vc1.c

mediastreamer2是开源的网络传输媒体流的库

    DSPContext *dsp = &v->s.dsp;

    if(v->rangeredfrm) {
        int i, j, k;
        for(k = 0; k < 6; k++)
            for(j = 0; j < 8; j++)
                for(i = 0; i < 8; i++)
                    block[k][i + j*8] = ((block[k][i + j*8] - 128) << 1) + 128;

    }
    ys = v->s.current_picture.linesize[0];
    us = v->s.current_picture.linesize[1];
    vs = v->s.current_picture.linesize[2];
    Y = v->s.dest[0];

    dsp->put_pixels_clamped(block[0], Y, ys);
    dsp->put_pixels_clamped(block[1], Y + 8, ys);
    Y += ys * 8;
    dsp->put_pixels_clamped(block[2], Y, ys);
    dsp->put_pixels_clamped(block[3], Y + 8, ys);

    if(!(v->s.flags & CODEC_FLAG_GRAY)) {
        dsp->put_pixels_clamped(block[4], v->s.dest[1], us);
        dsp->put_pixels_clamped(block[5], v->s.dest[2], vs);
    }
}

/** Do motion compensation over 1 macroblock
 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
 */
static void vc1_mc_1mv(VC1Context *v, int dir)
{
    MpegEncContext *s = &v->s;
    DSPContext *dsp = &v->s.dsp;
    uint8_t *srcY, *srcU, *srcV;
    int dxy, uvdxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;

    if(!v->s.last_picture.data[0])return;

    mx = s->mv[dir][0][0];
    my = s->mv[dir][0][1];

    // store motion vectors for further use in B frames
    if(s->pict_type == P_TYPE) {
        s->current_picture.motion_val[1][s->block_index[0]][0] = mx;
        s->current_picture.motion_val[1][s->block_index[0]][1] = my;
    }
    uvmx = (mx + ((mx & 3) == 3)) >> 1;
    uvmy = (my + ((my & 3) == 3)) >> 1;
    if(v->fastuvmc) {
        uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
        uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
    }
    if(!dir) {
        srcY = s->last_picture.data[0];
        srcU = s->last_picture.data[1];
        srcV = s->last_picture.data[2];
    } else {
        srcY = s->next_picture.data[0];
        srcU = s->next_picture.data[1];
        srcV = s->next_picture.data[2];
    }

    src_x = s->mb_x * 16 + (mx >> 2);
    src_y = s->mb_y * 16 + (my >> 2);
    uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
    uvsrc_y = s->mb_y * 8 + (uvmy >> 2);

    if(v->profile != PROFILE_ADVANCED){
        src_x   = av_clip(  src_x, -16, s->mb_width  * 16);
        src_y   = av_clip(  src_y, -16, s->mb_height * 16);
        uvsrc_x = av_clip(uvsrc_x,  -8, s->mb_width  *  8);
        uvsrc_y = av_clip(uvsrc_y,  -8, s->mb_height *  8);
    }else{
        src_x   = av_clip(  src_x, -17, s->avctx->coded_width);
        src_y   = av_clip(  src_y, -18, s->avctx->coded_height + 1);
        uvsrc_x = av_clip(uvsrc_x,  -8, s->avctx->coded_width  >> 1);
        uvsrc_y = av_clip(uvsrc_y,  -8, s->avctx->coded_height >> 1);
    }

    srcY += src_y * s->linesize + src_x;
    srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
    srcV += uvsrc_y * s->uvlinesize + uvsrc_x;

    /* for grayscale we should not try to read from unknown area */
    if(s->flags & CODEC_FLAG_GRAY) {
        srcU = s->edge_emu_buffer + 18 * s->linesize;
        srcV = s->edge_emu_buffer + 18 * s->linesize;
    }

    if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
       || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel*3
       || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 16 - s->mspel*3){
        uint8_t *uvbuf= s->edge_emu_buffer + 19 * s->linesize;

        srcY -= s->mspel * (1 + s->linesize);
        ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 17+s->mspel*2, 17+s->mspel*2,
                            src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
        srcY = s->edge_emu_buffer;
        ff_emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, 8+1, 8+1,
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
        ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8+1, 8+1,
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
        srcU = uvbuf;
        srcV = uvbuf + 16;
        /* if we deal with range reduction we need to scale source blocks */
        if(v->rangeredfrm) {
            int i, j;
            uint8_t *src, *src2;

            src = srcY;
            for(j = 0; j < 17 + s->mspel*2; j++) {
                for(i = 0; i < 17 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
                src += s->linesize;
            }
            src = srcU; src2 = srcV;
            for(j = 0; j < 9; j++) {
                for(i = 0; i < 9; i++) {
                    src[i] = ((src[i] - 128) >> 1) + 128;
                    src2[i] = ((src2[i] - 128) >> 1) + 128;
                }
                src += s->uvlinesize;
                src2 += s->uvlinesize;
            }
        }
        /* if we deal with intensity compensation we need to scale source blocks */
        if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
            int i, j;
            uint8_t *src, *src2;

            src = srcY;
            for(j = 0; j < 17 + s->mspel*2; j++) {
                for(i = 0; i < 17 + s->mspel*2; i++) src[i] = v->luty[src[i]];
                src += s->linesize;
            }
            src = srcU; src2 = srcV;
            for(j = 0; j < 9; j++) {
                for(i = 0; i < 9; i++) {
                    src[i] = v->lutuv[src[i]];
                    src2[i] = v->lutuv[src2[i]];
                }
                src += s->uvlinesize;
                src2 += s->uvlinesize;
            }
        }
        srcY += s->mspel * (1 + s->linesize);
    }

    if(s->mspel) {
        dxy = ((my & 3) << 2) | (mx & 3);
        dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0]    , srcY    , s->linesize, v->rnd);
        dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8, srcY + 8, s->linesize, v->rnd);
        srcY += s->linesize * 8;
        dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize    , srcY    , s->linesize, v->rnd);
        dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0] + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
    } else { // hpel mc - always used for luma
        dxy = (my & 2) | ((mx & 2) >> 1);

        if(!v->rnd)
            dsp->put_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
        else
            dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
    }

    if(s->flags & CODEC_FLAG_GRAY) return;
    /* Chroma MC always uses qpel bilinear */
    uvdxy = ((uvmy & 3) << 2) | (uvmx & 3);
    uvmx = (uvmx&3)<<1;
    uvmy = (uvmy&3)<<1;
    if(!v->rnd){
        dsp->put_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
        dsp->put_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
    }else{
        dsp->put_no_rnd_h264_chroma_pixels_tab[0](s->dest[1], srcU, s->uvlinesize, 8, uvmx, uvmy);
        dsp->put_no_rnd_h264_chroma_pixels_tab[0](s->dest[2], srcV, s->uvlinesize, 8, uvmx, uvmy);
    }
}

/** Do motion compensation for 4-MV macroblock - luminance block
 */
static void vc1_mc_4mv_luma(VC1Context *v, int n)
{
    MpegEncContext *s = &v->s;
    DSPContext *dsp = &v->s.dsp;
    uint8_t *srcY;
    int dxy, mx, my, src_x, src_y;
    int off;

    if(!v->s.last_picture.data[0])return;
    mx = s->mv[0][n][0];
    my = s->mv[0][n][1];
    srcY = s->last_picture.data[0];

    off = s->linesize * 4 * (n&2) + (n&1) * 8;

    src_x = s->mb_x * 16 + (n&1) * 8 + (mx >> 2);
    src_y = s->mb_y * 16 + (n&2) * 4 + (my >> 2);

    if(v->profile != PROFILE_ADVANCED){
        src_x   = av_clip(  src_x, -16, s->mb_width  * 16);
        src_y   = av_clip(  src_y, -16, s->mb_height * 16);
    }else{
        src_x   = av_clip(  src_x, -17, s->avctx->coded_width);
        src_y   = av_clip(  src_y, -18, s->avctx->coded_height + 1);
    }

    srcY += src_y * s->linesize + src_x;

    if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
       || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 8 - s->mspel*2
       || (unsigned)(src_y - s->mspel) > s->v_edge_pos - (my&3) - 8 - s->mspel*2){
        srcY -= s->mspel * (1 + s->linesize);
        ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, 9+s->mspel*2, 9+s->mspel*2,
                            src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, s->v_edge_pos);
        srcY = s->edge_emu_buffer;
        /* if we deal with range reduction we need to scale source blocks */
        if(v->rangeredfrm) {
            int i, j;
            uint8_t *src;

            src = srcY;
            for(j = 0; j < 9 + s->mspel*2; j++) {
                for(i = 0; i < 9 + s->mspel*2; i++) src[i] = ((src[i] - 128) >> 1) + 128;
                src += s->linesize;
            }
        }
        /* if we deal with intensity compensation we need to scale source blocks */
        if(v->mv_mode == MV_PMODE_INTENSITY_COMP) {
            int i, j;
            uint8_t *src;

            src = srcY;
            for(j = 0; j < 9 + s->mspel*2; j++) {
                for(i = 0; i < 9 + s->mspel*2; i++) src[i] = v->luty[src[i]];
                src += s->linesize;
            }
        }
        srcY += s->mspel * (1 + s->linesize);
    }

    if(s->mspel) {
        dxy = ((my & 3) << 2) | (mx & 3);
        dsp->put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, v->rnd);
    } else { // hpel mc - always used for luma
        dxy = (my & 2) | ((mx & 2) >> 1);
        if(!v->rnd)
            dsp->put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
        else
            dsp->put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
    }
}

static inline int median4(int a, int b, int c, int d)
{
    if(a < b) {
        if(c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
        else      return (FFMIN(b, c) + FFMAX(a, d)) / 2;
    } else {
        if(c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
        else      return (FFMIN(a, c) + FFMAX(b, d)) / 2;
    }
}


