Coverage Report

Created: 2024-05-21 06:09

/src/c-blosc/internal-complibs/zstd-1.5.6/compress/fse_compress.c
Line
Count
Source (jump to first uncovered line)
1
/* ******************************************************************
2
 * FSE : Finite State Entropy encoder
3
 * Copyright (c) Meta Platforms, Inc. and affiliates.
4
 *
5
 *  You can contact the author at :
6
 *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
7
 *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
8
 *
9
 * This source code is licensed under both the BSD-style license (found in the
10
 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
11
 * in the COPYING file in the root directory of this source tree).
12
 * You may select, at your option, one of the above-listed licenses.
13
****************************************************************** */
14
15
/* **************************************************************
16
*  Includes
17
****************************************************************/
18
#include "../common/compiler.h"
19
#include "../common/mem.h"        /* U32, U16, etc. */
20
#include "../common/debug.h"      /* assert, DEBUGLOG */
21
#include "hist.h"       /* HIST_count_wksp */
22
#include "../common/bitstream.h"
23
#define FSE_STATIC_LINKING_ONLY
24
#include "../common/fse.h"
25
#include "../common/error_private.h"
26
#define ZSTD_DEPS_NEED_MALLOC
27
#define ZSTD_DEPS_NEED_MATH64
28
#include "../common/zstd_deps.h"  /* ZSTD_memset */
29
#include "../common/bits.h" /* ZSTD_highbit32 */
30
31
32
/* **************************************************************
33
*  Error Management
34
****************************************************************/
35
173k
#define FSE_isError ERR_isError
36
37
38
/* **************************************************************
39
*  Templates
40
****************************************************************/
41
/*
42
  designed to be included
43
  for type-specific functions (template emulation in C)
44
  Objective is to write these functions only once, for improved maintenance
45
*/
46
47
/* safety checks */
48
#ifndef FSE_FUNCTION_EXTENSION
49
#  error "FSE_FUNCTION_EXTENSION must be defined"
50
#endif
51
#ifndef FSE_FUNCTION_TYPE
52
#  error "FSE_FUNCTION_TYPE must be defined"
53
#endif
54
55
/* Function names */
56
#define FSE_CAT(X,Y) X##Y
57
#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
58
#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
59
60
61
/* Function templates */
62
63
/* FSE_buildCTable_wksp() :
64
 * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
65
 * wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
66
 * workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
67
 */
68
size_t FSE_buildCTable_wksp(FSE_CTable* ct,
69
                      const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
70
                            void* workSpace, size_t wkspSize)
71
507k
{
72
507k
    U32 const tableSize = 1 << tableLog;
73
507k
    U32 const tableMask = tableSize - 1;
74
507k
    void* const ptr = ct;
75
507k
    U16* const tableU16 = ( (U16*) ptr) + 2;
76
507k
    void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
77
507k
    FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
78
507k
    U32 const step = FSE_TABLESTEP(tableSize);
79
507k
    U32 const maxSV1 = maxSymbolValue+1;
80
81
507k
    U16* cumul = (U16*)workSpace;   /* size = maxSV1 */
82
507k
    FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSV1+1));  /* size = tableSize */
83
84
507k
    U32 highThreshold = tableSize-1;
85
86
507k
    assert(((size_t)workSpace & 1) == 0);  /* Must be 2 bytes-aligned */
87
507k
    if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
88
    /* CTable header */
89
507k
    tableU16[-2] = (U16) tableLog;
90
507k
    tableU16[-1] = (U16) maxSymbolValue;
91
507k
    assert(tableLog < 16);   /* required for threshold strategy to work */
92
93
    /* For explanations on how to distribute symbol values over the table :
94
     * https://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
95
96
     #ifdef __clang_analyzer__
97
     ZSTD_memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
98
     #endif
99
100
    /* symbol start positions */
101
507k
    {   U32 u;
102
507k
        cumul[0] = 0;
103
11.2M
        for (u=1; u <= maxSV1; u++) {
