|
| From: | rail shafigulin |
| Subject: | Re: gtags-cscope/global finds incorrect callers |
| Date: | Fri, 25 Sep 2015 10:19:53 -0700 |
Hi,> Fist and third results are incorrect.Could you post the following source code so that all peoplein this list may be seen?llvm-or1k/include/llvm/CodeGen/MachineScheduler.h
//==- MachineScheduler.h - MachineInstr Scheduling Pass ----------*- C++ -*-==//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file provides an interface for customizing the standard MachineScheduler
// pass. Note that the entire pass may be replaced as follows:
//
// <Target>TargetMachine::createPassConfig(PassManagerBase &PM) {
// PM.substitutePass(&MachineSchedulerID, &CustomSchedulerPassID);
// ...}
//
// The MachineScheduler pass is only responsible for choosing the regions to be
// scheduled. Targets can override the DAG builder and scheduler without
// replacing the pass as follows:
//
// ScheduleDAGInstrs *<Target>PassConfig::
// createMachineScheduler(MachineSchedContext *C) {
// return new CustomMachineScheduler(C);
// }
//
// The default scheduler, ScheduleDAGMILive, builds the DAG and drives list
// scheduling while updating the instruction stream, register pressure, and live
// intervals. Most targets don't need to override the DAG builder and list
// schedulier, but subtargets that require custom scheduling heuristics may
// plugin an alternate MachineSchedStrategy. The strategy is responsible for
// selecting the highest priority node from the list:
//
// ScheduleDAGInstrs *<Target>PassConfig::
// createMachineScheduler(MachineSchedContext *C) {
// return new ScheduleDAGMI(C, CustomStrategy(C));
// }
//
// The DAG builder can also be customized in a sense by adding DAG mutations
// that will run after DAG building and before list scheduling. DAG mutations
// can adjust dependencies based on target-specific knowledge or add weak edges
// to aid heuristics:
//
// ScheduleDAGInstrs *<Target>PassConfig::
// createMachineScheduler(MachineSchedContext *C) {
// ScheduleDAGMI *DAG = new ScheduleDAGMI(C, CustomStrategy(C));
// DAG->addMutation(new CustomDependencies(DAG->TII, DAG->TRI));
// return DAG;
// }
//
// A target that supports alternative schedulers can use the
// MachineSchedRegistry to allow command line selection. This can be done by
// implementing the following boilerplate:
//
// static ScheduleDAGInstrs *createCustomMachineSched(MachineSchedContext *C) {
// return new CustomMachineScheduler(C);
// }
// static MachineSchedRegistry
// SchedCustomRegistry("custom", "Run my target's custom scheduler",
// createCustomMachineSched);
//
//
// Finally, subtargets that don't need to implement custom heuristics but would
// like to configure the GenericScheduler's policy for a given scheduler region,
// including scheduling direction and register pressure tracking policy, can do
// this:
//
// void <SubTarget>Subtarget::
// overrideSchedPolicy(MachineSchedPolicy &Policy,
// MachineInstr *begin,
// MachineInstr *end,
// unsigned NumRegionInstrs) const {
// Policy.<Flag> = true;
// }
//
//===----------------------------------------------------------------------===//
#ifndef LLVM_CODEGEN_MACHINESCHEDULER_H
#define LLVM_CODEGEN_MACHINESCHEDULER_H
#include "llvm/CodeGen/MachinePassRegistry.h"
#include "llvm/CodeGen/RegisterPressure.h"
#include "llvm/CodeGen/ScheduleDAGInstrs.h"
namespace llvm {
extern cl::opt<bool> ForceTopDown;
extern cl::opt<bool> ForceBottomUp;
class AliasAnalysis;
class LiveIntervals;
class MachineDominatorTree;
class MachineLoopInfo;
class RegisterClassInfo;
class ScheduleDAGInstrs;
class SchedDFSResult;
class ScheduleHazardRecognizer;
/// MachineSchedContext provides enough context from the MachineScheduler pass
/// for the target to instantiate a scheduler.
struct MachineSchedContext {
MachineFunction *MF;
const MachineLoopInfo *MLI;
const MachineDominatorTree *MDT;
const TargetPassConfig *PassConfig;
AliasAnalysis *AA;
LiveIntervals *LIS;
RegisterClassInfo *RegClassInfo;
MachineSchedContext();
virtual ~MachineSchedContext();
};
/// MachineSchedRegistry provides a selection of available machine instruction
/// schedulers.
class MachineSchedRegistry : public MachinePassRegistryNode {
public:
typedef ScheduleDAGInstrs *(*ScheduleDAGCtor)(MachineSchedContext *);
// RegisterPassParser requires a (misnamed) FunctionPassCtor type.
typedef ScheduleDAGCtor FunctionPassCtor;
static MachinePassRegistry Registry;
MachineSchedRegistry(const char *N, const char *D, ScheduleDAGCtor C)
: MachinePassRegistryNode(N, D, (MachinePassCtor)C) {
Registry.Add(this);
}
~MachineSchedRegistry() { Registry.Remove(this); }
// Accessors.
