diff --git a/include/llvm/CodeGen/MachineScheduler.h b/include/llvm/CodeGen/MachineScheduler.h
index acc8ef5abff..6003b1b6b82 100644
--- a/include/llvm/CodeGen/MachineScheduler.h
+++ b/include/llvm/CodeGen/MachineScheduler.h
@@ -466,6 +466,9 @@ public:
   PressureDiff &getPressureDiff(const SUnit *SU) {
     return SUPressureDiffs[SU->NodeNum];
   }
+  const PressureDiff &getPressureDiff(const SUnit *SU) const {
+    return SUPressureDiffs[SU->NodeNum];
+  }
 
   /// Compute a DFSResult after DAG building is complete, and before any
   /// queue comparisons.
@@ -491,6 +494,8 @@ public:
   /// Compute the cyclic critical path through the DAG.
   unsigned computeCyclicCriticalPath();
 
+  void dump() const override;
+
 protected:
   // Top-Level entry points for the schedule() driver...
 
diff --git a/include/llvm/CodeGen/ScheduleDAG.h b/include/llvm/CodeGen/ScheduleDAG.h
index 56adc2e2fbf..f2b072768b2 100644
--- a/include/llvm/CodeGen/ScheduleDAG.h
+++ b/include/llvm/CodeGen/ScheduleDAG.h
@@ -236,8 +236,7 @@ class TargetRegisterInfo;
       Contents.Reg = Reg;
     }
 
-    raw_ostream &print(raw_ostream &O,
-                       const TargetRegisterInfo *TRI = nullptr) const;
+    void dump(const TargetRegisterInfo *TRI = nullptr) const;
   };
 
   template <>
@@ -459,12 +458,7 @@ class TargetRegisterInfo;
     /// edge occurs first.
     void biasCriticalPath();
 
-    void dump(const ScheduleDAG *G) const;
-    void dumpAll(const ScheduleDAG *G) const;
-    raw_ostream &print(raw_ostream &O,
-                       const SUnit *Entry = nullptr,
-                       const SUnit *Exit = nullptr) const;
-    raw_ostream &print(raw_ostream &O, const ScheduleDAG *G) const;
+    void dumpAttributes() const;
 
   private:
     void ComputeDepth();
@@ -597,7 +591,9 @@ class TargetRegisterInfo;
     virtual void viewGraph(const Twine &Name, const Twine &Title);
     virtual void viewGraph();
 
-    virtual void dumpNode(const SUnit *SU) const = 0;
+    virtual void dumpNode(const SUnit &SU) const = 0;
+    virtual void dump() const = 0;
+    void dumpNodeName(const SUnit &SU) const;
 
     /// Returns a label for an SUnit node in a visualization of the ScheduleDAG.
     virtual std::string getGraphNodeLabel(const SUnit *SU) const = 0;
@@ -614,6 +610,9 @@ class TargetRegisterInfo;
     unsigned VerifyScheduledDAG(bool isBottomUp);
 #endif
 
+  protected:
+    void dumpNodeAll(const SUnit &SU) const;
+
   private:
     /// Returns the MCInstrDesc of this SDNode or NULL.
     const MCInstrDesc *getNodeDesc(const SDNode *Node) const;
diff --git a/include/llvm/CodeGen/ScheduleDAGInstrs.h b/include/llvm/CodeGen/ScheduleDAGInstrs.h
index 520a23846f6..daad18125db 100644
--- a/include/llvm/CodeGen/ScheduleDAGInstrs.h
+++ b/include/llvm/CodeGen/ScheduleDAGInstrs.h
@@ -327,7 +327,8 @@ namespace llvm {
     /// whole MachineFunction. By default does nothing.
     virtual void finalizeSchedule() {}
 
-    void dumpNode(const SUnit *SU) const override;
+    void dumpNode(const SUnit &SU) const override;
+    void dump() const override;
 
     /// Returns a label for a DAG node that points to an instruction.
     std::string getGraphNodeLabel(const SUnit *SU) const override;
diff --git a/lib/CodeGen/DFAPacketizer.cpp b/lib/CodeGen/DFAPacketizer.cpp
index cd302e78cc3..68034afe98d 100644
--- a/lib/CodeGen/DFAPacketizer.cpp
+++ b/lib/CodeGen/DFAPacketizer.cpp
@@ -250,8 +250,7 @@ void VLIWPacketizerList::PacketizeMIs(MachineBasicBlock *MBB,
 
   LLVM_DEBUG({
     dbgs() << "Scheduling DAG of the packetize region\n";
-    for (SUnit &SU : VLIWScheduler->SUnits)
-      SU.dumpAll(VLIWScheduler);
+    VLIWScheduler->dump();
   });
 
