LLVM 22.0.0git
llvm::SchedBoundary Class Reference

Each Scheduling boundary is associated with ready queues. More...

#include "llvm/CodeGen/MachineScheduler.h"

Public Types

enum  { TopQID = 1 , BotQID = 2 , LogMaxQID = 2 }
 SUnit::NodeQueueId: 0 (none), 1 (top), 2 (bot), 3 (both) More...

Public Member Functions

 SchedBoundary (unsigned ID, const Twine &Name)
 Pending queues extend the ready queues with the same ID and the PendingFlag set.
SchedBoundaryoperator= (const SchedBoundary &other)=delete
 SchedBoundary (const SchedBoundary &other)=delete
LLVM_ABI ~SchedBoundary ()
LLVM_ABI void reset ()
LLVM_ABI void init (ScheduleDAGMI *dag, const TargetSchedModel *smodel, SchedRemainder *rem)
bool isTop () const
unsigned getCurrCycle () const
 Number of cycles to issue the instructions scheduled in this zone.
unsigned getCurrMOps () const
 Micro-ops issued in the current cycle.
unsigned getDependentLatency () const
unsigned getScheduledLatency () const
 Get the number of latency cycles "covered" by the scheduled instructions.
unsigned getUnscheduledLatency (SUnit *SU) const
unsigned getResourceCount (unsigned ResIdx) const
unsigned getCriticalCount () const
 Get the scaled count of scheduled micro-ops and resources, including executed resources.
unsigned getExecutedCount () const
 Get a scaled count for the minimum execution time of the scheduled micro-ops that are ready to execute by getExecutedCount.
unsigned getZoneCritResIdx () const
bool isResourceLimited () const
LLVM_ABI unsigned getLatencyStallCycles (SUnit *SU)
 Get the difference between the given SUnit's ready time and the current cycle.
LLVM_ABI unsigned getNextResourceCycleByInstance (unsigned InstanceIndex, unsigned ReleaseAtCycle, unsigned AcquireAtCycle)
 Compute the next cycle at which the given processor resource unit can be scheduled.
LLVM_ABI std::pair< unsigned, unsignedgetNextResourceCycle (const MCSchedClassDesc *SC, unsigned PIdx, unsigned ReleaseAtCycle, unsigned AcquireAtCycle)
 Compute the next cycle at which the given processor resource can be scheduled.
bool isUnbufferedGroup (unsigned PIdx) const
LLVM_ABI bool checkHazard (SUnit *SU)
 Does this SU have a hazard within the current instruction group.
LLVM_ABI unsigned findMaxLatency (ArrayRef< SUnit * > ReadySUs)
LLVM_ABI unsigned getOtherResourceCount (unsigned &OtherCritIdx)
LLVM_ABI void releaseNode (SUnit *SU, unsigned ReadyCycle, bool InPQueue, unsigned Idx=0)
 Release SU to make it ready.
LLVM_ABI void bumpCycle (unsigned NextCycle)
 Move the boundary of scheduled code by one cycle.
LLVM_ABI void incExecutedResources (unsigned PIdx, unsigned Count)
LLVM_ABI unsigned countResource (const MCSchedClassDesc *SC, unsigned PIdx, unsigned Cycles, unsigned ReadyCycle, unsigned StartAtCycle)
 Add the given processor resource to this scheduled zone.
LLVM_ABI void bumpNode (SUnit *SU)
 Move the boundary of scheduled code by one SUnit.
LLVM_ABI void releasePending ()
 Release pending ready nodes in to the available queue.
LLVM_ABI void removeReady (SUnit *SU)
 Remove SU from the ready set for this boundary.
LLVM_ABI SUnitpickOnlyChoice ()
 Call this before applying any other heuristics to the Available queue.
LLVM_ABI void dumpReservedCycles () const
 Dump the state of the information that tracks resource usage.
LLVM_ABI void dumpScheduledState () const

Public Attributes

ScheduleDAGMIDAG = nullptr
const TargetSchedModelSchedModel = nullptr
SchedRemainderRem = nullptr
ReadyQueue Available
ReadyQueue Pending
ScheduleHazardRecognizerHazardRec = nullptr

Detailed Description

Each Scheduling boundary is associated with ready queues.

It tracks the current cycle in the direction of movement, and maintains the state of "hazards" and other interlocks at the current cycle.

Definition at line 856 of file MachineScheduler.h.

