/build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/lib/CodeGen/RegisterCoalescer.cpp

Bug Summary

File:	llvm/lib/CodeGen/RegisterCoalescer.cpp
Warning:	line 2586, column 11 Called C++ object pointer is null

Annotated Source Code

Press '?' to see keyboard shortcuts

Show analyzer invocation

clang -cc1 -cc1 -triple x86_64-pc-linux-gnu -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name RegisterCoalescer.cpp -analyzer-store=region -analyzer-opt-analyze-nested-blocks -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=cplusplus -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -setup-static-analyzer -analyzer-config-compatibility-mode=true -mrelocation-model pic -pic-level 2 -mframe-pointer=none -fmath-errno -fno-rounding-math -mconstructor-aliases -munwind-tables -target-cpu x86-64 -tune-cpu generic -debugger-tuning=gdb -ffunction-sections -fdata-sections -fcoverage-compilation-dir=/build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/build-llvm/lib/CodeGen -resource-dir /usr/lib/llvm-14/lib/clang/14.0.0 -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/build-llvm/lib/CodeGen -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/lib/CodeGen -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/build-llvm/include -I /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/include -D NDEBUG -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/x86_64-linux-gnu/c++/10 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../include/c++/10/backward -internal-isystem /usr/lib/llvm-14/lib/clang/14.0.0/include -internal-isystem /usr/local/include -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/10/../../../../x86_64-linux-gnu/include -internal-externc-isystem /usr/include/x86_64-linux-gnu -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -Wno-unused-parameter -Wwrite-strings -Wno-missing-field-initializers -Wno-long-long -Wno-maybe-uninitialized -Wno-class-memaccess -Wno-redundant-move -Wno-pessimizing-move -Wno-noexcept-type -Wno-comment -std=c++14 -fdeprecated-macro -fdebug-compilation-dir=/build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/build-llvm/lib/CodeGen -fdebug-prefix-map=/build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e=. -ferror-limit 19 -fvisibility-inlines-hidden -stack-protector 2 -fgnuc-version=4.2.1 -vectorize-loops -vectorize-slp -analyzer-output=html -analyzer-config stable-report-filename=true -faddrsig -D__GCC_HAVE_DWARF2_CFI_ASM=1 -o /tmp/scan-build-2021-09-04-040900-46481-1 -x c++ /build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/lib/CodeGen/RegisterCoalescer.cpp

/build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/lib/CodeGen/RegisterCoalescer.cpp

→

1//===- RegisterCoalescer.cpp - Generic Register Coalescing Interface ------===//
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 implements the generic RegisterCoalescer interface which
10// is used as the common interface used by all clients and
11// implementations of register coalescing.
12//
13//===----------------------------------------------------------------------===//

15#include "RegisterCoalescer.h"
16#include "llvm/ADT/ArrayRef.h"
17#include "llvm/ADT/BitVector.h"
18#include "llvm/ADT/DenseSet.h"
19#include "llvm/ADT/STLExtras.h"
20#include "llvm/ADT/SmallPtrSet.h"
21#include "llvm/ADT/SmallVector.h"
22#include "llvm/ADT/Statistic.h"
23#include "llvm/Analysis/AliasAnalysis.h"
24#include "llvm/CodeGen/LiveInterval.h"
25#include "llvm/CodeGen/LiveIntervals.h"
26#include "llvm/CodeGen/LiveRangeEdit.h"
27#include "llvm/CodeGen/MachineBasicBlock.h"
28#include "llvm/CodeGen/MachineFunction.h"
29#include "llvm/CodeGen/MachineFunctionPass.h"
30#include "llvm/CodeGen/MachineInstr.h"
31#include "llvm/CodeGen/MachineInstrBuilder.h"
32#include "llvm/CodeGen/MachineLoopInfo.h"
33#include "llvm/CodeGen/MachineOperand.h"
34#include "llvm/CodeGen/MachineRegisterInfo.h"
35#include "llvm/CodeGen/Passes.h"
36#include "llvm/CodeGen/RegisterClassInfo.h"
37#include "llvm/CodeGen/SlotIndexes.h"
38#include "llvm/CodeGen/TargetInstrInfo.h"
39#include "llvm/CodeGen/TargetOpcodes.h"
40#include "llvm/CodeGen/TargetRegisterInfo.h"
41#include "llvm/CodeGen/TargetSubtargetInfo.h"
42#include "llvm/IR/DebugLoc.h"
43#include "llvm/InitializePasses.h"
44#include "llvm/MC/LaneBitmask.h"
45#include "llvm/MC/MCInstrDesc.h"
46#include "llvm/MC/MCRegisterInfo.h"
47#include "llvm/Pass.h"
48#include "llvm/Support/CommandLine.h"
49#include "llvm/Support/Compiler.h"
50#include "llvm/Support/Debug.h"
51#include "llvm/Support/ErrorHandling.h"
52#include "llvm/Support/raw_ostream.h"
53#include <algorithm>
54#include <cassert>
55#include <iterator>
56#include <limits>
57#include <tuple>
58#include <utility>
59#include <vector>

61using namespace llvm;

63#define DEBUG_TYPE"regalloc" "regalloc"

65STATISTIC(numJoins    , "Number of interval joins performed")static llvm::Statistic numJoins = {"regalloc", "numJoins", "Number of interval joins performed"
};
66STATISTIC(numCrossRCs , "Number of cross class joins performed")static llvm::Statistic numCrossRCs = {"regalloc", "numCrossRCs"
, "Number of cross class joins performed"};
67STATISTIC(numCommutes , "Number of instruction commuting performed")static llvm::Statistic numCommutes = {"regalloc", "numCommutes"
, "Number of instruction commuting performed"};
68STATISTIC(numExtends  , "Number of copies extended")static llvm::Statistic numExtends = {"regalloc", "numExtends"
, "Number of copies extended"};
69STATISTIC(NumReMats   , "Number of instructions re-materialized")static llvm::Statistic NumReMats = {"regalloc", "NumReMats", "Number of instructions re-materialized"
};
70STATISTIC(NumInflated , "Number of register classes inflated")static llvm::Statistic NumInflated = {"regalloc", "NumInflated"
, "Number of register classes inflated"};
71STATISTIC(NumLaneConflicts, "Number of dead lane conflicts tested")static llvm::Statistic NumLaneConflicts = {"regalloc", "NumLaneConflicts"
, "Number of dead lane conflicts tested"};
72STATISTIC(NumLaneResolves,  "Number of dead lane conflicts resolved")static llvm::Statistic NumLaneResolves = {"regalloc", "NumLaneResolves"
, "Number of dead lane conflicts resolved"};
73STATISTIC(NumShrinkToUses,  "Number of shrinkToUses called")static llvm::Statistic NumShrinkToUses = {"regalloc", "NumShrinkToUses"
, "Number of shrinkToUses called"};

75static cl::opt<bool> EnableJoining("join-liveintervals",
                                 cl::desc("Coalesce copies (default=true)"),
                                 cl::init(true), cl::Hidden);

79static cl::opt<bool> UseTerminalRule("terminal-rule",
                                   cl::desc("Apply the terminal rule"),
                                   cl::init(false), cl::Hidden);

83/// Temporary flag to test critical edge unsplitting.
84static cl::opt<bool>
85EnableJoinSplits("join-splitedges",
cl::desc("Coalesce copies on split edges (default=subtarget)"), cl::Hidden);

88/// Temporary flag to test global copy optimization.
89static cl::opt<cl::boolOrDefault>
90EnableGlobalCopies("join-globalcopies",
cl::desc("Coalesce copies that span blocks (default=subtarget)"),
cl::init(cl::BOU_UNSET), cl::Hidden);

94static cl::opt<bool>
95VerifyCoalescing("verify-coalescing",
       cl::desc("Verify machine instrs before and after register coalescing"),
       cl::Hidden);

99static cl::opt<unsigned> LateRematUpdateThreshold(
  "late-remat-update-threshold", cl::Hidden,
  cl::desc("During rematerialization for a copy, if the def instruction has "
           "many other copy uses to be rematerialized, delay the multiple "
           "separate live interval update work and do them all at once after "
           "all those rematerialization are done. It will save a lot of "
           "repeated work. "),
  cl::init(100));

108static cl::opt<unsigned> LargeIntervalSizeThreshold(
  "large-interval-size-threshold", cl::Hidden,
  cl::desc("If the valnos size of an interval is larger than the threshold, "
           "it is regarded as a large interval. "),
  cl::init(100));

114static cl::opt<unsigned> LargeIntervalFreqThreshold(
  "large-interval-freq-threshold", cl::Hidden,
  cl::desc("For a large interval, if it is coalesed with other live "
           "intervals many times more than the threshold, stop its "
           "coalescing to control the compile time. "),
  cl::init(100));

121namespace {

class JoinVals;

class RegisterCoalescer : public MachineFunctionPass,
                          private LiveRangeEdit::Delegate {
  MachineFunction* MF = nullptr;
  MachineRegisterInfo* MRI = nullptr;
  const TargetRegisterInfo* TRI = nullptr;
  const TargetInstrInfo* TII = nullptr;
  LiveIntervals *LIS = nullptr;
  const MachineLoopInfo* Loops = nullptr;
  AliasAnalysis *AA = nullptr;
  RegisterClassInfo RegClassInfo;

  /// Position and VReg of a PHI instruction during coalescing.
  struct PHIValPos {
    SlotIndex SI;    ///< Slot where this PHI occurs.
    Register Reg;    ///< VReg the PHI occurs in.
    unsigned SubReg; ///< Qualifying subregister for Reg.
  };

  /// Map from debug instruction number to PHI position during coalescing.
  DenseMap<unsigned, PHIValPos> PHIValToPos;
  /// Index of, for each VReg, which debug instruction numbers and
  /// corresponding PHIs are sensitive to coalescing. Each VReg may have
  /// multiple PHI defs, at different positions.
  DenseMap<Register, SmallVector<unsigned, 2>> RegToPHIIdx;

  /// Debug variable location tracking -- for each VReg, maintain an
  /// ordered-by-slot-index set of DBG_VALUEs, to help quick
  /// identification of whether coalescing may change location validity.
  using DbgValueLoc = std::pair<SlotIndex, MachineInstr*>;
  DenseMap<Register, std::vector<DbgValueLoc>> DbgVRegToValues;

  /// VRegs may be repeatedly coalesced, and have many DBG_VALUEs attached.
  /// To avoid repeatedly merging sets of DbgValueLocs, instead record
  /// which vregs have been coalesced, and where to. This map is from
  /// vreg => {set of vregs merged in}.
  DenseMap<Register, SmallVector<Register, 4>> DbgMergedVRegNums;

  /// A LaneMask to remember on which subregister live ranges we need to call
  /// shrinkToUses() later.
  LaneBitmask ShrinkMask;

  /// True if the main range of the currently coalesced intervals should be
  /// checked for smaller live intervals.
  bool ShrinkMainRange = false;

  /// True if the coalescer should aggressively coalesce global copies
  /// in favor of keeping local copies.
  bool JoinGlobalCopies = false;

  /// True if the coalescer should aggressively coalesce fall-thru
  /// blocks exclusively containing copies.
  bool JoinSplitEdges = false;

  /// Copy instructions yet to be coalesced.
  SmallVector<MachineInstr*, 8> WorkList;
  SmallVector<MachineInstr*, 8> LocalWorkList;

  /// Set of instruction pointers that have been erased, and
  /// that may be present in WorkList.
  SmallPtrSet<MachineInstr*, 8> ErasedInstrs;

  /// Dead instructions that are about to be deleted.
  SmallVector<MachineInstr*, 8> DeadDefs;

  /// Virtual registers to be considered for register class inflation.
  SmallVector<Register, 8> InflateRegs;

  /// The collection of live intervals which should have been updated
  /// immediately after rematerialiation but delayed until
  /// lateLiveIntervalUpdate is called.
  DenseSet<Register> ToBeUpdated;

  /// Record how many times the large live interval with many valnos
  /// has been tried to join with other live interval.
  DenseMap<Register, unsigned long> LargeLIVisitCounter;

  /// Recursively eliminate dead defs in DeadDefs.
  void eliminateDeadDefs();

  /// allUsesAvailableAt - Return true if all registers used by OrigMI at
  /// OrigIdx are also available with the same value at UseIdx.
  bool allUsesAvailableAt(const MachineInstr *OrigMI, SlotIndex OrigIdx,
                          SlotIndex UseIdx);

  /// LiveRangeEdit callback for eliminateDeadDefs().
  void LRE_WillEraseInstruction(MachineInstr *MI) override;

  /// Coalesce the LocalWorkList.
  void coalesceLocals();

  /// Join compatible live intervals
  void joinAllIntervals();

  /// Coalesce copies in the specified MBB, putting
  /// copies that cannot yet be coalesced into WorkList.
  void copyCoalesceInMBB(MachineBasicBlock *MBB);

  /// Tries to coalesce all copies in CurrList. Returns true if any progress
  /// was made.
  bool copyCoalesceWorkList(MutableArrayRef<MachineInstr*> CurrList);

  /// If one def has many copy like uses, and those copy uses are all
  /// rematerialized, the live interval update needed for those
  /// rematerializations will be delayed and done all at once instead
  /// of being done multiple times. This is to save compile cost because
  /// live interval update is costly.
  void lateLiveIntervalUpdate();

  /// Check if the incoming value defined by a COPY at \p SLRQ in the subrange
  /// has no value defined in the predecessors. If the incoming value is the
  /// same as defined by the copy itself, the value is considered undefined.
  bool copyValueUndefInPredecessors(LiveRange &S,
                                    const MachineBasicBlock *MBB,
                                    LiveQueryResult SLRQ);

  /// Set necessary undef flags on subregister uses after pruning out undef
  /// lane segments from the subrange.
  void setUndefOnPrunedSubRegUses(LiveInterval &LI, Register Reg,
                                  LaneBitmask PrunedLanes);

  /// Attempt to join intervals corresponding to SrcReg/DstReg, which are the
  /// src/dst of the copy instruction CopyMI.  This returns true if the copy
  /// was successfully coalesced away. If it is not currently possible to
  /// coalesce this interval, but it may be possible if other things get
  /// coalesced, then it returns true by reference in 'Again'.
  bool joinCopy(MachineInstr *CopyMI, bool &Again);

  /// Attempt to join these two intervals.  On failure, this
  /// returns false.  The output "SrcInt" will not have been modified, so we
  /// can use this information below to update aliases.
  bool joinIntervals(CoalescerPair &CP);

  /// Attempt joining two virtual registers. Return true on success.
  bool joinVirtRegs(CoalescerPair &CP);

  /// If a live interval has many valnos and is coalesced with other
  /// live intervals many times, we regard such live interval as having
  /// high compile time cost.
  bool isHighCostLiveInterval(LiveInterval &LI);

  /// Attempt joining with a reserved physreg.
  bool joinReservedPhysReg(CoalescerPair &CP);

  /// Add the LiveRange @p ToMerge as a subregister liverange of @p LI.
  /// Subranges in @p LI which only partially interfere with the desired
  /// LaneMask are split as necessary. @p LaneMask are the lanes that
  /// @p ToMerge will occupy in the coalescer register. @p LI has its subrange
  /// lanemasks already adjusted to the coalesced register.
  void mergeSubRangeInto(LiveInterval &LI, const LiveRange &ToMerge,
                         LaneBitmask LaneMask, CoalescerPair &CP,
                         unsigned DstIdx);

  /// Join the liveranges of two subregisters. Joins @p RRange into
  /// @p LRange, @p RRange may be invalid afterwards.
  void joinSubRegRanges(LiveRange &LRange, LiveRange &RRange,
                        LaneBitmask LaneMask, const CoalescerPair &CP);

  /// We found a non-trivially-coalescable copy. If the source value number is
  /// defined by a copy from the destination reg see if we can merge these two
  /// destination reg valno# into a single value number, eliminating a copy.
  /// This returns true if an interval was modified.
  bool adjustCopiesBackFrom(const CoalescerPair &CP, MachineInstr *CopyMI);

  /// Return true if there are definitions of IntB
  /// other than BValNo val# that can reach uses of AValno val# of IntA.
  bool hasOtherReachingDefs(LiveInterval &IntA, LiveInterval &IntB,
                            VNInfo *AValNo, VNInfo *BValNo);

  /// We found a non-trivially-coalescable copy.
  /// If the source value number is defined by a commutable instruction and
  /// its other operand is coalesced to the copy dest register, see if we
  /// can transform the copy into a noop by commuting the definition.
  /// This returns a pair of two flags:
  /// - the first element is true if an interval was modified,
  /// - the second element is true if the destination interval needs
  ///   to be shrunk after deleting the copy.
  std::pair<bool,bool> removeCopyByCommutingDef(const CoalescerPair &CP,
                                                MachineInstr *CopyMI);

  /// We found a copy which can be moved to its less frequent predecessor.
  bool removePartialRedundancy(const CoalescerPair &CP, MachineInstr &CopyMI);

  /// If the source of a copy is defined by a
  /// trivial computation, replace the copy by rematerialize the definition.
  bool reMaterializeTrivialDef(const CoalescerPair &CP, MachineInstr *CopyMI,
                               bool &IsDefCopy);

  /// Return true if a copy involving a physreg should be joined.
  bool canJoinPhys(const CoalescerPair &CP);

  /// Replace all defs and uses of SrcReg to DstReg and update the subregister
  /// number if it is not zero. If DstReg is a physical register and the
  /// existing subregister number of the def / use being updated is not zero,
  /// make sure to set it to the correct physical subregister.
  void updateRegDefsUses(Register SrcReg, Register DstReg, unsigned SubIdx);

  /// If the given machine operand reads only undefined lanes add an undef
  /// flag.
  /// This can happen when undef uses were previously concealed by a copy
  /// which we coalesced. Example:
  ///    %0:sub0<def,read-undef> = ...
  ///    %1 = COPY %0           <-- Coalescing COPY reveals undef
  ///       = use %1:sub1       <-- hidden undef use
  void addUndefFlag(const LiveInterval &Int, SlotIndex UseIdx,
                    MachineOperand &MO, unsigned SubRegIdx);

  /// Handle copies of undef values. If the undef value is an incoming
  /// PHI value, it will convert @p CopyMI to an IMPLICIT_DEF.
  /// Returns nullptr if @p CopyMI was not in any way eliminable. Otherwise,
  /// it returns @p CopyMI (which could be an IMPLICIT_DEF at this point).
  MachineInstr *eliminateUndefCopy(MachineInstr *CopyMI);

  /// Check whether or not we should apply the terminal rule on the
  /// destination (Dst) of \p Copy.
  /// When the terminal rule applies, Copy is not profitable to
  /// coalesce.
  /// Dst is terminal if it has exactly one affinity (Dst, Src) and
  /// at least one interference (Dst, Dst2). If Dst is terminal, the
  /// terminal rule consists in checking that at least one of
  /// interfering node, say Dst2, has an affinity of equal or greater
  /// weight with Src.
  /// In that case, Dst2 and Dst will not be able to be both coalesced
  /// with Src. Since Dst2 exposes more coalescing opportunities than
  /// Dst, we can drop \p Copy.
  bool applyTerminalRule(const MachineInstr &Copy) const;

  /// Wrapper method for \see LiveIntervals::shrinkToUses.
  /// This method does the proper fixing of the live-ranges when the afore
  /// mentioned method returns true.
  void shrinkToUses(LiveInterval *LI,
                    SmallVectorImpl<MachineInstr * > *Dead = nullptr) {
    NumShrinkToUses++;
    if (LIS->shrinkToUses(LI, Dead)) {
      /// Check whether or not \p LI is composed by multiple connected
      /// components and if that is the case, fix that.
      SmallVector<LiveInterval*, 8> SplitLIs;
      LIS->splitSeparateComponents(*LI, SplitLIs);
    }
  }

  /// Wrapper Method to do all the necessary work when an Instruction is
  /// deleted.
  /// Optimizations should use this to make sure that deleted instructions
  /// are always accounted for.
  void deleteInstr(MachineInstr* MI) {
    ErasedInstrs.insert(MI);
    LIS->RemoveMachineInstrFromMaps(*MI);
    MI->eraseFromParent();
  }

  /// Walk over function and initialize the DbgVRegToValues map.
  void buildVRegToDbgValueMap(MachineFunction &MF);

  /// Test whether, after merging, any DBG_VALUEs would refer to a
  /// different value number than before merging, and whether this can
  /// be resolved. If not, mark the DBG_VALUE as being undef.
  void checkMergingChangesDbgValues(CoalescerPair &CP, LiveRange &LHS,
                                    JoinVals &LHSVals, LiveRange &RHS,
                                    JoinVals &RHSVals);

  void checkMergingChangesDbgValuesImpl(Register Reg, LiveRange &OtherRange,
                                        LiveRange &RegRange, JoinVals &Vals2);

public:
  static char ID; ///< Class identification, replacement for typeinfo

  RegisterCoalescer() : MachineFunctionPass(ID) {
    initializeRegisterCoalescerPass(*PassRegistry::getPassRegistry());
  }

  void getAnalysisUsage(AnalysisUsage &AU) const override;

  void releaseMemory() override;

  /// This is the pass entry point.
  bool runOnMachineFunction(MachineFunction&) override;

  /// Implement the dump method.
  void print(raw_ostream &O, const Module* = nullptr) const override;
};

406} // end anonymous namespace

408char RegisterCoalescer::ID = 0;

410char &llvm::RegisterCoalescerID = RegisterCoalescer::ID;

412INITIALIZE_PASS_BEGIN(RegisterCoalescer, "simple-register-coalescing",static void *initializeRegisterCoalescerPassOnce(PassRegistry
 &Registry) {
                    "Simple Register Coalescing", false, false)static void *initializeRegisterCoalescerPassOnce(PassRegistry
 &Registry) {
414INITIALIZE_PASS_DEPENDENCY(LiveIntervals)initializeLiveIntervalsPass(Registry);
415INITIALIZE_PASS_DEPENDENCY(SlotIndexes)initializeSlotIndexesPass(Registry);
416INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)initializeMachineLoopInfoPass(Registry);
417INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)initializeAAResultsWrapperPassPass(Registry);
418INITIALIZE_PASS_END(RegisterCoalescer, "simple-register-coalescing",PassInfo *PI = new PassInfo( "Simple Register Coalescing", "simple-register-coalescing"
, &RegisterCoalescer::ID, PassInfo::NormalCtor_t(callDefaultCtor
<RegisterCoalescer>), false, false); Registry.registerPass
(*PI, true); return PI; } static llvm::once_flag InitializeRegisterCoalescerPassFlag
; void llvm::initializeRegisterCoalescerPass(PassRegistry &
Registry) { llvm::call_once(InitializeRegisterCoalescerPassFlag
, initializeRegisterCoalescerPassOnce, std::ref(Registry)); }
                  "Simple Register Coalescing", false, false)PassInfo *PI = new PassInfo( "Simple Register Coalescing", "simple-register-coalescing"
, &RegisterCoalescer::ID, PassInfo::NormalCtor_t(callDefaultCtor
<RegisterCoalescer>), false, false); Registry.registerPass
(*PI, true); return PI; } static llvm::once_flag InitializeRegisterCoalescerPassFlag
; void llvm::initializeRegisterCoalescerPass(PassRegistry &
Registry) { llvm::call_once(InitializeRegisterCoalescerPassFlag
, initializeRegisterCoalescerPassOnce, std::ref(Registry)); }

421LLVM_NODISCARD[[clang::warn_unused_result]] static bool isMoveInstr(const TargetRegisterInfo &tri,
                                     const MachineInstr *MI, Register &Src,
                                     Register &Dst, unsigned &SrcSub,
                                     unsigned &DstSub) {
if (MI->isCopy()) {
  Dst = MI->getOperand(0).getReg();
  DstSub = MI->getOperand(0).getSubReg();
  Src = MI->getOperand(1).getReg();
  SrcSub = MI->getOperand(1).getSubReg();
} else if (MI->isSubregToReg()) {
  Dst = MI->getOperand(0).getReg();
  DstSub = tri.composeSubRegIndices(MI->getOperand(0).getSubReg(),
                                    MI->getOperand(3).getImm());
  Src = MI->getOperand(2).getReg();
  SrcSub = MI->getOperand(2).getSubReg();
} else
  return false;
return true;
439}

441/// Return true if this block should be vacated by the coalescer to eliminate
442/// branches. The important cases to handle in the coalescer are critical edges
443/// split during phi elimination which contain only copies. Simple blocks that
444/// contain non-branches should also be vacated, but this can be handled by an
445/// earlier pass similar to early if-conversion.
446static bool isSplitEdge(const MachineBasicBlock *MBB) {
if (MBB->pred_size() != 1 || MBB->succ_size() != 1)
  return false;

for (const auto &MI : *MBB) {
  if (!MI.isCopyLike() && !MI.isUnconditionalBranch())
    return false;
}
return true;
455}

457bool CoalescerPair::setRegisters(const MachineInstr *MI) {
SrcReg = DstReg = Register();
SrcIdx = DstIdx = 0;
NewRC = nullptr;
Flipped = CrossClass = false;

Register Src, Dst;
unsigned SrcSub = 0, DstSub = 0;
if (!isMoveInstr(TRI, MI, Src, Dst, SrcSub, DstSub))
  return false;
Partial = SrcSub || DstSub;

// If one register is a physreg, it must be Dst.
if (Register::isPhysicalRegister(Src)) {
  if (Register::isPhysicalRegister(Dst))
    return false;
  std::swap(Src, Dst);
  std::swap(SrcSub, DstSub);
  Flipped = true;
}

const MachineRegisterInfo &MRI = MI->getMF()->getRegInfo();

if (Register::isPhysicalRegister(Dst)) {
  // Eliminate DstSub on a physreg.
  if (DstSub) {
    Dst = TRI.getSubReg(Dst, DstSub);
    if (!Dst) return false;
    DstSub = 0;
  }

  // Eliminate SrcSub by picking a corresponding Dst superregister.
  if (SrcSub) {
    Dst = TRI.getMatchingSuperReg(Dst, SrcSub, MRI.getRegClass(Src));
    if (!Dst) return false;
  } else if (!MRI.getRegClass(Src)->contains(Dst)) {
    return false;
  }
} else {
  // Both registers are virtual.
  const TargetRegisterClass *SrcRC = MRI.getRegClass(Src);
  const TargetRegisterClass *DstRC = MRI.getRegClass(Dst);

  // Both registers have subreg indices.
  if (SrcSub && DstSub) {
    // Copies between different sub-registers are never coalescable.
    if (Src == Dst && SrcSub != DstSub)
      return false;

    NewRC = TRI.getCommonSuperRegClass(SrcRC, SrcSub, DstRC, DstSub,
                                       SrcIdx, DstIdx);
    if (!NewRC)
      return false;
  } else if (DstSub) {
    // SrcReg will be merged with a sub-register of DstReg.
    SrcIdx = DstSub;
    NewRC = TRI.getMatchingSuperRegClass(DstRC, SrcRC, DstSub);
  } else if (SrcSub) {
    // DstReg will be merged with a sub-register of SrcReg.
    DstIdx = SrcSub;
    NewRC = TRI.getMatchingSuperRegClass(SrcRC, DstRC, SrcSub);
  } else {
    // This is a straight copy without sub-registers.
    NewRC = TRI.getCommonSubClass(DstRC, SrcRC);
  }

  // The combined constraint may be impossible to satisfy.
  if (!NewRC)
    return false;

  // Prefer SrcReg to be a sub-register of DstReg.
  // FIXME: Coalescer should support subregs symmetrically.
  if (DstIdx && !SrcIdx) {
    std::swap(Src, Dst);
    std::swap(SrcIdx, DstIdx);
    Flipped = !Flipped;
  }

  CrossClass = NewRC != DstRC || NewRC != SrcRC;
}
// Check our invariants
assert(Register::isVirtualRegister(Src) && "Src must be virtual")(static_cast<void> (0));
assert(!(Register::isPhysicalRegister(Dst) && DstSub) &&(static_cast<void> (0))
       "Cannot have a physical SubIdx")(static_cast<void> (0));
SrcReg = Src;
DstReg = Dst;
return true;
544}

546bool CoalescerPair::flip() {
if (Register::isPhysicalRegister(DstReg))
  return false;
std::swap(SrcReg, DstReg);
std::swap(SrcIdx, DstIdx);
Flipped = !Flipped;
return true;
553}

555bool CoalescerPair::isCoalescable(const MachineInstr *MI) const {
if (!MI)
  return false;
Register Src, Dst;
unsigned SrcSub = 0, DstSub = 0;
if (!isMoveInstr(TRI, MI, Src, Dst, SrcSub, DstSub))
  return false;

// Find the virtual register that is SrcReg.
if (Dst == SrcReg) {
  std::swap(Src, Dst);
  std::swap(SrcSub, DstSub);
} else if (Src != SrcReg) {
  return false;
}

// Now check that Dst matches DstReg.
if (DstReg.isPhysical()) {
  if (!Dst.isPhysical())
    return false;
  assert(!DstIdx && !SrcIdx && "Inconsistent CoalescerPair state.")(static_cast<void> (0));
  // DstSub could be set for a physreg from INSERT_SUBREG.
  if (DstSub)
    Dst = TRI.getSubReg(Dst, DstSub);
  // Full copy of Src.
  if (!SrcSub)
    return DstReg == Dst;
  // This is a partial register copy. Check that the parts match.
  return Register(TRI.getSubReg(DstReg, SrcSub)) == Dst;
} else {
  // DstReg is virtual.
  if (DstReg != Dst)
    return false;
  // Registers match, do the subregisters line up?
  return TRI.composeSubRegIndices(SrcIdx, SrcSub) ==
         TRI.composeSubRegIndices(DstIdx, DstSub);
}
592}

594void RegisterCoalescer::getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesCFG();
AU.addRequired<AAResultsWrapperPass>();
AU.addRequired<LiveIntervals>();
AU.addPreserved<LiveIntervals>();
AU.addPreserved<SlotIndexes>();
AU.addRequired<MachineLoopInfo>();
AU.addPreserved<MachineLoopInfo>();
AU.addPreservedID(MachineDominatorsID);
MachineFunctionPass::getAnalysisUsage(AU);
604}

606void RegisterCoalescer::eliminateDeadDefs() {
SmallVector<Register, 8> NewRegs;
LiveRangeEdit(nullptr, NewRegs, *MF, *LIS,
              nullptr, this).eliminateDeadDefs(DeadDefs);
610}

612bool RegisterCoalescer::allUsesAvailableAt(const MachineInstr *OrigMI,
                                         SlotIndex OrigIdx,
                                         SlotIndex UseIdx) {
SmallVector<Register, 8> NewRegs;
return LiveRangeEdit(nullptr, NewRegs, *MF, *LIS, nullptr, this)
    .allUsesAvailableAt(OrigMI, OrigIdx, UseIdx);
618}

620void RegisterCoalescer::LRE_WillEraseInstruction(MachineInstr *MI) {
// MI may be in WorkList. Make sure we don't visit it.
ErasedInstrs.insert(MI);
623}

625bool RegisterCoalescer::adjustCopiesBackFrom(const CoalescerPair &CP,
                                           MachineInstr *CopyMI) {
assert(!CP.isPartial() && "This doesn't work for partial copies.")(static_cast<void> (0));
assert(!CP.isPhys() && "This doesn't work for physreg copies.")(static_cast<void> (0));

LiveInterval &IntA =
  LIS->getInterval(CP.isFlipped() ? CP.getDstReg() : CP.getSrcReg());
LiveInterval &IntB =
  LIS->getInterval(CP.isFlipped() ? CP.getSrcReg() : CP.getDstReg());
SlotIndex CopyIdx = LIS->getInstructionIndex(*CopyMI).getRegSlot();

// We have a non-trivially-coalescable copy with IntA being the source and
// IntB being the dest, thus this defines a value number in IntB.  If the
// source value number (in IntA) is defined by a copy from B, see if we can
// merge these two pieces of B into a single value number, eliminating a copy.
// For example:
//
//  A3 = B0
//    ...
//  B1 = A3      <- this copy
//
// In this case, B0 can be extended to where the B1 copy lives, allowing the
// B1 value number to be replaced with B0 (which simplifies the B
// liveinterval).

