xref: /freebsd/sys/contrib/zstd/lib/dictBuilder/zdict.c (revision 81ad6265)
1 /*
2  * Copyright (c) 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 /*-**************************************
13 *  Tuning parameters
14 ****************************************/
15 #define MINRATIO 4   /* minimum nb of apparition to be selected in dictionary */
16 #define ZDICT_MAX_SAMPLES_SIZE (2000U << 20)
17 #define ZDICT_MIN_SAMPLES_SIZE (ZDICT_CONTENTSIZE_MIN * MINRATIO)
18 
19 
20 /*-**************************************
21 *  Compiler Options
22 ****************************************/
23 /* Unix Large Files support (>4GB) */
24 #define _FILE_OFFSET_BITS 64
25 #if (defined(__sun__) && (!defined(__LP64__)))   /* Sun Solaris 32-bits requires specific definitions */
26 #  ifndef _LARGEFILE_SOURCE
27 #  define _LARGEFILE_SOURCE
28 #  endif
29 #elif ! defined(__LP64__)                        /* No point defining Large file for 64 bit */
30 #  ifndef _LARGEFILE64_SOURCE
31 #  define _LARGEFILE64_SOURCE
32 #  endif
33 #endif
34 
35 
36 /*-*************************************
37 *  Dependencies
38 ***************************************/
39 #include <stdlib.h>        /* malloc, free */
40 #include <string.h>        /* memset */
41 #include <stdio.h>         /* fprintf, fopen, ftello64 */
42 #include <time.h>          /* clock */
43 
44 #ifndef ZDICT_STATIC_LINKING_ONLY
45 #  define ZDICT_STATIC_LINKING_ONLY
46 #endif
47 #define HUF_STATIC_LINKING_ONLY
48 
49 #include "../common/mem.h"           /* read */
50 #include "../common/fse.h"           /* FSE_normalizeCount, FSE_writeNCount */
51 #include "../common/huf.h"           /* HUF_buildCTable, HUF_writeCTable */
52 #include "../common/zstd_internal.h" /* includes zstd.h */
53 #include "../common/xxhash.h"        /* XXH64 */
54 #include "../compress/zstd_compress_internal.h" /* ZSTD_loadCEntropy() */
55 #include "../zdict.h"
56 #include "divsufsort.h"
57 
58 
59 /*-*************************************
60 *  Constants
61 ***************************************/
62 #define KB *(1 <<10)
63 #define MB *(1 <<20)
64 #define GB *(1U<<30)
65 
66 #define DICTLISTSIZE_DEFAULT 10000
67 
68 #define NOISELENGTH 32
69 
70 static const U32 g_selectivity_default = 9;
71 
72 
73 /*-*************************************
74 *  Console display
75 ***************************************/
76 #undef  DISPLAY
77 #define DISPLAY(...)         { fprintf(stderr, __VA_ARGS__); fflush( stderr ); }
78 #undef  DISPLAYLEVEL
79 #define DISPLAYLEVEL(l, ...) if (notificationLevel>=l) { DISPLAY(__VA_ARGS__); }    /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
80 
81 static clock_t ZDICT_clockSpan(clock_t nPrevious) { return clock() - nPrevious; }
82 
83 static void ZDICT_printHex(const void* ptr, size_t length)
84 {
85     const BYTE* const b = (const BYTE*)ptr;
86     size_t u;
87     for (u=0; u<length; u++) {
88         BYTE c = b[u];
89         if (c<32 || c>126) c = '.';   /* non-printable char */
90         DISPLAY("%c", c);
91     }
92 }
93 
94 
95 /*-********************************************************
96 *  Helper functions
97 **********************************************************/
98 unsigned ZDICT_isError(size_t errorCode) { return ERR_isError(errorCode); }
99 
100 const char* ZDICT_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
101 
102 unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize)
103 {
104     if (dictSize < 8) return 0;
105     if (MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return 0;
106     return MEM_readLE32((const char*)dictBuffer + 4);
107 }
108 
109 size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize)
110 {
111     size_t headerSize;
112     if (dictSize <= 8 || MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return ERROR(dictionary_corrupted);
113 
114     {   ZSTD_compressedBlockState_t* bs = (ZSTD_compressedBlockState_t*)malloc(sizeof(ZSTD_compressedBlockState_t));
115         U32* wksp = (U32*)malloc(HUF_WORKSPACE_SIZE);
116         if (!bs || !wksp) {
117             headerSize = ERROR(memory_allocation);
118         } else {
119             ZSTD_reset_compressedBlockState(bs);
120             headerSize = ZSTD_loadCEntropy(bs, wksp, dictBuffer, dictSize);
121         }
122 
123         free(bs);
124         free(wksp);
125     }
126 
127     return headerSize;
128 }
129 
130 /*-********************************************************
131 *  Dictionary training functions
132 **********************************************************/
133 static unsigned ZDICT_NbCommonBytes (size_t val)
134 {
135     if (MEM_isLittleEndian()) {
136         if (MEM_64bits()) {
137 #       if defined(_MSC_VER) && defined(_WIN64)
138             if (val != 0) {
139                 unsigned long r;
140                 _BitScanForward64(&r, (U64)val);
141                 return (unsigned)(r >> 3);
142             } else {
143                 /* Should not reach this code path */
144                 __assume(0);
145             }
146 #       elif defined(__GNUC__) && (__GNUC__ >= 3)
147             return (unsigned)(__builtin_ctzll((U64)val) >> 3);
148 #       else
149             static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2, 0, 3, 1, 3, 1, 4, 2, 7, 0, 2, 3, 6, 1, 5, 3, 5, 1, 3, 4, 4, 2, 5, 6, 7, 7, 0, 1, 2, 3, 3, 4, 6, 2, 6, 5, 5, 3, 4, 5, 6, 7, 1, 2, 4, 6, 4, 4, 5, 7, 2, 6, 5, 7, 6, 7, 7 };
150             return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
151 #       endif
152         } else { /* 32 bits */
153 #       if defined(_MSC_VER)
154             if (val != 0) {
155                 unsigned long r;
156                 _BitScanForward(&r, (U32)val);
157                 return (unsigned)(r >> 3);
158             } else {
159                 /* Should not reach this code path */
160                 __assume(0);
161             }
162 #       elif defined(__GNUC__) && (__GNUC__ >= 3)
163             return (unsigned)(__builtin_ctz((U32)val) >> 3);
164 #       else
165             static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0, 3, 2, 2, 1, 3, 2, 0, 1, 3, 3, 1, 2, 2, 2, 2, 0, 3, 1, 2, 0, 1, 0, 1, 1 };
166             return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
167 #       endif
168         }
169     } else {  /* Big Endian CPU */
170         if (MEM_64bits()) {
171 #       if defined(_MSC_VER) && defined(_WIN64)
172             if (val != 0) {
173                 unsigned long r;
174                 _BitScanReverse64(&r, val);
175                 return (unsigned)(r >> 3);
176             } else {
177                 /* Should not reach this code path */
178                 __assume(0);
179             }
180 #       elif defined(__GNUC__) && (__GNUC__ >= 3)
181             return (unsigned)(__builtin_clzll(val) >> 3);
182 #       else
183             unsigned r;
184             const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
185             if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
186             if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
187             r += (!val);
188             return r;
189 #       endif
190         } else { /* 32 bits */
191 #       if defined(_MSC_VER)
192             if (val != 0) {
193                 unsigned long r;
194                 _BitScanReverse(&r, (unsigned long)val);
195                 return (unsigned)(r >> 3);
196             } else {
197                 /* Should not reach this code path */
198                 __assume(0);
199             }
200 #       elif defined(__GNUC__) && (__GNUC__ >= 3)
201             return (unsigned)(__builtin_clz((U32)val) >> 3);
202 #       else
203             unsigned r;
204             if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
205             r += (!val);
206             return r;
207 #       endif
208     }   }
209 }
210 
211 
212 /*! ZDICT_count() :
