| 1 | /* Copyright 2014 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 | /* Brotli bit stream functions to support the low level format. There are no |
| 8 | compression algorithms here, just the right ordering of bits to match the |
| 9 | specs. */ |
| 10 | |
| 11 | #include "./brotli_bit_stream.h" |
| 12 | |
| 13 | #include <string.h> /* memcpy, memset */ |
| 14 | |
| 15 | #include "../common/constants.h" |
| 16 | #include "../common/context.h" |
| 17 | #include "../common/platform.h" |
| 18 | #include <brotli/types.h> |
| 19 | #include "./entropy_encode.h" |
| 20 | #include "./entropy_encode_static.h" |
| 21 | #include "./fast_log.h" |
| 22 | #include "./histogram.h" |
| 23 | #include "./memory.h" |
| 24 | #include "./write_bits.h" |
| 25 | |
| 26 | #if defined(__cplusplus) || defined(c_plusplus) |
| 27 | extern "C" { |
| 28 | #endif |
| 29 | |
| 30 | #define MAX_HUFFMAN_TREE_SIZE (2 * BROTLI_NUM_COMMAND_SYMBOLS + 1) |
| 31 | /* The maximum size of Huffman dictionary for distances assuming that |
| 32 | NPOSTFIX = 0 and NDIRECT = 0. */ |
| 33 | #define MAX_SIMPLE_DISTANCE_ALPHABET_SIZE \ |
| 34 | BROTLI_DISTANCE_ALPHABET_SIZE(0, 0, BROTLI_LARGE_MAX_DISTANCE_BITS) |
| 35 | /* MAX_SIMPLE_DISTANCE_ALPHABET_SIZE == 140 */ |
| 36 | |
| 37 | /* Represents the range of values belonging to a prefix code: |
| 38 | [offset, offset + 2^nbits) */ |
| 39 | typedef struct PrefixCodeRange { |
| 40 | uint32_t offset; |
| 41 | uint32_t nbits; |
| 42 | } PrefixCodeRange; |
| 43 | |
| 44 | static const PrefixCodeRange |
| 45 | kBlockLengthPrefixCode[BROTLI_NUM_BLOCK_LEN_SYMBOLS] = { |
| 46 | { 1, 2}, { 5, 2}, { 9, 2}, {13, 2}, {17, 3}, { 25, 3}, { 33, 3}, |
| 47 | {41, 3}, {49, 4}, {65, 4}, {81, 4}, {97, 4}, {113, 5}, {145, 5}, |
| 48 | {177, 5}, { 209, 5}, { 241, 6}, { 305, 6}, { 369, 7}, { 497, 8}, |
| 49 | {753, 9}, {1265, 10}, {2289, 11}, {4337, 12}, {8433, 13}, {16625, 24} |
| 50 | }; |
| 51 | |
| 52 | static BROTLI_INLINE uint32_t BlockLengthPrefixCode(uint32_t len) { |
| 53 | uint32_t code = (len >= 177) ? (len >= 753 ? 20 : 14) : (len >= 41 ? 7 : 0); |
| 54 | while (code < (BROTLI_NUM_BLOCK_LEN_SYMBOLS - 1) && |
| 55 | len >= kBlockLengthPrefixCode[code + 1].offset) ++code; |
| 56 | return code; |
| 57 | } |
| 58 | |
| 59 | static BROTLI_INLINE void GetBlockLengthPrefixCode(uint32_t len, size_t* code, |
| 60 | uint32_t* , uint32_t* ) { |
| 61 | *code = BlockLengthPrefixCode(len); |
| 62 | *n_extra = kBlockLengthPrefixCode[*code].nbits; |
| 63 | *extra = len - kBlockLengthPrefixCode[*code].offset; |
| 64 | } |
| 65 | |
| 66 | typedef struct BlockTypeCodeCalculator { |
| 67 | size_t last_type; |
| 68 | size_t second_last_type; |
| 69 | } BlockTypeCodeCalculator; |
| 70 | |
| 71 | static void InitBlockTypeCodeCalculator(BlockTypeCodeCalculator* self) { |
| 72 | self->last_type = 1; |
| 73 | self->second_last_type = 0; |
| 74 | } |
| 75 | |
| 76 | static BROTLI_INLINE size_t NextBlockTypeCode( |
| 77 | BlockTypeCodeCalculator* calculator, uint8_t type) { |
| 78 | size_t type_code = (type == calculator->last_type + 1) ? 1u : |
| 79 | (type == calculator->second_last_type) ? 0u : type + 2u; |
| 80 | calculator->second_last_type = calculator->last_type; |
| 81 | calculator->last_type = type; |
| 82 | return type_code; |
| 83 | } |
| 84 | |
| 85 | /* |nibblesbits| represents the 2 bits to encode MNIBBLES (0-3) |
| 86 | REQUIRES: length > 0 |
| 87 | REQUIRES: length <= (1 << 24) */ |
| 88 | static void BrotliEncodeMlen(size_t length, uint64_t* bits, |
| 89 | size_t* numbits, uint64_t* nibblesbits) { |
| 90 | size_t lg = (length == 1) ? 1 : Log2FloorNonZero((uint32_t)(length - 1)) + 1; |
| 91 | size_t mnibbles = (lg < 16 ? 16 : (lg + 3)) / 4; |
| 92 | BROTLI_DCHECK(length > 0); |
| 93 | BROTLI_DCHECK(length <= (1 << 24)); |
| 94 | BROTLI_DCHECK(lg <= 24); |
| 95 | *nibblesbits = mnibbles - 4; |
| 96 | *numbits = mnibbles * 4; |
| 97 | *bits = length - 1; |
| 98 | } |
| 99 | |
| 100 | static BROTLI_INLINE void StoreCommandExtra( |
| 101 | const Command* cmd, size_t* storage_ix, uint8_t* storage) { |
| 102 | uint32_t copylen_code = CommandCopyLenCode(cmd); |
| 103 | uint16_t inscode = GetInsertLengthCode(cmd->insert_len_); |
| 104 | uint16_t copycode = GetCopyLengthCode(copylen_code); |
| 105 | uint32_t = GetInsertExtra(inscode); |
| 106 | uint64_t = cmd->insert_len_ - GetInsertBase(inscode); |
| 107 | uint64_t = copylen_code - GetCopyBase(copycode); |
| 108 | uint64_t bits = (copyextraval << insnumextra) | insextraval; |
| 109 | BrotliWriteBits( |
| 110 | insnumextra + GetCopyExtra(copycode), bits, storage_ix, storage); |
| 111 | } |
| 112 | |
| 113 | /* Data structure that stores almost everything that is needed to encode each |
| 114 | block switch command. */ |
| 115 | typedef struct BlockSplitCode { |
| 116 | BlockTypeCodeCalculator type_code_calculator; |
| 117 | uint8_t type_depths[BROTLI_MAX_BLOCK_TYPE_SYMBOLS]; |
| 118 | uint16_t type_bits[BROTLI_MAX_BLOCK_TYPE_SYMBOLS]; |
| 119 | uint8_t length_depths[BROTLI_NUM_BLOCK_LEN_SYMBOLS]; |
| 120 | uint16_t length_bits[BROTLI_NUM_BLOCK_LEN_SYMBOLS]; |
| 121 | } BlockSplitCode; |
| 122 | |
| 123 | /* Stores a number between 0 and 255. */ |
| 124 | static void StoreVarLenUint8(size_t n, size_t* storage_ix, uint8_t* storage) { |
| 125 | if (n == 0) { |
| 126 | BrotliWriteBits(1, 0, storage_ix, storage); |
| 127 | } else { |
| 128 | size_t nbits = Log2FloorNonZero(n); |
| 129 | BrotliWriteBits(1, 1, storage_ix, storage); |
| 130 | BrotliWriteBits(3, nbits, storage_ix, storage); |
| 131 | BrotliWriteBits(nbits, n - ((size_t)1 << nbits), storage_ix, storage); |
| 132 | } |
| 133 | } |
| 134 | |
| 135 | /* Stores the compressed meta-block header. |
| 136 | REQUIRES: length > 0 |
| 137 | REQUIRES: length <= (1 << 24) */ |
| 138 | static void (BROTLI_BOOL is_final_block, |
| 139 | size_t length, |
| 140 | size_t* storage_ix, |
| 141 | uint8_t* storage) { |
| 142 | uint64_t lenbits; |
| 143 | size_t nlenbits; |
| 144 | uint64_t nibblesbits; |
| 145 | |
| 146 | /* Write ISLAST bit. */ |
| 147 | BrotliWriteBits(1, (uint64_t)is_final_block, storage_ix, storage); |
| 148 | /* Write ISEMPTY bit. */ |
| 149 | if (is_final_block) { |
| 150 | BrotliWriteBits(1, 0, storage_ix, storage); |
| 151 | } |
| 152 | |
| 153 | BrotliEncodeMlen(length, &lenbits, &nlenbits, &nibblesbits); |
| 154 | BrotliWriteBits(2, nibblesbits, storage_ix, storage); |
| 155 | BrotliWriteBits(nlenbits, lenbits, storage_ix, storage); |
| 156 | |
| 157 | if (!is_final_block) { |
| 158 | /* Write ISUNCOMPRESSED bit. */ |
| 159 | BrotliWriteBits(1, 0, storage_ix, storage); |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | /* Stores the uncompressed meta-block header. |
| 164 | REQUIRES: length > 0 |
| 165 | REQUIRES: length <= (1 << 24) */ |
| 166 | static void (size_t length, |
