facebook/zstd
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #1144 from terrelln/fse-entropy

Browse Source 
Approximate FSE encoding costs for selection
This commit is contained in:
Yann Collet 2018-05-22 19:25:37 -07:00 committed by GitHub
commit 27dc078aa6
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 260 additions and 85 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
lib/common/fse.h
View File

@ -581,12 +581,13 @@ MEM_STATIC U32 FSE_getMaxNbBits(const void* symbolTTPtr, U32 symbolValue)
return (symbolTT[symbolValue].deltaNbBits + ((1<<16)-1)) >> 16; return (symbolTT[symbolValue].deltaNbBits + ((1<<16)-1)) >> 16;
} }
/* FSE_bitCost_b256() : /* FSE_bitCost() :
* Approximate symbol cost, * Approximate symbol cost,
* provide fractional value, using fixed-point format (accuracyLog fractional bits) * provide fractional value, using fixed-point format (accuracyLog fractional bits)
* note: assume symbolValue is valid */ * note: assume symbolValue is valid */
MEM_STATIC U32 FSE_bitCost(const FSE_symbolCompressionTransform* symbolTT, U32 tableLog, U32 symbolValue, U32 accuracyLog) MEM_STATIC U32 FSE_bitCost(const void* symbolTTPtr, U32 tableLog, U32 symbolValue, U32 accuracyLog)
{ {
const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
U32 const minNbBits = symbolTT[symbolValue].deltaNbBits >> 16; U32 const minNbBits = symbolTT[symbolValue].deltaNbBits >> 16;
U32 const threshold = (minNbBits+1) << 16; U32 const threshold = (minNbBits+1) << 16;
assert(tableLog < 16); assert(tableLog < 16);

274
lib/compress/zstd_compress.c
View File

@ -946,10 +946,10 @@ static void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
int i; int i;
for (i = 0; i < ZSTD_REP_NUM; ++i) for (i = 0; i < ZSTD_REP_NUM; ++i)
bs->rep[i] = repStartValue[i]; bs->rep[i] = repStartValue[i];
bs->entropy.hufCTable_repeatMode = HUF_repeat_none; bs->entropy.huf.repeatMode = HUF_repeat_none;
bs->entropy.offcode_repeatMode = FSE_repeat_none; bs->entropy.fse.offcode_repeatMode = FSE_repeat_none;
bs->entropy.matchlength_repeatMode = FSE_repeat_none; bs->entropy.fse.matchlength_repeatMode = FSE_repeat_none;
bs->entropy.litlength_repeatMode = FSE_repeat_none; bs->entropy.fse.litlength_repeatMode = FSE_repeat_none;
} }
/*! ZSTD_invalidateMatchState() /*! ZSTD_invalidateMatchState()
@ -1455,8 +1455,8 @@ static size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, cons
static size_t ZSTD_minGain(size_t srcSize) { return (srcSize >> 6) + 2; } static size_t ZSTD_minGain(size_t srcSize) { return (srcSize >> 6) + 2; }
static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t const* prevEntropy, static size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
ZSTD_entropyCTables_t* nextEntropy, ZSTD_hufCTables_t* nextHuf,
ZSTD_strategy strategy, int disableLiteralCompression, ZSTD_strategy strategy, int disableLiteralCompression,
void* dst, size_t dstCapacity, void* dst, size_t dstCapacity,
const void* src, size_t srcSize, const void* src, size_t srcSize,
@ -1473,27 +1473,25 @@ static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t const* prevEntropy,
disableLiteralCompression); disableLiteralCompression);
/* Prepare nextEntropy assuming reusing the existing table */ /* Prepare nextEntropy assuming reusing the existing table */
nextEntropy->hufCTable_repeatMode = prevEntropy->hufCTable_repeatMode; memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable,
sizeof(prevEntropy->hufCTable));
if (disableLiteralCompression) if (disableLiteralCompression)
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize); return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
/* small ? don't even attempt compression (speed opt) */ /* small ? don't even attempt compression (speed opt) */
# define COMPRESS_LITERALS_SIZE_MIN 63 # define COMPRESS_LITERALS_SIZE_MIN 63
{ size_t const minLitSize = (prevEntropy->hufCTable_repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN; { size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize); if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
} }
if (dstCapacity < lhSize+1) return ERROR(dstSize_tooSmall); /* not enough space for compression */ if (dstCapacity < lhSize+1) return ERROR(dstSize_tooSmall); /* not enough space for compression */
{ HUF_repeat repeat = prevEntropy->hufCTable_repeatMode; { HUF_repeat repeat = prevHuf->repeatMode;
int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0; int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1; if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11, cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat, bmi2) workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2)
: HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11, : HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat, bmi2); workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2);
if (repeat != HUF_repeat_none) { if (repeat != HUF_repeat_none) {
/* reused the existing table */ /* reused the existing table */
hType = set_repeat; hType = set_repeat;
@ -1501,17 +1499,17 @@ static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t const* prevEntropy,
} }
if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) { if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable)); memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize); return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
} }
if (cLitSize==1) { if (cLitSize==1) {
memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable)); memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize); return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
} }
if (hType == set_compressed) { if (hType == set_compressed) {
/* using a newly constructed table */ /* using a newly constructed table */
nextEntropy->hufCTable_repeatMode = HUF_repeat_check; nextHuf->repeatMode = HUF_repeat_check;
} }
/* Build header */ /* Build header */
@ -1561,6 +1559,137 @@ void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
mlCodeTable[seqStorePtr->longLengthPos] = MaxML; mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
} }
/**
* -log2(x / 256) lookup table for x in [0, 256).
* If x == 0: Return 0
* Else: Return floor(-log2(x / 256) * 256)
*/
static unsigned const kInverseProbabiltyLog256[256] = {
0, 2048, 1792, 1642, 1536, 1453, 1386, 1329, 1280, 1236, 1197, 1162,
1130, 1100, 1073, 1047, 1024, 1001, 980, 960, 941, 923, 906, 889,
874, 859, 844, 830, 817, 804, 791, 779, 768, 756, 745, 734,
724, 714, 704, 694, 685, 676, 667, 658, 650, 642, 633, 626,
618, 610, 603, 595, 588, 581, 574, 567, 561, 554, 548, 542,
535, 529, 523, 517, 512, 506, 500, 495, 489, 484, 478, 473,
468, 463, 458, 453, 448, 443, 438, 434, 429, 424, 420, 415,
411, 407, 402, 398, 394, 390, 386, 382, 377, 373, 370, 366,
362, 358, 354, 350, 347, 343, 339, 336, 332, 329, 325, 322,
318, 315, 311, 308, 305, 302, 298, 295, 292, 289, 286, 282,
279, 276, 273, 270, 267, 264, 261, 258, 256, 253, 250, 247,
244, 241, 239, 236, 233, 230, 228, 225, 222, 220, 217, 215,
212, 209, 207, 204, 202, 199, 197, 194, 192, 190, 187, 185,
182, 180, 178, 175, 173, 171, 168, 166, 164, 162, 159, 157,
155, 153, 151, 149, 146, 144, 142, 140, 138, 136, 134, 132,
130, 128, 126, 123, 121, 119, 117, 115, 114, 112, 110, 108,
106, 104, 102, 100, 98, 96, 94, 93, 91, 89, 87, 85,
83, 82, 80, 78, 76, 74, 73, 71, 69, 67, 66, 64,
62, 61, 59, 57, 55, 54, 52, 50, 49, 47, 46, 44,
42, 41, 39, 37, 36, 34, 33, 31, 30, 28, 26, 25,
23, 22, 20, 19, 17, 16, 14, 13, 11, 10, 8, 7,
5, 4, 2, 1,
};
/**
* Returns the cost in bits of encoding the distribution described by count
* using the entropy bound.
*/
static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t const total)
{
unsigned cost = 0;
unsigned s;
for (s = 0; s <= max; ++s) {
unsigned norm = (unsigned)((256 * count[s]) / total);
if (count[s] != 0 && norm == 0)
norm = 1;
assert(count[s] < total);
cost += count[s] * kInverseProbabiltyLog256[norm];
}
return cost >> 8;
}
/**
* Returns the cost in bits of encoding the distribution in count using the
* table described by norm. The max symbol support by norm is assumed >= max.
* norm must be valid for every symbol with non-zero probability in count.
