[clang] [llvm] [PowerPC] Enable _Float16 (PR #208542)

[Tony Varghese via cfe-commits]

tonykuttai wrote:

#196559 implements `_Float16` for PowerPC with FPR-based argument passing on Power8 and later.

Enabling `_Float16` through the soft-promotion path also silently defines its calling convention, half is passed as `i16` in GPRs. The ELFv2 psABI currently has no definition for `_Float16` at all, and no released compiler supports the type on Linux on Power, so whichever change merges first effectively creates the de-facto ABI.

On the IBM side, we are working with the GCC team to define the `_Float16` and `bfloat16` ABI for PowerPC. GCC's in-flight implementation ([Michael Meissner's gcc-patches series, Nov 2025](https://www.mail-archive.com/gcc-patches@gcc.gnu.org/msg388214.html)) passes scalar `_Float16` in FPRs in the native 16-bit format and explicitly marks the convention experimental (-Wpsabi) pending that definition. #196559 implements the same FPR convention and we plan to revisit and finalize it once the ABI definition is in place.

This applies to your other patch #208598 as well. 

https://github.com/llvm/llvm-project/pull/208542