Member Enumeration Documentation

◆ anonymous enum

anonymous enum

SUnit::NodeQueueId: 0 (none), 1 (top), 2 (bot), 3 (both)

Enumerator
TopQID 
BotQID 
LogMaxQID 

Definition at line 859 of file MachineScheduler.h.

Constructor & Destructor Documentation

◆ SchedBoundary() [1/2]

llvm::SchedBoundary::SchedBoundary ( unsigned ID,
const Twine & Name )
inline

Pending queues extend the ready queues with the same ID and the PendingFlag set.

Definition at line 967 of file MachineScheduler.h.

References Available, LogMaxQID, Pending, and reset().

Referenced by operator=(), and SchedBoundary().

◆ SchedBoundary() [2/2]

llvm::SchedBoundary::SchedBoundary ( const SchedBoundary & other)
delete

References LLVM_ABI, rem, and SchedBoundary().

◆ ~SchedBoundary()

SchedBoundary::~SchedBoundary ( )

Definition at line 2486 of file MachineScheduler.cpp.

References HazardRec.

Member Function Documentation

◆ bumpCycle()

void SchedBoundary::bumpCycle ( unsigned NextCycle)

Move the boundary of scheduled code by one cycle.

Definition at line 2862 of file MachineScheduler.cpp.

References assert(), Available, checkResourceLimit(), llvm::dbgs(), getCriticalCount(), getScheduledLatency(), HazardRec, isTop(), LLVM_DEBUG, and SchedModel.

Referenced by bumpNode(), and pickOnlyChoice().

◆ bumpNode()

◆ checkHazard()

bool SchedBoundary::checkHazard ( SUnit * SU)

Does this SU have a hazard within the current instruction group.

The scheduler supports two modes of hazard recognition. The first is the ScheduleHazardRecognizer API. It is a fully general hazard recognizer that supports highly complicated in-order reservation tables (ScoreboardHazardRecognizer) and arbitrary target-specific logic.

The second is a streamlined mechanism that checks for hazards based on simple counters that the scheduler itself maintains. It explicitly checks for instruction dispatch limitations, including the number of micro-ops that can dispatch per cycle.

TODO: Also check whether the SU must start a new group.

Definition at line 2712 of file MachineScheduler.cpp.

References DAG, llvm::dbgs(), llvm::SUnit::getInstr(), getNextResourceCycle(), llvm::SUnit::hasReservedResource, HazardRec, isTop(), LLVM_DEBUG, llvm::make_range(), llvm::SUnit::NodeNum, llvm::ScheduleHazardRecognizer::NoHazard, and SchedModel.

Referenced by pickOnlyChoice(), and releaseNode().

◆ countResource()

unsigned SchedBoundary::countResource ( const MCSchedClassDesc * SC,
unsigned PIdx,
unsigned ReleaseAtCycle,
unsigned NextCycle,
unsigned AcquireAtCycle )

Add the given processor resource to this scheduled zone.

Parameters
ReleaseAtCycleindicates the number of consecutive (non-pipelined) cycles during which this resource is released.
AcquireAtCycleindicates the number of consecutive (non-pipelined) cycles at which the resource is aquired after issue (assuming no stalls).
Returns
the next cycle at which the instruction may execute without oversubscribing resources.

Definition at line 2916 of file MachineScheduler.cpp.

References assert(), llvm::Count, llvm::dbgs(), getCriticalCount(), getNextResourceCycle(), getResourceCount(), incExecutedResources(), LLVM_DEBUG, Rem, and SchedModel.

Referenced by bumpNode().

◆ dumpReservedCycles()

LLVM_DUMP_METHOD void SchedBoundary::dumpReservedCycles ( ) const

Dump the state of the information that tracks resource usage.

Dump the content of the ReservedCycles vector for the resources that are used in the basic block.

Definition at line 3194 of file MachineScheduler.cpp.

References llvm::dbgs(), LLVM_DUMP_METHOD, and SchedModel.

Referenced by dumpScheduledState(), and getNextResourceCycle().

◆ dumpScheduledState()

LLVM_DUMP_METHOD void SchedBoundary::dumpScheduledState ( ) const

◆ findMaxLatency()

unsigned SchedBoundary::findMaxLatency ( ArrayRef< SUnit * > ReadySUs)

◆ getCriticalCount()

unsigned llvm::SchedBoundary::getCriticalCount ( ) const
inline

Get the scaled count of scheduled micro-ops and resources, including executed resources.

Definition at line 1010 of file MachineScheduler.h.

References getResourceCount(), and SchedModel.

