[llvm] [AMDGPU] Move EU and Wave Queries into TargetParser (PR #215681)

[Shilei Tian via llvm-commits]

================
@@ -194,6 +194,20 @@ LLVM_ABI unsigned getAddressableNumSGPRs(Triple::SubArchType SubArch);
 LLVM_ABI unsigned getSGPRAllocGranule(GPUKind AK);
 LLVM_ABI unsigned getSGPRAllocGranule(Triple::SubArchType SubArch);
 
+/// \returns Number of SIMDs a work-group's waves run on for the target and
+/// full-SIMD mode setting.
+LLVM_ABI unsigned getWorkGroupSIMDs(GPUKind AK, bool FullSIMDMode);
----------------
shiltian wrote:

Sorry for dropping the ball. Yeah, that's right. I was thinking that we should use a HW concept here instead of a SW one. Something like `getNumWorkGroupSIMDs`/`getNumWGSIMDs`, or whatever similar variant would work.

"Workgroup" or "WorkGroup", I don't know. This is an English question. :-) It's your call. I personally just think "WorkGroup" looks better, but that's just my personal opinion.

https://github.com/llvm/llvm-project/pull/215681