xref: /freebsd/sys/contrib/zstd/lib/compress/zstd_compress_sequences.c (revision 6966ac055c3b7a39266fb982493330df7a097997)
1 /*
2  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
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  /*-*************************************
12  *  Dependencies
13  ***************************************/
14 #include "zstd_compress_sequences.h"
15 
16 /**
17  * -log2(x / 256) lookup table for x in [0, 256).
18  * If x == 0: Return 0
19  * Else: Return floor(-log2(x / 256) * 256)
20  */
21 static unsigned const kInverseProbabilityLog256[256] = {
22     0,    2048, 1792, 1642, 1536, 1453, 1386, 1329, 1280, 1236, 1197, 1162,
23     1130, 1100, 1073, 1047, 1024, 1001, 980,  960,  941,  923,  906,  889,
24     874,  859,  844,  830,  817,  804,  791,  779,  768,  756,  745,  734,
25     724,  714,  704,  694,  685,  676,  667,  658,  650,  642,  633,  626,
26     618,  610,  603,  595,  588,  581,  574,  567,  561,  554,  548,  542,
27     535,  529,  523,  517,  512,  506,  500,  495,  489,  484,  478,  473,
28     468,  463,  458,  453,  448,  443,  438,  434,  429,  424,  420,  415,
29     411,  407,  402,  398,  394,  390,  386,  382,  377,  373,  370,  366,
30     362,  358,  354,  350,  347,  343,  339,  336,  332,  329,  325,  322,
31     318,  315,  311,  308,  305,  302,  298,  295,  292,  289,  286,  282,
32     279,  276,  273,  270,  267,  264,  261,  258,  256,  253,  250,  247,
33     244,  241,  239,  236,  233,  230,  228,  225,  222,  220,  217,  215,
34     212,  209,  207,  204,  202,  199,  197,  194,  192,  190,  187,  185,
35     182,  180,  178,  175,  173,  171,  168,  166,  164,  162,  159,  157,
36     155,  153,  151,  149,  146,  144,  142,  140,  138,  136,  134,  132,
37     130,  128,  126,  123,  121,  119,  117,  115,  114,  112,  110,  108,
38     106,  104,  102,  100,  98,   96,   94,   93,   91,   89,   87,   85,
39     83,   82,   80,   78,   76,   74,   73,   71,   69,   67,   66,   64,
40     62,   61,   59,   57,   55,   54,   52,   50,   49,   47,   46,   44,
41     42,   41,   39,   37,   36,   34,   33,   31,   30,   28,   26,   25,
42     23,   22,   20,   19,   17,   16,   14,   13,   11,   10,   8,    7,
43     5,    4,    2,    1,
44 };
45 
46 static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
47   void const* ptr = ctable;
48   U16 const* u16ptr = (U16 const*)ptr;
49   U32 const maxSymbolValue = MEM_read16(u16ptr + 1);
50   return maxSymbolValue;
51 }
52 
53 /**
54  * Returns the cost in bytes of encoding the normalized count header.
55  * Returns an error if any of the helper functions return an error.
56  */
57 static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
58                               size_t const nbSeq, unsigned const FSELog)
59 {
60     BYTE wksp[FSE_NCOUNTBOUND];
61     S16 norm[MaxSeq + 1];
62     const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
63     FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max));
64     return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
65 }
66 
67 /**
68  * Returns the cost in bits of encoding the distribution described by count
69  * using the entropy bound.
70  */
71 static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t const total)
72 {
73     unsigned cost = 0;
74     unsigned s;
75     for (s = 0; s <= max; ++s) {
76         unsigned norm = (unsigned)((256 * count[s]) / total);
77         if (count[s] != 0 && norm == 0)
78             norm = 1;
79         assert(count[s] < total);
80         cost += count[s] * kInverseProbabilityLog256[norm];
81     }
82     return cost >> 8;
83 }
84 
85 /**
86  * Returns the cost in bits of encoding the distribution in count using ctable.
87  * Returns an error if ctable cannot represent all the symbols in count.
