Mercurial > hg
comparison contrib/python-zstandard/zstd/compress/zstd_double_fast.c @ 37495:b1fb341d8a61
zstandard: vendor python-zstandard 0.9.0
This was just released. It features a number of goodies. More info at
https://gregoryszorc.com/blog/2018/04/09/release-of-python-zstandard-0.9/.
The clang-format ignore list was updated to reflect the new source
of files.
The project contains a vendored copy of zstandard 1.3.4. The old
version was 1.1.3. One of the changes between those versions is that
zstandard is now dual licensed BSD + GPLv2 and the patent rights grant
has been removed. Good riddance.
The API should be backwards compatible. So no changes in core
should be needed. However, there were a number of changes in the
library that we'll want to adapt to. Those will be addressed in
subsequent commits.
Differential Revision: https://phab.mercurial-scm.org/D3198
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Mon, 09 Apr 2018 10:13:29 -0700 |
parents | |
children | 73fef626dae3 |
comparison
equal
deleted
inserted
replaced
37494:1ce7a55b09d1 | 37495:b1fb341d8a61 |
---|---|
1 /* | |
2 * Copyright (c) 2016-present, 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 #include "zstd_compress_internal.h" | |
12 #include "zstd_double_fast.h" | |
13 | |
14 | |
15 void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms, | |
16 ZSTD_compressionParameters const* cParams, | |
17 void const* end) | |
18 { | |
19 U32* const hashLarge = ms->hashTable; | |
20 U32 const hBitsL = cParams->hashLog; | |
21 U32 const mls = cParams->searchLength; | |
22 U32* const hashSmall = ms->chainTable; | |
23 U32 const hBitsS = cParams->chainLog; | |
24 const BYTE* const base = ms->window.base; | |
25 const BYTE* ip = base + ms->nextToUpdate; | |
26 const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE; | |
27 const U32 fastHashFillStep = 3; | |
28 | |
29 /* Always insert every fastHashFillStep position into the hash tables. | |
30 * Insert the other positions into the large hash table if their entry | |
31 * is empty. | |
32 */ | |
33 for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) { | |
34 U32 const current = (U32)(ip - base); | |
35 U32 i; | |
36 for (i = 0; i < fastHashFillStep; ++i) { | |
37 size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls); | |
38 size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8); | |
39 if (i == 0) | |
40 hashSmall[smHash] = current + i; | |
41 if (i == 0 || hashLarge[lgHash] == 0) | |
42 hashLarge[lgHash] = current + i; | |
43 } | |
44 } | |
45 } | |
46 | |
47 | |
48 FORCE_INLINE_TEMPLATE | |
49 size_t ZSTD_compressBlock_doubleFast_generic( | |
50 ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | |
51 ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize, | |
52 U32 const mls /* template */) | |
53 { | |
54 U32* const hashLong = ms->hashTable; | |
55 const U32 hBitsL = cParams->hashLog; | |
56 U32* const hashSmall = ms->chainTable; | |
57 const U32 hBitsS = cParams->chainLog; | |
58 const BYTE* const base = ms->window.base; | |
59 const BYTE* const istart = (const BYTE*)src; | |
60 const BYTE* ip = istart; | |
61 const BYTE* anchor = istart; | |
62 const U32 lowestIndex = ms->window.dictLimit; | |
63 const BYTE* const lowest = base + lowestIndex; | |
64 const BYTE* const iend = istart + srcSize; | |
65 const BYTE* const ilimit = iend - HASH_READ_SIZE; | |
66 U32 offset_1=rep[0], offset_2=rep[1]; | |
67 U32 offsetSaved = 0; | |
68 | |
69 /* init */ | |
70 ip += (ip==lowest); | |
71 { U32 const maxRep = (U32)(ip-lowest); | |
72 if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0; | |
73 if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0; | |
74 } | |
75 | |
76 /* Main Search Loop */ | |
77 while (ip < ilimit) { /* < instead of <=, because repcode check at (ip+1) */ | |
78 size_t mLength; | |
79 size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8); | |
80 size_t const h = ZSTD_hashPtr(ip, hBitsS, mls); | |
81 U32 const current = (U32)(ip-base); | |
82 U32 const matchIndexL = hashLong[h2]; | |
83 U32 const matchIndexS = hashSmall[h]; | |
84 const BYTE* matchLong = base + matchIndexL; | |
85 const BYTE* match = base + matchIndexS; | |
86 hashLong[h2] = hashSmall[h] = current; /* update hash tables */ | |
87 | |
88 assert(offset_1 <= current); /* supposed guaranteed by construction */ | |
89 if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) { | |
90 /* favor repcode */ | |
91 mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4; | |
92 ip++; | |
93 ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH); | |
94 } else { | |
95 U32 offset; | |
96 if ( (matchIndexL > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip)) ) { | |
97 mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8; | |
98 offset = (U32)(ip-matchLong); | |
99 while (((ip>anchor) & (matchLong>lowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */ | |
100 } else if ( (matchIndexS > lowestIndex) && (MEM_read32(match) == MEM_read32(ip)) ) { | |
101 size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8); | |
102 U32 const matchIndexL3 = hashLong[hl3]; | |
103 const BYTE* matchL3 = base + matchIndexL3; | |
104 hashLong[hl3] = current + 1; | |
105 if ( (matchIndexL3 > lowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1)) ) { | |
106 mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8; | |
107 ip++; | |
108 offset = (U32)(ip-matchL3); | |
109 while (((ip>anchor) & (matchL3>lowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */ | |
110 } else { | |
111 mLength = ZSTD_count(ip+4, match+4, iend) + 4; | |
112 offset = (U32)(ip-match); | |
113 while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */ | |
114 } | |
115 } else { | |
116 ip += ((ip-anchor) >> kSearchStrength) + 1; | |
117 continue; | |
118 } | |
119 | |
120 offset_2 = offset_1; | |
121 offset_1 = offset; | |
122 | |
123 ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH); | |
124 } | |
125 | |
126 /* match found */ | |
127 ip += mLength; | |
128 anchor = ip; | |
129 | |
130 if (ip <= ilimit) { | |
131 /* Fill Table */ | |
132 hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = | |
133 hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2; /* here because current+2 could be > iend-8 */ | |
134 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = | |
135 hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base); | |
136 | |
137 /* check immediate repcode */ | |
138 while ( (ip <= ilimit) | |
139 && ( (offset_2>0) | |
140 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) { | |
141 /* store sequence */ | |
142 size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4; | |
143 { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */ | |
144 hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base); | |
145 hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base); | |
146 ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH); | |
147 ip += rLength; | |
148 anchor = ip; | |
149 continue; /* faster when present ... (?) */ | |
150 } } } | |
151 | |
152 /* save reps for next block */ | |
153 rep[0] = offset_1 ? offset_1 : offsetSaved; | |
154 rep[1] = offset_2 ? offset_2 : offsetSaved; | |
155 | |
156 /* Return the last literals size */ | |
157 return iend - anchor; | |
158 } | |
159 | |
160 | |
161 size_t ZSTD_compressBlock_doubleFast( | |
162 ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | |
163 ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize) | |
164 { | |
165 const U32 mls = cParams->searchLength; | |
166 switch(mls) | |
167 { | |
168 default: /* includes case 3 */ | |
169 case 4 : | |
170 return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 4); | |
171 case 5 : | |
172 return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 5); | |
173 case 6 : | |
174 return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 6); | |
175 case 7 : | |
176 return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 7); | |
177 } | |
178 } | |
179 | |
180 | |
181 static size_t ZSTD_compressBlock_doubleFast_extDict_generic( | |
182 ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | |
183 ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize, | |
184 U32 const mls /* template */) | |
185 { | |
186 U32* const hashLong = ms->hashTable; | |
187 U32 const hBitsL = cParams->hashLog; | |
188 U32* const hashSmall = ms->chainTable; | |
189 U32 const hBitsS = cParams->chainLog; | |
190 const BYTE* const base = ms->window.base; | |
191 const BYTE* const dictBase = ms->window.dictBase; | |
192 const BYTE* const istart = (const BYTE*)src; | |
193 const BYTE* ip = istart; | |
194 const BYTE* anchor = istart; | |
195 const U32 lowestIndex = ms->window.lowLimit; | |
196 const BYTE* const dictStart = dictBase + lowestIndex; | |
197 const U32 dictLimit = ms->window.dictLimit; | |
198 const BYTE* const lowPrefixPtr = base + dictLimit; | |
199 const BYTE* const dictEnd = dictBase + dictLimit; | |
200 const BYTE* const iend = istart + srcSize; | |
201 const BYTE* const ilimit = iend - 8; | |
202 U32 offset_1=rep[0], offset_2=rep[1]; | |
203 | |
204 /* Search Loop */ | |
205 while (ip < ilimit) { /* < instead of <=, because (ip+1) */ | |
206 const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls); | |
207 const U32 matchIndex = hashSmall[hSmall]; | |
208 const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base; | |
209 const BYTE* match = matchBase + matchIndex; | |
210 | |
211 const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8); | |
212 const U32 matchLongIndex = hashLong[hLong]; | |
213 const BYTE* matchLongBase = matchLongIndex < dictLimit ? dictBase : base; | |
214 const BYTE* matchLong = matchLongBase + matchLongIndex; | |
215 | |
216 const U32 current = (U32)(ip-base); | |
