pgcollect Access Violation (CUDA)

Hi,

I’m trying to use the pgcollect command to profile a CUDA program. I get this error for the matmul program as well so I’ll use that to illustrate. The source code is here http://www.pgroup.com/lit/samples/matmul.CUF

I compile my program using:

pgfortran -fast -Minfo=ccff matmul_CUDA.CUF -o matmul_CUDA

This works fine (I can run this executable with no problems). When I try to profile it using:

pgcollect -cuda matmul_CUDA

I get the following error:

target process abnormal termination: Error: ACCESS_VIOLATION in target program: Exception
no profile generated


I really need to be able to profile my code so if anyone can help that’d be great :)

I’m running Win XP 32bit and using NVIDIA GeForce 210 (compute v1.2)[/url]

phw_89:

I’ll try to reproduce the behavior you are seeing.

In the meantime, you might try

pgcollect -cuda=gmem,cc12 matmul_CUDA

to see if things work any better.

–Don

Thank you for the reply. Unfortunately I have already tried that with no luck.

There’s plenty of room on the Device for the arrays so it’s not like there’s memory issues. I’m pretty stumped to be honest…

I haven’t been able to reproduce the behavior you are seeing here, although unfortunately I don’t have the exact configuration you are using.

In order to make progress on this, please send an email to trs@pgroup.com. Refer to this User Forum thread (https://forums.developer.nvidia.com/t/pgcollect-access-violation-cuda/132234/1) and ask that a TPR be filed and assigned to “donb”.

Then, if you are willing, I will contact you and we can work out a plan to debug this problem.

thanks
–donb