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

[Florian Hahn via llvm-commits]

fhahn wrote:

> I'm seeing another crash following the reland. See #98413

Thanks, the issue was that the legacy cost model was querying the cost for extends that will be removed due to truncating its users. The query was done with the truncated types, which made TTI return an invalid cost (destination type <= source type), which meant the legacy cost model didn't consider the VF due to an invalid cost.

Should be fixed by 67f4968a57789fa07dcdb0c4421e4beeb725a0db

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