[XLA:GPU] Make SoftmaxRewriterTriton
use the new IsTritonSupportedInstruction
, as opposed to the one in legacy_triton::
.
#70895
+560
−1,074
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Outdated suggestions cannot be applied.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.
[XLA:GPU] Make
SoftmaxRewriterTriton
use the newIsTritonSupportedInstruction
, as opposed to the one inlegacy_triton::
.Notably, the old function conflated support with "support enabled by running
FloatNormalization
". Therefore, this change appears to impact what can befused end-to-end, but that should not be the case provided
FloatNormalization
is run appropriately.
Since I had to fix the tests in this direction, I took the opportunity to
edit tests in order to remain relevant, and to delete the ones that no longer
seemed useful. All in all, the tests now focus on the properties of the matcher
itself, instead of also testing Triton support which now has its own
disentangled logic. (Except for a couple of dedicated tests that ensure that
Triton support is indeed taken into account when deciding to fuse a diamond.)