SmallVector<Value *, 4> Objs;
GetUnderlyingObjects(const_cast<Value *>(V), Objs);
- for (SmallVector<Value *, 4>::iterator I = Objs.begin(), IE = Objs.end();
+ for (SmallVectorImpl<Value *>::iterator I = Objs.begin(), IE = Objs.end();
I != IE; ++I) {
V = *I;
if (!Visited.insert(V))
} while (!Working.empty());
}
+typedef SmallVector<PointerIntPair<const Value *, 1, bool>, 4>
+UnderlyingObjectsVector;
+
/// getUnderlyingObjectsForInstr - If this machine instr has memory reference
/// information and it can be tracked to a normal reference to a known
/// object, return the Value for that object.
static void getUnderlyingObjectsForInstr(const MachineInstr *MI,
- const MachineFrameInfo *MFI,
- SmallVectorImpl<std::pair<const Value *, bool> > &Objects) {
+ const MachineFrameInfo *MFI,
+ UnderlyingObjectsVector &Objects) {
if (!MI->hasOneMemOperand() ||
!(*MI->memoperands_begin())->getValue() ||
(*MI->memoperands_begin())->isVolatile())
SmallVector<Value *, 4> Objs;
getUnderlyingObjects(V, Objs);
- for (SmallVector<Value *, 4>::iterator I = Objs.begin(), IE = Objs.end();
- I != IE; ++I) {
+ for (SmallVectorImpl<Value *>::iterator I = Objs.begin(), IE = Objs.end();
+ I != IE; ++I) {
bool MayAlias = true;
V = *I;
return;
}
- Objects.push_back(std::make_pair(V, MayAlias));
+ Objects.push_back(UnderlyingObjectsVector::value_type(V, MayAlias));
}
}
if (UseOp < 0)
Dep = SDep(SU, SDep::Artificial);
else {
+ // Set the hasPhysRegDefs only for physreg defs that have a use within
+ // the scheduling region.
+ SU->hasPhysRegDefs = true;
Dep = SDep(SU, SDep::Data, *Alias);
RegUse = UseSU->getInstr();
- Dep.setMinLatency(
- SchedModel.computeOperandLatency(SU->getInstr(), OperIdx,
- RegUse, UseOp, /*FindMin=*/true));
}
Dep.setLatency(
- SchedModel.computeOperandLatency(SU->getInstr(), OperIdx,
- RegUse, UseOp, /*FindMin=*/false));
+ SchedModel.computeOperandLatency(SU->getInstr(), OperIdx, RegUse,
+ UseOp));
ST.adjustSchedDependency(SU, UseSU, Dep);
UseSU->addPred(Dep);
DefSU->addPred(SDep(SU, Kind, /*Reg=*/*Alias));
else {
SDep Dep(SU, Kind, /*Reg=*/*Alias);
- unsigned OutLatency =
- SchedModel.computeOutputLatency(MI, OperIdx, DefSU->getInstr());
- Dep.setMinLatency(OutLatency);
- Dep.setLatency(OutLatency);
+ Dep.setLatency(
+ SchedModel.computeOutputLatency(MI, OperIdx, DefSU->getInstr()));
DefSU->addPred(Dep);
}
}
}
if (!MO.isDef()) {
+ SU->hasPhysRegUses = true;
// Either insert a new Reg2SUnits entry with an empty SUnits list, or
// retrieve the existing SUnits list for this register's uses.
// Push this SUnit on the use list.
SUnit *DefSU = DefI->SU;
if (DefSU != SU && DefSU != &ExitSU) {
SDep Dep(SU, SDep::Output, Reg);
- unsigned OutLatency =
- SchedModel.computeOutputLatency(MI, OperIdx, DefSU->getInstr());
- Dep.setMinLatency(OutLatency);
- Dep.setLatency(OutLatency);
+ Dep.setLatency(
+ SchedModel.computeOutputLatency(MI, OperIdx, DefSU->getInstr()));
DefSU->addPred(Dep);
}
DefI->SU = SU;
// Adjust the dependence latency using operand def/use information, then
// allow the target to perform its own adjustments.
