?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
avformat/mpeg: demux ivtv captions
[ffmpeg.git] / libavcodec / webp.c
blob9f83b518ad5ad4f860f38cd217c2d839185c7662
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 * ICC profile
37 *
38 * Unimplemented:
39 * - Animation
40 * - XMP metadata
41 */
42
43 #include "libavutil/imgutils.h"
44 #include "libavutil/mem.h"
45
46 #define BITSTREAM_READER_LE
47 #include "avcodec.h"
48 #include "bytestream.h"
49 #include "codec_internal.h"
50 #include "decode.h"
51 #include "exif.h"
52 #include "get_bits.h"
53 #include "thread.h"
54 #include "tiff_common.h"
55 #include "vp8.h"
56
57 #define VP8X_FLAG_ANIMATION 0x02
58 #define VP8X_FLAG_XMP_METADATA 0x04
59 #define VP8X_FLAG_EXIF_METADATA 0x08
60 #define VP8X_FLAG_ALPHA 0x10
61 #define VP8X_FLAG_ICC 0x20
62
63 #define MAX_PALETTE_SIZE 256
64 #define MAX_CACHE_BITS 11
65 #define NUM_CODE_LENGTH_CODES 19
66 #define HUFFMAN_CODES_PER_META_CODE 5
67 #define NUM_LITERAL_CODES 256
68 #define NUM_LENGTH_CODES 24
69 #define NUM_DISTANCE_CODES 40
70 #define NUM_SHORT_DISTANCES 120
71 #define MAX_HUFFMAN_CODE_LENGTH 15
72
73 static const uint16_t alphabet_sizes[HUFFMAN_CODES_PER_META_CODE] = {
74 NUM_LITERAL_CODES + NUM_LENGTH_CODES,
75 NUM_LITERAL_CODES, NUM_LITERAL_CODES, NUM_LITERAL_CODES,
76 NUM_DISTANCE_CODES
77 };
78
79 static const uint8_t code_length_code_order[NUM_CODE_LENGTH_CODES] = {
80 17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
81 };
82
83 static const int8_t lz77_distance_offsets[NUM_SHORT_DISTANCES][2] = {
84 { 0, 1 }, { 1, 0 }, { 1, 1 }, { -1, 1 }, { 0, 2 }, { 2, 0 }, { 1, 2 }, { -1, 2 },
85 { 2, 1 }, { -2, 1 }, { 2, 2 }, { -2, 2 }, { 0, 3 }, { 3, 0 }, { 1, 3 }, { -1, 3 },
86 { 3, 1 }, { -3, 1 }, { 2, 3 }, { -2, 3 }, { 3, 2 }, { -3, 2 }, { 0, 4 }, { 4, 0 },
87 { 1, 4 }, { -1, 4 }, { 4, 1 }, { -4, 1 }, { 3, 3 }, { -3, 3 }, { 2, 4 }, { -2, 4 },
88 { 4, 2 }, { -4, 2 }, { 0, 5 }, { 3, 4 }, { -3, 4 }, { 4, 3 }, { -4, 3 }, { 5, 0 },
89 { 1, 5 }, { -1, 5 }, { 5, 1 }, { -5, 1 }, { 2, 5 }, { -2, 5 }, { 5, 2 }, { -5, 2 },
90 { 4, 4 }, { -4, 4 }, { 3, 5 }, { -3, 5 }, { 5, 3 }, { -5, 3 }, { 0, 6 }, { 6, 0 },
91 { 1, 6 }, { -1, 6 }, { 6, 1 }, { -6, 1 }, { 2, 6 }, { -2, 6 }, { 6, 2 }, { -6, 2 },
92 { 4, 5 }, { -4, 5 }, { 5, 4 }, { -5, 4 }, { 3, 6 }, { -3, 6 }, { 6, 3 }, { -6, 3 },
93 { 0, 7 }, { 7, 0 }, { 1, 7 }, { -1, 7 }, { 5, 5 }, { -5, 5 }, { 7, 1 }, { -7, 1 },
94 { 4, 6 }, { -4, 6 }, { 6, 4 }, { -6, 4 }, { 2, 7 }, { -2, 7 }, { 7, 2 }, { -7, 2 },
95 { 3, 7 }, { -3, 7 }, { 7, 3 }, { -7, 3 }, { 5, 6 }, { -5, 6 }, { 6, 5 }, { -6, 5 },
96 { 8, 0 }, { 4, 7 }, { -4, 7 }, { 7, 4 }, { -7, 4 }, { 8, 1 }, { 8, 2 }, { 6, 6 },
97 { -6, 6 }, { 8, 3 }, { 5, 7 }, { -5, 7 }, { 7, 5 }, { -7, 5 }, { 8, 4 }, { 6, 7 },
98 { -6, 7 }, { 7, 6 }, { -7, 6 }, { 8, 5 }, { 7, 7 }, { -7, 7 }, { 8, 6 }, { 8, 7 }
99 };
100
101 enum AlphaCompression {
102 ALPHA_COMPRESSION_NONE,
103 ALPHA_COMPRESSION_VP8L,
104 };
105
106 enum AlphaFilter {
107 ALPHA_FILTER_NONE,
108 ALPHA_FILTER_HORIZONTAL,
109 ALPHA_FILTER_VERTICAL,
110 ALPHA_FILTER_GRADIENT,
111 };
112
113 enum TransformType {
114 PREDICTOR_TRANSFORM = 0,
115 COLOR_TRANSFORM = 1,
116 SUBTRACT_GREEN = 2,
117 COLOR_INDEXING_TRANSFORM = 3,
118 };
119
120 enum PredictionMode {
121 PRED_MODE_BLACK,
122 PRED_MODE_L,
123 PRED_MODE_T,
124 PRED_MODE_TR,
125 PRED_MODE_TL,
126 PRED_MODE_AVG_T_AVG_L_TR,
127 PRED_MODE_AVG_L_TL,
128 PRED_MODE_AVG_L_T,
129 PRED_MODE_AVG_TL_T,
130 PRED_MODE_AVG_T_TR,
131 PRED_MODE_AVG_AVG_L_TL_AVG_T_TR,
132 PRED_MODE_SELECT,
133 PRED_MODE_ADD_SUBTRACT_FULL,
134 PRED_MODE_ADD_SUBTRACT_HALF,
135 };
136
137 enum HuffmanIndex {
138 HUFF_IDX_GREEN = 0,
139 HUFF_IDX_RED = 1,
140 HUFF_IDX_BLUE = 2,
141 HUFF_IDX_ALPHA = 3,
142 HUFF_IDX_DIST = 4
143 };
144
145 /* The structure of WebP lossless is an optional series of transformation data,
146 * followed by the primary image. The primary image also optionally contains
147 * an entropy group mapping if there are multiple entropy groups. There is a
148 * basic image type called an "entropy coded image" that is used for all of
149 * these. The type of each entropy coded image is referred to by the
150 * specification as its role. */
151 enum ImageRole {
152 /* Primary Image: Stores the actual pixels of the image. */
153 IMAGE_ROLE_ARGB,
154
155 /* Entropy Image: Defines which Huffman group to use for different areas of
156 * the primary image. */
157 IMAGE_ROLE_ENTROPY,
158
159 /* Predictors: Defines which predictor type to use for different areas of
160 * the primary image. */
161 IMAGE_ROLE_PREDICTOR,
162
163 /* Color Transform Data: Defines the color transformation for different
164 * areas of the primary image. */
165 IMAGE_ROLE_COLOR_TRANSFORM,
166
167 /* Color Index: Stored as an image of height == 1. */
168 IMAGE_ROLE_COLOR_INDEXING,
169
170 IMAGE_ROLE_NB,
171 };
172
173 typedef struct HuffReader {
174 VLC vlc; /* Huffman decoder context */
175 int simple; /* whether to use simple mode */
176 int nb_symbols; /* number of coded symbols */
177 uint16_t simple_symbols[2]; /* symbols for simple mode */
178 } HuffReader;
179
180 typedef struct ImageContext {
181 enum ImageRole role; /* role of this image */
182 AVFrame *frame; /* AVFrame for data */
183 int color_cache_bits; /* color cache size, log2 */
184 uint32_t *color_cache; /* color cache data */
185 int nb_huffman_groups; /* number of huffman groups */
186 HuffReader *huffman_groups; /* reader for each huffman group */
187 /* relative size compared to primary image, log2.
