Coverage Report

Created: 2025-11-16 06:36

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/src/perfetto/buildtools/zstd/lib/common/fse_decompress.c
Line
Count
Source
1
/* ******************************************************************
2
 * FSE : Finite State Entropy decoder
3
 * Copyright (c) Meta Platforms, Inc. and affiliates.
4
 *
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
8
 *
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
****************************************************************** */
14
15
16
/* **************************************************************
17
*  Includes
18
****************************************************************/
19
#include "debug.h"      /* assert */
20
#include "bitstream.h"
21
#include "compiler.h"
22
#define FSE_STATIC_LINKING_ONLY
23
#include "fse.h"
24
#include "error_private.h"
25
#define ZSTD_DEPS_NEED_MALLOC
26
#include "zstd_deps.h"
27
#include "bits.h"       /* ZSTD_highbit32 */
28
29
30
/* **************************************************************
31
*  Error Management
32
****************************************************************/
33
88
#define FSE_isError ERR_isError
34
#define FSE_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)   /* use only *after* variable declarations */
35
36
37
/* **************************************************************
38
*  Templates
39
****************************************************************/
40
/*
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
44
*/
45
46
/* safety checks */
47
#ifndef FSE_FUNCTION_EXTENSION
48
#  error "FSE_FUNCTION_EXTENSION must be defined"
49
#endif
50
#ifndef FSE_FUNCTION_TYPE
51
#  error "FSE_FUNCTION_TYPE must be defined"
52
#endif
53
54
/* Function names */
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)
58
59
static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
60
80
{
61
80
    void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
62
80
    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
63
80
    U16* symbolNext = (U16*)workSpace;
64
80
    BYTE* spread = (BYTE*)(symbolNext + maxSymbolValue + 1);
65
66
80
    U32 const maxSV1 = maxSymbolValue + 1;
67
80
    U32 const tableSize = 1 << tableLog;
68
80
    U32 highThreshold = tableSize-1;
69
70
    /* Sanity Checks */
71
80
    if (FSE_BUILD_DTABLE_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(maxSymbolValue_tooLarge);
72
80
    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
73
80
    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
74
75
    /* Init, lay down lowprob symbols */
76
80
    {   FSE_DTableHeader DTableH;
77
80
        DTableH.tableLog = (U16)tableLog;
78
80
        DTableH.fastMode = 1;
79
80
        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
80
80
            U32 s;
81
1.18k
            for (s=0; s<maxSV1; s++) {
82
1.10k
                if (normalizedCounter[s]==-1) {
83
182
                    tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
84
182
                    symbolNext[s] = 1;
85
919
                } else {
86
919
                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
87
919
                    symbolNext[s] = normalizedCounter[s];
88
919
        }   }   }
89
80
        ZSTD_memcpy(dt, &DTableH, sizeof(DTableH));
90
80
    }
91
92
    /* Spread symbols */
93
80
    if (highThreshold == tableSize - 1) {
94
40
        size_t const tableMask = tableSize-1;
95
40
        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.
101
         */
102
40
        {
103
40
            U64 const add = 0x0101010101010101ull;
104
40
            size_t pos = 0;
105
40
            U64 sv = 0;
106
40
            U32 s;
107
381
            for (s=0; s<maxSV1; ++s, sv += add) {
108
341
                int i;
109
341
                int const n = normalizedCounter[s];
110
341
                MEM_write64(spread + pos, sv);
111
460
                for (i = 8; i < n; i += 8) {
112
119
                    MEM_write64(spread + pos + i, sv);
113
119
                }
114
341
                pos += n;
115
341
            }
116
40
        }
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.
122
         */
123
40
        {
124
40
            size_t position = 0;
125
40
            size_t s;
126
40
            size_t const unroll = 2;
127
40
            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
128
808
            for (s = 0; s < (size_t)tableSize; s += unroll) {
129
768
                size_t u;
130
2.30k
                for (u = 0; u < unroll; ++u) {
131
1.53k
                    size_t const uPosition = (position + (u * step)) & tableMask;
132
1.53k
                    tableDecode[uPosition].symbol = spread[s + u];
133
1.53k
                }
134
768
                position = (position + (unroll * step)) & tableMask;
135
768
            }
136
40
            assert(position == 0);
137
40
        }
138
40
    } else {
139
40
        U32 const tableMask = tableSize-1;
140
40
        U32 const step = FSE_TABLESTEP(tableSize);
141
40
        U32 s, position = 0;
142
800
        for (s=0; s<maxSV1; s++) {
143
760
            int i;
144
2.05k
            for (i=0; i<normalizedCounter[s]; i++) {
145
1.29k
                tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
146
1.29k
                position = (position + step) & tableMask;
147
1.44k
                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
148
1.29k
        }   }
149
40
        if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
150
40
    }
151
152
    /* Build Decoding table */
153
80
    {   U32 u;
154
3.08k
        for (u=0; u<tableSize; u++) {
155