| 167 | size_t* storage_ix, |
| 168 | uint8_t* storage) { |
| 169 | uint64_t lenbits; |
| 170 | size_t nlenbits; |
| 171 | uint64_t nibblesbits; |
| 172 | |
| 173 | /* Write ISLAST bit. |
| 174 | Uncompressed block cannot be the last one, so set to 0. */ |
| 175 | BrotliWriteBits(1, 0, storage_ix, storage); |
| 176 | BrotliEncodeMlen(length, &lenbits, &nlenbits, &nibblesbits); |
| 177 | BrotliWriteBits(2, nibblesbits, storage_ix, storage); |
| 178 | BrotliWriteBits(nlenbits, lenbits, storage_ix, storage); |
| 179 | /* Write ISUNCOMPRESSED bit. */ |
| 180 | BrotliWriteBits(1, 1, storage_ix, storage); |
| 181 | } |
| 182 | |
| 183 | static void BrotliStoreHuffmanTreeOfHuffmanTreeToBitMask( |
| 184 | const int num_codes, const uint8_t* code_length_bitdepth, |
| 185 | size_t* storage_ix, uint8_t* storage) { |
| 186 | static const uint8_t kStorageOrder[BROTLI_CODE_LENGTH_CODES] = { |
| 187 | 1, 2, 3, 4, 0, 5, 17, 6, 16, 7, 8, 9, 10, 11, 12, 13, 14, 15 |
| 188 | }; |
| 189 | /* The bit lengths of the Huffman code over the code length alphabet |
| 190 | are compressed with the following static Huffman code: |
| 191 | Symbol Code |
| 192 | ------ ---- |
| 193 | 0 00 |
| 194 | 1 1110 |
| 195 | 2 110 |
| 196 | 3 01 |
| 197 | 4 10 |
| 198 | 5 1111 */ |
| 199 | static const uint8_t kHuffmanBitLengthHuffmanCodeSymbols[6] = { |
| 200 | 0, 7, 3, 2, 1, 15 |
| 201 | }; |
| 202 | static const uint8_t kHuffmanBitLengthHuffmanCodeBitLengths[6] = { |
| 203 | 2, 4, 3, 2, 2, 4 |
| 204 | }; |
| 205 | |
| 206 | size_t skip_some = 0; /* skips none. */ |
| 207 | |
| 208 | /* Throw away trailing zeros: */ |
| 209 | size_t codes_to_store = BROTLI_CODE_LENGTH_CODES; |
| 210 | if (num_codes > 1) { |
| 211 | for (; codes_to_store > 0; --codes_to_store) { |
| 212 | if (code_length_bitdepth[kStorageOrder[codes_to_store - 1]] != 0) { |
| 213 | break; |
| 214 | } |
| 215 | } |
| 216 | } |
| 217 | if (code_length_bitdepth[kStorageOrder[0]] == 0 && |
| 218 | code_length_bitdepth[kStorageOrder[1]] == 0) { |
| 219 | skip_some = 2; /* skips two. */ |
| 220 | if (code_length_bitdepth[kStorageOrder[2]] == 0) { |
| 221 | skip_some = 3; /* skips three. */ |
| 222 | } |
| 223 | } |
| 224 | BrotliWriteBits(2, skip_some, storage_ix, storage); |
| 225 | { |
| 226 | size_t i; |
| 227 | for (i = skip_some; i < codes_to_store; ++i) { |
| 228 | size_t l = code_length_bitdepth[kStorageOrder[i]]; |
| 229 | BrotliWriteBits(kHuffmanBitLengthHuffmanCodeBitLengths[l], |
| 230 | kHuffmanBitLengthHuffmanCodeSymbols[l], storage_ix, storage); |
| 231 | } |
| 232 | } |
| 233 | } |
| 234 | |
| 235 | static void BrotliStoreHuffmanTreeToBitMask( |
| 236 | const size_t huffman_tree_size, const uint8_t* huffman_tree, |
| 237 | const uint8_t* , const uint8_t* code_length_bitdepth, |
| 238 | const uint16_t* code_length_bitdepth_symbols, |
| 239 | size_t* BROTLI_RESTRICT storage_ix, uint8_t* BROTLI_RESTRICT storage) { |
| 240 | size_t i; |
| 241 | for (i = 0; i < huffman_tree_size; ++i) { |
| 242 | size_t ix = huffman_tree[i]; |
| 243 | BrotliWriteBits(code_length_bitdepth[ix], code_length_bitdepth_symbols[ix], |
| 244 | storage_ix, storage); |
| 245 | /* Extra bits */ |
| 246 | switch (ix) { |
| 247 | case BROTLI_REPEAT_PREVIOUS_CODE_LENGTH: |
| 248 | BrotliWriteBits(2, huffman_tree_extra_bits[i], storage_ix, storage); |
| 249 | break; |
| 250 | case BROTLI_REPEAT_ZERO_CODE_LENGTH: |
| 251 | BrotliWriteBits(3, huffman_tree_extra_bits[i], storage_ix, storage); |
| 252 | break; |
| 253 | } |
| 254 | } |
| 255 | } |
| 256 | |
| 257 | static void StoreSimpleHuffmanTree(const uint8_t* depths, |
| 258 | size_t symbols[4], |
| 259 | size_t num_symbols, |
| 260 | size_t max_bits, |
| 261 | size_t* storage_ix, uint8_t* storage) { |
| 262 | /* value of 1 indicates a simple Huffman code */ |
| 263 | BrotliWriteBits(2, 1, storage_ix, storage); |
| 264 | BrotliWriteBits(2, num_symbols - 1, storage_ix, storage); /* NSYM - 1 */ |
| 265 | |
| 266 | { |
| 267 | /* Sort */ |
| 268 | size_t i; |
| 269 | for (i = 0; i < num_symbols; i++) { |
| 270 | size_t j; |
| 271 | for (j = i + 1; j < num_symbols; j++) { |
| 272 | if (depths[symbols[j]] < depths[symbols[i]]) { |
| 273 | BROTLI_SWAP(size_t, symbols, j, i); |
| 274 | } |
| 275 | } |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | if (num_symbols == 2) { |
| 280 | BrotliWriteBits(max_bits, symbols[0], storage_ix, storage); |
| 281 | BrotliWriteBits(max_bits, symbols[1], storage_ix, storage); |
| 282 | } else if (num_symbols == 3) { |
| 283 | BrotliWriteBits(max_bits, symbols[0], storage_ix, storage); |
| 284 | BrotliWriteBits(max_bits, symbols[1], storage_ix, storage); |
| 285 | BrotliWriteBits(max_bits, symbols[2], storage_ix, storage); |
| 286 | } else { |
| 287 | BrotliWriteBits(max_bits, symbols[0], storage_ix, storage); |
| 288 | BrotliWriteBits(max_bits, symbols[1], storage_ix, storage); |
| 289 | BrotliWriteBits(max_bits, symbols[2], storage_ix, storage); |
| 290 | BrotliWriteBits(max_bits, symbols[3], storage_ix, storage); |
| 291 | /* tree-select */ |
| 292 | BrotliWriteBits(1, depths[symbols[0]] == 1 ? 1 : 0, storage_ix, storage); |
| 293 | } |
| 294 | } |
| 295 | |
| 296 | /* num = alphabet size |
| 297 | depths = symbol depths */ |
| 298 | void BrotliStoreHuffmanTree(const uint8_t* depths, size_t num, |
| 299 | HuffmanTree* tree, |
| 300 | size_t* storage_ix, uint8_t* storage) { |
| 301 | /* Write the Huffman tree into the brotli-representation. |
| 302 | The command alphabet is the largest, so this allocation will fit all |
| 303 | alphabets. */ |
| 304 | uint8_t huffman_tree[BROTLI_NUM_COMMAND_SYMBOLS]; |
| 305 | uint8_t [BROTLI_NUM_COMMAND_SYMBOLS]; |
| 306 | size_t huffman_tree_size = 0; |
| 307 | uint8_t code_length_bitdepth[BROTLI_CODE_LENGTH_CODES] = { 0 }; |
| 308 | uint16_t code_length_bitdepth_symbols[BROTLI_CODE_LENGTH_CODES]; |
| 309 | uint32_t huffman_tree_histogram[BROTLI_CODE_LENGTH_CODES] = { 0 }; |
| 310 | size_t i; |
| 311 | int num_codes = 0; |
| 312 | size_t code = 0; |
| 313 | |
| 314 | BROTLI_DCHECK(num <= BROTLI_NUM_COMMAND_SYMBOLS); |
| 315 | |
| 316 | BrotliWriteHuffmanTree(depths, num, &huffman_tree_size, huffman_tree, |
| 317 | huffman_tree_extra_bits); |
| 318 | |
| 319 | /* Calculate the statistics of the Huffman tree in brotli-representation. */ |
| 320 | for (i = 0; i < huffman_tree_size; ++i) { |
| 321 | ++huffman_tree_histogram[huffman_tree[i]]; |
| 322 | } |
| 323 | |
| 324 | for (i = 0; i < BROTLI_CODE_LENGTH_CODES; ++i) { |
| 325 | if (huffman_tree_histogram[i]) { |
| 326 | if (num_codes == 0) { |
| 327 | code = i; |
| 328 | num_codes = 1; |
| 329 | } else if (num_codes == 1) { |
| 330 | num_codes = 2; |
| 331 | break; |
| 332 | } |
| 333 | } |
| 334 | } |
| 335 | |
| 336 | /* Calculate another Huffman tree to use for compressing both the |
