[llvm-branch-commits] [llvm] [InstCombine] Fold fpto[su]i of llvm.trunc (PR #231414)

[Harrison Hao via llvm-branch-commits]

harrisonGPU wrote:

> Looks reasonable to me, but I'll leave it to the FP experts... Is it possible to add an alive proof (maybe using half to avoid timeouts?)
> 
> I'd also suggest adding some tests with value FMF to make it clear that their preservation was considered.

Thanks. I tried using Alive, but it timed out even with half precision. I also added FMF tests named `fptoui_trunc_fmf` and `fptosi_trunc_fmf`.

https://github.com/llvm/llvm-project/pull/231414