Coverage Report

Created: 2026-02-11 06:06

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/src/zstd/lib/compress/zstd_opt.c
Line
Count
Source
1
/*
2
 * Copyright (c) Meta Platforms, Inc. and affiliates.
3
 * All rights reserved.
4
 *
5
 * This source code is licensed under both the BSD-style license (found in the
6
 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7
 * in the COPYING file in the root directory of this source tree).
8
 * You may select, at your option, one of the above-listed licenses.
9
 */
10
11
#include "zstd_compress_internal.h"
12
#include "hist.h"
13
#include "zstd_opt.h"
14
15
#if !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR) \
16
 || !defined(ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR) \
17
 || !defined(ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR)
18
19
79.6M
#define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats */
20
157M
#define ZSTD_MAX_PRICE     (1<<30)
21
22
7.53k
#define ZSTD_PREDEF_THRESHOLD 8   /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */
23
24
25
/*-*************************************
26
*  Price functions for optimal parser
27
***************************************/
28
29
#if 0    /* approximation at bit level (for tests) */
30
#  define BITCOST_ACCURACY 0
31
#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
32
#  define WEIGHT(stat, opt) ((void)(opt), ZSTD_bitWeight(stat))
33
#elif 0  /* fractional bit accuracy (for tests) */
34
#  define BITCOST_ACCURACY 8
35
#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
36
#  define WEIGHT(stat,opt) ((void)(opt), ZSTD_fracWeight(stat))
37
#else    /* opt==approx, ultra==accurate */
38
6.50G
#  define BITCOST_ACCURACY 8
39
4.71G
#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
40
1.98G
#  define WEIGHT(stat,opt) ((opt) ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))
41
#endif
42
43
/* ZSTD_bitWeight() :
44
 * provide estimated "cost" of a stat in full bits only */
45
MEM_STATIC U32 ZSTD_bitWeight(U32 stat)
46
193M
{
47
193M
    return (ZSTD_highbit32(stat+1) * BITCOST_MULTIPLIER);
48
193M
}
49
50
/* ZSTD_fracWeight() :
51
 * provide fractional-bit "cost" of a stat,
52
 * using linear interpolation approximation */
53
MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)
54
1.79G
{
55
1.79G
    U32 const stat = rawStat + 1;
56
1.79G
    U32 const hb = ZSTD_highbit32(stat);
57
1.79G
    U32 const BWeight = hb * BITCOST_MULTIPLIER;
58
    /* Fweight was meant for "Fractional weight"
59
     * but it's effectively a value between 1 and 2
60
     * using fixed point arithmetic */
61
1.79G
    U32 const FWeight = (stat << BITCOST_ACCURACY) >> hb;
62
1.79G
    U32 const weight = BWeight + FWeight;
63
1.79G
    assert(hb + BITCOST_ACCURACY < 31);
64
1.79G
    return weight;
65
1.79G
}
66
67
#if (DEBUGLEVEL>=2)
68
/* debugging function,
69
 * @return price in bytes as fractional value
70
 * for debug messages only */
71
MEM_STATIC double ZSTD_fCost(int price)
72
{
73
    return (double)price / (BITCOST_MULTIPLIER*8);
74
}
75
#endif
76
77
static int ZSTD_compressedLiterals(optState_t const* const optPtr)
78
139M
{
79
139M
    return optPtr->literalCompressionMode != ZSTD_ps_disable;
80
139M
}
81
82
static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
83
5.02M
{
84
5.02M
    if (ZSTD_compressedLiterals(optPtr))
85
5.02M
        optPtr->litSumBasePrice = WEIGHT(optPtr->litSum, optLevel);
86
5.02M
    optPtr->litLengthSumBasePrice = WEIGHT(optPtr->litLengthSum, optLevel);
87
5.02M
    optPtr->matchLengthSumBasePrice = WEIGHT(optPtr->matchLengthSum, optLevel);
88
5.02M
    optPtr->offCodeSumBasePrice = WEIGHT(optPtr->offCodeSum, optLevel);
89
5.02M
}
90
91
92
static U32 sum_u32(const unsigned table[], size_t nbElts)
93
39.0k
{
94
39.0k
    size_t n;
95
39.0k
    U32 total = 0;
96
3.08M
    for (n=0; n<nbElts; n++) {
97
3.05M
        total += table[n];
98
3.05M
    }
99
39.0k
    return total;
100
39.0k
}
101
102
typedef enum { base_0possible=0, base_1guaranteed=1 } base_directive_e;
103
104
static U32
105
ZSTD_downscaleStats(unsigned* table, U32 lastEltIndex, U32 shift, base_directive_e base1)
106
11.9k
{
107
11.9k
    U32 s, sum=0;
108
11.9k
    DEBUGLOG(5, "ZSTD_downscaleStats (nbElts=%u, shift=%u)",
109
11.9k
            (unsigned)lastEltIndex+1, (unsigned)shift );
110
11.9k
    assert(shift < 30);
111
2.15M
    for (s=0; s<lastEltIndex+1; s++) {
112
2.13M
        unsigned const base = base1 ? 1 : (table[s]>0);
113
2.13M
        unsigned const newStat = base + (table[s] >> shift);
114
2.13M
        sum += newStat;
115
2.13M
        table[s] = newStat;
116
2.13M
    }
117
11.9k
    return sum;
118
11.9k
}
119
120
/* ZSTD_scaleStats() :
121
 * reduce all elt frequencies in table if sum too large
122
 * return the resulting sum of elements */
123
static U32 ZSTD_scaleStats(unsigned* table, U32 lastEltIndex, U32 logTarget)
124
28.5k
{
125
28.5k
    U32 const prevsum = sum_u32(table, lastEltIndex+1);
126
28.5k
    U32 const factor = prevsum >> logTarget;
127
28.5k
    DEBUGLOG(5, "ZSTD_scaleStats (nbElts=%u, target=%u)", (unsigned)lastEltIndex+1, (unsigned)logTarget);
128
28.5k
    assert(logTarget < 30);
129
28.5k
    if (factor <= 1) return prevsum;
130
6.70k
    return ZSTD_downscaleStats(table, lastEltIndex, ZSTD_highbit32(factor), base_1guaranteed);
131
28.5k
}
132
133
/* ZSTD_rescaleFreqs() :
134
 * if first block (detected by optPtr->litLengthSum == 0) : init statistics
135
 *    take hints from dictionary if there is one
136
 *    and init from zero if there is none,
137
 *    using src for literals stats, and baseline stats for sequence symbols
138
 * otherwise downscale existing stats, to be used as seed for next block.
139
 */
140
static void
141
ZSTD_rescaleFreqs(optState_t* const optPtr,
142
            const BYTE* const src, size_t const srcSize,
143
                  int const optLevel)
144
12.3k
{
145
12.3k
    int const compressedLiterals = ZSTD_compressedLiterals(optPtr);
146
12.3k
    DEBUGLOG(5, "ZSTD_rescaleFreqs (srcSize=%u)", (unsigned)srcSize);
147
12.3k
    optPtr->priceType = zop_dynamic;
148
149
12.3k
    if (optPtr->litLengthSum == 0) {  /* no literals stats collected -> first block assumed -> init */
150
151
        /* heuristic: use pre-defined stats for too small inputs */
152
5.25k
        if (srcSize <= ZSTD_PREDEF_THRESHOLD) {
153
69
            DEBUGLOG(5, "srcSize <= %i : use predefined stats", ZSTD_PREDEF_THRESHOLD);
154
69
            optPtr->priceType = zop_predef;
155
69
        }
156
157
5.25k
        assert(optPtr->symbolCosts != NULL);
158
5.25k
        if (optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid) {
159
160
            /* huffman stats covering the full value set : table presumed generated by dictionary */
161
0
            optPtr->priceType = zop_dynamic;
162
163
0
            if (compressedLiterals) {
164
                /* generate literals statistics from huffman table */
165
0
                unsigned lit;
166
0
                assert(optPtr->litFreq != NULL);
167
0
                optPtr->litSum = 0;
168
0
                for (lit=0; lit<=MaxLit; lit++) {
169
0
                    U32 const scaleLog = 11;   /* scale to 2K */
170
0
                    U32 const bitCost = HUF_getNbBitsFromCTable(optPtr->symbolCosts->huf.CTable, lit);
171
0
                    assert(bitCost <= scaleLog);
172
0
                    optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
173
0
                    optPtr->litSum += optPtr->litFreq[lit];
174
0
            }   }
175
176
0
            {   unsigned ll;
177
0
                FSE_CState_t llstate;
178
0
                FSE_initCState(&llstate, optPtr->symbolCosts->fse.litlengthCTable);
179
0
                optPtr->litLengthSum = 0;
180
0
                for (ll=0; ll<=MaxLL; ll++) {
181
0
                    U32 const scaleLog = 10;   /* scale to 1K */
182
0
                    U32 const bitCost = FSE_getMaxNbBits(llstate.symbolTT, ll);
183
0
                    assert(bitCost < scaleLog);
184
0
                    optPtr->litLengthFreq[ll] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
185
0
                    optPtr->litLengthSum += optPtr->litLengthFreq[ll];
186
0
            }   }
187
188
0
            {   unsigned ml;
189
0
                FSE_CState_t mlstate;
190
0
                FSE_initCState(&mlstate, optPtr->symbolCosts->fse.matchlengthCTable);
191
0
                optPtr->matchLengthSum = 0;
192
0
                for (ml=0; ml<=MaxML; ml++) {
193
0
                    U32 const scaleLog = 10;
194
0
                    U32 const bitCost = FSE_getMaxNbBits(mlstate.symbolTT, ml);
195
0
                    assert(bitCost < scaleLog);
196
0
                    optPtr->matchLengthFreq[ml] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
197
0
                    optPtr->matchLengthSum += optPtr->matchLengthFreq[ml];
198
0
            }   }
199
200
0
            {   unsigned of;
201
0
                FSE_CState_t ofstate;
202
0
                FSE_initCState(&ofstate, optPtr->symbolCosts->fse.offcodeCTable);
203
0
                optPtr->offCodeSum = 0;
204
0
                for (of=0; of<=MaxOff; of++) {
205
0
                    U32 const scaleLog = 10;
206
0
                    U32 const bitCost = FSE_getMaxNbBits(ofstate.symbolTT, of);
207
0
                    assert(bitCost < scaleLog);
208
0
                    optPtr->offCodeFreq[of] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
209
0
                    optPtr->offCodeSum += optPtr->offCodeFreq[of];
210
0
            }   }
211
212
5.25k
        } else {  /* first block, no dictionary */
213
214
5.25k
            assert(optPtr->litFreq != NULL);
215
5.25k
            if (compressedLiterals) {
216
                /* base initial cost of literals on direct frequency within src */
217
5.25k
                unsigned lit = MaxLit;
218
5.25k
                HIST_count_simple(optPtr->litFreq, &lit, src, srcSize);   /* use raw first block to init statistics */
219
5.25k
                optPtr->litSum = ZSTD_downscaleStats(optPtr->litFreq, MaxLit, 8, base_0possible);
220
5.25k
            }
221
222
5.25k
            {   unsigned const baseLLfreqs[MaxLL+1] = {
223
5.25k
                    4, 2, 1, 1, 1, 1, 1, 1,
224
5.25k
                    1, 1, 1, 1, 1, 1, 1, 1,
225
5.25k
                    1, 1, 1, 1, 1, 1, 1, 1,
226
5.25k
                    1, 1, 1, 1, 1, 1, 1, 1,
227
5.25k
                    1, 1, 1, 1
228
5.25k
                };
229
5.25k
                ZSTD_memcpy(optPtr->litLengthFreq, baseLLfreqs, sizeof(baseLLfreqs));
230
5.25k
                optPtr->litLengthSum = sum_u32(baseLLfreqs, MaxLL+1);
231
5.25k
            }
232
233
5.25k
            {   unsigned ml;
234
283k
                for (ml=0; ml<=MaxML; ml++)
235
278k
                    optPtr->matchLengthFreq[ml] = 1;
236
5.25k
            }
237
5.25k
            optPtr->matchLengthSum = MaxML+1;
238
239
5.25k
            {   unsigned const baseOFCfreqs[MaxOff+1] = {
240
5.25k
                    6, 2, 1, 1, 2, 3, 4, 4,
241
5.25k
                    4, 3, 2, 1, 1, 1, 1, 1,
242
5.25k
                    1, 1, 1, 1, 1, 1, 1, 1,
243
5.25k
                    1, 1, 1, 1, 1, 1, 1, 1
244
5.25k
                };
245
5.25k
                ZSTD_memcpy(optPtr->offCodeFreq, baseOFCfreqs, sizeof(baseOFCfreqs));
246
5.25k
                optPtr->offCodeSum = sum_u32(baseOFCfreqs, MaxOff+1);
247
5.25k
            }
248
249
5.25k
        }
250
251
7.14k
    } else {   /* new block : scale down accumulated statistics */
252
253
7.14k
        if (compressedLiterals)
254
7.14k
            optPtr->litSum = ZSTD_scaleStats(optPtr->litFreq, MaxLit, 12);
255
7.14k
        optPtr->litLengthSum = ZSTD_scaleStats(optPtr->litLengthFreq, MaxLL, 11);
256
7.14k
        optPtr->matchLengthSum = ZSTD_scaleStats(optPtr->matchLengthFreq, MaxML, 11);
257
7.14k
        optPtr->offCodeSum = ZSTD_scaleStats(optPtr->offCodeFreq, MaxOff, 11);
258
7.14k
    }
259
260
12.3k
    ZSTD_setBasePrices(optPtr, optLevel);
261
12.3k
}
262
263
/* ZSTD_rawLiteralsCost() :
264
 * price of literals (only) in specified segment (which length can be 0).
