zhangengru/tracy
1
0
Fork 0
mirror of https://github.com/wolfpld/tracy.git synced 2025-03-20 07:40:02 +08:00
Code Issues Projects Wiki Activity

Update zstd to 1.5.4.

Browse Source
This commit is contained in:
Bartosz Taudul
2023-02-10 20:25:40 +01:00
parent 69d0510255
commit 0d6fb703b4
66 changed files with 4541 additions and 3026 deletions
Download Patch File Download Diff File Expand all files Collapse all files
zstd/decompress/huf_decompress.c
+435 -442
View File
File diff suppressed because it is too large Load Diff
zstd/decompress/huf_decompress_amd64.S
+30 -39
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) Facebook, Inc.
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -30,14 +30,14 @@
* TODO: Support Windows calling convention.
*/
ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop)
ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop)
ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop)
ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop)
.global HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop
.global HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop
.global _HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop
.global _HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop
ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X1_usingDTable_internal_fast_asm_loop)
ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X2_usingDTable_internal_fast_asm_loop)
ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X2_usingDTable_internal_fast_asm_loop)
ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X1_usingDTable_internal_fast_asm_loop)
.global HUF_decompress4X1_usingDTable_internal_fast_asm_loop
.global HUF_decompress4X2_usingDTable_internal_fast_asm_loop
.global _HUF_decompress4X1_usingDTable_internal_fast_asm_loop
.global _HUF_decompress4X2_usingDTable_internal_fast_asm_loop
.text
/* Sets up register mappings for clarity.
@@ -95,8 +95,9 @@ ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop)
/* Define both _HUF_* & HUF_* symbols because MacOS
* C symbols are prefixed with '_' & Linux symbols aren't.
*/
_HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop:
HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop:
_HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
ZSTD_CET_ENDBRANCH
/* Save all registers - even if they are callee saved for simplicity. */
push %rax
push %rbx
@@ -350,8 +351,9 @@ HUF_decompress4X1_usingDTable_internal_bmi2_asm_loop:
pop %rax
ret
_HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop:
HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop:
_HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
ZSTD_CET_ENDBRANCH
/* Save all registers - even if they are callee saved for simplicity. */
push %rax
push %rbx
@@ -427,41 +429,30 @@ HUF_decompress4X2_usingDTable_internal_bmi2_asm_loop:
/* r15 = (ip0 - ilimit) / 7 */
movq %rdx, %r15
movabsq $-3689348814741910323, %rdx
movq 8(%rsp), %rax /* rax = oend0 */
subq %op0, %rax /* rax = oend0 - op0 */
mulq %rdx
shrq $3, %rdx /* rdx = rax / 10 */
/* r15 = min(r15, min(oend0 - op0, oend1 - op1, oend2 - op2, oend3 - op3) / 10) */
movq 8(%rsp), %rax /* rax = oend0 */
subq %op0, %rax /* rax = oend0 - op0 */
movq 16(%rsp), %rdx /* rdx = oend1 */
subq %op1, %rdx /* rdx = oend1 - op1 */
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
cmpq %rax, %rdx
cmova %rax, %rdx /* rdx = min(%rdx, %rax) */
movabsq $-3689348814741910323, %rdx
movq 16(%rsp), %rax /* rax = oend1 */
subq %op1, %rax /* rax = oend1 - op1 */
mulq %rdx
shrq $3, %rdx /* rdx = rax / 10 */
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
movabsq $-3689348814741910323, %rdx
movq 24(%rsp), %rax /* rax = oend2 */
subq %op2, %rax /* rax = oend2 - op2 */
mulq %rdx
shrq $3, %rdx /* rdx = rax / 10 */
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
cmpq %rax, %rdx
cmova %rax, %rdx /* rdx = min(%rdx, %rax) */
movabsq $-3689348814741910323, %rdx
movq 32(%rsp), %rax /* rax = oend3 */
subq %op3, %rax /* rax = oend3 - op3 */
cmpq %rax, %rdx
cmova %rax, %rdx /* rdx = min(%rdx, %rax) */
movabsq $-3689348814741910323, %rax
mulq %rdx
shrq $3, %rdx /* rdx = rax / 10 */
shrq $3, %rdx /* rdx = rdx / 10 */
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
zstd/decompress/zstd_ddict.c
+3 -4
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) Yann Collet, Facebook, Inc.
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -19,7 +19,6 @@
#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 "zstd_decompress_internal.h"
#include "zstd_ddict.h"
@@ -134,7 +133,7 @@ static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
ZSTD_memcpy(internalBuffer, dict, dictSize);
}
ddict->dictSize = dictSize;
ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001); /* cover both little and big endian */
ddict->entropy.hufTable[0] = (HUF_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001); /* cover both little and big endian */
/* parse dictionary content */
FORWARD_IF_ERROR( ZSTD_loadEntropy_intoDDict(ddict, dictContentType) , "");
@@ -240,5 +239,5 @@ size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict)
unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict)
{
if (ddict==NULL) return 0;
return ZSTD_getDictID_fromDict(ddict->dictContent, ddict->dictSize);
return ddict->dictID;
}
zstd/decompress/zstd_ddict.h
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) Yann Collet, Facebook, Inc.
