[libc-commits] [libc] [libc][math][c23] Temporarily disable float16 on 32-bit Arm (PR #95027)

[Nick Desaulniers via libc-commits]

nickdesaulniers wrote:

I'm always in favor of not requiring a dependency on a compiler runtime, though llvm-libc is capable of that at this point.  Here's a build of llvm-libc for 32b ARM:

```sh
$ llvm-nm libc/lib/libc.a | llvm-cxxfilt | grep aeabi
         U __aeabi_read_tp
         U __aeabi_ldivmod
         U __aeabi_uldivmod
         U __aeabi_ul2d
         U __aeabi_idiv
         U __aeabi_idiv
         U __aeabi_ldivmod
         U __aeabi_ul2d
         U __aeabi_uidiv
         U __aeabi_uidiv
         U __aeabi_ul2d
         U __aeabi_uldivmod
         U __aeabi_uidiv
         U __aeabi_uidivmod
```
Those symbols will be satisfied by the compiler runtime.

>From a quick discussion in #gcc on oftc.irc.net, it sounds like half precision ops are promoted to single precision for soft float.

---
That buildbot link above isn't loading for me; do you recall what the error was, specifically?

https://github.com/llvm/llvm-project/pull/95027