265
 * does not include price of literalLength symbol */
266
static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
267
                                const optState_t* const optPtr,
268
                                int optLevel)
269
117M
{
270
117M
    DEBUGLOG(8, "ZSTD_rawLiteralsCost (%u literals)", litLength);
271
117M
    if (litLength == 0) return 0;
272
273
117M
    if (!ZSTD_compressedLiterals(optPtr))
274
0
        return (litLength << 3) * BITCOST_MULTIPLIER;  /* Uncompressed - 8 bytes per literal. */
275
276
117M
    if (optPtr->priceType == zop_predef)
277
0
        return (litLength*6) * BITCOST_MULTIPLIER;  /* 6 bit per literal - no statistic used */
278
279
    /* dynamic statistics */
280
117M
    {   U32 price = optPtr->litSumBasePrice * litLength;
281
117M
        U32 const litPriceMax = optPtr->litSumBasePrice - BITCOST_MULTIPLIER;
282
117M
        U32 u;
283
117M
        assert(optPtr->litSumBasePrice >= BITCOST_MULTIPLIER);
284
234M
        for (u=0; u < litLength; u++) {
285
117M
            U32 litPrice = WEIGHT(optPtr->litFreq[literals[u]], optLevel);
286
117M
            if (UNLIKELY(litPrice > litPriceMax)) litPrice = litPriceMax;
287
117M
            price -= litPrice;
288
117M
        }
289
117M
        return price;
290
117M
    }
291
117M
}
292
293
/* ZSTD_litLengthPrice() :
294
 * cost of literalLength symbol */
295
static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr, int optLevel)
296
370M
{
297
370M
    assert(litLength <= ZSTD_BLOCKSIZE_MAX);
298
370M
    if (optPtr->priceType == zop_predef)
299
0
        return WEIGHT(litLength, optLevel);
300
301
    /* ZSTD_LLcode() can't compute litLength price for sizes >= ZSTD_BLOCKSIZE_MAX
302
     * because it isn't representable in the zstd format.
303
     * So instead just pretend it would cost 1 bit more than ZSTD_BLOCKSIZE_MAX - 1.
304
     * In such a case, the block would be all literals.
305
     */
306
370M
    if (litLength == ZSTD_BLOCKSIZE_MAX)
307
0
        return BITCOST_MULTIPLIER + ZSTD_litLengthPrice(ZSTD_BLOCKSIZE_MAX - 1, optPtr, optLevel);
308
309
    /* dynamic statistics */
310
370M
    {   U32 const llCode = ZSTD_LLcode(litLength);
311
370M
        return (LL_bits[llCode] * BITCOST_MULTIPLIER)
312
370M
             + optPtr->litLengthSumBasePrice
313
370M
             - WEIGHT(optPtr->litLengthFreq[llCode], optLevel);
314
370M
    }
315
370M
}
316
317
/* ZSTD_getMatchPrice() :
318
 * Provides the cost of the match part (offset + matchLength) of a sequence.
319
 * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
320
 * @offBase : sumtype, representing an offset or a repcode, and using numeric representation of ZSTD_storeSeq()
321
 * @optLevel: when <2, favors small offset for decompression speed (improved cache efficiency)
322
 */
323
FORCE_INLINE_TEMPLATE U32
324
ZSTD_getMatchPrice(U32 const offBase,
325
                   U32 const matchLength,
326
             const optState_t* const optPtr,
327
                   int const optLevel)
328
738M
{
329
738M
    U32 price;
330
738M
    U32 const offCode = ZSTD_highbit32(offBase);
331
738M
    U32 const mlBase = matchLength - MINMATCH;
332
738M
    assert(matchLength >= MINMATCH);
333
334
738M
    if (optPtr->priceType == zop_predef)  /* fixed scheme, does not use statistics */
335
0
        return WEIGHT(mlBase, optLevel)
336
0
             + ((16 + offCode) * BITCOST_MULTIPLIER); /* emulated offset cost */
337
338
    /* dynamic statistics */
339
738M
    price = (offCode * BITCOST_MULTIPLIER) + (optPtr->offCodeSumBasePrice - WEIGHT(optPtr->offCodeFreq[offCode], optLevel));
340
738M
    if ((optLevel<2) /*static*/ && offCode >= 20)
341
0
        price += (offCode-19)*2 * BITCOST_MULTIPLIER; /* handicap for long distance offsets, favor decompression speed */
342
343
    /* match Length */
344
738M
    {   U32 const mlCode = ZSTD_MLcode(mlBase);
345
738M
        price += (ML_bits[mlCode] * BITCOST_MULTIPLIER) + (optPtr->matchLengthSumBasePrice - WEIGHT(optPtr->matchLengthFreq[mlCode], optLevel));
346
738M
    }
347
348
738M
    price += BITCOST_MULTIPLIER / 5;   /* heuristic : make matches a bit more costly to favor less sequences -> faster decompression speed */
349
350
738M
    DEBUGLOG(8, "ZSTD_getMatchPrice(ml:%u) = %u", matchLength, price);
351
738M
    return price;
352
738M
}
353
354
/* ZSTD_updateStats() :
355
 * assumption : literals + litLength <= iend */
356
static void ZSTD_updateStats(optState_t* const optPtr,
357
                             U32 litLength, const BYTE* literals,
358
                             U32 offBase, U32 matchLength)
359
17.3M
{
360
    /* literals */
361
17.3M
    if (ZSTD_compressedLiterals(optPtr)) {
362
17.3M
        U32 u;
363
79.6M
        for (u=0; u < litLength; u++)
364
62.2M
            optPtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
365
17.3M
        optPtr->litSum += litLength*ZSTD_LITFREQ_ADD;
366
17.3M
    }
367
368
    /* literal Length */
369
17.3M
    {   U32 const llCode = ZSTD_LLcode(litLength);
370
17.3M
        optPtr->litLengthFreq[llCode]++;
371
17.3M
        optPtr->litLengthSum++;
372
17.3M
    }
373
374
    /* offset code : follows storeSeq() numeric representation */
375
17.3M
    {   U32 const offCode = ZSTD_highbit32(offBase);
376
17.3M
        assert(offCode <= MaxOff);
377
17.3M
        optPtr->offCodeFreq[offCode]++;
378
17.3M
        optPtr->offCodeSum++;
379
17.3M
    }
380
381
    /* match Length */
382
17.3M
    {   U32 const mlBase = matchLength - MINMATCH;
383
17.3M
        U32 const mlCode = ZSTD_MLcode(mlBase);
384
17.3M
        optPtr->matchLengthFreq[mlCode]++;
385
17.3M
        optPtr->matchLengthSum++;
386
17.3M
    }
387
17.3M
}
388
389
390
/* ZSTD_readMINMATCH() :
391
 * function safe only for comparisons
392
 * assumption : memPtr must be at least 4 bytes before end of buffer */
393
MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
394
856M
{
395
856M
    switch (length)
396
856M
    {
397
0
    default :
398
226M
    case 4 : return MEM_read32(memPtr);
399
630M
    case 3 : if (MEM_isLittleEndian())
400
630M
                return MEM_read32(memPtr)<<8;
401
0
             else
402
0
                return MEM_read32(memPtr)>>8;
403
856M
    }
404
856M
}
405
406
407
/* Update hashTable3 up to ip (excluded)
408
   Assumption : always within prefix (i.e. not within extDict) */
409
static
410
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
411
U32 ZSTD_insertAndFindFirstIndexHash3 (const ZSTD_MatchState_t* ms,
412
                                       U32* nextToUpdate3,
413
                                       const BYTE* const ip)
414
72.8M
{
415
72.8M
    U32* const hashTable3 = ms->hashTable3;
416
72.8M
    U32 const hashLog3 = ms->hashLog3;
417
72.8M
    const BYTE* const base = ms->window.base;
418
72.8M
    U32 idx = *nextToUpdate3;
419
72.8M
    U32 const target = (U32)(ip - base);
420
72.8M
    size_t const hash3 = ZSTD_hash3Ptr(ip, hashLog3);
421
72.8M
    assert(hashLog3 > 0);
422
423
261M
    while(idx < target) {
424
188M
        hashTable3[ZSTD_hash3Ptr(base+idx, hashLog3)] = idx;
425
188M
        idx++;
426
188M
    }
427
428
72.8M
    *nextToUpdate3 = target;
429
72.8M
    return hashTable3[hash3];
430
72.8M
}
431
432
433
/*-*************************************
434
*  Binary Tree search
435
***************************************/
436
/** ZSTD_insertBt1() : add one or multiple positions to tree.
