[llvm] [AMDGPU] Move `getVGPREncodingGranule` to TargetParser (PR #226817)

[Shilei Tian via llvm-commits]

================
@@ -438,6 +438,20 @@ unsigned AMDGPU::getVGPRAllocGranule(Triple::SubArchType SubArch,
   return getVGPRAllocGranule(getGPUKindFromSubArch(SubArch), IsWave32);
 }
 
+unsigned AMDGPU::getVGPREncodingGranule(GPUKind AK, bool IsWave32) {
----------------
shiltian wrote:

I wonder whether it makes more sense to make this an integer target feature? On targets with wave32/wave64, it is just divided by 2 or not.

https://github.com/llvm/llvm-project/pull/226817