zstd/lib/compress/fse_compress.c

706 lines
26 KiB
C
Raw Normal View History

2015-01-24 00:58:16 +00:00
/* ******************************************************************
* FSE : Finite State Entropy encoder
* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
*
* 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.
2015-01-24 00:58:16 +00:00
****************************************************************** */
2016-01-06 00:58:37 +00:00
/* **************************************************************
2015-01-24 00:58:16 +00:00
* Includes
****************************************************************/
#include "../common/compiler.h"
#include "../common/mem.h" /* U32, U16, etc. */
#include "../common/debug.h" /* assert, DEBUGLOG */
#include "hist.h" /* HIST_count_wksp */
#include "../common/bitstream.h"
#define FSE_STATIC_LINKING_ONLY
#include "../common/fse.h"
#include "../common/error_private.h"
2020-08-07 03:18:05 +00:00
#define ZSTD_DEPS_NEED_MALLOC
#define ZSTD_DEPS_NEED_MATH64
#include "../common/zstd_deps.h" /* ZSTD_malloc, ZSTD_free, ZSTD_memcpy, ZSTD_memset */
2015-01-24 00:58:16 +00:00
2016-01-06 00:58:37 +00:00
/* **************************************************************
2015-01-24 00:58:16 +00:00
* Error Management
****************************************************************/
#define FSE_isError ERR_isError
2015-01-24 00:58:16 +00:00
2016-01-06 00:58:37 +00:00
/* **************************************************************
* Templates
2015-01-24 00:58:16 +00:00
****************************************************************/
/*
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
2015-01-24 00:58:16 +00:00
/* 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)
2015-06-18 15:43:16 +00:00
/* 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)
2015-01-24 00:58:16 +00:00
{
2016-03-20 23:39:19 +00:00
U32 const tableSize = 1 << tableLog;
U32 const tableMask = tableSize - 1;
2016-01-06 00:58:37 +00:00
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);
2020-08-27 00:52:54 +00:00
U32* cumul = (U32*)workSpace;
FSE_FUNCTION_TYPE* tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSymbolValue + 2));
U32 highThreshold = tableSize-1;
2015-01-24 00:58:16 +00:00
2020-08-27 00:52:54 +00:00
if ((size_t)workSpace & 3) return ERROR(GENERIC); /* Must be 4 byte aligned */
if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
2016-03-20 23:39:19 +00:00
/* 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 :
* http://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 */
2016-03-20 23:39:19 +00:00
{ U32 u;
cumul[0] = 0;
for (u=1; u <= maxSymbolValue+1; u++) {
2016-03-20 23:39:19 +00:00
if (normalizedCounter[u-1]==-1) { /* Low proba symbol */
cumul[u] = cumul[u-1] + 1;
tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
} else {
cumul[u] = cumul[u-1] + normalizedCounter[u-1];
} }
cumul[maxSymbolValue+1] = tableSize+1;
}
/* Spread symbols */
2016-03-20 23:39:19 +00:00
{ U32 position = 0;
U32 symbol;
for (symbol=0; symbol<=maxSymbolValue; symbol++) {
Spelling (#1582) * spelling: accidentally * spelling: across * spelling: additionally * spelling: addresses * spelling: appropriate * spelling: assumed * spelling: available * spelling: builder * spelling: capacity * spelling: compiler * spelling: compressibility * spelling: compressor * spelling: compression * spelling: contract * spelling: convenience * spelling: decompress * spelling: description * spelling: deflate * spelling: deterministically * spelling: dictionary * spelling: display * spelling: eliminate * spelling: preemptively * spelling: exclude * spelling: failure * spelling: independence * spelling: independent * spelling: intentionally * spelling: matching * spelling: maximum * spelling: meaning * spelling: mishandled * spelling: memory * spelling: occasionally * spelling: occurrence * spelling: official * spelling: offsets * spelling: original * spelling: output * spelling: overflow * spelling: overridden * spelling: parameter * spelling: performance * spelling: probability * spelling: receives * spelling: redundant * spelling: recompression * spelling: resources * spelling: sanity * spelling: segment * spelling: series * spelling: specified * spelling: specify * spelling: subtracted * spelling: successful * spelling: return * spelling: translation * spelling: update * spelling: unrelated * spelling: useless * spelling: variables * spelling: variety * spelling: verbatim * spelling: verification * spelling: visited * spelling: warming * spelling: workers * spelling: with
