AuroraMiddleware/zstd
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge pull request #1824 from senhuang42/new_path_for_cdict

Browse Source 
Avoid using CDict params when input is large.
This commit is contained in:
Yann Collet 2019-10-23 12:04:40 -07:00 committed by GitHub
commit a9a216a846
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 45 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
lib/compress/zstd_compress.c
View File

@ -50,6 +50,7 @@ struct ZSTD_CDict_s {
ZSTD_compressedBlockState_t cBlockState;
ZSTD_customMem customMem;
U32 dictID;
int compressionLevel; /* 0 indicates that advanced API was used to select CDict params */
}; /* typedef'd to ZSTD_CDict within "zstd.h" */
ZSTD_CCtx* ZSTD_createCCtx(void)
@ -387,7 +388,7 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
case ZSTD_c_forceAttachDict:
ZSTD_STATIC_ASSERT(ZSTD_dictDefaultAttach < ZSTD_dictForceCopy);
bounds.lowerBound = ZSTD_dictDefaultAttach;
bounds.upperBound = ZSTD_dictForceCopy; /* note : how to ensure at compile time that this is the highest value enum ? */
bounds.upperBound = ZSTD_dictForceLoad; /* note : how to ensure at compile time that this is the highest value enum ? */
return bounds;
case ZSTD_c_literalCompressionMode:
@ -2912,6 +2913,9 @@ ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
bs, ms, ws, params, dict, dictSize, dtlm, workspace);
}
#define ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF (128 KB)
#define ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER (6)
/*! ZSTD_compressBegin_internal() :
* @return : 0, or an error code */
static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
@ -2926,17 +2930,26 @@ static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
/* 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) {
if ( (cdict)
&& (cdict->dictContentSize > 0)
&& ( pledgedSrcSize < ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF
|| pledgedSrcSize < cdict->dictContentSize * ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER
|| cdict->compressionLevel == 0)
&& (params->attachDictPref != ZSTD_dictForceLoad) ) {
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);
{ size_t const dictID = cdict ?
ZSTD_compress_insertDictionary(
cctx->blockState.prevCBlock, &cctx->blockState.matchState,
&cctx->workspace, params, cdict->dictContent, cdict->dictContentSize,
dictContentType, dtlm, cctx->entropyWorkspace)
: 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;
@ -3254,6 +3267,8 @@ ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
assert(cdict != NULL);
ZSTD_cwksp_move(&cdict->workspace, &ws);
cdict->customMem = customMem;
cdict->compressionLevel = 0; /* signals advanced API usage */
if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
dictBuffer, dictSize,
dictLoadMethod, dictContentType,
@ -3269,9 +3284,12 @@ ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
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);
ZSTD_CDict* cdict = ZSTD_createCDict_advanced(dict, dictSize,
ZSTD_dlm_byCopy, ZSTD_dct_auto,
cParams, ZSTD_defaultCMem);
if (cdict)
cdict->compressionLevel = compressionLevel == 0 ? ZSTD_CLEVEL_DEFAULT : compressionLevel;
return cdict;
}
ZSTD_CDict* ZSTD_createCDict_byReference(const void* dict, size_t dictSize, int compressionLevel)
@ -3361,7 +3379,14 @@ size_t ZSTD_compressBegin_usingCDict_advanced(
DEBUGLOG(4, "ZSTD_compressBegin_usingCDict_advanced");
RETURN_ERROR_IF(cdict==NULL, dictionary_wrong);
{ ZSTD_CCtx_params params = cctx->requestedParams;
params.cParams = ZSTD_getCParamsFromCDict(cdict);
params.cParams = ( pledgedSrcSize < ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF
|| pledgedSrcSize < cdict->dictContentSize * ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER
|| cdict->compressionLevel == 0 )
&& (params.attachDictPref != ZSTD_dictForceLoad) ?
ZSTD_getCParamsFromCDict(cdict)
: ZSTD_getCParams(cdict->compressionLevel,
pledgedSrcSize,
cdict->dictContentSize);
/* 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.

11
lib/zstd.h
View File

@ -1152,7 +1152,7 @@ typedef enum {
* to evolve and should be considered only in the context of extremely
* advanced performance tuning.
*
* Zstd currently supports the use of a CDict in two ways:
* Zstd currently supports the use of a CDict in three ways:
*
* - The contents of the CDict can be copied into the working context. This
* means that the compression can search both the dictionary and input
@ -1167,6 +1167,12 @@ typedef enum {
* tables. However, this model incurs no start-up cost (as long as the
* working context's tables can be reused). For small inputs, this can be
* faster than copying the CDict's tables.
*
* - The CDict's tables are not used at all, and instead we use the working
* context alone to reload the dictionary and use params based on the source
* size. See ZSTD_compress_insertDictionary() and ZSTD_compress_usingDict().
* This method is effective when the dictionary sizes are very small relative
* to the input size, and the input size is fairly large to begin with.
*
* Zstd has a simple internal heuristic that selects which strategy to use
* at the beginning of a compression. However, if experimentation shows that
@ -1175,7 +1181,8 @@ typedef enum {
*/
ZSTD_dictDefaultAttach = 0, /* Use the default heuristic. */
ZSTD_dictForceAttach = 1, /* Never copy the dictionary. */
ZSTD_dictForceCopy = 2 /* Always copy the dictionary. */
ZSTD_dictForceCopy = 2, /* Always copy the dictionary. */
ZSTD_dictForceLoad = 3 /* Always reload the dictionary */
} ZSTD_dictAttachPref_e;
typedef enum {