comparison h264_direct.c @ 11203:10c06a9bd3d9 libavcodec

H264: use alias-safe macros This eliminates all aliasing violation warnings in h264 code. No measurable speed difference with gcc-4.4.3 on i7.

11202:778139a5e058

                mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
                                     mid_pred(A[1], B[1], C[1]) );
            }else {
                assert(match_count==1);
                if(left_ref==ref[list]){
                    mv[list]= *(uint32_t*)A;
                }else if(top_ref==ref[list]){
                    mv[list]= *(uint32_t*)B;
                }else{
                    mv[list]= *(uint32_t*)C;
                }
            }
        }else{
            int mask= ~(MB_TYPE_L0 << (2*list));
            mv[list] = 0;

                        int m=0;
                    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;
                            m++;
                        }
                    }
                    if(!(m&3))
                        h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;

                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]);
                }
            }
        }
    }
}

11203:10c06a9bd3d9

                mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
                                     mid_pred(A[1], B[1], C[1]) );
            }else {
                assert(match_count==1);
                if(left_ref==ref[list]){
                    mv[list]= AV_RN32A(A);
                }else if(top_ref==ref[list]){
                    mv[list]= AV_RN32A(B);
                }else{
                    mv[list]= AV_RN32A(C);
                }
            }
        }else{
            int mask= ~(MB_TYPE_L0 << (2*list));
            mv[list] = 0;

                        int m=0;
                    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)
                                AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
                            if(ref[1] == 0)
                                AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
                            m++;
                        }
                    }
                    if(!(m&3))
                        h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;

                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;
                    AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
                }
            }
        }
    }
}