WSL and PGI compiler works great!

I put the WSL 2 on Tuesday night with Ubuntu 20.04. Added the PGI Compiler and everything works like a charm! Thanks,
Erin

Hi Erin,

That’s good news, but we didn’t do anything to specifically support WSL and haven’t tested it ourselves.

-Mat

That’s really interesting! I assumed that you helped in the process. I would be happy to help with this stuff if you need another set of eyes,

Hi Erin,

Did you happen to install CUDA on WSL and test it with your OpenACC code?

We had a report of some type of crash when someone else was trying this on WSL2. Our IT is in the process of set-up a system for us so we can try, but wanted to see if this worked for you.

Thanks,
Mat

Yes. No problems here.

Back again.

My OpenACC appears to work fine. Which doesn’t make sense now.

However, when I tried to run a CUDA Fortran program, I get the following:

status = 100(no CUDA-capable device is detected)

I tried all kinds of stuff with lspci, but no device appears, as well as ubuntu-drivers devices

So this must be where the snag is.

Thanks,
Erin

Ok, we’ll take a closer look once we get WSL installed someplace.

How do you compile your OpenACC codes? With just “-acc” or with “-ta=tesla”?

By default, a unified binary is created so if a NVIDIA GPU is found, it will run there, if not, it will run on the host.

-Mat

Hi Mat:

I am just using -acc and -Minfo=accel

Thanks,
Erin

I FINALLY got everything to work. Here is what I did after several unsuccessful attempts. Do the sudo install on the build-essential. Then sudo install on nvidia-cuda-toolkit. Download the tar ball from PGI community edition. Untar. Then sudo ./install.

Works with CUDA FORTRAN and OpenACC.

Thanks

That’s good news. I’ll pass the info along.

-Mat