More information: I have two modes in my program, one mode uses OpenCL events, and one does not.
The kernels are the same for both.
So, when I run GPU profiling for the mode with events, then CodeXL hangs. But, when I run the mode that doesn't use events,
it works fine.
So, it looks like the usage of events is causing this problem. I am using user events as well as runtime generated events.