| 1 | /* Copyright 2015 Google Inc. All Rights Reserved. |
| 2 | |
| 3 | Distributed under MIT license. |
| 4 | See file LICENSE for detail or copy at https://opensource.org/licenses/MIT |
| 5 | */ |
| 6 | |
| 7 | /* Function for fast encoding of an input fragment, independently from the input |
| 8 | history. This function uses two-pass processing: in the first pass we save |
| 9 | the found backward matches and literal bytes into a buffer, and in the |
| 10 | second pass we emit them into the bit stream using prefix codes built based |
| 11 | on the actual command and literal byte histograms. */ |
| 12 | |
| 13 | #include "./compress_fragment_two_pass.h" |
| 14 | |
| 15 | #include <string.h> /* memcmp, memcpy, memset */ |
| 16 | |
| 17 | #include "../common/constants.h" |
| 18 | #include "../common/platform.h" |
| 19 | #include <brotli/types.h> |
| 20 | #include "./bit_cost.h" |
| 21 | #include "./brotli_bit_stream.h" |
| 22 | #include "./entropy_encode.h" |
| 23 | #include "./fast_log.h" |
| 24 | #include "./find_match_length.h" |
| 25 | #include "./memory.h" |
| 26 | #include "./write_bits.h" |
| 27 | |
| 28 | #if defined(__cplusplus) || defined(c_plusplus) |
| 29 | extern "C" { |
| 30 | #endif |
| 31 | |
| 32 | #define MAX_DISTANCE (long)BROTLI_MAX_BACKWARD_LIMIT(18) |
| 33 | |
| 34 | /* kHashMul32 multiplier has these properties: |
| 35 | * The multiplier must be odd. Otherwise we may lose the highest bit. |
| 36 | * No long streaks of ones or zeros. |
| 37 | * There is no effort to ensure that it is a prime, the oddity is enough |
| 38 | for this use. |
| 39 | * The number has been tuned heuristically against compression benchmarks. */ |
| 40 | static const uint32_t kHashMul32 = 0x1E35A7BD; |
| 41 | |
| 42 | static BROTLI_INLINE uint32_t Hash(const uint8_t* p, |
| 43 | size_t shift, size_t length) { |
| 44 | const uint64_t h = |
| 45 | (BROTLI_UNALIGNED_LOAD64LE(p) << ((8 - length) * 8)) * kHashMul32; |
| 46 | return (uint32_t)(h >> shift); |
| 47 | } |
| 48 | |
| 49 | static BROTLI_INLINE uint32_t HashBytesAtOffset(uint64_t v, size_t offset, |
| 50 | size_t shift, size_t length) { |
| 51 | BROTLI_DCHECK(offset <= 8 - length); |
| 52 | { |
| 53 | const uint64_t h = ((v >> (8 * offset)) << ((8 - length) * 8)) * kHashMul32; |
| 54 | return (uint32_t)(h >> shift); |
| 55 | } |
| 56 | } |
| 57 | |
| 58 | static BROTLI_INLINE BROTLI_BOOL IsMatch(const uint8_t* p1, const uint8_t* p2, |
| 59 | size_t length) { |
| 60 | if (BrotliUnalignedRead32(p1) == BrotliUnalignedRead32(p2)) { |
| 61 | if (length == 4) return BROTLI_TRUE; |
| 62 | return TO_BROTLI_BOOL(p1[4] == p2[4] && p1[5] == p2[5]); |
| 63 | } |
| 64 | return BROTLI_FALSE; |
| 65 | } |
| 66 | |
| 67 | /* Builds a command and distance prefix code (each 64 symbols) into "depth" and |
| 68 | "bits" based on "histogram" and stores it into the bit stream. */ |
| 69 | static void BuildAndStoreCommandPrefixCode( |
| 70 | const uint32_t histogram[128], |
| 71 | uint8_t depth[128], uint16_t bits[128], |
| 72 | size_t* storage_ix, uint8_t* storage) { |
| 73 | /* Tree size for building a tree over 64 symbols is 2 * 64 + 1. */ |
| 74 | HuffmanTree tree[129]; |
| 75 | uint8_t cmd_depth[BROTLI_NUM_COMMAND_SYMBOLS] = { 0 }; |
| 76 | uint16_t cmd_bits[64]; |
| 77 | BrotliCreateHuffmanTree(histogram, 64, 15, tree, depth); |
| 78 | BrotliCreateHuffmanTree(&histogram[64], 64, 14, tree, &depth[64]); |
| 79 | /* We have to jump through a few hoops here in order to compute |
| 80 | the command bits because the symbols are in a different order than in |
