//===-- MSchedGraph.cpp - Scheduling Graph ----------------------*- C++ -*-===// // // The LLVM Compiler Infrastructure // // This file was developed by the LLVM research group and is distributed under // the University of Illinois Open Source License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // A graph class for dependencies. This graph only contains true, anti, and // output data dependencies for a given MachineBasicBlock. Dependencies // across iterations are also computed. Unless data dependence analysis // is provided, a conservative approach of adding dependencies between all // loads and stores is taken. //===----------------------------------------------------------------------===// #define DEBUG_TYPE "ModuloSched" #include "MSchedGraph.h" #include "../SparcV9RegisterInfo.h" #include "../MachineCodeForInstruction.h" #include "llvm/BasicBlock.h" #include "llvm/Constants.h" #include "llvm/Instructions.h" #include "llvm/Type.h" #include "llvm/CodeGen/MachineBasicBlock.h" #include "llvm/Target/TargetInstrInfo.h" #include "llvm/Support/Debug.h" #include #include #include #include using namespace llvm; //MSchedGraphNode constructor MSchedGraphNode::MSchedGraphNode(const MachineInstr* inst, MSchedGraph *graph, unsigned idx, unsigned late, bool isBranch) : Inst(inst), Parent(graph), index(idx), latency(late), isBranchInstr(isBranch) { //Add to the graph graph->addNode(inst, this); } //MSchedGraphNode copy constructor MSchedGraphNode::MSchedGraphNode(const MSchedGraphNode &N) : Predecessors(N.Predecessors), Successors(N.Successors) { Inst = N.Inst; Parent = N.Parent; index = N.index; latency = N.latency; isBranchInstr = N.isBranchInstr; } //Print the node (instruction and latency) void MSchedGraphNode::print(std::ostream &os) const { os << "MSchedGraphNode: Inst=" << *Inst << ", latency= " << latency << "\n"; } //Get the edge from a predecessor to this node MSchedGraphEdge MSchedGraphNode::getInEdge(MSchedGraphNode *pred) { //Loop over all the successors of our predecessor //return the edge the corresponds to this in edge for (MSchedGraphNode::succ_iterator I = pred->succ_begin(), E = pred->succ_end(); I != E; ++I) { if (*I == this) return I.getEdge(); } assert(0 && "Should have found edge between this node and its predecessor!"); abort(); } //Get the iteration difference for the edge from this node to its successor unsigned MSchedGraphNode::getIteDiff(MSchedGraphNode *succ) { for(std::vector::iterator I = Successors.begin(), E = Successors.end(); I != E; ++I) { if(I->getDest() == succ) return I->getIteDiff(); } return 0; } //Get the index into the vector of edges for the edge from pred to this node unsigned MSchedGraphNode::getInEdgeNum(MSchedGraphNode *pred) { //Loop over all the successors of our predecessor //return the edge the corresponds to this in edge int count = 0; for(MSchedGraphNode::succ_iterator I = pred->succ_begin(), E = pred->succ_end(); I != E; ++I) { if(*I == this) return count; count++; } assert(0 && "Should have found edge between this node and its predecessor!"); abort(); } //Determine if succ is a successor of this node bool MSchedGraphNode::isSuccessor(MSchedGraphNode *succ) { for(succ_iterator I = succ_begin(), E = succ_end(); I != E; ++I) if(*I == succ) return true; return false; } //Dtermine if pred is a predecessor of this node bool MSchedGraphNode::isPredecessor(MSchedGraphNode *pred) { if(std::find( Predecessors.begin(), Predecessors.end(), pred) != Predecessors.end()) return true; else return false; } //Add a node to the graph void MSchedGraph::addNode(const MachineInstr *MI, MSchedGraphNode *node) { //Make sure node does