[llvm] [AMDGPU] Skip register uses in AMDGPUResourceUsageAnalysis (PR #133242)

[Jay Foad via llvm-commits]

jayfoad wrote:

> Not a definitive answer, but I think it would be cleaner to report v0 as used.
> 
> 1. It allows to display the correct count in tools
> 2. and in case the hw behavior to always allocate at least one VGPR block changes, it still works correctly.

So you would distinguish between:
- Shader reads v0 which was initialized by hardware. This counts as "used".
- Shader reads an undefined value from v0 which was not initialized by hardware. This does not count as "used".

I guess we can do that. It just seems like extra effort in the compiler for (I claim) no tangible benefit.

https://github.com/llvm/llvm-project/pull/133242