lib/CodeGen/ScheduleDAG.cpp

   1 //===---- ScheduleDAG.cpp - Implement the ScheduleDAG class ---------------===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // This implements the ScheduleDAG class, which is a base class used by
  11 // scheduling implementation classes.
  12 //
  13 //===----------------------------------------------------------------------===//
  14
  15 #define DEBUG_TYPE "pre-RA-sched"
  16 #include "llvm/CodeGen/ScheduleDAG.h"
  17 #include "llvm/CodeGen/ScheduleHazardRecognizer.h"
  18 #include "llvm/Target/TargetMachine.h"
  19 #include "llvm/Target/TargetInstrInfo.h"
  20 #include "llvm/Target/TargetRegisterInfo.h"
  21 #include "llvm/Support/Debug.h"
  22 #include <climits>
  23 using namespace llvm;
  24
  25 ScheduleDAG::ScheduleDAG(MachineFunction &mf)
  26   : TM(mf.getTarget()),
  27     TII(TM.getInstrInfo()),
  28     TRI(TM.getRegisterInfo()),
  29     TLI(TM.getTargetLowering()),
  30     MF(mf), MRI(mf.getRegInfo()),
  31     ConstPool(MF.getConstantPool()),
  32     EntrySU(), ExitSU() {
  33 }
  34
  35 ScheduleDAG::~ScheduleDAG() {}
  36
  37 /// dump - dump the schedule.
  38 void ScheduleDAG::dumpSchedule() const {
  39   for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
  40     if (SUnit *SU = Sequence[i])
  41       SU->dump(this);
  42     else
  43       cerr << "**** NOOP ****\n";
  44   }
  45 }
  46
  47
  48 /// Run - perform scheduling.
  49 ///
  50 void ScheduleDAG::Run(MachineBasicBlock *bb,
  51                       MachineBasicBlock::iterator insertPos) {
  52   BB = bb;
  53   InsertPos = insertPos;
  54
  55   SUnits.clear();
  56   Sequence.clear();
  57   EntrySU = SUnit();
  58   ExitSU = SUnit();
  59
  60   Schedule();
  61
  62   DOUT << "*** Final schedule ***\n";
  63   DEBUG(dumpSchedule());
  64   DOUT << "\n";
  65 }
  66
  67 /// addPred - This adds the specified edge as a pred of the current node if
  68 /// not already.  It also adds the current node as a successor of the
  69 /// specified node.
  70 void SUnit::addPred(const SDep &D) {
  71   // If this node already has this depenence, don't add a redundant one.
  72   for (SmallVector<SDep, 4>::const_iterator I = Preds.begin(), E = Preds.end();
  73        I != E; ++I)
  74     if (*I == D)
  75       return;
  76   // Now add a corresponding succ to N.
  77   SDep P = D;
  78   P.setSUnit(this);
  79   SUnit *N = D.getSUnit();
  80   // Update the bookkeeping.
  81   if (D.getKind() == SDep::Data) {
  82     ++NumPreds;
  83     ++N->NumSuccs;
  84   }
  85   if (!N->isScheduled)
  86     ++NumPredsLeft;
  87   if (!isScheduled)
  88     ++N->NumSuccsLeft;
  89   Preds.push_back(D);
  90   N->Succs.push_back(P);
  91   if (P.getLatency() != 0) {
  92     this->setDepthDirty();
  93     N->setHeightDirty();
  94   }
  95 }
  96
  97 /// removePred - This removes the specified edge as a pred of the current
  98 /// node if it exists.  It also removes the current node as a successor of
  99 /// the specified node.
 100 void SUnit::removePred(const SDep &D) {
 101   // Find the matching predecessor.
 102   for (SmallVector<SDep, 4>::iterator I = Preds.begin(), E = Preds.end();
 103        I != E; ++I)
 104     if (*I == D) {
 105       bool FoundSucc = false;
 106       // Find the corresponding successor in N.
 107       SDep P = D;
 108       P.setSUnit(this);
 109       SUnit *N = D.getSUnit();
 110       for (SmallVector<SDep, 4>::iterator II = N->Succs.begin(),
 111              EE = N->Succs.end(); II != EE; ++II)
 112         if (*II == P) {
 113           FoundSucc = true;
 114           N->Succs.erase(II);
 115           break;
 116         }
 117       assert(FoundSucc && "Mismatching preds / succs lists!");
 118       Preds.erase(I);
 119       // Update the bookkeeping.
