[clang] [AMDGPU] Allow overload of __builtin_amdgcn_mov/update_dpp (PR #112447)

[Stanislav Mekhanoshin via cfe-commits]

rampitec wrote:

I actually wish a better way to have overloaded builtins in clang. I do not believe any user of these builtins is expecting that a wide integer will be silently truncated, and any fp will go through fptosi and backwards after, like we do now. We have much more builtins like that, which are supposed to work on a raw data and do not really take data type into account. Permute, swizzle, you name it.

https://github.com/llvm/llvm-project/pull/112447