macroblock.c

法国人的264代码 大家看看啊,里面有很多可以借鉴的东西啊

                h->mb.i_cbp_luma |= 1 << i;
        }
    }
    else
    {
        for( i = 0; i < 16; i++ )
        {
            const int nz = array_non_zero_count( h->dct.block[i].luma4x4, 16 );
            h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
            if( nz > 0 )
                h->mb.i_cbp_luma |= 1 << (i/4);
        }
    }

    /* Calculate the chroma patern */
    h->mb.i_cbp_chroma = 0x00;
    for( i = 0; i < 8; i++ )
    {
        const int nz = array_non_zero_count( h->dct.block[16+i].residual_ac, 15 );
        h->mb.cache.non_zero_count[x264_scan8[16+i]] = nz;
        if( nz > 0 )
        {
            h->mb.i_cbp_chroma = 0x02;    /* dc+ac (we can't do only ac) */
        }
    }
    if( h->mb.i_cbp_chroma == 0x00 && array_non_zero( h->dct.chroma_dc[0], 8 ) )
    {
        h->mb.i_cbp_chroma = 0x01;    /* dc only */
    }

    if( h->param.b_cabac )
    {
        i_cbp_dc = ( h->mb.i_type == I_16x16 && array_non_zero( h->dct.luma16x16_dc, 16 ) )
                 | array_non_zero( h->dct.chroma_dc[0], 4 ) << 1
                 | array_non_zero( h->dct.chroma_dc[1], 4 ) << 2;
    }

    /* store cbp */
    h->mb.cbp[h->mb.i_mb_xy] = (i_cbp_dc << 8) | (h->mb.i_cbp_chroma << 4) | h->mb.i_cbp_luma;

    /* Check for P_SKIP
     * XXX: in the me perhaps we should take x264_mb_predict_mv_pskip into account
     *      (if multiple mv give same result)*/
    if( h->mb.i_type == P_L0 && h->mb.i_partition == D_16x16 &&
        h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma== 0x00 &&
        h->mb.cache.ref[0][x264_scan8[0]] == 0 )
    {
        int mvp[2];

        x264_mb_predict_mv_pskip( h, mvp );
        if( h->mb.cache.mv[0][x264_scan8[0]][0] == mvp[0] &&
            h->mb.cache.mv[0][x264_scan8[0]][1] == mvp[1] )
        {
            h->mb.i_type = P_SKIP;
        }
    }

    /* Check for B_SKIP */
    if( h->mb.i_type == B_DIRECT &&
        h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma== 0x00 )
    {
        h->mb.i_type = B_SKIP;
    }
}

/*****************************************************************************
 * x264_macroblock_probe_skip:
 *  Check if the current MB could be encoded as a [PB]_SKIP (it supposes you use
 *  the previous QP
 *****************************************************************************/
int x264_macroblock_probe_skip( x264_t *h, int b_bidir )
{
    DECLARE_ALIGNED( int16_t, dct4x4[16][4][4], 16 );
    DECLARE_ALIGNED( int16_t, dct2x2[2][2], 16 );
    DECLARE_ALIGNED( int,     dctscan[16], 16 );

    int i_qp = h->mb.i_qp;
    int mvp[2];
    int ch;

    int i8x8, i4x4;
    int i_decimate_mb;

    if( !b_bidir )
    {
        /* Get the MV */
        x264_mb_predict_mv_pskip( h, mvp );
        mvp[0] = x264_clip3( mvp[0], h->mb.mv_min[0], h->mb.mv_max[0] );
        mvp[1] = x264_clip3( mvp[1], h->mb.mv_min[1], h->mb.mv_max[1] );

        /* Motion compensation */
        h->mc.mc_luma( h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
                       h->mb.pic.p_fdec[0],    FDEC_STRIDE,
                       mvp[0], mvp[1], 16, 16 );
    }

