-
Notifications
You must be signed in to change notification settings - Fork 415
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
Distributed matrix-matrix multiplication #16107
Conversation
We need an overload of |
Just realized there is a lot of stuff wrong with this PR. Converting back to draft |
Change to bulk communication since we're not creating domains for subarrays in each iteration anymore. This solves the privatization overhead we had before
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 changes look good 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.
LGTM! @ben-albrecht
I think LinearAlgebra tests don't use distributed arrays and as noted in the OP we need those tests to be modified to use distributed arrays (maybe in combination with non-distributed ones) for correctness. I also think that we need a performance test that uses distributed arrays with LinearAlgebra.dot even if the implementation is the same as PRK-DGEMM (it is not guaranteed to stay that way) |
Actually, I realized that it has already changed from PRK-DGEMM when I added the new block iterator and single initialization of local arrays EDIT -- I'll update the PR with the tests |
Will it be okay to copy the test from PRK-DGEMM as is and refactor it for distributed dot? |
@e-kayrakli I've updated the PR with tests (both performance and correctness). Can you do a sanity check? |
Looked good on a quick look, will test and merge on Monday. |
Fixed. I'm running the test now to make sure it compiles and runs |
Compiles and runs |
I'll re-test and review this this week. @rahulghangas -- sorry for the delay with the release and everything. If you could summarize what this PR is about in the PR message, that'd be great. Remember that they go into git history directly, so details can be helpful. Thanks! |
Based on silence (and conflicting files), I am closing this PR. @rahulghangas -- Please feel free to reopen when you have the time to work on this. |
A general distributed matrix-matrix multiplication procedure.
targetLocales
or entirely different distributions