| 81 | the full alphabet. This looks complicated, but having the symbols |
| 82 | in this order in the command bits saves a few branches in the Emit* |
| 83 | functions. */ |
| 84 | memcpy(cmd_depth, depth + 24, 24); |
| 85 | memcpy(cmd_depth + 24, depth, 8); |
| 86 | memcpy(cmd_depth + 32, depth + 48, 8); |
| 87 | memcpy(cmd_depth + 40, depth + 8, 8); |
| 88 | memcpy(cmd_depth + 48, depth + 56, 8); |
| 89 | memcpy(cmd_depth + 56, depth + 16, 8); |
| 90 | BrotliConvertBitDepthsToSymbols(cmd_depth, 64, cmd_bits); |
| 91 | memcpy(bits, cmd_bits + 24, 16); |
| 92 | memcpy(bits + 8, cmd_bits + 40, 16); |
| 93 | memcpy(bits + 16, cmd_bits + 56, 16); |
| 94 | memcpy(bits + 24, cmd_bits, 48); |
| 95 | memcpy(bits + 48, cmd_bits + 32, 16); |
| 96 | memcpy(bits + 56, cmd_bits + 48, 16); |
| 97 | BrotliConvertBitDepthsToSymbols(&depth[64], 64, &bits[64]); |
| 98 | { |
| 99 | /* Create the bit length array for the full command alphabet. */ |
| 100 | size_t i; |
| 101 | memset(cmd_depth, 0, 64); /* only 64 first values were used */ |
| 102 | memcpy(cmd_depth, depth + 24, 8); |
| 103 | memcpy(cmd_depth + 64, depth + 32, 8); |
| 104 | memcpy(cmd_depth + 128, depth + 40, 8); |
| 105 | memcpy(cmd_depth + 192, depth + 48, 8); |
| 106 | memcpy(cmd_depth + 384, depth + 56, 8); |
| 107 | for (i = 0; i < 8; ++i) { |
| 108 | cmd_depth[128 + 8 * i] = depth[i]; |
| 109 | cmd_depth[256 + 8 * i] = depth[8 + i]; |
| 110 | cmd_depth[448 + 8 * i] = depth[16 + i]; |
| 111 | } |
| 112 | BrotliStoreHuffmanTree( |
| 113 | cmd_depth, BROTLI_NUM_COMMAND_SYMBOLS, tree, storage_ix, storage); |
| 114 | } |
| 115 | BrotliStoreHuffmanTree(&depth[64], 64, tree, storage_ix, storage); |
| 116 | } |
| 117 | |
| 118 | static BROTLI_INLINE void EmitInsertLen( |
| 119 | uint32_t insertlen, uint32_t** commands) { |
| 120 | if (insertlen < 6) { |
| 121 | **commands = insertlen; |
| 122 | } else if (insertlen < 130) { |
| 123 | const uint32_t tail = insertlen - 2; |
| 124 | const uint32_t nbits = Log2FloorNonZero(tail) - 1u; |
| 125 | const uint32_t prefix = tail >> nbits; |
| 126 | const uint32_t inscode = (nbits << 1) + prefix + 2; |
| 127 | const uint32_t = tail - (prefix << nbits); |
| 128 | **commands = inscode | (extra << 8); |
| 129 | } else if (insertlen < 2114) { |
| 130 | const uint32_t tail = insertlen - 66; |
| 131 | const uint32_t nbits = Log2FloorNonZero(tail); |
| 132 | const uint32_t code = nbits + 10; |
| 133 | const uint32_t = tail - (1u << nbits); |
| 134 | **commands = code | (extra << 8); |
| 135 | } else if (insertlen < 6210) { |
| 136 | const uint32_t = insertlen - 2114; |
| 137 | **commands = 21 | (extra << 8); |
| 138 | } else if (insertlen < 22594) { |
| 139 | const uint32_t = insertlen - 6210; |
| 140 | **commands = 22 | (extra << 8); |
| 141 | } else { |
| 142 | const uint32_t = insertlen - 22594; |
| 143 | **commands = 23 | (extra << 8); |
| 144 | } |
| 145 | ++(*commands); |
| 146 | } |
| 147 | |
| 148 | static BROTLI_INLINE void EmitCopyLen(size_t copylen, uint32_t** commands) { |
| 149 | if (copylen < 10) { |
| 150 | **commands = (uint32_t)(copylen + 38); |
| 151 | } else if (copylen < 134) { |
| 152 | const size_t tail = copylen - 6; |
| 153 | const size_t nbits = Log2FloorNonZero(tail) - 1; |
| 154 | const size_t prefix = tail >> nbits; |
| 155 | const size_t code = (nbits << 1) + prefix + 44; |
| 156 | const size_t = tail - (prefix << nbits); |
| 157 | **commands = (uint32_t)(code | (extra << 8)); |
| 158 | } else if (copylen < 2118) { |
| 159 | const size_t tail = copylen - 70; |
| 160 | const size_t nbits = Log2FloorNonZero(tail); |
