void dct_blocks( BLOCK_YUV in, BLOCK_YUV out) {
    
    int u, v, x, y;
    float u_cs, v_cs, pi = 3.1415927f;

    for( u=0 ; u<8 ; u++ ) {
        for( v=0 ; v<8 ; v++ ) {
            for( x=0 ; x<8 ; x++ ) {
                for( y=0 ; y<8 ; y++ ) {
                    u_cs = cosf(((2.0f*FLOAT(x)+1.0f)*FLOAT(u)*pi)/16.0f);
                    if( u==0 ) u_cs = (1.0f/FLOAT(sqrt(2.0f)));
                    v_cs = cosf(((2.0f*FLOAT(y)+1.0f)*FLOAT(v)*pi)/16.0f);
                    if( v==0 ) v_cs = (1.0f/FLOAT(sqrt(2.0f)));
                    out.pixels[u][v].y += 0.25f * in.pixels[x][y].y * u_cs * v_cs;
                    out.pixels[u][v].v += 0.25f * in.pixels[x][y].v * u_cs * v_cs;
                    out.pixels[u][v].u += 0.25f * in.pixels[x][y].u * u_cs * v_cs;

                }
            }
        }
    }
}