| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 1 | package flate |
| 2 | |
| 3 | import "fmt" |
| 4 | |
| 5 | type fastEncL6 struct { |
| 6 | fastGen |
| 7 | table [tableSize]tableEntry |
| 8 | bTable [tableSize]tableEntryPrev |
| 9 | } |
| 10 | |
| 11 | func (e *fastEncL6) Encode(dst *tokens, src []byte) { |
| 12 | const ( |
| 13 | inputMargin = 12 - 1 |
| 14 | minNonLiteralBlockSize = 1 + 1 + inputMargin |
| 15 | hashShortBytes = 4 |
| 16 | ) |
| 17 | if debugDeflate && e.cur < 0 { |
| 18 | panic(fmt.Sprint("e.cur < 0: ", e.cur)) |
| 19 | } |
| 20 | |
| 21 | // Protect against e.cur wraparound. |
| 22 | for e.cur >= bufferReset { |
| 23 | if len(e.hist) == 0 { |
| 24 | for i := range e.table[:] { |
| 25 | e.table[i] = tableEntry{} |
| 26 | } |
| 27 | for i := range e.bTable[:] { |
| 28 | e.bTable[i] = tableEntryPrev{} |
| 29 | } |
| 30 | e.cur = maxMatchOffset |
| 31 | break |
| 32 | } |
| 33 | // Shift down everything in the table that isn't already too far away. |
| 34 | minOff := e.cur + int32(len(e.hist)) - maxMatchOffset |
| 35 | for i := range e.table[:] { |
| 36 | v := e.table[i].offset |
| 37 | if v <= minOff { |
| 38 | v = 0 |
| 39 | } else { |
| 40 | v = v - e.cur + maxMatchOffset |
| 41 | } |
| 42 | e.table[i].offset = v |
| 43 | } |
| 44 | for i := range e.bTable[:] { |
| 45 | v := e.bTable[i] |
| 46 | if v.Cur.offset <= minOff { |
| 47 | v.Cur.offset = 0 |
| 48 | v.Prev.offset = 0 |
| 49 | } else { |
| 50 | v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset |
| 51 | if v.Prev.offset <= minOff { |
| 52 | v.Prev.offset = 0 |
| 53 | } else { |
| 54 | v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset |
| 55 | } |
| 56 | } |
| 57 | e.bTable[i] = v |
| 58 | } |
| 59 | e.cur = maxMatchOffset |
| 60 | } |
| 61 | |
| 62 | s := e.addBlock(src) |
| 63 | |
| 64 | // This check isn't in the Snappy implementation, but there, the caller |
| 65 | // instead of the callee handles this case. |
| 66 | if len(src) < minNonLiteralBlockSize { |
| 67 | // We do not fill the token table. |
| 68 | // This will be picked up by caller. |
| 69 | dst.n = uint16(len(src)) |
| 70 | return |
| 71 | } |
| 72 | |
| 73 | // Override src |
| 74 | src = e.hist |
| 75 | nextEmit := s |
| 76 | |
| 77 | // sLimit is when to stop looking for offset/length copies. The inputMargin |
| 78 | // lets us use a fast path for emitLiteral in the main loop, while we are |
| 79 | // looking for copies. |
| 80 | sLimit := int32(len(src) - inputMargin) |
| 81 | |
| 82 | // nextEmit is where in src the next emitLiteral should start from. |
| 83 | cv := load6432(src, s) |
| 84 | // Repeat MUST be > 1 and within range |
| 85 | repeat := int32(1) |
| 86 | for { |
| 87 | const skipLog = 7 |
| 88 | const doEvery = 1 |
| 89 | |
| 90 | nextS := s |
| 91 | var l int32 |
| 92 | var t int32 |
| 93 | for { |
| 94 | nextHashS := hashLen(cv, tableBits, hashShortBytes) |
| 95 | nextHashL := hash7(cv, tableBits) |
| 96 | s = nextS |
| 97 | nextS = s + doEvery + (s-nextEmit)>>skipLog |
| 98 | if nextS > sLimit { |
| 99 | goto emitRemainder |
| 100 | } |
| 101 | // Fetch a short+long candidate |
| 102 | sCandidate := e.table[nextHashS] |
| 103 | lCandidate := e.bTable[nextHashL] |
| 104 | next := load6432(src, nextS) |
| 105 | entry := tableEntry{offset: s + e.cur} |
| 106 | e.table[nextHashS] = entry |
| 107 | eLong := &e.bTable[nextHashL] |
| 108 | eLong.Cur, eLong.Prev = entry, eLong.Cur |
| 109 | |
| 110 | // Calculate hashes of 'next' |
| 111 | nextHashS = hashLen(next, tableBits, hashShortBytes) |
