[PATCH] D109789: [SCEV] Stop blindly propagating flags from inbound geps to SCEV nodes

[Philip Reames via Phabricator via llvm-commits]

reames added a comment.

In D109789#3034640 <https://reviews.llvm.org/D109789#3034640>, @nikic wrote:

> If we get too many perf regression reports though, we may have to back this out and do some improvements to the poison reasoning first. I don't consider fixing this particularly urgent, as it's a very long-standing known bug.

We're caught in a bit of a catch-22 here.  If I strengthen the poison reasoning without fixing this and D109845 <https://reviews.llvm.org/D109845>, we start seeing miscompiles in practice.  I'd originally hoped to start with the stronger analysis, but my first attempt locally turned up some worrying results.  If I can get these in, the basic structure I used in D109845 <https://reviews.llvm.org/D109845> ports to the generic case exactly as you mention in comments.

At worst, I'll put both under a flag, implement the enhancements to the analysis under the same flag, then enable all of it in one big flag throw.  I'm really hoping not to have to do that though...


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D109789/new/

https://reviews.llvm.org/D109789