2019-04-12 18:18:11 +00:00
int nbOccurrences;
int const freq = normalizedCounter[symbol];
Spelling (#1582) * spelling: accidentally * spelling: across * spelling: additionally * spelling: addresses * spelling: appropriate * spelling: assumed * spelling: available * spelling: builder * spelling: capacity * spelling: compiler * spelling: compressibility * spelling: compressor * spelling: compression * spelling: contract * spelling: convenience * spelling: decompress * spelling: description * spelling: deflate * spelling: deterministically * spelling: dictionary * spelling: display * spelling: eliminate * spelling: preemptively * spelling: exclude * spelling: failure * spelling: independence * spelling: independent * spelling: intentionally * spelling: matching * spelling: maximum * spelling: meaning * spelling: mishandled * spelling: memory * spelling: occasionally * spelling: occurrence * spelling: official * spelling: offsets * spelling: original * spelling: output * spelling: overflow * spelling: overridden * spelling: parameter * spelling: performance * spelling: probability * spelling: receives * spelling: redundant * spelling: recompression * spelling: resources * spelling: sanity * spelling: segment * spelling: series * spelling: specified * spelling: specify * spelling: subtracted * spelling: successful * spelling: return * spelling: translation * spelling: update * spelling: unrelated * spelling: useless * spelling: variables * spelling: variety * spelling: verbatim * spelling: verification * spelling: visited * spelling: warming * spelling: workers * spelling: with
2019-04-12 18:18:11 +00:00
for (nbOccurrences=0; nbOccurrences<freq; nbOccurrences++) {
2016-03-20 23:39:19 +00:00
tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
position = (position + step) & tableMask;
while (position > highThreshold)
position = (position + step) & tableMask; /* Low proba area */
2016-03-20 23:39:19 +00:00
} }
assert(position==0); /* Must have initialized all positions */
2016-03-20 23:39:19 +00:00
}
/* Build table */
{ U32 u; for (u=0; u<tableSize; u++) {
2016-03-20 23:39:19 +00:00
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 */
2016-03-20 23:39:19 +00:00
{ unsigned total = 0;
unsigned s;
2016-01-28 02:24:53 +00:00
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:
2015-08-02 00:19:09 +00:00
case 1:
2016-01-30 23:58:06 +00:00
symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
symbolTT[s].deltaFindState = total - 1;
total ++;
break;
default :
2015-08-02 00:19:09 +00:00
{
2016-03-20 23:39:19 +00:00
U32 const maxBitsOut = tableLog - BIT_highbit32 (normalizedCounter[s]-1);
U32 const minStatePlus = normalizedCounter[s] << maxBitsOut;
2015-08-02 00:19:09 +00:00
symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
symbolTT[s].deltaFindState = total - normalizedCounter[s];
total += normalizedCounter[s];
2016-01-30 23:58:06 +00:00
} } } }
#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 ZSTD_NO_UNUSED_FUNCTIONS
size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
{
FSE_FUNCTION_TYPE tableSymbol[FSE_MAX_TABLESIZE]; /* memset() is not necessary, even if static analyzer complain about it */
return FSE_buildCTable_wksp(ct, normalizedCounter, maxSymbolValue, tableLog, tableSymbol, sizeof(tableSymbol));
}
#endif
2015-10-18 21:18:32 +00:00
#ifndef FSE_COMMONDEFS_ONLY
2016-01-28 02:24:53 +00:00
/*-**************************************************************
* FSE NCount encoding