int DefOp = Def->findRegisterDefOperandIdx(Reg);
- dep.setLatency(
- SchedModel.computeOperandLatency(Def, DefOp, MI, OperIdx, false));
- dep.setMinLatency(
- SchedModel.computeOperandLatency(Def, DefOp, MI, OperIdx, true));
+ dep.setLatency(SchedModel.computeOperandLatency(Def, DefOp, MI, OperIdx));
const TargetSubtargetInfo &ST = TM.getSubtarget<TargetSubtargetInfo>();
ST.adjustSchedDependency(DefSU, SU, const_cast<SDep &>(dep));
SmallVector<Value *, 4> Objs;
getUnderlyingObjects(V, Objs);
- for (SmallVector<Value *, 4>::iterator I = Objs.begin(),
- IE = Objs.end(); I != IE; ++I) {
+ for (SmallVectorImpl<Value *>::iterator I = Objs.begin(),
+ IE = Objs.end(); I != IE; ++I) {
V = *I;
if (const PseudoSourceValue *PSV = dyn_cast<PseudoSourceValue>(V)) {
assert(RPTracker->getPos() == prior(MII) && "RPTracker can't find MI");
}
- assert((!MI->isTerminator() || CanHandleTerminators) && !MI->isLabel() &&
+ assert((CanHandleTerminators || (!MI->isTerminator() && !MI->isLabel())) &&
"Cannot schedule terminators or labels!");
SUnit *SU = MISUnitMap[MI];
AliasMemDefs.clear();
AliasMemUses.clear();
} else if (MI->mayStore()) {
- SmallVector<std::pair<const Value *, bool>, 4> Objs;
+ UnderlyingObjectsVector Objs;
getUnderlyingObjectsForInstr(MI, MFI, Objs);
if (Objs.empty()) {
}
bool MayAlias = false;
- for (SmallVector<std::pair<const Value *, bool>, 4>::iterator
- K = Objs.begin(), KE = Objs.end(); K != KE; ++K) {
- const Value *V = K->first;
- bool ThisMayAlias = K->second;
+ for (UnderlyingObjectsVector::iterator K = Objs.begin(), KE = Objs.end();
+ K != KE; ++K) {
+ const Value *V = K->getPointer();
+ bool ThisMayAlias = K->getInt();
if (ThisMayAlias)
MayAlias = true;
if (MI->isInvariantLoad(AA)) {
// Invariant load, no chain dependencies needed!
} else {
- SmallVector<std::pair<const Value *, bool>, 4> Objs;
+ UnderlyingObjectsVector Objs;
getUnderlyingObjectsForInstr(MI, MFI, Objs);
if (Objs.empty()) {
MayAlias = false;
}
- for (SmallVector<std::pair<const Value *, bool>, 4>::iterator
+ for (UnderlyingObjectsVector::iterator
J = Objs.begin(), JE = Objs.end(); J != JE; ++J) {
- const Value *V = J->first;
- bool ThisMayAlias = J->second;
+ const Value *V = J->getPointer();
+ bool ThisMayAlias = J->getInt();
if (ThisMayAlias)
MayAlias = true;
else if (SU == &ExitSU)
oss << "<exit>";
else
- SU->getInstr()->print(oss);
+ SU->getInstr()->print(oss, &TM, /*SkipOpers=*/true);
return oss.str();
}
/// List PredSU, SuccSU pairs that represent data edges between subtrees.
std::vector<std::pair<const SUnit*, const SUnit*> > ConnectionPairs;
+ struct RootData {
+ unsigned NodeID;
+ unsigned ParentNodeID; // Parent node (member of the parent subtree).
+ unsigned SubInstrCount; // Instr count in this tree only, not children.
+
+ RootData(unsigned id): NodeID(id),
+ ParentNodeID(SchedDFSResult::InvalidSubtreeID),
+ SubInstrCount(0) {}
+
+ unsigned getSparseSetIndex() const { return NodeID; }
+ };
+
+ SparseSet<RootData> RootSet;
+
public:
- SchedDFSImpl(SchedDFSResult &r): R(r), SubtreeClasses(R.DFSData.size()) {}
+ SchedDFSImpl(SchedDFSResult &r): R(r), SubtreeClasses(R.DFSNodeData.size()) {
+ RootSet.setUniverse(R.DFSNodeData.size());
+ }
/// Return true if this node been visited by the DFS traversal.
