3 * VP5 and VP6 compatible video decoder (common features)
5 * Copyright (C) 2006 Aurelien Jacobs <aurel@gnuage.org>
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 #include "bytestream.h"
31 void vp56_init_dequant(vp56_context_t *s, int quantizer)
33 s->quantizer = quantizer;
34 s->dequant_dc = vp56_dc_dequant[quantizer] << 2;
35 s->dequant_ac = vp56_ac_dequant[quantizer] << 2;
38 static int vp56_get_vectors_predictors(vp56_context_t *s, int row, int col,
39 vp56_frame_t ref_frame)
42 vp56_mv_t vect[2] = {{0,0}, {0,0}};
46 for (pos=0; pos<12; pos++) {
47 mvp.x = col + vp56_candidate_predictor_pos[pos][0];
48 mvp.y = row + vp56_candidate_predictor_pos[pos][1];
49 if (mvp.x < 0 || mvp.x >= s->mb_width ||
50 mvp.y < 0 || mvp.y >= s->mb_height)
52 offset = mvp.x + s->mb_width*mvp.y;
54 if (vp56_reference_frame[s->macroblocks[offset].type] != ref_frame)
56 if ((s->macroblocks[offset].mv.x == vect[0].x &&
57 s->macroblocks[offset].mv.y == vect[0].y) ||
58 (s->macroblocks[offset].mv.x == 0 &&
59 s->macroblocks[offset].mv.y == 0))
62 vect[nb_pred++] = s->macroblocks[offset].mv;
67 s->vector_candidate_pos = pos;
70 s->vector_candidate[0] = vect[0];
71 s->vector_candidate[1] = vect[1];
76 static void vp56_parse_mb_type_models(vp56_context_t *s)
78 vp56_range_coder_t *c = &s->c;
79 vp56_model_t *model = s->modelp;
82 for (ctx=0; ctx<3; ctx++) {
83 if (vp56_rac_get_prob(c, 174)) {
84 int idx = vp56_rac_gets(c, 4);
85 memcpy(model->mb_types_stats[ctx],
86 vp56_pre_def_mb_type_stats[idx][ctx],
87 sizeof(model->mb_types_stats[ctx]));
89 if (vp56_rac_get_prob(c, 254)) {
90 for (type=0; type<10; type++) {
92 if (vp56_rac_get_prob(c, 205)) {
93 int delta, sign = vp56_rac_get(c);
95 delta = vp56_rac_get_tree(c, vp56_pmbtm_tree,
96 vp56_mb_type_model_model);
98 delta = 4 * vp56_rac_gets(c, 7);
99 model->mb_types_stats[ctx][type][i] += (delta ^ -sign) + sign;
106 /* compute MB type probability tables based on previous MB type */
107 for (ctx=0; ctx<3; ctx++) {
110 for (type=0; type<10; type++)
111 p[type] = 100 * model->mb_types_stats[ctx][type][1];
113 for (type=0; type<10; type++) {
114 int p02, p34, p0234, p17, p56, p89, p5689, p156789;
116 /* conservative MB type probability */
117 model->mb_type[ctx][type][0] = 255 - (255 * model->mb_types_stats[ctx][type][0]) / (1 + model->mb_types_stats[ctx][type][0] + model->mb_types_stats[ctx][type][1]);
119 p[type] = 0; /* same MB type => weight is null */
121 /* binary tree parsing probabilities */
129 p156789 = p17 + p5689;
131 model->mb_type[ctx][type][1] = 1 + 255 * p0234/(1+p0234+p156789);
132 model->mb_type[ctx][type][2] = 1 + 255 * p02 / (1+p0234);
133 model->mb_type[ctx][type][3] = 1 + 255 * p17 / (1+p156789);
134 model->mb_type[ctx][type][4] = 1 + 255 * p[0] / (1+p02);
135 model->mb_type[ctx][type][5] = 1 + 255 * p[3] / (1+p34);
136 model->mb_type[ctx][type][6] = 1 + 255 * p[1] / (1+p17);
137 model->mb_type[ctx][type][7] = 1 + 255 * p56 / (1+p5689);
138 model->mb_type[ctx][type][8] = 1 + 255 * p[5] / (1+p56);
139 model->mb_type[ctx][type][9] = 1 + 255 * p[8] / (1+p89);
141 /* restore initial value */
142 p[type] = 100 * model->mb_types_stats[ctx][type][1];
147 static vp56_mb_t vp56_parse_mb_type(vp56_context_t *s,
148 vp56_mb_t prev_type, int ctx)
150 uint8_t *mb_type_model = s->modelp->mb_type[ctx][prev_type];
151 vp56_range_coder_t *c = &s->c;
153 if (vp56_rac_get_prob(c, mb_type_model[0]))
156 return vp56_rac_get_tree(c, vp56_pmbt_tree, mb_type_model);
159 static void vp56_decode_4mv(vp56_context_t *s, int row, int col)
161 vp56_mv_t mv = {0,0};
165 /* parse each block type */
166 for (b=0; b<4; b++) {
167 type[b] = vp56_rac_gets(&s->c, 2);
169 type[b]++; /* only returns 0, 2, 3 or 4 (all INTER_PF) */
173 for (b=0; b<4; b++) {
175 case VP56_MB_INTER_NOVEC_PF:
176 s->mv[b] = (vp56_mv_t) {0,0};
178 case VP56_MB_INTER_DELTA_PF:
179 s->parse_vector_adjustment(s, &s->mv[b]);
181 case VP56_MB_INTER_V1_PF:
182 s->mv[b] = s->vector_candidate[0];
184 case VP56_MB_INTER_V2_PF:
185 s->mv[b] = s->vector_candidate[1];
192 /* this is the one selected for the whole MB for prediction */
193 s->macroblocks[row * s->mb_width + col].mv = s->mv[3];
195 /* chroma vectors are average luma vectors */
196 if (s->avctx->codec->id == CODEC_ID_VP5) {
197 s->mv[4].x = s->mv[5].x = RSHIFT(mv.x,2);
198 s->mv[4].y = s->mv[5].y = RSHIFT(mv.y,2);
200 s->mv[4] = s->mv[5] = (vp56_mv_t) {mv.x/4, mv.y/4};
204 static vp56_mb_t vp56_decode_mv(vp56_context_t *s, int row, int col)
206 vp56_mv_t *mv, vect = {0,0};
209 ctx = vp56_get_vectors_predictors(s, row, col, VP56_FRAME_PREVIOUS);
210 s->mb_type = vp56_parse_mb_type(s, s->mb_type, ctx);
211 s->macroblocks[row * s->mb_width + col].type = s->mb_type;
213 switch (s->mb_type) {
214 case VP56_MB_INTER_V1_PF:
215 mv = &s->vector_candidate[0];
218 case VP56_MB_INTER_V2_PF:
219 mv = &s->vector_candidate[1];
222 case VP56_MB_INTER_V1_GF:
223 vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN);
224 mv = &s->vector_candidate[0];
227 case VP56_MB_INTER_V2_GF:
228 vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN);
229 mv = &s->vector_candidate[1];
232 case VP56_MB_INTER_DELTA_PF:
233 s->parse_vector_adjustment(s, &vect);
237 case VP56_MB_INTER_DELTA_GF:
238 vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN);
239 s->parse_vector_adjustment(s, &vect);
243 case VP56_MB_INTER_4V:
244 vp56_decode_4mv(s, row, col);
252 s->macroblocks[row*s->mb_width + col].mv = *mv;
254 /* same vector for all blocks */
261 static void vp56_add_predictors_dc(vp56_context_t *s, vp56_frame_t ref_frame)
263 int idx = s->scantable.permutated[0];
266 for (i=0; i<6; i++) {
267 vp56_ref_dc_t *ab = &s->above_blocks[s->above_block_idx[i]];
268 vp56_ref_dc_t *lb = &s->left_block[vp56_b6to4[i]];
272 if (ref_frame == lb->ref_frame) {
276 if (ref_frame == ab->ref_frame) {
280 if (s->avctx->codec->id == CODEC_ID_VP5) {
281 if (count < 2 && ref_frame == ab[-1].ref_frame) {
282 dc += ab[-1].dc_coeff;
285 if (count < 2 && ref_frame == ab[1].ref_frame) {
286 dc += ab[1].dc_coeff;
291 dc = s->prev_dc[vp56_b2p[i]][ref_frame];
295 s->block_coeff[i][idx] += dc;
296 s->prev_dc[vp56_b2p[i]][ref_frame] = s->block_coeff[i][idx];
297 ab->dc_coeff = s->block_coeff[i][idx];
298 ab->ref_frame = ref_frame;
299 lb->dc_coeff = s->block_coeff[i][idx];
300 lb->ref_frame = ref_frame;
301 s->block_coeff[i][idx] *= s->dequant_dc;
305 static void vp56_edge_filter(vp56_context_t *s, uint8_t *yuv,
306 int pix_inc, int line_inc, int t)
308 int pix2_inc = 2 * pix_inc;
311 for (i=0; i<12; i++) {
312 v = (yuv[-pix2_inc] + 3*(yuv[0]-yuv[-pix_inc]) - yuv[pix_inc] + 4) >>3;
314 yuv[-pix_inc] = av_clip_uint8(yuv[-pix_inc] + v);
315 yuv[0] = av_clip_uint8(yuv[0] - v);
320 static void vp56_deblock_filter(vp56_context_t *s, uint8_t *yuv,
321 int stride, int dx, int dy)
323 int t = vp56_filter_threshold[s->quantizer];
324 if (dx) vp56_edge_filter(s, yuv + 10-dx , 1, stride, t);
325 if (dy) vp56_edge_filter(s, yuv + stride*(10-dy), stride, 1, t);
328 static void vp56_mc(vp56_context_t *s, int b, int plane, uint8_t *src,
329 int stride, int x, int y)
331 uint8_t *dst=s->framep[VP56_FRAME_CURRENT]->data[plane]+s->block_offset[b];
334 int overlap_offset = 0;
335 int mask = s->vp56_coord_div[b] - 1;
336 int deblock_filtering = s->deblock_filtering;
340 if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
341 (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY
342 && !s->framep[VP56_FRAME_CURRENT]->key_frame))
343 deblock_filtering = 0;
345 dx = s->mv[b].x / s->vp56_coord_div[b];
346 dy = s->mv[b].y / s->vp56_coord_div[b];
355 if (x<0 || x+12>=s->plane_width[plane] ||
356 y<0 || y+12>=s->plane_height[plane]) {
357 ff_emulated_edge_mc(s->edge_emu_buffer,
358 src + s->block_offset[b] + (dy-2)*stride + (dx-2),
359 stride, 12, 12, x, y,
360 s->plane_width[plane],
361 s->plane_height[plane]);
362 src_block = s->edge_emu_buffer;
363 src_offset = 2 + 2*stride;
364 } else if (deblock_filtering) {
365 /* only need a 12x12 block, but there is no such dsp function, */
366 /* so copy a 16x12 block */
367 s->dsp.put_pixels_tab[0][0](s->edge_emu_buffer,
368 src + s->block_offset[b] + (dy-2)*stride + (dx-2),
370 src_block = s->edge_emu_buffer;
371 src_offset = 2 + 2*stride;
374 src_offset = s->block_offset[b] + dy*stride + dx;
377 if (deblock_filtering)
378 vp56_deblock_filter(s, src_block, stride, dx&7, dy&7);
380 if (s->mv[b].x & mask)
381 overlap_offset += (s->mv[b].x > 0) ? 1 : -1;
382 if (s->mv[b].y & mask)
383 overlap_offset += (s->mv[b].y > 0) ? stride : -stride;
385 if (overlap_offset) {
387 s->filter(s, dst, src_block, src_offset, src_offset+overlap_offset,
388 stride, s->mv[b], mask, s->filter_selection, b<4);
390 s->dsp.put_no_rnd_pixels_l2[1](dst, src_block+src_offset,
391 src_block+src_offset+overlap_offset,
394 s->dsp.put_pixels_tab[1][0](dst, src_block+src_offset, stride, 8);
398 static void vp56_decode_mb(vp56_context_t *s, int row, int col, int is_alpha)
400 AVFrame *frame_current, *frame_ref;
402 vp56_frame_t ref_frame;
403 int b, ab, b_max, plane, off;
405 if (s->framep[VP56_FRAME_CURRENT]->key_frame)
406 mb_type = VP56_MB_INTRA;
408 mb_type = vp56_decode_mv(s, row, col);
409 ref_frame = vp56_reference_frame[mb_type];
411 memset(s->block_coeff, 0, sizeof(s->block_coeff));
415 vp56_add_predictors_dc(s, ref_frame);
417 frame_current = s->framep[VP56_FRAME_CURRENT];
418 frame_ref = s->framep[ref_frame];
421 b_max = 6 - 2*is_alpha;
425 for (b=0; b<b_max; b++) {
426 plane = vp56_b2p[b+ab];
427 s->dsp.idct_put(frame_current->data[plane] + s->block_offset[b],
428 s->stride[plane], s->block_coeff[b]);
432 case VP56_MB_INTER_NOVEC_PF:
433 case VP56_MB_INTER_NOVEC_GF:
434 for (b=0; b<b_max; b++) {
435 plane = vp56_b2p[b+ab];
436 off = s->block_offset[b];
437 s->dsp.put_pixels_tab[1][0](frame_current->data[plane] + off,
438 frame_ref->data[plane] + off,
439 s->stride[plane], 8);
440 s->dsp.idct_add(frame_current->data[plane] + off,
441 s->stride[plane], s->block_coeff[b]);
445 case VP56_MB_INTER_DELTA_PF:
446 case VP56_MB_INTER_V1_PF:
447 case VP56_MB_INTER_V2_PF:
448 case VP56_MB_INTER_DELTA_GF:
449 case VP56_MB_INTER_4V:
450 case VP56_MB_INTER_V1_GF:
451 case VP56_MB_INTER_V2_GF:
452 for (b=0; b<b_max; b++) {
453 int x_off = b==1 || b==3 ? 8 : 0;
454 int y_off = b==2 || b==3 ? 8 : 0;
455 plane = vp56_b2p[b+ab];
456 vp56_mc(s, b, plane, frame_ref->data[plane], s->stride[plane],
457 16*col+x_off, 16*row+y_off);
458 s->dsp.idct_add(frame_current->data[plane] + s->block_offset[b],
459 s->stride[plane], s->block_coeff[b]);
465 static int vp56_size_changed(AVCodecContext *avctx)
467 vp56_context_t *s = avctx->priv_data;
468 int stride = s->framep[VP56_FRAME_CURRENT]->linesize[0];
471 s->plane_width[0] = s->plane_width[3] = avctx->coded_width;
472 s->plane_width[1] = s->plane_width[2] = avctx->coded_width/2;
473 s->plane_height[0] = s->plane_height[3] = avctx->coded_height;
474 s->plane_height[1] = s->plane_height[2] = avctx->coded_height/2;
477 s->stride[i] = s->flip * s->framep[VP56_FRAME_CURRENT]->linesize[i];
479 s->mb_width = (avctx->coded_width +15) / 16;
480 s->mb_height = (avctx->coded_height+15) / 16;
482 if (s->mb_width > 1000 || s->mb_height > 1000) {
483 av_log(avctx, AV_LOG_ERROR, "picture too big\n");
487 s->above_blocks = av_realloc(s->above_blocks,
488 (4*s->mb_width+6) * sizeof(*s->above_blocks));
489 s->macroblocks = av_realloc(s->macroblocks,
490 s->mb_width*s->mb_height*sizeof(*s->macroblocks));
491 av_free(s->edge_emu_buffer_alloc);
492 s->edge_emu_buffer_alloc = av_malloc(16*stride);
493 s->edge_emu_buffer = s->edge_emu_buffer_alloc;
495 s->edge_emu_buffer += 15 * stride;
500 int vp56_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
501 uint8_t *buf, int buf_size)
503 vp56_context_t *s = avctx->priv_data;
504 AVFrame *const p = s->framep[VP56_FRAME_CURRENT];
505 int is_alpha, alpha_offset;
508 alpha_offset = bytestream_get_be24(&buf);
512 for (is_alpha=0; is_alpha < 1+s->has_alpha; is_alpha++) {
513 int mb_row, mb_col, mb_row_flip, mb_offset = 0;
514 int block, y, uv, stride_y, stride_uv;
515 int golden_frame = 0;
518 s->modelp = &s->models[is_alpha];
520 res = s->parse_header(s, buf, buf_size, &golden_frame);
526 if (avctx->get_buffer(avctx, p) < 0) {
527 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
532 if (vp56_size_changed(avctx)) {
533 avctx->release_buffer(avctx, p);
539 p->pict_type = FF_I_TYPE;
540 s->default_models_init(s);
541 for (block=0; block<s->mb_height*s->mb_width; block++)
542 s->macroblocks[block].type = VP56_MB_INTRA;
544 p->pict_type = FF_P_TYPE;
545 vp56_parse_mb_type_models(s);
546 s->parse_vector_models(s);
547 s->mb_type = VP56_MB_INTER_NOVEC_PF;
550 s->parse_coeff_models(s);
552 memset(s->prev_dc, 0, sizeof(s->prev_dc));
553 s->prev_dc[1][VP56_FRAME_CURRENT] = 128;
554 s->prev_dc[2][VP56_FRAME_CURRENT] = 128;
556 for (block=0; block < 4*s->mb_width+6; block++) {
557 s->above_blocks[block].ref_frame = -1;
558 s->above_blocks[block].dc_coeff = 0;
559 s->above_blocks[block].not_null_dc = 0;
561 s->above_blocks[2*s->mb_width + 2].ref_frame = 0;
562 s->above_blocks[3*s->mb_width + 4].ref_frame = 0;
564 stride_y = p->linesize[0];
565 stride_uv = p->linesize[1];
570 /* main macroblocks loop */
571 for (mb_row=0; mb_row<s->mb_height; mb_row++) {
573 mb_row_flip = s->mb_height - mb_row - 1;
575 mb_row_flip = mb_row;
577 for (block=0; block<4; block++) {
578 s->left_block[block].ref_frame = -1;
579 s->left_block[block].dc_coeff = 0;
580 s->left_block[block].not_null_dc = 0;
582 memset(s->coeff_ctx, 0, sizeof(s->coeff_ctx));
583 memset(s->coeff_ctx_last, 24, sizeof(s->coeff_ctx_last));
585 s->above_block_idx[0] = 1;
586 s->above_block_idx[1] = 2;
587 s->above_block_idx[2] = 1;
588 s->above_block_idx[3] = 2;
589 s->above_block_idx[4] = 2*s->mb_width + 2 + 1;
590 s->above_block_idx[5] = 3*s->mb_width + 4 + 1;
592 s->block_offset[s->frbi] = (mb_row_flip*16 + mb_offset) * stride_y;
593 s->block_offset[s->srbi] = s->block_offset[s->frbi] + 8*stride_y;
594 s->block_offset[1] = s->block_offset[0] + 8;
595 s->block_offset[3] = s->block_offset[2] + 8;
596 s->block_offset[4] = (mb_row_flip*8 + mb_offset) * stride_uv;
597 s->block_offset[5] = s->block_offset[4];
599 for (mb_col=0; mb_col<s->mb_width; mb_col++) {
600 vp56_decode_mb(s, mb_row, mb_col, is_alpha);
602 for (y=0; y<4; y++) {
603 s->above_block_idx[y] += 2;
604 s->block_offset[y] += 16;
607 for (uv=4; uv<6; uv++) {
608 s->above_block_idx[uv] += 1;
609 s->block_offset[uv] += 8;
614 if (p->key_frame || golden_frame) {
615 if (s->framep[VP56_FRAME_GOLDEN]->data[0] &&
616 s->framep[VP56_FRAME_GOLDEN] != s->framep[VP56_FRAME_GOLDEN2])
617 avctx->release_buffer(avctx, s->framep[VP56_FRAME_GOLDEN]);
618 s->framep[VP56_FRAME_GOLDEN] = p;
622 FFSWAP(AVFrame *, s->framep[VP56_FRAME_GOLDEN],
623 s->framep[VP56_FRAME_GOLDEN2]);
625 buf_size -= alpha_offset;
629 if (s->framep[VP56_FRAME_PREVIOUS] == s->framep[VP56_FRAME_GOLDEN] ||
630 s->framep[VP56_FRAME_PREVIOUS] == s->framep[VP56_FRAME_GOLDEN2]) {
631 if (s->framep[VP56_FRAME_UNUSED] != s->framep[VP56_FRAME_GOLDEN] &&
632 s->framep[VP56_FRAME_UNUSED] != s->framep[VP56_FRAME_GOLDEN2])
633 FFSWAP(AVFrame *, s->framep[VP56_FRAME_PREVIOUS],
634 s->framep[VP56_FRAME_UNUSED]);
636 FFSWAP(AVFrame *, s->framep[VP56_FRAME_PREVIOUS],
637 s->framep[VP56_FRAME_UNUSED2]);
638 } else if (s->framep[VP56_FRAME_PREVIOUS]->data[0])
639 avctx->release_buffer(avctx, s->framep[VP56_FRAME_PREVIOUS]);
640 FFSWAP(AVFrame *, s->framep[VP56_FRAME_CURRENT],
641 s->framep[VP56_FRAME_PREVIOUS]);
643 *(AVFrame*)data = *p;
644 *data_size = sizeof(AVFrame);
649 void vp56_init(AVCodecContext *avctx, int flip, int has_alpha)
651 vp56_context_t *s = avctx->priv_data;
655 avctx->pix_fmt = has_alpha ? PIX_FMT_YUVA420P : PIX_FMT_YUV420P;
657 if (avctx->idct_algo == FF_IDCT_AUTO)
658 avctx->idct_algo = FF_IDCT_VP3;
659 dsputil_init(&s->dsp, avctx);
660 ff_init_scantable(s->dsp.idct_permutation, &s->scantable,ff_zigzag_direct);
662 avcodec_set_dimensions(avctx, 0, 0);
665 s->framep[i] = &s->frames[i];
666 s->framep[VP56_FRAME_UNUSED] = s->framep[VP56_FRAME_GOLDEN];
667 s->framep[VP56_FRAME_UNUSED2] = s->framep[VP56_FRAME_GOLDEN2];
668 s->edge_emu_buffer_alloc = NULL;
670 s->above_blocks = NULL;
671 s->macroblocks = NULL;
673 s->deblock_filtering = 1;
677 s->has_alpha = has_alpha;
689 int vp56_free(AVCodecContext *avctx)
691 vp56_context_t *s = avctx->priv_data;
693 av_free(s->above_blocks);
694 av_free(s->macroblocks);
695 av_free(s->edge_emu_buffer_alloc);
696 if (s->framep[VP56_FRAME_GOLDEN]->data[0])
697 avctx->release_buffer(avctx, s->framep[VP56_FRAME_GOLDEN]);
698 if (s->framep[VP56_FRAME_GOLDEN2]->data[0])
699 avctx->release_buffer(avctx, s->framep[VP56_FRAME_GOLDEN2]);
700 if (s->framep[VP56_FRAME_PREVIOUS]->data[0])
701 avctx->release_buffer(avctx, s->framep[VP56_FRAME_PREVIOUS]);