[PATCH] D19558: Codegen: [X86] Set preferred loop alignment to 32 bytes.

[Zia Ansari via llvm-commits]

zansari added a subscriber: zansari.
zansari added a comment.

In general, I'm not a big fan of blindly aligning loops (on any boundary), as this can cause random effects +ve and -ve.
It's very simple to come up with examples where aligning a loop will cause regressions on certain architectures, specifically in loops that have control flow.

Having said that, I'd anticipate a change like this to cause pretty broad impact on code in a random way, and also bloat.

If the benchmark affected is important, it would be best to try and figure out the exact cause and see if there's a reasonable compiler heuristic that can be applied to target the specific issue (in many alignment-type regressions, there aren't). If you need help with this part of the analysis because it's tricky to determine the cause, I'd be happy to try and help.

Zia.


Repository:
  rL LLVM

http://reviews.llvm.org/D19558