comparison h264_loopfilter.c @ 11025:cd1f5f6a2e45 libavcodec

Use table to speedup access to non_zero_count in MBAFF with differing interlacing. ~4 cpu cycles speedup

11024:5ab861519e79

        if( IS_INTRA(mb_type) )
            *(uint64_t*)&bS[0]=
            *(uint64_t*)&bS[4]= 0x0004000400040004ULL;
        else {
            for( i = 0; i < 8; i++ ) {
                int j= MB_FIELD ? i>>2 : i&1;
                int mbn_xy = h->left_mb_xy[j];
                int mbn_type= h->left_type[j];

                else{
                    bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
                         ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
                            (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
                                                                       :
                            h->non_zero_count[mbn_xy][7+(MB_FIELD ? (i&3) : (i>>2)+(mb_y&1)*2)*8]));
                }
            }
        }

        mb_qp = s->current_picture.qscale_table[mb_xy];

11025:cd1f5f6a2e45

        if( IS_INTRA(mb_type) )
            *(uint64_t*)&bS[0]=
            *(uint64_t*)&bS[4]= 0x0004000400040004ULL;
        else {
            static const uint8_t offset[2][2][8]={
                {
                    {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1},
                    {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3},
                },{
                    {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
                    {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
                }
            };
            const uint8_t *off= offset[MB_FIELD][mb_y&1];
            for( i = 0; i < 8; i++ ) {
                int j= MB_FIELD ? i>>2 : i&1;
                int mbn_xy = h->left_mb_xy[j];
                int mbn_type= h->left_type[j];

                else{
                    bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
                         ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
                            (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
                                                                       :
                            h->non_zero_count[mbn_xy][ off[i] ]));
                }
            }
        }

        mb_qp = s->current_picture.qscale_table[mb_xy];