comparison mpegvideo.c @ 3216:4186eb8d29a1 libavcodec

prefetch pixels for future motion compensation. 4-10% faster mpeg1/2/4 decoding (on an athlon-xp).

3215:06f98047ff26

        ptr= s->edge_emu_buffer;
    }
    pix_op[lowres](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
}

/**
 * motion compensation of a single macroblock
 * @param s context
 * @param dest_y luma destination pointer
 * @param dest_cb chroma cb/u destination pointer

    int mb_x, mb_y, i;
    uint8_t *ptr, *dest;

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

    if(s->obmc && s->pict_type != B_TYPE){
        int16_t mv_cache[4][4][2];
        const int xy= s->mb_x + s->mb_y*s->mb_stride;
        const int mot_stride= s->b8_stride;

3216:4186eb8d29a1

        ptr= s->edge_emu_buffer;
    }
    pix_op[lowres](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
}

static inline void prefetch_motion(MpegEncContext *s, uint8_t **pix, int dir){
    /* fetch pixels for estimated mv 4 macroblocks ahead
     * optimized for 64byte cache lines */
    const int shift = s->quarter_sample ? 2 : 1;
    const int mx= (s->mv[dir][0][0]>>shift) + 16*s->mb_x + 8;
    const int my= (s->mv[dir][0][1]>>shift) + 16*s->mb_y;
    int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
    s->dsp.prefetch(pix[0]+off, s->linesize, 4);
    off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
    s->dsp.prefetch(pix[1]+off, pix[2]-pix[1], 2);
}

/**
 * motion compensation of a single macroblock
 * @param s context
 * @param dest_y luma destination pointer
 * @param dest_cb chroma cb/u destination pointer

    int mb_x, mb_y, i;
    uint8_t *ptr, *dest;

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

    prefetch_motion(s, ref_picture, dir);

    if(s->obmc && s->pict_type != B_TYPE){
        int16_t mv_cache[4][4][2];
        const int xy= s->mb_x + s->mb_y*s->mb_stride;
        const int mot_stride= s->b8_stride;