summaryrefslogtreecommitdiff
path: root/lib/CodeGen/SelectionDAG/ScheduleDAG.cpp
blob: 1d154017f51eb592b04ebb97879d3f553c11a6f7 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
//===---- ScheduleDAG.cpp - Implement the ScheduleDAG class ---------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file was developed by James M. Laskey and is distributed under the
// University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This implements a simple two pass scheduler.  The first pass attempts to push
// backward any lengthy instructions and critical paths.  The second pass packs
// instructions into semi-optimal time slots.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "sched"
#include "llvm/CodeGen/ScheduleDAG.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/SSARegMap.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include <iostream>
using namespace llvm;


/// BuildSchedUnits - Build SUnits from the selection dag that we are input.
/// This SUnit graph is similar to the SelectionDAG, but represents flagged
/// together nodes with a single SUnit.
void ScheduleDAG::BuildSchedUnits() {
  // Reserve entries in the vector for each of the SUnits we are creating.  This
  // ensure that reallocation of the vector won't happen, so SUnit*'s won't get
  // invalidated.
  SUnits.reserve(std::distance(DAG.allnodes_begin(), DAG.allnodes_end()));
  
  const InstrItineraryData &InstrItins = TM.getInstrItineraryData();
  
  for (SelectionDAG::allnodes_iterator NI = DAG.allnodes_begin(),
       E = DAG.allnodes_end(); NI != E; ++NI) {
    if (isPassiveNode(NI))  // Leaf node, e.g. a TargetImmediate.
      continue;
    
    // If this node has already been processed, stop now.
    if (SUnitMap[NI]) continue;
    
    SUnit *NodeSUnit = NewSUnit(NI);
    
    // See if anything is flagged to this node, if so, add them to flagged
    // nodes.  Nodes can have at most one flag input and one flag output.  Flags
    // are required the be the last operand and result of a node.
    
    // Scan up, adding flagged preds to FlaggedNodes.
    SDNode *N = NI;
    while (N->getNumOperands() &&
           N->getOperand(N->getNumOperands()-1).getValueType() == MVT::Flag) {
      N = N->getOperand(N->getNumOperands()-1).Val;
      NodeSUnit->FlaggedNodes.push_back(N);
      SUnitMap[N] = NodeSUnit;
    }
    
    // Scan down, adding this node and any flagged succs to FlaggedNodes if they
    // have a user of the flag operand.
    N = NI;
    while (N->getValueType(N->getNumValues()-1) == MVT::Flag) {
      SDOperand FlagVal(N, N->getNumValues()-1);
      
      // There are either zero or one users of the Flag result.
      bool HasFlagUse = false;
      for (SDNode::use_iterator UI = N->use_begin(), E = N->use_end(); 
           UI != E; ++UI)
        if (FlagVal.isOperand(*UI)) {
          HasFlagUse = true;
          NodeSUnit->FlaggedNodes.push_back(N);
          SUnitMap[N] = NodeSUnit;
          N = *UI;
          break;
        }
          if (!HasFlagUse) break;
    }
    
    // Now all flagged nodes are in FlaggedNodes and N is the bottom-most node.
    // Update the SUnit
    NodeSUnit->Node = N;
    SUnitMap[N] = NodeSUnit;
    
    // Compute the latency for the node.  We use the sum of the latencies for
    // all nodes flagged together into this SUnit.
    if (InstrItins.isEmpty()) {
      // No latency information.
      NodeSUnit->Latency = 1;
    } else {
      NodeSUnit->Latency = 0;
      if (N->isTargetOpcode()) {
        unsigned SchedClass = TII->getSchedClass(N->getTargetOpcode());
        InstrStage *S = InstrItins.begin(SchedClass);
        InstrStage *E = InstrItins.end(SchedClass);
        for (; S != E; ++S)
          NodeSUnit->Latency += S->Cycles;
      }
      for (unsigned i = 0, e = NodeSUnit->FlaggedNodes.size(); i != e; ++i) {
        SDNode *FNode = NodeSUnit->FlaggedNodes[i];
        if (FNode->isTargetOpcode()) {
          unsigned SchedClass = TII->getSchedClass(FNode->getTargetOpcode());
          InstrStage *S = InstrItins.begin(SchedClass);
          InstrStage *E = InstrItins.end(SchedClass);
          for (; S != E; ++S)
            NodeSUnit->Latency += S->Cycles;
        }
      }
    }
  }
  
