[PATCH] D59129: [SROA] WIP: Lowering alloca is not always beneficial

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

How hard would it be to try to handle your testcase late instead? Given a series of adjacent unaligned loads, transform that into a memcpy followed by aligned loads.  Maybe we don't have appropriate infrastructure...

If we are going to do this, we probably don't want to disable SROA completely for allocas that contain a memcpy source/destination, just avoid promoting the specific slices where it's relevant.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D59129/new/

https://reviews.llvm.org/D59129