1 //===- MacroFusion.cpp - Macro Fusion -------------------------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 /// \file This file contains the implementation of the DAG scheduling mutation 11 /// to pair instructions back to back. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "llvm/CodeGen/MacroFusion.h" 16 #include "llvm/ADT/STLExtras.h" 17 #include "llvm/ADT/Statistic.h" 18 #include "llvm/CodeGen/MachineInstr.h" 19 #include "llvm/CodeGen/MachineScheduler.h" 20 #include "llvm/CodeGen/ScheduleDAG.h" 21 #include "llvm/CodeGen/ScheduleDAGMutation.h" 22 #include "llvm/CodeGen/TargetInstrInfo.h" 23 #include "llvm/Support/CommandLine.h" 24 #include "llvm/Support/Debug.h" 25 #include "llvm/Support/raw_ostream.h" 26 27 #define DEBUG_TYPE "machine-scheduler" 28 29 STATISTIC(NumFused, "Number of instr pairs fused"); 30 31 using namespace llvm; 32 33 static cl::opt<bool> EnableMacroFusion("misched-fusion", cl::Hidden, 34 cl::desc("Enable scheduling for macro fusion."), cl::init(true)); 35 36 static bool isHazard(const SDep &Dep) { 37 return Dep.getKind() == SDep::Anti || Dep.getKind() == SDep::Output; 38 } 39 40 static bool fuseInstructionPair(ScheduleDAGMI &DAG, SUnit &FirstSU, 41 SUnit &SecondSU) { 42 // Check that neither instr is already paired with another along the edge 43 // between them. 44 for (SDep &SI : FirstSU.Succs) 45 if (SI.isCluster()) 46 return false; 47 48 for (SDep &SI : SecondSU.Preds) 49 if (SI.isCluster()) 50 return false; 51 // Though the reachability checks above could be made more generic, 52 // perhaps as part of ScheduleDAGMI::addEdge(), since such edges are valid, 53 // the extra computation cost makes it less interesting in general cases. 54 55 // Create a single weak edge between the adjacent instrs. The only effect is 56 // to cause bottom-up scheduling to heavily prioritize the clustered instrs. 57 if (!DAG.addEdge(&SecondSU, SDep(&FirstSU, SDep::Cluster))) 58 return false; 59 60 // Adjust the latency between both instrs. 61 for (SDep &SI : FirstSU.Succs) 62 if (SI.getSUnit() == &SecondSU) 63 SI.setLatency(0); 64 65 for (SDep &SI : SecondSU.Preds) 66 if (SI.getSUnit() == &FirstSU) 67 SI.setLatency(0); 68 69 DEBUG(dbgs() << "Macro fuse: "; 70 FirstSU.print(dbgs(), &DAG); dbgs() << " - "; 71 SecondSU.print(dbgs(), &DAG); dbgs() << " / "; 72 dbgs() << DAG.TII->getName(FirstSU.getInstr()->getOpcode()) << " - " << 73 DAG.TII->getName(SecondSU.getInstr()->getOpcode()) << '\n'; ); 74 75 // Make data dependencies from the FirstSU also dependent on the SecondSU to 76 // prevent them from being scheduled between the FirstSU and the SecondSU. 77 if (&SecondSU != &DAG.ExitSU) 78 for (const SDep &SI : FirstSU.Succs) { 79 SUnit *SU = SI.getSUnit(); 80 if (SI.isWeak() || isHazard(SI) || 81 SU == &DAG.ExitSU || SU == &SecondSU || SU->isPred(&SecondSU)) 82 continue; 83 DEBUG(dbgs() << " Bind "; 84 SecondSU.print(dbgs(), &DAG); dbgs() << " - "; 85 SU->print(dbgs(), &DAG); dbgs() << '\n';); 86 DAG.addEdge(SU, SDep(&SecondSU, SDep::Artificial)); 87 } 88 89 // Make the FirstSU also dependent on the dependencies of the SecondSU to 90 // prevent them from being scheduled between the FirstSU and the SecondSU. 