  // Pass 2: add the preds, succs, etc.
  for (unsigned su = 0, e = SUnits.size(); su != e; ++su) {
    SUnit *SU = &SUnits[su];
    SDNode *MainNode = SU->Node;
    
    if (MainNode->isTargetOpcode()) {
      unsigned Opc = MainNode->getTargetOpcode();
      if (TII->isTwoAddrInstr(Opc))
        SU->isTwoAddress = true;
      if (TII->isCommutableInstr(Opc))
        SU->isCommutable = true;
    }
    
    // Find all predecessors and successors of the group.
    // Temporarily add N to make code simpler.
    SU->FlaggedNodes.push_back(MainNode);
    
    for (unsigned n = 0, e = SU->FlaggedNodes.size(); n != e; ++n) {
      SDNode *N = SU->FlaggedNodes[n];
      
      for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
        SDNode *OpN = N->getOperand(i).Val;
        if (isPassiveNode(OpN)) continue;   // Not scheduled.
        SUnit *OpSU = SUnitMap[OpN];
        assert(OpSU && "Node has no SUnit!");
        if (OpSU == SU) continue;           // In the same group.

        MVT::ValueType OpVT = N->getOperand(i).getValueType();
        assert(OpVT != MVT::Flag && "Flagged nodes should be in same sunit!");
        bool isChain = OpVT == MVT::Other;
        
        if (SU->Preds.insert(std::make_pair(OpSU, isChain)).second) {
          if (!isChain) {
            SU->NumPreds++;
            SU->NumPredsLeft++;
          } else {
            SU->NumChainPredsLeft++;
          }
        }
        if (OpSU->Succs.insert(std::make_pair(SU, isChain)).second) {
          if (!isChain) {
            OpSU->NumSuccs++;
            OpSU->NumSuccsLeft++;
          } else {
            OpSU->NumChainSuccsLeft++;
          }
        }
      }
    }
    
    // Remove MainNode from FlaggedNodes again.
    SU->FlaggedNodes.pop_back();
  }
  
  return;
}

static void CalculateDepths(SUnit *SU, unsigned Depth) {
  if (SU->Depth == 0 || Depth > SU->Depth) {
    SU->Depth = Depth;
    for (std::set<std::pair<SUnit*, bool> >::iterator I = SU->Succs.begin(),
           E = SU->Succs.end(); I != E; ++I)
      CalculateDepths(I->first, Depth+1);
  }
}

void ScheduleDAG::CalculateDepths() {
  SUnit *Entry = SUnitMap[DAG.getEntryNode().Val];
  ::CalculateDepths(Entry, 0U);
  for (unsigned i = 0, e = SUnits.size(); i != e; ++i)
    if (SUnits[i].Preds.size() == 0 && &SUnits[i] != Entry) {
      ::CalculateDepths(&SUnits[i], 0U);
    }
}

static void CalculateHeights(SUnit *SU, unsigned Height) {
  if (SU->Height == 0 || Height > SU->Height) {
    SU->Height = Height;
    for (std::set<std::pair<SUnit*, bool> >::iterator I = SU->Preds.begin(),
           E = SU->Preds.end(); I != E; ++I)
      CalculateHeights(I->first, Height+1);
  }
}
void ScheduleDAG::CalculateHeights() {
  SUnit *Root = SUnitMap[DAG.getRoot().Val];
  ::CalculateHeights(Root, 0U);
}

/// CountResults - The results of target nodes have register or immediate
/// operands first, then an optional chain, and optional flag operands (which do
/// not go into the machine instrs.)
static unsigned CountResults(SDNode *Node) {
  unsigned N = Node->getNumValues();
  while (N && Node->getValueType(N - 1) == MVT::Flag)
    --N;
  if (N && Node->getValueType(N - 1) == MVT::Other)
    --N;    // Skip over chain result.
  return N;
}