   // Generate MI -> SU map.
diff --git a/lib/CodeGen/LatencyPriorityQueue.cpp b/lib/CodeGen/LatencyPriorityQueue.cpp
index 5dbce841cfd..f9f33a98a9d 100644
--- a/lib/CodeGen/LatencyPriorityQueue.cpp
+++ b/lib/CodeGen/LatencyPriorityQueue.cpp
@@ -145,9 +145,9 @@ void LatencyPriorityQueue::remove(SUnit *SU) {
 LLVM_DUMP_METHOD void LatencyPriorityQueue::dump(ScheduleDAG *DAG) const {
   dbgs() << "Latency Priority Queue\n";
   dbgs() << "  Number of Queue Entries: " << Queue.size() << "\n";
-  for (auto const &SU : Queue) {
+  for (const SUnit *SU : Queue) {
     dbgs() << "    ";
-    SU->dump(DAG);
+    DAG->dumpNode(*SU);
   }
 }
 #endif
diff --git a/lib/CodeGen/MachinePipeliner.cpp b/lib/CodeGen/MachinePipeliner.cpp
index a02669aa4e2..988861e3c90 100644
--- a/lib/CodeGen/MachinePipeliner.cpp
+++ b/lib/CodeGen/MachinePipeliner.cpp
@@ -886,10 +886,7 @@ void SwingSchedulerDAG::schedule() {
   Topo.InitDAGTopologicalSorting();
   postprocessDAG();
   changeDependences();
-  LLVM_DEBUG({
-    for (unsigned su = 0, e = SUnits.size(); su != e; ++su)
-      SUnits[su].dumpAll(this);
-  });
+  LLVM_DEBUG(dump());
 
   NodeSetType NodeSets;
   findCircuits(NodeSets);
@@ -1638,8 +1635,8 @@ void SwingSchedulerDAG::computeNodeFunctions(NodeSetType &NodeSets) {
     for (ScheduleDAGTopologicalSort::const_iterator I = Topo.begin(),
                                                     E = Topo.end();
          I != E; ++I) {
-      SUnit *SU = &SUnits[*I];
-      SU->dump(this);
+      const SUnit &SU = SUnits[*I];
+      dumpNode(SU);
     }
   });
 
diff --git a/lib/CodeGen/MachineScheduler.cpp b/lib/CodeGen/MachineScheduler.cpp
index 7fd3d298256..da01f6f5bbe 100644
--- a/lib/CodeGen/MachineScheduler.cpp
+++ b/lib/CodeGen/MachineScheduler.cpp
@@ -633,7 +633,7 @@ void ScheduleDAGMI::releaseSucc(SUnit *SU, SDep *SuccEdge) {
 #ifndef NDEBUG
   if (SuccSU->NumPredsLeft == 0) {
     dbgs() << "*** Scheduling failed! ***\n";
-    SuccSU->dump(this);
+    dumpNode(*SuccSU);
     dbgs() << " has been released too many times!\n";
     llvm_unreachable(nullptr);
   }
@@ -670,7 +670,7 @@ void ScheduleDAGMI::releasePred(SUnit *SU, SDep *PredEdge) {
 #ifndef NDEBUG
   if (PredSU->NumSuccsLeft == 0) {
     dbgs() << "*** Scheduling failed! ***\n";
-    PredSU->dump(this);
+    dumpNode(*PredSU);
     dbgs() << " has been released too many times!\n";
     llvm_unreachable(nullptr);
   }
@@ -764,10 +764,7 @@ void ScheduleDAGMI::schedule() {
   SmallVector<SUnit*, 8> TopRoots, BotRoots;
   findRootsAndBiasEdges(TopRoots, BotRoots);
 
-  LLVM_DEBUG(if (EntrySU.getInstr() != nullptr) EntrySU.dumpAll(this);
-             for (const SUnit &SU
-                  : SUnits) SU.dumpAll(this);
-             if (ExitSU.getInstr() != nullptr) ExitSU.dumpAll(this););
+  LLVM_DEBUG(dump());
   if (ViewMISchedDAGs) viewGraph();
 
   // Initialize the strategy before modifying the DAG.
@@ -920,7 +917,7 @@ void ScheduleDAGMI::placeDebugValues() {
 LLVM_DUMP_METHOD void ScheduleDAGMI::dumpSchedule() const {
   for (MachineBasicBlock::iterator MI = begin(), ME = end(); MI != ME; ++MI) {
     if (SUnit *SU = getSUnit(&(*MI)))
-      SU->dump(this);
+      dumpNode(*SU);
     else
       dbgs() << "Missing SUnit\n";
   }
@@ -1171,6 +1168,29 @@ void ScheduleDAGMILive::updatePressureDiffs(
   }
 }
 
+void ScheduleDAGMILive::dump() const {
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
+  if (EntrySU.getInstr() != nullptr)
+    dumpNodeAll(EntrySU);
+  for (const SUnit &SU : SUnits) {
+    dumpNodeAll(SU);
+    if (ShouldTrackPressure) {
+      dbgs() << "  Pressure Diff      : ";
+      getPressureDiff(&SU).dump(*TRI);
+    }
+    dbgs() << "  Single Issue       : ";
+    if (SchedModel.mustBeginGroup(SU.getInstr()) &&
+        SchedModel.mustEndGroup(SU.getInstr()))
+      dbgs() << "true;";
+    else
+      dbgs() << "false;";
+    dbgs() << '\n';
+  }
+  if (ExitSU.getInstr() != nullptr)
+    dumpNodeAll(ExitSU);
+#endif
+}
+
 /// schedule - Called back from MachineScheduler::runOnMachineFunction
 /// after setting up the current scheduling region. [RegionBegin, RegionEnd)
 /// only includes instructions that have DAG nodes, not scheduling boundaries.
@@ -1197,22 +1217,7 @@ void ScheduleDAGMILive::schedule() {
   // This may initialize a DFSResult to be used for queue priority.
   SchedImpl->initialize(this);
 