213     Count the nb of common bytes between 2 pointers.
214     Note : this function presumes end of buffer followed by noisy guard band.
215 */
216 static size_t ZDICT_count(const void* pIn, const void* pMatch)
217 {
218     const char* const pStart = (const char*)pIn;
219     for (;;) {
220         size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
221         if (!diff) {
222             pIn = (const char*)pIn+sizeof(size_t);
223             pMatch = (const char*)pMatch+sizeof(size_t);
224             continue;
225         }
226         pIn = (const char*)pIn+ZDICT_NbCommonBytes(diff);
227         return (size_t)((const char*)pIn - pStart);
228     }
229 }
230 
231 
232 typedef struct {
233     U32 pos;
234     U32 length;
235     U32 savings;
236 } dictItem;
237 
238 static void ZDICT_initDictItem(dictItem* d)
239 {
240     d->pos = 1;
241     d->length = 0;
242     d->savings = (U32)(-1);
243 }
244 
245 
246 #define LLIMIT 64          /* heuristic determined experimentally */
247 #define MINMATCHLENGTH 7   /* heuristic determined experimentally */
248 static dictItem ZDICT_analyzePos(
249                        BYTE* doneMarks,
250                        const int* suffix, U32 start,
251                        const void* buffer, U32 minRatio, U32 notificationLevel)
252 {
253     U32 lengthList[LLIMIT] = {0};
254     U32 cumulLength[LLIMIT] = {0};
255     U32 savings[LLIMIT] = {0};
256     const BYTE* b = (const BYTE*)buffer;
257     size_t maxLength = LLIMIT;
258     size_t pos = (size_t)suffix[start];
259     U32 end = start;
260     dictItem solution;
261 
262     /* init */
263     memset(&solution, 0, sizeof(solution));
264     doneMarks[pos] = 1;
265 
266     /* trivial repetition cases */
267     if ( (MEM_read16(b+pos+0) == MEM_read16(b+pos+2))
268        ||(MEM_read16(b+pos+1) == MEM_read16(b+pos+3))
269        ||(MEM_read16(b+pos+2) == MEM_read16(b+pos+4)) ) {
270         /* skip and mark segment */
271         U16 const pattern16 = MEM_read16(b+pos+4);
272         U32 u, patternEnd = 6;
273         while (MEM_read16(b+pos+patternEnd) == pattern16) patternEnd+=2 ;
274         if (b[pos+patternEnd] == b[pos+patternEnd-1]) patternEnd++;
275         for (u=1; u<patternEnd; u++)
276             doneMarks[pos+u] = 1;
277         return solution;
278     }
279 
280     /* look forward */
281     {   size_t length;
282         do {
283             end++;
284             length = ZDICT_count(b + pos, b + suffix[end]);
285         } while (length >= MINMATCHLENGTH);
286     }
287 
288     /* look backward */
289     {   size_t length;
290         do {
291             length = ZDICT_count(b + pos, b + *(suffix+start-1));
292             if (length >=MINMATCHLENGTH) start--;
293         } while(length >= MINMATCHLENGTH);
294     }
295 
296     /* exit if not found a minimum nb of repetitions */
297     if (end-start < minRatio) {
298         U32 idx;
299         for(idx=start; idx<end; idx++)
300             doneMarks[suffix[idx]] = 1;
301         return solution;
302     }
303 
304     {   int i;
305         U32 mml;
306         U32 refinedStart = start;
307         U32 refinedEnd = end;
308 
309         DISPLAYLEVEL(4, "\n");
310         DISPLAYLEVEL(4, "found %3u matches of length >= %i at pos %7u  ", (unsigned)(end-start), MINMATCHLENGTH, (unsigned)pos);
311         DISPLAYLEVEL(4, "\n");
312 
313         for (mml = MINMATCHLENGTH ; ; mml++) {
314             BYTE currentChar = 0;
315             U32 currentCount = 0;
316             U32 currentID = refinedStart;
317             U32 id;
318             U32 selectedCount = 0;
319             U32 selectedID = currentID;
320             for (id =refinedStart; id < refinedEnd; id++) {
321                 if (b[suffix[id] + mml] != currentChar) {
322                     if (currentCount > selectedCount) {
323                         selectedCount = currentCount;
324                         selectedID = currentID;
325                     }
326                     currentID = id;
327                     currentChar = b[ suffix[id] + mml];
328                     currentCount = 0;
329                 }
330                 currentCount ++;
331             }
332             if (currentCount > selectedCount) {  /* for last */
333                 selectedCount = currentCount;
334                 selectedID = currentID;
335             }
336 
337             if (selectedCount < minRatio)
338                 break;
339             refinedStart = selectedID;
340             refinedEnd = refinedStart + selectedCount;
341         }
342 
343         /* evaluate gain based on new dict */
344         start = refinedStart;
345         pos = suffix[refinedStart];
346         end = start;
347         memset(lengthList, 0, sizeof(lengthList));
348 
349         /* look forward */
350         {   size_t length;
351             do {
352                 end++;
353                 length = ZDICT_count(b + pos, b + suffix[end]);
354                 if (length >= LLIMIT) length = LLIMIT-1;
355                 lengthList[length]++;
356             } while (length >=MINMATCHLENGTH);
357         }
358 
359         /* look backward */
360         {   size_t length = MINMATCHLENGTH;
361             while ((length >= MINMATCHLENGTH) & (start > 0)) {
362                 length = ZDICT_count(b + pos, b + suffix[start - 1]);
363                 if (length >= LLIMIT) length = LLIMIT - 1;
364                 lengthList[length]++;
365                 if (length >= MINMATCHLENGTH) start--;
366             }
367         }
368 
369         /* largest useful length */
370         memset(cumulLength, 0, sizeof(cumulLength));
371         cumulLength[maxLength-1] = lengthList[maxLength-1];
