[llvm] [X86][AVX10.2] Lower fmininum/fmaximum to VMINMAX* (PR #121373)

[Evgenii Kudriashov via llvm-commits]

================
@@ -6753,9 +6756,9 @@ static const X86FoldTableEntry Table4[] = {
   {X86::VMINMAXPSZ128rrik, X86::VMINMAXPSZ128rmik, 0},
   {X86::VMINMAXPSZ256rrik, X86::VMINMAXPSZ256rmik, 0},
   {X86::VMINMAXPSZrrik, X86::VMINMAXPSZrmik, 0},
-  {X86::VMINMAXSDrrik, X86::VMINMAXSDrmik, TB_NO_REVERSE},
-  {X86::VMINMAXSHrrik, X86::VMINMAXSHrmik, TB_NO_REVERSE},
-  {X86::VMINMAXSSrrik, X86::VMINMAXSSrmik, TB_NO_REVERSE},
+  {X86::VMINMAXSDrri_Intk, X86::VMINMAXSDrmi_Intk, TB_NO_REVERSE},
----------------
e-kud wrote:

Slightly off topic, it looks like having a pattern `rri_Intk` is just a problem of adding `_Int` in a wrong place. Should we generally refactor all patterns that introduce more suffixes after `_Int` so that `_Int` is always the last? Or is it intended to show at which level an intrinsic was introduced (I assumed `Int` is for intrinsic).

https://github.com/llvm/llvm-project/pull/121373