*/
static size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
unsigned const* count, unsigned const max)
{
unsigned const shift = 8 - accuracyLog;
size_t cost = 0;
unsigned s;
assert(accuracyLog <= 8);
for (s = 0; s <= max; ++s) {
unsigned const normAcc = norm[s] != -1 ? norm[s] : 1;
unsigned const norm256 = normAcc << shift;
assert(norm256 > 0);
assert(norm256 < 256);
cost += count[s] * kInverseProbabiltyLog256[norm256];
}
return cost >> 8;
}
static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
void const* ptr = ctable;
U16 const* u16ptr = (U16 const*)ptr;
U32 const maxSymbolValue = MEM_read16(u16ptr + 1);
return maxSymbolValue;
}
/**
* Returns the cost in bits of encoding the distribution in count using ctable.
* Returns an error if ctable cannot represent all the symbols in count.
*/
static size_t ZSTD_fseBitCost(
FSE_CTable const* ctable,
unsigned const* count,
unsigned const max)
{
unsigned const kAccuracyLog = 8;
size_t cost = 0;
unsigned s;
FSE_CState_t cstate;
FSE_initCState(&cstate, ctable);
if (ZSTD_getFSEMaxSymbolValue(ctable) < max) {
DEBUGLOG(5, "Repeat FSE_CTable has maxSymbolValue %u < %u",
ZSTD_getFSEMaxSymbolValue(ctable), max);
return ERROR(GENERIC);
}
for (s = 0; s <= max; ++s) {
unsigned const tableLog = cstate.stateLog;
unsigned const badCost = (tableLog + 1) << kAccuracyLog;
unsigned const bitCost = FSE_bitCost(cstate.symbolTT, tableLog, s, kAccuracyLog);
if (count[s] == 0)
continue;
if (bitCost >= badCost) {
DEBUGLOG(5, "Repeat FSE_CTable has Prob[%u] == 0", s);
return ERROR(GENERIC);
}
cost += count[s] * bitCost;
}
return cost >> kAccuracyLog;
}
/**
* Returns the cost in bytes of encoding the normalized count header.
* Returns an error if any of the helper functions return an error.
*/
static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
size_t const nbSeq, unsigned const FSELog)
{
BYTE wksp[FSE_NCOUNTBOUND];
S16 norm[MaxSeq + 1];
const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
CHECK_F(FSE_normalizeCount(norm, tableLog, count, nbSeq, max));
return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
}
typedef enum { typedef enum {
ZSTD_defaultDisallowed = 0, ZSTD_defaultDisallowed = 0,
ZSTD_defaultAllowed = 1 ZSTD_defaultAllowed = 1
@ -1568,28 +1697,36 @@ typedef enum {
MEM_STATIC MEM_STATIC
symbolEncodingType_e ZSTD_selectEncodingType( symbolEncodingType_e ZSTD_selectEncodingType(
FSE_repeat* repeatMode, size_t const mostFrequent, size_t nbSeq, FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
U32 defaultNormLog, ZSTD_defaultPolicy_e const isDefaultAllowed) size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
FSE_CTable const* prevCTable,
short const* defaultNorm, U32 defaultNormLog,
ZSTD_defaultPolicy_e const isDefaultAllowed,
ZSTD_strategy const strategy)
{ {
#define MIN_SEQ_FOR_DYNAMIC_FSE 64 #define MIN_SEQ_FOR_DYNAMIC_FSE 64
#define MAX_SEQ_FOR_STATIC_FSE 1000 #define MAX_SEQ_FOR_STATIC_FSE 1000
ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0); ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
if ((mostFrequent == nbSeq) && (!isDefaultAllowed || nbSeq > 2)) { if (mostFrequent == nbSeq) {
DEBUGLOG(5, "Selected set_rle"); *repeatMode = FSE_repeat_none;
if (isDefaultAllowed && nbSeq <= 2) {
/* Prefer set_basic over set_rle when there are 2 or less symbols, /* Prefer set_basic over set_rle when there are 2 or less symbols,
* since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol. * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
* If basic encoding isn't possible, always choose RLE. * If basic encoding isn't possible, always choose RLE.
*/ */
*repeatMode = FSE_repeat_check; DEBUGLOG(5, "Selected set_basic");
return set_basic;
}
DEBUGLOG(5, "Selected set_rle");
return set_rle; return set_rle;
} }
if ( isDefaultAllowed if (strategy < ZSTD_lazy) {
&& (*repeatMode == FSE_repeat_valid) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) { if (isDefaultAllowed) {
if ((*repeatMode == FSE_repeat_valid) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
DEBUGLOG(5, "Selected set_repeat"); DEBUGLOG(5, "Selected set_repeat");
return set_repeat; return set_repeat;
} }
if ( isDefaultAllowed if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (defaultNormLog-1)))) {
&& ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (defaultNormLog-1)))) ) {
DEBUGLOG(5, "Selected set_basic"); DEBUGLOG(5, "Selected set_basic");
/* The format allows default tables to be repeated, but it isn't useful. /* The format allows default tables to be repeated, but it isn't useful.
* When using simple heuristics to select encoding type, we don't want * When using simple heuristics to select encoding type, we don't want
@ -1600,6 +1737,34 @@ symbolEncodingType_e ZSTD_selectEncodingType(
*repeatMode = FSE_repeat_none; *repeatMode = FSE_repeat_none;
return set_basic; return set_basic;
} }
}
} else {
size_t const basicCost = isDefaultAllowed ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, count, max) : ERROR(GENERIC);
size_t const repeatCost = *repeatMode != FSE_repeat_none ? ZSTD_fseBitCost(prevCTable, count, max) : ERROR(GENERIC);
size_t const NCountCost = ZSTD_NCountCost(count, max, nbSeq, FSELog);
size_t const compressedCost = (NCountCost << 3) + ZSTD_entropyCost(count, max, nbSeq);
if (isDefaultAllowed) {
assert(!ZSTD_isError(basicCost));
assert(!(*repeatMode == FSE_repeat_valid && ZSTD_isError(repeatCost)));
}
assert(!ZSTD_isError(NCountCost));
assert(compressedCost < ERROR(maxCode));
DEBUGLOG(5, "Estimated bit costs: basic=%u\trepeat=%u\tcompressed=%u",
(U32)basicCost, (U32)repeatCost, (U32)compressedCost);
if (basicCost <= repeatCost && basicCost <= compressedCost) {
DEBUGLOG(5, "Selected set_basic");
assert(isDefaultAllowed);
*repeatMode = FSE_repeat_none;
return set_basic;
}
if (repeatCost <= compressedCost) {
DEBUGLOG(5, "Selected set_repeat");
assert(!ZSTD_isError(repeatCost));
return set_repeat;
}
assert(compressedCost < basicCost && compressedCost < repeatCost);
}
DEBUGLOG(5, "Selected set_compressed"); DEBUGLOG(5, "Selected set_compressed");
*repeatMode = FSE_repeat_check; *repeatMode = FSE_repeat_check;
return set_compressed; return set_compressed;
@ -1803,10 +1968,11 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
const int bmi2) const int bmi2)
{ {
const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN; const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
ZSTD_strategy const strategy = cctxParams->cParams.strategy;
U32 count[MaxSeq+1]; U32 count[MaxSeq+1];
FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable; FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable; FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable; FSE_CTable* CTable_MatchLength = nextEntropy->fse.matchlengthCTable;
U32 LLtype, Offtype, MLtype; /* compressed, raw or rle */ U32 LLtype, Offtype, MLtype; /* compressed, raw or rle */
const seqDef* const sequences = seqStorePtr->sequencesStart; const seqDef* const sequences = seqStorePtr->sequencesStart;
const BYTE* const ofCodeTable = seqStorePtr->ofCode; const BYTE* const ofCodeTable = seqStorePtr->ofCode;
@ -1824,7 +1990,7 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
{ const BYTE* const literals = seqStorePtr->litStart; { const BYTE* const literals = seqStorePtr->litStart;
size_t const litSize = seqStorePtr->lit - literals; size_t const litSize = seqStorePtr->lit - literals;
size_t const cSize = ZSTD_compressLiterals( size_t const cSize = ZSTD_compressLiterals(
prevEntropy, nextEntropy, &prevEntropy->huf, &nextEntropy->huf,
cctxParams->cParams.strategy, cctxParams->disableLiteralCompression, cctxParams->cParams.strategy, cctxParams->disableLiteralCompression,
op, dstCapacity, op, dstCapacity,
literals, litSize, literals, litSize,
@ -1844,12 +2010,8 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
else else
op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3; op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
if (nbSeq==0) { if (nbSeq==0) {
memcpy(nextEntropy->litlengthCTable, prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable)); /* Copy the old tables over as if we repeated them */
nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode; memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
memcpy(nextEntropy->offcodeCTable, prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable));
nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
memcpy(nextEntropy->matchlengthCTable, prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable));
nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
return op - ostart; return op - ostart;
} }
@ -1862,11 +2024,13 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
{ U32 max = MaxLL; { U32 max = MaxLL;
size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, workspace); size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, workspace);
DEBUGLOG(5, "Building LL table"); DEBUGLOG(5, "Building LL table");
nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode; nextEntropy->fse.litlength_repeatMode = prevEntropy->fse.litlength_repeatMode;
LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode, mostFrequent, nbSeq, LL_defaultNormLog, ZSTD_defaultAllowed); 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, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype, { size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL, count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable), prevEntropy->fse.litlengthCTable, sizeof(prevEntropy->fse.litlengthCTable),
workspace, HUF_WORKSPACE_SIZE); workspace, HUF_WORKSPACE_SIZE);
if (ZSTD_isError(countSize)) return countSize; if (ZSTD_isError(countSize)) return countSize;
op += countSize; op += countSize;
@ -1877,11 +2041,12 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
/* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */ /* 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; ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
DEBUGLOG(5, "Building OF table"); DEBUGLOG(5, "Building OF table");
nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode; nextEntropy->fse.offcode_repeatMode = prevEntropy->fse.offcode_repeatMode;
Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode, mostFrequent, nbSeq, OF_defaultNormLog, defaultPolicy); 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, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype, { size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff, count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable), prevEntropy->fse.offcodeCTable, sizeof(prevEntropy->fse.offcodeCTable),
workspace, HUF_WORKSPACE_SIZE); workspace, HUF_WORKSPACE_SIZE);
if (ZSTD_isError(countSize)) return countSize; if (ZSTD_isError(countSize)) return countSize;
op += countSize; op += countSize;
@ -1890,11 +2055,12 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
{ U32 max = MaxML; { U32 max = MaxML;
size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, workspace); size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, workspace);
DEBUGLOG(5, "Building ML table"); DEBUGLOG(5, "Building ML table");
nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode; nextEntropy->fse.matchlength_repeatMode = prevEntropy->fse.matchlength_repeatMode;
MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode, mostFrequent, nbSeq, ML_defaultNormLog, ZSTD_defaultAllowed); 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, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype, { size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML, count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable), prevEntropy->fse.matchlengthCTable, sizeof(prevEntropy->fse.matchlengthCTable),
workspace, HUF_WORKSPACE_SIZE); workspace, HUF_WORKSPACE_SIZE);
if (ZSTD_isError(countSize)) return countSize; if (ZSTD_isError(countSize)) return countSize;
op += countSize; op += countSize;
@ -1942,8 +2108,8 @@ MEM_STATIC size_t ZSTD_compressSequences(seqStore_t* seqStorePtr,
* block. After the first block, the offcode table might not have large * block. After the first block, the offcode table might not have large
* enough codes to represent the offsets in the data. * enough codes to represent the offsets in the data.
*/ */
if (nextEntropy->offcode_repeatMode == FSE_repeat_valid) if (nextEntropy->fse.offcode_repeatMode == FSE_repeat_valid)
nextEntropy->offcode_repeatMode = FSE_repeat_check; nextEntropy->fse.offcode_repeatMode = FSE_repeat_check;
return cSize; return cSize;
} }
@ -2384,7 +2550,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
dictPtr += 4; dictPtr += 4;
{ unsigned maxSymbolValue = 255; { unsigned maxSymbolValue = 255;
size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.hufCTable, &maxSymbolValue, dictPtr, dictEnd-dictPtr); size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.huf.CTable, &maxSymbolValue, dictPtr, dictEnd-dictPtr);
if (HUF_isError(hufHeaderSize)) return ERROR(dictionary_corrupted); if (HUF_isError(hufHeaderSize)) return ERROR(dictionary_corrupted);
if (maxSymbolValue < 255) return ERROR(dictionary_corrupted); if (maxSymbolValue < 255) return ERROR(dictionary_corrupted);
dictPtr += hufHeaderSize; dictPtr += hufHeaderSize;
@ -2396,7 +2562,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted); if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
/* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */ /* 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 */ /* fill all offset symbols to avoid garbage at end of table */
CHECK_E( FSE_buildCTable_wksp(bs->entropy.offcodeCTable, offcodeNCount, MaxOff, offcodeLog, workspace, HUF_WORKSPACE_SIZE), CHECK_E( FSE_buildCTable_wksp(bs->entropy.fse.offcodeCTable, offcodeNCount, MaxOff, offcodeLog, workspace, HUF_WORKSPACE_SIZE),
dictionary_corrupted); dictionary_corrupted);
dictPtr += offcodeHeaderSize; dictPtr += offcodeHeaderSize;
} }
@ -2408,7 +2574,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted); if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
/* Every match length code must have non-zero probability */ /* Every match length code must have non-zero probability */
CHECK_F( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML)); CHECK_F( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
CHECK_E( FSE_buildCTable_wksp(bs->entropy.matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, workspace, HUF_WORKSPACE_SIZE), CHECK_E( FSE_buildCTable_wksp(bs->entropy.fse.matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, workspace, HUF_WORKSPACE_SIZE),
dictionary_corrupted); dictionary_corrupted);
dictPtr += matchlengthHeaderSize; dictPtr += matchlengthHeaderSize;
} }
@ -2420,7 +2586,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted); if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
/* Every literal length code must have non-zero probability */ /* Every literal length code must have non-zero probability */
CHECK_F( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL)); CHECK_F( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
CHECK_E( FSE_buildCTable_wksp(bs->entropy.litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, workspace, HUF_WORKSPACE_SIZE), CHECK_E( FSE_buildCTable_wksp(bs->entropy.fse.litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, workspace, HUF_WORKSPACE_SIZE),
dictionary_corrupted); dictionary_corrupted);
dictPtr += litlengthHeaderSize; dictPtr += litlengthHeaderSize;
} }
@ -2446,10 +2612,10 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
if (bs->rep[u] > dictContentSize) return ERROR(dictionary_corrupted); if (bs->rep[u] > dictContentSize) return ERROR(dictionary_corrupted);
} } } }
bs->entropy.hufCTable_repeatMode = HUF_repeat_valid; bs->entropy.huf.repeatMode = HUF_repeat_valid;
bs->entropy.offcode_repeatMode = FSE_repeat_valid; bs->entropy.fse.offcode_repeatMode = FSE_repeat_valid;
bs->entropy.matchlength_repeatMode = FSE_repeat_valid; bs->entropy.fse.matchlength_repeatMode = FSE_repeat_valid;
bs->entropy.litlength_repeatMode = FSE_repeat_valid; bs->entropy.fse.litlength_repeatMode = FSE_repeat_valid;
CHECK_F(ZSTD_loadDictionaryContent(ms, params, dictPtr, dictContentSize, dtlm)); CHECK_F(ZSTD_loadDictionaryContent(ms, params, dictPtr, dictContentSize, dtlm));
return dictID; return dictID;
} }

12
lib/compress/zstd_compress_internal.h
View File

@ -53,14 +53,22 @@ typedef struct ZSTD_prefixDict_s {
} ZSTD_prefixDict; } ZSTD_prefixDict;
typedef struct { typedef struct {
U32 hufCTable[HUF_CTABLE_SIZE_U32(255)]; U32 CTable[HUF_CTABLE_SIZE_U32(255)];
HUF_repeat repeatMode;
} ZSTD_hufCTables_t;
typedef struct {
FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)]; FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)]; FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)]; FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
HUF_repeat hufCTable_repeatMode;
FSE_repeat offcode_repeatMode; FSE_repeat offcode_repeatMode;
FSE_repeat matchlength_repeatMode; FSE_repeat matchlength_repeatMode;
FSE_repeat litlength_repeatMode; FSE_repeat litlength_repeatMode;
} ZSTD_fseCTables_t;
typedef struct {
ZSTD_hufCTables_t huf;
ZSTD_fseCTables_t fse;
} ZSTD_entropyCTables_t; } ZSTD_entropyCTables_t;
typedef struct { typedef struct {

22
lib/compress/zstd_opt.c
View File

@ -39,7 +39,7 @@ static void ZSTD_rescaleFreqs(optState_t* const optPtr,
optPtr->priceType = zop_predef; optPtr->priceType = zop_predef;
assert(optPtr->symbolCosts != NULL); assert(optPtr->symbolCosts != NULL);
if (optPtr->symbolCosts->hufCTable_repeatMode == HUF_repeat_valid) { /* huffman table presumed generated by dictionary */ if (optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid) { /* huffman table presumed generated by dictionary */
if (srcSize <= 8192) /* heuristic */ if (srcSize <= 8192) /* heuristic */
optPtr->priceType = zop_static; optPtr->priceType = zop_static;
else { else {
@ -52,7 +52,7 @@ static void ZSTD_rescaleFreqs(optState_t* const optPtr,
{ unsigned lit; { unsigned lit;
for (lit=0; lit<=MaxLit; lit++) { for (lit=0; lit<=MaxLit; lit++) {
U32 const scaleLog = 11; /* scale to 2K */ U32 const scaleLog = 11; /* scale to 2K */
U32 const bitCost = HUF_getNbBits(optPtr->symbolCosts->hufCTable, lit); U32 const bitCost = HUF_getNbBits(optPtr->symbolCosts->huf.CTable, lit);
assert(bitCost <= scaleLog); assert(bitCost <= scaleLog);
optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/; optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
optPtr->litSum += optPtr->litFreq[lit]; optPtr->litSum += optPtr->litFreq[lit];
@ -60,7 +60,7 @@ static void ZSTD_rescaleFreqs(optState_t* const optPtr,
{ unsigned ll; { unsigned ll;
FSE_CState_t llstate; FSE_CState_t llstate;
FSE_initCState(&llstate, optPtr->symbolCosts->litlengthCTable); FSE_initCState(&llstate, optPtr->symbolCosts->fse.litlengthCTable);
optPtr->litLengthSum = 0; optPtr->litLengthSum = 0;
for (ll=0; ll<=MaxLL; ll++) { for (ll=0; ll<=MaxLL; ll++) {
U32 const scaleLog = 10; /* scale to 1K */ U32 const scaleLog = 10; /* scale to 1K */
@ -72,7 +72,7 @@ static void ZSTD_rescaleFreqs(optState_t* const optPtr,
{ unsigned ml; { unsigned ml;
FSE_CState_t mlstate; FSE_CState_t mlstate;
FSE_initCState(&mlstate, optPtr->symbolCosts->matchlengthCTable); FSE_initCState(&mlstate, optPtr->symbolCosts->fse.matchlengthCTable);
optPtr->matchLengthSum = 0; optPtr->matchLengthSum = 0;
for (ml=0; ml<=MaxML; ml++) { for (ml=0; ml<=MaxML; ml++) {
U32 const scaleLog = 10; U32 const scaleLog = 10;
@ -84,7 +84,7 @@ static void ZSTD_rescaleFreqs(optState_t* const optPtr,
{ unsigned of; { unsigned of;
FSE_CState_t ofstate; FSE_CState_t ofstate;
FSE_initCState(&ofstate, optPtr->symbolCosts->offcodeCTable); FSE_initCState(&ofstate, optPtr->symbolCosts->fse.offcodeCTable);
optPtr->offCodeSum = 0; optPtr->offCodeSum = 0;
for (of=0; of<=MaxOff; of++) { for (of=0; of<=MaxOff; of++) {
U32 const scaleLog = 10; U32 const scaleLog = 10;
@ -180,9 +180,9 @@ static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
if (optPtr->priceType == zop_static) { if (optPtr->priceType == zop_static) {
U32 u, cost; U32 u, cost;
assert(optPtr->symbolCosts != NULL); assert(optPtr->symbolCosts != NULL);
assert(optPtr->symbolCosts->hufCTable_repeatMode == HUF_repeat_valid); assert(optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid);
for (u=0, cost=0; u < litLength; u++) for (u=0, cost=0; u < litLength; u++)
cost += HUF_getNbBits(optPtr->symbolCosts->hufCTable, literals[u]); cost += HUF_getNbBits(optPtr->symbolCosts->huf.CTable, literals[u]);
return cost * BITCOST_MULTIPLIER; return cost * BITCOST_MULTIPLIER;
} }
@ -202,7 +202,7 @@ static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optP
if (optPtr->priceType == zop_static) { if (optPtr->priceType == zop_static) {
U32 const llCode = ZSTD_LLcode(litLength); U32 const llCode = ZSTD_LLcode(litLength);
FSE_CState_t cstate; FSE_CState_t cstate;
FSE_initCState(&cstate, optPtr->symbolCosts->litlengthCTable); FSE_initCState(&cstate, optPtr->symbolCosts->fse.litlengthCTable);
{ U32 const price = LL_bits[llCode]*BITCOST_MULTIPLIER + BITCOST_SYMBOL(cstate.symbolTT, cstate.stateLog, llCode); { U32 const price = LL_bits[llCode]*BITCOST_MULTIPLIER + BITCOST_SYMBOL(cstate.symbolTT, cstate.stateLog, llCode);
DEBUGLOG(8, "ZSTD_litLengthPrice: ll=%u, bitCost=%.2f", litLength, (double)price / BITCOST_MULTIPLIER); DEBUGLOG(8, "ZSTD_litLengthPrice: ll=%u, bitCost=%.2f", litLength, (double)price / BITCOST_MULTIPLIER);
return price; return price;
@ -234,7 +234,7 @@ static int ZSTD_litLengthContribution(U32 const litLength, const optState_t* con
if (optPtr->priceType == zop_static) { if (optPtr->priceType == zop_static) {
U32 const llCode = ZSTD_LLcode(litLength); U32 const llCode = ZSTD_LLcode(litLength);
FSE_CState_t cstate; FSE_CState_t cstate;
FSE_initCState(&cstate, optPtr->symbolCosts->litlengthCTable); FSE_initCState(&cstate, optPtr->symbolCosts->fse.litlengthCTable);
return (int)(LL_bits[llCode] * BITCOST_MULTIPLIER) return (int)(LL_bits[llCode] * BITCOST_MULTIPLIER)
+ BITCOST_SYMBOL(cstate.symbolTT, cstate.stateLog, llCode) + BITCOST_SYMBOL(cstate.symbolTT, cstate.stateLog, llCode)
- BITCOST_SYMBOL(cstate.symbolTT, cstate.stateLog, 0); - BITCOST_SYMBOL(cstate.symbolTT, cstate.stateLog, 0);
@ -284,8 +284,8 @@ ZSTD_getMatchPrice(U32 const offset, U32 const matchLength,
if (optPtr->priceType == zop_static) { if (optPtr->priceType == zop_static) {
U32 const mlCode = ZSTD_MLcode(mlBase); U32 const mlCode = ZSTD_MLcode(mlBase);
FSE_CState_t mlstate, offstate; FSE_CState_t mlstate, offstate;
FSE_initCState(&mlstate, optPtr->symbolCosts->matchlengthCTable); FSE_initCState(&mlstate, optPtr->symbolCosts->fse.matchlengthCTable);
FSE_initCState(&offstate, optPtr->symbolCosts->offcodeCTable); FSE_initCState(&offstate, optPtr->symbolCosts->fse.offcodeCTable);
return BITCOST_SYMBOL(offstate.symbolTT, offstate.stateLog, offCode) + offCode*BITCOST_MULTIPLIER return BITCOST_SYMBOL(offstate.symbolTT, offstate.stateLog, offCode) + offCode*BITCOST_MULTIPLIER
+ BITCOST_SYMBOL(mlstate.symbolTT, mlstate.stateLog, mlCode) + ML_bits[mlCode]*BITCOST_MULTIPLIER; + BITCOST_SYMBOL(mlstate.symbolTT, mlstate.stateLog, mlCode) + ML_bits[mlCode]*BITCOST_MULTIPLIER;
} }