-  LLVM_DEBUG(if (EntrySU.getInstr() != nullptr) EntrySU.dumpAll(this);
-             for (const SUnit &SU
-                  : SUnits) {
-               SU.dumpAll(this);
-               if (ShouldTrackPressure) {
-                 dbgs() << "  Pressure Diff      : ";
-                 getPressureDiff(&SU).dump(*TRI);
-               }
-               dbgs() << "  Single Issue       : ";
-               if (SchedModel.mustBeginGroup(SU.getInstr()) &&
-                   SchedModel.mustEndGroup(SU.getInstr()))
-                 dbgs() << "true;";
-               else
-                 dbgs() << "false;";
-               dbgs() << '\n';
-             } if (ExitSU.getInstr() != nullptr) ExitSU.dumpAll(this););
+  LLVM_DEBUG(dump());
   if (ViewMISchedDAGs) viewGraph();
 
   // Initialize ready queues now that the DAG and priority data are finalized.
@@ -3186,7 +3191,7 @@ void GenericScheduler::reschedulePhysRegCopies(SUnit *SU, bool isTop) {
     if (!Copy->isCopy())
       continue;
     LLVM_DEBUG(dbgs() << "  Rescheduling physreg copy ";
-               Dep.getSUnit()->dump(DAG));
+               DAG->dumpNode(*Dep.getSUnit()));
     DAG->moveInstruction(Copy, InsertPos);
   }
 }
diff --git a/lib/CodeGen/MacroFusion.cpp b/lib/CodeGen/MacroFusion.cpp
index 62dadbba0c1..82b6d642c73 100644
--- a/lib/CodeGen/MacroFusion.cpp
+++ b/lib/CodeGen/MacroFusion.cpp
@@ -67,8 +67,8 @@ static bool fuseInstructionPair(ScheduleDAGMI &DAG, SUnit &FirstSU,
       SI.setLatency(0);
 
   LLVM_DEBUG(
-      dbgs() << "Macro fuse: "; FirstSU.print(dbgs(), &DAG); dbgs() << " - ";
-      SecondSU.print(dbgs(), &DAG); dbgs() << " /  ";
+      dbgs() << "Macro fuse: "; DAG.dumpNodeName(FirstSU); dbgs() << " - ";
+      DAG.dumpNodeName(SecondSU); dbgs() << " /  ";
       dbgs() << DAG.TII->getName(FirstSU.getInstr()->getOpcode()) << " - "
              << DAG.TII->getName(SecondSU.getInstr()->getOpcode()) << '\n';);
 
@@ -80,8 +80,8 @@ static bool fuseInstructionPair(ScheduleDAGMI &DAG, SUnit &FirstSU,
       if (SI.isWeak() || isHazard(SI) ||
           SU == &DAG.ExitSU || SU == &SecondSU || SU->isPred(&SecondSU))
         continue;
-      LLVM_DEBUG(dbgs() << "  Bind "; SecondSU.print(dbgs(), &DAG);
-                 dbgs() << " - "; SU->print(dbgs(), &DAG); dbgs() << '\n';);
+      LLVM_DEBUG(dbgs() << "  Bind "; DAG.dumpNodeName(SecondSU);
+                 dbgs() << " - "; DAG.dumpNodeName(*SU); dbgs() << '\n';);
       DAG.addEdge(SU, SDep(&SecondSU, SDep::Artificial));
     }
 
@@ -92,8 +92,8 @@ static bool fuseInstructionPair(ScheduleDAGMI &DAG, SUnit &FirstSU,
       SUnit *SU = SI.getSUnit();
       if (SI.isWeak() || isHazard(SI) || &FirstSU == SU || FirstSU.isSucc(SU))
         continue;
-      LLVM_DEBUG(dbgs() << "  Bind "; SU->print(dbgs(), &DAG); dbgs() << " - ";
-                 FirstSU.print(dbgs(), &DAG); dbgs() << '\n';);
+      LLVM_DEBUG(dbgs() << "  Bind "; DAG.dumpNodeName(*SU); dbgs() << " - ";
+                 DAG.dumpNodeName(FirstSU); dbgs() << '\n';);
       DAG.addEdge(&FirstSU, SDep(SU, SDep::Artificial));
     }
     // ExitSU comes last by design, which acts like an implicit dependency
diff --git a/lib/CodeGen/PostRASchedulerList.cpp b/lib/CodeGen/PostRASchedulerList.cpp
index 215da630caf..dd0a5fe1b39 100644
--- a/lib/CodeGen/PostRASchedulerList.cpp
+++ b/lib/CodeGen/PostRASchedulerList.cpp
@@ -256,7 +256,7 @@ void SchedulePostRATDList::exitRegion() {
 LLVM_DUMP_METHOD void SchedulePostRATDList::dumpSchedule() const {
   for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
     if (SUnit *SU = Sequence[i])
-      SU->dump(this);
+      dumpNode(*SU);
     else
       dbgs() << "**** NOOP ****\n";
   }
@@ -414,11 +414,7 @@ void SchedulePostRATDList::schedule() {
   postprocessDAG();
 
   LLVM_DEBUG(dbgs() << "********** List Scheduling **********\n");
-  LLVM_DEBUG(for (const SUnit &SU
-                  : SUnits) {
-    SU.dumpAll(this);
-    dbgs() << '\n';
-  });
+  LLVM_DEBUG(dump());
 