| 161 | const size_t code = nbits + 52; |
| 162 | const size_t = tail - ((size_t)1 << nbits); |
| 163 | **commands = (uint32_t)(code | (extra << 8)); |
| 164 | } else { |
| 165 | const size_t = copylen - 2118; |
| 166 | **commands = (uint32_t)(63 | (extra << 8)); |
| 167 | } |
| 168 | ++(*commands); |
| 169 | } |
| 170 | |
| 171 | static BROTLI_INLINE void EmitCopyLenLastDistance( |
| 172 | size_t copylen, uint32_t** commands) { |
| 173 | if (copylen < 12) { |
| 174 | **commands = (uint32_t)(copylen + 20); |
| 175 | ++(*commands); |
| 176 | } else if (copylen < 72) { |
| 177 | const size_t tail = copylen - 8; |
| 178 | const size_t nbits = Log2FloorNonZero(tail) - 1; |
| 179 | const size_t prefix = tail >> nbits; |
| 180 | const size_t code = (nbits << 1) + prefix + 28; |
| 181 | const size_t = tail - (prefix << nbits); |
| 182 | **commands = (uint32_t)(code | (extra << 8)); |
| 183 | ++(*commands); |
| 184 | } else if (copylen < 136) { |
| 185 | const size_t tail = copylen - 8; |
| 186 | const size_t code = (tail >> 5) + 54; |
| 187 | const size_t = tail & 31; |
| 188 | **commands = (uint32_t)(code | (extra << 8)); |
| 189 | ++(*commands); |
| 190 | **commands = 64; |
| 191 | ++(*commands); |
| 192 | } else if (copylen < 2120) { |
| 193 | const size_t tail = copylen - 72; |
| 194 | const size_t nbits = Log2FloorNonZero(tail); |
| 195 | const size_t code = nbits + 52; |
| 196 | const size_t = tail - ((size_t)1 << nbits); |
| 197 | **commands = (uint32_t)(code | (extra << 8)); |
| 198 | ++(*commands); |
| 199 | **commands = 64; |
| 200 | ++(*commands); |
| 201 | } else { |
| 202 | const size_t = copylen - 2120; |
| 203 | **commands = (uint32_t)(63 | (extra << 8)); |
| 204 | ++(*commands); |
| 205 | **commands = 64; |
| 206 | ++(*commands); |
| 207 | } |
| 208 | } |
| 209 | |
| 210 | static BROTLI_INLINE void EmitDistance(uint32_t distance, uint32_t** commands) { |
| 211 | uint32_t d = distance + 3; |
| 212 | uint32_t nbits = Log2FloorNonZero(d) - 1; |
| 213 | const uint32_t prefix = (d >> nbits) & 1; |
| 214 | const uint32_t offset = (2 + prefix) << nbits; |
| 215 | const uint32_t distcode = 2 * (nbits - 1) + prefix + 80; |
| 216 | uint32_t = d - offset; |
| 217 | **commands = distcode | (extra << 8); |
| 218 | ++(*commands); |
| 219 | } |
| 220 | |
| 221 | /* REQUIRES: len <= 1 << 24. */ |
| 222 | static void ( |
| 223 | size_t len, BROTLI_BOOL is_uncompressed, size_t* storage_ix, |
| 224 | uint8_t* storage) { |
| 225 | size_t nibbles = 6; |
| 226 | /* ISLAST */ |
| 227 | BrotliWriteBits(1, 0, storage_ix, storage); |
| 228 | if (len <= (1U << 16)) { |
| 229 | nibbles = 4; |
| 230 | } else if (len <= (1U << 20)) { |
| 231 | nibbles = 5; |
| 232 | } |
| 233 | BrotliWriteBits(2, nibbles - 4, storage_ix, storage); |
| 234 | BrotliWriteBits(nibbles * 4, len - 1, storage_ix, storage); |
| 235 | /* ISUNCOMPRESSED */ |
| 236 | BrotliWriteBits(1, (uint64_t)is_uncompressed, storage_ix, storage); |
| 237 | } |
| 238 | |
| 239 | static BROTLI_INLINE void CreateCommands(const uint8_t* input, |
| 240 | size_t block_size, size_t input_size, const uint8_t* base_ip, int* table, |
| 241 | size_t table_bits, size_t min_match, |
| 242 | uint8_t** literals, uint32_t** commands) { |
| 243 | /* "ip" is the input pointer. */ |
| 244 | const uint8_t* ip = input; |
| 245 | const size_t shift = 64u - table_bits; |
| 246 | const uint8_t* ip_end = input + block_size; |
| 247 | /* "next_emit" is a pointer to the first byte that is not covered by a |
| 248 | previous copy. Bytes between "next_emit" and the start of the next copy or |
