marsultor/email-tracker
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

should be it

Browse Source
This commit is contained in:
Mars Ultor
2025-10-24 19:21:19 -05:00
parent a4b23fc57c
commit f09560c7b1
14047 changed files with 3161551 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

628
external/duckdb/third_party/zstd/compress/fse_compress.cpp vendored Normal file
View File

@@ -0,0 +1,628 @@
/* ******************************************************************
* FSE : Finite State Entropy encoder
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
* - Public forum : https://groups.google.com/forum/#!forum/lz4c
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/* **************************************************************
* Includes
****************************************************************/
#include "zstd/common/compiler.h"
#include "zstd/common/mem.h" /* U32, U16, etc. */
#include "zstd/common/debug.h" /* assert, DEBUGLOG */
#include "zstd/compress/hist.h" /* HIST_count_wksp */
#include "zstd/common/bitstream.h"
#define FSE_STATIC_LINKING_ONLY
#include "zstd/common/fse.h"
#include "zstd/common/error_private.h"
#define ZSTD_DEPS_NEED_MALLOC
#define ZSTD_DEPS_NEED_MATH64
#include "zstd/common/zstd_deps.h" /* ZSTD_memset */
#include "zstd/common/bits.h" /* ZSTD_highbit32 */
namespace duckdb_zstd {
/* **************************************************************
* Error Management
****************************************************************/
#define FSE_isError ERR_isError
/* **************************************************************
* Templates
****************************************************************/
/*
designed to be included
for type-specific functions (template emulation in C)
Objective is to write these functions only once, for improved maintenance
*/
/* safety checks */
#ifndef FSE_FUNCTION_EXTENSION
# error "FSE_FUNCTION_EXTENSION must be defined"
#endif
#ifndef FSE_FUNCTION_TYPE
# error "FSE_FUNCTION_TYPE must be defined"
#endif
/* Function names */
#define FSE_CAT(X,Y) X##Y
#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
/* Function templates */
/* FSE_buildCTable_wksp() :
* Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
* wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
* workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
*/
size_t FSE_buildCTable_wksp(FSE_CTable* ct,
const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
void* workSpace, size_t wkspSize)
{
U32 const tableSize = 1 << tableLog;
U32 const tableMask = tableSize - 1;
void* const ptr = ct;
U16* const tableU16 = ( (U16*) ptr) + 2;
void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
U32 const step = FSE_TABLESTEP(tableSize);
U32 const maxSV1 = maxSymbolValue+1;
U16* cumul = (U16*)workSpace; /* size = maxSV1 */
FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSV1+1)); /* size = tableSize */
U32 highThreshold = tableSize-1;
assert(((size_t)workSpace & 1) == 0); /* Must be 2 bytes-aligned */
if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
/* CTable header */
tableU16[-2] = (U16) tableLog;
tableU16[-1] = (U16) maxSymbolValue;
assert(tableLog < 16); /* required for threshold strategy to work */
/* For explanations on how to distribute symbol values over the table :
* https://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
#ifdef __clang_analyzer__
ZSTD_memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize); /* useless initialization, just to keep scan-build happy */
#endif
/* symbol start positions */
{ U32 u;
cumul[0] = 0;
for (u=1; u <= maxSV1; u++) {
if (normalizedCounter[u-1]==-1) { /* Low proba symbol */
cumul[u] = cumul[u-1] + 1;
tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
} else {
assert(normalizedCounter[u-1] >= 0);
cumul[u] = cumul[u-1] + (U16)normalizedCounter[u-1];
assert(cumul[u] >= cumul[u-1]); /* no overflow */
} }
cumul[maxSV1] = (U16)(tableSize+1);
}
/* Spread symbols */
if (highThreshold == tableSize - 1) {
/* Case for no low prob count symbols. Lay down 8 bytes at a time
* to reduce branch misses since we are operating on a small block
*/
BYTE* const spread = tableSymbol + tableSize; /* size = tableSize + 8 (may write beyond tableSize) */
{ U64 const add = 0x0101010101010101ull;
size_t pos = 0;
U64 sv = 0;
U32 s;
for (s=0; s<maxSV1; ++s, sv += add) {
int i;
int const n = normalizedCounter[s];
MEM_write64(spread + pos, sv);
for (i = 8; i < n; i += 8) {
MEM_write64(spread + pos + i, sv);
}
assert(n>=0);
pos += (size_t)n;
}
}
/* Spread symbols across the table. Lack of lowprob symbols means that
* we don't need variable sized inner loop, so we can unroll the loop and
* reduce branch misses.
*/
{ size_t position = 0;
size_t s;
size_t const unroll = 2; /* Experimentally determined optimal unroll */
assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
for (s = 0; s < (size_t)tableSize; s += unroll) {
size_t u;
for (u = 0; u < unroll; ++u) {
size_t const uPosition = (position + (u * step)) & tableMask;
tableSymbol[uPosition] = spread[s + u];
}
position = (position + (unroll * step)) & tableMask;
}
assert(position == 0); /* Must have initialized all positions */
}
} else {
U32 position = 0;
U32 symbol;
for (symbol=0; symbol<maxSV1; symbol++) {
int nbOccurrences;
int const freq = normalizedCounter[symbol];
for (nbOccurrences=0; nbOccurrences<freq; nbOccurrences++) {
tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
position = (position + step) & tableMask;
while (position > highThreshold)
position = (position + step) & tableMask; /* Low proba area */
} }
assert(position==0); /* Must have initialized all positions */
}
/* Build table */
{ U32 u; for (u=0; u<tableSize; u++) {
FSE_FUNCTION_TYPE s = tableSymbol[u]; /* note : static analyzer may not understand tableSymbol is properly initialized */
tableU16[cumul[s]++] = (U16) (tableSize+u); /* TableU16 : sorted by symbol order; gives next state value */
} }
/* Build Symbol Transformation Table */
{ unsigned total = 0;
unsigned s;
for (s=0; s<=maxSymbolValue; s++) {
switch (normalizedCounter[s])
{
case 0:
/* filling nonetheless, for compatibility with FSE_getMaxNbBits() */
symbolTT[s].deltaNbBits = ((tableLog+1) << 16) - (1<<tableLog);
break;
case -1:
case 1:
symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
assert(total <= INT_MAX);
symbolTT[s].deltaFindState = (int)(total - 1);
total ++;
break;
default :
assert(normalizedCounter[s] > 1);
{ U32 const maxBitsOut = tableLog - ZSTD_highbit32 ((U32)normalizedCounter[s]-1);
U32 const minStatePlus = (U32)normalizedCounter[s] << maxBitsOut;
symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
symbolTT[s].deltaFindState = (int)(total - (unsigned)normalizedCounter[s]);
total += (unsigned)normalizedCounter[s];
} } } }
#if 0 /* debug : symbol costs */
DEBUGLOG(5, "\n --- table statistics : ");
{ U32 symbol;
for (symbol=0; symbol<=maxSymbolValue; symbol++) {
DEBUGLOG(5, "%3u: w=%3i, maxBits=%u, fracBits=%.2f",
symbol, normalizedCounter[symbol],
FSE_getMaxNbBits(symbolTT, symbol),
(double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
} }
#endif
return 0;
}
#ifndef FSE_COMMONDEFS_ONLY
/*-**************************************************************
* FSE NCount encoding
****************************************************************/
size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
{
size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog
+ 4 /* bitCount initialized at 4 */
+ 2 /* first two symbols may use one additional bit each */) / 8)
+ 1 /* round up to whole nb bytes */
+ 2 /* additional two bytes for bitstream flush */;
return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND; /* maxSymbolValue==0 ? use default */
}
static size_t
FSE_writeNCount_generic (void* header, size_t headerBufferSize,
const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
unsigned writeIsSafe)
{
BYTE* const ostart = (BYTE*) header;
BYTE* out = ostart;
BYTE* const oend = ostart + headerBufferSize;
int nbBits;
const int tableSize = 1 << tableLog;
int remaining;
int threshold;
U32 bitStream = 0;
int bitCount = 0;
unsigned symbol = 0;
unsigned const alphabetSize = maxSymbolValue + 1;
int previousIs0 = 0;
/* Table Size */
bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
bitCount += 4;
/* Init */
remaining = tableSize+1; /* +1 for extra accuracy */
threshold = tableSize;
nbBits = (int)tableLog+1;
while ((symbol < alphabetSize) && (remaining>1)) { /* stops at 1 */
if (previousIs0) {
unsigned start = symbol;
while ((symbol < alphabetSize) && !normalizedCounter[symbol]) symbol++;
if (symbol == alphabetSize) break; /* incorrect distribution */
while (symbol >= start+24) {
start+=24;
bitStream += 0xFFFFU << bitCount;
if ((!writeIsSafe) && (out > oend-2))
return ERROR(dstSize_tooSmall); /* Buffer overflow */
out[0] = (BYTE) bitStream;
out[1] = (BYTE)(bitStream>>8);
out+=2;
bitStream>>=16;
}
while (symbol >= start+3) {
start+=3;
bitStream += 3U << bitCount;
bitCount += 2;
}
bitStream += (symbol-start) << bitCount;
bitCount += 2;
if (bitCount>16) {
if ((!writeIsSafe) && (out > oend - 2))
return ERROR(dstSize_tooSmall); /* Buffer overflow */
out[0] = (BYTE)bitStream;
out[1] = (BYTE)(bitStream>>8);
out += 2;
bitStream >>= 16;
bitCount -= 16;
} }
{ int count = normalizedCounter[symbol++];
int const max = (2*threshold-1) - remaining;
remaining -= count < 0 ? -count : count;
count++; /* +1 for extra accuracy */
if (count>=threshold)
count += max; /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
bitStream += (U32)count << bitCount;
bitCount += nbBits;
bitCount -= (count<max);
previousIs0 = (count==1);
if (remaining<1) return ERROR(GENERIC);
while (remaining<threshold) { nbBits--; threshold>>=1; }
}
if (bitCount>16) {
if ((!writeIsSafe) && (out > oend - 2))
return ERROR(dstSize_tooSmall); /* Buffer overflow */
out[0] = (BYTE)bitStream;
out[1] = (BYTE)(bitStream>>8);
out += 2;
bitStream >>= 16;
bitCount -= 16;
} }
if (remaining != 1)
return ERROR(GENERIC); /* incorrect normalized distribution */
assert(symbol <= alphabetSize);
/* flush remaining bitStream */
if ((!writeIsSafe) && (out > oend - 2))
return ERROR(dstSize_tooSmall); /* Buffer overflow */
out[0] = (BYTE)bitStream;
out[1] = (BYTE)(bitStream>>8);
out+= (bitCount+7) /8;
assert(out >= ostart);
return (size_t)(out-ostart);
}
size_t FSE_writeNCount (void* buffer, size_t bufferSize,
const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
{
if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge); /* Unsupported */
if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC); /* Unsupported */
if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1 /* write in buffer is safe */);
}
/*-**************************************************************
* FSE Compression Code
****************************************************************/
/* provides the minimum logSize to safely represent a distribution */
static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
{
U32 minBitsSrc = ZSTD_highbit32((U32)(srcSize)) + 1;
U32 minBitsSymbols = ZSTD_highbit32(maxSymbolValue) + 2;
U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
assert(srcSize > 1); /* Not supported, RLE should be used instead */
return minBits;
}
unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
{
U32 maxBitsSrc = ZSTD_highbit32((U32)(srcSize - 1)) - minus;
U32 tableLog = maxTableLog;
U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
assert(srcSize > 1); /* Not supported, RLE should be used instead */
if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
if (maxBitsSrc < tableLog) tableLog = maxBitsSrc; /* Accuracy can be reduced */
if (minBits > tableLog) tableLog = minBits; /* Need a minimum to safely represent all symbol values */
if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
return tableLog;
}
unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
{
return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
}
/* Secondary normalization method.
To be used when primary method fails. */
static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue, short lowProbCount)
{
short const NOT_YET_ASSIGNED = -2;
U32 s;
U32 distributed = 0;
U32 ToDistribute;
/* Init */
U32 const lowThreshold = (U32)(total >> tableLog);
U32 lowOne = (U32)((total * 3) >> (tableLog + 1));
for (s=0; s<=maxSymbolValue; s++) {
if (count[s] == 0) {
norm[s]=0;
continue;
}
if (count[s] <= lowThreshold) {
norm[s] = lowProbCount;
distributed++;
total -= count[s];
continue;
}
if (count[s] <= lowOne) {
norm[s] = 1;
distributed++;
total -= count[s];
continue;
}
norm[s]=NOT_YET_ASSIGNED;
}
ToDistribute = (1 << tableLog) - distributed;
if (ToDistribute == 0)
return 0;
if ((total / ToDistribute) > lowOne) {
/* risk of rounding to zero */
lowOne = (U32)((total * 3) / (ToDistribute * 2));
for (s=0; s<=maxSymbolValue; s++) {
if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
norm[s] = 1;
distributed++;
total -= count[s];
continue;
} }
ToDistribute = (1 << tableLog) - distributed;
}
if (distributed == maxSymbolValue+1) {
/* all values are pretty poor;
probably incompressible data (should have already been detected);
find max, then give all remaining points to max */
U32 maxV = 0, maxC = 0;
for (s=0; s<=maxSymbolValue; s++)
if (count[s] > maxC) { maxV=s; maxC=count[s]; }
norm[maxV] += (short)ToDistribute;
return 0;
}
if (total == 0) {
/* all of the symbols were low enough for the lowOne or lowThreshold */
for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
if (norm[s] > 0) { ToDistribute--; norm[s]++; }
return 0;
}
{ U64 const vStepLog = 62 - tableLog;
U64 const mid = (1ULL << (vStepLog-1)) - 1;
U64 const rStep = ZSTD_div64((((U64)1<<vStepLog) * ToDistribute) + mid, (U32)total); /* scale on remaining */
U64 tmpTotal = mid;
for (s=0; s<=maxSymbolValue; s++) {
if (norm[s]==NOT_YET_ASSIGNED) {
U64 const end = tmpTotal + (count[s] * rStep);
U32 const sStart = (U32)(tmpTotal >> vStepLog);
U32 const sEnd = (U32)(end >> vStepLog);
U32 const weight = sEnd - sStart;
if (weight < 1)
return ERROR(GENERIC);
norm[s] = (short)weight;
tmpTotal = end;
} } }
return 0;
}
size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
const unsigned* count, size_t total,
unsigned maxSymbolValue, unsigned useLowProbCount)
{
/* Sanity checks */
if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC); /* Unsupported size */
if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge); /* Unsupported size */
if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC); /* Too small tableLog, compression potentially impossible */
{ static U32 const rtbTable[] = { 0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
short const lowProbCount = useLowProbCount ? -1 : 1;
U64 const scale = 62 - tableLog;
U64 const step = ZSTD_div64((U64)1<<62, (U32)total); /* <== here, one division ! */
U64 const vStep = 1ULL<<(scale-20);
int stillToDistribute = 1<<tableLog;
unsigned s;
unsigned largest=0;
short largestP=0;
U32 lowThreshold = (U32)(total >> tableLog);
for (s=0; s<=maxSymbolValue; s++) {
if (count[s] == total) return 0; /* rle special case */
if (count[s] == 0) { normalizedCounter[s]=0; continue; }
if (count[s] <= lowThreshold) {
normalizedCounter[s] = lowProbCount;
stillToDistribute--;
} else {
short proba = (short)((count[s]*step) >> scale);
if (proba<8) {
U64 restToBeat = vStep * rtbTable[proba];
proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
}
if (proba > largestP) { largestP=proba; largest=s; }
normalizedCounter[s] = proba;
stillToDistribute -= proba;
} }
if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
/* corner case, need another normalization method */
size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue, lowProbCount);
if (FSE_isError(errorCode)) return errorCode;
}
else normalizedCounter[largest] += (short)stillToDistribute;
}
#if 0
{ /* Print Table (debug) */
U32 s;
U32 nTotal = 0;
for (s=0; s<=maxSymbolValue; s++)
RAWLOG(2, "%3i: %4i \n", s, normalizedCounter[s]);
for (s=0; s<=maxSymbolValue; s++)
nTotal += abs(normalizedCounter[s]);
if (nTotal != (1U<<tableLog))
RAWLOG(2, "Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
getchar();
}
#endif
return tableLog;
}
/* fake FSE_CTable, for rle input (always same symbol) */
size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
{
void* ptr = ct;
U16* tableU16 = ( (U16*) ptr) + 2;
void* FSCTptr = (U32*)ptr + 2;
FSE_symbolCompressionTransform* symbolTT = (FSE_symbolCompressionTransform*) FSCTptr;
/* header */
tableU16[-2] = (U16) 0;
tableU16[-1] = (U16) symbolValue;
/* Build table */
tableU16[0] = 0;
tableU16[1] = 0; /* just in case */
/* Build Symbol Transformation Table */
symbolTT[symbolValue].deltaNbBits = 0;
symbolTT[symbolValue].deltaFindState = 0;
return 0;
}
static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
const void* src, size_t srcSize,
const FSE_CTable* ct, const unsigned fast)
{
const BYTE* const istart = (const BYTE*) src;
const BYTE* const iend = istart + srcSize;
const BYTE* ip=iend;
BIT_CStream_t bitC;
FSE_CState_t CState1, CState2;
/* init */
if (srcSize <= 2) return 0;
{ size_t const initError = BIT_initCStream(&bitC, dst, dstSize);
if (FSE_isError(initError)) return 0; /* not enough space available to write a bitstream */ }
#define FSE_FLUSHBITS(s) (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
if (srcSize & 1) {
FSE_initCState2(&CState1, ct, *--ip);
FSE_initCState2(&CState2, ct, *--ip);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
FSE_FLUSHBITS(&bitC);
} else {
FSE_initCState2(&CState2, ct, *--ip);
FSE_initCState2(&CState1, ct, *--ip);
}
/* join to mod 4 */
srcSize -= 2;
if ((sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) && (srcSize & 2)) { /* test bit 2 */
FSE_encodeSymbol(&bitC, &CState2, *--ip);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
FSE_FLUSHBITS(&bitC);
}
/* 2 or 4 encoding per loop */
while ( ip>istart ) {
FSE_encodeSymbol(&bitC, &CState2, *--ip);
if (sizeof(bitC.bitContainer)*8 < FSE_MAX_TABLELOG*2+7 ) /* this test must be static */
FSE_FLUSHBITS(&bitC);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
if (sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) { /* this test must be static */
FSE_encodeSymbol(&bitC, &CState2, *--ip);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
}
FSE_FLUSHBITS(&bitC);
}
FSE_flushCState(&bitC, &CState2);
FSE_flushCState(&bitC, &CState1);
return BIT_closeCStream(&bitC);
}
size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
const void* src, size_t srcSize,
const FSE_CTable* ct)
{
unsigned const fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
if (fast)
return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 1);
else
return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 0);
}
size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
#endif /* FSE_COMMONDEFS_ONLY */
} // namespace duckdb_zstd

