Commit | Line | Data |
---|---|---|
c4bfa098 JR |
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 Libav. | |
7 | * | |
8 | * Libav 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 | * Libav 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 Libav; 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 | * Unimplemented: | |
35 | * - Animation | |
36 | * - ICC profile | |
37 | * - Exif and XMP metadata | |
38 | */ | |
39 | ||
40 | #define BITSTREAM_READER_LE | |
41 | #include "libavutil/imgutils.h" | |
42 | #include "avcodec.h" | |
43 | #include "bytestream.h" | |
44 | #include "internal.h" | |
45 | #include "get_bits.h" | |
46 | #include "thread.h" | |
47 | #include "vp8.h" | |
48 | ||
49 | #define VP8X_FLAG_ANIMATION 0x02 | |
50 | #define VP8X_FLAG_XMP_METADATA 0x04 | |
51 | #define VP8X_FLAG_EXIF_METADATA 0x08 | |
52 | #define VP8X_FLAG_ALPHA 0x10 | |
53 | #define VP8X_FLAG_ICC 0x20 | |
54 | ||
55 | #define MAX_PALETTE_SIZE 256 | |
56 | #define MAX_CACHE_BITS 11 | |
57 | #define NUM_CODE_LENGTH_CODES 19 | |
58 | #define HUFFMAN_CODES_PER_META_CODE 5 | |
59 | #define NUM_LITERAL_CODES 256 | |
60 | #define NUM_LENGTH_CODES 24 | |
61 | #define NUM_DISTANCE_CODES 40 | |
62 | #define NUM_SHORT_DISTANCES 120 | |
63 | #define MAX_HUFFMAN_CODE_LENGTH 15 | |
64 | ||
65 | static const uint16_t alphabet_sizes[HUFFMAN_CODES_PER_META_CODE] = { | |
66 | NUM_LITERAL_CODES + NUM_LENGTH_CODES, | |
67 | NUM_LITERAL_CODES, NUM_LITERAL_CODES, NUM_LITERAL_CODES, | |
68 | NUM_DISTANCE_CODES | |
69 | }; | |
70 | ||
71 | static const uint8_t code_length_code_order[NUM_CODE_LENGTH_CODES] = { | |
72 | 17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 | |
73 | }; | |
74 | ||
75 | static const int8_t lz77_distance_offsets[NUM_SHORT_DISTANCES][2] = { | |
76 | { 0, 1 }, { 1, 0 }, { 1, 1 }, { -1, 1 }, { 0, 2 }, { 2, 0 }, { 1, 2 }, { -1, 2 }, | |
77 | { 2, 1 }, { -2, 1 }, { 2, 2 }, { -2, 2 }, { 0, 3 }, { 3, 0 }, { 1, 3 }, { -1, 3 }, | |
78 | { 3, 1 }, { -3, 1 }, { 2, 3 }, { -2, 3 }, { 3, 2 }, { -3, 2 }, { 0, 4 }, { 4, 0 }, | |
79 | { 1, 4 }, { -1, 4 }, { 4, 1 }, { -4, 1 }, { 3, 3 }, { -3, 3 }, { 2, 4 }, { -2, 4 }, | |
80 | { 4, 2 }, { -4, 2 }, { 0, 5 }, { 3, 4 }, { -3, 4 }, { 4, 3 }, { -4, 3 }, { 5, 0 }, | |
81 | { 1, 5 }, { -1, 5 }, { 5, 1 }, { -5, 1 }, { 2, 5 }, { -2, 5 }, { 5, 2 }, { -5, 2 }, | |
82 | { 4, 4 }, { -4, 4 }, { 3, 5 }, { -3, 5 }, { 5, 3 }, { -5, 3 }, { 0, 6 }, { 6, 0 }, | |
83 | { 1, 6 }, { -1, 6 }, { 6, 1 }, { -6, 1 }, { 2, 6 }, { -2, 6 }, { 6, 2 }, { -6, 2 }, | |
84 | { 4, 5 }, { -4, 5 }, { 5, 4 }, { -5, 4 }, { 3, 6 }, { -3, 6 }, { 6, 3 }, { -6, 3 }, | |
85 | { 0, 7 }, { 7, 0 }, { 1, 7 }, { -1, 7 }, { 5, 5 }, { -5, 5 }, { 7, 1 }, { -7, 1 }, | |
86 | { 4, 6 }, { -4, 6 }, { 6, 4 }, { -6, 4 }, { 2, 7 }, { -2, 7 }, { 7, 2 }, { -7, 2 }, | |
87 | { 3, 7 }, { -3, 7 }, { 7, 3 }, { -7, 3 }, { 5, 6 }, { -5, 6 }, { 6, 5 }, { -6, 5 }, | |
88 | { 8, 0 }, { 4, 7 }, { -4, 7 }, { 7, 4 }, { -7, 4 }, { 8, 1 }, { 8, 2 }, { 6, 6 }, | |
89 | { -6, 6 }, { 8, 3 }, { 5, 7 }, { -5, 7 }, { 7, 5 }, { -7, 5 }, { 8, 4 }, { 6, 7 }, | |
90 | { -6, 7 }, { 7, 6 }, { -7, 6 }, { 8, 5 }, { 7, 7 }, { -7, 7 }, { 8, 6 }, { 8, 7 } | |
91 | }; | |
92 | ||
93 | enum AlphaCompression { | |
94 | ALPHA_COMPRESSION_NONE, | |
95 | ALPHA_COMPRESSION_VP8L, | |
96 | }; | |
97 | ||
98 | enum AlphaFilter { | |
99 | ALPHA_FILTER_NONE, | |
100 | ALPHA_FILTER_HORIZONTAL, | |
101 | ALPHA_FILTER_VERTICAL, | |
102 | ALPHA_FILTER_GRADIENT, | |
103 | }; | |
104 | ||
105 | enum TransformType { | |
106 | PREDICTOR_TRANSFORM = 0, | |
107 | COLOR_TRANSFORM = 1, | |
108 | SUBTRACT_GREEN = 2, | |
109 | COLOR_INDEXING_TRANSFORM = 3, | |
110 | }; | |
111 | ||
112 | enum PredictionMode { | |
113 | PRED_MODE_BLACK, | |
114 | PRED_MODE_L, | |
115 | PRED_MODE_T, | |
116 | PRED_MODE_TR, | |
117 | PRED_MODE_TL, | |
118 | PRED_MODE_AVG_T_AVG_L_TR, | |
119 | PRED_MODE_AVG_L_TL, | |
120 | PRED_MODE_AVG_L_T, | |
121 | PRED_MODE_AVG_TL_T, | |
122 | PRED_MODE_AVG_T_TR, | |
123 | PRED_MODE_AVG_AVG_L_TL_AVG_T_TR, | |
124 | PRED_MODE_SELECT, | |
125 | PRED_MODE_ADD_SUBTRACT_FULL, | |
126 | PRED_MODE_ADD_SUBTRACT_HALF, | |
127 | }; | |
128 | ||
129 | enum HuffmanIndex { | |
130 | HUFF_IDX_GREEN = 0, | |
131 | HUFF_IDX_RED = 1, | |
132 | HUFF_IDX_BLUE = 2, | |
133 | HUFF_IDX_ALPHA = 3, | |
134 | HUFF_IDX_DIST = 4 | |
135 | }; | |
136 | ||
137 | /* The structure of WebP lossless is an optional series of transformation data, | |
138 | * followed by the primary image. The primary image also optionally contains | |
139 | * an entropy group mapping if there are multiple entropy groups. There is a | |
140 | * basic image type called an "entropy coded image" that is used for all of | |
141 | * these. The type of each entropy coded image is referred to by the | |
142 | * specification as its role. */ | |
143 | enum ImageRole { | |
144 | /* Primary Image: Stores the actual pixels of the image. */ | |
145 | IMAGE_ROLE_ARGB, | |
146 | ||
147 | /* Entropy Image: Defines which Huffman group to use for different areas of | |
148 | * the primary image. */ | |
149 | IMAGE_ROLE_ENTROPY, | |
150 | ||
151 | /* Predictors: Defines which predictor type to use for different areas of | |
152 | * the primary image. */ | |
153 | IMAGE_ROLE_PREDICTOR, | |
154 | ||
155 | /* Color Transform Data: Defines the color transformation for different | |
156 | * areas of the primary image. */ | |
157 | IMAGE_ROLE_COLOR_TRANSFORM, | |
158 | ||
159 | /* Color Index: Stored as an image of height == 1. */ | |
160 | IMAGE_ROLE_COLOR_INDEXING, | |
161 | ||
162 | IMAGE_ROLE_NB, | |
163 | }; | |
164 | ||
165 | typedef struct HuffReader { | |
166 | VLC vlc; /* Huffman decoder context */ | |
167 | int simple; /* whether to use simple mode */ | |
168 | int nb_symbols; /* number of coded symbols */ | |
169 | uint16_t simple_symbols[2]; /* symbols for simple mode */ | |
170 | } HuffReader; | |
171 | ||
172 | typedef struct ImageContext { | |
173 | enum ImageRole role; /* role of this image */ | |
174 | AVFrame *frame; /* AVFrame for data */ | |
175 | int color_cache_bits; /* color cache size, log2 */ | |
176 | uint32_t *color_cache; /* color cache data */ | |
177 | int nb_huffman_groups; /* number of huffman groups */ | |
178 | HuffReader *huffman_groups; /* reader for each huffman group */ | |
179 | int size_reduction; /* relative size compared to primary image, log2 */ | |
180 | int is_alpha_primary; | |
181 | } ImageContext; | |
182 | ||
183 | typedef struct WebPContext { | |
184 | VP8Context v; /* VP8 Context used for lossy decoding */ | |
185 | GetBitContext gb; /* bitstream reader for main image chunk */ | |
186 | AVFrame *alpha_frame; /* AVFrame for alpha data decompressed from VP8L */ | |
187 | AVCodecContext *avctx; /* parent AVCodecContext */ | |
188 | int initialized; /* set once the VP8 context is initialized */ | |
189 | int has_alpha; /* has a separate alpha chunk */ | |
190 | enum AlphaCompression alpha_compression; /* compression type for alpha chunk */ | |
191 | enum AlphaFilter alpha_filter; /* filtering method for alpha chunk */ | |
192 | uint8_t *alpha_data; /* alpha chunk data */ | |
193 | int alpha_data_size; /* alpha chunk data size */ | |
194 | int width; /* image width */ | |
195 | int height; /* image height */ | |
196 | int lossless; /* indicates lossless or lossy */ | |
197 | ||
198 | int nb_transforms; /* number of transforms */ | |
199 | enum TransformType transforms[4]; /* transformations used in the image, in order */ | |
200 | int reduced_width; /* reduced width for index image, if applicable */ | |
201 | int nb_huffman_groups; /* number of huffman groups in the primary image */ | |
202 | ImageContext image[IMAGE_ROLE_NB]; /* image context for each role */ | |
203 | } WebPContext; | |
204 | ||
205 | #define GET_PIXEL(frame, x, y) \ | |
206 | ((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x)) | |
207 | ||
208 | #define GET_PIXEL_COMP(frame, x, y, c) \ | |
209 | (*((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x) + c)) | |
210 | ||
211 | static void image_ctx_free(ImageContext *img) | |
212 | { | |
213 | int i, j; | |
214 | ||
215 | av_free(img->color_cache); | |
216 | if (img->role != IMAGE_ROLE_ARGB && !img->is_alpha_primary) | |
217 | av_frame_free(&img->frame); | |
218 | if (img->huffman_groups) { | |
219 | for (i = 0; i < img->nb_huffman_groups; i++) { | |
220 | for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++) | |
221 | ff_free_vlc(&img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE + j].vlc); | |
222 | } | |
223 | av_free(img->huffman_groups); | |
224 | } | |
225 | memset(img, 0, sizeof(*img)); | |
226 | } | |
227 | ||
228 | ||
229 | /* Differs from get_vlc2() in the following ways: | |
230 | * - codes are bit-reversed | |
231 | * - assumes 8-bit table to make reversal simpler | |
232 | * - assumes max depth of 2 since the max code length for WebP is 15 | |
233 | */ | |
234 | static av_always_inline int webp_get_vlc(GetBitContext *gb, VLC_TYPE (*table)[2]) | |
235 | { | |
236 | int n, nb_bits; | |
237 | unsigned int index; | |
238 | int code; | |
239 | ||
240 | OPEN_READER(re, gb); | |
241 | UPDATE_CACHE(re, gb); | |
242 | ||
243 | index = SHOW_UBITS(re, gb, 8); | |
244 | index = ff_reverse[index]; | |
245 | code = table[index][0]; | |
246 | n = table[index][1]; | |
247 | ||
248 | if (n < 0) { | |
249 | LAST_SKIP_BITS(re, gb, 8); | |
250 | UPDATE_CACHE(re, gb); | |
251 | ||
252 | nb_bits = -n; | |
253 | ||
254 | index = SHOW_UBITS(re, gb, nb_bits); | |
255 | index = (ff_reverse[index] >> (8 - nb_bits)) + code; | |
256 | code = table[index][0]; | |
257 | n = table[index][1]; | |
258 | } | |
259 | SKIP_BITS(re, gb, n); | |
260 | ||
261 | CLOSE_READER(re, gb); | |
262 | ||
263 | return code; | |
264 | } | |
265 | ||
266 | static int huff_reader_get_symbol(HuffReader *r, GetBitContext *gb) | |
267 | { | |
268 | if (r->simple) { | |
269 | if (r->nb_symbols == 1) | |
270 | return r->simple_symbols[0]; | |
271 | else | |
272 | return r->simple_symbols[get_bits1(gb)]; | |
273 | } else | |
274 | return webp_get_vlc(gb, r->vlc.table); | |
275 | } | |
276 | ||
277 | static int huff_reader_build_canonical(HuffReader *r, int *code_lengths, | |
278 | int alphabet_size) | |
279 | { | |
d085f80f | 280 | int len = 0, sym, code = 0, ret; |
c4bfa098 JR |
281 | int max_code_length = 0; |
282 | uint16_t *codes; | |
283 | ||
d085f80f JR |
284 | /* special-case 1 symbol since the vlc reader cannot handle it */ |
285 | for (sym = 0; sym < alphabet_size; sym++) { | |
286 | if (code_lengths[sym] > 0) { | |
287 | len++; | |
288 | code = sym; | |
289 | if (len > 1) | |
290 | break; | |
291 | } | |
292 | } | |
293 | if (len == 1) { | |
294 | r->nb_symbols = 1; | |
295 | r->simple_symbols[0] = code; | |
296 | r->simple = 1; | |
297 | return 0; | |
298 | } | |
299 | ||
c4bfa098 JR |
300 | for (sym = 0; sym < alphabet_size; sym++) |
301 | max_code_length = FFMAX(max_code_length, code_lengths[sym]); | |
302 | ||
303 | if (max_code_length == 0 || max_code_length > MAX_HUFFMAN_CODE_LENGTH) | |
304 | return AVERROR(EINVAL); | |
305 | ||
306 | codes = av_malloc(alphabet_size * sizeof(*codes)); | |
307 | if (!codes) | |
308 | return AVERROR(ENOMEM); | |
309 | ||
310 | code = 0; | |
311 | r->nb_symbols = 0; | |
312 | for (len = 1; len <= max_code_length; len++) { | |
313 | for (sym = 0; sym < alphabet_size; sym++) { | |
314 | if (code_lengths[sym] != len) | |
315 | continue; | |
316 | codes[sym] = code++; | |
317 | r->nb_symbols++; | |
318 | } | |
319 | code <<= 1; | |
320 | } | |
321 | if (!r->nb_symbols) { | |
322 | av_free(codes); | |
323 | return AVERROR_INVALIDDATA; | |
324 | } | |
325 | ||
326 | ret = init_vlc(&r->vlc, 8, alphabet_size, | |
327 | code_lengths, sizeof(*code_lengths), sizeof(*code_lengths), | |
328 | codes, sizeof(*codes), sizeof(*codes), 0); | |
329 | if (ret < 0) { | |
330 | av_free(codes); | |
331 | return ret; | |
332 | } | |
333 | r->simple = 0; | |
334 | ||
335 | av_free(codes); | |
336 | return 0; | |
337 | } | |
338 | ||
339 | static void read_huffman_code_simple(WebPContext *s, HuffReader *hc) | |
340 | { | |
341 | hc->nb_symbols = get_bits1(&s->gb) + 1; | |
342 | ||
343 | if (get_bits1(&s->gb)) | |
344 | hc->simple_symbols[0] = get_bits(&s->gb, 8); | |
345 | else | |
346 | hc->simple_symbols[0] = get_bits1(&s->gb); | |
347 | ||
348 | if (hc->nb_symbols == 2) | |
349 | hc->simple_symbols[1] = get_bits(&s->gb, 8); | |
350 | ||
351 | hc->simple = 1; | |
352 | } | |
353 | ||
354 | static int read_huffman_code_normal(WebPContext *s, HuffReader *hc, | |
355 | int alphabet_size) | |
356 | { | |
357 | HuffReader code_len_hc = { { 0 }, 0, 0, { 0 } }; | |
358 | int *code_lengths = NULL; | |
359 | int code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 }; | |
360 | int i, symbol, max_symbol, prev_code_len, ret; | |
361 | int num_codes = 4 + get_bits(&s->gb, 4); | |
362 | ||