Referenced by bumpCycle(), bumpNode(), and countResource().

◆ getCurrCycle()

unsigned llvm::SchedBoundary::getCurrCycle ( ) const
inline

Number of cycles to issue the instructions scheduled in this zone.

Definition at line 985 of file MachineScheduler.h.

Referenced by llvm::GenericSchedulerBase::setPolicy().

◆ getCurrMOps()

unsigned llvm::SchedBoundary::getCurrMOps ( ) const
inline

◆ getDependentLatency()

unsigned llvm::SchedBoundary::getDependentLatency ( ) const
inline

Definition at line 991 of file MachineScheduler.h.

Referenced by computeRemLatency().

◆ getExecutedCount()

unsigned llvm::SchedBoundary::getExecutedCount ( ) const
inline

Get a scaled count for the minimum execution time of the scheduled micro-ops that are ready to execute by getExecutedCount.

Notice the feedback loop.

Definition at line 1019 of file MachineScheduler.h.

References SchedModel.

Referenced by dumpScheduledState().

◆ getLatencyStallCycles()

unsigned SchedBoundary::getLatencyStallCycles ( SUnit * SU)

Get the difference between the given SUnit's ready time and the current cycle.

Compute the stall cycles based on this SUnit's ready time.

Heuristics treat these "soft stalls" differently than the hard stall cycles based on CPU resources and computed by checkHazard(). A fully in-order model (MicroOpBufferSize==0) will not make use of this since instructions are not available for scheduling until they are ready. However, a weaker in-order model may use this for heuristics. For example, if a processor has in-order behavior when reading certain resources, this may come into play.

Definition at line 2593 of file MachineScheduler.cpp.

References llvm::SUnit::BotReadyCycle, isTop(), llvm::SUnit::isUnbuffered, and llvm::SUnit::TopReadyCycle.

Referenced by llvm::GCNMaxILPSchedStrategy::tryCandidate(), llvm::GCNMaxMemoryClauseSchedStrategy::tryCandidate(), llvm::GenericScheduler::tryCandidate(), and llvm::PPCPreRASchedStrategy::tryCandidate().

◆ getNextResourceCycle()

std::pair< unsigned, unsigned > SchedBoundary::getNextResourceCycle ( const MCSchedClassDesc * SC,
unsigned PIdx,
unsigned ReleaseAtCycle,
unsigned AcquireAtCycle )

Compute the next cycle at which the given processor resource can be scheduled.

Returns the next cycle and the index of the processor resource instance in the reserved cycles vector.

Definition at line 2631 of file MachineScheduler.cpp.

References assert(), llvm::dbgs(), dumpReservedCycles(), getNextResourceCycle(), getNextResourceCycleByInstance(), I, InvalidCycle, isUnbufferedGroup(), LLVM_DEBUG, llvm::make_range(), llvm::MischedDetailResourceBooking, and SchedModel.

Referenced by bumpNode(), checkHazard(), countResource(), and getNextResourceCycle().

◆ getNextResourceCycleByInstance()

unsigned SchedBoundary::getNextResourceCycleByInstance ( unsigned InstanceIndex,
unsigned ReleaseAtCycle,
unsigned AcquireAtCycle )

Compute the next cycle at which the given processor resource unit can be scheduled.

Definition at line 2605 of file MachineScheduler.cpp.

References InvalidCycle, isTop(), and SchedModel.

Referenced by getNextResourceCycle().

◆ getOtherResourceCount()

unsigned SchedBoundary::getOtherResourceCount ( unsigned & OtherCritIdx)

◆ getResourceCount()

unsigned llvm::SchedBoundary::getResourceCount ( unsigned ResIdx) const
inline

◆ getScheduledLatency()

unsigned llvm::SchedBoundary::getScheduledLatency ( ) const
inline

Get the number of latency cycles "covered" by the scheduled instructions.

This is the larger of the critical path within the zone and the number of cycles required to issue the instructions.

Definition at line 996 of file MachineScheduler.h.

Referenced by bumpCycle(), bumpNode(), and llvm::tryLatency().

◆ getUnscheduledLatency()

unsigned llvm::SchedBoundary::getUnscheduledLatency ( SUnit * SU) const
inline

Definition at line 1000 of file MachineScheduler.h.

References llvm::SUnit::getDepth(), llvm::SUnit::getHeight(), and isTop().

Referenced by findMaxLatency().

◆ getZoneCritResIdx()

unsigned llvm::SchedBoundary::getZoneCritResIdx ( ) const
inline

Definition at line 1024 of file MachineScheduler.h.

