[PATCH] D47983: [IR][PatternMatch] m_APInt(): allow undef elements.

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

In https://reviews.llvm.org/D47983#1135568, @RKSimon wrote:

> Would it be better to implement m_APInt_or_Undef and m_APFloat_or_Undef - explaining that the undef is only for vectors (and naturally must still contain at least one APInt).
>
> Get that in with a couple of combines updated to use it (with thorough testing), We can then gradually audit the remaining cases.


Yes. That is the road i was planning on taking.
It is unfeasible to do this in one (more or less) go.


Repository:
  rL LLVM

https://reviews.llvm.org/D47983