// BValNo is a value number in B that is defined by a copy from A.  'B1' in
// the example above.
LiveInterval::iterator BS = IntB.FindSegmentContaining(CopyIdx);
if (BS == IntB.end()) return false;
VNInfo *BValNo = BS->valno;

// Get the location that B is defined at.  Two options: either this value has
// an unknown definition point or it is defined at CopyIdx.  If unknown, we
// can't process it.
if (BValNo->def != CopyIdx) return false;

// AValNo is the value number in A that defines the copy, A3 in the example.
SlotIndex CopyUseIdx = CopyIdx.getRegSlot(true);
LiveInterval::iterator AS = IntA.FindSegmentContaining(CopyUseIdx);
// The live segment might not exist after fun with physreg coalescing.
if (AS == IntA.end()) return false;
VNInfo *AValNo = AS->valno;

// If AValNo is defined as a copy from IntB, we can potentially process this.
// Get the instruction that defines this value number.
MachineInstr *ACopyMI = LIS->getInstructionFromIndex(AValNo->def);
// Don't allow any partial copies, even if isCoalescable() allows them.
if (!CP.isCoalescable(ACopyMI) || !ACopyMI->isFullCopy())
  return false;

// Get the Segment in IntB that this value number starts with.
LiveInterval::iterator ValS =
  IntB.FindSegmentContaining(AValNo->def.getPrevSlot());
if (ValS == IntB.end())
  return false;

// Make sure that the end of the live segment is inside the same block as
// CopyMI.
MachineInstr *ValSEndInst =
  LIS->getInstructionFromIndex(ValS->end.getPrevSlot());
if (!ValSEndInst || ValSEndInst->getParent() != CopyMI->getParent())
  return false;

// Okay, we now know that ValS ends in the same block that the CopyMI
// live-range starts.  If there are no intervening live segments between them
// in IntB, we can merge them.
if (ValS+1 != BS) return false;

LLVM_DEBUG(dbgs() << "Extending: " << printReg(IntB.reg(), TRI))do { } while (false);

SlotIndex FillerStart = ValS->end, FillerEnd = BS->start;
// We are about to delete CopyMI, so need to remove it as the 'instruction
// that defines this value #'. Update the valnum with the new defining
// instruction #.
BValNo->def = FillerStart;

// Okay, we can merge them.  We need to insert a new liverange:
// [ValS.end, BS.begin) of either value number, then we merge the
// two value numbers.
IntB.addSegment(LiveInterval::Segment(FillerStart, FillerEnd, BValNo));

// Okay, merge "B1" into the same value number as "B0".
if (BValNo != ValS->valno)
  IntB.MergeValueNumberInto(BValNo, ValS->valno);

// Do the same for the subregister segments.
for (LiveInterval::SubRange &S : IntB.subranges()) {
  // Check for SubRange Segments of the form [1234r,1234d:0) which can be
  // removed to prevent creating bogus SubRange Segments.
  LiveInterval::iterator SS = S.FindSegmentContaining(CopyIdx);
  if (SS != S.end() && SlotIndex::isSameInstr(SS->start, SS->end)) {
    S.removeSegment(*SS, true);
    continue;
  }
  // The subrange may have ended before FillerStart. If so, extend it.
  if (!S.getVNInfoAt(FillerStart)) {
    SlotIndex BBStart =
        LIS->getMBBStartIdx(LIS->getMBBFromIndex(FillerStart));
    S.extendInBlock(BBStart, FillerStart);
  }
  VNInfo *SubBValNo = S.getVNInfoAt(CopyIdx);
  S.addSegment(LiveInterval::Segment(FillerStart, FillerEnd, SubBValNo));
  VNInfo *SubValSNo = S.getVNInfoAt(AValNo->def.getPrevSlot());
  if (SubBValNo != SubValSNo)
    S.MergeValueNumberInto(SubBValNo, SubValSNo);
}

LLVM_DEBUG(dbgs() << "   result = " << IntB << '\n')do { } while (false);

// If the source instruction was killing the source register before the
// merge, unset the isKill marker given the live range has been extended.
int UIdx = ValSEndInst->findRegisterUseOperandIdx(IntB.reg(), true);
if (UIdx != -1) {
  ValSEndInst->getOperand(UIdx).setIsKill(false);
}

// Rewrite the copy.
CopyMI->substituteRegister(IntA.reg(), IntB.reg(), 0, *TRI);
// If the copy instruction was killing the destination register or any
// subrange before the merge trim the live range.
bool RecomputeLiveRange = AS->end == CopyIdx;
if (!RecomputeLiveRange) {
  for (LiveInterval::SubRange &S : IntA.subranges()) {
    LiveInterval::iterator SS = S.FindSegmentContaining(CopyUseIdx);
    if (SS != S.end() && SS->end == CopyIdx) {
      RecomputeLiveRange = true;
      break;
    }
  }
}
if (RecomputeLiveRange)
  shrinkToUses(&IntA);

++numExtends;
return true;
760}

762bool RegisterCoalescer::hasOtherReachingDefs(LiveInterval &IntA,
                                           LiveInterval &IntB,
                                           VNInfo *AValNo,
                                           VNInfo *BValNo) {
// If AValNo has PHI kills, conservatively assume that IntB defs can reach
// the PHI values.
if (LIS->hasPHIKill(IntA, AValNo))
  return true;

for (LiveRange::Segment &ASeg : IntA.segments) {
  if (ASeg.valno != AValNo) continue;
  LiveInterval::iterator BI = llvm::upper_bound(IntB, ASeg.start);
  if (BI != IntB.begin())
    --BI;
  for (; BI != IntB.end() && ASeg.end >= BI->start; ++BI) {
    if (BI->valno == BValNo)
      continue;
    if (BI->start <= ASeg.start && BI->end > ASeg.start)
      return true;
    if (BI->start > ASeg.start && BI->start < ASeg.end)
      return true;
  }
}
return false;
786}

788/// Copy segments with value number @p SrcValNo from liverange @p Src to live
789/// range @Dst and use value number @p DstValNo there.
790static std::pair<bool,bool>
791addSegmentsWithValNo(LiveRange &Dst, VNInfo *DstValNo, const LiveRange &Src,
                   const VNInfo *SrcValNo) {
bool Changed = false;
bool MergedWithDead = false;
for (const LiveRange::Segment &S : Src.segments) {
  if (S.valno != SrcValNo)
    continue;
  // This is adding a segment from Src that ends in a copy that is about
  // to be removed. This segment is going to be merged with a pre-existing
  // segment in Dst. This works, except in cases when the corresponding
  // segment in Dst is dead. For example: adding [192r,208r:1) from Src
  // to [208r,208d:1) in Dst would create [192r,208d:1) in Dst.
  // Recognized such cases, so that the segments can be shrunk.
  LiveRange::Segment Added = LiveRange::Segment(S.start, S.end, DstValNo);
  LiveRange::Segment &Merged = *Dst.addSegment(Added);
  if (Merged.end.isDead())
    MergedWithDead = true;
  Changed = true;
}
return std::make_pair(Changed, MergedWithDead);
811}

813std::pair<bool,bool>
814RegisterCoalescer::removeCopyByCommutingDef(const CoalescerPair &CP,
                                          MachineInstr *CopyMI) {
assert(!CP.isPhys())(static_cast<void> (0));

LiveInterval &IntA =
    LIS->getInterval(CP.isFlipped() ? CP.getDstReg() : CP.getSrcReg());
LiveInterval &IntB =
    LIS->getInterval(CP.isFlipped() ? CP.getSrcReg() : CP.getDstReg());

// We found a non-trivially-coalescable copy with IntA being the source and
// IntB being the dest, thus this defines a value number in IntB.  If the
// source value number (in IntA) is defined by a commutable instruction and
// its other operand is coalesced to the copy dest register, see if we can
// transform the copy into a noop by commuting the definition. For example,
//
//  A3 = op A2 killed B0
//    ...
//  B1 = A3      <- this copy
//    ...
//     = op A3   <- more uses
//
// ==>
//
//  B2 = op B0 killed A2
//    ...
//  B1 = B2      <- now an identity copy
//    ...
//     = op B2   <- more uses

// BValNo is a value number in B that is defined by a copy from A. 'B1' in
// the example above.
SlotIndex CopyIdx = LIS->getInstructionIndex(*CopyMI).getRegSlot();
VNInfo *BValNo = IntB.getVNInfoAt(CopyIdx);
assert(BValNo != nullptr && BValNo->def == CopyIdx)(static_cast<void> (0));

// AValNo is the value number in A that defines the copy, A3 in the example.
VNInfo *AValNo = IntA.getVNInfoAt(CopyIdx.getRegSlot(true));
assert(AValNo && !AValNo->isUnused() && "COPY source not live")(static_cast<void> (0));
if (AValNo->isPHIDef())
  return { false, false };
MachineInstr *DefMI = LIS->getInstructionFromIndex(AValNo->def);
if (!DefMI)
  return { false, false };
if (!DefMI->isCommutable())
  return { false, false };
// If DefMI is a two-address instruction then commuting it will change the
// destination register.
int DefIdx = DefMI->findRegisterDefOperandIdx(IntA.reg());
assert(DefIdx != -1)(static_cast<void> (0));
unsigned UseOpIdx;
if (!DefMI->isRegTiedToUseOperand(DefIdx, &UseOpIdx))
  return { false, false };

// FIXME: The code below tries to commute 'UseOpIdx' operand with some other
// commutable operand which is expressed by 'CommuteAnyOperandIndex'value
// passed to the method. That _other_ operand is chosen by
// the findCommutedOpIndices() method.
//
// That is obviously an area for improvement in case of instructions having
// more than 2 operands. For example, if some instruction has 3 commutable
// operands then all possible variants (i.e. op#1<->op#2, op#1<->op#3,
// op#2<->op#3) of commute transformation should be considered/tried here.
unsigned NewDstIdx = TargetInstrInfo::CommuteAnyOperandIndex;
if (!TII->findCommutedOpIndices(*DefMI, UseOpIdx, NewDstIdx))
  return { false, false };

MachineOperand &NewDstMO = DefMI->getOperand(NewDstIdx);
Register NewReg = NewDstMO.getReg();
if (NewReg != IntB.reg() || !IntB.Query(AValNo->def).isKill())
  return { false, false };

// Make sure there are no other definitions of IntB that would reach the
// uses which the new definition can reach.
if (hasOtherReachingDefs(IntA, IntB, AValNo, BValNo))
  return { false, false };

// If some of the uses of IntA.reg is already coalesced away, return false.
// It's not possible to determine whether it's safe to perform the coalescing.
for (MachineOperand &MO : MRI->use_nodbg_operands(IntA.reg())) {
  MachineInstr *UseMI = MO.getParent();
  unsigned OpNo = &MO - &UseMI->getOperand(0);
  SlotIndex UseIdx = LIS->getInstructionIndex(*UseMI);
  LiveInterval::iterator US = IntA.FindSegmentContaining(UseIdx);
  if (US == IntA.end() || US->valno != AValNo)
    continue;
  // If this use is tied to a def, we can't rewrite the register.
  if (UseMI->isRegTiedToDefOperand(OpNo))
    return { false, false };
}

LLVM_DEBUG(dbgs() << "\tremoveCopyByCommutingDef: " << AValNo->def << '\t'do { } while (false)
                  << *DefMI)do { } while (false);

// At this point we have decided that it is legal to do this
// transformation.  Start by commuting the instruction.
MachineBasicBlock *MBB = DefMI->getParent();
MachineInstr *NewMI =
    TII->commuteInstruction(*DefMI, false, UseOpIdx, NewDstIdx);
if (!NewMI)
  return { false, false };
if (Register::isVirtualRegister(IntA.reg()) &&
    Register::isVirtualRegister(IntB.reg()) &&
    !MRI->constrainRegClass(IntB.reg(), MRI->getRegClass(IntA.reg())))
  return { false, false };
if (NewMI != DefMI) {
  LIS->ReplaceMachineInstrInMaps(*DefMI, *NewMI);
  MachineBasicBlock::iterator Pos = DefMI;
  MBB->insert(Pos, NewMI);
  MBB->erase(DefMI);
}

// If ALR and BLR overlaps and end of BLR extends beyond end of ALR, e.g.
// A = or A, B
// ...
// B = A
// ...
// C = killed A
// ...
//   = B

// Update uses of IntA of the specific Val# with IntB.
for (MachineRegisterInfo::use_iterator UI = MRI->use_begin(IntA.reg()),
                                       UE = MRI->use_end();
     UI != UE;
     /* ++UI is below because of possible MI removal */) {
  MachineOperand &UseMO = *UI;
  ++UI;
  if (UseMO.isUndef())
    continue;
  MachineInstr *UseMI = UseMO.getParent();
  if (UseMI->isDebugInstr()) {
    // FIXME These don't have an instruction index.  Not clear we have enough
    // info to decide whether to do this replacement or not.  For now do it.
    UseMO.setReg(NewReg);
    continue;
  }
  SlotIndex UseIdx = LIS->getInstructionIndex(*UseMI).getRegSlot(true);
  LiveInterval::iterator US = IntA.FindSegmentContaining(UseIdx);
  assert(US != IntA.end() && "Use must be live")(static_cast<void> (0));
  if (US->valno != AValNo)
    continue;
  // Kill flags are no longer accurate. They are recomputed after RA.
  UseMO.setIsKill(false);
  if (Register::isPhysicalRegister(NewReg))
    UseMO.substPhysReg(NewReg, *TRI);
  else
    UseMO.setReg(NewReg);
  if (UseMI == CopyMI)
    continue;
  if (!UseMI->isCopy())
    continue;
  if (UseMI->getOperand(0).getReg() != IntB.reg() ||
      UseMI->getOperand(0).getSubReg())
    continue;

  // This copy will become a noop. If it's defining a new val#, merge it into
  // BValNo.
  SlotIndex DefIdx = UseIdx.getRegSlot();
  VNInfo *DVNI = IntB.getVNInfoAt(DefIdx);
  if (!DVNI)
    continue;
  LLVM_DEBUG(dbgs() << "\t\tnoop: " << DefIdx << '\t' << *UseMI)do { } while (false);
  assert(DVNI->def == DefIdx)(static_cast<void> (0));
  BValNo = IntB.MergeValueNumberInto(DVNI, BValNo);
  for (LiveInterval::SubRange &S : IntB.subranges()) {
    VNInfo *SubDVNI = S.getVNInfoAt(DefIdx);
    if (!SubDVNI)
      continue;
    VNInfo *SubBValNo = S.getVNInfoAt(CopyIdx);
    assert(SubBValNo->def == CopyIdx)(static_cast<void> (0));
    S.MergeValueNumberInto(SubDVNI, SubBValNo);
  }

  deleteInstr(UseMI);
}

// Extend BValNo by merging in IntA live segments of AValNo. Val# definition
// is updated.
bool ShrinkB = false;
BumpPtrAllocator &Allocator = LIS->getVNInfoAllocator();
if (IntA.hasSubRanges() || IntB.hasSubRanges()) {
  if (!IntA.hasSubRanges()) {
    LaneBitmask Mask = MRI->getMaxLaneMaskForVReg(IntA.reg());
    IntA.createSubRangeFrom(Allocator, Mask, IntA);
  } else if (!IntB.hasSubRanges()) {
    LaneBitmask Mask = MRI->getMaxLaneMaskForVReg(IntB.reg());
    IntB.createSubRangeFrom(Allocator, Mask, IntB);
  }
  SlotIndex AIdx = CopyIdx.getRegSlot(true);
  LaneBitmask MaskA;
  const SlotIndexes &Indexes = *LIS->getSlotIndexes();
  for (LiveInterval::SubRange &SA : IntA.subranges()) {
    VNInfo *ASubValNo = SA.getVNInfoAt(AIdx);
    // Even if we are dealing with a full copy, some lanes can
    // still be undefined.
    // E.g.,
    // undef A.subLow = ...
    // B = COPY A <== A.subHigh is undefined here and does
    //                not have a value number.
    if (!ASubValNo)
      continue;
    MaskA |= SA.LaneMask;

    IntB.refineSubRanges(
        Allocator, SA.LaneMask,
        [&Allocator, &SA, CopyIdx, ASubValNo,
         &ShrinkB](LiveInterval::SubRange &SR) {
          VNInfo *BSubValNo = SR.empty() ? SR.getNextValue(CopyIdx, Allocator)
                                         : SR.getVNInfoAt(CopyIdx);
          assert(BSubValNo != nullptr)(static_cast<void> (0));
          auto P = addSegmentsWithValNo(SR, BSubValNo, SA, ASubValNo);
          ShrinkB |= P.second;
          if (P.first)
            BSubValNo->def = ASubValNo->def;
        },
        Indexes, *TRI);
  }
  // Go over all subranges of IntB that have not been covered by IntA,
  // and delete the segments starting at CopyIdx. This can happen if
  // IntA has undef lanes that are defined in IntB.
  for (LiveInterval::SubRange &SB : IntB.subranges()) {
    if ((SB.LaneMask & MaskA).any())
      continue;
    if (LiveRange::Segment *S = SB.getSegmentContaining(CopyIdx))
      if (S->start.getBaseIndex() == CopyIdx.getBaseIndex())
        SB.removeSegment(*S, true);
  }
}

BValNo->def = AValNo->def;
auto P = addSegmentsWithValNo(IntB, BValNo, IntA, AValNo);
ShrinkB |= P.second;
LLVM_DEBUG(dbgs() << "\t\textended: " << IntB << '\n')do { } while (false);

LIS->removeVRegDefAt(IntA, AValNo->def);

LLVM_DEBUG(dbgs() << "\t\ttrimmed:  " << IntA << '\n')do { } while (false);
++numCommutes;
return { true, ShrinkB };
1053}

1055/// For copy B = A in BB2, if A is defined by A = B in BB0 which is a
1056/// predecessor of BB2, and if B is not redefined on the way from A = B
1057/// in BB0 to B = A in BB2, B = A in BB2 is partially redundant if the
1058/// execution goes through the path from BB0 to BB2. We may move B = A
1059/// to the predecessor without such reversed copy.
1060/// So we will transform the program from:
1061///   BB0:
1062///      A = B;    BB1:
1063///       ...         ...
1064///     /     \      /
1065///             BB2:
1066///               ...
1067///               B = A;
1068///
1069/// to:
1070///
1071///   BB0:         BB1:
1072///      A = B;        ...
1073///       ...          B = A;
1074///     /     \       /
1075///             BB2:
1076///               ...
1077///
1078/// A special case is when BB0 and BB2 are the same BB which is the only
1079/// BB in a loop:
1080///   BB1:
1081///        ...
1082///   BB0/BB2:  ----
1083///        B = A;   |
1084///        ...      |
1085///        A = B;   |
1086///          |-------
1087///          |
1088/// We may hoist B = A from BB0/BB2 to BB1.
1089///
1090/// The major preconditions for correctness to remove such partial
1091/// redundancy include:
1092/// 1. A in B = A in BB2 is defined by a PHI in BB2, and one operand of
1093///    the PHI is defined by the reversed copy A = B in BB0.
1094/// 2. No B is referenced from the start of BB2 to B = A.
1095/// 3. No B is defined from A = B to the end of BB0.
1096/// 4. BB1 has only one successor.
1097///
1098/// 2 and 4 implicitly ensure B is not live at the end of BB1.
1099/// 4 guarantees BB2 is hotter than BB1, so we can only move a copy to a
1100/// colder place, which not only prevent endless loop, but also make sure
1101/// the movement of copy is beneficial.
1102bool RegisterCoalescer::removePartialRedundancy(const CoalescerPair &CP,
                                              MachineInstr &CopyMI) {
assert(!CP.isPhys())(static_cast<void> (0));
if (!CopyMI.isFullCopy())
  return false;

MachineBasicBlock &MBB = *CopyMI.getParent();
// If this block is the target of an invoke/inlineasm_br, moving the copy into
// the predecessor is tricker, and we don't handle it.
if (MBB.isEHPad() || MBB.isInlineAsmBrIndirectTarget())
  return false;

if (MBB.pred_size() != 2)
  return false;

LiveInterval &IntA =
    LIS->getInterval(CP.isFlipped() ? CP.getDstReg() : CP.getSrcReg());
LiveInterval &IntB =
    LIS->getInterval(CP.isFlipped() ? CP.getSrcReg() : CP.getDstReg());

// A is defined by PHI at the entry of MBB.
SlotIndex CopyIdx = LIS->getInstructionIndex(CopyMI).getRegSlot(true);
VNInfo *AValNo = IntA.getVNInfoAt(CopyIdx);
assert(AValNo && !AValNo->isUnused() && "COPY source not live")(static_cast<void> (0));
if (!AValNo->isPHIDef())
  return false;

// No B is referenced before CopyMI in MBB.
if (IntB.overlaps(LIS->getMBBStartIdx(&MBB), CopyIdx))
  return false;

// MBB has two predecessors: one contains A = B so no copy will be inserted
// for it. The other one will have a copy moved from MBB.
bool FoundReverseCopy = false;
MachineBasicBlock *CopyLeftBB = nullptr;
for (MachineBasicBlock *Pred : MBB.predecessors()) {
  VNInfo *PVal = IntA.getVNInfoBefore(LIS->getMBBEndIdx(Pred));
  MachineInstr *DefMI = LIS->getInstructionFromIndex(PVal->def);
  if (!DefMI || !DefMI->isFullCopy()) {
    CopyLeftBB = Pred;
    continue;
  }
  // Check DefMI is a reverse copy and it is in BB Pred.
  if (DefMI->getOperand(0).getReg() != IntA.reg() ||
      DefMI->getOperand(1).getReg() != IntB.reg() ||
      DefMI->getParent() != Pred) {
    CopyLeftBB = Pred;
    continue;
  }
  // If there is any other def of B after DefMI and before the end of Pred,
  // we need to keep the copy of B = A at the end of Pred if we remove
  // B = A from MBB.
  bool ValB_Changed = false;
  for (auto VNI : IntB.valnos) {
    if (VNI->isUnused())
      continue;
    if (PVal->def < VNI->def && VNI->def < LIS->getMBBEndIdx(Pred)) {
      ValB_Changed = true;
      break;
    }
  }
  if (ValB_Changed) {
    CopyLeftBB = Pred;
    continue;
  }
  FoundReverseCopy = true;
}

// If no reverse copy is found in predecessors, nothing to do.
if (!FoundReverseCopy)
  return false;

// If CopyLeftBB is nullptr, it means every predecessor of MBB contains
// reverse copy, CopyMI can be removed trivially if only IntA/IntB is updated.
// If CopyLeftBB is not nullptr, move CopyMI from MBB to CopyLeftBB and
// update IntA/IntB.
//
// If CopyLeftBB is not nullptr, ensure CopyLeftBB has a single succ so
// MBB is hotter than CopyLeftBB.
if (CopyLeftBB && CopyLeftBB->succ_size() > 1)
  return false;

// Now (almost sure it's) ok to move copy.
if (CopyLeftBB) {
  // Position in CopyLeftBB where we should insert new copy.
  auto InsPos = CopyLeftBB->getFirstTerminator();

  // Make sure that B isn't referenced in the terminators (if any) at the end
  // of the predecessor since we're about to insert a new definition of B
  // before them.
  if (InsPos != CopyLeftBB->end()) {
    SlotIndex InsPosIdx = LIS->getInstructionIndex(*InsPos).getRegSlot(true);
    if (IntB.overlaps(InsPosIdx, LIS->getMBBEndIdx(CopyLeftBB)))
      return false;
  }

  LLVM_DEBUG(dbgs() << "\tremovePartialRedundancy: Move the copy to "do { } while (false)
                    << printMBBReference(*CopyLeftBB) << '\t' << CopyMI)do { } while (false);

  // Insert new copy to CopyLeftBB.
  MachineInstr *NewCopyMI = BuildMI(*CopyLeftBB, InsPos, CopyMI.getDebugLoc(),
                                    TII->get(TargetOpcode::COPY), IntB.reg())
                                .addReg(IntA.reg());
  SlotIndex NewCopyIdx =
      LIS->InsertMachineInstrInMaps(*NewCopyMI).getRegSlot();
  IntB.createDeadDef(NewCopyIdx, LIS->getVNInfoAllocator());
  for (LiveInterval::SubRange &SR : IntB.subranges())
    SR.createDeadDef(NewCopyIdx, LIS->getVNInfoAllocator());

  // If the newly created Instruction has an address of an instruction that was
  // deleted before (object recycled by the allocator) it needs to be removed from
  // the deleted list.
  ErasedInstrs.erase(NewCopyMI);
} else {
  LLVM_DEBUG(dbgs() << "\tremovePartialRedundancy: Remove the copy from "do { } while (false)
                    << printMBBReference(MBB) << '\t' << CopyMI)do { } while (false);
}

// Remove CopyMI.
// Note: This is fine to remove the copy before updating the live-ranges.
// While updating the live-ranges, we only look at slot indices and
// never go back to the instruction.
// Mark instructions as deleted.
deleteInstr(&CopyMI);

// Update the liveness.
SmallVector<SlotIndex, 8> EndPoints;
VNInfo *BValNo = IntB.Query(CopyIdx).valueOutOrDead();
LIS->pruneValue(*static_cast<LiveRange *>(&IntB), CopyIdx.getRegSlot(),
                &EndPoints);
BValNo->markUnused();
// Extend IntB to the EndPoints of its original live interval.
LIS->extendToIndices(IntB, EndPoints);

// Now, do the same for its subranges.
for (LiveInterval::SubRange &SR : IntB.subranges()) {
  EndPoints.clear();
  VNInfo *BValNo = SR.Query(CopyIdx).valueOutOrDead();
  assert(BValNo && "All sublanes should be live")(static_cast<void> (0));
  LIS->pruneValue(SR, CopyIdx.getRegSlot(), &EndPoints);
  BValNo->markUnused();
  // We can have a situation where the result of the original copy is live,
  // but is immediately dead in this subrange, e.g. [336r,336d:0). That makes
  // the copy appear as an endpoint from pruneValue(), but we don't want it
  // to because the copy has been removed.  We can go ahead and remove that
  // endpoint; there is no other situation here that there could be a use at
  // the same place as we know that the copy is a full copy.
  for (unsigned I = 0; I != EndPoints.size(); ) {
    if (SlotIndex::isSameInstr(EndPoints[I], CopyIdx)) {
      EndPoints[I] = EndPoints.back();
      EndPoints.pop_back();
      continue;
    }
    ++I;
  }
  SmallVector<SlotIndex, 8> Undefs;
  IntB.computeSubRangeUndefs(Undefs, SR.LaneMask, *MRI,
                             *LIS->getSlotIndexes());
  LIS->extendToIndices(SR, EndPoints, Undefs);
}
// If any dead defs were extended, truncate them.
shrinkToUses(&IntB);

// Finally, update the live-range of IntA.
shrinkToUses(&IntA);
return true;
1268}

1270/// Returns true if @p MI defines the full vreg @p Reg, as opposed to just
1271/// defining a subregister.
1272static bool definesFullReg(const MachineInstr &MI, Register Reg) {
assert(!Reg.isPhysical() && "This code cannot handle physreg aliasing")(static_cast<void> (0));

for (const MachineOperand &Op : MI.operands()) {
  if (!Op.isReg() || !Op.isDef() || Op.getReg() != Reg)
    continue;
  // Return true if we define the full register or don't care about the value
  // inside other subregisters.
  if (Op.getSubReg() == 0 || Op.isUndef())
    return true;
}
return false;
1284}

1286bool RegisterCoalescer::reMaterializeTrivialDef(const CoalescerPair &CP,
                                              MachineInstr *CopyMI,
                                              bool &IsDefCopy) {
IsDefCopy = false;
Register SrcReg = CP.isFlipped() ? CP.getDstReg() : CP.getSrcReg();
unsigned SrcIdx = CP.isFlipped() ? CP.getDstIdx() : CP.getSrcIdx();
Register DstReg = CP.isFlipped() ? CP.getSrcReg() : CP.getDstReg();
unsigned DstIdx = CP.isFlipped() ? CP.getSrcIdx() : CP.getDstIdx();
if (Register::isPhysicalRegister(SrcReg))
  return false;

LiveInterval &SrcInt = LIS->getInterval(SrcReg);
SlotIndex CopyIdx = LIS->getInstructionIndex(*CopyMI);
VNInfo *ValNo = SrcInt.Query(CopyIdx).valueIn();
if (!ValNo)
  return false;
if (ValNo->isPHIDef() || ValNo->isUnused())
  return false;
MachineInstr *DefMI = LIS->getInstructionFromIndex(ValNo->def);
if (!DefMI)
  return false;
if (DefMI->isCopyLike()) {
  IsDefCopy = true;
  return false;
}
if (!TII->isAsCheapAsAMove(*DefMI))
  return false;
if (!TII->isTriviallyReMaterializable(*DefMI, AA))
  return false;
if (!definesFullReg(*DefMI, SrcReg))
  return false;
bool SawStore = false;
if (!DefMI->isSafeToMove(AA, SawStore))
  return false;
const MCInstrDesc &MCID = DefMI->getDesc();
if (MCID.getNumDefs() != 1)
  return false;
// Only support subregister destinations when the def is read-undef.
MachineOperand &DstOperand = CopyMI->getOperand(0);
Register CopyDstReg = DstOperand.getReg();
if (DstOperand.getSubReg() && !DstOperand.isUndef())
  return false;

// If both SrcIdx and DstIdx are set, correct rematerialization would widen
// the register substantially (beyond both source and dest size). This is bad
// for performance since it can cascade through a function, introducing many
// extra spills and fills (e.g. ARM can easily end up copying QQQQPR registers
// around after a few subreg copies).
if (SrcIdx && DstIdx)
  return false;

const TargetRegisterClass *DefRC = TII->getRegClass(MCID, 0, TRI, *MF);
if (!DefMI->isImplicitDef()) {
  if (DstReg.isPhysical()) {
    Register NewDstReg = DstReg;

    unsigned NewDstIdx = TRI->composeSubRegIndices(CP.getSrcIdx(),
                                            DefMI->getOperand(0).getSubReg());
    if (NewDstIdx)
      NewDstReg = TRI->getSubReg(DstReg, NewDstIdx);

    // Finally, make sure that the physical subregister that will be
    // constructed later is permitted for the instruction.
    if (!DefRC->contains(NewDstReg))
      return false;
  } else {
    // Theoretically, some stack frame reference could exist. Just make sure
    // it hasn't actually happened.
    assert(Register::isVirtualRegister(DstReg) &&(static_cast<void> (0))
           "Only expect to deal with virtual or physical registers")(static_cast<void> (0));
  }
}

if (!allUsesAvailableAt(DefMI, ValNo->def, CopyIdx))
  return false;

DebugLoc DL = CopyMI->getDebugLoc();
MachineBasicBlock *MBB = CopyMI->getParent();
MachineBasicBlock::iterator MII =
  std::next(MachineBasicBlock::iterator(CopyMI));
TII->reMaterialize(*MBB, MII, DstReg, SrcIdx, *DefMI, *TRI);
MachineInstr &NewMI = *std::prev(MII);
NewMI.setDebugLoc(DL);

// In a situation like the following:
//     %0:subreg = instr              ; DefMI, subreg = DstIdx
//     %1        = copy %0:subreg ; CopyMI, SrcIdx = 0
// instead of widening %1 to the register class of %0 simply do:
//     %1 = instr
const TargetRegisterClass *NewRC = CP.getNewRC();
if (DstIdx != 0) {
  MachineOperand &DefMO = NewMI.getOperand(0);
  if (DefMO.getSubReg() == DstIdx) {
    assert(SrcIdx == 0 && CP.isFlipped()(static_cast<void> (0))
           && "Shouldn't have SrcIdx+DstIdx at this point")(static_cast<void> (0));
    const TargetRegisterClass *DstRC = MRI->getRegClass(DstReg);
    const TargetRegisterClass *CommonRC =
      TRI->getCommonSubClass(DefRC, DstRC);
    if (CommonRC != nullptr) {
      NewRC = CommonRC;
      DstIdx = 0;
      DefMO.setSubReg(0);
      DefMO.setIsUndef(false); // Only subregs can have def+undef.
    }
  }
}