| 337 | earlier Huffman tree with. */ |
| 338 | BrotliCreateHuffmanTree(huffman_tree_histogram, BROTLI_CODE_LENGTH_CODES, |
| 339 | 5, tree, code_length_bitdepth); |
| 340 | BrotliConvertBitDepthsToSymbols(code_length_bitdepth, |
| 341 | BROTLI_CODE_LENGTH_CODES, |
| 342 | code_length_bitdepth_symbols); |
| 343 | |
| 344 | /* Now, we have all the data, let's start storing it */ |
| 345 | BrotliStoreHuffmanTreeOfHuffmanTreeToBitMask(num_codes, code_length_bitdepth, |
| 346 | storage_ix, storage); |
| 347 | |
| 348 | if (num_codes == 1) { |
| 349 | code_length_bitdepth[code] = 0; |
| 350 | } |
| 351 | |
| 352 | /* Store the real Huffman tree now. */ |
| 353 | BrotliStoreHuffmanTreeToBitMask(huffman_tree_size, |
| 354 | huffman_tree, |
| 355 | huffman_tree_extra_bits, |
| 356 | code_length_bitdepth, |
| 357 | code_length_bitdepth_symbols, |
| 358 | storage_ix, storage); |
| 359 | } |
| 360 | |
| 361 | /* Builds a Huffman tree from histogram[0:length] into depth[0:length] and |
| 362 | bits[0:length] and stores the encoded tree to the bit stream. */ |
| 363 | static void BuildAndStoreHuffmanTree(const uint32_t* histogram, |
| 364 | const size_t histogram_length, |
| 365 | const size_t alphabet_size, |
| 366 | HuffmanTree* tree, |
| 367 | uint8_t* depth, |
| 368 | uint16_t* bits, |
| 369 | size_t* storage_ix, |
| 370 | uint8_t* storage) { |
| 371 | size_t count = 0; |
| 372 | size_t s4[4] = { 0 }; |
| 373 | size_t i; |
| 374 | size_t max_bits = 0; |
| 375 | for (i = 0; i < histogram_length; i++) { |
| 376 | if (histogram[i]) { |
| 377 | if (count < 4) { |
| 378 | s4[count] = i; |
| 379 | } else if (count > 4) { |
| 380 | break; |
| 381 | } |
| 382 | count++; |
| 383 | } |
| 384 | } |
| 385 | |
| 386 | { |
| 387 | size_t max_bits_counter = alphabet_size - 1; |
| 388 | while (max_bits_counter) { |
| 389 | max_bits_counter >>= 1; |
| 390 | ++max_bits; |
| 391 | } |
| 392 | } |
| 393 | |
| 394 | if (count <= 1) { |
| 395 | BrotliWriteBits(4, 1, storage_ix, storage); |
| 396 | BrotliWriteBits(max_bits, s4[0], storage_ix, storage); |
| 397 | depth[s4[0]] = 0; |
| 398 | bits[s4[0]] = 0; |
| 399 | return; |
| 400 | } |
| 401 | |
| 402 | memset(depth, 0, histogram_length * sizeof(depth[0])); |
| 403 | BrotliCreateHuffmanTree(histogram, histogram_length, 15, tree, depth); |
| 404 | BrotliConvertBitDepthsToSymbols(depth, histogram_length, bits); |
| 405 | |
| 406 | if (count <= 4) { |
| 407 | StoreSimpleHuffmanTree(depth, s4, count, max_bits, storage_ix, storage); |
| 408 | } else { |
| 409 | BrotliStoreHuffmanTree(depth, histogram_length, tree, storage_ix, storage); |
| 410 | } |
| 411 | } |
| 412 | |
| 413 | static BROTLI_INLINE BROTLI_BOOL SortHuffmanTree( |
| 414 | const HuffmanTree* v0, const HuffmanTree* v1) { |
| 415 | return TO_BROTLI_BOOL(v0->total_count_ < v1->total_count_); |
| 416 | } |
| 417 | |
| 418 | void BrotliBuildAndStoreHuffmanTreeFast(MemoryManager* m, |
| 419 | const uint32_t* histogram, |
| 420 | const size_t histogram_total, |
| 421 | const size_t max_bits, |
| 422 | uint8_t* depth, uint16_t* bits, |
| 423 | size_t* storage_ix, |
| 424 | uint8_t* storage) { |
| 425 | size_t count = 0; |
| 426 | size_t symbols[4] = { 0 }; |
| 427 | size_t length = 0; |
| 428 | size_t total = histogram_total; |
| 429 | while (total != 0) { |
| 430 | if (histogram[length]) { |
| 431 | if (count < 4) { |
| 432 | symbols[count] = length; |
| 433 | } |
| 434 | ++count; |
| 435 | total -= histogram[length]; |
| 436 | } |
| 437 | ++length; |
| 438 | } |
| 439 | |
| 440 | if (count <= 1) { |
| 441 | BrotliWriteBits(4, 1, storage_ix, storage); |
| 442 | BrotliWriteBits(max_bits, symbols[0], storage_ix, storage); |
| 443 | depth[symbols[0]] = 0; |
| 444 | bits[symbols[0]] = 0; |
| 445 | return; |
| 446 | } |
| 447 | |
| 448 | memset(depth, 0, length * sizeof(depth[0])); |
| 449 | { |
| 450 | const size_t max_tree_size = 2 * length + 1; |
| 451 | HuffmanTree* tree = BROTLI_ALLOC(m, HuffmanTree, max_tree_size); |
| 452 | uint32_t count_limit; |
| 453 | if (BROTLI_IS_OOM(m)) return; |
| 454 | for (count_limit = 1; ; count_limit *= 2) { |
| 455 | HuffmanTree* node = tree; |
| 456 | size_t l; |
| 457 | for (l = length; l != 0;) { |
| 458 | --l; |
| 459 | if (histogram[l]) { |
| 460 | if (BROTLI_PREDICT_TRUE(histogram[l] >= count_limit)) { |
| 461 | InitHuffmanTree(node, histogram[l], -1, (int16_t)l); |
| 462 | } else { |
| 463 | InitHuffmanTree(node, count_limit, -1, (int16_t)l); |
| 464 | } |
| 465 | ++node; |
| 466 | } |
| 467 | } |
| 468 | { |
| 469 | const int n = (int)(node - tree); |
| 470 | HuffmanTree sentinel; |
| 471 | int i = 0; /* Points to the next leaf node. */ |
| 472 | int j = n + 1; /* Points to the next non-leaf node. */ |
| 473 | int k; |
| 474 | |
| 475 | SortHuffmanTreeItems(tree, (size_t)n, SortHuffmanTree); |
| 476 | /* The nodes are: |
| 477 | [0, n): the sorted leaf nodes that we start with. |
| 478 | [n]: we add a sentinel here. |
| 479 | [n + 1, 2n): new parent nodes are added here, starting from |
| 480 | (n+1). These are naturally in ascending order. |
| 481 | [2n]: we add a sentinel at the end as well. |
| 482 | There will be (2n+1) elements at the end. */ |
| 483 | InitHuffmanTree(&sentinel, BROTLI_UINT32_MAX, -1, -1); |
| 484 | *node++ = sentinel; |
| 485 | *node++ = sentinel; |
| 486 | |
| 487 | for (k = n - 1; k > 0; --k) { |
| 488 | int left, right; |
| 489 | if (tree[i].total_count_ <= tree[j].total_count_) { |
| 490 | left = i; |
| 491 | ++i; |
| 492 | } else { |
| 493 | left = j; |
| 494 | ++j; |
| 495 | } |
| 496 | if (tree[i].total_count_ <= tree[j].total_count_) { |
| 497 | right = i; |
| 498 | ++i; |
| 499 | } else { |
| 500 | right = j; |
| 501 | ++j; |
| 502 | } |
| 503 | /* The sentinel node becomes the parent node. */ |
| 504 | node[-1].total_count_ = |
| 505 | tree[left].total_count_ + tree[right].total_count_; |
| 506 | node[-1].index_left_ = (int16_t)left; |
| 507 | node[-1].index_right_or_value_ = (int16_t)right; |
| 508 | /* Add back the last sentinel node. */ |
| 509 | *node++ = sentinel; |
| 510 | } |
| 511 | if (BrotliSetDepth(2 * n - 1, tree, depth, 14)) { |
| 512 | /* We need to pack the Huffman tree in 14 bits. If this was not |
| 513 | successful, add fake entities to the lowest values and retry. */ |
| 514 | break; |
| 515 | } |
| 516 | } |
| 517 | } |
| 518 | BROTLI_FREE(m, tree); |
| 519 | } |
| 520 | BrotliConvertBitDepthsToSymbols(depth, length, bits); |
| 521 | if (count <= 4) { |
| 522 | size_t i; |
| 523 | /* value of 1 indicates a simple Huffman code */ |
| 524 | BrotliWriteBits(2, 1, storage_ix, storage); |
| 525 | BrotliWriteBits(2, count - 1, storage_ix, storage); /* NSYM - 1 */ |
| 526 | |
| 527 | /* Sort */ |
| 528 | for (i = 0; i < count; i++) { |
| 529 | size_t j; |
| 530 | for (j = i + 1; j < count; j++) { |
