aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorAndrew Trick <atrick@apple.com>2011-01-14 21:11:41 +0000
committerAndrew Trick <atrick@apple.com>2011-01-14 21:11:41 +0000
commitf697c8a19adf962a933b055383952e72789a0e20 (patch)
tree7639dabdbfe586b58714693cc519dc383272823a
parentd0f56132cfa6e25fb9692e84ea12444c86b92ae4 (diff)
Support for precise scheduling of the instruction selection DAG,
disabled in this checkin. Sorry for the large diffs due to refactoring. New functionality is all guarded by EnableSchedCycles. Scheduling the isel DAG is inherently imprecise, but we give it a best effort: - Added MayReduceRegPressure to allow stalled nodes in the queue only if there is a regpressure need. - Added BUHasStall to allow checking for either dependence stalls due to latency or resource stalls due to pipeline hazards. - Added BUCompareLatency to encapsulate and standardize the heuristics for minimizing stall cycles (vs. reducing register pressure). - Modified the bottom-up heuristic (now in BUCompareLatency) to prioritize nodes by their depth rather than height. As long as it doesn't stall, height is irrelevant. Depth represents the critical path to the DAG root. - Added hybrid_ls_rr_sort::isReady to filter stalled nodes before adding them to the available queue. Related Cleanup: most of the register reduction routines do not need to be templates. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@123468 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r--lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp1200
1 files changed, 663 insertions, 537 deletions
diff --git a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
index a51595f1b0..efcf932a48 100644
--- a/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
+++ b/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp
@@ -137,6 +137,8 @@ public:
void Schedule();
+ ScheduleHazardRecognizer *getHazardRec() { return HazardRec; }
+
/// IsReachable - Checks if SU is reachable from TargetSU.
bool IsReachable(const SUnit *SU, const SUnit *TargetSU) {
return Topo.IsReachable(SU, TargetSU);
@@ -1246,101 +1248,286 @@ void ScheduleDAGRRList::ListScheduleTopDown() {
//===----------------------------------------------------------------------===//
-// RegReductionPriorityQueue Implementation
+// RegReductionPriorityQueue Definition
//===----------------------------------------------------------------------===//
//
// This is a SchedulingPriorityQueue that schedules using Sethi Ullman numbers
// to reduce register pressure.
//
namespace {
- template<class SF>
- class RegReductionPriorityQueue;
+class RegReductionPQBase;
+
+struct queue_sort : public std::binary_function<SUnit*, SUnit*, bool> {
+ bool isReady(SUnit* SU, unsigned CurCycle) const { return true; }
+};
- struct queue_sort : public std::binary_function<SUnit*, SUnit*, bool> {
- bool isReady(SUnit* SU, unsigned CurCycle) const { return true; }
+/// bu_ls_rr_sort - Priority function for bottom up register pressure
+// reduction scheduler.
+struct bu_ls_rr_sort : public queue_sort {
+ enum {
+ IsBottomUp = true,
+ HasReadyFilter = false
};
- /// bu_ls_rr_sort - Priority function for bottom up register pressure
- // reduction scheduler.
- struct bu_ls_rr_sort : public queue_sort {
- enum {
- IsBottomUp = true,
- HasReadyFilter = false
- };
+ RegReductionPQBase *SPQ;
+ bu_ls_rr_sort(RegReductionPQBase *spq) : SPQ(spq) {}
+ bu_ls_rr_sort(const bu_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {}
- RegReductionPriorityQueue<bu_ls_rr_sort> *SPQ;
- bu_ls_rr_sort(RegReductionPriorityQueue<bu_ls_rr_sort> *spq) : SPQ(spq) {}
- bu_ls_rr_sort(const bu_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {}
+ bool operator()(SUnit* left, SUnit* right) const;
+};
- bool operator()(const SUnit* left, const SUnit* right) const;
+// td_ls_rr_sort - Priority function for top down register pressure reduction
+// scheduler.
+struct td_ls_rr_sort : public queue_sort {
+ enum {
+ IsBottomUp = false,
+ HasReadyFilter = false
};
- // td_ls_rr_sort - Priority function for top down register pressure reduction
- // scheduler.
- struct td_ls_rr_sort : public queue_sort {
- enum {
- IsBottomUp = false,
- HasReadyFilter = false
- };
+ RegReductionPQBase *SPQ;
+ td_ls_rr_sort(RegReductionPQBase *spq) : SPQ(spq) {}
+ td_ls_rr_sort(const td_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {}
- RegReductionPriorityQueue<td_ls_rr_sort> *SPQ;
- td_ls_rr_sort(RegReductionPriorityQueue<td_ls_rr_sort> *spq) : SPQ(spq) {}
- td_ls_rr_sort(const td_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {}
+ bool operator()(const SUnit* left, const SUnit* right) const;
+};
- bool operator()(const SUnit* left, const SUnit* right) const;
+// src_ls_rr_sort - Priority function for source order scheduler.
