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

[Stanislav Mekhanoshin via cfe-commits]

rampitec wrote:

Note, there is also dpp8 with the similar problem. But dpp8 is not properly handled even if intrinsic is used with a 64-bit type (i.e. not split into 2 separate 32-bit dpp ops). This would be a nice to have, but not absolutely necessary like here, because there are no 64-bit real dpp8 operations. A best we can do is to split. On the contrary some of the 64-bit operations are legal with dpp16 and we need to give an usable interface.

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