| 531 | if (depth[symbols[j]] < depth[symbols[i]]) { |
| 532 | BROTLI_SWAP(size_t, symbols, j, i); |
| 533 | } |
| 534 | } |
| 535 | } |
| 536 | |
| 537 | if (count == 2) { |
| 538 | BrotliWriteBits(max_bits, symbols[0], storage_ix, storage); |
| 539 | BrotliWriteBits(max_bits, symbols[1], storage_ix, storage); |
| 540 | } else if (count == 3) { |
| 541 | BrotliWriteBits(max_bits, symbols[0], storage_ix, storage); |
| 542 | BrotliWriteBits(max_bits, symbols[1], storage_ix, storage); |
| 543 | BrotliWriteBits(max_bits, symbols[2], storage_ix, storage); |
| 544 | } else { |
| 545 | BrotliWriteBits(max_bits, symbols[0], storage_ix, storage); |
| 546 | BrotliWriteBits(max_bits, symbols[1], storage_ix, storage); |
| 547 | BrotliWriteBits(max_bits, symbols[2], storage_ix, storage); |
| 548 | BrotliWriteBits(max_bits, symbols[3], storage_ix, storage); |
| 549 | /* tree-select */ |
| 550 | BrotliWriteBits(1, depth[symbols[0]] == 1 ? 1 : 0, storage_ix, storage); |
| 551 | } |
| 552 | } else { |
| 553 | uint8_t previous_value = 8; |
| 554 | size_t i; |
| 555 | /* Complex Huffman Tree */ |
| 556 | StoreStaticCodeLengthCode(storage_ix, storage); |
| 557 | |
| 558 | /* Actual RLE coding. */ |
| 559 | for (i = 0; i < length;) { |
| 560 | const uint8_t value = depth[i]; |
| 561 | size_t reps = 1; |
| 562 | size_t k; |
| 563 | for (k = i + 1; k < length && depth[k] == value; ++k) { |
| 564 | ++reps; |
| 565 | } |
| 566 | i += reps; |
| 567 | if (value == 0) { |
| 568 | BrotliWriteBits(kZeroRepsDepth[reps], kZeroRepsBits[reps], |
| 569 | storage_ix, storage); |
| 570 | } else { |
| 571 | if (previous_value != value) { |
| 572 | BrotliWriteBits(kCodeLengthDepth[value], kCodeLengthBits[value], |
| 573 | storage_ix, storage); |
| 574 | --reps; |
| 575 | } |
| 576 | if (reps < 3) { |
| 577 | while (reps != 0) { |
| 578 | reps--; |
| 579 | BrotliWriteBits(kCodeLengthDepth[value], kCodeLengthBits[value], |
| 580 | storage_ix, storage); |
| 581 | } |
| 582 | } else { |
| 583 | reps -= 3; |
| 584 | BrotliWriteBits(kNonZeroRepsDepth[reps], kNonZeroRepsBits[reps], |
| 585 | storage_ix, storage); |
| 586 | } |
| 587 | previous_value = value; |
| 588 | } |
| 589 | } |
| 590 | } |
| 591 | } |
| 592 | |
| 593 | static size_t IndexOf(const uint8_t* v, size_t v_size, uint8_t value) { |
| 594 | size_t i = 0; |
| 595 | for (; i < v_size; ++i) { |
| 596 | if (v[i] == value) return i; |
| 597 | } |
| 598 | return i; |
| 599 | } |
| 600 | |
| 601 | static void MoveToFront(uint8_t* v, size_t index) { |
| 602 | uint8_t value = v[index]; |
| 603 | size_t i; |
| 604 | for (i = index; i != 0; --i) { |
| 605 | v[i] = v[i - 1]; |
| 606 | } |
| 607 | v[0] = value; |
| 608 | } |
| 609 | |
| 610 | static void MoveToFrontTransform(const uint32_t* BROTLI_RESTRICT v_in, |
| 611 | const size_t v_size, |
| 612 | uint32_t* v_out) { |
| 613 | size_t i; |
| 614 | uint8_t mtf[256]; |
| 615 | uint32_t max_value; |
| 616 | if (v_size == 0) { |
| 617 | return; |
| 618 | } |
| 619 | max_value = v_in[0]; |
| 620 | for (i = 1; i < v_size; ++i) { |
| 621 | if (v_in[i] > max_value) max_value = v_in[i]; |
| 622 | } |
| 623 | BROTLI_DCHECK(max_value < 256u); |
| 624 | for (i = 0; i <= max_value; ++i) { |
| 625 | mtf[i] = (uint8_t)i; |
| 626 | } |
| 627 | { |
| 628 | size_t mtf_size = max_value + 1; |
| 629 | for (i = 0; i < v_size; ++i) { |
| 630 | size_t index = IndexOf(mtf, mtf_size, (uint8_t)v_in[i]); |
| 631 | BROTLI_DCHECK(index < mtf_size); |
| 632 | v_out[i] = (uint32_t)index; |
| 633 | MoveToFront(mtf, index); |
| 634 | } |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | /* Finds runs of zeros in v[0..in_size) and replaces them with a prefix code of |
| 639 | the run length plus extra bits (lower 9 bits is the prefix code and the rest |
| 640 | are the extra bits). Non-zero values in v[] are shifted by |
| 641 | *max_length_prefix. Will not create prefix codes bigger than the initial |
| 642 | value of *max_run_length_prefix. The prefix code of run length L is simply |
| 643 | Log2Floor(L) and the number of extra bits is the same as the prefix code. */ |
| 644 | static void RunLengthCodeZeros(const size_t in_size, |
| 645 | uint32_t* BROTLI_RESTRICT v, size_t* BROTLI_RESTRICT out_size, |
| 646 | uint32_t* BROTLI_RESTRICT max_run_length_prefix) { |
| 647 | uint32_t max_reps = 0; |
| 648 | size_t i; |
| 649 | uint32_t max_prefix; |
| 650 | for (i = 0; i < in_size;) { |
| 651 | uint32_t reps = 0; |
| 652 | for (; i < in_size && v[i] != 0; ++i) ; |
| 653 | for (; i < in_size && v[i] == 0; ++i) { |
| 654 | ++reps; |
| 655 | } |
| 656 | max_reps = BROTLI_MAX(uint32_t, reps, max_reps); |
| 657 | } |
| 658 | max_prefix = max_reps > 0 ? Log2FloorNonZero(max_reps) : 0; |
| 659 | max_prefix = BROTLI_MIN(uint32_t, max_prefix, *max_run_length_prefix); |
| 660 | *max_run_length_prefix = max_prefix; |
| 661 | *out_size = 0; |
| 662 | for (i = 0; i < in_size;) { |
| 663 | BROTLI_DCHECK(*out_size <= i); |
| 664 | if (v[i] != 0) { |
| 665 | v[*out_size] = v[i] + *max_run_length_prefix; |
| 666 | ++i; |
| 667 | ++(*out_size); |
| 668 | } else { |
| 669 | uint32_t reps = 1; |
| 670 | size_t k; |
| 671 | for (k = i + 1; k < in_size && v[k] == 0; ++k) { |
| 672 | ++reps; |
| 673 | } |
| 674 | i += reps; |
| 675 | while (reps != 0) { |
| 676 | if (reps < (2u << max_prefix)) { |
| 677 | uint32_t run_length_prefix = Log2FloorNonZero(reps); |
| 678 | const uint32_t = reps - (1u << run_length_prefix); |
| 679 | v[*out_size] = run_length_prefix + (extra_bits << 9); |
| 680 | ++(*out_size); |
| 681 | break; |
| 682 | } else { |
| 683 | const uint32_t = (1u << max_prefix) - 1u; |
| 684 | v[*out_size] = max_prefix + (extra_bits << 9); |
| 685 | reps -= (2u << max_prefix) - 1u; |
| 686 | ++(*out_size); |
| 687 | } |
| 688 | } |
| 689 | } |
| 690 | } |
| 691 | } |
| 692 | |
| 693 | #define SYMBOL_BITS 9 |
| 694 | |
| 695 | static void EncodeContextMap(MemoryManager* m, |
| 696 | const uint32_t* context_map, |
| 697 | size_t context_map_size, |
| 698 | size_t num_clusters, |
| 699 | HuffmanTree* tree, |
| 700 | size_t* storage_ix, uint8_t* storage) { |
| 701 | size_t i; |
| 702 | uint32_t* rle_symbols; |
| 703 | uint32_t max_run_length_prefix = 6; |
| 704 | size_t num_rle_symbols = 0; |
| 705 | uint32_t histogram[BROTLI_MAX_CONTEXT_MAP_SYMBOLS]; |
| 706 | static const uint32_t kSymbolMask = (1u << SYMBOL_BITS) - 1u; |
| 707 | uint8_t depths[BROTLI_MAX_CONTEXT_MAP_SYMBOLS]; |
| 708 | uint16_t bits[BROTLI_MAX_CONTEXT_MAP_SYMBOLS]; |
| 709 | |
| 710 | StoreVarLenUint8(num_clusters - 1, storage_ix, storage); |
| 711 | |
| 712 | if (num_clusters == 1) { |
| 713 | return; |
| 714 | } |
| 715 | |
| 716 | rle_symbols = BROTLI_ALLOC(m, uint32_t, context_map_size); |
| 717 | if (BROTLI_IS_OOM(m)) return; |
| 718 | MoveToFrontTransform(context_map, context_map_size, rle_symbols); |