104
10.7M
            if (normalizedCounter[u-1]==-1) {  /* Low proba symbol */
105
838k
                cumul[u] = cumul[u-1] + 1;
106
838k
                tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
107
9.94M
            } else {
108
9.94M
                assert(normalizedCounter[u-1] >= 0);
109
9.94M
                cumul[u] = cumul[u-1] + (U16)normalizedCounter[u-1];
110
9.94M
                assert(cumul[u] >= cumul[u-1]);  /* no overflow */
111
9.94M
        }   }
112
507k
        cumul[maxSV1] = (U16)(tableSize+1);
113
507k
    }
114
115
    /* Spread symbols */
116
507k
    if (highThreshold == tableSize - 1) {
117
        /* Case for no low prob count symbols. Lay down 8 bytes at a time
118
         * to reduce branch misses since we are operating on a small block
119
         */
120
354k
        BYTE* const spread = tableSymbol + tableSize; /* size = tableSize + 8 (may write beyond tableSize) */
121
354k
        {   U64 const add = 0x0101010101010101ull;
122
354k
            size_t pos = 0;
123
354k
            U64 sv = 0;
124
354k
            U32 s;
125
4.92M
            for (s=0; s<maxSV1; ++s, sv += add) {
126
4.57M
                int i;
127
4.57M
                int const n = normalizedCounter[s];
128
4.57M
                MEM_write64(spread + pos, sv);
129
5.85M
                for (i = 8; i < n; i += 8) {
130
1.28M
                    MEM_write64(spread + pos + i, sv);
131
1.28M
                }
132
4.57M
                assert(n>=0);
133
4.57M
                pos += (size_t)n;
134
4.57M
            }
135
354k
        }
136
        /* Spread symbols across the table. Lack of lowprob symbols means that
137
         * we don't need variable sized inner loop, so we can unroll the loop and
138
         * reduce branch misses.
139
         */
140
354k
        {   size_t position = 0;
141
354k
            size_t s;
142
354k
            size_t const unroll = 2; /* Experimentally determined optimal unroll */
143
354k
            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
144
9.36M
            for (s = 0; s < (size_t)tableSize; s += unroll) {
145
9.00M
                size_t u;
146
27.0M
                for (u = 0; u < unroll; ++u) {
147
18.0M
                    size_t const uPosition = (position + (u * step)) & tableMask;
148
18.0M
                    tableSymbol[uPosition] = spread[s + u];
149
18.0M
                }
150
9.00M
                position = (position + (unroll * step)) & tableMask;
151
9.00M
            }
152
354k
            assert(position == 0);   /* Must have initialized all positions */
153
354k
        }
154
354k
    } else {
155
153k
        U32 position = 0;
156
153k
        U32 symbol;
157
6.36M
        for (symbol=0; symbol<maxSV1; symbol++) {
158
6.21M
            int nbOccurrences;
159
6.21M
            int const freq = normalizedCounter[symbol];
160
15.3M
            for (nbOccurrences=0; nbOccurrences<freq; nbOccurrences++) {
161
9.17M
                tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
162
9.17M
                position = (position + step) & tableMask;
163
10.0M
                while (position > highThreshold)
164
838k
                    position = (position + step) & tableMask;   /* Low proba area */
165
9.17M
        }   }
166
153k
        assert(position==0);  /* Must have initialized all positions */
167
153k
    }
168
169
    /* Build table */
170
28.5M
    {   U32 u; for (u=0; u<tableSize; u++) {
171
28.0M
        FSE_FUNCTION_TYPE s = tableSymbol[u];   /* note : static analyzer may not understand tableSymbol is properly initialized */
172
28.0M
        tableU16[cumul[s]++] = (U16) (tableSize+u);   /* TableU16 : sorted by symbol order; gives next state value */
173
28.0M
    }   }
174
175
    /* Build Symbol Transformation Table */
176
507k
    {   unsigned total = 0;
177
507k
        unsigned s;
178
11.2M
        for (s=0; s<=maxSymbolValue; s++) {
179
10.7M
            switch (normalizedCounter[s])
180
10.7M
            {
181
1.71M
            case  0:
182
                /* filling nonetheless, for compatibility with FSE_getMaxNbBits() */
183
1.71M
                symbolTT[s].deltaNbBits = ((tableLog+1) << 16) - (1<<tableLog);
184
1.71M
                break;
185
186
838k
            case -1:
187
5.51M
            case  1:
188
5.51M
                symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
189
5.51M
                assert(total <= INT_MAX);
190
5.51M
                symbolTT[s].deltaFindState = (int)(total - 1);
191
5.51M
                total ++;
192
5.51M
                break;
193
3.55M
            default :
194
3.55M
                assert(normalizedCounter[s] > 1);
195
3.55M
                {   U32 const maxBitsOut = tableLog - ZSTD_highbit32 ((U32)normalizedCounter[s]-1);
196
3.55M
                    U32 const minStatePlus = (U32)normalizedCounter[s] << maxBitsOut;
197
3.55M
                    symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
198
3.55M
                    symbolTT[s].deltaFindState = (int)(total - (unsigned)normalizedCounter[s]);
199
3.55M
                    total +=  (unsigned)normalizedCounter[s];
200
3.55M
    }   }   }   }
201
202
#if 0  /* debug : symbol costs */
203
    DEBUGLOG(5, "\n --- table statistics : ");
204
    {   U32 symbol;
205
        for (symbol=0; symbol<=maxSymbolValue; symbol++) {
206
            DEBUGLOG(5, "%3u: w=%3i,   maxBits=%u, fracBits=%.2f",
207
                symbol, normalizedCounter[symbol],
208
                FSE_getMaxNbBits(symbolTT, symbol),
209
                (double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
210
    }   }
211
#endif
212
213
507k
    return 0;
214
507k
}
215
216
217
218
#ifndef FSE_COMMONDEFS_ONLY
219
220
/*-**************************************************************
221
*  FSE NCount encoding
222
****************************************************************/
223
size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
224
586k
{
225
586k
    size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog
226
586k
                                   + 4 /* bitCount initialized at 4 */
227
586k
                                   + 2 /* first two symbols may use one additional bit each */) / 8)
228
586k
                                   + 1 /* round up to whole nb bytes */
229
586k
                                   + 2 /* additional two bytes for bitstream flush */;
230
586k
    return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
231
586k
}
232
233
static size_t
234
FSE_writeNCount_generic (void* header, size_t headerBufferSize,
235
                   const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
236
                         unsigned writeIsSafe)
237
586k
{
238
586k
    BYTE* const ostart = (BYTE*) header;
239
586k
    BYTE* out = ostart;
240
586k
    BYTE* const oend = ostart + headerBufferSize;
241
586k
    int nbBits;
242
586k
    const int tableSize = 1 << tableLog;
243
586k
    int remaining;
244
586k
    int threshold;
245
586k
    U32 bitStream = 0;
246
586k
    int bitCount = 0;
247
586k
    unsigned symbol = 0;
248
586k
    unsigned const alphabetSize = maxSymbolValue + 1;
249
586k
    int previousIs0 = 0;
250
251
    /* Table Size */
252
586k
    bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
253
586k
    bitCount  += 4;
254
255
    /* Init */
256
586k
    remaining = tableSize+1;   /* +1 for extra accuracy */
257
586k
    threshold = tableSize;
258
586k
    nbBits = (int)tableLog+1;
259
260
6.77M
    while ((symbol < alphabetSize) && (remaining>1)) {  /* stops at 1 */
261
6.18M
        if (previousIs0) {
262
1.15M
            unsigned start = symbol;
263
4.71M
            while ((symbol < alphabetSize) && !normalizedCounter[symbol]) symbol++;
264
1.15M
            if (symbol == alphabetSize) break;   /* incorrect distribution */
265
1.19M
            while (symbol >= start+24) {
266
38.4k
                start+=24;
267
38.4k
                bitStream += 0xFFFFU << bitCount;
268
38.4k
                if ((!writeIsSafe) && (out > oend-2))
269
1
                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
270
38.4k
                out[0] = (BYTE) bitStream;
271
38.4k
                out[1] = (BYTE)(bitStream>>8);
272
38.4k
                out+=2;
273
38.4k
                bitStream>>=16;
274
38.4k
            }
275
1.82M
            while (symbol >= start+3) {
276
674k
                start+=3;
277
674k
                bitStream += 3U << bitCount;
278
674k
                bitCount += 2;
279
674k
            }
280
1.15M
            bitStream += (symbol-start) << bitCount;
281
1.15M
            bitCount += 2;
282
1.15M