372         for (i=(int)(maxLength-2); i>=0; i--)
373             cumulLength[i] = cumulLength[i+1] + lengthList[i];
374 
375         for (i=LLIMIT-1; i>=MINMATCHLENGTH; i--) if (cumulLength[i]>=minRatio) break;
376         maxLength = i;
377 
378         /* reduce maxLength in case of final into repetitive data */
379         {   U32 l = (U32)maxLength;
380             BYTE const c = b[pos + maxLength-1];
381             while (b[pos+l-2]==c) l--;
382             maxLength = l;
383         }
384         if (maxLength < MINMATCHLENGTH) return solution;   /* skip : no long-enough solution */
385 
386         /* calculate savings */
387         savings[5] = 0;
388         for (i=MINMATCHLENGTH; i<=(int)maxLength; i++)
389             savings[i] = savings[i-1] + (lengthList[i] * (i-3));
390 
391         DISPLAYLEVEL(4, "Selected dict at position %u, of length %u : saves %u (ratio: %.2f)  \n",
392                      (unsigned)pos, (unsigned)maxLength, (unsigned)savings[maxLength], (double)savings[maxLength] / (double)maxLength);
393 
394         solution.pos = (U32)pos;
395         solution.length = (U32)maxLength;
396         solution.savings = savings[maxLength];
397 
398         /* mark positions done */
399         {   U32 id;
400             for (id=start; id<end; id++) {
401                 U32 p, pEnd, length;
402                 U32 const testedPos = (U32)suffix[id];
403                 if (testedPos == pos)
404                     length = solution.length;
405                 else {
406                     length = (U32)ZDICT_count(b+pos, b+testedPos);
407                     if (length > solution.length) length = solution.length;
408                 }
409                 pEnd = (U32)(testedPos + length);
410                 for (p=testedPos; p<pEnd; p++)
411                     doneMarks[p] = 1;
412     }   }   }
413 
414     return solution;
415 }
416 
417 
418 static int isIncluded(const void* in, const void* container, size_t length)
419 {
420     const char* const ip = (const char*) in;
421     const char* const into = (const char*) container;
422     size_t u;
423 
424     for (u=0; u<length; u++) {  /* works because end of buffer is a noisy guard band */
425         if (ip[u] != into[u]) break;
426     }
427 
428     return u==length;
429 }
430 
431 /*! ZDICT_tryMerge() :
432     check if dictItem can be merged, do it if possible
433     @return : id of destination elt, 0 if not merged
434 */
435 static U32 ZDICT_tryMerge(dictItem* table, dictItem elt, U32 eltNbToSkip, const void* buffer)
436 {
437     const U32 tableSize = table->pos;
438     const U32 eltEnd = elt.pos + elt.length;
439     const char* const buf = (const char*) buffer;
440 
441     /* tail overlap */
442     U32 u; for (u=1; u<tableSize; u++) {
443         if (u==eltNbToSkip) continue;
444         if ((table[u].pos > elt.pos) && (table[u].pos <= eltEnd)) {  /* overlap, existing > new */
445             /* append */
446             U32 const addedLength = table[u].pos - elt.pos;
447             table[u].length += addedLength;
448             table[u].pos = elt.pos;
449             table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
450             table[u].savings += elt.length / 8;    /* rough approx bonus */
451             elt = table[u];
452             /* sort : improve rank */
453             while ((u>1) && (table[u-1].savings < elt.savings))
454             table[u] = table[u-1], u--;
455             table[u] = elt;
456             return u;
457     }   }
458 
459     /* front overlap */
460     for (u=1; u<tableSize; u++) {
461         if (u==eltNbToSkip) continue;
462 
463         if ((table[u].pos + table[u].length >= elt.pos) && (table[u].pos < elt.pos)) {  /* overlap, existing < new */
464             /* append */
465             int const addedLength = (int)eltEnd - (int)(table[u].pos + table[u].length);
466             table[u].savings += elt.length / 8;    /* rough approx bonus */
467             if (addedLength > 0) {   /* otherwise, elt fully included into existing */
468                 table[u].length += addedLength;
469                 table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
470             }
471             /* sort : improve rank */
472             elt = table[u];
473             while ((u>1) && (table[u-1].savings < elt.savings))
474                 table[u] = table[u-1], u--;
475             table[u] = elt;
476             return u;
477         }
478 
479         if (MEM_read64(buf + table[u].pos) == MEM_read64(buf + elt.pos + 1)) {
480             if (isIncluded(buf + table[u].pos, buf + elt.pos + 1, table[u].length)) {
481                 size_t const addedLength = MAX( (int)elt.length - (int)table[u].length , 1 );
482                 table[u].pos = elt.pos;
483                 table[u].savings += (U32)(elt.savings * addedLength / elt.length);
484                 table[u].length = MIN(elt.length, table[u].length + 1);
485                 return u;
486             }
487         }
488     }
489 
490     return 0;
491 }
492 
493 
494 static void ZDICT_removeDictItem(dictItem* table, U32 id)
495 {
496     /* convention : table[0].pos stores nb of elts */
497     U32 const max = table[0].pos;
498     U32 u;
499     if (!id) return;   /* protection, should never happen */
500     for (u=id; u<max-1; u++)
501         table[u] = table[u+1];
502     table->pos--;
503 }
504 
505 
506 static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt, const void* buffer)
507 {
508     /* merge if possible */
509     U32 mergeId = ZDICT_tryMerge(table, elt, 0, buffer);
510     if (mergeId) {
511         U32 newMerge = 1;
512         while (newMerge) {
513             newMerge = ZDICT_tryMerge(table, table[mergeId], mergeId, buffer);