363 | if (num_codes > NUM_CODE_LENGTH_CODES) | |
364 | return AVERROR_INVALIDDATA; | |
365 | ||
366 | for (i = 0; i < num_codes; i++) | |
367 | code_length_code_lengths[code_length_code_order[i]] = get_bits(&s->gb, 3); | |
368 | ||
369 | ret = huff_reader_build_canonical(&code_len_hc, code_length_code_lengths, | |
370 | NUM_CODE_LENGTH_CODES); | |
371 | if (ret < 0) | |
372 | goto finish; | |
373 | ||
374 | code_lengths = av_mallocz_array(alphabet_size, sizeof(*code_lengths)); | |
375 | if (!code_lengths) { | |
376 | ret = AVERROR(ENOMEM); | |
377 | goto finish; | |
378 | } | |
379 | ||
380 | if (get_bits1(&s->gb)) { | |
381 | int bits = 2 + 2 * get_bits(&s->gb, 3); | |
382 | max_symbol = 2 + get_bits(&s->gb, bits); | |
383 | if (max_symbol > alphabet_size) { | |
384 | av_log(s->avctx, AV_LOG_ERROR, "max symbol %d > alphabet size %d\n", | |
385 | max_symbol, alphabet_size); | |
386 | ret = AVERROR_INVALIDDATA; | |
387 | goto finish; | |
388 | } | |
389 | } else { | |
390 | max_symbol = alphabet_size; | |
391 | } | |
392 | ||
393 | prev_code_len = 8; | |
394 | symbol = 0; | |
395 | while (symbol < alphabet_size) { | |
396 | int code_len; | |
397 | ||
398 | if (!max_symbol--) | |
399 | break; | |
400 | code_len = huff_reader_get_symbol(&code_len_hc, &s->gb); | |
401 | if (code_len < 16) { | |
402 | /* Code length code [0..15] indicates literal code lengths. */ | |
403 | code_lengths[symbol++] = code_len; | |
404 | if (code_len) | |
405 | prev_code_len = code_len; | |
406 | } else { | |
407 | int repeat = 0, length = 0; | |
408 | switch (code_len) { | |
409 | case 16: | |
410 | /* Code 16 repeats the previous non-zero value [3..6] times, | |
411 | * i.e., 3 + ReadBits(2) times. If code 16 is used before a | |
412 | * non-zero value has been emitted, a value of 8 is repeated. */ | |
413 | repeat = 3 + get_bits(&s->gb, 2); | |
414 | length = prev_code_len; | |
415 | break; | |
416 | case 17: | |
417 | /* Code 17 emits a streak of zeros [3..10], i.e., | |
418 | * 3 + ReadBits(3) times. */ | |
419 | repeat = 3 + get_bits(&s->gb, 3); | |
420 | break; | |
421 | case 18: | |
422 | /* Code 18 emits a streak of zeros of length [11..138], i.e., | |
423 | * 11 + ReadBits(7) times. */ | |
424 | repeat = 11 + get_bits(&s->gb, 7); | |
425 | break; | |
426 | } | |
427 | if (symbol + repeat > alphabet_size) { | |
428 | av_log(s->avctx, AV_LOG_ERROR, | |
429 | "invalid symbol %d + repeat %d > alphabet size %d\n", | |
430 | symbol, repeat, alphabet_size); | |
431 | ret = AVERROR_INVALIDDATA; | |
432 | goto finish; | |
433 | } | |
434 | while (repeat-- > 0) | |
435 | code_lengths[symbol++] = length; | |
436 | } | |
437 | } | |
438 | ||
439 | ret = huff_reader_build_canonical(hc, code_lengths, alphabet_size); | |
440 | ||
441 | finish: | |
442 | ff_free_vlc(&code_len_hc.vlc); | |
443 | av_free(code_lengths); | |
444 | return ret; | |
445 | } | |
446 | ||
447 | static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role, | |
448 | int w, int h); | |
449 | ||
450 | #define PARSE_BLOCK_SIZE(w, h) do { \ | |
451 | block_bits = get_bits(&s->gb, 3) + 2; \ | |
452 | blocks_w = FFALIGN((w), 1 << block_bits) >> block_bits; \ | |
453 | blocks_h = FFALIGN((h), 1 << block_bits) >> block_bits; \ | |
454 | } while (0) | |
455 | ||
456 | static int decode_entropy_image(WebPContext *s) | |
457 | { | |
458 | ImageContext *img; | |
459 | int ret, block_bits, width, blocks_w, blocks_h, x, y, max; | |
460 | ||
461 | width = s->width; | |
462 | if (s->reduced_width > 0) | |
463 | width = s->reduced_width; | |
464 | ||
465 | PARSE_BLOCK_SIZE(width, s->height); | |
466 | ||
467 | ret = decode_entropy_coded_image(s, IMAGE_ROLE_ENTROPY, blocks_w, blocks_h); | |
468 | if (ret < 0) | |
469 | return ret; | |
470 | ||
471 | img = &s->image[IMAGE_ROLE_ENTROPY]; | |
472 | img->size_reduction = block_bits; | |
473 | ||
474 | /* the number of huffman groups is determined by the maximum group number | |
475 | * coded in the entropy image */ | |
476 | max = 0; | |
477 | for (y = 0; y < img->frame->height; y++) { | |
478 | for (x = 0; x < img->frame->width; x++) { | |
c6698dfe JR |
479 | int p0 = GET_PIXEL_COMP(img->frame, x, y, 1); |
480 | int p1 = GET_PIXEL_COMP(img->frame, x, y, 2); | |
481 | int p = p0 << 8 | p1; | |
c4bfa098 JR |
482 | max = FFMAX(max, p); |
483 | } | |
484 | } | |
485 | s->nb_huffman_groups = max + 1; | |
486 | ||
487 | return 0; | |
488 | } | |
489 | ||
490 | static int parse_transform_predictor(WebPContext *s) | |
491 | { | |
492 | int block_bits, blocks_w, blocks_h, ret; | |
493 | ||
494 | PARSE_BLOCK_SIZE(s->width, s->height); | |
495 | ||
496 | ret = decode_entropy_coded_image(s, IMAGE_ROLE_PREDICTOR, blocks_w, | |
497 | blocks_h); | |
498 | if (ret < 0) | |
499 | return ret; | |
500 | ||
501 | s->image[IMAGE_ROLE_PREDICTOR].size_reduction = block_bits; | |
502 | ||
503 | return 0; | |
504 | } | |
505 | ||
506 | static int parse_transform_color(WebPContext *s) | |
507 | { | |
508 | int block_bits, blocks_w, blocks_h, ret; | |
509 | ||
510 | PARSE_BLOCK_SIZE(s->width, s->height); | |
511 | ||
512 | ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_TRANSFORM, blocks_w, | |
513 | blocks_h); | |
514 | if (ret < 0) | |
515 | return ret; | |
516 | ||
517 | s->image[IMAGE_ROLE_COLOR_TRANSFORM].size_reduction = block_bits; | |
518 | ||
519 | return 0; | |
520 | } | |
521 | ||
522 | static int parse_transform_color_indexing(WebPContext *s) | |
523 | { | |
524 | ImageContext *img; | |
525 | int width_bits, index_size, ret, x; | |
526 | uint8_t *ct; | |
527 | ||
528 | index_size = get_bits(&s->gb, 8) + 1; | |
529 | ||
530 | if (index_size <= 2) | |
531 | width_bits = 3; | |
532 | else if (index_size <= 4) | |
533 | width_bits = 2; | |
534 | else if (index_size <= 16) | |
535 | width_bits = 1; | |
536 | else | |
537 | width_bits = 0; | |
538 | ||
539 | ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_INDEXING, | |
540 | index_size, 1); | |
541 | if (ret < 0) | |
542 | return ret; | |
543 | ||
544 | img = &s->image[IMAGE_ROLE_COLOR_INDEXING]; | |
545 | img->size_reduction = width_bits; | |
546 | if (width_bits > 0) | |
547 | s->reduced_width = (s->width + ((1 << width_bits) - 1)) >> width_bits; | |
548 | ||
549 | /* color index values are delta-coded */ | |
550 | ct = img->frame->data[0] + 4; | |
551 | for (x = 4; x < img->frame->width * 4; x++, ct++) | |
552 | ct[0] += ct[-4]; | |
553 | ||
554 | return 0; | |
555 | } | |
556 | ||
557 | static HuffReader *get_huffman_group(WebPContext *s, ImageContext *img, | |
558 | int x, int y) | |
559 | { | |
560 | ImageContext *gimg = &s->image[IMAGE_ROLE_ENTROPY]; | |
561 | int group = 0; | |
562 | ||
563 | if (gimg->size_reduction > 0) { | |
564 | int group_x = x >> gimg->size_reduction; | |
565 | int group_y = y >> gimg->size_reduction; | |
c6698dfe JR |
566 | int g0 = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 1); |
567 | int g1 = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 2); | |
568 | group = g0 << 8 | g1; | |
c4bfa098 JR |
569 | } |
570 | ||
571 | return &img->huffman_groups[group * HUFFMAN_CODES_PER_META_CODE]; | |
572 | } | |
573 | ||
574 | static av_always_inline void color_cache_put(ImageContext *img, uint32_t c) | |
575 | { | |
576 | uint32_t cache_idx = (0x1E35A7BD * c) >> (32 - img->color_cache_bits); | |
577 | img->color_cache[cache_idx] = c; | |
578 | } | |
579 | ||
580 | static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role, | |
581 | int w, int h) | |
582 | { | |
583 | ImageContext *img; | |
584 | HuffReader *hg; | |
585 | int i, j, ret, x, y, width; | |
586 | ||
587 | img = &s->image[role]; | |
588 | img->role = role; | |
589 | ||
590 | if (!img->frame) { | |
591 | img->frame = av_frame_alloc(); | |
592 | if (!img->frame) | |
593 | return AVERROR(ENOMEM); | |
594 | } | |
595 | ||
596 | img->frame->format = AV_PIX_FMT_ARGB; | |
597 | img->frame->width = w; | |
598 | img->frame->height = h; | |
599 | ||
600 | if (role == IMAGE_ROLE_ARGB && !img->is_alpha_primary) { | |
601 | ThreadFrame pt = { .f = img->frame }; | |
602 | ret = ff_thread_get_buffer(s->avctx, &pt, 0); | |
603 | } else | |
604 | ret = av_frame_get_buffer(img->frame, 1); | |
605 | if (ret < 0) | |
606 | return ret; | |
607 | ||
608 | if (get_bits1(&s->gb)) { | |
609 | img->color_cache_bits = get_bits(&s->gb, 4); | |
610 | if (img->color_cache_bits < 1 || img->color_cache_bits > 11) { | |
611 | av_log(s->avctx, AV_LOG_ERROR, "invalid color cache bits: %d\n", | |
612 | img->color_cache_bits); | |
613 | return AVERROR_INVALIDDATA; | |
614 | } | |
615 | img->color_cache = av_mallocz_array(1 << img->color_cache_bits, | |
616 | sizeof(*img->color_cache)); | |
617 | if (!img->color_cache) | |
618 | return AVERROR(ENOMEM); | |
619 | } else { | |
620 | img->color_cache_bits = 0; | |
621 | } | |
622 | ||
623 | img->nb_huffman_groups = 1; | |
624 | if (role == IMAGE_ROLE_ARGB && get_bits1(&s->gb)) { | |
625 | ret = decode_entropy_image(s); | |
626 | if (ret < 0) | |
627 | return ret; | |
628 | img->nb_huffman_groups = s->nb_huffman_groups; | |
629 | } | |
630 | img->huffman_groups = av_mallocz_array(img->nb_huffman_groups * | |
631 | HUFFMAN_CODES_PER_META_CODE, | |
632 | sizeof(*img->huffman_groups)); | |
633 | if (!img->huffman_groups) | |
634 | return AVERROR(ENOMEM); | |
635 | ||
636 | for (i = 0; i < img->nb_huffman_groups; i++) { | |
637 | hg = &img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE]; | |
638 | for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++) { | |
639 | int alphabet_size = alphabet_sizes[j]; | |
640 | if (!j && img->color_cache_bits > 0) | |
641 | alphabet_size += 1 << img->color_cache_bits; | |
642 | ||
643 | if (get_bits1(&s->gb)) { | |
644 | read_huffman_code_simple(s, &hg[j]); | |
645 | } else { | |
646 | ret = read_huffman_code_normal(s, &hg[j], alphabet_size); | |
647 | if (ret < 0) | |
648 | return ret; | |
649 | } | |
650 | } | |
651 | } | |
652 | ||
653 | width = img->frame->width; | |
654 | if (role == IMAGE_ROLE_ARGB && s->reduced_width > 0) | |
655 | width = s->reduced_width; | |
656 | ||
657 | x = 0; y = 0; | |
658 | while (y < img->frame->height) { | |
659 | int v; | |
660 | ||
661 | hg = get_huffman_group(s, img, x, y); | |
662 | v = huff_reader_get_symbol(&hg[HUFF_IDX_GREEN], &s->gb); | |
663 | if (v < NUM_LITERAL_CODES) { | |
664 | /* literal pixel values */ | |
665 | uint8_t *p = GET_PIXEL(img->frame, x, y); | |
666 | p[2] = v; | |
667 | p[1] = huff_reader_get_symbol(&hg[HUFF_IDX_RED], &s->gb); | |
668 | p[3] = huff_reader_get_symbol(&hg[HUFF_IDX_BLUE], &s->gb); | |
669 | p[0] = huff_reader_get_symbol(&hg[HUFF_IDX_ALPHA], &s->gb); | |
670 | if (img->color_cache_bits) | |
671 | color_cache_put(img, AV_RB32(p)); | |
672 | x++; | |
673 | if (x == width) { | |
674 | x = 0; | |
675 | y++; | |
676 | } | |
677 | } else if (v < NUM_LITERAL_CODES + NUM_LENGTH_CODES) { | |
678 | /* LZ77 backwards mapping */ | |
679 | int prefix_code, length, distance, ref_x, ref_y; | |
680 | ||
681 | /* parse length and distance */ | |
682 | prefix_code = v - NUM_LITERAL_CODES; | |
683 | if (prefix_code < 4) { | |
684 | length = prefix_code + 1; | |
685 | } else { | |
686 | int extra_bits = (prefix_code - 2) >> 1; | |
687 | int offset = 2 + (prefix_code & 1) << extra_bits; | |
688 | length = offset + get_bits(&s->gb, extra_bits) + 1; | |
689 | } | |
690 | prefix_code = huff_reader_get_symbol(&hg[HUFF_IDX_DIST], &s->gb); | |
4f2ee9da AC |
691 | if (prefix_code > 39) { |
692 | av_log(s->avctx, AV_LOG_ERROR, | |
693 | "distance prefix code too large: %d\n", prefix_code); | |
694 | return AVERROR_INVALIDDATA; | |
695 | } | |
c4bfa098 JR |
696 | if (prefix_code < 4) { |
697 | distance = prefix_code + 1; | |
698 | } else { | |
699 | int extra_bits = prefix_code - 2 >> 1; | |
700 | int offset = 2 + (prefix_code & 1) << extra_bits; | |
701 | distance = offset + get_bits(&s->gb, extra_bits) + 1; | |
702 | } | |
703 | ||
704 | /* find reference location */ | |
705 | if (distance <= NUM_SHORT_DISTANCES) { | |
706 | int xi = lz77_distance_offsets[distance - 1][0]; | |
707 | int yi = lz77_distance_offsets[distance - 1][1]; | |
708 | distance = FFMAX(1, xi + yi * width); | |
709 | } else { | |
710 | distance -= NUM_SHORT_DISTANCES; | |
711 | } | |
712 | ref_x = x; | |
713 | ref_y = y; | |
714 | if (distance <= x) { | |
715 | ref_x -= distance; | |
716 | distance = 0; | |
717 | } else { | |
718 | ref_x = 0; | |
719 | distance -= x; | |
720 | } | |
721 | while (distance >= width) { | |
722 | ref_y--; | |
723 | distance -= width; | |
724 | } | |
725 | if (distance > 0) { | |
726 | ref_x = width - distance; | |
727 | ref_y--; | |
728 | } | |
729 | ref_x = FFMAX(0, ref_x); | |
730 | ref_y = FFMAX(0, ref_y); | |
731 | ||
732 | /* copy pixels | |
733 | * source and dest regions can overlap and wrap lines, so just | |
734 | * copy per-pixel */ | |
735 | for (i = 0; i < length; i++) { | |
736 | uint8_t *p_ref = GET_PIXEL(img->frame, ref_x, ref_y); | |
737 | uint8_t *p = GET_PIXEL(img->frame, x, y); | |
738 | ||
739 | AV_COPY32(p, p_ref); | |
740 | if (img->color_cache_bits) | |
741 | color_cache_put(img, AV_RB32(p)); | |
742 | x++; | |
743 | ref_x++; | |
744 | if (x == width) { | |
745 | x = 0; | |
746 | y++; | |
747 | } | |
748 | if (ref_x == width) { | |
749 | ref_x = 0; | |
750 | ref_y++; | |
751 | } | |
752 | if (y == img->frame->height || ref_y == img->frame->height) | |
753 | break; | |
754 | } | |
755 | } else { | |
756 | /* read from color cache */ | |
757 | uint8_t *p = GET_PIXEL(img->frame, x, y); | |
758 | int cache_idx = v - (NUM_LITERAL_CODES + NUM_LENGTH_CODES); | |
759 | ||
760 | if (!img->color_cache_bits) { | |
761 | av_log(s->avctx, AV_LOG_ERROR, "color cache not found\n"); | |
762 | return AVERROR_INVALIDDATA; | |
763 | } | |
764 | if (cache_idx >= 1 << img->color_cache_bits) { | |
765 | av_log(s->avctx, AV_LOG_ERROR, | |
766 | "color cache index out-of-bounds\n"); | |
767 | return AVERROR_INVALIDDATA; | |
768 | } | |
769 | AV_WB32(p, img->color_cache[cache_idx]); | |
770 | x++; | |
771 | if (x == width) { | |
772 | x = 0; | |
773 | y++; | |
774 | } | |
775 | } | |
776 | } | |
777 | ||
778 | return 0; | |
779 | } | |
780 | ||
781 | /* PRED_MODE_BLACK */ | |
782 | static void inv_predict_0(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_WB32(p, 0xFF000000); | |
786 | } | |
787 | ||
788 | /* PRED_MODE_L */ | |
789 | static void inv_predict_1(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 | AV_COPY32(p, p_l); | |
793 | } | |
794 | ||
795 | /* PRED_MODE_T */ | |
796 | static void inv_predict_2(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
797 | const uint8_t *p_t, const uint8_t *p_tr) | |
798 | { | |
799 | AV_COPY32(p, p_t); | |
800 | } | |
801 | ||
802 | /* PRED_MODE_TR */ | |
803 | static void inv_predict_3(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
804 | const uint8_t *p_t, const uint8_t *p_tr) | |
805 | { | |
806 | AV_COPY32(p, p_tr); | |
807 | } | |
808 | ||
809 | /* PRED_MODE_TL */ | |
810 | static void inv_predict_4(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
811 | const uint8_t *p_t, const uint8_t *p_tr) | |
812 | { | |
813 | AV_COPY32(p, p_tl); | |
814 | } | |
815 | ||
816 | /* PRED_MODE_AVG_T_AVG_L_TR */ | |
817 | static void inv_predict_5(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
818 | const uint8_t *p_t, const uint8_t *p_tr) | |
819 | { | |
820 | p[0] = p_t[0] + (p_l[0] + p_tr[0] >> 1) >> 1; | |
821 | p[1] = p_t[1] + (p_l[1] + p_tr[1] >> 1) >> 1; | |
822 | p[2] = p_t[2] + (p_l[2] + p_tr[2] >> 1) >> 1; | |
823 | p[3] = p_t[3] + (p_l[3] + p_tr[3] >> 1) >> 1; | |
824 | } | |
825 | ||
826 | /* PRED_MODE_AVG_L_TL */ | |
827 | static void inv_predict_6(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
828 | const uint8_t *p_t, const uint8_t *p_tr) | |
829 | { | |
830 | p[0] = p_l[0] + p_tl[0] >> 1; | |
831 | p[1] = p_l[1] + p_tl[1] >> 1; | |
832 | p[2] = p_l[2] + p_tl[2] >> 1; | |
833 | p[3] = p_l[3] + p_tl[3] >> 1; | |
834 | } | |
835 | ||
836 | /* PRED_MODE_AVG_L_T */ | |
837 | static void inv_predict_7(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
838 | const uint8_t *p_t, const uint8_t *p_tr) | |
839 | { | |
840 | p[0] = p_l[0] + p_t[0] >> 1; | |
841 | p[1] = p_l[1] + p_t[1] >> 1; | |
842 | p[2] = p_l[2] + p_t[2] >> 1; | |
843 | p[3] = p_l[3] + p_t[3] >> 1; | |
844 | } | |
845 | ||
846 | /* PRED_MODE_AVG_TL_T */ | |
847 | static void inv_predict_8(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
848 | const uint8_t *p_t, const uint8_t *p_tr) | |
849 | { | |
850 | p[0] = p_tl[0] + p_t[0] >> 1; | |
851 | p[1] = p_tl[1] + p_t[1] >> 1; | |
852 | p[2] = p_tl[2] + p_t[2] >> 1; | |
853 | p[3] = p_tl[3] + p_t[3] >> 1; | |
854 | } | |
855 | ||
856 | /* PRED_MODE_AVG_T_TR */ | |
857 | static void inv_predict_9(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
858 | const uint8_t *p_t, const uint8_t *p_tr) | |
859 | { | |
860 | p[0] = p_t[0] + p_tr[0] >> 1; | |
861 | p[1] = p_t[1] + p_tr[1] >> 1; | |
862 | p[2] = p_t[2] + p_tr[2] >> 1; | |
863 | p[3] = p_t[3] + p_tr[3] >> 1; | |
864 | } | |
865 | ||
866 | /* PRED_MODE_AVG_AVG_L_TL_AVG_T_TR */ | |
867 | static void inv_predict_10(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
868 | const uint8_t *p_t, const uint8_t *p_tr) | |
869 | { | |
870 | p[0] = (p_l[0] + p_tl[0] >> 1) + (p_t[0] + p_tr[0] >> 1) >> 1; | |
871 | p[1] = (p_l[1] + p_tl[1] >> 1) + (p_t[1] + p_tr[1] >> 1) >> 1; | |
872 | p[2] = (p_l[2] + p_tl[2] >> 1) + (p_t[2] + p_tr[2] >> 1) >> 1; | |
873 | p[3] = (p_l[3] + p_tl[3] >> 1) + (p_t[3] + p_tr[3] >> 1) >> 1; | |
874 | } | |
875 | ||
876 | /* PRED_MODE_SELECT */ | |
877 | static void inv_predict_11(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
878 | const uint8_t *p_t, const uint8_t *p_tr) | |
879 | { | |
880 | int diff = (FFABS(p_l[0] - p_tl[0]) - FFABS(p_t[0] - p_tl[0])) + | |
881 | (FFABS(p_l[1] - p_tl[1]) - FFABS(p_t[1] - p_tl[1])) + | |
882 | (FFABS(p_l[2] - p_tl[2]) - FFABS(p_t[2] - p_tl[2])) + | |
883 | (FFABS(p_l[3] - p_tl[3]) - FFABS(p_t[3] - p_tl[3])); | |
884 | if (diff <= 0) | |
885 | AV_COPY32(p, p_t); | |
886 | else | |
887 | AV_COPY32(p, p_l); | |
888 | } | |
889 | ||
890 | /* PRED_MODE_ADD_SUBTRACT_FULL */ | |
891 | static void inv_predict_12(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
892 | const uint8_t *p_t, const uint8_t *p_tr) | |
893 | { | |
894 | p[0] = av_clip_uint8(p_l[0] + p_t[0] - p_tl[0]); | |
895 | p[1] = av_clip_uint8(p_l[1] + p_t[1] - p_tl[1]); | |
896 | p[2] = av_clip_uint8(p_l[2] + p_t[2] - p_tl[2]); | |
897 | p[3] = av_clip_uint8(p_l[3] + p_t[3] - p_tl[3]); | |
898 | } | |
899 | ||
900 | static av_always_inline uint8_t clamp_add_subtract_half(int a, int b, int c) | |
901 | { | |
902 | int d = a + b >> 1; | |
903 | return av_clip_uint8(d + (d - c) / 2); | |
904 | } | |
905 | ||
906 | /* PRED_MODE_ADD_SUBTRACT_HALF */ | |
907 | static void inv_predict_13(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl, | |
908 | const uint8_t *p_t, const uint8_t *p_tr) | |
909 | { | |
910 | p[0] = clamp_add_subtract_half(p_l[0], p_t[0], p_tl[0]); | |
911 | p[1] = clamp_add_subtract_half(p_l[1], p_t[1], p_tl[1]); | |
912 | p[2] = clamp_add_subtract_half(p_l[2], p_t[2], p_tl[2]); | |
913 | p[3] = clamp_add_subtract_half(p_l[3], p_t[3], p_tl[3]); | |
914 | } | |
915 | ||
916 | typedef void (*inv_predict_func)(uint8_t *p, const uint8_t *p_l, | |
917 | const uint8_t *p_tl, const uint8_t *p_t, | |
918 | const uint8_t *p_tr); | |
919 | ||
920 | static const inv_predict_func inverse_predict[14] = { | |
921 | inv_predict_0, inv_predict_1, inv_predict_2, inv_predict_3, | |
922 | inv_predict_4, inv_predict_5, inv_predict_6, inv_predict_7, | |
923 | inv_predict_8, inv_predict_9, inv_predict_10, inv_predict_11, | |
924 | inv_predict_12, inv_predict_13, | |
925 | }; | |
926 | ||
927 | static void inverse_prediction(AVFrame *frame, enum PredictionMode m, int x, int y) | |
928 | { | |
929 | uint8_t *dec, *p_l, *p_tl, *p_t, *p_tr; | |
930 | uint8_t p[4]; | |
931 | ||
932 | dec = GET_PIXEL(frame, x, y); | |
933 | p_l = GET_PIXEL(frame, x - 1, y); | |
934 | p_tl = GET_PIXEL(frame, x - 1, y - 1); | |
935 | p_t = GET_PIXEL(frame, x, y - 1); | |
936 | if (x == frame->width - 1) | |
937 | p_tr = GET_PIXEL(frame, 0, y); | |
938 | else | |
939 | p_tr = GET_PIXEL(frame, x + 1, y - 1); | |
940 | ||
941 | inverse_predict[m](p, p_l, p_tl, p_t, p_tr); | |
942 | ||
943 | dec[0] += p[0]; | |
944 | dec[1] += p[1]; | |
945 | dec[2] += p[2]; | |
946 | dec[3] += p[3]; | |
947 | } | |
948 | ||
949 | static int apply_predictor_transform(WebPContext *s) | |
950 | { | |
951 | ImageContext *img = &s->image[IMAGE_ROLE_ARGB]; | |
952 | ImageContext *pimg = &s->image[IMAGE_ROLE_PREDICTOR]; | |
953 | int x, y; | |
954 | ||
955 | for (y = 0; y < img->frame->height; y++) { | |
956 | for (x = 0; x < img->frame->width; x++) { | |
957 | int tx = x >> pimg->size_reduction; | |
958 | int ty = y >> pimg->size_reduction; | |
959 | enum PredictionMode m = GET_PIXEL_COMP(pimg->frame, tx, ty, 2); | |
960 | ||
961 | if (x == 0) { | |
962 | if (y == 0) | |
963 | m = PRED_MODE_BLACK; | |
964 | else | |
965 | m = PRED_MODE_T; | |
966 | } else if (y == 0) | |
967 | m = PRED_MODE_L; | |
968 | ||
969 | if (m > 13) { | |
970 | av_log(s->avctx, AV_LOG_ERROR, | |
971 | "invalid predictor mode: %d\n", m); | |
972 | return AVERROR_INVALIDDATA; | |
973 | } | |
974 | inverse_prediction(img->frame, m, x, y); | |
975 | } | |
976 | } | |
977 | return 0; | |
978 | } | |
979 | ||
980 | static av_always_inline uint8_t color_transform_delta(uint8_t color_pred, | |
981 | uint8_t color) | |
982 | { | |
983 | return (int)ff_u8_to_s8(color_pred) * ff_u8_to_s8(color) >> 5; | |
984 | } | |
985 | ||
986 | static int apply_color_transform(WebPContext *s) | |
987 | { | |
988 | ImageContext *img, *cimg; | |
989 | int x, y, cx, cy; | |
990 | uint8_t *p, *cp; | |
991 | ||
992 | img = &s->image[IMAGE_ROLE_ARGB]; | |
993 | cimg = &s->image[IMAGE_ROLE_COLOR_TRANSFORM]; | |
994 | ||
995 | for (y = 0; y < img->frame->height; y++) { | |
996 | for (x = 0; x < img->frame->width; x++) { | |
997 | cx = x >> cimg->size_reduction; | |
998 | cy = y >> cimg->size_reduction; | |
999 | cp = GET_PIXEL(cimg->frame, cx, cy); | |
1000 | p = GET_PIXEL(img->frame, x, y); | |
1001 | ||
1002 | p[1] += color_transform_delta(cp[3], p[2]); | |
1003 | p[3] += color_transform_delta(cp[2], p[2]) + | |
1004 | color_transform_delta(cp[1], p[1]); | |
1005 | } | |
1006 | } | |
1007 | return 0; | |
1008 | } | |
1009 | ||
1010 | static int apply_subtract_green_transform(WebPContext *s) | |
1011 | { | |
1012 | int x, y; | |
1013 | ImageContext *img = &s->image[IMAGE_ROLE_ARGB]; | |
1014 | ||
1015 | for (y = 0; y < img->frame->height; y++) { | |
1016 | for (x = 0; x < img->frame->width; x++) { | |
1017 | uint8_t *p = GET_PIXEL(img->frame, x, y); | |
1018 | p[1] += p[2]; | |
1019 | p[3] += p[2]; | |
1020 | } | |
1021 | } | |
1022 | return 0; | |
1023 | } | |
1024 | ||
1025 | static int apply_color_indexing_transform(WebPContext *s) | |
1026 | { | |
1027 | ImageContext *img; | |
1028 | ImageContext *pal; | |
1029 | int i, x, y; | |
1030 | uint8_t *p, *pi; | |
1031 | ||
1032 | img = &s->image[IMAGE_ROLE_ARGB]; | |
1033 | pal = &s->image[IMAGE_ROLE_COLOR_INDEXING]; | |
1034 | ||
1035 | if (pal->size_reduction > 0) { | |
1036 | GetBitContext gb_g; | |
1037 | uint8_t *line; | |
1038 | int pixel_bits = 8 >> pal->size_reduction; | |
1039 | ||
1040 | line = av_malloc(img->frame->linesize[0]); | |
1041 | if (!line) | |
1042 | return AVERROR(ENOMEM); | |
1043 | ||
1044 | for (y = 0; y < img->frame->height; y++) { | |
1045 | p = GET_PIXEL(img->frame, 0, y); | |
1046 | memcpy(line, p, img->frame->linesize[0]); | |
1047 | init_get_bits(&gb_g, line, img->frame->linesize[0] * 8); | |
1048 | skip_bits(&gb_g, 16); | |
1049 | i = 0; | |
1050 | for (x = 0; x < img->frame->width; x++) { | |
1051 | p = GET_PIXEL(img->frame, x, y); | |
1052 | p[2] = get_bits(&gb_g, pixel_bits); | |
1053 | i++; | |
1054 | if (i == 1 << pal->size_reduction) { | |
1055 | skip_bits(&gb_g, 24); | |
1056 | i = 0; | |
1057 | } | |
1058 | } | |
1059 | } | |
1060 | av_free(line); | |
1061 | } | |
1062 | ||
1063 | for (y = 0; y < img->frame->height; y++) { | |
1064 | for (x = 0; x < img->frame->width; x++) { | |
1065 | p = GET_PIXEL(img->frame, x, y); | |
1066 | i = p[2]; | |
1067 | if (i >= pal->frame->width) { | |
1068 | av_log(s->avctx, AV_LOG_ERROR, "invalid palette index %d\n", i); | |
1069 | return AVERROR_INVALIDDATA; | |
1070 | } | |
1071 | pi = GET_PIXEL(pal->frame, i, 0); | |
1072 | AV_COPY32(p, pi); | |
1073 | } | |
1074 | } | |
1075 | ||
1076 | return 0; | |
1077 | } | |
1078 | ||
1079 | static int vp8_lossless_decode_frame(AVCodecContext *avctx, AVFrame *p, | |
1080 | int *got_frame, uint8_t *data_start, | |
1081 | unsigned int data_size, int is_alpha_chunk) | |
1082 | { | |
1083 | WebPContext *s = avctx->priv_data; | |
30e6abd1 | 1084 | int w, h, ret, i, used; |
c4bfa098 JR |
1085 | |
1086 | if (!is_alpha_chunk) { | |
1087 | s->lossless = 1; | |
1088 | avctx->pix_fmt = AV_PIX_FMT_ARGB; | |
1089 | } | |
1090 | ||
1091 | ret = init_get_bits(&s->gb, data_start, data_size * 8); | |
1092 | if (ret < 0) | |
1093 | return ret; | |
1094 | ||
1095 | if (!is_alpha_chunk) { | |
1096 | if (get_bits(&s->gb, 8) != 0x2F) { | |
1097 | av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless signature\n"); | |
1098 | return AVERROR_INVALIDDATA; | |
1099 | } | |
1100 | ||
1101 | w = get_bits(&s->gb, 14) + 1; | |
1102 | h = get_bits(&s->gb, 14) + 1; | |
1103 | if (s->width && s->width != w) { | |
1104 | av_log(avctx, AV_LOG_WARNING, "Width mismatch. %d != %d\n", | |
1105 | s->width, w); | |
1106 | } | |
1107 | s->width = w; | |
1108 | if (s->height && s->height != h) { | |
1109 | av_log(avctx, AV_LOG_WARNING, "Height mismatch. %d != %d\n", | |
1110 | s->width, w); | |
1111 | } | |
1112 | s->height = h; | |
0a9e94bb AK |
1113 | |
1114 | ret = ff_set_dimensions(avctx, s->width, s->height); | |
c4bfa098 JR |
1115 | if (ret < 0) |
1116 | return ret; | |
c4bfa098 JR |
1117 | |
1118 | s->has_alpha = get_bits1(&s->gb); | |
1119 | ||
1120 | if (get_bits(&s->gb, 3) != 0x0) { | |
1121 | av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless version\n"); | |
1122 | return AVERROR_INVALIDDATA; | |
1123 | } | |
1124 | } else { | |
1125 | if (!s->width || !s->height) | |
1126 | return AVERROR_BUG; | |
1127 | w = s->width; | |
1128 | h = s->height; | |
1129 | } | |
1130 | ||
1131 | /* parse transformations */ | |
1132 | s->nb_transforms = 0; | |
1133 | s->reduced_width = 0; | |
30e6abd1 | 1134 | used = 0; |
c4bfa098 JR |
1135 | while (get_bits1(&s->gb)) { |
1136 | enum TransformType transform = get_bits(&s->gb, 2); | |
1137 | s->transforms[s->nb_transforms++] = transform; | |
30e6abd1 AC |
1138 | if (used & (1 << transform)) { |
1139 | av_log(avctx, AV_LOG_ERROR, "Transform %d used more than once\n", | |
1140 | transform); | |
1141 | ret = AVERROR_INVALIDDATA; | |
1142 | goto free_and_return; | |
1143 | } | |
1144 | used |= (1 << transform); | |
c4bfa098 JR |
1145 | switch (transform) { |
1146 | case PREDICTOR_TRANSFORM: | |
1147 | ret = parse_transform_predictor(s); | |
1148 | break; | |
1149 | case COLOR_TRANSFORM: | |
1150 | ret = parse_transform_color(s); | |
1151 | break; | |
1152 | case COLOR_INDEXING_TRANSFORM: | |
1153 | ret = parse_transform_color_indexing(s); | |
1154 | break; | |
1155 | } | |
1156 | if (ret < 0) | |
1157 | goto free_and_return; | |
1158 | } | |
1159 | ||
1160 | /* decode primary image */ | |
1161 | s->image[IMAGE_ROLE_ARGB].frame = p; | |
1162 | if (is_alpha_chunk) | |
1163 | s->image[IMAGE_ROLE_ARGB].is_alpha_primary = 1; | |
1164 | ret = decode_entropy_coded_image(s, IMAGE_ROLE_ARGB, w, h); | |
f51e3a19 | 1165 | if (ret < 0) |
c4bfa098 | 1166 | goto free_and_return; |
c4bfa098 JR |
1167 | |
1168 | /* apply transformations */ | |
1169 | for (i = s->nb_transforms - 1; i >= 0; i--) { | |
1170 | switch (s->transforms[i]) { | |
1171 | case PREDICTOR_TRANSFORM: | |
1172 | ret = apply_predictor_transform(s); | |
1173 | break; | |
1174 | case COLOR_TRANSFORM: | |
1175 | ret = apply_color_transform(s); | |
1176 | break; | |
1177 | case SUBTRACT_GREEN: | |
1178 | ret = apply_subtract_green_transform(s); | |
1179 | break; | |
1180 | case COLOR_INDEXING_TRANSFORM: | |
1181 | ret = apply_color_indexing_transform(s); | |
1182 | break; | |
1183 | } | |
f51e3a19 | 1184 | if (ret < 0) |
c4bfa098 | 1185 | goto free_and_return; |
c4bfa098 JR |
1186 | } |
1187 | ||
1188 | *got_frame = 1; | |
1189 | p->pict_type = AV_PICTURE_TYPE_I; | |
1190 | p->key_frame = 1; | |
1191 | ret = data_size; | |
1192 | ||
1193 | free_and_return: | |
1194 | for (i = 0; i < IMAGE_ROLE_NB; i++) | |
1195 | image_ctx_free(&s->image[i]); | |
1196 | ||
1197 | return ret; | |
1198 | } | |
1199 | ||
1200 | static void alpha_inverse_prediction(AVFrame *frame, enum AlphaFilter m) | |
1201 | { | |
1202 | int x, y, ls; | |
1203 | uint8_t *dec; | |
1204 | ||
1205 | ls = frame->linesize[3]; | |
1206 | ||
1207 | /* filter first row using horizontal filter */ | |
1208 | dec = frame->data[3] + 1; | |
1209 | for (x = 1; x < frame->width; x++, dec++) | |
1210 | *dec += *(dec - 1); | |
1211 | ||
1212 | /* filter first column using vertical filter */ | |
1213 | dec = frame->data[3] + ls; | |
1214 | for (y = 1; y < frame->height; y++, dec += ls) | |
1215 | *dec += *(dec - ls); | |
1216 | ||
1217 | /* filter the rest using the specified filter */ | |
1218 | switch (m) { | |
1219 | case ALPHA_FILTER_HORIZONTAL: | |
1220 | for (y = 1; y < frame->height; y++) { | |
1221 | dec = frame->data[3] + y * ls + 1; | |
1222 | for (x = 1; x < frame->width; x++, dec++) | |
1223 | *dec += *(dec - 1); | |
1224 | } | |
1225 | break; | |
1226 | case ALPHA_FILTER_VERTICAL: | |
1227 | for (y = 1; y < frame->height; y++) { | |
1228 | dec = frame->data[3] + y * ls + 1; | |
1229 | for (x = 1; x < frame->width; x++, dec++) | |
1230 | *dec += *(dec - ls); | |
1231 | } | |
1232 | break; | |
1233 | case ALPHA_FILTER_GRADIENT: | |
1234 | for (y = 1; y < frame->height; y++) { | |
1235 | dec = frame->data[3] + y * ls + 1; | |
1236 | for (x = 1; x < frame->width; x++, dec++) | |
1237 | dec[0] += av_clip_uint8(*(dec - 1) + *(dec - ls) - *(dec - ls - 1)); | |
1238 | } | |
1239 | break; | |
1240 | } | |
1241 | } | |
1242 | ||
1243 | static int vp8_lossy_decode_alpha(AVCodecContext *avctx, AVFrame *p, | |
1244 | uint8_t *data_start, | |
1245 | unsigned int data_size) | |
1246 | { | |
1247 | WebPContext *s = avctx->priv_data; | |
1248 | int x, y, ret; | |
1249 | ||
1250 | if (s->alpha_compression == ALPHA_COMPRESSION_NONE) { | |
1251 | GetByteContext gb; | |
1252 | ||
1253 | bytestream2_init(&gb, data_start, data_size); | |
1254 | for (y = 0; y < s->height; y++) | |
1255 | bytestream2_get_buffer(&gb, p->data[3] + p->linesize[3] * y, | |
1256 | s->width); | |
1257 | } else if (s->alpha_compression == ALPHA_COMPRESSION_VP8L) { | |
1258 | uint8_t *ap, *pp; | |
1259 | int alpha_got_frame = 0; | |
1260 | ||
1261 | s->alpha_frame = av_frame_alloc(); | |
1262 | if (!s->alpha_frame) | |
1263 | return AVERROR(ENOMEM); | |
1264 | ||
1265 | ret = vp8_lossless_decode_frame(avctx, s->alpha_frame, &alpha_got_frame, | |
1266 | data_start, data_size, 1); | |
1267 | if (ret < 0) { | |
1268 | av_frame_free(&s->alpha_frame); | |
1269 | return ret; | |
1270 | } | |
1271 | if (!alpha_got_frame) { | |
1272 | av_frame_free(&s->alpha_frame); | |
1273 | return AVERROR_INVALIDDATA; | |
1274 | } | |
1275 | ||
1276 | /* copy green component of alpha image to alpha plane of primary image */ | |
1277 | for (y = 0; y < s->height; y++) { | |
1278 | ap = GET_PIXEL(s->alpha_frame, 0, y) + 2; | |
1279 | pp = p->data[3] + p->linesize[3] * y; | |
1280 | for (x = 0; x < s->width; x++) { | |
1281 | *pp = *ap; | |
1282 | pp++; | |
1283 | ap += 4; | |
1284 | } | |
1285 | } | |
1286 | av_frame_free(&s->alpha_frame); | |
1287 | } | |
1288 | ||
1289 | /* apply alpha filtering */ | |
1290 | if (s->alpha_filter) | |
1291 | alpha_inverse_prediction(p, s->alpha_filter); | |
1292 | ||
1293 | return 0; | |
1294 | } | |
1295 | ||
1296 | static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p, | |
1297 | int *got_frame, uint8_t *data_start, | |
1298 | unsigned int data_size) | |
1299 | { | |
1300 | WebPContext *s = avctx->priv_data; | |
1301 | AVPacket pkt; | |
1302 | int ret; | |
1303 | ||
1304 | if (!s->initialized) { | |
1305 | ff_vp8_decode_init(avctx); | |
1306 | s->initialized = 1; | |
1307 | if (s->has_alpha) | |
1308 | avctx->pix_fmt = AV_PIX_FMT_YUVA420P; | |
1309 | } | |
1310 | s->lossless = 0; | |
1311 | ||
1312 | if (data_size > INT_MAX) { | |
1313 | av_log(avctx, AV_LOG_ERROR, "unsupported chunk size\n"); | |
1314 | return AVERROR_PATCHWELCOME; | |
1315 | } | |
1316 | ||
1317 | av_init_packet(&pkt); | |
1318 | pkt.data = data_start; | |
1319 | pkt.size = data_size; | |
1320 | ||
1321 | ret = ff_vp8_decode_frame(avctx, p, got_frame, &pkt); | |
1322 | if (s->has_alpha) { | |
1323 | ret = vp8_lossy_decode_alpha(avctx, p, s->alpha_data, | |
1324 | s->alpha_data_size); | |
1325 | if (ret < 0) | |
1326 | return ret; | |
1327 | } | |
1328 | return ret; | |
1329 | } | |
1330 | ||
1331 | static int webp_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, | |
1332 | AVPacket *avpkt) | |
1333 | { | |
1334 | AVFrame * const p = data; | |
1335 | WebPContext *s = avctx->priv_data; | |
1336 | GetByteContext gb; | |
1337 | int ret; | |
1338 | uint32_t chunk_type, chunk_size; | |
1339 | int vp8x_flags = 0; | |
1340 | ||
1341 | s->avctx = avctx; | |
1342 | s->width = 0; | |
1343 | s->height = 0; | |
1344 | *got_frame = 0; | |
1345 | s->has_alpha = 0; | |
1346 | bytestream2_init(&gb, avpkt->data, avpkt->size); | |
1347 | ||
1348 | if (bytestream2_get_bytes_left(&gb) < 12) | |
1349 | return AVERROR_INVALIDDATA; | |
1350 | ||
1351 | if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) { | |
1352 | av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n"); | |
1353 | return AVERROR_INVALIDDATA; | |
1354 | } | |
1355 | ||
1356 | chunk_size = bytestream2_get_le32(&gb); | |
1357 | if (bytestream2_get_bytes_left(&gb) < chunk_size) | |
1358 | return AVERROR_INVALIDDATA; | |
1359 | ||
1360 | if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) { | |
1361 | av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n"); | |
1362 | return AVERROR_INVALIDDATA; | |
1363 | } | |
1364 | ||
76d4c627 | 1365 | while (bytestream2_get_bytes_left(&gb) > 8) { |
c4bfa098 JR |
1366 | char chunk_str[5] = { 0 }; |
1367 | ||
1368 | chunk_type = bytestream2_get_le32(&gb); | |
1369 | chunk_size = bytestream2_get_le32(&gb); | |
1370 | if (chunk_size == UINT32_MAX) | |
1371 | return AVERROR_INVALIDDATA; | |
1372 | chunk_size += chunk_size & 1; | |
1373 | ||
1374 | if (bytestream2_get_bytes_left(&gb) < chunk_size) | |
1375 | return AVERROR_INVALIDDATA; | |
1376 | ||
1377 | switch (chunk_type) { | |
1378 | case MKTAG('V', 'P', '8', ' '): | |
1379 | if (!*got_frame) { | |
1380 | ret = vp8_lossy_decode_frame(avctx, p, got_frame, | |
1381 | avpkt->data + bytestream2_tell(&gb), | |
1382 | chunk_size); | |
1383 | if (ret < 0) | |
1384 | return ret; | |
1385 | } | |
1386 | bytestream2_skip(&gb, chunk_size); | |
1387 | break; | |
1388 | case MKTAG('V', 'P', '8', 'L'): | |
1389 | if (!*got_frame) { | |
1390 | ret = vp8_lossless_decode_frame(avctx, p, got_frame, | |
1391 | avpkt->data + bytestream2_tell(&gb), | |
1392 | chunk_size, 0); | |
1393 | if (ret < 0) | |
1394 | return ret; | |
1395 | } | |
1396 | bytestream2_skip(&gb, chunk_size); | |
1397 | break; | |
1398 | case MKTAG('V', 'P', '8', 'X'): | |
1399 | vp8x_flags = bytestream2_get_byte(&gb); | |
1400 | bytestream2_skip(&gb, 3); | |
1401 | s->width = bytestream2_get_le24(&gb) + 1; | |
1402 | s->height = bytestream2_get_le24(&gb) + 1; | |
1403 | ret = av_image_check_size(s->width, s->height, 0, avctx); | |
1404 | if (ret < 0) | |
1405 | return ret; | |
1406 | break; | |
1407 | case MKTAG('A', 'L', 'P', 'H'): { | |
1408 | int alpha_header, filter_m, compression; | |
1409 | ||
1410 | if (!(vp8x_flags & VP8X_FLAG_ALPHA)) { | |
1411 | av_log(avctx, AV_LOG_WARNING, | |
1412 | "ALPHA chunk present, but alpha bit not set in the " | |
1413 | "VP8X header\n"); | |
1414 | } | |
1415 | if (chunk_size == 0) { | |
1416 | av_log(avctx, AV_LOG_ERROR, "invalid ALPHA chunk size\n"); | |
1417 | return AVERROR_INVALIDDATA; | |
1418 | } | |
1419 | alpha_header = bytestream2_get_byte(&gb); | |
1420 | s->alpha_data = avpkt->data + bytestream2_tell(&gb); | |
1421 | s->alpha_data_size = chunk_size - 1; | |
1422 | bytestream2_skip(&gb, s->alpha_data_size); | |
1423 | ||
1424 | filter_m = (alpha_header >> 2) & 0x03; | |
1425 | compression = alpha_header & 0x03; | |
1426 | ||
1427 | if (compression > ALPHA_COMPRESSION_VP8L) { | |
1428 | av_log(avctx, AV_LOG_VERBOSE, | |
1429 | "skipping unsupported ALPHA chunk\n"); | |
1430 | } else { | |
1431 | s->has_alpha = 1; | |
1432 | s->alpha_compression = compression; | |
1433 | s->alpha_filter = filter_m; | |
1434 | } | |
1435 | ||
1436 | break; | |
1437 | } | |
1438 | case MKTAG('I', 'C', 'C', 'P'): | |
1439 | case MKTAG('A', 'N', 'I', 'M'): | |
1440 | case MKTAG('A', 'N', 'M', 'F'): | |
1441 | case MKTAG('E', 'X', 'I', 'F'): | |
1442 | case MKTAG('X', 'M', 'P', ' '): | |
1443 | AV_WL32(chunk_str, chunk_type); | |
1444 | av_log(avctx, AV_LOG_VERBOSE, "skipping unsupported chunk: %s\n", | |
1445 | chunk_str); | |
1446 | bytestream2_skip(&gb, chunk_size); | |
1447 | break; | |
1448 | default: | |
1449 | AV_WL32(chunk_str, chunk_type); | |
1450 | av_log(avctx, AV_LOG_VERBOSE, "skipping unknown chunk: %s\n", | |
1451 | chunk_str); | |
1452 | bytestream2_skip(&gb, chunk_size); | |
1453 | break; | |
1454 | } | |
1455 | } | |
1456 | ||
1457 | if (!*got_frame) { | |
1458 | av_log(avctx, AV_LOG_ERROR, "image data not found\n"); | |
1459 | return AVERROR_INVALIDDATA; | |
1460 | } | |
1461 | ||
1462 | return avpkt->size; | |
1463 | } | |
1464 | ||
1465 | static av_cold int webp_decode_close(AVCodecContext *avctx) | |
1466 | { | |
1467 | WebPContext *s = avctx->priv_data; | |
1468 | ||
1469 | if (s->initialized) | |
1470 | return ff_vp8_decode_free(avctx); | |
1471 | ||
1472 | return 0; | |
1473 | } | |
1474 | ||
1475 | AVCodec ff_webp_decoder = { | |
1476 | .name = "webp", | |
1477 | .long_name = NULL_IF_CONFIG_SMALL("WebP image"), | |
1478 | .type = AVMEDIA_TYPE_VIDEO, | |
1479 | .id = AV_CODEC_ID_WEBP, | |
1480 | .priv_data_size = sizeof(WebPContext), | |
1481 | .decode = webp_decode_frame, | |
1482 | .close = webp_decode_close, | |
1483 | .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS, | |
1484 | }; |