zstd/lib/compress/zstd_compress.c

3386 lines
147 KiB
C
Raw Normal View History

/*
2016-08-30 17:04:33 +00:00
* Copyright (c) 2016-present, 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.
2016-08-30 17:04:33 +00:00
*/
2015-10-22 14:31:46 +00:00
2017-05-12 22:31:53 +00:00
/*-*************************************
* Tuning parameters
***************************************/
#ifndef ZSTD_CLEVEL_DEFAULT
# define ZSTD_CLEVEL_DEFAULT 3
2017-05-12 22:31:53 +00:00
#endif
2016-02-11 23:07:30 +00:00
/*-*************************************
2016-02-03 01:46:46 +00:00
* Dependencies
2015-10-22 14:31:46 +00:00
***************************************/
#include <string.h> /* memset */
2015-11-11 20:38:21 +00:00
#include "mem.h"
2016-08-11 23:20:36 +00:00
#define FSE_STATIC_LINKING_ONLY /* FSE_encodeSymbol */
#include "fse.h"
#define HUF_STATIC_LINKING_ONLY
#include "huf.h"
#include "zstd_compress_internal.h"
2017-09-02 01:28:35 +00:00
#include "zstd_fast.h"
#include "zstd_double_fast.h"
#include "zstd_lazy.h"
#include "zstd_opt.h"
2017-09-07 00:56:01 +00:00
#include "zstd_ldm.h"
2015-10-22 14:31:46 +00:00
2016-02-11 23:07:30 +00:00
/*-*************************************
2016-01-23 18:28:41 +00:00
* Helper functions
***************************************/
size_t ZSTD_compressBound(size_t srcSize) {
return ZSTD_COMPRESSBOUND(srcSize);
}
2016-01-23 18:28:41 +00:00
2016-02-11 23:07:30 +00:00
/*-*************************************
2015-11-11 20:38:21 +00:00
* Context memory management
2015-10-22 14:31:46 +00:00
***************************************/
struct ZSTD_CDict_s {
void* dictBuffer;
const void* dictContent;
size_t dictContentSize;
void* workspace;
size_t workspaceSize;
ZSTD_matchState_t matchState;
ZSTD_compressedBlockState_t cBlockState;
ZSTD_compressionParameters cParams;
ZSTD_customMem customMem;
U32 dictID;
}; /* typedef'd to ZSTD_CDict within "zstd.h" */
ZSTD_CCtx* ZSTD_createCCtx(void)
2015-10-22 14:31:46 +00:00
{
return ZSTD_createCCtx_advanced(ZSTD_defaultCMem);
2016-05-23 13:49:09 +00:00
}
ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
{
ZSTD_STATIC_ASSERT(zcss_init==0);
ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN==(0ULL - 1));
if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
{ ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_calloc(sizeof(ZSTD_CCtx), customMem);
if (!cctx) return NULL;
cctx->customMem = customMem;
cctx->requestedParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
cctx->requestedParams.fParams.contentSizeFlag = 1;
return cctx;
}
2015-10-22 14:31:46 +00:00
}
ZSTD_CCtx* ZSTD_initStaticCCtx(void *workspace, size_t workspaceSize)
{
ZSTD_CCtx* const cctx = (ZSTD_CCtx*) workspace;
if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL; /* minimum size */
if ((size_t)workspace & 7) return NULL; /* must be 8-aligned */
memset(workspace, 0, workspaceSize); /* may be a bit generous, could memset be smaller ? */
cctx->staticSize = workspaceSize;
cctx->workSpace = (void*)(cctx+1);
cctx->workSpaceSize = workspaceSize - sizeof(ZSTD_CCtx);
/* statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
2018-01-12 20:06:10 +00:00
if (cctx->workSpaceSize < HUF_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t)) return NULL;
assert(((size_t)cctx->workSpace & (sizeof(void*)-1)) == 0); /* ensure correct alignment */
2018-01-12 20:06:10 +00:00
cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)cctx->workSpace;
cctx->blockState.nextCBlock = cctx->blockState.prevCBlock + 1;
{
2018-01-12 20:06:10 +00:00
void* const ptr = cctx->blockState.nextCBlock + 1;
cctx->entropyWorkspace = (U32*)ptr;
}
2016-07-14 14:52:45 +00:00
return cctx;
2015-10-22 14:31:46 +00:00
}
size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
2015-10-25 13:06:35 +00:00
{
if (cctx==NULL) return 0; /* support free on NULL */
if (cctx->staticSize) return ERROR(memory_allocation); /* not compatible with static CCtx */
ZSTD_free(cctx->workSpace, cctx->customMem); cctx->workSpace = NULL;
ZSTD_freeCDict(cctx->cdictLocal); cctx->cdictLocal = NULL;
#ifdef ZSTD_MULTITHREAD
ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL;
#endif
ZSTD_free(cctx, cctx->customMem);
return 0; /* reserved as a potential error code in the future */
2015-10-25 13:06:35 +00:00
}
2015-10-22 14:31:46 +00:00
static size_t ZSTD_sizeof_mtctx(const ZSTD_CCtx* cctx)
{
#ifdef ZSTD_MULTITHREAD
return ZSTDMT_sizeof_CCtx(cctx->mtctx);
#else
(void) cctx;
return 0;
#endif
}
2016-08-22 23:18:06 +00:00
size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx)
2016-07-11 01:12:17 +00:00
{
if (cctx==NULL) return 0; /* support sizeof on NULL */
return sizeof(*cctx) + cctx->workSpaceSize
+ ZSTD_sizeof_CDict(cctx->cdictLocal)
+ ZSTD_sizeof_mtctx(cctx);
2016-07-11 01:12:17 +00:00
}
size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
{
return ZSTD_sizeof_CCtx(zcs); /* same object */
}
2017-05-12 22:31:53 +00:00
/* private API call, for dictBuilder only */
const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx) { return &(ctx->seqStore); }
2017-08-18 02:30:22 +00:00
#define ZSTD_CLEVEL_CUSTOM 999
2017-05-12 22:31:53 +00:00
2017-08-29 02:25:17 +00:00
static ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
ZSTD_CCtx_params CCtxParams, U64 srcSizeHint, size_t dictSize)
{
DEBUGLOG(4, "ZSTD_getCParamsFromCCtxParams: srcSize = %u, dictSize = %u",
(U32)srcSizeHint, (U32)dictSize);
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
return (CCtxParams.compressionLevel == ZSTD_CLEVEL_CUSTOM) ?
CCtxParams.cParams :
ZSTD_getCParams(CCtxParams.compressionLevel, srcSizeHint, dictSize);
}
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
static void ZSTD_cLevelToCCtxParams_srcSize(ZSTD_CCtx_params* CCtxParams, U64 srcSize)
{
DEBUGLOG(4, "ZSTD_cLevelToCCtxParams_srcSize: srcSize = %u",
(U32)srcSize);
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
CCtxParams->cParams = ZSTD_getCParamsFromCCtxParams(*CCtxParams, srcSize, 0);
CCtxParams->compressionLevel = ZSTD_CLEVEL_CUSTOM;
}
static void ZSTD_cLevelToCParams(ZSTD_CCtx* cctx)
{
DEBUGLOG(4, "ZSTD_cLevelToCParams: level=%i", cctx->requestedParams.compressionLevel);
ZSTD_cLevelToCCtxParams_srcSize(
&cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1);
}
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
static void ZSTD_cLevelToCCtxParams(ZSTD_CCtx_params* CCtxParams)
{
DEBUGLOG(4, "ZSTD_cLevelToCCtxParams");
ZSTD_cLevelToCCtxParams_srcSize(CCtxParams, ZSTD_CONTENTSIZE_UNKNOWN);
2017-08-18 02:30:22 +00:00
}
2017-08-21 17:09:06 +00:00
static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
ZSTD_compressionParameters cParams)
{
ZSTD_CCtx_params cctxParams;
memset(&cctxParams, 0, sizeof(cctxParams));
2017-08-21 17:09:06 +00:00
cctxParams.cParams = cParams;
cctxParams.compressionLevel = ZSTD_CLEVEL_CUSTOM;
return cctxParams;
}
static ZSTD_CCtx_params* ZSTD_createCCtxParams_advanced(
ZSTD_customMem customMem)
{
ZSTD_CCtx_params* params;
if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
params = (ZSTD_CCtx_params*)ZSTD_calloc(
sizeof(ZSTD_CCtx_params), customMem);
if (!params) { return NULL; }
params->customMem = customMem;
2017-08-18 02:30:22 +00:00
params->compressionLevel = ZSTD_CLEVEL_DEFAULT;
return params;
}
ZSTD_CCtx_params* ZSTD_createCCtxParams(void)
{
return ZSTD_createCCtxParams_advanced(ZSTD_defaultCMem);
}
size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params)
{
if (params == NULL) { return 0; }
ZSTD_free(params, params->customMem);
return 0;
}
2017-08-21 05:55:07 +00:00
size_t ZSTD_resetCCtxParams(ZSTD_CCtx_params* params)
{
2017-08-29 02:25:17 +00:00
return ZSTD_initCCtxParams(params, ZSTD_CLEVEL_DEFAULT);
2017-08-21 05:55:07 +00:00
}
size_t ZSTD_initCCtxParams(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
2017-08-29 02:25:17 +00:00
if (!cctxParams) { return ERROR(GENERIC); }
memset(cctxParams, 0, sizeof(*cctxParams));
2017-08-29 02:25:17 +00:00
cctxParams->compressionLevel = compressionLevel;
return 0;
}
size_t ZSTD_initCCtxParams_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
{
if (!cctxParams) { return ERROR(GENERIC); }
CHECK_F( ZSTD_checkCParams(params.cParams) );
memset(cctxParams, 0, sizeof(*cctxParams));
2017-08-21 08:59:08 +00:00
cctxParams->cParams = params.cParams;
cctxParams->fParams = params.fParams;
cctxParams->compressionLevel = ZSTD_CLEVEL_CUSTOM;
return 0;
}
static ZSTD_CCtx_params ZSTD_assignParamsToCCtxParams(
ZSTD_CCtx_params cctxParams, ZSTD_parameters params)
{
ZSTD_CCtx_params ret = cctxParams;
ret.cParams = params.cParams;
ret.fParams = params.fParams;
ret.compressionLevel = ZSTD_CLEVEL_CUSTOM;
return ret;
}
#define CLAMPCHECK(val,min,max) { \
if (((val)<(min)) | ((val)>(max))) { \
return ERROR(parameter_outOfBound); \
} }
2017-05-12 22:31:53 +00:00
size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned value)
{
DEBUGLOG(4, "ZSTD_CCtx_setParameter (%u, %u)", (U32)param, value);
2017-05-22 20:05:45 +00:00
if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
2017-05-12 22:31:53 +00:00
switch(param)
{
case ZSTD_p_format :
return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
2017-08-21 05:55:07 +00:00
case ZSTD_p_compressionLevel:
if (cctx->cdict) return ERROR(stage_wrong);
2017-08-21 08:59:08 +00:00
return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
2017-08-21 05:55:07 +00:00
case ZSTD_p_windowLog:
case ZSTD_p_hashLog:
case ZSTD_p_chainLog:
case ZSTD_p_searchLog:
case ZSTD_p_minMatch:
case ZSTD_p_targetLength:
case ZSTD_p_compressionStrategy:
if (cctx->cdict) return ERROR(stage_wrong);
if (value>0) ZSTD_cLevelToCParams(cctx); /* Can optimize if srcSize is known */
2017-08-21 05:55:07 +00:00
return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
2017-05-12 22:31:53 +00:00
2017-08-21 05:55:07 +00:00
case ZSTD_p_contentSizeFlag:
case ZSTD_p_checksumFlag:
case ZSTD_p_dictIDFlag:
return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
2017-05-12 22:31:53 +00:00
case ZSTD_p_forceMaxWindow : /* Force back-references to remain < windowSize,
* even when referencing into Dictionary content.
2017-05-12 22:31:53 +00:00
* default : 0 when using a CDict, 1 when using a Prefix */
return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
case ZSTD_p_nbThreads:
if ((value > 1) && cctx->staticSize) {
2017-08-29 23:18:21 +00:00
return ERROR(parameter_unsupported); /* MT not compatible with static alloc */
}
2017-08-29 23:18:21 +00:00
return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
case ZSTD_p_nonBlockingMode:
case ZSTD_p_jobSize:
case ZSTD_p_overlapSizeLog:
return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
2017-08-21 05:55:07 +00:00
2017-09-03 04:10:36 +00:00
case ZSTD_p_enableLongDistanceMatching:
if (cctx->cdict) return ERROR(stage_wrong);
if (value>0) ZSTD_cLevelToCParams(cctx);
return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
case ZSTD_p_ldmHashLog:
case ZSTD_p_ldmMinMatch:
2017-09-03 04:10:36 +00:00
case ZSTD_p_ldmBucketSizeLog:
case ZSTD_p_ldmHashEveryLog:
if (cctx->cdict) return ERROR(stage_wrong);
return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
default: return ERROR(parameter_unsupported);
}
}
2017-08-18 02:30:22 +00:00
size_t ZSTD_CCtxParam_setParameter(
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
ZSTD_CCtx_params* CCtxParams, ZSTD_cParameter param, unsigned value)
2017-08-18 02:30:22 +00:00
{
DEBUGLOG(4, "ZSTD_CCtxParam_setParameter (%u, %u)", (U32)param, value);
2017-08-18 02:30:22 +00:00
switch(param)
{
case ZSTD_p_format :
2017-09-27 19:22:22 +00:00
if (value > (unsigned)ZSTD_f_zstd1_magicless)
return ERROR(parameter_unsupported);
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
CCtxParams->format = (ZSTD_format_e)value;
return (size_t)CCtxParams->format;
2017-08-18 02:30:22 +00:00
case ZSTD_p_compressionLevel :
if ((int)value > ZSTD_maxCLevel()) value = ZSTD_maxCLevel();
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) /* 0 : does not change current level */
CCtxParams->compressionLevel = value;
return CCtxParams->compressionLevel;
2017-05-12 22:31:53 +00:00
case ZSTD_p_windowLog :
DEBUGLOG(4, "ZSTD_CCtxParam_setParameter: set windowLog=%u", value);
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) { /* 0 : does not change current windowLog */
CLAMPCHECK(value, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
ZSTD_cLevelToCCtxParams(CCtxParams);
CCtxParams->cParams.windowLog = value;
}
return CCtxParams->cParams.windowLog;
2017-05-12 22:31:53 +00:00
case ZSTD_p_hashLog :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) { /* 0 : does not change current hashLog */
CLAMPCHECK(value, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
ZSTD_cLevelToCCtxParams(CCtxParams);
CCtxParams->cParams.hashLog = value;
}
return CCtxParams->cParams.hashLog;
2017-05-12 22:31:53 +00:00
case ZSTD_p_chainLog :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) { /* 0 : does not change current chainLog */
CLAMPCHECK(value, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
ZSTD_cLevelToCCtxParams(CCtxParams);
CCtxParams->cParams.chainLog = value;
}
return CCtxParams->cParams.chainLog;
2017-05-12 22:31:53 +00:00
case ZSTD_p_searchLog :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) { /* 0 : does not change current searchLog */
CLAMPCHECK(value, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
ZSTD_cLevelToCCtxParams(CCtxParams);
CCtxParams->cParams.searchLog = value;
}
return value;
2017-05-12 22:31:53 +00:00
case ZSTD_p_minMatch :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) { /* 0 : does not change current minMatch length */
CLAMPCHECK(value, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
ZSTD_cLevelToCCtxParams(CCtxParams);
CCtxParams->cParams.searchLength = value;
}
return CCtxParams->cParams.searchLength;
2017-05-12 22:31:53 +00:00
case ZSTD_p_targetLength :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) { /* 0 : does not change current sufficient_len */
CLAMPCHECK(value, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
ZSTD_cLevelToCCtxParams(CCtxParams);
CCtxParams->cParams.targetLength = value;
}
return CCtxParams->cParams.targetLength;
2017-05-12 22:31:53 +00:00
case ZSTD_p_compressionStrategy :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) { /* 0 : does not change currentstrategy */
CLAMPCHECK(value, (unsigned)ZSTD_fast, (unsigned)ZSTD_btultra);
ZSTD_cLevelToCCtxParams(CCtxParams);
CCtxParams->cParams.strategy = (ZSTD_strategy)value;
}
return (size_t)CCtxParams->cParams.strategy;
2017-05-12 22:31:53 +00:00
case ZSTD_p_contentSizeFlag :
/* Content size written in frame header _when known_ (default:1) */
DEBUGLOG(4, "set content size flag = %u", (value>0));
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
CCtxParams->fParams.contentSizeFlag = value > 0;
return CCtxParams->fParams.contentSizeFlag;
2017-05-12 22:31:53 +00:00
case ZSTD_p_checksumFlag :
/* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
CCtxParams->fParams.checksumFlag = value > 0;
return CCtxParams->fParams.checksumFlag;
2017-05-12 22:31:53 +00:00
case ZSTD_p_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
DEBUGLOG(4, "set dictIDFlag = %u", (value>0));
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
CCtxParams->fParams.noDictIDFlag = (value == 0);
return !CCtxParams->fParams.noDictIDFlag;
2017-05-12 22:31:53 +00:00
2017-08-18 02:30:22 +00:00
case ZSTD_p_forceMaxWindow :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
CCtxParams->forceWindow = (value > 0);
return CCtxParams->forceWindow;
2017-08-18 02:30:22 +00:00
case ZSTD_p_nbThreads :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value == 0) return CCtxParams->nbThreads;
#ifndef ZSTD_MULTITHREAD
if (value > 1) return ERROR(parameter_unsupported);
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
return 1;
#else
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
return ZSTDMT_CCtxParam_setNbThreads(CCtxParams, value);
#endif
case ZSTD_p_nonBlockingMode :
#ifndef ZSTD_MULTITHREAD
return ERROR(parameter_unsupported);
#else
CCtxParams->nonBlockingMode = (value>0);
return CCtxParams->nonBlockingMode;
#endif
2017-08-18 02:30:22 +00:00
case ZSTD_p_jobSize :
#ifndef ZSTD_MULTITHREAD
return ERROR(parameter_unsupported);
#else
return ZSTDMT_CCtxParam_setMTCtxParameter(CCtxParams, ZSTDMT_p_jobSize, value);
#endif
2017-08-18 02:30:22 +00:00
case ZSTD_p_overlapSizeLog :
#ifndef ZSTD_MULTITHREAD
return ERROR(parameter_unsupported);
#else
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
return ZSTDMT_CCtxParam_setMTCtxParameter(CCtxParams, ZSTDMT_p_overlapSectionLog, value);
#endif
2017-05-12 22:31:53 +00:00
2017-09-03 04:10:36 +00:00
case ZSTD_p_enableLongDistanceMatching :
if (value) {
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
ZSTD_cLevelToCCtxParams(CCtxParams);
CCtxParams->cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
}
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
return ZSTD_ldm_initializeParameters(&CCtxParams->ldmParams, value);
case ZSTD_p_ldmHashLog :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) { /* 0 : does not change current ldmHashLog */
CLAMPCHECK(value, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
CCtxParams->ldmParams.hashLog = value;
}
return CCtxParams->ldmParams.hashLog;
case ZSTD_p_ldmMinMatch :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (value) { /* 0 : does not change current ldmMinMatch */
CLAMPCHECK(value, ZSTD_LDM_MINMATCH_MIN, ZSTD_LDM_MINMATCH_MAX);
CCtxParams->ldmParams.minMatchLength = value;
}
return CCtxParams->ldmParams.minMatchLength;
2017-09-03 04:10:36 +00:00
case ZSTD_p_ldmBucketSizeLog :
if (value > ZSTD_LDM_BUCKETSIZELOG_MAX) {
return ERROR(parameter_outOfBound);
}
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
CCtxParams->ldmParams.bucketSizeLog = value;
return value;
2017-09-03 04:10:36 +00:00
case ZSTD_p_ldmHashEveryLog :
if (value > ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN) {
return ERROR(parameter_outOfBound);
}
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
CCtxParams->ldmParams.hashEveryLog = value;
return value;
default: return ERROR(parameter_unsupported);
}
}
/** ZSTD_CCtx_setParametersUsingCCtxParams() :
* just applies `params` into `cctx`
* no action is performed, parameters are merely stored.
2017-08-21 08:59:08 +00:00
*/
2017-08-30 01:03:06 +00:00
size_t ZSTD_CCtx_setParametersUsingCCtxParams(
ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params)
2017-08-18 18:20:08 +00:00
{
2017-08-29 23:18:21 +00:00
if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
if (cctx->cdict) return ERROR(stage_wrong);
2017-08-18 20:01:55 +00:00
cctx->requestedParams = *params;
2017-08-18 18:20:08 +00:00
return 0;
}
2017-05-12 22:31:53 +00:00
ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize)
{
DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %u bytes", (U32)pledgedSrcSize);
2017-05-22 20:05:45 +00:00
if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
2017-05-12 22:31:53 +00:00
return 0;
}
size_t ZSTD_CCtx_loadDictionary_advanced(
ZSTD_CCtx* cctx, const void* dict, size_t dictSize,
ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictMode_e dictMode)
{
2017-05-22 20:05:45 +00:00
if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
if (cctx->staticSize) return ERROR(memory_allocation); /* no malloc for static CCtx */
DEBUGLOG(4, "ZSTD_CCtx_loadDictionary_advanced (size: %u)", (U32)dictSize);
ZSTD_freeCDict(cctx->cdictLocal); /* in case one already exists */
if (dict==NULL || dictSize==0) { /* no dictionary mode */
cctx->cdictLocal = NULL;
cctx->cdict = NULL;
} else {
ZSTD_compressionParameters const cParams =
ZSTD_getCParamsFromCCtxParams(cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1, dictSize);
cctx->cdictLocal = ZSTD_createCDict_advanced(
dict, dictSize,
dictLoadMethod, dictMode,
cParams, cctx->customMem);
cctx->cdict = cctx->cdictLocal;
if (cctx->cdictLocal == NULL)
return ERROR(memory_allocation);
}
return 0;
}
ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_byReference(
ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
{
return ZSTD_CCtx_loadDictionary_advanced(
cctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dm_auto);
}
ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
{
return ZSTD_CCtx_loadDictionary_advanced(
cctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dm_auto);
}
2017-06-27 22:49:12 +00:00
size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
2016-02-11 23:07:30 +00:00
{
2017-05-22 20:05:45 +00:00
if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
2017-06-27 22:49:12 +00:00
cctx->cdict = cdict;
memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict)); /* exclusive */
2017-06-27 22:49:12 +00:00
return 0;
2016-02-11 23:07:30 +00:00
}
2017-06-27 22:49:12 +00:00
size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize)
{
return ZSTD_CCtx_refPrefix_advanced(cctx, prefix, prefixSize, ZSTD_dm_rawContent);
}
size_t ZSTD_CCtx_refPrefix_advanced(
ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictMode_e dictMode)
2016-02-11 23:07:30 +00:00
{
2017-05-22 20:05:45 +00:00
if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
2017-06-27 22:49:12 +00:00
cctx->cdict = NULL; /* prefix discards any prior cdict */
cctx->prefixDict.dict = prefix;
cctx->prefixDict.dictSize = prefixSize;
cctx->prefixDict.dictMode = dictMode;
2017-06-27 22:49:12 +00:00
return 0;
2016-02-11 23:07:30 +00:00
}
2017-06-16 21:00:46 +00:00
static void ZSTD_startNewCompression(ZSTD_CCtx* cctx)
{
2017-06-16 17:17:50 +00:00
cctx->streamStage = zcss_init;
cctx->pledgedSrcSizePlusOne = 0;
}
2017-06-16 21:00:46 +00:00
/*! ZSTD_CCtx_reset() :
* Also dumps dictionary */
void ZSTD_CCtx_reset(ZSTD_CCtx* cctx)
{
ZSTD_startNewCompression(cctx);
2017-06-16 17:17:50 +00:00
cctx->cdict = NULL;
}
2015-11-04 11:05:27 +00:00
/** ZSTD_checkCParams() :
control CParam values remain within authorized range.
