frame.c.svn-base

H.264 source codes

                        pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
                    }
                }
                pix += i_pix_stride;
            }
        }
    }
}

static inline void deblocking_filter_edgecv( x264_t *h, uint8_t *pix, int i_pix_stride, int bS[4], int i_QP )
{
    int i, d;
    const int i_index_a = x264_clip3( i_QP + h->sh.i_alpha_c0_offset, 0, 51 );
    const int alpha = i_alpha_table[i_index_a];
    const int beta  = i_beta_table[x264_clip3( i_QP + h->sh.i_beta_offset, 0, 51 )];

    for( i = 0; i < 4; i++ )
    {
        if( bS[i] == 0 )
        {
            pix += 2 * i_pix_stride;
            continue;
        }

        if( bS[i] < 4 )
        {
            const int tc = i_tc0_table[i_index_a][bS[i] - 1] + 1;
            /* 2px edge length (because we use same bS than the one for luma) */
            for( d = 0; d < 2; d++ )
            {
                const int p0 = pix[-1];
                const int p1 = pix[-2];
                const int q0 = pix[0];
                const int q1 = pix[1];

                if( abs( p0 - q0 ) < alpha &&
                    abs( p1 - p0 ) < beta &&
                    abs( q1 - q0 ) < beta )
                {
                    const int i_delta = x264_clip3( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );

                    pix[-1] = clip_uint8( p0 + i_delta );    /* p0' */
                    pix[0]  = clip_uint8( q0 - i_delta );    /* q0' */
                }
                pix += i_pix_stride;
            }
        }
        else
        {
            /* 2px edge length (because we use same bS than the one for luma) */
            for( d = 0; d < 2; d++ )
            {
                const int p0 = pix[-1];
                const int p1 = pix[-2];
                const int q0 = pix[0];
                const int q1 = pix[1];

                if( abs( p0 - q0 ) < alpha &&
                    abs( p1 - p0 ) < beta &&
                    abs( q1 - q0 ) < beta )
                {
                    pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */
                    pix[0]  = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */
                }
                pix += i_pix_stride;
            }
        }
    }
}

static inline void deblocking_filter_edgeh( x264_t *h, uint8_t *pix, int i_pix_stride, int bS[4], int i_QP )
{
    int i, d;
    const int i_index_a = x264_clip3( i_QP + h->sh.i_alpha_c0_offset, 0, 51 );
    const int alpha = i_alpha_table[i_index_a];
    const int beta  = i_beta_table[x264_clip3( i_QP + h->sh.i_beta_offset, 0, 51 )];

    int i_pix_next  = i_pix_stride;

    for( i = 0; i < 4; i++ )
    {
        if( bS[i] == 0 )
        {
            pix += 4;
            continue;
        }

        if( bS[i] < 4 )
        {
            const int tc0 = i_tc0_table[i_index_a][bS[i] - 1];
            /* 4px edge length */
            for( d = 0; d < 4; d++ )
            {
                const int p0 = pix[-i_pix_next];
                const int p1 = pix[-2*i_pix_next];
                const int p2 = pix[-3*i_pix_next];
                const int q0 = pix[0];
                const int q1 = pix[1*i_pix_next];
                const int q2 = pix[2*i_pix_next];

                if( abs( p0 - q0 ) < alpha &&
                    abs( p1 - p0 ) < beta &&
                    abs( q1 - q0 ) < beta )
                {
                    int tc = tc0;
                    int i_delta;

                    if( abs( p2 - p0 ) < beta )
                    {
                        pix[-2*i_pix_next] = p1 + x264_clip3( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
                        tc++;
                    }
                    if( abs( q2 - q0 ) < beta )
                    {
                        pix[i_pix_next] = q1 + x264_clip3( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
                        tc++;
                    }

                    i_delta = x264_clip3( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
                    pix[-i_pix_next] = clip_uint8( p0 + i_delta );    /* p0' */
                    pix[0]           = clip_uint8( q0 - i_delta );    /* q0' */
                }
                pix++;
            }
        }
        else
        {
            /* 4px edge length */
            for( d = 0; d < 4; d++ )
            {
                const int p0 = pix[-i_pix_next];
                const int p1 = pix[-2*i_pix_next];
                const int p2 = pix[-3*i_pix_next];
                const int q0 = pix[0];
                const int q1 = pix[1*i_pix_next];
                const int q2 = pix[2*i_pix_next];

                if( abs( p0 - q0 ) < alpha &&
                    abs( p1 - p0 ) < beta &&
                    abs( q1 - q0 ) < beta )
                {
                    const int p3 = pix[-4*i_pix_next];
                    const int q3 = pix[ 3*i_pix_next];

