Commit | Line | Data |
---|---|---|
e0c1b49f NT |
1 | /* ****************************************************************** |
2 | * Common functions of New Generation Entropy library | |
3 | * Copyright (c) Yann Collet, Facebook, Inc. | |
4 | * | |
5 | * You can contact the author at : | |
6 | * - FSE+HUF 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 | * Dependencies | |
17 | ***************************************/ | |
637a642f | 18 | #include <linux/module.h> |
e0c1b49f NT |
19 | #include "mem.h" |
20 | #include "error_private.h" /* ERR_*, ERROR */ | |
21 | #define FSE_STATIC_LINKING_ONLY /* FSE_MIN_TABLELOG */ | |
22 | #include "fse.h" | |
23 | #define HUF_STATIC_LINKING_ONLY /* HUF_TABLELOG_ABSOLUTEMAX */ | |
24 | #include "huf.h" | |
25 | ||
26 | ||
27 | /*=== Version ===*/ | |
28 | unsigned FSE_versionNumber(void) { return FSE_VERSION_NUMBER; } | |
29 | ||
30 | ||
31 | /*=== Error Management ===*/ | |
32 | unsigned FSE_isError(size_t code) { return ERR_isError(code); } | |
33 | const char* FSE_getErrorName(size_t code) { return ERR_getErrorName(code); } | |
34 | ||
35 | unsigned HUF_isError(size_t code) { return ERR_isError(code); } | |
36 | const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); } | |
37 | ||
38 | ||
39 | /*-************************************************************** | |
40 | * FSE NCount encoding-decoding | |
41 | ****************************************************************/ | |
42 | static U32 FSE_ctz(U32 val) | |
43 | { | |
44 | assert(val != 0); | |
45 | { | |
46 | # if (__GNUC__ >= 3) /* GCC Intrinsic */ | |
47 | return __builtin_ctz(val); | |
48 | # else /* Software version */ | |
49 | U32 count = 0; | |
50 | while ((val & 1) == 0) { | |
51 | val >>= 1; | |
52 | ++count; | |
53 | } | |
54 | return count; | |
55 | # endif | |
56 | } | |
57 | } | |
58 | ||
59 | FORCE_INLINE_TEMPLATE | |
60 | size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr, | |
61 | const void* headerBuffer, size_t hbSize) | |
62 | { | |
63 | const BYTE* const istart = (const BYTE*) headerBuffer; | |
64 | const BYTE* const iend = istart + hbSize; | |
65 | const BYTE* ip = istart; | |
66 | int nbBits; | |
67 | int remaining; | |
68 | int threshold; | |
69 | U32 bitStream; | |
70 | int bitCount; | |
71 | unsigned charnum = 0; | |
72 | unsigned const maxSV1 = *maxSVPtr + 1; | |
73 | int previous0 = 0; | |
74 | ||
75 | if (hbSize < 8) { | |
76 | /* This function only works when hbSize >= 8 */ | |
77 | char buffer[8] = {0}; | |
78 | ZSTD_memcpy(buffer, headerBuffer, hbSize); | |
79 | { size_t const countSize = FSE_readNCount(normalizedCounter, maxSVPtr, tableLogPtr, | |
80 | buffer, sizeof(buffer)); | |
81 | if (FSE_isError(countSize)) return countSize; | |
82 | if (countSize > hbSize) return ERROR(corruption_detected); | |
83 | return countSize; | |
84 | } } | |
85 | assert(hbSize >= 8); | |
86 | ||
87 | /* init */ | |
88 | ZSTD_memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0])); /* all symbols not present in NCount have a frequency of 0 */ | |
89 | bitStream = MEM_readLE32(ip); | |
90 | nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG; /* extract tableLog */ | |
91 | if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge); | |
92 | bitStream >>= 4; | |
93 | bitCount = 4; | |
94 | *tableLogPtr = nbBits; | |
95 | remaining = (1<<nbBits)+1; | |
96 | threshold = 1<<nbBits; | |
97 | nbBits++; | |
98 | ||
99 | for (;;) { | |
100 | if (previous0) { | |
