Command profiler failed to launch

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hi,

I used the command profiler to profile my OpenCL kernels. Usually it works well but for the program “heartwall” in rodinia benchmark suit, the profiler says “NV_Error: Profiler launch failed.” in the output file. This is extremely weird. I checked the the return value of the clEnqueueNDRangeKernel call and it returns 0, meaning CL_SUCCESS. Can anyone tell me what can be wrong with this benchmark? Or is it a bug of the profiler?

My machine is GTX 580, running under Redhat Enterprise 5.5.

Thank you very much.