[llvm] [Passes] Run SimpleLoopUnswitch after introducing invariant branches. (PR #81271)

Florian Hahn via llvm-commits llvm-commits at lists.llvm.org
Fri Feb 9 12:21:18 PST 2024


fhahn wrote:

> This brings back the discussion that we really should have better infrastructure in the NPM to specify the conditional running of passes. The `extra-vectorizer-passes` change was supposed to be a one time approach.
> 

Sure, I am not really tied to the current approach and would be happy to explore alternative options; the current implementation was just the easiest way to get this done I was aware of.

> Do you have a measure of how this affects execution time for some workloads (e.g. spec)? Could you also test this on https://github.com/google/fleetbench?

Not yet, but I am planning to do this. I think the bounds check example highlights a good missed indvars -> SLU interaction

https://github.com/llvm/llvm-project/pull/81271


More information about the llvm-commits mailing list