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