///
/// During visitPostorderNode the Node's SubtreeID is assigned to the Node
/// ID. Later, SubtreeID is updated but remains valid.
bool isVisited(const SUnit *SU) const {
- return R.DFSData[SU->NodeNum].SubtreeID != SchedDFSResult::InvalidSubtreeID;
+ return R.DFSNodeData[SU->NodeNum].SubtreeID
+ != SchedDFSResult::InvalidSubtreeID;
}
/// Initialize this node's instruction count. We don't need to flag the node
/// visited until visitPostorder because the DAG cannot have cycles.
void visitPreorder(const SUnit *SU) {
- R.DFSData[SU->NodeNum].InstrCount = SU->getInstr()->isTransient() ? 0 : 1;
- R.DFSData[SU->NodeNum].SubInstrCount = R.DFSData[SU->NodeNum].InstrCount;
- }
-
- /// Called once for each tree edge after calling visitPostOrderNode on the
- /// predecessor. Increment the parent node's instruction count and
- /// preemptively join this subtree to its parent's if it is small enough.
- void visitPostorderEdge(const SDep &PredDep, const SUnit *Succ) {
- R.DFSData[Succ->NodeNum].InstrCount
- += R.DFSData[PredDep.getSUnit()->NodeNum].InstrCount;
- joinPredSubtree(PredDep, Succ);
+ R.DFSNodeData[SU->NodeNum].InstrCount =
+ SU->getInstr()->isTransient() ? 0 : 1;
}
/// Called once for each node after all predecessors are visited. Revisit this
void visitPostorderNode(const SUnit *SU) {
// Mark this node as the root of a subtree. It may be joined with its
// successors later.
- R.DFSData[SU->NodeNum].SubtreeID = SU->NodeNum;
+ R.DFSNodeData[SU->NodeNum].SubtreeID = SU->NodeNum;
+ RootData RData(SU->NodeNum);
+ RData.SubInstrCount = SU->getInstr()->isTransient() ? 0 : 1;
// If any predecessors are still in their own subtree, they either cannot be
// joined or are large enough to remain separate. If this parent node's
// total instruction count is not greater than a child subtree by at least
// the subtree limit, then try to join it now since splitting subtrees is
// only useful if multiple high-pressure paths are possible.
- unsigned InstrCount = R.DFSData[SU->NodeNum].InstrCount;
+ unsigned InstrCount = R.DFSNodeData[SU->NodeNum].InstrCount;
for (SUnit::const_pred_iterator
PI = SU->Preds.begin(), PE = SU->Preds.end(); PI != PE; ++PI) {
if (PI->getKind() != SDep::Data)
continue;
unsigned PredNum = PI->getSUnit()->NodeNum;
- if ((InstrCount - R.DFSData[PredNum].InstrCount) < R.SubtreeLimit)
+ if ((InstrCount - R.DFSNodeData[PredNum].InstrCount) < R.SubtreeLimit)
joinPredSubtree(*PI, SU, /*CheckLimit=*/false);
+
+ // Either link or merge the TreeData entry from the child to the parent.
+ if (R.DFSNodeData[PredNum].SubtreeID == PredNum) {
+ // If the predecessor's parent is invalid, this is a tree edge and the
+ // current node is the parent.
+ if (RootSet[PredNum].ParentNodeID == SchedDFSResult::InvalidSubtreeID)
+ RootSet[PredNum].ParentNodeID = SU->NodeNum;
+ }
+ else if (RootSet.count(PredNum)) {
+ // The predecessor is not a root, but is still in the root set. This
+ // must be the new parent that it was just joined to. Note that
+ // RootSet[PredNum].ParentNodeID may either be invalid or may still be
+ // set to the original parent.
+ RData.SubInstrCount += RootSet[PredNum].SubInstrCount;
+ RootSet.erase(PredNum);
+ }
}
+ RootSet[SU->NodeNum] = RData;
+ }
+
+ /// Called once for each tree edge after calling visitPostOrderNode on the
+ /// predecessor. Increment the parent node's instruction count and
+ /// preemptively join this subtree to its parent's if it is small enough.
+ void visitPostorderEdge(const SDep &PredDep, const SUnit *Succ) {
+ R.DFSNodeData[Succ->NodeNum].InstrCount
+ += R.DFSNodeData[PredDep.getSUnit()->NodeNum].InstrCount;
+ joinPredSubtree(PredDep, Succ);
}
/// Add a connection for cross edges.
/// between trees.
void finalize() {
SubtreeClasses.compress();
+ R.DFSTreeData.resize(SubtreeClasses.getNumClasses());
+ assert(SubtreeClasses.getNumClasses() == RootSet.size()
+ && "number of roots should match trees");
+ for (SparseSet<RootData>::const_iterator
+ RI = RootSet.begin(), RE = RootSet.end(); RI != RE; ++RI) {
+ unsigned TreeID = SubtreeClasses[RI->NodeID];
+ if (RI->ParentNodeID != SchedDFSResult::InvalidSubtreeID)
+ R.DFSTreeData[TreeID].ParentTreeID = SubtreeClasses[RI->ParentNodeID];
+ R.DFSTreeData[TreeID].SubInstrCount = RI->SubInstrCount;
+ // Note that SubInstrCount may be greater than InstrCount if we joined
+ // subtrees across a cross edge. InstrCount will be attributed to the
+ // original parent, while SubInstrCount will be attributed to the joined
+ // parent.