437
 * @param ip assumed <= iend-8 .
438
 * @param target The target of ZSTD_updateTree_internal() - we are filling to this position
439
 * @return : nb of positions added */
440
static
441
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
442
U32 ZSTD_insertBt1(
443
                const ZSTD_MatchState_t* ms,
444
                const BYTE* const ip, const BYTE* const iend,
445
                U32 const target,
446
                U32 const mls, const int extDict)
447
30.2M
{
448
30.2M
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
449
30.2M
    U32*   const hashTable = ms->hashTable;
450
30.2M
    U32    const hashLog = cParams->hashLog;
451
30.2M
    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
452
30.2M
    U32*   const bt = ms->chainTable;
453
30.2M
    U32    const btLog  = cParams->chainLog - 1;
454
30.2M
    U32    const btMask = (1 << btLog) - 1;
455
30.2M
    U32 matchIndex = hashTable[h];
456
30.2M
    size_t commonLengthSmaller=0, commonLengthLarger=0;
457
30.2M
    const BYTE* const base = ms->window.base;
458
30.2M
    const BYTE* const dictBase = ms->window.dictBase;
459
30.2M
    const U32 dictLimit = ms->window.dictLimit;
460
30.2M
    const BYTE* const dictEnd = dictBase + dictLimit;
461
30.2M
    const BYTE* const prefixStart = base + dictLimit;
462
30.2M
    const BYTE* match;
463
30.2M
    const U32 curr = (U32)(ip-base);
464
30.2M
    const U32 btLow = btMask >= curr ? 0 : curr - btMask;
465
30.2M
    U32* smallerPtr = bt + 2*(curr&btMask);
466
30.2M
    U32* largerPtr  = smallerPtr + 1;
467
30.2M
    U32 dummy32;   /* to be nullified at the end */
468
    /* windowLow is based on target because
469
     * we only need positions that will be in the window at the end of the tree update.
470
     */
471
30.2M
    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, target, cParams->windowLog);
472
30.2M
    U32 matchEndIdx = curr+8+1;
473
30.2M
    size_t bestLength = 8;
474
30.2M
    U32 nbCompares = 1U << cParams->searchLog;
475
#ifdef ZSTD_C_PREDICT
476
    U32 predictedSmall = *(bt + 2*((curr-1)&btMask) + 0);
477
    U32 predictedLarge = *(bt + 2*((curr-1)&btMask) + 1);
478
    predictedSmall += (predictedSmall>0);
479
    predictedLarge += (predictedLarge>0);
480
#endif /* ZSTD_C_PREDICT */
481
482
30.2M
    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", curr);
483
484
30.2M
    assert(curr <= target);
485
30.2M
    assert(ip <= iend-8);   /* required for h calculation */
486
30.2M
    hashTable[h] = curr;   /* Update Hash Table */
487
488
30.2M
    assert(windowLow > 0);
489
238M
    for (; nbCompares && (matchIndex >= windowLow); --nbCompares) {
490
207M
        U32* const nextPtr = bt + 2*(matchIndex & btMask);
491
207M
        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
492
207M
        assert(matchIndex < curr);
493
494
#ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */
495
        const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */
496
        if (matchIndex == predictedSmall) {
497
            /* no need to check length, result known */
498
            *smallerPtr = matchIndex;
499
            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
500
            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
501
            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
502
            predictedSmall = predictPtr[1] + (predictPtr[1]>0);
503
            continue;
504
        }
505
        if (matchIndex == predictedLarge) {
506
            *largerPtr = matchIndex;
507
            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
508
            largerPtr = nextPtr;
509
            matchIndex = nextPtr[0];
510
            predictedLarge = predictPtr[0] + (predictPtr[0]>0);
511
            continue;
512
        }
513
#endif
514
515
207M
        if (!extDict || (matchIndex+matchLength >= dictLimit)) {
516
207M
            assert(matchIndex+matchLength >= dictLimit);   /* might be wrong if actually extDict */
517
207M
            match = base + matchIndex;
518
207M
            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
519
207M
        } else {
520
0
            match = dictBase + matchIndex;
521
0
            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
522
0
            if (matchIndex+matchLength >= dictLimit)
523
0
                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
524
0
        }
525
526
207M
        if (matchLength > bestLength) {
527
37.5M
            bestLength = matchLength;
528
37.5M
            if (matchLength > matchEndIdx - matchIndex)
529
405k
                matchEndIdx = matchIndex + (U32)matchLength;
530
37.5M
        }
531
532
207M
        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
533
33.6k
            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
534
33.6k
        }
535
536
207M
        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
537
            /* match is smaller than current */
538
62.1M
            *smallerPtr = matchIndex;             /* update smaller idx */
539
62.1M
            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
540
62.1M
            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
541
62.1M
            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
542
62.1M
            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
543
145M
        } else {
544
            /* match is larger than current */
545
145M
            *largerPtr = matchIndex;
546
145M
            commonLengthLarger = matchLength;
547
145M
            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
548
145M
            largerPtr = nextPtr;
549
145M
            matchIndex = nextPtr[0];
550
145M
    }   }
551
552
30.2M
    *smallerPtr = *largerPtr = 0;
553
30.2M
    {   U32 positions = 0;
554
30.2M
        if (bestLength > 384) positions = MIN(192, (U32)(bestLength - 384));   /* speed optimization */
555
30.2M
        assert(matchEndIdx > curr + 8);
556
30.2M
        return MAX(positions, matchEndIdx - (curr + 8));
557
30.2M
    }
558
30.2M
}
559
560
FORCE_INLINE_TEMPLATE
561
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
562
void ZSTD_updateTree_internal(
563
                ZSTD_MatchState_t* ms,
564
                const BYTE* const ip, const BYTE* const iend,
565
                const U32 mls, const ZSTD_dictMode_e dictMode)
566
143M
{
567
143M
    const BYTE* const base = ms->window.base;
568
143M
    U32 const target = (U32)(ip - base);
569
143M
    U32 idx = ms->nextToUpdate;
570
143M
    DEBUGLOG(7, "ZSTD_updateTree_internal, from %u to %u  (dictMode:%u)",
571
143M
                idx, target, dictMode);
572
573
173M
    while(idx < target) {
574
30.2M
        U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, target, mls, dictMode == ZSTD_extDict);
575
30.2M
        assert(idx < (U32)(idx + forward));
576
30.2M
        idx += forward;
577
30.2M
    }
578
143M
    assert((size_t)(ip - base) <= (size_t)(U32)(-1));
579
143M
    assert((size_t)(iend - base) <= (size_t)(U32)(-1));
580
143M
    ms->nextToUpdate = target;
581
143M
}
582
583
0
void ZSTD_updateTree(ZSTD_MatchState_t* ms, const BYTE* ip, const BYTE* iend) {
584
0
    ZSTD_updateTree_internal(ms, ip, iend, ms->cParams.minMatch, ZSTD_noDict);
585
0
}
586
587
FORCE_INLINE_TEMPLATE
588
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
589
U32
590
ZSTD_insertBtAndGetAllMatches (
591
                ZSTD_match_t* matches,  /* store result (found matches) in this table (presumed large enough) */
592
                ZSTD_MatchState_t* ms,
593
                U32* nextToUpdate3,
594
                const BYTE* const ip, const BYTE* const iLimit,
595
                const ZSTD_dictMode_e dictMode,
596
                const U32 rep[ZSTD_REP_NUM],
597
                const U32 ll0,  /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
598
                const U32 lengthToBeat,
599
                const U32 mls /* template */)
600
143M
{
601
143M
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
602
143M
    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
603
143M
    const BYTE* const base = ms->window.base;
604
143M
    U32 const curr = (U32)(ip-base);
605
143M
    U32 const hashLog = cParams->hashLog;
606
143M
    U32 const minMatch = (mls==3) ? 3 : 4;
607
143M
    U32* const hashTable = ms->hashTable;
608
143M
    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
609
143M
    U32 matchIndex  = hashTable[h];
610
143M
    U32* const bt   = ms->chainTable;
611
143M
    U32 const btLog = cParams->chainLog - 1;
612
143M
    U32 const btMask= (1U << btLog) - 1;
613
143M
    size_t commonLengthSmaller=0, commonLengthLarger=0;
614
143M
    const BYTE* const dictBase = ms->window.dictBase;
615
143M
    U32 const dictLimit = ms->window.dictLimit;
616
143M
    const BYTE* const dictEnd = dictBase + dictLimit;
617
143M
    const BYTE* const prefixStart = base + dictLimit;
618
143M
    U32 const btLow = (btMask >= curr) ? 0 : curr - btMask;
619
143M
    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
620
143M
    U32 const matchLow = windowLow ? windowLow : 1;
621
143M
    U32* smallerPtr = bt + 2*(curr&btMask);
622
143M
    U32* largerPtr  = bt + 2*(curr&btMask) + 1;
623
143M
    U32 matchEndIdx = curr+8+1;   /* farthest referenced position of any match => detects repetitive patterns */
624
143M
    U32 dummy32;   /* to be nullified at the end */
625
143M
    U32 mnum = 0;
626
143M
    U32 nbCompares = 1U << cParams->searchLog;
627
628
143M
    const ZSTD_MatchState_t* dms    = dictMode == ZSTD_dictMatchState ? ms->dictMatchState : NULL;
629
143M
    const ZSTD_compressionParameters* const dmsCParams =
630
143M
                                      dictMode == ZSTD_dictMatchState ? &dms->cParams : NULL;
631
143M
    const BYTE* const dmsBase       = dictMode == ZSTD_dictMatchState ? dms->window.base : NULL;
632
143M
    const BYTE* const dmsEnd        = dictMode == ZSTD_dictMatchState ? dms->window.nextSrc : NULL;
633
143M
    U32         const dmsHighLimit  = dictMode == ZSTD_dictMatchState ? (U32)(dmsEnd - dmsBase) : 0;
634
143M
    U32         const dmsLowLimit   = dictMode == ZSTD_dictMatchState ? dms->window.lowLimit : 0;
635
143M
    U32         const dmsIndexDelta = dictMode == ZSTD_dictMatchState ? windowLow - dmsHighLimit : 0;
636
143M
    U32         const dmsHashLog    = dictMode == ZSTD_dictMatchState ? dmsCParams->hashLog : hashLog;
637
143M
    U32         const dmsBtLog      = dictMode == ZSTD_dictMatchState ? dmsCParams->chainLog - 1 : btLog;
638
143M
    U32         const dmsBtMask     = dictMode == ZSTD_dictMatchState ? (1U << dmsBtLog) - 1 : 0;
639
143M
    U32         const dmsBtLow      = dictMode == ZSTD_dictMatchState && dmsBtMask < dmsHighLimit - dmsLowLimit ? dmsHighLimit - dmsBtMask : dmsLowLimit;
640
641
143M
    size_t bestLength = lengthToBeat-1;
642
143M
    DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", curr);
643
644
    /* check repCode */
645
143M
    assert(ll0 <= 1);   /* necessarily 1 or 0 */
646
143M
    {   U32 const lastR = ZSTD_REP_NUM + ll0;
647
143M
        U32 repCode;
648
574M
        for (repCode = ll0; repCode < lastR; repCode++) {
649
430M
            U32 const repOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
650
430M
            U32 const repIndex = curr - repOffset;
651
430M
            U32 repLen = 0;
652
430M
            assert(curr >= dictLimit);
653
430M
            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < curr-dictLimit) {  /* equivalent to `curr > repIndex >= dictLimit` */
654
                /* We must validate the repcode offset because when we're using a dictionary the
655
                 * valid offset range shrinks when the dictionary goes out of bounds.