****************************************************************/
size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
{
size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog) >> 3) + 3;
2015-08-19 22:53:56 +00:00
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 = 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 += 3 << bitCount;
bitCount += 2;
}
bitStream += (symbol-start) << bitCount;
bitCount += 2;
2016-01-27 23:18:06 +00:00
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;
2016-01-27 23:18:06 +00:00
} }
{ 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 += count << bitCount;
bitCount += nbBits;
bitCount -= (count<max);
previousIs0 = (count==1);
if (remaining<1) return ERROR(GENERIC);
2018-02-22 23:44:26 +00:00
while (remaining<threshold) { nbBits--; threshold>>=1; }
}
2016-01-27 23:18:06 +00:00
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;
2016-01-30 23:58:06 +00:00
} }
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;
return (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 */
2015-10-18 21:18:32 +00:00
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 */);
}
2016-01-28 02:24:53 +00:00
/*-**************************************************************
2015-01-24 00:58:16 +00:00
* FSE Compression Code
****************************************************************/
2015-07-04 23:56:41 +00:00
FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
2015-01-24 00:58:16 +00:00
{
size_t size;
if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
size = FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
return (FSE_CTable*)ZSTD_malloc(size);
2015-01-24 00:58:16 +00:00
}
void FSE_freeCTable (FSE_CTable* ct) { ZSTD_free(ct); }
2015-01-24 00:58:16 +00:00
2015-08-21 02:59:31 +00:00
/* provides the minimum logSize to safely represent a distribution */
static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
2015-01-24 00:58:16 +00:00
{
2018-08-02 18:12:17 +00:00
U32 minBitsSrc = BIT_highbit32((U32)(srcSize)) + 1;
2017-03-30 01:51:58 +00:00
U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
assert(srcSize > 1); /* Not supported, RLE should be used instead */
2017-03-30 01:51:58 +00:00
return minBits;
2015-08-21 02:59:31 +00:00
}
2016-05-20 12:36:36 +00:00
unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
2015-08-21 02:59:31 +00:00
{
2017-03-30 01:51:58 +00:00
U32 maxBitsSrc = BIT_highbit32((U32)(srcSize - 1)) - minus;
2015-08-21 02:59:31 +00:00
U32 tableLog = maxTableLog;
2017-03-30 01:51:58 +00:00
U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
assert(srcSize > 1); /* Not supported, RLE should be used instead */
2015-01-24 00:58:16 +00:00
if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
2017-03-30 01:51:58 +00:00
if (maxBitsSrc < tableLog) tableLog = maxBitsSrc; /* Accuracy can be reduced */
if (minBits > tableLog) tableLog = minBits; /* Need a minimum to safely represent all symbol values */
2015-01-24 00:58:16 +00:00
if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
return tableLog;
}
2016-05-20 12:36:36 +00:00
unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
{
return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
}
2015-02-24 08:05:58 +00:00
/* Secondary normalization method.
To be used when primary method fails. */
2020-08-18 22:26:54 +00:00
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));
2016-01-27 23:18:06 +00:00
for (s=0; s<=maxSymbolValue; s++) {
if (count[s] == 0) {
norm[s]=0;
continue;
}
2020-08-18 22:26:54 +00:00
if (count[s] <= lowThreshold) {
norm[s] = lowProbCount;
distributed++;
total -= count[s];
continue;
}
2016-01-27 23:18:06 +00:00
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;
2016-01-27 23:18:06 +00:00
if ((total / ToDistribute) > lowOne) {
/* risk of rounding to zero */
lowOne = (U32)((total * 3) / (ToDistribute * 2));
2016-01-27 23:18:06 +00:00
for (s=0; s<=maxSymbolValue; s++) {
if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
norm[s] = 1;
distributed++;
total -= count[s];
continue;
2016-01-27 23:18:06 +00:00
} }