            if (bitCount>16) {
283
232k
                if ((!writeIsSafe) && (out > oend - 2))
284
1
                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
285
232k
                out[0] = (BYTE)bitStream;
286
232k
                out[1] = (BYTE)(bitStream>>8);
287
232k
                out += 2;
288
232k
                bitStream >>= 16;
289
232k
                bitCount -= 16;
290
232k
        }   }
291
6.18M
        {   int count = normalizedCounter[symbol++];
292
6.18M
            int const max = (2*threshold-1) - remaining;
293
6.18M
            remaining -= count < 0 ? -count : count;
294
6.18M
            count++;   /* +1 for extra accuracy */
295
6.18M
            if (count>=threshold)
296
1.05M
                count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
297
6.18M
            bitStream += (U32)count << bitCount;
298
6.18M
            bitCount  += nbBits;
299
6.18M
            bitCount  -= (count<max);
300
6.18M
            previousIs0  = (count==1);
301
6.18M
            if (remaining<1) return ERROR(GENERIC);
302
9.42M
            while (remaining<threshold) { nbBits--; threshold>>=1; }
303
6.18M
        }
304
6.18M
        if (bitCount>16) {
305
1.35M
            if ((!writeIsSafe) && (out > oend - 2))
306
5
                return ERROR(dstSize_tooSmall);   /* Buffer overflow */
307
1.35M
            out[0] = (BYTE)bitStream;
308
1.35M
            out[1] = (BYTE)(bitStream>>8);
309
1.35M
            out += 2;
310
1.35M
            bitStream >>= 16;
311
1.35M
            bitCount -= 16;
312
1.35M
    }   }
313
314
586k
    if (remaining != 1)
315
0
        return ERROR(GENERIC);  /* incorrect normalized distribution */
316
586k
    assert(symbol <= alphabetSize);
317
318
    /* flush remaining bitStream */
319
586k
    if ((!writeIsSafe) && (out > oend - 2))
320
115
        return ERROR(dstSize_tooSmall);   /* Buffer overflow */
321
586k
    out[0] = (BYTE)bitStream;
322
586k
    out[1] = (BYTE)(bitStream>>8);
323
586k
    out+= (bitCount+7) /8;
324
325
586k
    assert(out >= ostart);
326
586k
    return (size_t)(out-ostart);
327
586k
}
328
329
330
size_t FSE_writeNCount (void* buffer, size_t bufferSize,
331
                  const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
332
586k
{
333
586k
    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
334
586k
    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
335
336
586k
    if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
337
991
        return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
338
339
585k
    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1 /* write in buffer is safe */);
340
586k
}
341
342
343
/*-**************************************************************
344
*  FSE Compression Code
345
****************************************************************/
346
347
/* provides the minimum logSize to safely represent a distribution */
348
static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
349
1.23M
{
350
1.23M
    U32 minBitsSrc = ZSTD_highbit32((U32)(srcSize)) + 1;
351
1.23M
    U32 minBitsSymbols = ZSTD_highbit32(maxSymbolValue) + 2;
352
1.23M
    U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
353
1.23M
    assert(srcSize > 1); /* Not supported, RLE should be used instead */
354
1.23M
    return minBits;
355
1.23M
}
356
357
unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
358
644k
{
359
644k
    U32 maxBitsSrc = ZSTD_highbit32((U32)(srcSize - 1)) - minus;
360
644k
    U32 tableLog = maxTableLog;
361
644k
    U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
362
644k
    assert(srcSize > 1); /* Not supported, RLE should be used instead */
363
644k
    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
364
644k
    if (maxBitsSrc < tableLog) tableLog = maxBitsSrc;   /* Accuracy can be reduced */
365
644k
    if (minBits > tableLog) tableLog = minBits;   /* Need a minimum to safely represent all symbol values */
366
644k
    if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
367
644k
    if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
368
644k
    return tableLog;
369
644k
}
370
371
unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
372
586k
{
373
586k
    return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
374
586k
}
375
376
/* Secondary normalization method.