656
                 */
657
428M
                if ((repIndex >= windowLow) & (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repOffset, minMatch))) {
658
53.5M
                    repLen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repOffset, iLimit) + minMatch;
659
53.5M
                }
660
428M
            } else {  /* repIndex < dictLimit || repIndex >= curr */
661
2.32M
                const BYTE* const repMatch = dictMode == ZSTD_dictMatchState ?
662
0
                                             dmsBase + repIndex - dmsIndexDelta :
663
2.32M
                                             dictBase + repIndex;
664
2.32M
                assert(curr >= windowLow);
665
2.32M
                if ( dictMode == ZSTD_extDict
666
0
                  && ( ((repOffset-1) /*intentional overflow*/ < curr - windowLow)  /* equivalent to `curr > repIndex >= windowLow` */
667
0
                     & (ZSTD_index_overlap_check(dictLimit, repIndex)) )
668
0
                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
669
0
                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dictEnd, prefixStart) + minMatch;
670
0
                }
671
2.32M
                if (dictMode == ZSTD_dictMatchState
672
0
                  && ( ((repOffset-1) /*intentional overflow*/ < curr - (dmsLowLimit + dmsIndexDelta))  /* equivalent to `curr > repIndex >= dmsLowLimit` */
673
0
                     & (ZSTD_index_overlap_check(dictLimit, repIndex)) )
674
0
                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
675
0
                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dmsEnd, prefixStart) + minMatch;
676
0
            }   }
677
            /* save longer solution */
678
430M
            if (repLen > bestLength) {
679
41.0M
                DEBUGLOG(8, "found repCode %u (ll0:%u, offset:%u) of length %u",
680
41.0M
                            repCode, ll0, repOffset, repLen);
681
41.0M
                bestLength = repLen;
682
82.1M
                matches[mnum].off = REPCODE_TO_OFFBASE(repCode - ll0 + 1);  /* expect value between 1 and 3 */
683
82.1M
                matches[mnum].len = (U32)repLen;
684
82.1M
                mnum++;
685
82.1M
                if ( (repLen > sufficient_len)
686
41.0M
                   | (ip+repLen == iLimit) ) {  /* best possible */
687
118k
                    return mnum;
688
118k
    }   }   }   }
689
690
    /* HC3 match finder */
691
143M
    if ((mls == 3) /*static*/ && (bestLength < mls)) {
692
72.8M
        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, nextToUpdate3, ip);
693
72.8M
        if ((matchIndex3 >= matchLow)
694
72.8M
          & (curr - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
695
39.7M
            size_t mlen;
696
39.7M
            if ((dictMode == ZSTD_noDict) /*static*/ || (dictMode == ZSTD_dictMatchState) /*static*/ || (matchIndex3 >= dictLimit)) {
697
39.7M
                const BYTE* const match = base + matchIndex3;
698
39.7M
                mlen = ZSTD_count(ip, match, iLimit);
699
39.7M
            } else {
700
0
                const BYTE* const match = dictBase + matchIndex3;
701
0
                mlen = ZSTD_count_2segments(ip, match, iLimit, dictEnd, prefixStart);
702
0
            }
703
704
            /* save best solution */
705
39.7M
            if (mlen >= mls /* == 3 > bestLength */) {
706
28.2M
                DEBUGLOG(8, "found small match with hlog3, of length %u",
707
28.2M
                            (U32)mlen);
708
28.2M
                bestLength = mlen;
709
28.2M
                assert(curr > matchIndex3);
710
28.2M
                assert(mnum==0);  /* no prior solution */
711
28.2M
                matches[0].off = OFFSET_TO_OFFBASE(curr - matchIndex3);
712
28.2M
                matches[0].len = (U32)mlen;
713
28.2M
                mnum = 1;
714
28.2M
                if ( (mlen > sufficient_len) |
715
28.2M
                     (ip+mlen == iLimit) ) {  /* best possible length */
716
10.5k
                    ms->nextToUpdate = curr+1;  /* skip insertion */
717
10.5k
                    return 1;
718
10.5k
        }   }   }
719
        /* no dictMatchState lookup: dicts don't have a populated HC3 table */
720
72.8M
    }  /* if (mls == 3) */
721
722
143M
    hashTable[h] = curr;   /* Update Hash Table */
723
724
580M
    for (; nbCompares && (matchIndex >= matchLow); --nbCompares) {
725
437M
        U32* const nextPtr = bt + 2*(matchIndex & btMask);
726
437M
        const BYTE* match;
727
437M
        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
728
437M
        assert(curr > matchIndex);
729
730
437M
        if ((dictMode == ZSTD_noDict) || (dictMode == ZSTD_dictMatchState) || (matchIndex+matchLength >= dictLimit)) {
731
437M
            assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
732
437M
            match = base + matchIndex;
733
437M
            if (matchIndex >= dictLimit) assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
734
437M
            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iLimit);
735
437M
        } else {
736
0
            match = dictBase + matchIndex;
737
0
            assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
738
0
            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
739
0
            if (matchIndex+matchLength >= dictLimit)
740
0
                match = base + matchIndex;   /* prepare for match[matchLength] read */
741
0
        }
742
743
437M
        if (matchLength > bestLength) {
744
33.4M
            DEBUGLOG(8, "found match of length %u at distance %u (offBase=%u)",
745
33.4M
                    (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
746
33.4M
            assert(matchEndIdx > matchIndex);
747
33.4M
            if (matchLength > matchEndIdx - matchIndex)
748
126k
                matchEndIdx = matchIndex + (U32)matchLength;
749
33.4M
            bestLength = matchLength;
750
33.4M
            matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
751
33.4M
            matches[mnum].len = (U32)matchLength;
752
33.4M
            mnum++;
753
33.4M
            if ( (matchLength > ZSTD_OPT_NUM)
754
33.4M
               | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
755
6.66k
                if (dictMode == ZSTD_dictMatchState) nbCompares = 0; /* break should also skip searching dms */
756
6.66k
                break; /* drop, to preserve bt consistency (miss a little bit of compression) */
757
6.66k
        }   }
758
759
437M
        if (match[matchLength] < ip[matchLength]) {
760
            /* match smaller than current */
761
103M
            *smallerPtr = matchIndex;             /* update smaller idx */
762
103M
            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
763
103M
            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
764
103M
            smallerPtr = nextPtr+1;               /* new candidate => larger than match, which was smaller than current */
765
103M
            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous, closer to current */
766
334M
        } else {
767
334M
            *largerPtr = matchIndex;
768
334M
            commonLengthLarger = matchLength;
769
334M
            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
770
334M
            largerPtr = nextPtr;
771
334M
            matchIndex = nextPtr[0];
772
334M
    }   }
773
774
143M
    *smallerPtr = *largerPtr = 0;
775
776
143M
    assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
777
143M
    if (dictMode == ZSTD_dictMatchState && nbCompares) {
778
0
        size_t const dmsH = ZSTD_hashPtr(ip, dmsHashLog, mls);
779
0
        U32 dictMatchIndex = dms->hashTable[dmsH];
780
0
        const U32* const dmsBt = dms->chainTable;
781
0
        commonLengthSmaller = commonLengthLarger = 0;
782
0
        for (; nbCompares && (dictMatchIndex > dmsLowLimit); --nbCompares) {
783
0
            const U32* const nextPtr = dmsBt + 2*(dictMatchIndex & dmsBtMask);
784
0
            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
785
0
            const BYTE* match = dmsBase + dictMatchIndex;
786
0
            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dmsEnd, prefixStart);
787
0
            if (dictMatchIndex+matchLength >= dmsHighLimit)
788
0
                match = base + dictMatchIndex + dmsIndexDelta;   /* to prepare for next usage of match[matchLength] */
789
790
0
            if (matchLength > bestLength) {
791
0
                matchIndex = dictMatchIndex + dmsIndexDelta;
792
0
                DEBUGLOG(8, "found dms match of length %u at distance %u (offBase=%u)",
793
0
                        (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
794
0
                if (matchLength > matchEndIdx - matchIndex)
795
0
                    matchEndIdx = matchIndex + (U32)matchLength;
796
0
                bestLength = matchLength;
797
0
                matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
798
0
                matches[mnum].len = (U32)matchLength;
799
0
                mnum++;
800
0
                if ( (matchLength > ZSTD_OPT_NUM)
801
0
                   | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
802
0
                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
803
0
            }   }
804
805
0
            if (dictMatchIndex <= dmsBtLow) { break; }   /* beyond tree size, stop the search */
806
0
            if (match[matchLength] < ip[matchLength]) {
807
0
                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
808
0
                dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
809
0
            } else {
810
                /* match is larger than current */
811
0
                commonLengthLarger = matchLength;
812
0
                dictMatchIndex = nextPtr[0];
813
0
    }   }   }  /* if (dictMode == ZSTD_dictMatchState) */
814
815
143M
    assert(matchEndIdx > curr+8);
816
143M
    ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
817
143M
    return mnum;
818
143M
}
819
820
typedef U32 (*ZSTD_getAllMatchesFn)(
821
    ZSTD_match_t*,
