zstd/lib/compress/zstd_compress.c

3993 lines
164 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
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 <limits.h> /* INT_MAX */
#include <string.h> /* memset */
2018-02-15 03:20:32 +00:00
#include "cpu.h"
2015-11-11 20:38:21 +00:00
#include "mem.h"
#include "hist.h" /* HIST_countFast_wksp */
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"
2019-07-03 20:40:37 +00:00
#include "zstd_compress_sequences.h"
#include "zstd_compress_literals.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 {
const void* dictContent;
size_t dictContentSize;
2019-07-15 19:08:49 +00:00
U32* entropyWorkspace; /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
2019-08-15 16:51:24 +00:00
ZSTD_cwksp workspace;
ZSTD_matchState_t matchState;
ZSTD_compressedBlockState_t cBlockState;
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
}
static void ZSTD_initCCtx(ZSTD_CCtx* cctx, ZSTD_customMem memManager)
{
assert(cctx != NULL);
memset(cctx, 0, sizeof(*cctx));
cctx->customMem = memManager;
cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
{ size_t const err = ZSTD_CCtx_reset(cctx, ZSTD_reset_parameters);
assert(!ZSTD_isError(err));
(void)err;
}
}
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_malloc(sizeof(ZSTD_CCtx), customMem);
if (!cctx) return NULL;
ZSTD_initCCtx(cctx, customMem);
return cctx;
}
2015-10-22 14:31:46 +00:00
}
ZSTD_CCtx* ZSTD_initStaticCCtx(void *workspace, size_t workspaceSize)
{
2019-08-15 16:51:24 +00:00
ZSTD_cwksp ws;
ZSTD_CCtx* cctx;
if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL; /* minimum size */
if ((size_t)workspace & 7) return NULL; /* must be 8-aligned */
2019-08-15 16:51:24 +00:00
ZSTD_cwksp_init(&ws, workspace, workspaceSize);
2019-08-15 16:51:24 +00:00
cctx = (ZSTD_CCtx*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CCtx));
if (cctx == NULL) {
return NULL;
}
memset(cctx, 0, sizeof(ZSTD_CCtx));
ZSTD_cwksp_move(&cctx->workspace, &ws);
cctx->staticSize = workspaceSize;
/* statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
2019-08-15 16:51:24 +00:00
if (!ZSTD_cwksp_check_available(&cctx->workspace, HUF_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t))) return NULL;
cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
cctx->blockState.nextCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
cctx->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(
2019-07-15 19:08:49 +00:00
&cctx->workspace, HUF_WORKSPACE_SIZE);
2018-02-15 03:20:32 +00:00
cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
2016-07-14 14:52:45 +00:00
return cctx;
2015-10-22 14:31:46 +00:00
}
/**
* Clears and frees all of the dictionaries in the CCtx.
*/
static void ZSTD_clearAllDicts(ZSTD_CCtx* cctx)
{
ZSTD_free(cctx->localDict.dictBuffer, cctx->customMem);
ZSTD_freeCDict(cctx->localDict.cdict);
memset(&cctx->localDict, 0, sizeof(cctx->localDict));
memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));
cctx->cdict = NULL;
}
static size_t ZSTD_sizeof_localDict(ZSTD_localDict dict)
{
size_t const bufferSize = dict.dictBuffer != NULL ? dict.dictSize : 0;
size_t const cdictSize = ZSTD_sizeof_CDict(dict.cdict);
return bufferSize + cdictSize;
}
static void ZSTD_freeCCtxContent(ZSTD_CCtx* cctx)
2015-10-25 13:06:35 +00:00
{
assert(cctx != NULL);
assert(cctx->staticSize == 0);
/* Only free workspace if cctx not in workspace, otherwise the workspace
* will be freed when the cctx itself is freed. */
if ((void*)cctx->workspace.workspace != (void*)cctx) {
2019-08-15 16:51:24 +00:00
ZSTD_cwksp_free(&cctx->workspace, cctx->customMem);
}
ZSTD_clearAllDicts(cctx);
#ifdef ZSTD_MULTITHREAD
ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL;
#endif
}
size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
{
if (cctx==NULL) return 0; /* support free on NULL */
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
"not compatible with static CCtx");
ZSTD_freeCCtxContent(cctx);
ZSTD_free(cctx, cctx->customMem);
return 0;
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 */
/* cctx may be in the workspace */
return (cctx->workspace.workspace == cctx ? 0 : sizeof(*cctx))
+ ZSTD_cwksp_sizeof(&cctx->workspace)
+ ZSTD_sizeof_localDict(cctx->localDict)
+ ZSTD_sizeof_mtctx(cctx);
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); }
2017-08-21 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_DEFAULT; /* should not matter, as all cParams are presumed properly defined */
assert(!ZSTD_checkCParams(cParams));
cctxParams.fParams.contentSizeFlag = 1;
2017-08-21 17:09:06 +00:00
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;
params->fParams.contentSizeFlag = 1;
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;
}
size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params)
2017-08-21 05:55:07 +00:00
{
return ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
2017-08-21 05:55:07 +00:00
}
size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(!cctxParams, GENERIC);
memset(cctxParams, 0, sizeof(*cctxParams));
2017-08-29 02:25:17 +00:00
cctxParams->compressionLevel = compressionLevel;
cctxParams->fParams.contentSizeFlag = 1;
2017-08-29 02:25:17 +00:00
return 0;
}
size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
{
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(!cctxParams, GENERIC);
FORWARD_IF_ERROR( 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_DEFAULT; /* should not matter, as all cParams are presumed properly defined */
assert(!ZSTD_checkCParams(params.cParams));
return 0;
}
/* ZSTD_assignParamsToCCtxParams() :
* params is presumed valid at this stage */
static ZSTD_CCtx_params ZSTD_assignParamsToCCtxParams(
const ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
{
ZSTD_CCtx_params ret = *cctxParams;
ret.cParams = params.cParams;
ret.fParams = params.fParams;
ret.compressionLevel = ZSTD_CLEVEL_DEFAULT; /* should not matter, as all cParams are presumed properly defined */
assert(!ZSTD_checkCParams(params.cParams));
return ret;
}
ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
{
ZSTD_bounds bounds = { 0, 0, 0 };
switch(param)
{
case ZSTD_c_compressionLevel:
bounds.lowerBound = ZSTD_minCLevel();
bounds.upperBound = ZSTD_maxCLevel();
return bounds;
case ZSTD_c_windowLog:
bounds.lowerBound = ZSTD_WINDOWLOG_MIN;
bounds.upperBound = ZSTD_WINDOWLOG_MAX;
return bounds;
case ZSTD_c_hashLog:
bounds.lowerBound = ZSTD_HASHLOG_MIN;
bounds.upperBound = ZSTD_HASHLOG_MAX;
return bounds;
case ZSTD_c_chainLog:
bounds.lowerBound = ZSTD_CHAINLOG_MIN;
bounds.upperBound = ZSTD_CHAINLOG_MAX;
return bounds;
case ZSTD_c_searchLog:
bounds.lowerBound = ZSTD_SEARCHLOG_MIN;
bounds.upperBound = ZSTD_SEARCHLOG_MAX;
return bounds;
case ZSTD_c_minMatch:
bounds.lowerBound = ZSTD_MINMATCH_MIN;
bounds.upperBound = ZSTD_MINMATCH_MAX;
return bounds;
case ZSTD_c_targetLength:
bounds.lowerBound = ZSTD_TARGETLENGTH_MIN;
bounds.upperBound = ZSTD_TARGETLENGTH_MAX;
return bounds;
case ZSTD_c_strategy:
2018-12-07 00:51:17 +00:00
bounds.lowerBound = ZSTD_STRATEGY_MIN;
bounds.upperBound = ZSTD_STRATEGY_MAX;
return bounds;
case ZSTD_c_contentSizeFlag:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_checksumFlag:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_dictIDFlag:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_nbWorkers:
bounds.lowerBound = 0;
#ifdef ZSTD_MULTITHREAD
bounds.upperBound = ZSTDMT_NBWORKERS_MAX;
#else
bounds.upperBound = 0;
#endif
return bounds;
case ZSTD_c_jobSize:
bounds.lowerBound = 0;
#ifdef ZSTD_MULTITHREAD
bounds.upperBound = ZSTDMT_JOBSIZE_MAX;
#else
bounds.upperBound = 0;
#endif
2018-11-21 00:09:33 +00:00
return bounds;
case ZSTD_c_overlapLog:
bounds.lowerBound = ZSTD_OVERLAPLOG_MIN;
bounds.upperBound = ZSTD_OVERLAPLOG_MAX;
return bounds;
case ZSTD_c_enableLongDistanceMatching:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_ldmHashLog:
bounds.lowerBound = ZSTD_LDM_HASHLOG_MIN;
bounds.upperBound = ZSTD_LDM_HASHLOG_MAX;
return bounds;
case ZSTD_c_ldmMinMatch:
bounds.lowerBound = ZSTD_LDM_MINMATCH_MIN;
bounds.upperBound = ZSTD_LDM_MINMATCH_MAX;
return bounds;
case ZSTD_c_ldmBucketSizeLog:
bounds.lowerBound = ZSTD_LDM_BUCKETSIZELOG_MIN;
bounds.upperBound = ZSTD_LDM_BUCKETSIZELOG_MAX;
return bounds;
case ZSTD_c_ldmHashRateLog:
2018-11-21 22:36:57 +00:00
bounds.lowerBound = ZSTD_LDM_HASHRATELOG_MIN;
bounds.upperBound = ZSTD_LDM_HASHRATELOG_MAX;
return bounds;
/* experimental parameters */
case ZSTD_c_rsyncable:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_forceMaxWindow :
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_format:
2018-12-07 00:51:17 +00:00
ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
bounds.lowerBound = ZSTD_f_zstd1;
bounds.upperBound = ZSTD_f_zstd1_magicless; /* note : how to ensure at compile time that this is the highest value enum ? */
return bounds;
case ZSTD_c_forceAttachDict:
2018-12-07 00:51:17 +00:00
ZSTD_STATIC_ASSERT(ZSTD_dictDefaultAttach < ZSTD_dictForceCopy);
bounds.lowerBound = ZSTD_dictDefaultAttach;
2018-12-07 00:51:17 +00:00
bounds.upperBound = ZSTD_dictForceCopy; /* note : how to ensure at compile time that this is the highest value enum ? */
return bounds;
case ZSTD_c_literalCompressionMode:
ZSTD_STATIC_ASSERT(ZSTD_lcm_auto < ZSTD_lcm_huffman && ZSTD_lcm_huffman < ZSTD_lcm_uncompressed);
bounds.lowerBound = ZSTD_lcm_auto;
bounds.upperBound = ZSTD_lcm_uncompressed;
return bounds;
2019-06-24 20:40:52 +00:00
case ZSTD_c_targetCBlockSize:
bounds.lowerBound = ZSTD_TARGETCBLOCKSIZE_MIN;
bounds.upperBound = ZSTD_TARGETCBLOCKSIZE_MAX;
return bounds;
2019-08-19 15:52:08 +00:00
case ZSTD_c_srcSizeHint:
2019-08-20 20:06:15 +00:00
bounds.lowerBound = ZSTD_SRCSIZEHINT_MIN;
2019-08-19 15:52:08 +00:00
bounds.upperBound = ZSTD_SRCSIZEHINT_MAX;
return bounds;
default:
{ ZSTD_bounds const boundError = { ERROR(parameter_unsupported), 0, 0 };
return boundError;
}
}
}
/* ZSTD_cParam_clampBounds:
* Clamps the value into the bounded range.
*/
static size_t ZSTD_cParam_clampBounds(ZSTD_cParameter cParam, int* value)
{
ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);
if (ZSTD_isError(bounds.error)) return bounds.error;
if (*value < bounds.lowerBound) *value = bounds.lowerBound;
if (*value > bounds.upperBound) *value = bounds.upperBound;
return 0;
}
2019-01-28 22:05:18 +00:00
#define BOUNDCHECK(cParam, val) { \
RETURN_ERROR_IF(!ZSTD_cParam_withinBounds(cParam,val), \
parameter_outOfBound); \
}
static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
{
switch(param)
{
case ZSTD_c_compressionLevel:
case ZSTD_c_hashLog:
case ZSTD_c_chainLog:
case ZSTD_c_searchLog:
case ZSTD_c_minMatch:
case ZSTD_c_targetLength:
case ZSTD_c_strategy:
return 1;
case ZSTD_c_format:
case ZSTD_c_windowLog:
case ZSTD_c_contentSizeFlag:
case ZSTD_c_checksumFlag:
case ZSTD_c_dictIDFlag:
case ZSTD_c_forceMaxWindow :
case ZSTD_c_nbWorkers:
case ZSTD_c_jobSize:
case ZSTD_c_overlapLog:
case ZSTD_c_rsyncable:
case ZSTD_c_enableLongDistanceMatching:
case ZSTD_c_ldmHashLog:
case ZSTD_c_ldmMinMatch:
case ZSTD_c_ldmBucketSizeLog:
case ZSTD_c_ldmHashRateLog:
case ZSTD_c_forceAttachDict:
2019-02-15 18:29:03 +00:00
case ZSTD_c_literalCompressionMode:
2019-06-24 20:40:52 +00:00
case ZSTD_c_targetCBlockSize:
2019-08-19 15:52:08 +00:00
case ZSTD_c_srcSizeHint:
default:
return 0;
}
}
size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
2017-05-12 22:31:53 +00:00
{
DEBUGLOG(4, "ZSTD_CCtx_setParameter (%i, %i)", (int)param, value);
if (cctx->streamStage != zcss_init) {
if (ZSTD_isUpdateAuthorized(param)) {
cctx->cParamsChanged = 1;
} else {
2019-01-28 22:05:18 +00:00
RETURN_ERROR(stage_wrong);
} }
2017-05-12 22:31:53 +00:00
switch(param)
{
case ZSTD_c_nbWorkers:
RETURN_ERROR_IF((value!=0) && cctx->staticSize, parameter_unsupported,
"MT not compatible with static alloc");
break;
2017-08-21 08:59:08 +00:00
case ZSTD_c_compressionLevel:
case ZSTD_c_windowLog:
case ZSTD_c_hashLog:
case ZSTD_c_chainLog:
case ZSTD_c_searchLog:
case ZSTD_c_minMatch:
case ZSTD_c_targetLength:
case ZSTD_c_strategy:
case ZSTD_c_ldmHashRateLog:
case ZSTD_c_format:
case ZSTD_c_contentSizeFlag:
case ZSTD_c_checksumFlag:
case ZSTD_c_dictIDFlag:
case ZSTD_c_forceMaxWindow:
case ZSTD_c_forceAttachDict:
case ZSTD_c_literalCompressionMode:
case ZSTD_c_jobSize:
case ZSTD_c_overlapLog:
case ZSTD_c_rsyncable:
case ZSTD_c_enableLongDistanceMatching:
case ZSTD_c_ldmHashLog:
case ZSTD_c_ldmMinMatch:
case ZSTD_c_ldmBucketSizeLog:
2019-06-24 20:40:52 +00:00
case ZSTD_c_targetCBlockSize:
2019-08-19 15:52:08 +00:00
case ZSTD_c_srcSizeHint:
break;
2019-01-28 22:05:18 +00:00
default: RETURN_ERROR(parameter_unsupported);
}
return ZSTD_CCtxParams_setParameter(&cctx->requestedParams, param, value);
}
size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
ZSTD_cParameter param, int value)
2017-08-18 02:30:22 +00:00
{
DEBUGLOG(4, "ZSTD_CCtxParams_setParameter (%i, %i)", (int)param, value);
2017-08-18 02:30:22 +00:00
switch(param)
{
case ZSTD_c_format :
2018-12-07 00:51:17 +00:00
BOUNDCHECK(ZSTD_c_format, 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
CCtxParams->format = (ZSTD_format_e)value;
return (size_t)CCtxParams->format;
case ZSTD_c_compressionLevel : {
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value));
if (value) { /* 0 : does not change current level */
CCtxParams->compressionLevel = value;
}
if (CCtxParams->compressionLevel >= 0) return CCtxParams->compressionLevel;
return 0; /* return type (size_t) cannot represent negative values */
}
2017-05-12 22:31:53 +00:00
case ZSTD_c_windowLog :
if (value!=0) /* 0 => use default */
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_windowLog, value);
CCtxParams->cParams.windowLog = 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
return CCtxParams->cParams.windowLog;
2017-05-12 22:31:53 +00:00
case ZSTD_c_hashLog :
if (value!=0) /* 0 => use default */
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_hashLog, value);
CCtxParams->cParams.hashLog = 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
return CCtxParams->cParams.hashLog;
2017-05-12 22:31:53 +00:00
case ZSTD_c_chainLog :
if (value!=0) /* 0 => use default */
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_chainLog, value);
CCtxParams->cParams.chainLog = 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
return CCtxParams->cParams.chainLog;
2017-05-12 22:31:53 +00:00
case ZSTD_c_searchLog :
if (value!=0) /* 0 => use default */
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_searchLog, value);
CCtxParams->cParams.searchLog = 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
return value;
2017-05-12 22:31:53 +00:00
case ZSTD_c_minMatch :
if (value!=0) /* 0 => use default */
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_minMatch, value);
CCtxParams->cParams.minMatch = value;
return CCtxParams->cParams.minMatch;
2017-05-12 22:31:53 +00:00
case ZSTD_c_targetLength :
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_targetLength, value);
CCtxParams->cParams.targetLength = 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
return CCtxParams->cParams.targetLength;
2017-05-12 22:31:53 +00:00
case ZSTD_c_strategy :
if (value!=0) /* 0 => use default */
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_strategy, value);
CCtxParams->cParams.strategy = (ZSTD_strategy)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
return (size_t)CCtxParams->cParams.strategy;
2017-05-12 22:31:53 +00:00
case ZSTD_c_contentSizeFlag :
/* Content size written in frame header _when known_ (default:1) */
DEBUGLOG(4, "set content size flag = %u", (value!=0));
CCtxParams->fParams.contentSizeFlag = value != 0;
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->fParams.contentSizeFlag;
2017-05-12 22:31:53 +00:00
case ZSTD_c_checksumFlag :
/* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
CCtxParams->fParams.checksumFlag = value != 0;
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->fParams.checksumFlag;
2017-05-12 22:31:53 +00:00
case ZSTD_c_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
DEBUGLOG(4, "set dictIDFlag = %u", (value!=0));
CCtxParams->fParams.noDictIDFlag = !value;
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->fParams.noDictIDFlag;
2017-05-12 22:31:53 +00:00
case ZSTD_c_forceMaxWindow :
CCtxParams->forceWindow = (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
return CCtxParams->forceWindow;
case ZSTD_c_forceAttachDict : {
2018-11-12 22:48:42 +00:00
const ZSTD_dictAttachPref_e pref = (ZSTD_dictAttachPref_e)value;
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_forceAttachDict, pref);
2018-11-12 20:52:31 +00:00
CCtxParams->attachDictPref = pref;
return CCtxParams->attachDictPref;
}
case ZSTD_c_literalCompressionMode : {
const ZSTD_literalCompressionMode_e lcm = (ZSTD_literalCompressionMode_e)value;
BOUNDCHECK(ZSTD_c_literalCompressionMode, lcm);
CCtxParams->literalCompressionMode = lcm;
return CCtxParams->literalCompressionMode;
}
case ZSTD_c_nbWorkers :
#ifndef ZSTD_MULTITHREAD
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
return 0;
#else
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value));
CCtxParams->nbWorkers = value;
return CCtxParams->nbWorkers;
#endif
case ZSTD_c_jobSize :
#ifndef ZSTD_MULTITHREAD
RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
return 0;
#else
/* Adjust to the minimum non-default value. */
if (value != 0 && value < ZSTDMT_JOBSIZE_MIN)
value = ZSTDMT_JOBSIZE_MIN;
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value));
assert(value >= 0);
CCtxParams->jobSize = value;
return CCtxParams->jobSize;
#endif
case ZSTD_c_overlapLog :
#ifndef ZSTD_MULTITHREAD
RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
return 0;
#else
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value));
CCtxParams->overlapLog = value;
return CCtxParams->overlapLog;
#endif
2017-05-12 22:31:53 +00:00
case ZSTD_c_rsyncable :
#ifndef ZSTD_MULTITHREAD
RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
return 0;
#else
FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value));
CCtxParams->rsyncable = value;
return CCtxParams->rsyncable;
#endif
case ZSTD_c_enableLongDistanceMatching :
CCtxParams->ldmParams.enableLdm = (value!=0);
return CCtxParams->ldmParams.enableLdm;
case ZSTD_c_ldmHashLog :
if (value!=0) /* 0 ==> auto */
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_ldmHashLog, value);
CCtxParams->ldmParams.hashLog = 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
return CCtxParams->ldmParams.hashLog;
case ZSTD_c_ldmMinMatch :
if (value!=0) /* 0 ==> default */
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_ldmMinMatch, value);
CCtxParams->ldmParams.minMatchLength = 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
return CCtxParams->ldmParams.minMatchLength;
case ZSTD_c_ldmBucketSizeLog :
if (value!=0) /* 0 ==> default */
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_ldmBucketSizeLog, 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
CCtxParams->ldmParams.bucketSizeLog = value;
return CCtxParams->ldmParams.bucketSizeLog;
2017-09-03 04:10:36 +00:00
case ZSTD_c_ldmHashRateLog :
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(value > ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN,
parameter_outOfBound);
2018-11-21 22:36:57 +00:00
CCtxParams->ldmParams.hashRateLog = value;
return CCtxParams->ldmParams.hashRateLog;
2019-06-24 20:40:52 +00:00
case ZSTD_c_targetCBlockSize :
if (value!=0) /* 0 ==> default */
BOUNDCHECK(ZSTD_c_targetCBlockSize, value);
CCtxParams->targetCBlockSize = value;
return CCtxParams->targetCBlockSize;
2019-08-19 15:52:08 +00:00
case ZSTD_c_srcSizeHint :
if (value!=0) /* 0 ==> default */
BOUNDCHECK(ZSTD_c_srcSizeHint, value);
CCtxParams->srcSizeHint = value;
return CCtxParams->srcSizeHint;
2019-01-28 22:05:18 +00:00
default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
}
}
size_t ZSTD_CCtx_getParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int* value)
{
return ZSTD_CCtxParams_getParameter(&cctx->requestedParams, param, value);
}
size_t ZSTD_CCtxParams_getParameter(
ZSTD_CCtx_params* CCtxParams, ZSTD_cParameter param, int* value)
{
switch(param)
{
case ZSTD_c_format :
*value = CCtxParams->format;
break;
case ZSTD_c_compressionLevel :
*value = CCtxParams->compressionLevel;
break;
case ZSTD_c_windowLog :
*value = (int)CCtxParams->cParams.windowLog;
break;
case ZSTD_c_hashLog :
*value = (int)CCtxParams->cParams.hashLog;
break;
case ZSTD_c_chainLog :
*value = (int)CCtxParams->cParams.chainLog;
break;
case ZSTD_c_searchLog :
*value = CCtxParams->cParams.searchLog;
break;
case ZSTD_c_minMatch :
*value = CCtxParams->cParams.minMatch;
break;
case ZSTD_c_targetLength :
*value = CCtxParams->cParams.targetLength;
break;
case ZSTD_c_strategy :
*value = (unsigned)CCtxParams->cParams.strategy;
break;
case ZSTD_c_contentSizeFlag :
*value = CCtxParams->fParams.contentSizeFlag;
break;
case ZSTD_c_checksumFlag :
*value = CCtxParams->fParams.checksumFlag;
break;
case ZSTD_c_dictIDFlag :
*value = !CCtxParams->fParams.noDictIDFlag;
break;
case ZSTD_c_forceMaxWindow :
*value = CCtxParams->forceWindow;
break;
case ZSTD_c_forceAttachDict :
*value = CCtxParams->attachDictPref;
break;
case ZSTD_c_literalCompressionMode :
*value = CCtxParams->literalCompressionMode;
break;
case ZSTD_c_nbWorkers :
#ifndef ZSTD_MULTITHREAD
assert(CCtxParams->nbWorkers == 0);
#endif
*value = CCtxParams->nbWorkers;
break;
case ZSTD_c_jobSize :
#ifndef ZSTD_MULTITHREAD
2019-01-28 22:05:18 +00:00
RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
#else
assert(CCtxParams->jobSize <= INT_MAX);
*value = (int)CCtxParams->jobSize;
break;
#endif
case ZSTD_c_overlapLog :
#ifndef ZSTD_MULTITHREAD
2019-01-28 22:05:18 +00:00
RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
#else
*value = CCtxParams->overlapLog;
break;
#endif
case ZSTD_c_rsyncable :
#ifndef ZSTD_MULTITHREAD
2019-01-28 22:05:18 +00:00
RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
#else
*value = CCtxParams->rsyncable;
break;
#endif
case ZSTD_c_enableLongDistanceMatching :
*value = CCtxParams->ldmParams.enableLdm;
break;
case ZSTD_c_ldmHashLog :
*value = CCtxParams->ldmParams.hashLog;
break;
case ZSTD_c_ldmMinMatch :
*value = CCtxParams->ldmParams.minMatchLength;
break;
case ZSTD_c_ldmBucketSizeLog :
*value = CCtxParams->ldmParams.bucketSizeLog;
break;
case ZSTD_c_ldmHashRateLog :
2018-11-21 22:36:57 +00:00
*value = CCtxParams->ldmParams.hashRateLog;
break;
2019-06-24 20:40:52 +00:00
case ZSTD_c_targetCBlockSize :
*value = (int)CCtxParams->targetCBlockSize;
break;
2019-08-19 15:52:08 +00:00
case ZSTD_c_srcSizeHint :
*value = (int)CCtxParams->srcSizeHint;
2019-08-19 19:32:43 +00:00
break;
2019-01-28 22:05:18 +00:00
default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
}
return 0;
}
/** ZSTD_CCtx_setParametersUsingCCtxParams() :
* just applies `params` into `cctx`
* no action is performed, parameters are merely stored.
* If ZSTDMT is enabled, parameters are pushed to cctx->mtctx.
* This is possible even if a compression is ongoing.
* In which case, new parameters will be applied on the fly, starting with next compression job.
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
{
DEBUGLOG(4, "ZSTD_CCtx_setParametersUsingCCtxParams");
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong);
RETURN_ERROR_IF(cctx->cdict, 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);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong);
cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
2017-05-12 22:31:53 +00:00
return 0;
}
/**
* Initializes the local dict using the requested parameters.
* NOTE: This does not use the pledged src size, because it may be used for more
* than one compression.
*/
static size_t ZSTD_initLocalDict(ZSTD_CCtx* cctx)
{
ZSTD_localDict* const dl = &cctx->localDict;
ZSTD_compressionParameters const cParams = ZSTD_getCParamsFromCCtxParams(
&cctx->requestedParams, 0, dl->dictSize);
if (dl->dict == NULL) {
/* No local dictionary. */
assert(dl->dictBuffer == NULL);
assert(dl->cdict == NULL);
assert(dl->dictSize == 0);
return 0;
}
if (dl->cdict != NULL) {
assert(cctx->cdict == dl->cdict);
/* Local dictionary already initialized. */
return 0;
}
assert(dl->dictSize > 0);
assert(cctx->cdict == NULL);
assert(cctx->prefixDict.dict == NULL);
dl->cdict = ZSTD_createCDict_advanced(
dl->dict,
dl->dictSize,
ZSTD_dlm_byRef,
dl->dictContentType,
cParams,
cctx->customMem);
RETURN_ERROR_IF(!dl->cdict, memory_allocation);
cctx->cdict = dl->cdict;
return 0;
}
size_t ZSTD_CCtx_loadDictionary_advanced(
ZSTD_CCtx* cctx, const void* dict, size_t dictSize,
ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType)
{
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong);
RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
"no malloc for static CCtx");
DEBUGLOG(4, "ZSTD_CCtx_loadDictionary_advanced (size: %u)", (U32)dictSize);
ZSTD_clearAllDicts(cctx); /* in case one already exists */
if (dict == NULL || dictSize == 0) /* no dictionary mode */
return 0;
if (dictLoadMethod == ZSTD_dlm_byRef) {
cctx->localDict.dict = dict;
} else {
void* dictBuffer = ZSTD_malloc(dictSize, cctx->customMem);
RETURN_ERROR_IF(!dictBuffer, memory_allocation);
memcpy(dictBuffer, dict, dictSize);
cctx->localDict.dictBuffer = dictBuffer;
cctx->localDict.dict = dictBuffer;
}
cctx->localDict.dictSize = dictSize;
cctx->localDict.dictContentType = dictContentType;
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_dct_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_dct_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
{
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong);
/* Free the existing local cdict (if any) to save memory. */
ZSTD_clearAllDicts(cctx);
2017-06-27 22:49:12 +00:00
cctx->cdict = cdict;
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_dct_rawContent);
}
size_t ZSTD_CCtx_refPrefix_advanced(
ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
2016-02-11 23:07:30 +00:00
{
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong);
ZSTD_clearAllDicts(cctx);
cctx->prefixDict.dict = prefix;
cctx->prefixDict.dictSize = prefixSize;
cctx->prefixDict.dictContentType = dictContentType;
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
/*! ZSTD_CCtx_reset() :
* Also dumps dictionary */
size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset)
2017-06-16 21:00:46 +00:00
{
if ( (reset == ZSTD_reset_session_only)
|| (reset == ZSTD_reset_session_and_parameters) ) {
cctx->streamStage = zcss_init;
cctx->pledgedSrcSizePlusOne = 0;
}
if ( (reset == ZSTD_reset_parameters)
|| (reset == ZSTD_reset_session_and_parameters) ) {
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong);
ZSTD_clearAllDicts(cctx);
return ZSTD_CCtxParams_reset(&cctx->requestedParams);
}
return 0;
2017-06-16 17:17:50 +00:00
}
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)
{
BOUNDCHECK(ZSTD_c_windowLog, (int)cParams.windowLog);
BOUNDCHECK(ZSTD_c_chainLog, (int)cParams.chainLog);
BOUNDCHECK(ZSTD_c_hashLog, (int)cParams.hashLog);
BOUNDCHECK(ZSTD_c_searchLog, (int)cParams.searchLog);
BOUNDCHECK(ZSTD_c_minMatch, (int)cParams.minMatch);
BOUNDCHECK(ZSTD_c_targetLength,(int)cParams.targetLength);
2018-12-06 23:57:55 +00:00
BOUNDCHECK(ZSTD_c_strategy, cParams.strategy);
return 0;
}
/** ZSTD_clampCParams() :
* make CParam values within valid range.