+struct src_ls_rr_sort : public queue_sort {
+ enum {
+ IsBottomUp = true,
+ HasReadyFilter = false
};
- // src_ls_rr_sort - Priority function for source order scheduler.
- struct src_ls_rr_sort : public queue_sort {
- enum {
- IsBottomUp = true,
- HasReadyFilter = false
- };
+ RegReductionPQBase *SPQ;
+ src_ls_rr_sort(RegReductionPQBase *spq)
+ : SPQ(spq) {}
+ src_ls_rr_sort(const src_ls_rr_sort &RHS)
+ : SPQ(RHS.SPQ) {}
- RegReductionPriorityQueue<src_ls_rr_sort> *SPQ;
- src_ls_rr_sort(RegReductionPriorityQueue<src_ls_rr_sort> *spq)
- : SPQ(spq) {}
- src_ls_rr_sort(const src_ls_rr_sort &RHS)
- : SPQ(RHS.SPQ) {}
+ bool operator()(SUnit* left, SUnit* right) const;
+};
- bool operator()(const SUnit* left, const SUnit* right) const;
+// hybrid_ls_rr_sort - Priority function for hybrid scheduler.
+struct hybrid_ls_rr_sort : public queue_sort {
+ enum {
+ IsBottomUp = true,
+ HasReadyFilter = true
};
- // hybrid_ls_rr_sort - Priority function for hybrid scheduler.
- struct hybrid_ls_rr_sort : public queue_sort {
- enum {
- IsBottomUp = true,
- HasReadyFilter = false
- };
+ RegReductionPQBase *SPQ;
+ hybrid_ls_rr_sort(RegReductionPQBase *spq)
+ : SPQ(spq) {}
+ hybrid_ls_rr_sort(const hybrid_ls_rr_sort &RHS)
+ : SPQ(RHS.SPQ) {}
+
+ bool isReady(SUnit *SU, unsigned CurCycle) const;
- RegReductionPriorityQueue<hybrid_ls_rr_sort> *SPQ;
- hybrid_ls_rr_sort(RegReductionPriorityQueue<hybrid_ls_rr_sort> *spq)
- : SPQ(spq) {}
- hybrid_ls_rr_sort(const hybrid_ls_rr_sort &RHS)
- : SPQ(RHS.SPQ) {}
+ bool operator()(SUnit* left, SUnit* right) const;
+};
- bool operator()(const SUnit* left, const SUnit* right) const;
+// ilp_ls_rr_sort - Priority function for ILP (instruction level parallelism)
+// scheduler.
+struct ilp_ls_rr_sort : public queue_sort {
+ enum {
+ IsBottomUp = true,
+ HasReadyFilter = true
};
- // ilp_ls_rr_sort - Priority function for ILP (instruction level parallelism)
- // scheduler.
- struct ilp_ls_rr_sort : public queue_sort {
- enum {
- IsBottomUp = true,
- HasReadyFilter = true
- };
+ RegReductionPQBase *SPQ;
+ ilp_ls_rr_sort(RegReductionPQBase *spq)
+ : SPQ(spq) {}
+ ilp_ls_rr_sort(const ilp_ls_rr_sort &RHS)
+ : SPQ(RHS.SPQ) {}
+
+ bool isReady(SUnit *SU, unsigned CurCycle) const;
- RegReductionPriorityQueue<ilp_ls_rr_sort> *SPQ;
- ilp_ls_rr_sort(RegReductionPriorityQueue<ilp_ls_rr_sort> *spq)
- : SPQ(spq) {}
- ilp_ls_rr_sort(const ilp_ls_rr_sort &RHS)
- : SPQ(RHS.SPQ) {}
+ bool operator()(SUnit* left, SUnit* right) const;
+};
- bool isReady(SUnit *SU, unsigned CurCycle) const;
+class RegReductionPQBase : public SchedulingPriorityQueue {
+protected:
+ std::vector<SUnit*> Queue;
+ unsigned CurQueueId;
+ bool TracksRegPressure;
- bool operator()(const SUnit* left, const SUnit* right) const;
- };
-} // end anonymous namespace
+ // SUnits - The SUnits for the current graph.
+ std::vector<SUnit> *SUnits;
+
+ MachineFunction &MF;
+ const TargetInstrInfo *TII;
+ const TargetRegisterInfo *TRI;
+ const TargetLowering *TLI;
+ ScheduleDAGRRList *scheduleDAG;
+
+ // SethiUllmanNumbers - The SethiUllman number for each node.
+ std::vector<unsigned> SethiUllmanNumbers;
+
+ /// RegPressure - Tracking current reg pressure per register class.
+ ///
+ std::vector<unsigned> RegPressure;
+
+ /// RegLimit - Tracking the number of allocatable registers per register
+ /// class.
