kesavand | c71914f | 2022-03-25 11:19:03 +0530 | [diff] [blame] | 1 | // Copyright 2019+ Klaus Post. All rights reserved. |
| 2 | // License information can be found in the LICENSE file. |
| 3 | // Based on work by Yann Collet, released under BSD License. |
| 4 | |
| 5 | package zstd |
| 6 | |
| 7 | import ( |
| 8 | "fmt" |
| 9 | ) |
| 10 | |
| 11 | const ( |
| 12 | tableBits = 15 // Bits used in the table |
| 13 | tableSize = 1 << tableBits // Size of the table |
| 14 | tableShardCnt = 1 << (tableBits - dictShardBits) // Number of shards in the table |
| 15 | tableShardSize = tableSize / tableShardCnt // Size of an individual shard |
| 16 | tableFastHashLen = 6 |
| 17 | tableMask = tableSize - 1 // Mask for table indices. Redundant, but can eliminate bounds checks. |
| 18 | maxMatchLength = 131074 |
| 19 | ) |
| 20 | |
| 21 | type tableEntry struct { |
| 22 | val uint32 |
| 23 | offset int32 |
| 24 | } |
| 25 | |
| 26 | type fastEncoder struct { |
| 27 | fastBase |
| 28 | table [tableSize]tableEntry |
| 29 | } |
| 30 | |
| 31 | type fastEncoderDict struct { |
| 32 | fastEncoder |
| 33 | dictTable []tableEntry |
| 34 | tableShardDirty [tableShardCnt]bool |
| 35 | allDirty bool |
| 36 | } |
| 37 | |
| 38 | // Encode mimmics functionality in zstd_fast.c |
| 39 | func (e *fastEncoder) Encode(blk *blockEnc, src []byte) { |
| 40 | const ( |
| 41 | inputMargin = 8 |
| 42 | minNonLiteralBlockSize = 1 + 1 + inputMargin |
| 43 | ) |
| 44 | |
| 45 | // Protect against e.cur wraparound. |
| 46 | for e.cur >= bufferReset { |
| 47 | if len(e.hist) == 0 { |
| 48 | for i := range e.table[:] { |
| 49 | e.table[i] = tableEntry{} |
| 50 | } |
| 51 | e.cur = e.maxMatchOff |
| 52 | break |
| 53 | } |
| 54 | // Shift down everything in the table that isn't already too far away. |
| 55 | minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff |
| 56 | for i := range e.table[:] { |
| 57 | v := e.table[i].offset |
| 58 | if v < minOff { |
| 59 | v = 0 |
| 60 | } else { |
| 61 | v = v - e.cur + e.maxMatchOff |
| 62 | } |
| 63 | e.table[i].offset = v |
| 64 | } |
| 65 | e.cur = e.maxMatchOff |
| 66 | break |
| 67 | } |
| 68 | |
| 69 | s := e.addBlock(src) |
| 70 | blk.size = len(src) |
| 71 | if len(src) < minNonLiteralBlockSize { |
| 72 | blk.extraLits = len(src) |
| 73 | blk.literals = blk.literals[:len(src)] |
| 74 | copy(blk.literals, src) |
| 75 | return |
| 76 | } |
| 77 | |
| 78 | // Override src |
| 79 | src = e.hist |
| 80 | sLimit := int32(len(src)) - inputMargin |
| 81 | // stepSize is the number of bytes to skip on every main loop iteration. |
| 82 | // It should be >= 2. |
| 83 | const stepSize = 2 |
| 84 | |
| 85 | // TEMPLATE |
| 86 | const hashLog = tableBits |
| 87 | // seems global, but would be nice to tweak. |
| 88 | const kSearchStrength = 6 |
| 89 | |
| 90 | // nextEmit is where in src the next emitLiteral should start from. |
| 91 | nextEmit := s |
| 92 | cv := load6432(src, s) |
| 93 | |
| 94 | // Relative offsets |
| 95 | offset1 := int32(blk.recentOffsets[0]) |
| 96 | offset2 := int32(blk.recentOffsets[1]) |
| 97 | |
| 98 | addLiterals := func(s *seq, until int32) { |
| 99 | if until == nextEmit { |
| 100 | return |
| 101 | } |
| 102 | blk.literals = append(blk.literals, src[nextEmit:until]...) |
| 103 | s.litLen = uint32(until - nextEmit) |
| 104 | } |
| 105 | if debugEncoder { |
| 106 | println("recent offsets:", blk.recentOffsets) |
| 107 | } |
| 108 | |
| 109 | encodeLoop: |
| 110 | for { |
| 111 | // t will contain the match offset when we find one. |
| 112 | // When existing the search loop, we have already checked 4 bytes. |
| 113 | var t int32 |
| 114 | |
| 115 | // We will not use repeat offsets across blocks. |