* @return : valid CParams */
static ZSTD_compressionParameters
ZSTD_clampCParams(ZSTD_compressionParameters cParams)
{
2018-12-06 23:57:55 +00:00
# define CLAMP_TYPE(cParam, val, type) { \
ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam); \
if ((int)val<bounds.lowerBound) val=(type)bounds.lowerBound; \
else if ((int)val>bounds.upperBound) val=(type)bounds.upperBound; \
}
# define CLAMP(cParam, val) CLAMP_TYPE(cParam, val, unsigned)
CLAMP(ZSTD_c_windowLog, cParams.windowLog);
CLAMP(ZSTD_c_chainLog, cParams.chainLog);
CLAMP(ZSTD_c_hashLog, cParams.hashLog);
CLAMP(ZSTD_c_searchLog, cParams.searchLog);
CLAMP(ZSTD_c_minMatch, cParams.minMatch);
CLAMP(ZSTD_c_targetLength,cParams.targetLength);
CLAMP_TYPE(ZSTD_c_strategy,cParams.strategy, ZSTD_strategy);
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 specified input (`srcSize` and `dictSize`).
* mostly downsize to reduce memory consumption and initialization latency.
* `srcSize` can be ZSTD_CONTENTSIZE_UNKNOWN when not known.
* note : for the time being, `srcSize==0` means "unknown" too, for compatibility with older convention.
* condition : cPar is presumed validated (can be checked using ZSTD_checkCParams()). */
static 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) /* ZSTD_CONTENTSIZE_UNKNOWN and 0 mean "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+1) cPar.hashLog = cPar.windowLog+1;
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; /* minimum wlog required for valid 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); /* resulting cPar is necessarily valid (all parameters within range) */
return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize);
}
ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize)
{
2019-08-19 15:52:08 +00:00
ZSTD_compressionParameters cParams;
if (srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN && CCtxParams->srcSizeHint > 0) {
srcSizeHint = CCtxParams->srcSizeHint;
}
cParams = ZSTD_getCParams(CCtxParams->compressionLevel, srcSizeHint, dictSize);
if (CCtxParams->ldmParams.enableLdm) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
if (CCtxParams->cParams.windowLog) cParams.windowLog = CCtxParams->cParams.windowLog;
if (CCtxParams->cParams.hashLog) cParams.hashLog = CCtxParams->cParams.hashLog;
if (CCtxParams->cParams.chainLog) cParams.chainLog = CCtxParams->cParams.chainLog;
if (CCtxParams->cParams.searchLog) cParams.searchLog = CCtxParams->cParams.searchLog;
if (CCtxParams->cParams.minMatch) cParams.minMatch = CCtxParams->cParams.minMatch;
if (CCtxParams->cParams.targetLength) cParams.targetLength = CCtxParams->cParams.targetLength;
if (CCtxParams->cParams.strategy) cParams.strategy = CCtxParams->cParams.strategy;
assert(!ZSTD_checkCParams(cParams));
return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize);
}
static size_t
ZSTD_sizeof_matchState(const 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->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
size_t const h3Size = ((size_t)1) << hashLog3;
size_t const tableSpace = ZSTD_cwksp_alloc_size(chainSize * sizeof(U32))
+ ZSTD_cwksp_alloc_size(hSize * sizeof(U32))
+ ZSTD_cwksp_alloc_size(h3Size * sizeof(U32));
size_t const optPotentialSpace =
ZSTD_cwksp_alloc_size((MaxML+1) * sizeof(U32))
+ ZSTD_cwksp_alloc_size((MaxLL+1) * sizeof(U32))
+ ZSTD_cwksp_alloc_size((MaxOff+1) * sizeof(U32))
+ ZSTD_cwksp_alloc_size((1<<Litbits) * sizeof(U32))
+ ZSTD_cwksp_alloc_size((ZSTD_OPT_NUM+1) * sizeof(ZSTD_match_t))
+ ZSTD_cwksp_alloc_size((ZSTD_OPT_NUM+1) * sizeof(ZSTD_optimal_t));
size_t const optSpace = (forCCtx && (cParams->strategy >= ZSTD_btopt))
? 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
{
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
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.minMatch==3) ? 3 : 4;
size_t const maxNbSeq = blockSize / divider;
size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize + 11*maxNbSeq);
size_t const entropySpace = ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE);
size_t const blockStateSpace = 2 * ZSTD_cwksp_alloc_size(sizeof(ZSTD_compressedBlockState_t));
size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 1);
size_t const ldmSpace = ZSTD_ldm_getTableSize(params->ldmParams);
size_t const ldmSeqSpace = ZSTD_cwksp_alloc_size(ZSTD_ldm_getMaxNbSeq(params->ldmParams, blockSize) * sizeof(rawSeq));
size_t const neededSpace = entropySpace + blockStateSpace + tokenSpace +
matchStateSize + ldmSpace + ldmSeqSpace;
size_t const cctxSpace = ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx));
2016-07-11 01:12:17 +00:00
DEBUGLOG(5, "sizeof(ZSTD_CCtx) : %u", (U32)cctxSpace);
DEBUGLOG(5, "estimate workspace : %u", (U32)neededSpace);
return cctxSpace + 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=MIN(compressionLevel, 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)
{
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
{ ZSTD_compressionParameters const cParams =
ZSTD_getCParamsFromCCtxParams(params, 0, 0);
size_t const CCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(params);
size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
size_t const inBuffSize = ((size_t)1 << 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=MIN(compressionLevel, 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.nbWorkers > 0) {
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;
fp.flushed = cctx->producedCSize; /* simplified; some data might still be left within streaming output buffer */
fp.currentJobID = 0;
fp.nbActiveWorkers = 0;
return fp;
} }
/*! ZSTD_toFlushNow()
* Only useful for multithreading scenarios currently (nbWorkers >= 1).
*/
size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx)
{
#ifdef ZSTD_MULTITHREAD
if (cctx->appliedParams.nbWorkers > 0) {
return ZSTDMT_toFlushNow(cctx->mtctx);
}
#endif
(void)cctx;
return 0; /* over-simplification; could also check if context is currently running in streaming mode, and in which case, report how many bytes are left to be flushed within output buffer */
}
static void ZSTD_assertEqualCParams(ZSTD_compressionParameters cParams1,
ZSTD_compressionParameters cParams2)
{
(void)cParams1;
(void)cParams2;
assert(cParams1.windowLog == cParams2.windowLog);
assert(cParams1.chainLog == cParams2.chainLog);
assert(cParams1.hashLog == cParams2.hashLog);
assert(cParams1.searchLog == cParams2.searchLog);
assert(cParams1.minMatch == cParams2.minMatch);
assert(cParams1.targetLength == cParams2.targetLength);
assert(cParams1.strategy == cParams2.strategy);
}
static void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
{
int i;
for (i = 0; i < ZSTD_REP_NUM; ++i)
bs->rep[i] = repStartValue[i];
2018-05-22 23:06:33 +00:00
bs->entropy.huf.repeatMode = HUF_repeat_none;
bs->entropy.fse.offcode_repeatMode = FSE_repeat_none;
bs->entropy.fse.matchlength_repeatMode = FSE_repeat_none;
bs->entropy.fse.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)
{
ZSTD_window_clear(&ms->window);
ms->nextToUpdate = ms->window.dictLimit;
ms->loadedDictEnd = 0;
ms->opt.litLengthSum = 0; /* force reset of btopt stats */
ms->dictMatchState = NULL;
}
/**
* Indicates whether this compression proceeds directly from user-provided
* source buffer to user-provided destination buffer (ZSTDb_not_buffered), or
* whether the context needs to buffer the input/output (ZSTDb_buffered).
*/
typedef enum {
ZSTDb_not_buffered,
ZSTDb_buffered
} ZSTD_buffered_policy_e;
/**
* Controls, for this matchState reset, whether the tables need to be cleared /
* prepared for the coming compression (ZSTDcrp_makeClean), or whether the
* tables can be left unclean (ZSTDcrp_leaveDirty), because we know that a
* subsequent operation will overwrite the table space anyways (e.g., copying
* the matchState contents in from a CDict).
*/
typedef enum {
ZSTDcrp_makeClean,
ZSTDcrp_leaveDirty
} ZSTD_compResetPolicy_e;
/**
* Controls, for this matchState reset, whether indexing can continue where it
* left off (ZSTDirp_continue), or whether it needs to be restarted from zero
* (ZSTDirp_reset).
*/
typedef enum {
ZSTDirp_continue,
ZSTDirp_reset
} ZSTD_indexResetPolicy_e;
typedef enum {
ZSTD_resetTarget_CDict,
ZSTD_resetTarget_CCtx
} ZSTD_resetTarget_e;
2019-07-15 19:08:49 +00:00
static size_t
ZSTD_reset_matchState(ZSTD_matchState_t* ms,
2019-08-15 16:51:24 +00:00
ZSTD_cwksp* ws,
const ZSTD_compressionParameters* cParams,
const ZSTD_compResetPolicy_e crp,
const ZSTD_indexResetPolicy_e forceResetIndex,
const ZSTD_resetTarget_e forWho)
{
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 = ((forWho == ZSTD_resetTarget_CCtx) && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
size_t const h3Size = hashLog3 ? ((size_t)1) << hashLog3 : 0;
2019-09-11 15:21:00 +00:00
DEBUGLOG(4, "reset indices : %u", forceResetIndex == ZSTDirp_reset);
if (forceResetIndex == ZSTDirp_reset) {
memset(&ms->window, 0, sizeof(ms->window));
ms->window.dictLimit = 1; /* start from 1, so that 1st position is valid */
ms->window.lowLimit = 1; /* it ensures first and later CCtx usages compress the same */
ms->window.nextSrc = ms->window.base + 1; /* see issue #1241 */
ZSTD_cwksp_mark_tables_dirty(ws);
}
ms->hashLog3 = hashLog3;
ZSTD_invalidateMatchState(ms);
2019-08-15 16:51:24 +00:00
assert(!ZSTD_cwksp_reserve_failed(ws)); /* check that allocation hasn't already failed */
2019-07-15 19:08:49 +00:00
2019-08-15 16:51:24 +00:00
ZSTD_cwksp_clear_tables(ws);
2019-08-12 22:36:48 +00:00
DEBUGLOG(5, "reserving table space");
/* table Space */
2019-08-15 16:51:24 +00:00
ms->hashTable = (U32*)ZSTD_cwksp_reserve_table(ws, hSize * sizeof(U32));
ms->chainTable = (U32*)ZSTD_cwksp_reserve_table(ws, chainSize * sizeof(U32));
ms->hashTable3 = (U32*)ZSTD_cwksp_reserve_table(ws, h3Size * sizeof(U32));
RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation,
2019-08-12 22:36:48 +00:00
"failed a workspace allocation in ZSTD_reset_matchState");
DEBUGLOG(4, "reset table : %u", crp!=ZSTDcrp_leaveDirty);
if (crp!=ZSTDcrp_leaveDirty) {
2019-08-12 22:36:48 +00:00
/* reset tables only */
ZSTD_cwksp_clean_tables(ws);
2019-07-15 19:08:49 +00:00
}
/* opt parser space */
if ((forWho == ZSTD_resetTarget_CCtx) && (cParams->strategy >= ZSTD_btopt)) {
DEBUGLOG(4, "reserving optimal parser space");
2019-08-15 16:51:24 +00:00
ms->opt.litFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (1<<Litbits) * sizeof(unsigned));
ms->opt.litLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxLL+1) * sizeof(unsigned));
ms->opt.matchLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxML+1) * sizeof(unsigned));
ms->opt.offCodeFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxOff+1) * sizeof(unsigned));
ms->opt.matchTable = (ZSTD_match_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_match_t));
ms->opt.priceTable = (ZSTD_optimal_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_optimal_t));
}
ms->cParams = *cParams;
2019-08-15 16:51:24 +00:00
RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation,
2019-07-15 19:08:49 +00:00
"failed a workspace allocation in ZSTD_reset_matchState");
return 0;
}
/* ZSTD_indexTooCloseToMax() :
* minor optimization : prefer memset() rather than reduceIndex()
* which is measurably slow in some circumstances (reported for Visual Studio).
* Works when re-using a context for a lot of smallish inputs :
* if all inputs are smaller than ZSTD_INDEXOVERFLOW_MARGIN,
* memset() will be triggered before reduceIndex().
*/
#define ZSTD_INDEXOVERFLOW_MARGIN (16 MB)
static int ZSTD_indexTooCloseToMax(ZSTD_window_t w)
{
return (size_t)(w.nextSrc - w.base) > (ZSTD_CURRENT_MAX - ZSTD_INDEXOVERFLOW_MARGIN);
}
/*! ZSTD_resetCCtx_internal() :
note : `params` are assumed fully validated at this stage */
static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
ZSTD_CCtx_params params,
U64 const pledgedSrcSize,
ZSTD_compResetPolicy_e const crp,
ZSTD_buffered_policy_e const zbuff)
{
ZSTD_cwksp* const ws = &zc->workspace;
DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u",
(U32)pledgedSrcSize, params.cParams.windowLog);
assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
zc->isFirstBlock = 1;
if (params.ldmParams.enableLdm) {
2017-09-03 04:10:36 +00:00
/* Adjust long distance matching parameters */
ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
2017-09-07 00:56:01 +00:00
assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
2018-11-21 22:36:57 +00:00
assert(params.ldmParams.hashRateLog < 32);
zc->ldmState.hashPower = ZSTD_rollingHash_primePower(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.minMatch==3) ? 3 : 4;
size_t const maxNbSeq = blockSize / divider;
size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + 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);
size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(params.ldmParams, blockSize);
ZSTD_indexResetPolicy_e needsIndexReset = ZSTDirp_continue;
if (ZSTD_indexTooCloseToMax(zc->blockState.matchState.window)) {
needsIndexReset = ZSTDirp_reset;
}
ZSTD_cwksp_bump_oversized_duration(ws, 0);
/* Check if workspace is large enough, alloc a new one if needed */
{ size_t const cctxSpace = zc->staticSize ? ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx)) : 0;
size_t const entropySpace = ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE);
size_t const blockStateSpace = 2 * ZSTD_cwksp_alloc_size(sizeof(ZSTD_compressedBlockState_t));
size_t const bufferSpace = ZSTD_cwksp_alloc_size(buffInSize) + ZSTD_cwksp_alloc_size(buffOutSize);
size_t const ldmSpace = ZSTD_ldm_getTableSize(params.ldmParams);
size_t const ldmSeqSpace = ZSTD_cwksp_alloc_size(maxNbLdmSeq * sizeof(rawSeq));
2019-07-15 19:08:49 +00:00
size_t const neededSpace =
2019-09-09 19:12:14 +00:00
cctxSpace +
2019-07-15 19:08:49 +00:00
entropySpace +
blockStateSpace +
ldmSpace +
ldmSeqSpace +
matchStateSize +
tokenSpace +
bufferSpace;
2019-09-09 19:12:14 +00:00
int const workspaceTooSmall = ZSTD_cwksp_sizeof(ws) < neededSpace;
int const workspaceWasteful = ZSTD_cwksp_check_wasteful(ws, neededSpace);
2018-06-05 21:53:28 +00:00
DEBUGLOG(4, "Need %zuKB workspace, including %zuKB for match state, and %zuKB for buffers",
neededSpace>>10, matchStateSize>>10, bufferSpace>>10);
DEBUGLOG(4, "windowSize: %zu - blockSize: %zu", windowSize, blockSize);
if (workspaceTooSmall || workspaceWasteful) {
DEBUGLOG(4, "Resize workspaceSize from %zuKB to %zuKB",
ZSTD_cwksp_sizeof(ws) >> 10,
neededSpace >> 10);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(zc->staticSize, memory_allocation, "static cctx : no resize");
needsIndexReset = ZSTDirp_reset;
ZSTD_cwksp_free(ws, zc->customMem);
FORWARD_IF_ERROR(ZSTD_cwksp_create(ws, neededSpace, zc->customMem));
2019-07-15 19:08:49 +00:00
DEBUGLOG(5, "reserving object space");
/* Statically sized space.