101 | /* Count the number of repeats. Each time the | |
102 | * 2-bit repeat code is 0b11 there is another | |
103 | * repeat. | |
104 | * Avoid UB by setting the high bit to 1. | |
105 | */ | |
106 | int repeats = FSE_ctz(~bitStream | 0x80000000) >> 1; | |
107 | while (repeats >= 12) { | |
108 | charnum += 3 * 12; | |
109 | if (LIKELY(ip <= iend-7)) { | |
110 | ip += 3; | |
111 | } else { | |
112 | bitCount -= (int)(8 * (iend - 7 - ip)); | |
113 | bitCount &= 31; | |
114 | ip = iend - 4; | |
115 | } | |
116 | bitStream = MEM_readLE32(ip) >> bitCount; | |
117 | repeats = FSE_ctz(~bitStream | 0x80000000) >> 1; | |
118 | } | |
119 | charnum += 3 * repeats; | |
120 | bitStream >>= 2 * repeats; | |
121 | bitCount += 2 * repeats; | |
122 | ||
123 | /* Add the final repeat which isn't 0b11. */ | |
124 | assert((bitStream & 3) < 3); | |
125 | charnum += bitStream & 3; | |
126 | bitCount += 2; | |
127 | ||
128 | /* This is an error, but break and return an error | |
129 | * at the end, because returning out of a loop makes | |
130 | * it harder for the compiler to optimize. | |
131 | */ | |
132 | if (charnum >= maxSV1) break; | |
133 | ||
134 | /* We don't need to set the normalized count to 0 | |
135 | * because we already memset the whole buffer to 0. | |
136 | */ | |
137 | ||
138 | if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) { | |
139 | assert((bitCount >> 3) <= 3); /* For first condition to work */ | |
140 | ip += bitCount>>3; | |
141 | bitCount &= 7; | |
142 | } else { | |
143 | bitCount -= (int)(8 * (iend - 4 - ip)); | |
144 | bitCount &= 31; | |
145 | ip = iend - 4; | |
146 | } | |
147 | bitStream = MEM_readLE32(ip) >> bitCount; | |
148 | } | |
149 | { | |
150 | int const max = (2*threshold-1) - remaining; | |
151 | int count; | |
152 | ||
153 | if ((bitStream & (threshold-1)) < (U32)max) { | |
154 | count = bitStream & (threshold-1); | |
155 | bitCount += nbBits-1; | |
156 | } else { | |
157 | count = bitStream & (2*threshold-1); | |
158 | if (count >= threshold) count -= max; | |
159 | bitCount += nbBits; | |
160 | } | |
161 | ||
162 | count--; /* extra accuracy */ | |
163 | /* When it matters (small blocks), this is a | |
164 | * predictable branch, because we don't use -1. | |
165 | */ | |
166 | if (count >= 0) { | |
167 | remaining -= count; | |
168 | } else { | |
169 | assert(count == -1); | |
170 | remaining += count; | |
171 | } | |
172 | normalizedCounter[charnum++] = (short)count; | |
173 | previous0 = !count; | |
174 | ||
175 | assert(threshold > 1); | |
176 | if (remaining < threshold) { | |
177 | /* This branch can be folded into the | |
178 | * threshold update condition because we | |
179 | * know that threshold > 1. | |
180 | */ | |
181 | if (remaining <= 1) break; | |
182 | nbBits = BIT_highbit32(remaining) + 1; | |
183 | threshold = 1 << (nbBits - 1); | |
184 | } | |
185 | if (charnum >= maxSV1) break; | |
186 | ||
187 | if (LIKELY(ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) { | |
188 | ip += bitCount>>3; | |
189 | bitCount &= 7; | |
190 | } else { | |
191 | bitCount -= (int)(8 * (iend - 4 - ip)); | |
192 | bitCount &= 31; | |
193 | ip = iend - 4; | |
194 | } | |
195 | bitStream = MEM_readLE32(ip) >> bitCount; | |
196 | } } | |
197 | if (remaining != 1) return ERROR(corruption_detected); | |
198 | /* Only possible when there are too many zeros. */ | |
199 | if (charnum > maxSV1) return ERROR(maxSymbolValue_tooSmall); | |
