Merge pull request #2447 from senhuang42/block_splitter_v2

Recursive block splitting
This commit is contained in:
sen
2021-03-24 12:27:22 -04:00
committed by GitHub
6 changed files with 915 additions and 397 deletions
+843 -115
View File
@@ -211,6 +211,14 @@ static U32 ZSTD_CParams_shouldEnableLdm(const ZSTD_compressionParameters* const
return cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27;
}
/* Returns 1 if compression parameters are such that we should
* enable blockSplitter (wlog >= 17, strategy >= btopt).
* Returns 0 otherwise.
*/
static U32 ZSTD_CParams_useBlockSplitter(const ZSTD_compressionParameters* const cParams) {
return cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 17;
}
static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
ZSTD_compressionParameters cParams)
{
@@ -219,6 +227,7 @@ static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
ZSTD_CCtxParams_init(&cctxParams, ZSTD_CLEVEL_DEFAULT);
cctxParams.cParams = cParams;
/* Adjust advanced params according to cParams */
if (ZSTD_CParams_shouldEnableLdm(&cParams)) {
DEBUGLOG(4, "ZSTD_makeCCtxParamsFromCParams(): Including LDM into cctx params");
cctxParams.ldmParams.enableLdm = 1;
@@ -228,6 +237,11 @@ static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
assert(cctxParams.ldmParams.hashRateLog < 32);
}
if (ZSTD_CParams_useBlockSplitter(&cParams)) {
DEBUGLOG(4, "ZSTD_makeCCtxParamsFromCParams(): Including block splitting into cctx params");
cctxParams.splitBlocks = 1;
}
assert(!ZSTD_checkCParams(cParams));
return cctxParams;
}
@@ -485,6 +499,11 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
case ZSTD_c_splitBlocks:
bounds.lowerBound = 0;
bounds.upperBound = 1;
return bounds;
default:
bounds.error = ERROR(parameter_unsupported);
@@ -547,6 +566,7 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
case ZSTD_c_stableOutBuffer:
case ZSTD_c_blockDelimiters:
case ZSTD_c_validateSequences:
case ZSTD_c_splitBlocks:
default:
return 0;
}
@@ -599,6 +619,7 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
case ZSTD_c_stableOutBuffer:
case ZSTD_c_blockDelimiters:
case ZSTD_c_validateSequences:
case ZSTD_c_splitBlocks:
break;
default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
@@ -810,6 +831,11 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
CCtxParams->validateSequences = value;
return CCtxParams->validateSequences;
case ZSTD_c_splitBlocks:
BOUNDCHECK(ZSTD_c_splitBlocks, value);
CCtxParams->splitBlocks = value;
return CCtxParams->splitBlocks;
default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
}
}
@@ -933,6 +959,9 @@ size_t ZSTD_CCtxParams_getParameter(
case ZSTD_c_validateSequences :
*value = (int)CCtxParams->validateSequences;
break;
case ZSTD_c_splitBlocks :
*value = (int)CCtxParams->splitBlocks;
break;
default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
}
return 0;
@@ -2185,10 +2214,154 @@ static int ZSTD_useTargetCBlockSize(const ZSTD_CCtx_params* cctxParams)
return (cctxParams->targetCBlockSize != 0);
}
/* ZSTD_entropyCompressSequences_internal():
* actually compresses both literals and sequences */
/* ZSTD_blockSplitterEnabled():
* Returns if block splitting param is being used
* If used, compression will do best effort to split a block in order to improve compression ratio.
* Returns 1 if true, 0 otherwise. */
static int ZSTD_blockSplitterEnabled(ZSTD_CCtx_params* cctxParams)
{
DEBUGLOG(5, "ZSTD_blockSplitterEnabled(splitBlocks=%d)", cctxParams->splitBlocks);
return (cctxParams->splitBlocks != 0);
}
/* Type returned by ZSTD_buildSequencesStatistics containing finalized symbol encoding types
* and size of the sequences statistics
*/
typedef struct {
U32 LLtype;
U32 Offtype;
U32 MLtype;
size_t size;
} ZSTD_symbolEncodingTypeStats_t;
/* ZSTD_buildSequencesStatistics():
* Returns the size of the statistics for a given set of sequences, or a ZSTD error code,
* Also modifies LLtype, Offtype, MLtype, and lastNCount to the appropriate values.
*
* entropyWkspSize must be of size at least ENTROPY_WORKSPACE_SIZE - (MaxSeq + 1)*sizeof(U32)
*/
static ZSTD_symbolEncodingTypeStats_t
ZSTD_buildSequencesStatistics(seqStore_t* seqStorePtr, size_t nbSeq, size_t* lastCountSize,
const ZSTD_fseCTables_t* prevEntropy, ZSTD_fseCTables_t* nextEntropy,
BYTE* dst, const BYTE* const dstEnd,
ZSTD_strategy strategy, unsigned* countWorkspace,
void* entropyWorkspace, size_t entropyWkspSize) {
BYTE* const ostart = dst;
const BYTE* const oend = dstEnd;
BYTE* op = ostart;
FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
const BYTE* const ofCodeTable = seqStorePtr->ofCode;
const BYTE* const llCodeTable = seqStorePtr->llCode;
const BYTE* const mlCodeTable = seqStorePtr->mlCode;
ZSTD_symbolEncodingTypeStats_t stats;
/* convert length/distances into codes */
ZSTD_seqToCodes(seqStorePtr);
assert(op <= oend);
/* build CTable for Literal Lengths */
{ unsigned max = MaxLL;
size_t const mostFrequent = HIST_countFast_wksp(countWorkspace, &max, llCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */
DEBUGLOG(5, "Building LL table");
nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
stats.LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode,
countWorkspace, max, mostFrequent, nbSeq,
LLFSELog, prevEntropy->litlengthCTable,
LL_defaultNorm, LL_defaultNormLog,
ZSTD_defaultAllowed, strategy);
assert(set_basic < set_compressed && set_rle < set_compressed);
assert(!(stats.LLtype < set_compressed && nextEntropy->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)stats.LLtype,
countWorkspace, max, llCodeTable, nbSeq,
LL_defaultNorm, LL_defaultNormLog, MaxLL,
prevEntropy->litlengthCTable,
sizeof(prevEntropy->litlengthCTable),
entropyWorkspace, entropyWkspSize);
if (ZSTD_isError(countSize)) {
DEBUGLOG(3, "ZSTD_buildCTable for LitLens failed");
stats.size = countSize;
return stats;
}
if (stats.LLtype == set_compressed)
*lastCountSize = countSize;
op += countSize;
assert(op <= oend);
} }
/* build CTable for Offsets */
{ unsigned max = MaxOff;
size_t const mostFrequent = HIST_countFast_wksp(
countWorkspace, &max, ofCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */
/* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
DEBUGLOG(5, "Building OF table");
nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
stats.Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode,
countWorkspace, max, mostFrequent, nbSeq,
OffFSELog, prevEntropy->offcodeCTable,
OF_defaultNorm, OF_defaultNormLog,
defaultPolicy, strategy);
assert(!(stats.Offtype < set_compressed && nextEntropy->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)stats.Offtype,
countWorkspace, max, ofCodeTable, nbSeq,
OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
prevEntropy->offcodeCTable,
sizeof(prevEntropy->offcodeCTable),
entropyWorkspace, entropyWkspSize);
if (ZSTD_isError(countSize)) {
DEBUGLOG(3, "ZSTD_buildCTable for Offsets failed");
stats.size = countSize;
return stats;
}
if (stats.Offtype == set_compressed)
*lastCountSize = countSize;
op += countSize;
assert(op <= oend);
} }
/* build CTable for MatchLengths */
{ unsigned max = MaxML;
size_t const mostFrequent = HIST_countFast_wksp(
countWorkspace, &max, mlCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */
DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op));
nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
stats.MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode,
countWorkspace, max, mostFrequent, nbSeq,
MLFSELog, prevEntropy->matchlengthCTable,
ML_defaultNorm, ML_defaultNormLog,
ZSTD_defaultAllowed, strategy);
assert(!(stats.MLtype < set_compressed && nextEntropy->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)stats.MLtype,
countWorkspace, max, mlCodeTable, nbSeq,
ML_defaultNorm, ML_defaultNormLog, MaxML,
prevEntropy->matchlengthCTable,
sizeof(prevEntropy->matchlengthCTable),
entropyWorkspace, entropyWkspSize);
if (ZSTD_isError(countSize)) {
DEBUGLOG(3, "ZSTD_buildCTable for MatchLengths failed");
stats.size = countSize;
return stats;
}
if (stats.MLtype == set_compressed)
*lastCountSize = countSize;
op += countSize;
assert(op <= oend);
} }
stats.size = (size_t)(op-ostart);
return stats;
}
/* ZSTD_entropyCompressSeqStore_internal():
* compresses both literals and sequences
* Returns compressed size of block, or a zstd error.
