Merge pull request #194 from ot/dev070

Expose function to add entropy tables to pre-built dictionary.
This commit is contained in:
Yann Collet 2016-05-31 08:56:22 +02:00
commit 6ae270d772
2 changed files with 60 additions and 27 deletions

View File

@ -807,6 +807,33 @@ static size_t ZDICT_fastSampling(void* dictBuffer, size_t dictSize,
return nbSegments * DIB_FASTSEGMENTSIZE;
}
size_t ZDICT_addEntropyTablesFromBuffer_advanced(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_params_t params)
{
size_t hSize;
unsigned const compressionLevel = (params.compressionLevel == 0) ? g_compressionLevel_default : params.compressionLevel;
/* dictionary header */
MEM_writeLE32(dictBuffer, ZSTD_DICT_MAGIC);
{ U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
U32 const dictID = params.dictID ? params.dictID : (U32)(randomID>>11);
MEM_writeLE32((char*)dictBuffer+4, dictID);
}
hSize = 8;
/* entropy tables */
DISPLAYLEVEL(2, "\r%70s\r", ""); /* clean display line */
DISPLAYLEVEL(2, "statistics ... \n");
hSize += ZDICT_analyzeEntropy((char*)dictBuffer+hSize, dictBufferCapacity-hSize,
compressionLevel,
samplesBuffer, samplesSizes, nbSamples,
(char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
if (hSize + dictContentSize < dictBufferCapacity)
memmove((char*)dictBuffer + hSize, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
return MIN(dictBufferCapacity, hSize+dictContentSize);
}
#define DIB_MINSAMPLESSIZE (DIB_FASTSEGMENTSIZE*3)
/*! ZDICT_trainFromBuffer_unsafe() :
@ -815,13 +842,12 @@ static size_t ZDICT_fastSampling(void* dictBuffer, size_t dictSize,
*/
size_t ZDICT_trainFromBuffer_unsafe(
void* dictBuffer, size_t maxDictSize,
const void* samplesBuffer, const size_t* sampleSizes, unsigned nbSamples,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_params_t params)
{
U32 const dictListSize = MAX( MAX(DICTLISTSIZE, nbSamples), (U32)(maxDictSize/16));
dictItem* const dictList = (dictItem*)malloc(dictListSize * sizeof(*dictList));
unsigned selectivity = params.selectivityLevel;
unsigned compressionLevel = params.compressionLevel;
size_t const targetDictSize = maxDictSize;
size_t sBuffSize;
size_t dictSize = 0;
@ -831,18 +857,17 @@ size_t ZDICT_trainFromBuffer_unsafe(
if (!dictList) return ERROR(memory_allocation);
/* init */
{ unsigned u; for (u=0, sBuffSize=0; u<nbSamples; u++) sBuffSize += sampleSizes[u]; }
{ unsigned u; for (u=0, sBuffSize=0; u<nbSamples; u++) sBuffSize += samplesSizes[u]; }
if (sBuffSize < DIB_MINSAMPLESSIZE) return 0; /* not enough source to create dictionary */
ZDICT_initDictItem(dictList);
g_displayLevel = params.notificationLevel;
if (selectivity==0) selectivity = g_selectivity_default;
if (compressionLevel==0) compressionLevel = g_compressionLevel_default;
/* build dictionary */
if (selectivity>1) { /* selectivity == 1 => fast mode */
ZDICT_trainBuffer(dictList, dictListSize,
samplesBuffer, sBuffSize,
sampleSizes, nbSamples,
samplesSizes, nbSamples,
selectivity, (U32)targetDictSize);
/* display best matches */
@ -864,7 +889,6 @@ size_t ZDICT_trainFromBuffer_unsafe(
/* create dictionary */
{ U32 dictContentSize = ZDICT_dictSize(dictList);
size_t hSize;
/* build dict content */
{ U32 u;
@ -884,25 +908,9 @@ size_t ZDICT_trainFromBuffer_unsafe(
samplesBuffer, sBuffSize);
}
/* dictionary header */
MEM_writeLE32(dictBuffer, ZSTD_DICT_MAGIC);
{ U64 const randomID = XXH64((char*)dictBuffer + maxDictSize - dictContentSize, dictContentSize, 0);
U32 const dictID = params.dictID ? params.dictID : (U32)(randomID>>11);
MEM_writeLE32((char*)dictBuffer+4, dictID);
}
hSize = 8;
/* entropic tables */
DISPLAYLEVEL(2, "\r%70s\r", ""); /* clean display line */
DISPLAYLEVEL(2, "statistics ... \n");
hSize += ZDICT_analyzeEntropy((char*)dictBuffer+hSize, maxDictSize-hSize,
compressionLevel,
samplesBuffer, sampleSizes, nbSamples,
(char*)dictBuffer + maxDictSize - dictContentSize, dictContentSize);
if (hSize + dictContentSize < maxDictSize)
memmove((char*)dictBuffer + hSize, (char*)dictBuffer + maxDictSize - dictContentSize, dictContentSize);
dictSize = MIN(maxDictSize, hSize+dictContentSize);
dictSize = ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, maxDictSize,
samplesBuffer, samplesSizes, nbSamples,
params);
}
/* clean up */
@ -946,3 +954,13 @@ size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
samplesBuffer, samplesSizes, nbSamples,
params);
}
size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
{
ZDICT_params_t params;
memset(&params, 0, sizeof(params));
return ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, dictBufferCapacity,
samplesBuffer, samplesSizes, nbSamples,
params);
}

View File

@ -52,6 +52,21 @@ extern "C" {
size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
/*! ZDICT_addEntropyTablesFromBuffer() :
Given a content-only dictionary (built for example from common strings in
the input), add entropy tables computed from the memory buffer
`samplesBuffer`, where `nbSamples` samples have been stored concatenated.
Each sample size is provided into an orderly table `samplesSizes`.
The input dictionary is the last `dictContentSize` bytes of `dictBuffer`. The
resulting dictionary with added entropy tables will written back to
`dictBuffer`.
@return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`).
*/
size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
/*-*************************************
* Helper functions