184
external/duckdb/third_party/zstd/compress/hist.cpp vendored Normal file
View File

@@ -0,0 +1,184 @@
/* ******************************************************************
* hist : Histogram functions
* part of Finite State Entropy project
* Copyright (c) Meta Platforms, Inc. and affiliates.
*
* You can contact the author at :
* - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
* - Public forum : https://groups.google.com/forum/#!forum/lz4c
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
****************************************************************** */
/* --- dependencies --- */
#include "zstd/common/mem.h" /* U32, BYTE, etc. */
#include "zstd/common/debug.h" /* assert, DEBUGLOG */
#include "zstd/common/error_private.h" /* ERROR */
#include "zstd/compress/hist.h"
namespace duckdb_zstd {
/* --- Error management --- */
unsigned HIST_isError(size_t code) { return ERR_isError(code); }
/*-**************************************************************
* Histogram functions
****************************************************************/
unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize)
{
const BYTE* ip = (const BYTE*)src;
const BYTE* const end = ip + srcSize;
unsigned maxSymbolValue = *maxSymbolValuePtr;
unsigned largestCount=0;
ZSTD_memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
while (ip<end) {
assert(*ip <= maxSymbolValue);
count[*ip++]++;
}
while (!count[maxSymbolValue]) maxSymbolValue--;
*maxSymbolValuePtr = maxSymbolValue;
{ U32 s;
for (s=0; s<=maxSymbolValue; s++)
if (count[s] > largestCount) largestCount = count[s];
}
return largestCount;
}
typedef enum { trustInput, checkMaxSymbolValue } HIST_checkInput_e;
/* HIST_count_parallel_wksp() :
* store histogram into 4 intermediate tables, recombined at the end.
* this design makes better use of OoO cpus,
* and is noticeably faster when some values are heavily repeated.
* But it needs some additional workspace for intermediate tables.
* `workSpace` must be a U32 table of size >= HIST_WKSP_SIZE_U32.
* @return : largest histogram frequency,
* or an error code (notably when histogram's alphabet is larger than *maxSymbolValuePtr) */
static size_t HIST_count_parallel_wksp(
unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize,
HIST_checkInput_e check,
U32* const workSpace)
{
const BYTE* ip = (const BYTE*)source;
const BYTE* const iend = ip+sourceSize;
size_t const countSize = (*maxSymbolValuePtr + 1) * sizeof(*count);
unsigned max=0;
U32* const Counting1 = workSpace;
U32* const Counting2 = Counting1 + 256;
U32* const Counting3 = Counting2 + 256;
U32* const Counting4 = Counting3 + 256;
/* safety checks */
assert(*maxSymbolValuePtr <= 255);
if (!sourceSize) {
ZSTD_memset(count, 0, countSize);
*maxSymbolValuePtr = 0;
return 0;
}
ZSTD_memset(workSpace, 0, 4*256*sizeof(unsigned));
/* by stripes of 16 bytes */
{ U32 cached = MEM_read32(ip); ip += 4;
while (ip < iend-15) {
U32 c = cached; cached = MEM_read32(ip); ip += 4;
Counting1[(BYTE) c ]++;
Counting2[(BYTE)(c>>8) ]++;
Counting3[(BYTE)(c>>16)]++;
Counting4[ c>>24 ]++;
c = cached; cached = MEM_read32(ip); ip += 4;
Counting1[(BYTE) c ]++;
Counting2[(BYTE)(c>>8) ]++;
Counting3[(BYTE)(c>>16)]++;
Counting4[ c>>24 ]++;
c = cached; cached = MEM_read32(ip); ip += 4;
Counting1[(BYTE) c ]++;
Counting2[(BYTE)(c>>8) ]++;
Counting3[(BYTE)(c>>16)]++;
Counting4[ c>>24 ]++;
c = cached; cached = MEM_read32(ip); ip += 4;
Counting1[(BYTE) c ]++;
Counting2[(BYTE)(c>>8) ]++;
Counting3[(BYTE)(c>>16)]++;
Counting4[ c>>24 ]++;
}
ip-=4;
}
/* finish last symbols */
while (ip<iend) Counting1[*ip++]++;
{ U32 s;
for (s=0; s<256; s++) {
Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
if (Counting1[s] > max) max = Counting1[s];
} }
{ unsigned maxSymbolValue = 255;
while (!Counting1[maxSymbolValue]) maxSymbolValue--;
if (check && maxSymbolValue > *maxSymbolValuePtr) return ERROR(maxSymbolValue_tooSmall);
*maxSymbolValuePtr = maxSymbolValue;
ZSTD_memmove(count, Counting1, countSize); /* in case count & Counting1 are overlapping */
}
return (size_t)max;
}
/* HIST_countFast_wksp() :
* Same as HIST_countFast(), but using an externally provided scratch buffer.
* `workSpace` is a writable buffer which must be 4-bytes aligned,
* `workSpaceSize` must be >= HIST_WKSP_SIZE
*/
size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize,
void* workSpace, size_t workSpaceSize)
{
if (sourceSize < 1500) /* heuristic threshold */
return HIST_count_simple(count, maxSymbolValuePtr, source, sourceSize);
if ((size_t)workSpace & 3) return ERROR(GENERIC); /* must be aligned on 4-bytes boundaries */
if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, trustInput, (U32*)workSpace);
}
/* HIST_count_wksp() :
* Same as HIST_count(), but using an externally provided scratch buffer.
* `workSpace` size must be table of >= HIST_WKSP_SIZE_U32 unsigned */
size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize,
void* workSpace, size_t workSpaceSize)
{
if ((size_t)workSpace & 3) return ERROR(GENERIC); /* must be aligned on 4-bytes boundaries */
if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
if (*maxSymbolValuePtr < 255)
return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, checkMaxSymbolValue, (U32*)workSpace);
*maxSymbolValuePtr = 255;
return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace, workSpaceSize);
}
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
const void* source, size_t sourceSize)
{
unsigned tmpCounters[HIST_WKSP_SIZE_U32];
return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters, sizeof(tmpCounters));
}
size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
const void* src, size_t srcSize)
{
unsigned tmpCounters[HIST_WKSP_SIZE_U32];
return HIST_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters, sizeof(tmpCounters));
}
#endif
} // namespace duckdb_zstd

1467
external/duckdb/third_party/zstd/compress/huf_compress.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

7147
external/duckdb/third_party/zstd/compress/zstd_compress.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

238
external/duckdb/third_party/zstd/compress/zstd_compress_literals.cpp vendored Normal file
View File

@@ -0,0 +1,238 @@
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/*-*************************************
* Dependencies
***************************************/
#include "zstd/compress/zstd_compress_literals.h"
namespace duckdb_zstd {
/* **************************************************************
* Debug Traces
****************************************************************/
#if DEBUGLEVEL >= 2
static size_t showHexa(const void* src, size_t srcSize)
{
const BYTE* const ip = (const BYTE*)src;
size_t u;
for (u=0; u<srcSize; u++) {
RAWLOG(5, " %02X", ip[u]); (void)ip;
}
RAWLOG(5, " \n");
return srcSize;
}
#endif
/* **************************************************************
* Literals compression - special cases
****************************************************************/
size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
{
BYTE* const ostart = (BYTE*)dst;
U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
DEBUGLOG(5, "ZSTD_noCompressLiterals: srcSize=%zu, dstCapacity=%zu", srcSize, dstCapacity);
RETURN_ERROR_IF(srcSize + flSize > dstCapacity, dstSize_tooSmall, "");
switch(flSize)
{
case 1: /* 2 - 1 - 5 */
ostart[0] = (BYTE)((U32)set_basic + (srcSize<<3));
break;
case 2: /* 2 - 2 - 12 */
MEM_writeLE16(ostart, (U16)((U32)set_basic + (1<<2) + (srcSize<<4)));
break;
case 3: /* 2 - 2 - 20 */
MEM_writeLE32(ostart, (U32)((U32)set_basic + (3<<2) + (srcSize<<4)));
break;
default: /* not necessary : flSize is {1,2,3} */
assert(0);
}
ZSTD_memcpy(ostart + flSize, src, srcSize);
DEBUGLOG(5, "Raw (uncompressed) literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
return srcSize + flSize;
}
static int allBytesIdentical(const void* src, size_t srcSize)
{
assert(srcSize >= 1);
assert(src != NULL);
{ const BYTE b = ((const BYTE*)src)[0];
size_t p;
for (p=1; p<srcSize; p++) {
if (((const BYTE*)src)[p] != b) return 0;
}
return 1;
}
}
size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
{
BYTE* const ostart = (BYTE*)dst;
U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
assert(dstCapacity >= 4); (void)dstCapacity;
assert(allBytesIdentical(src, srcSize));
switch(flSize)
{
case 1: /* 2 - 1 - 5 */
ostart[0] = (BYTE)((U32)set_rle + (srcSize<<3));
break;
case 2: /* 2 - 2 - 12 */
MEM_writeLE16(ostart, (U16)((U32)set_rle + (1<<2) + (srcSize<<4)));
break;
case 3: /* 2 - 2 - 20 */
MEM_writeLE32(ostart, (U32)((U32)set_rle + (3<<2) + (srcSize<<4)));
break;
default: /* not necessary : flSize is {1,2,3} */
assert(0);
}
ostart[flSize] = *(const BYTE*)src;
DEBUGLOG(5, "RLE : Repeated Literal (%02X: %u times) -> %u bytes encoded", ((const BYTE*)src)[0], (U32)srcSize, (U32)flSize + 1);
return flSize+1;
}
/* ZSTD_minLiteralsToCompress() :
* returns minimal amount of literals
* for literal compression to even be attempted.
* Minimum is made tighter as compression strategy increases.
*/
static size_t
ZSTD_minLiteralsToCompress(ZSTD_strategy strategy, HUF_repeat huf_repeat)
{
assert((int)strategy >= 0);
assert((int)strategy <= 9);
/* btultra2 : min 8 bytes;
* then 2x larger for each successive compression strategy
* max threshold 64 bytes */
{ int const shift = MIN(9-(int)strategy, 3);
size_t const mintc = (huf_repeat == HUF_repeat_valid) ? 6 : (size_t)8 << shift;
DEBUGLOG(7, "minLiteralsToCompress = %zu", mintc);
return mintc;
}
}
size_t ZSTD_compressLiterals (
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
void* entropyWorkspace, size_t entropyWorkspaceSize,
const ZSTD_hufCTables_t* prevHuf,
ZSTD_hufCTables_t* nextHuf,
ZSTD_strategy strategy,
int disableLiteralCompression,
int suspectUncompressible,
int bmi2)
{
size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
BYTE* const ostart = (BYTE*)dst;
U32 singleStream = srcSize < 256;
symbolEncodingType_e hType = set_compressed;
size_t cLitSize;
DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i, srcSize=%u, dstCapacity=%zu)",
disableLiteralCompression, (U32)srcSize, dstCapacity);
DEBUGLOG(6, "Completed literals listing (%zu bytes)", showHexa(src, srcSize));
/* Prepare nextEntropy assuming reusing the existing table */
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
if (disableLiteralCompression)
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
/* if too small, don't even attempt compression (speed opt) */
if (srcSize < ZSTD_minLiteralsToCompress(strategy, prevHuf->repeatMode))
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
RETURN_ERROR_IF(dstCapacity < lhSize+1, dstSize_tooSmall, "not enough space for compression");
{ HUF_repeat repeat = prevHuf->repeatMode;
int const flags = 0
| (bmi2 ? HUF_flags_bmi2 : 0)
| (strategy < ZSTD_lazy && srcSize <= 1024 ? HUF_flags_preferRepeat : 0)
| (strategy >= HUF_OPTIMAL_DEPTH_THRESHOLD ? HUF_flags_optimalDepth : 0)
| (suspectUncompressible ? HUF_flags_suspectUncompressible : 0);
typedef size_t (*huf_compress_f)(void*, size_t, const void*, size_t, unsigned, unsigned, void*, size_t, HUF_CElt*, HUF_repeat*, int);
huf_compress_f huf_compress;
if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
huf_compress = singleStream ? HUF_compress1X_repeat : HUF_compress4X_repeat;
cLitSize = huf_compress(ostart+lhSize, dstCapacity-lhSize,
src, srcSize,
HUF_SYMBOLVALUE_MAX, LitHufLog,
entropyWorkspace, entropyWorkspaceSize,
(HUF_CElt*)nextHuf->CTable,
&repeat, flags);
DEBUGLOG(5, "%zu literals compressed into %zu bytes (before header)", srcSize, cLitSize);
if (repeat != HUF_repeat_none) {
/* reused the existing table */
DEBUGLOG(5, "reusing statistics from previous huffman block");
hType = set_repeat;
}
}
{ size_t const minGain = ZSTD_minGain(srcSize, strategy);
if ((cLitSize==0) || (cLitSize >= srcSize - minGain) || ERR_isError(cLitSize)) {
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
} }
if (cLitSize==1) {
/* A return value of 1 signals that the alphabet consists of a single symbol.
* However, in some rare circumstances, it could be the compressed size (a single byte).
* For that outcome to have a chance to happen, it's necessary that `srcSize < 8`.
* (it's also necessary to not generate statistics).
* Therefore, in such a case, actively check that all bytes are identical. */
if ((srcSize >= 8) || allBytesIdentical(src, srcSize)) {
ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
} }
if (hType == set_compressed) {
/* using a newly constructed table */
nextHuf->repeatMode = HUF_repeat_check;
}
/* Build header */
switch(lhSize)
{
case 3: /* 2 - 2 - 10 - 10 */
if (!singleStream) assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
{ U32 const lhc = hType + ((U32)(!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
MEM_writeLE24(ostart, lhc);
break;
}
case 4: /* 2 - 2 - 14 - 14 */
assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
{ U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
MEM_writeLE32(ostart, lhc);
break;
}
case 5: /* 2 - 2 - 18 - 18 */
assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
{ U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
MEM_writeLE32(ostart, lhc);
ostart[4] = (BYTE)(cLitSize >> 10);
break;
}
default: /* not possible : lhSize is {3,4,5} */
assert(0);
}
DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)srcSize, (U32)(lhSize+cLitSize));
return lhSize+cLitSize;
}
} // namespace duckdb_zstd

446
external/duckdb/third_party/zstd/compress/zstd_compress_sequences.cpp vendored Normal file
View File