// CopyMI may have implicit operands, save them so that we can transfer them
// over to the newly materialized instruction after CopyMI is removed.
SmallVector<MachineOperand, 4> ImplicitOps;
ImplicitOps.reserve(CopyMI->getNumOperands() -
                    CopyMI->getDesc().getNumOperands());
for (unsigned I = CopyMI->getDesc().getNumOperands(),
              E = CopyMI->getNumOperands();
     I != E; ++I) {
  MachineOperand &MO = CopyMI->getOperand(I);
  if (MO.isReg()) {
    assert(MO.isImplicit() && "No explicit operands after implicit operands.")(static_cast<void> (0));
    // Discard VReg implicit defs.
    if (Register::isPhysicalRegister(MO.getReg()))
      ImplicitOps.push_back(MO);
  }
}

LIS->ReplaceMachineInstrInMaps(*CopyMI, NewMI);
CopyMI->eraseFromParent();
ErasedInstrs.insert(CopyMI);

// NewMI may have dead implicit defs (E.g. EFLAGS for MOV<bits>r0 on X86).
// We need to remember these so we can add intervals once we insert
// NewMI into SlotIndexes.
SmallVector<MCRegister, 4> NewMIImplDefs;
for (unsigned i = NewMI.getDesc().getNumOperands(),
              e = NewMI.getNumOperands();
     i != e; ++i) {
  MachineOperand &MO = NewMI.getOperand(i);
  if (MO.isReg() && MO.isDef()) {
    assert(MO.isImplicit() && MO.isDead() &&(static_cast<void> (0))
           Register::isPhysicalRegister(MO.getReg()))(static_cast<void> (0));
    NewMIImplDefs.push_back(MO.getReg().asMCReg());
  }
}

if (DstReg.isVirtual()) {
  unsigned NewIdx = NewMI.getOperand(0).getSubReg();

  if (DefRC != nullptr) {
    if (NewIdx)
      NewRC = TRI->getMatchingSuperRegClass(NewRC, DefRC, NewIdx);
    else
      NewRC = TRI->getCommonSubClass(NewRC, DefRC);
    assert(NewRC && "subreg chosen for remat incompatible with instruction")(static_cast<void> (0));
  }
  // Remap subranges to new lanemask and change register class.
  LiveInterval &DstInt = LIS->getInterval(DstReg);
  for (LiveInterval::SubRange &SR : DstInt.subranges()) {
    SR.LaneMask = TRI->composeSubRegIndexLaneMask(DstIdx, SR.LaneMask);
  }
  MRI->setRegClass(DstReg, NewRC);

  // Update machine operands and add flags.
  updateRegDefsUses(DstReg, DstReg, DstIdx);
  NewMI.getOperand(0).setSubReg(NewIdx);
  // updateRegDefUses can add an "undef" flag to the definition, since
  // it will replace DstReg with DstReg.DstIdx. If NewIdx is 0, make
  // sure that "undef" is not set.
  if (NewIdx == 0)
    NewMI.getOperand(0).setIsUndef(false);
  // Add dead subregister definitions if we are defining the whole register
  // but only part of it is live.
  // This could happen if the rematerialization instruction is rematerializing
  // more than actually is used in the register.
  // An example would be:
  // %1 = LOAD CONSTANTS 5, 8 ; Loading both 5 and 8 in different subregs
  // ; Copying only part of the register here, but the rest is undef.
  // %2:sub_16bit<def, read-undef> = COPY %1:sub_16bit
  // ==>
  // ; Materialize all the constants but only using one
  // %2 = LOAD_CONSTANTS 5, 8
  //
  // at this point for the part that wasn't defined before we could have
  // subranges missing the definition.
  if (NewIdx == 0 && DstInt.hasSubRanges()) {
    SlotIndex CurrIdx = LIS->getInstructionIndex(NewMI);
    SlotIndex DefIndex =
        CurrIdx.getRegSlot(NewMI.getOperand(0).isEarlyClobber());
    LaneBitmask MaxMask = MRI->getMaxLaneMaskForVReg(DstReg);
    VNInfo::Allocator& Alloc = LIS->getVNInfoAllocator();
    for (LiveInterval::SubRange &SR : DstInt.subranges()) {
      if (!SR.liveAt(DefIndex))
        SR.createDeadDef(DefIndex, Alloc);
      MaxMask &= ~SR.LaneMask;
    }
    if (MaxMask.any()) {
      LiveInterval::SubRange *SR = DstInt.createSubRange(Alloc, MaxMask);
      SR->createDeadDef(DefIndex, Alloc);
    }
  }

  // Make sure that the subrange for resultant undef is removed
  // For example:
  //   %1:sub1<def,read-undef> = LOAD CONSTANT 1
  //   %2 = COPY %1
  // ==>
  //   %2:sub1<def, read-undef> = LOAD CONSTANT 1
  //     ; Correct but need to remove the subrange for %2:sub0
  //     ; as it is now undef
  if (NewIdx != 0 && DstInt.hasSubRanges()) {
    // The affected subregister segments can be removed.
    SlotIndex CurrIdx = LIS->getInstructionIndex(NewMI);
    LaneBitmask DstMask = TRI->getSubRegIndexLaneMask(NewIdx);
    bool UpdatedSubRanges = false;
    SlotIndex DefIndex =
        CurrIdx.getRegSlot(NewMI.getOperand(0).isEarlyClobber());
    VNInfo::Allocator &Alloc = LIS->getVNInfoAllocator();
    for (LiveInterval::SubRange &SR : DstInt.subranges()) {
      if ((SR.LaneMask & DstMask).none()) {
        LLVM_DEBUG(dbgs()do { } while (false)
                   << "Removing undefined SubRange "do { } while (false)
                   << PrintLaneMask(SR.LaneMask) << " : " << SR << "\n")do { } while (false);
        // VNI is in ValNo - remove any segments in this SubRange that have this ValNo
        if (VNInfo *RmValNo = SR.getVNInfoAt(CurrIdx.getRegSlot())) {
          SR.removeValNo(RmValNo);
          UpdatedSubRanges = true;
        }
      } else {
        // We know that this lane is defined by this instruction,
        // but at this point it may be empty because it is not used by
        // anything. This happens when updateRegDefUses adds the missing
        // lanes. Assign that lane a dead def so that the interferences
        // are properly modeled.
        if (SR.empty())
          SR.createDeadDef(DefIndex, Alloc);
      }
    }
    if (UpdatedSubRanges)
      DstInt.removeEmptySubRanges();
  }
} else if (NewMI.getOperand(0).getReg() != CopyDstReg) {
  // The New instruction may be defining a sub-register of what's actually
  // been asked for. If so it must implicitly define the whole thing.
  assert(Register::isPhysicalRegister(DstReg) &&(static_cast<void> (0))
         "Only expect virtual or physical registers in remat")(static_cast<void> (0));
  NewMI.getOperand(0).setIsDead(true);
  NewMI.addOperand(MachineOperand::CreateReg(
      CopyDstReg, true /*IsDef*/, true /*IsImp*/, false /*IsKill*/));
  // Record small dead def live-ranges for all the subregisters
  // of the destination register.
  // Otherwise, variables that live through may miss some
  // interferences, thus creating invalid allocation.
  // E.g., i386 code:
  // %1 = somedef ; %1 GR8
  // %2 = remat ; %2 GR32
  // CL = COPY %2.sub_8bit
  // = somedef %1 ; %1 GR8
  // =>
  // %1 = somedef ; %1 GR8
  // dead ECX = remat ; implicit-def CL
  // = somedef %1 ; %1 GR8
  // %1 will see the interferences with CL but not with CH since
  // no live-ranges would have been created for ECX.
  // Fix that!
  SlotIndex NewMIIdx = LIS->getInstructionIndex(NewMI);
  for (MCRegUnitIterator Units(NewMI.getOperand(0).getReg(), TRI);
       Units.isValid(); ++Units)
    if (LiveRange *LR = LIS->getCachedRegUnit(*Units))
      LR->createDeadDef(NewMIIdx.getRegSlot(), LIS->getVNInfoAllocator());
}

if (NewMI.getOperand(0).getSubReg())
  NewMI.getOperand(0).setIsUndef();

// Transfer over implicit operands to the rematerialized instruction.
for (MachineOperand &MO : ImplicitOps)
  NewMI.addOperand(MO);

SlotIndex NewMIIdx = LIS->getInstructionIndex(NewMI);
for (unsigned i = 0, e = NewMIImplDefs.size(); i != e; ++i) {
  MCRegister Reg = NewMIImplDefs[i];
  for (MCRegUnitIterator Units(Reg, TRI); Units.isValid(); ++Units)
    if (LiveRange *LR = LIS->getCachedRegUnit(*Units))
      LR->createDeadDef(NewMIIdx.getRegSlot(), LIS->getVNInfoAllocator());
}

LLVM_DEBUG(dbgs() << "Remat: " << NewMI)do { } while (false);
++NumReMats;

// If the virtual SrcReg is completely eliminated, update all DBG_VALUEs
// to describe DstReg instead.
if (MRI->use_nodbg_empty(SrcReg)) {
  for (MachineRegisterInfo::use_iterator UI = MRI->use_begin(SrcReg);
       UI != MRI->use_end();) {
    MachineOperand &UseMO = *UI++;
    MachineInstr *UseMI = UseMO.getParent();
    if (UseMI->isDebugInstr()) {
      if (Register::isPhysicalRegister(DstReg))
        UseMO.substPhysReg(DstReg, *TRI);
      else
        UseMO.setReg(DstReg);
      // Move the debug value directly after the def of the rematerialized
      // value in DstReg.
      MBB->splice(std::next(NewMI.getIterator()), UseMI->getParent(), UseMI);
      LLVM_DEBUG(dbgs() << "\t\tupdated: " << *UseMI)do { } while (false);
    }
  }
}

if (ToBeUpdated.count(SrcReg))
  return true;

unsigned NumCopyUses = 0;
for (MachineOperand &UseMO : MRI->use_nodbg_operands(SrcReg)) {
  if (UseMO.getParent()->isCopyLike())
    NumCopyUses++;
}
if (NumCopyUses < LateRematUpdateThreshold) {
  // The source interval can become smaller because we removed a use.
  shrinkToUses(&SrcInt, &DeadDefs);
  if (!DeadDefs.empty())
    eliminateDeadDefs();
} else {
  ToBeUpdated.insert(SrcReg);
}
return true;
1610}

1612MachineInstr *RegisterCoalescer::eliminateUndefCopy(MachineInstr *CopyMI) {
// ProcessImplicitDefs may leave some copies of <undef> values, it only
// removes local variables. When we have a copy like:
//
//   %1 = COPY undef %2
//
// We delete the copy and remove the corresponding value number from %1.
// Any uses of that value number are marked as <undef>.

// Note that we do not query CoalescerPair here but redo isMoveInstr as the
// CoalescerPair may have a new register class with adjusted subreg indices
// at this point.
Register SrcReg, DstReg;
unsigned SrcSubIdx = 0, DstSubIdx = 0;
if(!isMoveInstr(*TRI, CopyMI, SrcReg, DstReg, SrcSubIdx, DstSubIdx))
  return nullptr;

SlotIndex Idx = LIS->getInstructionIndex(*CopyMI);
const LiveInterval &SrcLI = LIS->getInterval(SrcReg);
// CopyMI is undef iff SrcReg is not live before the instruction.
if (SrcSubIdx != 0 && SrcLI.hasSubRanges()) {
  LaneBitmask SrcMask = TRI->getSubRegIndexLaneMask(SrcSubIdx);
  for (const LiveInterval::SubRange &SR : SrcLI.subranges()) {
    if ((SR.LaneMask & SrcMask).none())
      continue;
    if (SR.liveAt(Idx))
      return nullptr;
  }
} else if (SrcLI.liveAt(Idx))
  return nullptr;

// If the undef copy defines a live-out value (i.e. an input to a PHI def),
// then replace it with an IMPLICIT_DEF.
LiveInterval &DstLI = LIS->getInterval(DstReg);
SlotIndex RegIndex = Idx.getRegSlot();
LiveRange::Segment *Seg = DstLI.getSegmentContaining(RegIndex);
assert(Seg != nullptr && "No segment for defining instruction")(static_cast<void> (0));
if (VNInfo *V = DstLI.getVNInfoAt(Seg->end)) {
  if (V->isPHIDef()) {
    CopyMI->setDesc(TII->get(TargetOpcode::IMPLICIT_DEF));
    for (unsigned i = CopyMI->getNumOperands(); i != 0; --i) {
      MachineOperand &MO = CopyMI->getOperand(i-1);
      if (MO.isReg() && MO.isUse())
        CopyMI->RemoveOperand(i-1);
    }
    LLVM_DEBUG(dbgs() << "\tReplaced copy of <undef> value with an "do { } while (false)
                         "implicit def\n")do { } while (false);
    return CopyMI;
  }
}

// Remove any DstReg segments starting at the instruction.
LLVM_DEBUG(dbgs() << "\tEliminating copy of <undef> value\n")do { } while (false);

// Remove value or merge with previous one in case of a subregister def.
if (VNInfo *PrevVNI = DstLI.getVNInfoAt(Idx)) {
  VNInfo *VNI = DstLI.getVNInfoAt(RegIndex);
  DstLI.MergeValueNumberInto(VNI, PrevVNI);

  // The affected subregister segments can be removed.
  LaneBitmask DstMask = TRI->getSubRegIndexLaneMask(DstSubIdx);
  for (LiveInterval::SubRange &SR : DstLI.subranges()) {
    if ((SR.LaneMask & DstMask).none())
      continue;

    VNInfo *SVNI = SR.getVNInfoAt(RegIndex);
    assert(SVNI != nullptr && SlotIndex::isSameInstr(SVNI->def, RegIndex))(static_cast<void> (0));
    SR.removeValNo(SVNI);
  }
  DstLI.removeEmptySubRanges();
} else
  LIS->removeVRegDefAt(DstLI, RegIndex);

// Mark uses as undef.
for (MachineOperand &MO : MRI->reg_nodbg_operands(DstReg)) {
  if (MO.isDef() /*|| MO.isUndef()*/)
    continue;
  const MachineInstr &MI = *MO.getParent();
  SlotIndex UseIdx = LIS->getInstructionIndex(MI);
  LaneBitmask UseMask = TRI->getSubRegIndexLaneMask(MO.getSubReg());
  bool isLive;
  if (!UseMask.all() && DstLI.hasSubRanges()) {
    isLive = false;
    for (const LiveInterval::SubRange &SR : DstLI.subranges()) {
      if ((SR.LaneMask & UseMask).none())
        continue;
      if (SR.liveAt(UseIdx)) {
        isLive = true;
        break;
      }
    }
  } else
    isLive = DstLI.liveAt(UseIdx);
  if (isLive)
    continue;
  MO.setIsUndef(true);
  LLVM_DEBUG(dbgs() << "\tnew undef: " << UseIdx << '\t' << MI)do { } while (false);
}

// A def of a subregister may be a use of the other subregisters, so
// deleting a def of a subregister may also remove uses. Since CopyMI
// is still part of the function (but about to be erased), mark all
// defs of DstReg in it as <undef>, so that shrinkToUses would
// ignore them.
for (MachineOperand &MO : CopyMI->operands())
  if (MO.isReg() && MO.isDef() && MO.getReg() == DstReg)
    MO.setIsUndef(true);
LIS->shrinkToUses(&DstLI);

return CopyMI;
1722}

1724void RegisterCoalescer::addUndefFlag(const LiveInterval &Int, SlotIndex UseIdx,
                                   MachineOperand &MO, unsigned SubRegIdx) {
LaneBitmask Mask = TRI->getSubRegIndexLaneMask(SubRegIdx);
if (MO.isDef())
  Mask = ~Mask;
bool IsUndef = true;
for (const LiveInterval::SubRange &S : Int.subranges()) {
  if ((S.LaneMask & Mask).none())
    continue;
  if (S.liveAt(UseIdx)) {
    IsUndef = false;
    break;
  }
}
if (IsUndef) {
  MO.setIsUndef(true);
  // We found out some subregister use is actually reading an undefined
  // value. In some cases the whole vreg has become undefined at this
  // point so we have to potentially shrink the main range if the
  // use was ending a live segment there.
  LiveQueryResult Q = Int.Query(UseIdx);
  if (Q.valueOut() == nullptr)
    ShrinkMainRange = true;
}
1748}

1750void RegisterCoalescer::updateRegDefsUses(Register SrcReg, Register DstReg,
                                        unsigned SubIdx) {
bool DstIsPhys = Register::isPhysicalRegister(DstReg);
LiveInterval *DstInt = DstIsPhys ? nullptr : &LIS->getInterval(DstReg);

if (DstInt && DstInt->hasSubRanges() && DstReg != SrcReg) {
  for (MachineOperand &MO : MRI->reg_operands(DstReg)) {
    unsigned SubReg = MO.getSubReg();
    if (SubReg == 0 || MO.isUndef())
      continue;
    MachineInstr &MI = *MO.getParent();
    if (MI.isDebugInstr())
      continue;
    SlotIndex UseIdx = LIS->getInstructionIndex(MI).getRegSlot(true);
    addUndefFlag(*DstInt, UseIdx, MO, SubReg);
  }
}

SmallPtrSet<MachineInstr*, 8> Visited;
for (MachineRegisterInfo::reg_instr_iterator
     I = MRI->reg_instr_begin(SrcReg), E = MRI->reg_instr_end();
     I != E; ) {
  MachineInstr *UseMI = &*(I++);

  // Each instruction can only be rewritten once because sub-register
  // composition is not always idempotent. When SrcReg != DstReg, rewriting
  // the UseMI operands removes them from the SrcReg use-def chain, but when
  // SrcReg is DstReg we could encounter UseMI twice if it has multiple
  // operands mentioning the virtual register.
  if (SrcReg == DstReg && !Visited.insert(UseMI).second)
    continue;

  SmallVector<unsigned,8> Ops;
  bool Reads, Writes;
  std::tie(Reads, Writes) = UseMI->readsWritesVirtualRegister(SrcReg, &Ops);

  // If SrcReg wasn't read, it may still be the case that DstReg is live-in
  // because SrcReg is a sub-register.
  if (DstInt && !Reads && SubIdx && !UseMI->isDebugInstr())
    Reads = DstInt->liveAt(LIS->getInstructionIndex(*UseMI));

  // Replace SrcReg with DstReg in all UseMI operands.
  for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
    MachineOperand &MO = UseMI->getOperand(Ops[i]);

    // Adjust <undef> flags in case of sub-register joins. We don't want to
    // turn a full def into a read-modify-write sub-register def and vice
    // versa.
    if (SubIdx && MO.isDef())
      MO.setIsUndef(!Reads);

    // A subreg use of a partially undef (super) register may be a complete
    // undef use now and then has to be marked that way.
    if (MO.isUse() && !DstIsPhys) {
      unsigned SubUseIdx = TRI->composeSubRegIndices(SubIdx, MO.getSubReg());
      if (SubUseIdx != 0 && MRI->shouldTrackSubRegLiveness(DstReg)) {
        if (!DstInt->hasSubRanges()) {
          BumpPtrAllocator &Allocator = LIS->getVNInfoAllocator();
          LaneBitmask FullMask = MRI->getMaxLaneMaskForVReg(DstInt->reg());
          LaneBitmask UsedLanes = TRI->getSubRegIndexLaneMask(SubIdx);
          LaneBitmask UnusedLanes = FullMask & ~UsedLanes;
          DstInt->createSubRangeFrom(Allocator, UsedLanes, *DstInt);
          // The unused lanes are just empty live-ranges at this point.
          // It is the caller responsibility to set the proper
          // dead segments if there is an actual dead def of the
          // unused lanes. This may happen with rematerialization.
          DstInt->createSubRange(Allocator, UnusedLanes);
        }
        SlotIndex MIIdx = UseMI->isDebugInstr()
          ? LIS->getSlotIndexes()->getIndexBefore(*UseMI)
          : LIS->getInstructionIndex(*UseMI);
        SlotIndex UseIdx = MIIdx.getRegSlot(true);
        addUndefFlag(*DstInt, UseIdx, MO, SubUseIdx);
      }
    }

    if (DstIsPhys)
      MO.substPhysReg(DstReg, *TRI);
    else
      MO.substVirtReg(DstReg, SubIdx, *TRI);
  }

  LLVM_DEBUG({do { } while (false)
    dbgs() << "\t\tupdated: ";do { } while (false)
    if (!UseMI->isDebugInstr())do { } while (false)
      dbgs() << LIS->getInstructionIndex(*UseMI) << "\t";do { } while (false)
    dbgs() << *UseMI;do { } while (false)
  })do { } while (false);
}
1839}

1841bool RegisterCoalescer::canJoinPhys(const CoalescerPair &CP) {
// Always join simple intervals that are defined by a single copy from a
// reserved register. This doesn't increase register pressure, so it is
// always beneficial.
if (!MRI->isReserved(CP.getDstReg())) {
  LLVM_DEBUG(dbgs() << "\tCan only merge into reserved registers.\n")do { } while (false);
  return false;
}

LiveInterval &JoinVInt = LIS->getInterval(CP.getSrcReg());
if (JoinVInt.containsOneValue())
  return true;

LLVM_DEBUG(do { } while (false)
    dbgs() << "\tCannot join complex intervals into reserved register.\n")do { } while (false);
return false;
1857}

1859bool RegisterCoalescer::copyValueUndefInPredecessors(
  LiveRange &S, const MachineBasicBlock *MBB, LiveQueryResult SLRQ) {
for (const MachineBasicBlock *Pred : MBB->predecessors()) {
  SlotIndex PredEnd = LIS->getMBBEndIdx(Pred);
  if (VNInfo *V = S.getVNInfoAt(PredEnd.getPrevSlot())) {
    // If this is a self loop, we may be reading the same value.
    if (V->id != SLRQ.valueOutOrDead()->id)
      return false;
  }
}

return true;
1871}

1873void RegisterCoalescer::setUndefOnPrunedSubRegUses(LiveInterval &LI,
                                                 Register Reg,
                                                 LaneBitmask PrunedLanes) {
// If we had other instructions in the segment reading the undef sublane
// value, we need to mark them with undef.
for (MachineOperand &MO : MRI->use_nodbg_operands(Reg)) {
  unsigned SubRegIdx = MO.getSubReg();
  if (SubRegIdx == 0 || MO.isUndef())
    continue;

  LaneBitmask SubRegMask = TRI->getSubRegIndexLaneMask(SubRegIdx);
  SlotIndex Pos = LIS->getInstructionIndex(*MO.getParent());
  for (LiveInterval::SubRange &S : LI.subranges()) {
    if (!S.liveAt(Pos) && (PrunedLanes & SubRegMask).any()) {
      MO.setIsUndef();
      break;
    }
  }
}

LI.removeEmptySubRanges();

// A def of a subregister may be a use of other register lanes. Replacing
// such a def with a def of a different register will eliminate the use,
// and may cause the recorded live range to be larger than the actual
// liveness in the program IR.
LIS->shrinkToUses(&LI);
1900}

1902bool RegisterCoalescer::joinCopy(MachineInstr *CopyMI, bool &Again) {
Again = false;
LLVM_DEBUG(dbgs() << LIS->getInstructionIndex(*CopyMI) << '\t' << *CopyMI)do { } while (false);

CoalescerPair CP(*TRI);
if (!CP.setRegisters(CopyMI)) {
  LLVM_DEBUG(dbgs() << "\tNot coalescable.\n")do { } while (false);
  return false;
}

if (CP.getNewRC()) {
  auto SrcRC = MRI->getRegClass(CP.getSrcReg());
  auto DstRC = MRI->getRegClass(CP.getDstReg());
  unsigned SrcIdx = CP.getSrcIdx();
  unsigned DstIdx = CP.getDstIdx();
  if (CP.isFlipped()) {
    std::swap(SrcIdx, DstIdx);
    std::swap(SrcRC, DstRC);
  }
  if (!TRI->shouldCoalesce(CopyMI, SrcRC, SrcIdx, DstRC, DstIdx,
                           CP.getNewRC(), *LIS)) {
    LLVM_DEBUG(dbgs() << "\tSubtarget bailed on coalescing.\n")do { } while (false);
    return false;
  }
}

// Dead code elimination. This really should be handled by MachineDCE, but
// sometimes dead copies slip through, and we can't generate invalid live
// ranges.
if (!CP.isPhys() && CopyMI->allDefsAreDead()) {
  LLVM_DEBUG(dbgs() << "\tCopy is dead.\n")do { } while (false);
  DeadDefs.push_back(CopyMI);
  eliminateDeadDefs();
  return true;
}

// Eliminate undefs.
if (!CP.isPhys()) {
  // If this is an IMPLICIT_DEF, leave it alone, but don't try to coalesce.
  if (MachineInstr *UndefMI = eliminateUndefCopy(CopyMI)) {
    if (UndefMI->isImplicitDef())
      return false;
    deleteInstr(CopyMI);
    return false;  // Not coalescable.
  }
}

// Coalesced copies are normally removed immediately, but transformations
// like removeCopyByCommutingDef() can inadvertently create identity copies.
// When that happens, just join the values and remove the copy.
if (CP.getSrcReg() == CP.getDstReg()) {
  LiveInterval &LI = LIS->getInterval(CP.getSrcReg());
  LLVM_DEBUG(dbgs() << "\tCopy already coalesced: " << LI << '\n')do { } while (false);
  const SlotIndex CopyIdx = LIS->getInstructionIndex(*CopyMI);
  LiveQueryResult LRQ = LI.Query(CopyIdx);
  if (VNInfo *DefVNI = LRQ.valueDefined()) {
    VNInfo *ReadVNI = LRQ.valueIn();
    assert(ReadVNI && "No value before copy and no <undef> flag.")(static_cast<void> (0));
    assert(ReadVNI != DefVNI && "Cannot read and define the same value.")(static_cast<void> (0));

    // Track incoming undef lanes we need to eliminate from the subrange.
    LaneBitmask PrunedLanes;
    MachineBasicBlock *MBB = CopyMI->getParent();

    // Process subregister liveranges.
    for (LiveInterval::SubRange &S : LI.subranges()) {
      LiveQueryResult SLRQ = S.Query(CopyIdx);
      if (VNInfo *SDefVNI = SLRQ.valueDefined()) {
        if (VNInfo *SReadVNI = SLRQ.valueIn())
          SDefVNI = S.MergeValueNumberInto(SDefVNI, SReadVNI);

        // If this copy introduced an undef subrange from an incoming value,
        // we need to eliminate the undef live in values from the subrange.
        if (copyValueUndefInPredecessors(S, MBB, SLRQ)) {
          LLVM_DEBUG(dbgs() << "Incoming sublane value is undef at copy\n")do { } while (false);
          PrunedLanes |= S.LaneMask;
          S.removeValNo(SDefVNI);
        }
      }
    }

    LI.MergeValueNumberInto(DefVNI, ReadVNI);
    if (PrunedLanes.any()) {
      LLVM_DEBUG(dbgs() << "Pruning undef incoming lanes: "do { } while (false)
                        << PrunedLanes << '\n')do { } while (false);
      setUndefOnPrunedSubRegUses(LI, CP.getSrcReg(), PrunedLanes);
    }

    LLVM_DEBUG(dbgs() << "\tMerged values:          " << LI << '\n')do { } while (false);
  }
  deleteInstr(CopyMI);
  return true;
}

// Enforce policies.
if (CP.isPhys()) {
  LLVM_DEBUG(dbgs() << "\tConsidering merging "do { } while (false)
                    << printReg(CP.getSrcReg(), TRI) << " with "do { } while (false)
                    << printReg(CP.getDstReg(), TRI, CP.getSrcIdx()) << '\n')do { } while (false);
  if (!canJoinPhys(CP)) {
    // Before giving up coalescing, if definition of source is defined by
    // trivial computation, try rematerializing it.
    bool IsDefCopy = false;
    if (reMaterializeTrivialDef(CP, CopyMI, IsDefCopy))
      return true;
    if (IsDefCopy)
      Again = true;  // May be possible to coalesce later.
    return false;
  }
} else {
  // When possible, let DstReg be the larger interval.
  if (!CP.isPartial() && LIS->getInterval(CP.getSrcReg()).size() >
                         LIS->getInterval(CP.getDstReg()).size())
    CP.flip();

  LLVM_DEBUG({do { } while (false)
    dbgs() << "\tConsidering merging to "do { } while (false)
           << TRI->getRegClassName(CP.getNewRC()) << " with ";do { } while (false)
    if (CP.getDstIdx() && CP.getSrcIdx())do { } while (false)
      dbgs() << printReg(CP.getDstReg()) << " in "do { } while (false)
             << TRI->getSubRegIndexName(CP.getDstIdx()) << " and "do { } while (false)
             << printReg(CP.getSrcReg()) << " in "do { } while (false)
             << TRI->getSubRegIndexName(CP.getSrcIdx()) << '\n';do { } while (false)
    elsedo { } while (false)
      dbgs() << printReg(CP.getSrcReg(), TRI) << " in "do { } while (false)
             << printReg(CP.getDstReg(), TRI, CP.getSrcIdx()) << '\n';do { } while (false)
  })do { } while (false);
}

ShrinkMask = LaneBitmask::getNone();
ShrinkMainRange = false;

// Okay, attempt to join these two intervals.  On failure, this returns false.
// Otherwise, if one of the intervals being joined is a physreg, this method
// always canonicalizes DstInt to be it.  The output "SrcInt" will not have
// been modified, so we can use this information below to update aliases.
if (!joinIntervals(CP)) {
  // Coalescing failed.