200 | if (bitCount > 32) return ERROR(corruption_detected); | |
201 | *maxSVPtr = charnum-1; | |
202 | ||
203 | ip += (bitCount+7)>>3; | |
204 | return ip-istart; | |
205 | } | |
206 | ||
207 | /* Avoids the FORCE_INLINE of the _body() function. */ | |
208 | static size_t FSE_readNCount_body_default( | |
209 | short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr, | |
210 | const void* headerBuffer, size_t hbSize) | |
211 | { | |
212 | return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize); | |
213 | } | |
214 | ||
215 | #if DYNAMIC_BMI2 | |
216 | TARGET_ATTRIBUTE("bmi2") static size_t FSE_readNCount_body_bmi2( | |
217 | short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr, | |
218 | const void* headerBuffer, size_t hbSize) | |
219 | { | |
220 | return FSE_readNCount_body(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize); | |
221 | } | |
222 | #endif | |
223 | ||
224 | size_t FSE_readNCount_bmi2( | |
225 | short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr, | |
226 | const void* headerBuffer, size_t hbSize, int bmi2) | |
227 | { | |
228 | #if DYNAMIC_BMI2 | |
229 | if (bmi2) { | |
230 | return FSE_readNCount_body_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize); | |
231 | } | |
232 | #endif | |
233 | (void)bmi2; | |
234 | return FSE_readNCount_body_default(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize); | |
235 | } | |
236 | ||
237 | size_t FSE_readNCount( | |
238 | short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr, | |
239 | const void* headerBuffer, size_t hbSize) | |
240 | { | |
241 | return FSE_readNCount_bmi2(normalizedCounter, maxSVPtr, tableLogPtr, headerBuffer, hbSize, /* bmi2 */ 0); | |
242 | } | |
637a642f | 243 | EXPORT_SYMBOL_GPL(FSE_readNCount); |
e0c1b49f NT |
244 | |
245 | /*! HUF_readStats() : | |
246 | Read compact Huffman tree, saved by HUF_writeCTable(). | |
247 | `huffWeight` is destination buffer. | |
248 | `rankStats` is assumed to be a table of at least HUF_TABLELOG_MAX U32. | |
249 | @return : size read from `src` , or an error Code . | |
250 | Note : Needed by HUF_readCTable() and HUF_readDTableX?() . | |
251 | */ | |
252 | size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats, | |
253 | U32* nbSymbolsPtr, U32* tableLogPtr, | |
254 | const void* src, size_t srcSize) | |
255 | { | |
256 | U32 wksp[HUF_READ_STATS_WORKSPACE_SIZE_U32]; | |
257 | return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp), /* bmi2 */ 0); | |
258 | } | |
637a642f | 259 | EXPORT_SYMBOL_GPL(HUF_readStats); |
e0c1b49f NT |
260 | |
261 | FORCE_INLINE_TEMPLATE size_t | |
262 | HUF_readStats_body(BYTE* huffWeight, size_t hwSize, U32* rankStats, | |
263 | U32* nbSymbolsPtr, U32* tableLogPtr, | |
264 | const void* src, size_t srcSize, | |
265 | void* workSpace, size_t wkspSize, | |
266 | int bmi2) | |
267 | { | |
268 | U32 weightTotal; | |
269 | const BYTE* ip = (const BYTE*) src; | |
270 | size_t iSize; | |
271 | size_t oSize; | |
272 | ||
273 | if (!srcSize) return ERROR(srcSize_wrong); | |
274 | iSize = ip[0]; | |
275 | /* ZSTD_memset(huffWeight, 0, hwSize); *//* is not necessary, even though some analyzer complain ... */ | |
276 | ||
277 | if (iSize >= 128) { /* special header */ | |
278 | oSize = iSize - 127; | |
279 | iSize = ((oSize+1)/2); | |
280 | if (iSize+1 > srcSize) return ERROR(srcSize_wrong); | |
281 | if (oSize >= hwSize) return ERROR(corruption_detected); | |
282 | ip += 1; | |
