comparison snow.c @ 2197:ffa4fa6bebe9 libavcodec

10l more benchmarking code inline

2196:3eae46c131a3

    int x,y;
    const int qlog= clip(s->qlog + b->qlog, 0, 128);
    int qmul= qexp[qlog&7]<<(qlog>>3);
    int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
    
    if(b->buf == s->spatial_dwt_buffer || s->qlog == LOSSLESS_QLOG){
        qadd= 0;
        qmul= 1<<QEXPSHIFT;
    }

    START_TIMER

    if(1){
        int run;
        int index=0;
        int prev_index=-1;

mca( 0, 0,16)
mca( 8, 0,16)
mca( 0, 8,16)
mca( 8, 8,16)

static void add_xblock(SnowContext *s, DWTELEM *dst, uint8_t *src, uint8_t *obmc, int s_x, int s_y, int b_w, int b_h, int mv_x, int mv_y, int w, int h, int dst_stride, int src_stride, int obmc_stride, int mb_type, int add, int color){
    uint8_t tmp[src_stride*(b_h+5)]; //FIXME move to context to gurantee alignment
    int x,y;

    if(s_x<0){
        obmc -= s_x;

    int obmc_stride= plane_index ? block_size : 2*block_size;
    int ref_stride= s->last_picture.linesize[plane_index];
    uint8_t *ref  = s->last_picture.data[plane_index];
    int w= p->width;
    int h= p->height;
    
if(s->avctx->debug&512){
    for(y=0; y<h; y++){
        for(x=0; x<w; x++){
            if(add) buf[x + y*w]+= 128*256;

}
    for(mb_y=-1; mb_y<=mb_h; mb_y++){
        for(mb_x=-1; mb_x<=mb_w; mb_x++){
            int index= clip(mb_x, 0, mb_w-1) + clip(mb_y, 0, mb_h-1)*mb_stride;

            add_xblock(s, buf, ref, obmc, 
                       block_w*mb_x - block_w/2, 
                       block_w*mb_y - block_w/2,
                       2*block_w, 2*block_w,
                       s->block[index].mx*scale, s->block[index].my*scale,
                       w, h,
                       w, ref_stride, obmc_stride, 
                       s->block[index].type, add, s->block[index].color[plane_index]);

        }
    }
}

static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int bias){
    const int level= b->level;
    const int w= b->width;

2197:ffa4fa6bebe9

    int x,y;
    const int qlog= clip(s->qlog + b->qlog, 0, 128);
    int qmul= qexp[qlog&7]<<(qlog>>3);
    int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
    
    START_TIMER

    if(b->buf == s->spatial_dwt_buffer || s->qlog == LOSSLESS_QLOG){
        qadd= 0;
        qmul= 1<<QEXPSHIFT;
    }

    if(1){
        int run;
        int index=0;
        int prev_index=-1;

mca( 0, 0,16)
mca( 8, 0,16)
mca( 0, 8,16)
mca( 8, 8,16)

static always_inline void add_xblock(SnowContext *s, DWTELEM *dst, uint8_t *src, uint8_t *obmc, int s_x, int s_y, int b_w, int b_h, int mv_x, int mv_y, int w, int h, int dst_stride, int src_stride, int obmc_stride, int mb_type, int add, int color){
    uint8_t tmp[src_stride*(b_h+5)]; //FIXME move to context to gurantee alignment
    int x,y;

    if(s_x<0){
        obmc -= s_x;

    int obmc_stride= plane_index ? block_size : 2*block_size;
    int ref_stride= s->last_picture.linesize[plane_index];
    uint8_t *ref  = s->last_picture.data[plane_index];
    int w= p->width;
    int h= p->height;
    START_TIMER
    
if(s->avctx->debug&512){
    for(y=0; y<h; y++){
        for(x=0; x<w; x++){
            if(add) buf[x + y*w]+= 128*256;

}
    for(mb_y=-1; mb_y<=mb_h; mb_y++){
        for(mb_x=-1; mb_x<=mb_w; mb_x++){
            int index= clip(mb_x, 0, mb_w-1) + clip(mb_y, 0, mb_h-1)*mb_stride;

            START_TIMER
            add_xblock(s, buf, ref, obmc, 
                       block_w*mb_x - block_w/2, 
                       block_w*mb_y - block_w/2,
                       2*block_w, 2*block_w,
                       s->block[index].mx*scale, s->block[index].my*scale,
                       w, h,
                       w, ref_stride, obmc_stride, 
                       s->block[index].type, add, s->block[index].color[plane_index]);

            STOP_TIMER("add_xblock")
        }
    }
    
    STOP_TIMER("predict_plane")
}

static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int bias){
    const int level= b->level;
    const int w= b->width;