@@ -0,0 +1,446 @@
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/*-*************************************
* Dependencies
***************************************/
#include "zstd/compress/zstd_compress_sequences.h"
namespace duckdb_zstd {
/**
* -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 kInverseProbabilityLog256[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,
};
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 true if we should use ncount=-1 else we should
* use ncount=1 for low probability symbols instead.
*/
static unsigned ZSTD_useLowProbCount(size_t const nbSeq)
{
/* Heuristic: This should cover most blocks <= 16K and
* start to fade out after 16K to about 32K depending on
* compressibility.
*/
return nbSeq >= 2048;
}
/**
* 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);
FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max, ZSTD_useLowProbCount(nbSeq)), "");
return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
}
/**
* 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;
assert(total > 0);
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] * kInverseProbabilityLog256[norm];
}
return cost >> 8;
}
/**
* 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.
*/
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 += (size_t)count[s] * bitCost;
}
return cost >> kAccuracyLog;
}
/**
* 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.
*/
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) ? (unsigned)norm[s] : 1;
unsigned const norm256 = normAcc << shift;
assert(norm256 > 0);
assert(norm256 < 256);
cost += count[s] * kInverseProbabilityLog256[norm256];
}
return cost >> 8;
}
symbolEncodingType_e
ZSTD_selectEncodingType(
FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
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)
{
ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
if (mostFrequent == nbSeq) {
*repeatMode = FSE_repeat_none;
if (isDefaultAllowed && nbSeq <= 2) {
/* Prefer set_basic over set_rle when there are 2 or fewer symbols,
* since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
* If basic encoding isn't possible, always choose RLE.
*/
DEBUGLOG(5, "Selected set_basic");
return set_basic;
}
DEBUGLOG(5, "Selected set_rle");
return set_rle;
}
if (strategy < ZSTD_lazy) {
if (isDefaultAllowed) {
size_t const staticFse_nbSeq_max = 1000;
size_t const mult = 10 - strategy;
size_t const baseLog = 3;
size_t const dynamicFse_nbSeq_min = (((size_t)1 << defaultNormLog) * mult) >> baseLog; /* 28-36 for offset, 56-72 for lengths */
assert(defaultNormLog >= 5 && defaultNormLog <= 6); /* xx_DEFAULTNORMLOG */
assert(mult <= 9 && mult >= 7);
if ( (*repeatMode == FSE_repeat_valid)
&& (nbSeq < staticFse_nbSeq_max) ) {
DEBUGLOG(5, "Selected set_repeat");
return set_repeat;
}
if ( (nbSeq < dynamicFse_nbSeq_min)
|| (mostFrequent < (nbSeq >> (defaultNormLog-1))) ) {
DEBUGLOG(5, "Selected set_basic");
/* 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
* to confuse these tables with dictionaries. When running more careful
* analysis, we don't need to waste time checking both repeating tables
* and default tables.
*/
*repeatMode = FSE_repeat_none;
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",
(unsigned)basicCost, (unsigned)repeatCost, (unsigned)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");
*repeatMode = FSE_repeat_check;
return set_compressed;
}
typedef struct {
S16 norm[MaxSeq + 1];
U32 wksp[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(MaxSeq, MaxFSELog)];
} ZSTD_BuildCTableWksp;
size_t
ZSTD_buildCTable(void* dst, size_t dstCapacity,
FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
unsigned* count, U32 max,
const BYTE* codeTable, size_t nbSeq,
const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
const FSE_CTable* prevCTable, size_t prevCTableSize,
void* entropyWorkspace, size_t entropyWorkspaceSize)
{
BYTE* op = (BYTE*)dst;
const BYTE* const oend = op + dstCapacity;
DEBUGLOG(6, "ZSTD_buildCTable (dstCapacity=%u)", (unsigned)dstCapacity);
switch (type) {
case set_rle:
FORWARD_IF_ERROR(FSE_buildCTable_rle(nextCTable, (BYTE)max), "");
RETURN_ERROR_IF(dstCapacity==0, dstSize_tooSmall, "not enough space");
*op = codeTable[0];
return 1;
case set_repeat:
ZSTD_memcpy(nextCTable, prevCTable, prevCTableSize);
return 0;
case set_basic:
FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize), ""); /* note : could be pre-calculated */
return 0;
case set_compressed: {
ZSTD_BuildCTableWksp* wksp = (ZSTD_BuildCTableWksp*)entropyWorkspace;
size_t nbSeq_1 = nbSeq;
const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
if (count[codeTable[nbSeq-1]] > 1) {
count[codeTable[nbSeq-1]]--;
nbSeq_1--;
}
assert(nbSeq_1 > 1);
assert(entropyWorkspaceSize >= sizeof(ZSTD_BuildCTableWksp));
(void)entropyWorkspaceSize;
FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "FSE_normalizeCount failed");
assert(oend >= op);
{ size_t const NCountSize = FSE_writeNCount(op, (size_t)(oend - op), wksp->norm, max, tableLog); /* overflow protected */
FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "FSE_buildCTable_wksp failed");
return NCountSize;
}
}
default: assert(0); RETURN_ERROR(GENERIC, "impossible to reach");
}
}
FORCE_INLINE_TEMPLATE size_t
ZSTD_encodeSequences_body(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets)
{
BIT_CStream_t blockStream;
FSE_CState_t stateMatchLength;
FSE_CState_t stateOffsetBits;
FSE_CState_t stateLitLength;
RETURN_ERROR_IF(
ERR_isError(BIT_initCStream(&blockStream, dst, dstCapacity)),
dstSize_tooSmall, "not enough space remaining");
DEBUGLOG(6, "available space for bitstream : %i (dstCapacity=%u)",
(int)(blockStream.endPtr - blockStream.startPtr),
(unsigned)dstCapacity);
/* first symbols */
FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
FSE_initCState2(&stateOffsetBits, CTable_OffsetBits, ofCodeTable[nbSeq-1]);
FSE_initCState2(&stateLitLength, CTable_LitLength, llCodeTable[nbSeq-1]);
BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
if (MEM_32bits()) BIT_flushBits(&blockStream);
BIT_addBits(&blockStream, sequences[nbSeq-1].mlBase, ML_bits[mlCodeTable[nbSeq-1]]);
if (MEM_32bits()) BIT_flushBits(&blockStream);
if (longOffsets) {
U32 const ofBits = ofCodeTable[nbSeq-1];
unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
if (extraBits) {
BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, extraBits);
BIT_flushBits(&blockStream);
}
BIT_addBits(&blockStream, sequences[nbSeq-1].offBase >> extraBits,
ofBits - extraBits);
} else {
BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, ofCodeTable[nbSeq-1]);
}
BIT_flushBits(&blockStream);
{ size_t n;
for (n=nbSeq-2 ; n<nbSeq ; n--) { /* intentional underflow */
BYTE const llCode = llCodeTable[n];
BYTE const ofCode = ofCodeTable[n];
BYTE const mlCode = mlCodeTable[n];
U32 const llBits = LL_bits[llCode];
U32 const ofBits = ofCode;
U32 const mlBits = ML_bits[mlCode];
DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
(unsigned)sequences[n].litLength,
(unsigned)sequences[n].mlBase + MINMATCH,
(unsigned)sequences[n].offBase);
/* 32b*/ /* 64b*/
/* (7)*/ /* (7)*/
FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode); /* 15 */ /* 15 */
FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode); /* 24 */ /* 24 */
if (MEM_32bits()) BIT_flushBits(&blockStream); /* (7)*/
FSE_encodeSymbol(&blockStream, &stateLitLength, llCode); /* 16 */ /* 33 */
if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
BIT_flushBits(&blockStream); /* (7)*/
BIT_addBits(&blockStream, sequences[n].litLength, llBits);
if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
BIT_addBits(&blockStream, sequences[n].mlBase, mlBits);
if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
if (longOffsets) {
unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
if (extraBits) {
BIT_addBits(&blockStream, sequences[n].offBase, extraBits);
BIT_flushBits(&blockStream); /* (7)*/
}
BIT_addBits(&blockStream, sequences[n].offBase >> extraBits,
ofBits - extraBits); /* 31 */
} else {
BIT_addBits(&blockStream, sequences[n].offBase, ofBits); /* 31 */
}
BIT_flushBits(&blockStream); /* (7)*/
DEBUGLOG(7, "remaining space : %i", (int)(blockStream.endPtr - blockStream.ptr));
} }
DEBUGLOG(6, "ZSTD_encodeSequences: flushing ML state with %u bits", stateMatchLength.stateLog);
FSE_flushCState(&blockStream, &stateMatchLength);
DEBUGLOG(6, "ZSTD_encodeSequences: flushing Off state with %u bits", stateOffsetBits.stateLog);
FSE_flushCState(&blockStream, &stateOffsetBits);
DEBUGLOG(6, "ZSTD_encodeSequences: flushing LL state with %u bits", stateLitLength.stateLog);
FSE_flushCState(&blockStream, &stateLitLength);
{ size_t const streamSize = BIT_closeCStream(&blockStream);
RETURN_ERROR_IF(streamSize==0, dstSize_tooSmall, "not enough space");
return streamSize;
}
}
static size_t
ZSTD_encodeSequences_default(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets)
{
return ZSTD_encodeSequences_body(dst, dstCapacity,
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq, longOffsets);
}
#if DYNAMIC_BMI2
static BMI2_TARGET_ATTRIBUTE size_t
ZSTD_encodeSequences_bmi2(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets)
{
return ZSTD_encodeSequences_body(dst, dstCapacity,
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq, longOffsets);
}
#endif
size_t ZSTD_encodeSequences(
void* dst, size_t dstCapacity,
FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2)
{
DEBUGLOG(5, "ZSTD_encodeSequences: dstCapacity = %u", (unsigned)dstCapacity);
#if DYNAMIC_BMI2
if (bmi2) {
return ZSTD_encodeSequences_bmi2(dst, dstCapacity,
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq, longOffsets);
}
#endif
(void)bmi2;
return ZSTD_encodeSequences_default(dst, dstCapacity,
CTable_MatchLength, mlCodeTable,
CTable_OffsetBits, ofCodeTable,
CTable_LitLength, llCodeTable,
sequences, nbSeq, longOffsets);
}
} // namespace duckdb_zstd

692
external/duckdb/third_party/zstd/compress/zstd_compress_superblock.cpp vendored Normal file
View File

