Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 14 | pmbaty | 1 | //===- MachinePipeliner.h - Machine Software Pipeliner Pass -------------===// |
| 2 | // |
||
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
||
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
||
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
||
| 6 | // |
||
| 7 | //===----------------------------------------------------------------------===// |
||
| 8 | // |
||
| 9 | // An implementation of the Swing Modulo Scheduling (SMS) software pipeliner. |
||
| 10 | // |
||
| 11 | // Software pipelining (SWP) is an instruction scheduling technique for loops |
||
| 12 | // that overlap loop iterations and exploits ILP via a compiler transformation. |
||
| 13 | // |
||
| 14 | // Swing Modulo Scheduling is an implementation of software pipelining |
||
| 15 | // that generates schedules that are near optimal in terms of initiation |
||
| 16 | // interval, register requirements, and stage count. See the papers: |
||
| 17 | // |
||
| 18 | // "Swing Modulo Scheduling: A Lifetime-Sensitive Approach", by J. Llosa, |
||
| 19 | // A. Gonzalez, E. Ayguade, and M. Valero. In PACT '96 Proceedings of the 1996 |
||
| 20 | // Conference on Parallel Architectures and Compilation Techiniques. |
||
| 21 | // |
||
| 22 | // "Lifetime-Sensitive Modulo Scheduling in a Production Environment", by J. |
||
| 23 | // Llosa, E. Ayguade, A. Gonzalez, M. Valero, and J. Eckhardt. In IEEE |
||
| 24 | // Transactions on Computers, Vol. 50, No. 3, 2001. |
||
| 25 | // |
||
| 26 | // "An Implementation of Swing Modulo Scheduling With Extensions for |
||
| 27 | // Superblocks", by T. Lattner, Master's Thesis, University of Illinois at |
||
| 28 | // Urbana-Champaign, 2005. |
||
| 29 | // |
||
| 30 | // |
||
| 31 | // The SMS algorithm consists of three main steps after computing the minimal |
||
| 32 | // initiation interval (MII). |
||
| 33 | // 1) Analyze the dependence graph and compute information about each |
||
| 34 | // instruction in the graph. |
||
| 35 | // 2) Order the nodes (instructions) by priority based upon the heuristics |
||
| 36 | // described in the algorithm. |
||
| 37 | // 3) Attempt to schedule the nodes in the specified order using the MII. |
||
| 38 | // |
||
| 39 | //===----------------------------------------------------------------------===// |
||
| 40 | #ifndef LLVM_CODEGEN_MACHINEPIPELINER_H |
||
| 41 | #define LLVM_CODEGEN_MACHINEPIPELINER_H |
||
| 42 | |||
| 43 | #include "llvm/ADT/SetVector.h" |
||
| 44 | #include "llvm/CodeGen/DFAPacketizer.h" |
||
| 45 | #include "llvm/CodeGen/MachineDominators.h" |
||
| 46 | #include "llvm/CodeGen/MachineOptimizationRemarkEmitter.h" |
||
| 47 | #include "llvm/CodeGen/RegisterClassInfo.h" |
||
| 48 | #include "llvm/CodeGen/ScheduleDAGInstrs.h" |
||
| 49 | #include "llvm/CodeGen/ScheduleDAGMutation.h" |
||
| 50 | #include "llvm/CodeGen/TargetInstrInfo.h" |
||
| 51 | #include "llvm/InitializePasses.h" |
||
| 52 | |||
| 53 | #include <deque> |
||
| 54 | |||
| 55 | namespace llvm { |
||
| 56 | |||
| 57 | class AAResults; |
||
| 58 | class NodeSet; |
||
| 59 | class SMSchedule; |
||
| 60 | |||
| 61 | extern cl::opt<bool> SwpEnableCopyToPhi; |
||
| 62 | extern cl::opt<int> SwpForceIssueWidth; |
||
| 63 | |||
| 64 | /// The main class in the implementation of the target independent |
||
| 65 | /// software pipeliner pass. |
||
| 66 | class MachinePipeliner : public MachineFunctionPass { |
||
| 67 | public: |
||
| 68 | MachineFunction *MF = nullptr; |
||
| 69 | MachineOptimizationRemarkEmitter *ORE = nullptr; |
||
| 70 | const MachineLoopInfo *MLI = nullptr; |
||
| 71 | const MachineDominatorTree *MDT = nullptr; |
||
| 72 | const InstrItineraryData *InstrItins; |
||
