[clang] [llvm] [X86] Fix x87 excess precision escaping into f32/f64 values (PR #215104)

[Dennis Duda via cfe-commits]

================
@@ -129,6 +129,9 @@ const llvm::fltSemantics &llvm::getFltSemanticForLLT(LLT Ty) {
       return APFloat::IEEEsingle();
     case 64:
       return APFloat::IEEEdouble();
+    case 80:
+      // Not an IEEE format, but the only 80-bit float there is.
+      return APFloat::x87DoubleExtended();
----------------
seritools wrote:

this is obviously ugly here, any suggestions where to move it?

https://github.com/llvm/llvm-project/pull/215104