822
    ZSTD_MatchState_t*,
823
    U32*,
824
    const BYTE*,
825
    const BYTE*,
826
    const U32 rep[ZSTD_REP_NUM],
827
    U32 const ll0,
828
    U32 const lengthToBeat);
829
830
FORCE_INLINE_TEMPLATE
831
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
832
U32 ZSTD_btGetAllMatches_internal(
833
        ZSTD_match_t* matches,
834
        ZSTD_MatchState_t* ms,
835
        U32* nextToUpdate3,
836
        const BYTE* ip,
837
        const BYTE* const iHighLimit,
838
        const U32 rep[ZSTD_REP_NUM],
839
        U32 const ll0,
840
        U32 const lengthToBeat,
841
        const ZSTD_dictMode_e dictMode,
842
        const U32 mls)
843
146M
{
844
146M
    assert(BOUNDED(3, ms->cParams.minMatch, 6) == mls);
845
146M
    DEBUGLOG(8, "ZSTD_BtGetAllMatches(dictMode=%d, mls=%u)", (int)dictMode, mls);
846
146M
    if (ip < ms->window.base + ms->nextToUpdate)
847
3.07M
        return 0;   /* skipped area */
848
143M
    ZSTD_updateTree_internal(ms, ip, iHighLimit, mls, dictMode);
849
143M
    return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, mls);
850
146M
}
851
852
148k
#define ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, mls) ZSTD_btGetAllMatches_##dictMode##_##mls
853
854
#define GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, mls)            \
855
    static U32 ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, mls)(      \
856
            ZSTD_match_t* matches,                             \
857
            ZSTD_MatchState_t* ms,                             \
858
            U32* nextToUpdate3,                                \
859
            const BYTE* ip,                                    \
860
            const BYTE* const iHighLimit,                      \
861
            const U32 rep[ZSTD_REP_NUM],                       \
862
            U32 const ll0,                                     \
863
            U32 const lengthToBeat)                            \
864
146M
    {                                                          \
865
146M
        return ZSTD_btGetAllMatches_internal(                  \
866
146M
                matches, ms, nextToUpdate3, ip, iHighLimit,    \
867
146M
                rep, ll0, lengthToBeat, ZSTD_##dictMode, mls); \
868
146M
    }
zstd_opt.c:ZSTD_btGetAllMatches_noDict_3
Line
Count
Source
864
108M
    {                                                          \
865
108M
        return ZSTD_btGetAllMatches_internal(                  \
866
108M
                matches, ms, nextToUpdate3, ip, iHighLimit,    \
867
108M
                rep, ll0, lengthToBeat, ZSTD_##dictMode, mls); \
868
108M
    }
zstd_opt.c:ZSTD_btGetAllMatches_noDict_4
Line
Count
Source
864
19.4M
    {                                                          \
865
19.4M
        return ZSTD_btGetAllMatches_internal(                  \
866
19.4M
                matches, ms, nextToUpdate3, ip, iHighLimit,    \
867
19.4M
                rep, ll0, lengthToBeat, ZSTD_##dictMode, mls); \
868
19.4M
    }
zstd_opt.c:ZSTD_btGetAllMatches_noDict_5
Line
Count
Source
864
18.8M
    {                                                          \
865
18.8M
        return ZSTD_btGetAllMatches_internal(                  \
866
18.8M
                matches, ms, nextToUpdate3, ip, iHighLimit,    \
867
18.8M
                rep, ll0, lengthToBeat, ZSTD_##dictMode, mls); \
868
18.8M
    }
Unexecuted instantiation: zstd_opt.c:ZSTD_btGetAllMatches_noDict_6
Unexecuted instantiation: zstd_opt.c:ZSTD_btGetAllMatches_extDict_3
Unexecuted instantiation: zstd_opt.c:ZSTD_btGetAllMatches_extDict_4
Unexecuted instantiation: zstd_opt.c:ZSTD_btGetAllMatches_extDict_5
Unexecuted instantiation: zstd_opt.c:ZSTD_btGetAllMatches_extDict_6
Unexecuted instantiation: zstd_opt.c:ZSTD_btGetAllMatches_dictMatchState_3
Unexecuted instantiation: zstd_opt.c:ZSTD_btGetAllMatches_dictMatchState_4
Unexecuted instantiation: zstd_opt.c:ZSTD_btGetAllMatches_dictMatchState_5
Unexecuted instantiation: zstd_opt.c:ZSTD_btGetAllMatches_dictMatchState_6
869
870
#define GEN_ZSTD_BT_GET_ALL_MATCHES(dictMode)  \
871
    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 3)  \
872
    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 4)  \
873
    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 5)  \
874
    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 6)
875
876
GEN_ZSTD_BT_GET_ALL_MATCHES(noDict)
877
GEN_ZSTD_BT_GET_ALL_MATCHES(extDict)
878
GEN_ZSTD_BT_GET_ALL_MATCHES(dictMatchState)
879
880
#define ZSTD_BT_GET_ALL_MATCHES_ARRAY(dictMode)  \
881
37.1k
    {                                            \
882
37.1k
        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 3), \
883
37.1k
        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 4), \
884
37.1k
        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 5), \
885
37.1k
        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 6)  \
886
37.1k
    }
887
888
static ZSTD_getAllMatchesFn
889
ZSTD_selectBtGetAllMatches(ZSTD_MatchState_t const* ms, ZSTD_dictMode_e const dictMode)
890
12.3k
{
891
12.3k
    ZSTD_getAllMatchesFn const getAllMatchesFns[3][4] = {
892
12.3k
        ZSTD_BT_GET_ALL_MATCHES_ARRAY(noDict),
893
12.3k
        ZSTD_BT_GET_ALL_MATCHES_ARRAY(extDict),
894
12.3k
        ZSTD_BT_GET_ALL_MATCHES_ARRAY(dictMatchState)
895
12.3k
    };
896
12.3k
    U32 const mls = BOUNDED(3, ms->cParams.minMatch, 6);
897
12.3k
    assert((U32)dictMode < 3);
898
12.3k
    assert(mls - 3 < 4);
899
12.3k
    return getAllMatchesFns[(int)dictMode][mls - 3];
900
12.3k
}
901
902
/*************************
903
*  LDM helper functions  *
904
*************************/
905
906
/* Struct containing info needed to make decision about ldm inclusion */
907
typedef struct {
908
    RawSeqStore_t seqStore;   /* External match candidates store for this block */
909
    U32 startPosInBlock;      /* Start position of the current match candidate */
910
    U32 endPosInBlock;        /* End position of the current match candidate */
911
    U32 offset;               /* Offset of the match candidate */
912
} ZSTD_optLdm_t;
913
914
/* ZSTD_optLdm_skipRawSeqStoreBytes():
915
 * Moves forward in @rawSeqStore by @nbBytes,
916
 * which will update the fields 'pos' and 'posInSequence'.
917
 */
918
static void ZSTD_optLdm_skipRawSeqStoreBytes(RawSeqStore_t* rawSeqStore, size_t nbBytes)
919
0
{
920
0
    U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
921
0
    while (currPos && rawSeqStore->pos < rawSeqStore->size) {
922
0
        rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
923
0
        if (currPos >= currSeq.litLength + currSeq.matchLength) {
924
0
            currPos -= currSeq.litLength + currSeq.matchLength;
925
0
            rawSeqStore->pos++;
926
0
        } else {
927
0
            rawSeqStore->posInSequence = currPos;
928
0
            break;
929
0
        }
930
0
    }
931
0
    if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
932
0
        rawSeqStore->posInSequence = 0;
933
0
    }
934
0
}
935
936
/* ZSTD_opt_getNextMatchAndUpdateSeqStore():
937
 * Calculates the beginning and end of the next match in the current block.
938
 * Updates 'pos' and 'posInSequence' of the ldmSeqStore.
939
 */
940
static void
941
ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
942
                                       U32 blockBytesRemaining)
943
12.3k
{
944
12.3k
    rawSeq currSeq;
945
12.3k
    U32 currBlockEndPos;
946
12.3k
    U32 literalsBytesRemaining;
947
12.3k
    U32 matchBytesRemaining;
948
949
    /* Setting match end position to MAX to ensure we never use an LDM during this block */
950
12.3k
    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
951
12.3k
        optLdm->startPosInBlock = UINT_MAX;
952
12.3k
        optLdm->endPosInBlock = UINT_MAX;
953
12.3k
        return;
954
12.3k
    }
955
    /* Calculate appropriate bytes left in matchLength and litLength
956
     * after adjusting based on ldmSeqStore->posInSequence */
957
0
    currSeq = optLdm->seqStore.seq[optLdm->seqStore.pos];
958
0
    assert(optLdm->seqStore.posInSequence <= currSeq.litLength + currSeq.matchLength);
959
0
    currBlockEndPos = currPosInBlock + blockBytesRemaining;
960
0
    literalsBytesRemaining = (optLdm->seqStore.posInSequence < currSeq.litLength) ?
961
0
            currSeq.litLength - (U32)optLdm->seqStore.posInSequence :
962
0
            0;
963
0
    matchBytesRemaining = (literalsBytesRemaining == 0) ?
964
0
            currSeq.matchLength - ((U32)optLdm->seqStore.posInSequence - currSeq.litLength) :
965
0
            currSeq.matchLength;
966
967
    /* If there are more literal bytes than bytes remaining in block, no ldm is possible */
968
0
    if (literalsBytesRemaining >= blockBytesRemaining) {
969
0
        optLdm->startPosInBlock = UINT_MAX;
970
0
        optLdm->endPosInBlock = UINT_MAX;
971
0
        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, blockBytesRemaining);
972
0
        return;
973
0
    }
974
975
    /* Matches may be < minMatch by this process. In that case, we will reject them
976
       when we are deciding whether or not to add the ldm */
977
0
    optLdm->startPosInBlock = currPosInBlock + literalsBytesRemaining;
978
0
    optLdm->endPosInBlock = optLdm->startPosInBlock + matchBytesRemaining;
979
0
    optLdm->offset = currSeq.offset;
980
981
0
    if (optLdm->endPosInBlock > currBlockEndPos) {
982
        /* Match ends after the block ends, we can't use the whole match */
983
0
        optLdm->endPosInBlock = currBlockEndPos;
984
0
        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, currBlockEndPos - currPosInBlock);
985
0
    } else {
986
        /* Consume nb of bytes equal to size of sequence left */
987
0
        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, literalsBytesRemaining + matchBytesRemaining);
988
0
    }
989
0
}
990
991
/* ZSTD_optLdm_maybeAddMatch():
992
 * Adds a match if it's long enough,
993
 * based on it's 'matchStartPosInBlock' and 'matchEndPosInBlock',
994
 * into 'matches'. Maintains the correct ordering of 'matches'.