@@ -0,0 +1,692 @@
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
/*-*************************************
* Dependencies
***************************************/
#include "zstd/compress/zstd_compress_superblock.h"
#include "zstd/common/zstd_internal.h" /* ZSTD_getSequenceLength */
#include "zstd/compress/hist.h" /* HIST_countFast_wksp */
#include "zstd/compress/zstd_compress_internal.h" /* ZSTD_[huf|fse|entropy]CTablesMetadata_t */
#include "zstd/compress/zstd_compress_sequences.h"
#include "zstd/compress/zstd_compress_literals.h"
namespace duckdb_zstd {
/** ZSTD_compressSubBlock_literal() :
* Compresses literals section for a sub-block.
* When we have to write the Huffman table we will sometimes choose a header
* size larger than necessary. This is because we have to pick the header size
* before we know the table size + compressed size, so we have a bound on the
* table size. If we guessed incorrectly, we fall back to uncompressed literals.
*
* We write the header when writeEntropy=1 and set entropyWritten=1 when we succeeded
* in writing the header, otherwise it is set to 0.
*
* hufMetadata->hType has literals block type info.
* If it is set_basic, all sub-blocks literals section will be Raw_Literals_Block.
* If it is set_rle, all sub-blocks literals section will be RLE_Literals_Block.
* If it is set_compressed, first sub-block's literals section will be Compressed_Literals_Block
* If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
* and the following sub-blocks' literals sections will be Treeless_Literals_Block.
* @return : compressed size of literals section of a sub-block
* Or 0 if unable to compress.
* Or error code */
static size_t
ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
const ZSTD_hufCTablesMetadata_t* hufMetadata,
const BYTE* literals, size_t litSize,
void* dst, size_t dstSize,
const int bmi2, int writeEntropy, int* entropyWritten)
{
size_t const header = writeEntropy ? 200 : 0;
size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstSize;
BYTE* op = ostart + lhSize;
U32 const singleStream = lhSize == 3;
symbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
size_t cLitSize = 0;
DEBUGLOG(5, "ZSTD_compressSubBlock_literal (litSize=%zu, lhSize=%zu, writeEntropy=%d)", litSize, lhSize, writeEntropy);
*entropyWritten = 0;
if (litSize == 0 || hufMetadata->hType == set_basic) {
DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal");
return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
} else if (hufMetadata->hType == set_rle) {
DEBUGLOG(5, "ZSTD_compressSubBlock_literal using rle literal");
return ZSTD_compressRleLiteralsBlock(dst, dstSize, literals, litSize);
}
assert(litSize > 0);
assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);
if (writeEntropy && hufMetadata->hType == set_compressed) {
ZSTD_memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
op += hufMetadata->hufDesSize;
cLitSize += hufMetadata->hufDesSize;
DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
}
{ int const flags = bmi2 ? HUF_flags_bmi2 : 0;
const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, (size_t)(oend-op), literals, litSize, hufTable, flags)
: HUF_compress4X_usingCTable(op, (size_t)(oend-op), literals, litSize, hufTable, flags);
op += cSize;
cLitSize += cSize;
if (cSize == 0 || ERR_isError(cSize)) {
DEBUGLOG(5, "Failed to write entropy tables %s", ZSTD_getErrorName(cSize));
return 0;
}
/* If we expand and we aren't writing a header then emit uncompressed */
if (!writeEntropy && cLitSize >= litSize) {
DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal because uncompressible");
return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
}
/* If we are writing headers then allow expansion that doesn't change our header size. */
if (lhSize < (size_t)(3 + (cLitSize >= 1 KB) + (cLitSize >= 16 KB))) {
assert(cLitSize > litSize);
DEBUGLOG(5, "Literals expanded beyond allowed header size");
return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
}
DEBUGLOG(5, "ZSTD_compressSubBlock_literal (cSize=%zu)", cSize);
}
/* Build header */
switch(lhSize)
{
case 3: /* 2 - 2 - 10 - 10 */
{ U32 const lhc = hType + ((U32)(!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
MEM_writeLE24(ostart, lhc);
break;
}
case 4: /* 2 - 2 - 14 - 14 */
{ U32 const lhc = hType + (2 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<18);
MEM_writeLE32(ostart, lhc);
break;
}
case 5: /* 2 - 2 - 18 - 18 */
{ U32 const lhc = hType + (3 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<22);
MEM_writeLE32(ostart, lhc);
ostart[4] = (BYTE)(cLitSize >> 10);
break;
}
default: /* not possible : lhSize is {3,4,5} */
assert(0);
}
*entropyWritten = 1;
DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)litSize, (U32)(op-ostart));
return (size_t)(op-ostart);
}
static size_t
ZSTD_seqDecompressedSize(seqStore_t const* seqStore,
const seqDef* sequences, size_t nbSeqs,
size_t litSize, int lastSubBlock)
{
size_t matchLengthSum = 0;
size_t litLengthSum = 0;
size_t n;
for (n=0; n<nbSeqs; n++) {
const ZSTD_sequenceLength seqLen = ZSTD_getSequenceLength(seqStore, sequences+n);
litLengthSum += seqLen.litLength;
matchLengthSum += seqLen.matchLength;
}
DEBUGLOG(5, "ZSTD_seqDecompressedSize: %u sequences from %p: %u literals + %u matchlength",
(unsigned)nbSeqs, (const void*)sequences,
(unsigned)litLengthSum, (unsigned)matchLengthSum);
if (!lastSubBlock)
assert(litLengthSum == litSize);
else
assert(litLengthSum <= litSize);
(void)litLengthSum;
return matchLengthSum + litSize;
}
/** ZSTD_compressSubBlock_sequences() :
* Compresses sequences section for a sub-block.
* fseMetadata->llType, fseMetadata->ofType, and fseMetadata->mlType have
* symbol compression modes for the super-block.
* The first successfully compressed block will have these in its header.
* We set entropyWritten=1 when we succeed in compressing the sequences.
* The following sub-blocks will always have repeat mode.
* @return : compressed size of sequences section of a sub-block
* Or 0 if it is unable to compress
* Or error code. */
static size_t
ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
const ZSTD_fseCTablesMetadata_t* fseMetadata,
const seqDef* sequences, size_t nbSeq,
const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
const ZSTD_CCtx_params* cctxParams,
void* dst, size_t dstCapacity,
const int bmi2, int writeEntropy, int* entropyWritten)
{
const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstCapacity;
BYTE* op = ostart;
BYTE* seqHead;
DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (nbSeq=%zu, writeEntropy=%d, longOffsets=%d)", nbSeq, writeEntropy, longOffsets);
*entropyWritten = 0;
/* Sequences Header */
RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
dstSize_tooSmall, "");
if (nbSeq < 128)
*op++ = (BYTE)nbSeq;
else if (nbSeq < LONGNBSEQ)
op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
else
op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
if (nbSeq==0) {
return (size_t)(op - ostart);
}
/* seqHead : flags for FSE encoding type */
seqHead = op++;
DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (seqHeadSize=%u)", (unsigned)(op-ostart));
if (writeEntropy) {
const U32 LLtype = fseMetadata->llType;
const U32 Offtype = fseMetadata->ofType;
const U32 MLtype = fseMetadata->mlType;
DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
*seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
op += fseMetadata->fseTablesSize;
} else {
const U32 repeat = set_repeat;
*seqHead = (BYTE)((repeat<<6) + (repeat<<4) + (repeat<<2));
}
{ size_t const bitstreamSize = ZSTD_encodeSequences(
op, (size_t)(oend - op),
fseTables->matchlengthCTable, mlCode,
fseTables->offcodeCTable, ofCode,
fseTables->litlengthCTable, llCode,
sequences, nbSeq,
longOffsets, bmi2);
FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
op += bitstreamSize;
/* zstd versions <= 1.3.4 mistakenly report corruption when
* FSE_readNCount() receives a buffer < 4 bytes.
* Fixed by https://github.com/facebook/zstd/pull/1146.
* This can happen when the last set_compressed table present is 2
* bytes and the bitstream is only one byte.
* In this exceedingly rare case, we will simply emit an uncompressed
* block, since it isn't worth optimizing.
*/
#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
if (writeEntropy && fseMetadata->lastCountSize && fseMetadata->lastCountSize + bitstreamSize < 4) {
/* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
assert(fseMetadata->lastCountSize + bitstreamSize == 3);
DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
"emitting an uncompressed block.");
return 0;
}
#endif
DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (bitstreamSize=%zu)", bitstreamSize);
}
/* zstd versions <= 1.4.0 mistakenly report error when
* sequences section body size is less than 3 bytes.
* Fixed by https://github.com/facebook/zstd/pull/1664.
* This can happen when the previous sequences section block is compressed
* with rle mode and the current block's sequences section is compressed
* with repeat mode where sequences section body size can be 1 byte.
*/
#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
if (op-seqHead < 4) {
DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.4.0 by emitting "
"an uncompressed block when sequences are < 4 bytes");
return 0;
}
#endif
*entropyWritten = 1;
return (size_t)(op - ostart);
}
/** ZSTD_compressSubBlock() :
* Compresses a single sub-block.
* @return : compressed size of the sub-block
* Or 0 if it failed to compress. */
static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
const seqDef* sequences, size_t nbSeq,
const BYTE* literals, size_t litSize,
const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
const ZSTD_CCtx_params* cctxParams,
void* dst, size_t dstCapacity,
const int bmi2,
int writeLitEntropy, int writeSeqEntropy,
int* litEntropyWritten, int* seqEntropyWritten,
U32 lastBlock)
{
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstCapacity;
BYTE* op = ostart + ZSTD_blockHeaderSize;
DEBUGLOG(5, "ZSTD_compressSubBlock (litSize=%zu, nbSeq=%zu, writeLitEntropy=%d, writeSeqEntropy=%d, lastBlock=%d)",
litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
{ size_t cLitSize = ZSTD_compressSubBlock_literal((const HUF_CElt*)entropy->huf.CTable,
&entropyMetadata->hufMetadata, literals, litSize,
op, (size_t)(oend-op),
bmi2, writeLitEntropy, litEntropyWritten);
FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
if (cLitSize == 0) return 0;
op += cLitSize;
}
{ size_t cSeqSize = ZSTD_compressSubBlock_sequences(&entropy->fse,
&entropyMetadata->fseMetadata,
sequences, nbSeq,
llCode, mlCode, ofCode,
cctxParams,
op, (size_t)(oend-op),
bmi2, writeSeqEntropy, seqEntropyWritten);
FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
if (cSeqSize == 0) return 0;
op += cSeqSize;
}
/* Write block header */
{ size_t cSize = (size_t)(op-ostart) - ZSTD_blockHeaderSize;
U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
MEM_writeLE24(ostart, cBlockHeader24);
}
return (size_t)(op-ostart);
}
static size_t ZSTD_estimateSubBlockSize_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 = 255;
size_t literalSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
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;
return cLitSizeEstimate + literalSectionHeaderSize;
} }
assert(0); /* impossible */
return 0;
}
static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
const BYTE* codeTable, unsigned maxCode,
size_t nbSeq, const FSE_CTable* fseCTable,
const U8* 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);
cSymbolTypeSizeEstimateInBits = max <= defaultMax
? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
: ERROR(GENERIC);
} 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 / 8;
}
static size_t ZSTD_estimateSubBlockSize_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 const sequencesSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
size_t cSeqSizeEstimate = 0;
if (nbSeq == 0) return sequencesSectionHeaderSize;
cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
nbSeq, fseTables->offcodeCTable, NULL,
OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
nbSeq, fseTables->litlengthCTable, LL_bits,
LL_defaultNorm, LL_defaultNormLog, MaxLL,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
nbSeq, fseTables->matchlengthCTable, ML_bits,
ML_defaultNorm, ML_defaultNormLog, MaxML,
workspace, wkspSize);
if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
return cSeqSizeEstimate + sequencesSectionHeaderSize;
}
typedef struct {
size_t estLitSize;
size_t estBlockSize;
} EstimatedBlockSize;
static EstimatedBlockSize ZSTD_estimateSubBlockSize(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)
{
EstimatedBlockSize ebs;
ebs.estLitSize = ZSTD_estimateSubBlockSize_literal(literals, litSize,
&entropy->huf, &entropyMetadata->hufMetadata,
workspace, wkspSize, writeLitEntropy);
ebs.estBlockSize = ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
workspace, wkspSize, writeSeqEntropy);
ebs.estBlockSize += ebs.estLitSize + ZSTD_blockHeaderSize;
return ebs;
}
static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
{
if (fseMetadata->llType == set_compressed || fseMetadata->llType == set_rle)
return 1;
if (fseMetadata->mlType == set_compressed || fseMetadata->mlType == set_rle)
return 1;
if (fseMetadata->ofType == set_compressed || fseMetadata->ofType == set_rle)
return 1;
return 0;
}
static size_t countLiterals(seqStore_t const* seqStore, const seqDef* sp, size_t seqCount)
{
size_t n, total = 0;
assert(sp != NULL);
for (n=0; n<seqCount; n++) {
total += ZSTD_getSequenceLength(seqStore, sp+n).litLength;
}
DEBUGLOG(6, "countLiterals for %zu sequences from %p => %zu bytes", seqCount, (const void*)sp, total);
return total;
}
#define BYTESCALE 256
static size_t sizeBlockSequences(const seqDef* sp, size_t nbSeqs,
size_t targetBudget, size_t avgLitCost, size_t avgSeqCost,
int firstSubBlock)
{
size_t n, budget = 0, inSize=0;
/* entropy headers */
size_t const headerSize = (size_t)firstSubBlock * 120 * BYTESCALE; /* generous estimate */
assert(firstSubBlock==0 || firstSubBlock==1);
budget += headerSize;
/* first sequence => at least one sequence*/
budget += sp[0].litLength * avgLitCost + avgSeqCost;
if (budget > targetBudget) return 1;
inSize = sp[0].litLength + (sp[0].mlBase+MINMATCH);
/* loop over sequences */
for (n=1; n<nbSeqs; n++) {
size_t currentCost = sp[n].litLength * avgLitCost + avgSeqCost;
budget += currentCost;
inSize += sp[n].litLength + (sp[n].mlBase+MINMATCH);
/* stop when sub-block budget is reached */
if ( (budget > targetBudget)
/* though continue to expand until the sub-block is deemed compressible */
&& (budget < inSize * BYTESCALE) )
break;
}
return n;
}
/** ZSTD_compressSubBlock_multi() :
* Breaks super-block into multiple sub-blocks and compresses them.
* Entropy will be written into the first block.
* The following blocks use repeat_mode to compress.
* Sub-blocks are all compressed, except the last one when beneficial.
* @return : compressed size of the super block (which features multiple ZSTD blocks)
* or 0 if it failed to compress. */
static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
const ZSTD_compressedBlockState_t* prevCBlock,
ZSTD_compressedBlockState_t* nextCBlock,
const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
const ZSTD_CCtx_params* cctxParams,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
const int bmi2, U32 lastBlock,
void* workspace, size_t wkspSize)
{
const seqDef* const sstart = seqStorePtr->sequencesStart;
const seqDef* const send = seqStorePtr->sequences;
const seqDef* sp = sstart; /* tracks progresses within seqStorePtr->sequences */
size_t const nbSeqs = (size_t)(send - sstart);
const BYTE* const lstart = seqStorePtr->litStart;
const BYTE* const lend = seqStorePtr->lit;
const BYTE* lp = lstart;
size_t const nbLiterals = (size_t)(lend - lstart);
BYTE const* ip = (BYTE const*)src;
BYTE const* const iend = ip + srcSize;
BYTE* const ostart = (BYTE*)dst;
BYTE* const oend = ostart + dstCapacity;
BYTE* op = ostart;
const BYTE* llCodePtr = seqStorePtr->llCode;
const BYTE* mlCodePtr = seqStorePtr->mlCode;
const BYTE* ofCodePtr = seqStorePtr->ofCode;
size_t const minTarget = ZSTD_TARGETCBLOCKSIZE_MIN; /* enforce minimum size, to reduce undesirable side effects */
size_t const targetCBlockSize = MAX(minTarget, cctxParams->targetCBlockSize);
int writeLitEntropy = (entropyMetadata->hufMetadata.hType == set_compressed);
int writeSeqEntropy = 1;
DEBUGLOG(5, "ZSTD_compressSubBlock_multi (srcSize=%u, litSize=%u, nbSeq=%u)",
(unsigned)srcSize, (unsigned)(lend-lstart), (unsigned)(send-sstart));
/* let's start by a general estimation for the full block */
if (nbSeqs > 0) {
EstimatedBlockSize const ebs =
ZSTD_estimateSubBlockSize(lp, nbLiterals,
ofCodePtr, llCodePtr, mlCodePtr, nbSeqs,
&nextCBlock->entropy, entropyMetadata,
workspace, wkspSize,
writeLitEntropy, writeSeqEntropy);
/* quick estimation */
size_t const avgLitCost = nbLiterals ? (ebs.estLitSize * BYTESCALE) / nbLiterals : BYTESCALE;
size_t const avgSeqCost = ((ebs.estBlockSize - ebs.estLitSize) * BYTESCALE) / nbSeqs;
const size_t nbSubBlocks = MAX((ebs.estBlockSize + (targetCBlockSize/2)) / targetCBlockSize, 1);
size_t n, avgBlockBudget, blockBudgetSupp=0;
avgBlockBudget = (ebs.estBlockSize * BYTESCALE) / nbSubBlocks;
DEBUGLOG(5, "estimated fullblock size=%u bytes ; avgLitCost=%.2f ; avgSeqCost=%.2f ; targetCBlockSize=%u, nbSubBlocks=%u ; avgBlockBudget=%.0f bytes",
(unsigned)ebs.estBlockSize, (double)avgLitCost/BYTESCALE, (double)avgSeqCost/BYTESCALE,
(unsigned)targetCBlockSize, (unsigned)nbSubBlocks, (double)avgBlockBudget/BYTESCALE);
/* simplification: if estimates states that the full superblock doesn't compress, just bail out immediately
* this will result in the production of a single uncompressed block covering @srcSize.*/
if (ebs.estBlockSize > srcSize) return 0;
/* compress and write sub-blocks */
assert(nbSubBlocks>0);
for (n=0; n < nbSubBlocks-1; n++) {
/* determine nb of sequences for current sub-block + nbLiterals from next sequence */
size_t const seqCount = sizeBlockSequences(sp, (size_t)(send-sp),
avgBlockBudget + blockBudgetSupp, avgLitCost, avgSeqCost, n==0);
/* if reached last sequence : break to last sub-block (simplification) */
assert(seqCount <= (size_t)(send-sp));
if (sp + seqCount == send) break;
assert(seqCount > 0);
/* compress sub-block */
{ int litEntropyWritten = 0;
int seqEntropyWritten = 0;
size_t litSize = countLiterals(seqStorePtr, sp, seqCount);
const size_t decompressedSize =
ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, 0);
size_t const cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
sp, seqCount,
lp, litSize,
llCodePtr, mlCodePtr, ofCodePtr,
cctxParams,
op, (size_t)(oend-op),
bmi2, writeLitEntropy, writeSeqEntropy,
&litEntropyWritten, &seqEntropyWritten,
0);
FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
/* check compressibility, update state components */
if (cSize > 0 && cSize < decompressedSize) {
DEBUGLOG(5, "Committed sub-block compressing %u bytes => %u bytes",
(unsigned)decompressedSize, (unsigned)cSize);
assert(ip + decompressedSize <= iend);
ip += decompressedSize;
lp += litSize;
op += cSize;
llCodePtr += seqCount;
mlCodePtr += seqCount;
ofCodePtr += seqCount;
/* Entropy only needs to be written once */
if (litEntropyWritten) {
writeLitEntropy = 0;
}
if (seqEntropyWritten) {
writeSeqEntropy = 0;
}
sp += seqCount;
blockBudgetSupp = 0;
} }
/* otherwise : do not compress yet, coalesce current sub-block with following one */
}
} /* if (nbSeqs > 0) */
/* write last block */
DEBUGLOG(5, "Generate last sub-block: %u sequences remaining", (unsigned)(send - sp));
{ int litEntropyWritten = 0;
int seqEntropyWritten = 0;
size_t litSize = (size_t)(lend - lp);
size_t seqCount = (size_t)(send - sp);
const size_t decompressedSize =
ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, 1);
size_t const cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
sp, seqCount,
lp, litSize,
llCodePtr, mlCodePtr, ofCodePtr,
cctxParams,
op, (size_t)(oend-op),
bmi2, writeLitEntropy, writeSeqEntropy,
&litEntropyWritten, &seqEntropyWritten,
lastBlock);
FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
/* update pointers, the nb of literals borrowed from next sequence must be preserved */
if (cSize > 0 && cSize < decompressedSize) {
DEBUGLOG(5, "Last sub-block compressed %u bytes => %u bytes",
(unsigned)decompressedSize, (unsigned)cSize);
assert(ip + decompressedSize <= iend);
ip += decompressedSize;
lp += litSize;
op += cSize;
llCodePtr += seqCount;
mlCodePtr += seqCount;
ofCodePtr += seqCount;
/* Entropy only needs to be written once */
if (litEntropyWritten) {
writeLitEntropy = 0;
}
if (seqEntropyWritten) {
writeSeqEntropy = 0;
}
sp += seqCount;
}
}
if (writeLitEntropy) {
DEBUGLOG(5, "Literal entropy tables were never written");
ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
}
if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
/* If we haven't written our entropy tables, then we've violated our contract and
* must emit an uncompressed block.
*/
DEBUGLOG(5, "Sequence entropy tables were never written => cancel, emit an uncompressed block");
return 0;
}
if (ip < iend) {
/* some data left : last part of the block sent uncompressed */
size_t const rSize = (size_t)((iend - ip));
size_t const cSize = ZSTD_noCompressBlock(op, (size_t)(oend - op), ip, rSize, lastBlock);
DEBUGLOG(5, "Generate last uncompressed sub-block of %u bytes", (unsigned)(rSize));
FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
assert(cSize != 0);
op += cSize;
/* We have to regenerate the repcodes because we've skipped some sequences */
if (sp < send) {
const seqDef* seq;
repcodes_t rep;
ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
for (seq = sstart; seq < sp; ++seq) {
ZSTD_updateRep(rep.rep, seq->offBase, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
}
ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
}
}
DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed all subBlocks: total compressed size = %u",
(unsigned)(op-ostart));
return (size_t)(op-ostart);
}
size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
unsigned lastBlock)
{
ZSTD_entropyCTablesMetadata_t entropyMetadata;
FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
&zc->blockState.prevCBlock->entropy,
&zc->blockState.nextCBlock->entropy,
&zc->appliedParams,
&entropyMetadata,
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
return ZSTD_compressSubBlock_multi(&zc->seqStore,
zc->blockState.prevCBlock,
zc->blockState.nextCBlock,
&entropyMetadata,
&zc->appliedParams,
dst, dstCapacity,
src, srcSize,
zc->bmi2, lastBlock,
zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
}
} // namespace duckdb_zstd

774
external/duckdb/third_party/zstd/compress/zstd_double_fast.cpp vendored Normal file
View File

