Add kCall to GetFusibleComputations #21998
Open
+51
−1
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
|
This affects priority fusion, so we'll probably want a test in priority_fusion_test.cc as well. |
jreiffers
reviewed
Jan 29, 2025
xla/service/gpu/gpu_fusible.cc
Outdated
| @@ -858,6 +858,7 @@ std::vector<HloComputation*> GetFusibleComputations( | |||
| // flow. See also fusion_wrapper.cc, which does the same. | |||
| if (HloInstruction::MightHaveCalledComputations(instr->opcode()) && | |||
| instr->opcode() != HloOpcode::kWhile && | |||
| instr->opcode() != HloOpcode::kCall && | |||
There was a problem hiding this comment.
I'm not sure we want to fuse inside all calls. We might want to just enable the async ones you care about for now.
There was a problem hiding this comment.
I broke out this check, now it only applies to calls with stream annotations.
aded974 to
a748f44
Compare
a748f44 to
1ef97c8
Compare
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Outdated suggestions cannot be applied.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.
Previously, we had issues with the stream annotation being applied to non-Gemm operations, and having the compilation fail at IR emitter stage. Operations that normally should've been fused weren't getting fused inside the async computation. This PR is intended to fix this issue by including the call computations when searching for fusible computations.
This was suggested by @jreiffers