88  */
89 static size_t ZSTD_fseBitCost(
90     FSE_CTable const* ctable,
91     unsigned const* count,
92     unsigned const max)
93 {
94     unsigned const kAccuracyLog = 8;
95     size_t cost = 0;
96     unsigned s;
97     FSE_CState_t cstate;
98     FSE_initCState(&cstate, ctable);
99     RETURN_ERROR_IF(ZSTD_getFSEMaxSymbolValue(ctable) < max, GENERIC,
100                     "Repeat FSE_CTable has maxSymbolValue %u < %u",
101                     ZSTD_getFSEMaxSymbolValue(ctable), max);
102     for (s = 0; s <= max; ++s) {
103         unsigned const tableLog = cstate.stateLog;
104         unsigned const badCost = (tableLog + 1) << kAccuracyLog;
105         unsigned const bitCost = FSE_bitCost(cstate.symbolTT, tableLog, s, kAccuracyLog);
106         if (count[s] == 0)
107             continue;
108         RETURN_ERROR_IF(bitCost >= badCost, GENERIC,
109                         "Repeat FSE_CTable has Prob[%u] == 0", s);
110         cost += count[s] * bitCost;
111     }
112     return cost >> kAccuracyLog;
113 }
114 
115 /**
116  * Returns the cost in bits of encoding the distribution in count using the
117  * table described by norm. The max symbol support by norm is assumed >= max.
118  * norm must be valid for every symbol with non-zero probability in count.
119  */
120 static size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
121                                     unsigned const* count, unsigned const max)
122 {
123     unsigned const shift = 8 - accuracyLog;
124     size_t cost = 0;
125     unsigned s;
126     assert(accuracyLog <= 8);
127     for (s = 0; s <= max; ++s) {
128         unsigned const normAcc = norm[s] != -1 ? norm[s] : 1;
129         unsigned const norm256 = normAcc << shift;
130         assert(norm256 > 0);
131         assert(norm256 < 256);
132         cost += count[s] * kInverseProbabilityLog256[norm256];
133     }
134     return cost >> 8;
135 }
136 
137 symbolEncodingType_e
138 ZSTD_selectEncodingType(
139         FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
140         size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
141         FSE_CTable const* prevCTable,
142         short const* defaultNorm, U32 defaultNormLog,
143         ZSTD_defaultPolicy_e const isDefaultAllowed,
144         ZSTD_strategy const strategy)
145 {
146     ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
147     if (mostFrequent == nbSeq) {
148         *repeatMode = FSE_repeat_none;
149         if (isDefaultAllowed && nbSeq <= 2) {
150             /* Prefer set_basic over set_rle when there are 2 or less symbols,
151              * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
152              * If basic encoding isn't possible, always choose RLE.
153              */
154             DEBUGLOG(5, "Selected set_basic");
155             return set_basic;
156         }
157         DEBUGLOG(5, "Selected set_rle");
158         return set_rle;
159     }
160     if (strategy < ZSTD_lazy) {
161         if (isDefaultAllowed) {
162             size_t const staticFse_nbSeq_max = 1000;
163             size_t const mult = 10 - strategy;
164             size_t const baseLog = 3;
165             size_t const dynamicFse_nbSeq_min = (((size_t)1 << defaultNormLog) * mult) >> baseLog;  /* 28-36 for offset, 56-72 for lengths */
166             assert(defaultNormLog >= 5 && defaultNormLog <= 6);  /* xx_DEFAULTNORMLOG */
167             assert(mult <= 9 && mult >= 7);
168             if ( (*repeatMode == FSE_repeat_valid)
169               && (nbSeq < staticFse_nbSeq_max) ) {
170                 DEBUGLOG(5, "Selected set_repeat");
171                 return set_repeat;
172             }
173             if ( (nbSeq < dynamicFse_nbSeq_min)
174               || (mostFrequent < (nbSeq >> (defaultNormLog-1))) ) {
175                 DEBUGLOG(5, "Selected set_basic");
176                 /* The format allows default tables to be repeated, but it isn't useful.
177                  * When using simple heuristics to select encoding type, we don't want
178                  * to confuse these tables with dictionaries. When running more careful
179                  * analysis, we don't need to waste time checking both repeating tables
180                  * and default tables.