@return : 0, or an error code if one value is beyond authorized range */
size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
{
CLAMPCHECK(cParams.windowLog, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
2016-04-04 11:49:18 +00:00
CLAMPCHECK(cParams.chainLog, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
CLAMPCHECK(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
CLAMPCHECK(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
CLAMPCHECK(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
CLAMPCHECK(cParams.targetLength, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
if ((U32)(cParams.strategy) > (U32)ZSTD_btultra)
return ERROR(parameter_unsupported);
return 0;
}
/** ZSTD_clampCParams() :
* make CParam values within valid range.
* @return : valid CParams */
static ZSTD_compressionParameters ZSTD_clampCParams(ZSTD_compressionParameters cParams)
{
# define CLAMP(val,min,max) { \
if (val<min) val=min; \
else if (val>max) val=max; \
}
CLAMP(cParams.windowLog, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
CLAMP(cParams.chainLog, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
CLAMP(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
CLAMP(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
CLAMP(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
CLAMP(cParams.targetLength, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
if ((U32)(cParams.strategy) > (U32)ZSTD_btultra) cParams.strategy = ZSTD_btultra;
return cParams;
}
2016-02-10 12:37:52 +00:00
2016-12-11 23:47:30 +00:00
/** ZSTD_cycleLog() :
* condition for correct operation : hashLog > 1 */
static U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
{
U32 const btScale = ((U32)strat >= (U32)ZSTD_btlazy2);
return hashLog - btScale;
}
/** ZSTD_adjustCParams_internal() :
optimize `cPar` for a given input (`srcSize` and `dictSize`).
mostly downsizing to reduce memory consumption and initialization latency.
Both `srcSize` and `dictSize` are optional (use 0 if unknown).
Note : cPar is considered validated at this stage. Use ZSTD_checkCParams() to ensure that condition. */
ZSTD_compressionParameters ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
{
static const U64 minSrcSize = 513; /* (1<<9) + 1 */
static const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
assert(ZSTD_checkCParams(cPar)==0);
2015-11-04 11:05:27 +00:00
if (dictSize && (srcSize+1<2) /* srcSize unknown */ )
srcSize = minSrcSize; /* presumed small when there is a dictionary */
else if (srcSize == 0)
srcSize = ZSTD_CONTENTSIZE_UNKNOWN; /* 0 == unknown : presumed large */
/* resize windowLog if input is small enough, to use less memory */
if ( (srcSize < maxWindowResize)
&& (dictSize < maxWindowResize) ) {
U32 const tSize = (U32)(srcSize + dictSize);
static U32 const hashSizeMin = 1 << ZSTD_HASHLOG_MIN;
U32 const srcLog = (tSize < hashSizeMin) ? ZSTD_HASHLOG_MIN :
ZSTD_highbit32(tSize-1) + 1;
if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
}
if (cPar.hashLog > cPar.windowLog) cPar.hashLog = cPar.windowLog;
2016-12-11 23:47:30 +00:00
{ U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
if (cycleLog > cPar.windowLog)
cPar.chainLog -= (cycleLog - cPar.windowLog);
2016-12-11 23:47:30 +00:00
}
2016-03-19 16:18:00 +00:00
if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN)
cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN; /* required for frame header */
return cPar;
2015-11-04 11:05:27 +00:00
}
ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
{
cPar = ZSTD_clampCParams(cPar);
return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize);
}
static size_t ZSTD_sizeof_matchState(ZSTD_compressionParameters const* cParams, const U32 forCCtx)
{
size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
size_t const hSize = ((size_t)1) << cParams->hashLog;
U32 const hashLog3 = (forCCtx && cParams->searchLength==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
size_t const h3Size = ((size_t)1) << hashLog3;
size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
size_t const optPotentialSpace = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits)) * sizeof(U32)
+ (ZSTD_OPT_NUM+1) * (sizeof(ZSTD_match_t)+sizeof(ZSTD_optimal_t));
size_t const optSpace = (forCCtx && ((cParams->strategy == ZSTD_btopt) ||
(cParams->strategy == ZSTD_btultra)))
? optPotentialSpace
: 0;
DEBUGLOG(4, "chainSize: %u - hSize: %u - h3Size: %u",
(U32)chainSize, (U32)hSize, (U32)h3Size);
return tableSpace + optSpace;
}
size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
2016-03-19 15:09:09 +00:00
{
/* Estimate CCtx size is supported for single-threaded compression only. */
if (params->nbThreads > 1) { return ERROR(GENERIC); }
2017-08-29 02:25:17 +00:00
{ ZSTD_compressionParameters const cParams =
ZSTD_getCParamsFromCCtxParams(*params, 0, 0);
size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
U32 const divider = (cParams.searchLength==3) ? 3 : 4;
size_t const maxNbSeq = blockSize / divider;
size_t const tokenSpace = blockSize + 11*maxNbSeq;
size_t const entropySpace = HUF_WORKSPACE_SIZE;
2018-01-12 20:06:10 +00:00
size_t const blockStateSpace = 2 * sizeof(ZSTD_compressedBlockState_t);
size_t const matchStateSize = ZSTD_sizeof_matchState(&params->cParams, /* forCCtx */ 1);
size_t const ldmSpace = params->ldmParams.enableLdm ?
ZSTD_ldm_getTableSize(params->ldmParams.hashLog,
2017-09-03 04:10:36 +00:00
params->ldmParams.bucketSizeLog) : 0;
size_t const neededSpace = entropySpace + blockStateSpace + tokenSpace +
matchStateSize + ldmSpace;
2016-07-11 01:12:17 +00:00
DEBUGLOG(5, "sizeof(ZSTD_CCtx) : %u", (U32)sizeof(ZSTD_CCtx));
DEBUGLOG(5, "estimate workSpace : %u", (U32)neededSpace);
return sizeof(ZSTD_CCtx) + neededSpace;
}
}
2016-07-11 01:12:17 +00:00
size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
{
ZSTD_CCtx_params const params = ZSTD_makeCCtxParamsFromCParams(cParams);
return ZSTD_estimateCCtxSize_usingCCtxParams(&params);
}
static size_t ZSTD_estimateCCtxSize_internal(int compressionLevel)
{
ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, 0);
return ZSTD_estimateCCtxSize_usingCParams(cParams);
}
size_t ZSTD_estimateCCtxSize(int compressionLevel)
{
int level;
size_t memBudget = 0;
for (level=1; level<=compressionLevel; level++) {
size_t const newMB = ZSTD_estimateCCtxSize_internal(level);
if (newMB > memBudget) memBudget = newMB;
}
return memBudget;
}
size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
{
if (params->nbThreads > 1) { return ERROR(GENERIC); }
{ size_t const CCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(params);
size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << params->cParams.windowLog);
size_t const inBuffSize = ((size_t)1 << params->cParams.windowLog) + blockSize;
size_t const outBuffSize = ZSTD_compressBound(blockSize) + 1;
size_t const streamingSize = inBuffSize + outBuffSize;
return CCtxSize + streamingSize;
}
}
size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
{
ZSTD_CCtx_params const params = ZSTD_makeCCtxParamsFromCParams(cParams);
return ZSTD_estimateCStreamSize_usingCCtxParams(&params);
}
static size_t ZSTD_estimateCStreamSize_internal(int compressionLevel) {
2017-06-26 23:02:25 +00:00
ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, 0);
return ZSTD_estimateCStreamSize_usingCParams(cParams);
2017-06-26 23:02:25 +00:00
}
size_t ZSTD_estimateCStreamSize(int compressionLevel) {
int level;
size_t memBudget = 0;
for (level=1; level<=compressionLevel; level++) {
size_t const newMB = ZSTD_estimateCStreamSize_internal(level);
if (newMB > memBudget) memBudget = newMB;
}
return memBudget;
}
/* ZSTD_getFrameProgression():
* tells how much data has been consumed (input) and produced (output) for current frame.
* able to count progression inside worker threads (non-blocking mode).
*/
ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx)
{
#ifdef ZSTD_MULTITHREAD
if ((cctx->appliedParams.nbThreads > 1) || (cctx->appliedParams.nonBlockingMode)) {
return ZSTDMT_getFrameProgression(cctx->mtctx);
}
#endif
{ ZSTD_frameProgression fp;
size_t const buffered = (cctx->inBuff == NULL) ? 0 :
cctx->inBuffPos - cctx->inToCompress;
if (buffered) assert(cctx->inBuffPos >= cctx->inToCompress);
assert(buffered <= ZSTD_BLOCKSIZE_MAX);
fp.ingested = cctx->consumedSrcSize + buffered;
fp.consumed = cctx->consumedSrcSize;
fp.produced = cctx->producedCSize;
return fp;
} }
static U32 ZSTD_equivalentCParams(ZSTD_compressionParameters cParams1,
ZSTD_compressionParameters cParams2)
{
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
return (cParams1.hashLog == cParams2.hashLog)
& (cParams1.chainLog == cParams2.chainLog)
& (cParams1.strategy == cParams2.strategy) /* opt parser space */
& ((cParams1.searchLength==3) == (cParams2.searchLength==3)); /* hashlog3 space */
}
/** The parameters are equivalent if ldm is not enabled in both sets or
* all the parameters are equivalent. */
static U32 ZSTD_equivalentLdmParams(ldmParams_t ldmParams1,
ldmParams_t ldmParams2)
{
return (!ldmParams1.enableLdm && !ldmParams2.enableLdm) ||
(ldmParams1.enableLdm == ldmParams2.enableLdm &&
ldmParams1.hashLog == ldmParams2.hashLog &&
2017-09-03 04:10:36 +00:00
ldmParams1.bucketSizeLog == ldmParams2.bucketSizeLog &&
ldmParams1.minMatchLength == ldmParams2.minMatchLength &&
ldmParams1.hashEveryLog == ldmParams2.hashEveryLog);
}
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
typedef enum { ZSTDb_not_buffered, ZSTDb_buffered } ZSTD_buffered_policy_e;
/* ZSTD_sufficientBuff() :
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
* check internal buffers exist for streaming if buffPol == ZSTDb_buffered .
* Note : they are assumed to be correctly sized if ZSTD_equivalentCParams()==1 */
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
static U32 ZSTD_sufficientBuff(size_t bufferSize1, size_t blockSize1,
ZSTD_buffered_policy_e buffPol2,
ZSTD_compressionParameters cParams2,
U64 pledgedSrcSize)
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
{
size_t const windowSize2 = MAX(1, (size_t)MIN(((U64)1 << cParams2.windowLog), pledgedSrcSize));
size_t const blockSize2 = MIN(ZSTD_BLOCKSIZE_MAX, windowSize2);
size_t const neededBufferSize2 = (buffPol2==ZSTDb_buffered) ? windowSize2 + blockSize2 : 0;
DEBUGLOG(4, "ZSTD_sufficientBuff: is windowSize2=%u <= wlog1=%u",
(U32)windowSize2, cParams2.windowLog);
DEBUGLOG(4, "ZSTD_sufficientBuff: is blockSize2=%u <= blockSize1=%u",
(U32)blockSize2, (U32)blockSize1);
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
return (blockSize2 <= blockSize1) /* seqStore space depends on blockSize */
& (neededBufferSize2 <= bufferSize1);
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
}
/** Equivalence for resetCCtx purposes */
static U32 ZSTD_equivalentParams(ZSTD_CCtx_params params1,
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
ZSTD_CCtx_params params2,
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
size_t buffSize1, size_t blockSize1,
ZSTD_buffered_policy_e buffPol2,
U64 pledgedSrcSize)
{
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
DEBUGLOG(4, "ZSTD_equivalentParams: pledgedSrcSize=%u", (U32)pledgedSrcSize);
return ZSTD_equivalentCParams(params1.cParams, params2.cParams) &&
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
ZSTD_equivalentLdmParams(params1.ldmParams, params2.ldmParams) &&
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
ZSTD_sufficientBuff(buffSize1, blockSize1, buffPol2, params2.cParams, pledgedSrcSize);
}
static void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
{
int i;
for (i = 0; i < ZSTD_REP_NUM; ++i)
bs->rep[i] = repStartValue[i];
bs->entropy.hufCTable_repeatMode = HUF_repeat_none;
bs->entropy.offcode_repeatMode = FSE_repeat_none;
bs->entropy.matchlength_repeatMode = FSE_repeat_none;
bs->entropy.litlength_repeatMode = FSE_repeat_none;
}
/*! ZSTD_invalidateMatchState()
* Invalidate all the matches in the match finder tables.
* Requires nextSrc and base to be set (can be NULL).
*/
static void ZSTD_invalidateMatchState(ZSTD_matchState_t* ms)
{
size_t const endT = (size_t)(ms->nextSrc - ms->base);
U32 const end = (U32)endT;
assert(endT < (3U<<30));
ms->lowLimit = end;
ms->dictLimit = end;
ms->nextToUpdate = end + 1;
ms->loadedDictEnd = 0;
ms->opt.litLengthSum = 0; /* force reset of btopt stats */
}
/*! ZSTD_continueCCtx() :
* reuse CCtx without reset (note : requires no dictionary) */
2017-08-18 18:20:08 +00:00
static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_CCtx_params params, U64 pledgedSrcSize)
{
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params.cParams.windowLog), pledgedSrcSize));
size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
DEBUGLOG(4, "ZSTD_continueCCtx: re-use context in place");
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
cctx->blockSize = blockSize; /* previous block size could be different even for same windowLog, due to pledgedSrcSize */
cctx->appliedParams = params;
2017-06-16 21:00:46 +00:00
cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
cctx->consumedSrcSize = 0;
cctx->producedCSize = 0;
2017-06-16 21:00:46 +00:00
if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
cctx->appliedParams.fParams.contentSizeFlag = 0;
DEBUGLOG(4, "pledged content size : %u ; flag : %u",
2017-06-16 21:00:46 +00:00
(U32)pledgedSrcSize, cctx->appliedParams.fParams.contentSizeFlag);
cctx->stage = ZSTDcs_init;
cctx->dictID = 0;
2018-01-12 20:06:10 +00:00
ZSTD_invalidateMatchState(&cctx->blockState.matchState);
ZSTD_reset_compressedBlockState(cctx->blockState.prevCBlock);
2016-09-06 07:54:22 +00:00
XXH64_reset(&cctx->xxhState, 0);
return 0;
}
typedef enum { ZSTDcrp_continue, ZSTDcrp_noMemset } ZSTD_compResetPolicy_e;
static void* ZSTD_reset_matchState(ZSTD_matchState_t* ms, void* ptr, ZSTD_compressionParameters const* cParams, ZSTD_compResetPolicy_e const crp, U32 const forCCtx)
{
size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
size_t const hSize = ((size_t)1) << cParams->hashLog;
U32 const hashLog3 = (forCCtx && cParams->searchLength==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
size_t const h3Size = ((size_t)1) << hashLog3;
size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
assert(((size_t)ptr & 3) == 0);
ms->nextSrc = NULL;
ms->base = NULL;
ms->dictBase = NULL;
ms->hashLog3 = hashLog3;
ZSTD_invalidateMatchState(ms);
/* opt parser space */
if (forCCtx && ((cParams->strategy == ZSTD_btopt) | (cParams->strategy == ZSTD_btultra))) {
DEBUGLOG(4, "reserving optimal parser space");
ms->opt.litFreq = (U32*)ptr;
ms->opt.litLengthFreq = ms->opt.litFreq + (1<<Litbits);
ms->opt.matchLengthFreq = ms->opt.litLengthFreq + (MaxLL+1);
ms->opt.offCodeFreq = ms->opt.matchLengthFreq + (MaxML+1);
ptr = ms->opt.offCodeFreq + (MaxOff+1);
ms->opt.matchTable = (ZSTD_match_t*)ptr;
ptr = ms->opt.matchTable + ZSTD_OPT_NUM+1;
ms->opt.priceTable = (ZSTD_optimal_t*)ptr;
ptr = ms->opt.priceTable + ZSTD_OPT_NUM+1;
}
/* table Space */
DEBUGLOG(4, "reset table : %u", crp!=ZSTDcrp_noMemset);
assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */
if (crp!=ZSTDcrp_noMemset) memset(ptr, 0, tableSpace); /* reset tables only */
ms->hashTable = (U32*)(ptr);
ms->chainTable = ms->hashTable + hSize;
ms->hashTable3 = ms->chainTable + chainSize;
ptr = ms->hashTable3 + h3Size;
assert(((size_t)ptr & 3) == 0);
return ptr;
}
/*! ZSTD_resetCCtx_internal() :
note : `params` are assumed fully validated at this stage */
static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
2017-08-18 18:20:08 +00:00
ZSTD_CCtx_params params, U64 pledgedSrcSize,
ZSTD_compResetPolicy_e const crp,
ZSTD_buffered_policy_e const zbuff)
{
DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u",
(U32)pledgedSrcSize, params.cParams.windowLog);
assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
if (crp == ZSTDcrp_continue) {
if (ZSTD_equivalentParams(zc->appliedParams, params,
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
zc->inBuffSize, zc->blockSize,
zbuff, pledgedSrcSize)) {
DEBUGLOG(4, "ZSTD_equivalentParams()==1 -> continue mode (wLog1=%u, blockSize1=%u)",
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
zc->appliedParams.cParams.windowLog, (U32)zc->blockSize);
assert(!(params.ldmParams.enableLdm &&
2017-09-07 00:56:01 +00:00
params.ldmParams.hashEveryLog == ZSTD_LDM_HASHEVERYLOG_NOTSET));
return ZSTD_continueCCtx(zc, params, pledgedSrcSize);
} }
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "ZSTD_equivalentParams()==0 -> reset CCtx");
if (params.ldmParams.enableLdm) {
2017-09-03 04:10:36 +00:00
/* Adjust long distance matching parameters */
2017-09-07 00:56:01 +00:00
ZSTD_ldm_adjustParameters(&params.ldmParams, params.cParams.windowLog);
assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
assert(params.ldmParams.hashEveryLog < 32);
zc->ldmState.hashPower =
ZSTD_ldm_getHashPower(params.ldmParams.minMatchLength);
}
{ size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params.cParams.windowLog), pledgedSrcSize));
size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
U32 const divider = (params.cParams.searchLength==3) ? 3 : 4;
size_t const maxNbSeq = blockSize / divider;
size_t const tokenSpace = blockSize + 11*maxNbSeq;
size_t const buffOutSize = (zbuff==ZSTDb_buffered) ? ZSTD_compressBound(blockSize)+1 : 0;
size_t const buffInSize = (zbuff==ZSTDb_buffered) ? windowSize + blockSize : 0;
size_t const matchStateSize = ZSTD_sizeof_matchState(&params.cParams, /* forCCtx */ 1);
void* ptr;
/* Check if workSpace is large enough, alloc a new one if needed */
{ size_t const entropySpace = HUF_WORKSPACE_SIZE;
2018-01-12 20:06:10 +00:00
size_t const blockStateSpace = 2 * sizeof(ZSTD_compressedBlockState_t);
size_t const bufferSpace = buffInSize + buffOutSize;
2017-09-06 15:39:46 +00:00
size_t const ldmSpace = params.ldmParams.enableLdm
? ZSTD_ldm_getTableSize(params.ldmParams.hashLog, params.ldmParams.bucketSizeLog)
: 0;
size_t const neededSpace = entropySpace + blockStateSpace + ldmSpace +
matchStateSize + tokenSpace + bufferSpace;
DEBUGLOG(4, "Need %uKB workspace, including %uKB for match state, and %uKB for buffers",
(U32)(neededSpace>>10), (U32)(matchStateSize>>10), (U32)(bufferSpace>>10));
DEBUGLOG(4, "windowSize: %u - blockSize: %u", (U32)windowSize, (U32)blockSize);
if (zc->workSpaceSize < neededSpace) { /* too small : resize */
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "Need to update workSpaceSize from %uK to %uK",
(unsigned)(zc->workSpaceSize>>10),
(unsigned)(neededSpace>>10));
/* static cctx : no resize, error out */
if (zc->staticSize) return ERROR(memory_allocation);
zc->workSpaceSize = 0;
ZSTD_free(zc->workSpace, zc->customMem);
zc->workSpace = ZSTD_malloc(neededSpace, zc->customMem);
if (zc->workSpace == NULL) return ERROR(memory_allocation);
zc->workSpaceSize = neededSpace;
ptr = zc->workSpace;
/* Statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
assert(((size_t)zc->workSpace & 3) == 0); /* ensure correct alignment */
2018-01-12 20:06:10 +00:00
assert(zc->workSpaceSize >= 2 * sizeof(ZSTD_compressedBlockState_t));
zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)zc->workSpace;
zc->blockState.nextCBlock = zc->blockState.prevCBlock + 1;
ptr = zc->blockState.nextCBlock + 1;
zc->entropyWorkspace = (U32*)ptr;
} }
2016-03-19 15:09:09 +00:00
/* init params */
zc->appliedParams = params;
zc->pledgedSrcSizePlusOne = pledgedSrcSize+1;
zc->consumedSrcSize = 0;
zc->producedCSize = 0;
if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
zc->appliedParams.fParams.contentSizeFlag = 0;
DEBUGLOG(4, "pledged content size : %u ; flag : %u",
2017-06-16 21:00:46 +00:00
(U32)pledgedSrcSize, zc->appliedParams.fParams.contentSizeFlag);
zc->blockSize = blockSize;
XXH64_reset(&zc->xxhState, 0);
zc->stage = ZSTDcs_init;
zc->dictID = 0;
ZSTD_reset_compressedBlockState(zc->blockState.prevCBlock);
ptr = zc->entropyWorkspace + HUF_WORKSPACE_SIZE_U32;
2017-09-06 15:39:46 +00:00
/* ldm hash table */
/* initialize bucketOffsets table later for pointer alignment */
if (params.ldmParams.enableLdm) {
size_t const ldmHSize = ((size_t)1) << params.ldmParams.hashLog;
2017-09-06 15:39:46 +00:00
memset(ptr, 0, ldmHSize * sizeof(ldmEntry_t));
assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */
zc->ldmState.hashTable = (ldmEntry_t*)ptr;
ptr = zc->ldmState.hashTable + ldmHSize;
}
ptr = ZSTD_reset_matchState(&zc->blockState.matchState, ptr, &params.cParams, crp, /* forCCtx */ 1);
/* sequences storage */
zc->seqStore.sequencesStart = (seqDef*)ptr;
ptr = zc->seqStore.sequencesStart + maxNbSeq;
zc->seqStore.llCode = (BYTE*) ptr;
zc->seqStore.mlCode = zc->seqStore.llCode + maxNbSeq;
zc->seqStore.ofCode = zc->seqStore.mlCode + maxNbSeq;
zc->seqStore.litStart = zc->seqStore.ofCode + maxNbSeq;
ptr = zc->seqStore.litStart + blockSize;
2017-09-06 15:39:46 +00:00
/* ldm bucketOffsets table */
if (params.ldmParams.enableLdm) {
size_t const ldmBucketSize =
((size_t)1) << (params.ldmParams.hashLog -
params.ldmParams.bucketSizeLog);
memset(ptr, 0, ldmBucketSize);
zc->ldmState.bucketOffsets = (BYTE*)ptr;
ptr = zc->ldmState.bucketOffsets + ldmBucketSize;
}
/* buffers */
zc->inBuffSize = buffInSize;
zc->inBuff = (char*)ptr;
zc->outBuffSize = buffOutSize;
zc->outBuff = zc->inBuff + buffInSize;
return 0;
}
2015-10-22 14:31:46 +00:00
}
/* ZSTD_invalidateRepCodes() :
* ensures next compression will not use repcodes from previous block.
