2 * WebP (.webp) image decoder
3 * Copyright (c) 2013 Aneesh Dogra <aneesh@sugarlabs.org>
4 * Copyright (c) 2013 Justin Ruggles <justin.ruggles@gmail.com>
6 * This file is part of Libav.
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.
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.
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
27 * @author Aneesh Dogra <aneesh@sugarlabs.org>
28 * Container and Lossy decoding
30 * @author Justin Ruggles <justin.ruggles@gmail.com>
32 * Compressed alpha for lossy
37 * - Exif and XMP metadata
40 #include "libavutil/imgutils.h"
42 #define BITSTREAM_READER_LE
44 #include "bitstream.h"
45 #include "bytestream.h"
50 #define VP8X_FLAG_ANIMATION 0x02
51 #define VP8X_FLAG_XMP_METADATA 0x04
52 #define VP8X_FLAG_EXIF_METADATA 0x08
53 #define VP8X_FLAG_ALPHA 0x10
54 #define VP8X_FLAG_ICC 0x20
56 #define MAX_PALETTE_SIZE 256
57 #define MAX_CACHE_BITS 11
58 #define NUM_CODE_LENGTH_CODES 19
59 #define HUFFMAN_CODES_PER_META_CODE 5
60 #define NUM_LITERAL_CODES 256
61 #define NUM_LENGTH_CODES 24
62 #define NUM_DISTANCE_CODES 40
63 #define NUM_SHORT_DISTANCES 120
64 #define MAX_HUFFMAN_CODE_LENGTH 15
66 static const uint16_t alphabet_sizes[HUFFMAN_CODES_PER_META_CODE] = {
67 NUM_LITERAL_CODES + NUM_LENGTH_CODES,
68 NUM_LITERAL_CODES, NUM_LITERAL_CODES, NUM_LITERAL_CODES,
72 static const uint8_t code_length_code_order[NUM_CODE_LENGTH_CODES] = {
73 17, 18, 0, 1, 2, 3, 4, 5, 16, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
76 static const int8_t lz77_distance_offsets[NUM_SHORT_DISTANCES][2] = {
77 { 0, 1 }, { 1, 0 }, { 1, 1 }, { -1, 1 }, { 0, 2 }, { 2, 0 }, { 1, 2 }, { -1, 2 },
78 { 2, 1 }, { -2, 1 }, { 2, 2 }, { -2, 2 }, { 0, 3 }, { 3, 0 }, { 1, 3 }, { -1, 3 },
79 { 3, 1 }, { -3, 1 }, { 2, 3 }, { -2, 3 }, { 3, 2 }, { -3, 2 }, { 0, 4 }, { 4, 0 },
80 { 1, 4 }, { -1, 4 }, { 4, 1 }, { -4, 1 }, { 3, 3 }, { -3, 3 }, { 2, 4 }, { -2, 4 },
81 { 4, 2 }, { -4, 2 }, { 0, 5 }, { 3, 4 }, { -3, 4 }, { 4, 3 }, { -4, 3 }, { 5, 0 },
82 { 1, 5 }, { -1, 5 }, { 5, 1 }, { -5, 1 }, { 2, 5 }, { -2, 5 }, { 5, 2 }, { -5, 2 },
83 { 4, 4 }, { -4, 4 }, { 3, 5 }, { -3, 5 }, { 5, 3 }, { -5, 3 }, { 0, 6 }, { 6, 0 },
84 { 1, 6 }, { -1, 6 }, { 6, 1 }, { -6, 1 }, { 2, 6 }, { -2, 6 }, { 6, 2 }, { -6, 2 },
85 { 4, 5 }, { -4, 5 }, { 5, 4 }, { -5, 4 }, { 3, 6 }, { -3, 6 }, { 6, 3 }, { -6, 3 },
86 { 0, 7 }, { 7, 0 }, { 1, 7 }, { -1, 7 }, { 5, 5 }, { -5, 5 }, { 7, 1 }, { -7, 1 },
87 { 4, 6 }, { -4, 6 }, { 6, 4 }, { -6, 4 }, { 2, 7 }, { -2, 7 }, { 7, 2 }, { -7, 2 },
88 { 3, 7 }, { -3, 7 }, { 7, 3 }, { -7, 3 }, { 5, 6 }, { -5, 6 }, { 6, 5 }, { -6, 5 },
89 { 8, 0 }, { 4, 7 }, { -4, 7 }, { 7, 4 }, { -7, 4 }, { 8, 1 }, { 8, 2 }, { 6, 6 },
90 { -6, 6 }, { 8, 3 }, { 5, 7 }, { -5, 7 }, { 7, 5 }, { -7, 5 }, { 8, 4 }, { 6, 7 },
91 { -6, 7 }, { 7, 6 }, { -7, 6 }, { 8, 5 }, { 7, 7 }, { -7, 7 }, { 8, 6 }, { 8, 7 }
94 enum AlphaCompression {
95 ALPHA_COMPRESSION_NONE,
96 ALPHA_COMPRESSION_VP8L,
101 ALPHA_FILTER_HORIZONTAL,
102 ALPHA_FILTER_VERTICAL,
103 ALPHA_FILTER_GRADIENT,
107 PREDICTOR_TRANSFORM = 0,
110 COLOR_INDEXING_TRANSFORM = 3,
113 enum PredictionMode {
119 PRED_MODE_AVG_T_AVG_L_TR,
124 PRED_MODE_AVG_AVG_L_TL_AVG_T_TR,
126 PRED_MODE_ADD_SUBTRACT_FULL,
127 PRED_MODE_ADD_SUBTRACT_HALF,
138 /* The structure of WebP lossless is an optional series of transformation data,
139 * followed by the primary image. The primary image also optionally contains
140 * an entropy group mapping if there are multiple entropy groups. There is a
141 * basic image type called an "entropy coded image" that is used for all of
142 * these. The type of each entropy coded image is referred to by the
143 * specification as its role. */
145 /* Primary Image: Stores the actual pixels of the image. */
148 /* Entropy Image: Defines which Huffman group to use for different areas of
149 * the primary image. */
152 /* Predictors: Defines which predictor type to use for different areas of
153 * the primary image. */
154 IMAGE_ROLE_PREDICTOR,
156 /* Color Transform Data: Defines the color transformation for different
157 * areas of the primary image. */
158 IMAGE_ROLE_COLOR_TRANSFORM,
160 /* Color Index: Stored as an image of height == 1. */
161 IMAGE_ROLE_COLOR_INDEXING,
166 typedef struct HuffReader {
167 VLC vlc; /* Huffman decoder context */
168 int simple; /* whether to use simple mode */
169 int nb_symbols; /* number of coded symbols */
170 uint16_t simple_symbols[2]; /* symbols for simple mode */
173 typedef struct ImageContext {
174 enum ImageRole role; /* role of this image */
175 AVFrame *frame; /* AVFrame for data */
176 int color_cache_bits; /* color cache size, log2 */
177 uint32_t *color_cache; /* color cache data */
178 int nb_huffman_groups; /* number of huffman groups */
179 HuffReader *huffman_groups; /* reader for each huffman group */
180 int size_reduction; /* relative size compared to primary image, log2 */
181 int is_alpha_primary;
184 typedef struct WebPContext {
185 VP8Context v; /* VP8 Context used for lossy decoding */
186 BitstreamContext bc; /* bitstream reader for main image chunk */
187 AVFrame *alpha_frame; /* AVFrame for alpha data decompressed from VP8L */
188 AVCodecContext *avctx; /* parent AVCodecContext */
189 int initialized; /* set once the VP8 context is initialized */
190 int has_alpha; /* has a separate alpha chunk */
191 enum AlphaCompression alpha_compression; /* compression type for alpha chunk */
192 enum AlphaFilter alpha_filter; /* filtering method for alpha chunk */
193 uint8_t *alpha_data; /* alpha chunk data */
194 int alpha_data_size; /* alpha chunk data size */
195 int width; /* image width */
196 int height; /* image height */
197 int lossless; /* indicates lossless or lossy */
199 int nb_transforms; /* number of transforms */
200 enum TransformType transforms[4]; /* transformations used in the image, in order */
201 int reduced_width; /* reduced width for index image, if applicable */
202 int nb_huffman_groups; /* number of huffman groups in the primary image */
203 ImageContext image[IMAGE_ROLE_NB]; /* image context for each role */
206 #define GET_PIXEL(frame, x, y) \
207 ((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x))
209 #define GET_PIXEL_COMP(frame, x, y, c) \
210 (*((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x) + c))
212 static void image_ctx_free(ImageContext *img)
216 av_free(img->color_cache);
217 if (img->role != IMAGE_ROLE_ARGB && !img->is_alpha_primary)
218 av_frame_free(&img->frame);
219 if (img->huffman_groups) {
220 for (i = 0; i < img->nb_huffman_groups; i++) {
221 for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++)
222 ff_free_vlc(&img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE + j].vlc);
224 av_free(img->huffman_groups);
226 memset(img, 0, sizeof(*img));
230 /* Differs from get_vlc2() in the following ways:
231 * - codes are bit-reversed
232 * - assumes 8-bit table to make reversal simpler
233 * - assumes max depth of 2 since the max code length for WebP is 15
235 static av_always_inline int webp_get_vlc(BitstreamContext *bc, VLC_TYPE (*table)[2])
241 index = bitstream_peek(bc, 8);
242 index = ff_reverse[index];
243 code = table[index][0];
247 bitstream_skip(bc, 8);
251 index = bitstream_peek(bc, nb_bits);
252 index = (ff_reverse[index] >> (8 - nb_bits)) + code;
253 code = table[index][0];
256 bitstream_skip(bc, n);
261 static int huff_reader_get_symbol(HuffReader *r, BitstreamContext *bc)
264 if (r->nb_symbols == 1)
265 return r->simple_symbols[0];
267 return r->simple_symbols[bitstream_read_bit(bc)];
269 return webp_get_vlc(bc, r->vlc.table);
272 static int huff_reader_build_canonical(HuffReader *r, int *code_lengths,
275 int len = 0, sym, code = 0, ret;
276 int max_code_length = 0;
279 /* special-case 1 symbol since the vlc reader cannot handle it */
280 for (sym = 0; sym < alphabet_size; sym++) {
281 if (code_lengths[sym] > 0) {
290 r->simple_symbols[0] = code;
295 for (sym = 0; sym < alphabet_size; sym++)
296 max_code_length = FFMAX(max_code_length, code_lengths[sym]);
298 if (max_code_length == 0 || max_code_length > MAX_HUFFMAN_CODE_LENGTH)
299 return AVERROR(EINVAL);
301 codes = av_malloc(alphabet_size * sizeof(*codes));
303 return AVERROR(ENOMEM);
307 for (len = 1; len <= max_code_length; len++) {
308 for (sym = 0; sym < alphabet_size; sym++) {
309 if (code_lengths[sym] != len)
316 if (!r->nb_symbols) {
318 return AVERROR_INVALIDDATA;
321 ret = init_vlc(&r->vlc, 8, alphabet_size,
322 code_lengths, sizeof(*code_lengths), sizeof(*code_lengths),
323 codes, sizeof(*codes), sizeof(*codes), 0);
334 static void read_huffman_code_simple(WebPContext *s, HuffReader *hc)
336 hc->nb_symbols = bitstream_read_bit(&s->bc) + 1;
338 if (bitstream_read_bit(&s->bc))
339 hc->simple_symbols[0] = bitstream_read(&s->bc, 8);
341 hc->simple_symbols[0] = bitstream_read_bit(&s->bc);
343 if (hc->nb_symbols == 2)
344 hc->simple_symbols[1] = bitstream_read(&s->bc, 8);
349 static int read_huffman_code_normal(WebPContext *s, HuffReader *hc,
352 HuffReader code_len_hc = { { 0 }, 0, 0, { 0 } };
353 int *code_lengths = NULL;
354 int code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 };
355 int i, symbol, max_symbol, prev_code_len, ret;
356 int num_codes = 4 + bitstream_read(&s->bc, 4);
358 if (num_codes > NUM_CODE_LENGTH_CODES)
359 return AVERROR_INVALIDDATA;
361 for (i = 0; i < num_codes; i++)
362 code_length_code_lengths[code_length_code_order[i]] = bitstream_read(&s->bc, 3);
364 ret = huff_reader_build_canonical(&code_len_hc, code_length_code_lengths,
365 NUM_CODE_LENGTH_CODES);
369 code_lengths = av_mallocz_array(alphabet_size, sizeof(*code_lengths));
371 ret = AVERROR(ENOMEM);
375 if (bitstream_read_bit(&s->bc)) {
376 int bits = 2 + 2 * bitstream_read(&s->bc, 3);
377 max_symbol = 2 + bitstream_read(&s->bc, bits);
378 if (max_symbol > alphabet_size) {
379 av_log(s->avctx, AV_LOG_ERROR, "max symbol %d > alphabet size %d\n",
380 max_symbol, alphabet_size);
381 ret = AVERROR_INVALIDDATA;
385 max_symbol = alphabet_size;
390 while (symbol < alphabet_size) {
395 code_len = huff_reader_get_symbol(&code_len_hc, &s->bc);
397 /* Code length code [0..15] indicates literal code lengths. */
398 code_lengths[symbol++] = code_len;
400 prev_code_len = code_len;
402 int repeat = 0, length = 0;
405 /* Code 16 repeats the previous non-zero value [3..6] times,
406 * i.e., 3 + ReadBits(2) times. If code 16 is used before a
407 * non-zero value has been emitted, a value of 8 is repeated. */
408 repeat = 3 + bitstream_read(&s->bc, 2);
409 length = prev_code_len;
412 /* Code 17 emits a streak of zeros [3..10], i.e.,
413 * 3 + ReadBits(3) times. */
414 repeat = 3 + bitstream_read(&s->bc, 3);
417 /* Code 18 emits a streak of zeros of length [11..138], i.e.,
418 * 11 + ReadBits(7) times. */
419 repeat = 11 + bitstream_read(&s->bc, 7);
422 if (symbol + repeat > alphabet_size) {
423 av_log(s->avctx, AV_LOG_ERROR,
424 "invalid symbol %d + repeat %d > alphabet size %d\n",
425 symbol, repeat, alphabet_size);
426 ret = AVERROR_INVALIDDATA;
430 code_lengths[symbol++] = length;
434 ret = huff_reader_build_canonical(hc, code_lengths, alphabet_size);
437 ff_free_vlc(&code_len_hc.vlc);
438 av_free(code_lengths);
442 static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role,
445 #define PARSE_BLOCK_SIZE(w, h) do { \
446 block_bits = bitstream_read(&s->bc, 3) + 2; \
447 blocks_w = FFALIGN((w), 1 << block_bits) >> block_bits; \
448 blocks_h = FFALIGN((h), 1 << block_bits) >> block_bits; \
451 static int decode_entropy_image(WebPContext *s)
454 int ret, block_bits, width, blocks_w, blocks_h, x, y, max;
457 if (s->reduced_width > 0)
458 width = s->reduced_width;
460 PARSE_BLOCK_SIZE(width, s->height);
462 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ENTROPY, blocks_w, blocks_h);
466 img = &s->image[IMAGE_ROLE_ENTROPY];
467 img->size_reduction = block_bits;
469 /* the number of huffman groups is determined by the maximum group number
470 * coded in the entropy image */
472 for (y = 0; y < img->frame->height; y++) {
473 for (x = 0; x < img->frame->width; x++) {
474 int p0 = GET_PIXEL_COMP(img->frame, x, y, 1);
475 int p1 = GET_PIXEL_COMP(img->frame, x, y, 2);
476 int p = p0 << 8 | p1;
480 s->nb_huffman_groups = max + 1;
485 static int parse_transform_predictor(WebPContext *s)
487 int block_bits, blocks_w, blocks_h, ret;
489 PARSE_BLOCK_SIZE(s->width, s->height);
491 ret = decode_entropy_coded_image(s, IMAGE_ROLE_PREDICTOR, blocks_w,
496 s->image[IMAGE_ROLE_PREDICTOR].size_reduction = block_bits;
501 static int parse_transform_color(WebPContext *s)
503 int block_bits, blocks_w, blocks_h, ret;
505 PARSE_BLOCK_SIZE(s->width, s->height);
507 ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_TRANSFORM, blocks_w,
512 s->image[IMAGE_ROLE_COLOR_TRANSFORM].size_reduction = block_bits;
517 static int parse_transform_color_indexing(WebPContext *s)
520 int width_bits, index_size, ret, x;
523 index_size = bitstream_read(&s->bc, 8) + 1;
527 else if (index_size <= 4)
529 else if (index_size <= 16)
534 ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_INDEXING,
539 img = &s->image[IMAGE_ROLE_COLOR_INDEXING];
540 img->size_reduction = width_bits;
542 s->reduced_width = (s->width + ((1 << width_bits) - 1)) >> width_bits;
544 /* color index values are delta-coded */
545 ct = img->frame->data[0] + 4;
546 for (x = 4; x < img->frame->width * 4; x++, ct++)
552 static HuffReader *get_huffman_group(WebPContext *s, ImageContext *img,
555 ImageContext *gimg = &s->image[IMAGE_ROLE_ENTROPY];
558 if (gimg->size_reduction > 0) {
559 int group_x = x >> gimg->size_reduction;
560 int group_y = y >> gimg->size_reduction;
561 int g0 = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 1);
562 int g1 = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 2);
563 group = g0 << 8 | g1;
566 return &img->huffman_groups[group * HUFFMAN_CODES_PER_META_CODE];
569 static av_always_inline void color_cache_put(ImageContext *img, uint32_t c)
571 uint32_t cache_idx = (0x1E35A7BD * c) >> (32 - img->color_cache_bits);
572 img->color_cache[cache_idx] = c;
575 static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role,
580 int i, j, ret, x, y, width;
582 img = &s->image[role];
586 img->frame = av_frame_alloc();
588 return AVERROR(ENOMEM);
591 img->frame->format = AV_PIX_FMT_ARGB;
592 img->frame->width = w;
593 img->frame->height = h;
595 if (role == IMAGE_ROLE_ARGB && !img->is_alpha_primary) {
596 ThreadFrame pt = { .f = img->frame };
597 ret = ff_thread_get_buffer(s->avctx, &pt, 0);
599 ret = av_frame_get_buffer(img->frame, 1);
603 if (bitstream_read_bit(&s->bc)) {
604 img->color_cache_bits = bitstream_read(&s->bc, 4);
605 if (img->color_cache_bits < 1 || img->color_cache_bits > 11) {
606 av_log(s->avctx, AV_LOG_ERROR, "invalid color cache bits: %d\n",
607 img->color_cache_bits);
608 return AVERROR_INVALIDDATA;
610 img->color_cache = av_mallocz_array(1 << img->color_cache_bits,
611 sizeof(*img->color_cache));
612 if (!img->color_cache)
613 return AVERROR(ENOMEM);
615 img->color_cache_bits = 0;
618 img->nb_huffman_groups = 1;
619 if (role == IMAGE_ROLE_ARGB && bitstream_read_bit(&s->bc)) {
620 ret = decode_entropy_image(s);
623 img->nb_huffman_groups = s->nb_huffman_groups;
625 img->huffman_groups = av_mallocz_array(img->nb_huffman_groups *
626 HUFFMAN_CODES_PER_META_CODE,
627 sizeof(*img->huffman_groups));
628 if (!img->huffman_groups)
629 return AVERROR(ENOMEM);
631 for (i = 0; i < img->nb_huffman_groups; i++) {
632 hg = &img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE];
633 for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++) {
634 int alphabet_size = alphabet_sizes[j];
635 if (!j && img->color_cache_bits > 0)
636 alphabet_size += 1 << img->color_cache_bits;
638 if (bitstream_read_bit(&s->bc)) {
639 read_huffman_code_simple(s, &hg[j]);
641 ret = read_huffman_code_normal(s, &hg[j], alphabet_size);
648 width = img->frame->width;
649 if (role == IMAGE_ROLE_ARGB && s->reduced_width > 0)
650 width = s->reduced_width;
653 while (y < img->frame->height) {
656 hg = get_huffman_group(s, img, x, y);
657 v = huff_reader_get_symbol(&hg[HUFF_IDX_GREEN], &s->bc);
658 if (v < NUM_LITERAL_CODES) {
659 /* literal pixel values */
660 uint8_t *p = GET_PIXEL(img->frame, x, y);
662 p[1] = huff_reader_get_symbol(&hg[HUFF_IDX_RED], &s->bc);
663 p[3] = huff_reader_get_symbol(&hg[HUFF_IDX_BLUE], &s->bc);
664 p[0] = huff_reader_get_symbol(&hg[HUFF_IDX_ALPHA], &s->bc);
665 if (img->color_cache_bits)
666 color_cache_put(img, AV_RB32(p));
672 } else if (v < NUM_LITERAL_CODES + NUM_LENGTH_CODES) {
673 /* LZ77 backwards mapping */
674 int prefix_code, length, distance, ref_x, ref_y;
676 /* parse length and distance */
677 prefix_code = v - NUM_LITERAL_CODES;
678 if (prefix_code < 4) {
679 length = prefix_code + 1;
681 int extra_bits = (prefix_code - 2) >> 1;
682 int offset = 2 + (prefix_code & 1) << extra_bits;
683 length = offset + bitstream_read(&s->bc, extra_bits) + 1;
685 prefix_code = huff_reader_get_symbol(&hg[HUFF_IDX_DIST], &s->bc);
686 if (prefix_code > 39) {
687 av_log(s->avctx, AV_LOG_ERROR,
688 "distance prefix code too large: %d\n", prefix_code);
689 return AVERROR_INVALIDDATA;
691 if (prefix_code < 4) {
692 distance = prefix_code + 1;
694 int extra_bits = prefix_code - 2 >> 1;
695 int offset = 2 + (prefix_code & 1) << extra_bits;
696 distance = offset + bitstream_read(&s->bc, extra_bits) + 1;
699 /* find reference location */
700 if (distance <= NUM_SHORT_DISTANCES) {
701 int xi = lz77_distance_offsets[distance - 1][0];
702 int yi = lz77_distance_offsets[distance - 1][1];
703 distance = FFMAX(1, xi + yi * width);
705 distance -= NUM_SHORT_DISTANCES;
716 while (distance >= width) {
721 ref_x = width - distance;
724 ref_x = FFMAX(0, ref_x);
725 ref_y = FFMAX(0, ref_y);
728 * source and dest regions can overlap and wrap lines, so just
730 for (i = 0; i < length; i++) {
731 uint8_t *p_ref = GET_PIXEL(img->frame, ref_x, ref_y);
732 uint8_t *p = GET_PIXEL(img->frame, x, y);
735 if (img->color_cache_bits)
736 color_cache_put(img, AV_RB32(p));
743 if (ref_x == width) {
747 if (y == img->frame->height || ref_y == img->frame->height)
751 /* read from color cache */
752 uint8_t *p = GET_PIXEL(img->frame, x, y);
753 int cache_idx = v - (NUM_LITERAL_CODES + NUM_LENGTH_CODES);
755 if (!img->color_cache_bits) {
756 av_log(s->avctx, AV_LOG_ERROR, "color cache not found\n");
757 return AVERROR_INVALIDDATA;
759 if (cache_idx >= 1 << img->color_cache_bits) {
760 av_log(s->avctx, AV_LOG_ERROR,
761 "color cache index out-of-bounds\n");
762 return AVERROR_INVALIDDATA;
764 AV_WB32(p, img->color_cache[cache_idx]);
776 /* PRED_MODE_BLACK */
777 static void inv_predict_0(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
778 const uint8_t *p_t, const uint8_t *p_tr)
780 AV_WB32(p, 0xFF000000);
784 static void inv_predict_1(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
785 const uint8_t *p_t, const uint8_t *p_tr)
791 static void inv_predict_2(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
792 const uint8_t *p_t, const uint8_t *p_tr)
798 static void inv_predict_3(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
799 const uint8_t *p_t, const uint8_t *p_tr)
805 static void inv_predict_4(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
806 const uint8_t *p_t, const uint8_t *p_tr)
811 /* PRED_MODE_AVG_T_AVG_L_TR */
812 static void inv_predict_5(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
813 const uint8_t *p_t, const uint8_t *p_tr)
815 p[0] = p_t[0] + (p_l[0] + p_tr[0] >> 1) >> 1;
816 p[1] = p_t[1] + (p_l[1] + p_tr[1] >> 1) >> 1;
817 p[2] = p_t[2] + (p_l[2] + p_tr[2] >> 1) >> 1;
818 p[3] = p_t[3] + (p_l[3] + p_tr[3] >> 1) >> 1;
821 /* PRED_MODE_AVG_L_TL */
822 static void inv_predict_6(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
823 const uint8_t *p_t, const uint8_t *p_tr)
825 p[0] = p_l[0] + p_tl[0] >> 1;
826 p[1] = p_l[1] + p_tl[1] >> 1;
827 p[2] = p_l[2] + p_tl[2] >> 1;
828 p[3] = p_l[3] + p_tl[3] >> 1;
831 /* PRED_MODE_AVG_L_T */
832 static void inv_predict_7(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
833 const uint8_t *p_t, const uint8_t *p_tr)
835 p[0] = p_l[0] + p_t[0] >> 1;
836 p[1] = p_l[1] + p_t[1] >> 1;
837 p[2] = p_l[2] + p_t[2] >> 1;
838 p[3] = p_l[3] + p_t[3] >> 1;
841 /* PRED_MODE_AVG_TL_T */
842 static void inv_predict_8(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
843 const uint8_t *p_t, const uint8_t *p_tr)
845 p[0] = p_tl[0] + p_t[0] >> 1;
846 p[1] = p_tl[1] + p_t[1] >> 1;
847 p[2] = p_tl[2] + p_t[2] >> 1;
848 p[3] = p_tl[3] + p_t[3] >> 1;
851 /* PRED_MODE_AVG_T_TR */
852 static void inv_predict_9(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
853 const uint8_t *p_t, const uint8_t *p_tr)
855 p[0] = p_t[0] + p_tr[0] >> 1;
856 p[1] = p_t[1] + p_tr[1] >> 1;
857 p[2] = p_t[2] + p_tr[2] >> 1;
858 p[3] = p_t[3] + p_tr[3] >> 1;
861 /* PRED_MODE_AVG_AVG_L_TL_AVG_T_TR */
862 static void inv_predict_10(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
863 const uint8_t *p_t, const uint8_t *p_tr)
865 p[0] = (p_l[0] + p_tl[0] >> 1) + (p_t[0] + p_tr[0] >> 1) >> 1;
866 p[1] = (p_l[1] + p_tl[1] >> 1) + (p_t[1] + p_tr[1] >> 1) >> 1;
867 p[2] = (p_l[2] + p_tl[2] >> 1) + (p_t[2] + p_tr[2] >> 1) >> 1;
868 p[3] = (p_l[3] + p_tl[3] >> 1) + (p_t[3] + p_tr[3] >> 1) >> 1;
871 /* PRED_MODE_SELECT */
872 static void inv_predict_11(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
873 const uint8_t *p_t, const uint8_t *p_tr)
875 int diff = (FFABS(p_l[0] - p_tl[0]) - FFABS(p_t[0] - p_tl[0])) +
876 (FFABS(p_l[1] - p_tl[1]) - FFABS(p_t[1] - p_tl[1])) +
877 (FFABS(p_l[2] - p_tl[2]) - FFABS(p_t[2] - p_tl[2])) +
878 (FFABS(p_l[3] - p_tl[3]) - FFABS(p_t[3] - p_tl[3]));
885 /* PRED_MODE_ADD_SUBTRACT_FULL */
886 static void inv_predict_12(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
887 const uint8_t *p_t, const uint8_t *p_tr)
889 p[0] = av_clip_uint8(p_l[0] + p_t[0] - p_tl[0]);
890 p[1] = av_clip_uint8(p_l[1] + p_t[1] - p_tl[1]);
891 p[2] = av_clip_uint8(p_l[2] + p_t[2] - p_tl[2]);
892 p[3] = av_clip_uint8(p_l[3] + p_t[3] - p_tl[3]);
895 static av_always_inline uint8_t clamp_add_subtract_half(int a, int b, int c)
898 return av_clip_uint8(d + (d - c) / 2);
901 /* PRED_MODE_ADD_SUBTRACT_HALF */
902 static void inv_predict_13(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
903 const uint8_t *p_t, const uint8_t *p_tr)
905 p[0] = clamp_add_subtract_half(p_l[0], p_t[0], p_tl[0]);
906 p[1] = clamp_add_subtract_half(p_l[1], p_t[1], p_tl[1]);
907 p[2] = clamp_add_subtract_half(p_l[2], p_t[2], p_tl[2]);
908 p[3] = clamp_add_subtract_half(p_l[3], p_t[3], p_tl[3]);
911 typedef void (*inv_predict_func)(uint8_t *p, const uint8_t *p_l,
912 const uint8_t *p_tl, const uint8_t *p_t,
913 const uint8_t *p_tr);
915 static const inv_predict_func inverse_predict[14] = {
916 inv_predict_0, inv_predict_1, inv_predict_2, inv_predict_3,
917 inv_predict_4, inv_predict_5, inv_predict_6, inv_predict_7,
918 inv_predict_8, inv_predict_9, inv_predict_10, inv_predict_11,
919 inv_predict_12, inv_predict_13,
922 static void inverse_prediction(AVFrame *frame, enum PredictionMode m, int x, int y)
924 uint8_t *dec, *p_l, *p_tl, *p_t, *p_tr;
927 dec = GET_PIXEL(frame, x, y);
928 p_l = GET_PIXEL(frame, x - 1, y);
929 p_tl = GET_PIXEL(frame, x - 1, y - 1);
930 p_t = GET_PIXEL(frame, x, y - 1);
931 if (x == frame->width - 1)
932 p_tr = GET_PIXEL(frame, 0, y);
934 p_tr = GET_PIXEL(frame, x + 1, y - 1);
936 inverse_predict[m](p, p_l, p_tl, p_t, p_tr);
944 static int apply_predictor_transform(WebPContext *s)
946 ImageContext *img = &s->image[IMAGE_ROLE_ARGB];
947 ImageContext *pimg = &s->image[IMAGE_ROLE_PREDICTOR];
950 for (y = 0; y < img->frame->height; y++) {
951 for (x = 0; x < img->frame->width; x++) {
952 int tx = x >> pimg->size_reduction;
953 int ty = y >> pimg->size_reduction;
954 enum PredictionMode m = GET_PIXEL_COMP(pimg->frame, tx, ty, 2);
965 av_log(s->avctx, AV_LOG_ERROR,
966 "invalid predictor mode: %d\n", m);
967 return AVERROR_INVALIDDATA;
969 inverse_prediction(img->frame, m, x, y);
975 static av_always_inline uint8_t color_transform_delta(uint8_t color_pred,
978 return (int)ff_u8_to_s8(color_pred) * ff_u8_to_s8(color) >> 5;
981 static int apply_color_transform(WebPContext *s)
983 ImageContext *img, *cimg;
987 img = &s->image[IMAGE_ROLE_ARGB];
988 cimg = &s->image[IMAGE_ROLE_COLOR_TRANSFORM];
990 for (y = 0; y < img->frame->height; y++) {
991 for (x = 0; x < img->frame->width; x++) {
992 cx = x >> cimg->size_reduction;
993 cy = y >> cimg->size_reduction;
994 cp = GET_PIXEL(cimg->frame, cx, cy);
995 p = GET_PIXEL(img->frame, x, y);
997 p[1] += color_transform_delta(cp[3], p[2]);
998 p[3] += color_transform_delta(cp[2], p[2]) +
999 color_transform_delta(cp[1], p[1]);
1005 static int apply_subtract_green_transform(WebPContext *s)
1008 ImageContext *img = &s->image[IMAGE_ROLE_ARGB];
1010 for (y = 0; y < img->frame->height; y++) {
1011 for (x = 0; x < img->frame->width; x++) {
1012 uint8_t *p = GET_PIXEL(img->frame, x, y);
1020 static int apply_color_indexing_transform(WebPContext *s)
1027 img = &s->image[IMAGE_ROLE_ARGB];
1028 pal = &s->image[IMAGE_ROLE_COLOR_INDEXING];
1030 if (pal->size_reduction > 0) {
1031 BitstreamContext bc_g;
1033 int pixel_bits = 8 >> pal->size_reduction;
1035 line = av_malloc(img->frame->linesize[0]);
1037 return AVERROR(ENOMEM);
1039 for (y = 0; y < img->frame->height; y++) {
1040 p = GET_PIXEL(img->frame, 0, y);
1041 memcpy(line, p, img->frame->linesize[0]);
1042 bitstream_init8(&bc_g, line, img->frame->linesize[0]);
1043 bitstream_skip(&bc_g, 16);
1045 for (x = 0; x < img->frame->width; x++) {
1046 p = GET_PIXEL(img->frame, x, y);
1047 p[2] = bitstream_read(&bc_g, pixel_bits);
1049 if (i == 1 << pal->size_reduction) {
1050 bitstream_skip(&bc_g, 24);
1058 for (y = 0; y < img->frame->height; y++) {
1059 for (x = 0; x < img->frame->width; x++) {
1060 p = GET_PIXEL(img->frame, x, y);
1062 if (i >= pal->frame->width) {
1063 av_log(s->avctx, AV_LOG_ERROR, "invalid palette index %d\n", i);
1064 return AVERROR_INVALIDDATA;
1066 pi = GET_PIXEL(pal->frame, i, 0);
1074 static int vp8_lossless_decode_frame(AVCodecContext *avctx, AVFrame *p,
1075 int *got_frame, uint8_t *data_start,
1076 unsigned int data_size, int is_alpha_chunk)
1078 WebPContext *s = avctx->priv_data;
1079 int w, h, ret, i, used;
1081 if (!is_alpha_chunk) {
1083 avctx->pix_fmt = AV_PIX_FMT_ARGB;
1086 ret = bitstream_init8(&s->bc, data_start, data_size);
1090 if (!is_alpha_chunk) {
1091 if (bitstream_read(&s->bc, 8) != 0x2F) {
1092 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless signature\n");
1093 return AVERROR_INVALIDDATA;
1096 w = bitstream_read(&s->bc, 14) + 1;
1097 h = bitstream_read(&s->bc, 14) + 1;
1098 if (s->width && s->width != w) {
1099 av_log(avctx, AV_LOG_WARNING, "Width mismatch. %d != %d\n",
1103 if (s->height && s->height != h) {
1104 av_log(avctx, AV_LOG_WARNING, "Height mismatch. %d != %d\n",
1109 ret = ff_set_dimensions(avctx, s->width, s->height);
1113 s->has_alpha = bitstream_read_bit(&s->bc);
1115 if (bitstream_read(&s->bc, 3) != 0x0) {
1116 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless version\n");
1117 return AVERROR_INVALIDDATA;
1120 if (!s->width || !s->height)
1126 /* parse transformations */
1127 s->nb_transforms = 0;
1128 s->reduced_width = 0;
1130 while (bitstream_read_bit(&s->bc)) {
1131 enum TransformType transform = bitstream_read(&s->bc, 2);
1132 s->transforms[s->nb_transforms++] = transform;
1133 if (used & (1 << transform)) {
1134 av_log(avctx, AV_LOG_ERROR, "Transform %d used more than once\n",
1136 ret = AVERROR_INVALIDDATA;
1137 goto free_and_return;
1139 used |= (1 << transform);
1140 switch (transform) {
1141 case PREDICTOR_TRANSFORM:
1142 ret = parse_transform_predictor(s);
1144 case COLOR_TRANSFORM:
1145 ret = parse_transform_color(s);
1147 case COLOR_INDEXING_TRANSFORM:
1148 ret = parse_transform_color_indexing(s);
1152 goto free_and_return;
1155 /* decode primary image */
1156 s->image[IMAGE_ROLE_ARGB].frame = p;
1158 s->image[IMAGE_ROLE_ARGB].is_alpha_primary = 1;
1159 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ARGB, w, h);
1161 goto free_and_return;
1163 /* apply transformations */
1164 for (i = s->nb_transforms - 1; i >= 0; i--) {
1165 switch (s->transforms[i]) {
1166 case PREDICTOR_TRANSFORM:
1167 ret = apply_predictor_transform(s);
1169 case COLOR_TRANSFORM:
1170 ret = apply_color_transform(s);
1172 case SUBTRACT_GREEN:
1173 ret = apply_subtract_green_transform(s);
1175 case COLOR_INDEXING_TRANSFORM:
1176 ret = apply_color_indexing_transform(s);
1180 goto free_and_return;
1184 p->pict_type = AV_PICTURE_TYPE_I;
1189 for (i = 0; i < IMAGE_ROLE_NB; i++)
1190 image_ctx_free(&s->image[i]);
1195 static void alpha_inverse_prediction(AVFrame *frame, enum AlphaFilter m)
1200 ls = frame->linesize[3];
1202 /* filter first row using horizontal filter */
1203 dec = frame->data[3] + 1;
1204 for (x = 1; x < frame->width; x++, dec++)
1207 /* filter first column using vertical filter */
1208 dec = frame->data[3] + ls;
1209 for (y = 1; y < frame->height; y++, dec += ls)
1210 *dec += *(dec - ls);
1212 /* filter the rest using the specified filter */
1214 case ALPHA_FILTER_HORIZONTAL:
1215 for (y = 1; y < frame->height; y++) {
1216 dec = frame->data[3] + y * ls + 1;
1217 for (x = 1; x < frame->width; x++, dec++)
1221 case ALPHA_FILTER_VERTICAL:
1222 for (y = 1; y < frame->height; y++) {
1223 dec = frame->data[3] + y * ls + 1;
1224 for (x = 1; x < frame->width; x++, dec++)
1225 *dec += *(dec - ls);
1228 case ALPHA_FILTER_GRADIENT:
1229 for (y = 1; y < frame->height; y++) {
1230 dec = frame->data[3] + y * ls + 1;
1231 for (x = 1; x < frame->width; x++, dec++)
1232 dec[0] += av_clip_uint8(*(dec - 1) + *(dec - ls) - *(dec - ls - 1));
1238 static int vp8_lossy_decode_alpha(AVCodecContext *avctx, AVFrame *p,
1239 uint8_t *data_start,
1240 unsigned int data_size)
1242 WebPContext *s = avctx->priv_data;
1245 if (s->alpha_compression == ALPHA_COMPRESSION_NONE) {
1248 bytestream2_init(&gb, data_start, data_size);
1249 for (y = 0; y < s->height; y++)
1250 bytestream2_get_buffer(&gb, p->data[3] + p->linesize[3] * y,
1252 } else if (s->alpha_compression == ALPHA_COMPRESSION_VP8L) {
1254 int alpha_got_frame = 0;
1256 s->alpha_frame = av_frame_alloc();
1257 if (!s->alpha_frame)
1258 return AVERROR(ENOMEM);
1260 ret = vp8_lossless_decode_frame(avctx, s->alpha_frame, &alpha_got_frame,
1261 data_start, data_size, 1);
1263 av_frame_free(&s->alpha_frame);
1266 if (!alpha_got_frame) {
1267 av_frame_free(&s->alpha_frame);
1268 return AVERROR_INVALIDDATA;
1271 /* copy green component of alpha image to alpha plane of primary image */
1272 for (y = 0; y < s->height; y++) {
1273 ap = GET_PIXEL(s->alpha_frame, 0, y) + 2;
1274 pp = p->data[3] + p->linesize[3] * y;
1275 for (x = 0; x < s->width; x++) {
1281 av_frame_free(&s->alpha_frame);
1284 /* apply alpha filtering */
1285 if (s->alpha_filter)
1286 alpha_inverse_prediction(p, s->alpha_filter);
1291 static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p,
1292 int *got_frame, uint8_t *data_start,
1293 unsigned int data_size)
1295 WebPContext *s = avctx->priv_data;
1299 if (!s->initialized) {
1300 ff_vp8_decode_init(avctx);
1302 s->v.actually_webp = 1;
1304 avctx->pix_fmt = AV_PIX_FMT_YUVA420P;
1306 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1310 if (data_size > INT_MAX) {
1311 av_log(avctx, AV_LOG_ERROR, "unsupported chunk size\n");
1312 return AVERROR_PATCHWELCOME;
1315 av_init_packet(&pkt);
1316 pkt.data = data_start;
1317 pkt.size = data_size;
1319 ret = ff_vp8_decode_frame(avctx, p, got_frame, &pkt);
1321 ret = vp8_lossy_decode_alpha(avctx, p, s->alpha_data,
1322 s->alpha_data_size);
1329 static int webp_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
1332 AVFrame * const p = data;
1333 WebPContext *s = avctx->priv_data;
1336 uint32_t chunk_type, chunk_size;
1344 bytestream2_init(&gb, avpkt->data, avpkt->size);
1346 if (bytestream2_get_bytes_left(&gb) < 12)
1347 return AVERROR_INVALIDDATA;
1349 if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
1350 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
1351 return AVERROR_INVALIDDATA;
1354 chunk_size = bytestream2_get_le32(&gb);
1355 if (bytestream2_get_bytes_left(&gb) < chunk_size)
1356 return AVERROR_INVALIDDATA;
1358 if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) {
1359 av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n");
1360 return AVERROR_INVALIDDATA;
1363 while (bytestream2_get_bytes_left(&gb) > 8) {
1364 char chunk_str[5] = { 0 };
1366 chunk_type = bytestream2_get_le32(&gb);
1367 chunk_size = bytestream2_get_le32(&gb);
1368 if (chunk_size == UINT32_MAX)
1369 return AVERROR_INVALIDDATA;
1370 chunk_size += chunk_size & 1;
1372 if (bytestream2_get_bytes_left(&gb) < chunk_size)
1373 return AVERROR_INVALIDDATA;
1375 switch (chunk_type) {
1376 case MKTAG('V', 'P', '8', ' '):
1378 ret = vp8_lossy_decode_frame(avctx, p, got_frame,
1379 avpkt->data + bytestream2_tell(&gb),
1384 bytestream2_skip(&gb, chunk_size);
1386 case MKTAG('V', 'P', '8', 'L'):
1388 ret = vp8_lossless_decode_frame(avctx, p, got_frame,
1389 avpkt->data + bytestream2_tell(&gb),
1394 bytestream2_skip(&gb, chunk_size);
1396 case MKTAG('V', 'P', '8', 'X'):
1397 vp8x_flags = bytestream2_get_byte(&gb);
1398 bytestream2_skip(&gb, 3);
1399 s->width = bytestream2_get_le24(&gb) + 1;
1400 s->height = bytestream2_get_le24(&gb) + 1;
1401 ret = av_image_check_size(s->width, s->height, 0, avctx);
1405 case MKTAG('A', 'L', 'P', 'H'): {
1406 int alpha_header, filter_m, compression;
1408 if (!(vp8x_flags & VP8X_FLAG_ALPHA)) {
1409 av_log(avctx, AV_LOG_WARNING,
1410 "ALPHA chunk present, but alpha bit not set in the "
1413 if (chunk_size == 0) {
1414 av_log(avctx, AV_LOG_ERROR, "invalid ALPHA chunk size\n");
1415 return AVERROR_INVALIDDATA;
1417 alpha_header = bytestream2_get_byte(&gb);
1418 s->alpha_data = avpkt->data + bytestream2_tell(&gb);
1419 s->alpha_data_size = chunk_size - 1;
1420 bytestream2_skip(&gb, s->alpha_data_size);
1422 filter_m = (alpha_header >> 2) & 0x03;
1423 compression = alpha_header & 0x03;
1425 if (compression > ALPHA_COMPRESSION_VP8L) {
1426 av_log(avctx, AV_LOG_VERBOSE,
1427 "skipping unsupported ALPHA chunk\n");
1430 s->alpha_compression = compression;
1431 s->alpha_filter = filter_m;
1436 case MKTAG('I', 'C', 'C', 'P'):
1437 case MKTAG('A', 'N', 'I', 'M'):
1438 case MKTAG('A', 'N', 'M', 'F'):
1439 case MKTAG('E', 'X', 'I', 'F'):
1440 case MKTAG('X', 'M', 'P', ' '):
1441 AV_WL32(chunk_str, chunk_type);
1442 av_log(avctx, AV_LOG_VERBOSE, "skipping unsupported chunk: %s\n",
1444 bytestream2_skip(&gb, chunk_size);
1447 AV_WL32(chunk_str, chunk_type);
1448 av_log(avctx, AV_LOG_VERBOSE, "skipping unknown chunk: %s\n",
1450 bytestream2_skip(&gb, chunk_size);
1456 av_log(avctx, AV_LOG_ERROR, "image data not found\n");
1457 return AVERROR_INVALIDDATA;
1463 static av_cold int webp_decode_close(AVCodecContext *avctx)
1465 WebPContext *s = avctx->priv_data;
1468 return ff_vp8_decode_free(avctx);
1473 AVCodec ff_webp_decoder = {
1475 .long_name = NULL_IF_CONFIG_SMALL("WebP image"),
1476 .type = AVMEDIA_TYPE_VIDEO,
1477 .id = AV_CODEC_ID_WEBP,
1478 .priv_data_size = sizeof(WebPContext),
1479 .decode = webp_decode_frame,
1480 .close = webp_decode_close,
1481 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,