188 * for IMAGE_ROLE_COLOR_INDEXING with <= 16 colors, this is log2 of the
189 * number of pixels per byte in the primary image (pixel packing) */
190 int size_reduction;
191 int is_alpha_primary;
192 } ImageContext;
193
194 typedef struct WebPContext {
195 VP8Context v; /* VP8 Context used for lossy decoding */
196 GetBitContext gb; /* bitstream reader for main image chunk */
197 AVFrame *alpha_frame; /* AVFrame for alpha data decompressed from VP8L */
198 AVPacket *pkt; /* AVPacket to be passed to the underlying VP8 decoder */
199 AVCodecContext *avctx; /* parent AVCodecContext */
200 int initialized; /* set once the VP8 context is initialized */
201 int has_alpha; /* has a separate alpha chunk */
202 enum AlphaCompression alpha_compression; /* compression type for alpha chunk */
203 enum AlphaFilter alpha_filter; /* filtering method for alpha chunk */
204 const uint8_t *alpha_data; /* alpha chunk data */
205 int alpha_data_size; /* alpha chunk data size */
206 int has_exif; /* set after an EXIF chunk has been processed */
207 int has_iccp; /* set after an ICCP chunk has been processed */
208 int width; /* image width */
209 int height; /* image height */
210 int lossless; /* indicates lossless or lossy */
211
212 int nb_transforms; /* number of transforms */
213 enum TransformType transforms[4]; /* transformations used in the image, in order */
214 /* reduced width when using a color indexing transform with <= 16 colors (pixel packing)
215 * before pixels are unpacked, or same as width otherwise. */
216 int reduced_width;
217 int nb_huffman_groups; /* number of huffman groups in the primary image */
218 ImageContext image[IMAGE_ROLE_NB]; /* image context for each role */
219 } WebPContext;
220
221 #define GET_PIXEL(frame, x, y) \
222 ((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x))
223
224 #define GET_PIXEL_COMP(frame, x, y, c) \
225 (*((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x) + c))
226
227 static void image_ctx_free(ImageContext *img)
228 {
229 int i, j;
230
231 av_free(img->color_cache);
232 if (img->role != IMAGE_ROLE_ARGB && !img->is_alpha_primary)
233 av_frame_free(&img->frame);
234 if (img->huffman_groups) {
235 for (i = 0; i < img->nb_huffman_groups; i++) {
236 for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++)
237 ff_vlc_free(&img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE + j].vlc);
238 }
239 av_free(img->huffman_groups);
240 }
241 memset(img, 0, sizeof(*img));
242 }
243
244 static int huff_reader_get_symbol(HuffReader *r, GetBitContext *gb)
245 {
246 if (r->simple) {
247 if (r->nb_symbols == 1)
248 return r->simple_symbols[0];
249 else
250 return r->simple_symbols[get_bits1(gb)];
251 } else
252 return get_vlc2(gb, r->vlc.table, 8, 2);
253 }
254
255 static int huff_reader_build_canonical(HuffReader *r, const uint8_t *code_lengths,
256 int alphabet_size)
257 {
258 int len = 0, sym, code = 0, ret;
259 int max_code_length = 0;
260 uint16_t *codes;
261
262 /* special-case 1 symbol since the vlc reader cannot handle it */
263 for (sym = 0; sym < alphabet_size; sym++) {
264 if (code_lengths[sym] > 0) {
265 len++;
266 code = sym;
267 if (len > 1)
268 break;
269 }
270 }
271 if (len == 1) {
272 r->nb_symbols = 1;
273 r->simple_symbols[0] = code;
274 r->simple = 1;
275 return 0;
276 }
277
278 for (sym = 0; sym < alphabet_size; sym++)
279 max_code_length = FFMAX(max_code_length, code_lengths[sym]);
280
281 if (max_code_length == 0 || max_code_length > MAX_HUFFMAN_CODE_LENGTH)
282 return AVERROR(EINVAL);
283
284 codes = av_malloc_array(alphabet_size, sizeof(*codes));
285 if (!codes)
286 return AVERROR(ENOMEM);
287
288 code = 0;
289 r->nb_symbols = 0;
290 for (len = 1; len <= max_code_length; len++) {
291 for (sym = 0; sym < alphabet_size; sym++) {
292 if (code_lengths[sym] != len)
293 continue;
294 codes[sym] = code++;
295 r->nb_symbols++;
296 }
297 code <<= 1;
298 }
299 if (!r->nb_symbols) {
300 av_free(codes);
301 return AVERROR_INVALIDDATA;
302 }
303
304 ret = vlc_init(&r->vlc, 8, alphabet_size,
305 code_lengths, sizeof(*code_lengths), sizeof(*code_lengths),
306 codes, sizeof(*codes), sizeof(*codes), VLC_INIT_OUTPUT_LE);
307 if (ret < 0) {
308 av_free(codes);
309 return ret;
310 }
311 r->simple = 0;
312
313 av_free(codes);
314 return 0;
315 }
316
317 static void read_huffman_code_simple(WebPContext *s, HuffReader *hc)
318 {
319 hc->nb_symbols = get_bits1(&s->gb) + 1;
320
321 if (get_bits1(&s->gb))
322 hc->simple_symbols[0] = get_bits(&s->gb, 8);
323 else
324 hc->simple_symbols[0] = get_bits1(&s->gb);
325
326 if (hc->nb_symbols == 2)
327 hc->simple_symbols[1] = get_bits(&s->gb, 8);
328
329 hc->simple = 1;
330 }
331
332 static int read_huffman_code_normal(WebPContext *s, HuffReader *hc,
333 int alphabet_size)
334 {
335 HuffReader code_len_hc = { { 0 }, 0, 0, { 0 } };
336 uint8_t *code_lengths;
337 uint8_t code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 };
338 int i, symbol, max_symbol, prev_code_len, ret;
339 int num_codes = 4 + get_bits(&s->gb, 4);
340
341 av_assert1(num_codes <= NUM_CODE_LENGTH_CODES);
342
343 for (i = 0; i < num_codes; i++)
344 code_length_code_lengths[code_length_code_order[i]] = get_bits(&s->gb, 3);
345
346 ret = huff_reader_build_canonical(&code_len_hc, code_length_code_lengths,
347 NUM_CODE_LENGTH_CODES);
348 if (ret < 0)
349 return ret;
350
351 code_lengths = av_mallocz(alphabet_size);
352 if (!code_lengths) {
353 ret = AVERROR(ENOMEM);
354 goto finish;
355 }
356
357 if (get_bits1(&s->gb)) {
358 int bits = 2 + 2 * get_bits(&s->gb, 3);
359 max_symbol = 2 + get_bits(&s->gb, bits);
360 if (max_symbol > alphabet_size) {
361 av_log(s->avctx, AV_LOG_ERROR, "max symbol %d > alphabet size %d\n",
362 max_symbol, alphabet_size);
363 ret = AVERROR_INVALIDDATA;
364 goto finish;
365 }
366 } else {
367 max_symbol = alphabet_size;
368 }
369
370 prev_code_len = 8;
371 symbol = 0;
372 while (symbol < alphabet_size) {
373 int code_len;
374
375 if (!max_symbol--)
376 break;
377 code_len = huff_reader_get_symbol(&code_len_hc, &s->gb);
378 if (code_len < 16) {
379 /* Code length code [0..15] indicates literal code lengths. */
380 code_lengths[symbol++] = code_len;
381 if (code_len)
382 prev_code_len = code_len;
383 } else {
384 int repeat = 0, length = 0;
385 switch (code_len) {
386 case 16:
387 /* Code 16 repeats the previous non-zero value [3..6] times,
388 * i.e., 3 + ReadBits(2) times. If code 16 is used before a
389 * non-zero value has been emitted, a value of 8 is repeated. */
390 repeat = 3 + get_bits(&s->gb, 2);
391 length = prev_code_len;
392 break;
393 case 17:
394 /* Code 17 emits a streak of zeros [3..10], i.e.,
395 * 3 + ReadBits(3) times. */
396 repeat = 3 + get_bits(&s->gb, 3);
397 break;
398 case 18:
399 /* Code 18 emits a streak of zeros of length [11..138], i.e.,
400 * 11 + ReadBits(7) times. */
401 repeat = 11 + get_bits(&s->gb, 7);
402 break;
403 }
404 if (symbol + repeat > alphabet_size) {
405 av_log(s->avctx, AV_LOG_ERROR,
406 "invalid symbol %d + repeat %d > alphabet size %d\n",
407 symbol, repeat, alphabet_size);
408 ret = AVERROR_INVALIDDATA;
409 goto finish;
410 }
411 while (repeat-- > 0)
412 code_lengths[symbol++] = length;
413 }
414 }
415
416 ret = huff_reader_build_canonical(hc, code_lengths, alphabet_size);
417
418 finish:
419 ff_vlc_free(&code_len_hc.vlc);
420 av_free(code_lengths);
421 return ret;
422 }
423
424 static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role,
425 int w, int h);
426
427 #define PARSE_BLOCK_SIZE(w, h) do { \
428 block_bits = get_bits(&s->gb, 3) + 2; \
429 blocks_w = FFALIGN((w), 1 << block_bits) >> block_bits; \
430 blocks_h = FFALIGN((h), 1 << block_bits) >> block_bits; \
431 } while (0)
432
433 static int decode_entropy_image(WebPContext *s)
434 {
435 ImageContext *img;
436 int ret, block_bits, blocks_w, blocks_h, x, y, max;
437
438 PARSE_BLOCK_SIZE(s->reduced_width, s->height);
439
440 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ENTROPY, blocks_w, blocks_h);
441 if (ret < 0)
442 return ret;
443
444 img = &s->image[IMAGE_ROLE_ENTROPY];
445 img->size_reduction = block_bits;
446
447 /* the number of huffman groups is determined by the maximum group number
448 * coded in the entropy image */
449 max = 0;
450 for (y = 0; y < img->frame->height; y++) {
451 for (x = 0; x < img->frame->width; x++) {
452 int p0 = GET_PIXEL_COMP(img->frame, x, y, 1);
453 int p1 = GET_PIXEL_COMP(img->frame, x, y, 2);
454 int p = p0 << 8 | p1;
455 max = FFMAX(max, p);
456 }
457 }
458 s->nb_huffman_groups = max + 1;
459
460 return 0;
461 }
462
463 static int parse_transform_predictor(WebPContext *s)
464 {
465 int block_bits, blocks_w, blocks_h, ret;
466
467 PARSE_BLOCK_SIZE(s->reduced_width, s->height);
468
469 ret = decode_entropy_coded_image(s, IMAGE_ROLE_PREDICTOR, blocks_w,
470 blocks_h);
471 if (ret < 0)
472 return ret;
473
474 s->image[IMAGE_ROLE_PREDICTOR].size_reduction = block_bits;
475
476 return 0;
477 }
478
479 static int parse_transform_color(WebPContext *s)
480 {
481 int block_bits, blocks_w, blocks_h, ret;
482
483 PARSE_BLOCK_SIZE(s->reduced_width, s->height);
484
485 ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_TRANSFORM, blocks_w,
486 blocks_h);
487 if (ret < 0)
488 return ret;
489
490 s->image[IMAGE_ROLE_COLOR_TRANSFORM].size_reduction = block_bits;
491
492 return 0;
493 }
494
495 static int parse_transform_color_indexing(WebPContext *s)
496 {
497 ImageContext *img;
498 int width_bits, index_size, ret, x;
499 uint8_t *ct;
500
501 index_size = get_bits(&s->gb, 8) + 1;
502
503 if (index_size <= 2)
504 width_bits = 3;
505 else if (index_size <= 4)
506 width_bits = 2;
507 else if (index_size <= 16)
508 width_bits = 1;
509 else
510 width_bits = 0;
511
512 ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_INDEXING,
513 index_size, 1);
514 if (ret < 0)
515 return ret;
516
517 img = &s->image[IMAGE_ROLE_COLOR_INDEXING];
518 img->size_reduction = width_bits;
519 if (width_bits > 0)
520 s->reduced_width = (s->width + ((1 << width_bits) - 1)) >> width_bits;
521
522 /* color index values are delta-coded */
523 ct = img->frame->data[0] + 4;
524 for (x = 4; x < img->frame->width * 4; x++, ct++)
525 ct[0] += ct[-4];
526
527 return 0;
528 }
529
530 static HuffReader *get_huffman_group(WebPContext *s, ImageContext *img,
531 int x, int y)
532 {
533 ImageContext *gimg = &s->image[IMAGE_ROLE_ENTROPY];
534 int group = 0;
535
536 if (gimg->size_reduction > 0) {
537 int group_x = x >> gimg->size_reduction;
538 int group_y = y >> gimg->size_reduction;
539 int g0 = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 1);
540 int g1 = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 2);
541 group = g0 << 8 | g1;
542 }
543
544 return &img->huffman_groups[group * HUFFMAN_CODES_PER_META_CODE];
545 }
546
547 static av_always_inline void color_cache_put(ImageContext *img, uint32_t c)
548 {
549 uint32_t cache_idx = (0x1E35A7BD * c) >> (32 - img->color_cache_bits);
550 img->color_cache[cache_idx] = c;
551 }
552
553 static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role,
554 int w, int h)
555 {
556 ImageContext *img;
557 HuffReader *hg;
558 int i, j, ret, x, y, width;
559
560 img = &s->image[role];
561 img->role = role;
562
563 if (!img->frame) {
564 img->frame = av_frame_alloc();
565 if (!img->frame)
566 return AVERROR(ENOMEM);
567 }
568
569 img->frame->format = AV_PIX_FMT_ARGB;
570 img->frame->width = w;
571 img->frame->height = h;
572
573 if (role == IMAGE_ROLE_ARGB && !img->is_alpha_primary) {
574 ret = ff_thread_get_buffer(s->avctx, img->frame, 0);
575 } else
576 ret = av_frame_get_buffer(img->frame, 1);
577 if (ret < 0)
578 return ret;
579
580 if (get_bits1(&s->gb)) {
581 img->color_cache_bits = get_bits(&s->gb, 4);
582 if (img->color_cache_bits < 1 || img->color_cache_bits > 11) {
583 av_log(s->avctx, AV_LOG_ERROR, "invalid color cache bits: %d\n",
584 img->color_cache_bits);
585 return AVERROR_INVALIDDATA;
586 }
587 img->color_cache = av_calloc(1 << img->color_cache_bits,
588 sizeof(*img->color_cache));
589 if (!img->color_cache)
590 return AVERROR(ENOMEM);
591 } else {
592 img->color_cache_bits = 0;
593 }
594
595 img->nb_huffman_groups = 1;
596 if (role == IMAGE_ROLE_ARGB && get_bits1(&s->gb)) {
597 ret = decode_entropy_image(s);
598 if (ret < 0)
599 return ret;
600 img->nb_huffman_groups = s->nb_huffman_groups;
601 }
602 img->huffman_groups = av_calloc(img->nb_huffman_groups,
603 HUFFMAN_CODES_PER_META_CODE *
604 sizeof(*img->huffman_groups));
605 if (!img->huffman_groups)
606 return AVERROR(ENOMEM);
607
608 for (i = 0; i < img->nb_huffman_groups; i++) {
609 hg = &img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE];
610 for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++) {
611 int alphabet_size = alphabet_sizes[j];
612 if (!j && img->color_cache_bits > 0)
613 alphabet_size += 1 << img->color_cache_bits;
614
615 if (get_bits1(&s->gb)) {
616 read_huffman_code_simple(s, &hg[j]);
617 } else {
618 ret = read_huffman_code_normal(s, &hg[j], alphabet_size);
619 if (ret < 0)
620 return ret;
621 }
622 }
623 }
624
625 width = img->frame->width;
626 if (role == IMAGE_ROLE_ARGB)
627 width = s->reduced_width;
628
629 x = 0; y = 0;
630 while (y < img->frame->height) {
631 int v;
632
633 if (get_bits_left(&s->gb) < 0)
634 return AVERROR_INVALIDDATA;
635
636 hg = get_huffman_group(s, img, x, y);
637 v = huff_reader_get_symbol(&hg[HUFF_IDX_GREEN], &s->gb);
638 if (v < NUM_LITERAL_CODES) {
639 /* literal pixel values */
640 uint8_t *p = GET_PIXEL(img->frame, x, y);
641 p[2] = v;
642 p[1] = huff_reader_get_symbol(&hg[HUFF_IDX_RED], &s->gb);
643 p[3] = huff_reader_get_symbol(&hg[HUFF_IDX_BLUE], &s->gb);
644 p[0] = huff_reader_get_symbol(&hg[HUFF_IDX_ALPHA], &s->gb);
645 if (img->color_cache_bits)
646 color_cache_put(img, AV_RB32(p));
647 x++;
648 if (x == width) {
649 x = 0;
650 y++;
651 }
652 } else if (v < NUM_LITERAL_CODES + NUM_LENGTH_CODES) {
653 /* LZ77 backwards mapping */
654 int prefix_code, length, distance, ref_x, ref_y;
655
656 /* parse length and distance */
657 prefix_code = v - NUM_LITERAL_CODES;
658 if (prefix_code < 4) {
659 length = prefix_code + 1;
660 } else {
661 int extra_bits = (prefix_code - 2) >> 1;
662 int offset = 2 + (prefix_code & 1) << extra_bits;
663 length = offset + get_bits(&s->gb, extra_bits) + 1;
664 }
665 prefix_code = huff_reader_get_symbol(&hg[HUFF_IDX_DIST], &s->gb);
666 if (prefix_code > 39U) {
667 av_log(s->avctx, AV_LOG_ERROR,
668 "distance prefix code too large: %d\n", prefix_code);
669 return AVERROR_INVALIDDATA;
670 }
671 if (prefix_code < 4) {
672 distance = prefix_code + 1;
673 } else {
674 int extra_bits = prefix_code - 2 >> 1;
675 int offset = 2 + (prefix_code & 1) << extra_bits;
676 distance = offset + get_bits(&s->gb, extra_bits) + 1;
677 }
678
679 /* find reference location */
680 if (distance <= NUM_SHORT_DISTANCES) {
681 int xi = lz77_distance_offsets[distance - 1][0];
682 int yi = lz77_distance_offsets[distance - 1][1];
683 distance = FFMAX(1, xi + yi * width);
684 } else {
685 distance -= NUM_SHORT_DISTANCES;
686 }
687 ref_x = x;
688 ref_y = y;
689 if (distance <= x) {
690 ref_x -= distance;
691 distance = 0;
692 } else {
693 ref_x = 0;
694 distance -= x;
695 }
696 while (distance >= width) {
697 ref_y--;
698 distance -= width;
699 }
700 if (distance > 0) {
701 ref_x = width - distance;
702 ref_y--;
703 }
704 ref_x = FFMAX(0, ref_x);
705 ref_y = FFMAX(0, ref_y);
706
707 if (ref_y == y && ref_x >= x)
708 return AVERROR_INVALIDDATA;
709
710 /* copy pixels
711 * source and dest regions can overlap and wrap lines, so just
712 * copy per-pixel */
713 for (i = 0; i < length; i++) {
714 uint8_t *p_ref = GET_PIXEL(img->frame, ref_x, ref_y);
715 uint8_t *p = GET_PIXEL(img->frame, x, y);
716
717 AV_COPY32(p, p_ref);
718 if (img->color_cache_bits)
719 color_cache_put(img, AV_RB32(p));
720 x++;
721 ref_x++;
722 if (x == width) {
723 x = 0;
724 y++;
725 }
726 if (ref_x == width) {
727 ref_x = 0;
728 ref_y++;
729 }
730 if (y == img->frame->height || ref_y == img->frame->height)
731 break;
732 }
733 } else {
734 /* read from color cache */
735 uint8_t *p = GET_PIXEL(img->frame, x, y);
736 int cache_idx = v - (NUM_LITERAL_CODES + NUM_LENGTH_CODES);
737
738 if (!img->color_cache_bits) {
739 av_log(s->avctx, AV_LOG_ERROR, "color cache not found\n");
740 return AVERROR_INVALIDDATA;
741 }
742 if (cache_idx >= 1 << img->color_cache_bits) {
743 av_log(s->avctx, AV_LOG_ERROR,
744 "color cache index out-of-bounds\n");
745 return AVERROR_INVALIDDATA;
746 }
747 AV_WB32(p, img->color_cache[cache_idx]);
748 x++;
749 if (x == width) {
750 x = 0;
751 y++;
752 }
753 }
754 }
755
756 return 0;
757 }
758
759 /* PRED_MODE_BLACK */
760 static void inv_predict_0(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
761 const uint8_t *p_t, const uint8_t *p_tr)
762 {
763 AV_WB32(p, 0xFF000000);
764 }
765
766 /* PRED_MODE_L */
767 static void inv_predict_1(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
768 const uint8_t *p_t, const uint8_t *p_tr)
769 {
770 AV_COPY32(p, p_l);
771 }
772
773 /* PRED_MODE_T */
774 static void inv_predict_2(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
775 const uint8_t *p_t, const uint8_t *p_tr)
776 {
777 AV_COPY32(p, p_t);
778 }
779
780 /* PRED_MODE_TR */
781 static void inv_predict_3(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
782 const uint8_t *p_t, const uint8_t *p_tr)
783 {
784 AV_COPY32(p, p_tr);
785 }
786
787 /* PRED_MODE_TL */
788 static void inv_predict_4(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
789 const uint8_t *p_t, const uint8_t *p_tr)
790 {
791 AV_COPY32(p, p_tl);
792 }
793
794 /* PRED_MODE_AVG_T_AVG_L_TR */
795 static void inv_predict_5(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
796 const uint8_t *p_t, const uint8_t *p_tr)
797 {
798 p[0] = p_t[0] + (p_l[0] + p_tr[0] >> 1) >> 1;
799 p[1] = p_t[1] + (p_l[1] + p_tr[1] >> 1) >> 1;
800 p[2] = p_t[2] + (p_l[2] + p_tr[2] >> 1) >> 1;
801 p[3] = p_t[3] + (p_l[3] + p_tr[3] >> 1) >> 1;
802 }
803
804 /* PRED_MODE_AVG_L_TL */
805 static void inv_predict_6(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
806 const uint8_t *p_t, const uint8_t *p_tr)
807 {
808 p[0] = p_l[0] + p_tl[0] >> 1;
809 p[1] = p_l[1] + p_tl[1] >> 1;
810 p[2] = p_l[2] + p_tl[2] >> 1;
811 p[3] = p_l[3] + p_tl[3] >> 1;
812 }
813
814 /* PRED_MODE_AVG_L_T */
815 static void inv_predict_7(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
816 const uint8_t *p_t, const uint8_t *p_tr)
817 {
818 p[0] = p_l[0] + p_t[0] >> 1;
819 p[1] = p_l[1] + p_t[1] >> 1;
820 p[2] = p_l[2] + p_t[2] >> 1;
821 p[3] = p_l[3] + p_t[3] >> 1;
822 }
823
824 /* PRED_MODE_AVG_TL_T */
825 static void inv_predict_8(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
826 const uint8_t *p_t, const uint8_t *p_tr)
827 {
828 p[0] = p_tl[0] + p_t[0] >> 1;
829 p[1] = p_tl[1] + p_t[1] >> 1;
830 p[2] = p_tl[2] + p_t[2] >> 1;
831 p[3] = p_tl[3] + p_t[3] >> 1;
832 }
833
834 /* PRED_MODE_AVG_T_TR */
835 static void inv_predict_9(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
836 const uint8_t *p_t, const uint8_t *p_tr)
837 {
838 p[0] = p_t[0] + p_tr[0] >> 1;
839 p[1] = p_t[1] + p_tr[1] >> 1;
840 p[2] = p_t[2] + p_tr[2] >> 1;
841 p[3] = p_t[3] + p_tr[3] >> 1;
842 }
843
844 /* PRED_MODE_AVG_AVG_L_TL_AVG_T_TR */
845 static void inv_predict_10(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
846 const uint8_t *p_t, const uint8_t *p_tr)
847 {
848 p[0] = (p_l[0] + p_tl[0] >> 1) + (p_t[0] + p_tr[0] >> 1) >> 1;
849 p[1] = (p_l[1] + p_tl[1] >> 1) + (p_t[1] + p_tr[1] >> 1) >> 1;
850 p[2] = (p_l[2] + p_tl[2] >> 1) + (p_t[2] + p_tr[2] >> 1) >> 1;
851 p[3] = (p_l[3] + p_tl[3] >> 1) + (p_t[3] + p_tr[3] >> 1) >> 1;
852 }
853
854 /* PRED_MODE_SELECT */
855 static void inv_predict_11(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
856 const uint8_t *p_t, const uint8_t *p_tr)
857 {
858 int diff = (FFABS(p_l[0] - p_tl[0]) - FFABS(p_t[0] - p_tl[0])) +
859 (FFABS(p_l[1] - p_tl[1]) - FFABS(p_t[1] - p_tl[1])) +
860 (FFABS(p_l[2] - p_tl[2]) - FFABS(p_t[2] - p_tl[2])) +
861 (FFABS(p_l[3] - p_tl[3]) - FFABS(p_t[3] - p_tl[3]));
862 if (diff <= 0)
863 AV_COPY32(p, p_t);
864 else
865 AV_COPY32(p, p_l);
866 }
867
868 /* PRED_MODE_ADD_SUBTRACT_FULL */
869 static void inv_predict_12(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
870 const uint8_t *p_t, const uint8_t *p_tr)
871 {
872 p[0] = av_clip_uint8(p_l[0] + p_t[0] - p_tl[0]);
873 p[1] = av_clip_uint8(p_l[1] + p_t[1] - p_tl[1]);
874 p[2] = av_clip_uint8(p_l[2] + p_t[2] - p_tl[2]);
875 p[3] = av_clip_uint8(p_l[3] + p_t[3] - p_tl[3]);
876 }
877
878 static av_always_inline uint8_t clamp_add_subtract_half(int a, int b, int c)
879 {
880 int d = a + b >> 1;
881 return av_clip_uint8(d + (d - c) / 2);
882 }
883
884 /* PRED_MODE_ADD_SUBTRACT_HALF */
885 static void inv_predict_13(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
886 const uint8_t *p_t, const uint8_t *p_tr)
887 {
888 p[0] = clamp_add_subtract_half(p_l[0], p_t[0], p_tl[0]);
889 p[1] = clamp_add_subtract_half(p_l[1], p_t[1], p_tl[1]);
890 p[2] = clamp_add_subtract_half(p_l[2], p_t[2], p_tl[2]);
891 p[3] = clamp_add_subtract_half(p_l[3], p_t[3], p_tl[3]);
892 }
893
894 typedef void (*inv_predict_func)(uint8_t *p, const uint8_t *p_l,
895 const uint8_t *p_tl, const uint8_t *p_t,
896 const uint8_t *p_tr);
897
898 static const inv_predict_func inverse_predict[14] = {
899 inv_predict_0, inv_predict_1, inv_predict_2, inv_predict_3,
900 inv_predict_4, inv_predict_5, inv_predict_6, inv_predict_7,
901 inv_predict_8, inv_predict_9, inv_predict_10, inv_predict_11,
902 inv_predict_12, inv_predict_13,
903 };
904
905 static void inverse_prediction(AVFrame *frame, enum PredictionMode m, int x, int y)
906 {
907 uint8_t *dec, *p_l, *p_tl, *p_t, *p_tr;
908 uint8_t p[4];
909
910 dec = GET_PIXEL(frame, x, y);
911 p_l = GET_PIXEL(frame, x - 1, y);
912 p_tl = GET_PIXEL(frame, x - 1, y - 1);
913 p_t = GET_PIXEL(frame, x, y - 1);
914 if (x == frame->width - 1)
915 p_tr = GET_PIXEL(frame, 0, y);
916 else
917 p_tr = GET_PIXEL(frame, x + 1, y - 1);
918
919 inverse_predict[m](p, p_l, p_tl, p_t, p_tr);
920
921 dec[0] += p[0];
922 dec[1] += p[1];
923 dec[2] += p[2];
924 dec[3] += p[3];
925 }
926
927 static int apply_predictor_transform(WebPContext *s)
928 {
929 ImageContext *img = &s->image[IMAGE_ROLE_ARGB];
930 ImageContext *pimg = &s->image[IMAGE_ROLE_PREDICTOR];
931 int x, y;
932
933 for (y = 0; y < img->frame->height; y++) {
934 for (x = 0; x < s->reduced_width; x++) {
935 int tx = x >> pimg->size_reduction;
936 int ty = y >> pimg->size_reduction;
937 enum PredictionMode m = GET_PIXEL_COMP(pimg->frame, tx, ty, 2);
938
939 if (x == 0) {
940 if (y == 0)
941 m = PRED_MODE_BLACK;
942 else
943 m = PRED_MODE_T;
944 } else if (y == 0)
945 m = PRED_MODE_L;
946
947 if (m > 13) {
948 av_log(s->avctx, AV_LOG_ERROR,
949 "invalid predictor mode: %d\n", m);
950 return AVERROR_INVALIDDATA;
951 }
952 inverse_prediction(img->frame, m, x, y);
953 }
954 }
955 return 0;
956 }
957
958 static av_always_inline uint8_t color_transform_delta(uint8_t color_pred,
959 uint8_t color)
960 {
961 return (int)ff_u8_to_s8(color_pred) * ff_u8_to_s8(color) >> 5;
962 }
963
964 static int apply_color_transform(WebPContext *s)
965 {
966 ImageContext *img, *cimg;
967 int x, y, cx, cy;
968 uint8_t *p, *cp;
969
970 img = &s->image[IMAGE_ROLE_ARGB];
971 cimg = &s->image[IMAGE_ROLE_COLOR_TRANSFORM];
972
973 for (y = 0; y < img->frame->height; y++) {
974 for (x = 0; x < s->reduced_width; x++) {
975 cx = x >> cimg->size_reduction;
976 cy = y >> cimg->size_reduction;
977 cp = GET_PIXEL(cimg->frame, cx, cy);
978 p = GET_PIXEL(img->frame, x, y);
979
980 p[1] += color_transform_delta(cp[3], p[2]);
981 p[3] += color_transform_delta(cp[2], p[2]) +
982 color_transform_delta(cp[1], p[1]);
983 }
984 }
985 return 0;
986 }
987
988 static int apply_subtract_green_transform(WebPContext *s)
989 {
990 int x, y;
991 ImageContext *img = &s->image[IMAGE_ROLE_ARGB];
992
993 for (y = 0; y < img->frame->height; y++) {
994 for (x = 0; x < s->reduced_width; x++) {
995 uint8_t *p = GET_PIXEL(img->frame, x, y);
996 p[1] += p[2];
997 p[3] += p[2];
998 }
999 }
1000 return 0;
1001 }
1002
1003 static int apply_color_indexing_transform(WebPContext *s)
1004 {
1005 ImageContext *img;
1006 ImageContext *pal;
1007 int i, x, y;
1008 uint8_t *p;
1009
1010 img = &s->image[IMAGE_ROLE_ARGB];
1011 pal = &s->image[IMAGE_ROLE_COLOR_INDEXING];
1012
1013 if (pal->size_reduction > 0) { // undo pixel packing
1014 GetBitContext gb_g;
1015 uint8_t *line;
1016 int pixel_bits = 8 >> pal->size_reduction;
1017
1018 line = av_malloc(img->frame->linesize[0] + AV_INPUT_BUFFER_PADDING_SIZE);
1019 if (!line)
1020 return AVERROR(ENOMEM);
1021
1022 for (y = 0; y < img->frame->height; y++) {
1023 p = GET_PIXEL(img->frame, 0, y);
1024 memcpy(line, p, img->frame->linesize[0]);
1025 init_get_bits(&gb_g, line, img->frame->linesize[0] * 8);
1026 skip_bits(&gb_g, 16);
1027 i = 0;
1028 for (x = 0; x < img->frame->width; x++) {
1029 p = GET_PIXEL(img->frame, x, y);
1030 p[2] = get_bits(&gb_g, pixel_bits);
1031 i++;
1032 if (i == 1 << pal->size_reduction) {
1033 skip_bits(&gb_g, 24);
1034 i = 0;
1035 }
1036 }
1037 }
1038 av_free(line);
1039 s->reduced_width = s->width; // we are back to full size
1040 }
1041
1042 // switch to local palette if it's worth initializing it
1043 if (img->frame->height * img->frame->width > 300) {
1044 uint8_t palette[256 * 4];
1045 const int size = pal->frame->width * 4;
1046 av_assert0(size <= 1024U);
1047 memcpy(palette, GET_PIXEL(pal->frame, 0, 0), size); // copy palette
1048 // set extra entries to transparent black
1049 memset(palette + size, 0, 256 * 4 - size);
1050 for (y = 0; y < img->frame->height; y++) {
1051 for (x = 0; x < img->frame->width; x++) {
1052 p = GET_PIXEL(img->frame, x, y);
1053 i = p[2];
1054 AV_COPY32(p, &palette[i * 4]);
1055 }
1056 }
1057 } else {
1058 for (y = 0; y < img->frame->height; y++) {
1059 for (x = 0; x < img->frame->width; x++) {
1060 p = GET_PIXEL(img->frame, x, y);
1061 i = p[2];
1062 if (i >= pal->frame->width) {
1063 AV_WB32(p, 0x00000000);
1064 } else {
1065 const uint8_t *pi = GET_PIXEL(pal->frame, i, 0);
1066 AV_COPY32(p, pi);
1067 }
1068 }
1069 }
1070 }
1071
1072 return 0;
1073 }
1074
1075 static void update_canvas_size(AVCodecContext *avctx, int w, int h)
1076 {
1077 WebPContext *s = avctx->priv_data;
1078 if (s->width && s->width != w) {
1079 av_log(avctx, AV_LOG_WARNING, "Width mismatch. %d != %d\n",
1080 s->width, w);
1081 }
1082 s->width = w;
1083 if (s->height && s->height != h) {
1084 av_log(avctx, AV_LOG_WARNING, "Height mismatch. %d != %d\n",
1085 s->height, h);
1086 }
1087 s->height = h;
1088 }
1089
1090 static int vp8_lossless_decode_frame(AVCodecContext *avctx, AVFrame *p,
1091 int *got_frame, const uint8_t *data_start,
1092 unsigned int data_size, int is_alpha_chunk)
1093 {
1094 WebPContext *s = avctx->priv_data;
1095 int w, h, ret, i, used;
1096
1097 if (!is_alpha_chunk) {
1098 s->lossless = 1;
1099 avctx->pix_fmt = AV_PIX_FMT_ARGB;
1100 }
1101
1102 ret = init_get_bits8(&s->gb, data_start, data_size);
1103 if (ret < 0)
1104 return ret;
1105
1106 if (!is_alpha_chunk) {
1107 if (get_bits(&s->gb, 8) != 0x2F) {
1108 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless signature\n");
1109 return AVERROR_INVALIDDATA;
1110 }
1111
1112 w = get_bits(&s->gb, 14) + 1;
1113 h = get_bits(&s->gb, 14) + 1;
1114
1115 update_canvas_size(avctx, w, h);
1116
1117 ret = ff_set_dimensions(avctx, s->width, s->height);
1118 if (ret < 0)
1119 return ret;
1120
1121 s->has_alpha = get_bits1(&s->gb);
1122
1123 if (get_bits(&s->gb, 3) != 0x0) {
1124 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless version\n");
1125 return AVERROR_INVALIDDATA;
1126 }
1127 } else {
1128 if (!s->width || !s->height)
1129 return AVERROR_BUG;
1130 w = s->width;
1131 h = s->height;
1132 }
1133
1134 /* parse transformations */
1135 s->nb_transforms = 0;
1136 s->reduced_width = s->width;
1137 used = 0;
1138 while (get_bits1(&s->gb)) {
1139 enum TransformType transform = get_bits(&s->gb, 2);
1140 if (used & (1 << transform)) {
1141 av_log(avctx, AV_LOG_ERROR, "Transform %d used more than once\n",
1142 transform);
1143 ret = AVERROR_INVALIDDATA;
1144 goto free_and_return;
1145 }
1146 used |= (1 << transform);
1147 s->transforms[s->nb_transforms++] = transform;
1148 switch (transform) {
1149 case PREDICTOR_TRANSFORM:
1150 ret = parse_transform_predictor(s);
1151 break;
1152 case COLOR_TRANSFORM:
1153 ret = parse_transform_color(s);
1154 break;
1155 case COLOR_INDEXING_TRANSFORM:
1156 ret = parse_transform_color_indexing(s);
1157 break;
1158 }
1159 if (ret < 0)
1160 goto free_and_return;
1161 }
1162
1163 /* decode primary image */
1164 s->image[IMAGE_ROLE_ARGB].frame = p;
1165 if (is_alpha_chunk)
1166 s->image[IMAGE_ROLE_ARGB].is_alpha_primary = 1;
1167 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ARGB, w, h);
1168 if (ret < 0)
1169 goto free_and_return;
1170
1171 /* apply transformations */
1172 for (i = s->nb_transforms - 1; i >= 0; i--) {
1173 switch (s->transforms[i]) {
1174 case PREDICTOR_TRANSFORM:
1175 ret = apply_predictor_transform(s);
1176 break;
1177 case COLOR_TRANSFORM:
1178 ret = apply_color_transform(s);
1179 break;
1180 case SUBTRACT_GREEN:
1181 ret = apply_subtract_green_transform(s);
1182 break;
1183 case COLOR_INDEXING_TRANSFORM:
1184 ret = apply_color_indexing_transform(s);
1185 break;
1186 }
1187 if (ret < 0)
1188 goto free_and_return;
1189 }
1190
1191 *got_frame = 1;
1192 p->pict_type = AV_PICTURE_TYPE_I;
1193 p->flags |= AV_FRAME_FLAG_KEY;
1194 p->flags |= AV_FRAME_FLAG_LOSSLESS;
1195 ret = data_size;
1196
1197 free_and_return:
1198 for (i = 0; i < IMAGE_ROLE_NB; i++)
1199 image_ctx_free(&s->image[i]);
1200
1201 return ret;
1202 }
1203
1204 static void alpha_inverse_prediction(AVFrame *frame, enum AlphaFilter m)
1205 {
1206 int x, y, ls;
1207 uint8_t *dec;
1208
1209 ls = frame->linesize[3];
1210
1211 /* filter first row using horizontal filter */
1212 dec = frame->data[3] + 1;
1213 for (x = 1; x < frame->width; x++, dec++)
1214 *dec += *(dec - 1);
1215
1216 /* filter first column using vertical filter */
1217 dec = frame->data[3] + ls;
1218 for (y = 1; y < frame->height; y++, dec += ls)
1219 *dec += *(dec - ls);
1220
1221 /* filter the rest using the specified filter */
1222 switch (m) {
1223 case ALPHA_FILTER_HORIZONTAL:
1224 for (y = 1; y < frame->height; y++) {
1225 dec = frame->data[3] + y * ls + 1;
1226 for (x = 1; x < frame->width; x++, dec++)
1227 *dec += *(dec - 1);
1228 }
1229 break;
1230 case ALPHA_FILTER_VERTICAL:
1231 for (y = 1; y < frame->height; y++) {
1232 dec = frame->data[3] + y * ls + 1;
1233 for (x = 1; x < frame->width; x++, dec++)
1234 *dec += *(dec - ls);
1235 }
1236 break;
1237 case ALPHA_FILTER_GRADIENT:
1238 for (y = 1; y < frame->height; y++) {
1239 dec = frame->data[3] + y * ls + 1;
1240 for (x = 1; x < frame->width; x++, dec++)
1241 dec[0] += av_clip_uint8(*(dec - 1) + *(dec - ls) - *(dec - ls - 1));
1242 }
1243 break;
1244 }
1245 }
1246
1247 static int vp8_lossy_decode_alpha(AVCodecContext *avctx, AVFrame *p,
1248 const uint8_t *data_start,
1249 unsigned int data_size)
1250 {
1251 WebPContext *s = avctx->priv_data;
1252 int x, y, ret;
1253
1254 if (s->alpha_compression == ALPHA_COMPRESSION_NONE) {
1255 GetByteContext gb;
1256
1257 bytestream2_init(&gb, data_start, data_size);
1258 for (y = 0; y < s->height; y++)
1259 bytestream2_get_buffer(&gb, p->data[3] + p->linesize[3] * y,
1260 s->width);
1261 } else if (s->alpha_compression == ALPHA_COMPRESSION_VP8L) {
1262 uint8_t *ap, *pp;
1263 int alpha_got_frame = 0;
1264
1265 s->alpha_frame = av_frame_alloc();
1266 if (!s->alpha_frame)
1267 return AVERROR(ENOMEM);
1268
1269 ret = vp8_lossless_decode_frame(avctx, s->alpha_frame, &alpha_got_frame,
1270 data_start, data_size, 1);
1271 if (ret < 0) {
1272 av_frame_free(&s->alpha_frame);
1273 return ret;
1274 }
1275 if (!alpha_got_frame) {
1276 av_frame_free(&s->alpha_frame);
1277 return AVERROR_INVALIDDATA;
1278 }
1279
1280 /* copy green component of alpha image to alpha plane of primary image */
1281 for (y = 0; y < s->height; y++) {
1282 ap = GET_PIXEL(s->alpha_frame, 0, y) + 2;
1283 pp = p->data[3] + p->linesize[3] * y;
1284 for (x = 0; x < s->width; x++) {
1285 *pp = *ap;
1286 pp++;
1287 ap += 4;
1288 }
1289 }
1290 av_frame_free(&s->alpha_frame);
1291 }
1292
1293 /* apply alpha filtering */
1294 if (s->alpha_filter)
1295 alpha_inverse_prediction(p, s->alpha_filter);
1296
1297 return 0;
1298 }
1299
1300 static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p,
1301 int *got_frame, uint8_t *data_start,
1302 unsigned int data_size)
1303 {
1304 WebPContext *s = avctx->priv_data;
1305 int ret;
1306
1307 if (!s->initialized) {
1308 ff_vp8_decode_init(avctx);
1309 s->initialized = 1;
1310 s->v.actually_webp = 1;
1311 }
1312 avctx->pix_fmt = s->has_alpha ? AV_PIX_FMT_YUVA420P : AV_PIX_FMT_YUV420P;
1313 s->lossless = 0;
1314
1315 if (data_size > INT_MAX) {
1316 av_log(avctx, AV_LOG_ERROR, "unsupported chunk size\n");
1317 return AVERROR_PATCHWELCOME;
1318 }
1319
1320 av_packet_unref(s->pkt);
1321 s->pkt->data = data_start;
1322 s->pkt->size = data_size;
1323
1324 ret = ff_vp8_decode_frame(avctx, p, got_frame, s->pkt);
1325 if (ret < 0)
1326 return ret;
1327
1328 if (!*got_frame)
1329 return AVERROR_INVALIDDATA;
1330
1331 update_canvas_size(avctx, avctx->width, avctx->height);
1332
1333 if (s->has_alpha) {
1334 ret = vp8_lossy_decode_alpha(avctx, p, s->alpha_data,
1335 s->alpha_data_size);
1336 if (ret < 0)
1337 return ret;
1338 }
1339 return ret;
1340 }
1341
1342 static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
1343 int *got_frame, AVPacket *avpkt)
1344 {
1345 WebPContext *s = avctx->priv_data;
1346 GetByteContext gb;
1347 int ret;
1348 uint32_t chunk_type, chunk_size;
1349 int vp8x_flags = 0;
1350
1351 s->avctx = avctx;
1352 s->width = 0;
1353 s->height = 0;
1354 *got_frame = 0;
1355 s->has_alpha = 0;
1356 s->has_exif = 0;
1357 s->has_iccp = 0;
1358 bytestream2_init(&gb, avpkt->data, avpkt->size);
1359
1360 if (bytestream2_get_bytes_left(&gb) < 12)
1361 return AVERROR_INVALIDDATA;
1362
1363 if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
1364 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
1365 return AVERROR_INVALIDDATA;
1366 }
1367
1368 chunk_size = bytestream2_get_le32(&gb);
1369 if (bytestream2_get_bytes_left(&gb) < chunk_size)
1370 return AVERROR_INVALIDDATA;
1371
1372 if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) {
1373 av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n");
1374 return AVERROR_INVALIDDATA;
1375 }
1376
1377 while (bytestream2_get_bytes_left(&gb) > 8) {
1378 char chunk_str[5] = { 0 };
1379
1380 chunk_type = bytestream2_get_le32(&gb);
1381 chunk_size = bytestream2_get_le32(&gb);
1382 if (chunk_size == UINT32_MAX)
1383 return AVERROR_INVALIDDATA;
1384 chunk_size += chunk_size & 1;
1385