*/
MEM_STATIC size_t
ZSTD_entropyCompressSequences_internal(seqStore_t* seqStorePtr,
ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
const ZSTD_entropyCTables_t* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
@@ -2202,22 +2375,20 @@ ZSTD_entropyCompressSequences_internal(seqStore_t* seqStorePtr,
FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
FSE_CTable* CTable_MatchLength = nextEntropy->fse.matchlengthCTable;
U32 LLtype, Offtype, MLtype; /* compressed, raw or rle */
const seqDef* const sequences = seqStorePtr->sequencesStart;
const size_t nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
const BYTE* const ofCodeTable = seqStorePtr->ofCode;
const BYTE* const llCodeTable = seqStorePtr->llCode;
const BYTE* const mlCodeTable = seqStorePtr->mlCode;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstCapacity;
BYTE* op = ostart;
size_t const nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
BYTE* seqHead;
BYTE* lastNCount = NULL;
size_t lastCountSize = 0;
entropyWorkspace = count + (MaxSeq + 1);
entropyWkspSize -= (MaxSeq + 1) * sizeof(*count);
DEBUGLOG(4, "ZSTD_entropyCompressSequences_internal (nbSeq=%zu)", nbSeq);
DEBUGLOG(4, "ZSTD_entropyCompressSeqStore_internal (nbSeq=%zu)", nbSeq);
ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
assert(entropyWkspSize >= HUF_WORKSPACE_SIZE);
@@ -2257,95 +2428,19 @@ ZSTD_entropyCompressSequences_internal(seqStore_t* seqStorePtr,
ZSTD_memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
return (size_t)(op - ostart);
}
/* seqHead : flags for FSE encoding type */
seqHead = op++;
assert(op <= oend);
/* convert length/distances into codes */
ZSTD_seqToCodes(seqStorePtr);
/* build CTable for Literal Lengths */
{ unsigned max = MaxLL;
size_t const mostFrequent = HIST_countFast_wksp(count, &max, llCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */
DEBUGLOG(5, "Building LL table");
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);
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, "ZSTD_buildCTable for LitLens failed");
if (LLtype == set_compressed)
lastNCount = op;
op += countSize;
assert(op <= oend);
} }
/* build CTable for Offsets */
{ unsigned max = MaxOff;
size_t const mostFrequent = HIST_countFast_wksp(
count, &max, ofCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */
/* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
DEBUGLOG(5, "Building OF table");
nextEntropy->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);
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, "ZSTD_buildCTable for Offsets failed");
if (Offtype == set_compressed)
lastNCount = op;
op += countSize;
assert(op <= oend);
} }
/* build CTable for MatchLengths */
{ unsigned max = MaxML;
size_t const mostFrequent = HIST_countFast_wksp(
count, &max, mlCodeTable, nbSeq, entropyWorkspace, entropyWkspSize); /* can't fail */
DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op));
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);
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, "ZSTD_buildCTable for MatchLengths failed");
if (MLtype == set_compressed)
lastNCount = op;
op += countSize;
assert(op <= oend);
} }
*seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
{
ZSTD_symbolEncodingTypeStats_t stats;
BYTE* seqHead = op++;
/* build stats for sequences */
stats = ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq, &lastCountSize,
&prevEntropy->fse, &nextEntropy->fse,
op, oend,
strategy, count,
entropyWorkspace, entropyWkspSize);
FORWARD_IF_ERROR(stats.size, "ZSTD_buildSequencesStatistics failed!");
*seqHead = (BYTE)((stats.LLtype<<6) + (stats.Offtype<<4) + (stats.MLtype<<2));
op += stats.size;
}
{ size_t const bitstreamSize = ZSTD_encodeSequences(
op, (size_t)(oend - op),
@@ -2365,9 +2460,9 @@ ZSTD_entropyCompressSequences_internal(seqStore_t* seqStorePtr,
* 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);
if (lastCountSize && (lastCountSize + bitstreamSize) < 4) {
/* lastCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
assert(lastCountSize + bitstreamSize == 3);
DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
"emitting an uncompressed block.");
return 0;
@@ -2379,16 +2474,16 @@ ZSTD_entropyCompressSequences_internal(seqStore_t* seqStorePtr,
}
MEM_STATIC size_t
ZSTD_entropyCompressSequences(seqStore_t* seqStorePtr,
ZSTD_entropyCompressSeqStore(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)
int bmi2, U32 const canEmitUncompressed)
{
size_t const cSize = ZSTD_entropyCompressSequences_internal(
size_t const cSize = ZSTD_entropyCompressSeqStore_internal(
seqStorePtr, prevEntropy, nextEntropy, cctxParams,
dst, dstCapacity,
entropyWorkspace, entropyWkspSize, bmi2);
@@ -2396,15 +2491,17 @@ ZSTD_entropyCompressSequences(seqStore_t* seqStorePtr,
/* 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, "ZSTD_entropyCompressSequences_internal failed");
if (canEmitUncompressed) {
if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity))
return 0; /* block not compressed */
FORWARD_IF_ERROR(cSize, "ZSTD_entropyCompressSeqStore_internal failed");
/* Check compressibility */
{ size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, cctxParams->cParams.strategy);
if (cSize >= maxCSize) return 0; /* block not compressed */
/* Check compressibility */
{ size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, cctxParams->cParams.strategy);
if (cSize >= maxCSize) return 0; /* block not compressed */
}
}
DEBUGLOG(4, "ZSTD_entropyCompressSequences() cSize: %zu\n", cSize);
DEBUGLOG(4, "ZSTD_entropyCompressSeqStore() cSize: %zu", cSize);
return cSize;
}
@@ -2701,6 +2798,628 @@ static void ZSTD_confirmRepcodesAndEntropyTables(ZSTD_CCtx* zc)
zc->blockState.nextCBlock = tmp;
}
/* Writes the block header */
static void writeBlockHeader(void* op, size_t cSize, size_t blockSize, U32 lastBlock) {
U32 const cBlockHeader = cSize == 1 ?
lastBlock + (((U32)bt_rle)<<1) + (U32)(blockSize << 3) :
lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
MEM_writeLE24(op, cBlockHeader);
DEBUGLOG(3, "writeBlockHeader: cSize: %zu blockSize: %zu lastBlock: %u", cSize, blockSize, lastBlock);
}
/** ZSTD_buildBlockEntropyStats_literals() :
* Builds entropy for the literals.
* Stores literals block type (raw, rle, compressed, repeat) and
* huffman description table to hufMetadata.
