[llvm] [AMDGPU] Don't DEALLOC_VGPRS from callable functions (PR #72245)

[Diana Picus via llvm-commits]

rovka wrote:

> > Wasn't there also the issue that we shouldn't send the message if the current function contains calls, because a callee might have accessed scratch? Is this handled elsewhere?
> 
> It will be handled in a different patch. "Coming soon"

And finally: https://github.com/llvm/llvm-project/pull/77439

https://github.com/llvm/llvm-project/pull/72245