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
zstd/decompress/zstd_decompress.c
+164 -42
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) Yann Collet, Facebook, Inc.
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -59,13 +59,13 @@
#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 */
#include "zstd_decompress_block.h" /* ZSTD_decompressBlock_internal */
#include "../common/bits.h" /* ZSTD_highbit32 */
#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
# include "../legacy/zstd_legacy.h"
@@ -78,11 +78,11 @@
*************************************/
#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_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
@@ -243,6 +243,7 @@ static void ZSTD_DCtx_resetParameters(ZSTD_DCtx* dctx)
dctx->outBufferMode = ZSTD_bm_buffered;
dctx->forceIgnoreChecksum = ZSTD_d_validateChecksum;
dctx->refMultipleDDicts = ZSTD_rmd_refSingleDDict;
dctx->disableHufAsm = 0;
}
static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
@@ -438,16 +439,40 @@ size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
* note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless
* @return : 0, `zfhPtr` is correctly filled,
* >0, `srcSize` is too small, value is wanted `srcSize` amount,
* or an error code, which can be tested using ZSTD_isError() */
** or an error code, which can be tested using ZSTD_isError() */
size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
{
const BYTE* ip = (const BYTE*)src;
size_t const minInputSize = ZSTD_startingInputLength(format);
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");
DEBUGLOG(5, "ZSTD_getFrameHeader_advanced: minInputSize = %zu, srcSize = %zu", minInputSize, srcSize);
if (srcSize > 0) {
/* note : technically could be considered an assert(), since it's an invalid entry */
RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter : src==NULL, but srcSize>0");
}
if (srcSize < minInputSize) {
if (srcSize > 0 && format != ZSTD_f_zstd1_magicless) {
/* when receiving less than @minInputSize bytes,
* control these bytes at least correspond to a supported magic number
* in order to error out early if they don't.
**/
size_t const toCopy = MIN(4, srcSize);
unsigned char hbuf[4]; MEM_writeLE32(hbuf, ZSTD_MAGICNUMBER);
assert(src != NULL);
ZSTD_memcpy(hbuf, src, toCopy);
if ( MEM_readLE32(hbuf) != ZSTD_MAGICNUMBER ) {
/* not a zstd frame : let's check if it's a skippable frame */
MEM_writeLE32(hbuf, ZSTD_MAGIC_SKIPPABLE_START);
ZSTD_memcpy(hbuf, src, toCopy);
if ((MEM_readLE32(hbuf) & ZSTD_MAGIC_SKIPPABLE_MASK) != ZSTD_MAGIC_SKIPPABLE_START) {
RETURN_ERROR(prefix_unknown,
"first bytes don't correspond to any supported magic number");
} } }
return minInputSize;
}
ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr)); /* not strictly necessary, but static analyzers may not understand that zfhPtr will be read only if return value is zero, since they are 2 different signals */
if ( (format != ZSTD_f_zstd1_magicless)
&& (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
@@ -757,10 +782,11 @@ static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize
ip += 4;
}
frameSizeInfo.nbBlocks = nbBlocks;
frameSizeInfo.compressedSize = (size_t)(ip - ipstart);
frameSizeInfo.decompressedBound = (zfh.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN)
? zfh.frameContentSize
: nbBlocks * zfh.blockSizeMax;
: (unsigned long long)nbBlocks * zfh.blockSizeMax;
return frameSizeInfo;
}
}
@@ -800,6 +826,48 @@ unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
return bound;
}
size_t ZSTD_decompressionMargin(void const* src, size_t srcSize)
{
size_t margin = 0;
unsigned maxBlockSize = 0;
/* Iterate over each frame */
while (srcSize > 0) {
ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize);
size_t const compressedSize = frameSizeInfo.compressedSize;
unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
ZSTD_frameHeader zfh;
FORWARD_IF_ERROR(ZSTD_getFrameHeader(&zfh, src, srcSize), "");
if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
return ERROR(corruption_detected);
if (zfh.frameType == ZSTD_frame) {
/* Add the frame header to our margin */
margin += zfh.headerSize;
/* Add the checksum to our margin */
margin += zfh.checksumFlag ? 4 : 0;
/* Add 3 bytes per block */
margin += 3 * frameSizeInfo.nbBlocks;
/* Compute the max block size */
maxBlockSize = MAX(maxBlockSize, zfh.blockSizeMax);
} else {
assert(zfh.frameType == ZSTD_skippableFrame);
/* Add the entire skippable frame size to our margin. */
margin += compressedSize;
}
assert(srcSize >= compressedSize);
src = (const BYTE*)src + compressedSize;
srcSize -= compressedSize;
}
/* Add the max block size back to the margin. */
margin += maxBlockSize;
return margin;
}
/*-*************************************************************
* Frame decoding
@@ -825,7 +893,7 @@ static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
if (srcSize == 0) return 0;
RETURN_ERROR(dstBuffer_null, "");
}
ZSTD_memcpy(dst, src, srcSize);
ZSTD_memmove(dst, src, srcSize);
return srcSize;
}
@@ -903,6 +971,7 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
/* Loop on each block */
while (1) {
BYTE* oBlockEnd = oend;
size_t decodedSize;
blockProperties_t blockProperties;
size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSrcSize, &blockProperties);
@@ -912,16 +981,34 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
remainingSrcSize -= ZSTD_blockHeaderSize;
RETURN_ERROR_IF(cBlockSize > remainingSrcSize, srcSize_wrong, "");
if (ip >= op && ip < oBlockEnd) {
/* We are decompressing in-place. Limit the output pointer so that we
* don't overwrite the block that we are currently reading. This will
* fail decompression if the input & output pointers aren't spaced
* far enough apart.