ToDistribute = (1 << tableLog) - distributed;
}
2016-01-27 23:18:06 +00:00
if (distributed == maxSymbolValue+1) {
2015-02-24 08:05:58 +00:00
/* 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++)
2018-02-22 23:44:26 +00:00
if (count[s] > maxC) { maxV=s; maxC=count[s]; }
2015-07-04 23:56:41 +00:00
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))
2018-02-22 23:44:26 +00:00
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;
2016-01-27 23:18:06 +00:00
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)
2015-10-18 21:18:32 +00:00
return ERROR(GENERIC);
2015-06-18 15:43:16 +00:00
norm[s] = (short)weight;
tmpTotal = end;
2016-01-27 23:18:06 +00:00
} } }
return 0;
}
2015-01-24 00:58:16 +00:00
size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
const unsigned* count, size_t total,
2020-08-18 22:26:54 +00:00
unsigned maxSymbolValue, unsigned useLowProbCount)
2015-01-24 00:58:16 +00:00
{
/* Sanity checks */
if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
2015-10-18 21:18:32 +00:00
if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC); /* Unsupported size */
2016-01-27 23:18:06 +00:00
if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge); /* Unsupported size */
2015-10-18 21:18:32 +00:00
if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC); /* Too small tableLog, compression potentially impossible */
2015-01-24 00:58:16 +00:00
{ static U32 const rtbTable[] = { 0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
2020-08-18 22:26:54 +00:00
short const lowProbCount = useLowProbCount ? -1 : 1;
2015-01-24 00:58:16 +00:00
U64 const scale = 62 - tableLog;
U64 const step = ZSTD_div64((U64)1<<62, (U32)total); /* <== here, one division ! */
2015-01-24 00:58:16 +00:00
U64 const vStep = 1ULL<<(scale-20);
int stillToDistribute = 1<<tableLog;
unsigned s;
unsigned largest=0;
short largestP=0;
U32 lowThreshold = (U32)(total >> tableLog);
2016-01-27 23:18:06 +00:00
for (s=0; s<=maxSymbolValue; s++) {
if (count[s] == total) return 0; /* rle special case */
if (count[s] == 0) { normalizedCounter[s]=0; continue; }
2020-08-18 22:26:54 +00:00
if (count[s] <= lowThreshold) {
normalizedCounter[s] = lowProbCount;
2015-01-24 00:58:16 +00:00
stillToDistribute--;
2016-01-27 23:18:06 +00:00
} else {
2015-01-24 00:58:16 +00:00
short proba = (short)((count[s]*step) >> scale);
2016-01-27 23:18:06 +00:00
if (proba<8) {
U64 restToBeat = vStep * rtbTable[proba];
2015-01-24 00:58:16 +00:00
proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
}
2018-02-22 23:44:26 +00:00
if (proba > largestP) { largestP=proba; largest=s; }
2015-01-24 00:58:16 +00:00
normalizedCounter[s] = proba;
stillToDistribute -= proba;
2016-01-27 23:18:06 +00:00
} }
if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
2015-02-24 08:05:58 +00:00
/* corner case, need another normalization method */
2020-08-18 22:26:54 +00:00
size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue, lowProbCount);
2015-01-29 05:51:30 +00:00
if (FSE_isError(errorCode)) return errorCode;
2015-01-24 00:58:16 +00:00
}
else normalizedCounter[largest] += (short)stillToDistribute;
}
2020-08-24 19:24:45 +00:00
#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
2015-01-24 00:58:16 +00:00
return tableLog;
}
2015-07-04 23:56:41 +00:00
/* fake FSE_CTable, for raw (uncompressed) input */
size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits)
2015-01-24 00:58:16 +00:00
{
const unsigned tableSize = 1 << nbBits;
const unsigned tableMask = tableSize - 1;
const unsigned maxSymbolValue = tableMask;
2016-01-06 00:58:37 +00:00
void* const ptr = ct;
U16* const tableU16 = ( (U16*) ptr) + 2;
void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableSize>>1); /* assumption : tableLog >= 1 */
FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
2015-01-24 00:58:16 +00:00
unsigned s;
/* Sanity checks */
2015-10-18 21:18:32 +00:00
if (nbBits < 1) return ERROR(GENERIC); /* min size */
2015-01-24 00:58:16 +00:00
/* header */
tableU16[-2] = (U16) nbBits;
tableU16[-1] = (U16) maxSymbolValue;
