1 /* ******************************************************************
2 * FSE : Finite State Entropy decoder
3 * Copyright (c) Meta Platforms, Inc. and affiliates.
5 * You can contact the author at :
6 * - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
7 * - Public forum : https://groups.google.com/forum/#!forum/lz4c
9 * This source code is licensed under both the BSD-style license (found in the
10 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
11 * in the COPYING file in the root directory of this source tree).
12 * You may select, at your option, one of the above-listed licenses.
13 ****************************************************************** */
16 /* **************************************************************
18 ****************************************************************/
19 #include "debug.h" /* assert */
20 #include "bitstream.h"
22 #define FSE_STATIC_LINKING_ONLY
24 #include "error_private.h"
25 #define ZSTD_DEPS_NEED_MALLOC
26 #include "zstd_deps.h"
27 #include "bits.h" /* ZSTD_highbit32 */
30 /* **************************************************************
32 ****************************************************************/
33 #define FSE_isError ERR_isError
34 #define FSE_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c) /* use only *after* variable declarations */
37 /* **************************************************************
39 ****************************************************************/
41 designed to be included
42 for type-specific functions (template emulation in C)
43 Objective is to write these functions only once, for improved maintenance
47 #ifndef FSE_FUNCTION_EXTENSION
48 # error "FSE_FUNCTION_EXTENSION must be defined"
50 #ifndef FSE_FUNCTION_TYPE
51 # error "FSE_FUNCTION_TYPE must be defined"
55 #define FSE_CAT(X,Y) X##Y
56 #define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
57 #define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
59 static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
61 void* const tdPtr = dt+1; /* because *dt is unsigned, 32-bits aligned on 32-bits */
62 FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
63 U16* symbolNext = (U16*)workSpace;
64 BYTE* spread = (BYTE*)(symbolNext + maxSymbolValue + 1);
66 U32 const maxSV1 = maxSymbolValue + 1;
67 U32 const tableSize = 1 << tableLog;
68 U32 highThreshold = tableSize-1;
71 if (FSE_BUILD_DTABLE_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(maxSymbolValue_tooLarge);
72 if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
73 if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
75 /* Init, lay down lowprob symbols */
76 { FSE_DTableHeader DTableH;
77 DTableH.tableLog = (U16)tableLog;
79 { S16 const largeLimit= (S16)(1 << (tableLog-1));
81 for (s=0; s<maxSV1; s++) {
82 if (normalizedCounter[s]==-1) {
83 tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
86 if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
87 symbolNext[s] = normalizedCounter[s];
89 ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
93 if (highThreshold == tableSize - 1) {
94 size_t const tableMask = tableSize-1;
95 size_t const step = FSE_TABLESTEP(tableSize);
96 /* First lay down the symbols in order.
97 * We use a uint64_t to lay down 8 bytes at a time. This reduces branch
98 * misses since small blocks generally have small table logs, so nearly
99 * all symbols have counts <= 8. We ensure we have 8 bytes at the end of
100 * our buffer to handle the over-write.
103 U64 const add = 0x0101010101010101ull;
107 for (s=0; s<maxSV1; ++s, sv += add) {
109 int const n = normalizedCounter[s];
110 MEM_write64(spread + pos, sv);
111 for (i = 8; i < n; i += 8) {
112 MEM_write64(spread + pos + i, sv);
117 /* Now we spread those positions across the table.
118 * The benefit of doing it in two stages is that we avoid the
119 * variable size inner loop, which caused lots of branch misses.
120 * Now we can run through all the positions without any branch misses.
121 * We unroll the loop twice, since that is what empirically worked best.
