comparison h264_loopfilter.c @ 10960:10759fd39860 libavcodec

Gcc idiocy fixes related to filter_mb_edge*. Change order of operands as gcc uses a hardcoded register per operand it seems even for static functions thus reducing unneeded moved (now functions try to pass the same argument in the same spot). Change signed int to unsigned int for array indexes as signed requires signed extension while unsigned is free. move the +52 up and merge it where it will end as a lea instruction, gcc always splits the 52 out there turning the free +52 into an expensive one otherwise. The changed code becomes a little faster.

10959:d5320d1acaee

    {-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( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
    const int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = (alpha_table+52)[index_a];
    const int beta  = (beta_table+52)[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
        tc[0] = (tc0_table+52)[index_a][bS[0]];
        tc[1] = (tc0_table+52)[index_a][bS[1]];
        tc[2] = (tc0_table+52)[index_a][bS[2]];
        tc[3] = (tc0_table+52)[index_a][bS[3]];
        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( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
    const int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = (alpha_table+52)[index_a];
    const int beta  = (beta_table+52)[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
        tc[0] = (tc0_table+52)[index_a][bS[0]]+1;
        tc[1] = (tc0_table+52)[index_a][bS[1]]+1;
        tc[2] = (tc0_table+52)[index_a][bS[2]]+1;
        tc[3] = (tc0_table+52)[index_a][bS[3]]+1;
        h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
    }
}

            }
        }
    }
}

static void av_noinline filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
    const int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = (alpha_table+52)[index_a];
    const int beta  = (beta_table+52)[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
        tc[0] = (tc0_table+52)[index_a][bS[0]];
        tc[1] = (tc0_table+52)[index_a][bS[1]];
        tc[2] = (tc0_table+52)[index_a][bS[2]];
        tc[3] = (tc0_table+52)[index_a][bS[3]];
        h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
    }
}

static void av_noinline filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
    const int index_a = qp + h->slice_alpha_c0_offset;
    const int alpha = (alpha_table+52)[index_a];
    const int beta  = (beta_table+52)[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
        tc[0] = (tc0_table+52)[index_a][bS[0]]+1;
        tc[1] = (tc0_table+52)[index_a][bS[1]]+1;
        tc[2] = (tc0_table+52)[index_a][bS[2]]+1;
        tc[3] = (tc0_table+52)[index_a][bS[3]]+1;
        h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
    }
}

    if( IS_INTRA(mb_type) ) {
        int16_t bS4[4] = {4,4,4,4};
        int16_t bS3[4] = {3,3,3,3};
        int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
        if( IS_8x8DCT(mb_type) ) {
            filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );
            filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );
            filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 );
            filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );
        } else {
            filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );
            filter_mb_edgev( h, &img_y[4*1], linesize, bS3, qp );
            filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );
            filter_mb_edgev( h, &img_y[4*3], linesize, bS3, qp );
            filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 );
            filter_mb_edgeh( h, &img_y[4*1*linesize], linesize, bS3, qp );
            filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );
            filter_mb_edgeh( h, &img_y[4*3*linesize], linesize, bS3, qp );
        }
        filter_mb_edgecv( h, &img_cb[2*0], uvlinesize, bS4, qpc0 );
        filter_mb_edgecv( h, &img_cb[2*2], uvlinesize, bS3, qpc );
        filter_mb_edgecv( h, &img_cr[2*0], uvlinesize, bS4, qpc0 );
        filter_mb_edgecv( h, &img_cr[2*2], uvlinesize, bS3, qpc );
        filter_mb_edgech( h, &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1 );
        filter_mb_edgech( h, &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc );
        filter_mb_edgech( h, &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1 );
        filter_mb_edgech( h, &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc );
        return;
    } else {
        DECLARE_ALIGNED_8(int16_t, bS[2][4][4]);
        uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
        int edges;

        if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) )
            bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL;

#define FILTER(hv,dir,edge)\
        if(bSv[dir][edge]) {\
            filter_mb_edge##hv( h, &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir );\
            if(!(edge&1)) {\
                filter_mb_edgec##hv( h, &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\
                filter_mb_edgec##hv( h, &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\
            }\
        }
        if( edges == 1 ) {
            FILTER(v,0,0);
            FILTER(h,1,0);

            // Do not use s->qscale as luma quantizer because it has not the same
            // value in IPCM macroblocks.
            qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
            tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
            { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
            filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp );
            filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
                              ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
            filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
                              ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
        }

        start = 1;
    }

        qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
        //tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp[0], s->current_picture.qscale_table[mbn_xy]);
        tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
        //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
        if( dir == 0 ) {
            filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
            if( (edge&1) == 0 ) {
                int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
                filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, qp);
                if(h->pps.chroma_qp_diff)
                    qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
                filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, qp);
            }
        } else {
            filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
            if( (edge&1) == 0 ) {
                int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
                filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, qp);
                if(h->pps.chroma_qp_diff)
                    qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
                filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, qp);
            }
        }
    }
}

10960:10759fd39860

    {-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 = 52 + qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = (beta_table+52)[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
        tc[0] = tc0_table[index_a][bS[0]];
        tc[1] = tc0_table[index_a][bS[1]];
        tc[2] = tc0_table[index_a][bS[2]];
        tc[3] = tc0_table[index_a][bS[3]];
        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 = 52 + qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = (beta_table+52)[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
        tc[0] = tc0_table[index_a][bS[0]]+1;
        tc[1] = tc0_table[index_a][bS[1]]+1;
        tc[2] = tc0_table[index_a][bS[2]]+1;
        tc[3] = tc0_table[index_a][bS[3]]+1;
        h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
    }
}

            }
        }
    }
}

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 = 52 + qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = (beta_table+52)[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
        tc[0] = tc0_table[index_a][bS[0]];
        tc[1] = tc0_table[index_a][bS[1]];
        tc[2] = tc0_table[index_a][bS[2]];
        tc[3] = tc0_table[index_a][bS[3]];
        h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
    } 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 = 52 + qp + h->slice_alpha_c0_offset;
    const int alpha = alpha_table[index_a];
    const int beta  = (beta_table+52)[qp + h->slice_beta_offset];
    if (alpha ==0 || beta == 0) return;

    if( bS[0] < 4 ) {
        int8_t tc[4];
        tc[0] = tc0_table[index_a][bS[0]]+1;
        tc[1] = tc0_table[index_a][bS[1]]+1;
        tc[2] = tc0_table[index_a][bS[2]]+1;
        tc[3] = tc0_table[index_a][bS[3]]+1;
        h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
    } else {
        h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
    }
}

    if( IS_INTRA(mb_type) ) {
        int16_t bS4[4] = {4,4,4,4};
        int16_t bS3[4] = {3,3,3,3};
        int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
        if( IS_8x8DCT(mb_type) ) {
            filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
            filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
            filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
            filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
        } else {
            filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
            filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, h);
            filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
            filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, h);
            filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
            filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, h);
            filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
            filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h);
        }
        filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
        filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
        filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
        filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
        filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
        filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
        filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
        filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
        return;
    } else {
        DECLARE_ALIGNED_8(int16_t, bS[2][4][4]);
        uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
        int edges;

        if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) )
            bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL;

#define FILTER(hv,dir,edge)\
        if(bSv[dir][edge]) {\
            filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
            if(!(edge&1)) {\
                filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
                filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
            }\
        }
        if( edges == 1 ) {
            FILTER(v,0,0);
            FILTER(h,1,0);

            // Do not use s->qscale as luma quantizer because it has not the same
            // value in IPCM macroblocks.
            qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
            tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
            { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
            filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
            filter_mb_edgech( &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
                              ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
            filter_mb_edgech( &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
                              ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
        }

        start = 1;
    }

        qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
        //tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp[0], s->current_picture.qscale_table[mbn_xy]);
        tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
        //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
        if( dir == 0 ) {
            filter_mb_edgev( &img_y[4*edge], linesize, bS, qp, h );
            if( (edge&1) == 0 ) {
                int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
                filter_mb_edgecv( &img_cb[2*edge], uvlinesize, bS, qp, h);
                if(h->pps.chroma_qp_diff)
                    qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
                filter_mb_edgecv( &img_cr[2*edge], uvlinesize, bS, qp, h);
            }
        } else {
            filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
            if( (edge&1) == 0 ) {
                int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
                filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, qp, h);
                if(h->pps.chroma_qp_diff)
                    qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
                filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, qp, h);
            }
        }
    }
}