| 1 | // Copyright 2012 Google Inc. All Rights Reserved. |
| 2 | // |
| 3 | // Use of this source code is governed by a BSD-style license |
| 4 | // that can be found in the COPYING file in the root of the source |
| 5 | // tree. An additional intellectual property rights grant can be found |
| 6 | // in the file PATENTS. All contributing project authors may |
| 7 | // be found in the AUTHORS file in the root of the source tree. |
| 8 | // ----------------------------------------------------------------------------- |
| 9 | // |
| 10 | // main entry for the decoder |
| 11 | // |
| 12 | // Authors: Vikas Arora (vikaas.arora@gmail.com) |
| 13 | // Jyrki Alakuijala (jyrki@google.com) |
| 14 | |
| 15 | #include <stdlib.h> |
| 16 | |
| 17 | #include "./alphai_dec.h" |
| 18 | #include "./vp8li_dec.h" |
| 19 | #include "../dsp/dsp.h" |
| 20 | #include "../dsp/lossless.h" |
| 21 | #include "../dsp/lossless_common.h" |
| 22 | #include "../dsp/yuv.h" |
| 23 | #include "../utils/endian_inl_utils.h" |
| 24 | #include "../utils/huffman_utils.h" |
| 25 | #include "../utils/utils.h" |
| 26 | |
| 27 | #define NUM_ARGB_CACHE_ROWS 16 |
| 28 | |
| 29 | static const int kCodeLengthLiterals = 16; |
| 30 | static const int kCodeLengthRepeatCode = 16; |
| 31 | static const int [3] = { 2, 3, 7 }; |
| 32 | static const int kCodeLengthRepeatOffsets[3] = { 3, 3, 11 }; |
| 33 | |
| 34 | // ----------------------------------------------------------------------------- |
| 35 | // Five Huffman codes are used at each meta code: |
| 36 | // 1. green + length prefix codes + color cache codes, |
| 37 | // 2. alpha, |
| 38 | // 3. red, |
| 39 | // 4. blue, and, |
| 40 | // 5. distance prefix codes. |
| 41 | typedef enum { |
| 42 | GREEN = 0, |
| 43 | RED = 1, |
| 44 | BLUE = 2, |
| 45 | ALPHA = 3, |
| 46 | DIST = 4 |
| 47 | } HuffIndex; |
| 48 | |
| 49 | static const uint16_t kAlphabetSize[HUFFMAN_CODES_PER_META_CODE] = { |
| 50 | NUM_LITERAL_CODES + NUM_LENGTH_CODES, |
| 51 | NUM_LITERAL_CODES, NUM_LITERAL_CODES, NUM_LITERAL_CODES, |
| 52 | NUM_DISTANCE_CODES |
| 53 | }; |
| 54 | |
| 55 | static const uint8_t kLiteralMap[HUFFMAN_CODES_PER_META_CODE] = { |
| 56 | 0, 1, 1, 1, 0 |
| 57 | }; |
| 58 | |
| 59 | #define NUM_CODE_LENGTH_CODES 19 |
| 60 | static const uint8_t kCodeLengthCodeOrder[NUM_CODE_LENGTH_CODES] = { |
| 61 | 17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 |
| 62 | }; |
| 63 | |
| 64 | #define CODE_TO_PLANE_CODES 120 |
| 65 | static const uint8_t kCodeToPlane[CODE_TO_PLANE_CODES] = { |
| 66 | 0x18, 0x07, 0x17, 0x19, 0x28, 0x06, 0x27, 0x29, 0x16, 0x1a, |
| 67 | 0x26, 0x2a, 0x38, 0x05, 0x37, 0x39, 0x15, 0x1b, 0x36, 0x3a, |
| 68 | 0x25, 0x2b, 0x48, 0x04, 0x47, 0x49, 0x14, 0x1c, 0x35, 0x3b, |
| 69 | 0x46, 0x4a, 0x24, 0x2c, 0x58, 0x45, 0x4b, 0x34, 0x3c, 0x03, |
| 70 | 0x57, 0x59, 0x13, 0x1d, 0x56, 0x5a, 0x23, 0x2d, 0x44, 0x4c, |
| 71 | 0x55, 0x5b, 0x33, 0x3d, 0x68, 0x02, 0x67, 0x69, 0x12, 0x1e, |
| 72 | 0x66, 0x6a, 0x22, 0x2e, 0x54, 0x5c, 0x43, 0x4d, 0x65, 0x6b, |
| 73 | 0x32, 0x3e, 0x78, 0x01, 0x77, 0x79, 0x53, 0x5d, 0x11, 0x1f, |
| 74 | 0x64, 0x6c, 0x42, 0x4e, 0x76, 0x7a, 0x21, 0x2f, 0x75, 0x7b, |
| 75 | 0x31, 0x3f, 0x63, 0x6d, 0x52, 0x5e, 0x00, 0x74, 0x7c, 0x41, |
| 76 | 0x4f, 0x10, 0x20, 0x62, 0x6e, 0x30, 0x73, 0x7d, 0x51, 0x5f, |
| 77 | 0x40, 0x72, 0x7e, 0x61, 0x6f, 0x50, 0x71, 0x7f, 0x60, 0x70 |
| 78 | }; |
| 79 | |
| 80 | // Memory needed for lookup tables of one Huffman tree group. Red, blue, alpha |
| 81 | // and distance alphabets are constant (256 for red, blue and alpha, 40 for |
| 82 | // distance) and lookup table sizes for them in worst case are 630 and 410 |
| 83 | // respectively. Size of green alphabet depends on color cache size and is equal |
| 84 | // to 256 (green component values) + 24 (length prefix values) |
| 85 | // + color_cache_size (between 0 and 2048). |
| 86 | // All values computed for 8-bit first level lookup with Mark Adler's tool: |
| 87 | // http://www.hdfgroup.org/ftp/lib-external/zlib/zlib-1.2.5/examples/enough.c |
| 88 | #define FIXED_TABLE_SIZE (630 * 3 + 410) |
| 89 | static const int kTableSize[12] = { |
| 90 | FIXED_TABLE_SIZE + 654, |
| 91 | FIXED_TABLE_SIZE + 656, |
| 92 | FIXED_TABLE_SIZE + 658, |
| 93 | FIXED_TABLE_SIZE + 662, |
| 94 | FIXED_TABLE_SIZE + 670, |
| 95 | FIXED_TABLE_SIZE + 686, |
| 96 | FIXED_TABLE_SIZE + 718, |
| 97 | FIXED_TABLE_SIZE + 782, |
| 98 | FIXED_TABLE_SIZE + 912, |
| 99 | FIXED_TABLE_SIZE + 1168, |
| 100 | FIXED_TABLE_SIZE + 1680, |
| 101 | FIXED_TABLE_SIZE + 2704 |
| 102 | }; |
| 103 | |
| 104 | static int DecodeImageStream(int xsize, int ysize, |
| 105 | int is_level0, |
| 106 | VP8LDecoder* const dec, |
| 107 | uint32_t** const decoded_data); |
| 108 | |
| 109 | //------------------------------------------------------------------------------ |
| 110 | |
| 111 | int VP8LCheckSignature(const uint8_t* const data, size_t size) { |
| 112 | return (size >= VP8L_FRAME_HEADER_SIZE && |
| 113 | data[0] == VP8L_MAGIC_BYTE && |
| 114 | (data[4] >> 5) == 0); // version |
| 115 | } |
| 116 | |
| 117 | static int ReadImageInfo(VP8LBitReader* const br, |
| 118 | int* const width, int* const height, |
| 119 | int* const has_alpha) { |
| 120 | if (VP8LReadBits(br, 8) != VP8L_MAGIC_BYTE) return 0; |
| 121 | *width = VP8LReadBits(br, VP8L_IMAGE_SIZE_BITS) + 1; |
| 122 | *height = VP8LReadBits(br, VP8L_IMAGE_SIZE_BITS) + 1; |
| 123 | *has_alpha = VP8LReadBits(br, 1); |
| 124 | if (VP8LReadBits(br, VP8L_VERSION_BITS) != 0) return 0; |
| 125 | return !br->eos_; |
| 126 | } |
| 127 | |
| 128 | int VP8LGetInfo(const uint8_t* data, size_t data_size, |
| 129 | int* const width, int* const height, int* const has_alpha) { |
| 130 | if (data == NULL || data_size < VP8L_FRAME_HEADER_SIZE) { |
| 131 | return 0; // not enough data |
| 132 | } else if (!VP8LCheckSignature(data, data_size)) { |
| 133 | return 0; // bad signature |
| 134 | } else { |
| 135 | int w, h, a; |
| 136 | VP8LBitReader br; |
| 137 | VP8LInitBitReader(&br, data, data_size); |
| 138 | if (!ReadImageInfo(&br, &w, &h, &a)) { |
| 139 | return 0; |
| 140 | } |
| 141 | if (width != NULL) *width = w; |
| 142 | if (height != NULL) *height = h; |
| 143 | if (has_alpha != NULL) *has_alpha = a; |
| 144 | return 1; |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | //------------------------------------------------------------------------------ |
| 149 | |
| 150 | static WEBP_INLINE int GetCopyDistance(int distance_symbol, |
| 151 | VP8LBitReader* const br) { |
| 152 | int , offset; |
| 153 | if (distance_symbol < 4) { |
| 154 | return distance_symbol + 1; |
| 155 | } |
| 156 | extra_bits = (distance_symbol - 2) >> 1; |
| 157 | offset = (2 + (distance_symbol & 1)) << extra_bits; |
| 158 | return offset + VP8LReadBits(br, extra_bits) + 1; |
| 159 | } |
| 160 | |
| 161 | static WEBP_INLINE int GetCopyLength(int length_symbol, |
| 162 | VP8LBitReader* const br) { |
| 163 | // Length and distance prefixes are encoded the same way. |
| 164 | return GetCopyDistance(length_symbol, br); |
| 165 | } |
| 166 | |
| 167 | static WEBP_INLINE int PlaneCodeToDistance(int xsize, int plane_code) { |
| 168 | if (plane_code > CODE_TO_PLANE_CODES) { |
| 169 | return plane_code - CODE_TO_PLANE_CODES; |
| 170 | } else { |
| 171 | const int dist_code = kCodeToPlane[plane_code - 1]; |
| 172 | const int yoffset = dist_code >> 4; |
| 173 | const int xoffset = 8 - (dist_code & 0xf); |
| 174 | const int dist = yoffset * xsize + xoffset; |
| 175 | return (dist >= 1) ? dist : 1; // dist<1 can happen if xsize is very small |
| 176 | } |
| 177 | } |
| 178 | |
| 179 | //------------------------------------------------------------------------------ |
| 180 | // Decodes the next Huffman code from bit-stream. |
| 181 | // FillBitWindow(br) needs to be called at minimum every second call |
| 182 | // to ReadSymbol, in order to pre-fetch enough bits. |
| 183 | static WEBP_INLINE int ReadSymbol(const HuffmanCode* table, |
| 184 | VP8LBitReader* const br) { |
| 185 | int nbits; |
| 186 | uint32_t val = VP8LPrefetchBits(br); |
| 187 | table += val & HUFFMAN_TABLE_MASK; |
| 188 | nbits = table->bits - HUFFMAN_TABLE_BITS; |
| 189 | if (nbits > 0) { |
| 190 | VP8LSetBitPos(br, br->bit_pos_ + HUFFMAN_TABLE_BITS); |
| 191 | val = VP8LPrefetchBits(br); |
| 192 | table += table->value; |
| 193 | table += val & ((1 << nbits) - 1); |
| 194 | } |
| 195 | VP8LSetBitPos(br, br->bit_pos_ + table->bits); |
| 196 | return table->value; |
| 197 | } |
| 198 | |