Referenced by llvm::GenericSchedulerBase::setPolicy().

◆ incExecutedResources()

void SchedBoundary::incExecutedResources ( unsigned PIdx,
unsigned Count )

Definition at line 2900 of file MachineScheduler.cpp.

References llvm::Count.

Referenced by countResource().

◆ init()

void SchedBoundary::init ( ScheduleDAGMI * dag,
const TargetSchedModel * smodel,
SchedRemainder * rem )

Definition at line 2558 of file MachineScheduler.cpp.

References DAG, InvalidCycle, isUnbufferedGroup(), Rem, rem, reset(), and SchedModel.

◆ isResourceLimited()

bool llvm::SchedBoundary::isResourceLimited ( ) const
inline

Definition at line 1027 of file MachineScheduler.h.

Referenced by llvm::GenericSchedulerBase::setPolicy().

◆ isTop()

◆ isUnbufferedGroup()

bool llvm::SchedBoundary::isUnbufferedGroup ( unsigned PIdx) const
inline

Definition at line 1041 of file MachineScheduler.h.

References SchedModel.

Referenced by getNextResourceCycle(), and init().

◆ operator=()

SchedBoundary & llvm::SchedBoundary::operator= ( const SchedBoundary & other)
delete

References SchedBoundary().

◆ pickOnlyChoice()

SUnit * SchedBoundary::pickOnlyChoice ( )

Call this before applying any other heuristics to the Available queue.

If this queue only has one ready candidate, return it.

Updates the Available/Pending Q's if necessary and returns the single available instruction, or NULL if there are multiple candidates.

As a side effect, defer any nodes that now hit a hazard, and advance the cycle until at least one node is ready. If multiple instructions are ready, return NULL.

Definition at line 3159 of file MachineScheduler.cpp.

References Available, bumpCycle(), checkHazard(), I, LLVM_DEBUG, Pending, and releasePending().

◆ releaseNode()

void SchedBoundary::releaseNode ( SUnit * SU,
unsigned ReadyCycle,
bool InPQueue,
unsigned Idx = 0 )

Release SU to make it ready.

If it's not in hazard, remove it from pending queue (if already in) and push into available queue. Otherwise, push the SU into pending queue.

Parameters
SUThe unit to be released.
ReadyCycleUntil which cycle the unit is ready.
InPQueueWhether SU is already in pending queue.
IdxPosition offset in pending queue (if in it).

Definition at line 2814 of file MachineScheduler.cpp.

References assert(), Available, checkHazard(), llvm::dbgs(), llvm::SUnit::getInstr(), LLVM_DEBUG, llvm::SUnit::NodeNum, Pending, ReadyListLimit, and SchedModel.

Referenced by releasePending().

◆ releasePending()

void SchedBoundary::releasePending ( )

Release pending ready nodes in to the available queue.

This makes them visible to heuristics.

Definition at line 3118 of file MachineScheduler.cpp.

References Available, llvm::SUnit::BotReadyCycle, llvm::dbgs(), I, isTop(), LLVM_DEBUG, llvm::SUnit::NodeNum, Pending, ReadyListLimit, releaseNode(), and llvm::SUnit::TopReadyCycle.

Referenced by pickOnlyChoice().

◆ removeReady()

void SchedBoundary::removeReady ( SUnit * SU)

Remove SU from the ready set for this boundary.

Definition at line 3147 of file MachineScheduler.cpp.

References assert(), Available, and Pending.

◆ reset()

void SchedBoundary::reset ( )

Definition at line 2501 of file MachineScheduler.cpp.

References assert(), Available, HazardRec, and Pending.

Referenced by init(), and SchedBoundary().

Member Data Documentation

◆ Available

◆ DAG

ScheduleDAGMI* llvm::SchedBoundary::DAG = nullptr

◆ HazardRec

ScheduleHazardRecognizer* llvm::SchedBoundary::HazardRec = nullptr

Definition at line 872 of file MachineScheduler.h.

Referenced by bumpCycle(), bumpNode(), checkHazard(), reset(), and ~SchedBoundary().

◆ Pending

ReadyQueue llvm::SchedBoundary::Pending

◆ Rem

SchedRemainder* llvm::SchedBoundary::Rem = nullptr

Definition at line 867 of file MachineScheduler.h.

Referenced by bumpNode(), countResource(), getOtherResourceCount(), and init().

◆ SchedModel


The documentation for this class was generated from the following files: