Bump to triton-lang/triton@dbfbc1e1e6c

[enjustli]

This updates to latest Triton main which should resolve the nightly build issues.

Primary changes:

• Change llvm/mlir api usage
• Update driver.py

[enjustli]

this PR needs triton-lang/triton#8766 merged

[enjustli]

@nhat-nguyen, sorry for the explicit ping but I did not have access to request a formal review. Could you be pleaase review the PR / or request review from appropriate stakeholders?
Thanks!

[enjustli]

gently pin

[nhat-nguyen]

the api changes for op creation and driver.py look good to me. thanks a lot for helping us update this! as for the lit tests, do you know what are the main changes that caused the differences?

tangentially, we're not using the triton-to-linalg pass anymore (triton-to-linalg-experimental is the main one now), so it might be cleaner to remove the pass and those tests first before the triton bump so that we don't have to update the lit files.

[enjustli]

the api changes for op creation and driver.py look good to me. thanks a lot for helping us update this! as for the lit tests, do you know what are the main changes that caused the differences?

tangentially, we're not using the triton-to-linalg pass anymore (triton-to-linalg-experimental is the main one now), so it might be cleaner to remove the pass and those tests first before the triton bump so that we don't have to update the lit files.

In the new llvm version, memref.reinterpret_cast enhances the analysis of static offset. so the output IR changed. due to too many changed mlir files, so i use llvm-project/mlir/utils/generate-test-checks.py to generate mlir CHECK directly.

removing the triton-to-linalg pass helps a lot! 🎉

[enjustli]

already update.

[enjustli]

gently pin

[enjustli]

ping @bmyerz0 @red1bluelost

[nhat-nguyen]

i think it would be cleaner if we delete the TritonToLinalg pass in a different PR

[enjustli]

i think it would be cleaner if we delete the TritonToLinalg pass in a different PR

I see. 👌

[enjustli]

this pr depends on #366

[bmyerz0]

Please see #367