*
* This is important to set, even when the pointers are far enough
* apart, because ZSTD_decompressBlock_internal() can decide to store
* literals in the output buffer, after the block it is decompressing.
* Since we don't want anything to overwrite our input, we have to tell
* ZSTD_decompressBlock_internal to never write past ip.
*
* See ZSTD_allocateLiteralsBuffer() for reference.
*/
oBlockEnd = op + (ip - op);
}
switch(blockProperties.blockType)
{
case bt_compressed:
decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oend-op), ip, cBlockSize, /* frame */ 1, not_streaming);
decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oBlockEnd-op), ip, cBlockSize, /* frame */ 1, not_streaming);
break;
case bt_raw :
/* Use oend instead of oBlockEnd because this function is safe to overlap. It uses memmove. */
decodedSize = ZSTD_copyRawBlock(op, (size_t)(oend-op), ip, cBlockSize);
break;
case bt_rle :
decodedSize = ZSTD_setRleBlock(op, (size_t)(oend-op), *ip, blockProperties.origSize);
decodedSize = ZSTD_setRleBlock(op, (size_t)(oBlockEnd-op), *ip, blockProperties.origSize);
break;
case bt_reserved :
default:
@@ -956,6 +1043,7 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
}
ZSTD_DCtx_trace_end(dctx, (U64)(op-ostart), (U64)(ip-istart), /* streaming */ 0);
/* Allow caller to get size read */
DEBUGLOG(4, "ZSTD_decompressFrame: decompressed frame of size %zi, consuming %zi bytes of input", op-ostart, ip - (const BYTE*)*srcPtr);
*srcPtr = ip;
*srcSizePtr = remainingSrcSize;
return (size_t)(op-ostart);
@@ -1108,8 +1196,8 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx) { return dctx->expected; }
/**
* Similar to ZSTD_nextSrcSizeToDecompress(), but when when a block input can be streamed,
* we allow taking a partial block as the input. Currently only raw uncompressed blocks can
* Similar to ZSTD_nextSrcSizeToDecompress(), but when a block input can be streamed, we
* allow taking a partial block as the input. Currently only raw uncompressed blocks can
* be streamed.
*
* For blocks that can be streamed, this allows us to reduce the latency until we produce
@@ -1309,7 +1397,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
default:
assert(0); /* impossible */
RETURN_ERROR(GENERIC, "impossible to reach"); /* some compiler require default to do something */
RETURN_ERROR(GENERIC, "impossible to reach"); /* some compilers require default to do something */
}
}
@@ -1350,11 +1438,11 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
/* in minimal huffman, we always use X1 variants */
size_t const hSize = HUF_readDTableX1_wksp(entropy->hufTable,
dictPtr, dictEnd - dictPtr,
workspace, workspaceSize);
workspace, workspaceSize, /* flags */ 0);
#else
size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
dictPtr, (size_t)(dictEnd - dictPtr),
workspace, workspaceSize);
workspace, workspaceSize, /* flags */ 0);
#endif
RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted, "");
dictPtr += hSize;
@@ -1453,7 +1541,7 @@ size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
dctx->prefixStart = NULL;
dctx->virtualStart = NULL;
dctx->dictEnd = NULL;
dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001); /* cover both little and big endian */
dctx->entropy.hufTable[0] = (HUF_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001); /* cover both little and big endian */
dctx->litEntropy = dctx->fseEntropy = 0;
dctx->dictID = 0;
dctx->bType = bt_reserved;
@@ -1515,7 +1603,7 @@ unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
* This could for one of the following reasons :
* - The frame does not require a dictionary (most common case).
* - The frame was built with dictID intentionally removed.
* Needed dictionary is a hidden information.
* Needed dictionary is a hidden piece of information.
* Note : this use case also happens when using a non-conformant dictionary.
* - `srcSize` is too small, and as a result, frame header could not be decoded.
* Note : possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`.
