svq1enc.c.svn-base

mediastreamer2是开源的网络传输媒体流的库

        s->m.height= height;
        s->m.mb_width= block_width;
        s->m.mb_height= block_height;
        s->m.mb_stride= s->m.mb_width+1;
        s->m.b8_stride= 2*s->m.mb_width+1;
        s->m.f_code=1;
        s->m.pict_type= s->picture.pict_type;
        s->m.me_method= s->avctx->me_method;
        s->m.me.scene_change_score=0;
        s->m.flags= s->avctx->flags;
//        s->m.out_format = FMT_H263;
//        s->m.unrestricted_mv= 1;

        s->m.lambda= s->picture.quality;
        s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
        s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;

        if(!s->motion_val8[plane]){
            s->motion_val8 [plane]= av_mallocz((s->m.b8_stride*block_height*2 + 2)*2*sizeof(int16_t));
            s->motion_val16[plane]= av_mallocz((s->m.mb_stride*(block_height + 2) + 1)*2*sizeof(int16_t));
        }

        s->m.mb_type= s->mb_type;

        //dummies, to avoid segfaults
        s->m.current_picture.mb_mean=   (uint8_t *)s->dummy;
        s->m.current_picture.mb_var=    (uint16_t*)s->dummy;
        s->m.current_picture.mc_mb_var= (uint16_t*)s->dummy;
        s->m.current_picture.mb_type= s->dummy;

        s->m.current_picture.motion_val[0]= s->motion_val8[plane] + 2;
        s->m.p_mv_table= s->motion_val16[plane] + s->m.mb_stride + 1;
        s->m.dsp= s->dsp; //move
        ff_init_me(&s->m);

        s->m.me.dia_size= s->avctx->dia_size;
        s->m.first_slice_line=1;
        for (y = 0; y < block_height; y++) {
            uint8_t src[stride*16];

            s->m.new_picture.data[0]= src - y*16*stride; //ugly
            s->m.mb_y= y;

            for(i=0; i<16 && i + 16*y<height; i++){
                memcpy(&src[i*stride], &src_plane[(i+16*y)*src_stride], width);
                for(x=width; x<16*block_width; x++)
                    src[i*stride+x]= src[i*stride+x-1];
            }
            for(; i<16 && i + 16*y<16*block_height; i++)
                memcpy(&src[i*stride], &src[(i-1)*stride], 16*block_width);

            for (x = 0; x < block_width; x++) {
                s->m.mb_x= x;
                ff_init_block_index(&s->m);
                ff_update_block_index(&s->m);

                ff_estimate_p_frame_motion(&s->m, x, y);
            }
            s->m.first_slice_line=0;
        }

        ff_fix_long_p_mvs(&s->m);
        ff_fix_long_mvs(&s->m, NULL, 0, s->m.p_mv_table, s->m.f_code, CANDIDATE_MB_TYPE_INTER, 0);
    }

    s->m.first_slice_line=1;
    for (y = 0; y < block_height; y++) {
        uint8_t src[stride*16];

        for(i=0; i<16 && i + 16*y<height; i++){
            memcpy(&src[i*stride], &src_plane[(i+16*y)*src_stride], width);
            for(x=width; x<16*block_width; x++)
                src[i*stride+x]= src[i*stride+x-1];
        }
        for(; i<16 && i + 16*y<16*block_height; i++)
            memcpy(&src[i*stride], &src[(i-1)*stride], 16*block_width);

        s->m.mb_y= y;
        for (x = 0; x < block_width; x++) {
            uint8_t reorder_buffer[3][6][7*32];
            int count[3][6];
            int offset = y * 16 * stride + x * 16;
            uint8_t *decoded= decoded_plane + offset;
            uint8_t *ref= ref_plane + offset;
            int score[4]={0,0,0,0}, best;
            uint8_t temp[16*stride];

            if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < 3000){ //FIXME check size
                av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
                return -1;
            }

            s->m.mb_x= x;
            ff_init_block_index(&s->m);
            ff_update_block_index(&s->m);