| 199 | // Reads packed symbol depending on GREEN channel |
| 200 | #define BITS_SPECIAL_MARKER 0x100 // something large enough (and a bit-mask) |
| 201 | #define PACKED_NON_LITERAL_CODE 0 // must be < NUM_LITERAL_CODES |
| 202 | static WEBP_INLINE int ReadPackedSymbols(const HTreeGroup* group, |
| 203 | VP8LBitReader* const br, |
| 204 | uint32_t* const dst) { |
| 205 | const uint32_t val = VP8LPrefetchBits(br) & (HUFFMAN_PACKED_TABLE_SIZE - 1); |
| 206 | const HuffmanCode32 code = group->packed_table[val]; |
| 207 | assert(group->use_packed_table); |
| 208 | if (code.bits < BITS_SPECIAL_MARKER) { |
| 209 | VP8LSetBitPos(br, br->bit_pos_ + code.bits); |
| 210 | *dst = code.value; |
| 211 | return PACKED_NON_LITERAL_CODE; |
| 212 | } else { |
| 213 | VP8LSetBitPos(br, br->bit_pos_ + code.bits - BITS_SPECIAL_MARKER); |
| 214 | assert(code.value >= NUM_LITERAL_CODES); |
| 215 | return code.value; |
| 216 | } |
| 217 | } |
| 218 | |
| 219 | static int AccumulateHCode(HuffmanCode hcode, int shift, |
| 220 | HuffmanCode32* const huff) { |
| 221 | huff->bits += hcode.bits; |
| 222 | huff->value |= (uint32_t)hcode.value << shift; |
| 223 | assert(huff->bits <= HUFFMAN_TABLE_BITS); |
| 224 | return hcode.bits; |
| 225 | } |
| 226 | |
| 227 | static void BuildPackedTable(HTreeGroup* const htree_group) { |
| 228 | uint32_t code; |
| 229 | for (code = 0; code < HUFFMAN_PACKED_TABLE_SIZE; ++code) { |
| 230 | uint32_t bits = code; |
| 231 | HuffmanCode32* const huff = &htree_group->packed_table[bits]; |
| 232 | HuffmanCode hcode = htree_group->htrees[GREEN][bits]; |
| 233 | if (hcode.value >= NUM_LITERAL_CODES) { |
| 234 | huff->bits = hcode.bits + BITS_SPECIAL_MARKER; |
| 235 | huff->value = hcode.value; |
| 236 | } else { |
| 237 | huff->bits = 0; |
| 238 | huff->value = 0; |
| 239 | bits >>= AccumulateHCode(hcode, 8, huff); |
| 240 | bits >>= AccumulateHCode(htree_group->htrees[RED][bits], 16, huff); |
| 241 | bits >>= AccumulateHCode(htree_group->htrees[BLUE][bits], 0, huff); |
| 242 | bits >>= AccumulateHCode(htree_group->htrees[ALPHA][bits], 24, huff); |
| 243 | (void)bits; |
| 244 | } |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | static int ReadHuffmanCodeLengths( |
| 249 | VP8LDecoder* const dec, const int* const code_length_code_lengths, |
| 250 | int num_symbols, int* const code_lengths) { |
| 251 | int ok = 0; |
| 252 | VP8LBitReader* const br = &dec->br_; |
| 253 | int symbol; |
| 254 | int max_symbol; |
| 255 | int prev_code_len = DEFAULT_CODE_LENGTH; |
| 256 | HuffmanCode table[1 << LENGTHS_TABLE_BITS]; |
| 257 | |
| 258 | if (!VP8LBuildHuffmanTable(table, LENGTHS_TABLE_BITS, |
| 259 | code_length_code_lengths, |
| 260 | NUM_CODE_LENGTH_CODES)) { |
| 261 | goto End; |
| 262 | } |
| 263 | |
| 264 | if (VP8LReadBits(br, 1)) { // use length |
| 265 | const int length_nbits = 2 + 2 * VP8LReadBits(br, 3); |
| 266 | max_symbol = 2 + VP8LReadBits(br, length_nbits); |
| 267 | if (max_symbol > num_symbols) { |
| 268 | goto End; |
| 269 | } |
| 270 | } else { |
| 271 | max_symbol = num_symbols; |
| 272 | } |
| 273 | |
| 274 | symbol = 0; |
| 275 | while (symbol < num_symbols) { |
| 276 | const HuffmanCode* p; |
| 277 | int code_len; |
| 278 | if (max_symbol-- == 0) break; |
| 279 | VP8LFillBitWindow(br); |
| 280 | p = &table[VP8LPrefetchBits(br) & LENGTHS_TABLE_MASK]; |
| 281 | VP8LSetBitPos(br, br->bit_pos_ + p->bits); |
| 282 | code_len = p->value; |
| 283 | if (code_len < kCodeLengthLiterals) { |
| 284 | code_lengths[symbol++] = code_len; |
| 285 | if (code_len != 0) prev_code_len = code_len; |
| 286 | } else { |
| 287 | const int use_prev = (code_len == kCodeLengthRepeatCode); |
| 288 | const int slot = code_len - kCodeLengthLiterals; |
| 289 | const int = kCodeLengthExtraBits[slot]; |
| 290 | const int repeat_offset = kCodeLengthRepeatOffsets[slot]; |
| 291 | int repeat = VP8LReadBits(br, extra_bits) + repeat_offset; |
| 292 | if (symbol + repeat > num_symbols) { |
| 293 | goto End; |
| 294 | } else { |
| 295 | const int length = use_prev ? prev_code_len : 0; |
| 296 | while (repeat-- > 0) code_lengths[symbol++] = length; |
| 297 | } |
| 298 | } |
| 299 | } |
| 300 | ok = 1; |
| 301 | |
| 302 | End: |
| 303 | if (!ok) dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
| 304 | return ok; |
| 305 | } |
| 306 | |
| 307 | // 'code_lengths' is pre-allocated temporary buffer, used for creating Huffman |
| 308 | // tree. |
| 309 | static int ReadHuffmanCode(int alphabet_size, VP8LDecoder* const dec, |
| 310 | int* const code_lengths, HuffmanCode* const table) { |
| 311 | int ok = 0; |
| 312 | int size = 0; |
| 313 | VP8LBitReader* const br = &dec->br_; |
| 314 | const int simple_code = VP8LReadBits(br, 1); |
| 315 | |
| 316 | memset(code_lengths, 0, alphabet_size * sizeof(*code_lengths)); |
| 317 | |
| 318 | if (simple_code) { // Read symbols, codes & code lengths directly. |
| 319 | const int num_symbols = VP8LReadBits(br, 1) + 1; |
| 320 | const int first_symbol_len_code = VP8LReadBits(br, 1); |
| 321 | // The first code is either 1 bit or 8 bit code. |
| 322 | int symbol = VP8LReadBits(br, (first_symbol_len_code == 0) ? 1 : 8); |
| 323 | code_lengths[symbol] = 1; |
| 324 | // The second code (if present), is always 8 bit long. |
| 325 | if (num_symbols == 2) { |
| 326 | symbol = VP8LReadBits(br, 8); |
| 327 | code_lengths[symbol] = 1; |
| 328 | } |
| 329 | ok = 1; |
| 330 | } else { // Decode Huffman-coded code lengths. |
| 331 | int i; |
| 332 | int code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 }; |
| 333 | const int num_codes = VP8LReadBits(br, 4) + 4; |
| 334 | if (num_codes > NUM_CODE_LENGTH_CODES) { |
| 335 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
| 336 | return 0; |
| 337 | } |
| 338 | |
| 339 | for (i = 0; i < num_codes; ++i) { |
| 340 | code_length_code_lengths[kCodeLengthCodeOrder[i]] = VP8LReadBits(br, 3); |
| 341 | } |
| 342 | ok = ReadHuffmanCodeLengths(dec, code_length_code_lengths, alphabet_size, |
| 343 | code_lengths); |
| 344 | } |
| 345 | |
| 346 | ok = ok && !br->eos_; |
| 347 | if (ok) { |
| 348 | size = VP8LBuildHuffmanTable(table, HUFFMAN_TABLE_BITS, |
| 349 | code_lengths, alphabet_size); |
| 350 | } |
| 351 | if (!ok || size == 0) { |
| 352 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
| 353 | return 0; |
| 354 | } |
| 355 | return size; |
| 356 | } |
| 357 | |
| 358 | static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize, |
| 359 | int color_cache_bits, int allow_recursion) { |
| 360 | int i, j; |
| 361 | VP8LBitReader* const br = &dec->br_; |
| 362 | VP8LMetadata* const hdr = &dec->hdr_; |
| 363 | uint32_t* huffman_image = NULL; |
| 364 | HTreeGroup* htree_groups = NULL; |
| 365 | HuffmanCode* huffman_tables = NULL; |
| 366 | HuffmanCode* next = NULL; |
| 367 | int num_htree_groups = 1; |
| 368 | int max_alphabet_size = 0; |
| 369 | int* code_lengths = NULL; |
| 370 | const int table_size = kTableSize[color_cache_bits]; |
| 371 | |
| 372 | if (allow_recursion && VP8LReadBits(br, 1)) { |
| 373 | // use meta Huffman codes. |
| 374 | const int huffman_precision = VP8LReadBits(br, 3) + 2; |
| 375 | const int huffman_xsize = VP8LSubSampleSize(xsize, huffman_precision); |
| 376 | const int huffman_ysize = VP8LSubSampleSize(ysize, huffman_precision); |
| 377 | const int huffman_pixs = huffman_xsize * huffman_ysize; |
| 378 | if (!DecodeImageStream(huffman_xsize, huffman_ysize, 0, dec, |
| 379 | &huffman_image)) { |
| 380 | goto Error; |
| 381 | } |
| 382 | hdr->huffman_subsample_bits_ = huffman_precision; |
| 383 | for (i = 0; i < huffman_pixs; ++i) { |
| 384 | // The huffman data is stored in red and green bytes. |
| 385 | const int group = (huffman_image[i] >> 8) & 0xffff; |
| 386 | huffman_image[i] = group; |
| 387 | if (group >= num_htree_groups) { |
| 388 | num_htree_groups = group + 1; |
| 389 | } |
| 390 | } |
| 391 | } |
| 392 | |
| 393 | if (br->eos_) goto Error; |
| 394 | |
| 395 | // Find maximum alphabet size for the htree group. |
| 396 | for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) { |
| 397 | int alphabet_size = kAlphabetSize[j]; |
| 398 | if (j == 0 && color_cache_bits > 0) { |
| 399 | alphabet_size += 1 << color_cache_bits; |
| 400 | } |
| 401 | if (max_alphabet_size < alphabet_size) { |
| 402 | max_alphabet_size = alphabet_size; |
| 403 | } |
| 404 | } |
| 405 | |
| 406 | huffman_tables = (HuffmanCode*)WebPSafeMalloc(num_htree_groups * table_size, |
| 407 | sizeof(*huffman_tables)); |
