xref: /freebsd/contrib/llvm-project/llvm/lib/CodeGen/SelectionDAG/ScheduleDAGFast.cpp (revision e6bfd18d21b225af6a0ed67ceeaf1293b7b9eba5)
1 //===----- ScheduleDAGFast.cpp - Fast poor list scheduler -----------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This implements a fast scheduler.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "InstrEmitter.h"
14 #include "SDNodeDbgValue.h"
15 #include "ScheduleDAGSDNodes.h"
16 #include "llvm/ADT/SmallSet.h"
17 #include "llvm/ADT/Statistic.h"
18 #include "llvm/CodeGen/SchedulerRegistry.h"
19 #include "llvm/CodeGen/SelectionDAGISel.h"
20 #include "llvm/CodeGen/TargetInstrInfo.h"
21 #include "llvm/CodeGen/TargetRegisterInfo.h"
22 #include "llvm/IR/InlineAsm.h"
23 #include "llvm/Support/Debug.h"
24 #include "llvm/Support/ErrorHandling.h"
25 #include "llvm/Support/raw_ostream.h"
26 using namespace llvm;
27 
28 #define DEBUG_TYPE "pre-RA-sched"
29 
30 STATISTIC(NumUnfolds,    "Number of nodes unfolded");
31 STATISTIC(NumDups,       "Number of duplicated nodes");
32 STATISTIC(NumPRCopies,   "Number of physical copies");
33 
34 static RegisterScheduler
35   fastDAGScheduler("fast", "Fast suboptimal list scheduling",
36                    createFastDAGScheduler);
37 static RegisterScheduler
38   linearizeDAGScheduler("linearize", "Linearize DAG, no scheduling",
39                         createDAGLinearizer);
40 
41 
42 namespace {
43   /// FastPriorityQueue - A degenerate priority queue that considers
44   /// all nodes to have the same priority.
45   ///
46   struct FastPriorityQueue {
47     SmallVector<SUnit *, 16> Queue;
48 
49     bool empty() const { return Queue.empty(); }
50 
51     void push(SUnit *U) {
52       Queue.push_back(U);
53     }
54 
55     SUnit *pop() {
56       if (empty()) return nullptr;
57       return Queue.pop_back_val();
58     }
59   };
60 
61 //===----------------------------------------------------------------------===//
62 /// ScheduleDAGFast - The actual "fast" list scheduler implementation.
63 ///
64 class ScheduleDAGFast : public ScheduleDAGSDNodes {
65 private:
66   /// AvailableQueue - The priority queue to use for the available SUnits.
67   FastPriorityQueue AvailableQueue;
68 
69   /// LiveRegDefs - A set of physical registers and their definition
70   /// that are "live". These nodes must be scheduled before any other nodes that
71   /// modifies the registers can be scheduled.
72   unsigned NumLiveRegs;
73   std::vector<SUnit*> LiveRegDefs;
74   std::vector<unsigned> LiveRegCycles;
75 
76 public:
77   ScheduleDAGFast(MachineFunction &mf)
78     : ScheduleDAGSDNodes(mf) {}
79 
80   void Schedule() override;
81 
82   /// AddPred - adds a predecessor edge to SUnit SU.
83   /// This returns true if this is a new predecessor.
84   void AddPred(SUnit *SU, const SDep &D) {
85     SU->addPred(D);
86   }
87 
88   /// RemovePred - removes a predecessor edge from SUnit SU.
89   /// This returns true if an edge was removed.
90   void RemovePred(SUnit *SU, const SDep &D) {
91     SU->removePred(D);
92   }
93 
94 private:
95   void ReleasePred(SUnit *SU, SDep *PredEdge);
96   void ReleasePredecessors(SUnit *SU, unsigned CurCycle);
97   void ScheduleNodeBottomUp(SUnit*, unsigned);
98   SUnit *CopyAndMoveSuccessors(SUnit*);
99   void InsertCopiesAndMoveSuccs(SUnit*, unsigned,
100                                 const TargetRegisterClass*,
101                                 const TargetRegisterClass*,
102                                 SmallVectorImpl<SUnit*>&);
103   bool DelayForLiveRegsBottomUp(SUnit*, SmallVectorImpl<unsigned>&);
104   void ListScheduleBottomUp();
105 
106   /// forceUnitLatencies - The fast scheduler doesn't care about real latencies.
107   bool forceUnitLatencies() const override { return true; }
108 };
109 }  // end anonymous namespace
110 
111 
112 /// Schedule - Schedule the DAG using list scheduling.
113 void ScheduleDAGFast::Schedule() {
114   LLVM_DEBUG(dbgs() << "********** List Scheduling **********\n");
115 
116   NumLiveRegs = 0;
117   LiveRegDefs.resize(TRI->getNumRegs(), nullptr);
118   LiveRegCycles.resize(TRI->getNumRegs(), 0);
119 
120   // Build the scheduling graph.
121   BuildSchedGraph(nullptr);
122 
123   LLVM_DEBUG(dump());
124 
125   // Execute the actual scheduling loop.
126   ListScheduleBottomUp();
127 }
128 
129 //===----------------------------------------------------------------------===//
130 //  Bottom-Up Scheduling
131 //===----------------------------------------------------------------------===//
132 
133 /// ReleasePred - Decrement the NumSuccsLeft count of a predecessor. Add it to
134 /// the AvailableQueue if the count reaches zero. Also update its cycle bound.
135 void ScheduleDAGFast::ReleasePred(SUnit *SU, SDep *PredEdge) {
136   SUnit *PredSU = PredEdge->getSUnit();
137 
138 #ifndef NDEBUG
139   if (PredSU->NumSuccsLeft == 0) {
140     dbgs() << "*** Scheduling failed! ***\n";
141     dumpNode(*PredSU);
142     dbgs() << " has been released too many times!\n";
143     llvm_unreachable(nullptr);
144   }
145 #endif
146   --PredSU->NumSuccsLeft;
147 
148   // If all the node's successors are scheduled, this node is ready
149   // to be scheduled. Ignore the special EntrySU node.
150   if (PredSU->NumSuccsLeft == 0 && PredSU != &EntrySU) {
151     PredSU->isAvailable = true;
152     AvailableQueue.push(PredSU);
153   }
154 }
155 
156 void ScheduleDAGFast::ReleasePredecessors(SUnit *SU, unsigned CurCycle) {
157   // Bottom up: release predecessors
158   for (SDep &Pred : SU->Preds) {
159     ReleasePred(SU, &Pred);
160     if (Pred.isAssignedRegDep()) {
161       // This is a physical register dependency and it's impossible or
162       // expensive to copy the register. Make sure nothing that can
163       // clobber the register is scheduled between the predecessor and
164       // this node.
165       if (!LiveRegDefs[Pred.getReg()]) {
166         ++NumLiveRegs;
167         LiveRegDefs[Pred.getReg()] = Pred.getSUnit();
168         LiveRegCycles[Pred.getReg()] = CurCycle;
169       }
170     }
171   }
172 }
173 
174 /// ScheduleNodeBottomUp - Add the node to the schedule. Decrement the pending
175 /// count of its predecessors. If a predecessor pending count is zero, add it to
176 /// the Available queue.
177 void ScheduleDAGFast::ScheduleNodeBottomUp(SUnit *SU, unsigned CurCycle) {
178   LLVM_DEBUG(dbgs() << "*** Scheduling [" << CurCycle << "]: ");
179   LLVM_DEBUG(dumpNode(*SU));
180 
181   assert(CurCycle >= SU->getHeight() && "Node scheduled below its height!");
182   SU->setHeightToAtLeast(CurCycle);
183   Sequence.push_back(SU);
184 
185   ReleasePredecessors(SU, CurCycle);
186 
187   // Release all the implicit physical register defs that are live.
188   for (SDep &Succ : SU->Succs) {
189     if (Succ.isAssignedRegDep()) {
190       if (LiveRegCycles[Succ.getReg()] == Succ.getSUnit()->getHeight()) {
191         assert(NumLiveRegs > 0 && "NumLiveRegs is already zero!");
192         assert(LiveRegDefs[Succ.getReg()] == SU &&
193                "Physical register dependency violated?");
194         --NumLiveRegs;
195         LiveRegDefs[Succ.getReg()] = nullptr;
196         LiveRegCycles[Succ.getReg()] = 0;
197       }
198     }
199   }
200 
201   SU->isScheduled = true;
202 }
203 
204 /// CopyAndMoveSuccessors - Clone the specified node and move its scheduled
205 /// successors to the newly created node.
