[llvm-branch-commits] [llvm] [AMDGPU][InsertWaitCnts] Move HWEvent analysis code (PR #202887)

[Sameer Sahasrabuddhe via llvm-branch-commits]

================
@@ -25,5 +27,164 @@ void HWEventSet::dump() const {
   print(dbgs());
   dbgs() << "\n";
 }
+
+static HWEventSet getExpertSchedulingEventType(const MachineInstr &Inst,
+                                               const SIInstrInfo &TII) {
+  if (TII.isVALU(Inst)) {
+    // Core/Side-, DP-, XDL- and TRANS-MACC VALU instructions complete
+    // out-of-order with respect to each other, so each of these classes
+    // has its own event.
+
+    if (TII.isXDL(Inst))
+      return {HWEvent::VGPR_XDL_WRITE};
----------------
ssahasra wrote:

What's the motivation to keep it explicit?

https://github.com/llvm/llvm-project/pull/202887