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 /* zstd_decompress_block :
12  * this module takes care of decompressing _compressed_ block */
13 
14 /*-*******************************************************
15 *  Dependencies
16 *********************************************************/
17 #include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
18 #include "../common/compiler.h"    /* prefetch */
19 #include "../common/cpu.h"         /* bmi2 */
20 #include "../common/mem.h"         /* low level memory routines */
21 #define FSE_STATIC_LINKING_ONLY
22 #include "../common/fse.h"
23 #define HUF_STATIC_LINKING_ONLY
24 #include "../common/huf.h"
25 #include "../common/zstd_internal.h"
26 #include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
27 #include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
28 #include "zstd_decompress_block.h"
29 
30 /*_*******************************************************
31 *  Macros
32 **********************************************************/
33 
34 /* These two optional macros force the use one way or another of the two
35  * ZSTD_decompressSequences implementations. You can't force in both directions
36  * at the same time.
37  */
38 #if defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
39     defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
40 #error "Cannot force the use of the short and the long ZSTD_decompressSequences variants!"
41 #endif
42 
43 
44 /*_*******************************************************
45 *  Memory operations
46 **********************************************************/
47 static void ZSTD_copy4(void* dst, const void* src) { ZSTD_memcpy(dst, src, 4); }
48 
49 
50 /*-*************************************************************
51  *   Block decoding
52  ***************************************************************/
53 
54 /*! ZSTD_getcBlockSize() :
55  *  Provides the size of compressed block from block header `src` */
56 size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
57                           blockProperties_t* bpPtr)
58 {
59     RETURN_ERROR_IF(srcSize < ZSTD_blockHeaderSize, srcSize_wrong, "");
60 
61     {   U32 const cBlockHeader = MEM_readLE24(src);
62         U32 const cSize = cBlockHeader >> 3;
63         bpPtr->lastBlock = cBlockHeader & 1;
64         bpPtr->blockType = (blockType_e)((cBlockHeader >> 1) & 3);
65         bpPtr->origSize = cSize;   /* only useful for RLE */
66         if (bpPtr->blockType == bt_rle) return 1;
67         RETURN_ERROR_IF(bpPtr->blockType == bt_reserved, corruption_detected, "");
68         return cSize;
69     }
70 }
71 
72 /* Allocate buffer for literals, either overlapping current dst, or split between dst and litExtraBuffer, or stored entirely within litExtraBuffer */
73 static void ZSTD_allocateLiteralsBuffer(ZSTD_DCtx* dctx, void* const dst, const size_t dstCapacity, const size_t litSize,
74     const streaming_operation streaming, const size_t expectedWriteSize, const unsigned splitImmediately)
75 {
76     if (streaming == not_streaming && dstCapacity > ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH + litSize + WILDCOPY_OVERLENGTH)
77     {
78         /* room for litbuffer to fit without read faulting */
79         dctx->litBuffer = (BYTE*)dst + ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH;
80         dctx->litBufferEnd = dctx->litBuffer + litSize;
81         dctx->litBufferLocation = ZSTD_in_dst;
82     }
83     else if (litSize > ZSTD_LITBUFFEREXTRASIZE)
84     {
85         /* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
86         if (splitImmediately) {
87             /* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
88             dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
89             dctx->litBufferEnd = dctx->litBuffer + litSize - ZSTD_LITBUFFEREXTRASIZE;
90         }
91         else {
92             /* initially this will be stored entirely in dst during huffman decoding, it will partially shifted to litExtraBuffer after */
93             dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize;
94             dctx->litBufferEnd = (BYTE*)dst + expectedWriteSize;
95         }
96         dctx->litBufferLocation = ZSTD_split;
97     }
98     else
99     {
100         /* fits entirely within litExtraBuffer, so no split is necessary */
101         dctx->litBuffer = dctx->litExtraBuffer;
102         dctx->litBufferEnd = dctx->litBuffer + litSize;
103         dctx->litBufferLocation = ZSTD_not_in_dst;
104     }
105 }
106 
107 /* Hidden declaration for fullbench */
108 size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
109                           const void* src, size_t srcSize,
110                           void* dst, size_t dstCapacity, const streaming_operation streaming);
111 /*! ZSTD_decodeLiteralsBlock() :
112  * Where it is possible to do so without being stomped by the output during decompression, the literals block will be stored
113  * in the dstBuffer.  If there is room to do so, it will be stored in full in the excess dst space after where the current
114  * block will be output.  Otherwise it will be stored at the end of the current dst blockspace, with a small portion being
115  * stored in dctx->litExtraBuffer to help keep it "ahead" of the current output write.
116  *
117  * @return : nb of bytes read from src (< srcSize )
118  *  note : symbol not declared but exposed for fullbench */
119 size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
120                           const void* src, size_t srcSize,   /* note : srcSize < BLOCKSIZE */
121                           void* dst, size_t dstCapacity, const streaming_operation streaming)
122 {
123     DEBUGLOG(5, "ZSTD_decodeLiteralsBlock");
124     RETURN_ERROR_IF(srcSize < MIN_CBLOCK_SIZE, corruption_detected, "");
125 
126     {   const BYTE* const istart = (const BYTE*) src;
127         symbolEncodingType_e const litEncType = (symbolEncodingType_e)(istart[0] & 3);
128 
129         switch(litEncType)
130         {
131         case set_repeat:
132             DEBUGLOG(5, "set_repeat flag : re-using stats from previous compressed literals block");
133             RETURN_ERROR_IF(dctx->litEntropy==0, dictionary_corrupted, "");
134             ZSTD_FALLTHROUGH;
135 
136         case set_compressed:
137             RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3");
138             {   size_t lhSize, litSize, litCSize;
139                 U32 singleStream=0;
140                 U32 const lhlCode = (istart[0] >> 2) & 3;
141                 U32 const lhc = MEM_readLE32(istart);
142                 size_t hufSuccess;
143                 size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
144                 switch(lhlCode)
145                 {
146                 case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
147                     /* 2 - 2 - 10 - 10 */
148                     singleStream = !lhlCode;
149                     lhSize = 3;
150                     litSize  = (lhc >> 4) & 0x3FF;
151                     litCSize = (lhc >> 14) & 0x3FF;
152                     break;
153                 case 2:
154                     /* 2 - 2 - 14 - 14 */
155                     lhSize = 4;
156                     litSize  = (lhc >> 4) & 0x3FFF;
157                     litCSize = lhc >> 18;
158                     break;
159                 case 3:
160                     /* 2 - 2 - 18 - 18 */
161                     lhSize = 5;
162                     litSize  = (lhc >> 4) & 0x3FFFF;
163                     litCSize = (lhc >> 22) + ((size_t)istart[4] << 10);
164                     break;
165                 }
166                 RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
167                 RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
168                 RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected, "");
169                 RETURN_ERROR_IF(expectedWriteSize < litSize , dstSize_tooSmall, "");
170                 ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 0);
171 
172                 /* prefetch huffman table if cold */
173                 if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) {
174                     PREFETCH_AREA(dctx->HUFptr, sizeof(dctx->entropy.hufTable));
175                 }
176 
177                 if (litEncType==set_repeat) {
178                     if (singleStream) {
179                         hufSuccess = HUF_decompress1X_usingDTable_bmi2(
180                             dctx->litBuffer, litSize, istart+lhSize, litCSize,
181                             dctx->HUFptr, ZSTD_DCtx_get_bmi2(dctx));
182                     } else {
183                         hufSuccess = HUF_decompress4X_usingDTable_bmi2(
184                             dctx->litBuffer, litSize, istart+lhSize, litCSize,
185                             dctx->HUFptr, ZSTD_DCtx_get_bmi2(dctx));
186                     }
187                 } else {
188                     if (singleStream) {
189 #if defined(HUF_FORCE_DECOMPRESS_X2)
190                         hufSuccess = HUF_decompress1X_DCtx_wksp(
191                             dctx->entropy.hufTable, dctx->litBuffer, litSize,
192                             istart+lhSize, litCSize, dctx->workspace,
193                             sizeof(dctx->workspace));
194 #else
195                         hufSuccess = HUF_decompress1X1_DCtx_wksp_bmi2(
196                             dctx->entropy.hufTable, dctx->litBuffer, litSize,
197                             istart+lhSize, litCSize, dctx->workspace,
198                             sizeof(dctx->workspace), ZSTD_DCtx_get_bmi2(dctx));
199 #endif
200                     } else {
201                         hufSuccess = HUF_decompress4X_hufOnly_wksp_bmi2(
202                             dctx->entropy.hufTable, dctx->litBuffer, litSize,
203                             istart+lhSize, litCSize, dctx->workspace,
204                             sizeof(dctx->workspace), ZSTD_DCtx_get_bmi2(dctx));
205                     }
206                 }
207                 if (dctx->litBufferLocation == ZSTD_split)
208                 {
209                     ZSTD_memcpy(dctx->litExtraBuffer, dctx->litBufferEnd - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
210                     ZSTD_memmove(dctx->litBuffer + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH, dctx->litBuffer, litSize - ZSTD_LITBUFFEREXTRASIZE);
211                     dctx->litBuffer += ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
212                     dctx->litBufferEnd -= WILDCOPY_OVERLENGTH;
213                 }
214 
215                 RETURN_ERROR_IF(HUF_isError(hufSuccess), corruption_detected, "");
216 
217                 dctx->litPtr = dctx->litBuffer;
218                 dctx->litSize = litSize;
219                 dctx->litEntropy = 1;
220                 if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
221                 return litCSize + lhSize;
222             }
223 
224         case set_basic:
225             {   size_t litSize, lhSize;
226                 U32 const lhlCode = ((istart[0]) >> 2) & 3;
227                 size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
228                 switch(lhlCode)
229                 {
230                 case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
231                     lhSize = 1;
232                     litSize = istart[0] >> 3;
233                     break;
234                 case 1:
235                     lhSize = 2;
236                     litSize = MEM_readLE16(istart) >> 4;
237                     break;
238                 case 3:
239                     lhSize = 3;
240                     litSize = MEM_readLE24(istart) >> 4;
241                     break;
242                 }
243 
244                 RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
245                 RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
246                 ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
247                 if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
248                     RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected, "");
249                     if (dctx->litBufferLocation == ZSTD_split)
250                     {
251                         ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize - ZSTD_LITBUFFEREXTRASIZE);
252                         ZSTD_memcpy(dctx->litExtraBuffer, istart + lhSize + litSize - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
253                     }
254                     else
255                     {
256                         ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize);
257                     }
258                     dctx->litPtr = dctx->litBuffer;
259                     dctx->litSize = litSize;
260                     return lhSize+litSize;
261                 }
262                 /* direct reference into compressed stream */
263                 dctx->litPtr = istart+lhSize;
264                 dctx->litSize = litSize;
265                 dctx->litBufferEnd = dctx->litPtr + litSize;
266                 dctx->litBufferLocation = ZSTD_not_in_dst;
267                 return lhSize+litSize;
268             }
269 
270         case set_rle:
271             {   U32 const lhlCode = ((istart[0]) >> 2) & 3;
272                 size_t litSize, lhSize;
273                 size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
274                 switch(lhlCode)
275                 {
276                 case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
277                     lhSize = 1;
278                     litSize = istart[0] >> 3;
279                     break;
280                 case 1:
281                     lhSize = 2;
282                     litSize = MEM_readLE16(istart) >> 4;
283                     break;
284                 case 3:
285                     lhSize = 3;
286                     litSize = MEM_readLE24(istart) >> 4;
287                     RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4");
288                     break;
289                 }
290                 RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
291                 RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
292                 RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
293                 ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
294                 if (dctx->litBufferLocation == ZSTD_split)
295                 {
296                     ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize - ZSTD_LITBUFFEREXTRASIZE);
297                     ZSTD_memset(dctx->litExtraBuffer, istart[lhSize], ZSTD_LITBUFFEREXTRASIZE);
298                 }
299                 else
300                 {
301                     ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize);
302                 }
303                 dctx->litPtr = dctx->litBuffer;
304                 dctx->litSize = litSize;
305                 return lhSize+1;
306             }
307         default:
308             RETURN_ERROR(corruption_detected, "impossible");
309         }
310     }
311 }
312 
313 /* Default FSE distribution tables.
