comparison mpegvideo.c @ 481:29a7e17d19dd libavcodec

optimizing MPV_decode_mb

480:a7cbee351b55

/* add block[] to dest[] */
static inline void add_dct(MpegEncContext *s, 
                           DCTELEM *block, int i, UINT8 *dest, int line_size)
{
    /* skip dequant / idct if we are really late ;) */
    if(s->hurry_up>1) return;

    if (s->block_last_index[i] >= 0) {
        if (!s->mpeg2)
            if(s->encoding || (!s->h263_msmpeg4))
                s->dct_unquantize(s, block, i, s->qscale);

        ff_idct_add (dest, line_size, block);
    }
}

   s->interlaced_dct : true if interlaced dct used (mpeg2)
 */
void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
{
    int mb_x, mb_y;
    int dct_linesize, dct_offset;
    op_pixels_func *op_pix;
    qpel_mc_func *op_qpix;

    mb_x = s->mb_x;
    mb_y = s->mb_y;

#ifdef FF_POSTPROCESS

#endif

    /* update DC predictors for P macroblocks */
    if (!s->mb_intra) {
        if (s->h263_pred || s->h263_aic) {
            if(s->mbintra_table[mb_x + mb_y*s->mb_width])
                ff_clean_intra_table_entries(s);
        } else {
            s->last_dc[0] =
            s->last_dc[1] =
            s->last_dc[2] = 128 << s->intra_dc_precision;
        }
    }
    else if (s->h263_pred || s->h263_aic)
        s->mbintra_table[mb_x + mb_y*s->mb_width]=1;

    /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */
    if (s->out_format == FMT_H263) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here
      if(s->pict_type!=B_TYPE){
        int xy, wrap, motion_x, motion_y;
        
        wrap = 2 * s->mb_width + 2;
        xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap;
        if (s->mb_intra) {
            motion_x = 0;
            motion_y = 0;
            goto motion_init;
        } else if (s->mv_type == MV_TYPE_16X16) {

            s->motion_val[xy + wrap][0] = motion_x;
            s->motion_val[xy + wrap][1] = motion_y;
            s->motion_val[xy + 1 + wrap][0] = motion_x;
            s->motion_val[xy + 1 + wrap][1] = motion_y;
        }
      }
    }
    
    if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) {
        UINT8 *dest_y, *dest_cb, *dest_cr;
        UINT8 *mbskip_ptr;

        /* avoid copy if macroblock skipped in last frame too 
           dont touch it for B-frames as they need the skip info from the next p-frame */
        if (s->pict_type != B_TYPE) {
            mbskip_ptr = &s->mbskip_table[s->mb_y * s->mb_width + s->mb_x];
            if (s->mb_skiped) {
                s->mb_skiped = 0;
                /* if previous was skipped too, then nothing to do ! 
                   skip only during decoding as we might trash the buffers during encoding a bit */
                if (*mbskip_ptr != 0 && !s->encoding) 

            dct_linesize = s->linesize;
            dct_offset = s->linesize * 8;
        }

        if (!s->mb_intra) {
            const int xy= s->mb_y * s->mb_width + s->mb_x;
            /* motion handling */
            /* decoding or more than one mb_type (MC was allready done otherwise) */
            if((!s->encoding) || (s->mb_type[xy]&(s->mb_type[xy]-1))){
                if ((!s->no_rounding) || s->pict_type==B_TYPE){                
                    op_pix = put_pixels_tab;
                    op_qpix= qpel_mc_rnd_tab;
                }else{
                    op_pix = put_no_rnd_pixels_tab;

                if (s->mv_dir & MV_DIR_BACKWARD) {
                    MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
                }
            }

            /* add dct residue */
            add_dct(s, block[0], 0, dest_y, dct_linesize);
            add_dct(s, block[1], 1, dest_y + 8, dct_linesize);
            add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
            add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);

            add_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
            add_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
        } else {
            /* dct only in intra block */
            put_dct(s, block[0], 0, dest_y, dct_linesize);
            put_dct(s, block[1], 1, dest_y + 8, dct_linesize);
            put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);