* Note : only works with regular variant;
* do not use with extDict variant ! */
void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx) {
int i;
2018-01-12 20:06:10 +00:00
for (i=0; i<ZSTD_REP_NUM; i++) cctx->blockState.prevCBlock->rep[i] = 0;
assert(/* !extDict */ cctx->blockState.matchState.lowLimit == cctx->blockState.matchState.dictLimit);
}
2015-10-25 13:06:35 +00:00
static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
const ZSTD_CDict* cdict,
unsigned windowLog,
ZSTD_frameParameters fParams,
U64 pledgedSrcSize,
ZSTD_buffered_policy_e zbuff)
{
{ ZSTD_CCtx_params params = cctx->requestedParams;
/* Copy only compression parameters related to tables. */
params.cParams = cdict->cParams;
if (windowLog) params.cParams.windowLog = windowLog;
params.fParams = fParams;
ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
ZSTDcrp_noMemset, zbuff);
assert(cctx->appliedParams.cParams.strategy == cdict->cParams.strategy);
assert(cctx->appliedParams.cParams.hashLog == cdict->cParams.hashLog);
assert(cctx->appliedParams.cParams.chainLog == cdict->cParams.chainLog);
}
/* copy tables */
{ size_t const chainSize = (cdict->cParams.strategy == ZSTD_fast) ? 0 : ((size_t)1 << cdict->cParams.chainLog);
size_t const hSize = (size_t)1 << cdict->cParams.hashLog;
size_t const tableSpace = (chainSize + hSize) * sizeof(U32);
assert((U32*)cctx->blockState.matchState.chainTable == (U32*)cctx->blockState.matchState.hashTable + hSize); /* chainTable must follow hashTable */
assert((U32*)cctx->blockState.matchState.hashTable3 == (U32*)cctx->blockState.matchState.chainTable + chainSize);
assert((U32*)cdict->matchState.chainTable == (U32*)cdict->matchState.hashTable + hSize); /* chainTable must follow hashTable */
assert((U32*)cdict->matchState.hashTable3 == (U32*)cdict->matchState.chainTable + chainSize);
memcpy(cctx->blockState.matchState.hashTable, cdict->matchState.hashTable, tableSpace); /* presumes all tables follow each other */
}
/* Zero the hashTable3, since the cdict never fills it */
{ size_t const h3Size = (size_t)1 << cctx->blockState.matchState.hashLog3;
assert(cdict->matchState.hashLog3 == 0);
memset(cctx->blockState.matchState.hashTable3, 0, h3Size * sizeof(U32));
}
/* copy dictionary offsets */
{
ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
ZSTD_matchState_t* dstMatchState = &cctx->blockState.matchState;
dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
dstMatchState->nextToUpdate3= srcMatchState->nextToUpdate3;
dstMatchState->nextSrc = srcMatchState->nextSrc;
dstMatchState->base = srcMatchState->base;
dstMatchState->dictBase = srcMatchState->dictBase;
dstMatchState->dictLimit = srcMatchState->dictLimit;
dstMatchState->lowLimit = srcMatchState->lowLimit;
dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
}
cctx->dictID = cdict->dictID;
/* copy block state */
memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
return 0;
}
/*! ZSTD_copyCCtx_internal() :
* Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
* Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
* The "context", in this case, refers to the hash and chain tables,
* entropy tables, and dictionary references.
* `windowLog` value is enforced if != 0, otherwise value is copied from srcCCtx.
* @return : 0, or an error code */
static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
const ZSTD_CCtx* srcCCtx,
ZSTD_frameParameters fParams,
U64 pledgedSrcSize,
ZSTD_buffered_policy_e zbuff)
2016-01-26 14:58:49 +00:00
{
DEBUGLOG(5, "ZSTD_copyCCtx_internal");
2016-07-27 19:05:12 +00:00
if (srcCCtx->stage!=ZSTDcs_init) return ERROR(stage_wrong);
2016-01-26 14:58:49 +00:00
memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
{ ZSTD_CCtx_params params = dstCCtx->requestedParams;
/* Copy only compression parameters related to tables. */
params.cParams = srcCCtx->appliedParams.cParams;
params.fParams = fParams;
ZSTD_resetCCtx_internal(dstCCtx, params, pledgedSrcSize,
ZSTDcrp_noMemset, zbuff);
assert(dstCCtx->appliedParams.cParams.windowLog == srcCCtx->appliedParams.cParams.windowLog);
assert(dstCCtx->appliedParams.cParams.strategy == srcCCtx->appliedParams.cParams.strategy);
assert(dstCCtx->appliedParams.cParams.hashLog == srcCCtx->appliedParams.cParams.hashLog);
assert(dstCCtx->appliedParams.cParams.chainLog == srcCCtx->appliedParams.cParams.chainLog);
assert(dstCCtx->blockState.matchState.hashLog3 == srcCCtx->blockState.matchState.hashLog3);
}
2016-01-26 14:58:49 +00:00
/* copy tables */
2017-09-21 23:18:34 +00:00
{ size_t const chainSize = (srcCCtx->appliedParams.cParams.strategy == ZSTD_fast) ? 0 : ((size_t)1 << srcCCtx->appliedParams.cParams.chainLog);
size_t const hSize = (size_t)1 << srcCCtx->appliedParams.cParams.hashLog;
2018-01-12 20:06:10 +00:00
size_t const h3Size = (size_t)1 << srcCCtx->blockState.matchState.hashLog3;
2016-07-27 19:05:12 +00:00
size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
2018-01-12 20:06:10 +00:00
assert((U32*)dstCCtx->blockState.matchState.chainTable == (U32*)dstCCtx->blockState.matchState.hashTable + hSize); /* chainTable must follow hashTable */
assert((U32*)dstCCtx->blockState.matchState.hashTable3 == (U32*)dstCCtx->blockState.matchState.chainTable + chainSize);
memcpy(dstCCtx->blockState.matchState.hashTable, srcCCtx->blockState.matchState.hashTable, tableSpace); /* presumes all tables follow each other */
2016-03-19 16:18:00 +00:00
}
2016-01-26 14:58:49 +00:00
2016-05-29 03:01:04 +00:00
/* copy dictionary offsets */
2017-12-13 01:37:06 +00:00
{
2018-01-12 20:06:10 +00:00
ZSTD_matchState_t const* srcMatchState = &srcCCtx->blockState.matchState;
ZSTD_matchState_t* dstMatchState = &dstCCtx->blockState.matchState;
dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
dstMatchState->nextToUpdate3= srcMatchState->nextToUpdate3;
dstMatchState->nextSrc = srcMatchState->nextSrc;
dstMatchState->base = srcMatchState->base;
dstMatchState->dictBase = srcMatchState->dictBase;
dstMatchState->dictLimit = srcMatchState->dictLimit;
dstMatchState->lowLimit = srcMatchState->lowLimit;
dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
2017-12-13 01:37:06 +00:00
}
dstCCtx->dictID = srcCCtx->dictID;
/* copy block state */
2018-01-12 20:06:10 +00:00
memcpy(dstCCtx->blockState.prevCBlock, srcCCtx->blockState.prevCBlock, sizeof(*srcCCtx->blockState.prevCBlock));
2016-01-26 14:58:49 +00:00
return 0;
}
/*! ZSTD_copyCCtx() :
* Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
* Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
* pledgedSrcSize==0 means "unknown".
* @return : 0, or an error code */
size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx, unsigned long long pledgedSrcSize)
{
ZSTD_frameParameters fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
ZSTD_buffered_policy_e const zbuff = (ZSTD_buffered_policy_e)(srcCCtx->inBuffSize>0);
ZSTD_STATIC_ASSERT((U32)ZSTDb_buffered==1);
if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
fParams.contentSizeFlag = (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN);
return ZSTD_copyCCtx_internal(dstCCtx, srcCCtx,
fParams, pledgedSrcSize,
zbuff);
}
2016-01-26 14:58:49 +00:00
#define ZSTD_ROWSIZE 16
/*! ZSTD_reduceTable_internal() :
* reduce table indexes by `reducerValue`
* presume table size is a multiple of ZSTD_ROWSIZE.
* Helps auto-vectorization */
static void ZSTD_reduceTable_internal (U32* const table, int const nbRows, U32 const reducerValue)
{
int cellNb = 0;
int rowNb;
for (rowNb=0 ; rowNb < nbRows ; rowNb++) {
int column;
for (column=0; column<ZSTD_ROWSIZE; column++) {
if (table[cellNb] < reducerValue) table[cellNb] = 0;
else table[cellNb] -= reducerValue;
cellNb++;
} }
}
2016-03-20 15:20:06 +00:00
/*! ZSTD_reduceTable() :
* reduce table indexes by `reducerValue` */
2016-03-20 15:20:06 +00:00
static void ZSTD_reduceTable (U32* const table, U32 const size, U32 const reducerValue)
{
assert((size & (ZSTD_ROWSIZE-1)) == 0); /* multiple of ZSTD_ROWSIZE */
assert(size < (1U<<31)); /* can be casted to int */
ZSTD_reduceTable_internal(table, size/ZSTD_ROWSIZE, reducerValue);
}
/*! ZSTD_ldm_reduceTable() :
* reduce table indexes by `reducerValue` */
static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
U32 const reducerValue)
{
U32 u;
for (u = 0; u < size; u++) {
if (table[u].offset < reducerValue) table[u].offset = 0;
else table[u].offset -= reducerValue;
}
}
2016-03-20 15:20:06 +00:00
/*! ZSTD_reduceIndex() :
* rescale all indexes to avoid future overflow (indexes are U32) */
static void ZSTD_reduceIndex (ZSTD_CCtx* zc, const U32 reducerValue)
{
2018-01-12 20:06:10 +00:00
ZSTD_matchState_t* const ms = &zc->blockState.matchState;
{ U32 const hSize = (U32)1 << zc->appliedParams.cParams.hashLog;
ZSTD_reduceTable(ms->hashTable, hSize, reducerValue);
}
2016-03-20 15:20:06 +00:00
if (zc->appliedParams.cParams.strategy != ZSTD_fast) {
U32 const chainSize = (U32)1 << zc->appliedParams.cParams.chainLog;
if (zc->appliedParams.cParams.strategy == ZSTD_btlazy2)
ZSTD_preserveUnsortedMark(ms->chainTable, chainSize, reducerValue);
ZSTD_reduceTable(ms->chainTable, chainSize, reducerValue);
}
2016-03-20 15:20:06 +00:00
if (ms->hashLog3) {
U32 const h3Size = (U32)1 << ms->hashLog3;
ZSTD_reduceTable(ms->hashTable3, h3Size, reducerValue);
}
if (zc->appliedParams.ldmParams.enableLdm) {
U32 const ldmHSize = (U32)1 << zc->appliedParams.ldmParams.hashLog;
ZSTD_ldm_reduceTable(zc->ldmState.hashTable, ldmHSize, reducerValue);
}
2016-03-20 15:20:06 +00:00
}
2016-01-28 16:56:33 +00:00
/*-*******************************************************
2015-11-11 20:38:21 +00:00
* Block entropic compression
*********************************************************/
/* See doc/zstd_compression_format.md for detailed format description */
2015-11-11 20:38:21 +00:00
2016-03-15 00:24:33 +00:00
size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
2015-11-11 20:38:21 +00:00
{
2016-03-15 00:24:33 +00:00
if (srcSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
MEM_writeLE24(dst, (U32)(srcSize << 2) + (U32)bt_raw);
2015-11-11 20:38:21 +00:00
return ZSTD_blockHeaderSize+srcSize;
}
2016-03-15 00:24:33 +00:00
static size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
2015-11-11 20:38:21 +00:00
{
BYTE* const ostart = (BYTE* const)dst;
2016-07-27 19:05:12 +00:00
U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
2015-11-11 20:38:21 +00:00
2016-03-15 00:24:33 +00:00
if (srcSize + flSize > dstCapacity) return ERROR(dstSize_tooSmall);
2015-11-11 20:38:21 +00:00
2016-01-23 18:28:41 +00:00
switch(flSize)
{
case 1: /* 2 - 1 - 5 */
2016-07-23 14:31:49 +00:00
ostart[0] = (BYTE)((U32)set_basic + (srcSize<<3));
2016-01-23 18:28:41 +00:00
break;
case 2: /* 2 - 2 - 12 */
2016-07-23 14:31:49 +00:00
MEM_writeLE16(ostart, (U16)((U32)set_basic + (1<<2) + (srcSize<<4)));
2016-01-23 18:28:41 +00:00
break;
case 3: /* 2 - 2 - 20 */
2016-07-23 14:31:49 +00:00
MEM_writeLE32(ostart, (U32)((U32)set_basic + (3<<2) + (srcSize<<4)));
2016-01-23 18:28:41 +00:00
break;
default: /* not necessary : flSize is {1,2,3} */
assert(0);
2016-01-23 18:28:41 +00:00
}
memcpy(ostart + flSize, src, srcSize);
return srcSize + flSize;
2015-11-11 20:38:21 +00:00
}
2016-03-15 00:24:33 +00:00
static size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
2015-11-11 20:38:21 +00:00
{
BYTE* const ostart = (BYTE* const)dst;
2016-07-27 19:05:12 +00:00
U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
2016-01-23 18:28:41 +00:00
(void)dstCapacity; /* dstCapacity already guaranteed to be >=4, hence large enough */
2016-01-23 18:28:41 +00:00
switch(flSize)
{
case 1: /* 2 - 1 - 5 */
2016-07-23 14:31:49 +00:00
ostart[0] = (BYTE)((U32)set_rle + (srcSize<<3));
2016-01-23 18:28:41 +00:00
break;
case 2: /* 2 - 2 - 12 */
2016-07-23 14:31:49 +00:00
MEM_writeLE16(ostart, (U16)((U32)set_rle + (1<<2) + (srcSize<<4)));
2016-01-23 18:28:41 +00:00
break;
case 3: /* 2 - 2 - 20 */
2016-07-23 14:31:49 +00:00
MEM_writeLE32(ostart, (U32)((U32)set_rle + (3<<2) + (srcSize<<4)));
2016-01-23 18:28:41 +00:00
break;
default: /* not necessary : flSize is {1,2,3} */
assert(0);
2016-01-23 18:28:41 +00:00
}
2015-11-11 20:38:21 +00:00
2016-01-23 18:28:41 +00:00
ostart[flSize] = *(const BYTE*)src;
return flSize+1;
2015-11-11 20:38:21 +00:00
}
2016-01-23 18:28:41 +00:00
2016-03-20 00:09:18 +00:00
static size_t ZSTD_minGain(size_t srcSize) { return (srcSize >> 6) + 2; }
2015-11-11 20:38:21 +00:00
static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t const* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
ZSTD_strategy strategy,
2016-03-15 00:24:33 +00:00
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
U32* workspace)
2015-11-11 20:38:21 +00:00
{
2016-03-18 11:37:45 +00:00
size_t const minGain = ZSTD_minGain(srcSize);
size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
2016-07-27 19:05:12 +00:00
BYTE* const ostart = (BYTE*)dst;
2016-01-25 03:10:46 +00:00
U32 singleStream = srcSize < 256;
2016-07-23 14:31:49 +00:00
symbolEncodingType_e hType = set_compressed;
2016-03-18 11:37:45 +00:00
size_t cLitSize;
2015-11-11 20:38:21 +00:00
/* Prepare nextEntropy assuming reusing the existing table */
nextEntropy->hufCTable_repeatMode = prevEntropy->hufCTable_repeatMode;
memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable,
sizeof(prevEntropy->hufCTable));
2015-11-11 20:38:21 +00:00
2016-03-20 00:09:18 +00:00
/* small ? don't even attempt compression (speed opt) */
# define LITERAL_NOENTROPY 63
{ size_t const minLitSize = prevEntropy->hufCTable_repeatMode == HUF_repeat_valid ? 6 : LITERAL_NOENTROPY;
2016-03-20 00:09:18 +00:00
if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
}
if (dstCapacity < lhSize+1) return ERROR(dstSize_tooSmall); /* not enough space for compression */
{ HUF_repeat repeat = prevEntropy->hufCTable_repeatMode;
int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
Allow compressor to repeat Huffman tables * Compressor saves most recently used Huffman table and reuses it if it produces better results. * I attempted to preserve CPU usage profile. I intentionally left all of the existing heuristics in place. There is only a speed difference on the second block and later. When compressing large enough blocks (say >= 4 KiB) there is no significant difference in compression speed. Dictionary compression of one block is the same speed for blocks with literals <= 1 KiB, and after that the difference is not very significant. * In the synthetic data, with blocks 10 KB or smaller, most blocks can't use repeated tables because the previous block did not contain a symbol that the current block contains. Once blocks are about 12 KB or more, most previous blocks have valid Huffman tables for the current block, and the compression ratio and decompression speed jumped. * In silesia blocks as small as 4KB can frequently reuse the previous Huffman table (85%), but it isn't as profitable, and the previous Huffman table only gets used about 3% of the time. * Microbenchmarks show that `HUF_validateCTable()` takes ~55 ns and `HUF_estimateCompressedSize()` takes ~35 ns. They are decently well optimized, the first versions took 90 ns and 120 ns respectively. `HUF_validateCTable()` could be twice as fast, if we cast the `HUF_CElt*` to a `U32*` and compare to 0. However, `U32` has an alignment of 4 instead of 2, so I think that might be undefined behavior. * I've ran `zstreamtest` compiled normally, with UASAN and with MSAN for 4 hours each. The worst case for the speed difference is a bunch of small blocks in the same frame. I modified `bench.c` to compress the input in a single frame but with blocks of the given block size, set by `-B`. Benchmarks on level 1: | Program | Block size | Corpus | Ratio | Compression MB/s | Decompression MB/s | |-----------|------------|-----------|-------|------------------|--------------------| | zstd.base | 256 | synthetic | 2.364 | 110.0 | 297.0 | | zstd | 256 | synthetic | 2.367 | 108.9 | 297.0 | | zstd.base | 256 | silesia | 2.204 | 93.8 | 415.7 | | zstd | 256 | silesia | 2.204 | 93.4 | 415.7 | | zstd.base | 512 | synthetic | 2.594 | 144.2 | 420.0 | | zstd | 512 | synthetic | 2.599 | 141.5 | 425.7 | | zstd.base | 512 | silesia | 2.358 | 118.4 | 432.6 | | zstd | 512 | silesia | 2.358 | 119.8 | 432.6 | | zstd.base | 1024 | synthetic | 2.790 | 192.3 | 594.1 | | zstd | 1024 | synthetic | 2.794 | 192.3 | 600.0 | | zstd.base | 1024 | silesia | 2.524 | 148.2 | 464.2 | | zstd | 1024 | silesia | 2.525 | 148.2 | 467.6 | | zstd.base | 4096 | synthetic | 3.023 | 300.0 | 1000.0 | | zstd | 4096 | synthetic | 3.024 | 300.0 | 1010.1 | | zstd.base | 4096 | silesia | 2.779 | 223.1 | 623.5 | | zstd | 4096 | silesia | 2.779 | 223.1 | 636.0 | | zstd.base | 16384 | synthetic | 3.131 | 350.0 | 1150.1 | | zstd | 16384 | synthetic | 3.152 | 350.0 | 1630.3 | | zstd.base | 16384 | silesia | 2.871 | 296.5 | 883.3 | | zstd | 16384 | silesia | 2.872 | 294.4 | 898.3 |
2017-03-02 01:51:56 +00:00
if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
2017-04-20 18:14:13 +00:00
cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat)
2017-04-20 18:14:13 +00:00
: HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat);
if (repeat != HUF_repeat_none) {
/* reused the existing table */
hType = set_repeat;
}
2016-01-26 02:14:20 +00:00
}
2015-11-11 20:38:21 +00:00
if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable));
2016-03-18 11:37:45 +00:00
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
Allow compressor to repeat Huffman tables * Compressor saves most recently used Huffman table and reuses it if it produces better results. * I attempted to preserve CPU usage profile. I intentionally left all of the existing heuristics in place. There is only a speed difference on the second block and later. When compressing large enough blocks (say >= 4 KiB) there is no significant difference in compression speed. Dictionary compression of one block is the same speed for blocks with literals <= 1 KiB, and after that the difference is not very significant. * In the synthetic data, with blocks 10 KB or smaller, most blocks can't use repeated tables because the previous block did not contain a symbol that the current block contains. Once blocks are about 12 KB or more, most previous blocks have valid Huffman tables for the current block, and the compression ratio and decompression speed jumped. * In silesia blocks as small as 4KB can frequently reuse the previous Huffman table (85%), but it isn't as profitable, and the previous Huffman table only gets used about 3% of the time. * Microbenchmarks show that `HUF_validateCTable()` takes ~55 ns and `HUF_estimateCompressedSize()` takes ~35 ns. They are decently well optimized, the first versions took 90 ns and 120 ns respectively. `HUF_validateCTable()` could be twice as fast, if we cast the `HUF_CElt*` to a `U32*` and compare to 0. However, `U32` has an alignment of 4 instead of 2, so I think that might be undefined behavior. * I've ran `zstreamtest` compiled normally, with UASAN and with MSAN for 4 hours each. The worst case for the speed difference is a bunch of small blocks in the same frame. I modified `bench.c` to compress the input in a single frame but with blocks of the given block size, set by `-B`. Benchmarks on level 1: | Program | Block size | Corpus | Ratio | Compression MB/s | Decompression MB/s | |-----------|------------|-----------|-------|------------------|--------------------| | zstd.base | 256 | synthetic | 2.364 | 110.0 | 297.0 | | zstd | 256 | synthetic | 2.367 | 108.9 | 297.0 | | zstd.base | 256 | silesia | 2.204 | 93.8 | 415.7 | | zstd | 256 | silesia | 2.204 | 93.4 | 415.7 | | zstd.base | 512 | synthetic | 2.594 | 144.2 | 420.0 | | zstd | 512 | synthetic | 2.599 | 141.5 | 425.7 | | zstd.base | 512 | silesia | 2.358 | 118.4 | 432.6 | | zstd | 512 | silesia | 2.358 | 119.8 | 432.6 | | zstd.base | 1024 | synthetic | 2.790 | 192.3 | 594.1 | | zstd | 1024 | synthetic | 2.794 | 192.3 | 600.0 | | zstd.base | 1024 | silesia | 2.524 | 148.2 | 464.2 | | zstd | 1024 | silesia | 2.525 | 148.2 | 467.6 | | zstd.base | 4096 | synthetic | 3.023 | 300.0 | 1000.0 | | zstd | 4096 | synthetic | 3.024 | 300.0 | 1010.1 | | zstd.base | 4096 | silesia | 2.779 | 223.1 | 623.5 | | zstd | 4096 | silesia | 2.779 | 223.1 | 636.0 | | zstd.base | 16384 | synthetic | 3.131 | 350.0 | 1150.1 | | zstd | 16384 | synthetic | 3.152 | 350.0 | 1630.3 | | zstd.base | 16384 | silesia | 2.871 | 296.5 | 883.3 | | zstd | 16384 | silesia | 2.872 | 294.4 | 898.3 |
2017-03-02 01:51:56 +00:00
}
if (cLitSize==1) {
memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable));
2016-03-18 11:37:45 +00:00
return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
