frame.c

在dsp上实现h.264编解码

static void deblock_v_chroma_c( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
{   
    deblock_chroma_c( pix, stride, 1, alpha, beta, tc0 );
}
static void deblock_h_chroma_c( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
{   
    deblock_chroma_c( pix, 1, stride, alpha, beta, tc0 );
}

static inline void deblock_luma_intra_c( uint8_t *pix, int xstride, int ystride, int alpha, int beta )
{
    int d;
    for( d = 0; d < 16; d++ ) {
        const int p2 = pix[-3*xstride];
        const int p1 = pix[-2*xstride];
        const int p0 = pix[-1*xstride];
        const int q0 = pix[ 0*xstride];
        const int q1 = pix[ 1*xstride];
        const int q2 = pix[ 2*xstride];

        if( abs( p0 - q0 ) < alpha &&
            abs( p1 - p0 ) < beta &&
            abs( q1 - q0 ) < beta ) {

            if(abs( p0 - q0 ) < ((alpha >> 2) + 2) ){
                if( abs( p2 - p0 ) < beta)
                {
                    const int p3 = pix[-4*xstride];
                    /* p0', p1', p2' */
                    pix[-1*xstride] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
                    pix[-2*xstride] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
                    pix[-3*xstride] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
                } else {
                    /* p0' */
                    pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
                }
                if( abs( q2 - q0 ) < beta)
                {
                    const int q3 = pix[3*xstride];
                    /* q0', q1', q2' */
                    pix[0*xstride] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
                    pix[1*xstride] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
                    pix[2*xstride] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
                } else {
                    /* q0' */
                    pix[0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
                }
            }else{
                /* p0', q0' */
                pix[-1*xstride] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
                pix[ 0*xstride] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
            }
        }
        pix += ystride;
    }
}
static void deblock_v_luma_intra_c( uint8_t *pix, int stride, int alpha, int beta )
{   
    deblock_luma_intra_c( pix, stride, 1, alpha, beta );
}
static void deblock_h_luma_intra_c( uint8_t *pix, int stride, int alpha, int beta )
{   
    deblock_luma_intra_c( pix, 1, stride, alpha, beta );
}

static inline void deblock_chroma_intra_c( uint8_t *pix, int xstride, int ystride, int alpha, int beta )
{   
    int d; 
    for( d = 0; d < 8; d++ ) {
        const int p1 = pix[-2*xstride];
        const int p0 = pix[-1*xstride];
        const int q0 = pix[ 0*xstride];
        const int q1 = pix[ 1*xstride];

        if( abs( p0 - q0 ) < alpha &&
            abs( p1 - p0 ) < beta &&
            abs( q1 - q0 ) < beta ) {

            pix[-1*xstride] = (2*p1 + p0 + q1 + 2) >> 2;   /* p0' */
            pix[ 0*xstride] = (2*q1 + q0 + p1 + 2) >> 2;   /* q0' */
        }

        pix += ystride;
    }
}
static void deblock_v_chroma_intra_c( uint8_t *pix, int stride, int alpha, int beta )
{   
    deblock_chroma_intra_c( pix, stride, 1, alpha, beta );
}
static void deblock_h_chroma_intra_c( uint8_t *pix, int stride, int alpha, int beta )
{   
    deblock_chroma_intra_c( pix, 1, stride, alpha, beta );
}

static inline void deblock_edge( SHU264_t *h, uint8_t *pix, int i_stride, int bS[4], int i_qp, int b_chroma,
                                 SHU264_deblock_inter_t pf_inter, SHU264_deblock_intra_t pf_intra )
{
    int i;
    const int index_a = SHU264_clip3( i_qp + h->sh.i_alpha_c0_offset, 0, 51 );
    const int alpha = i_alpha_table[index_a];
    const int beta  = i_beta_table[SHU264_clip3( i_qp + h->sh.i_beta_offset, 0, 51 )];

    if( bS[0] < 4 ) {
        int8_t tc[4]; 
        for(i=0; i<4; i++)
            tc[i] = (bS[i] ? i_tc0_table[index_a][bS[i] - 1] : -1) + b_chroma;
        pf_inter( pix, i_stride, alpha, beta, tc );
    } else {
        pf_intra( pix, i_stride, alpha, beta );
    }
}

void SHU264_frame_deblocking_filter( SHU264_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;
        const int b_8x8_transform = h->mb.mb_transform_size[mb_xy];

        /* cavlc + 8x8 transform stores nnz per 16 coeffs for the purpose of
         * entropy coding, but per 64 coeffs for the purpose of deblocking */
        if( !h->param.b_cabac && b_8x8_transform )
        {
            uint32_t *nnz = (uint32_t*)h->mb.non_zero_count[mb_xy];
            if( nnz[0] ) nnz[0] = 0x01010101;
            if( nnz[1] ) nnz[1] = 0x01010101;
            if( nnz[2] ) nnz[2] = 0x01010101;
            if( nnz[3] ) nnz[3] = 0x01010101;
        }

        /* 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 */
                    if( !b_8x8_transform || !(i_edge & 1) )
                    {
                        deblock_edge( 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, 0,
                                      h->loopf.deblock_h_luma, h->loopf.deblock_h_luma_intra );
                    }
                    if( !(i_edge & 1) )
                    {
                        /* U/V planes */
                        int i_qpc = ( i_chroma_qp_table[SHU264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] +
                                      i_chroma_qp_table[SHU264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1;
                        deblock_edge( h, &h->fdec->plane[1][8*(mb_y*h->fdec->i_stride[1]+mb_x)+2*i_edge],
                                      h->fdec->i_stride[1], bS, i_qpc, 1,
                                      h->loopf.deblock_h_chroma, h->loopf.deblock_h_chroma_intra );
                        deblock_edge( h, &h->fdec->plane[2][8*(mb_y*h->fdec->i_stride[2]+mb_x)+2*i_edge],
                                      h->fdec->i_stride[2], bS, i_qpc, 1,
                                      h->loopf.deblock_h_chroma, h->loopf.deblock_h_chroma_intra );
                    }
                }
                else
                {
                    /* horizontal edge */
                    if( !b_8x8_transform || !(i_edge & 1) )
                    {
                        deblock_edge( 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, 0,
                                      h->loopf.deblock_v_luma, h->loopf.deblock_v_luma_intra );
                    }
                    /* U/V planes */
                    if( !(i_edge & 1) )
                    {
                        int i_qpc = ( i_chroma_qp_table[SHU264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )] +
                                      i_chroma_qp_table[SHU264_clip3( i_qpn + h->pps->i_chroma_qp_index_offset, 0, 51 )] + 1 ) >> 1;
                        deblock_edge( h, &h->fdec->plane[1][8*(mb_y*h->fdec->i_stride[1]+mb_x)+2*i_edge*h->fdec->i_stride[1]],
                                      h->fdec->i_stride[1], bS, i_qpc, 1,
                                      h->loopf.deblock_v_chroma, h->loopf.deblock_v_chroma_intra );
                        deblock_edge( h, &h->fdec->plane[2][8*(mb_y*h->fdec->i_stride[2]+mb_x)+2*i_edge*h->fdec->i_stride[2]],
                                      h->fdec->i_stride[2], bS, i_qpc, 1,
                                      h->loopf.deblock_v_chroma, h->loopf.deblock_v_chroma_intra );
                    }
                }
            }
        }

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

#ifdef HAVE_MMXEXT
void SHU264_deblock_v_chroma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
void SHU264_deblock_h_chroma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
void SHU264_deblock_v_chroma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta );
void SHU264_deblock_h_chroma_intra_mmxext( uint8_t *pix, int stride, int alpha, int beta );
#endif

#ifdef ARCH_X86_64
void SHU264_deblock_v_luma_sse2( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
void SHU264_deblock_h_luma_sse2( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
#elif defined( HAVE_MMXEXT )
void SHU264_deblock_h_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );
void SHU264_deblock_v8_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 );

void SHU264_deblock_v_luma_mmxext( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
{
    SHU264_deblock_v8_luma_mmxext( pix,   stride, alpha, beta, tc0   );
    SHU264_deblock_v8_luma_mmxext( pix+8, stride, alpha, beta, tc0+2 );
}
#endif

void SHU264_deblock_init( int cpu, SHU264_deblock_function_t *pf )
{
    pf->deblock_v_luma = deblock_v_luma_c;
    pf->deblock_h_luma = deblock_h_luma_c;
    pf->deblock_v_chroma = deblock_v_chroma_c;
    pf->deblock_h_chroma = deblock_h_chroma_c;
    pf->deblock_v_luma_intra = deblock_v_luma_intra_c;
    pf->deblock_h_luma_intra = deblock_h_luma_intra_c;
    pf->deblock_v_chroma_intra = deblock_v_chroma_intra_c;
    pf->deblock_h_chroma_intra = deblock_h_chroma_intra_c;

#ifdef HAVE_MMXEXT
    if( cpu&SHU264_CPU_MMXEXT )
    {
        pf->deblock_v_chroma = SHU264_deblock_v_chroma_mmxext;
        pf->deblock_h_chroma = SHU264_deblock_h_chroma_mmxext;
        pf->deblock_v_chroma_intra = SHU264_deblock_v_chroma_intra_mmxext;
        pf->deblock_h_chroma_intra = SHU264_deblock_h_chroma_intra_mmxext;

#ifdef ARCH_X86_64
        if( cpu&SHU264_CPU_SSE2 )
        {
            pf->deblock_v_luma = SHU264_deblock_v_luma_sse2;
            pf->deblock_h_luma = SHU264_deblock_h_luma_sse2;
        }
#else
        pf->deblock_v_luma = SHU264_deblock_v_luma_mmxext;
        pf->deblock_h_luma = SHU264_deblock_h_luma_mmxext;
#endif
    }
#endif
}