/* Build table */
for (s=0; s<tableSize; s++)
tableU16[s] = (U16)(tableSize + s);
/* Build Symbol Transformation Table */
{ const U32 deltaNbBits = (nbBits << 16) - (1 << nbBits);
2016-01-31 01:04:15 +00:00
for (s=0; s<=maxSymbolValue; s++) {
symbolTT[s].deltaNbBits = deltaNbBits;
symbolTT[s].deltaFindState = s-1;
} }
2015-01-24 00:58:16 +00:00
return 0;
}
/* fake FSE_CTable, for rle input (always same symbol) */
2015-07-04 23:56:41 +00:00
size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
2015-01-24 00:58:16 +00:00
{
2016-01-06 00:58:37 +00:00
void* ptr = ct;
U16* tableU16 = ( (U16*) ptr) + 2;
void* FSCTptr = (U32*)ptr + 2;
FSE_symbolCompressionTransform* symbolTT = (FSE_symbolCompressionTransform*) FSCTptr;
2015-01-24 00:58:16 +00:00
/* header */
tableU16[-2] = (U16) 0;
tableU16[-1] = (U16) symbolValue;
/* Build table */
tableU16[0] = 0;
tableU16[1] = 0; /* just in case */
/* Build Symbol Transformation Table */
2016-01-28 02:24:53 +00:00
symbolTT[symbolValue].deltaNbBits = 0;
symbolTT[symbolValue].deltaFindState = 0;
2015-01-24 00:58:16 +00:00
return 0;
}
2015-08-07 14:21:00 +00:00
static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
2015-01-24 00:58:16 +00:00
const void* src, size_t srcSize,
2015-08-07 14:21:00 +00:00
const FSE_CTable* ct, const unsigned fast)
2015-01-24 00:58:16 +00:00
{
const BYTE* const istart = (const BYTE*) src;
const BYTE* const iend = istart + srcSize;
2016-03-04 18:26:59 +00:00
const BYTE* ip=iend;
2015-10-18 21:18:32 +00:00
BIT_CStream_t bitC;
2015-01-24 00:58:16 +00:00
FSE_CState_t CState1, CState2;
/* init */
2016-03-04 18:26:59 +00:00
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 */ }
2015-01-24 00:58:16 +00:00
2015-10-18 21:18:32 +00:00
#define FSE_FLUSHBITS(s) (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
2015-08-07 14:21:00 +00:00
2016-01-28 02:24:53 +00:00
if (srcSize & 1) {
2016-03-04 18:26:59 +00:00
FSE_initCState2(&CState1, ct, *--ip);
FSE_initCState2(&CState2, ct, *--ip);
2015-07-04 23:56:41 +00:00
FSE_encodeSymbol(&bitC, &CState1, *--ip);
2015-08-07 14:21:00 +00:00
FSE_FLUSHBITS(&bitC);
2016-03-04 18:26:59 +00:00
} else {
FSE_initCState2(&CState2, ct, *--ip);
FSE_initCState2(&CState1, ct, *--ip);
2015-01-24 00:58:16 +00:00
}
/* join to mod 4 */
2016-03-04 18:26:59 +00:00
srcSize -= 2;
2016-01-28 02:24:53 +00:00
if ((sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) && (srcSize & 2)) { /* test bit 2 */
2015-07-04 23:56:41 +00:00
FSE_encodeSymbol(&bitC, &CState2, *--ip);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
2015-08-07 14:21:00 +00:00
FSE_FLUSHBITS(&bitC);
2015-01-24 00:58:16 +00:00
}
/* 2 or 4 encoding per loop */
while ( ip>istart ) {
2015-07-04 23:56:41 +00:00
FSE_encodeSymbol(&bitC, &CState2, *--ip);
2015-01-24 00:58:16 +00:00
2015-08-02 00:19:09 +00:00
if (sizeof(bitC.bitContainer)*8 < FSE_MAX_TABLELOG*2+7 ) /* this test must be static */
2015-08-07 14:21:00 +00:00
FSE_FLUSHBITS(&bitC);
2015-01-24 00:58:16 +00:00
2015-07-04 23:56:41 +00:00
FSE_encodeSymbol(&bitC, &CState1, *--ip);
2015-01-24 00:58:16 +00:00
2016-01-28 02:24:53 +00:00
if (sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) { /* this test must be static */
2015-07-04 23:56:41 +00:00
FSE_encodeSymbol(&bitC, &CState2, *--ip);
FSE_encodeSymbol(&bitC, &CState1, *--ip);
2015-01-24 00:58:16 +00:00
}
2015-08-07 14:21:00 +00:00
FSE_FLUSHBITS(&bitC);
2015-01-24 00:58:16 +00:00
}
FSE_flushCState(&bitC, &CState2);
FSE_flushCState(&bitC, &CState1);
2015-10-18 21:18:32 +00:00
return BIT_closeCStream(&bitC);
2015-01-24 00:58:16 +00:00
}
2015-08-07 14:21:00 +00:00
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));
2015-01-24 00:58:16 +00:00
2015-08-07 14:21:00 +00:00
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);
}
2015-01-24 00:58:16 +00:00
2015-08-07 14:21:00 +00:00
size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
#ifndef ZSTD_NO_UNUSED_FUNCTIONS
/* FSE_compress_wksp() :
* Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
* `wkspSize` size must be `(1<<tableLog)`.