   AvailableQueue.initNodes(SUnits);
   ListScheduleTopDown();
@@ -465,7 +461,7 @@ void SchedulePostRATDList::ReleaseSucc(SUnit *SU, SDep *SuccEdge) {
 #ifndef NDEBUG
   if (SuccSU->NumPredsLeft == 0) {
     dbgs() << "*** Scheduling failed! ***\n";
-    SuccSU->dump(this);
+    dumpNode(*SuccSU);
     dbgs() << " has been released too many times!\n";
     llvm_unreachable(nullptr);
   }
@@ -502,7 +498,7 @@ void SchedulePostRATDList::ReleaseSuccessors(SUnit *SU) {
 /// the Available queue.
 void SchedulePostRATDList::ScheduleNodeTopDown(SUnit *SU, unsigned CurCycle) {
   LLVM_DEBUG(dbgs() << "*** Scheduling [" << CurCycle << "]: ");
-  LLVM_DEBUG(SU->dump(this));
+  LLVM_DEBUG(dumpNode(*SU));
 
   Sequence.push_back(SU);
   assert(CurCycle >= SU->getDepth() &&
diff --git a/lib/CodeGen/ScheduleDAG.cpp b/lib/CodeGen/ScheduleDAG.cpp
index 46064012d9d..6c135b3d69d 100644
--- a/lib/CodeGen/ScheduleDAG.cpp
+++ b/lib/CodeGen/ScheduleDAG.cpp
@@ -68,39 +68,36 @@ const MCInstrDesc *ScheduleDAG::getNodeDesc(const SDNode *Node) const {
   return &TII->get(Node->getMachineOpcode());
 }
 
-LLVM_DUMP_METHOD
-raw_ostream &SDep::print(raw_ostream &OS, const TargetRegisterInfo *TRI) const {
+LLVM_DUMP_METHOD void SDep::dump(const TargetRegisterInfo *TRI) const {
   switch (getKind()) {
-  case Data:   OS << "Data"; break;
-  case Anti:   OS << "Anti"; break;
-  case Output: OS << "Out "; break;
-  case Order:  OS << "Ord "; break;
+  case Data:   dbgs() << "Data"; break;
+  case Anti:   dbgs() << "Anti"; break;
+  case Output: dbgs() << "Out "; break;
+  case Order:  dbgs() << "Ord "; break;
   }
 
   switch (getKind()) {
   case Data:
-    OS << " Latency=" << getLatency();
+    dbgs() << " Latency=" << getLatency();
     if (TRI && isAssignedRegDep())
-      OS << " Reg=" << printReg(getReg(), TRI);
+      dbgs() << " Reg=" << printReg(getReg(), TRI);
     break;
   case Anti:
   case Output:
-    OS << " Latency=" << getLatency();
+    dbgs() << " Latency=" << getLatency();
     break;
   case Order:
-    OS << " Latency=" << getLatency();
+    dbgs() << " Latency=" << getLatency();
     switch(Contents.OrdKind) {
-    case Barrier:      OS << " Barrier"; break;
+    case Barrier:      dbgs() << " Barrier"; break;
     case MayAliasMem:
-    case MustAliasMem: OS << " Memory"; break;
-    case Artificial:   OS << " Artificial"; break;
-    case Weak:         OS << " Weak"; break;
-    case Cluster:      OS << " Cluster"; break;
+    case MustAliasMem: dbgs() << " Memory"; break;
+    case Artificial:   dbgs() << " Artificial"; break;
+    case Weak:         dbgs() << " Weak"; break;
+    case Cluster:      dbgs() << " Cluster"; break;
     }
     break;
   }
-
-  return OS;
 }
 
 bool SUnit::addPred(const SDep &D, bool Required) {
@@ -337,33 +334,7 @@ void SUnit::biasCriticalPath() {
 }
 
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
-LLVM_DUMP_METHOD
-raw_ostream &SUnit::print(raw_ostream &OS,
-                          const SUnit *Entry, const SUnit *Exit) const {
-  if (this == Entry)
-    OS << "EntrySU";
-  else if (this == Exit)
-    OS << "ExitSU";
-  else
-    OS << "SU(" << NodeNum << ")";
-  return OS;
-}
-
-LLVM_DUMP_METHOD
-raw_ostream &SUnit::print(raw_ostream &OS, const ScheduleDAG *G) const {
-  return print(OS, &G->EntrySU, &G->ExitSU);
-}
-
-LLVM_DUMP_METHOD
-void SUnit::dump(const ScheduleDAG *G) const {
-  print(dbgs(), G);
-  dbgs() << ": ";
-  G->dumpNode(this);
-}
-
-LLVM_DUMP_METHOD void SUnit::dumpAll(const ScheduleDAG *G) const {
-  dump(G);
-
+LLVM_DUMP_METHOD void SUnit::dumpAttributes() const {
   dbgs() << "  # preds left       : " << NumPredsLeft << "\n";
   dbgs() << "  # succs left       : " << NumSuccsLeft << "\n";
   if (WeakPredsLeft)
@@ -374,21 +345,38 @@ LLVM_DUMP_METHOD void SUnit::dumpAll(const ScheduleDAG *G) const {
   dbgs() << "  Latency            : " << Latency << "\n";
   dbgs() << "  Depth              : " << getDepth() << "\n";
   dbgs() << "  Height             : " << getHeight() << "\n";
+}
 