3.00k
            FSE_FUNCTION_TYPE const symbol = (FSE_FUNCTION_TYPE)(tableDecode[u].symbol);
156
3.00k
            U32 const nextState = symbolNext[symbol]++;
157
3.00k
            tableDecode[u].nbBits = (BYTE) (tableLog - ZSTD_highbit32(nextState) );
158
3.00k
            tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
159
3.00k
    }   }
160
161
80
    return 0;
162
80
}
163
164
size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
165
0
{
166
0
    return FSE_buildDTable_internal(dt, normalizedCounter, maxSymbolValue, tableLog, workSpace, wkspSize);
167
0
}
168
169
170
#ifndef FSE_COMMONDEFS_ONLY
171
172
/*-*******************************************************
173
*  Decompression (Byte symbols)
174
*********************************************************/
175
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)
180
80
{
181
80
    BYTE* const ostart = (BYTE*) dst;
182
80
    BYTE* op = ostart;
183
80
    BYTE* const omax = op + maxDstSize;
184
80
    BYTE* const olimit = omax-3;
185
186
80
    BIT_DStream_t bitD;
187
80
    FSE_DState_t state1;
188
80
    FSE_DState_t state2;
189
190
    /* Init */
191
80
    CHECK_F(BIT_initDStream(&bitD, cSrc, cSrcSize));
192
193
72
    FSE_initDState(&state1, &bitD, dt);
194
72
    FSE_initDState(&state2, &bitD, dt);
195
196
4.73k
#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
197
198
    /* 4 symbols per loop */
199
450
    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) & (op<olimit) ; op+=4) {
200
378
        op[0] = FSE_GETSYMBOL(&state1);
201
202
378
        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
203
0
            BIT_reloadDStream(&bitD);
204
205
378
        op[1] = FSE_GETSYMBOL(&state2);
206
207
378
        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
208
0
            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
209
210
378
        op[2] = FSE_GETSYMBOL(&state1);
211
212
378
        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
213
0
            BIT_reloadDStream(&bitD);
214
215
378
        op[3] = FSE_GETSYMBOL(&state2);
216
378
    }
217
218
    /* tail */
219
    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
220
1.60k
    while (1) {
221
1.60k
        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
222
1.59k
        *op++ = FSE_GETSYMBOL(&state1);
223
1.59k
        if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
224
36
            *op++ = FSE_GETSYMBOL(&state2);
225
36
            break;
226
36
        }
227
228
1.55k
        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
229
1.55k
        *op++ = FSE_GETSYMBOL(&state2);
230
1.55k
        if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
231
30
            *op++ = FSE_GETSYMBOL(&state1);
232
30
            break;
233
30
    }   }
234
235
66
    return op-ostart;
236
72
}
237
238
typedef struct {
239
    short ncount[FSE_MAX_SYMBOL_VALUE + 1];
240
    FSE_DTable dtable[1]; /* Dynamically sized */
241
} FSE_DecompressWksp;
242
243
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,
248
        int bmi2)
249
88
{
250
88
    const BYTE* const istart = (const BYTE*)cSrc;
251
88
    const BYTE* ip = istart;
252
88
    unsigned tableLog;
253
88
    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
254
88
    FSE_DecompressWksp* const wksp = (FSE_DecompressWksp*)workSpace;
255
256
88
    DEBUG_STATIC_ASSERT((FSE_MAX_SYMBOL_VALUE + 1) % 2 == 0);
257
88
    if (wkspSize < sizeof(*wksp)) return ERROR(GENERIC);
258
259
    /* normal FSE decoding mode */
260
88
    {
261
88
        size_t const NCountLength = FSE_readNCount_bmi2(wksp->ncount, &maxSymbolValue, &tableLog, istart, cSrcSize, bmi2);
262
88
        if (FSE_isError(NCountLength)) return NCountLength;
263
82
        if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
264
81
        assert(NCountLength <= cSrcSize);
265
81
        ip += NCountLength;
266
81
        cSrcSize -= NCountLength;
267
81
    }
268
269
81
    if (FSE_DECOMPRESS_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(tableLog_tooLarge);
270
80
    assert(sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog) <= wkspSize);
271
80
    workSpace = (BYTE*)workSpace + sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
272
80
    wkspSize -= sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
273
274
80
    CHECK_F( FSE_buildDTable_internal(wksp->dtable, wksp->ncount, maxSymbolValue, tableLog, workSpace, wkspSize) );
275
276
80
    {
277
80
        const void* ptr = wksp->dtable;
278
80
        const FSE_DTableHeader* DTableH = (const FSE_DTableHeader*)ptr;
279
80
        const U32 fastMode = DTableH->fastMode;
280
281
        /* select fast mode (static) */
282
80
        if (fastMode) return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 1);
283
46
        return FSE_decompress_usingDTable_generic(dst, dstCapacity, ip, cSrcSize, wksp->dtable, 0);
284
80
    }
285
80
}
286
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)
289
88
{
290
88
    return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 0);
291
88
}
292
293
#if DYNAMIC_BMI2
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)
295
{
296
    return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 1);
297
}
298
#endif
299
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)
301
88
{
302
#if DYNAMIC_BMI2
303
    if (bmi2) {
304
        return FSE_decompress_wksp_body_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
305
    }
306
#endif
307
88
    (void)bmi2;
308
88
    return FSE_decompress_wksp_body_default(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
309
88
}
310
311
#endif   /* FSE_COMMONDEFS_ONLY */