//
MachineSchedRegistry *getNext() const {
return (MachineSchedRegistry *)MachinePassRegistryNode::getNext();
}
static MachineSchedRegistry *getList() {
return (MachineSchedRegistry *)Registry.getList();
}
static void setListener(MachinePassRegistryListener *L) {
Registry.setListener(L);
}
};
class ScheduleDAGMI;
/// Define a generic scheduling policy for targets that don't provide their own
/// MachineSchedStrategy. This can be overriden for each scheduling region
/// before building the DAG.
struct MachineSchedPolicy {
// Allow the scheduler to disable register pressure tracking.
bool ShouldTrackPressure;
// Allow the scheduler to force top-down or bottom-up scheduling. If neither
// is true, the scheduler runs in both directions and converges.
bool OnlyTopDown;
bool OnlyBottomUp;
MachineSchedPolicy(): ShouldTrackPressure(false), OnlyTopDown(false),
OnlyBottomUp(false) {}
};
/// MachineSchedStrategy - Interface to the scheduling algorithm used by
/// ScheduleDAGMI.
///
/// Initialization sequence:
/// initPolicy -> shouldTrackPressure -> initialize(DAG) -> registerRoots
class MachineSchedStrategy {
virtual void anchor();
public:
virtual ~MachineSchedStrategy() {}
/// Optionally override the per-region scheduling policy.
virtual void initPolicy(MachineBasicBlock::iterator Begin,
MachineBasicBlock::iterator End,
unsigned NumRegionInstrs) {}
/// Check if pressure tracking is needed before building the DAG and
/// initializing this strategy. Called after initPolicy.
virtual bool shouldTrackPressure() const { return true; }
/// Initialize the strategy after building the DAG for a new region.
virtual void initialize(ScheduleDAGMI *DAG) = 0;
/// Notify this strategy that all roots have been released (including those
/// that depend on EntrySU or ExitSU).
virtual void registerRoots() {}
/// Pick the next node to schedule, or return NULL. Set IsTopNode to true to
/// schedule the node at the top of the unscheduled region. Otherwise it will
/// be scheduled at the bottom.
virtual SUnit *pickNode(bool &IsTopNode) = 0;
/// \brief Scheduler callback to notify that a new subtree is scheduled.
virtual void scheduleTree(unsigned SubtreeID) {}
/// Notify MachineSchedStrategy that ScheduleDAGMI has scheduled an
/// instruction and updated scheduled/remaining flags in the DAG nodes.
virtual void schedNode(SUnit *SU, bool IsTopNode) = 0;
/// When all predecessor dependencies have been resolved, free this node for
/// top-down scheduling.
virtual void releaseTopNode(SUnit *SU) = 0;
/// When all successor dependencies have been resolved, free this node for
/// bottom-up scheduling.
virtual void releaseBottomNode(SUnit *SU) = 0;
};
/// Mutate the DAG as a postpass after normal DAG building.
class ScheduleDAGMutation {
virtual void anchor();
public:
virtual ~ScheduleDAGMutation() {}
virtual void apply(ScheduleDAGMI *DAG) = 0;
};
/// ScheduleDAGMI is an implementation of ScheduleDAGInstrs that simply
/// schedules machine instructions according to the given MachineSchedStrategy
/// without much extra book-keeping. This is the common functionality between
/// PreRA and PostRA MachineScheduler.
class ScheduleDAGMI : public ScheduleDAGInstrs {
protected:
AliasAnalysis *AA;
MachineSchedStrategy *SchedImpl;
/// Topo - A topological ordering for SUnits which permits fast IsReachable
/// and similar queries.
ScheduleDAGTopologicalSort Topo;
/// Ordered list of DAG postprocessing steps.
std::vector<ScheduleDAGMutation*> Mutations;
/// The top of the unscheduled zone.
MachineBasicBlock::iterator CurrentTop;
/// The bottom of the unscheduled zone.
MachineBasicBlock::iterator CurrentBottom;
/// Record the next node in a scheduled cluster.
const SUnit *NextClusterPred;
const SUnit *NextClusterSucc;
#ifndef NDEBUG
/// The number of instructions scheduled so far. Used to cut off the
/// scheduler at the point determined by misched-cutoff.
unsigned NumInstrsScheduled;
#endif
public:
ScheduleDAGMI(MachineSchedContext *C, MachineSchedStrategy *S, bool IsPostRA):
ScheduleDAGInstrs(*C->MF, *C->MLI, *C->MDT, IsPostRA,
/*RemoveKillFlags=*/IsPostRA, C->LIS),
AA(C->AA), SchedImpl(S), Topo(SUnits, &ExitSU), CurrentTop(),
CurrentBottom(), NextClusterPred(NULL), NextClusterSucc(NULL) {
#ifndef NDEBUG
NumInstrsScheduled = 0;
#endif
}
virtual ~ScheduleDAGMI();
/// Return true if this DAG supports VReg liveness and RegPressure.
virtual bool hasVRegLiveness() const { return false; }
/// Add a postprocessing step to the DAG builder.
/// Mutations are applied in the order that they are added after normal DAG
/// building and before MachineSchedStrategy initialization.
///
/// ScheduleDAGMI takes ownership of the Mutation object.
void addMutation(ScheduleDAGMutation *Mutation) {
Mutations.push_back(Mutation);
}
/// \brief True if an edge can be added from PredSU to SuccSU without creating
/// a cycle.
bool canAddEdge(SUnit *SuccSU, SUnit *PredSU);
/// \brief Add a DAG edge to the given SU with the given predecessor
/// dependence data.
///
/// \returns true if the edge may be added without creating a cycle OR if an
/// equivalent edge already existed (false indicates failure).
bool addEdge(SUnit *SuccSU, const SDep &PredDep);
MachineBasicBlock::iterator top() const { return CurrentTop; }
MachineBasicBlock::iterator bottom() const { return CurrentBottom; }
/// Implement the ScheduleDAGInstrs interface for handling the next scheduling
/// region. This covers all instructions in a block, while schedule() may only
/// cover a subset.
void enterRegion(MachineBasicBlock *bb,
MachineBasicBlock::iterator begin,
MachineBasicBlock::iterator end,
unsigned regioninstrs) override;
/// Implement ScheduleDAGInstrs interface for scheduling a sequence of
/// reorderable instructions.
void schedule() override;
/// Change the position of an instruction within the basic block and update
/// live ranges and region boundary iterators.
void moveInstruction(MachineInstr *MI, MachineBasicBlock::iterator InsertPos);
const SUnit *getNextClusterPred() const { return NextClusterPred; }
const SUnit *getNextClusterSucc() const { return NextClusterSucc; }
void viewGraph(const Twine &Name, const Twine &Title) override;
void viewGraph() override;
protected:
// Top-Level entry points for the schedule() driver...
/// Apply each ScheduleDAGMutation step in order. This allows different
/// instances of ScheduleDAGMI to perform custom DAG postprocessing.
void postprocessDAG();
/// Release ExitSU predecessors and setup scheduler queues.
void initQueues(ArrayRef<SUnit*> TopRoots, ArrayRef<SUnit*> BotRoots);
/// Update scheduler DAG and queues after scheduling an instruction.
void updateQueues(SUnit *SU, bool IsTopNode);
/// Reinsert debug_values recorded in ScheduleDAGInstrs::DbgValues.
void placeDebugValues();
/// \brief dump the scheduled Sequence.
void dumpSchedule() const;
// Lesser helpers...
bool checkSchedLimit();
void findRootsAndBiasEdges(SmallVectorImpl<SUnit*> &TopRoots,
SmallVectorImpl<SUnit*> &BotRoots);
void releaseSucc(SUnit *SU, SDep *SuccEdge);
void releaseSuccessors(SUnit *SU);
void releasePred(SUnit *SU, SDep *PredEdge);
void releasePredecessors(SUnit *SU);
};
/// ScheduleDAGMILive is an implementation of ScheduleDAGInstrs that schedules
/// machine instructions while updating LiveIntervals and tracking regpressure.
class ScheduleDAGMILive : public ScheduleDAGMI {
protected:
RegisterClassInfo *RegClassInfo;
/// Information about DAG subtrees. If DFSResult is NULL, then SchedulerTrees
/// will be empty.
SchedDFSResult *DFSResult;
BitVector ScheduledTrees;
MachineBasicBlock::iterator LiveRegionEnd;
// Map each SU to its summary of pressure changes. This array is updated for
// liveness during bottom-up scheduling. Top-down scheduling may proceed but
// has no affect on the pressure diffs.
PressureDiffs SUPressureDiffs;
/// Register pressure in this region computed by initRegPressure.
bool ShouldTrackPressure;
IntervalPressure RegPressure;
RegPressureTracker RPTracker;
/// List of pressure sets that exceed the target's pressure limit before
/// scheduling, listed in increasing set ID order. Each pressure set is paired
/// with its max pressure in the currently scheduled regions.
std::vector<PressureChange> RegionCriticalPSets;
/// The top of the unscheduled zone.
IntervalPressure TopPressure;
RegPressureTracker TopRPTracker;
/// The bottom of the unscheduled zone.
