[llvm] [AMDGPU][SIInsertWaitcnts] Cleanup: Remove WaitEventMaskForInst member variable (PR #178030)

[via llvm-commits]

================
@@ -652,6 +651,7 @@ class SIInsertWaitcnts {
                          bool ExpertMode) const;
   AtomicRMWState getAtomicRMWState(MachineInstr &MI,
                                    AtomicRMWState PrevState) const;
+  const WaitcntGenerator &getWCG() const { return *WCG; }
----------------
vporpo wrote:

It is indeed annoying that we write getWCG() everywhere, but I think I prefer it because it shows that the mask is target dependent.
I guess we could add WCG in the function name to convey that information too, like `const unsigned *getWCGWaitEventMask() const { return WCG->getWaitEventMask(); }` ? Wdyt ?

https://github.com/llvm/llvm-project/pull/178030