Allow compressor to repeat Huffman tables * Compressor saves most recently used Huffman table and reuses it if it produces better results. * I attempted to preserve CPU usage profile. I intentionally left all of the existing heuristics in place. There is only a speed difference on the second block and later. When compressing large enough blocks (say >= 4 KiB) there is no significant difference in compression speed. Dictionary compression of one block is the same speed for blocks with literals <= 1 KiB, and after that the difference is not very significant. * In the synthetic data, with blocks 10 KB or smaller, most blocks can't use repeated tables because the previous block did not contain a symbol that the current block contains. Once blocks are about 12 KB or more, most previous blocks have valid Huffman tables for the current block, and the compression ratio and decompression speed jumped. * In silesia blocks as small as 4KB can frequently reuse the previous Huffman table (85%), but it isn't as profitable, and the previous Huffman table only gets used about 3% of the time. * Microbenchmarks show that `HUF_validateCTable()` takes ~55 ns and `HUF_estimateCompressedSize()` takes ~35 ns. They are decently well optimized, the first versions took 90 ns and 120 ns respectively. `HUF_validateCTable()` could be twice as fast, if we cast the `HUF_CElt*` to a `U32*` and compare to 0. However, `U32` has an alignment of 4 instead of 2, so I think that might be undefined behavior. * I've ran `zstreamtest` compiled normally, with UASAN and with MSAN for 4 hours each. The worst case for the speed difference is a bunch of small blocks in the same frame. I modified `bench.c` to compress the input in a single frame but with blocks of the given block size, set by `-B`. Benchmarks on level 1: | Program | Block size | Corpus | Ratio | Compression MB/s | Decompression MB/s | |-----------|------------|-----------|-------|------------------|--------------------| | zstd.base | 256 | synthetic | 2.364 | 110.0 | 297.0 | | zstd | 256 | synthetic | 2.367 | 108.9 | 297.0 | | zstd.base | 256 | silesia | 2.204 | 93.8 | 415.7 | | zstd | 256 | silesia | 2.204 | 93.4 | 415.7 | | zstd.base | 512 | synthetic | 2.594 | 144.2 | 420.0 | | zstd | 512 | synthetic | 2.599 | 141.5 | 425.7 | | zstd.base | 512 | silesia | 2.358 | 118.4 | 432.6 | | zstd | 512 | silesia | 2.358 | 119.8 | 432.6 | | zstd.base | 1024 | synthetic | 2.790 | 192.3 | 594.1 | | zstd | 1024 | synthetic | 2.794 | 192.3 | 600.0 | | zstd.base | 1024 | silesia | 2.524 | 148.2 | 464.2 | | zstd | 1024 | silesia | 2.525 | 148.2 | 467.6 | | zstd.base | 4096 | synthetic | 3.023 | 300.0 | 1000.0 | | zstd | 4096 | synthetic | 3.024 | 300.0 | 1010.1 | | zstd.base | 4096 | silesia | 2.779 | 223.1 | 623.5 | | zstd | 4096 | silesia | 2.779 | 223.1 | 636.0 | | zstd.base | 16384 | synthetic | 3.131 | 350.0 | 1150.1 | | zstd | 16384 | synthetic | 3.152 | 350.0 | 1630.3 | | zstd.base | 16384 | silesia | 2.871 | 296.5 | 883.3 | | zstd | 16384 | silesia | 2.872 | 294.4 | 898.3 |
2017-03-02 01:51:56 +00:00
}
2015-11-11 20:38:21 +00:00
if (hType == set_compressed) {
/* using a newly constructed table */
nextEntropy->hufCTable_repeatMode = HUF_repeat_check;
}
2015-11-11 20:38:21 +00:00
/* Build header */
2016-01-23 18:28:41 +00:00
switch(lhSize)
2015-11-11 20:38:21 +00:00
{
2016-01-23 18:28:41 +00:00
case 3: /* 2 - 2 - 10 - 10 */
2016-07-22 15:30:52 +00:00
{ U32 const lhc = hType + ((!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
MEM_writeLE24(ostart, lhc);
break;
}
2016-01-23 18:28:41 +00:00
case 4: /* 2 - 2 - 14 - 14 */
2016-07-22 02:45:06 +00:00
{ U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
MEM_writeLE32(ostart, lhc);
break;
}
2016-01-23 18:28:41 +00:00
case 5: /* 2 - 2 - 18 - 18 */
2016-07-22 02:45:06 +00:00
{ U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
MEM_writeLE32(ostart, lhc);
ostart[4] = (BYTE)(cLitSize >> 10);
break;
}
default: /* not possible : lhSize is {3,4,5} */
assert(0);
2015-11-11 20:38:21 +00:00
}
2016-03-18 11:37:45 +00:00
return lhSize+cLitSize;
2015-11-11 20:38:21 +00:00
}
2016-07-29 19:22:17 +00:00
void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
{
2016-07-29 22:55:13 +00:00
const seqDef* const sequences = seqStorePtr->sequencesStart;
2016-07-29 19:22:17 +00:00
BYTE* const llCodeTable = seqStorePtr->llCode;
BYTE* const ofCodeTable = seqStorePtr->ofCode;
BYTE* const mlCodeTable = seqStorePtr->mlCode;
2016-07-29 22:55:13 +00:00
U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
2016-07-29 19:22:17 +00:00
U32 u;
for (u=0; u<nbSeq; u++) {
U32 const llv = sequences[u].litLength;
U32 const mlv = sequences[u].matchLength;
llCodeTable[u] = (BYTE)ZSTD_LLcode(llv);
2016-07-29 19:22:17 +00:00
ofCodeTable[u] = (BYTE)ZSTD_highbit32(sequences[u].offset);
mlCodeTable[u] = (BYTE)ZSTD_MLcode(mlv);
2016-04-07 15:19:00 +00:00
}
2016-07-29 19:22:17 +00:00
if (seqStorePtr->longLengthID==1)
llCodeTable[seqStorePtr->longLengthPos] = MaxLL;
if (seqStorePtr->longLengthID==2)
mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
}
2017-09-18 23:54:53 +00:00
typedef enum {
ZSTD_defaultDisallowed = 0,
ZSTD_defaultAllowed = 1
} ZSTD_defaultPolicy_e;
MEM_STATIC
symbolEncodingType_e ZSTD_selectEncodingType(
2017-09-18 23:54:53 +00:00
FSE_repeat* repeatMode, size_t const mostFrequent, size_t nbSeq,
U32 defaultNormLog, ZSTD_defaultPolicy_e const isDefaultAllowed)
{
#define MIN_SEQ_FOR_DYNAMIC_FSE 64
#define MAX_SEQ_FOR_STATIC_FSE 1000
2017-09-18 23:54:53 +00:00
ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
if ((mostFrequent == nbSeq) && (!isDefaultAllowed || nbSeq > 2)) {
DEBUGLOG(5, "Selected set_rle");
2017-09-18 23:54:53 +00:00
/* Prefer set_basic over set_rle when there are 2 or less symbols,
* since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
* If basic encoding isn't possible, always choose RLE.
*/
*repeatMode = FSE_repeat_check;
return set_rle;
}
if ( isDefaultAllowed
&& (*repeatMode == FSE_repeat_valid) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
DEBUGLOG(5, "Selected set_repeat");
return set_repeat;
}
if ( isDefaultAllowed
&& ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (defaultNormLog-1)))) ) {
DEBUGLOG(5, "Selected set_basic");
/* The format allows default tables to be repeated, but it isn't useful.
* When using simple heuristics to select encoding type, we don't want
* to confuse these tables with dictionaries. When running more careful
* analysis, we don't need to waste time checking both repeating tables
* and default tables.
*/
*repeatMode = FSE_repeat_none;
return set_basic;
}
DEBUGLOG(5, "Selected set_compressed");
*repeatMode = FSE_repeat_check;
return set_compressed;
}
MEM_STATIC
size_t ZSTD_buildCTable(void* dst, size_t dstCapacity,
FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
U32* count, U32 max,
BYTE const* codeTable, size_t nbSeq,
S16 const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
FSE_CTable const* prevCTable, size_t prevCTableSize,
void* workspace, size_t workspaceSize)
{
BYTE* op = (BYTE*)dst;
BYTE const* const oend = op + dstCapacity;
switch (type) {
case set_rle:
*op = codeTable[0];
CHECK_F(FSE_buildCTable_rle(nextCTable, (BYTE)max));
return 1;
case set_repeat:
memcpy(nextCTable, prevCTable, prevCTableSize);
return 0;
case set_basic:
CHECK_F(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, workspace, workspaceSize)); /* note : could be pre-calculated */
return 0;
case set_compressed: {
S16 norm[MaxSeq + 1];
size_t nbSeq_1 = nbSeq;
const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
if (count[codeTable[nbSeq-1]] > 1) {
count[codeTable[nbSeq-1]]--;
nbSeq_1--;
}
2017-09-18 23:54:53 +00:00
assert(nbSeq_1 > 1);
CHECK_F(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max));
{ size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog); /* overflow protected */
if (FSE_isError(NCountSize)) return NCountSize;
CHECK_F(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, workspace, workspaceSize));
return NCountSize;
}
}
default: return assert(0), ERROR(GENERIC);
}
}
MEM_STATIC
size_t ZSTD_encodeSequences(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets)
{
BIT_CStream_t blockStream;
FSE_CState_t stateMatchLength;
FSE_CState_t stateOffsetBits;
FSE_CState_t stateLitLength;
CHECK_E(BIT_initCStream(&blockStream, dst, dstCapacity), dstSize_tooSmall); /* not enough space remaining */
/* first symbols */
FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
FSE_initCState2(&stateOffsetBits, CTable_OffsetBits, ofCodeTable[nbSeq-1]);
FSE_initCState2(&stateLitLength, CTable_LitLength, llCodeTable[nbSeq-1]);
BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
if (MEM_32bits()) BIT_flushBits(&blockStream);
BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
if (MEM_32bits()) BIT_flushBits(&blockStream);
if (longOffsets) {
U32 const ofBits = ofCodeTable[nbSeq-1];
int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
if (extraBits) {
BIT_addBits(&blockStream, sequences[nbSeq-1].offset, extraBits);
BIT_flushBits(&blockStream);
}
BIT_addBits(&blockStream, sequences[nbSeq-1].offset >> extraBits,
ofBits - extraBits);
} else {
BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
}
BIT_flushBits(&blockStream);
{ size_t n;
for (n=nbSeq-2 ; n<nbSeq ; n--) { /* intentional underflow */
BYTE const llCode = llCodeTable[n];
BYTE const ofCode = ofCodeTable[n];
BYTE const mlCode = mlCodeTable[n];
U32 const llBits = LL_bits[llCode];
U32 const ofBits = ofCode;
U32 const mlBits = ML_bits[mlCode];
DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
sequences[n].litLength,
sequences[n].matchLength + MINMATCH,
sequences[n].offset); /* 32b*/ /* 64b*/
/* (7)*/ /* (7)*/
FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode); /* 15 */ /* 15 */
FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode); /* 24 */ /* 24 */
if (MEM_32bits()) BIT_flushBits(&blockStream); /* (7)*/
FSE_encodeSymbol(&blockStream, &stateLitLength, llCode); /* 16 */ /* 33 */
if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
BIT_flushBits(&blockStream); /* (7)*/
BIT_addBits(&blockStream, sequences[n].litLength, llBits);
if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
if (longOffsets) {
int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
if (extraBits) {
BIT_addBits(&blockStream, sequences[n].offset, extraBits);
BIT_flushBits(&blockStream); /* (7)*/
}
BIT_addBits(&blockStream, sequences[n].offset >> extraBits,
ofBits - extraBits); /* 31 */
} else {
BIT_addBits(&blockStream, sequences[n].offset, ofBits); /* 31 */
}
BIT_flushBits(&blockStream); /* (7)*/
} }
FSE_flushCState(&blockStream, &stateMatchLength);
FSE_flushCState(&blockStream, &stateOffsetBits);
FSE_flushCState(&blockStream, &stateLitLength);
{ size_t const streamSize = BIT_closeCStream(&blockStream);
if (streamSize==0) return ERROR(dstSize_tooSmall); /* not enough space */
return streamSize;
}
}
MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
ZSTD_entropyCTables_t const* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
ZSTD_compressionParameters const* cParams,
void* dst, size_t dstCapacity, U32* workspace)
2015-11-11 20:38:21 +00:00
{
const int longOffsets = cParams->windowLog > STREAM_ACCUMULATOR_MIN;
2015-11-11 20:38:21 +00:00
U32 count[MaxSeq+1];
FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
2015-11-11 20:38:21 +00:00
U32 LLtype, Offtype, MLtype; /* compressed, raw or rle */
2016-07-29 22:55:13 +00:00
const seqDef* const sequences = seqStorePtr->sequencesStart;
2016-07-29 19:22:17 +00:00
const BYTE* const ofCodeTable = seqStorePtr->ofCode;
const BYTE* const llCodeTable = seqStorePtr->llCode;
const BYTE* const mlCodeTable = seqStorePtr->mlCode;
2015-11-23 12:34:21 +00:00
BYTE* const ostart = (BYTE*)dst;
2016-03-15 00:24:33 +00:00
BYTE* const oend = ostart + dstCapacity;
2016-03-18 11:37:45 +00:00
BYTE* op = ostart;
2016-07-29 22:55:13 +00:00
size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
2015-11-11 20:38:21 +00:00
BYTE* seqHead;
ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
2015-11-11 20:38:21 +00:00
/* Compress literals */
2016-03-20 00:09:18 +00:00
{ const BYTE* const literals = seqStorePtr->litStart;
2016-03-18 11:37:45 +00:00
size_t const litSize = seqStorePtr->lit - literals;
size_t const cSize = ZSTD_compressLiterals(prevEntropy, nextEntropy,
cParams->strategy, op, dstCapacity, literals, litSize,
workspace);
if (ZSTD_isError(cSize))
return cSize;
assert(cSize <= dstCapacity);
2015-11-11 20:38:21 +00:00
op += cSize;
}
/* Sequences Header */
if ((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/) return ERROR(dstSize_tooSmall);
if (nbSeq < 0x7F)
*op++ = (BYTE)nbSeq;
else if (nbSeq < LONGNBSEQ)
op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
else
op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
if (nbSeq==0) {
memcpy(nextEntropy->litlengthCTable, prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable));
nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
memcpy(nextEntropy->offcodeCTable, prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable));
nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
memcpy(nextEntropy->matchlengthCTable, prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable));
nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
return op - ostart;
}
2015-11-11 20:38:21 +00:00
/* seqHead : flags for FSE encoding type */
seqHead = op++;
2015-11-11 20:38:21 +00:00
/* convert length/distances into codes */
2016-07-29 19:22:17 +00:00
ZSTD_seqToCodes(seqStorePtr);
/* build CTable for Literal Lengths */
{ U32 max = MaxLL;
size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, workspace);
DEBUGLOG(5, "Building LL table");
nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode, mostFrequent, nbSeq, LL_defaultNormLog, ZSTD_defaultAllowed);
{ size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable),
workspace, HUF_WORKSPACE_SIZE);
if (ZSTD_isError(countSize)) return countSize;
op += countSize;
} }
/* build CTable for Offsets */
{ U32 max = MaxOff;
size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, workspace);
2017-09-18 23:54:53 +00:00
/* 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, mostFrequent, nbSeq, OF_defaultNormLog, defaultPolicy);
{ size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
2017-09-18 23:54:53 +00:00
count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable),
workspace, HUF_WORKSPACE_SIZE);
if (ZSTD_isError(countSize)) return countSize;
op += countSize;
} }
/* build CTable for MatchLengths */
{ U32 max = MaxML;
size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, workspace);
DEBUGLOG(5, "Building ML table");
nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode, mostFrequent, nbSeq, ML_defaultNormLog, ZSTD_defaultAllowed);
{ size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable),
workspace, HUF_WORKSPACE_SIZE);
if (ZSTD_isError(countSize)) return countSize;
op += countSize;
} }
2015-11-11 20:38:21 +00:00
*seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
2015-11-11 20:38:21 +00:00
{ size_t const bitstreamSize = ZSTD_encodeSequences(
op, oend - op,
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq,
longOffsets);
if (ZSTD_isError(bitstreamSize)) return bitstreamSize;
op += bitstreamSize;
}
2015-11-11 20:38:21 +00:00
return op - ostart;
}
2015-11-11 20:38:21 +00:00
MEM_STATIC size_t ZSTD_compressSequences(seqStore_t* seqStorePtr,
ZSTD_entropyCTables_t const* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
ZSTD_compressionParameters const* cParams,
void* dst, size_t dstCapacity,
size_t srcSize, U32* workspace)
{
size_t const cSize = ZSTD_compressSequences_internal(
seqStorePtr, prevEntropy, nextEntropy, cParams, dst, dstCapacity,
workspace);
/* If the srcSize <= dstCapacity, then there is enough space to write a
* raw uncompressed block. Since we ran out of space, the block must not
* be compressible, so fall back to a raw uncompressed block.
*/
int const uncompressibleError = (cSize == ERROR(dstSize_tooSmall)) && (srcSize <= dstCapacity);
if (ZSTD_isError(cSize) && !uncompressibleError)
return cSize;
/* Check compressibility */
{ size_t const minGain = ZSTD_minGain(srcSize); /* note : fixed formula, maybe should depend on compression level, or strategy */
size_t const maxCSize = srcSize - minGain;
if (cSize >= maxCSize || uncompressibleError) {
return 0; /* block not compressed */
} }
assert(!ZSTD_isError(cSize));
2015-11-11 20:38:21 +00:00
/* We check that dictionaries have offset codes available for the first
* block. After the first block, the offcode table might not have large
* enough codes to represent the offsets in the data.
*/
if (nextEntropy->offcode_repeatMode == FSE_repeat_valid)
nextEntropy->offcode_repeatMode = FSE_repeat_check;
return cSize;
2015-11-11 20:38:21 +00:00
}
/* ZSTD_selectBlockCompressor() :
2017-09-07 00:56:01 +00:00
* Not static, but internal use only (used by long distance matcher)
* assumption : strat is a valid strategy */
2017-09-07 00:56:01 +00:00
ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict)
2015-10-31 11:57:14 +00:00
{
static const ZSTD_blockCompressor blockCompressor[2][(unsigned)ZSTD_btultra+1] = {
{ ZSTD_compressBlock_fast /* default for 0 */,
ZSTD_compressBlock_fast, ZSTD_compressBlock_doubleFast, ZSTD_compressBlock_greedy,
ZSTD_compressBlock_lazy, ZSTD_compressBlock_lazy2, ZSTD_compressBlock_btlazy2,
ZSTD_compressBlock_btopt, ZSTD_compressBlock_btultra },
{ ZSTD_compressBlock_fast_extDict /* default for 0 */,
ZSTD_compressBlock_fast_extDict, ZSTD_compressBlock_doubleFast_extDict, ZSTD_compressBlock_greedy_extDict,
ZSTD_compressBlock_lazy_extDict,ZSTD_compressBlock_lazy2_extDict, ZSTD_compressBlock_btlazy2_extDict,
ZSTD_compressBlock_btopt_extDict, ZSTD_compressBlock_btultra_extDict }
};
ZSTD_STATIC_ASSERT((unsigned)ZSTD_fast == 1);
assert((U32)strat >= (U32)ZSTD_fast);
assert((U32)strat <= (U32)ZSTD_btultra);
return blockCompressor[extDict!=0][(U32)strat];
2015-10-31 11:57:14 +00:00
}
static void ZSTD_storeLastLiterals(seqStore_t* seqStorePtr,
const BYTE* anchor, size_t lastLLSize)
{
memcpy(seqStorePtr->lit, anchor, lastLLSize);
seqStorePtr->lit += lastLLSize;
}
2015-10-31 11:57:14 +00:00
static void ZSTD_resetSeqStore(seqStore_t* ssPtr)
{
ssPtr->lit = ssPtr->litStart;
ssPtr->sequences = ssPtr->sequencesStart;
ssPtr->longLengthID = 0;
}
static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize)
2015-11-04 11:05:27 +00:00
{
DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
2018-01-12 20:06:10 +00:00
(U32)dstCapacity, zc->blockState.matchState.dictLimit, zc->blockState.matchState.nextToUpdate);
if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1)
return 0; /* don't even attempt compression below a certain srcSize */
ZSTD_resetSeqStore(&(zc->seqStore));
/* limited update after a very long match */
2018-01-12 20:06:10 +00:00
{ const BYTE* const base = zc->blockState.matchState.base;
const BYTE* const istart = (const BYTE*)src;
const U32 current = (U32)(istart-base);
2018-01-12 20:06:10 +00:00
if (current > zc->blockState.matchState.nextToUpdate + 384)
zc->blockState.matchState.nextToUpdate = current - MIN(192, (U32)(current - zc->blockState.matchState.nextToUpdate - 384));
}
/* select and store sequences */
{ U32 const extDict = zc->blockState.matchState.lowLimit < zc->blockState.matchState.dictLimit;
size_t lastLLSize;
2018-01-12 20:06:10 +00:00
{ int i; for (i = 0; i < ZSTD_REP_NUM; ++i) zc->blockState.nextCBlock->rep[i] = zc->blockState.prevCBlock->rep[i]; }
if (zc->appliedParams.ldmParams.enableLdm) {
typedef size_t (*ZSTD_ldmBlockCompressor)(
ldmState_t* ldms, ZSTD_matchState_t* ms, seqStore_t* seqStore,
U32 rep[ZSTD_REP_NUM], ZSTD_CCtx_params const* params,
void const* src, size_t srcSize);
ZSTD_ldmBlockCompressor const ldmBlockCompressor = extDict ? ZSTD_compressBlock_ldm_extDict : ZSTD_compressBlock_ldm;
2018-01-12 20:06:10 +00:00
lastLLSize = ldmBlockCompressor(&zc->ldmState, &zc->blockState.matchState, &zc->seqStore, zc->blockState.nextCBlock->rep, &zc->appliedParams, src, srcSize);
} else { /* not long range mode */
ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy, extDict);
2018-01-12 20:06:10 +00:00
lastLLSize = blockCompressor(&zc->blockState.matchState, &zc->seqStore, zc->blockState.nextCBlock->rep, &zc->appliedParams.cParams, src, srcSize);
}
{ const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
ZSTD_storeLastLiterals(&zc->seqStore, lastLiterals, lastLLSize);
} }
/* encode sequences and literals */
{ size_t const cSize = ZSTD_compressSequences(&zc->seqStore, &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy, &zc->appliedParams.cParams, dst, dstCapacity, srcSize, zc->entropyWorkspace);
if (ZSTD_isError(cSize) || cSize == 0) return cSize;
/* confirm repcodes and entropy tables */
{ ZSTD_compressedBlockState_t* const tmp = zc->blockState.prevCBlock;
2018-01-12 20:06:10 +00:00
zc->blockState.prevCBlock = zc->blockState.nextCBlock;
zc->blockState.nextCBlock = tmp;
}
return cSize;
}
2015-11-04 11:05:27 +00:00
}
/*! ZSTD_compress_frameChunk() :
2016-07-27 22:55:43 +00:00
* Compress a chunk of data into one or multiple blocks.