| 116 | // By not using them for the first 3 matches |
| 117 | canRepeat := len(blk.sequences) > 2 |
| 118 | |
| 119 | for { |
| 120 | if debugAsserts && canRepeat && offset1 == 0 { |
| 121 | panic("offset0 was 0") |
| 122 | } |
| 123 | |
| 124 | nextHash := hashLen(cv, hashLog, tableFastHashLen) |
| 125 | nextHash2 := hashLen(cv>>8, hashLog, tableFastHashLen) |
| 126 | candidate := e.table[nextHash] |
| 127 | candidate2 := e.table[nextHash2] |
| 128 | repIndex := s - offset1 + 2 |
| 129 | |
| 130 | e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} |
| 131 | e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)} |
| 132 | |
| 133 | if canRepeat && repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>16) { |
| 134 | // Consider history as well. |
| 135 | var seq seq |
| 136 | var length int32 |
| 137 | length = 4 + e.matchlen(s+6, repIndex+4, src) |
| 138 | seq.matchLen = uint32(length - zstdMinMatch) |
| 139 | |
| 140 | // We might be able to match backwards. |
| 141 | // Extend as long as we can. |
| 142 | start := s + 2 |
| 143 | // We end the search early, so we don't risk 0 literals |
| 144 | // and have to do special offset treatment. |
| 145 | startLimit := nextEmit + 1 |
| 146 | |
| 147 | sMin := s - e.maxMatchOff |
| 148 | if sMin < 0 { |
| 149 | sMin = 0 |
| 150 | } |
| 151 | for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch { |
| 152 | repIndex-- |
| 153 | start-- |
| 154 | seq.matchLen++ |
| 155 | } |
| 156 | addLiterals(&seq, start) |
| 157 | |
| 158 | // rep 0 |
| 159 | seq.offset = 1 |
| 160 | if debugSequences { |
| 161 | println("repeat sequence", seq, "next s:", s) |
| 162 | } |
| 163 | blk.sequences = append(blk.sequences, seq) |
| 164 | s += length + 2 |
| 165 | nextEmit = s |
| 166 | if s >= sLimit { |
| 167 | if debugEncoder { |
| 168 | println("repeat ended", s, length) |
| 169 | |
| 170 | } |
| 171 | break encodeLoop |
| 172 | } |
| 173 | cv = load6432(src, s) |
| 174 | continue |
| 175 | } |
| 176 | coffset0 := s - (candidate.offset - e.cur) |
| 177 | coffset1 := s - (candidate2.offset - e.cur) + 1 |
| 178 | if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val { |
| 179 | // found a regular match |
| 180 | t = candidate.offset - e.cur |
| 181 | if debugAsserts && s <= t { |
| 182 | panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) |
| 183 | } |
| 184 | if debugAsserts && s-t > e.maxMatchOff { |
| 185 | panic("s - t >e.maxMatchOff") |
| 186 | } |
| 187 | break |
| 188 | } |
| 189 | |
| 190 | if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val { |
| 191 | // found a regular match |
| 192 | t = candidate2.offset - e.cur |
| 193 | s++ |
| 194 | if debugAsserts && s <= t { |
| 195 | panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) |
| 196 | } |
| 197 | if debugAsserts && s-t > e.maxMatchOff { |
| 198 | panic("s - t >e.maxMatchOff") |
| 199 | } |
| 200 | if debugAsserts && t < 0 { |
| 201 | panic("t<0") |
| 202 | } |
| 203 | break |
| 204 | } |
| 205 | s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) |
| 206 | if s >= sLimit { |
| 207 | break encodeLoop |
| 208 | } |
| 209 | cv = load6432(src, s) |
| 210 | } |
| 211 | // A 4-byte match has been found. We'll later see if more than 4 bytes. |
| 212 | offset2 = offset1 |
| 213 | offset1 = s - t |
| 214 | |
| 215 | if debugAsserts && s <= t { |
| 216 | panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) |
| 217 | } |
| 218 | |
| 219 | if debugAsserts && canRepeat && int(offset1) > len(src) { |
| 220 | panic("invalid offset") |
| 221 | } |
| 222 | |
| 223 | // Extend the 4-byte match as long as possible. |
| 224 | l := e.matchlen(s+4, t+4, src) + 4 |
| 225 | |
| 226 | // Extend backwards |
| 227 | tMin := s - e.maxMatchOff |
