Matteo Scandolo | 9a2772a | 2018-11-19 14:56:26 -0800 | [diff] [blame^] | 1 | package lz4 |
| 2 | |
| 3 | import ( |
| 4 | "encoding/binary" |
| 5 | "errors" |
| 6 | ) |
| 7 | |
| 8 | var ( |
| 9 | // ErrInvalidSourceShortBuffer is returned by UncompressBlock or CompressBLock when a compressed |
| 10 | // block is corrupted or the destination buffer is not large enough for the uncompressed data. |
| 11 | ErrInvalidSourceShortBuffer = errors.New("lz4: invalid source or destination buffer too short") |
| 12 | // ErrInvalid is returned when reading an invalid LZ4 archive. |
| 13 | ErrInvalid = errors.New("lz4: bad magic number") |
| 14 | ) |
| 15 | |
| 16 | // blockHash hashes 4 bytes into a value < winSize. |
| 17 | func blockHash(x uint32) uint32 { |
| 18 | const hasher uint32 = 2654435761 // Knuth multiplicative hash. |
| 19 | return x * hasher >> hashShift |
| 20 | } |
| 21 | |
| 22 | // CompressBlockBound returns the maximum size of a given buffer of size n, when not compressible. |
| 23 | func CompressBlockBound(n int) int { |
| 24 | return n + n/255 + 16 |
| 25 | } |
| 26 | |
| 27 | // UncompressBlock uncompresses the source buffer into the destination one, |
| 28 | // and returns the uncompressed size. |
| 29 | // |
| 30 | // The destination buffer must be sized appropriately. |
| 31 | // |
| 32 | // An error is returned if the source data is invalid or the destination buffer is too small. |
| 33 | func UncompressBlock(src, dst []byte) (si int, err error) { |
| 34 | defer func() { |
| 35 | // It is now faster to let the runtime panic and recover on out of bound slice access |
| 36 | // than checking indices as we go along. |
| 37 | if recover() != nil { |
| 38 | err = ErrInvalidSourceShortBuffer |
| 39 | } |
| 40 | }() |
| 41 | sn := len(src) |
| 42 | if sn == 0 { |
| 43 | return 0, nil |
| 44 | } |
| 45 | var di int |
| 46 | |
| 47 | for { |
| 48 | // Literals and match lengths (token). |
| 49 | b := int(src[si]) |
| 50 | si++ |
| 51 | |
| 52 | // Literals. |
| 53 | if lLen := b >> 4; lLen > 0 { |
| 54 | if lLen == 0xF { |
| 55 | for src[si] == 0xFF { |
| 56 | lLen += 0xFF |
| 57 | si++ |
| 58 | } |
| 59 | lLen += int(src[si]) |
| 60 | si++ |
| 61 | } |
| 62 | i := si |
| 63 | si += lLen |
| 64 | di += copy(dst[di:], src[i:si]) |
| 65 | |
| 66 | if si >= sn { |
| 67 | return di, nil |
| 68 | } |
| 69 | } |
| 70 | |
| 71 | si++ |
| 72 | _ = src[si] // Bound check elimination. |
| 73 | offset := int(src[si-1]) | int(src[si])<<8 |
| 74 | si++ |
| 75 | |
| 76 | // Match. |
| 77 | mLen := b & 0xF |
| 78 | if mLen == 0xF { |
| 79 | for src[si] == 0xFF { |
| 80 | mLen += 0xFF |
| 81 | si++ |
| 82 | } |
| 83 | mLen += int(src[si]) |
| 84 | si++ |
| 85 | } |
| 86 | mLen += minMatch |
| 87 | |
| 88 | // Copy the match. |
| 89 | i := di - offset |
| 90 | if offset > 0 && mLen >= offset { |
| 91 | // Efficiently copy the match dst[di-offset:di] into the dst slice. |
| 92 | bytesToCopy := offset * (mLen / offset) |
| 93 | expanded := dst[i:] |
| 94 | for n := offset; n <= bytesToCopy+offset; n *= 2 { |
| 95 | copy(expanded[n:], expanded[:n]) |
| 96 | } |
| 97 | di += bytesToCopy |
| 98 | mLen -= bytesToCopy |
| 99 | } |
| 100 | di += copy(dst[di:], dst[i:i+mLen]) |
| 101 | } |
| 102 | } |
| 103 | |
| 104 | // CompressBlock compresses the source buffer into the destination one. |
| 105 | // This is the fast version of LZ4 compression and also the default one. |
| 106 | // The size of hashTable must be at least 64Kb. |
| 107 | // |
| 108 | // The size of the compressed data is returned. If it is 0 and no error, then the data is incompressible. |
| 109 | // |
| 110 | // An error is returned if the destination buffer is too small. |
| 111 | func CompressBlock(src, dst []byte, hashTable []int) (di int, err error) { |
| 112 | defer func() { |
| 113 | if recover() != nil { |
| 114 | err = ErrInvalidSourceShortBuffer |
| 115 | } |
| 116 | }() |
| 117 | |
| 118 | sn, dn := len(src)-mfLimit, len(dst) |
| 119 | if sn <= 0 || dn == 0 { |
| 120 | return 0, nil |
| 121 | } |
| 122 | var si int |
| 123 | |
| 124 | // Fast scan strategy: the hash table only stores the last 4 bytes sequences. |
| 125 | // const accInit = 1 << skipStrength |
| 126 | |
| 127 | anchor := si // Position of the current literals. |
| 128 | // acc := accInit // Variable step: improves performance on non-compressible data. |
| 129 | |
| 130 | for si < sn { |
| 131 | // Hash the next 4 bytes (sequence)... |
| 132 | match := binary.LittleEndian.Uint32(src[si:]) |
| 133 | h := blockHash(match) |
| 134 | |
| 135 | ref := hashTable[h] |
| 136 | hashTable[h] = si |
| 137 | if ref >= sn { // Invalid reference (dirty hashtable). |
| 138 | si++ |
| 139 | continue |
| 140 | } |
| 141 | offset := si - ref |
| 142 | if offset <= 0 || offset >= winSize || // Out of window. |
| 143 | match != binary.LittleEndian.Uint32(src[ref:]) { // Hash collision on different matches. |
| 144 | // si += acc >> skipStrength |
| 145 | // acc++ |
| 146 | si++ |
| 147 | continue |
| 148 | } |
| 149 | |
| 150 | // Match found. |
| 151 | // acc = accInit |
| 152 | lLen := si - anchor // Literal length. |
| 153 | |
| 154 | // Encode match length part 1. |
| 155 | si += minMatch |
| 156 | mLen := si // Match length has minMatch already. |
| 157 | // Find the longest match, first looking by batches of 8 bytes. |
| 158 | for si < sn && binary.LittleEndian.Uint64(src[si:]) == binary.LittleEndian.Uint64(src[si-offset:]) { |
| 159 | si += 8 |
| 160 | } |
| 161 | // Then byte by byte. |
| 162 | for si < sn && src[si] == src[si-offset] { |
| 163 | si++ |
| 164 | } |
| 165 | |
| 166 | mLen = si - mLen |
| 167 | if mLen < 0xF { |
| 168 | dst[di] = byte(mLen) |
| 169 | } else { |
| 170 | dst[di] = 0xF |
| 171 | } |
| 172 | |
| 173 | // Encode literals length. |
| 174 | if lLen < 0xF { |
| 175 | dst[di] |= byte(lLen << 4) |
| 176 | } else { |
| 177 | dst[di] |= 0xF0 |
| 178 | di++ |
| 179 | l := lLen - 0xF |
| 180 | for ; l >= 0xFF; l -= 0xFF { |
| 181 | dst[di] = 0xFF |
| 182 | di++ |
| 183 | } |
| 184 | dst[di] = byte(l) |
| 185 | } |
| 186 | di++ |
| 187 | |
| 188 | // Literals. |
| 189 | copy(dst[di:], src[anchor:anchor+lLen]) |
