diff options
author | Ronald S. Bultje <rsbultje@gmail.com> | 2011-05-04 15:40:53 +0400 |
---|---|---|
committer | Ronald S. Bultje <rsbultje@gmail.com> | 2011-05-04 15:40:53 +0400 |
commit | 7d2e03afc8573f959aa3641eea5c9f981466bcd8 (patch) | |
tree | 98cdee48456f83bf2ddddefb22c7ff949bf34f60 /libavcodec/vc1dec.c | |
parent | 5c9f147e055c732b1b2a0c8350b4d08b9b30e0cd (diff) |
vc1: make overlap filter for I-frames bit-exact.
Diffstat (limited to 'libavcodec/vc1dec.c')
-rw-r--r-- | libavcodec/vc1dec.c | 279 |
1 files changed, 236 insertions, 43 deletions
diff --git a/libavcodec/vc1dec.c b/libavcodec/vc1dec.c index ccdc3e8521..5c931b122a 100644 --- a/libavcodec/vc1dec.c +++ b/libavcodec/vc1dec.c @@ -160,6 +160,72 @@ enum Imode { /** @} */ //Bitplane group +static void vc1_put_signed_blocks_clamped(VC1Context *v) +{ + MpegEncContext *s = &v->s; + + /* The put pixels loop is always one MB row behind the decoding loop, + * because we can only put pixels when overlap filtering is done, and + * for filtering of the bottom edge of a MB, we need the next MB row + * present as well. + * Within the row, the put pixels loop is also one MB col behind the + * decoding loop. The reason for this is again, because for filtering + * of the right MB edge, we need the next MB present. */ + if (!s->first_slice_line) { + if (s->mb_x) { + s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][0], + s->dest[0] - 16 * s->linesize - 16, + s->linesize); + s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][1], + s->dest[0] - 16 * s->linesize - 8, + s->linesize); + s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][2], + s->dest[0] - 8 * s->linesize - 16, + s->linesize); + s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][3], + s->dest[0] - 8 * s->linesize - 8, + s->linesize); + s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][4], + s->dest[1] - 8 * s->uvlinesize - 8, + s->uvlinesize); + s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][5], + s->dest[2] - 8 * s->uvlinesize - 8, + s->uvlinesize); + } + if (s->mb_x == s->mb_width - 1) { + s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][0], + s->dest[0] - 16 * s->linesize, + s->linesize); + s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][1], + s->dest[0] - 16 * s->linesize + 8, + s->linesize); + s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][2], + s->dest[0] - 8 * s->linesize, + s->linesize); + s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][3], + s->dest[0] - 8 * s->linesize + 8, + s->linesize); + s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][4], + s->dest[1] - 8 * s->uvlinesize, + s->uvlinesize); + s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][5], + s->dest[2] - 8 * s->uvlinesize, + s->uvlinesize); + } + } + +#define inc_blk_idx(idx) do { \ + idx++; \ + if (idx >= v->n_allocated_blks) \ + idx = 0; \ + } while (0) + + inc_blk_idx(v->topleft_blk_idx); + inc_blk_idx(v->top_blk_idx); + inc_blk_idx(v->left_blk_idx); + inc_blk_idx(v->cur_blk_idx); +} + static void vc1_loop_filter_iblk(VC1Context *v, int pq) { MpegEncContext *s = &v->s; @@ -187,6 +253,151 @@ static void vc1_loop_filter_iblk(VC1Context *v, int pq) } } +static void vc1_loop_filter_iblk_delayed(VC1Context *v, int pq) +{ + MpegEncContext *s = &v->s; + int j; + + /* The loopfilter runs 1 row and 1 column behind the overlap filter, which + * means it runs two rows/cols behind the decoding loop. */ + if (!s->first_slice_line) { + if (s->mb_x) { + if (s->mb_y >= s->start_mb_y + 2) { + v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq); + + if (s->mb_x >= 2) + v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 16, s->linesize, pq); + v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 8, s->linesize, pq); + for(j = 0; j < 2; j++) { + v->vc1dsp.vc1_v_loop_filter8(s->dest[j+1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq); + if (s->mb_x >= 2) { + v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1] - 16 * s->uvlinesize - 8, s->uvlinesize, pq); + } + } + } + v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize - 16, s->linesize, pq); + } + + if (s->mb_x == s->mb_width - 1) { + if (s->mb_y >= s->start_mb_y + 2) { + v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq); + + if (s->mb_x) + v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize, s->linesize, pq); + v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize + 8, s->linesize, pq); + for(j = 0; j < 2; j++) { + v->vc1dsp.vc1_v_loop_filter8(s->dest[j+1] - 8 * s->uvlinesize, s->uvlinesize, pq); + if (s->mb_x >= 2) { + v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1] - 16 * s->uvlinesize, s->uvlinesize, pq); + } + } + } + v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize, s->linesize, pq); + } + + if (s->mb_y == s->mb_height) { + if (s->mb_x) { + if (s->mb_x >= 2) + v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq); + v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 8, s->linesize, pq); + if (s->mb_x >= 2) { + for(j = 0; j < 2; j++) { + v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq); + } + } + } + + if (s->mb_x == s->mb_width - 1) { + if (s->mb_x) + v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq); + v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq); + if (s->mb_x) { + for(j = 0; j < 2; j++) { + v->vc1dsp.vc1_h_loop_filter8(s->dest[j+1] - 8 * s->uvlinesize, s->uvlinesize, pq); + } + } + } + } + } +} + +static void vc1_smooth_overlap_filter_iblk(VC1Context *v) +{ + MpegEncContext *s = &v->s; + int mb_pos; + + if (v->condover == CONDOVER_NONE) + return; + + mb_pos = s->mb_x + s->mb_y * s->mb_stride; + + /* Within a MB, the horizontal overlap always runs before the vertical. + * To accomplish that, we run the H on left and internal borders of the + * currently decoded MB. Then, we wait for the next overlap iteration + * to do H overlap on the right edge of this MB, before moving over and + * running the V overlap. Therefore, the V overlap makes us trail by one + * MB col and the H overlap filter makes us trail by one MB row. This + * is reflected in the time at which we run the put_pixels loop. */ + if(v->condover == CONDOVER_ALL || v->pq >= 9 || v->over_flags_plane[mb_pos]) { + if(s->mb_x && (v->condover == CONDOVER_ALL || v->pq >= 9 || + v->over_flags_plane[mb_pos - 1])) { + v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][1], + v->block[v->cur_blk_idx][0]); + v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][3], + v->block[v->cur_blk_idx][2]); + if(!(s->flags & CODEC_FLAG_GRAY)) { + v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][4], + v->block[v->cur_blk_idx][4]); + v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][5], + v->block[v->cur_blk_idx][5]); + } + } + v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][0], + v->block[v->cur_blk_idx][1]); + v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][2], + v->block[v->cur_blk_idx][3]); + + if (s->mb_x == s->mb_width - 1) { + if(!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 || + v->over_flags_plane[mb_pos - s->mb_stride])) { + v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][2], + v->block[v->cur_blk_idx][0]); + v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][3], + v->block[v->cur_blk_idx][1]); + if(!(s->flags & CODEC_FLAG_GRAY)) { + v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][4], + v->block[v->cur_blk_idx][4]); + v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][5], + v->block[v->cur_blk_idx][5]); + } + } + v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][0], + v->block[v->cur_blk_idx][2]); + v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][1], + v->block[v->cur_blk_idx][3]); + } + } + if (s->mb_x && (v->condover == CONDOVER_ALL || v->over_flags_plane[mb_pos - 1])) { + if(!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 || + v->over_flags_plane[mb_pos - s->mb_stride - 1])) { + v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][2], + v->block[v->left_blk_idx][0]); + v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][3], + v->block[v->left_blk_idx][1]); + if(!(s->flags & CODEC_FLAG_GRAY)) { + v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][4], + v->block[v->left_blk_idx][4]); + v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][5], + v->block[v->left_blk_idx][5]); + } + } + v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][0], + v->block[v->left_blk_idx][2]); + v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][1], + v->block[v->left_blk_idx][3]); + } +} + /** Do motion compensation over 1 macroblock * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c */ @@ -2773,7 +2984,6 @@ static void vc1_decode_i_blocks_adv(VC1Context *v) int mb_pos; int mquant = v->pq; int mqdiff; - int overlap; GetBitContext *gb = &s->gb; /* select codingmode used for VLC tables selection */ @@ -2816,15 +3026,9 @@ static void vc1_decode_i_blocks_adv(VC1Context *v) s->mb_x = 