t; /* declared here for decompress and fullbench */
/*! ZSTD_getcBlockSize() :
* Provides the size of compressed block from block header `src` */
/* Used by: decompress, fullbench */
size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
blockProperties_t* bpPtr);
/*! ZSTD_decodeSeqHeaders() :
* decode sequence header from src */
/* Used by: zstd_decompress_block, fullbench */
size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
const void* src, size_t srcSize);
/**
* @returns true iff the CPU supports dynamic BMI2 dispatch.
*/
MEM_STATIC int ZSTD_cpuSupportsBmi2(void)
{
ZSTD_cpuid_t cpuid = ZSTD_cpuid();
return ZSTD_cpuid_bmi1(cpuid) && ZSTD_cpuid_bmi2(cpuid);
}
#if defined (__cplusplus)
}
#endif
#endif /* ZSTD_CCOMMON_H_MODULE */
/**** ended inlining zstd_internal.h ****/
/*-****************************************
* Version
******************************************/
unsigned ZSTD_versionNumber(void) { return ZSTD_VERSION_NUMBER; }
const char* ZSTD_versionString(void) { return ZSTD_VERSION_STRING; }
/*-****************************************
* ZSTD Error Management
******************************************/
#undef ZSTD_isError /* defined within zstd_internal.h */
/*! ZSTD_isError() :
* tells if a return value is an error code
* symbol is required for external callers */
unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
/*! ZSTD_getErrorName() :
* provides error code string from function result (useful for debugging) */
const char* ZSTD_getErrorName(size_t code) { return ERR_getErrorName(code); }
/*! ZSTD_getError() :
* convert a `size_t` function result into a proper ZSTD_errorCode enum */
ZSTD_ErrorCode ZSTD_getErrorCode(size_t code) { return ERR_getErrorCode(code); }
/*! ZSTD_getErrorString() :
* provides error code string from enum */
const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString(code); }
/**** ended inlining common/zstd_common.c ****/
/**** start inlining compress/fse_compress.c ****/
/* ******************************************************************
* FSE : Finite State Entropy encoder
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
* - Public forum : https://groups.google.com/forum/#!forum/lz4c
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/* **************************************************************
* Includes
****************************************************************/
/**** skipping file: ../common/compiler.h ****/
/**** skipping file: ../common/mem.h ****/
/**** skipping file: ../common/debug.h ****/
/**** start inlining hist.h ****/
/* ******************************************************************
* hist : Histogram functions
* part of Finite State Entropy project
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
* - Public forum : https://groups.google.com/forum/#!forum/lz4c
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/* --- dependencies --- */
/**** skipping file: ../common/zstd_deps.h ****/
/* --- simple histogram functions --- */
/*! HIST_count():
* Provides the precise count of each byte within a table 'count'.
* 'count' is a table of unsigned int, of minimum size (*maxSymbolValuePtr+1).
* Updates *maxSymbolValuePtr with actual largest symbol value detected.
* @return : count of the most frequent symbol (which isn't identified).
* or an error code, which can be tested using HIST_isError().
* note : if return == srcSize, there is only one symbol.
*/
size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize);
unsigned HIST_isError(size_t code); /**< tells if a return value is an error code */
/* --- advanced histogram functions --- */
#define HIST_WKSP_SIZE_U32 1024
#define HIST_WKSP_SIZE (HIST_WKSP_SIZE_U32 * sizeof(unsigned))
/** HIST_count_wksp() :
* Same as HIST_count(), but using an externally provided scratch buffer.
* Benefit is this function will use very little stack space.
* `workSpace` is a writable buffer which must be 4-bytes aligned,
* `workSpaceSize` must be >= HIST_WKSP_SIZE
*/
size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize,
void* workSpace, size_t workSpaceSize);
/** HIST_countFast() :
* same as HIST_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr.
* This function is unsafe, and will segfault if any value within `src` is `> *maxSymbolValuePtr`
*/
size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize);
/** HIST_countFast_wksp() :
* Same as HIST_countFast(), but using an externally provided scratch buffer.
* `workSpace` is a writable buffer which must be 4-bytes aligned,
* `workSpaceSize` must be >= HIST_WKSP_SIZE
*/
size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize,
void* workSpace, size_t workSpaceSize);
/*! HIST_count_simple() :
* Same as HIST_countFast(), this function is unsafe,
* and will segfault if any value within `src` is `> *maxSymbolValuePtr`.
* It is also a bit slower for large inputs.
* However, it does not need any additional memory (not even on stack).
* @return : count of the most frequent symbol.
* Note this function doesn't produce any error (i.e. it must succeed).
*/
unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize);
/**** ended inlining hist.h ****/
/**** skipping file: ../common/bitstream.h ****/
#define FSE_STATIC_LINKING_ONLY
/**** skipping file: ../common/fse.h ****/
/**** skipping file: ../common/error_private.h ****/
#define ZSTD_DEPS_NEED_MALLOC
#define ZSTD_DEPS_NEED_MATH64
/**** skipping file: ../common/zstd_deps.h ****/
/**** skipping file: ../common/bits.h ****/
/* **************************************************************
* Error Management
****************************************************************/
#define FSE_isError ERR_isError
/* **************************************************************
* Templates
****************************************************************/
/*
designed to be included
for type-specific functions (template emulation in C)
Objective is to write these functions only once, for improved maintenance
*/
/* safety checks */
#ifndef FSE_FUNCTION_EXTENSION
# error "FSE_FUNCTION_EXTENSION must be defined"
#endif
#ifndef FSE_FUNCTION_TYPE
# error "FSE_FUNCTION_TYPE must be defined"
#endif
/* Function names */
#define FSE_CAT(X,Y) X##Y
#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
/* Function templates */
/* FSE_buildCTable_wksp() :
* Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
* wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
* workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
*/
size_t FSE_buildCTable_wksp(FSE_CTable* ct,
const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
void* workSpace, size_t wkspSize)
{
U32 const tableSize = 1 << tableLog;
U32 const tableMask = tableSize - 1;
void* const ptr = ct;
U16* const tableU16 = ( (U16*) ptr) + 2;
void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
U32 const step = FSE_TABLESTEP(tableSize);
U32 const maxSV1 = maxSymbolValue+1;
U16* cumul = (U16*)workSpace; /* size = maxSV1 */
FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSV1+1)); /* size = tableSize */
U32 highThreshold = tableSize-1;
assert(((size_t)workSpace & 1) == 0); /* Must be 2 bytes-aligned */
if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
/* CTable header */
tableU16[-2] = (U16) tableLog;
tableU16[-1] = (U16) maxSymbolValue;
assert(tableLog < 16); /* required for threshold strategy to work */
/* For explanations on how to distribute symbol values over the table :
* https://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
#ifdef __clang_analyzer__
ZSTD_memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize); /* useless initialization, just to keep scan-build happy */
#endif
/* symbol start positions */
{ U32 u;
cumul[0] = 0;
for (u=1; u <= maxSV1; u++) {
if (normalizedCounter[u-1]==-1) { /* Low proba symbol */
cumul[u] = cumul[u-1] + 1;
tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
} else {
assert(normalizedCounter[u-1] >= 0);
cumul[u] = cumul[u-1] + (U16)normalizedCounter[u-1];
assert(cumul[u] >= cumul[u-1]); /* no overflow */
} }
cumul[maxSV1] = (U16)(tableSize+1);
}
/* Spread symbols */
if (highThreshold == tableSize - 1) {
/* Case for no low prob count symbols. Lay down 8 bytes at a time
* to reduce branch misses since we are operating on a small block
*/
BYTE* const spread = tableSymbol + tableSize; /* size = tableSize + 8 (may write beyond tableSize) */
{ U64 const add = 0x0101010101010101ull;
size_t pos = 0;
U64 sv = 0;
U32 s;
for (s=0; s<maxSV1; ++s, sv += add) {
int i;
int const n = normalizedCounter[s];
MEM_write64(spread + pos, sv);
for (i = 8; i < n; i += 8) {
MEM_write64(spread + pos + i, sv);
}
assert(n>=0);
pos += (size_t)n;
}
}
/* Spread symbols across the table. Lack of lowprob symbols means that
* we don't need variable sized inner loop, so we can unroll the loop and
* reduce branch misses.
*/
{ size_t position = 0;
size_t s;
size_t const unroll = 2; /* Experimentally determined optimal unroll */
assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
for (s = 0; s < (size_t)tableSize; s += unroll) {
size_t u;
for (u = 0; u < unroll; ++u) {
size_t const uPosition = (position + (u * step)) & tableMask;
tableSymbol[uPosition] = spread[s + u];
}
position = (position + (unroll * step)) & tableMask;
}
assert(position == 0); /* Must have initialized all positions */
}
} else {
U32 position = 0;
U32 symbol;
for (symbol=0; symbol<maxSV1; symbol++) {
int nbOccurrences;
int const freq = normalizedCounter[symbol];
for (nbOccurrences=0; nbOccurrences<freq; nbOccurrences++) {
tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
position = (position + step) & tableMask;
while (position > highThreshold)
position = (position + step) & tableMask; /* Low proba area */
} }
assert(position==0); /* Must have initialized all positions */
}
/* Build table */
{ U32 u; for (u=0; u<tableSize; u++) {
FSE_FUNCTION_TYPE s = tableSymbol[u]; /* note : static analyzer may not understand tableSymbol is properly initialized */
tableU16[cumul[s]++] = (U16) (tableSize+u); /* TableU16 : sorted by symbol order; gives next state value */
} }
/* Build Symbol Transformation Table */
{ unsigned total = 0;
unsigned s;
for (s=0; s<=maxSymbolValue; s++) {
switch (normalizedCounter[s])
{
case 0:
/* filling nonetheless, for compatibility with FSE_getMaxNbBits() */
symbolTT[s].deltaNbBits = ((tableLog+1) << 16) - (1<<tableLog);
break;
case -1:
case 1:
symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
assert(total <= INT_MAX);
symbolTT[s].deltaFindState = (int)(total - 1);
total ++;
break;
default :
assert(normalizedCounter[s] > 1);
{ U32 const maxBitsOut = tableLog - ZSTD_highbit32 ((U32)normalizedCounter[s]-1);
U32 const minStatePlus = (U32)normalizedCounter[s] << maxBitsOut;
symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
symbolTT[s].deltaFindState = (int)(total - (unsigned)normalizedCounter[s]);
total += (unsigned)normalizedCounter[s];
} } } }
#if 0 /* debug : symbol costs */
DEBUGLOG(5, "\n --- table statistics : ");
{ U32 symbol;
for (symbol=0; symbol<=maxSymbolValue; symbol++) {
DEBUGLOG(5, "%3u: w=%3i, maxBits=%u, fracBits=%.2f",
symbol, normalizedCounter[symbol],
FSE_getMaxNbBits(symbolTT, symbol),
(double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
} }
#endif
return 0;
}
#ifndef FSE_COMMONDEFS_ONLY
/*-**************************************************************
* FSE NCount encoding
****************************************************************/
size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
{
size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog
+ 4 /* bitCount initialized at 4 */
+ 2 /* first two symbols may use one additional bit each */) / 8)
+ 1 /* round up to whole nb bytes */
+ 2 /* additional two bytes for bitstream flush */;
return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND; /* maxSymbolValue==0 ? use default */
}
static size_t
FSE_writeNCount_generic (void* header, size_t headerBufferSize,
const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
unsigned writeIsSafe)
{
BYTE* const ostart = (BYTE*) header;
BYTE* out = ostart;
BYTE* const oend = ostart + headerBufferSize;
int nbBits;
const int tableSize = 1 << tableLog;
int remaining;
int threshold;
U32 bitStream = 0;
int bitCount = 0;
unsigned symbol = 0;
unsigned const alphabetSize = maxSymbolValue + 1;
int previousIs0 = 0;
/* Table Size */
bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
bitCount += 4;
/* Init */
remaining = tableSize+1; /* +1 for extra accuracy */
threshold = tableSize;
nbBits = tableLog+1;
while ((symbol < alphabetSize) && (remaining>1)) { /* stops at 1 */
if (previousIs0) {
unsigned start = symbol;
while ((symbol < alphabetSize) && !normalizedCounter[symbol]) symbol++;
if (symbol == alphabetSize) break; /* incorrect distribution */
while (symbol >= start+24) {
start+=24;
bitStream += 0xFFFFU << bitCount;
if ((!writeIsSafe) && (out > oend-2))
return ERROR(dstSize_tooSmall); /* Buffer overflow */
out[0] = (BYTE) bitStream;
out[1] = (BYTE)(bitStream>>8);
out+=2;
bitStream>>=16;
}
while (symbol >= start+3) {
start+=3;
bitStream += 3 << bitCount;
bitCount += 2;
}
bitStream += (symbol-start) << bitCount;
bitCount += 2;
if (bitCount>16) {
if ((!writeIsSafe) && (out > oend - 2))
return ERROR(dstSize_tooSmall); /* Buffer overflow */
out[0] = (BYTE)bitStream;
out[1] = (BYTE)(bitStream>>8);
out += 2;
bitStream >>= 16;
bitCount -= 16;
} }
{ int count = normalizedCounter[symbol++];
int const max = (2*threshold-1) - remaining;
remaining -= count < 0 ? -count : count;
count++; /* +1 for extra accuracy */
if (count>=threshold)
count += max; /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
bitStream += count << bitCount;
bitCount += nbBits;
bitCount -= (count<max);
previousIs0 = (count==1);
if (remaining<1) return ERROR(GENERIC);
while (remaining<threshold) { nbBits--; threshold>>=1; }
}
if (bitCount>16) {
if ((!writeIsSafe) && (out > oend - 2))
return ERROR(dstSize_tooSmall); /* Buffer overflow */
out[0] = (BYTE)bitStream;
out[1] = (BYTE)(bitStream>>8);
out += 2;
bitStream >>= 16;
bitCount -= 16;
} }
if (remaining != 1)
return ERROR(GENERIC); /* incorrect normalized distribution */
assert(symbol <= alphabetSize);
/* flush remaining bitStream */
if ((!writeIsSafe) && (out > oend - 2))
return ERROR(dstSize_tooSmall); /* Buffer overflow */
out[0] = (BYTE)bitStream;
out[1] = (BYTE)(bitStream>>8);
out+= (bitCount+7) /8;
return (out-ostart);
}
size_t FSE_writeNCount (void* buffer, size_t bufferSize,
const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
{
if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge); /* Unsupported */
if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC); /* Unsupported */
if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1 /* write in buffer is safe */);
}
/*-**************************************************************
* FSE Compression Code
****************************************************************/
/* provides the minimum logSize to safely represent a distribution */
static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
{
U32 minBitsSrc = ZSTD_highbit32((U32)(srcSize)) + 1;
U32 minBitsSymbols = ZSTD_highbit32(maxSymbolValue) + 2;
U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
assert(srcSize > 1); /* Not supported, RLE should be used instead */
return minBits;
}
unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
{
U32 maxBitsSrc = ZSTD_highbit32((U32)(srcSize - 1)) - minus;
U32 tableLog = maxTableLog;
U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
assert(srcSize > 1); /* Not supported, RLE should be used instead */
if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
if (maxBitsSrc < tableLog) tableLog = maxBitsSrc; /* Accuracy can be reduced */
if (minBits > tableLog) tableLog = minBits; /* Need a minimum to safely represent all symbol values */
if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
return tableLog;
}
unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
{
return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
}
/* Secondary normalization method.
To be used when primary method fails. */
static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue, short lowProbCount)
{
short const NOT_YET_ASSIGNED = -2;
U32 s;
U32 distributed = 0;
U32 ToDistribute;
/* Init */
U32 const lowThreshold = (U32)(total >> tableLog);
U32 lowOne = (U32)((total * 3) >> (tableLog + 1));
for (s=0; s<=maxSymbolValue; s++) {
if (count[s] == 0) {
norm[s]=0;
continue;
}
if (count[s] <= lowThreshold) {
norm[s] = lowProbCount;
distributed++;
total -= count[s];
continue;
}
if (count[s] <= lowOne) {
norm[s] = 1;
distributed++;
total -= count[s];
continue;
}
norm[s]=NOT_YET_ASSIGNED;
}
ToDistribute = (1 << tableLog) - distributed;
if (ToDistribute == 0)
return 0;
if ((total / ToDistribute) > lowOne) {
/* risk of rounding to zero */
lowOne = (U32)((total * 3) / (ToDistribute * 2));
for (s=0; s<=maxSymbolValue; s++) {
if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
norm[s] = 1;
distributed++;
total -= count[s];
continue;
} }
ToDistribute = (1 << tableLog) - distributed;
}
if (distributed == maxSymbolValue+1) {
/* all values are pretty poor;
probably incompressible data (should have already been detected);
find max, then give all remaining points to max */
U32 maxV = 0, maxC = 0;
for (s=0; s<=maxSymbolValue; s++)
if (count[s] > maxC) { maxV=s; maxC=count[s]; }
norm[maxV] += (short)ToDistribute;
return 0;
}
if (total == 0) {
/* all of the symbols were low enough for the lowOne or lowThreshold */
for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
if (norm[s] > 0) { ToDistribute--; norm[s]++; }
return 0;
}
{ U64 const vStepLog = 62 - tableLog;
U64 const mid = (1ULL << (vStepLog-1)) - 1;
U64 const rStep = ZSTD_div64((((U64)1<<vStepLog) * ToDistribute) + mid, (U32)total); /* scale on remaining */
U64 tmpTotal = mid;
for (s=0; s<=maxSymbolValue; s++) {
if (norm[s]==NOT_YET_ASSIGNED) {
U64 const end = tmpTotal + (count[s] * rStep);
U32 const sStart = (U32)(tmpTotal >> vStepLog);
U32 const sEnd = (U32)(end >> vStepLog);
U32 const weight = sEnd - sStart;
if (weight < 1)
return ERROR(GENERIC);
norm[s] = (short)weight;
tmpTotal = end;
} } }
return 0;
}
size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
const unsigned* count, size_t total,
unsigned maxSymbolValue, unsigned useLowProbCount)
{
/* Sanity checks */
if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC); /* Unsupported size */
if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge); /* Unsupported size */
if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC); /* Too small tableLog, compression potentially impossible */
{ static U32 const rtbTable[] = { 0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
short const lowProbCount = useLowProbCount ? -1 : 1;
U64 const scale = 62 - tableLog;
U64 const step = ZSTD_div64((U64)1<<62, (U32)total); /* <== here, one division ! */
U64 const vStep = 1ULL<<(scale-20);
int stillToDistribute = 1<<tableLog;
unsigned s;
unsigned largest=0;
short largestP=0;
U32 lowThreshold = (U32)(total >> tableLog);
for (s=0; s<=maxSymbolValue; s++) {
if (count[s] == total) return 0; /* rle special case */
if (count[s] == 0) { normalizedCounter[s]=0; continue; }
if (count[s] <= lowThreshold) {
normalizedCounter[s] = lowProbCount;
stillToDistribute--;
} else {
short proba = (short)((count[s]*step) >> scale);
if (proba<8) {
U64 restToBeat = vStep * rtbTable[proba];
proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
}
if (proba > largestP) { largestP=proba; largest=s; }
normalizedCounter[s] = proba;
stillToDistribute -= proba;
} }
if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
/* corner case, need another normalization method */
size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue, lowProbCount);
if (FSE_isError(errorCode)) return errorCode;
}
else normalizedCounter[largest] += (short)stillToDistribute;
}
#if 0
{ /* Print Table (debug) */
U32 s;
U32 nTotal = 0;
for (s=0; s<=maxSymbolValue; s++)
RAWLOG(2, "%3i: %4i \n", s, normalizedCounter[s]);
for (s=0; s<=maxSymbolValue; s++)
nTotal += abs(normalizedCounter[s]);
if (nTotal != (1U<<tableLog))
RAWLOG(2, "Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
getchar();
}
#endif
return tableLog;
}
/* fake FSE_CTable, for rle input (always same symbol) */
size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
{
void* ptr = ct;
U16* tableU16 = ( (U16*) ptr) + 2;
void* FSCTptr = (U32*)ptr + 2;
FSE_symbolCompressionTransform* symbolTT = (FSE_symbolCompressionTransform*) FSCTptr;
/* header */
tableU16[-2] = (U16) 0;
tableU16[-1] = (U16) symbolValue;
/* Build table */
tableU16[0] = 0;
tableU16[1] = 0; /* just in case */
/* Build Symbol Transformation Table */
symbolTT[symbolValue].deltaNbBits = 0;
symbolTT[symbolValue].deltaFindState = 0;
return 0;
}
static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
const void* src, size_t srcSize,
const FSE_CTable* ct, const unsigned fast)
{
const BYTE* const istart = (const BYTE*) src;
const BYTE* const iend = istart + srcSize;
const BYTE* ip=iend;
BIT_CStream_t bitC;
FSE_CState_t CState1, CState2;
/* init */
if (srcSize <= 2) return 0;
{ size_t const initError = BIT_initCStream(&bitC, dst, dstSize);
if (FSE_isError(initError)) return 0; /* not enough space available to write a bitstream */ }
#define FSE_FLUSHBITS(s) (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
if (srcSize & 1) {
FSE_initCState2(&CState1, ct, *--ip);
FSE_initCState2(&CState2, ct, *--ip);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
FSE_FLUSHBITS(&bitC);
} else {
FSE_initCState2(&CState2, ct, *--ip);
FSE_initCState2(&CState1, ct, *--ip);
}
/* join to mod 4 */
srcSize -= 2;
if ((sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) && (srcSize & 2)) { /* test bit 2 */
FSE_encodeSymbol(&bitC, &CState2, *--ip);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
FSE_FLUSHBITS(&bitC);
}
/* 2 or 4 encoding per loop */
while ( ip>istart ) {
FSE_encodeSymbol(&bitC, &CState2, *--ip);
if (sizeof(bitC.bitContainer)*8 < FSE_MAX_TABLELOG*2+7 ) /* this test must be static */
FSE_FLUSHBITS(&bitC);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
if (sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) { /* this test must be static */
FSE_encodeSymbol(&bitC, &CState2, *--ip);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
}
FSE_FLUSHBITS(&bitC);
}
FSE_flushCState(&bitC, &CState2);
FSE_flushCState(&bitC, &CState1);
return BIT_closeCStream(&bitC);
}
size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
const void* src, size_t srcSize,
const FSE_CTable* ct)
{
unsigned const fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
if (fast)
return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 1);
else
return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 0);
}
size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
#endif /* FSE_COMMONDEFS_ONLY */
/**** ended inlining compress/fse_compress.c ****/
/**** start inlining compress/hist.c ****/
/* ******************************************************************
* hist : Histogram functions
* part of Finite State Entropy project
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
* - Public forum : https://groups.google.com/forum/#!forum/lz4c
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/* --- dependencies --- */
/**** skipping file: ../common/mem.h ****/
/**** skipping file: ../common/debug.h ****/
/**** skipping file: ../common/error_private.h ****/
/**** skipping file: hist.h ****/
/* --- Error management --- */
unsigned HIST_isError(size_t code) { return ERR_isError(code); }
/*-**************************************************************
* Histogram functions
****************************************************************/
unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize)
{
const BYTE* ip = (const BYTE*)src;
const BYTE* const end = ip + srcSize;
unsigned maxSymbolValue = *maxSymbolValuePtr;
unsigned largestCount=0;
ZSTD_memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
while (ip<end) {
assert(*ip <= maxSymbolValue);
count[*ip++]++;
}
while (!count[maxSymbolValue]) maxSymbolValue--;
*maxSymbolValuePtr = maxSymbolValue;
{ U32 s;
for (s=0; s<=maxSymbolValue; s++)
if (count[s] > largestCount) largestCount = count[s];
}
return largestCount;
}
typedef enum { trustInput, checkMaxSymbolValue } HIST_checkInput_e;
/* HIST_count_parallel_wksp() :
* store histogram into 4 intermediate tables, recombined at the end.
* this design makes better use of OoO cpus,
* and is noticeably faster when some values are heavily repeated.
* But it needs some additional workspace for intermediate tables.
* `workSpace` must be a U32 table of size >= HIST_WKSP_SIZE_U32.
* @return : largest histogram frequency,
* or an error code (notably when histogram's alphabet is larger than *maxSymbolValuePtr) */
static size_t HIST_count_parallel_wksp(
unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize,
HIST_checkInput_e check,
U32* const workSpace)
{
const BYTE* ip = (const BYTE*)source;
const BYTE* const iend = ip+sourceSize;
size_t const countSize = (*maxSymbolValuePtr + 1) * sizeof(*count);
unsigned max=0;
U32* const Counting1 = workSpace;
U32* const Counting2 = Counting1 + 256;
U32* const Counting3 = Counting2 + 256;
U32* const Counting4 = Counting3 + 256;
/* safety checks */
assert(*maxSymbolValuePtr <= 255);
if (!sourceSize) {
ZSTD_memset(count, 0, countSize);
*maxSymbolValuePtr = 0;
return 0;
}
ZSTD_memset(workSpace, 0, 4*256*sizeof(unsigned));
/* by stripes of 16 bytes */
{ U32 cached = MEM_read32(ip); ip += 4;
while (ip < iend-15) {
U32 c = cached; cached = MEM_read32(ip); ip += 4;
Counting1[(BYTE) c ]++;
Counting2[(BYTE)(c>>8) ]++;
Counting3[(BYTE)(c>>16)]++;
Counting4[ c>>24 ]++;
c = cached; cached = MEM_read32(ip); ip += 4;
Counting1[(BYTE) c ]++;
Counting2[(BYTE)(c>>8) ]++;
Counting3[(BYTE)(c>>16)]++;
Counting4[ c>>24 ]++;
c = cached; cached = MEM_read32(ip); ip += 4;
Counting1[(BYTE) c ]++;
Counting2[(BYTE)(c>>8) ]++;
Counting3[(BYTE)(c>>16)]++;
Counting4[ c>>24 ]++;
c = cached; cached = MEM_read32(ip); ip += 4;
Counting1[(BYTE) c ]++;
Counting2[(BYTE)(c>>8) ]++;
Counting3[(BYTE)(c>>16)]++;
Counting4[ c>>24 ]++;
}
ip-=4;
}
/* finish last symbols */
while (ip<iend) Counting1[*ip++]++;
{ U32 s;
for (s=0; s<256; s++) {
Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
if (Counting1[s] > max) max = Counting1[s];
} }
{ unsigned maxSymbolValue = 255;
while (!Counting1[maxSymbolValue]) maxSymbolValue--;
if (check && maxSymbolValue > *maxSymbolValuePtr) return ERROR(maxSymbolValue_tooSmall);
*maxSymbolValuePtr = maxSymbolValue;
ZSTD_memmove(count, Counting1, countSize); /* in case count & Counting1 are overlapping */
}
return (size_t)max;
}
/* HIST_countFast_wksp() :
* Same as HIST_countFast(), but using an externally provided scratch buffer.
* `workSpace` is a writable buffer which must be 4-bytes aligned,
* `workSpaceSize` must be >= HIST_WKSP_SIZE
*/
size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize,
void* workSpace, size_t workSpaceSize)
{
if (sourceSize < 1500) /* heuristic threshold */
return HIST_count_simple(count, maxSymbolValuePtr, source, sourceSize);
if ((size_t)workSpace & 3) return ERROR(GENERIC); /* must be aligned on 4-bytes boundaries */
if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, trustInput, (U32*)workSpace);
}
/* HIST_count_wksp() :
* Same as HIST_count(), but using an externally provided scratch buffer.
* `workSpace` size must be table of >= HIST_WKSP_SIZE_U32 unsigned */
size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize,
void* workSpace, size_t workSpaceSize)
{
if ((size_t)workSpace & 3) return ERROR(GENERIC); /* must be aligned on 4-bytes boundaries */
if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
if (*maxSymbolValuePtr < 255)
return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, checkMaxSymbolValue, (U32*)workSpace);
*maxSymbolValuePtr = 255;
return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace, workSpaceSize);
}
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize)
{
unsigned tmpCounters[HIST_WKSP_SIZE_U32];
return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters, sizeof(tmpCounters));
}
size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize)
{
unsigned tmpCounters[HIST_WKSP_SIZE_U32];
return HIST_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters, sizeof(tmpCounters));
}
#endif
/**** ended inlining compress/hist.c ****/
/**** start inlining compress/huf_compress.c ****/
/* ******************************************************************
* Huffman encoder, part of New Generation Entropy library
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
* - Public forum : https://groups.google.com/forum/#!forum/lz4c
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/* **************************************************************
* Compiler specifics
****************************************************************/
#ifdef _MSC_VER /* Visual Studio */
# pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
#endif
/* **************************************************************
* Includes
****************************************************************/
/**** skipping file: ../common/zstd_deps.h ****/
/**** skipping file: ../common/compiler.h ****/
/**** skipping file: ../common/bitstream.h ****/
/**** skipping file: hist.h ****/
#define FSE_STATIC_LINKING_ONLY /* FSE_optimalTableLog_internal */
/**** skipping file: ../common/fse.h ****/
/**** skipping file: ../common/huf.h ****/
/**** skipping file: ../common/error_private.h ****/
/**** skipping file: ../common/bits.h ****/
/* **************************************************************
* Error Management
****************************************************************/
#define HUF_isError ERR_isError
#define HUF_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c) /* use only *after* variable declarations */
/* **************************************************************
* Required declarations
****************************************************************/
typedef struct nodeElt_s {
U32 count;
U16 parent;
BYTE byte;
BYTE nbBits;
} nodeElt;
/* **************************************************************
* Debug Traces
****************************************************************/
#if DEBUGLEVEL >= 2
static size_t showU32(const U32* arr, size_t size)
{
size_t u;
for (u=0; u<size; u++) {
RAWLOG(6, " %u", arr[u]); (void)arr;
}
RAWLOG(6, " \n");
return size;
}
static size_t HUF_getNbBits(HUF_CElt elt);
static size_t showCTableBits(const HUF_CElt* ctable, size_t size)
{
size_t u;
for (u=0; u<size; u++) {
RAWLOG(6, " %zu", HUF_getNbBits(ctable[u])); (void)ctable;
}
RAWLOG(6, " \n");
return size;
}
static size_t showHNodeSymbols(const nodeElt* hnode, size_t size)
{
size_t u;
for (u=0; u<size; u++) {
RAWLOG(6, " %u", hnode[u].byte); (void)hnode;
}
RAWLOG(6, " \n");
return size;
}
static size_t showHNodeBits(const nodeElt* hnode, size_t size)
{
size_t u;
for (u=0; u<size; u++) {
RAWLOG(6, " %u", hnode[u].nbBits); (void)hnode;
}
RAWLOG(6, " \n");
return size;
}
#endif
/* *******************************************************
* HUF : Huffman block compression
*********************************************************/
#define HUF_WORKSPACE_MAX_ALIGNMENT 8
static void* HUF_alignUpWorkspace(void* workspace, size_t* workspaceSizePtr, size_t align)
{
size_t const mask = align - 1;
size_t const rem = (size_t)workspace & mask;
size_t const add = (align - rem) & mask;
BYTE* const aligned = (BYTE*)workspace + add;
assert((align & (align - 1)) == 0); /* pow 2 */
assert(align <= HUF_WORKSPACE_MAX_ALIGNMENT);
if (*workspaceSizePtr >= add) {
assert(add < align);
assert(((size_t)aligned & mask) == 0);
*workspaceSizePtr -= add;
return aligned;
} else {
*workspaceSizePtr = 0;
return NULL;
}
}
/* HUF_compressWeights() :
* Same as FSE_compress(), but dedicated to huff0's weights compression.
* The use case needs much less stack memory.
* Note : all elements within weightTable are supposed to be <= HUF_TABLELOG_MAX.
*/
#define MAX_FSE_TABLELOG_FOR_HUFF_HEADER 6
typedef struct {
FSE_CTable CTable[FSE_CTABLE_SIZE_U32(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, HUF_TABLELOG_MAX)];
U32 scratchBuffer[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(HUF_TABLELOG_MAX, MAX_FSE_TABLELOG_FOR_HUFF_HEADER)];
unsigned count[HUF_TABLELOG_MAX+1];
S16 norm[HUF_TABLELOG_MAX+1];
} HUF_CompressWeightsWksp;
static size_t
HUF_compressWeights(void* dst, size_t dstSize,
const void* weightTable, size_t wtSize,
void* workspace, size_t workspaceSize)
{
BYTE* const ostart = (BYTE*) dst;
BYTE* op = ostart;
BYTE* const oend = ostart + dstSize;
unsigned maxSymbolValue = HUF_TABLELOG_MAX;
U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
HUF_CompressWeightsWksp* wksp = (HUF_CompressWeightsWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
if (workspaceSize < sizeof(HUF_CompressWeightsWksp)) return ERROR(GENERIC);
/* init conditions */
if (wtSize <= 1) return 0; /* Not compressible */
/* Scan input and build symbol stats */
{ unsigned const maxCount = HIST_count_simple(wksp->count, &maxSymbolValue, weightTable, wtSize); /* never fails */
if (maxCount == wtSize) return 1; /* only a single symbol in src : rle */
if (maxCount == 1) return 0; /* each symbol present maximum once => not compressible */
}
tableLog = FSE_optimalTableLog(tableLog, wtSize, maxSymbolValue);
CHECK_F( FSE_normalizeCount(wksp->norm, tableLog, wksp->count, wtSize, maxSymbolValue, /* useLowProbCount */ 0) );
/* Write table description header */
{ CHECK_V_F(hSize, FSE_writeNCount(op, (size_t)(oend-op), wksp->norm, maxSymbolValue, tableLog) );
op += hSize;
}
/* Compress */
CHECK_F( FSE_buildCTable_wksp(wksp->CTable, wksp->norm, maxSymbolValue, tableLog, wksp->scratchBuffer, sizeof(wksp->scratchBuffer)) );
{ CHECK_V_F(cSize, FSE_compress_usingCTable(op, (size_t)(oend - op), weightTable, wtSize, wksp->CTable) );
if (cSize == 0) return 0; /* not enough space for compressed data */
op += cSize;
}
return (size_t)(op-ostart);
}
static size_t HUF_getNbBits(HUF_CElt elt)
{
return elt & 0xFF;
}
static size_t HUF_getNbBitsFast(HUF_CElt elt)
{
return elt;
}
static size_t HUF_getValue(HUF_CElt elt)
{
return elt & ~(size_t)0xFF;
}
static size_t HUF_getValueFast(HUF_CElt elt)
{
return elt;
}
static void HUF_setNbBits(HUF_CElt* elt, size_t nbBits)
{
assert(nbBits <= HUF_TABLELOG_ABSOLUTEMAX);
*elt = nbBits;
}
static void HUF_setValue(HUF_CElt* elt, size_t value)
{
size_t const nbBits = HUF_getNbBits(*elt);
if (nbBits > 0) {
assert((value >> nbBits) == 0);
*elt |= value << (sizeof(HUF_CElt) * 8 - nbBits);
}
}
HUF_CTableHeader HUF_readCTableHeader(HUF_CElt const* ctable)
{
HUF_CTableHeader header;
ZSTD_memcpy(&header, ctable, sizeof(header));
return header;
}
static void HUF_writeCTableHeader(HUF_CElt* ctable, U32 tableLog, U32 maxSymbolValue)
{
HUF_CTableHeader header;
HUF_STATIC_ASSERT(sizeof(ctable[0]) == sizeof(header));
ZSTD_memset(&header, 0, sizeof(header));
assert(tableLog < 256);
header.tableLog = (BYTE)tableLog;
assert(maxSymbolValue < 256);
header.maxSymbolValue = (BYTE)maxSymbolValue;
ZSTD_memcpy(ctable, &header, sizeof(header));
}
typedef struct {
HUF_CompressWeightsWksp wksp;
BYTE bitsToWeight[HUF_TABLELOG_MAX + 1]; /* precomputed conversion table */
BYTE huffWeight[HUF_SYMBOLVALUE_MAX];
} HUF_WriteCTableWksp;
size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog,
void* workspace, size_t workspaceSize)
{
HUF_CElt const* const ct = CTable + 1;
BYTE* op = (BYTE*)dst;
U32 n;
HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE >= sizeof(HUF_WriteCTableWksp));
assert(HUF_readCTableHeader(CTable).maxSymbolValue == maxSymbolValue);
assert(HUF_readCTableHeader(CTable).tableLog == huffLog);
/* check conditions */
if (workspaceSize < sizeof(HUF_WriteCTableWksp)) return ERROR(GENERIC);
if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
/* convert to weight */
wksp->bitsToWeight[0] = 0;
for (n=1; n<huffLog+1; n++)
wksp->bitsToWeight[n] = (BYTE)(huffLog + 1 - n);
for (n=0; n<maxSymbolValue; n++)
wksp->huffWeight[n] = wksp->bitsToWeight[HUF_getNbBits(ct[n])];
/* attempt weights compression by FSE */
if (maxDstSize < 1) return ERROR(dstSize_tooSmall);
{ CHECK_V_F(hSize, HUF_compressWeights(op+1, maxDstSize-1, wksp->huffWeight, maxSymbolValue, &wksp->wksp, sizeof(wksp->wksp)) );
if ((hSize>1) & (hSize < maxSymbolValue/2)) { /* FSE compressed */
op[0] = (BYTE)hSize;
return hSize+1;
} }
/* write raw values as 4-bits (max : 15) */
if (maxSymbolValue > (256-128)) return ERROR(GENERIC); /* should not happen : likely means source cannot be compressed */
if (((maxSymbolValue+1)/2) + 1 > maxDstSize) return ERROR(dstSize_tooSmall); /* not enough space within dst buffer */
op[0] = (BYTE)(128 /*special case*/ + (maxSymbolValue-1));
wksp->huffWeight[maxSymbolValue] = 0; /* to be sure it doesn't cause msan issue in final combination */
for (n=0; n<maxSymbolValue; n+=2)
op[(n/2)+1] = (BYTE)((wksp->huffWeight[n] << 4) + wksp->huffWeight[n+1]);
return ((maxSymbolValue+1)/2) + 1;
}
size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* hasZeroWeights)
{
BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1]; /* init not required, even though some static analyzer may complain */
U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1]; /* large enough for values from 0 to 16 */
U32 tableLog = 0;
U32 nbSymbols = 0;
HUF_CElt* const ct = CTable + 1;
/* get symbol weights */
CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
*hasZeroWeights = (rankVal[0] > 0);
/* check result */
if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
*maxSymbolValuePtr = nbSymbols - 1;
HUF_writeCTableHeader(CTable, tableLog, *maxSymbolValuePtr);
/* Prepare base value per rank */
{ U32 n, nextRankStart = 0;
for (n=1; n<=tableLog; n++) {
U32 curr = nextRankStart;
nextRankStart += (rankVal[n] << (n-1));
rankVal[n] = curr;
} }
/* fill nbBits */
{ U32 n; for (n=0; n<nbSymbols; n++) {
const U32 w = huffWeight[n];
HUF_setNbBits(ct + n, (BYTE)(tableLog + 1 - w) & -(w != 0));
} }
/* fill val */
{ U16 nbPerRank[HUF_TABLELOG_MAX+2] = {0}; /* support w=0=>n=tableLog+1 */
U16 valPerRank[HUF_TABLELOG_MAX+2] = {0};
{ U32 n; for (n=0; n<nbSymbols; n++) nbPerRank[HUF_getNbBits(ct[n])]++; }
/* determine stating value per rank */
valPerRank[tableLog+1] = 0; /* for w==0 */
{ U16 min = 0;
U32 n; for (n=tableLog; n>0; n--) { /* start at n=tablelog <-> w=1 */
valPerRank[n] = min; /* get starting value within each rank */
min += nbPerRank[n];
min >>= 1;
} }
/* assign value within rank, symbol order */
{ U32 n; for (n=0; n<nbSymbols; n++) HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++); }
}
return readSize;
}
U32 HUF_getNbBitsFromCTable(HUF_CElt const* CTable, U32 symbolValue)
{
const HUF_CElt* const ct = CTable + 1;
assert(symbolValue <= HUF_SYMBOLVALUE_MAX);
if (symbolValue > HUF_readCTableHeader(CTable).maxSymbolValue)
return 0;
return (U32)HUF_getNbBits(ct[symbolValue]);
}
/**
* HUF_setMaxHeight():
* Try to enforce @targetNbBits on the Huffman tree described in @huffNode.
*
* It attempts to convert all nodes with nbBits > @targetNbBits
* to employ @targetNbBits instead. Then it adjusts the tree
* so that it remains a valid canonical Huffman tree.
*
* @pre The sum of the ranks of each symbol == 2^largestBits,
* where largestBits == huffNode[lastNonNull].nbBits.
* @post The sum of the ranks of each symbol == 2^largestBits,
* where largestBits is the return value (expected <= targetNbBits).
*
* @param huffNode The Huffman tree modified in place to enforce targetNbBits.
* It's presumed sorted, from most frequent to rarest symbol.
* @param lastNonNull The symbol with the lowest count in the Huffman tree.
* @param targetNbBits The allowed number of bits, which the Huffman tree
* may not respect. After this function the Huffman tree will
* respect targetNbBits.
* @return The maximum number of bits of the Huffman tree after adjustment.
*/
static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 targetNbBits)
{
const U32 largestBits = huffNode[lastNonNull].nbBits;
/* early exit : no elt > targetNbBits, so the tree is already valid. */
if (largestBits <= targetNbBits) return largestBits;
DEBUGLOG(5, "HUF_setMaxHeight (targetNbBits = %u)", targetNbBits);
/* there are several too large elements (at least >= 2) */
{ int totalCost = 0;
const U32 baseCost = 1 << (largestBits - targetNbBits);
int n = (int)lastNonNull;
/* Adjust any ranks > targetNbBits to targetNbBits.
* Compute totalCost, which is how far the sum of the ranks is
* we are over 2^largestBits after adjust the offending ranks.
*/
while (huffNode[n].nbBits > targetNbBits) {
totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
huffNode[n].nbBits = (BYTE)targetNbBits;
n--;
}
/* n stops at huffNode[n].nbBits <= targetNbBits */
assert(huffNode[n].nbBits <= targetNbBits);
/* n end at index of smallest symbol using < targetNbBits */
while (huffNode[n].nbBits == targetNbBits) --n;
/* renorm totalCost from 2^largestBits to 2^targetNbBits
* note : totalCost is necessarily a multiple of baseCost */
assert(((U32)totalCost & (baseCost - 1)) == 0);
totalCost >>= (largestBits - targetNbBits);
assert(totalCost > 0);
/* repay normalized cost */
{ U32 const noSymbol = 0xF0F0F0F0;
U32 rankLast[HUF_TABLELOG_MAX+2];
/* Get pos of last (smallest = lowest cum. count) symbol per rank */
ZSTD_memset(rankLast, 0xF0, sizeof(rankLast));
{ U32 currentNbBits = targetNbBits;
int pos;
for (pos=n ; pos >= 0; pos--) {
if (huffNode[pos].nbBits >= currentNbBits) continue;
currentNbBits = huffNode[pos].nbBits; /* < targetNbBits */
rankLast[targetNbBits-currentNbBits] = (U32)pos;
} }
while (totalCost > 0) {
/* Try to reduce the next power of 2 above totalCost because we
* gain back half the rank.
*/
U32 nBitsToDecrease = ZSTD_highbit32((U32)totalCost) + 1;
for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
U32 const highPos = rankLast[nBitsToDecrease];
U32 const lowPos = rankLast[nBitsToDecrease-1];
if (highPos == noSymbol) continue;
/* Decrease highPos if no symbols of lowPos or if it is
* not cheaper to remove 2 lowPos than highPos.
*/
if (lowPos == noSymbol) break;
{ U32 const highTotal = huffNode[highPos].count;
U32 const lowTotal = 2 * huffNode[lowPos].count;
if (highTotal <= lowTotal) break;
} }
/* only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) */
assert(rankLast[nBitsToDecrease] != noSymbol || nBitsToDecrease == 1);
/* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))
nBitsToDecrease++;
assert(rankLast[nBitsToDecrease] != noSymbol);
/* Increase the number of bits to gain back half the rank cost. */
totalCost -= 1 << (nBitsToDecrease-1);
huffNode[rankLast[nBitsToDecrease]].nbBits++;
/* Fix up the new rank.
* If the new rank was empty, this symbol is now its smallest.
* Otherwise, this symbol will be the largest in the new rank so no adjustment.
*/
if (rankLast[nBitsToDecrease-1] == noSymbol)
rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease];
/* Fix up the old rank.
* If the symbol was at position 0, meaning it was the highest weight symbol in the tree,
* it must be the only symbol in its rank, so the old rank now has no symbols.
* Otherwise, since the Huffman nodes are sorted by count, the previous position is now
* the smallest node in the rank. If the previous position belongs to a different rank,
* then the rank is now empty.
*/
if (rankLast[nBitsToDecrease] == 0) /* special case, reached largest symbol */
rankLast[nBitsToDecrease] = noSymbol;
else {
rankLast[nBitsToDecrease]--;
if (huffNode[rankLast[nBitsToDecrease]].nbBits != targetNbBits-nBitsToDecrease)
rankLast[nBitsToDecrease] = noSymbol; /* this rank is now empty */
}
} /* while (totalCost > 0) */
/* If we've removed too much weight, then we have to add it back.
* To avoid overshooting again, we only adjust the smallest rank.
* We take the largest nodes from the lowest rank 0 and move them
* to rank 1. There's guaranteed to be enough rank 0 symbols because
* TODO.
*/
while (totalCost < 0) { /* Sometimes, cost correction overshoot */
/* special case : no rank 1 symbol (using targetNbBits-1);
* let's create one from largest rank 0 (using targetNbBits).
*/
if (rankLast[1] == noSymbol) {
while (huffNode[n].nbBits == targetNbBits) n--;
huffNode[n+1].nbBits--;
assert(n >= 0);
rankLast[1] = (U32)(n+1);
totalCost++;
continue;
}
huffNode[ rankLast[1] + 1 ].nbBits--;
rankLast[1]++;
totalCost ++;
}
} /* repay normalized cost */
} /* there are several too large elements (at least >= 2) */
return targetNbBits;
}
typedef struct {
U16 base;
U16 curr;
} rankPos;
typedef nodeElt huffNodeTable[2 * (HUF_SYMBOLVALUE_MAX + 1)];
/* Number of buckets available for HUF_sort() */
#define RANK_POSITION_TABLE_SIZE 192
typedef struct {
huffNodeTable huffNodeTbl;
rankPos rankPosition[RANK_POSITION_TABLE_SIZE];
} HUF_buildCTable_wksp_tables;
/* RANK_POSITION_DISTINCT_COUNT_CUTOFF == Cutoff point in HUF_sort() buckets for which we use log2 bucketing.
* Strategy is to use as many buckets as possible for representing distinct
* counts while using the remainder to represent all "large" counts.
*
* To satisfy this requirement for 192 buckets, we can do the following:
* Let buckets 0-166 represent distinct counts of [0, 166]
* Let buckets 166 to 192 represent all remaining counts up to RANK_POSITION_MAX_COUNT_LOG using log2 bucketing.
*/
#define RANK_POSITION_MAX_COUNT_LOG 32
#define RANK_POSITION_LOG_BUCKETS_BEGIN ((RANK_POSITION_TABLE_SIZE - 1) - RANK_POSITION_MAX_COUNT_LOG - 1 /* == 158 */)
#define RANK_POSITION_DISTINCT_COUNT_CUTOFF (RANK_POSITION_LOG_BUCKETS_BEGIN + ZSTD_highbit32(RANK_POSITION_LOG_BUCKETS_BEGIN) /* == 166 */)
/* Return the appropriate bucket index for a given count. See definition of
* RANK_POSITION_DISTINCT_COUNT_CUTOFF for explanation of bucketing strategy.
*/
static U32 HUF_getIndex(U32 const count) {
return (count < RANK_POSITION_DISTINCT_COUNT_CUTOFF)
? count
: ZSTD_highbit32(count) + RANK_POSITION_LOG_BUCKETS_BEGIN;
}
/* Helper swap function for HUF_quickSortPartition() */
static void HUF_swapNodes(nodeElt* a, nodeElt* b) {
nodeElt tmp = *a;
*a = *b;
*b = tmp;
}
/* Returns 0 if the huffNode array is not sorted by descending count */
MEM_STATIC int HUF_isSorted(nodeElt huffNode[], U32 const maxSymbolValue1) {
U32 i;
for (i = 1; i < maxSymbolValue1; ++i) {
if (huffNode[i].count > huffNode[i-1].count) {
return 0;
}
}
return 1;
}
/* Insertion sort by descending order */
HINT_INLINE void HUF_insertionSort(nodeElt huffNode[], int const low, int const high) {
int i;
int const size = high-low+1;
huffNode += low;
for (i = 1; i < size; ++i) {
nodeElt const key = huffNode[i];
int j = i - 1;
while (j >= 0 && huffNode[j].count < key.count) {
huffNode[j + 1] = huffNode[j];
j--;
}
huffNode[j + 1] = key;
}
}
/* Pivot helper function for quicksort. */
static int HUF_quickSortPartition(nodeElt arr[], int const low, int const high) {
/* Simply select rightmost element as pivot. "Better" selectors like
* median-of-three don't experimentally appear to have any benefit.
*/
U32 const pivot = arr[high].count;
int i = low - 1;
int j = low;
for ( ; j < high; j++) {
if (arr[j].count > pivot) {
i++;
HUF_swapNodes(&arr[i], &arr[j]);
}
}
HUF_swapNodes(&arr[i + 1], &arr[high]);
return i + 1;
}
/* Classic quicksort by descending with partially iterative calls
* to reduce worst case callstack size.
*/
static void HUF_simpleQuickSort(nodeElt arr[], int low, int high) {
int const kInsertionSortThreshold = 8;
if (high - low < kInsertionSortThreshold) {
HUF_insertionSort(arr, low, high);
return;
}
while (low < high) {
int const idx = HUF_quickSortPartition(arr, low, high);
if (idx - low < high - idx) {
HUF_simpleQuickSort(arr, low, idx - 1);
low = idx + 1;
} else {
HUF_simpleQuickSort(arr, idx + 1, high);
high = idx - 1;
}
}
}
/**
* HUF_sort():
* Sorts the symbols [0, maxSymbolValue] by count[symbol] in decreasing order.
* This is a typical bucket sorting strategy that uses either quicksort or insertion sort to sort each bucket.
*
* @param[out] huffNode Sorted symbols by decreasing count. Only members `.count` and `.byte` are filled.
* Must have (maxSymbolValue + 1) entries.
* @param[in] count Histogram of the symbols.
* @param[in] maxSymbolValue Maximum symbol value.
* @param rankPosition This is a scratch workspace. Must have RANK_POSITION_TABLE_SIZE entries.
*/
static void HUF_sort(nodeElt huffNode[], const unsigned count[], U32 const maxSymbolValue, rankPos rankPosition[]) {
U32 n;
U32 const maxSymbolValue1 = maxSymbolValue+1;
/* Compute base and set curr to base.
* For symbol s let lowerRank = HUF_getIndex(count[n]) and rank = lowerRank + 1.
* See HUF_getIndex to see bucketing strategy.
* We attribute each symbol to lowerRank's base value, because we want to know where
* each rank begins in the output, so for rank R we want to count ranks R+1 and above.
*/
ZSTD_memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
for (n = 0; n < maxSymbolValue1; ++n) {
U32 lowerRank = HUF_getIndex(count[n]);
assert(lowerRank < RANK_POSITION_TABLE_SIZE - 1);
rankPosition[lowerRank].base++;
}
assert(rankPosition[RANK_POSITION_TABLE_SIZE - 1].base == 0);
/* Set up the rankPosition table */
for (n = RANK_POSITION_TABLE_SIZE - 1; n > 0; --n) {
rankPosition[n-1].base += rankPosition[n].base;
rankPosition[n-1].curr = rankPosition[n-1].base;
}
/* Insert each symbol into their appropriate bucket, setting up rankPosition table. */
for (n = 0; n < maxSymbolValue1; ++n) {
U32 const c = count[n];
U32 const r = HUF_getIndex(c) + 1;
U32 const pos = rankPosition[r].curr++;
assert(pos < maxSymbolValue1);
huffNode[pos].count = c;
huffNode[pos].byte = (BYTE)n;
}
/* Sort each bucket. */
for (n = RANK_POSITION_DISTINCT_COUNT_CUTOFF; n < RANK_POSITION_TABLE_SIZE - 1; ++n) {
int const bucketSize = rankPosition[n].curr - rankPosition[n].base;
U32 const bucketStartIdx = rankPosition[n].base;
if (bucketSize > 1) {
assert(bucketStartIdx < maxSymbolValue1);
HUF_simpleQuickSort(huffNode + bucketStartIdx, 0, bucketSize-1);
}
}
assert(HUF_isSorted(huffNode, maxSymbolValue1));
}
/** HUF_buildCTable_wksp() :
* Same as HUF_buildCTable(), but using externally allocated scratch buffer.
* `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as sizeof(HUF_buildCTable_wksp_tables).
*/
#define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
/* HUF_buildTree():
* Takes the huffNode array sorted by HUF_sort() and builds an unlimited-depth Huffman tree.
*
* @param huffNode The array sorted by HUF_sort(). Builds the Huffman tree in this array.
* @param maxSymbolValue The maximum symbol value.
* @return The smallest node in the Huffman tree (by count).
*/
static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
{
nodeElt* const huffNode0 = huffNode - 1;
int nonNullRank;
int lowS, lowN;
int nodeNb = STARTNODE;
int n, nodeRoot;
DEBUGLOG(5, "HUF_buildTree (alphabet size = %u)", maxSymbolValue + 1);
/* init for parents */
nonNullRank = (int)maxSymbolValue;
while(huffNode[nonNullRank].count == 0) nonNullRank--;
lowS = nonNullRank; nodeRoot = nodeNb + lowS - 1; lowN = nodeNb;
huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count;
huffNode[lowS].parent = huffNode[lowS-1].parent = (U16)nodeNb;
nodeNb++; lowS-=2;
for (n=nodeNb; n<=nodeRoot; n++) huffNode[n].count = (U32)(1U<<30);
huffNode0[0].count = (U32)(1U<<31); /* fake entry, strong barrier */
/* create parents */
while (nodeNb <= nodeRoot) {
int const n1 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
int const n2 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
huffNode[nodeNb].count = huffNode[n1].count + huffNode[n2].count;
huffNode[n1].parent = huffNode[n2].parent = (U16)nodeNb;
nodeNb++;
}
/* distribute weights (unlimited tree height) */
huffNode[nodeRoot].nbBits = 0;
for (n=nodeRoot-1; n>=STARTNODE; n--)
huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
for (n=0; n<=nonNullRank; n++)
huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
DEBUGLOG(6, "Initial distribution of bits completed (%zu sorted symbols)", showHNodeBits(huffNode, maxSymbolValue+1));
return nonNullRank;
}
/**
* HUF_buildCTableFromTree():
* Build the CTable given the Huffman tree in huffNode.
*
* @param[out] CTable The output Huffman CTable.
* @param huffNode The Huffman tree.
* @param nonNullRank The last and smallest node in the Huffman tree.
* @param maxSymbolValue The maximum symbol value.
* @param maxNbBits The exact maximum number of bits used in the Huffman tree.
*/
static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, int nonNullRank, U32 maxSymbolValue, U32 maxNbBits)
{
HUF_CElt* const ct = CTable + 1;
/* fill result into ctable (val, nbBits) */
int n;
U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
int const alphabetSize = (int)(maxSymbolValue + 1);
for (n=0; n<=nonNullRank; n++)
nbPerRank[huffNode[n].nbBits]++;
/* determine starting value per rank */
{ U16 min = 0;
for (n=(int)maxNbBits; n>0; n--) {
valPerRank[n] = min; /* get starting value within each rank */
min += nbPerRank[n];
min >>= 1;
} }
for (n=0; n<alphabetSize; n++)
HUF_setNbBits(ct + huffNode[n].byte, huffNode[n].nbBits); /* push nbBits per symbol, symbol order */
for (n=0; n<alphabetSize; n++)
HUF_setValue(ct + n, valPerRank[HUF_getNbBits(ct[n])]++); /* assign value within rank, symbol order */
HUF_writeCTableHeader(CTable, maxNbBits, maxSymbolValue);
}
size_t
HUF_buildCTable_wksp(HUF_CElt* CTable, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
void* workSpace, size_t wkspSize)
{
HUF_buildCTable_wksp_tables* const wksp_tables =
(HUF_buildCTable_wksp_tables*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(U32));
nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
nodeElt* const huffNode = huffNode0+1;
int nonNullRank;
HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE == sizeof(HUF_buildCTable_wksp_tables));
DEBUGLOG(5, "HUF_buildCTable_wksp (alphabet size = %u)", maxSymbolValue+1);
/* safety checks */
if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
return ERROR(workSpace_tooSmall);
if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
return ERROR(maxSymbolValue_tooLarge);
ZSTD_memset(huffNode0, 0, sizeof(huffNodeTable));
/* sort, decreasing order */
HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
DEBUGLOG(6, "sorted symbols completed (%zu symbols)", showHNodeSymbols(huffNode, maxSymbolValue+1));
/* build tree */
nonNullRank = HUF_buildTree(huffNode, maxSymbolValue);
/* determine and enforce maxTableLog */
maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC); /* check fit into table */
HUF_buildCTableFromTree(CTable, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
return maxNbBits;
}
size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
{
HUF_CElt const* ct = CTable + 1;
size_t nbBits = 0;
int s;
for (s = 0; s <= (int)maxSymbolValue; ++s) {
nbBits += HUF_getNbBits(ct[s]) * count[s];
}
return nbBits >> 3;
}
int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
HUF_CTableHeader header = HUF_readCTableHeader(CTable);
HUF_CElt const* ct = CTable + 1;
int bad = 0;
int s;
assert(header.tableLog <= HUF_TABLELOG_ABSOLUTEMAX);
if (header.maxSymbolValue < maxSymbolValue)
return 0;
for (s = 0; s <= (int)maxSymbolValue; ++s) {
bad |= (count[s] != 0) & (HUF_getNbBits(ct[s]) == 0);
}
return !bad;
}
size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
/** HUF_CStream_t:
* Huffman uses its own BIT_CStream_t implementation.
* There are three major differences from BIT_CStream_t:
* 1. HUF_addBits() takes a HUF_CElt (size_t) which is
* the pair (nbBits, value) in the format:
* format:
* - Bits [0, 4) = nbBits
* - Bits [4, 64 - nbBits) = 0
* - Bits [64 - nbBits, 64) = value
* 2. The bitContainer is built from the upper bits and
* right shifted. E.g. to add a new value of N bits
* you right shift the bitContainer by N, then or in
* the new value into the N upper bits.
* 3. The bitstream has two bit containers. You can add
* bits to the second container and merge them into
* the first container.
*/
#define HUF_BITS_IN_CONTAINER (sizeof(size_t) * 8)
typedef struct {
size_t bitContainer[2];
size_t bitPos[2];
BYTE* startPtr;
BYTE* ptr;
BYTE* endPtr;
} HUF_CStream_t;
/**! HUF_initCStream():
* Initializes the bitstream.
* @returns 0 or an error code.
*/
static size_t HUF_initCStream(HUF_CStream_t* bitC,
void* startPtr, size_t dstCapacity)
{
ZSTD_memset(bitC, 0, sizeof(*bitC));
bitC->startPtr = (BYTE*)startPtr;
bitC->ptr = bitC->startPtr;
bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer[0]);
if (dstCapacity <= sizeof(bitC->bitContainer[0])) return ERROR(dstSize_tooSmall);
return 0;
}
/*! HUF_addBits():
* Adds the symbol stored in HUF_CElt elt to the bitstream.
*
* @param elt The element we're adding. This is a (nbBits, value) pair.
* See the HUF_CStream_t docs for the format.
* @param idx Insert into the bitstream at this idx.
* @param kFast This is a template parameter. If the bitstream is guaranteed
* to have at least 4 unused bits after this call it may be 1,
* otherwise it must be 0. HUF_addBits() is faster when fast is set.
*/
FORCE_INLINE_TEMPLATE void HUF_addBits(HUF_CStream_t* bitC, HUF_CElt elt, int idx, int kFast)
{
assert(idx <= 1);
assert(HUF_getNbBits(elt) <= HUF_TABLELOG_ABSOLUTEMAX);
/* This is efficient on x86-64 with BMI2 because shrx
* only reads the low 6 bits of the register. The compiler
* knows this and elides the mask. When fast is set,
* every operation can use the same value loaded from elt.
*/
bitC->bitContainer[idx] >>= HUF_getNbBits(elt);
bitC->bitContainer[idx] |= kFast ? HUF_getValueFast(elt) : HUF_getValue(elt);
/* We only read the low 8 bits of bitC->bitPos[idx] so it
* doesn't matter that the high bits have noise from the value.
*/
bitC->bitPos[idx] += HUF_getNbBitsFast(elt);
assert((bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
/* The last 4-bits of elt are dirty if fast is set,
* so we must not be overwriting bits that have already been
* inserted into the bit container.
*/
#if DEBUGLEVEL >= 1
{
size_t const nbBits = HUF_getNbBits(elt);
size_t const dirtyBits = nbBits == 0 ? 0 : ZSTD_highbit32((U32)nbBits) + 1;
(void)dirtyBits;
/* Middle bits are 0. */
assert(((elt >> dirtyBits) << (dirtyBits + nbBits)) == 0);
/* We didn't overwrite any bits in the bit container. */
assert(!kFast || (bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
(void)dirtyBits;
}
#endif
}
FORCE_INLINE_TEMPLATE void HUF_zeroIndex1(HUF_CStream_t* bitC)
{
bitC->bitContainer[1] = 0;
bitC->bitPos[1] = 0;
}
/*! HUF_mergeIndex1() :
* Merges the bit container @ index 1 into the bit container @ index 0
* and zeros the bit container @ index 1.
*/
FORCE_INLINE_TEMPLATE void HUF_mergeIndex1(HUF_CStream_t* bitC)
{
assert((bitC->bitPos[1] & 0xFF) < HUF_BITS_IN_CONTAINER);
bitC->bitContainer[0] >>= (bitC->bitPos[1] & 0xFF);
bitC->bitContainer[0] |= bitC->bitContainer[1];
bitC->bitPos[0] += bitC->bitPos[1];
assert((bitC->bitPos[0] & 0xFF) <= HUF_BITS_IN_CONTAINER);
}
/*! HUF_flushBits() :
* Flushes the bits in the bit container @ index 0.
*
* @post bitPos will be < 8.
* @param kFast If kFast is set then we must know a-priori that
* the bit container will not overflow.
*/
FORCE_INLINE_TEMPLATE void HUF_flushBits(HUF_CStream_t* bitC, int kFast)
{
/* The upper bits of bitPos are noisy, so we must mask by 0xFF. */
size_t const nbBits = bitC->bitPos[0] & 0xFF;
size_t const nbBytes = nbBits >> 3;
/* The top nbBits bits of bitContainer are the ones we need. */
size_t const bitContainer = bitC->bitContainer[0] >> (HUF_BITS_IN_CONTAINER - nbBits);
/* Mask bitPos to account for the bytes we consumed. */
bitC->bitPos[0] &= 7;
assert(nbBits > 0);
assert(nbBits <= sizeof(bitC->bitContainer[0]) * 8);
assert(bitC->ptr <= bitC->endPtr);
MEM_writeLEST(bitC->ptr, bitContainer);
bitC->ptr += nbBytes;
assert(!kFast || bitC->ptr <= bitC->endPtr);
if (!kFast && bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
/* bitContainer doesn't need to be modified because the leftover
* bits are already the top bitPos bits. And we don't care about
* noise in the lower values.
*/
}
/*! HUF_endMark()
* @returns The Huffman stream end mark: A 1-bit value = 1.
*/
static HUF_CElt HUF_endMark(void)
{
HUF_CElt endMark;
HUF_setNbBits(&endMark, 1);
HUF_setValue(&endMark, 1);
return endMark;
}
/*! HUF_closeCStream() :
* @return Size of CStream, in bytes,
* or 0 if it could not fit into dstBuffer */
static size_t HUF_closeCStream(HUF_CStream_t* bitC)
{
HUF_addBits(bitC, HUF_endMark(), /* idx */ 0, /* kFast */ 0);
HUF_flushBits(bitC, /* kFast */ 0);
{
size_t const nbBits = bitC->bitPos[0] & 0xFF;
if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
return (size_t)(bitC->ptr - bitC->startPtr) + (nbBits > 0);
}
}
FORCE_INLINE_TEMPLATE void
HUF_encodeSymbol(HUF_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable, int idx, int fast)
{
HUF_addBits(bitCPtr, CTable[symbol], idx, fast);
}
FORCE_INLINE_TEMPLATE void
HUF_compress1X_usingCTable_internal_body_loop(HUF_CStream_t* bitC,
const BYTE* ip, size_t srcSize,
const HUF_CElt* ct,
int kUnroll, int kFastFlush, int kLastFast)
{
/* Join to kUnroll */
int n = (int)srcSize;
int rem = n % kUnroll;
if (rem > 0) {
for (; rem > 0; --rem) {
HUF_encodeSymbol(bitC, ip[--n], ct, 0, /* fast */ 0);
}
HUF_flushBits(bitC, kFastFlush);
}
assert(n % kUnroll == 0);
/* Join to 2 * kUnroll */
if (n % (2 * kUnroll)) {
int u;
for (u = 1; u < kUnroll; ++u) {
HUF_encodeSymbol(bitC, ip[n - u], ct, 0, 1);
}
HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, 0, kLastFast);
HUF_flushBits(bitC, kFastFlush);
n -= kUnroll;
}
assert(n % (2 * kUnroll) == 0);
for (; n>0; n-= 2 * kUnroll) {
/* Encode kUnroll symbols into the bitstream @ index 0. */
int u;
for (u = 1; u < kUnroll; ++u) {
HUF_encodeSymbol(bitC, ip[n - u], ct, /* idx */ 0, /* fast */ 1);
}
HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, /* idx */ 0, /* fast */ kLastFast);
HUF_flushBits(bitC, kFastFlush);
/* Encode kUnroll symbols into the bitstream @ index 1.
* This allows us to start filling the bit container
* without any data dependencies.
*/
HUF_zeroIndex1(bitC);
for (u = 1; u < kUnroll; ++u) {
HUF_encodeSymbol(bitC, ip[n - kUnroll - u], ct, /* idx */ 1, /* fast */ 1);
}
HUF_encodeSymbol(bitC, ip[n - kUnroll - kUnroll], ct, /* idx */ 1, /* fast */ kLastFast);
/* Merge bitstream @ index 1 into the bitstream @ index 0 */
HUF_mergeIndex1(bitC);
HUF_flushBits(bitC, kFastFlush);
}
assert(n == 0);
}
/**
* Returns a tight upper bound on the output space needed by Huffman
* with 8 bytes buffer to handle over-writes. If the output is at least
* this large we don't need to do bounds checks during Huffman encoding.
*/
static size_t HUF_tightCompressBound(size_t srcSize, size_t tableLog)
{
return ((srcSize * tableLog) >> 3) + 8;
}
FORCE_INLINE_TEMPLATE size_t
HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
const void* src, size_t srcSize,
const HUF_CElt* CTable)
{
U32 const tableLog = HUF_readCTableHeader(CTable).tableLog;
HUF_CElt const* ct = CTable + 1;
const BYTE* ip = (const BYTE*) src;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstSize;
BYTE* op = ostart;
HUF_CStream_t bitC;
/* init */
if (dstSize < 8) return 0; /* not enough space to compress */
{ size_t const initErr = HUF_initCStream(&bitC, op, (size_t)(oend-op));
if (HUF_isError(initErr)) return 0; }
if (dstSize < HUF_tightCompressBound(srcSize, (size_t)tableLog) || tableLog > 11)
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ MEM_32bits() ? 2 : 4, /* kFast */ 0, /* kLastFast */ 0);
else {
if (MEM_32bits()) {
switch (tableLog) {
case 11:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 10: ZSTD_FALLTHROUGH;
case 9: ZSTD_FALLTHROUGH;
case 8:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 1);
break;
case 7: ZSTD_FALLTHROUGH;
default:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 3, /* kFastFlush */ 1, /* kLastFast */ 1);
break;
}
} else {
switch (tableLog) {
case 11:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 10:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 1);
break;
case 9:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 6, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 8:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 7, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 7:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 8, /* kFastFlush */ 1, /* kLastFast */ 0);
break;
case 6: ZSTD_FALLTHROUGH;
default:
HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 9, /* kFastFlush */ 1, /* kLastFast */ 1);
break;
}
}
}
assert(bitC.ptr <= bitC.endPtr);
return HUF_closeCStream(&bitC);
}
#if DYNAMIC_BMI2
static BMI2_TARGET_ATTRIBUTE size_t
HUF_compress1X_usingCTable_internal_bmi2(void* dst, size_t dstSize,
const void* src, size_t srcSize,
const HUF_CElt* CTable)
{
return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
}
static size_t
HUF_compress1X_usingCTable_internal_default(void* dst, size_t dstSize,
const void* src, size_t srcSize,
const HUF_CElt* CTable)
{
return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
}
static size_t
HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
const void* src, size_t srcSize,
const HUF_CElt* CTable, const int flags)
{
if (flags & HUF_flags_bmi2) {
return HUF_compress1X_usingCTable_internal_bmi2(dst, dstSize, src, srcSize, CTable);
}
return HUF_compress1X_usingCTable_internal_default(dst, dstSize, src, srcSize, CTable);
}
#else
static size_t
HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
const void* src, size_t srcSize,
const HUF_CElt* CTable, const int flags)
{
(void)flags;
return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
}
#endif
size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
{
return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
}
static size_t
HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
const void* src, size_t srcSize,
const HUF_CElt* CTable, int flags)
{
size_t const segmentSize = (srcSize+3)/4; /* first 3 segments */
const BYTE* ip = (const BYTE*) src;
const BYTE* const iend = ip + srcSize;
BYTE* const ostart = (BYTE*) dst;
BYTE* const oend = ostart + dstSize;
BYTE* op = ostart;
if (dstSize < 6 + 1 + 1 + 1 + 8) return 0; /* minimum space to compress successfully */
if (srcSize < 12) return 0; /* no saving possible : too small input */
op += 6; /* jumpTable */
assert(op <= oend);
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
if (cSize == 0 || cSize > 65535) return 0;
MEM_writeLE16(ostart, (U16)cSize);
op += cSize;
}
ip += segmentSize;
assert(op <= oend);
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
if (cSize == 0 || cSize > 65535) return 0;
MEM_writeLE16(ostart+2, (U16)cSize);
op += cSize;
}
ip += segmentSize;
assert(op <= oend);
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
if (cSize == 0 || cSize > 65535) return 0;
MEM_writeLE16(ostart+4, (U16)cSize);
op += cSize;
}
ip += segmentSize;
assert(op <= oend);
assert(ip <= iend);
{ CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, flags) );
if (cSize == 0 || cSize > 65535) return 0;
op += cSize;
}
return (size_t)(op-ostart);
}
size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
{
return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
}
typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
static size_t HUF_compressCTable_internal(
BYTE* const ostart, BYTE* op, BYTE* const oend,
const void* src, size_t srcSize,
HUF_nbStreams_e nbStreams, const HUF_CElt* CTable, const int flags)
{
size_t const cSize = (nbStreams==HUF_singleStream) ?
HUF_compress1X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags) :
HUF_compress4X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags);
if (HUF_isError(cSize)) { return cSize; }
if (cSize==0) { return 0; } /* uncompressible */
op += cSize;
/* check compressibility */
assert(op >= ostart);
if ((size_t)(op-ostart) >= srcSize-1) { return 0; }
return (size_t)(op-ostart);
}
typedef struct {
unsigned count[HUF_SYMBOLVALUE_MAX + 1];
HUF_CElt CTable[HUF_CTABLE_SIZE_ST(HUF_SYMBOLVALUE_MAX)];
union {
HUF_buildCTable_wksp_tables buildCTable_wksp;
HUF_WriteCTableWksp writeCTable_wksp;
U32 hist_wksp[HIST_WKSP_SIZE_U32];
} wksps;
} HUF_compress_tables_t;
#define SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE 4096
#define SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO 10 /* Must be >= 2 */
unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue)
{
unsigned cardinality = 0;
unsigned i;
for (i = 0; i < maxSymbolValue + 1; i++) {
if (count[i] != 0) cardinality += 1;
}
return cardinality;
}
unsigned HUF_minTableLog(unsigned symbolCardinality)
{
U32 minBitsSymbols = ZSTD_highbit32(symbolCardinality) + 1;
return minBitsSymbols;
}
unsigned HUF_optimalTableLog(
unsigned maxTableLog,
size_t srcSize,
unsigned maxSymbolValue,
void* workSpace, size_t wkspSize,
HUF_CElt* table,
const unsigned* count,
int flags)
{
assert(srcSize > 1); /* Not supported, RLE should be used instead */
assert(wkspSize >= sizeof(HUF_buildCTable_wksp_tables));
if (!(flags & HUF_flags_optimalDepth)) {
/* cheap evaluation, based on FSE */
return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
}
{ BYTE* dst = (BYTE*)workSpace + sizeof(HUF_WriteCTableWksp);
size_t dstSize = wkspSize - sizeof(HUF_WriteCTableWksp);
size_t maxBits, hSize, newSize;
const unsigned symbolCardinality = HUF_cardinality(count, maxSymbolValue);
const unsigned minTableLog = HUF_minTableLog(symbolCardinality);
size_t optSize = ((size_t) ~0) - 1;
unsigned optLog = maxTableLog, optLogGuess;
DEBUGLOG(6, "HUF_optimalTableLog: probing huf depth (srcSize=%zu)", srcSize);
/* Search until size increases */
for (optLogGuess = minTableLog; optLogGuess <= maxTableLog; optLogGuess++) {
DEBUGLOG(7, "checking for huffLog=%u", optLogGuess);
maxBits = HUF_buildCTable_wksp(table, count, maxSymbolValue, optLogGuess, workSpace, wkspSize);
if (ERR_isError(maxBits)) continue;
if (maxBits < optLogGuess && optLogGuess > minTableLog) break;
hSize = HUF_writeCTable_wksp(dst, dstSize, table, maxSymbolValue, (U32)maxBits, workSpace, wkspSize);
if (ERR_isError(hSize)) continue;
newSize = HUF_estimateCompressedSize(table, count, maxSymbolValue) + hSize;
if (newSize > optSize + 1) {
break;
}
if (newSize < optSize) {
optSize = newSize;
optLog = optLogGuess;
}
}
assert(optLog <= HUF_TABLELOG_MAX);
return optLog;
}
}
/* HUF_compress_internal() :
* `workSpace_align4` must be aligned on 4-bytes boundaries,
* and occupies the same space as a table of HUF_WORKSPACE_SIZE_U64 unsigned */
static size_t
HUF_compress_internal (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog,
HUF_nbStreams_e nbStreams,
void* workSpace, size_t wkspSize,
HUF_CElt* oldHufTable, HUF_repeat* repeat, int flags)
{
HUF_compress_tables_t* const table = (HUF_compress_tables_t*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(size_t));
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstSize;
BYTE* op = ostart;
DEBUGLOG(5, "HUF_compress_internal (srcSize=%zu)", srcSize);
HUF_STATIC_ASSERT(sizeof(*table) + HUF_WORKSPACE_MAX_ALIGNMENT <= HUF_WORKSPACE_SIZE);
/* checks & inits */
if (wkspSize < sizeof(*table)) return ERROR(workSpace_tooSmall);
if (!srcSize) return 0; /* Uncompressed */
if (!dstSize) return 0; /* cannot fit anything within dst budget */
if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong); /* current block size limit */
if (huffLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
if (!maxSymbolValue) maxSymbolValue = HUF_SYMBOLVALUE_MAX;
if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
/* Heuristic : If old table is valid, use it for small inputs */
if ((flags & HUF_flags_preferRepeat) && repeat && *repeat == HUF_repeat_valid) {
return HUF_compressCTable_internal(ostart, op, oend,
src, srcSize,
nbStreams, oldHufTable, flags);
}
/* If uncompressible data is suspected, do a smaller sampling first */
DEBUG_STATIC_ASSERT(SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO >= 2);
if ((flags & HUF_flags_suspectUncompressible) && srcSize >= (SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE * SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO)) {
size_t largestTotal = 0;
DEBUGLOG(5, "input suspected incompressible : sampling to check");
{ unsigned maxSymbolValueBegin = maxSymbolValue;
CHECK_V_F(largestBegin, HIST_count_simple (table->count, &maxSymbolValueBegin, (const BYTE*)src, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
largestTotal += largestBegin;
}
{ unsigned maxSymbolValueEnd = maxSymbolValue;
CHECK_V_F(largestEnd, HIST_count_simple (table->count, &maxSymbolValueEnd, (const BYTE*)src + srcSize - SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
largestTotal += largestEnd;
}
if (largestTotal <= ((2 * SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) >> 7)+4) return 0; /* heuristic : probably not compressible enough */
}
/* Scan input and build symbol stats */
{ CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->wksps.hist_wksp, sizeof(table->wksps.hist_wksp)) );
if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; } /* single symbol, rle */
if (largest <= (srcSize >> 7)+4) return 0; /* heuristic : probably not compressible enough */
}
DEBUGLOG(6, "histogram detail completed (%zu symbols)", showU32(table->count, maxSymbolValue+1));
/* Check validity of previous table */
if ( repeat
&& *repeat == HUF_repeat_check
&& !HUF_validateCTable(oldHufTable, table->count, maxSymbolValue)) {
*repeat = HUF_repeat_none;
}
/* Heuristic : use existing table for small inputs */
if ((flags & HUF_flags_preferRepeat) && repeat && *repeat != HUF_repeat_none) {
return HUF_compressCTable_internal(ostart, op, oend,
src, srcSize,
nbStreams, oldHufTable, flags);
}
/* Build Huffman Tree */
huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, &table->wksps, sizeof(table->wksps), table->CTable, table->count, flags);
{ size_t const maxBits = HUF_buildCTable_wksp(table->CTable, table->count,
maxSymbolValue, huffLog,
&table->wksps.buildCTable_wksp, sizeof(table->wksps.buildCTable_wksp));
CHECK_F(maxBits);
huffLog = (U32)maxBits;
DEBUGLOG(6, "bit distribution completed (%zu symbols)", showCTableBits(table->CTable + 1, maxSymbolValue+1));
}
/* Write table description header */
{ CHECK_V_F(hSize, HUF_writeCTable_wksp(op, dstSize, table->CTable, maxSymbolValue, huffLog,
&table->wksps.writeCTable_wksp, sizeof(table->wksps.writeCTable_wksp)) );
/* Check if using previous huffman table is beneficial */
if (repeat && *repeat != HUF_repeat_none) {
size_t const oldSize = HUF_estimateCompressedSize(oldHufTable, table->count, maxSymbolValue);
size_t const newSize = HUF_estimateCompressedSize(table->CTable, table->count, maxSymbolValue);
if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
return HUF_compressCTable_internal(ostart, op, oend,
src, srcSize,
nbStreams, oldHufTable, flags);
} }
/* Use the new huffman table */
if (hSize + 12ul >= srcSize) { return 0; }
op += hSize;
if (repeat) { *repeat = HUF_repeat_none; }
if (oldHufTable)
ZSTD_memcpy(oldHufTable, table->CTable, sizeof(table->CTable)); /* Save new table */
}
return HUF_compressCTable_internal(ostart, op, oend,
src, srcSize,
nbStreams, table->CTable, flags);
}
size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog,
void* workSpace, size_t wkspSize,
HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
{
DEBUGLOG(5, "HUF_compress1X_repeat (srcSize = %zu)", srcSize);
return HUF_compress_internal(dst, dstSize, src, srcSize,
maxSymbolValue, huffLog, HUF_singleStream,
workSpace, wkspSize, hufTable,
repeat, flags);
}
/* HUF_compress4X_repeat():
* compress input using 4 streams.
* consider skipping quickly
* re-use an existing huffman compression table */
size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog,
void* workSpace, size_t wkspSize,
HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
{
DEBUGLOG(5, "HUF_compress4X_repeat (srcSize = %zu)", srcSize);
return HUF_compress_internal(dst, dstSize, src, srcSize,
maxSymbolValue, huffLog, HUF_fourStreams,
workSpace, wkspSize,
hufTable, repeat, flags);
}
/**** ended inlining compress/huf_compress.c ****/
/**** start inlining compress/zstd_compress_literals.c ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/*-*************************************
* Dependencies
***************************************/
/**** start inlining zstd_compress_literals.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_COMPRESS_LITERALS_H
#define ZSTD_COMPRESS_LITERALS_H
/**** start inlining zstd_compress_internal.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/* This header contains definitions
* that shall **only** be used by modules within lib/compress.
*/
#ifndef ZSTD_COMPRESS_H
#define ZSTD_COMPRESS_H
/*-*************************************
* Dependencies
***************************************/
/**** skipping file: ../common/zstd_internal.h ****/
/**** start inlining zstd_cwksp.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_CWKSP_H
#define ZSTD_CWKSP_H
/*-*************************************
* Dependencies
***************************************/
/**** skipping file: ../common/allocations.h ****/
/**** skipping file: ../common/zstd_internal.h ****/
/**** skipping file: ../common/portability_macros.h ****/
#if defined (__cplusplus)
extern "C" {
#endif
/*-*************************************
* Constants
***************************************/
/* Since the workspace is effectively its own little malloc implementation /
* arena, when we run under ASAN, we should similarly insert redzones between
* each internal element of the workspace, so ASAN will catch overruns that
* reach outside an object but that stay inside the workspace.
*
* This defines the size of that redzone.
*/
#ifndef ZSTD_CWKSP_ASAN_REDZONE_SIZE
#define ZSTD_CWKSP_ASAN_REDZONE_SIZE 128
#endif
/* Set our tables and aligneds to align by 64 bytes */
#define ZSTD_CWKSP_ALIGNMENT_BYTES 64
/*-*************************************
* Structures
***************************************/
typedef enum {
ZSTD_cwksp_alloc_objects,
ZSTD_cwksp_alloc_aligned_init_once,
ZSTD_cwksp_alloc_aligned,
ZSTD_cwksp_alloc_buffers
} ZSTD_cwksp_alloc_phase_e;
/**
* Used to describe whether the workspace is statically allocated (and will not
* necessarily ever be freed), or if it's dynamically allocated and we can
* expect a well-formed caller to free this.
*/
typedef enum {
ZSTD_cwksp_dynamic_alloc,
ZSTD_cwksp_static_alloc
} ZSTD_cwksp_static_alloc_e;
/**
* Zstd fits all its internal datastructures into a single continuous buffer,
* so that it only needs to perform a single OS allocation (or so that a buffer
* can be provided to it and it can perform no allocations at all). This buffer
* is called the workspace.
*
* Several optimizations complicate that process of allocating memory ranges
* from this workspace for each internal datastructure:
*
* - These different internal datastructures have different setup requirements:
*
* - The static objects need to be cleared once and can then be trivially
* reused for each compression.
*
* - Various buffers don't need to be initialized at all--they are always
* written into before they're read.
*
* - The matchstate tables have a unique requirement that they don't need
* their memory to be totally cleared, but they do need the memory to have
* some bound, i.e., a guarantee that all values in the memory they've been
* allocated is less than some maximum value (which is the starting value
* for the indices that they will then use for compression). When this
* guarantee is provided to them, they can use the memory without any setup
* work. When it can't, they have to clear the area.
*
* - These buffers also have different alignment requirements.
*
* - We would like to reuse the objects in the workspace for multiple
* compressions without having to perform any expensive reallocation or
* reinitialization work.
*
* - We would like to be able to efficiently reuse the workspace across
* multiple compressions **even when the compression parameters change** and
* we need to resize some of the objects (where possible).
*
* To attempt to manage this buffer, given these constraints, the ZSTD_cwksp
* abstraction was created. It works as follows:
*
* Workspace Layout:
*
* [ ... workspace ... ]
* [objects][tables ->] free space [<- buffers][<- aligned][<- init once]
*
* The various objects that live in the workspace are divided into the
* following categories, and are allocated separately:
*
* - Static objects: this is optionally the enclosing ZSTD_CCtx or ZSTD_CDict,
* so that literally everything fits in a single buffer. Note: if present,
* this must be the first object in the workspace, since ZSTD_customFree{CCtx,
* CDict}() rely on a pointer comparison to see whether one or two frees are
* required.
*
* - Fixed size objects: these are fixed-size, fixed-count objects that are
* nonetheless "dynamically" allocated in the workspace so that we can
* control how they're initialized separately from the broader ZSTD_CCtx.
* Examples:
* - Entropy Workspace
* - 2 x ZSTD_compressedBlockState_t
* - CDict dictionary contents
*
* - Tables: these are any of several different datastructures (hash tables,
* chain tables, binary trees) that all respect a common format: they are
* uint32_t arrays, all of whose values are between 0 and (nextSrc - base).
* Their sizes depend on the cparams. These tables are 64-byte aligned.
*
* - Init once: these buffers require to be initialized at least once before
* use. They should be used when we want to skip memory initialization
* while not triggering memory checkers (like Valgrind) when reading from
* from this memory without writing to it first.
* These buffers should be used carefully as they might contain data
* from previous compressions.
* Buffers are aligned to 64 bytes.
*
* - Aligned: these buffers don't require any initialization before they're
* used. The user of the buffer should make sure they write into a buffer
* location before reading from it.
* Buffers are aligned to 64 bytes.
*
* - Buffers: these buffers are used for various purposes that don't require
* any alignment or initialization before they're used. This means they can
* be moved around at no cost for a new compression.
*
* Allocating Memory:
*
* The various types of objects must be allocated in order, so they can be
* correctly packed into the workspace buffer. That order is:
*
* 1. Objects
* 2. Init once / Tables
* 3. Aligned / Tables
* 4. Buffers / Tables
*
* Attempts to reserve objects of different types out of order will fail.
*/
typedef struct {
void* workspace;
void* workspaceEnd;
void* objectEnd;
void* tableEnd;
void* tableValidEnd;
void* allocStart;
void* initOnceStart;
BYTE allocFailed;
int workspaceOversizedDuration;
ZSTD_cwksp_alloc_phase_e phase;
ZSTD_cwksp_static_alloc_e isStatic;
} ZSTD_cwksp;
/*-*************************************
* Functions
***************************************/
MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws);
MEM_STATIC void* ZSTD_cwksp_initialAllocStart(ZSTD_cwksp* ws);
MEM_STATIC void ZSTD_cwksp_assert_internal_consistency(ZSTD_cwksp* ws) {
(void)ws;
assert(ws->workspace <= ws->objectEnd);
assert(ws->objectEnd <= ws->tableEnd);
assert(ws->objectEnd <= ws->tableValidEnd);
assert(ws->tableEnd <= ws->allocStart);
assert(ws->tableValidEnd <= ws->allocStart);
assert(ws->allocStart <= ws->workspaceEnd);
assert(ws->initOnceStart <= ZSTD_cwksp_initialAllocStart(ws));
assert(ws->workspace <= ws->initOnceStart);
#if ZSTD_MEMORY_SANITIZER
{
intptr_t const offset = __msan_test_shadow(ws->initOnceStart,
(U8*)ZSTD_cwksp_initialAllocStart(ws) - (U8*)ws->initOnceStart);
(void)offset;
#if defined(ZSTD_MSAN_PRINT)
if(offset!=-1) {
__msan_print_shadow((U8*)ws->initOnceStart + offset - 8, 32);
}
#endif
assert(offset==-1);
};
#endif
}
/**
* Align must be a power of 2.
*/
MEM_STATIC size_t ZSTD_cwksp_align(size_t size, size_t const align) {
size_t const mask = align - 1;
assert((align & mask) == 0);
return (size + mask) & ~mask;
}
/**
* Use this to determine how much space in the workspace we will consume to
* allocate this object. (Normally it should be exactly the size of the object,
* but under special conditions, like ASAN, where we pad each object, it might
* be larger.)
*
* Since tables aren't currently redzoned, you don't need to call through this
* to figure out how much space you need for the matchState tables. Everything
* else is though.
*
* Do not use for sizing aligned buffers. Instead, use ZSTD_cwksp_aligned_alloc_size().
*/
MEM_STATIC size_t ZSTD_cwksp_alloc_size(size_t size) {
if (size == 0)
return 0;
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
return size + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
#else
return size;
#endif
}
/**
* Returns an adjusted alloc size that is the nearest larger multiple of 64 bytes.
* Used to determine the number of bytes required for a given "aligned".
*/
MEM_STATIC size_t ZSTD_cwksp_aligned_alloc_size(size_t size) {
return ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(size, ZSTD_CWKSP_ALIGNMENT_BYTES));
}
/**
* Returns the amount of additional space the cwksp must allocate
* for internal purposes (currently only alignment).
*/
MEM_STATIC size_t ZSTD_cwksp_slack_space_required(void) {
/* For alignment, the wksp will always allocate an additional 2*ZSTD_CWKSP_ALIGNMENT_BYTES
* bytes to align the beginning of tables section and end of buffers;
*/
size_t const slackSpace = ZSTD_CWKSP_ALIGNMENT_BYTES * 2;
return slackSpace;
}
/**
* Return the number of additional bytes required to align a pointer to the given number of bytes.
* alignBytes must be a power of two.
*/
MEM_STATIC size_t ZSTD_cwksp_bytes_to_align_ptr(void* ptr, const size_t alignBytes) {
size_t const alignBytesMask = alignBytes - 1;
size_t const bytes = (alignBytes - ((size_t)ptr & (alignBytesMask))) & alignBytesMask;
assert((alignBytes & alignBytesMask) == 0);
assert(bytes < alignBytes);
return bytes;
}
/**
* Returns the initial value for allocStart which is used to determine the position from
* which we can allocate from the end of the workspace.
*/
MEM_STATIC void* ZSTD_cwksp_initialAllocStart(ZSTD_cwksp* ws) {
return (void*)((size_t)ws->workspaceEnd & ~(ZSTD_CWKSP_ALIGNMENT_BYTES-1));
}
/**
* Internal function. Do not use directly.
* Reserves the given number of bytes within the aligned/buffer segment of the wksp,
* which counts from the end of the wksp (as opposed to the object/table segment).
*
* Returns a pointer to the beginning of that space.
*/
MEM_STATIC void*
ZSTD_cwksp_reserve_internal_buffer_space(ZSTD_cwksp* ws, size_t const bytes)
{
void* const alloc = (BYTE*)ws->allocStart - bytes;
void* const bottom = ws->tableEnd;
DEBUGLOG(5, "cwksp: reserving %p %zd bytes, %zd bytes remaining",
alloc, bytes, ZSTD_cwksp_available_space(ws) - bytes);
ZSTD_cwksp_assert_internal_consistency(ws);
assert(alloc >= bottom);
if (alloc < bottom) {
DEBUGLOG(4, "cwksp: alloc failed!");
ws->allocFailed = 1;
return NULL;
}
/* the area is reserved from the end of wksp.
* If it overlaps with tableValidEnd, it voids guarantees on values' range */
if (alloc < ws->tableValidEnd) {
ws->tableValidEnd = alloc;
}
ws->allocStart = alloc;
return alloc;
}
/**
* Moves the cwksp to the next phase, and does any necessary allocations.
* cwksp initialization must necessarily go through each phase in order.
* Returns a 0 on success, or zstd error
*/
MEM_STATIC size_t
ZSTD_cwksp_internal_advance_phase(ZSTD_cwksp* ws, ZSTD_cwksp_alloc_phase_e phase)
{
assert(phase >= ws->phase);
if (phase > ws->phase) {
/* Going from allocating objects to allocating initOnce / tables */
if (ws->phase < ZSTD_cwksp_alloc_aligned_init_once &&
phase >= ZSTD_cwksp_alloc_aligned_init_once) {
ws->tableValidEnd = ws->objectEnd;
ws->initOnceStart = ZSTD_cwksp_initialAllocStart(ws);
{ /* Align the start of the tables to 64 bytes. Use [0, 63] bytes */
void *const alloc = ws->objectEnd;
size_t const bytesToAlign = ZSTD_cwksp_bytes_to_align_ptr(alloc, ZSTD_CWKSP_ALIGNMENT_BYTES);
void *const objectEnd = (BYTE *) alloc + bytesToAlign;
DEBUGLOG(5, "reserving table alignment addtl space: %zu", bytesToAlign);
RETURN_ERROR_IF(objectEnd > ws->workspaceEnd, memory_allocation,
"table phase - alignment initial allocation failed!");
ws->objectEnd = objectEnd;
ws->tableEnd = objectEnd; /* table area starts being empty */
if (ws->tableValidEnd < ws->tableEnd) {
ws->tableValidEnd = ws->tableEnd;
}
}
}
ws->phase = phase;
ZSTD_cwksp_assert_internal_consistency(ws);
}
return 0;
}
/**
* Returns whether this object/buffer/etc was allocated in this workspace.
*/
MEM_STATIC int ZSTD_cwksp_owns_buffer(const ZSTD_cwksp* ws, const void* ptr)
{
return (ptr != NULL) && (ws->workspace <= ptr) && (ptr < ws->workspaceEnd);
}
/**
* Internal function. Do not use directly.
*/
MEM_STATIC void*
ZSTD_cwksp_reserve_internal(ZSTD_cwksp* ws, size_t bytes, ZSTD_cwksp_alloc_phase_e phase)
{
void* alloc;
if (ZSTD_isError(ZSTD_cwksp_internal_advance_phase(ws, phase)) || bytes == 0) {
return NULL;
}
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* over-reserve space */
bytes += 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
#endif
alloc = ZSTD_cwksp_reserve_internal_buffer_space(ws, bytes);
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
* either size. */
if (alloc) {
alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
/* We need to keep the redzone poisoned while unpoisoning the bytes that
* are actually allocated. */
__asan_unpoison_memory_region(alloc, bytes - 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE);
}
}
#endif
return alloc;
}
/**
* Reserves and returns unaligned memory.
*/
MEM_STATIC BYTE* ZSTD_cwksp_reserve_buffer(ZSTD_cwksp* ws, size_t bytes)
{
return (BYTE*)ZSTD_cwksp_reserve_internal(ws, bytes, ZSTD_cwksp_alloc_buffers);
}
/**
* Reserves and returns memory sized on and aligned on ZSTD_CWKSP_ALIGNMENT_BYTES (64 bytes).
* This memory has been initialized at least once in the past.
* This doesn't mean it has been initialized this time, and it might contain data from previous
* operations.
* The main usage is for algorithms that might need read access into uninitialized memory.
* The algorithm must maintain safety under these conditions and must make sure it doesn't
* leak any of the past data (directly or in side channels).
*/
MEM_STATIC void* ZSTD_cwksp_reserve_aligned_init_once(ZSTD_cwksp* ws, size_t bytes)
{
size_t const alignedBytes = ZSTD_cwksp_align(bytes, ZSTD_CWKSP_ALIGNMENT_BYTES);
void* ptr = ZSTD_cwksp_reserve_internal(ws, alignedBytes, ZSTD_cwksp_alloc_aligned_init_once);
assert(((size_t)ptr & (ZSTD_CWKSP_ALIGNMENT_BYTES-1))== 0);
if(ptr && ptr < ws->initOnceStart) {
/* We assume the memory following the current allocation is either:
* 1. Not usable as initOnce memory (end of workspace)
* 2. Another initOnce buffer that has been allocated before (and so was previously memset)
* 3. An ASAN redzone, in which case we don't want to write on it
* For these reasons it should be fine to not explicitly zero every byte up to ws->initOnceStart.
* Note that we assume here that MSAN and ASAN cannot run in the same time. */
ZSTD_memset(ptr, 0, MIN((size_t)((U8*)ws->initOnceStart - (U8*)ptr), alignedBytes));
ws->initOnceStart = ptr;
}
#if ZSTD_MEMORY_SANITIZER
assert(__msan_test_shadow(ptr, bytes) == -1);
#endif
return ptr;
}
/**
* Reserves and returns memory sized on and aligned on ZSTD_CWKSP_ALIGNMENT_BYTES (64 bytes).
*/
MEM_STATIC void* ZSTD_cwksp_reserve_aligned(ZSTD_cwksp* ws, size_t bytes)
{
void* ptr = ZSTD_cwksp_reserve_internal(ws, ZSTD_cwksp_align(bytes, ZSTD_CWKSP_ALIGNMENT_BYTES),
ZSTD_cwksp_alloc_aligned);
assert(((size_t)ptr & (ZSTD_CWKSP_ALIGNMENT_BYTES-1))== 0);
return ptr;
}
/**
* Aligned on 64 bytes. These buffers have the special property that
* their values remain constrained, allowing us to re-use them without
* memset()-ing them.
*/
MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes)
{
const ZSTD_cwksp_alloc_phase_e phase = ZSTD_cwksp_alloc_aligned_init_once;
void* alloc;
void* end;
void* top;
/* We can only start allocating tables after we are done reserving space for objects at the
* start of the workspace */
if(ws->phase < phase) {
if (ZSTD_isError(ZSTD_cwksp_internal_advance_phase(ws, phase))) {
return NULL;
}
}
alloc = ws->tableEnd;
end = (BYTE *)alloc + bytes;
top = ws->allocStart;
DEBUGLOG(5, "cwksp: reserving %p table %zd bytes, %zd bytes remaining",
alloc, bytes, ZSTD_cwksp_available_space(ws) - bytes);
assert((bytes & (sizeof(U32)-1)) == 0);
ZSTD_cwksp_assert_internal_consistency(ws);
assert(end <= top);
if (end > top) {
DEBUGLOG(4, "cwksp: table alloc failed!");
ws->allocFailed = 1;
return NULL;
}
ws->tableEnd = end;
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
__asan_unpoison_memory_region(alloc, bytes);
}
#endif
assert((bytes & (ZSTD_CWKSP_ALIGNMENT_BYTES-1)) == 0);
assert(((size_t)alloc & (ZSTD_CWKSP_ALIGNMENT_BYTES-1))== 0);
return alloc;
}
/**
* Aligned on sizeof(void*).
* Note : should happen only once, at workspace first initialization
*/
MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes)
{
size_t const roundedBytes = ZSTD_cwksp_align(bytes, sizeof(void*));
void* alloc = ws->objectEnd;
void* end = (BYTE*)alloc + roundedBytes;
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* over-reserve space */
end = (BYTE *)end + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
#endif
DEBUGLOG(4,
"cwksp: reserving %p object %zd bytes (rounded to %zd), %zd bytes remaining",
alloc, bytes, roundedBytes, ZSTD_cwksp_available_space(ws) - roundedBytes);
assert((size_t)alloc % ZSTD_ALIGNOF(void*) == 0);
assert(bytes % ZSTD_ALIGNOF(void*) == 0);
ZSTD_cwksp_assert_internal_consistency(ws);
/* we must be in the first phase, no advance is possible */
if (ws->phase != ZSTD_cwksp_alloc_objects || end > ws->workspaceEnd) {
DEBUGLOG(3, "cwksp: object alloc failed!");
ws->allocFailed = 1;
return NULL;
}
ws->objectEnd = end;
ws->tableEnd = end;
ws->tableValidEnd = end;
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
* either size. */
alloc = (BYTE*)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
__asan_unpoison_memory_region(alloc, bytes);
}
#endif
return alloc;
}
MEM_STATIC void ZSTD_cwksp_mark_tables_dirty(ZSTD_cwksp* ws)
{
DEBUGLOG(4, "cwksp: ZSTD_cwksp_mark_tables_dirty");
#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
/* To validate that the table re-use logic is sound, and that we don't
* access table space that we haven't cleaned, we re-"poison" the table
* space every time we mark it dirty.
* Since tableValidEnd space and initOnce space may overlap we don't poison
* the initOnce portion as it break its promise. This means that this poisoning
* check isn't always applied fully. */
{
size_t size = (BYTE*)ws->tableValidEnd - (BYTE*)ws->objectEnd;
assert(__msan_test_shadow(ws->objectEnd, size) == -1);
if((BYTE*)ws->tableValidEnd < (BYTE*)ws->initOnceStart) {
__msan_poison(ws->objectEnd, size);
} else {
assert(ws->initOnceStart >= ws->objectEnd);
__msan_poison(ws->objectEnd, (BYTE*)ws->initOnceStart - (BYTE*)ws->objectEnd);
}
}
#endif
assert(ws->tableValidEnd >= ws->objectEnd);
assert(ws->tableValidEnd <= ws->allocStart);
ws->tableValidEnd = ws->objectEnd;
ZSTD_cwksp_assert_internal_consistency(ws);
}
MEM_STATIC void ZSTD_cwksp_mark_tables_clean(ZSTD_cwksp* ws) {
DEBUGLOG(4, "cwksp: ZSTD_cwksp_mark_tables_clean");
assert(ws->tableValidEnd >= ws->objectEnd);
assert(ws->tableValidEnd <= ws->allocStart);
if (ws->tableValidEnd < ws->tableEnd) {
ws->tableValidEnd = ws->tableEnd;
}
ZSTD_cwksp_assert_internal_consistency(ws);
}
/**
* Zero the part of the allocated tables not already marked clean.
*/
MEM_STATIC void ZSTD_cwksp_clean_tables(ZSTD_cwksp* ws) {
DEBUGLOG(4, "cwksp: ZSTD_cwksp_clean_tables");
assert(ws->tableValidEnd >= ws->objectEnd);
assert(ws->tableValidEnd <= ws->allocStart);
if (ws->tableValidEnd < ws->tableEnd) {
ZSTD_memset(ws->tableValidEnd, 0, (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd));
}
ZSTD_cwksp_mark_tables_clean(ws);
}
/**
* Invalidates table allocations.
* All other allocations remain valid.
*/
MEM_STATIC void ZSTD_cwksp_clear_tables(ZSTD_cwksp* ws) {
DEBUGLOG(4, "cwksp: clearing tables!");
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* We don't do this when the workspace is statically allocated, because
* when that is the case, we have no capability to hook into the end of the
* workspace's lifecycle to unpoison the memory.
*/
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
size_t size = (BYTE*)ws->tableValidEnd - (BYTE*)ws->objectEnd;
__asan_poison_memory_region(ws->objectEnd, size);
}
#endif
ws->tableEnd = ws->objectEnd;
ZSTD_cwksp_assert_internal_consistency(ws);
}
/**
* Invalidates all buffer, aligned, and table allocations.
* Object allocations remain valid.
*/
MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
DEBUGLOG(4, "cwksp: clearing!");
#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
/* To validate that the context re-use logic is sound, and that we don't
* access stuff that this compression hasn't initialized, we re-"poison"
* the workspace except for the areas in which we expect memory re-use
* without initialization (objects, valid tables area and init once
* memory). */
{
if((BYTE*)ws->tableValidEnd < (BYTE*)ws->initOnceStart) {
size_t size = (BYTE*)ws->initOnceStart - (BYTE*)ws->tableValidEnd;
__msan_poison(ws->tableValidEnd, size);
}
}
#endif
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* We don't do this when the workspace is statically allocated, because
* when that is the case, we have no capability to hook into the end of the
* workspace's lifecycle to unpoison the memory.
*/
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
size_t size = (BYTE*)ws->workspaceEnd - (BYTE*)ws->objectEnd;
__asan_poison_memory_region(ws->objectEnd, size);
}
#endif
ws->tableEnd = ws->objectEnd;
ws->allocStart = ZSTD_cwksp_initialAllocStart(ws);
ws->allocFailed = 0;
if (ws->phase > ZSTD_cwksp_alloc_aligned_init_once) {
ws->phase = ZSTD_cwksp_alloc_aligned_init_once;
}
ZSTD_cwksp_assert_internal_consistency(ws);
}
MEM_STATIC size_t ZSTD_cwksp_sizeof(const ZSTD_cwksp* ws) {
return (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->workspace);
}
MEM_STATIC size_t ZSTD_cwksp_used(const ZSTD_cwksp* ws) {
return (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->workspace)
+ (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->allocStart);
}
/**
* The provided workspace takes ownership of the buffer [start, start+size).
* Any existing values in the workspace are ignored (the previously managed
* buffer, if present, must be separately freed).
*/
MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size, ZSTD_cwksp_static_alloc_e isStatic) {
DEBUGLOG(4, "cwksp: init'ing workspace with %zd bytes", size);
assert(((size_t)start & (sizeof(void*)-1)) == 0); /* ensure correct alignment */
ws->workspace = start;
ws->workspaceEnd = (BYTE*)start + size;
ws->objectEnd = ws->workspace;
ws->tableValidEnd = ws->objectEnd;
ws->initOnceStart = ZSTD_cwksp_initialAllocStart(ws);
ws->phase = ZSTD_cwksp_alloc_objects;
ws->isStatic = isStatic;
ZSTD_cwksp_clear(ws);
ws->workspaceOversizedDuration = 0;
ZSTD_cwksp_assert_internal_consistency(ws);
}
MEM_STATIC size_t ZSTD_cwksp_create(ZSTD_cwksp* ws, size_t size, ZSTD_customMem customMem) {
void* workspace = ZSTD_customMalloc(size, customMem);
DEBUGLOG(4, "cwksp: creating new workspace with %zd bytes", size);
RETURN_ERROR_IF(workspace == NULL, memory_allocation, "NULL pointer!");
ZSTD_cwksp_init(ws, workspace, size, ZSTD_cwksp_dynamic_alloc);
return 0;
}
MEM_STATIC void ZSTD_cwksp_free(ZSTD_cwksp* ws, ZSTD_customMem customMem) {
void *ptr = ws->workspace;
DEBUGLOG(4, "cwksp: freeing workspace");
#if ZSTD_MEMORY_SANITIZER && !defined(ZSTD_MSAN_DONT_POISON_WORKSPACE)
if (ptr != NULL && customMem.customFree != NULL) {
__msan_unpoison(ptr, ZSTD_cwksp_sizeof(ws));
}
#endif
ZSTD_memset(ws, 0, sizeof(ZSTD_cwksp));
ZSTD_customFree(ptr, customMem);
}
/**
* Moves the management of a workspace from one cwksp to another. The src cwksp
* is left in an invalid state (src must be re-init()'ed before it's used again).
*/
MEM_STATIC void ZSTD_cwksp_move(ZSTD_cwksp* dst, ZSTD_cwksp* src) {
*dst = *src;
ZSTD_memset(src, 0, sizeof(ZSTD_cwksp));
}
MEM_STATIC int ZSTD_cwksp_reserve_failed(const ZSTD_cwksp* ws) {
return ws->allocFailed;
}
/*-*************************************
* Functions Checking Free Space
***************************************/
/* ZSTD_alignmentSpaceWithinBounds() :
* Returns if the estimated space needed for a wksp is within an acceptable limit of the
* actual amount of space used.
*/
MEM_STATIC int ZSTD_cwksp_estimated_space_within_bounds(const ZSTD_cwksp *const ws, size_t const estimatedSpace) {
/* We have an alignment space between objects and tables between tables and buffers, so we can have up to twice
* the alignment bytes difference between estimation and actual usage */
return (estimatedSpace - ZSTD_cwksp_slack_space_required()) <= ZSTD_cwksp_used(ws) &&
ZSTD_cwksp_used(ws) <= estimatedSpace;
}
MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws) {
return (size_t)((BYTE*)ws->allocStart - (BYTE*)ws->tableEnd);
}
MEM_STATIC int ZSTD_cwksp_check_available(ZSTD_cwksp* ws, size_t additionalNeededSpace) {
return ZSTD_cwksp_available_space(ws) >= additionalNeededSpace;
}
MEM_STATIC int ZSTD_cwksp_check_too_large(ZSTD_cwksp* ws, size_t additionalNeededSpace) {
return ZSTD_cwksp_check_available(
ws, additionalNeededSpace * ZSTD_WORKSPACETOOLARGE_FACTOR);
}
MEM_STATIC int ZSTD_cwksp_check_wasteful(ZSTD_cwksp* ws, size_t additionalNeededSpace) {
return ZSTD_cwksp_check_too_large(ws, additionalNeededSpace)
&& ws->workspaceOversizedDuration > ZSTD_WORKSPACETOOLARGE_MAXDURATION;
}
MEM_STATIC void ZSTD_cwksp_bump_oversized_duration(
ZSTD_cwksp* ws, size_t additionalNeededSpace) {
if (ZSTD_cwksp_check_too_large(ws, additionalNeededSpace)) {
ws->workspaceOversizedDuration++;
} else {
ws->workspaceOversizedDuration = 0;
}
}
#if defined (__cplusplus)
}
#endif
#endif /* ZSTD_CWKSP_H */
/**** ended inlining zstd_cwksp.h ****/
#ifdef ZSTD_MULTITHREAD
/**** start inlining zstdmt_compress.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTDMT_COMPRESS_H
#define ZSTDMT_COMPRESS_H
#if defined (__cplusplus)
extern "C" {
#endif
/* Note : This is an internal API.
* These APIs used to be exposed with ZSTDLIB_API,
* because it used to be the only way to invoke MT compression.
* Now, you must use ZSTD_compress2 and ZSTD_compressStream2() instead.
*
* This API requires ZSTD_MULTITHREAD to be defined during compilation,
* otherwise ZSTDMT_createCCtx*() will fail.
*/
/* === Dependencies === */
/**** skipping file: ../common/zstd_deps.h ****/
#define ZSTD_STATIC_LINKING_ONLY /* ZSTD_parameters */
/**** skipping file: ../zstd.h ****/
/* === Constants === */
#ifndef ZSTDMT_NBWORKERS_MAX /* a different value can be selected at compile time */
# define ZSTDMT_NBWORKERS_MAX ((sizeof(void*)==4) /*32-bit*/ ? 64 : 256)
#endif
#ifndef ZSTDMT_JOBSIZE_MIN /* a different value can be selected at compile time */
# define ZSTDMT_JOBSIZE_MIN (512 KB)
#endif
#define ZSTDMT_JOBLOG_MAX (MEM_32bits() ? 29 : 30)
#define ZSTDMT_JOBSIZE_MAX (MEM_32bits() ? (512 MB) : (1024 MB))
/* ========================================================
* === Private interface, for use by ZSTD_compress.c ===
* === Not exposed in libzstd. Never invoke directly ===
* ======================================================== */
/* === Memory management === */
typedef struct ZSTDMT_CCtx_s ZSTDMT_CCtx;
/* Requires ZSTD_MULTITHREAD to be defined during compilation, otherwise it will return NULL. */
ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers,
ZSTD_customMem cMem,
ZSTD_threadPool *pool);
size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx);
size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx);
/* === Streaming functions === */
size_t ZSTDMT_nextInputSizeHint(const ZSTDMT_CCtx* mtctx);
/*! ZSTDMT_initCStream_internal() :
* Private use only. Init streaming operation.
* expects params to be valid.
* must receive dict, or cdict, or none, but not both.
* mtctx can be freshly constructed or reused from a prior compression.
* If mtctx is reused, memory allocations from the prior compression may not be freed,
* even if they are not needed for the current compression.
* @return : 0, or an error code */
size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* mtctx,
const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
const ZSTD_CDict* cdict,
ZSTD_CCtx_params params, unsigned long long pledgedSrcSize);
/*! ZSTDMT_compressStream_generic() :
* Combines ZSTDMT_compressStream() with optional ZSTDMT_flushStream() or ZSTDMT_endStream()
* depending on flush directive.
* @return : minimum amount of data still to be flushed
* 0 if fully flushed
* or an error code
* note : needs to be init using any ZSTD_initCStream*() variant */
size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
ZSTD_outBuffer* output,
ZSTD_inBuffer* input,
ZSTD_EndDirective endOp);
/*! ZSTDMT_toFlushNow()
* Tell how many bytes are ready to be flushed immediately.
* Probe the oldest active job (not yet entirely flushed) and check its output buffer.
* If return 0, it means there is no active job,
* or, it means oldest job is still active, but everything produced has been flushed so far,
* therefore flushing is limited by speed of oldest job. */
size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx);
/*! ZSTDMT_updateCParams_whileCompressing() :
* Updates only a selected set of compression parameters, to remain compatible with current frame.
* New parameters will be applied to next compression job. */
void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams);
/*! ZSTDMT_getFrameProgression():
* tells how much data has been consumed (input) and produced (output) for current frame.
* able to count progression inside worker threads.
*/
ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx);
#if defined (__cplusplus)
}
#endif
#endif /* ZSTDMT_COMPRESS_H */
/**** ended inlining zstdmt_compress.h ****/
#endif
/**** skipping file: ../common/bits.h ****/
#if defined (__cplusplus)
extern "C" {
#endif
/*-*************************************
* Constants
***************************************/
#define kSearchStrength 8
#define HASH_READ_SIZE 8
#define ZSTD_DUBT_UNSORTED_MARK 1 /* For btlazy2 strategy, index ZSTD_DUBT_UNSORTED_MARK==1 means "unsorted".
It could be confused for a real successor at index "1", if sorted as larger than its predecessor.
It's not a big deal though : candidate will just be sorted again.
Additionally, candidate position 1 will be lost.
But candidate 1 cannot hide a large tree of candidates, so it's a minimal loss.
The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be mishandled after table re-use with a different strategy.
This constant is required by ZSTD_compressBlock_btlazy2() and ZSTD_reduceTable_internal() */
/*-*************************************
* Context memory management
***************************************/
typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
typedef enum { zcss_init=0, zcss_load, zcss_flush } ZSTD_cStreamStage;
typedef struct ZSTD_prefixDict_s {
const void* dict;
size_t dictSize;
ZSTD_dictContentType_e dictContentType;
} ZSTD_prefixDict;
typedef struct {
void* dictBuffer;
void const* dict;
size_t dictSize;
ZSTD_dictContentType_e dictContentType;
ZSTD_CDict* cdict;
} ZSTD_localDict;
typedef struct {
HUF_CElt CTable[HUF_CTABLE_SIZE_ST(255)];
HUF_repeat repeatMode;
} ZSTD_hufCTables_t;
typedef struct {
FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
FSE_repeat offcode_repeatMode;
FSE_repeat matchlength_repeatMode;
FSE_repeat litlength_repeatMode;
} ZSTD_fseCTables_t;
typedef struct {
ZSTD_hufCTables_t huf;
ZSTD_fseCTables_t fse;
} ZSTD_entropyCTables_t;
/***********************************************
* Entropy buffer statistics structs and funcs *
***********************************************/
/** ZSTD_hufCTablesMetadata_t :
* Stores Literals Block Type for a super-block in hType, and
* huffman tree description in hufDesBuffer.
* hufDesSize refers to the size of huffman tree description in bytes.
* This metadata is populated in ZSTD_buildBlockEntropyStats_literals() */
typedef struct {
symbolEncodingType_e hType;
BYTE hufDesBuffer[ZSTD_MAX_HUF_HEADER_SIZE];
size_t hufDesSize;
} ZSTD_hufCTablesMetadata_t;
/** ZSTD_fseCTablesMetadata_t :
* Stores symbol compression modes for a super-block in {ll, ol, ml}Type, and
* fse tables in fseTablesBuffer.
* fseTablesSize refers to the size of fse tables in bytes.
* This metadata is populated in ZSTD_buildBlockEntropyStats_sequences() */
typedef struct {
symbolEncodingType_e llType;
symbolEncodingType_e ofType;
symbolEncodingType_e mlType;
BYTE fseTablesBuffer[ZSTD_MAX_FSE_HEADERS_SIZE];
size_t fseTablesSize;
size_t lastCountSize; /* This is to account for bug in 1.3.4. More detail in ZSTD_entropyCompressSeqStore_internal() */
} ZSTD_fseCTablesMetadata_t;
typedef struct {
ZSTD_hufCTablesMetadata_t hufMetadata;
ZSTD_fseCTablesMetadata_t fseMetadata;
} ZSTD_entropyCTablesMetadata_t;
/** ZSTD_buildBlockEntropyStats() :
* Builds entropy for the block.
* @return : 0 on success or error code */
size_t ZSTD_buildBlockEntropyStats(
const seqStore_t* seqStorePtr,
const ZSTD_entropyCTables_t* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
ZSTD_entropyCTablesMetadata_t* entropyMetadata,
void* workspace, size_t wkspSize);
/*********************************
* Compression internals structs *
*********************************/
typedef struct {
U32 off; /* Offset sumtype code for the match, using ZSTD_storeSeq() format */
U32 len; /* Raw length of match */
} ZSTD_match_t;
typedef struct {
U32 offset; /* Offset of sequence */
U32 litLength; /* Length of literals prior to match */
U32 matchLength; /* Raw length of match */
} rawSeq;
typedef struct {
rawSeq* seq; /* The start of the sequences */
size_t pos; /* The index in seq where reading stopped. pos <= size. */
size_t posInSequence; /* The position within the sequence at seq[pos] where reading
stopped. posInSequence <= seq[pos].litLength + seq[pos].matchLength */
size_t size; /* The number of sequences. <= capacity. */
size_t capacity; /* The capacity starting from `seq` pointer */
} rawSeqStore_t;
typedef struct {
U32 idx; /* Index in array of ZSTD_Sequence */
U32 posInSequence; /* Position within sequence at idx */
size_t posInSrc; /* Number of bytes given by sequences provided so far */
} ZSTD_sequencePosition;
UNUSED_ATTR static const rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0, 0};
typedef struct {
int price;
U32 off;
U32 mlen;
U32 litlen;
U32 rep[ZSTD_REP_NUM];
} ZSTD_optimal_t;
typedef enum { zop_dynamic=0, zop_predef } ZSTD_OptPrice_e;
typedef struct {
/* All tables are allocated inside cctx->workspace by ZSTD_resetCCtx_internal() */
unsigned* litFreq; /* table of literals statistics, of size 256 */
unsigned* litLengthFreq; /* table of litLength statistics, of size (MaxLL+1) */
unsigned* matchLengthFreq; /* table of matchLength statistics, of size (MaxML+1) */
unsigned* offCodeFreq; /* table of offCode statistics, of size (MaxOff+1) */
ZSTD_match_t* matchTable; /* list of found matches, of size ZSTD_OPT_NUM+1 */
ZSTD_optimal_t* priceTable; /* All positions tracked by optimal parser, of size ZSTD_OPT_NUM+1 */
U32 litSum; /* nb of literals */
U32 litLengthSum; /* nb of litLength codes */
U32 matchLengthSum; /* nb of matchLength codes */
U32 offCodeSum; /* nb of offset codes */
U32 litSumBasePrice; /* to compare to log2(litfreq) */
U32 litLengthSumBasePrice; /* to compare to log2(llfreq) */
U32 matchLengthSumBasePrice;/* to compare to log2(mlfreq) */
U32 offCodeSumBasePrice; /* to compare to log2(offreq) */
ZSTD_OptPrice_e priceType; /* prices can be determined dynamically, or follow a pre-defined cost structure */
const ZSTD_entropyCTables_t* symbolCosts; /* pre-calculated dictionary statistics */
ZSTD_paramSwitch_e literalCompressionMode;
} optState_t;
typedef struct {
ZSTD_entropyCTables_t entropy;
U32 rep[ZSTD_REP_NUM];
} ZSTD_compressedBlockState_t;
typedef struct {
BYTE const* nextSrc; /* next block here to continue on current prefix */
BYTE const* base; /* All regular indexes relative to this position */
BYTE const* dictBase; /* extDict indexes relative to this position */
U32 dictLimit; /* below that point, need extDict */
U32 lowLimit; /* below that point, no more valid data */
U32 nbOverflowCorrections; /* Number of times overflow correction has run since
* ZSTD_window_init(). Useful for debugging coredumps
* and for ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY.
*/
} ZSTD_window_t;
#define ZSTD_WINDOW_START_INDEX 2
typedef struct ZSTD_matchState_t ZSTD_matchState_t;
#define ZSTD_ROW_HASH_CACHE_SIZE 8 /* Size of prefetching hash cache for row-based matchfinder */
struct ZSTD_matchState_t {
ZSTD_window_t window; /* State for window round buffer management */
U32 loadedDictEnd; /* index of end of dictionary, within context's referential.
* When loadedDictEnd != 0, a dictionary is in use, and still valid.
* This relies on a mechanism to set loadedDictEnd=0 when dictionary is no longer within distance.
* Such mechanism is provided within ZSTD_window_enforceMaxDist() and ZSTD_checkDictValidity().
* When dict referential is copied into active context (i.e. not attached),
* loadedDictEnd == dictSize, since referential starts from zero.
*/
U32 nextToUpdate; /* index from which to continue table update */
U32 hashLog3; /* dispatch table for matches of len==3 : larger == faster, more memory */
U32 rowHashLog; /* For row-based matchfinder: Hashlog based on nb of rows in the hashTable.*/
BYTE* tagTable; /* For row-based matchFinder: A row-based table containing the hashes and head index. */
U32 hashCache[ZSTD_ROW_HASH_CACHE_SIZE]; /* For row-based matchFinder: a cache of hashes to improve speed */
U64 hashSalt; /* For row-based matchFinder: salts the hash for re-use of tag table */
U32 hashSaltEntropy; /* For row-based matchFinder: collects entropy for salt generation */
U32* hashTable;
U32* hashTable3;
U32* chainTable;
U32 forceNonContiguous; /* Non-zero if we should force non-contiguous load for the next window update. */
int dedicatedDictSearch; /* Indicates whether this matchState is using the
* dedicated dictionary search structure.
*/
optState_t opt; /* optimal parser state */
const ZSTD_matchState_t* dictMatchState;
ZSTD_compressionParameters cParams;
const rawSeqStore_t* ldmSeqStore;
/* Controls prefetching in some dictMatchState matchfinders.
* This behavior is controlled from the cctx ms.
* This parameter has no effect in the cdict ms. */
int prefetchCDictTables;
/* When == 0, lazy match finders insert every position.
* When != 0, lazy match finders only insert positions they search.
* This allows them to skip much faster over incompressible data,
* at a small cost to compression ratio.
*/
int lazySkipping;
};
typedef struct {
ZSTD_compressedBlockState_t* prevCBlock;
ZSTD_compressedBlockState_t* nextCBlock;
ZSTD_matchState_t matchState;
} ZSTD_blockState_t;
typedef struct {
U32 offset;
U32 checksum;
} ldmEntry_t;
typedef struct {
BYTE const* split;
U32 hash;
U32 checksum;
ldmEntry_t* bucket;
} ldmMatchCandidate_t;
#define LDM_BATCH_SIZE 64
typedef struct {
ZSTD_window_t window; /* State for the window round buffer management */
ldmEntry_t* hashTable;
U32 loadedDictEnd;
BYTE* bucketOffsets; /* Next position in bucket to insert entry */
size_t splitIndices[LDM_BATCH_SIZE];
ldmMatchCandidate_t matchCandidates[LDM_BATCH_SIZE];
} ldmState_t;
typedef struct {
ZSTD_paramSwitch_e enableLdm; /* ZSTD_ps_enable to enable LDM. ZSTD_ps_auto by default */
U32 hashLog; /* Log size of hashTable */
U32 bucketSizeLog; /* Log bucket size for collision resolution, at most 8 */
U32 minMatchLength; /* Minimum match length */
U32 hashRateLog; /* Log number of entries to skip */
U32 windowLog; /* Window log for the LDM */
} ldmParams_t;
typedef struct {
int collectSequences;
ZSTD_Sequence* seqStart;
size_t seqIndex;
size_t maxSequences;
} SeqCollector;
struct ZSTD_CCtx_params_s {
ZSTD_format_e format;
ZSTD_compressionParameters cParams;
ZSTD_frameParameters fParams;
int compressionLevel;
int forceWindow; /* force back-references to respect limit of
* 1<<wLog, even for dictionary */
size_t targetCBlockSize; /* Tries to fit compressed block size to be around targetCBlockSize.
* No target when targetCBlockSize == 0.
* There is no guarantee on compressed block size */
int srcSizeHint; /* User's best guess of source size.
* Hint is not valid when srcSizeHint == 0.
* There is no guarantee that hint is close to actual source size */
ZSTD_dictAttachPref_e attachDictPref;
ZSTD_paramSwitch_e literalCompressionMode;
/* Multithreading: used to pass parameters to mtctx */
int nbWorkers;
size_t jobSize;
int overlapLog;
int rsyncable;
/* Long distance matching parameters */
ldmParams_t ldmParams;
/* Dedicated dict search algorithm trigger */
int enableDedicatedDictSearch;
/* Input/output buffer modes */
ZSTD_bufferMode_e inBufferMode;
ZSTD_bufferMode_e outBufferMode;
/* Sequence compression API */
ZSTD_sequenceFormat_e blockDelimiters;
int validateSequences;
/* Block splitting */
ZSTD_paramSwitch_e useBlockSplitter;
/* Param for deciding whether to use row-based matchfinder */
ZSTD_paramSwitch_e useRowMatchFinder;
/* Always load a dictionary in ext-dict mode (not prefix mode)? */
int deterministicRefPrefix;
/* Internal use, for createCCtxParams() and freeCCtxParams() only */
ZSTD_customMem customMem;
/* Controls prefetching in some dictMatchState matchfinders */
ZSTD_paramSwitch_e prefetchCDictTables;
/* Controls whether zstd will fall back to an internal matchfinder
* if the external matchfinder returns an error code. */
int enableMatchFinderFallback;
/* Indicates whether an external matchfinder has been referenced.
* Users can't set this externally.
* It is set internally in ZSTD_registerSequenceProducer(). */
int useSequenceProducer;
/* Adjust the max block size*/
size_t maxBlockSize;
/* Controls repcode search in external sequence parsing */
ZSTD_paramSwitch_e searchForExternalRepcodes;
}; /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
#define COMPRESS_SEQUENCES_WORKSPACE_SIZE (sizeof(unsigned) * (MaxSeq + 2))
#define ENTROPY_WORKSPACE_SIZE (HUF_WORKSPACE_SIZE + COMPRESS_SEQUENCES_WORKSPACE_SIZE)
/**
* Indicates whether this compression proceeds directly from user-provided
* source buffer to user-provided destination buffer (ZSTDb_not_buffered), or
* whether the context needs to buffer the input/output (ZSTDb_buffered).
*/
typedef enum {
ZSTDb_not_buffered,
ZSTDb_buffered
} ZSTD_buffered_policy_e;
/**
* Struct that contains all elements of block splitter that should be allocated
* in a wksp.
*/
#define ZSTD_MAX_NB_BLOCK_SPLITS 196
typedef struct {
seqStore_t fullSeqStoreChunk;
seqStore_t firstHalfSeqStore;
seqStore_t secondHalfSeqStore;
seqStore_t currSeqStore;
seqStore_t nextSeqStore;
U32 partitions[ZSTD_MAX_NB_BLOCK_SPLITS];
ZSTD_entropyCTablesMetadata_t entropyMetadata;
} ZSTD_blockSplitCtx;
/* Context for block-level external matchfinder API */
typedef struct {
void* mState;
ZSTD_sequenceProducer_F* mFinder;
ZSTD_Sequence* seqBuffer;
size_t seqBufferCapacity;
} ZSTD_externalMatchCtx;
struct ZSTD_CCtx_s {
ZSTD_compressionStage_e stage;
int cParamsChanged; /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
int bmi2; /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
ZSTD_CCtx_params requestedParams;
ZSTD_CCtx_params appliedParams;
ZSTD_CCtx_params simpleApiParams; /* Param storage used by the simple API - not sticky. Must only be used in top-level simple API functions for storage. */
U32 dictID;
size_t dictContentSize;
ZSTD_cwksp workspace; /* manages buffer for dynamic allocations */
size_t blockSize;
unsigned long long pledgedSrcSizePlusOne; /* this way, 0 (default) == unknown */
unsigned long long consumedSrcSize;
unsigned long long producedCSize;
XXH64_state_t xxhState;
ZSTD_customMem customMem;
ZSTD_threadPool* pool;
size_t staticSize;
SeqCollector seqCollector;
int isFirstBlock;
int initialized;
seqStore_t seqStore; /* sequences storage ptrs */
ldmState_t ldmState; /* long distance matching state */
rawSeq* ldmSequences; /* Storage for the ldm output sequences */
size_t maxNbLdmSequences;
rawSeqStore_t externSeqStore; /* Mutable reference to external sequences */
ZSTD_blockState_t blockState;
U32* entropyWorkspace; /* entropy workspace of ENTROPY_WORKSPACE_SIZE bytes */
/* Whether we are streaming or not */
ZSTD_buffered_policy_e bufferedPolicy;
/* streaming */
char* inBuff;
size_t inBuffSize;
size_t inToCompress;
size_t inBuffPos;
size_t inBuffTarget;
char* outBuff;
size_t outBuffSize;
size_t outBuffContentSize;
size_t outBuffFlushedSize;
ZSTD_cStreamStage streamStage;
U32 frameEnded;
/* Stable in/out buffer verification */
ZSTD_inBuffer expectedInBuffer;
size_t stableIn_notConsumed; /* nb bytes within stable input buffer that are said to be consumed but are not */
size_t expectedOutBufferSize;
/* Dictionary */
ZSTD_localDict localDict;
const ZSTD_CDict* cdict;
ZSTD_prefixDict prefixDict; /* single-usage dictionary */
/* Multi-threading */
#ifdef ZSTD_MULTITHREAD
ZSTDMT_CCtx* mtctx;
#endif
/* Tracing */
#if ZSTD_TRACE
ZSTD_TraceCtx traceCtx;
#endif
/* Workspace for block splitter */
ZSTD_blockSplitCtx blockSplitCtx;
/* Workspace for external matchfinder */
ZSTD_externalMatchCtx externalMatchCtx;
};
typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
typedef enum { ZSTD_tfp_forCCtx, ZSTD_tfp_forCDict } ZSTD_tableFillPurpose_e;
typedef enum {
ZSTD_noDict = 0,
ZSTD_extDict = 1,
ZSTD_dictMatchState = 2,
ZSTD_dedicatedDictSearch = 3
} ZSTD_dictMode_e;
typedef enum {
ZSTD_cpm_noAttachDict = 0, /* Compression with ZSTD_noDict or ZSTD_extDict.
* In this mode we use both the srcSize and the dictSize
* when selecting and adjusting parameters.
*/
ZSTD_cpm_attachDict = 1, /* Compression with ZSTD_dictMatchState or ZSTD_dedicatedDictSearch.
* In this mode we only take the srcSize into account when selecting
* and adjusting parameters.
*/
ZSTD_cpm_createCDict = 2, /* Creating a CDict.
* In this mode we take both the source size and the dictionary size
* into account when selecting and adjusting the parameters.
*/
ZSTD_cpm_unknown = 3 /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
* We don't know what these parameters are for. We default to the legacy
* behavior of taking both the source size and the dict size into account
* when selecting and adjusting parameters.
*/
} ZSTD_cParamMode_e;
typedef size_t (*ZSTD_blockCompressor) (
ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramSwitch_e rowMatchfinderMode, ZSTD_dictMode_e dictMode);
MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
{
static const BYTE LL_Code[64] = { 0, 1, 2, 3, 4, 5, 6, 7,
8, 9, 10, 11, 12, 13, 14, 15,
16, 16, 17, 17, 18, 18, 19, 19,
20, 20, 20, 20, 21, 21, 21, 21,
22, 22, 22, 22, 22, 22, 22, 22,
23, 23, 23, 23, 23, 23, 23, 23,
24, 24, 24, 24, 24, 24, 24, 24,
24, 24, 24, 24, 24, 24, 24, 24 };
static const U32 LL_deltaCode = 19;
return (litLength > 63) ? ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
}
/* ZSTD_MLcode() :
* note : mlBase = matchLength - MINMATCH;
* because it's the format it's stored in seqStore->sequences */
MEM_STATIC U32 ZSTD_MLcode(U32 mlBase)
{
static const BYTE ML_Code[128] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37,
38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39,
40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41,
42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42,
42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42 };
static const U32 ML_deltaCode = 36;
return (mlBase > 127) ? ZSTD_highbit32(mlBase) + ML_deltaCode : ML_Code[mlBase];
}
/* ZSTD_cParam_withinBounds:
* @return 1 if value is within cParam bounds,
* 0 otherwise */
MEM_STATIC int ZSTD_cParam_withinBounds(ZSTD_cParameter cParam, int value)
{
ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);
if (ZSTD_isError(bounds.error)) return 0;
if (value < bounds.lowerBound) return 0;
if (value > bounds.upperBound) return 0;
return 1;
}
/* ZSTD_noCompressBlock() :
* Writes uncompressed block to dst buffer from given src.