IntervalPressure BotPressure;
RegPressureTracker BotRPTracker;
public:
ScheduleDAGMILive(MachineSchedContext *C, MachineSchedStrategy *S):
ScheduleDAGMI(C, S, /*IsPostRA=*/false), RegClassInfo(C->RegClassInfo),
DFSResult(0), ShouldTrackPressure(false), RPTracker(RegPressure),
TopRPTracker(TopPressure), BotRPTracker(BotPressure)
{}
virtual ~ScheduleDAGMILive();
/// Return true if this DAG supports VReg liveness and RegPressure.
bool hasVRegLiveness() const override { return true; }
/// \brief Return true if register pressure tracking is enabled.
bool isTrackingPressure() const { return ShouldTrackPressure; }
/// Get current register pressure for the top scheduled instructions.
const IntervalPressure &getTopPressure() const { return TopPressure; }
const RegPressureTracker &getTopRPTracker() const { return TopRPTracker; }
/// Get current register pressure for the bottom scheduled instructions.
const IntervalPressure &getBotPressure() const { return BotPressure; }
const RegPressureTracker &getBotRPTracker() const { return BotRPTracker; }
/// Get register pressure for the entire scheduling region before scheduling.
const IntervalPressure &getRegPressure() const { return RegPressure; }
const std::vector<PressureChange> &getRegionCriticalPSets() const {
return RegionCriticalPSets;
}
PressureDiff &getPressureDiff(const SUnit *SU) {
return SUPressureDiffs[SU->NodeNum];
}
/// Compute a DFSResult after DAG building is complete, and before any
/// queue comparisons.
void computeDFSResult();
/// Return a non-null DFS result if the scheduling strategy initialized it.
const SchedDFSResult *getDFSResult() const { return DFSResult; }
BitVector &getScheduledTrees() { return ScheduledTrees; }
/// Implement the ScheduleDAGInstrs interface for handling the next scheduling
/// region. This covers all instructions in a block, while schedule() may only
/// cover a subset.
void enterRegion(MachineBasicBlock *bb,
MachineBasicBlock::iterator begin,
MachineBasicBlock::iterator end,
unsigned regioninstrs) override;
/// Implement ScheduleDAGInstrs interface for scheduling a sequence of
/// reorderable instructions.
void schedule() override;
/// Compute the cyclic critical path through the DAG.
unsigned computeCyclicCriticalPath();
protected:
// Top-Level entry points for the schedule() driver...
/// Call ScheduleDAGInstrs::buildSchedGraph with register pressure tracking
/// enabled. This sets up three trackers. RPTracker will cover the entire DAG
/// region, TopTracker and BottomTracker will be initialized to the top and
/// bottom of the DAG region without covereing any unscheduled instruction.
void buildDAGWithRegPressure();
/// Move an instruction and update register pressure.
void scheduleMI(SUnit *SU, bool IsTopNode);
// Lesser helpers...
void initRegPressure();
void updatePressureDiffs(ArrayRef<unsigned> LiveUses);
void updateScheduledPressure(const SUnit *SU,
const std::vector<unsigned> &NewMaxPressure);
};
//===----------------------------------------------------------------------===//
///
/// Helpers for implementing custom MachineSchedStrategy classes. These take
/// care of the book-keeping associated with list scheduling heuristics.
///
//===----------------------------------------------------------------------===//
/// ReadyQueue encapsulates vector of "ready" SUnits with basic convenience
/// methods for pushing and removing nodes. ReadyQueue's are uniquely identified
/// by an ID. SUnit::NodeQueueId is a mask of the ReadyQueues the SUnit is in.
///
/// This is a convenience class that may be used by implementations of
/// MachineSchedStrategy.
class ReadyQueue {
unsigned ID;
std::string Name;
std::vector<SUnit*> Queue;
public:
ReadyQueue(unsigned id, const Twine &name): ID(id), Name(name.str()) {}
unsigned getID() const { return ID; }
StringRef getName() const { return Name; }
// SU is in this queue if it's NodeQueueID is a superset of this ID.
bool isInQueue(SUnit *SU) const { return (SU->NodeQueueId & ID); }
bool empty() const { return Queue.empty(); }
void clear() { Queue.clear(); }
unsigned size() const { return Queue.size(); }
typedef std::vector<SUnit*>::iterator iterator;
iterator begin() { return Queue.begin(); }
iterator end() { return Queue.end(); }
ArrayRef<SUnit*> elements() { return Queue; }
iterator find(SUnit *SU) {
return std::find(Queue.begin(), Queue.end(), SU);
}
void push(SUnit *SU) {
Queue.push_back(SU);
SU->NodeQueueId |= ID;
}
iterator remove(iterator I) {
(*I)->NodeQueueId &= ~ID;
*I = Queue.back();
unsigned idx = I - Queue.begin();
Queue.pop_back();
return Queue.begin() + idx;
}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
void dump();
#endif
};
/// Summarize the unscheduled region.
struct SchedRemainder {
// Critical path through the DAG in expected latency.
unsigned CriticalPath;
unsigned CyclicCritPath;
// Scaled count of micro-ops left to schedule.
unsigned RemIssueCount;
bool IsAcyclicLatencyLimited;
// Unscheduled resources
SmallVector<unsigned, 16> RemainingCounts;
void reset() {
CriticalPath = 0;
CyclicCritPath = 0;
RemIssueCount = 0;
IsAcyclicLatencyLimited = false;
RemainingCounts.clear();
}
SchedRemainder() { reset(); }
void init(ScheduleDAGMI *DAG, const TargetSchedModel *SchedModel);
};
/// 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.
class SchedBoundary {
public:
/// SUnit::NodeQueueId: 0 (none), 1 (top), 2 (bot), 3 (both)
enum {
TopQID = 1,
BotQID = 2,
LogMaxQID = 2
};
ScheduleDAGMI *DAG;
const TargetSchedModel *SchedModel;
SchedRemainder *Rem;
ReadyQueue Available;
ReadyQueue Pending;
ScheduleHazardRecognizer *HazardRec;
private:
/// True if the pending Q should be checked/updated before scheduling another
/// instruction.
bool CheckPending;
// For heuristics, keep a list of the nodes that immediately depend on the
// most recently scheduled node.
SmallPtrSet<const SUnit*, 8> NextSUs;
/// Number of cycles it takes to issue the instructions scheduled in this
/// zone. It is defined as: scheduled-micro-ops / issue-width + stalls.
/// See getStalls().
unsigned CurrCycle;
/// Micro-ops issued in the current cycle
unsigned CurrMOps;
/// MinReadyCycle - Cycle of the soonest available instruction.
unsigned MinReadyCycle;
// The expected latency of the critical path in this scheduled zone.
unsigned ExpectedLatency;
// The latency of dependence chains leading into this zone.
// For each node scheduled bottom-up: DLat = max DLat, N.Depth.
// For each cycle scheduled: DLat -= 1.
unsigned DependentLatency;
/// Count the scheduled (issued) micro-ops that can be retired by
/// time=CurrCycle assuming the first scheduled instr is retired at time=0.
unsigned RetiredMOps;
// Count scheduled resources that have been executed. Resources are
// considered executed if they become ready in the time that it takes to
// saturate any resource including the one in question. Counts are scaled
// for direct comparison with other resources. Counts can be compared with
// MOps * getMicroOpFactor and Latency * getLatencyFactor.
SmallVector<unsigned, 16> ExecutedResCounts;
/// Cache the max count for a single resource.
unsigned MaxExecutedResCount;
// Cache the critical resources ID in this scheduled zone.
unsigned ZoneCritResIdx;
// Is the scheduled region resource limited vs. latency limited.
bool IsResourceLimited;
// Record the highest cycle at which each resource has been reserved by a
// scheduled instruction.
SmallVector<unsigned, 16> ReservedCycles;
#ifndef NDEBUG
// Remember the greatest operand latency as an upper bound on the number of
// times we should retry the pending queue because of a hazard.
unsigned MaxObservedLatency;
#endif
public:
/// Pending queues extend the ready queues with the same ID and the
/// PendingFlag set.
SchedBoundary(unsigned ID, const Twine &Name):
DAG(0), SchedModel(0), Rem(0), Available(ID, Name+".A"),
Pending(ID << LogMaxQID, Name+".P"),
HazardRec(0) {
reset();
}
~SchedBoundary();
void reset();
void init(ScheduleDAGMI *dag, const TargetSchedModel *smodel,
SchedRemainder *rem);
bool isTop() const {
return Available.getID() == TopQID;
}
/// Number of cycles to issue the instructions scheduled in this zone.
unsigned getCurrCycle() const { return CurrCycle; }
/// Micro-ops issued in the current cycle
unsigned getCurrMOps() const { return CurrMOps; }
/// Return true if the given SU is used by the most recently scheduled
/// instruction.
bool isNextSU(const SUnit *SU) const { return NextSUs.count(SU); }
// The latency of dependence chains leading into this zone.
unsigned getDependentLatency() const { return DependentLatency; }
/// 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.
unsigned getScheduledLatency() const {
return std::max(ExpectedLatency, CurrCycle);
}
unsigned getUnscheduledLatency(SUnit *SU) const {
return isTop() ? SU->getHeight() : SU->getDepth();
}
unsigned getResourceCount(unsigned ResIdx) const {
return ExecutedResCounts[ResIdx];
}
/// Get the scaled count of scheduled micro-ops and resources, including
/// executed resources.
unsigned getCriticalCount() const {
if (!ZoneCritResIdx)
return RetiredMOps * SchedModel->getMicroOpFactor();
return getResourceCount(ZoneCritResIdx);
}
/// 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.