not already exist assert(GraphMap.find(MI) == GraphMap.end() && "New MSchedGraphNode already exists for this instruction"); GraphMap[MI] = node; } //Delete a node to the graph void MSchedGraph::deleteNode(MSchedGraphNode *node) { //Delete the edge to this node from all predecessors while(node->pred_size() > 0) { //DEBUG(std::cerr << "Delete edge from: " << **P << " to " << *node << "\n"); MSchedGraphNode *pred = *(node->pred_begin()); pred->deleteSuccessor(node); } //Remove this node from the graph GraphMap.erase(node->getInst()); } //Create a graph for a machine block. The ignoreInstrs map is so that //we ignore instructions associated to the index variable since this //is a special case in Modulo Scheduling. We only want to deal with //the body of the loop. MSchedGraph::MSchedGraph(const MachineBasicBlock *bb, const TargetMachine &targ, std::map &ignoreInstrs, DependenceAnalyzer &DA, std::map &machineTollvm) : Target(targ) { //Make sure BB is not null, assert(bb != NULL && "Basic Block is null"); BBs.push_back(bb); //Create nodes and edges for this BB buildNodesAndEdges(ignoreInstrs, DA, machineTollvm); //Experimental! //addBranchEdges(); } //Create a graph for a machine block. The ignoreInstrs map is so that //we ignore instructions associated to the index variable since this //is a special case in Modulo Scheduling. We only want to deal with //the body of the loop. MSchedGraph::MSchedGraph(std::vector &bbs, const TargetMachine &targ, std::map &ignoreInstrs, DependenceAnalyzer &DA, std::map &machineTollvm) : BBs(bbs), Target(targ) { //Make sure there is at least one BB and it is not null, assert(((bbs.size() >= 1) && bbs[1] != NULL) && "Basic Block is null"); //Create nodes and edges for this BB buildNodesAndEdges(ignoreInstrs, DA, machineTollvm); //Experimental! //addBranchEdges(); } //Copies the graph and keeps a map from old to new nodes MSchedGraph::MSchedGraph(const MSchedGraph &G, std::map &newNodes) : Target(G.Target) { BBs = G.BBs; std::map oldToNew; //Copy all nodes for(MSchedGraph::const_iterator N = G.GraphMap.begin(), NE = G.GraphMap.end(); N != NE; ++N) { MSchedGraphNode *newNode = new MSchedGraphNode(*(N->second)); oldToNew[&*(N->second)] = newNode; newNodes[newNode] = &*(N->second); GraphMap[&*(N->first)] = newNode; } //Loop over nodes and update edges to point to new nodes for(MSchedGraph::iterator N = GraphMap.begin(), NE = GraphMap.end(); N != NE; ++N) { //Get the node we are dealing with MSchedGraphNode *node = &*(N->second); node->setParent(this); //Loop over nodes successors and predecessors and update to the new nodes for(unsigned i = 0; i < node->pred_size(); ++i) { node->setPredecessor(i, oldToNew[node->getPredecessor(i)]); } for(unsigned i = 0; i < node->succ_size(); ++i) { MSchedGraphEdge *edge = node->getSuccessor(i); MSchedGraphNode *oldDest = edge->getDest(); edge->setDest(oldToNew[oldDest]); } } } //Deconstructor, deletes all nodes in the graph MSchedGraph::~MSchedGraph () { for(MSchedGraph::iterator I = GraphMap.begin(), E = GraphMap.end(); I != E; ++I) delete I->second; } //Print out graph void MSchedGraph::print(std::ostream &os) const { for(MSchedGraph::const_iterator N = GraphMap.begin(), NE = GraphMap.end(); N != NE; ++N) { //Get the node we are dealing with MSchedGraphNode *node = &*(N->second); os << "Node Start\n"; node->print(os); os << "Successors:\n"; //print successors for(unsigned i = 0; i < node->succ_size(); ++i) { MSchedGraphEdge *edge = node->getSuccessor(i); MSchedGraphNode *oldDest = edge->getDest(); oldDest->print(os); } os << "Node End\n"; } } //Calculate total delay int MSchedGraph::totalDelay() { int sum = 0; for(MSchedGraph::const_iterator N = GraphMap.begin(), NE = GraphMap.end(); N != NE; ++N) { //Get the node we are dealing with MSchedGraphNode *node = &*(N->second); sum += node->getLatency(); } return sum; } //Experimental code to add edges from the branch to all nodes dependent upon it. void hasPath(MSchedGraphNode *node, std::set &visited, std::set &branches, MSchedGraphNode *startNode, std::set > &newEdges ) { visited.insert(node); DEBUG(std::cerr << "Visiting: " << *node << "\n"); //Loop over successors for(unsigned i = 0; i < node->succ_size(); ++i) { MSchedGraphEdge *edge = node->getSuccessor(i); MSchedGraphNode *dest = edge->getDest(); if(branches.count(dest)) newEdges.insert(std::make_pair(dest, startNode)); //only visit if we have not already else if(!visited.count(dest)) { if(edge->getIteDiff() == 0) hasPath(dest, visited, branches, startNode, newEdges);} } } //Experimental code to add edges from the branch to all nodes dependent upon it. void MSchedGraph::addBranchEdges() { std::set branches; std::set nodes; for(MSchedGraph::iterator I = GraphMap.begin(), E = GraphMap.end(); I != E; ++I) { if(I->second->isBranch()) if(I->second->hasPredecessors()) branches.insert(I->second); } //See if there is a path first instruction to the branches, if so, add an //iteration dependence between that node and the branch std::set > newEdges; for(MSchedGraph::iterator I = GraphMap.begin(), E = GraphMap.end(); I != E; ++I) { std::set visited; hasPath((I->second), visited, branches, (I->second), newEdges); } //Spit out all edges we are going to add unsigned min = GraphMap.size(); if(newEdges.size() == 1) { ((newEdges.begin())->first)->addOutEdge(((newEdges.begin())->second), MSchedGraphEdge::BranchDep, MSchedGraphEdge::NonDataDep, 1); } else { unsigned count = 0; MSchedGraphNode *start; MSchedGraphNode *end; for(std::set >::iterator I = newEdges.begin(), E = newEdges.end(); I != E; ++I) { DEBUG(std::cerr << "Branch Edge from: " << *(I->first) << " to " << *(I->second) << "\n"); // if(I->second->getIndex() <= min) { start = I->first; end = I->second; //min = I->second->getIndex(); //} start->addOutEdge(end, MSchedGraphEdge::BranchDep, MSchedGraphEdge::NonDataDep, 1); } } } //Add edges between the nodes void MSchedGraph::buildNodesAndEdges(std::map &ignoreInstrs, DependenceAnalyzer &DA, std::map &machineTollvm) { //Get Machine target information for calculating latency const TargetInstrInfo *MTI = Target.getInstrInfo(); std::vector memInstructions; std::map > regNumtoNodeMap; std::map > valuetoNodeMap; //Save PHI instructions to deal with later std::vector phiInstrs; unsigned index = 0; for(std::vector::iterator B = BBs.begin(), BE = BBs.end(); B != BE; ++B) { const MachineBasicBlock *BB = *B; //Loop over instructions in MBB and add nodes and edges for (MachineBasicBlock::const_iterator MI = BB->begin(), e = BB->end(); MI != e; ++MI) { //Ignore indvar instructions if(ignoreInstrs.count(MI)) { ++index; continue; } //Get each instruction of machine basic block, get the delay //using the op code, create a new node for it, and add to the //graph. MachineOpCode opCode = MI->getOpcode(); int delay; #if 0 // FIXME: LOOK INTO THIS //Check if subsequent instructions can be issued before //the result is ready, if so use min delay. if(MTI->hasResultInterlock(MIopCode)) delay = MTI->minLatency(MIopCode); else #endif //Get delay delay = MTI->maxLatency(opCode); //Create new node for this machine instruction and add to the graph. //Create only if not a nop if(MTI->isNop(opCode)) continue; //Sparc BE does not use PHI opcode, so assert on this case assert(opCode != TargetInstrInfo::PHI && "Did not expect PHI opcode"); bool isBranch = false; //We want to flag the branch node to treat it special if(MTI->isBranch(opCode)) isBranch = true; //Node is created and added to the graph automatically MSchedGraphNode *node = new MSchedGraphNode(MI, this, index, delay, isBranch); DEBUG(std::cerr << "Created Node: " << *node << "\n"); //Check OpCode to keep track of memory operations to add memory //dependencies later. if(MTI->isLoad(opCode) || MTI->isStore(opCode)) memInstructions.push_back(node); //Loop over all operands, and put them into the register number to //graph node map for determining dependencies //If an operands is a use/def, we have an anti dependence to itself for(unsigned i=0; i < MI->getNumOperands(); ++i) { //Get Operand const MachineOperand &mOp = MI->getOperand(i); //Check if it has an allocated register if(mOp.hasAllocatedReg()) { int regNum = mOp.getReg(); if(regNum != SparcV9::g0) { //Put into our map regNumtoNodeMap[regNum].push_back(std::make_pair(i, node)); } continue; } //Add virtual registers dependencies //Check if any exist in the value map already and create dependencies //between them. if(mOp.getType() == MachineOperand::MO_VirtualRegister || mOp.getType() == MachineOperand::MO_CCRegister) { //Make sure virtual register value is not null assert((mOp.getVRegValue() != NULL) && "Null value is defined"); //Check if this is a read operation in a phi node, if so DO NOT PROCESS if(mOp.isUse() && (opCode == TargetInstrInfo::PHI)) { DEBUG(std::cerr << "Read Operation in a PHI node\n"); continue; } if (const Value* srcI = mOp.getVRegValue()) { //Find value in the map std::map >::iterator V = valuetoNodeMap.find(srcI); //If there is something in the map already, add edges from //those instructions //to this one we are processing if(V != valuetoNodeMap.end()) { addValueEdges(V->second, node, mOp.isUse(), mOp.isDef(), phiInstrs); //Add to value map V->second.push_back(std::make_pair(i,node)); } //Otherwise put it in the map else //Put into value map valuetoNodeMap[mOp.getVRegValue()].push_back(std::make_pair(i, node)); } } } ++index; } //Loop over LLVM BB, examine phi instructions, and add them to our //phiInstr list to process const BasicBlock *llvm_bb = BB->getBasicBlock(); for(BasicBlock::const_iterator I = llvm_bb->begin(), E = llvm_bb->end(); I != E; ++I) { if(const PHINode *PN = dyn_cast(I)) { MachineCodeForInstruction & tempMvec = MachineCodeForInstruction::get(PN); for (unsigned j = 0; j < tempMvec.size(); j++) { if(!ignoreInstrs.count(tempMvec[j])) { DEBUG(std::cerr << "Inserting phi instr into map: " << *tempMvec[j] << "\n"); phiInstrs.push_back((MachineInstr*) tempMvec[j]); } } } } addMemEdges(memInstructions, DA, machineTollvm); addMachRegEdges(regNumtoNodeMap); //Finally deal with PHI Nodes and Value* for(std::vector::iterator I = phiInstrs.begin(), E = phiInstrs.end(); I != E; ++I) { //Get Node for this instruction std::map::iterator X; X = find(*I); if(X == GraphMap.end()) continue; MSchedGraphNode *node = X->second; DEBUG(std::cerr << "Adding ite diff edges for node: " << *node << "\n"); //Loop over operands for this instruction and add value edges for(unsigned i=0; i < (*I)->getNumOperands(); ++i) { //Get Operand const MachineOperand &mOp = (*I)->getOperand(i); if((mOp.getType() == MachineOperand::MO_VirtualRegister || mOp.getType() == MachineOperand::MO_CCRegister) && mOp.isUse()) { //find the value in the map if (const Value* srcI = mOp.getVRegValue()) { //Find value in the map std::map >::iterator V = valuetoNodeMap.find(srcI); //If there is something in the map already, add edges from //those instructions //to