514             if (newMerge) ZDICT_removeDictItem(table, mergeId);
515             mergeId = newMerge;
516         }
517         return;
518     }
519 
520     /* insert */
521     {   U32 current;
522         U32 nextElt = table->pos;
523         if (nextElt >= maxSize) nextElt = maxSize-1;
524         current = nextElt-1;
525         while (table[current].savings < elt.savings) {
526             table[current+1] = table[current];
527             current--;
528         }
529         table[current+1] = elt;
530         table->pos = nextElt+1;
531     }
532 }
533 
534 
535 static U32 ZDICT_dictSize(const dictItem* dictList)
536 {
537     U32 u, dictSize = 0;
538     for (u=1; u<dictList[0].pos; u++)
539         dictSize += dictList[u].length;
540     return dictSize;
541 }
542 
543 
544 static size_t ZDICT_trainBuffer_legacy(dictItem* dictList, U32 dictListSize,
545                             const void* const buffer, size_t bufferSize,   /* buffer must end with noisy guard band */
546                             const size_t* fileSizes, unsigned nbFiles,
547                             unsigned minRatio, U32 notificationLevel)
548 {
549     int* const suffix0 = (int*)malloc((bufferSize+2)*sizeof(*suffix0));
550     int* const suffix = suffix0+1;
551     U32* reverseSuffix = (U32*)malloc((bufferSize)*sizeof(*reverseSuffix));
552     BYTE* doneMarks = (BYTE*)malloc((bufferSize+16)*sizeof(*doneMarks));   /* +16 for overflow security */
553     U32* filePos = (U32*)malloc(nbFiles * sizeof(*filePos));
554     size_t result = 0;
555     clock_t displayClock = 0;
556     clock_t const refreshRate = CLOCKS_PER_SEC * 3 / 10;
557 
558 #   undef  DISPLAYUPDATE
559 #   define DISPLAYUPDATE(l, ...) if (notificationLevel>=l) { \
560             if (ZDICT_clockSpan(displayClock) > refreshRate)  \
561             { displayClock = clock(); DISPLAY(__VA_ARGS__); \
562             if (notificationLevel>=4) fflush(stderr); } }
563 
564     /* init */
565     DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
566     if (!suffix0 || !reverseSuffix || !doneMarks || !filePos) {
567         result = ERROR(memory_allocation);
568         goto _cleanup;
569     }
570     if (minRatio < MINRATIO) minRatio = MINRATIO;
571     memset(doneMarks, 0, bufferSize+16);
572 
573     /* limit sample set size (divsufsort limitation)*/
574     if (bufferSize > ZDICT_MAX_SAMPLES_SIZE) DISPLAYLEVEL(3, "sample set too large : reduced to %u MB ...\n", (unsigned)(ZDICT_MAX_SAMPLES_SIZE>>20));
575     while (bufferSize > ZDICT_MAX_SAMPLES_SIZE) bufferSize -= fileSizes[--nbFiles];
576 
577     /* sort */
578     DISPLAYLEVEL(2, "sorting %u files of total size %u MB ...\n", nbFiles, (unsigned)(bufferSize>>20));
579     {   int const divSuftSortResult = divsufsort((const unsigned char*)buffer, suffix, (int)bufferSize, 0);
580         if (divSuftSortResult != 0) { result = ERROR(GENERIC); goto _cleanup; }
581     }
582     suffix[bufferSize] = (int)bufferSize;   /* leads into noise */
583     suffix0[0] = (int)bufferSize;           /* leads into noise */
584     /* build reverse suffix sort */
585     {   size_t pos;
586         for (pos=0; pos < bufferSize; pos++)
587             reverseSuffix[suffix[pos]] = (U32)pos;
588         /* note filePos tracks borders between samples.
589            It's not used at this stage, but planned to become useful in a later update */
590         filePos[0] = 0;
591         for (pos=1; pos<nbFiles; pos++)
592             filePos[pos] = (U32)(filePos[pos-1] + fileSizes[pos-1]);
593     }
594 
595     DISPLAYLEVEL(2, "finding patterns ... \n");
596     DISPLAYLEVEL(3, "minimum ratio : %u \n", minRatio);
597 
598     {   U32 cursor; for (cursor=0; cursor < bufferSize; ) {
599             dictItem solution;
600             if (doneMarks[cursor]) { cursor++; continue; }
601             solution = ZDICT_analyzePos(doneMarks, suffix, reverseSuffix[cursor], buffer, minRatio, notificationLevel);
602             if (solution.length==0) { cursor++; continue; }
603             ZDICT_insertDictItem(dictList, dictListSize, solution, buffer);
604             cursor += solution.length;
605             DISPLAYUPDATE(2, "\r%4.2f %% \r", (double)cursor / bufferSize * 100);
606     }   }
607 
608 _cleanup:
609     free(suffix0);
610     free(reverseSuffix);
611     free(doneMarks);
612     free(filePos);
613     return result;
614 }
615 
616 
617 static void ZDICT_fillNoise(void* buffer, size_t length)
618 {
619     unsigned const prime1 = 2654435761U;
620     unsigned const prime2 = 2246822519U;
621     unsigned acc = prime1;
622     size_t p=0;
623     for (p=0; p<length; p++) {
624         acc *= prime2;
625         ((unsigned char*)buffer)[p] = (unsigned char)(acc >> 21);
626     }
627 }
628 
629 
630 typedef struct
631 {
632     ZSTD_CDict* dict;    /* dictionary */
633     ZSTD_CCtx* zc;     /* working context */
634     void* workPlace;   /* must be ZSTD_BLOCKSIZE_MAX allocated */
635 } EStats_ress_t;
636 
637 #define MAXREPOFFSET 1024
638 
639 static void ZDICT_countEStats(EStats_ress_t esr, const ZSTD_parameters* params,
640                               unsigned* countLit, unsigned* offsetcodeCount, unsigned* matchlengthCount, unsigned* litlengthCount, U32* repOffsets,
641                               const void* src, size_t srcSize,
642                               U32 notificationLevel)
643 {
644     size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_MAX, 1 << params->cParams.windowLog);
645     size_t cSize;
646 
647     if (srcSize > blockSizeMax) srcSize = blockSizeMax;   /* protection vs large samples */
648     {   size_t const errorCode = ZSTD_compressBegin_usingCDict(esr.zc, esr.dict);