| 228 | if tMin < 0 { |
| 229 | tMin = 0 |
| 230 | } |
| 231 | for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { |
| 232 | s-- |
| 233 | t-- |
| 234 | l++ |
| 235 | } |
| 236 | |
| 237 | // Write our sequence. |
| 238 | var seq seq |
| 239 | seq.litLen = uint32(s - nextEmit) |
| 240 | seq.matchLen = uint32(l - zstdMinMatch) |
| 241 | if seq.litLen > 0 { |
| 242 | blk.literals = append(blk.literals, src[nextEmit:s]...) |
| 243 | } |
| 244 | // Don't use repeat offsets |
| 245 | seq.offset = uint32(s-t) + 3 |
| 246 | s += l |
| 247 | if debugSequences { |
| 248 | println("sequence", seq, "next s:", s) |
| 249 | } |
| 250 | blk.sequences = append(blk.sequences, seq) |
| 251 | nextEmit = s |
| 252 | if s >= sLimit { |
| 253 | break encodeLoop |
| 254 | } |
| 255 | cv = load6432(src, s) |
| 256 | |
| 257 | // Check offset 2 |
| 258 | if o2 := s - offset2; canRepeat && load3232(src, o2) == uint32(cv) { |
| 259 | // We have at least 4 byte match. |
| 260 | // No need to check backwards. We come straight from a match |
| 261 | l := 4 + e.matchlen(s+4, o2+4, src) |
| 262 | |
| 263 | // Store this, since we have it. |
| 264 | nextHash := hashLen(cv, hashLog, tableFastHashLen) |
| 265 | e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} |
| 266 | seq.matchLen = uint32(l) - zstdMinMatch |
| 267 | seq.litLen = 0 |
| 268 | // Since litlen is always 0, this is offset 1. |
| 269 | seq.offset = 1 |
| 270 | s += l |
| 271 | nextEmit = s |
| 272 | if debugSequences { |
| 273 | println("sequence", seq, "next s:", s) |
| 274 | } |
| 275 | blk.sequences = append(blk.sequences, seq) |
| 276 | |
| 277 | // Swap offset 1 and 2. |
| 278 | offset1, offset2 = offset2, offset1 |
| 279 | if s >= sLimit { |
| 280 | break encodeLoop |
| 281 | } |
| 282 | // Prepare next loop. |
| 283 | cv = load6432(src, s) |
| 284 | } |
| 285 | } |
| 286 | |
| 287 | if int(nextEmit) < len(src) { |
| 288 | blk.literals = append(blk.literals, src[nextEmit:]...) |
| 289 | blk.extraLits = len(src) - int(nextEmit) |
| 290 | } |
| 291 | blk.recentOffsets[0] = uint32(offset1) |
| 292 | blk.recentOffsets[1] = uint32(offset2) |
| 293 | if debugEncoder { |
| 294 | println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) |
| 295 | } |
| 296 | } |
| 297 | |
| 298 | // EncodeNoHist will encode a block with no history and no following blocks. |
| 299 | // Most notable difference is that src will not be copied for history and |
| 300 | // we do not need to check for max match length. |
| 301 | func (e *fastEncoder) EncodeNoHist(blk *blockEnc, src []byte) { |
| 302 | const ( |
| 303 | inputMargin = 8 |
| 304 | minNonLiteralBlockSize = 1 + 1 + inputMargin |
| 305 | ) |
| 306 | if debugEncoder { |
| 307 | if len(src) > maxBlockSize { |
| 308 | panic("src too big") |
| 309 | } |
| 310 | } |
| 311 | |
| 312 | // Protect against e.cur wraparound. |
| 313 | if e.cur >= bufferReset { |
| 314 | for i := range e.table[:] { |
| 315 | e.table[i] = tableEntry{} |
| 316 | } |
| 317 | e.cur = e.maxMatchOff |
| 318 | } |
| 319 | |
| 320 | s := int32(0) |
| 321 | blk.size = len(src) |
| 322 | if len(src) < minNonLiteralBlockSize { |
| 323 | blk.extraLits = len(src) |
| 324 | blk.literals = blk.literals[:len(src)] |
| 325 | copy(blk.literals, src) |
| 326 | return |
| 327 | } |
| 328 | |
| 329 | sLimit := int32(len(src)) - inputMargin |
| 330 | // stepSize is the number of bytes to skip on every main loop iteration. |
| 331 | // It should be >= 2. |
| 332 | const stepSize = 2 |
| 333 | |
| 334 | // TEMPLATE |
| 335 | const hashLog = tableBits |
| 336 | // seems global, but would be nice to tweak. |
| 337 | const kSearchStrength = 6 |
| 338 | |
| 339 | // nextEmit is where in src the next emitLiteral should start from. |
| 340 | nextEmit := s |
| 341 | cv := load6432(src, s) |
| 342 | |
| 343 | // Relative offsets |
| 344 | offset1 := int32(blk.recentOffsets[0]) |
| 345 | offset2 := int32(blk.recentOffsets[1]) |
| 346 | |
| 347 | addLiterals := func(s *seq, until int32) { |
| 348 | if until == nextEmit { |
| 349 | return |
| 350 | } |
| 351 | blk.literals = append(blk.literals, src[nextEmit:until]...) |
| 352 | s.litLen = uint32(until - nextEmit) |
| 353 | } |
| 354 | if debugEncoder { |
| 355 | println("recent offsets:", blk.recentOffsets) |
| 356 | } |
| 357 | |
| 358 | encodeLoop: |
| 359 | for { |
| 360 | // t will contain the match offset when we find one. |
| 361 | // When existing the search loop, we have already checked 4 bytes. |
| 362 | var t int32 |
| 363 | |
| 364 | // We will not use repeat offsets across blocks. |
| 365 | // By not using them for the first 3 matches |
| 366 | |
| 367 | for { |
| 368 | nextHash := hashLen(cv, hashLog, tableFastHashLen) |
| 369 | nextHash2 := hashLen(cv>>8, hashLog, tableFastHashLen) |
| 370 | candidate := e.table[nextHash] |
| 371 | candidate2 := e.table[nextHash2] |
| 372 | repIndex := s - offset1 + 2 |
| 373 | |
| 374 | e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} |
| 375 | e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)} |
| 376 | |
| 377 | if len(blk.sequences) > 2 && load3232(src, repIndex) == uint32(cv>>16) { |
| 378 | // Consider history as well. |
| 379 | var seq seq |
| 380 | length := 4 + e.matchlen(s+6, repIndex+4, src) |
| 381 | |
| 382 | seq.matchLen = uint32(length - zstdMinMatch) |
| 383 | |
| 384 | // We might be able to match backwards. |
| 385 | // Extend as long as we can. |
| 386 | start := s + 2 |
| 387 | // We end the search early, so we don't risk 0 literals |
| 388 | // and have to do special offset treatment. |
| 389 | startLimit := nextEmit + 1 |
| 390 | |
| 391 | sMin := s - e.maxMatchOff |
| 392 | if sMin < 0 { |
| 393 | sMin = 0 |
| 394 | } |
| 395 | for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] { |
| 396 | repIndex-- |
| 397 | start-- |
| 398 | seq.matchLen++ |
| 399 | } |
| 400 | addLiterals(&seq, start) |
| 401 | |
| 402 | // rep 0 |
| 403 | seq.offset = 1 |
| 404 | if debugSequences { |
| 405 | println("repeat sequence", seq, "next s:", s) |
| 406 | } |
| 407 | blk.sequences = append(blk.sequences, seq) |
| 408 | s += length + 2 |
| 409 | nextEmit = s |
| 410 | if s >= sLimit { |
| 411 | if debugEncoder { |
| 412 | println("repeat ended", s, length) |
| 413 | |
| 414 | } |
| 415 | break encodeLoop |
| 416 | } |
| 417 | cv = load6432(src, s) |
| 418 | continue |
| 419 | } |
| 420 | coffset0 := s - (candidate.offset - e.cur) |
| 421 | coffset1 := s - (candidate2.offset - e.cur) + 1 |
| 422 | if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val { |
| 423 | // found a regular match |
| 424 | t = candidate.offset - e.cur |
| 425 | if debugAsserts && s <= t { |
| 426 | panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) |
| 427 | } |
| 428 | if debugAsserts && s-t > e.maxMatchOff { |
| 429 | panic("s - t >e.maxMatchOff") |
| 430 | } |
| 431 | if debugAsserts && t < 0 { |
| 432 | panic(fmt.Sprintf("t (%d) < 0, candidate.offset: %d, e.cur: %d, coffset0: %d, e.maxMatchOff: %d", t, candidate.offset, e.cur, coffset0, e.maxMatchOff)) |
| 433 | } |
| 434 | break |
| 435 | } |
| 436 | |
| 437 | if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val { |
| 438 | // found a regular match |
| 439 | t = candidate2.offset - e.cur |
| 440 | s++ |
| 441 | if debugAsserts && s <= t { |
| 442 | panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) |
| 443 | } |
| 444 | if debugAsserts && s-t > e.maxMatchOff { |