+ std::vector<unsigned> RegLimit;
+
+public:
+ RegReductionPQBase(MachineFunction &mf,
+ bool hasReadyFilter,
+ bool tracksrp,
+ const TargetInstrInfo *tii,
+ const TargetRegisterInfo *tri,
+ const TargetLowering *tli)
+ : SchedulingPriorityQueue(hasReadyFilter),
+ CurQueueId(0), TracksRegPressure(tracksrp),
+ MF(mf), TII(tii), TRI(tri), TLI(tli), scheduleDAG(NULL) {
+ if (TracksRegPressure) {
+ unsigned NumRC = TRI->getNumRegClasses();
+ RegLimit.resize(NumRC);
+ RegPressure.resize(NumRC);
+ std::fill(RegLimit.begin(), RegLimit.end(), 0);
+ std::fill(RegPressure.begin(), RegPressure.end(), 0);
+ for (TargetRegisterInfo::regclass_iterator I = TRI->regclass_begin(),
+ E = TRI->regclass_end(); I != E; ++I)
+ RegLimit[(*I)->getID()] = tli->getRegPressureLimit(*I, MF);
+ }
+ }
+
+ void setScheduleDAG(ScheduleDAGRRList *scheduleDag) {
+ scheduleDAG = scheduleDag;
+ }
+
+ ScheduleHazardRecognizer* getHazardRec() {
+ return scheduleDAG->getHazardRec();
+ }
+
+ void initNodes(std::vector<SUnit> &sunits);
+
+ void addNode(const SUnit *SU);
+
+ void updateNode(const SUnit *SU);
+
+ void releaseState() {
+ SUnits = 0;
+ SethiUllmanNumbers.clear();
+ std::fill(RegPressure.begin(), RegPressure.end(), 0);
+ }
+
+ unsigned getNodePriority(const SUnit *SU) const;
+
+ unsigned getNodeOrdering(const SUnit *SU) const {
+ return scheduleDAG->DAG->GetOrdering(SU->getNode());
+ }
+
+ bool empty() const { return Queue.empty(); }
+
+ void push(SUnit *U) {
+ assert(!U->NodeQueueId && "Node in the queue already");
+ U->NodeQueueId = ++CurQueueId;
+ Queue.push_back(U);
+ }
+
+ void remove(SUnit *SU) {
+ assert(!Queue.empty() && "Queue is empty!");
+ assert(SU->NodeQueueId != 0 && "Not in queue!");
+ std::vector<SUnit *>::iterator I = std::find(Queue.begin(), Queue.end(),
+ SU);
+ if (I != prior(Queue.end()))
+ std::swap(*I, Queue.back());
+ Queue.pop_back();
+ SU->NodeQueueId = 0;
+ }
+
+ void dumpRegPressure() const;
+
+ bool HighRegPressure(const SUnit *SU) const;
+
+ bool MayReduceRegPressure(SUnit *SU);
+
+ void ScheduledNode(SUnit *SU);
+
+ void UnscheduledNode(SUnit *SU);
+
+protected:
+ bool canClobber(const SUnit *SU, const SUnit *Op);
+ void AddPseudoTwoAddrDeps();
+ void PrescheduleNodesWithMultipleUses();
+ void CalculateSethiUllmanNumbers();
+};
+
+template<class SF>
+class RegReductionPriorityQueue : public RegReductionPQBase {
+ static SUnit *popFromQueue(std::vector<SUnit*> &Q, SF &Picker) {
+ std::vector<SUnit *>::iterator Best = Q.begin();
+ for (std::vector<SUnit *>::iterator I = llvm::next(Q.begin()),
+ E = Q.end(); I != E; ++I)
+ if (Picker(*Best, *I))
+ Best = I;
+ SUnit *V = *Best;
+ if (Best != prior(Q.end()))
+ std::swap(*Best, Q.back());
+ Q.pop_back();
+ return V;
+ }
+
+ SF Picker;
+
+public:
+ RegReductionPriorityQueue(MachineFunction &mf,
+ bool tracksrp,
+ const TargetInstrInfo *tii,
+ const TargetRegisterInfo *tri,
+ const TargetLowering *tli)
+ : RegReductionPQBase(mf, SF::HasReadyFilter, tracksrp, tii, tri, tli),
+ Picker(this) {}
+
+ bool isBottomUp() const { return SF::IsBottomUp; }
+
+ bool isReady(SUnit *U) const {
+ return Picker.HasReadyFilter && Picker.isReady(U, getCurCycle());
+ }
+
+ SUnit *pop() {
+ if (Queue.empty()) return NULL;
+
+ SUnit *V = popFromQueue(Queue, Picker);
+ V->NodeQueueId = 0;
+ return V;
+ }
+
+ void dump(ScheduleDAG *DAG) const {
+ // Emulate pop() without clobbering NodeQueueIds.
+ std::vector<SUnit*> DumpQueue = Queue;
+ SF DumpPicker = Picker;
+ while (!DumpQueue.empty()) {
+ SUnit *SU = popFromQueue(DumpQueue, DumpPicker);
+ if (isBottomUp())
+ dbgs() << "Height " << SU->getHeight() << ": ";
+ else
+ dbgs() << "Depth " << SU->getDepth() << ": ";
+ SU->dump(DAG);
+ }
+ }
+};
+
+typedef RegReductionPriorityQueue<bu_ls_rr_sort>
+BURegReductionPriorityQueue;
+
+typedef RegReductionPriorityQueue<td_ls_rr_sort>
+TDRegReductionPriorityQueue;
+
+typedef RegReductionPriorityQueue<src_ls_rr_sort>
+SrcRegReductionPriorityQueue;
+
+typedef RegReductionPriorityQueue<hybrid_ls_rr_sort>
+HybridBURRPriorityQueue;
+
+typedef RegReductionPriorityQueue<ilp_ls_rr_sort>
+ILPBURRPriorityQueue;
+} // end anonymous namespace
+
+//===----------------------------------------------------------------------===//
+// Static Node Priority for Register Pressure Reduction
+//===----------------------------------------------------------------------===//
/// CalcNodeSethiUllmanNumber - Compute Sethi Ullman number.
/// Smaller number is the higher priority.
@@ -1371,436 +1558,324 @@ CalcNodeSethiUllmanNumber(const SUnit *SU, std::vector<unsigned> &SUNumbers) {
return SethiUllmanNumber;
}
-namespace {
- template<class SF>
- class RegReductionPriorityQueue : public SchedulingPriorityQueue {
- static SUnit *popFromQueue(std::vector<SUnit*> &Q, SF &Picker) {
- std::vector<SUnit *>::iterator Best = Q.begin();
- for (std::vector<SUnit *>::iterator I = llvm::next(Q.begin()),
- E = Q.end(); I != E; ++I)
- if (Picker(*Best, *I))
- Best = I;
- SUnit *V = *Best;
- if (Best != prior(Q.end()))
- std::swap(*Best, Q.back());
- Q.pop_back();
- return V;
- }
-
- std::vector<SUnit*> Queue;
- SF Picker;
- unsigned CurQueueId;
- bool TracksRegPressure;
-
- protected:
- // SUnits - The SUnits for the current graph.
- std::vector<SUnit> *SUnits;
-
- MachineFunction &MF;
- const TargetInstrInfo *TII;
- const TargetRegisterInfo *TRI;
- const TargetLowering *TLI;
- ScheduleDAGRRList *scheduleDAG;
-
- // SethiUllmanNumbers - The SethiUllman number for each node.
- std::vector<unsigned> SethiUllmanNumbers;
-
- /// RegPressure - Tracking current reg pressure per register class.
- ///
- std::vector<unsigned> RegPressure;
-
- /// RegLimit - Tracking the number of allocatable registers per register
- /// class.
- std::vector<unsigned> RegLimit;
-
- public:
- RegReductionPriorityQueue(MachineFunction &mf,
- bool tracksrp,
- const TargetInstrInfo *tii,
- const TargetRegisterInfo *tri,
- const TargetLowering *tli)
- : SchedulingPriorityQueue(SF::HasReadyFilter), Picker(this),
- CurQueueId(0), TracksRegPressure(tracksrp),
- MF(mf), TII(tii), TRI(tri), TLI(tli), scheduleDAG(NULL) {
- if (TracksRegPressure) {
- unsigned NumRC = TRI->getNumRegClasses();
- RegLimit.resize(NumRC);
- RegPressure.resize(NumRC);
- std::fill(RegLimit.begin(), RegLimit.end(), 0);
- std::fill(RegPressure.begin(), RegPressure.end(), 0);
- for (TargetRegisterInfo::regclass_iterator I = TRI->regclass_begin(),
- E = TRI->regclass_end(); I != E; ++I)
- RegLimit[(*I)->getID()] = tli->getRegPressureLimit(*I, MF);
- }
- }
+/// CalculateSethiUllmanNumbers - Calculate Sethi-Ullman numbers of all
+/// scheduling units.
+void RegReductionPQBase::CalculateSethiUllmanNumbers() {
+ SethiUllmanNumbers.assign(SUnits->size(), 0);
- bool isBottomUp() const { return SF::IsBottomUp; }
+ for (unsigned i = 0, e = SUnits->size(); i != e; ++i)
+ CalcNodeSethiUllmanNumber(&(*SUnits)[i], SethiUllmanNumbers);
+}
- void initNodes(std::vector<SUnit> &sunits) {
- SUnits = &sunits;
- // Add pseudo dependency edges for two-address nodes.
- AddPseudoTwoAddrDeps();
- // Reroute edges to nodes with multiple uses.
- PrescheduleNodesWithMultipleUses();
- // Calculate node priorities.
- CalculateSethiUllmanNumbers();
- }
+void RegReductionPQBase::initNodes(std::vector<SUnit> &sunits) {
+ SUnits = &sunits;
+ // Add pseudo dependency edges for two-address nodes.
+ AddPseudoTwoAddrDeps();
+ // Reroute edges to nodes with multiple uses.
+ PrescheduleNodesWithMultipleUses();
+ // Calculate node priorities.
+ CalculateSethiUllmanNumbers();
+}
- void addNode(const SUnit *SU) {
- unsigned SUSize = SethiUllmanNumbers.size();
- if (SUnits->size() > SUSize)
- SethiUllmanNumbers.resize(SUSize*2, 0);
- CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers);
- }
+void RegReductionPQBase::addNode(const SUnit *SU) {
+ unsigned SUSize = SethiUllmanNumbers.size();
+ if (SUnits->size() > SUSize)
+ SethiUllmanNumbers.resize(SUSize*2, 0);
+ CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers);
+}
- void updateNode(const SUnit *SU) {
- SethiUllmanNumbers[SU->NodeNum] = 0;
- CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers);
- }
+void RegReductionPQBase::updateNode(const SUnit *SU) {
+ SethiUllmanNumbers[SU->NodeNum] = 0;
+ CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers);
+}
- void releaseState() {
- SUnits = 0;
- SethiUllmanNumbers.clear();
- std::fill(RegPressure.begin(), RegPressure.end(), 0);
- }
+unsigned RegReductionPQBase::getNodePriority(const SUnit *SU) const {
+ assert(SU->NodeNum < SethiUllmanNumbers.size());
+ unsigned Opc = SU->getNode() ? SU->getNode()->getOpcode() : 0;
+ if (Opc == ISD::TokenFactor || Opc == ISD::CopyToReg)
+ // CopyToReg should be close to its uses to facilitate coalescing and
+ // avoid spilling.
+ return 0;
+ if (Opc == TargetOpcode::EXTRACT_SUBREG ||
+ Opc == TargetOpcode::SUBREG_TO_REG ||
+ Opc == TargetOpcode::INSERT_SUBREG)
+ // EXTRACT_SUBREG, INSERT_SUBREG, and SUBREG_TO_REG nodes should be
+ // close to their uses to facilitate coalescing.
+ return 0;
+ if (SU->NumSuccs == 0 && SU->NumPreds != 0)
+ // If SU does not have a register use, i.e. it doesn't produce a value
+ // that would be consumed (e.g. store), then it terminates a chain of
+ // computation. Give it a large SethiUllman number so it will be
+ // scheduled right before its predecessors that it doesn't lengthen
+ // their live ranges.
+ return 0xffff;
+ if (SU->NumPreds == 0 && SU->NumSuccs != 0)
+ // If SU does not have a register def, schedule it close to its uses
+ // because it does not lengthen any live ranges.
+ return 0;
+ return SethiUllmanNumbers[SU->NodeNum];
+}
- unsigned getNodePriority(const SUnit *SU) const {
- assert(SU->NodeNum < SethiUllmanNumbers.size());
- unsigned Opc = SU->getNode() ? SU->getNode()->getOpcode() : 0;
- if (Opc == ISD::TokenFactor || Opc == ISD::CopyToReg)
- // CopyToReg should be close to its uses to facilitate coalescing and
- // avoid spilling.
- return 0;
- if (Opc == TargetOpcode::EXTRACT_SUBREG ||
- Opc == TargetOpcode::SUBREG_TO_REG ||
- Opc == TargetOpcode::INSERT_SUBREG)
- // EXTRACT_SUBREG, INSERT_SUBREG, and SUBREG_TO_REG nodes should be
- // close to their uses to facilitate coalescing.
- return 0;
- if (SU->NumSuccs == 0 && SU->NumPreds != 0)
- // If SU does not have a register use, i.e. it doesn't produce a value
- // that would be consumed (e.g. store), then it terminates a chain of
- // computation. Give it a large SethiUllman number so it will be
- // scheduled right before its predecessors that it doesn't lengthen
- // their live ranges.
- return 0xffff;
- if (SU->NumPreds == 0 && SU->NumSuccs != 0)
- // If SU does not have a register def, schedule it close to its uses
- // because it does not lengthen any live ranges.