314  * These are pre-calculated FSE decoding tables using default distributions as defined in specification :
315  * https://github.com/facebook/zstd/blob/release/doc/zstd_compression_format.md#default-distributions
316  * They were generated programmatically with following method :
317  * - start from default distributions, present in /lib/common/zstd_internal.h
318  * - generate tables normally, using ZSTD_buildFSETable()
319  * - printout the content of tables
320  * - pretify output, report below, test with fuzzer to ensure it's correct */
321 
322 /* Default FSE distribution table for Literal Lengths */
323 static const ZSTD_seqSymbol LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = {
324      {  1,  1,  1, LL_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
325      /* nextState, nbAddBits, nbBits, baseVal */
326      {  0,  0,  4,    0},  { 16,  0,  4,    0},
327      { 32,  0,  5,    1},  {  0,  0,  5,    3},
328      {  0,  0,  5,    4},  {  0,  0,  5,    6},
329      {  0,  0,  5,    7},  {  0,  0,  5,    9},
330      {  0,  0,  5,   10},  {  0,  0,  5,   12},
331      {  0,  0,  6,   14},  {  0,  1,  5,   16},
332      {  0,  1,  5,   20},  {  0,  1,  5,   22},
333      {  0,  2,  5,   28},  {  0,  3,  5,   32},
334      {  0,  4,  5,   48},  { 32,  6,  5,   64},
335      {  0,  7,  5,  128},  {  0,  8,  6,  256},
336      {  0, 10,  6, 1024},  {  0, 12,  6, 4096},
337      { 32,  0,  4,    0},  {  0,  0,  4,    1},
338      {  0,  0,  5,    2},  { 32,  0,  5,    4},
339      {  0,  0,  5,    5},  { 32,  0,  5,    7},
340      {  0,  0,  5,    8},  { 32,  0,  5,   10},
341      {  0,  0,  5,   11},  {  0,  0,  6,   13},
342      { 32,  1,  5,   16},  {  0,  1,  5,   18},
343      { 32,  1,  5,   22},  {  0,  2,  5,   24},
344      { 32,  3,  5,   32},  {  0,  3,  5,   40},
345      {  0,  6,  4,   64},  { 16,  6,  4,   64},
346      { 32,  7,  5,  128},  {  0,  9,  6,  512},
347      {  0, 11,  6, 2048},  { 48,  0,  4,    0},
348      { 16,  0,  4,    1},  { 32,  0,  5,    2},
349      { 32,  0,  5,    3},  { 32,  0,  5,    5},
350      { 32,  0,  5,    6},  { 32,  0,  5,    8},
351      { 32,  0,  5,    9},  { 32,  0,  5,   11},
352      { 32,  0,  5,   12},  {  0,  0,  6,   15},
353      { 32,  1,  5,   18},  { 32,  1,  5,   20},
354      { 32,  2,  5,   24},  { 32,  2,  5,   28},
355      { 32,  3,  5,   40},  { 32,  4,  5,   48},
356      {  0, 16,  6,65536},  {  0, 15,  6,32768},
357      {  0, 14,  6,16384},  {  0, 13,  6, 8192},
358 };   /* LL_defaultDTable */
359 
360 /* Default FSE distribution table for Offset Codes */
361 static const ZSTD_seqSymbol OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = {
362     {  1,  1,  1, OF_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
363     /* nextState, nbAddBits, nbBits, baseVal */
364     {  0,  0,  5,    0},     {  0,  6,  4,   61},
365     {  0,  9,  5,  509},     {  0, 15,  5,32765},
366     {  0, 21,  5,2097149},   {  0,  3,  5,    5},
367     {  0,  7,  4,  125},     {  0, 12,  5, 4093},
368     {  0, 18,  5,262141},    {  0, 23,  5,8388605},
369     {  0,  5,  5,   29},     {  0,  8,  4,  253},
370     {  0, 14,  5,16381},     {  0, 20,  5,1048573},
371     {  0,  2,  5,    1},     { 16,  7,  4,  125},
372     {  0, 11,  5, 2045},     {  0, 17,  5,131069},
373     {  0, 22,  5,4194301},   {  0,  4,  5,   13},
374     { 16,  8,  4,  253},     {  0, 13,  5, 8189},
375     {  0, 19,  5,524285},    {  0,  1,  5,    1},
376     { 16,  6,  4,   61},     {  0, 10,  5, 1021},
377     {  0, 16,  5,65533},     {  0, 28,  5,268435453},
378     {  0, 27,  5,134217725}, {  0, 26,  5,67108861},
379     {  0, 25,  5,33554429},  {  0, 24,  5,16777213},
380 };   /* OF_defaultDTable */
381 
382 
383 /* Default FSE distribution table for Match Lengths */
384 static const ZSTD_seqSymbol ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = {
385     {  1,  1,  1, ML_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
386     /* nextState, nbAddBits, nbBits, baseVal */
387     {  0,  0,  6,    3},  {  0,  0,  4,    4},
388     { 32,  0,  5,    5},  {  0,  0,  5,    6},
389     {  0,  0,  5,    8},  {  0,  0,  5,    9},
390     {  0,  0,  5,   11},  {  0,  0,  6,   13},
391     {  0,  0,  6,   16},  {  0,  0,  6,   19},
392     {  0,  0,  6,   22},  {  0,  0,  6,   25},
393     {  0,  0,  6,   28},  {  0,  0,  6,   31},
394     {  0,  0,  6,   34},  {  0,  1,  6,   37},
395     {  0,  1,  6,   41},  {  0,  2,  6,   47},
396     {  0,  3,  6,   59},  {  0,  4,  6,   83},
397     {  0,  7,  6,  131},  {  0,  9,  6,  515},
398     { 16,  0,  4,    4},  {  0,  0,  4,    5},
399     { 32,  0,  5,    6},  {  0,  0,  5,    7},
400     { 32,  0,  5,    9},  {  0,  0,  5,   10},
401     {  0,  0,  6,   12},  {  0,  0,  6,   15},
402     {  0,  0,  6,   18},  {  0,  0,  6,   21},
403     {  0,  0,  6,   24},  {  0,  0,  6,   27},
404     {  0,  0,  6,   30},  {  0,  0,  6,   33},
405     {  0,  1,  6,   35},  {  0,  1,  6,   39},
406     {  0,  2,  6,   43},  {  0,  3,  6,   51},
407     {  0,  4,  6,   67},  {  0,  5,  6,   99},
408     {  0,  8,  6,  259},  { 32,  0,  4,    4},
409     { 48,  0,  4,    4},  { 16,  0,  4,    5},
410     { 32,  0,  5,    7},  { 32,  0,  5,    8},
411     { 32,  0,  5,   10},  { 32,  0,  5,   11},
412     {  0,  0,  6,   14},  {  0,  0,  6,   17},
413     {  0,  0,  6,   20},  {  0,  0,  6,   23},
414     {  0,  0,  6,   26},  {  0,  0,  6,   29},
415     {  0,  0,  6,   32},  {  0, 16,  6,65539},
416     {  0, 15,  6,32771},  {  0, 14,  6,16387},
417     {  0, 13,  6, 8195},  {  0, 12,  6, 4099},
418     {  0, 11,  6, 2051},  {  0, 10,  6, 1027},
419 };   /* ML_defaultDTable */
420 
421 
422 static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U8 nbAddBits)
423 {
424     void* ptr = dt;
425     ZSTD_seqSymbol_header* const DTableH = (ZSTD_seqSymbol_header*)ptr;
426     ZSTD_seqSymbol* const cell = dt + 1;
427 
428     DTableH->tableLog = 0;
429     DTableH->fastMode = 0;
430 
431     cell->nbBits = 0;
432     cell->nextState = 0;
433     assert(nbAddBits < 255);
434     cell->nbAdditionalBits = nbAddBits;
435     cell->baseValue = baseValue;
436 }
437 
438 
439 /* ZSTD_buildFSETable() :
440  * generate FSE decoding table for one symbol (ll, ml or off)
441  * cannot fail if input is valid =>
442  * all inputs are presumed validated at this stage */
443 FORCE_INLINE_TEMPLATE
444 void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
445             const short* normalizedCounter, unsigned maxSymbolValue,
446             const U32* baseValue, const U8* nbAdditionalBits,
447             unsigned tableLog, void* wksp, size_t wkspSize)
448 {
449     ZSTD_seqSymbol* const tableDecode = dt+1;
450     U32 const maxSV1 = maxSymbolValue + 1;
451     U32 const tableSize = 1 << tableLog;
452 
453     U16* symbolNext = (U16*)wksp;
454     BYTE* spread = (BYTE*)(symbolNext + MaxSeq + 1);
455     U32 highThreshold = tableSize - 1;
456 
457 
458     /* Sanity Checks */
459     assert(maxSymbolValue <= MaxSeq);
460     assert(tableLog <= MaxFSELog);
461     assert(wkspSize >= ZSTD_BUILD_FSE_TABLE_WKSP_SIZE);
462     (void)wkspSize;
463     /* Init, lay down lowprob symbols */
464     {   ZSTD_seqSymbol_header DTableH;
465         DTableH.tableLog = tableLog;
466         DTableH.fastMode = 1;
467         {   S16 const largeLimit= (S16)(1 << (tableLog-1));
468             U32 s;
469             for (s=0; s<maxSV1; s++) {
470                 if (normalizedCounter[s]==-1) {
471                     tableDecode[highThreshold--].baseValue = s;
472                     symbolNext[s] = 1;
473                 } else {
474                     if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
475                     assert(normalizedCounter[s]>=0);
476                     symbolNext[s] = (U16)normalizedCounter[s];
477         }   }   }
478         ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
479     }
480 
481     /* Spread symbols */
482     assert(tableSize <= 512);
483     /* Specialized symbol spreading for the case when there are
484      * no low probability (-1 count) symbols. When compressing
485      * small blocks we avoid low probability symbols to hit this
486      * case, since header decoding speed matters more.
487      */
488     if (highThreshold == tableSize - 1) {
489         size_t const tableMask = tableSize-1;
490         size_t const step = FSE_TABLESTEP(tableSize);
491         /* First lay down the symbols in order.
492          * We use a uint64_t to lay down 8 bytes at a time. This reduces branch
493          * misses since small blocks generally have small table logs, so nearly
494          * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
495          * our buffer to handle the over-write.
496          */
497         {
498             U64 const add = 0x0101010101010101ull;
499             size_t pos = 0;
500             U64 sv = 0;
501             U32 s;
502             for (s=0; s<maxSV1; ++s, sv += add) {
503                 int i;
504                 int const n = normalizedCounter[s];
505                 MEM_write64(spread + pos, sv);
506                 for (i = 8; i < n; i += 8) {
507                     MEM_write64(spread + pos + i, sv);
508                 }
509                 pos += n;
510             }
511         }
512         /* Now we spread those positions across the table.
513          * The benefit of doing it in two stages is that we avoid the the
514          * variable size inner loop, which caused lots of branch misses.
515          * Now we can run through all the positions without any branch misses.
516          * We unroll the loop twice, since that is what emperically worked best.
517          */
518         {
519             size_t position = 0;
520             size_t s;
521             size_t const unroll = 2;
522             assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
523             for (s = 0; s < (size_t)tableSize; s += unroll) {
524                 size_t u;
525                 for (u = 0; u < unroll; ++u) {
526                     size_t const uPosition = (position + (u * step)) & tableMask;
527                     tableDecode[uPosition].baseValue = spread[s + u];
528                 }
529                 position = (position + (unroll * step)) & tableMask;
530             }
531             assert(position == 0);
532         }
533     } else {
534         U32 const tableMask = tableSize-1;
535         U32 const step = FSE_TABLESTEP(tableSize);
536         U32 s, position = 0;
537         for (s=0; s<maxSV1; s++) {
538             int i;
539             int const n = normalizedCounter[s];
540             for (i=0; i<n; i++) {
541                 tableDecode[position].baseValue = s;
542                 position = (position + step) & tableMask;
543                 while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
544         }   }
545         assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
546     }
547 
548     /* Build Decoding table */
549     {
550         U32 u;
551         for (u=0; u<tableSize; u++) {
552             U32 const symbol = tableDecode[u].baseValue;
553             U32 const nextState = symbolNext[symbol]++;
554             tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
555             tableDecode[u].nextState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
556             assert(nbAdditionalBits[symbol] < 255);
557             tableDecode[u].nbAdditionalBits = nbAdditionalBits[symbol];
558             tableDecode[u].baseValue = baseValue[symbol];
559         }
560     }
561 }
562 
563 /* Avoids the FORCE_INLINE of the _body() function. */
564 static void ZSTD_buildFSETable_body_default(ZSTD_seqSymbol* dt,
565             const short* normalizedCounter, unsigned maxSymbolValue,
566             const U32* baseValue, const U8* nbAdditionalBits,
567             unsigned tableLog, void* wksp, size_t wkspSize)
568 {
569     ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
570             baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
571 }
572 
573 #if DYNAMIC_BMI2
574 BMI2_TARGET_ATTRIBUTE static void ZSTD_buildFSETable_body_bmi2(ZSTD_seqSymbol* dt,
575             const short* normalizedCounter, unsigned maxSymbolValue,
576             const U32* baseValue, const U8* nbAdditionalBits,
577             unsigned tableLog, void* wksp, size_t wkspSize)
578 {
579     ZSTD_buildFSETable_body(dt, normalizedCounter, maxSymbolValue,
580             baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
581 }
582 #endif
583 
584 void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
585             const short* normalizedCounter, unsigned maxSymbolValue,
586             const U32* baseValue, const U8* nbAdditionalBits,
587             unsigned tableLog, void* wksp, size_t wkspSize, int bmi2)
588 {
589 #if DYNAMIC_BMI2
590     if (bmi2) {
591         ZSTD_buildFSETable_body_bmi2(dt, normalizedCounter, maxSymbolValue,
592                 baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
593         return;
594     }
595 #endif
596     (void)bmi2;
597     ZSTD_buildFSETable_body_default(dt, normalizedCounter, maxSymbolValue,
598             baseValue, nbAdditionalBits, tableLog, wksp, wkspSize);
599 }
600 
601 
602 /*! ZSTD_buildSeqTable() :
603  * @return : nb bytes read from src,
604  *           or an error code if it fails */
605 static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymbol** DTablePtr,
606                                  symbolEncodingType_e type, unsigned max, U32 maxLog,
607                                  const void* src, size_t srcSize,
608                                  const U32* baseValue, const U8* nbAdditionalBits,
609                                  const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable,
610                                  int ddictIsCold, int nbSeq, U32* wksp, size_t wkspSize,
611                                  int bmi2)
612 {
613     switch(type)
614     {
615     case set_rle :
616         RETURN_ERROR_IF(!srcSize, srcSize_wrong, "");
617         RETURN_ERROR_IF((*(const BYTE*)src) > max, corruption_detected, "");
618         {   U32 const symbol = *(const BYTE*)src;
619             U32 const baseline = baseValue[symbol];
620             U8 const nbBits = nbAdditionalBits[symbol];
621             ZSTD_buildSeqTable_rle(DTableSpace, baseline, nbBits);
622         }
623         *DTablePtr = DTableSpace;
624         return 1;
625     case set_basic :
626         *DTablePtr = defaultTable;
627         return 0;
628     case set_repeat:
629         RETURN_ERROR_IF(!flagRepeatTable, corruption_detected, "");
630         /* prefetch FSE table if used */
631         if (ddictIsCold && (nbSeq > 24 /* heuristic */)) {
632             const void* const pStart = *DTablePtr;
633             size_t const pSize = sizeof(ZSTD_seqSymbol) * (SEQSYMBOL_TABLE_SIZE(maxLog));
634             PREFETCH_AREA(pStart, pSize);
635         }
636         return 0;
637     case set_compressed :
638         {   unsigned tableLog;
639             S16 norm[MaxSeq+1];
640             size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
641             RETURN_ERROR_IF(FSE_isError(headerSize), corruption_detected, "");
642             RETURN_ERROR_IF(tableLog > maxLog, corruption_detected, "");
643             ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog, wksp, wkspSize, bmi2);
644             *DTablePtr = DTableSpace;
645             return headerSize;
646         }
647     default :
648         assert(0);
649         RETURN_ERROR(GENERIC, "impossible");
650     }
651 }
652 
653 size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
654                              const void* src, size_t srcSize)
655 {
656     const BYTE* const istart = (const BYTE*)src;
657     const BYTE* const iend = istart + srcSize;
658     const BYTE* ip = istart;
659     int nbSeq;
660     DEBUGLOG(5, "ZSTD_decodeSeqHeaders");
661 
662     /* check */
663     RETURN_ERROR_IF(srcSize < MIN_SEQUENCES_SIZE, srcSize_wrong, "");
664 
665     /* SeqHead */
666     nbSeq = *ip++;
667     if (!nbSeq) {
668         *nbSeqPtr=0;
669         RETURN_ERROR_IF(srcSize != 1, srcSize_wrong, "");
670         return 1;
671     }
672     if (nbSeq > 0x7F) {
673         if (nbSeq == 0xFF) {
674             RETURN_ERROR_IF(ip+2 > iend, srcSize_wrong, "");
675             nbSeq = MEM_readLE16(ip) + LONGNBSEQ;
676             ip+=2;
677         } else {
678             RETURN_ERROR_IF(ip >= iend, srcSize_wrong, "");
679             nbSeq = ((nbSeq-0x80)<<8) + *ip++;
680         }
681     }
682     *nbSeqPtr = nbSeq;
683 
684     /* FSE table descriptors */
685     RETURN_ERROR_IF(ip+1 > iend, srcSize_wrong, ""); /* minimum possible size: 1 byte for symbol encoding types */
686     {   symbolEncodingType_e const LLtype = (symbolEncodingType_e)(*ip >> 6);
687         symbolEncodingType_e const OFtype = (symbolEncodingType_e)((*ip >> 4) & 3);
688         symbolEncodingType_e const MLtype = (symbolEncodingType_e)((*ip >> 2) & 3);
689         ip++;
690 
691         /* Build DTables */
692         {   size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr,
693                                                       LLtype, MaxLL, LLFSELog,
694                                                       ip, iend-ip,
695                                                       LL_base, LL_bits,
696                                                       LL_defaultDTable, dctx->fseEntropy,
697                                                       dctx->ddictIsCold, nbSeq,
698                                                       dctx->workspace, sizeof(dctx->workspace),
699                                                       ZSTD_DCtx_get_bmi2(dctx));
700             RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected, "ZSTD_buildSeqTable failed");
701             ip += llhSize;
702         }
703 
704         {   size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr,
705                                                       OFtype, MaxOff, OffFSELog,
706                                                       ip, iend-ip,
707                                                       OF_base, OF_bits,
708                                                       OF_defaultDTable, dctx->fseEntropy,
709                                                       dctx->ddictIsCold, nbSeq,
710                                                       dctx->workspace, sizeof(dctx->workspace),
711                                                       ZSTD_DCtx_get_bmi2(dctx));
712             RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected, "ZSTD_buildSeqTable failed");
713             ip += ofhSize;
714         }
715 
716         {   size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr,
717                                                       MLtype, MaxML, MLFSELog,
718                                                       ip, iend-ip,
719                                                       ML_base, ML_bits,
720                                                       ML_defaultDTable, dctx->fseEntropy,
721                                                       dctx->ddictIsCold, nbSeq,
722                                                       dctx->workspace, sizeof(dctx->workspace),
723                                                       ZSTD_DCtx_get_bmi2(dctx));
724             RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected, "ZSTD_buildSeqTable failed");
725             ip += mlhSize;
726         }
727     }
728 
729     return ip-istart;
730 }
731 
732 
733 typedef struct {
734     size_t litLength;
735     size_t matchLength;
736     size_t offset;
737 } seq_t;
738 
739 typedef struct {
740     size_t state;
741     const ZSTD_seqSymbol* table;
742 } ZSTD_fseState;
743 
744 typedef struct {
745     BIT_DStream_t DStream;
746     ZSTD_fseState stateLL;
747     ZSTD_fseState stateOffb;
748     ZSTD_fseState stateML;
749     size_t prevOffset[ZSTD_REP_NUM];
750 } seqState_t;
751 
752 /*! ZSTD_overlapCopy8() :
753  *  Copies 8 bytes from ip to op and updates op and ip where ip <= op.