* entropyWorkspace never moves,
* though prev/next block swap places */
assert(ZSTD_cwksp_check_available(ws, 2 * sizeof(ZSTD_compressedBlockState_t)));
zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
2019-07-15 19:08:49 +00:00
RETURN_ERROR_IF(zc->blockState.prevCBlock == NULL, memory_allocation, "couldn't allocate prevCBlock");
zc->blockState.nextCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
2019-07-15 19:08:49 +00:00
RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate nextCBlock");
zc->entropyWorkspace = (U32*) ZSTD_cwksp_reserve_object(ws, HUF_WORKSPACE_SIZE);
2019-07-15 19:08:49 +00:00
RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate entropyWorkspace");
} }
2016-03-19 15:09:09 +00:00
ZSTD_cwksp_clear(ws);
2019-07-15 19:08:49 +00:00
/* init params */
zc->appliedParams = params;
zc->blockState.matchState.cParams = params.cParams;
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",
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
(unsigned)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);
/* ZSTD_wildcopy() is used to copy into the literals buffer,
* so we have to oversize the buffer by WILDCOPY_OVERLENGTH bytes.
*/
zc->seqStore.litStart = ZSTD_cwksp_reserve_buffer(ws, blockSize + WILDCOPY_OVERLENGTH);
zc->seqStore.maxNbLit = blockSize;
2019-08-12 18:22:54 +00:00
/* buffers */
zc->inBuffSize = buffInSize;
zc->inBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffInSize);
2019-08-12 18:22:54 +00:00
zc->outBuffSize = buffOutSize;
zc->outBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffOutSize);
2017-09-06 15:39:46 +00:00
/* ldm bucketOffsets table */
if (params.ldmParams.enableLdm) {
2019-09-10 22:02:22 +00:00
/* TODO: avoid memset? */
2017-09-06 15:39:46 +00:00
size_t const ldmBucketSize =
((size_t)1) << (params.ldmParams.hashLog -
params.ldmParams.bucketSizeLog);
zc->ldmState.bucketOffsets = ZSTD_cwksp_reserve_buffer(ws, ldmBucketSize);
2019-08-12 18:22:54 +00:00
memset(zc->ldmState.bucketOffsets, 0, ldmBucketSize);
2017-09-06 15:39:46 +00:00
}
2019-08-12 18:22:54 +00:00
/* sequences storage */
ZSTD_referenceExternalSequences(zc, NULL, 0);
2019-08-12 18:22:54 +00:00
zc->seqStore.maxNbSeq = maxNbSeq;
zc->seqStore.llCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
zc->seqStore.mlCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
zc->seqStore.ofCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
zc->seqStore.sequencesStart = (seqDef*)ZSTD_cwksp_reserve_aligned(ws, maxNbSeq * sizeof(seqDef));
2017-09-06 15:39:46 +00:00
2019-07-15 19:08:49 +00:00
FORWARD_IF_ERROR(ZSTD_reset_matchState(
&zc->blockState.matchState,
ws,
2019-07-15 19:08:49 +00:00
&params.cParams,
crp,
needsIndexReset,
ZSTD_resetTarget_CCtx));
2019-07-15 19:08:49 +00:00
2017-09-06 15:39:46 +00:00
/* ldm hash table */
if (params.ldmParams.enableLdm) {
2019-09-10 22:02:22 +00:00
/* TODO: avoid memset? */
size_t const ldmHSize = ((size_t)1) << params.ldmParams.hashLog;
zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned(ws, ldmHSize * sizeof(ldmEntry_t));
2019-07-15 19:08:49 +00:00
memset(zc->ldmState.hashTable, 0, ldmHSize * sizeof(ldmEntry_t));
zc->ldmSequences = (rawSeq*)ZSTD_cwksp_reserve_aligned(ws, maxNbLdmSeq * sizeof(rawSeq));
zc->maxNbLdmSequences = maxNbLdmSeq;
memset(&zc->ldmState.window, 0, sizeof(zc->ldmState.window));
ZSTD_window_clear(&zc->ldmState.window);
2017-09-06 15:39:46 +00:00
}
DEBUGLOG(3, "wksp: finished allocating, %zd bytes remain available", ZSTD_cwksp_available_space(ws));
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(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
}
2015-10-25 13:06:35 +00:00
/* These are the approximate sizes for each strategy past which copying the
* dictionary tables into the working context is faster than using them
* in-place.
*/
static const size_t attachDictSizeCutoffs[ZSTD_STRATEGY_MAX+1] = {
8 KB, /* unused */
8 KB, /* ZSTD_fast */
16 KB, /* ZSTD_dfast */
32 KB, /* ZSTD_greedy */
32 KB, /* ZSTD_lazy */
32 KB, /* ZSTD_lazy2 */
32 KB, /* ZSTD_btlazy2 */
32 KB, /* ZSTD_btopt */
8 KB, /* ZSTD_btultra */
8 KB /* ZSTD_btultra2 */
};
static int ZSTD_shouldAttachDict(const ZSTD_CDict* cdict,
const ZSTD_CCtx_params* params,
U64 pledgedSrcSize)
{
size_t cutoff = attachDictSizeCutoffs[cdict->matchState.cParams.strategy];
return ( pledgedSrcSize <= cutoff
|| pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
|| params->attachDictPref == ZSTD_dictForceAttach )
&& params->attachDictPref != ZSTD_dictForceCopy
&& !params->forceWindow; /* dictMatchState isn't correctly
* handled in _enforceMaxDist */
}
static size_t
ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
const ZSTD_CDict* cdict,
ZSTD_CCtx_params params,
U64 pledgedSrcSize,
ZSTD_buffered_policy_e zbuff)
{
{ const ZSTD_compressionParameters* const cdict_cParams = &cdict->matchState.cParams;
unsigned const windowLog = params.cParams.windowLog;
assert(windowLog != 0);
/* Resize working context table params for input only, since the dict
* has its own tables. */
params.cParams = ZSTD_adjustCParams_internal(*cdict_cParams, pledgedSrcSize, 0);
params.cParams.windowLog = windowLog;
FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
ZSTDcrp_makeClean, zbuff));
2018-08-23 18:33:08 +00:00
assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy);
}
{ const U32 cdictEnd = (U32)( cdict->matchState.window.nextSrc
2018-05-21 22:27:08 +00:00
- cdict->matchState.window.base);
const U32 cdictLen = cdictEnd - cdict->matchState.window.dictLimit;
2018-05-21 22:27:08 +00:00
if (cdictLen == 0) {
/* don't even attach dictionaries with no contents */
DEBUGLOG(4, "skipping attaching empty dictionary");
} else {
DEBUGLOG(4, "attaching dictionary into context");
cctx->blockState.matchState.dictMatchState = &cdict->matchState;
/* prep working match state so dict matches never have negative indices
* when they are translated to the working context's index space. */
if (cctx->blockState.matchState.window.dictLimit < cdictEnd) {
cctx->blockState.matchState.window.nextSrc =
cctx->blockState.matchState.window.base + cdictEnd;
ZSTD_window_clear(&cctx->blockState.matchState.window);
}
/* loadedDictEnd is expressed within the referential of the active context */
cctx->blockState.matchState.loadedDictEnd = cctx->blockState.matchState.window.dictLimit;
} }
cctx->dictID = cdict->dictID;
/* copy block state */
memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
return 0;
}
static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
const ZSTD_CDict* cdict,
ZSTD_CCtx_params params,
U64 pledgedSrcSize,
ZSTD_buffered_policy_e zbuff)
{
const ZSTD_compressionParameters *cdict_cParams = &cdict->matchState.cParams;
DEBUGLOG(4, "copying dictionary into context");
{ unsigned const windowLog = params.cParams.windowLog;
assert(windowLog != 0);
/* Copy only compression parameters related to tables. */
params.cParams = *cdict_cParams;
params.cParams.windowLog = windowLog;
FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
ZSTDcrp_leaveDirty, 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);
}
ZSTD_cwksp_mark_tables_dirty(&cctx->workspace);
/* 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;
2019-09-12 22:39:46 +00:00
memcpy(cctx->blockState.matchState.hashTable,
cdict->matchState.hashTable,
hSize * sizeof(U32));
memcpy(cctx->blockState.matchState.chainTable,
cdict->matchState.chainTable,
chainSize * sizeof(U32));
}
/* Zero the hashTable3, since the cdict never fills it */
{ int const h3log = cctx->blockState.matchState.hashLog3;
size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
assert(cdict->matchState.hashLog3 == 0);
memset(cctx->blockState.matchState.hashTable3, 0, h3Size * sizeof(U32));
}
ZSTD_cwksp_mark_tables_clean(&cctx->workspace);
/* copy dictionary offsets */
{ ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
ZSTD_matchState_t* dstMatchState = &cctx->blockState.matchState;
dstMatchState->window = srcMatchState->window;
dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
}
cctx->dictID = cdict->dictID;
/* copy block state */
memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
return 0;
}
/* We have a choice between copying the dictionary context into the working
* context, or referencing the dictionary context from the working context
* in-place. We decide here which strategy to use. */
static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
const ZSTD_CDict* cdict,
const ZSTD_CCtx_params* params,
U64 pledgedSrcSize,
ZSTD_buffered_policy_e zbuff)
{
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(4, "ZSTD_resetCCtx_usingCDict (pledgedSrcSize=%u)",
(unsigned)pledgedSrcSize);
if (ZSTD_shouldAttachDict(cdict, params, pledgedSrcSize)) {
return ZSTD_resetCCtx_byAttachingCDict(
cctx, cdict, *params, pledgedSrcSize, zbuff);
} else {
return ZSTD_resetCCtx_byCopyingCDict(
cctx, cdict, *params, pledgedSrcSize, zbuff);
}
}
/*! 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");
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(srcCCtx->stage!=ZSTDcs_init, 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_leaveDirty, 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
ZSTD_cwksp_mark_tables_dirty(&dstCCtx->workspace);
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;
int const h3log = srcCCtx->blockState.matchState.hashLog3;
size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
2019-09-12 22:39:46 +00:00
memcpy(dstCCtx->blockState.matchState.hashTable,
srcCCtx->blockState.matchState.hashTable,
hSize * sizeof(U32));
memcpy(dstCCtx->blockState.matchState.chainTable,
srcCCtx->blockState.matchState.chainTable,
chainSize * sizeof(U32));
memcpy(dstCCtx->blockState.matchState.hashTable3,
srcCCtx->blockState.matchState.hashTable3,
h3Size * sizeof(U32));
2016-03-19 16:18:00 +00:00
}
2016-01-26 14:58:49 +00:00
ZSTD_cwksp_mark_tables_clean(&dstCCtx->workspace);
2016-05-29 03:01:04 +00:00
/* copy dictionary offsets */
2017-12-13 01:37:06 +00:00
{
const ZSTD_matchState_t* srcMatchState = &srcCCtx->blockState.matchState;
2018-01-12 20:06:10 +00:00
ZSTD_matchState_t* dstMatchState = &dstCCtx->blockState.matchState;
dstMatchState->window = srcMatchState->window;
dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
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() :
* reduce table indexes by `reducerValue`, or squash to zero.
* PreserveMark preserves "unsorted mark" for btlazy2 strategy.
* It must be set to a clear 0/1 value, to remove branch during inlining.
* Presume table size is a multiple of ZSTD_ROWSIZE
* to help auto-vectorization */
FORCE_INLINE_TEMPLATE void
ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerValue, int const preserveMark)
{
int const nbRows = (int)size / ZSTD_ROWSIZE;
int cellNb = 0;
int rowNb;
assert((size & (ZSTD_ROWSIZE-1)) == 0); /* multiple of ZSTD_ROWSIZE */
assert(size < (1U<<31)); /* can be casted to int */
#if defined (MEMORY_SANITIZER) && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
/* To validate that the table re-use logic is sound, and that we don't
* access table space that we haven't cleaned, we re-"poison" the table
* space every time we mark it dirty.
