| 1 | //=-- SystemZHazardRecognizer.h - SystemZ Hazard Recognizer -----*- C++ -*-===// |
| 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 | // This file declares a hazard recognizer for the SystemZ scheduler. |
| 10 | // |
| 11 | // This class is used by the SystemZ scheduling strategy to maintain |
| 12 | // the state during scheduling, and provide cost functions for |
| 13 | // scheduling candidates. This includes: |
| 14 | // |
| 15 | // * Decoder grouping. A decoder group can maximally hold 3 uops, and |
| 16 | // instructions that always begin a new group should be scheduled when |
| 17 | // the current decoder group is empty. |
| 18 | // * Processor resources usage. It is beneficial to balance the use of |
| 19 | // resources. |
| 20 | // |
| 21 | // A goal is to consider all instructions, also those outside of any |
| 22 | // scheduling region. Such instructions are "advanced" past and include |
| 23 | // single instructions before a scheduling region, branches etc. |
| 24 | // |
| 25 | // A block that has only one predecessor continues scheduling with the state |
| 26 | // of it (which may be updated by emitting branches). |
| 27 | // |
| 28 | // ===---------------------------------------------------------------------===// |
| 29 | |
| 30 | #ifndef LLVM_LIB_TARGET_SYSTEMZ_SYSTEMZHAZARDRECOGNIZER_H |
| 31 | #define LLVM_LIB_TARGET_SYSTEMZ_SYSTEMZHAZARDRECOGNIZER_H |
| 32 | |
| 33 | #include "SystemZSubtarget.h" |
| 34 | #include "llvm/ADT/SmallVector.h" |
| 35 | #include "llvm/CodeGen/MachineFunction.h" |
| 36 | #include "llvm/CodeGen/MachineScheduler.h" |
| 37 | #include "llvm/CodeGen/ScheduleHazardRecognizer.h" |
| 38 | #include "llvm/MC/MCInstrDesc.h" |
| 39 | #include "llvm/Support/raw_ostream.h" |
| 40 | #include <string> |
| 41 | |
| 42 | namespace llvm { |
| 43 | |
| 44 | /// SystemZHazardRecognizer maintains the state for one MBB during scheduling. |
| 45 | class SystemZHazardRecognizer : public ScheduleHazardRecognizer { |
| 46 | |
| 47 | const SystemZInstrInfo *TII; |
| 48 | const TargetSchedModel *SchedModel; |
| 49 | |
| 50 | /// Keep track of the number of decoder slots used in the current |
| 51 | /// decoder group. |
| 52 | unsigned CurrGroupSize; |
| 53 | |
| 54 | /// True if an instruction with four reg operands have been scheduled into |
| 55 | /// the current decoder group. |
| 56 | bool CurrGroupHas4RegOps; |
| 57 | |
| 58 | /// The tracking of resources here are quite similar to the common |
| 59 | /// code use of a critical resource. However, z13 differs in the way |
| 60 | /// that it has two processor sides which may be interesting to |
| 61 | /// model in the future (a work in progress). |
| 62 | |
| 63 | /// Counters for the number of uops scheduled per processor |
| 64 | /// resource. |
| 65 | SmallVector<int, 0> ProcResourceCounters; |
| 66 | |
| 67 | /// This is the resource with the greatest queue, which the |
| 68 | /// scheduler tries to avoid. |
| 69 | unsigned CriticalResourceIdx; |
| 70 | |
| 71 | /// Return the number of decoder slots MI requires. |
| 72 | inline unsigned getNumDecoderSlots(SUnit *SU) const; |
| 73 | |
| 74 | /// Return true if MI fits into current decoder group. |
| 75 | bool fitsIntoCurrentGroup(SUnit *SU) const; |
| 76 | |
| 77 | /// Return true if this instruction has four register operands. |
| 78 | bool has4RegOps(const MachineInstr *MI) const; |
| 79 | |
| 80 | /// Two decoder groups per cycle are formed (for z13), meaning 2x3 |
| 81 | /// instructions. This function returns a number between 0 and 5, |
| 82 | /// representing the current decoder slot of the current cycle. If an SU |
| 83 | /// is passed which will begin a new decoder group, the returned value is |
| 84 | /// the cycle index of the next group. |
| 85 | unsigned getCurrCycleIdx(SUnit *SU = nullptr) const; |
| 86 | |
| 87 | /// LastFPdOpCycleIdx stores the numbeer returned by getCurrCycleIdx() |
| 88 | /// when a stalling operation is scheduled (which uses the FPd resource). |
| 89 | unsigned LastFPdOpCycleIdx; |
| 90 | |
| 91 | /// A counter of decoder groups scheduled. |
| 92 | unsigned GrpCount; |
| 93 | |
| 94 | unsigned getCurrGroupSize() {return CurrGroupSize;}; |
| 95 | |
| 96 | /// Start next decoder group. |
| 97 | void nextGroup(); |
| 98 | |
| 99 | /// Clear all counters for processor resources. |
| 100 | void clearProcResCounters(); |
| 101 | |
| 102 | /// With the goal of alternating processor sides for stalling (FPd) |
| 103 | /// ops, return true if it seems good to schedule an FPd op next. |
| 104 | bool isFPdOpPreferred_distance(SUnit *SU) const; |
| 105 | |
| 106 | /// Last emitted instruction or nullptr. |
| 107 | MachineInstr *LastEmittedMI; |
| 108 | |
| 109 | public: |
| 110 | SystemZHazardRecognizer(const SystemZInstrInfo *tii, |
| 111 | const TargetSchedModel *SM) |
| 112 | : TII(tii), SchedModel(SM) { |
| 113 | Reset(); |
| 114 | } |
| 115 | |
| 116 | HazardType getHazardType(SUnit *SU, int Stalls = 0) override; |
| 117 | void Reset() override; |
| 118 | void EmitInstruction(SUnit *SU) override; |
| 119 | |
| 120 | /// Resolves and cache a resolved scheduling class for an SUnit. |
| 121 | const MCSchedClassDesc *getSchedClass(SUnit *SU) const { |
| 122 | if (!SU->SchedClass && SchedModel->hasInstrSchedModel()) |
| 123 | SU->SchedClass = SchedModel->resolveSchedClass(MI: SU->getInstr()); |
| 124 | return SU->SchedClass; |
| 125 | } |
| 126 | |
| 127 | /// Wrap a non-scheduled instruction in an SU and emit it. |
| 128 | void emitInstruction(MachineInstr *MI, bool TakenBranch = false); |
| 129 | |
| 130 | // Cost functions used by SystemZPostRASchedStrategy while |
| 131 | // evaluating candidates. |
| 132 | |
| 133 | /// Return the cost of decoder grouping for SU. If SU must start a |
| 134 | /// new decoder group, this is negative if this fits the schedule or |
| 135 | /// positive if it would mean ending a group prematurely. For normal |
| 136 | /// instructions this returns 0. |
| 137 | int groupingCost(SUnit *SU) const; |
| 138 | |
| 139 | /// Return the cost of SU in regards to processor resources usage. |
| 140 | /// A positive value means it would be better to wait with SU, while |
| 141 | /// a negative value means it would be good to schedule SU next. |
| 142 | int resourcesCost(SUnit *SU); |
| 143 | |
| 144 | #ifndef NDEBUG |
| 145 | // Debug dumping. |
| 146 | std::string CurGroupDbg; // current group as text |
| 147 | void dumpSU(SUnit *SU, raw_ostream &OS) const; |
| 148 | void dumpCurrGroup(std::string Msg = "" ) const; |
| 149 | void dumpProcResourceCounters() const; |
| 150 | void dumpState() const; |
| 151 | #endif |
| 152 | |
| 153 | MachineBasicBlock::iterator getLastEmittedMI() { return LastEmittedMI; } |
| 154 | |
| 155 | /// Copy counters from end of single predecessor. |
| 156 | void copyState(SystemZHazardRecognizer *Incoming); |
| 157 | }; |
| 158 | |
| 159 | } // namespace llvm |
| 160 | |
| 161 | #endif /* LLVM_LIB_TARGET_SYSTEMZ_SYSTEMZHAZARDRECOGNIZER_H */ |
| 162 | |