1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-19 02:52:53 +02:00

Teach pre-regalloc scheduler to schedule loads from nearby addresses. It may improve cache locality. This is controlled by -cluster-loads for now.

llvm-svn: 94148
This commit is contained in:
Evan Cheng 2010-01-22 03:36:51 +00:00
parent 8e25244a48
commit 72dbbce547
2 changed files with 134 additions and 0 deletions

View File

@ -20,10 +20,21 @@
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetSubtarget.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
STATISTIC(LoadsClustered, "Number of loads clustered together");
static cl::opt<bool>
ClusterLoads("cluster-loads", cl::Hidden,
cl::desc("Schedule nearby loads together and in order"));
ScheduleDAGSDNodes::ScheduleDAGSDNodes(MachineFunction &mf)
: ScheduleDAG(mf) {
}
@ -75,6 +86,122 @@ static void CheckForPhysRegDependency(SDNode *Def, SDNode *User, unsigned Op,
}
}
static void AddFlags(SDNode *N, SDValue Flag, bool AddFlag,
SelectionDAG *DAG) {
SmallVector<EVT, 4> VTs;
for (unsigned i = 0, e = N->getNumValues(); i != e; ++i)
VTs.push_back(N->getValueType(i));
if (AddFlag)
VTs.push_back(MVT::Flag);
SmallVector<SDValue, 4> Ops;
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
Ops.push_back(N->getOperand(i));
if (Flag.getNode())
Ops.push_back(Flag);
SDVTList VTList = DAG->getVTList(&VTs[0], VTs.size());
DAG->MorphNodeTo(N, N->getOpcode(), VTList, &Ops[0], Ops.size());
}
/// ClusterNeighboringLoads - Force nearby loads together by "flagging" them.
/// This function finds loads of the same base and different offsets. If the
/// offsets are not far apart (target specific), it add MVT::Flag inputs and
/// outputs to ensure they are scheduled together and in order. This
/// optimization may benefit some targets by improving cache locality.
void ScheduleDAGSDNodes::ClusterNeighboringLoads() {
SmallPtrSet<SDNode*, 16> Visited;
SmallVector<int64_t, 4> Offsets;
DenseMap<long long, SDNode*> O2SMap; // Map from offset to SDNode.
for (SelectionDAG::allnodes_iterator NI = DAG->allnodes_begin(),
E = DAG->allnodes_end(); NI != E; ++NI) {
SDNode *Node = &*NI;
if (!Node || !Node->isMachineOpcode())
continue;
unsigned Opc = Node->getMachineOpcode();
const TargetInstrDesc &TID = TII->get(Opc);
if (!TID.mayLoad())
continue;
SDNode *Chain = 0;
unsigned NumOps = Node->getNumOperands();
if (Node->getOperand(NumOps-1).getValueType() == MVT::Other)
Chain = Node->getOperand(NumOps-1).getNode();
if (!Chain)
continue;
// Look for other loads of the same chain. Find loads that are loading from
// the same base pointer and different offsets.
Visited.clear();
Offsets.clear();
O2SMap.clear();
bool Cluster = false;
SDNode *Base = Node;
int64_t BaseOffset;
for (SDNode::use_iterator I = Chain->use_begin(), E = Chain->use_end();
I != E; ++I) {
SDNode *User = *I;
if (User == Node || !Visited.insert(User))
continue;
int64_t Offset1, Offset2;
if (!TII->areLoadsFromSameBasePtr(Base, User, Offset1, Offset2) ||
Offset1 == Offset2)
// FIXME: Should be ok if they addresses are identical. But earlier
// optimizations really should have eliminated one of the loads.
continue;
if (O2SMap.insert(std::make_pair(Offset1, Base)).second)
Offsets.push_back(Offset1);
O2SMap.insert(std::make_pair(Offset2, User));
Offsets.push_back(Offset2);
if (Offset2 < Offset1) {
Base = User;
BaseOffset = Offset2;
} else {
BaseOffset = Offset1;
}
Cluster = true;
}
if (!Cluster)
continue;
// Sort them in increasing order.
std::sort(Offsets.begin(), Offsets.end());
// Check if the loads are close enough.
SmallVector<SDNode*, 4> Loads;
unsigned NumLoads = 0;
int64_t BaseOff = Offsets[0];
SDNode *BaseLoad = O2SMap[BaseOff];
Loads.push_back(BaseLoad);
for (unsigned i = 1, e = Offsets.size(); i != e; ++i) {
int64_t Offset = Offsets[i];
SDNode *Load = O2SMap[Offset];
if (!TII->shouldScheduleLoadsNear(BaseLoad, Load, BaseOff, Offset,
NumLoads))
break; // Stop right here. Ignore loads that are further away.
Loads.push_back(Load);
++NumLoads;
}
if (NumLoads == 0)
continue;
// Cluster loads by adding MVT::Flag outputs and inputs. This also
// ensure they are scheduled in order of increasing addresses.
SDNode *Lead = Loads[0];
AddFlags(Lead, SDValue(0,0), true, DAG);
SDValue InFlag = SDValue(Lead, Lead->getNumValues()-1);
for (unsigned i = 1, e = Loads.size(); i != e; ++i) {
bool OutFlag = i < e-1;
SDNode *Load = Loads[i];
AddFlags(Load, InFlag, OutFlag, DAG);
if (OutFlag)
InFlag = SDValue(Load, Load->getNumValues()-1);
++LoadsClustered;
}
}
}
void ScheduleDAGSDNodes::BuildSchedUnits() {
// During scheduling, the NodeId field of SDNode is used to map SDNodes
// to their associated SUnits by holding SUnits table indices. A value
@ -232,6 +359,9 @@ void ScheduleDAGSDNodes::AddSchedEdges() {
/// excludes nodes that aren't interesting to scheduling, and represents
/// flagged together nodes with a single SUnit.
void ScheduleDAGSDNodes::BuildSchedGraph(AliasAnalysis *AA) {
// Cluster loads from "near" addresses into combined SUnits.
if (ClusterLoads)
ClusterNeighboringLoads();
// Populate the SUnits array.
BuildSchedUnits();
// Compute all the scheduling dependencies between nodes.

View File

@ -108,6 +108,10 @@ namespace llvm {
virtual void getCustomGraphFeatures(GraphWriter<ScheduleDAG*> &GW) const;
private:
/// ClusterNeighboringLoads - Cluster loads from "near" addresses into
/// combined SUnits.
void ClusterNeighboringLoads();
/// BuildSchedUnits, AddSchedEdges - Helper functions for BuildSchedGraph.
void BuildSchedUnits();
void AddSchedEdges();