LLVM: llvm::TargetSchedModel Class Reference (original) (raw)

Provide an instruction scheduling machine model to CodeGen passes. More...

#include "[llvm/CodeGen/TargetSchedule.h](TargetSchedule%5F8h%5Fsource.html)"

Public Member Functions
	TargetSchedModel ()
LLVM_ABI void	init (const TargetSubtargetInfo *TSInfo, bool EnableSModel=true, bool EnableSItins=true)
	Initialize the machine model for instruction scheduling.
LLVM_ABI const MCSchedClassDesc *	resolveSchedClass (const MachineInstr *MI) const
	Return the MCSchedClassDesc for this instruction.
const TargetSubtargetInfo *	getSubtargetInfo () const
	TargetSubtargetInfo getter.
const TargetInstrInfo *	getInstrInfo () const
	TargetInstrInfo getter.
LLVM_ABI bool	hasInstrSchedModel () const
	Return true if this machine model includes an instruction-level scheduling model.
const MCSchedModel *	getMCSchedModel () const
LLVM_ABI bool	hasInstrItineraries () const
	Return true if this machine model includes cycle-to-cycle itinerary data.
const InstrItineraryData *	getInstrItineraries () const
bool	hasInstrSchedModelOrItineraries () const
	Return true if this machine model includes an instruction-level scheduling model or cycle-to-cycle itinerary data.
LLVM_ABI bool	enableIntervals () const
unsigned	getProcessorID () const
	Identify the processor corresponding to the current subtarget.
unsigned	getIssueWidth () const
	Maximum number of micro-ops that may be scheduled per cycle.
LLVM_ABI bool	mustBeginGroup (const MachineInstr MI, const MCSchedClassDesc SC=nullptr) const
	Return true if new group must begin.
LLVM_ABI bool	mustEndGroup (const MachineInstr MI, const MCSchedClassDesc SC=nullptr) const
	Return true if current group must end.
LLVM_ABI unsigned	getNumMicroOps (const MachineInstr MI, const MCSchedClassDesc SC=nullptr) const
	Return the number of issue slots required for this MI.
unsigned	getNumProcResourceKinds () const
	Get the number of kinds of resources for this target.
const MCProcResourceDesc *	getProcResource (unsigned PIdx) const
	Get a processor resource by ID for convenience.
const char *	getResourceName (unsigned PIdx) const
ProcResIter	getWriteProcResBegin (const MCSchedClassDesc *SC) const
ProcResIter	getWriteProcResEnd (const MCSchedClassDesc *SC) const
unsigned	getResourceFactor (unsigned ResIdx) const
	Multiply the number of units consumed for a resource by this factor to normalize it relative to other resources.
unsigned	getMicroOpFactor () const
	Multiply number of micro-ops by this factor to normalize it relative to other resources.
unsigned	getLatencyFactor () const
	Multiply cycle count by this factor to normalize it relative to other resources.
unsigned	getMicroOpBufferSize () const
	Number of micro-ops that may be buffered for OOO execution.
int	getResourceBufferSize (unsigned PIdx) const
	Number of resource units that may be buffered for OOO execution.
LLVM_ABI unsigned	computeOperandLatency (const MachineInstr DefMI, unsigned DefOperIdx, const MachineInstr UseMI, unsigned UseOperIdx) const
	Compute operand latency based on the available machine model.
LLVM_ABI unsigned	computeInstrLatency (const MachineInstr *MI, bool UseDefaultDefLatency=true) const
	Compute the instruction latency based on the available machine model.
LLVM_ABI unsigned	computeInstrLatency (const MCInst &Inst) const
LLVM_ABI unsigned	computeInstrLatency (unsigned Opcode) const
LLVM_ABI unsigned	computeOutputLatency (const MachineInstr DefMI, unsigned DefOperIdx, const MachineInstr DepMI) const
	Output dependency latency of a pair of defs of the same register.
LLVM_ABI double	computeReciprocalThroughput (const MachineInstr *MI) const
	Compute the reciprocal throughput of the given instruction.
LLVM_ABI double	computeReciprocalThroughput (const MCInst &MI) const
LLVM_ABI double	computeReciprocalThroughput (unsigned Opcode) const

Provide an instruction scheduling machine model to CodeGen passes.

Definition at line 31 of file TargetSchedule.h.

◆ ProcResIter

llvm::TargetSchedModel::TargetSchedModel ( )	inline

◆ computeInstrLatency() [1/3]

Compute the instruction latency based on the available machine model.

Compute and return the expected latency of this instruction independent of a particular use. computeOperandLatency is the preferred API, but this is occasionally useful to help estimate instruction cost.

