GCNRegPressure.h

Go to the documentation of this file.

//===- GCNRegPressure.h -----------------------------------------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
///
/// \file
/// This file defines the GCNRegPressure class, which tracks registry pressure
/// by bookkeeping number of SGPR/VGPRs used, weights for large SGPR/VGPRs. It
/// also implements a compare function, which compares different register
/// pressures, and declares one with max occupancy as winner.
///
//===----------------------------------------------------------------------===//
 
#ifndef LLVM_LIB_TARGET_AMDGPU_GCNREGPRESSURE_H
#define LLVM_LIB_TARGET_AMDGPU_GCNREGPRESSURE_H
 
#include "GCNSubtarget.h"
#include "llvm/CodeGen/LiveIntervals.h"
#include "llvm/CodeGen/RegisterPressure.h"
#include <algorithm>
#include <array>
#include <functional>
 
namespace llvm {
 
class MachineRegisterInfo;
class raw_ostream;
class SlotIndex;
 
struct GCNRegPressure {
  enum RegKind { SGPR, VGPR, AGPR, AVGPR, TOTAL_KINDS };
 
  static constexpr const char *getName(RegKind Kind) {
    const char *Names[] = {"SGPR", "VGPR", "AGPR", "AVGPR"};
    assert(Kind < TOTAL_KINDS);
    return Names[Kind];
  }
 
  GCNRegPressure() {
    clear();
  }
 
  bool empty() const {
    return !Value[SGPR] && !Value[VGPR] && !Value[AGPR] && !Value[AVGPR];
  }
 
  void clear() { Value.fill(0); }
 
  unsigned getNumRegs(RegKind Kind) const {
    assert(Kind < TOTAL_KINDS);
    return Value[Kind];
  }
 
  /// \returns the SGPR32 pressure
  unsigned getSGPRNum() const { return Value[SGPR]; }
  /// \returns the aggregated ArchVGPR32, AccVGPR32, and Pseudo AVGPR pressure
  /// dependent upon \p UnifiedVGPRFile
  unsigned getVGPRNum(bool UnifiedVGPRFile) const {
    if (UnifiedVGPRFile) {
      return Value[AGPR]
                 ? getUnifiedVGPRNum(Value[VGPR], Value[AGPR], Value[AVGPR])
                 : Value[VGPR] + Value[AVGPR];
    }
    // AVGPR assignment priority is based on the width of the register. Account
    // AVGPR pressure as VGPR.
    return std::max(Value[VGPR] + Value[AVGPR], Value[AGPR]);
  }
 
  /// Returns the aggregated VGPR pressure, assuming \p NumArchVGPRs ArchVGPRs
  /// \p NumAGPRs AGPRS, and \p NumAVGPRs AVGPRs for a target with a unified
  /// VGPR file.
  inline static unsigned getUnifiedVGPRNum(unsigned NumArchVGPRs,
                                           unsigned NumAGPRs,
                                           unsigned NumAVGPRs) {
 
    // Assume AVGPRs will be assigned as VGPRs.
    return alignTo(NumArchVGPRs + NumAVGPRs,
                   AMDGPU::IsaInfo::getArchVGPRAllocGranule()) +
           NumAGPRs;
  }
 
  /// \returns the ArchVGPR32 pressure, plus the AVGPRS which we assume will be
  /// allocated as VGPR
  unsigned getArchVGPRNum() const { return Value[VGPR] + Value[AVGPR]; }
  /// \returns the AccVGPR32 pressure
  unsigned getAGPRNum() const { return Value[AGPR]; }
 
  unsigned getVGPRTuplesWeight() const {
    return std::max(Value[TOTAL_KINDS + VGPR] + Value[TOTAL_KINDS + AVGPR],
                    Value[TOTAL_KINDS + AGPR]);
  }
  unsigned getSGPRTuplesWeight() const { return Value[TOTAL_KINDS + SGPR]; }
 
  unsigned getOccupancy(const GCNSubtarget &ST,
                        unsigned DynamicVGPRBlockSize) const {
    return std::min(ST.getOccupancyWithNumSGPRs(getSGPRNum()),
                    ST.getOccupancyWithNumVGPRs(getVGPRNum(ST.hasGFX90AInsts()),
                                                DynamicVGPRBlockSize));
  }
 
