[clang] [Clang] Add float type support to __builtin_reduce_add and __builtin_reduce_multipy (PR #120367)

[Farzon Lotfi via cfe-commits]

================
@@ -1754,6 +1755,17 @@ static bool interp__builtin_vector_reduce(InterpState &S, CodePtr OpPC,
   PrimType ElemT = *S.getContext().classify(ElemType);
   unsigned NumElems = Arg.getNumElems();
 
+  if (ElemType->isRealFloatingType()) {
+    if (ID != Builtin::BI__builtin_reduce_add &&
+        ID != Builtin::BI__builtin_reduce_mul)
+      llvm_unreachable("Only reduce_add and reduce_mul are supported for "
+                       "floating-point types.");
+    // Floating-point arithmetic is not valid for constant expression
+    // initialization. Returning false defers checks to integral constant
+    // expression validation, preventing a bad deref of Floating as an integer.
+    return false;
----------------
farzonl wrote:

I wanted to do something like this:

```cpp
if(ElemType->isRealFloatingType()) {
    FPOptions FPO = Call->getFPFeaturesInEffect(S.getASTContext().getLangOpts());
    llvm::RoundingMode RM = getRoundingMode(FPO);
    Floating Result = Arg.atIndex(0).deref<Floating>();
    APFloat currResult = Result.getAPFloat();
    for (unsigned I = 1; I != NumElems; ++I) {
      Floating Elem = Arg.atIndex(I).deref<Floating>();
      if (ID == Builtin::BI__builtin_reduce_add) {
        if(APFloat::opStatus::opOK != currResult.add(Elem.getAPFloat(),RM))
          return false;
      } else if (ID == Builtin::BI__builtin_reduce_mul) {
        if(APFloat::opStatus::opOK != currResult.multiply(Elem.getAPFloat(),RM))
          return false;
      } else
        llvm_unreachable("Only reduce_add and reduce_mul are supported for floating-point types.");
    }
    S.Stk.push<Floating>(Floating(currResult));
    return true;
  }
```
However there are checks further for `isIntegralPointer()` that make any code here to compute the reduction not make sense. 

https://github.com/llvm/llvm-project/pull/120367