* Returns the size of the block */
MEM_STATIC size_t
ZSTD_noCompressBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
{
U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(srcSize << 3);
DEBUGLOG(5, "ZSTD_noCompressBlock (srcSize=%zu, dstCapacity=%zu)", srcSize, dstCapacity);
RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
dstSize_tooSmall, "dst buf too small for uncompressed block");
MEM_writeLE24(dst, cBlockHeader24);
ZSTD_memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
return ZSTD_blockHeaderSize + srcSize;
}
MEM_STATIC size_t
ZSTD_rleCompressBlock(void* dst, size_t dstCapacity, BYTE src, size_t srcSize, U32 lastBlock)
{
BYTE* const op = (BYTE*)dst;
U32 const cBlockHeader = lastBlock + (((U32)bt_rle)<<1) + (U32)(srcSize << 3);
RETURN_ERROR_IF(dstCapacity < 4, dstSize_tooSmall, "");
MEM_writeLE24(op, cBlockHeader);
op[3] = src;
return 4;
}
/* ZSTD_minGain() :
* minimum compression required
* to generate a compress block or a compressed literals section.
* note : use same formula for both situations */
MEM_STATIC size_t ZSTD_minGain(size_t srcSize, ZSTD_strategy strat)
{
U32 const minlog = (strat>=ZSTD_btultra) ? (U32)(strat) - 1 : 6;
ZSTD_STATIC_ASSERT(ZSTD_btultra == 8);
assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, (int)strat));
return (srcSize >> minlog) + 2;
}
MEM_STATIC int ZSTD_literalsCompressionIsDisabled(const ZSTD_CCtx_params* cctxParams)
{
switch (cctxParams->literalCompressionMode) {
case ZSTD_ps_enable:
return 0;
case ZSTD_ps_disable:
return 1;
default:
assert(0 /* impossible: pre-validated */);
ZSTD_FALLTHROUGH;
case ZSTD_ps_auto:
return (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
}
}
/*! ZSTD_safecopyLiterals() :
* memcpy() function that won't read beyond more than WILDCOPY_OVERLENGTH bytes past ilimit_w.
* Only called when the sequence ends past ilimit_w, so it only needs to be optimized for single
* large copies.
*/
static void
ZSTD_safecopyLiterals(BYTE* op, BYTE const* ip, BYTE const* const iend, BYTE const* ilimit_w)
{
assert(iend > ilimit_w);
if (ip <= ilimit_w) {
ZSTD_wildcopy(op, ip, ilimit_w - ip, ZSTD_no_overlap);
op += ilimit_w - ip;
ip = ilimit_w;
}
while (ip < iend) *op++ = *ip++;
}
#define REPCODE1_TO_OFFBASE REPCODE_TO_OFFBASE(1)
#define REPCODE2_TO_OFFBASE REPCODE_TO_OFFBASE(2)
#define REPCODE3_TO_OFFBASE REPCODE_TO_OFFBASE(3)
#define REPCODE_TO_OFFBASE(r) (assert((r)>=1), assert((r)<=ZSTD_REP_NUM), (r)) /* accepts IDs 1,2,3 */
#define OFFSET_TO_OFFBASE(o) (assert((o)>0), o + ZSTD_REP_NUM)
#define OFFBASE_IS_OFFSET(o) ((o) > ZSTD_REP_NUM)
#define OFFBASE_IS_REPCODE(o) ( 1 <= (o) && (o) <= ZSTD_REP_NUM)
#define OFFBASE_TO_OFFSET(o) (assert(OFFBASE_IS_OFFSET(o)), (o) - ZSTD_REP_NUM)
#define OFFBASE_TO_REPCODE(o) (assert(OFFBASE_IS_REPCODE(o)), (o)) /* returns ID 1,2,3 */
/*! ZSTD_storeSeq() :
* Store a sequence (litlen, litPtr, offBase and matchLength) into seqStore_t.
* @offBase : Users should employ macros REPCODE_TO_OFFBASE() and OFFSET_TO_OFFBASE().
* @matchLength : must be >= MINMATCH
* Allowed to over-read literals up to litLimit.
*/
HINT_INLINE UNUSED_ATTR void
ZSTD_storeSeq(seqStore_t* seqStorePtr,
size_t litLength, const BYTE* literals, const BYTE* litLimit,
U32 offBase,
size_t matchLength)
{
BYTE const* const litLimit_w = litLimit - WILDCOPY_OVERLENGTH;
BYTE const* const litEnd = literals + litLength;
#if defined(DEBUGLEVEL) && (DEBUGLEVEL >= 6)
static const BYTE* g_start = NULL;
if (g_start==NULL) g_start = (const BYTE*)literals; /* note : index only works for compression within a single segment */
{ U32 const pos = (U32)((const BYTE*)literals - g_start);
DEBUGLOG(6, "Cpos%7u :%3u literals, match%4u bytes at offBase%7u",
pos, (U32)litLength, (U32)matchLength, (U32)offBase);
}
#endif
assert((size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart) < seqStorePtr->maxNbSeq);
/* copy Literals */
assert(seqStorePtr->maxNbLit <= 128 KB);
assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + seqStorePtr->maxNbLit);
assert(literals + litLength <= litLimit);
if (litEnd <= litLimit_w) {
/* Common case we can use wildcopy.
* First copy 16 bytes, because literals are likely short.
*/
ZSTD_STATIC_ASSERT(WILDCOPY_OVERLENGTH >= 16);
ZSTD_copy16(seqStorePtr->lit, literals);
if (litLength > 16) {
ZSTD_wildcopy(seqStorePtr->lit+16, literals+16, (ptrdiff_t)litLength-16, ZSTD_no_overlap);
}
} else {
ZSTD_safecopyLiterals(seqStorePtr->lit, literals, litEnd, litLimit_w);
}
seqStorePtr->lit += litLength;
/* literal Length */
if (litLength>0xFFFF) {
assert(seqStorePtr->longLengthType == ZSTD_llt_none); /* there can only be a single long length */
seqStorePtr->longLengthType = ZSTD_llt_literalLength;
seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
}
seqStorePtr->sequences[0].litLength = (U16)litLength;
/* match offset */
seqStorePtr->sequences[0].offBase = offBase;
/* match Length */
assert(matchLength >= MINMATCH);
{ size_t const mlBase = matchLength - MINMATCH;
if (mlBase>0xFFFF) {
assert(seqStorePtr->longLengthType == ZSTD_llt_none); /* there can only be a single long length */
seqStorePtr->longLengthType = ZSTD_llt_matchLength;
seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
}
seqStorePtr->sequences[0].mlBase = (U16)mlBase;
}
seqStorePtr->sequences++;
}
/* ZSTD_updateRep() :
* updates in-place @rep (array of repeat offsets)
* @offBase : sum-type, using numeric representation of ZSTD_storeSeq()
*/
MEM_STATIC void
ZSTD_updateRep(U32 rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
{
if (OFFBASE_IS_OFFSET(offBase)) { /* full offset */
rep[2] = rep[1];
rep[1] = rep[0];
rep[0] = OFFBASE_TO_OFFSET(offBase);
} else { /* repcode */
U32 const repCode = OFFBASE_TO_REPCODE(offBase) - 1 + ll0;
if (repCode > 0) { /* note : if repCode==0, no change */
U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
rep[2] = (repCode >= 2) ? rep[1] : rep[2];
rep[1] = rep[0];
rep[0] = currentOffset;
} else { /* repCode == 0 */
/* nothing to do */
}
}
}
typedef struct repcodes_s {
U32 rep[3];
} repcodes_t;
MEM_STATIC repcodes_t
ZSTD_newRep(U32 const rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
{
repcodes_t newReps;
ZSTD_memcpy(&newReps, rep, sizeof(newReps));
ZSTD_updateRep(newReps.rep, offBase, ll0);
return newReps;
}
/*-*************************************
* Match length counter
***************************************/
MEM_STATIC size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
{
const BYTE* const pStart = pIn;
const BYTE* const pInLoopLimit = pInLimit - (sizeof(size_t)-1);
if (pIn < pInLoopLimit) {
{ size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
if (diff) return ZSTD_NbCommonBytes(diff); }
pIn+=sizeof(size_t); pMatch+=sizeof(size_t);
while (pIn < pInLoopLimit) {
size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
if (!diff) { pIn+=sizeof(size_t); pMatch+=sizeof(size_t); continue; }
pIn += ZSTD_NbCommonBytes(diff);
return (size_t)(pIn - pStart);
} }
if (MEM_64bits() && (pIn<(pInLimit-3)) && (MEM_read32(pMatch) == MEM_read32(pIn))) { pIn+=4; pMatch+=4; }
if ((pIn<(pInLimit-1)) && (MEM_read16(pMatch) == MEM_read16(pIn))) { pIn+=2; pMatch+=2; }
if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
return (size_t)(pIn - pStart);
}
/** ZSTD_count_2segments() :
* can count match length with `ip` & `match` in 2 different segments.
* convention : on reaching mEnd, match count continue starting from iStart
*/
MEM_STATIC size_t
ZSTD_count_2segments(const BYTE* ip, const BYTE* match,
const BYTE* iEnd, const BYTE* mEnd, const BYTE* iStart)
{
const BYTE* const vEnd = MIN( ip + (mEnd - match), iEnd);
size_t const matchLength = ZSTD_count(ip, match, vEnd);
if (match + matchLength != mEnd) return matchLength;
DEBUGLOG(7, "ZSTD_count_2segments: found a 2-parts match (current length==%zu)", matchLength);
DEBUGLOG(7, "distance from match beginning to end dictionary = %zi", mEnd - match);
DEBUGLOG(7, "distance from current pos to end buffer = %zi", iEnd - ip);
DEBUGLOG(7, "next byte : ip==%02X, istart==%02X", ip[matchLength], *iStart);
DEBUGLOG(7, "final match length = %zu", matchLength + ZSTD_count(ip+matchLength, iStart, iEnd));
return matchLength + ZSTD_count(ip+matchLength, iStart, iEnd);
}
/*-*************************************
* Hashes
***************************************/
static const U32 prime3bytes = 506832829U;
static U32 ZSTD_hash3(U32 u, U32 h, U32 s) { assert(h <= 32); return (((u << (32-24)) * prime3bytes) ^ s) >> (32-h) ; }
MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h, 0); } /* only in zstd_opt.h */
MEM_STATIC size_t ZSTD_hash3PtrS(const void* ptr, U32 h, U32 s) { return ZSTD_hash3(MEM_readLE32(ptr), h, s); }
static const U32 prime4bytes = 2654435761U;
static U32 ZSTD_hash4(U32 u, U32 h, U32 s) { assert(h <= 32); return ((u * prime4bytes) ^ s) >> (32-h) ; }
static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_readLE32(ptr), h, 0); }
static size_t ZSTD_hash4PtrS(const void* ptr, U32 h, U32 s) { return ZSTD_hash4(MEM_readLE32(ptr), h, s); }
static const U64 prime5bytes = 889523592379ULL;
static size_t ZSTD_hash5(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u << (64-40)) * prime5bytes) ^ s) >> (64-h)) ; }
static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h, 0); }
static size_t ZSTD_hash5PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash5(MEM_readLE64(p), h, s); }
static const U64 prime6bytes = 227718039650203ULL;
static size_t ZSTD_hash6(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u << (64-48)) * prime6bytes) ^ s) >> (64-h)) ; }
static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h, 0); }
static size_t ZSTD_hash6PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash6(MEM_readLE64(p), h, s); }
static const U64 prime7bytes = 58295818150454627ULL;
static size_t ZSTD_hash7(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u << (64-56)) * prime7bytes) ^ s) >> (64-h)) ; }
static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h, 0); }
static size_t ZSTD_hash7PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash7(MEM_readLE64(p), h, s); }
static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
static size_t ZSTD_hash8(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u) * prime8bytes) ^ s) >> (64-h)) ; }
static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h, 0); }
static size_t ZSTD_hash8PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash8(MEM_readLE64(p), h, s); }
MEM_STATIC FORCE_INLINE_ATTR
size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
{
/* Although some of these hashes do support hBits up to 64, some do not.
* To be on the safe side, always avoid hBits > 32. */
assert(hBits <= 32);
switch(mls)
{
default:
case 4: return ZSTD_hash4Ptr(p, hBits);
case 5: return ZSTD_hash5Ptr(p, hBits);
case 6: return ZSTD_hash6Ptr(p, hBits);
case 7: return ZSTD_hash7Ptr(p, hBits);
case 8: return ZSTD_hash8Ptr(p, hBits);
}
}
MEM_STATIC FORCE_INLINE_ATTR
size_t ZSTD_hashPtrSalted(const void* p, U32 hBits, U32 mls, const U64 hashSalt) {
/* Although some of these hashes do support hBits up to 64, some do not.
* To be on the safe side, always avoid hBits > 32. */
assert(hBits <= 32);
switch(mls)
{
default:
case 4: return ZSTD_hash4PtrS(p, hBits, (U32)hashSalt);
case 5: return ZSTD_hash5PtrS(p, hBits, hashSalt);
case 6: return ZSTD_hash6PtrS(p, hBits, hashSalt);
case 7: return ZSTD_hash7PtrS(p, hBits, hashSalt);
case 8: return ZSTD_hash8PtrS(p, hBits, hashSalt);
}
}
/** ZSTD_ipow() :
* Return base^exponent.
*/
static U64 ZSTD_ipow(U64 base, U64 exponent)
{
U64 power = 1;
while (exponent) {
if (exponent & 1) power *= base;
exponent >>= 1;
base *= base;
}
return power;
}
#define ZSTD_ROLL_HASH_CHAR_OFFSET 10
/** ZSTD_rollingHash_append() :
* Add the buffer to the hash value.
*/
static U64 ZSTD_rollingHash_append(U64 hash, void const* buf, size_t size)
{
BYTE const* istart = (BYTE const*)buf;
size_t pos;
for (pos = 0; pos < size; ++pos) {
hash *= prime8bytes;
hash += istart[pos] + ZSTD_ROLL_HASH_CHAR_OFFSET;
}
return hash;
}
/** ZSTD_rollingHash_compute() :
* Compute the rolling hash value of the buffer.
*/
MEM_STATIC U64 ZSTD_rollingHash_compute(void const* buf, size_t size)
{
return ZSTD_rollingHash_append(0, buf, size);
}
/** ZSTD_rollingHash_primePower() :
* Compute the primePower to be passed to ZSTD_rollingHash_rotate() for a hash
* over a window of length bytes.
*/
MEM_STATIC U64 ZSTD_rollingHash_primePower(U32 length)
{
return ZSTD_ipow(prime8bytes, length - 1);
}
/** ZSTD_rollingHash_rotate() :
* Rotate the rolling hash by one byte.
*/
MEM_STATIC U64 ZSTD_rollingHash_rotate(U64 hash, BYTE toRemove, BYTE toAdd, U64 primePower)
{
hash -= (toRemove + ZSTD_ROLL_HASH_CHAR_OFFSET) * primePower;
hash *= prime8bytes;
hash += toAdd + ZSTD_ROLL_HASH_CHAR_OFFSET;
return hash;
}
/*-*************************************
* Round buffer management
***************************************/
#if (ZSTD_WINDOWLOG_MAX_64 > 31)
# error "ZSTD_WINDOWLOG_MAX is too large : would overflow ZSTD_CURRENT_MAX"
#endif
/* Max current allowed */
#define ZSTD_CURRENT_MAX ((3U << 29) + (1U << ZSTD_WINDOWLOG_MAX))
/* Maximum chunk size before overflow correction needs to be called again */
#define ZSTD_CHUNKSIZE_MAX \
( ((U32)-1) /* Maximum ending current index */ \
- ZSTD_CURRENT_MAX) /* Maximum beginning lowLimit */
/**
* ZSTD_window_clear():
* Clears the window containing the history by simply setting it to empty.
*/
MEM_STATIC void ZSTD_window_clear(ZSTD_window_t* window)
{
size_t const endT = (size_t)(window->nextSrc - window->base);
U32 const end = (U32)endT;
window->lowLimit = end;
window->dictLimit = end;
}
MEM_STATIC U32 ZSTD_window_isEmpty(ZSTD_window_t const window)
{
return window.dictLimit == ZSTD_WINDOW_START_INDEX &&
window.lowLimit == ZSTD_WINDOW_START_INDEX &&
(window.nextSrc - window.base) == ZSTD_WINDOW_START_INDEX;
}
/**
* ZSTD_window_hasExtDict():
* Returns non-zero if the window has a non-empty extDict.
*/
MEM_STATIC U32 ZSTD_window_hasExtDict(ZSTD_window_t const window)
{
return window.lowLimit < window.dictLimit;
}
/**
* ZSTD_matchState_dictMode():
* Inspects the provided matchState and figures out what dictMode should be
* passed to the compressor.
*/
MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
{
return ZSTD_window_hasExtDict(ms->window) ?
ZSTD_extDict :
ms->dictMatchState != NULL ?
(ms->dictMatchState->dedicatedDictSearch ? ZSTD_dedicatedDictSearch : ZSTD_dictMatchState) :
ZSTD_noDict;
}
/* Defining this macro to non-zero tells zstd to run the overflow correction
* code much more frequently. This is very inefficient, and should only be
* used for tests and fuzzers.
*/
#ifndef ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY
# ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
# define ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY 1
# else
# define ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY 0
# endif
#endif
/**
* ZSTD_window_canOverflowCorrect():
* Returns non-zero if the indices are large enough for overflow correction
* to work correctly without impacting compression ratio.
*/
MEM_STATIC U32 ZSTD_window_canOverflowCorrect(ZSTD_window_t const window,
U32 cycleLog,
U32 maxDist,
U32 loadedDictEnd,
void const* src)
{
U32 const cycleSize = 1u << cycleLog;
U32 const curr = (U32)((BYTE const*)src - window.base);
U32 const minIndexToOverflowCorrect = cycleSize
+ MAX(maxDist, cycleSize)
+ ZSTD_WINDOW_START_INDEX;
/* Adjust the min index to backoff the overflow correction frequency,
* so we don't waste too much CPU in overflow correction. If this
* computation overflows we don't really care, we just need to make
* sure it is at least minIndexToOverflowCorrect.
*/
U32 const adjustment = window.nbOverflowCorrections + 1;
U32 const adjustedIndex = MAX(minIndexToOverflowCorrect * adjustment,
minIndexToOverflowCorrect);
U32 const indexLargeEnough = curr > adjustedIndex;
/* Only overflow correct early if the dictionary is invalidated already,
* so we don't hurt compression ratio.
*/
U32 const dictionaryInvalidated = curr > maxDist + loadedDictEnd;
return indexLargeEnough && dictionaryInvalidated;
}
/**
* ZSTD_window_needOverflowCorrection():
* Returns non-zero if the indices are getting too large and need overflow
* protection.
*/
MEM_STATIC U32 ZSTD_window_needOverflowCorrection(ZSTD_window_t const window,
U32 cycleLog,
U32 maxDist,
U32 loadedDictEnd,
void const* src,
void const* srcEnd)
{
U32 const curr = (U32)((BYTE const*)srcEnd - window.base);
if (ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY) {
if (ZSTD_window_canOverflowCorrect(window, cycleLog, maxDist, loadedDictEnd, src)) {
return 1;
}
}
return curr > ZSTD_CURRENT_MAX;
}
/**
* ZSTD_window_correctOverflow():
* Reduces the indices to protect from index overflow.
* Returns the correction made to the indices, which must be applied to every
* stored index.
*
* The least significant cycleLog bits of the indices must remain the same,
* which may be 0. Every index up to maxDist in the past must be valid.
*/
MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
U32 maxDist, void const* src)
{
/* preemptive overflow correction:
* 1. correction is large enough:
* lowLimit > (3<<29) ==> current > 3<<29 + 1<<windowLog
* 1<<windowLog <= newCurrent < 1<<chainLog + 1<<windowLog
*
* current - newCurrent
* > (3<<29 + 1<<windowLog) - (1<<windowLog + 1<<chainLog)
* > (3<<29) - (1<<chainLog)
* > (3<<29) - (1<<30) (NOTE: chainLog <= 30)
* > 1<<29
*
* 2. (ip+ZSTD_CHUNKSIZE_MAX - cctx->base) doesn't overflow:
* After correction, current is less than (1<<chainLog + 1<<windowLog).
* In 64-bit mode we are safe, because we have 64-bit ptrdiff_t.
* In 32-bit mode we are safe, because (chainLog <= 29), so
* ip+ZSTD_CHUNKSIZE_MAX - cctx->base < 1<<32.
* 3. (cctx->lowLimit + 1<<windowLog) < 1<<32:
* windowLog <= 31 ==> 3<<29 + 1<<windowLog < 7<<29 < 1<<32.
*/
U32 const cycleSize = 1u << cycleLog;
U32 const cycleMask = cycleSize - 1;
U32 const curr = (U32)((BYTE const*)src - window->base);
U32 const currentCycle = curr & cycleMask;
/* Ensure newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX. */
U32 const currentCycleCorrection = currentCycle < ZSTD_WINDOW_START_INDEX
? MAX(cycleSize, ZSTD_WINDOW_START_INDEX)
: 0;
U32 const newCurrent = currentCycle
+ currentCycleCorrection
+ MAX(maxDist, cycleSize);
U32 const correction = curr - newCurrent;
/* maxDist must be a power of two so that:
* (newCurrent & cycleMask) == (curr & cycleMask)
* This is required to not corrupt the chains / binary tree.
*/
assert((maxDist & (maxDist - 1)) == 0);
assert((curr & cycleMask) == (newCurrent & cycleMask));
assert(curr > newCurrent);
if (!ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY) {
/* Loose bound, should be around 1<<29 (see above) */
assert(correction > 1<<28);
}
window->base += correction;
window->dictBase += correction;
if (window->lowLimit < correction + ZSTD_WINDOW_START_INDEX) {
window->lowLimit = ZSTD_WINDOW_START_INDEX;
} else {
window->lowLimit -= correction;
}
if (window->dictLimit < correction + ZSTD_WINDOW_START_INDEX) {
window->dictLimit = ZSTD_WINDOW_START_INDEX;
} else {
window->dictLimit -= correction;
}
/* Ensure we can still reference the full window. */
assert(newCurrent >= maxDist);
assert(newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX);
/* Ensure that lowLimit and dictLimit didn't underflow. */
assert(window->lowLimit <= newCurrent);
assert(window->dictLimit <= newCurrent);
++window->nbOverflowCorrections;
DEBUGLOG(4, "Correction of 0x%x bytes to lowLimit=0x%x", correction,
window->lowLimit);
return correction;
}
/**
* ZSTD_window_enforceMaxDist():
* Updates lowLimit so that:
* (srcEnd - base) - lowLimit == maxDist + loadedDictEnd
*
* It ensures index is valid as long as index >= lowLimit.
* This must be called before a block compression call.
*
* loadedDictEnd is only defined if a dictionary is in use for current compression.
* As the name implies, loadedDictEnd represents the index at end of dictionary.
* The value lies within context's referential, it can be directly compared to blockEndIdx.
*
* If loadedDictEndPtr is NULL, no dictionary is in use, and we use loadedDictEnd == 0.
* If loadedDictEndPtr is not NULL, we set it to zero after updating lowLimit.
* This is because dictionaries are allowed to be referenced fully
* as long as the last byte of the dictionary is in the window.
* Once input has progressed beyond window size, dictionary cannot be referenced anymore.
*
* In normal dict mode, the dictionary lies between lowLimit and dictLimit.
* In dictMatchState mode, lowLimit and dictLimit are the same,
* and the dictionary is below them.
* forceWindow and dictMatchState are therefore incompatible.
*/
MEM_STATIC void
ZSTD_window_enforceMaxDist(ZSTD_window_t* window,
const void* blockEnd,
U32 maxDist,
U32* loadedDictEndPtr,
const ZSTD_matchState_t** dictMatchStatePtr)
{
U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
U32 const loadedDictEnd = (loadedDictEndPtr != NULL) ? *loadedDictEndPtr : 0;
DEBUGLOG(5, "ZSTD_window_enforceMaxDist: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
(unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
/* - When there is no dictionary : loadedDictEnd == 0.
In which case, the test (blockEndIdx > maxDist) is merely to avoid
overflowing next operation `newLowLimit = blockEndIdx - maxDist`.
- When there is a standard dictionary :
Index referential is copied from the dictionary,
which means it starts from 0.
In which case, loadedDictEnd == dictSize,
and it makes sense to compare `blockEndIdx > maxDist + dictSize`
since `blockEndIdx` also starts from zero.
- When there is an attached dictionary :
loadedDictEnd is expressed within the referential of the context,
so it can be directly compared against blockEndIdx.
*/
if (blockEndIdx > maxDist + loadedDictEnd) {
U32 const newLowLimit = blockEndIdx - maxDist;
if (window->lowLimit < newLowLimit) window->lowLimit = newLowLimit;
if (window->dictLimit < window->lowLimit) {
DEBUGLOG(5, "Update dictLimit to match lowLimit, from %u to %u",
(unsigned)window->dictLimit, (unsigned)window->lowLimit);
window->dictLimit = window->lowLimit;
}
/* On reaching window size, dictionaries are invalidated */
if (loadedDictEndPtr) *loadedDictEndPtr = 0;
if (dictMatchStatePtr) *dictMatchStatePtr = NULL;
}
}
/* Similar to ZSTD_window_enforceMaxDist(),
* but only invalidates dictionary
* when input progresses beyond window size.
* assumption : loadedDictEndPtr and dictMatchStatePtr are valid (non NULL)
* loadedDictEnd uses same referential as window->base
* maxDist is the window size */
MEM_STATIC void
ZSTD_checkDictValidity(const ZSTD_window_t* window,
const void* blockEnd,
U32 maxDist,
U32* loadedDictEndPtr,
const ZSTD_matchState_t** dictMatchStatePtr)
{
assert(loadedDictEndPtr != NULL);
assert(dictMatchStatePtr != NULL);
{ U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
U32 const loadedDictEnd = *loadedDictEndPtr;
DEBUGLOG(5, "ZSTD_checkDictValidity: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
(unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
assert(blockEndIdx >= loadedDictEnd);
if (blockEndIdx > loadedDictEnd + maxDist || loadedDictEnd != window->dictLimit) {
/* On reaching window size, dictionaries are invalidated.
* For simplification, if window size is reached anywhere within next block,
* the dictionary is invalidated for the full block.
*
* We also have to invalidate the dictionary if ZSTD_window_update() has detected
* non-contiguous segments, which means that loadedDictEnd != window->dictLimit.
* loadedDictEnd may be 0, if forceWindow is true, but in that case we never use
* dictMatchState, so setting it to NULL is not a problem.
*/
DEBUGLOG(6, "invalidating dictionary for current block (distance > windowSize)");
*loadedDictEndPtr = 0;
*dictMatchStatePtr = NULL;
} else {
if (*loadedDictEndPtr != 0) {
DEBUGLOG(6, "dictionary considered valid for current block");
} } }
}
MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
ZSTD_memset(window, 0, sizeof(*window));
window->base = (BYTE const*)" ";
window->dictBase = (BYTE const*)" ";
ZSTD_STATIC_ASSERT(ZSTD_DUBT_UNSORTED_MARK < ZSTD_WINDOW_START_INDEX); /* Start above ZSTD_DUBT_UNSORTED_MARK */
window->dictLimit = ZSTD_WINDOW_START_INDEX; /* start from >0, so that 1st position is valid */
window->lowLimit = ZSTD_WINDOW_START_INDEX; /* it ensures first and later CCtx usages compress the same */
window->nextSrc = window->base + ZSTD_WINDOW_START_INDEX; /* see issue #1241 */
window->nbOverflowCorrections = 0;
}
/**
* ZSTD_window_update():
* Updates the window by appending [src, src + srcSize) to the window.
* If it is not contiguous, the current prefix becomes the extDict, and we
* forget about the extDict. Handles overlap of the prefix and extDict.
* Returns non-zero if the segment is contiguous.
*/
MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
void const* src, size_t srcSize,
int forceNonContiguous)
{
BYTE const* const ip = (BYTE const*)src;
U32 contiguous = 1;
DEBUGLOG(5, "ZSTD_window_update");
if (srcSize == 0)
return contiguous;
assert(window->base != NULL);
assert(window->dictBase != NULL);
/* Check if blocks follow each other */
if (src != window->nextSrc || forceNonContiguous) {
/* not contiguous */
size_t const distanceFromBase = (size_t)(window->nextSrc - window->base);
DEBUGLOG(5, "Non contiguous blocks, new segment starts at %u", window->dictLimit);
window->lowLimit = window->dictLimit;
assert(distanceFromBase == (size_t)(U32)distanceFromBase); /* should never overflow */
window->dictLimit = (U32)distanceFromBase;
window->dictBase = window->base;
window->base = ip - distanceFromBase;
/* ms->nextToUpdate = window->dictLimit; */
if (window->dictLimit - window->lowLimit < HASH_READ_SIZE) window->lowLimit = window->dictLimit; /* too small extDict */
contiguous = 0;
}
window->nextSrc = ip + srcSize;
/* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
if ( (ip+srcSize > window->dictBase + window->lowLimit)
& (ip < window->dictBase + window->dictLimit)) {
ptrdiff_t const highInputIdx = (ip + srcSize) - window->dictBase;
U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)window->dictLimit) ? window->dictLimit : (U32)highInputIdx;
window->lowLimit = lowLimitMax;
DEBUGLOG(5, "Overlapping extDict and input : new lowLimit = %u", window->lowLimit);
}
return contiguous;
}
/**
* Returns the lowest allowed match index. It may either be in the ext-dict or the prefix.
*/
MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
{
U32 const maxDistance = 1U << windowLog;
U32 const lowestValid = ms->window.lowLimit;
U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
U32 const isDictionary = (ms->loadedDictEnd != 0);
/* When using a dictionary the entire dictionary is valid if a single byte of the dictionary
* is within the window. We invalidate the dictionary (and set loadedDictEnd to 0) when it isn't
* valid for the entire block. So this check is sufficient to find the lowest valid match index.
*/
U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
return matchLowest;
}
/**
* Returns the lowest allowed match index in the prefix.
*/
MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
{
U32 const maxDistance = 1U << windowLog;
U32 const lowestValid = ms->window.dictLimit;
U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
U32 const isDictionary = (ms->loadedDictEnd != 0);
/* When computing the lowest prefix index we need to take the dictionary into account to handle
* the edge case where the dictionary and the source are contiguous in memory.
*/
U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
return matchLowest;
}
/* debug functions */
#if (DEBUGLEVEL>=2)
MEM_STATIC double ZSTD_fWeight(U32 rawStat)
{
U32 const fp_accuracy = 8;
U32 const fp_multiplier = (1 << fp_accuracy);
U32 const newStat = rawStat + 1;
U32 const hb = ZSTD_highbit32(newStat);
U32 const BWeight = hb * fp_multiplier;
U32 const FWeight = (newStat << fp_accuracy) >> hb;
U32 const weight = BWeight + FWeight;
assert(hb + fp_accuracy < 31);
return (double)weight / fp_multiplier;
}
/* display a table content,
* listing each element, its frequency, and its predicted bit cost */
MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
{
unsigned u, sum;
for (u=0, sum=0; u<=max; u++) sum += table[u];
DEBUGLOG(2, "total nb elts: %u", sum);
for (u=0; u<=max; u++) {
DEBUGLOG(2, "%2u: %5u (%.2f)",
u, table[u], ZSTD_fWeight(sum) - ZSTD_fWeight(table[u]) );
}
}
#endif
/* Short Cache */
/* Normally, zstd matchfinders follow this flow:
* 1. Compute hash at ip
* 2. Load index from hashTable[hash]
* 3. Check if *ip == *(base + index)
* In dictionary compression, loading *(base + index) is often an L2 or even L3 miss.
*
* Short cache is an optimization which allows us to avoid step 3 most of the time
* when the data doesn't actually match. With short cache, the flow becomes:
* 1. Compute (hash, currentTag) at ip. currentTag is an 8-bit independent hash at ip.
* 2. Load (index, matchTag) from hashTable[hash]. See ZSTD_writeTaggedIndex to understand how this works.
* 3. Only if currentTag == matchTag, check *ip == *(base + index). Otherwise, continue.
*
* Currently, short cache is only implemented in CDict hashtables. Thus, its use is limited to
* dictMatchState matchfinders.
*/
#define ZSTD_SHORT_CACHE_TAG_BITS 8
#define ZSTD_SHORT_CACHE_TAG_MASK ((1u << ZSTD_SHORT_CACHE_TAG_BITS) - 1)
/* Helper function for ZSTD_fillHashTable and ZSTD_fillDoubleHashTable.