-  if (Preds.size() != 0) {
+LLVM_DUMP_METHOD void ScheduleDAG::dumpNodeName(const SUnit &SU) const {
+  if (&SU == &EntrySU)
+    dbgs() << "EntrySU";
+  else if (&SU == &ExitSU)
+    dbgs() << "ExitSU";
+  else
+    dbgs() << "SU(" << SU.NodeNum << ")";
+}
+
+LLVM_DUMP_METHOD void ScheduleDAG::dumpNodeAll(const SUnit &SU) const {
+  dumpNode(SU);
+  SU.dumpAttributes();
+  if (SU.Preds.size() > 0) {
     dbgs() << "  Predecessors:\n";
-    for (const SDep &Dep : Preds) {
+    for (const SDep &Dep : SU.Preds) {
       dbgs() << "    ";
-      Dep.getSUnit()->print(dbgs(), G); dbgs() << ": ";
-      Dep.print(dbgs(), G->TRI); dbgs() << '\n';
+      dumpNodeName(*Dep.getSUnit());
+      dbgs() << ": ";
+      Dep.dump(TRI);
+      dbgs() << '\n';
     }
   }
-  if (Succs.size() != 0) {
+  if (SU.Succs.size() > 0) {
     dbgs() << "  Successors:\n";
-    for (const SDep &Dep : Succs) {
+    for (const SDep &Dep : SU.Succs) {
       dbgs() << "    ";
-      Dep.getSUnit()->print(dbgs(), G); dbgs() << ": ";
-      Dep.print(dbgs(), G->TRI); dbgs() << '\n';
+      dumpNodeName(*Dep.getSUnit());
+      dbgs() << ": ";
+      Dep.dump(TRI);
+      dbgs() << '\n';
     }
   }
 }
@@ -406,7 +394,7 @@ unsigned ScheduleDAG::VerifyScheduledDAG(bool isBottomUp) {
       }
       if (!AnyNotSched)
         dbgs() << "*** Scheduling failed! ***\n";
-      SUnit.dump(this);
+      dumpNode(SUnit);
       dbgs() << "has not been scheduled!\n";
       AnyNotSched = true;
     }
@@ -415,7 +403,7 @@ unsigned ScheduleDAG::VerifyScheduledDAG(bool isBottomUp) {
           unsigned(std::numeric_limits<int>::max())) {
       if (!AnyNotSched)
         dbgs() << "*** Scheduling failed! ***\n";
-      SUnit.dump(this);
+      dumpNode(SUnit);
       dbgs() << "has an unexpected "
            << (isBottomUp ? "Height" : "Depth") << " value!\n";
       AnyNotSched = true;
@@ -424,7 +412,7 @@ unsigned ScheduleDAG::VerifyScheduledDAG(bool isBottomUp) {
       if (SUnit.NumSuccsLeft != 0) {
         if (!AnyNotSched)
           dbgs() << "*** Scheduling failed! ***\n";
-        SUnit.dump(this);
+        dumpNode(SUnit);
         dbgs() << "has successors left!\n";
         AnyNotSched = true;
       }
@@ -432,7 +420,7 @@ unsigned ScheduleDAG::VerifyScheduledDAG(bool isBottomUp) {
       if (SUnit.NumPredsLeft != 0) {
         if (!AnyNotSched)
           dbgs() << "*** Scheduling failed! ***\n";
-        SUnit.dump(this);
+        dumpNode(SUnit);
         dbgs() << "has predecessors left!\n";
         AnyNotSched = true;
       }
diff --git a/lib/CodeGen/ScheduleDAGInstrs.cpp b/lib/CodeGen/ScheduleDAGInstrs.cpp
index d1c5ddabb97..5eb842eb04a 100644
--- a/lib/CodeGen/ScheduleDAGInstrs.cpp
+++ b/lib/CodeGen/ScheduleDAGInstrs.cpp
@@ -1097,10 +1097,22 @@ void ScheduleDAGInstrs::fixupKills(MachineBasicBlock &MBB) {
   }
 }
 
-void ScheduleDAGInstrs::dumpNode(const SUnit *SU) const {
-  // Cannot completely remove virtual function even in release mode.
+void ScheduleDAGInstrs::dumpNode(const SUnit &SU) const {
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
-  SU->getInstr()->dump();
+  dumpNodeName(SU);
+  dbgs() << ": ";
+  SU.getInstr()->dump();
+#endif
+}
+
+void ScheduleDAGInstrs::dump() const {
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
+  if (EntrySU.getInstr() != nullptr)
+    dumpNodeAll(EntrySU);
+  for (const SUnit &SU : SUnits)
+    dumpNodeAll(SU);
+  if (ExitSU.getInstr() != nullptr)
+    dumpNodeAll(ExitSU);
 #endif
 }
 
diff --git a/lib/CodeGen/ScoreboardHazardRecognizer.cpp b/lib/CodeGen/ScoreboardHazardRecognizer.cpp
index b8bfe69a76e..4301372179b 100644
--- a/lib/CodeGen/ScoreboardHazardRecognizer.cpp
+++ b/lib/CodeGen/ScoreboardHazardRecognizer.cpp
@@ -157,8 +157,7 @@ ScoreboardHazardRecognizer::getHazardType(SUnit *SU, int Stalls) {
 
       if (!freeUnits) {
         LLVM_DEBUG(dbgs() << "*** Hazard in cycle +" << StageCycle << ", ");
-        LLVM_DEBUG(dbgs() << "SU(" << SU->NodeNum << "): ");
-        LLVM_DEBUG(DAG->dumpNode(SU));
+        LLVM_DEBUG(DAG->dumpNode(*SU));
         return Hazard;
       }
     }
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGFast.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGFast.cpp
index 3944d7df286..4b06cfcc7f6 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGFast.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGFast.cpp
@@ -125,8 +125,7 @@ void ScheduleDAGFast::Schedule() {
   // Build the scheduling graph.
   BuildSchedGraph(nullptr);
 
-  LLVM_DEBUG(for (unsigned su = 0, e = SUnits.size(); su != e; ++su) SUnits[su]
-                 .dumpAll(this));
+  LLVM_DEBUG(dump());
 
   // Execute the actual scheduling loop.
   ListScheduleBottomUp();
@@ -144,7 +143,7 @@ void ScheduleDAGFast::ReleasePred(SUnit *SU, SDep *PredEdge) {
 #ifndef NDEBUG
   if (PredSU->NumSuccsLeft == 0) {
     dbgs() << "*** Scheduling failed! ***\n";
-    PredSU->dump(this);
+    dumpNode(*PredSU);
     dbgs() << " has been released too many times!\n";
     llvm_unreachable(nullptr);
   }
@@ -182,7 +181,7 @@ void ScheduleDAGFast::ReleasePredecessors(SUnit *SU, unsigned CurCycle) {
 /// the Available queue.
 void ScheduleDAGFast::ScheduleNodeBottomUp(SUnit *SU, unsigned CurCycle) {
   LLVM_DEBUG(dbgs() << "*** Scheduling [" << CurCycle << "]: ");
-  LLVM_DEBUG(SU->dump(this));
+  LLVM_DEBUG(dumpNode(*SU));
 
   assert(CurCycle >= SU->getHeight() && "Node scheduled below its height!");
   SU->setHeightToAtLeast(CurCycle);
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
index 43e8ffd3839..8d75b8133a3 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
@@ -365,7 +365,7 @@ void ScheduleDAGRRList::Schedule() {
   // Build the scheduling graph.
   BuildSchedGraph(nullptr);
 
-  LLVM_DEBUG(for (SUnit &SU : SUnits) SU.dumpAll(this));
+  LLVM_DEBUG(dump());
   Topo.InitDAGTopologicalSorting();
 
   AvailableQueue->initNodes(SUnits);
@@ -396,7 +396,7 @@ void ScheduleDAGRRList::ReleasePred(SUnit *SU, const SDep *PredEdge) {
 #ifndef NDEBUG
   if (PredSU->NumSuccsLeft == 0) {
     dbgs() << "*** Scheduling failed! ***\n";
-    PredSU->dump(this);
+    dumpNode(*PredSU);
     dbgs() << " has been released too many times!\n";
     llvm_unreachable(nullptr);
   }
@@ -729,7 +729,7 @@ static void resetVRegCycle(SUnit *SU);
 /// the Available queue.
 void ScheduleDAGRRList::ScheduleNodeBottomUp(SUnit *SU) {
   LLVM_DEBUG(dbgs() << "\n*** Scheduling [" << CurCycle << "]: ");
-  LLVM_DEBUG(SU->dump(this));
+  LLVM_DEBUG(dumpNode(*SU));
 
 #ifndef NDEBUG
   if (CurCycle < SU->getHeight())
@@ -828,7 +828,7 @@ void ScheduleDAGRRList::CapturePred(SDep *PredEdge) {
 /// its predecessor states to reflect the change.
 void ScheduleDAGRRList::UnscheduleNodeBottomUp(SUnit *SU) {
   LLVM_DEBUG(dbgs() << "*** Unscheduling [" << SU->getHeight() << "]: ");
-  LLVM_DEBUG(SU->dump(this));
+  LLVM_DEBUG(dumpNode(*SU));
 
   for (SDep &Pred : SU->Preds) {
     CapturePred(&Pred);
@@ -1130,7 +1130,7 @@ SUnit *ScheduleDAGRRList::CopyAndMoveSuccessors(SUnit *SU) {
     return nullptr;
 
   LLVM_DEBUG(dbgs() << "Considering duplicating the SU\n");
-  LLVM_DEBUG(SU->dump(this));
+  LLVM_DEBUG(dumpNode(*SU));
 
   if (N->getGluedNode() &&
       !TII->canCopyGluedNodeDuringSchedule(N)) {
@@ -1888,7 +1888,7 @@ public:
     while (!DumpQueue.empty()) {
       SUnit *SU = popFromQueue(DumpQueue, DumpPicker, scheduleDAG);
       dbgs() << "Height " << SU->getHeight() << ": ";
-      SU->dump(DAG);
+      DAG->dumpNode(*SU);
     }
   }
 #endif
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
index 6563d195e1f..90bc241ee5a 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp
@@ -648,18 +648,20 @@ void ScheduleDAGSDNodes::computeOperandLatency(SDNode *Def, SDNode *Use,
     dep.setLatency(Latency);
 }
 
-void ScheduleDAGSDNodes::dumpNode(const SUnit *SU) const {
-  // Cannot completely remove virtual function even in release mode.
+void ScheduleDAGSDNodes::dumpNode(const SUnit &SU) const {
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
-  if (!SU->getNode()) {
+  dumpNodeName(SU);
+  dbgs() << ": ";
+
+  if (!SU.getNode()) {
     dbgs() << "PHYS REG COPY\n";
     return;
   }
 
-  SU->getNode()->dump(DAG);
+  SU.getNode()->dump(DAG);
   dbgs() << "\n";
   SmallVector<SDNode *, 4> GluedNodes;
-  for (SDNode *N = SU->getNode()->getGluedNode(); N; N = N->getGluedNode())
+  for (SDNode *N = SU.getNode()->getGluedNode(); N; N = N->getGluedNode())
     GluedNodes.push_back(N);
   while (!GluedNodes.empty()) {
     dbgs() << "    ";
@@ -670,11 +672,22 @@ void ScheduleDAGSDNodes::dumpNode(const SUnit *SU) const {
 #endif
 }
 
+void ScheduleDAGSDNodes::dump() const {
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
+  if (EntrySU.getNode() != nullptr)
+    dumpNodeAll(EntrySU);
+  for (const SUnit &SU : SUnits)
+    dumpNodeAll(SU);
+  if (ExitSU.getNode() != nullptr)
+    dumpNodeAll(ExitSU);
+#endif
+}
+
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 void ScheduleDAGSDNodes::dumpSchedule() const {
   for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
     if (SUnit *SU = Sequence[i])
-      SU->dump(this);
+      dumpNode(*SU);
     else
       dbgs() << "**** NOOP ****\n";
   }
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.h b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.h
index 6417e16bd0f..3fa7ad89572 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.h
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.h
@@ -122,8 +122,8 @@ class InstrItineraryData;
     virtual MachineBasicBlock*
     EmitSchedule(MachineBasicBlock::iterator &InsertPos);
 
-    void dumpNode(const SUnit *SU) const override;
-
+    void dumpNode(const SUnit &SU) const override;
+    void dump() const override;
     void dumpSchedule() const;
 
     std::string getGraphNodeLabel(const SUnit *SU) const override;
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGVLIW.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGVLIW.cpp
index 84055f8ecc1..416061475b1 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGVLIW.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGVLIW.cpp
@@ -118,7 +118,7 @@ void ScheduleDAGVLIW::releaseSucc(SUnit *SU, const SDep &D) {
 #ifndef NDEBUG
   if (SuccSU->NumPredsLeft == 0) {
     dbgs() << "*** Scheduling failed! ***\n";
-    SuccSU->dump(this);
+    dumpNode(*SuccSU);
     dbgs() << " has been released too many times!\n";
     llvm_unreachable(nullptr);
   }
@@ -152,7 +152,7 @@ void ScheduleDAGVLIW::releaseSuccessors(SUnit *SU) {
 /// the Available queue.
 void ScheduleDAGVLIW::scheduleNodeTopDown(SUnit *SU, unsigned CurCycle) {
   LLVM_DEBUG(dbgs() << "*** Scheduling [" << CurCycle << "]: ");
-  LLVM_DEBUG(SU->dump(this));
+  LLVM_DEBUG(dumpNode(*SU));
 
   Sequence.push_back(SU);
   assert(CurCycle >= SU->getDepth() && "Node scheduled above its depth!");
diff --git a/lib/Target/AMDGPU/GCNILPSched.cpp b/lib/Target/AMDGPU/GCNILPSched.cpp
index 651091d4413..d62dc8d8678 100644
--- a/lib/Target/AMDGPU/GCNILPSched.cpp
+++ b/lib/Target/AMDGPU/GCNILPSched.cpp
@@ -335,7 +335,7 @@ GCNILPScheduler::schedule(ArrayRef<const SUnit*> BotRoots,
     assert(C);
     AvailQueue.remove(*C);
     auto SU = C->SU;
-    LLVM_DEBUG(dbgs() << "Selected "; SU->dump(&DAG));
+    LLVM_DEBUG(dbgs() << "Selected "; DAG.dumpNode(*SU));
 
     advanceToCycle(SU->getHeight());
 
diff --git a/lib/Target/AMDGPU/GCNMinRegStrategy.cpp b/lib/Target/AMDGPU/GCNMinRegStrategy.cpp
index 192d534bb9c..ec6bcae3355 100644
--- a/lib/Target/AMDGPU/GCNMinRegStrategy.cpp
+++ b/lib/Target/AMDGPU/GCNMinRegStrategy.cpp
@@ -258,7 +258,7 @@ GCNMinRegScheduler::schedule(ArrayRef<const SUnit*> TopRoots,
     assert(C);
     RQ.remove(*C);
     auto SU = C->SU;
-    LLVM_DEBUG(dbgs() << "Selected "; SU->dump(&DAG));
+    LLVM_DEBUG(dbgs() << "Selected "; DAG.dumpNode(*SU));
 
     releaseSuccessors(SU, StepNo);
     Schedule.push_back(SU);
diff --git a/lib/Target/AMDGPU/R600MachineScheduler.cpp b/lib/Target/AMDGPU/R600MachineScheduler.cpp
index a1429a2ac50..478a473a51b 100644
--- a/lib/Target/AMDGPU/R600MachineScheduler.cpp
+++ b/lib/Target/AMDGPU/R600MachineScheduler.cpp
@@ -127,13 +127,13 @@ SUnit* R600SchedStrategy::pickNode(bool &IsTopNode) {
 
   LLVM_DEBUG(if (SU) {
     dbgs() << " ** Pick node **\n";
-    SU->dump(DAG);
+    DAG->dumpNode(*SU);
   } else {
     dbgs() << "NO NODE \n";
     for (unsigned i = 0; i < DAG->SUnits.size(); i++) {
       const SUnit &S = DAG->SUnits[i];
       if (!S.isScheduled)
-        S.dump(DAG);
+        DAG->dumpNode(S);
     }
   });
 
@@ -188,11 +188,11 @@ isPhysicalRegCopy(MachineInstr *MI) {
 }
 
 void R600SchedStrategy::releaseTopNode(SUnit *SU) {
-  LLVM_DEBUG(dbgs() << "Top Releasing "; SU->dump(DAG););
+  LLVM_DEBUG(dbgs() << "Top Releasing "; DAG->dumpNode(*SU));
 }
 
 void R600SchedStrategy::releaseBottomNode(SUnit *SU) {
-  LLVM_DEBUG(dbgs() << "Bottom Releasing "; SU->dump(DAG););
+  LLVM_DEBUG(dbgs() << "Bottom Releasing "; DAG->dumpNode(*SU));
   if (isPhysicalRegCopy(SU->getInstr())) {
     PhysicalRegCopy.push_back(SU);
     return;
diff --git a/lib/Target/AMDGPU/SIMachineScheduler.cpp b/lib/Target/AMDGPU/SIMachineScheduler.cpp
index 18754442898..6670def7d09 100644
--- a/lib/Target/AMDGPU/SIMachineScheduler.cpp
+++ b/lib/Target/AMDGPU/SIMachineScheduler.cpp
@@ -471,7 +471,7 @@ void SIScheduleBlock::releaseSucc(SUnit *SU, SDep *SuccEdge) {
 #ifndef NDEBUG
   if (SuccSU->NumPredsLeft == 0) {
     dbgs() << "*** Scheduling failed! ***\n";
-    SuccSU->dump(DAG);
+    DAG->dumpNode(*SuccSU);
     dbgs() << " has been released too many times!\n";
     llvm_unreachable(nullptr);
   }
@@ -611,13 +611,11 @@ void SIScheduleBlock::printDebug(bool full) {
 
   dbgs() << "\nInstructions:\n";
   if (!Scheduled) {
-    for (SUnit* SU : SUnits) {
-      SU->dump(DAG);
-    }
+    for (const SUnit* SU : SUnits)
+      DAG->dumpNode(*SU);
   } else {
-    for (SUnit* SU : SUnits) {
-      SU->dump(DAG);
-    }
+    for (const SUnit* SU : SUnits)
+      DAG->dumpNode(*SU);
   }
 
   dbgs() << "///////////////////////\n";
@@ -1933,7 +1931,7 @@ void SIScheduleDAGMI::schedule()
   LLVM_DEBUG(dbgs() << "Preparing Scheduling\n");
 
   buildDAGWithRegPressure();
-  LLVM_DEBUG(for (SUnit &SU : SUnits) SU.dumpAll(this));
+  LLVM_DEBUG(dump());
 
   topologicalSort();
   findRootsAndBiasEdges(TopRoots, BotRoots);
diff --git a/lib/Target/Hexagon/HexagonMachineScheduler.cpp b/lib/Target/Hexagon/HexagonMachineScheduler.cpp
index 74c550ce822..ebfe21bd17d 100644
--- a/lib/Target/Hexagon/HexagonMachineScheduler.cpp
+++ b/lib/Target/Hexagon/HexagonMachineScheduler.cpp
@@ -215,8 +215,7 @@ void VLIWMachineScheduler::schedule() {
                   ++su) if (SUnits[su].getDepth() > maxD) maxD =
                  SUnits[su].getDepth();
              dbgs() << "Max Depth " << maxD << "\n";);
-  LLVM_DEBUG(for (unsigned su = 0, e = SUnits.size(); su != e; ++su) SUnits[su]
-                 .dumpAll(this));
+  LLVM_DEBUG(dump());
 
   initQueues(TopRoots, BotRoots);
 
@@ -489,7 +488,7 @@ void ConvergingVLIWScheduler::traceCandidate(const char *Label,
   else
     dbgs() << "     ";
   dbgs() << "cost(" << Cost << ")\t";
-  SU->dump(DAG);
+  DAG->dumpNode(*SU);
 }
 
 // Very detailed queue dump, to be used with higher verbosity levels.
@@ -982,7 +981,7 @@ SUnit *ConvergingVLIWScheduler::pickNode(bool &IsTopNode) {
                     << " Scheduling instruction in cycle "
                     << (IsTopNode ? Top.CurrCycle : Bot.CurrCycle) << " ("
                     << reportPackets() << ")\n";
-             SU->dump(DAG));
+             DAG->dumpNode(*SU));
   return SU;
 }
 
diff --git a/lib/Target/PowerPC/PPCHazardRecognizers.cpp b/lib/Target/PowerPC/PPCHazardRecognizers.cpp
index 793a4dd7f62..cf3547f3cfb 100644
--- a/lib/Target/PowerPC/PPCHazardRecognizers.cpp
+++ b/lib/Target/PowerPC/PPCHazardRecognizers.cpp
@@ -180,9 +180,8 @@ void PPCDispatchGroupSBHazardRecognizer::EmitInstruction(SUnit *SU) {
       CurGroup.clear();
       CurSlots = CurBranches = 0;
     } else {
-      LLVM_DEBUG(dbgs() << "**** Adding to dispatch group: SU(" << SU->NodeNum
-                        << "): ");
-      LLVM_DEBUG(DAG->dumpNode(SU));
+      LLVM_DEBUG(dbgs() << "**** Adding to dispatch group: ");
+      LLVM_DEBUG(DAG->dumpNode(*SU));
 
       unsigned NSlots;
       bool MustBeFirst = mustComeFirst(MCID, NSlots);