* Requires ENTROPY_WORKSPACE_SIZE workspace
* @return : size of huffman description table or error code */
static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSize,
const ZSTD_hufCTables_t* prevHuf,
ZSTD_hufCTables_t* nextHuf,
ZSTD_hufCTablesMetadata_t* hufMetadata,
const int disableLiteralsCompression,
void* workspace, size_t wkspSize)
{
BYTE* const wkspStart = (BYTE*)workspace;
BYTE* const wkspEnd = wkspStart + wkspSize;
BYTE* const countWkspStart = wkspStart;
unsigned* const countWksp = (unsigned*)workspace;
const size_t countWkspSize = (HUF_SYMBOLVALUE_MAX + 1) * sizeof(unsigned);
BYTE* const nodeWksp = countWkspStart + countWkspSize;
const size_t nodeWkspSize = wkspEnd-nodeWksp;
unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
unsigned huffLog = HUF_TABLELOG_DEFAULT;
HUF_repeat repeat = prevHuf->repeatMode;
DEBUGLOG(5, "ZSTD_buildBlockEntropyStats_literals (srcSize=%zu)", srcSize);
/* Prepare nextEntropy assuming reusing the existing table */
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
if (disableLiteralsCompression) {
DEBUGLOG(5, "set_basic - disabled");
hufMetadata->hType = set_basic;
return 0;
}
/* small ? don't even attempt compression (speed opt) */
#ifndef COMPRESS_LITERALS_SIZE_MIN
#define COMPRESS_LITERALS_SIZE_MIN 63
#endif
{ size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
if (srcSize <= minLitSize) {
DEBUGLOG(5, "set_basic - too small");
hufMetadata->hType = set_basic;
return 0;
}
}
/* Scan input and build symbol stats */
{ size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)src, srcSize, workspace, wkspSize);
FORWARD_IF_ERROR(largest, "HIST_count_wksp failed");
if (largest == srcSize) {
DEBUGLOG(5, "set_rle");
hufMetadata->hType = set_rle;
return 0;
}
if (largest <= (srcSize >> 7)+4) {
DEBUGLOG(5, "set_basic - no gain");
hufMetadata->hType = set_basic;
return 0;
}
}
/* Validate the previous Huffman table */
if (repeat == HUF_repeat_check && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
repeat = HUF_repeat_none;
}
/* Build Huffman Tree */
ZSTD_memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
{ size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
maxSymbolValue, huffLog,
nodeWksp, nodeWkspSize);
FORWARD_IF_ERROR(maxBits, "HUF_buildCTable_wksp");
huffLog = (U32)maxBits;
{ /* Build and write the CTable */
size_t const newCSize = HUF_estimateCompressedSize(
(HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
size_t const hSize = HUF_writeCTable_wksp(
hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
(HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog,
nodeWksp, nodeWkspSize);
/* Check against repeating the previous CTable */
if (repeat != HUF_repeat_none) {
size_t const oldCSize = HUF_estimateCompressedSize(
(HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
DEBUGLOG(5, "set_repeat - smaller");
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
hufMetadata->hType = set_repeat;
return 0;
}
}
if (newCSize + hSize >= srcSize) {
DEBUGLOG(5, "set_basic - no gains");
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
hufMetadata->hType = set_basic;
return 0;
}
DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
hufMetadata->hType = set_compressed;
nextHuf->repeatMode = HUF_repeat_check;
return hSize;
}
}
}
/** ZSTD_buildBlockEntropyStats_sequences() :
* Builds entropy for the sequences.
* Stores symbol compression modes and fse table to fseMetadata.
* Requires ENTROPY_WORKSPACE_SIZE wksp.
* @return : size of fse tables or error code */
static size_t ZSTD_buildBlockEntropyStats_sequences(seqStore_t* seqStorePtr,
const ZSTD_fseCTables_t* prevEntropy,
ZSTD_fseCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
ZSTD_fseCTablesMetadata_t* fseMetadata,
void* workspace, size_t wkspSize)
{
ZSTD_strategy const strategy = cctxParams->cParams.strategy;
size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
BYTE* const ostart = fseMetadata->fseTablesBuffer;
BYTE* const oend = ostart + sizeof(fseMetadata->fseTablesBuffer);
BYTE* op = ostart;
unsigned* countWorkspace = (unsigned*)workspace;
unsigned* entropyWorkspace = countWorkspace + (MaxSeq + 1);
size_t entropyWorkspaceSize = wkspSize - (MaxSeq + 1) * sizeof(*countWorkspace);
ZSTD_symbolEncodingTypeStats_t stats;
DEBUGLOG(5, "ZSTD_buildBlockEntropyStats_sequences (nbSeq=%zu)", nbSeq);
stats = ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq,
&fseMetadata->lastCountSize,
prevEntropy, nextEntropy, op, oend,
strategy, countWorkspace,
entropyWorkspace, entropyWorkspaceSize);
FORWARD_IF_ERROR(stats.size, "ZSTD_buildSequencesStatistics failed!");
fseMetadata->llType = (symbolEncodingType_e) stats.LLtype;
fseMetadata->ofType = (symbolEncodingType_e) stats.Offtype;
fseMetadata->mlType = (symbolEncodingType_e) stats.MLtype;
return stats.size;
}
/** ZSTD_buildBlockEntropyStats() :
* Builds entropy for the block.
* Requires workspace size ENTROPY_WORKSPACE_SIZE
*
* @return : 0 on success or error code
*/
size_t ZSTD_buildBlockEntropyStats(seqStore_t* seqStorePtr,
const ZSTD_entropyCTables_t* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
ZSTD_entropyCTablesMetadata_t* entropyMetadata,
void* workspace, size_t wkspSize)
{
size_t const litSize = seqStorePtr->lit - seqStorePtr->litStart;
entropyMetadata->hufMetadata.hufDesSize =
ZSTD_buildBlockEntropyStats_literals(seqStorePtr->litStart, litSize,
&prevEntropy->huf, &nextEntropy->huf,
&entropyMetadata->hufMetadata,
ZSTD_disableLiteralsCompression(cctxParams),
workspace, wkspSize);
FORWARD_IF_ERROR(entropyMetadata->hufMetadata.hufDesSize, "ZSTD_buildBlockEntropyStats_literals failed");
entropyMetadata->fseMetadata.fseTablesSize =
ZSTD_buildBlockEntropyStats_sequences(seqStorePtr,
&prevEntropy->fse, &nextEntropy->fse,
cctxParams,
&entropyMetadata->fseMetadata,
workspace, wkspSize);
FORWARD_IF_ERROR(entropyMetadata->fseMetadata.fseTablesSize, "ZSTD_buildBlockEntropyStats_sequences failed");
return 0;
}
/* Returns the size estimate for the literals section (header + content) of a block */
static size_t ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSize,
const ZSTD_hufCTables_t* huf,
const ZSTD_hufCTablesMetadata_t* hufMetadata,
void* workspace, size_t wkspSize,
int writeEntropy)
{
unsigned* const countWksp = (unsigned*)workspace;
unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
size_t literalSectionHeaderSize = 3 + (litSize >= 1 KB) + (litSize >= 16 KB);
U32 singleStream = litSize < 256;
if (hufMetadata->hType == set_basic) return litSize;
else if (hufMetadata->hType == set_rle) return 1;
else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
if (ZSTD_isError(largest)) return litSize;
{ size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
if (!singleStream) cLitSizeEstimate += 6; /* multi-stream huffman uses 6-byte jump table */
