/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp

Bug Summary

File:	build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp
Warning:	line 2932, column 21 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 -clear-ast-before-backend -disable-llvm-verifier -discard-value-names -main-file-name SimpleLoopUnswitch.cpp -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 -ffp-contract=on -fno-rounding-math -mconstructor-aliases -funwind-tables=2 -target-cpu x86-64 -tune-cpu generic -debugger-tuning=gdb -ffunction-sections -fdata-sections -fcoverage-compilation-dir=/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/build-llvm -resource-dir /usr/lib/llvm-16/lib/clang/16.0.0 -D _DEBUG -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I lib/Transforms/Scalar -I /build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/llvm/lib/Transforms/Scalar -I include -I /build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/llvm/include -D _FORTIFY_SOURCE=2 -D NDEBUG -U 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-16/lib/clang/16.0.0/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 -fmacro-prefix-map=/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/build-llvm=build-llvm -fmacro-prefix-map=/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/= -fcoverage-prefix-map=/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/build-llvm=build-llvm -fcoverage-prefix-map=/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/= -O3 -Wno-unused-command-line-argument -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 -Wno-misleading-indentation -std=c++17 -fdeprecated-macro -fdebug-compilation-dir=/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/build-llvm -fdebug-prefix-map=/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/build-llvm=build-llvm -fdebug-prefix-map=/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/= -ferror-limit 19 -fvisibility-inlines-hidden -stack-protector 2 -fgnuc-version=4.2.1 -fcolor-diagnostics -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-2022-10-03-140002-15933-1 -x c++ /build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp

/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp

→

1///===- SimpleLoopUnswitch.cpp - Hoist loop-invariant control flow ---------===//
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#include "llvm/Transforms/Scalar/SimpleLoopUnswitch.h"
10#include "llvm/ADT/DenseMap.h"
11#include "llvm/ADT/STLExtras.h"
12#include "llvm/ADT/Sequence.h"
13#include "llvm/ADT/SetVector.h"
14#include "llvm/ADT/SmallPtrSet.h"
15#include "llvm/ADT/SmallVector.h"
16#include "llvm/ADT/Statistic.h"
17#include "llvm/ADT/Twine.h"
18#include "llvm/Analysis/AssumptionCache.h"
19#include "llvm/Analysis/BlockFrequencyInfo.h"
20#include "llvm/Analysis/CFG.h"
21#include "llvm/Analysis/CodeMetrics.h"
22#include "llvm/Analysis/GuardUtils.h"
23#include "llvm/Analysis/LoopAnalysisManager.h"
24#include "llvm/Analysis/LoopInfo.h"
25#include "llvm/Analysis/LoopIterator.h"
26#include "llvm/Analysis/LoopPass.h"
27#include "llvm/Analysis/MemorySSA.h"
28#include "llvm/Analysis/MemorySSAUpdater.h"
29#include "llvm/Analysis/MustExecute.h"
30#include "llvm/Analysis/ProfileSummaryInfo.h"
31#include "llvm/Analysis/ScalarEvolution.h"
32#include "llvm/Analysis/TargetTransformInfo.h"
33#include "llvm/Analysis/ValueTracking.h"
34#include "llvm/IR/BasicBlock.h"
35#include "llvm/IR/Constant.h"
36#include "llvm/IR/Constants.h"
37#include "llvm/IR/Dominators.h"
38#include "llvm/IR/Function.h"
39#include "llvm/IR/IRBuilder.h"
40#include "llvm/IR/InstrTypes.h"
41#include "llvm/IR/Instruction.h"
42#include "llvm/IR/Instructions.h"
43#include "llvm/IR/IntrinsicInst.h"
44#include "llvm/IR/PatternMatch.h"
45#include "llvm/IR/Use.h"
46#include "llvm/IR/Value.h"
47#include "llvm/InitializePasses.h"
48#include "llvm/Pass.h"
49#include "llvm/Support/Casting.h"
50#include "llvm/Support/CommandLine.h"
51#include "llvm/Support/Debug.h"
52#include "llvm/Support/ErrorHandling.h"
53#include "llvm/Support/GenericDomTree.h"
54#include "llvm/Support/InstructionCost.h"
55#include "llvm/Support/raw_ostream.h"
56#include "llvm/Transforms/Scalar/LoopPassManager.h"
57#include "llvm/Transforms/Utils/BasicBlockUtils.h"
58#include "llvm/Transforms/Utils/Cloning.h"
59#include "llvm/Transforms/Utils/Local.h"
60#include "llvm/Transforms/Utils/LoopUtils.h"
61#include "llvm/Transforms/Utils/ValueMapper.h"
62#include <algorithm>
63#include <cassert>
64#include <iterator>
65#include <numeric>
66#include <utility>

68#define DEBUG_TYPE"simple-loop-unswitch" "simple-loop-unswitch"

70using namespace llvm;
71using namespace llvm::PatternMatch;

73STATISTIC(NumBranches, "Number of branches unswitched")static llvm::Statistic NumBranches = {"simple-loop-unswitch",
 "NumBranches", "Number of branches unswitched"};
74STATISTIC(NumSwitches, "Number of switches unswitched")static llvm::Statistic NumSwitches = {"simple-loop-unswitch",
 "NumSwitches", "Number of switches unswitched"};
75STATISTIC(NumGuards, "Number of guards turned into branches for unswitching")static llvm::Statistic NumGuards = {"simple-loop-unswitch", "NumGuards"
, "Number of guards turned into branches for unswitching"};
76STATISTIC(NumTrivial, "Number of unswitches that are trivial")static llvm::Statistic NumTrivial = {"simple-loop-unswitch", "NumTrivial"
, "Number of unswitches that are trivial"};
77STATISTIC(static llvm::Statistic NumCostMultiplierSkipped = {"simple-loop-unswitch"
, "NumCostMultiplierSkipped", "Number of unswitch candidates that had their cost multiplier skipped"
}
  NumCostMultiplierSkipped,static llvm::Statistic NumCostMultiplierSkipped = {"simple-loop-unswitch"
, "NumCostMultiplierSkipped", "Number of unswitch candidates that had their cost multiplier skipped"
}
  "Number of unswitch candidates that had their cost multiplier skipped")static llvm::Statistic NumCostMultiplierSkipped = {"simple-loop-unswitch"
, "NumCostMultiplierSkipped", "Number of unswitch candidates that had their cost multiplier skipped"
};

81static cl::opt<bool> EnableNonTrivialUnswitch(
  "enable-nontrivial-unswitch", cl::init(false), cl::Hidden,
  cl::desc("Forcibly enables non-trivial loop unswitching rather than "
           "following the configuration passed into the pass."));

86static cl::opt<int>
  UnswitchThreshold("unswitch-threshold", cl::init(50), cl::Hidden,
                    cl::desc("The cost threshold for unswitching a loop."));

90static cl::opt<bool> EnableUnswitchCostMultiplier(
  "enable-unswitch-cost-multiplier", cl::init(true), cl::Hidden,
  cl::desc("Enable unswitch cost multiplier that prohibits exponential "
           "explosion in nontrivial unswitch."));
94static cl::opt<int> UnswitchSiblingsToplevelDiv(
  "unswitch-siblings-toplevel-div", cl::init(2), cl::Hidden,
  cl::desc("Toplevel siblings divisor for cost multiplier."));
97static cl::opt<int> UnswitchNumInitialUnscaledCandidates(
  "unswitch-num-initial-unscaled-candidates", cl::init(8), cl::Hidden,
  cl::desc("Number of unswitch candidates that are ignored when calculating "
           "cost multiplier."));
101static cl::opt<bool> UnswitchGuards(
  "simple-loop-unswitch-guards", cl::init(true), cl::Hidden,
  cl::desc("If enabled, simple loop unswitching will also consider "
           "llvm.experimental.guard intrinsics as unswitch candidates."));
105static cl::opt<bool> DropNonTrivialImplicitNullChecks(
  "simple-loop-unswitch-drop-non-trivial-implicit-null-checks",
  cl::init(false), cl::Hidden,
  cl::desc("If enabled, drop make.implicit metadata in unswitched implicit "
           "null checks to save time analyzing if we can keep it."));
110static cl::opt<unsigned>
  MSSAThreshold("simple-loop-unswitch-memoryssa-threshold",
                cl::desc("Max number of memory uses to explore during "
                         "partial unswitching analysis"),
                cl::init(100), cl::Hidden);
115static cl::opt<bool> FreezeLoopUnswitchCond(
  "freeze-loop-unswitch-cond", cl::init(true), cl::Hidden,
  cl::desc("If enabled, the freeze instruction will be added to condition "
           "of loop unswitch to prevent miscompilation."));

120// Helper to skip (select x, true, false), which matches both a logical AND and
121// OR and can confuse code that tries to determine if \p Cond is either a
122// logical AND or OR but not both.
123static Value *skipTrivialSelect(Value *Cond) {
Value *CondNext;
while (match(Cond, m_Select(m_Value(CondNext), m_One(), m_Zero())))
  Cond = CondNext;
return Cond;
128}

130/// Collect all of the loop invariant input values transitively used by the
131/// homogeneous instruction graph from a given root.
132///
133/// This essentially walks from a root recursively through loop variant operands
134/// which have perform the same logical operation (AND or OR) and finds all
135/// inputs which are loop invariant. For some operations these can be
136/// re-associated and unswitched out of the loop entirely.
137static TinyPtrVector<Value *>
138collectHomogenousInstGraphLoopInvariants(Loop &L, Instruction &Root,
                                       LoopInfo &LI) {
assert(!L.isLoopInvariant(&Root) &&(static_cast <bool> (!L.isLoopInvariant(&Root) &&
 "Only need to walk the graph if root itself is not invariant."
) ? void (0) : __assert_fail ("!L.isLoopInvariant(&Root) && \"Only need to walk the graph if root itself is not invariant.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 141, __extension__
 __PRETTY_FUNCTION__))
       "Only need to walk the graph if root itself is not invariant.")(static_cast <bool> (!L.isLoopInvariant(&Root) &&
 "Only need to walk the graph if root itself is not invariant."
) ? void (0) : __assert_fail ("!L.isLoopInvariant(&Root) && \"Only need to walk the graph if root itself is not invariant.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 141, __extension__
 __PRETTY_FUNCTION__));
TinyPtrVector<Value *> Invariants;

bool IsRootAnd = match(&Root, m_LogicalAnd());
bool IsRootOr  = match(&Root, m_LogicalOr());

// Build a worklist and recurse through operators collecting invariants.
SmallVector<Instruction *, 4> Worklist;
SmallPtrSet<Instruction *, 8> Visited;
Worklist.push_back(&Root);
Visited.insert(&Root);
do {
  Instruction &I = *Worklist.pop_back_val();
  for (Value *OpV : I.operand_values()) {
    // Skip constants as unswitching isn't interesting for them.
    if (isa<Constant>(OpV))
      continue;

    // Add it to our result if loop invariant.
    if (L.isLoopInvariant(OpV)) {
      Invariants.push_back(OpV);
      continue;
    }

    // If not an instruction with the same opcode, nothing we can do.
    Instruction *OpI = dyn_cast<Instruction>(skipTrivialSelect(OpV));

    if (OpI && ((IsRootAnd && match(OpI, m_LogicalAnd())) ||
                (IsRootOr  && match(OpI, m_LogicalOr())))) {
      // Visit this operand.
      if (Visited.insert(OpI).second)
        Worklist.push_back(OpI);
    }
  }
} while (!Worklist.empty());

return Invariants;
178}

180static void replaceLoopInvariantUses(Loop &L, Value *Invariant,
                                   Constant &Replacement) {
assert(!isa<Constant>(Invariant) && "Why are we unswitching on a constant?")(static_cast <bool> (!isa<Constant>(Invariant) &&
 "Why are we unswitching on a constant?") ? void (0) : __assert_fail
 ("!isa<Constant>(Invariant) && \"Why are we unswitching on a constant?\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 182, __extension__
 __PRETTY_FUNCTION__));

// Replace uses of LIC in the loop with the given constant.
// We use make_early_inc_range as set invalidates the iterator.
for (Use &U : llvm::make_early_inc_range(Invariant->uses())) {
  Instruction *UserI = dyn_cast<Instruction>(U.getUser());

  // Replace this use within the loop body.
  if (UserI && L.contains(UserI))
    U.set(&Replacement);
}
193}

195/// Check that all the LCSSA PHI nodes in the loop exit block have trivial
196/// incoming values along this edge.
197static bool areLoopExitPHIsLoopInvariant(Loop &L, BasicBlock &ExitingBB,
                                       BasicBlock &ExitBB) {
for (Instruction &I : ExitBB) {
  auto *PN = dyn_cast<PHINode>(&I);
  if (!PN)
    // No more PHIs to check.
    return true;

  // If the incoming value for this edge isn't loop invariant the unswitch
  // won't be trivial.
  if (!L.isLoopInvariant(PN->getIncomingValueForBlock(&ExitingBB)))
    return false;
}
llvm_unreachable("Basic blocks should never be empty!")::llvm::llvm_unreachable_internal("Basic blocks should never be empty!"
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 210);
211}

213/// Copy a set of loop invariant values \p ToDuplicate and insert them at the
214/// end of \p BB and conditionally branch on the copied condition. We only
215/// branch on a single value.
216static void buildPartialUnswitchConditionalBranch(
  BasicBlock &BB, ArrayRef<Value *> Invariants, bool Direction,
  BasicBlock &UnswitchedSucc, BasicBlock &NormalSucc, bool InsertFreeze,
  Instruction *I, AssumptionCache *AC, DominatorTree &DT) {
IRBuilder<> IRB(&BB);

SmallVector<Value *> FrozenInvariants;
for (Value *Inv : Invariants) {
  if (InsertFreeze && !isGuaranteedNotToBeUndefOrPoison(Inv, AC, I, &DT))
    Inv = IRB.CreateFreeze(Inv, Inv->getName() + ".fr");
  FrozenInvariants.push_back(Inv);
}

Value *Cond = Direction ? IRB.CreateOr(FrozenInvariants)
                        : IRB.CreateAnd(FrozenInvariants);
IRB.CreateCondBr(Cond, Direction ? &UnswitchedSucc : &NormalSucc,
                 Direction ? &NormalSucc : &UnswitchedSucc);
233}

235/// Copy a set of loop invariant values, and conditionally branch on them.
236static void buildPartialInvariantUnswitchConditionalBranch(
  BasicBlock &BB, ArrayRef<Value *> ToDuplicate, bool Direction,
  BasicBlock &UnswitchedSucc, BasicBlock &NormalSucc, Loop &L,
  MemorySSAUpdater *MSSAU) {
ValueToValueMapTy VMap;
for (auto *Val : reverse(ToDuplicate)) {
  Instruction *Inst = cast<Instruction>(Val);
  Instruction *NewInst = Inst->clone();
  BB.getInstList().insert(BB.end(), NewInst);
  RemapInstruction(NewInst, VMap,
                   RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
  VMap[Val] = NewInst;

  if (!MSSAU)
    continue;

  MemorySSA *MSSA = MSSAU->getMemorySSA();
  if (auto *MemUse =
          dyn_cast_or_null<MemoryUse>(MSSA->getMemoryAccess(Inst))) {
    auto *DefiningAccess = MemUse->getDefiningAccess();
    // Get the first defining access before the loop.
    while (L.contains(DefiningAccess->getBlock())) {
      // If the defining access is a MemoryPhi, get the incoming
      // value for the pre-header as defining access.
      if (auto *MemPhi = dyn_cast<MemoryPhi>(DefiningAccess))
        DefiningAccess =
            MemPhi->getIncomingValueForBlock(L.getLoopPreheader());
      else
        DefiningAccess = cast<MemoryDef>(DefiningAccess)->getDefiningAccess();
    }
    MSSAU->createMemoryAccessInBB(NewInst, DefiningAccess,
                                  NewInst->getParent(),
                                  MemorySSA::BeforeTerminator);
  }
}

IRBuilder<> IRB(&BB);
Value *Cond = VMap[ToDuplicate[0]];
IRB.CreateCondBr(Cond, Direction ? &UnswitchedSucc : &NormalSucc,
                 Direction ? &NormalSucc : &UnswitchedSucc);
276}

278/// Rewrite the PHI nodes in an unswitched loop exit basic block.
279///
280/// Requires that the loop exit and unswitched basic block are the same, and
281/// that the exiting block was a unique predecessor of that block. Rewrites the
282/// PHI nodes in that block such that what were LCSSA PHI nodes become trivial
283/// PHI nodes from the old preheader that now contains the unswitched
284/// terminator.
285static void rewritePHINodesForUnswitchedExitBlock(BasicBlock &UnswitchedBB,
                                                BasicBlock &OldExitingBB,
                                                BasicBlock &OldPH) {
for (PHINode &PN : UnswitchedBB.phis()) {
  // When the loop exit is directly unswitched we just need to update the
  // incoming basic block. We loop to handle weird cases with repeated
  // incoming blocks, but expect to typically only have one operand here.
  for (auto i : seq<int>(0, PN.getNumOperands())) {
    assert(PN.getIncomingBlock(i) == &OldExitingBB &&(static_cast <bool> (PN.getIncomingBlock(i) == &OldExitingBB
 && "Found incoming block different from unique predecessor!"
) ? void (0) : __assert_fail ("PN.getIncomingBlock(i) == &OldExitingBB && \"Found incoming block different from unique predecessor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 294, __extension__
 __PRETTY_FUNCTION__))
           "Found incoming block different from unique predecessor!")(static_cast <bool> (PN.getIncomingBlock(i) == &OldExitingBB
 && "Found incoming block different from unique predecessor!"
) ? void (0) : __assert_fail ("PN.getIncomingBlock(i) == &OldExitingBB && \"Found incoming block different from unique predecessor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 294, __extension__
 __PRETTY_FUNCTION__));
    PN.setIncomingBlock(i, &OldPH);
  }
}
298}

300/// Rewrite the PHI nodes in the loop exit basic block and the split off
301/// unswitched block.
302///
303/// Because the exit block remains an exit from the loop, this rewrites the
304/// LCSSA PHI nodes in it to remove the unswitched edge and introduces PHI
305/// nodes into the unswitched basic block to select between the value in the
306/// old preheader and the loop exit.
307static void rewritePHINodesForExitAndUnswitchedBlocks(BasicBlock &ExitBB,
                                                    BasicBlock &UnswitchedBB,
                                                    BasicBlock &OldExitingBB,
                                                    BasicBlock &OldPH,
                                                    bool FullUnswitch) {
assert(&ExitBB != &UnswitchedBB &&(static_cast <bool> (&ExitBB != &UnswitchedBB &&
 "Must have different loop exit and unswitched blocks!") ? void
 (0) : __assert_fail ("&ExitBB != &UnswitchedBB && \"Must have different loop exit and unswitched blocks!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 313, __extension__
 __PRETTY_FUNCTION__))
       "Must have different loop exit and unswitched blocks!")(static_cast <bool> (&ExitBB != &UnswitchedBB &&
 "Must have different loop exit and unswitched blocks!") ? void
 (0) : __assert_fail ("&ExitBB != &UnswitchedBB && \"Must have different loop exit and unswitched blocks!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 313, __extension__
 __PRETTY_FUNCTION__));
Instruction *InsertPt = &*UnswitchedBB.begin();
for (PHINode &PN : ExitBB.phis()) {
  auto *NewPN = PHINode::Create(PN.getType(), /*NumReservedValues*/ 2,
                                PN.getName() + ".split", InsertPt);

  // Walk backwards over the old PHI node's inputs to minimize the cost of
  // removing each one. We have to do this weird loop manually so that we
  // create the same number of new incoming edges in the new PHI as we expect
  // each case-based edge to be included in the unswitched switch in some
  // cases.
  // FIXME: This is really, really gross. It would be much cleaner if LLVM
  // allowed us to create a single entry for a predecessor block without
  // having separate entries for each "edge" even though these edges are
  // required to produce identical results.
  for (int i = PN.getNumIncomingValues() - 1; i >= 0; --i) {
    if (PN.getIncomingBlock(i) != &OldExitingBB)
      continue;

    Value *Incoming = PN.getIncomingValue(i);
    if (FullUnswitch)
      // No more edge from the old exiting block to the exit block.
      PN.removeIncomingValue(i);

    NewPN->addIncoming(Incoming, &OldPH);
  }

  // Now replace the old PHI with the new one and wire the old one in as an
  // input to the new one.
  PN.replaceAllUsesWith(NewPN);
  NewPN->addIncoming(&PN, &ExitBB);
}
345}

347/// Hoist the current loop up to the innermost loop containing a remaining exit.
348///
349/// Because we've removed an exit from the loop, we may have changed the set of
350/// loops reachable and need to move the current loop up the loop nest or even
351/// to an entirely separate nest.
352static void hoistLoopToNewParent(Loop &L, BasicBlock &Preheader,
                               DominatorTree &DT, LoopInfo &LI,
                               MemorySSAUpdater *MSSAU, ScalarEvolution *SE) {
// If the loop is already at the top level, we can't hoist it anywhere.
Loop *OldParentL = L.getParentLoop();
if (!OldParentL)
  return;

SmallVector<BasicBlock *, 4> Exits;
L.getExitBlocks(Exits);
Loop *NewParentL = nullptr;
for (auto *ExitBB : Exits)
  if (Loop *ExitL = LI.getLoopFor(ExitBB))
    if (!NewParentL || NewParentL->contains(ExitL))
      NewParentL = ExitL;

if (NewParentL == OldParentL)
  return;

// The new parent loop (if different) should always contain the old one.
if (NewParentL)
  assert(NewParentL->contains(OldParentL) &&(static_cast <bool> (NewParentL->contains(OldParentL
) && "Can only hoist this loop up the nest!") ? void (
0) : __assert_fail ("NewParentL->contains(OldParentL) && \"Can only hoist this loop up the nest!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 374, __extension__
 __PRETTY_FUNCTION__))
         "Can only hoist this loop up the nest!")(static_cast <bool> (NewParentL->contains(OldParentL
) && "Can only hoist this loop up the nest!") ? void (
0) : __assert_fail ("NewParentL->contains(OldParentL) && \"Can only hoist this loop up the nest!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 374, __extension__
 __PRETTY_FUNCTION__));

// The preheader will need to move with the body of this loop. However,
// because it isn't in this loop we also need to update the primary loop map.
assert(OldParentL == LI.getLoopFor(&Preheader) &&(static_cast <bool> (OldParentL == LI.getLoopFor(&Preheader
) && "Parent loop of this loop should contain this loop's preheader!"
) ? void (0) : __assert_fail ("OldParentL == LI.getLoopFor(&Preheader) && \"Parent loop of this loop should contain this loop's preheader!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 379, __extension__
 __PRETTY_FUNCTION__))
       "Parent loop of this loop should contain this loop's preheader!")(static_cast <bool> (OldParentL == LI.getLoopFor(&Preheader
) && "Parent loop of this loop should contain this loop's preheader!"
) ? void (0) : __assert_fail ("OldParentL == LI.getLoopFor(&Preheader) && \"Parent loop of this loop should contain this loop's preheader!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 379, __extension__
 __PRETTY_FUNCTION__));
LI.changeLoopFor(&Preheader, NewParentL);

// Remove this loop from its old parent.
OldParentL->removeChildLoop(&L);

// Add the loop either to the new parent or as a top-level loop.
if (NewParentL)
  NewParentL->addChildLoop(&L);
else
  LI.addTopLevelLoop(&L);

// Remove this loops blocks from the old parent and every other loop up the
// nest until reaching the new parent. Also update all of these
// no-longer-containing loops to reflect the nesting change.
for (Loop *OldContainingL = OldParentL; OldContainingL != NewParentL;
     OldContainingL = OldContainingL->getParentLoop()) {
  llvm::erase_if(OldContainingL->getBlocksVector(),
                 [&](const BasicBlock *BB) {
                   return BB == &Preheader || L.contains(BB);
                 });

  OldContainingL->getBlocksSet().erase(&Preheader);
  for (BasicBlock *BB : L.blocks())
    OldContainingL->getBlocksSet().erase(BB);

  // Because we just hoisted a loop out of this one, we have essentially
  // created new exit paths from it. That means we need to form LCSSA PHI
  // nodes for values used in the no-longer-nested loop.
  formLCSSA(*OldContainingL, DT, &LI, SE);

  // We shouldn't need to form dedicated exits because the exit introduced
  // here is the (just split by unswitching) preheader. However, after trivial
  // unswitching it is possible to get new non-dedicated exits out of parent
  // loop so let's conservatively form dedicated exit blocks and figure out
  // if we can optimize later.
  formDedicatedExitBlocks(OldContainingL, &DT, &LI, MSSAU,
                          /*PreserveLCSSA*/ true);
}
418}

420// Return the top-most loop containing ExitBB and having ExitBB as exiting block
421// or the loop containing ExitBB, if there is no parent loop containing ExitBB
422// as exiting block.
423static Loop *getTopMostExitingLoop(BasicBlock *ExitBB, LoopInfo &LI) {
Loop *TopMost = LI.getLoopFor(ExitBB);
Loop *Current = TopMost;
while (Current) {
  if (Current->isLoopExiting(ExitBB))
    TopMost = Current;
  Current = Current->getParentLoop();
}
return TopMost;
432}

434/// Unswitch a trivial branch if the condition is loop invariant.
435///
436/// This routine should only be called when loop code leading to the branch has
437/// been validated as trivial (no side effects). This routine checks if the
438/// condition is invariant and one of the successors is a loop exit. This
439/// allows us to unswitch without duplicating the loop, making it trivial.
440///
441/// If this routine fails to unswitch the branch it returns false.
442///
443/// If the branch can be unswitched, this routine splits the preheader and
444/// hoists the branch above that split. Preserves loop simplified form
445/// (splitting the exit block as necessary). It simplifies the branch within
446/// the loop to an unconditional branch but doesn't remove it entirely. Further
447/// cleanup can be done with some simplifycfg like pass.
448///
449/// If `SE` is not null, it will be updated based on the potential loop SCEVs
450/// invalidated by this.
451static bool unswitchTrivialBranch(Loop &L, BranchInst &BI, DominatorTree &DT,
                                LoopInfo &LI, ScalarEvolution *SE,
                                MemorySSAUpdater *MSSAU) {
assert(BI.isConditional() && "Can only unswitch a conditional branch!")(static_cast <bool> (BI.isConditional() && "Can only unswitch a conditional branch!"
) ? void (0) : __assert_fail ("BI.isConditional() && \"Can only unswitch a conditional branch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 454, __extension__
 __PRETTY_FUNCTION__));
LLVM_DEBUG(dbgs() << "  Trying to unswitch branch: " << BI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Trying to unswitch branch: "
 << BI << "\n"; } } while (false);