649         if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_compressBegin_usingCDict failed \n"); return; }
650 
651     }
652     cSize = ZSTD_compressBlock(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_MAX, src, srcSize);
653     if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (unsigned)srcSize); return; }
654 
655     if (cSize) {  /* if == 0; block is not compressible */
656         const seqStore_t* const seqStorePtr = ZSTD_getSeqStore(esr.zc);
657 
658         /* literals stats */
659         {   const BYTE* bytePtr;
660             for(bytePtr = seqStorePtr->litStart; bytePtr < seqStorePtr->lit; bytePtr++)
661                 countLit[*bytePtr]++;
662         }
663 
664         /* seqStats */
665         {   U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
666             ZSTD_seqToCodes(seqStorePtr);
667 
668             {   const BYTE* codePtr = seqStorePtr->ofCode;
669                 U32 u;
670                 for (u=0; u<nbSeq; u++) offsetcodeCount[codePtr[u]]++;
671             }
672 
673             {   const BYTE* codePtr = seqStorePtr->mlCode;
674                 U32 u;
675                 for (u=0; u<nbSeq; u++) matchlengthCount[codePtr[u]]++;
676             }
677 
678             {   const BYTE* codePtr = seqStorePtr->llCode;
679                 U32 u;
680                 for (u=0; u<nbSeq; u++) litlengthCount[codePtr[u]]++;
681             }
682 
683             if (nbSeq >= 2) { /* rep offsets */
684                 const seqDef* const seq = seqStorePtr->sequencesStart;
685                 U32 offset1 = seq[0].offBase - ZSTD_REP_NUM;
686                 U32 offset2 = seq[1].offBase - ZSTD_REP_NUM;
687                 if (offset1 >= MAXREPOFFSET) offset1 = 0;
688                 if (offset2 >= MAXREPOFFSET) offset2 = 0;
689                 repOffsets[offset1] += 3;
690                 repOffsets[offset2] += 1;
691     }   }   }
692 }
693 
694 static size_t ZDICT_totalSampleSize(const size_t* fileSizes, unsigned nbFiles)
695 {
696     size_t total=0;
697     unsigned u;
698     for (u=0; u<nbFiles; u++) total += fileSizes[u];
699     return total;
700 }
701 
702 typedef struct { U32 offset; U32 count; } offsetCount_t;
703 
704 static void ZDICT_insertSortCount(offsetCount_t table[ZSTD_REP_NUM+1], U32 val, U32 count)
705 {
706     U32 u;
707     table[ZSTD_REP_NUM].offset = val;
708     table[ZSTD_REP_NUM].count = count;
709     for (u=ZSTD_REP_NUM; u>0; u--) {
710         offsetCount_t tmp;
711         if (table[u-1].count >= table[u].count) break;
712         tmp = table[u-1];
713         table[u-1] = table[u];
714         table[u] = tmp;
715     }
716 }
717 
718 /* ZDICT_flatLit() :
719  * rewrite `countLit` to contain a mostly flat but still compressible distribution of literals.
720  * necessary to avoid generating a non-compressible distribution that HUF_writeCTable() cannot encode.
721  */
722 static void ZDICT_flatLit(unsigned* countLit)
723 {
724     int u;
725     for (u=1; u<256; u++) countLit[u] = 2;
726     countLit[0]   = 4;
727     countLit[253] = 1;
728     countLit[254] = 1;
729 }
730 
731 #define OFFCODE_MAX 30  /* only applicable to first block */
732 static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
733                                    int compressionLevel,
734                              const void*  srcBuffer, const size_t* fileSizes, unsigned nbFiles,
735                              const void* dictBuffer, size_t  dictBufferSize,
736                                    unsigned notificationLevel)
737 {
738     unsigned countLit[256];
739     HUF_CREATE_STATIC_CTABLE(hufTable, 255);
740     unsigned offcodeCount[OFFCODE_MAX+1];
741     short offcodeNCount[OFFCODE_MAX+1];
742     U32 offcodeMax = ZSTD_highbit32((U32)(dictBufferSize + 128 KB));
743     unsigned matchLengthCount[MaxML+1];
744     short matchLengthNCount[MaxML+1];
745     unsigned litLengthCount[MaxLL+1];
746     short litLengthNCount[MaxLL+1];
747     U32 repOffset[MAXREPOFFSET];
748     offsetCount_t bestRepOffset[ZSTD_REP_NUM+1];
749     EStats_ress_t esr = { NULL, NULL, NULL };
750     ZSTD_parameters params;
751     U32 u, huffLog = 11, Offlog = OffFSELog, mlLog = MLFSELog, llLog = LLFSELog, total;
752     size_t pos = 0, errorCode;
753     size_t eSize = 0;
754     size_t const totalSrcSize = ZDICT_totalSampleSize(fileSizes, nbFiles);
755     size_t const averageSampleSize = totalSrcSize / (nbFiles + !nbFiles);
756     BYTE* dstPtr = (BYTE*)dstBuffer;
757 
758     /* init */
759     DEBUGLOG(4, "ZDICT_analyzeEntropy");
760     if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionaryCreation_failed); goto _cleanup; }   /* too large dictionary */
761     for (u=0; u<256; u++) countLit[u] = 1;   /* any character must be described */
762     for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1;
763     for (u=0; u<=MaxML; u++) matchLengthCount[u] = 1;
764     for (u=0; u<=MaxLL; u++) litLengthCount[u] = 1;
765     memset(repOffset, 0, sizeof(repOffset));
766     repOffset[1] = repOffset[4] = repOffset[8] = 1;
767     memset(bestRepOffset, 0, sizeof(bestRepOffset));
768     if (compressionLevel==0) compressionLevel = ZSTD_CLEVEL_DEFAULT;
769     params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
770 
771     esr.dict = ZSTD_createCDict_advanced(dictBuffer, dictBufferSize, ZSTD_dlm_byRef, ZSTD_dct_rawContent, params.cParams, ZSTD_defaultCMem);
772     esr.zc = ZSTD_createCCtx();
773     esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX);
774     if (!esr.dict || !esr.zc || !esr.workPlace) {
775         eSize = ERROR(memory_allocation);
776         DISPLAYLEVEL(1, "Not enough memory \n");
777         goto _cleanup;
778     }
779 