206 SUnit *ScheduleDAGFast::CopyAndMoveSuccessors(SUnit *SU) {
207   if (SU->getNode()->getGluedNode())
208     return nullptr;
209 
210   SDNode *N = SU->getNode();
211   if (!N)
212     return nullptr;
213 
214   SUnit *NewSU;
215   bool TryUnfold = false;
216   for (unsigned i = 0, e = N->getNumValues(); i != e; ++i) {
217     MVT VT = N->getSimpleValueType(i);
218     if (VT == MVT::Glue)
219       return nullptr;
220     else if (VT == MVT::Other)
221       TryUnfold = true;
222   }
223   for (const SDValue &Op : N->op_values()) {
224     MVT VT = Op.getNode()->getSimpleValueType(Op.getResNo());
225     if (VT == MVT::Glue)
226       return nullptr;
227   }
228 
229   if (TryUnfold) {
230     SmallVector<SDNode*, 2> NewNodes;
231     if (!TII->unfoldMemoryOperand(*DAG, N, NewNodes))
232       return nullptr;
233 
234     LLVM_DEBUG(dbgs() << "Unfolding SU # " << SU->NodeNum << "\n");
235     assert(NewNodes.size() == 2 && "Expected a load folding node!");
236 
237     N = NewNodes[1];
238     SDNode *LoadNode = NewNodes[0];
239     unsigned NumVals = N->getNumValues();
240     unsigned OldNumVals = SU->getNode()->getNumValues();
241     for (unsigned i = 0; i != NumVals; ++i)
242       DAG->ReplaceAllUsesOfValueWith(SDValue(SU->getNode(), i), SDValue(N, i));
243     DAG->ReplaceAllUsesOfValueWith(SDValue(SU->getNode(), OldNumVals-1),
244                                    SDValue(LoadNode, 1));
245 
246     SUnit *NewSU = newSUnit(N);
247     assert(N->getNodeId() == -1 && "Node already inserted!");
248     N->setNodeId(NewSU->NodeNum);
249 
250     const MCInstrDesc &MCID = TII->get(N->getMachineOpcode());
251     for (unsigned i = 0; i != MCID.getNumOperands(); ++i) {
252       if (MCID.getOperandConstraint(i, MCOI::TIED_TO) != -1) {
253         NewSU->isTwoAddress = true;
254         break;
255       }
256     }
257     if (MCID.isCommutable())
258       NewSU->isCommutable = true;
259 
260     // LoadNode may already exist. This can happen when there is another
261     // load from the same location and producing the same type of value
262     // but it has different alignment or volatileness.
263     bool isNewLoad = true;
264     SUnit *LoadSU;
265     if (LoadNode->getNodeId() != -1) {
266       LoadSU = &SUnits[LoadNode->getNodeId()];
267       isNewLoad = false;
268     } else {
269       LoadSU = newSUnit(LoadNode);
270       LoadNode->setNodeId(LoadSU->NodeNum);
271     }
272 
273     SDep ChainPred;
274     SmallVector<SDep, 4> ChainSuccs;
275     SmallVector<SDep, 4> LoadPreds;
276     SmallVector<SDep, 4> NodePreds;
277     SmallVector<SDep, 4> NodeSuccs;
278     for (SDep &Pred : SU->Preds) {
279       if (Pred.isCtrl())
280         ChainPred = Pred;
281       else if (Pred.getSUnit()->getNode() &&
282                Pred.getSUnit()->getNode()->isOperandOf(LoadNode))
283         LoadPreds.push_back(Pred);
284       else
285         NodePreds.push_back(Pred);
286     }
287     for (SDep &Succ : SU->Succs) {
288       if (Succ.isCtrl())
289         ChainSuccs.push_back(Succ);
290       else
291         NodeSuccs.push_back(Succ);
292     }
293 
294     if (ChainPred.getSUnit()) {
295       RemovePred(SU, ChainPred);
296       if (isNewLoad)
297         AddPred(LoadSU, ChainPred);
298     }
299     for (unsigned i = 0, e = LoadPreds.size(); i != e; ++i) {
300       const SDep &Pred = LoadPreds[i];
301       RemovePred(SU, Pred);
302       if (isNewLoad) {
303         AddPred(LoadSU, Pred);
304       }
305     }
306     for (unsigned i = 0, e = NodePreds.size(); i != e; ++i) {
307       const SDep &Pred = NodePreds[i];
308       RemovePred(SU, Pred);
309       AddPred(NewSU, Pred);
310     }
311     for (unsigned i = 0, e = NodeSuccs.size(); i != e; ++i) {
312       SDep D = NodeSuccs[i];
313       SUnit *SuccDep = D.getSUnit();
314       D.setSUnit(SU);
315       RemovePred(SuccDep, D);
316       D.setSUnit(NewSU);
317       AddPred(SuccDep, D);
318     }
319     for (unsigned i = 0, e = ChainSuccs.size(); i != e; ++i) {
320       SDep D = ChainSuccs[i];
321       SUnit *SuccDep = D.getSUnit();
322       D.setSUnit(SU);
323       RemovePred(SuccDep, D);
324       if (isNewLoad) {
325         D.setSUnit(LoadSU);
326         AddPred(SuccDep, D);
327       }
328     }
329     if (isNewLoad) {
330       SDep D(LoadSU, SDep::Barrier);
331       D.setLatency(LoadSU->Latency);
332       AddPred(NewSU, D);
333     }
334 
335     ++NumUnfolds;
336 
337     if (NewSU->NumSuccsLeft == 0) {
338       NewSU->isAvailable = true;
339       return NewSU;
340     }
341     SU = NewSU;
342   }
343 
344   LLVM_DEBUG(dbgs() << "Duplicating SU # " << SU->NodeNum << "\n");
345   NewSU = Clone(SU);
346 
347   // New SUnit has the exact same predecessors.