181                  */
182                 *repeatMode = FSE_repeat_none;
183                 return set_basic;
184             }
185         }
186     } else {
187         size_t const basicCost = isDefaultAllowed ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, count, max) : ERROR(GENERIC);
188         size_t const repeatCost = *repeatMode != FSE_repeat_none ? ZSTD_fseBitCost(prevCTable, count, max) : ERROR(GENERIC);
189         size_t const NCountCost = ZSTD_NCountCost(count, max, nbSeq, FSELog);
190         size_t const compressedCost = (NCountCost << 3) + ZSTD_entropyCost(count, max, nbSeq);
191 
192         if (isDefaultAllowed) {
193             assert(!ZSTD_isError(basicCost));
194             assert(!(*repeatMode == FSE_repeat_valid && ZSTD_isError(repeatCost)));
195         }
196         assert(!ZSTD_isError(NCountCost));
197         assert(compressedCost < ERROR(maxCode));
198         DEBUGLOG(5, "Estimated bit costs: basic=%u\trepeat=%u\tcompressed=%u",
199                     (unsigned)basicCost, (unsigned)repeatCost, (unsigned)compressedCost);
200         if (basicCost <= repeatCost && basicCost <= compressedCost) {
201             DEBUGLOG(5, "Selected set_basic");
202             assert(isDefaultAllowed);
203             *repeatMode = FSE_repeat_none;
204             return set_basic;
205         }
206         if (repeatCost <= compressedCost) {
207             DEBUGLOG(5, "Selected set_repeat");
208             assert(!ZSTD_isError(repeatCost));
209             return set_repeat;
210         }
211         assert(compressedCost < basicCost && compressedCost < repeatCost);
212     }
213     DEBUGLOG(5, "Selected set_compressed");
214     *repeatMode = FSE_repeat_check;
215     return set_compressed;
216 }
217 
218 size_t
219 ZSTD_buildCTable(void* dst, size_t dstCapacity,
220                 FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
221                 unsigned* count, U32 max,
222                 const BYTE* codeTable, size_t nbSeq,
223                 const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
224                 const FSE_CTable* prevCTable, size_t prevCTableSize,
225                 void* entropyWorkspace, size_t entropyWorkspaceSize)
226 {
227     BYTE* op = (BYTE*)dst;
228     const BYTE* const oend = op + dstCapacity;
229     DEBUGLOG(6, "ZSTD_buildCTable (dstCapacity=%u)", (unsigned)dstCapacity);
230 
231     switch (type) {
232     case set_rle:
233         FORWARD_IF_ERROR(FSE_buildCTable_rle(nextCTable, (BYTE)max));
234         RETURN_ERROR_IF(dstCapacity==0, dstSize_tooSmall);
235         *op = codeTable[0];
236         return 1;
237     case set_repeat:
238         memcpy(nextCTable, prevCTable, prevCTableSize);
239         return 0;
240     case set_basic:
241         FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize));  /* note : could be pre-calculated */
242         return 0;
243     case set_compressed: {
244         S16 norm[MaxSeq + 1];
245         size_t nbSeq_1 = nbSeq;
246         const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
247         if (count[codeTable[nbSeq-1]] > 1) {
248             count[codeTable[nbSeq-1]]--;
249             nbSeq_1--;
250         }
251         assert(nbSeq_1 > 1);
252         FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max));
253         {   size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog);   /* overflow protected */
254             FORWARD_IF_ERROR(NCountSize);
255             FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, entropyWorkspace, entropyWorkspaceSize));
256             return NCountSize;
257         }
258     }
259     default: assert(0); RETURN_ERROR(GENERIC);
260     }
261 }
262 
263 FORCE_INLINE_TEMPLATE size_t
264 ZSTD_encodeSequences_body(
265             void* dst, size_t dstCapacity,
266             FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
267             FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
268             FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
269             seqDef const* sequences, size_t nbSeq, int longOffsets)
270 {
271     BIT_CStream_t blockStream;
272     FSE_CState_t  stateMatchLength;
273     FSE_CState_t  stateOffsetBits;
274     FSE_CState_t  stateLitLength;
275 
276     RETURN_ERROR_IF(
277         ERR_isError(BIT_initCStream(&blockStream, dst, dstCapacity)),
278         dstSize_tooSmall, "not enough space remaining");
279     DEBUGLOG(6, "available space for bitstream : %i  (dstCapacity=%u)",
280                 (int)(blockStream.endPtr - blockStream.startPtr),
281                 (unsigned)dstCapacity);
282 
283     /* first symbols */
284     FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
285     FSE_initCState2(&stateOffsetBits,  CTable_OffsetBits,  ofCodeTable[nbSeq-1]);
286     FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
287     BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
288     if (MEM_32bits()) BIT_flushBits(&blockStream);
289     BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
290     if (MEM_32bits()) BIT_flushBits(&blockStream);
291     if (longOffsets) {
292         U32 const ofBits = ofCodeTable[nbSeq-1];
293         int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
294         if (extraBits) {
295             BIT_addBits(&blockStream, sequences[nbSeq-1].offset, extraBits);
296             BIT_flushBits(&blockStream);
297         }
298         BIT_addBits(&blockStream, sequences[nbSeq-1].offset >> extraBits,
299                     ofBits - extraBits);
300     } else {
301         BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
302     }
303     BIT_flushBits(&blockStream);
304 
305     {   size_t n;
306         for (n=nbSeq-2 ; n<nbSeq ; n--) {      /* intentional underflow */
307             BYTE const llCode = llCodeTable[n];
308             BYTE const ofCode = ofCodeTable[n];
