/// Apply cost model and heuristics to the if-conversion in IfConv.
/// Return true if the conversion is a good idea.
///
bool AArch64ConditionalCompares::shouldConvert() {
// Stress testing mode disables all cost considerations.
if (Stress)
return true;
if (!MinInstr)
MinInstr = Traces->getEnsemble(MachineTraceMetrics::TS_MinInstrCount);
// Head dominates CmpBB, so it is always included in its trace.
MachineTraceMetrics::Trace Trace = MinInstr->getTrace(CmpConv.CmpBB);
// If code size is the main concern
if (MinSize) {
int CodeSizeDelta = CmpConv.expectedCodeSizeDelta();
DEBUG(dbgs() << "Code size delta: " << CodeSizeDelta << '\n');
// If we are minimizing the code size, do the conversion whatever
// the cost is.
if (CodeSizeDelta < 0)
return true;
if (CodeSizeDelta > 0) {
DEBUG(dbgs() << "Code size is increasing, give up on this one.\n");
return false;
}
// CodeSizeDelta == 0, continue with the regular heuristics
}
// Heuristic: The compare conversion delays the execution of the branch
// instruction because we must wait for the inputs to the second compare as
// well. The branch has no dependent instructions, but delaying it increases
// the cost of a misprediction.
//
// Set a limit on the delay we will accept.
unsigned DelayLimit = SchedModel.MispredictPenalty * 3 / 4;
// Instruction depths can be computed for all trace instructions above CmpBB.
unsigned HeadDepth =
Trace.getInstrCycles(CmpConv.Head->getFirstTerminator()).Depth;
unsigned CmpBBDepth =
Trace.getInstrCycles(CmpConv.CmpBB->getFirstTerminator()).Depth;
DEBUG(dbgs() << "Head depth: " << HeadDepth
<< "\nCmpBB depth: " << CmpBBDepth << '\n');
if (CmpBBDepth > HeadDepth + DelayLimit) {
DEBUG(dbgs() << "Branch delay would be larger than " << DelayLimit
<< " cycles.\n");
return false;
}
// Check the resource depth at the bottom of CmpBB - these instructions will
// be speculated.
unsigned ResDepth = Trace.getResourceDepth(true);
DEBUG(dbgs() << "Resources: " << ResDepth << '\n');
// Heuristic: The speculatively executed instructions must all be able to
// merge into the Head block. The Head critical path should dominate the
// resource cost of the speculated instructions.
if (ResDepth > HeadDepth) {
DEBUG(dbgs() << "Too many instructions to speculate.\n");
return false;
}
return true;
}
/// Return true if an STP can be added to this block without increasing the
/// critical resource height. STP is good to form in Ld/St limited blocks and
/// bad to form in float-point limited blocks. This is true independent of the
/// critical path. If the critical path is longer than the resource height, the
/// extra vector ops can limit physreg renaming. Otherwise, it could simply
/// oversaturate the vector units.
bool ARM64StorePairSuppress::shouldAddSTPToBlock(const MachineBasicBlock *BB) {
if (!MinInstr)
MinInstr = Traces->getEnsemble(MachineTraceMetrics::TS_MinInstrCount);
MachineTraceMetrics::Trace BBTrace = MinInstr->getTrace(BB);
unsigned ResLength = BBTrace.getResourceLength();
// Get the machine model's scheduling class for STPQi.
// Bypass TargetSchedule's SchedClass resolution since we only have an opcode.
unsigned SCIdx = TII->get(ARM64::STPDi).getSchedClass();
const MCSchedClassDesc *SCDesc =
SchedModel.getMCSchedModel()->getSchedClassDesc(SCIdx);
// If a subtarget does not define resources for STPQi, bail here.
if (SCDesc->isValid() && !SCDesc->isVariant()) {
unsigned ResLenWithSTP = BBTrace.getResourceLength(
ArrayRef<const MachineBasicBlock *>(), SCDesc);
if (ResLenWithSTP > ResLength) {
DEBUG(dbgs() << " Suppress STP in BB: " << BB->getNumber()
<< " resources " << ResLength << " -> " << ResLenWithSTP
<< "\n");
return false;
}
}
return true;
}
/// getDepth - Computes depth of instructions in vector \InsInstr.
///
/// \param InsInstrs is a vector of machine instructions
/// \param InstrIdxForVirtReg is a dense map of virtual register to index
/// of defining machine instruction in \p InsInstrs
/// \param BlockTrace is a trace of machine instructions
///
/// \returns Depth of last instruction in \InsInstrs ("NewRoot")
unsigned
MachineCombiner::getDepth(SmallVectorImpl<MachineInstr *> &InsInstrs,
DenseMap<unsigned, unsigned> &InstrIdxForVirtReg,
MachineTraceMetrics::Trace BlockTrace) {
SmallVector<unsigned, 16> InstrDepth;
assert(TSchedModel.hasInstrSchedModel() && "Missing machine model\n");
// Foreach instruction in in the new sequence compute the depth based on the
// operands. Use the trace information when possible. For new operands which
// are tracked in the InstrIdxForVirtReg map depth is looked up in InstrDepth
for (auto *InstrPtr : InsInstrs) { // for each Use
unsigned IDepth = 0;
DEBUG(dbgs() << "NEW INSTR "; InstrPtr->dump(); dbgs() << "\n";);
for (unsigned i = 0, e = InstrPtr->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = InstrPtr->getOperand(i);
// Check for virtual register operand.
if (!(MO.isReg() && TargetRegisterInfo::isVirtualRegister(MO.getReg())))
continue;
if (!MO.isUse())
continue;
unsigned DepthOp = 0;
unsigned LatencyOp = 0;
DenseMap<unsigned, unsigned>::iterator II =
InstrIdxForVirtReg.find(MO.getReg());
if (II != InstrIdxForVirtReg.end()) {
// Operand is new virtual register not in trace
assert(II->second < InstrDepth.size() && "Bad Index");
MachineInstr *DefInstr = InsInstrs[II->second];
assert(DefInstr &&
"There must be a definition for a new virtual register");
DepthOp = InstrDepth[II->second];
LatencyOp = TSchedModel.computeOperandLatency(
DefInstr, DefInstr->findRegisterDefOperandIdx(MO.getReg()),
InstrPtr, InstrPtr->findRegisterUseOperandIdx(MO.getReg()));
} else {
MachineInstr *DefInstr = getOperandDef(MO);
if (DefInstr) {
DepthOp = BlockTrace.getInstrCycles(DefInstr).Depth;
LatencyOp = TSchedModel.computeOperandLatency(
DefInstr, DefInstr->findRegisterDefOperandIdx(MO.getReg()),
InstrPtr, InstrPtr->findRegisterUseOperandIdx(MO.getReg()));
}
}
IDepth = std::max(IDepth, DepthOp + LatencyOp);
}
InstrDepth.push_back(IDepth);
}