780     /* collect stats on all samples */
781     for (u=0; u<nbFiles; u++) {
782         ZDICT_countEStats(esr, &params,
783                           countLit, offcodeCount, matchLengthCount, litLengthCount, repOffset,
784                          (const char*)srcBuffer + pos, fileSizes[u],
785                           notificationLevel);
786         pos += fileSizes[u];
787     }
788 
789     if (notificationLevel >= 4) {
790         /* writeStats */
791         DISPLAYLEVEL(4, "Offset Code Frequencies : \n");
792         for (u=0; u<=offcodeMax; u++) {
793             DISPLAYLEVEL(4, "%2u :%7u \n", u, offcodeCount[u]);
794     }   }
795 
796     /* analyze, build stats, starting with literals */
797     {   size_t maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
798         if (HUF_isError(maxNbBits)) {
799             eSize = maxNbBits;
800             DISPLAYLEVEL(1, " HUF_buildCTable error \n");
801             goto _cleanup;
802         }
803         if (maxNbBits==8) {  /* not compressible : will fail on HUF_writeCTable() */
804             DISPLAYLEVEL(2, "warning : pathological dataset : literals are not compressible : samples are noisy or too regular \n");
805             ZDICT_flatLit(countLit);  /* replace distribution by a fake "mostly flat but still compressible" distribution, that HUF_writeCTable() can encode */
806             maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
807             assert(maxNbBits==9);
808         }
809         huffLog = (U32)maxNbBits;
810     }
811 
812     /* looking for most common first offsets */
813     {   U32 offset;
814         for (offset=1; offset<MAXREPOFFSET; offset++)
815             ZDICT_insertSortCount(bestRepOffset, offset, repOffset[offset]);
816     }
817     /* note : the result of this phase should be used to better appreciate the impact on statistics */
818 
819     total=0; for (u=0; u<=offcodeMax; u++) total+=offcodeCount[u];
820     errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax, /* useLowProbCount */ 1);
821     if (FSE_isError(errorCode)) {
822         eSize = errorCode;
823         DISPLAYLEVEL(1, "FSE_normalizeCount error with offcodeCount \n");
824         goto _cleanup;
825     }
826     Offlog = (U32)errorCode;
827 
828     total=0; for (u=0; u<=MaxML; u++) total+=matchLengthCount[u];
829     errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML, /* useLowProbCount */ 1);
830     if (FSE_isError(errorCode)) {
831         eSize = errorCode;
832         DISPLAYLEVEL(1, "FSE_normalizeCount error with matchLengthCount \n");
833         goto _cleanup;
834     }
835     mlLog = (U32)errorCode;
836 
837     total=0; for (u=0; u<=MaxLL; u++) total+=litLengthCount[u];
838     errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL, /* useLowProbCount */ 1);
839     if (FSE_isError(errorCode)) {
840         eSize = errorCode;
841         DISPLAYLEVEL(1, "FSE_normalizeCount error with litLengthCount \n");
842         goto _cleanup;
843     }
844     llLog = (U32)errorCode;
845 
846     /* write result to buffer */
847     {   size_t const hhSize = HUF_writeCTable(dstPtr, maxDstSize, hufTable, 255, huffLog);
848         if (HUF_isError(hhSize)) {
849             eSize = hhSize;
850             DISPLAYLEVEL(1, "HUF_writeCTable error \n");
851             goto _cleanup;
852         }
853         dstPtr += hhSize;
854         maxDstSize -= hhSize;
855         eSize += hhSize;
856     }
857 
858     {   size_t const ohSize = FSE_writeNCount(dstPtr, maxDstSize, offcodeNCount, OFFCODE_MAX, Offlog);
859         if (FSE_isError(ohSize)) {
860             eSize = ohSize;
861             DISPLAYLEVEL(1, "FSE_writeNCount error with offcodeNCount \n");
862             goto _cleanup;
863         }
864         dstPtr += ohSize;
865         maxDstSize -= ohSize;
866         eSize += ohSize;
867     }
868 
869     {   size_t const mhSize = FSE_writeNCount(dstPtr, maxDstSize, matchLengthNCount, MaxML, mlLog);
870         if (FSE_isError(mhSize)) {
871             eSize = mhSize;
872             DISPLAYLEVEL(1, "FSE_writeNCount error with matchLengthNCount \n");
873             goto _cleanup;
874         }
875         dstPtr += mhSize;
876         maxDstSize -= mhSize;
877         eSize += mhSize;
878     }
879 
880     {   size_t const lhSize = FSE_writeNCount(dstPtr, maxDstSize, litLengthNCount, MaxLL, llLog);
881         if (FSE_isError(lhSize)) {
882             eSize = lhSize;
883             DISPLAYLEVEL(1, "FSE_writeNCount error with litlengthNCount \n");
884             goto _cleanup;
885         }
886         dstPtr += lhSize;
887         maxDstSize -= lhSize;
888         eSize += lhSize;
889     }
890 
891     if (maxDstSize<12) {
892         eSize = ERROR(dstSize_tooSmall);
893         DISPLAYLEVEL(1, "not enough space to write RepOffsets \n");
894         goto _cleanup;
895     }
896 # if 0
897     MEM_writeLE32(dstPtr+0, bestRepOffset[0].offset);
898     MEM_writeLE32(dstPtr+4, bestRepOffset[1].offset);
899     MEM_writeLE32(dstPtr+8, bestRepOffset[2].offset);
900 #else
901     /* at this stage, we don't use the result of "most common first offset",
902      * as the impact of statistics is not properly evaluated */
903     MEM_writeLE32(dstPtr+0, repStartValue[0]);
904     MEM_writeLE32(dstPtr+4, repStartValue[1]);
905     MEM_writeLE32(dstPtr+8, repStartValue[2]);
906 #endif
907     eSize += 12;
908 
909 _cleanup:
910     ZSTD_freeCDict(esr.dict);
911     ZSTD_freeCCtx(esr.zc);
912     free(esr.workPlace);
913 
914     return eSize;
915 }
916 
917 
918 /**
919  * @returns the maximum repcode value
920  */
921 static U32 ZDICT_maxRep(U32 const reps[ZSTD_REP_NUM])
922 {
923     U32 maxRep = reps[0];
924     int r;
925     for (r = 1; r < ZSTD_REP_NUM; ++r)
