2 * Copyright (C) 2006 Aurelien Jacobs <aurel@gnuage.org>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * VP5 and VP6 compatible video decoder (common features)
27 #include "bytestream.h"
33 void ff_vp56_init_dequant(VP56Context *s, int quantizer)
35 s->quantizer = quantizer;
36 s->dequant_dc = vp56_dc_dequant[quantizer] << 2;
37 s->dequant_ac = vp56_ac_dequant[quantizer] << 2;
38 memset(s->qscale_table, quantizer, s->mb_width);
41 static int vp56_get_vectors_predictors(VP56Context *s, int row, int col,
45 VP56mv vect[2] = {{0,0}, {0,0}};
49 for (pos=0; pos<12; pos++) {
50 mvp.x = col + vp56_candidate_predictor_pos[pos][0];
51 mvp.y = row + vp56_candidate_predictor_pos[pos][1];
52 if (mvp.x < 0 || mvp.x >= s->mb_width ||
53 mvp.y < 0 || mvp.y >= s->mb_height)
55 offset = mvp.x + s->mb_width*mvp.y;
57 if (vp56_reference_frame[s->macroblocks[offset].type] != ref_frame)
59 if ((s->macroblocks[offset].mv.x == vect[0].x &&
60 s->macroblocks[offset].mv.y == vect[0].y) ||
61 (s->macroblocks[offset].mv.x == 0 &&
62 s->macroblocks[offset].mv.y == 0))
65 vect[nb_pred++] = s->macroblocks[offset].mv;
70 s->vector_candidate_pos = pos;
73 s->vector_candidate[0] = vect[0];
74 s->vector_candidate[1] = vect[1];
79 static void vp56_parse_mb_type_models(VP56Context *s)
81 VP56RangeCoder *c = &s->c;
82 VP56Model *model = s->modelp;
85 for (ctx=0; ctx<3; ctx++) {
86 if (vp56_rac_get_prob(c, 174)) {
87 int idx = vp56_rac_gets(c, 4);
88 memcpy(model->mb_types_stats[ctx],
89 vp56_pre_def_mb_type_stats[idx][ctx],
90 sizeof(model->mb_types_stats[ctx]));
92 if (vp56_rac_get_prob(c, 254)) {
93 for (type=0; type<10; type++) {
95 if (vp56_rac_get_prob(c, 205)) {
96 int delta, sign = vp56_rac_get(c);
98 delta = vp56_rac_get_tree(c, vp56_pmbtm_tree,
99 vp56_mb_type_model_model);
101 delta = 4 * vp56_rac_gets(c, 7);
102 model->mb_types_stats[ctx][type][i] += (delta ^ -sign) + sign;
109 /* compute MB type probability tables based on previous MB type */
110 for (ctx=0; ctx<3; ctx++) {
113 for (type=0; type<10; type++)
114 p[type] = 100 * model->mb_types_stats[ctx][type][1];
116 for (type=0; type<10; type++) {
117 int p02, p34, p0234, p17, p56, p89, p5689, p156789;
119 /* conservative MB type probability */
120 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]);
122 p[type] = 0; /* same MB type => weight is null */
124 /* binary tree parsing probabilities */
132 p156789 = p17 + p5689;
134 model->mb_type[ctx][type][1] = 1 + 255 * p0234/(1+p0234+p156789);
135 model->mb_type[ctx][type][2] = 1 + 255 * p02 / (1+p0234);
136 model->mb_type[ctx][type][3] = 1 + 255 * p17 / (1+p156789);
137 model->mb_type[ctx][type][4] = 1 + 255 * p[0] / (1+p02);
138 model->mb_type[ctx][type][5] = 1 + 255 * p[3] / (1+p34);
139 model->mb_type[ctx][type][6] = 1 + 255 * p[1] / (1+p17);
140 model->mb_type[ctx][type][7] = 1 + 255 * p56 / (1+p5689);
141 model->mb_type[ctx][type][8] = 1 + 255 * p[5] / (1+p56);
