[clang] [llvm] [AMDGPU][clang][CodeGen][opt] Add late-resolved feature identifying predicates (PR #134016)

[Joseph Huber via llvm-commits]

jhuber6 wrote:

> So your users today are building for generic AMDGPU but using builtins that are only available on a specific processor release? Presumably that code is protected _somehow_ and their programs are not simply crashing at runtime. Is that something you'd be able to leverage at all, or is completely ad hoc?

This is basically how the NVIDIA device library and ROCm device library already work. In the latter case we just accept that `globalopt,dce` is required to clean that up after injecting the library code into the user's application. Part of this formalizes that.

https://github.com/llvm/llvm-project/pull/134016