[PATCH] D46279: [LVI] Remove an assert on case which could happen in LazyValueInfo.

[Philip Reames via Phabricator via llvm-commits]

reames requested changes to this revision.
reames added a comment.
This revision now requires changes to proceed.

The assertion is checking the a value is not live-in to the entry block unless it is a global.  That assert is correct.   From your test case, it looks like you've actually found a place where we recurse on a non-local search we shouldn't have.  (i.e. a local add with two constant operands should never trigger the backwards walk.)

p.s. You can probably write your test as an IR test.


Repository:
  rL LLVM

https://reviews.llvm.org/D46279