754  *  If the offset is < 8 then the offset is spread to at least 8 bytes.
755  *
756  *  Precondition: *ip <= *op
757  *  Postcondition: *op - *op >= 8
758  */
759 HINT_INLINE void ZSTD_overlapCopy8(BYTE** op, BYTE const** ip, size_t offset) {
760     assert(*ip <= *op);
761     if (offset < 8) {
762         /* close range match, overlap */
763         static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
764         static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
765         int const sub2 = dec64table[offset];
766         (*op)[0] = (*ip)[0];
767         (*op)[1] = (*ip)[1];
768         (*op)[2] = (*ip)[2];
769         (*op)[3] = (*ip)[3];
770         *ip += dec32table[offset];
771         ZSTD_copy4(*op+4, *ip);
772         *ip -= sub2;
773     } else {
774         ZSTD_copy8(*op, *ip);
775     }
776     *ip += 8;
777     *op += 8;
778     assert(*op - *ip >= 8);
779 }
780 
781 /*! ZSTD_safecopy() :
782  *  Specialized version of memcpy() that is allowed to READ up to WILDCOPY_OVERLENGTH past the input buffer
783  *  and write up to 16 bytes past oend_w (op >= oend_w is allowed).
784  *  This function is only called in the uncommon case where the sequence is near the end of the block. It
785  *  should be fast for a single long sequence, but can be slow for several short sequences.
786  *
787  *  @param ovtype controls the overlap detection
788  *         - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
789  *         - ZSTD_overlap_src_before_dst: The src and dst may overlap and may be any distance apart.
790  *           The src buffer must be before the dst buffer.
791  */
792 static void ZSTD_safecopy(BYTE* op, const BYTE* const oend_w, BYTE const* ip, ptrdiff_t length, ZSTD_overlap_e ovtype) {
793     ptrdiff_t const diff = op - ip;
794     BYTE* const oend = op + length;
795 
796     assert((ovtype == ZSTD_no_overlap && (diff <= -8 || diff >= 8 || op >= oend_w)) ||
797            (ovtype == ZSTD_overlap_src_before_dst && diff >= 0));
798 
799     if (length < 8) {
800         /* Handle short lengths. */
801         while (op < oend) *op++ = *ip++;
802         return;
803     }
804     if (ovtype == ZSTD_overlap_src_before_dst) {
805         /* Copy 8 bytes and ensure the offset >= 8 when there can be overlap. */
806         assert(length >= 8);
807         ZSTD_overlapCopy8(&op, &ip, diff);
808         length -= 8;
809         assert(op - ip >= 8);
810         assert(op <= oend);
811     }
812 
813     if (oend <= oend_w) {
814         /* No risk of overwrite. */
815         ZSTD_wildcopy(op, ip, length, ovtype);
816         return;
817     }
818     if (op <= oend_w) {
819         /* Wildcopy until we get close to the end. */
820         assert(oend > oend_w);
821         ZSTD_wildcopy(op, ip, oend_w - op, ovtype);
822         ip += oend_w - op;
823         op += oend_w - op;
824     }
825     /* Handle the leftovers. */
826     while (op < oend) *op++ = *ip++;
827 }
828 
829 /* ZSTD_safecopyDstBeforeSrc():
830  * This version allows overlap with dst before src, or handles the non-overlap case with dst after src
831  * Kept separate from more common ZSTD_safecopy case to avoid performance impact to the safecopy common case */
832 static void ZSTD_safecopyDstBeforeSrc(BYTE* op, BYTE const* ip, ptrdiff_t length) {
833     ptrdiff_t const diff = op - ip;
834     BYTE* const oend = op + length;
835 
836     if (length < 8 || diff > -8) {
837         /* Handle short lengths, close overlaps, and dst not before src. */
838         while (op < oend) *op++ = *ip++;
839         return;
840     }
841 
842     if (op <= oend - WILDCOPY_OVERLENGTH && diff < -WILDCOPY_VECLEN) {
843         ZSTD_wildcopy(op, ip, oend - WILDCOPY_OVERLENGTH - op, ZSTD_no_overlap);
844         ip += oend - WILDCOPY_OVERLENGTH - op;
845         op += oend - WILDCOPY_OVERLENGTH - op;
846     }
847 
848     /* Handle the leftovers. */
849     while (op < oend) *op++ = *ip++;
850 }
851 
852 /* ZSTD_execSequenceEnd():
853  * This version handles cases that are near the end of the output buffer. It requires
854  * more careful checks to make sure there is no overflow. By separating out these hard
855  * and unlikely cases, we can speed up the common cases.
856  *
857  * NOTE: This function needs to be fast for a single long sequence, but doesn't need
858  * to be optimized for many small sequences, since those fall into ZSTD_execSequence().
859  */
860 FORCE_NOINLINE
861 size_t ZSTD_execSequenceEnd(BYTE* op,
862     BYTE* const oend, seq_t sequence,
863     const BYTE** litPtr, const BYTE* const litLimit,
864     const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
865 {
866     BYTE* const oLitEnd = op + sequence.litLength;
867     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
868     const BYTE* const iLitEnd = *litPtr + sequence.litLength;
869     const BYTE* match = oLitEnd - sequence.offset;
870     BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
871 
872     /* bounds checks : careful of address space overflow in 32-bit mode */
873     RETURN_ERROR_IF(sequenceLength > (size_t)(oend - op), dstSize_tooSmall, "last match must fit within dstBuffer");
874     RETURN_ERROR_IF(sequence.litLength > (size_t)(litLimit - *litPtr), corruption_detected, "try to read beyond literal buffer");
875     assert(op < op + sequenceLength);
876     assert(oLitEnd < op + sequenceLength);
877 
878     /* copy literals */
879     ZSTD_safecopy(op, oend_w, *litPtr, sequence.litLength, ZSTD_no_overlap);
880     op = oLitEnd;
881     *litPtr = iLitEnd;
882 
883     /* copy Match */
884     if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
885         /* offset beyond prefix */
886         RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
887         match = dictEnd - (prefixStart - match);
888         if (match + sequence.matchLength <= dictEnd) {
889             ZSTD_memmove(oLitEnd, match, sequence.matchLength);
890             return sequenceLength;
891         }
892         /* span extDict & currentPrefixSegment */
893         {   size_t const length1 = dictEnd - match;
894         ZSTD_memmove(oLitEnd, match, length1);
895         op = oLitEnd + length1;
896         sequence.matchLength -= length1;
897         match = prefixStart;
898         }
899     }
900     ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
901     return sequenceLength;
902 }
903 
904 /* ZSTD_execSequenceEndSplitLitBuffer():
905  * This version is intended to be used during instances where the litBuffer is still split.  It is kept separate to avoid performance impact for the good case.
906  */
907 FORCE_NOINLINE
908 size_t ZSTD_execSequenceEndSplitLitBuffer(BYTE* op,
909     BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
910     const BYTE** litPtr, const BYTE* const litLimit,
911     const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
912 {
913     BYTE* const oLitEnd = op + sequence.litLength;
914     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
915     const BYTE* const iLitEnd = *litPtr + sequence.litLength;
916     const BYTE* match = oLitEnd - sequence.offset;
917 
918 
919     /* bounds checks : careful of address space overflow in 32-bit mode */
920     RETURN_ERROR_IF(sequenceLength > (size_t)(oend - op), dstSize_tooSmall, "last match must fit within dstBuffer");
921     RETURN_ERROR_IF(sequence.litLength > (size_t)(litLimit - *litPtr), corruption_detected, "try to read beyond literal buffer");
922     assert(op < op + sequenceLength);
923     assert(oLitEnd < op + sequenceLength);
924 
925     /* copy literals */
926     RETURN_ERROR_IF(op > *litPtr && op < *litPtr + sequence.litLength, dstSize_tooSmall, "output should not catch up to and overwrite literal buffer");
927     ZSTD_safecopyDstBeforeSrc(op, *litPtr, sequence.litLength);
928     op = oLitEnd;
929     *litPtr = iLitEnd;
930 
931     /* copy Match */
932     if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
933         /* offset beyond prefix */
934         RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
935         match = dictEnd - (prefixStart - match);
936         if (match + sequence.matchLength <= dictEnd) {
937             ZSTD_memmove(oLitEnd, match, sequence.matchLength);
938             return sequenceLength;
939         }
940         /* span extDict & currentPrefixSegment */
941         {   size_t const length1 = dictEnd - match;
942         ZSTD_memmove(oLitEnd, match, length1);
943         op = oLitEnd + length1;
944         sequence.matchLength -= length1;
945         match = prefixStart;
946         }
947     }
948     ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
949     return sequenceLength;
950 }
951 
952 HINT_INLINE
953 size_t ZSTD_execSequence(BYTE* op,
954     BYTE* const oend, seq_t sequence,
955     const BYTE** litPtr, const BYTE* const litLimit,
956     const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
957 {
958     BYTE* const oLitEnd = op + sequence.litLength;
959     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
960     BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
961     BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;   /* risk : address space underflow on oend=NULL */
962     const BYTE* const iLitEnd = *litPtr + sequence.litLength;
963     const BYTE* match = oLitEnd - sequence.offset;
964 
965     assert(op != NULL /* Precondition */);
966     assert(oend_w < oend /* No underflow */);
967     /* Handle edge cases in a slow path:
968      *   - Read beyond end of literals
969      *   - Match end is within WILDCOPY_OVERLIMIT of oend
970      *   - 32-bit mode and the match length overflows
971      */
972     if (UNLIKELY(
973         iLitEnd > litLimit ||
974         oMatchEnd > oend_w ||
975         (MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
976         return ZSTD_execSequenceEnd(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
977 
978     /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
979     assert(op <= oLitEnd /* No overflow */);
980     assert(oLitEnd < oMatchEnd /* Non-zero match & no overflow */);
981     assert(oMatchEnd <= oend /* No underflow */);
982     assert(iLitEnd <= litLimit /* Literal length is in bounds */);
983     assert(oLitEnd <= oend_w /* Can wildcopy literals */);
984     assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
985 
986     /* Copy Literals:
987      * Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
988      * We likely don't need the full 32-byte wildcopy.