// The loop invariant values that we want to unswitch.
TinyPtrVector<Value *> Invariants;

// When true, we're fully unswitching the branch rather than just unswitching
// some input conditions to the branch.
bool FullUnswitch = false;

Value *Cond = skipTrivialSelect(BI.getCondition());
if (L.isLoopInvariant(Cond)) {
  Invariants.push_back(Cond);
  FullUnswitch = true;
} else {
  if (auto *CondInst = dyn_cast<Instruction>(Cond))
    Invariants = collectHomogenousInstGraphLoopInvariants(L, *CondInst, LI);
  if (Invariants.empty()) {
    LLVM_DEBUG(dbgs() << "   Couldn't find invariant inputs!\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "   Couldn't find invariant inputs!\n"
; } } while (false);
    return false;
  }
}

// Check that one of the branch's successors exits, and which one.
bool ExitDirection = true;
int LoopExitSuccIdx = 0;
auto *LoopExitBB = BI.getSuccessor(0);
if (L.contains(LoopExitBB)) {
  ExitDirection = false;
  LoopExitSuccIdx = 1;
  LoopExitBB = BI.getSuccessor(1);
  if (L.contains(LoopExitBB)) {
    LLVM_DEBUG(dbgs() << "   Branch doesn't exit the loop!\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "   Branch doesn't exit the loop!\n"
; } } while (false);
    return false;
  }
}
auto *ContinueBB = BI.getSuccessor(1 - LoopExitSuccIdx);
auto *ParentBB = BI.getParent();
if (!areLoopExitPHIsLoopInvariant(L, *ParentBB, *LoopExitBB)) {
  LLVM_DEBUG(dbgs() << "   Loop exit PHI's aren't loop-invariant!\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "   Loop exit PHI's aren't loop-invariant!\n"
; } } while (false);
  return false;
}

// When unswitching only part of the branch's condition, we need the exit
// block to be reached directly from the partially unswitched input. This can
// be done when the exit block is along the true edge and the branch condition
// is a graph of `or` operations, or the exit block is along the false edge
// and the condition is a graph of `and` operations.
if (!FullUnswitch) {
  if (ExitDirection ? !match(Cond, m_LogicalOr())
                    : !match(Cond, m_LogicalAnd())) {
    LLVM_DEBUG(dbgs() << "   Branch condition is in improper form for "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "   Branch condition is in improper form for "
 "non-full unswitch!\n"; } } while (false)
                         "non-full unswitch!\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "   Branch condition is in improper form for "
 "non-full unswitch!\n"; } } while (false);
    return false;
  }
}

LLVM_DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false)
  dbgs() << "    unswitching trivial invariant conditions for: " << BIdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false)
         << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false)
  for (Value *Invariant : Invariants) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false)
    dbgs() << "      " << *Invariant << " == true";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false)
    if (Invariant != Invariants.back())do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false)
      dbgs() << " ||";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false)
    dbgs() << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false)
  }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false)
})do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { { dbgs() << "    unswitching trivial invariant conditions for: "
 << BI << "\n"; for (Value *Invariant : Invariants
) { dbgs() << "      " << *Invariant << " == true"
; if (Invariant != Invariants.back()) dbgs() << " ||"; dbgs
() << "\n"; } }; } } while (false);

// If we have scalar evolutions, we need to invalidate them including this
// loop, the loop containing the exit block and the topmost parent loop
// exiting via LoopExitBB.
if (SE) {
  if (Loop *ExitL = getTopMostExitingLoop(LoopExitBB, LI))
    SE->forgetLoop(ExitL);
  else
    // Forget the entire nest as this exits the entire nest.
    SE->forgetTopmostLoop(&L);
}

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

// Split the preheader, so that we know that there is a safe place to insert
// the conditional branch. We will change the preheader to have a conditional
// branch on LoopCond.
BasicBlock *OldPH = L.getLoopPreheader();
BasicBlock *NewPH = SplitEdge(OldPH, L.getHeader(), &DT, &LI, MSSAU);

// Now that we have a place to insert the conditional branch, create a place
// to branch to: this is the exit block out of the loop that we are
// unswitching. We need to split this if there are other loop predecessors.
// Because the loop is in simplified form, *any* other predecessor is enough.
BasicBlock *UnswitchedBB;
if (FullUnswitch && LoopExitBB->getUniquePredecessor()) {
  assert(LoopExitBB->getUniquePredecessor() == BI.getParent() &&(static_cast <bool> (LoopExitBB->getUniquePredecessor
() == BI.getParent() && "A branch's parent isn't a predecessor!"
) ? void (0) : __assert_fail ("LoopExitBB->getUniquePredecessor() == BI.getParent() && \"A branch's parent isn't a predecessor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 549, __extension__
 __PRETTY_FUNCTION__))
         "A branch's parent isn't a predecessor!")(static_cast <bool> (LoopExitBB->getUniquePredecessor
() == BI.getParent() && "A branch's parent isn't a predecessor!"
) ? void (0) : __assert_fail ("LoopExitBB->getUniquePredecessor() == BI.getParent() && \"A branch's parent isn't a predecessor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 549, __extension__
 __PRETTY_FUNCTION__));
  UnswitchedBB = LoopExitBB;
} else {
  UnswitchedBB =
      SplitBlock(LoopExitBB, &LoopExitBB->front(), &DT, &LI, MSSAU);
}

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

// Actually move the invariant uses into the unswitched position. If possible,
// we do this by moving the instructions, but when doing partial unswitching
// we do it by building a new merge of the values in the unswitched position.
OldPH->getTerminator()->eraseFromParent();
if (FullUnswitch) {
  // If fully unswitching, we can use the existing branch instruction.
  // Splice it into the old PH to gate reaching the new preheader and re-point
  // its successors.
  OldPH->getInstList().splice(OldPH->end(), BI.getParent()->getInstList(),
                              BI);
  BI.setCondition(Cond);
  if (MSSAU) {
    // Temporarily clone the terminator, to make MSSA update cheaper by
    // separating "insert edge" updates from "remove edge" ones.
    ParentBB->getInstList().push_back(BI.clone());
  } else {
    // Create a new unconditional branch that will continue the loop as a new
    // terminator.
    BranchInst::Create(ContinueBB, ParentBB);
  }
  BI.setSuccessor(LoopExitSuccIdx, UnswitchedBB);
  BI.setSuccessor(1 - LoopExitSuccIdx, NewPH);
} else {
  // Only unswitching a subset of inputs to the condition, so we will need to
  // build a new branch that merges the invariant inputs.
  if (ExitDirection)
    assert(match(skipTrivialSelect(BI.getCondition()), m_LogicalOr()) &&(static_cast <bool> (match(skipTrivialSelect(BI.getCondition
()), m_LogicalOr()) && "Must have an `or` of `i1`s or `select i1 X, true, Y`s for the "
 "condition!") ? void (0) : __assert_fail ("match(skipTrivialSelect(BI.getCondition()), m_LogicalOr()) && \"Must have an `or` of `i1`s or `select i1 X, true, Y`s for the \" \"condition!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 587, __extension__
 __PRETTY_FUNCTION__))
           "Must have an `or` of `i1`s or `select i1 X, true, Y`s for the "(static_cast <bool> (match(skipTrivialSelect(BI.getCondition
()), m_LogicalOr()) && "Must have an `or` of `i1`s or `select i1 X, true, Y`s for the "
 "condition!") ? void (0) : __assert_fail ("match(skipTrivialSelect(BI.getCondition()), m_LogicalOr()) && \"Must have an `or` of `i1`s or `select i1 X, true, Y`s for the \" \"condition!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 587, __extension__
 __PRETTY_FUNCTION__))
           "condition!")(static_cast <bool> (match(skipTrivialSelect(BI.getCondition
()), m_LogicalOr()) && "Must have an `or` of `i1`s or `select i1 X, true, Y`s for the "
 "condition!") ? void (0) : __assert_fail ("match(skipTrivialSelect(BI.getCondition()), m_LogicalOr()) && \"Must have an `or` of `i1`s or `select i1 X, true, Y`s for the \" \"condition!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 587, __extension__
 __PRETTY_FUNCTION__));
  else
    assert(match(skipTrivialSelect(BI.getCondition()), m_LogicalAnd()) &&(static_cast <bool> (match(skipTrivialSelect(BI.getCondition
()), m_LogicalAnd()) && "Must have an `and` of `i1`s or `select i1 X, Y, false`s for the"
 " condition!") ? void (0) : __assert_fail ("match(skipTrivialSelect(BI.getCondition()), m_LogicalAnd()) && \"Must have an `and` of `i1`s or `select i1 X, Y, false`s for the\" \" condition!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 591, __extension__
 __PRETTY_FUNCTION__))
           "Must have an `and` of `i1`s or `select i1 X, Y, false`s for the"(static_cast <bool> (match(skipTrivialSelect(BI.getCondition
()), m_LogicalAnd()) && "Must have an `and` of `i1`s or `select i1 X, Y, false`s for the"
 " condition!") ? void (0) : __assert_fail ("match(skipTrivialSelect(BI.getCondition()), m_LogicalAnd()) && \"Must have an `and` of `i1`s or `select i1 X, Y, false`s for the\" \" condition!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 591, __extension__
 __PRETTY_FUNCTION__))
           " condition!")(static_cast <bool> (match(skipTrivialSelect(BI.getCondition
()), m_LogicalAnd()) && "Must have an `and` of `i1`s or `select i1 X, Y, false`s for the"
 " condition!") ? void (0) : __assert_fail ("match(skipTrivialSelect(BI.getCondition()), m_LogicalAnd()) && \"Must have an `and` of `i1`s or `select i1 X, Y, false`s for the\" \" condition!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 591, __extension__
 __PRETTY_FUNCTION__));
  buildPartialUnswitchConditionalBranch(
      *OldPH, Invariants, ExitDirection, *UnswitchedBB, *NewPH,
      FreezeLoopUnswitchCond, OldPH->getTerminator(), nullptr, DT);
}

// Update the dominator tree with the added edge.
DT.insertEdge(OldPH, UnswitchedBB);

// After the dominator tree was updated with the added edge, update MemorySSA
// if available.
if (MSSAU) {
  SmallVector<CFGUpdate, 1> Updates;
  Updates.push_back({cfg::UpdateKind::Insert, OldPH, UnswitchedBB});
  MSSAU->applyInsertUpdates(Updates, DT);
}

// Finish updating dominator tree and memory ssa for full unswitch.
if (FullUnswitch) {
  if (MSSAU) {
    // Remove the cloned branch instruction.
    ParentBB->getTerminator()->eraseFromParent();
    // Create unconditional branch now.
    BranchInst::Create(ContinueBB, ParentBB);
    MSSAU->removeEdge(ParentBB, LoopExitBB);
  }
  DT.deleteEdge(ParentBB, LoopExitBB);
}

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

// Rewrite the relevant PHI nodes.
if (UnswitchedBB == LoopExitBB)
  rewritePHINodesForUnswitchedExitBlock(*UnswitchedBB, *ParentBB, *OldPH);
else
  rewritePHINodesForExitAndUnswitchedBlocks(*LoopExitBB, *UnswitchedBB,
                                            *ParentBB, *OldPH, FullUnswitch);

// The constant we can replace all of our invariants with inside the loop
// body. If any of the invariants have a value other than this the loop won't
// be entered.
ConstantInt *Replacement = ExitDirection
                               ? ConstantInt::getFalse(BI.getContext())
                               : ConstantInt::getTrue(BI.getContext());

// Since this is an i1 condition we can also trivially replace uses of it
// within the loop with a constant.
for (Value *Invariant : Invariants)
  replaceLoopInvariantUses(L, Invariant, *Replacement);

// If this was full unswitching, we may have changed the nesting relationship
// for this loop so hoist it to its correct parent if needed.
if (FullUnswitch)
  hoistLoopToNewParent(L, *NewPH, DT, LI, MSSAU, SE);

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

LLVM_DEBUG(dbgs() << "    done: unswitching trivial branch...\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "    done: unswitching trivial branch...\n"
; } } while (false);
++NumTrivial;
++NumBranches;
return true;
654}

656/// Unswitch a trivial switch if the condition is loop invariant.
657///
658/// This routine should only be called when loop code leading to the switch has
659/// been validated as trivial (no side effects). This routine checks if the
660/// condition is invariant and that at least one of the successors is a loop
661/// exit. This allows us to unswitch without duplicating the loop, making it
662/// trivial.
663///
664/// If this routine fails to unswitch the switch it returns false.
665///
666/// If the switch can be unswitched, this routine splits the preheader and
667/// copies the switch above that split. If the default case is one of the
668/// exiting cases, it copies the non-exiting cases and points them at the new
669/// preheader. If the default case is not exiting, it copies the exiting cases
670/// and points the default at the preheader. It preserves loop simplified form
671/// (splitting the exit blocks as necessary). It simplifies the switch within
672/// the loop by removing now-dead cases. If the default case is one of those
673/// unswitched, it replaces its destination with a new basic block containing
674/// only unreachable. Such basic blocks, while technically loop exits, are not
675/// considered for unswitching so this is a stable transform and the same
676/// switch will not be revisited. If after unswitching there is only a single
677/// in-loop successor, the switch is further simplified to an unconditional
678/// branch. Still more cleanup can be done with some simplifycfg like pass.
679///
680/// If `SE` is not null, it will be updated based on the potential loop SCEVs
681/// invalidated by this.
682static bool unswitchTrivialSwitch(Loop &L, SwitchInst &SI, DominatorTree &DT,
                                LoopInfo &LI, ScalarEvolution *SE,
                                MemorySSAUpdater *MSSAU) {
LLVM_DEBUG(dbgs() << "  Trying to unswitch switch: " << SI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Trying to unswitch switch: "
 << SI << "\n"; } } while (false);
Value *LoopCond = SI.getCondition();

// If this isn't switching on an invariant condition, we can't unswitch it.
if (!L.isLoopInvariant(LoopCond))
  return false;

auto *ParentBB = SI.getParent();

// The same check must be used both for the default and the exit cases. We
// should never leave edges from the switch instruction to a basic block that
// we are unswitching, hence the condition used to determine the default case
// needs to also be used to populate ExitCaseIndices, which is then used to
// remove cases from the switch.
auto IsTriviallyUnswitchableExitBlock = [&](BasicBlock &BBToCheck) {
  // BBToCheck is not an exit block if it is inside loop L.
  if (L.contains(&BBToCheck))
    return false;
  // BBToCheck is not trivial to unswitch if its phis aren't loop invariant.
  if (!areLoopExitPHIsLoopInvariant(L, *ParentBB, BBToCheck))
    return false;
  // We do not unswitch a block that only has an unreachable statement, as
  // it's possible this is a previously unswitched block. Only unswitch if
  // either the terminator is not unreachable, or, if it is, it's not the only
  // instruction in the block.
  auto *TI = BBToCheck.getTerminator();
  bool isUnreachable = isa<UnreachableInst>(TI);
  return !isUnreachable ||
         (isUnreachable && (BBToCheck.getFirstNonPHIOrDbg() != TI));
};

SmallVector<int, 4> ExitCaseIndices;
for (auto Case : SI.cases())
  if (IsTriviallyUnswitchableExitBlock(*Case.getCaseSuccessor()))
    ExitCaseIndices.push_back(Case.getCaseIndex());
BasicBlock *DefaultExitBB = nullptr;
SwitchInstProfUpdateWrapper::CaseWeightOpt DefaultCaseWeight =
    SwitchInstProfUpdateWrapper::getSuccessorWeight(SI, 0);
if (IsTriviallyUnswitchableExitBlock(*SI.getDefaultDest())) {
  DefaultExitBB = SI.getDefaultDest();
} else if (ExitCaseIndices.empty())
  return false;

LLVM_DEBUG(dbgs() << "    unswitching trivial switch...\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "    unswitching trivial switch...\n"
; } } while (false);

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

// We may need to invalidate SCEVs for the outermost loop reached by any of
// the exits.
Loop *OuterL = &L;

if (DefaultExitBB) {
  // Clear out the default destination temporarily to allow accurate
  // predecessor lists to be examined below.
  SI.setDefaultDest(nullptr);
  // Check the loop containing this exit.
  Loop *ExitL = LI.getLoopFor(DefaultExitBB);
  if (!ExitL || ExitL->contains(OuterL))
    OuterL = ExitL;
}

// Store the exit cases into a separate data structure and remove them from
// the switch.
SmallVector<std::tuple<ConstantInt *, BasicBlock *,
                       SwitchInstProfUpdateWrapper::CaseWeightOpt>,
            4> ExitCases;
ExitCases.reserve(ExitCaseIndices.size());
SwitchInstProfUpdateWrapper SIW(SI);
// We walk the case indices backwards so that we remove the last case first
// and don't disrupt the earlier indices.
for (unsigned Index : reverse(ExitCaseIndices)) {
  auto CaseI = SI.case_begin() + Index;
  // Compute the outer loop from this exit.
  Loop *ExitL = LI.getLoopFor(CaseI->getCaseSuccessor());
  if (!ExitL || ExitL->contains(OuterL))
    OuterL = ExitL;
  // Save the value of this case.
  auto W = SIW.getSuccessorWeight(CaseI->getSuccessorIndex());
  ExitCases.emplace_back(CaseI->getCaseValue(), CaseI->getCaseSuccessor(), W);
  // Delete the unswitched cases.
  SIW.removeCase(CaseI);
}

if (SE) {
  if (OuterL)
    SE->forgetLoop(OuterL);
  else
    SE->forgetTopmostLoop(&L);
}

// Check if after this all of the remaining cases point at the same
// successor.
BasicBlock *CommonSuccBB = nullptr;
if (SI.getNumCases() > 0 &&
    all_of(drop_begin(SI.cases()), [&SI](const SwitchInst::CaseHandle &Case) {
      return Case.getCaseSuccessor() == SI.case_begin()->getCaseSuccessor();
    }))
  CommonSuccBB = SI.case_begin()->getCaseSuccessor();
if (!DefaultExitBB) {
  // If we're not unswitching the default, we need it to match any cases to
  // have a common successor or if we have no cases it is the common
  // successor.
  if (SI.getNumCases() == 0)
    CommonSuccBB = SI.getDefaultDest();
  else if (SI.getDefaultDest() != CommonSuccBB)
    CommonSuccBB = nullptr;
}

// Split the preheader, so that we know that there is a safe place to insert
// the switch.
BasicBlock *OldPH = L.getLoopPreheader();
BasicBlock *NewPH = SplitEdge(OldPH, L.getHeader(), &DT, &LI, MSSAU);
OldPH->getTerminator()->eraseFromParent();

// Now add the unswitched switch.
auto *NewSI = SwitchInst::Create(LoopCond, NewPH, ExitCases.size(), OldPH);
SwitchInstProfUpdateWrapper NewSIW(*NewSI);

// Rewrite the IR for the unswitched basic blocks. This requires two steps.
// First, we split any exit blocks with remaining in-loop predecessors. Then
// we update the PHIs in one of two ways depending on if there was a split.
// We walk in reverse so that we split in the same order as the cases
// appeared. This is purely for convenience of reading the resulting IR, but
// it doesn't cost anything really.
SmallPtrSet<BasicBlock *, 2> UnswitchedExitBBs;
SmallDenseMap<BasicBlock *, BasicBlock *, 2> SplitExitBBMap;
// Handle the default exit if necessary.
// FIXME: It'd be great if we could merge this with the loop below but LLVM's
// ranges aren't quite powerful enough yet.
if (DefaultExitBB) {
  if (pred_empty(DefaultExitBB)) {
    UnswitchedExitBBs.insert(DefaultExitBB);
    rewritePHINodesForUnswitchedExitBlock(*DefaultExitBB, *ParentBB, *OldPH);
  } else {
    auto *SplitBB =
        SplitBlock(DefaultExitBB, &DefaultExitBB->front(), &DT, &LI, MSSAU);
    rewritePHINodesForExitAndUnswitchedBlocks(*DefaultExitBB, *SplitBB,
                                              *ParentBB, *OldPH,
                                              /*FullUnswitch*/ true);
    DefaultExitBB = SplitExitBBMap[DefaultExitBB] = SplitBB;
  }
}
// Note that we must use a reference in the for loop so that we update the
// container.
for (auto &ExitCase : reverse(ExitCases)) {
  // Grab a reference to the exit block in the pair so that we can update it.
  BasicBlock *ExitBB = std::get<1>(ExitCase);

  // If this case is the last edge into the exit block, we can simply reuse it
  // as it will no longer be a loop exit. No mapping necessary.
  if (pred_empty(ExitBB)) {
    // Only rewrite once.
    if (UnswitchedExitBBs.insert(ExitBB).second)
      rewritePHINodesForUnswitchedExitBlock(*ExitBB, *ParentBB, *OldPH);
    continue;
  }

  // Otherwise we need to split the exit block so that we retain an exit
  // block from the loop and a target for the unswitched condition.
  BasicBlock *&SplitExitBB = SplitExitBBMap[ExitBB];
  if (!SplitExitBB) {
    // If this is the first time we see this, do the split and remember it.
    SplitExitBB = SplitBlock(ExitBB, &ExitBB->front(), &DT, &LI, MSSAU);
    rewritePHINodesForExitAndUnswitchedBlocks(*ExitBB, *SplitExitBB,
                                              *ParentBB, *OldPH,
                                              /*FullUnswitch*/ true);
  }
  // Update the case pair to point to the split block.
  std::get<1>(ExitCase) = SplitExitBB;
}

// Now add the unswitched cases. We do this in reverse order as we built them
// in reverse order.
for (auto &ExitCase : reverse(ExitCases)) {
  ConstantInt *CaseVal = std::get<0>(ExitCase);
  BasicBlock *UnswitchedBB = std::get<1>(ExitCase);

  NewSIW.addCase(CaseVal, UnswitchedBB, std::get<2>(ExitCase));
}

// If the default was unswitched, re-point it and add explicit cases for
// entering the loop.
if (DefaultExitBB) {
  NewSIW->setDefaultDest(DefaultExitBB);
  NewSIW.setSuccessorWeight(0, DefaultCaseWeight);

  // We removed all the exit cases, so we just copy the cases to the
  // unswitched switch.
  for (const auto &Case : SI.cases())
    NewSIW.addCase(Case.getCaseValue(), NewPH,
                   SIW.getSuccessorWeight(Case.getSuccessorIndex()));
} else if (DefaultCaseWeight) {
  // We have to set branch weight of the default case.
  uint64_t SW = *DefaultCaseWeight;
  for (const auto &Case : SI.cases()) {
    auto W = SIW.getSuccessorWeight(Case.getSuccessorIndex());
    assert(W &&(static_cast <bool> (W && "case weight must be defined as default case weight is defined"
) ? void (0) : __assert_fail ("W && \"case weight must be defined as default case weight is defined\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 883, __extension__
 __PRETTY_FUNCTION__))
           "case weight must be defined as default case weight is defined")(static_cast <bool> (W && "case weight must be defined as default case weight is defined"
) ? void (0) : __assert_fail ("W && \"case weight must be defined as default case weight is defined\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 883, __extension__
 __PRETTY_FUNCTION__));
    SW += *W;
  }
  NewSIW.setSuccessorWeight(0, SW);
}

// If we ended up with a common successor for every path through the switch
// after unswitching, rewrite it to an unconditional branch to make it easy
// to recognize. Otherwise we potentially have to recognize the default case
// pointing at unreachable and other complexity.
if (CommonSuccBB) {
  BasicBlock *BB = SI.getParent();
  // We may have had multiple edges to this common successor block, so remove
  // them as predecessors. We skip the first one, either the default or the
  // actual first case.
  bool SkippedFirst = DefaultExitBB == nullptr;
  for (auto Case : SI.cases()) {
    assert(Case.getCaseSuccessor() == CommonSuccBB &&(static_cast <bool> (Case.getCaseSuccessor() == CommonSuccBB
 && "Non-common successor!") ? void (0) : __assert_fail
 ("Case.getCaseSuccessor() == CommonSuccBB && \"Non-common successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 901, __extension__
 __PRETTY_FUNCTION__))
           "Non-common successor!")(static_cast <bool> (Case.getCaseSuccessor() == CommonSuccBB
 && "Non-common successor!") ? void (0) : __assert_fail
 ("Case.getCaseSuccessor() == CommonSuccBB && \"Non-common successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 901, __extension__
 __PRETTY_FUNCTION__));
    (void)Case;
    if (!SkippedFirst) {
      SkippedFirst = true;
      continue;
    }
    CommonSuccBB->removePredecessor(BB,
                                    /*KeepOneInputPHIs*/ true);
  }
  // Now nuke the switch and replace it with a direct branch.
  SIW.eraseFromParent();
  BranchInst::Create(CommonSuccBB, BB);
} else if (DefaultExitBB) {
  assert(SI.getNumCases() > 0 &&(static_cast <bool> (SI.getNumCases() > 0 &&
 "If we had no cases we'd have a common successor!") ? void (
0) : __assert_fail ("SI.getNumCases() > 0 && \"If we had no cases we'd have a common successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 915, __extension__
 __PRETTY_FUNCTION__))
         "If we had no cases we'd have a common successor!")(static_cast <bool> (SI.getNumCases() > 0 &&
 "If we had no cases we'd have a common successor!") ? void (
0) : __assert_fail ("SI.getNumCases() > 0 && \"If we had no cases we'd have a common successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 915, __extension__
 __PRETTY_FUNCTION__));
  // Move the last case to the default successor. This is valid as if the
  // default got unswitched it cannot be reached. This has the advantage of
  // being simple and keeping the number of edges from this switch to
  // successors the same, and avoiding any PHI update complexity.
  auto LastCaseI = std::prev(SI.case_end());

  SI.setDefaultDest(LastCaseI->getCaseSuccessor());
  SIW.setSuccessorWeight(
      0, SIW.getSuccessorWeight(LastCaseI->getSuccessorIndex()));
  SIW.removeCase(LastCaseI);
}

// Walk the unswitched exit blocks and the unswitched split blocks and update
// the dominator tree based on the CFG edits. While we are walking unordered
// containers here, the API for applyUpdates takes an unordered list of
// updates and requires them to not contain duplicates.
SmallVector<DominatorTree::UpdateType, 4> DTUpdates;
for (auto *UnswitchedExitBB : UnswitchedExitBBs) {
  DTUpdates.push_back({DT.Delete, ParentBB, UnswitchedExitBB});
  DTUpdates.push_back({DT.Insert, OldPH, UnswitchedExitBB});
}
for (auto SplitUnswitchedPair : SplitExitBBMap) {
  DTUpdates.push_back({DT.Delete, ParentBB, SplitUnswitchedPair.first});
  DTUpdates.push_back({DT.Insert, OldPH, SplitUnswitchedPair.second});
}

if (MSSAU) {
  MSSAU->applyUpdates(DTUpdates, DT, /*UpdateDT=*/true);
  if (VerifyMemorySSA)
    MSSAU->getMemorySSA()->verifyMemorySSA();
} else {
  DT.applyUpdates(DTUpdates);
}

assert(DT.verify(DominatorTree::VerificationLevel::Fast))(static_cast <bool> (DT.verify(DominatorTree::VerificationLevel
::Fast)) ? void (0) : __assert_fail ("DT.verify(DominatorTree::VerificationLevel::Fast)"
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 950, __extension__
 __PRETTY_FUNCTION__));