return cLitSizeEstimate + literalSectionHeaderSize;
} }
assert(0); /* impossible */
return 0;
}
/* Returns the size estimate for the FSE-compressed symbols (of, ml, ll) of a block */
static size_t ZSTD_estimateBlockSize_symbolType(symbolEncodingType_e type,
const BYTE* codeTable, size_t nbSeq, unsigned maxCode,
const FSE_CTable* fseCTable,
const U32* additionalBits,
short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
void* workspace, size_t wkspSize)
{
unsigned* const countWksp = (unsigned*)workspace;
const BYTE* ctp = codeTable;
const BYTE* const ctStart = ctp;
const BYTE* const ctEnd = ctStart + nbSeq;
size_t cSymbolTypeSizeEstimateInBits = 0;
unsigned max = maxCode;
HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize); /* can't fail */
if (type == set_basic) {
/* We selected this encoding type, so it must be valid. */
assert(max <= defaultMax);
(void)defaultMax;
cSymbolTypeSizeEstimateInBits = ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max);
} else if (type == set_rle) {
cSymbolTypeSizeEstimateInBits = 0;
} else if (type == set_compressed || type == set_repeat) {
cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
}
if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) {
return nbSeq * 10;
}
while (ctp < ctEnd) {
if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
ctp++;
}
return cSymbolTypeSizeEstimateInBits >> 3;
}
/* Returns the size estimate for the sequences section (header + content) of a block */
static size_t ZSTD_estimateBlockSize_sequences(const BYTE* ofCodeTable,
const BYTE* llCodeTable,
const BYTE* mlCodeTable,
size_t nbSeq,
const ZSTD_fseCTables_t* fseTables,
const ZSTD_fseCTablesMetadata_t* fseMetadata,
void* workspace, size_t wkspSize,
int writeEntropy)
{
size_t sequencesSectionHeaderSize = 1 /* seqHead */ + 1 /* min seqSize size */ + (nbSeq >= 128) + (nbSeq >= LONGNBSEQ);
size_t cSeqSizeEstimate = 0;
cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, nbSeq, MaxOff,
fseTables->offcodeCTable, NULL,
OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->llType, llCodeTable, nbSeq, MaxLL,
fseTables->litlengthCTable, LL_bits,
LL_defaultNorm, LL_defaultNormLog, MaxLL,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, nbSeq, MaxML,
fseTables->matchlengthCTable, ML_bits,
ML_defaultNorm, ML_defaultNormLog, MaxML,
workspace, wkspSize);
if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
return cSeqSizeEstimate + sequencesSectionHeaderSize;
}
/* Returns the size estimate for a given stream of literals, of, ll, ml */
static size_t ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
const BYTE* ofCodeTable,
const BYTE* llCodeTable,
const BYTE* mlCodeTable,
size_t nbSeq,
const ZSTD_entropyCTables_t* entropy,
const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
void* workspace, size_t wkspSize,
int writeLitEntropy, int writeSeqEntropy) {
size_t const literalsSize = ZSTD_estimateBlockSize_literal(literals, litSize,
&entropy->huf, &entropyMetadata->hufMetadata,
workspace, wkspSize, writeLitEntropy);
size_t const seqSize = ZSTD_estimateBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
workspace, wkspSize, writeSeqEntropy);
return seqSize + literalsSize + ZSTD_blockHeaderSize;
}
/* Builds entropy statistics and uses them for blocksize estimation.
*
* Returns the estimated compressed size of the seqStore, or a zstd error.
*/
static size_t ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(seqStore_t* seqStore, const ZSTD_CCtx* zc) {
ZSTD_entropyCTablesMetadata_t entropyMetadata;
FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(seqStore,
&zc->blockState.prevCBlock->entropy,
&zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
&entropyMetadata,
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
return ZSTD_estimateBlockSize(seqStore->litStart, (size_t)(seqStore->lit - seqStore->litStart),
seqStore->ofCode, seqStore->llCode, seqStore->mlCode,
(size_t)(seqStore->sequences - seqStore->sequencesStart),
&zc->blockState.nextCBlock->entropy, &entropyMetadata, zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE,
(int)(entropyMetadata.hufMetadata.hType == set_compressed), 1);
}
/* Returns literals bytes represented in a seqStore */
static size_t ZSTD_countSeqStoreLiteralsBytes(const seqStore_t* const seqStore) {
size_t literalsBytes = 0;
size_t const nbSeqs = seqStore->sequences - seqStore->sequencesStart;
size_t i;
for (i = 0; i < nbSeqs; ++i) {
seqDef seq = seqStore->sequencesStart[i];
literalsBytes += seq.litLength;
if (i == seqStore->longLengthPos && seqStore->longLengthID == 1) {
literalsBytes += 0x10000;
}
}
return literalsBytes;
}
/* Returns match bytes represented in a seqStore */
static size_t ZSTD_countSeqStoreMatchBytes(const seqStore_t* const seqStore) {
size_t matchBytes = 0;
size_t const nbSeqs = seqStore->sequences - seqStore->sequencesStart;
size_t i;
for (i = 0; i < nbSeqs; ++i) {
seqDef seq = seqStore->sequencesStart[i];
matchBytes += seq.matchLength + MINMATCH;
if (i == seqStore->longLengthPos && seqStore->longLengthID == 2) {
matchBytes += 0x10000;
}
}
return matchBytes;
}
/* Derives the seqStore that is a chunk of the originalSeqStore from [startIdx, endIdx).
* Stores the result in resultSeqStore.
*/
static void ZSTD_deriveSeqStoreChunk(seqStore_t* resultSeqStore,
const seqStore_t* originalSeqStore,
size_t startIdx, size_t endIdx) {
BYTE* const litEnd = originalSeqStore->lit;
size_t literalsBytes;
size_t literalsBytesPreceding = 0;
*resultSeqStore = *originalSeqStore;
if (startIdx > 0) {
resultSeqStore->sequences = originalSeqStore->sequencesStart + startIdx;
literalsBytesPreceding = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
}
/* Move longLengthPos into the correct position if necessary */
if (originalSeqStore->longLengthID != 0) {
if (originalSeqStore->longLengthPos < startIdx || originalSeqStore->longLengthPos > endIdx) {
resultSeqStore->longLengthID = 0;
} else {
resultSeqStore->longLengthPos -= (U32)startIdx;
}
}
resultSeqStore->sequencesStart = originalSeqStore->sequencesStart + startIdx;
resultSeqStore->sequences = originalSeqStore->sequencesStart + endIdx;
literalsBytes = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
resultSeqStore->litStart += literalsBytesPreceding;
if (endIdx == (size_t)(originalSeqStore->sequences - originalSeqStore->sequencesStart)) {
/* This accounts for possible last literals if the derived chunk reaches the end of the block */
resultSeqStore->lit = litEnd;
} else {
resultSeqStore->lit = resultSeqStore->litStart+literalsBytes;
}
resultSeqStore->llCode += startIdx;
resultSeqStore->mlCode += startIdx;
resultSeqStore->ofCode += startIdx;
}
/* ZSTD_compressSeqStore_singleBlock():
* Compresses a seqStore into a block with a block header, into the buffer dst.
*
* Returns the total size of that block (including header) or a ZSTD error code.