| 445 | panic("s - t >e.maxMatchOff") |
| 446 | } |
| 447 | if debugAsserts && t < 0 { |
| 448 | panic("t<0") |
| 449 | } |
| 450 | break |
| 451 | } |
| 452 | s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) |
| 453 | if s >= sLimit { |
| 454 | break encodeLoop |
| 455 | } |
| 456 | cv = load6432(src, s) |
| 457 | } |
| 458 | // A 4-byte match has been found. We'll later see if more than 4 bytes. |
| 459 | offset2 = offset1 |
| 460 | offset1 = s - t |
| 461 | |
| 462 | if debugAsserts && s <= t { |
| 463 | panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) |
| 464 | } |
| 465 | |
| 466 | if debugAsserts && t < 0 { |
| 467 | panic(fmt.Sprintf("t (%d) < 0 ", t)) |
| 468 | } |
| 469 | // Extend the 4-byte match as long as possible. |
| 470 | l := e.matchlen(s+4, t+4, src) + 4 |
| 471 | |
| 472 | // Extend backwards |
| 473 | tMin := s - e.maxMatchOff |
| 474 | if tMin < 0 { |
| 475 | tMin = 0 |
| 476 | } |
| 477 | for t > tMin && s > nextEmit && src[t-1] == src[s-1] { |
| 478 | s-- |
| 479 | t-- |
| 480 | l++ |
| 481 | } |
| 482 | |
| 483 | // Write our sequence. |
| 484 | var seq seq |
| 485 | seq.litLen = uint32(s - nextEmit) |
| 486 | seq.matchLen = uint32(l - zstdMinMatch) |
| 487 | if seq.litLen > 0 { |
| 488 | blk.literals = append(blk.literals, src[nextEmit:s]...) |
| 489 | } |
| 490 | // Don't use repeat offsets |
| 491 | seq.offset = uint32(s-t) + 3 |
| 492 | s += l |
| 493 | if debugSequences { |
| 494 | println("sequence", seq, "next s:", s) |
| 495 | } |
| 496 | blk.sequences = append(blk.sequences, seq) |
| 497 | nextEmit = s |
| 498 | if s >= sLimit { |
| 499 | break encodeLoop |
| 500 | } |
| 501 | cv = load6432(src, s) |
| 502 | |
| 503 | // Check offset 2 |
| 504 | if o2 := s - offset2; len(blk.sequences) > 2 && load3232(src, o2) == uint32(cv) { |
| 505 | // We have at least 4 byte match. |
| 506 | // No need to check backwards. We come straight from a match |
| 507 | l := 4 + e.matchlen(s+4, o2+4, src) |
| 508 | |
| 509 | // Store this, since we have it. |
| 510 | nextHash := hashLen(cv, hashLog, tableFastHashLen) |
| 511 | e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} |
| 512 | seq.matchLen = uint32(l) - zstdMinMatch |
| 513 | seq.litLen = 0 |
| 514 | // Since litlen is always 0, this is offset 1. |
| 515 | seq.offset = 1 |
| 516 | s += l |
| 517 | nextEmit = s |
| 518 | if debugSequences { |
| 519 | println("sequence", seq, "next s:", s) |
| 520 | } |
| 521 | blk.sequences = append(blk.sequences, seq) |
| 522 | |
| 523 | // Swap offset 1 and 2. |
| 524 | offset1, offset2 = offset2, offset1 |
| 525 | if s >= sLimit { |
| 526 | break encodeLoop |
| 527 | } |
| 528 | // Prepare next loop. |
| 529 | cv = load6432(src, s) |
| 530 | } |
| 531 | } |
| 532 | |
| 533 | if int(nextEmit) < len(src) { |
| 534 | blk.literals = append(blk.literals, src[nextEmit:]...) |
| 535 | blk.extraLits = len(src) - int(nextEmit) |
| 536 | } |
| 537 | if debugEncoder { |
| 538 | println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) |
| 539 | } |
| 540 | // We do not store history, so we must offset e.cur to avoid false matches for next user. |
| 541 | if e.cur < bufferReset { |
| 542 | e.cur += int32(len(src)) |
| 543 | } |
| 544 | } |
| 545 | |
| 546 | // Encode will encode the content, with a dictionary if initialized for it. |
| 547 | func (e *fastEncoderDict) Encode(blk *blockEnc, src []byte) { |
| 548 | const ( |
| 549 | inputMargin = 8 |
| 550 | minNonLiteralBlockSize = 1 + 1 + inputMargin |
| 551 | ) |
| 552 | if e.allDirty || len(src) > 32<<10 { |
| 553 | e.fastEncoder.Encode(blk, src) |
| 554 | e.allDirty = true |
| 555 | return |
| 556 | } |
| 557 | // Protect against e.cur wraparound. |
