Skew SUMMA communication to avoid bottlenecks #27982
Open
+11
−1
Conversation
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
The way I learned the SUMMA algorithm was essentially "Locales broadcast their blocks of data to all the other locales in their row/col." But in Chapel, there isn't really a broadcast option (nor would we necessarily want to use one here if there was), so I'd implemented this by having all locales do a remote read of the block in question. Engin pointed out (some time ago) that this could be rewritten to avoid bottlenecks by skewing each locale by its row/col ID such that each block copy is a 1:1 communication rather than a sqrt(numLocales):1 communication, which could bottleneck. This implements that transformation. --- Signed-off-by: Brad Chamberlain <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
The way I learned the SUMMA algorithm was essentially "Locales broadcast their blocks of data to all the other locales in their row/col." But in Chapel, there isn't really a broadcast option (nor would we necessarily want to use one here if there was), so I'd implemented this by having all locales do a remote read of the block in question.
Engin pointed out (some time ago) that this could be rewritten to avoid bottlenecks by skewing each locale by its row/col ID such that each block copy is a 1:1 communication rather than a sqrt(numLocales):1 communication, which could bottleneck. This implements that transformation.