  // If definition of source is defined by trivial computation, try
  // rematerializing it.
  bool IsDefCopy = false;
  if (reMaterializeTrivialDef(CP, CopyMI, IsDefCopy))
    return true;

  // If we can eliminate the copy without merging the live segments, do so
  // now.
  if (!CP.isPartial() && !CP.isPhys()) {
    bool Changed = adjustCopiesBackFrom(CP, CopyMI);
    bool Shrink = false;
    if (!Changed)
      std::tie(Changed, Shrink) = removeCopyByCommutingDef(CP, CopyMI);
    if (Changed) {
      deleteInstr(CopyMI);
      if (Shrink) {
        Register DstReg = CP.isFlipped() ? CP.getSrcReg() : CP.getDstReg();
        LiveInterval &DstLI = LIS->getInterval(DstReg);
        shrinkToUses(&DstLI);
        LLVM_DEBUG(dbgs() << "\t\tshrunk:   " << DstLI << '\n')do { } while (false);
      }
      LLVM_DEBUG(dbgs() << "\tTrivial!\n")do { } while (false);
      return true;
    }
  }

  // Try and see if we can partially eliminate the copy by moving the copy to
  // its predecessor.
  if (!CP.isPartial() && !CP.isPhys())
    if (removePartialRedundancy(CP, *CopyMI))
      return true;

  // Otherwise, we are unable to join the intervals.
  LLVM_DEBUG(dbgs() << "\tInterference!\n")do { } while (false);
  Again = true;  // May be possible to coalesce later.
  return false;
}

// Coalescing to a virtual register that is of a sub-register class of the
// other. Make sure the resulting register is set to the right register class.
if (CP.isCrossClass()) {
  ++numCrossRCs;
  MRI->setRegClass(CP.getDstReg(), CP.getNewRC());
}

// Removing sub-register copies can ease the register class constraints.
// Make sure we attempt to inflate the register class of DstReg.
if (!CP.isPhys() && RegClassInfo.isProperSubClass(CP.getNewRC()))
  InflateRegs.push_back(CP.getDstReg());

// CopyMI has been erased by joinIntervals at this point. Remove it from
// ErasedInstrs since copyCoalesceWorkList() won't add a successful join back
// to the work list. This keeps ErasedInstrs from growing needlessly.
ErasedInstrs.erase(CopyMI);

// Rewrite all SrcReg operands to DstReg.
// Also update DstReg operands to include DstIdx if it is set.
if (CP.getDstIdx())
  updateRegDefsUses(CP.getDstReg(), CP.getDstReg(), CP.getDstIdx());
updateRegDefsUses(CP.getSrcReg(), CP.getDstReg(), CP.getSrcIdx());

// Shrink subregister ranges if necessary.
if (ShrinkMask.any()) {
  LiveInterval &LI = LIS->getInterval(CP.getDstReg());
  for (LiveInterval::SubRange &S : LI.subranges()) {
    if ((S.LaneMask & ShrinkMask).none())
      continue;
    LLVM_DEBUG(dbgs() << "Shrink LaneUses (Lane " << PrintLaneMask(S.LaneMask)do { } while (false)
                      << ")\n")do { } while (false);
    LIS->shrinkToUses(S, LI.reg());
  }
  LI.removeEmptySubRanges();
}

// CP.getSrcReg()'s live interval has been merged into CP.getDstReg's live
// interval. Since CP.getSrcReg() is in ToBeUpdated set and its live interval
// is not up-to-date, need to update the merged live interval here.
if (ToBeUpdated.count(CP.getSrcReg()))
  ShrinkMainRange = true;

if (ShrinkMainRange) {
  LiveInterval &LI = LIS->getInterval(CP.getDstReg());
  shrinkToUses(&LI);
}

// SrcReg is guaranteed to be the register whose live interval that is
// being merged.
LIS->removeInterval(CP.getSrcReg());

// Update regalloc hint.
TRI->updateRegAllocHint(CP.getSrcReg(), CP.getDstReg(), *MF);

LLVM_DEBUG({do { } while (false)
  dbgs() << "\tSuccess: " << printReg(CP.getSrcReg(), TRI, CP.getSrcIdx())do { } while (false)
         << " -> " << printReg(CP.getDstReg(), TRI, CP.getDstIdx()) << '\n';do { } while (false)
  dbgs() << "\tResult = ";do { } while (false)
  if (CP.isPhys())do { } while (false)
    dbgs() << printReg(CP.getDstReg(), TRI);do { } while (false)
  elsedo { } while (false)
    dbgs() << LIS->getInterval(CP.getDstReg());do { } while (false)
  dbgs() << '\n';do { } while (false)
})do { } while (false);

++numJoins;
return true;
2146}

2148bool RegisterCoalescer::joinReservedPhysReg(CoalescerPair &CP) {
Register DstReg = CP.getDstReg();
Register SrcReg = CP.getSrcReg();
assert(CP.isPhys() && "Must be a physreg copy")(static_cast<void> (0));
assert(MRI->isReserved(DstReg) && "Not a reserved register")(static_cast<void> (0));
LiveInterval &RHS = LIS->getInterval(SrcReg);
LLVM_DEBUG(dbgs() << "\t\tRHS = " << RHS << '\n')do { } while (false);

assert(RHS.containsOneValue() && "Invalid join with reserved register")(static_cast<void> (0));

// Optimization for reserved registers like ESP. We can only merge with a
// reserved physreg if RHS has a single value that is a copy of DstReg.
// The live range of the reserved register will look like a set of dead defs
// - we don't properly track the live range of reserved registers.

// Deny any overlapping intervals.  This depends on all the reserved
// register live ranges to look like dead defs.
if (!MRI->isConstantPhysReg(DstReg)) {
  for (MCRegUnitIterator UI(DstReg, TRI); UI.isValid(); ++UI) {
    // Abort if not all the regunits are reserved.
    for (MCRegUnitRootIterator RI(*UI, TRI); RI.isValid(); ++RI) {
      if (!MRI->isReserved(*RI))
        return false;
    }
    if (RHS.overlaps(LIS->getRegUnit(*UI))) {
      LLVM_DEBUG(dbgs() << "\t\tInterference: " << printRegUnit(*UI, TRI)do { } while (false)
                        << '\n')do { } while (false);
      return false;
    }
  }

  // We must also check for overlaps with regmask clobbers.
  BitVector RegMaskUsable;
  if (LIS->checkRegMaskInterference(RHS, RegMaskUsable) &&
      !RegMaskUsable.test(DstReg)) {
    LLVM_DEBUG(dbgs() << "\t\tRegMask interference\n")do { } while (false);
    return false;
  }
}

// Skip any value computations, we are not adding new values to the
// reserved register.  Also skip merging the live ranges, the reserved
// register live range doesn't need to be accurate as long as all the
// defs are there.

// Delete the identity copy.
MachineInstr *CopyMI;
if (CP.isFlipped()) {
  // Physreg is copied into vreg
  //   %y = COPY %physreg_x
  //   ...  //< no other def of %physreg_x here
  //   use %y
  // =>
  //   ...
  //   use %physreg_x
  CopyMI = MRI->getVRegDef(SrcReg);
} else {
  // VReg is copied into physreg:
  //   %y = def
  //   ... //< no other def or use of %physreg_x here
  //   %physreg_x = COPY %y
  // =>
  //   %physreg_x = def
  //   ...
  if (!MRI->hasOneNonDBGUse(SrcReg)) {
    LLVM_DEBUG(dbgs() << "\t\tMultiple vreg uses!\n")do { } while (false);
    return false;
  }

  if (!LIS->intervalIsInOneMBB(RHS)) {
    LLVM_DEBUG(dbgs() << "\t\tComplex control flow!\n")do { } while (false);
    return false;
  }

  MachineInstr &DestMI = *MRI->getVRegDef(SrcReg);
  CopyMI = &*MRI->use_instr_nodbg_begin(SrcReg);
  SlotIndex CopyRegIdx = LIS->getInstructionIndex(*CopyMI).getRegSlot();
  SlotIndex DestRegIdx = LIS->getInstructionIndex(DestMI).getRegSlot();

  if (!MRI->isConstantPhysReg(DstReg)) {
    // We checked above that there are no interfering defs of the physical
    // register. However, for this case, where we intend to move up the def of
    // the physical register, we also need to check for interfering uses.
    SlotIndexes *Indexes = LIS->getSlotIndexes();
    for (SlotIndex SI = Indexes->getNextNonNullIndex(DestRegIdx);
         SI != CopyRegIdx; SI = Indexes->getNextNonNullIndex(SI)) {
      MachineInstr *MI = LIS->getInstructionFromIndex(SI);
      if (MI->readsRegister(DstReg, TRI)) {
        LLVM_DEBUG(dbgs() << "\t\tInterference (read): " << *MI)do { } while (false);
        return false;
      }
    }
  }

  // We're going to remove the copy which defines a physical reserved
  // register, so remove its valno, etc.
  LLVM_DEBUG(dbgs() << "\t\tRemoving phys reg def of "do { } while (false)
                    << printReg(DstReg, TRI) << " at " << CopyRegIdx << "\n")do { } while (false);

  LIS->removePhysRegDefAt(DstReg.asMCReg(), CopyRegIdx);
  // Create a new dead def at the new def location.
  for (MCRegUnitIterator UI(DstReg, TRI); UI.isValid(); ++UI) {
    LiveRange &LR = LIS->getRegUnit(*UI);
    LR.createDeadDef(DestRegIdx, LIS->getVNInfoAllocator());
  }
}

deleteInstr(CopyMI);

// We don't track kills for reserved registers.
MRI->clearKillFlags(CP.getSrcReg());

return true;
2261}

2263//===----------------------------------------------------------------------===//
2264//                 Interference checking and interval joining
2265//===----------------------------------------------------------------------===//
2266//
2267// In the easiest case, the two live ranges being joined are disjoint, and
2268// there is no interference to consider. It is quite common, though, to have
2269// overlapping live ranges, and we need to check if the interference can be
2270// resolved.
2271//
2272// The live range of a single SSA value forms a sub-tree of the dominator tree.
2273// This means that two SSA values overlap if and only if the def of one value
2274// is contained in the live range of the other value. As a special case, the
2275// overlapping values can be defined at the same index.
2276//
2277// The interference from an overlapping def can be resolved in these cases:
2278//
2279// 1. Coalescable copies. The value is defined by a copy that would become an
2280//    identity copy after joining SrcReg and DstReg. The copy instruction will
2281//    be removed, and the value will be merged with the source value.
2282//
2283//    There can be several copies back and forth, causing many values to be
2284//    merged into one. We compute a list of ultimate values in the joined live
2285//    range as well as a mappings from the old value numbers.
2286//
2287// 2. IMPLICIT_DEF. This instruction is only inserted to ensure all PHI
2288//    predecessors have a live out value. It doesn't cause real interference,
2289//    and can be merged into the value it overlaps. Like a coalescable copy, it
2290//    can be erased after joining.
2291//
2292// 3. Copy of external value. The overlapping def may be a copy of a value that
2293//    is already in the other register. This is like a coalescable copy, but
2294//    the live range of the source register must be trimmed after erasing the
2295//    copy instruction:
2296//
2297//      %src = COPY %ext
2298//      %dst = COPY %ext  <-- Remove this COPY, trim the live range of %ext.
2299//
2300// 4. Clobbering undefined lanes. Vector registers are sometimes built by
2301//    defining one lane at a time:
2302//
2303//      %dst:ssub0<def,read-undef> = FOO
2304//      %src = BAR
2305//      %dst:ssub1 = COPY %src
2306//
2307//    The live range of %src overlaps the %dst value defined by FOO, but
2308//    merging %src into %dst:ssub1 is only going to clobber the ssub1 lane
2309//    which was undef anyway.
2310//
2311//    The value mapping is more complicated in this case. The final live range
2312//    will have different value numbers for both FOO and BAR, but there is no
2313//    simple mapping from old to new values. It may even be necessary to add
2314//    new PHI values.
2315//
2316// 5. Clobbering dead lanes. A def may clobber a lane of a vector register that
2317//    is live, but never read. This can happen because we don't compute
2318//    individual live ranges per lane.
2319//
2320//      %dst = FOO
2321//      %src = BAR
2322//      %dst:ssub1 = COPY %src
2323//
2324//    This kind of interference is only resolved locally. If the clobbered
2325//    lane value escapes the block, the join is aborted.

2327namespace {

2329/// Track information about values in a single virtual register about to be
2330/// joined. Objects of this class are always created in pairs - one for each
2331/// side of the CoalescerPair (or one for each lane of a side of the coalescer
2332/// pair)
2333class JoinVals {
/// Live range we work on.
LiveRange &LR;

/// (Main) register we work on.
const Register Reg;

/// Reg (and therefore the values in this liverange) will end up as
/// subregister SubIdx in the coalesced register. Either CP.DstIdx or
/// CP.SrcIdx.
const unsigned SubIdx;

/// The LaneMask that this liverange will occupy the coalesced register. May
/// be smaller than the lanemask produced by SubIdx when merging subranges.
const LaneBitmask LaneMask;

/// This is true when joining sub register ranges, false when joining main
/// ranges.
const bool SubRangeJoin;

/// Whether the current LiveInterval tracks subregister liveness.
const bool TrackSubRegLiveness;

/// Values that will be present in the final live range.
SmallVectorImpl<VNInfo*> &NewVNInfo;

const CoalescerPair &CP;
LiveIntervals *LIS;
SlotIndexes *Indexes;
const TargetRegisterInfo *TRI;

/// Value number assignments. Maps value numbers in LI to entries in
/// NewVNInfo. This is suitable for passing to LiveInterval::join().
SmallVector<int, 8> Assignments;

public:
/// Conflict resolution for overlapping values.
enum ConflictResolution {
  /// No overlap, simply keep this value.
  CR_Keep,

  /// Merge this value into OtherVNI and erase the defining instruction.
  /// Used for IMPLICIT_DEF, coalescable copies, and copies from external
  /// values.
  CR_Erase,

  /// Merge this value into OtherVNI but keep the defining instruction.
  /// This is for the special case where OtherVNI is defined by the same
  /// instruction.
  CR_Merge,

  /// Keep this value, and have it replace OtherVNI where possible. This
  /// complicates value mapping since OtherVNI maps to two different values
  /// before and after this def.
  /// Used when clobbering undefined or dead lanes.
  CR_Replace,

  /// Unresolved conflict. Visit later when all values have been mapped.
  CR_Unresolved,

  /// Unresolvable conflict. Abort the join.
  CR_Impossible
};

private:
/// Per-value info for LI. The lane bit masks are all relative to the final
/// joined register, so they can be compared directly between SrcReg and
/// DstReg.
struct Val {
  ConflictResolution Resolution = CR_Keep;

  /// Lanes written by this def, 0 for unanalyzed values.
  LaneBitmask WriteLanes;

  /// Lanes with defined values in this register. Other lanes are undef and
  /// safe to clobber.
  LaneBitmask ValidLanes;

  /// Value in LI being redefined by this def.
  VNInfo *RedefVNI = nullptr;

  /// Value in the other live range that overlaps this def, if any.
  VNInfo *OtherVNI = nullptr;

  /// Is this value an IMPLICIT_DEF that can be erased?
  ///
  /// IMPLICIT_DEF values should only exist at the end of a basic block that
  /// is a predecessor to a phi-value. These IMPLICIT_DEF instructions can be
  /// safely erased if they are overlapping a live value in the other live
  /// interval.
  ///
  /// Weird control flow graphs and incomplete PHI handling in
  /// ProcessImplicitDefs can very rarely create IMPLICIT_DEF values with
  /// longer live ranges. Such IMPLICIT_DEF values should be treated like
  /// normal values.
  bool ErasableImplicitDef = false;

  /// True when the live range of this value will be pruned because of an
  /// overlapping CR_Replace value in the other live range.
  bool Pruned = false;

  /// True once Pruned above has been computed.
  bool PrunedComputed = false;

  /// True if this value is determined to be identical to OtherVNI
  /// (in valuesIdentical). This is used with CR_Erase where the erased
  /// copy is redundant, i.e. the source value is already the same as
  /// the destination. In such cases the subranges need to be updated
  /// properly. See comment at pruneSubRegValues for more info.
  bool Identical = false;

  Val() = default;

  bool isAnalyzed() const { return WriteLanes.any(); }
};

/// One entry per value number in LI.
SmallVector<Val, 8> Vals;

/// Compute the bitmask of lanes actually written by DefMI.
/// Set Redef if there are any partial register definitions that depend on the
/// previous value of the register.
LaneBitmask computeWriteLanes(const MachineInstr *DefMI, bool &Redef) const;

/// Find the ultimate value that VNI was copied from.
std::pair<const VNInfo *, Register> followCopyChain(const VNInfo *VNI) const;

bool valuesIdentical(VNInfo *Value0, VNInfo *Value1, const JoinVals &Other) const;

/// Analyze ValNo in this live range, and set all fields of Vals[ValNo].
/// Return a conflict resolution when possible, but leave the hard cases as
/// CR_Unresolved.
/// Recursively calls computeAssignment() on this and Other, guaranteeing that
/// both OtherVNI and RedefVNI have been analyzed and mapped before returning.
/// The recursion always goes upwards in the dominator tree, making loops
/// impossible.
ConflictResolution analyzeValue(unsigned ValNo, JoinVals &Other);

/// Compute the value assignment for ValNo in RI.
/// This may be called recursively by analyzeValue(), but never for a ValNo on
/// the stack.
void computeAssignment(unsigned ValNo, JoinVals &Other);

/// Assuming ValNo is going to clobber some valid lanes in Other.LR, compute
/// the extent of the tainted lanes in the block.
///
/// Multiple values in Other.LR can be affected since partial redefinitions
/// can preserve previously tainted lanes.
///
///   1 %dst = VLOAD           <-- Define all lanes in %dst
///   2 %src = FOO             <-- ValNo to be joined with %dst:ssub0
///   3 %dst:ssub1 = BAR       <-- Partial redef doesn't clear taint in ssub0
///   4 %dst:ssub0 = COPY %src <-- Conflict resolved, ssub0 wasn't read
///
/// For each ValNo in Other that is affected, add an (EndIndex, TaintedLanes)
/// entry to TaintedVals.
///
/// Returns false if the tainted lanes extend beyond the basic block.
bool
taintExtent(unsigned ValNo, LaneBitmask TaintedLanes, JoinVals &Other,
            SmallVectorImpl<std::pair<SlotIndex, LaneBitmask>> &TaintExtent);

/// Return true if MI uses any of the given Lanes from Reg.
/// This does not include partial redefinitions of Reg.
bool usesLanes(const MachineInstr &MI, Register, unsigned, LaneBitmask) const;

/// Determine if ValNo is a copy of a value number in LR or Other.LR that will
/// be pruned:
///
///   %dst = COPY %src
///   %src = COPY %dst  <-- This value to be pruned.
///   %dst = COPY %src  <-- This value is a copy of a pruned value.
bool isPrunedValue(unsigned ValNo, JoinVals &Other);

2507public:
JoinVals(LiveRange &LR, Register Reg, unsigned SubIdx, LaneBitmask LaneMask,
         SmallVectorImpl<VNInfo *> &newVNInfo, const CoalescerPair &cp,
         LiveIntervals *lis, const TargetRegisterInfo *TRI, bool SubRangeJoin,
         bool TrackSubRegLiveness)
    : LR(LR), Reg(Reg), SubIdx(SubIdx), LaneMask(LaneMask),
      SubRangeJoin(SubRangeJoin), TrackSubRegLiveness(TrackSubRegLiveness),
      NewVNInfo(newVNInfo), CP(cp), LIS(lis), Indexes(LIS->getSlotIndexes()),
      TRI(TRI), Assignments(LR.getNumValNums(), -1),
      Vals(LR.getNumValNums()) {}

/// Analyze defs in LR and compute a value mapping in NewVNInfo.
/// Returns false if any conflicts were impossible to resolve.
bool mapValues(JoinVals &Other);

/// Try to resolve conflicts that require all values to be mapped.
/// Returns false if any conflicts were impossible to resolve.
bool resolveConflicts(JoinVals &Other);

/// Prune the live range of values in Other.LR where they would conflict with
/// CR_Replace values in LR. Collect end points for restoring the live range
/// after joining.
void pruneValues(JoinVals &Other, SmallVectorImpl<SlotIndex> &EndPoints,
                 bool changeInstrs);

/// Removes subranges starting at copies that get removed. This sometimes
/// happens when undefined subranges are copied around. These ranges contain
/// no useful information and can be removed.
void pruneSubRegValues(LiveInterval &LI, LaneBitmask &ShrinkMask);

/// Pruning values in subranges can lead to removing segments in these
/// subranges started by IMPLICIT_DEFs. The corresponding segments in
/// the main range also need to be removed. This function will mark
/// the corresponding values in the main range as pruned, so that
/// eraseInstrs can do the final cleanup.
/// The parameter @p LI must be the interval whose main range is the
/// live range LR.
void pruneMainSegments(LiveInterval &LI, bool &ShrinkMainRange);

/// Erase any machine instructions that have been coalesced away.
/// Add erased instructions to ErasedInstrs.
/// Add foreign virtual registers to ShrinkRegs if their live range ended at
/// the erased instrs.
void eraseInstrs(SmallPtrSetImpl<MachineInstr*> &ErasedInstrs,
                 SmallVectorImpl<Register> &ShrinkRegs,
                 LiveInterval *LI = nullptr);

/// Remove liverange defs at places where implicit defs will be removed.
void removeImplicitDefs();

/// Get the value assignments suitable for passing to LiveInterval::join.
const int *getAssignments() const { return Assignments.data(); }

/// Get the conflict resolution for a value number.
ConflictResolution getResolution(unsigned Num) const {
  return Vals[Num].Resolution;
}
2564};

2566} // end anonymous namespace

2568LaneBitmask JoinVals::computeWriteLanes(const MachineInstr *DefMI, bool &Redef)
const {
LaneBitmask L;
for (const MachineOperand &MO : DefMI->operands()) {
  if (!MO.isReg() || MO.getReg() != Reg || !MO.isDef())
    continue;
  L |= TRI->getSubRegIndexLaneMask(
         TRI->composeSubRegIndices(SubIdx, MO.getSubReg()));
  if (MO.readsReg())
    Redef = true;
}
return L;
2580}

2582std::pair<const VNInfo *, Register>
2583JoinVals::followCopyChain(const VNInfo *VNI) const {
Register TrackReg = Reg;

while (!VNI->isPHIDef()) {
2
←
Calling 'VNInfo::isPHIDef'→
8
←
Returning from 'VNInfo::isPHIDef'→
9
←
Loop condition is true.  Entering loop body→
24
←
Called C++ object pointer is null
  SlotIndex Def = VNI->def;
  MachineInstr *MI = Indexes->getInstructionFromIndex(Def);
  assert(MI && "No defining instruction")(static_cast<void> (0));
  if (!MI->isFullCopy())
10
←
Taking false branch→
    return std::make_pair(VNI, TrackReg);
  Register SrcReg = MI->getOperand(1).getReg();
  if (!SrcReg.isVirtual())
11
←
Taking false branch→
    return std::make_pair(VNI, TrackReg);

  const LiveInterval &LI = LIS->getInterval(SrcReg);
  const VNInfo *ValueIn;
  // No subrange involved.
  if (!SubRangeJoin || !LI.hasSubRanges()) {
12
←
Assuming field 'SubRangeJoin' is false→
    LiveQueryResult LRQ = LI.Query(Def);
    ValueIn = LRQ.valueIn();
  } else {
    // Query subranges. Ensure that all matching ones take us to the same def
    // (allowing some of them to be undef).
    ValueIn = nullptr;
    for (const LiveInterval::SubRange &S : LI.subranges()) {
      // Transform lanemask to a mask in the joined live interval.
      LaneBitmask SMask = TRI->composeSubRegIndexLaneMask(SubIdx, S.LaneMask);
      if ((SMask & LaneMask).none())
        continue;
      LiveQueryResult LRQ = S.Query(Def);
      if (!ValueIn) {
        ValueIn = LRQ.valueIn();
        continue;
      }
      if (LRQ.valueIn() && ValueIn != LRQ.valueIn())
        return std::make_pair(VNI, TrackReg);
    }
  }
  if (ValueIn == nullptr) {
13
←
Assuming the condition is true→
14
←
Taking true branch→
    // Reaching an undefined value is legitimate, for example:
    //
    // 1   undef %0.sub1 = ...  ;; %0.sub0 == undef
    // 2   %1 = COPY %0         ;; %1 is defined here.
    // 3   %0 = COPY %1         ;; Now %0.sub0 has a definition,
    //                          ;; but it's equivalent to "undef".
    return std::make_pair(nullptr, SrcReg);
  }
  VNI = ValueIn;
  TrackReg = SrcReg;
}
return std::make_pair(VNI, TrackReg);
2633}

2635bool JoinVals::valuesIdentical(VNInfo *Value0, VNInfo *Value1,
                             const JoinVals &Other) const {
const VNInfo *Orig0;
Register Reg0;
std::tie(Orig0, Reg0) = followCopyChain(Value0);
1
Calling 'JoinVals::followCopyChain'→
15
←
Returning from 'JoinVals::followCopyChain'→
if (Orig0 == Value1 && Reg0 == Other.Reg)
16
←
Assuming 'Orig0' is equal to 'Value1'→
17
←
Calling 'Register::operator=='→
20
←
Returning from 'Register::operator=='→
21
←
Taking false branch→
  return true;

const VNInfo *Orig1;
Register Reg1;
std::tie(Orig1, Reg1) = Other.followCopyChain(Value1);
22
←
Passing null pointer value via 1st parameter 'VNI'→
23
←
Calling 'JoinVals::followCopyChain'→
// If both values are undefined, and the source registers are the same
// register, the values are identical. Filter out cases where only one
// value is defined.
if (Orig0 == nullptr || Orig1 == nullptr)
  return Orig0 == Orig1 && Reg0 == Reg1;

// The values are equal if they are defined at the same place and use the
// same register. Note that we cannot compare VNInfos directly as some of
// them might be from a copy created in mergeSubRangeInto()  while the other
// is from the original LiveInterval.
return Orig0->def == Orig1->def && Reg0 == Reg1;
2657}

2659JoinVals::ConflictResolution
2660JoinVals::analyzeValue(unsigned ValNo, JoinVals &Other) {
Val &V = Vals[ValNo];
assert(!V.isAnalyzed() && "Value has already been analyzed!")(static_cast<void> (0));
VNInfo *VNI = LR.getValNumInfo(ValNo);
if (VNI->isUnused()) {
  V.WriteLanes = LaneBitmask::getAll();
  return CR_Keep;
}

// Get the instruction defining this value, compute the lanes written.
const MachineInstr *DefMI = nullptr;
if (VNI->isPHIDef()) {
  // Conservatively assume that all lanes in a PHI are valid.
  LaneBitmask Lanes = SubRangeJoin ? LaneBitmask::getLane(0)
                                   : TRI->getSubRegIndexLaneMask(SubIdx);
  V.ValidLanes = V.WriteLanes = Lanes;
} else {
  DefMI = Indexes->getInstructionFromIndex(VNI->def);
  assert(DefMI != nullptr)(static_cast<void> (0));
  if (SubRangeJoin) {
    // We don't care about the lanes when joining subregister ranges.
    V.WriteLanes = V.ValidLanes = LaneBitmask::getLane(0);
    if (DefMI->isImplicitDef()) {
      V.ValidLanes = LaneBitmask::getNone();
      V.ErasableImplicitDef = true;
    }
  } else {
    bool Redef = false;
    V.ValidLanes = V.WriteLanes = computeWriteLanes(DefMI, Redef);

    // If this is a read-modify-write instruction, there may be more valid
    // lanes than the ones written by this instruction.
    // This only covers partial redef operands. DefMI may have normal use
    // operands reading the register. They don't contribute valid lanes.
    //
    // This adds ssub1 to the set of valid lanes in %src:
    //
    //   %src:ssub1 = FOO
    //
    // This leaves only ssub1 valid, making any other lanes undef:
    //
    //   %src:ssub1<def,read-undef> = FOO %src:ssub2
    //
    // The <read-undef> flag on the def operand means that old lane values are
    // not important.
    if (Redef) {
      V.RedefVNI = LR.Query(VNI->def).valueIn();
      assert((TrackSubRegLiveness || V.RedefVNI) &&(static_cast<void> (0))
             "Instruction is reading nonexistent value")(static_cast<void> (0));
      if (V.RedefVNI != nullptr) {
        computeAssignment(V.RedefVNI->id, Other);
        V.ValidLanes |= Vals[V.RedefVNI->id].ValidLanes;
      }
    }

    // An IMPLICIT_DEF writes undef values.
    if (DefMI->isImplicitDef()) {
      // We normally expect IMPLICIT_DEF values to be live only until the end
      // of their block. If the value is really live longer and gets pruned in
      // another block, this flag is cleared again.
      //
      // Clearing the valid lanes is deferred until it is sure this can be
      // erased.
      V.ErasableImplicitDef = true;
    }
  }
}

// Find the value in Other that overlaps VNI->def, if any.
LiveQueryResult OtherLRQ = Other.LR.Query(VNI->def);

// It is possible that both values are defined by the same instruction, or
// the values are PHIs defined in the same block. When that happens, the two
// values should be merged into one, but not into any preceding value.
// The first value defined or visited gets CR_Keep, the other gets CR_Merge.
if (VNInfo *OtherVNI = OtherLRQ.valueDefined()) {
  assert(SlotIndex::isSameInstr(VNI->def, OtherVNI->def) && "Broken LRQ")(static_cast<void> (0));

  // One value stays, the other is merged. Keep the earlier one, or the first
  // one we see.
  if (OtherVNI->def < VNI->def)
    Other.computeAssignment(OtherVNI->id, *this);
  else if (VNI->def < OtherVNI->def && OtherLRQ.valueIn()) {
    // This is an early-clobber def overlapping a live-in value in the other
    // register. Not mergeable.
    V.OtherVNI = OtherLRQ.valueIn();
    return CR_Impossible;
  }
  V.OtherVNI = OtherVNI;
  Val &OtherV = Other.Vals[OtherVNI->id];
  // Keep this value, check for conflicts when analyzing OtherVNI.
  if (!OtherV.isAnalyzed())
    return CR_Keep;
  // Both sides have been analyzed now.
  // Allow overlapping PHI values. Any real interference would show up in a
  // predecessor, the PHI itself can't introduce any conflicts.
  if (VNI->isPHIDef())
    return CR_Merge;
  if ((V.ValidLanes & OtherV.ValidLanes).any())
    // Overlapping lanes can't be resolved.
    return CR_Impossible;
  else
    return CR_Merge;
}

// No simultaneous def. Is Other live at the def?
V.OtherVNI = OtherLRQ.valueIn();
if (!V.OtherVNI)
  // No overlap, no conflict.
  return CR_Keep;

assert(!SlotIndex::isSameInstr(VNI->def, V.OtherVNI->def) && "Broken LRQ")(static_cast<void> (0));

// We have overlapping values, or possibly a kill of Other.
// Recursively compute assignments up the dominator tree.
Other.computeAssignment(V.OtherVNI->id, *this);
Val &OtherV = Other.Vals[V.OtherVNI->id];

if (OtherV.ErasableImplicitDef) {
  // Check if OtherV is an IMPLICIT_DEF that extends beyond its basic block.
  // This shouldn't normally happen, but ProcessImplicitDefs can leave such
  // IMPLICIT_DEF instructions behind, and there is nothing wrong with it
  // technically.
  //
  // When it happens, treat that IMPLICIT_DEF as a normal value, and don't try
  // to erase the IMPLICIT_DEF instruction.
  if (DefMI &&
      DefMI->getParent() != Indexes->getMBBFromIndex(V.OtherVNI->def)) {
    LLVM_DEBUG(dbgs() << "IMPLICIT_DEF defined at " << V.OtherVNI->defdo { } while (false)
               << " extends into "do { } while (false)
               << printMBBReference(*DefMI->getParent())do { } while (false)
               << ", keeping it.\n")do { } while (false);
    OtherV.ErasableImplicitDef = false;
  } else {
    // We deferred clearing these lanes in case we needed to save them
    OtherV.ValidLanes &= ~OtherV.WriteLanes;
  }
}

// Allow overlapping PHI values. Any real interference would show up in a
// predecessor, the PHI itself can't introduce any conflicts.
if (VNI->isPHIDef())
  return CR_Replace;

// Check for simple erasable conflicts.
if (DefMI->isImplicitDef())
  return CR_Erase;

// Include the non-conflict where DefMI is a coalescable copy that kills
// OtherVNI. We still want the copy erased and value numbers merged.
if (CP.isCoalescable(DefMI)) {
  // Some of the lanes copied from OtherVNI may be undef, making them undef
  // here too.
  V.ValidLanes &= ~V.WriteLanes | OtherV.ValidLanes;
  return CR_Erase;
}

// This may not be a real conflict if DefMI simply kills Other and defines
// VNI.
if (OtherLRQ.isKill() && OtherLRQ.endPoint() <= VNI->def)
  return CR_Keep;

// Handle the case where VNI and OtherVNI can be proven to be identical:
//
//   %other = COPY %ext
//   %this  = COPY %ext <-- Erase this copy
//
if (DefMI->isFullCopy() && !CP.isPartial() &&
    valuesIdentical(VNI, V.OtherVNI, Other)) {
  V.Identical = true;
  return CR_Erase;
}

// The remaining checks apply to the lanes, which aren't tracked here.  This
// was already decided to be OK via the following CR_Replace condition.
// CR_Replace.
if (SubRangeJoin)
  return CR_Replace;

// If the lanes written by this instruction were all undef in OtherVNI, it is
// still safe to join the live ranges. This can't be done with a simple value
// mapping, though - OtherVNI will map to multiple values:
//
//   1 %dst:ssub0 = FOO                <-- OtherVNI
//   2 %src = BAR                      <-- VNI
//   3 %dst:ssub1 = COPY killed %src    <-- Eliminate this copy.
//   4 BAZ killed %dst
//   5 QUUX killed %src
//
// Here OtherVNI will map to itself in [1;2), but to VNI in [2;5). CR_Replace
// handles this complex value mapping.
if ((V.WriteLanes & OtherV.ValidLanes).none())
  return CR_Replace;

// If the other live range is killed by DefMI and the live ranges are still
// overlapping, it must be because we're looking at an early clobber def:
//
//   %dst<def,early-clobber> = ASM killed %src
//
// In this case, it is illegal to merge the two live ranges since the early
// clobber def would clobber %src before it was read.
if (OtherLRQ.isKill()) {
  // This case where the def doesn't overlap the kill is handled above.
  assert(VNI->def.isEarlyClobber() &&(static_cast<void> (0))
         "Only early clobber defs can overlap a kill")(static_cast<void> (0));
  return CR_Impossible;
}

// VNI is clobbering live lanes in OtherVNI, but there is still the
// possibility that no instructions actually read the clobbered lanes.
// If we're clobbering all the lanes in OtherVNI, at least one must be read.
// Otherwise Other.RI wouldn't be live here.
if ((TRI->getSubRegIndexLaneMask(Other.SubIdx) & ~V.WriteLanes).none())
  return CR_Impossible;

if (TrackSubRegLiveness) {
  auto &OtherLI = LIS->getInterval(Other.Reg);
  // If OtherVNI does not have subranges, it means all the lanes of OtherVNI
  // share the same live range, so we just need to check whether they have
  // any conflict bit in their LaneMask.
  if (!OtherLI.hasSubRanges()) {
    LaneBitmask OtherMask = TRI->getSubRegIndexLaneMask(Other.SubIdx);
    return (OtherMask & V.WriteLanes).none() ? CR_Replace : CR_Impossible;
  }

  // If we are clobbering some active lanes of OtherVNI at VNI->def, it is
  // impossible to resolve the conflict. Otherwise, we can just replace
  // OtherVNI because of no real conflict.
  for (LiveInterval::SubRange &OtherSR : OtherLI.subranges()) {
    LaneBitmask OtherMask =
        TRI->composeSubRegIndexLaneMask(Other.SubIdx, OtherSR.LaneMask);
    if ((OtherMask & V.WriteLanes).none())
      continue;

    auto OtherSRQ = OtherSR.Query(VNI->def);
    if (OtherSRQ.valueIn() && OtherSRQ.endPoint() > VNI->def) {
      // VNI is clobbering some lanes of OtherVNI, they have real conflict.
      return CR_Impossible;
    }
  }

  // VNI is NOT clobbering any lane of OtherVNI, just replace OtherVNI.
  return CR_Replace;
}

// We need to verify that no instructions are reading the clobbered lanes.
// To save compile time, we'll only check that locally. Don't allow the
// tainted value to escape the basic block.
MachineBasicBlock *MBB = Indexes->getMBBFromIndex(VNI->def);
if (OtherLRQ.endPoint() >= Indexes->getMBBEndIdx(MBB))
  return CR_Impossible;

// There are still some things that could go wrong besides clobbered lanes
// being read, for example OtherVNI may be only partially redefined in MBB,
// and some clobbered lanes could escape the block. Save this analysis for
// resolveConflicts() when all values have been mapped. We need to know
// RedefVNI and WriteLanes for any later defs in MBB, and we can't compute
// that now - the recursive analyzeValue() calls must go upwards in the
// dominator tree.
return CR_Unresolved;
2920}

2922void JoinVals::computeAssignment(unsigned ValNo, JoinVals &Other) {
Val &V = Vals[ValNo];
if (V.isAnalyzed()) {
  // Recursion should always move up the dominator tree, so ValNo is not
  // supposed to reappear before it has been assigned.
  assert(Assignments[ValNo] != -1 && "Bad recursion?")(static_cast<void> (0));
  return;
}
switch ((V.Resolution = analyzeValue(ValNo, Other))) {
case CR_Erase:
case CR_Merge:
  // Merge this ValNo into OtherVNI.
  assert(V.OtherVNI && "OtherVNI not assigned, can't merge.")(static_cast<void> (0));
  assert(Other.Vals[V.OtherVNI->id].isAnalyzed() && "Missing recursion")(static_cast<void> (0));
  Assignments[ValNo] = Other.Assignments[V.OtherVNI->id];
  LLVM_DEBUG(dbgs() << "\t\tmerge " << printReg(Reg) << ':' << ValNo << '@'do { } while (false)
                    << LR.getValNumInfo(ValNo)->def << " into "do { } while (false)
                    << printReg(Other.Reg) << ':' << V.OtherVNI->id << '@'do { } while (false)
                    << V.OtherVNI->def << " --> @"do { } while (false)
                    << NewVNInfo[Assignments[ValNo]]->def << '\n')do { } while (false);
  break;
case CR_Replace:
case CR_Unresolved: {
  // The other value is going to be pruned if this join is successful.
  assert(V.OtherVNI && "OtherVNI not assigned, can't prune")(static_cast<void> (0));
  Val &OtherV = Other.Vals[V.OtherVNI->id];
  // We cannot erase an IMPLICIT_DEF if we don't have valid values for all
  // its lanes.
  if (OtherV.ErasableImplicitDef &&
      TrackSubRegLiveness &&
      (OtherV.WriteLanes & ~V.ValidLanes).any()) {
    LLVM_DEBUG(dbgs() << "Cannot erase implicit_def with missing values\n")do { } while (false);

    OtherV.ErasableImplicitDef = false;
    // The valid lanes written by the implicit_def were speculatively cleared
    // before, so make this more conservative. It may be better to track this,
    // I haven't found a testcase where it matters.
    OtherV.ValidLanes = LaneBitmask::getAll();
  }

  OtherV.Pruned = true;
  LLVM_FALLTHROUGH[[gnu::fallthrough]];
}
default:
  // This value number needs to go in the final joined live range.
  Assignments[ValNo] = NewVNInfo.size();
  NewVNInfo.push_back(LR.getValNumInfo(ValNo));
  break;
}
2971}

2973bool JoinVals::mapValues(JoinVals &Other) {
for (unsigned i = 0, e = LR.getNumValNums(); i != e; ++i) {
  computeAssignment(i, Other);
  if (Vals[i].Resolution == CR_Impossible) {
    LLVM_DEBUG(dbgs() << "\t\tinterference at " << printReg(Reg) << ':' << ido { } while (false)
                      << '@' << LR.getValNumInfo(i)->def << '\n')do { } while (false);
    return false;
  }
}
return true;
2983}

2985bool JoinVals::
2986taintExtent(unsigned ValNo, LaneBitmask TaintedLanes, JoinVals &Other,
          SmallVectorImpl<std::pair<SlotIndex, LaneBitmask>> &TaintExtent) {
VNInfo *VNI = LR.getValNumInfo(ValNo);
MachineBasicBlock *MBB = Indexes->getMBBFromIndex(VNI->def);
SlotIndex MBBEnd = Indexes->getMBBEndIdx(MBB);

// Scan Other.LR from VNI.def to MBBEnd.
LiveInterval::iterator OtherI = Other.LR.find(VNI->def);
assert(OtherI != Other.LR.end() && "No conflict?")(static_cast<void> (0));
do {
  // OtherI is pointing to a tainted value. Abort the join if the tainted
  // lanes escape the block.
  SlotIndex End = OtherI->end;
  if (End >= MBBEnd) {
    LLVM_DEBUG(dbgs() << "\t\ttaints global " << printReg(Other.Reg) << ':'do { } while (false)
                      << OtherI->valno->id << '@' << OtherI->start << '\n')do { } while (false);
    return false;
  }
  LLVM_DEBUG(dbgs() << "\t\ttaints local " << printReg(Other.Reg) << ':'do { } while (false)
                    << OtherI->valno->id << '@' << OtherI->start << " to "do { } while (false)
                    << End << '\n')do { } while (false);
  // A dead def is not a problem.
  if (End.isDead())
    break;
  TaintExtent.push_back(std::make_pair(End, TaintedLanes));

  // Check for another def in the MBB.
  if (++OtherI == Other.LR.end() || OtherI->start >= MBBEnd)
    break;

  // Lanes written by the new def are no longer tainted.
  const Val &OV = Other.Vals[OtherI->valno->id];
  TaintedLanes &= ~OV.WriteLanes;
  if (!OV.RedefVNI)
    break;
} while (TaintedLanes.any());
return true;
3023}

3025bool JoinVals::usesLanes(const MachineInstr &MI, Register Reg, unsigned SubIdx,
                       LaneBitmask Lanes) const {
if (MI.isDebugOrPseudoInstr())
  return false;
for (const MachineOperand &MO : MI.operands()) {
  if (!MO.isReg() || MO.isDef() || MO.getReg() != Reg)
    continue;
  if (!MO.readsReg())
    continue;
  unsigned S = TRI->composeSubRegIndices(SubIdx, MO.getSubReg());
  if ((Lanes & TRI->getSubRegIndexLaneMask(S)).any())
    return true;
}
return false;
3039}

3041bool JoinVals::resolveConflicts(JoinVals &Other) {
for (unsigned i = 0, e = LR.getNumValNums(); i != e; ++i) {
  Val &V = Vals[i];
  assert(V.Resolution != CR_Impossible && "Unresolvable conflict")(static_cast<void> (0));
  if (V.Resolution != CR_Unresolved)
    continue;
  LLVM_DEBUG(dbgs() << "\t\tconflict at " << printReg(Reg) << ':' << i << '@'do { } while (false)
                    << LR.getValNumInfo(i)->defdo { } while (false)
                    << ' ' << PrintLaneMask(LaneMask) << '\n')do { } while (false);
  if (SubRangeJoin)
    return false;

  ++NumLaneConflicts;
  assert(V.OtherVNI && "Inconsistent conflict resolution.")(static_cast<void> (0));
  VNInfo *VNI = LR.getValNumInfo(i);
  const Val &OtherV = Other.Vals[V.OtherVNI->id];

  // VNI is known to clobber some lanes in OtherVNI. If we go ahead with the
  // join, those lanes will be tainted with a wrong value. Get the extent of
  // the tainted lanes.
  LaneBitmask TaintedLanes = V.WriteLanes & OtherV.ValidLanes;
  SmallVector<std::pair<SlotIndex, LaneBitmask>, 8> TaintExtent;
  if (!taintExtent(i, TaintedLanes, Other, TaintExtent))
    // Tainted lanes would extend beyond the basic block.
    return false;

  assert(!TaintExtent.empty() && "There should be at least one conflict.")(static_cast<void> (0));

  // Now look at the instructions from VNI->def to TaintExtent (inclusive).
  MachineBasicBlock *MBB = Indexes->getMBBFromIndex(VNI->def);
  MachineBasicBlock::iterator MI = MBB->begin();
  if (!VNI->isPHIDef()) {
    MI = Indexes->getInstructionFromIndex(VNI->def);
    if (!VNI->def.isEarlyClobber()) {
      // No need to check the instruction defining VNI for reads.
      ++MI;
    }
  }
  assert(!SlotIndex::isSameInstr(VNI->def, TaintExtent.front().first) &&(static_cast<void> (0))
         "Interference ends on VNI->def. Should have been handled earlier")(static_cast<void> (0));
  MachineInstr *LastMI =
    Indexes->getInstructionFromIndex(TaintExtent.front().first);
  assert(LastMI && "Range must end at a proper instruction")(static_cast<void> (0));
  unsigned TaintNum = 0;
  while (true) {
    assert(MI != MBB->end() && "Bad LastMI")(static_cast<void> (0));
    if (usesLanes(*MI, Other.Reg, Other.SubIdx, TaintedLanes)) {
      LLVM_DEBUG(dbgs() << "\t\ttainted lanes used by: " << *MI)do { } while (false);
      return false;
    }
    // LastMI is the last instruction to use the current value.
    if (&*MI == LastMI) {
      if (++TaintNum == TaintExtent.size())
        break;
      LastMI = Indexes->getInstructionFromIndex(TaintExtent[TaintNum].first);
      assert(LastMI && "Range must end at a proper instruction")(static_cast<void> (0));
      TaintedLanes = TaintExtent[TaintNum].second;
    }
    ++MI;
  }

  // The tainted lanes are unused.
  V.Resolution = CR_Replace;
  ++NumLaneResolves;
}
return true;
3107}

3109bool JoinVals::isPrunedValue(unsigned ValNo, JoinVals &Other) {
Val &V = Vals[ValNo];
if (V.Pruned || V.PrunedComputed)
  return V.Pruned;

if (V.Resolution != CR_Erase && V.Resolution != CR_Merge)
  return V.Pruned;

// Follow copies up the dominator tree and check if any intermediate value
// has been pruned.
V.PrunedComputed = true;
V.Pruned = Other.isPrunedValue(V.OtherVNI->id, *this);
return V.Pruned;
3122}

3124void JoinVals::pruneValues(JoinVals &Other,
                         SmallVectorImpl<SlotIndex> &EndPoints,
                         bool changeInstrs) {
for (unsigned i = 0, e = LR.getNumValNums(); i != e; ++i) {
  SlotIndex Def = LR.getValNumInfo(i)->def;
  switch (Vals[i].Resolution) {
  case CR_Keep:
    break;
  case CR_Replace: {
    // This value takes precedence over the value in Other.LR.
    LIS->pruneValue(Other.LR, Def, &EndPoints);
    // Check if we're replacing an IMPLICIT_DEF value. The IMPLICIT_DEF
    // instructions are only inserted to provide a live-out value for PHI
    // predecessors, so the instruction should simply go away once its value
    // has been replaced.
    Val &OtherV = Other.Vals[Vals[i].OtherVNI->id];
    bool EraseImpDef = OtherV.ErasableImplicitDef &&
                       OtherV.Resolution == CR_Keep;
    if (!Def.isBlock()) {
      if (changeInstrs) {
        // Remove <def,read-undef> flags. This def is now a partial redef.
        // Also remove dead flags since the joined live range will
        // continue past this instruction.
        for (MachineOperand &MO :
             Indexes->getInstructionFromIndex(Def)->operands()) {
          if (MO.isReg() && MO.isDef() && MO.getReg() == Reg) {
            if (MO.getSubReg() != 0 && MO.isUndef() && !EraseImpDef)
              MO.setIsUndef(false);
            MO.setIsDead(false);
          }
        }
      }
      // This value will reach instructions below, but we need to make sure
      // the live range also reaches the instruction at Def.
      if (!EraseImpDef)
        EndPoints.push_back(Def);
    }
    LLVM_DEBUG(dbgs() << "\t\tpruned " << printReg(Other.Reg) << " at " << Defdo { } while (false)
                      << ": " << Other.LR << '\n')do { } while (false);
    break;
  }
  case CR_Erase:
  case CR_Merge:
    if (isPrunedValue(i, Other)) {
      // This value is ultimately a copy of a pruned value in LR or Other.LR.
      // We can no longer trust the value mapping computed by
      // computeAssignment(), the value that was originally copied could have
      // been replaced.
      LIS->pruneValue(LR, Def, &EndPoints);
      LLVM_DEBUG(dbgs() << "\t\tpruned all of " << printReg(Reg) << " at "do { } while (false)
                        << Def << ": " << LR << '\n')do { } while (false);
    }
    break;
  case CR_Unresolved:
  case CR_Impossible:
    llvm_unreachable("Unresolved conflicts")__builtin_unreachable();
  }
}
3182}

3184// Check if the segment consists of a copied live-through value (i.e. the copy
3185// in the block only extended the liveness, of an undef value which we may need
3186// to handle).
3187static bool isLiveThrough(const LiveQueryResult Q) {
return Q.valueIn() && Q.valueIn()->isPHIDef() && Q.valueIn() == Q.valueOut();
3189}

3191/// Consider the following situation when coalescing the copy between
3192/// %31 and %45 at 800. (The vertical lines represent live range segments.)
3193///
3194///                              Main range         Subrange 0004 (sub2)
3195///                              %31    %45           %31    %45
3196///  544    %45 = COPY %28               +                    +
3197///                                      | v1                 | v1
3198///  560B bb.1:                          +                    +
3199///  624        = %45.sub2               | v2                 | v2
3200///  800    %31 = COPY %45        +      +             +      +
3201///                               | v0                 | v0
3202///  816    %31.sub1 = ...        +                    |
3203///  880    %30 = COPY %31        | v1                 +
3204///  928    %45 = COPY %30        |      +                    +
3205///                               |      | v0                 | v0  <--+
3206///  992B   ; backedge -> bb.1    |      +                    +        |
3207/// 1040        = %31.sub0        +                                    |
3208///                                                 This value must remain
3209///                                                 live-out!
3210///
3211/// Assuming that %31 is coalesced into %45, the copy at 928 becomes
3212/// redundant, since it copies the value from %45 back into it. The
3213/// conflict resolution for the main range determines that %45.v0 is
3214/// to be erased, which is ok since %31.v1 is identical to it.
3215/// The problem happens with the subrange for sub2: it has to be live
3216/// on exit from the block, but since 928 was actually a point of
3217/// definition of %45.sub2, %45.sub2 was not live immediately prior
3218/// to that definition. As a result, when 928 was erased, the value v0
3219/// for %45.sub2 was pruned in pruneSubRegValues. Consequently, an
3220/// IMPLICIT_DEF was inserted as a "backedge" definition for %45.sub2,
3221/// providing an incorrect value to the use at 624.
3222///
3223/// Since the main-range values %31.v1 and %45.v0 were proved to be
3224/// identical, the corresponding values in subranges must also be the
3225/// same. A redundant copy is removed because it's not needed, and not
3226/// because it copied an undefined value, so any liveness that originated
3227/// from that copy cannot disappear. When pruning a value that started
3228/// at the removed copy, the corresponding identical value must be
3229/// extended to replace it.
3230void JoinVals::pruneSubRegValues(LiveInterval &LI, LaneBitmask &ShrinkMask) {
// Look for values being erased.
bool DidPrune = false;
for (unsigned i = 0, e = LR.getNumValNums(); i != e; ++i) {
  Val &V = Vals[i];
  // We should trigger in all cases in which eraseInstrs() does something.
  // match what eraseInstrs() is doing, print a message so
  if (V.Resolution != CR_Erase &&
      (V.Resolution != CR_Keep || !V.ErasableImplicitDef || !V.Pruned))
    continue;

  // Check subranges at the point where the copy will be removed.
  SlotIndex Def = LR.getValNumInfo(i)->def;
  SlotIndex OtherDef;
  if (V.Identical)
    OtherDef = V.OtherVNI->def;

  // Print message so mismatches with eraseInstrs() can be diagnosed.
  LLVM_DEBUG(dbgs() << "\t\tExpecting instruction removal at " << Defdo { } while (false)
                    << '\n')do { } while (false);
  for (LiveInterval::SubRange &S : LI.subranges()) {
    LiveQueryResult Q = S.Query(Def);

    // If a subrange starts at the copy then an undefined value has been
    // copied and we must remove that subrange value as well.
    VNInfo *ValueOut = Q.valueOutOrDead();
    if (ValueOut != nullptr && (Q.valueIn() == nullptr ||
                                (V.Identical && V.Resolution == CR_Erase &&
                                 ValueOut->def == Def))) {
      LLVM_DEBUG(dbgs() << "\t\tPrune sublane " << PrintLaneMask(S.LaneMask)do { } while (false)
                        << " at " << Def << "\n")do { } while (false);
      SmallVector<SlotIndex,8> EndPoints;
      LIS->pruneValue(S, Def, &EndPoints);
      DidPrune = true;
      // Mark value number as unused.
      ValueOut->markUnused();

      if (V.Identical && S.Query(OtherDef).valueOutOrDead()) {
        // If V is identical to V.OtherVNI (and S was live at OtherDef),
        // then we can't simply prune V from S. V needs to be replaced
        // with V.OtherVNI.
        LIS->extendToIndices(S, EndPoints);
      }

      // We may need to eliminate the subrange if the copy introduced a live
      // out undef value.
      if (ValueOut->isPHIDef())
        ShrinkMask |= S.LaneMask;
      continue;
    }

    // If a subrange ends at the copy, then a value was copied but only
    // partially used later. Shrink the subregister range appropriately.
    //
    // Ultimately this calls shrinkToUses, so assuming ShrinkMask is
    // conservatively correct.
    if ((Q.valueIn() != nullptr && Q.valueOut() == nullptr) ||
        (V.Resolution == CR_Erase && isLiveThrough(Q))) {
      LLVM_DEBUG(dbgs() << "\t\tDead uses at sublane "do { } while (false)
                        << PrintLaneMask(S.LaneMask) << " at " << Defdo { } while (false)
                        << "\n")do { } while (false);
      ShrinkMask |= S.LaneMask;
    }
  }
}
if (DidPrune)
  LI.removeEmptySubRanges();
3297}

3299/// Check if any of the subranges of @p LI contain a definition at @p Def.
3300static bool isDefInSubRange(LiveInterval &LI, SlotIndex Def) {
for (LiveInterval::SubRange &SR : LI.subranges()) {
  if (VNInfo *VNI = SR.Query(Def).valueOutOrDead())
    if (VNI->def == Def)
      return true;
}
return false;
3307}

3309void JoinVals::pruneMainSegments(LiveInterval &LI, bool &ShrinkMainRange) {
assert(&static_cast<LiveRange&>(LI) == &LR)(static_cast<void> (0));

for (unsigned i = 0, e = LR.getNumValNums(); i != e; ++i) {
  if (Vals[i].Resolution != CR_Keep)
    continue;
  VNInfo *VNI = LR.getValNumInfo(i);
  if (VNI->isUnused() || VNI->isPHIDef() || isDefInSubRange(LI, VNI->def))
    continue;
  Vals[i].Pruned = true;
  ShrinkMainRange = true;
}
3321}

3323void JoinVals::removeImplicitDefs() {
for (unsigned i = 0, e = LR.getNumValNums(); i != e; ++i) {
  Val &V = Vals[i];
  if (V.Resolution != CR_Keep || !V.ErasableImplicitDef || !V.Pruned)
    continue;

  VNInfo *VNI = LR.getValNumInfo(i);
  VNI->markUnused();
  LR.removeValNo(VNI);
}
3333}

3335void JoinVals::eraseInstrs(SmallPtrSetImpl<MachineInstr*> &ErasedInstrs,
                         SmallVectorImpl<Register> &ShrinkRegs,
                         LiveInterval *LI) {
for (unsigned i = 0, e = LR.getNumValNums(); i != e; ++i) {
  // Get the def location before markUnused() below invalidates it.
  VNInfo *VNI = LR.getValNumInfo(i);
  SlotIndex Def = VNI->def;
  switch (Vals[i].Resolution) {
  case CR_Keep: {
    // If an IMPLICIT_DEF value is pruned, it doesn't serve a purpose any
    // longer. The IMPLICIT_DEF instructions are only inserted by
    // PHIElimination to guarantee that all PHI predecessors have a value.
    if (!Vals[i].ErasableImplicitDef || !Vals[i].Pruned)
      break;
    // Remove value number i from LR.
    // For intervals with subranges, removing a segment from the main range
    // may require extending the previous segment: for each definition of
    // a subregister, there will be a corresponding def in the main range.
    // That def may fall in the middle of a segment from another subrange.
    // In such cases, removing this def from the main range must be
    // complemented by extending the main range to account for the liveness
    // of the other subrange.
    // The new end point of the main range segment to be extended.
    SlotIndex NewEnd;
    if (LI != nullptr) {
      LiveRange::iterator I = LR.FindSegmentContaining(Def);
      assert(I != LR.end())(static_cast<void> (0));
      // Do not extend beyond the end of the segment being removed.
      // The segment may have been pruned in preparation for joining
      // live ranges.
      NewEnd = I->end;
    }

    LR.removeValNo(VNI);
    // Note that this VNInfo is reused and still referenced in NewVNInfo,
    // make it appear like an unused value number.
    VNI->markUnused();

    if (LI != nullptr && LI->hasSubRanges()) {
      assert(static_cast<LiveRange*>(LI) == &LR)(static_cast<void> (0));
      // Determine the end point based on the subrange information:
      // minimum of (earliest def of next segment,
      //             latest end point of containing segment)
      SlotIndex ED, LE;
      for (LiveInterval::SubRange &SR : LI->subranges()) {
        LiveRange::iterator I = SR.find(Def);
        if (I == SR.end())
          continue;
        if (I->start > Def)
          ED = ED.isValid() ? std::min(ED, I->start) : I->start;
        else
          LE = LE.isValid() ? std::max(LE, I->end) : I->end;
      }
      if (LE.isValid())
        NewEnd = std::min(NewEnd, LE);
      if (ED.isValid())
        NewEnd = std::min(NewEnd, ED);

      // We only want to do the extension if there was a subrange that
      // was live across Def.
      if (LE.isValid()) {
        LiveRange::iterator S = LR.find(Def);
        if (S != LR.begin())
          std::prev(S)->end = NewEnd;
      }
    }
    LLVM_DEBUG({do { } while (false)
      dbgs() << "\t\tremoved " << i << '@' << Def << ": " << LR << '\n';do { } while (false)
      if (LI != nullptr)do { } while (false)
        dbgs() << "\t\t  LHS = " << *LI << '\n';do { } while (false)
    })do { } while (false);
    LLVM_FALLTHROUGH[[gnu::fallthrough]];
  }

  case CR_Erase: {
    MachineInstr *MI = Indexes->getInstructionFromIndex(Def);
    assert(MI && "No instruction to erase")(static_cast<void> (0));
    if (MI->isCopy()) {
      Register Reg = MI->getOperand(1).getReg();
      if (Register::isVirtualRegister(Reg) && Reg != CP.getSrcReg() &&
          Reg != CP.getDstReg())
        ShrinkRegs.push_back(Reg);
    }
    ErasedInstrs.insert(MI);
    LLVM_DEBUG(dbgs() << "\t\terased:\t" << Def << '\t' << *MI)do { } while (false);
    LIS->RemoveMachineInstrFromMaps(*MI);
    MI->eraseFromParent();
    break;
  }
  default:
    break;
  }
}
3428}

3430void RegisterCoalescer::joinSubRegRanges(LiveRange &LRange, LiveRange &RRange,
                                       LaneBitmask LaneMask,
                                       const CoalescerPair &CP) {
SmallVector<VNInfo*, 16> NewVNInfo;
JoinVals RHSVals(RRange, CP.getSrcReg(), CP.getSrcIdx(), LaneMask,
                 NewVNInfo, CP, LIS, TRI, true, true);
JoinVals LHSVals(LRange, CP.getDstReg(), CP.getDstIdx(), LaneMask,
                 NewVNInfo, CP, LIS, TRI, true, true);

// Compute NewVNInfo and resolve conflicts (see also joinVirtRegs())
// We should be able to resolve all conflicts here as we could successfully do
// it on the mainrange already. There is however a problem when multiple
// ranges get mapped to the "overflow" lane mask bit which creates unexpected
// interferences.
if (!LHSVals.mapValues(RHSVals) || !RHSVals.mapValues(LHSVals)) {
  // We already determined that it is legal to merge the intervals, so this
  // should never fail.
  llvm_unreachable("*** Couldn't join subrange!\n")__builtin_unreachable();
}
if (!LHSVals.resolveConflicts(RHSVals) ||
    !RHSVals.resolveConflicts(LHSVals)) {
  // We already determined that it is legal to merge the intervals, so this
  // should never fail.
  llvm_unreachable("*** Couldn't join subrange!\n")__builtin_unreachable();
}

// The merging algorithm in LiveInterval::join() can't handle conflicting
// value mappings, so we need to remove any live ranges that overlap a
// CR_Replace resolution. Collect a set of end points that can be used to
// restore the live range after joining.
SmallVector<SlotIndex, 8> EndPoints;
LHSVals.pruneValues(RHSVals, EndPoints, false);
RHSVals.pruneValues(LHSVals, EndPoints, false);

LHSVals.removeImplicitDefs();
RHSVals.removeImplicitDefs();

LRange.verify();
RRange.verify();

// Join RRange into LHS.
LRange.join(RRange, LHSVals.getAssignments(), RHSVals.getAssignments(),
            NewVNInfo);

LLVM_DEBUG(dbgs() << "\t\tjoined lanes: " << PrintLaneMask(LaneMask)do { } while (false)
                  << ' ' << LRange << "\n")do { } while (false);
if (EndPoints.empty())
  return;

// Recompute the parts of the live range we had to remove because of
// CR_Replace conflicts.
LLVM_DEBUG({do { } while (false)
  dbgs() << "\t\trestoring liveness to " << EndPoints.size() << " points: ";do { } while (false)
  for (unsigned i = 0, n = EndPoints.size(); i != n; ++i) {do { } while (false)
    dbgs() << EndPoints[i];do { } while (false)
    if (i != n-1)do { } while (false)
      dbgs() << ',';do { } while (false)
  }do { } while (false)
  dbgs() << ":  " << LRange << '\n';do { } while (false)
})do { } while (false);
LIS->extendToIndices(LRange, EndPoints);
3491}

3493void RegisterCoalescer::mergeSubRangeInto(LiveInterval &LI,
                                        const LiveRange &ToMerge,
                                        LaneBitmask LaneMask,
                                        CoalescerPair &CP,
                                        unsigned ComposeSubRegIdx) {
BumpPtrAllocator &Allocator = LIS->getVNInfoAllocator();
LI.refineSubRanges(
    Allocator, LaneMask,
    [this, &Allocator, &ToMerge, &CP](LiveInterval::SubRange &SR) {
      if (SR.empty()) {
        SR.assign(ToMerge, Allocator);
      } else {
        // joinSubRegRange() destroys the merged range, so we need a copy.
        LiveRange RangeCopy(ToMerge, Allocator);
        joinSubRegRanges(SR, RangeCopy, SR.LaneMask, CP);
      }
    },
    *LIS->getSlotIndexes(), *TRI, ComposeSubRegIdx);
3511}

3513bool RegisterCoalescer::isHighCostLiveInterval(LiveInterval &LI) {
if (LI.valnos.size() < LargeIntervalSizeThreshold)
  return false;
auto &Counter = LargeLIVisitCounter[LI.reg()];
if (Counter < LargeIntervalFreqThreshold) {
  Counter++;
  return false;
}
return true;
3522}

3524bool RegisterCoalescer::joinVirtRegs(CoalescerPair &CP) {
SmallVector<VNInfo*, 16> NewVNInfo;
LiveInterval &RHS = LIS->getInterval(CP.getSrcReg());
LiveInterval &LHS = LIS->getInterval(CP.getDstReg());
bool TrackSubRegLiveness = MRI->shouldTrackSubRegLiveness(*CP.getNewRC());
JoinVals RHSVals(RHS, CP.getSrcReg(), CP.getSrcIdx(), LaneBitmask::getNone(),
                 NewVNInfo, CP, LIS, TRI, false, TrackSubRegLiveness);
JoinVals LHSVals(LHS, CP.getDstReg(), CP.getDstIdx(), LaneBitmask::getNone(),
                 NewVNInfo, CP, LIS, TRI, false, TrackSubRegLiveness);

LLVM_DEBUG(dbgs() << "\t\tRHS = " << RHS << "\n\t\tLHS = " << LHS << '\n')do { } while (false);

if (isHighCostLiveInterval(LHS) || isHighCostLiveInterval(RHS))
  return false;

// First compute NewVNInfo and the simple value mappings.
// Detect impossible conflicts early.
if (!LHSVals.mapValues(RHSVals) || !RHSVals.mapValues(LHSVals))
  return false;

// Some conflicts can only be resolved after all values have been mapped.
if (!LHSVals.resolveConflicts(RHSVals) || !RHSVals.resolveConflicts(LHSVals))
  return false;

