Issues with atomicAdd when counting chars

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello.

I’m trying to count how many times the char exists in a array.

The kernel that I’m using is this one.

[codebox]global void charCounter(char *string, int *charCounter, int N)

{

int idx = blockIdx.x * blockDim.x + threadIdx.x;

if (idx < N) 

	atomicAdd(&charCounter[(int)string[idx]], 1);

}

[/codebox]

The result with the serial code is that all the characters got the same number. But the parallel results are different they are near the correct ones.

2

Mind that string being of char type will result in indices between -128 and +127 for your charCounter array.
Wouldn’t it be safer to use unsigned char instead?

3

Yes. It would be safer. It worked! External Media

thanks!