- return 0;
- return SethiUllmanNumbers[SU->NodeNum];
- }
+//===----------------------------------------------------------------------===//
+// Register Pressure Tracking
+//===----------------------------------------------------------------------===//
- unsigned getNodeOrdering(const SUnit *SU) const {
- return scheduleDAG->DAG->GetOrdering(SU->getNode());
- }
+void RegReductionPQBase::dumpRegPressure() const {
+ for (TargetRegisterInfo::regclass_iterator I = TRI->regclass_begin(),
+ E = TRI->regclass_end(); I != E; ++I) {
+ const TargetRegisterClass *RC = *I;
+ unsigned Id = RC->getID();
+ unsigned RP = RegPressure[Id];
+ if (!RP) continue;
+ DEBUG(dbgs() << RC->getName() << ": " << RP << " / " << RegLimit[Id]
+ << '\n');
+ }
+}
- bool empty() const { return Queue.empty(); }
+bool RegReductionPQBase::HighRegPressure(const SUnit *SU) const {
+ if (!TLI)
+ return false;
- bool isReady(SUnit *U) const {
- return Picker.HasReadyFilter && Picker.isReady(U, getCurCycle());
+ for (SUnit::const_pred_iterator I = SU->Preds.begin(),E = SU->Preds.end();
+ I != E; ++I) {
+ if (I->isCtrl())
+ continue;
+ SUnit *PredSU = I->getSUnit();
+ const SDNode *PN = PredSU->getNode();
+ if (!PN->isMachineOpcode()) {
+ if (PN->getOpcode() == ISD::CopyFromReg) {
+ EVT VT = PN->getValueType(0);
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ unsigned Cost = TLI->getRepRegClassCostFor(VT);
+ if ((RegPressure[RCId] + Cost) >= RegLimit[RCId])
+ return true;
+ }
+ continue;
}
-
- void push(SUnit *U) {
- assert(!U->NodeQueueId && "Node in the queue already");
- U->NodeQueueId = ++CurQueueId;
- Queue.push_back(U);
+ unsigned POpc = PN->getMachineOpcode();
+ if (POpc == TargetOpcode::IMPLICIT_DEF)
+ continue;
+ if (POpc == TargetOpcode::EXTRACT_SUBREG) {
+ EVT VT = PN->getOperand(0).getValueType();
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ unsigned Cost = TLI->getRepRegClassCostFor(VT);
+ // Check if this increases register pressure of the specific register
+ // class to the point where it would cause spills.
+ if ((RegPressure[RCId] + Cost) >= RegLimit[RCId])
+ return true;
+ continue;
+ } else if (POpc == TargetOpcode::INSERT_SUBREG ||
+ POpc == TargetOpcode::SUBREG_TO_REG) {
+ EVT VT = PN->getValueType(0);
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ unsigned Cost = TLI->getRepRegClassCostFor(VT);
+ // Check if this increases register pressure of the specific register
+ // class to the point where it would cause spills.
+ if ((RegPressure[RCId] + Cost) >= RegLimit[RCId])
+ return true;
+ continue;
}
-
- SUnit *pop() {
- if (Queue.empty()) return NULL;
-
- SUnit *V = popFromQueue(Queue, Picker);
- V->NodeQueueId = 0;
- return V;
+ unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs();
+ for (unsigned i = 0; i != NumDefs; ++i) {
+ EVT VT = PN->getValueType(i);
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ if (RegPressure[RCId] >= RegLimit[RCId])
+ return true; // Reg pressure already high.
+ unsigned Cost = TLI->getRepRegClassCostFor(VT);
+ if (!PN->hasAnyUseOfValue(i))
+ continue;
+ // Check if this increases register pressure of the specific register
+ // class to the point where it would cause spills.
+ if ((RegPressure[RCId] + Cost) >= RegLimit[RCId])
+ return true;
}
+ }
- void remove(SUnit *SU) {
- assert(!Queue.empty() && "Queue is empty!");
- assert(SU->NodeQueueId != 0 && "Not in queue!");
- std::vector<SUnit *>::iterator I = std::find(Queue.begin(), Queue.end(),
- SU);
- if (I != prior(Queue.end()))
- std::swap(*I, Queue.back());
- Queue.pop_back();
- SU->NodeQueueId = 0;
- }
+ return false;
+}
- bool HighRegPressure(const SUnit *SU) const {
- if (!TLI)
- return false;
+bool RegReductionPQBase::MayReduceRegPressure(SUnit *SU) {
+ const SDNode *N = SU->getNode();
- for (SUnit::const_pred_iterator I = SU->Preds.begin(),E = SU->Preds.end();
- I != E; ++I) {
- if (I->isCtrl())
- continue;
- SUnit *PredSU = I->getSUnit();
- const SDNode *PN = PredSU->getNode();
- if (!PN->isMachineOpcode()) {
- if (PN->getOpcode() == ISD::CopyFromReg) {
- EVT VT = PN->getValueType(0);
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- unsigned Cost = TLI->getRepRegClassCostFor(VT);
- if ((RegPressure[RCId] + Cost) >= RegLimit[RCId])
- return true;
- }
- continue;
- }
- unsigned POpc = PN->getMachineOpcode();
- if (POpc == TargetOpcode::IMPLICIT_DEF)
- continue;
- if (POpc == TargetOpcode::EXTRACT_SUBREG) {
- EVT VT = PN->getOperand(0).getValueType();
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- unsigned Cost = TLI->getRepRegClassCostFor(VT);
- // Check if this increases register pressure of the specific register
- // class to the point where it would cause spills.