| 73 | const TargetInstrInfo *TII = nullptr; |
||
| 74 | RegisterClassInfo RegClassInfo; |
||
| 75 | bool disabledByPragma = false; |
||
| 76 | unsigned II_setByPragma = 0; |
||
| 77 | |||
| 78 | #ifndef NDEBUG |
||
| 79 | static int NumTries; |
||
| 80 | #endif |
||
| 81 | |||
| 82 | /// Cache the target analysis information about the loop. |
||
| 83 | struct LoopInfo { |
||
| 84 | MachineBasicBlock *TBB = nullptr; |
||
| 85 | MachineBasicBlock *FBB = nullptr; |
||
| 86 | SmallVector<MachineOperand, 4> BrCond; |
||
| 87 | MachineInstr *LoopInductionVar = nullptr; |
||
| 88 | MachineInstr *LoopCompare = nullptr; |
||
| 89 | std::unique_ptr<TargetInstrInfo::PipelinerLoopInfo> LoopPipelinerInfo = |
||
| 90 | nullptr; |
||
| 91 | }; |
||
| 92 | LoopInfo LI; |
||
| 93 | |||
| 94 | static char ID; |
||
| 95 | |||
| 96 | MachinePipeliner() : MachineFunctionPass(ID) { |
||
| 97 | initializeMachinePipelinerPass(*PassRegistry::getPassRegistry()); |
||
| 98 | } |
||
| 99 | |||
| 100 | bool runOnMachineFunction(MachineFunction &MF) override; |
||
| 101 | |||
| 102 | void getAnalysisUsage(AnalysisUsage &AU) const override; |
||
| 103 | |||
| 104 | private: |
||
| 105 | void preprocessPhiNodes(MachineBasicBlock &B); |
||
| 106 | bool canPipelineLoop(MachineLoop &L); |
||
| 107 | bool scheduleLoop(MachineLoop &L); |
||
| 108 | bool swingModuloScheduler(MachineLoop &L); |
||
| 109 | void setPragmaPipelineOptions(MachineLoop &L); |
||
| 110 | }; |
||
| 111 | |||
| 112 | /// This class builds the dependence graph for the instructions in a loop, |
||
| 113 | /// and attempts to schedule the instructions using the SMS algorithm. |
||
| 114 | class SwingSchedulerDAG : public ScheduleDAGInstrs { |
||
| 115 | MachinePipeliner &Pass; |
||
| 116 | /// The minimum initiation interval between iterations for this schedule. |
||
| 117 | unsigned MII = 0; |
||
| 118 | /// The maximum initiation interval between iterations for this schedule. |
||
| 119 | unsigned MAX_II = 0; |
||
| 120 | /// Set to true if a valid pipelined schedule is found for the loop. |
||
| 121 | bool Scheduled = false; |
||
| 122 | MachineLoop &Loop; |
||
| 123 | LiveIntervals &LIS; |
||
| 124 | const RegisterClassInfo &RegClassInfo; |
||
| 125 | unsigned II_setByPragma = 0; |
||
| 126 | TargetInstrInfo::PipelinerLoopInfo *LoopPipelinerInfo = nullptr; |
||
| 127 | |||
| 128 | /// A toplogical ordering of the SUnits, which is needed for changing |
||
| 129 | /// dependences and iterating over the SUnits. |
||
| 130 | ScheduleDAGTopologicalSort Topo; |
||
| 131 | |||
| 132 | struct NodeInfo { |
||
| 133 | int ASAP = 0; |
||
| 134 | int ALAP = 0; |
||
| 135 | int ZeroLatencyDepth = 0; |
||
| 136 | int ZeroLatencyHeight = 0; |
||
| 137 | |||
| 138 | NodeInfo() = default; |
||
| 139 | }; |
||
| 140 | /// Computed properties for each node in the graph. |
||
| 141 | std::vector<NodeInfo> ScheduleInfo; |
||
| 142 | |||
| 143 | enum OrderKind { BottomUp = 0, TopDown = 1 }; |
||
| 144 | /// Computed node ordering for scheduling. |
||
| 145 | SetVector<SUnit *> NodeOrder; |
||
| 146 | |||
| 147 | using NodeSetType = SmallVector<NodeSet, 8>; |
||
| 148 | using ValueMapTy = DenseMap<unsigned, unsigned>; |
||
| 149 | using MBBVectorTy = SmallVectorImpl<MachineBasicBlock *>; |
||
| 150 | using InstrMapTy = DenseMap<MachineInstr *, MachineInstr *>; |
||
| 151 | |||
| 152 | /// Instructions to change when emitting the final schedule. |
||
| 153 | DenseMap<SUnit *, std::pair<unsigned, int64_t>> InstrChanges; |
||
| 154 | |||
| 155 | /// We may create a new instruction, so remember it because it |
||
| 156 | /// must be deleted when the pass is finished. |
||
| 157 | DenseMap<MachineInstr*, MachineInstr *> NewMIs; |
||
| 158 | |||
| 159 | /// Ordered list of DAG postprocessing steps. |
||
