[llvm] [ARM] Fix failure to register-allocate CMP_SWAP_64 pseudo-inst (PR #104039)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> @arsenm The AMDGPU backend uses AllocationPriority so this causes a lot of test changes, because that was previously ignored for fast-regalloc. Do you think this change is OK for AMDGPU, or should I make it specific to ARM?

As far as the AMDGPU changes are concerned, it doesn't matter. However I am more concerned by the implication that these priority values are required for correctness 

https://github.com/llvm/llvm-project/pull/104039