Determine if Memory Accesses are Being Coalesced

I am having problems getting my CUDA program to perform the way I expect it to. I think the problem is that I am not achieving coalescing in one of my global memory accesses.

Is there any way to determine (for sure) if memory accesses are being coalesced or not?


Try CUDA Visual Profiler

Thank you for this tool. This will be very helpful for my continued work in the field of GPUs.

However it looks like there is a bit of a learning curve associated with getting this installed and using it. The project I’m working on is already past due. Is there any other way to determine this information?

There isn’t much of a learning curve to it. Install it, select the .exe output from your last compiled run. Select a runtime, start it up and let it run. It will then display all the information in a nice chart for you.