/// CountOperands  The inputs to target nodes have any actual inputs first,
/// followed by an optional chain operand, then flag operands.  Compute the
/// number of actual operands that  will go into the machine instr.
static unsigned CountOperands(SDNode *Node) {
  unsigned N = Node->getNumOperands();
  while (N && Node->getOperand(N - 1).getValueType() == MVT::Flag)
    --N;
  if (N && Node->getOperand(N - 1).getValueType() == MVT::Other)
    --N; // Ignore chain if it exists.
  return N;
}

static unsigned CreateVirtualRegisters(MachineInstr *MI,
                                       unsigned NumResults,
                                       SSARegMap *RegMap,
                                       const TargetInstrInfo *TII,
                                       const TargetInstrDescriptor &II) {
  // Create the result registers for this node and add the result regs to
  // the machine instruction.
  unsigned ResultReg =
    RegMap->createVirtualRegister(TII->getInstrOperandRegClass(&II, 0));
  MI->addRegOperand(ResultReg, MachineOperand::Def);
  for (unsigned i = 1; i != NumResults; ++i) {
    const TargetRegisterClass *RC = TII->getInstrOperandRegClass(&II, i);
    assert(RC && "Isn't a register operand!");
    MI->addRegOperand(RegMap->createVirtualRegister(RC), MachineOperand::Def);
  }
  return ResultReg;
}

/// getVR - Return the virtual register corresponding to the specified result
/// of the specified node.
static unsigned getVR(SDOperand Op, std::map<SDNode*, unsigned> &VRBaseMap) {
  std::map<SDNode*, unsigned>::iterator I = VRBaseMap.find(Op.Val);
  assert(I != VRBaseMap.end() && "Node emitted out of order - late");
  return I->second + Op.ResNo;
}


/// AddOperand - Add the specified operand to the specified machine instr.  II
/// specifies the instruction information for the node, and IIOpNum is the
/// operand number (in the II) that we are adding. IIOpNum and II are used for 
/// assertions only.
void ScheduleDAG::AddOperand(MachineInstr *MI, SDOperand Op,
                             unsigned IIOpNum,
                             const TargetInstrDescriptor *II,
                             std::map<SDNode*, unsigned> &VRBaseMap) {
  if (Op.isTargetOpcode()) {
    // Note that this case is redundant with the final else block, but we
    // include it because it is the most common and it makes the logic
    // simpler here.
    assert(Op.getValueType() != MVT::Other &&
           Op.getValueType() != MVT::Flag &&
           "Chain and flag operands should occur at end of operand list!");
    
    // Get/emit the operand.
    unsigned VReg = getVR(Op, VRBaseMap);
    MI->addRegOperand(VReg, MachineOperand::Use);
    
    // Verify that it is right.
    assert(MRegisterInfo::isVirtualRegister(VReg) && "Not a vreg?");
    if (II) {
      const TargetRegisterClass *RC = TII->getInstrOperandRegClass(II, IIOpNum);
      assert(RC && "Don't have operand info for this instruction!");
      assert(RegMap->getRegClass(VReg) == RC &&
             "Register class of operand and regclass of use don't agree!");
    }
  } else if (ConstantSDNode *C =
             dyn_cast<ConstantSDNode>(Op)) {
    MI->addImmOperand(C->getValue());
  } else if (RegisterSDNode*R =
             dyn_cast<RegisterSDNode>(Op)) {
    MI->addRegOperand(R->getReg(), MachineOperand::Use);
  } else if (GlobalAddressSDNode *TGA =
             dyn_cast<GlobalAddressSDNode>(Op)) {
    MI->addGlobalAddressOperand(TGA->getGlobal(), TGA->getOffset());
  } else if (BasicBlockSDNode *BB =
             dyn_cast<BasicBlockSDNode>(Op)) {
    MI->addMachineBasicBlockOperand(BB->getBasicBlock());
  } else if (FrameIndexSDNode *FI =
             dyn_cast<FrameIndexSDNode>(Op)) {
    MI->addFrameIndexOperand(FI->getIndex());
  } else if (JumpTableSDNode *JT =
             dyn_cast<JumpTableSDNode>(Op)) {
    MI->addJumpTableIndexOperand(JT->getIndex());
  } else if (ConstantPoolSDNode *CP = 
             dyn_cast<ConstantPoolSDNode>(Op)) {
    int Offset = CP->getOffset();
    unsigned Align = CP->getAlignment();
    // MachineConstantPool wants an explicit alignment.
    if (Align == 0) {
      if (CP->get()->getType() == Type::DoubleTy)
        Align = 3;  // always 8-byte align doubles.
      else {
        Align = TM.getTargetData()
          ->getTypeAlignmentShift(CP->get()->getType());
        if (Align == 0) {
          // Alignment of packed types.  FIXME!
          Align = TM.getTargetData()->getTypeSize(CP->get()->getType());
          Align = Log2_64(Align);
        }
      }
    }
    