| 719 | RunLengthCodeZeros(context_map_size, rle_symbols, |
| 720 | &num_rle_symbols, &max_run_length_prefix); |
| 721 | memset(histogram, 0, sizeof(histogram)); |
| 722 | for (i = 0; i < num_rle_symbols; ++i) { |
| 723 | ++histogram[rle_symbols[i] & kSymbolMask]; |
| 724 | } |
| 725 | { |
| 726 | BROTLI_BOOL use_rle = TO_BROTLI_BOOL(max_run_length_prefix > 0); |
| 727 | BrotliWriteBits(1, (uint64_t)use_rle, storage_ix, storage); |
| 728 | if (use_rle) { |
| 729 | BrotliWriteBits(4, max_run_length_prefix - 1, storage_ix, storage); |
| 730 | } |
| 731 | } |
| 732 | BuildAndStoreHuffmanTree(histogram, num_clusters + max_run_length_prefix, |
| 733 | num_clusters + max_run_length_prefix, |
| 734 | tree, depths, bits, storage_ix, storage); |
| 735 | for (i = 0; i < num_rle_symbols; ++i) { |
| 736 | const uint32_t rle_symbol = rle_symbols[i] & kSymbolMask; |
| 737 | const uint32_t = rle_symbols[i] >> SYMBOL_BITS; |
| 738 | BrotliWriteBits(depths[rle_symbol], bits[rle_symbol], storage_ix, storage); |
| 739 | if (rle_symbol > 0 && rle_symbol <= max_run_length_prefix) { |
| 740 | BrotliWriteBits(rle_symbol, extra_bits_val, storage_ix, storage); |
| 741 | } |
| 742 | } |
| 743 | BrotliWriteBits(1, 1, storage_ix, storage); /* use move-to-front */ |
| 744 | BROTLI_FREE(m, rle_symbols); |
| 745 | } |
| 746 | |
| 747 | /* Stores the block switch command with index block_ix to the bit stream. */ |
| 748 | static BROTLI_INLINE void StoreBlockSwitch(BlockSplitCode* code, |
| 749 | const uint32_t block_len, |
| 750 | const uint8_t block_type, |
| 751 | BROTLI_BOOL is_first_block, |
| 752 | size_t* storage_ix, |
| 753 | uint8_t* storage) { |
| 754 | size_t typecode = NextBlockTypeCode(&code->type_code_calculator, block_type); |
| 755 | size_t lencode; |
| 756 | uint32_t ; |
| 757 | uint32_t ; |
| 758 | if (!is_first_block) { |
| 759 | BrotliWriteBits(code->type_depths[typecode], code->type_bits[typecode], |
| 760 | storage_ix, storage); |
| 761 | } |
| 762 | GetBlockLengthPrefixCode(block_len, &lencode, &len_nextra, &len_extra); |
| 763 | |
| 764 | BrotliWriteBits(code->length_depths[lencode], code->length_bits[lencode], |
| 765 | storage_ix, storage); |
| 766 | BrotliWriteBits(len_nextra, len_extra, storage_ix, storage); |
| 767 | } |
| 768 | |
| 769 | /* Builds a BlockSplitCode data structure from the block split given by the |
| 770 | vector of block types and block lengths and stores it to the bit stream. */ |
| 771 | static void BuildAndStoreBlockSplitCode(const uint8_t* types, |
| 772 | const uint32_t* lengths, |
| 773 | const size_t num_blocks, |
| 774 | const size_t num_types, |
| 775 | HuffmanTree* tree, |
| 776 | BlockSplitCode* code, |
| 777 | size_t* storage_ix, |
| 778 | uint8_t* storage) { |
| 779 | uint32_t type_histo[BROTLI_MAX_BLOCK_TYPE_SYMBOLS]; |
| 780 | uint32_t length_histo[BROTLI_NUM_BLOCK_LEN_SYMBOLS]; |
| 781 | size_t i; |
| 782 | BlockTypeCodeCalculator type_code_calculator; |
| 783 | memset(type_histo, 0, (num_types + 2) * sizeof(type_histo[0])); |
| 784 | memset(length_histo, 0, sizeof(length_histo)); |
| 785 | InitBlockTypeCodeCalculator(&type_code_calculator); |
| 786 | for (i = 0; i < num_blocks; ++i) { |
| 787 | size_t type_code = NextBlockTypeCode(&type_code_calculator, types[i]); |
| 788 | if (i != 0) ++type_histo[type_code]; |
| 789 | ++length_histo[BlockLengthPrefixCode(lengths[i])]; |
| 790 | } |
| 791 | StoreVarLenUint8(num_types - 1, storage_ix, storage); |
| 792 | if (num_types > 1) { /* TODO: else? could StoreBlockSwitch occur? */ |
| 793 | BuildAndStoreHuffmanTree(&type_histo[0], num_types + 2, num_types + 2, tree, |
| 794 | &code->type_depths[0], &code->type_bits[0], |
| 795 | storage_ix, storage); |
| 796 | BuildAndStoreHuffmanTree(&length_histo[0], BROTLI_NUM_BLOCK_LEN_SYMBOLS, |
| 797 | BROTLI_NUM_BLOCK_LEN_SYMBOLS, |
| 798 | tree, &code->length_depths[0], |
| 799 | &code->length_bits[0], storage_ix, storage); |
| 800 | StoreBlockSwitch(code, lengths[0], types[0], 1, storage_ix, storage); |
| 801 | } |
| 802 | } |
| 803 | |
| 804 | /* Stores a context map where the histogram type is always the block type. */ |
| 805 | static void StoreTrivialContextMap(size_t num_types, |
| 806 | size_t context_bits, |
| 807 | HuffmanTree* tree, |
| 808 | size_t* storage_ix, |
| 809 | uint8_t* storage) { |
| 810 | StoreVarLenUint8(num_types - 1, storage_ix, storage); |
| 811 | if (num_types > 1) { |
| 812 | size_t repeat_code = context_bits - 1u; |
| 813 | size_t repeat_bits = (1u << repeat_code) - 1u; |
| 814 | size_t alphabet_size = num_types + repeat_code; |
| 815 | uint32_t histogram[BROTLI_MAX_CONTEXT_MAP_SYMBOLS]; |
| 816 | uint8_t depths[BROTLI_MAX_CONTEXT_MAP_SYMBOLS]; |
| 817 | uint16_t bits[BROTLI_MAX_CONTEXT_MAP_SYMBOLS]; |
| 818 | size_t i; |
| 819 | memset(histogram, 0, alphabet_size * sizeof(histogram[0])); |
| 820 | /* Write RLEMAX. */ |
| 821 | BrotliWriteBits(1, 1, storage_ix, storage); |
| 822 | BrotliWriteBits(4, repeat_code - 1, storage_ix, storage); |
| 823 | histogram[repeat_code] = (uint32_t)num_types; |
| 824 | histogram[0] = 1; |
| 825 | for (i = context_bits; i < alphabet_size; ++i) { |
| 826 | histogram[i] = 1; |
| 827 | } |
| 828 | BuildAndStoreHuffmanTree(histogram, alphabet_size, alphabet_size, |
| 829 | tree, depths, bits, storage_ix, storage); |
| 830 | for (i = 0; i < num_types; ++i) { |
| 831 | size_t code = (i == 0 ? 0 : i + context_bits - 1); |
| 832 | BrotliWriteBits(depths[code], bits[code], storage_ix, storage); |
| 833 | BrotliWriteBits( |
| 834 | depths[repeat_code], bits[repeat_code], storage_ix, storage); |
| 835 | BrotliWriteBits(repeat_code, repeat_bits, storage_ix, storage); |
| 836 | } |
| 837 | /* Write IMTF (inverse-move-to-front) bit. */ |
| 838 | BrotliWriteBits(1, 1, storage_ix, storage); |
| 839 | } |
| 840 | } |
| 841 | |
| 842 | /* Manages the encoding of one block category (literal, command or distance). */ |
| 843 | typedef struct BlockEncoder { |
| 844 | size_t histogram_length_; |
| 845 | size_t num_block_types_; |
| 846 | const uint8_t* block_types_; /* Not owned. */ |
| 847 | const uint32_t* block_lengths_; /* Not owned. */ |
| 848 | size_t num_blocks_; |
| 849 | BlockSplitCode block_split_code_; |
| 850 | size_t block_ix_; |
| 851 | size_t block_len_; |
| 852 | size_t entropy_ix_; |
| 853 | uint8_t* depths_; |
| 854 | uint16_t* bits_; |
| 855 | } BlockEncoder; |
| 856 | |
| 857 | static void InitBlockEncoder(BlockEncoder* self, size_t histogram_length, |
| 858 | size_t num_block_types, const uint8_t* block_types, |
| 859 | const uint32_t* block_lengths, const size_t num_blocks) { |
| 860 | self->histogram_length_ = histogram_length; |
| 861 | self->num_block_types_ = num_block_types; |
| 862 | self->block_types_ = block_types; |
| 863 | self->block_lengths_ = block_lengths; |
| 864 | self->num_blocks_ = num_blocks; |
