[llvm] DAG: Improve fminimum/fmaximum vector expansion logic (PR #93579)

[Evgenii Kudriashov via llvm-commits]

================
@@ -28,35 +28,17 @@ define half @test_fminimum(half %x, half %y) {
 define <8 x half> @test_fminimum_scalarize(<8 x half> %x, <8 x half> %y) "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" {
----------------
e-kud wrote:

As you've said we test that only VMIN/VMAX is enough. Otherwise we  need to deal with `+/-0` using `fpclass` and `NaN`s with `vcmpunord`. It's called scalarize because already scalarized version of FMINIMUM comes to lowering. Probably this is the reason why it has no much support. Let me see what's going wrong and why we stop vectorizing it.

https://github.com/llvm/llvm-project/pull/93579