* All blocks will be terminated, all input will be consumed.
* Function will issue an error if there is not enough `dstCapacity` to hold the compressed content.
* Frame is supposed already started (header already produced)
* @return : compressed size, or an error code
*/
static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
2016-05-31 16:13:56 +00:00
void* dst, size_t dstCapacity,
2016-07-27 22:55:43 +00:00
const void* src, size_t srcSize,
U32 lastFrameChunk)
2015-10-22 14:31:46 +00:00
{
2016-05-31 16:13:56 +00:00
size_t blockSize = cctx->blockSize;
2015-10-22 14:31:46 +00:00
size_t remaining = srcSize;
const BYTE* ip = (const BYTE*)src;
BYTE* const ostart = (BYTE*)dst;
BYTE* op = ostart;
U32 const maxDist = (U32)1 << cctx->appliedParams.cParams.windowLog;
assert(cctx->appliedParams.cParams.windowLog <= 31);
2015-11-01 11:40:22 +00:00
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
DEBUGLOG(5, "ZSTD_compress_frameChunk (blockSize=%u)", (U32)blockSize);
if (cctx->appliedParams.fParams.checksumFlag && srcSize)
2016-05-31 16:13:56 +00:00
XXH64_update(&cctx->xxhState, src, srcSize);
2016-02-02 13:36:49 +00:00
while (remaining) {
2018-01-12 20:06:10 +00:00
ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
2016-07-27 22:55:43 +00:00
U32 const lastBlock = lastFrameChunk & (blockSize >= remaining);
2015-10-22 14:31:46 +00:00
if (dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE)
return ERROR(dstSize_tooSmall); /* not enough space to store compressed block */
2015-11-04 17:19:39 +00:00
if (remaining < blockSize) blockSize = remaining;
2017-09-21 23:18:34 +00:00
/* preemptive overflow correction:
* 1. correction is large enough:
* lowLimit > (3<<29) ==> current > 3<<29 + 1<<windowLog - blockSize
* 1<<windowLog <= newCurrent < 1<<chainLog + 1<<windowLog
*
* current - newCurrent
* > (3<<29 + 1<<windowLog - blockSize) - (1<<windowLog + 1<<chainLog)
* > (3<<29 - blockSize) - (1<<chainLog)
* > (3<<29 - blockSize) - (1<<30) (NOTE: chainLog <= 30)
* > 1<<29 - 1<<17
*
* 2. (ip+blockSize - cctx->base) doesn't overflow:
* In 32 bit mode we limit windowLog to 30 so we don't get
* differences larger than 1<<31-1.
* 3. cctx->lowLimit < 1<<32:
* windowLog <= 31 ==> 3<<29 + 1<<windowLog < 7<<29 < 1<<32.
*/
if (ms->lowLimit > (3U<<29)) {
U32 const cycleMask = ((U32)1 << ZSTD_cycleLog(cctx->appliedParams.cParams.chainLog, cctx->appliedParams.cParams.strategy)) - 1;
2018-01-12 20:06:10 +00:00
U32 const current = (U32)(ip - cctx->blockState.matchState.base);
U32 const newCurrent = (current & cycleMask) + ((U32)1 << cctx->appliedParams.cParams.windowLog);
2016-12-11 23:25:07 +00:00
U32 const correction = current - newCurrent;
2017-09-21 23:18:34 +00:00
ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
assert(current > newCurrent);
assert(correction > 1<<28); /* Loose bound, should be about 1<<29 */
2016-08-02 12:26:00 +00:00
ZSTD_reduceIndex(cctx, correction);
ms->base += correction;
ms->dictBase += correction;
ms->lowLimit -= correction;
ms->dictLimit -= correction;
if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
else ms->nextToUpdate -= correction;
DEBUGLOG(4, "Correction of 0x%x bytes to lowLimit=0x%x", correction, ms->lowLimit);
2016-08-02 12:26:00 +00:00
}
/* enforce maxDist */
if ((U32)(ip+blockSize - ms->base) > ms->loadedDictEnd + maxDist) {
U32 const newLowLimit = (U32)(ip+blockSize - ms->base) - maxDist;
if (ms->lowLimit < newLowLimit) ms->lowLimit = newLowLimit;
if (ms->dictLimit < ms->lowLimit)
DEBUGLOG(5, "ZSTD_compress_frameChunk : update dictLimit from %u to %u ",
ms->dictLimit, ms->lowLimit);
if (ms->dictLimit < ms->lowLimit) ms->dictLimit = ms->lowLimit;
if (ms->nextToUpdate < ms->lowLimit) ms->nextToUpdate = ms->lowLimit;
2015-11-24 13:06:07 +00:00
}
{ size_t cSize = ZSTD_compressBlock_internal(cctx,
op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize,
ip, blockSize);
if (ZSTD_isError(cSize)) return cSize;
if (cSize == 0) { /* block is not compressible */
U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(blockSize << 3);
if (blockSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
MEM_writeLE32(op, cBlockHeader24); /* 4th byte will be overwritten */
memcpy(op + ZSTD_blockHeaderSize, ip, blockSize);
cSize = ZSTD_blockHeaderSize + blockSize;
} else {
U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
MEM_writeLE24(op, cBlockHeader24);
cSize += ZSTD_blockHeaderSize;
}
2015-10-22 14:31:46 +00:00
ip += blockSize;
assert(remaining >= blockSize);
remaining -= blockSize;
op += cSize;
assert(dstCapacity >= cSize);
dstCapacity -= cSize;
DEBUGLOG(5, "ZSTD_compress_frameChunk: adding a block of size %u",
(U32)cSize);
} }
2015-10-22 14:31:46 +00:00
2016-07-28 13:29:08 +00:00
if (lastFrameChunk && (op>ostart)) cctx->stage = ZSTDcs_ending;
2015-10-22 14:31:46 +00:00
return op-ostart;
}
2016-04-12 13:52:33 +00:00
static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
2017-08-18 18:20:08 +00:00
ZSTD_CCtx_params params, U64 pledgedSrcSize, U32 dictID)
2016-04-12 13:52:33 +00:00
{ BYTE* const op = (BYTE*)dst;
U32 const dictIDSizeCodeLength = (dictID>0) + (dictID>=256) + (dictID>=65536); /* 0-3 */
U32 const dictIDSizeCode = params.fParams.noDictIDFlag ? 0 : dictIDSizeCodeLength; /* 0-3 */
2016-07-27 19:05:12 +00:00
U32 const checksumFlag = params.fParams.checksumFlag>0;
U32 const windowSize = (U32)1 << params.cParams.windowLog;
U32 const singleSegment = params.fParams.contentSizeFlag && (windowSize >= pledgedSrcSize);
2016-07-27 19:05:12 +00:00
BYTE const windowLogByte = (BYTE)((params.cParams.windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN) << 3);
U32 const fcsCode = params.fParams.contentSizeFlag ?
(pledgedSrcSize>=256) + (pledgedSrcSize>=65536+256) + (pledgedSrcSize>=0xFFFFFFFFU) : 0; /* 0-3 */
2016-07-27 19:05:12 +00:00
BYTE const frameHeaderDecriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
size_t pos=0;
2016-05-29 03:01:04 +00:00
if (dstCapacity < ZSTD_frameHeaderSize_max) return ERROR(dstSize_tooSmall);
DEBUGLOG(4, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
!params.fParams.noDictIDFlag, dictID, dictIDSizeCode);
2016-04-12 13:52:33 +00:00
if (params.format == ZSTD_f_zstd1) {
MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
pos = 4;
}
op[pos++] = frameHeaderDecriptionByte;
if (!singleSegment) op[pos++] = windowLogByte;
2016-05-29 03:01:04 +00:00
switch(dictIDSizeCode)
{
default: assert(0); /* impossible */
2016-05-29 03:01:04 +00:00
case 0 : break;
case 1 : op[pos] = (BYTE)(dictID); pos++; break;
2016-07-27 19:21:36 +00:00
case 2 : MEM_writeLE16(op+pos, (U16)dictID); pos+=2; break;
2016-05-29 03:01:04 +00:00
case 3 : MEM_writeLE32(op+pos, dictID); pos+=4; break;
}
switch(fcsCode)
2016-04-12 13:52:33 +00:00
{
default: assert(0); /* impossible */
case 0 : if (singleSegment) op[pos++] = (BYTE)(pledgedSrcSize); break;
case 1 : MEM_writeLE16(op+pos, (U16)(pledgedSrcSize-256)); pos+=2; break;
case 2 : MEM_writeLE32(op+pos, (U32)(pledgedSrcSize)); pos+=4; break;
2016-05-29 03:01:04 +00:00
case 3 : MEM_writeLE64(op+pos, (U64)(pledgedSrcSize)); pos+=8; break;
2016-04-12 13:52:33 +00:00
}
2016-05-29 03:01:04 +00:00
return pos;
2016-04-12 13:52:33 +00:00
}
/* ZSTD_writeLastEmptyBlock() :
* output an empty Block with end-of-frame mark to complete a frame
* @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
* or an error code if `dstCapcity` is too small (<ZSTD_blockHeaderSize)
*/
size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity)
{
if (dstCapacity < ZSTD_blockHeaderSize) return ERROR(dstSize_tooSmall);
{ U32 const cBlockHeader24 = 1 /*lastBlock*/ + (((U32)bt_raw)<<1); /* 0 size */
MEM_writeLE24(dst, cBlockHeader24);
return ZSTD_blockHeaderSize;
}
}
2016-04-12 13:52:33 +00:00
static void ZSTD_manageWindowContinuity(ZSTD_matchState_t* ms, void const* src, size_t srcSize)
{
const BYTE* const ip = (const BYTE*) src;
/* Check if blocks follow each other */
if (src != ms->nextSrc) {
/* not contiguous */
size_t const distanceFromBase = (size_t)(ms->nextSrc - ms->base);
DEBUGLOG(5, "ZSTD_manageWindowContinuity: non contiguous blocks, new segment starts at %u", ms->dictLimit);
ms->lowLimit = ms->dictLimit;
assert(distanceFromBase == (size_t)(U32)distanceFromBase); /* should never overflow */
ms->dictLimit = (U32)distanceFromBase;
ms->dictBase = ms->base;
ms->base = ip - distanceFromBase;
ms->nextToUpdate = ms->dictLimit;
if (ms->dictLimit - ms->lowLimit < HASH_READ_SIZE) ms->lowLimit = ms->dictLimit; /* too small extDict */
}
ms->nextSrc = ip + srcSize;
/* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
if ((ip+srcSize > ms->dictBase + ms->lowLimit) & (ip < ms->dictBase + ms->dictLimit)) {
ptrdiff_t const highInputIdx = (ip + srcSize) - ms->dictBase;
U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)ms->dictLimit) ? ms->dictLimit : (U32)highInputIdx;
ms->lowLimit = lowLimitMax;
}
}
2016-08-02 12:26:00 +00:00
static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
2016-03-23 21:31:57 +00:00
void* dst, size_t dstCapacity,
2016-01-09 00:08:23 +00:00
const void* src, size_t srcSize,
2016-07-27 22:55:43 +00:00
U32 frame, U32 lastFrameChunk)
2015-10-22 14:31:46 +00:00
{
2016-04-12 13:52:33 +00:00
size_t fhSize = 0;
2016-01-07 14:35:18 +00:00
DEBUGLOG(5, "ZSTD_compressContinue_internal, stage: %u, srcSize: %u",
cctx->stage, (U32)srcSize);
2016-08-02 12:26:00 +00:00
if (cctx->stage==ZSTDcs_created) return ERROR(stage_wrong); /* missing init (ZSTD_compressBegin) */
2016-07-27 19:21:36 +00:00
2016-08-02 12:26:00 +00:00
if (frame && (cctx->stage==ZSTDcs_init)) {
fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->appliedParams,
2017-08-23 17:24:19 +00:00
cctx->pledgedSrcSizePlusOne-1, cctx->dictID);
2016-04-12 13:52:33 +00:00
if (ZSTD_isError(fhSize)) return fhSize;
dstCapacity -= fhSize;
dst = (char*)dst + fhSize;
2016-08-02 12:26:00 +00:00
cctx->stage = ZSTDcs_ongoing;
2016-01-07 14:35:18 +00:00
}
2015-10-22 14:31:46 +00:00
if (!srcSize) return fhSize; /* do not generate an empty block if no input */
2018-01-12 20:06:10 +00:00
ZSTD_manageWindowContinuity(&cctx->blockState.matchState, src, srcSize);
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
DEBUGLOG(5, "ZSTD_compressContinue_internal (blockSize=%u)", (U32)cctx->blockSize);
{ size_t const cSize = frame ?
ZSTD_compress_frameChunk (cctx, dst, dstCapacity, src, srcSize, lastFrameChunk) :
2016-08-02 12:26:00 +00:00
ZSTD_compressBlock_internal (cctx, dst, dstCapacity, src, srcSize);
2016-01-07 14:35:18 +00:00
if (ZSTD_isError(cSize)) return cSize;
cctx->consumedSrcSize += srcSize;
cctx->producedCSize += (cSize + fhSize);
2016-04-12 13:52:33 +00:00
return cSize + fhSize;
}
2015-10-22 14:31:46 +00:00
}
2016-07-27 23:17:22 +00:00
size_t ZSTD_compressContinue (ZSTD_CCtx* cctx,
2016-03-23 21:31:57 +00:00
void* dst, size_t dstCapacity,
2016-01-09 00:08:23 +00:00
const void* src, size_t srcSize)
{
DEBUGLOG(5, "ZSTD_compressContinue (srcSize=%u)", (U32)srcSize);
return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 0 /* last chunk */);
2016-07-27 23:17:22 +00:00
}
size_t ZSTD_getBlockSize(const ZSTD_CCtx* cctx)
2016-01-09 00:08:23 +00:00
{
ZSTD_compressionParameters const cParams =
2017-08-29 02:25:17 +00:00
ZSTD_getCParamsFromCCtxParams(cctx->appliedParams, 0, 0);
return MIN (ZSTD_BLOCKSIZE_MAX, (U32)1 << cParams.windowLog);
2016-07-18 14:52:10 +00:00
}
size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
{
size_t const blockSizeMax = ZSTD_getBlockSize(cctx);
if (srcSize > blockSizeMax) return ERROR(srcSize_wrong);
return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0 /* frame mode */, 0 /* last chunk */);
2016-01-09 00:08:23 +00:00
}
/*! ZSTD_loadDictionaryContent() :
* @return : 0, or an error code
*/
static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const void* src, size_t srcSize)
{
const BYTE* const ip = (const BYTE*) src;
const BYTE* const iend = ip + srcSize;
ZSTD_compressionParameters const* cParams = &params->cParams;
/* input becomes current prefix */
ms->lowLimit = ms->dictLimit;
ms->dictLimit = (U32)(ms->nextSrc - ms->base);
ms->dictBase = ms->base;
ms->base = ip - ms->dictLimit;
ms->nextToUpdate = ms->dictLimit;
ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->base);
ms->nextSrc = iend;
2016-07-27 19:05:12 +00:00
if (srcSize <= HASH_READ_SIZE) return 0;
switch(params->cParams.strategy)
{
case ZSTD_fast:
ZSTD_fillHashTable(ms, cParams, iend);
break;
2016-07-12 07:47:31 +00:00
case ZSTD_dfast:
ZSTD_fillDoubleHashTable(ms, cParams, iend);
2016-07-12 07:47:31 +00:00
break;
case ZSTD_greedy:
case ZSTD_lazy:
case ZSTD_lazy2:
if (srcSize >= HASH_READ_SIZE)
ZSTD_insertAndFindFirstIndex(ms, cParams, iend-HASH_READ_SIZE);
break;
case ZSTD_btlazy2: /* we want the dictionary table fully sorted */
2016-02-15 06:21:54 +00:00
case ZSTD_btopt:
2017-03-09 19:44:25 +00:00
case ZSTD_btultra:
if (srcSize >= HASH_READ_SIZE)
ZSTD_updateTree(ms, cParams, iend-HASH_READ_SIZE, iend);
break;
default:
assert(0); /* not possible : not a valid strategy id */
}
ms->nextToUpdate = (U32)(iend - ms->base);
return 0;
}
2015-10-22 14:31:46 +00:00
/* Dictionaries that assign zero probability to symbols that show up causes problems
when FSE encoding. Refuse dictionaries that assign zero probability to symbols
that we may encounter during compression.