| 865 | InitBlockTypeCodeCalculator(&self->block_split_code_.type_code_calculator); |
| 866 | self->block_ix_ = 0; |
| 867 | self->block_len_ = num_blocks == 0 ? 0 : block_lengths[0]; |
| 868 | self->entropy_ix_ = 0; |
| 869 | self->depths_ = 0; |
| 870 | self->bits_ = 0; |
| 871 | } |
| 872 | |
| 873 | static void CleanupBlockEncoder(MemoryManager* m, BlockEncoder* self) { |
| 874 | BROTLI_FREE(m, self->depths_); |
| 875 | BROTLI_FREE(m, self->bits_); |
| 876 | } |
| 877 | |
| 878 | /* Creates entropy codes of block lengths and block types and stores them |
| 879 | to the bit stream. */ |
| 880 | static void BuildAndStoreBlockSwitchEntropyCodes(BlockEncoder* self, |
| 881 | HuffmanTree* tree, size_t* storage_ix, uint8_t* storage) { |
| 882 | BuildAndStoreBlockSplitCode(self->block_types_, self->block_lengths_, |
| 883 | self->num_blocks_, self->num_block_types_, tree, &self->block_split_code_, |
| 884 | storage_ix, storage); |
| 885 | } |
| 886 | |
| 887 | /* Stores the next symbol with the entropy code of the current block type. |
| 888 | Updates the block type and block length at block boundaries. */ |
| 889 | static void StoreSymbol(BlockEncoder* self, size_t symbol, size_t* storage_ix, |
| 890 | uint8_t* storage) { |
| 891 | if (self->block_len_ == 0) { |
| 892 | size_t block_ix = ++self->block_ix_; |
| 893 | uint32_t block_len = self->block_lengths_[block_ix]; |
| 894 | uint8_t block_type = self->block_types_[block_ix]; |
| 895 | self->block_len_ = block_len; |
| 896 | self->entropy_ix_ = block_type * self->histogram_length_; |
| 897 | StoreBlockSwitch(&self->block_split_code_, block_len, block_type, 0, |
| 898 | storage_ix, storage); |
| 899 | } |
| 900 | --self->block_len_; |
| 901 | { |
| 902 | size_t ix = self->entropy_ix_ + symbol; |
| 903 | BrotliWriteBits(self->depths_[ix], self->bits_[ix], storage_ix, storage); |
| 904 | } |
| 905 | } |
| 906 | |
| 907 | /* Stores the next symbol with the entropy code of the current block type and |
| 908 | context value. |
| 909 | Updates the block type and block length at block boundaries. */ |
| 910 | static void StoreSymbolWithContext(BlockEncoder* self, size_t symbol, |
| 911 | size_t context, const uint32_t* context_map, size_t* storage_ix, |
| 912 | uint8_t* storage, const size_t context_bits) { |
| 913 | if (self->block_len_ == 0) { |
| 914 | size_t block_ix = ++self->block_ix_; |
| 915 | uint32_t block_len = self->block_lengths_[block_ix]; |
| 916 | uint8_t block_type = self->block_types_[block_ix]; |
| 917 | self->block_len_ = block_len; |
| 918 | self->entropy_ix_ = (size_t)block_type << context_bits; |
| 919 | StoreBlockSwitch(&self->block_split_code_, block_len, block_type, 0, |
| 920 | storage_ix, storage); |
| 921 | } |
| 922 | --self->block_len_; |
| 923 | { |
| 924 | size_t histo_ix = context_map[self->entropy_ix_ + context]; |
| 925 | size_t ix = histo_ix * self->histogram_length_ + symbol; |
| 926 | BrotliWriteBits(self->depths_[ix], self->bits_[ix], storage_ix, storage); |
| 927 | } |
| 928 | } |
| 929 | |
| 930 | #define FN(X) X ## Literal |
| 931 | /* NOLINTNEXTLINE(build/include) */ |
| 932 | #include "./block_encoder_inc.h" |
| 933 | #undef FN |
| 934 | |
| 935 | #define FN(X) X ## Command |
| 936 | /* NOLINTNEXTLINE(build/include) */ |
| 937 | #include "./block_encoder_inc.h" |
| 938 | #undef FN |
| 939 | |
| 940 | #define FN(X) X ## Distance |
| 941 | /* NOLINTNEXTLINE(build/include) */ |
| 942 | #include "./block_encoder_inc.h" |
| 943 | #undef FN |
| 944 | |
| 945 | static void JumpToByteBoundary(size_t* storage_ix, uint8_t* storage) { |
| 946 | *storage_ix = (*storage_ix + 7u) & ~7u; |
| 947 | storage[*storage_ix >> 3] = 0; |
| 948 | } |
| 949 | |
| 950 | void BrotliStoreMetaBlock(MemoryManager* m, |
| 951 | const uint8_t* input, size_t start_pos, size_t length, size_t mask, |
| 952 | uint8_t prev_byte, uint8_t prev_byte2, BROTLI_BOOL is_last, |
| 953 | const BrotliEncoderParams* params, ContextType literal_context_mode, |
| 954 | const Command* commands, size_t n_commands, const MetaBlockSplit* mb, |
| 955 | size_t* storage_ix, uint8_t* storage) { |
| 956 | |
| 957 | size_t pos = start_pos; |
| 958 | size_t i; |
| 959 | uint32_t num_distance_symbols = params->dist.alphabet_size; |
| 960 | uint32_t num_effective_distance_symbols = num_distance_symbols; |
| 961 | HuffmanTree* tree; |
| 962 | ContextLut literal_context_lut = BROTLI_CONTEXT_LUT(literal_context_mode); |
| 963 | BlockEncoder literal_enc; |
| 964 | BlockEncoder command_enc; |
| 965 | BlockEncoder distance_enc; |
| 966 | const BrotliDistanceParams* dist = ¶ms->dist; |
| 967 | if (params->large_window && |
| 968 | num_effective_distance_symbols > BROTLI_NUM_HISTOGRAM_DISTANCE_SYMBOLS) { |
| 969 | num_effective_distance_symbols = BROTLI_NUM_HISTOGRAM_DISTANCE_SYMBOLS; |
| 970 | } |
| 971 | |
| 972 | StoreCompressedMetaBlockHeader(is_last, length, storage_ix, storage); |
| 973 | |
| 974 | tree = BROTLI_ALLOC(m, HuffmanTree, MAX_HUFFMAN_TREE_SIZE); |
| 975 | if (BROTLI_IS_OOM(m)) return; |
| 976 | InitBlockEncoder(&literal_enc, BROTLI_NUM_LITERAL_SYMBOLS, |
| 977 | mb->literal_split.num_types, mb->literal_split.types, |
| 978 | mb->literal_split.lengths, mb->literal_split.num_blocks); |
| 979 | InitBlockEncoder(&command_enc, BROTLI_NUM_COMMAND_SYMBOLS, |
| 980 | mb->command_split.num_types, mb->command_split.types, |
| 981 | mb->command_split.lengths, mb->command_split.num_blocks); |
| 982 | InitBlockEncoder(&distance_enc, num_effective_distance_symbols, |
| 983 | mb->distance_split.num_types, mb->distance_split.types, |
| 984 | mb->distance_split.lengths, mb->distance_split.num_blocks); |
| 985 | |
| 986 | BuildAndStoreBlockSwitchEntropyCodes(&literal_enc, tree, storage_ix, storage); |
| 987 | BuildAndStoreBlockSwitchEntropyCodes(&command_enc, tree, storage_ix, storage); |
| 988 | BuildAndStoreBlockSwitchEntropyCodes( |
| 989 | &distance_enc, tree, storage_ix, storage); |
| 990 | |
| 991 | BrotliWriteBits(2, dist->distance_postfix_bits, storage_ix, storage); |
| 992 | BrotliWriteBits( |
| 993 | 4, dist->num_direct_distance_codes >> dist->distance_postfix_bits, |
| 994 | storage_ix, storage); |
| 995 | for (i = 0; i < mb->literal_split.num_types; ++i) { |
| 996 | BrotliWriteBits(2, literal_context_mode, storage_ix, storage); |
| 997 | } |
| 998 | |
| 999 | if (mb->literal_context_map_size == 0) { |
| 1000 | StoreTrivialContextMap(mb->literal_histograms_size, |
| 1001 | BROTLI_LITERAL_CONTEXT_BITS, tree, storage_ix, storage); |
| 1002 | } else { |
| 1003 | EncodeContextMap(m, |
| 1004 | mb->literal_context_map, mb->literal_context_map_size, |
| 1005 | mb->literal_histograms_size, tree, storage_ix, storage); |
| 1006 | if (BROTLI_IS_OOM(m)) return; |
| 1007 | } |
| 1008 | |
| 1009 | if (mb->distance_context_map_size == 0) { |
