Problem with 2D block

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hello,

I’m trying to store each bock’s x and y in a 2D array, and turn them back to the host, my code is:

#include <iostream>

#define N 95

__global__ void add(int *ch) {
	*(ch + blockIdx.x * 2) = blockIdx.x;
	*(ch + (blockIdx.x * 2) + 1) = blockIdx.y;
}
int main(void) {

	int h_ch[N][2];
	int *dev_ch;

	cudaMalloc((void**)&dev_ch, sizeof(int[N][2]));

	dim3 numBlocks(N,N);

	add << <numBlocks, 1 >> > (dev_ch);

	cudaMemcpy(h_ch, dev_ch, sizeof(int[N][2]), cudaMemcpyDeviceToHost);

	for (int i = 0; i < N; i++) {
		printf("%d-%d\n", h_ch[i][0], h_ch[i][1]);
	}

	cudaFree(dev_ch);

	return 0;
}

The result of this code is:

cuda_test.exe
0-94
1-94
2-94
3-94
4-94
5-94
6-94
7-94
8-94
9-94
10-94
11-94
.
.
.
92-94
93-94
94-94

As can be seen, the value of x is changed (as expected), but the y is 94 for all rows. I was wondering what is wrong? Is something wrong with the code, or I understood the concept of grids/blocks/threads wrong?

Thanks in advance.

#2

So you have an NxN grid of thread blocks that try to store into an Nx2 array of results? Something tells me that’s not going to work well …

#3

Yes, exactly. what is the problem ?

#4

There are NxN = 95*95 = 9025 thread blocks. You want to store each block’s [x,y] coordinates in an array. How many elements does that array need? I would claim 9025 elements, because that’s how many blocks there are. How large an array does the code provide?

#5

So, I should allocate memory for sizeof(int[N*N][2]) ? And how should I calculate index for the storing?

#6

Hint: Much can be learned by (1) thinking and (2) experimenting. I am absolutely certain you can figure this out.

https://en.wikipedia.org/wiki/Think_(IBM)

#7

I think the correct calculation for indexes is:

int idx = (blockIdx.x * N) + (blockIdx.y * 2);

*(ch + idx) = blockIdx.x;
*(ch + ++idx) = blockIdx.y;

And also, allocated sizeof(int[N*N][2]) in memory. However, it didn’t work fine so I changed my code to see how x and y vary, the x changes from 0 to 94 and is repeated 94 times for each number, but, the y varies from 51 to 94 and gets repeated 94 times for each number.

What is wrong with y? shouldn’t it begin from 0 and end up to 94?