@@ -1524,7 +1612,7 @@ unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
* ZSTD_getFrameHeader(), which will provide a more precise error code. */
unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize)
{
ZSTD_frameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0 };
ZSTD_frameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0, 0, 0 };
size_t const hError = ZSTD_getFrameHeader(&zfp, src, srcSize);
if (ZSTD_isError(hError)) return 0;
return zfp.dictID;
@@ -1631,7 +1719,9 @@ size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t di
size_t ZSTD_initDStream(ZSTD_DStream* zds)
{
DEBUGLOG(4, "ZSTD_initDStream");
return ZSTD_initDStream_usingDDict(zds, NULL);
FORWARD_IF_ERROR(ZSTD_DCtx_reset(zds, ZSTD_reset_session_only), "");
FORWARD_IF_ERROR(ZSTD_DCtx_refDDict(zds, NULL), "");
return ZSTD_startingInputLength(zds->format);
}
/* ZSTD_initDStream_usingDDict() :
@@ -1639,6 +1729,7 @@ size_t ZSTD_initDStream(ZSTD_DStream* zds)
* this function cannot fail */
size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
{
DEBUGLOG(4, "ZSTD_initDStream_usingDDict");
FORWARD_IF_ERROR( ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only) , "");
FORWARD_IF_ERROR( ZSTD_DCtx_refDDict(dctx, ddict) , "");
return ZSTD_startingInputLength(dctx->format);
@@ -1649,6 +1740,7 @@ size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
* this function cannot fail */
size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
{
DEBUGLOG(4, "ZSTD_resetDStream");
FORWARD_IF_ERROR(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only), "");
return ZSTD_startingInputLength(dctx->format);
}
@@ -1720,6 +1812,11 @@ ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
bounds.lowerBound = (int)ZSTD_rmd_refSingleDDict;
bounds.upperBound = (int)ZSTD_rmd_refMultipleDDicts;
return bounds;
case ZSTD_d_disableHuffmanAssembly:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
default:;
}
bounds.error = ERROR(parameter_unsupported);
@@ -1760,6 +1857,9 @@ size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value
case ZSTD_d_refMultipleDDicts:
*value = (int)dctx->refMultipleDDicts;
return 0;
case ZSTD_d_disableHuffmanAssembly:
*value = (int)dctx->disableHufAsm;
return 0;
default:;
}
RETURN_ERROR(parameter_unsupported, "");
@@ -1793,6 +1893,10 @@ size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value
}
dctx->refMultipleDDicts = (ZSTD_refMultipleDDicts_e)value;
return 0;
case ZSTD_d_disableHuffmanAssembly:
CHECK_DBOUNDS(ZSTD_d_disableHuffmanAssembly, value);
dctx->disableHufAsm = value != 0;
return 0;
default:;
}
RETURN_ERROR(parameter_unsupported, "");
@@ -1980,7 +2084,6 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
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)
U32 const legacyVersion = ZSTD_isLegacy(istart, iend-istart);
@@ -2012,6 +2115,11 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
zds->lhSize += remainingInput;
}
input->pos = input->size;
/* check first few bytes */
FORWARD_IF_ERROR(
ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format),
"First few bytes detected incorrect" );
/* return hint 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);
@@ -2029,8 +2137,9 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
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()")
assert(istart != NULL);
ip = istart + cSize;
op += decompressedSize;
op = op ? op + decompressedSize : op; /* can occur if frameContentSize = 0 (empty frame) */
zds->expected = 0;
zds->streamStage = zdss_init;
someMoreWork = 0;
@@ -2114,6 +2223,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
}
if ((size_t)(iend-ip) >= neededInSize) { /* decode directly from src */
FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, ip, neededInSize), "");
assert(ip != NULL);
ip += neededInSize;
/* Function modifies the stage so we must break */
break;
@@ -2128,7 +2238,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
int const isSkipFrame = ZSTD_isSkipFrame(zds);
size_t loadedSize;
/* At this point we shouldn't be decompressing a block that we can stream. */
assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, iend - ip));
assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip)));
if (isSkipFrame) {
loadedSize = MIN(toLoad, (size_t)(iend-ip));
} else {
@@ -2137,8 +2247,11 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
"should never happen");
loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, (size_t)(iend-ip));
}
ip += loadedSize;
zds->inPos += loadedSize;
if (loadedSize != 0) {
/* ip may be NULL */
ip += loadedSize;
zds->inPos += loadedSize;
}
if (loadedSize < toLoad) { someMoreWork = 0; break; } /* not enough input, wait for more */
/* decode loaded input */
@@ -2148,14 +2261,17 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
break;
}
case zdss_flush:
{ size_t const toFlushSize = zds->outEnd - zds->outStart;
{
size_t const toFlushSize = zds->outEnd - zds->outStart;
size_t const flushedSize = ZSTD_limitCopy(op, (size_t)(oend-op), zds->outBuff + zds->outStart, toFlushSize);
op += flushedSize;
op = op ? op + flushedSize : op;
zds->outStart += flushedSize;
if (flushedSize == toFlushSize) { /* flush completed */
zds->streamStage = zdss_read;
if ( (zds->outBuffSize < zds->fParams.frameContentSize)
&& (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
&& (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
DEBUGLOG(5, "restart filling outBuff from beginning (left:%i, needed:%u)",
(int)(zds->outBuffSize - zds->outStart),
(U32)zds->fParams.blockSizeMax);
@@ -2169,7 +2285,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
default:
assert(0); /* impossible */
RETURN_ERROR(GENERIC, "impossible to reach"); /* some compiler require default to do something */
RETURN_ERROR(GENERIC, "impossible to reach"); /* some compilers require default to do something */
} }
/* result */
@@ -2182,8 +2298,8 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
if ((ip==istart) && (op==ostart)) { /* no forward progress */
zds->noForwardProgress ++;
if (zds->noForwardProgress >= ZSTD_NO_FORWARD_PROGRESS_MAX) {
RETURN_ERROR_IF(op==oend, dstSize_tooSmall, "");
RETURN_ERROR_IF(ip==iend, srcSize_wrong, "");
RETURN_ERROR_IF(op==oend, noForwardProgress_destFull, "");
RETURN_ERROR_IF(ip==iend, noForwardProgress_inputEmpty, "");
assert(0);
}
} else {
@@ -2220,11 +2336,17 @@ size_t ZSTD_decompressStream_simpleArgs (
void* dst, size_t dstCapacity, size_t* dstPos,
const void* src, size_t srcSize, size_t* srcPos)
{
ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
ZSTD_inBuffer input = { src, srcSize, *srcPos };
/* ZSTD_compress_generic() will check validity of dstPos and srcPos */
size_t const cErr = ZSTD_decompressStream(dctx, &output, &input);
*dstPos = output.pos;
*srcPos = input.pos;
return cErr;
ZSTD_outBuffer output;
ZSTD_inBuffer input;
output.dst = dst;
output.size = dstCapacity;
output.pos = *dstPos;
input.src = src;
input.size = srcSize;
input.pos = *srcPos;
{ size_t const cErr = ZSTD_decompressStream(dctx, &output, &input);
*dstPos = output.pos;
*srcPos = input.pos;
return cErr;
}
}
zstd/decompress/zstd_decompress_block.c
+186 -65
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) Yann Collet, Facebook, Inc.