481:29a7e17d19dd

/* add block[] to dest[] */
static inline void add_dct(MpegEncContext *s, 
                           DCTELEM *block, int i, UINT8 *dest, int line_size)
{
    if (s->block_last_index[i] >= 0) {
        ff_idct_add (dest, line_size, block);
    }
}

static inline void add_dequant_dct(MpegEncContext *s, 
                           DCTELEM *block, int i, UINT8 *dest, int line_size)
{
    if (s->block_last_index[i] >= 0) {
        s->dct_unquantize(s, block, i, s->qscale);

        ff_idct_add (dest, line_size, block);
    }
}

   s->interlaced_dct : true if interlaced dct used (mpeg2)
 */
void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
{
    int mb_x, mb_y;
    const int mb_xy = s->mb_y * s->mb_width + s->mb_x;

    mb_x = s->mb_x;
    mb_y = s->mb_y;

#ifdef FF_POSTPROCESS

#endif

    /* update DC predictors for P macroblocks */
    if (!s->mb_intra) {
        if (s->h263_pred || s->h263_aic) {
            if(s->mbintra_table[mb_xy])
                ff_clean_intra_table_entries(s);
        } else {
            s->last_dc[0] =
            s->last_dc[1] =
            s->last_dc[2] = 128 << s->intra_dc_precision;
        }
    }
    else if (s->h263_pred || s->h263_aic)
        s->mbintra_table[mb_xy]=1;

    /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */
    if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here
        int motion_x, motion_y;
        
        const int wrap = s->block_wrap[0];
        const int xy = s->block_index[0];
        if (s->mb_intra) {
            motion_x = 0;
            motion_y = 0;
            goto motion_init;
        } else if (s->mv_type == MV_TYPE_16X16) {

            s->motion_val[xy + wrap][0] = motion_x;
            s->motion_val[xy + wrap][1] = motion_y;
            s->motion_val[xy + 1 + wrap][0] = motion_x;
            s->motion_val[xy + 1 + wrap][1] = motion_y;
        }
    }
    
    if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) {
        UINT8 *dest_y, *dest_cb, *dest_cr;
        int dct_linesize, dct_offset;
        op_pixels_func *op_pix;
        qpel_mc_func *op_qpix;

        /* avoid copy if macroblock skipped in last frame too 
           dont touch it for B-frames as they need the skip info from the next p-frame */
        if (s->pict_type != B_TYPE) {
            UINT8 *mbskip_ptr = &s->mbskip_table[mb_xy];
            if (s->mb_skiped) {
                s->mb_skiped = 0;
                /* if previous was skipped too, then nothing to do ! 
                   skip only during decoding as we might trash the buffers during encoding a bit */
                if (*mbskip_ptr != 0 && !s->encoding) 

            dct_linesize = s->linesize;
            dct_offset = s->linesize * 8;
        }

        if (!s->mb_intra) {
            /* motion handling */
            /* decoding or more than one mb_type (MC was allready done otherwise) */
            if((!s->encoding) || (s->mb_type[mb_xy]&(s->mb_type[mb_xy]-1))){
                if ((!s->no_rounding) || s->pict_type==B_TYPE){                
                    op_pix = put_pixels_tab;
                    op_qpix= qpel_mc_rnd_tab;
                }else{
                    op_pix = put_no_rnd_pixels_tab;

                if (s->mv_dir & MV_DIR_BACKWARD) {
                    MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
                }
            }

            /* skip dequant / idct if we are really late ;) */
            if(s->hurry_up>1) goto the_end;

            /* add dct residue */
            if(!s->mpeg2 && (s->encoding || (!s->h263_msmpeg4))){
                add_dequant_dct(s, block[0], 0, dest_y, dct_linesize);
                add_dequant_dct(s, block[1], 1, dest_y + 8, dct_linesize);
                add_dequant_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
                add_dequant_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);

                add_dequant_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
                add_dequant_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
            } else {
                add_dct(s, block[0], 0, dest_y, dct_linesize);
                add_dct(s, block[1], 1, dest_y + 8, dct_linesize);
                add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
                add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);

                add_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
                add_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
            }
        } else {
            /* dct only in intra block */
            put_dct(s, block[0], 0, dest_y, dct_linesize);
            put_dct(s, block[1], 1, dest_y + 8, dct_linesize);
            put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);