 120       if (P.getKind() == SDep::Data) {
 121         --NumPreds;
 122         --N->NumSuccs;
 123       }
 124       if (!N->isScheduled)
 125         --NumPredsLeft;
 126       if (!isScheduled)
 127         --N->NumSuccsLeft;
 128       if (P.getLatency() != 0) {
 129         this->setDepthDirty();
 130         N->setHeightDirty();
 131       }
 132       return;
 133     }
 134 }
 135
 136 void SUnit::setDepthDirty() {
 137   if (!isDepthCurrent) return;
 138   SmallVector<SUnit*, 8> WorkList;
 139   WorkList.push_back(this);
 140   do {
 141     SUnit *SU = WorkList.pop_back_val();
 142     SU->isDepthCurrent = false;
 143     for (SUnit::const_succ_iterator I = SU->Succs.begin(),
 144          E = SU->Succs.end(); I != E; ++I) {
 145       SUnit *SuccSU = I->getSUnit();
 146       if (SuccSU->isDepthCurrent)
 147         WorkList.push_back(SuccSU);
 148     }
 149   } while (!WorkList.empty());
 150 }
 151
 152 void SUnit::setHeightDirty() {
 153   if (!isHeightCurrent) return;
 154   SmallVector<SUnit*, 8> WorkList;
 155   WorkList.push_back(this);
 156   do {
 157     SUnit *SU = WorkList.pop_back_val();
 158     SU->isHeightCurrent = false;
 159     for (SUnit::const_pred_iterator I = SU->Preds.begin(),
 160          E = SU->Preds.end(); I != E; ++I) {
 161       SUnit *PredSU = I->getSUnit();
 162       if (PredSU->isHeightCurrent)
 163         WorkList.push_back(PredSU);
 164     }
 165   } while (!WorkList.empty());
 166 }
 167
 168 /// setDepthToAtLeast - Update this node's successors to reflect the
 169 /// fact that this node's depth just increased.
 170 ///
 171 void SUnit::setDepthToAtLeast(unsigned NewDepth) {
 172   if (NewDepth <= getDepth())
 173     return;
 174   setDepthDirty();
 175   Depth = NewDepth;
 176   isDepthCurrent = true;
 177 }
 178
 179 /// setHeightToAtLeast - Update this node's predecessors to reflect the
 180 /// fact that this node's height just increased.
 181 ///
 182 void SUnit::setHeightToAtLeast(unsigned NewHeight) {
 183   if (NewHeight <= getHeight())
 184     return;
 185   setHeightDirty();
 186   Height = NewHeight;
 187   isHeightCurrent = true;
 188 }
 189
 190 /// ComputeDepth - Calculate the maximal path from the node to the exit.
 191 ///
 192 void SUnit::ComputeDepth() {
 193   SmallVector<SUnit*, 8> WorkList;
 194   WorkList.push_back(this);
 195   do {
 196     SUnit *Cur = WorkList.back();
 197
 198     bool Done = true;
 199     unsigned MaxPredDepth = 0;
 200     for (SUnit::const_pred_iterator I = Cur->Preds.begin(),
 201          E = Cur->Preds.end(); I != E; ++I) {
 202       SUnit *PredSU = I->getSUnit();
 203       if (PredSU->isDepthCurrent)
 204         MaxPredDepth = std::max(MaxPredDepth,
 205                                 PredSU->Depth + I->getLatency());
 206       else {
 207         Done = false;
 208         WorkList.push_back(PredSU);
 209       }
 210     }
 211
 212     if (Done) {
 213       WorkList.pop_back();
 214       if (MaxPredDepth != Cur->Depth) {
 215         Cur->setDepthDirty();
 216         Cur->Depth = MaxPredDepth;
 217       }
 218       Cur->isDepthCurrent = true;
 219     }
 220   } while (!WorkList.empty());
 221 }
 222
 223 /// ComputeHeight - Calculate the maximal path from the node to the entry.
 224 ///
 225 void SUnit::ComputeHeight() {
 226   SmallVector<SUnit*, 8> WorkList;
 227   WorkList.push_back(this);
 228   do {
 229     SUnit *Cur = WorkList.back();
 230
 231     bool Done = true;
 232     unsigned MaxSuccHeight = 0;
 233     for (SUnit::const_succ_iterator I = Cur->Succs.begin(),
 234          E = Cur->Succs.end(); I != E; ++I) {
 235       SUnit *SuccSU = I->getSUnit();
 236       if (SuccSU->isHeightCurrent)
 237         MaxSuccHeight = std::max(MaxSuccHeight,
 238                                  SuccSU->Height + I->getLatency());
 239       else {
 240         Done = false;
 241         WorkList.push_back(SuccSU);
 242       }
 243     }
 244
 245     if (Done) {
 246       WorkList.pop_back();
 247       if (MaxSuccHeight != Cur->Height) {
 248         Cur->setHeightDirty();
 249         Cur->Height = MaxSuccHeight;
 250       }
 251       Cur->isHeightCurrent = true;
 252     }
 253   } while (!WorkList.empty());
 254 }
 255
 256 /// SUnit - Scheduling unit. It's an wrapper around either a single SDNode or
 257 /// a group of nodes flagged together.