989      */
990     assert(WILDCOPY_OVERLENGTH >= 16);
991     ZSTD_copy16(op, (*litPtr));
992     if (UNLIKELY(sequence.litLength > 16)) {
993         ZSTD_wildcopy(op + 16, (*litPtr) + 16, sequence.litLength - 16, ZSTD_no_overlap);
994     }
995     op = oLitEnd;
996     *litPtr = iLitEnd;   /* update for next sequence */
997 
998     /* Copy Match */
999     if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
1000         /* offset beyond prefix -> go into extDict */
1001         RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
1002         match = dictEnd + (match - prefixStart);
1003         if (match + sequence.matchLength <= dictEnd) {
1004             ZSTD_memmove(oLitEnd, match, sequence.matchLength);
1005             return sequenceLength;
1006         }
1007         /* span extDict & currentPrefixSegment */
1008         {   size_t const length1 = dictEnd - match;
1009         ZSTD_memmove(oLitEnd, match, length1);
1010         op = oLitEnd + length1;
1011         sequence.matchLength -= length1;
1012         match = prefixStart;
1013         }
1014     }
1015     /* Match within prefix of 1 or more bytes */
1016     assert(op <= oMatchEnd);
1017     assert(oMatchEnd <= oend_w);
1018     assert(match >= prefixStart);
1019     assert(sequence.matchLength >= 1);
1020 
1021     /* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
1022      * without overlap checking.
1023      */
1024     if (LIKELY(sequence.offset >= WILDCOPY_VECLEN)) {
1025         /* We bet on a full wildcopy for matches, since we expect matches to be
1026          * longer than literals (in general). In silesia, ~10% of matches are longer
1027          * than 16 bytes.
1028          */
1029         ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength, ZSTD_no_overlap);
1030         return sequenceLength;
1031     }
1032     assert(sequence.offset < WILDCOPY_VECLEN);
1033 
1034     /* Copy 8 bytes and spread the offset to be >= 8. */
1035     ZSTD_overlapCopy8(&op, &match, sequence.offset);
1036 
1037     /* If the match length is > 8 bytes, then continue with the wildcopy. */
1038     if (sequence.matchLength > 8) {
1039         assert(op < oMatchEnd);
1040         ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength - 8, ZSTD_overlap_src_before_dst);
1041     }
1042     return sequenceLength;
1043 }
1044 
1045 HINT_INLINE
1046 size_t ZSTD_execSequenceSplitLitBuffer(BYTE* op,
1047     BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
1048     const BYTE** litPtr, const BYTE* const litLimit,
1049     const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
1050 {
1051     BYTE* const oLitEnd = op + sequence.litLength;
1052     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
1053     BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
1054     const BYTE* const iLitEnd = *litPtr + sequence.litLength;
1055     const BYTE* match = oLitEnd - sequence.offset;
1056 
1057     assert(op != NULL /* Precondition */);
1058     assert(oend_w < oend /* No underflow */);
1059     /* Handle edge cases in a slow path:
1060      *   - Read beyond end of literals
1061      *   - Match end is within WILDCOPY_OVERLIMIT of oend
1062      *   - 32-bit mode and the match length overflows
1063      */
1064     if (UNLIKELY(
1065             iLitEnd > litLimit ||
1066             oMatchEnd > oend_w ||
1067             (MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
1068         return ZSTD_execSequenceEndSplitLitBuffer(op, oend, oend_w, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
1069 
1070     /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
1071     assert(op <= oLitEnd /* No overflow */);
1072     assert(oLitEnd < oMatchEnd /* Non-zero match & no overflow */);
1073     assert(oMatchEnd <= oend /* No underflow */);
1074     assert(iLitEnd <= litLimit /* Literal length is in bounds */);
1075     assert(oLitEnd <= oend_w /* Can wildcopy literals */);
1076     assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
1077 
1078     /* Copy Literals:
1079      * Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
1080      * We likely don't need the full 32-byte wildcopy.
1081      */
1082     assert(WILDCOPY_OVERLENGTH >= 16);
1083     ZSTD_copy16(op, (*litPtr));
1084     if (UNLIKELY(sequence.litLength > 16)) {
1085         ZSTD_wildcopy(op+16, (*litPtr)+16, sequence.litLength-16, ZSTD_no_overlap);
1086     }
1087     op = oLitEnd;
1088     *litPtr = iLitEnd;   /* update for next sequence */
1089 
1090     /* Copy Match */
1091     if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
1092         /* offset beyond prefix -> go into extDict */
1093         RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
1094         match = dictEnd + (match - prefixStart);
1095         if (match + sequence.matchLength <= dictEnd) {
1096             ZSTD_memmove(oLitEnd, match, sequence.matchLength);
1097             return sequenceLength;
1098         }
1099         /* span extDict & currentPrefixSegment */
1100         {   size_t const length1 = dictEnd - match;
1101             ZSTD_memmove(oLitEnd, match, length1);
1102             op = oLitEnd + length1;
1103             sequence.matchLength -= length1;
1104             match = prefixStart;
1105     }   }
1106     /* Match within prefix of 1 or more bytes */
1107     assert(op <= oMatchEnd);
1108     assert(oMatchEnd <= oend_w);
1109     assert(match >= prefixStart);
1110     assert(sequence.matchLength >= 1);
1111 
1112     /* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
1113      * without overlap checking.
1114      */
1115     if (LIKELY(sequence.offset >= WILDCOPY_VECLEN)) {
1116         /* We bet on a full wildcopy for matches, since we expect matches to be
1117          * longer than literals (in general). In silesia, ~10% of matches are longer
1118          * than 16 bytes.
1119          */
1120         ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength, ZSTD_no_overlap);
1121         return sequenceLength;
1122     }
1123     assert(sequence.offset < WILDCOPY_VECLEN);
1124 
1125     /* Copy 8 bytes and spread the offset to be >= 8. */
1126     ZSTD_overlapCopy8(&op, &match, sequence.offset);
1127 
1128     /* If the match length is > 8 bytes, then continue with the wildcopy. */
1129     if (sequence.matchLength > 8) {
1130         assert(op < oMatchEnd);
1131         ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst);
1132     }
1133     return sequenceLength;
1134 }
1135 
1136 
1137 static void
1138 ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt)
1139 {
1140     const void* ptr = dt;
1141     const ZSTD_seqSymbol_header* const DTableH = (const ZSTD_seqSymbol_header*)ptr;
1142     DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
1143     DEBUGLOG(6, "ZSTD_initFseState : val=%u using %u bits",
1144                 (U32)DStatePtr->state, DTableH->tableLog);
1145     BIT_reloadDStream(bitD);
1146     DStatePtr->table = dt + 1;
1147 }
1148 
1149 FORCE_INLINE_TEMPLATE void
1150 ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, U16 nextState, U32 nbBits)
1151 {
1152     size_t const lowBits = BIT_readBits(bitD, nbBits);
1153     DStatePtr->state = nextState + lowBits;
1154 }
1155 
1156 /* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
1157  * offset bits. But we can only read at most (STREAM_ACCUMULATOR_MIN_32 - 1)
1158  * bits before reloading. This value is the maximum number of bytes we read
1159  * after reloading when we are decoding long offsets.
1160  */
1161 #define LONG_OFFSETS_MAX_EXTRA_BITS_32                       \
1162     (ZSTD_WINDOWLOG_MAX_32 > STREAM_ACCUMULATOR_MIN_32       \
1163         ? ZSTD_WINDOWLOG_MAX_32 - STREAM_ACCUMULATOR_MIN_32  \
1164         : 0)
1165 
1166 typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e;
1167 
1168 FORCE_INLINE_TEMPLATE seq_t
1169 ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
1170 {
1171     seq_t seq;
1172     const ZSTD_seqSymbol* const llDInfo = seqState->stateLL.table + seqState->stateLL.state;
1173     const ZSTD_seqSymbol* const mlDInfo = seqState->stateML.table + seqState->stateML.state;
1174     const ZSTD_seqSymbol* const ofDInfo = seqState->stateOffb.table + seqState->stateOffb.state;
1175     seq.matchLength = mlDInfo->baseValue;
1176     seq.litLength = llDInfo->baseValue;
1177     {   U32 const ofBase = ofDInfo->baseValue;
1178         BYTE const llBits = llDInfo->nbAdditionalBits;
1179         BYTE const mlBits = mlDInfo->nbAdditionalBits;
1180         BYTE const ofBits = ofDInfo->nbAdditionalBits;
1181         BYTE const totalBits = llBits+mlBits+ofBits;
1182 
1183         U16 const llNext = llDInfo->nextState;
1184         U16 const mlNext = mlDInfo->nextState;
1185         U16 const ofNext = ofDInfo->nextState;
1186         U32 const llnbBits = llDInfo->nbBits;
1187         U32 const mlnbBits = mlDInfo->nbBits;
1188         U32 const ofnbBits = ofDInfo->nbBits;
1189         /*
1190          * As gcc has better branch and block analyzers, sometimes it is only
1191          * valuable to mark likelyness for clang, it gives around 3-4% of
1192          * performance.
