cudaMemcpy error code 4

Accelerated Computing CUDA CUDA Programming and Performance
1

hi
i am trying to modify the in a way that i can specify number of threads, the is i get the error from cudaMemcpy with value of 4.
when i specify the thread with value of arraySize/2 it is ok but …
and i have no idea what is the problem.

__global__ void swapOnKernel(int *a, int size, int thread)
{
    int i = blockDim.x * blockIdx.x + threadIdx.x * 2;
	i = i* thread;

	int cacheFirst;
	int cacheSecond;
	int cacheThird;
	for (int k = 0; k < thread; k++) {
		for (int j = 0; j < size / 2 + 1; j++) {

			if (i + 1 < size) {
				cacheFirst = a[i];
				cacheSecond = a[i + 1];

				if (cacheFirst > cacheSecond) {
					int temp = cacheFirst;
					a[i] = cacheSecond;
					cacheSecond = a[i + 1] = temp;
				}
			}

			if (i + 2 < size) {
				cacheThird = a[i + 2];
				if (cacheSecond > cacheThird) {
					int temp = cacheSecond;
					a[i + 1] = cacheThird;
					a[i + 2] = temp;
				}
			}

			__syncthreads();
		}
		i++;
		__syncthreads();
	}
}

this fuction is a bubble sort
kernel.cu (5.02 KB)

2

When I run your code in kernel.cu as-is I get the message:

Sorting failed.

I can’t really grasp what your intent is with the kernel code, in what way this relates to bubble-sort, or why you wouuld choose to parallelize bubble sort. I haven’t studied your code carefully, but at first glance it looks like you would have separate threads reading and writing the same locations, and this would constitute a global memory race condition.

Do you have any version of this that actually sorts numbers correctly on the GPU?