  unsigned getVGPRSpills(MachineFunction &MF, unsigned ArchVGPRThreshold,
                         unsigned AGPRThreshold, unsigned CombinedThreshold) {
    const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
    if (!ST.hasGFX90AInsts())
      return 0;
 
    unsigned ArchPressure = getArchVGPRNum();
    unsigned AGPRPressure = getAGPRNum();
 
    unsigned ArchSpill = ArchPressure > ArchVGPRThreshold
                             ? (ArchPressure - ArchVGPRThreshold)
                             : 0;
    unsigned AGPRSpill =
        AGPRPressure > AGPRThreshold ? (AGPRPressure - AGPRThreshold) : 0;
 
    unsigned UnifiedPressure = getVGPRNum(/*UnifiedVGPRFile=*/true);
    unsigned UnifiedSpill = UnifiedPressure > CombinedThreshold
                                ? (UnifiedPressure - CombinedThreshold)
                                : 0;
 
    return std::max(UnifiedSpill, ArchSpill + AGPRSpill);
  }
 
  void inc(unsigned Reg,
           LaneBitmask PrevMask,
           LaneBitmask NewMask,
           const MachineRegisterInfo &MRI);
 
  /// Compares \p this GCNRegpressure to \p O, returning true if \p this is
  /// less. Since GCNRegpressure contains different types of pressures, and due
  /// to target-specific pecularities (e.g. we care about occupancy rather than
  /// raw register usage), we determine if \p this GCNRegPressure is less than
  /// \p O based on the following tiered comparisons (in order order of
  /// precedence):
  /// 1. Better occupancy
  /// 2. Less spilling (first preference to VGPR spills, then to SGPR spills)
  /// 3. Less tuple register pressure (first preference to VGPR tuples if we
  /// determine that SGPR pressure is not important)
  /// 4. Less raw register pressure (first preference to VGPR tuples if we
  /// determine that SGPR pressure is not important)
  bool less(const MachineFunction &MF, const GCNRegPressure &O,
            unsigned MaxOccupancy = std::numeric_limits<unsigned>::max()) const;
 
  bool operator==(const GCNRegPressure &O) const { return Value == O.Value; }
 
  bool operator!=(const GCNRegPressure &O) const {
    return !(*this == O);
  }
 
  GCNRegPressure &operator+=(const GCNRegPressure &RHS) {
    for (unsigned I = 0; I < ValueArraySize; ++I)
      Value[I] += RHS.Value[I];
    return *this;
  }
 
  GCNRegPressure &operator-=(const GCNRegPressure &RHS) {
    for (unsigned I = 0; I < ValueArraySize; ++I)
      Value[I] -= RHS.Value[I];
    return *this;
  }
 
  void dump() const;
 
  static RegKind getRegKind(unsigned Reg, const MachineRegisterInfo &MRI) {
    const TargetRegisterInfo *TRI = MRI.getTargetRegisterInfo();
    const SIRegisterInfo *STI = static_cast<const SIRegisterInfo *>(TRI);
    return (RegKind)getRegKind(MRI.getRegClass(Reg), STI);
  }
 
private:
  static constexpr unsigned ValueArraySize = TOTAL_KINDS * 2;
 
  /// Pressure for all register kinds (first all regular registers kinds, then
  /// all tuple register kinds).
  std::array<unsigned, ValueArraySize> Value;
 
  static unsigned getRegKind(const TargetRegisterClass *RC,
                             const SIRegisterInfo *STI);
 
  friend GCNRegPressure maxPerRegFile(const GCNRegPressure &P1,
                                      const GCNRegPressure &P2);
 
  friend Printable print(const GCNRegPressure &RP, const GCNSubtarget *ST,
                         unsigned DynamicVGPRBlockSize);
};
 
