I have a Windows 8.1 Pro machine (Xeon E5-2650 v2 @ 2.6GHz (16 processors), 32GB RAM, 256GB SSD, 1TB drive, OpenCL2.0 AMD-APP 1642.5) with the latest AMD releases of Catalyst (Driver Packaging Version: 14.501.1003-141120a-178000C, Catalyst Version:14.12 AMD Catalyst Omega Software) and CodeXL (1.78364). I am able to successfully debug and profile the Teapot example using CodeXL. I have an image processing application with a number of kernels that I have verified generate correct results. The application runs successfully on my machine, but I cannot profile it (GPU: Performance Counters). When debugging in CodeXL, I can stop in some kernels. Putting a breakpoint in others results in an error dialog with text "Could not debug kernel."<CR>"Error during kernel debugging.". The <CR> is mine to show that it is a two line error message.
My question is this: What attributes of a kernel or kernel invocation might make it a kernel that runs successfully but causes CodeXL to produce the error I am seeing?
My theories are:
1. Too many work items.
2. Code complexity.
3. Too many kernels.
4. Source file too long.
5. Dogs and cats sleeping together.
5. Aliens able to probe through my aluminum foil hat.
Please don't ask for more details of my software or system. I can't provide them.
I note again that both Teapot and my application run successfully on this system and that I am able to use CodeXL to debug and profile Teapot on this system