* Unpacks hashAndTag into (hash, tag), then packs (index, tag) into hashTable[hash]. */
MEM_STATIC void ZSTD_writeTaggedIndex(U32* const hashTable, size_t hashAndTag, U32 index) {
size_t const hash = hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
U32 const tag = (U32)(hashAndTag & ZSTD_SHORT_CACHE_TAG_MASK);
assert(index >> (32 - ZSTD_SHORT_CACHE_TAG_BITS) == 0);
hashTable[hash] = (index << ZSTD_SHORT_CACHE_TAG_BITS) | tag;
}
/* Helper function for short cache matchfinders.
* Unpacks tag1 and tag2 from lower bits of packedTag1 and packedTag2, then checks if the tags match. */
MEM_STATIC int ZSTD_comparePackedTags(size_t packedTag1, size_t packedTag2) {
U32 const tag1 = packedTag1 & ZSTD_SHORT_CACHE_TAG_MASK;
U32 const tag2 = packedTag2 & ZSTD_SHORT_CACHE_TAG_MASK;
return tag1 == tag2;
}
#if defined (__cplusplus)
}
#endif
/* ===============================================================
* Shared internal declarations
* These prototypes may be called from sources not in lib/compress
* =============================================================== */
/* ZSTD_loadCEntropy() :
* dict : must point at beginning of a valid zstd dictionary.
* return : size of dictionary header (size of magic number + dict ID + entropy tables)
* assumptions : magic number supposed already checked
* and dictSize >= 8 */
size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
const void* const dict, size_t dictSize);
void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
/* ==============================================================
* Private declarations
* These prototypes shall only be called from within lib/compress
* ============================================================== */
/* ZSTD_getCParamsFromCCtxParams() :
* cParams are built depending on compressionLevel, src size hints,
* LDM and manually set compression parameters.
* Note: srcSizeHint == 0 means 0!
*/
ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
/*! ZSTD_initCStream_internal() :
* Private use only. Init streaming operation.
* expects params to be valid.
* must receive dict, or cdict, or none, but not both.
* @return : 0, or an error code */
size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
const void* dict, size_t dictSize,
const ZSTD_CDict* cdict,
const ZSTD_CCtx_params* params, unsigned long long pledgedSrcSize);
void ZSTD_resetSeqStore(seqStore_t* ssPtr);
/*! ZSTD_getCParamsFromCDict() :
* as the name implies */
ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict);
/* ZSTD_compressBegin_advanced_internal() :
* Private use only. To be called from zstdmt_compress.c. */
size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
const void* dict, size_t dictSize,
ZSTD_dictContentType_e dictContentType,
ZSTD_dictTableLoadMethod_e dtlm,
const ZSTD_CDict* cdict,
const ZSTD_CCtx_params* params,
unsigned long long pledgedSrcSize);
/* ZSTD_compress_advanced_internal() :
* Private use only. To be called from zstdmt_compress.c. */
size_t ZSTD_compress_advanced_internal(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const void* dict,size_t dictSize,
const ZSTD_CCtx_params* params);
/* ZSTD_writeLastEmptyBlock() :
* output an empty Block with end-of-frame mark to complete a frame
* @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
* or an error code if `dstCapacity` is too small (<ZSTD_blockHeaderSize)
*/
size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity);
/* ZSTD_referenceExternalSequences() :
* Must be called before starting a compression operation.
* seqs must parse a prefix of the source.
* This cannot be used when long range matching is enabled.
* Zstd will use these sequences, and pass the literals to a secondary block
* compressor.
* NOTE: seqs are not verified! Invalid sequences can cause out-of-bounds memory
* access and data corruption.
*/
void ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq);
/** ZSTD_cycleLog() :
* condition for correct operation : hashLog > 1 */
U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat);
/** ZSTD_CCtx_trace() :
* Trace the end of a compression call.
*/
void ZSTD_CCtx_trace(ZSTD_CCtx* cctx, size_t extraCSize);
/* Returns 0 on success, and a ZSTD_error otherwise. This function scans through an array of
* ZSTD_Sequence, storing the sequences it finds, until it reaches a block delimiter.
* Note that the block delimiter must include the last literals of the block.
*/
size_t
ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
ZSTD_sequencePosition* seqPos,
const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
/* Returns the number of bytes to move the current read position back by.
* Only non-zero if we ended up splitting a sequence.
* Otherwise, it may return a ZSTD error if something went wrong.
*
* This function will attempt to scan through blockSize bytes
* represented by the sequences in @inSeqs,
* storing any (partial) sequences.
*
* Occasionally, we may want to change the actual number of bytes we consumed from inSeqs to
* avoid splitting a match, or to avoid splitting a match such that it would produce a match
* smaller than MINMATCH. In this case, we return the number of bytes that we didn't read from this block.
*/
size_t
ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
/* ===============================================================
* Deprecated definitions that are still used internally to avoid
* deprecation warnings. These functions are exactly equivalent to
* their public variants, but avoid the deprecation warnings.
* =============================================================== */
size_t ZSTD_compressBegin_usingCDict_deprecated(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
size_t ZSTD_compressContinue_public(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize);
size_t ZSTD_compressEnd_public(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize);
size_t ZSTD_compressBlock_deprecated(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
#endif /* ZSTD_COMPRESS_H */
/**** ended inlining zstd_compress_internal.h ****/
size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
/* ZSTD_compressRleLiteralsBlock() :
* Conditions :
* - All bytes in @src are identical
* - dstCapacity >= 4 */
size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
/* ZSTD_compressLiterals():
* @entropyWorkspace: must be aligned on 4-bytes boundaries
* @entropyWorkspaceSize : must be >= HUF_WORKSPACE_SIZE
* @suspectUncompressible: sampling checks, to potentially skip huffman coding
*/
size_t ZSTD_compressLiterals (void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
void* entropyWorkspace, size_t entropyWorkspaceSize,
const ZSTD_hufCTables_t* prevHuf,
ZSTD_hufCTables_t* nextHuf,
ZSTD_strategy strategy, int disableLiteralCompression,
int suspectUncompressible,
int bmi2);
#endif /* ZSTD_COMPRESS_LITERALS_H */
/**** ended inlining zstd_compress_literals.h ****/
/* **************************************************************
* Debug Traces
****************************************************************/
#if DEBUGLEVEL >= 2
static size_t showHexa(const void* src, size_t srcSize)
{
const BYTE* const ip = (const BYTE*)src;
size_t u;
for (u=0; u<srcSize; u++) {
RAWLOG(5, " %02X", ip[u]); (void)ip;
}
RAWLOG(5, " \n");
return srcSize;
}
#endif
/* **************************************************************
* Literals compression - special cases
****************************************************************/
size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
{
BYTE* const ostart = (BYTE*)dst;
U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
DEBUGLOG(5, "ZSTD_noCompressLiterals: srcSize=%zu, dstCapacity=%zu", srcSize, dstCapacity);
RETURN_ERROR_IF(srcSize + flSize > dstCapacity, dstSize_tooSmall, "");
switch(flSize)
{
case 1: /* 2 - 1 - 5 */
ostart[0] = (BYTE)((U32)set_basic + (srcSize<<3));
break;
case 2: /* 2 - 2 - 12 */
MEM_writeLE16(ostart, (U16)((U32)set_basic + (1<<2) + (srcSize<<4)));
break;
case 3: /* 2 - 2 - 20 */
MEM_writeLE32(ostart, (U32)((U32)set_basic + (3<<2) + (srcSize<<4)));
break;
default: /* not necessary : flSize is {1,2,3} */
assert(0);
}
ZSTD_memcpy(ostart + flSize, src, srcSize);
DEBUGLOG(5, "Raw (uncompressed) literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
return srcSize + flSize;
}
static int allBytesIdentical(const void* src, size_t srcSize)
{
assert(srcSize >= 1);
assert(src != NULL);
{ const BYTE b = ((const BYTE*)src)[0];
size_t p;
for (p=1; p<srcSize; p++) {
if (((const BYTE*)src)[p] != b) return 0;
}
return 1;
}
}
size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
{
BYTE* const ostart = (BYTE*)dst;
U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
assert(dstCapacity >= 4); (void)dstCapacity;
assert(allBytesIdentical(src, srcSize));
switch(flSize)
{
case 1: /* 2 - 1 - 5 */
ostart[0] = (BYTE)((U32)set_rle + (srcSize<<3));
break;
case 2: /* 2 - 2 - 12 */
MEM_writeLE16(ostart, (U16)((U32)set_rle + (1<<2) + (srcSize<<4)));
break;
case 3: /* 2 - 2 - 20 */
MEM_writeLE32(ostart, (U32)((U32)set_rle + (3<<2) + (srcSize<<4)));
break;
default: /* not necessary : flSize is {1,2,3} */
assert(0);
}
ostart[flSize] = *(const BYTE*)src;
DEBUGLOG(5, "RLE : Repeated Literal (%02X: %u times) -> %u bytes encoded", ((const BYTE*)src)[0], (U32)srcSize, (U32)flSize + 1);
return flSize+1;
}
/* ZSTD_minLiteralsToCompress() :
* returns minimal amount of literals
* for literal compression to even be attempted.
* Minimum is made tighter as compression strategy increases.
*/
static size_t
ZSTD_minLiteralsToCompress(ZSTD_strategy strategy, HUF_repeat huf_repeat)
{
assert((int)strategy >= 0);
assert((int)strategy <= 9);
/* btultra2 : min 8 bytes;
* then 2x larger for each successive compression strategy
* max threshold 64 bytes */
{ int const shift = MIN(9-(int)strategy, 3);
size_t const mintc = (huf_repeat == HUF_repeat_valid) ? 6 : (size_t)8 << shift;
DEBUGLOG(7, "minLiteralsToCompress = %zu", mintc);
return mintc;
}
}
size_t ZSTD_compressLiterals (
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
void* entropyWorkspace, size_t entropyWorkspaceSize,
const ZSTD_hufCTables_t* prevHuf,
ZSTD_hufCTables_t* nextHuf,
ZSTD_strategy strategy,
int disableLiteralCompression,
int suspectUncompressible,
int bmi2)
{
size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
BYTE* const ostart = (BYTE*)dst;
U32 singleStream = srcSize < 256;
symbolEncodingType_e hType = set_compressed;
size_t cLitSize;
DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i, srcSize=%u, dstCapacity=%zu)",
disableLiteralCompression, (U32)srcSize, dstCapacity);
DEBUGLOG(6, "Completed literals listing (%zu bytes)", showHexa(src, srcSize));
/* Prepare nextEntropy assuming reusing the existing table */
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
if (disableLiteralCompression)
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
/* if too small, don't even attempt compression (speed opt) */
if (srcSize < ZSTD_minLiteralsToCompress(strategy, prevHuf->repeatMode))
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
RETURN_ERROR_IF(dstCapacity < lhSize+1, dstSize_tooSmall, "not enough space for compression");
{ HUF_repeat repeat = prevHuf->repeatMode;
int const flags = 0
| (bmi2 ? HUF_flags_bmi2 : 0)
| (strategy < ZSTD_lazy && srcSize <= 1024 ? HUF_flags_preferRepeat : 0)
| (strategy >= HUF_OPTIMAL_DEPTH_THRESHOLD ? HUF_flags_optimalDepth : 0)
| (suspectUncompressible ? HUF_flags_suspectUncompressible : 0);
typedef size_t (*huf_compress_f)(void*, size_t, const void*, size_t, unsigned, unsigned, void*, size_t, HUF_CElt*, HUF_repeat*, int);
huf_compress_f huf_compress;
if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
huf_compress = singleStream ? HUF_compress1X_repeat : HUF_compress4X_repeat;
cLitSize = huf_compress(ostart+lhSize, dstCapacity-lhSize,
src, srcSize,
HUF_SYMBOLVALUE_MAX, LitHufLog,
entropyWorkspace, entropyWorkspaceSize,
(HUF_CElt*)nextHuf->CTable,
&repeat, flags);
DEBUGLOG(5, "%zu literals compressed into %zu bytes (before header)", srcSize, cLitSize);
if (repeat != HUF_repeat_none) {
/* reused the existing table */
DEBUGLOG(5, "reusing statistics from previous huffman block");
hType = set_repeat;
}
}
{ size_t const minGain = ZSTD_minGain(srcSize, strategy);
if ((cLitSize==0) || (cLitSize >= srcSize - minGain) || ERR_isError(cLitSize)) {
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
} }
if (cLitSize==1) {
/* A return value of 1 signals that the alphabet consists of a single symbol.
* However, in some rare circumstances, it could be the compressed size (a single byte).
* For that outcome to have a chance to happen, it's necessary that `srcSize < 8`.
* (it's also necessary to not generate statistics).
* Therefore, in such a case, actively check that all bytes are identical. */
if ((srcSize >= 8) || allBytesIdentical(src, srcSize)) {
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
} }
if (hType == set_compressed) {
/* using a newly constructed table */
nextHuf->repeatMode = HUF_repeat_check;
}
/* Build header */
switch(lhSize)
{
case 3: /* 2 - 2 - 10 - 10 */
if (!singleStream) assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
{ U32 const lhc = hType + ((U32)(!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
MEM_writeLE24(ostart, lhc);
break;
}
case 4: /* 2 - 2 - 14 - 14 */
assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
{ U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
MEM_writeLE32(ostart, lhc);
break;
}
case 5: /* 2 - 2 - 18 - 18 */
assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
{ U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
MEM_writeLE32(ostart, lhc);
ostart[4] = (BYTE)(cLitSize >> 10);
break;
}
default: /* not possible : lhSize is {3,4,5} */
assert(0);
}
DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)srcSize, (U32)(lhSize+cLitSize));
return lhSize+cLitSize;
}
/**** ended inlining compress/zstd_compress_literals.c ****/
/**** start inlining compress/zstd_compress_sequences.c ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/*-*************************************
* Dependencies
***************************************/
/**** start inlining zstd_compress_sequences.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_COMPRESS_SEQUENCES_H
#define ZSTD_COMPRESS_SEQUENCES_H
/**** skipping file: ../common/fse.h ****/
/**** skipping file: ../common/zstd_internal.h ****/
typedef enum {
ZSTD_defaultDisallowed = 0,
ZSTD_defaultAllowed = 1
} ZSTD_defaultPolicy_e;
symbolEncodingType_e
ZSTD_selectEncodingType(
FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
FSE_CTable const* prevCTable,
short const* defaultNorm, U32 defaultNormLog,
ZSTD_defaultPolicy_e const isDefaultAllowed,
ZSTD_strategy const strategy);
size_t
ZSTD_buildCTable(void* dst, size_t dstCapacity,
FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
unsigned* count, U32 max,
const BYTE* codeTable, size_t nbSeq,
const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
const FSE_CTable* prevCTable, size_t prevCTableSize,
void* entropyWorkspace, size_t entropyWorkspaceSize);
size_t ZSTD_encodeSequences(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2);
size_t ZSTD_fseBitCost(
FSE_CTable const* ctable,
unsigned const* count,
unsigned const max);
size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
unsigned const* count, unsigned const max);
#endif /* ZSTD_COMPRESS_SEQUENCES_H */
/**** ended inlining zstd_compress_sequences.h ****/
/**
* -log2(x / 256) lookup table for x in [0, 256).
* If x == 0: Return 0
* Else: Return floor(-log2(x / 256) * 256)
*/
static unsigned const kInverseProbabilityLog256[256] = {
0, 2048, 1792, 1642, 1536, 1453, 1386, 1329, 1280, 1236, 1197, 1162,
1130, 1100, 1073, 1047, 1024, 1001, 980, 960, 941, 923, 906, 889,
874, 859, 844, 830, 817, 804, 791, 779, 768, 756, 745, 734,
724, 714, 704, 694, 685, 676, 667, 658, 650, 642, 633, 626,
618, 610, 603, 595, 588, 581, 574, 567, 561, 554, 548, 542,
535, 529, 523, 517, 512, 506, 500, 495, 489, 484, 478, 473,
468, 463, 458, 453, 448, 443, 438, 434, 429, 424, 420, 415,
411, 407, 402, 398, 394, 390, 386, 382, 377, 373, 370, 366,
362, 358, 354, 350, 347, 343, 339, 336, 332, 329, 325, 322,
318, 315, 311, 308, 305, 302, 298, 295, 292, 289, 286, 282,
279, 276, 273, 270, 267, 264, 261, 258, 256, 253, 250, 247,
244, 241, 239, 236, 233, 230, 228, 225, 222, 220, 217, 215,
212, 209, 207, 204, 202, 199, 197, 194, 192, 190, 187, 185,
182, 180, 178, 175, 173, 171, 168, 166, 164, 162, 159, 157,
155, 153, 151, 149, 146, 144, 142, 140, 138, 136, 134, 132,
130, 128, 126, 123, 121, 119, 117, 115, 114, 112, 110, 108,
106, 104, 102, 100, 98, 96, 94, 93, 91, 89, 87, 85,
83, 82, 80, 78, 76, 74, 73, 71, 69, 67, 66, 64,
62, 61, 59, 57, 55, 54, 52, 50, 49, 47, 46, 44,
42, 41, 39, 37, 36, 34, 33, 31, 30, 28, 26, 25,
23, 22, 20, 19, 17, 16, 14, 13, 11, 10, 8, 7,
5, 4, 2, 1,
};
static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
void const* ptr = ctable;
U16 const* u16ptr = (U16 const*)ptr;
U32 const maxSymbolValue = MEM_read16(u16ptr + 1);
return maxSymbolValue;
}
/**
* Returns true if we should use ncount=-1 else we should
* use ncount=1 for low probability symbols instead.
*/
static unsigned ZSTD_useLowProbCount(size_t const nbSeq)
{
/* Heuristic: This should cover most blocks <= 16K and
* start to fade out after 16K to about 32K depending on
* compressibility.
*/
return nbSeq >= 2048;
}
/**
* Returns the cost in bytes of encoding the normalized count header.
* Returns an error if any of the helper functions return an error.
*/
static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
size_t const nbSeq, unsigned const FSELog)
{
BYTE wksp[FSE_NCOUNTBOUND];
S16 norm[MaxSeq + 1];
const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max, ZSTD_useLowProbCount(nbSeq)), "");
return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
}
/**
* Returns the cost in bits of encoding the distribution described by count
* using the entropy bound.
*/
static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t const total)
{
unsigned cost = 0;
unsigned s;
assert(total > 0);
for (s = 0; s <= max; ++s) {
unsigned norm = (unsigned)((256 * count[s]) / total);
if (count[s] != 0 && norm == 0)
norm = 1;
assert(count[s] < total);
cost += count[s] * kInverseProbabilityLog256[norm];
}
return cost >> 8;
}
/**
* Returns the cost in bits of encoding the distribution in count using ctable.
* Returns an error if ctable cannot represent all the symbols in count.
*/
size_t ZSTD_fseBitCost(
FSE_CTable const* ctable,
unsigned const* count,
unsigned const max)
{
unsigned const kAccuracyLog = 8;
size_t cost = 0;
unsigned s;
FSE_CState_t cstate;
FSE_initCState(&cstate, ctable);
if (ZSTD_getFSEMaxSymbolValue(ctable) < max) {
DEBUGLOG(5, "Repeat FSE_CTable has maxSymbolValue %u < %u",
ZSTD_getFSEMaxSymbolValue(ctable), max);
return ERROR(GENERIC);
}
for (s = 0; s <= max; ++s) {
unsigned const tableLog = cstate.stateLog;
unsigned const badCost = (tableLog + 1) << kAccuracyLog;
unsigned const bitCost = FSE_bitCost(cstate.symbolTT, tableLog, s, kAccuracyLog);
if (count[s] == 0)
continue;
if (bitCost >= badCost) {
DEBUGLOG(5, "Repeat FSE_CTable has Prob[%u] == 0", s);
return ERROR(GENERIC);
}
cost += (size_t)count[s] * bitCost;
}
return cost >> kAccuracyLog;
}
/**
* Returns the cost in bits of encoding the distribution in count using the
* table described by norm. The max symbol support by norm is assumed >= max.
* norm must be valid for every symbol with non-zero probability in count.
*/
size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
unsigned const* count, unsigned const max)
{
unsigned const shift = 8 - accuracyLog;
size_t cost = 0;
unsigned s;
assert(accuracyLog <= 8);
for (s = 0; s <= max; ++s) {
unsigned const normAcc = (norm[s] != -1) ? (unsigned)norm[s] : 1;
unsigned const norm256 = normAcc << shift;
assert(norm256 > 0);
assert(norm256 < 256);
cost += count[s] * kInverseProbabilityLog256[norm256];
}
return cost >> 8;
}
symbolEncodingType_e
ZSTD_selectEncodingType(
FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
FSE_CTable const* prevCTable,
short const* defaultNorm, U32 defaultNormLog,
ZSTD_defaultPolicy_e const isDefaultAllowed,
ZSTD_strategy const strategy)
{
ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
if (mostFrequent == nbSeq) {
*repeatMode = FSE_repeat_none;
if (isDefaultAllowed && nbSeq <= 2) {
/* Prefer set_basic over set_rle when there are 2 or fewer symbols,
* since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
* If basic encoding isn't possible, always choose RLE.
*/
DEBUGLOG(5, "Selected set_basic");
return set_basic;
}
DEBUGLOG(5, "Selected set_rle");
return set_rle;
}
if (strategy < ZSTD_lazy) {
if (isDefaultAllowed) {
size_t const staticFse_nbSeq_max = 1000;
size_t const mult = 10 - strategy;
size_t const baseLog = 3;
size_t const dynamicFse_nbSeq_min = (((size_t)1 << defaultNormLog) * mult) >> baseLog; /* 28-36 for offset, 56-72 for lengths */
assert(defaultNormLog >= 5 && defaultNormLog <= 6); /* xx_DEFAULTNORMLOG */
assert(mult <= 9 && mult >= 7);
if ( (*repeatMode == FSE_repeat_valid)
&& (nbSeq < staticFse_nbSeq_max) ) {
DEBUGLOG(5, "Selected set_repeat");
return set_repeat;
}
if ( (nbSeq < dynamicFse_nbSeq_min)
|| (mostFrequent < (nbSeq >> (defaultNormLog-1))) ) {
DEBUGLOG(5, "Selected set_basic");
/* The format allows default tables to be repeated, but it isn't useful.
* When using simple heuristics to select encoding type, we don't want
* to confuse these tables with dictionaries. When running more careful
* analysis, we don't need to waste time checking both repeating tables
* and default tables.
*/
*repeatMode = FSE_repeat_none;
return set_basic;
}
}
} else {
size_t const basicCost = isDefaultAllowed ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, count, max) : ERROR(GENERIC);
size_t const repeatCost = *repeatMode != FSE_repeat_none ? ZSTD_fseBitCost(prevCTable, count, max) : ERROR(GENERIC);
size_t const NCountCost = ZSTD_NCountCost(count, max, nbSeq, FSELog);
size_t const compressedCost = (NCountCost << 3) + ZSTD_entropyCost(count, max, nbSeq);
if (isDefaultAllowed) {
assert(!ZSTD_isError(basicCost));
assert(!(*repeatMode == FSE_repeat_valid && ZSTD_isError(repeatCost)));
}
assert(!ZSTD_isError(NCountCost));
assert(compressedCost < ERROR(maxCode));
DEBUGLOG(5, "Estimated bit costs: basic=%u\trepeat=%u\tcompressed=%u",
(unsigned)basicCost, (unsigned)repeatCost, (unsigned)compressedCost);
if (basicCost <= repeatCost && basicCost <= compressedCost) {
DEBUGLOG(5, "Selected set_basic");
assert(isDefaultAllowed);
*repeatMode = FSE_repeat_none;
return set_basic;
}
if (repeatCost <= compressedCost) {
DEBUGLOG(5, "Selected set_repeat");
assert(!ZSTD_isError(repeatCost));
return set_repeat;
}
assert(compressedCost < basicCost && compressedCost < repeatCost);
}
DEBUGLOG(5, "Selected set_compressed");
*repeatMode = FSE_repeat_check;
return set_compressed;
}
typedef struct {
S16 norm[MaxSeq + 1];
U32 wksp[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(MaxSeq, MaxFSELog)];
} ZSTD_BuildCTableWksp;
size_t
ZSTD_buildCTable(void* dst, size_t dstCapacity,
FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
unsigned* count, U32 max,
const BYTE* codeTable, size_t nbSeq,
const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
const FSE_CTable* prevCTable, size_t prevCTableSize,
void* entropyWorkspace, size_t entropyWorkspaceSize)
{
BYTE* op = (BYTE*)dst;
const BYTE* const oend = op + dstCapacity;
DEBUGLOG(6, "ZSTD_buildCTable (dstCapacity=%u)", (unsigned)dstCapacity);
switch (type) {
case set_rle:
FORWARD_IF_ERROR(FSE_buildCTable_rle(nextCTable, (BYTE)max), "");
RETURN_ERROR_IF(dstCapacity==0, dstSize_tooSmall, "not enough space");
*op = codeTable[0];
return 1;
case set_repeat:
ZSTD_memcpy(nextCTable, prevCTable, prevCTableSize);
return 0;
case set_basic:
FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize), ""); /* note : could be pre-calculated */
return 0;
case set_compressed: {
ZSTD_BuildCTableWksp* wksp = (ZSTD_BuildCTableWksp*)entropyWorkspace;
size_t nbSeq_1 = nbSeq;
const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
if (count[codeTable[nbSeq-1]] > 1) {
count[codeTable[nbSeq-1]]--;
nbSeq_1--;
}
assert(nbSeq_1 > 1);
assert(entropyWorkspaceSize >= sizeof(ZSTD_BuildCTableWksp));
(void)entropyWorkspaceSize;
FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "FSE_normalizeCount failed");
assert(oend >= op);
{ size_t const NCountSize = FSE_writeNCount(op, (size_t)(oend - op), wksp->norm, max, tableLog); /* overflow protected */
FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "FSE_buildCTable_wksp failed");
return NCountSize;
}
}
default: assert(0); RETURN_ERROR(GENERIC, "impossible to reach");
}
}
FORCE_INLINE_TEMPLATE size_t
ZSTD_encodeSequences_body(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets)
{
BIT_CStream_t blockStream;
FSE_CState_t stateMatchLength;
FSE_CState_t stateOffsetBits;
FSE_CState_t stateLitLength;
RETURN_ERROR_IF(
ERR_isError(BIT_initCStream(&blockStream, dst, dstCapacity)),
dstSize_tooSmall, "not enough space remaining");
DEBUGLOG(6, "available space for bitstream : %i (dstCapacity=%u)",
(int)(blockStream.endPtr - blockStream.startPtr),
(unsigned)dstCapacity);
/* first symbols */
FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
FSE_initCState2(&stateOffsetBits, CTable_OffsetBits, ofCodeTable[nbSeq-1]);
FSE_initCState2(&stateLitLength, CTable_LitLength, llCodeTable[nbSeq-1]);
BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
if (MEM_32bits()) BIT_flushBits(&blockStream);
BIT_addBits(&blockStream, sequences[nbSeq-1].mlBase, ML_bits[mlCodeTable[nbSeq-1]]);
if (MEM_32bits()) BIT_flushBits(&blockStream);
if (longOffsets) {
U32 const ofBits = ofCodeTable[nbSeq-1];
unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
if (extraBits) {
BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, extraBits);
BIT_flushBits(&blockStream);
}
BIT_addBits(&blockStream, sequences[nbSeq-1].offBase >> extraBits,
ofBits - extraBits);
} else {
BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, ofCodeTable[nbSeq-1]);
}
BIT_flushBits(&blockStream);
{ size_t n;
for (n=nbSeq-2 ; n<nbSeq ; n--) { /* intentional underflow */
BYTE const llCode = llCodeTable[n];
BYTE const ofCode = ofCodeTable[n];
BYTE const mlCode = mlCodeTable[n];
U32 const llBits = LL_bits[llCode];
U32 const ofBits = ofCode;
U32 const mlBits = ML_bits[mlCode];
DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
(unsigned)sequences[n].litLength,
(unsigned)sequences[n].mlBase + MINMATCH,
(unsigned)sequences[n].offBase);
/* 32b*/ /* 64b*/
/* (7)*/ /* (7)*/
FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode); /* 15 */ /* 15 */
FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode); /* 24 */ /* 24 */
if (MEM_32bits()) BIT_flushBits(&blockStream); /* (7)*/
FSE_encodeSymbol(&blockStream, &stateLitLength, llCode); /* 16 */ /* 33 */
if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
BIT_flushBits(&blockStream); /* (7)*/
BIT_addBits(&blockStream, sequences[n].litLength, llBits);
if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
BIT_addBits(&blockStream, sequences[n].mlBase, mlBits);
if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
if (longOffsets) {
unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
if (extraBits) {
BIT_addBits(&blockStream, sequences[n].offBase, extraBits);
BIT_flushBits(&blockStream); /* (7)*/
}
BIT_addBits(&blockStream, sequences[n].offBase >> extraBits,
ofBits - extraBits); /* 31 */
} else {
BIT_addBits(&blockStream, sequences[n].offBase, ofBits); /* 31 */
}
BIT_flushBits(&blockStream); /* (7)*/
DEBUGLOG(7, "remaining space : %i", (int)(blockStream.endPtr - blockStream.ptr));
} }
DEBUGLOG(6, "ZSTD_encodeSequences: flushing ML state with %u bits", stateMatchLength.stateLog);
FSE_flushCState(&blockStream, &stateMatchLength);
DEBUGLOG(6, "ZSTD_encodeSequences: flushing Off state with %u bits", stateOffsetBits.stateLog);
FSE_flushCState(&blockStream, &stateOffsetBits);
DEBUGLOG(6, "ZSTD_encodeSequences: flushing LL state with %u bits", stateLitLength.stateLog);
FSE_flushCState(&blockStream, &stateLitLength);
{ size_t const streamSize = BIT_closeCStream(&blockStream);
RETURN_ERROR_IF(streamSize==0, dstSize_tooSmall, "not enough space");
return streamSize;
}
}
static size_t
ZSTD_encodeSequences_default(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets)
{
return ZSTD_encodeSequences_body(dst, dstCapacity,
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq, longOffsets);
}
#if DYNAMIC_BMI2
static BMI2_TARGET_ATTRIBUTE size_t
ZSTD_encodeSequences_bmi2(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets)
{
return ZSTD_encodeSequences_body(dst, dstCapacity,
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq, longOffsets);
}
#endif
size_t ZSTD_encodeSequences(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2)
{
DEBUGLOG(5, "ZSTD_encodeSequences: dstCapacity = %u", (unsigned)dstCapacity);
#if DYNAMIC_BMI2
if (bmi2) {
return ZSTD_encodeSequences_bmi2(dst, dstCapacity,
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq, longOffsets);
}
#endif
(void)bmi2;
return ZSTD_encodeSequences_default(dst, dstCapacity,
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq, longOffsets);
}
/**** ended inlining compress/zstd_compress_sequences.c ****/
/**** start inlining compress/zstd_compress_superblock.c ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/*-*************************************
* Dependencies
***************************************/
/**** start inlining zstd_compress_superblock.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_COMPRESS_ADVANCED_H
#define ZSTD_COMPRESS_ADVANCED_H
/*-*************************************
* Dependencies
***************************************/
/**** skipping file: ../zstd.h ****/
/*-*************************************
* Target Compressed Block Size
***************************************/
/* ZSTD_compressSuperBlock() :
* Used to compress a super block when targetCBlockSize is being used.
* The given block will be compressed into multiple sub blocks that are around targetCBlockSize. */
size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
void* dst, size_t dstCapacity,
void const* src, size_t srcSize,
unsigned lastBlock);
#endif /* ZSTD_COMPRESS_ADVANCED_H */
/**** ended inlining zstd_compress_superblock.h ****/
/**** skipping file: ../common/zstd_internal.h ****/
/**** skipping file: hist.h ****/
/**** skipping file: zstd_compress_internal.h ****/
/**** skipping file: zstd_compress_sequences.h ****/
/**** skipping file: zstd_compress_literals.h ****/
/** ZSTD_compressSubBlock_literal() :
* Compresses literals section for a sub-block.
* When we have to write the Huffman table we will sometimes choose a header
* size larger than necessary. This is because we have to pick the header size
* before we know the table size + compressed size, so we have a bound on the
* table size. If we guessed incorrectly, we fall back to uncompressed literals.
*
* We write the header when writeEntropy=1 and set entropyWritten=1 when we succeeded
* in writing the header, otherwise it is set to 0.
*
* hufMetadata->hType has literals block type info.
* If it is set_basic, all sub-blocks literals section will be Raw_Literals_Block.
* If it is set_rle, all sub-blocks literals section will be RLE_Literals_Block.
* If it is set_compressed, first sub-block's literals section will be Compressed_Literals_Block
* If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
* and the following sub-blocks' literals sections will be Treeless_Literals_Block.
* @return : compressed size of literals section of a sub-block
* Or 0 if unable to compress.