/// \returns the maximum of \p P1 and \p P2 for each register file. The VGPR and
/// AVGPR components are queried as one sum, so both come from the operand with
/// the larger sum. The AGPR maximum stays separate also for a unified VGPR
/// file: ArchVGPRs and AGPRs use separate ranges of it, so the unified pressure
/// is the sum of both peaks even when they are at different points.
inline GCNRegPressure maxPerRegFile(const GCNRegPressure &P1,
                                    const GCNRegPressure &P2) {
  GCNRegPressure Res;
  // Value holds the register counts, then the tuple weights, in the same
  // RegKind order.
  for (unsigned I = 0; I < GCNRegPressure::ValueArraySize; ++I) {
    switch (I % GCNRegPressure::TOTAL_KINDS) {
    case GCNRegPressure::VGPR: {
      unsigned AV = I - GCNRegPressure::VGPR + GCNRegPressure::AVGPR;
      const GCNRegPressure &From =
          P1.Value[I] + P1.Value[AV] >= P2.Value[I] + P2.Value[AV] ? P1 : P2;
      Res.Value[I] = From.Value[I];
      Res.Value[AV] = From.Value[AV];
      break;
    }
    case GCNRegPressure::AVGPR:
      // Already set with VGPR, which comes first in RegKind.
      break;
    default:
      Res.Value[I] = std::max(P1.Value[I], P2.Value[I]);
    }
  }
  return Res;
}
 
inline GCNRegPressure operator+(const GCNRegPressure &P1,
                                const GCNRegPressure &P2) {
  GCNRegPressure Sum = P1;
  Sum += P2;
  return Sum;
}
 
inline GCNRegPressure operator-(const GCNRegPressure &P1,
                                const GCNRegPressure &P2) {
  GCNRegPressure Diff = P1;
  Diff -= P2;
  return Diff;
}
 
////////////////////////////////////////////////////////////////////////////////
// GCNRPTarget
 
/// Models a register pressure target, allowing to evaluate and track register
/// savings against that target from a starting \ref GCNRegPressure.
class GCNRPTarget {
public:
  /// Sets up the target such that the register pressure starting at \p RP does
  /// not show register spilling on function \p MF (w.r.t. the function's
  /// mininum target occupancy).
  GCNRPTarget(const MachineFunction &MF, const GCNRegPressure &RP);
 
  /// Sets up the target such that the register pressure starting at \p RP does
  /// not use more than \p NumSGPRs SGPRs and \p NumVGPRs VGPRs on function \p
  /// MF.
  GCNRPTarget(unsigned NumSGPRs, unsigned NumVGPRs, const MachineFunction &MF,
              const GCNRegPressure &RP);
 
  /// Sets up the target such that the register pressure starting at \p RP does
  /// not prevent achieving an occupancy of at least \p Occupancy on function
  /// \p MF.
  GCNRPTarget(unsigned Occupancy, const MachineFunction &MF,
              const GCNRegPressure &RP);
 
  /// Changes the target (same semantics as constructor).
  void setTarget(unsigned NumSGPRs, unsigned NumVGPRs);
 
  const GCNRegPressure &getCurrentRP() const { return RP; }
 
  void setRP(const GCNRegPressure &NewRP) { RP = NewRP; }
 
  /// Determines whether saving virtual register \p Reg will be beneficial
  /// towards achieving the RP target.
  bool isSaveBeneficial(Register Reg) const;
 
  /// Returns whether the benefit that saving \p SaveRP represents will be
  /// beneficial towards achieving the RP target.
  bool isSaveBeneficial(const GCNRegPressure &SaveRP) const;
 
  /// Returns the benefit towards achieving the RP target that saving \p SaveRP
  /// represents, in total number of registers saved across all classes.
  unsigned getNumRegsBenefit(const GCNRegPressure &SaveRP) const;
 
  /// Saves a total pressure of \p SaveRP.
  void saveRP(const GCNRegPressure &SaveRP) {
    assert(!RP.less(MF, SaveRP) && "saving beyond current RP");
    RP -= SaveRP;
  }
 
  /// Whether \p TestRP is at or below the defined pressure target.
  bool satisfied(const GCNRegPressure &TestRP) const;
  /// Whether the current RP is at or below the defined pressure target.
  bool satisfied() const { return satisfied(RP); }
  bool hasVectorRegisterExcess() const;
 
  unsigned getMaxSGPRs() const { return MaxSGPRs; }
  unsigned getMaxVGPRs() const {
    return UnifiedRF ? MaxUnifiedVGPRs : MaxVGPRs;
  }
 
