[PATCH] D48330: [GVN] Avoid casting a vector of size less than 8 bits to i8

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

It should be possible to implement this sort of coercion with an appropriate cast, rather than just bailing out, I think.  But I won't block the patch on that.

Please add a testcase for the load-load case, in addition to the load-store case.


Repository:
  rL LLVM

https://reviews.llvm.org/D48330