[llvm] [AMDGPU] Ignore inactive VGPRs in .vgpr_count (PR #144855)

[Diana Picus via llvm-commits]

rovka wrote:

I updated this so we always skip calls and IMPLICIT_DEFs. This is because the entry point for our use case does not use init.whole.wave, and we still want it to ignore any inactive VGPRs passed to the chain call. 

It's also more generic that way and that seems to be preferred :) 

https://github.com/llvm/llvm-project/pull/144855