            if(s->picture.pict_type == I_TYPE || (s->m.mb_type[x + y*s->m.mb_stride]&CANDIDATE_MB_TYPE_INTRA)){
                for(i=0; i<6; i++){
                    init_put_bits(&s->reorder_pb[i], reorder_buffer[0][i], 7*32);
                }
                if(s->picture.pict_type == P_TYPE){
                    const uint8_t *vlc= ff_svq1_block_type_vlc[SVQ1_BLOCK_INTRA];
                    put_bits(&s->reorder_pb[5], vlc[1], vlc[0]);
                    score[0]= vlc[1]*lambda;
                }
                score[0]+= encode_block(s, src+16*x, NULL, temp, stride, 5, 64, lambda, 1);
                for(i=0; i<6; i++){
                    count[0][i]= put_bits_count(&s->reorder_pb[i]);
                    flush_put_bits(&s->reorder_pb[i]);
                }
            }else
                score[0]= INT_MAX;

            best=0;

            if(s->picture.pict_type == P_TYPE){
                const uint8_t *vlc= ff_svq1_block_type_vlc[SVQ1_BLOCK_INTER];
                int mx, my, pred_x, pred_y, dxy;
                int16_t *motion_ptr;

                motion_ptr= h263_pred_motion(&s->m, 0, 0, &pred_x, &pred_y);
                if(s->m.mb_type[x + y*s->m.mb_stride]&CANDIDATE_MB_TYPE_INTER){
                    for(i=0; i<6; i++)
                        init_put_bits(&s->reorder_pb[i], reorder_buffer[1][i], 7*32);

                    put_bits(&s->reorder_pb[5], vlc[1], vlc[0]);

                    s->m.pb= s->reorder_pb[5];
                    mx= motion_ptr[0];
                    my= motion_ptr[1];
                    assert(mx>=-32 && mx<=31);
                    assert(my>=-32 && my<=31);
                    assert(pred_x>=-32 && pred_x<=31);
                    assert(pred_y>=-32 && pred_y<=31);
                    ff_h263_encode_motion(&s->m, mx - pred_x, 1);
                    ff_h263_encode_motion(&s->m, my - pred_y, 1);
                    s->reorder_pb[5]= s->m.pb;
                    score[1] += lambda*put_bits_count(&s->reorder_pb[5]);

                    dxy= (mx&1) + 2*(my&1);

                    s->dsp.put_pixels_tab[0][dxy](temp+16, ref + (mx>>1) + stride*(my>>1), stride, 16);

                    score[1]+= encode_block(s, src+16*x, temp+16, decoded, stride, 5, 64, lambda, 0);
                    best= score[1] <= score[0];

                    vlc= ff_svq1_block_type_vlc[SVQ1_BLOCK_SKIP];
                    score[2]= s->dsp.sse[0](NULL, src+16*x, ref, stride, 16);
                    score[2]+= vlc[1]*lambda;
                    if(score[2] < score[best] && mx==0 && my==0){
                        best=2;
                        s->dsp.put_pixels_tab[0][0](decoded, ref, stride, 16);
                        for(i=0; i<6; i++){
                            count[2][i]=0;
                        }
                        put_bits(&s->pb, vlc[1], vlc[0]);
                    }
                }

                if(best==1){
                    for(i=0; i<6; i++){
                        count[1][i]= put_bits_count(&s->reorder_pb[i]);
                        flush_put_bits(&s->reorder_pb[i]);
                    }
                }else{
                    motion_ptr[0                 ] = motion_ptr[1                 ]=
                    motion_ptr[2                 ] = motion_ptr[3                 ]=
                    motion_ptr[0+2*s->m.b8_stride] = motion_ptr[1+2*s->m.b8_stride]=
                    motion_ptr[2+2*s->m.b8_stride] = motion_ptr[3+2*s->m.b8_stride]=0;
                }
            }

            s->rd_total += score[best];