@@ -0,0 +1,774 @@
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#include "zstd/compress/zstd_compress_internal.h"
#include "zstd/compress/zstd_double_fast.h"
namespace duckdb_zstd {
#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
void ZSTD_fillDoubleHashTableForCDict(ZSTD_matchState_t* ms,
void const* end, ZSTD_dictTableLoadMethod_e dtlm)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
U32* const hashLarge = ms->hashTable;
U32 const hBitsL = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
U32 const mls = cParams->minMatch;
U32* const hashSmall = ms->chainTable;
U32 const hBitsS = cParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
const BYTE* const base = ms->window.base;
const BYTE* ip = base + ms->nextToUpdate;
const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
const U32 fastHashFillStep = 3;
/* Always insert every fastHashFillStep position into the hash tables.
* Insert the other positions into the large hash table if their entry
* is empty.
*/
for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
U32 const curr = (U32)(ip - base);
U32 i;
for (i = 0; i < fastHashFillStep; ++i) {
size_t const smHashAndTag = ZSTD_hashPtr(ip + i, hBitsS, mls);
size_t const lgHashAndTag = ZSTD_hashPtr(ip + i, hBitsL, 8);
if (i == 0) {
ZSTD_writeTaggedIndex(hashSmall, smHashAndTag, curr + i);
}
if (i == 0 || hashLarge[lgHashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {
ZSTD_writeTaggedIndex(hashLarge, lgHashAndTag, curr + i);
}
/* Only load extra positions for ZSTD_dtlm_full */
if (dtlm == ZSTD_dtlm_fast)
break;
} }
}
static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
void ZSTD_fillDoubleHashTableForCCtx(ZSTD_matchState_t* ms,
void const* end, ZSTD_dictTableLoadMethod_e dtlm)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
U32* const hashLarge = ms->hashTable;
U32 const hBitsL = cParams->hashLog;
U32 const mls = cParams->minMatch;
U32* const hashSmall = ms->chainTable;
U32 const hBitsS = cParams->chainLog;
const BYTE* const base = ms->window.base;
const BYTE* ip = base + ms->nextToUpdate;
const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
const U32 fastHashFillStep = 3;
/* Always insert every fastHashFillStep position into the hash tables.
* Insert the other positions into the large hash table if their entry
* is empty.
*/
for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
U32 const curr = (U32)(ip - base);
U32 i;
for (i = 0; i < fastHashFillStep; ++i) {
size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
if (i == 0)
hashSmall[smHash] = curr + i;
if (i == 0 || hashLarge[lgHash] == 0)
hashLarge[lgHash] = curr + i;
/* Only load extra positions for ZSTD_dtlm_full */
if (dtlm == ZSTD_dtlm_fast)
break;
} }
}
void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
const void* const end,
ZSTD_dictTableLoadMethod_e dtlm,
ZSTD_tableFillPurpose_e tfp)
{
if (tfp == ZSTD_tfp_forCDict) {
ZSTD_fillDoubleHashTableForCDict(ms, end, dtlm);
} else {
ZSTD_fillDoubleHashTableForCCtx(ms, end, dtlm);
}
}
FORCE_INLINE_TEMPLATE
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_compressBlock_doubleFast_noDict_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize, U32 const mls /* template */)
{
ZSTD_compressionParameters const* cParams = &ms->cParams;
U32* const hashLong = ms->hashTable;
const U32 hBitsL = cParams->hashLog;
U32* const hashSmall = ms->chainTable;
const U32 hBitsS = cParams->chainLog;
const BYTE* const base = ms->window.base;
const BYTE* const istart = (const BYTE*)src;
const BYTE* anchor = istart;
const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
/* presumes that, if there is a dictionary, it must be using Attach mode */
const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
const BYTE* const prefixLowest = base + prefixLowestIndex;
const BYTE* const iend = istart + srcSize;
const BYTE* const ilimit = iend - HASH_READ_SIZE;
U32 offset_1=rep[0], offset_2=rep[1];
U32 offsetSaved1 = 0, offsetSaved2 = 0;
size_t mLength;
U32 offset;
U32 curr;
/* how many positions to search before increasing step size */
const size_t kStepIncr = 1 << kSearchStrength;
/* the position at which to increment the step size if no match is found */
const BYTE* nextStep;
size_t step; /* the current step size */
size_t hl0; /* the long hash at ip */
size_t hl1; /* the long hash at ip1 */
U32 idxl0; /* the long match index for ip */
U32 idxl1; /* the long match index for ip1 */
const BYTE* matchl0; /* the long match for ip */
const BYTE* matchs0; /* the short match for ip */
const BYTE* matchl1; /* the long match for ip1 */
const BYTE* ip = istart; /* the current position */
const BYTE* ip1; /* the next position */
DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_noDict_generic");
/* init */
ip += ((ip - prefixLowest) == 0);
{
U32 const current = (U32)(ip - base);
U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
U32 const maxRep = current - windowLow;
if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
}
/* Outer Loop: one iteration per match found and stored */
while (1) {
step = 1;
nextStep = ip + kStepIncr;
ip1 = ip + step;
if (ip1 > ilimit) {
goto _cleanup;
}
hl0 = ZSTD_hashPtr(ip, hBitsL, 8);
idxl0 = hashLong[hl0];
matchl0 = base + idxl0;
/* Inner Loop: one iteration per search / position */
do {
const size_t hs0 = ZSTD_hashPtr(ip, hBitsS, mls);
const U32 idxs0 = hashSmall[hs0];
curr = (U32)(ip-base);
matchs0 = base + idxs0;
hashLong[hl0] = hashSmall[hs0] = curr; /* update hash tables */
/* check noDict repcode */
if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
ip++;
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
goto _match_stored;
}
hl1 = ZSTD_hashPtr(ip1, hBitsL, 8);
if (idxl0 > prefixLowestIndex) {
/* check prefix long match */
if (MEM_read64(matchl0) == MEM_read64(ip)) {
mLength = ZSTD_count(ip+8, matchl0+8, iend) + 8;
offset = (U32)(ip-matchl0);
while (((ip>anchor) & (matchl0>prefixLowest)) && (ip[-1] == matchl0[-1])) { ip--; matchl0--; mLength++; } /* catch up */
goto _match_found;
}
}
idxl1 = hashLong[hl1];
matchl1 = base + idxl1;
if (idxs0 > prefixLowestIndex) {
/* check prefix short match */
if (MEM_read32(matchs0) == MEM_read32(ip)) {
goto _search_next_long;
}
}
if (ip1 >= nextStep) {
PREFETCH_L1(ip1 + 64);
PREFETCH_L1(ip1 + 128);
step++;
nextStep += kStepIncr;
}
ip = ip1;
ip1 += step;
hl0 = hl1;
idxl0 = idxl1;
matchl0 = matchl1;
#if defined(__aarch64__)
PREFETCH_L1(ip+256);
#endif
} while (ip1 <= ilimit);
_cleanup:
/* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
* rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
/* save reps for next block */
rep[0] = offset_1 ? offset_1 : offsetSaved1;
rep[1] = offset_2 ? offset_2 : offsetSaved2;
/* Return the last literals size */
return (size_t)(iend - anchor);
_search_next_long:
/* check prefix long +1 match */
if (idxl1 > prefixLowestIndex) {
if (MEM_read64(matchl1) == MEM_read64(ip1)) {
ip = ip1;
mLength = ZSTD_count(ip+8, matchl1+8, iend) + 8;
offset = (U32)(ip-matchl1);
while (((ip>anchor) & (matchl1>prefixLowest)) && (ip[-1] == matchl1[-1])) { ip--; matchl1--; mLength++; } /* catch up */
goto _match_found;
}
}
/* if no long +1 match, explore the short match we found */
mLength = ZSTD_count(ip+4, matchs0+4, iend) + 4;
offset = (U32)(ip - matchs0);
while (((ip>anchor) & (matchs0>prefixLowest)) && (ip[-1] == matchs0[-1])) { ip--; matchs0--; mLength++; } /* catch up */
/* fall-through */
_match_found: /* requires ip, offset, mLength */
offset_2 = offset_1;
offset_1 = offset;
if (step < 4) {
/* It is unsafe to write this value back to the hashtable when ip1 is
* greater than or equal to the new ip we will have after we're done
* processing this match. Rather than perform that test directly
* (ip1 >= ip + mLength), which costs speed in practice, we do a simpler
* more predictable test. The minmatch even if we take a short match is
* 4 bytes, so as long as step, the distance between ip and ip1
* (initially) is less than 4, we know ip1 < new ip. */
hashLong[hl1] = (U32)(ip1 - base);
}
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
_match_stored:
/* match found */
ip += mLength;
anchor = ip;
if (ip <= ilimit) {
/* Complementary insertion */
/* done after iLimit test, as candidates could be > iend-8 */
{ U32 const indexToInsert = curr+2;
hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
}
/* check immediate repcode */
while ( (ip <= ilimit)
&& ( (offset_2>0)
& (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
/* store sequence */
size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; /* swap offset_2 <=> offset_1 */
hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
ip += rLength;
anchor = ip;
continue; /* faster when present ... (?) */
}
}
}
}
FORCE_INLINE_TEMPLATE
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize,
U32 const mls /* template */)
{
ZSTD_compressionParameters const* cParams = &ms->cParams;
U32* const hashLong = ms->hashTable;
const U32 hBitsL = cParams->hashLog;
U32* const hashSmall = ms->chainTable;
const U32 hBitsS = cParams->chainLog;
const BYTE* const base = ms->window.base;
const BYTE* const istart = (const BYTE*)src;
const BYTE* ip = istart;
const BYTE* anchor = istart;
const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
/* presumes that, if there is a dictionary, it must be using Attach mode */
const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
const BYTE* const prefixLowest = base + prefixLowestIndex;
const BYTE* const iend = istart + srcSize;
const BYTE* const ilimit = iend - HASH_READ_SIZE;
U32 offset_1=rep[0], offset_2=rep[1];
const ZSTD_matchState_t* const dms = ms->dictMatchState;
const ZSTD_compressionParameters* const dictCParams = &dms->cParams;
const U32* const dictHashLong = dms->hashTable;
const U32* const dictHashSmall = dms->chainTable;
const U32 dictStartIndex = dms->window.dictLimit;
const BYTE* const dictBase = dms->window.base;
const BYTE* const dictStart = dictBase + dictStartIndex;
const BYTE* const dictEnd = dms->window.nextSrc;
const U32 dictIndexDelta = prefixLowestIndex - (U32)(dictEnd - dictBase);
const U32 dictHBitsL = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
const U32 dictHBitsS = dictCParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");
/* if a dictionary is attached, it must be within window range */
assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
if (ms->prefetchCDictTables) {
size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
size_t const chainTableBytes = (((size_t)1) << dictCParams->chainLog) * sizeof(U32);
PREFETCH_AREA(dictHashLong, hashTableBytes);
PREFETCH_AREA(dictHashSmall, chainTableBytes);
}
/* init */
ip += (dictAndPrefixLength == 0);
/* dictMatchState repCode checks don't currently handle repCode == 0
* disabling. */
assert(offset_1 <= dictAndPrefixLength);
assert(offset_2 <= dictAndPrefixLength);
/* Main Search Loop */
while (ip < ilimit) { /* < instead of <=, because repcode check at (ip+1) */
size_t mLength;
U32 offset;
size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
size_t const dictHashAndTagL = ZSTD_hashPtr(ip, dictHBitsL, 8);
size_t const dictHashAndTagS = ZSTD_hashPtr(ip, dictHBitsS, mls);
U32 const dictMatchIndexAndTagL = dictHashLong[dictHashAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS];
U32 const dictMatchIndexAndTagS = dictHashSmall[dictHashAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS];
int const dictTagsMatchL = ZSTD_comparePackedTags(dictMatchIndexAndTagL, dictHashAndTagL);
int const dictTagsMatchS = ZSTD_comparePackedTags(dictMatchIndexAndTagS, dictHashAndTagS);
U32 const curr = (U32)(ip-base);
U32 const matchIndexL = hashLong[h2];
U32 matchIndexS = hashSmall[h];
const BYTE* matchLong = base + matchIndexL;
const BYTE* match = base + matchIndexS;
const U32 repIndex = curr + 1 - offset_1;
const BYTE* repMatch = (repIndex < prefixLowestIndex) ?
dictBase + (repIndex - dictIndexDelta) :
base + repIndex;
hashLong[h2] = hashSmall[h] = curr; /* update hash tables */
/* check repcode */
if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
&& (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
ip++;
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
goto _match_stored;
}
if (matchIndexL > prefixLowestIndex) {
/* check prefix long match */
if (MEM_read64(matchLong) == MEM_read64(ip)) {
mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
offset = (U32)(ip-matchLong);
while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
goto _match_found;
}
} else if (dictTagsMatchL) {
/* check dictMatchState long match */
U32 const dictMatchIndexL = dictMatchIndexAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS;
const BYTE* dictMatchL = dictBase + dictMatchIndexL;
assert(dictMatchL < dictEnd);
if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
offset = (U32)(curr - dictMatchIndexL - dictIndexDelta);
while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
goto _match_found;
} }
if (matchIndexS > prefixLowestIndex) {
/* check prefix short match */
if (MEM_read32(match) == MEM_read32(ip)) {
goto _search_next_long;
}
} else if (dictTagsMatchS) {
/* check dictMatchState short match */
U32 const dictMatchIndexS = dictMatchIndexAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS;
match = dictBase + dictMatchIndexS;
matchIndexS = dictMatchIndexS + dictIndexDelta;
if (match > dictStart && MEM_read32(match) == MEM_read32(ip)) {
goto _search_next_long;
} }
ip += ((ip-anchor) >> kSearchStrength) + 1;
#if defined(__aarch64__)
PREFETCH_L1(ip+256);
#endif
continue;
_search_next_long:
{ size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
size_t const dictHashAndTagL3 = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
U32 const matchIndexL3 = hashLong[hl3];
U32 const dictMatchIndexAndTagL3 = dictHashLong[dictHashAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS];
int const dictTagsMatchL3 = ZSTD_comparePackedTags(dictMatchIndexAndTagL3, dictHashAndTagL3);
const BYTE* matchL3 = base + matchIndexL3;
hashLong[hl3] = curr + 1;
/* check prefix long +1 match */
if (matchIndexL3 > prefixLowestIndex) {
if (MEM_read64(matchL3) == MEM_read64(ip+1)) {
mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
ip++;
offset = (U32)(ip-matchL3);
while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
goto _match_found;
}
} else if (dictTagsMatchL3) {
/* check dict long +1 match */
U32 const dictMatchIndexL3 = dictMatchIndexAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS;
const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
assert(dictMatchL3 < dictEnd);
if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
ip++;
offset = (U32)(curr + 1 - dictMatchIndexL3 - dictIndexDelta);
while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
goto _match_found;
} } }
/* if no long +1 match, explore the short match we found */
if (matchIndexS < prefixLowestIndex) {
mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
offset = (U32)(curr - matchIndexS);
while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
} else {
mLength = ZSTD_count(ip+4, match+4, iend) + 4;
offset = (U32)(ip - match);
while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
}
_match_found:
offset_2 = offset_1;
offset_1 = offset;
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
_match_stored:
/* match found */
ip += mLength;
anchor = ip;
if (ip <= ilimit) {
/* Complementary insertion */
/* done after iLimit test, as candidates could be > iend-8 */
{ U32 const indexToInsert = curr+2;
hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
}
/* check immediate repcode */
while (ip <= ilimit) {
U32 const current2 = (U32)(ip-base);
U32 const repIndex2 = current2 - offset_2;
const BYTE* repMatch2 = repIndex2 < prefixLowestIndex ?
dictBase + repIndex2 - dictIndexDelta :
base + repIndex2;
if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
&& (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */
ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
ip += repLength2;
anchor = ip;
continue;
}
break;
}
}
} /* while (ip < ilimit) */
/* save reps for next block */
rep[0] = offset_1;
rep[1] = offset_2;
/* Return the last literals size */
return (size_t)(iend - anchor);
}
#define ZSTD_GEN_DFAST_FN(dictMode, mls) \
static size_t ZSTD_compressBlock_doubleFast_##dictMode##_##mls( \
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], \
void const* src, size_t srcSize) \
{ \
return ZSTD_compressBlock_doubleFast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls); \
}
ZSTD_GEN_DFAST_FN(noDict, 4)
ZSTD_GEN_DFAST_FN(noDict, 5)
ZSTD_GEN_DFAST_FN(noDict, 6)
ZSTD_GEN_DFAST_FN(noDict, 7)
ZSTD_GEN_DFAST_FN(dictMatchState, 4)
ZSTD_GEN_DFAST_FN(dictMatchState, 5)
ZSTD_GEN_DFAST_FN(dictMatchState, 6)
ZSTD_GEN_DFAST_FN(dictMatchState, 7)
size_t ZSTD_compressBlock_doubleFast(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize)
{
const U32 mls = ms->cParams.minMatch;
switch(mls)
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_doubleFast_noDict_4(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_doubleFast_noDict_5(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_doubleFast_noDict_6(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_doubleFast_noDict_7(ms, seqStore, rep, src, srcSize);
}
}
size_t ZSTD_compressBlock_doubleFast_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize)
{
const U32 mls = ms->cParams.minMatch;
switch(mls)
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_doubleFast_dictMatchState_4(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_doubleFast_dictMatchState_5(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_doubleFast_dictMatchState_6(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_doubleFast_dictMatchState_7(ms, seqStore, rep, src, srcSize);
}
}
static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_compressBlock_doubleFast_extDict_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize,
U32 const mls /* template */)
{
ZSTD_compressionParameters const* cParams = &ms->cParams;
U32* const hashLong = ms->hashTable;
U32 const hBitsL = cParams->hashLog;
U32* const hashSmall = ms->chainTable;
U32 const hBitsS = cParams->chainLog;
const BYTE* const istart = (const BYTE*)src;
const BYTE* ip = istart;
const BYTE* anchor = istart;
const BYTE* const iend = istart + srcSize;
const BYTE* const ilimit = iend - 8;
const BYTE* const base = ms->window.base;
const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
const U32 lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
const U32 dictStartIndex = lowLimit;
const U32 dictLimit = ms->window.dictLimit;
const U32 prefixStartIndex = (dictLimit > lowLimit) ? dictLimit : lowLimit;
const BYTE* const prefixStart = base + prefixStartIndex;
const BYTE* const dictBase = ms->window.dictBase;
const BYTE* const dictStart = dictBase + dictStartIndex;
const BYTE* const dictEnd = dictBase + prefixStartIndex;
U32 offset_1=rep[0], offset_2=rep[1];
DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_extDict_generic (srcSize=%zu)", srcSize);
/* if extDict is invalidated due to maxDistance, switch to "regular" variant */
if (prefixStartIndex == dictStartIndex)
return ZSTD_compressBlock_doubleFast(ms, seqStore, rep, src, srcSize);
/* Search Loop */
while (ip < ilimit) { /* < instead of <=, because (ip+1) */
const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
const U32 matchIndex = hashSmall[hSmall];
const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
const BYTE* match = matchBase + matchIndex;
const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
const U32 matchLongIndex = hashLong[hLong];
const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? dictBase : base;
const BYTE* matchLong = matchLongBase + matchLongIndex;
const U32 curr = (U32)(ip-base);
const U32 repIndex = curr + 1 - offset_1; /* offset_1 expected <= curr +1 */
const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
const BYTE* const repMatch = repBase + repIndex;
size_t mLength;
hashSmall[hSmall] = hashLong[hLong] = curr; /* update hash table */
if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
& (offset_1 <= curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
&& (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
ip++;
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
} else {
if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
U32 offset;
mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
offset = curr - matchLongIndex;
while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
offset_2 = offset_1;
offset_1 = offset;
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
} else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
U32 const matchIndex3 = hashLong[h3];
const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
const BYTE* match3 = match3Base + matchIndex3;
U32 offset;
hashLong[h3] = curr + 1;
if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
ip++;
offset = curr+1 - matchIndex3;
while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
} else {
const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
offset = curr - matchIndex;
while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
}
offset_2 = offset_1;
offset_1 = offset;
ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
} else {
ip += ((ip-anchor) >> kSearchStrength) + 1;
continue;
} }
/* move to next sequence start */
ip += mLength;
anchor = ip;
if (ip <= ilimit) {
/* Complementary insertion */
/* done after iLimit test, as candidates could be > iend-8 */
{ U32 const indexToInsert = curr+2;
hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
}
/* check immediate repcode */
while (ip <= ilimit) {
U32 const current2 = (U32)(ip-base);
U32 const repIndex2 = current2 - offset_2;
const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) /* intentional overflow : ensure repIndex2 doesn't overlap dict + prefix */
& (offset_2 <= current2 - dictStartIndex))
&& (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */
ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
ip += repLength2;
anchor = ip;
continue;
}
break;
} } }
/* save reps for next block */
rep[0] = offset_1;
rep[1] = offset_2;
/* Return the last literals size */
return (size_t)(iend - anchor);
}
ZSTD_GEN_DFAST_FN(extDict, 4)
ZSTD_GEN_DFAST_FN(extDict, 5)
ZSTD_GEN_DFAST_FN(extDict, 6)
ZSTD_GEN_DFAST_FN(extDict, 7)
size_t ZSTD_compressBlock_doubleFast_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize)
{
U32 const mls = ms->cParams.minMatch;
switch(mls)
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_doubleFast_extDict_4(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_doubleFast_extDict_5(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_doubleFast_extDict_6(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_doubleFast_extDict_7(ms, seqStore, rep, src, srcSize);
}
}
#endif /* ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR */
} // namespace duckdb_zstd

972
external/duckdb/third_party/zstd/compress/zstd_fast.cpp vendored Normal file
View File

