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 #define BITSTREAM_READER_LE
41 #include "libavutil/imgutils.h"
43 #include "bytestream.h"
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
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
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,
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
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 }
93 enum AlphaCompression {
94 ALPHA_COMPRESSION_NONE,
95 ALPHA_COMPRESSION_VP8L,
100 ALPHA_FILTER_HORIZONTAL,
101 ALPHA_FILTER_VERTICAL,
102 ALPHA_FILTER_GRADIENT,
106 PREDICTOR_TRANSFORM = 0,
109 COLOR_INDEXING_TRANSFORM = 3,
112 enum PredictionMode {
118 PRED_MODE_AVG_T_AVG_L_TR,
123 PRED_MODE_AVG_AVG_L_TL_AVG_T_TR,
125 PRED_MODE_ADD_SUBTRACT_FULL,
126 PRED_MODE_ADD_SUBTRACT_HALF,
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. */
144 /* Primary Image: Stores the actual pixels of the image. */
147 /* Entropy Image: Defines which Huffman group to use for different areas of
148 * the primary image. */
151 /* Predictors: Defines which predictor type to use for different areas of
152 * the primary image. */
153 IMAGE_ROLE_PREDICTOR,
155 /* Color Transform Data: Defines the color transformation for different
156 * areas of the primary image. */
157 IMAGE_ROLE_COLOR_TRANSFORM,
159 /* Color Index: Stored as an image of height == 1. */
160 IMAGE_ROLE_COLOR_INDEXING,
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 */
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;
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 */
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 */
205 #define GET_PIXEL(frame, x, y) \
206 ((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x))
208 #define GET_PIXEL_COMP(frame, x, y, c) \
209 (*((frame)->data[0] + (y) * frame->linesize[0] + 4 * (x) + c))
211 static void image_ctx_free(ImageContext *img)
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);
223 av_free(img->huffman_groups);
225 memset(img, 0, sizeof(*img));
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
234 static av_always_inline int webp_get_vlc(GetBitContext *gb, VLC_TYPE (*table)[2])
241 UPDATE_CACHE(re, gb);
243 index = SHOW_UBITS(re, gb, 8);
244 index = ff_reverse[index];
245 code = table[index][0];
249 LAST_SKIP_BITS(re, gb, 8);
250 UPDATE_CACHE(re, gb);
254 index = SHOW_UBITS(re, gb, nb_bits);
255 index = (ff_reverse[index] >> (8 - nb_bits)) + code;
256 code = table[index][0];
259 SKIP_BITS(re, gb, n);
261 CLOSE_READER(re, gb);
266 static int huff_reader_get_symbol(HuffReader *r, GetBitContext *gb)
269 if (r->nb_symbols == 1)
270 return r->simple_symbols[0];
272 return r->simple_symbols[get_bits1(gb)];
274 return webp_get_vlc(gb, r->vlc.table);
277 static int huff_reader_build_canonical(HuffReader *r, int *code_lengths,
280 int len = 0, sym, code = 0, ret;
281 int max_code_length = 0;
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) {
295 r->simple_symbols[0] = code;
300 for (sym = 0; sym < alphabet_size; sym++)
301 max_code_length = FFMAX(max_code_length, code_lengths[sym]);
303 if (max_code_length == 0 || max_code_length > MAX_HUFFMAN_CODE_LENGTH)
304 return AVERROR(EINVAL);
306 codes = av_malloc(alphabet_size * sizeof(*codes));
308 return AVERROR(ENOMEM);
312 for (len = 1; len <= max_code_length; len++) {
313 for (sym = 0; sym < alphabet_size; sym++) {
314 if (code_lengths[sym] != len)
321 if (!r->nb_symbols) {
323 return AVERROR_INVALIDDATA;
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);
339 static void read_huffman_code_simple(WebPContext *s, HuffReader *hc)
341 hc->nb_symbols = get_bits1(&s->gb) + 1;
343 if (get_bits1(&s->gb))
344 hc->simple_symbols[0] = get_bits(&s->gb, 8);
346 hc->simple_symbols[0] = get_bits1(&s->gb);
348 if (hc->nb_symbols == 2)
349 hc->simple_symbols[1] = get_bits(&s->gb, 8);
354 static int read_huffman_code_normal(WebPContext *s, HuffReader *hc,
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);
363 if (num_codes > NUM_CODE_LENGTH_CODES)
364 return AVERROR_INVALIDDATA;
366 for (i = 0; i < num_codes; i++)
367 code_length_code_lengths[code_length_code_order[i]] = get_bits(&s->gb, 3);
369 ret = huff_reader_build_canonical(&code_len_hc, code_length_code_lengths,
370 NUM_CODE_LENGTH_CODES);
374 code_lengths = av_mallocz_array(alphabet_size, sizeof(*code_lengths));
376 ret = AVERROR(ENOMEM);
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;
390 max_symbol = alphabet_size;
395 while (symbol < alphabet_size) {
400 code_len = huff_reader_get_symbol(&code_len_hc, &s->gb);
402 /* Code length code [0..15] indicates literal code lengths. */
403 code_lengths[symbol++] = code_len;
405 prev_code_len = code_len;
407 int repeat = 0, length = 0;
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;
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);
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);
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;
435 code_lengths[symbol++] = length;
439 ret = huff_reader_build_canonical(hc, code_lengths, alphabet_size);
442 ff_free_vlc(&code_len_hc.vlc);
443 av_free(code_lengths);
447 static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role,
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; \
456 static int decode_entropy_image(WebPContext *s)
459 int ret, block_bits, width, blocks_w, blocks_h, x, y, max;
462 if (s->reduced_width > 0)
463 width = s->reduced_width;
465 PARSE_BLOCK_SIZE(width, s->height);
467 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ENTROPY, blocks_w, blocks_h);
471 img = &s->image[IMAGE_ROLE_ENTROPY];
472 img->size_reduction = block_bits;
474 /* the number of huffman groups is determined by the maximum group number
475 * coded in the entropy image */
477 for (y = 0; y < img->frame->height; y++) {
478 for (x = 0; x < img->frame->width; x++) {
479 int p = GET_PIXEL_COMP(img->frame, x, y, 2);
483 s->nb_huffman_groups = max + 1;
488 static int parse_transform_predictor(WebPContext *s)
490 int block_bits, blocks_w, blocks_h, ret;
492 PARSE_BLOCK_SIZE(s->width, s->height);
494 ret = decode_entropy_coded_image(s, IMAGE_ROLE_PREDICTOR, blocks_w,
499 s->image[IMAGE_ROLE_PREDICTOR].size_reduction = block_bits;
504 static int parse_transform_color(WebPContext *s)
506 int block_bits, blocks_w, blocks_h, ret;
508 PARSE_BLOCK_SIZE(s->width, s->height);
510 ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_TRANSFORM, blocks_w,
515 s->image[IMAGE_ROLE_COLOR_TRANSFORM].size_reduction = block_bits;
520 static int parse_transform_color_indexing(WebPContext *s)
523 int width_bits, index_size, ret, x;
526 index_size = get_bits(&s->gb, 8) + 1;
530 else if (index_size <= 4)
532 else if (index_size <= 16)
537 ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_INDEXING,
542 img = &s->image[IMAGE_ROLE_COLOR_INDEXING];
543 img->size_reduction = width_bits;
545 s->reduced_width = (s->width + ((1 << width_bits) - 1)) >> width_bits;
547 /* color index values are delta-coded */
548 ct = img->frame->data[0] + 4;
549 for (x = 4; x < img->frame->width * 4; x++, ct++)
555 static HuffReader *get_huffman_group(WebPContext *s, ImageContext *img,
558 ImageContext *gimg = &s->image[IMAGE_ROLE_ENTROPY];
561 if (gimg->size_reduction > 0) {
562 int group_x = x >> gimg->size_reduction;
563 int group_y = y >> gimg->size_reduction;
564 group = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 2);
567 return &img->huffman_groups[group * HUFFMAN_CODES_PER_META_CODE];
570 static av_always_inline void color_cache_put(ImageContext *img, uint32_t c)
572 uint32_t cache_idx = (0x1E35A7BD * c) >> (32 - img->color_cache_bits);
573 img->color_cache[cache_idx] = c;
576 static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role,
581 int i, j, ret, x, y, width;
583 img = &s->image[role];
587 img->frame = av_frame_alloc();
589 return AVERROR(ENOMEM);
592 img->frame->format = AV_PIX_FMT_ARGB;
593 img->frame->width = w;
594 img->frame->height = h;
596 if (role == IMAGE_ROLE_ARGB && !img->is_alpha_primary) {
597 ThreadFrame pt = { .f = img->frame };
598 ret = ff_thread_get_buffer(s->avctx, &pt, 0);
600 ret = av_frame_get_buffer(img->frame, 1);
604 if (get_bits1(&s->gb)) {
605 img->color_cache_bits = get_bits(&s->gb, 4);
606 if (img->color_cache_bits < 1 || img->color_cache_bits > 11) {
607 av_log(s->avctx, AV_LOG_ERROR, "invalid color cache bits: %d\n",
608 img->color_cache_bits);
609 return AVERROR_INVALIDDATA;
611 img->color_cache = av_mallocz_array(1 << img->color_cache_bits,
612 sizeof(*img->color_cache));
613 if (!img->color_cache)
614 return AVERROR(ENOMEM);
616 img->color_cache_bits = 0;
619 img->nb_huffman_groups = 1;
620 if (role == IMAGE_ROLE_ARGB && get_bits1(&s->gb)) {
621 ret = decode_entropy_image(s);
624 img->nb_huffman_groups = s->nb_huffman_groups;
626 img->huffman_groups = av_mallocz_array(img->nb_huffman_groups *
627 HUFFMAN_CODES_PER_META_CODE,
628 sizeof(*img->huffman_groups));
629 if (!img->huffman_groups)
630 return AVERROR(ENOMEM);
632 for (i = 0; i < img->nb_huffman_groups; i++) {
633 hg = &img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE];
634 for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++) {
635 int alphabet_size = alphabet_sizes[j];
636 if (!j && img->color_cache_bits > 0)
637 alphabet_size += 1 << img->color_cache_bits;
639 if (get_bits1(&s->gb)) {
640 read_huffman_code_simple(s, &hg[j]);
642 ret = read_huffman_code_normal(s, &hg[j], alphabet_size);
649 width = img->frame->width;
650 if (role == IMAGE_ROLE_ARGB && s->reduced_width > 0)
651 width = s->reduced_width;
654 while (y < img->frame->height) {
657 hg = get_huffman_group(s, img, x, y);
658 v = huff_reader_get_symbol(&hg[HUFF_IDX_GREEN], &s->gb);
659 if (v < NUM_LITERAL_CODES) {
660 /* literal pixel values */
661 uint8_t *p = GET_PIXEL(img->frame, x, y);
663 p[1] = huff_reader_get_symbol(&hg[HUFF_IDX_RED], &s->gb);
664 p[3] = huff_reader_get_symbol(&hg[HUFF_IDX_BLUE], &s->gb);
665 p[0] = huff_reader_get_symbol(&hg[HUFF_IDX_ALPHA], &s->gb);
666 if (img->color_cache_bits)
667 color_cache_put(img, AV_RB32(p));
673 } else if (v < NUM_LITERAL_CODES + NUM_LENGTH_CODES) {
674 /* LZ77 backwards mapping */
675 int prefix_code, length, distance, ref_x, ref_y;
677 /* parse length and distance */
678 prefix_code = v - NUM_LITERAL_CODES;
679 if (prefix_code < 4) {
680 length = prefix_code + 1;
682 int extra_bits = (prefix_code - 2) >> 1;
683 int offset = 2 + (prefix_code & 1) << extra_bits;
684 length = offset + get_bits(&s->gb, extra_bits) + 1;
686 prefix_code = huff_reader_get_symbol(&hg[HUFF_IDX_DIST], &s->gb);
687 if (prefix_code < 4) {
688 distance = prefix_code + 1;
690 int extra_bits = prefix_code - 2 >> 1;
691 int offset = 2 + (prefix_code & 1) << extra_bits;
692 distance = offset + get_bits(&s->gb, extra_bits) + 1;
695 /* find reference location */
696 if (distance <= NUM_SHORT_DISTANCES) {
697 int xi = lz77_distance_offsets[distance - 1][0];
698 int yi = lz77_distance_offsets[distance - 1][1];
699 distance = FFMAX(1, xi + yi * width);
701 distance -= NUM_SHORT_DISTANCES;
712 while (distance >= width) {
717 ref_x = width - distance;
720 ref_x = FFMAX(0, ref_x);
721 ref_y = FFMAX(0, ref_y);
724 * source and dest regions can overlap and wrap lines, so just
726 for (i = 0; i < length; i++) {
727 uint8_t *p_ref = GET_PIXEL(img->frame, ref_x, ref_y);
728 uint8_t *p = GET_PIXEL(img->frame, x, y);
731 if (img->color_cache_bits)
732 color_cache_put(img, AV_RB32(p));
739 if (ref_x == width) {
743 if (y == img->frame->height || ref_y == img->frame->height)
747 /* read from color cache */
748 uint8_t *p = GET_PIXEL(img->frame, x, y);
749 int cache_idx = v - (NUM_LITERAL_CODES + NUM_LENGTH_CODES);
751 if (!img->color_cache_bits) {
752 av_log(s->avctx, AV_LOG_ERROR, "color cache not found\n");
753 return AVERROR_INVALIDDATA;
755 if (cache_idx >= 1 << img->color_cache_bits) {
756 av_log(s->avctx, AV_LOG_ERROR,
757 "color cache index out-of-bounds\n");
758 return AVERROR_INVALIDDATA;
760 AV_WB32(p, img->color_cache[cache_idx]);
772 /* PRED_MODE_BLACK */
773 static void inv_predict_0(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
774 const uint8_t *p_t, const uint8_t *p_tr)
776 AV_WB32(p, 0xFF000000);
780 static void inv_predict_1(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
781 const uint8_t *p_t, const uint8_t *p_tr)
787 static void inv_predict_2(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
788 const uint8_t *p_t, const uint8_t *p_tr)
794 static void inv_predict_3(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
795 const uint8_t *p_t, const uint8_t *p_tr)
801 static void inv_predict_4(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
802 const uint8_t *p_t, const uint8_t *p_tr)
807 /* PRED_MODE_AVG_T_AVG_L_TR */
808 static void inv_predict_5(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
809 const uint8_t *p_t, const uint8_t *p_tr)
811 p[0] = p_t[0] + (p_l[0] + p_tr[0] >> 1) >> 1;
812 p[1] = p_t[1] + (p_l[1] + p_tr[1] >> 1) >> 1;
813 p[2] = p_t[2] + (p_l[2] + p_tr[2] >> 1) >> 1;
814 p[3] = p_t[3] + (p_l[3] + p_tr[3] >> 1) >> 1;
817 /* PRED_MODE_AVG_L_TL */
818 static void inv_predict_6(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
819 const uint8_t *p_t, const uint8_t *p_tr)
821 p[0] = p_l[0] + p_tl[0] >> 1;
822 p[1] = p_l[1] + p_tl[1] >> 1;
823 p[2] = p_l[2] + p_tl[2] >> 1;
824 p[3] = p_l[3] + p_tl[3] >> 1;
827 /* PRED_MODE_AVG_L_T */
828 static void inv_predict_7(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
829 const uint8_t *p_t, const uint8_t *p_tr)
831 p[0] = p_l[0] + p_t[0] >> 1;
832 p[1] = p_l[1] + p_t[1] >> 1;
833 p[2] = p_l[2] + p_t[2] >> 1;
834 p[3] = p_l[3] + p_t[3] >> 1;
837 /* PRED_MODE_AVG_TL_T */
838 static void inv_predict_8(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
839 const uint8_t *p_t, const uint8_t *p_tr)
841 p[0] = p_tl[0] + p_t[0] >> 1;
842 p[1] = p_tl[1] + p_t[1] >> 1;
843 p[2] = p_tl[2] + p_t[2] >> 1;
844 p[3] = p_tl[3] + p_t[3] >> 1;
847 /* PRED_MODE_AVG_T_TR */
848 static void inv_predict_9(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
849 const uint8_t *p_t, const uint8_t *p_tr)
851 p[0] = p_t[0] + p_tr[0] >> 1;
852 p[1] = p_t[1] + p_tr[1] >> 1;
853 p[2] = p_t[2] + p_tr[2] >> 1;
854 p[3] = p_t[3] + p_tr[3] >> 1;
857 /* PRED_MODE_AVG_AVG_L_TL_AVG_T_TR */
858 static void inv_predict_10(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
859 const uint8_t *p_t, const uint8_t *p_tr)
861 p[0] = (p_l[0] + p_tl[0] >> 1) + (p_t[0] + p_tr[0] >> 1) >> 1;
862 p[1] = (p_l[1] + p_tl[1] >> 1) + (p_t[1] + p_tr[1] >> 1) >> 1;
863 p[2] = (p_l[2] + p_tl[2] >> 1) + (p_t[2] + p_tr[2] >> 1) >> 1;
864 p[3] = (p_l[3] + p_tl[3] >> 1) + (p_t[3] + p_tr[3] >> 1) >> 1;
867 /* PRED_MODE_SELECT */
868 static void inv_predict_11(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
869 const uint8_t *p_t, const uint8_t *p_tr)
871 int diff = (FFABS(p_l[0] - p_tl[0]) - FFABS(p_t[0] - p_tl[0])) +
872 (FFABS(p_l[1] - p_tl[1]) - FFABS(p_t[1] - p_tl[1])) +
873 (FFABS(p_l[2] - p_tl[2]) - FFABS(p_t[2] - p_tl[2])) +
874 (FFABS(p_l[3] - p_tl[3]) - FFABS(p_t[3] - p_tl[3]));
881 /* PRED_MODE_ADD_SUBTRACT_FULL */
882 static void inv_predict_12(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
883 const uint8_t *p_t, const uint8_t *p_tr)
885 p[0] = av_clip_uint8(p_l[0] + p_t[0] - p_tl[0]);
886 p[1] = av_clip_uint8(p_l[1] + p_t[1] - p_tl[1]);
887 p[2] = av_clip_uint8(p_l[2] + p_t[2] - p_tl[2]);
888 p[3] = av_clip_uint8(p_l[3] + p_t[3] - p_tl[3]);
891 static av_always_inline uint8_t clamp_add_subtract_half(int a, int b, int c)
894 return av_clip_uint8(d + (d - c) / 2);
897 /* PRED_MODE_ADD_SUBTRACT_HALF */
898 static void inv_predict_13(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
899 const uint8_t *p_t, const uint8_t *p_tr)
901 p[0] = clamp_add_subtract_half(p_l[0], p_t[0], p_tl[0]);
902 p[1] = clamp_add_subtract_half(p_l[1], p_t[1], p_tl[1]);
903 p[2] = clamp_add_subtract_half(p_l[2], p_t[2], p_tl[2]);
904 p[3] = clamp_add_subtract_half(p_l[3], p_t[3], p_tl[3]);
907 typedef void (*inv_predict_func)(uint8_t *p, const uint8_t *p_l,
908 const uint8_t *p_tl, const uint8_t *p_t,
909 const uint8_t *p_tr);
911 static const inv_predict_func inverse_predict[14] = {
912 inv_predict_0, inv_predict_1, inv_predict_2, inv_predict_3,
913 inv_predict_4, inv_predict_5, inv_predict_6, inv_predict_7,
914 inv_predict_8, inv_predict_9, inv_predict_10, inv_predict_11,
915 inv_predict_12, inv_predict_13,
918 static void inverse_prediction(AVFrame *frame, enum PredictionMode m, int x, int y)
920 uint8_t *dec, *p_l, *p_tl, *p_t, *p_tr;
923 dec = GET_PIXEL(frame, x, y);
924 p_l = GET_PIXEL(frame, x - 1, y);
925 p_tl = GET_PIXEL(frame, x - 1, y - 1);
926 p_t = GET_PIXEL(frame, x, y - 1);
927 if (x == frame->width - 1)
928 p_tr = GET_PIXEL(frame, 0, y);
930 p_tr = GET_PIXEL(frame, x + 1, y - 1);
932 inverse_predict[m](p, p_l, p_tl, p_t, p_tr);
940 static int apply_predictor_transform(WebPContext *s)
942 ImageContext *img = &s->image[IMAGE_ROLE_ARGB];
943 ImageContext *pimg = &s->image[IMAGE_ROLE_PREDICTOR];
946 for (y = 0; y < img->frame->height; y++) {
947 for (x = 0; x < img->frame->width; x++) {
948 int tx = x >> pimg->size_reduction;
949 int ty = y >> pimg->size_reduction;
950 enum PredictionMode m = GET_PIXEL_COMP(pimg->frame, tx, ty, 2);
961 av_log(s->avctx, AV_LOG_ERROR,
962 "invalid predictor mode: %d\n", m);
963 return AVERROR_INVALIDDATA;
965 inverse_prediction(img->frame, m, x, y);
971 static av_always_inline uint8_t color_transform_delta(uint8_t color_pred,
974 return (int)ff_u8_to_s8(color_pred) * ff_u8_to_s8(color) >> 5;
977 static int apply_color_transform(WebPContext *s)
979 ImageContext *img, *cimg;
983 img = &s->image[IMAGE_ROLE_ARGB];
984 cimg = &s->image[IMAGE_ROLE_COLOR_TRANSFORM];
986 for (y = 0; y < img->frame->height; y++) {
987 for (x = 0; x < img->frame->width; x++) {
988 cx = x >> cimg->size_reduction;
989 cy = y >> cimg->size_reduction;
990 cp = GET_PIXEL(cimg->frame, cx, cy);
991 p = GET_PIXEL(img->frame, x, y);
993 p[1] += color_transform_delta(cp[3], p[2]);
994 p[3] += color_transform_delta(cp[2], p[2]) +
995 color_transform_delta(cp[1], p[1]);
1001 static int apply_subtract_green_transform(WebPContext *s)
1004 ImageContext *img = &s->image[IMAGE_ROLE_ARGB];
1006 for (y = 0; y < img->frame->height; y++) {
1007 for (x = 0; x < img->frame->width; x++) {
1008 uint8_t *p = GET_PIXEL(img->frame, x, y);
1016 static int apply_color_indexing_transform(WebPContext *s)
1023 img = &s->image[IMAGE_ROLE_ARGB];
1024 pal = &s->image[IMAGE_ROLE_COLOR_INDEXING];
1026 if (pal->size_reduction > 0) {
1029 int pixel_bits = 8 >> pal->size_reduction;
1031 line = av_malloc(img->frame->linesize[0]);
1033 return AVERROR(ENOMEM);
1035 for (y = 0; y < img->frame->height; y++) {
1036 p = GET_PIXEL(img->frame, 0, y);
1037 memcpy(line, p, img->frame->linesize[0]);
1038 init_get_bits(&gb_g, line, img->frame->linesize[0] * 8);
1039 skip_bits(&gb_g, 16);
1041 for (x = 0; x < img->frame->width; x++) {
1042 p = GET_PIXEL(img->frame, x, y);
1043 p[2] = get_bits(&gb_g, pixel_bits);
1045 if (i == 1 << pal->size_reduction) {
1046 skip_bits(&gb_g, 24);
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);
1058 if (i >= pal->frame->width) {
1059 av_log(s->avctx, AV_LOG_ERROR, "invalid palette index %d\n", i);
1060 return AVERROR_INVALIDDATA;
1062 pi = GET_PIXEL(pal->frame, i, 0);
1070 static int vp8_lossless_decode_frame(AVCodecContext *avctx, AVFrame *p,
1071 int *got_frame, uint8_t *data_start,
1072 unsigned int data_size, int is_alpha_chunk)
1074 WebPContext *s = avctx->priv_data;
1077 if (!is_alpha_chunk) {
1079 avctx->pix_fmt = AV_PIX_FMT_ARGB;
1082 ret = init_get_bits(&s->gb, data_start, data_size * 8);
1086 if (!is_alpha_chunk) {
1087 if (get_bits(&s->gb, 8) != 0x2F) {
1088 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless signature\n");
1089 return AVERROR_INVALIDDATA;
1092 w = get_bits(&s->gb, 14) + 1;
1093 h = get_bits(&s->gb, 14) + 1;
1094 if (s->width && s->width != w) {
1095 av_log(avctx, AV_LOG_WARNING, "Width mismatch. %d != %d\n",
1099 if (s->height && s->height != h) {
1100 av_log(avctx, AV_LOG_WARNING, "Height mismatch. %d != %d\n",
1105 ret = ff_set_dimensions(avctx, s->width, s->height);
1109 s->has_alpha = get_bits1(&s->gb);
1111 if (get_bits(&s->gb, 3) != 0x0) {
1112 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless version\n");
1113 return AVERROR_INVALIDDATA;
1116 if (!s->width || !s->height)
1122 /* parse transformations */
1123 s->nb_transforms = 0;
1124 s->reduced_width = 0;
1125 while (get_bits1(&s->gb)) {
1126 enum TransformType transform = get_bits(&s->gb, 2);
1127 s->transforms[s->nb_transforms++] = transform;
1128 switch (transform) {
1129 case PREDICTOR_TRANSFORM:
1130 ret = parse_transform_predictor(s);
1132 case COLOR_TRANSFORM:
1133 ret = parse_transform_color(s);
1135 case COLOR_INDEXING_TRANSFORM:
1136 ret = parse_transform_color_indexing(s);
1140 goto free_and_return;
1143 /* decode primary image */
1144 s->image[IMAGE_ROLE_ARGB].frame = p;
1146 s->image[IMAGE_ROLE_ARGB].is_alpha_primary = 1;
1147 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ARGB, w, h);
1149 goto free_and_return;
1151 /* apply transformations */
1152 for (i = s->nb_transforms - 1; i >= 0; i--) {
1153 switch (s->transforms[i]) {
1154 case PREDICTOR_TRANSFORM:
1155 ret = apply_predictor_transform(s);
1157 case COLOR_TRANSFORM:
1158 ret = apply_color_transform(s);
1160 case SUBTRACT_GREEN:
1161 ret = apply_subtract_green_transform(s);
1163 case COLOR_INDEXING_TRANSFORM:
1164 ret = apply_color_indexing_transform(s);
1168 goto free_and_return;
1172 p->pict_type = AV_PICTURE_TYPE_I;
1177 for (i = 0; i < IMAGE_ROLE_NB; i++)
1178 image_ctx_free(&s->image[i]);
1183 static void alpha_inverse_prediction(AVFrame *frame, enum AlphaFilter m)
1188 ls = frame->linesize[3];
1190 /* filter first row using horizontal filter */
1191 dec = frame->data[3] + 1;
1192 for (x = 1; x < frame->width; x++, dec++)
1195 /* filter first column using vertical filter */
1196 dec = frame->data[3] + ls;
1197 for (y = 1; y < frame->height; y++, dec += ls)
1198 *dec += *(dec - ls);
1200 /* filter the rest using the specified filter */
1202 case ALPHA_FILTER_HORIZONTAL:
1203 for (y = 1; y < frame->height; y++) {
1204 dec = frame->data[3] + y * ls + 1;
1205 for (x = 1; x < frame->width; x++, dec++)
1209 case ALPHA_FILTER_VERTICAL:
1210 for (y = 1; y < frame->height; y++) {
1211 dec = frame->data[3] + y * ls + 1;
1212 for (x = 1; x < frame->width; x++, dec++)
1213 *dec += *(dec - ls);
1216 case ALPHA_FILTER_GRADIENT:
1217 for (y = 1; y < frame->height; y++) {
1218 dec = frame->data[3] + y * ls + 1;
1219 for (x = 1; x < frame->width; x++, dec++)
1220 dec[0] += av_clip_uint8(*(dec - 1) + *(dec - ls) - *(dec - ls - 1));
1226 static int vp8_lossy_decode_alpha(AVCodecContext *avctx, AVFrame *p,
1227 uint8_t *data_start,
1228 unsigned int data_size)
1230 WebPContext *s = avctx->priv_data;
1233 if (s->alpha_compression == ALPHA_COMPRESSION_NONE) {
1236 bytestream2_init(&gb, data_start, data_size);
1237 for (y = 0; y < s->height; y++)
1238 bytestream2_get_buffer(&gb, p->data[3] + p->linesize[3] * y,
1240 } else if (s->alpha_compression == ALPHA_COMPRESSION_VP8L) {
1242 int alpha_got_frame = 0;
1244 s->alpha_frame = av_frame_alloc();
1245 if (!s->alpha_frame)
1246 return AVERROR(ENOMEM);
1248 ret = vp8_lossless_decode_frame(avctx, s->alpha_frame, &alpha_got_frame,
1249 data_start, data_size, 1);
1251 av_frame_free(&s->alpha_frame);
1254 if (!alpha_got_frame) {
1255 av_frame_free(&s->alpha_frame);
1256 return AVERROR_INVALIDDATA;
1259 /* copy green component of alpha image to alpha plane of primary image */
1260 for (y = 0; y < s->height; y++) {
1261 ap = GET_PIXEL(s->alpha_frame, 0, y) + 2;
1262 pp = p->data[3] + p->linesize[3] * y;
1263 for (x = 0; x < s->width; x++) {
1269 av_frame_free(&s->alpha_frame);
1272 /* apply alpha filtering */
1273 if (s->alpha_filter)
1274 alpha_inverse_prediction(p, s->alpha_filter);
1279 static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p,
1280 int *got_frame, uint8_t *data_start,
1281 unsigned int data_size)
1283 WebPContext *s = avctx->priv_data;
1287 if (!s->initialized) {
1288 ff_vp8_decode_init(avctx);
1291 avctx->pix_fmt = AV_PIX_FMT_YUVA420P;
1295 if (data_size > INT_MAX) {
1296 av_log(avctx, AV_LOG_ERROR, "unsupported chunk size\n");
1297 return AVERROR_PATCHWELCOME;
1300 av_init_packet(&pkt);
1301 pkt.data = data_start;
1302 pkt.size = data_size;
1304 ret = ff_vp8_decode_frame(avctx, p, got_frame, &pkt);
1306 ret = vp8_lossy_decode_alpha(avctx, p, s->alpha_data,
1307 s->alpha_data_size);
1314 static int webp_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
1317 AVFrame * const p = data;
1318 WebPContext *s = avctx->priv_data;
1321 uint32_t chunk_type, chunk_size;
1329 bytestream2_init(&gb, avpkt->data, avpkt->size);
1331 if (bytestream2_get_bytes_left(&gb) < 12)
1332 return AVERROR_INVALIDDATA;
1334 if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
1335 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
1336 return AVERROR_INVALIDDATA;
1339 chunk_size = bytestream2_get_le32(&gb);
1340 if (bytestream2_get_bytes_left(&gb) < chunk_size)
1341 return AVERROR_INVALIDDATA;
1343 if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) {
1344 av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n");
1345 return AVERROR_INVALIDDATA;
1348 while (bytestream2_get_bytes_left(&gb) > 0) {
1349 char chunk_str[5] = { 0 };
1351 chunk_type = bytestream2_get_le32(&gb);
1352 chunk_size = bytestream2_get_le32(&gb);
1353 if (chunk_size == UINT32_MAX)
1354 return AVERROR_INVALIDDATA;
1355 chunk_size += chunk_size & 1;
1357 if (bytestream2_get_bytes_left(&gb) < chunk_size)
1358 return AVERROR_INVALIDDATA;
1360 switch (chunk_type) {
1361 case MKTAG('V', 'P', '8', ' '):
1363 ret = vp8_lossy_decode_frame(avctx, p, got_frame,
1364 avpkt->data + bytestream2_tell(&gb),
1369 bytestream2_skip(&gb, chunk_size);
1371 case MKTAG('V', 'P', '8', 'L'):
1373 ret = vp8_lossless_decode_frame(avctx, p, got_frame,
1374 avpkt->data + bytestream2_tell(&gb),
1379 bytestream2_skip(&gb, chunk_size);
1381 case MKTAG('V', 'P', '8', 'X'):
1382 vp8x_flags = bytestream2_get_byte(&gb);
1383 bytestream2_skip(&gb, 3);
1384 s->width = bytestream2_get_le24(&gb) + 1;
1385 s->height = bytestream2_get_le24(&gb) + 1;
1386 ret = av_image_check_size(s->width, s->height, 0, avctx);
1390 case MKTAG('A', 'L', 'P', 'H'): {
1391 int alpha_header, filter_m, compression;
1393 if (!(vp8x_flags & VP8X_FLAG_ALPHA)) {
1394 av_log(avctx, AV_LOG_WARNING,
1395 "ALPHA chunk present, but alpha bit not set in the "
1398 if (chunk_size == 0) {
1399 av_log(avctx, AV_LOG_ERROR, "invalid ALPHA chunk size\n");
1400 return AVERROR_INVALIDDATA;
1402 alpha_header = bytestream2_get_byte(&gb);
1403 s->alpha_data = avpkt->data + bytestream2_tell(&gb);
1404 s->alpha_data_size = chunk_size - 1;
1405 bytestream2_skip(&gb, s->alpha_data_size);
1407 filter_m = (alpha_header >> 2) & 0x03;
1408 compression = alpha_header & 0x03;
1410 if (compression > ALPHA_COMPRESSION_VP8L) {
1411 av_log(avctx, AV_LOG_VERBOSE,
1412 "skipping unsupported ALPHA chunk\n");
1415 s->alpha_compression = compression;
1416 s->alpha_filter = filter_m;
1421 case MKTAG('I', 'C', 'C', 'P'):
1422 case MKTAG('A', 'N', 'I', 'M'):
1423 case MKTAG('A', 'N', 'M', 'F'):
1424 case MKTAG('E', 'X', 'I', 'F'):
1425 case MKTAG('X', 'M', 'P', ' '):
1426 AV_WL32(chunk_str, chunk_type);
1427 av_log(avctx, AV_LOG_VERBOSE, "skipping unsupported chunk: %s\n",
1429 bytestream2_skip(&gb, chunk_size);
1432 AV_WL32(chunk_str, chunk_type);
1433 av_log(avctx, AV_LOG_VERBOSE, "skipping unknown chunk: %s\n",
1435 bytestream2_skip(&gb, chunk_size);
1441 av_log(avctx, AV_LOG_ERROR, "image data not found\n");
1442 return AVERROR_INVALIDDATA;
1448 static av_cold int webp_decode_close(AVCodecContext *avctx)
1450 WebPContext *s = avctx->priv_data;
1453 return ff_vp8_decode_free(avctx);
1458 AVCodec ff_webp_decoder = {
1460 .long_name = NULL_IF_CONFIG_SMALL("WebP image"),
1461 .type = AVMEDIA_TYPE_VIDEO,
1462 .id = AV_CODEC_ID_WEBP,
1463 .priv_data_size = sizeof(WebPContext),
1464 .decode = webp_decode_frame,
1465 .close = webp_decode_close,
1466 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,