377
   To be used when primary method fails. */
378
379
static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue, short lowProbCount)
380
4.11k
{
381
4.11k
    short const NOT_YET_ASSIGNED = -2;
382
4.11k
    U32 s;
383
4.11k
    U32 distributed = 0;
384
4.11k
    U32 ToDistribute;
385
386
    /* Init */
387
4.11k
    U32 const lowThreshold = (U32)(total >> tableLog);
388
4.11k
    U32 lowOne = (U32)((total * 3) >> (tableLog + 1));
389
390
128k
    for (s=0; s<=maxSymbolValue; s++) {
391
124k
        if (count[s] == 0) {
392
49.7k
            norm[s]=0;
393
49.7k
            continue;
394
49.7k
        }
395
74.5k
        if (count[s] <= lowThreshold) {
396
16.9k
            norm[s] = lowProbCount;
397
16.9k
            distributed++;
398
16.9k
            total -= count[s];
399
16.9k
            continue;
400
16.9k
        }
401
57.6k
        if (count[s] <= lowOne) {
402
10.4k
            norm[s] = 1;
403
10.4k
            distributed++;
404
10.4k
            total -= count[s];
405
10.4k
            continue;
406
10.4k
        }
407
408
47.1k
        norm[s]=NOT_YET_ASSIGNED;
409
47.1k
    }
410
4.11k
    ToDistribute = (1 << tableLog) - distributed;
411
412
4.11k
    if (ToDistribute == 0)
413
0
        return 0;
414
415
4.11k
    if ((total / ToDistribute) > lowOne) {
416
        /* risk of rounding to zero */
417
14
        lowOne = (U32)((total * 3) / (ToDistribute * 2));
418
318
        for (s=0; s<=maxSymbolValue; s++) {
419
304
            if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
420
9
                norm[s] = 1;
421
9
                distributed++;
422
9
                total -= count[s];
423
9
                continue;
424
9
        }   }
425
14
        ToDistribute = (1 << tableLog) - distributed;
426
14
    }
427
428
4.11k
    if (distributed == maxSymbolValue+1) {
429
        /* all values are pretty poor;
430
           probably incompressible data (should have already been detected);
431
           find max, then give all remaining points to max */
432
0
        U32 maxV = 0, maxC = 0;
433
0
        for (s=0; s<=maxSymbolValue; s++)
434
0
            if (count[s] > maxC) { maxV=s; maxC=count[s]; }
435
0
        norm[maxV] += (short)ToDistribute;
436
0
        return 0;
437
0
    }
438
439
4.11k
    if (total == 0) {
440
        /* all of the symbols were low enough for the lowOne or lowThreshold */
441
3.16k
        for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
442
3.02k
            if (norm[s] > 0) { ToDistribute--; norm[s]++; }
443
146
        return 0;
444
146
    }
445
446
3.96k
    {   U64 const vStepLog = 62 - tableLog;
447
3.96k
        U64 const mid = (1ULL << (vStepLog-1)) - 1;
448
3.96k
        U64 const rStep = ZSTD_div64((((U64)1<<vStepLog) * ToDistribute) + mid, (U32)total);   /* scale on remaining */
449
3.96k
        U64 tmpTotal = mid;
450
121k
        for (s=0; s<=maxSymbolValue; s++) {
451
117k
            if (norm[s]==NOT_YET_ASSIGNED) {
452
47.1k
                U64 const end = tmpTotal + (count[s] * rStep);
453
47.1k
                U32 const sStart = (U32)(tmpTotal >> vStepLog);
454
47.1k
                U32 const sEnd = (U32)(end >> vStepLog);
455
47.1k
                U32 const weight = sEnd - sStart;
456
47.1k
                if (weight < 1)
457
0
                    return ERROR(GENERIC);
458
47.1k
                norm[s] = (short)weight;
459
47.1k
                tmpTotal = end;
460
47.1k
    }   }   }
