Something wrong about using cuda texture processing colorful image

[github2016-yuan]

I have learned your code and write down some code to do some image processing. But my code could get right result and when debugging it I found that my code even can't show the colorful image.I tried a few days but still cannot get the right result image. May I post the code here to get some help for you?
(I delete the code of image processing and just I/O operation remained)
global void Kernel2(float* output, int width, int height)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
int y = blockIdx.y * blockDim.y + threadIdx.y;
// int idx = (y * width + x) * 4;
int idx = y * width + x;

if(x < width && y < height)
{

// float u = x / (float)width;
// float v = y / (float)height;

	 output[idx] = float(tex2D(texRef, x + 0.5,y + 0.5)) ;
	 output[idx+1] = float(tex2D(texRef, x + 0.5 + 1,y + 0.5));
	 output[idx+2] = float(tex2D(texRef,x + 0.5 + 2,y + 0.5));
	 output[idx+3] = float(tex2D(texRef, x + 0.5 + 3, y + 0.5));
}

}
const int width = 512;
const int height = 512;
int main()
{
cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
cudaArray* cuArray;
cudaMallocArray(&cuArray, &channelDesc, width*4, height);

Mat image = imread("lena.jpg");
imshow("source-texture1", image);
cvtColor(image,image,CV_BGR2RGBA);
//convert from uchar to float
image.convertTo(image,CV_32FC4);

cudaMemcpyToArray(cuArray, 0, 0, image.data, width*height*4*sizeof(float), cudaMemcpyHostToDevice);

   //ste texture
texRef.addressMode[0] = cudaAddressModeWrap; 
texRef.addressMode[1] = cudaAddressModeWrap;
texRef.filterMode = cudaFilterModePoint; 
texRef.normalized = false; 

   //bind
cudaBindTextureToArray(texRef, cuArray, channelDesc);



Mat imageOutput = Mat(Size(width, height), CV_32FC4);
float * output = NULL;

cudaMalloc((void**)&output, width * height * sizeof(float)*4);

dim3 dimBlock(16, 16);
dim3 dimGrid((width + dimBlock.x - 1) / dimBlock.x, (height + dimBlock.y - 1) / dimBlock.y);

Kernel2<<<dimGrid,dimBlock>>>(output,width,height);

cudaMemcpy(imageOutput.data, output, 4*height*width*sizeof(float), cudaMemcpyDeviceToHost);

imageOutput.convertTo(imageOutput,CV_8UC4);
cvtColor(imageOutput,imageOutput,CV_RGBA2BGR);
imwrite("texture1.jpg",imageOutput);
imshow("texture1", imageOutput);
waitKey();

cudaFreeArray(cuArray);
cudaFree(output);

}