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
  friend raw_ostream &operator<<(raw_ostream &OS, const GCNRPTarget &Target) {
    OS << "Actual/Target: " << Target.RP.getSGPRNum() << '/' << Target.MaxSGPRs
       << " SGPRs, " << Target.RP.getArchVGPRNum() << '/' << Target.MaxVGPRs
       << " ArchVGPRs, " << Target.RP.getAGPRNum() << '/' << Target.MaxVGPRs
       << " AGPRs";
 
    if (Target.MaxUnifiedVGPRs) {
      OS << ", " << Target.RP.getVGPRNum(true) << '/' << Target.MaxUnifiedVGPRs
         << " VGPRs (unified)";
    }
    return OS;
  }
#endif
 
private:
  const MachineFunction &MF;
  const bool UnifiedRF;
 
  /// Current register pressure.
  GCNRegPressure RP;
 
  /// Target number of SGPRs.
  unsigned MaxSGPRs = 0;
  /// Target number of ArchVGPRs and AGPRs.
  unsigned MaxVGPRs = 0;
  /// Target number of overall VGPRs for subtargets with unified RFs. Always 0
  /// for subtargets with non-unified RFs.
  unsigned MaxUnifiedVGPRs = 0;
 
  GCNRPTarget(const GCNRegPressure &RP, const MachineFunction &MF)
      : MF(MF), UnifiedRF(MF.getSubtarget<GCNSubtarget>().hasGFX90AInsts()),
        RP(RP) {}
};
 
///////////////////////////////////////////////////////////////////////////////
// GCNRPTracker
 
class GCNRPTracker {
public:
  using LiveRegSet = DenseMap<unsigned, LaneBitmask>;
 
protected:
  LiveIntervals &LIS;
  LiveRegSet LiveRegs;
  GCNRegPressure CurPressure, MaxPressure;
  MachineBasicBlock::const_iterator LastTrackedMI;
  mutable const MachineRegisterInfo *MRI = nullptr;
  std::function<void(const GCNRegPressure &)> PressureCallback;
 
  GCNRPTracker(LiveIntervals &LIS_) : LIS(LIS_) {}
 
  /// Folds the pressure \p RP of one program point into MaxPressure and passes
  /// it to the pressure callback.
  void updateMaxPressure(const GCNRegPressure &RP) {
    MaxPressure = maxPerRegFile(MaxPressure, RP);
    if (PressureCallback)
      PressureCallback(RP);
  }
 
  /// Resets tracker before or \p After the provided \p MI, which can be a debug
  /// instruction.
  void reset(const MachineInstr &MI, bool After);
 
  /// Resets tracker at the start or \p End of the \p MBB.
  void reset(const MachineBasicBlock &MBB, bool End);
 
  /// Resets tracker at the specified slot index \p SI.
  void reset(const MachineRegisterInfo &MRI, SlotIndex SI);
 
  LaneBitmask getLastUsedLanes(Register Reg, SlotIndex Pos) const;
 
public:
  /// Resets tracker with the provided \p LiveRegs.
  void reset(const MachineRegisterInfo &MRI, const LiveRegSet &LiveRegs);
 
  // live regs for the current state
  const decltype(LiveRegs) &getLiveRegs() const { return LiveRegs; }
  const MachineInstr *getLastTrackedMI() const {
    return LastTrackedMI.isValid() ? &*LastTrackedMI : nullptr;
  }
 
  void clearMaxPressure() { MaxPressure.clear(); }
 
  const GCNRegPressure &getMaxPressure() const { return MaxPressure; }
 
  void resetMaxPressure() { MaxPressure = CurPressure; }
 
  /// Sets a function that the tracker calls with the pressure of every program
  /// point that it folds into the maximum pressure. A user can compute there a
  /// maximum that maxPerRegFile() does not keep, e.g. of one register kind.
  void setPressureCallback(std::function<void(const GCNRegPressure &)> CB) {
    PressureCallback = std::move(CB);
  }
 
