launching fail detection

Accelerated Computing CUDA CUDA Programming and Performance
1

Greetings!

If I call utility functions from CUDA, I can use the CUDA_SAFE_CALL() to collect error messages with ease. But how can I detect the launching error of a user-defined kernel function? I means, functions like func<<<dimGrid,dimBlock>>>().

I would like to know whether a certain kernel is correctly launched, or better, what’s the reason of a failure? Say, register memory exceeds or what. Can someone please tell me how to check?

Thank you!

2

cudaThreadSynchronize() after the kernel and check its return value or the return value of cudaGetLastError().

The CUT_CHECK_ERROR macro does this in debug builds.