[llvm] [AMDGPU] Report only local per-function resource usage when object linking is enabled (PR #192594)

[Shilei Tian via llvm-commits]

shiltian wrote:

> Probably should just do this unconditionally?

We can't, because these symbols are used by kernel descriptor and HSA metadata. If linker doesn't patch them later, we will have wrong numbers.

https://github.com/llvm/llvm-project/pull/192594