*
* This function however is intended to operate on those dirty tables and
* re-clean them. So when this function is used correctly, we can unpoison
* the memory it operated on. This introduces a blind spot though, since
* if we now try to operate on __actually__ poisoned memory, we will not
* detect that. */
__msan_unpoison(table, size * sizeof(U32));
#endif
for (rowNb=0 ; rowNb < nbRows ; rowNb++) {
int column;
for (column=0; column<ZSTD_ROWSIZE; column++) {
if (preserveMark) {
U32 const adder = (table[cellNb] == ZSTD_DUBT_UNSORTED_MARK) ? reducerValue : 0;
table[cellNb] += adder;
}
if (table[cellNb] < reducerValue) table[cellNb] = 0;
else table[cellNb] -= reducerValue;
cellNb++;
} }
}
static void ZSTD_reduceTable(U32* const table, U32 const size, U32 const reducerValue)
{
ZSTD_reduceTable_internal(table, size, reducerValue, 0);
}
static void ZSTD_reduceTable_btlazy2(U32* const table, U32 const size, U32 const reducerValue)
{
ZSTD_reduceTable_internal(table, size, reducerValue, 1);
}
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_matchState_t* ms, ZSTD_CCtx_params const* params, const U32 reducerValue)
2016-03-20 15:20:06 +00:00
{
{ U32 const hSize = (U32)1 << params->cParams.hashLog;
ZSTD_reduceTable(ms->hashTable, hSize, reducerValue);
}
2016-03-20 15:20:06 +00:00
if (params->cParams.strategy != ZSTD_fast) {
U32 const chainSize = (U32)1 << params->cParams.chainLog;
if (params->cParams.strategy == ZSTD_btlazy2)
ZSTD_reduceTable_btlazy2(ms->chainTable, chainSize, reducerValue);
else
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);
}
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
static size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
{
U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(srcSize << 3);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
dstSize_tooSmall);
MEM_writeLE24(dst, cBlockHeader24);
memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
return ZSTD_blockHeaderSize + srcSize;
}
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;
assert(nbSeq <= seqStorePtr->maxNbSeq);
2016-07-29 19:22:17 +00:00
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;
}
static int ZSTD_disableLiteralsCompression(const ZSTD_CCtx_params* cctxParams)
{
switch (cctxParams->literalCompressionMode) {
case ZSTD_lcm_huffman:
return 0;
case ZSTD_lcm_uncompressed:
return 1;
default:
assert(0 /* impossible: pre-validated */);
/* fall-through */
case ZSTD_lcm_auto:
return (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
}
}
/* ZSTD_compressSequences_internal():
* actually compresses both literals and sequences */
MEM_STATIC size_t
ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
const ZSTD_entropyCTables_t* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
void* dst, size_t dstCapacity,
void* entropyWorkspace, size_t entropyWkspSize,
const int bmi2)
2015-11-11 20:38:21 +00:00
{
const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
ZSTD_strategy const strategy = cctxParams->cParams.strategy;
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
unsigned count[MaxSeq+1];
2018-05-22 23:06:33 +00:00
FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
FSE_CTable* CTable_MatchLength = nextEntropy->fse.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;
size_t const nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
2015-11-11 20:38:21 +00:00
BYTE* seqHead;
BYTE* lastNCount = NULL;
2019-07-03 20:40:37 +00:00
DEBUGLOG(5, "ZSTD_compressSequences_internal (nbSeq=%zu)", nbSeq);
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;
size_t const litSize = (size_t)(seqStorePtr->lit - literals);
size_t const cSize = ZSTD_compressLiterals(
2018-05-22 23:06:33 +00:00
&prevEntropy->huf, &nextEntropy->huf,
cctxParams->cParams.strategy,
ZSTD_disableLiteralsCompression(cctxParams),
op, dstCapacity,
literals, litSize,
entropyWorkspace, entropyWkspSize,
bmi2);
FORWARD_IF_ERROR(cSize);
assert(cSize <= dstCapacity);
2015-11-11 20:38:21 +00:00
op += cSize;
}
/* Sequences Header */
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
dstSize_tooSmall);
if (nbSeq < 128) {
*op++ = (BYTE)nbSeq;
} else if (nbSeq < LONGNBSEQ) {
op[0] = (BYTE)((nbSeq>>8) + 0x80);
op[1] = (BYTE)nbSeq;
op+=2;
} else {
op[0]=0xFF;
MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ));
op+=3;
}
assert(op <= oend);
if (nbSeq==0) {
2018-05-22 23:06:33 +00:00
/* Copy the old tables over as if we repeated them */
memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
return (size_t)(op - ostart);
}
2015-11-11 20:38:21 +00:00
/* seqHead : flags for FSE encoding type */
seqHead = op++;
assert(op <= oend);
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 */
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
{ unsigned max = MaxLL;
size_t const mostFrequent = HIST_countFast_wksp(count, &max, llCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */
DEBUGLOG(5, "Building LL table");
2018-05-22 23:06:33 +00:00
nextEntropy->fse.litlength_repeatMode = prevEntropy->fse.litlength_repeatMode;
LLtype = ZSTD_selectEncodingType(&nextEntropy->fse.litlength_repeatMode,
count, max, mostFrequent, nbSeq,
LLFSELog, prevEntropy->fse.litlengthCTable,
LL_defaultNorm, LL_defaultNormLog,
ZSTD_defaultAllowed, strategy);
assert(set_basic < set_compressed && set_rle < set_compressed);
2018-05-22 23:06:33 +00:00
assert(!(LLtype < set_compressed && nextEntropy->fse.litlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
{ size_t const countSize = ZSTD_buildCTable(
op, (size_t)(oend - op),
CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
count, max, llCodeTable, nbSeq,
LL_defaultNorm, LL_defaultNormLog, MaxLL,
prevEntropy->fse.litlengthCTable,
sizeof(prevEntropy->fse.litlengthCTable),
entropyWorkspace, entropyWkspSize);
FORWARD_IF_ERROR(countSize);
if (LLtype == set_compressed)
lastNCount = op;
op += countSize;
assert(op <= oend);
} }
/* build CTable for Offsets */
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
{ unsigned max = MaxOff;
size_t const mostFrequent = HIST_countFast_wksp(
count, &max, ofCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */
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");
2018-05-22 23:06:33 +00:00
nextEntropy->fse.offcode_repeatMode = prevEntropy->fse.offcode_repeatMode;
Offtype = ZSTD_selectEncodingType(&nextEntropy->fse.offcode_repeatMode,
count, max, mostFrequent, nbSeq,
OffFSELog, prevEntropy->fse.offcodeCTable,
OF_defaultNorm, OF_defaultNormLog,
defaultPolicy, strategy);
2018-05-22 23:06:33 +00:00
assert(!(Offtype < set_compressed && nextEntropy->fse.offcode_repeatMode != FSE_repeat_none)); /* We don't copy tables */
{ size_t const countSize = ZSTD_buildCTable(
op, (size_t)(oend - op),
CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
count, max, ofCodeTable, nbSeq,
OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
prevEntropy->fse.offcodeCTable,
sizeof(prevEntropy->fse.offcodeCTable),
entropyWorkspace, entropyWkspSize);
FORWARD_IF_ERROR(countSize);
if (Offtype == set_compressed)
lastNCount = op;
op += countSize;
assert(op <= oend);
} }
/* build CTable for MatchLengths */
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
{ unsigned max = MaxML;
size_t const mostFrequent = HIST_countFast_wksp(
count, &max, mlCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */
2018-11-06 01:50:30 +00:00
DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op));
2018-05-22 23:06:33 +00:00
nextEntropy->fse.matchlength_repeatMode = prevEntropy->fse.matchlength_repeatMode;
MLtype = ZSTD_selectEncodingType(&nextEntropy->fse.matchlength_repeatMode,
count, max, mostFrequent, nbSeq,
MLFSELog, prevEntropy->fse.matchlengthCTable,
ML_defaultNorm, ML_defaultNormLog,
ZSTD_defaultAllowed, strategy);
2018-05-22 23:06:33 +00:00
assert(!(MLtype < set_compressed && nextEntropy->fse.matchlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
{ size_t const countSize = ZSTD_buildCTable(
op, (size_t)(oend - op),
CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
count, max, mlCodeTable, nbSeq,
ML_defaultNorm, ML_defaultNormLog, MaxML,
prevEntropy->fse.matchlengthCTable,
sizeof(prevEntropy->fse.matchlengthCTable),
entropyWorkspace, entropyWkspSize);
FORWARD_IF_ERROR(countSize);
if (MLtype == set_compressed)
lastNCount = op;
op += countSize;
assert(op <= oend);
} }
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, (size_t)(oend - op),
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq,
2018-02-15 03:20:32 +00:00
longOffsets, bmi2);
FORWARD_IF_ERROR(bitstreamSize);
op += bitstreamSize;
assert(op <= oend);
/* zstd versions <= 1.3.4 mistakenly report corruption when
Spelling (#1582) * spelling: accidentally * spelling: across * spelling: additionally * spelling: addresses * spelling: appropriate * spelling: assumed * spelling: available * spelling: builder * spelling: capacity * spelling: compiler * spelling: compressibility * spelling: compressor * spelling: compression * spelling: contract * spelling: convenience * spelling: decompress * spelling: description * spelling: deflate * spelling: deterministically * spelling: dictionary * spelling: display * spelling: eliminate * spelling: preemptively * spelling: exclude * spelling: failure * spelling: independence * spelling: independent * spelling: intentionally * spelling: matching * spelling: maximum * spelling: meaning * spelling: mishandled * spelling: memory * spelling: occasionally * spelling: occurrence * spelling: official * spelling: offsets * spelling: original * spelling: output * spelling: overflow * spelling: overridden * spelling: parameter * spelling: performance * spelling: probability * spelling: receives * spelling: redundant * spelling: recompression * spelling: resources * spelling: sanity * spelling: segment * spelling: series * spelling: specified * spelling: specify * spelling: subtracted * spelling: successful * spelling: return * spelling: translation * spelling: update * spelling: unrelated * spelling: useless * spelling: variables * spelling: variety * spelling: verbatim * spelling: verification * spelling: visited * spelling: warming * spelling: workers * spelling: with
2019-04-12 18:18:11 +00:00
* FSE_readNCount() receives a buffer < 4 bytes.
* Fixed by https://github.com/facebook/zstd/pull/1146.
* This can happen when the last set_compressed table present is 2
* bytes and the bitstream is only one byte.
* In this exceedingly rare case, we will simply emit an uncompressed
* block, since it isn't worth optimizing.
*/
if (lastNCount && (op - lastNCount) < 4) {
/* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
assert(op - lastNCount == 3);
DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
"emitting an uncompressed block.");
return 0;
}
}
2015-11-11 20:38:21 +00:00
DEBUGLOG(5, "compressed block size : %u", (unsigned)(op - ostart));
return (size_t)(op - ostart);
}
2015-11-11 20:38:21 +00:00
MEM_STATIC size_t
ZSTD_compressSequences(seqStore_t* seqStorePtr,
const ZSTD_entropyCTables_t* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
void* dst, size_t dstCapacity,
size_t srcSize,
void* entropyWorkspace, size_t entropyWkspSize,
int bmi2)
{
size_t const cSize = ZSTD_compressSequences_internal(
seqStorePtr, prevEntropy, nextEntropy, cctxParams,
dst, dstCapacity,
entropyWorkspace, entropyWkspSize, bmi2);
if (cSize == 0) return 0;
/* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block.
* Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block.
*/
if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity))
return 0; /* block not compressed */
FORWARD_IF_ERROR(cSize);
/* Check compressibility */
{ size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, cctxParams->cParams.strategy);
if (cSize >= maxCSize) return 0; /* block not compressed */
}
2015-11-11 20:38:21 +00:00
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 */
ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_dictMode_e dictMode)
2015-10-31 11:57:14 +00:00
{
static const ZSTD_blockCompressor blockCompressor[3][ZSTD_STRATEGY_MAX+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_btultra2 },
{ 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_compressBlock_btultra_extDict },
{ ZSTD_compressBlock_fast_dictMatchState /* default for 0 */,
ZSTD_compressBlock_fast_dictMatchState,
ZSTD_compressBlock_doubleFast_dictMatchState,
ZSTD_compressBlock_greedy_dictMatchState,
ZSTD_compressBlock_lazy_dictMatchState,
ZSTD_compressBlock_lazy2_dictMatchState,
ZSTD_compressBlock_btlazy2_dictMatchState,
2018-06-13 21:10:37 +00:00
ZSTD_compressBlock_btopt_dictMatchState,
ZSTD_compressBlock_btultra_dictMatchState,
2018-06-13 21:10:37 +00:00
ZSTD_compressBlock_btultra_dictMatchState }
};
ZSTD_blockCompressor selectedCompressor;
ZSTD_STATIC_ASSERT((unsigned)ZSTD_fast == 1);
assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, strat));
selectedCompressor = blockCompressor[(int)dictMode][(int)strat];
assert(selectedCompressor != NULL);
return selectedCompressor;
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
void ZSTD_resetSeqStore(seqStore_t* ssPtr)
{
ssPtr->lit = ssPtr->litStart;
ssPtr->sequences = ssPtr->sequencesStart;
ssPtr->longLengthID = 0;
}
typedef enum { ZSTDbss_compress, ZSTDbss_noCompress } ZSTD_buildSeqStore_e;
static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
2015-11-04 11:05:27 +00:00
{
ZSTD_matchState_t* const ms = &zc->blockState.matchState;
DEBUGLOG(5, "ZSTD_buildSeqStore (srcSize=%zu)", srcSize);
assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
/* Assert that we have correctly flushed the ctx params into the ms's copy */
ZSTD_assertEqualCParams(zc->appliedParams.cParams, ms->cParams);
if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.minMatch);
return ZSTDbss_noCompress; /* don't even attempt compression below a certain srcSize */
}
ZSTD_resetSeqStore(&(zc->seqStore));
2019-02-15 18:29:03 +00:00
/* required for optimal parser to read stats from dictionary */
ms->opt.symbolCosts = &zc->blockState.prevCBlock->entropy;
/* tell the optimal parser how we expect to compress literals */
ms->opt.literalCompressionMode = zc->appliedParams.literalCompressionMode;
/* a gap between an attached dict and the current window is not safe,
* they must remain adjacent,
* and when that stops being the case, the dict must be unset */
assert(ms->dictMatchState == NULL || ms->loadedDictEnd == ms->window.dictLimit);
/* limited update after a very long match */
{ const BYTE* const base = ms->window.base;
const BYTE* const istart = (const BYTE*)src;
const U32 current = (U32)(istart-base);
if (sizeof(ptrdiff_t)==8) assert(istart - base < (ptrdiff_t)(U32)(-1)); /* ensure no overflow */
if (current > ms->nextToUpdate + 384)
ms->nextToUpdate = current - MIN(192, (U32)(current - ms->nextToUpdate - 384));
}
/* select and store sequences */
{ ZSTD_dictMode_e const dictMode = ZSTD_matchState_dictMode(ms);
size_t lastLLSize;
{ int i;
for (i = 0; i < ZSTD_REP_NUM; ++i)
zc->blockState.nextCBlock->rep[i] = zc->blockState.prevCBlock->rep[i];
}
if (zc->externSeqStore.pos < zc->externSeqStore.size) {
assert(!zc->appliedParams.ldmParams.enableLdm);
/* Updates ldmSeqStore.pos */
lastLLSize =
ZSTD_ldm_blockCompress(&zc->externSeqStore,
ms, &zc->seqStore,
zc->blockState.nextCBlock->rep,
src, srcSize);
assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
} else if (zc->appliedParams.ldmParams.enableLdm) {
rawSeqStore_t ldmSeqStore = {NULL, 0, 0, 0};
ldmSeqStore.seq = zc->ldmSequences;
ldmSeqStore.capacity = zc->maxNbLdmSequences;
/* Updates ldmSeqStore.size */
FORWARD_IF_ERROR(ZSTD_ldm_generateSequences(&zc->ldmState, &ldmSeqStore,
&zc->appliedParams.ldmParams,
src, srcSize));
/* Updates ldmSeqStore.pos */
lastLLSize =
ZSTD_ldm_blockCompress(&ldmSeqStore,
ms, &zc->seqStore,
zc->blockState.nextCBlock->rep,
src, srcSize);
assert(ldmSeqStore.pos == ldmSeqStore.size);
} else { /* not long range mode */
ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy, dictMode);
lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
}
{ const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
ZSTD_storeLastLiterals(&zc->seqStore, lastLiterals, lastLLSize);
} }
return ZSTDbss_compress;
}
/* Returns true if the given block is a RLE block */
static int ZSTD_isRLE(const BYTE *ip, size_t length) {
size_t i;
if (length < 2) return 1;
for (i = 1; i < length; ++i) {
if (ip[0] != ip[i]) return 0;
}
return 1;
}
static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize, U32 frame)
{
/* This the upper bound for the length of an rle block.
* This isn't the actual upper bound. Finding the real threshold
* needs further investigation.
*/
2019-08-26 16:00:22 +00:00
const U32 rleMaxLength = 25;
size_t cSize;
2019-08-22 20:46:15 +00:00
const BYTE* ip = (const BYTE*)src;
BYTE* op = (BYTE*)dst;
DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
(unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
(unsigned)zc->blockState.matchState.nextToUpdate);
{ const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
FORWARD_IF_ERROR(bss);
if (bss == ZSTDbss_noCompress) { cSize = 0; goto out; }
}
/* encode sequences and literals */
2018-09-28 19:09:28 +00:00
cSize = ZSTD_compressSequences(&zc->seqStore,
&zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
dst, dstCapacity,
srcSize,
zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
zc->bmi2);
2018-09-28 19:09:28 +00:00
if (frame &&
/* We don't want to emit our first block as a RLE even if it qualifies because
* doing so will cause the decoder (cli only) to throw a "should consume all input error."
