[flang-commits] [clang] [flang] [llvm] [AsmWriter] Change the output syntax of floating-point literals. (PR #190649)

[Joshua Cranmer via flang-commits]

================
@@ -5780,6 +5787,9 @@ DoubleAPFloat frexp(const DoubleAPFloat &Arg, int &Exp,
                        std::move(Second));
 }
 
+APInt DoubleAPFloat::getNaNPayload() const {
+  return Floats[0].getNaNPayload();
----------------
jcranmer-intel wrote:

I'm not entirely certain what the standard interpretation of NaN payloads is for the `ppc_fp128` type is, and I'd welcome feedback on the answer.

As best as I can tell from trying to read glibc source code, they only consider the payload of the first double to be part of the NaN payload, so this appears to be the "correct answer" for the payload of double-double.

I can look into making sure that we only emit the values for APFloat if the second double is all 0, and add some tests for weirder `ppc_fp128` values.

https://github.com/llvm/llvm-project/pull/190649