91 if (&FirstSU != &DAG.EntrySU) 92 for (const SDep &SI : SecondSU.Preds) { 93 SUnit *SU = SI.getSUnit(); 94 if (SI.isWeak() || isHazard(SI) || &FirstSU == SU || FirstSU.isSucc(SU)) 95 continue; 96 DEBUG(dbgs() << " Bind "; 97 SU->print(dbgs(), &DAG); dbgs() << " - "; 98 FirstSU.print(dbgs(), &DAG); dbgs() << '\n';); 99 DAG.addEdge(&FirstSU, SDep(SU, SDep::Artificial)); 100 } 101 102 ++NumFused; 103 return true; 104 } 105 106 namespace { 107 108 /// Post-process the DAG to create cluster edges between instrs that may 109 /// be fused by the processor into a single operation. 110 class MacroFusion : public ScheduleDAGMutation { 111 ShouldSchedulePredTy shouldScheduleAdjacent; 112 bool FuseBlock; 113 bool scheduleAdjacentImpl(ScheduleDAGMI &DAG, SUnit &AnchorSU); 114 115 public: 116 MacroFusion(ShouldSchedulePredTy shouldScheduleAdjacent, bool FuseBlock) 117 : shouldScheduleAdjacent(shouldScheduleAdjacent), FuseBlock(FuseBlock) {} 118 119 void apply(ScheduleDAGInstrs *DAGInstrs) override; 120 }; 121 122 } // end anonymous namespace 123 124 void MacroFusion::apply(ScheduleDAGInstrs *DAGInstrs) { 125 ScheduleDAGMI *DAG = static_cast<ScheduleDAGMI*>(DAGInstrs); 126 127 if (FuseBlock) 128 // For each of the SUnits in the scheduling block, try to fuse the instr in 129 // it with one in its predecessors. 130 for (SUnit &ISU : DAG->SUnits) 131 scheduleAdjacentImpl(*DAG, ISU); 132 133 if (DAG->ExitSU.getInstr()) 134 // Try to fuse the instr in the ExitSU with one in its predecessors. 135 scheduleAdjacentImpl(*DAG, DAG->ExitSU); 136 } 137 138 /// Implement the fusion of instr pairs in the scheduling DAG, 139 /// anchored at the instr in AnchorSU.. 140 bool MacroFusion::scheduleAdjacentImpl(ScheduleDAGMI &DAG, SUnit &AnchorSU) { 141 const MachineInstr &AnchorMI = *AnchorSU.getInstr(); 142 const TargetInstrInfo &TII = *DAG.TII; 143 const TargetSubtargetInfo &ST = DAG.MF.getSubtarget(); 144 145 // Check if the anchor instr may be fused. 146 if (!shouldScheduleAdjacent(TII, ST, nullptr, AnchorMI)) 147 return false; 148 149 // Explorer for fusion candidates among the dependencies of the anchor instr. 150 for (SDep &Dep : AnchorSU.Preds) { 151 // Ignore dependencies other than data or strong ordering. 152 if (Dep.isWeak() || isHazard(Dep)) 153 continue; 154 155 SUnit &DepSU = *Dep.getSUnit(); 156 if (DepSU.isBoundaryNode()) 157 continue; 158 159 const MachineInstr *DepMI = DepSU.getInstr(); 160 if (!shouldScheduleAdjacent(TII, ST, DepMI, AnchorMI)) 161 continue; 162 163 if (fuseInstructionPair(DAG, DepSU, AnchorSU)) 164 return true; 165 } 166 167 return false; 168 } 169 170 std::unique_ptr<ScheduleDAGMutation> 171 llvm::createMacroFusionDAGMutation( 172 ShouldSchedulePredTy shouldScheduleAdjacent) { 173 if(EnableMacroFusion) 174 return llvm::make_unique<MacroFusion>(shouldScheduleAdjacent, true); 175 return nullptr; 176 } 177 178 std::unique_ptr<ScheduleDAGMutation> 179 llvm::createBranchMacroFusionDAGMutation( 180 ShouldSchedulePredTy shouldScheduleAdjacent) { 181 if(EnableMacroFusion) 182 return llvm::make_unique<MacroFusion>(shouldScheduleAdjacent, false); 183 return nullptr; 184 } 185