// We may have changed the nesting relationship for this loop so hoist it to
// its correct parent if needed.
hoistLoopToNewParent(L, *NewPH, DT, LI, MSSAU, SE);

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

++NumTrivial;
++NumSwitches;
LLVM_DEBUG(dbgs() << "    done: unswitching trivial switch...\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "    done: unswitching trivial switch...\n"
; } } while (false);
return true;
963}

965/// This routine scans the loop to find a branch or switch which occurs before
966/// any side effects occur. These can potentially be unswitched without
967/// duplicating the loop. If a branch or switch is successfully unswitched the
968/// scanning continues to see if subsequent branches or switches have become
969/// trivial. Once all trivial candidates have been unswitched, this routine
970/// returns.
971///
972/// The return value indicates whether anything was unswitched (and therefore
973/// changed).
974///
975/// If `SE` is not null, it will be updated based on the potential loop SCEVs
976/// invalidated by this.
977static bool unswitchAllTrivialConditions(Loop &L, DominatorTree &DT,
                                       LoopInfo &LI, ScalarEvolution *SE,
                                       MemorySSAUpdater *MSSAU) {
bool Changed = false;

// If loop header has only one reachable successor we should keep looking for
// trivial condition candidates in the successor as well. An alternative is
// to constant fold conditions and merge successors into loop header (then we
// only need to check header's terminator). The reason for not doing this in
// LoopUnswitch pass is that it could potentially break LoopPassManager's
// invariants. Folding dead branches could either eliminate the current loop
// or make other loops unreachable. LCSSA form might also not be preserved
// after deleting branches. The following code keeps traversing loop header's
// successors until it finds the trivial condition candidate (condition that
// is not a constant). Since unswitching generates branches with constant
// conditions, this scenario could be very common in practice.
BasicBlock *CurrentBB = L.getHeader();
SmallPtrSet<BasicBlock *, 8> Visited;
Visited.insert(CurrentBB);
do {
  // Check if there are any side-effecting instructions (e.g. stores, calls,
  // volatile loads) in the part of the loop that the code *would* execute
  // without unswitching.
  if (MSSAU) // Possible early exit with MSSA
    if (auto *Defs = MSSAU->getMemorySSA()->getBlockDefs(CurrentBB))
      if (!isa<MemoryPhi>(*Defs->begin()) || (++Defs->begin() != Defs->end()))
        return Changed;
  if (llvm::any_of(*CurrentBB,
                   [](Instruction &I) { return I.mayHaveSideEffects(); }))
    return Changed;

  Instruction *CurrentTerm = CurrentBB->getTerminator();

  if (auto *SI = dyn_cast<SwitchInst>(CurrentTerm)) {
    // Don't bother trying to unswitch past a switch with a constant
    // condition. This should be removed prior to running this pass by
    // simplifycfg.
    if (isa<Constant>(SI->getCondition()))
      return Changed;

    if (!unswitchTrivialSwitch(L, *SI, DT, LI, SE, MSSAU))
      // Couldn't unswitch this one so we're done.
      return Changed;

    // Mark that we managed to unswitch something.
    Changed = true;

    // If unswitching turned the terminator into an unconditional branch then
    // we can continue. The unswitching logic specifically works to fold any
    // cases it can into an unconditional branch to make it easier to
    // recognize here.
    auto *BI = dyn_cast<BranchInst>(CurrentBB->getTerminator());
    if (!BI || BI->isConditional())
      return Changed;

    CurrentBB = BI->getSuccessor(0);
    continue;
  }

  auto *BI = dyn_cast<BranchInst>(CurrentTerm);
  if (!BI)
    // We do not understand other terminator instructions.
    return Changed;

  // Don't bother trying to unswitch past an unconditional branch or a branch
  // with a constant value. These should be removed by simplifycfg prior to
  // running this pass.
  if (!BI->isConditional() ||
      isa<Constant>(skipTrivialSelect(BI->getCondition())))
    return Changed;

  // Found a trivial condition candidate: non-foldable conditional branch. If
  // we fail to unswitch this, we can't do anything else that is trivial.
  if (!unswitchTrivialBranch(L, *BI, DT, LI, SE, MSSAU))
    return Changed;

  // Mark that we managed to unswitch something.
  Changed = true;

  // If we only unswitched some of the conditions feeding the branch, we won't
  // have collapsed it to a single successor.
  BI = cast<BranchInst>(CurrentBB->getTerminator());
  if (BI->isConditional())
    return Changed;

  // Follow the newly unconditional branch into its successor.
  CurrentBB = BI->getSuccessor(0);

  // When continuing, if we exit the loop or reach a previous visited block,
  // then we can not reach any trivial condition candidates (unfoldable
  // branch instructions or switch instructions) and no unswitch can happen.
} while (L.contains(CurrentBB) && Visited.insert(CurrentBB).second);

return Changed;
1071}

1073/// Build the cloned blocks for an unswitched copy of the given loop.
1074///
1075/// The cloned blocks are inserted before the loop preheader (`LoopPH`) and
1076/// after the split block (`SplitBB`) that will be used to select between the
1077/// cloned and original loop.
1078///
1079/// This routine handles cloning all of the necessary loop blocks and exit
1080/// blocks including rewriting their instructions and the relevant PHI nodes.
1081/// Any loop blocks or exit blocks which are dominated by a different successor
1082/// than the one for this clone of the loop blocks can be trivially skipped. We
1083/// use the `DominatingSucc` map to determine whether a block satisfies that
1084/// property with a simple map lookup.
1085///
1086/// It also correctly creates the unconditional branch in the cloned
1087/// unswitched parent block to only point at the unswitched successor.
1088///
1089/// This does not handle most of the necessary updates to `LoopInfo`. Only exit
1090/// block splitting is correctly reflected in `LoopInfo`, essentially all of
1091/// the cloned blocks (and their loops) are left without full `LoopInfo`
1092/// updates. This also doesn't fully update `DominatorTree`. It adds the cloned
1093/// blocks to them but doesn't create the cloned `DominatorTree` structure and
1094/// instead the caller must recompute an accurate DT. It *does* correctly
1095/// update the `AssumptionCache` provided in `AC`.
1096static BasicBlock *buildClonedLoopBlocks(
  Loop &L, BasicBlock *LoopPH, BasicBlock *SplitBB,
  ArrayRef<BasicBlock *> ExitBlocks, BasicBlock *ParentBB,
  BasicBlock *UnswitchedSuccBB, BasicBlock *ContinueSuccBB,
  const SmallDenseMap<BasicBlock *, BasicBlock *, 16> &DominatingSucc,
  ValueToValueMapTy &VMap,
  SmallVectorImpl<DominatorTree::UpdateType> &DTUpdates, AssumptionCache &AC,
  DominatorTree &DT, LoopInfo &LI, MemorySSAUpdater *MSSAU) {
SmallVector<BasicBlock *, 4> NewBlocks;
NewBlocks.reserve(L.getNumBlocks() + ExitBlocks.size());

// We will need to clone a bunch of blocks, wrap up the clone operation in
// a helper.
auto CloneBlock = [&](BasicBlock *OldBB) {
  // Clone the basic block and insert it before the new preheader.
  BasicBlock *NewBB = CloneBasicBlock(OldBB, VMap, ".us", OldBB->getParent());
  NewBB->moveBefore(LoopPH);

  // Record this block and the mapping.
  NewBlocks.push_back(NewBB);
  VMap[OldBB] = NewBB;

  return NewBB;
};

// We skip cloning blocks when they have a dominating succ that is not the
// succ we are cloning for.
auto SkipBlock = [&](BasicBlock *BB) {
  auto It = DominatingSucc.find(BB);
  return It != DominatingSucc.end() && It->second != UnswitchedSuccBB;
};

// First, clone the preheader.
auto *ClonedPH = CloneBlock(LoopPH);

// Then clone all the loop blocks, skipping the ones that aren't necessary.
for (auto *LoopBB : L.blocks())
  if (!SkipBlock(LoopBB))
    CloneBlock(LoopBB);

// Split all the loop exit edges so that when we clone the exit blocks, if
// any of the exit blocks are *also* a preheader for some other loop, we
// don't create multiple predecessors entering the loop header.
for (auto *ExitBB : ExitBlocks) {
  if (SkipBlock(ExitBB))
    continue;

  // When we are going to clone an exit, we don't need to clone all the
  // instructions in the exit block and we want to ensure we have an easy
  // place to merge the CFG, so split the exit first. This is always safe to
  // do because there cannot be any non-loop predecessors of a loop exit in
  // loop simplified form.
  auto *MergeBB = SplitBlock(ExitBB, &ExitBB->front(), &DT, &LI, MSSAU);

  // Rearrange the names to make it easier to write test cases by having the
  // exit block carry the suffix rather than the merge block carrying the
  // suffix.
  MergeBB->takeName(ExitBB);
  ExitBB->setName(Twine(MergeBB->getName()) + ".split");

  // Now clone the original exit block.
  auto *ClonedExitBB = CloneBlock(ExitBB);
  assert(ClonedExitBB->getTerminator()->getNumSuccessors() == 1 &&(static_cast <bool> (ClonedExitBB->getTerminator()->
getNumSuccessors() == 1 && "Exit block should have been split to have one successor!"
) ? void (0) : __assert_fail ("ClonedExitBB->getTerminator()->getNumSuccessors() == 1 && \"Exit block should have been split to have one successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1159, __extension__
 __PRETTY_FUNCTION__))
         "Exit block should have been split to have one successor!")(static_cast <bool> (ClonedExitBB->getTerminator()->
getNumSuccessors() == 1 && "Exit block should have been split to have one successor!"
) ? void (0) : __assert_fail ("ClonedExitBB->getTerminator()->getNumSuccessors() == 1 && \"Exit block should have been split to have one successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1159, __extension__
 __PRETTY_FUNCTION__));
  assert(ClonedExitBB->getTerminator()->getSuccessor(0) == MergeBB &&(static_cast <bool> (ClonedExitBB->getTerminator()->
getSuccessor(0) == MergeBB && "Cloned exit block has the wrong successor!"
) ? void (0) : __assert_fail ("ClonedExitBB->getTerminator()->getSuccessor(0) == MergeBB && \"Cloned exit block has the wrong successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1161, __extension__
 __PRETTY_FUNCTION__))
         "Cloned exit block has the wrong successor!")(static_cast <bool> (ClonedExitBB->getTerminator()->
getSuccessor(0) == MergeBB && "Cloned exit block has the wrong successor!"
) ? void (0) : __assert_fail ("ClonedExitBB->getTerminator()->getSuccessor(0) == MergeBB && \"Cloned exit block has the wrong successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1161, __extension__
 __PRETTY_FUNCTION__));

  // Remap any cloned instructions and create a merge phi node for them.
  for (auto ZippedInsts : llvm::zip_first(
           llvm::make_range(ExitBB->begin(), std::prev(ExitBB->end())),
           llvm::make_range(ClonedExitBB->begin(),
                            std::prev(ClonedExitBB->end())))) {
    Instruction &I = std::get<0>(ZippedInsts);
    Instruction &ClonedI = std::get<1>(ZippedInsts);

    // The only instructions in the exit block should be PHI nodes and
    // potentially a landing pad.
    assert((static_cast <bool> ((isa<PHINode>(I) || isa<LandingPadInst
>(I) || isa<CatchPadInst>(I)) && "Bad instruction in exit block!"
) ? void (0) : __assert_fail ("(isa<PHINode>(I) || isa<LandingPadInst>(I) || isa<CatchPadInst>(I)) && \"Bad instruction in exit block!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1175, __extension__
 __PRETTY_FUNCTION__))
        (isa<PHINode>(I) || isa<LandingPadInst>(I) || isa<CatchPadInst>(I)) &&(static_cast <bool> ((isa<PHINode>(I) || isa<LandingPadInst
>(I) || isa<CatchPadInst>(I)) && "Bad instruction in exit block!"
) ? void (0) : __assert_fail ("(isa<PHINode>(I) || isa<LandingPadInst>(I) || isa<CatchPadInst>(I)) && \"Bad instruction in exit block!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1175, __extension__
 __PRETTY_FUNCTION__))
        "Bad instruction in exit block!")(static_cast <bool> ((isa<PHINode>(I) || isa<LandingPadInst
>(I) || isa<CatchPadInst>(I)) && "Bad instruction in exit block!"
) ? void (0) : __assert_fail ("(isa<PHINode>(I) || isa<LandingPadInst>(I) || isa<CatchPadInst>(I)) && \"Bad instruction in exit block!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1175, __extension__
 __PRETTY_FUNCTION__));
    // We should have a value map between the instruction and its clone.
    assert(VMap.lookup(&I) == &ClonedI && "Mismatch in the value map!")(static_cast <bool> (VMap.lookup(&I) == &ClonedI
 && "Mismatch in the value map!") ? void (0) : __assert_fail
 ("VMap.lookup(&I) == &ClonedI && \"Mismatch in the value map!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1177, __extension__
 __PRETTY_FUNCTION__));

    auto *MergePN =
        PHINode::Create(I.getType(), /*NumReservedValues*/ 2, ".us-phi",
                        &*MergeBB->getFirstInsertionPt());
    I.replaceAllUsesWith(MergePN);
    MergePN->addIncoming(&I, ExitBB);
    MergePN->addIncoming(&ClonedI, ClonedExitBB);
  }
}

// Rewrite the instructions in the cloned blocks to refer to the instructions
// in the cloned blocks. We have to do this as a second pass so that we have
// everything available. Also, we have inserted new instructions which may
// include assume intrinsics, so we update the assumption cache while
// processing this.
for (auto *ClonedBB : NewBlocks)
  for (Instruction &I : *ClonedBB) {
    RemapInstruction(&I, VMap,
                     RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
    if (auto *II = dyn_cast<AssumeInst>(&I))
      AC.registerAssumption(II);
  }

// Update any PHI nodes in the cloned successors of the skipped blocks to not
// have spurious incoming values.
for (auto *LoopBB : L.blocks())
  if (SkipBlock(LoopBB))
    for (auto *SuccBB : successors(LoopBB))
      if (auto *ClonedSuccBB = cast_or_null<BasicBlock>(VMap.lookup(SuccBB)))
        for (PHINode &PN : ClonedSuccBB->phis())
          PN.removeIncomingValue(LoopBB, /*DeletePHIIfEmpty*/ false);

// Remove the cloned parent as a predecessor of any successor we ended up
// cloning other than the unswitched one.
auto *ClonedParentBB = cast<BasicBlock>(VMap.lookup(ParentBB));
for (auto *SuccBB : successors(ParentBB)) {
  if (SuccBB == UnswitchedSuccBB)
    continue;

  auto *ClonedSuccBB = cast_or_null<BasicBlock>(VMap.lookup(SuccBB));
  if (!ClonedSuccBB)
    continue;

  ClonedSuccBB->removePredecessor(ClonedParentBB,
                                  /*KeepOneInputPHIs*/ true);
}

// Replace the cloned branch with an unconditional branch to the cloned
// unswitched successor.
auto *ClonedSuccBB = cast<BasicBlock>(VMap.lookup(UnswitchedSuccBB));
Instruction *ClonedTerminator = ClonedParentBB->getTerminator();
// Trivial Simplification. If Terminator is a conditional branch and
// condition becomes dead - erase it.
Value *ClonedConditionToErase = nullptr;
if (auto *BI = dyn_cast<BranchInst>(ClonedTerminator))
  ClonedConditionToErase = BI->getCondition();
else if (auto *SI = dyn_cast<SwitchInst>(ClonedTerminator))
  ClonedConditionToErase = SI->getCondition();

ClonedTerminator->eraseFromParent();
BranchInst::Create(ClonedSuccBB, ClonedParentBB);

if (ClonedConditionToErase)
  RecursivelyDeleteTriviallyDeadInstructions(ClonedConditionToErase, nullptr,
                                             MSSAU);

// If there are duplicate entries in the PHI nodes because of multiple edges
// to the unswitched successor, we need to nuke all but one as we replaced it
// with a direct branch.
for (PHINode &PN : ClonedSuccBB->phis()) {
  bool Found = false;
  // Loop over the incoming operands backwards so we can easily delete as we
  // go without invalidating the index.
  for (int i = PN.getNumOperands() - 1; i >= 0; --i) {
    if (PN.getIncomingBlock(i) != ClonedParentBB)
      continue;
    if (!Found) {
      Found = true;
      continue;
    }
    PN.removeIncomingValue(i, /*DeletePHIIfEmpty*/ false);
  }
}

// Record the domtree updates for the new blocks.
SmallPtrSet<BasicBlock *, 4> SuccSet;
for (auto *ClonedBB : NewBlocks) {
  for (auto *SuccBB : successors(ClonedBB))
    if (SuccSet.insert(SuccBB).second)
      DTUpdates.push_back({DominatorTree::Insert, ClonedBB, SuccBB});
  SuccSet.clear();
}

return ClonedPH;
1272}

1274/// Recursively clone the specified loop and all of its children.
1275///
1276/// The target parent loop for the clone should be provided, or can be null if
1277/// the clone is a top-level loop. While cloning, all the blocks are mapped
1278/// with the provided value map. The entire original loop must be present in
1279/// the value map. The cloned loop is returned.
1280static Loop *cloneLoopNest(Loop &OrigRootL, Loop *RootParentL,
                         const ValueToValueMapTy &VMap, LoopInfo &LI) {
auto AddClonedBlocksToLoop = [&](Loop &OrigL, Loop &ClonedL) {
  assert(ClonedL.getBlocks().empty() && "Must start with an empty loop!")(static_cast <bool> (ClonedL.getBlocks().empty() &&
 "Must start with an empty loop!") ? void (0) : __assert_fail
 ("ClonedL.getBlocks().empty() && \"Must start with an empty loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1283, __extension__
 __PRETTY_FUNCTION__));
  ClonedL.reserveBlocks(OrigL.getNumBlocks());
  for (auto *BB : OrigL.blocks()) {
    auto *ClonedBB = cast<BasicBlock>(VMap.lookup(BB));
    ClonedL.addBlockEntry(ClonedBB);
    if (LI.getLoopFor(BB) == &OrigL)
      LI.changeLoopFor(ClonedBB, &ClonedL);
  }
};

// We specially handle the first loop because it may get cloned into
// a different parent and because we most commonly are cloning leaf loops.
Loop *ClonedRootL = LI.AllocateLoop();
if (RootParentL)
  RootParentL->addChildLoop(ClonedRootL);
else
  LI.addTopLevelLoop(ClonedRootL);
AddClonedBlocksToLoop(OrigRootL, *ClonedRootL);

if (OrigRootL.isInnermost())
  return ClonedRootL;

// If we have a nest, we can quickly clone the entire loop nest using an
// iterative approach because it is a tree. We keep the cloned parent in the
// data structure to avoid repeatedly querying through a map to find it.
SmallVector<std::pair<Loop *, Loop *>, 16> LoopsToClone;
// Build up the loops to clone in reverse order as we'll clone them from the
// back.
for (Loop *ChildL : llvm::reverse(OrigRootL))
  LoopsToClone.push_back({ClonedRootL, ChildL});
do {
  Loop *ClonedParentL, *L;
  std::tie(ClonedParentL, L) = LoopsToClone.pop_back_val();
  Loop *ClonedL = LI.AllocateLoop();
  ClonedParentL->addChildLoop(ClonedL);
  AddClonedBlocksToLoop(*L, *ClonedL);
  for (Loop *ChildL : llvm::reverse(*L))
    LoopsToClone.push_back({ClonedL, ChildL});
} while (!LoopsToClone.empty());

return ClonedRootL;
1324}

1326/// Build the cloned loops of an original loop from unswitching.
1327///
1328/// Because unswitching simplifies the CFG of the loop, this isn't a trivial
1329/// operation. We need to re-verify that there even is a loop (as the backedge
1330/// may not have been cloned), and even if there are remaining backedges the
1331/// backedge set may be different. However, we know that each child loop is
1332/// undisturbed, we only need to find where to place each child loop within
1333/// either any parent loop or within a cloned version of the original loop.
1334///
1335/// Because child loops may end up cloned outside of any cloned version of the
1336/// original loop, multiple cloned sibling loops may be created. All of them
1337/// are returned so that the newly introduced loop nest roots can be
1338/// identified.
1339static void buildClonedLoops(Loop &OrigL, ArrayRef<BasicBlock *> ExitBlocks,
                           const ValueToValueMapTy &VMap, LoopInfo &LI,
                           SmallVectorImpl<Loop *> &NonChildClonedLoops) {
Loop *ClonedL = nullptr;

auto *OrigPH = OrigL.getLoopPreheader();
auto *OrigHeader = OrigL.getHeader();

auto *ClonedPH = cast<BasicBlock>(VMap.lookup(OrigPH));
auto *ClonedHeader = cast<BasicBlock>(VMap.lookup(OrigHeader));

// We need to know the loops of the cloned exit blocks to even compute the
// accurate parent loop. If we only clone exits to some parent of the
// original parent, we want to clone into that outer loop. We also keep track
// of the loops that our cloned exit blocks participate in.
Loop *ParentL = nullptr;
SmallVector<BasicBlock *, 4> ClonedExitsInLoops;
SmallDenseMap<BasicBlock *, Loop *, 16> ExitLoopMap;
ClonedExitsInLoops.reserve(ExitBlocks.size());
for (auto *ExitBB : ExitBlocks)
  if (auto *ClonedExitBB = cast_or_null<BasicBlock>(VMap.lookup(ExitBB)))
    if (Loop *ExitL = LI.getLoopFor(ExitBB)) {
      ExitLoopMap[ClonedExitBB] = ExitL;
      ClonedExitsInLoops.push_back(ClonedExitBB);
      if (!ParentL || (ParentL != ExitL && ParentL->contains(ExitL)))
        ParentL = ExitL;
    }
assert((!ParentL || ParentL == OrigL.getParentLoop() ||(static_cast <bool> ((!ParentL || ParentL == OrigL.getParentLoop
() || ParentL->contains(OrigL.getParentLoop())) &&
 "The computed parent loop should always contain (or be) the parent of "
 "the original loop.") ? void (0) : __assert_fail ("(!ParentL || ParentL == OrigL.getParentLoop() || ParentL->contains(OrigL.getParentLoop())) && \"The computed parent loop should always contain (or be) the parent of \" \"the original loop.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1369, __extension__
 __PRETTY_FUNCTION__))
        ParentL->contains(OrigL.getParentLoop())) &&(static_cast <bool> ((!ParentL || ParentL == OrigL.getParentLoop
() || ParentL->contains(OrigL.getParentLoop())) &&
 "The computed parent loop should always contain (or be) the parent of "
 "the original loop.") ? void (0) : __assert_fail ("(!ParentL || ParentL == OrigL.getParentLoop() || ParentL->contains(OrigL.getParentLoop())) && \"The computed parent loop should always contain (or be) the parent of \" \"the original loop.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1369, __extension__
 __PRETTY_FUNCTION__))
       "The computed parent loop should always contain (or be) the parent of "(static_cast <bool> ((!ParentL || ParentL == OrigL.getParentLoop
() || ParentL->contains(OrigL.getParentLoop())) &&
 "The computed parent loop should always contain (or be) the parent of "
 "the original loop.") ? void (0) : __assert_fail ("(!ParentL || ParentL == OrigL.getParentLoop() || ParentL->contains(OrigL.getParentLoop())) && \"The computed parent loop should always contain (or be) the parent of \" \"the original loop.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1369, __extension__
 __PRETTY_FUNCTION__))
       "the original loop.")(static_cast <bool> ((!ParentL || ParentL == OrigL.getParentLoop
() || ParentL->contains(OrigL.getParentLoop())) &&
 "The computed parent loop should always contain (or be) the parent of "
 "the original loop.") ? void (0) : __assert_fail ("(!ParentL || ParentL == OrigL.getParentLoop() || ParentL->contains(OrigL.getParentLoop())) && \"The computed parent loop should always contain (or be) the parent of \" \"the original loop.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1369, __extension__
 __PRETTY_FUNCTION__));

// We build the set of blocks dominated by the cloned header from the set of
// cloned blocks out of the original loop. While not all of these will
// necessarily be in the cloned loop, it is enough to establish that they
// aren't in unreachable cycles, etc.
SmallSetVector<BasicBlock *, 16> ClonedLoopBlocks;
for (auto *BB : OrigL.blocks())
  if (auto *ClonedBB = cast_or_null<BasicBlock>(VMap.lookup(BB)))
    ClonedLoopBlocks.insert(ClonedBB);

// Rebuild the set of blocks that will end up in the cloned loop. We may have
// skipped cloning some region of this loop which can in turn skip some of
// the backedges so we have to rebuild the blocks in the loop based on the
// backedges that remain after cloning.
SmallVector<BasicBlock *, 16> Worklist;
SmallPtrSet<BasicBlock *, 16> BlocksInClonedLoop;
for (auto *Pred : predecessors(ClonedHeader)) {
  // The only possible non-loop header predecessor is the preheader because
  // we know we cloned the loop in simplified form.
  if (Pred == ClonedPH)
    continue;

  // Because the loop was in simplified form, the only non-loop predecessor
  // should be the preheader.
  assert(ClonedLoopBlocks.count(Pred) && "Found a predecessor of the loop "(static_cast <bool> (ClonedLoopBlocks.count(Pred) &&
 "Found a predecessor of the loop " "header other than the preheader "
 "that is not part of the loop!") ? void (0) : __assert_fail (
"ClonedLoopBlocks.count(Pred) && \"Found a predecessor of the loop \" \"header other than the preheader \" \"that is not part of the loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1396, __extension__
 __PRETTY_FUNCTION__))
                                         "header other than the preheader "(static_cast <bool> (ClonedLoopBlocks.count(Pred) &&
 "Found a predecessor of the loop " "header other than the preheader "
 "that is not part of the loop!") ? void (0) : __assert_fail (
"ClonedLoopBlocks.count(Pred) && \"Found a predecessor of the loop \" \"header other than the preheader \" \"that is not part of the loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1396, __extension__
 __PRETTY_FUNCTION__))
                                         "that is not part of the loop!")(static_cast <bool> (ClonedLoopBlocks.count(Pred) &&
 "Found a predecessor of the loop " "header other than the preheader "
 "that is not part of the loop!") ? void (0) : __assert_fail (
"ClonedLoopBlocks.count(Pred) && \"Found a predecessor of the loop \" \"header other than the preheader \" \"that is not part of the loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1396, __extension__
 __PRETTY_FUNCTION__));

