1 | /* |
2 | lz4opt.h - Optimal Mode of LZ4 |
3 | Copyright (C) 2015-2017, Przemyslaw Skibinski <inikep@gmail.com> |
4 | Note : this file is intended to be included within lz4hc.c |
5 | |
6 | BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php) |
7 | |
8 | Redistribution and use in source and binary forms, with or without |
9 | modification, are permitted provided that the following conditions are |
10 | met: |
11 | |
12 | * Redistributions of source code must retain the above copyright |
13 | notice, this list of conditions and the following disclaimer. |
14 | * Redistributions in binary form must reproduce the above |
15 | copyright notice, this list of conditions and the following disclaimer |
16 | in the documentation and/or other materials provided with the |
17 | distribution. |
18 | |
19 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
20 | "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
21 | LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
22 | A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
23 | OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
24 | SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
25 | LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
26 | DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
27 | THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
28 | (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
29 | OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
30 | |
31 | You can contact the author at : |
32 | - LZ4 source repository : https://github.com/lz4/lz4 |
33 | - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c |
34 | */ |
35 | |
36 | #define LZ4_OPT_NUM (1<<12) |
37 | |
38 | typedef struct { |
39 | int price; |
40 | int off; |
41 | int mlen; |
42 | int litlen; |
43 | } LZ4HC_optimal_t; |
44 | |
45 | |
46 | /* price in bytes */ |
47 | LZ4_FORCE_INLINE int LZ4HC_literalsPrice(int const litlen) |
48 | { |
49 | int price = litlen; |
50 | if (litlen >= (int)RUN_MASK) |
51 | price += 1 + (litlen-RUN_MASK)/255; |
52 | return price; |
53 | } |
54 | |
55 | |
56 | /* requires mlen >= MINMATCH */ |
57 | LZ4_FORCE_INLINE int LZ4HC_sequencePrice(int litlen, int mlen) |
58 | { |
59 | int price = 1 + 2 ; /* token + 16-bit offset */ |
60 | |
61 | price += LZ4HC_literalsPrice(litlen); |
62 | |
63 | if (mlen >= (int)(ML_MASK+MINMATCH)) |
64 | price += 1 + (mlen-(ML_MASK+MINMATCH))/255; |
65 | |
66 | return price; |
67 | } |
68 | |
69 | |
70 | /*-************************************* |
71 | * Match finder |
72 | ***************************************/ |
73 | typedef struct { |
74 | int off; |
75 | int len; |
76 | } LZ4HC_match_t; |
77 | |
78 | LZ4_FORCE_INLINE |
79 | LZ4HC_match_t LZ4HC_FindLongerMatch(LZ4HC_CCtx_internal* const ctx, |
80 | const BYTE* ip, const BYTE* const iHighLimit, |
81 | int minLen, int nbSearches) |
82 | { |
83 | LZ4HC_match_t match = { 0 , 0 }; |
84 | const BYTE* matchPtr = NULL; |
85 | /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos), |
86 | * but this won't be the case here, as we define iLowLimit==ip, |
87 | * so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */ |
88 | int const matchLength = LZ4HC_InsertAndGetWiderMatch(ctx, ip, ip, iHighLimit, minLen, &matchPtr, &ip, nbSearches); |
89 | if (matchLength <= minLen) return match; |
90 | match.len = matchLength; |
91 | match.off = (int)(ip-matchPtr); |
92 | return match; |
93 | } |
94 | |
95 | |
96 | static int LZ4HC_compress_optimal ( |
97 | LZ4HC_CCtx_internal* ctx, |
98 | const char* const source, |
99 | char* dst, |
100 | int inputSize, |
101 | int dstCapacity, |
102 | limitedOutput_directive limit, |
103 | int const nbSearches, |
104 | size_t sufficient_len, |
105 | int const fullUpdate |
106 | ) |
107 | { |
108 | #define TRAILING_LITERALS 3 |
109 | LZ4HC_optimal_t opt[LZ4_OPT_NUM + TRAILING_LITERALS]; /* this uses a bit too much stack memory to my taste ... */ |
110 | |
111 | const BYTE* ip = (const BYTE*) source; |
112 | const BYTE* anchor = ip; |
113 | const BYTE* const iend = ip + inputSize; |
114 | const BYTE* const mflimit = iend - MFLIMIT; |
115 | const BYTE* const matchlimit = iend - LASTLITERALS; |
116 | BYTE* op = (BYTE*) dst; |
117 | BYTE* const oend = op + dstCapacity; |
118 | |
119 | /* init */ |
120 | DEBUGLOG(5, "LZ4HC_compress_optimal" ); |
121 | if (sufficient_len >= LZ4_OPT_NUM) sufficient_len = LZ4_OPT_NUM-1; |
122 | |
123 | /* Main Loop */ |