 258 void SUnit::dump(const ScheduleDAG *G) const {
 259   cerr << "SU(" << NodeNum << "): ";
 260   G->dumpNode(this);
 261 }
 262
 263 void SUnit::dumpAll(const ScheduleDAG *G) const {
 264   dump(G);
 265
 266   cerr << "  # preds left       : " << NumPredsLeft << "\n";
 267   cerr << "  # succs left       : " << NumSuccsLeft << "\n";
 268   cerr << "  Latency            : " << Latency << "\n";
 269   cerr << "  Depth              : " << Depth << "\n";
 270   cerr << "  Height             : " << Height << "\n";
 271
 272   if (Preds.size() != 0) {
 273     cerr << "  Predecessors:\n";
 274     for (SUnit::const_succ_iterator I = Preds.begin(), E = Preds.end();
 275          I != E; ++I) {
 276       cerr << "   ";
 277       switch (I->getKind()) {
 278       case SDep::Data:        cerr << "val "; break;
 279       case SDep::Anti:        cerr << "anti"; break;
 280       case SDep::Output:      cerr << "out "; break;
 281       case SDep::Order:       cerr << "ch  "; break;
 282       }
 283       cerr << "#";
 284       cerr << I->getSUnit() << " - SU(" << I->getSUnit()->NodeNum << ")";
 285       if (I->isArtificial())
 286         cerr << " *";
 287       cerr << "\n";
 288     }
 289   }
 290   if (Succs.size() != 0) {
 291     cerr << "  Successors:\n";
 292     for (SUnit::const_succ_iterator I = Succs.begin(), E = Succs.end();
 293          I != E; ++I) {
 294       cerr << "   ";
 295       switch (I->getKind()) {
 296       case SDep::Data:        cerr << "val "; break;
 297       case SDep::Anti:        cerr << "anti"; break;
 298       case SDep::Output:      cerr << "out "; break;
 299       case SDep::Order:       cerr << "ch  "; break;
 300       }
 301       cerr << "#";
 302       cerr << I->getSUnit() << " - SU(" << I->getSUnit()->NodeNum << ")";
 303       if (I->isArtificial())
 304         cerr << " *";
 305       cerr << "\n";
 306     }
 307   }
 308   cerr << "\n";
 309 }
 310
 311 #ifndef NDEBUG
 312 /// VerifySchedule - Verify that all SUnits were scheduled and that
 313 /// their state is consistent.
 314 ///
 315 void ScheduleDAG::VerifySchedule(bool isBottomUp) {
 316   bool AnyNotSched = false;
 317   unsigned DeadNodes = 0;
 318   unsigned Noops = 0;
 319   for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
 320     if (!SUnits[i].isScheduled) {
 321       if (SUnits[i].NumPreds == 0 && SUnits[i].NumSuccs == 0) {
 322         ++DeadNodes;
 323         continue;
 324       }
 325       if (!AnyNotSched)
 326         cerr << "*** Scheduling failed! ***\n";
 327       SUnits[i].dump(this);
 328       cerr << "has not been scheduled!\n";
 329       AnyNotSched = true;
 330     }
 331     if (SUnits[i].isScheduled &&
 332         (isBottomUp ? SUnits[i].getHeight() : SUnits[i].getHeight()) >
 333           unsigned(INT_MAX)) {
 334       if (!AnyNotSched)
 335         cerr << "*** Scheduling failed! ***\n";
 336       SUnits[i].dump(this);
 337       cerr << "has an unexpected "
 338            << (isBottomUp ? "Height" : "Depth") << " value!\n";
 339       AnyNotSched = true;
 340     }
 341     if (isBottomUp) {
 342       if (SUnits[i].NumSuccsLeft != 0) {
 343         if (!AnyNotSched)
 344           cerr << "*** Scheduling failed! ***\n";
 345         SUnits[i].dump(this);
 346         cerr << "has successors left!\n";
 347         AnyNotSched = true;
 348       }
 349     } else {
 350       if (SUnits[i].NumPredsLeft != 0) {
 351         if (!AnyNotSched)
 352           cerr << "*** Scheduling failed! ***\n";
 353         SUnits[i].dump(this);
 354         cerr << "has predecessors left!\n";
 355         AnyNotSched = true;
 356       }
 357     }
 358   }
 359   for (unsigned i = 0, e = Sequence.size(); i != e; ++i)
 360     if (!Sequence[i])
 361       ++Noops;
 362   assert(!AnyNotSched);
 363   assert(Sequence.size() + DeadNodes - Noops == SUnits.size() &&
 364          "The number of nodes scheduled doesn't match the expected number!");
 365 }
 366 #endif
 367
 368 /// InitDAGTopologicalSorting - create the initial topological
 369 /// ordering from the DAG to be scheduled.
 370 ///
 371 /// The idea of the algorithm is taken from
 372 /// "Online algorithms for managing the topological order of
 373 /// a directed acyclic graph" by David J. Pearce and Paul H.J. Kelly
 374 /// This is the MNR algorithm, which was first introduced by
 375 /// A. Marchetti-Spaccamela, U. Nanni and H. Rohnert in
 376 /// "Maintaining a topological order under edge insertions".
 377 ///
 378 /// Short description of the algorithm:
 379 ///
 380 /// Topological ordering, ord, of a DAG maps each node to a topological
 381 /// index so that for all edges X->Y it is the case that ord(X) < ord(Y).
 382 ///
 383 /// This means that if there is a path from the node X to the node Z,
 384 /// then ord(X) < ord(Z).
 385 ///
 386 /// This property can be used to check for reachability of nodes:
 387 /// if Z is reachable from X, then an insertion of the edge Z->X would
 388 /// create a cycle.
 389 ///
 390 /// The algorithm first computes a topological ordering for the DAG by
 391 /// initializing the Index2Node and Node2Index arrays and then tries to keep
 392 /// the ordering up-to-date after edge insertions by reordering the DAG.
 393 ///
 394 /// On insertion of the edge X->Y, the algorithm first marks by calling DFS
 395 /// the nodes reachable from Y, and then shifts them using Shift to lie
 396 /// immediately after X in Index2Node.
 397 void ScheduleDAGTopologicalSort::InitDAGTopologicalSorting() {
 398   unsigned DAGSize = SUnits.size();
 399   std::vector<SUnit*> WorkList;
 400   WorkList.reserve(DAGSize);
 401
 402   Index2Node.resize(DAGSize);
 403   Node2Index.resize(DAGSize);
 404
 405   // Initialize the data structures.
 406   for (unsigned i = 0, e = DAGSize; i != e; ++i) {
 407     SUnit *SU = &SUnits[i];
 408     int NodeNum = SU->NodeNum;
 409     unsigned Degree = SU->Succs.size();
 410     // Temporarily use the Node2Index array as scratch space for degree counts.
 411     Node2Index[NodeNum] = Degree;
 412
 413     // Is it a node without dependencies?
 414     if (Degree == 0) {
 415       assert(SU->Succs.empty() && "SUnit should have no successors");
 416       // Collect leaf nodes.
 417       WorkList.push_back(SU);
 418     }
 419   }
 420
 421   int Id = DAGSize;
 422   while (!WorkList.empty()) {
 423     SUnit *SU = WorkList.back();
 424     WorkList.pop_back();
 425     Allocate(SU->NodeNum, --Id);
 426     for (SUnit::const_pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 427          I != E; ++I) {
 428       SUnit *SU = I->getSUnit();
 429       if (!--Node2Index[SU->NodeNum])
 430         // If all dependencies of the node are processed already,
 431         // then the node can be computed now.
 432         WorkList.push_back(SU);
 433     }
 434   }
 435
 436   Visited.resize(DAGSize);
 437
 438 #ifndef NDEBUG
 439   // Check correctness of the ordering
 440   for (unsigned i = 0, e = DAGSize; i != e; ++i) {
 441     SUnit *SU = &SUnits[i];
 442     for (SUnit::const_pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 443          I != E; ++I) {
 444       assert(Node2Index[SU->NodeNum] > Node2Index[I->getSUnit()->NodeNum] &&
 445       "Wrong topological sorting");
 446     }
 447   }
 448 #endif
 449 }
 450
 451 /// AddPred - Updates the topological ordering to accomodate an edge
 452 /// to be added from SUnit X to SUnit Y.
 453 void ScheduleDAGTopologicalSort::AddPred(SUnit *Y, SUnit *X) {
 454   int UpperBound, LowerBound;
 455   LowerBound = Node2Index[Y->NodeNum];
 456   UpperBound = Node2Index[X->NodeNum];
 457   bool HasLoop = false;
 458   // Is Ord(X) < Ord(Y) ?
 459   if (LowerBound < UpperBound) {
 460     // Update the topological order.
 461     Visited.reset();
 462     DFS(Y, UpperBound, HasLoop);
 463     assert(!HasLoop && "Inserted edge creates a loop!");
 464     // Recompute topological indexes.
 465     Shift(Visited, LowerBound, UpperBound);
 466   }
 467 }
 468
 469 /// RemovePred - Updates the topological ordering to accomodate an
 470 /// an edge to be removed from the specified node N from the predecessors
 471 /// of the current node M.
 472 void ScheduleDAGTopologicalSort::RemovePred(SUnit *M, SUnit *N) {
 473   // InitDAGTopologicalSorting();
 474 }
 475
 476 /// DFS - Make a DFS traversal to mark all nodes reachable from SU and mark
 477 /// all nodes affected by the edge insertion. These nodes will later get new
 478 /// topological indexes by means of the Shift method.
 479 void ScheduleDAGTopologicalSort::DFS(const SUnit *SU, int UpperBound,
 480                                      bool& HasLoop) {
 481   std::vector<const SUnit*> WorkList;
 482   WorkList.reserve(SUnits.size());
 483
 484   WorkList.push_back(SU);
 485   do {
 486     SU = WorkList.back();
 487     WorkList.pop_back();
 488     Visited.set(SU->NodeNum);
 489     for (int I = SU->Succs.size()-1; I >= 0; --I) {
 490       int s = SU->Succs[I].getSUnit()->NodeNum;
 491       if (Node2Index[s] == UpperBound) {
 492         HasLoop = true;
 493         return;
 494       }
 495       // Visit successors if not already and in affected region.
 496       if (!Visited.test(s) && Node2Index[s] < UpperBound) {
 497         WorkList.push_back(SU->Succs[I].getSUnit());
 498       }
 499     }
 500   } while (!WorkList.empty());
 501 }
 502
 503 /// Shift - Renumber the nodes so that the topological ordering is
 504 /// preserved.
 505 void ScheduleDAGTopologicalSort::Shift(BitVector& Visited, int LowerBound,
 506                                        int UpperBound) {
 507   std::vector<int> L;
 508   int shift = 0;
 509   int i;
 510
 511   for (i = LowerBound; i <= UpperBound; ++i) {
 512     // w is node at topological index i.
 513     int w = Index2Node[i];
 514     if (Visited.test(w)) {
 515       // Unmark.
 516       Visited.reset(w);
 517       L.push_back(w);
 518       shift = shift + 1;
 519     } else {
 520       Allocate(w, i - shift);
 521     }
 522   }
 523
 524   for (unsigned j = 0; j < L.size(); ++j) {
 525     Allocate(L[j], i - shift);
 526     i = i + 1;
 527   }
 528 }
 529
 530
 531 /// WillCreateCycle - Returns true if adding an edge from SU to TargetSU will
 532 /// create a cycle.
 533 bool ScheduleDAGTopologicalSort::WillCreateCycle(SUnit *SU, SUnit *TargetSU) {
 534   if (IsReachable(TargetSU, SU))
 535     return true;
 536   for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
 537        I != E; ++I)
 538     if (I->isAssignedRegDep() &&
 539         IsReachable(TargetSU, I->getSUnit()))
 540       return true;
 541   return false;
 542 }
 543
 544 /// IsReachable - Checks if SU is reachable from TargetSU.
 545 bool ScheduleDAGTopologicalSort::IsReachable(const SUnit *SU,
 546                                              const SUnit *TargetSU) {
 547   // If insertion of the edge SU->TargetSU would create a cycle
 548   // then there is a path from TargetSU to SU.
 549   int UpperBound, LowerBound;
 550   LowerBound = Node2Index[TargetSU->NodeNum];
 551   UpperBound = Node2Index[SU->NodeNum];
 552   bool HasLoop = false;
 553   // Is Ord(TargetSU) < Ord(SU) ?
 554   if (LowerBound < UpperBound) {
 555     Visited.reset();
 556     // There may be a path from TargetSU to SU. Check for it.
 557     DFS(TargetSU, UpperBound, HasLoop);
 558   }
 559   return HasLoop;
 560 }
 561
 562 /// Allocate - assign the topological index to the node n.
 563 void ScheduleDAGTopologicalSort::Allocate(int n, int index) {
 564   Node2Index[n] = index;
 565   Index2Node[index] = n;
 566 }
 567
 568 ScheduleDAGTopologicalSort::ScheduleDAGTopologicalSort(
 569                                                      std::vector<SUnit> &sunits)
 570  : SUnits(sunits) {}
 571
 572 ScheduleHazardRecognizer::~ScheduleHazardRecognizer() {}