NOTE: This behavior is not standard and could be improved in the future. */
static size_t ZSTD_checkDictNCount(short* normalizedCounter, unsigned dictMaxSymbolValue, unsigned maxSymbolValue) {
U32 s;
if (dictMaxSymbolValue < maxSymbolValue) return ERROR(dictionary_corrupted);
for (s = 0; s <= maxSymbolValue; ++s) {
if (normalizedCounter[s] == 0) return ERROR(dictionary_corrupted);
}
return 0;
}
2016-01-26 02:14:20 +00:00
/* Dictionary format :
* See :
* https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format
*/
/*! ZSTD_loadZstdDictionary() :
* @return : dictID, or an error code
* assumptions : magic number supposed already checked
* dictSize supposed > 8
*/
2018-01-12 20:06:10 +00:00
static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const void* dict, size_t dictSize, void* workspace)
2016-01-26 02:14:20 +00:00
{
const BYTE* dictPtr = (const BYTE*)dict;
const BYTE* const dictEnd = dictPtr + dictSize;
short offcodeNCount[MaxOff+1];
unsigned offcodeMaxValue = MaxOff;
size_t dictID;
ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
2016-05-31 16:13:56 +00:00
dictPtr += 4; /* skip magic number */
dictID = params->fParams.noDictIDFlag ? 0 : MEM_readLE32(dictPtr);
dictPtr += 4;
{ unsigned maxSymbolValue = 255;
size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.hufCTable, &maxSymbolValue, dictPtr, dictEnd-dictPtr);
2016-05-31 16:13:56 +00:00
if (HUF_isError(hufHeaderSize)) return ERROR(dictionary_corrupted);
if (maxSymbolValue < 255) return ERROR(dictionary_corrupted);
dictPtr += hufHeaderSize;
2016-05-31 16:13:56 +00:00
}
{ unsigned offcodeLog;
size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
2016-05-31 16:13:56 +00:00
if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
Fix buffer overrun in ZSTD_loadDictEntropyStats() The table log set by `FSE_readNCount()` was not checked in `ZSTD_loadDictEntropyStats()`. This caused `FSE_buildCTable()` to stack/heap overflow in a few places. The benchmarks look good, there is no obvious compression performance regression: > ./zstds/zstd.opt.0 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 271.6 MB/s , 716.8 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 204.8 MB/s , 671.1 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 156.8 MB/s , 658.6 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 136.4 MB/s , 665.3 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 98.9 MB/s , 647.0 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 65.2 MB/s , 670.4 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.9 MB/s , 688.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 32.4 MB/s , 711.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.1 MB/s , 718.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 22.2 MB/s , 721.8 MB/s > ./compress_zstds/zstd.opt.1 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 273.8 MB/s , 722.0 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 203.2 MB/s , 666.6 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 157.4 MB/s , 666.5 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 132.1 MB/s , 661.9 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 96.8 MB/s , 641.6 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 63.1 MB/s , 677.0 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.3 MB/s , 678.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 33.1 MB/s , 708.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.5 MB/s , 710.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 21.9 MB/s , 723.9 MB/s
2016-10-17 23:55:52 +00:00
if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
/* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
CHECK_E( FSE_buildCTable_wksp(bs->entropy.offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog, workspace, HUF_WORKSPACE_SIZE),
dictionary_corrupted);
dictPtr += offcodeHeaderSize;
2016-05-31 16:13:56 +00:00
}
{ short matchlengthNCount[MaxML+1];
Fix buffer overrun in ZSTD_loadDictEntropyStats() The table log set by `FSE_readNCount()` was not checked in `ZSTD_loadDictEntropyStats()`. This caused `FSE_buildCTable()` to stack/heap overflow in a few places. The benchmarks look good, there is no obvious compression performance regression: > ./zstds/zstd.opt.0 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 271.6 MB/s , 716.8 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 204.8 MB/s , 671.1 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 156.8 MB/s , 658.6 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 136.4 MB/s , 665.3 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 98.9 MB/s , 647.0 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 65.2 MB/s , 670.4 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.9 MB/s , 688.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 32.4 MB/s , 711.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.1 MB/s , 718.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 22.2 MB/s , 721.8 MB/s > ./compress_zstds/zstd.opt.1 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 273.8 MB/s , 722.0 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 203.2 MB/s , 666.6 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 157.4 MB/s , 666.5 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 132.1 MB/s , 661.9 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 96.8 MB/s , 641.6 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 63.1 MB/s , 677.0 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.3 MB/s , 678.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 33.1 MB/s , 708.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.5 MB/s , 710.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 21.9 MB/s , 723.9 MB/s
2016-10-17 23:55:52 +00:00
unsigned matchlengthMaxValue = MaxML, matchlengthLog;
size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
2016-05-31 16:13:56 +00:00
if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
Fix buffer overrun in ZSTD_loadDictEntropyStats() The table log set by `FSE_readNCount()` was not checked in `ZSTD_loadDictEntropyStats()`. This caused `FSE_buildCTable()` to stack/heap overflow in a few places. The benchmarks look good, there is no obvious compression performance regression: > ./zstds/zstd.opt.0 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 271.6 MB/s , 716.8 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 204.8 MB/s , 671.1 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 156.8 MB/s , 658.6 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 136.4 MB/s , 665.3 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 98.9 MB/s , 647.0 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 65.2 MB/s , 670.4 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.9 MB/s , 688.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 32.4 MB/s , 711.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.1 MB/s , 718.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 22.2 MB/s , 721.8 MB/s > ./compress_zstds/zstd.opt.1 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 273.8 MB/s , 722.0 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 203.2 MB/s , 666.6 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 157.4 MB/s , 666.5 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 132.1 MB/s , 661.9 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 96.8 MB/s , 641.6 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 63.1 MB/s , 677.0 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.3 MB/s , 678.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 33.1 MB/s , 708.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.5 MB/s , 710.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 21.9 MB/s , 723.9 MB/s
2016-10-17 23:55:52 +00:00
if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
/* Every match length code must have non-zero probability */
CHECK_F( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
CHECK_E( FSE_buildCTable_wksp(bs->entropy.matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, workspace, HUF_WORKSPACE_SIZE),
dictionary_corrupted);
dictPtr += matchlengthHeaderSize;
2016-05-31 16:13:56 +00:00
}
{ short litlengthNCount[MaxLL+1];
Fix buffer overrun in ZSTD_loadDictEntropyStats() The table log set by `FSE_readNCount()` was not checked in `ZSTD_loadDictEntropyStats()`. This caused `FSE_buildCTable()` to stack/heap overflow in a few places. The benchmarks look good, there is no obvious compression performance regression: > ./zstds/zstd.opt.0 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 271.6 MB/s , 716.8 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 204.8 MB/s , 671.1 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 156.8 MB/s , 658.6 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 136.4 MB/s , 665.3 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 98.9 MB/s , 647.0 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 65.2 MB/s , 670.4 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.9 MB/s , 688.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 32.4 MB/s , 711.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.1 MB/s , 718.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 22.2 MB/s , 721.8 MB/s > ./compress_zstds/zstd.opt.1 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 273.8 MB/s , 722.0 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 203.2 MB/s , 666.6 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 157.4 MB/s , 666.5 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 132.1 MB/s , 661.9 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 96.8 MB/s , 641.6 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 63.1 MB/s , 677.0 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.3 MB/s , 678.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 33.1 MB/s , 708.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.5 MB/s , 710.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 21.9 MB/s , 723.9 MB/s
2016-10-17 23:55:52 +00:00
unsigned litlengthMaxValue = MaxLL, litlengthLog;
size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
2016-05-31 16:13:56 +00:00
if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
Fix buffer overrun in ZSTD_loadDictEntropyStats() The table log set by `FSE_readNCount()` was not checked in `ZSTD_loadDictEntropyStats()`. This caused `FSE_buildCTable()` to stack/heap overflow in a few places. The benchmarks look good, there is no obvious compression performance regression: > ./zstds/zstd.opt.0 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 271.6 MB/s , 716.8 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 204.8 MB/s , 671.1 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 156.8 MB/s , 658.6 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 136.4 MB/s , 665.3 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 98.9 MB/s , 647.0 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 65.2 MB/s , 670.4 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.9 MB/s , 688.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 32.4 MB/s , 711.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.1 MB/s , 718.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 22.2 MB/s , 721.8 MB/s > ./compress_zstds/zstd.opt.1 -i10 -b1 -e10 ~/bench/silesia.tar 1#silesia.tar : 211988480 -> 73656930 (2.878), 273.8 MB/s , 722.0 MB/s 2#silesia.tar : 211988480 -> 70162842 (3.021), 203.2 MB/s , 666.6 MB/s 3#silesia.tar : 211988480 -> 66997986 (3.164), 157.4 MB/s , 666.5 MB/s 4#silesia.tar : 211988480 -> 66002591 (3.212), 132.1 MB/s , 661.9 MB/s 5#silesia.tar : 211988480 -> 65008480 (3.261), 96.8 MB/s , 641.6 MB/s 6#silesia.tar : 211988480 -> 62979643 (3.366), 63.1 MB/s , 677.0 MB/s 7#silesia.tar : 211988480 -> 61974560 (3.421), 44.3 MB/s , 678.2 MB/s 8#silesia.tar : 211988480 -> 61028308 (3.474), 33.1 MB/s , 708.9 MB/s 9#silesia.tar : 211988480 -> 60416751 (3.509), 21.5 MB/s , 710.1 MB/s 10#silesia.tar : 211988480 -> 60174239 (3.523), 21.9 MB/s , 723.9 MB/s
2016-10-17 23:55:52 +00:00
if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
/* Every literal length code must have non-zero probability */
CHECK_F( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
CHECK_E( FSE_buildCTable_wksp(bs->entropy.litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, workspace, HUF_WORKSPACE_SIZE),
dictionary_corrupted);
dictPtr += litlengthHeaderSize;
2016-05-31 16:13:56 +00:00
}
if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
bs->rep[0] = MEM_readLE32(dictPtr+0);
bs->rep[1] = MEM_readLE32(dictPtr+4);
bs->rep[2] = MEM_readLE32(dictPtr+8);
dictPtr += 12;
{ size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
U32 offcodeMax = MaxOff;
if (dictContentSize <= ((U32)-1) - 128 KB) {
U32 const maxOffset = (U32)dictContentSize + 128 KB; /* The maximum offset that must be supported */
offcodeMax = ZSTD_highbit32(maxOffset); /* Calculate minimum offset code required to represent maxOffset */
2016-10-24 21:11:27 +00:00
}
/* All offset values <= dictContentSize + 128 KB must be representable */
CHECK_F (ZSTD_checkDictNCount(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff)));
/* All repCodes must be <= dictContentSize and != 0*/
{ U32 u;
for (u=0; u<3; u++) {
if (bs->rep[u] == 0) return ERROR(dictionary_corrupted);
if (bs->rep[u] > dictContentSize) return ERROR(dictionary_corrupted);
} }
bs->entropy.hufCTable_repeatMode = HUF_repeat_valid;
bs->entropy.offcode_repeatMode = FSE_repeat_valid;
bs->entropy.matchlength_repeatMode = FSE_repeat_valid;
bs->entropy.litlength_repeatMode = FSE_repeat_valid;
CHECK_F(ZSTD_loadDictionaryContent(ms, params, dictPtr, dictContentSize));
return dictID;
}
2016-01-26 02:14:20 +00:00
}
2016-03-15 00:24:33 +00:00
/** ZSTD_compress_insertDictionary() :
* @return : dictID, or an error code */
static size_t ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matchState_t* ms,
ZSTD_CCtx_params const* params,
const void* dict, size_t dictSize,
ZSTD_dictMode_e dictMode,
void* workspace)
2016-01-26 02:14:20 +00:00
{
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "ZSTD_compress_insertDictionary (dictSize=%u)", (U32)dictSize);
2016-05-29 03:01:04 +00:00
if ((dict==NULL) || (dictSize<=8)) return 0;
2016-01-26 14:58:49 +00:00
ZSTD_reset_compressedBlockState(bs);
/* dict restricted modes */
if (dictMode==ZSTD_dm_rawContent)
return ZSTD_loadDictionaryContent(ms, params, dict, dictSize);
2016-03-15 00:24:33 +00:00
if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
if (dictMode == ZSTD_dm_auto) {
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "raw content dictionary detected");
return ZSTD_loadDictionaryContent(ms, params, dict, dictSize);
}
if (dictMode == ZSTD_dm_fullDict)
return ERROR(dictionary_wrong);
assert(0); /* impossible */
}
/* dict as full zstd dictionary */
return ZSTD_loadZstdDictionary(bs, ms, params, dict, dictSize, workspace);
2016-01-07 14:35:18 +00:00
}
2016-04-01 13:48:48 +00:00
/*! ZSTD_compressBegin_internal() :
* @return : 0, or an error code */
size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
2016-01-26 15:31:22 +00:00
const void* dict, size_t dictSize,
ZSTD_dictMode_e dictMode,
const ZSTD_CDict* cdict,
ZSTD_CCtx_params params, U64 pledgedSrcSize,
ZSTD_buffered_policy_e zbuff)
2015-10-22 14:31:46 +00:00
{
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
DEBUGLOG(4, "ZSTD_compressBegin_internal: wlog=%u", params.cParams.windowLog);
/* params are supposed to be fully validated at this point */
assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
assert(!((dict) && (cdict))); /* either dict or cdict, not both */
if (cdict && cdict->dictContentSize>0) {
cctx->requestedParams = params;
return ZSTD_resetCCtx_usingCDict(cctx, cdict, params.cParams.windowLog,
params.fParams, pledgedSrcSize, zbuff);
}
CHECK_F( ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
ZSTDcrp_continue, zbuff) );
{
size_t const dictID = ZSTD_compress_insertDictionary(
cctx->blockState.prevCBlock, &cctx->blockState.matchState,
&params, dict, dictSize, dictMode, cctx->entropyWorkspace);
if (ZSTD_isError(dictID)) return dictID;
assert(dictID <= (size_t)(U32)-1);
cctx->dictID = (U32)dictID;
}
return 0;
2015-10-22 14:31:46 +00:00
}
size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
2017-08-18 23:17:24 +00:00
const void* dict, size_t dictSize,
ZSTD_dictMode_e dictMode,
const ZSTD_CDict* cdict,
2017-08-18 23:17:24 +00:00
ZSTD_CCtx_params params,
unsigned long long pledgedSrcSize)
{
fix a subtle issue in continue mode The deep fuzzer tests caught a subtle bug that was probably there for a long time. The impact of the bug is not a crash, or any other clear error signal, rather, it reduces performance, by cutting data into smaller blocks. Eventually, the following test would fail because it produces too many 1-byte blocks, requiring more space than buffer can provide : `./zstreamtest_asan --mt -s3514 -t1678312 -i1678314` The root scenario is as follows : - Create context, initialize it using explicit parameters or a `cdict` to pin them down, set `pledgedSrcSize=1` - The compression parameters will not be adapted, but `windowSize` and `blockSize` will be automatically set to `1`. `windowSize` and `blockSize` are dynamic values, set within `ZSTD_resetCCtx_internal()`. The automatic adaptation makes it possible to generate smaller contexts for smaller input sizes. - Complete compression - New compression with same context, using same parameters, but `pledgedSrcSize=ZSTD_CONTENTSIZE_UNKNOWN` trigger "continue mode" - Continue mode doesn't modify blockSize, because it used to depend on `windowLog` only, but in fact, it also depends on `pledgedSrcSize`. - The "old" blocksize (1) is still there, next compression will use this value to cut input into blocks, resulting in more blocks and worse performance than necessary performance. Given the scenario, and its possible variants, I'm surprised it did not show up before. But I suspect it did show up, it's just that it never triggered an error, because "worse performance" is not a trigger. The above test is a special corner case, where performance is so impacted that it reaches an error case. The fix works, but I'm not completely pleased. I think the current code relies too much on implied relations between variables. This will likely break again in the future when some related part of the code change. Unfortunately, no time to make larger changes if we want to keep the release target for zstd v1.3.3. So a longer term fix will have to be considered after the release. To do : create a reliable test case which triggers this scenario for CI tests.
2017-12-19 08:43:03 +00:00
DEBUGLOG(4, "ZSTD_compressBegin_advanced_internal: wlog=%u", params.cParams.windowLog);
2017-08-18 23:17:24 +00:00
/* compression parameters verification and optimization */
CHECK_F( ZSTD_checkCParams(params.cParams) );
return ZSTD_compressBegin_internal(cctx,
dict, dictSize, dictMode,
cdict,
2017-08-18 23:17:24 +00:00
params, pledgedSrcSize,
ZSTDb_not_buffered);
}
2015-10-25 13:06:35 +00:00
2016-04-01 13:48:48 +00:00
/*! ZSTD_compressBegin_advanced() :
* @return : 0, or an error code */
size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
2016-04-01 13:48:48 +00:00
const void* dict, size_t dictSize,
ZSTD_parameters params, unsigned long long pledgedSrcSize)
2016-04-01 13:48:48 +00:00
{
ZSTD_CCtx_params const cctxParams =
ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
return ZSTD_compressBegin_advanced_internal(cctx,
dict, dictSize, ZSTD_dm_auto,
NULL /*cdict*/,
cctxParams, pledgedSrcSize);
2016-04-01 13:48:48 +00:00
}
size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
2016-01-26 02:14:20 +00:00
{
ZSTD_parameters const params = ZSTD_getParams(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
ZSTD_CCtx_params const cctxParams =
2017-08-23 19:30:47 +00:00
ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (U32)dictSize);
return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dm_auto, NULL,
cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, ZSTDb_not_buffered);
2016-01-26 15:31:22 +00:00
}
2015-11-25 13:42:45 +00:00
size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel)
2015-11-25 13:42:45 +00:00
{
return ZSTD_compressBegin_usingDict(cctx, NULL, 0, compressionLevel);
2015-10-25 13:06:35 +00:00
}
2016-07-28 13:29:08 +00:00
/*! ZSTD_writeEpilogue() :
* Ends a frame.
2015-11-25 13:42:45 +00:00
* @return : nb of bytes written into dst (or an error code) */
2016-07-28 13:29:08 +00:00
static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
{
2016-07-27 22:55:43 +00:00
BYTE* const ostart = (BYTE*)dst;
BYTE* op = ostart;
2016-04-12 13:52:33 +00:00
size_t fhSize = 0;
DEBUGLOG(4, "ZSTD_writeEpilogue");
if (cctx->stage == ZSTDcs_created) return ERROR(stage_wrong); /* init missing */
/* special case : empty frame */
2016-07-27 22:55:43 +00:00
if (cctx->stage == ZSTDcs_init) {
fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->appliedParams, 0, 0);
2016-04-12 13:52:33 +00:00
if (ZSTD_isError(fhSize)) return fhSize;
dstCapacity -= fhSize;
op += fhSize;
2016-07-27 19:05:12 +00:00
cctx->stage = ZSTDcs_ongoing;
2016-01-07 14:35:18 +00:00
}
2016-07-27 22:55:43 +00:00
if (cctx->stage != ZSTDcs_ending) {
/* write one last empty block, make it the "last" block */
U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1) + 0;
if (dstCapacity<4) return ERROR(dstSize_tooSmall);
MEM_writeLE32(op, cBlockHeader24);
op += ZSTD_blockHeaderSize;
dstCapacity -= ZSTD_blockHeaderSize;
}
if (cctx->appliedParams.fParams.checksumFlag) {
2016-07-27 22:55:43 +00:00
U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
if (dstCapacity<4) return ERROR(dstSize_tooSmall);
DEBUGLOG(4, "ZSTD_writeEpilogue: write checksum : %08X", checksum);
2016-07-27 22:55:43 +00:00
MEM_writeLE32(op, checksum);
op += 4;
2016-05-31 16:13:56 +00:00
}
2016-07-27 19:05:12 +00:00
cctx->stage = ZSTDcs_created; /* return to "created but no init" status */
2016-07-27 22:55:43 +00:00
return op-ostart;
}
2016-07-28 13:29:08 +00:00
size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize)
{
size_t endResult;
size_t const cSize = ZSTD_compressContinue_internal(cctx,
dst, dstCapacity, src, srcSize,
1 /* frame mode */, 1 /* last chunk */);
2016-07-28 13:29:08 +00:00
if (ZSTD_isError(cSize)) return cSize;
endResult = ZSTD_writeEpilogue(cctx, (char*)dst + cSize, dstCapacity-cSize);
if (ZSTD_isError(endResult)) return endResult;
if (cctx->appliedParams.fParams.contentSizeFlag) { /* control src size */
DEBUGLOG(4, "end of frame : controlling src size");
if (cctx->pledgedSrcSizePlusOne != cctx->consumedSrcSize+1) {
DEBUGLOG(4, "error : pledgedSrcSize = %u, while realSrcSize = %u",
(U32)cctx->pledgedSrcSizePlusOne-1, (U32)cctx->consumedSrcSize);
return ERROR(srcSize_wrong);
} }
2016-07-28 13:29:08 +00:00
return cSize + endResult;
}
2016-07-27 23:25:46 +00:00
static size_t ZSTD_compress_internal (ZSTD_CCtx* cctx,
2016-03-15 00:24:33 +00:00
void* dst, size_t dstCapacity,
2015-11-25 13:42:45 +00:00
const void* src, size_t srcSize,
2015-12-18 00:26:48 +00:00
const void* dict,size_t dictSize,
2015-11-25 13:42:45 +00:00
ZSTD_parameters params)
2015-10-22 14:31:46 +00:00
{
ZSTD_CCtx_params const cctxParams =
ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "ZSTD_compress_internal");
return ZSTD_compress_advanced_internal(cctx,
dst, dstCapacity,
src, srcSize,
dict, dictSize,
cctxParams);
2015-10-22 14:31:46 +00:00
}
size_t ZSTD_compress_advanced (ZSTD_CCtx* ctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const void* dict,size_t dictSize,
ZSTD_parameters params)
{
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "ZSTD_compress_advanced");
CHECK_F(ZSTD_checkCParams(params.cParams));
return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
}
2017-08-18 23:17:24 +00:00
/* Internal */
size_t ZSTD_compress_advanced_internal(
2017-08-21 05:55:07 +00:00
ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const void* dict,size_t dictSize,
ZSTD_CCtx_params params)
2017-08-18 23:17:24 +00:00
{
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "ZSTD_compress_advanced_internal");
CHECK_F( ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dm_auto, NULL,
2017-08-18 23:17:24 +00:00
params, srcSize, ZSTDb_not_buffered) );
return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
}
size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize,
const void* dict, size_t dictSize, int compressionLevel)
2015-12-18 00:26:48 +00:00
{
ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize ? srcSize : 1, dict ? dictSize : 0);
params.fParams.contentSizeFlag = 1;
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "ZSTD_compress_usingDict (level=%i, srcSize=%u, dictSize=%u)",
compressionLevel, (U32)srcSize, (U32)dictSize);
return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
2015-12-18 00:26:48 +00:00
}
2016-03-15 00:24:33 +00:00
size_t ZSTD_compressCCtx (ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
2015-10-25 13:06:35 +00:00
{
DEBUGLOG(4, "ZSTD_compressCCtx (srcSize=%u)", (U32)srcSize);
return ZSTD_compress_usingDict(ctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
2015-10-25 13:06:35 +00:00
}
2016-03-15 00:24:33 +00:00
size_t ZSTD_compress(void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
2015-10-22 14:31:46 +00:00
{
2015-10-29 21:02:40 +00:00
size_t result;
ZSTD_CCtx ctxBody;
2015-10-29 17:41:45 +00:00
memset(&ctxBody, 0, sizeof(ctxBody));
ctxBody.customMem = ZSTD_defaultCMem;
2016-03-15 00:24:33 +00:00
result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
ZSTD_free(ctxBody.workSpace, ZSTD_defaultCMem); /* can't free ctxBody itself, as it's on stack; free only heap content */
2015-10-29 21:02:40 +00:00
return result;
2015-10-22 14:31:46 +00:00
}
2015-12-17 22:50:15 +00:00
/* ===== Dictionary API ===== */
/*! ZSTD_estimateCDictSize_advanced() :
* Estimate amount of memory that will be needed to create a dictionary with following arguments */
2017-08-29 18:55:02 +00:00
size_t ZSTD_estimateCDictSize_advanced(
size_t dictSize, ZSTD_compressionParameters cParams,
ZSTD_dictLoadMethod_e dictLoadMethod)
{
DEBUGLOG(5, "sizeof(ZSTD_CDict) : %u", (U32)sizeof(ZSTD_CDict));
return sizeof(ZSTD_CDict) + HUF_WORKSPACE_SIZE + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0)
2017-08-29 18:55:02 +00:00
+ (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
}
size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel)
{
ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
return ZSTD_estimateCDictSize_advanced(dictSize, cParams, ZSTD_dlm_byCopy);
}
size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict)
{
if (cdict==NULL) return 0; /* support sizeof on NULL */
DEBUGLOG(5, "sizeof(*cdict) : %u", (U32)sizeof(*cdict));
return cdict->workspaceSize + (cdict->dictBuffer ? cdict->dictContentSize : 0) + sizeof(*cdict);
}
2017-05-26 01:05:49 +00:00
static size_t ZSTD_initCDict_internal(
ZSTD_CDict* cdict,
const void* dictBuffer, size_t dictSize,
2017-08-29 18:55:02 +00:00
ZSTD_dictLoadMethod_e dictLoadMethod,
ZSTD_dictMode_e dictMode,
2017-05-26 01:05:49 +00:00
ZSTD_compressionParameters cParams)
{
DEBUGLOG(3, "ZSTD_initCDict_internal, mode %u", (U32)dictMode);
cdict->cParams = cParams;
2017-08-29 18:55:02 +00:00
if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dictBuffer) || (!dictSize)) {
2017-05-26 01:05:49 +00:00
cdict->dictBuffer = NULL;
cdict->dictContent = dictBuffer;
} else {
void* const internalBuffer = ZSTD_malloc(dictSize, cdict->customMem);
2017-05-26 01:05:49 +00:00
cdict->dictBuffer = internalBuffer;
cdict->dictContent = internalBuffer;
if (!internalBuffer) return ERROR(memory_allocation);
memcpy(internalBuffer, dictBuffer, dictSize);
2017-05-26 01:05:49 +00:00
}
cdict->dictContentSize = dictSize;
/* Reset the state to no dictionary */
ZSTD_reset_compressedBlockState(&cdict->cBlockState);
{
void* const end = ZSTD_reset_matchState(
&cdict->matchState,
(U32*)cdict->workspace + HUF_WORKSPACE_SIZE_U32,
&cParams, ZSTDcrp_continue, /* forCCtx */ 0);
assert(end == (char*)cdict->workspace + cdict->workspaceSize);
(void)end;
}
/* (Maybe) load the dictionary
* Skips loading the dictionary if it is <= 8 bytes.
*/
{
ZSTD_CCtx_params params;
memset(&params, 0, sizeof(params));
params.compressionLevel = ZSTD_CLEVEL_DEFAULT;
params.fParams.contentSizeFlag = 1;
params.cParams = cParams;
{
size_t const dictID = ZSTD_compress_insertDictionary(
&cdict->cBlockState, &cdict->matchState,&params,
cdict->dictContent, cdict->dictContentSize,
dictMode, cdict->workspace);
if (ZSTD_isError(dictID)) return dictID;
assert(dictID <= (size_t)(U32)-1);
cdict->dictID = (U32)dictID;
}
2017-05-26 01:05:49 +00:00
}
return 0;
}
ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
2017-08-29 18:55:02 +00:00
ZSTD_dictLoadMethod_e dictLoadMethod,
ZSTD_dictMode_e dictMode,
ZSTD_compressionParameters cParams, ZSTD_customMem customMem)
{
DEBUGLOG(3, "ZSTD_createCDict_advanced, mode %u", (U32)dictMode);
if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
{ ZSTD_CDict* const cdict = (ZSTD_CDict*)ZSTD_malloc(sizeof(ZSTD_CDict), customMem);
size_t const workspaceSize = HUF_WORKSPACE_SIZE + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0);
void* const workspace = ZSTD_malloc(workspaceSize, customMem);
if (!cdict || !workspace) {
ZSTD_free(cdict, customMem);
ZSTD_free(workspace, customMem);
return NULL;
}
cdict->customMem = customMem;
cdict->workspace = workspace;
cdict->workspaceSize = workspaceSize;
2017-05-26 01:05:49 +00:00
if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
dictBuffer, dictSize,
2017-08-29 18:55:02 +00:00
dictLoadMethod, dictMode,
2017-05-26 01:05:49 +00:00
cParams) )) {
ZSTD_freeCDict(cdict);
return NULL;
}
2016-12-21 15:20:11 +00:00
return cdict;
}
}
ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel)
{
ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
return ZSTD_createCDict_advanced(dict, dictSize,
2017-08-29 18:55:02 +00:00
ZSTD_dlm_byCopy, ZSTD_dm_auto,
cParams, ZSTD_defaultCMem);
2016-12-21 15:20:11 +00:00
}
ZSTD_CDict* ZSTD_createCDict_byReference(const void* dict, size_t dictSize, int compressionLevel)
{
ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
return ZSTD_createCDict_advanced(dict, dictSize,
2017-08-29 18:55:02 +00:00
ZSTD_dlm_byRef, ZSTD_dm_auto,
cParams, ZSTD_defaultCMem);
}
size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
{
if (cdict==NULL) return 0; /* support free on NULL */
{ ZSTD_customMem const cMem = cdict->customMem;
ZSTD_free(cdict->workspace, cMem);
2016-12-21 15:44:35 +00:00
ZSTD_free(cdict->dictBuffer, cMem);
ZSTD_free(cdict, cMem);
return 0;
}
}
2017-05-26 01:05:49 +00:00
/*! ZSTD_initStaticCDict_advanced() :
* Generate a digested dictionary in provided memory area.
* workspace: The memory area to emplace the dictionary into.
* Provided pointer must 8-bytes aligned.
* It must outlive dictionary usage.
* workspaceSize: Use ZSTD_estimateCDictSize()
* to determine how large workspace must be.
* cParams : use ZSTD_getCParams() to transform a compression level
* into its relevants cParams.
* @return : pointer to ZSTD_CDict*, or NULL if error (size too small)
* Note : there is no corresponding "free" function.
* Since workspace was allocated externally, it must be freed externally.
*/
const ZSTD_CDict* ZSTD_initStaticCDict(
void* workspace, size_t workspaceSize,
const void* dict, size_t dictSize,
2017-08-29 18:55:02 +00:00
ZSTD_dictLoadMethod_e dictLoadMethod,
ZSTD_dictMode_e dictMode,
2017-05-26 01:05:49 +00:00
ZSTD_compressionParameters cParams)
{
size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0);
2017-08-29 18:55:02 +00:00
size_t const neededSize = sizeof(ZSTD_CDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize)
+ HUF_WORKSPACE_SIZE + matchStateSize;
2017-05-26 01:05:49 +00:00
ZSTD_CDict* const cdict = (ZSTD_CDict*) workspace;
void* ptr;
if ((size_t)workspace & 7) return NULL; /* 8-aligned */
DEBUGLOG(4, "(workspaceSize < neededSize) : (%u < %u) => %u",
2017-05-26 01:05:49 +00:00
(U32)workspaceSize, (U32)neededSize, (U32)(workspaceSize < neededSize));
if (workspaceSize < neededSize) return NULL;
2017-08-29 18:55:02 +00:00
if (dictLoadMethod == ZSTD_dlm_byCopy) {
2017-05-26 01:05:49 +00:00
memcpy(cdict+1, dict, dictSize);
dict = cdict+1;
ptr = (char*)workspace + sizeof(ZSTD_CDict) + dictSize;
} else {
ptr = cdict+1;
}
cdict->workspace = ptr;
cdict->workspaceSize = HUF_WORKSPACE_SIZE + matchStateSize;
2017-05-26 01:05:49 +00:00
if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
dict, dictSize,
2017-08-29 18:55:02 +00:00
ZSTD_dlm_byRef, dictMode,
cParams) ))
2017-05-26 01:05:49 +00:00
return NULL;
return cdict;
}
ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict)
{
assert(cdict != NULL);
return cdict->cParams;
}
/* ZSTD_compressBegin_usingCDict_advanced() :
* cdict must be != NULL */
size_t ZSTD_compressBegin_usingCDict_advanced(
ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict,
ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize)
2016-09-15 12:54:07 +00:00
{
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "ZSTD_compressBegin_usingCDict_advanced");
if (cdict==NULL) return ERROR(dictionary_wrong);
2017-08-26 00:58:28 +00:00
{ ZSTD_CCtx_params params = cctx->requestedParams;
params.cParams = ZSTD_getCParamsFromCDict(cdict);
/* Increase window log to fit the entire dictionary and source if the
* source size is known. Limit the increase to 19, which is the
* window log for compression level 1 with the largest source size.
*/
if (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN) {
U32 const limitedSrcSize = (U32)MIN(pledgedSrcSize, 1U << 19);
U32 const limitedSrcLog = limitedSrcSize > 1 ? ZSTD_highbit32(limitedSrcSize - 1) + 1 : 1;
params.cParams.windowLog = MAX(params.cParams.windowLog, limitedSrcLog);
}
params.fParams = fParams;
return ZSTD_compressBegin_internal(cctx,
NULL, 0, ZSTD_dm_auto,
cdict,
params, pledgedSrcSize,
ZSTDb_not_buffered);
}
2016-09-15 12:54:07 +00:00
}
/* ZSTD_compressBegin_usingCDict() :
* pledgedSrcSize=0 means "unknown"
* if pledgedSrcSize>0, it will enable contentSizeFlag */
size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
{
ZSTD_frameParameters const fParams = { 0 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
DEBUGLOG(4, "ZSTD_compressBegin_usingCDict : dictIDFlag == %u", !fParams.noDictIDFlag);
return ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, 0);
}
size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
{
CHECK_F (ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, srcSize)); /* will check if cdict != NULL */
return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
}
/*! ZSTD_compress_usingCDict() :
* Compression using a digested Dictionary.
* Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
* Note that compression parameters are decided at CDict creation time
* while frame parameters are hardcoded */
2016-09-15 12:54:07 +00:00
size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const ZSTD_CDict* cdict)
{
ZSTD_frameParameters const fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, fParams);
}
2016-08-12 11:04:27 +00:00
/* ******************************************************************
* Streaming
********************************************************************/
2016-08-11 23:20:36 +00:00
ZSTD_CStream* ZSTD_createCStream(void)
{
DEBUGLOG(3, "ZSTD_createCStream");
return ZSTD_createCStream_advanced(ZSTD_defaultCMem);
2016-08-11 23:20:36 +00:00
}
ZSTD_CStream* ZSTD_initStaticCStream(void *workspace, size_t workspaceSize)
{
return ZSTD_initStaticCCtx(workspace, workspaceSize);
}
2016-08-11 23:20:36 +00:00
ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem)
{ /* CStream and CCtx are now same object */
return ZSTD_createCCtx_advanced(customMem);
2016-08-11 23:20:36 +00:00
}
size_t ZSTD_freeCStream(ZSTD_CStream* zcs)
{
return ZSTD_freeCCtx(zcs); /* same object */
2016-08-11 23:20:36 +00:00
}
2016-08-12 11:04:27 +00:00
/*====== Initialization ======*/
size_t ZSTD_CStreamInSize(void) { return ZSTD_BLOCKSIZE_MAX; }
2016-08-11 23:20:36 +00:00
size_t ZSTD_CStreamOutSize(void)
{
return ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + 4 /* 32-bits hash */ ;
}
2016-08-11 23:20:36 +00:00
static size_t ZSTD_resetCStream_internal(ZSTD_CStream* zcs,
const void* const dict, size_t const dictSize, ZSTD_dictMode_e const dictMode,
const ZSTD_CDict* const cdict,
ZSTD_CCtx_params const params, unsigned long long const pledgedSrcSize)
2016-09-15 12:54:07 +00:00
{
DEBUGLOG(4, "ZSTD_resetCStream_internal");
2017-06-27 22:49:12 +00:00
/* params are supposed to be fully validated at this point */
assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
assert(!((dict) && (cdict))); /* either dict or cdict, not both */
CHECK_F( ZSTD_compressBegin_internal(zcs,
dict, dictSize, dictMode,
cdict,
params, pledgedSrcSize,
ZSTDb_buffered) );
2016-09-15 12:54:07 +00:00
zcs->inToCompress = 0;
zcs->inBuffPos = 0;
saves 3-bytes on small input with streaming API zstd streaming API was adding a null-block at end of frame for small input. Reason is : on small input, a single block is enough. ZSTD_CStream would size its input buffer to expect a single block of this size, automatically triggering a flush on reaching this size. Unfortunately, that last byte was generally received before the "end" directive (at least in `fileio`). The later "end" directive would force the creation of a 3-bytes last block to indicate end of frame. The solution is to not flush automatically, which is btw the expected behavior. It happens in this case because blocksize is defined with exactly the same size as input. Just adding one-byte is enough to stop triggering the automatic flush. I initially looked at another solution, solving the problem directly in the compression context. But it felt awkward. Now, the underlying compression API `ZSTD_compressContinue()` would take the decision the close a frame on reaching its expected end (`pledgedSrcSize`). This feels awkward, a responsability over-reach, beyond the definition of this API. ZSTD_compressContinue() is clearly documented as a guaranteed flush, with ZSTD_compressEnd() generating a guaranteed end. I faced similar issue when trying to port a similar mechanism at the higher streaming layer. Having ZSTD_CStream end a frame automatically on reaching `pledgedSrcSize` can surprise the caller, since it did not explicitly requested an end of frame. The only sensible action remaining after that is to end the frame with no additional input. This adds additional logic in the ZSTD_CStream state to check this condition. Plus some potential confusion on the meaning of ZSTD_endStream() with no additional input (ending confirmation ? new 0-size frame ?) In the end, just enlarging input buffer by 1 byte feels the least intrusive change. It's also a contract remaining inside the streaming layer, so the logic is contained in this part of the code. The patch also introduces a new test checking that size of small frame is as expected, without additional 3-bytes null block.
2017-12-14 19:47:02 +00:00
zcs->inBuffTarget = zcs->blockSize
+ (zcs->blockSize == pledgedSrcSize); /* for small input: avoid automatic flush on reaching end of block, since it would require to add a 3-bytes null block to end frame */
2016-09-15 12:54:07 +00:00
zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
zcs->streamStage = zcss_load;
2016-09-15 12:54:07 +00:00
zcs->frameEnded = 0;
return 0; /* ready to go */
}
/* ZSTD_resetCStream():
* pledgedSrcSize == 0 means "unknown" */
size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize)
{
ZSTD_CCtx_params params = zcs->requestedParams;
DEBUGLOG(4, "ZSTD_resetCStream: pledgedSrcSize = %u", (U32)pledgedSrcSize);
if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
params.fParams.contentSizeFlag = 1;
2017-08-29 02:25:17 +00:00
params.cParams = ZSTD_getCParamsFromCCtxParams(params, pledgedSrcSize, 0);
return ZSTD_resetCStream_internal(zcs, NULL, 0, ZSTD_dm_auto, zcs->cdict, params, pledgedSrcSize);
}
/*! ZSTD_initCStream_internal() :
* Note : for lib/compress only. Used by zstdmt_compress.c.
* Assumption 1 : params are valid
* Assumption 2 : either dict, or cdict, is defined, not both */
size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
const void* dict, size_t dictSize, const ZSTD_CDict* cdict,
ZSTD_CCtx_params params, unsigned long long pledgedSrcSize)
2016-08-11 23:20:36 +00:00
{
DEBUGLOG(4, "ZSTD_initCStream_internal");
assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
assert(!((dict) && (cdict))); /* either dict or cdict, not both */
2016-08-11 23:20:36 +00:00
if (dict && dictSize >= 8) {
DEBUGLOG(4, "loading dictionary of size %u", (U32)dictSize);
if (zcs->staticSize) { /* static CCtx : never uses malloc */
/* incompatible with internal cdict creation */
return ERROR(memory_allocation);
2016-08-11 23:20:36 +00:00
}
ZSTD_freeCDict(zcs->cdictLocal);
zcs->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
ZSTD_dlm_byCopy, ZSTD_dm_auto,
params.cParams, zcs->customMem);
zcs->cdict = zcs->cdictLocal;
if (zcs->cdictLocal == NULL) return ERROR(memory_allocation);
} else {
if (cdict) {
2017-12-13 23:35:49 +00:00
params.cParams = ZSTD_getCParamsFromCDict(cdict); /* cParams are enforced from cdict; it includes windowLog */
}
ZSTD_freeCDict(zcs->cdictLocal);
zcs->cdictLocal = NULL;
zcs->cdict = cdict;
2016-08-11 23:20:36 +00:00
}
params.compressionLevel = ZSTD_CLEVEL_CUSTOM; /* enforce usage of cParams, instead of a dynamic derivation from cLevel (but does that happen ?) */
zcs->requestedParams = params;
2017-08-18 18:43:31 +00:00
return ZSTD_resetCStream_internal(zcs, NULL, 0, ZSTD_dm_auto, zcs->cdict, params, pledgedSrcSize);
}
/* ZSTD_initCStream_usingCDict_advanced() :
* same as ZSTD_initCStream_usingCDict(), with control over frame parameters */
size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
const ZSTD_CDict* cdict,
ZSTD_frameParameters fParams,
unsigned long long pledgedSrcSize)
{
DEBUGLOG(4, "ZSTD_initCStream_usingCDict_advanced");
if (!cdict) return ERROR(dictionary_wrong); /* cannot handle NULL cdict (does not know what to do) */
2017-08-26 00:58:28 +00:00
{ ZSTD_CCtx_params params = zcs->requestedParams;
params.cParams = ZSTD_getCParamsFromCDict(cdict);
params.fParams = fParams;
return ZSTD_initCStream_internal(zcs,
NULL, 0, cdict,
params, pledgedSrcSize);
2016-08-11 23:20:36 +00:00
}
}
/* note : cdict must outlive compression session */
size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict)
{
ZSTD_frameParameters const fParams = { 0 /* contentSizeFlag */, 0 /* checksum */, 0 /* hideDictID */ };
DEBUGLOG(4, "ZSTD_initCStream_usingCDict");
return ZSTD_initCStream_usingCDict_advanced(zcs, cdict, fParams, ZSTD_CONTENTSIZE_UNKNOWN); /* note : will check that cdict != NULL */
2016-08-11 23:20:36 +00:00
}
/* ZSTD_initCStream_advanced() :
2017-12-08 17:16:49 +00:00
* pledgedSrcSize must be correct.
* if srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN.
* dict is loaded with default parameters ZSTD_dm_auto and ZSTD_dlm_byCopy. */
size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
const void* dict, size_t dictSize,
ZSTD_parameters params, unsigned long long pledgedSrcSize)
{
ZSTD_CCtx_params const cctxParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
DEBUGLOG(4, "ZSTD_initCStream_advanced: pledgedSrcSize=%u, flag=%u",
(U32)pledgedSrcSize, params.fParams.contentSizeFlag);
CHECK_F( ZSTD_checkCParams(params.cParams) );
2017-12-08 17:16:49 +00:00
if ((pledgedSrcSize==0) && (params.fParams.contentSizeFlag==0)) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN; /* for compatibility with older programs relying on this behavior. Users should now specify ZSTD_CONTENTSIZE_UNKNOWN. This line will be removed in the future. */
return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL /*cdict*/, cctxParams, pledgedSrcSize);
}
2016-08-11 23:20:36 +00:00
size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel)
{
ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
ZSTD_CCtx_params const cctxParams =
ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL, cctxParams, ZSTD_CONTENTSIZE_UNKNOWN);
2016-08-11 23:20:36 +00:00
}
size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pss)
{
U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss; /* temporary : 0 interpreted as "unknown" during transition period. Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN. `0` will be interpreted as "empty" in the future */
ZSTD_parameters const params = ZSTD_getParams(compressionLevel, pledgedSrcSize, 0);
ZSTD_CCtx_params const cctxParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
2017-08-23 19:30:47 +00:00
return ZSTD_initCStream_internal(zcs, NULL, 0, NULL, cctxParams, pledgedSrcSize);
}
2016-08-11 23:20:36 +00:00
size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
{
DEBUGLOG(4, "ZSTD_initCStream");
return ZSTD_initCStream_srcSize(zcs, compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN);
2016-08-22 22:30:31 +00:00
}
2016-08-11 23:20:36 +00:00
2016-08-12 11:04:27 +00:00
/*====== Compression ======*/
2016-08-11 23:20:36 +00:00
MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity,
const void* src, size_t srcSize)
2016-08-11 23:20:36 +00:00
{
size_t const length = MIN(dstCapacity, srcSize);
2017-05-31 16:59:22 +00:00
if (length) memcpy(dst, src, length);
2016-08-11 23:20:36 +00:00
return length;
}
/** ZSTD_compressStream_generic():
* internal function for all *compressStream*() variants and *compress_generic()
* non-static, because can be called from zstdmt.c
* @return : hint size for next input */
size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
ZSTD_outBuffer* output,
ZSTD_inBuffer* input,
ZSTD_EndDirective const flushMode)
2016-08-11 23:20:36 +00:00
{
const char* const istart = (const char*)input->src;
const char* const iend = istart + input->size;
const char* ip = istart + input->pos;
char* const ostart = (char*)output->dst;
char* const oend = ostart + output->size;
char* op = ostart + output->pos;
2016-08-11 23:20:36 +00:00
U32 someMoreWork = 1;
/* check expectations */
DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%u", (U32)flushMode);
assert(zcs->inBuff != NULL);
saves 3-bytes on small input with streaming API zstd streaming API was adding a null-block at end of frame for small input. Reason is : on small input, a single block is enough. ZSTD_CStream would size its input buffer to expect a single block of this size, automatically triggering a flush on reaching this size. Unfortunately, that last byte was generally received before the "end" directive (at least in `fileio`). The later "end" directive would force the creation of a 3-bytes last block to indicate end of frame. The solution is to not flush automatically, which is btw the expected behavior. It happens in this case because blocksize is defined with exactly the same size as input. Just adding one-byte is enough to stop triggering the automatic flush. I initially looked at another solution, solving the problem directly in the compression context. But it felt awkward. Now, the underlying compression API `ZSTD_compressContinue()` would take the decision the close a frame on reaching its expected end (`pledgedSrcSize`). This feels awkward, a responsability over-reach, beyond the definition of this API. ZSTD_compressContinue() is clearly documented as a guaranteed flush, with ZSTD_compressEnd() generating a guaranteed end. I faced similar issue when trying to port a similar mechanism at the higher streaming layer. Having ZSTD_CStream end a frame automatically on reaching `pledgedSrcSize` can surprise the caller, since it did not explicitly requested an end of frame. The only sensible action remaining after that is to end the frame with no additional input. This adds additional logic in the ZSTD_CStream state to check this condition. Plus some potential confusion on the meaning of ZSTD_endStream() with no additional input (ending confirmation ? new 0-size frame ?) In the end, just enlarging input buffer by 1 byte feels the least intrusive change. It's also a contract remaining inside the streaming layer, so the logic is contained in this part of the code. The patch also introduces a new test checking that size of small frame is as expected, without additional 3-bytes null block.
2017-12-14 19:47:02 +00:00
assert(zcs->inBuffSize > 0);
assert(zcs->outBuff != NULL);
assert(zcs->outBuffSize > 0);
assert(output->pos <= output->size);
assert(input->pos <= input->size);
2016-08-11 23:20:36 +00:00
while (someMoreWork) {
switch(zcs->streamStage)
2016-08-11 23:20:36 +00:00
{
case zcss_init:
/* call ZSTD_initCStream() first ! */
return ERROR(init_missing);
2016-08-11 23:20:36 +00:00
case zcss_load:
if ( (flushMode == ZSTD_e_end)
2017-07-03 22:52:19 +00:00
&& ((size_t)(oend-op) >= ZSTD_compressBound(iend-ip)) /* enough dstCapacity */
&& (zcs->inBuffPos == 0) ) {
/* shortcut to compression pass directly into output buffer */
size_t const cSize = ZSTD_compressEnd(zcs,
op, oend-op, ip, iend-ip);
DEBUGLOG(4, "ZSTD_compressEnd : %u", (U32)cSize);
if (ZSTD_isError(cSize)) return cSize;
ip = iend;
op += cSize;
zcs->frameEnded = 1;
ZSTD_startNewCompression(zcs);
someMoreWork = 0; break;
2017-07-03 22:52:19 +00:00
}
/* complete loading into inBuffer */
2016-08-11 23:20:36 +00:00
{ size_t const toLoad = zcs->inBuffTarget - zcs->inBuffPos;
size_t const loaded = ZSTD_limitCopy(
zcs->inBuff + zcs->inBuffPos, toLoad,
ip, iend-ip);
2016-08-11 23:20:36 +00:00
zcs->inBuffPos += loaded;
ip += loaded;
if ( (flushMode == ZSTD_e_continue)
&& (zcs->inBuffPos < zcs->inBuffTarget) ) {
/* not enough input to fill full block : stop here */
someMoreWork = 0; break;
}
if ( (flushMode == ZSTD_e_flush)
&& (zcs->inBuffPos == zcs->inToCompress) ) {
/* empty */
someMoreWork = 0; break;
}
}
2016-08-11 23:20:36 +00:00
/* compress current block (note : this stage cannot be stopped in the middle) */
DEBUGLOG(5, "stream compression stage (flushMode==%u)", flushMode);
2016-08-11 23:20:36 +00:00
{ void* cDst;
size_t cSize;
size_t const iSize = zcs->inBuffPos - zcs->inToCompress;
size_t oSize = oend-op;
unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip==iend);
2016-08-11 23:20:36 +00:00
if (oSize >= ZSTD_compressBound(iSize))
cDst = op; /* compress into output buffer, to skip flush stage */
2016-08-11 23:20:36 +00:00
else
cDst = zcs->outBuff, oSize = zcs->outBuffSize;
cSize = lastBlock ?
