[clang] Revert "[Clang][AMDGPU] Change __fp16 to _Float16 in builtin definitions" (PR #185861)

[Joseph Huber via cfe-commits]

jhuber6 wrote:

Yeah, we honestly need to get the builtins in agreement. There's fp16 and ext_vector issues that basically mean you can't compile the runtimes with an upstream compiler.

Honestly if someone had a build script for all the runtimes you could probably just run an LLM in a loop on it until it fixed everything and then just check the results.

https://github.com/llvm/llvm-project/pull/185861