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -20,12 +20,12 @@
#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/zstd_internal.h"
#include "zstd_decompress_internal.h" /* ZSTD_DCtx */
#include "zstd_ddict.h" /* ZSTD_DDictDictContent */
#include "zstd_decompress_block.h"
#include "../common/bits.h" /* ZSTD_highbit32 */
/*_*******************************************************
* Macros
@@ -89,7 +89,7 @@ static void ZSTD_allocateLiteralsBuffer(ZSTD_DCtx* dctx, void* const dst, const
dctx->litBufferEnd = dctx->litBuffer + litSize - ZSTD_LITBUFFEREXTRASIZE;
}
else {
/* initially this will be stored entirely in dst during huffman decoding, it will partially shifted to litExtraBuffer after */
/* initially this will be stored entirely in dst during huffman decoding, it will partially be shifted to litExtraBuffer after */
dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize;
dctx->litBufferEnd = (BYTE*)dst + expectedWriteSize;
}
@@ -134,13 +134,16 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
ZSTD_FALLTHROUGH;
case set_compressed:
RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3");
RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need up to 5 for case 3");
{ size_t lhSize, litSize, litCSize;
U32 singleStream=0;
U32 const lhlCode = (istart[0] >> 2) & 3;
U32 const lhc = MEM_readLE32(istart);
size_t hufSuccess;
size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
int const flags = 0
| (ZSTD_DCtx_get_bmi2(dctx) ? HUF_flags_bmi2 : 0)
| (dctx->disableHufAsm ? HUF_flags_disableAsm : 0);
switch(lhlCode)
{
case 0: case 1: default: /* note : default is impossible, since lhlCode into [0..3] */
@@ -165,6 +168,10 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
}
RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
if (!singleStream)
RETURN_ERROR_IF(litSize < MIN_LITERALS_FOR_4_STREAMS, literals_headerWrong,
"Not enough literals (%zu) for the 4-streams mode (min %u)",
litSize, MIN_LITERALS_FOR_4_STREAMS);
RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected, "");
RETURN_ERROR_IF(expectedWriteSize < litSize , dstSize_tooSmall, "");
ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 0);
@@ -176,13 +183,14 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
if (litEncType==set_repeat) {
if (singleStream) {
hufSuccess = HUF_decompress1X_usingDTable_bmi2(
hufSuccess = HUF_decompress1X_usingDTable(
dctx->litBuffer, litSize, istart+lhSize, litCSize,
dctx->HUFptr, ZSTD_DCtx_get_bmi2(dctx));
dctx->HUFptr, flags);
} else {
hufSuccess = HUF_decompress4X_usingDTable_bmi2(
assert(litSize >= MIN_LITERALS_FOR_4_STREAMS);
hufSuccess = HUF_decompress4X_usingDTable(
dctx->litBuffer, litSize, istart+lhSize, litCSize,
dctx->HUFptr, ZSTD_DCtx_get_bmi2(dctx));
dctx->HUFptr, flags);
}
} else {
if (singleStream) {
@@ -190,18 +198,18 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
hufSuccess = HUF_decompress1X_DCtx_wksp(
dctx->entropy.hufTable, dctx->litBuffer, litSize,
istart+lhSize, litCSize, dctx->workspace,
sizeof(dctx->workspace));
sizeof(dctx->workspace), flags);
#else
hufSuccess = HUF_decompress1X1_DCtx_wksp_bmi2(
hufSuccess = HUF_decompress1X1_DCtx_wksp(
dctx->entropy.hufTable, dctx->litBuffer, litSize,
istart+lhSize, litCSize, dctx->workspace,
sizeof(dctx->workspace), ZSTD_DCtx_get_bmi2(dctx));
sizeof(dctx->workspace), flags);
#endif
} else {
hufSuccess = HUF_decompress4X_hufOnly_wksp_bmi2(
hufSuccess = HUF_decompress4X_hufOnly_wksp(
dctx->entropy.hufTable, dctx->litBuffer, litSize,
istart+lhSize, litCSize, dctx->workspace,
sizeof(dctx->workspace), ZSTD_DCtx_get_bmi2(dctx));
sizeof(dctx->workspace), flags);
}
}
if (dctx->litBufferLocation == ZSTD_split)
@@ -237,6 +245,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
break;
case 3:
lhSize = 3;
RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize = 3");
litSize = MEM_readLE24(istart) >> 4;
break;
}
@@ -279,12 +288,13 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
break;
case 1:
lhSize = 2;
RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 3");
litSize = MEM_readLE16(istart) >> 4;
break;
case 3:
lhSize = 3;
RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 4");
litSize = MEM_readLE24(istart) >> 4;
RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4");
break;
}
RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
@@ -506,14 +516,15 @@ void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
for (i = 8; i < n; i += 8) {
MEM_write64(spread + pos + i, sv);
}
pos += n;
assert(n>=0);
pos += (size_t)n;
}
}
/* Now we spread those positions across the table.
