[llvm] [MachinePipeliner] Introduce a new class for loop-carried deps (PR #137663)

[Ryotaro Kasuga via llvm-commits]

================
@@ -941,28 +1014,116 @@ static bool hasLoopCarriedMemDep(const SUnitWithMemInfo &Src,
   return false;
 }
 
+void LoopCarriedOrderDepsTracker::LoadStoreChunk::append(SUnit *SU) {
+  const MachineInstr *MI = SU->getInstr();
+  if (!MI->mayLoadOrStore())
+    return;
+  (MI->mayStore() ? Stores : Loads).emplace_back(SU);
+}
+
+LoopCarriedOrderDepsTracker::LoopCarriedOrderDepsTracker(
+    SwingSchedulerDAG *SSD, BatchAAResults *BAA, const TargetInstrInfo *TII,
+    const TargetRegisterInfo *TRI)
+    : DAG(SSD), BAA(BAA), SUnits(DAG->SUnits), N(SUnits.size()),
+      LoopCarried(N, BitVector(N)), TII(TII), TRI(TRI) {}
+
+void LoopCarriedOrderDepsTracker::computeDependencies() {
+  // Traverse all instructions and extract only what we are targetting.
+  for (auto &SU : SUnits) {
+    auto Tagged = checkInstrType(&SU);
+
+    // This instruction has no loop-carried order-dependencies.
+    if (!Tagged)
+      continue;
+    TaggedSUnits.push_back(*Tagged);
+  }
+
+  computeDependenciesAux();
+
+  LLVM_DEBUG({
+    for (unsigned I = 0; I != N; I++)
+      assert(!LoopCarried[I].test(I) && "Unexpected self-loop");
+  });
----------------
kasuga-fj wrote:

Correct, deleted (this can be here, but it’s not very meaningful).

https://github.com/llvm/llvm-project/pull/137663