[clang] [Clang][bytecode] Add interp__builtin_elementwise_triop_fp to handle general 3-operand floating point intrinsics (PR #157106)

[Timm Baeder via cfe-commits]

================
@@ -2736,12 +2736,14 @@ static bool interp__builtin_ia32_pmul(InterpState &S, CodePtr OpPC,
   return true;
 }
 
-static bool interp__builtin_elementwise_fma(InterpState &S, CodePtr OpPC,
-                                            const CallExpr *Call) {
+static bool interp__builtin_elementwise_triop_fp(
+    InterpState &S, CodePtr OpPC, const CallExpr *Call,
+    llvm::function_ref<APFloat(const APFloat &, const APFloat &,
+                               const APFloat &, const FPOptions &)>
----------------
tbaederr wrote:

We might end up just always passing the rounding mode instead of the `FPOptions`, I'm not sure what else we need it for. Otherwise, we should pass them as a copy since they're only 4 bytes and other code does copies as well.

https://github.com/llvm/llvm-project/pull/157106