[llvm] [AMDGPU] Fix legacy fmin/fmax combines to preserve NaN and zero ties (PR #220519)

[Arseniy Obolenskiy via llvm-commits]

================
@@ -299,6 +301,231 @@ define float @v_test_fmax_legacy_uge_f32_nnan_nsz_flag(float %a, float %b) {
   ret float %val
 }
 
----------------
aobolensk wrote:

added 

https://github.com/llvm/llvm-project/pull/220519