unsigned getExecutedCount() const {
return std::max(CurrCycle * SchedModel->getLatencyFactor(),
MaxExecutedResCount);
}
unsigned getZoneCritResIdx() const { return ZoneCritResIdx; }
// Is the scheduled region resource limited vs. latency limited.
bool isResourceLimited() const { return IsResourceLimited; }
/// Get the difference between the given SUnit's ready time and the current
/// cycle.
unsigned getLatencyStallCycles(SUnit *SU);
unsigned getNextResourceCycle(unsigned PIdx, unsigned Cycles);
bool checkHazard(SUnit *SU);
unsigned findMaxLatency(ArrayRef<SUnit*> ReadySUs);
unsigned getOtherResourceCount(unsigned &OtherCritIdx);
void releaseNode(SUnit *SU, unsigned ReadyCycle);
void releaseTopNode(SUnit *SU);
void releaseBottomNode(SUnit *SU);
void bumpCycle(unsigned NextCycle);
void incExecutedResources(unsigned PIdx, unsigned Count);
unsigned countResource(unsigned PIdx, unsigned Cycles, unsigned ReadyCycle);
void bumpNode(SUnit *SU);
void releasePending();
void removeReady(SUnit *SU);
/// Call this before applying any other heuristics to the Available queue.
/// Updates the Available/Pending Q's if necessary and returns the single
/// available instruction, or NULL if there are multiple candidates.
SUnit *pickOnlyChoice();
#ifndef NDEBUG
void dumpScheduledState();
#endif
};
} // namespace llvm
llvm-or1k/lib/CodeGen/SelectionDAG/ScheduleDAGVLIW.cpp
//===- ScheduleDAGVLIW.cpp - SelectionDAG list scheduler for VLIW -*- C++ -*-=//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This implements a top-down list scheduler, using standard algorithms.
// The basic approach uses a priority queue of available nodes to schedule.
// One at a time, nodes are taken from the priority queue (thus in priority
// order), checked for legality to schedule, and emitted if legal.
//
// Nodes may not be legal to schedule either due to structural hazards (e.g.
// pipeline or resource constraints) or because an input to the instruction has
// not completed execution.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "pre-RA-sched"
#include "llvm/CodeGen/SchedulerRegistry.h"
#include "ScheduleDAGSDNodes.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/LatencyPriorityQueue.h"
#include "llvm/CodeGen/ResourcePriorityQueue.h"
#include "llvm/CodeGen/ScheduleHazardRecognizer.h"
#include "llvm/CodeGen/SelectionDAGISel.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include <climits>
using namespace llvm;
STATISTIC(NumNoops , "Number of noops inserted");
STATISTIC(NumStalls, "Number of pipeline stalls");
static RegisterScheduler
VLIWScheduler("vliw-td", "VLIW scheduler",
createVLIWDAGScheduler);
namespace {
//===----------------------------------------------------------------------===//
/// ScheduleDAGVLIW - The actual DFA list scheduler implementation. This
/// supports / top-down scheduling.
///
class ScheduleDAGVLIW : public ScheduleDAGSDNodes {
private:
/// AvailableQueue - The priority queue to use for the available SUnits.
///
SchedulingPriorityQueue *AvailableQueue;
/// PendingQueue - This contains all of the instructions whose operands have
/// been issued, but their results are not ready yet (due to the latency of
/// the operation). Once the operands become available, the instruction is
/// added to the AvailableQueue.
std::vector<SUnit*> PendingQueue;
/// HazardRec - The hazard recognizer to use.
ScheduleHazardRecognizer *HazardRec;
/// AA - AliasAnalysis for making memory reference queries.
AliasAnalysis *AA;
public:
ScheduleDAGVLIW(MachineFunction &mf,
AliasAnalysis *aa,
SchedulingPriorityQueue *availqueue)
: ScheduleDAGSDNodes(mf), AvailableQueue(availqueue), AA(aa) {
const TargetMachine &tm = mf.getTarget();
HazardRec = tm.getInstrInfo()->CreateTargetHazardRecognizer(&tm, this);
}
~ScheduleDAGVLIW() {
delete HazardRec;
delete AvailableQueue;
}
void Schedule() override;
private:
void releaseSucc(SUnit *SU, const SDep &D);
void releaseSuccessors(SUnit *SU);
void scheduleNodeTopDown(SUnit *SU, unsigned CurCycle);
void listScheduleTopDown();
};
} // end anonymous namespace
/// Schedule - Schedule the DAG using list scheduling.
void ScheduleDAGVLIW::Schedule() {
DEBUG(dbgs()
<< "********** List Scheduling BB#" << BB->getNumber()
<< " '" << BB->getName() << "' **********\n");
// Build the scheduling graph.