  // Insert this block into the loop set and on the first visit (and if it
  // isn't the header we're currently walking) put it into the worklist to
  // recurse through.
  if (BlocksInClonedLoop.insert(Pred).second && Pred != ClonedHeader)
    Worklist.push_back(Pred);
}

// If we had any backedges then there *is* a cloned loop. Put the header into
// the loop set and then walk the worklist backwards to find all the blocks
// that remain within the loop after cloning.
if (!BlocksInClonedLoop.empty()) {
  BlocksInClonedLoop.insert(ClonedHeader);

  while (!Worklist.empty()) {
    BasicBlock *BB = Worklist.pop_back_val();
    assert(BlocksInClonedLoop.count(BB) &&(static_cast <bool> (BlocksInClonedLoop.count(BB) &&
 "Didn't put block into the loop set!") ? void (0) : __assert_fail
 ("BlocksInClonedLoop.count(BB) && \"Didn't put block into the loop set!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1414, __extension__
 __PRETTY_FUNCTION__))
           "Didn't put block into the loop set!")(static_cast <bool> (BlocksInClonedLoop.count(BB) &&
 "Didn't put block into the loop set!") ? void (0) : __assert_fail
 ("BlocksInClonedLoop.count(BB) && \"Didn't put block into the loop set!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1414, __extension__
 __PRETTY_FUNCTION__));

    // Insert any predecessors that are in the possible set into the cloned
    // set, and if the insert is successful, add them to the worklist. Note
    // that we filter on the blocks that are definitely reachable via the
    // backedge to the loop header so we may prune out dead code within the
    // cloned loop.
    for (auto *Pred : predecessors(BB))
      if (ClonedLoopBlocks.count(Pred) &&
          BlocksInClonedLoop.insert(Pred).second)
        Worklist.push_back(Pred);
  }

  ClonedL = LI.AllocateLoop();
  if (ParentL) {
    ParentL->addBasicBlockToLoop(ClonedPH, LI);
    ParentL->addChildLoop(ClonedL);
  } else {
    LI.addTopLevelLoop(ClonedL);
  }
  NonChildClonedLoops.push_back(ClonedL);

  ClonedL->reserveBlocks(BlocksInClonedLoop.size());
  // We don't want to just add the cloned loop blocks based on how we
  // discovered them. The original order of blocks was carefully built in
  // a way that doesn't rely on predecessor ordering. Rather than re-invent
  // that logic, we just re-walk the original blocks (and those of the child
  // loops) and filter them as we add them into the cloned loop.
  for (auto *BB : OrigL.blocks()) {
    auto *ClonedBB = cast_or_null<BasicBlock>(VMap.lookup(BB));
    if (!ClonedBB || !BlocksInClonedLoop.count(ClonedBB))
      continue;

    // Directly add the blocks that are only in this loop.
    if (LI.getLoopFor(BB) == &OrigL) {
      ClonedL->addBasicBlockToLoop(ClonedBB, LI);
      continue;
    }

    // We want to manually add it to this loop and parents.
    // Registering it with LoopInfo will happen when we clone the top
    // loop for this block.
    for (Loop *PL = ClonedL; PL; PL = PL->getParentLoop())
      PL->addBlockEntry(ClonedBB);
  }

  // Now add each child loop whose header remains within the cloned loop. All
  // of the blocks within the loop must satisfy the same constraints as the
  // header so once we pass the header checks we can just clone the entire
  // child loop nest.
  for (Loop *ChildL : OrigL) {
    auto *ClonedChildHeader =
        cast_or_null<BasicBlock>(VMap.lookup(ChildL->getHeader()));
    if (!ClonedChildHeader || !BlocksInClonedLoop.count(ClonedChildHeader))
      continue;

1470#ifndef NDEBUG
    // We should never have a cloned child loop header but fail to have
    // all of the blocks for that child loop.
    for (auto *ChildLoopBB : ChildL->blocks())
      assert(BlocksInClonedLoop.count((static_cast <bool> (BlocksInClonedLoop.count( cast<
BasicBlock>(VMap.lookup(ChildLoopBB))) && "Child cloned loop has a header within the cloned outer "
 "loop but not all of its blocks!") ? void (0) : __assert_fail
 ("BlocksInClonedLoop.count( cast<BasicBlock>(VMap.lookup(ChildLoopBB))) && \"Child cloned loop has a header within the cloned outer \" \"loop but not all of its blocks!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1477, __extension__
 __PRETTY_FUNCTION__))
                 cast<BasicBlock>(VMap.lookup(ChildLoopBB))) &&(static_cast <bool> (BlocksInClonedLoop.count( cast<
BasicBlock>(VMap.lookup(ChildLoopBB))) && "Child cloned loop has a header within the cloned outer "
 "loop but not all of its blocks!") ? void (0) : __assert_fail
 ("BlocksInClonedLoop.count( cast<BasicBlock>(VMap.lookup(ChildLoopBB))) && \"Child cloned loop has a header within the cloned outer \" \"loop but not all of its blocks!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1477, __extension__
 __PRETTY_FUNCTION__))
             "Child cloned loop has a header within the cloned outer "(static_cast <bool> (BlocksInClonedLoop.count( cast<
BasicBlock>(VMap.lookup(ChildLoopBB))) && "Child cloned loop has a header within the cloned outer "
 "loop but not all of its blocks!") ? void (0) : __assert_fail
 ("BlocksInClonedLoop.count( cast<BasicBlock>(VMap.lookup(ChildLoopBB))) && \"Child cloned loop has a header within the cloned outer \" \"loop but not all of its blocks!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1477, __extension__
 __PRETTY_FUNCTION__))
             "loop but not all of its blocks!")(static_cast <bool> (BlocksInClonedLoop.count( cast<
BasicBlock>(VMap.lookup(ChildLoopBB))) && "Child cloned loop has a header within the cloned outer "
 "loop but not all of its blocks!") ? void (0) : __assert_fail
 ("BlocksInClonedLoop.count( cast<BasicBlock>(VMap.lookup(ChildLoopBB))) && \"Child cloned loop has a header within the cloned outer \" \"loop but not all of its blocks!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1477, __extension__
 __PRETTY_FUNCTION__));
1478#endif

    cloneLoopNest(*ChildL, ClonedL, VMap, LI);
  }
}

// Now that we've handled all the components of the original loop that were
// cloned into a new loop, we still need to handle anything from the original
// loop that wasn't in a cloned loop.

// Figure out what blocks are left to place within any loop nest containing
// the unswitched loop. If we never formed a loop, the cloned PH is one of
// them.
SmallPtrSet<BasicBlock *, 16> UnloopedBlockSet;
if (BlocksInClonedLoop.empty())
  UnloopedBlockSet.insert(ClonedPH);
for (auto *ClonedBB : ClonedLoopBlocks)
  if (!BlocksInClonedLoop.count(ClonedBB))
    UnloopedBlockSet.insert(ClonedBB);

// Copy the cloned exits and sort them in ascending loop depth, we'll work
// backwards across these to process them inside out. The order shouldn't
// matter as we're just trying to build up the map from inside-out; we use
// the map in a more stably ordered way below.
auto OrderedClonedExitsInLoops = ClonedExitsInLoops;
llvm::sort(OrderedClonedExitsInLoops, [&](BasicBlock *LHS, BasicBlock *RHS) {
  return ExitLoopMap.lookup(LHS)->getLoopDepth() <
         ExitLoopMap.lookup(RHS)->getLoopDepth();
});

// Populate the existing ExitLoopMap with everything reachable from each
// exit, starting from the inner most exit.
while (!UnloopedBlockSet.empty() && !OrderedClonedExitsInLoops.empty()) {
  assert(Worklist.empty() && "Didn't clear worklist!")(static_cast <bool> (Worklist.empty() && "Didn't clear worklist!"
) ? void (0) : __assert_fail ("Worklist.empty() && \"Didn't clear worklist!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1511, __extension__
 __PRETTY_FUNCTION__));

  BasicBlock *ExitBB = OrderedClonedExitsInLoops.pop_back_val();
  Loop *ExitL = ExitLoopMap.lookup(ExitBB);

  // Walk the CFG back until we hit the cloned PH adding everything reachable
  // and in the unlooped set to this exit block's loop.
  Worklist.push_back(ExitBB);
  do {
    BasicBlock *BB = Worklist.pop_back_val();
    // We can stop recursing at the cloned preheader (if we get there).
    if (BB == ClonedPH)
      continue;

    for (BasicBlock *PredBB : predecessors(BB)) {
      // If this pred has already been moved to our set or is part of some
      // (inner) loop, no update needed.
      if (!UnloopedBlockSet.erase(PredBB)) {
        assert((static_cast <bool> ((BlocksInClonedLoop.count(PredBB) ||
 ExitLoopMap.count(PredBB)) && "Predecessor not mapped to a loop!"
) ? void (0) : __assert_fail ("(BlocksInClonedLoop.count(PredBB) || ExitLoopMap.count(PredBB)) && \"Predecessor not mapped to a loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1531, __extension__
 __PRETTY_FUNCTION__))
            (BlocksInClonedLoop.count(PredBB) || ExitLoopMap.count(PredBB)) &&(static_cast <bool> ((BlocksInClonedLoop.count(PredBB) ||
 ExitLoopMap.count(PredBB)) && "Predecessor not mapped to a loop!"
) ? void (0) : __assert_fail ("(BlocksInClonedLoop.count(PredBB) || ExitLoopMap.count(PredBB)) && \"Predecessor not mapped to a loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1531, __extension__
 __PRETTY_FUNCTION__))
            "Predecessor not mapped to a loop!")(static_cast <bool> ((BlocksInClonedLoop.count(PredBB) ||
 ExitLoopMap.count(PredBB)) && "Predecessor not mapped to a loop!"
) ? void (0) : __assert_fail ("(BlocksInClonedLoop.count(PredBB) || ExitLoopMap.count(PredBB)) && \"Predecessor not mapped to a loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1531, __extension__
 __PRETTY_FUNCTION__));
        continue;
      }

      // We just insert into the loop set here. We'll add these blocks to the
      // exit loop after we build up the set in an order that doesn't rely on
      // predecessor order (which in turn relies on use list order).
      bool Inserted = ExitLoopMap.insert({PredBB, ExitL}).second;
      (void)Inserted;
      assert(Inserted && "Should only visit an unlooped block once!")(static_cast <bool> (Inserted && "Should only visit an unlooped block once!"
) ? void (0) : __assert_fail ("Inserted && \"Should only visit an unlooped block once!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1540, __extension__
 __PRETTY_FUNCTION__));

      // And recurse through to its predecessors.
      Worklist.push_back(PredBB);
    }
  } while (!Worklist.empty());
}

// Now that the ExitLoopMap gives as  mapping for all the non-looping cloned
// blocks to their outer loops, walk the cloned blocks and the cloned exits
// in their original order adding them to the correct loop.

// We need a stable insertion order. We use the order of the original loop
// order and map into the correct parent loop.
for (auto *BB : llvm::concat<BasicBlock *const>(
         makeArrayRef(ClonedPH), ClonedLoopBlocks, ClonedExitsInLoops))
  if (Loop *OuterL = ExitLoopMap.lookup(BB))
    OuterL->addBasicBlockToLoop(BB, LI);

1559#ifndef NDEBUG
for (auto &BBAndL : ExitLoopMap) {
  auto *BB = BBAndL.first;
  auto *OuterL = BBAndL.second;
  assert(LI.getLoopFor(BB) == OuterL &&(static_cast <bool> (LI.getLoopFor(BB) == OuterL &&
 "Failed to put all blocks into outer loops!") ? void (0) : __assert_fail
 ("LI.getLoopFor(BB) == OuterL && \"Failed to put all blocks into outer loops!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1564, __extension__
 __PRETTY_FUNCTION__))
         "Failed to put all blocks into outer loops!")(static_cast <bool> (LI.getLoopFor(BB) == OuterL &&
 "Failed to put all blocks into outer loops!") ? void (0) : __assert_fail
 ("LI.getLoopFor(BB) == OuterL && \"Failed to put all blocks into outer loops!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1564, __extension__
 __PRETTY_FUNCTION__));
}
1566#endif

// Now that all the blocks are placed into the correct containing loop in the
// absence of child loops, find all the potentially cloned child loops and
// clone them into whatever outer loop we placed their header into.
for (Loop *ChildL : OrigL) {
  auto *ClonedChildHeader =
      cast_or_null<BasicBlock>(VMap.lookup(ChildL->getHeader()));
  if (!ClonedChildHeader || BlocksInClonedLoop.count(ClonedChildHeader))
    continue;

1577#ifndef NDEBUG
  for (auto *ChildLoopBB : ChildL->blocks())
    assert(VMap.count(ChildLoopBB) &&(static_cast <bool> (VMap.count(ChildLoopBB) &&
 "Cloned a child loop header but not all of that loops blocks!"
) ? void (0) : __assert_fail ("VMap.count(ChildLoopBB) && \"Cloned a child loop header but not all of that loops blocks!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1580, __extension__
 __PRETTY_FUNCTION__))
           "Cloned a child loop header but not all of that loops blocks!")(static_cast <bool> (VMap.count(ChildLoopBB) &&
 "Cloned a child loop header but not all of that loops blocks!"
) ? void (0) : __assert_fail ("VMap.count(ChildLoopBB) && \"Cloned a child loop header but not all of that loops blocks!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1580, __extension__
 __PRETTY_FUNCTION__));
1581#endif

  NonChildClonedLoops.push_back(cloneLoopNest(
      *ChildL, ExitLoopMap.lookup(ClonedChildHeader), VMap, LI));
}
1586}

1588static void
1589deleteDeadClonedBlocks(Loop &L, ArrayRef<BasicBlock *> ExitBlocks,
                     ArrayRef<std::unique_ptr<ValueToValueMapTy>> VMaps,
                     DominatorTree &DT, MemorySSAUpdater *MSSAU) {
// Find all the dead clones, and remove them from their successors.
SmallVector<BasicBlock *, 16> DeadBlocks;
for (BasicBlock *BB : llvm::concat<BasicBlock *const>(L.blocks(), ExitBlocks))
  for (const auto &VMap : VMaps)
    if (BasicBlock *ClonedBB = cast_or_null<BasicBlock>(VMap->lookup(BB)))
      if (!DT.isReachableFromEntry(ClonedBB)) {
        for (BasicBlock *SuccBB : successors(ClonedBB))
          SuccBB->removePredecessor(ClonedBB);
        DeadBlocks.push_back(ClonedBB);
      }

// Remove all MemorySSA in the dead blocks
if (MSSAU) {
  SmallSetVector<BasicBlock *, 8> DeadBlockSet(DeadBlocks.begin(),
                                               DeadBlocks.end());
  MSSAU->removeBlocks(DeadBlockSet);
}

// Drop any remaining references to break cycles.
for (BasicBlock *BB : DeadBlocks)
  BB->dropAllReferences();
// Erase them from the IR.
for (BasicBlock *BB : DeadBlocks)
  BB->eraseFromParent();
1616}

1618static void
1619deleteDeadBlocksFromLoop(Loop &L,
                       SmallVectorImpl<BasicBlock *> &ExitBlocks,
                       DominatorTree &DT, LoopInfo &LI,
                       MemorySSAUpdater *MSSAU,
                       function_ref<void(Loop &, StringRef)> DestroyLoopCB) {
// Find all the dead blocks tied to this loop, and remove them from their
// successors.
SmallSetVector<BasicBlock *, 8> DeadBlockSet;

// Start with loop/exit blocks and get a transitive closure of reachable dead
// blocks.
SmallVector<BasicBlock *, 16> DeathCandidates(ExitBlocks.begin(),
                                              ExitBlocks.end());
DeathCandidates.append(L.blocks().begin(), L.blocks().end());
while (!DeathCandidates.empty()) {
  auto *BB = DeathCandidates.pop_back_val();
  if (!DeadBlockSet.count(BB) && !DT.isReachableFromEntry(BB)) {
    for (BasicBlock *SuccBB : successors(BB)) {
      SuccBB->removePredecessor(BB);
      DeathCandidates.push_back(SuccBB);
    }
    DeadBlockSet.insert(BB);
  }
}

// Remove all MemorySSA in the dead blocks
if (MSSAU)
  MSSAU->removeBlocks(DeadBlockSet);

// Filter out the dead blocks from the exit blocks list so that it can be
// used in the caller.
llvm::erase_if(ExitBlocks,
               [&](BasicBlock *BB) { return DeadBlockSet.count(BB); });

// Walk from this loop up through its parents removing all of the dead blocks.
for (Loop *ParentL = &L; ParentL; ParentL = ParentL->getParentLoop()) {
  for (auto *BB : DeadBlockSet)
    ParentL->getBlocksSet().erase(BB);
  llvm::erase_if(ParentL->getBlocksVector(),
                 [&](BasicBlock *BB) { return DeadBlockSet.count(BB); });
}

// Now delete the dead child loops. This raw delete will clear them
// recursively.
llvm::erase_if(L.getSubLoopsVector(), [&](Loop *ChildL) {
  if (!DeadBlockSet.count(ChildL->getHeader()))
    return false;

  assert(llvm::all_of(ChildL->blocks(),(static_cast <bool> (llvm::all_of(ChildL->blocks(), [
&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB
); }) && "If the child loop header is dead all blocks in the child loop must "
 "be dead as well!") ? void (0) : __assert_fail ("llvm::all_of(ChildL->blocks(), [&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB); }) && \"If the child loop header is dead all blocks in the child loop must \" \"be dead as well!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1672, __extension__
 __PRETTY_FUNCTION__))
                      [&](BasicBlock *ChildBB) {(static_cast <bool> (llvm::all_of(ChildL->blocks(), [
&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB
); }) && "If the child loop header is dead all blocks in the child loop must "
 "be dead as well!") ? void (0) : __assert_fail ("llvm::all_of(ChildL->blocks(), [&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB); }) && \"If the child loop header is dead all blocks in the child loop must \" \"be dead as well!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1672, __extension__
 __PRETTY_FUNCTION__))
                        return DeadBlockSet.count(ChildBB);(static_cast <bool> (llvm::all_of(ChildL->blocks(), [
&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB
); }) && "If the child loop header is dead all blocks in the child loop must "
 "be dead as well!") ? void (0) : __assert_fail ("llvm::all_of(ChildL->blocks(), [&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB); }) && \"If the child loop header is dead all blocks in the child loop must \" \"be dead as well!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1672, __extension__
 __PRETTY_FUNCTION__))
                      }) &&(static_cast <bool> (llvm::all_of(ChildL->blocks(), [
&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB
); }) && "If the child loop header is dead all blocks in the child loop must "
 "be dead as well!") ? void (0) : __assert_fail ("llvm::all_of(ChildL->blocks(), [&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB); }) && \"If the child loop header is dead all blocks in the child loop must \" \"be dead as well!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1672, __extension__
 __PRETTY_FUNCTION__))
         "If the child loop header is dead all blocks in the child loop must "(static_cast <bool> (llvm::all_of(ChildL->blocks(), [
&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB
); }) && "If the child loop header is dead all blocks in the child loop must "
 "be dead as well!") ? void (0) : __assert_fail ("llvm::all_of(ChildL->blocks(), [&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB); }) && \"If the child loop header is dead all blocks in the child loop must \" \"be dead as well!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1672, __extension__
 __PRETTY_FUNCTION__))
         "be dead as well!")(static_cast <bool> (llvm::all_of(ChildL->blocks(), [
&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB
); }) && "If the child loop header is dead all blocks in the child loop must "
 "be dead as well!") ? void (0) : __assert_fail ("llvm::all_of(ChildL->blocks(), [&](BasicBlock *ChildBB) { return DeadBlockSet.count(ChildBB); }) && \"If the child loop header is dead all blocks in the child loop must \" \"be dead as well!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1672, __extension__
 __PRETTY_FUNCTION__));
  DestroyLoopCB(*ChildL, ChildL->getName());
  LI.destroy(ChildL);
  return true;
});

// Remove the loop mappings for the dead blocks and drop all the references
// from these blocks to others to handle cyclic references as we start
// deleting the blocks themselves.
for (auto *BB : DeadBlockSet) {
  // Check that the dominator tree has already been updated.
  assert(!DT.getNode(BB) && "Should already have cleared domtree!")(static_cast <bool> (!DT.getNode(BB) && "Should already have cleared domtree!"
) ? void (0) : __assert_fail ("!DT.getNode(BB) && \"Should already have cleared domtree!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1683, __extension__
 __PRETTY_FUNCTION__));
  LI.changeLoopFor(BB, nullptr);
  // Drop all uses of the instructions to make sure we won't have dangling
  // uses in other blocks.
  for (auto &I : *BB)
    if (!I.use_empty())
      I.replaceAllUsesWith(PoisonValue::get(I.getType()));
  BB->dropAllReferences();
}

// Actually delete the blocks now that they've been fully unhooked from the
// IR.
for (auto *BB : DeadBlockSet)
  BB->eraseFromParent();
1697}

1699/// Recompute the set of blocks in a loop after unswitching.
1700///
1701/// This walks from the original headers predecessors to rebuild the loop. We
1702/// take advantage of the fact that new blocks can't have been added, and so we
1703/// filter by the original loop's blocks. This also handles potentially
1704/// unreachable code that we don't want to explore but might be found examining
1705/// the predecessors of the header.
1706///
1707/// If the original loop is no longer a loop, this will return an empty set. If
1708/// it remains a loop, all the blocks within it will be added to the set
1709/// (including those blocks in inner loops).
1710static SmallPtrSet<const BasicBlock *, 16> recomputeLoopBlockSet(Loop &L,
                                                               LoopInfo &LI) {
SmallPtrSet<const BasicBlock *, 16> LoopBlockSet;

auto *PH = L.getLoopPreheader();
auto *Header = L.getHeader();

// A worklist to use while walking backwards from the header.
SmallVector<BasicBlock *, 16> Worklist;

// First walk the predecessors of the header to find the backedges. This will
// form the basis of our walk.
for (auto *Pred : predecessors(Header)) {
  // Skip the preheader.
  if (Pred == PH)
    continue;

  // Because the loop was in simplified form, the only non-loop predecessor
  // is the preheader.
  assert(L.contains(Pred) && "Found a predecessor of the loop header other "(static_cast <bool> (L.contains(Pred) && "Found a predecessor of the loop header other "
 "than the preheader that is not part of the " "loop!") ? void
 (0) : __assert_fail ("L.contains(Pred) && \"Found a predecessor of the loop header other \" \"than the preheader that is not part of the \" \"loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1731, __extension__
 __PRETTY_FUNCTION__))
                             "than the preheader that is not part of the "(static_cast <bool> (L.contains(Pred) && "Found a predecessor of the loop header other "
 "than the preheader that is not part of the " "loop!") ? void
 (0) : __assert_fail ("L.contains(Pred) && \"Found a predecessor of the loop header other \" \"than the preheader that is not part of the \" \"loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1731, __extension__
 __PRETTY_FUNCTION__))
                             "loop!")(static_cast <bool> (L.contains(Pred) && "Found a predecessor of the loop header other "
 "than the preheader that is not part of the " "loop!") ? void
 (0) : __assert_fail ("L.contains(Pred) && \"Found a predecessor of the loop header other \" \"than the preheader that is not part of the \" \"loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1731, __extension__
 __PRETTY_FUNCTION__));

  // Insert this block into the loop set and on the first visit and, if it
  // isn't the header we're currently walking, put it into the worklist to
  // recurse through.
  if (LoopBlockSet.insert(Pred).second && Pred != Header)
    Worklist.push_back(Pred);
}

// If no backedges were found, we're done.
if (LoopBlockSet.empty())
  return LoopBlockSet;

// We found backedges, recurse through them to identify the loop blocks.
while (!Worklist.empty()) {
  BasicBlock *BB = Worklist.pop_back_val();
  assert(LoopBlockSet.count(BB) && "Didn't put block into the loop set!")(static_cast <bool> (LoopBlockSet.count(BB) && "Didn't put block into the loop set!"
) ? void (0) : __assert_fail ("LoopBlockSet.count(BB) && \"Didn't put block into the loop set!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1747, __extension__
 __PRETTY_FUNCTION__));

  // No need to walk past the header.
  if (BB == Header)
    continue;

  // Because we know the inner loop structure remains valid we can use the
  // loop structure to jump immediately across the entire nested loop.
  // Further, because it is in loop simplified form, we can directly jump
  // to its preheader afterward.
  if (Loop *InnerL = LI.getLoopFor(BB))
    if (InnerL != &L) {
      assert(L.contains(InnerL) &&(static_cast <bool> (L.contains(InnerL) && "Should not reach a loop *outside* this loop!"
) ? void (0) : __assert_fail ("L.contains(InnerL) && \"Should not reach a loop *outside* this loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1760, __extension__
 __PRETTY_FUNCTION__))
             "Should not reach a loop *outside* this loop!")(static_cast <bool> (L.contains(InnerL) && "Should not reach a loop *outside* this loop!"
) ? void (0) : __assert_fail ("L.contains(InnerL) && \"Should not reach a loop *outside* this loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1760, __extension__
 __PRETTY_FUNCTION__));
      // The preheader is the only possible predecessor of the loop so
      // insert it into the set and check whether it was already handled.
      auto *InnerPH = InnerL->getLoopPreheader();
      assert(L.contains(InnerPH) && "Cannot contain an inner loop block "(static_cast <bool> (L.contains(InnerPH) && "Cannot contain an inner loop block "
 "but not contain the inner loop " "preheader!") ? void (0) :
 __assert_fail ("L.contains(InnerPH) && \"Cannot contain an inner loop block \" \"but not contain the inner loop \" \"preheader!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1766, __extension__
 __PRETTY_FUNCTION__))
                                    "but not contain the inner loop "(static_cast <bool> (L.contains(InnerPH) && "Cannot contain an inner loop block "
 "but not contain the inner loop " "preheader!") ? void (0) :
 __assert_fail ("L.contains(InnerPH) && \"Cannot contain an inner loop block \" \"but not contain the inner loop \" \"preheader!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1766, __extension__
 __PRETTY_FUNCTION__))
                                    "preheader!")(static_cast <bool> (L.contains(InnerPH) && "Cannot contain an inner loop block "
 "but not contain the inner loop " "preheader!") ? void (0) :
 __assert_fail ("L.contains(InnerPH) && \"Cannot contain an inner loop block \" \"but not contain the inner loop \" \"preheader!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1766, __extension__
 __PRETTY_FUNCTION__));
      if (!LoopBlockSet.insert(InnerPH).second)
        // The only way to reach the preheader is through the loop body
        // itself so if it has been visited the loop is already handled.
        continue;

      // Insert all of the blocks (other than those already present) into
      // the loop set. We expect at least the block that led us to find the
      // inner loop to be in the block set, but we may also have other loop
      // blocks if they were already enqueued as predecessors of some other
      // outer loop block.
      for (auto *InnerBB : InnerL->blocks()) {
        if (InnerBB == BB) {
          assert(LoopBlockSet.count(InnerBB) &&(static_cast <bool> (LoopBlockSet.count(InnerBB) &&
 "Block should already be in the set!") ? void (0) : __assert_fail
 ("LoopBlockSet.count(InnerBB) && \"Block should already be in the set!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1780, __extension__
 __PRETTY_FUNCTION__))
                 "Block should already be in the set!")(static_cast <bool> (LoopBlockSet.count(InnerBB) &&
 "Block should already be in the set!") ? void (0) : __assert_fail
 ("LoopBlockSet.count(InnerBB) && \"Block should already be in the set!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1780, __extension__
 __PRETTY_FUNCTION__));
          continue;
        }

        LoopBlockSet.insert(InnerBB);
      }

      // Add the preheader to the worklist so we will continue past the
      // loop body.
      Worklist.push_back(InnerPH);
      continue;
    }

  // Insert any predecessors that were in the original loop into the new
  // set, and if the insert is successful, add them to the worklist.
  for (auto *Pred : predecessors(BB))
    if (L.contains(Pred) && LoopBlockSet.insert(Pred).second)
      Worklist.push_back(Pred);
}

assert(LoopBlockSet.count(Header) && "Cannot fail to add the header!")(static_cast <bool> (LoopBlockSet.count(Header) &&
 "Cannot fail to add the header!") ? void (0) : __assert_fail
 ("LoopBlockSet.count(Header) && \"Cannot fail to add the header!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1800, __extension__
 __PRETTY_FUNCTION__));

// We've found all the blocks participating in the loop, return our completed
// set.
return LoopBlockSet;
1805}

1807/// Rebuild a loop after unswitching removes some subset of blocks and edges.
1808///
1809/// The removal may have removed some child loops entirely but cannot have
1810/// disturbed any remaining child loops. However, they may need to be hoisted
1811/// to the parent loop (or to be top-level loops). The original loop may be
1812/// completely removed.
1813///
1814/// The sibling loops resulting from this update are returned. If the original
1815/// loop remains a valid loop, it will be the first entry in this list with all
1816/// of the newly sibling loops following it.
1817///
1818/// Returns true if the loop remains a loop after unswitching, and false if it
1819/// is no longer a loop after unswitching (and should not continue to be
1820/// referenced).
1821static bool rebuildLoopAfterUnswitch(Loop &L, ArrayRef<BasicBlock *> ExitBlocks,
                                   LoopInfo &LI,
                                   SmallVectorImpl<Loop *> &HoistedLoops) {
auto *PH = L.getLoopPreheader();

// Compute the actual parent loop from the exit blocks. Because we may have
// pruned some exits the loop may be different from the original parent.
Loop *ParentL = nullptr;
SmallVector<Loop *, 4> ExitLoops;
SmallVector<BasicBlock *, 4> ExitsInLoops;
ExitsInLoops.reserve(ExitBlocks.size());
for (auto *ExitBB : ExitBlocks)
  if (Loop *ExitL = LI.getLoopFor(ExitBB)) {
    ExitLoops.push_back(ExitL);
    ExitsInLoops.push_back(ExitBB);
    if (!ParentL || (ParentL != ExitL && ParentL->contains(ExitL)))
      ParentL = ExitL;
  }

// Recompute the blocks participating in this loop. This may be empty if it
// is no longer a loop.
auto LoopBlockSet = recomputeLoopBlockSet(L, LI);

// If we still have a loop, we need to re-set the loop's parent as the exit
// block set changing may have moved it within the loop nest. Note that this
// can only happen when this loop has a parent as it can only hoist the loop
// *up* the nest.
if (!LoopBlockSet.empty() && L.getParentLoop() != ParentL) {
  // Remove this loop's (original) blocks from all of the intervening loops.
  for (Loop *IL = L.getParentLoop(); IL != ParentL;
       IL = IL->getParentLoop()) {
    IL->getBlocksSet().erase(PH);
    for (auto *BB : L.blocks())
      IL->getBlocksSet().erase(BB);
    llvm::erase_if(IL->getBlocksVector(), [&](BasicBlock *BB) {
      return BB == PH || L.contains(BB);
    });
  }

  LI.changeLoopFor(PH, ParentL);
  L.getParentLoop()->removeChildLoop(&L);
  if (ParentL)
    ParentL->addChildLoop(&L);
  else
    LI.addTopLevelLoop(&L);
}

// Now we update all the blocks which are no longer within the loop.
auto &Blocks = L.getBlocksVector();
auto BlocksSplitI =
    LoopBlockSet.empty()
        ? Blocks.begin()
        : std::stable_partition(
              Blocks.begin(), Blocks.end(),
              [&](BasicBlock *BB) { return LoopBlockSet.count(BB); });

// Before we erase the list of unlooped blocks, build a set of them.
SmallPtrSet<BasicBlock *, 16> UnloopedBlocks(BlocksSplitI, Blocks.end());
if (LoopBlockSet.empty())
  UnloopedBlocks.insert(PH);

// Now erase these blocks from the loop.
for (auto *BB : make_range(BlocksSplitI, Blocks.end()))
  L.getBlocksSet().erase(BB);
Blocks.erase(BlocksSplitI, Blocks.end());

// Sort the exits in ascending loop depth, we'll work backwards across these
// to process them inside out.
llvm::stable_sort(ExitsInLoops, [&](BasicBlock *LHS, BasicBlock *RHS) {
  return LI.getLoopDepth(LHS) < LI.getLoopDepth(RHS);
});

// We'll build up a set for each exit loop.
SmallPtrSet<BasicBlock *, 16> NewExitLoopBlocks;
Loop *PrevExitL = L.getParentLoop(); // The deepest possible exit loop.

auto RemoveUnloopedBlocksFromLoop =
    [](Loop &L, SmallPtrSetImpl<BasicBlock *> &UnloopedBlocks) {
      for (auto *BB : UnloopedBlocks)
        L.getBlocksSet().erase(BB);
      llvm::erase_if(L.getBlocksVector(), [&](BasicBlock *BB) {
        return UnloopedBlocks.count(BB);
      });
    };

SmallVector<BasicBlock *, 16> Worklist;
while (!UnloopedBlocks.empty() && !ExitsInLoops.empty()) {
  assert(Worklist.empty() && "Didn't clear worklist!")(static_cast <bool> (Worklist.empty() && "Didn't clear worklist!"
) ? void (0) : __assert_fail ("Worklist.empty() && \"Didn't clear worklist!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1908, __extension__
 __PRETTY_FUNCTION__));
  assert(NewExitLoopBlocks.empty() && "Didn't clear loop set!")(static_cast <bool> (NewExitLoopBlocks.empty() &&
 "Didn't clear loop set!") ? void (0) : __assert_fail ("NewExitLoopBlocks.empty() && \"Didn't clear loop set!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1909, __extension__
 __PRETTY_FUNCTION__));

  // Grab the next exit block, in decreasing loop depth order.
  BasicBlock *ExitBB = ExitsInLoops.pop_back_val();
  Loop &ExitL = *LI.getLoopFor(ExitBB);
  assert(ExitL.contains(&L) && "Exit loop must contain the inner loop!")(static_cast <bool> (ExitL.contains(&L) && "Exit loop must contain the inner loop!"
) ? void (0) : __assert_fail ("ExitL.contains(&L) && \"Exit loop must contain the inner loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1914, __extension__
 __PRETTY_FUNCTION__));

  // Erase all of the unlooped blocks from the loops between the previous
  // exit loop and this exit loop. This works because the ExitInLoops list is
  // sorted in increasing order of loop depth and thus we visit loops in
  // decreasing order of loop depth.
  for (; PrevExitL != &ExitL; PrevExitL = PrevExitL->getParentLoop())
    RemoveUnloopedBlocksFromLoop(*PrevExitL, UnloopedBlocks);

  // Walk the CFG back until we hit the cloned PH adding everything reachable
  // and in the unlooped set to this exit block's loop.
  Worklist.push_back(ExitBB);
  do {
    BasicBlock *BB = Worklist.pop_back_val();
    // We can stop recursing at the cloned preheader (if we get there).
    if (BB == PH)
      continue;

    for (BasicBlock *PredBB : predecessors(BB)) {
      // If this pred has already been moved to our set or is part of some
      // (inner) loop, no update needed.
      if (!UnloopedBlocks.erase(PredBB)) {
        assert((NewExitLoopBlocks.count(PredBB) ||(static_cast <bool> ((NewExitLoopBlocks.count(PredBB) ||
 ExitL.contains(LI.getLoopFor(PredBB))) && "Predecessor not in a nested loop (or already visited)!"
) ? void (0) : __assert_fail ("(NewExitLoopBlocks.count(PredBB) || ExitL.contains(LI.getLoopFor(PredBB))) && \"Predecessor not in a nested loop (or already visited)!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1938, __extension__
 __PRETTY_FUNCTION__))
                ExitL.contains(LI.getLoopFor(PredBB))) &&(static_cast <bool> ((NewExitLoopBlocks.count(PredBB) ||
 ExitL.contains(LI.getLoopFor(PredBB))) && "Predecessor not in a nested loop (or already visited)!"
) ? void (0) : __assert_fail ("(NewExitLoopBlocks.count(PredBB) || ExitL.contains(LI.getLoopFor(PredBB))) && \"Predecessor not in a nested loop (or already visited)!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1938, __extension__
 __PRETTY_FUNCTION__))
               "Predecessor not in a nested loop (or already visited)!")(static_cast <bool> ((NewExitLoopBlocks.count(PredBB) ||
 ExitL.contains(LI.getLoopFor(PredBB))) && "Predecessor not in a nested loop (or already visited)!"
) ? void (0) : __assert_fail ("(NewExitLoopBlocks.count(PredBB) || ExitL.contains(LI.getLoopFor(PredBB))) && \"Predecessor not in a nested loop (or already visited)!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1938, __extension__
 __PRETTY_FUNCTION__));
        continue;
      }

      // We just insert into the loop set here. We'll add these blocks to the
      // exit loop after we build up the set in a deterministic order rather
      // than the predecessor-influenced visit order.
      bool Inserted = NewExitLoopBlocks.insert(PredBB).second;
      (void)Inserted;
      assert(Inserted && "Should only visit an unlooped block once!")(static_cast <bool> (Inserted && "Should only visit an unlooped block once!"
) ? void (0) : __assert_fail ("Inserted && \"Should only visit an unlooped block once!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 1947, __extension__
 __PRETTY_FUNCTION__));

      // And recurse through to its predecessors.
      Worklist.push_back(PredBB);
    }
  } while (!Worklist.empty());

  // If blocks in this exit loop were directly part of the original loop (as
  // opposed to a child loop) update the map to point to this exit loop. This
  // just updates a map and so the fact that the order is unstable is fine.
  for (auto *BB : NewExitLoopBlocks)
    if (Loop *BBL = LI.getLoopFor(BB))
      if (BBL == &L || !L.contains(BBL))
        LI.changeLoopFor(BB, &ExitL);

  // We will remove the remaining unlooped blocks from this loop in the next
  // iteration or below.
  NewExitLoopBlocks.clear();
}

// Any remaining unlooped blocks are no longer part of any loop unless they
// are part of some child loop.
for (; PrevExitL; PrevExitL = PrevExitL->getParentLoop())
  RemoveUnloopedBlocksFromLoop(*PrevExitL, UnloopedBlocks);
for (auto *BB : UnloopedBlocks)
  if (Loop *BBL = LI.getLoopFor(BB))
    if (BBL == &L || !L.contains(BBL))
      LI.changeLoopFor(BB, nullptr);

// Sink all the child loops whose headers are no longer in the loop set to
// the parent (or to be top level loops). We reach into the loop and directly
// update its subloop vector to make this batch update efficient.
auto &SubLoops = L.getSubLoopsVector();
auto SubLoopsSplitI =
    LoopBlockSet.empty()
        ? SubLoops.begin()
        : std::stable_partition(
              SubLoops.begin(), SubLoops.end(), [&](Loop *SubL) {
                return LoopBlockSet.count(SubL->getHeader());
              });
for (auto *HoistedL : make_range(SubLoopsSplitI, SubLoops.end())) {
  HoistedLoops.push_back(HoistedL);
  HoistedL->setParentLoop(nullptr);

  // To compute the new parent of this hoisted loop we look at where we
  // placed the preheader above. We can't lookup the header itself because we
  // retained the mapping from the header to the hoisted loop. But the
  // preheader and header should have the exact same new parent computed
  // based on the set of exit blocks from the original loop as the preheader
  // is a predecessor of the header and so reached in the reverse walk. And
  // because the loops were all in simplified form the preheader of the
  // hoisted loop can't be part of some *other* loop.
  if (auto *NewParentL = LI.getLoopFor(HoistedL->getLoopPreheader()))
    NewParentL->addChildLoop(HoistedL);
  else
    LI.addTopLevelLoop(HoistedL);
}
SubLoops.erase(SubLoopsSplitI, SubLoops.end());

// Actually delete the loop if nothing remained within it.
if (Blocks.empty()) {
  assert(SubLoops.empty() &&(static_cast <bool> (SubLoops.empty() && "Failed to remove all subloops from the original loop!"
) ? void (0) : __assert_fail ("SubLoops.empty() && \"Failed to remove all subloops from the original loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2009, __extension__
 __PRETTY_FUNCTION__))
         "Failed to remove all subloops from the original loop!")(static_cast <bool> (SubLoops.empty() && "Failed to remove all subloops from the original loop!"
) ? void (0) : __assert_fail ("SubLoops.empty() && \"Failed to remove all subloops from the original loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2009, __extension__
 __PRETTY_FUNCTION__));
  if (Loop *ParentL = L.getParentLoop())
    ParentL->removeChildLoop(llvm::find(*ParentL, &L));
  else
    LI.removeLoop(llvm::find(LI, &L));
  // markLoopAsDeleted for L should be triggered by the caller (it is typically
  // done by using the UnswitchCB callback).
  LI.destroy(&L);
  return false;
}

return true;
2021}

2023/// Helper to visit a dominator subtree, invoking a callable on each node.
2024///
2025/// Returning false at any point will stop walking past that node of the tree.
2026template <typename CallableT>
2027void visitDomSubTree(DominatorTree &DT, BasicBlock *BB, CallableT Callable) {
SmallVector<DomTreeNode *, 4> DomWorklist;
DomWorklist.push_back(DT[BB]);
2030#ifndef NDEBUG
SmallPtrSet<DomTreeNode *, 4> Visited;
Visited.insert(DT[BB]);
2033#endif
do {
  DomTreeNode *N = DomWorklist.pop_back_val();

  // Visit this node.
  if (!Callable(N->getBlock()))
    continue;

  // Accumulate the child nodes.
  for (DomTreeNode *ChildN : *N) {
    assert(Visited.insert(ChildN).second &&(static_cast <bool> (Visited.insert(ChildN).second &&
 "Cannot visit a node twice when walking a tree!") ? void (0)
 : __assert_fail ("Visited.insert(ChildN).second && \"Cannot visit a node twice when walking a tree!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2044, __extension__
 __PRETTY_FUNCTION__))
           "Cannot visit a node twice when walking a tree!")(static_cast <bool> (Visited.insert(ChildN).second &&
 "Cannot visit a node twice when walking a tree!") ? void (0)
 : __assert_fail ("Visited.insert(ChildN).second && \"Cannot visit a node twice when walking a tree!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2044, __extension__
 __PRETTY_FUNCTION__));
    DomWorklist.push_back(ChildN);
  }
} while (!DomWorklist.empty());
2048}

2050static void unswitchNontrivialInvariants(
  Loop &L, Instruction &TI, ArrayRef<Value *> Invariants,
  SmallVectorImpl<BasicBlock *> &ExitBlocks, IVConditionInfo &PartialIVInfo,
  DominatorTree &DT, LoopInfo &LI, AssumptionCache &AC,
  function_ref<void(bool, bool, ArrayRef<Loop *>)> UnswitchCB,
  ScalarEvolution *SE, MemorySSAUpdater *MSSAU,
  function_ref<void(Loop &, StringRef)> DestroyLoopCB) {
auto *ParentBB = TI.getParent();
BranchInst *BI = dyn_cast<BranchInst>(&TI);
SwitchInst *SI = BI ? nullptr : cast<SwitchInst>(&TI);

// We can only unswitch switches, conditional branches with an invariant
// condition, or combining invariant conditions with an instruction or
// partially invariant instructions.
assert((SI || (BI && BI->isConditional())) &&(static_cast <bool> ((SI || (BI && BI->isConditional
())) && "Can only unswitch switches and conditional branch!"
) ? void (0) : __assert_fail ("(SI || (BI && BI->isConditional())) && \"Can only unswitch switches and conditional branch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2065, __extension__
 __PRETTY_FUNCTION__))
       "Can only unswitch switches and conditional branch!")(static_cast <bool> ((SI || (BI && BI->isConditional
())) && "Can only unswitch switches and conditional branch!"
) ? void (0) : __assert_fail ("(SI || (BI && BI->isConditional())) && \"Can only unswitch switches and conditional branch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2065, __extension__
 __PRETTY_FUNCTION__));
bool PartiallyInvariant = !PartialIVInfo.InstToDuplicate.empty();
bool FullUnswitch =
    SI || (skipTrivialSelect(BI->getCondition()) == Invariants[0] &&
           !PartiallyInvariant);
if (FullUnswitch)
  assert(Invariants.size() == 1 &&(static_cast <bool> (Invariants.size() == 1 && "Cannot have other invariants with full unswitching!"
) ? void (0) : __assert_fail ("Invariants.size() == 1 && \"Cannot have other invariants with full unswitching!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2072, __extension__
 __PRETTY_FUNCTION__))
         "Cannot have other invariants with full unswitching!")(static_cast <bool> (Invariants.size() == 1 && "Cannot have other invariants with full unswitching!"
) ? void (0) : __assert_fail ("Invariants.size() == 1 && \"Cannot have other invariants with full unswitching!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2072, __extension__
 __PRETTY_FUNCTION__));
else
  assert(isa<Instruction>(skipTrivialSelect(BI->getCondition())) &&(static_cast <bool> (isa<Instruction>(skipTrivialSelect
(BI->getCondition())) && "Partial unswitching requires an instruction as the condition!"
) ? void (0) : __assert_fail ("isa<Instruction>(skipTrivialSelect(BI->getCondition())) && \"Partial unswitching requires an instruction as the condition!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2075, __extension__
 __PRETTY_FUNCTION__))
         "Partial unswitching requires an instruction as the condition!")(static_cast <bool> (isa<Instruction>(skipTrivialSelect
(BI->getCondition())) && "Partial unswitching requires an instruction as the condition!"
) ? void (0) : __assert_fail ("isa<Instruction>(skipTrivialSelect(BI->getCondition())) && \"Partial unswitching requires an instruction as the condition!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2075, __extension__
 __PRETTY_FUNCTION__));

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

// Constant and BBs tracking the cloned and continuing successor. When we are
// unswitching the entire condition, this can just be trivially chosen to
// unswitch towards `true`. However, when we are unswitching a set of
// invariants combined with `and` or `or` or partially invariant instructions,
// the combining operation determines the best direction to unswitch: we want
// to unswitch the direction that will collapse the branch.
bool Direction = true;
int ClonedSucc = 0;
if (!FullUnswitch) {
  Value *Cond = skipTrivialSelect(BI->getCondition());
  (void)Cond;
  assert(((match(Cond, m_LogicalAnd()) ^ match(Cond, m_LogicalOr())) ||(static_cast <bool> (((match(Cond, m_LogicalAnd()) ^ match
(Cond, m_LogicalOr())) || PartiallyInvariant) && "Only `or`, `and`, an `select`, partially invariant instructions "
 "can combine invariants being unswitched.") ? void (0) : __assert_fail
 ("((match(Cond, m_LogicalAnd()) ^ match(Cond, m_LogicalOr())) || PartiallyInvariant) && \"Only `or`, `and`, an `select`, partially invariant instructions \" \"can combine invariants being unswitched.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2094, __extension__
 __PRETTY_FUNCTION__))
          PartiallyInvariant) &&(static_cast <bool> (((match(Cond, m_LogicalAnd()) ^ match
(Cond, m_LogicalOr())) || PartiallyInvariant) && "Only `or`, `and`, an `select`, partially invariant instructions "
 "can combine invariants being unswitched.") ? void (0) : __assert_fail
 ("((match(Cond, m_LogicalAnd()) ^ match(Cond, m_LogicalOr())) || PartiallyInvariant) && \"Only `or`, `and`, an `select`, partially invariant instructions \" \"can combine invariants being unswitched.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2094, __extension__
 __PRETTY_FUNCTION__))
         "Only `or`, `and`, an `select`, partially invariant instructions "(static_cast <bool> (((match(Cond, m_LogicalAnd()) ^ match
(Cond, m_LogicalOr())) || PartiallyInvariant) && "Only `or`, `and`, an `select`, partially invariant instructions "
 "can combine invariants being unswitched.") ? void (0) : __assert_fail
 ("((match(Cond, m_LogicalAnd()) ^ match(Cond, m_LogicalOr())) || PartiallyInvariant) && \"Only `or`, `and`, an `select`, partially invariant instructions \" \"can combine invariants being unswitched.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2094, __extension__
 __PRETTY_FUNCTION__))
         "can combine invariants being unswitched.")(static_cast <bool> (((match(Cond, m_LogicalAnd()) ^ match
(Cond, m_LogicalOr())) || PartiallyInvariant) && "Only `or`, `and`, an `select`, partially invariant instructions "
 "can combine invariants being unswitched.") ? void (0) : __assert_fail
 ("((match(Cond, m_LogicalAnd()) ^ match(Cond, m_LogicalOr())) || PartiallyInvariant) && \"Only `or`, `and`, an `select`, partially invariant instructions \" \"can combine invariants being unswitched.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2094, __extension__
 __PRETTY_FUNCTION__));
  if (!match(Cond, m_LogicalOr())) {
    if (match(Cond, m_LogicalAnd()) ||
        (PartiallyInvariant && !PartialIVInfo.KnownValue->isOneValue())) {
      Direction = false;
      ClonedSucc = 1;
    }
  }
}

BasicBlock *RetainedSuccBB =
    BI ? BI->getSuccessor(1 - ClonedSucc) : SI->getDefaultDest();
SmallSetVector<BasicBlock *, 4> UnswitchedSuccBBs;
if (BI)
  UnswitchedSuccBBs.insert(BI->getSuccessor(ClonedSucc));
else
  for (auto Case : SI->cases())
    if (Case.getCaseSuccessor() != RetainedSuccBB)
      UnswitchedSuccBBs.insert(Case.getCaseSuccessor());

assert(!UnswitchedSuccBBs.count(RetainedSuccBB) &&(static_cast <bool> (!UnswitchedSuccBBs.count(RetainedSuccBB
) && "Should not unswitch the same successor we are retaining!"
) ? void (0) : __assert_fail ("!UnswitchedSuccBBs.count(RetainedSuccBB) && \"Should not unswitch the same successor we are retaining!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2115, __extension__
 __PRETTY_FUNCTION__))
       "Should not unswitch the same successor we are retaining!")(static_cast <bool> (!UnswitchedSuccBBs.count(RetainedSuccBB
) && "Should not unswitch the same successor we are retaining!"
) ? void (0) : __assert_fail ("!UnswitchedSuccBBs.count(RetainedSuccBB) && \"Should not unswitch the same successor we are retaining!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2115, __extension__
 __PRETTY_FUNCTION__));

// The branch should be in this exact loop. Any inner loop's invariant branch
// should be handled by unswitching that inner loop. The caller of this
// routine should filter out any candidates that remain (but were skipped for
// whatever reason).
assert(LI.getLoopFor(ParentBB) == &L && "Branch in an inner loop!")(static_cast <bool> (LI.getLoopFor(ParentBB) == &L &&
 "Branch in an inner loop!") ? void (0) : __assert_fail ("LI.getLoopFor(ParentBB) == &L && \"Branch in an inner loop!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2121, __extension__
 __PRETTY_FUNCTION__));

// Compute the parent loop now before we start hacking on things.
Loop *ParentL = L.getParentLoop();
// Get blocks in RPO order for MSSA update, before changing the CFG.
LoopBlocksRPO LBRPO(&L);
if (MSSAU)
  LBRPO.perform(&LI);

// Compute the outer-most loop containing one of our exit blocks. This is the
// furthest up our loopnest which can be mutated, which we will use below to
// update things.
Loop *OuterExitL = &L;
for (auto *ExitBB : ExitBlocks) {
  Loop *NewOuterExitL = LI.getLoopFor(ExitBB);
  if (!NewOuterExitL) {
    // We exited the entire nest with this block, so we're done.
    OuterExitL = nullptr;
    break;
  }
  if (NewOuterExitL != OuterExitL && NewOuterExitL->contains(OuterExitL))
    OuterExitL = NewOuterExitL;
}

// At this point, we're definitely going to unswitch something so invalidate
// any cached information in ScalarEvolution for the outer most loop
// containing an exit block and all nested loops.
if (SE) {
  if (OuterExitL)
    SE->forgetLoop(OuterExitL);
  else
    SE->forgetTopmostLoop(&L);
}

bool InsertFreeze = false;
if (FreezeLoopUnswitchCond) {
  ICFLoopSafetyInfo SafetyInfo;
  SafetyInfo.computeLoopSafetyInfo(&L);
  InsertFreeze = !SafetyInfo.isGuaranteedToExecute(TI, &DT, &L);
}

// If the edge from this terminator to a successor dominates that successor,
// store a map from each block in its dominator subtree to it. This lets us
// tell when cloning for a particular successor if a block is dominated by
// some *other* successor with a single data structure. We use this to
// significantly reduce cloning.
SmallDenseMap<BasicBlock *, BasicBlock *, 16> DominatingSucc;
for (auto *SuccBB : llvm::concat<BasicBlock *const>(
         makeArrayRef(RetainedSuccBB), UnswitchedSuccBBs))
  if (SuccBB->getUniquePredecessor() ||
      llvm::all_of(predecessors(SuccBB), [&](BasicBlock *PredBB) {
        return PredBB == ParentBB || DT.dominates(SuccBB, PredBB);
      }))
    visitDomSubTree(DT, SuccBB, [&](BasicBlock *BB) {
      DominatingSucc[BB] = SuccBB;
      return true;
    });