| 160 | std::vector<std::unique_ptr<ScheduleDAGMutation>> Mutations; |
||
| 161 | |||
| 162 | /// Helper class to implement Johnson's circuit finding algorithm. |
||
| 163 | class Circuits { |
||
| 164 | std::vector<SUnit> &SUnits; |
||
| 165 | SetVector<SUnit *> Stack; |
||
| 166 | BitVector Blocked; |
||
| 167 | SmallVector<SmallPtrSet<SUnit *, 4>, 10> B; |
||
| 168 | SmallVector<SmallVector<int, 4>, 16> AdjK; |
||
| 169 | // Node to Index from ScheduleDAGTopologicalSort |
||
| 170 | std::vector<int> *Node2Idx; |
||
| 171 | unsigned NumPaths; |
||
| 172 | static unsigned MaxPaths; |
||
| 173 | |||
| 174 | public: |
||
| 175 | Circuits(std::vector<SUnit> &SUs, ScheduleDAGTopologicalSort &Topo) |
||
| 176 | : SUnits(SUs), Blocked(SUs.size()), B(SUs.size()), AdjK(SUs.size()) { |
||
| 177 | Node2Idx = new std::vector<int>(SUs.size()); |
||
| 178 | unsigned Idx = 0; |
||
| 179 | for (const auto &NodeNum : Topo) |
||
| 180 | Node2Idx->at(NodeNum) = Idx++; |
||
| 181 | } |
||
| 182 | |||
| 183 | ~Circuits() { delete Node2Idx; } |
||
| 184 | |||
| 185 | /// Reset the data structures used in the circuit algorithm. |
||
| 186 | void reset() { |
||
| 187 | Stack.clear(); |
||
| 188 | Blocked.reset(); |
||
| 189 | B.assign(SUnits.size(), SmallPtrSet<SUnit *, 4>()); |
||
| 190 | NumPaths = 0; |
||
| 191 | } |
||
| 192 | |||
| 193 | void createAdjacencyStructure(SwingSchedulerDAG *DAG); |
||
| 194 | bool circuit(int V, int S, NodeSetType &NodeSets, bool HasBackedge = false); |
||
| 195 | void unblock(int U); |
||
| 196 | }; |
||
| 197 | |||
| 198 | struct CopyToPhiMutation : public ScheduleDAGMutation { |
||
| 199 | void apply(ScheduleDAGInstrs *DAG) override; |
||
| 200 | }; |
||
| 201 | |||
| 202 | public: |
||
| 203 | SwingSchedulerDAG(MachinePipeliner &P, MachineLoop &L, LiveIntervals &lis, |
||
| 204 | const RegisterClassInfo &rci, unsigned II, |
||
| 205 | TargetInstrInfo::PipelinerLoopInfo *PLI) |
||
| 206 | : ScheduleDAGInstrs(*P.MF, P.MLI, false), Pass(P), Loop(L), LIS(lis), |
||
| 207 | RegClassInfo(rci), II_setByPragma(II), LoopPipelinerInfo(PLI), |
||
| 208 | Topo(SUnits, &ExitSU) { |
||
| 209 | P.MF->getSubtarget().getSMSMutations(Mutations); |
||
| 210 | if (SwpEnableCopyToPhi) |
||
| 211 | Mutations.push_back(std::make_unique<CopyToPhiMutation>()); |
||
| 212 | } |
||
| 213 | |||
| 214 | void schedule() override; |
||
| 215 | void finishBlock() override; |
||
| 216 | |||
| 217 | /// Return true if the loop kernel has been scheduled. |
||
| 218 | bool hasNewSchedule() { return Scheduled; } |
||
| 219 | |||
| 220 | /// Return the earliest time an instruction may be scheduled. |
||
| 221 | int getASAP(SUnit *Node) { return ScheduleInfo[Node->NodeNum].ASAP; } |
||
| 222 | |||
| 223 | /// Return the latest time an instruction my be scheduled. |
||
| 224 | int getALAP(SUnit *Node) { return ScheduleInfo[Node->NodeNum].ALAP; } |
||
| 225 | |||
| 226 | /// The mobility function, which the number of slots in which |
||
| 227 | /// an instruction may be scheduled. |
||
| 228 | int getMOV(SUnit *Node) { return getALAP(Node) - getASAP(Node); } |
||
| 229 | |||
| 230 | /// The depth, in the dependence graph, for a node. |
||
| 231 | unsigned getDepth(SUnit *Node) { return Node->getDepth(); } |
||
| 232 | |||
| 233 | /// The maximum unweighted length of a path from an arbitrary node to the |
||
| 234 | /// given node in which each edge has latency 0 |
||
| 235 | int getZeroLatencyDepth(SUnit *Node) { |
||
| 236 | return ScheduleInfo[Node->NodeNum].ZeroLatencyDepth; |
||
| 237 | } |
||
| 238 | |||
| 239 | /// The height, in the dependence graph, for a node. |
||
| 240 | unsigned getHeight(SUnit *Node) { return Node->getHeight(); } |
||
| 241 | |||
| 242 | /// The maximum unweighted length of a path from the given node to an |