BuildSchedGraph(AA);
AvailableQueue->initNodes(SUnits);
listScheduleTopDown();
AvailableQueue->releaseState();
}
//===----------------------------------------------------------------------===//
// Top-Down Scheduling
//===----------------------------------------------------------------------===//
/// releaseSucc - Decrement the NumPredsLeft count of a successor. Add it to
/// the PendingQueue if the count reaches zero. Also update its cycle bound.
void ScheduleDAGVLIW::releaseSucc(SUnit *SU, const SDep &D) {
SUnit *SuccSU = D.getSUnit();
#ifndef NDEBUG
if (SuccSU->NumPredsLeft == 0) {
dbgs() << "*** Scheduling failed! ***\n";
SuccSU->dump(this);
dbgs() << " has been released too many times!\n";
llvm_unreachable(0);
}
#endif
assert(!D.isWeak() && "unexpected artificial DAG edge");
--SuccSU->NumPredsLeft;
SuccSU->setDepthToAtLeast(SU->getDepth() + D.getLatency());
// If all the node's predecessors are scheduled, this node is ready
// to be scheduled. Ignore the special ExitSU node.
if (SuccSU->NumPredsLeft == 0 && SuccSU != &ExitSU) {
PendingQueue.push_back(SuccSU);
}
}
void ScheduleDAGVLIW::releaseSuccessors(SUnit *SU) {
// Top down: release successors.
for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
I != E; ++I) {
assert(!I->isAssignedRegDep() &&
"The list-td scheduler doesn't yet support physreg dependencies!");
releaseSucc(SU, *I);
}
}
/// scheduleNodeTopDown - Add the node to the schedule. Decrement the pending
/// count of its successors. If a successor pending count is zero, add it to
/// the Available queue.
void ScheduleDAGVLIW::scheduleNodeTopDown(SUnit *SU, unsigned CurCycle) {
DEBUG(dbgs() << "*** Scheduling [" << CurCycle << "]: ");
DEBUG(SU->dump(this));
Sequence.push_back(SU);
assert(CurCycle >= SU->getDepth() && "Node scheduled above its depth!");
SU->setDepthToAtLeast(CurCycle);
releaseSuccessors(SU);
SU->isScheduled = true;
AvailableQueue->scheduledNode(SU);
}
/// listScheduleTopDown - The main loop of list scheduling for top-down
/// schedulers.
void ScheduleDAGVLIW::listScheduleTopDown() {
unsigned CurCycle = 0;
// Release any successors of the special Entry node.
releaseSuccessors(&EntrySU);
// All leaves to AvailableQueue.
for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
// It is available if it has no predecessors.
if (SUnits[i].Preds.empty()) {
AvailableQueue->push(&SUnits[i]);
SUnits[i].isAvailable = true;
}
}
// While AvailableQueue is not empty, grab the node with the highest
// priority. If it is not ready put it back. Schedule the node.
std::vector<SUnit*> NotReady;
Sequence.reserve(SUnits.size());
while (!AvailableQueue->empty() || !PendingQueue.empty()) {
// Check to see if any of the pending instructions are ready to issue. If
// so, add them to the available queue.
for (unsigned i = 0, e = PendingQueue.size(); i != e; ++i) {
if (PendingQueue[i]->getDepth() == CurCycle) {
AvailableQueue->push(PendingQueue[i]);
PendingQueue[i]->isAvailable = true;
PendingQueue[i] = PendingQueue.back();
PendingQueue.pop_back();
--i; --e;
}
else {
assert(PendingQueue[i]->getDepth() > CurCycle && "Negative latency?");
}
}
// If there are no instructions available, don't try to issue anything, and
// don't advance the hazard recognizer.
if (AvailableQueue->empty()) {
// Reset DFA state.
AvailableQueue->scheduledNode(0);
++CurCycle;
continue;
}
SUnit *FoundSUnit = 0;
bool HasNoopHazards = false;
while (!AvailableQueue->empty()) {
SUnit *CurSUnit = AvailableQueue->pop();
ScheduleHazardRecognizer::HazardType HT =
HazardRec->getHazardType(CurSUnit, 0/*no stalls*/);
if (HT == ScheduleHazardRecognizer::NoHazard) {
FoundSUnit = CurSUnit;
break;
}
// Remember if this is a noop hazard.
HasNoopHazards |= HT == ScheduleHazardRecognizer::NoopHazard;
NotReady.push_back(CurSUnit);
}
// Add the nodes that aren't ready back onto the available list.
if (!NotReady.empty()) {
AvailableQueue->push_all(NotReady);
NotReady.clear();
}
// If we found a node to schedule, do it now.
if (FoundSUnit) {
scheduleNodeTopDown(FoundSUnit, CurCycle);
HazardRec->EmitInstruction(FoundSUnit);
// If this is a pseudo-op node, we don't want to increment the current
// cycle.
if (FoundSUnit->Latency) // Don't increment CurCycle for pseudo-ops!