| 249 | the end of the input will be emitted as literal bytes. */ |
| 250 | const uint8_t* next_emit = input; |
| 251 | |
| 252 | int last_distance = -1; |
| 253 | const size_t kInputMarginBytes = BROTLI_WINDOW_GAP; |
| 254 | |
| 255 | if (BROTLI_PREDICT_TRUE(block_size >= kInputMarginBytes)) { |
| 256 | /* For the last block, we need to keep a 16 bytes margin so that we can be |
| 257 | sure that all distances are at most window size - 16. |
| 258 | For all other blocks, we only need to keep a margin of 5 bytes so that |
| 259 | we don't go over the block size with a copy. */ |
| 260 | const size_t len_limit = BROTLI_MIN(size_t, block_size - min_match, |
| 261 | input_size - kInputMarginBytes); |
| 262 | const uint8_t* ip_limit = input + len_limit; |
| 263 | |
| 264 | uint32_t next_hash; |
| 265 | for (next_hash = Hash(++ip, shift, min_match); ; ) { |
| 266 | /* Step 1: Scan forward in the input looking for a 6-byte-long match. |
| 267 | If we get close to exhausting the input then goto emit_remainder. |
| 268 | |
| 269 | Heuristic match skipping: If 32 bytes are scanned with no matches |
| 270 | found, start looking only at every other byte. If 32 more bytes are |
| 271 | scanned, look at every third byte, etc.. When a match is found, |
| 272 | immediately go back to looking at every byte. This is a small loss |
| 273 | (~5% performance, ~0.1% density) for compressible data due to more |
| 274 | bookkeeping, but for non-compressible data (such as JPEG) it's a huge |
| 275 | win since the compressor quickly "realizes" the data is incompressible |
| 276 | and doesn't bother looking for matches everywhere. |
| 277 | |
| 278 | The "skip" variable keeps track of how many bytes there are since the |
| 279 | last match; dividing it by 32 (ie. right-shifting by five) gives the |
| 280 | number of bytes to move ahead for each iteration. */ |
| 281 | uint32_t skip = 32; |
| 282 | |
| 283 | const uint8_t* next_ip = ip; |
| 284 | const uint8_t* candidate; |
| 285 | |
| 286 | BROTLI_DCHECK(next_emit < ip); |
| 287 | trawl: |
| 288 | do { |
| 289 | uint32_t hash = next_hash; |
| 290 | uint32_t bytes_between_hash_lookups = skip++ >> 5; |
| 291 | ip = next_ip; |
| 292 | BROTLI_DCHECK(hash == Hash(ip, shift, min_match)); |
| 293 | next_ip = ip + bytes_between_hash_lookups; |
| 294 | if (BROTLI_PREDICT_FALSE(next_ip > ip_limit)) { |
| 295 | goto emit_remainder; |
| 296 | } |
| 297 | next_hash = Hash(next_ip, shift, min_match); |
| 298 | candidate = ip - last_distance; |
| 299 | if (IsMatch(ip, candidate, min_match)) { |
| 300 | if (BROTLI_PREDICT_TRUE(candidate < ip)) { |
| 301 | table[hash] = (int)(ip - base_ip); |
| 302 | break; |
| 303 | } |
| 304 | } |
| 305 | candidate = base_ip + table[hash]; |
| 306 | BROTLI_DCHECK(candidate >= base_ip); |
| 307 | BROTLI_DCHECK(candidate < ip); |
| 308 | |
| 309 | table[hash] = (int)(ip - base_ip); |
| 310 | } while (BROTLI_PREDICT_TRUE(!IsMatch(ip, candidate, min_match))); |
| 311 | |
| 312 | /* Check copy distance. If candidate is not feasible, continue search. |
| 313 | Checking is done outside of hot loop to reduce overhead. */ |
| 314 | if (ip - candidate > MAX_DISTANCE) goto trawl; |
| 315 | |
| 316 | /* Step 2: Emit the found match together with the literal bytes from |
| 317 | "next_emit", and then see if we can find a next match immediately |
| 318 | afterwards. Repeat until we find no match for the input |
| 319 | without emitting some literal bytes. */ |
| 320 | |
| 321 | { |
| 322 | /* We have a 6-byte match at ip, and we need to emit bytes in |
| 323 | [next_emit, ip). */ |
