[llvm-branch-commits] AlwaysInliner: A new inlining algorithm to interleave alloca promotion with inlines. (PR #145613)

[Mircea Trofin via llvm-branch-commits]

mtrofin wrote:

> Final ping. Does anyone have _objections_ to this patch?

It seems very odd to tackle this in the always inliner, as it adds a new concern to a fairly scoped pass. Are you sure there are no alternatives, for instance, having a pre-AlwaysInliner pass to mark off callsites that shouldn't be inlined, and then re-enabling them afterwards? 

Maybe a RFC would help gather more feedback and consensus?

https://github.com/llvm/llvm-project/pull/145613