[llvm] [AMDGPU] Have VCC as a first-class member of the SGPR pool. (PR #173870)

[Matt Arsenault via llvm-commits]

arsenm wrote:

I think there are hazards here, and changing the core class is probably the wrong solution. Can you instead change the definitions of the relevant instructions on relevant targets to use a larger class?

I recall on older targets there were additional hazards when accessing the numbered SGPRs aliasing VCC through the numbered names, and we deliberately avoided ever referring to VCC through them. Is this a new property on gfx10+ that you can just intermix these? 

https://github.com/llvm/llvm-project/pull/173870