1193          */
1194 
1195         /* sequence */
1196         {   size_t offset;
1197     #if defined(__clang__)
1198             if (LIKELY(ofBits > 1)) {
1199     #else
1200             if (ofBits > 1) {
1201     #endif
1202                 ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
1203                 ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
1204                 assert(ofBits <= MaxOff);
1205                 if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
1206                     U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed);
1207                     offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
1208                     BIT_reloadDStream(&seqState->DStream);
1209                     if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
1210                     assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32);   /* to avoid another reload */
1211                 } else {
1212                     offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
1213                     if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
1214                 }
1215                 seqState->prevOffset[2] = seqState->prevOffset[1];
1216                 seqState->prevOffset[1] = seqState->prevOffset[0];
1217                 seqState->prevOffset[0] = offset;
1218             } else {
1219                 U32 const ll0 = (llDInfo->baseValue == 0);
1220                 if (LIKELY((ofBits == 0))) {
1221                     offset = seqState->prevOffset[ll0];
1222                     seqState->prevOffset[1] = seqState->prevOffset[!ll0];
1223                     seqState->prevOffset[0] = offset;
1224                 } else {
1225                     offset = ofBase + ll0 + BIT_readBitsFast(&seqState->DStream, 1);
1226                     {   size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
1227                         temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
1228                         if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
1229                         seqState->prevOffset[1] = seqState->prevOffset[0];
1230                         seqState->prevOffset[0] = offset = temp;
1231             }   }   }
1232             seq.offset = offset;
1233         }
1234 
1235     #if defined(__clang__)
1236         if (UNLIKELY(mlBits > 0))
1237     #else
1238         if (mlBits > 0)
1239     #endif
1240             seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
1241 
1242         if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
1243             BIT_reloadDStream(&seqState->DStream);
1244         if (MEM_64bits() && UNLIKELY(totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
1245             BIT_reloadDStream(&seqState->DStream);
1246         /* Ensure there are enough bits to read the rest of data in 64-bit mode. */
1247         ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
1248 
1249     #if defined(__clang__)
1250         if (UNLIKELY(llBits > 0))
1251     #else
1252         if (llBits > 0)
1253     #endif
1254             seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
1255 
1256         if (MEM_32bits())
1257             BIT_reloadDStream(&seqState->DStream);
1258 
1259         DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
1260                     (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
1261 
1262         ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llNext, llnbBits);    /* <=  9 bits */
1263         ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlNext, mlnbBits);    /* <=  9 bits */
1264         if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
1265         ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofNext, ofnbBits);  /* <=  8 bits */
1266     }
1267 
1268     return seq;
1269 }
1270 
1271 #ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
1272 MEM_STATIC int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
1273 {
1274     size_t const windowSize = dctx->fParams.windowSize;
1275     /* No dictionary used. */
1276     if (dctx->dictContentEndForFuzzing == NULL) return 0;
1277     /* Dictionary is our prefix. */
1278     if (prefixStart == dctx->dictContentBeginForFuzzing) return 1;
1279     /* Dictionary is not our ext-dict. */
1280     if (dctx->dictEnd != dctx->dictContentEndForFuzzing) return 0;
1281     /* Dictionary is not within our window size. */
1282     if ((size_t)(oLitEnd - prefixStart) >= windowSize) return 0;
1283     /* Dictionary is active. */
1284     return 1;
1285 }
1286 
1287 MEM_STATIC void ZSTD_assertValidSequence(
1288         ZSTD_DCtx const* dctx,
1289         BYTE const* op, BYTE const* oend,
1290         seq_t const seq,
1291         BYTE const* prefixStart, BYTE const* virtualStart)
1292 {
1293 #if DEBUGLEVEL >= 1
1294     size_t const windowSize = dctx->fParams.windowSize;
1295     size_t const sequenceSize = seq.litLength + seq.matchLength;
1296     BYTE const* const oLitEnd = op + seq.litLength;
1297     DEBUGLOG(6, "Checking sequence: litL=%u matchL=%u offset=%u",
1298             (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
1299     assert(op <= oend);
1300     assert((size_t)(oend - op) >= sequenceSize);
1301     assert(sequenceSize <= ZSTD_BLOCKSIZE_MAX);
1302     if (ZSTD_dictionaryIsActive(dctx, prefixStart, oLitEnd)) {
1303         size_t const dictSize = (size_t)((char const*)dctx->dictContentEndForFuzzing - (char const*)dctx->dictContentBeginForFuzzing);
1304         /* Offset must be within the dictionary. */
1305         assert(seq.offset <= (size_t)(oLitEnd - virtualStart));
1306         assert(seq.offset <= windowSize + dictSize);
1307     } else {
1308         /* Offset must be within our window. */
1309         assert(seq.offset <= windowSize);
1310     }
1311 #else
1312     (void)dctx, (void)op, (void)oend, (void)seq, (void)prefixStart, (void)virtualStart;
1313 #endif
1314 }
1315 #endif
1316 
1317 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
1318 
1319 
1320 FORCE_INLINE_TEMPLATE size_t
1321 DONT_VECTORIZE
1322 ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
1323                                void* dst, size_t maxDstSize,
1324                          const void* seqStart, size_t seqSize, int nbSeq,
1325                          const ZSTD_longOffset_e isLongOffset,
1326                          const int frame)
1327 {
1328     const BYTE* ip = (const BYTE*)seqStart;
1329     const BYTE* const iend = ip + seqSize;
1330     BYTE* const ostart = (BYTE*)dst;
1331     BYTE* const oend = ostart + maxDstSize;
1332     BYTE* op = ostart;
1333     const BYTE* litPtr = dctx->litPtr;
1334     const BYTE* litBufferEnd = dctx->litBufferEnd;
1335     const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
1336     const BYTE* const vBase = (const BYTE*) (dctx->virtualStart);
1337     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
1338     DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer");
1339     (void)frame;
1340 
1341     /* Regen sequences */
1342     if (nbSeq) {
1343         seqState_t seqState;
1344         dctx->fseEntropy = 1;
1345         { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
1346         RETURN_ERROR_IF(
1347             ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)),
1348             corruption_detected, "");
1349         ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
1350         ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
1351         ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
1352         assert(dst != NULL);
1353 
1354         ZSTD_STATIC_ASSERT(
1355                 BIT_DStream_unfinished < BIT_DStream_completed &&
1356                 BIT_DStream_endOfBuffer < BIT_DStream_completed &&
1357                 BIT_DStream_completed < BIT_DStream_overflow);
1358 
1359         /* decompress without overrunning litPtr begins */
1360         {
1361             seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
1362             /* Align the decompression loop to 32 + 16 bytes.
1363                 *
1364                 * zstd compiled with gcc-9 on an Intel i9-9900k shows 10% decompression
1365                 * speed swings based on the alignment of the decompression loop. This
1366                 * performance swing is caused by parts of the decompression loop falling
1367                 * out of the DSB. The entire decompression loop should fit in the DSB,
1368                 * when it can't we get much worse performance. You can measure if you've
1369                 * hit the good case or the bad case with this perf command for some
1370                 * compressed file test.zst:
1371                 *
1372                 *   perf stat -e cycles -e instructions -e idq.all_dsb_cycles_any_uops \
1373                 *             -e idq.all_mite_cycles_any_uops -- ./zstd -tq test.zst
1374                 *
1375                 * If you see most cycles served out of the MITE you've hit the bad case.
1376                 * If you see most cycles served out of the DSB you've hit the good case.
1377                 * If it is pretty even then you may be in an okay case.
1378                 *
1379                 * This issue has been reproduced on the following CPUs:
1380                 *   - Kabylake: Macbook Pro (15-inch, 2019) 2.4 GHz Intel Core i9
1381                 *               Use Instruments->Counters to get DSB/MITE cycles.
1382                 *               I never got performance swings, but I was able to
1383                 *               go from the good case of mostly DSB to half of the
1384                 *               cycles served from MITE.
1385                 *   - Coffeelake: Intel i9-9900k
1386                 *   - Coffeelake: Intel i7-9700k
1387                 *
1388                 * I haven't been able to reproduce the instability or DSB misses on any
1389                 * of the following CPUS:
1390                 *   - Haswell
1391                 *   - Broadwell: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GH
1392                 *   - Skylake
1393                 *
1394                 * Alignment is done for each of the three major decompression loops:
1395                 *   - ZSTD_decompressSequences_bodySplitLitBuffer - presplit section of the literal buffer
1396                 *   - ZSTD_decompressSequences_bodySplitLitBuffer - postsplit section of the literal buffer
1397                 *   - ZSTD_decompressSequences_body
1398                 * Alignment choices are made to minimize large swings on bad cases and influence on performance
1399                 * from changes external to this code, rather than to overoptimize on the current commit.
1400                 *
1401                 * If you are seeing performance stability this script can help test.
1402                 * It tests on 4 commits in zstd where I saw performance change.
