Mercurial > libavcodec.hg
view h264_direct.c @ 10865:bcdc5343a577 libavcodec
Use MULH instead of 64x64 multiplication, around 5% overall speedup on Intel Atom.
Patch by myself and Yuriy Kaminskiy [yumkam mail ru]
author | reimar |
---|---|
date | Tue, 12 Jan 2010 23:02:07 +0000 |
parents | e3f5eb016712 |
children | b847f02d5b03 |
line wrap: on
line source
/* * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at> * * 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 */ /** * @file libavcodec/h264_direct.c * H.264 / AVC / MPEG4 part10 direct mb/block decoding. * @author Michael Niedermayer <michaelni@gmx.at> */ #include "internal.h" #include "dsputil.h" #include "avcodec.h" #include "mpegvideo.h" #include "h264.h" #include "h264_mvpred.h" #include "rectangle.h" #if ARCH_X86 #include "x86/h264_i386.h" #endif //#undef NDEBUG #include <assert.h> static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){ int poc0 = h->ref_list[0][i].poc; int td = av_clip(poc1 - poc0, -128, 127); if(td == 0 || h->ref_list[0][i].long_ref){ return 256; }else{ int tb = av_clip(poc - poc0, -128, 127); int tx = (16384 + (FFABS(td) >> 1)) / td; return av_clip((tb*tx + 32) >> 6, -1024, 1023); } } void ff_h264_direct_dist_scale_factor(H264Context * const h){ MpegEncContext * const s = &h->s; const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ]; const int poc1 = h->ref_list[1][0].poc; int i, field; for(field=0; field<2; field++){ const int poc = h->s.current_picture_ptr->field_poc[field]; const int poc1 = h->ref_list[1][0].field_poc[field]; for(i=0; i < 2*h->ref_count[0]; i++) h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16); } for(i=0; i<h->ref_count[0]; i++){ h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i); } } static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){ MpegEncContext * const s = &h->s; Picture * const ref1 = &h->ref_list[1][0]; int j, old_ref, rfield; int start= mbafi ? 16 : 0; int end = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list]; int interl= mbafi || s->picture_structure != PICT_FRAME; /* bogus; fills in for missing frames */ memset(map[list], 0, sizeof(map[list])); for(rfield=0; rfield<2; rfield++){ for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){ int poc = ref1->ref_poc[colfield][list][old_ref]; if (!interl) poc |= 3; else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed poc= (poc&~3) + rfield + 1; for(j=start; j<end; j++){ if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){ int cur_ref= mbafi ? (j-16)^field : j; map[list][2*old_ref + (rfield^field) + 16] = cur_ref; if(rfield == field) map[list][old_ref] = cur_ref; break; } } } } } void ff_h264_direct_ref_list_init(H264Context * const h){ MpegEncContext * const s = &h->s; Picture * const ref1 = &h->ref_list[1][0]; Picture * const cur = s->current_picture_ptr; int list, j, field; int sidx= (s->picture_structure&1)^1; int ref1sidx= (ref1->reference&1)^1; for(list=0; list<2; list++){ cur->ref_count[sidx][list] = h->ref_count[list]; for(j=0; j<h->ref_count[list]; j++) cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3); } if(s->picture_structure == PICT_FRAME){ memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0])); memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0])); } cur->mbaff= FRAME_MBAFF; if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred) return; for(list=0; list<2; list++){ fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0); for(field=0; field<2; field++) fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1); } } void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){ MpegEncContext * const s = &h->s; int b8_stride = h->b8_stride; int b4_stride = h->b_stride; int mb_xy = h->mb_xy; int mb_type_col[2]; const int16_t (*l1mv0)[2], (*l1mv1)[2]; const int8_t *l1ref0, *l1ref1; const int is_b8x8 = IS_8X8(*mb_type); unsigned int sub_mb_type; int i8, i4; assert(h->ref_list[1][0].reference&3); #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM) if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL if(!IS_INTERLACED(*mb_type)){ // AFR/FR -> AFL/FL int cur_poc = s->current_picture_ptr->poc; int *col_poc = h->ref_list[1]->field_poc; int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc); mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride; b8_stride = 0; }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity int fieldoff= 2*(h->ref_list[1][0].reference)-3; mb_xy += s->mb_stride*fieldoff; } goto single_col; }else{ // AFL/AFR/FR/FL -> AFR/FR if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride; mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy]; mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride]; b8_stride *= 3; b4_stride *= 6; //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) && !is_b8x8){ sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */ }else{ sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1; } }else{ // AFR/FR -> AFR/FR single_col: mb_type_col[0] = mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy]; if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){ /* FIXME save sub mb types from previous frames (or derive from MVs) * so we know exactly what block size to use */ sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */ *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1; }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){ sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */ }else{ sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1; } } } l1mv0 = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]]; l1mv1 = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]]; l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]]; l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]]; if(!b8_stride){ if(s->mb_y&1){ l1ref0 += h->b8_stride; l1ref1 += h->b8_stride; l1mv0 += 2*b4_stride; l1mv1 += 2*b4_stride; } } if(h->direct_spatial_mv_pred){ int ref[2]; int mv[2][2]; int list; /* FIXME interlacing + spatial direct uses wrong colocated block positions */ /* ref = min(neighbors) */ for(list=0; list<2; list++){ int refa = h->ref_cache[list][scan8[0] - 1]; int refb = h->ref_cache[list][scan8[0] - 8]; int refc = h->ref_cache[list][scan8[0] - 8 + 4]; if(refc == PART_NOT_AVAILABLE) refc = h->ref_cache[list][scan8[0] - 8 - 1]; ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc); if(ref[list] < 0) ref[list] = -1; } if(ref[0] < 0 && ref[1] < 0){ ref[0] = ref[1] = 0; mv[0][0] = mv[0][1] = mv[1][0] = mv[1][1] = 0; }else{ for(list=0; list<2; list++){ if(ref[list] >= 0) pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]); else mv[list][0] = mv[list][1] = 0; } } if(ref[1] < 0){ if(!is_b8x8) *mb_type &= ~MB_TYPE_L1; sub_mb_type &= ~MB_TYPE_L1; }else if(ref[0] < 0){ if(!is_b8x8) *mb_type &= ~MB_TYPE_L0; sub_mb_type &= ~MB_TYPE_L0; } if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){ for(i8=0; i8<4; i8++){ int x8 = i8&1; int y8 = i8>>1; int xy8 = x8+y8*b8_stride; int xy4 = 3*x8+y8*b4_stride; int a=0, b=0; if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) continue; h->sub_mb_type[i8] = sub_mb_type; fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1); fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1); if(!IS_INTRA(mb_type_col[y8]) && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1) || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){ if(ref[0] > 0) a= pack16to32(mv[0][0],mv[0][1]); if(ref[1] > 0) b= pack16to32(mv[1][0],mv[1][1]); }else{ a= pack16to32(mv[0][0],mv[0][1]); b= pack16to32(mv[1][0],mv[1][1]); } fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4); fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4); } }else if(IS_16X16(*mb_type)){ int a=0, b=0; fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1); fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1); if(!IS_INTRA(mb_type_col[0]) && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1) || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1 && (h->x264_build>33 || !h->x264_build)))){ if(ref[0] > 0) a= pack16to32(mv[0][0],mv[0][1]); if(ref[1] > 0) b= pack16to32(mv[1][0],mv[1][1]); }else{ a= pack16to32(mv[0][0],mv[0][1]); b= pack16to32(mv[1][0],mv[1][1]); } fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4); fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4); }else{ for(i8=0; i8<4; i8++){ const int x8 = i8&1; const int y8 = i8>>1; if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) continue; h->sub_mb_type[i8] = sub_mb_type; fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4); fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4); fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1); fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1); /* col_zero_flag */ if(!IS_INTRA(mb_type_col[0]) && ( l1ref0[x8 + y8*b8_stride] == 0 || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0 && (h->x264_build>33 || !h->x264_build)))){ const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1; if(IS_SUB_8X8(sub_mb_type)){ const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride]; if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ if(ref[0] == 0) fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); if(ref[1] == 0) fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); } }else for(i4=0; i4<4; i4++){ const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride]; if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ if(ref[0] == 0) *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0; if(ref[1] == 0) *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0; } } } } } }else{ /* direct temporal mv pred */ const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]}; const int *dist_scale_factor = h->dist_scale_factor; int ref_offset= 0; if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){ map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0]; map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1]; dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1]; } if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0])) ref_offset += 16; if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){ /* FIXME assumes direct_8x8_inference == 1 */ int y_shift = 2*!IS_INTERLACED(*mb_type); for(i8=0; i8<4; i8++){ const int x8 = i8&1; const int y8 = i8>>1; int ref0, scale; const int16_t (*l1mv)[2]= l1mv0; if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) continue; h->sub_mb_type[i8] = sub_mb_type; fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); if(IS_INTRA(mb_type_col[y8])){ fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); continue; } ref0 = l1ref0[x8 + y8*b8_stride]; if(ref0 >= 0) ref0 = map_col_to_list0[0][ref0 + ref_offset]; else{ ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset]; l1mv= l1mv1; } scale = dist_scale_factor[ref0]; fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); { const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride]; int my_col = (mv_col[1]<<y_shift)/2; int mx = (scale * mv_col[0] + 128) >> 8; int my = (scale * my_col + 128) >> 8; fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4); } } return; } /* one-to-one mv scaling */ if(IS_16X16(*mb_type)){ int ref, mv0, mv1; fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1); if(IS_INTRA(mb_type_col[0])){ ref=mv0=mv1=0; }else{ const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset] : map_col_to_list0[1][l1ref1[0] + ref_offset]; const int scale = dist_scale_factor[ref0]; const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0]; int mv_l0[2]; mv_l0[0] = (scale * mv_col[0] + 128) >> 8; mv_l0[1] = (scale * mv_col[1] + 128) >> 8; ref= ref0; mv0= pack16to32(mv_l0[0],mv_l0[1]); mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]); } fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1); fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4); fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4); }else{ for(i8=0; i8<4; i8++){ const int x8 = i8&1; const int y8 = i8>>1; int ref0, scale; const int16_t (*l1mv)[2]= l1mv0; if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) continue; h->sub_mb_type[i8] = sub_mb_type; fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); if(IS_INTRA(mb_type_col[0])){ fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); continue; } ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset; if(ref0 >= 0) ref0 = map_col_to_list0[0][ref0]; else{ ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset]; l1mv= l1mv1; } scale = dist_scale_factor[ref0]; fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); if(IS_SUB_8X8(sub_mb_type)){ const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride]; int mx = (scale * mv_col[0] + 128) >> 8; int my = (scale * mv_col[1] + 128) >> 8; fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4); }else for(i4=0; i4<4; i4++){ const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride]; int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]]; mv_l0[0] = (scale * mv_col[0] + 128) >> 8; mv_l0[1] = (scale * mv_col[1] + 128) >> 8; *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]); } } } } }