||
| 243 | /// arbitrary node in which each edge has latency 0 |
||
| 244 | int getZeroLatencyHeight(SUnit *Node) { |
||
| 245 | return ScheduleInfo[Node->NodeNum].ZeroLatencyHeight; |
||
| 246 | } |
||
| 247 | |||
| 248 | /// Return true if the dependence is a back-edge in the data dependence graph. |
||
| 249 | /// Since the DAG doesn't contain cycles, we represent a cycle in the graph |
||
| 250 | /// using an anti dependence from a Phi to an instruction. |
||
| 251 | bool isBackedge(SUnit *Source, const SDep &Dep) { |
||
| 252 | if (Dep.getKind() != SDep::Anti) |
||
| 253 | return false; |
||
| 254 | return Source->getInstr()->isPHI() || Dep.getSUnit()->getInstr()->isPHI(); |
||
| 255 | } |
||
| 256 | |||
| 257 | bool isLoopCarriedDep(SUnit *Source, const SDep &Dep, bool isSucc = true); |
||
| 258 | |||
| 259 | /// The distance function, which indicates that operation V of iteration I |
||
| 260 | /// depends on operations U of iteration I-distance. |
||
| 261 | unsigned getDistance(SUnit *U, SUnit *V, const SDep &Dep) { |
||
| 262 | // Instructions that feed a Phi have a distance of 1. Computing larger |
||
| 263 | // values for arrays requires data dependence information. |
||
| 264 | if (V->getInstr()->isPHI() && Dep.getKind() == SDep::Anti) |
||
| 265 | return 1; |
||
| 266 | return 0; |
||
| 267 | } |
||
| 268 | |||
| 269 | void applyInstrChange(MachineInstr *MI, SMSchedule &Schedule); |
||
| 270 | |||
| 271 | void fixupRegisterOverlaps(std::deque<SUnit *> &Instrs); |
||
| 272 | |||
| 273 | /// Return the new base register that was stored away for the changed |
||
| 274 | /// instruction. |
||
| 275 | unsigned getInstrBaseReg(SUnit *SU) { |
||
| 276 | DenseMap<SUnit *, std::pair<unsigned, int64_t>>::iterator It = |
||
| 277 | InstrChanges.find(SU); |
||
| 278 | if (It != InstrChanges.end()) |
||
| 279 | return It->second.first; |
||
| 280 | return 0; |
||
| 281 | } |
||
| 282 | |||
| 283 | void addMutation(std::unique_ptr<ScheduleDAGMutation> Mutation) { |
||
| 284 | Mutations.push_back(std::move(Mutation)); |
||
| 285 | } |
||
| 286 | |||
| 287 | static bool classof(const ScheduleDAGInstrs *DAG) { return true; } |
||
| 288 | |||
| 289 | private: |
||
| 290 | void addLoopCarriedDependences(AAResults *AA); |
||
| 291 | void updatePhiDependences(); |
||
| 292 | void changeDependences(); |
||
| 293 | unsigned calculateResMII(); |
||
| 294 | unsigned calculateRecMII(NodeSetType &RecNodeSets); |
||
| 295 | void findCircuits(NodeSetType &NodeSets); |
||
| 296 | void fuseRecs(NodeSetType &NodeSets); |
||
| 297 | void removeDuplicateNodes(NodeSetType &NodeSets); |
||
| 298 | void computeNodeFunctions(NodeSetType &NodeSets); |
||
| 299 | void registerPressureFilter(NodeSetType &NodeSets); |
||
| 300 | void colocateNodeSets(NodeSetType &NodeSets); |
||
| 301 | void checkNodeSets(NodeSetType &NodeSets); |
||
| 302 | void groupRemainingNodes(NodeSetType &NodeSets); |
||
| 303 | void addConnectedNodes(SUnit *SU, NodeSet &NewSet, |
||
| 304 | SetVector<SUnit *> &NodesAdded); |
||
| 305 | void computeNodeOrder(NodeSetType &NodeSets); |
||
| 306 | void checkValidNodeOrder(const NodeSetType &Circuits) const; |
||
| 307 | bool schedulePipeline(SMSchedule &Schedule); |
||
| 308 | bool computeDelta(MachineInstr &MI, unsigned &Delta); |
||
| 309 | MachineInstr *findDefInLoop(Register Reg); |
||
| 310 | bool canUseLastOffsetValue(MachineInstr *MI, unsigned &BasePos, |
||
| 311 | unsigned &OffsetPos, unsigned &NewBase, |
||
| 312 | int64_t &NewOffset); |
||
| 313 | void postprocessDAG(); |
||
| 314 | /// Set the Minimum Initiation Interval for this schedule attempt. |
||
| 315 | void setMII(unsigned ResMII, unsigned RecMII); |
||
| 316 | /// Set the Maximum Initiation Interval for this schedule attempt. |
||
| 317 | void setMAX_II(); |
||
| 318 | }; |
||
| 319 | |||
| 320 | /// A NodeSet contains a set of SUnit DAG nodes with additional information |
||
| 321 | /// that assigns a priority to the set. |
||
| 322 | class NodeSet { |
||
| 323 | SetVector<SUnit *> Nodes; |
||
| 324 | bool HasRecurrence = false; |
||
| 325 | unsigned RecMII = 0; |
||
| 326 | int MaxMOV = 0; |
||
| 327 | unsigned MaxDepth = 0; |
||
| 328 | unsigned Colocate = 0; |
||
| 329 | SUnit *ExceedPressure = nullptr; |
||
| 330 | unsigned Latency = 0; |
||
| 331 | |||
| 332 | public: |
||
| 333 | using iterator = SetVector<SUnit *>::const_iterator; |
||
| 334 | |||
| 335 | NodeSet() = default; |
||
| 336 | NodeSet(iterator S, iterator E) : Nodes(S, E), HasRecurrence(true) { |
||
| 337 | Latency = 0; |
||
| 338 | for (const SUnit *Node : Nodes) { |
||
| 339 | DenseMap<SUnit *, unsigned> SuccSUnitLatency; |
||
| 340 | for (const SDep &Succ : Node->Succs) { |
||
| 341 | auto SuccSUnit = Succ.getSUnit(); |
||
| 342 | if (!Nodes.count(SuccSUnit)) |
||
| 343 | continue; |
||
| 344 | unsigned CurLatency = Succ.getLatency(); |
||
| 345 | unsigned MaxLatency = 0; |
||
| 346 | if (SuccSUnitLatency.count(SuccSUnit)) |
||
| 347 | MaxLatency = SuccSUnitLatency[SuccSUnit]; |
||
| 348 | if (CurLatency > MaxLatency) |
||
| 349 | SuccSUnitLatency[SuccSUnit] = CurLatency; |
||
| 350 | } |
||
| 351 | for (auto SUnitLatency : SuccSUnitLatency) |
||
| 352 | Latency += SUnitLatency.second; |
||
| 353 | } |
||
| 354 | } |
||
| 355 | |||
| 356 | bool insert(SUnit *SU) { return Nodes.insert(SU); } |
||
| 357 | |||
| 358 | void insert(iterator S, iterator E) { Nodes.insert(S, E); } |
||
| 359 | |||
| 360 | template <typename UnaryPredicate> bool remove_if(UnaryPredicate P) { |
||
| 361 | return Nodes.remove_if(P); |
||
| 362 | } |
||
| 363 | |||
| 364 | unsigned count(SUnit *SU) const { return Nodes.count(SU); } |
||
| 365 | |||
| 366 | bool hasRecurrence() { return HasRecurrence; }; |
||
| 367 | |||
| 368 | unsigned size() const { return Nodes.size(); } |
||
| 369 | |||
| 370 | bool empty() const { return Nodes.empty(); } |
||
| 371 | |||
| 372 | SUnit *getNode(unsigned i) const { return Nodes[i]; }; |
||
| 373 | |||
| 374 | void setRecMII(unsigned mii) { RecMII = mii; }; |
||
| 375 | |||
| 376 | void setColocate(unsigned c) { Colocate = c; }; |
||
| 377 | |||
| 378 | void setExceedPressure(SUnit *SU) { ExceedPressure = SU; } |
||
| 379 | |||
| 380 | bool isExceedSU(SUnit *SU) { return ExceedPressure == SU; } |
||
| 381 | |||
| 382 | int compareRecMII(NodeSet &RHS) { return RecMII - RHS.RecMII; } |
||
| 383 | |||
| 384 | int getRecMII() { return RecMII; } |
||
| 385 | |||
| 386 | /// Summarize node functions for the entire node set. |
||
| 387 | void computeNodeSetInfo(SwingSchedulerDAG *SSD) { |
||
| 388 | for (SUnit *SU : *this) { |
||
| 389 | MaxMOV = std::max(MaxMOV, SSD->getMOV(SU)); |
||
| 390 | MaxDepth = std::max(MaxDepth, SSD->getDepth(SU)); |
||
| 391 | } |
||
| 392 | } |
||
| 393 | |||
| 394 | unsigned getLatency() { return Latency; } |
||
| 395 | |||
| 396 | unsigned getMaxDepth() { return MaxDepth; } |
||
| 397 | |||
| 398 | void clear() { |
||
| 399 | Nodes.clear(); |
||
| 400 | RecMII = 0; |
||
| 401 | HasRecurrence = false; |
||
| 402 | MaxMOV = 0; |
||
| 403 | MaxDepth = 0; |
||
| 404 | Colocate = 0; |
||
| 405 | ExceedPressure = nullptr; |
||
| 406 | } |
||
| 407 | |||
| 408 | operator SetVector<SUnit *> &() { return Nodes; } |
||
| 409 | |||
| 410 | /// Sort the node sets by importance. First, rank them by recurrence MII, |
||
| 411 | /// then by mobility (least mobile done first), and finally by depth. |
||
| 412 | /// Each node set may contain a colocate value which is used as the first |
||