If UseDefaultDefLatency is false and no new machine sched model is present this method falls back to TII->getInstrLatency with an empty instruction itinerary (this is so we preserve the previous behavior of the if converter after moving it to TargetSchedModel).

Definition at line 253 of file TargetSchedule.cpp.

References hasInstrItineraries(), hasInstrSchedModel(), llvm::MCSchedClassDesc::isValid(), MI, and resolveSchedClass().

◆ computeInstrLatency() [2/3]

◆ computeInstrLatency() [3/3]

◆ computeOperandLatency()

Compute operand latency based on the available machine model.

Compute and return the latency of the given data dependent def and use when the operand indices are already known. UseMI may be NULL for an unknown user.

Definition at line 169 of file TargetSchedule.cpp.

References capLatency(), llvm::MCWriteLatencyEntry::Cycles, DefMI, llvm::errs(), findDefIdx(), findUseIdx(), hasInstrItineraries(), hasInstrSchedModel(), llvm::MCSchedClassDesc::isValid(), llvm::Latency, llvm_unreachable, llvm::MCSchedClassDesc::NumReadAdvanceEntries, resolveSchedClass(), UseMI, and llvm::MCWriteLatencyEntry::WriteResourceID.

Referenced by llvm::AArch64Subtarget::adjustSchedDependency(), pushDepHeight(), and updatePhysDepsUpwards().

◆ computeOutputLatency()

◆ computeReciprocalThroughput() [1/3]

double TargetSchedModel::computeReciprocalThroughput	(	const MachineInstr *	MI	)	const

◆ computeReciprocalThroughput() [2/3]

double TargetSchedModel::computeReciprocalThroughput	(	const MCInst &	MI	)	const

◆ computeReciprocalThroughput() [3/3]

double TargetSchedModel::computeReciprocalThroughput	(	unsigned	Opcode	)	const

◆ enableIntervals()

bool TargetSchedModel::enableIntervals	(	)	const

◆ getInstrInfo()

◆ getInstrItineraries()

◆ getIssueWidth()

unsigned llvm::TargetSchedModel::getIssueWidth ( ) const	inline

◆ getLatencyFactor()

unsigned llvm::TargetSchedModel::getLatencyFactor ( ) const	inline

Multiply cycle count by this factor to normalize it relative to other resources.

This is the number of resource units per cycle.

Definition at line 168 of file TargetSchedule.h.

◆ getMCSchedModel()

◆ getMicroOpBufferSize()

unsigned llvm::TargetSchedModel::getMicroOpBufferSize ( ) const	inline

◆ getMicroOpFactor()

unsigned llvm::TargetSchedModel::getMicroOpFactor ( ) const	inline

◆ getNumMicroOps()

◆ getNumProcResourceKinds()

unsigned llvm::TargetSchedModel::getNumProcResourceKinds ( ) const	inline

◆ getProcessorID()

unsigned llvm::TargetSchedModel::getProcessorID ( ) const	inline

Identify the processor corresponding to the current subtarget.

Definition at line 108 of file TargetSchedule.h.

◆ getProcResource()

◆ getResourceBufferSize()

int llvm::TargetSchedModel::getResourceBufferSize ( unsigned PIdx) const	inline

Number of resource units that may be buffered for OOO execution.

Returns

The buffer size in resource units or -1 for unlimited.

Definition at line 177 of file TargetSchedule.h.

◆ getResourceFactor()

unsigned llvm::TargetSchedModel::getResourceFactor ( unsigned ResIdx) const	inline

◆ getResourceName()

◆ getSubtargetInfo()

◆ getWriteProcResBegin()

◆ getWriteProcResEnd()

◆ hasInstrItineraries()

bool TargetSchedModel::hasInstrItineraries	(	)	const

◆ hasInstrSchedModel()

bool TargetSchedModel::hasInstrSchedModel	(	)	const

Return true if this machine model includes an instruction-level scheduling model.

This is more detailed than the course grain IssueWidth and default latency properties, but separate from the per-cycle itinerary data.

Definition at line 35 of file TargetSchedule.cpp.

Referenced by computeInstrLatency(), computeInstrLatency(), computeInstrLatency(), computeOperandLatency(), computeOutputLatency(), computeReciprocalThroughput(), computeReciprocalThroughput(), computeReciprocalThroughput(), getNumMicroOps(), hasInstrSchedModelOrItineraries(), llvm::SchedRemainder::init(), mustBeginGroup(), mustEndGroup(), and shouldCheckPending().

◆ hasInstrSchedModelOrItineraries()

bool llvm::TargetSchedModel::hasInstrSchedModelOrItineraries ( ) const	inline