*/
static size_t ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* seqStore,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
U32 lastBlock, U32 canEmitRLEorNoCompress) {
const U32 rleMaxLength = 25;
BYTE* op = (BYTE*)dst;
const BYTE* ip = (const BYTE*)src;
size_t cSize;
size_t cSeqsSize = ZSTD_entropyCompressSeqStore(seqStore,
&zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
op + ZSTD_blockHeaderSize, dstCapacity - ZSTD_blockHeaderSize,
srcSize,
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
zc->bmi2, canEmitRLEorNoCompress);
FORWARD_IF_ERROR(cSeqsSize, "ZSTD_entropyCompressSeqStore failed!");
if (!zc->isFirstBlock &&
cSeqsSize < rleMaxLength &&
ZSTD_isRLE((BYTE const*)src, srcSize)&&
canEmitRLEorNoCompress) {
/* 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
*/
cSeqsSize = 1;
}
if (zc->seqCollector.collectSequences) {
ZSTD_copyBlockSequences(zc);
ZSTD_confirmRepcodesAndEntropyTables(zc);
return 0;
}
if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
if (cSeqsSize == 0 && canEmitRLEorNoCompress) {
cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, srcSize, lastBlock);
FORWARD_IF_ERROR(cSize, "Nocompress block failed");
DEBUGLOG(4, "Writing out nocompress block, size: %zu", cSize);
} else if (cSeqsSize == 1 && canEmitRLEorNoCompress) {
cSize = ZSTD_rleCompressBlock(op, dstCapacity, *ip, srcSize, lastBlock);
FORWARD_IF_ERROR(cSize, "RLE compress block failed");
DEBUGLOG(4, "Writing out RLE block, size: %zu", cSize);
} else {
/* Error checking and repcodes update */
ZSTD_confirmRepcodesAndEntropyTables(zc);
writeBlockHeader(op, cSeqsSize, srcSize, lastBlock);
cSize = ZSTD_blockHeaderSize + cSeqsSize;
DEBUGLOG(4, "Writing out compressed block, size: %zu", cSize);
}
return cSize;
}
/* Struct to keep track of where we are in our recursive calls. */
typedef struct {
U32* splitLocations; /* Array of split indices */
size_t idx; /* The current index within splitLocations being worked on */
} seqStoreSplits;
#define MIN_SEQUENCES_BLOCK_SPLITTING 300
#define MAX_NB_SPLITS 196
/* Helper function to perform the recursive search for block splits.
* Estimates the cost of seqStore prior to split, and estimates the cost of splitting the sequences in half.
* If advantageous to split, then we recurse down the two sub-blocks. If not, or if an error occurred in estimation, then
* we do not recurse.
*
* Note: The recursion depth is capped by a heuristic minimum number of sequences, defined by MIN_SEQUENCES_BLOCK_SPLITTING.
* In theory, this means the absolute largest recursion depth is 10 == log2(maxNbSeqInBlock/MIN_SEQUENCES_BLOCK_SPLITTING).
* In practice, recursion depth usually doesn't go beyond 4.
*
* Furthermore, the number of splits is capped by MAX_NB_SPLITS. At MAX_NB_SPLITS == 196 with the current existing blockSize
* maximum of 128 KB, this value is actually impossible to reach.
*/
static void ZSTD_deriveBlockSplitsHelper(seqStoreSplits* splits, size_t startIdx, size_t endIdx,
const ZSTD_CCtx* zc, const seqStore_t* origSeqStore) {
seqStore_t fullSeqStoreChunk;
seqStore_t firstHalfSeqStore;
seqStore_t secondHalfSeqStore;
size_t estimatedOriginalSize;
size_t estimatedFirstHalfSize;
size_t estimatedSecondHalfSize;
size_t midIdx = (startIdx + endIdx)/2;
if (endIdx - startIdx < MIN_SEQUENCES_BLOCK_SPLITTING || splits->idx >= MAX_NB_SPLITS) {
return;
}
ZSTD_deriveSeqStoreChunk(&fullSeqStoreChunk, origSeqStore, startIdx, endIdx);
ZSTD_deriveSeqStoreChunk(&firstHalfSeqStore, origSeqStore, startIdx, midIdx);
ZSTD_deriveSeqStoreChunk(&secondHalfSeqStore, origSeqStore, midIdx, endIdx);
estimatedOriginalSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&fullSeqStoreChunk, zc);
estimatedFirstHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&firstHalfSeqStore, zc);
estimatedSecondHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&secondHalfSeqStore, zc);
DEBUGLOG(5, "Estimated original block size: %zu -- First half split: %zu -- Second half split: %zu",
estimatedOriginalSize, estimatedFirstHalfSize, estimatedSecondHalfSize);
if (ZSTD_isError(estimatedOriginalSize) || ZSTD_isError(estimatedFirstHalfSize) || ZSTD_isError(estimatedSecondHalfSize)) {
return;
}
if (estimatedFirstHalfSize + estimatedSecondHalfSize < estimatedOriginalSize) {
ZSTD_deriveBlockSplitsHelper(splits, startIdx, midIdx, zc, origSeqStore);
splits->splitLocations[splits->idx] = (U32)midIdx;
splits->idx++;
ZSTD_deriveBlockSplitsHelper(splits, midIdx, endIdx, zc, origSeqStore);
}
}
/* Base recursive function. Populates a table with intra-block partition indices that can improve compression ratio.
*
* Returns the number of splits made (which equals the size of the partition table - 1).
*/
static size_t ZSTD_deriveBlockSplits(ZSTD_CCtx* zc, U32 partitions[], U32 nbSeq) {
seqStoreSplits splits = {partitions, 0};
if (nbSeq <= 4) {
DEBUGLOG(4, "ZSTD_deriveBlockSplits: Too few sequences to split");
/* Refuse to try and split anything with less than 4 sequences */
return 0;
}
ZSTD_deriveBlockSplitsHelper(&splits, 0, nbSeq, zc, &zc->seqStore);
splits.splitLocations[splits.idx] = nbSeq;
DEBUGLOG(5, "ZSTD_deriveBlockSplits: final nb splits: %zu", splits.idx-1);
return splits.idx;
}
/* Return 1 if if the first three sequences of seqstore/block use repcodes */
static U32 ZSTD_seqStore_firstThreeContainRepcodes(const seqStore_t* const seqStore) {
U32 const seqLimit = MIN((U32)(seqStore->sequences - seqStore->sequencesStart), ZSTD_REP_NUM);
U32 seqIdx = 0;
for (; seqIdx < seqLimit; ++seqIdx) {
if (seqStore->sequencesStart[seqIdx].offset <= ZSTD_REP_MOVE) {
return 1;
}
}
return 0;
}
/* ZSTD_compressBlock_splitBlock():
* Attempts to split a given block into multiple blocks to improve compression ratio.
*
* Returns combined size of all blocks (which includes headers), or a ZSTD error code.
*/
static size_t ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCapacity,
const void* src, size_t blockSize, U32 lastBlock, U32 nbSeq) {
size_t cSize = 0;
const BYTE* ip = (const BYTE*)src;
BYTE* op = (BYTE*)dst;
U32 partitions[MAX_NB_SPLITS];
size_t i = 0;
size_t srcBytesTotal = 0;
size_t numSplits = ZSTD_deriveBlockSplits(zc, partitions, nbSeq);
seqStore_t nextSeqStore;
seqStore_t currSeqStore;
U32 canEmitRLEorNoCompress = 1;
const size_t dstCapacityInitial = dstCapacity;
DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
(unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
(unsigned)zc->blockState.matchState.nextToUpdate);
if (numSplits == 0) {
size_t cSizeSingleBlock = ZSTD_compressSeqStore_singleBlock(zc, &zc->seqStore, op, dstCapacity, ip, blockSize, lastBlock, 1);
FORWARD_IF_ERROR(cSizeSingleBlock, "Compressing single block from splitBlock_internal() failed!");
DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal: No splits");
return cSizeSingleBlock;
}
ZSTD_deriveSeqStoreChunk(&currSeqStore, &zc->seqStore, 0, partitions[0]);
for (i = 0; i <= numSplits; ++i) {
size_t srcBytes;
size_t cSizeChunk;
U32 lastBlockActual;
srcBytes = ZSTD_countSeqStoreLiteralsBytes(&currSeqStore) + ZSTD_countSeqStoreMatchBytes(&currSeqStore);
lastBlockActual = lastBlock && (i == numSplits);
srcBytesTotal += srcBytes;
if (i == numSplits) {
/* This is the final partition, need to account for possible last literals */
srcBytes += blockSize - srcBytesTotal;
} else {
ZSTD_deriveSeqStoreChunk(&nextSeqStore, &zc->seqStore, partitions[i], partitions[i+1]);
if (ZSTD_seqStore_firstThreeContainRepcodes(&nextSeqStore)) {
DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal: Next block contains rep in first three seqs!");
canEmitRLEorNoCompress = 0;
}
}
cSizeChunk = ZSTD_compressSeqStore_singleBlock(zc, &currSeqStore, op, dstCapacity, ip, srcBytes, lastBlockActual, canEmitRLEorNoCompress);
DEBUGLOG(5, "Estimated size: %zu actual size: %zu", ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&currSeqStore, zc), cSizeChunk);
FORWARD_IF_ERROR(cSizeChunk, "Compressing chunk failed!");
ZSTD_memcpy(zc->blockState.nextCBlock->rep, zc->blockState.prevCBlock->rep, sizeof(U32)*ZSTD_REP_NUM);
ip += srcBytes;
op += cSizeChunk;
dstCapacity -= cSizeChunk;
cSize += cSizeChunk;
currSeqStore = nextSeqStore;
}
if (cSize > ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize) {
/* If too large, recompress the original block to avoid any chance of a single block exceeding ZSTD_BLOCKSIZE_MAX */
cSize = ZSTD_compressSeqStore_singleBlock(zc, &zc->seqStore, (BYTE*)dst, dstCapacityInitial, (const BYTE*)src, blockSize, lastBlock, 1);
FORWARD_IF_ERROR(cSize, "Compressing single block from splitBlock_internal() fallback failed!");
DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal: Compressed split block too large, recompressed");
}
assert(cSize <= ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize);
return cSize;
}
static size_t ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize, U32 lastBlock) {
const BYTE* ip = (const BYTE*)src;
BYTE* op = (BYTE*)dst;
U32 nbSeq;
size_t cSize;
DEBUGLOG(4, "ZSTD_compressBlock_splitBlock");
{ const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
if (bss == ZSTDbss_noCompress) {
if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, srcSize, lastBlock);
FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
DEBUGLOG(4, "ZSTD_compressBlock_splitBlock: Nocompress block");
return cSize;
}
nbSeq = (U32)(zc->seqStore.sequences - zc->seqStore.sequencesStart);
}
assert(zc->appliedParams.splitBlocks == 1);
cSize = ZSTD_compressBlock_splitBlock_internal(zc, dst, dstCapacity, src, srcSize, lastBlock, nbSeq);
FORWARD_IF_ERROR(cSize, "Splitting blocks failed!");
return cSize;
}
static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize, U32 frame)
@@ -2729,13 +3448,13 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
}
/* encode sequences and literals */
cSize = ZSTD_entropyCompressSequences(&zc->seqStore,
cSize = ZSTD_entropyCompressSeqStore(&zc->seqStore,
&zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
dst, dstCapacity,
srcSize,
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
zc->bmi2);
zc->bmi2, 1 /* Can emit uncompressed blocks */);
if (zc->seqCollector.collectSequences) {
ZSTD_copyBlockSequences(zc);
@@ -2876,7 +3595,7 @@ static void ZSTD_overflowCorrectIfNeeded(ZSTD_matchState_t* ms,
* Frame is supposed already started (header already produced)
* @return : compressed size, or an error code
*/
static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
U32 lastFrameChunk)
@@ -2916,6 +3635,10 @@ static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_targetCBlockSize failed");
assert(cSize > 0);
assert(cSize <= blockSize + ZSTD_blockHeaderSize);
} else if (ZSTD_blockSplitterEnabled(&cctx->appliedParams)) {
cSize = ZSTD_compressBlock_splitBlock(cctx, op, dstCapacity, ip, blockSize, lastBlock);
FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_splitBlock failed");
assert(cSize > 0 || cctx->seqCollector.collectSequences == 1);
} else {
cSize = ZSTD_compressBlock_internal(cctx,
op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize,
@@ -4455,6 +5178,11 @@ static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
params.ldmParams.enableLdm = 1;
}
if (ZSTD_CParams_useBlockSplitter(&params.cParams)) {
DEBUGLOG(4, "Block splitter enabled by default (window size >= 128K, strategy >= btopt)");
params.splitBlocks = 1;
}
#ifdef ZSTD_MULTITHREAD
if ((cctx->pledgedSrcSizePlusOne-1) <= ZSTDMT_JOBSIZE_MIN) {
params.nbWorkers = 0; /* do not invoke multi-threading when src size is too small */
@@ -4929,13 +5657,13 @@ static size_t ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
continue;
}
compressedSeqsSize = ZSTD_entropyCompressSequences(&cctx->seqStore,
compressedSeqsSize = ZSTD_entropyCompressSeqStore(&cctx->seqStore,
&cctx->blockState.prevCBlock->entropy, &cctx->blockState.nextCBlock->entropy,
&cctx->appliedParams,
op + ZSTD_blockHeaderSize /* Leave space for block header */, dstCapacity - ZSTD_blockHeaderSize,
blockSize,
cctx->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
cctx->bmi2);
cctx->bmi2, 1 /* Can emit uncompressed blocks */);
FORWARD_IF_ERROR(compressedSeqsSize, "Compressing sequences of block failed");
DEBUGLOG(4, "Compressed sequences size: %zu", compressedSeqsSize);
+50
View File
@@ -82,6 +82,53 @@ typedef struct {
ZSTD_fseCTables_t fse;
} ZSTD_entropyCTables_t;
/***********************************************
* Entropy buffer statistics structs and funcs *
***********************************************/
/** ZSTD_hufCTablesMetadata_t :
* Stores Literals Block Type for a super-block in hType, and
* huffman tree description in hufDesBuffer.
* hufDesSize refers to the size of huffman tree description in bytes.
* This metadata is populated in ZSTD_buildBlockEntropyStats_literals() */
typedef struct {
symbolEncodingType_e hType;
BYTE hufDesBuffer[ZSTD_MAX_HUF_HEADER_SIZE];
size_t hufDesSize;
} ZSTD_hufCTablesMetadata_t;
/** ZSTD_fseCTablesMetadata_t :
* Stores symbol compression modes for a super-block in {ll, ol, ml}Type, and
* fse tables in fseTablesBuffer.
* fseTablesSize refers to the size of fse tables in bytes.
* This metadata is populated in ZSTD_buildBlockEntropyStats_sequences() */
typedef struct {
symbolEncodingType_e llType;
symbolEncodingType_e ofType;
symbolEncodingType_e mlType;
BYTE fseTablesBuffer[ZSTD_MAX_FSE_HEADERS_SIZE];
size_t fseTablesSize;
size_t lastCountSize; /* This is to account for bug in 1.3.4. More detail in ZSTD_entropyCompressSequences_internal() */
} ZSTD_fseCTablesMetadata_t;
typedef struct {
ZSTD_hufCTablesMetadata_t hufMetadata;
ZSTD_fseCTablesMetadata_t fseMetadata;
} ZSTD_entropyCTablesMetadata_t;
/** ZSTD_buildBlockEntropyStats() :
* Builds entropy for the block.
* @return : 0 on success or error code */
size_t ZSTD_buildBlockEntropyStats(seqStore_t* seqStorePtr,
const ZSTD_entropyCTables_t* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
ZSTD_entropyCTablesMetadata_t* entropyMetadata,
void* workspace, size_t wkspSize);
/*********************************
* Compression internals structs *
*********************************/
typedef struct {
U32 off; /* Offset code (offset + ZSTD_REP_MOVE) for the match */
U32 len; /* Raw length of match */
@@ -256,6 +303,9 @@ struct ZSTD_CCtx_params_s {
ZSTD_sequenceFormat_e blockDelimiters;
int validateSequences;
/* Block splitting */
int splitBlocks;
/* Internal use, for createCCtxParams() and freeCCtxParams() only */
ZSTD_customMem customMem;
}; /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
+2 -281
View File
@@ -15,289 +15,10 @@
#include "../common/zstd_internal.h" /* ZSTD_getSequenceLength */
#include "hist.h" /* HIST_countFast_wksp */
#include "zstd_compress_internal.h"
#include "zstd_compress_internal.h" /* ZSTD_[huf|fse|entropy]CTablesMetadata_t */
#include "zstd_compress_sequences.h"
#include "zstd_compress_literals.h"
/*-*************************************
* Superblock entropy buffer structs
***************************************/
/** ZSTD_hufCTablesMetadata_t :
* Stores Literals Block Type for a super-block in hType, and
* huffman tree description in hufDesBuffer.
* hufDesSize refers to the size of huffman tree description in bytes.
* This metadata is populated in ZSTD_buildSuperBlockEntropy_literal() */
typedef struct {
symbolEncodingType_e hType;
BYTE hufDesBuffer[ZSTD_MAX_HUF_HEADER_SIZE];
size_t hufDesSize;
} ZSTD_hufCTablesMetadata_t;
/** ZSTD_fseCTablesMetadata_t :
* Stores symbol compression modes for a super-block in {ll, ol, ml}Type, and
* fse tables in fseTablesBuffer.
* fseTablesSize refers to the size of fse tables in bytes.
* This metadata is populated in ZSTD_buildSuperBlockEntropy_sequences() */
typedef struct {
symbolEncodingType_e llType;
symbolEncodingType_e ofType;
symbolEncodingType_e mlType;
BYTE fseTablesBuffer[ZSTD_MAX_FSE_HEADERS_SIZE];
size_t fseTablesSize;
size_t lastCountSize; /* This is to account for bug in 1.3.4. More detail in ZSTD_compressSubBlock_sequences() */
} ZSTD_fseCTablesMetadata_t;
typedef struct {
ZSTD_hufCTablesMetadata_t hufMetadata;
ZSTD_fseCTablesMetadata_t fseMetadata;
} ZSTD_entropyCTablesMetadata_t;
/** ZSTD_buildSuperBlockEntropy_literal() :
* Builds entropy for the super-block literals.
* Stores literals block type (raw, rle, compressed, repeat) and
* huffman description table to hufMetadata.
* @return : size of huffman description table or error code */
static size_t ZSTD_buildSuperBlockEntropy_literal(void* const src, size_t srcSize,
const ZSTD_hufCTables_t* prevHuf,
ZSTD_hufCTables_t* nextHuf,
ZSTD_hufCTablesMetadata_t* hufMetadata,
const int disableLiteralsCompression,
void* workspace, size_t wkspSize)
{
BYTE* const wkspStart = (BYTE*)workspace;
BYTE* const wkspEnd = wkspStart + wkspSize;
BYTE* const countWkspStart = wkspStart;
unsigned* const countWksp = (unsigned*)workspace;
const size_t countWkspSize = (HUF_SYMBOLVALUE_MAX + 1) * sizeof(unsigned);
BYTE* const nodeWksp = countWkspStart + countWkspSize;
const size_t nodeWkspSize = wkspEnd-nodeWksp;
unsigned maxSymbolValue = 255;
unsigned huffLog = HUF_TABLELOG_DEFAULT;
HUF_repeat repeat = prevHuf->repeatMode;
DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_literal (srcSize=%zu)", srcSize);
/* Prepare nextEntropy assuming reusing the existing table */
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
if (disableLiteralsCompression) {
DEBUGLOG(5, "set_basic - disabled");
hufMetadata->hType = set_basic;
return 0;
}
/* small ? don't even attempt compression (speed opt) */
# define COMPRESS_LITERALS_SIZE_MIN 63
{ size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
if (srcSize <= minLitSize) {
DEBUGLOG(5, "set_basic - too small");
hufMetadata->hType = set_basic;
return 0;
}
}
/* Scan input and build symbol stats */
{ size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)src, srcSize, workspace, wkspSize);
FORWARD_IF_ERROR(largest, "HIST_count_wksp failed");
if (largest == srcSize) {
DEBUGLOG(5, "set_rle");
hufMetadata->hType = set_rle;
return 0;
}
if (largest <= (srcSize >> 7)+4) {
DEBUGLOG(5, "set_basic - no gain");
hufMetadata->hType = set_basic;
return 0;
}
}
/* Validate the previous Huffman table */
if (repeat == HUF_repeat_check && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
repeat = HUF_repeat_none;
}
/* Build Huffman Tree */
ZSTD_memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
{ size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
maxSymbolValue, huffLog,
nodeWksp, nodeWkspSize);
FORWARD_IF_ERROR(maxBits, "HUF_buildCTable_wksp");
huffLog = (U32)maxBits;
{ /* Build and write the CTable */
size_t const newCSize = HUF_estimateCompressedSize(
(HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
size_t const hSize = HUF_writeCTable_wksp(
hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
(HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog,
nodeWksp, nodeWkspSize);
/* Check against repeating the previous CTable */
if (repeat != HUF_repeat_none) {
size_t const oldCSize = HUF_estimateCompressedSize(
(HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
DEBUGLOG(5, "set_repeat - smaller");
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
hufMetadata->hType = set_repeat;
return 0;
}
}
if (newCSize + hSize >= srcSize) {
DEBUGLOG(5, "set_basic - no gains");
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
hufMetadata->hType = set_basic;
return 0;
}
DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
hufMetadata->hType = set_compressed;
nextHuf->repeatMode = HUF_repeat_check;
return hSize;
}
}
}
/** ZSTD_buildSuperBlockEntropy_sequences() :
* Builds entropy for the super-block sequences.
* Stores symbol compression modes and fse table to fseMetadata.