* Or error code */
static size_t
ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
const ZSTD_hufCTablesMetadata_t* hufMetadata,
const BYTE* literals, size_t litSize,
void* dst, size_t dstSize,
const int bmi2, int writeEntropy, int* entropyWritten)
{
size_t const header = writeEntropy ? 200 : 0;
size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstSize;
BYTE* op = ostart + lhSize;
U32 const singleStream = lhSize == 3;
symbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
size_t cLitSize = 0;
DEBUGLOG(5, "ZSTD_compressSubBlock_literal (litSize=%zu, lhSize=%zu, writeEntropy=%d)", litSize, lhSize, writeEntropy);
*entropyWritten = 0;
if (litSize == 0 || hufMetadata->hType == set_basic) {
DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal");
return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
} else if (hufMetadata->hType == set_rle) {
DEBUGLOG(5, "ZSTD_compressSubBlock_literal using rle literal");
return ZSTD_compressRleLiteralsBlock(dst, dstSize, literals, litSize);
}
assert(litSize > 0);
assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);
if (writeEntropy && hufMetadata->hType == set_compressed) {
ZSTD_memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
op += hufMetadata->hufDesSize;
cLitSize += hufMetadata->hufDesSize;
DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
}
{ int const flags = bmi2 ? HUF_flags_bmi2 : 0;
const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, oend-op, literals, litSize, hufTable, flags)
: HUF_compress4X_usingCTable(op, oend-op, literals, litSize, hufTable, flags);
op += cSize;
cLitSize += cSize;
if (cSize == 0 || ERR_isError(cSize)) {
DEBUGLOG(5, "Failed to write entropy tables %s", ZSTD_getErrorName(cSize));
return 0;
}
/* If we expand and we aren't writing a header then emit uncompressed */
if (!writeEntropy && cLitSize >= litSize) {
DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal because uncompressible");
return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
}
/* If we are writing headers then allow expansion that doesn't change our header size. */
if (lhSize < (size_t)(3 + (cLitSize >= 1 KB) + (cLitSize >= 16 KB))) {
assert(cLitSize > litSize);
DEBUGLOG(5, "Literals expanded beyond allowed header size");
return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
}
DEBUGLOG(5, "ZSTD_compressSubBlock_literal (cSize=%zu)", cSize);
}
/* Build header */
switch(lhSize)
{
case 3: /* 2 - 2 - 10 - 10 */
{ U32 const lhc = hType + ((!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
MEM_writeLE24(ostart, lhc);
break;
}
case 4: /* 2 - 2 - 14 - 14 */
{ U32 const lhc = hType + (2 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<18);
MEM_writeLE32(ostart, lhc);
break;
}
case 5: /* 2 - 2 - 18 - 18 */
{ U32 const lhc = hType + (3 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<22);
MEM_writeLE32(ostart, lhc);
ostart[4] = (BYTE)(cLitSize >> 10);
break;
}
default: /* not possible : lhSize is {3,4,5} */
assert(0);
}
*entropyWritten = 1;
DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)litSize, (U32)(op-ostart));
return op-ostart;
}
static size_t
ZSTD_seqDecompressedSize(seqStore_t const* seqStore,
const seqDef* sequences, size_t nbSeq,
size_t litSize, int lastSequence)
{
const seqDef* const sstart = sequences;
const seqDef* const send = sequences + nbSeq;
const seqDef* sp = sstart;
size_t matchLengthSum = 0;
size_t litLengthSum = 0;
(void)(litLengthSum); /* suppress unused variable warning on some environments */
while (send-sp > 0) {
ZSTD_sequenceLength const seqLen = ZSTD_getSequenceLength(seqStore, sp);
litLengthSum += seqLen.litLength;
matchLengthSum += seqLen.matchLength;
sp++;
}
assert(litLengthSum <= litSize);
if (!lastSequence) {
assert(litLengthSum == litSize);
}
return matchLengthSum + litSize;
}
/** ZSTD_compressSubBlock_sequences() :
* Compresses sequences section for a sub-block.
* fseMetadata->llType, fseMetadata->ofType, and fseMetadata->mlType have
* symbol compression modes for the super-block.
* The first successfully compressed block will have these in its header.
* We set entropyWritten=1 when we succeed in compressing the sequences.
* The following sub-blocks will always have repeat mode.
* @return : compressed size of sequences section of a sub-block
* Or 0 if it is unable to compress
* Or error code. */
static size_t
ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
const ZSTD_fseCTablesMetadata_t* fseMetadata,
const seqDef* sequences, size_t nbSeq,
const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
const ZSTD_CCtx_params* cctxParams,
void* dst, size_t dstCapacity,
const int bmi2, int writeEntropy, int* entropyWritten)
{
const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstCapacity;
BYTE* op = ostart;
BYTE* seqHead;
DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (nbSeq=%zu, writeEntropy=%d, longOffsets=%d)", nbSeq, writeEntropy, longOffsets);
*entropyWritten = 0;
/* Sequences Header */
RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
dstSize_tooSmall, "");
if (nbSeq < 128)
*op++ = (BYTE)nbSeq;
else if (nbSeq < LONGNBSEQ)
op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
else
op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
if (nbSeq==0) {
return op - ostart;
}
/* seqHead : flags for FSE encoding type */
seqHead = op++;
DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (seqHeadSize=%u)", (unsigned)(op-ostart));
if (writeEntropy) {
const U32 LLtype = fseMetadata->llType;
const U32 Offtype = fseMetadata->ofType;
const U32 MLtype = fseMetadata->mlType;
DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
*seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
op += fseMetadata->fseTablesSize;
} else {
const U32 repeat = set_repeat;
*seqHead = (BYTE)((repeat<<6) + (repeat<<4) + (repeat<<2));
}
{ size_t const bitstreamSize = ZSTD_encodeSequences(
op, oend - op,
fseTables->matchlengthCTable, mlCode,
fseTables->offcodeCTable, ofCode,
fseTables->litlengthCTable, llCode,
sequences, nbSeq,
longOffsets, bmi2);
FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
op += bitstreamSize;
/* zstd versions <= 1.3.4 mistakenly report corruption when
* FSE_readNCount() receives a buffer < 4 bytes.
* Fixed by https://github.com/facebook/zstd/pull/1146.
* This can happen when the last set_compressed table present is 2
* bytes and the bitstream is only one byte.
* In this exceedingly rare case, we will simply emit an uncompressed
* block, since it isn't worth optimizing.
*/
#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
if (writeEntropy && fseMetadata->lastCountSize && fseMetadata->lastCountSize + bitstreamSize < 4) {
/* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
assert(fseMetadata->lastCountSize + bitstreamSize == 3);
DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
"emitting an uncompressed block.");
return 0;
}
#endif
DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (bitstreamSize=%zu)", bitstreamSize);
}
/* zstd versions <= 1.4.0 mistakenly report error when
* sequences section body size is less than 3 bytes.
* Fixed by https://github.com/facebook/zstd/pull/1664.
* This can happen when the previous sequences section block is compressed
* with rle mode and the current block's sequences section is compressed
* with repeat mode where sequences section body size can be 1 byte.
*/
#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
if (op-seqHead < 4) {
DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.4.0 by emitting "
"an uncompressed block when sequences are < 4 bytes");
return 0;
}
#endif
*entropyWritten = 1;
return op - ostart;
}
/** ZSTD_compressSubBlock() :
* Compresses a single sub-block.
* @return : compressed size of the sub-block
* Or 0 if it failed to compress. */
static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
const seqDef* sequences, size_t nbSeq,
const BYTE* literals, size_t litSize,
const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
const ZSTD_CCtx_params* cctxParams,
void* dst, size_t dstCapacity,
const int bmi2,
int writeLitEntropy, int writeSeqEntropy,
int* litEntropyWritten, int* seqEntropyWritten,
U32 lastBlock)
{
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstCapacity;
BYTE* op = ostart + ZSTD_blockHeaderSize;
DEBUGLOG(5, "ZSTD_compressSubBlock (litSize=%zu, nbSeq=%zu, writeLitEntropy=%d, writeSeqEntropy=%d, lastBlock=%d)",
litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
{ size_t cLitSize = ZSTD_compressSubBlock_literal((const HUF_CElt*)entropy->huf.CTable,
&entropyMetadata->hufMetadata, literals, litSize,
op, oend-op, bmi2, writeLitEntropy, litEntropyWritten);
FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
if (cLitSize == 0) return 0;
op += cLitSize;
}
{ size_t cSeqSize = ZSTD_compressSubBlock_sequences(&entropy->fse,
&entropyMetadata->fseMetadata,
sequences, nbSeq,
llCode, mlCode, ofCode,
cctxParams,
op, oend-op,
bmi2, writeSeqEntropy, seqEntropyWritten);
FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
if (cSeqSize == 0) return 0;
op += cSeqSize;
}
/* Write block header */
{ size_t cSize = (op-ostart)-ZSTD_blockHeaderSize;
U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
MEM_writeLE24(ostart, cBlockHeader24);
}
return op-ostart;
}
static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t litSize,
const ZSTD_hufCTables_t* huf,
const ZSTD_hufCTablesMetadata_t* hufMetadata,
void* workspace, size_t wkspSize,
int writeEntropy)
{
unsigned* const countWksp = (unsigned*)workspace;
unsigned maxSymbolValue = 255;
size_t literalSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
if (hufMetadata->hType == set_basic) return litSize;
else if (hufMetadata->hType == set_rle) return 1;
else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
if (ZSTD_isError(largest)) return litSize;
{ size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
return cLitSizeEstimate + literalSectionHeaderSize;
} }
assert(0); /* impossible */
return 0;
}
static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
const BYTE* codeTable, unsigned maxCode,
size_t nbSeq, const FSE_CTable* fseCTable,
const U8* additionalBits,
short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
void* workspace, size_t wkspSize)
{
unsigned* const countWksp = (unsigned*)workspace;
const BYTE* ctp = codeTable;
const BYTE* const ctStart = ctp;
const BYTE* const ctEnd = ctStart + nbSeq;
size_t cSymbolTypeSizeEstimateInBits = 0;
unsigned max = maxCode;
HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize); /* can't fail */
if (type == set_basic) {
/* We selected this encoding type, so it must be valid. */
assert(max <= defaultMax);
cSymbolTypeSizeEstimateInBits = max <= defaultMax
? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
: ERROR(GENERIC);
} else if (type == set_rle) {
cSymbolTypeSizeEstimateInBits = 0;
} else if (type == set_compressed || type == set_repeat) {
cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
}
if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) return nbSeq * 10;
while (ctp < ctEnd) {
if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
ctp++;
}
return cSymbolTypeSizeEstimateInBits / 8;
}
static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
const BYTE* llCodeTable,
const BYTE* mlCodeTable,
size_t nbSeq,
const ZSTD_fseCTables_t* fseTables,
const ZSTD_fseCTablesMetadata_t* fseMetadata,
void* workspace, size_t wkspSize,
int writeEntropy)
{
size_t const sequencesSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
size_t cSeqSizeEstimate = 0;
if (nbSeq == 0) return sequencesSectionHeaderSize;
cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
nbSeq, fseTables->offcodeCTable, NULL,
OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
nbSeq, fseTables->litlengthCTable, LL_bits,
LL_defaultNorm, LL_defaultNormLog, MaxLL,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
nbSeq, fseTables->matchlengthCTable, ML_bits,
ML_defaultNorm, ML_defaultNormLog, MaxML,
workspace, wkspSize);
if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
return cSeqSizeEstimate + sequencesSectionHeaderSize;
}
static size_t ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
const BYTE* ofCodeTable,
const BYTE* llCodeTable,
const BYTE* mlCodeTable,
size_t nbSeq,
const ZSTD_entropyCTables_t* entropy,
const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
void* workspace, size_t wkspSize,
int writeLitEntropy, int writeSeqEntropy) {
size_t cSizeEstimate = 0;
cSizeEstimate += ZSTD_estimateSubBlockSize_literal(literals, litSize,
&entropy->huf, &entropyMetadata->hufMetadata,
workspace, wkspSize, writeLitEntropy);
cSizeEstimate += ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
workspace, wkspSize, writeSeqEntropy);
return cSizeEstimate + ZSTD_blockHeaderSize;
}
static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
{
if (fseMetadata->llType == set_compressed || fseMetadata->llType == set_rle)
return 1;
if (fseMetadata->mlType == set_compressed || fseMetadata->mlType == set_rle)
return 1;
if (fseMetadata->ofType == set_compressed || fseMetadata->ofType == set_rle)
return 1;
return 0;
}
/** ZSTD_compressSubBlock_multi() :
* Breaks super-block into multiple sub-blocks and compresses them.
* Entropy will be written to the first block.
* The following blocks will use repeat mode to compress.
* All sub-blocks are compressed blocks (no raw or rle blocks).
* @return : compressed size of the super block (which is multiple ZSTD blocks)
* Or 0 if it failed to compress. */
static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
const ZSTD_compressedBlockState_t* prevCBlock,
ZSTD_compressedBlockState_t* nextCBlock,
const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
const ZSTD_CCtx_params* cctxParams,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const int bmi2, U32 lastBlock,
void* workspace, size_t wkspSize)
{
const seqDef* const sstart = seqStorePtr->sequencesStart;
const seqDef* const send = seqStorePtr->sequences;
const seqDef* sp = sstart;
const BYTE* const lstart = seqStorePtr->litStart;
const BYTE* const lend = seqStorePtr->lit;
const BYTE* lp = lstart;
BYTE const* ip = (BYTE const*)src;
BYTE const* const iend = ip + srcSize;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstCapacity;
BYTE* op = ostart;
const BYTE* llCodePtr = seqStorePtr->llCode;
const BYTE* mlCodePtr = seqStorePtr->mlCode;
const BYTE* ofCodePtr = seqStorePtr->ofCode;
size_t targetCBlockSize = cctxParams->targetCBlockSize;
size_t litSize, seqCount;
int writeLitEntropy = entropyMetadata->hufMetadata.hType == set_compressed;
int writeSeqEntropy = 1;
int lastSequence = 0;
DEBUGLOG(5, "ZSTD_compressSubBlock_multi (litSize=%u, nbSeq=%u)",
(unsigned)(lend-lp), (unsigned)(send-sstart));
litSize = 0;
seqCount = 0;
do {
size_t cBlockSizeEstimate = 0;
if (sstart == send) {
lastSequence = 1;
} else {
const seqDef* const sequence = sp + seqCount;
lastSequence = sequence == send - 1;
litSize += ZSTD_getSequenceLength(seqStorePtr, sequence).litLength;
seqCount++;
}
if (lastSequence) {
assert(lp <= lend);
assert(litSize <= (size_t)(lend - lp));
litSize = (size_t)(lend - lp);
}
/* I think there is an optimization opportunity here.
* Calling ZSTD_estimateSubBlockSize for every sequence can be wasteful
* since it recalculates estimate from scratch.
* For example, it would recount literal distribution and symbol codes every time.
*/
cBlockSizeEstimate = ZSTD_estimateSubBlockSize(lp, litSize, ofCodePtr, llCodePtr, mlCodePtr, seqCount,
&nextCBlock->entropy, entropyMetadata,
workspace, wkspSize, writeLitEntropy, writeSeqEntropy);
if (cBlockSizeEstimate > targetCBlockSize || lastSequence) {
int litEntropyWritten = 0;
int seqEntropyWritten = 0;
const size_t decompressedSize = ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, lastSequence);
const size_t cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
sp, seqCount,
lp, litSize,
llCodePtr, mlCodePtr, ofCodePtr,
cctxParams,
op, oend-op,
bmi2, writeLitEntropy, writeSeqEntropy,
&litEntropyWritten, &seqEntropyWritten,
lastBlock && lastSequence);
FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
if (cSize > 0 && cSize < decompressedSize) {
DEBUGLOG(5, "Committed the sub-block");
assert(ip + decompressedSize <= iend);
ip += decompressedSize;
sp += seqCount;
lp += litSize;
op += cSize;
llCodePtr += seqCount;
mlCodePtr += seqCount;
ofCodePtr += seqCount;
litSize = 0;
seqCount = 0;
/* Entropy only needs to be written once */
if (litEntropyWritten) {
writeLitEntropy = 0;
}
if (seqEntropyWritten) {
writeSeqEntropy = 0;
}
}
}
} while (!lastSequence);
if (writeLitEntropy) {
DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
}
if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
/* If we haven't written our entropy tables, then we've violated our contract and
* must emit an uncompressed block.
*/
DEBUGLOG(5, "ZSTD_compressSubBlock_multi has sequence entropy tables unwritten");
return 0;
}
if (ip < iend) {
size_t const cSize = ZSTD_noCompressBlock(op, oend - op, ip, iend - ip, lastBlock);
DEBUGLOG(5, "ZSTD_compressSubBlock_multi last sub-block uncompressed, %zu bytes", (size_t)(iend - ip));
FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
assert(cSize != 0);
op += cSize;
/* We have to regenerate the repcodes because we've skipped some sequences */
if (sp < send) {
seqDef const* seq;
repcodes_t rep;
ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
for (seq = sstart; seq < sp; ++seq) {
ZSTD_updateRep(rep.rep, seq->offBase, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
}
ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
}
}
DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
return op-ostart;
}
size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
void* dst, size_t dstCapacity,
void const* src, size_t srcSize,
unsigned lastBlock) {
ZSTD_entropyCTablesMetadata_t entropyMetadata;
FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
&zc->blockState.prevCBlock->entropy,
&zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
&entropyMetadata,
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
return ZSTD_compressSubBlock_multi(&zc->seqStore,
zc->blockState.prevCBlock,
zc->blockState.nextCBlock,
&entropyMetadata,
&zc->appliedParams,
dst, dstCapacity,
src, srcSize,
zc->bmi2, lastBlock,
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
}
/**** ended inlining compress/zstd_compress_superblock.c ****/
/**** start inlining compress/zstd_compress.c ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/*-*************************************
* Dependencies
***************************************/
/**** skipping file: ../common/allocations.h ****/
/**** skipping file: ../common/zstd_deps.h ****/
/**** skipping file: ../common/mem.h ****/
/**** skipping file: hist.h ****/
#define FSE_STATIC_LINKING_ONLY /* FSE_encodeSymbol */
/**** skipping file: ../common/fse.h ****/
/**** skipping file: ../common/huf.h ****/
/**** skipping file: zstd_compress_internal.h ****/
/**** skipping file: zstd_compress_sequences.h ****/
/**** skipping file: zstd_compress_literals.h ****/
/**** start inlining zstd_fast.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_FAST_H
#define ZSTD_FAST_H
#if defined (__cplusplus)
extern "C" {
#endif
/**** skipping file: ../common/mem.h ****/
/**** skipping file: zstd_compress_internal.h ****/
void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
void const* end, ZSTD_dictTableLoadMethod_e dtlm,
ZSTD_tableFillPurpose_e tfp);
size_t ZSTD_compressBlock_fast(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_fast_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_fast_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
#if defined (__cplusplus)
}
#endif
#endif /* ZSTD_FAST_H */
/**** ended inlining zstd_fast.h ****/
/**** start inlining zstd_double_fast.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_DOUBLE_FAST_H
#define ZSTD_DOUBLE_FAST_H
#if defined (__cplusplus)
extern "C" {
#endif
/**** skipping file: ../common/mem.h ****/
/**** skipping file: zstd_compress_internal.h ****/
#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
void const* end, ZSTD_dictTableLoadMethod_e dtlm,
ZSTD_tableFillPurpose_e tfp);
size_t ZSTD_compressBlock_doubleFast(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_doubleFast_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_doubleFast_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
#define ZSTD_COMPRESSBLOCK_DOUBLEFAST ZSTD_compressBlock_doubleFast
#define ZSTD_COMPRESSBLOCK_DOUBLEFAST_DICTMATCHSTATE ZSTD_compressBlock_doubleFast_dictMatchState
#define ZSTD_COMPRESSBLOCK_DOUBLEFAST_EXTDICT ZSTD_compressBlock_doubleFast_extDict
#else
#define ZSTD_COMPRESSBLOCK_DOUBLEFAST NULL
#define ZSTD_COMPRESSBLOCK_DOUBLEFAST_DICTMATCHSTATE NULL
#define ZSTD_COMPRESSBLOCK_DOUBLEFAST_EXTDICT NULL
#endif /* ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR */
#if defined (__cplusplus)
}
#endif
#endif /* ZSTD_DOUBLE_FAST_H */
/**** ended inlining zstd_double_fast.h ****/
/**** start inlining zstd_lazy.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_LAZY_H
#define ZSTD_LAZY_H
#if defined (__cplusplus)
extern "C" {
#endif
/**** skipping file: zstd_compress_internal.h ****/
/**
* Dedicated Dictionary Search Structure bucket log. In the
* ZSTD_dedicatedDictSearch mode, the hashTable has
* 2 ** ZSTD_LAZY_DDSS_BUCKET_LOG entries in each bucket, rather than just
* one.
*/
#define ZSTD_LAZY_DDSS_BUCKET_LOG 2
#define ZSTD_ROW_HASH_TAG_BITS 8 /* nb bits to use for the tag */
#if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
|| !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
|| !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \
|| !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR)
U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip);
void ZSTD_row_update(ZSTD_matchState_t* const ms, const BYTE* ip);
void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_matchState_t* ms, const BYTE* const ip);
void ZSTD_preserveUnsortedMark (U32* const table, U32 const size, U32 const reducerValue); /*! used in ZSTD_reduceIndex(). preemptively increase value of ZSTD_DUBT_UNSORTED_MARK */
#endif
#ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
size_t ZSTD_compressBlock_greedy(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_greedy_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_greedy_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_greedy_dictMatchState_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_greedy_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_greedy_extDict_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
#define ZSTD_COMPRESSBLOCK_GREEDY ZSTD_compressBlock_greedy
#define ZSTD_COMPRESSBLOCK_GREEDY_ROW ZSTD_compressBlock_greedy_row
#define ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE ZSTD_compressBlock_greedy_dictMatchState
#define ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE_ROW ZSTD_compressBlock_greedy_dictMatchState_row
#define ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH ZSTD_compressBlock_greedy_dedicatedDictSearch
#define ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH_ROW ZSTD_compressBlock_greedy_dedicatedDictSearch_row
#define ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT ZSTD_compressBlock_greedy_extDict
#define ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT_ROW ZSTD_compressBlock_greedy_extDict_row
#else
#define ZSTD_COMPRESSBLOCK_GREEDY NULL
#define ZSTD_COMPRESSBLOCK_GREEDY_ROW NULL
#define ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE NULL
#define ZSTD_COMPRESSBLOCK_GREEDY_DICTMATCHSTATE_ROW NULL
#define ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH NULL
#define ZSTD_COMPRESSBLOCK_GREEDY_DEDICATEDDICTSEARCH_ROW NULL
#define ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT NULL
#define ZSTD_COMPRESSBLOCK_GREEDY_EXTDICT_ROW NULL
#endif
#ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
size_t ZSTD_compressBlock_lazy(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy_dictMatchState_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy_extDict_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
#define ZSTD_COMPRESSBLOCK_LAZY ZSTD_compressBlock_lazy
#define ZSTD_COMPRESSBLOCK_LAZY_ROW ZSTD_compressBlock_lazy_row
#define ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE ZSTD_compressBlock_lazy_dictMatchState
#define ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE_ROW ZSTD_compressBlock_lazy_dictMatchState_row
#define ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH ZSTD_compressBlock_lazy_dedicatedDictSearch
#define ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH_ROW ZSTD_compressBlock_lazy_dedicatedDictSearch_row
#define ZSTD_COMPRESSBLOCK_LAZY_EXTDICT ZSTD_compressBlock_lazy_extDict
#define ZSTD_COMPRESSBLOCK_LAZY_EXTDICT_ROW ZSTD_compressBlock_lazy_extDict_row
#else
#define ZSTD_COMPRESSBLOCK_LAZY NULL
#define ZSTD_COMPRESSBLOCK_LAZY_ROW NULL
#define ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE NULL
#define ZSTD_COMPRESSBLOCK_LAZY_DICTMATCHSTATE_ROW NULL
#define ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH NULL
#define ZSTD_COMPRESSBLOCK_LAZY_DEDICATEDDICTSEARCH_ROW NULL
#define ZSTD_COMPRESSBLOCK_LAZY_EXTDICT NULL
#define ZSTD_COMPRESSBLOCK_LAZY_EXTDICT_ROW NULL
#endif
#ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
size_t ZSTD_compressBlock_lazy2(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy2_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy2_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy2_dictMatchState_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy2_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy2_extDict_row(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
#define ZSTD_COMPRESSBLOCK_LAZY2 ZSTD_compressBlock_lazy2
#define ZSTD_COMPRESSBLOCK_LAZY2_ROW ZSTD_compressBlock_lazy2_row
#define ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE ZSTD_compressBlock_lazy2_dictMatchState
#define ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE_ROW ZSTD_compressBlock_lazy2_dictMatchState_row
#define ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH ZSTD_compressBlock_lazy2_dedicatedDictSearch
#define ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH_ROW ZSTD_compressBlock_lazy2_dedicatedDictSearch_row
#define ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT ZSTD_compressBlock_lazy2_extDict
#define ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT_ROW ZSTD_compressBlock_lazy2_extDict_row
#else
#define ZSTD_COMPRESSBLOCK_LAZY2 NULL
#define ZSTD_COMPRESSBLOCK_LAZY2_ROW NULL
#define ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE NULL
#define ZSTD_COMPRESSBLOCK_LAZY2_DICTMATCHSTATE_ROW NULL
#define ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH NULL
#define ZSTD_COMPRESSBLOCK_LAZY2_DEDICATEDDICTSEARCH_ROW NULL
#define ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT NULL
#define ZSTD_COMPRESSBLOCK_LAZY2_EXTDICT_ROW NULL
#endif
#ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
size_t ZSTD_compressBlock_btlazy2(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_btlazy2_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_btlazy2_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
#define ZSTD_COMPRESSBLOCK_BTLAZY2 ZSTD_compressBlock_btlazy2
#define ZSTD_COMPRESSBLOCK_BTLAZY2_DICTMATCHSTATE ZSTD_compressBlock_btlazy2_dictMatchState
#define ZSTD_COMPRESSBLOCK_BTLAZY2_EXTDICT ZSTD_compressBlock_btlazy2_extDict
#else
#define ZSTD_COMPRESSBLOCK_BTLAZY2 NULL
#define ZSTD_COMPRESSBLOCK_BTLAZY2_DICTMATCHSTATE NULL
#define ZSTD_COMPRESSBLOCK_BTLAZY2_EXTDICT NULL
#endif
#if defined (__cplusplus)
}
#endif
#endif /* ZSTD_LAZY_H */
/**** ended inlining zstd_lazy.h ****/
/**** start inlining zstd_opt.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_OPT_H
#define ZSTD_OPT_H
#if defined (__cplusplus)
extern "C" {
#endif
/**** skipping file: zstd_compress_internal.h ****/
#if !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR) \
|| !defined(ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR) \
|| !defined(ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR)
/* used in ZSTD_loadDictionaryContent() */
void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend);
#endif
#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
size_t ZSTD_compressBlock_btopt(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_btopt_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_btopt_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
#define ZSTD_COMPRESSBLOCK_BTOPT ZSTD_compressBlock_btopt
#define ZSTD_COMPRESSBLOCK_BTOPT_DICTMATCHSTATE ZSTD_compressBlock_btopt_dictMatchState
#define ZSTD_COMPRESSBLOCK_BTOPT_EXTDICT ZSTD_compressBlock_btopt_extDict
#else
#define ZSTD_COMPRESSBLOCK_BTOPT NULL
#define ZSTD_COMPRESSBLOCK_BTOPT_DICTMATCHSTATE NULL
#define ZSTD_COMPRESSBLOCK_BTOPT_EXTDICT NULL
#endif
#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
size_t ZSTD_compressBlock_btultra(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_btultra_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_btultra_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
/* note : no btultra2 variant for extDict nor dictMatchState,
* because btultra2 is not meant to work with dictionaries
* and is only specific for the first block (no prefix) */
size_t ZSTD_compressBlock_btultra2(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
#define ZSTD_COMPRESSBLOCK_BTULTRA ZSTD_compressBlock_btultra
#define ZSTD_COMPRESSBLOCK_BTULTRA_DICTMATCHSTATE ZSTD_compressBlock_btultra_dictMatchState
#define ZSTD_COMPRESSBLOCK_BTULTRA_EXTDICT ZSTD_compressBlock_btultra_extDict
#define ZSTD_COMPRESSBLOCK_BTULTRA2 ZSTD_compressBlock_btultra2
#else
#define ZSTD_COMPRESSBLOCK_BTULTRA NULL
#define ZSTD_COMPRESSBLOCK_BTULTRA_DICTMATCHSTATE NULL
#define ZSTD_COMPRESSBLOCK_BTULTRA_EXTDICT NULL
#define ZSTD_COMPRESSBLOCK_BTULTRA2 NULL
#endif
#if defined (__cplusplus)
}
#endif
#endif /* ZSTD_OPT_H */
/**** ended inlining zstd_opt.h ****/
/**** start inlining zstd_ldm.h ****/
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_LDM_H
#define ZSTD_LDM_H
#if defined (__cplusplus)
extern "C" {
#endif
/**** skipping file: zstd_compress_internal.h ****/
/**** skipping file: ../zstd.h ****/
/*-*************************************
* Long distance matching
***************************************/
#define ZSTD_LDM_DEFAULT_WINDOW_LOG ZSTD_WINDOWLOG_LIMIT_DEFAULT
void ZSTD_ldm_fillHashTable(
ldmState_t* state, const BYTE* ip,
const BYTE* iend, ldmParams_t const* params);
/**
* ZSTD_ldm_generateSequences():
*
* Generates the sequences using the long distance match finder.
* Generates long range matching sequences in `sequences`, which parse a prefix
* of the source. `sequences` must be large enough to store every sequence,
* which can be checked with `ZSTD_ldm_getMaxNbSeq()`.
* @returns 0 or an error code.
*
* NOTE: The user must have called ZSTD_window_update() for all of the input
* they have, even if they pass it to ZSTD_ldm_generateSequences() in chunks.
* NOTE: This function returns an error if it runs out of space to store
* sequences.
*/
size_t ZSTD_ldm_generateSequences(
ldmState_t* ldms, rawSeqStore_t* sequences,
ldmParams_t const* params, void const* src, size_t srcSize);
/**
* ZSTD_ldm_blockCompress():
*
* Compresses a block using the predefined sequences, along with a secondary
* block compressor. The literals section of every sequence is passed to the
* secondary block compressor, and those sequences are interspersed with the
* predefined sequences. Returns the length of the last literals.
* Updates `rawSeqStore.pos` to indicate how many sequences have been consumed.
* `rawSeqStore.seq` may also be updated to split the last sequence between two
* blocks.
* @return The length of the last literals.
*
* NOTE: The source must be at most the maximum block size, but the predefined
* sequences can be any size, and may be longer than the block. In the case that
* they are longer than the block, the last sequences may need to be split into
* two. We handle that case correctly, and update `rawSeqStore` appropriately.
* NOTE: This function does not return any errors.
*/
size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
ZSTD_paramSwitch_e useRowMatchFinder,
void const* src, size_t srcSize);
/**
* ZSTD_ldm_skipSequences():
*
* Skip past `srcSize` bytes worth of sequences in `rawSeqStore`.
* Avoids emitting matches less than `minMatch` bytes.
* Must be called for data that is not passed to ZSTD_ldm_blockCompress().
*/
void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize,
U32 const minMatch);
/* ZSTD_ldm_skipRawSeqStoreBytes():
* Moves forward in rawSeqStore by nbBytes, updating fields 'pos' and 'posInSequence'.
* Not to be used in conjunction with ZSTD_ldm_skipSequences().
* Must be called for data with is not passed to ZSTD_ldm_blockCompress().
*/
void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes);
/** ZSTD_ldm_getTableSize() :
* Estimate the space needed for long distance matching tables or 0 if LDM is
* disabled.
*/
size_t ZSTD_ldm_getTableSize(ldmParams_t params);
/** ZSTD_ldm_getSeqSpace() :
* Return an upper bound on the number of sequences that can be produced by
* the long distance matcher, or 0 if LDM is disabled.
*/
size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize);
/** ZSTD_ldm_adjustParameters() :
* If the params->hashRateLog is not set, set it to its default value based on
* windowLog and params->hashLog.
*
* Ensures that params->bucketSizeLog is <= params->hashLog (setting it to
* params->hashLog if it is not).
*
* Ensures that the minMatchLength >= targetLength during optimal parsing.
*/
void ZSTD_ldm_adjustParameters(ldmParams_t* params,
ZSTD_compressionParameters const* cParams);
#if defined (__cplusplus)
}
#endif
#endif /* ZSTD_FAST_H */
/**** ended inlining zstd_ldm.h ****/
/**** skipping file: zstd_compress_superblock.h ****/
/**** skipping file: ../common/bits.h ****/
/* ***************************************************************
* Tuning parameters
*****************************************************************/
/*!
* COMPRESS_HEAPMODE :
* Select how default decompression function ZSTD_compress() allocates its context,
* on stack (0, default), or into heap (1).
* Note that functions with explicit context such as ZSTD_compressCCtx() are unaffected.
*/
#ifndef ZSTD_COMPRESS_HEAPMODE
# define ZSTD_COMPRESS_HEAPMODE 0
#endif
/*!
* ZSTD_HASHLOG3_MAX :
* Maximum size of the hash table dedicated to find 3-bytes matches,
* in log format, aka 17 => 1 << 17 == 128Ki positions.
* This structure is only used in zstd_opt.
* Since allocation is centralized for all strategies, it has to be known here.
* The actual (selected) size of the hash table is then stored in ZSTD_matchState_t.hashLog3,
* so that zstd_opt.c doesn't need to know about this constant.
*/
#ifndef ZSTD_HASHLOG3_MAX
# define ZSTD_HASHLOG3_MAX 17
#endif
/*-*************************************
* Helper functions
***************************************/
/* ZSTD_compressBound()
* Note that the result from this function is only valid for
* the one-pass compression functions.
* When employing the streaming mode,
* if flushes are frequently altering the size of blocks,
* the overhead from block headers can make the compressed data larger
* than the return value of ZSTD_compressBound().
*/
size_t ZSTD_compressBound(size_t srcSize) {
size_t const r = ZSTD_COMPRESSBOUND(srcSize);
if (r==0) return ERROR(srcSize_wrong);
return r;
}
/*-*************************************
* Context memory management
***************************************/
struct ZSTD_CDict_s {
const void* dictContent;
size_t dictContentSize;
ZSTD_dictContentType_e dictContentType; /* The dictContentType the CDict was created with */
U32* entropyWorkspace; /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
ZSTD_cwksp workspace;
ZSTD_matchState_t matchState;
ZSTD_compressedBlockState_t cBlockState;
ZSTD_customMem customMem;
U32 dictID;
int compressionLevel; /* 0 indicates that advanced API was used to select CDict params */
ZSTD_paramSwitch_e useRowMatchFinder; /* Indicates whether the CDict was created with params that would use
* row-based matchfinder. Unless the cdict is reloaded, we will use
* the same greedy/lazy matchfinder at compression time.
*/
}; /* typedef'd to ZSTD_CDict within "zstd.h" */
ZSTD_CCtx* ZSTD_createCCtx(void)
{
return ZSTD_createCCtx_advanced(ZSTD_defaultCMem);
}
static void ZSTD_initCCtx(ZSTD_CCtx* cctx, ZSTD_customMem memManager)
{
assert(cctx != NULL);
ZSTD_memset(cctx, 0, sizeof(*cctx));
cctx->customMem = memManager;
cctx->bmi2 = ZSTD_cpuSupportsBmi2();
{ size_t const err = ZSTD_CCtx_reset(cctx, ZSTD_reset_parameters);
assert(!ZSTD_isError(err));
(void)err;
}
}
ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
{
ZSTD_STATIC_ASSERT(zcss_init==0);
ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN==(0ULL - 1));
if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
{ ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_customMalloc(sizeof(ZSTD_CCtx), customMem);
if (!cctx) return NULL;
ZSTD_initCCtx(cctx, customMem);
return cctx;
}
}
ZSTD_CCtx* ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize)
{
ZSTD_cwksp ws;
ZSTD_CCtx* cctx;
if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL; /* minimum size */
if ((size_t)workspace & 7) return NULL; /* must be 8-aligned */
ZSTD_cwksp_init(&ws, workspace, workspaceSize, ZSTD_cwksp_static_alloc);
cctx = (ZSTD_CCtx*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CCtx));
if (cctx == NULL) return NULL;
ZSTD_memset(cctx, 0, sizeof(ZSTD_CCtx));
ZSTD_cwksp_move(&cctx->workspace, &ws);
cctx->staticSize = workspaceSize;
/* statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
if (!ZSTD_cwksp_check_available(&cctx->workspace, ENTROPY_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t))) return NULL;
cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
cctx->blockState.nextCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
cctx->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cctx->workspace, ENTROPY_WORKSPACE_SIZE);
cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
return cctx;
}
/**
* Clears and frees all of the dictionaries in the CCtx.
*/
static void ZSTD_clearAllDicts(ZSTD_CCtx* cctx)
{
ZSTD_customFree(cctx->localDict.dictBuffer, cctx->customMem);
ZSTD_freeCDict(cctx->localDict.cdict);
ZSTD_memset(&cctx->localDict, 0, sizeof(cctx->localDict));
ZSTD_memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));
cctx->cdict = NULL;
}
static size_t ZSTD_sizeof_localDict(ZSTD_localDict dict)
{
size_t const bufferSize = dict.dictBuffer != NULL ? dict.dictSize : 0;
size_t const cdictSize = ZSTD_sizeof_CDict(dict.cdict);
return bufferSize + cdictSize;
}
static void ZSTD_freeCCtxContent(ZSTD_CCtx* cctx)
{
assert(cctx != NULL);
assert(cctx->staticSize == 0);
ZSTD_clearAllDicts(cctx);
#ifdef ZSTD_MULTITHREAD
ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL;
#endif
ZSTD_cwksp_free(&cctx->workspace, cctx->customMem);
}
size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
{
if (cctx==NULL) return 0; /* support free on NULL */
RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
"not compatible with static CCtx");
{ int cctxInWorkspace = ZSTD_cwksp_owns_buffer(&cctx->workspace, cctx);
ZSTD_freeCCtxContent(cctx);
if (!cctxInWorkspace) ZSTD_customFree(cctx, cctx->customMem);
}
return 0;
}
static size_t ZSTD_sizeof_mtctx(const ZSTD_CCtx* cctx)
{
#ifdef ZSTD_MULTITHREAD
return ZSTDMT_sizeof_CCtx(cctx->mtctx);
#else
(void)cctx;
return 0;
#endif
}
size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx)
{
if (cctx==NULL) return 0; /* support sizeof on NULL */
/* cctx may be in the workspace */
return (cctx->workspace.workspace == cctx ? 0 : sizeof(*cctx))
+ ZSTD_cwksp_sizeof(&cctx->workspace)
+ ZSTD_sizeof_localDict(cctx->localDict)
+ ZSTD_sizeof_mtctx(cctx);
}
size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
{
return ZSTD_sizeof_CCtx(zcs); /* same object */
}
/* private API call, for dictBuilder only */
const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx) { return &(ctx->seqStore); }
/* Returns true if the strategy supports using a row based matchfinder */
static int ZSTD_rowMatchFinderSupported(const ZSTD_strategy strategy) {
return (strategy >= ZSTD_greedy && strategy <= ZSTD_lazy2);
}
/* Returns true if the strategy and useRowMatchFinder mode indicate that we will use the row based matchfinder
* for this compression.
*/
static int ZSTD_rowMatchFinderUsed(const ZSTD_strategy strategy, const ZSTD_paramSwitch_e mode) {
assert(mode != ZSTD_ps_auto);
return ZSTD_rowMatchFinderSupported(strategy) && (mode == ZSTD_ps_enable);
}
/* Returns row matchfinder usage given an initial mode and cParams */
static ZSTD_paramSwitch_e ZSTD_resolveRowMatchFinderMode(ZSTD_paramSwitch_e mode,
const ZSTD_compressionParameters* const cParams) {
#if defined(ZSTD_ARCH_X86_SSE2) || defined(ZSTD_ARCH_ARM_NEON)
int const kHasSIMD128 = 1;
#else
int const kHasSIMD128 = 0;
#endif
if (mode != ZSTD_ps_auto) return mode; /* if requested enabled, but no SIMD, we still will use row matchfinder */
mode = ZSTD_ps_disable;
if (!ZSTD_rowMatchFinderSupported(cParams->strategy)) return mode;
if (kHasSIMD128) {
if (cParams->windowLog > 14) mode = ZSTD_ps_enable;
} else {
if (cParams->windowLog > 17) mode = ZSTD_ps_enable;
}
return mode;
}
/* Returns block splitter usage (generally speaking, when using slower/stronger compression modes) */
static ZSTD_paramSwitch_e ZSTD_resolveBlockSplitterMode(ZSTD_paramSwitch_e mode,
const ZSTD_compressionParameters* const cParams) {
if (mode != ZSTD_ps_auto) return mode;
return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 17) ? ZSTD_ps_enable : ZSTD_ps_disable;
}
/* Returns 1 if the arguments indicate that we should allocate a chainTable, 0 otherwise */
static int ZSTD_allocateChainTable(const ZSTD_strategy strategy,
const ZSTD_paramSwitch_e useRowMatchFinder,
const U32 forDDSDict) {
assert(useRowMatchFinder != ZSTD_ps_auto);
/* We always should allocate a chaintable if we are allocating a matchstate for a DDS dictionary matchstate.
* We do not allocate a chaintable if we are using ZSTD_fast, or are using the row-based matchfinder.
*/
return forDDSDict || ((strategy != ZSTD_fast) && !ZSTD_rowMatchFinderUsed(strategy, useRowMatchFinder));
}
/* Returns ZSTD_ps_enable if compression parameters are such that we should
* enable long distance matching (wlog >= 27, strategy >= btopt).
* Returns ZSTD_ps_disable otherwise.
*/
static ZSTD_paramSwitch_e ZSTD_resolveEnableLdm(ZSTD_paramSwitch_e mode,
const ZSTD_compressionParameters* const cParams) {
if (mode != ZSTD_ps_auto) return mode;
return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27) ? ZSTD_ps_enable : ZSTD_ps_disable;
}
static int ZSTD_resolveExternalSequenceValidation(int mode) {
return mode;
}
/* Resolves maxBlockSize to the default if no value is present. */
static size_t ZSTD_resolveMaxBlockSize(size_t maxBlockSize) {
if (maxBlockSize == 0) {
return ZSTD_BLOCKSIZE_MAX;
} else {
return maxBlockSize;
}
}
static ZSTD_paramSwitch_e ZSTD_resolveExternalRepcodeSearch(ZSTD_paramSwitch_e value, int cLevel) {
if (value != ZSTD_ps_auto) return value;
if (cLevel < 10) {
return ZSTD_ps_disable;
} else {
return ZSTD_ps_enable;
}
}
/* Returns 1 if compression parameters are such that CDict hashtable and chaintable indices are tagged.
* If so, the tags need to be removed in ZSTD_resetCCtx_byCopyingCDict. */
static int ZSTD_CDictIndicesAreTagged(const ZSTD_compressionParameters* const cParams) {
return cParams->strategy == ZSTD_fast || cParams->strategy == ZSTD_dfast;
}
static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
ZSTD_compressionParameters cParams)
{
ZSTD_CCtx_params cctxParams;
/* should not matter, as all cParams are presumed properly defined */
ZSTD_CCtxParams_init(&cctxParams, ZSTD_CLEVEL_DEFAULT);
cctxParams.cParams = cParams;
/* Adjust advanced params according to cParams */
cctxParams.ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams.ldmParams.enableLdm, &cParams);
if (cctxParams.ldmParams.enableLdm == ZSTD_ps_enable) {
ZSTD_ldm_adjustParameters(&cctxParams.ldmParams, &cParams);
assert(cctxParams.ldmParams.hashLog >= cctxParams.ldmParams.bucketSizeLog);
assert(cctxParams.ldmParams.hashRateLog < 32);
}
cctxParams.useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams.useBlockSplitter, &cParams);
cctxParams.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams.useRowMatchFinder, &cParams);
cctxParams.validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams.validateSequences);
cctxParams.maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams.maxBlockSize);
cctxParams.searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams.searchForExternalRepcodes,
cctxParams.compressionLevel);
assert(!ZSTD_checkCParams(cParams));
return cctxParams;
}
static ZSTD_CCtx_params* ZSTD_createCCtxParams_advanced(
ZSTD_customMem customMem)
{
ZSTD_CCtx_params* params;
if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
params = (ZSTD_CCtx_params*)ZSTD_customCalloc(
sizeof(ZSTD_CCtx_params), customMem);
if (!params) { return NULL; }
ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
params->customMem = customMem;
return params;
}
ZSTD_CCtx_params* ZSTD_createCCtxParams(void)
{
return ZSTD_createCCtxParams_advanced(ZSTD_defaultCMem);
}
size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params)
{
if (params == NULL) { return 0; }
ZSTD_customFree(params, params->customMem);
return 0;
}
size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params)
{
return ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
}
size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
cctxParams->compressionLevel = compressionLevel;
cctxParams->fParams.contentSizeFlag = 1;
return 0;
}
#define ZSTD_NO_CLEVEL 0
/**
* Initializes `cctxParams` from `params` and `compressionLevel`.
* @param compressionLevel If params are derived from a compression level then that compression level, otherwise ZSTD_NO_CLEVEL.
*/
static void
ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams,
const ZSTD_parameters* params,
int compressionLevel)
{
assert(!ZSTD_checkCParams(params->cParams));
ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
cctxParams->cParams = params->cParams;
cctxParams->fParams = params->fParams;
/* Should not matter, as all cParams are presumed properly defined.
* But, set it for tracing anyway.
*/
cctxParams->compressionLevel = compressionLevel;
cctxParams->useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams->useRowMatchFinder, &params->cParams);
cctxParams->useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams->useBlockSplitter, &params->cParams);
cctxParams->ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams->ldmParams.enableLdm, &params->cParams);
cctxParams->validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams->validateSequences);
cctxParams->maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams->maxBlockSize);
cctxParams->searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams->searchForExternalRepcodes, compressionLevel);
DEBUGLOG(4, "ZSTD_CCtxParams_init_internal: useRowMatchFinder=%d, useBlockSplitter=%d ldm=%d",
cctxParams->useRowMatchFinder, cctxParams->useBlockSplitter, cctxParams->ldmParams.enableLdm);
}
size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
{
RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
FORWARD_IF_ERROR( ZSTD_checkCParams(params.cParams) , "");
ZSTD_CCtxParams_init_internal(cctxParams, &params, ZSTD_NO_CLEVEL);
return 0;
}
/**
* Sets cctxParams' cParams and fParams from params, but otherwise leaves them alone.
* @param params Validated zstd parameters.
*/
static void ZSTD_CCtxParams_setZstdParams(
ZSTD_CCtx_params* cctxParams, const ZSTD_parameters* params)
{
assert(!ZSTD_checkCParams(params->cParams));
cctxParams->cParams = params->cParams;
cctxParams->fParams = params->fParams;
/* Should not matter, as all cParams are presumed properly defined.
* But, set it for tracing anyway.
*/
cctxParams->compressionLevel = ZSTD_NO_CLEVEL;
}
ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
{
ZSTD_bounds bounds = { 0, 0, 0 };
switch(param)
{
case ZSTD_c_compressionLevel:
bounds.lowerBound = ZSTD_minCLevel();
bounds.upperBound = ZSTD_maxCLevel();
return bounds;
case ZSTD_c_windowLog:
bounds.lowerBound = ZSTD_WINDOWLOG_MIN;
bounds.upperBound = ZSTD_WINDOWLOG_MAX;
return bounds;
case ZSTD_c_hashLog:
bounds.lowerBound = ZSTD_HASHLOG_MIN;
bounds.upperBound = ZSTD_HASHLOG_MAX;
return bounds;
case ZSTD_c_chainLog:
bounds.lowerBound = ZSTD_CHAINLOG_MIN;
bounds.upperBound = ZSTD_CHAINLOG_MAX;
return bounds;
case ZSTD_c_searchLog:
bounds.lowerBound = ZSTD_SEARCHLOG_MIN;
bounds.upperBound = ZSTD_SEARCHLOG_MAX;
return bounds;
case ZSTD_c_minMatch:
bounds.lowerBound = ZSTD_MINMATCH_MIN;
bounds.upperBound = ZSTD_MINMATCH_MAX;
return bounds;
case ZSTD_c_targetLength:
bounds.lowerBound = ZSTD_TARGETLENGTH_MIN;
bounds.upperBound = ZSTD_TARGETLENGTH_MAX;
return bounds;
case ZSTD_c_strategy:
bounds.lowerBound = ZSTD_STRATEGY_MIN;
bounds.upperBound = ZSTD_STRATEGY_MAX;
return bounds;
case ZSTD_c_contentSizeFlag:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_checksumFlag:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_dictIDFlag:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_nbWorkers:
bounds.lowerBound = 0;
#ifdef ZSTD_MULTITHREAD
bounds.upperBound = ZSTDMT_NBWORKERS_MAX;
#else
bounds.upperBound = 0;
#endif
return bounds;
case ZSTD_c_jobSize:
bounds.lowerBound = 0;
#ifdef ZSTD_MULTITHREAD
bounds.upperBound = ZSTDMT_JOBSIZE_MAX;
#else
bounds.upperBound = 0;
#endif
return bounds;
case ZSTD_c_overlapLog:
#ifdef ZSTD_MULTITHREAD
bounds.lowerBound = ZSTD_OVERLAPLOG_MIN;
bounds.upperBound = ZSTD_OVERLAPLOG_MAX;
#else
bounds.lowerBound = 0;
bounds.upperBound = 0;
#endif
return bounds;
case ZSTD_c_enableDedicatedDictSearch:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_enableLongDistanceMatching:
bounds.lowerBound = (int)ZSTD_ps_auto;
bounds.upperBound = (int)ZSTD_ps_disable;
return bounds;
case ZSTD_c_ldmHashLog:
bounds.lowerBound = ZSTD_LDM_HASHLOG_MIN;
bounds.upperBound = ZSTD_LDM_HASHLOG_MAX;
return bounds;
case ZSTD_c_ldmMinMatch:
bounds.lowerBound = ZSTD_LDM_MINMATCH_MIN;
bounds.upperBound = ZSTD_LDM_MINMATCH_MAX;
return bounds;
case ZSTD_c_ldmBucketSizeLog:
bounds.lowerBound = ZSTD_LDM_BUCKETSIZELOG_MIN;
bounds.upperBound = ZSTD_LDM_BUCKETSIZELOG_MAX;
return bounds;
case ZSTD_c_ldmHashRateLog:
bounds.lowerBound = ZSTD_LDM_HASHRATELOG_MIN;
bounds.upperBound = ZSTD_LDM_HASHRATELOG_MAX;
return bounds;
/* experimental parameters */
case ZSTD_c_rsyncable:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_forceMaxWindow :
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_format:
ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
bounds.lowerBound = ZSTD_f_zstd1;
bounds.upperBound = ZSTD_f_zstd1_magicless; /* note : how to ensure at compile time that this is the highest value enum ? */
return bounds;
case ZSTD_c_forceAttachDict:
ZSTD_STATIC_ASSERT(ZSTD_dictDefaultAttach < ZSTD_dictForceLoad);
bounds.lowerBound = ZSTD_dictDefaultAttach;
bounds.upperBound = ZSTD_dictForceLoad; /* note : how to ensure at compile time that this is the highest value enum ? */
return bounds;
case ZSTD_c_literalCompressionMode:
ZSTD_STATIC_ASSERT(ZSTD_ps_auto < ZSTD_ps_enable && ZSTD_ps_enable < ZSTD_ps_disable);
bounds.lowerBound = (int)ZSTD_ps_auto;
bounds.upperBound = (int)ZSTD_ps_disable;
return bounds;
case ZSTD_c_targetCBlockSize:
bounds.lowerBound = ZSTD_TARGETCBLOCKSIZE_MIN;
bounds.upperBound = ZSTD_TARGETCBLOCKSIZE_MAX;
return bounds;
case ZSTD_c_srcSizeHint:
bounds.lowerBound = ZSTD_SRCSIZEHINT_MIN;
bounds.upperBound = ZSTD_SRCSIZEHINT_MAX;
return bounds;
case ZSTD_c_stableInBuffer:
case ZSTD_c_stableOutBuffer:
bounds.lowerBound = (int)ZSTD_bm_buffered;
bounds.upperBound = (int)ZSTD_bm_stable;
return bounds;
case ZSTD_c_blockDelimiters:
bounds.lowerBound = (int)ZSTD_sf_noBlockDelimiters;
bounds.upperBound = (int)ZSTD_sf_explicitBlockDelimiters;
return bounds;
case ZSTD_c_validateSequences:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_useBlockSplitter:
bounds.lowerBound = (int)ZSTD_ps_auto;
bounds.upperBound = (int)ZSTD_ps_disable;
return bounds;
case ZSTD_c_useRowMatchFinder:
bounds.lowerBound = (int)ZSTD_ps_auto;
bounds.upperBound = (int)ZSTD_ps_disable;
return bounds;
case ZSTD_c_deterministicRefPrefix:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_prefetchCDictTables:
bounds.lowerBound = (int)ZSTD_ps_auto;
bounds.upperBound = (int)ZSTD_ps_disable;
return bounds;
case ZSTD_c_enableSeqProducerFallback:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_maxBlockSize:
bounds.lowerBound = ZSTD_BLOCKSIZE_MAX_MIN;
bounds.upperBound = ZSTD_BLOCKSIZE_MAX;
return bounds;
case ZSTD_c_searchForExternalRepcodes:
bounds.lowerBound = (int)ZSTD_ps_auto;
bounds.upperBound = (int)ZSTD_ps_disable;
return bounds;
default:
bounds.error = ERROR(parameter_unsupported);
return bounds;
}
}
/* ZSTD_cParam_clampBounds:
* Clamps the value into the bounded range.
*/
static size_t ZSTD_cParam_clampBounds(ZSTD_cParameter cParam, int* value)
{
ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);
if (ZSTD_isError(bounds.error)) return bounds.error;
if (*value < bounds.lowerBound) *value = bounds.lowerBound;
if (*value > bounds.upperBound) *value = bounds.upperBound;
return 0;
}
#define BOUNDCHECK(cParam, val) { \
RETURN_ERROR_IF(!ZSTD_cParam_withinBounds(cParam,val), \
parameter_outOfBound, "Param out of bounds"); \
}
static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
{
switch(param)
{
case ZSTD_c_compressionLevel:
case ZSTD_c_hashLog:
case ZSTD_c_chainLog:
case ZSTD_c_searchLog:
case ZSTD_c_minMatch:
case ZSTD_c_targetLength:
case ZSTD_c_strategy:
return 1;
case ZSTD_c_format:
case ZSTD_c_windowLog:
case ZSTD_c_contentSizeFlag:
case ZSTD_c_checksumFlag:
case ZSTD_c_dictIDFlag:
case ZSTD_c_forceMaxWindow :
case ZSTD_c_nbWorkers:
case ZSTD_c_jobSize:
case ZSTD_c_overlapLog:
case ZSTD_c_rsyncable:
case ZSTD_c_enableDedicatedDictSearch:
case ZSTD_c_enableLongDistanceMatching:
case ZSTD_c_ldmHashLog:
case ZSTD_c_ldmMinMatch:
case ZSTD_c_ldmBucketSizeLog:
case ZSTD_c_ldmHashRateLog:
case ZSTD_c_forceAttachDict:
case ZSTD_c_literalCompressionMode:
case ZSTD_c_targetCBlockSize:
case ZSTD_c_srcSizeHint:
case ZSTD_c_stableInBuffer:
case ZSTD_c_stableOutBuffer:
case ZSTD_c_blockDelimiters:
case ZSTD_c_validateSequences:
case ZSTD_c_useBlockSplitter:
case ZSTD_c_useRowMatchFinder:
case ZSTD_c_deterministicRefPrefix:
case ZSTD_c_prefetchCDictTables:
case ZSTD_c_enableSeqProducerFallback:
case ZSTD_c_maxBlockSize:
case ZSTD_c_searchForExternalRepcodes:
default:
return 0;
}
}
size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
{
DEBUGLOG(4, "ZSTD_CCtx_setParameter (%i, %i)", (int)param, value);
if (cctx->streamStage != zcss_init) {
if (ZSTD_isUpdateAuthorized(param)) {
cctx->cParamsChanged = 1;
} else {
RETURN_ERROR(stage_wrong, "can only set params in cctx init stage");
} }
switch(param)
{
case ZSTD_c_nbWorkers:
RETURN_ERROR_IF((value!=0) && cctx->staticSize, parameter_unsupported,
"MT not compatible with static alloc");
break;
case ZSTD_c_compressionLevel:
case ZSTD_c_windowLog:
case ZSTD_c_hashLog:
case ZSTD_c_chainLog:
case ZSTD_c_searchLog:
case ZSTD_c_minMatch:
case ZSTD_c_targetLength:
case ZSTD_c_strategy:
case ZSTD_c_ldmHashRateLog:
case ZSTD_c_format:
case ZSTD_c_contentSizeFlag:
case ZSTD_c_checksumFlag:
case ZSTD_c_dictIDFlag:
case ZSTD_c_forceMaxWindow:
case ZSTD_c_forceAttachDict:
case ZSTD_c_literalCompressionMode:
case ZSTD_c_jobSize:
case ZSTD_c_overlapLog:
case ZSTD_c_rsyncable:
case ZSTD_c_enableDedicatedDictSearch:
case ZSTD_c_enableLongDistanceMatching:
case ZSTD_c_ldmHashLog:
case ZSTD_c_ldmMinMatch:
case ZSTD_c_ldmBucketSizeLog:
case ZSTD_c_targetCBlockSize:
case ZSTD_c_srcSizeHint:
case ZSTD_c_stableInBuffer:
case ZSTD_c_stableOutBuffer:
case ZSTD_c_blockDelimiters:
case ZSTD_c_validateSequences:
case ZSTD_c_useBlockSplitter:
case ZSTD_c_useRowMatchFinder:
case ZSTD_c_deterministicRefPrefix:
case ZSTD_c_prefetchCDictTables:
case ZSTD_c_enableSeqProducerFallback:
case ZSTD_c_maxBlockSize:
case ZSTD_c_searchForExternalRepcodes:
break;
default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
}
return ZSTD_CCtxParams_setParameter(&cctx->requestedParams, param, value);
}
size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
ZSTD_cParameter param, int value)
{
DEBUGLOG(4, "ZSTD_CCtxParams_setParameter (%i, %i)", (int)param, value);
switch(param)
{
case ZSTD_c_format :
BOUNDCHECK(ZSTD_c_format, value);
CCtxParams->format = (ZSTD_format_e)value;
return (size_t)CCtxParams->format;
case ZSTD_c_compressionLevel : {
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
if (value == 0)
CCtxParams->compressionLevel = ZSTD_CLEVEL_DEFAULT; /* 0 == default */
else
CCtxParams->compressionLevel = value;
if (CCtxParams->compressionLevel >= 0) return (size_t)CCtxParams->compressionLevel;
return 0; /* return type (size_t) cannot represent negative values */
}
case ZSTD_c_windowLog :
if (value!=0) /* 0 => use default */
BOUNDCHECK(ZSTD_c_windowLog, value);
CCtxParams->cParams.windowLog = (U32)value;
return CCtxParams->cParams.windowLog;
case ZSTD_c_hashLog :
if (value!=0) /* 0 => use default */
BOUNDCHECK(ZSTD_c_hashLog, value);
CCtxParams->cParams.hashLog = (U32)value;
return CCtxParams->cParams.hashLog;
case ZSTD_c_chainLog :
if (value!=0) /* 0 => use default */
BOUNDCHECK(ZSTD_c_chainLog, value);
CCtxParams->cParams.chainLog = (U32)value;
return CCtxParams->cParams.chainLog;
case ZSTD_c_searchLog :
if (value!=0) /* 0 => use default */
BOUNDCHECK(ZSTD_c_searchLog, value);
CCtxParams->cParams.searchLog = (U32)value;
return (size_t)value;
case ZSTD_c_minMatch :
if (value!=0) /* 0 => use default */
BOUNDCHECK(ZSTD_c_minMatch, value);
CCtxParams->cParams.minMatch = (U32)value;
return CCtxParams->cParams.minMatch;
case ZSTD_c_targetLength :
BOUNDCHECK(ZSTD_c_targetLength, value);
CCtxParams->cParams.targetLength = (U32)value;
return CCtxParams->cParams.targetLength;
case ZSTD_c_strategy :
if (value!=0) /* 0 => use default */
BOUNDCHECK(ZSTD_c_strategy, value);
CCtxParams->cParams.strategy = (ZSTD_strategy)value;
return (size_t)CCtxParams->cParams.strategy;
case ZSTD_c_contentSizeFlag :
/* Content size written in frame header _when known_ (default:1) */
DEBUGLOG(4, "set content size flag = %u", (value!=0));
CCtxParams->fParams.contentSizeFlag = value != 0;
return (size_t)CCtxParams->fParams.contentSizeFlag;
case ZSTD_c_checksumFlag :
/* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
CCtxParams->fParams.checksumFlag = value != 0;
return (size_t)CCtxParams->fParams.checksumFlag;
case ZSTD_c_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
DEBUGLOG(4, "set dictIDFlag = %u", (value!=0));
CCtxParams->fParams.noDictIDFlag = !value;
return !CCtxParams->fParams.noDictIDFlag;
case ZSTD_c_forceMaxWindow :
CCtxParams->forceWindow = (value != 0);
return (size_t)CCtxParams->forceWindow;
case ZSTD_c_forceAttachDict : {
const ZSTD_dictAttachPref_e pref = (ZSTD_dictAttachPref_e)value;
BOUNDCHECK(ZSTD_c_forceAttachDict, (int)pref);
CCtxParams->attachDictPref = pref;
return CCtxParams->attachDictPref;
}
case ZSTD_c_literalCompressionMode : {
const ZSTD_paramSwitch_e lcm = (ZSTD_paramSwitch_e)value;
BOUNDCHECK(ZSTD_c_literalCompressionMode, (int)lcm);
CCtxParams->literalCompressionMode = lcm;
return CCtxParams->literalCompressionMode;
}
case ZSTD_c_nbWorkers :
#ifndef ZSTD_MULTITHREAD
RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
return 0;
#else
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
CCtxParams->nbWorkers = value;
return CCtxParams->nbWorkers;
#endif
case ZSTD_c_jobSize :
#ifndef ZSTD_MULTITHREAD
RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
return 0;
#else
/* Adjust to the minimum non-default value. */
if (value != 0 && value < ZSTDMT_JOBSIZE_MIN)
value = ZSTDMT_JOBSIZE_MIN;
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
assert(value >= 0);
CCtxParams->jobSize = value;
return CCtxParams->jobSize;
#endif
case ZSTD_c_overlapLog :
#ifndef ZSTD_MULTITHREAD
RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
return 0;
#else
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value), "");
CCtxParams->overlapLog = value;
return CCtxParams->overlapLog;
#endif
case ZSTD_c_rsyncable :
#ifndef ZSTD_MULTITHREAD
RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
return 0;
#else
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value), "");
CCtxParams->rsyncable = value;
return CCtxParams->rsyncable;
#endif
case ZSTD_c_enableDedicatedDictSearch :
CCtxParams->enableDedicatedDictSearch = (value!=0);
return (size_t)CCtxParams->enableDedicatedDictSearch;
case ZSTD_c_enableLongDistanceMatching :
BOUNDCHECK(ZSTD_c_enableLongDistanceMatching, value);
CCtxParams->ldmParams.enableLdm = (ZSTD_paramSwitch_e)value;
return CCtxParams->ldmParams.enableLdm;
case ZSTD_c_ldmHashLog :
if (value!=0) /* 0 ==> auto */
BOUNDCHECK(ZSTD_c_ldmHashLog, value);
CCtxParams->ldmParams.hashLog = (U32)value;
return CCtxParams->ldmParams.hashLog;
case ZSTD_c_ldmMinMatch :
if (value!=0) /* 0 ==> default */
BOUNDCHECK(ZSTD_c_ldmMinMatch, value);
CCtxParams->ldmParams.minMatchLength = (U32)value;
return CCtxParams->ldmParams.minMatchLength;
case ZSTD_c_ldmBucketSizeLog :
if (value!=0) /* 0 ==> default */
BOUNDCHECK(ZSTD_c_ldmBucketSizeLog, value);
CCtxParams->ldmParams.bucketSizeLog = (U32)value;
return CCtxParams->ldmParams.bucketSizeLog;
case ZSTD_c_ldmHashRateLog :
if (value!=0) /* 0 ==> default */
BOUNDCHECK(ZSTD_c_ldmHashRateLog, value);
CCtxParams->ldmParams.hashRateLog = (U32)value;
return CCtxParams->ldmParams.hashRateLog;
case ZSTD_c_targetCBlockSize :
if (value!=0) /* 0 ==> default */
BOUNDCHECK(ZSTD_c_targetCBlockSize, value);
CCtxParams->targetCBlockSize = (U32)value;
return CCtxParams->targetCBlockSize;
case ZSTD_c_srcSizeHint :
if (value!=0) /* 0 ==> default */
BOUNDCHECK(ZSTD_c_srcSizeHint, value);
CCtxParams->srcSizeHint = value;
return (size_t)CCtxParams->srcSizeHint;
case ZSTD_c_stableInBuffer:
BOUNDCHECK(ZSTD_c_stableInBuffer, value);
CCtxParams->inBufferMode = (ZSTD_bufferMode_e)value;
return CCtxParams->inBufferMode;
case ZSTD_c_stableOutBuffer:
BOUNDCHECK(ZSTD_c_stableOutBuffer, value);
CCtxParams->outBufferMode = (ZSTD_bufferMode_e)value;
return CCtxParams->outBufferMode;
case ZSTD_c_blockDelimiters:
BOUNDCHECK(ZSTD_c_blockDelimiters, value);
CCtxParams->blockDelimiters = (ZSTD_sequenceFormat_e)value;
return CCtxParams->blockDelimiters;
case ZSTD_c_validateSequences:
BOUNDCHECK(ZSTD_c_validateSequences, value);
CCtxParams->validateSequences = value;
return CCtxParams->validateSequences;
case ZSTD_c_useBlockSpli
Raw
zstd.h
View raw

(Sorry about that, but we can’t show files that are this big right now.)

Raw
zstd_errors.h
View raw

(Sorry about that, but we can’t show files that are this big right now.)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment