mirror/CMake
1
0
Fork 0
mirror of https://github.com/Kitware/CMake.git synced 2026-05-07 06:40:16 -05:00
Code Issues Packages Projects Releases Wiki Activity

Merge topic 'update-zstd' into release-3.22

Browse Source 
ddced0c6b6 zstd: Fix trace compilation on AIX
ee98d9378f zstd: Fix ZSTD_VecMask_next compilation of software ctz implementation
eb8db99596 Merge branch 'upstream-zstd' into update-zstd
da1e841ccf zstd 2021-05-14 (a488ba11)
8c65ea4744 zstd: Update script to get zstd 1.5.0

Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !6622
This commit is contained in:
Brad King
2021-10-14 11:45:49 +00:00
committed by Kitware Robot
parent 127cb6481c ddced0c6b6
commit 9b8a551d84
68 changed files with 8570 additions and 3322 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Utilities/Scripts/update-zstd.bash
+2 -2
View File
@@ -8,7 +8,7 @@ readonly name="zstd"
readonly ownership="zstd upstream <kwrobot@kitware.com>"
readonly subtree="Utilities/cmzstd"
readonly repo="https://github.com/facebook/zstd.git"
readonly tag="v1.4.5"
readonly tag="v1.5.0"
readonly shortlog=false
readonly paths="
LICENSE
@@ -23,7 +23,7 @@ readonly paths="
lib/deprecated/*.h
lib/dictBuilder/*.c
lib/dictBuilder/*.h
lib/zstd.h
lib/*.h
"
extract_source () {
Utilities/cmzstd/README.md
+9 -3
View File
@@ -176,6 +176,12 @@ Going into `build` directory, you will find additional possibilities:
You can build the zstd binary via buck by executing: `buck build programs:zstd` from the root of the repo.
The output binary will be in `buck-out/gen/programs/`.
## Testing
You can run quick local smoke tests by executing the `playTest.sh` script from the `src/tests` directory.
Two env variables `$ZSTD_BIN` and `$DATAGEN_BIN` are needed for the test script to locate the zstd and datagen binary.
For information on CI testing, please refer to TESTING.md
## Status
Zstandard is currently deployed within Facebook. It is used continuously to compress large amounts of data in multiple formats and use cases.
@@ -187,7 +193,7 @@ Zstandard is dual-licensed under [BSD](LICENSE) and [GPLv2](COPYING).
## Contributing
The "dev" branch is the one where all contributions are merged before reaching "master".
If you plan to propose a patch, please commit into the "dev" branch, or its own feature branch.
Direct commit to "master" are not permitted.
The `dev` branch is the one where all contributions are merged before reaching `release`.
If you plan to propose a patch, please commit into the `dev` branch, or its own feature branch.
Direct commit to `release` are not permitted.
For more information, please read [CONTRIBUTING](CONTRIBUTING.md).
Utilities/cmzstd/lib/common/bitstream.h
+25 -16
View File
@@ -1,7 +1,7 @@
/* ******************************************************************
* bitstream
* Part of FSE library
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -17,7 +17,6 @@
#if defined (__cplusplus)
extern "C" {
#endif
/*
* This API consists of small unitary functions, which must be inlined for best performance.
* Since link-time-optimization is not available for all compilers,
@@ -36,10 +35,12 @@ extern "C" {
/*=========================================
* Target specific
=========================================*/
#if defined(__BMI__) && defined(__GNUC__)
# include <immintrin.h> /* support for bextr (experimental) */
#elif defined(__ICCARM__)
# include <intrinsics.h>
#ifndef ZSTD_NO_INTRINSICS
# if defined(__BMI__) && defined(__GNUC__)
# include <immintrin.h> /* support for bextr (experimental) */
# elif defined(__ICCARM__)
# include <intrinsics.h>
# endif
#endif
#define STREAM_ACCUMULATOR_MIN_32 25
@@ -141,8 +142,12 @@ MEM_STATIC unsigned BIT_highbit32 (U32 val)
assert(val != 0);
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
return _BitScanReverse ( &r, val ) ? (unsigned)r : 0;
# if STATIC_BMI2 == 1
return _lzcnt_u32(val) ^ 31;
# else
unsigned long r = 0;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* Use GCC Intrinsic */
return __builtin_clz (val) ^ 31;
# elif defined(__ICCARM__) /* IAR Intrinsic */
@@ -198,7 +203,7 @@ MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
size_t value, unsigned nbBits)
{
MEM_STATIC_ASSERT(BIT_MASK_SIZE == 32);
DEBUG_STATIC_ASSERT(BIT_MASK_SIZE == 32);
assert(nbBits < BIT_MASK_SIZE);
assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
bitC->bitContainer |= (value & BIT_mask[nbBits]) << bitC->bitPos;
@@ -271,7 +276,7 @@ MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
*/
MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
{
if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
if (srcSize < 1) { ZSTD_memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
bitD->start = (const char*)srcBuffer;
bitD->limitPtr = bitD->start + sizeof(bitD->bitContainer);
@@ -317,12 +322,12 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
return srcSize;
}
MEM_STATIC size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
{
return bitContainer >> start;
}
MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
{
U32 const regMask = sizeof(bitContainer)*8 - 1;
/* if start > regMask, bitstream is corrupted, and result is undefined */
@@ -330,10 +335,14 @@ MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 co
return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
}
MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
{
#if defined(STATIC_BMI2) && STATIC_BMI2 == 1
return _bzhi_u64(bitContainer, nbBits);
#else
assert(nbBits < BIT_MASK_SIZE);
return bitContainer & BIT_mask[nbBits];
#endif
}
/*! BIT_lookBits() :
@@ -342,7 +351,7 @@ MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
* On 32-bits, maxNbBits==24.
* On 64-bits, maxNbBits==56.
* @return : value extracted */
MEM_STATIC size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
MEM_STATIC FORCE_INLINE_ATTR size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
{
/* arbitrate between double-shift and shift+mask */
#if 1
@@ -365,7 +374,7 @@ MEM_STATIC size_t BIT_lookBitsFast(const BIT_DStream_t* bitD, U32 nbBits)
return (bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> (((regMask+1)-nbBits) & regMask);
}
MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
MEM_STATIC FORCE_INLINE_ATTR void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
{
bitD->bitsConsumed += nbBits;
}
@@ -374,7 +383,7 @@ MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
* Read (consume) next n bits from local register and update.
* Pay attention to not read more than nbBits contained into local register.
* @return : extracted value. */
MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
MEM_STATIC FORCE_INLINE_ATTR size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
{
size_t const value = BIT_lookBits(bitD, nbBits);
BIT_skipBits(bitD, nbBits);
Utilities/cmzstd/lib/common/compiler.h
+118 -4
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -38,6 +38,17 @@
#endif
/**
On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC).
This explictly marks such functions as __cdecl so that the code will still compile
if a CC other than __cdecl has been made the default.
*/
#if defined(_MSC_VER)
# define WIN_CDECL __cdecl
#else
# define WIN_CDECL
#endif
/**
* FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
* parameters. They must be inlined for the compiler to eliminate the constant
@@ -79,6 +90,7 @@
# endif
#endif
/* target attribute */
#ifndef __has_attribute
#define __has_attribute(x) 0 /* Compatibility with non-clang compilers. */
@@ -114,12 +126,12 @@
# include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
# define PREFETCH_L1(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
# define PREFETCH_L2(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T1)
# elif defined(__aarch64__)
# define PREFETCH_L1(ptr) __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
# define PREFETCH_L2(ptr) __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
# elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
# define PREFETCH_L1(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
# define PREFETCH_L2(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
# elif defined(__aarch64__)
# define PREFETCH_L1(ptr) __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
# define PREFETCH_L2(ptr) __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
# else
# define PREFETCH_L1(ptr) (void)(ptr) /* disabled */
# define PREFETCH_L2(ptr) (void)(ptr) /* disabled */
@@ -172,4 +184,106 @@
# pragma warning(disable : 4324) /* disable: C4324: padded structure */
#endif
/*Like DYNAMIC_BMI2 but for compile time determination of BMI2 support*/
#ifndef STATIC_BMI2
# if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))
# ifdef __AVX2__ //MSVC does not have a BMI2 specific flag, but every CPU that supports AVX2 also supports BMI2
# define STATIC_BMI2 1
# endif
# endif
#endif
#ifndef STATIC_BMI2
#define STATIC_BMI2 0
#endif
/* compat. with non-clang compilers */
#ifndef __has_builtin
# define __has_builtin(x) 0
#endif
/* compat. with non-clang compilers */
#ifndef __has_feature
# define __has_feature(x) 0
#endif
/* detects whether we are being compiled under msan */
#ifndef ZSTD_MEMORY_SANITIZER
# if __has_feature(memory_sanitizer)
# define ZSTD_MEMORY_SANITIZER 1
# else
# define ZSTD_MEMORY_SANITIZER 0
# endif
#endif
#if ZSTD_MEMORY_SANITIZER
/* Not all platforms that support msan provide sanitizers/msan_interface.h.
* We therefore declare the functions we need ourselves, rather than trying to
* include the header file... */
#include <stddef.h> /* size_t */
#define ZSTD_DEPS_NEED_STDINT
#include "zstd_deps.h" /* intptr_t */
/* Make memory region fully initialized (without changing its contents). */
void __msan_unpoison(const volatile void *a, size_t size);
/* Make memory region fully uninitialized (without changing its contents).
This is a legacy interface that does not update origin information. Use
__msan_allocated_memory() instead. */
void __msan_poison(const volatile void *a, size_t size);
/* Returns the offset of the first (at least partially) poisoned byte in the
memory range, or -1 if the whole range is good. */
intptr_t __msan_test_shadow(const volatile void *x, size_t size);
#endif
/* detects whether we are being compiled under asan */
#ifndef ZSTD_ADDRESS_SANITIZER
# if __has_feature(address_sanitizer)
# define ZSTD_ADDRESS_SANITIZER 1
# elif defined(__SANITIZE_ADDRESS__)
# define ZSTD_ADDRESS_SANITIZER 1
# else
# define ZSTD_ADDRESS_SANITIZER 0
# endif
#endif
#if ZSTD_ADDRESS_SANITIZER
/* Not all platforms that support asan provide sanitizers/asan_interface.h.
* We therefore declare the functions we need ourselves, rather than trying to
* include the header file... */
#include <stddef.h> /* size_t */
/**
* Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable.
*
* This memory must be previously allocated by your program. Instrumented
* code is forbidden from accessing addresses in this region until it is
* unpoisoned. This function is not guaranteed to poison the entire region -
* it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan
* alignment restrictions.
*
* \note This function is not thread-safe because no two threads can poison or
* unpoison memory in the same memory region simultaneously.
*
* \param addr Start of memory region.
* \param size Size of memory region. */
void __asan_poison_memory_region(void const volatile *addr, size_t size);
/**
* Marks a memory region (<c>[addr, addr+size)</c>) as addressable.
*
* This memory must be previously allocated by your program. Accessing
* addresses in this region is allowed until this region is poisoned again.
* This function could unpoison a super-region of <c>[addr, addr+size)</c> due
* to ASan alignment restrictions.
*
* \note This function is not thread-safe because no two threads can
* poison or unpoison memory in the same memory region simultaneously.
*
* \param addr Start of memory region.
* \param size Size of memory region. */
void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
#endif
#endif /* ZSTD_COMPILER_H */
Utilities/cmzstd/lib/common/cpu.h
+1 -3
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2018-2020, Facebook, Inc.
* Copyright (c) Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -16,8 +16,6 @@
* https://github.com/facebook/folly/blob/master/folly/CpuId.h
*/
#include <string.h>
#include "mem.h"
#ifdef _MSC_VER
Utilities/cmzstd/lib/common/debug.c
+1 -1
View File
@@ -1,7 +1,7 @@
/* ******************************************************************
* debug
* Part of FSE library
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
Utilities/cmzstd/lib/common/debug.h
+12 -19
View File
@@ -1,7 +1,7 @@
/* ******************************************************************
* debug
* Part of FSE library
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -51,15 +51,6 @@ extern "C" {
#endif
/* DEBUGFILE can be defined externally,
* typically through compiler command line.
* note : currently useless.
* Value must be stderr or stdout */
#ifndef DEBUGFILE
# define DEBUGFILE stderr
#endif
/* recommended values for DEBUGLEVEL :
* 0 : release mode, no debug, all run-time checks disabled
* 1 : enables assert() only, no display
@@ -76,7 +67,8 @@ extern "C" {
*/
#if (DEBUGLEVEL>=1)
# include <assert.h>
# define ZSTD_DEPS_NEED_ASSERT
# include "zstd_deps.h"
#else
# ifndef assert /* assert may be already defined, due to prior #include <assert.h> */
# define assert(condition) ((void)0) /* disable assert (default) */
@@ -84,7 +76,8 @@ extern "C" {
#endif
#if (DEBUGLEVEL>=2)
# include <stdio.h>
# define ZSTD_DEPS_NEED_IO
# include "zstd_deps.h"
extern int g_debuglevel; /* the variable is only declared,
it actually lives in debug.c,
and is shared by the whole process.
@@ -92,14 +85,14 @@ extern int g_debuglevel; /* the variable is only declared,
It's useful when enabling very verbose levels
on selective conditions (such as position in src) */
# define RAWLOG(l, ...) { \
if (l<=g_debuglevel) { \
fprintf(stderr, __VA_ARGS__); \
# define RAWLOG(l, ...) { \
if (l<=g_debuglevel) { \
ZSTD_DEBUG_PRINT(__VA_ARGS__); \
} }
# define DEBUGLOG(l, ...) { \
if (l<=g_debuglevel) { \
fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
fprintf(stderr, " \n"); \
# define DEBUGLOG(l, ...) { \
if (l<=g_debuglevel) { \
ZSTD_DEBUG_PRINT(__FILE__ ": " __VA_ARGS__); \
ZSTD_DEBUG_PRINT(" \n"); \
} }
#else
# define RAWLOG(l, ...) {} /* disabled */
Utilities/cmzstd/lib/common/entropy_common.c
+190 -44
View File
@@ -1,6 +1,6 @@
/* ******************************************************************
* Common functions of New Generation Entropy library
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -38,8 +38,31 @@ const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
/*-**************************************************************
* FSE NCount encoding-decoding
****************************************************************/
size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
static U32 FSE_ctz(U32 val)
{
assert(val != 0);
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
return _BitScanForward(&r, val) ? (unsigned)r : 0;
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* GCC Intrinsic */
return __builtin_ctz(val);
# elif defined(__ICCARM__) /* IAR Intrinsic */
return __CTZ(val);
# else /* Software version */
U32 count = 0;
while ((val & 1) == 0) {
val >>= 1;
++count;
}
return count;
# endif
}
}
FORCE_INLINE_TEMPLATE
size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
{
const BYTE* const istart = (const BYTE*) headerBuffer;
const BYTE* const iend = istart + hbSize;
@@ -50,23 +73,23 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
U32 bitStream;
int bitCount;
unsigned charnum = 0;
unsigned const maxSV1 = *maxSVPtr + 1;
int previous0 = 0;
if (hbSize < 4) {
/* This function only works when hbSize >= 4 */
char buffer[4];
memset(buffer, 0, sizeof(buffer));
memcpy(buffer, headerBuffer, hbSize);
if (hbSize < 8) {
/* This function only works when hbSize >= 8 */
char buffer[8] = {0};
ZSTD_memcpy(buffer, headerBuffer, hbSize);
{ size_t const countSize = FSE_readNCount(normalizedCounter, maxSVPtr, tableLogPtr,
buffer, sizeof(buffer));
if (FSE_isError(countSize)) return countSize;
if (countSize > hbSize) return ERROR(corruption_detected);
return countSize;
} }
assert(hbSize >= 4);
assert(hbSize >= 8);
/* init */
memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0])); /* all symbols not present in NCount have a frequency of 0 */
ZSTD_memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0])); /* all symbols not present in NCount have a frequency of 0 */
bitStream = MEM_readLE32(ip);
nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG; /* extract tableLog */
if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
@@ -77,36 +100,58 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
threshold = 1<<nbBits;
nbBits++;
while ((remaining>1) & (charnum<=*maxSVPtr)) {
for (;;) {
if (previous0) {
unsigned n0 = charnum;
while ((bitStream & 0xFFFF) == 0xFFFF) {
n0 += 24;
if (ip < iend-5) {
ip += 2;
bitStream = MEM_readLE32(ip) >> bitCount;
/* Count the number of repeats. Each time the
* 2-bit repeat code is 0b11 there is another
* repeat.
* Avoid UB by setting the high bit to 1.
*/
int repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
while (repeats >= 12) {
charnum += 3 * 12;
if (LIKELY(ip <= iend-7)) {
ip += 3;
} else {
bitStream >>= 16;
bitCount += 16;
} }
while ((bitStream & 3) == 3) {
n0 += 3;
bitStream >>= 2;
bitCount += 2;
bitCount -= (int)(8 * (iend - 7 - ip));
bitCount &= 31;
ip = iend - 4;
}
bitStream = MEM_readLE32(ip) >> bitCount;
repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
}
n0 += bitStream & 3;
charnum += 3 * repeats;
bitStream >>= 2 * repeats;
bitCount += 2 * repeats;
/* Add the final repeat which isn't 0b11. */
assert((bitStream & 3) < 3);
charnum += bitStream & 3;
bitCount += 2;
if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
while (charnum < n0) normalizedCounter[charnum++] = 0;
if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
/* This is an error, but break and return an error
* at the end, because returning out of a loop makes
* it harder for the compiler to optimize.
*/
if (charnum >= maxSV1) break;
/* We don't need to set the normalized count to 0
* because we already memset the whole buffer to 0.
*/
if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
assert((bitCount >> 3) <= 3); /* For first condition to work */
ip += bitCount>>3;
bitCount &= 7;
bitStream = MEM_readLE32(ip) >> bitCount;
} else {
bitStream >>= 2;
} }
{ int const max = (2*threshold-1) - remaining;
bitCount -= (int)(8 * (iend - 4 - ip));
bitCount &= 31;
ip = iend - 4;
}
bitStream = MEM_readLE32(ip) >> bitCount;
}
{
int const max = (2*threshold-1) - remaining;
int count;
if ((bitStream & (threshold-1)) < (U32)max) {
@@ -119,24 +164,43 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
}
count--; /* extra accuracy */
remaining -= count < 0 ? -count : count; /* -1 means +1 */
/* When it matters (small blocks), this is a
* predictable branch, because we don't use -1.
*/
if (count >= 0) {
remaining -= count;
} else {
assert(count == -1);
remaining += count;
}
normalizedCounter[charnum++] = (short)count;
previous0 = !count;
while (remaining < threshold) {
nbBits--;
threshold >>= 1;
}
if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
assert(threshold > 1);
if (remaining < threshold) {
/* This branch can be folded into the
* threshold update condition because we
* know that threshold > 1.
*/
if (remaining <= 1) break;
nbBits = BIT_highbit32(remaining) + 1;
threshold = 1 << (nbBits - 1);
}
if (charnum >= maxSV1) break;
if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
ip += bitCount>>3;
bitCount &= 7;
} else {
bitCount -= (int)(8 * (iend - 4 - ip));
bitCount &= 31;
ip = iend - 4;
}
bitStream = MEM_readLE32(ip) >> (bitCount & 31);
} } /* while ((remaining>1) & (charnum<=*maxSVPtr)) */
bitStream = MEM_readLE32(ip) >> bitCount;
} }
if (remaining != 1) return ERROR(corruption_detected);
/* Only possible when there are too many zeros. */
if (charnum > maxSV1) return ERROR(maxSymbolValue_tooSmall);
if (bitCount > 32) return ERROR(corruption_detected);
*maxSVPtr = charnum-1;
@@ -144,6 +208,43 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
return ip-istart;
}
/* Avoids the FORCE_INLINE of the _body() function. */
static size_t FSE_readNCount_body_default(
short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
{
return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
}
#if DYNAMIC_BMI2
TARGET_ATTRIBUTE("bmi2") static size_t FSE_readNCount_body_bmi2(
short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
{
return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
}
#endif
size_t FSE_readNCount_bmi2(
short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize, int bmi2)
{
#if DYNAMIC_BMI2
if (bmi2) {
return FSE_readNCount_body_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
}
#endif
(void)bmi2;
return FSE_readNCount_body_default(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize);
}
size_t FSE_readNCount(
short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
const void* headerBuffer, size_t hbSize)
{
return FSE_readNCount_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize, /* bmi2 */ 0);
}
/*! HUF_readStats() :
Read compact Huffman tree, saved by HUF_writeCTable().
@@ -155,6 +256,17 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize)
{
U32 wksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp), /* bmi2 */ 0);
}
FORCE_INLINE_TEMPLATE size_t
HUF_readStats_body(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize,
int bmi2)
{
U32 weightTotal;
const BYTE* ip = (const BYTE*) src;
@@ -163,7 +275,7 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
if (!srcSize) return ERROR(srcSize_wrong);
iSize = ip[0];
/* memset(huffWeight, 0, hwSize); *//* is not necessary, even though some analyzer complain ... */
/* ZSTD_memset(huffWeight, 0, hwSize); *//* is not necessary, even though some analyzer complain ... */
if (iSize >= 128) { /* special header */
oSize = iSize - 127;
@@ -177,14 +289,14 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
huffWeight[n+1] = ip[n/2] & 15;
} } }
else { /* header compressed with FSE (normal case) */
FSE_DTable fseWorkspace[FSE_DTABLE_SIZE_U32(6)]; /* 6 is max possible tableLog for HUF header (maybe even 5, to be tested) */
if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
oSize = FSE_decompress_wksp(huffWeight, hwSize-1, ip+1, iSize, fseWorkspace, 6); /* max (hwSize-1) values decoded, as last one is implied */
/* max (hwSize-1) values decoded, as last one is implied */
oSize = FSE_decompress_wksp_bmi2(huffWeight, hwSize-1, ip+1, iSize, 6, workSpace, wkspSize, bmi2);
if (FSE_isError(oSize)) return oSize;
}
/* collect weight stats */
memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
ZSTD_memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
weightTotal = 0;
{ U32 n; for (n=0; n<oSize; n++) {
if (huffWeight[n] >= HUF_TABLELOG_MAX) return ERROR(corruption_detected);
@@ -214,3 +326,37 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
*nbSymbolsPtr = (U32)(oSize+1);
return iSize+1;
}
/* Avoids the FORCE_INLINE of the _body() function. */
static size_t HUF_readStats_body_default(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize)
{
return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 0);
}
#if DYNAMIC_BMI2
static TARGET_ATTRIBUTE("bmi2") size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize)
{
return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 1);
}
#endif
size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize, U32* rankStats,
U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize,
int bmi2)
{
#if DYNAMIC_BMI2
if (bmi2) {
return HUF_readStats_body_bmi2(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
}
#endif
(void)bmi2;
return HUF_readStats_body_default(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
}
Utilities/cmzstd/lib/common/error_private.c
+2 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -48,6 +48,7 @@ const char* ERR_getErrorString(ERR_enum code)
case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
case PREFIX(dstBuffer_wrong): return "Destination buffer is wrong";
case PREFIX(srcBuffer_wrong): return "Source buffer is wrong";
case PREFIX(maxCode):
default: return notErrorCode;
}
Utilities/cmzstd/lib/common/error_private.h
+3 -3
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -21,8 +21,8 @@ extern "C" {
/* ****************************************
* Dependencies
******************************************/
#include <stddef.h> /* size_t */
#include "zstd_errors.h" /* enum list */
#include "../zstd_errors.h" /* enum list */
#include "zstd_deps.h" /* size_t */
/* ****************************************
Utilities/cmzstd/lib/common/fse.h
+40 -12
View File
@@ -1,7 +1,7 @@
/* ******************************************************************
* FSE : Finite State Entropy codec
* Public Prototypes declaration
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -23,7 +23,7 @@ extern "C" {
/*-*****************************************
* Dependencies
******************************************/
#include <stddef.h> /* size_t, ptrdiff_t */
#include "zstd_deps.h" /* size_t, ptrdiff_t */
/*-*****************************************
@@ -137,10 +137,16 @@ FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize
/*! FSE_normalizeCount():
normalize counts so that sum(count[]) == Power_of_2 (2^tableLog)
'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
useLowProbCount is a boolean parameter which trades off compressed size for
faster header decoding. When it is set to 1, the compressed data will be slightly
smaller. And when it is set to 0, FSE_readNCount() and FSE_buildDTable() will be
faster. If you are compressing a small amount of data (< 2 KB) then useLowProbCount=0
is a good default, since header deserialization makes a big speed difference.
Otherwise, useLowProbCount=1 is a good default, since the speed difference is small.
@return : tableLog,
or an errorCode, which can be tested using FSE_isError() */
FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog,
const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
const unsigned* count, size_t srcSize, unsigned maxSymbolValue, unsigned useLowProbCount);
/*! FSE_NCountWriteBound():
Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
@@ -228,6 +234,13 @@ FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter,
unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
const void* rBuffer, size_t rBuffSize);
/*! FSE_readNCount_bmi2():
* Same as FSE_readNCount() but pass bmi2=1 when your CPU supports BMI2 and 0 otherwise.
*/
FSE_PUBLIC_API size_t FSE_readNCount_bmi2(short* normalizedCounter,
unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
const void* rBuffer, size_t rBuffSize, int bmi2);
/*! Constructor and Destructor of FSE_DTable.
Note that its size depends on 'tableLog' */
typedef unsigned FSE_DTable; /* don't allocate that. It's just a way to be more restrictive than void* */
@@ -288,12 +301,12 @@ If there is an error, the function will return an error code, which can be teste
*******************************************/
/* FSE buffer bounds */
#define FSE_NCOUNTBOUND 512
#define FSE_BLOCKBOUND(size) (size + (size>>7) + 4 /* fse states */ + sizeof(size_t) /* bitContainer */)
#define FSE_BLOCKBOUND(size) ((size) + ((size)>>7) + 4 /* fse states */ + sizeof(size_t) /* bitContainer */)
#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size)) /* Macro version, useful for static allocation */
/* It is possible to statically allocate FSE CTable/DTable as a table of FSE_CTable/FSE_DTable using below macros */
#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
#define FSE_DTABLE_SIZE_U32(maxTableLog) (1 + (1<<maxTableLog))
#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) (1 + (1<<((maxTableLog)-1)) + (((maxSymbolValue)+1)*2))
#define FSE_DTABLE_SIZE_U32(maxTableLog) (1 + (1<<(maxTableLog)))
/* or use the size to malloc() space directly. Pay attention to alignment restrictions though */
#define FSE_CTABLE_SIZE(maxTableLog, maxSymbolValue) (FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(FSE_CTable))
@@ -309,9 +322,9 @@ unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsi
/* FSE_compress_wksp() :
* Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
* FSE_WKSP_SIZE_U32() provides the minimum size required for `workSpace` as a table of FSE_CTable.
* FSE_COMPRESS_WKSP_SIZE_U32() provides the minimum size required for `workSpace` as a table of FSE_CTable.
*/
#define FSE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
#define FSE_COMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits);
@@ -322,18 +335,30 @@ size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
/* FSE_buildCTable_wksp() :
* Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
* `wkspSize` must be >= `(1<<tableLog)`.
* `wkspSize` must be >= `FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog)` of `unsigned`.
*/
#define FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog) (maxSymbolValue + 2 + (1ull << (tableLog - 2)))
#define FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) (sizeof(unsigned) * FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog))
size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
#define FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) (sizeof(short) * (maxSymbolValue + 1) + (1ULL << maxTableLog) + 8)
#define FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) ((FSE_BUILD_DTABLE_WKSP_SIZE(maxTableLog, maxSymbolValue) + sizeof(unsigned) - 1) / sizeof(unsigned))
FSE_PUBLIC_API size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
/**< Same as FSE_buildDTable(), using an externally allocated `workspace` produced with `FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxSymbolValue)` */
size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
/**< build a fake FSE_DTable, designed to read a flat distribution where each symbol uses nbBits */
size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
/**< build a fake FSE_DTable, designed to always generate the same symbolValue */
size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog);
/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DTABLE_SIZE_U32(maxLog)` */
#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) + (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(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize);
/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)` */
size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2);
/**< Same as FSE_decompress_wksp() but with dynamic BMI2 support. Pass 1 if your CPU supports BMI2 or 0 if it doesn't. */
typedef enum {
FSE_repeat_none, /**< Cannot use the previous table */
@@ -644,6 +669,9 @@ MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
#ifndef FSE_DEFAULT_MEMORY_USAGE
# define FSE_DEFAULT_MEMORY_USAGE 13
#endif
#if (FSE_DEFAULT_MEMORY_USAGE > FSE_MAX_MEMORY_USAGE)
# error "FSE_DEFAULT_MEMORY_USAGE must be <= FSE_MAX_MEMORY_USAGE"
#endif
/*!FSE_MAX_SYMBOL_VALUE :
* Maximum symbol value authorized.
@@ -677,7 +705,7 @@ MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
# error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
#endif
#define FSE_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)
#define FSE_TABLESTEP(tableSize) (((tableSize)>>1) + ((tableSize)>>3) + 3)
#endif /* FSE_STATIC_LINKING_ONLY */
Utilities/cmzstd/lib/common/fse_decompress.c
+140 -23
View File
@@ -1,6 +1,6 @@
/* ******************************************************************
* FSE : Finite State Entropy decoder
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -16,13 +16,14 @@
/* **************************************************************
* Includes
****************************************************************/
#include <stdlib.h> /* malloc, free, qsort */
#include <string.h> /* memcpy, memset */
#include "debug.h" /* assert */
#include "bitstream.h"
#include "compiler.h"
#define FSE_STATIC_LINKING_ONLY
#include "fse.h"
#include "error_private.h"
#define ZSTD_DEPS_NEED_MALLOC
#include "zstd_deps.h"
/* **************************************************************
@@ -59,25 +60,27 @@
FSE_DTable* FSE_createDTable (unsigned tableLog)
{
if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
return (FSE_DTable*)malloc( FSE_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
return (FSE_DTable*)ZSTD_malloc( FSE_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
}
void FSE_freeDTable (FSE_DTable* dt)
{
free(dt);
ZSTD_free(dt);
}
size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
{
void* const tdPtr = dt+1; /* because *dt is unsigned, 32-bits aligned on 32-bits */
FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
U16* symbolNext = (U16*)workSpace;
BYTE* spread = (BYTE*)(symbolNext + maxSymbolValue + 1);
U32 const maxSV1 = maxSymbolValue + 1;
U32 const tableSize = 1 << tableLog;
U32 highThreshold = tableSize-1;
/* Sanity Checks */
if (FSE_BUILD_DTABLE_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(maxSymbolValue_tooLarge);
if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
@@ -95,11 +98,57 @@ size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned
if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
symbolNext[s] = normalizedCounter[s];
} } }
memcpy(dt, &DTableH, sizeof(DTableH));
ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
}
/* Spread symbols */
{ U32 const tableMask = tableSize-1;
if (highThreshold == tableSize - 1) {
size_t const tableMask = tableSize-1;
size_t const step = FSE_TABLESTEP(tableSize);
/* First lay down the symbols in order.
* We use a uint64_t to lay down 8 bytes at a time. This reduces branch
* misses since small blocks generally have small table logs, so nearly
* all symbols have counts <= 8. We ensure we have 8 bytes at the end of
* our buffer to handle the over-write.
*/
{
U64 const add = 0x0101010101010101ull;
size_t pos = 0;
U64 sv = 0;
U32 s;
for (s=0; s<maxSV1; ++s, sv += add) {
int i;
int const n = normalizedCounter[s];
MEM_write64(spread + pos, sv);
for (i = 8; i < n; i += 8) {
MEM_write64(spread + pos + i, sv);
}
pos += n;
}
}
/* Now we spread those positions across the table.
* The benefit of doing it in two stages is that we avoid the the
* variable size inner loop, which caused lots of branch misses.
* Now we can run through all the positions without any branch misses.
* We unroll the loop twice, since that is what emperically worked best.
*/
{
size_t position = 0;
size_t s;
size_t const unroll = 2;
assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
for (s = 0; s < (size_t)tableSize; s += unroll) {
size_t u;
for (u = 0; u < unroll; ++u) {
size_t const uPosition = (position + (u * step)) & tableMask;
tableDecode[uPosition].symbol = spread[s + u];
}
position = (position + (unroll * step)) & tableMask;
}
assert(position == 0);
}
} else {
U32 const tableMask = tableSize-1;
U32 const step = FSE_TABLESTEP(tableSize);
U32 s, position = 0;
for (s=0; s<maxSV1; s++) {
@@ -124,6 +173,11 @@ size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned
return 0;
}
size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
{
return FSE_buildDTable_internal(dt, normalizedCounter, maxSymbolValue, tableLog, workSpace, wkspSize);
}
#ifndef FSE_COMMONDEFS_ONLY
@@ -251,36 +305,99 @@ size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
}
size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog)
size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
{
return FSE_decompress_wksp_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, /* bmi2 */ 0);
}
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;
short counting[FSE_MAX_SYMBOL_VALUE+1];
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 (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
if (FSE_isError(NCountLength)) return NCountLength;
/* if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong); */ /* too small input size; supposed to be already checked in NCountLength, only remaining case : NCountLength==cSrcSize */
if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
ip += NCountLength;
cSrcSize -= NCountLength;
{
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;
}
CHECK_F( FSE_buildDTable (workSpace, counting, maxSymbolValue, tableLog) );
if (FSE_DECOMPRESS_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(tableLog_tooLarge);
workSpace = wksp->dtable + FSE_DTABLE_SIZE_U32(tableLog);
wkspSize -= sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
return FSE_decompress_usingDTable (dst, dstCapacity, ip, cSrcSize, workSpace); /* always return, even if it is an error code */
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
TARGET_ATTRIBUTE("bmi2") static size_t FSE_decompress_wksp_body_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
{
return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 1);
}
#endif
size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2)
{
#if DYNAMIC_BMI2
if (bmi2) {
return FSE_decompress_wksp_body_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
}
#endif
(void)bmi2;
return FSE_decompress_wksp_body_default(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
}
typedef FSE_DTable DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
size_t FSE_decompress(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize)
{
DTable_max_t dt; /* Static analyzer seems unable to understand this table will be properly initialized later */
return FSE_decompress_wksp(dst, dstCapacity, cSrc, cSrcSize, dt, FSE_MAX_TABLELOG);
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog) {
U32 wksp[FSE_BUILD_DTABLE_WKSP_SIZE_U32(FSE_TABLELOG_ABSOLUTE_MAX, FSE_MAX_SYMBOL_VALUE)];
return FSE_buildDTable_wksp(dt, normalizedCounter, maxSymbolValue, tableLog, wksp, sizeof(wksp));
}
size_t FSE_decompress(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize)
{
/* Static analyzer seems unable to understand this table will be properly initialized later */
U32 wksp[FSE_DECOMPRESS_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
return FSE_decompress_wksp(dst, dstCapacity, cSrc, cSrcSize, FSE_MAX_TABLELOG, wksp, sizeof(wksp));
}
#endif
#endif /* FSE_COMMONDEFS_ONLY */
Utilities/cmzstd/lib/common/huf.h
+29 -7
View File
@@ -1,7 +1,7 @@
/* ******************************************************************
* huff0 huffman codec,
* part of Finite State Entropy library
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -20,7 +20,7 @@ extern "C" {
#define HUF_H_298734234
/* *** Dependencies *** */
#include <stddef.h> /* size_t */
#include "zstd_deps.h" /* size_t */
/* *** library symbols visibility *** */
@@ -111,6 +111,8 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
/* *** Dependencies *** */
#include "mem.h" /* U32 */
#define FSE_STATIC_LINKING_ONLY
#include "fse.h"
/* *** Constants *** */
@@ -133,12 +135,16 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size)) /* Macro version, useful for static allocation */
/* static allocation of HUF's Compression Table */
/* this is a private definition, just exposed for allocation and strict aliasing purpose. never EVER access its members directly */
struct HUF_CElt_s {
U16 val;
BYTE nbBits;
}; /* typedef'd to HUF_CElt */
typedef struct HUF_CElt_s HUF_CElt; /* consider it an incomplete type */
#define HUF_CTABLE_SIZE_U32(maxSymbolValue) ((maxSymbolValue)+1) /* Use tables of U32, for proper alignment */
#define HUF_CTABLE_SIZE(maxSymbolValue) (HUF_CTABLE_SIZE_U32(maxSymbolValue) * sizeof(U32))
#define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
U32 name##hb[HUF_CTABLE_SIZE_U32(maxSymbolValue)]; \
void* name##hv = &(name##hb); \
HUF_CElt* name = (HUF_CElt*)(name##hv) /* no final ; */
HUF_CElt name[HUF_CTABLE_SIZE_U32(maxSymbolValue)] /* no final ; */
/* static allocation of HUF's DTable */
typedef U32 HUF_DTable;
@@ -184,9 +190,9 @@ size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
* or to save and regenerate 'CTable' using external methods.
*/
unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
typedef struct HUF_CElt_s HUF_CElt; /* incomplete type */
size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits); /* @return : maxNbBits; CTable and count can overlap. In which case, CTable will overwrite count content */
size_t HUF_writeCTable (void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog);
size_t HUF_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);
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);
@@ -226,6 +232,19 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize);
/*! HUF_readStats_wksp() :
* Same as HUF_readStats() but takes an external workspace which must be
* 4-byte aligned and its size must be >= HUF_READ_STATS_WORKSPACE_SIZE.
* If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
*/
#define HUF_READ_STATS_WORKSPACE_SIZE_U32 FSE_DECOMPRESS_WKSP_SIZE_U32(6, HUF_TABLELOG_MAX-1)
#define HUF_READ_STATS_WORKSPACE_SIZE (HUF_READ_STATS_WORKSPACE_SIZE_U32 * sizeof(unsigned))
size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
const void* src, size_t srcSize,
void* workspace, size_t wkspSize,
int bmi2);
/** HUF_readCTable() :
* Loading a CTable saved with HUF_writeCTable() */
size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);
@@ -260,7 +279,7 @@ U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
* a required workspace size greater than that specified in the following
* macro.
*/
#define HUF_DECOMPRESS_WORKSPACE_SIZE (2 << 10)
#define HUF_DECOMPRESS_WORKSPACE_SIZE ((2 << 10) + (1 << 9))
#define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
#ifndef HUF_FORCE_DECOMPRESS_X2
@@ -332,6 +351,9 @@ size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstS
#endif
size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
#ifndef HUF_FORCE_DECOMPRESS_X2
size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2);
#endif
#endif /* HUF_STATIC_LINKING_ONLY */
Utilities/cmzstd/lib/common/mem.h
+69 -98
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -18,8 +18,10 @@ extern "C" {
/*-****************************************
* Dependencies
******************************************/
#include <stddef.h> /* size_t, ptrdiff_t */
#include <string.h> /* memcpy */
#include <stddef.h> /* size_t, ptrdiff_t */
#include "compiler.h" /* __has_builtin */
#include "debug.h" /* DEBUG_STATIC_ASSERT */
#include "zstd_deps.h" /* ZSTD_memcpy */
/*-****************************************
@@ -39,93 +41,15 @@ extern "C" {
# define MEM_STATIC static /* this version may generate warnings for unused static functions; disable the relevant warning */
#endif
#ifndef __has_builtin
# define __has_builtin(x) 0 /* compat. with non-clang compilers */
#endif
/* code only tested on 32 and 64 bits systems */
#define MEM_STATIC_ASSERT(c) { enum { MEM_static_assert = 1/(int)(!!(c)) }; }
MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
/* detects whether we are being compiled under msan */
#if defined (__has_feature)
# if __has_feature(memory_sanitizer)
# define MEMORY_SANITIZER 1
# endif
#endif
#if defined (MEMORY_SANITIZER)
/* Not all platforms that support msan provide sanitizers/msan_interface.h.
* We therefore declare the functions we need ourselves, rather than trying to
* include the header file... */
#include <stdint.h> /* intptr_t */
/* Make memory region fully initialized (without changing its contents). */
void __msan_unpoison(const volatile void *a, size_t size);
/* Make memory region fully uninitialized (without changing its contents).
This is a legacy interface that does not update origin information. Use
__msan_allocated_memory() instead. */
void __msan_poison(const volatile void *a, size_t size);
/* Returns the offset of the first (at least partially) poisoned byte in the
memory range, or -1 if the whole range is good. */
intptr_t __msan_test_shadow(const volatile void *x, size_t size);
#endif
/* detects whether we are being compiled under asan */
#if defined (__has_feature)
# if __has_feature(address_sanitizer)
# define ADDRESS_SANITIZER 1
# endif
#elif defined(__SANITIZE_ADDRESS__)
# define ADDRESS_SANITIZER 1
#endif
#if defined (ADDRESS_SANITIZER)
/* Not all platforms that support asan provide sanitizers/asan_interface.h.
* We therefore declare the functions we need ourselves, rather than trying to
* include the header file... */
/**
* Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable.
*
* This memory must be previously allocated by your program. Instrumented
* code is forbidden from accessing addresses in this region until it is
* unpoisoned. This function is not guaranteed to poison the entire region -
* it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan
* alignment restrictions.
*
* \note This function is not thread-safe because no two threads can poison or
* unpoison memory in the same memory region simultaneously.
*
* \param addr Start of memory region.
* \param size Size of memory region. */
void __asan_poison_memory_region(void const volatile *addr, size_t size);
/**
* Marks a memory region (<c>[addr, addr+size)</c>) as addressable.
*
* This memory must be previously allocated by your program. Accessing
* addresses in this region is allowed until this region is poisoned again.
* This function could unpoison a super-region of <c>[addr, addr+size)</c> due
* to ASan alignment restrictions.
*
* \note This function is not thread-safe because no two threads can
* poison or unpoison memory in the same memory region simultaneously.
*
* \param addr Start of memory region.
* \param size Size of memory region. */
void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
#endif
/*-**************************************************************
* Basic Types
*****************************************************************/
#if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
# include <stdint.h>
# if defined(_AIX)
# include <inttypes.h>
# else
# include <stdint.h> /* intptr_t */
# endif
typedef uint8_t BYTE;
typedef uint16_t U16;
typedef int16_t S16;
@@ -157,7 +81,53 @@ void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
/*-**************************************************************
* Memory I/O
* Memory I/O API
*****************************************************************/
/*=== Static platform detection ===*/
MEM_STATIC unsigned MEM_32bits(void);
MEM_STATIC unsigned MEM_64bits(void);
MEM_STATIC unsigned MEM_isLittleEndian(void);
/*=== Native unaligned read/write ===*/
MEM_STATIC U16 MEM_read16(const void* memPtr);
MEM_STATIC U32 MEM_read32(const void* memPtr);
MEM_STATIC U64 MEM_read64(const void* memPtr);
MEM_STATIC size_t MEM_readST(const void* memPtr);
MEM_STATIC void MEM_write16(void* memPtr, U16 value);
MEM_STATIC void MEM_write32(void* memPtr, U32 value);
MEM_STATIC void MEM_write64(void* memPtr, U64 value);
/*=== Little endian unaligned read/write ===*/
MEM_STATIC U16 MEM_readLE16(const void* memPtr);
MEM_STATIC U32 MEM_readLE24(const void* memPtr);
MEM_STATIC U32 MEM_readLE32(const void* memPtr);
MEM_STATIC U64 MEM_readLE64(const void* memPtr);
MEM_STATIC size_t MEM_readLEST(const void* memPtr);
MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val);
MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val);
MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32);
MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64);
MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val);
/*=== Big endian unaligned read/write ===*/
MEM_STATIC U32 MEM_readBE32(const void* memPtr);
MEM_STATIC U64 MEM_readBE64(const void* memPtr);
MEM_STATIC size_t MEM_readBEST(const void* memPtr);
MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32);
MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64);
MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val);
/*=== Byteswap ===*/
MEM_STATIC U32 MEM_swap32(U32 in);
MEM_STATIC U64 MEM_swap64(U64 in);
MEM_STATIC size_t MEM_swapST(size_t in);
/*-**************************************************************
* Memory I/O Implementation
*****************************************************************/
/* MEM_FORCE_MEMORY_ACCESS :
* By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
@@ -173,9 +143,7 @@ void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
* Prefer these methods in priority order (0 > 1 > 2)
*/
#ifndef MEM_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */
# if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
# define MEM_FORCE_MEMORY_ACCESS 2
# elif defined(__INTEL_COMPILER) || defined(__GNUC__) || defined(__ICCARM__)
# if defined(__INTEL_COMPILER) || defined(__GNUC__) || defined(__ICCARM__)
# define MEM_FORCE_MEMORY_ACCESS 1
# endif
#endif
@@ -236,37 +204,37 @@ MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign64*)memPtr)->v =
MEM_STATIC U16 MEM_read16(const void* memPtr)
{
U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
U16 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC U32 MEM_read32(const void* memPtr)
{
U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
U32 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC U64 MEM_read64(const void* memPtr)
{
U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
U64 val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC size_t MEM_readST(const void* memPtr)
{
size_t val; memcpy(&val, memPtr, sizeof(val)); return val;
size_t val; ZSTD_memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC void MEM_write16(void* memPtr, U16 value)
{
memcpy(memPtr, &value, sizeof(value));
ZSTD_memcpy(memPtr, &value, sizeof(value));
}
MEM_STATIC void MEM_write32(void* memPtr, U32 value)
{
memcpy(memPtr, &value, sizeof(value));
ZSTD_memcpy(memPtr, &value, sizeof(value));
}
MEM_STATIC void MEM_write64(void* memPtr, U64 value)
{
memcpy(memPtr, &value, sizeof(value));
ZSTD_memcpy(memPtr, &value, sizeof(value));
}
#endif /* MEM_FORCE_MEMORY_ACCESS */
@@ -338,7 +306,7 @@ MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
MEM_STATIC U32 MEM_readLE24(const void* memPtr)
{
return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
return (U32)MEM_readLE16(memPtr) + ((U32)(((const BYTE*)memPtr)[2]) << 16);
}
MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val)
@@ -445,6 +413,9 @@ MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val)
MEM_writeBE64(memPtr, (U64)val);
}
/* code only tested on 32 and 64 bits systems */
MEM_STATIC void MEM_check(void) { DEBUG_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
#if defined (__cplusplus)
}
Utilities/cmzstd/lib/common/pool.c
+23 -17
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -10,9 +10,9 @@
/* ====== Dependencies ======= */
#include <stddef.h> /* size_t */
#include "zstd_deps.h" /* size_t */
#include "debug.h" /* assert */
#include "zstd_internal.h" /* ZSTD_malloc, ZSTD_free */
#include "zstd_internal.h" /* ZSTD_customMalloc, ZSTD_customFree */
#include "pool.h"
/* ====== Compiler specifics ====== */
@@ -105,6 +105,10 @@ static void* POOL_thread(void* opaque) {
assert(0); /* Unreachable */
}
POOL_ctx* ZSTD_createThreadPool(size_t numThreads) {
return POOL_create (numThreads, 0);
}
POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
}
@@ -115,14 +119,14 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
/* Check parameters */
if (!numThreads) { return NULL; }
/* Allocate the context and zero initialize */
ctx = (POOL_ctx*)ZSTD_calloc(sizeof(POOL_ctx), customMem);
ctx = (POOL_ctx*)ZSTD_customCalloc(sizeof(POOL_ctx), customMem);
if (!ctx) { return NULL; }
/* Initialize the job queue.
* It needs one extra space since one space is wasted to differentiate
* empty and full queues.
*/
ctx->queueSize = queueSize + 1;
ctx->queue = (POOL_job*)ZSTD_malloc(ctx->queueSize * sizeof(POOL_job), customMem);
ctx->queue = (POOL_job*)ZSTD_customMalloc(ctx->queueSize * sizeof(POOL_job), customMem);
ctx->queueHead = 0;
ctx->queueTail = 0;
ctx->numThreadsBusy = 0;
@@ -136,7 +140,7 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
}
ctx->shutdown = 0;
/* Allocate space for the thread handles */
ctx->threads = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
ctx->threads = (ZSTD_pthread_t*)ZSTD_customMalloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
ctx->threadCapacity = 0;
ctx->customMem = customMem;
/* Check for errors */
@@ -179,12 +183,14 @@ void POOL_free(POOL_ctx *ctx) {
ZSTD_pthread_mutex_destroy(&ctx->queueMutex);
ZSTD_pthread_cond_destroy(&ctx->queuePushCond);
ZSTD_pthread_cond_destroy(&ctx->queuePopCond);
ZSTD_free(ctx->queue, ctx->customMem);
ZSTD_free(ctx->threads, ctx->customMem);
ZSTD_free(ctx, ctx->customMem);
ZSTD_customFree(ctx->queue, ctx->customMem);
ZSTD_customFree(ctx->threads, ctx->customMem);
ZSTD_customFree(ctx, ctx->customMem);
}
void ZSTD_freeThreadPool (ZSTD_threadPool* pool) {
POOL_free (pool);
}
size_t POOL_sizeof(POOL_ctx *ctx) {
if (ctx==NULL) return 0; /* supports sizeof NULL */
@@ -203,11 +209,11 @@ static int POOL_resize_internal(POOL_ctx* ctx, size_t numThreads)
return 0;
}
/* numThreads > threadCapacity */
{ ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
{ ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_customMalloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
if (!threadPool) return 1;
/* replace existing thread pool */
memcpy(threadPool, ctx->threads, ctx->threadCapacity * sizeof(*threadPool));
ZSTD_free(ctx->threads, ctx->customMem);
ZSTD_memcpy(threadPool, ctx->threads, ctx->threadCapacity * sizeof(*threadPool));
ZSTD_customFree(ctx->threads, ctx->customMem);
ctx->threads = threadPool;
/* Initialize additional threads */
{ size_t threadId;
@@ -301,7 +307,7 @@ int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque)
struct POOL_ctx_s {
int dummy;
};
static POOL_ctx g_ctx;
static POOL_ctx g_poolCtx;
POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
@@ -311,11 +317,11 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customM
(void)numThreads;
(void)queueSize;
(void)customMem;
return &g_ctx;
return &g_poolCtx;
}
void POOL_free(POOL_ctx* ctx) {
assert(!ctx || ctx == &g_ctx);
assert(!ctx || ctx == &g_poolCtx);
(void)ctx;
}
@@ -337,7 +343,7 @@ int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque) {
size_t POOL_sizeof(POOL_ctx* ctx) {
if (ctx==NULL) return 0; /* supports sizeof NULL */
assert(ctx == &g_ctx);
assert(ctx == &g_poolCtx);
return sizeof(*ctx);
}
Utilities/cmzstd/lib/common/pool.h
+2 -2
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -16,7 +16,7 @@ extern "C" {
#endif
#include <stddef.h> /* size_t */
#include "zstd_deps.h"
#define ZSTD_STATIC_LINKING_ONLY /* ZSTD_customMem */
#include "../zstd.h"
Utilities/cmzstd/lib/common/threading.c
+6 -5
View File
@@ -78,11 +78,12 @@ int ZSTD_pthread_join(ZSTD_pthread_t thread, void **value_ptr)
#if defined(ZSTD_MULTITHREAD) && DEBUGLEVEL >= 1 && !defined(_WIN32)
#include <stdlib.h>
#define ZSTD_DEPS_NEED_MALLOC
#include "zstd_deps.h"
int ZSTD_pthread_mutex_init(ZSTD_pthread_mutex_t* mutex, pthread_mutexattr_t const* attr)
{
*mutex = (pthread_mutex_t*)malloc(sizeof(pthread_mutex_t));
*mutex = (pthread_mutex_t*)ZSTD_malloc(sizeof(pthread_mutex_t));
if (!*mutex)
return 1;
return pthread_mutex_init(*mutex, attr);
@@ -94,14 +95,14 @@ int ZSTD_pthread_mutex_destroy(ZSTD_pthread_mutex_t* mutex)
return 0;
{
int const ret = pthread_mutex_destroy(*mutex);
free(*mutex);
ZSTD_free(*mutex);
return ret;
}
}
int ZSTD_pthread_cond_init(ZSTD_pthread_cond_t* cond, pthread_condattr_t const* attr)
{
*cond = (pthread_cond_t*)malloc(sizeof(pthread_cond_t));
*cond = (pthread_cond_t*)ZSTD_malloc(sizeof(pthread_cond_t));
if (!*cond)
return 1;
return pthread_cond_init(*cond, attr);
@@ -113,7 +114,7 @@ int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond)
return 0;
{
int const ret = pthread_cond_destroy(*cond);
free(*cond);
ZSTD_free(*cond);
return ret;
}
}
Utilities/cmzstd/lib/common/xxhash.c
+20 -60
View File
@@ -1,6 +1,6 @@
/*
* xxHash - Fast Hash algorithm
* Copyright (c) 2012-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - xxHash homepage: http://www.xxhash.com
@@ -30,9 +30,7 @@
* Prefer these methods in priority order (0 > 1 > 2)
*/
#ifndef XXH_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */
# if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
# define XXH_FORCE_MEMORY_ACCESS 2
# elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
# if (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
(defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) )) || \
defined(__ICCARM__)
# define XXH_FORCE_MEMORY_ACCESS 1
@@ -77,14 +75,12 @@
* Includes & Memory related functions
***************************************/
/* Modify the local functions below should you wish to use some other memory routines */
/* for malloc(), free() */
#include <stdlib.h>
#include <stddef.h> /* size_t */
static void* XXH_malloc(size_t s) { return malloc(s); }
static void XXH_free (void* p) { free(p); }
/* for memcpy() */
#include <string.h>
static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); }
/* for ZSTD_malloc(), ZSTD_free() */
#define ZSTD_DEPS_NEED_MALLOC
#include "zstd_deps.h" /* size_t, ZSTD_malloc, ZSTD_free, ZSTD_memcpy */
static void* XXH_malloc(size_t s) { return ZSTD_malloc(s); }
static void XXH_free (void* p) { ZSTD_free(p); }
static void* XXH_memcpy(void* dest, const void* src, size_t size) { return ZSTD_memcpy(dest,src,size); }
#ifndef XXH_STATIC_LINKING_ONLY
# define XXH_STATIC_LINKING_ONLY
@@ -95,49 +91,13 @@ static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcp
/* *************************************
* Compiler Specific Options
***************************************/
#if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */
# define INLINE_KEYWORD inline
#else
# define INLINE_KEYWORD
#endif
#if defined(__GNUC__) || defined(__ICCARM__)
# define FORCE_INLINE_ATTR __attribute__((always_inline))
#elif defined(_MSC_VER)
# define FORCE_INLINE_ATTR __forceinline
#else
# define FORCE_INLINE_ATTR
#endif
#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
#ifdef _MSC_VER
# pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
#endif
#include "compiler.h"
/* *************************************
* Basic Types
***************************************/
#ifndef MEM_MODULE
# define MEM_MODULE
# if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
# include <stdint.h>
typedef uint8_t BYTE;
typedef uint16_t U16;
typedef uint32_t U32;
typedef int32_t S32;
typedef uint64_t U64;
# else
typedef unsigned char BYTE;
typedef unsigned short U16;
typedef unsigned int U32;
typedef signed int S32;
typedef unsigned long long U64; /* if your compiler doesn't support unsigned long long, replace by another 64-bit type here. Note that xxhash.h will also need to be updated. */
# endif
#endif
#include "mem.h" /* BYTE, U32, U64, size_t */
#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
@@ -163,14 +123,14 @@ static U64 XXH_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
static U32 XXH_read32(const void* memPtr)
{
U32 val;
memcpy(&val, memPtr, sizeof(val));
ZSTD_memcpy(&val, memPtr, sizeof(val));
return val;
}
static U64 XXH_read64(const void* memPtr)
{
U64 val;
memcpy(&val, memPtr, sizeof(val));
ZSTD_memcpy(&val, memPtr, sizeof(val));
return val;
}
@@ -307,12 +267,12 @@ XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
****************************/
XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dstState, const XXH32_state_t* restrict srcState)
{
memcpy(dstState, srcState, sizeof(*dstState));
ZSTD_memcpy(dstState, srcState, sizeof(*dstState));
}
XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dstState, const XXH64_state_t* restrict srcState)
{
memcpy(dstState, srcState, sizeof(*dstState));
ZSTD_memcpy(dstState, srcState, sizeof(*dstState));
}
@@ -554,12 +514,12 @@ XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed)
{
XXH32_state_t state; /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
memset(&state, 0, sizeof(state)-4); /* do not write into reserved, for future removal */
ZSTD_memset(&state, 0, sizeof(state)-4); /* do not write into reserved, for future removal */
state.v1 = seed + PRIME32_1 + PRIME32_2;
state.v2 = seed + PRIME32_2;
state.v3 = seed + 0;
state.v4 = seed - PRIME32_1;
memcpy(statePtr, &state, sizeof(state));
ZSTD_memcpy(statePtr, &state, sizeof(state));
return XXH_OK;
}
@@ -567,12 +527,12 @@ XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int s
XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed)
{
XXH64_state_t state; /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
memset(&state, 0, sizeof(state)-8); /* do not write into reserved, for future removal */
ZSTD_memset(&state, 0, sizeof(state)-8); /* do not write into reserved, for future removal */
state.v1 = seed + PRIME64_1 + PRIME64_2;
state.v2 = seed + PRIME64_2;
state.v3 = seed + 0;
state.v4 = seed - PRIME64_1;
memcpy(statePtr, &state, sizeof(state));
ZSTD_memcpy(statePtr, &state, sizeof(state));
return XXH_OK;
}
@@ -843,14 +803,14 @@ XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t
{
XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
memcpy(dst, &hash, sizeof(*dst));
ZSTD_memcpy(dst, &hash, sizeof(*dst));
}
XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
{
XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
memcpy(dst, &hash, sizeof(*dst));
ZSTD_memcpy(dst, &hash, sizeof(*dst));
}
XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
Utilities/cmzstd/lib/common/xxhash.h
+2 -2
View File
@@ -1,7 +1,7 @@
/*
* xxHash - Extremely Fast Hash algorithm
* Header File
* Copyright (c) 2012-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - xxHash source repository : https://github.com/Cyan4973/xxHash
@@ -55,7 +55,7 @@ extern "C" {
/* ****************************
* Definitions
******************************/
#include <stddef.h> /* size_t */
#include "zstd_deps.h"
typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
Utilities/cmzstd/lib/common/zstd_common.c
+10 -10
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -13,8 +13,8 @@
/*-*************************************
* Dependencies
***************************************/
#include <stdlib.h> /* malloc, calloc, free */
#include <string.h> /* memset */
#define ZSTD_DEPS_NEED_MALLOC
#include "zstd_deps.h" /* ZSTD_malloc, ZSTD_calloc, ZSTD_free, ZSTD_memset */
#include "error_private.h"
#include "zstd_internal.h"
@@ -53,31 +53,31 @@ const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString
/*=**************************************************************
* Custom allocator
****************************************************************/
void* ZSTD_malloc(size_t size, ZSTD_customMem customMem)
void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem)
{
if (customMem.customAlloc)
return customMem.customAlloc(customMem.opaque, size);
return malloc(size);
return ZSTD_malloc(size);
}
void* ZSTD_calloc(size_t size, ZSTD_customMem customMem)
void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem)
{
if (customMem.customAlloc) {
/* calloc implemented as malloc+memset;
* not as efficient as calloc, but next best guess for custom malloc */
void* const ptr = customMem.customAlloc(customMem.opaque, size);
memset(ptr, 0, size);
ZSTD_memset(ptr, 0, size);
return ptr;
}
return calloc(1, size);
return ZSTD_calloc(1, size);
}
void ZSTD_free(void* ptr, ZSTD_customMem customMem)
void ZSTD_customFree(void* ptr, ZSTD_customMem customMem)
{
if (ptr!=NULL) {
if (customMem.customFree)
customMem.customFree(customMem.opaque, ptr);
else
free(ptr);
ZSTD_free(ptr);
}
}
Utilities/cmzstd/lib/common/zstd_deps.h
+111
View File
@@ -0,0 +1,111 @@
/*
* Copyright (c) Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/* This file provides common libc dependencies that zstd requires.
* The purpose is to allow replacing this file with a custom implementation
* to compile zstd without libc support.
*/
/* Need:
* NULL
* INT_MAX
* UINT_MAX
* ZSTD_memcpy()
* ZSTD_memset()
* ZSTD_memmove()
*/
#ifndef ZSTD_DEPS_COMMON
#define ZSTD_DEPS_COMMON
#include <limits.h>
#include <stddef.h>
#include <string.h>
#if defined(__GNUC__) && __GNUC__ >= 4
# define ZSTD_memcpy(d,s,l) __builtin_memcpy((d),(s),(l))
# define ZSTD_memmove(d,s,l) __builtin_memmove((d),(s),(l))
# define ZSTD_memset(p,v,l) __builtin_memset((p),(v),(l))
#else
# define ZSTD_memcpy(d,s,l) memcpy((d),(s),(l))
# define ZSTD_memmove(d,s,l) memmove((d),(s),(l))
# define ZSTD_memset(p,v,l) memset((p),(v),(l))
#endif
#endif /* ZSTD_DEPS_COMMON */
/* Need:
* ZSTD_malloc()
* ZSTD_free()
* ZSTD_calloc()
*/
#ifdef ZSTD_DEPS_NEED_MALLOC
#ifndef ZSTD_DEPS_MALLOC
#define ZSTD_DEPS_MALLOC
#include <stdlib.h>
#define ZSTD_malloc(s) malloc(s)
#define ZSTD_calloc(n,s) calloc((n), (s))
#define ZSTD_free(p) free((p))
#endif /* ZSTD_DEPS_MALLOC */
#endif /* ZSTD_DEPS_NEED_MALLOC */
/*
* Provides 64-bit math support.
* Need:
* U64 ZSTD_div64(U64 dividend, U32 divisor)
*/
#ifdef ZSTD_DEPS_NEED_MATH64
#ifndef ZSTD_DEPS_MATH64
#define ZSTD_DEPS_MATH64
#define ZSTD_div64(dividend, divisor) ((dividend) / (divisor))
#endif /* ZSTD_DEPS_MATH64 */
#endif /* ZSTD_DEPS_NEED_MATH64 */
/* Need:
* assert()
*/
#ifdef ZSTD_DEPS_NEED_ASSERT
#ifndef ZSTD_DEPS_ASSERT
#define ZSTD_DEPS_ASSERT
#include <assert.h>
#endif /* ZSTD_DEPS_ASSERT */
#endif /* ZSTD_DEPS_NEED_ASSERT */
/* Need:
* ZSTD_DEBUG_PRINT()
*/
#ifdef ZSTD_DEPS_NEED_IO
#ifndef ZSTD_DEPS_IO
#define ZSTD_DEPS_IO
#include <stdio.h>
#define ZSTD_DEBUG_PRINT(...) fprintf(stderr, __VA_ARGS__)
#endif /* ZSTD_DEPS_IO */
#endif /* ZSTD_DEPS_NEED_IO */
/* Only requested when <stdint.h> is known to be present.
* Need:
* intptr_t
*/
#ifdef ZSTD_DEPS_NEED_STDINT
#ifndef ZSTD_DEPS_STDINT
#define ZSTD_DEPS_STDINT
#include <stdint.h>
#endif /* ZSTD_DEPS_STDINT */
#endif /* ZSTD_DEPS_NEED_STDINT */
Utilities/cmzstd/lib/common/zstd_internal.h
+103 -60
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -19,7 +19,7 @@
/*-*************************************
* Dependencies
***************************************/
#ifdef __aarch64__
#if !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON)
#include <arm_neon.h>
#endif
#include "compiler.h"
@@ -36,6 +36,11 @@
# define XXH_STATIC_LINKING_ONLY /* XXH64_state_t */
#endif
#include "xxhash.h" /* XXH_reset, update, digest */
#ifndef ZSTD_NO_TRACE
# include "zstd_trace.h"
#else
# define ZSTD_TRACE 0
#endif
#if defined (__cplusplus)
extern "C" {
@@ -139,7 +144,7 @@ void _force_has_format_string(const char *format, ...) {
#define ZSTD_REP_NUM 3 /* number of repcodes */
#define ZSTD_REP_MOVE (ZSTD_REP_NUM-1)
static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
static UNUSED_ATTR const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
#define KB *(1 <<10)
#define MB *(1 <<20)
@@ -153,13 +158,13 @@ static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
#define BIT0 1
#define ZSTD_WINDOWLOG_ABSOLUTEMIN 10
static const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
static const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
static UNUSED_ATTR const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
static UNUSED_ATTR const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
#define ZSTD_FRAMEIDSIZE 4 /* magic number size */
#define ZSTD_BLOCKHEADERSIZE 3 /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
static const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
static UNUSED_ATTR const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
#define ZSTD_FRAMECHECKSUMSIZE 4
@@ -186,61 +191,75 @@ typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingTy
#define OffFSELog 8
#define MaxFSELog MAX(MAX(MLFSELog, LLFSELog), OffFSELog)
static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
1, 1, 1, 1, 2, 2, 3, 3,
4, 6, 7, 8, 9,10,11,12,
13,14,15,16 };
static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 1, 1, 1,
2, 2, 2, 2, 2, 2, 2, 2,
2, 3, 2, 1, 1, 1, 1, 1,
-1,-1,-1,-1 };
#define ZSTD_MAX_HUF_HEADER_SIZE 128 /* header + <= 127 byte tree description */
/* Each table cannot take more than #symbols * FSELog bits */
#define ZSTD_MAX_FSE_HEADERS_SIZE (((MaxML + 1) * MLFSELog + (MaxLL + 1) * LLFSELog + (MaxOff + 1) * OffFSELog + 7) / 8)
static UNUSED_ATTR const U32 LL_bits[MaxLL+1] = {
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
1, 1, 1, 1, 2, 2, 3, 3,
4, 6, 7, 8, 9,10,11,12,
13,14,15,16
};
static UNUSED_ATTR const S16 LL_defaultNorm[MaxLL+1] = {
4, 3, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 1, 1, 1,
2, 2, 2, 2, 2, 2, 2, 2,
2, 3, 2, 1, 1, 1, 1, 1,
-1,-1,-1,-1
};
#define LL_DEFAULTNORMLOG 6 /* for static allocation */
static const U32 LL_defaultNormLog = LL_DEFAULTNORMLOG;
static UNUSED_ATTR const U32 LL_defaultNormLog = LL_DEFAULTNORMLOG;
static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
1, 1, 1, 1, 2, 2, 3, 3,
4, 4, 5, 7, 8, 9,10,11,
12,13,14,15,16 };
static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2,
2, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1,-1,-1,
-1,-1,-1,-1,-1 };
static UNUSED_ATTR const U32 ML_bits[MaxML+1] = {
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
1, 1, 1, 1, 2, 2, 3, 3,
4, 4, 5, 7, 8, 9,10,11,
12,13,14,15,16
};
static UNUSED_ATTR const S16 ML_defaultNorm[MaxML+1] = {
1, 4, 3, 2, 2, 2, 2, 2,
2, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1,-1,-1,
-1,-1,-1,-1,-1
};
#define ML_DEFAULTNORMLOG 6 /* for static allocation */
static const U32 ML_defaultNormLog = ML_DEFAULTNORMLOG;
static UNUSED_ATTR const U32 ML_defaultNormLog = ML_DEFAULTNORMLOG;
static const S16 OF_defaultNorm[DefaultMaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2,
2, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
-1,-1,-1,-1,-1 };
static UNUSED_ATTR const S16 OF_defaultNorm[DefaultMaxOff+1] = {
1, 1, 1, 1, 1, 1, 2, 2,
2, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
-1,-1,-1,-1,-1
};
#define OF_DEFAULTNORMLOG 5 /* for static allocation */
static const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
static UNUSED_ATTR const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
/*-*******************************************
* Shared functions to include for inlining
*********************************************/
static void ZSTD_copy8(void* dst, const void* src) {
#ifdef __aarch64__
#if !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON)
vst1_u8((uint8_t*)dst, vld1_u8((const uint8_t*)src));
#else
memcpy(dst, src, 8);
ZSTD_memcpy(dst, src, 8);
#endif
}
#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
static void ZSTD_copy16(void* dst, const void* src) {
#ifdef __aarch64__
#if !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON)
vst1q_u8((uint8_t*)dst, vld1q_u8((const uint8_t*)src));
#else
memcpy(dst, src, 16);
ZSTD_memcpy(dst, src, 16);
#endif
}
#define COPY16(d,s) { ZSTD_copy16(d,s); d+=16; s+=16; }
@@ -255,13 +274,13 @@ typedef enum {
} ZSTD_overlap_e;
/*! ZSTD_wildcopy() :
* Custom version of memcpy(), can over read/write up to WILDCOPY_OVERLENGTH bytes (if length==0)
* Custom version of ZSTD_memcpy(), can over read/write up to WILDCOPY_OVERLENGTH bytes (if length==0)
* @param ovtype controls the overlap detection
* - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
* - ZSTD_overlap_src_before_dst: The src and dst may overlap, but they MUST be at least 8 bytes apart.
* The src buffer must be before the dst buffer.
*/
MEM_STATIC FORCE_INLINE_ATTR
MEM_STATIC FORCE_INLINE_ATTR
void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e const ovtype)
{
ptrdiff_t diff = (BYTE*)dst - (const BYTE*)src;
@@ -284,14 +303,16 @@ void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e
* one COPY16() in the first call. Then, do two calls per loop since
* at that point it is more likely to have a high trip count.
*/
#ifndef __aarch64__
#ifdef __aarch64__
do {
COPY16(op, ip);
}
while (op < oend);
#else
COPY16(op, ip);
if (op >= oend) return;
ZSTD_copy16(op, ip);
if (16 >= length) return;
op += 16;
ip += 16;
do {
COPY16(op, ip);
COPY16(op, ip);
@@ -305,7 +326,7 @@ MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src,
{
size_t const length = MIN(dstCapacity, srcSize);
if (length > 0) {
memcpy(dst, src, length);
ZSTD_memcpy(dst, src, length);
}
return length;
}
@@ -320,28 +341,46 @@ MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src,
* In which case, resize it down to free some memory */
#define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128
/* Controls whether the input/output buffer is buffered or stable. */
typedef enum {
ZSTD_bm_buffered = 0, /* Buffer the input/output */
ZSTD_bm_stable = 1 /* ZSTD_inBuffer/ZSTD_outBuffer is stable */
} ZSTD_bufferMode_e;
/*-*******************************************
* Private declarations
*********************************************/
typedef struct seqDef_s {
U32 offset;
U32 offset; /* offset == rawOffset + ZSTD_REP_NUM, or equivalently, offCode + 1 */
U16 litLength;
U16 matchLength;
} 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;
seqDef* sequences; /* ptr to end of sequences */
BYTE* litStart;
BYTE* lit;
BYTE* lit; /* ptr to end of literals */
BYTE* llCode;
BYTE* mlCode;
BYTE* ofCode;
size_t maxNbSeq;
size_t maxNbLit;
U32 longLengthID; /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
U32 longLengthPos;
/* longLengthPos and 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 {
@@ -351,7 +390,7 @@ typedef struct {
/**
* Returns the ZSTD_sequenceLength for the given sequences. It handles the decoding of long sequences
* indicated by longLengthPos and longLengthID, and adds MINMATCH back to matchLength.
* indicated by longLengthPos and longLengthType, and adds MINMATCH back to matchLength.
*/
MEM_STATIC ZSTD_sequenceLength ZSTD_getSequenceLength(seqStore_t const* seqStore, seqDef const* seq)
{
@@ -359,10 +398,10 @@ MEM_STATIC ZSTD_sequenceLength ZSTD_getSequenceLength(seqStore_t const* seqStore
seqLen.litLength = seq->litLength;
seqLen.matchLength = seq->matchLength + MINMATCH;
if (seqStore->longLengthPos == (U32)(seq - seqStore->sequencesStart)) {
if (seqStore->longLengthID == 1) {
if (seqStore->longLengthType == ZSTD_llt_literalLength) {
seqLen.litLength += 0xFFFF;
}
if (seqStore->longLengthID == 2) {
if (seqStore->longLengthType == ZSTD_llt_matchLength) {
seqLen.matchLength += 0xFFFF;
}
}
@@ -384,9 +423,9 @@ const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx); /* compress & dictBu
void ZSTD_seqToCodes(const seqStore_t* seqStorePtr); /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
/* custom memory allocation functions */
void* ZSTD_malloc(size_t size, ZSTD_customMem customMem);
void* ZSTD_calloc(size_t size, ZSTD_customMem customMem);
void ZSTD_free(void* ptr, ZSTD_customMem customMem);
void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem);
void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem);
void ZSTD_customFree(void* ptr, ZSTD_customMem customMem);
MEM_STATIC U32 ZSTD_highbit32(U32 val) /* compress, dictBuilder, decodeCorpus */
@@ -394,8 +433,12 @@ MEM_STATIC U32 ZSTD_highbit32(U32 val) /* compress, dictBuilder, decodeCorpus
assert(val != 0);
{
# if defined(_MSC_VER) /* Visual */
unsigned long r=0;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# if STATIC_BMI2 == 1
return _lzcnt_u32(val)^31;
# else
unsigned long r=0;
return _BitScanReverse(&r, val) ? (unsigned)r : 0;
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 3) /* GCC Intrinsic */
return __builtin_clz (val) ^ 31;
# elif defined(__ICCARM__) /* IAR Intrinsic */
Utilities/cmzstd/lib/common/zstd_trace.h
+154
View File
@@ -0,0 +1,154 @@
/*
* Copyright (c) Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_TRACE_H
#define ZSTD_TRACE_H
#if defined (__cplusplus)
extern "C" {
#endif
#include <stddef.h>
/* weak symbol support */
#if !defined(ZSTD_HAVE_WEAK_SYMBOLS) && defined(__GNUC__) && \
!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 */
Utilities/cmzstd/lib/compress/fse_compress.c
+31 -24
View File
@@ -1,6 +1,6 @@
/* ******************************************************************
* FSE : Finite State Entropy encoder
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -15,8 +15,6 @@
/* **************************************************************
* Includes
****************************************************************/
#include <stdlib.h> /* malloc, free, qsort */
#include <string.h> /* memcpy, memset */
#include "../common/compiler.h"
#include "../common/mem.h" /* U32, U16, etc. */
#include "../common/debug.h" /* assert, DEBUGLOG */
@@ -25,6 +23,9 @@
#define FSE_STATIC_LINKING_ONLY
#include "../common/fse.h"
#include "../common/error_private.h"
#define ZSTD_DEPS_NEED_MALLOC
#define ZSTD_DEPS_NEED_MATH64
#include "../common/zstd_deps.h" /* ZSTD_malloc, ZSTD_free, ZSTD_memcpy, ZSTD_memset */
/* **************************************************************
@@ -74,13 +75,15 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
U32 const step = FSE_TABLESTEP(tableSize);
U32 cumul[FSE_MAX_SYMBOL_VALUE+2];
FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)workSpace;
U32* cumul = (U32*)workSpace;
FSE_FUNCTION_TYPE* tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSymbolValue + 2));
U32 highThreshold = tableSize-1;
if ((size_t)workSpace & 3) return ERROR(GENERIC); /* Must be 4 byte aligned */
if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
/* CTable header */
if (((size_t)1 << tableLog) * sizeof(FSE_FUNCTION_TYPE) > wkspSize) return ERROR(tableLog_tooLarge);
tableU16[-2] = (U16) tableLog;
tableU16[-1] = (U16) maxSymbolValue;
assert(tableLog < 16); /* required for threshold strategy to work */
@@ -89,7 +92,7 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
* http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
#ifdef __clang_analyzer__
memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize); /* useless initialization, just to keep scan-build happy */
ZSTD_memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize); /* useless initialization, just to keep scan-build happy */
#endif
/* symbol start positions */
@@ -168,12 +171,13 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
return 0;
}
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
{
FSE_FUNCTION_TYPE tableSymbol[FSE_MAX_TABLESIZE]; /* memset() is not necessary, even if static analyzer complain about it */
return FSE_buildCTable_wksp(ct, normalizedCounter, maxSymbolValue, tableLog, tableSymbol, sizeof(tableSymbol));
}
#endif
@@ -307,10 +311,10 @@ FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
size_t size;
if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
size = FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
return (FSE_CTable*)malloc(size);
return (FSE_CTable*)ZSTD_malloc(size);
}
void FSE_freeCTable (FSE_CTable* ct) { free(ct); }
void FSE_freeCTable (FSE_CTable* ct) { ZSTD_free(ct); }
/* provides the minimum logSize to safely represent a distribution */
static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
@@ -341,11 +345,10 @@ unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
}
/* Secondary normalization method.
To be used when primary method fails. */
static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue)
static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue, short lowProbCount)
{
short const NOT_YET_ASSIGNED = -2;
U32 s;
@@ -362,7 +365,7 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
continue;
}
if (count[s] <= lowThreshold) {
norm[s] = -1;
norm[s] = lowProbCount;
distributed++;
total -= count[s];
continue;
@@ -414,7 +417,7 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
{ U64 const vStepLog = 62 - tableLog;
U64 const mid = (1ULL << (vStepLog-1)) - 1;
U64 const rStep = ((((U64)1<<vStepLog) * ToDistribute) + mid) / total; /* scale on remaining */
U64 const rStep = ZSTD_div64((((U64)1<<vStepLog) * ToDistribute) + mid, (U32)total); /* scale on remaining */
U64 tmpTotal = mid;
for (s=0; s<=maxSymbolValue; s++) {
if (norm[s]==NOT_YET_ASSIGNED) {
@@ -431,10 +434,9 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
return 0;
}
size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
const unsigned* count, size_t total,
unsigned maxSymbolValue)
unsigned maxSymbolValue, unsigned useLowProbCount)
{
/* Sanity checks */
if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
@@ -443,8 +445,9 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC); /* Too small tableLog, compression potentially impossible */
{ static U32 const rtbTable[] = { 0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
short const lowProbCount = useLowProbCount ? -1 : 1;
U64 const scale = 62 - tableLog;
U64 const step = ((U64)1<<62) / total; /* <== here, one division ! */
U64 const step = ZSTD_div64((U64)1<<62, (U32)total); /* <== here, one division ! */
U64 const vStep = 1ULL<<(scale-20);
int stillToDistribute = 1<<tableLog;
unsigned s;
@@ -456,7 +459,7 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
if (count[s] == total) return 0; /* rle special case */
if (count[s] == 0) { normalizedCounter[s]=0; continue; }
if (count[s] <= lowThreshold) {
normalizedCounter[s] = -1;
normalizedCounter[s] = lowProbCount;
stillToDistribute--;
} else {
short proba = (short)((count[s]*step) >> scale);
@@ -470,7 +473,7 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
} }
if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
/* corner case, need another normalization method */
size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue);
size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue, lowProbCount);
if (FSE_isError(errorCode)) return errorCode;
}
else normalizedCounter[largest] += (short)stillToDistribute;
@@ -625,6 +628,7 @@ size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
/* FSE_compress_wksp() :
* Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
* `wkspSize` size must be `(1<<tableLog)`.
@@ -643,7 +647,7 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
size_t const scratchBufferSize = wkspSize - (CTableSize * sizeof(FSE_CTable));
/* init conditions */
if (wkspSize < FSE_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
if (wkspSize < FSE_COMPRESS_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
if (srcSize <= 1) return 0; /* Not compressible */
if (!maxSymbolValue) maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
@@ -656,7 +660,7 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
}
tableLog = FSE_optimalTableLog(tableLog, srcSize, maxSymbolValue);
CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue) );
CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue, /* useLowProbCount */ srcSize >= 2048) );
/* Write table description header */
{ CHECK_V_F(nc_err, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
@@ -678,13 +682,16 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
typedef struct {
FSE_CTable CTable_max[FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
BYTE scratchBuffer[1 << FSE_MAX_TABLELOG];
union {
U32 hist_wksp[HIST_WKSP_SIZE_U32];
BYTE scratchBuffer[1 << FSE_MAX_TABLELOG];
} workspace;
} fseWkspMax_t;
size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
{
fseWkspMax_t scratchBuffer;
DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)); /* compilation failures here means scratchBuffer is not large enough */
DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_COMPRESS_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)); /* compilation failures here means scratchBuffer is not large enough */
if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
}
@@ -693,6 +700,6 @@ size_t FSE_compress (void* dst, size_t dstCapacity, const void* src, size_t srcS
{
return FSE_compress2(dst, dstCapacity, src, srcSize, FSE_MAX_SYMBOL_VALUE, FSE_DEFAULT_TABLELOG);
}
#endif
#endif /* FSE_COMMONDEFS_ONLY */
Utilities/cmzstd/lib/compress/hist.c
+27 -29
View File
@@ -1,7 +1,7 @@
/* ******************************************************************
* hist : Histogram functions
* part of Finite State Entropy project
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -34,7 +34,7 @@ unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
unsigned maxSymbolValue = *maxSymbolValuePtr;
unsigned largestCount=0;
memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
ZSTD_memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
while (ip<end) {
@@ -60,9 +60,9 @@ typedef enum { trustInput, checkMaxSymbolValue } HIST_checkInput_e;
* this design makes better use of OoO cpus,
* and is noticeably faster when some values are heavily repeated.
* But it needs some additional workspace for intermediate tables.
* `workSpace` size must be a table of size >= HIST_WKSP_SIZE_U32.
* `workSpace` must be a U32 table of size >= HIST_WKSP_SIZE_U32.
* @return : largest histogram frequency,
* or an error code (notably when histogram would be larger than *maxSymbolValuePtr). */
* or an error code (notably when histogram's alphabet is larger than *maxSymbolValuePtr) */
static size_t HIST_count_parallel_wksp(
unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize,
@@ -71,22 +71,21 @@ static size_t HIST_count_parallel_wksp(
{
const BYTE* ip = (const BYTE*)source;
const BYTE* const iend = ip+sourceSize;
unsigned maxSymbolValue = *maxSymbolValuePtr;
size_t const countSize = (*maxSymbolValuePtr + 1) * sizeof(*count);
unsigned max=0;
U32* const Counting1 = workSpace;
U32* const Counting2 = Counting1 + 256;
U32* const Counting3 = Counting2 + 256;
U32* const Counting4 = Counting3 + 256;
memset(workSpace, 0, 4*256*sizeof(unsigned));
/* safety checks */
assert(*maxSymbolValuePtr <= 255);
if (!sourceSize) {
memset(count, 0, maxSymbolValue + 1);
ZSTD_memset(count, 0, countSize);
*maxSymbolValuePtr = 0;
return 0;
}
if (!maxSymbolValue) maxSymbolValue = 255; /* 0 == default */
ZSTD_memset(workSpace, 0, 4*256*sizeof(unsigned));
/* by stripes of 16 bytes */
{ U32 cached = MEM_read32(ip); ip += 4;
@@ -118,21 +117,18 @@ static size_t HIST_count_parallel_wksp(
/* finish last symbols */
while (ip<iend) Counting1[*ip++]++;
if (check) { /* verify stats will fit into destination table */
U32 s; for (s=255; s>maxSymbolValue; s--) {
Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
if (Counting1[s]) return ERROR(maxSymbolValue_tooSmall);
} }
{ U32 s;
if (maxSymbolValue > 255) maxSymbolValue = 255;
for (s=0; s<=maxSymbolValue; s++) {
count[s] = Counting1[s] + Counting2[s] + Counting3[s] + Counting4[s];
if (count[s] > max) max = count[s];
for (s=0; s<256; s++) {
Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
if (Counting1[s] > max) max = Counting1[s];
} }
while (!count[maxSymbolValue]) maxSymbolValue--;
*maxSymbolValuePtr = maxSymbolValue;
{ unsigned maxSymbolValue = 255;
while (!Counting1[maxSymbolValue]) maxSymbolValue--;
if (check && maxSymbolValue > *maxSymbolValuePtr) return ERROR(maxSymbolValue_tooSmall);
*maxSymbolValuePtr = maxSymbolValue;
ZSTD_memmove(count, Counting1, countSize); /* in case count & Counting1 are overlapping */
}
return (size_t)max;
}
@@ -152,14 +148,6 @@ size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, trustInput, (U32*)workSpace);
}
/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize)
{
unsigned tmpCounters[HIST_WKSP_SIZE_U32];
return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters, sizeof(tmpCounters));
}
/* HIST_count_wksp() :
* Same as HIST_count(), but using an externally provided scratch buffer.
* `workSpace` size must be table of >= HIST_WKSP_SIZE_U32 unsigned */
@@ -175,9 +163,19 @@ size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace, workSpaceSize);
}
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize)
{
unsigned tmpCounters[HIST_WKSP_SIZE_U32];
return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters, sizeof(tmpCounters));
}
size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize)
{
unsigned tmpCounters[HIST_WKSP_SIZE_U32];
return HIST_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters, sizeof(tmpCounters));
}
#endif
Utilities/cmzstd/lib/compress/hist.h
+2 -2
View File
@@ -1,7 +1,7 @@
/* ******************************************************************
* hist : Histogram functions
* part of Finite State Entropy project
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -14,7 +14,7 @@
****************************************************************** */
/* --- dependencies --- */
#include <stddef.h> /* size_t */
#include "../common/zstd_deps.h" /* size_t */
/* --- simple histogram functions --- */
Utilities/cmzstd/lib/compress/huf_compress.c
+264 -125
View File
@@ -1,6 +1,6 @@
/* ******************************************************************
* Huffman encoder, part of New Generation Entropy library
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -23,8 +23,7 @@
/* **************************************************************
* Includes
****************************************************************/
#include <string.h> /* memcpy, memset */
#include <stdio.h> /* printf (debug) */
#include "../common/zstd_deps.h" /* ZSTD_memcpy, ZSTD_memset */
#include "../common/compiler.h"
#include "../common/bitstream.h"
#include "hist.h"
@@ -60,7 +59,15 @@ unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
* Note : all elements within weightTable are supposed to be <= HUF_TABLELOG_MAX.
*/
#define MAX_FSE_TABLELOG_FOR_HUFF_HEADER 6
static size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weightTable, size_t wtSize)
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;
@@ -68,33 +75,30 @@ static size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weight
unsigned maxSymbolValue = HUF_TABLELOG_MAX;
U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
HUF_CompressWeightsWksp* wksp = (HUF_CompressWeightsWksp*)workspace;
FSE_CTable CTable[FSE_CTABLE_SIZE_U32(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, HUF_TABLELOG_MAX)];
BYTE scratchBuffer[1<<MAX_FSE_TABLELOG_FOR_HUFF_HEADER];
unsigned count[HUF_TABLELOG_MAX+1];
S16 norm[HUF_TABLELOG_MAX+1];
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(count, &maxSymbolValue, weightTable, wtSize); /* never fails */
{ 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(norm, tableLog, count, 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), norm, maxSymbolValue, tableLog) );
{ CHECK_V_F(hSize, FSE_writeNCount(op, (size_t)(oend-op), wksp->norm, maxSymbolValue, tableLog) );
op += hSize;
}
/* Compress */
CHECK_F( FSE_buildCTable_wksp(CTable, norm, maxSymbolValue, tableLog, scratchBuffer, sizeof(scratchBuffer)) );
{ CHECK_V_F(cSize, FSE_compress_usingCTable(op, (size_t)(oend - op), weightTable, wtSize, CTable) );
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;
}
@@ -103,34 +107,33 @@ static size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weight
}
struct HUF_CElt_s {
U16 val;
BYTE nbBits;
}; /* typedef'd to HUF_CElt within "huf.h" */
/*! HUF_writeCTable() :
`CTable` : Huffman tree to save, using huf representation.
@return : size of saved CTable */
size_t HUF_writeCTable (void* dst, size_t maxDstSize,
const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog)
{
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)
{
BYTE* op = (BYTE*)dst;
U32 n;
HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)workspace;
/* check conditions */
/* check conditions */
if (workspaceSize < sizeof(HUF_WriteCTableWksp)) return ERROR(GENERIC);
if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
/* convert to weight */
bitsToWeight[0] = 0;
wksp->bitsToWeight[0] = 0;
for (n=1; n<huffLog+1; n++)
bitsToWeight[n] = (BYTE)(huffLog + 1 - n);
wksp->bitsToWeight[n] = (BYTE)(huffLog + 1 - n);
for (n=0; n<maxSymbolValue; n++)
huffWeight[n] = bitsToWeight[CTable[n].nbBits];
wksp->huffWeight[n] = wksp->bitsToWeight[CTable[n].nbBits];
/* attempt weights compression by FSE */
{ CHECK_V_F(hSize, HUF_compressWeights(op+1, maxDstSize-1, huffWeight, maxSymbolValue) );
{ 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;
@@ -140,12 +143,22 @@ size_t HUF_writeCTable (void* dst, size_t maxDstSize,
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));
huffWeight[maxSymbolValue] = 0; /* to be sure it doesn't cause msan issue in final combination */
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)((huffWeight[n] << 4) + huffWeight[n+1]);
op[(n/2)+1] = (BYTE)((wksp->huffWeight[n] << 4) + wksp->huffWeight[n+1]);
return ((maxSymbolValue+1)/2) + 1;
}
/*! HUF_writeCTable() :
`CTable` : Huffman tree to save, using huf representation.
@return : size of saved CTable */
size_t HUF_writeCTable (void* dst, size_t maxDstSize,
const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog)
{
HUF_WriteCTableWksp wksp;
return HUF_writeCTable_wksp(dst, maxDstSize, CTable, maxSymbolValue, huffLog, &wksp, sizeof(wksp));
}
size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* hasZeroWeights)
{
@@ -156,6 +169,7 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
/* get symbol weights */
CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
*hasZeroWeights = (rankVal[0] > 0);
/* check result */
if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
@@ -164,16 +178,14 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
/* Prepare base value per rank */
{ U32 n, nextRankStart = 0;
for (n=1; n<=tableLog; n++) {
U32 current = nextRankStart;
U32 curr = nextRankStart;
nextRankStart += (rankVal[n] << (n-1));
rankVal[n] = current;
rankVal[n] = curr;
} }
/* fill nbBits */
*hasZeroWeights = 0;
{ U32 n; for (n=0; n<nbSymbols; n++) {
const U32 w = huffWeight[n];
*hasZeroWeights |= (w == 0);
CTable[n].nbBits = (BYTE)(tableLog + 1 - w) & -(w != 0);
} }
@@ -212,32 +224,63 @@ typedef struct nodeElt_s {
BYTE nbBits;
} nodeElt;
/**
* HUF_setMaxHeight():
* Enforces maxNbBits on the Huffman tree described in huffNode.
*
* It sets all nodes with nbBits > maxNbBits to be maxNbBits. Then it adjusts
* the tree to so that it is a valid canonical Huffman tree.
*
* @pre The sum of the ranks of each symbol == 2^largestBits,
* where largestBits == huffNode[lastNonNull].nbBits.
* @post The sum of the ranks of each symbol == 2^largestBits,
* where largestBits is the return value <= maxNbBits.
*
* @param huffNode The Huffman tree modified in place to enforce maxNbBits.
* @param lastNonNull The symbol with the lowest count in the Huffman tree.
* @param maxNbBits The maximum allowed number of bits, which the Huffman tree
* may not respect. After this function the Huffman tree will
* respect maxNbBits.
* @return The maximum number of bits of the Huffman tree after adjustment,
* necessarily no more than maxNbBits.
*/
static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
{
const U32 largestBits = huffNode[lastNonNull].nbBits;
if (largestBits <= maxNbBits) return largestBits; /* early exit : no elt > maxNbBits */
/* early exit : no elt > maxNbBits, so the tree is already valid. */
if (largestBits <= maxNbBits) return largestBits;
/* there are several too large elements (at least >= 2) */
{ int totalCost = 0;
const U32 baseCost = 1 << (largestBits - maxNbBits);
int n = (int)lastNonNull;
/* Adjust any ranks > maxNbBits to maxNbBits.
* Compute totalCost, which is how far the sum of the ranks is
* we are over 2^largestBits after adjust the offending ranks.
*/
while (huffNode[n].nbBits > maxNbBits) {
totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
huffNode[n].nbBits = (BYTE)maxNbBits;
n --;
} /* n stops at huffNode[n].nbBits <= maxNbBits */
while (huffNode[n].nbBits == maxNbBits) n--; /* n end at index of smallest symbol using < maxNbBits */
n--;
}
/* n stops at huffNode[n].nbBits <= maxNbBits */
assert(huffNode[n].nbBits <= maxNbBits);
/* n end at index of smallest symbol using < maxNbBits */
while (huffNode[n].nbBits == maxNbBits) --n;
/* renorm totalCost */
totalCost >>= (largestBits - maxNbBits); /* note : totalCost is necessarily a multiple of baseCost */
/* renorm totalCost from 2^largestBits to 2^maxNbBits
* note : totalCost is necessarily a multiple of baseCost */
assert((totalCost & (baseCost - 1)) == 0);
totalCost >>= (largestBits - maxNbBits);
assert(totalCost > 0);
/* repay normalized cost */
{ U32 const noSymbol = 0xF0F0F0F0;
U32 rankLast[HUF_TABLELOG_MAX+2];
/* Get pos of last (smallest) symbol per rank */
memset(rankLast, 0xF0, sizeof(rankLast));
/* Get pos of last (smallest = lowest cum. count) symbol per rank */
ZSTD_memset(rankLast, 0xF0, sizeof(rankLast));
{ U32 currentNbBits = maxNbBits;
int pos;
for (pos=n ; pos >= 0; pos--) {
@@ -247,34 +290,65 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
} }
while (totalCost > 0) {
/* Try to reduce the next power of 2 above totalCost because we
* gain back half the rank.
*/
U32 nBitsToDecrease = BIT_highbit32((U32)totalCost) + 1;
for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
U32 const highPos = rankLast[nBitsToDecrease];
U32 const lowPos = rankLast[nBitsToDecrease-1];
if (highPos == noSymbol) continue;
/* Decrease highPos if no symbols of lowPos or if it is
* not cheaper to remove 2 lowPos than highPos.
*/
if (lowPos == noSymbol) break;
{ U32 const highTotal = huffNode[highPos].count;
U32 const lowTotal = 2 * huffNode[lowPos].count;
if (highTotal <= lowTotal) break;
} }
/* only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) */
assert(rankLast[nBitsToDecrease] != noSymbol || nBitsToDecrease == 1);
/* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))
nBitsToDecrease ++;
nBitsToDecrease++;
assert(rankLast[nBitsToDecrease] != noSymbol);
/* Increase the number of bits to gain back half the rank cost. */
totalCost -= 1 << (nBitsToDecrease-1);
huffNode[rankLast[nBitsToDecrease]].nbBits++;
/* Fix up the new rank.
* If the new rank was empty, this symbol is now its smallest.
* Otherwise, this symbol will be the largest in the new rank so no adjustment.
*/
if (rankLast[nBitsToDecrease-1] == noSymbol)
rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease]; /* this rank is no longer empty */
huffNode[rankLast[nBitsToDecrease]].nbBits ++;
rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease];
/* Fix up the old rank.
* If the symbol was at position 0, meaning it was the highest weight symbol in the tree,
* it must be the only symbol in its rank, so the old rank now has no symbols.
* Otherwise, since the Huffman nodes are sorted by count, the previous position is now
* the smallest node in the rank. If the previous position belongs to a different rank,
* then the rank is now empty.
*/
if (rankLast[nBitsToDecrease] == 0) /* special case, reached largest symbol */
rankLast[nBitsToDecrease] = noSymbol;
else {
rankLast[nBitsToDecrease]--;
if (huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease)
rankLast[nBitsToDecrease] = noSymbol; /* this rank is now empty */
} } /* while (totalCost > 0) */
}
} /* while (totalCost > 0) */
/* If we've removed too much weight, then we have to add it back.
* To avoid overshooting again, we only adjust the smallest rank.
* We take the largest nodes from the lowest rank 0 and move them
* to rank 1. There's guaranteed to be enough rank 0 symbols because
* TODO.
*/
while (totalCost < 0) { /* Sometimes, cost correction overshoot */
if (rankLast[1] == noSymbol) { /* special case : no rank 1 symbol (using maxNbBits-1); let's create one from largest rank 0 (using maxNbBits) */
/* special case : no rank 1 symbol (using maxNbBits-1);
* let's create one from largest rank 0 (using maxNbBits).
*/
if (rankLast[1] == noSymbol) {
while (huffNode[n].nbBits == maxNbBits) n--;
huffNode[n+1].nbBits--;
assert(n >= 0);
@@ -285,14 +359,16 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
huffNode[ rankLast[1] + 1 ].nbBits--;
rankLast[1]++;
totalCost ++;
} } } /* there are several too large elements (at least >= 2) */
}
} /* repay normalized cost */
} /* there are several too large elements (at least >= 2) */
return maxNbBits;
}
typedef struct {
U32 base;
U32 current;
U32 curr;
} rankPos;
typedef nodeElt huffNodeTable[HUF_CTABLE_WORKSPACE_SIZE_U32];
@@ -304,21 +380,45 @@ typedef struct {
rankPos rankPosition[RANK_POSITION_TABLE_SIZE];
} HUF_buildCTable_wksp_tables;
/**
* HUF_sort():
* Sorts the symbols [0, maxSymbolValue] by count[symbol] in decreasing order.
*
* @param[out] huffNode Sorted symbols by decreasing count. Only members `.count` and `.byte` are filled.
* Must have (maxSymbolValue + 1) entries.
* @param[in] count Histogram of the symbols.
* @param[in] maxSymbolValue Maximum symbol value.
* @param rankPosition This is a scratch workspace. Must have RANK_POSITION_TABLE_SIZE entries.
*/
static void HUF_sort(nodeElt* huffNode, const unsigned* count, U32 maxSymbolValue, rankPos* rankPosition)
{
U32 n;
int n;
int const maxSymbolValue1 = (int)maxSymbolValue + 1;
memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
for (n=0; n<=maxSymbolValue; n++) {
U32 r = BIT_highbit32(count[n] + 1);
rankPosition[r].base ++;
/* Compute base and set curr to base.
* For symbol s let lowerRank = BIT_highbit32(count[n]+1) and rank = lowerRank + 1.
* Then 2^lowerRank <= count[n]+1 <= 2^rank.
* We attribute each symbol to lowerRank's base value, because we want to know where
* each rank begins in the output, so for rank R we want to count ranks R+1 and above.
*/
ZSTD_memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
for (n = 0; n < maxSymbolValue1; ++n) {
U32 lowerRank = BIT_highbit32(count[n] + 1);
rankPosition[lowerRank].base++;
}
for (n=30; n>0; n--) rankPosition[n-1].base += rankPosition[n].base;
for (n=0; n<32; n++) rankPosition[n].current = rankPosition[n].base;
for (n=0; n<=maxSymbolValue; n++) {
assert(rankPosition[RANK_POSITION_TABLE_SIZE - 1].base == 0);
for (n = RANK_POSITION_TABLE_SIZE - 1; n > 0; --n) {
rankPosition[n-1].base += rankPosition[n].base;
rankPosition[n-1].curr = rankPosition[n-1].base;
}
/* Sort */
for (n = 0; n < maxSymbolValue1; ++n) {
U32 const c = count[n];
U32 const r = BIT_highbit32(c+1) + 1;
U32 pos = rankPosition[r].current++;
U32 pos = rankPosition[r].curr++;
/* Insert into the correct position in the rank.
* We have at most 256 symbols, so this insertion should be fine.
*/
while ((pos > rankPosition[r].base) && (c > huffNode[pos-1].count)) {
huffNode[pos] = huffNode[pos-1];
pos--;
@@ -335,28 +435,20 @@ static void HUF_sort(nodeElt* huffNode, const unsigned* count, U32 maxSymbolValu
*/
#define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
size_t HUF_buildCTable_wksp (HUF_CElt* tree, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
/* HUF_buildTree():
* Takes the huffNode array sorted by HUF_sort() and builds an unlimited-depth Huffman tree.
*
* @param huffNode The array sorted by HUF_sort(). Builds the Huffman tree in this array.
* @param maxSymbolValue The maximum symbol value.
* @return The smallest node in the Huffman tree (by count).
*/
static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
{
HUF_buildCTable_wksp_tables* const wksp_tables = (HUF_buildCTable_wksp_tables*)workSpace;
nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
nodeElt* const huffNode = huffNode0+1;
nodeElt* const huffNode0 = huffNode - 1;
int nonNullRank;
int lowS, lowN;
int nodeNb = STARTNODE;
int n, nodeRoot;
/* safety checks */
if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC); /* must be aligned on 4-bytes boundaries */
if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
return ERROR(workSpace_tooSmall);
if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
return ERROR(maxSymbolValue_tooLarge);
memset(huffNode0, 0, sizeof(huffNodeTable));
/* sort, decreasing order */
HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
/* init for parents */
nonNullRank = (int)maxSymbolValue;
while(huffNode[nonNullRank].count == 0) nonNullRank--;
@@ -383,40 +475,70 @@ size_t HUF_buildCTable_wksp (HUF_CElt* tree, const unsigned* count, U32 maxSymbo
for (n=0; n<=nonNullRank; n++)
huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
/* enforce maxTableLog */
maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
/* fill result into tree (val, nbBits) */
{ U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
int const alphabetSize = (int)(maxSymbolValue + 1);
if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC); /* check fit into table */
for (n=0; n<=nonNullRank; n++)
nbPerRank[huffNode[n].nbBits]++;
/* determine stating value per rank */
{ U16 min = 0;
for (n=(int)maxNbBits; n>0; n--) {
valPerRank[n] = min; /* get starting value within each rank */
min += nbPerRank[n];
min >>= 1;
} }
for (n=0; n<alphabetSize; n++)
tree[huffNode[n].byte].nbBits = huffNode[n].nbBits; /* push nbBits per symbol, symbol order */
for (n=0; n<alphabetSize; n++)
tree[n].val = valPerRank[tree[n].nbBits]++; /* assign value within rank, symbol order */
}
return maxNbBits;
return nonNullRank;
}
/** HUF_buildCTable() :
* @return : maxNbBits
* Note : count is used before tree is written, so they can safely overlap
/**
* 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.
*/
size_t HUF_buildCTable (HUF_CElt* tree, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits)
static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, int nonNullRank, U32 maxSymbolValue, U32 maxNbBits)
{
HUF_buildCTable_wksp_tables workspace;
return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, &workspace, sizeof(workspace));
/* fill result into ctable (val, nbBits) */
int n;
U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
int const alphabetSize = (int)(maxSymbolValue + 1);
for (n=0; n<=nonNullRank; n++)
nbPerRank[huffNode[n].nbBits]++;
/* determine starting value per rank */
{ U16 min = 0;
for (n=(int)maxNbBits; n>0; n--) {
valPerRank[n] = min; /* get starting value within each rank */
min += nbPerRank[n];
min >>= 1;
} }
for (n=0; n<alphabetSize; n++)
CTable[huffNode[n].byte].nbBits = huffNode[n].nbBits; /* push nbBits per symbol, symbol order */
for (n=0; n<alphabetSize; n++)
CTable[n].val = valPerRank[CTable[n].nbBits]++; /* assign value within rank, symbol order */
}
size_t HUF_buildCTable_wksp (HUF_CElt* tree, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
{
HUF_buildCTable_wksp_tables* const wksp_tables = (HUF_buildCTable_wksp_tables*)workSpace;
nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
nodeElt* const huffNode = huffNode0+1;
int nonNullRank;
/* safety checks */
if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC); /* must be aligned on 4-bytes boundaries */
if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
return ERROR(workSpace_tooSmall);
if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
return ERROR(maxSymbolValue_tooLarge);
ZSTD_memset(huffNode0, 0, sizeof(huffNodeTable));
/* sort, decreasing order */
HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
/* build tree */
nonNullRank = HUF_buildTree(huffNode, maxSymbolValue);
/* enforce maxTableLog */
maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC); /* check fit into table */
HUF_buildCTableFromTree(tree, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
return maxNbBits;
}
size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
@@ -629,29 +751,33 @@ static size_t HUF_compressCTable_internal(
typedef struct {
unsigned count[HUF_SYMBOLVALUE_MAX + 1];
HUF_CElt CTable[HUF_SYMBOLVALUE_MAX + 1];
HUF_buildCTable_wksp_tables buildCTable_wksp;
union {
HUF_buildCTable_wksp_tables buildCTable_wksp;
HUF_WriteCTableWksp writeCTable_wksp;
} wksps;
} HUF_compress_tables_t;
/* HUF_compress_internal() :
* `workSpace` must a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
* `workSpace_align4` must be aligned on 4-bytes boundaries,
* and occupies the same space as a table of HUF_WORKSPACE_SIZE_U32 unsigned */
static size_t
HUF_compress_internal (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog,
HUF_nbStreams_e nbStreams,
void* workSpace, size_t wkspSize,
void* workSpace_align4, size_t wkspSize,
HUF_CElt* oldHufTable, HUF_repeat* repeat, int preferRepeat,
const int bmi2)
{
HUF_compress_tables_t* const table = (HUF_compress_tables_t*)workSpace;
HUF_compress_tables_t* const table = (HUF_compress_tables_t*)workSpace_align4;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstSize;
BYTE* op = ostart;
HUF_STATIC_ASSERT(sizeof(*table) <= HUF_WORKSPACE_SIZE);
assert(((size_t)workSpace_align4 & 3) == 0); /* must be aligned on 4-bytes boundaries */
/* checks & inits */
if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC); /* must be aligned on 4-bytes boundaries */
if (wkspSize < HUF_WORKSPACE_SIZE) return ERROR(workSpace_tooSmall);
if (!srcSize) return 0; /* Uncompressed */
if (!dstSize) return 0; /* cannot fit anything within dst budget */
@@ -669,7 +795,7 @@ HUF_compress_internal (void* dst, size_t dstSize,
}
/* Scan input and build symbol stats */
{ CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, workSpace, wkspSize) );
{ CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, workSpace_align4, wkspSize) );
if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; } /* single symbol, rle */
if (largest <= (srcSize >> 7)+4) return 0; /* heuristic : probably not compressible enough */
}
@@ -691,16 +817,17 @@ HUF_compress_internal (void* dst, size_t dstSize,
huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
{ size_t const maxBits = HUF_buildCTable_wksp(table->CTable, table->count,
maxSymbolValue, huffLog,
&table->buildCTable_wksp, sizeof(table->buildCTable_wksp));
&table->wksps.buildCTable_wksp, sizeof(table->wksps.buildCTable_wksp));
CHECK_F(maxBits);
huffLog = (U32)maxBits;
/* Zero unused symbols in CTable, so we can check it for validity */
memset(table->CTable + (maxSymbolValue + 1), 0,
ZSTD_memset(table->CTable + (maxSymbolValue + 1), 0,
sizeof(table->CTable) - ((maxSymbolValue + 1) * sizeof(HUF_CElt)));
}
/* Write table description header */
{ CHECK_V_F(hSize, HUF_writeCTable (op, dstSize, table->CTable, maxSymbolValue, huffLog) );
{ 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);
@@ -716,7 +843,7 @@ HUF_compress_internal (void* dst, size_t dstSize,
op += hSize;
if (repeat) { *repeat = HUF_repeat_none; }
if (oldHufTable)
memcpy(oldHufTable, table->CTable, sizeof(table->CTable)); /* Save new table */
ZSTD_memcpy(oldHufTable, table->CTable, sizeof(table->CTable)); /* Save new table */
}
return HUF_compressCTable_internal(ostart, op, oend,
src, srcSize,
@@ -747,14 +874,6 @@ size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
repeat, preferRepeat, bmi2);
}
size_t HUF_compress1X (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog)
{
unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
return HUF_compress1X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
}
/* HUF_compress4X_repeat():
* compress input using 4 streams.
* provide workspace to generate compression tables */
@@ -784,6 +903,25 @@ size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
hufTable, repeat, preferRepeat, bmi2);
}
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
/** HUF_buildCTable() :
* @return : maxNbBits
* Note : count is used before tree is written, so they can safely overlap
*/
size_t HUF_buildCTable (HUF_CElt* tree, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits)
{
HUF_buildCTable_wksp_tables workspace;
return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, &workspace, sizeof(workspace));
}
size_t HUF_compress1X (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog)
{
unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
return HUF_compress1X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
}
size_t HUF_compress2 (void* dst, size_t dstSize,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned huffLog)
@@ -796,3 +934,4 @@ size_t HUF_compress (void* dst, size_t maxDstSize, const void* src, size_t srcSi
{
return HUF_compress2(dst, maxDstSize, src, srcSize, 255, HUF_TABLELOG_DEFAULT);
}
#endif
Utilities/cmzstd/lib/compress/zstd_compress.c
+2827 -712
View File
File diff suppressed because it is too large Load Diff
Utilities/cmzstd/lib/compress/zstd_compress_internal.h
+304 -62
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -28,7 +28,6 @@
extern "C" {
#endif
/*-*************************************
* Constants
***************************************/
@@ -64,7 +63,7 @@ typedef struct {
} ZSTD_localDict;
typedef struct {
U32 CTable[HUF_CTABLE_SIZE_U32(255)];
HUF_CElt CTable[HUF_CTABLE_SIZE_U32(255)];
HUF_repeat repeatMode;
} ZSTD_hufCTables_t;
@@ -82,11 +81,75 @@ typedef struct {
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 {
U32 off;
U32 len;
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(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 code (offset + ZSTD_REP_MOVE) for the match */
U32 len; /* Raw length of match */
} ZSTD_match_t;
typedef struct {
U32 offset; /* Offset of sequence */
U32 litLength; /* Length of literals prior to match */
U32 matchLength; /* Raw length of match */
} rawSeq;
typedef struct {
rawSeq* seq; /* The start of the sequences */
size_t pos; /* The index in seq where reading stopped. pos <= size. */
size_t posInSequence; /* The position within the sequence at seq[pos] where reading
stopped. posInSequence <= seq[pos].litLength + seq[pos].matchLength */
size_t size; /* The number of sequences. <= capacity. */
size_t capacity; /* The capacity starting from `seq` pointer */
} rawSeqStore_t;
UNUSED_ATTR static const rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0, 0};
typedef struct {
int price;
U32 off;
@@ -125,14 +188,21 @@ typedef struct {
} 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 */
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;
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.
@@ -144,12 +214,24 @@ struct ZSTD_matchState_t {
*/
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.*/
U16* 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 */
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;
};
typedef struct {
@@ -163,13 +245,22 @@ typedef struct {
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 */
U64 hashPower; /* Used to compute the rolling hash.
* Depends on ldmParams.minMatchLength */
size_t splitIndices[LDM_BATCH_SIZE];
ldmMatchCandidate_t matchCandidates[LDM_BATCH_SIZE];
} ldmState_t;
typedef struct {
@@ -181,19 +272,6 @@ typedef struct {
U32 windowLog; /* Window log for the LDM */
} ldmParams_t;
typedef struct {
U32 offset;
U32 litLength;
U32 matchLength;
} rawSeq;
typedef struct {
rawSeq* seq; /* The start of the sequences */
size_t pos; /* The position where reading stopped. <= size. */
size_t size; /* The number of sequences. <= capacity. */
size_t capacity; /* The capacity starting from `seq` pointer */
} rawSeqStore_t;
typedef struct {
int collectSequences;
ZSTD_Sequence* seqStart;
@@ -228,17 +306,52 @@ struct ZSTD_CCtx_params_s {
/* Long distance matching parameters */
ldmParams_t ldmParams;
/* Dedicated dict search algorithm trigger */
int enableDedicatedDictSearch;
/* Input/output buffer modes */
ZSTD_bufferMode_e inBufferMode;
ZSTD_bufferMode_e outBufferMode;
/* Sequence compression API */
ZSTD_sequenceFormat_e blockDelimiters;
int validateSequences;
/* Block splitting */
int splitBlocks;
/* Param for deciding whether to use row-based matchfinder */
ZSTD_useRowMatchFinderMode_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;
}; /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
#define COMPRESS_SEQUENCES_WORKSPACE_SIZE (sizeof(unsigned) * (MaxSeq + 2))
#define ENTROPY_WORKSPACE_SIZE (HUF_WORKSPACE_SIZE + COMPRESS_SEQUENCES_WORKSPACE_SIZE)
/**
* Indicates whether this compression proceeds directly from user-provided
* source buffer to user-provided destination buffer (ZSTDb_not_buffered), or
* whether the context needs to buffer the input/output (ZSTDb_buffered).
*/
typedef enum {
ZSTDb_not_buffered,
ZSTDb_buffered
} ZSTD_buffered_policy_e;
struct ZSTD_CCtx_s {
ZSTD_compressionStage_e stage;
int cParamsChanged; /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
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;
@@ -247,6 +360,7 @@ struct ZSTD_CCtx_s {
unsigned long long producedCSize;
XXH64_state_t xxhState;
ZSTD_customMem customMem;
ZSTD_threadPool* pool;
size_t staticSize;
SeqCollector seqCollector;
int isFirstBlock;
@@ -258,7 +372,10 @@ struct ZSTD_CCtx_s {
size_t maxNbLdmSequences;
rawSeqStore_t externSeqStore; /* Mutable reference to external sequences */
ZSTD_blockState_t blockState;
U32* entropyWorkspace; /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
U32* entropyWorkspace; /* entropy workspace of ENTROPY_WORKSPACE_SIZE bytes */
/* Wether we are streaming or not */
ZSTD_buffered_policy_e bufferedPolicy;
/* streaming */
char* inBuff;
@@ -273,6 +390,10 @@ struct ZSTD_CCtx_s {
ZSTD_cStreamStage streamStage;
U32 frameEnded;
/* Stable in/out buffer verification */
ZSTD_inBuffer expectedInBuffer;
size_t expectedOutBufferSize;
/* Dictionary */
ZSTD_localDict localDict;
const ZSTD_CDict* cdict;
@@ -282,17 +403,46 @@ struct ZSTD_CCtx_s {
#ifdef ZSTD_MULTITHREAD
ZSTDMT_CCtx* mtctx;
#endif
/* Tracing */
#if ZSTD_TRACE
ZSTD_TraceCtx traceCtx;
#endif
};
typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
typedef enum { ZSTD_noDict = 0, ZSTD_extDict = 1, ZSTD_dictMatchState = 2 } ZSTD_dictMode_e;
typedef enum {
ZSTD_noDict = 0,
ZSTD_extDict = 1,
ZSTD_dictMatchState = 2,
ZSTD_dedicatedDictSearch = 3
} ZSTD_dictMode_e;
typedef enum {
ZSTD_cpm_noAttachDict = 0, /* Compression with ZSTD_noDict or ZSTD_extDict.
* In this mode we use both the srcSize and the dictSize
* when selecting and adjusting parameters.
*/
ZSTD_cpm_attachDict = 1, /* Compression with ZSTD_dictMatchState or ZSTD_dedicatedDictSearch.
* In this mode we only take the srcSize into account when selecting
* and adjusting parameters.
*/
ZSTD_cpm_createCDict = 2, /* Creating a CDict.
* In this mode we take both the source size and the dictionary size
* into account when selecting and adjusting the parameters.
*/
ZSTD_cpm_unknown = 3, /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
* We don't know what these parameters are for. We default to the legacy
* behavior of taking both the source size and the dict size into account
* when selecting and adjusting parameters.
*/
} ZSTD_cParamMode_e;
typedef size_t (*ZSTD_blockCompressor) (
ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_dictMode_e dictMode);
ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_useRowMatchFinderMode_e rowMatchfinderMode, ZSTD_dictMode_e dictMode);
MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
@@ -345,7 +495,7 @@ MEM_STATIC repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 con
newReps.rep[1] = rep[0];
newReps.rep[0] = currentOffset;
} else { /* repCode == 0 */
memcpy(&newReps, rep, sizeof(newReps));
ZSTD_memcpy(&newReps, rep, sizeof(newReps));
}
}
return newReps;
@@ -372,7 +522,7 @@ MEM_STATIC size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const voi
RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
dstSize_tooSmall, "dst buf too small for uncompressed block");
MEM_writeLE24(dst, cBlockHeader24);
memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
ZSTD_memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
return ZSTD_blockHeaderSize + srcSize;
}
@@ -469,8 +619,8 @@ void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const BYTE* litera
/* literal Length */
if (litLength>0xFFFF) {
assert(seqStorePtr->longLengthID == 0); /* there can only be a single long length */
seqStorePtr->longLengthID = 1;
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;
@@ -480,8 +630,8 @@ void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const BYTE* litera
/* match Length */
if (mlBase>0xFFFF) {
assert(seqStorePtr->longLengthID == 0); /* there can only be a single long length */
seqStorePtr->longLengthID = 2;
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].matchLength = (U16)mlBase;
@@ -498,8 +648,12 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
if (MEM_isLittleEndian()) {
if (MEM_64bits()) {
# if defined(_MSC_VER) && defined(_WIN64)
unsigned long r = 0;
return _BitScanForward64( &r, (U64)val ) ? (unsigned)(r >> 3) : 0;
# if STATIC_BMI2
return _tzcnt_u64(val) >> 3;
# else
unsigned long r = 0;
return _BitScanForward64( &r, (U64)val ) ? (unsigned)(r >> 3) : 0;
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 4)
return (__builtin_ctzll((U64)val) >> 3);
# else
@@ -530,8 +684,12 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
} else { /* Big Endian CPU */
if (MEM_64bits()) {
# if defined(_MSC_VER) && defined(_WIN64)
unsigned long r = 0;
return _BitScanReverse64( &r, val ) ? (unsigned)(r >> 3) : 0;
# if STATIC_BMI2
return _lzcnt_u64(val) >> 3;
# else
unsigned long r = 0;
return _BitScanReverse64(&r, (U64)val) ? (unsigned)(r >> 3) : 0;
# endif
# elif defined(__GNUC__) && (__GNUC__ >= 4)
return (__builtin_clzll(val) >> 3);
# else
@@ -626,7 +784,8 @@ static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
MEM_STATIC size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
MEM_STATIC FORCE_INLINE_ATTR
size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
{
switch(mls)
{
@@ -723,6 +882,13 @@ MEM_STATIC void ZSTD_window_clear(ZSTD_window_t* window)
window->dictLimit = end;
}
MEM_STATIC U32 ZSTD_window_isEmpty(ZSTD_window_t const window)
{
return window.dictLimit == 1 &&
window.lowLimit == 1 &&
(window.nextSrc - window.base) == 1;
}
/**
* ZSTD_window_hasExtDict():
* Returns non-zero if the window has a non-empty extDict.
@@ -742,20 +908,74 @@ MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
return ZSTD_window_hasExtDict(ms->window) ?
ZSTD_extDict :
ms->dictMatchState != NULL ?
ZSTD_dictMatchState :
(ms->dictMatchState->dedicatedDictSearch ? ZSTD_dedicatedDictSearch : ZSTD_dictMatchState) :
ZSTD_noDict;
}
/* 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);
/* 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 current = (U32)((BYTE const*)srcEnd - window.base);
return current > ZSTD_CURRENT_MAX;
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;
}
/**
@@ -766,7 +986,6 @@ MEM_STATIC U32 ZSTD_window_needOverflowCorrection(ZSTD_window_t const window,
*
* The least significant cycleLog bits of the indices must remain the same,
* which may be 0. Every index up to maxDist in the past must be valid.
* NOTE: (maxDist & cycleMask) must be zero.
*/
MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
U32 maxDist, void const* src)
@@ -790,17 +1009,25 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
* 3. (cctx->lowLimit + 1<<windowLog) < 1<<32:
* windowLog <= 31 ==> 3<<29 + 1<<windowLog < 7<<29 < 1<<32.
*/
U32 const cycleMask = (1U << cycleLog) - 1;
U32 const current = (U32)((BYTE const*)src - window->base);
U32 const currentCycle0 = current & cycleMask;
U32 const cycleSize = 1u << cycleLog;
U32 const cycleMask = cycleSize - 1;
U32 const curr = (U32)((BYTE const*)src - window->base);
U32 const currentCycle0 = curr & cycleMask;
/* Exclude zero so that newCurrent - maxDist >= 1. */
U32 const currentCycle1 = currentCycle0 == 0 ? (1U << cycleLog) : currentCycle0;
U32 const newCurrent = currentCycle1 + maxDist;
U32 const correction = current - newCurrent;
assert((maxDist & cycleMask) == 0);
assert(current > newCurrent);
/* Loose bound, should be around 1<<29 (see above) */
assert(correction > 1<<28);
U32 const currentCycle1 = currentCycle0 == 0 ? cycleSize : currentCycle0;
U32 const newCurrent = currentCycle1 + 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;
@@ -816,6 +1043,8 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
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;
@@ -919,12 +1148,13 @@ ZSTD_checkDictValidity(const ZSTD_window_t* window,
}
MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
memset(window, 0, sizeof(*window));
ZSTD_memset(window, 0, sizeof(*window));
window->base = (BYTE const*)"";
window->dictBase = (BYTE const*)"";
window->dictLimit = 1; /* start from 1, so that 1st position is valid */
window->lowLimit = 1; /* it ensures first and later CCtx usages compress the same */
window->nextSrc = window->base + 1; /* see issue #1241 */
window->nbOverflowCorrections = 0;
}
/**
@@ -935,7 +1165,8 @@ MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
* Returns non-zero if the segment is contiguous.
*/
MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
void const* src, size_t srcSize)
void const* src, size_t srcSize,
int forceNonContiguous)
{
BYTE const* const ip = (BYTE const*)src;
U32 contiguous = 1;
@@ -945,7 +1176,7 @@ MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
assert(window->base != NULL);
assert(window->dictBase != NULL);
/* Check if blocks follow each other */
if (src != window->nextSrc) {
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);
@@ -973,12 +1204,16 @@ MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
/**
* Returns the lowest allowed match index. It may either be in the ext-dict or the prefix.
*/
MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 current, unsigned windowLog)
MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
{
U32 const maxDistance = 1U << windowLog;
U32 const lowestValid = ms->window.lowLimit;
U32 const withinWindow = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
U32 const isDictionary = (ms->loadedDictEnd != 0);
/* When using a dictionary the entire dictionary is valid if a single byte of the dictionary
* is within the window. We invalidate the dictionary (and set loadedDictEnd to 0) when it isn't
* valid for the entire block. So this check is sufficient to find the lowest valid match index.
*/
U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
return matchLowest;
}
@@ -986,12 +1221,15 @@ MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 current
/**
* Returns the lowest allowed match index in the prefix.
*/
MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 current, unsigned windowLog)
MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
{
U32 const maxDistance = 1U << windowLog;
U32 const lowestValid = ms->window.dictLimit;
U32 const withinWindow = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
U32 const isDictionary = (ms->loadedDictEnd != 0);
/* When computing the lowest prefix index we need to take the dictionary into account to handle
* the edge case where the dictionary and the source are contiguous in memory.
*/
U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
return matchLowest;
}
@@ -1045,7 +1283,6 @@ MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
* assumptions : magic number supposed already checked
* and dictSize >= 8 */
size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
short* offcodeNCount, unsigned* offcodeMaxValue,
const void* const dict, size_t dictSize);
void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
@@ -1061,7 +1298,7 @@ void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
* Note: srcSizeHint == 0 means 0!
*/
ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize);
const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
/*! ZSTD_initCStream_internal() :
* Private use only. Init streaming operation.
@@ -1122,4 +1359,9 @@ size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSe
* 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);
#endif /* ZSTD_COMPRESS_H */
Utilities/cmzstd/lib/compress/zstd_compress_literals.c
+8 -8
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -15,7 +15,7 @@
size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
{
BYTE* const ostart = (BYTE* const)dst;
BYTE* const ostart = (BYTE*)dst;
U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
RETURN_ERROR_IF(srcSize + flSize > dstCapacity, dstSize_tooSmall, "");
@@ -35,14 +35,14 @@ size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src,
assert(0);
}
memcpy(ostart + flSize, src, srcSize);
ZSTD_memcpy(ostart + flSize, src, srcSize);
DEBUGLOG(5, "Raw literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
return srcSize + flSize;
}
size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
{
BYTE* const ostart = (BYTE* const)dst;
BYTE* const ostart = (BYTE*)dst;
U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
(void)dstCapacity; /* dstCapacity already guaranteed to be >=4, hence large enough */
@@ -86,7 +86,7 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
disableLiteralCompression, (U32)srcSize);
/* Prepare nextEntropy assuming reusing the existing table */
memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
if (disableLiteralCompression)
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
@@ -117,12 +117,12 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
}
}
if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
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) {
memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
}
Utilities/cmzstd/lib/compress/zstd_compress_literals.h
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/compress/zstd_compress_sequences.c
+29 -7
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -50,6 +50,19 @@ static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
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
* comprssibility.
*/
return nbSeq >= 2048;
}
/**
* Returns the cost in bytes of encoding the normalized count header.
* Returns an error if any of the helper functions return an error.
@@ -60,7 +73,7 @@ static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
BYTE wksp[FSE_NCOUNTBOUND];
S16 norm[MaxSeq + 1];
const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max), "");
FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max, ZSTD_useLowProbCount(nbSeq)), "");
return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
}
@@ -72,6 +85,8 @@ static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t
{
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)
@@ -219,6 +234,11 @@ ZSTD_selectEncodingType(
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,
@@ -239,13 +259,13 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
*op = codeTable[0];
return 1;
case set_repeat:
memcpy(nextCTable, prevCTable, prevCTableSize);
ZSTD_memcpy(nextCTable, prevCTable, prevCTableSize);
return 0;
case set_basic:
FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize), ""); /* note : could be pre-calculated */
return 0;
case set_compressed: {
S16 norm[MaxSeq + 1];
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) {
@@ -253,10 +273,12 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
nbSeq_1--;
}
assert(nbSeq_1 > 1);
FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max), "");
{ size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog); /* overflow protected */
assert(entropyWorkspaceSize >= sizeof(ZSTD_BuildCTableWksp));
(void)entropyWorkspaceSize;
FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "");
{ size_t const NCountSize = FSE_writeNCount(op, oend - op, wksp->norm, max, tableLog); /* overflow protected */
FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, entropyWorkspace, entropyWorkspaceSize), "");
FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "");
return NCountSize;
}
}
Utilities/cmzstd/lib/compress/zstd_compress_sequences.h
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/compress/zstd_compress_superblock.c
+22 -295
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -15,288 +15,10 @@
#include "../common/zstd_internal.h" /* ZSTD_getSequenceLength */
#include "hist.h" /* HIST_countFast_wksp */
#include "zstd_compress_internal.h"
#include "zstd_compress_internal.h" /* ZSTD_[huf|fse|entropy]CTablesMetadata_t */
#include "zstd_compress_sequences.h"
#include "zstd_compress_literals.h"
/*-*************************************
* Superblock entropy buffer structs
***************************************/
/** 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_buildSuperBlockEntropy_literal() */
typedef struct {
symbolEncodingType_e hType;
BYTE hufDesBuffer[500]; /* TODO give name to this value */
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_buildSuperBlockEntropy_sequences() */
typedef struct {
symbolEncodingType_e llType;
symbolEncodingType_e ofType;
symbolEncodingType_e mlType;
BYTE fseTablesBuffer[500]; /* TODO give name to this value */
size_t fseTablesSize;
size_t lastCountSize; /* This is to account for bug in 1.3.4. More detail in ZSTD_compressSubBlock_sequences() */
} ZSTD_fseCTablesMetadata_t;
typedef struct {
ZSTD_hufCTablesMetadata_t hufMetadata;
ZSTD_fseCTablesMetadata_t fseMetadata;
} ZSTD_entropyCTablesMetadata_t;
/** ZSTD_buildSuperBlockEntropy_literal() :
* Builds entropy for the super-block literals.
* Stores literals block type (raw, rle, compressed, repeat) and
* huffman description table to hufMetadata.
* @return : size of huffman description table or error code */
static size_t ZSTD_buildSuperBlockEntropy_literal(void* const src, size_t srcSize,
const ZSTD_hufCTables_t* prevHuf,
ZSTD_hufCTables_t* nextHuf,
ZSTD_hufCTablesMetadata_t* hufMetadata,
const int disableLiteralsCompression,
void* workspace, size_t wkspSize)
{
BYTE* const wkspStart = (BYTE*)workspace;
BYTE* const wkspEnd = wkspStart + wkspSize;
BYTE* const countWkspStart = wkspStart;
unsigned* const countWksp = (unsigned*)workspace;
const size_t countWkspSize = (HUF_SYMBOLVALUE_MAX + 1) * sizeof(unsigned);
BYTE* const nodeWksp = countWkspStart + countWkspSize;
const size_t nodeWkspSize = wkspEnd-nodeWksp;
unsigned maxSymbolValue = 255;
unsigned huffLog = HUF_TABLELOG_DEFAULT;
HUF_repeat repeat = prevHuf->repeatMode;
DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_literal (srcSize=%zu)", srcSize);
/* Prepare nextEntropy assuming reusing the existing table */
memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
if (disableLiteralsCompression) {
DEBUGLOG(5, "set_basic - disabled");
hufMetadata->hType = set_basic;
return 0;
}
/* small ? don't even attempt compression (speed opt) */
# define COMPRESS_LITERALS_SIZE_MIN 63
{ size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
if (srcSize <= minLitSize) {
DEBUGLOG(5, "set_basic - too small");
hufMetadata->hType = set_basic;
return 0;
}
}
/* Scan input and build symbol stats */
{ size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)src, srcSize, workspace, wkspSize);
FORWARD_IF_ERROR(largest, "HIST_count_wksp failed");
if (largest == srcSize) {
DEBUGLOG(5, "set_rle");
hufMetadata->hType = set_rle;
return 0;
}
if (largest <= (srcSize >> 7)+4) {
DEBUGLOG(5, "set_basic - no gain");
hufMetadata->hType = set_basic;
return 0;
}
}
/* Validate the previous Huffman table */
if (repeat == HUF_repeat_check && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
repeat = HUF_repeat_none;
}
/* Build Huffman Tree */
memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
{ size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
maxSymbolValue, huffLog,
nodeWksp, nodeWkspSize);
FORWARD_IF_ERROR(maxBits, "HUF_buildCTable_wksp");
huffLog = (U32)maxBits;
{ /* Build and write the CTable */
size_t const newCSize = HUF_estimateCompressedSize(
(HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
size_t const hSize = HUF_writeCTable(
hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
(HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog);
/* Check against repeating the previous CTable */
if (repeat != HUF_repeat_none) {
size_t const oldCSize = HUF_estimateCompressedSize(
(HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
DEBUGLOG(5, "set_repeat - smaller");
memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
hufMetadata->hType = set_repeat;
return 0;
}
}
if (newCSize + hSize >= srcSize) {
DEBUGLOG(5, "set_basic - no gains");
memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
hufMetadata->hType = set_basic;
return 0;
}
DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
hufMetadata->hType = set_compressed;
nextHuf->repeatMode = HUF_repeat_check;
return hSize;
}
}
}
/** ZSTD_buildSuperBlockEntropy_sequences() :
* Builds entropy for the super-block sequences.
* Stores symbol compression modes and fse table to fseMetadata.
* @return : size of fse tables or error code */
static size_t ZSTD_buildSuperBlockEntropy_sequences(seqStore_t* seqStorePtr,
const ZSTD_fseCTables_t* prevEntropy,
ZSTD_fseCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
ZSTD_fseCTablesMetadata_t* fseMetadata,
void* workspace, size_t wkspSize)
{
BYTE* const wkspStart = (BYTE*)workspace;
BYTE* const wkspEnd = wkspStart + wkspSize;
BYTE* const countWkspStart = wkspStart;
unsigned* const countWksp = (unsigned*)workspace;
const size_t countWkspSize = (MaxSeq + 1) * sizeof(unsigned);
BYTE* const cTableWksp = countWkspStart + countWkspSize;
const size_t cTableWkspSize = wkspEnd-cTableWksp;
ZSTD_strategy const strategy = cctxParams->cParams.strategy;
FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
const BYTE* const ofCodeTable = seqStorePtr->ofCode;
const BYTE* const llCodeTable = seqStorePtr->llCode;
const BYTE* const mlCodeTable = seqStorePtr->mlCode;
size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
BYTE* const ostart = fseMetadata->fseTablesBuffer;
BYTE* const oend = ostart + sizeof(fseMetadata->fseTablesBuffer);
BYTE* op = ostart;
assert(cTableWkspSize >= (1 << MaxFSELog) * sizeof(FSE_FUNCTION_TYPE));
DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_sequences (nbSeq=%zu)", nbSeq);
memset(workspace, 0, wkspSize);
fseMetadata->lastCountSize = 0;
/* convert length/distances into codes */
ZSTD_seqToCodes(seqStorePtr);
/* build CTable for Literal Lengths */
{ U32 LLtype;
unsigned max = MaxLL;
size_t const mostFrequent = HIST_countFast_wksp(countWksp, &max, llCodeTable, nbSeq, workspace, wkspSize); /* can't fail */
DEBUGLOG(5, "Building LL table");
nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode,
countWksp, max, mostFrequent, nbSeq,
LLFSELog, prevEntropy->litlengthCTable,
LL_defaultNorm, LL_defaultNormLog,
ZSTD_defaultAllowed, strategy);
assert(set_basic < set_compressed && set_rle < set_compressed);
assert(!(LLtype < set_compressed && nextEntropy->litlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
{ size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
countWksp, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable),
cTableWksp, cTableWkspSize);
FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for LitLens failed");
if (LLtype == set_compressed)
fseMetadata->lastCountSize = countSize;
op += countSize;
fseMetadata->llType = (symbolEncodingType_e) LLtype;
} }
/* build CTable for Offsets */
{ U32 Offtype;
unsigned max = MaxOff;
size_t const mostFrequent = HIST_countFast_wksp(countWksp, &max, ofCodeTable, nbSeq, workspace, wkspSize); /* can't fail */
/* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
DEBUGLOG(5, "Building OF table");
nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode,
countWksp, max, mostFrequent, nbSeq,
OffFSELog, prevEntropy->offcodeCTable,
OF_defaultNorm, OF_defaultNormLog,
defaultPolicy, strategy);
assert(!(Offtype < set_compressed && nextEntropy->offcode_repeatMode != FSE_repeat_none)); /* We don't copy tables */
{ size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
countWksp, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable),
cTableWksp, cTableWkspSize);
FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for Offsets failed");
if (Offtype == set_compressed)
fseMetadata->lastCountSize = countSize;
op += countSize;
fseMetadata->ofType = (symbolEncodingType_e) Offtype;
} }
/* build CTable for MatchLengths */
{ U32 MLtype;
unsigned max = MaxML;
size_t const mostFrequent = HIST_countFast_wksp(countWksp, &max, mlCodeTable, nbSeq, workspace, wkspSize); /* can't fail */
DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op));
nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode,
countWksp, max, mostFrequent, nbSeq,
MLFSELog, prevEntropy->matchlengthCTable,
ML_defaultNorm, ML_defaultNormLog,
ZSTD_defaultAllowed, strategy);
assert(!(MLtype < set_compressed && nextEntropy->matchlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
{ size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
countWksp, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable),
cTableWksp, cTableWkspSize);
FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for MatchLengths failed");
if (MLtype == set_compressed)
fseMetadata->lastCountSize = countSize;
op += countSize;
fseMetadata->mlType = (symbolEncodingType_e) MLtype;
} }
assert((size_t) (op-ostart) <= sizeof(fseMetadata->fseTablesBuffer));
return op-ostart;
}
/** ZSTD_buildSuperBlockEntropy() :
* Builds entropy for the super-block.
* @return : 0 on success or error code */
static size_t
ZSTD_buildSuperBlockEntropy(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)
{
size_t const litSize = seqStorePtr->lit - seqStorePtr->litStart;
DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy");
entropyMetadata->hufMetadata.hufDesSize =
ZSTD_buildSuperBlockEntropy_literal(seqStorePtr->litStart, litSize,
&prevEntropy->huf, &nextEntropy->huf,
&entropyMetadata->hufMetadata,
ZSTD_disableLiteralsCompression(cctxParams),
workspace, wkspSize);
FORWARD_IF_ERROR(entropyMetadata->hufMetadata.hufDesSize, "ZSTD_buildSuperBlockEntropy_literal failed");
entropyMetadata->fseMetadata.fseTablesSize =
ZSTD_buildSuperBlockEntropy_sequences(seqStorePtr,
&prevEntropy->fse, &nextEntropy->fse,
cctxParams,
&entropyMetadata->fseMetadata,
workspace, wkspSize);
FORWARD_IF_ERROR(entropyMetadata->fseMetadata.fseTablesSize, "ZSTD_buildSuperBlockEntropy_sequences failed");
return 0;
}
/** ZSTD_compressSubBlock_literal() :
* Compresses literals section for a sub-block.
* When we have to write the Huffman table we will sometimes choose a header
@@ -304,7 +26,7 @@ ZSTD_buildSuperBlockEntropy(seqStore_t* seqStorePtr,
* 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 entropyWrriten=1 when we succeeded
* 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.
@@ -348,7 +70,7 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);
if (writeEntropy && hufMetadata->hType == set_compressed) {
memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
ZSTD_memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
op += hufMetadata->hufDesSize;
cLitSize += hufMetadata->hufDesSize;
DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
@@ -474,7 +196,7 @@ static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables
const U32 MLtype = fseMetadata->mlType;
DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
*seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
op += fseMetadata->fseTablesSize;
} else {
const U32 repeat = set_repeat;
@@ -603,7 +325,7 @@ static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
const BYTE* codeTable, unsigned maxCode,
size_t nbSeq, const FSE_CTable* fseCTable,
const U32* additionalBits,
short const* defaultNorm, U32 defaultNormLog,
short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
void* workspace, size_t wkspSize)
{
unsigned* const countWksp = (unsigned*)workspace;
@@ -615,7 +337,11 @@ static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize); /* can't fail */
if (type == set_basic) {
cSymbolTypeSizeEstimateInBits = ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max);
/* We selected this encoding type, so it must be valid. */
assert(max <= defaultMax);
cSymbolTypeSizeEstimateInBits = max <= defaultMax
? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
: ERROR(GENERIC);
} else if (type == set_rle) {
cSymbolTypeSizeEstimateInBits = 0;
} else if (type == set_compressed || type == set_repeat) {
@@ -639,19 +365,20 @@ static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
void* workspace, size_t wkspSize,
int writeEntropy)
{
size_t sequencesSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
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,
OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
nbSeq, fseTables->litlengthCTable, LL_bits,
LL_defaultNorm, LL_defaultNormLog,
LL_defaultNorm, LL_defaultNormLog, MaxLL,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
nbSeq, fseTables->matchlengthCTable, ML_bits,
ML_defaultNorm, ML_defaultNormLog,
ML_defaultNorm, ML_defaultNormLog, MaxML,
workspace, wkspSize);
if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
return cSeqSizeEstimate + sequencesSectionHeaderSize;
@@ -790,7 +517,7 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
} while (!lastSequence);
if (writeLitEntropy) {
DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
}
if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
/* If we haven't written our entropy tables, then we've violated our contract and
@@ -809,11 +536,11 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
if (sp < send) {
seqDef const* seq;
repcodes_t rep;
memcpy(&rep, prevCBlock->rep, sizeof(rep));
ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
for (seq = sstart; seq < sp; ++seq) {
rep = ZSTD_updateRep(rep.rep, seq->offset - 1, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
}
memcpy(nextCBlock->rep, &rep, sizeof(rep));
ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
}
}
DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
@@ -826,12 +553,12 @@ size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
unsigned lastBlock) {
ZSTD_entropyCTablesMetadata_t entropyMetadata;
FORWARD_IF_ERROR(ZSTD_buildSuperBlockEntropy(&zc->seqStore,
FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
&zc->blockState.prevCBlock->entropy,
&zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
&entropyMetadata,
zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
return ZSTD_compressSubBlock_multi(&zc->seqStore,
zc->blockState.prevCBlock,
@@ -841,5 +568,5 @@ size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
dst, dstCapacity,
src, srcSize,
zc->bmi2, lastBlock,
zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
}
Utilities/cmzstd/lib/compress/zstd_compress_superblock.h
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/compress/zstd_cwksp.h
+206 -69
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -35,6 +35,10 @@ extern "C" {
#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
***************************************/
@@ -44,6 +48,16 @@ typedef enum {
ZSTD_cwksp_alloc_aligned
} 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
@@ -92,7 +106,7 @@ typedef enum {
*
* - Static objects: this is optionally the enclosing ZSTD_CCtx or ZSTD_CDict,
* so that literally everything fits in a single buffer. Note: if present,
* this must be the first object in the workspace, since ZSTD_free{CCtx,
* this must be the first object in the workspace, since ZSTD_customFree{CCtx,
* CDict}() rely on a pointer comparison to see whether one or two frees are
* required.
*
@@ -107,10 +121,11 @@ typedef enum {
* - 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.
* Their sizes depend on the cparams. These tables are 64-byte aligned.
*
* - Aligned: these buffers are used for various purposes that require 4 byte
* alignment, but don't require any initialization before they're used.
* alignment, but don't require any initialization before they're used. These
* buffers are each 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
@@ -123,8 +138,7 @@ typedef enum {
*
* 1. Objects
* 2. Buffers
* 3. Aligned
* 4. Tables
* 3. Aligned/Tables
*
* Attempts to reserve objects of different types out of order will fail.
*/
@@ -137,9 +151,10 @@ typedef struct {
void* tableValidEnd;
void* allocStart;
int allocFailed;
BYTE allocFailed;
int workspaceOversizedDuration;
ZSTD_cwksp_alloc_phase_e phase;
ZSTD_cwksp_static_alloc_e isStatic;
} ZSTD_cwksp;
/*-*************************************
@@ -176,39 +191,123 @@ MEM_STATIC size_t ZSTD_cwksp_align(size_t size, size_t const align) {
* 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 defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
if (size == 0)
return 0;
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
return size + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
#else
return size;
#endif
}
MEM_STATIC void ZSTD_cwksp_internal_advance_phase(
/**
* 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 n_1=[1, 64] bytes
* to align the beginning of tables section, as well as another n_2=[0, 63] bytes
* to align the beginning of the aligned secion.
*
* n_1 + n_2 == 64 bytes if the cwksp is freshly allocated, due to tables and
* aligneds being sized in multiples of 64 bytes.
*/
size_t const slackSpace = ZSTD_CWKSP_ALIGNMENT_BYTES;
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 != ZSTD_CWKSP_ALIGNMENT_BYTES);
return bytes;
}
/**
* 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;
}
if (alloc < ws->tableValidEnd) {
ws->tableValidEnd = alloc;
}
ws->allocStart = alloc;
return alloc;
}
/**
* Moves the cwksp to the next phase, and does any necessary allocations.
* 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 buffers */
if (ws->phase < ZSTD_cwksp_alloc_buffers &&
phase >= ZSTD_cwksp_alloc_buffers) {
ws->tableValidEnd = ws->objectEnd;
}
/* Going from allocating buffers to allocating aligneds/tables */
if (ws->phase < ZSTD_cwksp_alloc_aligned &&
phase >= ZSTD_cwksp_alloc_aligned) {
/* If unaligned allocations down from a too-large top have left us
* unaligned, we need to realign our alloc ptr. Technically, this
* can consume space that is unaccounted for in the neededSpace
* calculation. However, I believe this can only happen when the
* workspace is too large, and specifically when it is too large
* by a larger margin than the space that will be consumed. */
/* TODO: cleaner, compiler warning friendly way to do this??? */
ws->allocStart = (BYTE*)ws->allocStart - ((size_t)ws->allocStart & (sizeof(U32)-1));
if (ws->allocStart < ws->tableValidEnd) {
ws->tableValidEnd = ws->allocStart;
{ /* Align the start of the "aligned" to 64 bytes. Use [1, 64] bytes. */
size_t const bytesToAlign =
ZSTD_CWKSP_ALIGNMENT_BYTES - ZSTD_cwksp_bytes_to_align_ptr(ws->allocStart, ZSTD_CWKSP_ALIGNMENT_BYTES);
DEBUGLOG(5, "reserving aligned alignment addtl space: %zu", bytesToAlign);
ZSTD_STATIC_ASSERT((ZSTD_CWKSP_ALIGNMENT_BYTES & (ZSTD_CWKSP_ALIGNMENT_BYTES - 1)) == 0); /* power of 2 */
RETURN_ERROR_IF(!ZSTD_cwksp_reserve_internal_buffer_space(ws, bytesToAlign),
memory_allocation, "aligned phase - alignment initial allocation failed!");
}
{ /* 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 end = (BYTE*)alloc + bytesToAlign;
DEBUGLOG(5, "reserving table alignment addtl space: %zu", bytesToAlign);
RETURN_ERROR_IF(end > ws->workspaceEnd, memory_allocation,
"table phase - alignment initial allocation failed!");
ws->objectEnd = end;
ws->tableEnd = end;
ws->tableValidEnd = end;
}
}
ws->phase = phase;
ZSTD_cwksp_assert_internal_consistency(ws);
}
return 0;
}
/**
@@ -224,34 +323,26 @@ MEM_STATIC int ZSTD_cwksp_owns_buffer(const ZSTD_cwksp* ws, const void* ptr) {
MEM_STATIC void* ZSTD_cwksp_reserve_internal(
ZSTD_cwksp* ws, size_t bytes, ZSTD_cwksp_alloc_phase_e phase) {
void* alloc;
void* bottom = ws->tableEnd;
ZSTD_cwksp_internal_advance_phase(ws, phase);
alloc = (BYTE *)ws->allocStart - bytes;
#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* over-reserve space */
alloc = (BYTE *)alloc - 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
#endif
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;
if (ZSTD_isError(ZSTD_cwksp_internal_advance_phase(ws, phase)) || bytes == 0) {
return NULL;
}
if (alloc < ws->tableValidEnd) {
ws->tableValidEnd = alloc;
}
ws->allocStart = alloc;
#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
#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. */
alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
__asan_unpoison_memory_region(alloc, bytes);
if (alloc) {
alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
__asan_unpoison_memory_region(alloc, bytes);
}
}
#endif
return alloc;
@@ -265,28 +356,36 @@ MEM_STATIC BYTE* ZSTD_cwksp_reserve_buffer(ZSTD_cwksp* ws, size_t bytes) {
}
/**
* Reserves and returns memory sized on and aligned on sizeof(unsigned).
* 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) {
assert((bytes & (sizeof(U32)-1)) == 0);
return ZSTD_cwksp_reserve_internal(ws, ZSTD_cwksp_align(bytes, sizeof(U32)), ZSTD_cwksp_alloc_aligned);
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 sizeof(unsigned). These buffers have the special property that
* 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;
void* alloc = ws->tableEnd;
void* end = (BYTE *)alloc + bytes;
void* top = ws->allocStart;
void* alloc;
void* end;
void* top;
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_internal_advance_phase(ws, phase);
ZSTD_cwksp_assert_internal_consistency(ws);
assert(end <= top);
if (end > top) {
@@ -296,10 +395,14 @@ MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes) {
}
ws->tableEnd = end;
#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
__asan_unpoison_memory_region(alloc, bytes);
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
__asan_unpoison_memory_region(alloc, bytes);
}
#endif
assert((bytes & (ZSTD_CWKSP_ALIGNMENT_BYTES-1)) == 0);
assert(((size_t)alloc & (ZSTD_CWKSP_ALIGNMENT_BYTES-1))== 0);
return alloc;
}
@@ -311,7 +414,7 @@ MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
void* alloc = ws->objectEnd;
void* end = (BYTE*)alloc + roundedBytes;
#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* over-reserve space */
end = (BYTE *)end + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
#endif
@@ -332,11 +435,13 @@ MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
ws->tableEnd = end;
ws->tableValidEnd = end;
#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
* either size. */
alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
__asan_unpoison_memory_region(alloc, bytes);
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
__asan_unpoison_memory_region(alloc, bytes);
}
#endif
return alloc;
@@ -345,7 +450,7 @@ MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
MEM_STATIC void ZSTD_cwksp_mark_tables_dirty(ZSTD_cwksp* ws) {
DEBUGLOG(4, "cwksp: ZSTD_cwksp_mark_tables_dirty");
#if defined (MEMORY_SANITIZER) && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
/* To validate that the table re-use logic is sound, and that we don't
* access table space that we haven't cleaned, we re-"poison" the table
* space every time we mark it dirty. */
@@ -380,7 +485,7 @@ MEM_STATIC void ZSTD_cwksp_clean_tables(ZSTD_cwksp* ws) {
assert(ws->tableValidEnd >= ws->objectEnd);
assert(ws->tableValidEnd <= ws->allocStart);
if (ws->tableValidEnd < ws->tableEnd) {
memset(ws->tableValidEnd, 0, (BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd);
ZSTD_memset(ws->tableValidEnd, 0, (BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd);
}
ZSTD_cwksp_mark_tables_clean(ws);
}
@@ -392,8 +497,12 @@ MEM_STATIC void ZSTD_cwksp_clean_tables(ZSTD_cwksp* ws) {
MEM_STATIC void ZSTD_cwksp_clear_tables(ZSTD_cwksp* ws) {
DEBUGLOG(4, "cwksp: clearing tables!");
#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
{
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* We don't do this when the workspace is statically allocated, because
* when that is the case, we have no capability to hook into the end of the
* workspace's lifecycle to unpoison the memory.
*/
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
size_t size = (BYTE*)ws->tableValidEnd - (BYTE*)ws->objectEnd;
__asan_poison_memory_region(ws->objectEnd, size);
}
@@ -410,7 +519,7 @@ MEM_STATIC void ZSTD_cwksp_clear_tables(ZSTD_cwksp* ws) {
MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
DEBUGLOG(4, "cwksp: clearing!");
#if defined (MEMORY_SANITIZER) && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
#if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
/* To validate that the context re-use logic is sound, and that we don't
* access stuff that this compression hasn't initialized, we re-"poison"
* the workspace (or at least the non-static, non-table parts of it)
@@ -421,8 +530,12 @@ MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
}
#endif
#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
{
#if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
/* We don't do this when the workspace is statically allocated, because
* when that is the case, we have no capability to hook into the end of the
* workspace's lifecycle to unpoison the memory.
*/
if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
size_t size = (BYTE*)ws->workspaceEnd - (BYTE*)ws->objectEnd;
__asan_poison_memory_region(ws->objectEnd, size);
}
@@ -442,7 +555,7 @@ MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
* Any existing values in the workspace are ignored (the previously managed
* buffer, if present, must be separately freed).
*/
MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size) {
MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size, ZSTD_cwksp_static_alloc_e isStatic) {
DEBUGLOG(4, "cwksp: init'ing workspace with %zd bytes", size);
assert(((size_t)start & (sizeof(void*)-1)) == 0); /* ensure correct alignment */
ws->workspace = start;
@@ -450,39 +563,45 @@ MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size) {
ws->objectEnd = ws->workspace;
ws->tableValidEnd = ws->objectEnd;
ws->phase = ZSTD_cwksp_alloc_objects;
ws->isStatic = isStatic;
ZSTD_cwksp_clear(ws);
ws->workspaceOversizedDuration = 0;
ZSTD_cwksp_assert_internal_consistency(ws);
}
MEM_STATIC size_t ZSTD_cwksp_create(ZSTD_cwksp* ws, size_t size, ZSTD_customMem customMem) {
void* workspace = ZSTD_malloc(size, customMem);
void* workspace = ZSTD_customMalloc(size, customMem);
DEBUGLOG(4, "cwksp: creating new workspace with %zd bytes", size);
RETURN_ERROR_IF(workspace == NULL, memory_allocation, "NULL pointer!");
ZSTD_cwksp_init(ws, workspace, size);
ZSTD_cwksp_init(ws, workspace, size, ZSTD_cwksp_dynamic_alloc);
return 0;
}
MEM_STATIC void ZSTD_cwksp_free(ZSTD_cwksp* ws, ZSTD_customMem customMem) {
void *ptr = ws->workspace;
DEBUGLOG(4, "cwksp: freeing workspace");
memset(ws, 0, sizeof(ZSTD_cwksp));
ZSTD_free(ptr, customMem);
ZSTD_memset(ws, 0, sizeof(ZSTD_cwksp));
ZSTD_customFree(ptr, customMem);
}
/**
* 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 its used again).
* 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;
memset(src, 0, sizeof(ZSTD_cwksp));
ZSTD_memset(src, 0, sizeof(ZSTD_cwksp));
}
MEM_STATIC size_t ZSTD_cwksp_sizeof(const ZSTD_cwksp* ws) {
return (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->workspace);
}
MEM_STATIC size_t ZSTD_cwksp_used(const ZSTD_cwksp* ws) {
return (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->workspace)
+ (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->allocStart);
}
MEM_STATIC int ZSTD_cwksp_reserve_failed(const ZSTD_cwksp* ws) {
return ws->allocFailed;
}
@@ -491,6 +610,24 @@ MEM_STATIC int ZSTD_cwksp_reserve_failed(const ZSTD_cwksp* ws) {
* 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, int resizedWorkspace) {
if (resizedWorkspace) {
/* Resized/newly allocated wksp should have exact bounds */
return ZSTD_cwksp_used(ws) == estimatedSpace;
} else {
/* Due to alignment, when reusing a workspace, we can actually consume 63 fewer or more bytes
* than estimatedSpace. See the comments in zstd_cwksp.h for details.
*/
return (ZSTD_cwksp_used(ws) >= estimatedSpace - 63) && (ZSTD_cwksp_used(ws) <= estimatedSpace + 63);
}
}
MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws) {
return (size_t)((BYTE*)ws->allocStart - (BYTE*)ws->tableEnd);
}
Utilities/cmzstd/lib/compress/zstd_double_fast.c
+25 -25
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -31,15 +31,15 @@ void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
* is empty.
*/
for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
U32 const current = (U32)(ip - base);
U32 const curr = (U32)(ip - base);
U32 i;
for (i = 0; i < fastHashFillStep; ++i) {
size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
if (i == 0)
hashSmall[smHash] = current + i;
hashSmall[smHash] = curr + i;
if (i == 0 || hashLarge[lgHash] == 0)
hashLarge[lgHash] = current + i;
hashLarge[lgHash] = curr + i;
/* Only load extra positions for ZSTD_dtlm_full */
if (dtlm == ZSTD_dtlm_fast)
break;
@@ -108,9 +108,9 @@ size_t ZSTD_compressBlock_doubleFast_generic(
/* init */
ip += (dictAndPrefixLength == 0);
if (dictMode == ZSTD_noDict) {
U32 const current = (U32)(ip - base);
U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
U32 const maxRep = current - windowLow;
U32 const curr = (U32)(ip - base);
U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
U32 const maxRep = curr - windowLow;
if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
}
@@ -129,17 +129,17 @@ size_t ZSTD_compressBlock_doubleFast_generic(
size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
size_t const dictHL = ZSTD_hashPtr(ip, dictHBitsL, 8);
size_t const dictHS = ZSTD_hashPtr(ip, dictHBitsS, mls);
U32 const current = (U32)(ip-base);
U32 const curr = (U32)(ip-base);
U32 const matchIndexL = hashLong[h2];
U32 matchIndexS = hashSmall[h];
const BYTE* matchLong = base + matchIndexL;
const BYTE* match = base + matchIndexS;
const U32 repIndex = current + 1 - offset_1;
const U32 repIndex = curr + 1 - offset_1;
const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
&& repIndex < prefixLowestIndex) ?
dictBase + (repIndex - dictIndexDelta) :
base + repIndex;
hashLong[h2] = hashSmall[h] = current; /* update hash tables */
hashLong[h2] = hashSmall[h] = curr; /* update hash tables */
/* check dictMatchState repcode */
if (dictMode == ZSTD_dictMatchState
@@ -177,7 +177,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
offset = (U32)(current - dictMatchIndexL - dictIndexDelta);
offset = (U32)(curr - dictMatchIndexL - dictIndexDelta);
while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
goto _match_found;
} }
@@ -209,7 +209,7 @@ _search_next_long:
size_t const dictHLNext = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
U32 const matchIndexL3 = hashLong[hl3];
const BYTE* matchL3 = base + matchIndexL3;
hashLong[hl3] = current + 1;
hashLong[hl3] = curr + 1;
/* check prefix long +1 match */
if (matchIndexL3 > prefixLowestIndex) {
@@ -228,7 +228,7 @@ _search_next_long:
if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
ip++;
offset = (U32)(current + 1 - dictMatchIndexL3 - dictIndexDelta);
offset = (U32)(curr + 1 - dictMatchIndexL3 - dictIndexDelta);
while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
goto _match_found;
} } }
@@ -236,7 +236,7 @@ _search_next_long:
/* if no long +1 match, explore the short match we found */
if (dictMode == ZSTD_dictMatchState && matchIndexS < prefixLowestIndex) {
mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
offset = (U32)(current - matchIndexS);
offset = (U32)(curr - matchIndexS);
while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
} else {
mLength = ZSTD_count(ip+4, match+4, iend) + 4;
@@ -260,7 +260,7 @@ _match_stored:
if (ip <= ilimit) {
/* Complementary insertion */
/* done after iLimit test, as candidates could be > iend-8 */
{ U32 const indexToInsert = current+2;
{ U32 const indexToInsert = curr+2;
hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
@@ -401,15 +401,15 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? dictBase : base;
const BYTE* matchLong = matchLongBase + matchLongIndex;
const U32 current = (U32)(ip-base);
const U32 repIndex = current + 1 - offset_1; /* offset_1 expected <= current +1 */
const U32 curr = (U32)(ip-base);
const U32 repIndex = curr + 1 - offset_1; /* offset_1 expected <= curr +1 */
const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
const BYTE* const repMatch = repBase + repIndex;
size_t mLength;
hashSmall[hSmall] = hashLong[hLong] = current; /* update hash table */
hashSmall[hSmall] = hashLong[hLong] = curr; /* update hash table */
if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
& (repIndex > dictStartIndex))
& (offset_1 < curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
&& (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
@@ -421,7 +421,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
U32 offset;
mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
offset = current - matchLongIndex;
offset = curr - matchLongIndex;
while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
offset_2 = offset_1;
offset_1 = offset;
@@ -433,19 +433,19 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
const BYTE* match3 = match3Base + matchIndex3;
U32 offset;
hashLong[h3] = current + 1;
hashLong[h3] = curr + 1;
if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
ip++;
offset = current+1 - matchIndex3;
offset = curr+1 - matchIndex3;
while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
} else {
const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
offset = current - matchIndex;
offset = curr - matchIndex;
while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
}
offset_2 = offset_1;
@@ -464,7 +464,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
if (ip <= ilimit) {
/* Complementary insertion */
/* done after iLimit test, as candidates could be > iend-8 */
{ U32 const indexToInsert = current+2;
{ U32 const indexToInsert = curr+2;
hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
@@ -477,7 +477,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
U32 const repIndex2 = current2 - offset_2;
const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) /* intentional overflow : ensure repIndex2 doesn't overlap dict + prefix */
& (repIndex2 > dictStartIndex))
& (offset_2 < current2 - dictStartIndex))
&& (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
Utilities/cmzstd/lib/compress/zstd_double_fast.h
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/compress/zstd_fast.c
+23 -23
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -29,16 +29,16 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
* Insert the other positions if their hash entry is empty.
*/
for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
U32 const current = (U32)(ip - base);
U32 const curr = (U32)(ip - base);
size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls);
hashTable[hash0] = current;
hashTable[hash0] = curr;
if (dtlm == ZSTD_dtlm_fast) continue;
/* Only load extra positions for ZSTD_dtlm_full */
{ U32 p;
for (p = 1; p < fastHashFillStep; ++p) {
size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls);
if (hashTable[hash] == 0) { /* not yet filled */
hashTable[hash] = current + p;
hashTable[hash] = curr + p;
} } } }
}
@@ -72,9 +72,9 @@ ZSTD_compressBlock_fast_generic(
DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
ip0 += (ip0 == prefixStart);
ip1 = ip0 + 1;
{ U32 const current = (U32)(ip0 - base);
U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
U32 const maxRep = current - windowLow;
{ U32 const curr = (U32)(ip0 - base);
U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
U32 const maxRep = curr - windowLow;
if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
}
@@ -242,7 +242,7 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
assert(endIndex - prefixStartIndex <= maxDistance);
(void)maxDistance; (void)endIndex; /* these variables are not used when assert() is disabled */
/* ensure there will be no no underflow
/* ensure there will be no underflow
* when translating a dict index into a local index */
assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
@@ -258,14 +258,14 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
while (ip < ilimit) { /* < instead of <=, because repcode check at (ip+1) */
size_t mLength;
size_t const h = ZSTD_hashPtr(ip, hlog, mls);
U32 const current = (U32)(ip-base);
U32 const curr = (U32)(ip-base);
U32 const matchIndex = hashTable[h];
const BYTE* match = base + matchIndex;
const U32 repIndex = current + 1 - offset_1;
const U32 repIndex = curr + 1 - offset_1;
const BYTE* repMatch = (repIndex < prefixStartIndex) ?
dictBase + (repIndex - dictIndexDelta) :
base + repIndex;
hashTable[h] = current; /* update hash table */
hashTable[h] = curr; /* update hash table */
if ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
&& (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
@@ -284,7 +284,7 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
continue;
} else {
/* found a dict match */
U32 const offset = (U32)(current-dictMatchIndex-dictIndexDelta);
U32 const offset = (U32)(curr-dictMatchIndex-dictIndexDelta);
mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4;
while (((ip>anchor) & (dictMatch>dictStart))
&& (ip[-1] == dictMatch[-1])) {
@@ -316,8 +316,8 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
if (ip <= ilimit) {
/* Fill Table */
assert(base+current+2 > istart); /* check base overflow */
hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2; /* here because current+2 could be > iend-8 */
assert(base+curr+2 > istart); /* check base overflow */
hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2; /* here because curr+2 could be > iend-8 */
hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
/* check immediate repcode */
@@ -410,15 +410,15 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
const U32 matchIndex = hashTable[h];
const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
const BYTE* match = matchBase + matchIndex;
const U32 current = (U32)(ip-base);
const U32 repIndex = current + 1 - offset_1;
const U32 curr = (U32)(ip-base);
const U32 repIndex = curr + 1 - offset_1;
const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
const BYTE* const repMatch = repBase + repIndex;
hashTable[h] = current; /* update hash table */
DEBUGLOG(7, "offset_1 = %u , current = %u", offset_1, current);
assert(offset_1 <= current +1); /* check repIndex */
hashTable[h] = curr; /* update hash table */
DEBUGLOG(7, "offset_1 = %u , curr = %u", offset_1, curr);
if ( (((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > dictStartIndex))
if ( ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */
& (offset_1 < curr+1 - dictStartIndex) ) /* note: we are searching at curr+1 */
&& (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
size_t const rLength = ZSTD_count_2segments(ip+1 +4, repMatch +4, iend, repMatchEnd, prefixStart) + 4;
@@ -435,7 +435,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
}
{ const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
U32 const offset = current - matchIndex;
U32 const offset = curr - matchIndex;
size_t mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
offset_2 = offset_1; offset_1 = offset; /* update offset history */
@@ -446,14 +446,14 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
if (ip <= ilimit) {
/* Fill Table */
hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;
hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;
hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
/* check immediate repcode */
while (ip <= ilimit) {
U32 const current2 = (U32)(ip-base);
U32 const repIndex2 = current2 - offset_2;
const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (repIndex2 > dictStartIndex)) /* intentional overflow */
if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 < curr - dictStartIndex)) /* intentional overflow */
&& (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
Utilities/cmzstd/lib/compress/zstd_fast.h
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/compress/zstd_lazy.c
+1139 -92
View File
File diff suppressed because it is too large Load Diff
Utilities/cmzstd/lib/compress/zstd_lazy.h
+59 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -17,7 +17,18 @@ extern "C" {
#include "zstd_compress_internal.h"
/**
* Dedicated Dictionary Search Structure bucket log. In the
* ZSTD_dedicatedDictSearch mode, the hashTable has
* 2 ** ZSTD_LAZY_DDSS_BUCKET_LOG entries in each bucket, rather than just
* one.
*/
#define ZSTD_LAZY_DDSS_BUCKET_LOG 2
U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip);
void ZSTD_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 */
@@ -33,6 +44,15 @@ size_t ZSTD_compressBlock_lazy(
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_lazy2_row(
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_greedy_row(
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],
@@ -46,6 +66,34 @@ size_t ZSTD_compressBlock_lazy_dictMatchState(
size_t ZSTD_compressBlock_greedy_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_lazy2_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_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_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_lazy_dedicatedDictSearch(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize);
size_t ZSTD_compressBlock_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_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_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],
@@ -56,9 +104,19 @@ size_t ZSTD_compressBlock_lazy_extDict(
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_greedy_extDict_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_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_row(
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);
#if defined (__cplusplus)
}
Utilities/cmzstd/lib/compress/zstd_ldm.c
+312 -209
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -11,13 +11,126 @@
#include "zstd_ldm.h"
#include "../common/debug.h"
#include "../common/xxhash.h"
#include "zstd_fast.h" /* ZSTD_fillHashTable() */
#include "zstd_double_fast.h" /* ZSTD_fillDoubleHashTable() */
#include "zstd_ldm_geartab.h"
#define LDM_BUCKET_SIZE_LOG 3
#define LDM_MIN_MATCH_LENGTH 64
#define LDM_HASH_RLOG 7
#define LDM_HASH_CHAR_OFFSET 10
typedef struct {
U64 rolling;
U64 stopMask;
} ldmRollingHashState_t;
/** ZSTD_ldm_gear_init():
*
* Initializes the rolling hash state such that it will honor the
* settings in params. */
static void ZSTD_ldm_gear_init(ldmRollingHashState_t* state, ldmParams_t const* params)
{
unsigned maxBitsInMask = MIN(params->minMatchLength, 64);
unsigned hashRateLog = params->hashRateLog;
state->rolling = ~(U32)0;
/* The choice of the splitting criterion is subject to two conditions:
* 1. it has to trigger on average every 2^(hashRateLog) bytes;
* 2. ideally, it has to depend on a window of minMatchLength bytes.
*
* In the gear hash algorithm, bit n depends on the last n bytes;
* so in order to obtain a good quality splitting criterion it is
* preferable to use bits with high weight.
*
* To match condition 1 we use a mask with hashRateLog bits set
* and, because of the previous remark, we make sure these bits
* have the highest possible weight while still respecting
* condition 2.
*/
if (hashRateLog > 0 && hashRateLog <= maxBitsInMask) {
state->stopMask = (((U64)1 << hashRateLog) - 1) << (maxBitsInMask - hashRateLog);
} else {
/* In this degenerate case we simply honor the hash rate. */
state->stopMask = ((U64)1 << hashRateLog) - 1;
}
}
/** ZSTD_ldm_gear_reset()
* Feeds [data, data + minMatchLength) into the hash without registering any
* splits. This effectively resets the hash state. This is used when skipping
* over data, either at the beginning of a block, or skipping sections.
*/
static void ZSTD_ldm_gear_reset(ldmRollingHashState_t* state,
BYTE const* data, size_t minMatchLength)
{
U64 hash = state->rolling;
size_t n = 0;
#define GEAR_ITER_ONCE() do { \
hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
n += 1; \
} while (0)
while (n + 3 < minMatchLength) {
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
}
while (n < minMatchLength) {
GEAR_ITER_ONCE();
}
#undef GEAR_ITER_ONCE
}
/** ZSTD_ldm_gear_feed():
*
* Registers in the splits array all the split points found in the first
* size bytes following the data pointer. This function terminates when
* either all the data has been processed or LDM_BATCH_SIZE splits are
* present in the splits array.
*
* Precondition: The splits array must not be full.
* Returns: The number of bytes processed. */
static size_t ZSTD_ldm_gear_feed(ldmRollingHashState_t* state,
BYTE const* data, size_t size,
size_t* splits, unsigned* numSplits)
{
size_t n;
U64 hash, mask;
hash = state->rolling;
mask = state->stopMask;
n = 0;
#define GEAR_ITER_ONCE() do { \
hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
n += 1; \
if (UNLIKELY((hash & mask) == 0)) { \
splits[*numSplits] = n; \
*numSplits += 1; \
if (*numSplits == LDM_BATCH_SIZE) \
goto done; \
} \
} while (0)
while (n + 3 < size) {
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
}
while (n < size) {
GEAR_ITER_ONCE();
}
#undef GEAR_ITER_ONCE
done:
state->rolling = hash;
return n;
}
void ZSTD_ldm_adjustParameters(ldmParams_t* params,
ZSTD_compressionParameters const* cParams)
@@ -27,13 +140,6 @@ void ZSTD_ldm_adjustParameters(ldmParams_t* params,
DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
if (cParams->strategy >= ZSTD_btopt) {
/* Get out of the way of the optimal parser */
U32 const minMatch = MAX(cParams->targetLength, params->minMatchLength);
assert(minMatch >= ZSTD_LDM_MINMATCH_MIN);
assert(minMatch <= ZSTD_LDM_MINMATCH_MAX);
params->minMatchLength = minMatch;
}
if (params->hashLog == 0) {
params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
assert(params->hashLog <= ZSTD_HASHLOG_MAX);
@@ -61,41 +167,6 @@ size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
}
/** ZSTD_ldm_getSmallHash() :
* numBits should be <= 32
* If numBits==0, returns 0.
* @return : the most significant numBits of value. */
static U32 ZSTD_ldm_getSmallHash(U64 value, U32 numBits)
{
assert(numBits <= 32);
return numBits == 0 ? 0 : (U32)(value >> (64 - numBits));
}
/** ZSTD_ldm_getChecksum() :
* numBitsToDiscard should be <= 32
* @return : the next most significant 32 bits after numBitsToDiscard */
static U32 ZSTD_ldm_getChecksum(U64 hash, U32 numBitsToDiscard)
{
assert(numBitsToDiscard <= 32);
return (hash >> (64 - 32 - numBitsToDiscard)) & 0xFFFFFFFF;
}
/** ZSTD_ldm_getTag() ;
* Given the hash, returns the most significant numTagBits bits
* after (32 + hbits) bits.
*
* If there are not enough bits remaining, return the last
* numTagBits bits. */
static U32 ZSTD_ldm_getTag(U64 hash, U32 hbits, U32 numTagBits)
{
assert(numTagBits < 32 && hbits <= 32);
if (32 - hbits < numTagBits) {
return hash & (((U32)1 << numTagBits) - 1);
} else {
return (hash >> (32 - hbits - numTagBits)) & (((U32)1 << numTagBits) - 1);
}
}
/** ZSTD_ldm_getBucket() :
* Returns a pointer to the start of the bucket associated with hash. */
static ldmEntry_t* ZSTD_ldm_getBucket(
@@ -110,38 +181,12 @@ static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
size_t const hash, const ldmEntry_t entry,
ldmParams_t const ldmParams)
{
BYTE* const bucketOffsets = ldmState->bucketOffsets;
*(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + bucketOffsets[hash]) = entry;
bucketOffsets[hash]++;
bucketOffsets[hash] &= ((U32)1 << ldmParams.bucketSizeLog) - 1;
}
BYTE* const pOffset = ldmState->bucketOffsets + hash;
unsigned const offset = *pOffset;
*(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + offset) = entry;
*pOffset = (BYTE)((offset + 1) & ((1u << ldmParams.bucketSizeLog) - 1));
/** ZSTD_ldm_makeEntryAndInsertByTag() :
*
* Gets the small hash, checksum, and tag from the rollingHash.
*
* If the tag matches (1 << ldmParams.hashRateLog)-1, then
* creates an ldmEntry from the offset, and inserts it into the hash table.
*
* hBits is the length of the small hash, which is the most significant hBits
* of rollingHash. The checksum is the next 32 most significant bits, followed
* by ldmParams.hashRateLog bits that make up the tag. */
static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
U64 const rollingHash,
U32 const hBits,
U32 const offset,
ldmParams_t const ldmParams)
{
U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashRateLog);
U32 const tagMask = ((U32)1 << ldmParams.hashRateLog) - 1;
if (tag == tagMask) {
U32 const hash = ZSTD_ldm_getSmallHash(rollingHash, hBits);
U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
ldmEntry_t entry;
entry.offset = offset;
entry.checksum = checksum;
ZSTD_ldm_insertEntry(ldmState, hash, entry, ldmParams);
}
}
/** ZSTD_ldm_countBackwardsMatch() :
@@ -150,10 +195,10 @@ static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
* We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
static size_t ZSTD_ldm_countBackwardsMatch(
const BYTE* pIn, const BYTE* pAnchor,
const BYTE* pMatch, const BYTE* pBase)
const BYTE* pMatch, const BYTE* pMatchBase)
{
size_t matchLength = 0;
while (pIn > pAnchor && pMatch > pBase && pIn[-1] == pMatch[-1]) {
while (pIn > pAnchor && pMatch > pMatchBase && pIn[-1] == pMatch[-1]) {
pIn--;
pMatch--;
matchLength++;
@@ -161,6 +206,27 @@ static size_t ZSTD_ldm_countBackwardsMatch(
return matchLength;
}
/** ZSTD_ldm_countBackwardsMatch_2segments() :
* Returns the number of bytes that match backwards from pMatch,
* even with the backwards match spanning 2 different segments.
*
* On reaching `pMatchBase`, start counting from mEnd */
static size_t ZSTD_ldm_countBackwardsMatch_2segments(
const BYTE* pIn, const BYTE* pAnchor,
const BYTE* pMatch, const BYTE* pMatchBase,
const BYTE* pExtDictStart, const BYTE* pExtDictEnd)
{
size_t matchLength = ZSTD_ldm_countBackwardsMatch(pIn, pAnchor, pMatch, pMatchBase);
if (pMatch - matchLength != pMatchBase || pMatchBase == pExtDictStart) {
/* If backwards match is entirely in the extDict or prefix, immediately return */
return matchLength;
}
DEBUGLOG(7, "ZSTD_ldm_countBackwardsMatch_2segments: found 2-parts backwards match (length in prefix==%zu)", matchLength);
matchLength += ZSTD_ldm_countBackwardsMatch(pIn - matchLength, pAnchor, pExtDictEnd, pExtDictStart);
DEBUGLOG(7, "final backwards match length = %zu", matchLength);
return matchLength;
}
/** ZSTD_ldm_fillFastTables() :
*
* Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
@@ -198,43 +264,42 @@ static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
return 0;
}
/** ZSTD_ldm_fillLdmHashTable() :
*
* Fills hashTable from (lastHashed + 1) to iend (non-inclusive).
* lastHash is the rolling hash that corresponds to lastHashed.
*
* Returns the rolling hash corresponding to position iend-1. */
static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
U64 lastHash, const BYTE* lastHashed,
const BYTE* iend, const BYTE* base,
U32 hBits, ldmParams_t const ldmParams)
{
U64 rollingHash = lastHash;
const BYTE* cur = lastHashed + 1;
while (cur < iend) {
rollingHash = ZSTD_rollingHash_rotate(rollingHash, cur[-1],
cur[ldmParams.minMatchLength-1],
state->hashPower);
ZSTD_ldm_makeEntryAndInsertByTag(state,
rollingHash, hBits,
(U32)(cur - base), ldmParams);
++cur;
}
return rollingHash;
}
void ZSTD_ldm_fillHashTable(
ldmState_t* state, const BYTE* ip,
ldmState_t* ldmState, const BYTE* ip,
const BYTE* iend, ldmParams_t const* params)
{
U32 const minMatchLength = params->minMatchLength;
U32 const hBits = params->hashLog - params->bucketSizeLog;
BYTE const* const base = ldmState->window.base;
BYTE const* const istart = ip;
ldmRollingHashState_t hashState;
size_t* const splits = ldmState->splitIndices;
unsigned numSplits;
DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
if ((size_t)(iend - ip) >= params->minMatchLength) {
U64 startingHash = ZSTD_rollingHash_compute(ip, params->minMatchLength);
ZSTD_ldm_fillLdmHashTable(
state, startingHash, ip, iend - params->minMatchLength, state->window.base,
params->hashLog - params->bucketSizeLog,
*params);
ZSTD_ldm_gear_init(&hashState, params);
while (ip < iend) {
size_t hashed;
unsigned n;
numSplits = 0;
hashed = ZSTD_ldm_gear_feed(&hashState, ip, iend - ip, splits, &numSplits);
for (n = 0; n < numSplits; n++) {
if (ip + splits[n] >= istart + minMatchLength) {
BYTE const* const split = ip + splits[n] - minMatchLength;
U64 const xxhash = XXH64(split, minMatchLength, 0);
U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
ldmEntry_t entry;
entry.offset = (U32)(split - base);
entry.checksum = (U32)(xxhash >> 32);
ZSTD_ldm_insertEntry(ldmState, hash, entry, *params);
}
}
ip += hashed;
}
}
@@ -246,10 +311,10 @@ void ZSTD_ldm_fillHashTable(
* (after a long match, only update tables a limited amount). */
static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
{
U32 const current = (U32)(anchor - ms->window.base);
if (current > ms->nextToUpdate + 1024) {
U32 const curr = (U32)(anchor - ms->window.base);
if (curr > ms->nextToUpdate + 1024) {
ms->nextToUpdate =
current - MIN(512, current - ms->nextToUpdate - 1024);
curr - MIN(512, curr - ms->nextToUpdate - 1024);
}
}
@@ -260,11 +325,8 @@ static size_t ZSTD_ldm_generateSequences_internal(
/* LDM parameters */
int const extDict = ZSTD_window_hasExtDict(ldmState->window);
U32 const minMatchLength = params->minMatchLength;
U64 const hashPower = ldmState->hashPower;
U32 const entsPerBucket = 1U << params->bucketSizeLog;
U32 const hBits = params->hashLog - params->bucketSizeLog;
U32 const ldmBucketSize = 1U << params->bucketSizeLog;
U32 const hashRateLog = params->hashRateLog;
U32 const ldmTagMask = (1U << params->hashRateLog) - 1;
/* Prefix and extDict parameters */
U32 const dictLimit = ldmState->window.dictLimit;
U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
@@ -276,45 +338,69 @@ static size_t ZSTD_ldm_generateSequences_internal(
/* Input bounds */
BYTE const* const istart = (BYTE const*)src;
BYTE const* const iend = istart + srcSize;
BYTE const* const ilimit = iend - MAX(minMatchLength, HASH_READ_SIZE);
BYTE const* const ilimit = iend - HASH_READ_SIZE;
/* Input positions */
BYTE const* anchor = istart;
BYTE const* ip = istart;
/* Rolling hash */
BYTE const* lastHashed = NULL;
U64 rollingHash = 0;
/* Rolling hash state */
ldmRollingHashState_t hashState;
/* Arrays for staged-processing */
size_t* const splits = ldmState->splitIndices;
ldmMatchCandidate_t* const candidates = ldmState->matchCandidates;
unsigned numSplits;
while (ip <= ilimit) {
size_t mLength;
U32 const current = (U32)(ip - base);
size_t forwardMatchLength = 0, backwardMatchLength = 0;
ldmEntry_t* bestEntry = NULL;
if (ip != istart) {
rollingHash = ZSTD_rollingHash_rotate(rollingHash, lastHashed[0],
lastHashed[minMatchLength],
hashPower);
} else {
rollingHash = ZSTD_rollingHash_compute(ip, minMatchLength);
}
lastHashed = ip;
if (srcSize < minMatchLength)
return iend - anchor;
/* Do not insert and do not look for a match */
if (ZSTD_ldm_getTag(rollingHash, hBits, hashRateLog) != ldmTagMask) {
ip++;
continue;
/* Initialize the rolling hash state with the first minMatchLength bytes */
ZSTD_ldm_gear_init(&hashState, params);
ZSTD_ldm_gear_reset(&hashState, ip, minMatchLength);
ip += minMatchLength;
while (ip < ilimit) {
size_t hashed;
unsigned n;
numSplits = 0;
hashed = ZSTD_ldm_gear_feed(&hashState, ip, ilimit - ip,
splits, &numSplits);
for (n = 0; n < numSplits; n++) {
BYTE const* const split = ip + splits[n] - minMatchLength;
U64 const xxhash = XXH64(split, minMatchLength, 0);
U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
candidates[n].split = split;
candidates[n].hash = hash;
candidates[n].checksum = (U32)(xxhash >> 32);
candidates[n].bucket = ZSTD_ldm_getBucket(ldmState, hash, *params);
PREFETCH_L1(candidates[n].bucket);
}
/* Get the best entry and compute the match lengths */
{
ldmEntry_t* const bucket =
ZSTD_ldm_getBucket(ldmState,
ZSTD_ldm_getSmallHash(rollingHash, hBits),
*params);
ldmEntry_t* cur;
size_t bestMatchLength = 0;
U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
for (n = 0; n < numSplits; n++) {
size_t forwardMatchLength = 0, backwardMatchLength = 0,
bestMatchLength = 0, mLength;
U32 offset;
BYTE const* const split = candidates[n].split;
U32 const checksum = candidates[n].checksum;
U32 const hash = candidates[n].hash;
ldmEntry_t* const bucket = candidates[n].bucket;
ldmEntry_t const* cur;
ldmEntry_t const* bestEntry = NULL;
ldmEntry_t newEntry;
for (cur = bucket; cur < bucket + ldmBucketSize; ++cur) {
newEntry.offset = (U32)(split - base);
newEntry.checksum = checksum;
/* If a split point would generate a sequence overlapping with
* the previous one, we merely register it in the hash table and
* move on */
if (split < anchor) {
ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
continue;
}
for (cur = bucket; cur < bucket + entsPerBucket; cur++) {
size_t curForwardMatchLength, curBackwardMatchLength,
curTotalMatchLength;
if (cur->checksum != checksum || cur->offset <= lowestIndex) {
@@ -328,30 +414,23 @@ static size_t ZSTD_ldm_generateSequences_internal(
cur->offset < dictLimit ? dictEnd : iend;
BYTE const* const lowMatchPtr =
cur->offset < dictLimit ? dictStart : lowPrefixPtr;
curForwardMatchLength = ZSTD_count_2segments(
ip, pMatch, iend,
matchEnd, lowPrefixPtr);
curForwardMatchLength =
ZSTD_count_2segments(split, pMatch, iend, matchEnd, lowPrefixPtr);
if (curForwardMatchLength < minMatchLength) {
continue;
}
curBackwardMatchLength =
ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
lowMatchPtr);
curTotalMatchLength = curForwardMatchLength +
curBackwardMatchLength;
curBackwardMatchLength = ZSTD_ldm_countBackwardsMatch_2segments(
split, anchor, pMatch, lowMatchPtr, dictStart, dictEnd);
} else { /* !extDict */
BYTE const* const pMatch = base + cur->offset;
curForwardMatchLength = ZSTD_count(ip, pMatch, iend);
curForwardMatchLength = ZSTD_count(split, pMatch, iend);
if (curForwardMatchLength < minMatchLength) {
continue;
}
curBackwardMatchLength =
ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
lowPrefixPtr);
curTotalMatchLength = curForwardMatchLength +
curBackwardMatchLength;
ZSTD_ldm_countBackwardsMatch(split, anchor, pMatch, lowPrefixPtr);
}
curTotalMatchLength = curForwardMatchLength + curBackwardMatchLength;
if (curTotalMatchLength > bestMatchLength) {
bestMatchLength = curTotalMatchLength;
@@ -360,57 +439,54 @@ static size_t ZSTD_ldm_generateSequences_internal(
bestEntry = cur;
}
}
}
/* No match found -- continue searching */
if (bestEntry == NULL) {
ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash,
hBits, current,
*params);
ip++;
continue;
}
/* No match found -- insert an entry into the hash table
* and process the next candidate match */
if (bestEntry == NULL) {
ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
continue;
}
/* Match found */
mLength = forwardMatchLength + backwardMatchLength;
ip -= backwardMatchLength;
/* Match found */
offset = (U32)(split - base) - bestEntry->offset;
mLength = forwardMatchLength + backwardMatchLength;
{
rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
{
/* Store the sequence:
* ip = current - backwardMatchLength
* The match is at (bestEntry->offset - backwardMatchLength)
/* Out of sequence storage */
if (rawSeqStore->size == rawSeqStore->capacity)
return ERROR(dstSize_tooSmall);
seq->litLength = (U32)(split - backwardMatchLength - anchor);
seq->matchLength = (U32)mLength;
seq->offset = offset;
rawSeqStore->size++;
}
/* Insert the current entry into the hash table --- it must be
* done after the previous block to avoid clobbering bestEntry */
ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
anchor = split + forwardMatchLength;
/* If we find a match that ends after the data that we've hashed
* then we have a repeating, overlapping, pattern. E.g. all zeros.
* If one repetition of the pattern matches our `stopMask` then all
* repetitions will. We don't need to insert them all into out table,
* only the first one. So skip over overlapping matches.
* This is a major speed boost (20x) for compressing a single byte
* repeated, when that byte ends up in the table.
*/
U32 const matchIndex = bestEntry->offset;
U32 const offset = current - matchIndex;
rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
/* Out of sequence storage */
if (rawSeqStore->size == rawSeqStore->capacity)
return ERROR(dstSize_tooSmall);
seq->litLength = (U32)(ip - anchor);
seq->matchLength = (U32)mLength;
seq->offset = offset;
rawSeqStore->size++;
if (anchor > ip + hashed) {
ZSTD_ldm_gear_reset(&hashState, anchor - minMatchLength, minMatchLength);
/* Continue the outter loop at anchor (ip + hashed == anchor). */
ip = anchor - hashed;
break;
}
}
/* Insert the current entry into the hash table */
ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
(U32)(lastHashed - base),
*params);
assert(ip + backwardMatchLength == lastHashed);
/* Fill the hash table from lastHashed+1 to ip+mLength*/
/* Heuristic: don't need to fill the entire table at end of block */
if (ip + mLength <= ilimit) {
rollingHash = ZSTD_ldm_fillLdmHashTable(
ldmState, rollingHash, lastHashed,
ip + mLength, base, hBits, *params);
lastHashed = ip + mLength - 1;
}
ip += mLength;
anchor = ip;
ip += hashed;
}
return iend - anchor;
}
@@ -459,7 +535,7 @@ size_t ZSTD_ldm_generateSequences(
assert(chunkStart < iend);
/* 1. Perform overflow correction if necessary. */
if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
if (ZSTD_window_needOverflowCorrection(ldmState->window, 0, maxDist, ldmState->loadedDictEnd, chunkStart, chunkEnd)) {
U32 const ldmHSize = 1U << params->hashLog;
U32 const correction = ZSTD_window_correctOverflow(
&ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
@@ -562,14 +638,32 @@ static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
return sequence;
}
void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
while (currPos && rawSeqStore->pos < rawSeqStore->size) {
rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
if (currPos >= currSeq.litLength + currSeq.matchLength) {
currPos -= currSeq.litLength + currSeq.matchLength;
rawSeqStore->pos++;
} else {
rawSeqStore->posInSequence = currPos;
break;
}
}
if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
rawSeqStore->posInSequence = 0;
}
}
size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
void const* src, size_t srcSize)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
unsigned const minMatch = cParams->minMatch;
ZSTD_blockCompressor const blockCompressor =
ZSTD_selectBlockCompressor(cParams->strategy, ZSTD_matchState_dictMode(ms));
ZSTD_selectBlockCompressor(cParams->strategy, useRowMatchFinder, ZSTD_matchState_dictMode(ms));
/* Input bounds */
BYTE const* const istart = (BYTE const*)src;
BYTE const* const iend = istart + srcSize;
@@ -577,9 +671,18 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
BYTE const* ip = istart;
DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
/* If using opt parser, use LDMs only as candidates rather than always accepting them */
if (cParams->strategy >= ZSTD_btopt) {
size_t lastLLSize;
ms->ldmSeqStore = rawSeqStore;
lastLLSize = blockCompressor(ms, seqStore, rep, src, srcSize);
ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore, srcSize);
return lastLLSize;
}
assert(rawSeqStore->pos <= rawSeqStore->size);
assert(rawSeqStore->size <= rawSeqStore->capacity);
/* Loop through each sequence and apply the block compressor to the lits */
/* Loop through each sequence and apply the block compressor to the literals */
while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
/* maybeSplitSequence updates rawSeqStore->pos */
rawSeq const sequence = maybeSplitSequence(rawSeqStore,
Utilities/cmzstd/lib/compress/zstd_ldm.h
+9 -2
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -66,6 +66,7 @@ size_t ZSTD_ldm_generateSequences(
*/
size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
void const* src, size_t srcSize);
/**
@@ -73,11 +74,17 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
*
* Skip past `srcSize` bytes worth of sequences in `rawSeqStore`.
* Avoids emitting matches less than `minMatch` bytes.
* Must be called for data with is not passed to ZSTD_ldm_blockCompress().
* 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
Utilities/cmzstd/lib/compress/zstd_ldm_geartab.h
+103
View File
@@ -0,0 +1,103 @@
/*
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZSTD_LDM_GEARTAB_H
#define ZSTD_LDM_GEARTAB_H
static U64 ZSTD_ldm_gearTab[256] = {
0xf5b8f72c5f77775c, 0x84935f266b7ac412, 0xb647ada9ca730ccc,
0xb065bb4b114fb1de, 0x34584e7e8c3a9fd0, 0x4e97e17c6ae26b05,
0x3a03d743bc99a604, 0xcecd042422c4044f, 0x76de76c58524259e,
0x9c8528f65badeaca, 0x86563706e2097529, 0x2902475fa375d889,
0xafb32a9739a5ebe6, 0xce2714da3883e639, 0x21eaf821722e69e,
0x37b628620b628, 0x49a8d455d88caf5, 0x8556d711e6958140,
0x4f7ae74fc605c1f, 0x829f0c3468bd3a20, 0x4ffdc885c625179e,
0x8473de048a3daf1b, 0x51008822b05646b2, 0x69d75d12b2d1cc5f,
0x8c9d4a19159154bc, 0xc3cc10f4abbd4003, 0xd06ddc1cecb97391,
0xbe48e6e7ed80302e, 0x3481db31cee03547, 0xacc3f67cdaa1d210,
0x65cb771d8c7f96cc, 0x8eb27177055723dd, 0xc789950d44cd94be,
0x934feadc3700b12b, 0x5e485f11edbdf182, 0x1e2e2a46fd64767a,
0x2969ca71d82efa7c, 0x9d46e9935ebbba2e, 0xe056b67e05e6822b,
0x94d73f55739d03a0, 0xcd7010bdb69b5a03, 0x455ef9fcd79b82f4,
0x869cb54a8749c161, 0x38d1a4fa6185d225, 0xb475166f94bbe9bb,
0xa4143548720959f1, 0x7aed4780ba6b26ba, 0xd0ce264439e02312,
0x84366d746078d508, 0xa8ce973c72ed17be, 0x21c323a29a430b01,
0x9962d617e3af80ee, 0xab0ce91d9c8cf75b, 0x530e8ee6d19a4dbc,
0x2ef68c0cf53f5d72, 0xc03a681640a85506, 0x496e4e9f9c310967,
0x78580472b59b14a0, 0x273824c23b388577, 0x66bf923ad45cb553,
0x47ae1a5a2492ba86, 0x35e304569e229659, 0x4765182a46870b6f,
0x6cbab625e9099412, 0xddac9a2e598522c1, 0x7172086e666624f2,
0xdf5003ca503b7837, 0x88c0c1db78563d09, 0x58d51865acfc289d,
0x177671aec65224f1, 0xfb79d8a241e967d7, 0x2be1e101cad9a49a,
0x6625682f6e29186b, 0x399553457ac06e50, 0x35dffb4c23abb74,
0x429db2591f54aade, 0xc52802a8037d1009, 0x6acb27381f0b25f3,
0xf45e2551ee4f823b, 0x8b0ea2d99580c2f7, 0x3bed519cbcb4e1e1,
0xff452823dbb010a, 0x9d42ed614f3dd267, 0x5b9313c06257c57b,
0xa114b8008b5e1442, 0xc1fe311c11c13d4b, 0x66e8763ea34c5568,
0x8b982af1c262f05d, 0xee8876faaa75fbb7, 0x8a62a4d0d172bb2a,
0xc13d94a3b7449a97, 0x6dbbba9dc15d037c, 0xc786101f1d92e0f1,
0xd78681a907a0b79b, 0xf61aaf2962c9abb9, 0x2cfd16fcd3cb7ad9,
0x868c5b6744624d21, 0x25e650899c74ddd7, 0xba042af4a7c37463,
0x4eb1a539465a3eca, 0xbe09dbf03b05d5ca, 0x774e5a362b5472ba,
0x47a1221229d183cd, 0x504b0ca18ef5a2df, 0xdffbdfbde2456eb9,
0x46cd2b2fbee34634, 0xf2aef8fe819d98c3, 0x357f5276d4599d61,
0x24a5483879c453e3, 0x88026889192b4b9, 0x28da96671782dbec,
0x4ef37c40588e9aaa, 0x8837b90651bc9fb3, 0xc164f741d3f0e5d6,
0xbc135a0a704b70ba, 0x69cd868f7622ada, 0xbc37ba89e0b9c0ab,
0x47c14a01323552f6, 0x4f00794bacee98bb, 0x7107de7d637a69d5,
0x88af793bb6f2255e, 0xf3c6466b8799b598, 0xc288c616aa7f3b59,
0x81ca63cf42fca3fd, 0x88d85ace36a2674b, 0xd056bd3792389e7,
0xe55c396c4e9dd32d, 0xbefb504571e6c0a6, 0x96ab32115e91e8cc,
0xbf8acb18de8f38d1, 0x66dae58801672606, 0x833b6017872317fb,
0xb87c16f2d1c92864, 0xdb766a74e58b669c, 0x89659f85c61417be,
0xc8daad856011ea0c, 0x76a4b565b6fe7eae, 0xa469d085f6237312,
0xaaf0365683a3e96c, 0x4dbb746f8424f7b8, 0x638755af4e4acc1,
0x3d7807f5bde64486, 0x17be6d8f5bbb7639, 0x903f0cd44dc35dc,
0x67b672eafdf1196c, 0xa676ff93ed4c82f1, 0x521d1004c5053d9d,
0x37ba9ad09ccc9202, 0x84e54d297aacfb51, 0xa0b4b776a143445,
0x820d471e20b348e, 0x1874383cb83d46dc, 0x97edeec7a1efe11c,
0xb330e50b1bdc42aa, 0x1dd91955ce70e032, 0xa514cdb88f2939d5,
0x2791233fd90db9d3, 0x7b670a4cc50f7a9b, 0x77c07d2a05c6dfa5,
0xe3778b6646d0a6fa, 0xb39c8eda47b56749, 0x933ed448addbef28,
0xaf846af6ab7d0bf4, 0xe5af208eb666e49, 0x5e6622f73534cd6a,
0x297daeca42ef5b6e, 0x862daef3d35539a6, 0xe68722498f8e1ea9,
0x981c53093dc0d572, 0xfa09b0bfbf86fbf5, 0x30b1e96166219f15,
0x70e7d466bdc4fb83, 0x5a66736e35f2a8e9, 0xcddb59d2b7c1baef,
0xd6c7d247d26d8996, 0xea4e39eac8de1ba3, 0x539c8bb19fa3aff2,
0x9f90e4c5fd508d8, 0xa34e5956fbaf3385, 0x2e2f8e151d3ef375,
0x173691e9b83faec1, 0xb85a8d56bf016379, 0x8382381267408ae3,
0xb90f901bbdc0096d, 0x7c6ad32933bcec65, 0x76bb5e2f2c8ad595,
0x390f851a6cf46d28, 0xc3e6064da1c2da72, 0xc52a0c101cfa5389,
0xd78eaf84a3fbc530, 0x3781b9e2288b997e, 0x73c2f6dea83d05c4,
0x4228e364c5b5ed7, 0x9d7a3edf0da43911, 0x8edcfeda24686756,
0x5e7667a7b7a9b3a1, 0x4c4f389fa143791d, 0xb08bc1023da7cddc,
0x7ab4be3ae529b1cc, 0x754e6132dbe74ff9, 0x71635442a839df45,
0x2f6fb1643fbe52de, 0x961e0a42cf7a8177, 0xf3b45d83d89ef2ea,
0xee3de4cf4a6e3e9b, 0xcd6848542c3295e7, 0xe4cee1664c78662f,
0x9947548b474c68c4, 0x25d73777a5ed8b0b, 0xc915b1d636b7fc,
0x21c2ba75d9b0d2da, 0x5f6b5dcf608a64a1, 0xdcf333255ff9570c,
0x633b922418ced4ee, 0xc136dde0b004b34a, 0x58cc83b05d4b2f5a,
0x5eb424dda28e42d2, 0x62df47369739cd98, 0xb4e0b42485e4ce17,
0x16e1f0c1f9a8d1e7, 0x8ec3916707560ebf, 0x62ba6e2df2cc9db3,
0xcbf9f4ff77d83a16, 0x78d9d7d07d2bbcc4, 0xef554ce1e02c41f4,
0x8d7581127eccf94d, 0xa9b53336cb3c8a05, 0x38c42c0bf45c4f91,
0x640893cdf4488863, 0x80ec34bc575ea568, 0x39f324f5b48eaa40,
0xe9d9ed1f8eff527f, 0x9224fc058cc5a214, 0xbaba00b04cfe7741,
0x309a9f120fcf52af, 0xa558f3ec65626212, 0x424bec8b7adabe2f,
0x41622513a6aea433, 0xb88da2d5324ca798, 0xd287733b245528a4,
0x9a44697e6d68aec3, 0x7b1093be2f49bb28, 0x50bbec632e3d8aad,
0x6cd90723e1ea8283, 0x897b9e7431b02bf3, 0x219efdcb338a7047,
0x3b0311f0a27c0656, 0xdb17bf91c0db96e7, 0x8cd4fd6b4e85a5b2,
0xfab071054ba6409d, 0x40d6fe831fa9dfd9, 0xaf358debad7d791e,
0xeb8d0e25a65e3e58, 0xbbcbd3df14e08580, 0xcf751f27ecdab2b,
0x2b4da14f2613d8f4
};
#endif /* ZSTD_LDM_GEARTAB_H */
Utilities/cmzstd/lib/compress/zstd_opt.c
+191 -46
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
* Copyright (c) Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -386,32 +386,32 @@ static U32 ZSTD_insertBt1(
const BYTE* const dictEnd = dictBase + dictLimit;
const BYTE* const prefixStart = base + dictLimit;
const BYTE* match;
const U32 current = (U32)(ip-base);
const U32 btLow = btMask >= current ? 0 : current - btMask;
U32* smallerPtr = bt + 2*(current&btMask);
const U32 curr = (U32)(ip-base);
const U32 btLow = btMask >= curr ? 0 : curr - btMask;
U32* smallerPtr = bt + 2*(curr&btMask);
U32* largerPtr = smallerPtr + 1;
U32 dummy32; /* to be nullified at the end */
U32 const windowLow = ms->window.lowLimit;
U32 matchEndIdx = current+8+1;
U32 matchEndIdx = curr+8+1;
size_t bestLength = 8;
U32 nbCompares = 1U << cParams->searchLog;
#ifdef ZSTD_C_PREDICT
U32 predictedSmall = *(bt + 2*((current-1)&btMask) + 0);
U32 predictedLarge = *(bt + 2*((current-1)&btMask) + 1);
U32 predictedSmall = *(bt + 2*((curr-1)&btMask) + 0);
U32 predictedLarge = *(bt + 2*((curr-1)&btMask) + 1);
predictedSmall += (predictedSmall>0);
predictedLarge += (predictedLarge>0);
#endif /* ZSTD_C_PREDICT */
DEBUGLOG(8, "ZSTD_insertBt1 (%u)", current);
DEBUGLOG(8, "ZSTD_insertBt1 (%u)", curr);
assert(ip <= iend-8); /* required for h calculation */
hashTable[h] = current; /* Update Hash Table */
hashTable[h] = curr; /* Update Hash Table */
assert(windowLow > 0);
while (nbCompares-- && (matchIndex >= windowLow)) {
U32* const nextPtr = bt + 2*(matchIndex & btMask);
size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */
assert(matchIndex < current);
assert(matchIndex < curr);
#ifdef ZSTD_C_PREDICT /* note : can create issues when hlog small <= 11 */
const U32* predictPtr = bt + 2*((matchIndex-1) & btMask); /* written this way, as bt is a roll buffer */
@@ -474,8 +474,8 @@ static U32 ZSTD_insertBt1(
*smallerPtr = *largerPtr = 0;
{ U32 positions = 0;
if (bestLength > 384) positions = MIN(192, (U32)(bestLength - 384)); /* speed optimization */
assert(matchEndIdx > current + 8);
return MAX(positions, matchEndIdx - (current + 8));
assert(matchEndIdx > curr + 8);
return MAX(positions, matchEndIdx - (curr + 8));
}
}
@@ -519,7 +519,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
const ZSTD_compressionParameters* const cParams = &ms->cParams;
U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
const BYTE* const base = ms->window.base;
U32 const current = (U32)(ip-base);
U32 const curr = (U32)(ip-base);
U32 const hashLog = cParams->hashLog;
U32 const minMatch = (mls==3) ? 3 : 4;
U32* const hashTable = ms->hashTable;
@@ -533,12 +533,12 @@ U32 ZSTD_insertBtAndGetAllMatches (
U32 const dictLimit = ms->window.dictLimit;
const BYTE* const dictEnd = dictBase + dictLimit;
const BYTE* const prefixStart = base + dictLimit;
U32 const btLow = (btMask >= current) ? 0 : current - btMask;
U32 const windowLow = ZSTD_getLowestMatchIndex(ms, current, cParams->windowLog);
U32 const btLow = (btMask >= curr) ? 0 : curr - btMask;
U32 const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
U32 const matchLow = windowLow ? windowLow : 1;
U32* smallerPtr = bt + 2*(current&btMask);
U32* largerPtr = bt + 2*(current&btMask) + 1;
U32 matchEndIdx = current+8+1; /* farthest referenced position of any match => detects repetitive patterns */
U32* smallerPtr = bt + 2*(curr&btMask);
U32* largerPtr = bt + 2*(curr&btMask) + 1;
U32 matchEndIdx = curr+8+1; /* farthest referenced position of any match => detects repetitive patterns */
U32 dummy32; /* to be nullified at the end */
U32 mnum = 0;
U32 nbCompares = 1U << cParams->searchLog;
@@ -557,7 +557,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
U32 const dmsBtLow = dictMode == ZSTD_dictMatchState && dmsBtMask < dmsHighLimit - dmsLowLimit ? dmsHighLimit - dmsBtMask : dmsLowLimit;
size_t bestLength = lengthToBeat-1;
DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", current);
DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", curr);
/* check repCode */
assert(ll0 <= 1); /* necessarily 1 or 0 */
@@ -565,29 +565,29 @@ U32 ZSTD_insertBtAndGetAllMatches (
U32 repCode;
for (repCode = ll0; repCode < lastR; repCode++) {
U32 const repOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
U32 const repIndex = current - repOffset;
U32 const repIndex = curr - repOffset;
U32 repLen = 0;
assert(current >= dictLimit);
if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < current-dictLimit) { /* equivalent to `current > repIndex >= dictLimit` */
assert(curr >= dictLimit);
if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < curr-dictLimit) { /* equivalent to `curr > repIndex >= dictLimit` */
/* We must validate the repcode offset because when we're using a dictionary the
* valid offset range shrinks when the dictionary goes out of bounds.
*/
if ((repIndex >= windowLow) & (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repOffset, minMatch))) {
repLen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repOffset, iLimit) + minMatch;
}
} else { /* repIndex < dictLimit || repIndex >= current */
} else { /* repIndex < dictLimit || repIndex >= curr */
const BYTE* const repMatch = dictMode == ZSTD_dictMatchState ?
dmsBase + repIndex - dmsIndexDelta :
dictBase + repIndex;
assert(current >= windowLow);
assert(curr >= windowLow);
if ( dictMode == ZSTD_extDict
&& ( ((repOffset-1) /*intentional overflow*/ < current - windowLow) /* equivalent to `current > repIndex >= windowLow` */
&& ( ((repOffset-1) /*intentional overflow*/ < curr - windowLow) /* equivalent to `curr > repIndex >= windowLow` */
& (((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */)
&& (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dictEnd, prefixStart) + minMatch;
}
if (dictMode == ZSTD_dictMatchState
&& ( ((repOffset-1) /*intentional overflow*/ < current - (dmsLowLimit + dmsIndexDelta)) /* equivalent to `current > repIndex >= dmsLowLimit` */
&& ( ((repOffset-1) /*intentional overflow*/ < curr - (dmsLowLimit + dmsIndexDelta)) /* equivalent to `curr > repIndex >= dmsLowLimit` */
& ((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */
&& (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dmsEnd, prefixStart) + minMatch;
@@ -609,7 +609,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
if ((mls == 3) /*static*/ && (bestLength < mls)) {
U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, nextToUpdate3, ip);
if ((matchIndex3 >= matchLow)
& (current - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
& (curr - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
size_t mlen;
if ((dictMode == ZSTD_noDict) /*static*/ || (dictMode == ZSTD_dictMatchState) /*static*/ || (matchIndex3 >= dictLimit)) {
const BYTE* const match = base + matchIndex3;
@@ -624,26 +624,26 @@ U32 ZSTD_insertBtAndGetAllMatches (
DEBUGLOG(8, "found small match with hlog3, of length %u",
(U32)mlen);
bestLength = mlen;
assert(current > matchIndex3);
assert(curr > matchIndex3);
assert(mnum==0); /* no prior solution */
matches[0].off = (current - matchIndex3) + ZSTD_REP_MOVE;
matches[0].off = (curr - matchIndex3) + ZSTD_REP_MOVE;
matches[0].len = (U32)mlen;
mnum = 1;
if ( (mlen > sufficient_len) |
(ip+mlen == iLimit) ) { /* best possible length */
ms->nextToUpdate = current+1; /* skip insertion */
ms->nextToUpdate = curr+1; /* skip insertion */
return 1;
} } }
/* no dictMatchState lookup: dicts don't have a populated HC3 table */
}
hashTable[h] = current; /* Update Hash Table */
hashTable[h] = curr; /* Update Hash Table */
while (nbCompares-- && (matchIndex >= matchLow)) {
U32* const nextPtr = bt + 2*(matchIndex & btMask);
const BYTE* match;
size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */
assert(current > matchIndex);
assert(curr > matchIndex);
if ((dictMode == ZSTD_noDict) || (dictMode == ZSTD_dictMatchState) || (matchIndex+matchLength >= dictLimit)) {
assert(matchIndex+matchLength >= dictLimit); /* ensure the condition is correct when !extDict */
@@ -660,12 +660,12 @@ U32 ZSTD_insertBtAndGetAllMatches (
if (matchLength > bestLength) {
DEBUGLOG(8, "found match of length %u at distance %u (offCode=%u)",
(U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
(U32)matchLength, curr - matchIndex, curr - matchIndex + ZSTD_REP_MOVE);
assert(matchEndIdx > matchIndex);
if (matchLength > matchEndIdx - matchIndex)
matchEndIdx = matchIndex + (U32)matchLength;
bestLength = matchLength;
matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
matches[mnum].off = (curr - matchIndex) + ZSTD_REP_MOVE;
matches[mnum].len = (U32)matchLength;
mnum++;
if ( (matchLength > ZSTD_OPT_NUM)
@@ -708,11 +708,11 @@ U32 ZSTD_insertBtAndGetAllMatches (
if (matchLength > bestLength) {
matchIndex = dictMatchIndex + dmsIndexDelta;
DEBUGLOG(8, "found dms match of length %u at distance %u (offCode=%u)",
(U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
(U32)matchLength, curr - matchIndex, curr - matchIndex + ZSTD_REP_MOVE);
if (matchLength > matchEndIdx - matchIndex)
matchEndIdx = matchIndex + (U32)matchLength;
bestLength = matchLength;
matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
matches[mnum].off = (curr - matchIndex) + ZSTD_REP_MOVE;
matches[mnum].len = (U32)matchLength;
mnum++;
if ( (matchLength > ZSTD_OPT_NUM)
@@ -733,7 +733,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
}
}
assert(matchEndIdx > current+8);
assert(matchEndIdx > curr+8);
ms->nextToUpdate = matchEndIdx - 8; /* skip repetitive patterns */
return mnum;
}
@@ -764,6 +764,140 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
}
}
/*************************
* LDM helper functions *
*************************/
/* Struct containing info needed to make decision about ldm inclusion */
typedef struct {
rawSeqStore_t seqStore; /* External match candidates store for this block */
U32 startPosInBlock; /* Start position of the current match candidate */
U32 endPosInBlock; /* End position of the current match candidate */
U32 offset; /* Offset of the match candidate */
} ZSTD_optLdm_t;
/* ZSTD_optLdm_skipRawSeqStoreBytes():
* Moves forward in rawSeqStore by nbBytes, which will update the fields 'pos' and 'posInSequence'.
*/
static void ZSTD_optLdm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
while (currPos && rawSeqStore->pos < rawSeqStore->size) {
rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
if (currPos >= currSeq.litLength + currSeq.matchLength) {
currPos -= currSeq.litLength + currSeq.matchLength;
rawSeqStore->pos++;
} else {
rawSeqStore->posInSequence = currPos;
break;
}
}
if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
rawSeqStore->posInSequence = 0;
}
}
/* ZSTD_opt_getNextMatchAndUpdateSeqStore():
* Calculates the beginning and end of the next match in the current block.
* Updates 'pos' and 'posInSequence' of the ldmSeqStore.
*/
static void ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
U32 blockBytesRemaining) {
rawSeq currSeq;
U32 currBlockEndPos;
U32 literalsBytesRemaining;
U32 matchBytesRemaining;
/* Setting match end position to MAX to ensure we never use an LDM during this block */
if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
optLdm->startPosInBlock = UINT_MAX;
optLdm->endPosInBlock = UINT_MAX;
return;
}
/* Calculate appropriate bytes left in matchLength and litLength after adjusting
based on ldmSeqStore->posInSequence */
currSeq = optLdm->seqStore.seq[optLdm->seqStore.pos];
assert(optLdm->seqStore.posInSequence <= currSeq.litLength + currSeq.matchLength);
currBlockEndPos = currPosInBlock + blockBytesRemaining;
literalsBytesRemaining = (optLdm->seqStore.posInSequence < currSeq.litLength) ?
currSeq.litLength - (U32)optLdm->seqStore.posInSequence :
0;
matchBytesRemaining = (literalsBytesRemaining == 0) ?
currSeq.matchLength - ((U32)optLdm->seqStore.posInSequence - currSeq.litLength) :
currSeq.matchLength;
/* If there are more literal bytes than bytes remaining in block, no ldm is possible */
if (literalsBytesRemaining >= blockBytesRemaining) {
optLdm->startPosInBlock = UINT_MAX;
optLdm->endPosInBlock = UINT_MAX;
ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, blockBytesRemaining);
return;
}
/* Matches may be < MINMATCH by this process. In that case, we will reject them
when we are deciding whether or not to add the ldm */
optLdm->startPosInBlock = currPosInBlock + literalsBytesRemaining;
optLdm->endPosInBlock = optLdm->startPosInBlock + matchBytesRemaining;
optLdm->offset = currSeq.offset;
if (optLdm->endPosInBlock > currBlockEndPos) {
/* Match ends after the block ends, we can't use the whole match */
optLdm->endPosInBlock = currBlockEndPos;
ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, currBlockEndPos - currPosInBlock);
} else {
/* Consume nb of bytes equal to size of sequence left */
ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, literalsBytesRemaining + matchBytesRemaining);
}
}
/* ZSTD_optLdm_maybeAddMatch():
* Adds a match if it's long enough, based on it's 'matchStartPosInBlock'
* and 'matchEndPosInBlock', into 'matches'. Maintains the correct ordering of 'matches'
*/
static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
ZSTD_optLdm_t* optLdm, U32 currPosInBlock) {
U32 posDiff = currPosInBlock - optLdm->startPosInBlock;
/* Note: ZSTD_match_t actually contains offCode and matchLength (before subtracting MINMATCH) */
U32 candidateMatchLength = optLdm->endPosInBlock - optLdm->startPosInBlock - posDiff;
U32 candidateOffCode = optLdm->offset + ZSTD_REP_MOVE;
/* Ensure that current block position is not outside of the match */
if (currPosInBlock < optLdm->startPosInBlock
|| currPosInBlock >= optLdm->endPosInBlock
|| candidateMatchLength < MINMATCH) {
return;
}
if (*nbMatches == 0 || ((candidateMatchLength > matches[*nbMatches-1].len) && *nbMatches < ZSTD_OPT_NUM)) {
DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offCode: %u matchLength %u) at block position=%u",
candidateOffCode, candidateMatchLength, currPosInBlock);
matches[*nbMatches].len = candidateMatchLength;
matches[*nbMatches].off = candidateOffCode;
(*nbMatches)++;
}
}
/* ZSTD_optLdm_processMatchCandidate():
* Wrapper function to update ldm seq store and call ldm functions as necessary.
*/
static void ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm, ZSTD_match_t* matches, U32* nbMatches,
U32 currPosInBlock, U32 remainingBytes) {
if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
return;
}
if (currPosInBlock >= optLdm->endPosInBlock) {
if (currPosInBlock > optLdm->endPosInBlock) {
/* The position at which ZSTD_optLdm_processMatchCandidate() is called is not necessarily
* at the end of a match from the ldm seq store, and will often be some bytes
* over beyond matchEndPosInBlock. As such, we need to correct for these "overshoots"
*/
U32 posOvershoot = currPosInBlock - optLdm->endPosInBlock;
ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, posOvershoot);
}
ZSTD_opt_getNextMatchAndUpdateSeqStore(optLdm, currPosInBlock, remainingBytes);
}
ZSTD_optLdm_maybeAddMatch(matches, nbMatches, optLdm, currPosInBlock);
}
/*-*******************************
* Optimal parser
@@ -817,6 +951,11 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
ZSTD_optimal_t* const opt = optStatePtr->priceTable;
ZSTD_match_t* const matches = optStatePtr->matchTable;
ZSTD_optimal_t lastSequence;
ZSTD_optLdm_t optLdm;
optLdm.seqStore = ms->ldmSeqStore ? *ms->ldmSeqStore : kNullRawSeqStore;
optLdm.endPosInBlock = optLdm.startPosInBlock = optLdm.offset = 0;
ZSTD_opt_getNextMatchAndUpdateSeqStore(&optLdm, (U32)(ip-istart), (U32)(iend-ip));
/* init */
DEBUGLOG(5, "ZSTD_compressBlock_opt_generic: current=%u, prefix=%u, nextToUpdate=%u",
@@ -832,7 +971,9 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
/* find first match */
{ U32 const litlen = (U32)(ip - anchor);
U32 const ll0 = !litlen;
U32 const nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, ip, iend, dictMode, rep, ll0, minMatch);
U32 nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, ip, iend, dictMode, rep, ll0, minMatch);
ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
(U32)(ip-istart), (U32)(iend - ip));
if (!nbMatches) { ip++; continue; }
/* initialize opt[0] */
@@ -925,9 +1066,9 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
if (opt[cur].mlen != 0) {
U32 const prev = cur - opt[cur].mlen;
repcodes_t newReps = ZSTD_updateRep(opt[prev].rep, opt[cur].off, opt[cur].litlen==0);
memcpy(opt[cur].rep, &newReps, sizeof(repcodes_t));
ZSTD_memcpy(opt[cur].rep, &newReps, sizeof(repcodes_t));
} else {
memcpy(opt[cur].rep, opt[cur - 1].rep, sizeof(repcodes_t));
ZSTD_memcpy(opt[cur].rep, opt[cur - 1].rep, sizeof(repcodes_t));
}
/* last match must start at a minimum distance of 8 from oend */
@@ -945,8 +1086,12 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
U32 const previousPrice = opt[cur].price;
U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
U32 const nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, inr, iend, dictMode, opt[cur].rep, ll0, minMatch);
U32 nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, inr, iend, dictMode, opt[cur].rep, ll0, minMatch);
U32 matchNb;
ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
(U32)(inr-istart), (U32)(iend-inr));
if (!nbMatches) {
DEBUGLOG(7, "rPos:%u : no match found", cur);
continue;
@@ -1010,9 +1155,9 @@ _shortestPath: /* cur, last_pos, best_mlen, best_off have to be set */
*/
if (lastSequence.mlen != 0) {
repcodes_t reps = ZSTD_updateRep(opt[cur].rep, lastSequence.off, lastSequence.litlen==0);
memcpy(rep, &reps, sizeof(reps));
ZSTD_memcpy(rep, &reps, sizeof(reps));
} else {
memcpy(rep, opt[cur].rep, sizeof(repcodes_t));
ZSTD_memcpy(rep, opt[cur].rep, sizeof(repcodes_t));
}
{ U32 const storeEnd = cur + 1;
@@ -1110,7 +1255,7 @@ ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
const void* src, size_t srcSize)
{
U32 tmpRep[ZSTD_REP_NUM]; /* updated rep codes will sink here */
memcpy(tmpRep, rep, sizeof(tmpRep));
ZSTD_memcpy(tmpRep, rep, sizeof(tmpRep));
DEBUGLOG(4, "ZSTD_initStats_ultra (srcSize=%zu)", srcSize);
assert(ms->opt.litLengthSum == 0); /* first block */
@@ -1143,7 +1288,7 @@ size_t ZSTD_compressBlock_btultra2(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
const void* src, size_t srcSize)
{
U32 const current = (U32)((const BYTE*)src - ms->window.base);
U32 const curr = (U32)((const BYTE*)src - ms->window.base);
DEBUGLOG(5, "ZSTD_compressBlock_btultra2 (srcSize=%zu)", srcSize);
/* 2-pass strategy:
@@ -1158,7 +1303,7 @@ size_t ZSTD_compressBlock_btultra2(
if ( (ms->opt.litLengthSum==0) /* first block */
&& (seqStore->sequences == seqStore->sequencesStart) /* no ldm */
&& (ms->window.dictLimit == ms->window.lowLimit) /* no dictionary */
&& (current == ms->window.dictLimit) /* start of frame, nothing already loaded nor skipped */
&& (curr == ms->window.dictLimit) /* start of frame, nothing already loaded nor skipped */
&& (srcSize > ZSTD_PREDEF_THRESHOLD)
) {
ZSTD_initStats_ultra(ms, seqStore, rep, src, srcSize);
Utilities/cmzstd/lib/compress/zstd_opt.h
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/compress/zstdmt_compress.c
+93 -415
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -20,8 +20,7 @@
/* ====== Dependencies ====== */
#include <string.h> /* memcpy, memset */
#include <limits.h> /* INT_MAX, UINT_MAX */
#include "../common/zstd_deps.h" /* ZSTD_memcpy, ZSTD_memset, INT_MAX, UINT_MAX */
#include "../common/mem.h" /* MEM_STATIC */
#include "../common/pool.h" /* threadpool */
#include "../common/threading.h" /* mutex */
@@ -106,11 +105,11 @@ typedef struct ZSTDMT_bufferPool_s {
static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbWorkers, ZSTD_customMem cMem)
{
unsigned const maxNbBuffers = 2*nbWorkers + 3;
ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_calloc(
ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_customCalloc(
sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem);
if (bufPool==NULL) return NULL;
if (ZSTD_pthread_mutex_init(&bufPool->poolMutex, NULL)) {
ZSTD_free(bufPool, cMem);
ZSTD_customFree(bufPool, cMem);
return NULL;
}
bufPool->bufferSize = 64 KB;
@@ -127,10 +126,10 @@ static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
if (!bufPool) return; /* compatibility with free on NULL */
for (u=0; u<bufPool->totalBuffers; u++) {
DEBUGLOG(4, "free buffer %2u (address:%08X)", u, (U32)(size_t)bufPool->bTable[u].start);
ZSTD_free(bufPool->bTable[u].start, bufPool->cMem);
ZSTD_customFree(bufPool->bTable[u].start, bufPool->cMem);
}
ZSTD_pthread_mutex_destroy(&bufPool->poolMutex);
ZSTD_free(bufPool, bufPool->cMem);
ZSTD_customFree(bufPool, bufPool->cMem);
}
/* only works at initialization, not during compression */
@@ -201,13 +200,13 @@ static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
}
/* size conditions not respected : scratch this buffer, create new one */
DEBUGLOG(5, "ZSTDMT_getBuffer: existing buffer does not meet size conditions => freeing");
ZSTD_free(buf.start, bufPool->cMem);
ZSTD_customFree(buf.start, bufPool->cMem);
}
ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
/* create new buffer */
DEBUGLOG(5, "ZSTDMT_getBuffer: create a new buffer");
{ buffer_t buffer;
void* const start = ZSTD_malloc(bSize, bufPool->cMem);
void* const start = ZSTD_customMalloc(bSize, bufPool->cMem);
buffer.start = start; /* note : start can be NULL if malloc fails ! */
buffer.capacity = (start==NULL) ? 0 : bSize;
if (start==NULL) {
@@ -229,13 +228,13 @@ static buffer_t ZSTDMT_resizeBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buffer)
{
size_t const bSize = bufPool->bufferSize;
if (buffer.capacity < bSize) {
void* const start = ZSTD_malloc(bSize, bufPool->cMem);
void* const start = ZSTD_customMalloc(bSize, bufPool->cMem);
buffer_t newBuffer;
newBuffer.start = start;
newBuffer.capacity = start == NULL ? 0 : bSize;
if (start != NULL) {
assert(newBuffer.capacity >= buffer.capacity);
memcpy(newBuffer.start, buffer.start, buffer.capacity);
ZSTD_memcpy(newBuffer.start, buffer.start, buffer.capacity);
DEBUGLOG(5, "ZSTDMT_resizeBuffer: created buffer of size %u", (U32)bSize);
return newBuffer;
}
@@ -261,14 +260,12 @@ static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
/* Reached bufferPool capacity (should not happen) */
DEBUGLOG(5, "ZSTDMT_releaseBuffer: pool capacity reached => freeing ");
ZSTD_free(buf.start, bufPool->cMem);
ZSTD_customFree(buf.start, bufPool->cMem);
}
/* ===== Seq Pool Wrapper ====== */
static rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0};
typedef ZSTDMT_bufferPool ZSTDMT_seqPool;
static size_t ZSTDMT_sizeof_seqPool(ZSTDMT_seqPool* seqPool)
@@ -278,7 +275,7 @@ static size_t ZSTDMT_sizeof_seqPool(ZSTDMT_seqPool* seqPool)
static rawSeqStore_t bufferToSeq(buffer_t buffer)
{
rawSeqStore_t seq = {NULL, 0, 0, 0};
rawSeqStore_t seq = kNullRawSeqStore;
seq.seq = (rawSeq*)buffer.start;
seq.capacity = buffer.capacity / sizeof(rawSeq);
return seq;
@@ -354,7 +351,7 @@ static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
for (cid=0; cid<pool->totalCCtx; cid++)
ZSTD_freeCCtx(pool->cctx[cid]); /* note : compatible with free on NULL */
ZSTD_pthread_mutex_destroy(&pool->poolMutex);
ZSTD_free(pool, pool->cMem);
ZSTD_customFree(pool, pool->cMem);
}
/* ZSTDMT_createCCtxPool() :
@@ -362,12 +359,12 @@ static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(int nbWorkers,
ZSTD_customMem cMem)
{
ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_calloc(
ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_customCalloc(
sizeof(ZSTDMT_CCtxPool) + (nbWorkers-1)*sizeof(ZSTD_CCtx*), cMem);
assert(nbWorkers > 0);
if (!cctxPool) return NULL;
if (ZSTD_pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
ZSTD_free(cctxPool, cMem);
ZSTD_customFree(cctxPool, cMem);
return NULL;
}
cctxPool->cMem = cMem;
@@ -475,10 +472,8 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
assert(params.ldmParams.hashRateLog < 32);
serialState->ldmState.hashPower =
ZSTD_rollingHash_primePower(params.ldmParams.minMatchLength);
} else {
memset(&params.ldmParams, 0, sizeof(params.ldmParams));
ZSTD_memset(&params.ldmParams, 0, sizeof(params.ldmParams));
}
serialState->nextJobID = 0;
if (params.fParams.checksumFlag)
@@ -489,35 +484,35 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
size_t const hashSize = ((size_t)1 << hashLog) * sizeof(ldmEntry_t);
unsigned const bucketLog =
params.ldmParams.hashLog - params.ldmParams.bucketSizeLog;
size_t const bucketSize = (size_t)1 << bucketLog;
unsigned const prevBucketLog =
serialState->params.ldmParams.hashLog -
serialState->params.ldmParams.bucketSizeLog;
size_t const numBuckets = (size_t)1 << bucketLog;
/* Size the seq pool tables */
ZSTDMT_setNbSeq(seqPool, ZSTD_ldm_getMaxNbSeq(params.ldmParams, jobSize));
/* Reset the window */
ZSTD_window_init(&serialState->ldmState.window);
/* Resize tables and output space if necessary. */
if (serialState->ldmState.hashTable == NULL || serialState->params.ldmParams.hashLog < hashLog) {
ZSTD_free(serialState->ldmState.hashTable, cMem);
serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_malloc(hashSize, cMem);
ZSTD_customFree(serialState->ldmState.hashTable, cMem);
serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_customMalloc(hashSize, cMem);
}
if (serialState->ldmState.bucketOffsets == NULL || prevBucketLog < bucketLog) {
ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_malloc(bucketSize, cMem);
ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_customMalloc(numBuckets, cMem);
}
if (!serialState->ldmState.hashTable || !serialState->ldmState.bucketOffsets)
return 1;
/* Zero the tables */
memset(serialState->ldmState.hashTable, 0, hashSize);
memset(serialState->ldmState.bucketOffsets, 0, bucketSize);
ZSTD_memset(serialState->ldmState.hashTable, 0, hashSize);
ZSTD_memset(serialState->ldmState.bucketOffsets, 0, numBuckets);
/* Update window state and fill hash table with dict */
serialState->ldmState.loadedDictEnd = 0;
if (dictSize > 0) {
if (dictContentType == ZSTD_dct_rawContent) {
BYTE const* const dictEnd = (const BYTE*)dict + dictSize;
ZSTD_window_update(&serialState->ldmState.window, dict, dictSize);
ZSTD_window_update(&serialState->ldmState.window, dict, dictSize, /* forceNonContiguous */ 0);
ZSTD_ldm_fillHashTable(&serialState->ldmState, (const BYTE*)dict, dictEnd, &params.ldmParams);
serialState->ldmState.loadedDictEnd = params.forceWindow ? 0 : (U32)(dictEnd - serialState->ldmState.window.base);
} else {
@@ -537,7 +532,7 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
static int ZSTDMT_serialState_init(serialState_t* serialState)
{
int initError = 0;
memset(serialState, 0, sizeof(*serialState));
ZSTD_memset(serialState, 0, sizeof(*serialState));
initError |= ZSTD_pthread_mutex_init(&serialState->mutex, NULL);
initError |= ZSTD_pthread_cond_init(&serialState->cond, NULL);
initError |= ZSTD_pthread_mutex_init(&serialState->ldmWindowMutex, NULL);
@@ -552,8 +547,8 @@ static void ZSTDMT_serialState_free(serialState_t* serialState)
ZSTD_pthread_cond_destroy(&serialState->cond);
ZSTD_pthread_mutex_destroy(&serialState->ldmWindowMutex);
ZSTD_pthread_cond_destroy(&serialState->ldmWindowCond);
ZSTD_free(serialState->ldmState.hashTable, cMem);
ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
ZSTD_customFree(serialState->ldmState.hashTable, cMem);
ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
}
static void ZSTDMT_serialState_update(serialState_t* serialState,
@@ -574,7 +569,7 @@ static void ZSTDMT_serialState_update(serialState_t* serialState,
assert(seqStore.seq != NULL && seqStore.pos == 0 &&
seqStore.size == 0 && seqStore.capacity > 0);
assert(src.size <= serialState->params.jobSize);
ZSTD_window_update(&serialState->ldmState.window, src.start, src.size);
ZSTD_window_update(&serialState->ldmState.window, src.start, src.size, /* forceNonContiguous */ 0);
error = ZSTD_ldm_generateSequences(
&serialState->ldmState, &seqStore,
&serialState->params.ldmParams, src.start, src.size);
@@ -686,6 +681,8 @@ static void ZSTDMT_compressionJob(void* jobDescription)
if (job->jobID != 0) jobParams.fParams.checksumFlag = 0;
/* Don't run LDM for the chunks, since we handle it externally */
jobParams.ldmParams.enableLdm = 0;
/* Correct nbWorkers to 0. */
jobParams.nbWorkers = 0;
/* init */
@@ -698,6 +695,10 @@ static void ZSTDMT_compressionJob(void* jobDescription)
{ size_t const forceWindowError = ZSTD_CCtxParams_setParameter(&jobParams, ZSTD_c_forceMaxWindow, !job->firstJob);
if (ZSTD_isError(forceWindowError)) JOB_ERROR(forceWindowError);
}
if (!job->firstJob) {
size_t const err = ZSTD_CCtxParams_setParameter(&jobParams, ZSTD_c_deterministicRefPrefix, 0);
if (ZSTD_isError(err)) JOB_ERROR(err);
}
{ size_t const initError = ZSTD_compressBegin_advanced_internal(cctx,
job->prefix.start, job->prefix.size, ZSTD_dct_rawContent, /* load dictionary in "content-only" mode (no header analysis) */
ZSTD_dtlm_fast,
@@ -753,6 +754,13 @@ static void ZSTDMT_compressionJob(void* jobDescription)
if (ZSTD_isError(cSize)) JOB_ERROR(cSize);
lastCBlockSize = cSize;
} }
if (!job->firstJob) {
/* Double check that we don't have an ext-dict, because then our
* repcode invalidation doesn't work.
*/
assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
}
ZSTD_CCtx_trace(cctx, 0);
_endJob:
ZSTDMT_serialState_ensureFinished(job->serial, job->jobID, job->cSize);
@@ -820,7 +828,6 @@ struct ZSTDMT_CCtx_s {
roundBuff_t roundBuff;
serialState_t serial;
rsyncState_t rsync;
unsigned singleBlockingThread;
unsigned jobIDMask;
unsigned doneJobID;
unsigned nextJobID;
@@ -832,6 +839,7 @@ struct ZSTDMT_CCtx_s {
ZSTD_customMem cMem;
ZSTD_CDict* cdictLocal;
const ZSTD_CDict* cdict;
unsigned providedFactory: 1;
};
static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZSTD_customMem cMem)
@@ -842,7 +850,7 @@ static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZS
ZSTD_pthread_mutex_destroy(&jobTable[jobNb].job_mutex);
ZSTD_pthread_cond_destroy(&jobTable[jobNb].job_cond);
}
ZSTD_free(jobTable, cMem);
ZSTD_customFree(jobTable, cMem);
}
/* ZSTDMT_allocJobsTable()
@@ -854,7 +862,7 @@ static ZSTDMT_jobDescription* ZSTDMT_createJobsTable(U32* nbJobsPtr, ZSTD_custom
U32 const nbJobs = 1 << nbJobsLog2;
U32 jobNb;
ZSTDMT_jobDescription* const jobTable = (ZSTDMT_jobDescription*)
ZSTD_calloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
ZSTD_customCalloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
int initError = 0;
if (jobTable==NULL) return NULL;
*nbJobsPtr = nbJobs;
@@ -885,12 +893,12 @@ static size_t ZSTDMT_expandJobsTable (ZSTDMT_CCtx* mtctx, U32 nbWorkers) {
/* ZSTDMT_CCtxParam_setNbWorkers():
* Internal use only */
size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
static size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
{
return ZSTD_CCtxParams_setParameter(params, ZSTD_c_nbWorkers, (int)nbWorkers);
}
MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers, ZSTD_customMem cMem)
MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
{
ZSTDMT_CCtx* mtctx;
U32 nbJobs = nbWorkers + 2;
@@ -903,12 +911,19 @@ MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers,
/* invalid custom allocator */
return NULL;
mtctx = (ZSTDMT_CCtx*) ZSTD_calloc(sizeof(ZSTDMT_CCtx), cMem);
mtctx = (ZSTDMT_CCtx*) ZSTD_customCalloc(sizeof(ZSTDMT_CCtx), cMem);
if (!mtctx) return NULL;
ZSTDMT_CCtxParam_setNbWorkers(&mtctx->params, nbWorkers);
mtctx->cMem = cMem;
mtctx->allJobsCompleted = 1;
mtctx->factory = POOL_create_advanced(nbWorkers, 0, cMem);
if (pool != NULL) {
mtctx->factory = pool;
mtctx->providedFactory = 1;
}
else {
mtctx->factory = POOL_create_advanced(nbWorkers, 0, cMem);
mtctx->providedFactory = 0;
}
mtctx->jobs = ZSTDMT_createJobsTable(&nbJobs, cMem);
assert(nbJobs > 0); assert((nbJobs & (nbJobs - 1)) == 0); /* ensure nbJobs is a power of 2 */
mtctx->jobIDMask = nbJobs - 1;
@@ -925,22 +940,18 @@ MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers,
return mtctx;
}
ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem)
ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
{
#ifdef ZSTD_MULTITHREAD
return ZSTDMT_createCCtx_advanced_internal(nbWorkers, cMem);
return ZSTDMT_createCCtx_advanced_internal(nbWorkers, cMem, pool);
#else
(void)nbWorkers;
(void)cMem;
(void)pool;
return NULL;
#endif
}
ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbWorkers)
{
return ZSTDMT_createCCtx_advanced(nbWorkers, ZSTD_defaultCMem);
}
/* ZSTDMT_releaseAllJobResources() :
* note : ensure all workers are killed first ! */
@@ -957,7 +968,7 @@ static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
/* Clear the job description, but keep the mutex/cond */
memset(&mtctx->jobs[jobID], 0, sizeof(mtctx->jobs[jobID]));
ZSTD_memset(&mtctx->jobs[jobID], 0, sizeof(mtctx->jobs[jobID]));
mtctx->jobs[jobID].job_mutex = mutex;
mtctx->jobs[jobID].job_cond = cond;
}
@@ -984,7 +995,8 @@ static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* mtctx)
size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
{
if (mtctx==NULL) return 0; /* compatible with free on NULL */
POOL_free(mtctx->factory); /* stop and free worker threads */
if (!mtctx->providedFactory)
POOL_free(mtctx->factory); /* stop and free worker threads */
ZSTDMT_releaseAllJobResources(mtctx); /* release job resources into pools first */
ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
ZSTDMT_freeBufferPool(mtctx->bufPool);
@@ -993,8 +1005,8 @@ size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
ZSTDMT_serialState_free(&mtctx->serial);
ZSTD_freeCDict(mtctx->cdictLocal);
if (mtctx->roundBuff.buffer)
ZSTD_free(mtctx->roundBuff.buffer, mtctx->cMem);
ZSTD_free(mtctx, mtctx->cMem);
ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
ZSTD_customFree(mtctx, mtctx->cMem);
return 0;
}
@@ -1011,65 +1023,6 @@ size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
+ mtctx->roundBuff.capacity;
}
/* Internal only */
size_t
ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params,
ZSTDMT_parameter parameter,
int value)
{
DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter");
switch(parameter)
{
case ZSTDMT_p_jobSize :
DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter : set jobSize to %i", value);
return ZSTD_CCtxParams_setParameter(params, ZSTD_c_jobSize, value);
case ZSTDMT_p_overlapLog :
DEBUGLOG(4, "ZSTDMT_p_overlapLog : %i", value);
return ZSTD_CCtxParams_setParameter(params, ZSTD_c_overlapLog, value);
case ZSTDMT_p_rsyncable :
DEBUGLOG(4, "ZSTD_p_rsyncable : %i", value);
return ZSTD_CCtxParams_setParameter(params, ZSTD_c_rsyncable, value);
default :
return ERROR(parameter_unsupported);
}
}
size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int value)
{
DEBUGLOG(4, "ZSTDMT_setMTCtxParameter");
return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
}
size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int* value)
{
switch (parameter) {
case ZSTDMT_p_jobSize:
return ZSTD_CCtxParams_getParameter(&mtctx->params, ZSTD_c_jobSize, value);
case ZSTDMT_p_overlapLog:
return ZSTD_CCtxParams_getParameter(&mtctx->params, ZSTD_c_overlapLog, value);
case ZSTDMT_p_rsyncable:
return ZSTD_CCtxParams_getParameter(&mtctx->params, ZSTD_c_rsyncable, value);
default:
return ERROR(parameter_unsupported);
}
}
/* Sets parameters relevant to the compression job,
* initializing others to default values. */
static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(const ZSTD_CCtx_params* params)
{
ZSTD_CCtx_params jobParams = *params;
/* Clear parameters related to multithreading */
jobParams.forceWindow = 0;
jobParams.nbWorkers = 0;
jobParams.jobSize = 0;
jobParams.overlapLog = 0;
jobParams.rsyncable = 0;
memset(&jobParams.ldmParams, 0, sizeof(ldmParams_t));
memset(&jobParams.customMem, 0, sizeof(ZSTD_customMem));
return jobParams;
}
/* ZSTDMT_resize() :
* @return : error code if fails, 0 on success */
@@ -1098,7 +1051,7 @@ void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_p
DEBUGLOG(5, "ZSTDMT_updateCParams_whileCompressing (level:%i)",
compressionLevel);
mtctx->params.compressionLevel = compressionLevel;
{ ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, 0);
{ ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
cParams.windowLog = saved_wlog;
mtctx->params.cParams = cParams;
}
@@ -1185,8 +1138,8 @@ static unsigned ZSTDMT_computeTargetJobLog(const ZSTD_CCtx_params* params)
if (params->ldmParams.enableLdm) {
/* In Long Range Mode, the windowLog is typically oversized.
* In which case, it's preferable to determine the jobSize
* based on chainLog instead. */
jobLog = MAX(21, params->cParams.chainLog + 4);
* based on cycleLog instead. */
jobLog = MAX(21, ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy) + 3);
} else {
jobLog = MAX(20, params->cParams.windowLog + 2);
}
@@ -1240,174 +1193,6 @@ static size_t ZSTDMT_computeOverlapSize(const ZSTD_CCtx_params* params)
return (ovLog==0) ? 0 : (size_t)1 << ovLog;
}
static unsigned
ZSTDMT_computeNbJobs(const ZSTD_CCtx_params* params, size_t srcSize, unsigned nbWorkers)
{
assert(nbWorkers>0);
{ size_t const jobSizeTarget = (size_t)1 << ZSTDMT_computeTargetJobLog(params);
size_t const jobMaxSize = jobSizeTarget << 2;
size_t const passSizeMax = jobMaxSize * nbWorkers;
unsigned const multiplier = (unsigned)(srcSize / passSizeMax) + 1;
unsigned const nbJobsLarge = multiplier * nbWorkers;
unsigned const nbJobsMax = (unsigned)(srcSize / jobSizeTarget) + 1;
unsigned const nbJobsSmall = MIN(nbJobsMax, nbWorkers);
return (multiplier>1) ? nbJobsLarge : nbJobsSmall;
} }
/* ZSTDMT_compress_advanced_internal() :
* This is a blocking function : it will only give back control to caller after finishing its compression job.
*/
static size_t
ZSTDMT_compress_advanced_internal(
ZSTDMT_CCtx* mtctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const ZSTD_CDict* cdict,
ZSTD_CCtx_params params)
{
ZSTD_CCtx_params const jobParams = ZSTDMT_initJobCCtxParams(&params);
size_t const overlapSize = ZSTDMT_computeOverlapSize(&params);
unsigned const nbJobs = ZSTDMT_computeNbJobs(&params, srcSize, params.nbWorkers);
size_t const proposedJobSize = (srcSize + (nbJobs-1)) / nbJobs;
size_t const avgJobSize = (((proposedJobSize-1) & 0x1FFFF) < 0x7FFF) ? proposedJobSize + 0xFFFF : proposedJobSize; /* avoid too small last block */
const char* const srcStart = (const char*)src;
size_t remainingSrcSize = srcSize;
unsigned const compressWithinDst = (dstCapacity >= ZSTD_compressBound(srcSize)) ? nbJobs : (unsigned)(dstCapacity / ZSTD_compressBound(avgJobSize)); /* presumes avgJobSize >= 256 KB, which should be the case */
size_t frameStartPos = 0, dstBufferPos = 0;
assert(jobParams.nbWorkers == 0);
assert(mtctx->cctxPool->totalCCtx == params.nbWorkers);
params.jobSize = (U32)avgJobSize;
DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: nbJobs=%2u (rawSize=%u bytes; fixedSize=%u) ",
nbJobs, (U32)proposedJobSize, (U32)avgJobSize);
if ((nbJobs==1) | (params.nbWorkers<=1)) { /* fallback to single-thread mode : this is a blocking invocation anyway */
ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];
DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: fallback to single-thread mode");
if (cdict) return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, jobParams.fParams);
return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, NULL, 0, &jobParams);
}
assert(avgJobSize >= 256 KB); /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), required to compress directly into Dst (no additional buffer) */
ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgJobSize) );
/* LDM doesn't even try to load the dictionary in single-ingestion mode */
if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params, avgJobSize, NULL, 0, ZSTD_dct_auto))
return ERROR(memory_allocation);
FORWARD_IF_ERROR( ZSTDMT_expandJobsTable(mtctx, nbJobs) , ""); /* only expands if necessary */
{ unsigned u;
for (u=0; u<nbJobs; u++) {
size_t const jobSize = MIN(remainingSrcSize, avgJobSize);
size_t const dstBufferCapacity = ZSTD_compressBound(jobSize);
buffer_t const dstAsBuffer = { (char*)dst + dstBufferPos, dstBufferCapacity };
buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : g_nullBuffer;
size_t dictSize = u ? overlapSize : 0;
mtctx->jobs[u].prefix.start = srcStart + frameStartPos - dictSize;
mtctx->jobs[u].prefix.size = dictSize;
mtctx->jobs[u].src.start = srcStart + frameStartPos;
mtctx->jobs[u].src.size = jobSize; assert(jobSize > 0); /* avoid job.src.size == 0 */
mtctx->jobs[u].consumed = 0;
mtctx->jobs[u].cSize = 0;
mtctx->jobs[u].cdict = (u==0) ? cdict : NULL;
mtctx->jobs[u].fullFrameSize = srcSize;
mtctx->jobs[u].params = jobParams;
/* do not calculate checksum within sections, but write it in header for first section */
mtctx->jobs[u].dstBuff = dstBuffer;
mtctx->jobs[u].cctxPool = mtctx->cctxPool;
mtctx->jobs[u].bufPool = mtctx->bufPool;
mtctx->jobs[u].seqPool = mtctx->seqPool;
mtctx->jobs[u].serial = &mtctx->serial;
mtctx->jobs[u].jobID = u;
mtctx->jobs[u].firstJob = (u==0);
mtctx->jobs[u].lastJob = (u==nbJobs-1);
DEBUGLOG(5, "ZSTDMT_compress_advanced_internal: posting job %u (%u bytes)", u, (U32)jobSize);
DEBUG_PRINTHEX(6, mtctx->jobs[u].prefix.start, 12);
POOL_add(mtctx->factory, ZSTDMT_compressionJob, &mtctx->jobs[u]);
frameStartPos += jobSize;
dstBufferPos += dstBufferCapacity;
remainingSrcSize -= jobSize;
} }
/* collect result */
{ size_t error = 0, dstPos = 0;
unsigned jobID;
for (jobID=0; jobID<nbJobs; jobID++) {
DEBUGLOG(5, "waiting for job %u ", jobID);
ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[jobID].job_mutex);
while (mtctx->jobs[jobID].consumed < mtctx->jobs[jobID].src.size) {
DEBUGLOG(5, "waiting for jobCompleted signal from job %u", jobID);
ZSTD_pthread_cond_wait(&mtctx->jobs[jobID].job_cond, &mtctx->jobs[jobID].job_mutex);
}
ZSTD_pthread_mutex_unlock(&mtctx->jobs[jobID].job_mutex);
DEBUGLOG(5, "ready to write job %u ", jobID);
{ size_t const cSize = mtctx->jobs[jobID].cSize;
if (ZSTD_isError(cSize)) error = cSize;
if ((!error) && (dstPos + cSize > dstCapacity)) error = ERROR(dstSize_tooSmall);
if (jobID) { /* note : job 0 is written directly at dst, which is correct position */
if (!error)
memmove((char*)dst + dstPos, mtctx->jobs[jobID].dstBuff.start, cSize); /* may overlap when job compressed within dst */
if (jobID >= compressWithinDst) { /* job compressed into its own buffer, which must be released */
DEBUGLOG(5, "releasing buffer %u>=%u", jobID, compressWithinDst);
ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
} }
mtctx->jobs[jobID].dstBuff = g_nullBuffer;
mtctx->jobs[jobID].cSize = 0;
dstPos += cSize ;
}
} /* for (jobID=0; jobID<nbJobs; jobID++) */
DEBUGLOG(4, "checksumFlag : %u ", params.fParams.checksumFlag);
if (params.fParams.checksumFlag) {
U32 const checksum = (U32)XXH64_digest(&mtctx->serial.xxhState);
if (dstPos + 4 > dstCapacity) {
error = ERROR(dstSize_tooSmall);
} else {
DEBUGLOG(4, "writing checksum : %08X \n", checksum);
MEM_writeLE32((char*)dst + dstPos, checksum);
dstPos += 4;
} }
if (!error) DEBUGLOG(4, "compressed size : %u ", (U32)dstPos);
return error ? error : dstPos;
}
}
size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const ZSTD_CDict* cdict,
ZSTD_parameters params,
int overlapLog)
{
ZSTD_CCtx_params cctxParams = mtctx->params;
cctxParams.cParams = params.cParams;
cctxParams.fParams = params.fParams;
assert(ZSTD_OVERLAPLOG_MIN <= overlapLog && overlapLog <= ZSTD_OVERLAPLOG_MAX);
cctxParams.overlapLog = overlapLog;
return ZSTDMT_compress_advanced_internal(mtctx,
dst, dstCapacity,
src, srcSize,
cdict, cctxParams);
}
size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
int compressionLevel)
{
ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);
int const overlapLog = ZSTDMT_overlapLog_default(params.cParams.strategy);
params.fParams.contentSizeFlag = 1;
return ZSTDMT_compress_advanced(mtctx, dst, dstCapacity, src, srcSize, NULL, params, overlapLog);
}
/* ====================================== */
/* ======= Streaming API ======= */
/* ====================================== */
@@ -1432,16 +1217,6 @@ size_t ZSTDMT_initCStream_internal(
if (params.jobSize != 0 && params.jobSize < ZSTDMT_JOBSIZE_MIN) params.jobSize = ZSTDMT_JOBSIZE_MIN;
if (params.jobSize > (size_t)ZSTDMT_JOBSIZE_MAX) params.jobSize = (size_t)ZSTDMT_JOBSIZE_MAX;
mtctx->singleBlockingThread = (pledgedSrcSize <= ZSTDMT_JOBSIZE_MIN); /* do not trigger multi-threading when srcSize is too small */
if (mtctx->singleBlockingThread) {
ZSTD_CCtx_params const singleThreadParams = ZSTDMT_initJobCCtxParams(&params);
DEBUGLOG(5, "ZSTDMT_initCStream_internal: switch to single blocking thread mode");
assert(singleThreadParams.nbWorkers == 0);
return ZSTD_initCStream_internal(mtctx->cctxPool->cctx[0],
dict, dictSize, cdict,
&singleThreadParams, pledgedSrcSize);
}
DEBUGLOG(4, "ZSTDMT_initCStream_internal: %u workers", params.nbWorkers);
if (mtctx->allJobsCompleted == 0) { /* previous compression not correctly finished */
@@ -1475,9 +1250,8 @@ size_t ZSTDMT_initCStream_internal(
if (params.rsyncable) {
/* Aim for the targetsectionSize as the average job size. */
U32 const jobSizeMB = (U32)(mtctx->targetSectionSize >> 20);
U32 const rsyncBits = ZSTD_highbit32(jobSizeMB) + 20;
assert(jobSizeMB >= 1);
U32 const jobSizeKB = (U32)(mtctx->targetSectionSize >> 10);
U32 const rsyncBits = (assert(jobSizeKB >= 1), ZSTD_highbit32(jobSizeKB) + 10);
DEBUGLOG(4, "rsyncLog = %u", rsyncBits);
mtctx->rsync.hash = 0;
mtctx->rsync.hitMask = (1ULL << rsyncBits) - 1;
@@ -1504,8 +1278,8 @@ size_t ZSTDMT_initCStream_internal(
size_t const capacity = MAX(windowSize, sectionsSize) + slackSize;
if (mtctx->roundBuff.capacity < capacity) {
if (mtctx->roundBuff.buffer)
ZSTD_free(mtctx->roundBuff.buffer, mtctx->cMem);
mtctx->roundBuff.buffer = (BYTE*)ZSTD_malloc(capacity, mtctx->cMem);
ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
mtctx->roundBuff.buffer = (BYTE*)ZSTD_customMalloc(capacity, mtctx->cMem);
if (mtctx->roundBuff.buffer == NULL) {
mtctx->roundBuff.capacity = 0;
return ERROR(memory_allocation);
@@ -1530,53 +1304,6 @@ size_t ZSTDMT_initCStream_internal(
return 0;
}
size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
const void* dict, size_t dictSize,
ZSTD_parameters params,
unsigned long long pledgedSrcSize)
{
ZSTD_CCtx_params cctxParams = mtctx->params; /* retrieve sticky params */
DEBUGLOG(4, "ZSTDMT_initCStream_advanced (pledgedSrcSize=%u)", (U32)pledgedSrcSize);
cctxParams.cParams = params.cParams;
cctxParams.fParams = params.fParams;
return ZSTDMT_initCStream_internal(mtctx, dict, dictSize, ZSTD_dct_auto, NULL,
cctxParams, pledgedSrcSize);
}
size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
const ZSTD_CDict* cdict,
ZSTD_frameParameters fParams,
unsigned long long pledgedSrcSize)
{
ZSTD_CCtx_params cctxParams = mtctx->params;
if (cdict==NULL) return ERROR(dictionary_wrong); /* method incompatible with NULL cdict */
cctxParams.cParams = ZSTD_getCParamsFromCDict(cdict);
cctxParams.fParams = fParams;
return ZSTDMT_initCStream_internal(mtctx, NULL, 0 /*dictSize*/, ZSTD_dct_auto, cdict,
cctxParams, pledgedSrcSize);
}
/* ZSTDMT_resetCStream() :
* pledgedSrcSize can be zero == unknown (for the time being)
* prefer using ZSTD_CONTENTSIZE_UNKNOWN,
* as `0` might mean "empty" in the future */
size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize)
{
if (!pledgedSrcSize) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
return ZSTDMT_initCStream_internal(mtctx, NULL, 0, ZSTD_dct_auto, 0, mtctx->params,
pledgedSrcSize);
}
size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel) {
ZSTD_parameters const params = ZSTD_getParams(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0);
ZSTD_CCtx_params cctxParams = mtctx->params; /* retrieve sticky params */
DEBUGLOG(4, "ZSTDMT_initCStream (cLevel=%i)", compressionLevel);
cctxParams.cParams = params.cParams;
cctxParams.fParams = params.fParams;
return ZSTDMT_initCStream_internal(mtctx, NULL, 0, ZSTD_dct_auto, NULL, cctxParams, ZSTD_CONTENTSIZE_UNKNOWN);
}
/* ZSTDMT_writeLastEmptyBlock()
* Write a single empty block with an end-of-frame to finish a frame.
@@ -1740,7 +1467,7 @@ static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, u
assert(cSize >= mtctx->jobs[wJobID].dstFlushed);
assert(mtctx->jobs[wJobID].dstBuff.start != NULL);
if (toFlush > 0) {
memcpy((char*)output->dst + output->pos,
ZSTD_memcpy((char*)output->dst + output->pos,
(const char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].dstFlushed,
toFlush);
}
@@ -1894,7 +1621,7 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
return 0;
}
ZSTDMT_waitForLdmComplete(mtctx, buffer);
memmove(start, mtctx->inBuff.prefix.start, prefixSize);
ZSTD_memmove(start, mtctx->inBuff.prefix.start, prefixSize);
mtctx->inBuff.prefix.start = start;
mtctx->roundBuff.pos = prefixSize;
}
@@ -1968,6 +1695,16 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
pos = 0;
prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
if ((hash & hitMask) == hitMask) {
/* We're already at a sync point so don't load any more until
* we're able to flush this sync point.
* This likely happened because the job table was full so we
* couldn't add our job.
*/
syncPoint.toLoad = 0;
syncPoint.flush = 1;
return syncPoint;
}
} else {
/* We don't have enough bytes buffered to initialize the hash, but
* we know we have at least RSYNC_LENGTH bytes total.
@@ -2022,34 +1759,11 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
assert(output->pos <= output->size);
assert(input->pos <= input->size);
if (mtctx->singleBlockingThread) { /* delegate to single-thread (synchronous) */
return ZSTD_compressStream2(mtctx->cctxPool->cctx[0], output, input, endOp);
}
if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) {
/* current frame being ended. Only flush/end are allowed */
return ERROR(stage_wrong);
}
/* single-pass shortcut (note : synchronous-mode) */
if ( (!mtctx->params.rsyncable) /* rsyncable mode is disabled */
&& (mtctx->nextJobID == 0) /* just started */
&& (mtctx->inBuff.filled == 0) /* nothing buffered */
&& (!mtctx->jobReady) /* no job already created */
&& (endOp == ZSTD_e_end) /* end order */
&& (output->size - output->pos >= ZSTD_compressBound(input->size - input->pos)) ) { /* enough space in dst */
size_t const cSize = ZSTDMT_compress_advanced_internal(mtctx,
(char*)output->dst + output->pos, output->size - output->pos,
(const char*)input->src + input->pos, input->size - input->pos,
mtctx->cdict, mtctx->params);
if (ZSTD_isError(cSize)) return cSize;
input->pos = input->size;
output->pos += cSize;
mtctx->allJobsCompleted = 1;
mtctx->frameEnded = 1;
return 0;
}
/* fill input buffer */
if ( (!mtctx->jobReady)
&& (input->size > input->pos) ) { /* support NULL input */
@@ -2072,13 +1786,21 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
assert(mtctx->inBuff.buffer.capacity >= mtctx->targetSectionSize);
DEBUGLOG(5, "ZSTDMT_compressStream_generic: adding %u bytes on top of %u to buffer of size %u",
(U32)syncPoint.toLoad, (U32)mtctx->inBuff.filled, (U32)mtctx->targetSectionSize);
memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, syncPoint.toLoad);
ZSTD_memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, syncPoint.toLoad);
input->pos += syncPoint.toLoad;
mtctx->inBuff.filled += syncPoint.toLoad;
forwardInputProgress = syncPoint.toLoad>0;
}
if ((input->pos < input->size) && (endOp == ZSTD_e_end))
endOp = ZSTD_e_flush; /* can't end now : not all input consumed */
}
if ((input->pos < input->size) && (endOp == ZSTD_e_end)) {
/* Can't end yet because the input is not fully consumed.
* We are in one of these cases:
* - mtctx->inBuff is NULL & empty: we couldn't get an input buffer so don't create a new job.
* - We filled the input buffer: flush this job but don't end the frame.
* - We hit a synchronization point: flush this job but don't end the frame.
*/
assert(mtctx->inBuff.filled == 0 || mtctx->inBuff.filled == mtctx->targetSectionSize || mtctx->params.rsyncable);
endOp = ZSTD_e_flush;
}
if ( (mtctx->jobReady)
@@ -2097,47 +1819,3 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
return remainingToFlush;
}
}
size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
{
FORWARD_IF_ERROR( ZSTDMT_compressStream_generic(mtctx, output, input, ZSTD_e_continue) , "");
/* recommended next input size : fill current input buffer */
return mtctx->targetSectionSize - mtctx->inBuff.filled; /* note : could be zero when input buffer is fully filled and no more availability to create new job */
}
static size_t ZSTDMT_flushStream_internal(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_EndDirective endFrame)
{
size_t const srcSize = mtctx->inBuff.filled;
DEBUGLOG(5, "ZSTDMT_flushStream_internal");
if ( mtctx->jobReady /* one job ready for a worker to pick up */
|| (srcSize > 0) /* still some data within input buffer */
|| ((endFrame==ZSTD_e_end) && !mtctx->frameEnded)) { /* need a last 0-size block to end frame */
DEBUGLOG(5, "ZSTDMT_flushStream_internal : create a new job (%u bytes, end:%u)",
(U32)srcSize, (U32)endFrame);
FORWARD_IF_ERROR( ZSTDMT_createCompressionJob(mtctx, srcSize, endFrame) , "");
}
/* check if there is any data available to flush */
return ZSTDMT_flushProduced(mtctx, output, 1 /* blockToFlush */, endFrame);
}
size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
{
DEBUGLOG(5, "ZSTDMT_flushStream");
if (mtctx->singleBlockingThread)
return ZSTD_flushStream(mtctx->cctxPool->cctx[0], output);
return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_flush);
}
size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
{
DEBUGLOG(4, "ZSTDMT_endStream");
if (mtctx->singleBlockingThread)
return ZSTD_endStream(mtctx->cctxPool->cctx[0], output);
return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_end);
}
Utilities/cmzstd/lib/compress/zstdmt_compress.h
+31 -113
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -19,113 +19,57 @@
/* Note : This is an internal API.
* These APIs used to be exposed with ZSTDLIB_API,
* because it used to be the only way to invoke MT compression.
* Now, it's recommended to use ZSTD_compress2 and ZSTD_compressStream2()
* instead.
*
* If you depend on these APIs and can't switch, then define
* ZSTD_LEGACY_MULTITHREADED_API when making the dynamic library.
* However, we may completely remove these functions in a future
* release, so please switch soon.
* Now, you must use ZSTD_compress2 and ZSTD_compressStream2() instead.
*
* This API requires ZSTD_MULTITHREAD to be defined during compilation,
* otherwise ZSTDMT_createCCtx*() will fail.
*/
#ifdef ZSTD_LEGACY_MULTITHREADED_API
# define ZSTDMT_API ZSTDLIB_API
#else
# define ZSTDMT_API
#endif
/* === Dependencies === */
#include <stddef.h> /* size_t */
#include "../common/zstd_deps.h" /* size_t */
#define ZSTD_STATIC_LINKING_ONLY /* ZSTD_parameters */
#include "../zstd.h" /* ZSTD_inBuffer, ZSTD_outBuffer, ZSTDLIB_API */
/* === Constants === */
#ifndef ZSTDMT_NBWORKERS_MAX
# define ZSTDMT_NBWORKERS_MAX 200
#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
# define ZSTDMT_JOBSIZE_MIN (1 MB)
#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_API ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbWorkers);
/* Requires ZSTD_MULTITHREAD to be defined during compilation, otherwise it will return NULL. */
ZSTDMT_API ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers,
ZSTD_customMem cMem);
ZSTDMT_API size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx);
ZSTDMT_API size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx);
/* === Simple one-pass compression function === */
ZSTDMT_API size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
int compressionLevel);
ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers,
ZSTD_customMem cMem,
ZSTD_threadPool *pool);
size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx);
size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx);
/* === Streaming functions === */
ZSTDMT_API size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel);
ZSTDMT_API size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize); /**< if srcSize is not known at reset time, use ZSTD_CONTENTSIZE_UNKNOWN. Note: for compatibility with older programs, 0 means the same as ZSTD_CONTENTSIZE_UNKNOWN, but it will change in the future to mean "empty" */
ZSTDMT_API size_t ZSTDMT_nextInputSizeHint(const ZSTDMT_CCtx* mtctx);
ZSTDMT_API size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
ZSTDMT_API size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output); /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
ZSTDMT_API size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output); /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
/* === Advanced functions and parameters === */
ZSTDMT_API size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const ZSTD_CDict* cdict,
ZSTD_parameters params,
int overlapLog);
ZSTDMT_API size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
const void* dict, size_t dictSize, /* dict can be released after init, a local copy is preserved within zcs */
ZSTD_parameters params,
unsigned long long pledgedSrcSize); /* pledgedSrcSize is optional and can be zero == unknown */
ZSTDMT_API size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
const ZSTD_CDict* cdict,
ZSTD_frameParameters fparams,
unsigned long long pledgedSrcSize); /* note : zero means empty */
/* ZSTDMT_parameter :
* List of parameters that can be set using ZSTDMT_setMTCtxParameter() */
typedef enum {
ZSTDMT_p_jobSize, /* Each job is compressed in parallel. By default, this value is dynamically determined depending on compression parameters. Can be set explicitly here. */
ZSTDMT_p_overlapLog, /* Each job may reload a part of previous job to enhance compression ratio; 0 == no overlap, 6(default) == use 1/8th of window, >=9 == use full window. This is a "sticky" parameter : its value will be re-used on next compression job */
ZSTDMT_p_rsyncable /* Enables rsyncable mode. */
} ZSTDMT_parameter;
/* ZSTDMT_setMTCtxParameter() :
* allow setting individual parameters, one at a time, among a list of enums defined in ZSTDMT_parameter.
* The function must be called typically after ZSTD_createCCtx() but __before ZSTDMT_init*() !__
* Parameters not explicitly reset by ZSTDMT_init*() remain the same in consecutive compression sessions.
* @return : 0, or an error code (which can be tested using ZSTD_isError()) */
ZSTDMT_API size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int value);
/* ZSTDMT_getMTCtxParameter() :
* Query the ZSTDMT_CCtx for a parameter value.
* @return : 0, or an error code (which can be tested using ZSTD_isError()) */
ZSTDMT_API size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int* value);
size_t ZSTDMT_nextInputSizeHint(const ZSTDMT_CCtx* mtctx);
/*! ZSTDMT_initCStream_internal() :
* Private use only. Init streaming operation.
* expects params to be valid.
* must receive dict, or cdict, or none, but not both.
* @return : 0, or an error code */
size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
const ZSTD_CDict* cdict,
ZSTD_CCtx_params params, unsigned long long pledgedSrcSize);
/*! ZSTDMT_compressStream_generic() :
* Combines ZSTDMT_compressStream() with optional ZSTDMT_flushStream() or ZSTDMT_endStream()
@@ -134,16 +78,10 @@ ZSTDMT_API size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter
* 0 if fully flushed
* or an error code
* note : needs to be init using any ZSTD_initCStream*() variant */
ZSTDMT_API size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
ZSTD_outBuffer* output,
ZSTD_inBuffer* input,
ZSTD_EndDirective endOp);
/* ========================================================
* === Private interface, for use by ZSTD_compress.c ===
* === Not exposed in libzstd. Never invoke directly ===
* ======================================================== */
size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
ZSTD_outBuffer* output,
ZSTD_inBuffer* input,
ZSTD_EndDirective endOp);
/*! ZSTDMT_toFlushNow()
* Tell how many bytes are ready to be flushed immediately.
@@ -153,15 +91,6 @@ ZSTDMT_API size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
* therefore flushing is limited by speed of oldest job. */
size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx);
/*! ZSTDMT_CCtxParam_setMTCtxParameter()
* like ZSTDMT_setMTCtxParameter(), but into a ZSTD_CCtx_Params */
size_t ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params, ZSTDMT_parameter parameter, int value);
/*! ZSTDMT_CCtxParam_setNbWorkers()
* Set nbWorkers, and clamp it.
* Also reset jobSize and overlapLog */
size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers);
/*! ZSTDMT_updateCParams_whileCompressing() :
* Updates only a selected set of compression parameters, to remain compatible with current frame.
* New parameters will be applied to next compression job. */
@@ -174,17 +103,6 @@ void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_p
ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx);
/*! ZSTDMT_initCStream_internal() :
* Private use only. Init streaming operation.
* expects params to be valid.
* must receive dict, or cdict, or none, but not both.
* @return : 0, or an error code */
size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
const ZSTD_CDict* cdict,
ZSTD_CCtx_params params, unsigned long long pledgedSrcSize);
#if defined (__cplusplus)
}
#endif
Utilities/cmzstd/lib/decompress/huf_decompress.c
+340 -237
View File
@@ -1,7 +1,7 @@
/* ******************************************************************
* huff0 huffman decoder,
* part of Finite State Entropy library
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
*
* You can contact the author at :
* - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -15,7 +15,7 @@
/* **************************************************************
* Dependencies
****************************************************************/
#include <string.h> /* memcpy, memset */
#include "../common/zstd_deps.h" /* ZSTD_memcpy, ZSTD_memset */
#include "../common/compiler.h"
#include "../common/bitstream.h" /* BIT_* */
#include "../common/fse.h" /* to compress headers */
@@ -103,7 +103,7 @@ typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved;
static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
{
DTableDesc dtd;
memcpy(&dtd, table, sizeof(dtd));
ZSTD_memcpy(&dtd, table, sizeof(dtd));
return dtd;
}
@@ -115,29 +115,51 @@ static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
/*-***************************/
typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX1; /* single-symbol decoding */
/**
* Packs 4 HUF_DEltX1 structs into a U64. This is used to lay down 4 entries at
* a time.
*/
static U64 HUF_DEltX1_set4(BYTE symbol, BYTE nbBits) {
U64 D4;
if (MEM_isLittleEndian()) {
D4 = symbol + (nbBits << 8);
} else {
D4 = (symbol << 8) + nbBits;
}
D4 *= 0x0001000100010001ULL;
return D4;
}
typedef struct {
U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];
U32 rankStart[HUF_TABLELOG_ABSOLUTEMAX + 1];
U32 statsWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
BYTE symbols[HUF_SYMBOLVALUE_MAX + 1];
BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];
} HUF_ReadDTableX1_Workspace;
size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
{
return HUF_readDTableX1_wksp_bmi2(DTable, src, srcSize, workSpace, wkspSize, /* bmi2 */ 0);
}
size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2)
{
U32 tableLog = 0;
U32 nbSymbols = 0;
size_t iSize;
void* const dtPtr = DTable + 1;
HUF_DEltX1* const dt = (HUF_DEltX1*)dtPtr;
HUF_ReadDTableX1_Workspace* wksp = (HUF_ReadDTableX1_Workspace*)workSpace;
U32* rankVal;
BYTE* huffWeight;
size_t spaceUsed32 = 0;
rankVal = (U32 *)workSpace + spaceUsed32;
spaceUsed32 += HUF_TABLELOG_ABSOLUTEMAX + 1;
huffWeight = (BYTE *)((U32 *)workSpace + spaceUsed32);
spaceUsed32 += HUF_ALIGN(HUF_SYMBOLVALUE_MAX + 1, sizeof(U32)) >> 2;
if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
DEBUG_STATIC_ASSERT(HUF_DECOMPRESS_WORKSPACE_SIZE >= sizeof(*wksp));
if (sizeof(*wksp) > wkspSize) return ERROR(tableLog_tooLarge);
DEBUG_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
/* memset(huffWeight, 0, sizeof(huffWeight)); */ /* is not necessary, even though some analyzer complain ... */
/* ZSTD_memset(huffWeight, 0, sizeof(huffWeight)); */ /* is not necessary, even though some analyzer complain ... */
iSize = HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp), bmi2);
if (HUF_isError(iSize)) return iSize;
/* Table header */
@@ -145,52 +167,117 @@ size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize
if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge); /* DTable too small, Huffman tree cannot fit in */
dtd.tableType = 0;
dtd.tableLog = (BYTE)tableLog;
memcpy(DTable, &dtd, sizeof(dtd));
ZSTD_memcpy(DTable, &dtd, sizeof(dtd));
}
/* Calculate starting value for each rank */
{ U32 n, nextRankStart = 0;
for (n=1; n<tableLog+1; n++) {
U32 const current = nextRankStart;
nextRankStart += (rankVal[n] << (n-1));
rankVal[n] = current;
} }
/* Compute symbols and rankStart given rankVal:
*
* rankVal already contains the number of values of each weight.
*
* symbols contains the symbols ordered by weight. First are the rankVal[0]
* weight 0 symbols, followed by the rankVal[1] weight 1 symbols, and so on.
* symbols[0] is filled (but unused) to avoid a branch.
*
* rankStart contains the offset where each rank belongs in the DTable.
* rankStart[0] is not filled because there are no entries in the table for
* weight 0.
*/
{
int n;
int nextRankStart = 0;
int const unroll = 4;
int const nLimit = (int)nbSymbols - unroll + 1;
for (n=0; n<(int)tableLog+1; n++) {
U32 const curr = nextRankStart;
nextRankStart += wksp->rankVal[n];
wksp->rankStart[n] = curr;
}
for (n=0; n < nLimit; n += unroll) {
int u;
for (u=0; u < unroll; ++u) {
size_t const w = wksp->huffWeight[n+u];
wksp->symbols[wksp->rankStart[w]++] = (BYTE)(n+u);
}
}
for (; n < (int)nbSymbols; ++n) {
size_t const w = wksp->huffWeight[n];
wksp->symbols[wksp->rankStart[w]++] = (BYTE)n;
}
}
/* fill DTable */
{ U32 n;
size_t const nEnd = nbSymbols;
for (n=0; n<nEnd; n++) {
size_t const w = huffWeight[n];
size_t const length = (1 << w) >> 1;
size_t const uStart = rankVal[w];
size_t const uEnd = uStart + length;
size_t u;
HUF_DEltX1 D;
D.byte = (BYTE)n;
D.nbBits = (BYTE)(tableLog + 1 - w);
rankVal[w] = (U32)uEnd;
if (length < 4) {
/* Use length in the loop bound so the compiler knows it is short. */
for (u = 0; u < length; ++u)
dt[uStart + u] = D;
} else {
/* Unroll the loop 4 times, we know it is a power of 2. */
for (u = uStart; u < uEnd; u += 4) {
dt[u + 0] = D;
dt[u + 1] = D;
dt[u + 2] = D;
dt[u + 3] = D;
} } } }
/* fill DTable
* We fill all entries of each weight in order.
* That way length is a constant for each iteration of the outter loop.
* We can switch based on the length to a different inner loop which is
* optimized for that particular case.
*/
{
U32 w;
int symbol=wksp->rankVal[0];
int rankStart=0;
for (w=1; w<tableLog+1; ++w) {
int const symbolCount = wksp->rankVal[w];
int const length = (1 << w) >> 1;
int uStart = rankStart;
BYTE const nbBits = (BYTE)(tableLog + 1 - w);
int s;
int u;
switch (length) {
case 1:
for (s=0; s<symbolCount; ++s) {
HUF_DEltX1 D;
D.byte = wksp->symbols[symbol + s];
D.nbBits = nbBits;
dt[uStart] = D;
uStart += 1;
}
break;
case 2:
for (s=0; s<symbolCount; ++s) {
HUF_DEltX1 D;
D.byte = wksp->symbols[symbol + s];
D.nbBits = nbBits;
dt[uStart+0] = D;
dt[uStart+1] = D;
uStart += 2;
}
break;
case 4:
for (s=0; s<symbolCount; ++s) {
U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
MEM_write64(dt + uStart, D4);
uStart += 4;
}
break;
case 8:
for (s=0; s<symbolCount; ++s) {
U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
MEM_write64(dt + uStart, D4);
MEM_write64(dt + uStart + 4, D4);
uStart += 8;
}
break;
default:
for (s=0; s<symbolCount; ++s) {
U64 const D4 = HUF_DEltX1_set4(wksp->symbols[symbol + s], nbBits);
for (u=0; u < length; u += 16) {
MEM_write64(dt + uStart + u + 0, D4);
MEM_write64(dt + uStart + u + 4, D4);
MEM_write64(dt + uStart + u + 8, D4);
MEM_write64(dt + uStart + u + 12, D4);
}
assert(u == length);
uStart += length;
}
break;
}
symbol += symbolCount;
rankStart += symbolCount * length;
}
}
return iSize;
}
size_t HUF_readDTableX1(HUF_DTable* DTable, const void* src, size_t srcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_readDTableX1_wksp(DTable, src, srcSize,
workSpace, sizeof(workSpace));
}
FORCE_INLINE_TEMPLATE BYTE
HUF_decodeSymbolX1(BIT_DStream_t* Dstream, const HUF_DEltX1* dt, const U32 dtLog)
{
@@ -389,20 +476,6 @@ size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
}
size_t HUF_decompress1X1_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress1X1_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
return HUF_decompress1X1_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
}
size_t HUF_decompress4X1_usingDTable(
void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize,
@@ -419,8 +492,7 @@ static size_t HUF_decompress4X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size
{
const BYTE* ip = (const BYTE*) cSrc;
size_t const hSize = HUF_readDTableX1_wksp (dctx, cSrc, cSrcSize,
workSpace, wkspSize);
size_t const hSize = HUF_readDTableX1_wksp_bmi2(dctx, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
if (HUF_isError(hSize)) return hSize;
if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
ip += hSize; cSrcSize -= hSize;
@@ -436,18 +508,6 @@ size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
}
size_t HUF_decompress4X1_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
return HUF_decompress4X1_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
}
#endif /* HUF_FORCE_DECOMPRESS_X2 */
@@ -468,13 +528,15 @@ typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 sizeLog, const U32 consumed,
const U32* rankValOrigin, const int minWeight,
const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
U32 nbBitsBaseline, U16 baseSeq)
U32 nbBitsBaseline, U16 baseSeq, U32* wksp, size_t wkspSize)
{
HUF_DEltX2 DElt;
U32 rankVal[HUF_TABLELOG_MAX + 1];
U32* rankVal = wksp;
assert(wkspSize >= HUF_TABLELOG_MAX + 1);
(void)wkspSize;
/* get pre-calculated rankVal */
memcpy(rankVal, rankValOrigin, sizeof(rankVal));
ZSTD_memcpy(rankVal, rankValOrigin, sizeof(U32) * (HUF_TABLELOG_MAX + 1));
/* fill skipped values */
if (minWeight>1) {
@@ -509,14 +571,18 @@ static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 sizeLog, const U32 co
static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
const sortedSymbol_t* sortedList, const U32 sortedListSize,
const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
const U32 nbBitsBaseline)
const U32 nbBitsBaseline, U32* wksp, size_t wkspSize)
{
U32 rankVal[HUF_TABLELOG_MAX + 1];
U32* rankVal = wksp;
const int scaleLog = nbBitsBaseline - targetLog; /* note : targetLog >= srcLog, hence scaleLog <= 1 */
const U32 minBits = nbBitsBaseline - maxWeight;
U32 s;
memcpy(rankVal, rankValOrigin, sizeof(rankVal));
assert(wkspSize >= HUF_TABLELOG_MAX + 1);
wksp += HUF_TABLELOG_MAX + 1;
wkspSize -= HUF_TABLELOG_MAX + 1;
ZSTD_memcpy(rankVal, rankValOrigin, sizeof(U32) * (HUF_TABLELOG_MAX + 1));
/* fill DTable */
for (s=0; s<sortedListSize; s++) {
@@ -534,7 +600,7 @@ static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
HUF_fillDTableX2Level2(DTable+start, targetLog-nbBits, nbBits,
rankValOrigin[nbBits], minWeight,
sortedList+sortedRank, sortedListSize-sortedRank,
nbBitsBaseline, symbol);
nbBitsBaseline, symbol, wksp, wkspSize);
} else {
HUF_DEltX2 DElt;
MEM_writeLE16(&(DElt.sequence), symbol);
@@ -548,6 +614,15 @@ static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
}
}
typedef struct {
rankValCol_t rankVal[HUF_TABLELOG_MAX];
U32 rankStats[HUF_TABLELOG_MAX + 1];
U32 rankStart0[HUF_TABLELOG_MAX + 2];
sortedSymbol_t sortedSymbol[HUF_SYMBOLVALUE_MAX + 1];
BYTE weightList[HUF_SYMBOLVALUE_MAX + 1];
U32 calleeWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
} HUF_ReadDTableX2_Workspace;
size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
const void* src, size_t srcSize,
void* workSpace, size_t wkspSize)
@@ -560,48 +635,33 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
U32 *rankStart;
rankValCol_t* rankVal;
U32* rankStats;
U32* rankStart0;
sortedSymbol_t* sortedSymbol;
BYTE* weightList;
size_t spaceUsed32 = 0;
HUF_ReadDTableX2_Workspace* const wksp = (HUF_ReadDTableX2_Workspace*)workSpace;
rankVal = (rankValCol_t *)((U32 *)workSpace + spaceUsed32);
spaceUsed32 += (sizeof(rankValCol_t) * HUF_TABLELOG_MAX) >> 2;
rankStats = (U32 *)workSpace + spaceUsed32;
spaceUsed32 += HUF_TABLELOG_MAX + 1;
rankStart0 = (U32 *)workSpace + spaceUsed32;
spaceUsed32 += HUF_TABLELOG_MAX + 2;
sortedSymbol = (sortedSymbol_t *)workSpace + (spaceUsed32 * sizeof(U32)) / sizeof(sortedSymbol_t);
spaceUsed32 += HUF_ALIGN(sizeof(sortedSymbol_t) * (HUF_SYMBOLVALUE_MAX + 1), sizeof(U32)) >> 2;
weightList = (BYTE *)((U32 *)workSpace + spaceUsed32);
spaceUsed32 += HUF_ALIGN(HUF_SYMBOLVALUE_MAX + 1, sizeof(U32)) >> 2;
if (sizeof(*wksp) > wkspSize) return ERROR(GENERIC);
if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
rankStart = rankStart0 + 1;
memset(rankStats, 0, sizeof(U32) * (2 * HUF_TABLELOG_MAX + 2 + 1));
rankStart = wksp->rankStart0 + 1;
ZSTD_memset(wksp->rankStats, 0, sizeof(wksp->rankStats));
ZSTD_memset(wksp->rankStart0, 0, sizeof(wksp->rankStart0));
DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(HUF_DTable)); /* if compiler fails here, assertion is wrong */
if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
/* memset(weightList, 0, sizeof(weightList)); */ /* is not necessary, even though some analyzer complain ... */
/* ZSTD_memset(weightList, 0, sizeof(weightList)); */ /* is not necessary, even though some analyzer complain ... */
iSize = HUF_readStats(weightList, HUF_SYMBOLVALUE_MAX + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
iSize = HUF_readStats_wksp(wksp->weightList, HUF_SYMBOLVALUE_MAX + 1, wksp->rankStats, &nbSymbols, &tableLog, src, srcSize, wksp->calleeWksp, sizeof(wksp->calleeWksp), /* bmi2 */ 0);
if (HUF_isError(iSize)) return iSize;
/* check result */
if (tableLog > maxTableLog) return ERROR(tableLog_tooLarge); /* DTable can't fit code depth */
/* find maxWeight */
for (maxW = tableLog; rankStats[maxW]==0; maxW--) {} /* necessarily finds a solution before 0 */
for (maxW = tableLog; wksp->rankStats[maxW]==0; maxW--) {} /* necessarily finds a solution before 0 */
/* Get start index of each weight */
{ U32 w, nextRankStart = 0;
for (w=1; w<maxW+1; w++) {
U32 current = nextRankStart;
nextRankStart += rankStats[w];
rankStart[w] = current;
U32 curr = nextRankStart;
nextRankStart += wksp->rankStats[w];
rankStart[w] = curr;
}
rankStart[0] = nextRankStart; /* put all 0w symbols at the end of sorted list*/
sizeOfSort = nextRankStart;
@@ -610,57 +670,51 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
/* sort symbols by weight */
{ U32 s;
for (s=0; s<nbSymbols; s++) {
U32 const w = weightList[s];
U32 const w = wksp->weightList[s];
U32 const r = rankStart[w]++;
sortedSymbol[r].symbol = (BYTE)s;
sortedSymbol[r].weight = (BYTE)w;
wksp->sortedSymbol[r].symbol = (BYTE)s;
wksp->sortedSymbol[r].weight = (BYTE)w;
}
rankStart[0] = 0; /* forget 0w symbols; this is beginning of weight(1) */
}
/* Build rankVal */
{ U32* const rankVal0 = rankVal[0];
{ U32* const rankVal0 = wksp->rankVal[0];
{ int const rescale = (maxTableLog-tableLog) - 1; /* tableLog <= maxTableLog */
U32 nextRankVal = 0;
U32 w;
for (w=1; w<maxW+1; w++) {
U32 current = nextRankVal;
nextRankVal += rankStats[w] << (w+rescale);
rankVal0[w] = current;
U32 curr = nextRankVal;
nextRankVal += wksp->rankStats[w] << (w+rescale);
rankVal0[w] = curr;
} }
{ U32 const minBits = tableLog+1 - maxW;
U32 consumed;
for (consumed = minBits; consumed < maxTableLog - minBits + 1; consumed++) {
U32* const rankValPtr = rankVal[consumed];
U32* const rankValPtr = wksp->rankVal[consumed];
U32 w;
for (w = 1; w < maxW+1; w++) {
rankValPtr[w] = rankVal0[w] >> consumed;
} } } }
HUF_fillDTableX2(dt, maxTableLog,
sortedSymbol, sizeOfSort,
rankStart0, rankVal, maxW,
tableLog+1);
wksp->sortedSymbol, sizeOfSort,
wksp->rankStart0, wksp->rankVal, maxW,
tableLog+1,
wksp->calleeWksp, sizeof(wksp->calleeWksp) / sizeof(U32));
dtd.tableLog = (BYTE)maxTableLog;
dtd.tableType = 1;
memcpy(DTable, &dtd, sizeof(dtd));
ZSTD_memcpy(DTable, &dtd, sizeof(dtd));
return iSize;
}
size_t HUF_readDTableX2(HUF_DTable* DTable, const void* src, size_t srcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_readDTableX2_wksp(DTable, src, srcSize,
workSpace, sizeof(workSpace));
}
FORCE_INLINE_TEMPLATE U32
HUF_decodeSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
{
size_t const val = BIT_lookBitsFast(DStream, dtLog); /* note : dtLog >= 1 */
memcpy(op, dt+val, 2);
ZSTD_memcpy(op, dt+val, 2);
BIT_skipBits(DStream, dt[val].nbBits);
return dt[val].length;
}
@@ -669,7 +723,7 @@ FORCE_INLINE_TEMPLATE U32
HUF_decodeLastSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
{
size_t const val = BIT_lookBitsFast(DStream, dtLog); /* note : dtLog >= 1 */
memcpy(op, dt+val, 1);
ZSTD_memcpy(op, dt+val, 1);
if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
else {
if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
@@ -890,20 +944,6 @@ size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
}
size_t HUF_decompress1X2_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress1X2_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
return HUF_decompress1X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
}
size_t HUF_decompress4X2_usingDTable(
void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize,
@@ -937,20 +977,6 @@ size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
}
size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
}
#endif /* HUF_FORCE_DECOMPRESS_X1 */
@@ -1051,67 +1077,6 @@ U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
}
typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
static const decompressionAlgo decompress[2] = { HUF_decompress4X1, HUF_decompress4X2 };
#endif
/* validation checks */
if (dstSize == 0) return ERROR(dstSize_tooSmall);
if (cSrcSize > dstSize) return ERROR(corruption_detected); /* invalid */
if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; } /* not compressed */
if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; } /* RLE */
{ U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
#if defined(HUF_FORCE_DECOMPRESS_X1)
(void)algoNb;
assert(algoNb == 0);
return HUF_decompress4X1(dst, dstSize, cSrc, cSrcSize);
#elif defined(HUF_FORCE_DECOMPRESS_X2)
(void)algoNb;
assert(algoNb == 1);
return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);
#else
return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
#endif
}
}
size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
/* validation checks */
if (dstSize == 0) return ERROR(dstSize_tooSmall);
if (cSrcSize > dstSize) return ERROR(corruption_detected); /* invalid */
if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; } /* not compressed */
if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; } /* RLE */
{ U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
#if defined(HUF_FORCE_DECOMPRESS_X1)
(void)algoNb;
assert(algoNb == 0);
return HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
#elif defined(HUF_FORCE_DECOMPRESS_X2)
(void)algoNb;
assert(algoNb == 1);
return HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
#else
return algoNb ? HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
#endif
}
}
size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress4X_hufOnly_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst,
size_t dstSize, const void* cSrc,
size_t cSrcSize, void* workSpace,
@@ -1145,8 +1110,8 @@ size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
/* validation checks */
if (dstSize == 0) return ERROR(dstSize_tooSmall);
if (cSrcSize > dstSize) return ERROR(corruption_detected); /* invalid */
if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; } /* not compressed */
if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; } /* RLE */
if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; } /* not compressed */
if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; } /* RLE */
{ U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
#if defined(HUF_FORCE_DECOMPRESS_X1)
@@ -1168,14 +1133,6 @@ size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
}
}
size_t HUF_decompress1X_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress1X_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
{
@@ -1199,7 +1156,7 @@ size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstS
{
const BYTE* ip = (const BYTE*) cSrc;
size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize);
size_t const hSize = HUF_readDTableX1_wksp_bmi2(dctx, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
if (HUF_isError(hSize)) return hSize;
if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
ip += hSize; cSrcSize -= hSize;
@@ -1246,3 +1203,149 @@ size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t ds
#endif
}
}
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
#ifndef HUF_FORCE_DECOMPRESS_X2
size_t HUF_readDTableX1(HUF_DTable* DTable, const void* src, size_t srcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_readDTableX1_wksp(DTable, src, srcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress1X1_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress1X1_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
return HUF_decompress1X1_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
}
#endif
#ifndef HUF_FORCE_DECOMPRESS_X1
size_t HUF_readDTableX2(HUF_DTable* DTable, const void* src, size_t srcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_readDTableX2_wksp(DTable, src, srcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress1X2_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress1X2_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
return HUF_decompress1X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
}
#endif
#ifndef HUF_FORCE_DECOMPRESS_X2
size_t HUF_decompress4X1_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
return HUF_decompress4X1_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
}
#endif
#ifndef HUF_FORCE_DECOMPRESS_X1
size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
}
#endif
typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
static const decompressionAlgo decompress[2] = { HUF_decompress4X1, HUF_decompress4X2 };
#endif
/* validation checks */
if (dstSize == 0) return ERROR(dstSize_tooSmall);
if (cSrcSize > dstSize) return ERROR(corruption_detected); /* invalid */
if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; } /* not compressed */
if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; } /* RLE */
{ U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
#if defined(HUF_FORCE_DECOMPRESS_X1)
(void)algoNb;
assert(algoNb == 0);
return HUF_decompress4X1(dst, dstSize, cSrc, cSrcSize);
#elif defined(HUF_FORCE_DECOMPRESS_X2)
(void)algoNb;
assert(algoNb == 1);
return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);
#else
return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
#endif
}
}
size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
/* validation checks */
if (dstSize == 0) return ERROR(dstSize_tooSmall);
if (cSrcSize > dstSize) return ERROR(corruption_detected); /* invalid */
if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; } /* not compressed */
if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; } /* RLE */
{ U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
#if defined(HUF_FORCE_DECOMPRESS_X1)
(void)algoNb;
assert(algoNb == 0);
return HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
#elif defined(HUF_FORCE_DECOMPRESS_X2)
(void)algoNb;
assert(algoNb == 1);
return HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
#else
return algoNb ? HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
#endif
}
}
size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress4X_hufOnly_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
size_t HUF_decompress1X_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
const void* cSrc, size_t cSrcSize)
{
U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
return HUF_decompress1X_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
workSpace, sizeof(workSpace));
}
#endif
Utilities/cmzstd/lib/decompress/zstd_ddict.c
+9 -9
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -14,7 +14,7 @@
/*-*******************************************************
* Dependencies
*********************************************************/
#include <string.h> /* memcpy, memmove, memset */
#include "../common/zstd_deps.h" /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
#include "../common/cpu.h" /* bmi2 */
#include "../common/mem.h" /* low level memory routines */
#define FSE_STATIC_LINKING_ONLY
@@ -127,11 +127,11 @@ static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
ddict->dictContent = dict;
if (!dict) dictSize = 0;
} else {
void* const internalBuffer = ZSTD_malloc(dictSize, ddict->cMem);
void* const internalBuffer = ZSTD_customMalloc(dictSize, ddict->cMem);
ddict->dictBuffer = internalBuffer;
ddict->dictContent = internalBuffer;
if (!internalBuffer) return ERROR(memory_allocation);
memcpy(internalBuffer, dict, dictSize);
ZSTD_memcpy(internalBuffer, dict, dictSize);
}
ddict->dictSize = dictSize;
ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001); /* cover both little and big endian */
@@ -147,9 +147,9 @@ ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
ZSTD_dictContentType_e dictContentType,
ZSTD_customMem customMem)
{
if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
{ ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
{ ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_customMalloc(sizeof(ZSTD_DDict), customMem);
if (ddict == NULL) return NULL;
ddict->cMem = customMem;
{ size_t const initResult = ZSTD_initDDict_internal(ddict,
@@ -198,7 +198,7 @@ const ZSTD_DDict* ZSTD_initStaticDDict(
if ((size_t)sBuffer & 7) return NULL; /* 8-aligned */
if (sBufferSize < neededSpace) return NULL;
if (dictLoadMethod == ZSTD_dlm_byCopy) {
memcpy(ddict+1, dict, dictSize); /* local copy */
ZSTD_memcpy(ddict+1, dict, dictSize); /* local copy */
dict = ddict+1;
}
if (ZSTD_isError( ZSTD_initDDict_internal(ddict,
@@ -213,8 +213,8 @@ size_t ZSTD_freeDDict(ZSTD_DDict* ddict)
{
if (ddict==NULL) return 0; /* support free on NULL */
{ ZSTD_customMem const cMem = ddict->cMem;
ZSTD_free(ddict->dictBuffer, cMem);
ZSTD_free(ddict, cMem);
ZSTD_customFree(ddict->dictBuffer, cMem);
ZSTD_customFree(ddict, cMem);
return 0;
}
}
Utilities/cmzstd/lib/decompress/zstd_ddict.h
+2 -2
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -15,7 +15,7 @@
/*-*******************************************************
* Dependencies
*********************************************************/
#include <stddef.h> /* size_t */
#include "../common/zstd_deps.h" /* size_t */
#include "../zstd.h" /* ZSTD_DDict, and several public functions */
Utilities/cmzstd/lib/decompress/zstd_decompress.c
+369 -87
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -55,13 +55,14 @@
/*-*******************************************************
* Dependencies
*********************************************************/
#include <string.h> /* memcpy, memmove, memset */
#include "../common/zstd_deps.h" /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
#include "../common/cpu.h" /* bmi2 */
#include "../common/mem.h" /* low level memory routines */
#define FSE_STATIC_LINKING_ONLY
#include "../common/fse.h"
#define HUF_STATIC_LINKING_ONLY
#include "../common/huf.h"
#include "../common/xxhash.h" /* XXH64_reset, XXH64_update, XXH64_digest, XXH64 */
#include "../common/zstd_internal.h" /* blockProperties_t */
#include "zstd_decompress_internal.h" /* ZSTD_DCtx */
#include "zstd_ddict.h" /* ZSTD_DDictDictContent */
@@ -72,6 +73,144 @@
#endif
/*************************************
* Multiple DDicts Hashset internals *
*************************************/
#define DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT 4
#define DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT 3 /* These two constants represent SIZE_MULT/COUNT_MULT load factor without using a float.
