hello guys,
I’m writing a cuda program under windows xp with visual studio.
the program runs fine in the debug mode, but now i need to test the performance, so i compiled it into the release mode. however, unpredictable CUDA_LAUNCH_FAILED problem happens. each time, it happens on different cuda function call at different time.
i have noticed that, there are few lines in my code:
right=1;
for(i=10;i>=right;--i)
{
list[i*2]=list[(i-1)*2];
list[i*2+1]=list[(i-1)*2+1];
}
if i replace “right” with the constant number 1, the program will not generate the CUDA_LAUNCH_FAILED problem.
I don’t know if this has things to do with the loop unroll of the cuda compiler. but i have already disabled any optimization.
i don’t know how to debug this, since it happens differently each time and is fine under the debug mode.
do you have any suggestion?
thank you so much.