[clang] [OpenMP] Basic BumpAllocator for (AMD)GPUs (PR #69806)

[Joseph Huber via cfe-commits]

https://github.com/jhuber6 commented:

I figured we would just have an internal implementation of `malloc` provided by the OpenMP device runtime library. The variants make this more difficult, but we could just use `#ifdef __AMDGPU__` instead of an OpenMP variant so we can declare it as `extern "C"` inside the AMDGPU device RTL. Since we use a two-step compilation and always compile with clang it should be legal.

https://github.com/llvm/llvm-project/pull/69806