As far as I know Nvidia Nsight 2.1 gives possibility to debug OpenCL code. I had passed all Nsight Monitor requirements (disable WDDM, disable aero, install development drivers, connect to second graphics card) but I still cannot debug my code.
Using 'Start CUDA debugging' in VS 2010 ends with notification:
A CUDA context was created on a GPU that is not currently debuggable. Breakpoints will be disabled.
Adapter: GeForce GTX 590
My C++ solution was created using cmake, and I didn't add kernel sources (*.cl) to my solution. My application reads kernels sources from file and compiles them. So, how can I set any breakpoints in my kernels?
Parallel Nsight 2.1 supports tracing OpenCL 1.1 using Nsight > New Analysis Seasion.
Parallel Nsight 2.1 CUDA Debugger does not support source level debugging of OpenCL kernels. Source level debugging is only supported for CUDA kernels launched through the CUDA Runtime API and CUDA Driver API.