* The benefit of doing it in two stages is that we avoid the the
* The benefit of doing it in two stages is that we avoid the
* variable size inner loop, which caused lots of branch misses.
* Now we can run through all the positions without any branch misses.
* We unroll the loop twice, since that is what emperically worked best.
* We unroll the loop twice, since that is what empirically worked best.
*/
{
size_t position = 0;
@@ -540,7 +551,7 @@ void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
for (i=0; i<n; i++) {
tableDecode[position].baseValue = s;
position = (position + step) & tableMask;
while (position > highThreshold) position = (position + step) & tableMask; /* lowprob area */
while (UNLIKELY(position > highThreshold)) position = (position + step) & tableMask; /* lowprob area */
} }
assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
}
@@ -551,7 +562,7 @@ void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
for (u=0; u<tableSize; u++) {
U32 const symbol = tableDecode[u].baseValue;
U32 const nextState = symbolNext[symbol]++;
tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
tableDecode[u].nbBits = (BYTE) (tableLog - ZSTD_highbit32(nextState) );
tableDecode[u].nextState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
assert(nbAdditionalBits[symbol] < 255);
tableDecode[u].nbAdditionalBits = nbAdditionalBits[symbol];
@@ -964,6 +975,11 @@ size_t ZSTD_execSequence(BYTE* op,
assert(op != NULL /* Precondition */);
assert(oend_w < oend /* No underflow */);
#if defined(__aarch64__)
/* prefetch sequence starting from match that will be used for copy later */
PREFETCH_L1(match);
#endif
/* Handle edge cases in a slow path:
* - Read beyond end of literals
* - Match end is within WILDCOPY_OVERLIMIT of oend
@@ -1154,7 +1170,7 @@ ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, U16
}
/* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
* offset bits. But we can only read at most (STREAM_ACCUMULATOR_MIN_32 - 1)
* offset bits. But we can only read at most STREAM_ACCUMULATOR_MIN_32
* bits before reloading. This value is the maximum number of bytes we read
* after reloading when we are decoding long offsets.
*/
@@ -1169,9 +1185,27 @@ FORCE_INLINE_TEMPLATE seq_t
ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
{
seq_t seq;
/*
* ZSTD_seqSymbol is a structure with a total of 64 bits wide. So it can be
* loaded in one operation and extracted its fields by simply shifting or
* bit-extracting on aarch64.
* GCC doesn't recognize this and generates more unnecessary ldr/ldrb/ldrh
* operations that cause performance drop. This can be avoided by using this
* ZSTD_memcpy hack.
*/
#if defined(__aarch64__) && (defined(__GNUC__) && !defined(__clang__))
ZSTD_seqSymbol llDInfoS, mlDInfoS, ofDInfoS;
ZSTD_seqSymbol* const llDInfo = &llDInfoS;
ZSTD_seqSymbol* const mlDInfo = &mlDInfoS;
ZSTD_seqSymbol* const ofDInfo = &ofDInfoS;
ZSTD_memcpy(llDInfo, seqState->stateLL.table + seqState->stateLL.state, sizeof(ZSTD_seqSymbol));
ZSTD_memcpy(mlDInfo, seqState->stateML.table + seqState->stateML.state, sizeof(ZSTD_seqSymbol));
ZSTD_memcpy(ofDInfo, seqState->stateOffb.table + seqState->stateOffb.state, sizeof(ZSTD_seqSymbol));
#else
const ZSTD_seqSymbol* const llDInfo = seqState->stateLL.table + seqState->stateLL.state;
const ZSTD_seqSymbol* const mlDInfo = seqState->stateML.table + seqState->stateML.state;
const ZSTD_seqSymbol* const ofDInfo = seqState->stateOffb.table + seqState->stateOffb.state;
#endif
seq.matchLength = mlDInfo->baseValue;
seq.litLength = llDInfo->baseValue;
{ U32 const ofBase = ofDInfo->baseValue;
@@ -1186,9 +1220,13 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
U32 const llnbBits = llDInfo->nbBits;
U32 const mlnbBits = mlDInfo->nbBits;
U32 const ofnbBits = ofDInfo->nbBits;
assert(llBits <= MaxLLBits);
assert(mlBits <= MaxMLBits);
assert(ofBits <= MaxOff);
/*
* As gcc has better branch and block analyzers, sometimes it is only
* valuable to mark likelyness for clang, it gives around 3-4% of
* valuable to mark likeliness for clang, it gives around 3-4% of
* performance.
