-
Notifications
You must be signed in to change notification settings - Fork 405
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
Make in-order queues the default via macro #6189
Conversation
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.
Make it more obvious in the description that this is in some sort reverting back to using in-order queues. Please add a reference to the previous work (PR link) that had lifted the requirement.
I add some text in the pull request description. |
Will ignore HIP timeouts. Everything else passed on Jenkins. |
da97221
to
cf76d50
Compare
I just collapsed all the commits into one. |
Hi 👋. I see 4.1 (https://github.com/kokkos/kokkos/releases/tag/4.1.00) was released without this PR. This continues to be a substantial performance regression for vector operations in PETSc. Was that intentional and is there a time when in-order queues will go in a release? |
This was merged to |
We'll be asking PETSc users to |
We could make this part of a 4.1.1 |
This pull request makes in-order queues the default but retains the option for us to change this later easily.
If we are using in-order queues, we don't need any of the
depends_on
orsubmit_barrier
calls and for benchmarks up to moderate sizes, I seezeCommandListAppendBarrier
dominating without this patch.Related to #6035.
This pull request is in some way reverting back to in-order queues after we lifted the requirement in #5065 (but also guarding a bunch of other synchronization points necessary for out-of-order queues). Thus, this is for performance, not correctness (which was the issue before).