995
 */
996
static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
997
                                      const ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
998
                                      U32 minMatch)
999
0
{
1000
0
    U32 const posDiff = currPosInBlock - optLdm->startPosInBlock;
1001
    /* Note: ZSTD_match_t actually contains offBase and matchLength (before subtracting MINMATCH) */
1002
0
    U32 const candidateMatchLength = optLdm->endPosInBlock - optLdm->startPosInBlock - posDiff;
1003
1004
    /* Ensure that current block position is not outside of the match */
1005
0
    if (currPosInBlock < optLdm->startPosInBlock
1006
0
      || currPosInBlock >= optLdm->endPosInBlock
1007
0
      || candidateMatchLength < minMatch) {
1008
0
        return;
1009
0
    }
1010
1011
0
    if (*nbMatches == 0 || ((candidateMatchLength > matches[*nbMatches-1].len) && *nbMatches < ZSTD_OPT_NUM)) {
1012
0
        U32 const candidateOffBase = OFFSET_TO_OFFBASE(optLdm->offset);
1013
0
        DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offBase: %u matchLength %u) at block position=%u",
1014
0
                 candidateOffBase, candidateMatchLength, currPosInBlock);
1015
0
        matches[*nbMatches].len = candidateMatchLength;
1016
0
        matches[*nbMatches].off = candidateOffBase;
1017
0
        (*nbMatches)++;
1018
0
    }
1019
0
}
1020
1021
/* ZSTD_optLdm_processMatchCandidate():
1022
 * Wrapper function to update ldm seq store and call ldm functions as necessary.
1023
 */
1024
static void
1025
ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm,
1026
                                  ZSTD_match_t* matches, U32* nbMatches,
1027
                                  U32 currPosInBlock, U32 remainingBytes,
1028
                                  U32 minMatch)
1029
146M
{
1030
146M
    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
1031
146M
        return;
1032
146M
    }
1033
1034
0
    if (currPosInBlock >= optLdm->endPosInBlock) {
1035
0
        if (currPosInBlock > optLdm->endPosInBlock) {
1036
            /* The position at which ZSTD_optLdm_processMatchCandidate() is called is not necessarily
1037
             * at the end of a match from the ldm seq store, and will often be some bytes
1038
             * over beyond matchEndPosInBlock. As such, we need to correct for these "overshoots"
1039
             */
1040
0
            U32 const posOvershoot = currPosInBlock - optLdm->endPosInBlock;
1041
0
            ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, posOvershoot);
1042
0
        }
1043
0
        ZSTD_opt_getNextMatchAndUpdateSeqStore(optLdm, currPosInBlock, remainingBytes);
1044
0
    }
1045
0
    ZSTD_optLdm_maybeAddMatch(matches, nbMatches, optLdm, currPosInBlock, minMatch);
1046
0
}
1047
1048
1049
/*-*******************************
1050
*  Optimal parser
1051
*********************************/
1052
1053
#if 0 /* debug */
1054
1055
static void
1056
listStats(const U32* table, int lastEltID)
1057
{
1058
    int const nbElts = lastEltID + 1;
1059
    int enb;
1060
    for (enb=0; enb < nbElts; enb++) {
1061
        (void)table;
1062
        /* RAWLOG(2, "%3i:%3i,  ", enb, table[enb]); */
1063
        RAWLOG(2, "%4i,", table[enb]);
1064
    }
1065
    RAWLOG(2, " \n");
1066
}
1067
1068
#endif
1069
1070
117M
#define LIT_PRICE(_p) (int)ZSTD_rawLiteralsCost(_p, 1, optStatePtr, optLevel)
1071
370M
#define LL_PRICE(_l) (int)ZSTD_litLengthPrice(_l, optStatePtr, optLevel)
1072
128M
#define LL_INCPRICE(_l) (LL_PRICE(_l) - LL_PRICE(_l-1))
1073
1074
FORCE_INLINE_TEMPLATE
1075
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
1076
size_t
1077
ZSTD_compressBlock_opt_generic(ZSTD_MatchState_t* ms,
1078
                               SeqStore_t* seqStore,
1079
                               U32 rep[ZSTD_REP_NUM],
1080
                         const void* src, size_t srcSize,
1081
                         const int optLevel,
1082
                         const ZSTD_dictMode_e dictMode)
1083
12.3k
{
1084
12.3k
    optState_t* const optStatePtr = &ms->opt;
1085
12.3k
    const BYTE* const istart = (const BYTE*)src;
1086
12.3k
    const BYTE* ip = istart;
1087
12.3k
    const BYTE* anchor = istart;
1088
12.3k
    const BYTE* const iend = istart + srcSize;
1089
12.3k
    const BYTE* const ilimit = iend - 8;
1090
12.3k
    const BYTE* const base = ms->window.base;
1091
12.3k
    const BYTE* const prefixStart = base + ms->window.dictLimit;
1092
12.3k
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
1093
1094
12.3k
    ZSTD_getAllMatchesFn getAllMatches = ZSTD_selectBtGetAllMatches(ms, dictMode);
1095
1096
12.3k
    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
1097
12.3k
    U32 const minMatch = (cParams->minMatch == 3) ? 3 : 4;
1098
12.3k
    U32 nextToUpdate3 = ms->nextToUpdate;
1099
1100
12.3k
    ZSTD_optimal_t* const opt = optStatePtr->priceTable;
1101
12.3k
    ZSTD_match_t* const matches = optStatePtr->matchTable;
1102
12.3k
    ZSTD_optimal_t lastStretch;
1103
12.3k
    ZSTD_optLdm_t optLdm;
1104
1105
12.3k
    ZSTD_memset(&lastStretch, 0, sizeof(ZSTD_optimal_t));
1106
1107
12.3k
    optLdm.seqStore = ms->ldmSeqStore ? *ms->ldmSeqStore : kNullRawSeqStore;
1108
12.3k
    optLdm.endPosInBlock = optLdm.startPosInBlock = optLdm.offset = 0;
1109
12.3k
    ZSTD_opt_getNextMatchAndUpdateSeqStore(&optLdm, (U32)(ip-istart), (U32)(iend-ip));
1110
1111
    /* init */
1112
12.3k
    DEBUGLOG(5, "ZSTD_compressBlock_opt_generic: current=%u, prefix=%u, nextToUpdate=%u",
1113
12.3k
                (U32)(ip - base), ms->window.dictLimit, ms->nextToUpdate);
1114
12.3k
    assert(optLevel <= 2);
1115
12.3k
    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize, optLevel);
1116
12.3k
    ip += (ip==prefixStart);
1117
1118
    /* Match Loop */
1119
59.0M
    while (ip < ilimit) {
1120
59.0M
        U32 cur, last_pos = 0;
1121
1122
        /* find first match */
1123
59.0M
        {   U32 const litlen = (U32)(ip - anchor);
1124
59.0M
            U32 const ll0 = !litlen;
1125
59.0M
            U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, ip, iend, rep, ll0, minMatch);
1126
59.0M
            ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
1127
59.0M
                                              (U32)(ip-istart), (U32)(iend-ip),
1128
59.0M
                                              minMatch);
1129
59.0M
            if (!nbMatches) {
1130
53.3M
                DEBUGLOG(8, "no match found at cPos %u", (unsigned)(ip-istart));
1131
53.3M
                ip++;
1132
53.3M
                continue;
1133
53.3M
            }
1134
1135
            /* Match found: let's store this solution, and eventually find more candidates.
1136
             * During this forward pass, @opt is used to store stretches,
1137
             * defined as "a match followed by N literals".
1138
             * Note how this is different from a Sequence, which is "N literals followed by a match".
1139
             * Storing stretches allows us to store different match predecessors
1140
             * for each literal position part of a literals run. */
1141
1142
            /* initialize opt[0] */
1143
5.68M
            opt[0].mlen = 0;  /* there are only literals so far */
1144
5.68M
            opt[0].litlen = litlen;
1145
            /* No need to include the actual price of the literals before the first match
1146
             * because it is static for the duration of the forward pass, and is included
1147
             * in every subsequent price. But, we include the literal length because
1148
             * the cost variation of litlen depends on the value of litlen.