| 1010 | StoreTrivialContextMap(mb->distance_histograms_size, |
| 1011 | BROTLI_DISTANCE_CONTEXT_BITS, tree, storage_ix, storage); |
| 1012 | } else { |
| 1013 | EncodeContextMap(m, |
| 1014 | mb->distance_context_map, mb->distance_context_map_size, |
| 1015 | mb->distance_histograms_size, tree, storage_ix, storage); |
| 1016 | if (BROTLI_IS_OOM(m)) return; |
| 1017 | } |
| 1018 | |
| 1019 | BuildAndStoreEntropyCodesLiteral(m, &literal_enc, mb->literal_histograms, |
| 1020 | mb->literal_histograms_size, BROTLI_NUM_LITERAL_SYMBOLS, tree, |
| 1021 | storage_ix, storage); |
| 1022 | if (BROTLI_IS_OOM(m)) return; |
| 1023 | BuildAndStoreEntropyCodesCommand(m, &command_enc, mb->command_histograms, |
| 1024 | mb->command_histograms_size, BROTLI_NUM_COMMAND_SYMBOLS, tree, |
| 1025 | storage_ix, storage); |
| 1026 | if (BROTLI_IS_OOM(m)) return; |
| 1027 | BuildAndStoreEntropyCodesDistance(m, &distance_enc, mb->distance_histograms, |
| 1028 | mb->distance_histograms_size, num_distance_symbols, tree, |
| 1029 | storage_ix, storage); |
| 1030 | if (BROTLI_IS_OOM(m)) return; |
| 1031 | BROTLI_FREE(m, tree); |
| 1032 | |
| 1033 | for (i = 0; i < n_commands; ++i) { |
| 1034 | const Command cmd = commands[i]; |
| 1035 | size_t cmd_code = cmd.cmd_prefix_; |
| 1036 | StoreSymbol(&command_enc, cmd_code, storage_ix, storage); |
| 1037 | StoreCommandExtra(&cmd, storage_ix, storage); |
| 1038 | if (mb->literal_context_map_size == 0) { |
| 1039 | size_t j; |
| 1040 | for (j = cmd.insert_len_; j != 0; --j) { |
| 1041 | StoreSymbol(&literal_enc, input[pos & mask], storage_ix, storage); |
| 1042 | ++pos; |
| 1043 | } |
| 1044 | } else { |
| 1045 | size_t j; |
| 1046 | for (j = cmd.insert_len_; j != 0; --j) { |
| 1047 | size_t context = |
| 1048 | BROTLI_CONTEXT(prev_byte, prev_byte2, literal_context_lut); |
| 1049 | uint8_t literal = input[pos & mask]; |
| 1050 | StoreSymbolWithContext(&literal_enc, literal, context, |
| 1051 | mb->literal_context_map, storage_ix, storage, |
| 1052 | BROTLI_LITERAL_CONTEXT_BITS); |
| 1053 | prev_byte2 = prev_byte; |
| 1054 | prev_byte = literal; |
| 1055 | ++pos; |
| 1056 | } |
| 1057 | } |
| 1058 | pos += CommandCopyLen(&cmd); |
| 1059 | if (CommandCopyLen(&cmd)) { |
| 1060 | prev_byte2 = input[(pos - 2) & mask]; |
| 1061 | prev_byte = input[(pos - 1) & mask]; |
| 1062 | if (cmd.cmd_prefix_ >= 128) { |
| 1063 | size_t dist_code = cmd.dist_prefix_ & 0x3FF; |
| 1064 | uint32_t = cmd.dist_prefix_ >> 10; |
| 1065 | uint64_t = cmd.dist_extra_; |
| 1066 | if (mb->distance_context_map_size == 0) { |
| 1067 | StoreSymbol(&distance_enc, dist_code, storage_ix, storage); |
| 1068 | } else { |
| 1069 | size_t context = CommandDistanceContext(&cmd); |
| 1070 | StoreSymbolWithContext(&distance_enc, dist_code, context, |
| 1071 | mb->distance_context_map, storage_ix, storage, |
| 1072 | BROTLI_DISTANCE_CONTEXT_BITS); |
| 1073 | } |
| 1074 | BrotliWriteBits(distnumextra, distextra, storage_ix, storage); |
| 1075 | } |
| 1076 | } |
| 1077 | } |
| 1078 | CleanupBlockEncoder(m, &distance_enc); |
| 1079 | CleanupBlockEncoder(m, &command_enc); |
| 1080 | CleanupBlockEncoder(m, &literal_enc); |
| 1081 | if (is_last) { |
| 1082 | JumpToByteBoundary(storage_ix, storage); |
| 1083 | } |
| 1084 | } |
| 1085 | |
| 1086 | static void BuildHistograms(const uint8_t* input, |
| 1087 | size_t start_pos, |
| 1088 | size_t mask, |
| 1089 | const Command* commands, |
| 1090 | size_t n_commands, |
| 1091 | HistogramLiteral* lit_histo, |
| 1092 | HistogramCommand* cmd_histo, |
| 1093 | HistogramDistance* dist_histo) { |
| 1094 | size_t pos = start_pos; |
| 1095 | size_t i; |
| 1096 | for (i = 0; i < n_commands; ++i) { |
| 1097 | const Command cmd = commands[i]; |
| 1098 | size_t j; |
| 1099 | HistogramAddCommand(cmd_histo, cmd.cmd_prefix_); |
| 1100 | for (j = cmd.insert_len_; j != 0; --j) { |
| 1101 | HistogramAddLiteral(lit_histo, input[pos & mask]); |
| 1102 | ++pos; |
| 1103 | } |
| 1104 | pos += CommandCopyLen(&cmd); |
| 1105 | if (CommandCopyLen(&cmd) && cmd.cmd_prefix_ >= 128) { |
| 1106 | HistogramAddDistance(dist_histo, cmd.dist_prefix_ & 0x3FF); |
| 1107 | } |
| 1108 | } |
| 1109 | } |
| 1110 | |
| 1111 | static void StoreDataWithHuffmanCodes(const uint8_t* input, |
| 1112 | size_t start_pos, |
| 1113 | size_t mask, |
| 1114 | const Command* commands, |
| 1115 | size_t n_commands, |
| 1116 | const uint8_t* lit_depth, |
| 1117 | const uint16_t* lit_bits, |
| 1118 | const uint8_t* cmd_depth, |
| 1119 | const uint16_t* cmd_bits, |
| 1120 | const uint8_t* dist_depth, |
| 1121 | const uint16_t* dist_bits, |
| 1122 | size_t* storage_ix, |
| 1123 | uint8_t* storage) { |
| 1124 | size_t pos = start_pos; |
| 1125 | size_t i; |
| 1126 | for (i = 0; i < n_commands; ++i) { |
| 1127 | const Command cmd = commands[i]; |
| 1128 | const size_t cmd_code = cmd.cmd_prefix_; |
| 1129 | size_t j; |
| 1130 | BrotliWriteBits( |
| 1131 | cmd_depth[cmd_code], cmd_bits[cmd_code], storage_ix, storage); |
| 1132 | StoreCommandExtra(&cmd, storage_ix, storage); |
| 1133 | for (j = cmd.insert_len_; j != 0; --j) { |
| 1134 | const uint8_t literal = input[pos & mask]; |
| 1135 | BrotliWriteBits( |
| 1136 | lit_depth[literal], lit_bits[literal], storage_ix, storage); |
| 1137 | ++pos; |
| 1138 | } |
| 1139 | pos += CommandCopyLen(&cmd); |
| 1140 | if (CommandCopyLen(&cmd) && cmd.cmd_prefix_ >= 128) { |
| 1141 | const size_t dist_code = cmd.dist_prefix_ & 0x3FF; |
| 1142 | const uint32_t = cmd.dist_prefix_ >> 10; |
| 1143 | const uint32_t = cmd.dist_extra_; |
| 1144 | BrotliWriteBits(dist_depth[dist_code], dist_bits[dist_code], |
| 1145 | storage_ix, storage); |
| 1146 | BrotliWriteBits(distnumextra, distextra, storage_ix, storage); |
| 1147 | } |
| 1148 | } |
| 1149 | } |
| 1150 | |
| 1151 | void BrotliStoreMetaBlockTrivial(MemoryManager* m, |
| 1152 | const uint8_t* input, size_t start_pos, size_t length, size_t mask, |
| 1153 | BROTLI_BOOL is_last, const BrotliEncoderParams* params, |
| 1154 | const Command* commands, size_t n_commands, |
| 1155 | size_t* storage_ix, uint8_t* storage) { |
| 1156 | HistogramLiteral lit_histo; |
| 1157 | HistogramCommand cmd_histo; |
| 1158 | HistogramDistance dist_histo; |
| 1159 | uint8_t lit_depth[BROTLI_NUM_LITERAL_SYMBOLS]; |
| 1160 | uint16_t lit_bits[BROTLI_NUM_LITERAL_SYMBOLS]; |
| 1161 | uint8_t cmd_depth[BROTLI_NUM_COMMAND_SYMBOLS]; |
| 1162 | uint16_t cmd_bits[BROTLI_NUM_COMMAND_SYMBOLS]; |
| 1163 | uint8_t dist_depth[MAX_SIMPLE_DISTANCE_ALPHABET_SIZE]; |
| 1164 | uint16_t dist_bits[MAX_SIMPLE_DISTANCE_ALPHABET_SIZE]; |
| 1165 | HuffmanTree* tree; |
| 1166 | uint32_t num_distance_symbols = params->dist.alphabet_size; |
| 1167 | |
| 1168 | StoreCompressedMetaBlockHeader(is_last, length, storage_ix, storage); |
| 1169 | |
| 1170 | HistogramClearLiteral(&lit_histo); |
