-
Notifications
You must be signed in to change notification settings - Fork 94
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Update cuda arch 8.9 #1733
Update cuda arch 8.9 #1733
Conversation
08da2d3
to
ac0251f
Compare
.gitlab-ci.yml
Outdated
@@ -108,6 +108,64 @@ build/cuda110/nompi/gcc/cuda/release/shared: | |||
# disable spurious unused argument warning | |||
EXTRA_CMAKE_FLAGS: "-DCMAKE_CUDA_FLAGS=-diag-suppress=177" | |||
|
|||
# cuda 11.4 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why are these versions necessary? We already test 11.0, 11.7, 12.0. I could understand if we also add the latest cuda version (12.6), but these seem quite random and unnecessary to me.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Not necessary but good to have.
I consider all versions are different, so it will be great if we test all.
Also, something in nvhpc is different from the normal setup in my opinion.
We do not do that actually just limited by the resource.
Also, I only pick up those in the docker hub now because we do not process the our container currently.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We can't test all versions, so my question still remains, are these versions so important that we need to add extra jobs, each might take ~40--60min, to our CI? Until I see a compelling reason, I will block this particular change.
Also, I wouldn't want to change our CI this close to the release. After the release, we should restructure our CI anyway to be a bit less random in which versions we are using.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I am seeing it in the opposite side. I do not have idea why we drop these version accendidently. The version jump is roughly 3 minor version in nvcc side. They change some components version when updating the version. As #1730 , we do not test it a lot because only two of them only in nvhpc (I still think nvhpc is different from nvcc with the same cuda version). To get more confident with the thrust without giving the user option to turn it off by comments, which actually made something broken and fix recently, I manually run additionaly 4 version to check.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The CI changes should be removed, see the comment thread.
Otherwise LGTM.
ac0251f
to
425b189
Compare
457fe1c
to
812843f
Compare
This pr updates cuda arch 8.9 and pick the missing change from custom thrust namespace Related PR: ginkgo-project#1733
This PR adds Ada to sSMtoCores, adds ada and hopper in selector