| 558 | for e.cur >= bufferReset { |
| 559 | if len(e.hist) == 0 { |
| 560 | for i := range e.table[:] { |
| 561 | e.table[i] = tableEntry{} |
| 562 | } |
| 563 | e.cur = e.maxMatchOff |
| 564 | break |
| 565 | } |
| 566 | // Shift down everything in the table that isn't already too far away. |
| 567 | minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff |
| 568 | for i := range e.table[:] { |
| 569 | v := e.table[i].offset |
| 570 | if v < minOff { |
| 571 | v = 0 |
| 572 | } else { |
| 573 | v = v - e.cur + e.maxMatchOff |
| 574 | } |
| 575 | e.table[i].offset = v |
| 576 | } |
| 577 | e.cur = e.maxMatchOff |
| 578 | break |
| 579 | } |
| 580 | |
| 581 | s := e.addBlock(src) |
| 582 | blk.size = len(src) |
| 583 | if len(src) < minNonLiteralBlockSize { |
| 584 | blk.extraLits = len(src) |
| 585 | blk.literals = blk.literals[:len(src)] |
| 586 | copy(blk.literals, src) |
| 587 | return |
| 588 | } |
| 589 | |
| 590 | // Override src |
| 591 | src = e.hist |
| 592 | sLimit := int32(len(src)) - inputMargin |
| 593 | // stepSize is the number of bytes to skip on every main loop iteration. |
| 594 | // It should be >= 2. |
| 595 | const stepSize = 2 |
| 596 | |
| 597 | // TEMPLATE |
| 598 | const hashLog = tableBits |
| 599 | // seems global, but would be nice to tweak. |
| 600 | const kSearchStrength = 7 |
| 601 | |
| 602 | // nextEmit is where in src the next emitLiteral should start from. |
| 603 | nextEmit := s |
| 604 | cv := load6432(src, s) |
| 605 | |
| 606 | // Relative offsets |
| 607 | offset1 := int32(blk.recentOffsets[0]) |
| 608 | offset2 := int32(blk.recentOffsets[1]) |
| 609 | |
| 610 | addLiterals := func(s *seq, until int32) { |
| 611 | if until == nextEmit { |
| 612 | return |
| 613 | } |
| 614 | blk.literals = append(blk.literals, src[nextEmit:until]...) |
| 615 | s.litLen = uint32(until - nextEmit) |
| 616 | } |
| 617 | if debugEncoder { |
| 618 | println("recent offsets:", blk.recentOffsets) |
| 619 | } |
| 620 | |
| 621 | encodeLoop: |
| 622 | for { |
| 623 | // t will contain the match offset when we find one. |
| 624 | // When existing the search loop, we have already checked 4 bytes. |
| 625 | var t int32 |
| 626 | |
| 627 | // We will not use repeat offsets across blocks. |
| 628 | // By not using them for the first 3 matches |
| 629 | canRepeat := len(blk.sequences) > 2 |
| 630 | |
| 631 | for { |
| 632 | if debugAsserts && canRepeat && offset1 == 0 { |
| 633 | panic("offset0 was 0") |
| 634 | } |
| 635 | |
| 636 | nextHash := hashLen(cv, hashLog, tableFastHashLen) |
| 637 | nextHash2 := hashLen(cv>>8, hashLog, tableFastHashLen) |
| 638 | candidate := e.table[nextHash] |
| 639 | candidate2 := e.table[nextHash2] |
| 640 | repIndex := s - offset1 + 2 |
| 641 | |
| 642 | e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} |
| 643 | e.markShardDirty(nextHash) |
| 644 | e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)} |
| 645 | e.markShardDirty(nextHash2) |
| 646 | |
| 647 | if canRepeat && repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>16) { |
| 648 | // Consider history as well. |
| 649 | var seq seq |
| 650 | var length int32 |
| 651 | length = 4 + e.matchlen(s+6, repIndex+4, src) |
| 652 | |
| 653 | seq.matchLen = uint32(length - zstdMinMatch) |
| 654 | |
| 655 | // We might be able to match backwards. |
| 656 | // Extend as long as we can. |
| 657 | start := s + 2 |
| 658 | // We end the search early, so we don't risk 0 literals |
| 659 | // and have to do special offset treatment. |
| 660 | startLimit := nextEmit + 1 |
| 661 | |
| 662 | sMin := s - e.maxMatchOff |
| 663 | if sMin < 0 { |
| 664 | sMin = 0 |
| 665 | } |
| 666 | for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch { |
| 667 | repIndex-- |
| 668 | start-- |
| 669 | seq.matchLen++ |
| 670 | } |
| 671 | addLiterals(&seq, start) |
| 672 | |
| 673 | // rep 0 |
| 674 | seq.offset = 1 |
| 675 | if debugSequences { |
| 676 | println("repeat sequence", seq, "next s:", s) |
| 677 | } |
| 678 | blk.sequences = append(blk.sequences, seq) |
| 679 | s += length + 2 |
| 680 | nextEmit = s |
| 681 | if s >= sLimit { |
| 682 | if debugEncoder { |
| 683 | println("repeat ended", s, length) |
| 684 | |
| 685 | } |
| 686 | break encodeLoop |
| 687 | } |
| 688 | cv = load6432(src, s) |
| 689 | continue |
| 690 | } |
| 691 | coffset0 := s - (candidate.offset - e.cur) |
| 692 | coffset1 := s - (candidate2.offset - e.cur) + 1 |
| 693 | if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val { |
| 694 | // found a regular match |
| 695 | t = candidate.offset - e.cur |
| 696 | if debugAsserts && s <= t { |
| 697 | panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) |
| 698 | } |
| 699 | if debugAsserts && s-t > e.maxMatchOff { |
| 700 | panic("s - t >e.maxMatchOff") |
| 701 | } |
| 702 | break |
| 703 | } |
| 704 | |
| 705 | if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val { |
| 706 | // found a regular match |
| 707 | t = candidate2.offset - e.cur |
| 708 | s++ |
| 709 | if debugAsserts && s <= t { |
| 710 | panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) |
| 711 | } |
| 712 | if debugAsserts && s-t > e.maxMatchOff { |
| 713 | panic("s - t >e.maxMatchOff") |
| 714 | } |
| 715 | if debugAsserts && t < 0 { |
| 716 | panic("t<0") |
| 717 | } |
| 718 | break |
| 719 | } |
| 720 | s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) |
| 721 | if s >= sLimit { |
| 722 | break encodeLoop |
| 723 | } |
| 724 | cv = load6432(src, s) |
| 725 | } |
| 726 | // A 4-byte match has been found. We'll later see if more than 4 bytes. |
| 727 | offset2 = offset1 |
| 728 | offset1 = s - t |
| 729 | |
| 730 | if debugAsserts && s <= t { |
| 731 | panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) |
| 732 | } |
| 733 | |
| 734 | if debugAsserts && canRepeat && int(offset1) > len(src) { |
| 735 | panic("invalid offset") |
| 736 | } |
| 737 | |
| 738 | // Extend the 4-byte match as long as possible. |
| 739 | l := e.matchlen(s+4, t+4, src) + 4 |
| 740 | |
| 741 | // Extend backwards |
| 742 | tMin := s - e.maxMatchOff |
| 743 | if tMin < 0 { |
| 744 | tMin = 0 |
| 745 | } |
| 746 | for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { |
| 747 | s-- |
| 748 | t-- |
| 749 | l++ |
| 750 | } |
| 751 | |
| 752 | // Write our sequence. |
| 753 | var seq seq |
| 754 | seq.litLen = uint32(s - nextEmit) |
| 755 | seq.matchLen = uint32(l - zstdMinMatch) |
| 756 | if seq.litLen > 0 { |
| 757 | blk.literals = append(blk.literals, src[nextEmit:s]...) |
| 758 | } |
| 759 | // Don't use repeat offsets |
| 760 | seq.offset = uint32(s-t) + 3 |
| 761 | s += l |
| 762 | if debugSequences { |
| 763 | println("sequence", seq, "next s:", s) |
| 764 | } |
| 765 | blk.sequences = append(blk.sequences, seq) |
| 766 | nextEmit = s |
| 767 | if s >= sLimit { |
| 768 | break encodeLoop |
| 769 | } |
| 770 | cv = load6432(src, s) |
| 771 | |
| 772 | // Check offset 2 |
| 773 | if o2 := s - offset2; canRepeat && load3232(src, o2) == uint32(cv) { |
| 774 | // We have at least 4 byte match. |
| 775 | // No need to check backwards. We come straight from a match |
| 776 | l := 4 + e.matchlen(s+4, o2+4, src) |
| 777 | |
| 778 | // Store this, since we have it. |
| 779 | nextHash := hashLen(cv, hashLog, tableFastHashLen) |
| 780 | e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} |
| 781 | e.markShardDirty(nextHash) |
| 782 | seq.matchLen = uint32(l) - zstdMinMatch |
| 783 | seq.litLen = 0 |
