[llvm] [X86] Don't fold fp_round pairs into f64->f16 with F16C (PR #230079)

[Aochang Liu via llvm-commits]

================
@@ -3461,6 +3461,12 @@ class LLVM_ABI TargetLoweringBase {
     return false;
   }
 
+  /// Return true if (fp_round (fp_round x)) should be folded into a single
+  /// fp_round from \p SrcVT to \p DestVT.
+  virtual bool shouldFoldFPRoundPair(EVT DestVT, EVT SrcVT) const {
+    return true;
+  }
----------------
HelloWorldU wrote:

I got you, let me try another approach.

https://github.com/llvm/llvm-project/pull/230079