[llvm] [AMDGPU] Implement IR expansion for frem instruction (PR #130988)

[Frederik Harwath via llvm-commits]

================
@@ -5671,6 +5671,10 @@ class TargetLowering : public TargetLoweringBase {
                                        LoadSDNode *OriginalLoad,
                                        SelectionDAG &DAG) const;
 
+  /// Indicates whether the FRem instruction should be expanded before
+  /// ISel in the LLVM IR.
+  virtual bool shouldExpandFRemInIR() const { return false; };
----------------
frederik-h wrote:

I see that I can check whether the operation is going to expand using `TargetLowering::isOperationExpand`.  But there is no `RTLIB::Libcall` enum value for the `frem` instruction in `llvm/include/llvm/IR/RuntimeLibcalls.def"` which is what we are trying to expand here; of course, we could include the `fmod` variants here. So, for `frem`, I would just check for  `!TargetLowering::isOperationExpand`, right?

https://github.com/llvm/llvm-project/pull/130988