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 FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
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, sym, code, ret;
281 int max_code_length = 0;
284 for (sym = 0; sym < alphabet_size; sym++)
285 max_code_length = FFMAX(max_code_length, code_lengths[sym]);
287 if (max_code_length == 0 || max_code_length > MAX_HUFFMAN_CODE_LENGTH)
288 return AVERROR(EINVAL);
290 codes = av_malloc(alphabet_size * sizeof(*codes));
292 return AVERROR(ENOMEM);
296 for (len = 1; len <= max_code_length; len++) {
297 for (sym = 0; sym < alphabet_size; sym++) {
298 if (code_lengths[sym] != len)
305 if (!r->nb_symbols) {
307 return AVERROR_INVALIDDATA;
310 ret = init_vlc(&r->vlc, 8, alphabet_size,
311 code_lengths, sizeof(*code_lengths), sizeof(*code_lengths),
312 codes, sizeof(*codes), sizeof(*codes), 0);
323 static void read_huffman_code_simple(WebPContext *s, HuffReader *hc)
325 hc->nb_symbols = get_bits1(&s->gb) + 1;
327 if (get_bits1(&s->gb))
328 hc->simple_symbols[0] = get_bits(&s->gb, 8);
330 hc->simple_symbols[0] = get_bits1(&s->gb);
332 if (hc->nb_symbols == 2)
333 hc->simple_symbols[1] = get_bits(&s->gb, 8);
338 static int read_huffman_code_normal(WebPContext *s, HuffReader *hc,
341 HuffReader code_len_hc = { { 0 }, 0, 0, { 0 } };
342 int *code_lengths = NULL;
343 int code_length_code_lengths[NUM_CODE_LENGTH_CODES] = { 0 };
344 int i, symbol, max_symbol, prev_code_len, ret;
345 int num_codes = 4 + get_bits(&s->gb, 4);
347 if (num_codes > NUM_CODE_LENGTH_CODES)
348 return AVERROR_INVALIDDATA;
350 for (i = 0; i < num_codes; i++)
351 code_length_code_lengths[code_length_code_order[i]] = get_bits(&s->gb, 3);
353 ret = huff_reader_build_canonical(&code_len_hc, code_length_code_lengths,
354 NUM_CODE_LENGTH_CODES);
358 code_lengths = av_mallocz_array(alphabet_size, sizeof(*code_lengths));
360 ret = AVERROR(ENOMEM);
364 if (get_bits1(&s->gb)) {
365 int bits = 2 + 2 * get_bits(&s->gb, 3);
366 max_symbol = 2 + get_bits(&s->gb, bits);
367 if (max_symbol > alphabet_size) {
368 av_log(s->avctx, AV_LOG_ERROR, "max symbol %d > alphabet size %d\n",
369 max_symbol, alphabet_size);
370 ret = AVERROR_INVALIDDATA;
374 max_symbol = alphabet_size;
379 while (symbol < alphabet_size) {
384 code_len = huff_reader_get_symbol(&code_len_hc, &s->gb);
386 /* Code length code [0..15] indicates literal code lengths. */
387 code_lengths[symbol++] = code_len;
389 prev_code_len = code_len;
391 int repeat = 0, length = 0;
394 /* Code 16 repeats the previous non-zero value [3..6] times,
395 * i.e., 3 + ReadBits(2) times. If code 16 is used before a
396 * non-zero value has been emitted, a value of 8 is repeated. */
397 repeat = 3 + get_bits(&s->gb, 2);
398 length = prev_code_len;
401 /* Code 17 emits a streak of zeros [3..10], i.e.,
402 * 3 + ReadBits(3) times. */
403 repeat = 3 + get_bits(&s->gb, 3);
406 /* Code 18 emits a streak of zeros of length [11..138], i.e.,
407 * 11 + ReadBits(7) times. */
408 repeat = 11 + get_bits(&s->gb, 7);
411 if (symbol + repeat > alphabet_size) {
412 av_log(s->avctx, AV_LOG_ERROR,
413 "invalid symbol %d + repeat %d > alphabet size %d\n",
414 symbol, repeat, alphabet_size);
415 ret = AVERROR_INVALIDDATA;
419 code_lengths[symbol++] = length;
423 ret = huff_reader_build_canonical(hc, code_lengths, alphabet_size);
426 ff_free_vlc(&code_len_hc.vlc);
427 av_free(code_lengths);
431 static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role,
434 #define PARSE_BLOCK_SIZE(w, h) do { \
435 block_bits = get_bits(&s->gb, 3) + 2; \
436 blocks_w = FFALIGN((w), 1 << block_bits) >> block_bits; \
437 blocks_h = FFALIGN((h), 1 << block_bits) >> block_bits; \
440 static int decode_entropy_image(WebPContext *s)
443 int ret, block_bits, width, blocks_w, blocks_h, x, y, max;
446 if (s->reduced_width > 0)
447 width = s->reduced_width;
449 PARSE_BLOCK_SIZE(width, s->height);
451 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ENTROPY, blocks_w, blocks_h);
455 img = &s->image[IMAGE_ROLE_ENTROPY];
456 img->size_reduction = block_bits;
458 /* the number of huffman groups is determined by the maximum group number
459 * coded in the entropy image */
461 for (y = 0; y < img->frame->height; y++) {
462 for (x = 0; x < img->frame->width; x++) {
463 int p = GET_PIXEL_COMP(img->frame, x, y, 2);
467 s->nb_huffman_groups = max + 1;
472 static int parse_transform_predictor(WebPContext *s)
474 int block_bits, blocks_w, blocks_h, ret;
476 PARSE_BLOCK_SIZE(s->width, s->height);
478 ret = decode_entropy_coded_image(s, IMAGE_ROLE_PREDICTOR, blocks_w,
483 s->image[IMAGE_ROLE_PREDICTOR].size_reduction = block_bits;
488 static int parse_transform_color(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_COLOR_TRANSFORM, blocks_w,
499 s->image[IMAGE_ROLE_COLOR_TRANSFORM].size_reduction = block_bits;
504 static int parse_transform_color_indexing(WebPContext *s)
507 int width_bits, index_size, ret, x;
510 index_size = get_bits(&s->gb, 8) + 1;
514 else if (index_size <= 4)
516 else if (index_size <= 16)
521 ret = decode_entropy_coded_image(s, IMAGE_ROLE_COLOR_INDEXING,
526 img = &s->image[IMAGE_ROLE_COLOR_INDEXING];
527 img->size_reduction = width_bits;
529 s->reduced_width = (s->width + ((1 << width_bits) - 1)) >> width_bits;
531 /* color index values are delta-coded */
532 ct = img->frame->data[0] + 4;
533 for (x = 4; x < img->frame->width * 4; x++, ct++)
539 static HuffReader *get_huffman_group(WebPContext *s, ImageContext *img,
542 ImageContext *gimg = &s->image[IMAGE_ROLE_ENTROPY];
545 if (gimg->size_reduction > 0) {
546 int group_x = x >> gimg->size_reduction;
547 int group_y = y >> gimg->size_reduction;
548 group = GET_PIXEL_COMP(gimg->frame, group_x, group_y, 2);
551 return &img->huffman_groups[group * HUFFMAN_CODES_PER_META_CODE];
554 static av_always_inline void color_cache_put(ImageContext *img, uint32_t c)
556 uint32_t cache_idx = (0x1E35A7BD * c) >> (32 - img->color_cache_bits);
557 img->color_cache[cache_idx] = c;
560 static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role,
565 int i, j, ret, x, y, width;
567 img = &s->image[role];
571 img->frame = av_frame_alloc();
573 return AVERROR(ENOMEM);
576 img->frame->format = AV_PIX_FMT_ARGB;
577 img->frame->width = w;
578 img->frame->height = h;
580 if (role == IMAGE_ROLE_ARGB && !img->is_alpha_primary) {
581 ThreadFrame pt = { .f = img->frame };
582 ret = ff_thread_get_buffer(s->avctx, &pt, 0);
584 ret = av_frame_get_buffer(img->frame, 1);
588 if (get_bits1(&s->gb)) {
589 img->color_cache_bits = get_bits(&s->gb, 4);
590 if (img->color_cache_bits < 1 || img->color_cache_bits > 11) {
591 av_log(s->avctx, AV_LOG_ERROR, "invalid color cache bits: %d\n",
592 img->color_cache_bits);
593 return AVERROR_INVALIDDATA;
595 img->color_cache = av_mallocz_array(1 << img->color_cache_bits,
596 sizeof(*img->color_cache));
597 if (!img->color_cache)
598 return AVERROR(ENOMEM);
600 img->color_cache_bits = 0;
603 img->nb_huffman_groups = 1;
604 if (role == IMAGE_ROLE_ARGB && get_bits1(&s->gb)) {
605 ret = decode_entropy_image(s);
608 img->nb_huffman_groups = s->nb_huffman_groups;
610 img->huffman_groups = av_mallocz_array(img->nb_huffman_groups *
611 HUFFMAN_CODES_PER_META_CODE,
612 sizeof(*img->huffman_groups));
613 if (!img->huffman_groups)
614 return AVERROR(ENOMEM);
616 for (i = 0; i < img->nb_huffman_groups; i++) {
617 hg = &img->huffman_groups[i * HUFFMAN_CODES_PER_META_CODE];
618 for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; j++) {
619 int alphabet_size = alphabet_sizes[j];
620 if (!j && img->color_cache_bits > 0)
621 alphabet_size += 1 << img->color_cache_bits;
623 if (get_bits1(&s->gb)) {
624 read_huffman_code_simple(s, &hg[j]);
626 ret = read_huffman_code_normal(s, &hg[j], alphabet_size);
633 width = img->frame->width;
634 if (role == IMAGE_ROLE_ARGB && s->reduced_width > 0)
635 width = s->reduced_width;
638 while (y < img->frame->height) {
641 hg = get_huffman_group(s, img, x, y);
642 v = huff_reader_get_symbol(&hg[HUFF_IDX_GREEN], &s->gb);
643 if (v < NUM_LITERAL_CODES) {
644 /* literal pixel values */
645 uint8_t *p = GET_PIXEL(img->frame, x, y);
647 p[1] = huff_reader_get_symbol(&hg[HUFF_IDX_RED], &s->gb);
648 p[3] = huff_reader_get_symbol(&hg[HUFF_IDX_BLUE], &s->gb);
649 p[0] = huff_reader_get_symbol(&hg[HUFF_IDX_ALPHA], &s->gb);
650 if (img->color_cache_bits)
651 color_cache_put(img, AV_RB32(p));
657 } else if (v < NUM_LITERAL_CODES + NUM_LENGTH_CODES) {
658 /* LZ77 backwards mapping */
659 int prefix_code, length, distance, ref_x, ref_y;
661 /* parse length and distance */
662 prefix_code = v - NUM_LITERAL_CODES;
663 if (prefix_code < 4) {
664 length = prefix_code + 1;
666 int extra_bits = (prefix_code - 2) >> 1;
667 int offset = 2 + (prefix_code & 1) << extra_bits;
668 length = offset + get_bits(&s->gb, extra_bits) + 1;
670 prefix_code = huff_reader_get_symbol(&hg[HUFF_IDX_DIST], &s->gb);
671 if (prefix_code < 4) {
672 distance = prefix_code + 1;
674 int extra_bits = prefix_code - 2 >> 1;
675 int offset = 2 + (prefix_code & 1) << extra_bits;
676 distance = offset + get_bits(&s->gb, extra_bits) + 1;
679 /* find reference location */
680 if (distance <= NUM_SHORT_DISTANCES) {
681 int xi = lz77_distance_offsets[distance - 1][0];
682 int yi = lz77_distance_offsets[distance - 1][1];
683 distance = FFMAX(1, xi + yi * width);
685 distance -= NUM_SHORT_DISTANCES;
696 while (distance >= width) {
701 ref_x = width - distance;
704 ref_x = FFMAX(0, ref_x);
705 ref_y = FFMAX(0, ref_y);
708 * source and dest regions can overlap and wrap lines, so just
710 for (i = 0; i < length; i++) {
711 uint8_t *p_ref = GET_PIXEL(img->frame, ref_x, ref_y);
712 uint8_t *p = GET_PIXEL(img->frame, x, y);
715 if (img->color_cache_bits)
716 color_cache_put(img, AV_RB32(p));
723 if (ref_x == width) {
727 if (y == img->frame->height || ref_y == img->frame->height)
731 /* read from color cache */
732 uint8_t *p = GET_PIXEL(img->frame, x, y);
733 int cache_idx = v - (NUM_LITERAL_CODES + NUM_LENGTH_CODES);
735 if (!img->color_cache_bits) {
736 av_log(s->avctx, AV_LOG_ERROR, "color cache not found\n");
737 return AVERROR_INVALIDDATA;
739 if (cache_idx >= 1 << img->color_cache_bits) {
740 av_log(s->avctx, AV_LOG_ERROR,
741 "color cache index out-of-bounds\n");
742 return AVERROR_INVALIDDATA;
744 AV_WB32(p, img->color_cache[cache_idx]);
756 /* PRED_MODE_BLACK */
757 static void inv_predict_0(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
758 const uint8_t *p_t, const uint8_t *p_tr)
760 AV_WB32(p, 0xFF000000);
764 static void inv_predict_1(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
765 const uint8_t *p_t, const uint8_t *p_tr)
771 static void inv_predict_2(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
772 const uint8_t *p_t, const uint8_t *p_tr)
778 static void inv_predict_3(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
779 const uint8_t *p_t, const uint8_t *p_tr)
785 static void inv_predict_4(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
786 const uint8_t *p_t, const uint8_t *p_tr)
791 /* PRED_MODE_AVG_T_AVG_L_TR */
792 static void inv_predict_5(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
793 const uint8_t *p_t, const uint8_t *p_tr)
795 p[0] = p_t[0] + (p_l[0] + p_tr[0] >> 1) >> 1;
796 p[1] = p_t[1] + (p_l[1] + p_tr[1] >> 1) >> 1;
797 p[2] = p_t[2] + (p_l[2] + p_tr[2] >> 1) >> 1;
798 p[3] = p_t[3] + (p_l[3] + p_tr[3] >> 1) >> 1;
801 /* PRED_MODE_AVG_L_TL */
802 static void inv_predict_6(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
803 const uint8_t *p_t, const uint8_t *p_tr)
805 p[0] = p_l[0] + p_tl[0] >> 1;
806 p[1] = p_l[1] + p_tl[1] >> 1;
807 p[2] = p_l[2] + p_tl[2] >> 1;
808 p[3] = p_l[3] + p_tl[3] >> 1;
811 /* PRED_MODE_AVG_L_T */
812 static void inv_predict_7(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_l[0] + p_t[0] >> 1;
816 p[1] = p_l[1] + p_t[1] >> 1;
817 p[2] = p_l[2] + p_t[2] >> 1;
818 p[3] = p_l[3] + p_t[3] >> 1;
821 /* PRED_MODE_AVG_TL_T */
822 static void inv_predict_8(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_tl[0] + p_t[0] >> 1;
826 p[1] = p_tl[1] + p_t[1] >> 1;
827 p[2] = p_tl[2] + p_t[2] >> 1;
828 p[3] = p_tl[3] + p_t[3] >> 1;
831 /* PRED_MODE_AVG_T_TR */
832 static void inv_predict_9(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_t[0] + p_tr[0] >> 1;
836 p[1] = p_t[1] + p_tr[1] >> 1;
837 p[2] = p_t[2] + p_tr[2] >> 1;
838 p[3] = p_t[3] + p_tr[3] >> 1;
841 /* PRED_MODE_AVG_AVG_L_TL_AVG_T_TR */
842 static void inv_predict_10(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_l[0] + p_tl[0] >> 1) + (p_t[0] + p_tr[0] >> 1) >> 1;
846 p[1] = (p_l[1] + p_tl[1] >> 1) + (p_t[1] + p_tr[1] >> 1) >> 1;
847 p[2] = (p_l[2] + p_tl[2] >> 1) + (p_t[2] + p_tr[2] >> 1) >> 1;
848 p[3] = (p_l[3] + p_tl[3] >> 1) + (p_t[3] + p_tr[3] >> 1) >> 1;
851 /* PRED_MODE_SELECT */
852 static void inv_predict_11(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 int diff = (FFABS(p_l[0] - p_tl[0]) - FFABS(p_t[0] - p_tl[0])) +
856 (FFABS(p_l[1] - p_tl[1]) - FFABS(p_t[1] - p_tl[1])) +
857 (FFABS(p_l[2] - p_tl[2]) - FFABS(p_t[2] - p_tl[2])) +
858 (FFABS(p_l[3] - p_tl[3]) - FFABS(p_t[3] - p_tl[3]));
865 /* PRED_MODE_ADD_SUBTRACT_FULL */
866 static void inv_predict_12(uint8_t *p, const uint8_t *p_l, const uint8_t *p_tl,
867 const uint8_t *p_t, const uint8_t *p_tr)
869 p[0] = av_clip_uint8(p_l[0] + p_t[0] - p_tl[0]);
870 p[1] = av_clip_uint8(p_l[1] + p_t[1] - p_tl[1]);
871 p[2] = av_clip_uint8(p_l[2] + p_t[2] - p_tl[2]);
872 p[3] = av_clip_uint8(p_l[3] + p_t[3] - p_tl[3]);
875 static av_always_inline uint8_t clamp_add_subtract_half(int a, int b, int c)
878 return av_clip_uint8(d + (d - c) / 2);
881 /* PRED_MODE_ADD_SUBTRACT_HALF */
882 static void inv_predict_13(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] = clamp_add_subtract_half(p_l[0], p_t[0], p_tl[0]);
886 p[1] = clamp_add_subtract_half(p_l[1], p_t[1], p_tl[1]);
887 p[2] = clamp_add_subtract_half(p_l[2], p_t[2], p_tl[2]);
888 p[3] = clamp_add_subtract_half(p_l[3], p_t[3], p_tl[3]);
891 typedef void (*inv_predict_func)(uint8_t *p, const uint8_t *p_l,
892 const uint8_t *p_tl, const uint8_t *p_t,
893 const uint8_t *p_tr);
895 static const inv_predict_func inverse_predict[14] = {
896 inv_predict_0, inv_predict_1, inv_predict_2, inv_predict_3,
897 inv_predict_4, inv_predict_5, inv_predict_6, inv_predict_7,
898 inv_predict_8, inv_predict_9, inv_predict_10, inv_predict_11,
899 inv_predict_12, inv_predict_13,
902 static void inverse_prediction(AVFrame *frame, enum PredictionMode m, int x, int y)
904 uint8_t *dec, *p_l, *p_tl, *p_t, *p_tr;
907 dec = GET_PIXEL(frame, x, y);
908 p_l = GET_PIXEL(frame, x - 1, y);
909 p_tl = GET_PIXEL(frame, x - 1, y - 1);
910 p_t = GET_PIXEL(frame, x, y - 1);
911 if (x == frame->width - 1)
912 p_tr = GET_PIXEL(frame, 0, y);
914 p_tr = GET_PIXEL(frame, x + 1, y - 1);
916 inverse_predict[m](p, p_l, p_tl, p_t, p_tr);
924 static int apply_predictor_transform(WebPContext *s)
926 ImageContext *img = &s->image[IMAGE_ROLE_ARGB];
927 ImageContext *pimg = &s->image[IMAGE_ROLE_PREDICTOR];
930 for (y = 0; y < img->frame->height; y++) {
931 for (x = 0; x < img->frame->width; x++) {
932 int tx = x >> pimg->size_reduction;
933 int ty = y >> pimg->size_reduction;
934 enum PredictionMode m = GET_PIXEL_COMP(pimg->frame, tx, ty, 2);
945 av_log(s->avctx, AV_LOG_ERROR,
946 "invalid predictor mode: %d\n", m);
947 return AVERROR_INVALIDDATA;
949 inverse_prediction(img->frame, m, x, y);
955 static av_always_inline uint8_t color_transform_delta(uint8_t color_pred,
958 return (int)ff_u8_to_s8(color_pred) * ff_u8_to_s8(color) >> 5;
961 static int apply_color_transform(WebPContext *s)
963 ImageContext *img, *cimg;
967 img = &s->image[IMAGE_ROLE_ARGB];
968 cimg = &s->image[IMAGE_ROLE_COLOR_TRANSFORM];
970 for (y = 0; y < img->frame->height; y++) {
971 for (x = 0; x < img->frame->width; x++) {
972 cx = x >> cimg->size_reduction;
973 cy = y >> cimg->size_reduction;
974 cp = GET_PIXEL(cimg->frame, cx, cy);
975 p = GET_PIXEL(img->frame, x, y);
977 p[1] += color_transform_delta(cp[3], p[2]);
978 p[3] += color_transform_delta(cp[2], p[2]) +
979 color_transform_delta(cp[1], p[1]);
985 static int apply_subtract_green_transform(WebPContext *s)
988 ImageContext *img = &s->image[IMAGE_ROLE_ARGB];
990 for (y = 0; y < img->frame->height; y++) {
991 for (x = 0; x < img->frame->width; x++) {
992 uint8_t *p = GET_PIXEL(img->frame, x, y);
1000 static int apply_color_indexing_transform(WebPContext *s)
1007 img = &s->image[IMAGE_ROLE_ARGB];
1008 pal = &s->image[IMAGE_ROLE_COLOR_INDEXING];
1010 if (pal->size_reduction > 0) {
1013 int pixel_bits = 8 >> pal->size_reduction;
1015 line = av_malloc(img->frame->linesize[0]);
1017 return AVERROR(ENOMEM);
1019 for (y = 0; y < img->frame->height; y++) {
1020 p = GET_PIXEL(img->frame, 0, y);
1021 memcpy(line, p, img->frame->linesize[0]);
1022 init_get_bits(&gb_g, line, img->frame->linesize[0] * 8);
1023 skip_bits(&gb_g, 16);
1025 for (x = 0; x < img->frame->width; x++) {
1026 p = GET_PIXEL(img->frame, x, y);
1027 p[2] = get_bits(&gb_g, pixel_bits);
1029 if (i == 1 << pal->size_reduction) {
1030 skip_bits(&gb_g, 24);
1038 for (y = 0; y < img->frame->height; y++) {
1039 for (x = 0; x < img->frame->width; x++) {
1040 p = GET_PIXEL(img->frame, x, y);
1042 if (i >= pal->frame->width) {
1043 av_log(s->avctx, AV_LOG_ERROR, "invalid palette index %d\n", i);
1044 return AVERROR_INVALIDDATA;
1046 pi = GET_PIXEL(pal->frame, i, 0);
1054 static int vp8_lossless_decode_frame(AVCodecContext *avctx, AVFrame *p,
1055 int *got_frame, uint8_t *data_start,
1056 unsigned int data_size, int is_alpha_chunk)
1058 WebPContext *s = avctx->priv_data;
1061 if (!is_alpha_chunk) {
1063 avctx->pix_fmt = AV_PIX_FMT_ARGB;
1066 ret = init_get_bits(&s->gb, data_start, data_size * 8);
1070 if (!is_alpha_chunk) {
1071 if (get_bits(&s->gb, 8) != 0x2F) {
1072 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless signature\n");
1073 return AVERROR_INVALIDDATA;
1076 w = get_bits(&s->gb, 14) + 1;
1077 h = get_bits(&s->gb, 14) + 1;
1078 if (s->width && s->width != w) {
1079 av_log(avctx, AV_LOG_WARNING, "Width mismatch. %d != %d\n",
1083 if (s->height && s->height != h) {
1084 av_log(avctx, AV_LOG_WARNING, "Height mismatch. %d != %d\n",
1088 ret = av_image_check_size(s->width, s->height, 0, avctx);
1091 avcodec_set_dimensions(avctx, s->width, s->height);
1093 s->has_alpha = get_bits1(&s->gb);
1095 if (get_bits(&s->gb, 3) != 0x0) {
1096 av_log(avctx, AV_LOG_ERROR, "Invalid WebP Lossless version\n");
1097 return AVERROR_INVALIDDATA;
1100 if (!s->width || !s->height)
1106 /* parse transformations */
1107 s->nb_transforms = 0;
1108 s->reduced_width = 0;
1109 while (get_bits1(&s->gb)) {
1110 enum TransformType transform = get_bits(&s->gb, 2);
1111 s->transforms[s->nb_transforms++] = transform;
1112 switch (transform) {
1113 case PREDICTOR_TRANSFORM:
1114 ret = parse_transform_predictor(s);
1116 case COLOR_TRANSFORM:
1117 ret = parse_transform_color(s);
1119 case COLOR_INDEXING_TRANSFORM:
1120 ret = parse_transform_color_indexing(s);
1124 goto free_and_return;
1127 /* decode primary image */
1128 s->image[IMAGE_ROLE_ARGB].frame = p;
1130 s->image[IMAGE_ROLE_ARGB].is_alpha_primary = 1;
1131 ret = decode_entropy_coded_image(s, IMAGE_ROLE_ARGB, w, h);
1134 goto free_and_return;
1137 /* apply transformations */
1138 for (i = s->nb_transforms - 1; i >= 0; i--) {
1139 switch (s->transforms[i]) {
1140 case PREDICTOR_TRANSFORM:
1141 ret = apply_predictor_transform(s);
1143 case COLOR_TRANSFORM:
1144 ret = apply_color_transform(s);
1146 case SUBTRACT_GREEN:
1147 ret = apply_subtract_green_transform(s);
1149 case COLOR_INDEXING_TRANSFORM:
1150 ret = apply_color_indexing_transform(s);
1155 goto free_and_return;
1160 p->pict_type = AV_PICTURE_TYPE_I;
1165 for (i = 0; i < IMAGE_ROLE_NB; i++)
1166 image_ctx_free(&s->image[i]);
1171 static void alpha_inverse_prediction(AVFrame *frame, enum AlphaFilter m)
1176 ls = frame->linesize[3];
1178 /* filter first row using horizontal filter */
1179 dec = frame->data[3] + 1;
1180 for (x = 1; x < frame->width; x++, dec++)
1183 /* filter first column using vertical filter */
1184 dec = frame->data[3] + ls;
1185 for (y = 1; y < frame->height; y++, dec += ls)
1186 *dec += *(dec - ls);
1188 /* filter the rest using the specified filter */
1190 case ALPHA_FILTER_HORIZONTAL:
1191 for (y = 1; y < frame->height; y++) {
1192 dec = frame->data[3] + y * ls + 1;
1193 for (x = 1; x < frame->width; x++, dec++)
1197 case ALPHA_FILTER_VERTICAL:
1198 for (y = 1; y < frame->height; y++) {
1199 dec = frame->data[3] + y * ls + 1;
1200 for (x = 1; x < frame->width; x++, dec++)
1201 *dec += *(dec - ls);
1204 case ALPHA_FILTER_GRADIENT:
1205 for (y = 1; y < frame->height; y++) {
1206 dec = frame->data[3] + y * ls + 1;
1207 for (x = 1; x < frame->width; x++, dec++)
1208 dec[0] += av_clip_uint8(*(dec - 1) + *(dec - ls) - *(dec - ls - 1));
1214 static int vp8_lossy_decode_alpha(AVCodecContext *avctx, AVFrame *p,
1215 uint8_t *data_start,
1216 unsigned int data_size)
1218 WebPContext *s = avctx->priv_data;
1221 if (s->alpha_compression == ALPHA_COMPRESSION_NONE) {
1224 bytestream2_init(&gb, data_start, data_size);
1225 for (y = 0; y < s->height; y++)
1226 bytestream2_get_buffer(&gb, p->data[3] + p->linesize[3] * y,
1228 } else if (s->alpha_compression == ALPHA_COMPRESSION_VP8L) {
1230 int alpha_got_frame = 0;
1232 s->alpha_frame = av_frame_alloc();
1233 if (!s->alpha_frame)
1234 return AVERROR(ENOMEM);
1236 ret = vp8_lossless_decode_frame(avctx, s->alpha_frame, &alpha_got_frame,
1237 data_start, data_size, 1);
1239 av_frame_free(&s->alpha_frame);
1242 if (!alpha_got_frame) {
1243 av_frame_free(&s->alpha_frame);
1244 return AVERROR_INVALIDDATA;
1247 /* copy green component of alpha image to alpha plane of primary image */
1248 for (y = 0; y < s->height; y++) {
1249 ap = GET_PIXEL(s->alpha_frame, 0, y) + 2;
1250 pp = p->data[3] + p->linesize[3] * y;
1251 for (x = 0; x < s->width; x++) {
1257 av_frame_free(&s->alpha_frame);
1260 /* apply alpha filtering */
1261 if (s->alpha_filter)
1262 alpha_inverse_prediction(p, s->alpha_filter);
1267 static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p,
1268 int *got_frame, uint8_t *data_start,
1269 unsigned int data_size)
1271 WebPContext *s = avctx->priv_data;
1275 if (!s->initialized) {
1276 ff_vp8_decode_init(avctx);
1279 avctx->pix_fmt = AV_PIX_FMT_YUVA420P;
1283 if (data_size > INT_MAX) {
1284 av_log(avctx, AV_LOG_ERROR, "unsupported chunk size\n");
1285 return AVERROR_PATCHWELCOME;
1288 av_init_packet(&pkt);
1289 pkt.data = data_start;
1290 pkt.size = data_size;
1292 ret = ff_vp8_decode_frame(avctx, p, got_frame, &pkt);
1294 ret = vp8_lossy_decode_alpha(avctx, p, s->alpha_data,
1295 s->alpha_data_size);
1302 static int webp_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
1305 AVFrame * const p = data;
1306 WebPContext *s = avctx->priv_data;
1309 uint32_t chunk_type, chunk_size;
1317 bytestream2_init(&gb, avpkt->data, avpkt->size);
1319 if (bytestream2_get_bytes_left(&gb) < 12)
1320 return AVERROR_INVALIDDATA;
1322 if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
1323 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
1324 return AVERROR_INVALIDDATA;
1327 chunk_size = bytestream2_get_le32(&gb);
1328 if (bytestream2_get_bytes_left(&gb) < chunk_size)
1329 return AVERROR_INVALIDDATA;
1331 if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) {
1332 av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n");
1333 return AVERROR_INVALIDDATA;
1336 while (bytestream2_get_bytes_left(&gb) > 0) {
1337 char chunk_str[5] = { 0 };
1339 chunk_type = bytestream2_get_le32(&gb);
1340 chunk_size = bytestream2_get_le32(&gb);
1341 if (chunk_size == UINT32_MAX)
1342 return AVERROR_INVALIDDATA;
1343 chunk_size += chunk_size & 1;
1345 if (bytestream2_get_bytes_left(&gb) < chunk_size)
1346 return AVERROR_INVALIDDATA;
1348 switch (chunk_type) {
1349 case MKTAG('V', 'P', '8', ' '):
1351 ret = vp8_lossy_decode_frame(avctx, p, got_frame,
1352 avpkt->data + bytestream2_tell(&gb),
1357 bytestream2_skip(&gb, chunk_size);
1359 case MKTAG('V', 'P', '8', 'L'):
1361 ret = vp8_lossless_decode_frame(avctx, p, got_frame,
1362 avpkt->data + bytestream2_tell(&gb),
1367 bytestream2_skip(&gb, chunk_size);
1369 case MKTAG('V', 'P', '8', 'X'):
1370 vp8x_flags = bytestream2_get_byte(&gb);
1371 bytestream2_skip(&gb, 3);
1372 s->width = bytestream2_get_le24(&gb) + 1;
1373 s->height = bytestream2_get_le24(&gb) + 1;
1374 ret = av_image_check_size(s->width, s->height, 0, avctx);
1378 case MKTAG('A', 'L', 'P', 'H'): {
1379 int alpha_header, filter_m, compression;
1381 if (!(vp8x_flags & VP8X_FLAG_ALPHA)) {
1382 av_log(avctx, AV_LOG_WARNING,
1383 "ALPHA chunk present, but alpha bit not set in the "
1386 if (chunk_size == 0) {
1387 av_log(avctx, AV_LOG_ERROR, "invalid ALPHA chunk size\n");
1388 return AVERROR_INVALIDDATA;
1390 alpha_header = bytestream2_get_byte(&gb);
1391 s->alpha_data = avpkt->data + bytestream2_tell(&gb);
1392 s->alpha_data_size = chunk_size - 1;
1393 bytestream2_skip(&gb, s->alpha_data_size);
1395 filter_m = (alpha_header >> 2) & 0x03;
1396 compression = alpha_header & 0x03;
1398 if (compression > ALPHA_COMPRESSION_VP8L) {
1399 av_log(avctx, AV_LOG_VERBOSE,
1400 "skipping unsupported ALPHA chunk\n");
1403 s->alpha_compression = compression;
1404 s->alpha_filter = filter_m;
1409 case MKTAG('I', 'C', 'C', 'P'):
1410 case MKTAG('A', 'N', 'I', 'M'):
1411 case MKTAG('A', 'N', 'M', 'F'):
1412 case MKTAG('E', 'X', 'I', 'F'):
1413 case MKTAG('X', 'M', 'P', ' '):
1414 AV_WL32(chunk_str, chunk_type);
1415 av_log(avctx, AV_LOG_VERBOSE, "skipping unsupported chunk: %s\n",
1417 bytestream2_skip(&gb, chunk_size);
1420 AV_WL32(chunk_str, chunk_type);
1421 av_log(avctx, AV_LOG_VERBOSE, "skipping unknown chunk: %s\n",
1423 bytestream2_skip(&gb, chunk_size);
1429 av_log(avctx, AV_LOG_ERROR, "image data not found\n");
1430 return AVERROR_INVALIDDATA;
1436 static av_cold int webp_decode_close(AVCodecContext *avctx)
1438 WebPContext *s = avctx->priv_data;
1441 return ff_vp8_decode_free(avctx);
1446 AVCodec ff_webp_decoder = {
1448 .long_name = NULL_IF_CONFIG_SMALL("WebP image"),
1449 .type = AVMEDIA_TYPE_VIDEO,
1450 .id = AV_CODEC_ID_WEBP,
1451 .priv_data_size = sizeof(WebPContext),
1452 .decode = webp_decode_frame,
1453 .close = webp_decode_close,
1454 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_FRAME_THREADS,