283 | { U32 n; | |
284 | for (n=0; n<oSize; n+=2) { | |
285 | huffWeight[n] = ip[n/2] >> 4; | |
286 | huffWeight[n+1] = ip[n/2] & 15; | |
287 | } } } | |
288 | else { /* header compressed with FSE (normal case) */ | |
289 | if (iSize+1 > srcSize) return ERROR(srcSize_wrong); | |
290 | /* max (hwSize-1) values decoded, as last one is implied */ | |
291 | oSize = FSE_decompress_wksp_bmi2(huffWeight, hwSize-1, ip+1, iSize, 6, workSpace, wkspSize, bmi2); | |
292 | if (FSE_isError(oSize)) return oSize; | |
293 | } | |
294 | ||
295 | /* collect weight stats */ | |
296 | ZSTD_memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32)); | |
297 | weightTotal = 0; | |
298 | { U32 n; for (n=0; n<oSize; n++) { | |
299 | if (huffWeight[n] >= HUF_TABLELOG_MAX) return ERROR(corruption_detected); | |
300 | rankStats[huffWeight[n]]++; | |
301 | weightTotal += (1 << huffWeight[n]) >> 1; | |
302 | } } | |
303 | if (weightTotal == 0) return ERROR(corruption_detected); | |
304 | ||
305 | /* get last non-null symbol weight (implied, total must be 2^n) */ | |
306 | { U32 const tableLog = BIT_highbit32(weightTotal) + 1; | |
307 | if (tableLog > HUF_TABLELOG_MAX) return ERROR(corruption_detected); | |
308 | *tableLogPtr = tableLog; | |
309 | /* determine last weight */ | |
310 | { U32 const total = 1 << tableLog; | |
311 | U32 const rest = total - weightTotal; | |
312 | U32 const verif = 1 << BIT_highbit32(rest); | |
313 | U32 const lastWeight = BIT_highbit32(rest) + 1; | |
314 | if (verif != rest) return ERROR(corruption_detected); /* last value must be a clean power of 2 */ | |
315 | huffWeight[oSize] = (BYTE)lastWeight; | |
316 | rankStats[lastWeight]++; | |
317 | } } | |
318 | ||
319 | /* check tree construction validity */ | |
320 | if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected); /* by construction : at least 2 elts of rank 1, must be even */ | |
321 | ||
322 | /* results */ | |
323 | *nbSymbolsPtr = (U32)(oSize+1); | |
324 | return iSize+1; | |
325 | } | |
326 | ||
327 | /* Avoids the FORCE_INLINE of the _body() function. */ | |
328 | static size_t HUF_readStats_body_default(BYTE* huffWeight, size_t hwSize, U32* rankStats, | |
329 | U32* nbSymbolsPtr, U32* tableLogPtr, | |
330 | const void* src, size_t srcSize, | |
331 | void* workSpace, size_t wkspSize) | |
332 | { | |
333 | return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 0); | |
334 | } | |
335 | ||
336 | #if DYNAMIC_BMI2 | |
337 | static TARGET_ATTRIBUTE("bmi2") size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats, | |
338 | U32* nbSymbolsPtr, U32* tableLogPtr, | |
339 | const void* src, size_t srcSize, | |
340 | void* workSpace, size_t wkspSize) | |
341 | { | |
342 | return HUF_readStats_body(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize, 1); | |
343 | } | |
344 | #endif | |
345 | ||
346 | size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize, U32* rankStats, | |
347 | U32* nbSymbolsPtr, U32* tableLogPtr, | |
348 | const void* src, size_t srcSize, | |
349 | void* workSpace, size_t wkspSize, | |
350 | int bmi2) | |
351 | { | |
352 | #if DYNAMIC_BMI2 | |
353 | if (bmi2) { | |
354 | return HUF_readStats_body_bmi2(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize); | |
355 | } | |
356 | #endif | |
357 | (void)bmi2; | |
358 | return HUF_readStats_body_default(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize); | |
359 | } | |
637a642f | 360 | EXPORT_SYMBOL_GPL(HUF_readStats_wksp); |