利用CUDA加速opencv cvtColor

__global__ void rgb_to_ycrcb(uchar3 *input,uchar3 *out,int h,int w,float *y)
{
	unsigned int x=blockDim.x*blockIdx.x+threadIdx.x;
	unsigned int y=blockDim.y*blockIdx.y+threadIdx.y;
	if(x<w&&y<h)
	{
		uchar3 rgb=input[y*w+x];
		 yy[y*w+x]=(16  + 0.257 * rgb.x + 0.504 * rgb.y + 0.098 * rgb.z)/255.0;
        if(yy[y*w+x]<0) yy[y*w+x]=0;
        if(yy[y*w+x]>1) yy[y*w+x]=1;  //防止溢出
        out[y*w+x].x=yy[y*w+x];
        out[y*w+x].y=(128 - 0.148 * rgb.x - 0.291 * rgb.y + 0.439 * rgb.z);
        out[y*w+x].z=(128 + 0.439 * rgb.x - 0.368  * rgb.y - 0.071 * rgb.z);
	}
}

__global__ void ycrcb_to_rgb(uchar3 *in,uchar3 *out,int h,int w)
{
	unsigned int x=blockDim.x*blockIdx.x+threadIdx.x;
	unsigned int y=blockDim.y*blockIdx.y+threadIdx.y;
	if(x<w&&y<h)
	{
		uchar3 yuv = in[y*w+x]
		float rr,gg,bb;
        rr=(1.164 *(yuv.x - 16) + 1.596 *(yuv.z- 128));
        if(rr<0) rr=0;  if(rr>255) rr=255;
        gg=(1.164 *(yuv.x - 16) - 0.392 *(yuv.y - 128) - 0.812 *(yuv.z - 128));
        if(gg<0) gg=0;  if(gg>255) gg=255;
        bb=(1.164 *(yuv.x - 16) + 2.016 *(yuv.y - 128));
        if(bb<0) bb=0;  if(bb>255) bb=255;
		//防止溢出
        out[y*w+x].x=rr;
        out[y*w+x].y=gg;
        out[y*w+x].z=bb;
	}
}

//grid block 大小(超分前)
    dim3 block(32,32);
    dim3 grid((block.x + w - 1)/ block.x,
              (block.y + h - 1)/block.y);
//grid block 大小(超分后)
	dim3 block(32,32);
    dim3 grid((block.x + w*2 - 1)/ block.x,
              (block.y + h*2 - 1)/block.y);