| 1171 | HistogramClearCommand(&cmd_histo); |
| 1172 | HistogramClearDistance(&dist_histo); |
| 1173 | |
| 1174 | BuildHistograms(input, start_pos, mask, commands, n_commands, |
| 1175 | &lit_histo, &cmd_histo, &dist_histo); |
| 1176 | |
| 1177 | BrotliWriteBits(13, 0, storage_ix, storage); |
| 1178 | |
| 1179 | tree = BROTLI_ALLOC(m, HuffmanTree, MAX_HUFFMAN_TREE_SIZE); |
| 1180 | if (BROTLI_IS_OOM(m)) return; |
| 1181 | BuildAndStoreHuffmanTree(lit_histo.data_, BROTLI_NUM_LITERAL_SYMBOLS, |
| 1182 | BROTLI_NUM_LITERAL_SYMBOLS, tree, |
| 1183 | lit_depth, lit_bits, |
| 1184 | storage_ix, storage); |
| 1185 | BuildAndStoreHuffmanTree(cmd_histo.data_, BROTLI_NUM_COMMAND_SYMBOLS, |
| 1186 | BROTLI_NUM_COMMAND_SYMBOLS, tree, |
| 1187 | cmd_depth, cmd_bits, |
| 1188 | storage_ix, storage); |
| 1189 | BuildAndStoreHuffmanTree(dist_histo.data_, MAX_SIMPLE_DISTANCE_ALPHABET_SIZE, |
| 1190 | num_distance_symbols, tree, |
| 1191 | dist_depth, dist_bits, |
| 1192 | storage_ix, storage); |
| 1193 | BROTLI_FREE(m, tree); |
| 1194 | StoreDataWithHuffmanCodes(input, start_pos, mask, commands, |
| 1195 | n_commands, lit_depth, lit_bits, |
| 1196 | cmd_depth, cmd_bits, |
| 1197 | dist_depth, dist_bits, |
| 1198 | storage_ix, storage); |
| 1199 | if (is_last) { |
| 1200 | JumpToByteBoundary(storage_ix, storage); |
| 1201 | } |
| 1202 | } |
| 1203 | |
| 1204 | void BrotliStoreMetaBlockFast(MemoryManager* m, |
| 1205 | const uint8_t* input, size_t start_pos, size_t length, size_t mask, |
| 1206 | BROTLI_BOOL is_last, const BrotliEncoderParams* params, |
| 1207 | const Command* commands, size_t n_commands, |
| 1208 | size_t* storage_ix, uint8_t* storage) { |
| 1209 | uint32_t num_distance_symbols = params->dist.alphabet_size; |
| 1210 | uint32_t distance_alphabet_bits = |
| 1211 | Log2FloorNonZero(num_distance_symbols - 1) + 1; |
| 1212 | |
| 1213 | StoreCompressedMetaBlockHeader(is_last, length, storage_ix, storage); |
| 1214 | |
| 1215 | BrotliWriteBits(13, 0, storage_ix, storage); |
| 1216 | |
| 1217 | if (n_commands <= 128) { |
| 1218 | uint32_t histogram[BROTLI_NUM_LITERAL_SYMBOLS] = { 0 }; |
| 1219 | size_t pos = start_pos; |
| 1220 | size_t num_literals = 0; |
| 1221 | size_t i; |
| 1222 | uint8_t lit_depth[BROTLI_NUM_LITERAL_SYMBOLS]; |
| 1223 | uint16_t lit_bits[BROTLI_NUM_LITERAL_SYMBOLS]; |
| 1224 | for (i = 0; i < n_commands; ++i) { |
| 1225 | const Command cmd = commands[i]; |
| 1226 | size_t j; |
| 1227 | for (j = cmd.insert_len_; j != 0; --j) { |
| 1228 | ++histogram[input[pos & mask]]; |
| 1229 | ++pos; |
| 1230 | } |
| 1231 | num_literals += cmd.insert_len_; |
| 1232 | pos += CommandCopyLen(&cmd); |
| 1233 | } |
| 1234 | BrotliBuildAndStoreHuffmanTreeFast(m, histogram, num_literals, |
| 1235 | /* max_bits = */ 8, |
| 1236 | lit_depth, lit_bits, |
| 1237 | storage_ix, storage); |
| 1238 | if (BROTLI_IS_OOM(m)) return; |
| 1239 | StoreStaticCommandHuffmanTree(storage_ix, storage); |
| 1240 | StoreStaticDistanceHuffmanTree(storage_ix, storage); |
| 1241 | StoreDataWithHuffmanCodes(input, start_pos, mask, commands, |
| 1242 | n_commands, lit_depth, lit_bits, |
| 1243 | kStaticCommandCodeDepth, |
| 1244 | kStaticCommandCodeBits, |
| 1245 | kStaticDistanceCodeDepth, |
| 1246 | kStaticDistanceCodeBits, |
| 1247 | storage_ix, storage); |
| 1248 | } else { |
| 1249 | HistogramLiteral lit_histo; |
| 1250 | HistogramCommand cmd_histo; |
| 1251 | HistogramDistance dist_histo; |
| 1252 | uint8_t lit_depth[BROTLI_NUM_LITERAL_SYMBOLS]; |
| 1253 | uint16_t lit_bits[BROTLI_NUM_LITERAL_SYMBOLS]; |
| 1254 | uint8_t cmd_depth[BROTLI_NUM_COMMAND_SYMBOLS]; |
| 1255 | uint16_t cmd_bits[BROTLI_NUM_COMMAND_SYMBOLS]; |
| 1256 | uint8_t dist_depth[MAX_SIMPLE_DISTANCE_ALPHABET_SIZE]; |
| 1257 | uint16_t dist_bits[MAX_SIMPLE_DISTANCE_ALPHABET_SIZE]; |
| 1258 | HistogramClearLiteral(&lit_histo); |
| 1259 | HistogramClearCommand(&cmd_histo); |
| 1260 | HistogramClearDistance(&dist_histo); |
| 1261 | BuildHistograms(input, start_pos, mask, commands, n_commands, |
| 1262 | &lit_histo, &cmd_histo, &dist_histo); |
| 1263 | BrotliBuildAndStoreHuffmanTreeFast(m, lit_histo.data_, |
| 1264 | lit_histo.total_count_, |
| 1265 | /* max_bits = */ 8, |
| 1266 | lit_depth, lit_bits, |
| 1267 | storage_ix, storage); |
| 1268 | if (BROTLI_IS_OOM(m)) return; |
| 1269 | BrotliBuildAndStoreHuffmanTreeFast(m, cmd_histo.data_, |
| 1270 | cmd_histo.total_count_, |
| 1271 | /* max_bits = */ 10, |
| 1272 | cmd_depth, cmd_bits, |
| 1273 | storage_ix, storage); |
| 1274 | if (BROTLI_IS_OOM(m)) return; |
| 1275 | BrotliBuildAndStoreHuffmanTreeFast(m, dist_histo.data_, |
| 1276 | dist_histo.total_count_, |
| 1277 | /* max_bits = */ |
| 1278 | distance_alphabet_bits, |
| 1279 | dist_depth, dist_bits, |
| 1280 | storage_ix, storage); |
| 1281 | if (BROTLI_IS_OOM(m)) return; |
| 1282 | StoreDataWithHuffmanCodes(input, start_pos, mask, commands, |
| 1283 | n_commands, lit_depth, lit_bits, |
| 1284 | cmd_depth, cmd_bits, |
| 1285 | dist_depth, dist_bits, |
| 1286 | storage_ix, storage); |
| 1287 | } |
| 1288 | |
| 1289 | if (is_last) { |
| 1290 | JumpToByteBoundary(storage_ix, storage); |
| 1291 | } |
| 1292 | } |
| 1293 | |
| 1294 | /* This is for storing uncompressed blocks (simple raw storage of |
| 1295 | bytes-as-bytes). */ |
| 1296 | void BrotliStoreUncompressedMetaBlock(BROTLI_BOOL is_final_block, |
| 1297 | const uint8_t* BROTLI_RESTRICT input, |
| 1298 | size_t position, size_t mask, |
| 1299 | size_t len, |
| 1300 | size_t* BROTLI_RESTRICT storage_ix, |
| 1301 | uint8_t* BROTLI_RESTRICT storage) { |
| 1302 | size_t masked_pos = position & mask; |
| 1303 | BrotliStoreUncompressedMetaBlockHeader(len, storage_ix, storage); |
| 1304 | JumpToByteBoundary(storage_ix, storage); |
| 1305 | |
| 1306 | if (masked_pos + len > mask + 1) { |
| 1307 | size_t len1 = mask + 1 - masked_pos; |
| 1308 | memcpy(&storage[*storage_ix >> 3], &input[masked_pos], len1); |
| 1309 | *storage_ix += len1 << 3; |
| 1310 | len -= len1; |
| 1311 | masked_pos = 0; |
| 1312 | } |
| 1313 | memcpy(&storage[*storage_ix >> 3], &input[masked_pos], len); |
| 1314 | *storage_ix += len << 3; |
| 1315 | |
| 1316 | /* We need to clear the next 4 bytes to continue to be |
| 1317 | compatible with BrotliWriteBits. */ |
| 1318 | BrotliWriteBitsPrepareStorage(*storage_ix, storage); |
| 1319 | |
| 1320 | /* Since the uncompressed block itself may not be the final block, add an |
| 1321 | empty one after this. */ |
| 1322 | if (is_final_block) { |
| 1323 | BrotliWriteBits(1, 1, storage_ix, storage); /* islast */ |
| 1324 | BrotliWriteBits(1, 1, storage_ix, storage); /* isempty */ |
| 1325 | JumpToByteBoundary(storage_ix, storage); |
| 1326 | } |
| 1327 | } |
| 1328 | |
| 1329 | #if defined(__cplusplus) || defined(c_plusplus) |
| 1330 | } /* extern "C" */ |
| 1331 | #endif |
| 1332 | |