| 413 | /// tie breaker, if it's set. |
||
| 414 | bool operator>(const NodeSet &RHS) const { |
||
| 415 | if (RecMII == RHS.RecMII) { |
||
| 416 | if (Colocate != 0 && RHS.Colocate != 0 && Colocate != RHS.Colocate) |
||
| 417 | return Colocate < RHS.Colocate; |
||
| 418 | if (MaxMOV == RHS.MaxMOV) |
||
| 419 | return MaxDepth > RHS.MaxDepth; |
||
| 420 | return MaxMOV < RHS.MaxMOV; |
||
| 421 | } |
||
| 422 | return RecMII > RHS.RecMII; |
||
| 423 | } |
||
| 424 | |||
| 425 | bool operator==(const NodeSet &RHS) const { |
||
| 426 | return RecMII == RHS.RecMII && MaxMOV == RHS.MaxMOV && |
||
| 427 | MaxDepth == RHS.MaxDepth; |
||
| 428 | } |
||
| 429 | |||
| 430 | bool operator!=(const NodeSet &RHS) const { return !operator==(RHS); } |
||
| 431 | |||
| 432 | iterator begin() { return Nodes.begin(); } |
||
| 433 | iterator end() { return Nodes.end(); } |
||
| 434 | void print(raw_ostream &os) const; |
||
| 435 | |||
| 436 | #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) |
||
| 437 | LLVM_DUMP_METHOD void dump() const; |
||
| 438 | #endif |
||
| 439 | }; |
||
| 440 | |||
| 441 | // 16 was selected based on the number of ProcResource kinds for all |
||
| 442 | // existing Subtargets, so that SmallVector don't need to resize too often. |
||
| 443 | static const int DefaultProcResSize = 16; |
||
| 444 | |||
| 445 | class ResourceManager { |
||
| 446 | private: |
||
| 447 | const MCSubtargetInfo *STI; |
||
| 448 | const MCSchedModel &SM; |
||
| 449 | const TargetSubtargetInfo *ST; |
||
| 450 | const TargetInstrInfo *TII; |
||
| 451 | SwingSchedulerDAG *DAG; |
||
| 452 | const bool UseDFA; |
||
| 453 | /// DFA resources for each slot |
||
| 454 | llvm::SmallVector<std::unique_ptr<DFAPacketizer>> DFAResources; |
||
| 455 | /// Modulo Reservation Table. When a resource with ID R is consumed in cycle |
||
| 456 | /// C, it is counted in MRT[C mod II][R]. (Used when UseDFA == F) |
||
| 457 | llvm::SmallVector<llvm::SmallVector<uint64_t, DefaultProcResSize>> MRT; |
||
| 458 | /// The number of scheduled micro operations for each slot. Micro operations |
||
| 459 | /// are assumed to be scheduled one per cycle, starting with the cycle in |
||
| 460 | /// which the instruction is scheduled. |
||
| 461 | llvm::SmallVector<int> NumScheduledMops; |
||
| 462 | /// Each processor resource is associated with a so-called processor resource |
||
| 463 | /// mask. This vector allows to correlate processor resource IDs with |
||
| 464 | /// processor resource masks. There is exactly one element per each processor |
||
| 465 | /// resource declared by the scheduling model. |
||
| 466 | llvm::SmallVector<uint64_t, DefaultProcResSize> ProcResourceMasks; |
||
| 467 | int InitiationInterval; |
||
| 468 | /// The number of micro operations that can be scheduled at a cycle. |
||
| 469 | int IssueWidth; |
||
| 470 | |||
| 471 | int calculateResMIIDFA() const; |
||
| 472 | /// Check if MRT is overbooked |
||
| 473 | bool isOverbooked() const; |
||
| 474 | /// Reserve resources on MRT |
||
| 475 | void reserveResources(const MCSchedClassDesc *SCDesc, int Cycle); |
||
| 476 | /// Unreserve resources on MRT |
||
| 477 | void unreserveResources(const MCSchedClassDesc *SCDesc, int Cycle); |
||
| 478 | |||
| 479 | /// Return M satisfying Dividend = Divisor * X + M, 0 < M < Divisor. |
||
| 480 | /// The slot on MRT to reserve a resource for the cycle C is positiveModulo(C, |
||
| 481 | /// II). |
||
| 482 | int positiveModulo(int Dividend, int Divisor) const { |
||
| 483 | assert(Divisor > 0); |
||
| 484 | int R = Dividend % Divisor; |
||
| 485 | if (R < 0) |
||
| 486 | R += Divisor; |
||
| 487 | return R; |
||
| 488 | } |
||
| 489 | |||
| 490 | #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) |
||
| 491 | LLVM_DUMP_METHOD void dumpMRT() const; |
