//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "machine-trace-metrics"
#include "llvm/CodeGen/MachineTraceMetrics.h"
#include "llvm/ADT/PostOrderIterator.h"
#include "llvm/ADT/SparseSet.h"
using namespace llvm;
+#define DEBUG_TYPE "machine-trace-metrics"
+
char MachineTraceMetrics::ID = 0;
char &llvm::MachineTraceMetricsID = MachineTraceMetrics::ID;
"machine-trace-metrics", "Machine Trace Metrics", false, true)
MachineTraceMetrics::MachineTraceMetrics()
- : MachineFunctionPass(ID), MF(0), TII(0), TRI(0), MRI(0), Loops(0) {
- std::fill(Ensembles, array_endof(Ensembles), (Ensemble*)0);
+ : MachineFunctionPass(ID), MF(nullptr), TII(nullptr), TRI(nullptr),
+ MRI(nullptr), Loops(nullptr) {
+ std::fill(std::begin(Ensembles), std::end(Ensembles), nullptr);
}
void MachineTraceMetrics::getAnalysisUsage(AnalysisUsage &AU) const {
bool MachineTraceMetrics::runOnMachineFunction(MachineFunction &Func) {
MF = &Func;
- TII = MF->getTarget().getInstrInfo();
- TRI = MF->getTarget().getRegisterInfo();
+ TII = MF->getSubtarget().getInstrInfo();
+ TRI = MF->getSubtarget().getRegisterInfo();
MRI = &MF->getRegInfo();
Loops = &getAnalysis<MachineLoopInfo>();
const TargetSubtargetInfo &ST =
MF->getTarget().getSubtarget<TargetSubtargetInfo>();
- SchedModel.init(*ST.getSchedModel(), &ST, TII);
+ SchedModel.init(ST.getSchedModel(), &ST, TII);
BlockInfo.resize(MF->getNumBlockIDs());
ProcResourceCycles.resize(MF->getNumBlockIDs() *
SchedModel.getNumProcResourceKinds());
}
void MachineTraceMetrics::releaseMemory() {
- MF = 0;
+ MF = nullptr;
BlockInfo.clear();
for (unsigned i = 0; i != TS_NumStrategies; ++i) {
delete Ensembles[i];
- Ensembles[i] = 0;
+ Ensembles[i] = nullptr;
}
}
unsigned PRKinds = SchedModel.getNumProcResourceKinds();
SmallVector<unsigned, 32> PRCycles(PRKinds);
- for (MachineBasicBlock::const_iterator I = MBB->begin(), E = MBB->end();
- I != E; ++I) {
- const MachineInstr *MI = I;
- if (MI->isTransient())
+ for (const auto &MI : *MBB) {
+ if (MI.isTransient())
continue;
++InstrCount;
- if (MI->isCall())
+ if (MI.isCall())
FBI->HasCalls = true;
// Count processor resources used.
if (!SchedModel.hasInstrSchedModel())
continue;
- const MCSchedClassDesc *SC = SchedModel.resolveSchedClass(MI);
+ const MCSchedClassDesc *SC = SchedModel.resolveSchedClass(&MI);
if (!SC->isValid())
continue;
"getResources() must be called before getProcResourceCycles()");
unsigned PRKinds = SchedModel.getNumProcResourceKinds();
assert((MBBNum+1) * PRKinds <= ProcResourceCycles.size());
- return ArrayRef<unsigned>(ProcResourceCycles.data() + MBBNum * PRKinds,
- PRKinds);
+ return makeArrayRef(ProcResourceCycles.data() + MBBNum * PRKinds, PRKinds);
}
MachineTraceMetrics::Ensemble::
getDepthResources(const MachineBasicBlock *MBB) const {
const TraceBlockInfo *TBI = &BlockInfo[MBB->getNumber()];
- return TBI->hasValidDepth() ? TBI : 0;
+ return TBI->hasValidDepth() ? TBI : nullptr;
}
// Check if height resources for MBB are valid and return the TBI.
MachineTraceMetrics::Ensemble::
getHeightResources(const MachineBasicBlock *MBB) const {
const TraceBlockInfo *TBI = &BlockInfo[MBB->getNumber()];
- return TBI->hasValidHeight() ? TBI : 0;
+ return TBI->hasValidHeight() ? TBI : nullptr;
}
/// Get an array of processor resource depths for MBB. Indexed by processor
getProcResourceDepths(unsigned MBBNum) const {
unsigned PRKinds = MTM.SchedModel.getNumProcResourceKinds();
assert((MBBNum+1) * PRKinds <= ProcResourceDepths.size());
- return ArrayRef<unsigned>(ProcResourceDepths.data() + MBBNum * PRKinds,
- PRKinds);
+ return makeArrayRef(ProcResourceDepths.data() + MBBNum * PRKinds, PRKinds);
}
/// Get an array of processor resource heights for MBB. Indexed by processor
getProcResourceHeights(unsigned MBBNum) const {
unsigned PRKinds = MTM.SchedModel.getNumProcResourceKinds();
assert((MBBNum+1) * PRKinds <= ProcResourceHeights.size());
- return ArrayRef<unsigned>(ProcResourceHeights.data() + MBBNum * PRKinds,
- PRKinds);
+ return makeArrayRef(ProcResourceHeights.data() + MBBNum * PRKinds, PRKinds);
}
//===----------------------------------------------------------------------===//
const MachineBasicBlock*
MinInstrCountEnsemble::pickTracePred(const MachineBasicBlock *MBB) {
if (MBB->pred_empty())
- return 0;
+ return nullptr;
const MachineLoop *CurLoop = getLoopFor(MBB);
// Don't leave loops, and never follow back-edges.
if (CurLoop && MBB == CurLoop->getHeader())
- return 0;
+ return nullptr;
unsigned CurCount = MTM.getResources(MBB)->InstrCount;
- const MachineBasicBlock *Best = 0;
+ const MachineBasicBlock *Best = nullptr;
unsigned BestDepth = 0;
for (MachineBasicBlock::const_pred_iterator
I = MBB->pred_begin(), E = MBB->pred_end(); I != E; ++I) {
const MachineBasicBlock*
MinInstrCountEnsemble::pickTraceSucc(const MachineBasicBlock *MBB) {
if (MBB->pred_empty())
- return 0;
+ return nullptr;
const MachineLoop *CurLoop = getLoopFor(MBB);
- const MachineBasicBlock *Best = 0;
+ const MachineBasicBlock *Best = nullptr;
unsigned BestHeight = 0;
for (MachineBasicBlock::const_succ_iterator
I = MBB->succ_begin(), E = MBB->succ_end(); I != E; ++I) {
}
// To is a new block. Mark the block as visited in case the CFG has cycles
// that MachineLoopInfo didn't recognize as a natural loop.
- return LB.Visited.insert(To);
+ return LB.Visited.insert(To).second;
}
};
}
// invalidated, but their instructions will stay the same, so there is no
// need to erase the Cycle entries. They will be overwritten when we
// recompute.
- for (MachineBasicBlock::const_iterator I = BadMBB->begin(), E = BadMBB->end();
- I != E; ++I)
- Cycles.erase(I);
+ for (const auto &I : *BadMBB)
+ Cycles.erase(&I);
}
void MachineTraceMetrics::Ensemble::verify() const {
assert(TargetRegisterInfo::isVirtualRegister(VirtReg));
MachineRegisterInfo::def_iterator DefI = MRI->def_begin(VirtReg);
assert(!DefI.atEnd() && "Register has no defs");
- DefMI = &*DefI;
+ DefMI = DefI->getParent();
DefOp = DefI.getOperandNo();
assert((++DefI).atEnd() && "Register has multiple defs");
}
unsigned getSparseSetIndex() const { return RegUnit; }
- LiveRegUnit(unsigned RU) : RegUnit(RU), Cycle(0), MI(0), Op(0) {}
+ LiveRegUnit(unsigned RU) : RegUnit(RU), Cycle(0), MI(nullptr), Op(0) {}
};
}
if (TBI.HasValidInstrHeights)
TBI.CriticalPath = computeCrossBlockCriticalPath(TBI);
- for (MachineBasicBlock::const_iterator I = MBB->begin(), E = MBB->end();
- I != E; ++I) {
- const MachineInstr *UseMI = I;
-
+ for (const auto &UseMI : *MBB) {
// Collect all data dependencies.
