[llvm] [Draft] Data flow based shrink wrapping (PR #191784)

[Mikhail Gudim via llvm-commits]

mgudim wrote:

> > We know that shrink-wrapping is important improvement and we have to approaches: this one (lets' call it "data-flow approach") and mine, for the lack of better name let's call it "early-csrs". I feel like we really need to figure out what to do with them. The question is: do we want (1) both of them? This would make sense if for example there is no way to produce correct CFIs always with "early-csrs", so we have to use the "data-flow" when CFIs are needed. (2) one of them?
> > If we want both of them, can we concentrate on one at a time? My patches are up for months if not years, which I find pretty frustrating btw, and now there are two approaches to review?
> 
> Can you add a link to your patches?
@jayfoad 
https://github.com/llvm/llvm-project/pull/90819#issuecomment-4042035694

this Friday is the original PR's birthday - 2 years)

https://github.com/llvm/llvm-project/pull/191784