| 408 | htree_groups = VP8LHtreeGroupsNew(num_htree_groups); |
| 409 | code_lengths = (int*)WebPSafeCalloc((uint64_t)max_alphabet_size, |
| 410 | sizeof(*code_lengths)); |
| 411 | |
| 412 | if (htree_groups == NULL || code_lengths == NULL || huffman_tables == NULL) { |
| 413 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
| 414 | goto Error; |
| 415 | } |
| 416 | |
| 417 | next = huffman_tables; |
| 418 | for (i = 0; i < num_htree_groups; ++i) { |
| 419 | HTreeGroup* const htree_group = &htree_groups[i]; |
| 420 | HuffmanCode** const htrees = htree_group->htrees; |
| 421 | int size; |
| 422 | int total_size = 0; |
| 423 | int is_trivial_literal = 1; |
| 424 | int max_bits = 0; |
| 425 | for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) { |
| 426 | int alphabet_size = kAlphabetSize[j]; |
| 427 | htrees[j] = next; |
| 428 | if (j == 0 && color_cache_bits > 0) { |
| 429 | alphabet_size += 1 << color_cache_bits; |
| 430 | } |
| 431 | size = ReadHuffmanCode(alphabet_size, dec, code_lengths, next); |
| 432 | if (size == 0) { |
| 433 | goto Error; |
| 434 | } |
| 435 | if (is_trivial_literal && kLiteralMap[j] == 1) { |
| 436 | is_trivial_literal = (next->bits == 0); |
| 437 | } |
| 438 | total_size += next->bits; |
| 439 | next += size; |
| 440 | if (j <= ALPHA) { |
| 441 | int local_max_bits = code_lengths[0]; |
| 442 | int k; |
| 443 | for (k = 1; k < alphabet_size; ++k) { |
| 444 | if (code_lengths[k] > local_max_bits) { |
| 445 | local_max_bits = code_lengths[k]; |
| 446 | } |
| 447 | } |
| 448 | max_bits += local_max_bits; |
| 449 | } |
| 450 | } |
| 451 | htree_group->is_trivial_literal = is_trivial_literal; |
| 452 | htree_group->is_trivial_code = 0; |
| 453 | if (is_trivial_literal) { |
| 454 | const int red = htrees[RED][0].value; |
| 455 | const int blue = htrees[BLUE][0].value; |
| 456 | const int alpha = htrees[ALPHA][0].value; |
| 457 | htree_group->literal_arb = |
| 458 | ((uint32_t)alpha << 24) | (red << 16) | blue; |
| 459 | if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) { |
| 460 | htree_group->is_trivial_code = 1; |
| 461 | htree_group->literal_arb |= htrees[GREEN][0].value << 8; |
| 462 | } |
| 463 | } |
| 464 | htree_group->use_packed_table = !htree_group->is_trivial_code && |
| 465 | (max_bits < HUFFMAN_PACKED_BITS); |
| 466 | if (htree_group->use_packed_table) BuildPackedTable(htree_group); |
| 467 | } |
| 468 | WebPSafeFree(code_lengths); |
| 469 | |
| 470 | // All OK. Finalize pointers and return. |
| 471 | hdr->huffman_image_ = huffman_image; |
| 472 | hdr->num_htree_groups_ = num_htree_groups; |
| 473 | hdr->htree_groups_ = htree_groups; |
| 474 | hdr->huffman_tables_ = huffman_tables; |
| 475 | return 1; |
| 476 | |
| 477 | Error: |
| 478 | WebPSafeFree(code_lengths); |
| 479 | WebPSafeFree(huffman_image); |
| 480 | WebPSafeFree(huffman_tables); |
| 481 | VP8LHtreeGroupsFree(htree_groups); |
| 482 | return 0; |
| 483 | } |
| 484 | |
| 485 | //------------------------------------------------------------------------------ |
| 486 | // Scaling. |
| 487 | |
| 488 | static int AllocateAndInitRescaler(VP8LDecoder* const dec, VP8Io* const io) { |
| 489 | const int num_channels = 4; |
| 490 | const int in_width = io->mb_w; |
| 491 | const int out_width = io->scaled_width; |
| 492 | const int in_height = io->mb_h; |
| 493 | const int out_height = io->scaled_height; |
| 494 | const uint64_t work_size = 2 * num_channels * (uint64_t)out_width; |
| 495 | rescaler_t* work; // Rescaler work area. |
| 496 | const uint64_t scaled_data_size = (uint64_t)out_width; |
| 497 | uint32_t* scaled_data; // Temporary storage for scaled BGRA data. |
| 498 | const uint64_t memory_size = sizeof(*dec->rescaler) + |
| 499 | work_size * sizeof(*work) + |
| 500 | scaled_data_size * sizeof(*scaled_data); |
| 501 | uint8_t* memory = (uint8_t*)WebPSafeMalloc(memory_size, sizeof(*memory)); |
| 502 | if (memory == NULL) { |
| 503 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
| 504 | return 0; |
| 505 | } |
| 506 | assert(dec->rescaler_memory == NULL); |
| 507 | dec->rescaler_memory = memory; |
| 508 | |
| 509 | dec->rescaler = (WebPRescaler*)memory; |
| 510 | memory += sizeof(*dec->rescaler); |
| 511 | work = (rescaler_t*)memory; |
| 512 | memory += work_size * sizeof(*work); |
| 513 | scaled_data = (uint32_t*)memory; |
| 514 | |
| 515 | WebPRescalerInit(dec->rescaler, in_width, in_height, (uint8_t*)scaled_data, |
| 516 | out_width, out_height, 0, num_channels, work); |
| 517 | return 1; |
| 518 | } |
| 519 | |
| 520 | //------------------------------------------------------------------------------ |
| 521 | // Export to ARGB |
| 522 | |
| 523 | // We have special "export" function since we need to convert from BGRA |
| 524 | static int Export(WebPRescaler* const rescaler, WEBP_CSP_MODE colorspace, |
| 525 | int rgba_stride, uint8_t* const rgba) { |
| 526 | uint32_t* const src = (uint32_t*)rescaler->dst; |
| 527 | const int dst_width = rescaler->dst_width; |
| 528 | int num_lines_out = 0; |
| 529 | while (WebPRescalerHasPendingOutput(rescaler)) { |
| 530 | uint8_t* const dst = rgba + num_lines_out * rgba_stride; |
| 531 | WebPRescalerExportRow(rescaler); |
| 532 | WebPMultARGBRow(src, dst_width, 1); |
| 533 | VP8LConvertFromBGRA(src, dst_width, colorspace, dst); |
| 534 | ++num_lines_out; |
| 535 | } |
| 536 | return num_lines_out; |
| 537 | } |
| 538 | |
| 539 | // Emit scaled rows. |
| 540 | static int EmitRescaledRowsRGBA(const VP8LDecoder* const dec, |
| 541 | uint8_t* in, int in_stride, int mb_h, |
| 542 | uint8_t* const out, int out_stride) { |
| 543 | const WEBP_CSP_MODE colorspace = dec->output_->colorspace; |
| 544 | int num_lines_in = 0; |
| 545 | int num_lines_out = 0; |
| 546 | while (num_lines_in < mb_h) { |
| 547 | uint8_t* const row_in = in + num_lines_in * in_stride; |
| 548 | uint8_t* const row_out = out + num_lines_out * out_stride; |
| 549 | const int lines_left = mb_h - num_lines_in; |
| 550 | const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left); |
| 551 | int lines_imported; |
| 552 | assert(needed_lines > 0 && needed_lines <= lines_left); |
| 553 | WebPMultARGBRows(row_in, in_stride, |
| 554 | dec->rescaler->src_width, needed_lines, 0); |
| 555 | lines_imported = |
| 556 | WebPRescalerImport(dec->rescaler, lines_left, row_in, in_stride); |
| 557 | assert(lines_imported == needed_lines); |
| 558 | num_lines_in += lines_imported; |
| 559 | num_lines_out += Export(dec->rescaler, colorspace, out_stride, row_out); |
| 560 | } |
| 561 | return num_lines_out; |
| 562 | } |
| 563 | |
| 564 | // Emit rows without any scaling. |
| 565 | static int EmitRows(WEBP_CSP_MODE colorspace, |
| 566 | const uint8_t* row_in, int in_stride, |
| 567 | int mb_w, int mb_h, |
| 568 | uint8_t* const out, int out_stride) { |
| 569 | int lines = mb_h; |
| 570 | uint8_t* row_out = out; |
| 571 | while (lines-- > 0) { |
| 572 | VP8LConvertFromBGRA((const uint32_t*)row_in, mb_w, colorspace, row_out); |
| 573 | row_in += in_stride; |
| 574 | row_out += out_stride; |
| 575 | } |
| 576 | return mb_h; // Num rows out == num rows in. |
| 577 | } |
| 578 | |
| 579 | //------------------------------------------------------------------------------ |
| 580 | // Export to YUVA |
| 581 | |
| 582 | static void ConvertToYUVA(const uint32_t* const src, int width, int y_pos, |
| 583 | const WebPDecBuffer* const output) { |
| 584 | const WebPYUVABuffer* const buf = &output->u.YUVA; |
| 585 | |
| 586 | // first, the luma plane |
| 587 | WebPConvertARGBToY(src, buf->y + y_pos * buf->y_stride, width); |
| 588 | |
| 589 | // then U/V planes |
| 590 | { |
| 591 | uint8_t* const u = buf->u + (y_pos >> 1) * buf->u_stride; |
| 592 | uint8_t* const v = buf->v + (y_pos >> 1) * buf->v_stride; |
| 593 | // even lines: store values |
| 594 | // odd lines: average with previous values |
| 595 | WebPConvertARGBToUV(src, u, v, width, !(y_pos & 1)); |
| 596 | } |
| 597 | // Lastly, store alpha if needed. |
| 598 | if (buf->a != NULL) { |
| 599 | uint8_t* const a = buf->a + y_pos * buf->a_stride; |
| 600 | #if defined(WORDS_BIGENDIAN) |
| 601 | WebPExtractAlpha((uint8_t*)src + 0, 0, width, 1, a, 0); |
| 602 | #else |
| 603 | WebPExtractAlpha((uint8_t*)src + 3, 0, width, 1, a, 0); |
| 604 | #endif |
| 605 | } |
| 606 | } |
| 607 | |
| 608 | static int ExportYUVA(const VP8LDecoder* const dec, int y_pos) { |
| 609 | WebPRescaler* const rescaler = dec->rescaler; |
| 610 | uint32_t* const src = (uint32_t*)rescaler->dst; |
| 611 | const int dst_width = rescaler->dst_width; |
| 612 | int num_lines_out = 0; |
| 613 | while (WebPRescalerHasPendingOutput(rescaler)) { |
| 614 | WebPRescalerExportRow(rescaler); |
| 615 | WebPMultARGBRow(src, dst_width, 1); |
| 616 | ConvertToYUVA(src, dst_width, y_pos, dec->output_); |
| 617 | ++y_pos; |
| 618 | ++num_lines_out; |
| 619 | } |
| 620 | return num_lines_out; |
| 621 | } |
| 622 | |
| 623 | static int EmitRescaledRowsYUVA(const VP8LDecoder* const dec, |
| 624 | uint8_t* in, int in_stride, int mb_h) { |
| 625 | int num_lines_in = 0; |
| 626 | int y_pos = dec->last_out_row_; |
| 627 | while (num_lines_in < mb_h) { |
| 628 | const int lines_left = mb_h - num_lines_in; |
| 629 | const int needed_lines = WebPRescaleNeededLines(dec->rescaler, lines_left); |
| 630 | int lines_imported; |
| 631 | WebPMultARGBRows(in, in_stride, dec->rescaler->src_width, needed_lines, 0); |
| 632 | lines_imported = |
| 633 | WebPRescalerImport(dec->rescaler, lines_left, in, in_stride); |
| 634 | assert(lines_imported == needed_lines); |
| 635 | num_lines_in += lines_imported; |
| 636 | in += needed_lines * in_stride; |
| 637 | y_pos += ExportYUVA(dec, y_pos); |
| 638 | } |
| 639 | return y_pos; |
| 640 | } |
| 641 | |
| 642 | static int EmitRowsYUVA(const VP8LDecoder* const dec, |
| 643 | const uint8_t* in, int in_stride, |
| 644 | int mb_w, int num_rows) { |
| 645 | int y_pos = dec->last_out_row_; |
| 646 | while (num_rows-- > 0) { |
| 647 | ConvertToYUVA((const uint32_t*)in, mb_w, y_pos, dec->output_); |
| 648 | in += in_stride; |
| 649 | ++y_pos; |
| 650 | } |
| 651 | return y_pos; |
| 652 | } |
| 653 | |
| 654 | //------------------------------------------------------------------------------ |
| 655 | // Cropping. |
| 656 | |
| 657 | // Sets io->mb_y, io->mb_h & io->mb_w according to start row, end row and |
| 658 | // crop options. Also updates the input data pointer, so that it points to the |
| 659 | // start of the cropped window. Note that pixels are in ARGB format even if |
| 660 | // 'in_data' is uint8_t*. |
| 661 | // Returns true if the crop window is not empty. |
| 662 | static int SetCropWindow(VP8Io* const io, int y_start, int y_end, |
| 663 | uint8_t** const in_data, int pixel_stride) { |
| 664 | assert(y_start < y_end); |
| 665 | assert(io->crop_left < io->crop_right); |
| 666 | if (y_end > io->crop_bottom) { |
| 667 | y_end = io->crop_bottom; // make sure we don't overflow on last row. |
| 668 | } |
| 669 | if (y_start < io->crop_top) { |
| 670 | const int delta = io->crop_top - y_start; |
| 671 | y_start = io->crop_top; |
| 672 | *in_data += delta * pixel_stride; |
| 673 | } |
| 674 | if (y_start >= y_end) return 0; // Crop window is empty. |
| 675 | |
| 676 | *in_data += io->crop_left * sizeof(uint32_t); |
| 677 | |
| 678 | io->mb_y = y_start - io->crop_top; |
| 679 | io->mb_w = io->crop_right - io->crop_left; |
| 680 | io->mb_h = y_end - y_start; |
| 681 | return 1; // Non-empty crop window. |
| 682 | } |
| 683 | |
| 684 | //------------------------------------------------------------------------------ |
| 685 | |
| 686 | static WEBP_INLINE int GetMetaIndex( |
| 687 | const uint32_t* const image, int xsize, int bits, int x, int y) { |
| 688 | if (bits == 0) return 0; |
| 689 | return image[xsize * (y >> bits) + (x >> bits)]; |
| 690 | } |
| 691 | |
| 692 | static WEBP_INLINE HTreeGroup* GetHtreeGroupForPos(VP8LMetadata* const hdr, |
| 693 | int x, int y) { |
| 694 | const int meta_index = GetMetaIndex(hdr->huffman_image_, hdr->huffman_xsize_, |
| 695 | hdr->huffman_subsample_bits_, x, y); |
| 696 | assert(meta_index < hdr->num_htree_groups_); |
| 697 | return hdr->htree_groups_ + meta_index; |
| 698 | } |
| 699 | |
| 700 | //------------------------------------------------------------------------------ |
| 701 | // Main loop, with custom row-processing function |
| 702 | |
| 703 | typedef void (*ProcessRowsFunc)(VP8LDecoder* const dec, int row); |
| 704 | |
| 705 | static void ApplyInverseTransforms(VP8LDecoder* const dec, int num_rows, |
| 706 | const uint32_t* const rows) { |
| 707 | int n = dec->next_transform_; |
| 708 | const int cache_pixs = dec->width_ * num_rows; |
| 709 | const int start_row = dec->last_row_; |
| 710 | const int end_row = start_row + num_rows; |
| 711 | const uint32_t* rows_in = rows; |
| 712 | uint32_t* const rows_out = dec->argb_cache_; |
| 713 | |
| 714 | // Inverse transforms. |
| 715 | while (n-- > 0) { |
| 716 | VP8LTransform* const transform = &dec->transforms_[n]; |
| 717 | VP8LInverseTransform(transform, start_row, end_row, rows_in, rows_out); |
| 718 | rows_in = rows_out; |
| 719 | } |
| 720 | if (rows_in != rows_out) { |
| 721 | // No transform called, hence just copy. |
| 722 | memcpy(rows_out, rows_in, cache_pixs * sizeof(*rows_out)); |
| 723 | } |
| 724 | } |
| 725 | |
| 726 | // Processes (transforms, scales & color-converts) the rows decoded after the |
| 727 | // last call. |
| 728 | static void ProcessRows(VP8LDecoder* const dec, int row) { |
| 729 | const uint32_t* const rows = dec->pixels_ + dec->width_ * dec->last_row_; |
| 730 | const int num_rows = row - dec->last_row_; |
| 731 | |
| 732 | assert(row <= dec->io_->crop_bottom); |
| 733 | // We can't process more than NUM_ARGB_CACHE_ROWS at a time (that's the size |
| 734 | // of argb_cache_), but we currently don't need more than that. |
| 735 | assert(num_rows <= NUM_ARGB_CACHE_ROWS); |
| 736 | if (num_rows > 0) { // Emit output. |
| 737 | VP8Io* const io = dec->io_; |
| 738 | uint8_t* rows_data = (uint8_t*)dec->argb_cache_; |
| 739 | const int in_stride = io->width * sizeof(uint32_t); // in unit of RGBA |
| 740 | |
| 741 | ApplyInverseTransforms(dec, num_rows, rows); |
| 742 | if (!SetCropWindow(io, dec->last_row_, row, &rows_data, in_stride)) { |
| 743 | // Nothing to output (this time). |
| 744 | } else { |
| 745 | const WebPDecBuffer* const output = dec->output_; |
| 746 | if (WebPIsRGBMode(output->colorspace)) { // convert to RGBA |
| 747 | const WebPRGBABuffer* const buf = &output->u.RGBA; |
| 748 | uint8_t* const rgba = buf->rgba + dec->last_out_row_ * buf->stride; |
| 749 | const int num_rows_out = io->use_scaling ? |
| 750 | EmitRescaledRowsRGBA(dec, rows_data, in_stride, io->mb_h, |
| 751 | rgba, buf->stride) : |
| 752 | EmitRows(output->colorspace, rows_data, in_stride, |
| 753 | io->mb_w, io->mb_h, rgba, buf->stride); |
| 754 | // Update 'last_out_row_'. |
| 755 | dec->last_out_row_ += num_rows_out; |
| 756 | } else { // convert to YUVA |
| 757 | dec->last_out_row_ = io->use_scaling ? |
| 758 | EmitRescaledRowsYUVA(dec, rows_data, in_stride, io->mb_h) : |
| 759 | EmitRowsYUVA(dec, rows_data, in_stride, io->mb_w, io->mb_h); |
| 760 | } |
| 761 | assert(dec->last_out_row_ <= output->height); |
| 762 | } |
| 763 | } |
| 764 | |
| 765 | // Update 'last_row_'. |
| 766 | dec->last_row_ = row; |
| 767 | assert(dec->last_row_ <= dec->height_); |
| 768 | } |
| 769 | |
| 770 | // Row-processing for the special case when alpha data contains only one |
| 771 | // transform (color indexing), and trivial non-green literals. |
| 772 | static int Is8bOptimizable(const VP8LMetadata* const hdr) { |
| 773 | int i; |
| 774 | if (hdr->color_cache_size_ > 0) return 0; |
| 775 | // When the Huffman tree contains only one symbol, we can skip the |
| 776 | // call to ReadSymbol() for red/blue/alpha channels. |
| 777 | for (i = 0; i < hdr->num_htree_groups_; ++i) { |
| 778 | HuffmanCode** const htrees = hdr->htree_groups_[i].htrees; |
| 779 | if (htrees[RED][0].bits > 0) return 0; |
| 780 | if (htrees[BLUE][0].bits > 0) return 0; |
| 781 | if (htrees[ALPHA][0].bits > 0) return 0; |
| 782 | } |
| 783 | return 1; |
| 784 | } |
| 785 | |
| 786 | static void AlphaApplyFilter(ALPHDecoder* const alph_dec, |
| 787 | int first_row, int last_row, |
| 788 | uint8_t* out, int stride) { |
| 789 | if (alph_dec->filter_ != WEBP_FILTER_NONE) { |
| 790 | int y; |
| 791 | const uint8_t* prev_line = alph_dec->prev_line_; |
| 792 | assert(WebPUnfilters[alph_dec->filter_] != NULL); |
| 793 | for (y = first_row; y < last_row; ++y) { |
| 794 | WebPUnfilters[alph_dec->filter_](prev_line, out, out, stride); |
| 795 | prev_line = out; |
| 796 | out += stride; |
| 797 | } |
| 798 | alph_dec->prev_line_ = prev_line; |
| 799 | } |
| 800 | } |
| 801 | |
| 802 | static void (VP8LDecoder* const dec, int last_row) { |
| 803 | // For vertical and gradient filtering, we need to decode the part above the |
| 804 | // crop_top row, in order to have the correct spatial predictors. |