1403                 *
1404                 *   https://gist.github.com/terrelln/9889fc06a423fd5ca6e99351564473f4
1405                 */
1406 #if defined(__x86_64__)
1407             __asm__(".p2align 6");
1408 #  if __GNUC__ >= 7
1409 	    /* good for gcc-7, gcc-9, and gcc-11 */
1410             __asm__("nop");
1411             __asm__(".p2align 5");
1412             __asm__("nop");
1413             __asm__(".p2align 4");
1414 #    if __GNUC__ == 8 || __GNUC__ == 10
1415 	    /* good for gcc-8 and gcc-10 */
1416             __asm__("nop");
1417             __asm__(".p2align 3");
1418 #    endif
1419 #  endif
1420 #endif
1421 
1422             /* Handle the initial state where litBuffer is currently split between dst and litExtraBuffer */
1423             for (; litPtr + sequence.litLength <= dctx->litBufferEnd; ) {
1424                 size_t const oneSeqSize = ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence.litLength - WILDCOPY_OVERLENGTH, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
1425 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
1426                 assert(!ZSTD_isError(oneSeqSize));
1427                 if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
1428 #endif
1429                 if (UNLIKELY(ZSTD_isError(oneSeqSize)))
1430                     return oneSeqSize;
1431                 DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
1432                 op += oneSeqSize;
1433                 if (UNLIKELY(!--nbSeq))
1434                     break;
1435                 BIT_reloadDStream(&(seqState.DStream));
1436                 sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
1437             }
1438 
1439             /* If there are more sequences, they will need to read literals from litExtraBuffer; copy over the remainder from dst and update litPtr and litEnd */
1440             if (nbSeq > 0) {
1441                 const size_t leftoverLit = dctx->litBufferEnd - litPtr;
1442                 if (leftoverLit)
1443                 {
1444                     RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
1445                     ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
1446                     sequence.litLength -= leftoverLit;
1447                     op += leftoverLit;
1448                 }
1449                 litPtr = dctx->litExtraBuffer;
1450                 litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
1451                 dctx->litBufferLocation = ZSTD_not_in_dst;
1452                 {
1453                     size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
1454 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
1455                     assert(!ZSTD_isError(oneSeqSize));
1456                     if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
1457 #endif
1458                     if (UNLIKELY(ZSTD_isError(oneSeqSize)))
1459                         return oneSeqSize;
1460                     DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
1461                     op += oneSeqSize;
1462                     if (--nbSeq)
1463                         BIT_reloadDStream(&(seqState.DStream));
1464                 }
1465             }
1466         }
1467 
1468         if (nbSeq > 0) /* there is remaining lit from extra buffer */
1469         {
1470 
1471 #if defined(__x86_64__)
1472             __asm__(".p2align 6");
1473             __asm__("nop");
1474 #  if __GNUC__ != 7
1475             /* worse for gcc-7 better for gcc-8, gcc-9, and gcc-10 and clang */
1476             __asm__(".p2align 4");
1477             __asm__("nop");
1478             __asm__(".p2align 3");
1479 #  elif __GNUC__ >= 11
1480             __asm__(".p2align 3");
1481 #  else
1482             __asm__(".p2align 5");
1483             __asm__("nop");
1484             __asm__(".p2align 3");
1485 #  endif
1486 #endif
1487 
1488             for (; ; ) {
1489                 seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
1490                 size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
1491 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
1492                 assert(!ZSTD_isError(oneSeqSize));
1493                 if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
1494 #endif
1495                 if (UNLIKELY(ZSTD_isError(oneSeqSize)))
1496                     return oneSeqSize;
1497                 DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
1498                 op += oneSeqSize;
1499                 if (UNLIKELY(!--nbSeq))
1500                     break;
1501                 BIT_reloadDStream(&(seqState.DStream));
1502             }
1503         }
1504 
1505         /* check if reached exact end */
1506         DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer: after decode loop, remaining nbSeq : %i", nbSeq);
1507         RETURN_ERROR_IF(nbSeq, corruption_detected, "");
1508         RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
1509         /* save reps for next block */
1510         { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
1511     }
1512 
1513     /* last literal segment */
1514     if (dctx->litBufferLocation == ZSTD_split)  /* split hasn't been reached yet, first get dst then copy litExtraBuffer */
1515     {
1516         size_t const lastLLSize = litBufferEnd - litPtr;
1517         RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
1518         if (op != NULL) {
1519             ZSTD_memmove(op, litPtr, lastLLSize);
1520             op += lastLLSize;
1521         }
1522         litPtr = dctx->litExtraBuffer;
1523         litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
1524         dctx->litBufferLocation = ZSTD_not_in_dst;
1525     }
1526     {   size_t const lastLLSize = litBufferEnd - litPtr;
1527         RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
1528         if (op != NULL) {
1529             ZSTD_memcpy(op, litPtr, lastLLSize);
1530             op += lastLLSize;
1531         }
1532     }
1533 
1534     return op-ostart;
1535 }
1536 
1537 FORCE_INLINE_TEMPLATE size_t
1538 DONT_VECTORIZE
1539 ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
1540     void* dst, size_t maxDstSize,
1541     const void* seqStart, size_t seqSize, int nbSeq,
1542     const ZSTD_longOffset_e isLongOffset,
1543     const int frame)
1544 {
1545     const BYTE* ip = (const BYTE*)seqStart;
1546     const BYTE* const iend = ip + seqSize;
1547     BYTE* const ostart = (BYTE*)dst;
1548     BYTE* const oend = dctx->litBufferLocation == ZSTD_not_in_dst ? ostart + maxDstSize : dctx->litBuffer;
1549     BYTE* op = ostart;
1550     const BYTE* litPtr = dctx->litPtr;
1551     const BYTE* const litEnd = litPtr + dctx->litSize;
1552     const BYTE* const prefixStart = (const BYTE*)(dctx->prefixStart);
1553     const BYTE* const vBase = (const BYTE*)(dctx->virtualStart);
1554     const BYTE* const dictEnd = (const BYTE*)(dctx->dictEnd);
1555     DEBUGLOG(5, "ZSTD_decompressSequences_body");
1556     (void)frame;
1557 
1558     /* Regen sequences */
1559     if (nbSeq) {
1560         seqState_t seqState;
1561         dctx->fseEntropy = 1;
1562         { U32 i; for (i = 0; i < ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
1563         RETURN_ERROR_IF(
1564             ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend - ip)),
1565             corruption_detected, "");
1566         ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
1567         ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
1568         ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
1569         assert(dst != NULL);
1570 
1571         ZSTD_STATIC_ASSERT(
1572             BIT_DStream_unfinished < BIT_DStream_completed &&
1573             BIT_DStream_endOfBuffer < BIT_DStream_completed &&
1574             BIT_DStream_completed < BIT_DStream_overflow);
1575 
1576 #if defined(__x86_64__)
1577             __asm__(".p2align 6");
1578             __asm__("nop");
1579 #  if __GNUC__ >= 7
1580             __asm__(".p2align 5");
1581             __asm__("nop");
1582             __asm__(".p2align 3");
1583 #  else
1584             __asm__(".p2align 4");
1585             __asm__("nop");
1586             __asm__(".p2align 3");
1587 #  endif
1588 #endif
1589 
1590         for ( ; ; ) {
1591             seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
1592             size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
1593 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
1594             assert(!ZSTD_isError(oneSeqSize));
1595             if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
1596 #endif
1597             if (UNLIKELY(ZSTD_isError(oneSeqSize)))
1598                 return oneSeqSize;
1599             DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
1600             op += oneSeqSize;
1601             if (UNLIKELY(!--nbSeq))
1602                 break;
1603             BIT_reloadDStream(&(seqState.DStream));
1604         }
1605 
1606         /* check if reached exact end */
1607         DEBUGLOG(5, "ZSTD_decompressSequences_body: after decode loop, remaining nbSeq : %i", nbSeq);
1608         RETURN_ERROR_IF(nbSeq, corruption_detected, "");
1609         RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
1610         /* save reps for next block */
1611         { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
1612     }
1613 
1614     /* last literal segment */
1615     {   size_t const lastLLSize = litEnd - litPtr;
1616         RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
1617         if (op != NULL) {
1618             ZSTD_memcpy(op, litPtr, lastLLSize);
1619             op += lastLLSize;
1620         }
1621     }
1622 
1623     return op-ostart;
1624 }
1625 
1626 static size_t
1627 ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
1628                                  void* dst, size_t maxDstSize,
1629                            const void* seqStart, size_t seqSize, int nbSeq,
1630                            const ZSTD_longOffset_e isLongOffset,
1631                            const int frame)
1632 {
1633     return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1634 }
1635 
1636 static size_t
1637 ZSTD_decompressSequencesSplitLitBuffer_default(ZSTD_DCtx* dctx,
1638                                                void* dst, size_t maxDstSize,
1639                                          const void* seqStart, size_t seqSize, int nbSeq,
1640                                          const ZSTD_longOffset_e isLongOffset,
1641                                          const int frame)
1642 {
1643     return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1644 }
1645 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
1646 
1647 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
1648 
1649 FORCE_INLINE_TEMPLATE size_t
1650 ZSTD_prefetchMatch(size_t prefetchPos, seq_t const sequence,
1651                    const BYTE* const prefixStart, const BYTE* const dictEnd)
1652 {
1653     prefetchPos += sequence.litLength;
1654     {   const BYTE* const matchBase = (sequence.offset > prefetchPos) ? dictEnd : prefixStart;
1655         const BYTE* const match = matchBase + prefetchPos - sequence.offset; /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
1656                                                                               * No consequence though : memory address is only used for prefetching, not for dereferencing */
1657         PREFETCH_L1(match); PREFETCH_L1(match+CACHELINE_SIZE);   /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
1658     }
1659     return prefetchPos + sequence.matchLength;
1660 }
1661 
1662 /* This decoding function employs prefetching
1663  * to reduce latency impact of cache misses.
1664  * It's generally employed when block contains a significant portion of long-distance matches
1665  * or when coupled with a "cold" dictionary */
1666 FORCE_INLINE_TEMPLATE size_t
1667 ZSTD_decompressSequencesLong_body(
1668                                ZSTD_DCtx* dctx,
1669                                void* dst, size_t maxDstSize,
1670                          const void* seqStart, size_t seqSize, int nbSeq,
1671                          const ZSTD_longOffset_e isLongOffset,
1672                          const int frame)
1673 {
1674     const BYTE* ip = (const BYTE*)seqStart;
1675     const BYTE* const iend = ip + seqSize;
1676     BYTE* const ostart = (BYTE*)dst;
1677     BYTE* const oend = dctx->litBufferLocation == ZSTD_in_dst ? dctx->litBuffer : ostart + maxDstSize;
1678     BYTE* op = ostart;
1679     const BYTE* litPtr = dctx->litPtr;
1680     const BYTE* litBufferEnd = dctx->litBufferEnd;
1681     const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
1682     const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart);
1683     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
1684     (void)frame;
1685 
1686     /* Regen sequences */
1687     if (nbSeq) {
1688 #define STORED_SEQS 8
1689 #define STORED_SEQS_MASK (STORED_SEQS-1)
1690 #define ADVANCED_SEQS STORED_SEQS
1691         seq_t sequences[STORED_SEQS];
1692         int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS);
1693         seqState_t seqState;
1694         int seqNb;
1695         size_t prefetchPos = (size_t)(op-prefixStart); /* track position relative to prefixStart */
1696 
1697         dctx->fseEntropy = 1;
1698         { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
1699         assert(dst != NULL);
1700         assert(iend >= ip);
1701         RETURN_ERROR_IF(
1702             ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)),
1703             corruption_detected, "");
1704         ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
1705         ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
1706         ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
1707 
1708         /* prepare in advance */
1709         for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && (seqNb<seqAdvance); seqNb++) {
1710             seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
1711             prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
1712             sequences[seqNb] = sequence;
1713         }
1714         RETURN_ERROR_IF(seqNb<seqAdvance, corruption_detected, "");
1715 
1716         /* decompress without stomping litBuffer */
1717         for (; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb < nbSeq); seqNb++) {
1718             seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
1719             size_t oneSeqSize;
1720 
1721             if (dctx->litBufferLocation == ZSTD_split && litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength > dctx->litBufferEnd)
1722             {
1723                 /* lit buffer is reaching split point, empty out the first buffer and transition to litExtraBuffer */
1724                 const size_t leftoverLit = dctx->litBufferEnd - litPtr;
1725                 if (leftoverLit)
1726                 {
1727                     RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
1728                     ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
1729                     sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength -= leftoverLit;
1730                     op += leftoverLit;
1731                 }
1732                 litPtr = dctx->litExtraBuffer;
1733                 litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