217 const U32 repIndex = current + 1 - offset_1; /* offset_1 expected <= current +1 */ | |
218 const BYTE* repBase = repIndex < dictLimit ? dictBase : base; | |
219 const BYTE* repMatch = repBase + repIndex; | |
220 size_t mLength; | |
221 hashSmall[hSmall] = hashLong[hLong] = current; /* update hash table */ | |
222 | |
223 if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex)) | |
224 && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) { | |
225 const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend; | |
226 mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4; | |
227 ip++; | |
228 ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH); | |
229 } else { | |
230 if ((matchLongIndex > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) { | |
231 const BYTE* matchEnd = matchLongIndex < dictLimit ? dictEnd : iend; | |
232 const BYTE* lowMatchPtr = matchLongIndex < dictLimit ? dictStart : lowPrefixPtr; | |
233 U32 offset; | |
234 mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, lowPrefixPtr) + 8; | |
235 offset = current - matchLongIndex; | |
236 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */ | |
237 offset_2 = offset_1; | |
238 offset_1 = offset; | |
239 ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH); | |
240 | |
241 } else if ((matchIndex > lowestIndex) && (MEM_read32(match) == MEM_read32(ip))) { | |
242 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8); | |
243 U32 const matchIndex3 = hashLong[h3]; | |
244 const BYTE* const match3Base = matchIndex3 < dictLimit ? dictBase : base; | |
245 const BYTE* match3 = match3Base + matchIndex3; | |
246 U32 offset; | |
247 hashLong[h3] = current + 1; | |
248 if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) { | |
249 const BYTE* matchEnd = matchIndex3 < dictLimit ? dictEnd : iend; | |
250 const BYTE* lowMatchPtr = matchIndex3 < dictLimit ? dictStart : lowPrefixPtr; | |
251 mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, lowPrefixPtr) + 8; | |
252 ip++; | |
253 offset = current+1 - matchIndex3; | |
254 while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */ | |
255 } else { | |
256 const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend; | |
257 const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr; | |
258 mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4; | |
259 offset = current - matchIndex; | |
260 while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */ | |
261 } | |
262 offset_2 = offset_1; | |
263 offset_1 = offset; | |
264 ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH); | |
265 | |
266 } else { | |
267 ip += ((ip-anchor) >> kSearchStrength) + 1; | |
268 continue; | |
269 } } | |
270 | |
271 /* found a match : store it */ | |
272 ip += mLength; | |
273 anchor = ip; | |
274 | |
275 if (ip <= ilimit) { | |
276 /* Fill Table */ | |
277 hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2; | |
278 hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = current+2; | |
279 hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base); | |
280 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base); | |
281 /* check immediate repcode */ | |
282 while (ip <= ilimit) { | |
283 U32 const current2 = (U32)(ip-base); | |
284 U32 const repIndex2 = current2 - offset_2; | |
285 const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2; | |
286 if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex)) /* intentional overflow */ | |
287 && (MEM_read32(repMatch2) == MEM_read32(ip)) ) { | |
288 const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend; | |
289 size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4; | |
290 U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */ | |
291 ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH); | |
292 hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2; | |
293 hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2; | |
294 ip += repLength2; | |
295 anchor = ip; | |
296 continue; | |
297 } | |
298 break; | |
299 } } } | |
300 | |
301 /* save reps for next block */ | |
302 rep[0] = offset_1; | |
303 rep[1] = offset_2; | |
304 | |
305 /* Return the last literals size */ | |
306 return iend - anchor; | |
307 } | |
308 | |
309 | |
310 size_t ZSTD_compressBlock_doubleFast_extDict( | |
311 ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], | |
312 ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize) | |
313 { | |
314 U32 const mls = cParams->searchLength; | |
315 switch(mls) | |
316 { | |
317 default: /* includes case 3 */ | |
318 case 4 : | |
319 return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 4); | |
320 case 5 : | |
321 return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 5); | |
322 case 6 : | |
323 return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 6); | |
324 case 7 : | |
325 return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 7); | |
326 } | |
327 } |