  GCNRegPressure getPressure() const { return CurPressure; }
 
  decltype(LiveRegs) moveLiveRegs() {
    return std::move(LiveRegs);
  }
};
 
GCNRPTracker::LiveRegSet
getLiveRegs(SlotIndex SI, const LiveIntervals &LIS,
            const MachineRegisterInfo &MRI,
            GCNRegPressure::RegKind RegKind = GCNRegPressure::TOTAL_KINDS);
 
////////////////////////////////////////////////////////////////////////////////
// GCNUpwardRPTracker
 
class GCNUpwardRPTracker : public GCNRPTracker {
public:
  GCNUpwardRPTracker(LiveIntervals &LIS_) : GCNRPTracker(LIS_) {}
 
  using GCNRPTracker::reset;
 
  /// Resets tracker to the point just after \p MI (in program order), which can
  /// be a debug instruction.
  void reset(const MachineInstr &MI) { reset(MI, /*After=*/true); }
 
  /// Move to the state of RP just before the \p MI.
  void recede(const MachineInstr &MI);
 
  /// \p returns whether the tracker's state after receding MI corresponds
  /// to reported by LIS.
  bool isValid() const;
 
  GCNRegPressure getMaxPressureAndReset() {
    GCNRegPressure RP = MaxPressure;
    resetMaxPressure();
    return RP;
  }
};
 
////////////////////////////////////////////////////////////////////////////////
// GCNDownwardRPTracker
 
/// Common behavior for downward RP trackers. Unconstructible directly.
class GCNDownwardRPTrackerBase : public GCNRPTracker {
public:
  /// Mostly copy/paste from CodeGen/RegisterPressure.cpp
  /// Calculate the impact \p MI will have on CurPressure and \return the
  /// speculated pressure. In order to support RP Speculation, this does not
  /// rely on the implicit program ordering in the LiveIntervals.
  GCNRegPressure bumpDownwardPressure(const MachineInstr *MI,
                                      const SIRegisterInfo *TRI) const;
 
protected:
  GCNDownwardRPTrackerBase(LiveIntervals &LIS) : GCNRPTracker(LIS) {}
 
  /// Retires the lanes of \p MI's used registers that die at \p MI, mirroring
  /// RegPressureTracker::advance(). Early-clobber defs of \p MI are accounted
  /// in \ref MaxPressure as live together with all of its uses. The uses of a
  /// PHI are not retired since they do not die there.
  void retireDeadUses(const MachineInstr &MI);
 
  /// Retires the lanes defined by \p MI that are dead at \p SI, according to
  /// LIS.
  void retireDeadDefs(const MachineInstr &MI, SlotIndex SI);
 
  /// Adds all lanes (dead or not) defined by \p MI's register operands to \ref
  /// CurPressure.
  void addDefs(const MachineInstr &MI);
 
private:
  /// Drop the lanes of \p Reg that are no longer live at \p SI, decreasing
  /// CurPressure accordingly. \p Reg must be a virtual register that is
  /// currently tracked as live.
  void retireVirtReg(Register Reg, SlotIndex SI);
};
 
/// Downward RP tracker that relies on program order to advance down the
/// instruction stream.
class GCNDownwardRPTracker : public GCNDownwardRPTrackerBase {
private:
  /// Next MI to advance over.
  MachineBasicBlock::const_iterator NextMI;
  /// End iterator for basic block being advanced in.
  MachineBasicBlock::const_iterator MBBEnd;
 
public:
  GCNDownwardRPTracker(LiveIntervals &LIS) : GCNDownwardRPTrackerBase(LIS) {}
 
  using GCNDownwardRPTrackerBase::addDefs;
  using GCNRPTracker::reset;
 
  /// Returns the next MI that will be advanced over.
  MachineBasicBlock::const_iterator getNext() const { return NextMI; }
 
  /// \p return MaxPressure and clear it.
  GCNRegPressure moveMaxPressure() {
    auto Res = MaxPressure;
    MaxPressure.clear();
    return Res;
  }
 
