How to Step-by-Step Debug OpenCL GPU Applications under Windows with a NVidia GPU

Question

I would like to know wether you know of any way to step-by-step debug OpenCL Kernel using Windows (my IDE is Visual Studio) and running OpenCL Kernels on a NVidia GPU.

What i found so far is:

• with NVidias NSight you can only profile OpenCL Applications, but not debug them
• the current version of the gDEBugger from AMD only supports ATI/AMD GPUs
• the old version of gDEBugger supports NVidia GPUs but work is discontinued in Dec '10
• the GDB debugger seems to support it, but is only available under Linux
• the Intel OpenCL SDK brings a Debugger, but it only works while running the code on the CPU, not while running it on the GPU

Is there really a gap for this configuration (Windows + NVidia GPU + OpenCL)?

Answer 1

There is currently no method of debugging into OpenCL kernels on nVidia hardware.

nSight will only debug CUDA kernels. It will profile ("trace") CUDA and OpenCL kernels. I have heard rumour that nVidia have an internal version of nSight that will debug OpenCL kernels but it has not been (and imho will is unlikely to be) released due to commercial reasons, i.e. pushing CUDA as the compute platform of choice.

gDEBugger no longer exists. It was bought a few years ago and became CodeXL. AMD will not support other vendors' hardware for obvious reasons.