| 112 | nextHashL = hash7(next, tableBits) |
| 113 | |
| 114 | t = lCandidate.Cur.offset - e.cur |
| 115 | if s-t < maxMatchOffset { |
| 116 | if uint32(cv) == load3232(src, t) { |
| 117 | // Long candidate matches at least 4 bytes. |
| 118 | |
| 119 | // Store the next match |
| 120 | e.table[nextHashS] = tableEntry{offset: nextS + e.cur} |
| 121 | eLong := &e.bTable[nextHashL] |
| 122 | eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur |
| 123 | |
| 124 | // Check the previous long candidate as well. |
| 125 | t2 := lCandidate.Prev.offset - e.cur |
| 126 | if s-t2 < maxMatchOffset && uint32(cv) == load3232(src, t2) { |
| 127 | l = e.matchlen(int(s+4), int(t+4), src) + 4 |
| 128 | ml1 := e.matchlen(int(s+4), int(t2+4), src) + 4 |
| 129 | if ml1 > l { |
| 130 | t = t2 |
| 131 | l = ml1 |
| 132 | break |
| 133 | } |
| 134 | } |
| 135 | break |
| 136 | } |
| 137 | // Current value did not match, but check if previous long value does. |
| 138 | t = lCandidate.Prev.offset - e.cur |
| 139 | if s-t < maxMatchOffset && uint32(cv) == load3232(src, t) { |
| 140 | // Store the next match |
| 141 | e.table[nextHashS] = tableEntry{offset: nextS + e.cur} |
| 142 | eLong := &e.bTable[nextHashL] |
| 143 | eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur |
| 144 | break |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | t = sCandidate.offset - e.cur |
| 149 | if s-t < maxMatchOffset && uint32(cv) == load3232(src, t) { |
| 150 | // Found a 4 match... |
| 151 | l = e.matchlen(int(s+4), int(t+4), src) + 4 |
| 152 | |
| 153 | // Look up next long candidate (at nextS) |
| 154 | lCandidate = e.bTable[nextHashL] |
| 155 | |
| 156 | // Store the next match |
| 157 | e.table[nextHashS] = tableEntry{offset: nextS + e.cur} |
| 158 | eLong := &e.bTable[nextHashL] |
| 159 | eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur |
| 160 | |
| 161 | // Check repeat at s + repOff |
| 162 | const repOff = 1 |
| 163 | t2 := s - repeat + repOff |
| 164 | if load3232(src, t2) == uint32(cv>>(8*repOff)) { |
| 165 | ml := e.matchlen(int(s+4+repOff), int(t2+4), src) + 4 |
| 166 | if ml > l { |
| 167 | t = t2 |
| 168 | l = ml |
| 169 | s += repOff |
| 170 | // Not worth checking more. |
| 171 | break |
| 172 | } |
| 173 | } |
| 174 | |
| 175 | // If the next long is a candidate, use that... |
| 176 | t2 = lCandidate.Cur.offset - e.cur |
| 177 | if nextS-t2 < maxMatchOffset { |
| 178 | if load3232(src, t2) == uint32(next) { |
| 179 | ml := e.matchlen(int(nextS+4), int(t2+4), src) + 4 |
| 180 | if ml > l { |
| 181 | t = t2 |
| 182 | s = nextS |
| 183 | l = ml |
| 184 | // This is ok, but check previous as well. |
| 185 | } |
| 186 | } |
| 187 | // If the previous long is a candidate, use that... |
| 188 | t2 = lCandidate.Prev.offset - e.cur |
| 189 | if nextS-t2 < maxMatchOffset && load3232(src, t2) == uint32(next) { |
| 190 | ml := e.matchlen(int(nextS+4), int(t2+4), src) + 4 |
| 191 | if ml > l { |
| 192 | t = t2 |
| 193 | s = nextS |
| 194 | l = ml |
| 195 | break |
| 196 | } |
| 197 | } |
| 198 | } |
| 199 | break |
| 200 | } |
| 201 | cv = next |
| 202 | } |
| 203 | |
| 204 | // A 4-byte match has been found. We'll later see if more than 4 bytes |
| 205 | // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit |
| 206 | // them as literal bytes. |
| 207 | |
| 208 | // Extend the 4-byte match as long as possible. |
| 209 | if l == 0 { |
| 210 | l = e.matchlenLong(int(s+4), int(t+4), src) + 4 |
| 211 | } else if l == maxMatchLength { |
| 212 | l += e.matchlenLong(int(s+l), int(t+l), src) |
| 213 | } |
| 214 | |
