comparison vp8.c @ 12241:c7f6ddcc5c01 libavcodec

VP8: optimize DC-only chroma case in the same way as luma. Add MMX idct_dc_add4uv function for this case. ~40% faster chroma idct.

12240:e6ade5e849c9

                            else
                                s->vp8dsp.vp8_idct_add(y_dst+4*x, s->block[y][x], s->linesize);
                        }
                    }
                } else {
                    s->vp8dsp.vp8_idct_dc_add4(y_dst, s->block[y], s->linesize);
                }
            }
            y_dst += 4*s->linesize;
        }
    }

    for (ch = 0; ch < 2; ch++) {
        if (AV_RN32A(s->non_zero_count_cache[4+ch])) {
            uint8_t *ch_dst = dst[1+ch];
            for (y = 0; y < 2; y++) {
                for (x = 0; x < 2; x++) {
                    int nnz = s->non_zero_count_cache[4+ch][(y<<1)+x];
                    if (nnz) {
                        if (nnz == 1)
                            s->vp8dsp.vp8_idct_dc_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
                        else
                            s->vp8dsp.vp8_idct_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
                    }
                }
                ch_dst += 4*s->uvlinesize;
            }
        }
    }
}

12241:c7f6ddcc5c01

                            else
                                s->vp8dsp.vp8_idct_add(y_dst+4*x, s->block[y][x], s->linesize);
                        }
                    }
                } else {
                    s->vp8dsp.vp8_idct_dc_add4y(y_dst, s->block[y], s->linesize);
                }
            }
            y_dst += 4*s->linesize;
        }
    }

    for (ch = 0; ch < 2; ch++) {
        uint32_t nnz4 = AV_RN32A(s->non_zero_count_cache[4+ch]);
        if (nnz4) {
            uint8_t *ch_dst = dst[1+ch];
            if (nnz4&~0x01010101) {
                for (y = 0; y < 2; y++) {
                    for (x = 0; x < 2; x++) {
                        int nnz = s->non_zero_count_cache[4+ch][(y<<1)+x];
                        if (nnz) {
                            if (nnz == 1)
                                s->vp8dsp.vp8_idct_dc_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
                            else
                                s->vp8dsp.vp8_idct_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
                        }
                    }
                    ch_dst += 4*s->uvlinesize;
                }
            } else {
                s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, s->block[4+ch], s->uvlinesize);
            }
        }
    }
}