mpegvideo_enc.c

ffmpeg的完整源代码和作者自己写的文档。不但有在Linux的工程哦

    ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
    ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);

    if (ENABLE_H261_ENCODER && s->out_format == FMT_H261)
        ff_h261_encode_init(s);
    if (ENABLE_ANY_H263_ENCODER && s->out_format == FMT_H263)
        h263_encode_init(s);
    if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version)
        ff_msmpeg4_encode_init(s);
    if ((ENABLE_MPEG1VIDEO_ENCODER || ENABLE_MPEG2VIDEO_ENCODER)
        && s->out_format == FMT_MPEG1)
        ff_mpeg1_encode_init(s);

    /* init q matrix */
    for(i=0;i<64;i++) {
        int j= s->dsp.idct_permutation[i];
        if(ENABLE_MPEG4_ENCODER && s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
            s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
            s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
        }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
            s->intra_matrix[j] =
            s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
        }else
        { /* mpeg1/2 */
            s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
            s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
        }
        if(s->avctx->intra_matrix)
            s->intra_matrix[j] = s->avctx->intra_matrix[i];
        if(s->avctx->inter_matrix)
            s->inter_matrix[j] = s->avctx->inter_matrix[i];
    }

    /* precompute matrix */
    /* for mjpeg, we do include qscale in the matrix */
    if (s->out_format != FMT_MJPEG) {
        convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
                       s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
        convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
                       s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
    }

    if(ff_rate_control_init(s) < 0)
        return -1;

    return 0;
}

int MPV_encode_end(AVCodecContext *avctx)
{
    MpegEncContext *s = avctx->priv_data;

    ff_rate_control_uninit(s);

    MPV_common_end(s);
    if ((ENABLE_MJPEG_ENCODER || ENABLE_LJPEG_ENCODER) && s->out_format == FMT_MJPEG)
        ff_mjpeg_encode_close(s);

    av_freep(&avctx->extradata);

    return 0;
}

static int get_sae(uint8_t *src, int ref, int stride){
    int x,y;
    int acc=0;

    for(y=0; y<16; y++){
        for(x=0; x<16; x++){
            acc+= FFABS(src[x+y*stride] - ref);
        }
    }

    return acc;
}

static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
    int x, y, w, h;
    int acc=0;

    w= s->width &~15;
    h= s->height&~15;

    for(y=0; y<h; y+=16){
        for(x=0; x<w; x+=16){
            int offset= x + y*stride;
            int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
            int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
            int sae = get_sae(src + offset, mean, stride);

            acc+= sae + 500 < sad;
        }
    }
    return acc;
}


static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
    AVFrame *pic=NULL;
    int64_t pts;
    int i;
    const int encoding_delay= s->max_b_frames;
    int direct=1;

    if(pic_arg){
        pts= pic_arg->pts;
        pic_arg->display_picture_number= s->input_picture_number++;

        if(pts != AV_NOPTS_VALUE){
            if(s->user_specified_pts != AV_NOPTS_VALUE){
                int64_t time= pts;
                int64_t last= s->user_specified_pts;

                if(time <= last){
                    av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
                    return -1;
                }
            }
            s->user_specified_pts= pts;
        }else{
            if(s->user_specified_pts != AV_NOPTS_VALUE){
                s->user_specified_pts=
                pts= s->user_specified_pts + 1;
                av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
            }else{
                pts= pic_arg->display_picture_number;
            }
        }
    }

  if(pic_arg){
    if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
    if(pic_arg->linesize[0] != s->linesize) direct=0;
    if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
    if(pic_arg->linesize[2] != s->uvlinesize) direct=0;

//    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);

    if(direct){
        i= ff_find_unused_picture(s, 1);

        pic= (AVFrame*)&s->picture[i];
        pic->reference= 3;

        for(i=0; i<4; i++){
            pic->data[i]= pic_arg->data[i];
            pic->linesize[i]= pic_arg->linesize[i];
        }
        alloc_picture(s, (Picture*)pic, 1);
    }else{
        i= ff_find_unused_picture(s, 0);

        pic= (AVFrame*)&s->picture[i];
        pic->reference= 3;

        alloc_picture(s, (Picture*)pic, 0);

        if(   pic->data[0] + INPLACE_OFFSET == pic_arg->data[0]
           && pic->data[1] + INPLACE_OFFSET == pic_arg->data[1]
           && pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]){
       // empty
        }else{
            int h_chroma_shift, v_chroma_shift;
            avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);

            for(i=0; i<3; i++){
                int src_stride= pic_arg->linesize[i];
                int dst_stride= i ? s->uvlinesize : s->linesize;
                int h_shift= i ? h_chroma_shift : 0;
                int v_shift= i ? v_chroma_shift : 0;
                int w= s->width >>h_shift;
                int h= s->height>>v_shift;
                uint8_t *src= pic_arg->data[i];
                uint8_t *dst= pic->data[i];

                if(!s->avctx->rc_buffer_size)
                    dst +=INPLACE_OFFSET;

                if(src_stride==dst_stride)
                    memcpy(dst, src, src_stride*h);
                else{
                    while(h--){
                        memcpy(dst, src, w);
                        dst += dst_stride;
                        src += src_stride;
                    }
                }
            }
        }
    }
    copy_picture_attributes(s, pic, pic_arg);
    pic->pts= pts; //we set this here to avoid modifiying pic_arg
  }

    /* shift buffer entries */
    for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
        s->input_picture[i-1]= s->input_picture[i];

    s->input_picture[encoding_delay]= (Picture*)pic;

    return 0;
}

static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
    int x, y, plane;
    int score=0;
    int64_t score64=0;

    for(plane=0; plane<3; plane++){
        const int stride= p->linesize[plane];
        const int bw= plane ? 1 : 2;
        for(y=0; y<s->mb_height*bw; y++){
            for(x=0; x<s->mb_width*bw; x++){
                int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
                int v= s->dsp.frame_skip_cmp[1](s, p->data[plane] + 8*(x + y*stride)+off, ref->data[plane] + 8*(x + y*stride), stride, 8);

                switch(s->avctx->frame_skip_exp){
                    case 0: score= FFMAX(score, v); break;
                    case 1: score+= FFABS(v);break;
                    case 2: score+= v*v;break;
                    case 3: score64+= FFABS(v*v*(int64_t)v);break;
                    case 4: score64+= v*v*(int64_t)(v*v);break;
                }
            }
        }
    }

    if(score) score64= score;

    if(score64 < s->avctx->frame_skip_threshold)
        return 1;
    if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
        return 1;
    return 0;
}

static int estimate_best_b_count(MpegEncContext *s){
    AVCodec *codec= avcodec_find_encoder(s->avctx->codec_id);
    AVCodecContext *c= avcodec_alloc_context();
    AVFrame input[FF_MAX_B_FRAMES+2];
    const int scale= s->avctx->brd_scale;
    int i, j, out_size, p_lambda, b_lambda, lambda2;
    int outbuf_size= s->width * s->height; //FIXME
    uint8_t *outbuf= av_malloc(outbuf_size);
    int64_t best_rd= INT64_MAX;
    int best_b_count= -1;

    assert(scale>=0 && scale <=3);

//    emms_c();
    p_lambda= s->last_lambda_for[P_TYPE]; //s->next_picture_ptr->quality;
    b_lambda= s->last_lambda_for[B_TYPE]; //p_lambda *FFABS(s->avctx->b_quant_factor) + s->avctx->b_quant_offset;
    if(!b_lambda) b_lambda= p_lambda; //FIXME we should do this somewhere else
    lambda2= (b_lambda*b_lambda + (1<<FF_LAMBDA_SHIFT)/2 ) >> FF_LAMBDA_SHIFT;

    c->width = s->width >> scale;
    c->height= s->height>> scale;
    c->flags= CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR | CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
    c->flags|= s->avctx->flags & CODEC_FLAG_QPEL;
    c->mb_decision= s->avctx->mb_decision;
    c->me_cmp= s->avctx->me_cmp;
    c->mb_cmp= s->avctx->mb_cmp;
    c->me_sub_cmp= s->avctx->me_sub_cmp;
    c->pix_fmt = PIX_FMT_YUV420P;
    c->time_base= s->avctx->time_base;
    c->max_b_frames= s->max_b_frames;

    if (avcodec_open(c, codec) < 0)
        return -1;

    for(i=0; i<s->max_b_frames+2; i++){
        int ysize= c->width*c->height;
        int csize= (c->width/2)*(c->height/2);
        Picture pre_input, *pre_input_ptr= i ? s->input_picture[i-1] : s->next_picture_ptr;

        avcodec_get_frame_defaults(&input[i]);
        input[i].data[0]= av_malloc(ysize + 2*csize);
        input[i].data[1]= input[i].data[0] + ysize;
        input[i].data[2]= input[i].data[1] + csize;
        input[i].linesize[0]= c->width;
        input[i].linesize[1]=
        input[i].linesize[2]= c->width/2;

        if(pre_input_ptr && (!i || s->input_picture[i-1])) {
            pre_input= *pre_input_ptr;

            if(pre_input.type != FF_BUFFER_TYPE_SHARED && i) {
                pre_input.data[0]+=INPLACE_OFFSET;
                pre_input.data[1]+=INPLACE_OFFSET;
                pre_input.data[2]+=INPLACE_OFFSET;
            }

            s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0], pre_input.data[0], pre_input.linesize[0], c->width, c->height);
            s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1], pre_input.data[1], pre_input.linesize[1], c->width>>1, c->height>>1);
            s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2], pre_input.data[2], pre_input.linesize[2], c->width>>1, c->height>>1);
        }
    }

    for(j=0; j<s->max_b_frames+1; j++){
        int64_t rd=0;

        if(!s->input_picture[j])
            break;

        c->error[0]= c->error[1]= c->error[2]= 0;

        input[0].pict_type= I_TYPE;
        input[0].quality= 1 * FF_QP2LAMBDA;
        out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[0]);
//        rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;

        for(i=0; i<s->max_b_frames+1; i++){
            int is_p= i % (j+1) == j || i==s->max_b_frames;

            input[i+1].pict_type= is_p ? P_TYPE : B_TYPE;
            input[i+1].quality= is_p ? p_lambda : b_lambda;
            out_size = avcodec_encode_video(c, outbuf, outbuf_size, &input[i+1]);
            rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
        }

        /* get the delayed frames */
        while(out_size){
            out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
            rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
        }

        rd += c->error[0] + c->error[1] + c->error[2];

        if(rd < best_rd){
            best_rd= rd;
            best_b_count= j;
        }
    }

    av_freep(&outbuf);
    avcodec_close(c);
    av_freep(&c);

    for(i=0; i<s->max_b_frames+2; i++){
        av_freep(&input[i].data[0]);
    }