@@ -0,0 +1,972 @@
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#include "zstd/compress/zstd_compress_internal.h" /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
#include "zstd/compress/zstd_fast.h"
namespace duckdb_zstd {
static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
void ZSTD_fillHashTableForCDict(ZSTD_matchState_t* ms,
const void* const end,
ZSTD_dictTableLoadMethod_e dtlm)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
U32* const hashTable = ms->hashTable;
U32 const hBits = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
U32 const mls = cParams->minMatch;
const BYTE* const base = ms->window.base;
const BYTE* ip = base + ms->nextToUpdate;
const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
const U32 fastHashFillStep = 3;
/* Currently, we always use ZSTD_dtlm_full for filling CDict tables.
* Feel free to remove this assert if there's a good reason! */
assert(dtlm == ZSTD_dtlm_full);
/* Always insert every fastHashFillStep position into the hash table.
* Insert the other positions if their hash entry is empty.
*/
for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
U32 const curr = (U32)(ip - base);
{ size_t const hashAndTag = ZSTD_hashPtr(ip, hBits, mls);
ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr); }
if (dtlm == ZSTD_dtlm_fast) continue;
/* Only load extra positions for ZSTD_dtlm_full */
{ U32 p;
for (p = 1; p < fastHashFillStep; ++p) {
size_t const hashAndTag = ZSTD_hashPtr(ip + p, hBits, mls);
if (hashTable[hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) { /* not yet filled */
ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr + p);
} } } }
}
static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
void ZSTD_fillHashTableForCCtx(ZSTD_matchState_t* ms,
const void* const end,
ZSTD_dictTableLoadMethod_e dtlm)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
U32* const hashTable = ms->hashTable;
U32 const hBits = cParams->hashLog;
U32 const mls = cParams->minMatch;
const BYTE* const base = ms->window.base;
const BYTE* ip = base + ms->nextToUpdate;
const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
const U32 fastHashFillStep = 3;
/* Currently, we always use ZSTD_dtlm_fast for filling CCtx tables.
* Feel free to remove this assert if there's a good reason! */
assert(dtlm == ZSTD_dtlm_fast);
/* Always insert every fastHashFillStep position into the hash table.
* Insert the other positions if their hash entry is empty.
*/
for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
U32 const curr = (U32)(ip - base);
size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls);
hashTable[hash0] = curr;
if (dtlm == ZSTD_dtlm_fast) continue;
/* Only load extra positions for ZSTD_dtlm_full */
{ U32 p;
for (p = 1; p < fastHashFillStep; ++p) {
size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls);
if (hashTable[hash] == 0) { /* not yet filled */
hashTable[hash] = curr + p;
} } } }
}
void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
const void* const end,
ZSTD_dictTableLoadMethod_e dtlm,
ZSTD_tableFillPurpose_e tfp)
{
if (tfp == ZSTD_tfp_forCDict) {
ZSTD_fillHashTableForCDict(ms, end, dtlm);
} else {
ZSTD_fillHashTableForCCtx(ms, end, dtlm);
}
}
/**
* If you squint hard enough (and ignore repcodes), the search operation at any
* given position is broken into 4 stages:
*
* 1. Hash (map position to hash value via input read)
* 2. Lookup (map hash val to index via hashtable read)
* 3. Load (map index to value at that position via input read)
* 4. Compare
*
* Each of these steps involves a memory read at an address which is computed
* from the previous step. This means these steps must be sequenced and their
* latencies are cumulative.
*
* Rather than do 1->2->3->4 sequentially for a single position before moving
* onto the next, this implementation interleaves these operations across the
* next few positions:
*
* R = Repcode Read & Compare
* H = Hash
* T = Table Lookup
* M = Match Read & Compare
*
* Pos | Time -->
* ----+-------------------
* N | ... M
* N+1 | ... TM
* N+2 | R H T M
* N+3 | H TM
* N+4 | R H T M
* N+5 | H ...
* N+6 | R ...
*
* This is very much analogous to the pipelining of execution in a CPU. And just
* like a CPU, we have to dump the pipeline when we find a match (i.e., take a
* branch).
*
* When this happens, we throw away our current state, and do the following prep
* to re-enter the loop:
*
* Pos | Time -->
* ----+-------------------
* N | H T
* N+1 | H
*
* This is also the work we do at the beginning to enter the loop initially.
*/
FORCE_INLINE_TEMPLATE
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_compressBlock_fast_noDict_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize,
U32 const mls, U32 const hasStep)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
U32* const hashTable = ms->hashTable;
U32 const hlog = cParams->hashLog;
/* support stepSize of 0 */
size_t const stepSize = hasStep ? (cParams->targetLength + !(cParams->targetLength) + 1) : 2;
const BYTE* const base = ms->window.base;
const BYTE* const istart = (const BYTE*)src;
const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
const U32 prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
const BYTE* const prefixStart = base + prefixStartIndex;
const BYTE* const iend = istart + srcSize;
const BYTE* const ilimit = iend - HASH_READ_SIZE;
const BYTE* anchor = istart;
const BYTE* ip0 = istart;
const BYTE* ip1;
const BYTE* ip2;
const BYTE* ip3;
U32 current0;
U32 rep_offset1 = rep[0];
U32 rep_offset2 = rep[1];
U32 offsetSaved1 = 0, offsetSaved2 = 0;
size_t hash0; /* hash for ip0 */
size_t hash1; /* hash for ip1 */
U32 idx; /* match idx for ip0 */
U32 mval; /* src value at match idx */
U32 offcode;
const BYTE* match0;
size_t mLength;
/* ip0 and ip1 are always adjacent. The targetLength skipping and
* uncompressibility acceleration is applied to every other position,
* matching the behavior of #1562. step therefore represents the gap
* between pairs of positions, from ip0 to ip2 or ip1 to ip3. */
size_t step;
const BYTE* nextStep;
const size_t kStepIncr = (1 << (kSearchStrength - 1));
DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
ip0 += (ip0 == prefixStart);
{ U32 const curr = (U32)(ip0 - base);
U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
U32 const maxRep = curr - windowLow;
if (rep_offset2 > maxRep) offsetSaved2 = rep_offset2, rep_offset2 = 0;
if (rep_offset1 > maxRep) offsetSaved1 = rep_offset1, rep_offset1 = 0;
}
/* start each op */
_start: /* Requires: ip0 */
step = stepSize;
nextStep = ip0 + kStepIncr;
/* calculate positions, ip0 - anchor == 0, so we skip step calc */
ip1 = ip0 + 1;
ip2 = ip0 + step;
ip3 = ip2 + 1;
if (ip3 >= ilimit) {
goto _cleanup;
}
hash0 = ZSTD_hashPtr(ip0, hlog, mls);
hash1 = ZSTD_hashPtr(ip1, hlog, mls);
idx = hashTable[hash0];
do {
/* load repcode match for ip[2]*/
const U32 rval = MEM_read32(ip2 - rep_offset1);
/* write back hash table entry */
current0 = (U32)(ip0 - base);
hashTable[hash0] = current0;
/* check repcode at ip[2] */
if ((MEM_read32(ip2) == rval) & (rep_offset1 > 0)) {
ip0 = ip2;
match0 = ip0 - rep_offset1;
mLength = ip0[-1] == match0[-1];
ip0 -= mLength;
match0 -= mLength;
offcode = REPCODE1_TO_OFFBASE;
mLength += 4;
/* First write next hash table entry; we've already calculated it.
* This write is known to be safe because the ip1 is before the
* repcode (ip2). */
hashTable[hash1] = (U32)(ip1 - base);
goto _match;
}
/* load match for ip[0] */
if (idx >= prefixStartIndex) {
mval = MEM_read32(base + idx);
} else {
mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
}
/* check match at ip[0] */
if (MEM_read32(ip0) == mval) {
/* found a match! */
/* First write next hash table entry; we've already calculated it.
* This write is known to be safe because the ip1 == ip0 + 1, so
* we know we will resume searching after ip1 */
hashTable[hash1] = (U32)(ip1 - base);
goto _offset;
}
/* lookup ip[1] */
idx = hashTable[hash1];
/* hash ip[2] */
hash0 = hash1;
hash1 = ZSTD_hashPtr(ip2, hlog, mls);
/* advance to next positions */
ip0 = ip1;
ip1 = ip2;
ip2 = ip3;
/* write back hash table entry */
current0 = (U32)(ip0 - base);
hashTable[hash0] = current0;
/* load match for ip[0] */
if (idx >= prefixStartIndex) {
mval = MEM_read32(base + idx);
} else {
mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
}
/* check match at ip[0] */
if (MEM_read32(ip0) == mval) {
/* found a match! */
/* first write next hash table entry; we've already calculated it */
if (step <= 4) {
/* We need to avoid writing an index into the hash table >= the
* position at which we will pick up our searching after we've
* taken this match.
*
* The minimum possible match has length 4, so the earliest ip0
* can be after we take this match will be the current ip0 + 4.
* ip1 is ip0 + step - 1. If ip1 is >= ip0 + 4, we can't safely
* write this position.
*/
hashTable[hash1] = (U32)(ip1 - base);
}
goto _offset;
}
/* lookup ip[1] */
idx = hashTable[hash1];
/* hash ip[2] */
hash0 = hash1;
hash1 = ZSTD_hashPtr(ip2, hlog, mls);
/* advance to next positions */
ip0 = ip1;
ip1 = ip2;
ip2 = ip0 + step;
ip3 = ip1 + step;
/* calculate step */
if (ip2 >= nextStep) {
step++;
PREFETCH_L1(ip1 + 64);
PREFETCH_L1(ip1 + 128);
nextStep += kStepIncr;
}
} while (ip3 < ilimit);
_cleanup:
/* Note that there are probably still a couple positions we could search.
* However, it seems to be a meaningful performance hit to try to search
* them. So let's not. */
/* When the repcodes are outside of the prefix, we set them to zero before the loop.
* When the offsets are still zero, we need to restore them after the block to have a correct
* repcode history. If only one offset was invalid, it is easy. The tricky case is when both
* offsets were invalid. We need to figure out which offset to refill with.
* - If both offsets are zero they are in the same order.
* - If both offsets are non-zero, we won't restore the offsets from `offsetSaved[12]`.
* - If only one is zero, we need to decide which offset to restore.
* - If rep_offset1 is non-zero, then rep_offset2 must be offsetSaved1.
* - It is impossible for rep_offset2 to be non-zero.
*
* So if rep_offset1 started invalid (offsetSaved1 != 0) and became valid (rep_offset1 != 0), then
* set rep[0] = rep_offset1 and rep[1] = offsetSaved1.
*/
offsetSaved2 = ((offsetSaved1 != 0) && (rep_offset1 != 0)) ? offsetSaved1 : offsetSaved2;
/* save reps for next block */
rep[0] = rep_offset1 ? rep_offset1 : offsetSaved1;
rep[1] = rep_offset2 ? rep_offset2 : offsetSaved2;
/* Return the last literals size */
return (size_t)(iend - anchor);
_offset: /* Requires: ip0, idx */
/* Compute the offset code. */
match0 = base + idx;
rep_offset2 = rep_offset1;
rep_offset1 = (U32)(ip0-match0);
offcode = OFFSET_TO_OFFBASE(rep_offset1);
mLength = 4;
/* Count the backwards match length. */
while (((ip0>anchor) & (match0>prefixStart)) && (ip0[-1] == match0[-1])) {
ip0--;
match0--;
mLength++;
}
_match: /* Requires: ip0, match0, offcode */
/* Count the forward length. */
mLength += ZSTD_count(ip0 + mLength, match0 + mLength, iend);
ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);
ip0 += mLength;
anchor = ip0;
/* Fill table and check for immediate repcode. */
if (ip0 <= ilimit) {
/* Fill Table */
assert(base+current0+2 > istart); /* check base overflow */
hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2; /* here because current+2 could be > iend-8 */
hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
if (rep_offset2 > 0) { /* rep_offset2==0 means rep_offset2 is invalidated */
while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - rep_offset2)) ) {
/* store sequence */
size_t const rLength = ZSTD_count(ip0+4, ip0+4-rep_offset2, iend) + 4;
{ U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */
hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
ip0 += rLength;
ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
anchor = ip0;
continue; /* faster when present (confirmed on gcc-8) ... (?) */
} } }
goto _start;
}
#define ZSTD_GEN_FAST_FN(dictMode, mls, step) \
static size_t ZSTD_compressBlock_fast_##dictMode##_##mls##_##step( \
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], \
void const* src, size_t srcSize) \
{ \
return ZSTD_compressBlock_fast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls, step); \
}
ZSTD_GEN_FAST_FN(noDict, 4, 1)
ZSTD_GEN_FAST_FN(noDict, 5, 1)
ZSTD_GEN_FAST_FN(noDict, 6, 1)
ZSTD_GEN_FAST_FN(noDict, 7, 1)
ZSTD_GEN_FAST_FN(noDict, 4, 0)
ZSTD_GEN_FAST_FN(noDict, 5, 0)
ZSTD_GEN_FAST_FN(noDict, 6, 0)
ZSTD_GEN_FAST_FN(noDict, 7, 0)
size_t ZSTD_compressBlock_fast(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize)
{
U32 const mls = ms->cParams.minMatch;
assert(ms->dictMatchState == NULL);
if (ms->cParams.targetLength > 1) {
switch(mls)
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_fast_noDict_4_1(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_fast_noDict_5_1(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_fast_noDict_6_1(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_fast_noDict_7_1(ms, seqStore, rep, src, srcSize);
}
} else {
switch(mls)
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_fast_noDict_4_0(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_fast_noDict_5_0(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_fast_noDict_6_0(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_fast_noDict_7_0(ms, seqStore, rep, src, srcSize);
}
}
}
FORCE_INLINE_TEMPLATE
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_compressBlock_fast_dictMatchState_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
U32* const hashTable = ms->hashTable;
U32 const hlog = cParams->hashLog;
/* support stepSize of 0 */
U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
const BYTE* const base = ms->window.base;
const BYTE* const istart = (const BYTE*)src;
const BYTE* ip0 = istart;
const BYTE* ip1 = ip0 + stepSize; /* we assert below that stepSize >= 1 */
const BYTE* anchor = istart;
const U32 prefixStartIndex = ms->window.dictLimit;