461
462
3.96k
    return 0;
463
3.96k
}
464
465
size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
466
                           const unsigned* count, size_t total,
467
                           unsigned maxSymbolValue, unsigned useLowProbCount)
468
586k
{
469
    /* Sanity checks */
470
586k
    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
471
586k
    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported size */
472
586k
    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported size */
473
586k
    if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC);   /* Too small tableLog, compression potentially impossible */
474
475
586k
    {   static U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
476
586k
        short const lowProbCount = useLowProbCount ? -1 : 1;
477
586k
        U64 const scale = 62 - tableLog;
478
586k
        U64 const step = ZSTD_div64((U64)1<<62, (U32)total);   /* <== here, one division ! */
479
586k
        U64 const vStep = 1ULL<<(scale-20);
480
586k
        int stillToDistribute = 1<<tableLog;
481
586k
        unsigned s;
482
586k
        unsigned largest=0;
483
586k
        short largestP=0;
484
586k
        U32 lowThreshold = (U32)(total >> tableLog);
485
486
10.3M
        for (s=0; s<=maxSymbolValue; s++) {
487
9.74M
            if (count[s] == total) return 0;   /* rle special case */
488
9.74M
            if (count[s] == 0) { normalizedCounter[s]=0; continue; }
489
5.03M
            if (count[s] <= lowThreshold) {
490
1.46M
                normalizedCounter[s] = lowProbCount;
491
1.46M
                stillToDistribute--;
492
3.57M
            } else {
493
3.57M
                short proba = (short)((count[s]*step) >> scale);
494
3.57M
                if (proba<8) {
495
2.54M
                    U64 restToBeat = vStep * rtbTable[proba];
496
2.54M
                    proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
497
2.54M
                }
498
3.57M
                if (proba > largestP) { largestP=proba; largest=s; }
499
3.57M
                normalizedCounter[s] = proba;
500
3.57M
                stillToDistribute -= proba;
501
3.57M
        }   }
502
586k
        if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
503
            /* corner case, need another normalization method */
504
4.11k
            size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue, lowProbCount);
505
4.11k
            if (FSE_isError(errorCode)) return errorCode;
506
4.11k
        }
507
582k
        else normalizedCounter[largest] += (short)stillToDistribute;
508
586k
    }
509
510
#if 0
511
    {   /* Print Table (debug) */
512
        U32 s;
513
        U32 nTotal = 0;
514
        for (s=0; s<=maxSymbolValue; s++)
515
            RAWLOG(2, "%3i: %4i \n", s, normalizedCounter[s]);
516
        for (s=0; s<=maxSymbolValue; s++)
517
            nTotal += abs(normalizedCounter[s]);
518
        if (nTotal != (1U<<tableLog))
519
            RAWLOG(2, "Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
520
        getchar();
521
    }
522
#endif
523
524
586k
    return tableLog;
525
586k
}
526
527
/* fake FSE_CTable, for rle input (always same symbol) */
528
size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
529
14.5k
{
530
14.5k
    void* ptr = ct;
531
14.5k
    U16* tableU16 = ( (U16*) ptr) + 2;
532
14.5k
    void* FSCTptr = (U32*)ptr + 2;
533
14.5k
    FSE_symbolCompressionTransform* symbolTT = (FSE_symbolCompressionTransform*) FSCTptr;
