web123456

x264 Loop Filtering Principles Series: x264_macroblock_deblock_strength Function

void x264_macroblock_deblock_strength( x264_t *h ) { uint8_t (*bs)[8][4] = h->mb.cache.deblock_strength; if( IS_INTRA( h->mb.i_type ) ) { M32( bs[0][1] ) = 0x03030303; M64( bs[0][2] ) = 0x0303030303030303ULL; M32( bs[1][1] ) = 0x03030303; M64( bs[1][2] ) = 0x0303030303030303ULL; return; } /* Early termination: in this case, nnz guarantees all edges use strength 2.*/ if( h->mb.b_transform_8x8 && !CHROMA444 ) { int cbp_mask = 0xf >> CHROMA_V_SHIFT; if( (h->mb.i_cbp_luma&cbp_mask) == cbp_mask ) { M32( bs[0][0] ) = 0x02020202; M32( bs[0][2] ) = 0x02020202; M32( bs[0][4] ) = 0x02020202; M64( bs[1][0] ) = 0x0202020202020202ULL; /* [1][1] and [1][3] has to be set for 4:2:2 */ M64( bs[1][2] ) = 0x0202020202020202ULL; M32( bs[1][4] ) = 0x02020202; return; } } int neighbour_changed = 0; if( h->sh.i_disable_deblocking_filter_idc != 2 ) { neighbour_changed = h->mb.i_neighbour_frame&~h->mb.i_neighbour; h->mb.i_neighbour = h->mb.i_neighbour_frame; } /* MBAFF deblock uses different left neighbors from encoding */ if( SLICE_MBAFF && (h->mb.i_neighbour & MB_LEFT) && (h->mb.field[h->mb.i_mb_xy - 1] != MB_INTERLACED) ) { h->mb.i_mb_left_xy[1] = h->mb.i_mb_left_xy[0] = h->mb.i_mb_xy - 1; if( h->mb.i_mb_y&1 ) h->mb.i_mb_left_xy[0] -= h->mb.i_mb_stride; else h->mb.i_mb_left_xy[1] += h->mb.i_mb_stride; } /* If we have multiple slices and we're deblocking on slice edges, we * have to reload neighbour data. */ if( neighbour_changed ) { int top_y = h->mb.i_mb_top_y; int top_8x8 = (2*top_y+1) * h->mb.i_b8_stride + 2*h->mb.i_mb_x; int top_4x4 = (4*top_y+3) * h->mb.i_b4_stride + 4*h->mb.i_mb_x; int s8x8 = h->mb.i_b8_stride; int s4x4 = h->mb.i_b4_stride; uint8_t (*nnz)[48] = h->mb.non_zero_count; const x264_left_table_t *left_index_table = SLICE_MBAFF ? h->mb.left_index_table : &left_indices[3]; if( neighbour_changed & MB_TOP ) CP32( &h->mb.cache.non_zero_count[x264_scan8[0] - 8], &nnz[h->mb.i_mb_top_xy][12] ); if( neighbour_changed & MB_LEFT ) { int *left = h->mb.i_mb_left_xy; h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left[0]][left_index_table->nnz[0]]; h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left[0]][left_index_table->nnz[1]]; h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left[1]][left_index_table->nnz[2]]; h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left[1]][left_index_table->nnz[3]]; } for( int l = 0; l <= (h->sh.i_type == SLICE_TYPE_B); l++ ) { int16_t (*mv)[2] = h->mb.mv[l]; int8_t *ref = h->mb.ref[l]; int i8 = x264_scan8[0] - 8; if( neighbour_changed & MB_TOP ) { h->mb.cache.ref[l][i8+0] = h->mb.cache.ref[l][i8+1] = ref[top_8x8 + 0]; h->mb.cache.ref[l][i8+2] = h->mb.cache.ref[l][i8+3] = ref[top_8x8 + 1]; CP128( h->mb.cache.mv[l][i8], mv[top_4x4] ); } i8 = x264_scan8[0] - 1; if( neighbour_changed & MB_LEFT ) { h->mb.cache.ref[l][i8+0*8] = h->mb.cache.ref[l][i8+1*8] = ref[h->mb.left_b8[0] + 1 + s8x8*left_index_table->ref[0]]; h->mb.cache.ref[l][i8+2*8] = h->mb.cache.ref[l][i8+3*8] = ref[h->mb.left_b8[1] + 1 + s8x8*left_index_table->ref[2]]; CP32( h->mb.cache.mv[l][i8+0*8], mv[h->mb.left_b4[0] + 3 + s4x4*left_index_table->mv[0]] ); CP32( h->mb.cache.mv[l][i8+1*8], mv[h->mb.left_b4[0] + 3 + s4x4*left_index_table->mv[1]] ); CP32( h->mb.cache.mv[l][i8+2*8], mv[h->mb.left_b4[1] + 3 + s4x4*left_index_table->mv[2]] ); CP32( h->mb.cache.mv[l][i8+3*8], mv[h->mb.left_b4[1] + 3 + s4x4*left_index_table->mv[3]] ); } } } if( h->param.analyse.i_weighted_pred == X264_WEIGHTP_SMART && h->sh.i_type == SLICE_TYPE_P ) { /* Handle reference frame duplicates */ int i8 = x264_scan8[0] - 8; h->mb.cache.ref[0][i8+0] = h->mb.cache.ref[0][i8+1] = deblock_ref_table(h->mb.cache.ref[0][i8+0]); h->mb.cache.ref[0][i8+2] = h->mb.cache.ref[0][i8+3] = deblock_ref_table(h->mb.cache.ref[0][i8+2]); i8 = x264_scan8[0] - 1; h->mb.cache.ref[0][i8+0*8] = h->mb.cache.ref[0][i8+1*8] = deblock_ref_table(h->mb.cache.ref[0][i8+0*8]); h->mb.cache.ref[0][i8+2*8] = h->mb.cache.ref[0][i8+3*8] = deblock_ref_table(h->mb.cache.ref[0][i8+2*8]); int ref0 = deblock_ref_table(h->mb.cache.ref[0][x264_scan8[ 0]]); int ref1 = deblock_ref_table(h->mb.cache.ref[0][x264_scan8[ 4]]); int ref2 = deblock_ref_table(h->mb.cache.ref[0][x264_scan8[ 8]]); int ref3 = deblock_ref_table(h->mb.cache.ref[0][x264_scan8[12]]); uint32_t reftop = pack16to32( (uint8_t)ref0, (uint8_t)ref1 ) * 0x0101; uint32_t refbot = pack16to32( (uint8_t)ref2, (uint8_t)ref3 ) * 0x0101; M32( &h->mb.cache.ref[0][x264_scan8[0]+8*0] ) = reftop; M32( &h->mb.cache.ref[0][x264_scan8[0]+8*1] ) = reftop; M32( &h->mb.cache.ref[0][x264_scan8[0]+8*2] ) = refbot; M32( &h->mb.cache.ref[0][x264_scan8[0]+8*3] ) = refbot; } /* Munge NNZ for cavlc + 8x8dct */ if( !h->param.b_cabac && h->pps->b_transform_8x8_mode ) { uint8_t (*nnz)[48] = h->mb.non_zero_count; int top = h->mb.i_mb_top_xy; int *left = h->mb.i_mb_left_xy; if( (h->mb.i_neighbour & MB_TOP) && h->mb.mb_transform_size[top] ) { int i8 = x264_scan8[0] - 8; int nnz_top0 = M16( &nnz[top][8] ) | M16( &nnz[top][12] ); int nnz_top1 = M16( &nnz[top][10] ) | M16( &nnz[top][14] ); M16( &h->mb.cache.non_zero_count[i8+0] ) = nnz_top0 ? 0x0101 : 0; M16( &h->mb.cache.non_zero_count[i8+2] ) = nnz_top1 ? 0x0101 : 0; } if( h->mb.i_neighbour & MB_LEFT ) { int i8 = x264_scan8[0] - 1; if( h->mb.mb_transform_size[left[0]] ) { int nnz_left0 = M16( &nnz[left[0]][2] ) | M16( &nnz[left[0]][6] ); h->mb.cache.non_zero_count[i8+8*0] = !!nnz_left0; h->mb.cache.non_zero_count[i8+8*1] = !!nnz_left0; } if( h->mb.mb_transform_size[left[1]] ) { int nnz_left1 = M16( &nnz[left[1]][10] ) | M16( &nnz[left[1]][14] ); h->mb.cache.non_zero_count[i8+8*2] = !!nnz_left1; h->mb.cache.non_zero_count[i8+8*3] = !!nnz_left1; } } if( h->mb.b_transform_8x8 ) { int nnz0 = M16( &h->mb.cache.non_zero_count[x264_scan8[ 0]] ) | M16( &h->mb.cache.non_zero_count[x264_scan8[ 2]] ); int nnz1 = M16( &h->mb.cache.non_zero_count[x264_scan8[ 4]] ) | M16( &h->mb.cache.non_zero_count[x264_scan8[ 6]] ); int nnz2 = M16( &h->mb.cache.non_zero_count[x264_scan8[ 8]] ) | M16( &h->mb.cache.non_zero_count[x264_scan8[10]] ); int nnz3 = M16( &h->mb.cache.non_zero_count[x264_scan8[12]] ) | M16( &h->mb.cache.non_zero_count[x264_scan8[14]] ); uint32_t nnztop = pack16to32( !!nnz0, !!nnz1 ) * 0x0101; uint32_t nnzbot = pack16to32( !!nnz2, !!nnz3 ) * 0x0101; M32( &h->mb.cache.non_zero_count[x264_scan8[0]+8*0] ) = nnztop; M32( &h->mb.cache.non_zero_count[x264_scan8[0]+8*1] ) = nnztop; M32( &h->mb.cache.non_zero_count[x264_scan8[0]+8*2] ) = nnzbot; M32( &h->mb.cache.non_zero_count[x264_scan8[0]+8*3] ) = nnzbot; } } h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv, bs, 4 >> MB_INTERLACED, h->sh.i_type == SLICE_TYPE_B ); if( SLICE_MBAFF ) macroblock_deblock_strength_mbaff( h, bs ); }