| 805 | ALPHDecoder* const alph_dec = (ALPHDecoder*)dec->io_->opaque; |
| 806 | const int top_row = |
| 807 | (alph_dec->filter_ == WEBP_FILTER_NONE || |
| 808 | alph_dec->filter_ == WEBP_FILTER_HORIZONTAL) ? dec->io_->crop_top |
| 809 | : dec->last_row_; |
| 810 | const int first_row = (dec->last_row_ < top_row) ? top_row : dec->last_row_; |
| 811 | assert(last_row <= dec->io_->crop_bottom); |
| 812 | if (last_row > first_row) { |
| 813 | // Special method for paletted alpha data. We only process the cropped area. |
| 814 | const int width = dec->io_->width; |
| 815 | uint8_t* out = alph_dec->output_ + width * first_row; |
| 816 | const uint8_t* const in = |
| 817 | (uint8_t*)dec->pixels_ + dec->width_ * first_row; |
| 818 | VP8LTransform* const transform = &dec->transforms_[0]; |
| 819 | assert(dec->next_transform_ == 1); |
| 820 | assert(transform->type_ == COLOR_INDEXING_TRANSFORM); |
| 821 | VP8LColorIndexInverseTransformAlpha(transform, first_row, last_row, |
| 822 | in, out); |
| 823 | AlphaApplyFilter(alph_dec, first_row, last_row, out, width); |
| 824 | } |
| 825 | dec->last_row_ = dec->last_out_row_ = last_row; |
| 826 | } |
| 827 | |
| 828 | //------------------------------------------------------------------------------ |
| 829 | // Helper functions for fast pattern copy (8b and 32b) |
| 830 | |
| 831 | // cyclic rotation of pattern word |
| 832 | static WEBP_INLINE uint32_t Rotate8b(uint32_t V) { |
| 833 | #if defined(WORDS_BIGENDIAN) |
| 834 | return ((V & 0xff000000u) >> 24) | (V << 8); |
| 835 | #else |
| 836 | return ((V & 0xffu) << 24) | (V >> 8); |
| 837 | #endif |
| 838 | } |
| 839 | |
| 840 | // copy 1, 2 or 4-bytes pattern |
| 841 | static WEBP_INLINE void CopySmallPattern8b(const uint8_t* src, uint8_t* dst, |
| 842 | int length, uint32_t pattern) { |
| 843 | int i; |
| 844 | // align 'dst' to 4-bytes boundary. Adjust the pattern along the way. |
| 845 | while ((uintptr_t)dst & 3) { |
| 846 | *dst++ = *src++; |
| 847 | pattern = Rotate8b(pattern); |
| 848 | --length; |
| 849 | } |
| 850 | // Copy the pattern 4 bytes at a time. |
| 851 | for (i = 0; i < (length >> 2); ++i) { |
| 852 | ((uint32_t*)dst)[i] = pattern; |
| 853 | } |
| 854 | // Finish with left-overs. 'pattern' is still correctly positioned, |
| 855 | // so no Rotate8b() call is needed. |
| 856 | for (i <<= 2; i < length; ++i) { |
| 857 | dst[i] = src[i]; |
| 858 | } |
| 859 | } |
| 860 | |
| 861 | static WEBP_INLINE void CopyBlock8b(uint8_t* const dst, int dist, int length) { |
| 862 | const uint8_t* src = dst - dist; |
| 863 | if (length >= 8) { |
| 864 | uint32_t pattern = 0; |
| 865 | switch (dist) { |
| 866 | case 1: |
| 867 | pattern = src[0]; |
| 868 | #if defined(__arm__) || defined(_M_ARM) // arm doesn't like multiply that much |
| 869 | pattern |= pattern << 8; |
| 870 | pattern |= pattern << 16; |
| 871 | #elif defined(WEBP_USE_MIPS_DSP_R2) |
| 872 | __asm__ volatile ("replv.qb %0, %0" : "+r" (pattern)); |
| 873 | #else |
| 874 | pattern = 0x01010101u * pattern; |
| 875 | #endif |
| 876 | break; |
| 877 | case 2: |
| 878 | memcpy(&pattern, src, sizeof(uint16_t)); |
| 879 | #if defined(__arm__) || defined(_M_ARM) |
| 880 | pattern |= pattern << 16; |
| 881 | #elif defined(WEBP_USE_MIPS_DSP_R2) |
| 882 | __asm__ volatile ("replv.ph %0, %0" : "+r" (pattern)); |
| 883 | #else |
| 884 | pattern = 0x00010001u * pattern; |
| 885 | #endif |
| 886 | break; |
| 887 | case 4: |
| 888 | memcpy(&pattern, src, sizeof(uint32_t)); |
| 889 | break; |
| 890 | default: |
| 891 | goto Copy; |
| 892 | break; |
| 893 | } |
| 894 | CopySmallPattern8b(src, dst, length, pattern); |
| 895 | return; |
| 896 | } |
| 897 | Copy: |
| 898 | if (dist >= length) { // no overlap -> use memcpy() |
| 899 | memcpy(dst, src, length * sizeof(*dst)); |
| 900 | } else { |
| 901 | int i; |
| 902 | for (i = 0; i < length; ++i) dst[i] = src[i]; |
| 903 | } |
| 904 | } |
| 905 | |
| 906 | // copy pattern of 1 or 2 uint32_t's |
| 907 | static WEBP_INLINE void CopySmallPattern32b(const uint32_t* src, |
| 908 | uint32_t* dst, |
| 909 | int length, uint64_t pattern) { |
| 910 | int i; |
| 911 | if ((uintptr_t)dst & 4) { // Align 'dst' to 8-bytes boundary. |
| 912 | *dst++ = *src++; |
| 913 | pattern = (pattern >> 32) | (pattern << 32); |
| 914 | --length; |
| 915 | } |
| 916 | assert(0 == ((uintptr_t)dst & 7)); |
| 917 | for (i = 0; i < (length >> 1); ++i) { |
| 918 | ((uint64_t*)dst)[i] = pattern; // Copy the pattern 8 bytes at a time. |
| 919 | } |
| 920 | if (length & 1) { // Finish with left-over. |
| 921 | dst[i << 1] = src[i << 1]; |
| 922 | } |
| 923 | } |
| 924 | |
| 925 | static WEBP_INLINE void CopyBlock32b(uint32_t* const dst, |
| 926 | int dist, int length) { |
| 927 | const uint32_t* const src = dst - dist; |
| 928 | if (dist <= 2 && length >= 4 && ((uintptr_t)dst & 3) == 0) { |
| 929 | uint64_t pattern; |
| 930 | if (dist == 1) { |
| 931 | pattern = (uint64_t)src[0]; |
| 932 | pattern |= pattern << 32; |
| 933 | } else { |
| 934 | memcpy(&pattern, src, sizeof(pattern)); |
| 935 | } |
| 936 | CopySmallPattern32b(src, dst, length, pattern); |
| 937 | } else if (dist >= length) { // no overlap |
| 938 | memcpy(dst, src, length * sizeof(*dst)); |
| 939 | } else { |
| 940 | int i; |
| 941 | for (i = 0; i < length; ++i) dst[i] = src[i]; |
| 942 | } |
| 943 | } |
| 944 | |
| 945 | //------------------------------------------------------------------------------ |
| 946 | |
| 947 | static int DecodeAlphaData(VP8LDecoder* const dec, uint8_t* const data, |
| 948 | int width, int height, int last_row) { |
| 949 | int ok = 1; |
| 950 | int row = dec->last_pixel_ / width; |
| 951 | int col = dec->last_pixel_ % width; |
| 952 | VP8LBitReader* const br = &dec->br_; |
| 953 | VP8LMetadata* const hdr = &dec->hdr_; |
| 954 | int pos = dec->last_pixel_; // current position |
| 955 | const int end = width * height; // End of data |
| 956 | const int last = width * last_row; // Last pixel to decode |
| 957 | const int len_code_limit = NUM_LITERAL_CODES + NUM_LENGTH_CODES; |
| 958 | const int mask = hdr->huffman_mask_; |
| 959 | const HTreeGroup* htree_group = |
| 960 | (pos < last) ? GetHtreeGroupForPos(hdr, col, row) : NULL; |
| 961 | assert(pos <= end); |
| 962 | assert(last_row <= height); |
| 963 | assert(Is8bOptimizable(hdr)); |
| 964 | |
| 965 | while (!br->eos_ && pos < last) { |
| 966 | int code; |
| 967 | // Only update when changing tile. |
| 968 | if ((col & mask) == 0) { |
| 969 | htree_group = GetHtreeGroupForPos(hdr, col, row); |
| 970 | } |
| 971 | assert(htree_group != NULL); |
| 972 | VP8LFillBitWindow(br); |
| 973 | code = ReadSymbol(htree_group->htrees[GREEN], br); |
| 974 | if (code < NUM_LITERAL_CODES) { // Literal |
| 975 | data[pos] = code; |
| 976 | ++pos; |
| 977 | ++col; |
| 978 | if (col >= width) { |
| 979 | col = 0; |
| 980 | ++row; |
| 981 | if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { |
| 982 | ExtractPalettedAlphaRows(dec, row); |
| 983 | } |
| 984 | } |
| 985 | } else if (code < len_code_limit) { // Backward reference |
| 986 | int dist_code, dist; |
| 987 | const int length_sym = code - NUM_LITERAL_CODES; |
| 988 | const int length = GetCopyLength(length_sym, br); |
| 989 | const int dist_symbol = ReadSymbol(htree_group->htrees[DIST], br); |
| 990 | VP8LFillBitWindow(br); |
| 991 | dist_code = GetCopyDistance(dist_symbol, br); |
| 992 | dist = PlaneCodeToDistance(width, dist_code); |
| 993 | if (pos >= dist && end - pos >= length) { |
| 994 | CopyBlock8b(data + pos, dist, length); |
| 995 | } else { |
| 996 | ok = 0; |
| 997 | goto End; |
| 998 | } |
| 999 | pos += length; |
| 1000 | col += length; |
| 1001 | while (col >= width) { |
| 1002 | col -= width; |
| 1003 | ++row; |
| 1004 | if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { |
| 1005 | ExtractPalettedAlphaRows(dec, row); |
| 1006 | } |
| 1007 | } |
| 1008 | if (pos < last && (col & mask)) { |
| 1009 | htree_group = GetHtreeGroupForPos(hdr, col, row); |
| 1010 | } |
| 1011 | } else { // Not reached |
| 1012 | ok = 0; |
| 1013 | goto End; |
| 1014 | } |
| 1015 | assert(br->eos_ == VP8LIsEndOfStream(br)); |
| 1016 | } |
| 1017 | // Process the remaining rows corresponding to last row-block. |
| 1018 | ExtractPalettedAlphaRows(dec, row > last_row ? last_row : row); |
| 1019 | |
| 1020 | End: |
| 1021 | if (!ok || (br->eos_ && pos < end)) { |
| 1022 | ok = 0; |
| 1023 | dec->status_ = br->eos_ ? VP8_STATUS_SUSPENDED |