* This is only an issue for zstd <= v1.4.3
*/
!zc->isFirstBlock &&
cSize < rleMaxLength &&
ZSTD_isRLE(ip, srcSize))
{
cSize = 1;
2019-08-22 20:46:15 +00:00
op[0] = ip[0];
}
2018-09-28 19:09:28 +00:00
out:
if (!ZSTD_isError(cSize) && cSize > 1) {
2018-09-28 19:09:28 +00:00
/* confirm repcodes and entropy tables when emitting a compressed block */
ZSTD_compressedBlockState_t* const tmp = zc->blockState.prevCBlock;
zc->blockState.prevCBlock = zc->blockState.nextCBlock;
zc->blockState.nextCBlock = tmp;
}
2018-09-28 19:09:28 +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 (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
return cSize;
2015-11-04 11:05:27 +00:00
}
static void ZSTD_overflowCorrectIfNeeded(ZSTD_matchState_t* ms,
ZSTD_cwksp* ws,
ZSTD_CCtx_params const* params,
void const* ip,
void const* iend)
{
if (ZSTD_window_needOverflowCorrection(ms->window, iend)) {
U32 const maxDist = (U32)1 << params->cParams.windowLog;
U32 const cycleLog = ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy);
U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, maxDist, ip);
ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
ZSTD_cwksp_mark_tables_dirty(ws);
ZSTD_reduceIndex(ms, params, correction);
ZSTD_cwksp_mark_tables_clean(ws);
if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
else ms->nextToUpdate -= correction;
/* invalidate dictionaries on overflow correction */
ms->loadedDictEnd = 0;
ms->dictMatchState = NULL;
}
}
/*! 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 <= ZSTD_WINDOWLOG_MAX);
2015-11-01 11:40:22 +00:00
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(5, "ZSTD_compress_frameChunk (blockSize=%u)", (unsigned)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
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE,
dstSize_tooSmall,
"not enough space to store compressed block");
2015-11-04 17:19:39 +00:00
if (remaining < blockSize) blockSize = remaining;
ZSTD_overflowCorrectIfNeeded(
ms, &cctx->workspace, &cctx->appliedParams, ip, ip + blockSize);
ZSTD_checkDictValidity(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
/* Ensure hash/chain table insertion resumes no sooner than lowlimit */
if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
{ size_t cSize = ZSTD_compressBlock_internal(cctx,
op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize,
ip, blockSize, 1 /* frame */);
FORWARD_IF_ERROR(cSize);
if (cSize == 0) { /* block is not compressible */
cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
FORWARD_IF_ERROR(cSize);
} else {
const U32 cBlockHeader = cSize == 1 ?
lastBlock + (((U32)bt_rle)<<1) + (U32)(blockSize << 3) :
lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
2019-08-22 20:46:15 +00:00
MEM_writeLE24(op, cBlockHeader);
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;
cctx->isFirstBlock = 0;
DEBUGLOG(5, "ZSTD_compress_frameChunk: adding a block of size %u",
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
(unsigned)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;
return (size_t)(op-ostart);
2015-10-22 14:31:46 +00:00
}
2016-04-12 13:52:33 +00:00
static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
const 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 */
U32 const checksumFlag = params->fParams.checksumFlag>0;
U32 const windowSize = (U32)1 << params->cParams.windowLog;
U32 const singleSegment = params->fParams.contentSizeFlag && (windowSize >= pledgedSrcSize);
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 */
Spelling (#1582) * spelling: accidentally * spelling: across * spelling: additionally * spelling: addresses * spelling: appropriate * spelling: assumed * spelling: available * spelling: builder * spelling: capacity * spelling: compiler * spelling: compressibility * spelling: compressor * spelling: compression * spelling: contract * spelling: convenience * spelling: decompress * spelling: description * spelling: deflate * spelling: deterministically * spelling: dictionary * spelling: display * spelling: eliminate * spelling: preemptively * spelling: exclude * spelling: failure * spelling: independence * spelling: independent * spelling: intentionally * spelling: matching * spelling: maximum * spelling: meaning * spelling: mishandled * spelling: memory * spelling: occasionally * spelling: occurrence * spelling: official * spelling: offsets * spelling: original * spelling: output * spelling: overflow * spelling: overridden * spelling: parameter * spelling: performance * spelling: probability * spelling: receives * spelling: redundant * spelling: recompression * spelling: resources * spelling: sanity * spelling: segment * spelling: series * spelling: specified * spelling: specify * spelling: subtracted * spelling: successful * spelling: return * spelling: translation * spelling: update * spelling: unrelated * spelling: useless * spelling: variables * spelling: variety * spelling: verbatim * spelling: verification * spelling: visited * spelling: warming * spelling: workers * spelling: with
2019-04-12 18:18:11 +00:00
BYTE const frameHeaderDescriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
size_t pos=0;
2016-05-29 03:01:04 +00:00
assert(!(params->fParams.contentSizeFlag && pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN));
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(dstCapacity < ZSTD_FRAMEHEADERSIZE_MAX, dstSize_tooSmall);
DEBUGLOG(4, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
!params->fParams.noDictIDFlag, (unsigned)dictID, (unsigned)dictIDSizeCode);
2016-04-12 13:52:33 +00:00
if (params->format == ZSTD_f_zstd1) {
MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
pos = 4;
}
Spelling (#1582) * spelling: accidentally * spelling: across * spelling: additionally * spelling: addresses * spelling: appropriate * spelling: assumed * spelling: available * spelling: builder * spelling: capacity * spelling: compiler * spelling: compressibility * spelling: compressor * spelling: compression * spelling: contract * spelling: convenience * spelling: decompress * spelling: description * spelling: deflate * spelling: deterministically * spelling: dictionary * spelling: display * spelling: eliminate * spelling: preemptively * spelling: exclude * spelling: failure * spelling: independence * spelling: independent * spelling: intentionally * spelling: matching * spelling: maximum * spelling: meaning * spelling: mishandled * spelling: memory * spelling: occasionally * spelling: occurrence * spelling: official * spelling: offsets * spelling: original * spelling: output * spelling: overflow * spelling: overridden * spelling: parameter * spelling: performance * spelling: probability * spelling: receives * spelling: redundant * spelling: recompression * spelling: resources * spelling: sanity * spelling: segment * spelling: series * spelling: specified * spelling: specify * spelling: subtracted * spelling: successful * spelling: return * spelling: translation * spelling: update * spelling: unrelated * spelling: useless * spelling: variables * spelling: variety * spelling: verbatim * spelling: verification * spelling: visited * spelling: warming * spelling: workers * spelling: with
2019-04-12 18:18:11 +00:00
op[pos++] = frameHeaderDescriptionByte;
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))
Spelling (#1582) * spelling: accidentally * spelling: across * spelling: additionally * spelling: addresses * spelling: appropriate * spelling: assumed * spelling: available * spelling: builder * spelling: capacity * spelling: compiler * spelling: compressibility * spelling: compressor * spelling: compression * spelling: contract * spelling: convenience * spelling: decompress * spelling: description * spelling: deflate * spelling: deterministically * spelling: dictionary * spelling: display * spelling: eliminate * spelling: preemptively * spelling: exclude * spelling: failure * spelling: independence * spelling: independent * spelling: intentionally * spelling: matching * spelling: maximum * spelling: meaning * spelling: mishandled * spelling: memory * spelling: occasionally * spelling: occurrence * spelling: official * spelling: offsets * spelling: original * spelling: output * spelling: overflow * spelling: overridden * spelling: parameter * spelling: performance * spelling: probability * spelling: receives * spelling: redundant * spelling: recompression * spelling: resources * spelling: sanity * spelling: segment * spelling: series * spelling: specified * spelling: specify * spelling: subtracted * spelling: successful * spelling: return * spelling: translation * spelling: update * spelling: unrelated * spelling: useless * spelling: variables * spelling: variety * spelling: verbatim * spelling: verification * spelling: visited * spelling: warming * spelling: workers * spelling: with
2019-04-12 18:18:11 +00:00
* or an error code if `dstCapacity` is too small (<ZSTD_blockHeaderSize)
*/
size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity)
{
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall);
{ U32 const cBlockHeader24 = 1 /*lastBlock*/ + (((U32)bt_raw)<<1); /* 0 size */
MEM_writeLE24(dst, cBlockHeader24);
return ZSTD_blockHeaderSize;
}
}
size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq)
{
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->stage != ZSTDcs_init, stage_wrong);
RETURN_ERROR_IF(cctx->appliedParams.ldmParams.enableLdm,
parameter_unsupported);
cctx->externSeqStore.seq = seq;
cctx->externSeqStore.size = nbSeq;
cctx->externSeqStore.capacity = nbSeq;
cctx->externSeqStore.pos = 0;
return 0;
}
2016-04-12 13:52:33 +00:00
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
{
ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
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",
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
cctx->stage, (unsigned)srcSize);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->stage==ZSTDcs_created, 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);
FORWARD_IF_ERROR(fhSize);
assert(fhSize <= dstCapacity);
2016-04-12 13:52:33 +00:00
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 */
if (!ZSTD_window_update(&ms->window, src, srcSize)) {
ms->nextToUpdate = ms->window.dictLimit;
}
if (cctx->appliedParams.ldmParams.enableLdm) {
ZSTD_window_update(&cctx->ldmState.window, src, srcSize);
}
if (!frame) {
/* overflow check and correction for block mode */
ZSTD_overflowCorrectIfNeeded(
ms, &cctx->workspace, &cctx->appliedParams,
src, (BYTE const*)src + srcSize);
}
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(5, "ZSTD_compressContinue_internal (blockSize=%u)", (unsigned)cctx->blockSize);
{ size_t const cSize = frame ?
ZSTD_compress_frameChunk (cctx, dst, dstCapacity, src, srcSize, lastFrameChunk) :
ZSTD_compressBlock_internal (cctx, dst, dstCapacity, src, srcSize, 0 /* frame */);
FORWARD_IF_ERROR(cSize);
cctx->consumedSrcSize += srcSize;
cctx->producedCSize += (cSize + fhSize);
assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
if (cctx->pledgedSrcSizePlusOne != 0) { /* control src size */
ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(
cctx->consumedSrcSize+1 > cctx->pledgedSrcSizePlusOne,
srcSize_wrong,
"error : pledgedSrcSize = %u, while realSrcSize >= %u",
(unsigned)cctx->pledgedSrcSizePlusOne-1,
(unsigned)cctx->consumedSrcSize);
}
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)
{
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(5, "ZSTD_compressContinue (srcSize=%u)", (unsigned)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
{
2018-03-20 23:35:14 +00:00
ZSTD_compressionParameters const cParams = cctx->appliedParams.cParams;
assert(!ZSTD_checkCParams(cParams));
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)
{
DEBUGLOG(5, "ZSTD_compressBlock: srcSize = %u", (unsigned)srcSize);
{ size_t const blockSizeMax = ZSTD_getBlockSize(cctx);
RETURN_ERROR_IF(srcSize > blockSizeMax, 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_cwksp* ws,
ZSTD_CCtx_params const* params,
const void* src, size_t srcSize,
ZSTD_dictTableLoadMethod_e dtlm)
{
const BYTE* ip = (const BYTE*) src;
const BYTE* const iend = ip + srcSize;
ZSTD_window_update(&ms->window, src, srcSize);
ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->window.base);
/* Assert that we the ms params match the params we're being given */
ZSTD_assertEqualCParams(params->cParams, ms->cParams);
2016-07-27 19:05:12 +00:00
if (srcSize <= HASH_READ_SIZE) return 0;
while (iend - ip > HASH_READ_SIZE) {
size_t const remaining = (size_t)(iend - ip);
size_t const chunk = MIN(remaining, ZSTD_CHUNKSIZE_MAX);
const BYTE* const ichunk = ip + chunk;
2016-07-12 07:47:31 +00:00
ZSTD_overflowCorrectIfNeeded(ms, ws, params, ip, ichunk);
switch(params->cParams.strategy)
{
case ZSTD_fast:
ZSTD_fillHashTable(ms, ichunk, dtlm);
break;
case ZSTD_dfast:
ZSTD_fillDoubleHashTable(ms, ichunk, dtlm);
break;
case ZSTD_greedy:
case ZSTD_lazy:
case ZSTD_lazy2:
if (chunk >= HASH_READ_SIZE)
ZSTD_insertAndFindFirstIndex(ms, ichunk-HASH_READ_SIZE);
break;
case ZSTD_btlazy2: /* we want the dictionary table fully sorted */
case ZSTD_btopt:
case ZSTD_btultra:
case ZSTD_btultra2:
if (chunk >= HASH_READ_SIZE)
ZSTD_updateTree(ms, ichunk-HASH_READ_SIZE, ichunk);
break;
default:
assert(0); /* not possible : not a valid strategy id */
}
ip = ichunk;
}
ms->nextToUpdate = (U32)(iend - ms->window.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;
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(dictMaxSymbolValue < maxSymbolValue, dictionary_corrupted);
for (s = 0; s <= maxSymbolValue; ++s) {
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(normalizedCounter[s] == 0, 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
*/
static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
ZSTD_matchState_t* ms,
ZSTD_cwksp* ws,
ZSTD_CCtx_params const* params,
const void* dict, size_t dictSize,
ZSTD_dictTableLoadMethod_e dtlm,
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)));
assert(dictSize > 8);
assert(MEM_readLE32(dictPtr) == ZSTD_MAGIC_DICTIONARY);
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;
2018-05-22 23:06:33 +00:00
size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.huf.CTable, &maxSymbolValue, dictPtr, dictEnd-dictPtr);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(HUF_isError(hufHeaderSize), dictionary_corrupted);
RETURN_ERROR_IF(maxSymbolValue < 255, 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);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted);
RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted);
/* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
/* fill all offset symbols to avoid garbage at end of table */
RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
bs->entropy.fse.offcodeCTable,
offcodeNCount, MaxOff, 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);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted);
RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted);
/* Every match length code must have non-zero probability */
FORWARD_IF_ERROR( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
bs->entropy.fse.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);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted);
RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted);
/* Every literal length code must have non-zero probability */
FORWARD_IF_ERROR( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
bs->entropy.fse.litlengthCTable,
litlengthNCount, litlengthMaxValue, litlengthLog,
workspace, HUF_WORKSPACE_SIZE)),
dictionary_corrupted);
dictPtr += litlengthHeaderSize;
2016-05-31 16:13:56 +00:00
}
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(dictPtr+12 > dictEnd, 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 */
FORWARD_IF_ERROR(ZSTD_checkDictNCount(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff)));
/* All repCodes must be <= dictContentSize and != 0*/
{ U32 u;
for (u=0; u<3; u++) {
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(bs->rep[u] == 0, dictionary_corrupted);
RETURN_ERROR_IF(bs->rep[u] > dictContentSize, dictionary_corrupted);
} }
2018-05-22 23:06:33 +00:00
bs->entropy.huf.repeatMode = HUF_repeat_valid;
bs->entropy.fse.offcode_repeatMode = FSE_repeat_valid;
bs->entropy.fse.matchlength_repeatMode = FSE_repeat_valid;
bs->entropy.fse.litlength_repeatMode = FSE_repeat_valid;
FORWARD_IF_ERROR(ZSTD_loadDictionaryContent(
ms, ws, params, dictPtr, dictContentSize, dtlm));
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_cwksp* ws,
const ZSTD_CCtx_params* params,
const void* dict, size_t dictSize,
ZSTD_dictContentType_e dictContentType,
ZSTD_dictTableLoadMethod_e dtlm,
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 (dictContentType == ZSTD_dct_rawContent)
return ZSTD_loadDictionaryContent(ms, ws, params, dict, dictSize, dtlm);
2016-03-15 00:24:33 +00:00
if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
if (dictContentType == ZSTD_dct_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, ws, params, dict, dictSize, dtlm);
}
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(dictContentType == ZSTD_dct_fullDict, dictionary_wrong);
assert(0); /* impossible */
}
/* dict as full zstd dictionary */
return ZSTD_loadZstdDictionary(
bs, ms, ws, params, dict, dictSize, dtlm, 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 */
2018-09-27 22:13:43 +00:00
static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
const void* dict, size_t dictSize,
ZSTD_dictContentType_e dictContentType,
ZSTD_dictTableLoadMethod_e dtlm,
const ZSTD_CDict* cdict,
const ZSTD_CCtx_params* params, U64 pledgedSrcSize,
2018-09-27 22:13:43 +00:00
ZSTD_buffered_policy_e zbuff)
2015-10-22 14:31:46 +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) {
return ZSTD_resetCCtx_usingCDict(cctx, cdict, params, pledgedSrcSize, zbuff);
}
FORWARD_IF_ERROR( ZSTD_resetCCtx_internal(cctx, *params, pledgedSrcSize,
ZSTDcrp_makeClean, zbuff) );
{ size_t const dictID = ZSTD_compress_insertDictionary(
cctx->blockState.prevCBlock, &cctx->blockState.matchState,
&cctx->workspace, params, dict, dictSize, dictContentType, dtlm,
cctx->entropyWorkspace);
FORWARD_IF_ERROR(dictID);
assert(dictID <= UINT_MAX);
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_dictContentType_e dictContentType,
ZSTD_dictTableLoadMethod_e dtlm,
const ZSTD_CDict* cdict,
const ZSTD_CCtx_params* params,
2017-08-18 23:17:24 +00:00
unsigned long long pledgedSrcSize)
{
DEBUGLOG(4, "ZSTD_compressBegin_advanced_internal: wlog=%u", params->cParams.windowLog);
2017-08-18 23:17:24 +00:00
/* compression parameters verification and optimization */
FORWARD_IF_ERROR( ZSTD_checkCParams(params->cParams) );
return ZSTD_compressBegin_internal(cctx,
dict, dictSize, dictContentType, dtlm,
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_dct_auto, ZSTD_dtlm_fast,
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 =
ZSTD_assignParamsToCCtxParams(&cctx->requestedParams, params);
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (unsigned)dictSize);
return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, 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");
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->stage == ZSTDcs_created, 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);