926         maxRep = MAX(maxRep, reps[r]);
927     return maxRep;
928 }
929 
930 size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
931                           const void* customDictContent, size_t dictContentSize,
932                           const void* samplesBuffer, const size_t* samplesSizes,
933                           unsigned nbSamples, ZDICT_params_t params)
934 {
935     size_t hSize;
936 #define HBUFFSIZE 256   /* should prove large enough for all entropy headers */
937     BYTE header[HBUFFSIZE];
938     int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
939     U32 const notificationLevel = params.notificationLevel;
940     /* The final dictionary content must be at least as large as the largest repcode */
941     size_t const minContentSize = (size_t)ZDICT_maxRep(repStartValue);
942     size_t paddingSize;
943 
944     /* check conditions */
945     DEBUGLOG(4, "ZDICT_finalizeDictionary");
946     if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
947     if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
948 
949     /* dictionary header */
950     MEM_writeLE32(header, ZSTD_MAGIC_DICTIONARY);
951     {   U64 const randomID = XXH64(customDictContent, dictContentSize, 0);
952         U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
953         U32 const dictID = params.dictID ? params.dictID : compliantID;
954         MEM_writeLE32(header+4, dictID);
955     }
956     hSize = 8;
957 
958     /* entropy tables */
959     DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
960     DISPLAYLEVEL(2, "statistics ... \n");
961     {   size_t const eSize = ZDICT_analyzeEntropy(header+hSize, HBUFFSIZE-hSize,
962                                   compressionLevel,
963                                   samplesBuffer, samplesSizes, nbSamples,
964                                   customDictContent, dictContentSize,
965                                   notificationLevel);
966         if (ZDICT_isError(eSize)) return eSize;
967         hSize += eSize;
968     }
969 
970     /* Shrink the content size if it doesn't fit in the buffer */
971     if (hSize + dictContentSize > dictBufferCapacity) {
972         dictContentSize = dictBufferCapacity - hSize;
973     }
974 
975     /* Pad the dictionary content with zeros if it is too small */
976     if (dictContentSize < minContentSize) {
977         RETURN_ERROR_IF(hSize + minContentSize > dictBufferCapacity, dstSize_tooSmall,
978                         "dictBufferCapacity too small to fit max repcode");
979         paddingSize = minContentSize - dictContentSize;
980     } else {
981         paddingSize = 0;
982     }
983 
984     {
985         size_t const dictSize = hSize + paddingSize + dictContentSize;
986 
987         /* The dictionary consists of the header, optional padding, and the content.
988          * The padding comes before the content because the "best" position in the
989          * dictionary is the last byte.
990          */
991         BYTE* const outDictHeader = (BYTE*)dictBuffer;
992         BYTE* const outDictPadding = outDictHeader + hSize;
993         BYTE* const outDictContent = outDictPadding + paddingSize;
994 
995         assert(dictSize <= dictBufferCapacity);
996         assert(outDictContent + dictContentSize == (BYTE*)dictBuffer + dictSize);
997 
998         /* First copy the customDictContent into its final location.
999          * `customDictContent` and `dictBuffer` may overlap, so we must
1000          * do this before any other writes into the output buffer.
1001          * Then copy the header & padding into the output buffer.
1002          */
1003         memmove(outDictContent, customDictContent, dictContentSize);
1004         memcpy(outDictHeader, header, hSize);
1005         memset(outDictPadding, 0, paddingSize);
1006 
1007         return dictSize;
1008     }
1009 }
1010 
1011 
1012 static size_t ZDICT_addEntropyTablesFromBuffer_advanced(
1013         void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
1014         const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
1015         ZDICT_params_t params)
1016 {
1017     int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
1018     U32 const notificationLevel = params.notificationLevel;
1019     size_t hSize = 8;
1020 
1021     /* calculate entropy tables */
1022     DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
1023     DISPLAYLEVEL(2, "statistics ... \n");
1024     {   size_t const eSize = ZDICT_analyzeEntropy((char*)dictBuffer+hSize, dictBufferCapacity-hSize,
1025                                   compressionLevel,
1026                                   samplesBuffer, samplesSizes, nbSamples,
1027                                   (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize,
1028                                   notificationLevel);
1029         if (ZDICT_isError(eSize)) return eSize;
1030         hSize += eSize;
1031     }
1032 
1033     /* add dictionary header (after entropy tables) */
1034     MEM_writeLE32(dictBuffer, ZSTD_MAGIC_DICTIONARY);
1035     {   U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
1036         U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
1037         U32 const dictID = params.dictID ? params.dictID : compliantID;
1038         MEM_writeLE32((char*)dictBuffer+4, dictID);
1039     }
1040 
1041     if (hSize + dictContentSize < dictBufferCapacity)
1042         memmove((char*)dictBuffer + hSize, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
1043     return MIN(dictBufferCapacity, hSize+dictContentSize);
1044 }
1045 
1046 /*! ZDICT_trainFromBuffer_unsafe_legacy() :
1047 *   Warning : `samplesBuffer` must be followed by noisy guard band !!!
