Added ROCm 7.0 support ( fix #170)

[DevManpreet5]

Motivation

Extend CI matrix to test against ROCm 7.0. Fix #170

Technical Details

• .github/workflows/iris-tests-apptainer.yml: Extended CI matrix to test both ROCm 6.3.1 and 7.0
• apptainer/iris-rocm6.3.1.def: Apptainer definition for ROCm 6.3.1
• apptainer/iris-rocm7.0.def: Apptainer definition for ROCm 7.0

Test Plan

Could not perform local testing since all AMD cloud droplets are currently out of stock.
Submitting as a Draft PR to validate changes via CI.

Test Result

N/A (pending)

Submission Checklist

• Reviewed contributing guidelines.

[DevManpreet5]

@neoblizz @maawad Hi, could you please approve and run the workflow for this draft PR? Thanks!

[mawad-amd]

Thanks for the PR Manpreet!

[mawad-amd]

Looks like the ROCm 7.0 Apptainer is broken. I suggest trying to build that locally on any system.

[DevManpreet5]

@mawad-amd can you look into new checks , old failing check worked but now 1 -rank is failing. thats Interesting , 6.3 worked but 7.0 failed because pip tried writing to a read-only path. Maybe something different in the 7.0 Docker image? so should I force user install ?

[DevManpreet5]

@mawad-amd Hi , I am getting [Errno 30] Read-only file system: '/opt/venv/lib/python3.10/site-packages/urllib3'
I was thinking to use pip install --user -e . (https://luminousmen.medium.com/why-use-pip-install-user-2df0259c8fb7) while copilot is suggesting Modify your workflow step to create and activate a new virtual environment before installing packages. ,

Which method will you suggest?

[mawad-amd]

@mawad-amd Hi , I am getting [Errno 30] Read-only file system: '/opt/venv/lib/python3.10/site-packages/urllib3' I was thinking to use pip install --user -e . (https://luminousmen.medium.com/why-use-pip-install-user-2df0259c8fb7) while copilot is suggesting Modify your workflow step to create and activate a new virtual environment before installing packages. ,

Which method will you suggest?

There seems to be something wrong with the apptainer image itself. See log here.

I think we will have to fix that first, then see if the other problem you are pointing out still persist. Do you still have problems with getting AMD GPU access (single GPU is fine)?

[DevManpreet5]

@mawad-amd Thanks , Ohh the testcase build-apptainer-7.0 passed so I didn’t check the logs I’ll look into it. Yes, I do have access to an AMD GPU now, but just building the Apptainer image fails with:

INFO:    Extracting OCI image...
INFO:    Inserting Apptainer configuration...
INFO:    Running post scriptlet
ERROR  : Failed to set mount propagation: Permission denied
FATAL:   While performing build: while running engine: while running %post section: exit status 1

for all apptainers so i will have to figure some stuff , thanks for your time !