this one we are processing if(V != valuetoNodeMap.end()) { addValueEdges(V->second, node, mOp.isUse(), mOp.isDef(), phiInstrs, 1); } } } } } } } //Add dependencies for Value*s void MSchedGraph::addValueEdges(std::vector &NodesInMap, MSchedGraphNode *destNode, bool nodeIsUse, bool nodeIsDef, std::vector &phiInstrs, int diff) { for(std::vector::iterator I = NodesInMap.begin(), E = NodesInMap.end(); I != E; ++I) { //Get node in vectors machine operand that is the same value as node MSchedGraphNode *srcNode = I->second; MachineOperand mOp = srcNode->getInst()->getOperand(I->first); if(diff > 0) if(std::find(phiInstrs.begin(), phiInstrs.end(), srcNode->getInst()) == phiInstrs.end()) continue; //Node is a Def, so add output dep. if(nodeIsDef) { if(mOp.isUse()) { DEBUG(std::cerr << "Edge from " << *srcNode << " to " << *destNode << " (itediff=" << diff << ", type=anti)\n"); srcNode->addOutEdge(destNode, MSchedGraphEdge::ValueDep, MSchedGraphEdge::AntiDep, diff); } if(mOp.isDef()) { DEBUG(std::cerr << "Edge from " << *srcNode << " to " << *destNode << " (itediff=" << diff << ", type=output)\n"); srcNode->addOutEdge(destNode, MSchedGraphEdge::ValueDep, MSchedGraphEdge::OutputDep, diff); } } if(nodeIsUse) { if(mOp.isDef()) { DEBUG(std::cerr << "Edge from " << *srcNode << " to " << *destNode << " (itediff=" << diff << ", type=true)\n"); srcNode->addOutEdge(destNode, MSchedGraphEdge::ValueDep, MSchedGraphEdge::TrueDep, diff); } } } } //Add dependencies for machine registers across iterations void MSchedGraph::addMachRegEdges(std::map >& regNumtoNodeMap) { //Loop over all machine registers in the map, and add dependencies //between the instructions that use it typedef std::map > regNodeMap; for(regNodeMap::iterator I = regNumtoNodeMap.begin(); I != regNumtoNodeMap.end(); ++I) { //Get the register number int regNum = (*I).first; //Get Vector of nodes that use this register std::vector Nodes = (*I).second; //Loop over nodes and determine the dependence between the other //nodes in the vector for(unsigned i =0; i < Nodes.size(); ++i) { //Get src node operator index that uses this machine register int srcOpIndex = Nodes[i].first; //Get the actual src Node MSchedGraphNode *srcNode = Nodes[i].second; //Get Operand const MachineOperand &srcMOp = srcNode->getInst()->getOperand(srcOpIndex); bool srcIsUseandDef = srcMOp.isDef() && srcMOp.isUse(); bool srcIsUse = srcMOp.isUse() && !srcMOp.isDef(); //Look at all instructions after this in execution order for(unsigned j=i+1; j < Nodes.size(); ++j) { //Sink node is a write if(Nodes[j].second->getInst()->getOperand(Nodes[j].first).isDef()) { //Src only uses the register (read) if(srcIsUse) srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::AntiDep); else if(srcIsUseandDef) { srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::AntiDep); srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::OutputDep); } else srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::OutputDep); } //Dest node is a read else { if(!srcIsUse || srcIsUseandDef) srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::TrueDep); } } //Look at all the instructions before this one since machine registers //could live across iterations. for(unsigned j = 0; j < i; ++j) { //Sink node is a write if(Nodes[j].second->getInst()->getOperand(Nodes[j].first).isDef()) { //Src only uses the register (read) if(srcIsUse) srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::AntiDep, 1); else if(srcIsUseandDef) { srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::AntiDep, 1); srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::OutputDep, 1); } else srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::OutputDep, 1); } //Dest node is a read else { if(!srcIsUse || srcIsUseandDef) srcNode->addOutEdge(Nodes[j].second, MSchedGraphEdge::MachineRegister, MSchedGraphEdge::TrueDep,1 ); } } } } } //Add edges between all loads and stores //Can be less strict with alias analysis and data dependence analysis. void MSchedGraph::addMemEdges(const std::vector& memInst, DependenceAnalyzer &DA, std::map &machineTollvm) { //Get Target machine instruction info const TargetInstrInfo *TMI = Target.getInstrInfo(); //Loop over all memory instructions in the vector //Knowing that they are in execution, add true, anti, and output dependencies for (unsigned srcIndex = 0; srcIndex < memInst.size(); ++srcIndex) { MachineInstr *srcInst = (MachineInstr*) memInst[srcIndex]->getInst(); //Get the machine opCode to determine type of memory instruction MachineOpCode srcNodeOpCode = srcInst->getOpcode(); //All instructions after this one in execution order have an //iteration delay of 0 for(unsigned destIndex = 0; destIndex < memInst.size(); ++destIndex) { //No self loops if(destIndex == srcIndex) continue; MachineInstr *destInst = (MachineInstr*) memInst[destIndex]->getInst(); DEBUG(std::cerr << "MInst1: " << *srcInst << "\n"); DEBUG(std::cerr << "MInst2: " << *destInst << "\n"); //Assuming instructions without corresponding llvm instructions //are from constant pools. if (!machineTollvm.count(srcInst) || !machineTollvm.count(destInst)) continue; bool useDepAnalyzer = true; //Some machine loads and stores are generated by casts, so be //conservative and always add deps Instruction *srcLLVM = machineTollvm[srcInst]; Instruction *destLLVM = machineTollvm[destInst]; if(!isa(srcLLVM) && !isa(srcLLVM)) { if(isa(srcLLVM)) { if(isa(srcLLVM->getOperand(0)) || isa(srcLLVM->getOperand(1))) continue; } useDepAnalyzer = false; } if(!isa(destLLVM) && !isa(destLLVM)) { if(isa(destLLVM)) { if(isa(destLLVM->getOperand(0)) || isa(destLLVM->getOperand(1))) continue; } useDepAnalyzer = false; } //Use dep analysis when we have corresponding llvm loads/stores if(useDepAnalyzer) { bool srcBeforeDest = true; if(destIndex < srcIndex) srcBeforeDest = false; DependenceResult dr = DA.getDependenceInfo(machineTollvm[srcInst], machineTollvm[destInst], srcBeforeDest); for(std::vector::iterator d = dr.dependences.begin(), de = dr.dependences.end(); d != de; ++d) { //Add edge from load to store memInst[srcIndex]->addOutEdge(memInst[destIndex], MSchedGraphEdge::MemoryDep, d->getDepType(), d->getIteDiff()); } } //Otherwise, we can not do any further analysis and must make a dependence else { //Get the machine opCode to determine type of memory instruction MachineOpCode destNodeOpCode = destInst->getOpcode(); //Get the Value* that we are reading from the load, always the first op const MachineOperand &mOp = srcInst->getOperand(0); const MachineOperand &mOp2 = destInst->getOperand(0); if(mOp.hasAllocatedReg()) if(mOp.getReg() == SparcV9::g0) continue; if(mOp2.hasAllocatedReg()) if(mOp2.getReg() == SparcV9::g0) continue; DEBUG(std::cerr << "Adding dependence for machine instructions\n"); //Load-Store deps if(TMI->isLoad(srcNodeOpCode)) { if(TMI->isStore(destNodeOpCode)) memInst[srcIndex]->addOutEdge(memInst[destIndex], MSchedGraphEdge::MemoryDep, MSchedGraphEdge::AntiDep, 0); } else if(TMI->isStore(srcNodeOpCode)) { if(TMI->isStore(destNodeOpCode)) memInst[srcIndex]->addOutEdge(memInst[destIndex], MSchedGraphEdge::MemoryDep, MSchedGraphEdge::OutputDep, 0); else memInst[srcIndex]->addOutEdge(memInst[destIndex], MSchedGraphEdge::MemoryDep, MSchedGraphEdge::TrueDep, 0); } } } } }