// All clear, the live ranges can be merged.
if (RHS.hasSubRanges() || LHS.hasSubRanges()) {
  BumpPtrAllocator &Allocator = LIS->getVNInfoAllocator();

  // Transform lanemasks from the LHS to masks in the coalesced register and
  // create initial subranges if necessary.
  unsigned DstIdx = CP.getDstIdx();
  if (!LHS.hasSubRanges()) {
    LaneBitmask Mask = DstIdx == 0 ? CP.getNewRC()->getLaneMask()
                                   : TRI->getSubRegIndexLaneMask(DstIdx);
    // LHS must support subregs or we wouldn't be in this codepath.
    assert(Mask.any())(static_cast<void> (0));
    LHS.createSubRangeFrom(Allocator, Mask, LHS);
  } else if (DstIdx != 0) {
    // Transform LHS lanemasks to new register class if necessary.
    for (LiveInterval::SubRange &R : LHS.subranges()) {
      LaneBitmask Mask = TRI->composeSubRegIndexLaneMask(DstIdx, R.LaneMask);
      R.LaneMask = Mask;
    }
  }
  LLVM_DEBUG(dbgs() << "\t\tLHST = " << printReg(CP.getDstReg()) << ' ' << LHSdo { } while (false)
                    << '\n')do { } while (false);

  // Determine lanemasks of RHS in the coalesced register and merge subranges.
  unsigned SrcIdx = CP.getSrcIdx();
  if (!RHS.hasSubRanges()) {
    LaneBitmask Mask = SrcIdx == 0 ? CP.getNewRC()->getLaneMask()
                                   : TRI->getSubRegIndexLaneMask(SrcIdx);
    mergeSubRangeInto(LHS, RHS, Mask, CP, DstIdx);
  } else {
    // Pair up subranges and merge.
    for (LiveInterval::SubRange &R : RHS.subranges()) {
      LaneBitmask Mask = TRI->composeSubRegIndexLaneMask(SrcIdx, R.LaneMask);
      mergeSubRangeInto(LHS, R, Mask, CP, DstIdx);
    }
  }
  LLVM_DEBUG(dbgs() << "\tJoined SubRanges " << LHS << "\n")do { } while (false);

  // Pruning implicit defs from subranges may result in the main range
  // having stale segments.
  LHSVals.pruneMainSegments(LHS, ShrinkMainRange);

  LHSVals.pruneSubRegValues(LHS, ShrinkMask);
  RHSVals.pruneSubRegValues(LHS, ShrinkMask);
}

// The merging algorithm in LiveInterval::join() can't handle conflicting
// value mappings, so we need to remove any live ranges that overlap a
// CR_Replace resolution. Collect a set of end points that can be used to
// restore the live range after joining.
SmallVector<SlotIndex, 8> EndPoints;
LHSVals.pruneValues(RHSVals, EndPoints, true);
RHSVals.pruneValues(LHSVals, EndPoints, true);

// Erase COPY and IMPLICIT_DEF instructions. This may cause some external
// registers to require trimming.
SmallVector<Register, 8> ShrinkRegs;
LHSVals.eraseInstrs(ErasedInstrs, ShrinkRegs, &LHS);
RHSVals.eraseInstrs(ErasedInstrs, ShrinkRegs);
while (!ShrinkRegs.empty())
  shrinkToUses(&LIS->getInterval(ShrinkRegs.pop_back_val()));

// Scan and mark undef any DBG_VALUEs that would refer to a different value.
checkMergingChangesDbgValues(CP, LHS, LHSVals, RHS, RHSVals);

// If the RHS covers any PHI locations that were tracked for debug-info, we
// must update tracking information to reflect the join.
auto RegIt = RegToPHIIdx.find(CP.getSrcReg());
if (RegIt != RegToPHIIdx.end()) {
  // Iterate over all the debug instruction numbers assigned this register.
  for (unsigned InstID : RegIt->second) {
    auto PHIIt = PHIValToPos.find(InstID);
    assert(PHIIt != PHIValToPos.end())(static_cast<void> (0));
    const SlotIndex &SI = PHIIt->second.SI;

    // Does the RHS cover the position of this PHI?
    auto LII = RHS.find(SI);
    if (LII == RHS.end() || LII->start > SI)
      continue;

    // Accept two kinds of subregister movement:
    //  * When we merge from one register class into a larger register:
    //        %1:gr16 = some-inst
    //                ->
    //        %2:gr32.sub_16bit = some-inst
    //  * When the PHI is already in a subregister, and the larger class
    //    is coalesced:
    //        %2:gr32.sub_16bit = some-inst
    //        %3:gr32 = COPY %2
    //                ->
    //        %3:gr32.sub_16bit = some-inst
    // Test for subregister move:
    if (CP.getSrcIdx() != 0 || CP.getDstIdx() != 0)
      // If we're moving between different subregisters, ignore this join.
      // The PHI will not get a location, dropping variable locations.
      if (PHIIt->second.SubReg && PHIIt->second.SubReg != CP.getSrcIdx())
        continue;

    // Update our tracking of where the PHI is.
    PHIIt->second.Reg = CP.getDstReg();

    // If we merge into a sub-register of a larger class (test above),
    // update SubReg.
    if (CP.getSrcIdx() != 0)
      PHIIt->second.SubReg = CP.getSrcIdx();
  }

  // Rebuild the register index in RegToPHIIdx to account for PHIs tracking
  // different VRegs now. Copy old collection of debug instruction numbers and
  // erase the old one:
  auto InstrNums = RegIt->second;
  RegToPHIIdx.erase(RegIt);

  // There might already be PHIs being tracked in the destination VReg. Insert
  // into an existing tracking collection, or insert a new one.
  RegIt = RegToPHIIdx.find(CP.getDstReg());
  if (RegIt != RegToPHIIdx.end())
    RegIt->second.insert(RegIt->second.end(), InstrNums.begin(),
                         InstrNums.end());
  else
    RegToPHIIdx.insert({CP.getDstReg(), InstrNums});
}

// Join RHS into LHS.
LHS.join(RHS, LHSVals.getAssignments(), RHSVals.getAssignments(), NewVNInfo);

// Kill flags are going to be wrong if the live ranges were overlapping.
// Eventually, we should simply clear all kill flags when computing live
// ranges. They are reinserted after register allocation.
MRI->clearKillFlags(LHS.reg());
MRI->clearKillFlags(RHS.reg());

if (!EndPoints.empty()) {
  // Recompute the parts of the live range we had to remove because of
  // CR_Replace conflicts.
  LLVM_DEBUG({do { } while (false)
    dbgs() << "\t\trestoring liveness to " << EndPoints.size() << " points: ";do { } while (false)
    for (unsigned i = 0, n = EndPoints.size(); i != n; ++i) {do { } while (false)
      dbgs() << EndPoints[i];do { } while (false)
      if (i != n-1)do { } while (false)
        dbgs() << ',';do { } while (false)
    }do { } while (false)
    dbgs() << ":  " << LHS << '\n';do { } while (false)
  })do { } while (false);
  LIS->extendToIndices((LiveRange&)LHS, EndPoints);
}

return true;
3696}

3698bool RegisterCoalescer::joinIntervals(CoalescerPair &CP) {
return CP.isPhys() ? joinReservedPhysReg(CP) : joinVirtRegs(CP);
3700}

3702void RegisterCoalescer::buildVRegToDbgValueMap(MachineFunction &MF)
3703{
const SlotIndexes &Slots = *LIS->getSlotIndexes();
SmallVector<MachineInstr *, 8> ToInsert;

// After collecting a block of DBG_VALUEs into ToInsert, enter them into the
// vreg => DbgValueLoc map.
auto CloseNewDVRange = [this, &ToInsert](SlotIndex Slot) {
  for (auto *X : ToInsert) {
    for (auto Op : X->debug_operands()) {
      if (Op.isReg() && Op.getReg().isVirtual())
        DbgVRegToValues[Op.getReg()].push_back({Slot, X});
    }
  }

  ToInsert.clear();
};

// Iterate over all instructions, collecting them into the ToInsert vector.
// Once a non-debug instruction is found, record the slot index of the
// collected DBG_VALUEs.
for (auto &MBB : MF) {
  SlotIndex CurrentSlot = Slots.getMBBStartIdx(&MBB);

  for (auto &MI : MBB) {
    if (MI.isDebugValue()) {
      if (any_of(MI.debug_operands(), [](const MachineOperand &MO) {
            return MO.isReg() && MO.getReg().isVirtual();
          }))
        ToInsert.push_back(&MI);
    } else if (!MI.isDebugOrPseudoInstr()) {
      CurrentSlot = Slots.getInstructionIndex(MI);
      CloseNewDVRange(CurrentSlot);
    }
  }

  // Close range of DBG_VALUEs at the end of blocks.
  CloseNewDVRange(Slots.getMBBEndIdx(&MBB));
}

// Sort all DBG_VALUEs we've seen by slot number.
for (auto &Pair : DbgVRegToValues)
  llvm::sort(Pair.second);
3745}

3747void RegisterCoalescer::checkMergingChangesDbgValues(CoalescerPair &CP,
                                                   LiveRange &LHS,
                                                   JoinVals &LHSVals,
                                                   LiveRange &RHS,
                                                   JoinVals &RHSVals) {
auto ScanForDstReg = [&](Register Reg) {
  checkMergingChangesDbgValuesImpl(Reg, RHS, LHS, LHSVals);
};

auto ScanForSrcReg = [&](Register Reg) {
  checkMergingChangesDbgValuesImpl(Reg, LHS, RHS, RHSVals);
};

// Scan for potentially unsound DBG_VALUEs: examine first the register number
// Reg, and then any other vregs that may have been merged into  it.
auto PerformScan = [this](Register Reg, std::function<void(Register)> Func) {
  Func(Reg);
  if (DbgMergedVRegNums.count(Reg))
    for (Register X : DbgMergedVRegNums[Reg])
      Func(X);
};

// Scan for unsound updates of both the source and destination register.
PerformScan(CP.getSrcReg(), ScanForSrcReg);
PerformScan(CP.getDstReg(), ScanForDstReg);
3772}

3774void RegisterCoalescer::checkMergingChangesDbgValuesImpl(Register Reg,
                                                       LiveRange &OtherLR,
                                                       LiveRange &RegLR,
                                                       JoinVals &RegVals) {
// Are there any DBG_VALUEs to examine?
auto VRegMapIt = DbgVRegToValues.find(Reg);
if (VRegMapIt == DbgVRegToValues.end())
  return;

auto &DbgValueSet = VRegMapIt->second;
auto DbgValueSetIt = DbgValueSet.begin();
auto SegmentIt = OtherLR.begin();

bool LastUndefResult = false;
SlotIndex LastUndefIdx;

// If the "Other" register is live at a slot Idx, test whether Reg can
// safely be merged with it, or should be marked undef.
auto ShouldUndef = [&RegVals, &RegLR, &LastUndefResult,
                    &LastUndefIdx](SlotIndex Idx) -> bool {
  // Our worst-case performance typically happens with asan, causing very
  // many DBG_VALUEs of the same location. Cache a copy of the most recent
  // result for this edge-case.
  if (LastUndefIdx == Idx)
    return LastUndefResult;

  // If the other range was live, and Reg's was not, the register coalescer
  // will not have tried to resolve any conflicts. We don't know whether
  // the DBG_VALUE will refer to the same value number, so it must be made
  // undef.
  auto OtherIt = RegLR.find(Idx);
  if (OtherIt == RegLR.end())
    return true;

  // Both the registers were live: examine the conflict resolution record for
  // the value number Reg refers to. CR_Keep meant that this value number
  // "won" and the merged register definitely refers to that value. CR_Erase
  // means the value number was a redundant copy of the other value, which
  // was coalesced and Reg deleted. It's safe to refer to the other register
  // (which will be the source of the copy).
  auto Resolution = RegVals.getResolution(OtherIt->valno->id);
  LastUndefResult = Resolution != JoinVals::CR_Keep &&
                    Resolution != JoinVals::CR_Erase;
  LastUndefIdx = Idx;
  return LastUndefResult;
};

// Iterate over both the live-range of the "Other" register, and the set of
// DBG_VALUEs for Reg at the same time. Advance whichever one has the lowest
// slot index. This relies on the DbgValueSet being ordered.
while (DbgValueSetIt != DbgValueSet.end() && SegmentIt != OtherLR.end()) {
  if (DbgValueSetIt->first < SegmentIt->end) {
    // "Other" is live and there is a DBG_VALUE of Reg: test if we should
    // set it undef.
    if (DbgValueSetIt->first >= SegmentIt->start) {
      bool HasReg = DbgValueSetIt->second->hasDebugOperandForReg(Reg);
      bool ShouldUndefReg = ShouldUndef(DbgValueSetIt->first);
      if (HasReg && ShouldUndefReg) {
        // Mark undef, erase record of this DBG_VALUE to avoid revisiting.
        DbgValueSetIt->second->setDebugValueUndef();
        continue;
      }
    }
    ++DbgValueSetIt;
  } else {
    ++SegmentIt;
  }
}
3842}

3844namespace {

3846/// Information concerning MBB coalescing priority.
3847struct MBBPriorityInfo {
MachineBasicBlock *MBB;
unsigned Depth;
bool IsSplit;

MBBPriorityInfo(MachineBasicBlock *mbb, unsigned depth, bool issplit)
  : MBB(mbb), Depth(depth), IsSplit(issplit) {}
3854};

3856} // end anonymous namespace

3858/// C-style comparator that sorts first based on the loop depth of the basic
3859/// block (the unsigned), and then on the MBB number.
3860///
3861/// EnableGlobalCopies assumes that the primary sort key is loop depth.
3862static int compareMBBPriority(const MBBPriorityInfo *LHS,
                            const MBBPriorityInfo *RHS) {
// Deeper loops first
if (LHS->Depth != RHS->Depth)
  return LHS->Depth > RHS->Depth ? -1 : 1;

// Try to unsplit critical edges next.
if (LHS->IsSplit != RHS->IsSplit)
  return LHS->IsSplit ? -1 : 1;

// Prefer blocks that are more connected in the CFG. This takes care of
// the most difficult copies first while intervals are short.
unsigned cl = LHS->MBB->pred_size() + LHS->MBB->succ_size();
unsigned cr = RHS->MBB->pred_size() + RHS->MBB->succ_size();
if (cl != cr)
  return cl > cr ? -1 : 1;

// As a last resort, sort by block number.
return LHS->MBB->getNumber() < RHS->MBB->getNumber() ? -1 : 1;
3881}

3883/// \returns true if the given copy uses or defines a local live range.
3884static bool isLocalCopy(MachineInstr *Copy, const LiveIntervals *LIS) {
if (!Copy->isCopy())
  return false;

if (Copy->getOperand(1).isUndef())
  return false;

Register SrcReg = Copy->getOperand(1).getReg();
Register DstReg = Copy->getOperand(0).getReg();
if (Register::isPhysicalRegister(SrcReg) ||
    Register::isPhysicalRegister(DstReg))
  return false;

return LIS->intervalIsInOneMBB(LIS->getInterval(SrcReg))
  || LIS->intervalIsInOneMBB(LIS->getInterval(DstReg));
3899}

3901void RegisterCoalescer::lateLiveIntervalUpdate() {
for (Register reg : ToBeUpdated) {
  if (!LIS->hasInterval(reg))
    continue;
  LiveInterval &LI = LIS->getInterval(reg);
  shrinkToUses(&LI, &DeadDefs);
  if (!DeadDefs.empty())
    eliminateDeadDefs();
}
ToBeUpdated.clear();
3911}

3913bool RegisterCoalescer::
3914copyCoalesceWorkList(MutableArrayRef<MachineInstr*> CurrList) {
bool Progress = false;
for (unsigned i = 0, e = CurrList.size(); i != e; ++i) {
  if (!CurrList[i])
    continue;
  // Skip instruction pointers that have already been erased, for example by
  // dead code elimination.
  if (ErasedInstrs.count(CurrList[i])) {
    CurrList[i] = nullptr;
    continue;
  }
  bool Again = false;
  bool Success = joinCopy(CurrList[i], Again);
  Progress |= Success;
  if (Success || !Again)
    CurrList[i] = nullptr;
}
return Progress;
3932}

3934/// Check if DstReg is a terminal node.
3935/// I.e., it does not have any affinity other than \p Copy.
3936static bool isTerminalReg(Register DstReg, const MachineInstr &Copy,
                        const MachineRegisterInfo *MRI) {
assert(Copy.isCopyLike())(static_cast<void> (0));
// Check if the destination of this copy as any other affinity.
for (const MachineInstr &MI : MRI->reg_nodbg_instructions(DstReg))
  if (&MI != &Copy && MI.isCopyLike())
    return false;
return true;
3944}

3946bool RegisterCoalescer::applyTerminalRule(const MachineInstr &Copy) const {
assert(Copy.isCopyLike())(static_cast<void> (0));
if (!UseTerminalRule)
  return false;
Register SrcReg, DstReg;
unsigned SrcSubReg = 0, DstSubReg = 0;
if (!isMoveInstr(*TRI, &Copy, SrcReg, DstReg, SrcSubReg, DstSubReg))
  return false;
// Check if the destination of this copy has any other affinity.
if (DstReg.isPhysical() ||
    // If SrcReg is a physical register, the copy won't be coalesced.
    // Ignoring it may have other side effect (like missing
    // rematerialization). So keep it.
    SrcReg.isPhysical() || !isTerminalReg(DstReg, Copy, MRI))
  return false;

// DstReg is a terminal node. Check if it interferes with any other
// copy involving SrcReg.
const MachineBasicBlock *OrigBB = Copy.getParent();
const LiveInterval &DstLI = LIS->getInterval(DstReg);
for (const MachineInstr &MI : MRI->reg_nodbg_instructions(SrcReg)) {
  // Technically we should check if the weight of the new copy is
  // interesting compared to the other one and update the weight
  // of the copies accordingly. However, this would only work if
  // we would gather all the copies first then coalesce, whereas
  // right now we interleave both actions.
  // For now, just consider the copies that are in the same block.
  if (&MI == &Copy || !MI.isCopyLike() || MI.getParent() != OrigBB)
    continue;
  Register OtherSrcReg, OtherReg;
  unsigned OtherSrcSubReg = 0, OtherSubReg = 0;
  if (!isMoveInstr(*TRI, &Copy, OtherSrcReg, OtherReg, OtherSrcSubReg,
              OtherSubReg))
    return false;
  if (OtherReg == SrcReg)
    OtherReg = OtherSrcReg;
  // Check if OtherReg is a non-terminal.
  if (Register::isPhysicalRegister(OtherReg) ||
      isTerminalReg(OtherReg, MI, MRI))
    continue;
  // Check that OtherReg interfere with DstReg.
  if (LIS->getInterval(OtherReg).overlaps(DstLI)) {
    LLVM_DEBUG(dbgs() << "Apply terminal rule for: " << printReg(DstReg)do { } while (false)
                      << '\n')do { } while (false);
    return true;
  }
}
return false;
3994}

3996void
3997RegisterCoalescer::copyCoalesceInMBB(MachineBasicBlock *MBB) {
LLVM_DEBUG(dbgs() << MBB->getName() << ":\n")do { } while (false);

// Collect all copy-like instructions in MBB. Don't start coalescing anything
// yet, it might invalidate the iterator.
const unsigned PrevSize = WorkList.size();
if (JoinGlobalCopies) {
  SmallVector<MachineInstr*, 2> LocalTerminals;
  SmallVector<MachineInstr*, 2> GlobalTerminals;
  // Coalesce copies bottom-up to coalesce local defs before local uses. They
  // are not inherently easier to resolve, but slightly preferable until we
  // have local live range splitting. In particular this is required by
  // cmp+jmp macro fusion.
  for (MachineInstr &MI : *MBB) {
    if (!MI.isCopyLike())
      continue;
    bool ApplyTerminalRule = applyTerminalRule(MI);
    if (isLocalCopy(&MI, LIS)) {
      if (ApplyTerminalRule)
        LocalTerminals.push_back(&MI);
      else
        LocalWorkList.push_back(&MI);
    } else {
      if (ApplyTerminalRule)
        GlobalTerminals.push_back(&MI);
      else
        WorkList.push_back(&MI);
    }
  }
  // Append the copies evicted by the terminal rule at the end of the list.
  LocalWorkList.append(LocalTerminals.begin(), LocalTerminals.end());
  WorkList.append(GlobalTerminals.begin(), GlobalTerminals.end());
}
else {
  SmallVector<MachineInstr*, 2> Terminals;
  for (MachineInstr &MII : *MBB)
    if (MII.isCopyLike()) {
      if (applyTerminalRule(MII))
        Terminals.push_back(&MII);
      else
        WorkList.push_back(&MII);
    }
  // Append the copies evicted by the terminal rule at the end of the list.
  WorkList.append(Terminals.begin(), Terminals.end());
}
// Try coalescing the collected copies immediately, and remove the nulls.
// This prevents the WorkList from getting too large since most copies are
// joinable on the first attempt.
MutableArrayRef<MachineInstr*>
  CurrList(WorkList.begin() + PrevSize, WorkList.end());
if (copyCoalesceWorkList(CurrList))
  WorkList.erase(std::remove(WorkList.begin() + PrevSize, WorkList.end(),
                             nullptr), WorkList.end());
4050}

4052void RegisterCoalescer::coalesceLocals() {
copyCoalesceWorkList(LocalWorkList);
for (unsigned j = 0, je = LocalWorkList.size(); j != je; ++j) {
  if (LocalWorkList[j])
    WorkList.push_back(LocalWorkList[j]);
}
LocalWorkList.clear();
4059}

4061void RegisterCoalescer::joinAllIntervals() {
LLVM_DEBUG(dbgs() << "********** JOINING INTERVALS ***********\n")do { } while (false);
assert(WorkList.empty() && LocalWorkList.empty() && "Old data still around.")(static_cast<void> (0));

std::vector<MBBPriorityInfo> MBBs;
MBBs.reserve(MF->size());
for (MachineBasicBlock &MBB : *MF) {
  MBBs.push_back(MBBPriorityInfo(&MBB, Loops->getLoopDepth(&MBB),
                                 JoinSplitEdges && isSplitEdge(&MBB)));
}
array_pod_sort(MBBs.begin(), MBBs.end(), compareMBBPriority);

// Coalesce intervals in MBB priority order.
unsigned CurrDepth = std::numeric_limits<unsigned>::max();
for (unsigned i = 0, e = MBBs.size(); i != e; ++i) {
  // Try coalescing the collected local copies for deeper loops.
  if (JoinGlobalCopies && MBBs[i].Depth < CurrDepth) {
    coalesceLocals();
    CurrDepth = MBBs[i].Depth;
  }
  copyCoalesceInMBB(MBBs[i].MBB);
}
lateLiveIntervalUpdate();
coalesceLocals();

// Joining intervals can allow other intervals to be joined.  Iteratively join
// until we make no progress.
while (copyCoalesceWorkList(WorkList))
  /* empty */ ;
lateLiveIntervalUpdate();
4091}

4093void RegisterCoalescer::releaseMemory() {
ErasedInstrs.clear();
WorkList.clear();
DeadDefs.clear();
InflateRegs.clear();
LargeLIVisitCounter.clear();
4099}

4101bool RegisterCoalescer::runOnMachineFunction(MachineFunction &fn) {
LLVM_DEBUG(dbgs() << "********** SIMPLE REGISTER COALESCING **********\n"do { } while (false)
                  << "********** Function: " << fn.getName() << '\n')do { } while (false);

// Variables changed between a setjmp and a longjump can have undefined value
// after the longjmp. This behaviour can be observed if such a variable is
// spilled, so longjmp won't restore the value in the spill slot.
// RegisterCoalescer should not run in functions with a setjmp to avoid
// merging such undefined variables with predictable ones.
//
// TODO: Could specifically disable coalescing registers live across setjmp
// calls
if (fn.exposesReturnsTwice()) {
  LLVM_DEBUG(do { } while (false)
      dbgs() << "* Skipped as it exposes funcions that returns twice.\n")do { } while (false);
  return false;
}

MF = &fn;
MRI = &fn.getRegInfo();
const TargetSubtargetInfo &STI = fn.getSubtarget();
TRI = STI.getRegisterInfo();
TII = STI.getInstrInfo();
LIS = &getAnalysis<LiveIntervals>();
AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();
Loops = &getAnalysis<MachineLoopInfo>();
if (EnableGlobalCopies == cl::BOU_UNSET)
  JoinGlobalCopies = STI.enableJoinGlobalCopies();
else
  JoinGlobalCopies = (EnableGlobalCopies == cl::BOU_TRUE);

// If there are PHIs tracked by debug-info, they will need updating during
// coalescing. Build an index of those PHIs to ease updating.
SlotIndexes *Slots = LIS->getSlotIndexes();
for (const auto &DebugPHI : MF->DebugPHIPositions) {
  MachineBasicBlock *MBB = DebugPHI.second.MBB;
  Register Reg = DebugPHI.second.Reg;
  unsigned SubReg = DebugPHI.second.SubReg;
  SlotIndex SI = Slots->getMBBStartIdx(MBB);
  PHIValPos P = {SI, Reg, SubReg};
  PHIValToPos.insert(std::make_pair(DebugPHI.first, P));
  RegToPHIIdx[Reg].push_back(DebugPHI.first);
}

// The MachineScheduler does not currently require JoinSplitEdges. This will
// either be enabled unconditionally or replaced by a more general live range
// splitting optimization.
JoinSplitEdges = EnableJoinSplits;

if (VerifyCoalescing)
  MF->verify(this, "Before register coalescing");

DbgVRegToValues.clear();
DbgMergedVRegNums.clear();
buildVRegToDbgValueMap(fn);

RegClassInfo.runOnMachineFunction(fn);

// Join (coalesce) intervals if requested.
if (EnableJoining)
  joinAllIntervals();

// After deleting a lot of copies, register classes may be less constrained.
// Removing sub-register operands may allow GR32_ABCD -> GR32 and DPR_VFP2 ->
// DPR inflation.
array_pod_sort(InflateRegs.begin(), InflateRegs.end());
InflateRegs.erase(std::unique(InflateRegs.begin(), InflateRegs.end()),
                  InflateRegs.end());
LLVM_DEBUG(dbgs() << "Trying to inflate " << InflateRegs.size()do { } while (false)
                  << " regs.\n")do { } while (false);
for (unsigned i = 0, e = InflateRegs.size(); i != e; ++i) {
  Register Reg = InflateRegs[i];
  if (MRI->reg_nodbg_empty(Reg))
    continue;
  if (MRI->recomputeRegClass(Reg)) {
    LLVM_DEBUG(dbgs() << printReg(Reg) << " inflated to "do { } while (false)
                      << TRI->getRegClassName(MRI->getRegClass(Reg)) << '\n')do { } while (false);
    ++NumInflated;

    LiveInterval &LI = LIS->getInterval(Reg);
    if (LI.hasSubRanges()) {
      // If the inflated register class does not support subregisters anymore
      // remove the subranges.
      if (!MRI->shouldTrackSubRegLiveness(Reg)) {
        LI.clearSubRanges();
      } else {
4187#ifndef NDEBUG1
        LaneBitmask MaxMask = MRI->getMaxLaneMaskForVReg(Reg);
        // If subranges are still supported, then the same subregs
        // should still be supported.
        for (LiveInterval::SubRange &S : LI.subranges()) {
          assert((S.LaneMask & ~MaxMask).none())(static_cast<void> (0));
        }
4194#endif
      }
    }
  }
}

// After coalescing, update any PHIs that are being tracked by debug-info
// with their new VReg locations.
for (auto &p : MF->DebugPHIPositions) {
  auto it = PHIValToPos.find(p.first);
  assert(it != PHIValToPos.end())(static_cast<void> (0));
  p.second.Reg = it->second.Reg;
  p.second.SubReg = it->second.SubReg;
}

PHIValToPos.clear();
RegToPHIIdx.clear();

LLVM_DEBUG(dump())do { } while (false);
if (VerifyCoalescing)
  MF->verify(this, "After register coalescing");
return true;
4216}

4218void RegisterCoalescer::print(raw_ostream &O, const Module* m) const {
 LIS->print(O, m);
4220}

←

/build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/include/llvm/CodeGen/LiveInterval.h

→

1//===- llvm/CodeGen/LiveInterval.h - Interval representation ----*- 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 implements the LiveRange and LiveInterval classes.  Given some
10// numbering of each the machine instructions an interval [i, j) is said to be a
11// live range for register v if there is no instruction with number j' >= j
12// such that v is live at j' and there is no instruction with number i' < i such
13// that v is live at i'. In this implementation ranges can have holes,
14// i.e. a range might look like [1,20), [50,65), [1000,1001).  Each
15// individual segment is represented as an instance of LiveRange::Segment,
16// and the whole range is represented as an instance of LiveRange.
17//
18//===----------------------------------------------------------------------===//

20#ifndef LLVM_CODEGEN_LIVEINTERVAL_H
21#define LLVM_CODEGEN_LIVEINTERVAL_H

23#include "llvm/ADT/ArrayRef.h"
24#include "llvm/ADT/IntEqClasses.h"
25#include "llvm/ADT/STLExtras.h"
26#include "llvm/ADT/SmallVector.h"
27#include "llvm/ADT/iterator_range.h"
28#include "llvm/CodeGen/Register.h"
29#include "llvm/CodeGen/SlotIndexes.h"
30#include "llvm/MC/LaneBitmask.h"
31#include "llvm/Support/Allocator.h"
32#include "llvm/Support/MathExtras.h"
33#include <algorithm>
34#include <cassert>
35#include <cstddef>
36#include <functional>
37#include <memory>
38#include <set>
39#include <tuple>
40#include <utility>

42namespace llvm {

class CoalescerPair;
class LiveIntervals;
class MachineRegisterInfo;
class raw_ostream;

/// VNInfo - Value Number Information.
/// This class holds information about a machine level values, including
/// definition and use points.
///
class VNInfo {
public:
  using Allocator = BumpPtrAllocator;

  /// The ID number of this value.
  unsigned id;

  /// The index of the defining instruction.
  SlotIndex def;

  /// VNInfo constructor.
  VNInfo(unsigned i, SlotIndex d) : id(i), def(d) {}

  /// VNInfo constructor, copies values from orig, except for the value number.
  VNInfo(unsigned i, const VNInfo &orig) : id(i), def(orig.def) {}

  /// Copy from the parameter into this VNInfo.
  void copyFrom(VNInfo &src) {
    def = src.def;
  }

  /// Returns true if this value is defined by a PHI instruction (or was,
  /// PHI instructions may have been eliminated).
  /// PHI-defs begin at a block boundary, all other defs begin at register or
  /// EC slots.
  bool isPHIDef() const { return def.isBlock(); }
3
←
Calling 'SlotIndex::isBlock'→
6
←
Returning from 'SlotIndex::isBlock'→
7
←
Returning zero, which participates in a condition later→

  /// Returns true if this value is unused.
  bool isUnused() const { return !def.isValid(); }

  /// Mark this value as unused.
  void markUnused() { def = SlotIndex(); }
};

/// Result of a LiveRange query. This class hides the implementation details
/// of live ranges, and it should be used as the primary interface for
/// examining live ranges around instructions.
class LiveQueryResult {
  VNInfo *const EarlyVal;
  VNInfo *const LateVal;
  const SlotIndex EndPoint;
  const bool Kill;

public:
  LiveQueryResult(VNInfo *EarlyVal, VNInfo *LateVal, SlotIndex EndPoint,
                  bool Kill)
    : EarlyVal(EarlyVal), LateVal(LateVal), EndPoint(EndPoint), Kill(Kill)
  {}

  /// Return the value that is live-in to the instruction. This is the value
  /// that will be read by the instruction's use operands. Return NULL if no
  /// value is live-in.
  VNInfo *valueIn() const {
    return EarlyVal;
  }

  /// Return true if the live-in value is killed by this instruction. This
  /// means that either the live range ends at the instruction, or it changes
  /// value.
  bool isKill() const {
    return Kill;
  }

  /// Return true if this instruction has a dead def.
  bool isDeadDef() const {
    return EndPoint.isDead();
  }

  /// Return the value leaving the instruction, if any. This can be a
  /// live-through value, or a live def. A dead def returns NULL.
  VNInfo *valueOut() const {
    return isDeadDef() ? nullptr : LateVal;
  }

  /// Returns the value alive at the end of the instruction, if any. This can
  /// be a live-through value, a live def or a dead def.
  VNInfo *valueOutOrDead() const {
    return LateVal;
  }

  /// Return the value defined by this instruction, if any. This includes
  /// dead defs, it is the value created by the instruction's def operands.
  VNInfo *valueDefined() const {
    return EarlyVal == LateVal ? nullptr : LateVal;
  }

  /// Return the end point of the last live range segment to interact with
  /// the instruction, if any.
  ///
  /// The end point is an invalid SlotIndex only if the live range doesn't
  /// intersect the instruction at all.
  ///
  /// The end point may be at or past the end of the instruction's basic
  /// block. That means the value was live out of the block.
  SlotIndex endPoint() const {
    return EndPoint;
  }
};

/// This class represents the liveness of a register, stack slot, etc.
/// It manages an ordered list of Segment objects.
/// The Segments are organized in a static single assignment form: At places
/// where a new value is defined or different values reach a CFG join a new
/// segment with a new value number is used.
class LiveRange {
public:
  /// This represents a simple continuous liveness interval for a value.
  /// The start point is inclusive, the end point exclusive. These intervals
  /// are rendered as [start,end).
  struct Segment {
    SlotIndex start;  // Start point of the interval (inclusive)
    SlotIndex end;    // End point of the interval (exclusive)
    VNInfo *valno = nullptr; // identifier for the value contained in this
                             // segment.