1048 *   @return : size of dictionary, or an error code which can be tested with ZDICT_isError()
1049 */
1050 /* Begin FreeBSD - This symbol is needed by dll-linked CLI zstd(1). */
1051 ZSTDLIB_API
1052 /* End FreeBSD */
1053 static size_t ZDICT_trainFromBuffer_unsafe_legacy(
1054                             void* dictBuffer, size_t maxDictSize,
1055                             const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
1056                             ZDICT_legacy_params_t params)
1057 {
1058     U32 const dictListSize = MAX(MAX(DICTLISTSIZE_DEFAULT, nbSamples), (U32)(maxDictSize/16));
1059     dictItem* const dictList = (dictItem*)malloc(dictListSize * sizeof(*dictList));
1060     unsigned const selectivity = params.selectivityLevel == 0 ? g_selectivity_default : params.selectivityLevel;
1061     unsigned const minRep = (selectivity > 30) ? MINRATIO : nbSamples >> selectivity;
1062     size_t const targetDictSize = maxDictSize;
1063     size_t const samplesBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
1064     size_t dictSize = 0;
1065     U32 const notificationLevel = params.zParams.notificationLevel;
1066 
1067     /* checks */
1068     if (!dictList) return ERROR(memory_allocation);
1069     if (maxDictSize < ZDICT_DICTSIZE_MIN) { free(dictList); return ERROR(dstSize_tooSmall); }   /* requested dictionary size is too small */
1070     if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* not enough source to create dictionary */
1071 
1072     /* init */
1073     ZDICT_initDictItem(dictList);
1074 
1075     /* build dictionary */
1076     ZDICT_trainBuffer_legacy(dictList, dictListSize,
1077                        samplesBuffer, samplesBuffSize,
1078                        samplesSizes, nbSamples,
1079                        minRep, notificationLevel);
1080 
1081     /* display best matches */
1082     if (params.zParams.notificationLevel>= 3) {
1083         unsigned const nb = MIN(25, dictList[0].pos);
1084         unsigned const dictContentSize = ZDICT_dictSize(dictList);
1085         unsigned u;
1086         DISPLAYLEVEL(3, "\n %u segments found, of total size %u \n", (unsigned)dictList[0].pos-1, dictContentSize);
1087         DISPLAYLEVEL(3, "list %u best segments \n", nb-1);
1088         for (u=1; u<nb; u++) {
1089             unsigned const pos = dictList[u].pos;
1090             unsigned const length = dictList[u].length;
1091             U32 const printedLength = MIN(40, length);
1092             if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize)) {
1093                 free(dictList);
1094                 return ERROR(GENERIC);   /* should never happen */
1095             }
1096             DISPLAYLEVEL(3, "%3u:%3u bytes at pos %8u, savings %7u bytes |",
1097                          u, length, pos, (unsigned)dictList[u].savings);
1098             ZDICT_printHex((const char*)samplesBuffer+pos, printedLength);
1099             DISPLAYLEVEL(3, "| \n");
1100     }   }
1101 
1102 
1103     /* create dictionary */
1104     {   unsigned dictContentSize = ZDICT_dictSize(dictList);
1105         if (dictContentSize < ZDICT_CONTENTSIZE_MIN) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* dictionary content too small */
1106         if (dictContentSize < targetDictSize/4) {
1107             DISPLAYLEVEL(2, "!  warning : selected content significantly smaller than requested (%u < %u) \n", dictContentSize, (unsigned)maxDictSize);
1108             if (samplesBuffSize < 10 * targetDictSize)
1109                 DISPLAYLEVEL(2, "!  consider increasing the number of samples (total size : %u MB)\n", (unsigned)(samplesBuffSize>>20));
1110             if (minRep > MINRATIO) {
1111                 DISPLAYLEVEL(2, "!  consider increasing selectivity to produce larger dictionary (-s%u) \n", selectivity+1);
1112                 DISPLAYLEVEL(2, "!  note : larger dictionaries are not necessarily better, test its efficiency on samples \n");
1113             }
1114         }
1115 
1116         if ((dictContentSize > targetDictSize*3) && (nbSamples > 2*MINRATIO) && (selectivity>1)) {
1117             unsigned proposedSelectivity = selectivity-1;
1118             while ((nbSamples >> proposedSelectivity) <= MINRATIO) { proposedSelectivity--; }
1119             DISPLAYLEVEL(2, "!  note : calculated dictionary significantly larger than requested (%u > %u) \n", dictContentSize, (unsigned)maxDictSize);
1120             DISPLAYLEVEL(2, "!  consider increasing dictionary size, or produce denser dictionary (-s%u) \n", proposedSelectivity);
1121             DISPLAYLEVEL(2, "!  always test dictionary efficiency on real samples \n");
1122         }
1123 
1124         /* limit dictionary size */
1125         {   U32 const max = dictList->pos;   /* convention : nb of useful elts within dictList */
1126             U32 currentSize = 0;
1127             U32 n; for (n=1; n<max; n++) {
1128                 currentSize += dictList[n].length;
1129                 if (currentSize > targetDictSize) { currentSize -= dictList[n].length; break; }
1130             }
1131             dictList->pos = n;
1132             dictContentSize = currentSize;
1133         }
1134 
1135         /* build dict content */
1136         {   U32 u;
1137             BYTE* ptr = (BYTE*)dictBuffer + maxDictSize;
1138             for (u=1; u<dictList->pos; u++) {
1139                 U32 l = dictList[u].length;
1140                 ptr -= l;
1141                 if (ptr<(BYTE*)dictBuffer) { free(dictList); return ERROR(GENERIC); }   /* should not happen */
1142                 memcpy(ptr, (const char*)samplesBuffer+dictList[u].pos, l);
1143         }   }
1144 
1145         dictSize = ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, maxDictSize,
1146                                                              samplesBuffer, samplesSizes, nbSamples,
1147                                                              params.zParams);
1148     }
1149 
1150     /* clean up */
1151     free(dictList);
1152     return dictSize;
1153 }
1154 
1155 
1156 /* ZDICT_trainFromBuffer_legacy() :
1157  * issue : samplesBuffer need to be followed by a noisy guard band.
1158  * work around : duplicate the buffer, and add the noise */
1159 size_t ZDICT_trainFromBuffer_legacy(void* dictBuffer, size_t dictBufferCapacity,
1160                               const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
1161                               ZDICT_legacy_params_t params)
1162 {
1163     size_t result;
1164     void* newBuff;
1165     size_t const sBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
1166     if (sBuffSize < ZDICT_MIN_SAMPLES_SIZE) return 0;   /* not enough content => no dictionary */
1167 
1168     newBuff = malloc(sBuffSize + NOISELENGTH);
1169     if (!newBuff) return ERROR(memory_allocation);
1170 
1171     memcpy(newBuff, samplesBuffer, sBuffSize);
1172     ZDICT_fillNoise((char*)newBuff + sBuffSize, NOISELENGTH);   /* guard band, for end of buffer condition */
1173 
1174     result =
1175         ZDICT_trainFromBuffer_unsafe_legacy(dictBuffer, dictBufferCapacity, newBuff,
1176                                             samplesSizes, nbSamples, params);
1177     free(newBuff);
1178     return result;
1179 }
1180 
1181 
1182 size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
1183                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
1184 {
1185     ZDICT_fastCover_params_t params;
1186     DEBUGLOG(3, "ZDICT_trainFromBuffer");
1187     memset(&params, 0, sizeof(params));
1188     params.d = 8;
1189     params.steps = 4;
1190     /* Use default level since no compression level information is available */
1191     params.zParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
1192 #if defined(DEBUGLEVEL) && (DEBUGLEVEL>=1)
1193     params.zParams.notificationLevel = DEBUGLEVEL;
1194 #endif
1195     return ZDICT_optimizeTrainFromBuffer_fastCover(dictBuffer, dictBufferCapacity,
1196                                                samplesBuffer, samplesSizes, nbSamples,
1197                                                &params);
1198 }
1199 
1200 size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
1201                                   const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
1202 {
1203     ZDICT_params_t params;
1204     memset(&params, 0, sizeof(params));
1205     return ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, dictBufferCapacity,
1206                                                      samplesBuffer, samplesSizes, nbSamples,
1207                                                      params);
1208 }
1209