++CurCycle;
} else if (!HasNoopHazards) {
// Otherwise, we have a pipeline stall, but no other problem, just advance
// the current cycle and try again.
DEBUG(dbgs() << "*** Advancing cycle, no work to do\n");
HazardRec->AdvanceCycle();
++NumStalls;
++CurCycle;
} else {
// Otherwise, we have no instructions to issue and we have instructions
// that will fault if we don't do this right. This is the case for
// processors without pipeline interlocks and other cases.
DEBUG(dbgs() << "*** Emitting noop\n");
HazardRec->EmitNoop();
Sequence.push_back(0); // NULL here means noop
++NumNoops;
++CurCycle;
}
}
#ifndef NDEBUG
VerifyScheduledSequence(/*isBottomUp=*/false);
#endif
}
//===----------------------------------------------------------------------===//
// Public Constructor Functions
//===----------------------------------------------------------------------===//
/// createVLIWDAGScheduler - This creates a top-down list scheduler.
ScheduleDAGSDNodes *
llvm::createVLIWDAGScheduler(SelectionDAGISel *IS, CodeGenOpt::Level) {
return new ScheduleDAGVLIW(*IS->MF, IS->AA, new ResourcePriorityQueue(IS));
}Thank you in advance.Regards,Shigio2015-09-25 9:37 GMT+09:00 rail shafigulin <address@hidden>:On Thu, Sep 24, 2015 at 4:08 PM, Shigio YAMAGUCHI <address@hidden> wrote:Hello,> I'm navigating through https://github.com/openrisc/llvm-or1k...> Fist and third results are incorrect. They are declarations of these functions and not the callers.Could you show the source code in which the problem has occurred?This mailing list is recorded in the archive. But we won't know whetheror not the external site exists tomorrow.Cone the repo from https://github.com/openrisc/llvm-or1k using gitcd to llvm-or1k directoryexecute ctags in this direcotryopen vimexecute the follwoing commands:cs add CTAGS:cs f c releaseSucc:copenYou will see that some of the results are not functions that call a given function, they are function declarations.Thanks> Output also doesn't seem to show what method calls a searched method.> Note: <<releaseSucc>> releaseSucc(SU, *I); It would be nice to have something like> <<ScheduleDAGVLIW::releaseSuccessors(SUnit *SU)>> releaseSucc(SU, *I)That's the present specification.In cscope, ':cs f c' is 'Find functions calling this function:'.But in gtags-cscope, it is 'Find locations calling this function:'.Regards,Shigio2015-09-25 3:38 GMT+09:00 rail shafigulin <address@hidden>:_______________________________________________The output looked like the following::cs f c releaseSuccI've tried to do a caller search for releaseSucc using the following commandjust like it is stated in the manual.:set cscopeprg=gtags-cscopeand set my cscope program to gtags-cscope usingI've connected the GTAGS database to my Vim using:cs add GTAGSllvm-or1k/include/llvm/CodeGen/MachineScheduler.h|335| <<releaseSucc>> void releaseSucc(SUnit *SU, SDep *SuccEdge);
llvm-or1k/lib/CodeGen/MachineScheduler.cpp|539| <<releaseSucc>> releaseSucc(SU, &*I);
llvm-or1k/lib/CodeGen/SelectionDAG/ScheduleDAGVLIW.cpp|86| <<releaseSucc>> void releaseSucc(SUnit *SU, const SDep &D);
llvm-or1k/lib/CodeGen/SelectionDAG/ScheduleDAGVLIW.cpp|146| <<releaseSucc>> releaseSucc(SU, *I);Fist and third results are incorrect. They are declarations of these functions and not the callers.Output also doesn't seem to show what method calls a searched method. Note: <<releaseSucc>> releaseSucc(SU, *I); It would be nice to have something like <<ScheduleDAGVLIW::releaseSuccessors(SUnit *SU)>> releaseSucc(SU, *I)As an additional note:It would be nice to give the global tool awareness what function callers are being search for. The result above demonstrates that search found two methods named releaseSucc, one from ScheduleDAGMI class (2nd result) and one from ScheduleDAGVLIW class (4th result). If a user could pass information to the global tool stating that the user is looking for callers of ScheduleDAGVLIW::rereleaseSucc(...) instead of a caller of <any class>::releaseSucc(...), I believe this would be a very good improvement of the tool.
Bug-global mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/bug-global
--Shigio YAMAGUCHI <address@hidden>PGP fingerprint: D1CB 0B89 B346 4AB6 5663 C4B6 3CA5 BBB3 57BE DDA3--Shigio YAMAGUCHI <address@hidden>PGP fingerprint: D1CB 0B89 B346 4AB6 5663 C4B6 3CA5 BBB3 57BE DDA3
| [Prev in Thread] | Current Thread | [Next in Thread] |