Deps.clear();
- if (UseMI->isPHI())
- getPHIDeps(UseMI, Deps, TBI.Pred, MTM.MRI);
- else if (getDataDeps(UseMI, Deps, MTM.MRI))
- updatePhysDepsDownwards(UseMI, Deps, RegUnits, MTM.TRI);
+ if (UseMI.isPHI())
+ getPHIDeps(&UseMI, Deps, TBI.Pred, MTM.MRI);
+ else if (getDataDeps(&UseMI, Deps, MTM.MRI))
+ updatePhysDepsDownwards(&UseMI, Deps, RegUnits, MTM.TRI);
// Filter and process dependencies, computing the earliest issue cycle.
unsigned Cycle = 0;
// Add latency if DefMI is a real instruction. Transients get latency 0.
if (!Dep.DefMI->isTransient())
DepCycle += MTM.SchedModel
- .computeOperandLatency(Dep.DefMI, Dep.DefOp, UseMI, Dep.UseOp);
+ .computeOperandLatency(Dep.DefMI, Dep.DefOp, &UseMI, Dep.UseOp);
Cycle = std::max(Cycle, DepCycle);
}
// Remember the instruction depth.
- InstrCycles &MICycles = Cycles[UseMI];
+ InstrCycles &MICycles = Cycles[&UseMI];
MICycles.Depth = Cycle;
if (!TBI.HasValidInstrHeights) {
- DEBUG(dbgs() << Cycle << '\t' << *UseMI);
+ DEBUG(dbgs() << Cycle << '\t' << UseMI);
continue;
}
// Update critical path length.
TBI.CriticalPath = std::max(TBI.CriticalPath, Cycle + MICycles.Height);
- DEBUG(dbgs() << TBI.CriticalPath << '\t' << Cycle << '\t' << *UseMI);
+ DEBUG(dbgs() << TBI.CriticalPath << '\t' << Cycle << '\t' << UseMI);
}
}
}
Succ = Loop->getHeader();
if (Succ) {
- for (MachineBasicBlock::const_iterator I = Succ->begin(), E = Succ->end();
- I != E && I->isPHI(); ++I) {
- const MachineInstr *PHI = I;
+ for (const auto &PHI : *Succ) {
+ if (!PHI.isPHI())
+ break;
Deps.clear();
- getPHIDeps(PHI, Deps, MBB, MTM.MRI);
+ getPHIDeps(&PHI, Deps, MBB, MTM.MRI);
if (!Deps.empty()) {
// Loop header PHI heights are all 0.
- unsigned Height = TBI.Succ ? Cycles.lookup(PHI).Height : 0;
- DEBUG(dbgs() << "pred\t" << Height << '\t' << *PHI);
- if (pushDepHeight(Deps.front(), PHI, Height,
+ unsigned Height = TBI.Succ ? Cycles.lookup(&PHI).Height : 0;
+ DEBUG(dbgs() << "pred\t" << Height << '\t' << PHI);
+ if (pushDepHeight(Deps.front(), &PHI, Height,
Heights, MTM.SchedModel, MTM.TII))
addLiveIns(Deps.front().DefMI, Deps.front().DefOp, Stack);
}
return DepCycle;
}
+/// When bottom is set include instructions in current block in estimate.
unsigned MachineTraceMetrics::Trace::getResourceDepth(bool Bottom) const {
// Find the limiting processor resource.
// Numbers have been pre-scaled to be comparable.
// Convert to cycle count.
PRMax = TE.MTM.getCycles(PRMax);
+ /// All instructions before current block
unsigned Instrs = TBI.InstrDepth;
+ // plus instructions in current block
if (Bottom)
Instrs += TE.MTM.BlockInfo[getBlockNum()].InstrCount;
if (unsigned IW = TE.MTM.SchedModel.getIssueWidth())
return std::max(Instrs, PRMax);
}
-
-unsigned MachineTraceMetrics::Trace::
-getResourceLength(ArrayRef<const MachineBasicBlock*> Extrablocks,
- ArrayRef<const MCSchedClassDesc*> ExtraInstrs) const {
+unsigned MachineTraceMetrics::Trace::getResourceLength(
+ ArrayRef<const MachineBasicBlock *> Extrablocks,
+ ArrayRef<const MCSchedClassDesc *> ExtraInstrs,
+ ArrayRef<const MCSchedClassDesc *> RemoveInstrs) const {
// Add up resources above and below the center block.
