[PATCH] D110529: [FuncSpec] Don't crash and specialise on poison or constexpr values

Mon Sep 27 03:55:59 PDT 2021

ChuanqiXu accepted this revision.
ChuanqiXu added a comment.
This revision is now accepted and ready to land.

In D110529#3023997 <https://reviews.llvm.org/D110529#3023997>, @SjoerdMeijer wrote:

> In D110529#3023974 <https://reviews.llvm.org/D110529#3023974>, @ChuanqiXu wrote:
>
>> I met the problem about `ConstantExpr` before: https://reviews.llvm.org/D107067. So it looks good to me basically. But I am wondering if it is better to fix the solver to solve these kind of problem fundamentally?
>
> Yep, agreed. That's why I called this change to function specialisation a stopgap, but I think this makes sense because I don't think we want to support constexprs at this point.
>
> In addition, we would indeed need to fix the solver (or how it is used), which I think is a separate issue. I don't know if the solver should exit more gracefully and returns a "don't know", and/or really support these constexprs. That needs some looking into...

Agreed. I think we should add corresponding `TODO` or `FIXME` in SCCPSolver.cpp in this patch or in another NFC patch.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D110529/new/

https://reviews.llvm.org/D110529