llvm-6502/lib/CodeGen/SelectionDAG/ScheduleDAGList.cpp
Evan Cheng 86ec7d1d07 - Fixed some priority calculation bugs that were causing bug 478. Among them:
a predecessor appearing more than once in the operand list was counted as
  multiple predecessor; priority1 should be updated during scheduling;
  CycleBound was updated after the node is inserted into priority queue; one
  of the tie breaking condition was flipped.
- Take into consideration of two address opcodes. If a predecessor is a def&use
  operand, it should have a higher priority.
- Scheduler should also favor floaters, i.e. nodes that do not have real
  predecessors such as MOV32ri.
- The scheduling fixes / tweaks fixed bug 478:
        .text
        .align  4
        .globl  _f
_f:
        movl 4(%esp), %eax
        movl 8(%esp), %ecx
        movl %eax, %edx
        imull %ecx, %edx
        imull %eax, %eax
        imull %ecx, %ecx
        addl %eax, %ecx
        leal (%ecx,%edx,2), %eax
        ret

  It is also a slight performance win (1% - 3%) for most tests.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26470 91177308-0d34-0410-b5e6-96231b3b80d8
2006-03-02 21:38:29 +00:00

502 lines
16 KiB
C++

//===---- ScheduleDAGList.cpp - Implement a list scheduler for isel DAG ---===//
//
// The LLVM Compiler Infrastructure
//
// This file was developed by Evan Cheng and is distributed under the
// University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This implements a simple two pass scheduler. The first pass attempts to push
// backward any lengthy instructions and critical paths. The second pass packs
// instructions into semi-optimal time slots.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "sched"
#include "llvm/CodeGen/ScheduleDAG.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Support/Debug.h"
#include <climits>
#include <iostream>
#include <queue>
#include <set>
#include <vector>
using namespace llvm;
namespace {
/// SUnit - Scheduling unit. It's an wrapper around either a single SDNode or a
/// group of nodes flagged together.
struct SUnit {
SDNode *Node; // Representative node.
std::vector<SDNode*> FlaggedNodes; // All nodes flagged to Node.
std::set<SUnit*> Preds; // All real predecessors.
std::set<SUnit*> ChainPreds; // All chain predecessors.
std::set<SUnit*> Succs; // All real successors.
std::set<SUnit*> ChainSuccs; // All chain successors.
int NumPredsLeft; // # of preds not scheduled.
int NumSuccsLeft; // # of succs not scheduled.
int NumChainPredsLeft; // # of chain preds not scheduled.
int NumChainSuccsLeft; // # of chain succs not scheduled.
int Priority1; // Scheduling priority 1.
int Priority2; // Scheduling priority 2.
bool isDefNUseOperand; // Is a def&use operand.
unsigned Latency; // Node latency.
unsigned CycleBound; // Upper/lower cycle to be scheduled at.
unsigned Slot; // Cycle node is scheduled at.
SUnit *Next;
SUnit(SDNode *node)
: Node(node), NumPredsLeft(0), NumSuccsLeft(0),
NumChainPredsLeft(0), NumChainSuccsLeft(0),
Priority1(INT_MIN), Priority2(INT_MIN), isDefNUseOperand(false),
Latency(0), CycleBound(0), Slot(0), Next(NULL) {}
void dump(const SelectionDAG *G, bool All=true) const;
};
void SUnit::dump(const SelectionDAG *G, bool All) const {
std::cerr << "SU: ";
Node->dump(G);
std::cerr << "\n";
if (FlaggedNodes.size() != 0) {
for (unsigned i = 0, e = FlaggedNodes.size(); i != e; i++) {
std::cerr << " ";
FlaggedNodes[i]->dump(G);
std::cerr << "\n";
}
}
if (All) {
std::cerr << "# preds left : " << NumPredsLeft << "\n";
std::cerr << "# succs left : " << NumSuccsLeft << "\n";
std::cerr << "# chain preds left : " << NumChainPredsLeft << "\n";
std::cerr << "# chain succs left : " << NumChainSuccsLeft << "\n";
std::cerr << "Latency : " << Latency << "\n";
std::cerr << "Priority : " << Priority1 << " , " << Priority2 << "\n";
if (Preds.size() != 0) {
std::cerr << "Predecessors :\n";
for (std::set<SUnit*>::iterator I = Preds.begin(),
E = Preds.end(); I != E; ++I) {
std::cerr << " ";
(*I)->dump(G, false);
}
}
if (ChainPreds.size() != 0) {
std::cerr << "Chained Preds :\n";
for (std::set<SUnit*>::iterator I = ChainPreds.begin(),
E = ChainPreds.end(); I != E; ++I) {
std::cerr << " ";
(*I)->dump(G, false);
}
}
if (Succs.size() != 0) {
std::cerr << "Successors :\n";
for (std::set<SUnit*>::iterator I = Succs.begin(),
E = Succs.end(); I != E; ++I) {
std::cerr << " ";
(*I)->dump(G, false);
}
}
if (ChainSuccs.size() != 0) {
std::cerr << "Chained succs :\n";
for (std::set<SUnit*>::iterator I = ChainSuccs.begin(),
E = ChainSuccs.end(); I != E; ++I) {
std::cerr << " ";
(*I)->dump(G, false);
}
}
}
}
/// Sorting functions for the Available queue.