348   for (SDep &Pred : SU->Preds)
349     if (!Pred.isArtificial())
350       AddPred(NewSU, Pred);
351 
352   // Only copy scheduled successors. Cut them from old node's successor
353   // list and move them over.
354   SmallVector<std::pair<SUnit *, SDep>, 4> DelDeps;
355   for (SDep &Succ : SU->Succs) {
356     if (Succ.isArtificial())
357       continue;
358     SUnit *SuccSU = Succ.getSUnit();
359     if (SuccSU->isScheduled) {
360       SDep D = Succ;
361       D.setSUnit(NewSU);
362       AddPred(SuccSU, D);
363       D.setSUnit(SU);
364       DelDeps.push_back(std::make_pair(SuccSU, D));
365     }
366   }
367   for (unsigned i = 0, e = DelDeps.size(); i != e; ++i)
368     RemovePred(DelDeps[i].first, DelDeps[i].second);
369 
370   ++NumDups;
371   return NewSU;
372 }
373 
374 /// InsertCopiesAndMoveSuccs - Insert register copies and move all
375 /// scheduled successors of the given SUnit to the last copy.
376 void ScheduleDAGFast::InsertCopiesAndMoveSuccs(SUnit *SU, unsigned Reg,
377                                               const TargetRegisterClass *DestRC,
378                                               const TargetRegisterClass *SrcRC,
379                                               SmallVectorImpl<SUnit*> &Copies) {
380   SUnit *CopyFromSU = newSUnit(static_cast<SDNode *>(nullptr));
381   CopyFromSU->CopySrcRC = SrcRC;
382   CopyFromSU->CopyDstRC = DestRC;
383 
384   SUnit *CopyToSU = newSUnit(static_cast<SDNode *>(nullptr));
385   CopyToSU->CopySrcRC = DestRC;
386   CopyToSU->CopyDstRC = SrcRC;
387 
388   // Only copy scheduled successors. Cut them from old node's successor
389   // list and move them over.
390   SmallVector<std::pair<SUnit *, SDep>, 4> DelDeps;
391   for (SDep &Succ : SU->Succs) {
392     if (Succ.isArtificial())
393       continue;
394     SUnit *SuccSU = Succ.getSUnit();
395     if (SuccSU->isScheduled) {
396       SDep D = Succ;
397       D.setSUnit(CopyToSU);
398       AddPred(SuccSU, D);
399       DelDeps.push_back(std::make_pair(SuccSU, Succ));
400     }
401   }
402   for (unsigned i = 0, e = DelDeps.size(); i != e; ++i) {
403     RemovePred(DelDeps[i].first, DelDeps[i].second);
404   }
405   SDep FromDep(SU, SDep::Data, Reg);
406   FromDep.setLatency(SU->Latency);
407   AddPred(CopyFromSU, FromDep);
408   SDep ToDep(CopyFromSU, SDep::Data, 0);
409   ToDep.setLatency(CopyFromSU->Latency);
410   AddPred(CopyToSU, ToDep);
411 
412   Copies.push_back(CopyFromSU);
413   Copies.push_back(CopyToSU);
414 
415   ++NumPRCopies;
416 }
417 
418 /// getPhysicalRegisterVT - Returns the ValueType of the physical register
419 /// definition of the specified node.