| 190 | di += lLen + 2 |
| 191 | anchor = si |
| 192 | |
| 193 | // Encode offset. |
| 194 | _ = dst[di] // Bound check elimination. |
| 195 | dst[di-2], dst[di-1] = byte(offset), byte(offset>>8) |
| 196 | |
| 197 | // Encode match length part 2. |
| 198 | if mLen >= 0xF { |
| 199 | for mLen -= 0xF; mLen >= 0xFF; mLen -= 0xFF { |
| 200 | dst[di] = 0xFF |
| 201 | di++ |
| 202 | } |
| 203 | dst[di] = byte(mLen) |
| 204 | di++ |
| 205 | } |
| 206 | } |
| 207 | |
| 208 | if anchor == 0 { |
| 209 | // Incompressible. |
| 210 | return 0, nil |
| 211 | } |
| 212 | |
| 213 | // Last literals. |
| 214 | lLen := len(src) - anchor |
| 215 | if lLen < 0xF { |
| 216 | dst[di] = byte(lLen << 4) |
| 217 | } else { |
| 218 | dst[di] = 0xF0 |
| 219 | di++ |
| 220 | for lLen -= 0xF; lLen >= 0xFF; lLen -= 0xFF { |
| 221 | dst[di] = 0xFF |
| 222 | di++ |
| 223 | } |
| 224 | dst[di] = byte(lLen) |
| 225 | } |
| 226 | di++ |
| 227 | |
| 228 | // Write the last literals. |
| 229 | if di >= anchor { |
| 230 | // Incompressible. |
| 231 | return 0, nil |
| 232 | } |
| 233 | di += copy(dst[di:], src[anchor:]) |
| 234 | return di, nil |
| 235 | } |
| 236 | |
| 237 | // CompressBlockHC compresses the source buffer src into the destination dst |
| 238 | // with max search depth (use 0 or negative value for no max). |
| 239 | // |
| 240 | // CompressBlockHC compression ratio is better than CompressBlock but it is also slower. |
| 241 | // |
| 242 | // The size of the compressed data is returned. If it is 0 and no error, then the data is not compressible. |
| 243 | // |
| 244 | // An error is returned if the destination buffer is too small. |
| 245 | func CompressBlockHC(src, dst []byte, depth int) (di int, err error) { |
| 246 | defer func() { |
| 247 | if recover() != nil { |
| 248 | err = ErrInvalidSourceShortBuffer |
| 249 | } |
| 250 | }() |
| 251 | |
| 252 | sn, dn := len(src)-mfLimit, len(dst) |
| 253 | if sn <= 0 || dn == 0 { |
| 254 | return 0, nil |
| 255 | } |
| 256 | var si int |
| 257 | |
| 258 | // hashTable: stores the last position found for a given hash |
| 259 | // chaingTable: stores previous positions for a given hash |
| 260 | var hashTable, chainTable [winSize]int |
| 261 | |
| 262 | if depth <= 0 { |
| 263 | depth = winSize |
| 264 | } |
| 265 | |
| 266 | anchor := si |
| 267 | for si < sn { |
| 268 | // Hash the next 4 bytes (sequence). |
| 269 | match := binary.LittleEndian.Uint32(src[si:]) |
| 270 | h := blockHash(match) |
| 271 | |
| 272 | // Follow the chain until out of window and give the longest match. |
| 273 | mLen := 0 |
| 274 | offset := 0 |
| 275 | for next, try := hashTable[h], depth; try > 0 && next > 0 && si-next < winSize; next = chainTable[next&winMask] { |
| 276 | // The first (mLen==0) or next byte (mLen>=minMatch) at current match length |
| 277 | // must match to improve on the match length. |
| 278 | if src[next+mLen] != src[si+mLen] { |
| 279 | continue |
| 280 | } |
| 281 | ml := 0 |
| 282 | // Compare the current position with a previous with the same hash. |
| 283 | for ml < sn-si && binary.LittleEndian.Uint64(src[next+ml:]) == binary.LittleEndian.Uint64(src[si+ml:]) { |
| 284 | ml += 8 |
| 285 | } |
| 286 | for ml < sn-si && src[next+ml] == src[si+ml] { |
| 287 | ml++ |
| 288 | } |
| 289 | if ml < minMatch || ml <= mLen { |
| 290 | // Match too small (<minMath) or smaller than the current match. |
| 291 | continue |
| 292 | } |
| 293 | // Found a longer match, keep its position and length. |
| 294 | mLen = ml |
| 295 | offset = si - next |
| 296 | // Try another previous position with the same hash. |
| 297 | try-- |
| 298 | } |
| 299 | chainTable[si&winMask] = hashTable[h] |
| 300 | hashTable[h] = si |
| 301 | |
| 302 | // No match found. |
| 303 | if mLen == 0 { |
| 304 | si++ |
| 305 | continue |
| 306 | } |
| 307 | |
| 308 | // Match found. |
| 309 | // Update hash/chain tables with overlapping bytes: |
| 310 | // si already hashed, add everything from si+1 up to the match length. |
| 311 | winStart := si + 1 |
| 312 | if ws := si + mLen - winSize; ws > winStart { |
| 313 | winStart = ws |
| 314 | } |
| 315 | for si, ml := winStart, si+mLen; si < ml; { |
| 316 | match >>= 8 |
| 317 | match |= uint32(src[si+3]) << 24 |
| 318 | h := blockHash(match) |
| 319 | chainTable[si&winMask] = hashTable[h] |
| 320 | hashTable[h] = si |
| 321 | si++ |
| 322 | } |
| 323 | |
| 324 | lLen := si - anchor |
| 325 | si += mLen |
| 326 | mLen -= minMatch // Match length does not include minMatch. |
| 327 | |
| 328 | if mLen < 0xF { |
| 329 | dst[di] = byte(mLen) |
| 330 | } else { |
| 331 | dst[di] = 0xF |
| 332 | } |
| 333 | |
| 334 | // Encode literals length. |
| 335 | if lLen < 0xF { |
| 336 | dst[di] |= byte(lLen << 4) |
| 337 | } else { |
| 338 | dst[di] |= 0xF0 |
| 339 | di++ |
| 340 | l := lLen - 0xF |
| 341 | for ; l >= 0xFF; l -= 0xFF { |
| 342 | dst[di] = 0xFF |
| 343 | di++ |
| 344 | } |
| 345 | dst[di] = byte(l) |
| 346 | } |
| 347 | di++ |
| 348 | |
| 349 | // Literals. |
| 350 | copy(dst[di:], src[anchor:anchor+lLen]) |
| 351 | di += lLen |
| 352 | anchor = si |
| 353 | |
| 354 | // Encode offset. |
| 355 | di += 2 |
| 356 | dst[di-2], dst[di-1] = byte(offset), byte(offset>>8) |
| 357 | |
| 358 | // Encode match length part 2. |
| 359 | if mLen >= 0xF { |
| 360 | for mLen -= 0xF; mLen >= 0xFF; mLen -= 0xFF { |
| 361 | dst[di] = 0xFF |
| 362 | di++ |
| 363 | } |
| 364 | dst[di] = byte(mLen) |
| 365 | di++ |
| 366 | } |
| 367 | } |
| 368 | |
| 369 | if anchor == 0 { |
| 370 | // Incompressible. |
| 371 | return 0, nil |
| 372 | } |
| 373 | |
| 374 | // Last literals. |
| 375 | lLen := len(src) - anchor |
| 376 | if lLen < 0xF { |
| 377 | dst[di] = byte(lLen << 4) |
| 378 | } else { |
| 379 | dst[di] = 0xF0 |
| 380 | di++ |
| 381 | lLen -= 0xF |
| 382 | for ; lLen >= 0xFF; lLen -= 0xFF { |
| 383 | dst[di] = 0xFF |
| 384 | di++ |
| 385 | } |
| 386 | dst[di] = byte(lLen) |
| 387 | } |
| 388 | di++ |
| 389 | |
| 390 | // Write the last literals. |
| 391 | if di >= anchor { |
| 392 | // Incompressible. |
| 393 | return 0, nil |
| 394 | } |
| 395 | di += copy(dst[di:], src[anchor:]) |
| 396 | return di, nil |
| 397 | } |