| 215 | // Try to locate a better match by checking the end-of-match... |
| 216 | if sAt := s + l; sAt < sLimit { |
| 217 | // Allow some bytes at the beginning to mismatch. |
| 218 | // Sweet spot is 2/3 bytes depending on input. |
| 219 | // 3 is only a little better when it is but sometimes a lot worse. |
| 220 | // The skipped bytes are tested in Extend backwards, |
| 221 | // and still picked up as part of the match if they do. |
| 222 | const skipBeginning = 2 |
| 223 | eLong := &e.bTable[hash7(load6432(src, sAt), tableBits)] |
| 224 | // Test current |
| 225 | t2 := eLong.Cur.offset - e.cur - l + skipBeginning |
| 226 | s2 := s + skipBeginning |
| 227 | off := s2 - t2 |
| 228 | if off < maxMatchOffset { |
| 229 | if off > 0 && t2 >= 0 { |
| 230 | if l2 := e.matchlenLong(int(s2), int(t2), src); l2 > l { |
| 231 | t = t2 |
| 232 | l = l2 |
| 233 | s = s2 |
| 234 | } |
| 235 | } |
| 236 | // Test next: |
| 237 | t2 = eLong.Prev.offset - e.cur - l + skipBeginning |
| 238 | off := s2 - t2 |
| 239 | if off > 0 && off < maxMatchOffset && t2 >= 0 { |
| 240 | if l2 := e.matchlenLong(int(s2), int(t2), src); l2 > l { |
| 241 | t = t2 |
| 242 | l = l2 |
| 243 | s = s2 |
| 244 | } |
| 245 | } |
| 246 | } |
| 247 | } |
| 248 | |
| 249 | // Extend backwards |
| 250 | for t > 0 && s > nextEmit && src[t-1] == src[s-1] { |
| 251 | s-- |
| 252 | t-- |
| 253 | l++ |
| 254 | } |
| 255 | if nextEmit < s { |
| 256 | if false { |
| 257 | emitLiteral(dst, src[nextEmit:s]) |
| 258 | } else { |
| 259 | for _, v := range src[nextEmit:s] { |
| 260 | dst.tokens[dst.n] = token(v) |
| 261 | dst.litHist[v]++ |
| 262 | dst.n++ |
| 263 | } |
| 264 | } |
| 265 | } |
| 266 | if false { |
| 267 | if t >= s { |
| 268 | panic(fmt.Sprintln("s-t", s, t)) |
| 269 | } |
| 270 | if (s - t) > maxMatchOffset { |
| 271 | panic(fmt.Sprintln("mmo", s-t)) |
| 272 | } |
| 273 | if l < baseMatchLength { |
| 274 | panic("bml") |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) |
| 279 | repeat = s - t |
| 280 | s += l |
| 281 | nextEmit = s |
| 282 | if nextS >= s { |
| 283 | s = nextS + 1 |
| 284 | } |
| 285 | |
| 286 | if s >= sLimit { |
| 287 | // Index after match end. |
| 288 | for i := nextS + 1; i < int32(len(src))-8; i += 2 { |
| 289 | cv := load6432(src, i) |
| 290 | e.table[hashLen(cv, tableBits, hashShortBytes)] = tableEntry{offset: i + e.cur} |
| 291 | eLong := &e.bTable[hash7(cv, tableBits)] |
| 292 | eLong.Cur, eLong.Prev = tableEntry{offset: i + e.cur}, eLong.Cur |
| 293 | } |
| 294 | goto emitRemainder |
| 295 | } |
| 296 | |
| 297 | // Store every long hash in-between and every second short. |
| 298 | if true { |
| 299 | for i := nextS + 1; i < s-1; i += 2 { |
| 300 | cv := load6432(src, i) |
| 301 | t := tableEntry{offset: i + e.cur} |
| 302 | t2 := tableEntry{offset: t.offset + 1} |
| 303 | eLong := &e.bTable[hash7(cv, tableBits)] |
| 304 | eLong2 := &e.bTable[hash7(cv>>8, tableBits)] |
| 305 | e.table[hashLen(cv, tableBits, hashShortBytes)] = t |
| 306 | eLong.Cur, eLong.Prev = t, eLong.Cur |
| 307 | eLong2.Cur, eLong2.Prev = t2, eLong2.Cur |
| 308 | } |
| 309 | } |
| 310 | |
| 311 | // We could immediately start working at s now, but to improve |
| 312 | // compression we first update the hash table at s-1 and at s. |
| 313 | cv = load6432(src, s) |
| 314 | } |
| 315 | |
| 316 | emitRemainder: |
| 317 | if int(nextEmit) < len(src) { |
| 318 | // If nothing was added, don't encode literals. |
| 319 | if dst.n == 0 { |
| 320 | return |
| 321 | } |
| 322 | |
| 323 | emitLiteral(dst, src[nextEmit:]) |
| 324 | } |
| 325 | } |