Update Dockerfiles to rocm 7.0

[djramic]

Motivation

Preparing CI Dockerfiles for rocm 7.0

Technical Details

• ROCm version changed form 6.4.2 to 7.0
• Update dependency: use graphics instead of amdgpu (libdrm-amdgpu-dev is no longer packaged under amdgpu)

Test Plan

Manually tested

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1991      +/-   ##
===========================================
+ Coverage    79.50%   79.76%   +0.26%     
===========================================
  Files          100      102       +2     
  Lines        31016    32361    +1345     
  Branches      4819     5104     +285     
===========================================
+ Hits         24659    25811    +1152     
- Misses        4245     4340      +95     
- Partials      2112     2210      +98     

Flag	Coverage Δ
mfma	?
navi4x	79.76% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.
see 54 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[pabloantoniom]

May I ask why this change is needed? Buildbot is actually stuck in 6.3, it's not even in 6.4. More background here: https://github.com/ROCm/rocMLIR-internal/issues/1979#issuecomment-3270870992

[dhernandez0]

I guess he replaced all rocm6.4 found in the project. Let's a add a comment here saying this should stay in 6.3 to avoid this happening in the future.

[pabloantoniom]

So in conclusion I think this should be 6.3 and, at some point, should be updated to 7.0, for which we will need a newer machine to run the buildbot, see https://github.com/ROCm/rocMLIR-internal/issues/1981, so adding a comment here makes sense to me 👍

[pabloantoniom]

I think we should discuss if changes on the buildbot Dockerfile are needed before merging this

[dhernandez0]

do we want to use 24.04?

[umangyadav]

Yes

[umangyadav]

I am not sure though if that would lead to different python version and trigger more changes.

[dhernandez0]

can we update dockerImage() and similar functions in Jenkins files?

[dhernandez0]

also, I think perfRunner.py and tuningRunner.py might have some issues. rocprofv3 needs "-f csv" at least on MI350, so I suspect it's needed for rocm7.

[djramic]

can we update dockerImage() and similar functions in Jenkins files?

I'll cover this in the next PR once the Docker image has been created and pushed

[dhernandez0]

I think we should run nightly or weekly CI?

[djramic]

I think we should run nightly or weekly CI?

The build image job gets triggered after merging a PR that modifies the Dockerfile, so this one needs to be merged first. I'll add these checks in a new PR with Jenkinsfile changes.