FORWARD_IF_ERROR(fhSize);
2016-04-12 13:52:33 +00:00
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;
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall);
2016-07-27 22:55:43 +00:00
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);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall);
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(4, "ZSTD_writeEpilogue: write checksum : %08X", (unsigned)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 */);
FORWARD_IF_ERROR(cSize);
2016-07-28 13:29:08 +00:00
endResult = ZSTD_writeEpilogue(cctx, (char*)dst + cSize, dstCapacity-cSize);
FORWARD_IF_ERROR(endResult);
assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
if (cctx->pledgedSrcSizePlusOne != 0) { /* control src size */
ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
DEBUGLOG(4, "end of frame : controlling src size");
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(
cctx->pledgedSrcSizePlusOne != cctx->consumedSrcSize+1,
srcSize_wrong,
"error : pledgedSrcSize = %u, while realSrcSize = %u",
(unsigned)cctx->pledgedSrcSizePlusOne-1,
(unsigned)cctx->consumedSrcSize);
}
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,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const void* dict,size_t dictSize,
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* cctx,
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");
FORWARD_IF_ERROR(ZSTD_checkCParams(params.cParams));
return ZSTD_compress_internal(cctx,
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,
const ZSTD_CCtx_params* params)
2017-08-18 23:17:24 +00:00
{
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(4, "ZSTD_compress_advanced_internal (srcSize:%u)", (unsigned)srcSize);
FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
params, srcSize, ZSTDb_not_buffered) );
2017-08-18 23:17:24 +00:00
return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
}
size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx,
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 const params = ZSTD_getParams(compressionLevel, srcSize + (!srcSize), dict ? dictSize : 0);
ZSTD_CCtx_params cctxParams = ZSTD_assignParamsToCCtxParams(&cctx->requestedParams, params);
assert(params.fParams.contentSizeFlag == 1);
return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, dict, dictSize, &cctxParams);
2015-12-18 00:26:48 +00:00
}
size_t ZSTD_compressCCtx(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
int compressionLevel)
2015-10-25 13:06:35 +00:00
{
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(4, "ZSTD_compressCCtx (srcSize=%u)", (unsigned)srcSize);
assert(cctx != NULL);
return ZSTD_compress_usingDict(cctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
2015-10-25 13:06:35 +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;
ZSTD_initCCtx(&ctxBody, ZSTD_defaultCMem);
2016-03-15 00:24:33 +00:00
result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
ZSTD_freeCCtxContent(&ctxBody); /* 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)
{
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(5, "sizeof(ZSTD_CDict) : %u", (unsigned)sizeof(ZSTD_CDict));
return ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
+ ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
+ ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0)
+ (dictLoadMethod == ZSTD_dlm_byRef ? 0
: ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void *))));
}
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 */
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(5, "sizeof(*cdict) : %u", (unsigned)sizeof(*cdict));
/* cdict may be in the workspace */
return (cdict->workspace.workspace == cdict ? 0 : sizeof(*cdict))
+ ZSTD_cwksp_sizeof(&cdict->workspace);
}
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_dictContentType_e dictContentType,
2017-05-26 01:05:49 +00:00
ZSTD_compressionParameters cParams)
{
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(3, "ZSTD_initCDict_internal (dictContentType:%u)", (unsigned)dictContentType);
assert(!ZSTD_checkCParams(cParams));
2018-08-23 18:33:08 +00:00
cdict->matchState.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->dictContent = dictBuffer;
} else {
2019-08-15 16:51:24 +00:00
void *internalBuffer = ZSTD_cwksp_reserve_object(&cdict->workspace, ZSTD_cwksp_align(dictSize, sizeof(void*)));
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(!internalBuffer, memory_allocation);
cdict->dictContent = internalBuffer;
memcpy(internalBuffer, dictBuffer, dictSize);
2017-05-26 01:05:49 +00:00
}
cdict->dictContentSize = dictSize;
2019-08-15 16:51:24 +00:00
cdict->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cdict->workspace, HUF_WORKSPACE_SIZE);
2019-07-15 19:08:49 +00:00
/* Reset the state to no dictionary */
ZSTD_reset_compressedBlockState(&cdict->cBlockState);
2019-07-15 19:08:49 +00:00
FORWARD_IF_ERROR(ZSTD_reset_matchState(
&cdict->matchState,
&cdict->workspace,
&cParams,
ZSTDcrp_makeClean,
ZSTDirp_reset,
ZSTD_resetTarget_CDict));
/* (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, &cdict->workspace,
&params, cdict->dictContent, cdict->dictContentSize,
2019-07-15 19:08:49 +00:00
dictContentType, ZSTD_dtlm_full, cdict->entropyWorkspace);
FORWARD_IF_ERROR(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_dictContentType_e dictContentType,
ZSTD_compressionParameters cParams, ZSTD_customMem customMem)
{
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(3, "ZSTD_createCDict_advanced, mode %u", (unsigned)dictContentType);
if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
{ size_t const workspaceSize =
ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict)) +
ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE) +
ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0) +
(dictLoadMethod == ZSTD_dlm_byRef ? 0
: ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void*))));
void* const workspace = ZSTD_malloc(workspaceSize, customMem);
2019-08-15 16:51:24 +00:00
ZSTD_cwksp ws;
ZSTD_CDict* cdict;
if (!workspace) {
ZSTD_free(workspace, customMem);
return NULL;
}
2019-08-15 16:51:24 +00:00
ZSTD_cwksp_init(&ws, workspace, workspaceSize);
2019-08-15 16:51:24 +00:00
cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
assert(cdict != NULL);
ZSTD_cwksp_move(&cdict->workspace, &ws);
cdict->customMem = customMem;
2017-05-26 01:05:49 +00:00
if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
dictBuffer, dictSize,
dictLoadMethod, dictContentType,
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,
ZSTD_dlm_byCopy, ZSTD_dct_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,
ZSTD_dlm_byRef, ZSTD_dct_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;
/* Only free workspace if cdict not in workspace, otherwise the
* workspace will be freed when the cdict itself is freed. */
if ((void*)cdict->workspace.workspace != (void*)cdict) {
2019-08-15 16:51:24 +00:00
ZSTD_cwksp_free(&cdict->workspace, 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_dictContentType_e dictContentType,
2017-05-26 01:05:49 +00:00
ZSTD_compressionParameters cParams)
{
size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0);
size_t const neededSize = ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
+ (dictLoadMethod == ZSTD_dlm_byRef ? 0
: ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void*))))
+ ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
+ matchStateSize;
2019-07-15 19:08:49 +00:00
ZSTD_CDict* cdict;
2017-05-26 01:05:49 +00:00
if ((size_t)workspace & 7) return NULL; /* 8-aligned */
2019-07-15 19:08:49 +00:00
{
2019-08-15 16:51:24 +00:00
ZSTD_cwksp ws;
ZSTD_cwksp_init(&ws, workspace, workspaceSize);
cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
2019-07-15 19:08:49 +00:00
if (cdict == NULL) return NULL;
ZSTD_cwksp_move(&cdict->workspace, &ws);
2019-07-15 19:08:49 +00:00
}
DEBUGLOG(4, "(workspaceSize < neededSize) : (%u < %u) => %u",
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
(unsigned)workspaceSize, (unsigned)neededSize, (unsigned)(workspaceSize < neededSize));
2017-05-26 01:05:49 +00:00
if (workspaceSize < neededSize) return NULL;
if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
dict, dictSize,
dictLoadMethod, dictContentType,
cParams) ))
2017-05-26 01:05:49 +00:00
return NULL;
return cdict;
}
ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict)
{
assert(cdict != NULL);
2018-08-23 18:33:08 +00:00
return cdict->matchState.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");
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cdict==NULL, 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_dct_auto, ZSTD_dtlm_fast,
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, ZSTD_CONTENTSIZE_UNKNOWN);
}
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)
{
FORWARD_IF_ERROR(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* cctx,
const void* const dict, size_t const dictSize, ZSTD_dictContentType_e const dictContentType,
const ZSTD_CDict* const cdict,
ZSTD_CCtx_params params, unsigned long long const pledgedSrcSize)
2016-09-15 12:54:07 +00:00
{
DEBUGLOG(4, "ZSTD_resetCStream_internal");
/* Finalize the compression parameters */
params.cParams = ZSTD_getCParamsFromCCtxParams(&params, pledgedSrcSize, dictSize);
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 */
FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
dict, dictSize, dictContentType, ZSTD_dtlm_fast,
cdict,
&params, pledgedSrcSize,
ZSTDb_buffered) );
2016-09-15 12:54:07 +00:00
cctx->inToCompress = 0;
cctx->inBuffPos = 0;
cctx->inBuffTarget = cctx->blockSize
+ (cctx->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 */
cctx->outBuffContentSize = cctx->outBuffFlushedSize = 0;
cctx->streamStage = zcss_load;
cctx->frameEnded = 0;
2016-09-15 12:54:07 +00:00
return 0; /* ready to go */
}
/* ZSTD_resetCStream():
* pledgedSrcSize == 0 means "unknown" */
2019-03-13 21:05:18 +00:00
size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pss)
{
2019-03-13 21:05:18 +00:00
/* 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.
*/
U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(4, "ZSTD_resetCStream: pledgedSrcSize = %u", (unsigned)pledgedSrcSize);
2019-03-13 21:05:18 +00:00
FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) );
FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) );
return 0;
}
/*! 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,
const ZSTD_CCtx_params* params,
unsigned long long pledgedSrcSize)
2016-08-11 23:20:36 +00:00
{
DEBUGLOG(4, "ZSTD_initCStream_internal");
2019-03-13 21:05:18 +00:00
FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) );
FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) );
assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
zcs->requestedParams = *params;
assert(!((dict) && (cdict))); /* either dict or cdict, not both */
2019-03-13 21:05:18 +00:00
if (dict) {
FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) );
} else {
2019-03-13 21:05:18 +00:00
/* Dictionary is cleared if !cdict */
FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) );
2016-08-11 23:20:36 +00:00
}
2019-03-13 21:05:18 +00:00
return 0;
}
/* 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");
2019-03-13 21:05:18 +00:00
FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) );
FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) );
zcs->requestedParams.fParams = fParams;
FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) );
return 0;
}
/* note : cdict must outlive compression session */
size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict)
{
DEBUGLOG(4, "ZSTD_initCStream_usingCDict");
2019-03-13 21:05:18 +00:00
FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) );
FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) );
return 0;
2016-08-11 23:20:36 +00:00
}
/* ZSTD_initCStream_advanced() :
* pledgedSrcSize must be exact.
2017-12-08 17:16:49 +00:00
* 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,
2019-03-13 21:05:18 +00:00
ZSTD_parameters params, unsigned long long pss)
{
2019-03-13 21:05:18 +00:00
/* for compatibility with older programs relying on this behavior.
* Users should now specify ZSTD_CONTENTSIZE_UNKNOWN.
* This line will be removed in the future.
*/
U64 const pledgedSrcSize = (pss==0 && params.fParams.contentSizeFlag==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
DEBUGLOG(4, "ZSTD_initCStream_advanced");
FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) );
FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) );
FORWARD_IF_ERROR( ZSTD_checkCParams(params.cParams) );
zcs->requestedParams = ZSTD_assignParamsToCCtxParams(&zcs->requestedParams, params);
2019-03-13 21:05:18 +00:00
FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) );
return 0;
}
2016-08-11 23:20:36 +00:00
size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel)
{
2019-03-13 21:05:18 +00:00
DEBUGLOG(4, "ZSTD_initCStream_usingDict");
FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) );
FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) );
FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) );
return 0;
2016-08-11 23:20:36 +00:00
}
size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pss)
{
2019-03-13 21:05:18 +00:00
/* 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.
*/
U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
DEBUGLOG(4, "ZSTD_initCStream_srcSize");
FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) );
FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, NULL) );
FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) );
FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) );
return 0;
}
2016-08-11 23:20:36 +00:00
size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
{
DEBUGLOG(4, "ZSTD_initCStream");
2019-03-13 21:05:18 +00:00
FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) );
FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, NULL) );
FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) );
return 0;
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
static size_t ZSTD_nextInputSizeHint(const ZSTD_CCtx* cctx)
{
size_t hintInSize = cctx->inBuffTarget - cctx->inBuffPos;
if (hintInSize==0) hintInSize = cctx->blockSize;
return hintInSize;
}
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
* non-static, because can be called from zstdmt_compress.c
* @return : hint size for next input */
2019-03-13 21:05:18 +00:00
static 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 */
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%u", (unsigned)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:
2019-01-28 22:05:18 +00:00
RETURN_ERROR(init_missing, "call ZSTD_initCStream() first!");
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);
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(4, "ZSTD_compressEnd : cSize=%u", (unsigned)cSize);
FORWARD_IF_ERROR(cSize);
ip = iend;
op += cSize;
zcs->frameEnded = 1;
ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
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);
FORWARD_IF_ERROR(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",
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
(unsigned)zcs->inBuffTarget, (unsigned)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;
ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
}
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, (size_t)(oend-op),
zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
DEBUGLOG(5, "toFlush: %u into %u ==> flushed: %u",
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
(unsigned)toFlush, (unsigned)(oend-op), (unsigned)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;
ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
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;
return ZSTD_nextInputSizeHint(zcs);
}
static size_t ZSTD_nextInputSizeHint_MTorST(const ZSTD_CCtx* cctx)
{
#ifdef ZSTD_MULTITHREAD
if (cctx->appliedParams.nbWorkers >= 1) {
assert(cctx->mtctx != NULL);
return ZSTDMT_nextInputSizeHint(cctx->mtctx);
2016-08-11 23:20:36 +00:00
}
#endif
return ZSTD_nextInputSizeHint(cctx);
2016-08-11 23:20:36 +00:00
}
2016-08-16 23:39:22 +00:00
size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
{
FORWARD_IF_ERROR( ZSTD_compressStream2(zcs, output, input, ZSTD_e_continue) );
return ZSTD_nextInputSizeHint_MTorST(zcs);
2016-08-11 23:20:36 +00:00
}
2017-06-06 01:32:48 +00:00
size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
ZSTD_outBuffer* output,
ZSTD_inBuffer* input,
ZSTD_EndDirective endOp)
{
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(5, "ZSTD_compressStream2, endOp=%u ", (unsigned)endOp);
/* check conditions */
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(output->pos > output->size, GENERIC);
RETURN_ERROR_IF(input->pos > input->size, 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;
FORWARD_IF_ERROR( ZSTD_initLocalDict(cctx) ); /* Init the local dict if present. */
memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict)); /* single usage */
assert(prefixDict.dict==NULL || cctx->cdict==NULL); /* only one can be set */
DEBUGLOG(4, "ZSTD_compressStream2 : 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.nbWorkers = 0; /* do not invoke multi-threading when src size is too small */
}
if (params.nbWorkers > 0) {
/* mt context creation */
if (cctx->mtctx == NULL) {
DEBUGLOG(4, "ZSTD_compressStream2: creating new mtctx for nbWorkers=%u",
params.nbWorkers);
cctx->mtctx = ZSTDMT_createCCtx_advanced(params.nbWorkers, cctx->customMem);
2019-01-28 22:05:18 +00:00
RETURN_ERROR_IF(cctx->mtctx == NULL, memory_allocation);
2017-08-29 23:18:21 +00:00
}
/* mt compression */
DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbWorkers=%u", params.nbWorkers);
FORWARD_IF_ERROR( ZSTDMT_initCStream_internal(
cctx->mtctx,
prefixDict.dict, prefixDict.dictSize, ZSTD_dct_rawContent,
cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) );
2017-06-12 01:32:36 +00:00
cctx->streamStage = zcss_load;
cctx->appliedParams.nbWorkers = params.nbWorkers;
} else
#endif
{ FORWARD_IF_ERROR( ZSTD_resetCStream_internal(cctx,
prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType,
cctx->cdict,
params, cctx->pledgedSrcSizePlusOne-1) );
assert(cctx->streamStage == zcss_load);
assert(cctx->appliedParams.nbWorkers == 0);
2017-06-06 01:32:48 +00:00
} }
/* end of transparent initialization stage */
2017-06-06 01:32:48 +00:00
/* compression stage */
#ifdef ZSTD_MULTITHREAD
if (cctx->appliedParams.nbWorkers > 0) {
int const forceMaxProgress = (endOp == ZSTD_e_flush || endOp == ZSTD_e_end);
size_t flushMin;
assert(forceMaxProgress || endOp == ZSTD_e_continue /* Protection for a new flush type */);
if (cctx->cParamsChanged) {
ZSTDMT_updateCParams_whileCompressing(cctx->mtctx, &cctx->requestedParams);
cctx->cParamsChanged = 0;
}
do {
flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
if ( ZSTD_isError(flushMin)
|| (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
}
FORWARD_IF_ERROR(flushMin);
} while (forceMaxProgress && flushMin != 0 && output->pos < output->size);
DEBUGLOG(5, "completed ZSTD_compressStream2 delegating to ZSTDMT_compressStream_generic");
/* Either we don't require maximum forward progress, we've finished the
* flush, or we are out of output space.