  /// Reset tracker to the point before the \p MI (filling \ref LiveRegs upon
  /// this point using \p LiveRegsCopy if provided or LIS otherwise) point using
  /// LIS. \p End must be between the MI and the end of its parent block
  /// (inclusive). \p returns false if the range [MI, End) is empty except debug
  /// values.
  bool reset(const MachineInstr &MI, MachineBasicBlock::const_iterator End,
             const LiveRegSet *LiveRegsCopy = nullptr);
 
  /// Move to the state right before \ref NextMI or the end of the MBB. This
  /// retires defs of \ref LastTrackedMI that are dead after the instruction.
  /// Its dead uses were already retired by \ref advanceToNext. \p returns
  /// false if it reached the end of the block.
  bool advanceBeforeNext();
 
  /// Move to the state at \ref NextMI. This retires the uses that die at \p
  /// NextMI, then adds the register lanes it defines to \ref CurPressure. Has
  /// to be called after \ref advanceBeforeNext.
  void advanceToNext();
 
  /// Advance to the state at the next MI in program order. \p returns false if
  /// it reached end of block.
  bool advance();
 
  /// Advance in program order until before \p End. Returns whether it actually
  /// had to advance to reach \p End.
  bool advance(MachineBasicBlock::const_iterator End);
 
  /// Reset tracker to \p Begin (filling \ref LiveRegs upon this point using \p
  /// LiveRegsCopy if provided or LIS otherwise) and advance to \p End, which
  /// must be between \p Begin and the end of its parent block (inclusive). \p
  /// returns false if the range [Begin, End) is empty except debug values.
  bool advance(MachineBasicBlock::const_iterator Begin,
               MachineBasicBlock::const_iterator End,
               const LiveRegSet *LiveRegsCopy = nullptr);
};
 
/// Downward RP tracker that relies on the user to provide the next instruction
/// down the instruction stream. It still uses LIS for liveness queries.
class GCNManualDownwardRPTracker : public GCNDownwardRPTrackerBase {
public:
  GCNManualDownwardRPTracker(LiveIntervals &LIS)
      : GCNDownwardRPTrackerBase(LIS) {}
 
  using GCNRPTracker::reset;
 
  /// Advances to the state after \p MI.
  void advance(const MachineInstr &MI);
};
 
/// \returns the LaneMask of live lanes of \p Reg at position \p SI. Only the
/// active lanes of \p LaneMaskFilter will be set in the return value. This is
/// used, for example, to limit the live lanes to a specific subreg when
/// calculating use masks.
LaneBitmask getLiveLaneMask(unsigned Reg, SlotIndex SI,
                            const LiveIntervals &LIS,
                            const MachineRegisterInfo &MRI,
                            LaneBitmask LaneMaskFilter = LaneBitmask::getAll());
 
LaneBitmask getLiveLaneMask(const LiveInterval &LI, SlotIndex SI,
                            const MachineRegisterInfo &MRI,
                            LaneBitmask LaneMaskFilter = LaneBitmask::getAll());
 
/// creates a map MachineInstr -> LiveRegSet
/// R - range of iterators on instructions
/// After - upon entry or exit of every instruction
/// Note: there is no entry in the map for instructions with empty live reg set
/// Complexity = O(NumVirtRegs * averageLiveRangeSegmentsPerReg * lg(R))
template <typename Range>
DenseMap<MachineInstr*, GCNRPTracker::LiveRegSet>
getLiveRegMap(Range &&R, bool After, LiveIntervals &LIS) {
  std::vector<SlotIndex> Indexes;
  Indexes.reserve(llvm::size(R));
  auto &SII = *LIS.getSlotIndexes();
  for (MachineInstr *I : R) {
    auto SI = SII.getInstructionIndex(*I);
    Indexes.push_back(After ? SI.getDeadSlot() : SI.getBaseIndex());
  }
  llvm::sort(Indexes);
 
  auto &MRI = (*R.begin())->getMF()->getRegInfo();
  DenseMap<MachineInstr *, GCNRPTracker::LiveRegSet> LiveRegMap;
  SmallVector<SlotIndex, 32> LiveIdxs, SRLiveIdxs;
  for (unsigned I = 0, E = MRI.getNumVirtRegs(); I != E; ++I) {
    auto Reg = Register::index2VirtReg(I);
    if (!LIS.hasInterval(Reg))
      continue;
    auto &LI = LIS.getInterval(Reg);
    LiveIdxs.clear();
    if (!LI.findIndexesLiveAt(Indexes, std::back_inserter(LiveIdxs)))
      continue;
    if (!LI.hasSubRanges()) {
      for (auto SI : LiveIdxs)
        LiveRegMap[SII.getInstructionFromIndex(SI)][Reg] =
            MRI.getMaxLaneMaskForVReg(Reg);
    } else
      for (const auto &S : LI.subranges()) {
        // constrain search for subranges by indexes live at main range
        SRLiveIdxs.clear();
        S.findIndexesLiveAt(LiveIdxs, std::back_inserter(SRLiveIdxs));
        for (auto SI : SRLiveIdxs)
          LiveRegMap[SII.getInstructionFromIndex(SI)][Reg] |= S.LaneMask;
      }
  }
  return LiveRegMap;
}
 
inline GCNRPTracker::LiveRegSet getLiveRegsAfter(const MachineInstr &MI,
                                                 const LiveIntervals &LIS) {
  return getLiveRegs(LIS.getInstructionIndex(MI).getDeadSlot(), LIS,
                     MI.getMF()->getRegInfo());
}
 
inline GCNRPTracker::LiveRegSet getLiveRegsBefore(const MachineInstr &MI,
                                                  const LiveIntervals &LIS) {
  return getLiveRegs(LIS.getInstructionIndex(MI).getBaseIndex(), LIS,
                     MI.getMF()->getRegInfo());
}
 
template <typename Range>
GCNRegPressure getRegPressure(const MachineRegisterInfo &MRI,
                              Range &&LiveRegs) {
  GCNRegPressure Res;
  for (const auto &RM : LiveRegs)
    Res.inc(RM.first, LaneBitmask::getNone(), RM.second, MRI);
  return Res;
}
 
bool isEqual(const GCNRPTracker::LiveRegSet &S1,
             const GCNRPTracker::LiveRegSet &S2);
 
Printable print(const GCNRegPressure &RP, const GCNSubtarget *ST = nullptr,
                unsigned DynamicVGPRBlockSize = 0);
 
Printable print(const GCNRPTracker::LiveRegSet &LiveRegs,
                const MachineRegisterInfo &MRI);
 
Printable reportMismatch(const GCNRPTracker::LiveRegSet &LISLR,
                         const GCNRPTracker::LiveRegSet &TrackedL,
                         const TargetRegisterInfo *TRI, StringRef Pfx = "  ");
 
struct GCNRegPressurePrinter : public MachineFunctionPass {
  static char ID;
 
public:
  GCNRegPressurePrinter() : MachineFunctionPass(ID) {}
 
  bool runOnMachineFunction(MachineFunction &MF) override;
 
  void getAnalysisUsage(AnalysisUsage &AU) const override {
    AU.addRequired<LiveIntervalsWrapperPass>();
    AU.setPreservesAll();
    MachineFunctionPass::getAnalysisUsage(AU);
  }
};
 
LLVM_ABI void dumpMaxRegPressure(MachineFunction &MF,
                                 GCNRegPressure::RegKind Kind,
                                 LiveIntervals &LIS,
                                 const MachineLoopInfo *MLI);
 
/// Estimate VGPR pressure using greedy, non-splitting register allocation
/// simulation, accounting for live interval interference.
/// \param RegionBegin Start iterator of the region
/// \param RegionEnd End iterator of the region
/// \param LiveIns Live-in registers for the region
/// \returns estimated VGPR pressure
unsigned estimateGreedyVGPRPressure(
    MachineBasicBlock::const_iterator RegionBegin,
    MachineBasicBlock::const_iterator RegionEnd,
    const GCNRPTracker::LiveRegSet &LiveIns, const LiveIntervals &LIS,
    const MachineRegisterInfo &MRI, const SIRegisterInfo &TRI);
 
} // end namespace llvm
 
#endif // LLVM_LIB_TARGET_AMDGPU_GCNREGPRESSURE_H