| 324 | const uint8_t* base = ip; |
| 325 | size_t matched = min_match + FindMatchLengthWithLimit( |
| 326 | candidate + min_match, ip + min_match, |
| 327 | (size_t)(ip_end - ip) - min_match); |
| 328 | int distance = (int)(base - candidate); /* > 0 */ |
| 329 | int insert = (int)(base - next_emit); |
| 330 | ip += matched; |
| 331 | BROTLI_DCHECK(0 == memcmp(base, candidate, matched)); |
| 332 | EmitInsertLen((uint32_t)insert, commands); |
| 333 | memcpy(*literals, next_emit, (size_t)insert); |
| 334 | *literals += insert; |
| 335 | if (distance == last_distance) { |
| 336 | **commands = 64; |
| 337 | ++(*commands); |
| 338 | } else { |
| 339 | EmitDistance((uint32_t)distance, commands); |
| 340 | last_distance = distance; |
| 341 | } |
| 342 | EmitCopyLenLastDistance(matched, commands); |
| 343 | |
| 344 | next_emit = ip; |
| 345 | if (BROTLI_PREDICT_FALSE(ip >= ip_limit)) { |
| 346 | goto emit_remainder; |
| 347 | } |
| 348 | { |
| 349 | /* We could immediately start working at ip now, but to improve |
| 350 | compression we first update "table" with the hashes of some |
| 351 | positions within the last copy. */ |
| 352 | uint64_t input_bytes; |
| 353 | uint32_t cur_hash; |
| 354 | uint32_t prev_hash; |
| 355 | if (min_match == 4) { |
| 356 | input_bytes = BROTLI_UNALIGNED_LOAD64LE(ip - 3); |
| 357 | cur_hash = HashBytesAtOffset(input_bytes, 3, shift, min_match); |
| 358 | prev_hash = HashBytesAtOffset(input_bytes, 0, shift, min_match); |
| 359 | table[prev_hash] = (int)(ip - base_ip - 3); |
| 360 | prev_hash = HashBytesAtOffset(input_bytes, 1, shift, min_match); |
| 361 | table[prev_hash] = (int)(ip - base_ip - 2); |
| 362 | prev_hash = HashBytesAtOffset(input_bytes, 0, shift, min_match); |
| 363 | table[prev_hash] = (int)(ip - base_ip - 1); |
| 364 | } else { |
| 365 | input_bytes = BROTLI_UNALIGNED_LOAD64LE(ip - 5); |
| 366 | prev_hash = HashBytesAtOffset(input_bytes, 0, shift, min_match); |
| 367 | table[prev_hash] = (int)(ip - base_ip - 5); |
| 368 | prev_hash = HashBytesAtOffset(input_bytes, 1, shift, min_match); |
| 369 | table[prev_hash] = (int)(ip - base_ip - 4); |
| 370 | prev_hash = HashBytesAtOffset(input_bytes, 2, shift, min_match); |
| 371 | table[prev_hash] = (int)(ip - base_ip - 3); |
| 372 | input_bytes = BROTLI_UNALIGNED_LOAD64LE(ip - 2); |
| 373 | cur_hash = HashBytesAtOffset(input_bytes, 2, shift, min_match); |
| 374 | prev_hash = HashBytesAtOffset(input_bytes, 0, shift, min_match); |
| 375 | table[prev_hash] = (int)(ip - base_ip - 2); |
| 376 | prev_hash = HashBytesAtOffset(input_bytes, 1, shift, min_match); |
| 377 | table[prev_hash] = (int)(ip - base_ip - 1); |
| 378 | } |
| 379 | |
| 380 | candidate = base_ip + table[cur_hash]; |
| 381 | table[cur_hash] = (int)(ip - base_ip); |
| 382 | } |
| 383 | } |
| 384 | |
| 385 | while (ip - candidate <= MAX_DISTANCE && |
| 386 | IsMatch(ip, candidate, min_match)) { |
| 387 | /* We have a 6-byte match at ip, and no need to emit any |
| 388 | literal bytes prior to ip. */ |
| 389 | const uint8_t* base = ip; |
| 390 | size_t matched = min_match + FindMatchLengthWithLimit( |
| 391 | candidate + min_match, ip + min_match, |
| 392 | (size_t)(ip_end - ip) - min_match); |
| 393 | ip += matched; |
| 394 | last_distance = (int)(base - candidate); /* > 0 */ |
| 395 | BROTLI_DCHECK(0 == memcmp(base, candidate, matched)); |
| 396 | EmitCopyLen(matched, commands); |
| 397 | EmitDistance((uint32_t)last_distance, commands); |
| 398 | |
| 399 | next_emit = ip; |
| 400 | if (BROTLI_PREDICT_FALSE(ip >= ip_limit)) { |
| 401 | goto emit_remainder; |
| 402 | } |
| 403 | { |
| 404 | /* We could immediately start working at ip now, but to improve |
| 405 | compression we first update "table" with the hashes of some |
| 406 | positions within the last copy. */ |
| 407 | uint64_t input_bytes; |
| 408 | uint32_t cur_hash; |
| 409 | uint32_t prev_hash; |
| 410 | if (min_match == 4) { |
| 411 | input_bytes = BROTLI_UNALIGNED_LOAD64LE(ip - 3); |
| 412 | cur_hash = HashBytesAtOffset(input_bytes, 3, shift, min_match); |
| 413 | prev_hash = HashBytesAtOffset(input_bytes, 0, shift, min_match); |
| 414 | table[prev_hash] = (int)(ip - base_ip - 3); |
| 415 | prev_hash = HashBytesAtOffset(input_bytes, 1, shift, min_match); |
| 416 | table[prev_hash] = (int)(ip - base_ip - 2); |
| 417 | prev_hash = HashBytesAtOffset(input_bytes, 2, shift, min_match); |
| 418 | table[prev_hash] = (int)(ip - base_ip - 1); |
| 419 | } else { |
| 420 | input_bytes = BROTLI_UNALIGNED_LOAD64LE(ip - 5); |
| 421 | prev_hash = HashBytesAtOffset(input_bytes, 0, shift, min_match); |
| 422 | table[prev_hash] = (int)(ip - base_ip - 5); |
| 423 | prev_hash = HashBytesAtOffset(input_bytes, 1, shift, min_match); |
| 424 | table[prev_hash] = (int)(ip - base_ip - 4); |
| 425 | prev_hash = HashBytesAtOffset(input_bytes, 2, shift, min_match); |
| 426 | table[prev_hash] = (int)(ip - base_ip - 3); |
| 427 | input_bytes = BROTLI_UNALIGNED_LOAD64LE(ip - 2); |
| 428 | cur_hash = HashBytesAtOffset(input_bytes, 2, shift, min_match); |
| 429 | prev_hash = HashBytesAtOffset(input_bytes, 0, shift, min_match); |
| 430 | table[prev_hash] = (int)(ip - base_ip - 2); |
| 431 | prev_hash = HashBytesAtOffset(input_bytes, 1, shift, min_match); |
| 432 | table[prev_hash] = (int)(ip - base_ip - 1); |
| 433 | } |
| 434 | |
| 435 | candidate = base_ip + table[cur_hash]; |
| 436 | table[cur_hash] = (int)(ip - base_ip); |
| 437 | } |
| 438 | } |
| 439 | |
| 440 | next_hash = Hash(++ip, shift, min_match); |
| 441 | } |
| 442 | } |
| 443 | |
| 444 | emit_remainder: |
| 445 | BROTLI_DCHECK(next_emit <= ip_end); |
| 446 | /* Emit the remaining bytes as literals. */ |
| 447 | if (next_emit < ip_end) { |
| 448 | const uint32_t insert = (uint32_t)(ip_end - next_emit); |
| 449 | EmitInsertLen(insert, commands); |
| 450 | memcpy(*literals, next_emit, insert); |
| 451 | *literals += insert; |
| 452 | } |
| 453 | } |
| 454 | |
| 455 | static void StoreCommands(MemoryManager* m, |
| 456 | const uint8_t* literals, const size_t num_literals, |
| 457 | const uint32_t* commands, const size_t num_commands, |
| 458 | size_t* storage_ix, uint8_t* storage) { |
| 459 | static const uint32_t [128] = { |
| 460 | 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 7, 8, 9, 10, 12, 14, 24, |
| 461 | 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, |
| 462 | 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 7, 8, 9, 10, 24, |
| 463 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 464 | 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, |
| 465 | 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14, 14, 15, 15, 16, 16, |
| 466 | 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, 23, 23, 24, 24, |
| 467 | }; |
| 468 | static const uint32_t kInsertOffset[24] = { |
| 469 | 0, 1, 2, 3, 4, 5, 6, 8, 10, 14, 18, 26, 34, 50, 66, 98, 130, 194, 322, 578, |
| 470 | 1090, 2114, 6210, 22594, |
| 471 | }; |
| 472 | |
| 473 | uint8_t lit_depths[256]; |
| 474 | uint16_t lit_bits[256]; |
| 475 | uint32_t lit_histo[256] = { 0 }; |
| 476 | uint8_t cmd_depths[128] = { 0 }; |
| 477 | uint16_t cmd_bits[128] = { 0 }; |
| 478 | uint32_t cmd_histo[128] = { 0 }; |
| 479 | size_t i; |