    unsigned Idx = ConstPool->getConstantPoolIndex(CP->get(), Align);
    MI->addConstantPoolIndexOperand(Idx, Offset);
  } else if (ExternalSymbolSDNode *ES = 
             dyn_cast<ExternalSymbolSDNode>(Op)) {
    MI->addExternalSymbolOperand(ES->getSymbol());
  } else {
    assert(Op.getValueType() != MVT::Other &&
           Op.getValueType() != MVT::Flag &&
           "Chain and flag operands should occur at end of operand list!");
    unsigned VReg = getVR(Op, VRBaseMap);
    MI->addRegOperand(VReg, MachineOperand::Use);
    
    // Verify that it is right.
    assert(MRegisterInfo::isVirtualRegister(VReg) && "Not a vreg?");
    if (II) {
      const TargetRegisterClass *RC = TII->getInstrOperandRegClass(II, IIOpNum);
      assert(RC && "Don't have operand info for this instruction!");
      assert(RegMap->getRegClass(VReg) == RC &&
             "Register class of operand and regclass of use don't agree!");
    }
  }
  
}


/// EmitNode - Generate machine code for an node and needed dependencies.
///
void ScheduleDAG::EmitNode(SDNode *Node, 
                           std::map<SDNode*, unsigned> &VRBaseMap) {
  unsigned VRBase = 0;                 // First virtual register for node
  
  // If machine instruction
  if (Node->isTargetOpcode()) {
    unsigned Opc = Node->getTargetOpcode();
    const TargetInstrDescriptor &II = TII->get(Opc);

    unsigned NumResults = CountResults(Node);
    unsigned NodeOperands = CountOperands(Node);
    unsigned NumMIOperands = NodeOperands + NumResults;
#ifndef NDEBUG
    assert((unsigned(II.numOperands) == NumMIOperands || II.numOperands == -1)&&
           "#operands for dag node doesn't match .td file!"); 
#endif

    // Create the new machine instruction.
    MachineInstr *MI = new MachineInstr(Opc, NumMIOperands);
    
    // Add result register values for things that are defined by this
    // instruction.
    
    // If the node is only used by a CopyToReg and the dest reg is a vreg, use
    // the CopyToReg'd destination register instead of creating a new vreg.
    if (NumResults == 1) {
      for (SDNode::use_iterator UI = Node->use_begin(), E = Node->use_end();
           UI != E; ++UI) {
        SDNode *Use = *UI;
        if (Use->getOpcode() == ISD::CopyToReg && 
            Use->getOperand(2).Val == Node) {
          unsigned Reg = cast<RegisterSDNode>(Use->getOperand(1))->getReg();
          if (MRegisterInfo::isVirtualRegister(Reg)) {
            VRBase = Reg;
            MI->addRegOperand(Reg, MachineOperand::Def);
            break;
          }
        }
      }
    }
    
    // Otherwise, create new virtual registers.
    if (NumResults && VRBase == 0)
      VRBase = CreateVirtualRegisters(MI, NumResults, RegMap, TII, II);
    
    // Emit all of the actual operands of this instruction, adding them to the
    // instruction as appropriate.
    for (unsigned i = 0; i != NodeOperands; ++i)
      AddOperand(MI, Node->getOperand(i), i+NumResults, &II, VRBaseMap);

    // Commute node if it has been determined to be profitable.
    if (CommuteSet.count(Node)) {
      MachineInstr *NewMI = TII->commuteInstruction(MI);
      if (NewMI == 0)
        DEBUG(std::cerr << "Sched: COMMUTING FAILED!\n");
      else {
        DEBUG(std::cerr << "Sched: COMMUTED TO: " << *NewMI);
        if (MI != NewMI) {
          delete MI;
          MI = NewMI;
        }
      }
    }

    // Now that we have emitted all operands, emit this instruction itself.
    if ((II.Flags & M_USES_CUSTOM_DAG_SCHED_INSERTION) == 0) {
      BB->insert(BB->end(), MI);
    } else {
      // Insert this instruction into the end of the basic block, potentially
      // taking some custom action.
      BB = DAG.getTargetLoweringInfo().InsertAtEndOfBasicBlock(MI, BB);
    }
  } else {
    switch (Node->getOpcode()) {
    default:
      Node->dump(); 
      assert(0 && "This target-independent node should have been selected!");
    case ISD::EntryToken: // fall thru
    case ISD::TokenFactor:
      break;
    case ISD::CopyToReg: {
      unsigned InReg = getVR(Node->getOperand(2), VRBaseMap);
      unsigned DestReg = cast<RegisterSDNode>(Node->getOperand(1))->getReg();
      if (InReg != DestReg)   // Coalesced away the copy?
        MRI->copyRegToReg(*BB, BB->end(), DestReg, InReg,
                          RegMap->getRegClass(InReg));
      break;
    }
    case ISD::CopyFromReg: {
      unsigned SrcReg = cast<RegisterSDNode>(Node->getOperand(1))->getReg();
      if (MRegisterInfo::isVirtualRegister(SrcReg)) {
        VRBase = SrcReg;  // Just use the input register directly!
        break;
      }

      // If the node is only used by a CopyToReg and the dest reg is a vreg, use
      // the CopyToReg'd destination register instead of creating a new vreg.
      for (SDNode::use_iterator UI = Node->use_begin(), E = Node->use_end();
           UI != E; ++UI) {
        SDNode *Use = *UI;
        if (Use->getOpcode() == ISD::CopyToReg && 
            Use->getOperand(2).Val == Node) {
          unsigned DestReg = cast<RegisterSDNode>(Use->getOperand(1))->getReg();
          if (MRegisterInfo::isVirtualRegister(DestReg)) {
            VRBase = DestReg;
            break;
          }
        }
      }

      // Figure out the register class to create for the destreg.
      const TargetRegisterClass *TRC = 0;
      if (VRBase) {
        TRC = RegMap->getRegClass(VRBase);
      } else {

        // Pick the register class of the right type that contains this physreg.
        for (MRegisterInfo::regclass_iterator I = MRI->regclass_begin(),
             E = MRI->regclass_end(); I != E; ++I)
          if ((*I)->hasType(Node->getValueType(0)) &&
              (*I)->contains(SrcReg)) {
            TRC = *I;
            break;
          }
        assert(TRC && "Couldn't find register class for reg copy!");
      
        // Create the reg, emit the copy.
        VRBase = RegMap->createVirtualRegister(TRC);
      }
      MRI->copyRegToReg(*BB, BB->end(), VRBase, SrcReg, TRC);
      break;
    }
    case ISD::INLINEASM: {
      unsigned NumOps = Node->getNumOperands();
      if (Node->getOperand(NumOps-1).getValueType() == MVT::Flag)
        --NumOps;  // Ignore the flag operand.
      
      // Create the inline asm machine instruction.
      MachineInstr *MI =
        new MachineInstr(BB, TargetInstrInfo::INLINEASM, (NumOps-2)/2+1);

      // Add the asm string as an external symbol operand.
      const char *AsmStr =
        cast<ExternalSymbolSDNode>(Node->getOperand(1))->getSymbol();
      MI->addExternalSymbolOperand(AsmStr);
      
      // Add all of the operand registers to the instruction.
      for (unsigned i = 2; i != NumOps;) {
        unsigned Flags = cast<ConstantSDNode>(Node->getOperand(i))->getValue();
        unsigned NumVals = Flags >> 3;
        
        MI->addImmOperand(Flags);
        ++i;  // Skip the ID value.
        
        switch (Flags & 7) {
        default: assert(0 && "Bad flags!");
        case 1:  // Use of register.
          for (; NumVals; --NumVals, ++i) {
            unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
            MI->addRegOperand(Reg, MachineOperand::Use);
          }
          break;
        case 2:   // Def of register.
          for (; NumVals; --NumVals, ++i) {
            unsigned Reg = cast<RegisterSDNode>(Node->getOperand(i))->getReg();
            MI->addRegOperand(Reg, MachineOperand::Def);
          }
          break;
        case 3: { // Immediate.
          assert(NumVals == 1 && "Unknown immediate value!");
          uint64_t Val = cast<ConstantSDNode>(Node->getOperand(i))->getValue();
          MI->addImmOperand(Val);
          ++i;
          break;
        }
        case 4:  // Addressing mode.
          // The addressing mode has been selected, just add all of the
          // operands to the machine instruction.
          for (; NumVals; --NumVals, ++i)
            AddOperand(MI, Node->getOperand(i), 0, 0, VRBaseMap);
          break;
        }
      }
      break;
    }
    }
  }