*/
assert(!forceMaxProgress || flushMin == 0 || output->pos == output->size);
return flushMin;
}
#endif
FORWARD_IF_ERROR( ZSTD_compressStream_generic(cctx, output, input, endOp) );
DEBUGLOG(5, "completed ZSTD_compressStream2");
return cctx->outBuffContentSize - cctx->outBuffFlushedSize; /* remaining to flush */
}
size_t ZSTD_compressStream2_simpleArgs (
ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity, size_t* dstPos,
const void* src, size_t srcSize, size_t* srcPos,
ZSTD_EndDirective endOp)
{
ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
ZSTD_inBuffer input = { src, srcSize, *srcPos };
/* ZSTD_compressStream2() will check validity of dstPos and srcPos */
size_t const cErr = ZSTD_compressStream2(cctx, &output, &input, endOp);
*dstPos = output.pos;
*srcPos = input.pos;
return cErr;
2016-08-11 23:20:36 +00:00
}
size_t ZSTD_compress2(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize)
{
ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
{ size_t oPos = 0;
size_t iPos = 0;
size_t const result = ZSTD_compressStream2_simpleArgs(cctx,
dst, dstCapacity, &oPos,
src, srcSize, &iPos,
ZSTD_e_end);
FORWARD_IF_ERROR(result);
if (result != 0) { /* compression not completed, due to lack of output space */
assert(oPos == dstCapacity);
2019-01-28 22:05:18 +00:00
RETURN_ERROR(dstSize_tooSmall);
}
assert(iPos == srcSize); /* all input is expected consumed */
return oPos;
}
}
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 };
return ZSTD_compressStream2(zcs, output, &input, ZSTD_e_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 };
size_t const remainingToFlush = ZSTD_compressStream2(zcs, output, &input, ZSTD_e_end);
FORWARD_IF_ERROR( remainingToFlush );
if (zcs->appliedParams.nbWorkers > 0) return remainingToFlush; /* minimal estimation */
/* single thread mode : attempt to calculate remaining to flush more precisely */
{ size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
size_t const checksumSize = (size_t)(zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4);
size_t const toFlush = remainingToFlush + lastBlockSize + checksumSize;
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
DEBUGLOG(4, "ZSTD_endStream : remaining to flush : %u", (unsigned)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; }
int ZSTD_minCLevel(void) { return (int)-ZSTD_TARGETLENGTH_MAX; }
static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
{ /* "default" - for any srcSize > 256 KB */
2016-04-09 18:32:00 +00:00
/* W, C, H, S, L, TL, strat */
{ 19, 12, 13, 1, 6, 1, ZSTD_fast }, /* base for negative levels */
{ 19, 13, 14, 1, 7, 0, ZSTD_fast }, /* level 1 */
2018-12-10 06:38:05 +00:00
{ 20, 15, 16, 1, 6, 0, ZSTD_fast }, /* level 2 */
{ 21, 16, 17, 1, 5, 1, ZSTD_dfast }, /* level 3 */
{ 21, 18, 18, 1, 5, 1, ZSTD_dfast }, /* level 4 */
{ 21, 18, 19, 2, 5, 2, ZSTD_greedy }, /* level 5 */
{ 21, 19, 19, 3, 5, 4, ZSTD_greedy }, /* level 6 */
{ 21, 19, 19, 3, 5, 8, ZSTD_lazy }, /* level 7 */
{ 21, 19, 19, 3, 5, 16, ZSTD_lazy2 }, /* level 8 */
{ 21, 19, 20, 4, 5, 16, ZSTD_lazy2 }, /* level 9 */
2018-12-10 06:38:05 +00:00
{ 22, 20, 21, 4, 5, 16, ZSTD_lazy2 }, /* level 10 */
{ 22, 21, 22, 4, 5, 16, ZSTD_lazy2 }, /* level 11 */
{ 22, 21, 22, 5, 5, 16, ZSTD_lazy2 }, /* level 12 */
{ 22, 21, 22, 5, 5, 32, ZSTD_btlazy2 }, /* level 13 */
{ 22, 22, 23, 5, 5, 32, ZSTD_btlazy2 }, /* level 14 */
{ 22, 23, 23, 6, 5, 32, ZSTD_btlazy2 }, /* level 15 */
{ 22, 22, 22, 5, 5, 48, ZSTD_btopt }, /* level 16 */
{ 23, 23, 22, 5, 4, 64, ZSTD_btopt }, /* level 17 */
{ 23, 23, 22, 6, 3, 64, ZSTD_btultra }, /* level 18 */
{ 23, 24, 22, 7, 3,256, ZSTD_btultra2}, /* level 19 */
{ 25, 25, 23, 7, 3,256, ZSTD_btultra2}, /* level 20 */
{ 26, 26, 24, 7, 3,512, ZSTD_btultra2}, /* level 21 */
{ 27, 27, 25, 9, 3,999, ZSTD_btultra2}, /* level 22 */
},
{ /* for srcSize <= 256 KB */
/* W, C, H, S, L, T, strat */
{ 18, 12, 13, 1, 5, 1, ZSTD_fast }, /* base for negative levels */
{ 18, 13, 14, 1, 6, 0, ZSTD_fast }, /* level 1 */
{ 18, 14, 14, 1, 5, 1, ZSTD_dfast }, /* level 2 */
{ 18, 16, 16, 1, 4, 1, ZSTD_dfast }, /* level 3 */
{ 18, 16, 17, 2, 5, 2, ZSTD_greedy }, /* level 4.*/
{ 18, 18, 18, 3, 5, 2, ZSTD_greedy }, /* level 5.*/
{ 18, 18, 19, 3, 5, 4, ZSTD_lazy }, /* level 6.*/
{ 18, 18, 19, 4, 4, 4, ZSTD_lazy }, /* level 7 */
{ 18, 18, 19, 4, 4, 8, ZSTD_lazy2 }, /* level 8 */
{ 18, 18, 19, 5, 4, 8, ZSTD_lazy2 }, /* level 9 */
{ 18, 18, 19, 6, 4, 8, ZSTD_lazy2 }, /* level 10 */
2018-12-09 05:40:08 +00:00
{ 18, 18, 19, 5, 4, 12, ZSTD_btlazy2 }, /* level 11.*/
{ 18, 19, 19, 7, 4, 12, ZSTD_btlazy2 }, /* level 12.*/
{ 18, 18, 19, 4, 4, 16, ZSTD_btopt }, /* level 13 */
{ 18, 18, 19, 4, 3, 32, ZSTD_btopt }, /* level 14.*/
{ 18, 18, 19, 6, 3,128, ZSTD_btopt }, /* level 15.*/
{ 18, 19, 19, 6, 3,128, ZSTD_btultra }, /* level 16.*/
{ 18, 19, 19, 8, 3,256, ZSTD_btultra }, /* level 17.*/
{ 18, 19, 19, 6, 3,128, ZSTD_btultra2}, /* level 18.*/
{ 18, 19, 19, 8, 3,256, ZSTD_btultra2}, /* level 19.*/
{ 18, 19, 19, 10, 3,512, ZSTD_btultra2}, /* level 20.*/
{ 18, 19, 19, 12, 3,512, ZSTD_btultra2}, /* level 21.*/
{ 18, 19, 19, 13, 3,999, ZSTD_btultra2}, /* level 22.*/
},
{ /* for srcSize <= 128 KB */
/* W, C, H, S, L, T, strat */
2018-05-14 00:15:07 +00:00
{ 17, 12, 12, 1, 5, 1, ZSTD_fast }, /* base for negative levels */
{ 17, 12, 13, 1, 6, 0, ZSTD_fast }, /* level 1 */
{ 17, 13, 15, 1, 5, 0, ZSTD_fast }, /* level 2 */
{ 17, 15, 16, 2, 5, 1, ZSTD_dfast }, /* level 3 */
{ 17, 17, 17, 2, 4, 1, ZSTD_dfast }, /* level 4 */
{ 17, 16, 17, 3, 4, 2, ZSTD_greedy }, /* level 5 */
{ 17, 17, 17, 3, 4, 4, ZSTD_lazy }, /* level 6 */
2018-05-14 00:15:07 +00:00
{ 17, 17, 17, 3, 4, 8, ZSTD_lazy2 }, /* level 7 */
2016-07-22 12:36:46 +00:00
{ 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 */
2018-12-08 18:42:55 +00:00
{ 17, 17, 17, 5, 4, 8, ZSTD_btlazy2 }, /* level 11 */
{ 17, 18, 17, 7, 4, 12, ZSTD_btlazy2 }, /* level 12 */
{ 17, 18, 17, 3, 4, 12, ZSTD_btopt }, /* level 13.*/
{ 17, 18, 17, 4, 3, 32, ZSTD_btopt }, /* level 14.*/
{ 17, 18, 17, 6, 3,256, ZSTD_btopt }, /* level 15.*/
{ 17, 18, 17, 6, 3,128, ZSTD_btultra }, /* level 16.*/
{ 17, 18, 17, 8, 3,256, ZSTD_btultra }, /* level 17.*/
{ 17, 18, 17, 10, 3,512, ZSTD_btultra }, /* level 18.*/
{ 17, 18, 17, 5, 3,256, ZSTD_btultra2}, /* level 19.*/
{ 17, 18, 17, 7, 3,512, ZSTD_btultra2}, /* level 20.*/
{ 17, 18, 17, 9, 3,512, ZSTD_btultra2}, /* level 21.*/
{ 17, 18, 17, 11, 3,999, ZSTD_btultra2}, /* level 22.*/
},
{ /* for srcSize <= 16 KB */
/* W, C, H, S, L, T, strat */
{ 14, 12, 13, 1, 5, 1, ZSTD_fast }, /* base for negative levels */
{ 14, 14, 15, 1, 5, 0, ZSTD_fast }, /* level 1 */
{ 14, 14, 15, 1, 4, 0, ZSTD_fast }, /* level 2 */
{ 14, 14, 15, 2, 4, 1, ZSTD_dfast }, /* level 3 */
{ 14, 14, 14, 4, 4, 2, ZSTD_greedy }, /* level 4 */
{ 14, 14, 14, 3, 4, 4, ZSTD_lazy }, /* level 5.*/
{ 14, 14, 14, 4, 4, 8, ZSTD_lazy2 }, /* level 6 */
{ 14, 14, 14, 6, 4, 8, ZSTD_lazy2 }, /* level 7 */
{ 14, 14, 14, 8, 4, 8, ZSTD_lazy2 }, /* level 8.*/
{ 14, 15, 14, 5, 4, 8, ZSTD_btlazy2 }, /* level 9.*/
{ 14, 15, 14, 9, 4, 8, ZSTD_btlazy2 }, /* level 10.*/
{ 14, 15, 14, 3, 4, 12, ZSTD_btopt }, /* level 11.*/
{ 14, 15, 14, 4, 3, 24, ZSTD_btopt }, /* level 12.*/
{ 14, 15, 14, 5, 3, 32, ZSTD_btultra }, /* level 13.*/
{ 14, 15, 15, 6, 3, 64, ZSTD_btultra }, /* level 14.*/
{ 14, 15, 15, 7, 3,256, ZSTD_btultra }, /* level 15.*/
{ 14, 15, 15, 5, 3, 48, ZSTD_btultra2}, /* level 16.*/
{ 14, 15, 15, 6, 3,128, ZSTD_btultra2}, /* level 17.*/
{ 14, 15, 15, 7, 3,256, ZSTD_btultra2}, /* level 18.*/
{ 14, 15, 15, 8, 3,256, ZSTD_btultra2}, /* level 19.*/
{ 14, 15, 15, 8, 3,512, ZSTD_btultra2}, /* level 20.*/
{ 14, 15, 15, 9, 3,512, ZSTD_btultra2}, /* level 21.*/
{ 14, 15, 15, 10, 3,999, ZSTD_btultra2}, /* 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 : ZSTD_CONTENTSIZE_UNKNOWN; /* intentional overflow for srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN */
U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB);
int row = compressionLevel;
DEBUGLOG(5, "ZSTD_getCParams (cLevel=%i)", compressionLevel);
if (compressionLevel == 0) row = ZSTD_CLEVEL_DEFAULT; /* 0 == default */
if (compressionLevel < 0) row = 0; /* entry 0 is baseline for fast mode */
if (compressionLevel > ZSTD_MAX_CLEVEL) row = ZSTD_MAX_CLEVEL;
{ ZSTD_compressionParameters cp = ZSTD_defaultCParameters[tableID][row];
if (compressionLevel < 0) cp.targetLength = (unsigned)(-compressionLevel); /* acceleration factor */
return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize); /* refine parameters based on srcSize & dictSize */
}
}
/*! ZSTD_getParams() :
* same idea as ZSTD_getCParams()
* @return a `ZSTD_parameters` structure (instead of `ZSTD_compressionParameters`).
* Fields of `ZSTD_frameParameters` are set to default values */
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);
DEBUGLOG(5, "ZSTD_getParams (cLevel=%i)", compressionLevel);
memset(&params, 0, sizeof(params));
params.cParams = cParams;
params.fParams.contentSizeFlag = 1;
return params;
}