| 480 | for (i = 0; i < num_literals; ++i) { |
| 481 | ++lit_histo[literals[i]]; |
| 482 | } |
| 483 | BrotliBuildAndStoreHuffmanTreeFast(m, lit_histo, num_literals, |
| 484 | /* max_bits = */ 8, |
| 485 | lit_depths, lit_bits, |
| 486 | storage_ix, storage); |
| 487 | if (BROTLI_IS_OOM(m)) return; |
| 488 | |
| 489 | for (i = 0; i < num_commands; ++i) { |
| 490 | const uint32_t code = commands[i] & 0xFF; |
| 491 | BROTLI_DCHECK(code < 128); |
| 492 | ++cmd_histo[code]; |
| 493 | } |
| 494 | cmd_histo[1] += 1; |
| 495 | cmd_histo[2] += 1; |
| 496 | cmd_histo[64] += 1; |
| 497 | cmd_histo[84] += 1; |
| 498 | BuildAndStoreCommandPrefixCode(cmd_histo, cmd_depths, cmd_bits, |
| 499 | storage_ix, storage); |
| 500 | |
| 501 | for (i = 0; i < num_commands; ++i) { |
| 502 | const uint32_t cmd = commands[i]; |
| 503 | const uint32_t code = cmd & 0xFF; |
| 504 | const uint32_t = cmd >> 8; |
| 505 | BROTLI_DCHECK(code < 128); |
| 506 | BrotliWriteBits(cmd_depths[code], cmd_bits[code], storage_ix, storage); |
| 507 | BrotliWriteBits(kNumExtraBits[code], extra, storage_ix, storage); |
| 508 | if (code < 24) { |
| 509 | const uint32_t insert = kInsertOffset[code] + extra; |
| 510 | uint32_t j; |
| 511 | for (j = 0; j < insert; ++j) { |
| 512 | const uint8_t lit = *literals; |
| 513 | BrotliWriteBits(lit_depths[lit], lit_bits[lit], storage_ix, storage); |
| 514 | ++literals; |
| 515 | } |
| 516 | } |
| 517 | } |
| 518 | } |
| 519 | |
| 520 | /* Acceptable loss for uncompressible speedup is 2% */ |
| 521 | #define MIN_RATIO 0.98 |
| 522 | #define SAMPLE_RATE 43 |
| 523 | |
| 524 | static BROTLI_BOOL ShouldCompress( |
| 525 | const uint8_t* input, size_t input_size, size_t num_literals) { |
| 526 | double corpus_size = (double)input_size; |
| 527 | if (num_literals < MIN_RATIO * corpus_size) { |
| 528 | return BROTLI_TRUE; |
| 529 | } else { |
| 530 | uint32_t literal_histo[256] = { 0 }; |
| 531 | const double max_total_bit_cost = corpus_size * 8 * MIN_RATIO / SAMPLE_RATE; |
| 532 | size_t i; |
| 533 | for (i = 0; i < input_size; i += SAMPLE_RATE) { |
| 534 | ++literal_histo[input[i]]; |
| 535 | } |
| 536 | return TO_BROTLI_BOOL(BitsEntropy(literal_histo, 256) < max_total_bit_cost); |
| 537 | } |
| 538 | } |
| 539 | |
| 540 | static void RewindBitPosition(const size_t new_storage_ix, |
| 541 | size_t* storage_ix, uint8_t* storage) { |
| 542 | const size_t bitpos = new_storage_ix & 7; |
| 543 | const size_t mask = (1u << bitpos) - 1; |
| 544 | storage[new_storage_ix >> 3] &= (uint8_t)mask; |
| 545 | *storage_ix = new_storage_ix; |
| 546 | } |
| 547 | |
| 548 | static void EmitUncompressedMetaBlock(const uint8_t* input, size_t input_size, |
| 549 | size_t* storage_ix, uint8_t* storage) { |
| 550 | BrotliStoreMetaBlockHeader(input_size, 1, storage_ix, storage); |
| 551 | *storage_ix = (*storage_ix + 7u) & ~7u; |
| 552 | memcpy(&storage[*storage_ix >> 3], input, input_size); |
| 553 | *storage_ix += input_size << 3; |
| 554 | storage[*storage_ix >> 3] = 0; |
| 555 | } |
| 556 | |
| 557 | static BROTLI_INLINE void BrotliCompressFragmentTwoPassImpl( |
| 558 | MemoryManager* m, const uint8_t* input, size_t input_size, |
| 559 | BROTLI_BOOL is_last, uint32_t* command_buf, uint8_t* literal_buf, |
| 560 | int* table, size_t table_bits, size_t min_match, |
| 561 | size_t* storage_ix, uint8_t* storage) { |
| 562 | /* Save the start of the first block for position and distance computations. |
| 563 | */ |
| 564 | const uint8_t* base_ip = input; |
| 565 | BROTLI_UNUSED(is_last); |
| 566 | |
| 567 | while (input_size > 0) { |
| 568 | size_t block_size = |