    Segment() = default;

    Segment(SlotIndex S, SlotIndex E, VNInfo *V)
      : start(S), end(E), valno(V) {
      assert(S < E && "Cannot create empty or backwards segment")(static_cast<void> (0));
    }

    /// Return true if the index is covered by this segment.
    bool contains(SlotIndex I) const {
      return start <= I && I < end;
    }

    /// Return true if the given interval, [S, E), is covered by this segment.
    bool containsInterval(SlotIndex S, SlotIndex E) const {
      assert((S < E) && "Backwards interval?")(static_cast<void> (0));
      return (start <= S && S < end) && (start < E && E <= end);
    }

    bool operator<(const Segment &Other) const {
      return std::tie(start, end) < std::tie(Other.start, Other.end);
    }
    bool operator==(const Segment &Other) const {
      return start == Other.start && end == Other.end;
    }

    bool operator!=(const Segment &Other) const {
      return !(*this == Other);
    }

    void dump() const;
  };

  using Segments = SmallVector<Segment, 2>;
  using VNInfoList = SmallVector<VNInfo *, 2>;

  Segments segments;   // the liveness segments
  VNInfoList valnos;   // value#'s

  // The segment set is used temporarily to accelerate initial computation
  // of live ranges of physical registers in computeRegUnitRange.
  // After that the set is flushed to the segment vector and deleted.
  using SegmentSet = std::set<Segment>;
  std::unique_ptr<SegmentSet> segmentSet;

  using iterator = Segments::iterator;
  using const_iterator = Segments::const_iterator;

  iterator begin() { return segments.begin(); }
  iterator end()   { return segments.end(); }

  const_iterator begin() const { return segments.begin(); }
  const_iterator end() const  { return segments.end(); }

  using vni_iterator = VNInfoList::iterator;
  using const_vni_iterator = VNInfoList::const_iterator;

  vni_iterator vni_begin() { return valnos.begin(); }
  vni_iterator vni_end()   { return valnos.end(); }

  const_vni_iterator vni_begin() const { return valnos.begin(); }
  const_vni_iterator vni_end() const   { return valnos.end(); }

  /// Constructs a new LiveRange object.
  LiveRange(bool UseSegmentSet = false)
      : segmentSet(UseSegmentSet ? std::make_unique<SegmentSet>()
                                 : nullptr) {}

  /// Constructs a new LiveRange object by copying segments and valnos from
  /// another LiveRange.
  LiveRange(const LiveRange &Other, BumpPtrAllocator &Allocator) {
    assert(Other.segmentSet == nullptr &&(static_cast<void> (0))
           "Copying of LiveRanges with active SegmentSets is not supported")(static_cast<void> (0));
    assign(Other, Allocator);
  }

  /// Copies values numbers and live segments from \p Other into this range.
  void assign(const LiveRange &Other, BumpPtrAllocator &Allocator) {
    if (this == &Other)
      return;

    assert(Other.segmentSet == nullptr &&(static_cast<void> (0))
           "Copying of LiveRanges with active SegmentSets is not supported")(static_cast<void> (0));
    // Duplicate valnos.
    for (const VNInfo *VNI : Other.valnos)
      createValueCopy(VNI, Allocator);
    // Now we can copy segments and remap their valnos.
    for (const Segment &S : Other.segments)
      segments.push_back(Segment(S.start, S.end, valnos[S.valno->id]));
  }

  /// advanceTo - Advance the specified iterator to point to the Segment
  /// containing the specified position, or end() if the position is past the
  /// end of the range.  If no Segment contains this position, but the
  /// position is in a hole, this method returns an iterator pointing to the
  /// Segment immediately after the hole.
  iterator advanceTo(iterator I, SlotIndex Pos) {
    assert(I != end())(static_cast<void> (0));
    if (Pos >= endIndex())
      return end();
    while (I->end <= Pos) ++I;
    return I;
  }

  const_iterator advanceTo(const_iterator I, SlotIndex Pos) const {
    assert(I != end())(static_cast<void> (0));
    if (Pos >= endIndex())
      return end();
    while (I->end <= Pos) ++I;
    return I;
  }

  /// find - Return an iterator pointing to the first segment that ends after
  /// Pos, or end(). This is the same as advanceTo(begin(), Pos), but faster
  /// when searching large ranges.
  ///
  /// If Pos is contained in a Segment, that segment is returned.
  /// If Pos is in a hole, the following Segment is returned.
  /// If Pos is beyond endIndex, end() is returned.
  iterator find(SlotIndex Pos);

  const_iterator find(SlotIndex Pos) const {
    return const_cast<LiveRange*>(this)->find(Pos);
  }

  void clear() {
    valnos.clear();
    segments.clear();
  }

  size_t size() const {
    return segments.size();
  }

  bool hasAtLeastOneValue() const { return !valnos.empty(); }

  bool containsOneValue() const { return valnos.size() == 1; }

  unsigned getNumValNums() const { return (unsigned)valnos.size(); }

  /// getValNumInfo - Returns pointer to the specified val#.
  ///
  inline VNInfo *getValNumInfo(unsigned ValNo) {
    return valnos[ValNo];
  }
  inline const VNInfo *getValNumInfo(unsigned ValNo) const {
    return valnos[ValNo];
  }

  /// containsValue - Returns true if VNI belongs to this range.
  bool containsValue(const VNInfo *VNI) const {
    return VNI && VNI->id < getNumValNums() && VNI == getValNumInfo(VNI->id);
  }

  /// getNextValue - Create a new value number and return it.  MIIdx specifies
  /// the instruction that defines the value number.
  VNInfo *getNextValue(SlotIndex def, VNInfo::Allocator &VNInfoAllocator) {
    VNInfo *VNI =
      new (VNInfoAllocator) VNInfo((unsigned)valnos.size(), def);
    valnos.push_back(VNI);
    return VNI;
  }

  /// createDeadDef - Make sure the range has a value defined at Def.
  /// If one already exists, return it. Otherwise allocate a new value and
  /// add liveness for a dead def.
  VNInfo *createDeadDef(SlotIndex Def, VNInfo::Allocator &VNIAlloc);

  /// Create a def of value @p VNI. Return @p VNI. If there already exists
  /// a definition at VNI->def, the value defined there must be @p VNI.
  VNInfo *createDeadDef(VNInfo *VNI);

  /// Create a copy of the given value. The new value will be identical except
  /// for the Value number.
  VNInfo *createValueCopy(const VNInfo *orig,
                          VNInfo::Allocator &VNInfoAllocator) {
    VNInfo *VNI =
      new (VNInfoAllocator) VNInfo((unsigned)valnos.size(), *orig);
    valnos.push_back(VNI);
    return VNI;
  }

  /// RenumberValues - Renumber all values in order of appearance and remove
  /// unused values.
  void RenumberValues();

  /// MergeValueNumberInto - This method is called when two value numbers
  /// are found to be equivalent.  This eliminates V1, replacing all
  /// segments with the V1 value number with the V2 value number.  This can
  /// cause merging of V1/V2 values numbers and compaction of the value space.
  VNInfo* MergeValueNumberInto(VNInfo *V1, VNInfo *V2);

  /// Merge all of the live segments of a specific val# in RHS into this live
  /// range as the specified value number. The segments in RHS are allowed
  /// to overlap with segments in the current range, it will replace the
  /// value numbers of the overlaped live segments with the specified value
  /// number.
  void MergeSegmentsInAsValue(const LiveRange &RHS, VNInfo *LHSValNo);

  /// MergeValueInAsValue - Merge all of the segments of a specific val#
  /// in RHS into this live range as the specified value number.
  /// The segments in RHS are allowed to overlap with segments in the
  /// current range, but only if the overlapping segments have the
  /// specified value number.
  void MergeValueInAsValue(const LiveRange &RHS,
                           const VNInfo *RHSValNo, VNInfo *LHSValNo);

  bool empty() const { return segments.empty(); }

  /// beginIndex - Return the lowest numbered slot covered.
  SlotIndex beginIndex() const {
    assert(!empty() && "Call to beginIndex() on empty range.")(static_cast<void> (0));
    return segments.front().start;
  }

  /// endNumber - return the maximum point of the range of the whole,
  /// exclusive.
  SlotIndex endIndex() const {
    assert(!empty() && "Call to endIndex() on empty range.")(static_cast<void> (0));
    return segments.back().end;
  }

  bool expiredAt(SlotIndex index) const {
    return index >= endIndex();
  }

  bool liveAt(SlotIndex index) const {
    const_iterator r = find(index);
    return r != end() && r->start <= index;
  }

  /// Return the segment that contains the specified index, or null if there
  /// is none.
  const Segment *getSegmentContaining(SlotIndex Idx) const {
    const_iterator I = FindSegmentContaining(Idx);
    return I == end() ? nullptr : &*I;
  }

  /// Return the live segment that contains the specified index, or null if
  /// there is none.
  Segment *getSegmentContaining(SlotIndex Idx) {
    iterator I = FindSegmentContaining(Idx);
    return I == end() ? nullptr : &*I;
  }

  /// getVNInfoAt - Return the VNInfo that is live at Idx, or NULL.
  VNInfo *getVNInfoAt(SlotIndex Idx) const {
    const_iterator I = FindSegmentContaining(Idx);
    return I == end() ? nullptr : I->valno;
  }

  /// getVNInfoBefore - Return the VNInfo that is live up to but not
  /// necessarilly including Idx, or NULL. Use this to find the reaching def
  /// used by an instruction at this SlotIndex position.
  VNInfo *getVNInfoBefore(SlotIndex Idx) const {
    const_iterator I = FindSegmentContaining(Idx.getPrevSlot());
    return I == end() ? nullptr : I->valno;
  }

  /// Return an iterator to the segment that contains the specified index, or
  /// end() if there is none.
  iterator FindSegmentContaining(SlotIndex Idx) {
    iterator I = find(Idx);
    return I != end() && I->start <= Idx ? I : end();
  }

  const_iterator FindSegmentContaining(SlotIndex Idx) const {
    const_iterator I = find(Idx);
    return I != end() && I->start <= Idx ? I : end();
  }

  /// overlaps - Return true if the intersection of the two live ranges is
  /// not empty.
  bool overlaps(const LiveRange &other) const {
    if (other.empty())
      return false;
    return overlapsFrom(other, other.begin());
  }

  /// overlaps - Return true if the two ranges have overlapping segments
  /// that are not coalescable according to CP.
  ///
  /// Overlapping segments where one range is defined by a coalescable
  /// copy are allowed.
  bool overlaps(const LiveRange &Other, const CoalescerPair &CP,
                const SlotIndexes&) const;

  /// overlaps - Return true if the live range overlaps an interval specified
  /// by [Start, End).
  bool overlaps(SlotIndex Start, SlotIndex End) const;

  /// overlapsFrom - Return true if the intersection of the two live ranges
  /// is not empty.  The specified iterator is a hint that we can begin
  /// scanning the Other range starting at I.
  bool overlapsFrom(const LiveRange &Other, const_iterator StartPos) const;

  /// Returns true if all segments of the @p Other live range are completely
  /// covered by this live range.
  /// Adjacent live ranges do not affect the covering:the liverange
  /// [1,5](5,10] covers (3,7].
  bool covers(const LiveRange &Other) const;

  /// Add the specified Segment to this range, merging segments as
  /// appropriate.  This returns an iterator to the inserted segment (which
  /// may have grown since it was inserted).
  iterator addSegment(Segment S);

  /// Attempt to extend a value defined after @p StartIdx to include @p Use.
  /// Both @p StartIdx and @p Use should be in the same basic block. In case
  /// of subranges, an extension could be prevented by an explicit "undef"
  /// caused by a <def,read-undef> on a non-overlapping lane. The list of
  /// location of such "undefs" should be provided in @p Undefs.
  /// The return value is a pair: the first element is VNInfo of the value
  /// that was extended (possibly nullptr), the second is a boolean value
  /// indicating whether an "undef" was encountered.
  /// If this range is live before @p Use in the basic block that starts at
  /// @p StartIdx, and there is no intervening "undef", extend it to be live
  /// up to @p Use, and return the pair {value, false}. If there is no
  /// segment before @p Use and there is no "undef" between @p StartIdx and
  /// @p Use, return {nullptr, false}. If there is an "undef" before @p Use,
  /// return {nullptr, true}.
  std::pair<VNInfo*,bool> extendInBlock(ArrayRef<SlotIndex> Undefs,
      SlotIndex StartIdx, SlotIndex Kill);

  /// Simplified version of the above "extendInBlock", which assumes that
  /// no register lanes are undefined by <def,read-undef> operands.
  /// If this range is live before @p Use in the basic block that starts
  /// at @p StartIdx, extend it to be live up to @p Use, and return the
  /// value. If there is no segment before @p Use, return nullptr.
  VNInfo *extendInBlock(SlotIndex StartIdx, SlotIndex Kill);

  /// join - Join two live ranges (this, and other) together.  This applies
  /// mappings to the value numbers in the LHS/RHS ranges as specified.  If
  /// the ranges are not joinable, this aborts.
  void join(LiveRange &Other,
            const int *ValNoAssignments,
            const int *RHSValNoAssignments,
            SmallVectorImpl<VNInfo *> &NewVNInfo);

  /// True iff this segment is a single segment that lies between the
  /// specified boundaries, exclusively. Vregs live across a backedge are not
  /// considered local. The boundaries are expected to lie within an extended
  /// basic block, so vregs that are not live out should contain no holes.
  bool isLocal(SlotIndex Start, SlotIndex End) const {
    return beginIndex() > Start.getBaseIndex() &&
      endIndex() < End.getBoundaryIndex();
  }

  /// Remove the specified segment from this range.  Note that the segment
  /// must be a single Segment in its entirety.
  void removeSegment(SlotIndex Start, SlotIndex End,
                     bool RemoveDeadValNo = false);

  void removeSegment(Segment S, bool RemoveDeadValNo = false) {
    removeSegment(S.start, S.end, RemoveDeadValNo);
  }

  /// Remove segment pointed to by iterator @p I from this range.  This does
  /// not remove dead value numbers.
  iterator removeSegment(iterator I) {
    return segments.erase(I);
  }

  /// Query Liveness at Idx.
  /// The sub-instruction slot of Idx doesn't matter, only the instruction
  /// it refers to is considered.
  LiveQueryResult Query(SlotIndex Idx) const {
    // Find the segment that enters the instruction.
    const_iterator I = find(Idx.getBaseIndex());
    const_iterator E = end();
    if (I == E)
      return LiveQueryResult(nullptr, nullptr, SlotIndex(), false);

    // Is this an instruction live-in segment?
    // If Idx is the start index of a basic block, include live-in segments
    // that start at Idx.getBaseIndex().
    VNInfo *EarlyVal = nullptr;
    VNInfo *LateVal  = nullptr;
    SlotIndex EndPoint;
    bool Kill = false;
    if (I->start <= Idx.getBaseIndex()) {
      EarlyVal = I->valno;
      EndPoint = I->end;
      // Move to the potentially live-out segment.
      if (SlotIndex::isSameInstr(Idx, I->end)) {
        Kill = true;
        if (++I == E)
          return LiveQueryResult(EarlyVal, LateVal, EndPoint, Kill);
      }
      // Special case: A PHIDef value can have its def in the middle of a
      // segment if the value happens to be live out of the layout
      // predecessor.
      // Such a value is not live-in.
      if (EarlyVal->def == Idx.getBaseIndex())
        EarlyVal = nullptr;
    }
    // I now points to the segment that may be live-through, or defined by
    // this instr. Ignore segments starting after the current instr.
    if (!SlotIndex::isEarlierInstr(Idx, I->start)) {
      LateVal = I->valno;
      EndPoint = I->end;
    }
    return LiveQueryResult(EarlyVal, LateVal, EndPoint, Kill);
  }

  /// removeValNo - Remove all the segments defined by the specified value#.
  /// Also remove the value# from value# list.
  void removeValNo(VNInfo *ValNo);

  /// Returns true if the live range is zero length, i.e. no live segments
  /// span instructions. It doesn't pay to spill such a range.
  bool isZeroLength(SlotIndexes *Indexes) const {
    for (const Segment &S : segments)
      if (Indexes->getNextNonNullIndex(S.start).getBaseIndex() <
          S.end.getBaseIndex())
        return false;
    return true;
  }

  // Returns true if any segment in the live range contains any of the
  // provided slot indexes.  Slots which occur in holes between
  // segments will not cause the function to return true.
  bool isLiveAtIndexes(ArrayRef<SlotIndex> Slots) const;

  bool operator<(const LiveRange& other) const {
    const SlotIndex &thisIndex = beginIndex();
    const SlotIndex &otherIndex = other.beginIndex();
    return thisIndex < otherIndex;
  }

  /// Returns true if there is an explicit "undef" between @p Begin
  /// @p End.
  bool isUndefIn(ArrayRef<SlotIndex> Undefs, SlotIndex Begin,
                 SlotIndex End) const {
    return llvm::any_of(Undefs, [Begin, End](SlotIndex Idx) -> bool {
      return Begin <= Idx && Idx < End;
    });
  }

  /// Flush segment set into the regular segment vector.
  /// The method is to be called after the live range
  /// has been created, if use of the segment set was
  /// activated in the constructor of the live range.
  void flushSegmentSet();

  /// Stores indexes from the input index sequence R at which this LiveRange
  /// is live to the output O iterator.
  /// R is a range of _ascending sorted_ _random_ access iterators
  /// to the input indexes. Indexes stored at O are ascending sorted so it
  /// can be used directly in the subsequent search (for example for
  /// subranges). Returns true if found at least one index.
  template <typename Range, typename OutputIt>
  bool findIndexesLiveAt(Range &&R, OutputIt O) const {
    assert(llvm::is_sorted(R))(static_cast<void> (0));
    auto Idx = R.begin(), EndIdx = R.end();
    auto Seg = segments.begin(), EndSeg = segments.end();
    bool Found = false;
    while (Idx != EndIdx && Seg != EndSeg) {
      // if the Seg is lower find first segment that is above Idx using binary
      // search
      if (Seg->end <= *Idx) {
        Seg = std::upper_bound(
            ++Seg, EndSeg, *Idx,
            [=](std::remove_reference_t<decltype(*Idx)> V,
                const std::remove_reference_t<decltype(*Seg)> &S) {
              return V < S.end;
            });
        if (Seg == EndSeg)
          break;
      }
      auto NotLessStart = std::lower_bound(Idx, EndIdx, Seg->start);
      if (NotLessStart == EndIdx)
        break;
      auto NotLessEnd = std::lower_bound(NotLessStart, EndIdx, Seg->end);
      if (NotLessEnd != NotLessStart) {
        Found = true;
        O = std::copy(NotLessStart, NotLessEnd, O);
      }
      Idx = NotLessEnd;
      ++Seg;
    }
    return Found;
  }

  void print(raw_ostream &OS) const;
  void dump() const;

  /// Walk the range and assert if any invariants fail to hold.
  ///
  /// Note that this is a no-op when asserts are disabled.
657#ifdef NDEBUG1
  void verify() const {}
659#else
  void verify() const;
661#endif

protected:
  /// Append a segment to the list of segments.
  void append(const LiveRange::Segment S);

private:
  friend class LiveRangeUpdater;
  void addSegmentToSet(Segment S);
  void markValNoForDeletion(VNInfo *V);
};

inline raw_ostream &operator<<(raw_ostream &OS, const LiveRange &LR) {
  LR.print(OS);
  return OS;
}

/// LiveInterval - This class represents the liveness of a register,
/// or stack slot.
class LiveInterval : public LiveRange {
public:
  using super = LiveRange;

  /// A live range for subregisters. The LaneMask specifies which parts of the
  /// super register are covered by the interval.
  /// (@sa TargetRegisterInfo::getSubRegIndexLaneMask()).
  class SubRange : public LiveRange {
  public:
    SubRange *Next = nullptr;
    LaneBitmask LaneMask;

    /// Constructs a new SubRange object.
    SubRange(LaneBitmask LaneMask) : LaneMask(LaneMask) {}

    /// Constructs a new SubRange object by copying liveness from @p Other.
    SubRange(LaneBitmask LaneMask, const LiveRange &Other,
             BumpPtrAllocator &Allocator)
      : LiveRange(Other, Allocator), LaneMask(LaneMask) {}

    void print(raw_ostream &OS) const;
    void dump() const;
  };

private:
  SubRange *SubRanges = nullptr; ///< Single linked list of subregister live
                                 /// ranges.
  const Register Reg; // the register or stack slot of this interval.
  float Weight = 0.0; // weight of this interval

public:
  Register reg() const { return Reg; }
  float weight() const { return Weight; }
  void incrementWeight(float Inc) { Weight += Inc; }
  void setWeight(float Value) { Weight = Value; }

  LiveInterval(unsigned Reg, float Weight) : Reg(Reg), Weight(Weight) {}

  ~LiveInterval() {
    clearSubRanges();
  }

  template<typename T>
  class SingleLinkedListIterator {
    T *P;

  public:
    SingleLinkedListIterator<T>(T *P) : P(P) {}

    SingleLinkedListIterator<T> &operator++() {
      P = P->Next;
      return *this;
    }
    SingleLinkedListIterator<T> operator++(int) {
      SingleLinkedListIterator res = *this;
      ++*this;
      return res;
    }
    bool operator!=(const SingleLinkedListIterator<T> &Other) const {
      return P != Other.operator->();
    }
    bool operator==(const SingleLinkedListIterator<T> &Other) const {
      return P == Other.operator->();
    }
    T &operator*() const {
      return *P;
    }
    T *operator->() const {
      return P;
    }
  };

  using subrange_iterator = SingleLinkedListIterator<SubRange>;
  using const_subrange_iterator = SingleLinkedListIterator<const SubRange>;

  subrange_iterator subrange_begin() {
    return subrange_iterator(SubRanges);
  }
  subrange_iterator subrange_end() {
    return subrange_iterator(nullptr);
  }

  const_subrange_iterator subrange_begin() const {
    return const_subrange_iterator(SubRanges);
  }
  const_subrange_iterator subrange_end() const {
    return const_subrange_iterator(nullptr);
  }

  iterator_range<subrange_iterator> subranges() {
    return make_range(subrange_begin(), subrange_end());
  }

  iterator_range<const_subrange_iterator> subranges() const {
    return make_range(subrange_begin(), subrange_end());
  }

  /// Creates a new empty subregister live range. The range is added at the
  /// beginning of the subrange list; subrange iterators stay valid.
  SubRange *createSubRange(BumpPtrAllocator &Allocator,
                           LaneBitmask LaneMask) {
    SubRange *Range = new (Allocator) SubRange(LaneMask);
    appendSubRange(Range);
    return Range;
  }

  /// Like createSubRange() but the new range is filled with a copy of the
  /// liveness information in @p CopyFrom.
  SubRange *createSubRangeFrom(BumpPtrAllocator &Allocator,
                               LaneBitmask LaneMask,
                               const LiveRange &CopyFrom) {
    SubRange *Range = new (Allocator) SubRange(LaneMask, CopyFrom, Allocator);
    appendSubRange(Range);
    return Range;
  }

  /// Returns true if subregister liveness information is available.
  bool hasSubRanges() const {
    return SubRanges != nullptr;
  }

  /// Removes all subregister liveness information.
  void clearSubRanges();

  /// Removes all subranges without any segments (subranges without segments
  /// are not considered valid and should only exist temporarily).
  void removeEmptySubRanges();

  /// getSize - Returns the sum of sizes of all the LiveRange's.
  ///
  unsigned getSize() const;

  /// isSpillable - Can this interval be spilled?
  bool isSpillable() const { return Weight != huge_valf; }

  /// markNotSpillable - Mark interval as not spillable
  void markNotSpillable() { Weight = huge_valf; }

  /// For a given lane mask @p LaneMask, compute indexes at which the
  /// lane is marked undefined by subregister <def,read-undef> definitions.
  void computeSubRangeUndefs(SmallVectorImpl<SlotIndex> &Undefs,
                             LaneBitmask LaneMask,
                             const MachineRegisterInfo &MRI,
                             const SlotIndexes &Indexes) const;

  /// Refines the subranges to support \p LaneMask. This may only be called
  /// for LI.hasSubrange()==true. Subregister ranges are split or created
  /// until \p LaneMask can be matched exactly. \p Mod is executed on the
  /// matching subranges.
  ///
  /// Example:
  ///    Given an interval with subranges with lanemasks L0F00, L00F0 and
  ///    L000F, refining for mask L0018. Will split the L00F0 lane into
  ///    L00E0 and L0010 and the L000F lane into L0007 and L0008. The Mod
  ///    function will be applied to the L0010 and L0008 subranges.
  ///
  /// \p Indexes and \p TRI are required to clean up the VNIs that
  /// don't define the related lane masks after they get shrunk. E.g.,
  /// when L000F gets split into L0007 and L0008 maybe only a subset
  /// of the VNIs that defined L000F defines L0007.
  ///
  /// The clean up of the VNIs need to look at the actual instructions
  /// to decide what is or is not live at a definition point. If the
  /// update of the subranges occurs while the IR does not reflect these
  /// changes, \p ComposeSubRegIdx can be used to specify how the
  /// definition are going to be rewritten.
  /// E.g., let say we want to merge:
  ///     V1.sub1:<2 x s32> = COPY V2.sub3:<4 x s32>
  /// We do that by choosing a class where sub1:<2 x s32> and sub3:<4 x s32>
  /// overlap, i.e., by choosing a class where we can find "offset + 1 == 3".
  /// Put differently we align V2's sub3 with V1's sub1:
  /// V2: sub0 sub1 sub2 sub3
  /// V1: <offset>  sub0 sub1
  ///
  /// This offset will look like a composed subregidx in the the class:
  ///     V1.(composed sub2 with sub1):<4 x s32> = COPY V2.sub3:<4 x s32>
  /// =>  V1.(composed sub2 with sub1):<4 x s32> = COPY V2.sub3:<4 x s32>
  ///
  /// Now if we didn't rewrite the uses and def of V1, all the checks for V1
  /// need to account for this offset.
  /// This happens during coalescing where we update the live-ranges while
  /// still having the old IR around because updating the IR on-the-fly
  /// would actually clobber some information on how the live-ranges that
  /// are being updated look like.
  void refineSubRanges(BumpPtrAllocator &Allocator, LaneBitmask LaneMask,
                       std::function<void(LiveInterval::SubRange &)> Apply,
                       const SlotIndexes &Indexes,
                       const TargetRegisterInfo &TRI,
                       unsigned ComposeSubRegIdx = 0);

  bool operator<(const LiveInterval& other) const {
    const SlotIndex &thisIndex = beginIndex();
    const SlotIndex &otherIndex = other.beginIndex();
    return std::tie(thisIndex, Reg) < std::tie(otherIndex, other.Reg);
  }

  void print(raw_ostream &OS) const;
  void dump() const;

  /// Walks the interval and assert if any invariants fail to hold.
  ///
  /// Note that this is a no-op when asserts are disabled.
882#ifdef NDEBUG1
  void verify(const MachineRegisterInfo *MRI = nullptr) const {}
884#else
  void verify(const MachineRegisterInfo *MRI = nullptr) const;
886#endif

private:
  /// Appends @p Range to SubRanges list.
  void appendSubRange(SubRange *Range) {
    Range->Next = SubRanges;
    SubRanges = Range;
  }

  /// Free memory held by SubRange.
  void freeSubRange(SubRange *S);
};

inline raw_ostream &operator<<(raw_ostream &OS,
                               const LiveInterval::SubRange &SR) {
  SR.print(OS);
  return OS;
}

inline raw_ostream &operator<<(raw_ostream &OS, const LiveInterval &LI) {
  LI.print(OS);
  return OS;
}

raw_ostream &operator<<(raw_ostream &OS, const LiveRange::Segment &S);

inline bool operator<(SlotIndex V, const LiveRange::Segment &S) {
  return V < S.start;
}

inline bool operator<(const LiveRange::Segment &S, SlotIndex V) {
  return S.start < V;
}

/// Helper class for performant LiveRange bulk updates.
///
/// Calling LiveRange::addSegment() repeatedly can be expensive on large
/// live ranges because segments after the insertion point may need to be
/// shifted. The LiveRangeUpdater class can defer the shifting when adding
/// many segments in order.
///
/// The LiveRange will be in an invalid state until flush() is called.
class LiveRangeUpdater {
  LiveRange *LR;
  SlotIndex LastStart;
  LiveRange::iterator WriteI;
  LiveRange::iterator ReadI;
  SmallVector<LiveRange::Segment, 16> Spills;
  void mergeSpills();

public:
  /// Create a LiveRangeUpdater for adding segments to LR.
  /// LR will temporarily be in an invalid state until flush() is called.
  LiveRangeUpdater(LiveRange *lr = nullptr) : LR(lr) {}

  ~LiveRangeUpdater() { flush(); }

  /// Add a segment to LR and coalesce when possible, just like
  /// LR.addSegment(). Segments should be added in increasing start order for
  /// best performance.
  void add(LiveRange::Segment);

  void add(SlotIndex Start, SlotIndex End, VNInfo *VNI) {
    add(LiveRange::Segment(Start, End, VNI));
  }

  /// Return true if the LR is currently in an invalid state, and flush()
  /// needs to be called.
  bool isDirty() const { return LastStart.isValid(); }

  /// Flush the updater state to LR so it is valid and contains all added
  /// segments.
  void flush();

  /// Select a different destination live range.
  void setDest(LiveRange *lr) {
    if (LR != lr && isDirty())
      flush();
    LR = lr;
  }

  /// Get the current destination live range.
  LiveRange *getDest() const { return LR; }

  void dump() const;
  void print(raw_ostream&) const;
};

inline raw_ostream &operator<<(raw_ostream &OS, const LiveRangeUpdater &X) {
  X.print(OS);
  return OS;
}

/// ConnectedVNInfoEqClasses - Helper class that can divide VNInfos in a
/// LiveInterval into equivalence clases of connected components. A
/// LiveInterval that has multiple connected components can be broken into
/// multiple LiveIntervals.
///
/// Given a LiveInterval that may have multiple connected components, run:
///
///   unsigned numComps = ConEQ.Classify(LI);
///   if (numComps > 1) {
///     // allocate numComps-1 new LiveIntervals into LIS[1..]
///     ConEQ.Distribute(LIS);
/// }

class ConnectedVNInfoEqClasses {
  LiveIntervals &LIS;
  IntEqClasses EqClass;

public:
  explicit ConnectedVNInfoEqClasses(LiveIntervals &lis) : LIS(lis) {}

  /// Classify the values in \p LR into connected components.
  /// Returns the number of connected components.
  unsigned Classify(const LiveRange &LR);

