About cudaMemcpy()

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,
I am very new to cuda programming. I am now using CUDA8.0 within Visual Studio 2015 and the GPU is Geforce 1060.
As I know the global memory of 1060 is 6GB. However, when I tried to copy the memory from my host to the device by using cudaMemcpy(), such as code below:
cudaMalloc((void **)&d_index_array, 10000 * 3 * sizeof(int)); //copy the index array from host to device;
cudaMalloc((void **)&d_hist, 10000sizeof(float));
cudaMalloc((void **)&d_img,10000sizeof(float));

cudaMemcpy(d_hist, &hist, 10000*sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_img, &img, 10000*sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_index_array, &index_array, 10000 * 3*sizeof(int), cudaMemcpyHostToDevice);

I will encounter an error as below:
Exception thrown at 0x00007FFCF04090D0 (nvcuda.dll) in xxxx.exe: 0xC0000005: Access violation reading location 0x0000002153120000.

If there is a handler for this exception, the program may be safely continued.

Has anyone encountered this type of problem before? I have searched the issue but it does not seem that I could find any useful solutions. Any advice will be highly appreciated!!Many thanks!

2

These seem unlikely to be correct:

cudaMemcpy(d_hist, &hist, 10000*sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_img, &img, 10000*sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(d_index_array, &index_array, 10000 * 3*sizeof(int), cudaMemcpyHostToDevice);

It’s impossible to say without a more complete code, showing the actual allocations for hist, img, and index_array, but it’s unlikely that taking the address of those is the correct thing to do, e.g.:

&hist

3

Thanks so much. It works. Yes, i made a simple mistake. Thanks for pointing out this.