struct ls_rr_sort : public std::binary_function<SUnit*, SUnit*, bool> {
bool operator()(const SUnit* left, const SUnit* right) const {
bool LFloater = (left ->Preds.size() == 0);
bool RFloater = (right->Preds.size() == 0);
int LBonus = (int)left ->isDefNUseOperand;
int RBonus = (int)right->isDefNUseOperand;
int LPriority1 = left ->Priority1 - LBonus;
int RPriority1 = right->Priority1 - RBonus;
int LPriority2 = left ->Priority2 + LBonus;
int RPriority2 = right->Priority2 + RBonus;
// Favor floaters (i.e. node with no non-passive predecessors):
// e.g. MOV32ri.
if (!LFloater && RFloater)
return true;
else if (LFloater == RFloater)
if (LPriority1 > RPriority1)
return true;
else if (LPriority1 == RPriority1)
if (LPriority2 < RPriority2)
return true;
else if (LPriority1 == RPriority1)
if (left->CycleBound > right->CycleBound)
return true;
else
return left->Node->getNodeDepth() < right->Node->getNodeDepth();
return false;
}
};
/// ScheduleDAGList - List scheduler.
class ScheduleDAGList : public ScheduleDAG {
private:
// SDNode to SUnit mapping (many to one).
std::map<SDNode*, SUnit*> SUnitMap;
// Available queue.
std::priority_queue<SUnit*, std::vector<SUnit*>, ls_rr_sort> Available;
// The schedule.
std::vector<SUnit*> Sequence;
// Current scheduling cycle.
unsigned CurrCycle;
// First and last SUnit created.
SUnit *HeadSUnit, *TailSUnit;
public:
ScheduleDAGList(SelectionDAG &dag, MachineBasicBlock *bb,
const TargetMachine &tm)
: ScheduleDAG(listSchedulingBURR, dag, bb, tm),
CurrCycle(0), HeadSUnit(NULL), TailSUnit(NULL) {};
~ScheduleDAGList() {
SUnit *SU = HeadSUnit;
while (SU) {
SUnit *NextSU = SU->Next;
delete SU;
SU = NextSU;
}
}
void Schedule();
void dump() const;
private:
SUnit *NewSUnit(SDNode *N);
void ReleasePred(SUnit *PredSU, bool isChain = false);
void ScheduleNode(SUnit *SU);
int CalcNodePriority(SUnit *SU);
void CalculatePriorities();
void ListSchedule();
void BuildSchedUnits();
void EmitSchedule();
};
} // end namespace
/// NewSUnit - Creates a new SUnit and return a ptr to it.
SUnit *ScheduleDAGList::NewSUnit(SDNode *N) {
SUnit *CurrSUnit = new SUnit(N);
if (HeadSUnit == NULL)
HeadSUnit = CurrSUnit;
if (TailSUnit != NULL)
TailSUnit->Next = CurrSUnit;
TailSUnit = CurrSUnit;
return CurrSUnit;
}
/// ReleasePred - Decrement the NumSuccsLeft count of a predecessor. Add it to
/// the Available queue is the count reaches zero. Also update its cycle bound.
void ScheduleDAGList::ReleasePred(SUnit *PredSU, bool isChain) {
SDNode *PredNode = PredSU->Node;
// FIXME: the distance between two nodes is not always == the predecessor's
// latency. For example, the reader can very well read the register written
// by the predecessor later than the issue cycle. It also depends on the
// interrupt model (drain vs. freeze).