1734                 dctx->litBufferLocation = ZSTD_not_in_dst;
1735                 oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
1736 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
1737                 assert(!ZSTD_isError(oneSeqSize));
1738                 if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
1739 #endif
1740                 if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
1741 
1742                 prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
1743                 sequences[seqNb & STORED_SEQS_MASK] = sequence;
1744                 op += oneSeqSize;
1745             }
1746             else
1747             {
1748                 /* lit buffer is either wholly contained in first or second split, or not split at all*/
1749                 oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
1750                     ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength - WILDCOPY_OVERLENGTH, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
1751                     ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
1752 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
1753                 assert(!ZSTD_isError(oneSeqSize));
1754                 if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
1755 #endif
1756                 if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
1757 
1758                 prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
1759                 sequences[seqNb & STORED_SEQS_MASK] = sequence;
1760                 op += oneSeqSize;
1761             }
1762         }
1763         RETURN_ERROR_IF(seqNb<nbSeq, corruption_detected, "");
1764 
1765         /* finish queue */
1766         seqNb -= seqAdvance;
1767         for ( ; seqNb<nbSeq ; seqNb++) {
1768             seq_t *sequence = &(sequences[seqNb&STORED_SEQS_MASK]);
1769             if (dctx->litBufferLocation == ZSTD_split && litPtr + sequence->litLength > dctx->litBufferEnd)
1770             {
1771                 const size_t leftoverLit = dctx->litBufferEnd - litPtr;
1772                 if (leftoverLit)
1773                 {
1774                     RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
1775                     ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
1776                     sequence->litLength -= leftoverLit;
1777                     op += leftoverLit;
1778                 }
1779                 litPtr = dctx->litExtraBuffer;
1780                 litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
1781                 dctx->litBufferLocation = ZSTD_not_in_dst;
1782                 {
1783                     size_t const oneSeqSize = ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
1784 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
1785                     assert(!ZSTD_isError(oneSeqSize));
1786                     if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
1787 #endif
1788                     if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
1789                     op += oneSeqSize;
1790                 }
1791             }
1792             else
1793             {
1794                 size_t const oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
1795                     ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence->litLength - WILDCOPY_OVERLENGTH, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
1796                     ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
1797 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
1798                 assert(!ZSTD_isError(oneSeqSize));
1799                 if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
1800 #endif
1801                 if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
1802                 op += oneSeqSize;
1803             }
1804         }
1805 
1806         /* save reps for next block */
1807         { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
1808     }
1809 
1810     /* last literal segment */
1811     if (dctx->litBufferLocation == ZSTD_split)  /* first deplete literal buffer in dst, then copy litExtraBuffer */
1812     {
1813         size_t const lastLLSize = litBufferEnd - litPtr;
1814         RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
1815         if (op != NULL) {
1816             ZSTD_memmove(op, litPtr, lastLLSize);
1817             op += lastLLSize;
1818         }
1819         litPtr = dctx->litExtraBuffer;
1820         litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
1821     }
1822     {   size_t const lastLLSize = litBufferEnd - litPtr;
1823         RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
1824         if (op != NULL) {
1825             ZSTD_memmove(op, litPtr, lastLLSize);
1826             op += lastLLSize;
1827         }
1828     }
1829 
1830     return op-ostart;
1831 }
1832 
1833 static size_t
1834 ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
1835                                  void* dst, size_t maxDstSize,
1836                            const void* seqStart, size_t seqSize, int nbSeq,
1837                            const ZSTD_longOffset_e isLongOffset,
1838                            const int frame)
1839 {
1840     return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1841 }
1842 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
1843 
1844 
1845 
1846 #if DYNAMIC_BMI2
1847 
1848 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
1849 static BMI2_TARGET_ATTRIBUTE size_t
1850 DONT_VECTORIZE
1851 ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
1852                                  void* dst, size_t maxDstSize,
1853                            const void* seqStart, size_t seqSize, int nbSeq,
1854                            const ZSTD_longOffset_e isLongOffset,
1855                            const int frame)
1856 {
1857     return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1858 }
1859 static BMI2_TARGET_ATTRIBUTE size_t
1860 DONT_VECTORIZE
1861 ZSTD_decompressSequencesSplitLitBuffer_bmi2(ZSTD_DCtx* dctx,
1862                                  void* dst, size_t maxDstSize,
1863                            const void* seqStart, size_t seqSize, int nbSeq,
1864                            const ZSTD_longOffset_e isLongOffset,
1865                            const int frame)
1866 {
1867     return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1868 }
1869 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
1870 
1871 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
1872 static BMI2_TARGET_ATTRIBUTE size_t
1873 ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx,
1874                                  void* dst, size_t maxDstSize,
1875                            const void* seqStart, size_t seqSize, int nbSeq,
1876                            const ZSTD_longOffset_e isLongOffset,
1877                            const int frame)
1878 {
1879     return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1880 }
1881 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
1882 
1883 #endif /* DYNAMIC_BMI2 */
1884 
1885 typedef size_t (*ZSTD_decompressSequences_t)(
1886                             ZSTD_DCtx* dctx,
1887                             void* dst, size_t maxDstSize,
1888                             const void* seqStart, size_t seqSize, int nbSeq,
1889                             const ZSTD_longOffset_e isLongOffset,
1890                             const int frame);
1891 
1892 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
1893 static size_t
1894 ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
1895                    const void* seqStart, size_t seqSize, int nbSeq,
1896                    const ZSTD_longOffset_e isLongOffset,
1897                    const int frame)
1898 {
1899     DEBUGLOG(5, "ZSTD_decompressSequences");
1900 #if DYNAMIC_BMI2
1901     if (ZSTD_DCtx_get_bmi2(dctx)) {
1902         return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1903     }
1904 #endif
1905     return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1906 }
1907 static size_t
1908 ZSTD_decompressSequencesSplitLitBuffer(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
1909                                  const void* seqStart, size_t seqSize, int nbSeq,
1910                                  const ZSTD_longOffset_e isLongOffset,
1911                                  const int frame)
1912 {
1913     DEBUGLOG(5, "ZSTD_decompressSequencesSplitLitBuffer");
1914 #if DYNAMIC_BMI2
1915     if (ZSTD_DCtx_get_bmi2(dctx)) {
1916         return ZSTD_decompressSequencesSplitLitBuffer_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1917     }
1918 #endif
1919     return ZSTD_decompressSequencesSplitLitBuffer_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1920 }
1921 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
1922 
1923 
1924 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
1925 /* ZSTD_decompressSequencesLong() :
1926  * decompression function triggered when a minimum share of offsets is considered "long",
1927  * aka out of cache.
1928  * note : "long" definition seems overloaded here, sometimes meaning "wider than bitstream register", and sometimes meaning "farther than memory cache distance".
1929  * This function will try to mitigate main memory latency through the use of prefetching */
1930 static size_t
1931 ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
1932                              void* dst, size_t maxDstSize,
1933                              const void* seqStart, size_t seqSize, int nbSeq,
1934                              const ZSTD_longOffset_e isLongOffset,
1935                              const int frame)
1936 {
1937     DEBUGLOG(5, "ZSTD_decompressSequencesLong");
1938 #if DYNAMIC_BMI2
1939     if (ZSTD_DCtx_get_bmi2(dctx)) {
1940         return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1941     }
1942 #endif
1943   return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
1944 }
1945 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
1946 
1947 
1948 
1949 #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
1950     !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
1951 /* ZSTD_getLongOffsetsShare() :
1952  * condition : offTable must be valid
1953  * @return : "share" of long offsets (arbitrarily defined as > (1<<23))
1954  *           compared to maximum possible of (1<<OffFSELog) */
1955 static unsigned
1956 ZSTD_getLongOffsetsShare(const ZSTD_seqSymbol* offTable)
1957 {
1958     const void* ptr = offTable;
1959     U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
1960     const ZSTD_seqSymbol* table = offTable + 1;
1961     U32 const max = 1 << tableLog;
1962     U32 u, total = 0;
1963     DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
1964 
1965     assert(max <= (1 << OffFSELog));  /* max not too large */
1966     for (u=0; u<max; u++) {
1967         if (table[u].nbAdditionalBits > 22) total += 1;
1968     }
1969 
1970     assert(tableLog <= OffFSELog);
1971     total <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
1972 
1973     return total;
1974 }
1975 #endif
1976 
1977 size_t
1978 ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
1979                               void* dst, size_t dstCapacity,
1980                         const void* src, size_t srcSize, const int frame, const streaming_operation streaming)
1981 {   /* blockType == blockCompressed */
1982     const BYTE* ip = (const BYTE*)src;
1983     /* isLongOffset must be true if there are long offsets.
1984      * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
1985      * We don't expect that to be the case in 64-bit mode.
1986      * In block mode, window size is not known, so we have to be conservative.
1987      * (note: but it could be evaluated from current-lowLimit)
1988      */
1989     ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || (dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN))));
1990     DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
1991 
1992     RETURN_ERROR_IF(srcSize >= ZSTD_BLOCKSIZE_MAX, srcSize_wrong, "");
1993 
1994     /* Decode literals section */
1995     {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize, dst, dstCapacity, streaming);
1996         DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize);
1997         if (ZSTD_isError(litCSize)) return litCSize;
1998         ip += litCSize;
1999         srcSize -= litCSize;
2000     }
2001 
2002     /* Build Decoding Tables */
2003     {
2004         /* These macros control at build-time which decompressor implementation
2005          * we use. If neither is defined, we do some inspection and dispatch at
2006          * runtime.
2007          */
2008 #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
2009     !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
2010         int usePrefetchDecoder = dctx->ddictIsCold;
2011 #endif
2012         int nbSeq;
2013         size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
2014         if (ZSTD_isError(seqHSize)) return seqHSize;
2015         ip += seqHSize;
2016         srcSize -= seqHSize;
2017 
2018         RETURN_ERROR_IF(dst == NULL && nbSeq > 0, dstSize_tooSmall, "NULL not handled");
2019 
2020 #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
2021     !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
2022         if ( !usePrefetchDecoder
2023           && (!frame || (dctx->fParams.windowSize > (1<<24)))
2024           && (nbSeq>ADVANCED_SEQS) ) {  /* could probably use a larger nbSeq limit */
2025             U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr);
2026             U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
2027             usePrefetchDecoder = (shareLongOffsets >= minShare);
2028         }
2029 #endif
2030 
2031         dctx->ddictIsCold = 0;
2032 
2033 #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
2034     !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
2035         if (usePrefetchDecoder)
2036 #endif
2037 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
2038             return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
2039 #endif
2040 
2041 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
2042         /* else */
2043         if (dctx->litBufferLocation == ZSTD_split)
2044             return ZSTD_decompressSequencesSplitLitBuffer(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
2045         else
2046             return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
2047 #endif
2048     }
2049 }
2050 
2051 
2052 void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst, size_t dstSize)
2053 {
2054     if (dst != dctx->previousDstEnd && dstSize > 0) {   /* not contiguous */
2055         dctx->dictEnd = dctx->previousDstEnd;
2056         dctx->virtualStart = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
2057         dctx->prefixStart = dst;
2058         dctx->previousDstEnd = dst;
2059     }
2060 }
2061 
2062 
2063 size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
2064                             void* dst, size_t dstCapacity,
2065                       const void* src, size_t srcSize)
2066 {
2067     size_t dSize;
2068     ZSTD_checkContinuity(dctx, dst, dstCapacity);
2069     dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0, not_streaming);
2070     dctx->previousDstEnd = (char*)dst + dSize;
2071     return dSize;
2072 }
2073