[PATCH] D63489: [InstSimplify] LCSSA PHIs should not be simplified away

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a comment.

In D63489#1550529 <https://reviews.llvm.org/D63489#1550529>, @lebedev.ri wrote:

> Since this problem is being encountered, i'm guessing that even without this workaround,
>  the same problem can still be encountered if the original IR is not in LCSSA form to begin with,
>  i.e. i think the D60834 <https://reviews.llvm.org/D60834> sounds like the properer fix (ignoring the roadblocks on the way there.)


Agree - InstSimplify is a very early (but frequently used) target-independent (no TTI/TLI) canonicalization pass/analysis. I don't see why it should be crippled as proposed in this patch. 
The problem is that target-independent IR canonicalization passes like EarlyCSE are running late in the IR codegen pipeline?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63489/new/

https://reviews.llvm.org/D63489