[XLA:GPU] propagate the algorithm flag of dot op to cublasGemm custom call. #17595
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
![copybara-service[bot]](https://avatars.githubusercontent.com/in/44061?s=60&v=4){kind=small}
copybara-service
bot
force-pushed
the
test_678604330
branch
2 times, most recently
from
e6af262
to
a965de6
Compare
copybara-service
bot
changed the title
copybara-service
bot
force-pushed
the
test_678604330
branch
from
a965de6
to
729182a
Compare
copybara-service
bot
changed the title
copybara-service
bot
force-pushed
the
test_678604330
branch
2 times, most recently
from
faa4741
to
ed0fa94
Compare
copybara-service
bot
changed the title
copybara-service
bot
force-pushed
the
test_678604330
branch
10 times, most recently
from
f988fe1
to
a83908c
Compare
… call. we have the algorithm flag of dot op. we handle it in triton emitter, now let's push it to cublas via gemm_rewriter. Otherwise the cublas call uses the default f32_f32_f32 algorithm and loses the competition with triton. As a result of this change it get clear that only Ampere ran bf16 version of cublas kernel. Hopper uses tf32 for that because it does not have the b16 version for this case. DotBF16ForBf16Bf16F32Tests was removed because the algorithm BF16_BF16_F32 expects F32 input and F32 output with the BF16 arithmetics inside cublas. PiperOrigin-RevId: 679595014
copybara-service
bot
force-pushed
the
test_678604330
branch
from
a83908c
to
31b180c
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Outdated suggestions cannot be applied.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.
[XLA:GPU] propagate the algorithm flag of dot op to cublasGemm custom call.
we have the algorithm flag of dot op. we handle it in triton emitter, now let's push it to cublas via gemm_rewriter. Otherwise the cublas call uses the default f32_f32_f32 algorithm and loses the competition with triton.
As a result of this change it get clear that only Ampere ran bf16 version of cublas kernel. Hopper uses tf32 for that because it does not have the b16 version for this case.
DotBF16ForBf16Bf16F32Tests was removed because the algorithm BF16_BF16_F32 expects F32 input and F32 output with the BF16 arithmetics inside cublas.