[Mlir-commits] [clang] [lld] [llvm] [mlir] [polly] [SetVector] Make SetVector::insert consistent (NFC) (PR #232372)

[Florian Hahn]

fhahn wrote:

> > > his is a nice-to-have that does not degrade compile time, as the iterator is never used.
> > 
> > 
> > It looks like the compile-time tracker shows a +0.10% for stage2-O3 AFAICT?
> 
> Yeah, not sure what it means: does it mean that clang is not optimizing away the find from the first member of the pair on O3, while gcc is? Seems a bit surprising, as it's just straight-forward DCE?

stage2 is built with ThinLTO, so presumably the new version cannot be optimized as well as the original version with ThinLTO. This unfortunately happens somewhat frequently :/ 

https://github.com/llvm/llvm-project/pull/232372