142 model->mb_type[ctx][type][9] = 1 + 255 * p[8] / (1+p89);
144 /* restore initial value */
145 p[type] = 100 * model->mb_types_stats[ctx][type][1];
150 static VP56mb vp56_parse_mb_type(VP56Context *s,
151 VP56mb prev_type, int ctx)
153 uint8_t *mb_type_model = s->modelp->mb_type[ctx][prev_type];
154 VP56RangeCoder *c = &s->c;
156 if (vp56_rac_get_prob(c, mb_type_model[0]))
159 return vp56_rac_get_tree(c, vp56_pmbt_tree, mb_type_model);
162 static void vp56_decode_4mv(VP56Context *s, int row, int col)
168 /* parse each block type */
169 for (b=0; b<4; b++) {
170 type[b] = vp56_rac_gets(&s->c, 2);
172 type[b]++; /* only returns 0, 2, 3 or 4 (all INTER_PF) */
176 for (b=0; b<4; b++) {
178 case VP56_MB_INTER_NOVEC_PF:
179 s->mv[b] = (VP56mv) {0,0};
181 case VP56_MB_INTER_DELTA_PF:
182 s->parse_vector_adjustment(s, &s->mv[b]);
184 case VP56_MB_INTER_V1_PF:
185 s->mv[b] = s->vector_candidate[0];
187 case VP56_MB_INTER_V2_PF:
188 s->mv[b] = s->vector_candidate[1];
195 /* this is the one selected for the whole MB for prediction */
196 s->macroblocks[row * s->mb_width + col].mv = s->mv[3];
198 /* chroma vectors are average luma vectors */
199 if (s->avctx->codec->id == AV_CODEC_ID_VP5) {
200 s->mv[4].x = s->mv[5].x = RSHIFT(mv.x,2);
201 s->mv[4].y = s->mv[5].y = RSHIFT(mv.y,2);
203 s->mv[4] = s->mv[5] = (VP56mv) {mv.x/4, mv.y/4};
207 static VP56mb vp56_decode_mv(VP56Context *s, int row, int col)
209 VP56mv *mv, vect = {0,0};
212 ctx = vp56_get_vectors_predictors(s, row, col, VP56_FRAME_PREVIOUS);
213 s->mb_type = vp56_parse_mb_type(s, s->mb_type, ctx);
214 s->macroblocks[row * s->mb_width + col].type = s->mb_type;
216 switch (s->mb_type) {
217 case VP56_MB_INTER_V1_PF:
218 mv = &s->vector_candidate[0];
221 case VP56_MB_INTER_V2_PF:
222 mv = &s->vector_candidate[1];
225 case VP56_MB_INTER_V1_GF:
226 vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN);
227 mv = &s->vector_candidate[0];
230 case VP56_MB_INTER_V2_GF:
231 vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN);
232 mv = &s->vector_candidate[1];
235 case VP56_MB_INTER_DELTA_PF:
236 s->parse_vector_adjustment(s, &vect);
240 case VP56_MB_INTER_DELTA_GF:
241 vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN);
242 s->parse_vector_adjustment(s, &vect);
246 case VP56_MB_INTER_4V:
247 vp56_decode_4mv(s, row, col);
255 s->macroblocks[row*s->mb_width + col].mv = *mv;
257 /* same vector for all blocks */
264 static void vp56_add_predictors_dc(VP56Context *s, VP56Frame ref_frame)
266 int idx = s->scantable.permutated[0];
269 for (b=0; b<6; b++) {
270 VP56RefDc *ab = &s->above_blocks[s->above_block_idx[b]];
271 VP56RefDc *lb = &s->left_block[ff_vp56_b6to4[b]];
276 if (ref_frame == lb->ref_frame) {
280 if (ref_frame == ab->ref_frame) {
284 if (s->avctx->codec->id == AV_CODEC_ID_VP5)
286 if (count < 2 && ref_frame == ab[-1+2*i].ref_frame) {
287 dc += ab[-1+2*i].dc_coeff;
291 dc = s->prev_dc[ff_vp56_b2p[b]][ref_frame];
295 s->block_coeff[b][idx] += dc;
296 s->prev_dc[ff_vp56_b2p[b]][ref_frame] = s->block_coeff[b][idx];
297 ab->dc_coeff = s->block_coeff[b][idx];
298 ab->ref_frame = ref_frame;
299 lb->dc_coeff = s->block_coeff[b][idx];
300 lb->ref_frame = ref_frame;
301 s->block_coeff[b][idx] *= s->dequant_dc;
305 static void vp56_deblock_filter(VP56Context *s, uint8_t *yuv,
306 int stride, int dx, int dy)
308 int t = vp56_filter_threshold[s->quantizer];
309 if (dx) s->vp56dsp.edge_filter_hor(yuv + 10-dx , stride, t);
310 if (dy) s->vp56dsp.edge_filter_ver(yuv + stride*(10-dy), stride, t);
313 static void vp56_mc(VP56Context *s, int b, int plane, uint8_t *src,
314 int stride, int x, int y)
316 uint8_t *dst=s->framep[VP56_FRAME_CURRENT]->data[plane]+s->block_offset[b];
319 int overlap_offset = 0;
320 int mask = s->vp56_coord_div[b] - 1;
321 int deblock_filtering = s->deblock_filtering;
325 if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
326 (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY
327 && !s->framep[VP56_FRAME_CURRENT]->key_frame))
328 deblock_filtering = 0;
330 dx = s->mv[b].x / s->vp56_coord_div[b];
331 dy = s->mv[b].y / s->vp56_coord_div[b];
340 if (x<0 || x+12>=s->plane_width[plane] ||
341 y<0 || y+12>=s->plane_height[plane]) {
342 s->dsp.emulated_edge_mc(s->edge_emu_buffer,
343 src + s->block_offset[b] + (dy-2)*stride + (dx-2),
344 stride, 12, 12, x, y,
345 s->plane_width[plane],
346 s->plane_height[plane]);
347 src_block = s->edge_emu_buffer;
348 src_offset = 2 + 2*stride;
349 } else if (deblock_filtering) {
350 /* only need a 12x12 block, but there is no such dsp function, */
351 /* so copy a 16x12 block */
352 s->dsp.put_pixels_tab[0][0](s->edge_emu_buffer,
353 src + s->block_offset[b] + (dy-2)*stride + (dx-2),
355 src_block = s->edge_emu_buffer;
356 src_offset = 2 + 2*stride;
359 src_offset = s->block_offset[b] + dy*stride + dx;
362 if (deblock_filtering)
363 vp56_deblock_filter(s, src_block, stride, dx&7, dy&7);
365 if (s->mv[b].x & mask)
366 overlap_offset += (s->mv[b].x > 0) ? 1 : -1;
367 if (s->mv[b].y & mask)
368 overlap_offset += (s->mv[b].y > 0) ? stride : -stride;
370 if (overlap_offset) {
372 s->filter(s, dst, src_block, src_offset, src_offset+overlap_offset,
373 stride, s->mv[b], mask, s->filter_selection, b<4);
375 s->dsp.put_no_rnd_pixels_l2[1](dst, src_block+src_offset,
376 src_block+src_offset+overlap_offset,
379 s->dsp.put_pixels_tab[1][0](dst, src_block+src_offset, stride, 8);
383 static void vp56_decode_mb(VP56Context *s, int row, int col, int is_alpha)
385 AVFrame *frame_current, *frame_ref;
388 int b, ab, b_max, plane, off;
390 if (s->framep[VP56_FRAME_CURRENT]->key_frame)
391 mb_type = VP56_MB_INTRA;
393 mb_type = vp56_decode_mv(s, row, col);
394 ref_frame = vp56_reference_frame[mb_type];
396 s->dsp.clear_blocks(*s->block_coeff);
400 vp56_add_predictors_dc(s, ref_frame);
402 frame_current = s->framep[VP56_FRAME_CURRENT];
403 frame_ref = s->framep[ref_frame];
404 if (mb_type != VP56_MB_INTRA && !frame_ref->data[0])
408 b_max = 6 - 2*is_alpha;
412 for (b=0; b<b_max; b++) {
413 plane = ff_vp56_b2p[b+ab];
414 s->vp3dsp.idct_put(frame_current->data[plane] + s->block_offset[b],
415 s->stride[plane], s->block_coeff[b]);
419 case VP56_MB_INTER_NOVEC_PF:
420 case VP56_MB_INTER_NOVEC_GF:
421 for (b=0; b<b_max; b++) {
422 plane = ff_vp56_b2p[b+ab];
423 off = s->block_offset[b];
424 s->dsp.put_pixels_tab[1][0](frame_current->data[plane] + off,
425 frame_ref->data[plane] + off,
426 s->stride[plane], 8);
427 s->vp3dsp.idct_add(frame_current->data[plane] + off,
428 s->stride[plane], s->block_coeff[b]);
432 case VP56_MB_INTER_DELTA_PF:
433 case VP56_MB_INTER_V1_PF:
434 case VP56_MB_INTER_V2_PF:
435 case VP56_MB_INTER_DELTA_GF:
436 case VP56_MB_INTER_4V:
437 case VP56_MB_INTER_V1_GF:
438 case VP56_MB_INTER_V2_GF:
439 for (b=0; b<b_max; b++) {
440 int x_off = b==1 || b==3 ? 8 : 0;
441 int y_off = b==2 || b==3 ? 8 : 0;
442 plane = ff_vp56_b2p[b+ab];
443 vp56_mc(s, b, plane, frame_ref->data[plane], s->stride[plane],
444 16*col+x_off, 16*row+y_off);
445 s->vp3dsp.idct_add(frame_current->data[plane] + s->block_offset[b],
446 s->stride[plane], s->block_coeff[b]);
452 static int vp56_size_changed(VP56Context *s)
454 AVCodecContext *avctx = s->avctx;
455 int stride = s->framep[VP56_FRAME_CURRENT]->linesize[0];
458 s->plane_width[0] = s->plane_width[3] = avctx->coded_width;
459 s->plane_width[1] = s->plane_width[2] = avctx->coded_width/2;
460 s->plane_height[0] = s->plane_height[3] = avctx->coded_height;
461 s->plane_height[1] = s->plane_height[2] = avctx->coded_height/2;
464 s->stride[i] = s->flip * s->framep[VP56_FRAME_CURRENT]->linesize[i];
466 s->mb_width = (avctx->coded_width +15) / 16;
467 s->mb_height = (avctx->coded_height+15) / 16;
469 if (s->mb_width > 1000 || s->mb_height > 1000) {
470 avcodec_set_dimensions(avctx, 0, 0);
471 av_log(avctx, AV_LOG_ERROR, "picture too big\n");
475 s->qscale_table = av_realloc(s->qscale_table, s->mb_width);
476 s->above_blocks = av_realloc(s->above_blocks,
477 (4*s->mb_width+6) * sizeof(*s->above_blocks));
478 s->macroblocks = av_realloc(s->macroblocks,
479 s->mb_width*s->mb_height*sizeof(*s->macroblocks));
480 av_free(s->edge_emu_buffer_alloc);
481 s->edge_emu_buffer_alloc = av_malloc(16*stride);
482 s->edge_emu_buffer = s->edge_emu_buffer_alloc;
484 s->edge_emu_buffer += 15 * stride;
486 if (s->alpha_context)
487 return vp56_size_changed(s->alpha_context);
492 static int ff_vp56_decode_mbs(AVCodecContext *avctx, void *, int, int);
494 int ff_vp56_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
497 const uint8_t *buf = avpkt->data;
498 VP56Context *s = avctx->priv_data;
500 int remaining_buf_size = avpkt->size;
501 int av_uninit(alpha_offset);
504 /* select a current frame from the unused frames */
505 for (i = 0; i < 4; ++i) {
506 if (!s->frames[i].data[0]) {
512 s->framep[VP56_FRAME_CURRENT] = p;
513 if (s->alpha_context)
514 s->alpha_context->framep[VP56_FRAME_CURRENT] = p;
517 if (remaining_buf_size < 3)
519 alpha_offset = bytestream_get_be24(&buf);
520 remaining_buf_size -= 3;
521 if (remaining_buf_size < alpha_offset)
525 res = s->parse_header(s, buf, remaining_buf_size);
530 for (i = 0; i < 4; i++) {
531 if (s->frames[i].data[0])
532 avctx->release_buffer(avctx, &s->frames[i]);
537 if (avctx->get_buffer(avctx, p) < 0) {
538 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
543 if (vp56_size_changed(s)) {
544 avctx->release_buffer(avctx, p);
551 remaining_buf_size -= alpha_offset;
553 res = s->alpha_context->parse_header(s->alpha_context, buf, remaining_buf_size);
555 avctx->release_buffer(avctx, p);
560 avctx->execute2(avctx, ff_vp56_decode_mbs, 0, 0, s->has_alpha + 1);
562 /* release frames that aren't in use */
563 for (i = 0; i < 4; ++i) {
564 AVFrame *victim = &s->frames[i];
565 if (!victim->data[0])
567 if (victim != s->framep[VP56_FRAME_PREVIOUS] &&
568 victim != s->framep[VP56_FRAME_GOLDEN] &&
569 (!s->has_alpha || victim != s->alpha_context->framep[VP56_FRAME_GOLDEN]))
570 avctx->release_buffer(avctx, victim);
574 p->qscale_table = s->qscale_table;
575 p->qscale_type = FF_QSCALE_TYPE_VP56;
576 *(AVFrame*)data = *p;
577 *data_size = sizeof(AVFrame);
582 static int ff_vp56_decode_mbs(AVCodecContext *avctx, void *data,
583 int jobnr, int threadnr)
585 VP56Context *s0 = avctx->priv_data;
586 int is_alpha = (jobnr == 1);
587 VP56Context *s = is_alpha ? s0->alpha_context : s0;
588 AVFrame *const p = s->framep[VP56_FRAME_CURRENT];
589 int mb_row, mb_col, mb_row_flip, mb_offset = 0;
590 int block, y, uv, stride_y, stride_uv;
593 p->pict_type = AV_PICTURE_TYPE_I;
594 s->default_models_init(s);
595 for (block=0; block<s->mb_height*s->mb_width; block++)
596 s->macroblocks[block].type = VP56_MB_INTRA;
598 p->pict_type = AV_PICTURE_TYPE_P;
599 vp56_parse_mb_type_models(s);
600 s->parse_vector_models(s);
601 s->mb_type = VP56_MB_INTER_NOVEC_PF;
604 if (s->parse_coeff_models(s))
607 memset(s->prev_dc, 0, sizeof(s->prev_dc));
608 s->prev_dc[1][VP56_FRAME_CURRENT] = 128;
609 s->prev_dc[2][VP56_FRAME_CURRENT] = 128;
611 for (block=0; block < 4*s->mb_width+6; block++) {
612 s->above_blocks[block].ref_frame = VP56_FRAME_NONE;
613 s->above_blocks[block].dc_coeff = 0;
614 s->above_blocks[block].not_null_dc = 0;
616 s->above_blocks[2*s->mb_width + 2].ref_frame = VP56_FRAME_CURRENT;
617 s->above_blocks[3*s->mb_width + 4].ref_frame = VP56_FRAME_CURRENT;
619 stride_y = p->linesize[0];
620 stride_uv = p->linesize[1];
625 /* main macroblocks loop */
626 for (mb_row=0; mb_row<s->mb_height; mb_row++) {
628 mb_row_flip = s->mb_height - mb_row - 1;
630 mb_row_flip = mb_row;
632 for (block=0; block<4; block++) {
633 s->left_block[block].ref_frame = VP56_FRAME_NONE;
634 s->left_block[block].dc_coeff = 0;
635 s->left_block[block].not_null_dc = 0;
637 memset(s->coeff_ctx, 0, sizeof(s->coeff_ctx));
638 memset(s->coeff_ctx_last, 24, sizeof(s->coeff_ctx_last));
640 s->above_block_idx[0] = 1;
641 s->above_block_idx[1] = 2;
642 s->above_block_idx[2] = 1;
643 s->above_block_idx[3] = 2;
644 s->above_block_idx[4] = 2*s->mb_width + 2 + 1;
645 s->above_block_idx[5] = 3*s->mb_width + 4 + 1;
647 s->block_offset[s->frbi] = (mb_row_flip*16 + mb_offset) * stride_y;
648 s->block_offset[s->srbi] = s->block_offset[s->frbi] + 8*stride_y;
649 s->block_offset[1] = s->block_offset[0] + 8;
650 s->block_offset[3] = s->block_offset[2] + 8;
651 s->block_offset[4] = (mb_row_flip*8 + mb_offset) * stride_uv;
652 s->block_offset[5] = s->block_offset[4];
654 for (mb_col=0; mb_col<s->mb_width; mb_col++) {
655 vp56_decode_mb(s, mb_row, mb_col, is_alpha);
657 for (y=0; y<4; y++) {
658 s->above_block_idx[y] += 2;
659 s->block_offset[y] += 16;
662 for (uv=4; uv<6; uv++) {
663 s->above_block_idx[uv] += 1;
664 s->block_offset[uv] += 8;
670 if (p->key_frame || s->golden_frame) {
671 s->framep[VP56_FRAME_GOLDEN] = p;
674 FFSWAP(AVFrame *, s->framep[VP56_FRAME_CURRENT],
675 s->framep[VP56_FRAME_PREVIOUS]);
679 av_cold void ff_vp56_init(AVCodecContext *avctx, int flip, int has_alpha)
681 VP56Context *s = avctx->priv_data;
682 ff_vp56_init_context(avctx, s, flip, has_alpha);
685 av_cold void ff_vp56_init_context(AVCodecContext *avctx, VP56Context *s,
686 int flip, int has_alpha)
691 avctx->pix_fmt = has_alpha ? PIX_FMT_YUVA420P : PIX_FMT_YUV420P;
693 ff_dsputil_init(&s->dsp, avctx);
694 ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
695 ff_vp56dsp_init(&s->vp56dsp, avctx->codec->id);
696 ff_init_scantable_permutation(s->dsp.idct_permutation, s->vp3dsp.idct_perm);
697 ff_init_scantable(s->dsp.idct_permutation, &s->scantable,ff_zigzag_direct);
699 for (i=0; i<4; i++) {
700 s->framep[i] = &s->frames[i];
701 avcodec_get_frame_defaults(&s->frames[i]);
703 s->framep[VP56_FRAME_UNUSED] = s->framep[VP56_FRAME_GOLDEN];
704 s->framep[VP56_FRAME_UNUSED2] = s->framep[VP56_FRAME_GOLDEN2];
705 s->edge_emu_buffer_alloc = NULL;
707 s->above_blocks = NULL;
708 s->macroblocks = NULL;
710 s->deblock_filtering = 1;
715 s->has_alpha = has_alpha;
717 s->modelp = &s->model;
730 av_cold int ff_vp56_free(AVCodecContext *avctx)
732 VP56Context *s = avctx->priv_data;
733 return ff_vp56_free_context(s);
736 av_cold int ff_vp56_free_context(VP56Context *s)
738 AVCodecContext *avctx = s->avctx;
741 av_freep(&s->qscale_table);
742 av_freep(&s->above_blocks);
743 av_freep(&s->macroblocks);
744 av_freep(&s->edge_emu_buffer_alloc);
745 for (i = 0; i < 4; ++i) {
746 if (s->frames[i].data[0])
747 avctx->release_buffer(avctx, &s->frames[i]);