*/
@@ -1201,13 +1239,16 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
#endif
ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
assert(ofBits <= MaxOff);
ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 > LONG_OFFSETS_MAX_EXTRA_BITS_32);
ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 - LONG_OFFSETS_MAX_EXTRA_BITS_32 >= MaxMLBits);
if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed);
/* Always read extra bits, this keeps the logic simple,
* avoids branches, and avoids accidentally reading 0 bits.
*/
U32 const extraBits = LONG_OFFSETS_MAX_EXTRA_BITS_32;
offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
BIT_reloadDStream(&seqState->DStream);
if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32); /* to avoid another reload */
offset += BIT_readBitsFast(&seqState->DStream, extraBits);
} else {
offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/); /* <= (ZSTD_WINDOWLOG_MAX-1) bits */
if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
@@ -1552,7 +1593,7 @@ ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
const BYTE* const prefixStart = (const BYTE*)(dctx->prefixStart);
const BYTE* const vBase = (const BYTE*)(dctx->virtualStart);
const BYTE* const dictEnd = (const BYTE*)(dctx->dictEnd);
DEBUGLOG(5, "ZSTD_decompressSequences_body");
DEBUGLOG(5, "ZSTD_decompressSequences_body: nbSeq = %d", nbSeq);
(void)frame;
/* Regen sequences */
@@ -1945,34 +1986,79 @@ ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
/**
* @returns The total size of the history referencable by zstd, including
* both the prefix and the extDict. At @p op any offset larger than this
* is invalid.
*/
static size_t ZSTD_totalHistorySize(BYTE* op, BYTE const* virtualStart)
{
return (size_t)(op - virtualStart);
}
#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
!defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
/* ZSTD_getLongOffsetsShare() :
typedef struct {
unsigned longOffsetShare;
unsigned maxNbAdditionalBits;
} ZSTD_OffsetInfo;
/* ZSTD_getOffsetInfo() :
* condition : offTable must be valid
* @return : "share" of long offsets (arbitrarily defined as > (1<<23))
* compared to maximum possible of (1<<OffFSELog) */
static unsigned
ZSTD_getLongOffsetsShare(const ZSTD_seqSymbol* offTable)
* compared to maximum possible of (1<<OffFSELog),
* as well as the maximum number additional bits required.
*/
static ZSTD_OffsetInfo
ZSTD_getOffsetInfo(const ZSTD_seqSymbol* offTable, int nbSeq)
{
const void* ptr = offTable;
U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
const ZSTD_seqSymbol* table = offTable + 1;
U32 const max = 1 << tableLog;
U32 u, total = 0;
DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
ZSTD_OffsetInfo info = {0, 0};
/* If nbSeq == 0, then the offTable is uninitialized, but we have
* no sequences, so both values should be 0.
*/
if (nbSeq != 0) {
const void* ptr = offTable;
U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
const ZSTD_seqSymbol* table = offTable + 1;
U32 const max = 1 << tableLog;
U32 u;
DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
assert(max <= (1 << OffFSELog)); /* max not too large */
for (u=0; u<max; u++) {
if (table[u].nbAdditionalBits > 22) total += 1;
assert(max <= (1 << OffFSELog)); /* max not too large */
for (u=0; u<max; u++) {
info.maxNbAdditionalBits = MAX(info.maxNbAdditionalBits, table[u].nbAdditionalBits);
if (table[u].nbAdditionalBits > 22) info.longOffsetShare += 1;
}
assert(tableLog <= OffFSELog);
info.longOffsetShare <<= (OffFSELog - tableLog); /* scale to OffFSELog */
}
assert(tableLog <= OffFSELog);
total <<= (OffFSELog - tableLog); /* scale to OffFSELog */
return total;
return info;
}
/**
* @returns The maximum offset we can decode in one read of our bitstream, without
* reloading more bits in the middle of the offset bits read. Any offsets larger
* than this must use the long offset decoder.
*/
static size_t ZSTD_maxShortOffset(void)
{
if (MEM_64bits()) {
/* We can decode any offset without reloading bits.
* This might change if the max window size grows.
*/
ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
return (size_t)-1;
} else {
/* The maximum offBase is (1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1.
* This offBase would require STREAM_ACCUMULATOR_MIN extra bits.
* Then we have to subtract ZSTD_REP_NUM to get the maximum possible offset.
*/
size_t const maxOffbase = ((size_t)1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1;
size_t const maxOffset = maxOffbase - ZSTD_REP_NUM;
assert(ZSTD_highbit32((U32)maxOffbase) == STREAM_ACCUMULATOR_MIN);
return maxOffset;
}
}
#endif
size_t
ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
@@ -1980,20 +2066,21 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
const void* src, size_t srcSize, const int frame, const streaming_operation streaming)
{ /* blockType == blockCompressed */
const BYTE* ip = (const BYTE*)src;
/* isLongOffset must be true if there are long offsets.
* Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
* We don't expect that to be the case in 64-bit mode.
* In block mode, window size is not known, so we have to be conservative.