1149
             */
1150
5.68M
            opt[0].price = LL_PRICE(litlen);
1151
5.68M
            ZSTD_STATIC_ASSERT(sizeof(opt[0].rep[0]) == sizeof(rep[0]));
1152
5.68M
            ZSTD_memcpy(&opt[0].rep, rep, sizeof(opt[0].rep));
1153
1154
            /* large match -> immediate encoding */
1155
5.68M
            {   U32 const maxML = matches[nbMatches-1].len;
1156
5.68M
                U32 const maxOffBase = matches[nbMatches-1].off;
1157
5.68M
                DEBUGLOG(6, "found %u matches of maxLength=%u and maxOffBase=%u at cPos=%u => start new series",
1158
5.68M
                            nbMatches, maxML, maxOffBase, (U32)(ip-prefixStart));
1159
1160
5.68M
                if (maxML > sufficient_len) {
1161
184k
                    lastStretch.litlen = 0;
1162
184k
                    lastStretch.mlen = maxML;
1163
184k
                    lastStretch.off = maxOffBase;
1164
184k
                    DEBUGLOG(6, "large match (%u>%u) => immediate encoding",
1165
184k
                                maxML, sufficient_len);
1166
184k
                    cur = 0;
1167
184k
                    last_pos = maxML;
1168
184k
                    goto _shortestPath;
1169
184k
            }   }
1170
1171
            /* set prices for first matches starting position == 0 */
1172
5.68M
            assert(opt[0].price >= 0);
1173
5.50M
            {   U32 pos;
1174
5.50M
                U32 matchNb;
1175
18.2M
                for (pos = 1; pos < minMatch; pos++) {
1176
12.7M
                    opt[pos].price = ZSTD_MAX_PRICE;
1177
12.7M
                    opt[pos].mlen = 0;
1178
12.7M
                    opt[pos].litlen = litlen + pos;
1179
12.7M
                }
1180
12.4M
                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
1181
6.93M
                    U32 const offBase = matches[matchNb].off;
1182
6.93M
                    U32 const end = matches[matchNb].len;
1183
26.4M
                    for ( ; pos <= end ; pos++ ) {
1184
19.5M
                        int const matchPrice = (int)ZSTD_getMatchPrice(offBase, pos, optStatePtr, optLevel);
1185
19.5M
                        int const sequencePrice = opt[0].price + matchPrice;
1186
19.5M
                        DEBUGLOG(7, "rPos:%u => set initial price : %.2f",
1187
19.5M
                                    pos, ZSTD_fCost(sequencePrice));
1188
19.5M
                        opt[pos].mlen = pos;
1189
19.5M
                        opt[pos].off = offBase;
1190
19.5M
                        opt[pos].litlen = 0; /* end of match */
1191
19.5M
                        opt[pos].price = sequencePrice + LL_PRICE(0);
1192
19.5M
                    }
1193
6.93M
                }
1194
5.50M
                last_pos = pos-1;
1195
5.50M
                opt[pos].price = ZSTD_MAX_PRICE;
1196
5.50M
            }
1197
5.50M
        }
1198
1199
        /* check further positions */
1200
104M
        for (cur = 1; cur <= last_pos; cur++) {
1201
104M
            const BYTE* const inr = ip + cur;
1202
104M
            assert(cur <= ZSTD_OPT_NUM);
1203
104M
            DEBUGLOG(7, "cPos:%i==rPos:%u", (int)(inr-istart), cur);
1204
1205
            /* Fix current position with one literal if cheaper */
1206
104M
            {   U32 const litlen = opt[cur-1].litlen + 1;
1207
104M
                int const price = opt[cur-1].price
1208
104M
                                + LIT_PRICE(ip+cur-1)
1209
104M
                                + LL_INCPRICE(litlen);
1210
104M
                assert(price < 1000000000); /* overflow check */
1211
104M
                if (price <= opt[cur].price) {
1212
39.5M
                    ZSTD_optimal_t const prevMatch = opt[cur];
1213
39.5M
                    DEBUGLOG(7, "cPos:%i==rPos:%u : better price (%.2f<=%.2f) using literal (ll==%u) (hist:%u,%u,%u)",
1214
39.5M
                                (int)(inr-istart), cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price), litlen,
1215
39.5M
                                opt[cur-1].rep[0], opt[cur-1].rep[1], opt[cur-1].rep[2]);
1216
39.5M
                    opt[cur] = opt[cur-1];
1217
39.5M
                    opt[cur].litlen = litlen;
1218
39.5M
                    opt[cur].price = price;
1219
39.5M
                    if ( (optLevel >= 1) /* additional check only for higher modes */
1220
26.1M
                      && (prevMatch.litlen == 0) /* replace a match */
1221
11.5M
                      && (LL_INCPRICE(1) < 0) /* ll1 is cheaper than ll0 */
1222
6.68M
                      && LIKELY(ip + cur < iend)
1223
39.5M
                    ) {
1224
                        /* check next position, in case it would be cheaper */
1225
6.68M
                        int with1literal = prevMatch.price + LIT_PRICE(ip+cur) + LL_INCPRICE(1);
1226
6.68M
                        int withMoreLiterals = price + LIT_PRICE(ip+cur) + LL_INCPRICE(litlen+1);
1227
6.68M
                        DEBUGLOG(7, "then at next rPos %u : match+1lit %.2f vs %ulits %.2f",
1228
6.68M
                                cur+1, ZSTD_fCost(with1literal), litlen+1, ZSTD_fCost(withMoreLiterals));
1229
6.68M
                        if ( (with1literal < withMoreLiterals)
1230
2.20M
                          && (with1literal < opt[cur+1].price) ) {
1231
                            /* update offset history - before it disappears */
1232
1.02M
                            U32 const prev = cur - prevMatch.mlen;
1233
1.02M
                            Repcodes_t const newReps = ZSTD_newRep(opt[prev].rep, prevMatch.off, opt[prev].litlen==0);
1234
1.02M
                            assert(cur >= prevMatch.mlen);
1235
1.02M
                            DEBUGLOG(7, "==> match+1lit is cheaper (%.2f < %.2f) (hist:%u,%u,%u) !",
1236
1.02M
                                        ZSTD_fCost(with1literal), ZSTD_fCost(withMoreLiterals),
1237
1.02M
                                        newReps.rep[0], newReps.rep[1], newReps.rep[2] );
1238
1.02M
                            opt[cur+1] = prevMatch;  /* mlen & offbase */
1239
1.02M
                            ZSTD_memcpy(opt[cur+1].rep, &newReps, sizeof(Repcodes_t));
1240
1.02M
                            opt[cur+1].litlen = 1;
1241
1.02M
                            opt[cur+1].price = with1literal;
1242
1.02M
                            if (last_pos < cur+1) last_pos = cur+1;
1243
1.02M
                        }
1244
6.68M
                    }
1245
64.4M
                } else {
1246
64.4M
                    DEBUGLOG(7, "cPos:%i==rPos:%u : literal would cost more (%.2f>%.2f)",
1247
64.4M
                                (int)(inr-istart), cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price));
1248
64.4M
                }
1249
104M
            }
1250
1251
            /* Offset history is not updated during match comparison.
1252
             * Do it here, now that the match is selected and confirmed.
1253
             */
1254
104M
            ZSTD_STATIC_ASSERT(sizeof(opt[cur].rep) == sizeof(Repcodes_t));
1255
104M
            assert(cur >= opt[cur].mlen);
1256
104M
            if (opt[cur].litlen == 0) {
1257
                /* just finished a match => alter offset history */
1258
63.4M
                U32 const prev = cur - opt[cur].mlen;
1259
63.4M
                Repcodes_t const newReps = ZSTD_newRep(opt[prev].rep, opt[cur].off, opt[prev].litlen==0);
1260
63.4M
                ZSTD_memcpy(opt[cur].rep, &newReps, sizeof(Repcodes_t));
1261
63.4M
            }
1262
1263
            /* last match must start at a minimum distance of 8 from oend */
1264
104M
            if (inr > ilimit) continue;
1265
1266
104M
            if (cur == last_pos) break;
1267
1268
98.5M
            if ( (optLevel==0) /*static_test*/
1269
26.1M
              && (opt[cur+1].price <= opt[cur].price + (BITCOST_MULTIPLIER/2)) ) {
1270
10.9M
                DEBUGLOG(7, "skip current position : next rPos(%u) price is cheaper", cur+1);
1271
10.9M
                continue;  /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
1272
10.9M
            }
1273
1274
98.5M
            assert(opt[cur].price >= 0);
1275
87.6M
            {   U32 const ll0 = (opt[cur].litlen == 0);
1276
87.6M
                int const previousPrice = opt[cur].price;
1277
87.6M
                int const basePrice = previousPrice + LL_PRICE(0);
1278
87.6M
                U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, inr, iend, opt[cur].rep, ll0, minMatch);
1279
87.6M
                U32 matchNb;
1280
1281
87.6M
                ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
1282
87.6M
                                                  (U32)(inr-istart), (U32)(iend-inr),
1283
87.6M
                                                  minMatch);
1284
1285
87.6M
                if (!nbMatches) {
1286
21.1M
                    DEBUGLOG(7, "rPos:%u : no match found", cur);
1287
21.1M
                    continue;
1288
21.1M
                }
1289
1290
66.5M
                {   U32 const longestML = matches[nbMatches-1].len;
1291
66.5M
                    DEBUGLOG(7, "cPos:%i==rPos:%u, found %u matches, of longest ML=%u",
1292
66.5M
                                (int)(inr-istart), cur, nbMatches, longestML);
1293
1294
66.5M
                    if ( (longestML > sufficient_len)
1295
66.4M
                      || (cur + longestML >= ZSTD_OPT_NUM)
1296
66.4M
                      || (ip + cur + longestML >= iend) ) {
1297
76.2k
                        lastStretch.mlen = longestML;
1298
76.2k
                        lastStretch.off = matches[nbMatches-1].off;
1299
76.2k
                        lastStretch.litlen = 0;
1300
76.2k
                        last_pos = cur + longestML;
1301
76.2k
                        goto _shortestPath;
1302
76.2k
                }   }
1303
1304
                /* set prices using matches found at position == cur */
1305
161M
                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
1306
95.1M
                    U32 const offset = matches[matchNb].off;
1307
95.1M
                    U32 const lastML = matches[matchNb].len;
1308
95.1M
                    U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
1309
95.1M
                    U32 mlen;
1310
1311
95.1M
                    DEBUGLOG(7, "testing match %u => offBase=%4u, mlen=%2u, llen=%2u",
1312
95.1M
                                matchNb, matches[matchNb].off, lastML, opt[cur].litlen);
1313
1314
803M
                    for (mlen = lastML; mlen >= startML; mlen--) {  /* scan downward */
1315
718M
                        U32 const pos = cur + mlen;
1316
718M
                        int const price = basePrice + (int)ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
1317
1318
718M
                        if ((pos > last_pos) || (price < opt[pos].price)) {
1319
109M
                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new better price (%.2f<%.2f)",
1320
109M
                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
1321
182M
                            while (last_pos < pos) {
1322
                                /* fill empty positions, for future comparisons */
1323
72.7M
                                last_pos++;
1324
72.7M
                                opt[last_pos].price = ZSTD_MAX_PRICE;
1325
72.7M
                                opt[last_pos].litlen = !0;  /* just needs to be != 0, to mean "not an end of match" */
1326
72.7M
                            }
1327
109M
                            opt[pos].mlen = mlen;
1328
109M
                            opt[pos].off = offset;
1329
109M
                            opt[pos].litlen = 0;
1330
109M
                            opt[pos].price = price;
1331
609M
                        } else {
1332
609M
                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new price is worse (%.2f>=%.2f)",
1333
609M
                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
1334
609M
                            if (optLevel==0) break;  /* early update abort; gets ~+10% speed for about -0.01 ratio loss */
1335
609M
                        }
1336
718M
            }   }   }
1337
66.4M
            opt[last_pos+1].price = ZSTD_MAX_PRICE;
1338
66.4M
        }  /* for (cur = 1; cur <= last_pos; cur++) */
1339
1340
5.42M
        lastStretch = opt[last_pos];
1341
5.42M
        assert(cur >= lastStretch.mlen);
1342
5.42M
        cur = last_pos - lastStretch.mlen;
1343
1344
5.68M
_shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
1345
5.68M
        assert(opt[0].mlen == 0);
1346
5.68M
        assert(last_pos >= lastStretch.mlen);
1347
5.68M
        assert(cur == last_pos - lastStretch.mlen);
1348
1349
5.68M
        if (lastStretch.mlen==0) {
1350
            /* no solution : all matches have been converted into literals */
1351
676k
            assert(lastStretch.litlen == (ip - anchor) + last_pos);
1352
676k
            ip += last_pos;
1353
676k
            continue;
1354
676k
        }
1355
5.68M
        assert(lastStretch.off > 0);
1356
1357
        /* Update offset history */
1358
5.01M
        if (lastStretch.litlen == 0) {
1359
            /* finishing on a match : update offset history */
1360
3.92M
            Repcodes_t const reps = ZSTD_newRep(opt[cur].rep, lastStretch.off, opt[cur].litlen==0);
1361
3.92M
            ZSTD_memcpy(rep, &reps, sizeof(Repcodes_t));
1362
3.92M
        } else {
1363
1.09M
            ZSTD_memcpy(rep, lastStretch.rep, sizeof(Repcodes_t));
1364
1.09M
            assert(cur >= lastStretch.litlen);
1365
1.09M
            cur -= lastStretch.litlen;
1366
1.09M
        }
1367
1368
        /* Let's write the shortest path solution.
1369
         * It is stored in @opt in reverse order,
1370
         * starting from @storeEnd (==cur+2),
1371
         * effectively partially @opt overwriting.
1372
         * Content is changed too:
1373
         * - So far, @opt stored stretches, aka a match followed by literals
1374
         * - Now, it will store sequences, aka literals followed by a match
1375
         */
1376
5.01M
        {   U32 const storeEnd = cur + 2;
1377
5.01M
            U32 storeStart = storeEnd;
1378
5.01M
            U32 stretchPos = cur;
1379
1380
5.01M
            DEBUGLOG(6, "start reverse traversal (last_pos:%u, cur:%u)",
1381
5.01M
                        last_pos, cur); (void)last_pos;
1382
5.01M
            assert(storeEnd < ZSTD_OPT_SIZE);
1383
5.01M
            DEBUGLOG(6, "last stretch copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
1384
5.01M
                        storeEnd, lastStretch.litlen, lastStretch.mlen, lastStretch.off);
1385
5.01M
            opt[storeEnd] = lastStretch;  /* note: litlen will be fixed */
1386
5.01M
            storeStart = storeEnd;
1387
17.3M
            while (1) {
1388
17.3M
                ZSTD_optimal_t nextStretch = opt[stretchPos];
1389
17.3M
                opt[storeStart].litlen = nextStretch.litlen;
1390
17.3M
                DEBUGLOG(6, "selected sequence (llen=%u,mlen=%u,ofc=%u)",
1391
17.3M
                            opt[storeStart].litlen, opt[storeStart].mlen, opt[storeStart].off);
1392
17.3M
                if (nextStretch.mlen == 0) {
1393
                    /* reaching beginning of segment */
1394
5.01M
                    break;
1395
5.01M
                }
1396
12.3M
                storeStart--;
1397
12.3M
                opt[storeStart] = nextStretch; /* note: litlen will be fixed */
1398
12.3M
                assert(nextStretch.litlen + nextStretch.mlen <= stretchPos);
1399
12.3M
                stretchPos -= nextStretch.litlen + nextStretch.mlen;
1400
12.3M
            }
1401
1402
            /* save sequences */
1403
5.01M
            DEBUGLOG(6, "sending selected sequences into seqStore");
1404
5.01M
            {   U32 storePos;
1405
22.4M
                for (storePos=storeStart; storePos <= storeEnd; storePos++) {
1406
17.3M
                    U32 const llen = opt[storePos].litlen;
1407
17.3M
                    U32 const mlen = opt[storePos].mlen;
1408
17.3M
                    U32 const offBase = opt[storePos].off;
1409
17.3M
                    U32 const advance = llen + mlen;
1410
17.3M
                    DEBUGLOG(6, "considering seq starting at %i, llen=%u, mlen=%u",
1411
17.3M
                                (int)(anchor - istart), (unsigned)llen, (unsigned)mlen);
1412
1413
17.3M
                    if (mlen==0) {  /* only literals => must be last "sequence", actually starting a new stream of sequences */
1414
0
                        assert(storePos == storeEnd);   /* must be last sequence */
1415
0
                        ip = anchor + llen;     /* last "sequence" is a bunch of literals => don't progress anchor */
1416
0
                        continue;   /* will finish */
1417
0
                    }
1418
1419
17.3M
                    assert(anchor + llen <= iend);
1420
17.3M
                    ZSTD_updateStats(optStatePtr, llen, anchor, offBase, mlen);
1421
17.3M
                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offBase, mlen);
1422
17.3M
                    anchor += advance;
1423
17.3M
                    ip = anchor;
1424
17.3M
            }   }
1425
5.01M
            DEBUGLOG(7, "new offset history : %u, %u, %u", rep[0], rep[1], rep[2]);
1426
1427
            /* update all costs */
1428
5.01M
            ZSTD_setBasePrices(optStatePtr, optLevel);
1429
5.01M
        }
1430
5.01M
    }   /* while (ip < ilimit) */
1431
1432
    /* Return the last literals size */
1433
12.3k
    return (size_t)(iend - anchor);
1434
12.3k
}
1435
#endif /* build exclusions */
1436
1437
#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
1438
static size_t ZSTD_compressBlock_opt0(
1439
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1440
        const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
1441
4.82k
{
1442
4.82k
    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /* optLevel */, dictMode);
1443
4.82k
}
1444
#endif
1445
1446
#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
1447
static size_t ZSTD_compressBlock_opt2(
1448
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1449
        const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
1450
7.57k
{
1451
7.57k
    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /* optLevel */, dictMode);
1452
7.57k
}
1453
#endif
1454
1455
#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
1456
size_t ZSTD_compressBlock_btopt(
1457
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1458
        const void* src, size_t srcSize)
1459
4.82k
{
1460
4.82k
    DEBUGLOG(5, "ZSTD_compressBlock_btopt");
1461
4.82k
    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
1462
4.82k
}
1463
#endif
1464
1465
1466
1467
1468
#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
1469
/* ZSTD_initStats_ultra():
1470
 * make a first compression pass, just to seed stats with more accurate starting values.
1471
 * only works on first block, with no dictionary and no ldm.
1472
 * this function cannot error out, its narrow contract must be respected.
1473
 */
1474
static
1475
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
1476
void ZSTD_initStats_ultra(ZSTD_MatchState_t* ms,
1477
                          SeqStore_t* seqStore,
1478
                          U32 rep[ZSTD_REP_NUM],
1479
                    const void* src, size_t srcSize)
1480
2.22k
{
1481
2.22k
    U32 tmpRep[ZSTD_REP_NUM];  /* updated rep codes will sink here */
1482
2.22k
    ZSTD_memcpy(tmpRep, rep, sizeof(tmpRep));
1483
1484
2.22k
    DEBUGLOG(4, "ZSTD_initStats_ultra (srcSize=%zu)", srcSize);
1485
2.22k
    assert(ms->opt.litLengthSum == 0);    /* first block */
1486
2.22k
    assert(seqStore->sequences == seqStore->sequencesStart);   /* no ldm */
1487
2.22k
    assert(ms->window.dictLimit == ms->window.lowLimit);   /* no dictionary */
1488
2.22k
    assert(ms->window.dictLimit - ms->nextToUpdate <= 1);  /* no prefix (note: intentional overflow, defined as 2-complement) */
1489
1490
2.22k
    ZSTD_compressBlock_opt2(ms, seqStore, tmpRep, src, srcSize, ZSTD_noDict);   /* generate stats into ms->opt*/
1491
1492
    /* invalidate first scan from history, only keep entropy stats */
1493
2.22k
    ZSTD_resetSeqStore(seqStore);
1494
2.22k
    ms->window.base -= srcSize;
1495
2.22k
    ms->window.dictLimit += (U32)srcSize;
1496
2.22k
    ms->window.lowLimit = ms->window.dictLimit;
1497
2.22k
    ms->nextToUpdate = ms->window.dictLimit;
1498
1499
2.22k
}
1500
1501
size_t ZSTD_compressBlock_btultra(
1502
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1503
        const void* src, size_t srcSize)
1504
886
{
1505
886
    DEBUGLOG(5, "ZSTD_compressBlock_btultra (srcSize=%zu)", srcSize);
1506
886
    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
1507
886
}
1508
1509
size_t ZSTD_compressBlock_btultra2(
1510
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1511
        const void* src, size_t srcSize)
1512
4.45k
{
1513
4.45k
    U32 const curr = (U32)((const BYTE*)src - ms->window.base);
1514
4.45k
    DEBUGLOG(5, "ZSTD_compressBlock_btultra2 (srcSize=%zu)", srcSize);
1515
1516
    /* 2-passes strategy:
1517
     * this strategy makes a first pass over first block to collect statistics
1518
     * in order to seed next round's statistics with it.
1519
     * After 1st pass, function forgets history, and starts a new block.
1520
     * Consequently, this can only work if no data has been previously loaded in tables,
1521
     * aka, no dictionary, no prefix, no ldm preprocessing.
1522
     * The compression ratio gain is generally small (~0.5% on first block),
1523
     * the cost is 2x cpu time on first block. */
1524
4.45k
    assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
1525
4.45k
    if ( (ms->opt.litLengthSum==0)   /* first block */
1526
2.28k
      && (seqStore->sequences == seqStore->sequencesStart)  /* no ldm */
1527
2.28k
      && (ms->window.dictLimit == ms->window.lowLimit)   /* no dictionary */
1528
2.28k
      && (curr == ms->window.dictLimit)    /* start of frame, nothing already loaded nor skipped */
1529
2.28k
      && (srcSize > ZSTD_PREDEF_THRESHOLD) /* input large enough to not employ default stats */
1530
4.45k
      ) {
1531
2.22k
        ZSTD_initStats_ultra(ms, seqStore, rep, src, srcSize);
1532
2.22k
    }
1533
1534
4.45k
    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
1535
4.45k
}
1536
#endif
1537
1538
#ifndef ZSTD_EXCLUDE_BTOPT_BLOCK_COMPRESSOR
1539
size_t ZSTD_compressBlock_btopt_dictMatchState(
1540
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1541
        const void* src, size_t srcSize)
1542
0
{
1543
0
    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
1544
0
}
1545
1546
size_t ZSTD_compressBlock_btopt_extDict(
1547
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1548
        const void* src, size_t srcSize)
1549
0
{
1550
0
    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
1551
0
}
1552
#endif
1553
1554
#ifndef ZSTD_EXCLUDE_BTULTRA_BLOCK_COMPRESSOR
1555
size_t ZSTD_compressBlock_btultra_dictMatchState(
1556
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1557
        const void* src, size_t srcSize)
1558
0
{
1559
0
    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
1560
0
}
1561
1562
size_t ZSTD_compressBlock_btultra_extDict(
1563
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1564
        const void* src, size_t srcSize)
1565
0
{
1566
0
    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
1567
0
}
1568
#endif
1569
1570
/* note : no btultra2 variant for extDict nor dictMatchState,
1571
 * because btultra2 is not meant to work with dictionaries
1572
 * and is only specific for the first block (no prefix) */