[DAG] SDPatternMatch - m_Zero can't see through bitcasts

[RKSimon]

X86 uses ISD::isBuildVectorAllZeros to peek through bitcasts to match zero vectors in a large number of combines.

Ideally we'd use SDPatternMatch to match these combines and use m_Zero() instead, but regrettably this fails to match the zero vectors through bitcasts,

84397c6 shows an example of the regression

If we convert m_Zero() to something closer to m_AllOnes / AllOnes_match we might be able to make this work.

[llvmbot]

Hi!

This issue may be a good introductory issue for people new to working on LLVM. If you would like to work on this issue, your first steps are:

1. Check that no other contributor has already been assigned to this issue. If you believe that no one is actually working on it despite an assignment, ping the person. After one week without a response, the assignee may be changed.
2. In the comments of this issue, request for it to be assigned to you, or just create a pull request after following the steps below. Mention this issue in the description of the pull request.
3. Fix the issue locally.
4. Run the test suite locally. Remember that the subdirectories under test/ create fine-grained testing targets, so you can e.g. use make check-clang-ast to only run Clang's AST tests.
5. Create a Git commit.
6. Run git clang-format HEAD~1 to format your changes.
7. Open a pull request to the upstream repository on GitHub. Detailed instructions can be found in GitHub's documentation. Mention this issue in the description of the pull request.

If you have any further questions about this issue, don't hesitate to ask via a comment in the thread below.

[llvmbot]

@llvm/issue-subscribers-good-first-issue

Author: Simon Pilgrim (RKSimon)

X86 uses ISD::isBuildVectorAllZeros to peek through bitcasts to match zero vectors in a large number of combines.

Ideally we'd use SDPatternMatch to match these combines and use m_Zero() instead, but regrettably this fails to match the zero vectors through bitcasts,

84397c6 shows an example of the regression

If we convert m_Zero() to something closer to m_AllOnes / AllOnes_match we might be able to make this work.

[average-avancha]

Hi, I'd like to work on this issue. Could the issue be assigned to me? Thanks!

[RKSimon]

Good luck! Please keep this ticket updated with your progress so nobody thinks it has been abandoned.

[RKSimon]

@average-avancha Any progress?

[average-avancha]

Hi @RKSimon, sorry for the delay but I managed to put together a fix that should allow m_Zero to peek through bitcasts. I'm still testing the solution locally but I think it should work.

Here's my fix on a fork of the repo: average-avancha#1

I can move forward with an actual PR once I'm done locally testing.
I'd appreciate any feedback, thanks!

[RKSimon]

@average-avancha The patch looks like it's just for IR PatternMatch, have you made any progress on the DAG SDPatternMatch equivalent?

[average-avancha]

Not at the moment, seems like my effort was misguided. I'll work through the DAG SDPatternMatch section. Sorry about the confusion.

[RKSimon]

@average-avancha reverse-ping

[laurenmchin]

@RKSimon if @average-avancha isn’t actively working on this, I’d like to claim this issue

[RKSimon]

ping @average-avancha