| 1024 | : VP8_STATUS_BITSTREAM_ERROR; |
| 1025 | } else { |
| 1026 | dec->last_pixel_ = pos; |
| 1027 | } |
| 1028 | return ok; |
| 1029 | } |
| 1030 | |
| 1031 | static void SaveState(VP8LDecoder* const dec, int last_pixel) { |
| 1032 | assert(dec->incremental_); |
| 1033 | dec->saved_br_ = dec->br_; |
| 1034 | dec->saved_last_pixel_ = last_pixel; |
| 1035 | if (dec->hdr_.color_cache_size_ > 0) { |
| 1036 | VP8LColorCacheCopy(&dec->hdr_.color_cache_, &dec->hdr_.saved_color_cache_); |
| 1037 | } |
| 1038 | } |
| 1039 | |
| 1040 | static void RestoreState(VP8LDecoder* const dec) { |
| 1041 | assert(dec->br_.eos_); |
| 1042 | dec->status_ = VP8_STATUS_SUSPENDED; |
| 1043 | dec->br_ = dec->saved_br_; |
| 1044 | dec->last_pixel_ = dec->saved_last_pixel_; |
| 1045 | if (dec->hdr_.color_cache_size_ > 0) { |
| 1046 | VP8LColorCacheCopy(&dec->hdr_.saved_color_cache_, &dec->hdr_.color_cache_); |
| 1047 | } |
| 1048 | } |
| 1049 | |
| 1050 | #define SYNC_EVERY_N_ROWS 8 // minimum number of rows between check-points |
| 1051 | static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data, |
| 1052 | int width, int height, int last_row, |
| 1053 | ProcessRowsFunc process_func) { |
| 1054 | int row = dec->last_pixel_ / width; |
| 1055 | int col = dec->last_pixel_ % width; |
| 1056 | VP8LBitReader* const br = &dec->br_; |
| 1057 | VP8LMetadata* const hdr = &dec->hdr_; |
| 1058 | uint32_t* src = data + dec->last_pixel_; |
| 1059 | uint32_t* last_cached = src; |
| 1060 | uint32_t* const src_end = data + width * height; // End of data |
| 1061 | uint32_t* const src_last = data + width * last_row; // Last pixel to decode |
| 1062 | const int len_code_limit = NUM_LITERAL_CODES + NUM_LENGTH_CODES; |
| 1063 | const int color_cache_limit = len_code_limit + hdr->color_cache_size_; |
| 1064 | int next_sync_row = dec->incremental_ ? row : 1 << 24; |
| 1065 | VP8LColorCache* const color_cache = |
| 1066 | (hdr->color_cache_size_ > 0) ? &hdr->color_cache_ : NULL; |
| 1067 | const int mask = hdr->huffman_mask_; |
| 1068 | const HTreeGroup* htree_group = |
| 1069 | (src < src_last) ? GetHtreeGroupForPos(hdr, col, row) : NULL; |
| 1070 | assert(dec->last_row_ < last_row); |
| 1071 | assert(src_last <= src_end); |
| 1072 | |
| 1073 | while (src < src_last) { |
| 1074 | int code; |
| 1075 | if (row >= next_sync_row) { |
| 1076 | SaveState(dec, (int)(src - data)); |
| 1077 | next_sync_row = row + SYNC_EVERY_N_ROWS; |
| 1078 | } |
| 1079 | // Only update when changing tile. Note we could use this test: |
| 1080 | // if "((((prev_col ^ col) | prev_row ^ row)) > mask)" -> tile changed |
| 1081 | // but that's actually slower and needs storing the previous col/row. |
| 1082 | if ((col & mask) == 0) { |
| 1083 | htree_group = GetHtreeGroupForPos(hdr, col, row); |
| 1084 | } |
| 1085 | assert(htree_group != NULL); |
| 1086 | if (htree_group->is_trivial_code) { |
| 1087 | *src = htree_group->literal_arb; |
| 1088 | goto AdvanceByOne; |
| 1089 | } |
| 1090 | VP8LFillBitWindow(br); |
| 1091 | if (htree_group->use_packed_table) { |
| 1092 | code = ReadPackedSymbols(htree_group, br, src); |
| 1093 | if (code == PACKED_NON_LITERAL_CODE) goto AdvanceByOne; |
| 1094 | } else { |
| 1095 | code = ReadSymbol(htree_group->htrees[GREEN], br); |
| 1096 | } |
| 1097 | if (br->eos_) break; // early out |
| 1098 | if (code < NUM_LITERAL_CODES) { // Literal |
| 1099 | if (htree_group->is_trivial_literal) { |
| 1100 | *src = htree_group->literal_arb | (code << 8); |
| 1101 | } else { |
| 1102 | int red, blue, alpha; |
| 1103 | red = ReadSymbol(htree_group->htrees[RED], br); |
| 1104 | VP8LFillBitWindow(br); |
| 1105 | blue = ReadSymbol(htree_group->htrees[BLUE], br); |
| 1106 | alpha = ReadSymbol(htree_group->htrees[ALPHA], br); |
| 1107 | if (br->eos_) break; |
| 1108 | *src = ((uint32_t)alpha << 24) | (red << 16) | (code << 8) | blue; |
| 1109 | } |
| 1110 | AdvanceByOne: |
| 1111 | ++src; |
| 1112 | ++col; |
| 1113 | if (col >= width) { |
| 1114 | col = 0; |
| 1115 | ++row; |
| 1116 | if (process_func != NULL) { |
| 1117 | if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { |
| 1118 | process_func(dec, row); |
| 1119 | } |
| 1120 | } |
| 1121 | if (color_cache != NULL) { |
| 1122 | while (last_cached < src) { |
| 1123 | VP8LColorCacheInsert(color_cache, *last_cached++); |
| 1124 | } |
| 1125 | } |
| 1126 | } |
| 1127 | } else if (code < len_code_limit) { // Backward reference |
| 1128 | int dist_code, dist; |
| 1129 | const int length_sym = code - NUM_LITERAL_CODES; |
| 1130 | const int length = GetCopyLength(length_sym, br); |
| 1131 | const int dist_symbol = ReadSymbol(htree_group->htrees[DIST], br); |
| 1132 | VP8LFillBitWindow(br); |
| 1133 | dist_code = GetCopyDistance(dist_symbol, br); |
| 1134 | dist = PlaneCodeToDistance(width, dist_code); |
| 1135 | if (br->eos_) break; |
| 1136 | if (src - data < (ptrdiff_t)dist || src_end - src < (ptrdiff_t)length) { |
| 1137 | goto Error; |
| 1138 | } else { |
| 1139 | CopyBlock32b(src, dist, length); |
| 1140 | } |
| 1141 | src += length; |
| 1142 | col += length; |
| 1143 | while (col >= width) { |
| 1144 | col -= width; |
| 1145 | ++row; |
| 1146 | if (process_func != NULL) { |
| 1147 | if (row <= last_row && (row % NUM_ARGB_CACHE_ROWS == 0)) { |
| 1148 | process_func(dec, row); |
| 1149 | } |
| 1150 | } |
| 1151 | } |
| 1152 | // Because of the check done above (before 'src' was incremented by |
| 1153 | // 'length'), the following holds true. |
| 1154 | assert(src <= src_end); |
| 1155 | if (col & mask) htree_group = GetHtreeGroupForPos(hdr, col, row); |
| 1156 | if (color_cache != NULL) { |
| 1157 | while (last_cached < src) { |
| 1158 | VP8LColorCacheInsert(color_cache, *last_cached++); |
| 1159 | } |
| 1160 | } |
| 1161 | } else if (code < color_cache_limit) { // Color cache |
| 1162 | const int key = code - len_code_limit; |
| 1163 | assert(color_cache != NULL); |
| 1164 | while (last_cached < src) { |
| 1165 | VP8LColorCacheInsert(color_cache, *last_cached++); |
| 1166 | } |
| 1167 | *src = VP8LColorCacheLookup(color_cache, key); |
| 1168 | goto AdvanceByOne; |
| 1169 | } else { // Not reached |
| 1170 | goto Error; |
| 1171 | } |
| 1172 | assert(br->eos_ == VP8LIsEndOfStream(br)); |
| 1173 | } |
| 1174 | |
| 1175 | if (dec->incremental_ && br->eos_ && src < src_end) { |
| 1176 | RestoreState(dec); |
| 1177 | } else if (!br->eos_) { |
| 1178 | // Process the remaining rows corresponding to last row-block. |
| 1179 | if (process_func != NULL) { |
| 1180 | process_func(dec, row > last_row ? last_row : row); |
| 1181 | } |
| 1182 | dec->status_ = VP8_STATUS_OK; |
| 1183 | dec->last_pixel_ = (int)(src - data); // end-of-scan marker |
| 1184 | } else { |
| 1185 | // if not incremental, and we are past the end of buffer (eos_=1), then this |
| 1186 | // is a real bitstream error. |
| 1187 | goto Error; |
| 1188 | } |
| 1189 | return 1; |
| 1190 | |
| 1191 | Error: |
| 1192 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
| 1193 | return 0; |
| 1194 | } |
| 1195 | |
| 1196 | // ----------------------------------------------------------------------------- |
| 1197 | // VP8LTransform |
| 1198 | |
| 1199 | static void ClearTransform(VP8LTransform* const transform) { |
| 1200 | WebPSafeFree(transform->data_); |
| 1201 | transform->data_ = NULL; |
| 1202 | } |
| 1203 | |
| 1204 | // For security reason, we need to remap the color map to span |
| 1205 | // the total possible bundled values, and not just the num_colors. |
| 1206 | static int ExpandColorMap(int num_colors, VP8LTransform* const transform) { |
| 1207 | int i; |
| 1208 | const int final_num_colors = 1 << (8 >> transform->bits_); |
| 1209 | uint32_t* const new_color_map = |
| 1210 | (uint32_t*)WebPSafeMalloc((uint64_t)final_num_colors, |
| 1211 | sizeof(*new_color_map)); |
| 1212 | if (new_color_map == NULL) { |
| 1213 | return 0; |
| 1214 | } else { |
| 1215 | uint8_t* const data = (uint8_t*)transform->data_; |
| 1216 | uint8_t* const new_data = (uint8_t*)new_color_map; |
| 1217 | new_color_map[0] = transform->data_[0]; |
| 1218 | for (i = 4; i < 4 * num_colors; ++i) { |
| 1219 | // Equivalent to AddPixelEq(), on a byte-basis. |
| 1220 | new_data[i] = (data[i] + new_data[i - 4]) & 0xff; |
| 1221 | } |
| 1222 | for (; i < 4 * final_num_colors; ++i) { |
| 1223 | new_data[i] = 0; // black tail. |
| 1224 | } |
| 1225 | WebPSafeFree(transform->data_); |
| 1226 | transform->data_ = new_color_map; |
| 1227 | } |
| 1228 | return 1; |
| 1229 | } |
| 1230 | |
| 1231 | static int ReadTransform(int* const xsize, int const* ysize, |
| 1232 | VP8LDecoder* const dec) { |
| 1233 | int ok = 1; |
| 1234 | VP8LBitReader* const br = &dec->br_; |
| 1235 | VP8LTransform* transform = &dec->transforms_[dec->next_transform_]; |
| 1236 | const VP8LImageTransformType type = |
| 1237 | (VP8LImageTransformType)VP8LReadBits(br, 2); |
| 1238 | |
| 1239 | // Each transform type can only be present once in the stream. |
| 1240 | if (dec->transforms_seen_ & (1U << type)) { |
| 1241 | return 0; // Already there, let's not accept the second same transform. |
| 1242 | } |
| 1243 | dec->transforms_seen_ |= (1U << type); |
| 1244 | |
| 1245 | transform->type_ = type; |
| 1246 | transform->xsize_ = *xsize; |
| 1247 | transform->ysize_ = *ysize; |
| 1248 | transform->data_ = NULL; |
| 1249 | ++dec->next_transform_; |
| 1250 | assert(dec->next_transform_ <= NUM_TRANSFORMS); |
| 1251 | |
| 1252 | switch (type) { |
| 1253 | case PREDICTOR_TRANSFORM: |
| 1254 | case CROSS_COLOR_TRANSFORM: |
| 1255 | transform->bits_ = VP8LReadBits(br, 3) + 2; |
| 1256 | ok = DecodeImageStream(VP8LSubSampleSize(transform->xsize_, |
| 1257 | transform->bits_), |
| 1258 | VP8LSubSampleSize(transform->ysize_, |
| 1259 | transform->bits_), |
| 1260 | 0, dec, &transform->data_); |
| 1261 | break; |
| 1262 | case COLOR_INDEXING_TRANSFORM: { |
| 1263 | const int num_colors = VP8LReadBits(br, 8) + 1; |
| 1264 | const int bits = (num_colors > 16) ? 0 |
| 1265 | : (num_colors > 4) ? 1 |
| 1266 | : (num_colors > 2) ? 2 |
| 1267 | : 3; |
| 1268 | *xsize = VP8LSubSampleSize(transform->xsize_, bits); |
| 1269 | transform->bits_ = bits; |
| 1270 | ok = DecodeImageStream(num_colors, 1, 0, dec, &transform->data_); |
| 1271 | ok = ok && ExpandColorMap(num_colors, transform); |
| 1272 | break; |
| 1273 | } |
| 1274 | case SUBTRACT_GREEN: |
| 1275 | break; |
| 1276 | default: |
| 1277 | assert(0); // can't happen |
| 1278 | break; |
| 1279 | } |
| 1280 | |
| 1281 | return ok; |
| 1282 | } |
| 1283 | |
| 1284 | // ----------------------------------------------------------------------------- |
| 1285 | // VP8LMetadata |
| 1286 | |
| 1287 | static void InitMetadata(VP8LMetadata* const hdr) { |
| 1288 | assert(hdr != NULL); |
| 1289 | memset(hdr, 0, sizeof(*hdr)); |
| 1290 | } |
| 1291 | |
| 1292 | static void ClearMetadata(VP8LMetadata* const hdr) { |
| 1293 | assert(hdr != NULL); |
| 1294 | |
| 1295 | WebPSafeFree(hdr->huffman_image_); |
| 1296 | WebPSafeFree(hdr->huffman_tables_); |
| 1297 | VP8LHtreeGroupsFree(hdr->htree_groups_); |
| 1298 | VP8LColorCacheClear(&hdr->color_cache_); |
| 1299 | VP8LColorCacheClear(&hdr->saved_color_cache_); |
| 1300 | InitMetadata(hdr); |
| 1301 | } |
| 1302 | |
| 1303 | // ----------------------------------------------------------------------------- |
| 1304 | // VP8LDecoder |
| 1305 | |
| 1306 | VP8LDecoder* VP8LNew(void) { |
| 1307 | VP8LDecoder* const dec = (VP8LDecoder*)WebPSafeCalloc(1ULL, sizeof(*dec)); |
| 1308 | if (dec == NULL) return NULL; |
| 1309 | dec->status_ = VP8_STATUS_OK; |
| 1310 | dec->state_ = READ_DIM; |
| 1311 | |
| 1312 | VP8LDspInit(); // Init critical function pointers. |
| 1313 | |
| 1314 | return dec; |
| 1315 | } |
| 1316 | |
| 1317 | void VP8LClear(VP8LDecoder* const dec) { |
| 1318 | int i; |
| 1319 | if (dec == NULL) return; |
| 1320 | ClearMetadata(&dec->hdr_); |
| 1321 | |
| 1322 | WebPSafeFree(dec->pixels_); |
| 1323 | dec->pixels_ = NULL; |
| 1324 | for (i = 0; i < dec->next_transform_; ++i) { |
| 1325 | ClearTransform(&dec->transforms_[i]); |
| 1326 | } |
| 1327 | dec->next_transform_ = 0; |
| 1328 | dec->transforms_seen_ = 0; |
| 1329 | |
| 1330 | WebPSafeFree(dec->rescaler_memory); |
| 1331 | dec->rescaler_memory = NULL; |
| 1332 | |
| 1333 | dec->output_ = NULL; // leave no trace behind |
| 1334 | } |
| 1335 | |
| 1336 | void VP8LDelete(VP8LDecoder* const dec) { |
| 1337 | if (dec != NULL) { |
| 1338 | VP8LClear(dec); |
| 1339 | WebPSafeFree(dec); |
| 1340 | } |
| 1341 | } |
| 1342 | |
| 1343 | static void UpdateDecoder(VP8LDecoder* const dec, int width, int height) { |
| 1344 | VP8LMetadata* const hdr = &dec->hdr_; |
| 1345 | const int num_bits = hdr->huffman_subsample_bits_; |
| 1346 | dec->width_ = width; |
| 1347 | dec->height_ = height; |
| 1348 | |
| 1349 | hdr->huffman_xsize_ = VP8LSubSampleSize(width, num_bits); |
| 1350 | hdr->huffman_mask_ = (num_bits == 0) ? ~0 : (1 << num_bits) - 1; |
| 1351 | } |
| 1352 | |
| 1353 | static int DecodeImageStream(int xsize, int ysize, |
| 1354 | int is_level0, |
| 1355 | VP8LDecoder* const dec, |
| 1356 | uint32_t** const decoded_data) { |
| 1357 | int ok = 1; |
| 1358 | int transform_xsize = xsize; |
| 1359 | int transform_ysize = ysize; |
| 1360 | VP8LBitReader* const br = &dec->br_; |
| 1361 | VP8LMetadata* const hdr = &dec->hdr_; |
| 1362 | uint32_t* data = NULL; |
| 1363 | int color_cache_bits = 0; |
| 1364 | |
| 1365 | // Read the transforms (may recurse). |
| 1366 | if (is_level0) { |
| 1367 | while (ok && VP8LReadBits(br, 1)) { |
| 1368 | ok = ReadTransform(&transform_xsize, &transform_ysize, dec); |
| 1369 | } |
| 1370 | } |
| 1371 | |
| 1372 | // Color cache |
| 1373 | if (ok && VP8LReadBits(br, 1)) { |
| 1374 | color_cache_bits = VP8LReadBits(br, 4); |
| 1375 | ok = (color_cache_bits >= 1 && color_cache_bits <= MAX_CACHE_BITS); |
| 1376 | if (!ok) { |
| 1377 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
| 1378 | goto End; |
| 1379 | } |
| 1380 | } |
| 1381 | |
| 1382 | // Read the Huffman codes (may recurse). |
| 1383 | ok = ok && ReadHuffmanCodes(dec, transform_xsize, transform_ysize, |
| 1384 | color_cache_bits, is_level0); |
| 1385 | if (!ok) { |
| 1386 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
| 1387 | goto End; |
| 1388 | } |
| 1389 | |
| 1390 | // Finish setting up the color-cache |
| 1391 | if (color_cache_bits > 0) { |
| 1392 | hdr->color_cache_size_ = 1 << color_cache_bits; |
| 1393 | if (!VP8LColorCacheInit(&hdr->color_cache_, color_cache_bits)) { |
| 1394 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
| 1395 | ok = 0; |
| 1396 | goto End; |
| 1397 | } |
| 1398 | } else { |
| 1399 | hdr->color_cache_size_ = 0; |
| 1400 | } |
| 1401 | UpdateDecoder(dec, transform_xsize, transform_ysize); |
| 1402 | |
| 1403 | if (is_level0) { // level 0 complete |
| 1404 | dec->state_ = READ_HDR; |
| 1405 | goto End; |
| 1406 | } |
| 1407 | |
| 1408 | { |
| 1409 | const uint64_t total_size = (uint64_t)transform_xsize * transform_ysize; |
| 1410 | data = (uint32_t*)WebPSafeMalloc(total_size, sizeof(*data)); |
| 1411 | if (data == NULL) { |
| 1412 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
| 1413 | ok = 0; |
| 1414 | goto End; |
| 1415 | } |
| 1416 | } |
| 1417 | |
| 1418 | // Use the Huffman trees to decode the LZ77 encoded data. |
| 1419 | ok = DecodeImageData(dec, data, transform_xsize, transform_ysize, |
| 1420 | transform_ysize, NULL); |
| 1421 | ok = ok && !br->eos_; |
| 1422 | |
| 1423 | End: |
| 1424 | if (!ok) { |
| 1425 | WebPSafeFree(data); |
| 1426 | ClearMetadata(hdr); |
| 1427 | } else { |
| 1428 | if (decoded_data != NULL) { |
| 1429 | *decoded_data = data; |
| 1430 | } else { |
| 1431 | // We allocate image data in this function only for transforms. At level 0 |
| 1432 | // (that is: not the transforms), we shouldn't have allocated anything. |
| 1433 | assert(data == NULL); |
| 1434 | assert(is_level0); |
| 1435 | } |
| 1436 | dec->last_pixel_ = 0; // Reset for future DECODE_DATA_FUNC() calls. |
| 1437 | if (!is_level0) ClearMetadata(hdr); // Clean up temporary data behind. |
| 1438 | } |
| 1439 | return ok; |
| 1440 | } |
| 1441 | |
| 1442 | //------------------------------------------------------------------------------ |
| 1443 | // Allocate internal buffers dec->pixels_ and dec->argb_cache_. |
| 1444 | static int AllocateInternalBuffers32b(VP8LDecoder* const dec, int final_width) { |
| 1445 | const uint64_t num_pixels = (uint64_t)dec->width_ * dec->height_; |
| 1446 | // Scratch buffer corresponding to top-prediction row for transforming the |
| 1447 | // first row in the row-blocks. Not needed for paletted alpha. |
| 1448 | const uint64_t cache_top_pixels = (uint16_t)final_width; |
| 1449 | // Scratch buffer for temporary BGRA storage. Not needed for paletted alpha. |
| 1450 | const uint64_t cache_pixels = (uint64_t)final_width * NUM_ARGB_CACHE_ROWS; |
| 1451 | const uint64_t total_num_pixels = |
| 1452 | num_pixels + cache_top_pixels + cache_pixels; |
| 1453 | |
| 1454 | assert(dec->width_ <= final_width); |
| 1455 | dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint32_t)); |
| 1456 | if (dec->pixels_ == NULL) { |
| 1457 | dec->argb_cache_ = NULL; // for sanity check |
| 1458 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
| 1459 | return 0; |
| 1460 | } |
| 1461 | dec->argb_cache_ = dec->pixels_ + num_pixels + cache_top_pixels; |
| 1462 | return 1; |
| 1463 | } |
| 1464 | |
| 1465 | static int AllocateInternalBuffers8b(VP8LDecoder* const dec) { |
| 1466 | const uint64_t total_num_pixels = (uint64_t)dec->width_ * dec->height_; |
| 1467 | dec->argb_cache_ = NULL; // for sanity check |
| 1468 | dec->pixels_ = (uint32_t*)WebPSafeMalloc(total_num_pixels, sizeof(uint8_t)); |
| 1469 | if (dec->pixels_ == NULL) { |
| 1470 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
| 1471 | return 0; |
| 1472 | } |
| 1473 | return 1; |
| 1474 | } |
| 1475 | |
| 1476 | //------------------------------------------------------------------------------ |
| 1477 | |
| 1478 | // Special row-processing that only stores the alpha data. |
| 1479 | static void (VP8LDecoder* const dec, int last_row) { |
| 1480 | int cur_row = dec->last_row_; |
| 1481 | int num_rows = last_row - cur_row; |
| 1482 | const uint32_t* in = dec->pixels_ + dec->width_ * cur_row; |
| 1483 | |
| 1484 | assert(last_row <= dec->io_->crop_bottom); |
| 1485 | while (num_rows > 0) { |
| 1486 | const int num_rows_to_process = |
| 1487 | (num_rows > NUM_ARGB_CACHE_ROWS) ? NUM_ARGB_CACHE_ROWS : num_rows; |
| 1488 | // Extract alpha (which is stored in the green plane). |
| 1489 | ALPHDecoder* const alph_dec = (ALPHDecoder*)dec->io_->opaque; |
| 1490 | uint8_t* const output = alph_dec->output_; |
| 1491 | const int width = dec->io_->width; // the final width (!= dec->width_) |
| 1492 | const int cache_pixs = width * num_rows_to_process; |
| 1493 | uint8_t* const dst = output + width * cur_row; |
| 1494 | const uint32_t* const src = dec->argb_cache_; |
| 1495 | ApplyInverseTransforms(dec, num_rows_to_process, in); |
| 1496 | WebPExtractGreen(src, dst, cache_pixs); |
| 1497 | AlphaApplyFilter(alph_dec, |
| 1498 | cur_row, cur_row + num_rows_to_process, dst, width); |
| 1499 | num_rows -= num_rows_to_process; |
| 1500 | in += num_rows_to_process * dec->width_; |
| 1501 | cur_row += num_rows_to_process; |
| 1502 | } |
| 1503 | assert(cur_row == last_row); |
| 1504 | dec->last_row_ = dec->last_out_row_ = last_row; |
| 1505 | } |
| 1506 | |
| 1507 | int (ALPHDecoder* const alph_dec, |
| 1508 | const uint8_t* const data, size_t data_size) { |
| 1509 | int ok = 0; |
| 1510 | VP8LDecoder* dec = VP8LNew(); |
| 1511 | |
| 1512 | if (dec == NULL) return 0; |
| 1513 | |
| 1514 | assert(alph_dec != NULL); |
| 1515 | alph_dec->vp8l_dec_ = dec; |
| 1516 | |
| 1517 | dec->width_ = alph_dec->width_; |
| 1518 | dec->height_ = alph_dec->height_; |
| 1519 | dec->io_ = &alph_dec->io_; |
| 1520 | dec->io_->opaque = alph_dec; |
| 1521 | dec->io_->width = alph_dec->width_; |
| 1522 | dec->io_->height = alph_dec->height_; |
| 1523 | |
| 1524 | dec->status_ = VP8_STATUS_OK; |
| 1525 | VP8LInitBitReader(&dec->br_, data, data_size); |
| 1526 | |
| 1527 | if (!DecodeImageStream(alph_dec->width_, alph_dec->height_, 1, dec, NULL)) { |
| 1528 | goto Err; |
| 1529 | } |
| 1530 | |
| 1531 | // Special case: if alpha data uses only the color indexing transform and |
| 1532 | // doesn't use color cache (a frequent case), we will use DecodeAlphaData() |
| 1533 | // method that only needs allocation of 1 byte per pixel (alpha channel). |
| 1534 | if (dec->next_transform_ == 1 && |
| 1535 | dec->transforms_[0].type_ == COLOR_INDEXING_TRANSFORM && |
| 1536 | Is8bOptimizable(&dec->hdr_)) { |
| 1537 | alph_dec->use_8b_decode_ = 1; |
| 1538 | ok = AllocateInternalBuffers8b(dec); |
| 1539 | } else { |
| 1540 | // Allocate internal buffers (note that dec->width_ may have changed here). |
| 1541 | alph_dec->use_8b_decode_ = 0; |
| 1542 | ok = AllocateInternalBuffers32b(dec, alph_dec->width_); |
| 1543 | } |
| 1544 | |
| 1545 | if (!ok) goto Err; |
| 1546 | |
| 1547 | return 1; |
| 1548 | |
| 1549 | Err: |
| 1550 | VP8LDelete(alph_dec->vp8l_dec_); |
| 1551 | alph_dec->vp8l_dec_ = NULL; |
| 1552 | return 0; |
| 1553 | } |
| 1554 | |
| 1555 | int VP8LDecodeAlphaImageStream(ALPHDecoder* const alph_dec, int last_row) { |
| 1556 | VP8LDecoder* const dec = alph_dec->vp8l_dec_; |
| 1557 | assert(dec != NULL); |
| 1558 | assert(last_row <= dec->height_); |
| 1559 | |
| 1560 | if (dec->last_row_ >= last_row) { |
| 1561 | return 1; // done |
| 1562 | } |
| 1563 | |
| 1564 | if (!alph_dec->use_8b_decode_) WebPInitAlphaProcessing(); |
| 1565 | |
| 1566 | // Decode (with special row processing). |
| 1567 | return alph_dec->use_8b_decode_ ? |
| 1568 | DecodeAlphaData(dec, (uint8_t*)dec->pixels_, dec->width_, dec->height_, |
| 1569 | last_row) : |
| 1570 | DecodeImageData(dec, dec->pixels_, dec->width_, dec->height_, |
| 1571 | last_row, ExtractAlphaRows); |
| 1572 | } |
| 1573 | |
| 1574 | //------------------------------------------------------------------------------ |
| 1575 | |
| 1576 | int (VP8LDecoder* const dec, VP8Io* const io) { |
| 1577 | int width, height, has_alpha; |
| 1578 | |
| 1579 | if (dec == NULL) return 0; |
| 1580 | if (io == NULL) { |
| 1581 | dec->status_ = VP8_STATUS_INVALID_PARAM; |
| 1582 | return 0; |
| 1583 | } |
| 1584 | |
| 1585 | dec->io_ = io; |
| 1586 | dec->status_ = VP8_STATUS_OK; |
| 1587 | VP8LInitBitReader(&dec->br_, io->data, io->data_size); |
| 1588 | if (!ReadImageInfo(&dec->br_, &width, &height, &has_alpha)) { |
| 1589 | dec->status_ = VP8_STATUS_BITSTREAM_ERROR; |
| 1590 | goto Error; |
| 1591 | } |
| 1592 | dec->state_ = READ_DIM; |
| 1593 | io->width = width; |
| 1594 | io->height = height; |
| 1595 | |
| 1596 | if (!DecodeImageStream(width, height, 1, dec, NULL)) goto Error; |
| 1597 | return 1; |
| 1598 | |
| 1599 | Error: |
| 1600 | VP8LClear(dec); |
| 1601 | assert(dec->status_ != VP8_STATUS_OK); |
| 1602 | return 0; |
| 1603 | } |
| 1604 | |
| 1605 | int VP8LDecodeImage(VP8LDecoder* const dec) { |
| 1606 | VP8Io* io = NULL; |
| 1607 | WebPDecParams* params = NULL; |
| 1608 | |
| 1609 | // Sanity checks. |
| 1610 | if (dec == NULL) return 0; |
| 1611 | |
| 1612 | assert(dec->hdr_.huffman_tables_ != NULL); |
| 1613 | assert(dec->hdr_.htree_groups_ != NULL); |
| 1614 | assert(dec->hdr_.num_htree_groups_ > 0); |
| 1615 | |
| 1616 | io = dec->io_; |
| 1617 | assert(io != NULL); |
| 1618 | params = (WebPDecParams*)io->opaque; |
| 1619 | assert(params != NULL); |
| 1620 | |
| 1621 | // Initialization. |
| 1622 | if (dec->state_ != READ_DATA) { |
| 1623 | dec->output_ = params->output; |
| 1624 | assert(dec->output_ != NULL); |
| 1625 | |
| 1626 | if (!WebPIoInitFromOptions(params->options, io, MODE_BGRA)) { |
| 1627 | dec->status_ = VP8_STATUS_INVALID_PARAM; |
| 1628 | goto Err; |
| 1629 | } |
| 1630 | |
| 1631 | if (!AllocateInternalBuffers32b(dec, io->width)) goto Err; |
| 1632 | |
| 1633 | if (io->use_scaling && !AllocateAndInitRescaler(dec, io)) goto Err; |
| 1634 | |
| 1635 | if (io->use_scaling || WebPIsPremultipliedMode(dec->output_->colorspace)) { |
| 1636 | // need the alpha-multiply functions for premultiplied output or rescaling |
| 1637 | WebPInitAlphaProcessing(); |
| 1638 | } |
| 1639 | if (!WebPIsRGBMode(dec->output_->colorspace)) { |
| 1640 | WebPInitConvertARGBToYUV(); |
| 1641 | if (dec->output_->u.YUVA.a != NULL) WebPInitAlphaProcessing(); |
| 1642 | } |
| 1643 | if (dec->incremental_) { |
| 1644 | if (dec->hdr_.color_cache_size_ > 0 && |
| 1645 | dec->hdr_.saved_color_cache_.colors_ == NULL) { |
| 1646 | if (!VP8LColorCacheInit(&dec->hdr_.saved_color_cache_, |
| 1647 | dec->hdr_.color_cache_.hash_bits_)) { |
| 1648 | dec->status_ = VP8_STATUS_OUT_OF_MEMORY; |
| 1649 | goto Err; |
| 1650 | } |
| 1651 | } |
| 1652 | } |
| 1653 | dec->state_ = READ_DATA; |
| 1654 | } |
| 1655 | |
| 1656 | // Decode. |
| 1657 | if (!DecodeImageData(dec, dec->pixels_, dec->width_, dec->height_, |
| 1658 | io->crop_bottom, ProcessRows)) { |
| 1659 | goto Err; |
| 1660 | } |
| 1661 | |
| 1662 | params->last_y = dec->last_out_row_; |
| 1663 | return 1; |
| 1664 | |
| 1665 | Err: |
| 1666 | VP8LClear(dec); |
| 1667 | assert(dec->status_ != VP8_STATUS_OK); |
| 1668 | return 0; |
| 1669 | } |
| 1670 | |
| 1671 | //------------------------------------------------------------------------------ |
| 1672 | |