| 784 | // Since litlen is always 0, this is offset 1. |
| 785 | seq.offset = 1 |
| 786 | s += l |
| 787 | nextEmit = s |
| 788 | if debugSequences { |
| 789 | println("sequence", seq, "next s:", s) |
| 790 | } |
| 791 | blk.sequences = append(blk.sequences, seq) |
| 792 | |
| 793 | // Swap offset 1 and 2. |
| 794 | offset1, offset2 = offset2, offset1 |
| 795 | if s >= sLimit { |
| 796 | break encodeLoop |
| 797 | } |
| 798 | // Prepare next loop. |
| 799 | cv = load6432(src, s) |
| 800 | } |
| 801 | } |
| 802 | |
| 803 | if int(nextEmit) < len(src) { |
| 804 | blk.literals = append(blk.literals, src[nextEmit:]...) |
| 805 | blk.extraLits = len(src) - int(nextEmit) |
| 806 | } |
| 807 | blk.recentOffsets[0] = uint32(offset1) |
| 808 | blk.recentOffsets[1] = uint32(offset2) |
| 809 | if debugEncoder { |
| 810 | println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) |
| 811 | } |
| 812 | } |
| 813 | |
| 814 | // ResetDict will reset and set a dictionary if not nil |
| 815 | func (e *fastEncoder) Reset(d *dict, singleBlock bool) { |
| 816 | e.resetBase(d, singleBlock) |
| 817 | if d != nil { |
| 818 | panic("fastEncoder: Reset with dict") |
| 819 | } |
| 820 | } |
| 821 | |
| 822 | // ResetDict will reset and set a dictionary if not nil |
| 823 | func (e *fastEncoderDict) Reset(d *dict, singleBlock bool) { |
| 824 | e.resetBase(d, singleBlock) |
| 825 | if d == nil { |
| 826 | return |
| 827 | } |
| 828 | |
| 829 | // Init or copy dict table |
| 830 | if len(e.dictTable) != len(e.table) || d.id != e.lastDictID { |
| 831 | if len(e.dictTable) != len(e.table) { |
| 832 | e.dictTable = make([]tableEntry, len(e.table)) |
| 833 | } |
| 834 | if true { |
| 835 | end := e.maxMatchOff + int32(len(d.content)) - 8 |
| 836 | for i := e.maxMatchOff; i < end; i += 3 { |
| 837 | const hashLog = tableBits |
| 838 | |
| 839 | cv := load6432(d.content, i-e.maxMatchOff) |
| 840 | nextHash := hashLen(cv, hashLog, tableFastHashLen) // 0 -> 5 |
| 841 | nextHash1 := hashLen(cv>>8, hashLog, tableFastHashLen) // 1 -> 6 |
| 842 | nextHash2 := hashLen(cv>>16, hashLog, tableFastHashLen) // 2 -> 7 |
| 843 | e.dictTable[nextHash] = tableEntry{ |
| 844 | val: uint32(cv), |
| 845 | offset: i, |
| 846 | } |
| 847 | e.dictTable[nextHash1] = tableEntry{ |
| 848 | val: uint32(cv >> 8), |
| 849 | offset: i + 1, |
| 850 | } |
| 851 | e.dictTable[nextHash2] = tableEntry{ |
| 852 | val: uint32(cv >> 16), |
| 853 | offset: i + 2, |
| 854 | } |
| 855 | } |
| 856 | } |
| 857 | e.lastDictID = d.id |
| 858 | e.allDirty = true |
| 859 | } |
| 860 | |
| 861 | e.cur = e.maxMatchOff |
| 862 | dirtyShardCnt := 0 |
| 863 | if !e.allDirty { |
| 864 | for i := range e.tableShardDirty { |
| 865 | if e.tableShardDirty[i] { |
| 866 | dirtyShardCnt++ |
| 867 | } |
| 868 | } |
| 869 | } |
| 870 | |
| 871 | const shardCnt = tableShardCnt |
| 872 | const shardSize = tableShardSize |
| 873 | if e.allDirty || dirtyShardCnt > shardCnt*4/6 { |
| 874 | copy(e.table[:], e.dictTable) |
| 875 | for i := range e.tableShardDirty { |
| 876 | e.tableShardDirty[i] = false |
| 877 | } |
| 878 | e.allDirty = false |
| 879 | return |
| 880 | } |
| 881 | for i := range e.tableShardDirty { |
| 882 | if !e.tableShardDirty[i] { |
| 883 | continue |
| 884 | } |
| 885 | |
| 886 | copy(e.table[i*shardSize:(i+1)*shardSize], e.dictTable[i*shardSize:(i+1)*shardSize]) |
| 887 | e.tableShardDirty[i] = false |
| 888 | } |
| 889 | e.allDirty = false |
| 890 | } |
| 891 | |
| 892 | func (e *fastEncoderDict) markAllShardsDirty() { |
| 893 | e.allDirty = true |
| 894 | } |
| 895 | |
| 896 | func (e *fastEncoderDict) markShardDirty(entryNum uint32) { |
| 897 | e.tableShardDirty[entryNum/tableShardSize] = true |
| 898 | } |