* (note: but it could be evaluated from current-lowLimit)
*/
ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || (dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN))));
DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
RETURN_ERROR_IF(srcSize >= ZSTD_BLOCKSIZE_MAX, srcSize_wrong, "");
/* Note : the wording of the specification
* allows compressed block to be sized exactly ZSTD_BLOCKSIZE_MAX.
* This generally does not happen, as it makes little sense,
* since an uncompressed block would feature same size and have no decompression cost.
* Also, note that decoder from reference libzstd before < v1.5.4
* would consider this edge case as an error.
* As a consequence, avoid generating compressed blocks of size ZSTD_BLOCKSIZE_MAX
* for broader compatibility with the deployed ecosystem of zstd decoders */
RETURN_ERROR_IF(srcSize > ZSTD_BLOCKSIZE_MAX, srcSize_wrong, "");
/* Decode literals section */
{ size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize, dst, dstCapacity, streaming);
DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize);
DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : cSize=%u, nbLiterals=%zu", (U32)litCSize, dctx->litSize);
if (ZSTD_isError(litCSize)) return litCSize;
ip += litCSize;
srcSize -= litCSize;
@@ -2001,6 +2088,23 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
/* Build Decoding Tables */
{
/* Compute the maximum block size, which must also work when !frame and fParams are unset.
* Additionally, take the min with dstCapacity to ensure that the totalHistorySize fits in a size_t.
*/
size_t const blockSizeMax = MIN(dstCapacity, (frame ? dctx->fParams.blockSizeMax : ZSTD_BLOCKSIZE_MAX));
size_t const totalHistorySize = ZSTD_totalHistorySize((BYTE*)dst + blockSizeMax, (BYTE const*)dctx->virtualStart);
/* isLongOffset must be true if there are long offsets.
* Offsets are long if they are larger than ZSTD_maxShortOffset().
* We don't expect that to be the case in 64-bit mode.
*
* We check here to see if our history is large enough to allow long offsets.
* If it isn't, then we can't possible have (valid) long offsets. If the offset
* is invalid, then it is okay to read it incorrectly.
*
* If isLongOffsets is true, then we will later check our decoding table to see
* if it is even possible to generate long offsets.
*/
ZSTD_longOffset_e isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (totalHistorySize > ZSTD_maxShortOffset()));
/* These macros control at build-time which decompressor implementation
* we use. If neither is defined, we do some inspection and dispatch at
* runtime.
@@ -2008,6 +2112,11 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
!defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
int usePrefetchDecoder = dctx->ddictIsCold;
#else
/* Set to 1 to avoid computing offset info if we don't need to.
* Otherwise this value is ignored.
*/
int usePrefetchDecoder = 1;
#endif
int nbSeq;
size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
@@ -2017,26 +2126,38 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
RETURN_ERROR_IF(dst == NULL && nbSeq > 0, dstSize_tooSmall, "NULL not handled");
#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
!defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
if ( !usePrefetchDecoder
&& (!frame || (dctx->fParams.windowSize > (1<<24)))
&& (nbSeq>ADVANCED_SEQS) ) { /* could probably use a larger nbSeq limit */
U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr);
U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
usePrefetchDecoder = (shareLongOffsets >= minShare);
/* If we could potentially have long offsets, or we might want to use the prefetch decoder,
* compute information about the share of long offsets, and the maximum nbAdditionalBits.
* NOTE: could probably use a larger nbSeq limit
*/
if (isLongOffset || (!usePrefetchDecoder && (totalHistorySize > (1u << 24)) && (nbSeq > 8))) {
ZSTD_OffsetInfo const info = ZSTD_getOffsetInfo(dctx->OFTptr, nbSeq);
if (isLongOffset && info.maxNbAdditionalBits <= STREAM_ACCUMULATOR_MIN) {
/* If isLongOffset, but the maximum number of additional bits that we see in our table is small
* enough, then we know it is impossible to have too long an offset in this block, so we can
* use the regular offset decoder.
*/
isLongOffset = ZSTD_lo_isRegularOffset;
}
if (!usePrefetchDecoder) {
U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
usePrefetchDecoder = (info.longOffsetShare >= minShare);
}
}
#endif
dctx->ddictIsCold = 0;
#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
!defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
if (usePrefetchDecoder)
if (usePrefetchDecoder) {
#else
(void)usePrefetchDecoder;
{
#endif
#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
#endif
}
#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
/* else */
zstd/decompress/zstd_decompress_block.h
+1 -1
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) Yann Collet, Facebook, Inc.
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
zstd/decompress/zstd_decompress_internal.h
+4 -2
View File
@@ -1,5 +1,5 @@
/*
* Copyright (c) Yann Collet, Facebook, Inc.
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
@@ -75,12 +75,13 @@ static UNUSED_ATTR const U32 ML_base[MaxML+1] = {
#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))
#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
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 */
HUF_DTable hufTable[HUF_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)]; /* can accommodate HUF_decompress4X */
U32 rep[ZSTD_REP_NUM];
U32 workspace[ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32];
} ZSTD_entropyDTables_t;
@@ -164,6 +165,7 @@ struct ZSTD_DCtx_s
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) */
int disableHufAsm;
/* streaming */
ZSTD_dStreamStage streamStage;