// Split the preheader, so that we know that there is a safe place to insert
// the conditional branch. We will change the preheader to have a conditional
// branch on LoopCond. The original preheader will become the split point
// between the unswitched versions, and we will have a new preheader for the
// original loop.
BasicBlock *SplitBB = L.getLoopPreheader();
BasicBlock *LoopPH = SplitEdge(SplitBB, L.getHeader(), &DT, &LI, MSSAU);

// Keep track of the dominator tree updates needed.
SmallVector<DominatorTree::UpdateType, 4> DTUpdates;

// Clone the loop for each unswitched successor.
SmallVector<std::unique_ptr<ValueToValueMapTy>, 4> VMaps;
VMaps.reserve(UnswitchedSuccBBs.size());
SmallDenseMap<BasicBlock *, BasicBlock *, 4> ClonedPHs;
for (auto *SuccBB : UnswitchedSuccBBs) {
  VMaps.emplace_back(new ValueToValueMapTy());
  ClonedPHs[SuccBB] = buildClonedLoopBlocks(
      L, LoopPH, SplitBB, ExitBlocks, ParentBB, SuccBB, RetainedSuccBB,
      DominatingSucc, *VMaps.back(), DTUpdates, AC, DT, LI, MSSAU);
}

// Drop metadata if we may break its semantics by moving this instr into the
// split block.
if (TI.getMetadata(LLVMContext::MD_make_implicit)) {
  if (DropNonTrivialImplicitNullChecks)
    // Do not spend time trying to understand if we can keep it, just drop it
    // to save compile time.
    TI.setMetadata(LLVMContext::MD_make_implicit, nullptr);
  else {
    // It is only legal to preserve make.implicit metadata if we are
    // guaranteed no reach implicit null check after following this branch.
    ICFLoopSafetyInfo SafetyInfo;
    SafetyInfo.computeLoopSafetyInfo(&L);
    if (!SafetyInfo.isGuaranteedToExecute(TI, &DT, &L))
      TI.setMetadata(LLVMContext::MD_make_implicit, nullptr);
  }
}

// The stitching of the branched code back together depends on whether we're
// doing full unswitching or not with the exception that we always want to
// nuke the initial terminator placed in the split block.
SplitBB->getTerminator()->eraseFromParent();
if (FullUnswitch) {
  // Splice the terminator from the original loop and rewrite its
  // successors.
  SplitBB->getInstList().splice(SplitBB->end(), ParentBB->getInstList(), TI);

  // Keep a clone of the terminator for MSSA updates.
  Instruction *NewTI = TI.clone();
  ParentBB->getInstList().push_back(NewTI);

  // First wire up the moved terminator to the preheaders.
  if (BI) {
    BasicBlock *ClonedPH = ClonedPHs.begin()->second;
    BI->setSuccessor(ClonedSucc, ClonedPH);
    BI->setSuccessor(1 - ClonedSucc, LoopPH);
    Value *Cond = skipTrivialSelect(BI->getCondition());
    if (InsertFreeze) {
      if (!isGuaranteedNotToBeUndefOrPoison(Cond, &AC, BI, &DT))
        Cond = new FreezeInst(Cond, Cond->getName() + ".fr", BI);
    }
    BI->setCondition(Cond);
    DTUpdates.push_back({DominatorTree::Insert, SplitBB, ClonedPH});
  } else {
    assert(SI && "Must either be a branch or switch!")(static_cast <bool> (SI && "Must either be a branch or switch!"
) ? void (0) : __assert_fail ("SI && \"Must either be a branch or switch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2244, __extension__
 __PRETTY_FUNCTION__));

    // Walk the cases and directly update their successors.
    assert(SI->getDefaultDest() == RetainedSuccBB &&(static_cast <bool> (SI->getDefaultDest() == RetainedSuccBB
 && "Not retaining default successor!") ? void (0) : __assert_fail
 ("SI->getDefaultDest() == RetainedSuccBB && \"Not retaining default successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2248, __extension__
 __PRETTY_FUNCTION__))
           "Not retaining default successor!")(static_cast <bool> (SI->getDefaultDest() == RetainedSuccBB
 && "Not retaining default successor!") ? void (0) : __assert_fail
 ("SI->getDefaultDest() == RetainedSuccBB && \"Not retaining default successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2248, __extension__
 __PRETTY_FUNCTION__));
    SI->setDefaultDest(LoopPH);
    for (const auto &Case : SI->cases())
      if (Case.getCaseSuccessor() == RetainedSuccBB)
        Case.setSuccessor(LoopPH);
      else
        Case.setSuccessor(ClonedPHs.find(Case.getCaseSuccessor())->second);

    if (InsertFreeze) {
      auto Cond = SI->getCondition();
      if (!isGuaranteedNotToBeUndefOrPoison(Cond, &AC, SI, &DT))
        SI->setCondition(new FreezeInst(Cond, Cond->getName() + ".fr", SI));
    }
    // We need to use the set to populate domtree updates as even when there
    // are multiple cases pointing at the same successor we only want to
    // remove and insert one edge in the domtree.
    for (BasicBlock *SuccBB : UnswitchedSuccBBs)
      DTUpdates.push_back(
          {DominatorTree::Insert, SplitBB, ClonedPHs.find(SuccBB)->second});
  }

  if (MSSAU) {
    DT.applyUpdates(DTUpdates);
    DTUpdates.clear();

    // Remove all but one edge to the retained block and all unswitched
    // blocks. This is to avoid having duplicate entries in the cloned Phis,
    // when we know we only keep a single edge for each case.
    MSSAU->removeDuplicatePhiEdgesBetween(ParentBB, RetainedSuccBB);
    for (BasicBlock *SuccBB : UnswitchedSuccBBs)
      MSSAU->removeDuplicatePhiEdgesBetween(ParentBB, SuccBB);

    for (auto &VMap : VMaps)
      MSSAU->updateForClonedLoop(LBRPO, ExitBlocks, *VMap,
                                 /*IgnoreIncomingWithNoClones=*/true);
    MSSAU->updateExitBlocksForClonedLoop(ExitBlocks, VMaps, DT);

    // Remove all edges to unswitched blocks.
    for (BasicBlock *SuccBB : UnswitchedSuccBBs)
      MSSAU->removeEdge(ParentBB, SuccBB);
  }

  // Now unhook the successor relationship as we'll be replacing
  // the terminator with a direct branch. This is much simpler for branches
  // than switches so we handle those first.
  if (BI) {
    // Remove the parent as a predecessor of the unswitched successor.
    assert(UnswitchedSuccBBs.size() == 1 &&(static_cast <bool> (UnswitchedSuccBBs.size() == 1 &&
 "Only one possible unswitched block for a branch!") ? void (
0) : __assert_fail ("UnswitchedSuccBBs.size() == 1 && \"Only one possible unswitched block for a branch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2296, __extension__
 __PRETTY_FUNCTION__))
           "Only one possible unswitched block for a branch!")(static_cast <bool> (UnswitchedSuccBBs.size() == 1 &&
 "Only one possible unswitched block for a branch!") ? void (
0) : __assert_fail ("UnswitchedSuccBBs.size() == 1 && \"Only one possible unswitched block for a branch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2296, __extension__
 __PRETTY_FUNCTION__));
    BasicBlock *UnswitchedSuccBB = *UnswitchedSuccBBs.begin();
    UnswitchedSuccBB->removePredecessor(ParentBB,
                                        /*KeepOneInputPHIs*/ true);
    DTUpdates.push_back({DominatorTree::Delete, ParentBB, UnswitchedSuccBB});
  } else {
    // Note that we actually want to remove the parent block as a predecessor
    // of *every* case successor. The case successor is either unswitched,
    // completely eliminating an edge from the parent to that successor, or it
    // is a duplicate edge to the retained successor as the retained successor
    // is always the default successor and as we'll replace this with a direct
    // branch we no longer need the duplicate entries in the PHI nodes.
    SwitchInst *NewSI = cast<SwitchInst>(NewTI);
    assert(NewSI->getDefaultDest() == RetainedSuccBB &&(static_cast <bool> (NewSI->getDefaultDest() == RetainedSuccBB
 && "Not retaining default successor!") ? void (0) : __assert_fail
 ("NewSI->getDefaultDest() == RetainedSuccBB && \"Not retaining default successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2310, __extension__
 __PRETTY_FUNCTION__))
           "Not retaining default successor!")(static_cast <bool> (NewSI->getDefaultDest() == RetainedSuccBB
 && "Not retaining default successor!") ? void (0) : __assert_fail
 ("NewSI->getDefaultDest() == RetainedSuccBB && \"Not retaining default successor!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2310, __extension__
 __PRETTY_FUNCTION__));
    for (const auto &Case : NewSI->cases())
      Case.getCaseSuccessor()->removePredecessor(
          ParentBB,
          /*KeepOneInputPHIs*/ true);

    // We need to use the set to populate domtree updates as even when there
    // are multiple cases pointing at the same successor we only want to
    // remove and insert one edge in the domtree.
    for (BasicBlock *SuccBB : UnswitchedSuccBBs)
      DTUpdates.push_back({DominatorTree::Delete, ParentBB, SuccBB});
  }

  // After MSSAU update, remove the cloned terminator instruction NewTI.
  ParentBB->getTerminator()->eraseFromParent();

  // Create a new unconditional branch to the continuing block (as opposed to
  // the one cloned).
  BranchInst::Create(RetainedSuccBB, ParentBB);
} else {
  assert(BI && "Only branches have partial unswitching.")(static_cast <bool> (BI && "Only branches have partial unswitching."
) ? void (0) : __assert_fail ("BI && \"Only branches have partial unswitching.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2330, __extension__
 __PRETTY_FUNCTION__));
  assert(UnswitchedSuccBBs.size() == 1 &&(static_cast <bool> (UnswitchedSuccBBs.size() == 1 &&
 "Only one possible unswitched block for a branch!") ? void (
0) : __assert_fail ("UnswitchedSuccBBs.size() == 1 && \"Only one possible unswitched block for a branch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2332, __extension__
 __PRETTY_FUNCTION__))
         "Only one possible unswitched block for a branch!")(static_cast <bool> (UnswitchedSuccBBs.size() == 1 &&
 "Only one possible unswitched block for a branch!") ? void (
0) : __assert_fail ("UnswitchedSuccBBs.size() == 1 && \"Only one possible unswitched block for a branch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2332, __extension__
 __PRETTY_FUNCTION__));
  BasicBlock *ClonedPH = ClonedPHs.begin()->second;
  // When doing a partial unswitch, we have to do a bit more work to build up
  // the branch in the split block.
  if (PartiallyInvariant)
    buildPartialInvariantUnswitchConditionalBranch(
        *SplitBB, Invariants, Direction, *ClonedPH, *LoopPH, L, MSSAU);
  else {
    buildPartialUnswitchConditionalBranch(
        *SplitBB, Invariants, Direction, *ClonedPH, *LoopPH,
        FreezeLoopUnswitchCond, BI, &AC, DT);
  }
  DTUpdates.push_back({DominatorTree::Insert, SplitBB, ClonedPH});

  if (MSSAU) {
    DT.applyUpdates(DTUpdates);
    DTUpdates.clear();

    // Perform MSSA cloning updates.
    for (auto &VMap : VMaps)
      MSSAU->updateForClonedLoop(LBRPO, ExitBlocks, *VMap,
                                 /*IgnoreIncomingWithNoClones=*/true);
    MSSAU->updateExitBlocksForClonedLoop(ExitBlocks, VMaps, DT);
  }
}

// Apply the updates accumulated above to get an up-to-date dominator tree.
DT.applyUpdates(DTUpdates);

// Now that we have an accurate dominator tree, first delete the dead cloned
// blocks so that we can accurately build any cloned loops. It is important to
// not delete the blocks from the original loop yet because we still want to
// reference the original loop to understand the cloned loop's structure.
deleteDeadClonedBlocks(L, ExitBlocks, VMaps, DT, MSSAU);

// Build the cloned loop structure itself. This may be substantially
// different from the original structure due to the simplified CFG. This also
// handles inserting all the cloned blocks into the correct loops.
SmallVector<Loop *, 4> NonChildClonedLoops;
for (std::unique_ptr<ValueToValueMapTy> &VMap : VMaps)
  buildClonedLoops(L, ExitBlocks, *VMap, LI, NonChildClonedLoops);

// Now that our cloned loops have been built, we can update the original loop.
// First we delete the dead blocks from it and then we rebuild the loop
// structure taking these deletions into account.
deleteDeadBlocksFromLoop(L, ExitBlocks, DT, LI, MSSAU, DestroyLoopCB);

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

SmallVector<Loop *, 4> HoistedLoops;
bool IsStillLoop = rebuildLoopAfterUnswitch(L, ExitBlocks, LI, HoistedLoops);

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

// This transformation has a high risk of corrupting the dominator tree, and
// the below steps to rebuild loop structures will result in hard to debug
// errors in that case so verify that the dominator tree is sane first.
// FIXME: Remove this when the bugs stop showing up and rely on existing
// verification steps.
assert(DT.verify(DominatorTree::VerificationLevel::Fast))(static_cast <bool> (DT.verify(DominatorTree::VerificationLevel
::Fast)) ? void (0) : __assert_fail ("DT.verify(DominatorTree::VerificationLevel::Fast)"
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2393, __extension__
 __PRETTY_FUNCTION__));

if (BI && !PartiallyInvariant) {
  // If we unswitched a branch which collapses the condition to a known
  // constant we want to replace all the uses of the invariants within both
  // the original and cloned blocks. We do this here so that we can use the
  // now updated dominator tree to identify which side the users are on.
  assert(UnswitchedSuccBBs.size() == 1 &&(static_cast <bool> (UnswitchedSuccBBs.size() == 1 &&
 "Only one possible unswitched block for a branch!") ? void (
0) : __assert_fail ("UnswitchedSuccBBs.size() == 1 && \"Only one possible unswitched block for a branch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2401, __extension__
 __PRETTY_FUNCTION__))
         "Only one possible unswitched block for a branch!")(static_cast <bool> (UnswitchedSuccBBs.size() == 1 &&
 "Only one possible unswitched block for a branch!") ? void (
0) : __assert_fail ("UnswitchedSuccBBs.size() == 1 && \"Only one possible unswitched block for a branch!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2401, __extension__
 __PRETTY_FUNCTION__));
  BasicBlock *ClonedPH = ClonedPHs.begin()->second;

  // When considering multiple partially-unswitched invariants
  // we cant just go replace them with constants in both branches.
  //
  // For 'AND' we infer that true branch ("continue") means true
  // for each invariant operand.
  // For 'OR' we can infer that false branch ("continue") means false
  // for each invariant operand.
  // So it happens that for multiple-partial case we dont replace
  // in the unswitched branch.
  bool ReplaceUnswitched =
      FullUnswitch || (Invariants.size() == 1) || PartiallyInvariant;

  ConstantInt *UnswitchedReplacement =
      Direction ? ConstantInt::getTrue(BI->getContext())
                : ConstantInt::getFalse(BI->getContext());
  ConstantInt *ContinueReplacement =
      Direction ? ConstantInt::getFalse(BI->getContext())
                : ConstantInt::getTrue(BI->getContext());
  for (Value *Invariant : Invariants) {
    assert(!isa<Constant>(Invariant) &&(static_cast <bool> (!isa<Constant>(Invariant) &&
 "Should not be replacing constant values!") ? void (0) : __assert_fail
 ("!isa<Constant>(Invariant) && \"Should not be replacing constant values!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2424, __extension__
 __PRETTY_FUNCTION__))
           "Should not be replacing constant values!")(static_cast <bool> (!isa<Constant>(Invariant) &&
 "Should not be replacing constant values!") ? void (0) : __assert_fail
 ("!isa<Constant>(Invariant) && \"Should not be replacing constant values!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2424, __extension__
 __PRETTY_FUNCTION__));
    // Use make_early_inc_range here as set invalidates the iterator.
    for (Use &U : llvm::make_early_inc_range(Invariant->uses())) {
      Instruction *UserI = dyn_cast<Instruction>(U.getUser());
      if (!UserI)
        continue;

      // Replace it with the 'continue' side if in the main loop body, and the
      // unswitched if in the cloned blocks.
      if (DT.dominates(LoopPH, UserI->getParent()))
        U.set(ContinueReplacement);
      else if (ReplaceUnswitched &&
               DT.dominates(ClonedPH, UserI->getParent()))
        U.set(UnswitchedReplacement);
    }
  }
}

// We can change which blocks are exit blocks of all the cloned sibling
// loops, the current loop, and any parent loops which shared exit blocks
// with the current loop. As a consequence, we need to re-form LCSSA for
// them. But we shouldn't need to re-form LCSSA for any child loops.
// FIXME: This could be made more efficient by tracking which exit blocks are
// new, and focusing on them, but that isn't likely to be necessary.
//
// In order to reasonably rebuild LCSSA we need to walk inside-out across the
// loop nest and update every loop that could have had its exits changed. We
// also need to cover any intervening loops. We add all of these loops to
// a list and sort them by loop depth to achieve this without updating
// unnecessary loops.
auto UpdateLoop = [&](Loop &UpdateL) {
2455#ifndef NDEBUG
  UpdateL.verifyLoop();
  for (Loop *ChildL : UpdateL) {
    ChildL->verifyLoop();
    assert(ChildL->isRecursivelyLCSSAForm(DT, LI) &&(static_cast <bool> (ChildL->isRecursivelyLCSSAForm(
DT, LI) && "Perturbed a child loop's LCSSA form!") ? void
 (0) : __assert_fail ("ChildL->isRecursivelyLCSSAForm(DT, LI) && \"Perturbed a child loop's LCSSA form!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2460, __extension__
 __PRETTY_FUNCTION__))
           "Perturbed a child loop's LCSSA form!")(static_cast <bool> (ChildL->isRecursivelyLCSSAForm(
DT, LI) && "Perturbed a child loop's LCSSA form!") ? void
 (0) : __assert_fail ("ChildL->isRecursivelyLCSSAForm(DT, LI) && \"Perturbed a child loop's LCSSA form!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2460, __extension__
 __PRETTY_FUNCTION__));
  }
2462#endif
  // First build LCSSA for this loop so that we can preserve it when
  // forming dedicated exits. We don't want to perturb some other loop's
  // LCSSA while doing that CFG edit.
  formLCSSA(UpdateL, DT, &LI, SE);

  // For loops reached by this loop's original exit blocks we may
  // introduced new, non-dedicated exits. At least try to re-form dedicated
  // exits for these loops. This may fail if they couldn't have dedicated
  // exits to start with.
  formDedicatedExitBlocks(&UpdateL, &DT, &LI, MSSAU, /*PreserveLCSSA*/ true);
};

// For non-child cloned loops and hoisted loops, we just need to update LCSSA
// and we can do it in any order as they don't nest relative to each other.
//
// Also check if any of the loops we have updated have become top-level loops
// as that will necessitate widening the outer loop scope.
for (Loop *UpdatedL :
     llvm::concat<Loop *>(NonChildClonedLoops, HoistedLoops)) {
  UpdateLoop(*UpdatedL);
  if (UpdatedL->isOutermost())
    OuterExitL = nullptr;
}
if (IsStillLoop) {
  UpdateLoop(L);
  if (L.isOutermost())
    OuterExitL = nullptr;
}

// If the original loop had exit blocks, walk up through the outer most loop
// of those exit blocks to update LCSSA and form updated dedicated exits.
if (OuterExitL != &L)
  for (Loop *OuterL = ParentL; OuterL != OuterExitL;
       OuterL = OuterL->getParentLoop())
    UpdateLoop(*OuterL);

2499#ifndef NDEBUG
// Verify the entire loop structure to catch any incorrect updates before we
// progress in the pass pipeline.
LI.verify(DT);
2503#endif

// Now that we've unswitched something, make callbacks to report the changes.
// For that we need to merge together the updated loops and the cloned loops
// and check whether the original loop survived.
SmallVector<Loop *, 4> SibLoops;
for (Loop *UpdatedL : llvm::concat<Loop *>(NonChildClonedLoops, HoistedLoops))
  if (UpdatedL->getParentLoop() == ParentL)
    SibLoops.push_back(UpdatedL);
UnswitchCB(IsStillLoop, PartiallyInvariant, SibLoops);

if (MSSAU && VerifyMemorySSA)
  MSSAU->getMemorySSA()->verifyMemorySSA();

if (BI)
  ++NumBranches;
else
  ++NumSwitches;
2521}

2523/// Recursively compute the cost of a dominator subtree based on the per-block
2524/// cost map provided.
2525///
2526/// The recursive computation is memozied into the provided DT-indexed cost map
2527/// to allow querying it for most nodes in the domtree without it becoming
2528/// quadratic.
2529static InstructionCost computeDomSubtreeCost(
  DomTreeNode &N,
  const SmallDenseMap<BasicBlock *, InstructionCost, 4> &BBCostMap,
  SmallDenseMap<DomTreeNode *, InstructionCost, 4> &DTCostMap) {
// Don't accumulate cost (or recurse through) blocks not in our block cost
// map and thus not part of the duplication cost being considered.
auto BBCostIt = BBCostMap.find(N.getBlock());
if (BBCostIt == BBCostMap.end())
  return 0;

// Lookup this node to see if we already computed its cost.
auto DTCostIt = DTCostMap.find(&N);
if (DTCostIt != DTCostMap.end())
  return DTCostIt->second;

// If not, we have to compute it. We can't use insert above and update
// because computing the cost may insert more things into the map.
InstructionCost Cost = std::accumulate(
    N.begin(), N.end(), BBCostIt->second,
    [&](InstructionCost Sum, DomTreeNode *ChildN) -> InstructionCost {
      return Sum + computeDomSubtreeCost(*ChildN, BBCostMap, DTCostMap);
    });
bool Inserted = DTCostMap.insert({&N, Cost}).second;
(void)Inserted;
assert(Inserted && "Should not insert a node while visiting children!")(static_cast <bool> (Inserted && "Should not insert a node while visiting children!"
) ? void (0) : __assert_fail ("Inserted && \"Should not insert a node while visiting children!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2553, __extension__
 __PRETTY_FUNCTION__));
return Cost;
2555}

2557/// Turns a llvm.experimental.guard intrinsic into implicit control flow branch,
2558/// making the following replacement:
2559///
2560///   --code before guard--
2561///   call void (i1, ...) @llvm.experimental.guard(i1 %cond) [ "deopt"() ]
2562///   --code after guard--
2563///
2564/// into
2565///
2566///   --code before guard--
2567///   br i1 %cond, label %guarded, label %deopt
2568///
2569/// guarded:
2570///   --code after guard--
2571///
2572/// deopt:
2573///   call void (i1, ...) @llvm.experimental.guard(i1 false) [ "deopt"() ]
2574///   unreachable
2575///
2576/// It also makes all relevant DT and LI updates, so that all structures are in
2577/// valid state after this transform.
2578static BranchInst *
2579turnGuardIntoBranch(IntrinsicInst *GI, Loop &L,
                  SmallVectorImpl<BasicBlock *> &ExitBlocks,
                  DominatorTree &DT, LoopInfo &LI, MemorySSAUpdater *MSSAU) {
SmallVector<DominatorTree::UpdateType, 4> DTUpdates;
LLVM_DEBUG(dbgs() << "Turning " << *GI << " into a branch.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Turning " <<
 *GI << " into a branch.\n"; } } while (false);
BasicBlock *CheckBB = GI->getParent();

if (MSSAU && VerifyMemorySSA)
   MSSAU->getMemorySSA()->verifyMemorySSA();

// Remove all CheckBB's successors from DomTree. A block can be seen among
// successors more than once, but for DomTree it should be added only once.
SmallPtrSet<BasicBlock *, 4> Successors;
for (auto *Succ : successors(CheckBB))
  if (Successors.insert(Succ).second)
    DTUpdates.push_back({DominatorTree::Delete, CheckBB, Succ});

Instruction *DeoptBlockTerm =
    SplitBlockAndInsertIfThen(GI->getArgOperand(0), GI, true);
BranchInst *CheckBI = cast<BranchInst>(CheckBB->getTerminator());
// SplitBlockAndInsertIfThen inserts control flow that branches to
// DeoptBlockTerm if the condition is true.  We want the opposite.
CheckBI->swapSuccessors();

BasicBlock *GuardedBlock = CheckBI->getSuccessor(0);
GuardedBlock->setName("guarded");
CheckBI->getSuccessor(1)->setName("deopt");
BasicBlock *DeoptBlock = CheckBI->getSuccessor(1);

// We now have a new exit block.
ExitBlocks.push_back(CheckBI->getSuccessor(1));

if (MSSAU)
  MSSAU->moveAllAfterSpliceBlocks(CheckBB, GuardedBlock, GI);

GI->moveBefore(DeoptBlockTerm);
GI->setArgOperand(0, ConstantInt::getFalse(GI->getContext()));

// Add new successors of CheckBB into DomTree.
for (auto *Succ : successors(CheckBB))
  DTUpdates.push_back({DominatorTree::Insert, CheckBB, Succ});

// Now the blocks that used to be CheckBB's successors are GuardedBlock's
// successors.
for (auto *Succ : Successors)
  DTUpdates.push_back({DominatorTree::Insert, GuardedBlock, Succ});

// Make proper changes to DT.
DT.applyUpdates(DTUpdates);
// Inform LI of a new loop block.
L.addBasicBlockToLoop(GuardedBlock, LI);

if (MSSAU) {
  MemoryDef *MD = cast<MemoryDef>(MSSAU->getMemorySSA()->getMemoryAccess(GI));
  MSSAU->moveToPlace(MD, DeoptBlock, MemorySSA::BeforeTerminator);
  if (VerifyMemorySSA)
    MSSAU->getMemorySSA()->verifyMemorySSA();
}

++NumGuards;
return CheckBI;
2640}

2642/// Cost multiplier is a way to limit potentially exponential behavior
2643/// of loop-unswitch. Cost is multipied in proportion of 2^number of unswitch
2644/// candidates available. Also accounting for the number of "sibling" loops with
2645/// the idea to account for previous unswitches that already happened on this
2646/// cluster of loops. There was an attempt to keep this formula simple,
2647/// just enough to limit the worst case behavior. Even if it is not that simple
2648/// now it is still not an attempt to provide a detailed heuristic size
2649/// prediction.
2650///
2651/// TODO: Make a proper accounting of "explosion" effect for all kinds of
2652/// unswitch candidates, making adequate predictions instead of wild guesses.
2653/// That requires knowing not just the number of "remaining" candidates but
2654/// also costs of unswitching for each of these candidates.
2655static int CalculateUnswitchCostMultiplier(
  Instruction &TI, Loop &L, LoopInfo &LI, DominatorTree &DT,
  ArrayRef<std::pair<Instruction *, TinyPtrVector<Value *>>>
      UnswitchCandidates) {

