1/* 2 * WebP (.webp) image decoder 3 * Copyright (c) 2013 Aneesh Dogra <aneesh@sugarlabs.org> 4 * Copyright (c) 2013 Justin Ruggles <justin.ruggles@gmail.com> 5 * 6 * This file is part of FFmpeg. 7 * 8 * FFmpeg is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU Lesser General Public 10 * License as published by the Free Software Foundation; either 11 * version 2.1 of the License, or (at your option) any later version. 12 * 13 * FFmpeg is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 16 * Lesser General Public License for more details. 17 * 18 * You should have received a copy of the GNU Lesser General Public 19 * License along with FFmpeg; if not, write to the Free Software 20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 21 */ 22 23/** 24 * @file 25 * WebP image decoder 26 * 27 * @author Aneesh Dogra <aneesh@sugarlabs.org> 28 * Container and Lossy decoding 29 * 30 * @author Justin Ruggles <justin.ruggles@gmail.com> 31 * Lossless decoder 32 * Compressed alpha for lossy 33 * 34 * @author James Almer <jamrial@gmail.com> 35 * Exif metadata 36 * 37 * Unimplemented: 38 * - Animation 39 * - ICC profile 40 * - XMP metadata 41 */ 42 43#define BITSTREAM_READER_LE 44#include "libavutil/imgutils.h" 45#include "avcodec.h" 46#include "bytestream.h" 47#include "exif.h" 48#include "internal.h" 49#include "get_bits.h" 50#include "thread.h" 51#include "vp8.h" 52 53#define VP8X_FLAG_ANIMATION 0x02 54#define VP8X_FLAG_XMP_METADATA 0x04 55#define VP8X_FLAG_EXIF_METADATA 0x08 56#define VP8X_FLAG_ALPHA 0x10 57#define VP8X_FLAG_ICC 0x20 58 59#define MAX_PALETTE_SIZE 256 60#define MAX_CACHE_BITS 11 61#define NUM_CODE_LENGTH_CODES 19 62#define HUFFMAN_CODES_PER_META_CODE 5 63#define NUM_LITERAL_CODES 256 64#define NUM_LENGTH_CODES 24 65#define NUM_DISTANCE_CODES 40 66#define NUM_SHORT_DISTANCES 120 67#define MAX_HUFFMAN_CODE_LENGTH 15 68 69static const uint16_t alphabet_sizes[HUFFMAN_CODES_PER_META_CODE] = { 70 NUM_LITERAL_CODES + NUM_LENGTH_CODES, 71 NUM_LITERAL_CODES, NUM_LITERAL_CODES, NUM_LITERAL_CODES, 72 NUM_DISTANCE_CODES 73}; 74 75static const uint8_t code_length_code_order[NUM_CODE_LENGTH_CODES] = { 76 17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 77}; 78 79static const int8_t lz77_distance_offsets[NUM_SHORT_DISTANCES][2] = { 80 { 0, 1 }, { 1, 0 }, { 1, 1 }, { -1, 1 }, { 0, 2 }, { 2, 0 }, { 1, 2 }, { -1, 2 }, 81 { 2, 1 }, { -2, 1 }, { 2, 2 }, { -2, 2 }, { 0, 3 }, { 3, 0 }, { 1, 3 }, { -1, 3 }, 82 { 3, 1 }, { -3, 1 }, { 2, 3 }, { -2, 3 }, { 3, 2 }, { -3, 2 }, { 0, 4 }, { 4, 0 }, 83 { 1, 4 }, { -1, 4 }, { 4, 1 }, { -4, 1 }, { 3, 3 }, { -3, 3 }, { 2, 4 }, { -2, 4 }, 84 { 4, 2 }, { -4, 2 }, { 0, 5 }, { 3, 4 }, { -3, 4 }, { 4, 3 }, { -4, 3 }, { 5, 0 }, 85 { 1, 5 }, { -1, 5 }, { 5, 1 }, { -5, 1 }, { 2, 5 }, { -2, 5 }, { 5, 2 }, { -5, 2 }, 86 { 4, 4 }, { -4, 4 }, { 3, 5 }, { -3, 5 }, { 5, 3 }, { -5, 3 }, { 0, 6 }, { 6, 0 }, 87 { 1, 6 }, { -1, 6 }, { 6, 1 }, { -6, 1 }, { 2, 6 }, { -2, 6 }, { 6, 2 }, { -6, 2 }, 88 { 4, 5 }, { -4, 5 }, { 5, 4 }, { -5, 4 }, { 3, 6 }, { -3, 6 }, { 6, 3 }, { -6, 3 }, 89 { 0, 7 }, { 7, 0 }, { 1, 7 }, { -1, 7 }, { 5, 5 }, { -5, 5 }, { 7, 1 }, { -7, 1 }, 90 { 4, 6 }, { -4, 6 }, { 6, 4 }, { -6, 4 }, { 2, 7 }, { -2, 7 }, { 7, 2 }, { -7, 2 }, 91 { 3, 7 }, { -3, 7 }, { 7, 3 }, { -7, 3 }, { 5, 6 }, { -5, 6 }, { 6, 5 }, { -6, 5 }, 92 { 8, 0 }, { 4, 7 }, { -4, 7 }, { 7, 4 }, { -7, 4 }, { 8, 1 }, { 8, 2 }, { 6, 6 }, 93 { -6, 6 }, { 8, 3 }, { 5, 7 }, { -5, 7 }, { 7, 5 }, { -7, 5 }, { 8, 4 }, { 6, 7 }, 94 { -6, 7 }, { 7, 6 }, { -7, 6 }, { 8, 5 }, { 7, 7 }, { -7, 7 }, { 8, 6 }, { 8, 7 } 95}; 96 97enum AlphaCompression { 98 ALPHA_COMPRESSION_NONE, 99 ALPHA_COMPRESSION_VP8L, 100}; 101 102enum AlphaFilter { 103 ALPHA_FILTER_NONE, 104 ALPHA_FILTER_HORIZONTAL, 105 ALPHA_FILTER_VERTICAL, 106 ALPHA_FILTER_GRADIENT, 107}; 108 109enum TransformType { 110 PREDICTOR_TRANSFORM = 0, 111 COLOR_TRANSFORM = 1, 112 SUBTRACT_GREEN = 2, 113 COLOR_INDEXING_TRANSFORM = 3, 114}; 115 116enum PredictionMode { 117 PRED_MODE_BLACK, 118 PRED_MODE_L, 119 PRED_MODE_T, 120 PRED_MODE_TR, 121 PRED_MODE_TL, 122 PRED_MODE_AVG_T_AVG_L_TR, 123 PRED_MODE_AVG_L_TL, 124 PRED_MODE_AVG_L_T, 125 PRED_MODE_AVG_TL_T, 126 PRED_MODE_AVG_T_TR, 127 PRED_MODE_AVG_AVG_L_TL_AVG_T_TR, 128 PRED_MODE_SELECT, 129 PRED_MODE_ADD_SUBTRACT_FULL, 130 PRED_MODE_ADD_SUBTRACT_HALF, 131}; 132 133enum HuffmanIndex { 134 HUFF_IDX_GREEN = 0, 135 HUFF_IDX_RED = 1, 136 HUFF_IDX_BLUE = 2, 137 HUFF_IDX_ALPHA = 3, 138 HUFF_IDX_DIST = 4 139}; 140 141/* The structure of WebP lossless is an optional series of transformation data, 142 * followed by the primary image. The primary image also optionally contains 143 * an entropy group mapping if there are multiple entropy groups. There is a 144 * basic image type called an "entropy coded image" that is used for all of 145 * these. The type of each entropy coded image is referred to by the 146 * specification as its role. */ 147enum ImageRole { 148 /* Primary Image: Stores the actual pixels of the image. */ 149 IMAGE_ROLE_ARGB, 150 151 /* Entropy Image: Defines which Huffman group to use for different areas of 152 * the primary image. */ 153 IMAGE_ROLE_ENTROPY, 154 155 /* Predictors: Defines which predictor type to use for different areas of 156 * the primary image. */ 157 IMAGE_ROLE_PREDICTOR, 158 159 /* Color Transform Data: Defines the color transformation for different 160 * areas of the primary image. */ 161 IMAGE_ROLE_COLOR_TRANSFORM, 162 163 /* Color Index: Stored as an image of height == 1. */ 164 IMAGE_ROLE_COLOR_INDEXING, 165 166 IMAGE_ROLE_NB, 167}; 168 169typedef struct HuffReader { 170 VLC vlc; /* Huffman decoder context */ 171 int simple; /* whether to use simple mode */ 172 int nb_symbols; /* number of coded symbols */ 173 uint16_t simple_symbols[2]; /* symbols for simple mode */ 174} HuffReader; 175 176typedef struct ImageContext { 177 enum ImageRole role; /* role of this image */ 178 AVFrame *frame; /* AVFrame for data */ 179 int color_cache_bits; /* color cache size, log2 */ 180 uint32_t *color_cache; /* color cache data */ 181 int nb_huffman_groups; /* number of huffman groups */ 182 HuffReader *huffman_groups; /* reader for each huffman group */ 183 int size_reduction; /* relative size compared to primary image, log2 */ 184 int is_alpha_primary; 185} ImageContext; 186 187typedef struct WebPContext { 188 VP8Context v; /* VP8 Context used for lossy decoding */ 189 GetBitContext gb; /* bitstream reader for main image chunk */ 190 AVFrame *alpha_frame; /* AVFrame for alpha data decompressed from VP8L */ 191 AVCodecContext *avctx; /* parent AVCodecContext */ 192 int initialized; /* set once the VP8 context is initialized */ 193 int has_alpha; /* has a separate alpha chunk */ 194 enum AlphaCompression alpha_compression; /* compression type for alpha chunk */ 195 enum AlphaFilter alpha_filter; /* filtering method for alpha chunk */ 196 uint8_t *alpha_data; /* alpha chunk data */ 197 int alpha_data_size; /* alpha chunk data size */ 198 int has_exif; /* set after an EXIF chunk has been processed */ 199 AVDictionary *exif_metadata; /* EXIF chunk data */ 200 int width; /* image width */ 201 int height; /* image height */ 202 int lossless; /* indicates lossless or lossy */ 203 204 int nb_transforms; /* number of transforms */ 205 enum TransformType transforms[4]; /* transformations used in the image, in order */ 206 int reduced_width; /* reduced width for index image, if applicable */ 207 int nb_huffman_groups; /* number of huffman groups in the primary image */ 208 ImageContext image[IMAGE_ROLE_NB]; /* image context for each role */ 209} WebPContext; 210 211#define GET_PIXEL(frame, x, y) \ 212 ((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x)) 213 214#define GET_PIXEL_COMP(frame, x, y, c) \ 215 (*((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x) + c)) 216 217static void image_ctx_free(ImageContext *img) 218{ 219 int i, j; 220 221 av_free(img->color_cache); 222 if (img->role != IMAGE_ROLE_ARGB && !img->is_alpha_primary) 223 av_frame_free(&img->frame); 224 if (img->huffman_groups) { 225 for (i = 0; i < img->nb_huffman_groups; i++) { 226 for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++) 227 ff_free_vlc(&img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE + j].vlc); 228 } 229 av_free(img->huffman_groups); 230 } 231 memset(img, 0, sizeof(*img)); 232} 233 234 235/* Differs from get_vlc2() in the following ways: 236 * - codes are bit-reversed 237 * - assumes 8-bit table to make reversal simpler 238 * - assumes max depth of 2 since the max code length for WebP is 15 239 */ 240static av_always_inline int webp_get_vlc(GetBitContext *gb, VLC_TYPE (*table)[2]) 241{ 242 int n, nb_bits; 243 unsigned int index; 244 int code; 245 246 OPEN_READER(re, gb); 247 UPDATE_CACHE(re, gb); 248 249 index = SHOW_UBITS(re, gb, 8); 250 index = ff_reverse[index]; 251 code = table[index][0]; 252 n = table[index][1]; 253 254 if (n < 0) { 255 LAST_SKIP_BITS(re, gb, 8); 256 UPDATE_CACHE(re, gb); 257 258 nb_bits = -n; 259 260 index = SHOW_UBITS(re, gb, nb_bits); 261 index = (ff_reverse[index] >> (8 - nb_bits)) + code; 262 code = table[index][0]; 263 n = table[index][1]; 264 } 265 SKIP_BITS(re, gb, n); 266 267 CLOSE_READER(re, gb); 268 269 return code; 270} 271 272static int huff_reader_get_symbol(HuffReader *r, GetBitContext *gb) 273{ 274 if (r->simple) { 275 if (r->nb_symbols == 1) 276 return r->simple_symbols[0]; 277 else 278 return r->simple_symbols[get_bits1(gb)]; 279 } else 280 return webp_get_vlc(gb, r->vlc.table); 281} 282 283static int huff_reader_build_canonical(HuffReader *r, int *code_lengths, 284 int alphabet_size) 285{ 286 int len = 0, sym, code = 0, ret; 287 int max_code_length = 0; 288 uint16_t *codes; 289 290 /* special-case 1 symbol since the vlc reader cannot handle it */ 291 for (sym = 0; sym < alphabet_size; sym++) { 292 if (code_lengths[sym] > 0) { 293 len++; 294 code = sym; 295 if (len > 1) 296 break; 297 } 298 } 299 if (len == 1) { 300 r->nb_symbols = 1; 301 r->simple_symbols[0] = code; 302 r->simple = 1; 303 return 0; 304 } 305 306 for (sym = 0; sym < alphabet_size; sym++) 307 max_code_length = FFMAX(max_code_length, code_lengths[sym]); 308 309 if (max_code_length == 0 || max_code_length > MAX_HUFFMAN_CODE_LENGTH) 310 return AVERROR(EINVAL); 311 312 codes = av_malloc_array(alphabet_size, sizeof(*codes)); 313 if (!codes) 314 return AVERROR(ENOMEM); 315 316 code = 0; 317 r->nb_symbols = 0; 318 for (len = 1; len <= max_code_length; len++) { 319 for (sym = 0; sym < alphabet_size; sym++) { 320 if (code_lengths[sym] != len) 321 continue; 322 codes[sym] = code++; 323 r->nb_symbols++; 324 } 325 code <<= 1; 326 } 327 if (!r->nb_symbols) { 328 av_free(codes); 329 return AVERROR_INVALIDDATA; 330 } 331 332 ret = init_vlc(&r->vlc, 8, alphabet_size, 333 code_lengths, sizeof(*code_lengths), sizeof(*code_lengths), 334 codes, sizeof(*codes), sizeof(*codes), 0); 335 if (ret < 0) { 336 av_free(codes); 337 return ret; 338 } 339 r->simple = 0; 340 341 av_free(codes); 342 return 0; 343} 344 345static void read_huffman_code_simple(WebPContext *s, HuffReader *hc) 346{ 347 hc->nb_symbols = get_bits1(&s->gb) + 1; 348 349 if (get_bits1(&s->gb)) 350 hc->simple_symbols[0] = get_bits(&s->gb, 8); 351 else 352 hc->simple_symbols[0] = get_bits1(&s->gb); 353 354 if (hc->nb_symbols == 2) 355 hc->simple_symbols[1] = get_bits(&s->gb, 8); 356 357 hc->simple = 1; 358} 359 360static int read_huffman_code_normal(WebPContext *s, HuffReader *hc, 361 int alphabet_size) 362{ 363 HuffReader code_len_hc = { { 0 }, 0, 0, { 0 } }; 364 int *code_lengths = NULL; 365 int code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 }; 366 int i, symbol, max_symbol, prev_code_len, ret; 367 int num_codes = 4 + get_bits(&s->gb, 4); 368 369 if (num_codes > NUM_CODE_LENGTH_CODES) 370 return AVERROR_INVALIDDATA; 371 372 for (i = 0; i < num_codes; i++) 373 code_length_code_lengths[code_length_code_order[i]] = get_bits(&s->gb, 3); 374 375 ret = huff_reader_build_canonical(&code_len_hc, code_length_code_lengths, 376 NUM_CODE_LENGTH_CODES); 377 if (ret < 0) 378 goto finish; 379 380 code_lengths = av_mallocz_array(alphabet_size, sizeof(*code_lengths)); 381 if (!code_lengths) { 382 ret = AVERROR(ENOMEM); 383 goto finish; 384 } 385 386 if (get_bits1(&s->gb)) { 387 int bits = 2 + 2 * get_bits(&s->gb, 3); 388 max_symbol = 2 + get_bits(&s->gb, bits); 389 if (max_symbol > alphabet_size) { 390 av_log(s->avctx, AV_LOG_ERROR, "max symbol %d > alphabet size %d\n", 391 max_symbol, alphabet_size); 392 ret = AVERROR_INVALIDDATA; 393 goto finish; 394 } 395 } else { 396 max_symbol = alphabet_size; 397 } 398 399 prev_code_len = 8; 400 symbol = 0; 401 while (symbol < alphabet_size) { 402 int code_len; 403 404 if (!max_symbol--) 405 break; 406 code_len = huff_reader_get_symbol(&code_len_hc, &s->gb); 407 if (code_len < 16) { 408 /* Code length code [0..15] indicates literal code lengths. */ 409 code_lengths[symbol++] = code_len; 410 if (code_len) 411 prev_code_len = code_len; 412 } else { 413 int repeat = 0, length = 0; 414 switch (code_len) { 415 case 16: 416 /* Code 16 repeats the previous non-zero value [3..6] times, 417 * i.e., 3 + ReadBits(2) times. If code 16 is used before a 418 * non-zero value has been emitted, a value of 8 is repeated. */ 419 repeat = 3 + get_bits(&s->gb, 2); 420 length = prev_code_len; 421 break; 422 case 17: 423 /* Code 17 emits a streak of zeros [3..10], i.e., 424 * 3 + ReadBits(3) times. */ 425 repeat = 3 + get_bits(&s->gb, 3); 426 break; 427 case 18: 428 /* Code 18 emits a streak of zeros of length [11..138], i.e., 429 * 11 + ReadBits(7) times. */ 430 repeat = 11 + get_bits(&s->gb, 7); 431 break; 432 } 433 if (symbol + repeat > alphabet_size) { 434 av_log(s->avctx, AV_LOG_ERROR, 435 "invalid symbol %d + repeat %d > alphabet size %d\n", 436 symbol, repeat, alphabet_size); 437 ret = AVERROR_INVALIDDATA; 438 goto finish; 439 } 440 while (repeat-- > 0) 441 code_lengths[symbol++] = length; 442 } 443 } 444 445 ret = huff_reader_build_canonical(hc, code_lengths, alphabet_size); 446 447finish: 448 ff_free_vlc(&code_len_hc.vlc); 449 av_free(code_lengths); 450 return ret; 451} 452 453static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role, 454 int w, int h); 455 456#define PARSE_BLOCK_SIZE(w, h) do { \ 457 block_bits = get_bits(&s->gb, 3) + 2; \ 458 blocks_w = FFALIGN((w), 1 << block_bits) >> block_bits; \ 459 blocks_h = FFALIGN((h), 1 << block_bits) >> block_bits; \ 460} while (0) 461 462static int decode_entropy_image(WebPContext *s) 463{ 464 ImageContext *img; 465 int ret, block_bits, width, blocks_w, blocks_h, x, y, max; 466 467 width = s->width; 468 if (s->reduced_width > 0) 469 width = s->reduced_width; 470 471 PARSE_BLOCK_SIZE(width, s->height); 472 473 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ENTROPY, blocks_w, blocks_h); 474 if (ret < 0) 475 return ret; 476 477 img = &s->image[IMAGE_ROLE_ENTROPY]; 478 img->size_reduction = block_bits; 479 480 /* the number of huffman groups is determined by the maximum group number 481 * coded in the entropy image */ 482 max = 0; 483 for (y = 0; y < img->frame->height; y++) { 484 for (x = 0; x < img->frame->width; x++) { 485 int p0 = GET_PIXEL_COMP(img->frame, x, y, 1); 486 int p1 = GET_PIXEL_COMP(img->frame, x, y, 2); 487 int p = p0 << 8 | p1; 488 max = FFMAX(max, p); 489 } 490 } 491 s->nb_huffman_groups = max + 1; 492 493 return 0; 494} 495 496static int parse_transform_predictor(WebPContext *s) 497{ 498 int block_bits, blocks_w, blocks_h, ret; 499 500 PARSE_BLOCK_SIZE(s->width, s->height); 501 502 ret = decode_entropy_coded_image(s, IMAGE_ROLE_PREDICTOR, blocks_w, 503 blocks_h); 504 if (ret < 0) 505 return ret; 506 507 s->image[IMAGE_ROLE_PREDICTOR].size_reduction = block_bits; 508 509 return 0; 510} 511 512static int parse_transform_color(WebPContext *s) 513{ 514 int block_bits, blocks_w, blocks_h, ret; 515 516 PARSE_BLOCK_SIZE(s->width, s->height); 517 518 ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_TRANSFORM, blocks_w, 519 blocks_h); 520 if (ret < 0) 521 return ret; 522 523 s->image[IMAGE_ROLE_COLOR_TRANSFORM].size_reduction = block_bits; 524 525 return 0; 526} 527 528static int parse_transform_color_indexing(WebPContext *s) 529{ 530 ImageContext *img; 531 int width_bits, index_size, ret, x; 532 uint8_t *ct; 533 534 index_size = get_bits(&s->gb, 8) + 1; 535 536 if (index_size <= 2) 537 width_bits = 3; 538 else if (index_size <= 4) 539 width_bits = 2; 540 else if (index_size <= 16) 541 width_bits = 1; 542 else 543 width_bits = 0; 544 545 ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_INDEXING, 546 index_size, 1); 547 if (ret < 0) 548 return ret; 549 550 img = &s->image[IMAGE_ROLE_COLOR_INDEXING]; 551 img->size_reduction = width_bits; 552 if (width_bits > 0) 553 s->reduced_width = (s->width + ((1 << width_bits) - 1)) >> width_bits; 554 555 /* color index values are delta-coded */ 556 ct = img->frame->data[0] + 4; 557 for (x = 4; x < img->frame->width * 4; x++, ct++) 558 ct[0] += ct[-4]; 559 560 return 0; 561} 562 563static HuffReader *get_huffman_group(WebPContext *s, ImageContext *img, 564 int x, int y) 565{ 566 ImageContext *gimg = &s->image[IMAGE_ROLE_ENTROPY]; 567 int group = 0; 568 569 if (gimg->size_reduction > 0) { 570 int group_x = x >> gimg->size_reduction; 571 int group_y = y >> gimg->size_reduction; 572 int g0 = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 1); 573 int g1 = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 2); 574 group = g0 << 8 | g1; 575 } 576 577 return &img->huffman_groups[group * HUFFMAN_CODES_PER_META_CODE]; 578} 579 580static av_always_inline void color_cache_put(ImageContext *img, uint32_t c) 581{ 582 uint32_t cache_idx = (0x1E35A7BD * c) >> (32 - img->color_cache_bits); 583 img->color_cache[cache_idx] = c; 584} 585 586static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role, 587 int w, int h) 588{ 589 ImageContext *img; 590 HuffReader *hg; 591 int i, j, ret, x, y, width; 592 593 img = &s->image[role]; 594 img->role = role; 595 596 if (!img->frame) { 597 img->frame = av_frame_alloc(); 598 if (!img->frame) 599 return AVERROR(ENOMEM); 600 } 601 602 img->frame->format = AV_PIX_FMT_ARGB; 603 img->frame->width = w; 604 img->frame->height = h; 605 606 if (role == IMAGE_ROLE_ARGB && !img->is_alpha_primary) { 607 ThreadFrame pt = { .f = img->frame }; 608 ret = ff_thread_get_buffer(s->avctx, &pt, 0); 609 } else 610 ret = av_frame_get_buffer(img->frame, 1); 611 if (ret < 0) 612 return ret; 613 614 if (get_bits1(&s->gb)) { 615 img->color_cache_bits = get_bits(&s->gb, 4); 616 if (img->color_cache_bits < 1 || img->color_cache_bits > 11) { 617 av_log(s->avctx, AV_LOG_ERROR, "invalid color cache bits: %d\n", 618 img->color_cache_bits); 619 return AVERROR_INVALIDDATA; 620 } 621 img->color_cache = av_mallocz_array(1 << img->color_cache_bits, 622 sizeof(*img->color_cache)); 623 if (!img->color_cache) 624 return AVERROR(ENOMEM); 625 } else { 626 img->color_cache_bits = 0; 627 } 628 629 img->nb_huffman_groups = 1; 630 if (role == IMAGE_ROLE_ARGB && get_bits1(&s->gb)) { 631 ret = decode_entropy_image(s); 632 if (ret < 0) 633 return ret; 634 img->nb_huffman_groups = s->nb_huffman_groups; 635 } 636 img->huffman_groups = av_mallocz_array(img->nb_huffman_groups * 637 HUFFMAN_CODES_PER_META_CODE, 638 sizeof(*img->huffman_groups)); 639 if (!img->huffman_groups) 640 return AVERROR(ENOMEM); 641 642 for (i = 0; i < img->nb_huffman_groups; i++) { 643 hg = &img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE]; 644 for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++) { 645 int alphabet_size = alphabet_sizes[j]; 646 if (!j && img->color_cache_bits > 0) 647 alphabet_size += 1 << img->color_cache_bits; 648 649 if (get_bits1(&s->gb)) { 650 read_huffman_code_simple(s, &hg[j]); 651 } else { 652 ret = read_huffman_code_normal(s, &hg[j], alphabet_size); 653 if (ret < 0) 654 return ret; 655 } 656 } 657 } 658 659 width = img->frame->width; 660 if (role == IMAGE_ROLE_ARGB && s->reduced_width > 0) 661 width = s->reduced_width; 662 663 x = 0; y = 0; 664 while (y < img->frame->height) { 665 int v; 666 667 hg = get_huffman_group(s, img, x, y); 668 v = huff_reader_get_symbol(&hg[HUFF_IDX_GREEN], &s->gb); 669 if (v < NUM_LITERAL_CODES) { 670 /* literal pixel values */ 671 uint8_t *p = GET_PIXEL(img->frame, x, y); 672 p[2] = v; 673 p[1] = huff_reader_get_symbol(&hg[HUFF_IDX_RED], &s->gb); 674 p[3] = huff_reader_get_symbol(&hg[HUFF_IDX_BLUE], &s->gb); 675 p[0] = huff_reader_get_symbol(&hg[HUFF_IDX_ALPHA], &s->gb); 676 if (img->color_cache_bits) 677 color_cache_put(img, AV_RB32(p)); 678 x++; 679 if (x == width) { 680 x = 0; 681 y++; 682 } 683 } else if (v < NUM_LITERAL_CODES + NUM_LENGTH_CODES) { 684 /* LZ77 backwards mapping */ 685 int prefix_code, length, distance, ref_x, ref_y; 686 687 /* parse length and distance */ 688 prefix_code = v - NUM_LITERAL_CODES; 689 if (prefix_code < 4) { 690 length = prefix_code + 1; 691 } else { 692 int extra_bits = (prefix_code - 2) >> 1; 693 int offset = 2 + (prefix_code & 1) << extra_bits; 694 length = offset + get_bits(&s->gb, extra_bits) + 1; 695 } 696 prefix_code = huff_reader_get_symbol(&hg[HUFF_IDX_DIST], &s->gb); 697 if (prefix_code < 4) { 698 distance = prefix_code + 1; 699 } else { 700 int extra_bits = prefix_code - 2 >> 1; 701 int offset = 2 + (prefix_code & 1) << extra_bits; 702 distance = offset + get_bits(&s->gb, extra_bits) + 1; 703 } 704 705 /* find reference location */ 706 if (distance <= NUM_SHORT_DISTANCES) { 707 int xi = lz77_distance_offsets[distance - 1][0]; 708 int yi = lz77_distance_offsets[distance - 1][1]; 709 distance = FFMAX(1, xi + yi * width); 710 } else { 711 distance -= NUM_SHORT_DISTANCES; 712 } 713 ref_x = x; 714 ref_y = y; 715 if (distance <= x) { 716 ref_x -= distance; 717 distance = 0; 718 } else { 719 ref_x = 0; 720 distance -= x; 721 } 722 while (distance >= width) { 723 ref_y--; 724 distance -= width; 725 } 726 if (distance > 0) { 727 ref_x = width - distance; 728 ref_y--; 729 } 730 ref_x = FFMAX(0, ref_x); 731 ref_y = FFMAX(0, ref_y); 732 733 /* copy pixels 734 * source and dest regions can overlap and wrap lines, so just 735 * copy per-pixel */ 736 for (i = 0; i < length; i++) { 737 uint8_t *p_ref = GET_PIXEL(img->frame, ref_x, ref_y); 738 uint8_t *p = GET_PIXEL(img->frame, x, y); 739 740 AV_COPY32(p, p_ref); 741 if (img->color_cache_bits) 742 color_cache_put(img, AV_RB32(p)); 743 x++; 744 ref_x++; 745 if (x == width) { 746 x = 0; 747 y++; 748 } 749 if (ref_x == width) { 750 ref_x = 0; 751 ref_y++; 752 } 753 if (y == img->frame->height || ref_y == img->frame->height) 754 break; 755 } 756 } else { 757 /* read from color cache */ 758 uint8_t *p = GET_PIXEL(img->frame, x, y); 759 int cache_idx = v - (NUM_LITERAL_CODES + NUM_LENGTH_CODES); 760 761 if (!img->color_cache_bits) { 762 av_log(s->avctx, AV_LOG_ERROR, "color cache not found\n"); 763 return AVERROR_INVALIDDATA; 764 } 765 if (cache_idx >= 1 << img->color_cache_bits) { 766 av_log(s->avctx, AV_LOG_ERROR, 767 "color cache index out-of-bounds\n"); 768 return AVERROR_INVALIDDATA; 769 } 770 AV_WB32(p, img->color_cache[cache_idx]); 771 x++; 772 if (x == width) { 773 x = 0; 774 y++; 775 } 776 } 777 } 778 779 return 0; 780} 781 782/* PRED_MODE_BLACK */ 783static void inv_predict_0(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 784 const uint8_t *p_t, const uint8_t *p_tr) 785{ 786 AV_WB32(p, 0xFF000000); 787} 788 789/* PRED_MODE_L */ 790static void inv_predict_1(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 791 const uint8_t *p_t, const uint8_t *p_tr) 792{ 793 AV_COPY32(p, p_l); 794} 795 796/* PRED_MODE_T */ 797static void inv_predict_2(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 798 const uint8_t *p_t, const uint8_t *p_tr) 799{ 800 AV_COPY32(p, p_t); 801} 802 803/* PRED_MODE_TR */ 804static void inv_predict_3(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 805 const uint8_t *p_t, const uint8_t *p_tr) 806{ 807 AV_COPY32(p, p_tr); 808} 809 810/* PRED_MODE_TL */ 811static void inv_predict_4(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 812 const uint8_t *p_t, const uint8_t *p_tr) 813{ 814 AV_COPY32(p, p_tl); 815} 816 817/* PRED_MODE_AVG_T_AVG_L_TR */ 818static void inv_predict_5(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 819 const uint8_t *p_t, const uint8_t *p_tr) 820{ 821 p[0] = p_t[0] + (p_l[0] + p_tr[0] >> 1) >> 1; 822 p[1] = p_t[1] + (p_l[1] + p_tr[1] >> 1) >> 1; 823 p[2] = p_t[2] + (p_l[2] + p_tr[2] >> 1) >> 1; 824 p[3] = p_t[3] + (p_l[3] + p_tr[3] >> 1) >> 1; 825} 826 827/* PRED_MODE_AVG_L_TL */ 828static void inv_predict_6(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 829 const uint8_t *p_t, const uint8_t *p_tr) 830{ 831 p[0] = p_l[0] + p_tl[0] >> 1; 832 p[1] = p_l[1] + p_tl[1] >> 1; 833 p[2] = p_l[2] + p_tl[2] >> 1; 834 p[3] = p_l[3] + p_tl[3] >> 1; 835} 836 837/* PRED_MODE_AVG_L_T */ 838static void inv_predict_7(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 839 const uint8_t *p_t, const uint8_t *p_tr) 840{ 841 p[0] = p_l[0] + p_t[0] >> 1; 842 p[1] = p_l[1] + p_t[1] >> 1; 843 p[2] = p_l[2] + p_t[2] >> 1; 844 p[3] = p_l[3] + p_t[3] >> 1; 845} 846 847/* PRED_MODE_AVG_TL_T */ 848static void inv_predict_8(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 849 const uint8_t *p_t, const uint8_t *p_tr) 850{ 851 p[0] = p_tl[0] + p_t[0] >> 1; 852 p[1] = p_tl[1] + p_t[1] >> 1; 853 p[2] = p_tl[2] + p_t[2] >> 1; 854 p[3] = p_tl[3] + p_t[3] >> 1; 855} 856 857/* PRED_MODE_AVG_T_TR */ 858static void inv_predict_9(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 859 const uint8_t *p_t, const uint8_t *p_tr) 860{ 861 p[0] = p_t[0] + p_tr[0] >> 1; 862 p[1] = p_t[1] + p_tr[1] >> 1; 863 p[2] = p_t[2] + p_tr[2] >> 1; 864 p[3] = p_t[3] + p_tr[3] >> 1; 865} 866 867/* PRED_MODE_AVG_AVG_L_TL_AVG_T_TR */ 868static void inv_predict_10(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 869 const uint8_t *p_t, const uint8_t *p_tr) 870{ 871 p[0] = (p_l[0] + p_tl[0] >> 1) + (p_t[0] + p_tr[0] >> 1) >> 1; 872 p[1] = (p_l[1] + p_tl[1] >> 1) + (p_t[1] + p_tr[1] >> 1) >> 1; 873 p[2] = (p_l[2] + p_tl[2] >> 1) + (p_t[2] + p_tr[2] >> 1) >> 1; 874 p[3] = (p_l[3] + p_tl[3] >> 1) + (p_t[3] + p_tr[3] >> 1) >> 1; 875} 876 877/* PRED_MODE_SELECT */ 878static void inv_predict_11(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 879 const uint8_t *p_t, const uint8_t *p_tr) 880{ 881 int diff = (FFABS(p_l[0] - p_tl[0]) - FFABS(p_t[0] - p_tl[0])) + 882 (FFABS(p_l[1] - p_tl[1]) - FFABS(p_t[1] - p_tl[1])) + 883 (FFABS(p_l[2] - p_tl[2]) - FFABS(p_t[2] - p_tl[2])) + 884 (FFABS(p_l[3] - p_tl[3]) - FFABS(p_t[3] - p_tl[3])); 885 if (diff <= 0) 886 AV_COPY32(p, p_t); 887 else 888 AV_COPY32(p, p_l); 889} 890 891/* PRED_MODE_ADD_SUBTRACT_FULL */ 892static void inv_predict_12(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 893 const uint8_t *p_t, const uint8_t *p_tr) 894{ 895 p[0] = av_clip_uint8(p_l[0] + p_t[0] - p_tl[0]); 896 p[1] = av_clip_uint8(p_l[1] + p_t[1] - p_tl[1]); 897 p[2] = av_clip_uint8(p_l[2] + p_t[2] - p_tl[2]); 898 p[3] = av_clip_uint8(p_l[3] + p_t[3] - p_tl[3]); 899} 900 901static av_always_inline uint8_t clamp_add_subtract_half(int a, int b, int c) 902{ 903 int d = a + b >> 1; 904 return av_clip_uint8(d + (d - c) / 2); 905} 906 907/* PRED_MODE_ADD_SUBTRACT_HALF */ 908static void inv_predict_13(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, 909 const uint8_t *p_t, const uint8_t *p_tr) 910{ 911 p[0] = clamp_add_subtract_half(p_l[0], p_t[0], p_tl[0]); 912 p[1] = clamp_add_subtract_half(p_l[1], p_t[1], p_tl[1]); 913 p[2] = clamp_add_subtract_half(p_l[2], p_t[2], p_tl[2]); 914 p[3] = clamp_add_subtract_half(p_l[3], p_t[3], p_tl[3]); 915} 916 917typedef void (*inv_predict_func)(uint8_t *p, const uint8_t *p_l, 918 const uint8_t *p_tl, const uint8_t *p_t, 919 const uint8_t *p_tr); 920 921static const inv_predict_func inverse_predict[14] = { 922 inv_predict_0, inv_predict_1, inv_predict_2, inv_predict_3, 923 inv_predict_4, inv_predict_5, inv_predict_6, inv_predict_7, 924 inv_predict_8, inv_predict_9, inv_predict_10, inv_predict_11, 925 inv_predict_12, inv_predict_13, 926}; 927 928static void inverse_prediction(AVFrame *frame, enum PredictionMode m, int x, int y) 929{ 930 uint8_t *dec, *p_l, *p_tl, *p_t, *p_tr; 931 uint8_t p[4]; 932 933 dec = GET_PIXEL(frame, x, y); 934 p_l = GET_PIXEL(frame, x - 1, y); 935 p_tl = GET_PIXEL(frame, x - 1, y - 1); 936 p_t = GET_PIXEL(frame, x, y - 1); 937 if (x == frame->width - 1) 938 p_tr = GET_PIXEL(frame, 0, y); 939 else 940 p_tr = GET_PIXEL(frame, x + 1, y - 1); 941 942 inverse_predict[m](p, p_l, p_tl, p_t, p_tr); 943 944 dec[0] += p[0]; 945 dec[1] += p[1]; 946 dec[2] += p[2]; 947 dec[3] += p[3]; 948} 949 950static int apply_predictor_transform(WebPContext *s) 951{ 952 ImageContext *img = &s->image[IMAGE_ROLE_ARGB]; 953 ImageContext *pimg = &s->image[IMAGE_ROLE_PREDICTOR]; 954 int x, y; 955 956 for (y = 0; y < img->frame->height; y++) { 957 for (x = 0; x < img->frame->width; x++) { 958 int tx = x >> pimg->size_reduction; 959 int ty = y >> pimg->size_reduction; 960 enum PredictionMode m = GET_PIXEL_COMP(pimg->frame, tx, ty, 2); 961 962 if (x == 0) { 963 if (y == 0) 964 m = PRED_MODE_BLACK; 965 else 966 m = PRED_MODE_T; 967 } else if (y == 0) 968 m = PRED_MODE_L; 969 970 if (m > 13) { 971 av_log(s->avctx, AV_LOG_ERROR, 972 "invalid predictor mode: %d\n", m); 973 return AVERROR_INVALIDDATA; 974 } 975 inverse_prediction(img->frame, m, x, y); 976 } 977 } 978 return 0; 979} 980 981static av_always_inline uint8_t color_transform_delta(uint8_t color_pred, 982 uint8_t color) 983{ 984 return (int)ff_u8_to_s8(color_pred) * ff_u8_to_s8(color) >> 5; 985} 986 987static int apply_color_transform(WebPContext *s) 988{ 989 ImageContext *img, *cimg; 990 int x, y, cx, cy; 991 uint8_t *p, *cp; 992 993 img = &s->image[IMAGE_ROLE_ARGB]; 994 cimg = &s->image[IMAGE_ROLE_COLOR_TRANSFORM]; 995 996 for (y = 0; y < img->frame->height; y++) { 997 for (x = 0; x < img->frame->width; x++) { 998 cx = x >> cimg->size_reduction; 999 cy = y >> cimg->size_reduction; 1000 cp = GET_PIXEL(cimg->frame, cx, cy); 1001 p = GET_PIXEL(img->frame, x, y); 1002 1003 p[1] += color_transform_delta(cp[3], p[2]); 1004 p[3] += color_transform_delta(cp[2], p[2]) + 1005 color_transform_delta(cp[1], p[1]); 1006 } 1007 } 1008 return 0; 1009} 1010 1011static int apply_subtract_green_transform(WebPContext *s) 1012{ 1013 int x, y; 1014 ImageContext *img = &s->image[IMAGE_ROLE_ARGB]; 1015 1016 for (y = 0; y < img->frame->height; y++) { 1017 for (x = 0; x < img->frame->width; x++) { 1018 uint8_t *p = GET_PIXEL(img->frame, x, y); 1019 p[1] += p[2]; 1020 p[3] += p[2]; 1021 } 1022 } 1023 return 0; 1024} 1025 1026static int apply_color_indexing_transform(WebPContext *s) 1027{ 1028 ImageContext *img; 1029 ImageContext *pal; 1030 int i, x, y; 1031 uint8_t *p; 1032 1033 img = &s->image[IMAGE_ROLE_ARGB]; 1034 pal = &s->image[IMAGE_ROLE_COLOR_INDEXING]; 1035 1036 if (pal->size_reduction > 0) { 1037 GetBitContext gb_g; 1038 uint8_t *line; 1039 int pixel_bits = 8 >> pal->size_reduction; 1040 1041 line = av_malloc(img->frame->linesize[0]); 1042 if (!line) 1043 return AVERROR(ENOMEM); 1044 1045 for (y = 0; y < img->frame->height; y++) { 1046 p = GET_PIXEL(img->frame, 0, y); 1047 memcpy(line, p, img->frame->linesize[0]); 1048 init_get_bits(&gb_g, line, img->frame->linesize[0] * 8); 1049 skip_bits(&gb_g, 16); 1050 i = 0; 1051 for (x = 0; x < img->frame->width; x++) { 1052 p = GET_PIXEL(img->frame, x, y); 1053 p[2] = get_bits(&gb_g, pixel_bits); 1054 i++; 1055 if (i == 1 << pal->size_reduction) { 1056 skip_bits(&gb_g, 24); 1057 i = 0; 1058 } 1059 } 1060 } 1061 av_free(line); 1062 } 1063 1064 for (y = 0; y < img->frame->height; y++) { 1065 for (x = 0; x < img->frame->width; x++) { 1066 p = GET_PIXEL(img->frame, x, y); 1067 i = p[2]; 1068 if (i >= pal->frame->width) { 1069 AV_WB32(p, 0x00000000); 1070 } else { 1071 const uint8_t *pi = GET_PIXEL(pal->frame, i, 0); 1072 AV_COPY32(p, pi); 1073 } 1074 } 1075 } 1076 1077 return 0; 1078} 1079 1080static int vp8_lossless_decode_frame(AVCodecContext *avctx, AVFrame *p, 1081 int *got_frame, uint8_t *data_start, 1082 unsigned int data_size, int is_alpha_chunk) 1083{ 1084 WebPContext *s = avctx->priv_data; 1085 int w, h, ret, i; 1086 1087 if (!is_alpha_chunk) { 1088 s->lossless = 1; 1089 avctx->pix_fmt = AV_PIX_FMT_ARGB; 1090 } 1091 1092 ret = init_get_bits(&s->gb, data_start, data_size * 8); 1093 if (ret < 0) 1094 return ret; 1095 1096 if (!is_alpha_chunk) { 1097 if (get_bits(&s->gb, 8) != 0x2F) { 1098 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless signature\n"); 1099 return AVERROR_INVALIDDATA; 1100 } 1101 1102 w = get_bits(&s->gb, 14) + 1; 1103 h = get_bits(&s->gb, 14) + 1; 1104 if (s->width && s->width != w) { 1105 av_log(avctx, AV_LOG_WARNING, "Width mismatch. %d != %d\n", 1106 s->width, w); 1107 } 1108 s->width = w; 1109 if (s->height && s->height != h) { 1110 av_log(avctx, AV_LOG_WARNING, "Height mismatch. %d != %d\n", 1111 s->width, w); 1112 } 1113 s->height = h; 1114 1115 ret = ff_set_dimensions(avctx, s->width, s->height); 1116 if (ret < 0) 1117 return ret; 1118 1119 s->has_alpha = get_bits1(&s->gb); 1120 1121 if (get_bits(&s->gb, 3) != 0x0) { 1122 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless version\n"); 1123 return AVERROR_INVALIDDATA; 1124 } 1125 } else { 1126 if (!s->width || !s->height) 1127 return AVERROR_BUG; 1128 w = s->width; 1129 h = s->height; 1130 } 1131 1132 /* parse transformations */ 1133 s->nb_transforms = 0; 1134 s->reduced_width = 0; 1135 while (get_bits1(&s->gb)) { 1136 enum TransformType transform = get_bits(&s->gb, 2); 1137 s->transforms[s->nb_transforms++] = transform; 1138 switch (transform) { 1139 case PREDICTOR_TRANSFORM: 1140 ret = parse_transform_predictor(s); 1141 break; 1142 case COLOR_TRANSFORM: 1143 ret = parse_transform_color(s); 1144 break; 1145 case COLOR_INDEXING_TRANSFORM: 1146 ret = parse_transform_color_indexing(s); 1147 break; 1148 } 1149 if (ret < 0) 1150 goto free_and_return; 1151 } 1152 1153 /* decode primary image */ 1154 s->image[IMAGE_ROLE_ARGB].frame = p; 1155 if (is_alpha_chunk) 1156 s->image[IMAGE_ROLE_ARGB].is_alpha_primary = 1; 1157 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ARGB, w, h); 1158 if (ret < 0) 1159 goto free_and_return; 1160 1161 /* apply transformations */ 1162 for (i = s->nb_transforms - 1; i >= 0; i--) { 1163 switch (s->transforms[i]) { 1164 case PREDICTOR_TRANSFORM: 1165 ret = apply_predictor_transform(s); 1166 break; 1167 case COLOR_TRANSFORM: 1168 ret = apply_color_transform(s); 1169 break; 1170 case SUBTRACT_GREEN: 1171 ret = apply_subtract_green_transform(s); 1172 break; 1173 case COLOR_INDEXING_TRANSFORM: 1174 ret = apply_color_indexing_transform(s); 1175 break; 1176 } 1177 if (ret < 0) 1178 goto free_and_return; 1179 } 1180 1181 *got_frame = 1; 1182 p->pict_type = AV_PICTURE_TYPE_I; 1183 p->key_frame = 1; 1184 ret = data_size; 1185 1186free_and_return: 1187 for (i = 0; i < IMAGE_ROLE_NB; i++) 1188 image_ctx_free(&s->image[i]); 1189 1190 return ret; 1191} 1192 1193static void alpha_inverse_prediction(AVFrame *frame, enum AlphaFilter m) 1194{ 1195 int x, y, ls; 1196 uint8_t *dec; 1197 1198 ls = frame->linesize[3]; 1199 1200 /* filter first row using horizontal filter */ 1201 dec = frame->data[3] + 1; 1202 for (x = 1; x < frame->width; x++, dec++) 1203 *dec += *(dec - 1); 1204 1205 /* filter first column using vertical filter */ 1206 dec = frame->data[3] + ls; 1207 for (y = 1; y < frame->height; y++, dec += ls) 1208 *dec += *(dec - ls); 1209 1210 /* filter the rest using the specified filter */ 1211 switch (m) { 1212 case ALPHA_FILTER_HORIZONTAL: 1213 for (y = 1; y < frame->height; y++) { 1214 dec = frame->data[3] + y * ls + 1; 1215 for (x = 1; x < frame->width; x++, dec++) 1216 *dec += *(dec - 1); 1217 } 1218 break; 1219 case ALPHA_FILTER_VERTICAL: 1220 for (y = 1; y < frame->height; y++) { 1221 dec = frame->data[3] + y * ls + 1; 1222 for (x = 1; x < frame->width; x++, dec++) 1223 *dec += *(dec - ls); 1224 } 1225 break; 1226 case ALPHA_FILTER_GRADIENT: 1227 for (y = 1; y < frame->height; y++) { 1228 dec = frame->data[3] + y * ls + 1; 1229 for (x = 1; x < frame->width; x++, dec++) 1230 dec[0] += av_clip_uint8(*(dec - 1) + *(dec - ls) - *(dec - ls - 1)); 1231 } 1232 break; 1233 } 1234} 1235 1236static int vp8_lossy_decode_alpha(AVCodecContext *avctx, AVFrame *p, 1237 uint8_t *data_start, 1238 unsigned int data_size) 1239{ 1240 WebPContext *s = avctx->priv_data; 1241 int x, y, ret; 1242 1243 if (s->alpha_compression == ALPHA_COMPRESSION_NONE) { 1244 GetByteContext gb; 1245 1246 bytestream2_init(&gb, data_start, data_size); 1247 for (y = 0; y < s->height; y++) 1248 bytestream2_get_buffer(&gb, p->data[3] + p->linesize[3] * y, 1249 s->width); 1250 } else if (s->alpha_compression == ALPHA_COMPRESSION_VP8L) { 1251 uint8_t *ap, *pp; 1252 int alpha_got_frame = 0; 1253 1254 s->alpha_frame = av_frame_alloc(); 1255 if (!s->alpha_frame) 1256 return AVERROR(ENOMEM); 1257 1258 ret = vp8_lossless_decode_frame(avctx, s->alpha_frame, &alpha_got_frame, 1259 data_start, data_size, 1); 1260 if (ret < 0) { 1261 av_frame_free(&s->alpha_frame); 1262 return ret; 1263 } 1264 if (!alpha_got_frame) { 1265 av_frame_free(&s->alpha_frame); 1266 return AVERROR_INVALIDDATA; 1267 } 1268 1269 /* copy green component of alpha image to alpha plane of primary image */ 1270 for (y = 0; y < s->height; y++) { 1271 ap = GET_PIXEL(s->alpha_frame, 0, y) + 2; 1272 pp = p->data[3] + p->linesize[3] * y; 1273 for (x = 0; x < s->width; x++) { 1274 *pp = *ap; 1275 pp++; 1276 ap += 4; 1277 } 1278 } 1279 av_frame_free(&s->alpha_frame); 1280 } 1281 1282 /* apply alpha filtering */ 1283 if (s->alpha_filter) 1284 alpha_inverse_prediction(p, s->alpha_filter); 1285 1286 return 0; 1287} 1288 1289static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p, 1290 int *got_frame, uint8_t *data_start, 1291 unsigned int data_size) 1292{ 1293 WebPContext *s = avctx->priv_data; 1294 AVPacket pkt; 1295 int ret; 1296 1297 if (!s->initialized) { 1298 ff_vp8_decode_init(avctx); 1299 s->initialized = 1; 1300 if (s->has_alpha) 1301 avctx->pix_fmt = AV_PIX_FMT_YUVA420P; 1302 } 1303 s->lossless = 0; 1304 1305 if (data_size > INT_MAX) { 1306 av_log(avctx, AV_LOG_ERROR, "unsupported chunk size\n"); 1307 return AVERROR_PATCHWELCOME; 1308 } 1309 1310 av_init_packet(&pkt); 1311 pkt.data = data_start; 1312 pkt.size = data_size; 1313 1314 ret = ff_vp8_decode_frame(avctx, p, got_frame, &pkt); 1315 if (s->has_alpha) { 1316 ret = vp8_lossy_decode_alpha(avctx, p, s->alpha_data, 1317 s->alpha_data_size); 1318 if (ret < 0) 1319 return ret; 1320 } 1321 return ret; 1322} 1323 1324static int webp_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, 1325 AVPacket *avpkt) 1326{ 1327 AVFrame * const p = data; 1328 WebPContext *s = avctx->priv_data; 1329 GetByteContext gb; 1330 int ret; 1331 uint32_t chunk_type, chunk_size; 1332 int vp8x_flags = 0; 1333 1334 s->avctx = avctx; 1335 s->width = 0; 1336 s->height = 0; 1337 *got_frame = 0; 1338 s->has_alpha = 0; 1339 s->has_exif = 0; 1340 bytestream2_init(&gb, avpkt->data, avpkt->size); 1341 1342 if (bytestream2_get_bytes_left(&gb) < 12) 1343 return AVERROR_INVALIDDATA; 1344 1345 if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) { 1346 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n"); 1347 return AVERROR_INVALIDDATA; 1348 } 1349 1350 chunk_size = bytestream2_get_le32(&gb); 1351 if (bytestream2_get_bytes_left(&gb) < chunk_size) 1352 return AVERROR_INVALIDDATA; 1353 1354 if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) { 1355 av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n"); 1356 return AVERROR_INVALIDDATA; 1357 } 1358 1359 av_dict_free(&s->exif_metadata); 1360 while (bytestream2_get_bytes_left(&gb) > 0) { 1361 char chunk_str[5] = { 0 }; 1362 1363 chunk_type = bytestream2_get_le32(&gb); 1364 chunk_size = bytestream2_get_le32(&gb); 1365 if (chunk_size == UINT32_MAX) 1366 return AVERROR_INVALIDDATA; 1367 chunk_size += chunk_size & 1; 1368 1369 if (bytestream2_get_bytes_left(&gb) < chunk_size) 1370 return AVERROR_INVALIDDATA; 1371 1372 switch (chunk_type) { 1373 case MKTAG('V', 'P', '8', ' '): 1374 if (!*got_frame) { 1375 ret = vp8_lossy_decode_frame(avctx, p, got_frame, 1376 avpkt->data + bytestream2_tell(&gb), 1377 chunk_size); 1378 if (ret < 0) 1379 return ret; 1380 } 1381 bytestream2_skip(&gb, chunk_size); 1382 break; 1383 case MKTAG('V', 'P', '8', 'L'): 1384 if (!*got_frame) { 1385 ret = vp8_lossless_decode_frame(avctx, p, got_frame, 1386 avpkt->data + bytestream2_tell(&gb), 1387 chunk_size, 0); 1388 if (ret < 0) 1389 return ret; 1390 } 1391 bytestream2_skip(&gb, chunk_size); 1392 break; 1393 case MKTAG('V', 'P', '8', 'X'): 1394 vp8x_flags = bytestream2_get_byte(&gb); 1395 bytestream2_skip(&gb, 3); 1396 s->width = bytestream2_get_le24(&gb) + 1; 1397 s->height = bytestream2_get_le24(&gb) + 1; 1398 ret = av_image_check_size(s->width, s->height, 0, avctx); 1399 if (ret < 0) 1400 return ret; 1401 break; 1402 case MKTAG('A', 'L', 'P', 'H'): { 1403 int alpha_header, filter_m, compression; 1404 1405 if (!(vp8x_flags & VP8X_FLAG_ALPHA)) { 1406 av_log(avctx, AV_LOG_WARNING, 1407 "ALPHA chunk present, but alpha bit not set in the " 1408 "VP8X header\n"); 1409 } 1410 if (chunk_size == 0) { 1411 av_log(avctx, AV_LOG_ERROR, "invalid ALPHA chunk size\n"); 1412 return AVERROR_INVALIDDATA; 1413 } 1414 alpha_header = bytestream2_get_byte(&gb); 1415 s->alpha_data = avpkt->data + bytestream2_tell(&gb); 1416 s->alpha_data_size = chunk_size - 1; 1417 bytestream2_skip(&gb, s->alpha_data_size); 1418 1419 filter_m = (alpha_header >> 2) & 0x03; 1420 compression = alpha_header & 0x03; 1421 1422 if (compression > ALPHA_COMPRESSION_VP8L) { 1423 av_log(avctx, AV_LOG_VERBOSE, 1424 "skipping unsupported ALPHA chunk\n"); 1425 } else { 1426 s->has_alpha = 1; 1427 s->alpha_compression = compression; 1428 s->alpha_filter = filter_m; 1429 } 1430 1431 break; 1432 } 1433 case MKTAG('E', 'X', 'I', 'F'): { 1434 int le, ifd_offset, exif_offset = bytestream2_tell(&gb); 1435 GetByteContext exif_gb; 1436 1437 if (s->has_exif) { 1438 av_log(avctx, AV_LOG_VERBOSE, "Ignoring extra EXIF chunk\n"); 1439 goto exif_end; 1440 } 1441 if (!(vp8x_flags & VP8X_FLAG_EXIF_METADATA)) 1442 av_log(avctx, AV_LOG_WARNING, 1443 "EXIF chunk present, but Exif bit not set in the " 1444 "VP8X header\n"); 1445 1446 s->has_exif = 1; 1447 bytestream2_init(&exif_gb, avpkt->data + exif_offset, 1448 avpkt->size - exif_offset); 1449 if (ff_tdecode_header(&exif_gb, &le, &ifd_offset) < 0) { 1450 av_log(avctx, AV_LOG_ERROR, "invalid TIFF header " 1451 "in Exif data\n"); 1452 goto exif_end; 1453 } 1454 1455 bytestream2_seek(&exif_gb, ifd_offset, SEEK_SET); 1456 if (avpriv_exif_decode_ifd(avctx, &exif_gb, le, 0, &s->exif_metadata) < 0) { 1457 av_log(avctx, AV_LOG_ERROR, "error decoding Exif data\n"); 1458 goto exif_end; 1459 } 1460 1461 av_dict_copy(avpriv_frame_get_metadatap(data), s->exif_metadata, 0); 1462 1463exif_end: 1464 av_dict_free(&s->exif_metadata); 1465 bytestream2_skip(&gb, chunk_size); 1466 break; 1467 } 1468 case MKTAG('I', 'C', 'C', 'P'): 1469 case MKTAG('A', 'N', 'I', 'M'): 1470 case MKTAG('A', 'N', 'M', 'F'): 1471 case MKTAG('X', 'M', 'P', ' '): 1472 AV_WL32(chunk_str, chunk_type); 1473 av_log(avctx, AV_LOG_VERBOSE, "skipping unsupported chunk: %s\n", 1474 chunk_str); 1475 bytestream2_skip(&gb, chunk_size); 1476 break; 1477 default: 1478 AV_WL32(chunk_str, chunk_type); 1479 av_log(avctx, AV_LOG_VERBOSE, "skipping unknown chunk: %s\n", 1480 chunk_str); 1481 bytestream2_skip(&gb, chunk_size); 1482 break; 1483 } 1484 } 1485 1486 if (!*got_frame) { 1487 av_log(avctx, AV_LOG_ERROR, "image data not found\n"); 1488 return AVERROR_INVALIDDATA; 1489 } 1490 1491 return avpkt->size; 1492} 1493 1494static av_cold int webp_decode_close(AVCodecContext *avctx) 1495{ 1496 WebPContext *s = avctx->priv_data; 1497 1498 if (s->initialized) 1499 return ff_vp8_decode_free(avctx); 1500 1501 return 0; 1502} 1503 1504AVCodec ff_webp_decoder = { 1505 .name = "webp", 1506 .long_name = NULL_IF_CONFIG_SMALL("WebP image"), 1507 .type = AVMEDIA_TYPE_VIDEO, 1508 .id = AV_CODEC_ID_WEBP, 1509 .priv_data_size = sizeof(WebPContext), 1510 .decode = webp_decode_frame, 1511 .close = webp_decode_close, 1512 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS, 1513}; 1514