const BYTE* const prefixStart = base + prefixStartIndex;
const BYTE* const iend = istart + srcSize;
const BYTE* const ilimit = iend - HASH_READ_SIZE;
U32 offset_1=rep[0], offset_2=rep[1];
const ZSTD_matchState_t* const dms = ms->dictMatchState;
const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;
const U32* const dictHashTable = dms->hashTable;
const U32 dictStartIndex = dms->window.dictLimit;
const BYTE* const dictBase = dms->window.base;
const BYTE* const dictStart = dictBase + dictStartIndex;
const BYTE* const dictEnd = dms->window.nextSrc;
const U32 dictIndexDelta = prefixStartIndex - (U32)(dictEnd - dictBase);
const U32 dictAndPrefixLength = (U32)(istart - prefixStart + dictEnd - dictStart);
const U32 dictHBits = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
/* if a dictionary is still attached, it necessarily means that
* it is within window size. So we just check it. */
const U32 maxDistance = 1U << cParams->windowLog;
const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
assert(endIndex - prefixStartIndex <= maxDistance);
(void)maxDistance; (void)endIndex; /* these variables are not used when assert() is disabled */
(void)hasStep; /* not currently specialized on whether it's accelerated */
/* ensure there will be no underflow
* when translating a dict index into a local index */
assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
if (ms->prefetchCDictTables) {
size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
PREFETCH_AREA(dictHashTable, hashTableBytes);
}
/* init */
DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
ip0 += (dictAndPrefixLength == 0);
/* dictMatchState repCode checks don't currently handle repCode == 0
* disabling. */
assert(offset_1 <= dictAndPrefixLength);
assert(offset_2 <= dictAndPrefixLength);
/* Outer search loop */
assert(stepSize >= 1);
while (ip1 <= ilimit) { /* repcode check at (ip0 + 1) is safe because ip0 < ip1 */
size_t mLength;
size_t hash0 = ZSTD_hashPtr(ip0, hlog, mls);
size_t const dictHashAndTag0 = ZSTD_hashPtr(ip0, dictHBits, mls);
U32 dictMatchIndexAndTag = dictHashTable[dictHashAndTag0 >> ZSTD_SHORT_CACHE_TAG_BITS];
int dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag0);
U32 matchIndex = hashTable[hash0];
U32 curr = (U32)(ip0 - base);
size_t step = stepSize;
const size_t kStepIncr = 1 << kSearchStrength;
const BYTE* nextStep = ip0 + kStepIncr;
/* Inner search loop */
while (1) {
const BYTE* match = base + matchIndex;
const U32 repIndex = curr + 1 - offset_1;
const BYTE* repMatch = (repIndex < prefixStartIndex) ?
dictBase + (repIndex - dictIndexDelta) :
base + repIndex;
const size_t hash1 = ZSTD_hashPtr(ip1, hlog, mls);
size_t const dictHashAndTag1 = ZSTD_hashPtr(ip1, dictHBits, mls);
hashTable[hash0] = curr; /* update hash table */
if (((U32) ((prefixStartIndex - 1) - repIndex) >=
3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
&& (MEM_read32(repMatch) == MEM_read32(ip0 + 1))) {
const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
mLength = ZSTD_count_2segments(ip0 + 1 + 4, repMatch + 4, iend, repMatchEnd, prefixStart) + 4;
ip0++;
ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
break;
}
if (dictTagsMatch) {
/* Found a possible dict match */
const U32 dictMatchIndex = dictMatchIndexAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
const BYTE* dictMatch = dictBase + dictMatchIndex;
if (dictMatchIndex > dictStartIndex &&
MEM_read32(dictMatch) == MEM_read32(ip0)) {
/* To replicate extDict parse behavior, we only use dict matches when the normal matchIndex is invalid */
if (matchIndex <= prefixStartIndex) {
U32 const offset = (U32) (curr - dictMatchIndex - dictIndexDelta);
mLength = ZSTD_count_2segments(ip0 + 4, dictMatch + 4, iend, dictEnd, prefixStart) + 4;
while (((ip0 > anchor) & (dictMatch > dictStart))
&& (ip0[-1] == dictMatch[-1])) {
ip0--;
dictMatch--;
mLength++;
} /* catch up */
offset_2 = offset_1;
offset_1 = offset;
ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
break;
}
}
}
if (matchIndex > prefixStartIndex && MEM_read32(match) == MEM_read32(ip0)) {
/* found a regular match */
U32 const offset = (U32) (ip0 - match);
mLength = ZSTD_count(ip0 + 4, match + 4, iend) + 4;
while (((ip0 > anchor) & (match > prefixStart))
&& (ip0[-1] == match[-1])) {
ip0--;
match--;
mLength++;
} /* catch up */
offset_2 = offset_1;
offset_1 = offset;
ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
break;
}
/* Prepare for next iteration */
dictMatchIndexAndTag = dictHashTable[dictHashAndTag1 >> ZSTD_SHORT_CACHE_TAG_BITS];
dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag1);
matchIndex = hashTable[hash1];
if (ip1 >= nextStep) {
step++;
nextStep += kStepIncr;
}
ip0 = ip1;
ip1 = ip1 + step;
if (ip1 > ilimit) goto _cleanup;
curr = (U32)(ip0 - base);
hash0 = hash1;
} /* end inner search loop */
/* match found */
assert(mLength);
ip0 += mLength;
anchor = ip0;
if (ip0 <= ilimit) {
/* Fill Table */
assert(base+curr+2 > istart); /* check base overflow */
hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2; /* here because curr+2 could be > iend-8 */
hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
/* check immediate repcode */
while (ip0 <= ilimit) {
U32 const current2 = (U32)(ip0-base);
U32 const repIndex2 = current2 - offset_2;
const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
dictBase - dictIndexDelta + repIndex2 :
base + repIndex2;
if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
&& (MEM_read32(repMatch2) == MEM_read32(ip0))) {
const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */
ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = current2;
ip0 += repLength2;
anchor = ip0;
continue;
}
break;
}
}
/* Prepare for next iteration */
assert(ip0 == anchor);
ip1 = ip0 + stepSize;
}
_cleanup:
/* save reps for next block */
rep[0] = offset_1;
rep[1] = offset_2;
/* Return the last literals size */
return (size_t)(iend - anchor);
}
ZSTD_GEN_FAST_FN(dictMatchState, 4, 0)
ZSTD_GEN_FAST_FN(dictMatchState, 5, 0)
ZSTD_GEN_FAST_FN(dictMatchState, 6, 0)
ZSTD_GEN_FAST_FN(dictMatchState, 7, 0)
size_t ZSTD_compressBlock_fast_dictMatchState(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize)
{
U32 const mls = ms->cParams.minMatch;
assert(ms->dictMatchState != NULL);
switch(mls)
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_fast_dictMatchState_4_0(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_fast_dictMatchState_5_0(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_fast_dictMatchState_6_0(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_fast_dictMatchState_7_0(ms, seqStore, rep, src, srcSize);
}
}
static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_compressBlock_fast_extDict_generic(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
U32* const hashTable = ms->hashTable;
U32 const hlog = cParams->hashLog;
/* support stepSize of 0 */
size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
const BYTE* const base = ms->window.base;
const BYTE* const dictBase = ms->window.dictBase;
const BYTE* const istart = (const BYTE*)src;
const BYTE* anchor = istart;
const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
const U32 lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
const U32 dictStartIndex = lowLimit;
const BYTE* const dictStart = dictBase + dictStartIndex;
const U32 dictLimit = ms->window.dictLimit;
const U32 prefixStartIndex = dictLimit < lowLimit ? lowLimit : dictLimit;
const BYTE* const prefixStart = base + prefixStartIndex;
const BYTE* const dictEnd = dictBase + prefixStartIndex;
const BYTE* const iend = istart + srcSize;
const BYTE* const ilimit = iend - 8;
U32 offset_1=rep[0], offset_2=rep[1];
U32 offsetSaved1 = 0, offsetSaved2 = 0;
const BYTE* ip0 = istart;
const BYTE* ip1;
const BYTE* ip2;
const BYTE* ip3;
U32 current0;
size_t hash0; /* hash for ip0 */
size_t hash1; /* hash for ip1 */
U32 idx; /* match idx for ip0 */
const BYTE* idxBase; /* base pointer for idx */
U32 offcode;
const BYTE* match0;
size_t mLength;
const BYTE* matchEnd = 0; /* initialize to avoid warning, assert != 0 later */
size_t step;
const BYTE* nextStep;
const size_t kStepIncr = (1 << (kSearchStrength - 1));
(void)hasStep; /* not currently specialized on whether it's accelerated */
DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1);
/* switch to "regular" variant if extDict is invalidated due to maxDistance */
if (prefixStartIndex == dictStartIndex)
return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);
{ U32 const curr = (U32)(ip0 - base);
U32 const maxRep = curr - dictStartIndex;
if (offset_2 >= maxRep) offsetSaved2 = offset_2, offset_2 = 0;
if (offset_1 >= maxRep) offsetSaved1 = offset_1, offset_1 = 0;
}
/* start each op */
_start: /* Requires: ip0 */
step = stepSize;
nextStep = ip0 + kStepIncr;
/* calculate positions, ip0 - anchor == 0, so we skip step calc */
ip1 = ip0 + 1;
ip2 = ip0 + step;
ip3 = ip2 + 1;
if (ip3 >= ilimit) {
goto _cleanup;
}
hash0 = ZSTD_hashPtr(ip0, hlog, mls);
hash1 = ZSTD_hashPtr(ip1, hlog, mls);
idx = hashTable[hash0];
idxBase = idx < prefixStartIndex ? dictBase : base;
do {
{ /* load repcode match for ip[2] */
U32 const current2 = (U32)(ip2 - base);
U32 const repIndex = current2 - offset_1;
const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
U32 rval;
if ( ((U32)(prefixStartIndex - repIndex) >= 4) /* intentional underflow */
& (offset_1 > 0) ) {
rval = MEM_read32(repBase + repIndex);
} else {
rval = MEM_read32(ip2) ^ 1; /* guaranteed to not match. */
}
/* write back hash table entry */
current0 = (U32)(ip0 - base);
hashTable[hash0] = current0;
/* check repcode at ip[2] */
if (MEM_read32(ip2) == rval) {
ip0 = ip2;
match0 = repBase + repIndex;
matchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
assert((match0 != prefixStart) & (match0 != dictStart));
mLength = ip0[-1] == match0[-1];
ip0 -= mLength;
match0 -= mLength;
offcode = REPCODE1_TO_OFFBASE;
mLength += 4;
goto _match;
} }
{ /* load match for ip[0] */
U32 const mval = idx >= dictStartIndex ?
MEM_read32(idxBase + idx) :
MEM_read32(ip0) ^ 1; /* guaranteed not to match */
/* check match at ip[0] */
if (MEM_read32(ip0) == mval) {
/* found a match! */
goto _offset;
} }
/* lookup ip[1] */
idx = hashTable[hash1];
idxBase = idx < prefixStartIndex ? dictBase : base;
/* hash ip[2] */
hash0 = hash1;
hash1 = ZSTD_hashPtr(ip2, hlog, mls);
/* advance to next positions */
ip0 = ip1;
ip1 = ip2;
ip2 = ip3;
/* write back hash table entry */
current0 = (U32)(ip0 - base);
hashTable[hash0] = current0;
{ /* load match for ip[0] */
U32 const mval = idx >= dictStartIndex ?
MEM_read32(idxBase + idx) :
MEM_read32(ip0) ^ 1; /* guaranteed not to match */
/* check match at ip[0] */
if (MEM_read32(ip0) == mval) {
/* found a match! */
goto _offset;
} }
/* lookup ip[1] */
idx = hashTable[hash1];
idxBase = idx < prefixStartIndex ? dictBase : base;
/* hash ip[2] */
hash0 = hash1;
hash1 = ZSTD_hashPtr(ip2, hlog, mls);
/* advance to next positions */
ip0 = ip1;
ip1 = ip2;
ip2 = ip0 + step;
ip3 = ip1 + step;
/* calculate step */
if (ip2 >= nextStep) {
step++;
PREFETCH_L1(ip1 + 64);
PREFETCH_L1(ip1 + 128);
nextStep += kStepIncr;
}
} while (ip3 < ilimit);
_cleanup:
/* Note that there are probably still a couple positions we could search.
* However, it seems to be a meaningful performance hit to try to search
* them. So let's not. */
/* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
* rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
/* save reps for next block */
rep[0] = offset_1 ? offset_1 : offsetSaved1;
rep[1] = offset_2 ? offset_2 : offsetSaved2;
/* Return the last literals size */
return (size_t)(iend - anchor);
_offset: /* Requires: ip0, idx, idxBase */
/* Compute the offset code. */
{ U32 const offset = current0 - idx;
const BYTE* const lowMatchPtr = idx < prefixStartIndex ? dictStart : prefixStart;
matchEnd = idx < prefixStartIndex ? dictEnd : iend;
match0 = idxBase + idx;
offset_2 = offset_1;
offset_1 = offset;
offcode = OFFSET_TO_OFFBASE(offset);
mLength = 4;
/* Count the backwards match length. */
while (((ip0>anchor) & (match0>lowMatchPtr)) && (ip0[-1] == match0[-1])) {
ip0--;
match0--;
mLength++;
} }
_match: /* Requires: ip0, match0, offcode, matchEnd */
/* Count the forward length. */
assert(matchEnd != 0);
mLength += ZSTD_count_2segments(ip0 + mLength, match0 + mLength, iend, matchEnd, prefixStart);
ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);
ip0 += mLength;
anchor = ip0;
/* write next hash table entry */
if (ip1 < ip0) {
hashTable[hash1] = (U32)(ip1 - base);
}
/* Fill table and check for immediate repcode. */
if (ip0 <= ilimit) {
/* Fill Table */
assert(base+current0+2 > istart); /* check base overflow */
hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2; /* here because current+2 could be > iend-8 */
hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
while (ip0 <= ilimit) {
U32 const repIndex2 = (U32)(ip0-base) - offset_2;
const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 > 0)) /* intentional underflow */
&& (MEM_read32(repMatch2) == MEM_read32(ip0)) ) {
const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
{ U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; } /* swap offset_2 <=> offset_1 */
ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
ip0 += repLength2;
anchor = ip0;
continue;
}
break;
} }
goto _start;
}
ZSTD_GEN_FAST_FN(extDict, 4, 0)
ZSTD_GEN_FAST_FN(extDict, 5, 0)
ZSTD_GEN_FAST_FN(extDict, 6, 0)
ZSTD_GEN_FAST_FN(extDict, 7, 0)
size_t ZSTD_compressBlock_fast_extDict(
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
void const* src, size_t srcSize)
{
U32 const mls = ms->cParams.minMatch;
assert(ms->dictMatchState == NULL);
switch(mls)
{
default: /* includes case 3 */
case 4 :
return ZSTD_compressBlock_fast_extDict_4_0(ms, seqStore, rep, src, srcSize);
case 5 :
return ZSTD_compressBlock_fast_extDict_5_0(ms, seqStore, rep, src, srcSize);
case 6 :
return ZSTD_compressBlock_fast_extDict_6_0(ms, seqStore, rep, src, srcSize);
case 7 :
return ZSTD_compressBlock_fast_extDict_7_0(ms, seqStore, rep, src, srcSize);
}
}
} // namespace duckdb_zstd