  /// getEqClass - Classify creates equivalence classes numbered 0..N. Return
  /// the equivalence class assigned the VNI.
  unsigned getEqClass(const VNInfo *VNI) const { return EqClass[VNI->id]; }

  /// Distribute values in \p LI into a separate LiveIntervals
  /// for each connected component. LIV must have an empty LiveInterval for
  /// each additional connected component. The first connected component is
  /// left in \p LI.
  void Distribute(LiveInterval &LI, LiveInterval *LIV[],
                  MachineRegisterInfo &MRI);
};

1015} // end namespace llvm

1017#endif // LLVM_CODEGEN_LIVEINTERVAL_H

←

/build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/include/llvm/CodeGen/SlotIndexes.h

→

1//===- llvm/CodeGen/SlotIndexes.h - Slot indexes representation -*- 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 implements SlotIndex and related classes. The purpose of SlotIndex
10// is to describe a position at which a register can become live, or cease to
11// be live.
12//
13// SlotIndex is mostly a proxy for entries of the SlotIndexList, a class which
14// is held is LiveIntervals and provides the real numbering. This allows
15// LiveIntervals to perform largely transparent renumbering.
16//===----------------------------------------------------------------------===//

18#ifndef LLVM_CODEGEN_SLOTINDEXES_H
19#define LLVM_CODEGEN_SLOTINDEXES_H

21#include "llvm/ADT/DenseMap.h"
22#include "llvm/ADT/IntervalMap.h"
23#include "llvm/ADT/PointerIntPair.h"
24#include "llvm/ADT/SmallVector.h"
25#include "llvm/ADT/ilist.h"
26#include "llvm/CodeGen/MachineBasicBlock.h"
27#include "llvm/CodeGen/MachineFunction.h"
28#include "llvm/CodeGen/MachineFunctionPass.h"
29#include "llvm/CodeGen/MachineInstr.h"
30#include "llvm/CodeGen/MachineInstrBundle.h"
31#include "llvm/Pass.h"
32#include "llvm/Support/Allocator.h"
33#include <algorithm>
34#include <cassert>
35#include <iterator>
36#include <utility>

38namespace llvm {

40class raw_ostream;

/// This class represents an entry in the slot index list held in the
/// SlotIndexes pass. It should not be used directly. See the
/// SlotIndex & SlotIndexes classes for the public interface to this
/// information.
class IndexListEntry : public ilist_node<IndexListEntry> {
  MachineInstr *mi;
  unsigned index;

public:
  IndexListEntry(MachineInstr *mi, unsigned index) : mi(mi), index(index) {}

  MachineInstr* getInstr() const { return mi; }
  void setInstr(MachineInstr *mi) {
    this->mi = mi;
  }

  unsigned getIndex() const { return index; }
  void setIndex(unsigned index) {
    this->index = index;
  }

63#ifdef EXPENSIVE_CHECKS
  // When EXPENSIVE_CHECKS is defined, "erased" index list entries will
  // actually be moved to a "graveyard" list, and have their pointers
  // poisoned, so that dangling SlotIndex access can be reliably detected.
  void setPoison() {
    intptr_t tmp = reinterpret_cast<intptr_t>(mi);
    assert(((tmp & 0x1) == 0x0) && "Pointer already poisoned?")(static_cast<void> (0));
    tmp |= 0x1;
    mi = reinterpret_cast<MachineInstr*>(tmp);
  }

  bool isPoisoned() const { return (reinterpret_cast<intptr_t>(mi) & 0x1) == 0x1; }
75#endif // EXPENSIVE_CHECKS
};

template <>
struct ilist_alloc_traits<IndexListEntry>
    : public ilist_noalloc_traits<IndexListEntry> {};

/// SlotIndex - An opaque wrapper around machine indexes.
class SlotIndex {
  friend class SlotIndexes;

  enum Slot {
    /// Basic block boundary.  Used for live ranges entering and leaving a
    /// block without being live in the layout neighbor.  Also used as the
    /// def slot of PHI-defs.
    Slot_Block,

    /// Early-clobber register use/def slot.  A live range defined at
    /// Slot_EarlyClobber interferes with normal live ranges killed at
    /// Slot_Register.  Also used as the kill slot for live ranges tied to an
    /// early-clobber def.
    Slot_EarlyClobber,

    /// Normal register use/def slot.  Normal instructions kill and define
    /// register live ranges at this slot.
    Slot_Register,

    /// Dead def kill point.  Kill slot for a live range that is defined by
    /// the same instruction (Slot_Register or Slot_EarlyClobber), but isn't
    /// used anywhere.
    Slot_Dead,

    Slot_Count
  };

  PointerIntPair<IndexListEntry*, 2, unsigned> lie;

  SlotIndex(IndexListEntry *entry, unsigned slot)
    : lie(entry, slot) {}

  IndexListEntry* listEntry() const {
    assert(isValid() && "Attempt to compare reserved index.")(static_cast<void> (0));
117#ifdef EXPENSIVE_CHECKS
    assert(!lie.getPointer()->isPoisoned() &&(static_cast<void> (0))
           "Attempt to access deleted list-entry.")(static_cast<void> (0));
120#endif // EXPENSIVE_CHECKS
    return lie.getPointer();
  }

  unsigned getIndex() const {
    return listEntry()->getIndex() | getSlot();
  }

  /// Returns the slot for this SlotIndex.
  Slot getSlot() const {
    return static_cast<Slot>(lie.getInt());
  }

public:
  enum {
    /// The default distance between instructions as returned by distance().
    /// This may vary as instructions are inserted and removed.
    InstrDist = 4 * Slot_Count
  };

  /// Construct an invalid index.
  SlotIndex() = default;

  // Construct a new slot index from the given one, and set the slot.
  SlotIndex(const SlotIndex &li, Slot s) : lie(li.listEntry(), unsigned(s)) {
    assert(lie.getPointer() != nullptr &&(static_cast<void> (0))
           "Attempt to construct index with 0 pointer.")(static_cast<void> (0));
  }

  /// Returns true if this is a valid index. Invalid indices do
  /// not point into an index table, and cannot be compared.
  bool isValid() const {
    return lie.getPointer();
  }

  /// Return true for a valid index.
  explicit operator bool() const { return isValid(); }

  /// Print this index to the given raw_ostream.
  void print(raw_ostream &os) const;

  /// Dump this index to stderr.
  void dump() const;

  /// Compare two SlotIndex objects for equality.
  bool operator==(SlotIndex other) const {
    return lie == other.lie;
  }
  /// Compare two SlotIndex objects for inequality.
  bool operator!=(SlotIndex other) const {
    return lie != other.lie;
  }

  /// Compare two SlotIndex objects. Return true if the first index
  /// is strictly lower than the second.
  bool operator<(SlotIndex other) const {
    return getIndex() < other.getIndex();
  }
  /// Compare two SlotIndex objects. Return true if the first index
  /// is lower than, or equal to, the second.
  bool operator<=(SlotIndex other) const {
    return getIndex() <= other.getIndex();
  }

  /// Compare two SlotIndex objects. Return true if the first index
  /// is greater than the second.
  bool operator>(SlotIndex other) const {
    return getIndex() > other.getIndex();
  }

  /// Compare two SlotIndex objects. Return true if the first index
  /// is greater than, or equal to, the second.
  bool operator>=(SlotIndex other) const {
    return getIndex() >= other.getIndex();
  }

  /// isSameInstr - Return true if A and B refer to the same instruction.
  static bool isSameInstr(SlotIndex A, SlotIndex B) {
    return A.lie.getPointer() == B.lie.getPointer();
  }

  /// isEarlierInstr - Return true if A refers to an instruction earlier than
  /// B. This is equivalent to A < B && !isSameInstr(A, B).
  static bool isEarlierInstr(SlotIndex A, SlotIndex B) {
    return A.listEntry()->getIndex() < B.listEntry()->getIndex();
  }

  /// Return true if A refers to the same instruction as B or an earlier one.
  /// This is equivalent to !isEarlierInstr(B, A).
  static bool isEarlierEqualInstr(SlotIndex A, SlotIndex B) {
    return !isEarlierInstr(B, A);
  }

  /// Return the distance from this index to the given one.
  int distance(SlotIndex other) const {
    return other.getIndex() - getIndex();
  }

  /// Return the scaled distance from this index to the given one, where all
  /// slots on the same instruction have zero distance.
  int getInstrDistance(SlotIndex other) const {
    return (other.listEntry()->getIndex() - listEntry()->getIndex())
      / Slot_Count;
  }

  /// isBlock - Returns true if this is a block boundary slot.
  bool isBlock() const { return getSlot() == Slot_Block; }
4
←
Assuming the condition is false→
5
←
Returning zero, which participates in a condition later→

  /// isEarlyClobber - Returns true if this is an early-clobber slot.
  bool isEarlyClobber() const { return getSlot() == Slot_EarlyClobber; }

  /// isRegister - Returns true if this is a normal register use/def slot.
  /// Note that early-clobber slots may also be used for uses and defs.
  bool isRegister() const { return getSlot() == Slot_Register; }

  /// isDead - Returns true if this is a dead def kill slot.
  bool isDead() const { return getSlot() == Slot_Dead; }

  /// Returns the base index for associated with this index. The base index
  /// is the one associated with the Slot_Block slot for the instruction
  /// pointed to by this index.
  SlotIndex getBaseIndex() const {
    return SlotIndex(listEntry(), Slot_Block);
  }

  /// Returns the boundary index for associated with this index. The boundary
  /// index is the one associated with the Slot_Block slot for the instruction
  /// pointed to by this index.
  SlotIndex getBoundaryIndex() const {
    return SlotIndex(listEntry(), Slot_Dead);
  }

  /// Returns the register use/def slot in the current instruction for a
  /// normal or early-clobber def.
  SlotIndex getRegSlot(bool EC = false) const {
    return SlotIndex(listEntry(), EC ? Slot_EarlyClobber : Slot_Register);
  }

  /// Returns the dead def kill slot for the current instruction.
  SlotIndex getDeadSlot() const {
    return SlotIndex(listEntry(), Slot_Dead);
  }

  /// Returns the next slot in the index list. This could be either the
  /// next slot for the instruction pointed to by this index or, if this
  /// index is a STORE, the first slot for the next instruction.
  /// WARNING: This method is considerably more expensive than the methods
  /// that return specific slots (getUseIndex(), etc). If you can - please
  /// use one of those methods.
  SlotIndex getNextSlot() const {
    Slot s = getSlot();
    if (s == Slot_Dead) {
      return SlotIndex(&*++listEntry()->getIterator(), Slot_Block);
    }
    return SlotIndex(listEntry(), s + 1);
  }

  /// Returns the next index. This is the index corresponding to the this
  /// index's slot, but for the next instruction.
  SlotIndex getNextIndex() const {
    return SlotIndex(&*++listEntry()->getIterator(), getSlot());
  }

  /// Returns the previous slot in the index list. This could be either the
  /// previous slot for the instruction pointed to by this index or, if this
  /// index is a Slot_Block, the last slot for the previous instruction.
  /// WARNING: This method is considerably more expensive than the methods
  /// that return specific slots (getUseIndex(), etc). If you can - please
  /// use one of those methods.
  SlotIndex getPrevSlot() const {
    Slot s = getSlot();
    if (s == Slot_Block) {
      return SlotIndex(&*--listEntry()->getIterator(), Slot_Dead);
    }
    return SlotIndex(listEntry(), s - 1);
  }

  /// Returns the previous index. This is the index corresponding to this
  /// index's slot, but for the previous instruction.
  SlotIndex getPrevIndex() const {
    return SlotIndex(&*--listEntry()->getIterator(), getSlot());
  }
};

inline raw_ostream& operator<<(raw_ostream &os, SlotIndex li) {
  li.print(os);
  return os;
}

using IdxMBBPair = std::pair<SlotIndex, MachineBasicBlock *>;

/// SlotIndexes pass.
///
/// This pass assigns indexes to each instruction.
class SlotIndexes : public MachineFunctionPass {
private:
  // IndexListEntry allocator.
  BumpPtrAllocator ileAllocator;

  using IndexList = ilist<IndexListEntry>;
  IndexList indexList;

  MachineFunction *mf;

  using Mi2IndexMap = DenseMap<const MachineInstr *, SlotIndex>;
  Mi2IndexMap mi2iMap;

  /// MBBRanges - Map MBB number to (start, stop) indexes.
  SmallVector<std::pair<SlotIndex, SlotIndex>, 8> MBBRanges;

  /// Idx2MBBMap - Sorted list of pairs of index of first instruction
  /// and MBB id.
  SmallVector<IdxMBBPair, 8> idx2MBBMap;

  IndexListEntry* createEntry(MachineInstr *mi, unsigned index) {
    IndexListEntry *entry =
        static_cast<IndexListEntry *>(ileAllocator.Allocate(
            sizeof(IndexListEntry), alignof(IndexListEntry)));

    new (entry) IndexListEntry(mi, index);

    return entry;
  }

  /// Renumber locally after inserting curItr.
  void renumberIndexes(IndexList::iterator curItr);

public:
  static char ID;

  SlotIndexes();

  ~SlotIndexes() override;

  void getAnalysisUsage(AnalysisUsage &au) const override;
  void releaseMemory() override;

  bool runOnMachineFunction(MachineFunction &fn) override;

  /// Dump the indexes.
  void dump() const;

  /// Repair indexes after adding and removing instructions.
  void repairIndexesInRange(MachineBasicBlock *MBB,
                            MachineBasicBlock::iterator Begin,
                            MachineBasicBlock::iterator End);

  /// Returns the zero index for this analysis.
  SlotIndex getZeroIndex() {
    assert(indexList.front().getIndex() == 0 && "First index is not 0?")(static_cast<void> (0));
    return SlotIndex(&indexList.front(), 0);
  }

  /// Returns the base index of the last slot in this analysis.
  SlotIndex getLastIndex() {
    return SlotIndex(&indexList.back(), 0);
  }

  /// Returns true if the given machine instr is mapped to an index,
  /// otherwise returns false.
  bool hasIndex(const MachineInstr &instr) const {
    return mi2iMap.count(&instr);
  }

  /// Returns the base index for the given instruction.
  SlotIndex getInstructionIndex(const MachineInstr &MI,
                                bool IgnoreBundle = false) const {
    // Instructions inside a bundle have the same number as the bundle itself.
    auto BundleStart = getBundleStart(MI.getIterator());
    auto BundleEnd = getBundleEnd(MI.getIterator());
    // Use the first non-debug instruction in the bundle to get SlotIndex.
    const MachineInstr &BundleNonDebug =
        IgnoreBundle ? MI
                     : *skipDebugInstructionsForward(BundleStart, BundleEnd);
    assert(!BundleNonDebug.isDebugInstr() &&(static_cast<void> (0))
           "Could not use a debug instruction to query mi2iMap.")(static_cast<void> (0));
    Mi2IndexMap::const_iterator itr = mi2iMap.find(&BundleNonDebug);
    assert(itr != mi2iMap.end() && "Instruction not found in maps.")(static_cast<void> (0));
    return itr->second;
  }

  /// Returns the instruction for the given index, or null if the given
  /// index has no instruction associated with it.
  MachineInstr* getInstructionFromIndex(SlotIndex index) const {
    return index.isValid() ? index.listEntry()->getInstr() : nullptr;
  }

  /// Returns the next non-null index, if one exists.
  /// Otherwise returns getLastIndex().
  SlotIndex getNextNonNullIndex(SlotIndex Index) {
    IndexList::iterator I = Index.listEntry()->getIterator();
    IndexList::iterator E = indexList.end();
    while (++I != E)
      if (I->getInstr())
        return SlotIndex(&*I, Index.getSlot());
    // We reached the end of the function.
    return getLastIndex();
  }

  /// getIndexBefore - Returns the index of the last indexed instruction
  /// before MI, or the start index of its basic block.
  /// MI is not required to have an index.
  SlotIndex getIndexBefore(const MachineInstr &MI) const {
    const MachineBasicBlock *MBB = MI.getParent();
    assert(MBB && "MI must be inserted in a basic block")(static_cast<void> (0));
    MachineBasicBlock::const_iterator I = MI, B = MBB->begin();
    while (true) {
      if (I == B)
        return getMBBStartIdx(MBB);
      --I;
      Mi2IndexMap::const_iterator MapItr = mi2iMap.find(&*I);
      if (MapItr != mi2iMap.end())
        return MapItr->second;
    }
  }

  /// getIndexAfter - Returns the index of the first indexed instruction
  /// after MI, or the end index of its basic block.
  /// MI is not required to have an index.
  SlotIndex getIndexAfter(const MachineInstr &MI) const {
    const MachineBasicBlock *MBB = MI.getParent();
    assert(MBB && "MI must be inserted in a basic block")(static_cast<void> (0));
    MachineBasicBlock::const_iterator I = MI, E = MBB->end();
    while (true) {
      ++I;
      if (I == E)
        return getMBBEndIdx(MBB);
      Mi2IndexMap::const_iterator MapItr = mi2iMap.find(&*I);
      if (MapItr != mi2iMap.end())
        return MapItr->second;
    }
  }

  /// Return the (start,end) range of the given basic block number.
  const std::pair<SlotIndex, SlotIndex> &
  getMBBRange(unsigned Num) const {
    return MBBRanges[Num];
  }

  /// Return the (start,end) range of the given basic block.
  const std::pair<SlotIndex, SlotIndex> &
  getMBBRange(const MachineBasicBlock *MBB) const {
    return getMBBRange(MBB->getNumber());
  }

  /// Returns the first index in the given basic block number.
  SlotIndex getMBBStartIdx(unsigned Num) const {
    return getMBBRange(Num).first;
  }

  /// Returns the first index in the given basic block.
  SlotIndex getMBBStartIdx(const MachineBasicBlock *mbb) const {
    return getMBBRange(mbb).first;
  }

  /// Returns the last index in the given basic block number.
  SlotIndex getMBBEndIdx(unsigned Num) const {
    return getMBBRange(Num).second;
  }

  /// Returns the last index in the given basic block.
  SlotIndex getMBBEndIdx(const MachineBasicBlock *mbb) const {
    return getMBBRange(mbb).second;
  }

  /// Iterator over the idx2MBBMap (sorted pairs of slot index of basic block
  /// begin and basic block)
  using MBBIndexIterator = SmallVectorImpl<IdxMBBPair>::const_iterator;

  /// Move iterator to the next IdxMBBPair where the SlotIndex is greater or
  /// equal to \p To.
  MBBIndexIterator advanceMBBIndex(MBBIndexIterator I, SlotIndex To) const {
    return std::partition_point(
        I, idx2MBBMap.end(),
        [=](const IdxMBBPair &IM) { return IM.first < To; });
  }

  /// Get an iterator pointing to the IdxMBBPair with the biggest SlotIndex
  /// that is greater or equal to \p Idx.
  MBBIndexIterator findMBBIndex(SlotIndex Idx) const {
    return advanceMBBIndex(idx2MBBMap.begin(), Idx);
  }

  /// Returns an iterator for the begin of the idx2MBBMap.
  MBBIndexIterator MBBIndexBegin() const {
    return idx2MBBMap.begin();
  }

  /// Return an iterator for the end of the idx2MBBMap.
  MBBIndexIterator MBBIndexEnd() const {
    return idx2MBBMap.end();
  }

  /// Returns the basic block which the given index falls in.
  MachineBasicBlock* getMBBFromIndex(SlotIndex index) const {
    if (MachineInstr *MI = getInstructionFromIndex(index))
      return MI->getParent();

    MBBIndexIterator I = findMBBIndex(index);
    // Take the pair containing the index
    MBBIndexIterator J =
      ((I != MBBIndexEnd() && I->first > index) ||
       (I == MBBIndexEnd() && !idx2MBBMap.empty())) ? std::prev(I) : I;

    assert(J != MBBIndexEnd() && J->first <= index &&(static_cast<void> (0))
           index < getMBBEndIdx(J->second) &&(static_cast<void> (0))
           "index does not correspond to an MBB")(static_cast<void> (0));
    return J->second;
  }

  /// Insert the given machine instruction into the mapping. Returns the
  /// assigned index.
  /// If Late is set and there are null indexes between mi's neighboring
  /// instructions, create the new index after the null indexes instead of
  /// before them.
  SlotIndex insertMachineInstrInMaps(MachineInstr &MI, bool Late = false) {
    assert(!MI.isInsideBundle() &&(static_cast<void> (0))
           "Instructions inside bundles should use bundle start's slot.")(static_cast<void> (0));
    assert(mi2iMap.find(&MI) == mi2iMap.end() && "Instr already indexed.")(static_cast<void> (0));
    // Numbering debug instructions could cause code generation to be
    // affected by debug information.
    assert(!MI.isDebugInstr() && "Cannot number debug instructions.")(static_cast<void> (0));

    assert(MI.getParent() != nullptr && "Instr must be added to function.")(static_cast<void> (0));

    // Get the entries where MI should be inserted.
    IndexList::iterator prevItr, nextItr;
    if (Late) {
      // Insert MI's index immediately before the following instruction.
      nextItr = getIndexAfter(MI).listEntry()->getIterator();
      prevItr = std::prev(nextItr);
    } else {
      // Insert MI's index immediately after the preceding instruction.
      prevItr = getIndexBefore(MI).listEntry()->getIterator();
      nextItr = std::next(prevItr);
    }

    // Get a number for the new instr, or 0 if there's no room currently.
    // In the latter case we'll force a renumber later.
    unsigned dist = ((nextItr->getIndex() - prevItr->getIndex())/2) & ~3u;
    unsigned newNumber = prevItr->getIndex() + dist;

    // Insert a new list entry for MI.
    IndexList::iterator newItr =
        indexList.insert(nextItr, createEntry(&MI, newNumber));

    // Renumber locally if we need to.
    if (dist == 0)
      renumberIndexes(newItr);

    SlotIndex newIndex(&*newItr, SlotIndex::Slot_Block);
    mi2iMap.insert(std::make_pair(&MI, newIndex));
    return newIndex;
  }

  /// Removes machine instruction (bundle) \p MI from the mapping.
  /// This should be called before MachineInstr::eraseFromParent() is used to
  /// remove a whole bundle or an unbundled instruction.
  /// If \p AllowBundled is set then this can be used on a bundled
  /// instruction; however, this exists to support handleMoveIntoBundle,
  /// and in general removeSingleMachineInstrFromMaps should be used instead.
  void removeMachineInstrFromMaps(MachineInstr &MI,
                                  bool AllowBundled = false);

  /// Removes a single machine instruction \p MI from the mapping.
  /// This should be called before MachineInstr::eraseFromBundle() is used to
  /// remove a single instruction (out of a bundle).
  void removeSingleMachineInstrFromMaps(MachineInstr &MI);

  /// ReplaceMachineInstrInMaps - Replacing a machine instr with a new one in
  /// maps used by register allocator. \returns the index where the new
  /// instruction was inserted.
  SlotIndex replaceMachineInstrInMaps(MachineInstr &MI, MachineInstr &NewMI) {
    Mi2IndexMap::iterator mi2iItr = mi2iMap.find(&MI);
    if (mi2iItr == mi2iMap.end())
      return SlotIndex();
    SlotIndex replaceBaseIndex = mi2iItr->second;
    IndexListEntry *miEntry(replaceBaseIndex.listEntry());
    assert(miEntry->getInstr() == &MI &&(static_cast<void> (0))
           "Mismatched instruction in index tables.")(static_cast<void> (0));
    miEntry->setInstr(&NewMI);
    mi2iMap.erase(mi2iItr);
    mi2iMap.insert(std::make_pair(&NewMI, replaceBaseIndex));
    return replaceBaseIndex;
  }

  /// Add the given MachineBasicBlock into the maps.
  /// If it contains any instructions then they must already be in the maps.
  /// This is used after a block has been split by moving some suffix of its
  /// instructions into a newly created block.
  void insertMBBInMaps(MachineBasicBlock *mbb) {
    assert(mbb != &mbb->getParent()->front() &&(static_cast<void> (0))
           "Can't insert a new block at the beginning of a function.")(static_cast<void> (0));
    auto prevMBB = std::prev(MachineFunction::iterator(mbb));

    // Create a new entry to be used for the start of mbb and the end of
    // prevMBB.
    IndexListEntry *startEntry = createEntry(nullptr, 0);
    IndexListEntry *endEntry = getMBBEndIdx(&*prevMBB).listEntry();
    IndexListEntry *insEntry =
        mbb->empty() ? endEntry
                     : getInstructionIndex(mbb->front()).listEntry();
    IndexList::iterator newItr =
        indexList.insert(insEntry->getIterator(), startEntry);

    SlotIndex startIdx(startEntry, SlotIndex::Slot_Block);
    SlotIndex endIdx(endEntry, SlotIndex::Slot_Block);

    MBBRanges[prevMBB->getNumber()].second = startIdx;

    assert(unsigned(mbb->getNumber()) == MBBRanges.size() &&(static_cast<void> (0))
           "Blocks must be added in order")(static_cast<void> (0));
    MBBRanges.push_back(std::make_pair(startIdx, endIdx));
    idx2MBBMap.push_back(IdxMBBPair(startIdx, mbb));

    renumberIndexes(newItr);
    llvm::sort(idx2MBBMap, less_first());
  }
};

// Specialize IntervalMapInfo for half-open slot index intervals.
template <>
struct IntervalMapInfo<SlotIndex> : IntervalMapHalfOpenInfo<SlotIndex> {
};

645} // end namespace llvm

647#endif // LLVM_CODEGEN_SLOTINDEXES_H

←

/build/llvm-toolchain-snapshot-14~++20210903100615+fd66b44ec19e/llvm/include/llvm/CodeGen/Register.h

1//===-- llvm/CodeGen/Register.h ---------------------------------*- 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//===----------------------------------------------------------------------===//

9#ifndef LLVM_CODEGEN_REGISTER_H
10#define LLVM_CODEGEN_REGISTER_H

12#include "llvm/MC/MCRegister.h"
13#include <cassert>

15namespace llvm {

17/// Wrapper class representing virtual and physical registers. Should be passed
18/// by value.
19class Register {
unsigned Reg;

22public:
constexpr Register(unsigned Val = 0): Reg(Val) {}
constexpr Register(MCRegister Val): Reg(Val) {}

// Register numbers can represent physical registers, virtual registers, and
// sometimes stack slots. The unsigned values are divided into these ranges:
//
//   0           Not a register, can be used as a sentinel.
//   [1;2^30)    Physical registers assigned by TableGen.
//   [2^30;2^31) Stack slots. (Rarely used.)
//   [2^31;2^32) Virtual registers assigned by MachineRegisterInfo.
//
// Further sentinels can be allocated from the small negative integers.
// DenseMapInfo<unsigned> uses -1u and -2u.
static_assert(std::numeric_limits<decltype(Reg)>::max() >= 0xFFFFFFFF,
              "Reg isn't large enough to hold full range.");

/// isStackSlot - Sometimes it is useful the be able to store a non-negative
/// frame index in a variable that normally holds a register. isStackSlot()
/// returns true if Reg is in the range used for stack slots.
///
/// FIXME: remove in favor of member.
static bool isStackSlot(unsigned Reg) {
  return MCRegister::isStackSlot(Reg);
}

/// Return true if this is a stack slot.
bool isStack() const { return MCRegister::isStackSlot(Reg); }

/// Compute the frame index from a register value representing a stack slot.
static int stackSlot2Index(Register Reg) {
  assert(Reg.isStack() && "Not a stack slot")(static_cast<void> (0));
  return int(Reg - MCRegister::FirstStackSlot);
}

/// Convert a non-negative frame index to a stack slot register value.
static Register index2StackSlot(int FI) {
  assert(FI >= 0 && "Cannot hold a negative frame index.")(static_cast<void> (0));
  return Register(FI + MCRegister::FirstStackSlot);
}

/// Return true if the specified register number is in
/// the physical register namespace.
static bool isPhysicalRegister(unsigned Reg) {
  return MCRegister::isPhysicalRegister(Reg);
}

/// Return true if the specified register number is in
/// the virtual register namespace.
static bool isVirtualRegister(unsigned Reg) {
  return Reg & MCRegister::VirtualRegFlag && !isStackSlot(Reg);
}

/// Convert a virtual register number to a 0-based index.
/// The first virtual register in a function will get the index 0.
static unsigned virtReg2Index(Register Reg) {
  assert(isVirtualRegister(Reg) && "Not a virtual register")(static_cast<void> (0));
  return Reg & ~MCRegister::VirtualRegFlag;
}

/// Convert a 0-based index to a virtual register number.
/// This is the inverse operation of VirtReg2IndexFunctor below.
static Register index2VirtReg(unsigned Index) {
  assert(Index < (1u << 31) && "Index too large for virtual register range.")(static_cast<void> (0));
  return Index | MCRegister::VirtualRegFlag;
}

/// Return true if the specified register number is in the virtual register
/// namespace.
bool isVirtual() const {
  return isVirtualRegister(Reg);
}

/// Return true if the specified register number is in the physical register
/// namespace.
bool isPhysical() const {
  return isPhysicalRegister(Reg);
}

/// Convert a virtual register number to a 0-based index. The first virtual
/// register in a function will get the index 0.
unsigned virtRegIndex() const {
  return virtReg2Index(Reg);
}

constexpr operator unsigned() const {
  return Reg;
}

unsigned id() const { return Reg; }

operator MCRegister() const {
  return MCRegister(Reg);
}

/// Utility to check-convert this value to a MCRegister. The caller is
/// expected to have already validated that this Register is, indeed,
/// physical.
MCRegister asMCReg() const {
  assert(Reg == MCRegister::NoRegister ||(static_cast<void> (0))
         MCRegister::isPhysicalRegister(Reg))(static_cast<void> (0));
  return MCRegister(Reg);
}

bool isValid() const { return Reg != MCRegister::NoRegister; }

/// Comparisons between register objects
bool operator==(const Register &Other) const { return Reg == Other.Reg; }
18
←
Assuming 'Reg' is not equal to 'Other.Reg'→
19
←
Returning zero, which participates in a condition later→
bool operator!=(const Register &Other) const { return Reg != Other.Reg; }
bool operator==(const MCRegister &Other) const { return Reg == Other.id(); }
bool operator!=(const MCRegister &Other) const { return Reg != Other.id(); }

/// Comparisons against register constants. E.g.
/// * R == AArch64::WZR
/// * R == 0
/// * R == VirtRegMap::NO_PHYS_REG
bool operator==(unsigned Other) const { return Reg == Other; }
bool operator!=(unsigned Other) const { return Reg != Other; }
bool operator==(int Other) const { return Reg == unsigned(Other); }
bool operator!=(int Other) const { return Reg != unsigned(Other); }
// MSVC requires that we explicitly declare these two as well.
bool operator==(MCPhysReg Other) const { return Reg == unsigned(Other); }
bool operator!=(MCPhysReg Other) const { return Reg != unsigned(Other); }
145};

147// Provide DenseMapInfo for Register
148template<> struct DenseMapInfo<Register> {
static inline unsigned getEmptyKey() {
  return DenseMapInfo<unsigned>::getEmptyKey();
}
static inline unsigned getTombstoneKey() {
  return DenseMapInfo<unsigned>::getTombstoneKey();
}
static unsigned getHashValue(const Register &Val) {
  return DenseMapInfo<unsigned>::getHashValue(Val.id());
}
static bool isEqual(const Register &LHS, const Register &RHS) {
  return DenseMapInfo<unsigned>::isEqual(LHS.id(), RHS.id());
}
161};

163}

165#endif // LLVM_CODEGEN_REGISTER_H