+ }
R.SubtreeConnections.resize(SubtreeClasses.getNumClasses());
R.SubtreeConnectLevels.resize(SubtreeClasses.getNumClasses());
DEBUG(dbgs() << R.getNumSubtrees() << " subtrees:\n");
- for (unsigned Idx = 0, End = R.DFSData.size(); Idx != End; ++Idx) {
- R.DFSData[Idx].SubtreeID = SubtreeClasses[Idx];
+ for (unsigned Idx = 0, End = R.DFSNodeData.size(); Idx != End; ++Idx) {
+ R.DFSNodeData[Idx].SubtreeID = SubtreeClasses[Idx];
DEBUG(dbgs() << " SU(" << Idx << ") in tree "
- << R.DFSData[Idx].SubtreeID << '\n');
+ << R.DFSNodeData[Idx].SubtreeID << '\n');
}
for (std::vector<std::pair<const SUnit*, const SUnit*> >::const_iterator
I = ConnectionPairs.begin(), E = ConnectionPairs.end();
// Check if the predecessor is already joined.
const SUnit *PredSU = PredDep.getSUnit();
unsigned PredNum = PredSU->NodeNum;
- if (R.DFSData[PredNum].SubtreeID != PredNum)
+ if (R.DFSNodeData[PredNum].SubtreeID != PredNum)
return false;
// Four is the magic number of successors before a node is considered a
return false;
}
}
- if (CheckLimit && R.DFSData[PredNum].SubInstrCount > R.SubtreeLimit)
+ if (CheckLimit && R.DFSNodeData[PredNum].InstrCount > R.SubtreeLimit)
return false;
-
- R.DFSData[PredNum].SubtreeID = Succ->NodeNum;
- R.DFSData[Succ->NodeNum].SubInstrCount += R.DFSData[PredNum].SubInstrCount;
+ R.DFSNodeData[PredNum].SubtreeID = Succ->NodeNum;
SubtreeClasses.join(Succ->NodeNum, PredNum);
return true;
}
if (!Depth)
return;
- SmallVectorImpl<SchedDFSResult::Connection> &Connections =
- R.SubtreeConnections[FromTree];
- for (SmallVectorImpl<SchedDFSResult::Connection>::iterator
- I = Connections.begin(), E = Connections.end(); I != E; ++I) {
- if (I->TreeID == ToTree) {
- I->Level = std::max(I->Level, Depth);
- return;
+ do {
+ SmallVectorImpl<SchedDFSResult::Connection> &Connections =
+ R.SubtreeConnections[FromTree];
+ for (SmallVectorImpl<SchedDFSResult::Connection>::iterator
+ I = Connections.begin(), E = Connections.end(); I != E; ++I) {
+ if (I->TreeID == ToTree) {
+ I->Level = std::max(I->Level, Depth);
+ return;
+ }
}
- }
- Connections.push_back(SchedDFSResult::Connection(ToTree, Depth));
+ Connections.push_back(SchedDFSResult::Connection(ToTree, Depth));
+ FromTree = R.DFSTreeData[FromTree].ParentTreeID;
+ } while (FromTree != SchedDFSResult::InvalidSubtreeID);
}
};
} // namespace llvm
static bool hasDataSucc(const SUnit *SU) {
for (SUnit::const_succ_iterator
SI = SU->Succs.begin(), SE = SU->Succs.end(); SI != SE; ++SI) {
- if (SI->getKind() == SDep::Data)
+ if (SI->getKind() == SDep::Data && !SI->getSUnit()->isBoundaryNode())
return true;
}
return false;
const SDep &PredDep = *DFS.getPred();
DFS.advance();
// Ignore non-data edges.
- if (PredDep.getKind() != SDep::Data)
+ if (PredDep.getKind() != SDep::Data
+ || PredDep.getSUnit()->isBoundaryNode()) {
continue;
+ }
// An already visited edge is a cross edge, assuming an acyclic DAG.
if (Impl.isVisited(PredDep.getSUnit())) {
Impl.visitCrossEdge(PredDep, DFS.getCurr());