ZSTD_compressEnd(zcs, cDst, oSize,
zcs->inBuff + zcs->inToCompress, iSize) :
ZSTD_compressContinue(zcs, cDst, oSize,
zcs->inBuff + zcs->inToCompress, iSize);
2016-08-11 23:20:36 +00:00
if (ZSTD_isError(cSize)) return cSize;
zcs->frameEnded = lastBlock;
2016-08-11 23:20:36 +00:00
/* prepare next block */
zcs->inBuffTarget = zcs->inBuffPos + zcs->blockSize;
if (zcs->inBuffTarget > zcs->inBuffSize)
zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;
DEBUGLOG(5, "inBuffTarget:%u / inBuffSize:%u",
(U32)zcs->inBuffTarget, (U32)zcs->inBuffSize);
if (!lastBlock)
assert(zcs->inBuffTarget <= zcs->inBuffSize);
2016-08-11 23:20:36 +00:00
zcs->inToCompress = zcs->inBuffPos;
if (cDst == op) { /* no need to flush */
op += cSize;
if (zcs->frameEnded) {
DEBUGLOG(5, "Frame completed directly in outBuffer");
someMoreWork = 0;
2017-06-16 21:00:46 +00:00
ZSTD_startNewCompression(zcs);
}
break;
}
2016-08-11 23:20:36 +00:00
zcs->outBuffContentSize = cSize;
zcs->outBuffFlushedSize = 0;
zcs->streamStage = zcss_flush; /* pass-through to flush stage */
2016-08-11 23:20:36 +00:00
}
/* fall-through */
2016-08-11 23:20:36 +00:00
case zcss_flush:
DEBUGLOG(5, "flush stage");
2016-08-11 23:20:36 +00:00
{ size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
size_t const flushed = ZSTD_limitCopy(op, oend-op,
zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
DEBUGLOG(5, "toFlush: %u into %u ==> flushed: %u",
(U32)toFlush, (U32)(oend-op), (U32)flushed);
2016-08-11 23:20:36 +00:00
op += flushed;
zcs->outBuffFlushedSize += flushed;
if (toFlush!=flushed) {
/* flush not fully completed, presumably because dst is too small */
assert(op==oend);
someMoreWork = 0;
break;
}
2016-08-11 23:20:36 +00:00
zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
if (zcs->frameEnded) {
DEBUGLOG(5, "Frame completed on flush");
someMoreWork = 0;
2017-06-16 21:00:46 +00:00
ZSTD_startNewCompression(zcs);
break;
}
zcs->streamStage = zcss_load;
2016-08-11 23:20:36 +00:00
break;
}
default: /* impossible */
assert(0);
2016-08-11 23:20:36 +00:00
}
}
input->pos = ip - istart;
output->pos = op - ostart;
2016-08-11 23:20:36 +00:00
if (zcs->frameEnded) return 0;
{ size_t hintInSize = zcs->inBuffTarget - zcs->inBuffPos;
if (hintInSize==0) hintInSize = zcs->blockSize;
return hintInSize;
}
}
2016-08-16 23:39:22 +00:00
size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
{
/* check conditions */
if (output->pos > output->size) return ERROR(GENERIC);
if (input->pos > input->size) return ERROR(GENERIC);
return ZSTD_compressStream_generic(zcs, output, input, ZSTD_e_continue);
2016-08-11 23:20:36 +00:00
}
2017-06-06 01:32:48 +00:00
size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
ZSTD_outBuffer* output,
ZSTD_inBuffer* input,
ZSTD_EndDirective endOp)
{
DEBUGLOG(5, "ZSTD_compress_generic, endOp=%u ", (U32)endOp);
/* check conditions */
if (output->pos > output->size) return ERROR(GENERIC);
if (input->pos > input->size) return ERROR(GENERIC);
assert(cctx!=NULL);
/* transparent initialization stage */
if (cctx->streamStage == zcss_init) {
ZSTD_CCtx_params params = cctx->requestedParams;
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
ZSTD_prefixDict const prefixDict = cctx->prefixDict;
memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict)); /* single usage */
assert(prefixDict.dict==NULL || cctx->cdict==NULL); /* only one can be set */
DEBUGLOG(4, "ZSTD_compress_generic : transparent init stage");
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = input->size + 1; /* auto-fix pledgedSrcSize */
params.cParams = ZSTD_getCParamsFromCCtxParams(
cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1, 0 /*dictSize*/);
#ifdef ZSTD_MULTITHREAD
if ((cctx->pledgedSrcSizePlusOne-1) <= ZSTDMT_JOBSIZE_MIN) {
params.nbThreads = 1; /* do not invoke multi-threading when src size is too small */
params.nonBlockingMode = 0;
}
if ((params.nbThreads > 1) | (params.nonBlockingMode == 1)) {
zstdmt via compress_generic: reduce opportunity to free/create mtctx `zstreamtest --newapi` (and `--opaqueapi`) create and destroy way too many threads resulting in failure of tsan tests, and potentially connected to the qemu flaky tests. This is because, at each test, the nb of threads can be changed (random). The `--no-big-tests` directive reduce this choice to 1/2 threads, in order to limit memory usage, especially for qemu and 32-bits builds. Unfortunately, swapping between 1 and 2 threads is enough to constantly create/destroy new mtctx. This patch takes advantage of the following property : via compress_generic, no internal mtctx is needed for nbThreads < 2. As a consequence, when nbThreads == 2, the currently active mtctx is necessarily good. This dramatically reduces the nb of thread creations when invoking `zstreamtest --newapi --no-big-tests` (only when parent cctx itself is created, which is randomized to 1/256 tests). Expected outcome : - at a minimum : tsan tests shall now work continuously without exploding the thread counter - at best : flaky qemu tests on `zstreamtest --newapi --no-big-tests` may stop being flaky, due to less stress from constant thread creation/destruction Real world impact : minimal, I don't expect users to constantly change `nbThreads` between each invocation. If `nbThreads` remains stable, existing implementation re-uses existing mtctx. Also : `zstreamtest --newapi` but without `--no-big-tests` doesn't benefit as much, since this test can select a random `nbThreads` value between 1 and 4. The current patch only reduces opportunity to free/create mtctx (for example : 2->1->2 doesn't need a new mtctx) but doesn't completely eliminate it, since `nbThreads` can still change between 2/3/4. A more complete solution could be to only use 2 out of 4 allocated threads, thus keeping the pool at a constant size. This would require a larger change to `POOL_*` api though.
2017-12-16 20:48:13 +00:00
if (cctx->mtctx == NULL || (params.nbThreads != ZSTDMT_getNbThreads(cctx->mtctx))) {
DEBUGLOG(4, "ZSTD_compress_generic: creating new mtctx for nbThreads=%u",
params.nbThreads);
if (cctx->mtctx != NULL)
DEBUGLOG(4, "ZSTD_compress_generic: previous nbThreads was %u",
ZSTDMT_getNbThreads(cctx->mtctx));
2017-08-29 23:18:21 +00:00
ZSTDMT_freeCCtx(cctx->mtctx);
cctx->mtctx = ZSTDMT_createCCtx_advanced(params.nbThreads, cctx->customMem);
if (cctx->mtctx == NULL) return ERROR(memory_allocation);
}
DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbThreads=%u", params.nbThreads);
CHECK_F( ZSTDMT_initCStream_internal(
cctx->mtctx,
prefixDict.dict, prefixDict.dictSize, ZSTD_dm_rawContent,
cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) );
2017-06-12 01:32:36 +00:00
cctx->streamStage = zcss_load;
2017-08-29 23:18:21 +00:00
cctx->appliedParams.nbThreads = params.nbThreads;
cctx->appliedParams.nonBlockingMode = params.nonBlockingMode;
} else
#endif
{ CHECK_F( ZSTD_resetCStream_internal(
cctx, prefixDict.dict, prefixDict.dictSize,
prefixDict.dictMode, cctx->cdict, params,
cctx->pledgedSrcSizePlusOne-1) );
assert(cctx->streamStage == zcss_load);
assert(cctx->appliedParams.nbThreads <= 1);
2017-06-06 01:32:48 +00:00
} }
/* compression stage */
#ifdef ZSTD_MULTITHREAD
if ((cctx->appliedParams.nbThreads > 1) | (cctx->appliedParams.nonBlockingMode==1)) {
2017-06-12 01:32:36 +00:00
size_t const flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
2017-06-16 21:00:46 +00:00
if ( ZSTD_isError(flushMin)
|| (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
ZSTD_startNewCompression(cctx);
}
2017-06-12 01:32:36 +00:00
return flushMin;
}
#endif
CHECK_F( ZSTD_compressStream_generic(cctx, output, input, endOp) );
DEBUGLOG(5, "completed ZSTD_compress_generic");
return cctx->outBuffContentSize - cctx->outBuffFlushedSize; /* remaining to flush */
}
size_t ZSTD_compress_generic_simpleArgs (
ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity, size_t* dstPos,
const void* src, size_t srcSize, size_t* srcPos,
ZSTD_EndDirective endOp)
{
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_compress_generic(cctx, &output, &input, endOp);
*dstPos = output.pos;
*srcPos = input.pos;
return cErr;
2016-08-11 23:20:36 +00:00
}
2016-08-12 11:04:27 +00:00
/*====== Finalize ======*/
2016-08-11 23:20:36 +00:00
/*! ZSTD_flushStream() :
* @return : amount of data remaining to flush */
2016-08-16 23:39:22 +00:00
size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
2016-08-11 23:20:36 +00:00
{
2017-05-31 16:59:22 +00:00
ZSTD_inBuffer input = { NULL, 0, 0 };
if (output->pos > output->size) return ERROR(GENERIC);
CHECK_F( ZSTD_compressStream_generic(zcs, output, &input, ZSTD_e_flush) );
return zcs->outBuffContentSize - zcs->outBuffFlushedSize; /* remaining to flush */
2016-08-11 23:20:36 +00:00
}
2016-08-16 23:39:22 +00:00
size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
2016-08-11 23:20:36 +00:00
{
2017-05-31 16:59:22 +00:00
ZSTD_inBuffer input = { NULL, 0, 0 };
if (output->pos > output->size) return ERROR(GENERIC);
CHECK_F( ZSTD_compressStream_generic(zcs, output, &input, ZSTD_e_end) );
{ size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
size_t const checksumSize = zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4;
size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize + lastBlockSize + checksumSize;
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
DEBUGLOG(4, "ZSTD_endStream : remaining to flush : %u", (U32)toFlush);
return toFlush;
}
2016-08-11 23:20:36 +00:00
}
2016-02-10 12:37:52 +00:00
/*-===== Pre-defined compression levels =====-*/
#define ZSTD_MAX_CLEVEL 22
int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
{ /* "default" - guarantees a monotonically increasing memory budget */
2016-04-09 18:32:00 +00:00
/* W, C, H, S, L, TL, strat */
2016-08-26 18:02:49 +00:00
{ 18, 12, 12, 1, 7, 16, ZSTD_fast }, /* level 0 - never used */
2016-07-13 12:56:24 +00:00
{ 19, 13, 14, 1, 7, 16, ZSTD_fast }, /* level 1 */
{ 19, 15, 16, 1, 6, 16, ZSTD_fast }, /* level 2 */
{ 20, 16, 17, 1, 5, 16, ZSTD_dfast }, /* level 3 */
{ 20, 17, 18, 1, 5, 16, ZSTD_dfast }, /* level 4 */
{ 20, 17, 18, 2, 5, 16, ZSTD_greedy }, /* level 5 */
{ 21, 17, 19, 2, 5, 16, ZSTD_lazy }, /* level 6 */
{ 21, 18, 19, 3, 5, 16, ZSTD_lazy }, /* level 7 */
2016-08-26 18:02:49 +00:00
{ 21, 18, 20, 3, 5, 16, ZSTD_lazy2 }, /* level 8 */
{ 21, 19, 20, 3, 5, 16, ZSTD_lazy2 }, /* level 9 */
2016-07-13 12:56:24 +00:00
{ 21, 19, 21, 4, 5, 16, ZSTD_lazy2 }, /* level 10 */
{ 22, 20, 22, 4, 5, 16, ZSTD_lazy2 }, /* level 11 */
{ 22, 20, 22, 5, 5, 16, ZSTD_lazy2 }, /* level 12 */
{ 22, 21, 22, 4, 5, 16, ZSTD_btlazy2 }, /* level 13 */
{ 22, 21, 22, 5, 5, 16, ZSTD_btlazy2 }, /* level 14 */
{ 22, 22, 22, 6, 5, 16, ZSTD_btlazy2 }, /* level 15 */
{ 22, 21, 22, 4, 5, 48, ZSTD_btopt }, /* level 16 */
{ 23, 22, 22, 4, 4, 48, ZSTD_btopt }, /* level 17 */
{ 23, 22, 22, 5, 3, 64, ZSTD_btopt }, /* level 18 */
{ 23, 23, 22, 7, 3,128, ZSTD_btopt }, /* level 19 */
{ 25, 25, 23, 7, 3,128, ZSTD_btultra }, /* level 20 */
{ 26, 26, 24, 7, 3,256, ZSTD_btultra }, /* level 21 */
2017-05-11 00:48:42 +00:00
{ 27, 27, 25, 9, 3,512, ZSTD_btultra }, /* level 22 */
},
{ /* for srcSize <= 256 KB */
/* W, C, H, S, L, T, strat */
2016-08-26 18:02:49 +00:00
{ 0, 0, 0, 0, 0, 0, ZSTD_fast }, /* level 0 - not used */
2016-08-24 17:42:15 +00:00
{ 18, 13, 14, 1, 6, 8, ZSTD_fast }, /* level 1 */
2016-08-24 12:22:26 +00:00
{ 18, 14, 13, 1, 5, 8, ZSTD_dfast }, /* level 2 */
{ 18, 16, 15, 1, 5, 8, ZSTD_dfast }, /* level 3 */
{ 18, 15, 17, 1, 5, 8, ZSTD_greedy }, /* level 4.*/
{ 18, 16, 17, 4, 5, 8, ZSTD_greedy }, /* level 5.*/
{ 18, 16, 17, 3, 5, 8, ZSTD_lazy }, /* level 6.*/
{ 18, 17, 17, 4, 4, 8, ZSTD_lazy }, /* level 7 */
{ 18, 17, 17, 4, 4, 8, ZSTD_lazy2 }, /* level 8 */
{ 18, 17, 17, 5, 4, 8, ZSTD_lazy2 }, /* level 9 */
{ 18, 17, 17, 6, 4, 8, ZSTD_lazy2 }, /* level 10 */
{ 18, 18, 17, 6, 4, 8, ZSTD_lazy2 }, /* level 11.*/
{ 18, 18, 17, 7, 4, 8, ZSTD_lazy2 }, /* level 12.*/
{ 18, 19, 17, 6, 4, 8, ZSTD_btlazy2 }, /* level 13 */
2016-04-08 10:36:19 +00:00
{ 18, 18, 18, 4, 4, 16, ZSTD_btopt }, /* level 14.*/
2016-08-24 12:22:26 +00:00
{ 18, 18, 18, 4, 3, 16, ZSTD_btopt }, /* level 15.*/
{ 18, 19, 18, 6, 3, 32, ZSTD_btopt }, /* level 16.*/
{ 18, 19, 18, 8, 3, 64, ZSTD_btopt }, /* level 17.*/
2016-04-08 10:36:19 +00:00
{ 18, 19, 18, 9, 3,128, ZSTD_btopt }, /* level 18.*/
{ 18, 19, 18, 10, 3,256, ZSTD_btopt }, /* level 19.*/
2017-05-11 00:48:42 +00:00
{ 18, 19, 18, 11, 3,512, ZSTD_btultra }, /* level 20.*/
{ 18, 19, 18, 12, 3,512, ZSTD_btultra }, /* level 21.*/
{ 18, 19, 18, 13, 3,512, ZSTD_btultra }, /* level 22.*/
},
{ /* for srcSize <= 128 KB */
/* W, C, H, S, L, T, strat */
2016-07-22 12:36:46 +00:00
{ 17, 12, 12, 1, 7, 8, ZSTD_fast }, /* level 0 - not used */
{ 17, 12, 13, 1, 6, 8, ZSTD_fast }, /* level 1 */
{ 17, 13, 16, 1, 5, 8, ZSTD_fast }, /* level 2 */
{ 17, 16, 16, 2, 5, 8, ZSTD_dfast }, /* level 3 */
{ 17, 13, 15, 3, 4, 8, ZSTD_greedy }, /* level 4 */
{ 17, 15, 17, 4, 4, 8, ZSTD_greedy }, /* level 5 */
{ 17, 16, 17, 3, 4, 8, ZSTD_lazy }, /* level 6 */
{ 17, 15, 17, 4, 4, 8, ZSTD_lazy2 }, /* level 7 */
{ 17, 17, 17, 4, 4, 8, ZSTD_lazy2 }, /* level 8 */
{ 17, 17, 17, 5, 4, 8, ZSTD_lazy2 }, /* level 9 */
{ 17, 17, 17, 6, 4, 8, ZSTD_lazy2 }, /* level 10 */
{ 17, 17, 17, 7, 4, 8, ZSTD_lazy2 }, /* level 11 */
{ 17, 17, 17, 8, 4, 8, ZSTD_lazy2 }, /* level 12 */
{ 17, 18, 17, 6, 4, 8, ZSTD_btlazy2 }, /* level 13.*/
{ 17, 17, 17, 7, 3, 8, ZSTD_btopt }, /* level 14.*/
{ 17, 17, 17, 7, 3, 16, ZSTD_btopt }, /* level 15.*/
{ 17, 18, 17, 7, 3, 32, ZSTD_btopt }, /* level 16.*/
{ 17, 18, 17, 7, 3, 64, ZSTD_btopt }, /* level 17.*/
{ 17, 18, 17, 7, 3,256, ZSTD_btopt }, /* level 18.*/
{ 17, 18, 17, 8, 3,256, ZSTD_btopt }, /* level 19.*/
2017-05-11 00:48:42 +00:00
{ 17, 18, 17, 9, 3,256, ZSTD_btultra }, /* level 20.*/
{ 17, 18, 17, 10, 3,256, ZSTD_btultra }, /* level 21.*/
{ 17, 18, 17, 11, 3,512, ZSTD_btultra }, /* level 22.*/
},
{ /* for srcSize <= 16 KB */
/* W, C, H, S, L, T, strat */
{ 14, 12, 12, 1, 7, 6, ZSTD_fast }, /* level 0 - not used */
2016-07-17 14:21:37 +00:00
{ 14, 14, 14, 1, 6, 6, ZSTD_fast }, /* level 1 */
{ 14, 14, 14, 1, 4, 6, ZSTD_fast }, /* level 2 */
{ 14, 14, 14, 1, 4, 6, ZSTD_dfast }, /* level 3.*/
{ 14, 14, 14, 4, 4, 6, ZSTD_greedy }, /* level 4.*/
{ 14, 14, 14, 3, 4, 6, ZSTD_lazy }, /* level 5.*/
{ 14, 14, 14, 4, 4, 6, ZSTD_lazy2 }, /* level 6 */
{ 14, 14, 14, 5, 4, 6, ZSTD_lazy2 }, /* level 7 */
{ 14, 14, 14, 6, 4, 6, ZSTD_lazy2 }, /* level 8.*/
{ 14, 15, 14, 6, 4, 6, ZSTD_btlazy2 }, /* level 9.*/
{ 14, 15, 14, 3, 3, 6, ZSTD_btopt }, /* level 10.*/
{ 14, 15, 14, 6, 3, 8, ZSTD_btopt }, /* level 11.*/
{ 14, 15, 14, 6, 3, 16, ZSTD_btopt }, /* level 12.*/
{ 14, 15, 14, 6, 3, 24, ZSTD_btopt }, /* level 13.*/
{ 14, 15, 15, 6, 3, 48, ZSTD_btopt }, /* level 14.*/
{ 14, 15, 15, 6, 3, 64, ZSTD_btopt }, /* level 15.*/
{ 14, 15, 15, 6, 3, 96, ZSTD_btopt }, /* level 16.*/
{ 14, 15, 15, 6, 3,128, ZSTD_btopt }, /* level 17.*/
{ 14, 15, 15, 6, 3,256, ZSTD_btopt }, /* level 18.*/
{ 14, 15, 15, 7, 3,256, ZSTD_btopt }, /* level 19.*/
2017-05-11 00:48:42 +00:00
{ 14, 15, 15, 8, 3,256, ZSTD_btultra }, /* level 20.*/
{ 14, 15, 15, 9, 3,256, ZSTD_btultra }, /* level 21.*/
{ 14, 15, 15, 10, 3,256, ZSTD_btultra }, /* level 22.*/
},
};
/*! ZSTD_getCParams() :
* @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
* Size values are optional, provide 0 if not known or unused */
ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
{
size_t const addedSize = srcSizeHint ? 0 : 500;
U64 const rSize = srcSizeHint+dictSize ? srcSizeHint+dictSize+addedSize : (U64)-1;
U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB); /* intentional underflow for srcSizeHint == 0 */
DEBUGLOG(4, "ZSTD_getCParams: cLevel=%i, srcSize=%u, dictSize=%u => table %u",
compressionLevel, (U32)srcSizeHint, (U32)dictSize, tableID);
if (compressionLevel <= 0) compressionLevel = ZSTD_CLEVEL_DEFAULT; /* 0 == default; no negative compressionLevel yet */
if (compressionLevel > ZSTD_MAX_CLEVEL) compressionLevel = ZSTD_MAX_CLEVEL;
2017-06-21 00:44:55 +00:00
{ ZSTD_compressionParameters const cp = ZSTD_defaultCParameters[tableID][compressionLevel];
return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize); }
2017-08-29 02:25:17 +00:00
}
/*! ZSTD_getParams() :
2016-07-12 11:42:10 +00:00
* same as ZSTD_getCParams(), but @return a `ZSTD_parameters` object (instead of `ZSTD_compressionParameters`).
* All fields of `ZSTD_frameParameters` are set to default (0) */
ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize) {
ZSTD_parameters params;
ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, srcSizeHint, dictSize);
memset(&params, 0, sizeof(params));
params.cParams = cParams;
params.fParams.contentSizeFlag = 1;
return params;
}