[SYCL][Graph] Breadth-first schedule #182
Open
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
EwanC
reviewed
May 23, 2023
There was a problem hiding this comment.
I have a bunch of superficial comments on this PR, and I'm seeing a build fail, but fundamentally I'm okay with using the breadth-first approach.
A quick description (can just append to PR description) of why this algorithm is preferable would be useful though, since we may want to document this implementation detail in #183 and justify why it's the default to other internal & external stakeholders.
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Outdated suggestions cannot be applied.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.
The intent is to change the "reversed depth-first" scheduling/ordering (as implemented in topology_sort) to a strict breadth-first scheduling/ordering. The schedule is now computed by building a multimap of nodes in layers.
Draft documentation wording for the change of schedule ordering:
"Breadth-first traversal when scheduling a graph maximizes the number of independent kernels that are submitted concurrently, which enables more efficient work scheduling on massively parallel devices by overlapping kernel launch overheads with independent computation and data movement actions."
A side-effect is to change the trigger that creates the schedule so the executable graph no longer needs an internal reference to the modifiable graph from which it was created. The schedule is now created during the constructor for the executable graph rather than a later call to a method on the modifiable graph.