309             BYTE const mlCode = mlCodeTable[n];
310             U32  const llBits = LL_bits[llCode];
311             U32  const ofBits = ofCode;
312             U32  const mlBits = ML_bits[mlCode];
313             DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
314                         (unsigned)sequences[n].litLength,
315                         (unsigned)sequences[n].matchLength + MINMATCH,
316                         (unsigned)sequences[n].offset);
317                                                                             /* 32b*/  /* 64b*/
318                                                                             /* (7)*/  /* (7)*/
319             FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
320             FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode);      /* 24 */  /* 24 */
321             if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
322             FSE_encodeSymbol(&blockStream, &stateLitLength, llCode);        /* 16 */  /* 33 */
323             if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
324                 BIT_flushBits(&blockStream);                                /* (7)*/
325             BIT_addBits(&blockStream, sequences[n].litLength, llBits);
326             if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
327             BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
328             if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
329             if (longOffsets) {
330                 int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
331                 if (extraBits) {
332                     BIT_addBits(&blockStream, sequences[n].offset, extraBits);
333                     BIT_flushBits(&blockStream);                            /* (7)*/
334                 }
335                 BIT_addBits(&blockStream, sequences[n].offset >> extraBits,
336                             ofBits - extraBits);                            /* 31 */
337             } else {
338                 BIT_addBits(&blockStream, sequences[n].offset, ofBits);     /* 31 */
339             }
340             BIT_flushBits(&blockStream);                                    /* (7)*/
341             DEBUGLOG(7, "remaining space : %i", (int)(blockStream.endPtr - blockStream.ptr));
342     }   }
343 
344     DEBUGLOG(6, "ZSTD_encodeSequences: flushing ML state with %u bits", stateMatchLength.stateLog);
345     FSE_flushCState(&blockStream, &stateMatchLength);
346     DEBUGLOG(6, "ZSTD_encodeSequences: flushing Off state with %u bits", stateOffsetBits.stateLog);
347     FSE_flushCState(&blockStream, &stateOffsetBits);
348     DEBUGLOG(6, "ZSTD_encodeSequences: flushing LL state with %u bits", stateLitLength.stateLog);
349     FSE_flushCState(&blockStream, &stateLitLength);
350 
351     {   size_t const streamSize = BIT_closeCStream(&blockStream);
352         RETURN_ERROR_IF(streamSize==0, dstSize_tooSmall, "not enough space");
353         return streamSize;
354     }
355 }
356 
357 static size_t
358 ZSTD_encodeSequences_default(
359             void* dst, size_t dstCapacity,
360             FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
361             FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
362             FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
363             seqDef const* sequences, size_t nbSeq, int longOffsets)
364 {
365     return ZSTD_encodeSequences_body(dst, dstCapacity,
366                                     CTable_MatchLength, mlCodeTable,
367                                     CTable_OffsetBits, ofCodeTable,
368                                     CTable_LitLength, llCodeTable,
369                                     sequences, nbSeq, longOffsets);
370 }
371 
372 
373 #if DYNAMIC_BMI2
374 
375 static TARGET_ATTRIBUTE("bmi2") size_t
376 ZSTD_encodeSequences_bmi2(
377             void* dst, size_t dstCapacity,
378             FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
379             FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
380             FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
381             seqDef const* sequences, size_t nbSeq, int longOffsets)
382 {
383     return ZSTD_encodeSequences_body(dst, dstCapacity,
384                                     CTable_MatchLength, mlCodeTable,
385                                     CTable_OffsetBits, ofCodeTable,
386                                     CTable_LitLength, llCodeTable,
387                                     sequences, nbSeq, longOffsets);
388 }
389 
390 #endif
391 
392 size_t ZSTD_encodeSequences(
393             void* dst, size_t dstCapacity,
394             FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
395             FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
396             FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
397             seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2)
398 {
399     DEBUGLOG(5, "ZSTD_encodeSequences: dstCapacity = %u", (unsigned)dstCapacity);
400 #if DYNAMIC_BMI2
401     if (bmi2) {
402         return ZSTD_encodeSequences_bmi2(dst, dstCapacity,
403                                          CTable_MatchLength, mlCodeTable,
404                                          CTable_OffsetBits, ofCodeTable,
405                                          CTable_LitLength, llCodeTable,
406                                          sequences, nbSeq, longOffsets);
407     }
408 #endif
409     (void)bmi2;
410     return ZSTD_encodeSequences_default(dst, dstCapacity,
411                                         CTable_MatchLength, mlCodeTable,
412                                         CTable_OffsetBits, ofCodeTable,
413                                         CTable_LitLength, llCodeTable,
414                                         sequences, nbSeq, longOffsets);
415 }
416