1
votes

I'm currently writing a code in MATLAB using CUDA from a MEX file. I attached the MATLAB process to Visual Studio, both on native (x64) and CUDA, but when I stop at a breakpoint in a kernel definition I'm unable to inspect the values in global memory, even when I can see any other variable defined on thread (local) memory.

Is this something expected or I'm doing anything wrong? If not, there is something I can do to inspect the global memory while debugging a kernel?

1

1 Answers

1
votes

I figured it out that Visual Studio, while attaching the process both in native and CUDA, it can debug and stop in breakpoints in kernel definitions and native code, but it cannot inspect global memory AND host memory in this way. It works properly if I only attach to the CUDA process.

It would be helpful though if someone can point out why this happens.