2 * @file libavcodec/vp56.c
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(VP56Context *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;
36 memset(s->qscale_table, quantizer, s->mb_width);
39 static int vp56_get_vectors_predictors(VP56Context *s, int row, int col,
43 VP56mv vect[2] = {{0,0}, {0,0}};
47 for (pos=0; pos<12; pos++) {
48 mvp.x = col + vp56_candidate_predictor_pos[pos][0];
49 mvp.y = row + vp56_candidate_predictor_pos[pos][1];
50 if (mvp.x < 0 || mvp.x >= s->mb_width ||
51 mvp.y < 0 || mvp.y >= s->mb_height)
53 offset = mvp.x + s->mb_width*mvp.y;
55 if (vp56_reference_frame[s->macroblocks[offset].type] != ref_frame)
57 if ((s->macroblocks[offset].mv.x == vect[0].x &&
58 s->macroblocks[offset].mv.y == vect[0].y) ||
59 (s->macroblocks[offset].mv.x == 0 &&
60 s->macroblocks[offset].mv.y == 0))
63 vect[nb_pred++] = s->macroblocks[offset].mv;
68 s->vector_candidate_pos = pos;
71 s->vector_candidate[0] = vect[0];
72 s->vector_candidate[1] = vect[1];
77 static void vp56_parse_mb_type_models(VP56Context *s)
79 VP56RangeCoder *c = &s->c;
80 VP56Model *model = s->modelp;
83 for (ctx=0; ctx<3; ctx++) {
84 if (vp56_rac_get_prob(c, 174)) {
85 int idx = vp56_rac_gets(c, 4);
86 memcpy(model->mb_types_stats[ctx],
87 vp56_pre_def_mb_type_stats[idx][ctx],
88 sizeof(model->mb_types_stats[ctx]));
90 if (vp56_rac_get_prob(c, 254)) {
91 for (type=0; type<10; type++) {
93 if (vp56_rac_get_prob(c, 205)) {
94 int delta, sign = vp56_rac_get(c);
96 delta = vp56_rac_get_tree(c, vp56_pmbtm_tree,
97 vp56_mb_type_model_model);
99 delta = 4 * vp56_rac_gets(c, 7);
100 model->mb_types_stats[ctx][type][i] += (delta ^ -sign) + sign;
107 /* compute MB type probability tables based on previous MB type */
108 for (ctx=0; ctx<3; ctx++) {
111 for (type=0; type<10; type++)
112 p[type] = 100 * model->mb_types_stats[ctx][type][1];
114 for (type=0; type<10; type++) {
115 int p02, p34, p0234, p17, p56, p89, p5689, p156789;
117 /* conservative MB type probability */
118 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]);
120 p[type] = 0; /* same MB type => weight is null */
122 /* binary tree parsing probabilities */
130 p156789 = p17 + p5689;
132 model->mb_type[ctx][type][1] = 1 + 255 * p0234/(1+p0234+p156789);
133 model->mb_type[ctx][type][2] = 1 + 255 * p02 / (1+p0234);
134 model->mb_type[ctx][type][3] = 1 + 255 * p17 / (1+p156789);
135 model->mb_type[ctx][type][4] = 1 + 255 * p[0] / (1+p02);
136 model->mb_type[ctx][type][5] = 1 + 255 * p[3] / (1+p34);
137 model->mb_type[ctx][type][6] = 1 + 255 * p[1] / (1+p17);
138 model->mb_type[ctx][type][7] = 1 + 255 * p56 / (1+p5689);
139 model->mb_type[ctx][type][8] = 1 + 255 * p[5] / (1+p56);
140 model->mb_type[ctx][type][9] = 1 + 255 * p[8] / (1+p89);
142 /* restore initial value */
143 p[type] = 100 * model->mb_types_stats[ctx][type][1];
148 static VP56mb vp56_parse_mb_type(VP56Context *s,
149 VP56mb prev_type, int ctx)
151 uint8_t *mb_type_model = s->modelp->mb_type[ctx][prev_type];
152 VP56RangeCoder *c = &s->c;
154 if (vp56_rac_get_prob(c, mb_type_model[0]))
157 return vp56_rac_get_tree(c, vp56_pmbt_tree, mb_type_model);
160 static void vp56_decode_4mv(VP56Context *s, int row, int col)
166 /* parse each block type */
167 for (b=0; b<4; b++) {
168 type[b] = vp56_rac_gets(&s->c, 2);
170 type[b]++; /* only returns 0, 2, 3 or 4 (all INTER_PF) */
174 for (b=0; b<4; b++) {
176 case VP56_MB_INTER_NOVEC_PF:
177 s->mv[b] = (VP56mv) {0,0};
179 case VP56_MB_INTER_DELTA_PF:
180 s->parse_vector_adjustment(s, &s->mv[b]);
182 case VP56_MB_INTER_V1_PF:
183 s->mv[b] = s->vector_candidate[0];
185 case VP56_MB_INTER_V2_PF:
186 s->mv[b] = s->vector_candidate[1];
193 /* this is the one selected for the whole MB for prediction */
194 s->macroblocks[row * s->mb_width + col].mv = s->mv[3];
196 /* chroma vectors are average luma vectors */
197 if (s->avctx->codec->id == CODEC_ID_VP5) {
198 s->mv[4].x = s->mv[5].x = RSHIFT(mv.x,2);
199 s->mv[4].y = s->mv[5].y = RSHIFT(mv.y,2);
201 s->mv[4] = s->mv[5] = (VP56mv) {mv.x/4, mv.y/4};
205 static VP56mb vp56_decode_mv(VP56Context *s, int row, int col)
207 VP56mv *mv, vect = {0,0};
210 ctx = vp56_get_vectors_predictors(s, row, col, VP56_FRAME_PREVIOUS);
211 s->mb_type = vp56_parse_mb_type(s, s->mb_type, ctx);
212 s->macroblocks[row * s->mb_width + col].type = s->mb_type;
214 switch (s->mb_type) {
215 case VP56_MB_INTER_V1_PF:
216 mv = &s->vector_candidate[0];
219 case VP56_MB_INTER_V2_PF:
220 mv = &s->vector_candidate[1];
223 case VP56_MB_INTER_V1_GF:
224 vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN);
225 mv = &s->vector_candidate[0];
228 case VP56_MB_INTER_V2_GF:
229 vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN);
230 mv = &s->vector_candidate[1];
233 case VP56_MB_INTER_DELTA_PF:
234 s->parse_vector_adjustment(s, &vect);
238 case VP56_MB_INTER_DELTA_GF:
239 vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN);
240 s->parse_vector_adjustment(s, &vect);
244 case VP56_MB_INTER_4V:
245 vp56_decode_4mv(s, row, col);
253 s->macroblocks[row*s->mb_width + col].mv = *mv;
255 /* same vector for all blocks */
262 static void vp56_add_predictors_dc(VP56Context *s, VP56Frame ref_frame)
264 int idx = s->scantable.permutated[0];
267 for (b=0; b<6; b++) {
268 VP56RefDc *ab = &s->above_blocks[s->above_block_idx[b]];
269 VP56RefDc *lb = &s->left_block[vp56_b6to4[b]];
274 if (ref_frame == lb->ref_frame) {
278 if (ref_frame == ab->ref_frame) {
282 if (s->avctx->codec->id == CODEC_ID_VP5)
284 if (count < 2 && ref_frame == ab[-1+2*i].ref_frame) {
285 dc += ab[-1+2*i].dc_coeff;
289 dc = s->prev_dc[vp56_b2p[b]][ref_frame];
293 s->block_coeff[b][idx] += dc;
294 s->prev_dc[vp56_b2p[b]][ref_frame] = s->block_coeff[b][idx];
295 ab->dc_coeff = s->block_coeff[b][idx];
296 ab->ref_frame = ref_frame;
297 lb->dc_coeff = s->block_coeff[b][idx];
298 lb->ref_frame = ref_frame;
299 s->block_coeff[b][idx] *= s->dequant_dc;
303 static void vp56_edge_filter(VP56Context *s, uint8_t *yuv,
304 int pix_inc, int line_inc, int t)
306 int pix2_inc = 2 * pix_inc;
309 for (i=0; i<12; i++) {
310 v = (yuv[-pix2_inc] + 3*(yuv[0]-yuv[-pix_inc]) - yuv[pix_inc] + 4) >>3;
312 yuv[-pix_inc] = av_clip_uint8(yuv[-pix_inc] + v);
313 yuv[0] = av_clip_uint8(yuv[0] - v);
318 static void vp56_deblock_filter(VP56Context *s, uint8_t *yuv,
319 int stride, int dx, int dy)
321 int t = vp56_filter_threshold[s->quantizer];
322 if (dx) vp56_edge_filter(s, yuv + 10-dx , 1, stride, t);
323 if (dy) vp56_edge_filter(s, yuv + stride*(10-dy), stride, 1, t);
326 static void vp56_mc(VP56Context *s, int b, int plane, uint8_t *src,
327 int stride, int x, int y)
329 uint8_t *dst=s->framep[VP56_FRAME_CURRENT]->data[plane]+s->block_offset[b];
332 int overlap_offset = 0;
333 int mask = s->vp56_coord_div[b] - 1;
334 int deblock_filtering = s->deblock_filtering;
338 if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
339 (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY
340 && !s->framep[VP56_FRAME_CURRENT]->key_frame))
341 deblock_filtering = 0;
343 dx = s->mv[b].x / s->vp56_coord_div[b];
344 dy = s->mv[b].y / s->vp56_coord_div[b];
353 if (x<0 || x+12>=s->plane_width[plane] ||
354 y<0 || y+12>=s->plane_height[plane]) {
355 ff_emulated_edge_mc(s->edge_emu_buffer,
356 src + s->block_offset[b] + (dy-2)*stride + (dx-2),
357 stride, 12, 12, x, y,
358 s->plane_width[plane],
359 s->plane_height[plane]);
360 src_block = s->edge_emu_buffer;
361 src_offset = 2 + 2*stride;
362 } else if (deblock_filtering) {
363 /* only need a 12x12 block, but there is no such dsp function, */
364 /* so copy a 16x12 block */
365 s->dsp.put_pixels_tab[0][0](s->edge_emu_buffer,
366 src + s->block_offset[b] + (dy-2)*stride + (dx-2),
368 src_block = s->edge_emu_buffer;
369 src_offset = 2 + 2*stride;
372 src_offset = s->block_offset[b] + dy*stride + dx;
375 if (deblock_filtering)
376 vp56_deblock_filter(s, src_block, stride, dx&7, dy&7);
378 if (s->mv[b].x & mask)
379 overlap_offset += (s->mv[b].x > 0) ? 1 : -1;
380 if (s->mv[b].y & mask)
381 overlap_offset += (s->mv[b].y > 0) ? stride : -stride;
383 if (overlap_offset) {
385 s->filter(s, dst, src_block, src_offset, src_offset+overlap_offset,
386 stride, s->mv[b], mask, s->filter_selection, b<4);
388 s->dsp.put_no_rnd_pixels_l2[1](dst, src_block+src_offset,
389 src_block+src_offset+overlap_offset,
392 s->dsp.put_pixels_tab[1][0](dst, src_block+src_offset, stride, 8);
396 static void vp56_decode_mb(VP56Context *s, int row, int col, int is_alpha)
398 AVFrame *frame_current, *frame_ref;
401 int b, ab, b_max, plane, off;
403 if (s->framep[VP56_FRAME_CURRENT]->key_frame)
404 mb_type = VP56_MB_INTRA;
406 mb_type = vp56_decode_mv(s, row, col);
407 ref_frame = vp56_reference_frame[mb_type];
409 s->dsp.clear_blocks(*s->block_coeff);
413 vp56_add_predictors_dc(s, ref_frame);
415 frame_current = s->framep[VP56_FRAME_CURRENT];
416 frame_ref = s->framep[ref_frame];
419 b_max = 6 - 2*is_alpha;
423 for (b=0; b<b_max; b++) {
424 plane = vp56_b2p[b+ab];
425 s->dsp.idct_put(frame_current->data[plane] + s->block_offset[b],
426 s->stride[plane], s->block_coeff[b]);
430 case VP56_MB_INTER_NOVEC_PF:
431 case VP56_MB_INTER_NOVEC_GF:
432 for (b=0; b<b_max; b++) {
433 plane = vp56_b2p[b+ab];
434 off = s->block_offset[b];
435 s->dsp.put_pixels_tab[1][0](frame_current->data[plane] + off,
436 frame_ref->data[plane] + off,
437 s->stride[plane], 8);
438 s->dsp.idct_add(frame_current->data[plane] + off,
439 s->stride[plane], s->block_coeff[b]);
443 case VP56_MB_INTER_DELTA_PF:
444 case VP56_MB_INTER_V1_PF:
445 case VP56_MB_INTER_V2_PF:
446 case VP56_MB_INTER_DELTA_GF:
447 case VP56_MB_INTER_4V:
448 case VP56_MB_INTER_V1_GF:
449 case VP56_MB_INTER_V2_GF:
450 for (b=0; b<b_max; b++) {
451 int x_off = b==1 || b==3 ? 8 : 0;
452 int y_off = b==2 || b==3 ? 8 : 0;
453 plane = vp56_b2p[b+ab];
454 vp56_mc(s, b, plane, frame_ref->data[plane], s->stride[plane],
455 16*col+x_off, 16*row+y_off);
456 s->dsp.idct_add(frame_current->data[plane] + s->block_offset[b],
457 s->stride[plane], s->block_coeff[b]);
463 static int vp56_size_changed(AVCodecContext *avctx)
465 VP56Context *s = avctx->priv_data;
466 int stride = s->framep[VP56_FRAME_CURRENT]->linesize[0];
469 s->plane_width[0] = s->plane_width[3] = avctx->coded_width;
470 s->plane_width[1] = s->plane_width[2] = avctx->coded_width/2;
471 s->plane_height[0] = s->plane_height[3] = avctx->coded_height;
472 s->plane_height[1] = s->plane_height[2] = avctx->coded_height/2;
475 s->stride[i] = s->flip * s->framep[VP56_FRAME_CURRENT]->linesize[i];
477 s->mb_width = (avctx->coded_width +15) / 16;
478 s->mb_height = (avctx->coded_height+15) / 16;
480 if (s->mb_width > 1000 || s->mb_height > 1000) {
481 av_log(avctx, AV_LOG_ERROR, "picture too big\n");
485 s->qscale_table = av_realloc(s->qscale_table, s->mb_width);
486 s->above_blocks = av_realloc(s->above_blocks,
487 (4*s->mb_width+6) * sizeof(*s->above_blocks));
488 s->macroblocks = av_realloc(s->macroblocks,
489 s->mb_width*s->mb_height*sizeof(*s->macroblocks));
490 av_free(s->edge_emu_buffer_alloc);
491 s->edge_emu_buffer_alloc = av_malloc(16*stride);
492 s->edge_emu_buffer = s->edge_emu_buffer_alloc;
494 s->edge_emu_buffer += 15 * stride;
499 int vp56_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
502 const uint8_t *buf = avpkt->data;
503 VP56Context *s = avctx->priv_data;
504 AVFrame *const p = s->framep[VP56_FRAME_CURRENT];
505 int remaining_buf_size = avpkt->size;
506 int is_alpha, av_uninit(alpha_offset);
509 if (remaining_buf_size < 3)
511 alpha_offset = bytestream_get_be24(&buf);
512 remaining_buf_size -= 3;
513 if (remaining_buf_size < alpha_offset)
517 for (is_alpha=0; is_alpha < 1+s->has_alpha; is_alpha++) {
518 int mb_row, mb_col, mb_row_flip, mb_offset = 0;
519 int block, y, uv, stride_y, stride_uv;
520 int golden_frame = 0;
523 s->modelp = &s->models[is_alpha];
525 res = s->parse_header(s, buf, remaining_buf_size, &golden_frame);
531 if (avctx->get_buffer(avctx, p) < 0) {
532 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
537 if (vp56_size_changed(avctx)) {
538 avctx->release_buffer(avctx, p);
544 p->pict_type = FF_I_TYPE;
545 s->default_models_init(s);
546 for (block=0; block<s->mb_height*s->mb_width; block++)
547 s->macroblocks[block].type = VP56_MB_INTRA;
549 p->pict_type = FF_P_TYPE;
550 vp56_parse_mb_type_models(s);
551 s->parse_vector_models(s);
552 s->mb_type = VP56_MB_INTER_NOVEC_PF;
555 s->parse_coeff_models(s);
557 memset(s->prev_dc, 0, sizeof(s->prev_dc));
558 s->prev_dc[1][VP56_FRAME_CURRENT] = 128;
559 s->prev_dc[2][VP56_FRAME_CURRENT] = 128;
561 for (block=0; block < 4*s->mb_width+6; block++) {
562 s->above_blocks[block].ref_frame = VP56_FRAME_NONE;
563 s->above_blocks[block].dc_coeff = 0;
564 s->above_blocks[block].not_null_dc = 0;
566 s->above_blocks[2*s->mb_width + 2].ref_frame = VP56_FRAME_CURRENT;
567 s->above_blocks[3*s->mb_width + 4].ref_frame = VP56_FRAME_CURRENT;
569 stride_y = p->linesize[0];
570 stride_uv = p->linesize[1];
575 /* main macroblocks loop */
576 for (mb_row=0; mb_row<s->mb_height; mb_row++) {
578 mb_row_flip = s->mb_height - mb_row - 1;
580 mb_row_flip = mb_row;
582 for (block=0; block<4; block++) {
583 s->left_block[block].ref_frame = VP56_FRAME_NONE;
584 s->left_block[block].dc_coeff = 0;
585 s->left_block[block].not_null_dc = 0;
587 memset(s->coeff_ctx, 0, sizeof(s->coeff_ctx));
588 memset(s->coeff_ctx_last, 24, sizeof(s->coeff_ctx_last));
590 s->above_block_idx[0] = 1;
591 s->above_block_idx[1] = 2;
592 s->above_block_idx[2] = 1;
593 s->above_block_idx[3] = 2;
594 s->above_block_idx[4] = 2*s->mb_width + 2 + 1;
595 s->above_block_idx[5] = 3*s->mb_width + 4 + 1;
597 s->block_offset[s->frbi] = (mb_row_flip*16 + mb_offset) * stride_y;
598 s->block_offset[s->srbi] = s->block_offset[s->frbi] + 8*stride_y;
599 s->block_offset[1] = s->block_offset[0] + 8;
600 s->block_offset[3] = s->block_offset[2] + 8;
601 s->block_offset[4] = (mb_row_flip*8 + mb_offset) * stride_uv;
602 s->block_offset[5] = s->block_offset[4];
604 for (mb_col=0; mb_col<s->mb_width; mb_col++) {
605 vp56_decode_mb(s, mb_row, mb_col, is_alpha);
607 for (y=0; y<4; y++) {
608 s->above_block_idx[y] += 2;
609 s->block_offset[y] += 16;
612 for (uv=4; uv<6; uv++) {
613 s->above_block_idx[uv] += 1;
614 s->block_offset[uv] += 8;
619 if (p->key_frame || golden_frame) {
620 if (s->framep[VP56_FRAME_GOLDEN]->data[0] &&
621 s->framep[VP56_FRAME_GOLDEN] != s->framep[VP56_FRAME_GOLDEN2])
622 avctx->release_buffer(avctx, s->framep[VP56_FRAME_GOLDEN]);
623 s->framep[VP56_FRAME_GOLDEN] = p;
627 FFSWAP(AVFrame *, s->framep[VP56_FRAME_GOLDEN],
628 s->framep[VP56_FRAME_GOLDEN2]);
630 remaining_buf_size -= alpha_offset;
634 if (s->framep[VP56_FRAME_PREVIOUS] == s->framep[VP56_FRAME_GOLDEN] ||
635 s->framep[VP56_FRAME_PREVIOUS] == s->framep[VP56_FRAME_GOLDEN2]) {
636 if (s->framep[VP56_FRAME_UNUSED] != s->framep[VP56_FRAME_GOLDEN] &&
637 s->framep[VP56_FRAME_UNUSED] != s->framep[VP56_FRAME_GOLDEN2])
638 FFSWAP(AVFrame *, s->framep[VP56_FRAME_PREVIOUS],
639 s->framep[VP56_FRAME_UNUSED]);
641 FFSWAP(AVFrame *, s->framep[VP56_FRAME_PREVIOUS],
642 s->framep[VP56_FRAME_UNUSED2]);
643 } else if (s->framep[VP56_FRAME_PREVIOUS]->data[0])
644 avctx->release_buffer(avctx, s->framep[VP56_FRAME_PREVIOUS]);
645 FFSWAP(AVFrame *, s->framep[VP56_FRAME_CURRENT],
646 s->framep[VP56_FRAME_PREVIOUS]);
649 p->qscale_table = s->qscale_table;
650 p->qscale_type = FF_QSCALE_TYPE_VP56;
651 *(AVFrame*)data = *p;
652 *data_size = sizeof(AVFrame);
657 av_cold void vp56_init(AVCodecContext *avctx, int flip, int has_alpha)
659 VP56Context *s = avctx->priv_data;
663 avctx->pix_fmt = has_alpha ? PIX_FMT_YUVA420P : PIX_FMT_YUV420P;
665 if (avctx->idct_algo == FF_IDCT_AUTO)
666 avctx->idct_algo = FF_IDCT_VP3;
667 dsputil_init(&s->dsp, avctx);
668 ff_init_scantable(s->dsp.idct_permutation, &s->scantable,ff_zigzag_direct);
671 s->framep[i] = &s->frames[i];
672 s->framep[VP56_FRAME_UNUSED] = s->framep[VP56_FRAME_GOLDEN];
673 s->framep[VP56_FRAME_UNUSED2] = s->framep[VP56_FRAME_GOLDEN2];
674 s->edge_emu_buffer_alloc = NULL;
676 s->above_blocks = NULL;
677 s->macroblocks = NULL;
679 s->deblock_filtering = 1;
683 s->has_alpha = has_alpha;
695 av_cold int vp56_free(AVCodecContext *avctx)
697 VP56Context *s = avctx->priv_data;
699 av_freep(&s->qscale_table);
700 av_freep(&s->above_blocks);
701 av_freep(&s->macroblocks);
702 av_freep(&s->edge_emu_buffer_alloc);
703 if (s->framep[VP56_FRAME_GOLDEN]->data[0])
704 avctx->release_buffer(avctx, s->framep[VP56_FRAME_GOLDEN]);
705 if (s->framep[VP56_FRAME_GOLDEN2]->data[0])
706 avctx->release_buffer(avctx, s->framep[VP56_FRAME_GOLDEN2]);
707 if (s->framep[VP56_FRAME_PREVIOUS]->data[0])
708 avctx->release_buffer(avctx, s->framep[VP56_FRAME_PREVIOUS]);