Joe: 32-bit targets don't support Int128, and Windows and non-x86 targets don't support Float80. It should be an error to codegen a runtime value of these types if the platform doesn't support them. (However, simply referencing the type name should still be allowed.)