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