[PATCH] D48851: [PM/LoopUnswitch] Fix the first and immediate components of PR37889.

[Chandler Carruth via Phabricator via llvm-commits]

chandlerc added a comment.

In https://reviews.llvm.org/D48851#1151279, @fedor.sergeev wrote:

> LGTM for the code.
>
> I'm not sure whats your understanding of the state of this issue for non-trivial?
>  This code does not address the issue, but the tests supplied do not fail?


I think there is existing code that at least handles the cases in these tests. We had to handle similar stuff w/ non-trivial unswitching because we create new loops as well and we got lots of assert failures w/o handling those correctly. I mostly added the test cases here to demonstrate to myself (especially my future self) that at least these cases are already handled correctly.

> Do you need a test that fails before addressing non-trivial part?

I mean, if there is a bug in the non-trivial part, yeah, we'll need to find it and the test cases that actually hit that path. So far, it looks OK.


Repository:
  rL LLVM

https://reviews.llvm.org/D48851