comparison h264_loopfilter.c @ 11042:74f672a1f763 libavcodec

Change wraper functions to always inline, they are faster now that way. 1% faster MBAFF decoding overall, maybe ~0.1% faster for the cathedral sample.

11041:6d74270c4f6f

    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
};

static void av_noinline filter_mb_edgev( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h) {
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = beta_table[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

        h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
    }
}
static void av_noinline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = beta_table[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

            }
        }
    }
}

static void av_noinline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = beta_table[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    } else {
        h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
    }
}

static void av_noinline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = beta_table[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

11042:74f672a1f763

    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
    {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
};

static void av_always_inline filter_mb_edgev( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h) {
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = beta_table[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

        h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
    }
}
static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = beta_table[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

            }
        }
    }
}

static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = beta_table[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    } else {
        h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
    }
}

static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
    const unsigned int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = beta_table[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;