* Currently, that means a 0.75 load factor.
* So, if count * COUNT_MULT / size * SIZE_MULT != 0, then we've exceeded
* the load factor of the ddict hash set.
*/
#define DDICT_HASHSET_TABLE_BASE_SIZE 64
#define DDICT_HASHSET_RESIZE_FACTOR 2
/* Hash function to determine starting position of dict insertion within the table
* Returns an index between [0, hashSet->ddictPtrTableSize]
*/
static size_t ZSTD_DDictHashSet_getIndex(const ZSTD_DDictHashSet* hashSet, U32 dictID) {
const U64 hash = XXH64(&dictID, sizeof(U32), 0);
/* DDict ptr table size is a multiple of 2, use size - 1 as mask to get index within [0, hashSet->ddictPtrTableSize) */
return hash & (hashSet->ddictPtrTableSize - 1);
}
/* Adds DDict to a hashset without resizing it.
* If inserting a DDict with a dictID that already exists in the set, replaces the one in the set.
* Returns 0 if successful, or a zstd error code if something went wrong.
*/
static size_t ZSTD_DDictHashSet_emplaceDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict) {
const U32 dictID = ZSTD_getDictID_fromDDict(ddict);
size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
RETURN_ERROR_IF(hashSet->ddictPtrCount == hashSet->ddictPtrTableSize, GENERIC, "Hash set is full!");
DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
while (hashSet->ddictPtrTable[idx] != NULL) {
/* Replace existing ddict if inserting ddict with same dictID */
if (ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]) == dictID) {
DEBUGLOG(4, "DictID already exists, replacing rather than adding");
hashSet->ddictPtrTable[idx] = ddict;
return 0;
}
idx &= idxRangeMask;
idx++;
}
DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
hashSet->ddictPtrTable[idx] = ddict;
hashSet->ddictPtrCount++;
return 0;
}
/* Expands hash table by factor of DDICT_HASHSET_RESIZE_FACTOR and
* rehashes all values, allocates new table, frees old table.
* Returns 0 on success, otherwise a zstd error code.
*/
static size_t ZSTD_DDictHashSet_expand(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
size_t newTableSize = hashSet->ddictPtrTableSize * DDICT_HASHSET_RESIZE_FACTOR;
const ZSTD_DDict** newTable = (const ZSTD_DDict**)ZSTD_customCalloc(sizeof(ZSTD_DDict*) * newTableSize, customMem);
const ZSTD_DDict** oldTable = hashSet->ddictPtrTable;
size_t oldTableSize = hashSet->ddictPtrTableSize;
size_t i;
DEBUGLOG(4, "Expanding DDict hash table! Old size: %zu new size: %zu", oldTableSize, newTableSize);
RETURN_ERROR_IF(!newTable, memory_allocation, "Expanded hashset allocation failed!");
hashSet->ddictPtrTable = newTable;
hashSet->ddictPtrTableSize = newTableSize;
hashSet->ddictPtrCount = 0;
for (i = 0; i < oldTableSize; ++i) {
if (oldTable[i] != NULL) {
FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, oldTable[i]), "");
}
}
ZSTD_customFree((void*)oldTable, customMem);
DEBUGLOG(4, "Finished re-hash");
return 0;
}
/* Fetches a DDict with the given dictID
* Returns the ZSTD_DDict* with the requested dictID. If it doesn't exist, then returns NULL.
*/
static const ZSTD_DDict* ZSTD_DDictHashSet_getDDict(ZSTD_DDictHashSet* hashSet, U32 dictID) {
size_t idx = ZSTD_DDictHashSet_getIndex(hashSet, dictID);
const size_t idxRangeMask = hashSet->ddictPtrTableSize - 1;
DEBUGLOG(4, "Hashed index: for dictID: %u is %zu", dictID, idx);
for (;;) {
size_t currDictID = ZSTD_getDictID_fromDDict(hashSet->ddictPtrTable[idx]);
if (currDictID == dictID || currDictID == 0) {
/* currDictID == 0 implies a NULL ddict entry */
break;
} else {
idx &= idxRangeMask; /* Goes to start of table when we reach the end */
idx++;
}
}
DEBUGLOG(4, "Final idx after probing for dictID %u is: %zu", dictID, idx);
return hashSet->ddictPtrTable[idx];
}
/* Allocates space for and returns a ddict hash set
* The hash set's ZSTD_DDict* table has all values automatically set to NULL to begin with.
* Returns NULL if allocation failed.
*/
static ZSTD_DDictHashSet* ZSTD_createDDictHashSet(ZSTD_customMem customMem) {
ZSTD_DDictHashSet* ret = (ZSTD_DDictHashSet*)ZSTD_customMalloc(sizeof(ZSTD_DDictHashSet), customMem);
DEBUGLOG(4, "Allocating new hash set");
ret->ddictPtrTable = (const ZSTD_DDict**)ZSTD_customCalloc(DDICT_HASHSET_TABLE_BASE_SIZE * sizeof(ZSTD_DDict*), customMem);
ret->ddictPtrTableSize = DDICT_HASHSET_TABLE_BASE_SIZE;
ret->ddictPtrCount = 0;
if (!ret || !ret->ddictPtrTable) {
return NULL;
}
return ret;
}
/* Frees the table of ZSTD_DDict* within a hashset, then frees the hashset itself.
* Note: The ZSTD_DDict* within the table are NOT freed.
*/
static void ZSTD_freeDDictHashSet(ZSTD_DDictHashSet* hashSet, ZSTD_customMem customMem) {
DEBUGLOG(4, "Freeing ddict hash set");
if (hashSet && hashSet->ddictPtrTable) {
ZSTD_customFree((void*)hashSet->ddictPtrTable, customMem);
}
if (hashSet) {
ZSTD_customFree(hashSet, customMem);
}
}
/* Public function: Adds a DDict into the ZSTD_DDictHashSet, possibly triggering a resize of the hash set.
* Returns 0 on success, or a ZSTD error.
*/
static size_t ZSTD_DDictHashSet_addDDict(ZSTD_DDictHashSet* hashSet, const ZSTD_DDict* ddict, ZSTD_customMem customMem) {
DEBUGLOG(4, "Adding dict ID: %u to hashset with - Count: %zu Tablesize: %zu", ZSTD_getDictID_fromDDict(ddict), hashSet->ddictPtrCount, hashSet->ddictPtrTableSize);
if (hashSet->ddictPtrCount * DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT / hashSet->ddictPtrTableSize * DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT != 0) {
FORWARD_IF_ERROR(ZSTD_DDictHashSet_expand(hashSet, customMem), "");
}
FORWARD_IF_ERROR(ZSTD_DDictHashSet_emplaceDDict(hashSet, ddict), "");
return 0;
}
/*-*************************************************************
* Context management
***************************************************************/
@@ -94,11 +233,19 @@ static size_t ZSTD_startingInputLength(ZSTD_format_e format)
return startingInputLength;
}
static void ZSTD_DCtx_resetParameters(ZSTD_DCtx* dctx)
{
assert(dctx->streamStage == zdss_init);
dctx->format = ZSTD_f_zstd1;
dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
dctx->outBufferMode = ZSTD_bm_buffered;
dctx->forceIgnoreChecksum = ZSTD_d_validateChecksum;
dctx->refMultipleDDicts = ZSTD_rmd_refSingleDDict;
}
static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
{
dctx->format = ZSTD_f_zstd1; /* ZSTD_decompressBegin() invokes ZSTD_startingInputLength() with argument dctx->format */
dctx->staticSize = 0;
dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
dctx->ddict = NULL;
dctx->ddictLocal = NULL;
dctx->dictEnd = NULL;
@@ -113,7 +260,8 @@ static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
dctx->noForwardProgress = 0;
dctx->oversizedDuration = 0;
dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
dctx->outBufferMode = ZSTD_obm_buffered;
dctx->ddictSet = NULL;
ZSTD_DCtx_resetParameters(dctx);
#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
dctx->dictContentEndForFuzzing = NULL;
#endif
@@ -134,9 +282,9 @@ ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
{
if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
{ ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_malloc(sizeof(*dctx), customMem);
{ ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_customMalloc(sizeof(*dctx), customMem);
if (!dctx) return NULL;
dctx->customMem = customMem;
ZSTD_initDCtx_internal(dctx);
@@ -164,13 +312,17 @@ size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
RETURN_ERROR_IF(dctx->staticSize, memory_allocation, "not compatible with static DCtx");
{ ZSTD_customMem const cMem = dctx->customMem;
ZSTD_clearDict(dctx);
ZSTD_free(dctx->inBuff, cMem);
ZSTD_customFree(dctx->inBuff, cMem);
dctx->inBuff = NULL;
#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
if (dctx->legacyContext)
ZSTD_freeLegacyStreamContext(dctx->legacyContext, dctx->previousLegacyVersion);
#endif
ZSTD_free(dctx, cMem);
if (dctx->ddictSet) {
ZSTD_freeDDictHashSet(dctx->ddictSet, cMem);
dctx->ddictSet = NULL;
}
ZSTD_customFree(dctx, cMem);
return 0;
}
}
@@ -179,7 +331,30 @@ size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
{
size_t const toCopy = (size_t)((char*)(&dstDCtx->inBuff) - (char*)dstDCtx);
memcpy(dstDCtx, srcDCtx, toCopy); /* no need to copy workspace */
ZSTD_memcpy(dstDCtx, srcDCtx, toCopy); /* no need to copy workspace */
}
/* Given a dctx with a digested frame params, re-selects the correct ZSTD_DDict based on
* the requested dict ID from the frame. If there exists a reference to the correct ZSTD_DDict, then
* accordingly sets the ddict to be used to decompress the frame.
*
* If no DDict is found, then no action is taken, and the ZSTD_DCtx::ddict remains as-is.
*
* ZSTD_d_refMultipleDDicts must be enabled for this function to be called.
*/
static void ZSTD_DCtx_selectFrameDDict(ZSTD_DCtx* dctx) {
assert(dctx->refMultipleDDicts && dctx->ddictSet);
DEBUGLOG(4, "Adjusting DDict based on requested dict ID from frame");
if (dctx->ddict) {
const ZSTD_DDict* frameDDict = ZSTD_DDictHashSet_getDDict(dctx->ddictSet, dctx->fParams.dictID);
if (frameDDict) {
DEBUGLOG(4, "DDict found!");
ZSTD_clearDict(dctx);
dctx->dictID = dctx->fParams.dictID;
dctx->ddict = frameDDict;
dctx->dictUses = ZSTD_use_indefinitely;
}
}
}
@@ -246,7 +421,7 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
const BYTE* ip = (const BYTE*)src;
size_t const minInputSize = ZSTD_startingInputLength(format);
memset(zfhPtr, 0, sizeof(*zfhPtr)); /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr)); /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
if (srcSize < minInputSize) return minInputSize;
RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter");
@@ -256,7 +431,7 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
/* skippable frame */
if (srcSize < ZSTD_SKIPPABLEHEADERSIZE)
return ZSTD_SKIPPABLEHEADERSIZE; /* magic number + frame length */
memset(zfhPtr, 0, sizeof(*zfhPtr));
ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));
zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_FRAMEIDSIZE);
zfhPtr->frameType = ZSTD_skippableFrame;
return 0;
@@ -433,12 +608,19 @@ unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
/** ZSTD_decodeFrameHeader() :
* `headerSize` must be the size provided by ZSTD_frameHeaderSize().
* If multiple DDict references are enabled, also will choose the correct DDict to use.
* @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
{
size_t const result = ZSTD_getFrameHeader_advanced(&(dctx->fParams), src, headerSize, dctx->format);
if (ZSTD_isError(result)) return result; /* invalid header */
RETURN_ERROR_IF(result>0, srcSize_wrong, "headerSize too small");
/* Reference DDict requested by frame if dctx references multiple ddicts */
if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts && dctx->ddictSet) {
ZSTD_DCtx_selectFrameDDict(dctx);
}
#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
/* Skip the dictID check in fuzzing mode, because it makes the search
* harder.
@@ -446,7 +628,9 @@ static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t he
RETURN_ERROR_IF(dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID),
dictionary_wrong, "");
#endif
if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
dctx->validateChecksum = (dctx->fParams.checksumFlag && !dctx->forceIgnoreChecksum) ? 1 : 0;
if (dctx->validateChecksum) XXH64_reset(&dctx->xxhState, 0);
dctx->processedCSize += headerSize;
return 0;
}
@@ -461,7 +645,7 @@ static ZSTD_frameSizeInfo ZSTD_errorFrameSizeInfo(size_t ret)
static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize)
{
ZSTD_frameSizeInfo frameSizeInfo;
memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
ZSTD_memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
if (ZSTD_isLegacy(src, srcSize))
@@ -516,7 +700,7 @@ static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize
ip += 4;
}
frameSizeInfo.compressedSize = ip - ipstart;
frameSizeInfo.compressedSize = (size_t)(ip - ipstart);
frameSizeInfo.decompressedBound = (zfh.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN)
? zfh.frameContentSize
: nbBlocks * zfh.blockSizeMax;
@@ -569,7 +753,7 @@ unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize)
{
DEBUGLOG(5, "ZSTD_insertBlock: %u bytes", (unsigned)blockSize);
ZSTD_checkContinuity(dctx, blockStart);
ZSTD_checkContinuity(dctx, blockStart, blockSize);
dctx->previousDstEnd = (const char*)blockStart + blockSize;
return blockSize;
}
@@ -579,12 +763,12 @@ static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
const void* src, size_t srcSize)
{
DEBUGLOG(5, "ZSTD_copyRawBlock");
RETURN_ERROR_IF(srcSize > dstCapacity, dstSize_tooSmall, "");
if (dst == NULL) {
if (srcSize == 0) return 0;
RETURN_ERROR(dstBuffer_null, "");
}
RETURN_ERROR_IF(srcSize > dstCapacity, dstSize_tooSmall, "");
memcpy(dst, src, srcSize);
ZSTD_memcpy(dst, src, srcSize);
return srcSize;
}
@@ -592,15 +776,41 @@ static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
BYTE b,
size_t regenSize)
{
RETURN_ERROR_IF(regenSize > dstCapacity, dstSize_tooSmall, "");
if (dst == NULL) {
if (regenSize == 0) return 0;
RETURN_ERROR(dstBuffer_null, "");
}
RETURN_ERROR_IF(regenSize > dstCapacity, dstSize_tooSmall, "");
memset(dst, b, regenSize);
ZSTD_memset(dst, b, regenSize);
return regenSize;
}
static void ZSTD_DCtx_trace_end(ZSTD_DCtx const* dctx, U64 uncompressedSize, U64 compressedSize, unsigned streaming)
{
#if ZSTD_TRACE
if (dctx->traceCtx && ZSTD_trace_decompress_end != NULL) {
ZSTD_Trace trace;
ZSTD_memset(&trace, 0, sizeof(trace));
trace.version = ZSTD_VERSION_NUMBER;
trace.streaming = streaming;
if (dctx->ddict) {
trace.dictionaryID = ZSTD_getDictID_fromDDict(dctx->ddict);
trace.dictionarySize = ZSTD_DDict_dictSize(dctx->ddict);
trace.dictionaryIsCold = dctx->ddictIsCold;
}
trace.uncompressedSize = (size_t)uncompressedSize;
trace.compressedSize = (size_t)compressedSize;
trace.dctx = dctx;
ZSTD_trace_decompress_end(dctx->traceCtx, &trace);
}
#else
(void)dctx;
(void)uncompressedSize;
(void)compressedSize;
(void)streaming;
#endif
}
/*! ZSTD_decompressFrame() :
* @dctx must be properly initialized
@@ -610,8 +820,9 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
void* dst, size_t dstCapacity,
const void** srcPtr, size_t *srcSizePtr)
{
const BYTE* ip = (const BYTE*)(*srcPtr);
BYTE* const ostart = (BYTE* const)dst;
const BYTE* const istart = (const BYTE*)(*srcPtr);
const BYTE* ip = istart;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = dstCapacity != 0 ? ostart + dstCapacity : ostart;
BYTE* op = ostart;
size_t remainingSrcSize = *srcSizePtr;
@@ -647,13 +858,13 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
switch(blockProperties.blockType)
{
case bt_compressed:
decodedSize = ZSTD_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize, /* frame */ 1);
decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oend-op), ip, cBlockSize, /* frame */ 1);
break;
case bt_raw :
decodedSize = ZSTD_copyRawBlock(op, oend-op, ip, cBlockSize);
decodedSize = ZSTD_copyRawBlock(op, (size_t)(oend-op), ip, cBlockSize);
break;
case bt_rle :
decodedSize = ZSTD_setRleBlock(op, oend-op, *ip, blockProperties.origSize);
decodedSize = ZSTD_setRleBlock(op, (size_t)(oend-op), *ip, blockProperties.origSize);
break;
case bt_reserved :
default:
@@ -661,7 +872,7 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
}
if (ZSTD_isError(decodedSize)) return decodedSize;
if (dctx->fParams.checksumFlag)
if (dctx->validateChecksum)
XXH64_update(&dctx->xxhState, op, decodedSize);
if (decodedSize != 0)
op += decodedSize;
@@ -676,19 +887,21 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
corruption_detected, "");
}
if (dctx->fParams.checksumFlag) { /* Frame content checksum verification */
U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
U32 checkRead;
RETURN_ERROR_IF(remainingSrcSize<4, checksum_wrong, "");
checkRead = MEM_readLE32(ip);
RETURN_ERROR_IF(checkRead != checkCalc, checksum_wrong, "");
if (!dctx->forceIgnoreChecksum) {
U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
U32 checkRead;
checkRead = MEM_readLE32(ip);
RETURN_ERROR_IF(checkRead != checkCalc, checksum_wrong, "");
}
ip += 4;
remainingSrcSize -= 4;
}
ZSTD_DCtx_trace_end(dctx, (U64)(op-ostart), (U64)(ip-istart), /* streaming */ 0);
/* Allow caller to get size read */
*srcPtr = ip;
*srcSizePtr = remainingSrcSize;
return op-ostart;
return (size_t)(op-ostart);
}
static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
@@ -721,7 +934,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
decodedSize = ZSTD_decompressLegacy(dst, dstCapacity, src, frameSize, dict, dictSize);
if (ZSTD_isError(decodedSize)) return decodedSize;
assert(decodedSize <=- dstCapacity);
assert(decodedSize <= dstCapacity);
dst = (BYTE*)dst + decodedSize;
dstCapacity -= decodedSize;
@@ -753,7 +966,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
* use this in all cases but ddict */
FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize), "");
}
ZSTD_checkContinuity(dctx, dst);
ZSTD_checkContinuity(dctx, dst, dstCapacity);
{ const size_t res = ZSTD_decompressFrame(dctx, dst, dstCapacity,
&src, &srcSize);
@@ -761,15 +974,13 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
(ZSTD_getErrorCode(res) == ZSTD_error_prefix_unknown)
&& (moreThan1Frame==1),
srcSize_wrong,
"at least one frame successfully completed, but following "
"bytes are garbage: it's more likely to be a srcSize error, "
"specifying more bytes than compressed size of frame(s). This "
"error message replaces ERROR(prefix_unknown), which would be "
"confusing, as the first header is actually correct. Note that "
"one could be unlucky, it might be a corruption error instead, "
"happening right at the place where we expect zstd magic "
"bytes. But this is _much_ less likely than a srcSize field "
"error.");
"At least one frame successfully completed, "
"but following bytes are garbage: "
"it's more likely to be a srcSize error, "
"specifying more input bytes than size of frame(s). "
"Note: one could be unlucky, it might be a corruption error instead, "
"happening right at the place where we expect zstd magic bytes. "
"But this is _much_ less likely than a srcSize field error.");
if (ZSTD_isError(res)) return res;
assert(res <= dstCapacity);
if (res != 0)
@@ -781,7 +992,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
RETURN_ERROR_IF(srcSize, srcSize_wrong, "input not entirely consumed");
return (BYTE*)dst - (BYTE*)dststart;
return (size_t)((BYTE*)dst - (BYTE*)dststart);
}
size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
@@ -890,7 +1101,9 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
DEBUGLOG(5, "ZSTD_decompressContinue (srcSize:%u)", (unsigned)srcSize);
/* Sanity check */
RETURN_ERROR_IF(srcSize != ZSTD_nextSrcSizeToDecompressWithInputSize(dctx, srcSize), srcSize_wrong, "not allowed");
if (dstCapacity) ZSTD_checkContinuity(dctx, dst);
ZSTD_checkContinuity(dctx, dst, dstCapacity);
dctx->processedCSize += srcSize;
switch (dctx->stage)
{
@@ -899,21 +1112,21 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
if (dctx->format == ZSTD_f_zstd1) { /* allows header */
assert(srcSize >= ZSTD_FRAMEIDSIZE); /* to read skippable magic number */
if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) { /* skippable frame */
memcpy(dctx->headerBuffer, src, srcSize);
ZSTD_memcpy(dctx->headerBuffer, src, srcSize);
dctx->expected = ZSTD_SKIPPABLEHEADERSIZE - srcSize; /* remaining to load to get full skippable frame header */
dctx->stage = ZSTDds_decodeSkippableHeader;
return 0;
} }
dctx->headerSize = ZSTD_frameHeaderSize_internal(src, srcSize, dctx->format);
if (ZSTD_isError(dctx->headerSize)) return dctx->headerSize;
memcpy(dctx->headerBuffer, src, srcSize);
ZSTD_memcpy(dctx->headerBuffer, src, srcSize);
dctx->expected = dctx->headerSize - srcSize;
dctx->stage = ZSTDds_decodeFrameHeader;
return 0;
case ZSTDds_decodeFrameHeader:
assert(src != NULL);
memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
ZSTD_memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize), "");
dctx->expected = ZSTD_blockHeaderSize;
dctx->stage = ZSTDds_decodeBlockHeader;
@@ -977,7 +1190,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
RETURN_ERROR_IF(rSize > dctx->fParams.blockSizeMax, corruption_detected, "Decompressed Block Size Exceeds Maximum");
DEBUGLOG(5, "ZSTD_decompressContinue: decoded size from block : %u", (unsigned)rSize);
dctx->decodedSize += rSize;
if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, dst, rSize);
if (dctx->validateChecksum) XXH64_update(&dctx->xxhState, dst, rSize);
dctx->previousDstEnd = (char*)dst + rSize;
/* Stay on the same stage until we are finished streaming the block. */
@@ -995,6 +1208,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
dctx->expected = 4;
dctx->stage = ZSTDds_checkChecksum;
} else {
ZSTD_DCtx_trace_end(dctx, dctx->decodedSize, dctx->processedCSize, /* streaming */ 1);
dctx->expected = 0; /* ends here */
dctx->stage = ZSTDds_getFrameHeaderSize;
}
@@ -1007,10 +1221,14 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
case ZSTDds_checkChecksum:
assert(srcSize == 4); /* guaranteed by dctx->expected */
{ U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
U32 const check32 = MEM_readLE32(src);
DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", (unsigned)h32, (unsigned)check32);
RETURN_ERROR_IF(check32 != h32, checksum_wrong, "");
{
if (dctx->validateChecksum) {
U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
U32 const check32 = MEM_readLE32(src);
DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", (unsigned)h32, (unsigned)check32);
RETURN_ERROR_IF(check32 != h32, checksum_wrong, "");
}
ZSTD_DCtx_trace_end(dctx, dctx->decodedSize, dctx->processedCSize, /* streaming */ 1);
dctx->expected = 0;
dctx->stage = ZSTDds_getFrameHeaderSize;
return 0;
@@ -1019,7 +1237,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
case ZSTDds_decodeSkippableHeader:
assert(src != NULL);
assert(srcSize <= ZSTD_SKIPPABLEHEADERSIZE);
memcpy(dctx->headerBuffer + (ZSTD_SKIPPABLEHEADERSIZE - srcSize), src, srcSize); /* complete skippable header */
ZSTD_memcpy(dctx->headerBuffer + (ZSTD_SKIPPABLEHEADERSIZE - srcSize), src, srcSize); /* complete skippable header */
dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_FRAMEIDSIZE); /* note : dctx->expected can grow seriously large, beyond local buffer size */
dctx->stage = ZSTDds_skipFrame;
return 0;
@@ -1075,7 +1293,7 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
workspace, workspaceSize);
#else
size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
dictPtr, dictEnd - dictPtr,
dictPtr, (size_t)(dictEnd - dictPtr),
workspace, workspaceSize);
#endif
RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted, "");
@@ -1084,40 +1302,46 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
{ short offcodeNCount[MaxOff+1];
unsigned offcodeMaxValue = MaxOff, offcodeLog;
size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, (size_t)(dictEnd-dictPtr));
RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
RETURN_ERROR_IF(offcodeMaxValue > MaxOff, dictionary_corrupted, "");
RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
ZSTD_buildFSETable( entropy->OFTable,
offcodeNCount, offcodeMaxValue,
OF_base, OF_bits,
offcodeLog);
offcodeLog,
entropy->workspace, sizeof(entropy->workspace),
/* bmi2 */0);
dictPtr += offcodeHeaderSize;
}
{ short matchlengthNCount[MaxML+1];
unsigned matchlengthMaxValue = MaxML, matchlengthLog;
size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
RETURN_ERROR_IF(matchlengthMaxValue > MaxML, dictionary_corrupted, "");
RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
ZSTD_buildFSETable( entropy->MLTable,
matchlengthNCount, matchlengthMaxValue,
ML_base, ML_bits,
matchlengthLog);
matchlengthLog,
entropy->workspace, sizeof(entropy->workspace),
/* bmi2 */ 0);
dictPtr += matchlengthHeaderSize;
}
{ short litlengthNCount[MaxLL+1];
unsigned litlengthMaxValue = MaxLL, litlengthLog;
size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, (size_t)(dictEnd-dictPtr));
RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
RETURN_ERROR_IF(litlengthMaxValue > MaxLL, dictionary_corrupted, "");
RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
ZSTD_buildFSETable( entropy->LLTable,
litlengthNCount, litlengthMaxValue,
LL_base, LL_bits,
litlengthLog);
litlengthLog,
entropy->workspace, sizeof(entropy->workspace),
/* bmi2 */ 0);
dictPtr += litlengthHeaderSize;
}
@@ -1131,7 +1355,7 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
entropy->rep[i] = rep;
} }
return dictPtr - (const BYTE*)dict;
return (size_t)(dictPtr - (const BYTE*)dict);
}
static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
@@ -1158,8 +1382,12 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
{
assert(dctx != NULL);
#if ZSTD_TRACE
dctx->traceCtx = (ZSTD_trace_decompress_begin != NULL) ? ZSTD_trace_decompress_begin(dctx) : 0;
#endif
dctx->expected = ZSTD_startingInputLength(dctx->format); /* dctx->format must be properly set */
dctx->stage = ZSTDds_getFrameHeaderSize;
dctx->processedCSize = 0;
dctx->decodedSize = 0;
dctx->previousDstEnd = NULL;
dctx->prefixStart = NULL;
@@ -1170,7 +1398,7 @@ size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
dctx->dictID = 0;
dctx->bType = bt_reserved;
ZSTD_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue)); /* initial repcodes */
ZSTD_memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue)); /* initial repcodes */
dctx->LLTptr = dctx->entropy.LLTable;
dctx->MLTptr = dctx->entropy.MLTable;
dctx->OFTptr = dctx->entropy.OFTable;
@@ -1373,6 +1601,16 @@ size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
if (ddict) {
dctx->ddict = ddict;
dctx->dictUses = ZSTD_use_indefinitely;
if (dctx->refMultipleDDicts == ZSTD_rmd_refMultipleDDicts) {
if (dctx->ddictSet == NULL) {
dctx->ddictSet = ZSTD_createDDictHashSet(dctx->customMem);
if (!dctx->ddictSet) {
RETURN_ERROR(memory_allocation, "Failed to allocate memory for hash set!");
}
}
assert(!dctx->staticSize); /* Impossible: ddictSet cannot have been allocated if static dctx */
FORWARD_IF_ERROR(ZSTD_DDictHashSet_addDDict(dctx->ddictSet, ddict, dctx->customMem), "");
}
}
return 0;
}
@@ -1394,7 +1632,7 @@ size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format)
{
return ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, format);
return ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, (int)format);
}
ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
@@ -1411,8 +1649,16 @@ ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
return bounds;
case ZSTD_d_stableOutBuffer:
bounds.lowerBound = (int)ZSTD_obm_buffered;
bounds.upperBound = (int)ZSTD_obm_stable;
bounds.lowerBound = (int)ZSTD_bm_buffered;
bounds.upperBound = (int)ZSTD_bm_stable;
return bounds;
case ZSTD_d_forceIgnoreChecksum:
bounds.lowerBound = (int)ZSTD_d_validateChecksum;
bounds.upperBound = (int)ZSTD_d_ignoreChecksum;
return bounds;
case ZSTD_d_refMultipleDDicts:
bounds.lowerBound = (int)ZSTD_rmd_refSingleDDict;
bounds.upperBound = (int)ZSTD_rmd_refMultipleDDicts;
return bounds;
default:;
}
@@ -1436,6 +1682,29 @@ static int ZSTD_dParam_withinBounds(ZSTD_dParameter dParam, int value)
RETURN_ERROR_IF(!ZSTD_dParam_withinBounds(p, v), parameter_outOfBound, ""); \
}
size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value)
{
switch (param) {
case ZSTD_d_windowLogMax:
*value = (int)ZSTD_highbit32((U32)dctx->maxWindowSize);
return 0;
case ZSTD_d_format:
*value = (int)dctx->format;
return 0;
case ZSTD_d_stableOutBuffer:
*value = (int)dctx->outBufferMode;
return 0;
case ZSTD_d_forceIgnoreChecksum:
*value = (int)dctx->forceIgnoreChecksum;
return 0;
case ZSTD_d_refMultipleDDicts:
*value = (int)dctx->refMultipleDDicts;
return 0;
default:;
}
RETURN_ERROR(parameter_unsupported, "");
}
size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value)
{
RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
@@ -1451,7 +1720,18 @@ size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value
return 0;
case ZSTD_d_stableOutBuffer:
CHECK_DBOUNDS(ZSTD_d_stableOutBuffer, value);
dctx->outBufferMode = (ZSTD_outBufferMode_e)value;
dctx->outBufferMode = (ZSTD_bufferMode_e)value;
return 0;
case ZSTD_d_forceIgnoreChecksum:
CHECK_DBOUNDS(ZSTD_d_forceIgnoreChecksum, value);
dctx->forceIgnoreChecksum = (ZSTD_forceIgnoreChecksum_e)value;
return 0;
case ZSTD_d_refMultipleDDicts:
CHECK_DBOUNDS(ZSTD_d_refMultipleDDicts, value);
if (dctx->staticSize != 0) {
RETURN_ERROR(parameter_unsupported, "Static dctx does not support multiple DDicts!");
}
dctx->refMultipleDDicts = (ZSTD_refMultipleDDicts_e)value;
return 0;
default:;
}
@@ -1469,8 +1749,7 @@ size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset)
|| (reset == ZSTD_reset_session_and_parameters) ) {
RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
ZSTD_clearDict(dctx);
dctx->format = ZSTD_f_zstd1;
dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
ZSTD_DCtx_resetParameters(dctx);
}
return 0;
}
@@ -1524,7 +1803,7 @@ static void ZSTD_DCtx_updateOversizedDuration(ZSTD_DStream* zds, size_t const ne
{
if (ZSTD_DCtx_isOverflow(zds, neededInBuffSize, neededOutBuffSize))
zds->oversizedDuration++;
else
else
zds->oversizedDuration = 0;
}
@@ -1538,7 +1817,7 @@ static size_t ZSTD_checkOutBuffer(ZSTD_DStream const* zds, ZSTD_outBuffer const*
{
ZSTD_outBuffer const expect = zds->expectedOutBuffer;
/* No requirement when ZSTD_obm_stable is not enabled. */
if (zds->outBufferMode != ZSTD_obm_stable)
if (zds->outBufferMode != ZSTD_bm_stable)
return 0;
/* Any buffer is allowed in zdss_init, this must be the same for every other call until
* the context is reset.
@@ -1548,7 +1827,7 @@ static size_t ZSTD_checkOutBuffer(ZSTD_DStream const* zds, ZSTD_outBuffer const*
/* The buffer must match our expectation exactly. */
if (expect.dst == output->dst && expect.pos == output->pos && expect.size == output->size)
return 0;
RETURN_ERROR(dstBuffer_wrong, "ZSTD_obm_stable enabled but output differs!");
RETURN_ERROR(dstBuffer_wrong, "ZSTD_d_stableOutBuffer enabled but output differs!");
}
/* Calls ZSTD_decompressContinue() with the right parameters for ZSTD_decompressStream()
@@ -1560,7 +1839,7 @@ static size_t ZSTD_decompressContinueStream(
ZSTD_DStream* zds, char** op, char* oend,
void const* src, size_t srcSize) {
int const isSkipFrame = ZSTD_isSkipFrame(zds);
if (zds->outBufferMode == ZSTD_obm_buffered) {
if (zds->outBufferMode == ZSTD_bm_buffered) {
size_t const dstSize = isSkipFrame ? 0 : zds->outBuffSize - zds->outStart;
size_t const decodedSize = ZSTD_decompressContinue(zds,
zds->outBuff + zds->outStart, dstSize, src, srcSize);
@@ -1573,14 +1852,14 @@ static size_t ZSTD_decompressContinueStream(
}
} else {
/* Write directly into the output buffer */
size_t const dstSize = isSkipFrame ? 0 : oend - *op;
size_t const dstSize = isSkipFrame ? 0 : (size_t)(oend - *op);
size_t const decodedSize = ZSTD_decompressContinue(zds, *op, dstSize, src, srcSize);
FORWARD_IF_ERROR(decodedSize, "");
*op += decodedSize;
/* Flushing is not needed. */
zds->streamStage = zdss_read;
assert(*op <= oend);
assert(zds->outBufferMode == ZSTD_obm_stable);
assert(zds->outBufferMode == ZSTD_bm_stable);
}
return 0;
}
@@ -1635,6 +1914,9 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
} }
#endif
{ size_t const hSize = ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
if (zds->refMultipleDDicts && zds->ddictSet) {
ZSTD_DCtx_selectFrameDDict(zds);
}
DEBUGLOG(5, "header size : %u", (U32)hSize);
if (ZSTD_isError(hSize)) {
#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
@@ -1663,14 +1945,14 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
assert(iend >= ip);
if (toLoad > remainingInput) { /* not enough input to load full header */
if (remainingInput > 0) {
memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
zds->lhSize += remainingInput;
}
input->pos = input->size;
return (MAX((size_t)ZSTD_FRAMEHEADERSIZE_MIN(zds->format), hSize) - zds->lhSize) + ZSTD_blockHeaderSize; /* remaining header bytes + next block header */
}
assert(ip != NULL);
memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
ZSTD_memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
break;
} }
@@ -1678,10 +1960,10 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
if (zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
&& zds->fParams.frameType != ZSTD_skippableFrame
&& (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
size_t const cSize = ZSTD_findFrameCompressedSize(istart, iend-istart);
size_t const cSize = ZSTD_findFrameCompressedSize(istart, (size_t)(iend-istart));
if (cSize <= (size_t)(iend-istart)) {
/* shortcut : using single-pass mode */
size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, oend-op, istart, cSize, ZSTD_getDDict(zds));
size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, (size_t)(oend-op), istart, cSize, ZSTD_getDDict(zds));
if (ZSTD_isError(decompressedSize)) return decompressedSize;
DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()")
ip = istart + cSize;
@@ -1693,7 +1975,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
} }
/* Check output buffer is large enough for ZSTD_odm_stable. */
if (zds->outBufferMode == ZSTD_obm_stable
if (zds->outBufferMode == ZSTD_bm_stable
&& zds->fParams.frameType != ZSTD_skippableFrame
&& zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
&& (U64)(size_t)(oend-op) < zds->fParams.frameContentSize) {
@@ -1723,7 +2005,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
/* Adapt buffer sizes to frame header instructions */
{ size_t const neededInBuffSize = MAX(zds->fParams.blockSizeMax, 4 /* frame checksum */);
size_t const neededOutBuffSize = zds->outBufferMode == ZSTD_obm_buffered
size_t const neededOutBuffSize = zds->outBufferMode == ZSTD_bm_buffered
? ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize)
: 0;
@@ -1731,7 +2013,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
{ int const tooSmall = (zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize);
int const tooLarge = ZSTD_DCtx_isOversizedTooLong(zds);
if (tooSmall || tooLarge) {
size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
DEBUGLOG(4, "inBuff : from %u to %u",
@@ -1745,10 +2027,10 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
bufferSize > zds->staticSize - sizeof(ZSTD_DCtx),
memory_allocation, "");
} else {
ZSTD_free(zds->inBuff, zds->customMem);
ZSTD_customFree(zds->inBuff, zds->customMem);
zds->inBuffSize = 0;
zds->outBuffSize = 0;
zds->inBuff = (char*)ZSTD_malloc(bufferSize, zds->customMem);
zds->inBuff = (char*)ZSTD_customMalloc(bufferSize, zds->customMem);
RETURN_ERROR_IF(zds->inBuff == NULL, memory_allocation, "");
}
zds->inBuffSize = neededInBuffSize;
@@ -1760,7 +2042,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
case zdss_read:
DEBUGLOG(5, "stage zdss_read");
{ size_t const neededInSize = ZSTD_nextSrcSizeToDecompressWithInputSize(zds, iend - ip);
{ size_t const neededInSize = ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip));
DEBUGLOG(5, "neededInSize = %u", (U32)neededInSize);
if (neededInSize==0) { /* end of frame */
zds->streamStage = zdss_init;
@@ -1790,7 +2072,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
RETURN_ERROR_IF(toLoad > zds->inBuffSize - zds->inPos,
corruption_detected,
"should never happen");
loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, iend-ip);
loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, (size_t)(iend-ip));
}
ip += loadedSize;
zds->inPos += loadedSize;
@@ -1804,7 +2086,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
}
case zdss_flush:
{ size_t const toFlushSize = zds->outEnd - zds->outStart;
size_t const flushedSize = ZSTD_limitCopy(op, oend-op, zds->outBuff + zds->outStart, toFlushSize);
size_t const flushedSize = ZSTD_limitCopy(op, (size_t)(oend-op), zds->outBuff + zds->outStart, toFlushSize);
op += flushedSize;
zds->outStart += flushedSize;
if (flushedSize == toFlushSize) { /* flush completed */
Utilities/cmzstd/lib/decompress/zstd_decompress_block.c
+191 -75
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -14,7 +14,7 @@
/*-*******************************************************
* Dependencies
*********************************************************/
#include <string.h> /* memcpy, memmove, memset */
#include "../common/zstd_deps.h" /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
#include "../common/compiler.h" /* prefetch */
#include "../common/cpu.h" /* bmi2 */
#include "../common/mem.h" /* low level memory routines */
@@ -44,7 +44,7 @@
/*_*******************************************************
* Memory operations
**********************************************************/
static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
static void ZSTD_copy4(void* dst, const void* src) { ZSTD_memcpy(dst, src, 4); }
/*-*************************************************************
@@ -166,7 +166,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
dctx->litSize = litSize;
dctx->litEntropy = 1;
if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
ZSTD_memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
return litCSize + lhSize;
}
@@ -191,10 +191,10 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) { /* risk reading beyond src buffer with wildcopy */
RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected, "");
memcpy(dctx->litBuffer, istart+lhSize, litSize);
ZSTD_memcpy(dctx->litBuffer, istart+lhSize, litSize);
dctx->litPtr = dctx->litBuffer;
dctx->litSize = litSize;
memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
ZSTD_memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
return lhSize+litSize;
}
/* direct reference into compressed stream */
@@ -223,7 +223,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
break;
}
RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
dctx->litPtr = dctx->litBuffer;
dctx->litSize = litSize;
return lhSize+1;
@@ -236,7 +236,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
/* Default FSE distribution tables.
* These are pre-calculated FSE decoding tables using default distributions as defined in specification :
* https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#default-distributions
* https://github.com/facebook/zstd/blob/release/doc/zstd_compression_format.md#default-distributions
* They were generated programmatically with following method :
* - start from default distributions, present in /lib/common/zstd_internal.h
* - generate tables normally, using ZSTD_buildFSETable()
@@ -364,23 +364,26 @@ static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U32 nbAddB
* generate FSE decoding table for one symbol (ll, ml or off)
* cannot fail if input is valid =>
* all inputs are presumed validated at this stage */
void
ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
FORCE_INLINE_TEMPLATE
void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
const short* normalizedCounter, unsigned maxSymbolValue,
const U32* baseValue, const U32* nbAdditionalBits,
unsigned tableLog)
unsigned tableLog, void* wksp, size_t wkspSize)
{
ZSTD_seqSymbol* const tableDecode = dt+1;
U16 symbolNext[MaxSeq+1];
U32 const maxSV1 = maxSymbolValue + 1;
U32 const tableSize = 1 << tableLog;
U32 highThreshold = tableSize-1;
U16* symbolNext = (U16*)wksp;
BYTE* spread = (BYTE*)(symbolNext + MaxSeq + 1);
U32 highThreshold = tableSize - 1;
/* Sanity Checks */
assert(maxSymbolValue <= MaxSeq);
assert(tableLog <= MaxFSELog);
assert(wkspSize >= ZSTD_BUILD_FSE_TABLE_WKSP_SIZE);
(void)wkspSize;
/* Init, lay down lowprob symbols */
{ ZSTD_seqSymbol_header DTableH;
DTableH.tableLog = tableLog;
@@ -396,16 +399,69 @@ ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
assert(normalizedCounter[s]>=0);
symbolNext[s] = (U16)normalizedCounter[s];
} } }
memcpy(dt, &DTableH, sizeof(DTableH));
ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
}
/* Spread symbols */
{ U32 const tableMask = tableSize-1;
assert(tableSize <= 512);
/* Specialized symbol spreading for the case when there are
* no low probability (-1 count) symbols. When compressing
* small blocks we avoid low probability symbols to hit this
* case, since header decoding speed matters more.
*/
if (highThreshold == tableSize - 1) {
size_t const tableMask = tableSize-1;
size_t const step = FSE_TABLESTEP(tableSize);
/* First lay down the symbols in order.
* We use a uint64_t to lay down 8 bytes at a time. This reduces branch
* misses since small blocks generally have small table logs, so nearly
* all symbols have counts <= 8. We ensure we have 8 bytes at the end of
* our buffer to handle the over-write.
*/
{
U64 const add = 0x0101010101010101ull;
size_t pos = 0;
U64 sv = 0;
U32 s;
for (s=0; s<maxSV1; ++s, sv += add) {
int i;
int const n = normalizedCounter[s];
MEM_write64(spread + pos, sv);
for (i = 8; i < n; i += 8) {
MEM_write64(spread + pos + i, sv);
}
pos += n;
}
}
/* Now we spread those positions across the table.
* The benefit of doing it in two stages is that we avoid the the
* variable size inner loop, which caused lots of branch misses.
* Now we can run through all the positions without any branch misses.
* We unroll the loop twice, since that is what emperically worked best.