534
535
    /* header */
536
14.5k
    tableU16[-2] = (U16) 0;
537
14.5k
    tableU16[-1] = (U16) symbolValue;
538
539
    /* Build table */
540
14.5k
    tableU16[0] = 0;
541
14.5k
    tableU16[1] = 0;   /* just in case */
542
543
    /* Build Symbol Transformation Table */
544
14.5k
    symbolTT[symbolValue].deltaNbBits = 0;
545
14.5k
    symbolTT[symbolValue].deltaFindState = 0;
546
547
14.5k
    return 0;
548
14.5k
}
549
550
551
static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
552
                           const void* src, size_t srcSize,
553
                           const FSE_CTable* ct, const unsigned fast)
554
169k
{
555
169k
    const BYTE* const istart = (const BYTE*) src;
556
169k
    const BYTE* const iend = istart + srcSize;
557
169k
    const BYTE* ip=iend;
558
559
169k
    BIT_CStream_t bitC;
560
169k
    FSE_CState_t CState1, CState2;
561
562
    /* init */
563
169k
    if (srcSize <= 2) return 0;
564
169k
    { size_t const initError = BIT_initCStream(&bitC, dst, dstSize);
565
169k
      if (FSE_isError(initError)) return 0; /* not enough space available to write a bitstream */ }
566
567
9.89M
#define FSE_FLUSHBITS(s)  (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
568
569
169k
    if (srcSize & 1) {
570
127k
        FSE_initCState2(&CState1, ct, *--ip);
571
127k
        FSE_initCState2(&CState2, ct, *--ip);
572
127k
        FSE_encodeSymbol(&bitC, &CState1, *--ip);
573
127k
        FSE_FLUSHBITS(&bitC);
574
127k
    } else {
575
41.8k
        FSE_initCState2(&CState2, ct, *--ip);
576
41.8k
        FSE_initCState2(&CState1, ct, *--ip);
577
41.8k
    }
578
579
    /* join to mod 4 */
580
169k
    srcSize -= 2;
581
169k
    if ((sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) && (srcSize & 2)) {  /* test bit 2 */
582
32.1k
        FSE_encodeSymbol(&bitC, &CState2, *--ip);
583
32.1k
        FSE_encodeSymbol(&bitC, &CState1, *--ip);
584
32.1k
        FSE_FLUSHBITS(&bitC);
585
32.1k
    }
586
587
    /* 2 or 4 encoding per loop */
588
9.90M
    while ( ip>istart ) {
589
590
9.73M
        FSE_encodeSymbol(&bitC, &CState2, *--ip);
591
592
9.73M
        if (sizeof(bitC.bitContainer)*8 < FSE_MAX_TABLELOG*2+7 )   /* this test must be static */
593
0
            FSE_FLUSHBITS(&bitC);
594
595
9.73M
        FSE_encodeSymbol(&bitC, &CState1, *--ip);
596
597
9.73M
        if (sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) {  /* this test must be static */
598
9.73M
            FSE_encodeSymbol(&bitC, &CState2, *--ip);
599
9.73M
            FSE_encodeSymbol(&bitC, &CState1, *--ip);
600
9.73M
        }
601
602
9.73M
        FSE_FLUSHBITS(&bitC);
603
9.73M
    }
604
605
169k
    FSE_flushCState(&bitC, &CState2);
606
169k
    FSE_flushCState(&bitC, &CState1);
607
169k
    return BIT_closeCStream(&bitC);
608
169k
}
609
610
size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
611
                           const void* src, size_t srcSize,
612
                           const FSE_CTable* ct)
613
169k
{
614
169k
    unsigned const fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
615
616
169k
    if (fast)
617
167k
        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 1);
618
1.80k
    else
619
1.80k
        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 0);
620
169k
}
621
622
623
0
size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
624
625
#endif   /* FSE_COMMONDEFS_ONLY */