- if ((RegPressure[RCId] + Cost) >= RegLimit[RCId])
- return true;
- continue;
- } else if (POpc == TargetOpcode::INSERT_SUBREG ||
- POpc == TargetOpcode::SUBREG_TO_REG) {
- EVT VT = PN->getValueType(0);
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- unsigned Cost = TLI->getRepRegClassCostFor(VT);
- // Check if this increases register pressure of the specific register
- // class to the point where it would cause spills.
- if ((RegPressure[RCId] + Cost) >= RegLimit[RCId])
- return true;
- continue;
- }
- unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs();
- for (unsigned i = 0; i != NumDefs; ++i) {
- EVT VT = PN->getValueType(i);
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- if (RegPressure[RCId] >= RegLimit[RCId])
- return true; // Reg pressure already high.
- unsigned Cost = TLI->getRepRegClassCostFor(VT);
- if (!PN->hasAnyUseOfValue(i))
- continue;
- // Check if this increases register pressure of the specific register
- // class to the point where it would cause spills.
- if ((RegPressure[RCId] + Cost) >= RegLimit[RCId])
- return true;
- }
- }
+ if (!N->isMachineOpcode() || !SU->NumSuccs)
+ return false;
- return false;
- }
+ unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs();
+ for (unsigned i = 0; i != NumDefs; ++i) {
+ EVT VT = N->getValueType(i);
+ if (!N->hasAnyUseOfValue(i))
+ continue;
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ if (RegPressure[RCId] >= RegLimit[RCId])
+ return true;
+ }
+ return false;
+}
- void ScheduledNode(SUnit *SU) {
- if (!TracksRegPressure)
- return;
-
- const SDNode *N = SU->getNode();
- if (!N->isMachineOpcode()) {
- if (N->getOpcode() != ISD::CopyToReg)
- return;
- } else {
- unsigned Opc = N->getMachineOpcode();
- if (Opc == TargetOpcode::EXTRACT_SUBREG ||
- Opc == TargetOpcode::INSERT_SUBREG ||
- Opc == TargetOpcode::SUBREG_TO_REG ||
- Opc == TargetOpcode::REG_SEQUENCE ||
- Opc == TargetOpcode::IMPLICIT_DEF)
- return;
- }
+void RegReductionPQBase::ScheduledNode(SUnit *SU) {
+ if (!TracksRegPressure)
+ return;
- for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
- I != E; ++I) {
- if (I->isCtrl())
- continue;
- SUnit *PredSU = I->getSUnit();
- if (PredSU->NumSuccsLeft != PredSU->NumSuccs)
- continue;
- const SDNode *PN = PredSU->getNode();
- if (!PN->isMachineOpcode()) {
- if (PN->getOpcode() == ISD::CopyFromReg) {
- EVT VT = PN->getValueType(0);
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
- }
- continue;
- }
- unsigned POpc = PN->getMachineOpcode();
- if (POpc == TargetOpcode::IMPLICIT_DEF)
- continue;
- if (POpc == TargetOpcode::EXTRACT_SUBREG) {
- EVT VT = PN->getOperand(0).getValueType();
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
- continue;
- } else if (POpc == TargetOpcode::INSERT_SUBREG ||
- POpc == TargetOpcode::SUBREG_TO_REG) {
- EVT VT = PN->getValueType(0);
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
- continue;
- }
- unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs();
- for (unsigned i = 0; i != NumDefs; ++i) {
- EVT VT = PN->getValueType(i);
- if (!PN->hasAnyUseOfValue(i))
- continue;
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
- }
- }
+ const SDNode *N = SU->getNode();
+ if (!N->isMachineOpcode()) {
+ if (N->getOpcode() != ISD::CopyToReg)
+ return;
+ } else {
+ unsigned Opc = N->getMachineOpcode();
+ if (Opc == TargetOpcode::EXTRACT_SUBREG ||
+ Opc == TargetOpcode::INSERT_SUBREG ||
+ Opc == TargetOpcode::SUBREG_TO_REG ||
+ Opc == TargetOpcode::REG_SEQUENCE ||
+ Opc == TargetOpcode::IMPLICIT_DEF)
+ return;
+ }
- // Check for isMachineOpcode() as PrescheduleNodesWithMultipleUses()
- // may transfer data dependencies to CopyToReg.
- if (SU->NumSuccs && N->isMachineOpcode()) {
- unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs();
- for (unsigned i = 0; i != NumDefs; ++i) {
- EVT VT = N->getValueType(i);
- if (!N->hasAnyUseOfValue(i))
- continue;
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- if (RegPressure[RCId] < TLI->getRepRegClassCostFor(VT))
- // Register pressure tracking is imprecise. This can happen.