* @return : size of fse tables or error code */
static size_t ZSTD_buildSuperBlockEntropy_sequences(seqStore_t* seqStorePtr,
const ZSTD_fseCTables_t* prevEntropy,
ZSTD_fseCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
ZSTD_fseCTablesMetadata_t* fseMetadata,
void* workspace, size_t wkspSize)
{
BYTE* const wkspStart = (BYTE*)workspace;
BYTE* const wkspEnd = wkspStart + wkspSize;
BYTE* const countWkspStart = wkspStart;
unsigned* const countWksp = (unsigned*)workspace;
const size_t countWkspSize = (MaxSeq + 1) * sizeof(unsigned);
BYTE* const cTableWksp = countWkspStart + countWkspSize;
const size_t cTableWkspSize = wkspEnd-cTableWksp;
ZSTD_strategy const strategy = cctxParams->cParams.strategy;
FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
const BYTE* const ofCodeTable = seqStorePtr->ofCode;
const BYTE* const llCodeTable = seqStorePtr->llCode;
const BYTE* const mlCodeTable = seqStorePtr->mlCode;
size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
BYTE* const ostart = fseMetadata->fseTablesBuffer;
BYTE* const oend = ostart + sizeof(fseMetadata->fseTablesBuffer);
BYTE* op = ostart;
assert(cTableWkspSize >= (1 << MaxFSELog) * sizeof(FSE_FUNCTION_TYPE));
DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_sequences (nbSeq=%zu)", nbSeq);
ZSTD_memset(workspace, 0, wkspSize);
fseMetadata->lastCountSize = 0;
/* convert length/distances into codes */
ZSTD_seqToCodes(seqStorePtr);
/* build CTable for Literal Lengths */
{ U32 LLtype;
unsigned max = MaxLL;
size_t const mostFrequent = HIST_countFast_wksp(countWksp, &max, llCodeTable, nbSeq, workspace, wkspSize); /* can't fail */
DEBUGLOG(5, "Building LL table");
nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode,
countWksp, max, mostFrequent, nbSeq,
LLFSELog, prevEntropy->litlengthCTable,
LL_defaultNorm, LL_defaultNormLog,
ZSTD_defaultAllowed, strategy);
assert(set_basic < set_compressed && set_rle < set_compressed);
assert(!(LLtype < set_compressed && nextEntropy->litlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
{ size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
countWksp, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable),
cTableWksp, cTableWkspSize);
FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for LitLens failed");
if (LLtype == set_compressed)
fseMetadata->lastCountSize = countSize;
op += countSize;
fseMetadata->llType = (symbolEncodingType_e) LLtype;
} }
/* build CTable for Offsets */
{ U32 Offtype;
unsigned max = MaxOff;
size_t const mostFrequent = HIST_countFast_wksp(countWksp, &max, ofCodeTable, nbSeq, workspace, wkspSize); /* can't fail */
/* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
DEBUGLOG(5, "Building OF table");
nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode,
countWksp, max, mostFrequent, nbSeq,
OffFSELog, prevEntropy->offcodeCTable,
OF_defaultNorm, OF_defaultNormLog,
defaultPolicy, strategy);
assert(!(Offtype < set_compressed && nextEntropy->offcode_repeatMode != FSE_repeat_none)); /* We don't copy tables */
{ size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
countWksp, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable),
cTableWksp, cTableWkspSize);
FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for Offsets failed");
if (Offtype == set_compressed)
fseMetadata->lastCountSize = countSize;
op += countSize;
fseMetadata->ofType = (symbolEncodingType_e) Offtype;
} }
/* build CTable for MatchLengths */
{ U32 MLtype;
unsigned max = MaxML;
size_t const mostFrequent = HIST_countFast_wksp(countWksp, &max, mlCodeTable, nbSeq, workspace, wkspSize); /* can't fail */
DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op));
nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode,
countWksp, max, mostFrequent, nbSeq,
MLFSELog, prevEntropy->matchlengthCTable,
ML_defaultNorm, ML_defaultNormLog,
ZSTD_defaultAllowed, strategy);
assert(!(MLtype < set_compressed && nextEntropy->matchlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
{ size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
countWksp, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable),
cTableWksp, cTableWkspSize);
FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for MatchLengths failed");
if (MLtype == set_compressed)
fseMetadata->lastCountSize = countSize;
op += countSize;
fseMetadata->mlType = (symbolEncodingType_e) MLtype;
} }
assert((size_t) (op-ostart) <= sizeof(fseMetadata->fseTablesBuffer));
return op-ostart;
}
/** ZSTD_buildSuperBlockEntropy() :
* Builds entropy for the super-block.
* @return : 0 on success or error code */
static size_t
ZSTD_buildSuperBlockEntropy(seqStore_t* seqStorePtr,
const ZSTD_entropyCTables_t* prevEntropy,
ZSTD_entropyCTables_t* nextEntropy,
const ZSTD_CCtx_params* cctxParams,
ZSTD_entropyCTablesMetadata_t* entropyMetadata,
void* workspace, size_t wkspSize)
{
size_t const litSize = seqStorePtr->lit - seqStorePtr->litStart;
DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy");
entropyMetadata->hufMetadata.hufDesSize =
ZSTD_buildSuperBlockEntropy_literal(seqStorePtr->litStart, litSize,
&prevEntropy->huf, &nextEntropy->huf,
&entropyMetadata->hufMetadata,
ZSTD_disableLiteralsCompression(cctxParams),
workspace, wkspSize);
FORWARD_IF_ERROR(entropyMetadata->hufMetadata.hufDesSize, "ZSTD_buildSuperBlockEntropy_literal failed");
entropyMetadata->fseMetadata.fseTablesSize =
ZSTD_buildSuperBlockEntropy_sequences(seqStorePtr,
&prevEntropy->fse, &nextEntropy->fse,
cctxParams,
&entropyMetadata->fseMetadata,
workspace, wkspSize);
FORWARD_IF_ERROR(entropyMetadata->fseMetadata.fseTablesSize, "ZSTD_buildSuperBlockEntropy_sequences failed");
return 0;
}
/** ZSTD_compressSubBlock_literal() :
* Compresses literals section for a sub-block.
* When we have to write the Huffman table we will sometimes choose a header
@@ -831,7 +552,7 @@ size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
unsigned lastBlock) {
ZSTD_entropyCTablesMetadata_t entropyMetadata;
FORWARD_IF_ERROR(ZSTD_buildSuperBlockEntropy(&zc->seqStore,
FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
&zc->blockState.prevCBlock->entropy,
&zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
+10 -1
View File
@@ -419,6 +419,7 @@ typedef enum {
* ZSTD_c_stableOutBuffer
* ZSTD_c_blockDelimiters
* ZSTD_c_validateSequences
* ZSTD_c_splitBlocks
* Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
* note : never ever use experimentalParam? names directly;
* also, the enums values themselves are unstable and can still change.
@@ -434,7 +435,8 @@ typedef enum {
ZSTD_c_experimentalParam9=1006,
ZSTD_c_experimentalParam10=1007,
ZSTD_c_experimentalParam11=1008,
ZSTD_c_experimentalParam12=1009
ZSTD_c_experimentalParam12=1009,
ZSTD_c_experimentalParam13=1010
} ZSTD_cParameter;
typedef struct {
@@ -1834,6 +1836,13 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
*/
#define ZSTD_c_validateSequences ZSTD_c_experimentalParam12
/* ZSTD_c_splitBlocks
* Default is 0 == disabled. Set to 1 to enable block splitting.
*
* Will attempt to split blocks in order to improve compression ratio at the cost of speed.
*/
#define ZSTD_c_splitBlocks ZSTD_c_experimentalParam13
/*! ZSTD_CCtx_getParameter() :
* Get the requested compression parameter value, selected by enum ZSTD_cParameter,
* and store it into int* value.
+1
View File
@@ -94,6 +94,7 @@ void FUZZ_setRandomParameters(ZSTD_CCtx *cctx, size_t srcSize, FUZZ_dataProducer
setRand(cctx, ZSTD_c_forceMaxWindow, 0, 1, producer);
setRand(cctx, ZSTD_c_literalCompressionMode, 0, 2, producer);
setRand(cctx, ZSTD_c_forceAttachDict, 0, 2, producer);
setRand(cctx, ZSTD_c_splitBlocks, 0, 1, producer);
if (FUZZ_dataProducer_uint32Range(producer, 0, 1) == 0) {
setRand(cctx, ZSTD_c_srcSizeHint, ZSTD_SRCSIZEHINT_MIN, 2 * srcSize, producer);
}
+9
View File
@@ -1544,6 +1544,15 @@ static int basicUnitTests(U32 const seed, double compressibility)
ZSTD_freeCCtx(cctx);
}
DISPLAYLEVEL(3, "test%3i : compress with block splitting : ", testNb++)
{ ZSTD_CCtx* cctx = ZSTD_createCCtx();
CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_splitBlocks, 1) );
cSize = ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize);
CHECK(cSize);
ZSTD_freeCCtx(cctx);
}
DISPLAYLEVEL(3, "OK \n");
DISPLAYLEVEL(3, "test%3i : compress -T2 with/without literals compression : ", testNb++)
{ ZSTD_CCtx* cctx = ZSTD_createCCtx();
size_t cSize1, cSize2;