    /* get luma diff */
    h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0],
                                  h->mb.pic.p_fdec[0] );

    for( i8x8 = 0, i_decimate_mb = 0; i8x8 < 4; i8x8++ )
    {
        /* encode one 4x4 block */
        for( i4x4 = 0; i4x4 < 4; i4x4++ )
        {
            const int idx = i8x8 * 4 + i4x4;

            quant_4x4( h, dct4x4[idx], (int(*)[4][4])def_quant4_mf, i_qp, 0 );
            scan_zigzag_4x4full( dctscan, dct4x4[idx] );

            i_decimate_mb += x264_mb_decimate_score( dctscan, 16 );

            if( i_decimate_mb >= 6 )
            {
                /* not as P_SKIP */
                return 0;
            }
        }
    }

    /* encode chroma */
    i_qp = i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )];

    for( ch = 0; ch < 2; ch++ )
    {
        uint8_t  *p_src = h->mb.pic.p_fenc[1+ch];
        uint8_t  *p_dst = h->mb.pic.p_fdec[1+ch];

        if( !b_bidir )
        {
            h->mc.mc_chroma( h->mb.pic.p_fref[0][0][4+ch], h->mb.pic.i_stride[1+ch],
                             h->mb.pic.p_fdec[1+ch],       FDEC_STRIDE,
                             mvp[0], mvp[1], 8, 8 );
        }

        h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );

        /* calculate dct DC */
        dct2x2[0][0] = dct4x4[0][0][0];
        dct2x2[0][1] = dct4x4[1][0][0];
        dct2x2[1][0] = dct4x4[2][0][0];
        dct2x2[1][1] = dct4x4[3][0][0];
        h->dctf.dct2x2dc( dct2x2 );
        quant_2x2_dc( h, dct2x2, (int(*)[4][4])def_quant4_mf, i_qp, 0 );
        if( dct2x2[0][0] || dct2x2[0][1] || dct2x2[1][0] || dct2x2[1][1]  )
        {
            /* can't be */
            return 0;
        }

        /* calculate dct coeffs */
        for( i4x4 = 0, i_decimate_mb = 0; i4x4 < 4; i4x4++ )
        {
            quant_4x4( h, dct4x4[i4x4], (int(*)[4][4])def_quant4_mf, i_qp, 0 );
            scan_zigzag_4x4( dctscan, dct4x4[i4x4] );

            i_decimate_mb += x264_mb_decimate_score( dctscan, 15 );
            if( i_decimate_mb >= 7 )
            {
                return 0;
            }
        }
    }

    return 1;
}

/****************************************************************************
 * DCT-domain noise reduction / adaptive deadzone
 * from libavcodec
 ****************************************************************************/

void x264_noise_reduction_update( x264_t *h )
{
    int cat, i;
    for( cat = 0; cat < 2; cat++ )
    {
        int size = cat ? 64 : 16;
        const int *weight = cat ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;

        if( h->nr_count[cat] > (cat ? (1<<16) : (1<<18)) )
        {
            for( i = 0; i < size; i++ )
                h->nr_residual_sum[cat][i] >>= 1;
            h->nr_count[cat] >>= 1;
        }

        for( i = 0; i < size; i++ )
            h->nr_offset[cat][i] =
                ((uint64_t)h->param.analyse.i_noise_reduction * h->nr_count[cat]
                 + h->nr_residual_sum[cat][i]/2)
              / ((uint64_t)h->nr_residual_sum[cat][i] * weight[i]/256 + 1);
    }
}

void x264_denoise_dct( x264_t *h, int16_t *dct )
{
    const int cat = h->mb.b_transform_8x8;
    int i;

    h->nr_count[cat]++;

    for( i = (cat ? 63 : 15); i >= 1; i-- )
    {
        int level = dct[i];
        if( level )
        {
            if( level > 0 )
            {
                h->nr_residual_sum[cat][i] += level;
                level -= h->nr_offset[cat][i];
                if( level < 0 )
                    level = 0;
            }
            else
            {
                h->nr_residual_sum[cat][i] -= level;
                level += h->nr_offset[cat][i];
                if( level > 0 )
                    level = 0;
            }
            dct[i] = level;
        }
    }
}

/*****************************************************************************
 * RD only; 4 calls to this do not make up for one macroblock_encode.
 * doesn't transform chroma dc.
 *****************************************************************************/
void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
{
    int i_qp = h->mb.i_qp;
    uint8_t *p_fenc = h->mb.pic.p_fenc[0] + (i8&1)*8 + (i8>>1)*8*FENC_STRIDE;
    uint8_t *p_fdec = h->mb.pic.p_fdec[0] + (i8&1)*8 + (i8>>1)*8*FDEC_STRIDE;
    int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
    int nnz8x8;
    int ch;

    x264_mb_mc_8x8( h, i8 );

    if( h->mb.b_transform_8x8 )
    {
        int16_t dct8x8[8][8];
        h->dctf.sub8x8_dct8( dct8x8, p_fenc, p_fdec );
        quant_8x8( h, dct8x8, h->quant8_mf[CQM_8PY], i_qp, 0 );
        scan_zigzag_8x8full( h->dct.luma8x8[i8], dct8x8 );

        if( b_decimate )
            nnz8x8 = 4 <= x264_mb_decimate_score( h->dct.luma8x8[i8], 64 );
        else
            nnz8x8 = array_non_zero( (int*)dct8x8, sizeof(dct8x8)/sizeof(int) );

        if( nnz8x8 )
        {
            h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8PY], i_qp );
            h->dctf.add8x8_idct8( p_fdec, dct8x8 );
        }
    }
    else
    {
        int i4;
        int16_t dct4x4[4][4][4];
        h->dctf.sub8x8_dct( dct4x4, p_fenc, p_fdec );
        quant_4x4( h, dct4x4[0], h->quant4_mf[CQM_4PY], i_qp, 0 );
        quant_4x4( h, dct4x4[1], h->quant4_mf[CQM_4PY], i_qp, 0 );
        quant_4x4( h, dct4x4[2], h->quant4_mf[CQM_4PY], i_qp, 0 );
        quant_4x4( h, dct4x4[3], h->quant4_mf[CQM_4PY], i_qp, 0 );
        for( i4 = 0; i4 < 4; i4++ )
            scan_zigzag_4x4full( h->dct.block[i8*4+i4].luma4x4, dct4x4[i4] );

        if( b_decimate )
        {
            int i_decimate_8x8 = 0;
            for( i4 = 0; i4 < 4 && i_decimate_8x8 < 4; i4++ )
                i_decimate_8x8 += x264_mb_decimate_score( h->dct.block[i8*4+i4].luma4x4, 16 );
            nnz8x8 = 4 <= i_decimate_8x8;
        }
        else
            nnz8x8 = array_non_zero( (int*)dct4x4, sizeof(dct4x4)/sizeof(int) );

        if( nnz8x8 )
        {
            for( i4 = 0; i4 < 4; i4++ )
                h->quantf.dequant_4x4( dct4x4[i4], h->dequant4_mf[CQM_4PY], i_qp );
            h->dctf.add8x8_idct( p_fdec, dct4x4 );
        }
    }

    i_qp = i_chroma_qp_table[x264_clip3( i_qp + h->pps->i_chroma_qp_index_offset, 0, 51 )];

    for( ch = 0; ch < 2; ch++ )
    {
        int16_t dct4x4[4][4];
        p_fenc = h->mb.pic.p_fenc[1+ch] + (i8&1)*4 + (i8>>1)*4*FENC_STRIDE;
        p_fdec = h->mb.pic.p_fdec[1+ch] + (i8&1)*4 + (i8>>1)*4*FDEC_STRIDE;

        h->dctf.sub4x4_dct( dct4x4, p_fenc, p_fdec );
        quant_4x4( h, dct4x4, h->quant4_mf[CQM_4PC], i_qp, 0 );
        scan_zigzag_4x4( h->dct.block[16+i8+ch*4].residual_ac, dct4x4 );
        if( array_non_zero( (int*)dct4x4, sizeof(dct4x4)/sizeof(int) ) )
        {
            h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PC], i_qp );
            h->dctf.add4x4_idct( p_fdec, dct4x4 );
        }
    }

    if( nnz8x8 )
        h->mb.i_cbp_luma |= (1 << i8);
    else
        h->mb.i_cbp_luma &= ~(1 << i8);
    h->mb.i_cbp_chroma = 0x02;
}