| 569 | BROTLI_MIN(size_t, input_size, kCompressFragmentTwoPassBlockSize); |
| 570 | uint32_t* commands = command_buf; |
| 571 | uint8_t* literals = literal_buf; |
| 572 | size_t num_literals; |
| 573 | CreateCommands(input, block_size, input_size, base_ip, table, |
| 574 | table_bits, min_match, &literals, &commands); |
| 575 | num_literals = (size_t)(literals - literal_buf); |
| 576 | if (ShouldCompress(input, block_size, num_literals)) { |
| 577 | const size_t num_commands = (size_t)(commands - command_buf); |
| 578 | BrotliStoreMetaBlockHeader(block_size, 0, storage_ix, storage); |
| 579 | /* No block splits, no contexts. */ |
| 580 | BrotliWriteBits(13, 0, storage_ix, storage); |
| 581 | StoreCommands(m, literal_buf, num_literals, command_buf, num_commands, |
| 582 | storage_ix, storage); |
| 583 | if (BROTLI_IS_OOM(m)) return; |
| 584 | } else { |
| 585 | /* Since we did not find many backward references and the entropy of |
| 586 | the data is close to 8 bits, we can simply emit an uncompressed block. |
| 587 | This makes compression speed of uncompressible data about 3x faster. */ |
| 588 | EmitUncompressedMetaBlock(input, block_size, storage_ix, storage); |
| 589 | } |
| 590 | input += block_size; |
| 591 | input_size -= block_size; |
| 592 | } |
| 593 | } |
| 594 | |
| 595 | #define FOR_TABLE_BITS_(X) \ |
| 596 | X(8) X(9) X(10) X(11) X(12) X(13) X(14) X(15) X(16) X(17) |
| 597 | |
| 598 | #define BAKE_METHOD_PARAM_(B) \ |
| 599 | static BROTLI_NOINLINE void BrotliCompressFragmentTwoPassImpl ## B( \ |
| 600 | MemoryManager* m, const uint8_t* input, size_t input_size, \ |
| 601 | BROTLI_BOOL is_last, uint32_t* command_buf, uint8_t* literal_buf, \ |
| 602 | int* table, size_t* storage_ix, uint8_t* storage) { \ |
| 603 | size_t min_match = (B <= 15) ? 4 : 6; \ |
| 604 | BrotliCompressFragmentTwoPassImpl(m, input, input_size, is_last, command_buf,\ |
| 605 | literal_buf, table, B, min_match, storage_ix, storage); \ |
| 606 | } |
| 607 | FOR_TABLE_BITS_(BAKE_METHOD_PARAM_) |
| 608 | #undef BAKE_METHOD_PARAM_ |
| 609 | |
| 610 | void BrotliCompressFragmentTwoPass( |
| 611 | MemoryManager* m, const uint8_t* input, size_t input_size, |
| 612 | BROTLI_BOOL is_last, uint32_t* command_buf, uint8_t* literal_buf, |
| 613 | int* table, size_t table_size, size_t* storage_ix, uint8_t* storage) { |
| 614 | const size_t initial_storage_ix = *storage_ix; |
| 615 | const size_t table_bits = Log2FloorNonZero(table_size); |
| 616 | switch (table_bits) { |
| 617 | #define CASE_(B) \ |
| 618 | case B: \ |
| 619 | BrotliCompressFragmentTwoPassImpl ## B( \ |
| 620 | m, input, input_size, is_last, command_buf, \ |
| 621 | literal_buf, table, storage_ix, storage); \ |
| 622 | break; |
| 623 | FOR_TABLE_BITS_(CASE_) |
| 624 | #undef CASE_ |
| 625 | default: BROTLI_DCHECK(0); break; |
| 626 | } |
| 627 | |
| 628 | /* If output is larger than single uncompressed block, rewrite it. */ |
| 629 | if (*storage_ix - initial_storage_ix > 31 + (input_size << 3)) { |
| 630 | RewindBitPosition(initial_storage_ix, storage_ix, storage); |
| 631 | EmitUncompressedMetaBlock(input, input_size, storage_ix, storage); |
| 632 | } |
| 633 | |
| 634 | if (is_last) { |
| 635 | BrotliWriteBits(1, 1, storage_ix, storage); /* islast */ |
| 636 | BrotliWriteBits(1, 1, storage_ix, storage); /* isempty */ |
| 637 | *storage_ix = (*storage_ix + 7u) & ~7u; |
| 638 | } |
| 639 | } |
| 640 | |
| 641 | #undef FOR_TABLE_BITS_ |
| 642 | |
| 643 | #if defined(__cplusplus) || defined(c_plusplus) |
| 644 | } /* extern "C" */ |
| 645 | #endif |
| 646 | |