- RegPressure[RCId] = 0;
- else
- RegPressure[RCId] -= TLI->getRepRegClassCostFor(VT);
- }
+ for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
+ I != E; ++I) {
+ if (I->isCtrl())
+ continue;
+ SUnit *PredSU = I->getSUnit();
+ if (PredSU->NumSuccsLeft != PredSU->NumSuccs)
+ continue;
+ const SDNode *PN = PredSU->getNode();
+ if (!PN->isMachineOpcode()) {
+ if (PN->getOpcode() == ISD::CopyFromReg) {
+ EVT VT = PN->getValueType(0);
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
}
+ continue;
+ }
+ unsigned POpc = PN->getMachineOpcode();
+ if (POpc == TargetOpcode::IMPLICIT_DEF)
+ continue;
+ if (POpc == TargetOpcode::EXTRACT_SUBREG) {
+ EVT VT = PN->getOperand(0).getValueType();
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
+ continue;
+ } else if (POpc == TargetOpcode::INSERT_SUBREG ||
+ POpc == TargetOpcode::SUBREG_TO_REG) {
+ EVT VT = PN->getValueType(0);
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
+ continue;
+ }
+ unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs();
+ for (unsigned i = 0; i != NumDefs; ++i) {
+ EVT VT = PN->getValueType(i);
+ if (!PN->hasAnyUseOfValue(i))
+ continue;
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
+ }
+ }
- dumpRegPressure();
+ // Check for isMachineOpcode() as PrescheduleNodesWithMultipleUses()
+ // may transfer data dependencies to CopyToReg.
+ if (SU->NumSuccs && N->isMachineOpcode()) {
+ unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs();
+ for (unsigned i = 0; i != NumDefs; ++i) {
+ EVT VT = N->getValueType(i);
+ if (!N->hasAnyUseOfValue(i))
+ continue;
+ unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
+ if (RegPressure[RCId] < TLI->getRepRegClassCostFor(VT))
+ // Register pressure tracking is imprecise. This can happen.
+ RegPressure[RCId] = 0;
+ else
+ RegPressure[RCId] -= TLI->getRepRegClassCostFor(VT);
}
+ }
- void UnscheduledNode(SUnit *SU) {
- if (!TracksRegPressure)
- return;
-
- const SDNode *N = SU->getNode();
- if (!N->isMachineOpcode()) {
- if (N->getOpcode() != ISD::CopyToReg)
- return;
- } else {
- unsigned Opc = N->getMachineOpcode();
- if (Opc == TargetOpcode::EXTRACT_SUBREG ||
- Opc == TargetOpcode::INSERT_SUBREG ||
- Opc == TargetOpcode::SUBREG_TO_REG ||
- Opc == TargetOpcode::REG_SEQUENCE ||
- Opc == TargetOpcode::IMPLICIT_DEF)
- return;
- }
+ dumpRegPressure();
+}
- for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end();
- I != E; ++I) {
- if (I->isCtrl())
- continue;
- SUnit *PredSU = I->getSUnit();
- if (PredSU->NumSuccsLeft != PredSU->NumSuccs)
- continue;
- const SDNode *PN = PredSU->getNode();
- if (!PN->isMachineOpcode()) {
- if (PN->getOpcode() == ISD::CopyFromReg) {
- EVT VT = PN->getValueType(0);
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
- }
- continue;
- }
- unsigned POpc = PN->getMachineOpcode();
- if (POpc == TargetOpcode::IMPLICIT_DEF)
- continue;
- if (POpc == TargetOpcode::EXTRACT_SUBREG) {
- EVT VT = PN->getOperand(0).getValueType();
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
- continue;
- } else if (POpc == TargetOpcode::INSERT_SUBREG ||
- POpc == TargetOpcode::SUBREG_TO_REG) {
- EVT VT = PN->getValueType(0);
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- RegPressure[RCId] += TLI->getRepRegClassCostFor(VT);
- continue;
- }
- unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs();
- for (unsigned i = 0; i != NumDefs; ++i) {
- EVT VT = PN->getValueType(i);
- if (!PN->hasAnyUseOfValue(i))
- continue;
- unsigned RCId = TLI->getRepRegClassFor(VT)->getID();
- if (RegPressure[RCId] < TLI->getRepRegClassCostFor(VT))
- // Register pressure tracking is imprecise. This can happen.
- RegPressure[RCId] = 0;
- else
- RegPressure[RCId] -= TLI->getRepRegClassCostFor(VT);
- }
- }
+void RegReductionPQBase::UnscheduledNode(SUnit *SU) {
+ if (!TracksRegPressure)
+ return;
- // Check for isMachineOpcode() as PrescheduleNodesWithMultipleUses()
- // may transfer data dependencies to CopyToReg.
- if (SU->NumSuccs && N->isMachineOpcode()) {
- unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs();
- for (unsigned i = NumDefs, e = N->getNumValues(); i != e; ++i) {
- EVT VT = N->getValueType(i);
- if (VT == MVT::Glue || VT == MVT::Other)
- continue;
- if (!N->hasAnyUseOfValue(i))