[llvm] [AMDGPU] Examine instructions in pending queues during scheduling (PR #147653)

[Austin Kerbow via llvm-commits]

================
@@ -450,35 +534,46 @@ SUnit *GCNSchedStrategy::pickNode(bool &IsTopNode) {
            Bot.Available.empty() && Bot.Pending.empty() && "ReadyQ garbage");
     return nullptr;
   }
+  bool PickedPending;
   SUnit *SU;
   do {
+    PickedPending = false;
     if (RegionPolicy.OnlyTopDown) {
-      SU = Top.pickOnlyChoice();
+      SU = pickOnlyChoice(Top, SchedModel);
       if (!SU) {
         CandPolicy NoPolicy;
         TopCand.reset(NoPolicy);
         pickNodeFromQueue(Top, NoPolicy, DAG->getTopRPTracker(), TopCand,
+                          PickedPending,
                           /*IsBottomUp=*/false);
         assert(TopCand.Reason != NoCand && "failed to find a candidate");
         SU = TopCand.SU;
       }
       IsTopNode = true;
     } else if (RegionPolicy.OnlyBottomUp) {
-      SU = Bot.pickOnlyChoice();
+      SU = pickOnlyChoice(Bot, SchedModel);
       if (!SU) {
         CandPolicy NoPolicy;
         BotCand.reset(NoPolicy);
         pickNodeFromQueue(Bot, NoPolicy, DAG->getBotRPTracker(), BotCand,
+                          PickedPending,
                           /*IsBottomUp=*/true);
         assert(BotCand.Reason != NoCand && "failed to find a candidate");
         SU = BotCand.SU;
       }
       IsTopNode = false;
     } else {
-      SU = pickNodeBidirectional(IsTopNode);
+      SU = pickNodeBidirectional(IsTopNode, PickedPending);
     }
   } while (SU->isScheduled);
 
+  if (PickedPending) {
+    unsigned ReadyCycle = IsTopNode ? SU->TopReadyCycle : SU->BotReadyCycle;
+    SchedBoundary &Zone = IsTopNode ? Top : Bot;
+    Zone.bumpCycle(ReadyCycle);
----------------
kerbowa wrote:

Fixed. Thanks for finding this. There were two issues with the previous implementation.

1. The scheduler doesn't keep SU's next ready cycle in sync with possible hazards, and `checkHazard(SU)` needs to be queried to confirm whether the instruction can be issued in the current cycle. There is not currently a mechanism to identify exactly which cycle the hazard will resolve.
2. There is additional bookkeeping in the base implementation of `pickOnlyChoice()` that really should exist in other functions, but for our purposes it makes the most sense to always call `pickOnlyChoice()` before selecting any candidate SU. This matches the behavior of the generic scheduler.

https://github.com/llvm/llvm-project/pull/147653