||
| 492 | #endif |
||
| 493 | |||
| 494 | public: |
||
| 495 | ResourceManager(const TargetSubtargetInfo *ST, SwingSchedulerDAG *DAG) |
||
| 496 | : STI(ST), SM(ST->getSchedModel()), ST(ST), TII(ST->getInstrInfo()), |
||
| 497 | DAG(DAG), UseDFA(ST->useDFAforSMS()), |
||
| 498 | ProcResourceMasks(SM.getNumProcResourceKinds(), 0), |
||
| 499 | IssueWidth(SM.IssueWidth) { |
||
| 500 | initProcResourceVectors(SM, ProcResourceMasks); |
||
| 501 | if (IssueWidth <= 0) |
||
| 502 | // If IssueWidth is not specified, set a sufficiently large value |
||
| 503 | IssueWidth = 100; |
||
| 504 | if (SwpForceIssueWidth > 0) |
||
| 505 | IssueWidth = SwpForceIssueWidth; |
||
| 506 | } |
||
| 507 | |||
| 508 | void initProcResourceVectors(const MCSchedModel &SM, |
||
| 509 | SmallVectorImpl<uint64_t> &Masks); |
||
| 510 | |||
| 511 | /// Check if the resources occupied by a machine instruction are available |
||
| 512 | /// in the current state. |
||
| 513 | bool canReserveResources(SUnit &SU, int Cycle); |
||
| 514 | |||
| 515 | /// Reserve the resources occupied by a machine instruction and change the |
||
| 516 | /// current state to reflect that change. |
||
| 517 | void reserveResources(SUnit &SU, int Cycle); |
||
| 518 | |||
| 519 | int calculateResMII() const; |
||
| 520 | |||
| 521 | /// Initialize resources with the initiation interval II. |
||
| 522 | void init(int II); |
||
| 523 | }; |
||
| 524 | |||
| 525 | /// This class represents the scheduled code. The main data structure is a |
||
| 526 | /// map from scheduled cycle to instructions. During scheduling, the |
||
| 527 | /// data structure explicitly represents all stages/iterations. When |
||
| 528 | /// the algorithm finshes, the schedule is collapsed into a single stage, |
||
| 529 | /// which represents instructions from different loop iterations. |
||
| 530 | /// |
||
| 531 | /// The SMS algorithm allows negative values for cycles, so the first cycle |
||
| 532 | /// in the schedule is the smallest cycle value. |
||
| 533 | class SMSchedule { |
||
| 534 | private: |
||
| 535 | /// Map from execution cycle to instructions. |
||
| 536 | DenseMap<int, std::deque<SUnit *>> ScheduledInstrs; |
||
| 537 | |||
| 538 | /// Map from instruction to execution cycle. |
||
| 539 | std::map<SUnit *, int> InstrToCycle; |
||
| 540 | |||
| 541 | /// Keep track of the first cycle value in the schedule. It starts |
||
| 542 | /// as zero, but the algorithm allows negative values. |
||
| 543 | int FirstCycle = 0; |
||
| 544 | |||
| 545 | /// Keep track of the last cycle value in the schedule. |
||
| 546 | int LastCycle = 0; |
||
| 547 | |||
| 548 | /// The initiation interval (II) for the schedule. |
||
| 549 | int InitiationInterval = 0; |
||
| 550 | |||
| 551 | /// Target machine information. |
||
| 552 | const TargetSubtargetInfo &ST; |
||
| 553 | |||
| 554 | /// Virtual register information. |
||
| 555 | MachineRegisterInfo &MRI; |
||
| 556 | |||
| 557 | ResourceManager ProcItinResources; |
||
| 558 | |||
| 559 | public: |
||
| 560 | SMSchedule(MachineFunction *mf, SwingSchedulerDAG *DAG) |
||
| 561 | : ST(mf->getSubtarget()), MRI(mf->getRegInfo()), |
||
| 562 | ProcItinResources(&ST, DAG) {} |
||
| 563 | |||
| 564 | void reset() { |
||
| 565 | ScheduledInstrs.clear(); |
||
| 566 | InstrToCycle.clear(); |
||
| 567 | FirstCycle = 0; |
||
| 568 | LastCycle = 0; |
||
| 569 | InitiationInterval = 0; |
||
| 570 | } |
||
| 571 | |||
| 572 | /// Set the initiation interval for this schedule. |
||
| 573 | void setInitiationInterval(int ii) { |
||
| 574 | InitiationInterval = ii; |
||
| 575 | ProcItinResources.init(ii); |
||
| 576 | } |
||
| 577 | |||
| 578 | /// Return the initiation interval for this schedule. |
||
| 579 | int getInitiationInterval() const { return InitiationInterval; } |
||
| 580 | |||
| 581 | /// Return the first cycle in the completed schedule. This |
||
| 582 | /// can be a negative value. |
||
| 583 | int getFirstCycle() const { return FirstCycle; } |
||
| 584 | |||
| 585 | /// Return the last cycle in the finalized schedule. |
||
| 586 | int getFinalCycle() const { return FirstCycle + InitiationInterval - 1; } |
||
| 587 | |||
| 588 | /// Return the cycle of the earliest scheduled instruction in the dependence |
||
| 589 | /// chain. |
||
| 590 | int earliestCycleInChain(const SDep &Dep); |
||
| 591 | |||
| 592 | /// Return the cycle of the latest scheduled instruction in the dependence |
||
| 593 | /// chain. |
||
| 594 | int latestCycleInChain(const SDep &Dep); |
||
| 595 | |||
| 596 | void computeStart(SUnit *SU, int *MaxEarlyStart, int *MinLateStart, |
||
| 597 | int *MinEnd, int *MaxStart, int II, SwingSchedulerDAG *DAG); |
||
| 598 | bool insert(SUnit *SU, int StartCycle, int EndCycle, int II); |
||
| 599 | |||
| 600 | /// Iterators for the cycle to instruction map. |
||
| 601 | using sched_iterator = DenseMap<int, std::deque<SUnit *>>::iterator; |
||
| 602 | using const_sched_iterator = |
||
| 603 | DenseMap<int, std::deque<SUnit *>>::const_iterator; |
||
| 604 | |||
| 605 | /// Return true if the instruction is scheduled at the specified stage. |
||
| 606 | bool isScheduledAtStage(SUnit *SU, unsigned StageNum) { |
||
| 607 | return (stageScheduled(SU) == (int)StageNum); |
||
| 608 | } |
||
| 609 | |||
| 610 | /// Return the stage for a scheduled instruction. Return -1 if |
||
| 611 | /// the instruction has not been scheduled. |
||
| 612 | int stageScheduled(SUnit *SU) const { |
||
| 613 | std::map<SUnit *, int>::const_iterator it = InstrToCycle.find(SU); |
||
| 614 | if (it == InstrToCycle.end()) |
||
| 615 | return -1; |
||
| 616 | return (it->second - FirstCycle) / InitiationInterval; |
||
| 617 | } |
||
| 618 | |||
| 619 | /// Return the cycle for a scheduled instruction. This function normalizes |
||
| 620 | /// the first cycle to be 0. |
||
| 621 | unsigned cycleScheduled(SUnit *SU) const { |
||
| 622 | std::map<SUnit *, int>::const_iterator it = InstrToCycle.find(SU); |
||
| 623 | assert(it != InstrToCycle.end() && "Instruction hasn't been scheduled."); |
||
| 624 | return (it->second - FirstCycle) % InitiationInterval; |
||
| 625 | } |
||
| 626 | |||
| 627 | /// Return the maximum stage count needed for this schedule. |
||
| 628 | unsigned getMaxStageCount() { |
||
| 629 | return (LastCycle - FirstCycle) / InitiationInterval; |
||
| 630 | } |
||
| 631 | |||
| 632 | /// Return the instructions that are scheduled at the specified cycle. |
||
| 633 | std::deque<SUnit *> &getInstructions(int cycle) { |
||
| 634 | return ScheduledInstrs[cycle]; |
||
| 635 | } |
||
| 636 | |||
| 637 | SmallSet<SUnit *, 8> |
||
| 638 | computeUnpipelineableNodes(SwingSchedulerDAG *SSD, |
||
| 639 | TargetInstrInfo::PipelinerLoopInfo *PLI); |
||
| 640 | |||
| 641 | bool |
||
| 642 | normalizeNonPipelinedInstructions(SwingSchedulerDAG *SSD, |
||
| 643 | TargetInstrInfo::PipelinerLoopInfo *PLI); |
||
| 644 | bool isValidSchedule(SwingSchedulerDAG *SSD); |
||
| 645 | void finalizeSchedule(SwingSchedulerDAG *SSD); |
||
| 646 | void orderDependence(SwingSchedulerDAG *SSD, SUnit *SU, |
||
| 647 | std::deque<SUnit *> &Insts); |
||
| 648 | bool isLoopCarried(SwingSchedulerDAG *SSD, MachineInstr &Phi); |
||
| 649 | bool isLoopCarriedDefOfUse(SwingSchedulerDAG *SSD, MachineInstr *Def, |
||
| 650 | MachineOperand &MO); |
||
| 651 | void print(raw_ostream &os) const; |
||
| 652 | void dump() const; |
||
| 653 | }; |
||
| 654 | |||
| 655 | } // end namespace llvm |
||
| 656 | |||
| 657 | #endif // LLVM_CODEGEN_MACHINEPIPELINER_H |