124 | assert(ip - anchor < LZ4_MAX_INPUT_SIZE); |
125 | while (ip < mflimit) { |
126 | int const llen = (int)(ip - anchor); |
127 | int best_mlen, best_off; |
128 | int cur, last_match_pos = 0; |
129 | |
130 | LZ4HC_match_t const firstMatch = LZ4HC_FindLongerMatch(ctx, ip, matchlimit, MINMATCH-1, nbSearches); |
131 | if (firstMatch.len==0) { ip++; continue; } |
132 | |
133 | if ((size_t)firstMatch.len > sufficient_len) { |
134 | /* good enough solution : immediate encoding */ |
135 | int const firstML = firstMatch.len; |
136 | const BYTE* const matchPos = ip - firstMatch.off; |
137 | if ( LZ4HC_encodeSequence(&ip, &op, &anchor, firstML, matchPos, limit, oend) ) /* updates ip, op and anchor */ |
138 | return 0; /* error */ |
139 | continue; |
140 | } |
141 | |
142 | /* set prices for first positions (literals) */ |
143 | { int rPos; |
144 | for (rPos = 0 ; rPos < MINMATCH ; rPos++) { |
145 | int const cost = LZ4HC_literalsPrice(llen + rPos); |
146 | opt[rPos].mlen = 1; |
147 | opt[rPos].off = 0; |
148 | opt[rPos].litlen = llen + rPos; |
149 | opt[rPos].price = cost; |
150 | DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup" , |
151 | rPos, cost, opt[rPos].litlen); |
152 | } } |
153 | /* set prices using initial match */ |
154 | { int mlen = MINMATCH; |
155 | int const matchML = firstMatch.len; /* necessarily < sufficient_len < LZ4_OPT_NUM */ |
156 | int const offset = firstMatch.off; |
157 | assert(matchML < LZ4_OPT_NUM); |
158 | for ( ; mlen <= matchML ; mlen++) { |
159 | int const cost = LZ4HC_sequencePrice(llen, mlen); |
160 | opt[mlen].mlen = mlen; |
161 | opt[mlen].off = offset; |
162 | opt[mlen].litlen = llen; |
163 | opt[mlen].price = cost; |
164 | DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i) -- initial setup" , |
165 | mlen, cost, mlen); |
166 | } } |
167 | last_match_pos = firstMatch.len; |
168 | { int addLit; |
169 | for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) { |
170 | opt[last_match_pos+addLit].mlen = 1; /* literal */ |
171 | opt[last_match_pos+addLit].off = 0; |
172 | opt[last_match_pos+addLit].litlen = addLit; |
173 | opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit); |
174 | DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup" , |
175 | last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit); |
176 | } } |
177 | |
178 | /* check further positions */ |
179 | for (cur = 1; cur < last_match_pos; cur++) { |
180 | const BYTE* const curPtr = ip + cur; |
181 | LZ4HC_match_t newMatch; |
182 | |
183 | if (curPtr >= mflimit) break; |
184 | DEBUGLOG(7, "rPos:%u[%u] vs [%u]%u" , |
185 | cur, opt[cur].price, opt[cur+1].price, cur+1); |
186 | if (fullUpdate) { |
187 | /* not useful to search here if next position has same (or lower) cost */ |
188 | if ( (opt[cur+1].price <= opt[cur].price) |
189 | /* in some cases, next position has same cost, but cost rises sharply after, so a small match would still be beneficial */ |
190 | && (opt[cur+MINMATCH].price < opt[cur].price + 3/*min seq price*/) ) |
191 | continue; |
192 | } else { |
193 | /* not useful to search here if next position has same (or lower) cost */ |
194 | if (opt[cur+1].price <= opt[cur].price) continue; |
195 | } |
196 | |
197 | DEBUGLOG(7, "search at rPos:%u" , cur); |
198 | if (fullUpdate) |
199 | newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, MINMATCH-1, nbSearches); |
200 | else |
201 | /* only test matches of minimum length; slightly faster, but misses a few bytes */ |
202 | newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, last_match_pos - cur, nbSearches); |
203 | if (!newMatch.len) continue; |
204 | |
205 | if ( ((size_t)newMatch.len > sufficient_len) |
206 | || (newMatch.len + cur >= LZ4_OPT_NUM) ) { |
207 | /* immediate encoding */ |
208 | best_mlen = newMatch.len; |
209 | best_off = newMatch.off; |
210 | last_match_pos = cur + 1; |
211 | goto encode; |
212 | } |
213 | |
214 | /* before match : set price with literals at beginning */ |
215 | { int const baseLitlen = opt[cur].litlen; |
216 | int litlen; |
217 | for (litlen = 1; litlen < MINMATCH; litlen++) { |
218 | int const price = opt[cur].price - LZ4HC_literalsPrice(baseLitlen) + LZ4HC_literalsPrice(baseLitlen+litlen); |
219 | int const pos = cur + litlen; |
220 | if (price < opt[pos].price) { |
221 | opt[pos].mlen = 1; /* literal */ |
222 | opt[pos].off = 0; |
223 | opt[pos].litlen = baseLitlen+litlen; |
224 | opt[pos].price = price; |
225 | DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)" , |
