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