*/
{
size_t position = 0;
size_t s;
size_t const unroll = 2;
assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
for (s = 0; s < (size_t)tableSize; s += unroll) {
size_t u;
for (u = 0; u < unroll; ++u) {
size_t const uPosition = (position + (u * step)) & tableMask;
tableDecode[uPosition].baseValue = spread[s + u];
}
position = (position + (unroll * step)) & tableMask;
}
assert(position == 0);
}
} else {
U32 const tableMask = tableSize-1;
U32 const step = FSE_TABLESTEP(tableSize);
U32 s, position = 0;
for (s=0; s<maxSV1; s++) {
int i;
for (i=0; i<normalizedCounter[s]; i++) {
int const n = normalizedCounter[s];
for (i=0; i<n; i++) {
tableDecode[position].baseValue = s;
position = (position + step) & tableMask;
while (position > highThreshold) position = (position + step) & tableMask; /* lowprob area */
@@ -414,7 +470,8 @@ ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
}
/* Build Decoding table */
{ U32 u;
{
U32 u;
for (u=0; u<tableSize; u++) {
U32 const symbol = tableDecode[u].baseValue;
U32 const nextState = symbolNext[symbol]++;
@@ -423,7 +480,46 @@ ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
assert(nbAdditionalBits[symbol] < 255);
tableDecode[u].nbAdditionalBits = (BYTE)nbAdditionalBits[symbol];
tableDecode[u].baseValue = baseValue[symbol];
} }
}
}
}
/* Avoids the FORCE_INLINE of the _body() function. */
static void ZSTD_buildFSETable_body_default(ZSTD_seqSymbol* dt,
const short* normalizedCounter, unsigned maxSymbolValue,
const U32* baseValue, const U32* nbAdditionalBits,
unsigned tableLog, void* wksp, size_t wkspSize)
{
ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
}
#if DYNAMIC_BMI2
TARGET_ATTRIBUTE("bmi2") static void ZSTD_buildFSETable_body_bmi2(ZSTD_seqSymbol* dt,
const short* normalizedCounter, unsigned maxSymbolValue,
const U32* baseValue, const U32* nbAdditionalBits,
unsigned tableLog, void* wksp, size_t wkspSize)
{
ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
}
#endif
void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
const short* normalizedCounter, unsigned maxSymbolValue,
const U32* baseValue, const U32* nbAdditionalBits,
unsigned tableLog, void* wksp, size_t wkspSize, int bmi2)
{
#if DYNAMIC_BMI2
if (bmi2) {
ZSTD_buildFSETable_body_bmi2(dt, normalizedCounter, maxSymbolValue,
baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
return;
}
#endif
(void)bmi2;
ZSTD_buildFSETable_body_default(dt, normalizedCounter, maxSymbolValue,
baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
}
@@ -435,7 +531,8 @@ static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymb
const void* src, size_t srcSize,
const U32* baseValue, const U32* nbAdditionalBits,
const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable,
int ddictIsCold, int nbSeq)
int ddictIsCold, int nbSeq, U32* wksp, size_t wkspSize,
int bmi2)
{
switch(type)
{
@@ -467,7 +564,7 @@ static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymb
size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
RETURN_ERROR_IF(FSE_isError(headerSize), corruption_detected, "");
RETURN_ERROR_IF(tableLog > maxLog, corruption_detected, "");
ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog);
ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog, wksp, wkspSize, bmi2);
*DTablePtr = DTableSpace;
return headerSize;
}
@@ -480,7 +577,7 @@ static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymb
size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
const void* src, size_t srcSize)
{
const BYTE* const istart = (const BYTE* const)src;
const BYTE* const istart = (const BYTE*)src;
const BYTE* const iend = istart + srcSize;
const BYTE* ip = istart;
int nbSeq;
@@ -499,7 +596,8 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
if (nbSeq > 0x7F) {
if (nbSeq == 0xFF) {
RETURN_ERROR_IF(ip+2 > iend, srcSize_wrong, "");
nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
nbSeq = MEM_readLE16(ip) + LONGNBSEQ;
ip+=2;
} else {
RETURN_ERROR_IF(ip >= iend, srcSize_wrong, "");
nbSeq = ((nbSeq-0x80)<<8) + *ip++;
@@ -520,7 +618,9 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
ip, iend-ip,
LL_base, LL_bits,
LL_defaultDTable, dctx->fseEntropy,
dctx->ddictIsCold, nbSeq);
dctx->ddictIsCold, nbSeq,
dctx->workspace, sizeof(dctx->workspace),
dctx->bmi2);
RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected, "ZSTD_buildSeqTable failed");
ip += llhSize;
}
@@ -530,7 +630,9 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
ip, iend-ip,
OF_base, OF_bits,
OF_defaultDTable, dctx->fseEntropy,
dctx->ddictIsCold, nbSeq);
dctx->ddictIsCold, nbSeq,
dctx->workspace, sizeof(dctx->workspace),
dctx->bmi2);
RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected, "ZSTD_buildSeqTable failed");
ip += ofhSize;
}
@@ -540,7 +642,9 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
ip, iend-ip,
ML_base, ML_bits,
ML_defaultDTable, dctx->fseEntropy,
dctx->ddictIsCold, nbSeq);
dctx->ddictIsCold, nbSeq,
dctx->workspace, sizeof(dctx->workspace),
dctx->bmi2);
RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected, "ZSTD_buildSeqTable failed");
ip += mlhSize;
}
@@ -554,7 +658,6 @@ typedef struct {
size_t litLength;
size_t matchLength;
size_t offset;
const BYTE* match;
} seq_t;
typedef struct {
@@ -568,9 +671,6 @@ typedef struct {
ZSTD_fseState stateOffb;
ZSTD_fseState stateML;
size_t prevOffset[ZSTD_REP_NUM];
const BYTE* prefixStart;
const BYTE* dictEnd;
size_t pos;
} seqState_t;
/*! ZSTD_overlapCopy8() :
@@ -686,12 +786,12 @@ size_t ZSTD_execSequenceEnd(BYTE* op,
RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
match = dictEnd - (prefixStart-match);
if (match + sequence.matchLength <= dictEnd) {
memmove(oLitEnd, match, sequence.matchLength);
ZSTD_memmove(oLitEnd, match, sequence.matchLength);
return sequenceLength;
}
/* span extDict & currentPrefixSegment */
{ size_t const length1 = dictEnd - match;
memmove(oLitEnd, match, length1);
ZSTD_memmove(oLitEnd, match, length1);
op = oLitEnd + length1;
sequence.matchLength -= length1;
match = prefixStart;
@@ -752,12 +852,12 @@ size_t ZSTD_execSequence(BYTE* op,
RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
match = dictEnd + (match - prefixStart);
if (match + sequence.matchLength <= dictEnd) {
memmove(oLitEnd, match, sequence.matchLength);
ZSTD_memmove(oLitEnd, match, sequence.matchLength);
return sequenceLength;
}
/* span extDict & currentPrefixSegment */
{ size_t const length1 = dictEnd - match;
memmove(oLitEnd, match, length1);
ZSTD_memmove(oLitEnd, match, length1);
op = oLitEnd + length1;
sequence.matchLength -= length1;
match = prefixStart;
@@ -832,10 +932,9 @@ ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, ZSTD
: 0)
typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e;
typedef enum { ZSTD_p_noPrefetch=0, ZSTD_p_prefetch=1 } ZSTD_prefetch_e;
FORCE_INLINE_TEMPLATE seq_t
ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets, const ZSTD_prefetch_e prefetch)
ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
{
seq_t seq;
ZSTD_seqSymbol const llDInfo = seqState->stateLL.table[seqState->stateLL.state];
@@ -910,14 +1009,6 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets, c
DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
(U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
if (prefetch == ZSTD_p_prefetch) {
size_t const pos = seqState->pos + seq.litLength;
const BYTE* const matchBase = (seq.offset > pos) ? seqState->dictEnd : seqState->prefixStart;
seq.match = matchBase + pos - seq.offset; /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
* No consequence though : no memory access will occur, offset is only used for prefetching */
seqState->pos = pos + seq.matchLength;
}
/* ANS state update
* gcc-9.0.0 does 2.5% worse with ZSTD_updateFseStateWithDInfo().
* clang-9.2.0 does 7% worse with ZSTD_updateFseState().
@@ -948,7 +1039,7 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets, c
}
#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
static int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
MEM_STATIC int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
{
size_t const windowSize = dctx->fParams.windowSize;
/* No dictionary used. */
@@ -969,6 +1060,7 @@ MEM_STATIC void ZSTD_assertValidSequence(
seq_t const seq,
BYTE const* prefixStart, BYTE const* virtualStart)
{
#if DEBUGLEVEL >= 1
size_t const windowSize = dctx->fParams.windowSize;
size_t const sequenceSize = seq.litLength + seq.matchLength;
BYTE const* const oLitEnd = op + seq.litLength;
@@ -986,6 +1078,9 @@ MEM_STATIC void ZSTD_assertValidSequence(
/* Offset must be within our window. */
assert(seq.offset <= windowSize);
}
#else
(void)dctx, (void)op, (void)oend, (void)seq, (void)prefixStart, (void)virtualStart;
#endif
}
#endif
@@ -1000,7 +1095,7 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
{
const BYTE* ip = (const BYTE*)seqStart;
const BYTE* const iend = ip + seqSize;
BYTE* const ostart = (BYTE* const)dst;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + maxDstSize;
BYTE* op = ostart;
const BYTE* litPtr = dctx->litPtr;
@@ -1014,7 +1109,6 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
/* Regen sequences */
if (nbSeq) {
seqState_t seqState;
size_t error = 0;
dctx->fseEntropy = 1;
{ U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
RETURN_ERROR_IF(
@@ -1048,13 +1142,14 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
* If you see most cycles served out of the DSB you've hit the good case.
* If it is pretty even then you may be in an okay case.
*
* I've been able to reproduce this issue on the following CPUs:
* This issue has been reproduced on the following CPUs:
* - Kabylake: Macbook Pro (15-inch, 2019) 2.4 GHz Intel Core i9
* Use Instruments->Counters to get DSB/MITE cycles.
* I never got performance swings, but I was able to
* go from the good case of mostly DSB to half of the
* cycles served from MITE.
* - Coffeelake: Intel i9-9900k
* - Coffeelake: Intel i7-9700k
*
* I haven't been able to reproduce the instability or DSB misses on any
* of the following CPUS:
@@ -1067,33 +1162,35 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
*
* https://gist.github.com/terrelln/9889fc06a423fd5ca6e99351564473f4
*/
__asm__(".p2align 6");
__asm__("nop");
__asm__(".p2align 5");
__asm__("nop");
# if __GNUC__ >= 9
/* better for gcc-9 and gcc-10, worse for clang and gcc-8 */
__asm__(".p2align 3");
# else
__asm__(".p2align 4");
# endif
#endif
for ( ; ; ) {
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, ZSTD_p_noPrefetch);
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
#endif
if (UNLIKELY(ZSTD_isError(oneSeqSize)))
return oneSeqSize;
DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
op += oneSeqSize;
if (UNLIKELY(!--nbSeq))
break;
BIT_reloadDStream(&(seqState.DStream));
/* gcc and clang both don't like early returns in this loop.
* gcc doesn't like early breaks either.
* Instead save an error and report it at the end.
* When there is an error, don't increment op, so we don't
* overwrite.
*/
if (UNLIKELY(ZSTD_isError(oneSeqSize))) error = oneSeqSize;
else op += oneSeqSize;
if (UNLIKELY(!--nbSeq)) break;
}
/* check if reached exact end */
DEBUGLOG(5, "ZSTD_decompressSequences_body: after decode loop, remaining nbSeq : %i", nbSeq);
if (ZSTD_isError(error)) return error;
RETURN_ERROR_IF(nbSeq, corruption_detected, "");
RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
/* save reps for next block */
@@ -1104,7 +1201,7 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
{ size_t const lastLLSize = litEnd - litPtr;
RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
if (op != NULL) {
memcpy(op, litPtr, lastLLSize);
ZSTD_memcpy(op, litPtr, lastLLSize);
op += lastLLSize;
}
}
@@ -1124,6 +1221,24 @@ ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
FORCE_INLINE_TEMPLATE size_t
ZSTD_prefetchMatch(size_t prefetchPos, seq_t const sequence,
const BYTE* const prefixStart, const BYTE* const dictEnd)
{
prefetchPos += sequence.litLength;
{ const BYTE* const matchBase = (sequence.offset > prefetchPos) ? dictEnd : prefixStart;
const BYTE* const match = matchBase + prefetchPos - sequence.offset; /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
* No consequence though : memory address is only used for prefetching, not for dereferencing */
PREFETCH_L1(match); PREFETCH_L1(match+CACHELINE_SIZE); /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
}
return prefetchPos + sequence.matchLength;
}
/* This decoding function employs prefetching
* to reduce latency impact of cache misses.
* It's generally employed when block contains a significant portion of long-distance matches
* or when coupled with a "cold" dictionary */
FORCE_INLINE_TEMPLATE size_t
ZSTD_decompressSequencesLong_body(
ZSTD_DCtx* dctx,
@@ -1134,7 +1249,7 @@ ZSTD_decompressSequencesLong_body(
{
const BYTE* ip = (const BYTE*)seqStart;
const BYTE* const iend = ip + seqSize;
BYTE* const ostart = (BYTE* const)dst;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + maxDstSize;
BYTE* op = ostart;
const BYTE* litPtr = dctx->litPtr;
@@ -1146,18 +1261,17 @@ ZSTD_decompressSequencesLong_body(
/* Regen sequences */
if (nbSeq) {
#define STORED_SEQS 4
#define STORED_SEQS 8
#define STORED_SEQS_MASK (STORED_SEQS-1)
#define ADVANCED_SEQS 4
#define ADVANCED_SEQS STORED_SEQS
seq_t sequences[STORED_SEQS];
int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS);
seqState_t seqState;
int seqNb;
size_t prefetchPos = (size_t)(op-prefixStart); /* track position relative to prefixStart */
dctx->fseEntropy = 1;
{ int i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
seqState.prefixStart = prefixStart;
seqState.pos = (size_t)(op-prefixStart);
seqState.dictEnd = dictEnd;
assert(dst != NULL);
assert(iend >= ip);
RETURN_ERROR_IF(
@@ -1169,21 +1283,23 @@ ZSTD_decompressSequencesLong_body(
/* prepare in advance */
for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && (seqNb<seqAdvance); seqNb++) {
sequences[seqNb] = ZSTD_decodeSequence(&seqState, isLongOffset, ZSTD_p_prefetch);
PREFETCH_L1(sequences[seqNb].match); PREFETCH_L1(sequences[seqNb].match + sequences[seqNb].matchLength - 1); /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
sequences[seqNb] = sequence;
}
RETURN_ERROR_IF(seqNb<seqAdvance, corruption_detected, "");
/* decode and decompress */
for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb<nbSeq) ; seqNb++) {
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, ZSTD_p_prefetch);
seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
assert(!ZSTD_isError(oneSeqSize));
if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
#endif
if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
PREFETCH_L1(sequence.match); PREFETCH_L1(sequence.match + sequence.matchLength - 1); /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
sequences[seqNb & STORED_SEQS_MASK] = sequence;
op += oneSeqSize;
}
@@ -1209,7 +1325,7 @@ ZSTD_decompressSequencesLong_body(
{ size_t const lastLLSize = litEnd - litPtr;
RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
if (op != NULL) {
memcpy(op, litPtr, lastLLSize);
ZSTD_memcpy(op, litPtr, lastLLSize);
op += lastLLSize;
}
}
@@ -1409,9 +1525,9 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
}
void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst, size_t dstSize)
{
if (dst != dctx->previousDstEnd) { /* not contiguous */
if (dst != dctx->previousDstEnd && dstSize > 0) { /* not contiguous */
dctx->dictEnd = dctx->previousDstEnd;
dctx->virtualStart = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
dctx->prefixStart = dst;
@@ -1425,7 +1541,7 @@ size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
const void* src, size_t srcSize)
{
size_t dSize;
ZSTD_checkContinuity(dctx, dst);
ZSTD_checkContinuity(dctx, dst, dstCapacity);
dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0);
dctx->previousDstEnd = (char*)dst + dSize;
return dSize;
Utilities/cmzstd/lib/decompress/zstd_decompress_block.h
+6 -3
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -15,7 +15,7 @@
/*-*******************************************************
* Dependencies
*********************************************************/
#include <stddef.h> /* size_t */
#include "../common/zstd_deps.h" /* size_t */
#include "../zstd.h" /* DCtx, and some public functions */
#include "../common/zstd_internal.h" /* blockProperties_t, and some public functions */
#include "zstd_decompress_internal.h" /* ZSTD_seqSymbol */
@@ -48,12 +48,15 @@ size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
* this function must be called with valid parameters only
* (dt is large enough, normalizedCounter distribution total is a power of 2, max is within range, etc.)
* in which case it cannot fail.
* The workspace must be 4-byte aligned and at least ZSTD_BUILD_FSE_TABLE_WKSP_SIZE bytes, which is
* defined in zstd_decompress_internal.h.
* Internal use only.
*/
void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
const short* normalizedCounter, unsigned maxSymbolValue,
const U32* baseValue, const U32* nbAdditionalBits,
unsigned tableLog);
unsigned tableLog, void* wksp, size_t wkspSize,
int bmi2);
#endif /* ZSTD_DEC_BLOCK_H */
Utilities/cmzstd/lib/decompress/zstd_decompress_internal.h
+27 -11
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -27,26 +27,26 @@
/*-*******************************************************
* Constants
*********************************************************/
static const U32 LL_base[MaxLL+1] = {
static UNUSED_ATTR const U32 LL_base[MaxLL+1] = {
0, 1, 2, 3, 4, 5, 6, 7,
8, 9, 10, 11, 12, 13, 14, 15,
16, 18, 20, 22, 24, 28, 32, 40,
48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
0x2000, 0x4000, 0x8000, 0x10000 };
static const U32 OF_base[MaxOff+1] = {
static UNUSED_ATTR const U32 OF_base[MaxOff+1] = {
0, 1, 1, 5, 0xD, 0x1D, 0x3D, 0x7D,
0xFD, 0x1FD, 0x3FD, 0x7FD, 0xFFD, 0x1FFD, 0x3FFD, 0x7FFD,
0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD, 0x1FFFFFFD, 0x3FFFFFFD, 0x7FFFFFFD };
static const U32 OF_bits[MaxOff+1] = {
static UNUSED_ATTR const U32 OF_bits[MaxOff+1] = {
0, 1, 2, 3, 4, 5, 6, 7,
8, 9, 10, 11, 12, 13, 14, 15,
16, 17, 18, 19, 20, 21, 22, 23,
24, 25, 26, 27, 28, 29, 30, 31 };
static const U32 ML_base[MaxML+1] = {
static UNUSED_ATTR const U32 ML_base[MaxML+1] = {
3, 4, 5, 6, 7, 8, 9, 10,
11, 12, 13, 14, 15, 16, 17, 18,
19, 20, 21, 22, 23, 24, 25, 26,
@@ -73,12 +73,16 @@ static const U32 ML_base[MaxML+1] = {
#define SEQSYMBOL_TABLE_SIZE(log) (1 + (1 << (log)))
#define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE (sizeof(S16) * (MaxSeq + 1) + (1u << MaxFSELog) + sizeof(U64))
#define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32 ((ZSTD_BUILD_FSE_TABLE_WKSP_SIZE + sizeof(U32) - 1) / sizeof(U32))
typedef struct {
ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)]; /* Note : Space reserved for FSE Tables */
ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)]; /* is also used as temporary workspace while building hufTable during DDict creation */
ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)]; /* and therefore must be at least HUF_DECOMPRESS_WORKSPACE_SIZE large */
HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)]; /* can accommodate HUF_decompress4X */
U32 rep[ZSTD_REP_NUM];
U32 workspace[ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32];
} ZSTD_entropyDTables_t;
typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
@@ -95,10 +99,12 @@ typedef enum {
ZSTD_use_once = 1 /* Use the dictionary once and set to ZSTD_dont_use */
} ZSTD_dictUses_e;
typedef enum {
ZSTD_obm_buffered = 0, /* Buffer the output */
ZSTD_obm_stable = 1 /* ZSTD_outBuffer is stable */
} ZSTD_outBufferMode_e;
/* Hashset for storing references to multiple ZSTD_DDict within ZSTD_DCtx */
typedef struct {
const ZSTD_DDict** ddictPtrTable;
size_t ddictPtrTableSize;
size_t ddictPtrCount;
} ZSTD_DDictHashSet;
struct ZSTD_DCtx_s
{
@@ -114,6 +120,7 @@ struct ZSTD_DCtx_s
const void* dictEnd; /* end of previous segment */
size_t expected;
ZSTD_frameHeader fParams;
U64 processedCSize;
U64 decodedSize;
blockType_e bType; /* used in ZSTD_decompressContinue(), store blockType between block header decoding and block decompression stages */
ZSTD_dStage stage;
@@ -122,6 +129,8 @@ struct ZSTD_DCtx_s
XXH64_state_t xxhState;
size_t headerSize;
ZSTD_format_e format;
ZSTD_forceIgnoreChecksum_e forceIgnoreChecksum; /* User specified: if == 1, will ignore checksums in compressed frame. Default == 0 */
U32 validateChecksum; /* if == 1, will validate checksum. Is == 1 if (fParams.checksumFlag == 1) and (forceIgnoreChecksum == 0). */
const BYTE* litPtr;
ZSTD_customMem customMem;
size_t litSize;
@@ -135,6 +144,8 @@ struct ZSTD_DCtx_s
U32 dictID;
int ddictIsCold; /* if == 1 : dictionary is "new" for working context, and presumed "cold" (not in cpu cache) */
ZSTD_dictUses_e dictUses;
ZSTD_DDictHashSet* ddictSet; /* Hash set for multiple ddicts */
ZSTD_refMultipleDDicts_e refMultipleDDicts; /* User specified: if == 1, will allow references to multiple DDicts. Default == 0 (disabled) */
/* streaming */
ZSTD_dStreamStage streamStage;
@@ -152,7 +163,7 @@ struct ZSTD_DCtx_s
U32 legacyVersion;
U32 hostageByte;
int noForwardProgress;
ZSTD_outBufferMode_e outBufferMode;
ZSTD_bufferMode_e outBufferMode;
ZSTD_outBuffer expectedOutBuffer;
/* workspace */
@@ -165,6 +176,11 @@ struct ZSTD_DCtx_s
void const* dictContentBeginForFuzzing;
void const* dictContentEndForFuzzing;
#endif
/* Tracing */
#if ZSTD_TRACE
ZSTD_TraceCtx traceCtx;
#endif
}; /* typedef'd to ZSTD_DCtx within "zstd.h" */
@@ -183,7 +199,7 @@ size_t ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
* If yes, do nothing (continue on current segment).
* If not, classify previous segment as "external dictionary", and start a new segment.
* This function cannot fail. */
void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst);
void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst, size_t dstSize);
#endif /* ZSTD_DECOMPRESS_INTERNAL_H */
Utilities/cmzstd/lib/deprecated/zbuff.h
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/deprecated/zbuff_common.c
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/deprecated/zbuff_compress.c
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/deprecated/zbuff_decompress.c
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
Utilities/cmzstd/lib/dictBuilder/cover.c
+46 -36
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -26,47 +26,57 @@
#include <string.h> /* memset */
#include <time.h> /* clock */
#ifndef ZDICT_STATIC_LINKING_ONLY
# define ZDICT_STATIC_LINKING_ONLY
#endif
#include "../common/mem.h" /* read */
#include "../common/pool.h"
#include "../common/threading.h"
#include "cover.h"
#include "../common/zstd_internal.h" /* includes zstd.h */
#ifndef ZDICT_STATIC_LINKING_ONLY
#define ZDICT_STATIC_LINKING_ONLY
#endif
#include "zdict.h"
#include "../zdict.h"
#include "cover.h"
/*-*************************************
* Constants
***************************************/
#define COVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
#define DEFAULT_SPLITPOINT 1.0
#define COVER_DEFAULT_SPLITPOINT 1.0
/*-*************************************
* Console display
***************************************/
#ifndef LOCALDISPLAYLEVEL
static int g_displayLevel = 2;
#endif
#undef DISPLAY
#define DISPLAY(...) \
{ \
fprintf(stderr, __VA_ARGS__); \
fflush(stderr); \
}
#undef LOCALDISPLAYLEVEL
#define LOCALDISPLAYLEVEL(displayLevel, l, ...) \
if (displayLevel >= l) { \
DISPLAY(__VA_ARGS__); \
} /* 0 : no display; 1: errors; 2: default; 3: details; 4: debug */
#undef DISPLAYLEVEL
#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
#ifndef LOCALDISPLAYUPDATE
static const clock_t g_refreshRate = CLOCKS_PER_SEC * 15 / 100;
static clock_t g_time = 0;
#endif
#undef LOCALDISPLAYUPDATE
#define LOCALDISPLAYUPDATE(displayLevel, l, ...) \
if (displayLevel >= l) { \
if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) { \
if ((clock() - g_time > g_refreshRate) || (displayLevel >= 4)) { \
g_time = clock(); \
DISPLAY(__VA_ARGS__); \
} \
}
#undef DISPLAYUPDATE
#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
static clock_t g_time = 0;
/*-*************************************
* Hash table
@@ -120,9 +130,9 @@ static int COVER_map_init(COVER_map_t *map, U32 size) {
/**
* Internal hash function
*/
static const U32 prime4bytes = 2654435761U;
static const U32 COVER_prime4bytes = 2654435761U;
static U32 COVER_map_hash(COVER_map_t *map, U32 key) {
return (key * prime4bytes) >> (32 - map->sizeLog);
return (key * COVER_prime4bytes) >> (32 - map->sizeLog);
}
/**
@@ -215,7 +225,7 @@ typedef struct {
} COVER_ctx_t;
/* We need a global context for qsort... */
static COVER_ctx_t *g_ctx = NULL;
static COVER_ctx_t *g_coverCtx = NULL;
/*-*************************************
* Helper functions
@@ -258,11 +268,11 @@ static int COVER_cmp8(COVER_ctx_t *ctx, const void *lp, const void *rp) {
/**
* Same as COVER_cmp() except ties are broken by pointer value
* NOTE: g_ctx must be set to call this function. A global is required because
* NOTE: g_coverCtx must be set to call this function. A global is required because
* qsort doesn't take an opaque pointer.
*/
static int COVER_strict_cmp(const void *lp, const void *rp) {
int result = COVER_cmp(g_ctx, lp, rp);
static int WIN_CDECL COVER_strict_cmp(const void *lp, const void *rp) {
int result = COVER_cmp(g_coverCtx, lp, rp);
if (result == 0) {
result = lp < rp ? -1 : 1;
}
@@ -271,8 +281,8 @@ static int COVER_strict_cmp(const void *lp, const void *rp) {
/**
* Faster version for d <= 8.
*/
static int COVER_strict_cmp8(const void *lp, const void *rp) {
int result = COVER_cmp8(g_ctx, lp, rp);
static int WIN_CDECL COVER_strict_cmp8(const void *lp, const void *rp) {
int result = COVER_cmp8(g_coverCtx, lp, rp);
if (result == 0) {
result = lp < rp ? -1 : 1;
}
@@ -603,7 +613,7 @@ static size_t COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
/* qsort doesn't take an opaque pointer, so pass as a global.
* On OpenBSD qsort() is not guaranteed to be stable, their mergesort() is.
*/
g_ctx = ctx;
g_coverCtx = ctx;
#if defined(__OpenBSD__)
mergesort(ctx->suffix, ctx->suffixSize, sizeof(U32),
(ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
@@ -946,7 +956,7 @@ void COVER_dictSelectionFree(COVER_dictSelection_t selection){
free(selection.dictContent);
}
COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent, size_t dictBufferCapacity,
size_t dictContentSize, const BYTE* samplesBuffer, const size_t* samplesSizes, unsigned nbFinalizeSamples,
size_t nbCheckSamples, size_t nbSamples, ZDICT_cover_params_t params, size_t* offsets, size_t totalCompressedSize) {
@@ -954,8 +964,8 @@ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
size_t largestCompressed = 0;
BYTE* customDictContentEnd = customDictContent + dictContentSize;
BYTE * largestDictbuffer = (BYTE *)malloc(dictContentSize);
BYTE * candidateDictBuffer = (BYTE *)malloc(dictContentSize);
BYTE * largestDictbuffer = (BYTE *)malloc(dictBufferCapacity);
BYTE * candidateDictBuffer = (BYTE *)malloc(dictBufferCapacity);
double regressionTolerance = ((double)params.shrinkDictMaxRegression / 100.0) + 1.00;
if (!largestDictbuffer || !candidateDictBuffer) {
@@ -967,7 +977,7 @@ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
/* Initial dictionary size and compressed size */
memcpy(largestDictbuffer, customDictContent, dictContentSize);
dictContentSize = ZDICT_finalizeDictionary(
largestDictbuffer, dictContentSize, customDictContent, dictContentSize,
largestDictbuffer, dictBufferCapacity, customDictContent, dictContentSize,
samplesBuffer, samplesSizes, nbFinalizeSamples, params.zParams);
if (ZDICT_isError(dictContentSize)) {
@@ -1001,7 +1011,7 @@ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
while (dictContentSize < largestDict) {
memcpy(candidateDictBuffer, largestDictbuffer, largestDict);
dictContentSize = ZDICT_finalizeDictionary(
candidateDictBuffer, dictContentSize, customDictContentEnd - dictContentSize, dictContentSize,
candidateDictBuffer, dictBufferCapacity, customDictContentEnd - dictContentSize, dictContentSize,
samplesBuffer, samplesSizes, nbFinalizeSamples, params.zParams);
if (ZDICT_isError(dictContentSize)) {
@@ -1053,18 +1063,19 @@ typedef struct COVER_tryParameters_data_s {
* This function is thread safe if zstd is compiled with multithreaded support.
* It takes its parameters as an *OWNING* opaque pointer to support threading.
*/
static void COVER_tryParameters(void *opaque) {
static void COVER_tryParameters(void *opaque)
{
/* Save parameters as local variables */
COVER_tryParameters_data_t *const data = (COVER_tryParameters_data_t *)opaque;
COVER_tryParameters_data_t *const data = (COVER_tryParameters_data_t*)opaque;
const COVER_ctx_t *const ctx = data->ctx;
const ZDICT_cover_params_t parameters = data->parameters;
size_t dictBufferCapacity = data->dictBufferCapacity;
size_t totalCompressedSize = ERROR(GENERIC);
/* Allocate space for hash table, dict, and freqs */
COVER_map_t activeDmers;
BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
BYTE* const dict = (BYTE*)malloc(dictBufferCapacity);
COVER_dictSelection_t selection = COVER_dictSelectionError(ERROR(GENERIC));
U32 *freqs = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
U32* const freqs = (U32*)malloc(ctx->suffixSize * sizeof(U32));
if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
DISPLAYLEVEL(1, "Failed to allocate dmer map: out of memory\n");
goto _cleanup;
@@ -1079,7 +1090,7 @@ static void COVER_tryParameters(void *opaque) {
{
const size_t tail = COVER_buildDictionary(ctx, freqs, &activeDmers, dict,
dictBufferCapacity, parameters);
selection = COVER_selectDict(dict + tail, dictBufferCapacity - tail,
selection = COVER_selectDict(dict + tail, dictBufferCapacity, dictBufferCapacity - tail,
ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbTrainSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
totalCompressedSize);
@@ -1094,19 +1105,18 @@ _cleanup:
free(data);
COVER_map_destroy(&activeDmers);
COVER_dictSelectionFree(selection);
if (freqs) {
free(freqs);
}
free(freqs);
}
ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
const size_t *samplesSizes, unsigned nbSamples,
ZDICT_cover_params_t *parameters) {
void* dictBuffer, size_t dictBufferCapacity, const void* samplesBuffer,
const size_t* samplesSizes, unsigned nbSamples,
ZDICT_cover_params_t* parameters)
{
/* constants */
const unsigned nbThreads = parameters->nbThreads;
const double splitPoint =
parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
parameters->splitPoint <= 0.0 ? COVER_DEFAULT_SPLITPOINT : parameters->splitPoint;
const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
Utilities/cmzstd/lib/dictBuilder/cover.h
+7 -6
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2017-2020, Facebook, Inc.
* Copyright (c) Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -8,6 +8,10 @@
* You may select, at your option, one of the above-listed licenses.
*/
#ifndef ZDICT_STATIC_LINKING_ONLY
# define ZDICT_STATIC_LINKING_ONLY
#endif
#include <stdio.h> /* fprintf */
#include <stdlib.h> /* malloc, free, qsort */
#include <string.h> /* memset */
@@ -16,10 +20,7 @@
#include "../common/pool.h"
#include "../common/threading.h"
#include "../common/zstd_internal.h" /* includes zstd.h */
#ifndef ZDICT_STATIC_LINKING_ONLY
#define ZDICT_STATIC_LINKING_ONLY
#endif
#include "zdict.h"
#include "../zdict.h"
/**
* COVER_best_t is used for two purposes:
@@ -152,6 +153,6 @@ void COVER_dictSelectionFree(COVER_dictSelection_t selection);
* smallest dictionary within a specified regression of the compressed size
* from the largest dictionary.
*/
COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent, size_t dictBufferCapacity,
size_t dictContentSize, const BYTE* samplesBuffer, const size_t* samplesSizes, unsigned nbFinalizeSamples,
size_t nbCheckSamples, size_t nbSamples, ZDICT_cover_params_t params, size_t* offsets, size_t totalCompressedSize);
Utilities/cmzstd/lib/dictBuilder/divsufsort.c
+1 -1
View File
@@ -1576,7 +1576,7 @@ note:
/* Construct the inverse suffix array of type B* suffixes using trsort. */
trsort(ISAb, SA, m, 1);
/* Set the sorted order of tyoe B* suffixes. */
/* Set the sorted order of type B* suffixes. */
for(i = n - 1, j = m, c0 = T[n - 1]; 0 <= i;) {
for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) >= c1); --i, c1 = c0) { }
if(0 <= i) {
Utilities/cmzstd/lib/dictBuilder/fastcover.c
+32 -30
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2018-2020, Facebook, Inc.
* Copyright (c) Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -16,15 +16,17 @@
#include <string.h> /* memset */
#include <time.h> /* clock */
#ifndef ZDICT_STATIC_LINKING_ONLY
# define ZDICT_STATIC_LINKING_ONLY
#endif
#include "../common/mem.h" /* read */
#include "../common/pool.h"
#include "../common/threading.h"
#include "cover.h"
#include "../common/zstd_internal.h" /* includes zstd.h */
#ifndef ZDICT_STATIC_LINKING_ONLY
#define ZDICT_STATIC_LINKING_ONLY
#endif
#include "zdict.h"
#include "../compress/zstd_compress_internal.h" /* ZSTD_hash*() */
#include "../zdict.h"
#include "cover.h"
/*-*************************************
@@ -33,7 +35,7 @@
#define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
#define FASTCOVER_MAX_F 31
#define FASTCOVER_MAX_ACCEL 10
#define DEFAULT_SPLITPOINT 0.75
#define FASTCOVER_DEFAULT_SPLITPOINT 0.75
#define DEFAULT_F 20
#define DEFAULT_ACCEL 1
@@ -41,50 +43,50 @@
/*-*************************************
* Console display
***************************************/
#ifndef LOCALDISPLAYLEVEL
static int g_displayLevel = 2;
#endif
#undef DISPLAY
#define DISPLAY(...) \
{ \
fprintf(stderr, __VA_ARGS__); \
fflush(stderr); \
}
#undef LOCALDISPLAYLEVEL
#define LOCALDISPLAYLEVEL(displayLevel, l, ...) \
if (displayLevel >= l) { \
DISPLAY(__VA_ARGS__); \
} /* 0 : no display; 1: errors; 2: default; 3: details; 4: debug */
#undef DISPLAYLEVEL
#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
#ifndef LOCALDISPLAYUPDATE
static const clock_t g_refreshRate = CLOCKS_PER_SEC * 15 / 100;
static clock_t g_time = 0;
#endif
#undef LOCALDISPLAYUPDATE
#define LOCALDISPLAYUPDATE(displayLevel, l, ...) \
if (displayLevel >= l) { \
if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) { \
if ((clock() - g_time > g_refreshRate) || (displayLevel >= 4)) { \
g_time = clock(); \
DISPLAY(__VA_ARGS__); \
} \
}
#undef DISPLAYUPDATE
#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
static clock_t g_time = 0;
/*-*************************************
* Hash Functions
***************************************/
static const U64 prime6bytes = 227718039650203ULL;
static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u << (64-48)) * prime6bytes) >> (64-h)) ; }
static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
/**
* Hash the d-byte value pointed to by p and mod 2^f
* Hash the d-byte value pointed to by p and mod 2^f into the frequency vector
*/
static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 h, unsigned d) {
static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 f, unsigned d) {
if (d == 6) {
return ZSTD_hash6Ptr(p, h) & ((1 << h) - 1);
return ZSTD_hash6Ptr(p, f);
}
return ZSTD_hash8Ptr(p, h) & ((1 << h) - 1);
return ZSTD_hash8Ptr(p, f);
}
@@ -461,20 +463,20 @@ typedef struct FASTCOVER_tryParameters_data_s {
* This function is thread safe if zstd is compiled with multithreaded support.
* It takes its parameters as an *OWNING* opaque pointer to support threading.
*/
static void FASTCOVER_tryParameters(void *opaque)
static void FASTCOVER_tryParameters(void* opaque)
{
/* Save parameters as local variables */
FASTCOVER_tryParameters_data_t *const data = (FASTCOVER_tryParameters_data_t *)opaque;
FASTCOVER_tryParameters_data_t *const data = (FASTCOVER_tryParameters_data_t*)opaque;
const FASTCOVER_ctx_t *const ctx = data->ctx;
const ZDICT_cover_params_t parameters = data->parameters;
size_t dictBufferCapacity = data->dictBufferCapacity;
size_t totalCompressedSize = ERROR(GENERIC);
/* Initialize array to keep track of frequency of dmer within activeSegment */
U16* segmentFreqs = (U16 *)calloc(((U64)1 << ctx->f), sizeof(U16));
U16* segmentFreqs = (U16*)calloc(((U64)1 << ctx->f), sizeof(U16));
/* Allocate space for hash table, dict, and freqs */
BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
BYTE *const dict = (BYTE*)malloc(dictBufferCapacity);
COVER_dictSelection_t selection = COVER_dictSelectionError(ERROR(GENERIC));
U32 *freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
U32* freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
if (!segmentFreqs || !dict || !freqs) {
DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
goto _cleanup;
@@ -486,7 +488,7 @@ static void FASTCOVER_tryParameters(void *opaque)
parameters, segmentFreqs);
const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
selection = COVER_selectDict(dict + tail, dictBufferCapacity - tail,
selection = COVER_selectDict(dict + tail, dictBufferCapacity, dictBufferCapacity - tail,
ctx->samples, ctx->samplesSizes, nbFinalizeSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
totalCompressedSize);
@@ -617,7 +619,7 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
/* constants */
const unsigned nbThreads = parameters->nbThreads;
const double splitPoint =
parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
parameters->splitPoint <= 0.0 ? FASTCOVER_DEFAULT_SPLITPOINT : parameters->splitPoint;
const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
Utilities/cmzstd/lib/dictBuilder/zdict.c
+31 -32
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -23,9 +23,13 @@
/* Unix Large Files support (>4GB) */
#define _FILE_OFFSET_BITS 64
#if (defined(__sun__) && (!defined(__LP64__))) /* Sun Solaris 32-bits requires specific definitions */
# ifndef _LARGEFILE_SOURCE
# define _LARGEFILE_SOURCE
# endif
#elif ! defined(__LP64__) /* No point defining Large file for 64 bit */
# ifndef _LARGEFILE64_SOURCE
# define _LARGEFILE64_SOURCE
# endif
#endif
@@ -37,18 +41,19 @@
#include <stdio.h> /* fprintf, fopen, ftello64 */
#include <time.h> /* clock */
#include "../common/mem.h" /* read */
#include "../common/fse.h" /* FSE_normalizeCount, FSE_writeNCount */
#define HUF_STATIC_LINKING_ONLY
#include "../common/huf.h" /* HUF_buildCTable, HUF_writeCTable */
#include "../common/zstd_internal.h" /* includes zstd.h */
#include "../common/xxhash.h" /* XXH64 */
#include "divsufsort.h"
#ifndef ZDICT_STATIC_LINKING_ONLY
# define ZDICT_STATIC_LINKING_ONLY
#endif
#include "zdict.h"
#define HUF_STATIC_LINKING_ONLY
#include "../common/mem.h" /* read */
#include "../common/fse.h" /* FSE_normalizeCount, FSE_writeNCount */
#include "../common/huf.h" /* HUF_buildCTable, HUF_writeCTable */
#include "../common/zstd_internal.h" /* includes zstd.h */
#include "../common/xxhash.h" /* XXH64 */
#include "../compress/zstd_compress_internal.h" /* ZSTD_loadCEntropy() */
#include "../zdict.h"
#include "divsufsort.h"
/*-*************************************
@@ -62,14 +67,15 @@
#define NOISELENGTH 32
static const int g_compressionLevel_default = 3;
static const U32 g_selectivity_default = 9;
/*-*************************************
* Console display
***************************************/
#undef DISPLAY
#define DISPLAY(...) { fprintf(stderr, __VA_ARGS__); fflush( stderr ); }
#undef DISPLAYLEVEL
#define DISPLAYLEVEL(l, ...) if (notificationLevel>=l) { DISPLAY(__VA_ARGS__); } /* 0 : no display; 1: errors; 2: default; 3: details; 4: debug */
static clock_t ZDICT_clockSpan(clock_t nPrevious) { return clock() - nPrevious; }
@@ -105,20 +111,17 @@ size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize)
size_t headerSize;
if (dictSize <= 8 || MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return ERROR(dictionary_corrupted);
{ unsigned offcodeMaxValue = MaxOff;
ZSTD_compressedBlockState_t* bs = (ZSTD_compressedBlockState_t*)malloc(sizeof(ZSTD_compressedBlockState_t));
{ ZSTD_compressedBlockState_t* bs = (ZSTD_compressedBlockState_t*)malloc(sizeof(ZSTD_compressedBlockState_t));
U32* wksp = (U32*)malloc(HUF_WORKSPACE_SIZE);
short* offcodeNCount = (short*)malloc((MaxOff+1)*sizeof(short));
if (!bs || !wksp || !offcodeNCount) {
if (!bs || !wksp) {
headerSize = ERROR(memory_allocation);
} else {
ZSTD_reset_compressedBlockState(bs);
headerSize = ZSTD_loadCEntropy(bs, wksp, offcodeNCount, &offcodeMaxValue, dictBuffer, dictSize);
headerSize = ZSTD_loadCEntropy(bs, wksp, dictBuffer, dictSize);
}
free(bs);
free(wksp);
free(offcodeNCount);
}
return headerSize;
@@ -532,6 +535,7 @@ static size_t ZDICT_trainBuffer_legacy(dictItem* dictList, U32 dictListSize,
clock_t displayClock = 0;
clock_t const refreshRate = CLOCKS_PER_SEC * 3 / 10;
# undef DISPLAYUPDATE
# define DISPLAYUPDATE(l, ...) if (notificationLevel>=l) { \
if (ZDICT_clockSpan(displayClock) > refreshRate) \
{ displayClock = clock(); DISPLAY(__VA_ARGS__); \
@@ -706,7 +710,7 @@ static void ZDICT_flatLit(unsigned* countLit)
#define OFFCODE_MAX 30 /* only applicable to first block */
static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
unsigned compressionLevel,
int compressionLevel,
const void* srcBuffer, const size_t* fileSizes, unsigned nbFiles,
const void* dictBuffer, size_t dictBufferSize,
unsigned notificationLevel)
@@ -741,7 +745,7 @@ static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
memset(repOffset, 0, sizeof(repOffset));
repOffset[1] = repOffset[4] = repOffset[8] = 1;
memset(bestRepOffset, 0, sizeof(bestRepOffset));
if (compressionLevel==0) compressionLevel = g_compressionLevel_default;
if (compressionLevel==0) compressionLevel = ZSTD_CLEVEL_DEFAULT;
params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
esr.dict = ZSTD_createCDict_advanced(dictBuffer, dictBufferSize, ZSTD_dlm_byRef, ZSTD_dct_rawContent, params.cParams, ZSTD_defaultCMem);
@@ -786,7 +790,7 @@ static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
/* note : the result of this phase should be used to better appreciate the impact on statistics */
total=0; for (u=0; u<=offcodeMax; u++) total+=offcodeCount[u];
errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax);
errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax, /* useLowProbCount */ 1);
if (FSE_isError(errorCode)) {
eSize = errorCode;
DISPLAYLEVEL(1, "FSE_normalizeCount error with offcodeCount \n");
@@ -795,7 +799,7 @@ static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
Offlog = (U32)errorCode;
total=0; for (u=0; u<=MaxML; u++) total+=matchLengthCount[u];
errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML);
errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML, /* useLowProbCount */ 1);
if (FSE_isError(errorCode)) {
eSize = errorCode;
DISPLAYLEVEL(1, "FSE_normalizeCount error with matchLengthCount \n");
@@ -804,7 +808,7 @@ static size_t ZDICT_analyzeEntropy(void* dstBuffer, size_t maxDstSize,
mlLog = (U32)errorCode;
total=0; for (u=0; u<=MaxLL; u++) total+=litLengthCount[u];
errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL);
errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL, /* useLowProbCount */ 1);
if (FSE_isError(errorCode)) {
eSize = errorCode;
DISPLAYLEVEL(1, "FSE_normalizeCount error with litLengthCount \n");
@@ -893,7 +897,7 @@ size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
size_t hSize;
#define HBUFFSIZE 256 /* should prove large enough for all entropy headers */
BYTE header[HBUFFSIZE];
int const compressionLevel = (params.compressionLevel == 0) ? g_compressionLevel_default : params.compressionLevel;
int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
U32 const notificationLevel = params.notificationLevel;
/* check conditions */
@@ -939,7 +943,7 @@ static size_t ZDICT_addEntropyTablesFromBuffer_advanced(
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_params_t params)
{
int const compressionLevel = (params.compressionLevel == 0) ? g_compressionLevel_default : params.compressionLevel;
int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
U32 const notificationLevel = params.notificationLevel;
size_t hSize = 8;
@@ -968,16 +972,11 @@ static size_t ZDICT_addEntropyTablesFromBuffer_advanced(
return MIN(dictBufferCapacity, hSize+dictContentSize);
}
/* Hidden declaration for dbio.c */
size_t ZDICT_trainFromBuffer_unsafe_legacy(
void* dictBuffer, size_t maxDictSize,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_legacy_params_t params);
/*! ZDICT_trainFromBuffer_unsafe_legacy() :
* Warning : `samplesBuffer` must be followed by noisy guard band.
* Warning : `samplesBuffer` must be followed by noisy guard band !!!
* @return : size of dictionary, or an error code which can be tested with ZDICT_isError()
*/
size_t ZDICT_trainFromBuffer_unsafe_legacy(
static size_t ZDICT_trainFromBuffer_unsafe_legacy(
void* dictBuffer, size_t maxDictSize,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_legacy_params_t params)
@@ -1114,8 +1113,8 @@ size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
memset(&params, 0, sizeof(params));
params.d = 8;
params.steps = 4;
/* Default to level 6 since no compression level information is available */
params.zParams.compressionLevel = 3;
/* Use default level since no compression level information is available */
params.zParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
#if defined(DEBUGLEVEL) && (DEBUGLEVEL>=1)
params.zParams.notificationLevel = DEBUGLEVEL;
#endif
Utilities/cmzstd/lib/dictBuilder/zdict.h → Utilities/cmzstd/lib/zdict.h
+152 -5
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -36,6 +36,145 @@ extern "C" {
# define ZDICTLIB_API ZDICTLIB_VISIBILITY
#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
* repetion 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 a 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 dictioanry
* 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.
@@ -64,7 +203,14 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCap
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) */
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():
@@ -264,10 +410,11 @@ typedef struct {
* Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0.
*/
ZDICTLIB_API size_t ZDICT_trainFromBuffer_legacy(
void *dictBuffer, size_t dictBufferCapacity,
const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
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
@@ -279,7 +426,7 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_legacy(
# define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
# if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
# define ZDICT_DEPRECATED(message) [[deprecated(message)]] ZDICTLIB_API
# elif (ZDICT_GCC_VERSION >= 405) || defined(__clang__)
# elif defined(__clang__) || (ZDICT_GCC_VERSION >= 405)
# define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated(message)))
# elif (ZDICT_GCC_VERSION >= 301)
# define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated))
Utilities/cmzstd/lib/zstd.h
+568 -126
View File
File diff suppressed because it is too large Load Diff
Utilities/cmzstd/lib/common/zstd_errors.h → Utilities/cmzstd/lib/zstd_errors.h
+2 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
* Copyright (c) Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -77,6 +77,7 @@ typedef enum {
ZSTD_error_frameIndex_tooLarge = 100,
ZSTD_error_seekableIO = 102,
ZSTD_error_dstBuffer_wrong = 104,
ZSTD_error_srcBuffer_wrong = 105,
ZSTD_error_maxCode = 120 /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
} ZSTD_ErrorCode;