/** Do motion compensation for 4-MV macroblock - both chroma blocks
 */
static void vc1_mc_4mv_chroma(VC1Context *v)
{
    MpegEncContext *s = &v->s;
    DSPContext *dsp = &v->s.dsp;
    uint8_t *srcU, *srcV;
    int uvdxy, uvmx, uvmy, uvsrc_x, uvsrc_y;
    int i, idx, tx = 0, ty = 0;
    int mvx[4], mvy[4], intra[4];
    static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};

    if(!v->s.last_picture.data[0])return;
    if(s->flags & CODEC_FLAG_GRAY) return;

    for(i = 0; i < 4; i++) {
        mvx[i] = s->mv[0][i][0];
        mvy[i] = s->mv[0][i][1];
        intra[i] = v->mb_type[0][s->block_index[i]];
    }

    /* calculate chroma MV vector from four luma MVs */
    idx = (intra[3] << 3) | (intra[2] << 2) | (intra[1] << 1) | intra[0];
    if(!idx) { // all blocks are inter
        tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
        ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
    } else if(count[idx] == 1) { // 3 inter blocks
        switch(idx) {
        case 0x1:
            tx = mid_pred(mvx[1], mvx[2], mvx[3]);
            ty = mid_pred(mvy[1], mvy[2], mvy[3]);
            break;
        case 0x2:
            tx = mid_pred(mvx[0], mvx[2], mvx[3]);
            ty = mid_pred(mvy[0], mvy[2], mvy[3]);
            break;
        case 0x4:
            tx = mid_pred(mvx[0], mvx[1], mvx[3]);
            ty = mid_pred(mvy[0], mvy[1], mvy[3]);
            break;
        case 0x8:
            tx = mid_pred(mvx[0], mvx[1], mvx[2]);
            ty = mid_pred(mvy[0], mvy[1], mvy[2]);
            break;
        }
    } else if(count[idx] == 2) {
        int t1 = 0, t2 = 0;
        for(i=0; i<3;i++) if(!intra[i]) {t1 = i; break;}
        for(i= t1+1; i<4; i++)if(!intra[i]) {t2 = i; break;}
        tx = (mvx[t1] + mvx[t2]) / 2;
        ty = (mvy[t1] + mvy[t2]) / 2;
    } else {
        s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
        s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
        return; //no need to do MC for inter blocks
    }

    s->current_picture.motion_val[1][s->block_index[0]][0] = tx;
    s->current_picture.motion_val[1][s->block_index[0]][1] = ty;
    uvmx = (tx + ((tx&3) == 3)) >> 1;
    uvmy = (ty + ((ty&3) == 3)) >> 1;
    if(v->fastuvmc) {
        uvmx = uvmx + ((uvmx<0)?(uvmx&1):-(uvmx&1));
        uvmy = uvmy + ((uvmy<0)?(uvmy&1):-(uvmy&1));
    }

    uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
    uvsrc_y = s->mb_y * 8 + (uvmy >> 2);

    if(v->profile != PROFILE_ADVANCED){
        uvsrc_x = av_clip(uvsrc_x,  -8, s->mb_width  *  8);
        uvsrc_y = av_clip(uvsrc_y,  -8, s->mb_height *  8);
    }else{
        uvsrc_x = av_clip(uvsrc_x,  -8, s->avctx->coded_width  >> 1);
        uvsrc_y = av_clip(uvsrc_y,  -8, s->avctx->coded_height >> 1);
    }

    srcU = s->last_picture.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
    srcV = s->last_picture.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
    if(v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
       || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
       || (unsigned)uvsrc_y > (s->v_edge_pos >> 1) - 9){
        ff_emulated_edge_mc(s->edge_emu_buffer     , srcU, s->uvlinesize, 8+1, 8+1,
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
        ff_emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize, 8+1, 8+1,
                            uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
        srcU = s->edge_emu_buffer;
        srcV = s->edge_emu_buffer + 16;

        /* if we deal with range reduction we need to scale source blocks */
        if(v->rangeredfrm) {
            int i, j;
            uint8_t *src, *src2;

            src = srcU; src2 = srcV;