126 size_t const unroll = 2;
127 assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
128 for (s = 0; s < (size_t)tableSize; s += unroll) {
130 for (u = 0; u < unroll; ++u) {
131 size_t const uPosition = (position + (u * step)) & tableMask;
132 tableDecode[uPosition].symbol = spread[s + u];
134 position = (position + (unroll * step)) & tableMask;
136 assert(position == 0);
139 U32 const tableMask = tableSize-1;
140 U32 const step = FSE_TABLESTEP(tableSize);
142 for (s=0; s<maxSV1; s++) {
144 for (i=0; i<normalizedCounter[s]; i++) {
145 tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
146 position = (position + step) & tableMask;
147 while (position > highThreshold) position = (position + step) & tableMask; /* lowprob area */
149 if (position!=0) return ERROR(GENERIC); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
152 /* Build Decoding table */
154 for (u=0; u<tableSize; u++) {
155 FSE_FUNCTION_TYPE const symbol = (FSE_FUNCTION_TYPE)(tableDecode[u].symbol);
156 U32 const nextState = symbolNext[symbol]++;
157 tableDecode[u].nbBits = (BYTE) (tableLog - ZSTD_highbit32(nextState) );
158 tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
164 size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
166 return FSE_buildDTable_internal(dt, normalizedCounter, maxSymbolValue, tableLog, workSpace, wkspSize);
170 #ifndef FSE_COMMONDEFS_ONLY
172 /*-*******************************************************
173 * Decompression (Byte symbols)
174 *********************************************************/
176 FORCE_INLINE_TEMPLATE size_t FSE_decompress_usingDTable_generic(
177 void* dst, size_t maxDstSize,
178 const void* cSrc, size_t cSrcSize,
179 const FSE_DTable* dt, const unsigned fast)
181 BYTE* const ostart = (BYTE*) dst;
183 BYTE* const omax = op + maxDstSize;
184 BYTE* const olimit = omax-3;
191 CHECK_F(BIT_initDStream(&bitD, cSrc, cSrcSize));
193 FSE_initDState(&state1, &bitD, dt);
194 FSE_initDState(&state2, &bitD, dt);
196 #define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
198 /* 4 symbols per loop */
199 for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) & (op<olimit) ; op+=4) {
200 op[0] = FSE_GETSYMBOL(&state1);
202 if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8) /* This test must be static */
203 BIT_reloadDStream(&bitD);
205 op[1] = FSE_GETSYMBOL(&state2);
207 if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8) /* This test must be static */
208 { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
210 op[2] = FSE_GETSYMBOL(&state1);
212 if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8) /* This test must be static */
213 BIT_reloadDStream(&bitD);
215 op[3] = FSE_GETSYMBOL(&state2);
219 /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
221 if (op>(omax-2)) return ERROR(dstSize_tooSmall);
222 *op++ = FSE_GETSYMBOL(&state1);
223 if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
224 *op++ = FSE_GETSYMBOL(&state2);
228 if (op>(omax-2)) return ERROR(dstSize_tooSmall);
229 *op++ = FSE_GETSYMBOL(&state2);
230 if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
231 *op++ = FSE_GETSYMBOL(&state1);
239 short ncount[FSE_MAX_SYMBOL_VALUE + 1];
240 FSE_DTable dtable[1]; /* Dynamically sized */
241 } FSE_DecompressWksp;
244 FORCE_INLINE_TEMPLATE size_t FSE_decompress_wksp_body(
245 void* dst, size_t dstCapacity,
246 const void* cSrc, size_t cSrcSize,
247 unsigned maxLog, void* workSpace, size_t wkspSize,
250 const BYTE* const istart = (const BYTE*)cSrc;
251 const BYTE* ip = istart;
253 unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
254 FSE_DecompressWksp* const wksp = (FSE_DecompressWksp*)workSpace;
256 DEBUG_STATIC_ASSERT((FSE_MAX_SYMBOL_VALUE + 1) % 2 == 0);
257 if (wkspSize < sizeof(*wksp)) return ERROR(GENERIC);
259 /* normal FSE decoding mode */
261 size_t const NCountLength = FSE_readNCount_bmi2(wksp->ncount, &maxSymbolValue, &tableLog, istart, cSrcSize, bmi2);
262 if (FSE_isError(NCountLength)) return NCountLength;
263 if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
264 assert(NCountLength <= cSrcSize);
266 cSrcSize -= NCountLength;
269 if (FSE_DECOMPRESS_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(tableLog_tooLarge);
270 assert(sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog) <= wkspSize);
271 workSpace = (BYTE*)workSpace + sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
272 wkspSize -= sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
274 CHECK_F( FSE_buildDTable_internal(wksp->dtable, wksp->ncount, maxSymbolValue, tableLog, workSpace, wkspSize) );
277 const void* ptr = wksp->dtable;
278 const FSE_DTableHeader* DTableH = (const FSE_DTableHeader*)ptr;
279 const U32 fastMode = DTableH->fastMode;
281 /* select fast mode (static) */
282 if (fastMode) return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 1);
283 return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 0);
287 /* Avoids the FORCE_INLINE of the _body() function. */
288 static size_t FSE_decompress_wksp_body_default(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
290 return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 0);
294 BMI2_TARGET_ATTRIBUTE static size_t FSE_decompress_wksp_body_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
296 return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 1);
300 size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2)
304 return FSE_decompress_wksp_body_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
308 return FSE_decompress_wksp_body_default(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
311 #endif /* FSE_COMMONDEFS_ONLY */