OpenCL profiler aborts

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I was just trying out the OpenCL profiler 64bit on Linux on my freshly ported OpenCL application (a Brownian dynamics
simulation).

My program (which runs normally outside the profiler) always aborts on the fourth profiling pass (the first three runs are successful), upon calling clEnqueueNDRangeKernel() with error CL_OUT_OF_RESOURCES.

Reducing the work group size from 128 to 64 did not help.

Any hints?

best regards,
Jens.

EDIT: It is of course the program inside the profiler which aborts, not the profiler itself! :)