                    if( abs( p0 - q0 ) < (( alpha >> 2 ) + 2 ) )
                    {
                        if( abs( p2 - p0 ) < beta )
                        {
                            /* p0', p1', p2' */
                            pix[-1*i_pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
                            pix[-2*i_pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
                            pix[-3*i_pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
                        }
                        else
                        {
                            /* p0' */
                            pix[-1*i_pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
                        }
                        if( abs( q2 - q0 ) < beta )
                        {
                            /* q0', q1', q2' */
                            pix[0*i_pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
                            pix[1*i_pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
                            pix[2*i_pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
                        }
                        else
                        {
                            /* q0' */
                            pix[0*i_pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
                        }
                    }
                    else
                    {
                        /* p0' */
                        pix[-1*i_pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
                        /* q0' */
                        pix[0*i_pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
                    }
                }
                pix++;
            }

        }
    }
}

static inline void deblocking_filter_edgech( x264_t *h, uint8_t *pix, int i_pix_stride, int bS[4], int i_QP )
{
    int i, d;
    const int i_index_a = x264_clip3( i_QP + h->sh.i_alpha_c0_offset, 0, 51 );
    const int alpha = i_alpha_table[i_index_a];
    const int beta  = i_beta_table[x264_clip3( i_QP + h->sh.i_beta_offset, 0, 51 )];

    int i_pix_next  = i_pix_stride;

    for( i = 0; i < 4; i++ )
    {
        if( bS[i] == 0 )
        {
            pix += 2;
            continue;
        }
        if( bS[i] < 4 )
        {
            int tc = i_tc0_table[i_index_a][bS[i] - 1] + 1;
            /* 2px edge length (see deblocking_filter_edgecv) */
            for( d = 0; d < 2; d++ )
            {
                const int p0 = pix[-1*i_pix_next];
                const int p1 = pix[-2*i_pix_next];
                const int q0 = pix[0];
                const int q1 = pix[1*i_pix_next];

                if( abs( p0 - q0 ) < alpha &&
                    abs( p1 - p0 ) < beta &&
                    abs( q1 - q0 ) < beta )
                {
                    int i_delta = x264_clip3( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );

                    pix[-i_pix_next] = clip_uint8( p0 + i_delta );    /* p0' */
                    pix[0]           = clip_uint8( q0 - i_delta );    /* q0' */
                }
                pix++;
            }
        }
        else
        {
            /* 2px edge length (see deblocking_filter_edgecv) */
            for( d = 0; d < 2; d++ )
            {
                const int p0 = pix[-1*i_pix_next];
                const int p1 = pix[-2*i_pix_next];
                const int q0 = pix[0];
                const int q1 = pix[1*i_pix_next];

                if( abs( p0 - q0 ) < alpha &&
                    abs( p1 - p0 ) < beta &&
                    abs( q1 - q0 ) < beta )
                {
                    pix[-i_pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */
                    pix[0]           = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */
                }
                pix++;
            }
        }
    }
}

void x264_frame_deblocking_filter( x264_t *h, int i_slice_type )
{
    const int s8x8 = 2 * h->mb.i_mb_stride;
    const int s4x4 = 4 * h->mb.i_mb_stride;

    int mb_y, mb_x;

    for( mb_y = 0, mb_x = 0; mb_y < h->sps->i_mb_height; )
    {
        const int mb_xy  = mb_y * h->mb.i_mb_stride + mb_x;
        const int mb_8x8 = 2 * s8x8 * mb_y + 2 * mb_x;
        const int mb_4x4 = 4 * s4x4 * mb_y + 4 * mb_x;
        int i_edge;
        int i_dir;

        /* i_dir == 0 -> vertical edge
         * i_dir == 1 -> horizontal edge */
        for( i_dir = 0; i_dir < 2; i_dir++ )
        {
            int i_start;
            int i_qp, i_qpn;

            i_start = (( i_dir == 0 && mb_x != 0 ) || ( i_dir == 1 && mb_y != 0 ) ) ? 0 : 1;

            for( i_edge = i_start; i_edge < 4; i_edge++ )
            {
                int mbn_xy  = i_edge > 0 ? mb_xy  : ( i_dir == 0 ? mb_xy  - 1 : mb_xy - h->mb.i_mb_stride );
                int mbn_8x8 = i_edge > 0 ? mb_8x8 : ( i_dir == 0 ? mb_8x8 - 2 : mb_8x8 - 2 * s8x8 );
                int mbn_4x4 = i_edge > 0 ? mb_4x4 : ( i_dir == 0 ? mb_4x4 - 4 : mb_4x4 - 4 * s4x4 );

                int bS[4];  /* filtering strength */

                /* *** Get bS for each 4px for the current edge *** */
                if( IS_INTRA( h->mb.type[mb_xy] ) || IS_INTRA( h->mb.type[mbn_xy] ) )
                {
                    bS[0] = bS[1] = bS[2] = bS[3] = ( i_edge == 0 ? 4 : 3 );
                }
                else
                {
                    int i;
                    for( i = 0; i < 4; i++ )
                    {
                        int x  = i_dir == 0 ? i_edge : i;
                        int y  = i_dir == 0 ? i      : i_edge;
                        int xn = (x - (i_dir == 0 ? 1 : 0 ))&0x03;
                        int yn = (y - (i_dir == 0 ? 0 : 1 ))&0x03;

                        if( h->mb.non_zero_count[mb_xy][block_idx_xy[x][y]] != 0 ||
                            h->mb.non_zero_count[mbn_xy][block_idx_xy[xn][yn]] != 0 )
                        {
                            bS[i] = 2;
                        }
                        else
                        {
                            /* FIXME: A given frame may occupy more than one position in
                             * the reference list. So we should compare the frame numbers,
                             * not the indices in the ref list.
                             * No harm yet, as we don't generate that case.*/

                            int i8p= mb_8x8+(x/2)+(y/2)*s8x8;
                            int i8q= mbn_8x8+(xn/2)+(yn/2)*s8x8;
                            int i4p= mb_4x4+x+y*s4x4;
                            int i4q= mbn_4x4+xn+yn*s4x4;
                            int l;

                            bS[i] = 0;

                            for( l = 0; l < 1 + (i_slice_type == SLICE_TYPE_B); l++ )
                            {
                                if( h->mb.ref[l][i8p] != h->mb.ref[l][i8q] ||
                                    abs( h->mb.mv[l][i4p][0] - h->mb.mv[l][i4q][0] ) >= 4 ||
                                    abs( h->mb.mv[l][i4p][1] - h->mb.mv[l][i4q][1] ) >= 4 )
                                {
                                    bS[i] = 1;
                                    break;
                                }
                            }
                        }
                    }
                }

                /* *** filter *** */
                /* Y plane */
                i_qp = h->mb.qp[mb_xy];
                i_qpn= h->mb.qp[mbn_xy];

                if( i_dir == 0 )
                {
                    /* vertical edge */
                    deblocking_filter_edgev( h, &h->fdec->plane[0][16 * mb_y * h->fdec->i_stride[0]+ 16 * mb_x + 4 * i_edge],
                                                h->fdec->i_stride[0], bS, (i_qp+i_qpn+1) >> 1);
                    if( (i_edge % 2) == 0  )
                    {
                        /* U/V planes */
                        int i_qpc = ( i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] +
                                      i_chroma_qp_table[x264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1;
                        deblocking_filter_edgecv( h, &h->fdec->plane[1][8*(mb_y*h->fdec->i_stride[1]+mb_x)+i_edge*2],
                                                      h->fdec->i_stride[1], bS, i_qpc );
                        deblocking_filter_edgecv( h, &h->fdec->plane[2][8*(mb_y*h->fdec->i_stride[2]+mb_x)+i_edge*2],
                                                  h->fdec->i_stride[2], bS, i_qpc );
                    }
                }
                else
                {
                    /* horizontal edge */
                    deblocking_filter_edgeh( h, &h->fdec->plane[0][(16*mb_y + 4 * i_edge) * h->fdec->i_stride[0]+ 16 * mb_x],
                                                h->fdec->i_stride[0], bS, (i_qp+i_qpn+1) >> 1 );
                    /* U/V planes */
                    if( ( i_edge % 2  ) == 0 )
                    {
                        int i_qpc = ( i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] +
                                      i_chroma_qp_table[x264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1;
                        deblocking_filter_edgech( h, &h->fdec->plane[1][8*(mb_y*h->fdec->i_stride[1]+mb_x)+i_edge*2*h->fdec->i_stride[1]],
                                                 h->fdec->i_stride[1], bS, i_qpc );
                        deblocking_filter_edgech( h, &h->fdec->plane[2][8*(mb_y*h->fdec->i_stride[2]+mb_x)+i_edge*2*h->fdec->i_stride[2]],
                                                 h->fdec->i_stride[2], bS, i_qpc );
                    }
                }
            }
        }

        /* newt mb */
        mb_x++;
        if( mb_x >= h->sps->i_mb_width )
        {
            mb_x = 0;
            mb_y++;
        }
    }
}