Hello dear all,
I am working on the HD image processing using CUDA. I have a 3750*3750 image, and I have troubles to initialize an array of this dimension.
-----------------------------------------------------------------------------------------------
__device__ int *B;
__global__ void initialize(int *tab, int v, int s)
{
int k = blockDim.x*blockIdx.x + threadIdx.x ;
if ( k < s )
tab[k] = v;
}
__global__ void parentKernel(int sizeImage)
{
B = (int*)malloc(dim*dim*sizeof(int));
int threadsPerBlock = 256;
int blocksPerGrid = (sizeImage + threadsPerBlock -1) / threadsPerBlock;
initialize(B,0,sizeImage);
cudaDeviceSynchronize();
}
-----------------------------------------------------------------------------------------
Thank you
You are right, it was a lack of memory that I solved with this line.
cudaDeviceSetLimit(cudaLimitMallocHeapSize, 128*1024*1024);
Better ask this kind of very programmatic and non-research question on StackOverflow.
Also:
- You do not give any error message or warning;
- You do not say anything about your hardware (do you even have enough memory on your graphic card to allocate an array of this size?)
Good luck!
http://stackoverflow.com/
I don't have problem with programmation because my program works very well with other sizes of image. I need to know if there is a specific method to deal with HD images rather than what is basically used.
Maybe I should ask this question differently. Thank you for the remark
Have you pinpointed the exact image size at which your program fails? Is it the malloc call that is failing? If so, maybe you lack memory on the GPU. In that case, you have two choices:
Finally, just to be sure we understand each other, your program working on other image size is definitely not a proof that there is no programming problem. In my experience, assuming this could lock your investigation on bad leads and make you loose much time debugging "elsewhere".
Thank you very much, dear Bruno. I have an nvidia geforce 840M, it means that I have 5.0 computing capability (2^31 - 1 blocks per grid ). For this reason, I asked if there are other methods specific to HD images.
You are right, it was a lack of memory that I solved with this line.
cudaDeviceSetLimit(cudaLimitMallocHeapSize, 128*1024*1024);
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing