[PATCH] D61604: [InstCombine] sink FP negation of operands through select

[Kevin Neal via llvm-commits]

I still need to finish D55897 (fptrunc and fpext) and then get back to wrapping up D43515 (fptosi and fptoui at this point).

I was working on fneg while waiting for reviews. So it'll be a while before I can put a lot of time in it. Any idea on how much more needs to be done in InstCombine? 

If there's not much more to do in InstCombine then I could probably finish it before D43515. But if someone beats me to it then that's fine with me.

How should we handle regressions in code quality that are exposed by tests?

-----Original Message-----
From: Cameron McInally via Phabricator <reviews at reviews.llvm.org> 
Sent: Monday, May 06, 2019 3:37 PM
To: spatel at rotateright.com; Kevin Neal <Kevin.Neal at sas.com>; cameron.mcinally at nyu.edu; Matthew.Arsenault at amd.com; andrew.kaylor at intel.com
Cc: mcrosier at codeaurora.org; wei.ding2 at amd.com; hiraditya at msn.com; llvm-commits at lists.llvm.org; jun.l at samsung.com
Subject: [PATCH] D61604: [InstCombine] sink FP negation of operands through select

EXTERNAL

cameron.mcinally added inline comments.


================
Comment at: llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp:347
+  }
+
   // Only handle binary operators (including two-operand getelementptr) with
----------------
spatel wrote:
> cameron.mcinally wrote:
> > Do you know what pass needs updating so that we can use the UnOp form? I could work on that next...
> >
> > If you don't know, that's fine. I can dig it up.
> I think it's mostly within instcombine - anywhere we're implicitly or explicitly (m_BinOp) matching generic binop patterns will not match with the new fneg in IR.
Ok. @kpn already has the bulk of InstCombine covered, so I'll leave that to him. If that's okay with you, Kevin.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61604/new/

https://reviews.llvm.org/D61604