Adding CloneToConsumersPass and ElideAsyncTransfersPass. #20103
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
benvanik
force-pushed
the
users/benvanik/clone-no-yuck
branch
4 times, most recently
from
57a757d to
54eeac4
Compare
These transfers can arise during stream conversion if there was ambiguity in the source IR.
This performs affinity analysis and clones any clonable producer that is used by multiple affinities per-user. It's effectively the inverse of a CSE step and is intended to undo CSE that created the affinity ambiguity. The primary case where this arises today is CSE collapsing a splat or splat-like dispatch that is consumed by transfers or dispatches on different affinities.
There's larger issues to fix in resource usage analysis and lifetime assignment.
benvanik
force-pushed
the
users/benvanik/clone-no-yuck
branch
from
54eeac4 to
aef6e37
Compare
|
@rsuderman after banging my head against the desk for a week I need to switch to other stuff - this should be an improvement for you by having splats/splat-like dispatches cloned to consumer affinities but may still have unneeded copies (no more than before, just unfortunately no fewer). I'll loop back and see if I can remove some of the transfers in a bit. |
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Outdated suggestions cannot be applied.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.
CloneToConsumersPass performs affinity analysis and clones any clonable producer that is used by multiple affinities per-user. It's effectively the inverse of a CSE step and is intended to undo CSE that created the affinity ambiguity.
ElideAsyncTransfersPass does a global analysis to try to find transfers that are redundant and elides them. Unfortunately resource usage refinement is somewhat broken when transfers are turned into clones so the pass is currently disabled. The pass on its own is fine and shows a large reduction in transfers when applied.
The primary case where this arises today is CSE collapsing a splat or splat-like dispatch that is consumed by transfers or dispatches on different affinities.