*/
size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
2015-01-24 00:58:16 +00:00
{
BYTE* const ostart = (BYTE*) dst;
BYTE* op = ostart;
BYTE* const oend = ostart + dstSize;
fix confusion between unsigned <-> U32 as suggested in #1441. generally U32 and unsigned are the same thing, except when they are not ... case : 32-bit compilation for MIPS (uint32_t == unsigned long) A vast majority of transformation consists in transforming U32 into unsigned. In rare cases, it's the other way around (typically for internal code, such as seeds). Among a few issues this patches solves : - some parameters were declared with type `unsigned` in *.h, but with type `U32` in their implementation *.c . - some parameters have type unsigned*, but the caller user a pointer to U32 instead. These fixes are useful. However, the bulk of changes is about %u formating, which requires unsigned type, but generally receives U32 values instead, often just for brevity (U32 is shorter than unsigned). These changes are generally minor, or even annoying. As a consequence, the amount of code changed is larger than I would expect for such a patch. Testing is also a pain : it requires manually modifying `mem.h`, in order to lie about `U32` and force it to be an `unsigned long` typically. On a 64-bit system, this will break the equivalence unsigned == U32. Unfortunately, it will also break a few static_assert(), controlling structure sizes. So it also requires modifying `debug.h` to make `static_assert()` a noop. And then reverting these changes. So it's inconvenient, and as a consequence, this property is currently not checked during CI tests. Therefore, these problems can emerge again in the future. I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests. It's another restriction for coding, adding more frustration during merge tests, since most platforms don't need this distinction (hence contributor will not see it), and while this can matter in theory, the number of platforms impacted seems minimal. Thoughts ?
2018-12-22 00:19:44 +00:00
unsigned count[FSE_MAX_SYMBOL_VALUE+1];
2015-01-24 00:58:16 +00:00
S16 norm[FSE_MAX_SYMBOL_VALUE+1];
FSE_CTable* CTable = (FSE_CTable*)workSpace;
size_t const CTableSize = FSE_CTABLE_SIZE_U32(tableLog, maxSymbolValue);
void* scratchBuffer = (void*)(CTable + CTableSize);
size_t const scratchBufferSize = wkspSize - (CTableSize * sizeof(FSE_CTable));
2015-01-24 00:58:16 +00:00
2015-08-07 14:21:00 +00:00
/* init conditions */
2020-08-17 05:22:33 +00:00
if (wkspSize < FSE_COMPRESS_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
if (srcSize <= 1) return 0; /* Not compressible */
2015-01-24 00:58:16 +00:00
if (!maxSymbolValue) maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
/* Scan input and build symbol stats */
{ CHECK_V_F(maxCount, HIST_count_wksp(count, &maxSymbolValue, src, srcSize, scratchBuffer, scratchBufferSize) );
if (maxCount == srcSize) return 1; /* only a single symbol in src : rle */
if (maxCount == 1) return 0; /* each symbol present maximum once => not compressible */
if (maxCount < (srcSize >> 7)) return 0; /* Heuristic : not compressible enough */
}
2015-01-24 00:58:16 +00:00
tableLog = FSE_optimalTableLog(tableLog, srcSize, maxSymbolValue);
2020-08-18 22:26:54 +00:00
CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue, /* useLowProbCount */ srcSize >= 2048) );
2015-01-24 00:58:16 +00:00
/* Write table description header */
{ CHECK_V_F(nc_err, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
op += nc_err;
}
2015-01-24 00:58:16 +00:00
/* Compress */
CHECK_F( FSE_buildCTable_wksp(CTable, norm, maxSymbolValue, tableLog, scratchBuffer, scratchBufferSize) );
{ CHECK_V_F(cSize, FSE_compress_usingCTable(op, oend - op, src, srcSize, CTable) );
if (cSize == 0) return 0; /* not enough space for compressed data */
op += cSize;
}
2015-01-24 00:58:16 +00:00
/* check compressibility */
if ( (size_t)(op-ostart) >= srcSize-1 ) return 0;
2015-01-24 00:58:16 +00:00
return op-ostart;
}
typedef struct {
FSE_CTable CTable_max[FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
union {
U32 hist_wksp[HIST_WKSP_SIZE_U32];
BYTE scratchBuffer[1 << FSE_MAX_TABLELOG];
} workspace;
} fseWkspMax_t;
size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
{
fseWkspMax_t scratchBuffer;
2020-08-17 05:22:33 +00:00
DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_COMPRESS_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)); /* compilation failures here means scratchBuffer is not large enough */
if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
}
size_t FSE_compress (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
2015-01-24 00:58:16 +00:00
{
return FSE_compress2(dst, dstCapacity, src, srcSize, FSE_MAX_SYMBOL_VALUE, FSE_DEFAULT_TABLELOG);
2015-01-24 00:58:16 +00:00
}
#endif
2015-01-24 00:58:16 +00:00
#endif /* FSE_COMMONDEFS_ONLY */