2203
external/duckdb/third_party/zstd/compress/zstd_lazy.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

735
external/duckdb/third_party/zstd/compress/zstd_ldm.cpp vendored Normal file
View File

@@ -0,0 +1,735 @@
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#include "zstd/compress/zstd_ldm.h"
#include "zstd/common/debug.h"
#include "zstd/common/xxhash.hpp"
#include "zstd/common/xxhash_static.hpp"
#include "zstd/compress/zstd_fast.h" /* ZSTD_fillHashTable() */
#include "zstd/compress/zstd_double_fast.h" /* ZSTD_fillDoubleHashTable() */
#include "zstd/compress/zstd_ldm_geartab.h"
#define LDM_BUCKET_SIZE_LOG 3
#define LDM_MIN_MATCH_LENGTH 64
#define LDM_HASH_RLOG 7
namespace duckdb_zstd {
typedef struct {
U64 rolling;
U64 stopMask;
} ldmRollingHashState_t;
/** ZSTD_ldm_gear_init():
*
* Initializes the rolling hash state such that it will honor the
* settings in params. */
static void ZSTD_ldm_gear_init(ldmRollingHashState_t* state, ldmParams_t const* params)
{
unsigned maxBitsInMask = MIN(params->minMatchLength, 64);
unsigned hashRateLog = params->hashRateLog;
state->rolling = ~(U32)0;
/* The choice of the splitting criterion is subject to two conditions:
* 1. it has to trigger on average every 2^(hashRateLog) bytes;
* 2. ideally, it has to depend on a window of minMatchLength bytes.
*
* In the gear hash algorithm, bit n depends on the last n bytes;
* so in order to obtain a good quality splitting criterion it is
* preferable to use bits with high weight.
*
* To match condition 1 we use a mask with hashRateLog bits set
* and, because of the previous remark, we make sure these bits
* have the highest possible weight while still respecting
* condition 2.
*/
if (hashRateLog > 0 && hashRateLog <= maxBitsInMask) {
state->stopMask = (((U64)1 << hashRateLog) - 1) << (maxBitsInMask - hashRateLog);
} else {
/* In this degenerate case we simply honor the hash rate. */
state->stopMask = ((U64)1 << hashRateLog) - 1;
}
}
/** ZSTD_ldm_gear_reset()
* Feeds [data, data + minMatchLength) into the hash without registering any
* splits. This effectively resets the hash state. This is used when skipping
* over data, either at the beginning of a block, or skipping sections.
*/
static void ZSTD_ldm_gear_reset(ldmRollingHashState_t* state,
BYTE const* data, size_t minMatchLength)
{
U64 hash = state->rolling;
size_t n = 0;
#define GEAR_ITER_ONCE() do { \
hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
n += 1; \
} while (0)
while (n + 3 < minMatchLength) {
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
}
while (n < minMatchLength) {
GEAR_ITER_ONCE();
}
#undef GEAR_ITER_ONCE
}
/** ZSTD_ldm_gear_feed():
*
* Registers in the splits array all the split points found in the first
* size bytes following the data pointer. This function terminates when
* either all the data has been processed or LDM_BATCH_SIZE splits are
* present in the splits array.
*
* Precondition: The splits array must not be full.
* Returns: The number of bytes processed. */
static size_t ZSTD_ldm_gear_feed(ldmRollingHashState_t* state,
BYTE const* data, size_t size,
size_t* splits, unsigned* numSplits)
{
size_t n;
U64 hash, mask;
hash = state->rolling;
mask = state->stopMask;
n = 0;
#define GEAR_ITER_ONCE() do { \
hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
n += 1; \
if (UNLIKELY((hash & mask) == 0)) { \
splits[*numSplits] = n; \
*numSplits += 1; \
if (*numSplits == LDM_BATCH_SIZE) \
goto done; \
} \
} while (0)
while (n + 3 < size) {
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
GEAR_ITER_ONCE();
}
while (n < size) {
GEAR_ITER_ONCE();
}
#undef GEAR_ITER_ONCE
done:
state->rolling = hash;
return n;
}
void ZSTD_ldm_adjustParameters(ldmParams_t* params,
ZSTD_compressionParameters const* cParams)
{
params->windowLog = cParams->windowLog;
ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
if (params->hashLog == 0) {
params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
assert(params->hashLog <= ZSTD_HASHLOG_MAX);
}
if (params->hashRateLog == 0) {
params->hashRateLog = params->windowLog < params->hashLog
? 0
: params->windowLog - params->hashLog;
}
params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
}
size_t ZSTD_ldm_getTableSize(ldmParams_t params)
{
size_t const ldmHSize = ((size_t)1) << params.hashLog;
size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
+ ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
return params.enableLdm == ZSTD_ps_enable ? totalSize : 0;
}
size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
{
return params.enableLdm == ZSTD_ps_enable ? (maxChunkSize / params.minMatchLength) : 0;
}
/** ZSTD_ldm_getBucket() :
* Returns a pointer to the start of the bucket associated with hash. */
static ldmEntry_t* ZSTD_ldm_getBucket(
ldmState_t* ldmState, size_t hash, ldmParams_t const ldmParams)
{
return ldmState->hashTable + (hash << ldmParams.bucketSizeLog);
}
/** ZSTD_ldm_insertEntry() :
* Insert the entry with corresponding hash into the hash table */
static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
size_t const hash, const ldmEntry_t entry,
ldmParams_t const ldmParams)
{
BYTE* const pOffset = ldmState->bucketOffsets + hash;
unsigned const offset = *pOffset;
*(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + offset) = entry;
*pOffset = (BYTE)((offset + 1) & ((1u << ldmParams.bucketSizeLog) - 1));
}
/** ZSTD_ldm_countBackwardsMatch() :
* Returns the number of bytes that match backwards before pIn and pMatch.
*
* We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
static size_t ZSTD_ldm_countBackwardsMatch(
const BYTE* pIn, const BYTE* pAnchor,
const BYTE* pMatch, const BYTE* pMatchBase)
{
size_t matchLength = 0;
while (pIn > pAnchor && pMatch > pMatchBase && pIn[-1] == pMatch[-1]) {
pIn--;
pMatch--;
matchLength++;
}
return matchLength;
}
/** ZSTD_ldm_countBackwardsMatch_2segments() :
* Returns the number of bytes that match backwards from pMatch,
* even with the backwards match spanning 2 different segments.
*
* On reaching `pMatchBase`, start counting from mEnd */
static size_t ZSTD_ldm_countBackwardsMatch_2segments(
const BYTE* pIn, const BYTE* pAnchor,
const BYTE* pMatch, const BYTE* pMatchBase,
const BYTE* pExtDictStart, const BYTE* pExtDictEnd)
{
size_t matchLength = ZSTD_ldm_countBackwardsMatch(pIn, pAnchor, pMatch, pMatchBase);
if (pMatch - matchLength != pMatchBase || pMatchBase == pExtDictStart) {
/* If backwards match is entirely in the extDict or prefix, immediately return */
return matchLength;
}
DEBUGLOG(7, "ZSTD_ldm_countBackwardsMatch_2segments: found 2-parts backwards match (length in prefix==%zu)", matchLength);
matchLength += ZSTD_ldm_countBackwardsMatch(pIn - matchLength, pAnchor, pExtDictEnd, pExtDictStart);
DEBUGLOG(7, "final backwards match length = %zu", matchLength);
return matchLength;
}
/** ZSTD_ldm_fillFastTables() :
*
* Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
* This is similar to ZSTD_loadDictionaryContent.
*
* The tables for the other strategies are filled within their
* block compressors. */
static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
void const* end)
{
const BYTE* const iend = (const BYTE*)end;
switch(ms->cParams.strategy)
{
case ZSTD_fast:
ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
break;
case ZSTD_dfast:
#ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
#else
assert(0); /* shouldn't be called: cparams should've been adjusted. */
#endif
break;
case ZSTD_greedy:
case ZSTD_lazy:
case ZSTD_lazy2:
case ZSTD_btlazy2:
case ZSTD_btopt:
case ZSTD_btultra:
case ZSTD_btultra2:
break;
default:
assert(0); /* not possible : not a valid strategy id */
}
return 0;
}
void ZSTD_ldm_fillHashTable(
ldmState_t* ldmState, const BYTE* ip,
const BYTE* iend, ldmParams_t const* params)
{
U32 const minMatchLength = params->minMatchLength;
U32 const hBits = params->hashLog - params->bucketSizeLog;
BYTE const* const base = ldmState->window.base;
BYTE const* const istart = ip;
ldmRollingHashState_t hashState;
size_t* const splits = ldmState->splitIndices;
unsigned numSplits;
DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
ZSTD_ldm_gear_init(&hashState, params);
while (ip < iend) {
size_t hashed;
unsigned n;
numSplits = 0;
hashed = ZSTD_ldm_gear_feed(&hashState, ip, iend - ip, splits, &numSplits);
for (n = 0; n < numSplits; n++) {
if (ip + splits[n] >= istart + minMatchLength) {
BYTE const* const split = ip + splits[n] - minMatchLength;
U64 const xxhash = XXH64(split, minMatchLength, 0);
U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
ldmEntry_t entry;
entry.offset = (U32)(split - base);
entry.checksum = (U32)(xxhash >> 32);
ZSTD_ldm_insertEntry(ldmState, hash, entry, *params);
}
}
ip += hashed;
}
}
/** ZSTD_ldm_limitTableUpdate() :
*
* Sets cctx->nextToUpdate to a position corresponding closer to anchor
* if it is far way
* (after a long match, only update tables a limited amount). */
static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
{
U32 const curr = (U32)(anchor - ms->window.base);
if (curr > ms->nextToUpdate + 1024) {
ms->nextToUpdate =
curr - MIN(512, curr - ms->nextToUpdate - 1024);
}
}
static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_ldm_generateSequences_internal(
ldmState_t* ldmState, rawSeqStore_t* rawSeqStore,
ldmParams_t const* params, void const* src, size_t srcSize)
{
/* LDM parameters */
int const extDict = ZSTD_window_hasExtDict(ldmState->window);
U32 const minMatchLength = params->minMatchLength;
U32 const entsPerBucket = 1U << params->bucketSizeLog;
U32 const hBits = params->hashLog - params->bucketSizeLog;
/* Prefix and extDict parameters */
U32 const dictLimit = ldmState->window.dictLimit;
U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
BYTE const* const base = ldmState->window.base;
BYTE const* const dictBase = extDict ? ldmState->window.dictBase : NULL;
BYTE const* const dictStart = extDict ? dictBase + lowestIndex : NULL;
BYTE const* const dictEnd = extDict ? dictBase + dictLimit : NULL;
BYTE const* const lowPrefixPtr = base + dictLimit;
/* Input bounds */
BYTE const* const istart = (BYTE const*)src;
BYTE const* const iend = istart + srcSize;
BYTE const* const ilimit = iend - HASH_READ_SIZE;
/* Input positions */
BYTE const* anchor = istart;
BYTE const* ip = istart;
/* Rolling hash state */
ldmRollingHashState_t hashState;
/* Arrays for staged-processing */
size_t* const splits = ldmState->splitIndices;
ldmMatchCandidate_t* const candidates = ldmState->matchCandidates;
unsigned numSplits;
if (srcSize < minMatchLength)
return iend - anchor;
/* Initialize the rolling hash state with the first minMatchLength bytes */
ZSTD_ldm_gear_init(&hashState, params);
ZSTD_ldm_gear_reset(&hashState, ip, minMatchLength);
ip += minMatchLength;
while (ip < ilimit) {
size_t hashed;
unsigned n;
numSplits = 0;
hashed = ZSTD_ldm_gear_feed(&hashState, ip, ilimit - ip,
splits, &numSplits);
for (n = 0; n < numSplits; n++) {
BYTE const* const split = ip + splits[n] - minMatchLength;
U64 const xxhash = XXH64(split, minMatchLength, 0);
U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
candidates[n].split = split;
candidates[n].hash = hash;
candidates[n].checksum = (U32)(xxhash >> 32);
candidates[n].bucket = ZSTD_ldm_getBucket(ldmState, hash, *params);
PREFETCH_L1(candidates[n].bucket);
}
for (n = 0; n < numSplits; n++) {
size_t forwardMatchLength = 0, backwardMatchLength = 0,
bestMatchLength = 0, mLength;
U32 offset;
BYTE const* const split = candidates[n].split;
U32 const checksum = candidates[n].checksum;
U32 const hash = candidates[n].hash;
ldmEntry_t* const bucket = candidates[n].bucket;
ldmEntry_t const* cur;
ldmEntry_t const* bestEntry = NULL;
ldmEntry_t newEntry;
newEntry.offset = (U32)(split - base);
newEntry.checksum = checksum;
/* If a split point would generate a sequence overlapping with
* the previous one, we merely register it in the hash table and
* move on */
if (split < anchor) {
ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
continue;
}
for (cur = bucket; cur < bucket + entsPerBucket; cur++) {
size_t curForwardMatchLength, curBackwardMatchLength,
curTotalMatchLength;
if (cur->checksum != checksum || cur->offset <= lowestIndex) {
continue;
}
if (extDict) {
BYTE const* const curMatchBase =
cur->offset < dictLimit ? dictBase : base;
BYTE const* const pMatch = curMatchBase + cur->offset;
BYTE const* const matchEnd =
cur->offset < dictLimit ? dictEnd : iend;
BYTE const* const lowMatchPtr =
cur->offset < dictLimit ? dictStart : lowPrefixPtr;
curForwardMatchLength =
ZSTD_count_2segments(split, pMatch, iend, matchEnd, lowPrefixPtr);
if (curForwardMatchLength < minMatchLength) {
continue;
}
curBackwardMatchLength = ZSTD_ldm_countBackwardsMatch_2segments(
split, anchor, pMatch, lowMatchPtr, dictStart, dictEnd);
} else { /* !extDict */
BYTE const* const pMatch = base + cur->offset;
curForwardMatchLength = ZSTD_count(split, pMatch, iend);
if (curForwardMatchLength < minMatchLength) {
continue;
}
curBackwardMatchLength =
ZSTD_ldm_countBackwardsMatch(split, anchor, pMatch, lowPrefixPtr);
}
curTotalMatchLength = curForwardMatchLength + curBackwardMatchLength;
if (curTotalMatchLength > bestMatchLength) {
bestMatchLength = curTotalMatchLength;
forwardMatchLength = curForwardMatchLength;
backwardMatchLength = curBackwardMatchLength;
bestEntry = cur;
}
}
/* No match found -- insert an entry into the hash table
* and process the next candidate match */
if (bestEntry == NULL) {
ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
continue;
}
/* Match found */
offset = (U32)(split - base) - bestEntry->offset;
mLength = forwardMatchLength + backwardMatchLength;
{
rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
/* Out of sequence storage */
if (rawSeqStore->size == rawSeqStore->capacity)
return ERROR(dstSize_tooSmall);
seq->litLength = (U32)(split - backwardMatchLength - anchor);
seq->matchLength = (U32)mLength;
seq->offset = offset;
rawSeqStore->size++;
}
/* Insert the current entry into the hash table --- it must be
* done after the previous block to avoid clobbering bestEntry */
ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
anchor = split + forwardMatchLength;
/* If we find a match that ends after the data that we've hashed
* then we have a repeating, overlapping, pattern. E.g. all zeros.
* If one repetition of the pattern matches our `stopMask` then all
* repetitions will. We don't need to insert them all into out table,
* only the first one. So skip over overlapping matches.
* This is a major speed boost (20x) for compressing a single byte
* repeated, when that byte ends up in the table.
*/
if (anchor > ip + hashed) {
ZSTD_ldm_gear_reset(&hashState, anchor - minMatchLength, minMatchLength);
/* Continue the outer loop at anchor (ip + hashed == anchor). */
ip = anchor - hashed;
break;
}
}
ip += hashed;
}
return iend - anchor;
}
/*! ZSTD_ldm_reduceTable() :
* reduce table indexes by `reducerValue` */
static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
U32 const reducerValue)
{
U32 u;
for (u = 0; u < size; u++) {
if (table[u].offset < reducerValue) table[u].offset = 0;
else table[u].offset -= reducerValue;
}
}
size_t ZSTD_ldm_generateSequences(
ldmState_t* ldmState, rawSeqStore_t* sequences,
ldmParams_t const* params, void const* src, size_t srcSize)
{
U32 const maxDist = 1U << params->windowLog;
BYTE const* const istart = (BYTE const*)src;
BYTE const* const iend = istart + srcSize;
size_t const kMaxChunkSize = 1 << 20;
size_t const nbChunks = (srcSize / kMaxChunkSize) + ((srcSize % kMaxChunkSize) != 0);
size_t chunk;
size_t leftoverSize = 0;
assert(ZSTD_CHUNKSIZE_MAX >= kMaxChunkSize);
/* Check that ZSTD_window_update() has been called for this chunk prior
* to passing it to this function.
*/
assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
/* The input could be very large (in zstdmt), so it must be broken up into
* chunks to enforce the maximum distance and handle overflow correction.
*/
assert(sequences->pos <= sequences->size);
assert(sequences->size <= sequences->capacity);
for (chunk = 0; chunk < nbChunks && sequences->size < sequences->capacity; ++chunk) {
BYTE const* const chunkStart = istart + chunk * kMaxChunkSize;
size_t const remaining = (size_t)(iend - chunkStart);
BYTE const *const chunkEnd =
(remaining < kMaxChunkSize) ? iend : chunkStart + kMaxChunkSize;
size_t const chunkSize = chunkEnd - chunkStart;
size_t newLeftoverSize;
size_t const prevSize = sequences->size;
assert(chunkStart < iend);
/* 1. Perform overflow correction if necessary. */
if (ZSTD_window_needOverflowCorrection(ldmState->window, 0, maxDist, ldmState->loadedDictEnd, chunkStart, chunkEnd)) {
U32 const ldmHSize = 1U << params->hashLog;
U32 const correction = ZSTD_window_correctOverflow(
&ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
/* invalidate dictionaries on overflow correction */
ldmState->loadedDictEnd = 0;
}
/* 2. We enforce the maximum offset allowed.
*
* kMaxChunkSize should be small enough that we don't lose too much of
* the window through early invalidation.
* TODO: * Test the chunk size.
* * Try invalidation after the sequence generation and test the
* offset against maxDist directly.
*
* NOTE: Because of dictionaries + sequence splitting we MUST make sure
* that any offset used is valid at the END of the sequence, since it may
* be split into two sequences. This condition holds when using
* ZSTD_window_enforceMaxDist(), but if we move to checking offsets
* against maxDist directly, we'll have to carefully handle that case.
*/
ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, &ldmState->loadedDictEnd, NULL);
/* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
newLeftoverSize = ZSTD_ldm_generateSequences_internal(
ldmState, sequences, params, chunkStart, chunkSize);
if (ZSTD_isError(newLeftoverSize))
return newLeftoverSize;
/* 4. We add the leftover literals from previous iterations to the first
* newly generated sequence, or add the `newLeftoverSize` if none are
* generated.
*/
/* Prepend the leftover literals from the last call */
if (prevSize < sequences->size) {
sequences->seq[prevSize].litLength += (U32)leftoverSize;
leftoverSize = newLeftoverSize;
} else {
assert(newLeftoverSize == chunkSize);
leftoverSize += chunkSize;
}
}
return 0;
}
void
ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch)
{
while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
if (srcSize <= seq->litLength) {
/* Skip past srcSize literals */
seq->litLength -= (U32)srcSize;
return;
}
srcSize -= seq->litLength;
seq->litLength = 0;
if (srcSize < seq->matchLength) {
/* Skip past the first srcSize of the match */
seq->matchLength -= (U32)srcSize;
if (seq->matchLength < minMatch) {
/* The match is too short, omit it */
if (rawSeqStore->pos + 1 < rawSeqStore->size) {
seq[1].litLength += seq[0].matchLength;
}
rawSeqStore->pos++;
}
return;
}
srcSize -= seq->matchLength;
seq->matchLength = 0;
rawSeqStore->pos++;
}
}
/**
* If the sequence length is longer than remaining then the sequence is split
* between this block and the next.
*
* Returns the current sequence to handle, or if the rest of the block should
* be literals, it returns a sequence with offset == 0.
*/
static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
U32 const remaining, U32 const minMatch)
{
rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
assert(sequence.offset > 0);
/* Likely: No partial sequence */
if (remaining >= sequence.litLength + sequence.matchLength) {
rawSeqStore->pos++;
return sequence;
}
/* Cut the sequence short (offset == 0 ==> rest is literals). */
if (remaining <= sequence.litLength) {
sequence.offset = 0;
} else if (remaining < sequence.litLength + sequence.matchLength) {
sequence.matchLength = remaining - sequence.litLength;
if (sequence.matchLength < minMatch) {
sequence.offset = 0;
}
}
/* Skip past `remaining` bytes for the future sequences. */
ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
return sequence;
}
void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
while (currPos && rawSeqStore->pos < rawSeqStore->size) {
rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
if (currPos >= currSeq.litLength + currSeq.matchLength) {
currPos -= currSeq.litLength + currSeq.matchLength;
rawSeqStore->pos++;
} else {
rawSeqStore->posInSequence = currPos;
break;
}
}
if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
rawSeqStore->posInSequence = 0;
}
}
size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
ZSTD_paramSwitch_e useRowMatchFinder,
void const* src, size_t srcSize)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
unsigned const minMatch = cParams->minMatch;
ZSTD_blockCompressor const blockCompressor =
ZSTD_selectBlockCompressor(cParams->strategy, useRowMatchFinder, ZSTD_matchState_dictMode(ms));
/* Input bounds */
BYTE const* const istart = (BYTE const*)src;
BYTE const* const iend = istart + srcSize;
/* Input positions */
BYTE const* ip = istart;
DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
/* If using opt parser, use LDMs only as candidates rather than always accepting them */
if (cParams->strategy >= ZSTD_btopt) {
size_t lastLLSize;
ms->ldmSeqStore = rawSeqStore;
lastLLSize = blockCompressor(ms, seqStore, rep, src, srcSize);
ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore, srcSize);
return lastLLSize;
}
assert(rawSeqStore->pos <= rawSeqStore->size);
assert(rawSeqStore->size <= rawSeqStore->capacity);
/* Loop through each sequence and apply the block compressor to the literals */
while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
/* maybeSplitSequence updates rawSeqStore->pos */
rawSeq const sequence = maybeSplitSequence(rawSeqStore,
(U32)(iend - ip), minMatch);
/* End signal */
if (sequence.offset == 0)
break;
assert(ip + sequence.litLength + sequence.matchLength <= iend);
/* Fill tables for block compressor */
ZSTD_ldm_limitTableUpdate(ms, ip);
ZSTD_ldm_fillFastTables(ms, ip);
/* Run the block compressor */
DEBUGLOG(5, "pos %u : calling block compressor on segment of size %u", (unsigned)(ip-istart), sequence.litLength);
{
int i;
size_t const newLitLength =
blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
ip += sequence.litLength;
/* Update the repcodes */
for (i = ZSTD_REP_NUM - 1; i > 0; i--)
rep[i] = rep[i-1];
rep[0] = sequence.offset;
/* Store the sequence */
ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
OFFSET_TO_OFFBASE(sequence.offset),
sequence.matchLength);
ip += sequence.matchLength;
}
}
/* Fill the tables for the block compressor */
ZSTD_ldm_limitTableUpdate(ms, ip);
ZSTD_ldm_fillFastTables(ms, ip);
/* Compress the last literals */
return blockCompressor(ms, seqStore, rep, ip, iend - ip);
}
} // namespace duckdb_zstd

1580
external/duckdb/third_party/zstd/compress/zstd_opt.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1885
external/duckdb/third_party/zstd/compress/zstdmt_compress.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff