[clang] [HLSL] Vector standard conversions (PR #71098)

[Chris B via cfe-commits]

================
@@ -1843,13 +1843,86 @@ bool Sema::IsFunctionConversion(QualType FromType, QualType ToType,
   return true;
 }
 
+/// Determine whether the conversion from FromType to ToType is a valid
+/// floating point conversion.
+///
+static bool IsFloatingPointConversion(Sema &S, QualType FromType,
+                                      QualType ToType) {
+  if (!FromType->isRealFloatingType() || !ToType->isRealFloatingType())
+    return false;
+  // FIXME: disable conversions between long double, __ibm128 and __float128
+  // if their representation is different until there is back end support
+  // We of course allow this conversion if long double is really double.
+
+  // Conversions between bfloat16 and float16 are currently not supported.
+  if ((FromType->isBFloat16Type() &&
+       (ToType->isFloat16Type() || ToType->isHalfType())) ||
+      (ToType->isBFloat16Type() &&
+       (FromType->isFloat16Type() || FromType->isHalfType())))
+    return false;
+
+  // Conversions between IEEE-quad and IBM-extended semantics are not
+  // permitted.
+  const llvm::fltSemantics &FromSem = S.Context.getFloatTypeSemantics(FromType);
+  const llvm::fltSemantics &ToSem = S.Context.getFloatTypeSemantics(ToType);
+  if ((&FromSem == &llvm::APFloat::PPCDoubleDouble() &&
+       &ToSem == &llvm::APFloat::IEEEquad()) ||
+      (&FromSem == &llvm::APFloat::IEEEquad() &&
+       &ToSem == &llvm::APFloat::PPCDoubleDouble()))
+    return false;
+  return true;
+}
+
+static bool IsVectorElementConversion(Sema &S, QualType FromType,
+                                      QualType ToType,
+                                      ImplicitConversionKind &ICK, Expr *From) {
+  if (S.Context.hasSameUnqualifiedType(FromType, ToType))
+    return true;
+
+  if (IsFloatingPointConversion(S, FromType, ToType)) {
+    ICK = ICK_Floating_Conversion;
+    return true;
+  }
+
+  if (S.IsFloatingPointPromotion(FromType, ToType)) {
+    ICK = ICK_Floating_Promotion;
+    return true;
+  }
----------------
llvm-beanz wrote:

Good catch! We do need this distinction to for overload resolution because we prefer promotions over conversions:

```c++
void Fn(vector<double,2> D);
void Fn(vector<half,2> H);

void Call(vector<float,2> F) {
  Fn(F); // Should call the double version.
}
```

That showed that I also missed a place where I need to factor in the element conversion to the conversion sequence rank. I'll get the fixed up.

https://github.com/llvm/llvm-project/pull/71098