[libc-commits] [libc] [libc][math][c23] Enable C23 _Float16 math functions on GPUs (PR #99248)

[Joseph Huber via libc-commits]

jhuber6 wrote:

> Should we try to run some benchmarks before enabling builtins on GPUs?

If the builitin is supported it's pretty much always going to be faster than the bit-twidding generic implementation. Maybe there's some exceptions for things like `sin` or `sqrt` which might have precision issues. @arsenm the f16 builtins are generally supported in hardware, right?

https://github.com/llvm/llvm-project/pull/99248