ArrayRef<unsigned> PRDepths = TE.getProcResourceDepths(getBlockNum());
ArrayRef<unsigned> PRHeights = TE.getProcResourceHeights(getBlockNum());
unsigned PRMax = 0;
- for (unsigned K = 0; K != PRDepths.size(); ++K) {
- unsigned PRCycles = PRDepths[K] + PRHeights[K];
- for (unsigned I = 0; I != Extrablocks.size(); ++I)
- PRCycles += TE.MTM.getProcResourceCycles(Extrablocks[I]->getNumber())[K];
- for (unsigned I = 0; I != ExtraInstrs.size(); ++I) {
- const MCSchedClassDesc* SC = ExtraInstrs[I];
+
+ // Capture computing cycles from extra instructions
+ auto extraCycles = [this](ArrayRef<const MCSchedClassDesc *> Instrs,
+ unsigned ResourceIdx)
+ ->unsigned {
+ unsigned Cycles = 0;
+ for (unsigned I = 0; I != Instrs.size(); ++I) {
+ const MCSchedClassDesc *SC = Instrs[I];
if (!SC->isValid())
continue;
for (TargetSchedModel::ProcResIter
- PI = TE.MTM.SchedModel.getWriteProcResBegin(SC),
- PE = TE.MTM.SchedModel.getWriteProcResEnd(SC); PI != PE; ++PI) {
- if (PI->ProcResourceIdx != K)
+ PI = TE.MTM.SchedModel.getWriteProcResBegin(SC),
+ PE = TE.MTM.SchedModel.getWriteProcResEnd(SC);
+ PI != PE; ++PI) {
+ if (PI->ProcResourceIdx != ResourceIdx)
continue;
- PRCycles += (PI->Cycles * TE.MTM.SchedModel.getResourceFactor(K));
+ Cycles +=
+ (PI->Cycles * TE.MTM.SchedModel.getResourceFactor(ResourceIdx));
}
}
+ return Cycles;
+ };
+
+ for (unsigned K = 0; K != PRDepths.size(); ++K) {
+ unsigned PRCycles = PRDepths[K] + PRHeights[K];
+ for (unsigned I = 0; I != Extrablocks.size(); ++I)
+ PRCycles += TE.MTM.getProcResourceCycles(Extrablocks[I]->getNumber())[K];
+ PRCycles += extraCycles(ExtraInstrs, K);
+ PRCycles -= extraCycles(RemoveInstrs, K);
PRMax = std::max(PRMax, PRCycles);
}
// Convert to cycle count.
PRMax = TE.MTM.getCycles(PRMax);
+ // Instrs: #instructions in current trace outside current block.
unsigned Instrs = TBI.InstrDepth + TBI.InstrHeight;
+ // Add instruction count from the extra blocks.
for (unsigned i = 0, e = Extrablocks.size(); i != e; ++i)
Instrs += TE.MTM.getResources(Extrablocks[i])->InstrCount;
+ Instrs += ExtraInstrs.size();
+ Instrs -= RemoveInstrs.size();
if (unsigned IW = TE.MTM.SchedModel.getIssueWidth())
Instrs /= IW;
// Assume issue width 1 without a schedule model.
return std::max(Instrs, PRMax);
}
+bool MachineTraceMetrics::Trace::isDepInTrace(const MachineInstr *DefMI,
+ const MachineInstr *UseMI) const {
+ if (DefMI->getParent() == UseMI->getParent())
+ return true;
+
+ const TraceBlockInfo &DepTBI = TE.BlockInfo[DefMI->getParent()->getNumber()];
+ const TraceBlockInfo &TBI = TE.BlockInfo[UseMI->getParent()->getNumber()];
+
+ return DepTBI.isUsefulDominator(TBI);
+}
+
void MachineTraceMetrics::Ensemble::print(raw_ostream &OS) const {
OS << getName() << " ensemble:\n";
for (unsigned i = 0, e = BlockInfo.size(); i != e; ++i) {
projects / oota-llvm.git / blobdiff