Add support for C23 interchange float types (_Float32, _Float64, _Float128, ...)

[tgross35]

Newer versions of C allow the following relevant interchange types:

• _Float16: 16-bit IEEE floating point numbers
• _Float32: 32-bit IEEE floating point numbers
• _Float64: 64-bit IEEE floating point numbers
• _Float64x: Extended 64-bit IEEE floating point numbers (e.g. x86 f80 long double)
• _Float128: 128-bit IEEE floating point numbers
• _FloatN _Complex
• Suffix F16/f16
• Suffix F32/f32
• Suffix F64/f64
• Suffix F64x/f64x
• Suffix F128/f128

GCC currently supports all of these. Clang currently supports only _Float16, but should add the rest.

Link to the draft: https://www.open-std.org/jtc1/sc22/wg14/www/docs/n2601.pdf
Link to demo: https://clang.godbolt.org/z/5xbq9919z

C++ also supports the same suffixes for the std types, https://en.cppreference.com/w/cpp/types/floating-point

It sounds like these can't simply be redefines of float and double, see #50110 (cc @AaronBallman)

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Trevor Gross (tgross35)

GCC supports `_Float32` and `_Float64` in both C and C++, with the associated `F32`, `F64`, and `F128` literal suffixes. Clang should add support for these to be compatible.

It sounds like these can't simply be redefines of float and double, see #50110 (cc @AaronBallman)

[pinskia]

Note this is not exactly a GNU extension but for compatibility with GCC (via C23 defined extended floating point types).

[tgross35]

Yes thanks, I updated the top post to give more context. It seems like GCC does not restrict these types to only --std=c23.

[fweimer-rh]

Please note that glibc headers define _Float64 et al. by default and have a built-in list of compiler versions which provide their own definitions of these types. We can certainly backport the required header changes to make this conditional on Clang version, but it might be more developer-friendly to accept compatible redefinitions of _Float64 et al.