  assert(!VRBaseMap.count(Node) && "Node emitted out of order - early");
  VRBaseMap[Node] = VRBase;
}

void ScheduleDAG::EmitNoop() {
  TII->insertNoop(*BB, BB->end());
}

/// EmitSchedule - Emit the machine code in scheduled order.
void ScheduleDAG::EmitSchedule() {
  // If this is the first basic block in the function, and if it has live ins
  // that need to be copied into vregs, emit the copies into the top of the
  // block before emitting the code for the block.
  MachineFunction &MF = DAG.getMachineFunction();
  if (&MF.front() == BB && MF.livein_begin() != MF.livein_end()) {
    for (MachineFunction::livein_iterator LI = MF.livein_begin(),
         E = MF.livein_end(); LI != E; ++LI)
      if (LI->second)
        MRI->copyRegToReg(*MF.begin(), MF.begin()->end(), LI->second,
                          LI->first, RegMap->getRegClass(LI->second));
  }
  
  
  // Finally, emit the code for all of the scheduled instructions.
  std::map<SDNode*, unsigned> VRBaseMap;
  for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
    if (SUnit *SU = Sequence[i]) {
      for (unsigned j = 0, ee = SU->FlaggedNodes.size(); j != ee; j++)
        EmitNode(SU->FlaggedNodes[j], VRBaseMap);
      EmitNode(SU->Node, VRBaseMap);
    } else {
      // Null SUnit* is a noop.
      EmitNoop();
    }
  }
}

/// dump - dump the schedule.
void ScheduleDAG::dumpSchedule() const {
  for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
    if (SUnit *SU = Sequence[i])
      SU->dump(&DAG);
    else
      std::cerr << "**** NOOP ****\n";
  }
}


/// Run - perform scheduling.
///
MachineBasicBlock *ScheduleDAG::Run() {
  TII = TM.getInstrInfo();
  MRI = TM.getRegisterInfo();
  RegMap = BB->getParent()->getSSARegMap();
  ConstPool = BB->getParent()->getConstantPool();

  Schedule();
  return BB;
}

/// SUnit - Scheduling unit. It's an wrapper around either a single SDNode or
/// a group of nodes flagged together.
void SUnit::dump(const SelectionDAG *G) const {
  std::cerr << "SU(" << NodeNum << "): ";
  Node->dump(G);
  std::cerr << "\n";
  if (FlaggedNodes.size() != 0) {
    for (unsigned i = 0, e = FlaggedNodes.size(); i != e; i++) {
      std::cerr << "    ";
      FlaggedNodes[i]->dump(G);
      std::cerr << "\n";
    }
  }
}

void SUnit::dumpAll(const SelectionDAG *G) const {
  dump(G);

  std::cerr << "  # preds left       : " << NumPredsLeft << "\n";
  std::cerr << "  # succs left       : " << NumSuccsLeft << "\n";
  std::cerr << "  # chain preds left : " << NumChainPredsLeft << "\n";
  std::cerr << "  # chain succs left : " << NumChainSuccsLeft << "\n";
  std::cerr << "  Latency            : " << Latency << "\n";
  std::cerr << "  Depth              : " << Depth << "\n";
  std::cerr << "  Height             : " << Height << "\n";

  if (Preds.size() != 0) {
    std::cerr << "  Predecessors:\n";
    for (std::set<std::pair<SUnit*,bool> >::const_iterator I = Preds.begin(),
           E = Preds.end(); I != E; ++I) {
      if (I->second)
        std::cerr << "   ch  ";
      else
        std::cerr << "   val ";
      I->first->dump(G);
    }
  }
  if (Succs.size() != 0) {
    std::cerr << "  Successors:\n";
    for (std::set<std::pair<SUnit*, bool> >::const_iterator I = Succs.begin(),
           E = Succs.end(); I != E; ++I) {
      if (I->second)
        std::cerr << "   ch  ";
      else
        std::cerr << "   val ";
      I->first->dump(G);
    }
  }
  std::cerr << "\n";
}