            for(i=5; i>=0; i--){
                ff_copy_bits(&s->pb, reorder_buffer[best][i], count[best][i]);
            }
            if(best==0){
                s->dsp.put_pixels_tab[0][0](decoded, temp, stride, 16);
            }
        }
        s->m.first_slice_line=0;
    }
    return 0;
}

static int svq1_encode_init(AVCodecContext *avctx)
{
    SVQ1Context * const s = avctx->priv_data;

    dsputil_init(&s->dsp, avctx);
    avctx->coded_frame= (AVFrame*)&s->picture;

    s->frame_width = avctx->width;
    s->frame_height = avctx->height;

    s->y_block_width = (s->frame_width + 15) / 16;
    s->y_block_height = (s->frame_height + 15) / 16;

    s->c_block_width = (s->frame_width / 4 + 15) / 16;
    s->c_block_height = (s->frame_height / 4 + 15) / 16;

    s->avctx= avctx;
    s->m.avctx= avctx;
    s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t));
    s->m.me.map       = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
    s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
    s->mb_type        = av_mallocz((s->y_block_width+1)*s->y_block_height*sizeof(int16_t));
    s->dummy          = av_mallocz((s->y_block_width+1)*s->y_block_height*sizeof(int32_t));
    h263_encode_init(&s->m); //mv_penalty

    return 0;
}

static int svq1_encode_frame(AVCodecContext *avctx, unsigned char *buf,
    int buf_size, void *data)
{
    SVQ1Context * const s = avctx->priv_data;
    AVFrame *pict = data;
    AVFrame * const p= (AVFrame*)&s->picture;
    AVFrame temp;
    int i;

    if(avctx->pix_fmt != PIX_FMT_YUV410P){
        av_log(avctx, AV_LOG_ERROR, "unsupported pixel format\n");
        return -1;
    }

    if(!s->current_picture.data[0]){
        avctx->get_buffer(avctx, &s->current_picture);
        avctx->get_buffer(avctx, &s->last_picture);
    }

    temp= s->current_picture;
    s->current_picture= s->last_picture;
    s->last_picture= temp;

    init_put_bits(&s->pb, buf, buf_size);

    *p = *pict;
    p->pict_type = avctx->gop_size && avctx->frame_number % avctx->gop_size ? P_TYPE : I_TYPE;
    p->key_frame = p->pict_type == I_TYPE;

    svq1_write_header(s, p->pict_type);
    for(i=0; i<3; i++){
        if(svq1_encode_plane(s, i,
            s->picture.data[i], s->last_picture.data[i], s->current_picture.data[i],
            s->frame_width / (i?4:1), s->frame_height / (i?4:1),
            s->picture.linesize[i], s->current_picture.linesize[i]) < 0)
                return -1;
    }

//    align_put_bits(&s->pb);
    while(put_bits_count(&s->pb) & 31)
        put_bits(&s->pb, 1, 0);

    flush_put_bits(&s->pb);

    return (put_bits_count(&s->pb) / 8);
}

static int svq1_encode_end(AVCodecContext *avctx)
{
    SVQ1Context * const s = avctx->priv_data;
    int i;

    av_log(avctx, AV_LOG_DEBUG, "RD: %f\n", s->rd_total/(double)(avctx->width*avctx->height*avctx->frame_number));

    av_freep(&s->m.me.scratchpad);
    av_freep(&s->m.me.map);
    av_freep(&s->m.me.score_map);
    av_freep(&s->mb_type);
    av_freep(&s->dummy);

    for(i=0; i<3; i++){
        av_freep(&s->motion_val8[i]);
        av_freep(&s->motion_val16[i]);
    }

    return 0;
}


AVCodec svq1_encoder = {
    "svq1",
    CODEC_TYPE_VIDEO,
    CODEC_ID_SVQ1,
    sizeof(SVQ1Context),
    svq1_encode_init,
    svq1_encode_frame,
    svq1_encode_end,
    .pix_fmts= (enum PixelFormat[]){PIX_FMT_YUV410P, -1},
};