1386 if (bytestream2_get_bytes_left(&gb) < chunk_size) {
1387 /* we seem to be running out of data, but it could also be that the
1388 bitstream has trailing junk leading to bogus chunk_size. */
1389 break;
1390 }
1391
1392 switch (chunk_type) {
1393 case MKTAG('V', 'P', '8', ' '):
1394 if (!*got_frame) {
1395 ret = vp8_lossy_decode_frame(avctx, p, got_frame,
1396 avpkt->data + bytestream2_tell(&gb),
1397 chunk_size);
1398 if (ret < 0)
1399 return ret;
1400 }
1401 bytestream2_skip(&gb, chunk_size);
1402 break;
1403 case MKTAG('V', 'P', '8', 'L'):
1404 if (!*got_frame) {
1405 ret = vp8_lossless_decode_frame(avctx, p, got_frame,
1406 avpkt->data + bytestream2_tell(&gb),
1407 chunk_size, 0);
1408 if (ret < 0)
1409 return ret;
1410 #if FF_API_CODEC_PROPS
1411 FF_DISABLE_DEPRECATION_WARNINGS
1412 avctx->properties |= FF_CODEC_PROPERTY_LOSSLESS;
1413 FF_ENABLE_DEPRECATION_WARNINGS
1414 #endif
1415 }
1416 bytestream2_skip(&gb, chunk_size);
1417 break;
1418 case MKTAG('V', 'P', '8', 'X'):
1419 if (s->width || s->height || *got_frame) {
1420 av_log(avctx, AV_LOG_ERROR, "Canvas dimensions are already set\n");
1421 return AVERROR_INVALIDDATA;
1422 }
1423 vp8x_flags = bytestream2_get_byte(&gb);
1424 bytestream2_skip(&gb, 3);
1425 s->width = bytestream2_get_le24(&gb) + 1;
1426 s->height = bytestream2_get_le24(&gb) + 1;
1427 ret = av_image_check_size(s->width, s->height, 0, avctx);
1428 if (ret < 0)
1429 return ret;
1430 break;
1431 case MKTAG('A', 'L', 'P', 'H'): {
1432 int alpha_header, filter_m, compression;
1433
1434 if (!(vp8x_flags & VP8X_FLAG_ALPHA)) {
1435 av_log(avctx, AV_LOG_WARNING,
1436 "ALPHA chunk present, but alpha bit not set in the "
1437 "VP8X header\n");
1438 }
1439 if (chunk_size == 0) {
1440 av_log(avctx, AV_LOG_ERROR, "invalid ALPHA chunk size\n");
1441 return AVERROR_INVALIDDATA;
1442 }
1443 alpha_header = bytestream2_get_byte(&gb);
1444 s->alpha_data = avpkt->data + bytestream2_tell(&gb);
1445 s->alpha_data_size = chunk_size - 1;
1446 bytestream2_skip(&gb, s->alpha_data_size);
1447
1448 filter_m = (alpha_header >> 2) & 0x03;
1449 compression = alpha_header & 0x03;
1450
1451 if (compression > ALPHA_COMPRESSION_VP8L) {
1452 av_log(avctx, AV_LOG_VERBOSE,
1453 "skipping unsupported ALPHA chunk\n");
1454 } else {
1455 s->has_alpha = 1;
1456 s->alpha_compression = compression;
1457 s->alpha_filter = filter_m;
1458 }
1459
1460 break;
1461 }
1462 case MKTAG('E', 'X', 'I', 'F'): {
1463 int le, ifd_offset, exif_offset = bytestream2_tell(&gb);
1464 AVDictionary *exif_metadata = NULL;
1465 GetByteContext exif_gb;
1466
1467 if (s->has_exif) {
1468 av_log(avctx, AV_LOG_VERBOSE, "Ignoring extra EXIF chunk\n");
1469 goto exif_end;
1470 }
1471 if (!(vp8x_flags & VP8X_FLAG_EXIF_METADATA))
1472 av_log(avctx, AV_LOG_WARNING,
1473 "EXIF chunk present, but Exif bit not set in the "
1474 "VP8X header\n");
1475
1476 s->has_exif = 1;
1477 bytestream2_init(&exif_gb, avpkt->data + exif_offset,
1478 avpkt->size - exif_offset);
1479 if (ff_tdecode_header(&exif_gb, &le, &ifd_offset) < 0) {
1480 av_log(avctx, AV_LOG_ERROR, "invalid TIFF header "
1481 "in Exif data\n");
1482 goto exif_end;
1483 }
1484
1485 bytestream2_seek(&exif_gb, ifd_offset, SEEK_SET);
1486 if (ff_exif_decode_ifd(avctx, &exif_gb, le, 0, &exif_metadata) < 0) {
1487 av_log(avctx, AV_LOG_ERROR, "error decoding Exif data\n");
1488 goto exif_end;
1489 }
1490
1491 av_dict_copy(&p->metadata, exif_metadata, 0);
1492
1493 exif_end:
1494 av_dict_free(&exif_metadata);
1495 bytestream2_skip(&gb, chunk_size);
1496 break;
1497 }
1498 case MKTAG('I', 'C', 'C', 'P'): {
1499 AVFrameSideData *sd;
1500
1501 if (s->has_iccp) {
1502 av_log(avctx, AV_LOG_VERBOSE, "Ignoring extra ICCP chunk\n");
1503 bytestream2_skip(&gb, chunk_size);
1504 break;
1505 }
1506 if (!(vp8x_flags & VP8X_FLAG_ICC))
1507 av_log(avctx, AV_LOG_WARNING,
1508 "ICCP chunk present, but ICC Profile bit not set in the "
1509 "VP8X header\n");
1510
1511 s->has_iccp = 1;
1512
1513 ret = ff_frame_new_side_data(avctx, p, AV_FRAME_DATA_ICC_PROFILE, chunk_size, &sd);
1514 if (ret < 0)
1515 return ret;
1516
1517 if (sd) {
1518 bytestream2_get_buffer(&gb, sd->data, chunk_size);
1519 } else {
1520 bytestream2_skip(&gb, chunk_size);
1521 }
1522 break;
1523 }
1524 case MKTAG('A', 'N', 'I', 'M'):
1525 case MKTAG('A', 'N', 'M', 'F'):
1526 case MKTAG('X', 'M', 'P', ' '):
1527 AV_WL32(chunk_str, chunk_type);
1528 av_log(avctx, AV_LOG_WARNING, "skipping unsupported chunk: %s\n",
1529 chunk_str);
1530 bytestream2_skip(&gb, chunk_size);
1531 break;
1532 default:
1533 AV_WL32(chunk_str, chunk_type);
1534 av_log(avctx, AV_LOG_VERBOSE, "skipping unknown chunk: %s\n",
1535 chunk_str);
1536 bytestream2_skip(&gb, chunk_size);
1537 break;
1538 }
1539 }
1540
1541 if (!*got_frame) {
1542 av_log(avctx, AV_LOG_ERROR, "image data not found\n");
1543 return AVERROR_INVALIDDATA;
1544 }
1545
1546 return avpkt->size;
1547 }
1548
1549 static av_cold int webp_decode_init(AVCodecContext *avctx)
1550 {
1551 WebPContext *s = avctx->priv_data;
1552
1553 s->pkt = av_packet_alloc();
1554 if (!s->pkt)
1555 return AVERROR(ENOMEM);
1556
1557 return 0;
1558 }
1559
1560 static av_cold int webp_decode_close(AVCodecContext *avctx)
1561 {
1562 WebPContext *s = avctx->priv_data;
1563
1564 av_packet_free(&s->pkt);
1565
1566 if (s->initialized)
1567 return ff_vp8_decode_free(avctx);
1568
1569 return 0;
1570 }
1571
1572 const FFCodec ff_webp_decoder = {
1573 .p.name = "webp",
1574 CODEC_LONG_NAME("WebP image"),
1575 .p.type = AVMEDIA_TYPE_VIDEO,
1576 .p.id = AV_CODEC_ID_WEBP,
1577 .priv_data_size = sizeof(WebPContext),
1578 .init = webp_decode_init,
1579 FF_CODEC_DECODE_CB(webp_decode_frame),
1580 .close = webp_decode_close,
1581 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,
1582 .caps_internal = FF_CODEC_CAP_ICC_PROFILES |
1583 FF_CODEC_CAP_USES_PROGRESSFRAMES,
1584 };