226 | pos, price, opt[pos].litlen); |
227 | } } } |
228 | |
229 | /* set prices using match at position = cur */ |
230 | { int const matchML = newMatch.len; |
231 | int ml = MINMATCH; |
232 | |
233 | assert(cur + newMatch.len < LZ4_OPT_NUM); |
234 | for ( ; ml <= matchML ; ml++) { |
235 | int const pos = cur + ml; |
236 | int const offset = newMatch.off; |
237 | int price; |
238 | int ll; |
239 | DEBUGLOG(7, "testing price rPos %i (last_match_pos=%i)" , |
240 | pos, last_match_pos); |
241 | if (opt[cur].mlen == 1) { |
242 | ll = opt[cur].litlen; |
243 | price = ((cur > ll) ? opt[cur - ll].price : 0) |
244 | + LZ4HC_sequencePrice(ll, ml); |
245 | } else { |
246 | ll = 0; |
247 | price = opt[cur].price + LZ4HC_sequencePrice(0, ml); |
248 | } |
249 | |
250 | if (pos > last_match_pos+TRAILING_LITERALS || price <= opt[pos].price) { |
251 | DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i)" , |
252 | pos, price, ml); |
253 | assert(pos < LZ4_OPT_NUM); |
254 | if ( (ml == matchML) /* last pos of last match */ |
255 | && (last_match_pos < pos) ) |
256 | last_match_pos = pos; |
257 | opt[pos].mlen = ml; |
258 | opt[pos].off = offset; |
259 | opt[pos].litlen = ll; |
260 | opt[pos].price = price; |
261 | } } } |
262 | /* complete following positions with literals */ |
263 | { int addLit; |
264 | for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) { |
265 | opt[last_match_pos+addLit].mlen = 1; /* literal */ |
266 | opt[last_match_pos+addLit].off = 0; |
267 | opt[last_match_pos+addLit].litlen = addLit; |
268 | opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit); |
269 | DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)" , last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit); |
270 | } } |
271 | } /* for (cur = 1; cur <= last_match_pos; cur++) */ |
272 | |
273 | best_mlen = opt[last_match_pos].mlen; |
274 | best_off = opt[last_match_pos].off; |
275 | cur = last_match_pos - best_mlen; |
276 | |
277 | encode: /* cur, last_match_pos, best_mlen, best_off must be set */ |
278 | assert(cur < LZ4_OPT_NUM); |
279 | assert(last_match_pos >= 1); /* == 1 when only one candidate */ |
280 | DEBUGLOG(6, "reverse traversal, looking for shortest path" ) |
281 | DEBUGLOG(6, "last_match_pos = %i" , last_match_pos); |
282 | { int candidate_pos = cur; |
283 | int selected_matchLength = best_mlen; |
284 | int selected_offset = best_off; |
285 | while (1) { /* from end to beginning */ |
286 | int const next_matchLength = opt[candidate_pos].mlen; /* can be 1, means literal */ |
287 | int const next_offset = opt[candidate_pos].off; |
288 | DEBUGLOG(6, "pos %i: sequence length %i" , candidate_pos, selected_matchLength); |
289 | opt[candidate_pos].mlen = selected_matchLength; |
290 | opt[candidate_pos].off = selected_offset; |
291 | selected_matchLength = next_matchLength; |
292 | selected_offset = next_offset; |
293 | if (next_matchLength > candidate_pos) break; /* last match elected, first match to encode */ |
294 | assert(next_matchLength > 0); /* can be 1, means literal */ |
295 | candidate_pos -= next_matchLength; |
296 | } } |
297 | |
298 | /* encode all recorded sequences in order */ |
299 | { int rPos = 0; /* relative position (to ip) */ |
300 | while (rPos < last_match_pos) { |
301 | int const ml = opt[rPos].mlen; |
302 | int const offset = opt[rPos].off; |
303 | if (ml == 1) { ip++; rPos++; continue; } /* literal; note: can end up with several literals, in which case, skip them */ |
304 | rPos += ml; |
305 | assert(ml >= MINMATCH); |
306 | assert((offset >= 1) && (offset <= MAX_DISTANCE)); |
307 | if ( LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ip - offset, limit, oend) ) /* updates ip, op and anchor */ |
308 | return 0; /* error */ |
309 | } } |
310 | } /* while (ip < mflimit) */ |
311 | |
312 | /* Encode Last Literals */ |
313 | { int lastRun = (int)(iend - anchor); |
314 | if ( (limit) |
315 | && (((char*)op - dst) + lastRun + 1 + ((lastRun+255-RUN_MASK)/255) > (U32)dstCapacity)) |
316 | return 0; /* Check output limit */ |
317 | if (lastRun >= (int)RUN_MASK) { |
318 | *op++=(RUN_MASK<<ML_BITS); |
319 | lastRun-=RUN_MASK; |
320 | for (; lastRun > 254 ; lastRun-=255) *op++ = 255; |
321 | *op++ = (BYTE) lastRun; |
322 | } else *op++ = (BYTE)(lastRun<<ML_BITS); |
323 | memcpy(op, anchor, iend - anchor); |
324 | op += iend-anchor; |
325 | } |
326 | |
327 | /* End */ |
328 | return (int) ((char*)op-dst); |
329 | } |
330 | |