PredSU->CycleBound = std::max(PredSU->CycleBound, CurrCycle + PredSU->Latency);
if (!isChain) {
PredSU->NumSuccsLeft--;
PredSU->Priority1++;
} else
PredSU->NumChainSuccsLeft--;
if (PredSU->NumSuccsLeft == 0 && PredSU->NumChainSuccsLeft == 0) {
// EntryToken has to go last!
if (PredNode->getOpcode() != ISD::EntryToken)
Available.push(PredSU);
} else if (PredSU->NumSuccsLeft < 0) {
#ifndef NDEBUG
std::cerr << "*** List scheduling failed! ***\n";
PredSU->dump(&DAG);
std::cerr << " has been released too many times!\n";
assert(0);
#endif
}
}
/// ScheduleNode - Add the node to the schedule. Decrement the pending count of
/// its predecessors. If a predecessor pending count is zero, add it to the
/// Available queue.
void ScheduleDAGList::ScheduleNode(SUnit *SU) {
Sequence.push_back(SU);
SU->Slot = CurrCycle;
// Bottom up: release predecessors
for (std::set<SUnit*>::iterator I1 = SU->Preds.begin(),
E1 = SU->Preds.end(); I1 != E1; ++I1) {
ReleasePred(*I1);
SU->NumPredsLeft--;
SU->Priority1--;
}
for (std::set<SUnit*>::iterator I2 = SU->ChainPreds.begin(),
E2 = SU->ChainPreds.end(); I2 != E2; ++I2)
ReleasePred(*I2, true);
CurrCycle++;
}
/// isReady - True if node's lower cycle bound is less or equal to the current
/// scheduling cycle. Always true if all nodes have uniform latency 1.
static inline bool isReady(SUnit *SU, unsigned CurrCycle) {
return SU->CycleBound <= CurrCycle;
}
/// ListSchedule - The main loop of list scheduling.
void ScheduleDAGList::ListSchedule() {
// Add root to Available queue
SUnit *Root = SUnitMap[DAG.getRoot().Val];
Available.push(Root);
// While Available queue 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;
while (!Available.empty()) {
SUnit *CurrNode = Available.top();
Available.pop();
NotReady.clear();
while (!isReady(CurrNode, CurrCycle)) {
NotReady.push_back(CurrNode);
CurrNode = Available.top();
Available.pop();
}
for (unsigned i = 0, e = NotReady.size(); i != e; ++i)
Available.push(NotReady[i]);
DEBUG(std::cerr << "*** Scheduling: ");
DEBUG(CurrNode->dump(&DAG, false));
ScheduleNode(CurrNode);
}
// Add entry node last
if (DAG.getEntryNode().Val != DAG.getRoot().Val) {
SUnit *Entry = SUnitMap[DAG.getEntryNode().Val];
Entry->Slot = CurrCycle;
Sequence.push_back(Entry);
}
#ifndef NDEBUG
bool AnyNotSched = false;
for (SUnit *SU = HeadSUnit; SU != NULL; SU = SU->Next) {
if (SU->NumSuccsLeft != 0 || SU->NumChainSuccsLeft != 0) {
if (!AnyNotSched)
std::cerr << "*** List scheduling failed! ***\n";
SU->dump(&DAG);
std::cerr << "has not been scheduled!\n";
AnyNotSched = true;
}
}
assert(!AnyNotSched);
#endif
// Reverse the order if it is bottom up.
std::reverse(Sequence.begin(), Sequence.end());
DEBUG(std::cerr << "*** Final schedule ***\n");
DEBUG(dump());
DEBUG(std::cerr << "\n");
}
/// CalcNodePriority - Priority1 is just the number of live range genned -
/// number of live range killed. Priority2 is the Sethi Ullman number. It
/// returns Priority2 since it is calculated recursively.
/// Smaller number is the higher priority for Priority2. Reverse is true for
/// Priority1.
int ScheduleDAGList::CalcNodePriority(SUnit *SU) {
if (SU->Priority2 != INT_MIN)
return SU->Priority2;
SU->Priority1 = SU->NumPredsLeft - SU->NumSuccsLeft;
if (SU->Preds.size() == 0) {
SU->Priority2 = 1;
} else {
int Extra = 0;
for (std::set<SUnit*>::iterator I = SU->Preds.begin(),
E = SU->Preds.end(); I != E; ++I) {
SUnit *PredSU = *I;
int PredPriority = CalcNodePriority(PredSU);
if (PredPriority > SU->Priority2) {
SU->Priority2 = PredPriority;
Extra = 0;
} else if (PredPriority == SU->Priority2)
Extra++;
}
if (SU->Node->getOpcode() != ISD::TokenFactor)
SU->Priority2 += Extra;
else
SU->Priority2 = (Extra == 1) ? 0 : Extra-1;
}
return SU->Priority2;
}
/// CalculatePriorities - Calculate priorities of all scheduling units.
void ScheduleDAGList::CalculatePriorities() {
for (SUnit *SU = HeadSUnit; SU != NULL; SU = SU->Next) {
// FIXME: assumes uniform latency for now.
SU->Latency = 1;
(void)CalcNodePriority(SU);
DEBUG(SU->dump(&DAG));
DEBUG(std::cerr << "\n");
}
}
void ScheduleDAGList::BuildSchedUnits() {
// Pass 1: create the SUnit's.
for (unsigned i = 0, NC = NodeCount; i < NC; i++) {
NodeInfo *NI = &Info[i];
SDNode *N = NI->Node;
if (isPassiveNode(N))
continue;
SUnit *SU;
if (NI->isInGroup()) {
if (NI != NI->Group->getBottom()) // Bottom up, so only look at bottom
continue; // node of the NodeGroup
SU = NewSUnit(N);
// Find the flagged nodes.
SDOperand FlagOp = N->getOperand(N->getNumOperands() - 1);
SDNode *Flag = FlagOp.Val;
unsigned ResNo = FlagOp.ResNo;
while (Flag->getValueType(ResNo) == MVT::Flag) {
NodeInfo *FNI = getNI(Flag);
assert(FNI->Group == NI->Group);
SU->FlaggedNodes.insert(SU->FlaggedNodes.begin(), Flag);
SUnitMap[Flag] = SU;
FlagOp = Flag->getOperand(Flag->getNumOperands() - 1);
Flag = FlagOp.Val;
ResNo = FlagOp.ResNo;
}
} else {
SU = NewSUnit(N);
}
SUnitMap[N] = SU;
}
// Pass 2: add the preds, succs, etc.
for (SUnit *SU = HeadSUnit; SU != NULL; SU = SU->Next) {
SDNode *N = SU->Node;
NodeInfo *NI = getNI(N);
if (NI->isInGroup()) {
// Find all predecessors (of the group).
NodeGroupOpIterator NGOI(NI);
while (!NGOI.isEnd()) {
SDOperand Op = NGOI.next();
SDNode *OpN = Op.Val;
MVT::ValueType VT = OpN->getValueType(Op.ResNo);
NodeInfo *OpNI = getNI(OpN);
if (OpNI->Group != NI->Group && !isPassiveNode(OpN)) {
assert(VT != MVT::Flag);
SUnit *OpSU = SUnitMap[OpN];
if (VT == MVT::Other) {
if (SU->ChainPreds.insert(OpSU).second)
SU->NumChainPredsLeft++;
if (OpSU->ChainSuccs.insert(SU).second)
OpSU->NumChainSuccsLeft++;
} else {
if (SU->Preds.insert(OpSU).second)
SU->NumPredsLeft++;
if (OpSU->Succs.insert(SU).second)
OpSU->NumSuccsLeft++;
}
}
}
} else {
// Find node predecessors.
for (unsigned j = 0, e = N->getNumOperands(); j != e; j++) {
SDOperand Op = N->getOperand(j);
SDNode *OpN = Op.Val;
MVT::ValueType VT = OpN->getValueType(Op.ResNo);
if (!isPassiveNode(OpN)) {
assert(VT != MVT::Flag);
SUnit *OpSU = SUnitMap[OpN];
if (VT == MVT::Other) {
if (SU->ChainPreds.insert(OpSU).second)
SU->NumChainPredsLeft++;
if (OpSU->ChainSuccs.insert(SU).second)
OpSU->NumChainSuccsLeft++;
} else {
if (SU->Preds.insert(OpSU).second)
SU->NumPredsLeft++;
if (OpSU->Succs.insert(SU).second)
OpSU->NumSuccsLeft++;
if (j == 0 && TII->isTwoAddrInstr(N->getTargetOpcode()))
OpSU->isDefNUseOperand = true;
}
}
}
}
}
}
/// EmitSchedule - Emit the machine code in scheduled order.
void ScheduleDAGList::EmitSchedule() {
for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
SDNode *N;
SUnit *SU = Sequence[i];
for (unsigned j = 0, ee = SU->FlaggedNodes.size(); j != ee; j++) {
N = SU->FlaggedNodes[j];
EmitNode(getNI(N));
}
EmitNode(getNI(SU->Node));
}
}
/// dump - dump the schedule.
void ScheduleDAGList::dump() const {
for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
SUnit *SU = Sequence[i];
SU->dump(&DAG, false);
}
}
/// Schedule - Schedule the DAG using list scheduling.
/// FIXME: Right now it only supports the burr (bottom up register reducing)
/// heuristic.
void ScheduleDAGList::Schedule() {
DEBUG(std::cerr << "********** List Scheduling **********\n");
// Build scheduling units.
BuildSchedUnits();
// Calculate node prirorities.
CalculatePriorities();
// Execute the actual scheduling loop.
ListSchedule();
// Emit in scheduled order
EmitSchedule();
}
llvm::ScheduleDAG* llvm::createBURRListDAGScheduler(SelectionDAG &DAG,
MachineBasicBlock *BB) {
return new ScheduleDAGList(DAG, BB, DAG.getTarget());
}