Mercurial > libavcodec.hg
view vp56.c @ 5723:49a5d44423ef libavcodec
h264/PAFF preparation: use DELAYED_PIC_REF to mark non-refs frames as held for delayed output
patch by Jeff Downs, heydowns a borg d com
original thread:
Subject: [FFmpeg-devel] [PATCH] Implement PAFF in H.264
Date: 18/09/07 20:30
author | andoma |
---|---|
date | Wed, 26 Sep 2007 13:28:13 +0000 |
parents | 1643aa6ac147 |
children | 1b91ce6c758c |
line wrap: on
line source
/** * @file vp56.c * VP5 and VP6 compatible video decoder (common features) * * Copyright (C) 2006 Aurelien Jacobs <aurel@gnuage.org> * * This file is part of FFmpeg. * * FFmpeg is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2.1 of the License, or (at your option) any later version. * * FFmpeg is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ #include "avcodec.h" #include "bytestream.h" #include "vp56.h" #include "vp56data.h" void vp56_init_dequant(vp56_context_t *s, int quantizer) { s->quantizer = quantizer; s->dequant_dc = vp56_dc_dequant[quantizer] << 2; s->dequant_ac = vp56_ac_dequant[quantizer] << 2; } static int vp56_get_vectors_predictors(vp56_context_t *s, int row, int col, vp56_frame_t ref_frame) { int nb_pred = 0; vp56_mv_t vect[2] = {{0,0}, {0,0}}; int pos, offset; vp56_mv_t mvp; for (pos=0; pos<12; pos++) { mvp.x = col + vp56_candidate_predictor_pos[pos][0]; mvp.y = row + vp56_candidate_predictor_pos[pos][1]; if (mvp.x < 0 || mvp.x >= s->mb_width || mvp.y < 0 || mvp.y >= s->mb_height) continue; offset = mvp.x + s->mb_width*mvp.y; if (vp56_reference_frame[s->macroblocks[offset].type] != ref_frame) continue; if ((s->macroblocks[offset].mv.x == vect[0].x && s->macroblocks[offset].mv.y == vect[0].y) || (s->macroblocks[offset].mv.x == 0 && s->macroblocks[offset].mv.y == 0)) continue; vect[nb_pred++] = s->macroblocks[offset].mv; if (nb_pred > 1) { nb_pred = -1; break; } s->vector_candidate_pos = pos; } s->vector_candidate[0] = vect[0]; s->vector_candidate[1] = vect[1]; return nb_pred+1; } static void vp56_parse_mb_type_models(vp56_context_t *s) { vp56_range_coder_t *c = &s->c; vp56_model_t *model = s->modelp; int i, ctx, type; for (ctx=0; ctx<3; ctx++) { if (vp56_rac_get_prob(c, 174)) { int idx = vp56_rac_gets(c, 4); memcpy(model->mb_types_stats[ctx], vp56_pre_def_mb_type_stats[idx][ctx], sizeof(model->mb_types_stats[ctx])); } if (vp56_rac_get_prob(c, 254)) { for (type=0; type<10; type++) { for(i=0; i<2; i++) { if (vp56_rac_get_prob(c, 205)) { int delta, sign = vp56_rac_get(c); delta = vp56_rac_get_tree(c, vp56_pmbtm_tree, vp56_mb_type_model_model); if (!delta) delta = 4 * vp56_rac_gets(c, 7); model->mb_types_stats[ctx][type][i] += (delta ^ -sign) + sign; } } } } } /* compute MB type probability tables based on previous MB type */ for (ctx=0; ctx<3; ctx++) { int p[10]; for (type=0; type<10; type++) p[type] = 100 * model->mb_types_stats[ctx][type][1]; for (type=0; type<10; type++) { int p02, p34, p0234, p17, p56, p89, p5689, p156789; /* conservative MB type probability */ 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]); p[type] = 0; /* same MB type => weight is null */ /* binary tree parsing probabilities */ p02 = p[0] + p[2]; p34 = p[3] + p[4]; p0234 = p02 + p34; p17 = p[1] + p[7]; p56 = p[5] + p[6]; p89 = p[8] + p[9]; p5689 = p56 + p89; p156789 = p17 + p5689; model->mb_type[ctx][type][1] = 1 + 255 * p0234/(1+p0234+p156789); model->mb_type[ctx][type][2] = 1 + 255 * p02 / (1+p0234); model->mb_type[ctx][type][3] = 1 + 255 * p17 / (1+p156789); model->mb_type[ctx][type][4] = 1 + 255 * p[0] / (1+p02); model->mb_type[ctx][type][5] = 1 + 255 * p[3] / (1+p34); model->mb_type[ctx][type][6] = 1 + 255 * p[1] / (1+p17); model->mb_type[ctx][type][7] = 1 + 255 * p56 / (1+p5689); model->mb_type[ctx][type][8] = 1 + 255 * p[5] / (1+p56); model->mb_type[ctx][type][9] = 1 + 255 * p[8] / (1+p89); /* restore initial value */ p[type] = 100 * model->mb_types_stats[ctx][type][1]; } } } static vp56_mb_t vp56_parse_mb_type(vp56_context_t *s, vp56_mb_t prev_type, int ctx) { uint8_t *mb_type_model = s->modelp->mb_type[ctx][prev_type]; vp56_range_coder_t *c = &s->c; if (vp56_rac_get_prob(c, mb_type_model[0])) return prev_type; else return vp56_rac_get_tree(c, vp56_pmbt_tree, mb_type_model); } static void vp56_decode_4mv(vp56_context_t *s, int row, int col) { vp56_mv_t mv = {0,0}; int type[4]; int b; /* parse each block type */ for (b=0; b<4; b++) { type[b] = vp56_rac_gets(&s->c, 2); if (type[b]) type[b]++; /* only returns 0, 2, 3 or 4 (all INTER_PF) */ } /* get vectors */ for (b=0; b<4; b++) { switch (type[b]) { case VP56_MB_INTER_NOVEC_PF: s->mv[b] = (vp56_mv_t) {0,0}; break; case VP56_MB_INTER_DELTA_PF: s->parse_vector_adjustment(s, &s->mv[b]); break; case VP56_MB_INTER_V1_PF: s->mv[b] = s->vector_candidate[0]; break; case VP56_MB_INTER_V2_PF: s->mv[b] = s->vector_candidate[1]; break; } mv.x += s->mv[b].x; mv.y += s->mv[b].y; } /* this is the one selected for the whole MB for prediction */ s->macroblocks[row * s->mb_width + col].mv = s->mv[3]; /* chroma vectors are average luma vectors */ if (s->avctx->codec->id == CODEC_ID_VP5) { s->mv[4].x = s->mv[5].x = RSHIFT(mv.x,2); s->mv[4].y = s->mv[5].y = RSHIFT(mv.y,2); } else { s->mv[4] = s->mv[5] = (vp56_mv_t) {mv.x/4, mv.y/4}; } } static vp56_mb_t vp56_decode_mv(vp56_context_t *s, int row, int col) { vp56_mv_t *mv, vect = {0,0}; int ctx, b; ctx = vp56_get_vectors_predictors(s, row, col, VP56_FRAME_PREVIOUS); s->mb_type = vp56_parse_mb_type(s, s->mb_type, ctx); s->macroblocks[row * s->mb_width + col].type = s->mb_type; switch (s->mb_type) { case VP56_MB_INTER_V1_PF: mv = &s->vector_candidate[0]; break; case VP56_MB_INTER_V2_PF: mv = &s->vector_candidate[1]; break; case VP56_MB_INTER_V1_GF: vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN); mv = &s->vector_candidate[0]; break; case VP56_MB_INTER_V2_GF: vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN); mv = &s->vector_candidate[1]; break; case VP56_MB_INTER_DELTA_PF: s->parse_vector_adjustment(s, &vect); mv = &vect; break; case VP56_MB_INTER_DELTA_GF: vp56_get_vectors_predictors(s, row, col, VP56_FRAME_GOLDEN); s->parse_vector_adjustment(s, &vect); mv = &vect; break; case VP56_MB_INTER_4V: vp56_decode_4mv(s, row, col); return s->mb_type; default: mv = &vect; break; } s->macroblocks[row*s->mb_width + col].mv = *mv; /* same vector for all blocks */ for (b=0; b<6; b++) s->mv[b] = *mv; return s->mb_type; } static void vp56_add_predictors_dc(vp56_context_t *s, vp56_frame_t ref_frame) { int idx = s->scantable.permutated[0]; int b; for (b=0; b<6; b++) { vp56_ref_dc_t *ab = &s->above_blocks[s->above_block_idx[b]]; vp56_ref_dc_t *lb = &s->left_block[vp56_b6to4[b]]; int count = 0; int dc = 0; int i; if (ref_frame == lb->ref_frame) { dc += lb->dc_coeff; count++; } if (ref_frame == ab->ref_frame) { dc += ab->dc_coeff; count++; } if (s->avctx->codec->id == CODEC_ID_VP5) for (i=0; i<2; i++) if (count < 2 && ref_frame == ab[-1+2*i].ref_frame) { dc += ab[-1+2*i].dc_coeff; count++; } if (count == 0) dc = s->prev_dc[vp56_b2p[b]][ref_frame]; else if (count == 2) dc /= 2; s->block_coeff[b][idx] += dc; s->prev_dc[vp56_b2p[b]][ref_frame] = s->block_coeff[b][idx]; ab->dc_coeff = s->block_coeff[b][idx]; ab->ref_frame = ref_frame; lb->dc_coeff = s->block_coeff[b][idx]; lb->ref_frame = ref_frame; s->block_coeff[b][idx] *= s->dequant_dc; } } static void vp56_edge_filter(vp56_context_t *s, uint8_t *yuv, int pix_inc, int line_inc, int t) { int pix2_inc = 2 * pix_inc; int i, v; for (i=0; i<12; i++) { v = (yuv[-pix2_inc] + 3*(yuv[0]-yuv[-pix_inc]) - yuv[pix_inc] + 4) >>3; v = s->adjust(v, t); yuv[-pix_inc] = av_clip_uint8(yuv[-pix_inc] + v); yuv[0] = av_clip_uint8(yuv[0] - v); yuv += line_inc; } } static void vp56_deblock_filter(vp56_context_t *s, uint8_t *yuv, int stride, int dx, int dy) { int t = vp56_filter_threshold[s->quantizer]; if (dx) vp56_edge_filter(s, yuv + 10-dx , 1, stride, t); if (dy) vp56_edge_filter(s, yuv + stride*(10-dy), stride, 1, t); } static void vp56_mc(vp56_context_t *s, int b, int plane, uint8_t *src, int stride, int x, int y) { uint8_t *dst=s->framep[VP56_FRAME_CURRENT]->data[plane]+s->block_offset[b]; uint8_t *src_block; int src_offset; int overlap_offset = 0; int mask = s->vp56_coord_div[b] - 1; int deblock_filtering = s->deblock_filtering; int dx; int dy; if (s->avctx->skip_loop_filter >= AVDISCARD_ALL || (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && !s->framep[VP56_FRAME_CURRENT]->key_frame)) deblock_filtering = 0; dx = s->mv[b].x / s->vp56_coord_div[b]; dy = s->mv[b].y / s->vp56_coord_div[b]; if (b >= 4) { x /= 2; y /= 2; } x += dx - 2; y += dy - 2; if (x<0 || x+12>=s->plane_width[plane] || y<0 || y+12>=s->plane_height[plane]) { ff_emulated_edge_mc(s->edge_emu_buffer, src + s->block_offset[b] + (dy-2)*stride + (dx-2), stride, 12, 12, x, y, s->plane_width[plane], s->plane_height[plane]); src_block = s->edge_emu_buffer; src_offset = 2 + 2*stride; } else if (deblock_filtering) { /* only need a 12x12 block, but there is no such dsp function, */ /* so copy a 16x12 block */ s->dsp.put_pixels_tab[0][0](s->edge_emu_buffer, src + s->block_offset[b] + (dy-2)*stride + (dx-2), stride, 12); src_block = s->edge_emu_buffer; src_offset = 2 + 2*stride; } else { src_block = src; src_offset = s->block_offset[b] + dy*stride + dx; } if (deblock_filtering) vp56_deblock_filter(s, src_block, stride, dx&7, dy&7); if (s->mv[b].x & mask) overlap_offset += (s->mv[b].x > 0) ? 1 : -1; if (s->mv[b].y & mask) overlap_offset += (s->mv[b].y > 0) ? stride : -stride; if (overlap_offset) { if (s->filter) s->filter(s, dst, src_block, src_offset, src_offset+overlap_offset, stride, s->mv[b], mask, s->filter_selection, b<4); else s->dsp.put_no_rnd_pixels_l2[1](dst, src_block+src_offset, src_block+src_offset+overlap_offset, stride, 8); } else { s->dsp.put_pixels_tab[1][0](dst, src_block+src_offset, stride, 8); } } static void vp56_decode_mb(vp56_context_t *s, int row, int col, int is_alpha) { AVFrame *frame_current, *frame_ref; vp56_mb_t mb_type; vp56_frame_t ref_frame; int b, ab, b_max, plane, off; if (s->framep[VP56_FRAME_CURRENT]->key_frame) mb_type = VP56_MB_INTRA; else mb_type = vp56_decode_mv(s, row, col); ref_frame = vp56_reference_frame[mb_type]; memset(s->block_coeff, 0, sizeof(s->block_coeff)); s->parse_coeff(s); vp56_add_predictors_dc(s, ref_frame); frame_current = s->framep[VP56_FRAME_CURRENT]; frame_ref = s->framep[ref_frame]; ab = 6*is_alpha; b_max = 6 - 2*is_alpha; switch (mb_type) { case VP56_MB_INTRA: for (b=0; b<b_max; b++) { plane = vp56_b2p[b+ab]; s->dsp.idct_put(frame_current->data[plane] + s->block_offset[b], s->stride[plane], s->block_coeff[b]); } break; case VP56_MB_INTER_NOVEC_PF: case VP56_MB_INTER_NOVEC_GF: for (b=0; b<b_max; b++) { plane = vp56_b2p[b+ab]; off = s->block_offset[b]; s->dsp.put_pixels_tab[1][0](frame_current->data[plane] + off, frame_ref->data[plane] + off, s->stride[plane], 8); s->dsp.idct_add(frame_current->data[plane] + off, s->stride[plane], s->block_coeff[b]); } break; case VP56_MB_INTER_DELTA_PF: case VP56_MB_INTER_V1_PF: case VP56_MB_INTER_V2_PF: case VP56_MB_INTER_DELTA_GF: case VP56_MB_INTER_4V: case VP56_MB_INTER_V1_GF: case VP56_MB_INTER_V2_GF: for (b=0; b<b_max; b++) { int x_off = b==1 || b==3 ? 8 : 0; int y_off = b==2 || b==3 ? 8 : 0; plane = vp56_b2p[b+ab]; vp56_mc(s, b, plane, frame_ref->data[plane], s->stride[plane], 16*col+x_off, 16*row+y_off); s->dsp.idct_add(frame_current->data[plane] + s->block_offset[b], s->stride[plane], s->block_coeff[b]); } break; } } static int vp56_size_changed(AVCodecContext *avctx) { vp56_context_t *s = avctx->priv_data; int stride = s->framep[VP56_FRAME_CURRENT]->linesize[0]; int i; s->plane_width[0] = s->plane_width[3] = avctx->coded_width; s->plane_width[1] = s->plane_width[2] = avctx->coded_width/2; s->plane_height[0] = s->plane_height[3] = avctx->coded_height; s->plane_height[1] = s->plane_height[2] = avctx->coded_height/2; for (i=0; i<4; i++) s->stride[i] = s->flip * s->framep[VP56_FRAME_CURRENT]->linesize[i]; s->mb_width = (avctx->coded_width +15) / 16; s->mb_height = (avctx->coded_height+15) / 16; if (s->mb_width > 1000 || s->mb_height > 1000) { av_log(avctx, AV_LOG_ERROR, "picture too big\n"); return -1; } s->above_blocks = av_realloc(s->above_blocks, (4*s->mb_width+6) * sizeof(*s->above_blocks)); s->macroblocks = av_realloc(s->macroblocks, s->mb_width*s->mb_height*sizeof(*s->macroblocks)); av_free(s->edge_emu_buffer_alloc); s->edge_emu_buffer_alloc = av_malloc(16*stride); s->edge_emu_buffer = s->edge_emu_buffer_alloc; if (s->flip < 0) s->edge_emu_buffer += 15 * stride; return 0; } int vp56_decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8_t *buf, int buf_size) { vp56_context_t *s = avctx->priv_data; AVFrame *const p = s->framep[VP56_FRAME_CURRENT]; int is_alpha, alpha_offset; if (s->has_alpha) { alpha_offset = bytestream_get_be24(&buf); buf_size -= 3; } for (is_alpha=0; is_alpha < 1+s->has_alpha; is_alpha++) { int mb_row, mb_col, mb_row_flip, mb_offset = 0; int block, y, uv, stride_y, stride_uv; int golden_frame = 0; int res; s->modelp = &s->models[is_alpha]; res = s->parse_header(s, buf, buf_size, &golden_frame); if (!res) return -1; if (!is_alpha) { p->reference = 1; if (avctx->get_buffer(avctx, p) < 0) { av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n"); return -1; } if (res == 2) if (vp56_size_changed(avctx)) { avctx->release_buffer(avctx, p); return -1; } } if (p->key_frame) { p->pict_type = FF_I_TYPE; s->default_models_init(s); for (block=0; block<s->mb_height*s->mb_width; block++) s->macroblocks[block].type = VP56_MB_INTRA; } else { p->pict_type = FF_P_TYPE; vp56_parse_mb_type_models(s); s->parse_vector_models(s); s->mb_type = VP56_MB_INTER_NOVEC_PF; } s->parse_coeff_models(s); memset(s->prev_dc, 0, sizeof(s->prev_dc)); s->prev_dc[1][VP56_FRAME_CURRENT] = 128; s->prev_dc[2][VP56_FRAME_CURRENT] = 128; for (block=0; block < 4*s->mb_width+6; block++) { s->above_blocks[block].ref_frame = -1; s->above_blocks[block].dc_coeff = 0; s->above_blocks[block].not_null_dc = 0; } s->above_blocks[2*s->mb_width + 2].ref_frame = 0; s->above_blocks[3*s->mb_width + 4].ref_frame = 0; stride_y = p->linesize[0]; stride_uv = p->linesize[1]; if (s->flip < 0) mb_offset = 7; /* main macroblocks loop */ for (mb_row=0; mb_row<s->mb_height; mb_row++) { if (s->flip < 0) mb_row_flip = s->mb_height - mb_row - 1; else mb_row_flip = mb_row; for (block=0; block<4; block++) { s->left_block[block].ref_frame = -1; s->left_block[block].dc_coeff = 0; s->left_block[block].not_null_dc = 0; } memset(s->coeff_ctx, 0, sizeof(s->coeff_ctx)); memset(s->coeff_ctx_last, 24, sizeof(s->coeff_ctx_last)); s->above_block_idx[0] = 1; s->above_block_idx[1] = 2; s->above_block_idx[2] = 1; s->above_block_idx[3] = 2; s->above_block_idx[4] = 2*s->mb_width + 2 + 1; s->above_block_idx[5] = 3*s->mb_width + 4 + 1; s->block_offset[s->frbi] = (mb_row_flip*16 + mb_offset) * stride_y; s->block_offset[s->srbi] = s->block_offset[s->frbi] + 8*stride_y; s->block_offset[1] = s->block_offset[0] + 8; s->block_offset[3] = s->block_offset[2] + 8; s->block_offset[4] = (mb_row_flip*8 + mb_offset) * stride_uv; s->block_offset[5] = s->block_offset[4]; for (mb_col=0; mb_col<s->mb_width; mb_col++) { vp56_decode_mb(s, mb_row, mb_col, is_alpha); for (y=0; y<4; y++) { s->above_block_idx[y] += 2; s->block_offset[y] += 16; } for (uv=4; uv<6; uv++) { s->above_block_idx[uv] += 1; s->block_offset[uv] += 8; } } } if (p->key_frame || golden_frame) { if (s->framep[VP56_FRAME_GOLDEN]->data[0] && s->framep[VP56_FRAME_GOLDEN] != s->framep[VP56_FRAME_GOLDEN2]) avctx->release_buffer(avctx, s->framep[VP56_FRAME_GOLDEN]); s->framep[VP56_FRAME_GOLDEN] = p; } if (s->has_alpha) { FFSWAP(AVFrame *, s->framep[VP56_FRAME_GOLDEN], s->framep[VP56_FRAME_GOLDEN2]); buf += alpha_offset; buf_size -= alpha_offset; } } if (s->framep[VP56_FRAME_PREVIOUS] == s->framep[VP56_FRAME_GOLDEN] || s->framep[VP56_FRAME_PREVIOUS] == s->framep[VP56_FRAME_GOLDEN2]) { if (s->framep[VP56_FRAME_UNUSED] != s->framep[VP56_FRAME_GOLDEN] && s->framep[VP56_FRAME_UNUSED] != s->framep[VP56_FRAME_GOLDEN2]) FFSWAP(AVFrame *, s->framep[VP56_FRAME_PREVIOUS], s->framep[VP56_FRAME_UNUSED]); else FFSWAP(AVFrame *, s->framep[VP56_FRAME_PREVIOUS], s->framep[VP56_FRAME_UNUSED2]); } else if (s->framep[VP56_FRAME_PREVIOUS]->data[0]) avctx->release_buffer(avctx, s->framep[VP56_FRAME_PREVIOUS]); FFSWAP(AVFrame *, s->framep[VP56_FRAME_CURRENT], s->framep[VP56_FRAME_PREVIOUS]); *(AVFrame*)data = *p; *data_size = sizeof(AVFrame); return buf_size; } void vp56_init(AVCodecContext *avctx, int flip, int has_alpha) { vp56_context_t *s = avctx->priv_data; int i; s->avctx = avctx; avctx->pix_fmt = has_alpha ? PIX_FMT_YUVA420P : PIX_FMT_YUV420P; if (avctx->idct_algo == FF_IDCT_AUTO) avctx->idct_algo = FF_IDCT_VP3; dsputil_init(&s->dsp, avctx); ff_init_scantable(s->dsp.idct_permutation, &s->scantable,ff_zigzag_direct); avcodec_set_dimensions(avctx, 0, 0); for (i=0; i<4; i++) s->framep[i] = &s->frames[i]; s->framep[VP56_FRAME_UNUSED] = s->framep[VP56_FRAME_GOLDEN]; s->framep[VP56_FRAME_UNUSED2] = s->framep[VP56_FRAME_GOLDEN2]; s->edge_emu_buffer_alloc = NULL; s->above_blocks = NULL; s->macroblocks = NULL; s->quantizer = -1; s->deblock_filtering = 1; s->filter = NULL; s->has_alpha = has_alpha; if (flip) { s->flip = -1; s->frbi = 2; s->srbi = 0; } else { s->flip = 1; s->frbi = 0; s->srbi = 2; } } int vp56_free(AVCodecContext *avctx) { vp56_context_t *s = avctx->priv_data; av_free(s->above_blocks); av_free(s->macroblocks); av_free(s->edge_emu_buffer_alloc); if (s->framep[VP56_FRAME_GOLDEN]->data[0]) avctx->release_buffer(avctx, s->framep[VP56_FRAME_GOLDEN]); if (s->framep[VP56_FRAME_GOLDEN2]->data[0]) avctx->release_buffer(avctx, s->framep[VP56_FRAME_GOLDEN2]); if (s->framep[VP56_FRAME_PREVIOUS]->data[0]) avctx->release_buffer(avctx, s->framep[VP56_FRAME_PREVIOUS]); return 0; }