New AlltoAllV (Imbalanced AlltoAll) benchmark. #157
Conversation
This file contains 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
Changes: - Added variable count of elements to send/recv based on sending/recieving peers - Added new file to make file Notes: - Current method of uniquely identifying the peers that are sending (thread_local of thread number) may not work correctly. Not sure if that is the appropriate way to determine rank.
- alltoallv2.cu testfile: Parameterizes with alltoallv_param.csv - run_a2av.sh script: -- Runs the built test with an arbitrarily named CSV instead of the static name -- Passes through other arguments to the testfile
Each Rank is guaranteed to send X/nranks data in some distribution.
|
cc: @AddyLaddy |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Alltoallv benchmark
This pull request incorporates a new AllToAllV benchmark. This allows for testing a wider range of communication patterns found in real-world workloads, and allows end users to easily create custom benchmarks by adjusting test parameters. It can even be used to simulate communication patterns that have dedicated tests currently, without having to compile all of them.
The benchmark requires a parameterization CSV file which contains a square matrix of dimension NxN where N is the number of ranks. Each entry in this matrix must contain a fraction between 0 and 1 (inclusive). Entry I,J determines the amount of data sent from Rank I to Rank J. The amount is equal to the fraction times the data size specified for the test. For example, if the benchmark is being run at 256M, and a particular entry has a value of 0.5, then the sending rank will send 128M to the receiving rank.
This PR also adds the ability to parameterize any test with a setup file with the
-sflag. The AllToAllV benchmark implementation can be referenced for an example on how to use this.cc: @sjeaugey (I don't have the ability to add reviewers)