0; ff_init_block_index(s); for(;s->mb_x < s->mb_width; s->mb_x++) { - uint8_t *dst[6]; + DCTELEM (*block)[64] = v->block[v->cur_blk_idx]; ff_update_block_index(s); - dst[0] = s->dest[0]; - dst[1] = dst[0] + 8; - dst[2] = s->dest[0] + s->linesize * 8; - dst[3] = dst[2] + 8; - dst[4] = s->dest[1]; - dst[5] = s->dest[2]; - s->dsp.clear_blocks(s->block[0]); + s->dsp.clear_blocks(block[0]); mb_pos = s->mb_x + s->mb_y * s->mb_stride; s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA; s->current_picture.motion_val[1][s->block_index[0]][0] = 0; @@ -2837,13 +3041,8 @@ static void vc1_decode_i_blocks_adv(VC1Context *v) else v->s.ac_pred = v->acpred_plane[mb_pos]; - if(v->condover == CONDOVER_SELECT) { - if(v->overflg_is_raw) - overlap = get_bits1(&v->s.gb); - else - overlap = v->over_flags_plane[mb_pos]; - } else - overlap = (v->condover == CONDOVER_ALL); + if (v->condover == CONDOVER_SELECT && v->overflg_is_raw) + v->over_flags_plane[mb_pos] = get_bits1(&v->s.gb); GET_MQUANT(); @@ -2865,37 +3064,15 @@ static void vc1_decode_i_blocks_adv(VC1Context *v) v->a_avail = !s->first_slice_line || (k==2 || k==3); v->c_avail = !!s->mb_x || (k==1 || k==3); - vc1_decode_i_block_adv(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2, mquant); + vc1_decode_i_block_adv(v, block[k], k, val, (k<4)? v->codingset : v->codingset2, mquant); if (k > 3 && (s->flags & CODEC_FLAG_GRAY)) continue; - v->vc1dsp.vc1_inv_trans_8x8(s->block[k]); - s->dsp.put_signed_pixels_clamped(s->block[k], dst[k], - k & 4 ? s->uvlinesize : s->linesize); + v->vc1dsp.vc1_inv_trans_8x8(block[k]); } - if(overlap) { - if(s->mb_x) { - v->vc1dsp.vc1_h_overlap(s->dest[0], s->linesize); - v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize); - if(!(s->flags & CODEC_FLAG_GRAY)) { - v->vc1dsp.vc1_h_overlap(s->dest[1], s->uvlinesize); - v->vc1dsp.vc1_h_overlap(s->dest[2], s->uvlinesize); - } - } - v->vc1dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize); - v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize); - if(!s->first_slice_line) { - v->vc1dsp.vc1_v_overlap(s->dest[0], s->linesize); - v->vc1dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize); - if(!(s->flags & CODEC_FLAG_GRAY)) { - v->vc1dsp.vc1_v_overlap(s->dest[1], s->uvlinesize); - v->vc1dsp.vc1_v_overlap(s->dest[2], s->uvlinesize); - } - } - v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize); - v->vc1dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize); - } - if(v->s.loop_filter) vc1_loop_filter_iblk(v, v->pq); + vc1_smooth_overlap_filter_iblk(v); + vc1_put_signed_blocks_clamped(v); + if(v->s.loop_filter) vc1_loop_filter_iblk_delayed(v, v->pq); if(get_bits_count(&s->gb) > v->bits) { ff_er_add_slice(s, 0, s->start_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)); @@ -2909,6 +3086,15 @@ static void vc1_decode_i_blocks_adv(VC1Context *v) ff_draw_horiz_band(s, (s->mb_y-1) * 16, 16); s->first_slice_line = 0; } + + /* raw bottom MB row */ + s->mb_x = 0; + ff_init_block_index(s); + for(;s->mb_x < s->mb_width; s->mb_x++) { + ff_update_block_index(s); + vc1_put_signed_blocks_clamped(v); + if(v->s.loop_filter) vc1_loop_filter_iblk_delayed(v, v->pq); + } if (v->s.loop_filter) ff_draw_horiz_band(s, (s->mb_height-1)*16, 16); ff_er_add_slice(s, 0, s->start_mb_y, s->mb_width - 1, s->end_mb_y - 1, (AC_END|DC_END|MV_END)); @@ -3063,6 +3249,10 @@ static void vc1_decode_blocks(VC1Context *v) if(v->x8_type){ ff_intrax8_decode_picture(&v->x8, 2*v->pq+v->halfpq, v->pq*(!v->pquantizer) ); }else{ + v->cur_blk_idx = 0; + v->left_blk_idx = -1; + v->topleft_blk_idx = 1; + v->top_blk_idx = 2; switch(v->s.pict_type) { case AV_PICTURE_TYPE_I: if(v->profile == PROFILE_ADVANCED) @@ -3333,6 +3523,8 @@ static av_cold int vc1_decode_init(AVCodecContext *avctx) v->acpred_plane = av_malloc(s->mb_stride * s->mb_height); v->over_flags_plane = av_malloc(s->mb_stride * s->mb_height); + v->n_allocated_blks = s->mb_width + 2; + v->block = av_malloc(sizeof(*v->block) * v->n_allocated_blks); v->cbp_base = av_malloc(sizeof(v->cbp_base[0]) * 2 * s->mb_stride); v->cbp = v->cbp_base + s->mb_stride; v->ttblk_base = av_malloc(sizeof(v->ttblk_base[0]) * 2 * s->mb_stride); @@ -3598,6 +3790,7 @@ static av_cold int vc1_decode_end(AVCodecContext *avctx) av_freep(&v->acpred_plane); av_freep(&v->over_flags_plane); av_freep(&v->mb_type_base); + av_freep(&v->block); av_freep(&v->cbp_base); av_freep(&v->ttblk_base); av_freep(&v->is_intra_base); // FIXME use v->mb_type[] |