[llvm] [DAGCombiner] Extend FP-to-Int cast without requiring nsz (PR #161093)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> I've tested the code generation of -mtriple=amdgcn. It seems that the assembly code is expected, but I'm not sure how to write a general testcase for every target.

It's missing from the PR? 

https://github.com/llvm/llvm-project/pull/161093