[llvm] [VPlan] First step towards VPlan cost modeling. (PR #92555)

[via llvm-commits]

ayalz wrote:

> (Submit responses to pending comments from earlier)
> 
> 
> 
> > OK. Measuring sub-optimal cost of initial VPlan would help demonstrate the current state of CM, and VPlan's ability to measure more accurate costs after various VPlan transforms are applied.
> 
> 
> 
> Right, the current approach has the benefit of grouping together all cases where we already diverge from the legacy cost model, which hopefully makes those gaps clearer. Unless you prefer to move the cost-computation to an earlier stage (and also introduce a mapping between cost <-> VPlan) I'd go ahead and land the current version a bit later today.

Sure, current version is fine and accepted. Your call as computing cost earlier may be somewhat safer and more gradual, though would still need some skippings.

https://github.com/llvm/llvm-project/pull/92555