420 /// FIXME: Move to SelectionDAG?
421 static MVT getPhysicalRegisterVT(SDNode *N, unsigned Reg,
422                                  const TargetInstrInfo *TII) {
423   unsigned NumRes;
424   if (N->getOpcode() == ISD::CopyFromReg) {
425     // CopyFromReg has: "chain, Val, glue" so operand 1 gives the type.
426     NumRes = 1;
427   } else {
428     const MCInstrDesc &MCID = TII->get(N->getMachineOpcode());
429     assert(MCID.ImplicitDefs && "Physical reg def must be in implicit def list!");
430     NumRes = MCID.getNumDefs();
431     for (const MCPhysReg *ImpDef = MCID.getImplicitDefs(); *ImpDef; ++ImpDef) {
432       if (Reg == *ImpDef)
433         break;
434       ++NumRes;
435     }
436   }
437   return N->getSimpleValueType(NumRes);
438 }
439 
440 /// CheckForLiveRegDef - Return true and update live register vector if the
441 /// specified register def of the specified SUnit clobbers any "live" registers.
442 static bool CheckForLiveRegDef(SUnit *SU, unsigned Reg,
443                                std::vector<SUnit *> &LiveRegDefs,
444                                SmallSet<unsigned, 4> &RegAdded,
445                                SmallVectorImpl<unsigned> &LRegs,
446                                const TargetRegisterInfo *TRI,
447                                const SDNode *Node = nullptr) {
448   bool Added = false;
449   for (MCRegAliasIterator AI(Reg, TRI, true); AI.isValid(); ++AI) {
450     // Check if Ref is live.
451     if (!LiveRegDefs[*AI])
452       continue;
453 
454     // Allow multiple uses of the same def.
455     if (LiveRegDefs[*AI] == SU)
456       continue;
457 
458     // Allow multiple uses of same def
459     if (Node && LiveRegDefs[*AI]->getNode() == Node)
460       continue;
461 
462     // Add Reg to the set of interfering live regs.
463     if (RegAdded.insert(*AI).second) {
464       LRegs.push_back(*AI);
465       Added = true;
466     }
467   }
468   return Added;
469 }
470 
471 /// DelayForLiveRegsBottomUp - Returns true if it is necessary to delay
472 /// scheduling of the given node to satisfy live physical register dependencies.
473 /// If the specific node is the last one that's available to schedule, do
474 /// whatever is necessary (i.e. backtracking or cloning) to make it possible.
475 bool ScheduleDAGFast::DelayForLiveRegsBottomUp(SUnit *SU,
476                                               SmallVectorImpl<unsigned> &LRegs){
477   if (NumLiveRegs == 0)
478     return false;
479 
480   SmallSet<unsigned, 4> RegAdded;
481   // If this node would clobber any "live" register, then it's not ready.
482   for (SDep &Pred : SU->Preds) {
483     if (Pred.isAssignedRegDep()) {
484       CheckForLiveRegDef(Pred.getSUnit(), Pred.getReg(), LiveRegDefs,
485                          RegAdded, LRegs, TRI);
486     }
487   }
488 
489   for (SDNode *Node = SU->getNode(); Node; Node = Node->getGluedNode()) {
490     if (Node->getOpcode() == ISD::INLINEASM ||
491         Node->getOpcode() == ISD::INLINEASM_BR) {
492       // Inline asm can clobber physical defs.
493       unsigned NumOps = Node->getNumOperands();
494       if (Node->getOperand(NumOps-1).getValueType() == MVT::Glue)
495         --NumOps;  // Ignore the glue operand.
496 
497       for (unsigned i = InlineAsm::Op_FirstOperand; i != NumOps;) {
498         unsigned Flags =
499           cast<ConstantSDNode>(Node->getOperand(i))->getZExtValue();
500         unsigned NumVals = InlineAsm::getNumOperandRegisters(Flags);
501 
502         ++i; // Skip the ID value.
503         if (InlineAsm::isRegDefKind(Flags) ||
504             InlineAsm::isRegDefEarlyClobberKind(Flags) ||
505             InlineAsm::isClobberKind(Flags)) {
506           // Check for def of register or earlyclobber register.
507           for (; NumVals; --NumVals, ++i) {
508             unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
509             if (Register::isPhysicalRegister(Reg))
510               CheckForLiveRegDef(SU, Reg, LiveRegDefs, RegAdded, LRegs, TRI);
511           }
512         } else
513           i += NumVals;
514       }
515       continue;
516     }
517 
518     if (Node->getOpcode() == ISD::CopyToReg) {
519       Register Reg = cast<RegisterSDNode>(Node->getOperand(1))->getReg();
520       if (Reg.isPhysical()) {
521         SDNode *SrcNode = Node->getOperand(2).getNode();
522         CheckForLiveRegDef(SU, Reg, LiveRegDefs, RegAdded, LRegs, TRI, SrcNode);
523       }
524     }
525 
526     if (!Node->isMachineOpcode())
527       continue;
528     const MCInstrDesc &MCID = TII->get(Node->getMachineOpcode());
529     if (!MCID.ImplicitDefs)
530       continue;
531     for (const MCPhysReg *Reg = MCID.getImplicitDefs(); *Reg; ++Reg) {
532       CheckForLiveRegDef(SU, *Reg, LiveRegDefs, RegAdded, LRegs, TRI);
533     }
534   }
535   return !LRegs.empty();
536 }
537 
538 
539 /// ListScheduleBottomUp - The main loop of list scheduling for bottom-up
540 /// schedulers.
541 void ScheduleDAGFast::ListScheduleBottomUp() {
542   unsigned CurCycle = 0;
543 
544   // Release any predecessors of the special Exit node.
545   ReleasePredecessors(&ExitSU, CurCycle);
546 
547   // Add root to Available queue.
548   if (!SUnits.empty()) {
549     SUnit *RootSU = &SUnits[DAG->getRoot().getNode()->getNodeId()];
550     assert(RootSU->Succs.empty() && "Graph root shouldn't have successors!");
551     RootSU->isAvailable = true;
552     AvailableQueue.push(RootSU);
553   }
554 
555   // While Available queue is not empty, grab the node with the highest
556   // priority. If it is not ready put it back.  Schedule the node.
557   SmallVector<SUnit*, 4> NotReady;
558   DenseMap<SUnit*, SmallVector<unsigned, 4> > LRegsMap;
559   Sequence.reserve(SUnits.size());
560   while (!AvailableQueue.empty()) {
561     bool Delayed = false;
562     LRegsMap.clear();
563     SUnit *CurSU = AvailableQueue.pop();
564     while (CurSU) {
565       SmallVector<unsigned, 4> LRegs;
566       if (!DelayForLiveRegsBottomUp(CurSU, LRegs))
567         break;
568       Delayed = true;
569       LRegsMap.insert(std::make_pair(CurSU, LRegs));
570 
571       CurSU->isPending = true;  // This SU is not in AvailableQueue right now.
572       NotReady.push_back(CurSU);
573       CurSU = AvailableQueue.pop();
574     }
575 
576     // All candidates are delayed due to live physical reg dependencies.
577     // Try code duplication or inserting cross class copies
578     // to resolve it.
579     if (Delayed && !CurSU) {
580       if (!CurSU) {
581         // Try duplicating the nodes that produces these
582         // "expensive to copy" values to break the dependency. In case even
583         // that doesn't work, insert cross class copies.
584         SUnit *TrySU = NotReady[0];
585         SmallVectorImpl<unsigned> &LRegs = LRegsMap[TrySU];
586         assert(LRegs.size() == 1 && "Can't handle this yet!");
587         unsigned Reg = LRegs[0];
588         SUnit *LRDef = LiveRegDefs[Reg];
589         MVT VT = getPhysicalRegisterVT(LRDef->getNode(), Reg, TII);
590         const TargetRegisterClass *RC =
591           TRI->getMinimalPhysRegClass(Reg, VT);
592         const TargetRegisterClass *DestRC = TRI->getCrossCopyRegClass(RC);
593 
594         // If cross copy register class is the same as RC, then it must be
595         // possible copy the value directly. Do not try duplicate the def.
596         // If cross copy register class is not the same as RC, then it's
597         // possible to copy the value but it require cross register class copies
598         // and it is expensive.
599         // If cross copy register class is null, then it's not possible to copy
600         // the value at all.
601         SUnit *NewDef = nullptr;
602         if (DestRC != RC) {
603           NewDef = CopyAndMoveSuccessors(LRDef);
604           if (!DestRC && !NewDef)
605             report_fatal_error("Can't handle live physical "
606                                "register dependency!");
607         }
608         if (!NewDef) {
609           // Issue copies, these can be expensive cross register class copies.
610           SmallVector<SUnit*, 2> Copies;
611           InsertCopiesAndMoveSuccs(LRDef, Reg, DestRC, RC, Copies);
612           LLVM_DEBUG(dbgs() << "Adding an edge from SU # " << TrySU->NodeNum
613                             << " to SU #" << Copies.front()->NodeNum << "\n");
614           AddPred(TrySU, SDep(Copies.front(), SDep::Artificial));
615           NewDef = Copies.back();
616         }
617 
618         LLVM_DEBUG(dbgs() << "Adding an edge from SU # " << NewDef->NodeNum
619                           << " to SU #" << TrySU->NodeNum << "\n");
620         LiveRegDefs[Reg] = NewDef;
621         AddPred(NewDef, SDep(TrySU, SDep::Artificial));
622         TrySU->isAvailable = false;
623         CurSU = NewDef;
624       }
625 
626       if (!CurSU) {
627         llvm_unreachable("Unable to resolve live physical register dependencies!");
628       }
629     }
630 
631     // Add the nodes that aren't ready back onto the available list.
632     for (unsigned i = 0, e = NotReady.size(); i != e; ++i) {
633       NotReady[i]->isPending = false;
634       // May no longer be available due to backtracking.
635       if (NotReady[i]->isAvailable)
636         AvailableQueue.push(NotReady[i]);
637     }
638     NotReady.clear();
639 
640     if (CurSU)
641       ScheduleNodeBottomUp(CurSU, CurCycle);
642     ++CurCycle;
643   }
644 
645   // Reverse the order since it is bottom up.
646   std::reverse(Sequence.begin(), Sequence.end());
647 
648 #ifndef NDEBUG
649   VerifyScheduledSequence(/*isBottomUp=*/true);
650 #endif
651 }
652 
653 
654 namespace {
655 //===----------------------------------------------------------------------===//
656 // ScheduleDAGLinearize - No scheduling scheduler, it simply linearize the
657 // DAG in topological order.
658 // IMPORTANT: this may not work for targets with phyreg dependency.
659 //
660 class ScheduleDAGLinearize : public ScheduleDAGSDNodes {
661 public:
662   ScheduleDAGLinearize(MachineFunction &mf) : ScheduleDAGSDNodes(mf) {}
663 
664   void Schedule() override;
665 
666   MachineBasicBlock *
667     EmitSchedule(MachineBasicBlock::iterator &InsertPos) override;
668 
669 private:
670   std::vector<SDNode*> Sequence;
671   DenseMap<SDNode*, SDNode*> GluedMap;  // Cache glue to its user
672 
673   void ScheduleNode(SDNode *N);
674 };
675 } // end anonymous namespace
676 
677 void ScheduleDAGLinearize::ScheduleNode(SDNode *N) {
678   if (N->getNodeId() != 0)
679     llvm_unreachable(nullptr);
680 
681   if (!N->isMachineOpcode() &&
682       (N->getOpcode() == ISD::EntryToken || isPassiveNode(N)))
683     // These nodes do not need to be translated into MIs.
684     return;
685 
686   LLVM_DEBUG(dbgs() << "\n*** Scheduling: ");
687   LLVM_DEBUG(N->dump(DAG));
688   Sequence.push_back(N);
689 
690   unsigned NumOps = N->getNumOperands();
691   if (unsigned NumLeft = NumOps) {
692     SDNode *GluedOpN = nullptr;
693     do {
694       const SDValue &Op = N->getOperand(NumLeft-1);
695       SDNode *OpN = Op.getNode();
696 
697       if (NumLeft == NumOps && Op.getValueType() == MVT::Glue) {
698         // Schedule glue operand right above N.
699         GluedOpN = OpN;
700         assert(OpN->getNodeId() != 0 && "Glue operand not ready?");
701         OpN->setNodeId(0);
702         ScheduleNode(OpN);
703         continue;
704       }
705 
706       if (OpN == GluedOpN)
707         // Glue operand is already scheduled.
708         continue;
709 
710       DenseMap<SDNode*, SDNode*>::iterator DI = GluedMap.find(OpN);
711       if (DI != GluedMap.end() && DI->second != N)
712         // Users of glues are counted against the glued users.
713         OpN = DI->second;
714 
715       unsigned Degree = OpN->getNodeId();
716       assert(Degree > 0 && "Predecessor over-released!");
717       OpN->setNodeId(--Degree);
718       if (Degree == 0)
719         ScheduleNode(OpN);
720     } while (--NumLeft);
721   }
722 }
723 
724 /// findGluedUser - Find the representative use of a glue value by walking
725 /// the use chain.
726 static SDNode *findGluedUser(SDNode *N) {
727   while (SDNode *Glued = N->getGluedUser())
728     N = Glued;
729   return N;
730 }
731 
732 void ScheduleDAGLinearize::Schedule() {
733   LLVM_DEBUG(dbgs() << "********** DAG Linearization **********\n");
734 
735   SmallVector<SDNode*, 8> Glues;
736   unsigned DAGSize = 0;
737   for (SDNode &Node : DAG->allnodes()) {
738     SDNode *N = &Node;
739 
740     // Use node id to record degree.
741     unsigned Degree = N->use_size();
742     N->setNodeId(Degree);
743     unsigned NumVals = N->getNumValues();
744     if (NumVals && N->getValueType(NumVals-1) == MVT::Glue &&
745         N->hasAnyUseOfValue(NumVals-1)) {
746       SDNode *User = findGluedUser(N);
747       if (User) {
748         Glues.push_back(N);
749         GluedMap.insert(std::make_pair(N, User));
750       }
751     }
752 
753     if (N->isMachineOpcode() ||
754         (N->getOpcode() != ISD::EntryToken && !isPassiveNode(N)))
755       ++DAGSize;
756   }
757 
758   for (unsigned i = 0, e = Glues.size(); i != e; ++i) {
759     SDNode *Glue = Glues[i];
760     SDNode *GUser = GluedMap[Glue];
761     unsigned Degree = Glue->getNodeId();
762     unsigned UDegree = GUser->getNodeId();
763 
764     // Glue user must be scheduled together with the glue operand. So other
765     // users of the glue operand must be treated as its users.
766     SDNode *ImmGUser = Glue->getGluedUser();
767     for (const SDNode *U : Glue->uses())
768       if (U == ImmGUser)
769         --Degree;
770     GUser->setNodeId(UDegree + Degree);
771     Glue->setNodeId(1);
772   }
773 
774   Sequence.reserve(DAGSize);
775   ScheduleNode(DAG->getRoot().getNode());
776 }
777 
778 MachineBasicBlock*
779 ScheduleDAGLinearize::EmitSchedule(MachineBasicBlock::iterator &InsertPos) {
780   InstrEmitter Emitter(DAG->getTarget(), BB, InsertPos,
781                        DAG->getUseInstrRefDebugInfo());
782   DenseMap<SDValue, Register> VRBaseMap;
783 
784   LLVM_DEBUG({ dbgs() << "\n*** Final schedule ***\n"; });
785 
786   unsigned NumNodes = Sequence.size();
787   MachineBasicBlock *BB = Emitter.getBlock();
788   for (unsigned i = 0; i != NumNodes; ++i) {
789     SDNode *N = Sequence[NumNodes-i-1];
790     LLVM_DEBUG(N->dump(DAG));
791     Emitter.EmitNode(N, false, false, VRBaseMap);
792 
793     // Emit any debug values associated with the node.
794     if (N->getHasDebugValue()) {
795       MachineBasicBlock::iterator InsertPos = Emitter.getInsertPos();
796       for (auto *DV : DAG->GetDbgValues(N)) {
797         if (!DV->isEmitted())
798           if (auto *DbgMI = Emitter.EmitDbgValue(DV, VRBaseMap))
799             BB->insert(InsertPos, DbgMI);
800       }
801     }
802   }
803 
804   LLVM_DEBUG(dbgs() << '\n');
805 
806   InsertPos = Emitter.getInsertPos();
807   return Emitter.getBlock();
808 }
809 
810 //===----------------------------------------------------------------------===//
811 //                         Public Constructor Functions
812 //===----------------------------------------------------------------------===//
813 
814 llvm::ScheduleDAGSDNodes *
815 llvm::createFastDAGScheduler(SelectionDAGISel *IS, CodeGenOpt::Level) {
816   return new ScheduleDAGFast(*IS->MF);
817 }
818 
819 llvm::ScheduleDAGSDNodes *
820 llvm::createDAGLinearizer(SelectionDAGISel *IS, CodeGenOpt::Level) {
821   return new ScheduleDAGLinearize(*IS->MF);
822 }
823