feat: add fast variant of CI build and smoke test - #109
Merged
Conversation
john00003
requested review from
amd-ivaganev,
gaoikawa,
sbates130272 and
vcave
as code owners
July 29, 2026 16:40
Collaborator
Author
|
/run-ci |
Collaborator
Author
|
We'll need to update the workflows in /usr/ on the CI runners in order for the CI changes to reflect. |
|
✅ Hardware CI passed for dafc0cd (dist-build + smoke-test + tiny-test). |
sbates130272
requested changes
Jul 29, 2026
john00003
force-pushed
the
feat/dist-build-fast
branch
from
July 29, 2026 21:02
2b207b0 to
3d69848
Compare
Collaborator
Author
|
@sbates130272 I addressed your comments! Sorry, I accidentally squashed the fixup commits, but I tried to address all of your comments. |
sbates130272
previously approved these changes
Jul 30, 2026
sbates130272
left a comment
sbates130272
left a comment
Collaborator
There was a problem hiding this comment.
@john00003 looks good! Let me run /run-ci. Do you have the updated scripts installed on your runner? Do we want a differnt unique trigger for this? /run-ci-fast perhaps?
Oh while we are at it can we rename the workflows to follow the AIC convention. AIC Hardware Test - <testname>.
- only use gfx950 - disable compiling some components of UCX and NIXL
Groups the SPUR-backed workflows under a common prefix so they are identifiable in the Actions list. workflow_run triggers match on display name, so the chained nightly references are updated in the same commit to avoid breaking the chain.
Commit 3d69848 converted the only /run-ci workflow to the single-arch fast variant, leaving no way to run the full multi-arch gate from a PR. Restore it as its own workflow so /run-ci means the full build again. The trigger must exclude /run-ci-fast explicitly: contains() is a substring match, so a bare '/run-ci' check would also fire on a fast request and start both workflows.
Frees /run-ci for the full gate and makes the fast single-arch run opt-in. The display name is suffixed "(fast)" so the two workflows are distinguishable in the Actions list and as workflow_run targets.
Both workflows previously posted the same 'hardware-ci / <stage>' contexts, so a single-arch fast run would overwrite — and could satisfy — the full multi-arch gate. Give the fast run its own '-fast' contexts and label its PR comments accordingly, so a green check always names the variant that earned it.
Both workflows key their SPUR state solely on the commit SHA — image tag rocm-aic-ci-<short>, tarball images/aic-ci-<short>, clone rocm-aic.<short>. With a single workflow that collision was unreachable, but /run-ci and /run-ci-fast on the same commit now race: the fast single-arch tarball could be what the full smoke-test loads. Both use the same group name so they queue. cancel-in-progress is false because discarding an in-flight hardware run is expensive.
john00003
force-pushed
the
feat/dist-build-fast
branch
from
July 30, 2026 20:55
da1788a to
5b00cf2
Compare
sbates130272
approved these changes
Jul 30, 2026
sbates130272
left a comment
sbates130272
left a comment
Collaborator
There was a problem hiding this comment.
@john00003 looks great thanks!
Collaborator
|
/run-ci |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Motivation
We want a faster version of the
dist-buildandsmoke-testMakefile targets.Technical Details
The
dist-build-fasttarget improves the build time by:gfx950(the machines our CI targets)Additionally, we disable building the tests/examples for both hipFile and NIXL to reduce the build time.
Currently the
smoke-test-fastMakefile target doesn't run any different tests than the normal variant, it just uses the docker container built by thedist-build-fastvariant.Test Plan
Try the
dist-build-fastandsmoke-test-fast, ensure the build and tests pass. Measure the amount of time the reduced Makefile targets take compared to the regular targets.Test Result
The reduced compile targets and disabling the tests and examples save ~30s. The reduced gfx compile targets save a couple minutes.
Submission Checklist