kernel return

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

I have an issue with kernel return ! I know they must be void, I do the following :

kernel <<<  ...  >>> (a, b, c, d);

cudaThreadSynchronize();

cudaMemcpy(h_d, d, 50, cudaMemcpyDeviceToHost);

printf(" return : %s\n", h_d);

The kernel launch N threads and only one can modify the d variable. It’s something like :

if( (threadId.x == a) && (threadId.y == b) )

{

	d[0] = 'o'; 

	d[1] = 'k';

}

If I compile & execute the code in emulation mode (make emu=1) there is no problems and I get the result ‘ok’ on the host. If I compile & execute in ‘normal’ mode (to be executed on the device) I get nothing (I mean an empty string) !

Does anyone have an idea ?

Thanks,

2

Actually that’s a bug in my app. Sorry :)