[llvm] [AMDGPU] Introduce GCNPostGenericScheduler to model MFMA-VALU coexecution (PR #184084)

[Frederik Harwath via llvm-commits]

================
@@ -802,6 +802,25 @@ class MemoryClauseInitialScheduleStage : public GCNSchedStage {
       : GCNSchedStage(StageID, DAG) {}
 };
 
+class GCNPostGenericScheduler : public PostGenericScheduler {
+protected:
+  bool LastTopScheduledIsMFMA = false;
+  bool LastBottomScheduledIsMFMA = false;
+
+public:
+  GCNPostGenericScheduler(const MachineSchedContext *C)
+      : PostGenericScheduler(C) {}
+
+  void initPolicy(MachineBasicBlock::iterator Begin,
+                  MachineBasicBlock::iterator End,
+                  unsigned NumRegionInstrs) override;
+
+  void schedNode(SUnit *SU, bool IsTopNode) override;
+
+  bool isLastTopScheduledIsMFMA() const { return LastTopScheduledIsMFMA; }
+  bool isLastBottomScheduledIsMFMA() const { return LastBottomScheduledIsMFMA; }
----------------
frederik-h wrote:

I think you should just use the variables directly since there are only internal uses within the class. If you prefer to keep them, I think they shouldn't be public and they could be renamed to, for instance,  `lastTopScheduledIsMFMA` or `isLastTopScheduledMFMA` if you prefer to keep the `is` prefix.

https://github.com/llvm/llvm-project/pull/184084