// Guards and other exiting conditions do not contribute to exponential
// explosion as soon as they dominate the latch (otherwise there might be
// another path to the latch remaining that does not allow to eliminate the
// loop copy on unswitch).
BasicBlock *Latch = L.getLoopLatch();
BasicBlock *CondBlock = TI.getParent();
if (DT.dominates(CondBlock, Latch) &&
    (isGuard(&TI) ||
     llvm::count_if(successors(&TI), [&L](BasicBlock *SuccBB) {
       return L.contains(SuccBB);
     }) <= 1)) {
  NumCostMultiplierSkipped++;
  return 1;
}

auto *ParentL = L.getParentLoop();
int SiblingsCount = (ParentL ? ParentL->getSubLoopsVector().size()
                             : std::distance(LI.begin(), LI.end()));
// Count amount of clones that all the candidates might cause during
// unswitching. Branch/guard counts as 1, switch counts as log2 of its cases.
int UnswitchedClones = 0;
for (auto Candidate : UnswitchCandidates) {
  Instruction *CI = Candidate.first;
  BasicBlock *CondBlock = CI->getParent();
  bool SkipExitingSuccessors = DT.dominates(CondBlock, Latch);
  if (isGuard(CI)) {
    if (!SkipExitingSuccessors)
      UnswitchedClones++;
    continue;
  }
  int NonExitingSuccessors = llvm::count_if(
      successors(CondBlock), [SkipExitingSuccessors, &L](BasicBlock *SuccBB) {
        return !SkipExitingSuccessors || L.contains(SuccBB);
      });
  UnswitchedClones += Log2_32(NonExitingSuccessors);
}

// Ignore up to the "unscaled candidates" number of unswitch candidates
// when calculating the power-of-two scaling of the cost. The main idea
// with this control is to allow a small number of unswitches to happen
// and rely more on siblings multiplier (see below) when the number
// of candidates is small.
unsigned ClonesPower =
    std::max(UnswitchedClones - (int)UnswitchNumInitialUnscaledCandidates, 0);

// Allowing top-level loops to spread a bit more than nested ones.
int SiblingsMultiplier =
    std::max((ParentL ? SiblingsCount
                      : SiblingsCount / (int)UnswitchSiblingsToplevelDiv),
             1);
// Compute the cost multiplier in a way that won't overflow by saturating
// at an upper bound.
int CostMultiplier;
if (ClonesPower > Log2_32(UnswitchThreshold) ||
    SiblingsMultiplier > UnswitchThreshold)
  CostMultiplier = UnswitchThreshold;
else
  CostMultiplier = std::min(SiblingsMultiplier * (1 << ClonesPower),
                            (int)UnswitchThreshold);

LLVM_DEBUG(dbgs() << "  Computed multiplier  " << CostMultiplierdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Computed multiplier  "
 << CostMultiplier << " (siblings " << SiblingsMultiplier
 << " * clones " << (1 << ClonesPower) <<
 ")" << " for unswitch candidate: " << TI <<
 "\n"; } } while (false)
                  << " (siblings " << SiblingsMultiplier << " * clones "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Computed multiplier  "
 << CostMultiplier << " (siblings " << SiblingsMultiplier
 << " * clones " << (1 << ClonesPower) <<
 ")" << " for unswitch candidate: " << TI <<
 "\n"; } } while (false)
                  << (1 << ClonesPower) << ")"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Computed multiplier  "
 << CostMultiplier << " (siblings " << SiblingsMultiplier
 << " * clones " << (1 << ClonesPower) <<
 ")" << " for unswitch candidate: " << TI <<
 "\n"; } } while (false)
                  << " for unswitch candidate: " << TI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Computed multiplier  "
 << CostMultiplier << " (siblings " << SiblingsMultiplier
 << " * clones " << (1 << ClonesPower) <<
 ")" << " for unswitch candidate: " << TI <<
 "\n"; } } while (false);
return CostMultiplier;
2725}

2727static bool unswitchBestCondition(
  Loop &L, DominatorTree &DT, LoopInfo &LI, AssumptionCache &AC,
  AAResults &AA, TargetTransformInfo &TTI,
  function_ref<void(bool, bool, ArrayRef<Loop *>)> UnswitchCB,
  ScalarEvolution *SE, MemorySSAUpdater *MSSAU,
  function_ref<void(Loop &, StringRef)> DestroyLoopCB) {
// Collect all invariant conditions within this loop (as opposed to an inner
// loop which would be handled when visiting that inner loop).
SmallVector<std::pair<Instruction *, TinyPtrVector<Value *>>, 4>
    UnswitchCandidates;

// Whether or not we should also collect guards in the loop.
bool CollectGuards = false;
if (UnswitchGuards) {
1
Assuming the condition is false→
2
←
Taking false branch→
  auto *GuardDecl = L.getHeader()->getParent()->getParent()->getFunction(
      Intrinsic::getName(Intrinsic::experimental_guard));
  if (GuardDecl && !GuardDecl->use_empty())
    CollectGuards = true;
}

IVConditionInfo PartialIVInfo;
3
←
Calling implicit default constructor for 'IVConditionInfo'→
5
←
Returning from default constructor for 'IVConditionInfo'→
for (auto *BB : L.blocks()) {
6
←
Assuming '__begin1' is equal to '__end1'→
  if (LI.getLoopFor(BB) != &L)
    continue;

  if (CollectGuards)
    for (auto &I : *BB)
      if (isGuard(&I)) {
        auto *Cond = cast<IntrinsicInst>(&I)->getArgOperand(0);
        // TODO: Support AND, OR conditions and partial unswitching.
        if (!isa<Constant>(Cond) && L.isLoopInvariant(Cond))
          UnswitchCandidates.push_back({&I, {Cond}});
      }

  if (auto *SI = dyn_cast<SwitchInst>(BB->getTerminator())) {
    // We can only consider fully loop-invariant switch conditions as we need
    // to completely eliminate the switch after unswitching.
    if (!isa<Constant>(SI->getCondition()) &&
        L.isLoopInvariant(SI->getCondition()) && !BB->getUniqueSuccessor())
      UnswitchCandidates.push_back({SI, {SI->getCondition()}});
    continue;
  }

  auto *BI = dyn_cast<BranchInst>(BB->getTerminator());
  if (!BI || !BI->isConditional() || isa<Constant>(BI->getCondition()) ||
      BI->getSuccessor(0) == BI->getSuccessor(1))
    continue;

  Value *Cond = skipTrivialSelect(BI->getCondition());
  if (isa<Constant>(Cond))
    continue;

  if (L.isLoopInvariant(Cond)) {
    UnswitchCandidates.push_back({BI, {Cond}});
    continue;
  }

  Instruction &CondI = *cast<Instruction>(Cond);
  if (match(&CondI, m_CombineOr(m_LogicalAnd(), m_LogicalOr()))) {
    TinyPtrVector<Value *> Invariants =
        collectHomogenousInstGraphLoopInvariants(L, CondI, LI);
    if (Invariants.empty())
      continue;

    UnswitchCandidates.push_back({BI, std::move(Invariants)});
    continue;
  }
}

Instruction *PartialIVCondBranch = nullptr;
if (MSSAU && !findOptionMDForLoop(&L, "llvm.loop.unswitch.partial.disable") &&
7
←
Assuming 'MSSAU' is null→
    !any_of(UnswitchCandidates, [&L](auto &TerminatorAndInvariants) {
      return TerminatorAndInvariants.first == L.getHeader()->getTerminator();
    })) {
  MemorySSA *MSSA = MSSAU->getMemorySSA();
  if (auto Info = hasPartialIVCondition(L, MSSAThreshold, *MSSA, AA)) {
    LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "simple-loop-unswitch: Found partially invariant condition "
 << *Info->InstToDuplicate[0] << "\n"; } } while
 (false)
        dbgs() << "simple-loop-unswitch: Found partially invariant condition "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "simple-loop-unswitch: Found partially invariant condition "
 << *Info->InstToDuplicate[0] << "\n"; } } while
 (false)
               << *Info->InstToDuplicate[0] << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "simple-loop-unswitch: Found partially invariant condition "
 << *Info->InstToDuplicate[0] << "\n"; } } while
 (false);
    PartialIVInfo = *Info;
    PartialIVCondBranch = L.getHeader()->getTerminator();
    TinyPtrVector<Value *> ValsToDuplicate;
    llvm::append_range(ValsToDuplicate, Info->InstToDuplicate);
    UnswitchCandidates.push_back(
        {L.getHeader()->getTerminator(), std::move(ValsToDuplicate)});
  }
}

// If we didn't find any candidates, we're done.
if (UnswitchCandidates.empty())
8
←
Taking false branch→
  return false;

// Check if there are irreducible CFG cycles in this loop. If so, we cannot
// easily unswitch non-trivial edges out of the loop. Doing so might turn the
// irreducible control flow into reducible control flow and introduce new
// loops "out of thin air". If we ever discover important use cases for doing
// this, we can add support to loop unswitch, but it is a lot of complexity
// for what seems little or no real world benefit.
LoopBlocksRPO RPOT(&L);
RPOT.perform(&LI);
if (containsIrreducibleCFG<const BasicBlock *>(RPOT, LI))
9
←
Taking false branch→
  return false;

SmallVector<BasicBlock *, 4> ExitBlocks;
L.getUniqueExitBlocks(ExitBlocks);

// We cannot unswitch if exit blocks contain a cleanuppad/catchswitch
// instruction as we don't know how to split those exit blocks.
// FIXME: We should teach SplitBlock to handle this and remove this
// restriction.
for (auto *ExitBB : ExitBlocks) {
10
←
Assuming '__begin1' is equal to '__end1'→
  auto *I = ExitBB->getFirstNonPHI();
  if (isa<CleanupPadInst>(I) || isa<CatchSwitchInst>(I)) {
    LLVM_DEBUG(dbgs() << "Cannot unswitch because of cleanuppad/catchswitch "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Cannot unswitch because of cleanuppad/catchswitch "
 "in exit block\n"; } } while (false)
                         "in exit block\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Cannot unswitch because of cleanuppad/catchswitch "
 "in exit block\n"; } } while (false);
    return false;
  }
}

LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Considering " <<
 UnswitchCandidates.size() << " non-trivial loop invariant conditions for unswitching.\n"
; } } while (false)
11
←
Assuming 'DebugFlag' is false→
12
←
Loop condition is false.  Exiting loop→
    dbgs() << "Considering " << UnswitchCandidates.size()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Considering " <<
 UnswitchCandidates.size() << " non-trivial loop invariant conditions for unswitching.\n"
; } } while (false)
           << " non-trivial loop invariant conditions for unswitching.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Considering " <<
 UnswitchCandidates.size() << " non-trivial loop invariant conditions for unswitching.\n"
; } } while (false);

// Given that unswitching these terminators will require duplicating parts of
// the loop, so we need to be able to model that cost. Compute the ephemeral
// values and set up a data structure to hold per-BB costs. We cache each
// block's cost so that we don't recompute this when considering different
// subsets of the loop for duplication during unswitching.
SmallPtrSet<const Value *, 4> EphValues;
CodeMetrics::collectEphemeralValues(&L, &AC, EphValues);
SmallDenseMap<BasicBlock *, InstructionCost, 4> BBCostMap;

// Compute the cost of each block, as well as the total loop cost. Also, bail
// out if we see instructions which are incompatible with loop unswitching
// (convergent, noduplicate, or cross-basic-block tokens).
// FIXME: We might be able to safely handle some of these in non-duplicated
// regions.
TargetTransformInfo::TargetCostKind CostKind =
    L.getHeader()->getParent()->hasMinSize()
13
←
Assuming the condition is false→
14
←
'?' condition is false→
    ? TargetTransformInfo::TCK_CodeSize
    : TargetTransformInfo::TCK_SizeAndLatency;
InstructionCost LoopCost = 0;
for (auto *BB : L.blocks()) {
15
←
Assuming '__begin1' is equal to '__end1'→
  InstructionCost Cost = 0;
  for (auto &I : *BB) {
    if (EphValues.count(&I))
      continue;

    if (I.getType()->isTokenTy() && I.isUsedOutsideOfBlock(BB))
      return false;
    if (auto *CB = dyn_cast<CallBase>(&I))
      if (CB->isConvergent() || CB->cannotDuplicate())
        return false;

    Cost += TTI.getInstructionCost(&I, CostKind);
  }
  assert(Cost >= 0 && "Must not have negative costs!")(static_cast <bool> (Cost >= 0 && "Must not have negative costs!"
) ? void (0) : __assert_fail ("Cost >= 0 && \"Must not have negative costs!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2883, __extension__
 __PRETTY_FUNCTION__));
  LoopCost += Cost;
  assert(LoopCost >= 0 && "Must not have negative loop costs!")(static_cast <bool> (LoopCost >= 0 && "Must not have negative loop costs!"
) ? void (0) : __assert_fail ("LoopCost >= 0 && \"Must not have negative loop costs!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2885, __extension__
 __PRETTY_FUNCTION__));
  BBCostMap[BB] = Cost;
}
LLVM_DEBUG(dbgs() << "  Total loop cost: " << LoopCost << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Total loop cost: "
 << LoopCost << "\n"; } } while (false);
16
←
Assuming 'DebugFlag' is false→
17
←
Loop condition is false.  Exiting loop→

// Now we find the best candidate by searching for the one with the following
// properties in order:
//
// 1) An unswitching cost below the threshold
// 2) The smallest number of duplicated unswitch candidates (to avoid
//    creating redundant subsequent unswitching)
// 3) The smallest cost after unswitching.
//
// We prioritize reducing fanout of unswitch candidates provided the cost
// remains below the threshold because this has a multiplicative effect.
//
// This requires memoizing each dominator subtree to avoid redundant work.
//
// FIXME: Need to actually do the number of candidates part above.
SmallDenseMap<DomTreeNode *, InstructionCost, 4> DTCostMap;
// Given a terminator which might be unswitched, computes the non-duplicated
// cost for that terminator.
auto ComputeUnswitchedCost = [&](Instruction &TI,
                                 bool FullUnswitch) -> InstructionCost {
  BasicBlock &BB = *TI.getParent();
  SmallPtrSet<BasicBlock *, 4> Visited;

  InstructionCost Cost = 0;
  for (BasicBlock *SuccBB : successors(&BB)) {
    // Don't count successors more than once.
    if (!Visited.insert(SuccBB).second)
22
←
Assuming field 'second' is true→
23
←
Taking false branch→
      continue;

    // If this is a partial unswitch candidate, then it must be a conditional
    // branch with a condition of either `or`, `and`, their corresponding
    // select forms or partially invariant instructions. In that case, one of
    // the successors is necessarily duplicated, so don't even try to remove
    // its cost.
    if (!FullUnswitch23.1
'FullUnswitch' is false
1
'FullUnswitch' is false
) {
24
←
Taking true branch→
      auto &BI = cast<BranchInst>(TI);
25
←
'TI' is a 'BranchInst'→
      Value *Cond = skipTrivialSelect(BI.getCondition());
      if (match(Cond, m_LogicalAnd())) {
26
←
Taking false branch→
        if (SuccBB == BI.getSuccessor(1))
          continue;
      } else if (match(Cond, m_LogicalOr())) {
        if (SuccBB == BI.getSuccessor(0))
          continue;
      } else if ((PartialIVInfo.KnownValue->isOneValue() &&
27
←
Called C++ object pointer is null
                  SuccBB == BI.getSuccessor(0)) ||
                 (!PartialIVInfo.KnownValue->isOneValue() &&
                  SuccBB == BI.getSuccessor(1)))
        continue;
    }

    // This successor's domtree will not need to be duplicated after
    // unswitching if the edge to the successor dominates it (and thus the
    // entire tree). This essentially means there is no other path into this
    // subtree and so it will end up live in only one clone of the loop.
    if (SuccBB->getUniquePredecessor() ||
        llvm::all_of(predecessors(SuccBB), [&](BasicBlock *PredBB) {
          return PredBB == &BB || DT.dominates(SuccBB, PredBB);
        })) {
      Cost += computeDomSubtreeCost(*DT[SuccBB], BBCostMap, DTCostMap);
      assert(Cost <= LoopCost &&(static_cast <bool> (Cost <= LoopCost && "Non-duplicated cost should never exceed total loop cost!"
) ? void (0) : __assert_fail ("Cost <= LoopCost && \"Non-duplicated cost should never exceed total loop cost!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2949, __extension__
 __PRETTY_FUNCTION__))
             "Non-duplicated cost should never exceed total loop cost!")(static_cast <bool> (Cost <= LoopCost && "Non-duplicated cost should never exceed total loop cost!"
) ? void (0) : __assert_fail ("Cost <= LoopCost && \"Non-duplicated cost should never exceed total loop cost!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2949, __extension__
 __PRETTY_FUNCTION__));
    }
  }

  // Now scale the cost by the number of unique successors minus one. We
  // subtract one because there is already at least one copy of the entire
  // loop. This is computing the new cost of unswitching a condition.
  // Note that guards always have 2 unique successors that are implicit and
  // will be materialized if we decide to unswitch it.
  int SuccessorsCount = isGuard(&TI) ? 2 : Visited.size();
  assert(SuccessorsCount > 1 &&(static_cast <bool> (SuccessorsCount > 1 && "Cannot unswitch a condition without multiple distinct successors!"
) ? void (0) : __assert_fail ("SuccessorsCount > 1 && \"Cannot unswitch a condition without multiple distinct successors!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2960, __extension__
 __PRETTY_FUNCTION__))
         "Cannot unswitch a condition without multiple distinct successors!")(static_cast <bool> (SuccessorsCount > 1 && "Cannot unswitch a condition without multiple distinct successors!"
) ? void (0) : __assert_fail ("SuccessorsCount > 1 && \"Cannot unswitch a condition without multiple distinct successors!\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2960, __extension__
 __PRETTY_FUNCTION__));
  return (LoopCost - Cost) * (SuccessorsCount - 1);
};
Instruction *BestUnswitchTI = nullptr;
InstructionCost BestUnswitchCost = 0;
ArrayRef<Value *> BestUnswitchInvariants;
for (auto &TerminatorAndInvariants : UnswitchCandidates) {
18
←
Assuming '__begin1' is not equal to '__end1'→
  Instruction &TI = *TerminatorAndInvariants.first;
  ArrayRef<Value *> Invariants = TerminatorAndInvariants.second;
  BranchInst *BI = dyn_cast<BranchInst>(&TI);
19
←
Assuming the object is a 'CastReturnType'→
  InstructionCost CandidateCost = ComputeUnswitchedCost(
21
←
Calling 'operator()'→
      TI, /*FullUnswitch*/ !BI19.1
'BI' is non-null
1
'BI' is non-null
 ||
              (Invariants.size() == 1 &&
20
←
Assuming the condition is false→
               Invariants[0] == skipTrivialSelect(BI->getCondition())));
  // Calculate cost multiplier which is a tool to limit potentially
  // exponential behavior of loop-unswitch.
  if (EnableUnswitchCostMultiplier) {
    int CostMultiplier =
        CalculateUnswitchCostMultiplier(TI, L, LI, DT, UnswitchCandidates);
    assert((static_cast <bool> ((CostMultiplier > 0 && CostMultiplier
 <= UnswitchThreshold) && "cost multiplier needs to be in the range of 1..UnswitchThreshold"
) ? void (0) : __assert_fail ("(CostMultiplier > 0 && CostMultiplier <= UnswitchThreshold) && \"cost multiplier needs to be in the range of 1..UnswitchThreshold\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2981, __extension__
 __PRETTY_FUNCTION__))
        (CostMultiplier > 0 && CostMultiplier <= UnswitchThreshold) &&(static_cast <bool> ((CostMultiplier > 0 && CostMultiplier
 <= UnswitchThreshold) && "cost multiplier needs to be in the range of 1..UnswitchThreshold"
) ? void (0) : __assert_fail ("(CostMultiplier > 0 && CostMultiplier <= UnswitchThreshold) && \"cost multiplier needs to be in the range of 1..UnswitchThreshold\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2981, __extension__
 __PRETTY_FUNCTION__))
        "cost multiplier needs to be in the range of 1..UnswitchThreshold")(static_cast <bool> ((CostMultiplier > 0 && CostMultiplier
 <= UnswitchThreshold) && "cost multiplier needs to be in the range of 1..UnswitchThreshold"
) ? void (0) : __assert_fail ("(CostMultiplier > 0 && CostMultiplier <= UnswitchThreshold) && \"cost multiplier needs to be in the range of 1..UnswitchThreshold\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2981, __extension__
 __PRETTY_FUNCTION__));
    CandidateCost *= CostMultiplier;
    LLVM_DEBUG(dbgs() << "  Computed cost of " << CandidateCostdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Computed cost of "
 << CandidateCost << " (multiplier: " << CostMultiplier
 << ")" << " for unswitch candidate: " << TI
 << "\n"; } } while (false)
                      << " (multiplier: " << CostMultiplier << ")"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Computed cost of "
 << CandidateCost << " (multiplier: " << CostMultiplier
 << ")" << " for unswitch candidate: " << TI
 << "\n"; } } while (false)
                      << " for unswitch candidate: " << TI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Computed cost of "
 << CandidateCost << " (multiplier: " << CostMultiplier
 << ")" << " for unswitch candidate: " << TI
 << "\n"; } } while (false);
  } else {
    LLVM_DEBUG(dbgs() << "  Computed cost of " << CandidateCostdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Computed cost of "
 << CandidateCost << " for unswitch candidate: " <<
 TI << "\n"; } } while (false)
                      << " for unswitch candidate: " << TI << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Computed cost of "
 << CandidateCost << " for unswitch candidate: " <<
 TI << "\n"; } } while (false);
  }

  if (!BestUnswitchTI || CandidateCost < BestUnswitchCost) {
    BestUnswitchTI = &TI;
    BestUnswitchCost = CandidateCost;
    BestUnswitchInvariants = Invariants;
  }
}
assert(BestUnswitchTI && "Failed to find loop unswitch candidate")(static_cast <bool> (BestUnswitchTI && "Failed to find loop unswitch candidate"
) ? void (0) : __assert_fail ("BestUnswitchTI && \"Failed to find loop unswitch candidate\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 2997, __extension__
 __PRETTY_FUNCTION__));

if (BestUnswitchCost >= UnswitchThreshold) {
  LLVM_DEBUG(dbgs() << "Cannot unswitch, lowest cost found: "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Cannot unswitch, lowest cost found: "
 << BestUnswitchCost << "\n"; } } while (false)
                    << BestUnswitchCost << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Cannot unswitch, lowest cost found: "
 << BestUnswitchCost << "\n"; } } while (false);
  return false;
}

if (BestUnswitchTI != PartialIVCondBranch)
  PartialIVInfo.InstToDuplicate.clear();

// If the best candidate is a guard, turn it into a branch.
if (isGuard(BestUnswitchTI))
  BestUnswitchTI = turnGuardIntoBranch(cast<IntrinsicInst>(BestUnswitchTI), L,
                                       ExitBlocks, DT, LI, MSSAU);

LLVM_DEBUG(dbgs() << "  Unswitching non-trivial (cost = "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Unswitching non-trivial (cost = "
 << BestUnswitchCost << ") terminator: " <<
 *BestUnswitchTI << "\n"; } } while (false)
                  << BestUnswitchCost << ") terminator: " << *BestUnswitchTIdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Unswitching non-trivial (cost = "
 << BestUnswitchCost << ") terminator: " <<
 *BestUnswitchTI << "\n"; } } while (false)
                  << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "  Unswitching non-trivial (cost = "
 << BestUnswitchCost << ") terminator: " <<
 *BestUnswitchTI << "\n"; } } while (false);
unswitchNontrivialInvariants(L, *BestUnswitchTI, BestUnswitchInvariants,
                             ExitBlocks, PartialIVInfo, DT, LI, AC,
                             UnswitchCB, SE, MSSAU, DestroyLoopCB);
return true;
3020}

3022/// Unswitch control flow predicated on loop invariant conditions.
3023///
3024/// This first hoists all branches or switches which are trivial (IE, do not
3025/// require duplicating any part of the loop) out of the loop body. It then
3026/// looks at other loop invariant control flows and tries to unswitch those as
3027/// well by cloning the loop if the result is small enough.
3028///
3029/// The `DT`, `LI`, `AC`, `AA`, `TTI` parameters are required analyses that are
3030/// also updated based on the unswitch. The `MSSA` analysis is also updated if
3031/// valid (i.e. its use is enabled).
3032///
3033/// If either `NonTrivial` is true or the flag `EnableNonTrivialUnswitch` is
3034/// true, we will attempt to do non-trivial unswitching as well as trivial
3035/// unswitching.
3036///
3037/// The `UnswitchCB` callback provided will be run after unswitching is
3038/// complete, with the first parameter set to `true` if the provided loop
3039/// remains a loop, and a list of new sibling loops created.
3040///
3041/// If `SE` is non-null, we will update that analysis based on the unswitching
3042/// done.
3043static bool
3044unswitchLoop(Loop &L, DominatorTree &DT, LoopInfo &LI, AssumptionCache &AC,
           AAResults &AA, TargetTransformInfo &TTI, bool Trivial,
           bool NonTrivial,
           function_ref<void(bool, bool, ArrayRef<Loop *>)> UnswitchCB,
           ScalarEvolution *SE, MemorySSAUpdater *MSSAU,
           ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI,
           function_ref<void(Loop &, StringRef)> DestroyLoopCB) {
assert(L.isRecursivelyLCSSAForm(DT, LI) &&(static_cast <bool> (L.isRecursivelyLCSSAForm(DT, LI) &&
 "Loops must be in LCSSA form before unswitching.") ? void (0
) : __assert_fail ("L.isRecursivelyLCSSAForm(DT, LI) && \"Loops must be in LCSSA form before unswitching.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 3052, __extension__
 __PRETTY_FUNCTION__))
       "Loops must be in LCSSA form before unswitching.")(static_cast <bool> (L.isRecursivelyLCSSAForm(DT, LI) &&
 "Loops must be in LCSSA form before unswitching.") ? void (0
) : __assert_fail ("L.isRecursivelyLCSSAForm(DT, LI) && \"Loops must be in LCSSA form before unswitching.\""
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 3052, __extension__
 __PRETTY_FUNCTION__));

// Must be in loop simplified form: we need a preheader and dedicated exits.
if (!L.isLoopSimplifyForm())
  return false;

// Try trivial unswitch first before loop over other basic blocks in the loop.
if (Trivial && unswitchAllTrivialConditions(L, DT, LI, SE, MSSAU)) {
  // If we unswitched successfully we will want to clean up the loop before
  // processing it further so just mark it as unswitched and return.
  UnswitchCB(/*CurrentLoopValid*/ true, false, {});
  return true;
}

// Check whether we should continue with non-trivial conditions.
// EnableNonTrivialUnswitch: Global variable that forces non-trivial
//                           unswitching for testing and debugging.
// NonTrivial: Parameter that enables non-trivial unswitching for this
//             invocation of the transform. But this should be allowed only
//             for targets without branch divergence.
//
// FIXME: If divergence analysis becomes available to a loop
// transform, we should allow unswitching for non-trivial uniform
// branches even on targets that have divergence.
// https://bugs.llvm.org/show_bug.cgi?id=48819
bool ContinueWithNonTrivial =
    EnableNonTrivialUnswitch || (NonTrivial && !TTI.hasBranchDivergence());
if (!ContinueWithNonTrivial)
  return false;

// Skip non-trivial unswitching for optsize functions.
if (L.getHeader()->getParent()->hasOptSize())
  return false;

// Skip cold loops, as unswitching them brings little benefit
// but increases the code size
if (PSI && PSI->hasProfileSummary() && BFI &&
    PSI->isFunctionColdInCallGraph(L.getHeader()->getParent(), *BFI)) {
  LLVM_DEBUG(dbgs() << " Skip cold loop: " << L << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << " Skip cold loop: "
 << L << "\n"; } } while (false);
  return false;
}

// Skip non-trivial unswitching for loops that cannot be cloned.
if (!L.isSafeToClone())
  return false;

// For non-trivial unswitching, because it often creates new loops, we rely on
// the pass manager to iterate on the loops rather than trying to immediately
// reach a fixed point. There is no substantial advantage to iterating
// internally, and if any of the new loops are simplified enough to contain
// trivial unswitching we want to prefer those.

// Try to unswitch the best invariant condition. We prefer this full unswitch to
// a partial unswitch when possible below the threshold.
if (unswitchBestCondition(L, DT, LI, AC, AA, TTI, UnswitchCB, SE, MSSAU,
                          DestroyLoopCB))
  return true;

// No other opportunities to unswitch.
return false;
3112}

3114PreservedAnalyses SimpleLoopUnswitchPass::run(Loop &L, LoopAnalysisManager &AM,
                                            LoopStandardAnalysisResults &AR,
                                            LPMUpdater &U) {
Function &F = *L.getHeader()->getParent();
(void)F;
ProfileSummaryInfo *PSI = nullptr;
if (auto OuterProxy =
        AM.getResult<FunctionAnalysisManagerLoopProxy>(L, AR)
            .getCachedResult<ModuleAnalysisManagerFunctionProxy>(F))
  PSI = OuterProxy->getCachedResult<ProfileSummaryAnalysis>(*F.getParent());
LLVM_DEBUG(dbgs() << "Unswitching loop in " << F.getName() << ": " << Ldo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Unswitching loop in "
 << F.getName() << ": " << L << "\n";
 } } while (false)
                  << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Unswitching loop in "
 << F.getName() << ": " << L << "\n";
 } } while (false);

// Save the current loop name in a variable so that we can report it even
// after it has been deleted.
std::string LoopName = std::string(L.getName());

auto UnswitchCB = [&L, &U, &LoopName](bool CurrentLoopValid,
                                      bool PartiallyInvariant,
                                      ArrayRef<Loop *> NewLoops) {
  // If we did a non-trivial unswitch, we have added new (cloned) loops.
  if (!NewLoops.empty())
    U.addSiblingLoops(NewLoops);

  // If the current loop remains valid, we should revisit it to catch any
  // other unswitch opportunities. Otherwise, we need to mark it as deleted.
  if (CurrentLoopValid) {
    if (PartiallyInvariant) {
      // Mark the new loop as partially unswitched, to avoid unswitching on
      // the same condition again.
      auto &Context = L.getHeader()->getContext();
      MDNode *DisableUnswitchMD = MDNode::get(
          Context,
          MDString::get(Context, "llvm.loop.unswitch.partial.disable"));
      MDNode *NewLoopID = makePostTransformationMetadata(
          Context, L.getLoopID(), {"llvm.loop.unswitch.partial"},
          {DisableUnswitchMD});
      L.setLoopID(NewLoopID);
    } else
      U.revisitCurrentLoop();
  } else
    U.markLoopAsDeleted(L, LoopName);
};

auto DestroyLoopCB = [&U](Loop &L, StringRef Name) {
  U.markLoopAsDeleted(L, Name);
};

Optional<MemorySSAUpdater> MSSAU;
if (AR.MSSA) {
  MSSAU = MemorySSAUpdater(AR.MSSA);
  if (VerifyMemorySSA)
    AR.MSSA->verifyMemorySSA();
}
if (!unswitchLoop(L, AR.DT, AR.LI, AR.AC, AR.AA, AR.TTI, Trivial, NonTrivial,
                  UnswitchCB, &AR.SE, MSSAU ? MSSAU.getPointer() : nullptr,
                  PSI, AR.BFI, DestroyLoopCB))
  return PreservedAnalyses::all();

if (AR.MSSA && VerifyMemorySSA)
  AR.MSSA->verifyMemorySSA();

// Historically this pass has had issues with the dominator tree so verify it
// in asserts builds.
assert(AR.DT.verify(DominatorTree::VerificationLevel::Fast))(static_cast <bool> (AR.DT.verify(DominatorTree::VerificationLevel
::Fast)) ? void (0) : __assert_fail ("AR.DT.verify(DominatorTree::VerificationLevel::Fast)"
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 3178, __extension__
 __PRETTY_FUNCTION__));

auto PA = getLoopPassPreservedAnalyses();
if (AR.MSSA)
  PA.preserve<MemorySSAAnalysis>();
return PA;
3184}

3186void SimpleLoopUnswitchPass::printPipeline(
  raw_ostream &OS, function_ref<StringRef(StringRef)> MapClassName2PassName) {
static_cast<PassInfoMixin<SimpleLoopUnswitchPass> *>(this)->printPipeline(
    OS, MapClassName2PassName);

OS << "<";
OS << (NonTrivial ? "" : "no-") << "nontrivial;";
OS << (Trivial ? "" : "no-") << "trivial";
OS << ">";
3195}

3197namespace {

3199class SimpleLoopUnswitchLegacyPass : public LoopPass {
bool NonTrivial;

3202public:
static char ID; // Pass ID, replacement for typeid

explicit SimpleLoopUnswitchLegacyPass(bool NonTrivial = false)
    : LoopPass(ID), NonTrivial(NonTrivial) {
  initializeSimpleLoopUnswitchLegacyPassPass(
      *PassRegistry::getPassRegistry());
}

bool runOnLoop(Loop *L, LPPassManager &LPM) override;

void getAnalysisUsage(AnalysisUsage &AU) const override {
  AU.addRequired<AssumptionCacheTracker>();
  AU.addRequired<TargetTransformInfoWrapperPass>();
  AU.addRequired<MemorySSAWrapperPass>();
  AU.addPreserved<MemorySSAWrapperPass>();
  getLoopAnalysisUsage(AU);
}
3220};

3222} // end anonymous namespace

3224bool SimpleLoopUnswitchLegacyPass::runOnLoop(Loop *L, LPPassManager &LPM) {
if (skipLoop(L))
  return false;

Function &F = *L->getHeader()->getParent();

LLVM_DEBUG(dbgs() << "Unswitching loop in " << F.getName() << ": " << *Ldo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Unswitching loop in "
 << F.getName() << ": " << *L << "\n"
; } } while (false)
                  << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("simple-loop-unswitch")) { dbgs() << "Unswitching loop in "
 << F.getName() << ": " << *L << "\n"
; } } while (false);
auto &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
auto &LI = getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
auto &AC = getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
auto &AA = getAnalysis<AAResultsWrapperPass>().getAAResults();
auto &TTI = getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
MemorySSA *MSSA = &getAnalysis<MemorySSAWrapperPass>().getMSSA();
MemorySSAUpdater MSSAU(MSSA);

auto *SEWP = getAnalysisIfAvailable<ScalarEvolutionWrapperPass>();
auto *SE = SEWP ? &SEWP->getSE() : nullptr;

auto UnswitchCB = [&L, &LPM](bool CurrentLoopValid, bool PartiallyInvariant,
                             ArrayRef<Loop *> NewLoops) {
  // If we did a non-trivial unswitch, we have added new (cloned) loops.
  for (auto *NewL : NewLoops)
    LPM.addLoop(*NewL);

  // If the current loop remains valid, re-add it to the queue. This is
  // a little wasteful as we'll finish processing the current loop as well,
  // but it is the best we can do in the old PM.
  if (CurrentLoopValid) {
    // If the current loop has been unswitched using a partially invariant
    // condition, we should not re-add the current loop to avoid unswitching
    // on the same condition again.
    if (!PartiallyInvariant)
      LPM.addLoop(*L);
  } else
    LPM.markLoopAsDeleted(*L);
};

auto DestroyLoopCB = [&LPM](Loop &L, StringRef /* Name */) {
  LPM.markLoopAsDeleted(L);
};

if (VerifyMemorySSA)
  MSSA->verifyMemorySSA();
bool Changed =
    unswitchLoop(*L, DT, LI, AC, AA, TTI, true, NonTrivial, UnswitchCB, SE,
                 &MSSAU, nullptr, nullptr, DestroyLoopCB);

if (VerifyMemorySSA)
  MSSA->verifyMemorySSA();

// Historically this pass has had issues with the dominator tree so verify it
// in asserts builds.
assert(DT.verify(DominatorTree::VerificationLevel::Fast))(static_cast <bool> (DT.verify(DominatorTree::VerificationLevel
::Fast)) ? void (0) : __assert_fail ("DT.verify(DominatorTree::VerificationLevel::Fast)"
, "llvm/lib/Transforms/Scalar/SimpleLoopUnswitch.cpp", 3277, __extension__
 __PRETTY_FUNCTION__));

return Changed;
3280}

3282char SimpleLoopUnswitchLegacyPass::ID = 0;
3283INITIALIZE_PASS_BEGIN(SimpleLoopUnswitchLegacyPass, "simple-loop-unswitch",static void *initializeSimpleLoopUnswitchLegacyPassPassOnce(PassRegistry
 &Registry) {
                    "Simple unswitch loops", false, false)static void *initializeSimpleLoopUnswitchLegacyPassPassOnce(PassRegistry
 &Registry) {
3285INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)initializeAssumptionCacheTrackerPass(Registry);
3286INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)initializeDominatorTreeWrapperPassPass(Registry);
3287INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)initializeLoopInfoWrapperPassPass(Registry);
3288INITIALIZE_PASS_DEPENDENCY(LoopPass)initializeLoopPassPass(Registry);
3289INITIALIZE_PASS_DEPENDENCY(MemorySSAWrapperPass)initializeMemorySSAWrapperPassPass(Registry);
3290INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)initializeTargetTransformInfoWrapperPassPass(Registry);
3291INITIALIZE_PASS_END(SimpleLoopUnswitchLegacyPass, "simple-loop-unswitch",PassInfo *PI = new PassInfo( "Simple unswitch loops", "simple-loop-unswitch"
, &SimpleLoopUnswitchLegacyPass::ID, PassInfo::NormalCtor_t
(callDefaultCtor<SimpleLoopUnswitchLegacyPass>), false,
 false); Registry.registerPass(*PI, true); return PI; } static
 llvm::once_flag InitializeSimpleLoopUnswitchLegacyPassPassFlag
; void llvm::initializeSimpleLoopUnswitchLegacyPassPass(PassRegistry
 &Registry) { llvm::call_once(InitializeSimpleLoopUnswitchLegacyPassPassFlag
, initializeSimpleLoopUnswitchLegacyPassPassOnce, std::ref(Registry
)); }
                  "Simple unswitch loops", false, false)PassInfo *PI = new PassInfo( "Simple unswitch loops", "simple-loop-unswitch"
, &SimpleLoopUnswitchLegacyPass::ID, PassInfo::NormalCtor_t
(callDefaultCtor<SimpleLoopUnswitchLegacyPass>), false,
 false); Registry.registerPass(*PI, true); return PI; } static
 llvm::once_flag InitializeSimpleLoopUnswitchLegacyPassPassFlag
; void llvm::initializeSimpleLoopUnswitchLegacyPassPass(PassRegistry
 &Registry) { llvm::call_once(InitializeSimpleLoopUnswitchLegacyPassPassFlag
, initializeSimpleLoopUnswitchLegacyPassPassOnce, std::ref(Registry
)); }

3294Pass *llvm::createSimpleLoopUnswitchLegacyPass(bool NonTrivial) {
return new SimpleLoopUnswitchLegacyPass(NonTrivial);
3296}

←

/build/llvm-toolchain-snapshot-16~++20221003111214+1fa2019828ca/llvm/include/llvm/Transforms/Utils/LoopUtils.h

1//===- llvm/Transforms/Utils/LoopUtils.h - Loop utilities -------*- 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 defines some loop transformation utilities.
10//
11//===----------------------------------------------------------------------===//
12 
13#ifndef LLVM_TRANSFORMS_UTILS_LOOPUTILS_H
14#define LLVM_TRANSFORMS_UTILS_LOOPUTILS_H
15 
16#include "llvm/Analysis/IVDescriptors.h"
17#include "llvm/Analysis/LoopAccessAnalysis.h"
18#include "llvm/Transforms/Utils/ValueMapper.h"
19 
20namespace llvm {
21 
22template <typename T> class DomTreeNodeBase;
23using DomTreeNode = DomTreeNodeBase<BasicBlock>;
24class AssumptionCache;
25class StringRef;
26class AnalysisUsage;
27class TargetTransformInfo;
28class AAResults;
29class BasicBlock;
30class ICFLoopSafetyInfo;
31class IRBuilderBase;
32class Loop;
33class LoopInfo;
34class MemoryAccess;
35class MemorySSA;
36class MemorySSAUpdater;
37class OptimizationRemarkEmitter;
38class PredIteratorCache;
39class ScalarEvolution;
40class SCEV;
41class SCEVExpander;
42class TargetLibraryInfo;
43class LPPassManager;
44class Instruction;
45struct RuntimeCheckingPtrGroup;
46typedef std::pair<const RuntimeCheckingPtrGroup *,
47                  const RuntimeCheckingPtrGroup *>
48    RuntimePointerCheck;
49 
50template <typename T> class Optional;
51template <typename T, unsigned N> class SmallSetVector;
52template <typename T, unsigned N> class SmallPriorityWorklist;
53 
54BasicBlock *InsertPreheaderForLoop(Loop *L, DominatorTree *DT, LoopInfo *LI,
55                                   MemorySSAUpdater *MSSAU, bool PreserveLCSSA);
56 
57/// Ensure that all exit blocks of the loop are dedicated exits.
58///
59/// For any loop exit block with non-loop predecessors, we split the loop
60/// predecessors to use a dedicated loop exit block. We update the dominator
61/// tree and loop info if provided, and will preserve LCSSA if requested.
62bool formDedicatedExitBlocks(Loop *L, DominatorTree *DT, LoopInfo *LI,
63                             MemorySSAUpdater *MSSAU, bool PreserveLCSSA);
64 
65/// Ensures LCSSA form for every instruction from the Worklist in the scope of
66/// innermost containing loop.
67///
68/// For the given instruction which have uses outside of the loop, an LCSSA PHI
69/// node is inserted and the uses outside the loop are rewritten to use this
70/// node.
71///
72/// LoopInfo and DominatorTree are required and, since the routine makes no
73/// changes to CFG, preserved.
74///
75/// Returns true if any modifications are made.
76///
77/// This function may introduce unused PHI nodes. If \p PHIsToRemove is not
78/// nullptr, those are added to it (before removing, the caller has to check if
79/// they still do not have any uses). Otherwise the PHIs are directly removed.
80bool formLCSSAForInstructions(
81    SmallVectorImpl<Instruction *> &Worklist, const DominatorTree &DT,
82    const LoopInfo &LI, ScalarEvolution *SE, IRBuilderBase &Builder,
83    SmallVectorImpl<PHINode *> *PHIsToRemove = nullptr);
84 
85/// Put loop into LCSSA form.
86///
87/// Looks at all instructions in the loop which have uses outside of the
88/// current loop. For each, an LCSSA PHI node is inserted and the uses outside
89/// the loop are rewritten to use this node. Sub-loops must be in LCSSA form
90/// already.
91///
92/// LoopInfo and DominatorTree are required and preserved.
93///
94/// If ScalarEvolution is passed in, it will be preserved.
95///
96/// Returns true if any modifications are made to the loop.
97bool formLCSSA(Loop &L, const DominatorTree &DT, const LoopInfo *LI,
98               ScalarEvolution *SE);
99 
100/// Put a loop nest into LCSSA form.
101///
102/// This recursively forms LCSSA for a loop nest.
103///
104/// LoopInfo and DominatorTree are required and preserved.
105///
106/// If ScalarEvolution is passed in, it will be preserved.
107///
108/// Returns true if any modifications are made to the loop.
109bool formLCSSARecursively(Loop &L, const DominatorTree &DT, const LoopInfo *LI,
110                          ScalarEvolution *SE);
111 
112/// Flags controlling how much is checked when sinking or hoisting
113/// instructions.  The number of memory access in the loop (and whether there
114/// are too many) is determined in the constructors when using MemorySSA.
115class SinkAndHoistLICMFlags {
116public:
117  // Explicitly set limits.
118  SinkAndHoistLICMFlags(unsigned LicmMssaOptCap,
119                        unsigned LicmMssaNoAccForPromotionCap, bool IsSink,
120                        Loop *L = nullptr, MemorySSA *MSSA = nullptr);
121  // Use default limits.
122  SinkAndHoistLICMFlags(bool IsSink, Loop *L = nullptr,
123                        MemorySSA *MSSA = nullptr);
124 
125  void setIsSink(bool B) { IsSink = B; }
126  bool getIsSink() { return IsSink; }
127  bool tooManyMemoryAccesses() { return NoOfMemAccTooLarge; }
128  bool tooManyClobberingCalls() { return LicmMssaOptCounter >= LicmMssaOptCap; }
129  void incrementClobberingCalls() { ++LicmMssaOptCounter; }
130 
131protected:
132  bool NoOfMemAccTooLarge = false;
133  unsigned LicmMssaOptCounter = 0;
134  unsigned LicmMssaOptCap;
135  unsigned LicmMssaNoAccForPromotionCap;
136  bool IsSink;
137};
138 
139/// Walk the specified region of the CFG (defined by all blocks
140/// dominated by the specified block, and that are in the current loop) in
141/// reverse depth first order w.r.t the DominatorTree. This allows us to visit
142/// uses before definitions, allowing us to sink a loop body in one pass without
143/// iteration. Takes DomTreeNode, AAResults, LoopInfo, DominatorTree,
144/// TargetLibraryInfo, Loop, AliasSet information for all
145/// instructions of the loop and loop safety information as
146/// arguments. Diagnostics is emitted via \p ORE. It returns changed status.
147/// \p CurLoop is a loop to do sinking on. \p OutermostLoop is used only when
148/// this function is called by \p sinkRegionForLoopNest.
149bool sinkRegion(DomTreeNode *, AAResults *, LoopInfo *, DominatorTree *,
150                TargetLibraryInfo *, TargetTransformInfo *, Loop *CurLoop,
151                MemorySSAUpdater &, ICFLoopSafetyInfo *,
152                SinkAndHoistLICMFlags &, OptimizationRemarkEmitter *,
153                Loop *OutermostLoop = nullptr);
154 
155/// Call sinkRegion on loops contained within the specified loop
156/// in order from innermost to outermost.
157bool sinkRegionForLoopNest(DomTreeNode *, AAResults *, LoopInfo *,
158                           DominatorTree *, TargetLibraryInfo *,
159                           TargetTransformInfo *, Loop *, MemorySSAUpdater &,
160                           ICFLoopSafetyInfo *, SinkAndHoistLICMFlags &,
161                           OptimizationRemarkEmitter *);
162 
163/// Walk the specified region of the CFG (defined by all blocks
164/// dominated by the specified block, and that are in the current loop) in depth
165/// first order w.r.t the DominatorTree.  This allows us to visit definitions
166/// before uses, allowing us to hoist a loop body in one pass without iteration.
167/// Takes DomTreeNode, AAResults, LoopInfo, DominatorTree,
168/// TargetLibraryInfo, Loop, AliasSet information for all
169/// instructions of the loop and loop safety information as arguments.
170/// Diagnostics is emitted via \p ORE. It returns changed status.
171/// \p AllowSpeculation is whether values should be hoisted even if they are not
172/// guaranteed to execute in the loop, but are safe to speculatively execute.
173bool hoistRegion(DomTreeNode *, AAResults *, LoopInfo *, DominatorTree *,
174                 AssumptionCache *, TargetLibraryInfo *, Loop *,
175                 MemorySSAUpdater &, ScalarEvolution *, ICFLoopSafetyInfo *,
176                 SinkAndHoistLICMFlags &, OptimizationRemarkEmitter *, bool,
177                 bool AllowSpeculation);
178 
179/// This function deletes dead loops. The caller of this function needs to
180/// guarantee that the loop is infact dead.
181/// The function requires a bunch or prerequisites to be present:
182///   - The loop needs to be in LCSSA form
183///   - The loop needs to have a Preheader
184///   - A unique dedicated exit block must exist
185///
186/// This also updates the relevant analysis information in \p DT, \p SE, \p LI
187/// and \p MSSA if pointers to those are provided.
188/// It also updates the loop PM if an updater struct is provided.
189 
190void deleteDeadLoop(Loop *L, DominatorTree *DT, ScalarEvolution *SE,
191                    LoopInfo *LI, MemorySSA *MSSA = nullptr);
192 
193/// Remove the backedge of the specified loop.  Handles loop nests and general
194/// loop structures subject to the precondition that the loop has no parent
195/// loop and has a single latch block.  Preserves all listed analyses.
196void breakLoopBackedge(Loop *L, DominatorTree &DT, ScalarEvolution &SE,
197                       LoopInfo &LI, MemorySSA *MSSA);
198 
199/// Try to promote memory values to scalars by sinking stores out of
200/// the loop and moving loads to before the loop.  We do this by looping over
201/// the stores in the loop, looking for stores to Must pointers which are
202/// loop invariant. It takes a set of must-alias values, Loop exit blocks
203/// vector, loop exit blocks insertion point vector, PredIteratorCache,
204/// LoopInfo, DominatorTree, Loop, AliasSet information for all instructions
205/// of the loop and loop safety information as arguments.
206/// Diagnostics is emitted via \p ORE. It returns changed status.
207/// \p AllowSpeculation is whether values should be hoisted even if they are not
208/// guaranteed to execute in the loop, but are safe to speculatively execute.
209bool promoteLoopAccessesToScalars(
210    const SmallSetVector<Value *, 8> &, SmallVectorImpl<BasicBlock *> &,
211    SmallVectorImpl<Instruction *> &, SmallVectorImpl<MemoryAccess *> &,
212    PredIteratorCache &, LoopInfo *, DominatorTree *, AssumptionCache *AC,
213    const TargetLibraryInfo *, Loop *, MemorySSAUpdater &, ICFLoopSafetyInfo *,
214    OptimizationRemarkEmitter *, bool AllowSpeculation);
215 
216/// Does a BFS from a given node to all of its children inside a given loop.
217/// The returned vector of nodes includes the starting point.
218SmallVector<DomTreeNode *, 16> collectChildrenInLoop(DomTreeNode *N,
219                                                     const Loop *CurLoop);
220 
221/// Returns the instructions that use values defined in the loop.
222SmallVector<Instruction *, 8> findDefsUsedOutsideOfLoop(Loop *L);
223 
224/// Find a combination of metadata ("llvm.loop.vectorize.width" and
225/// "llvm.loop.vectorize.scalable.enable") for a loop and use it to construct a
226/// ElementCount. If the metadata "llvm.loop.vectorize.width" cannot be found
227/// then None is returned.
228Optional<ElementCount>
229getOptionalElementCountLoopAttribute(const Loop *TheLoop);
230 
231/// Create a new loop identifier for a loop created from a loop transformation.
232///
233/// @param OrigLoopID The loop ID of the loop before the transformation.
234/// @param FollowupAttrs List of attribute names that contain attributes to be
235///                      added to the new loop ID.
236/// @param InheritOptionsAttrsPrefix Selects which attributes should be inherited
237///                                  from the original loop. The following values
238///                                  are considered:
239///        nullptr   : Inherit all attributes from @p OrigLoopID.
240///        ""        : Do not inherit any attribute from @p OrigLoopID; only use
241///                    those specified by a followup attribute.
242///        "<prefix>": Inherit all attributes except those which start with
243///                    <prefix>; commonly used to remove metadata for the
244///                    applied transformation.
245/// @param AlwaysNew If true, do not try to reuse OrigLoopID and never return
246///                  None.
247///
248/// @return The loop ID for the after-transformation loop. The following values
249///         can be returned:
250///         None         : No followup attribute was found; it is up to the
251///                        transformation to choose attributes that make sense.
252///         @p OrigLoopID: The original identifier can be reused.
253///         nullptr      : The new loop has no attributes.
254///         MDNode*      : A new unique loop identifier.
255Optional<MDNode *>
256makeFollowupLoopID(MDNode *OrigLoopID, ArrayRef<StringRef> FollowupAttrs,
257                   const char *InheritOptionsAttrsPrefix = "",
258                   bool AlwaysNew = false);
259 
260/// Look for the loop attribute that disables all transformation heuristic.
261bool hasDisableAllTransformsHint(const Loop *L);
262 
263/// Look for the loop attribute that disables the LICM transformation heuristics.
264bool hasDisableLICMTransformsHint(const Loop *L);
265 
266/// The mode sets how eager a transformation should be applied.
267enum TransformationMode {
268  /// The pass can use heuristics to determine whether a transformation should
269  /// be applied.
270  TM_Unspecified,
271 
272  /// The transformation should be applied without considering a cost model.
273  TM_Enable,
274 
275  /// The transformation should not be applied.
276  TM_Disable,
277 
278  /// Force is a flag and should not be used alone.
279  TM_Force = 0x04,
280 
281  /// The transformation was directed by the user, e.g. by a #pragma in
282  /// the source code. If the transformation could not be applied, a
283  /// warning should be emitted.
284  TM_ForcedByUser = TM_Enable | TM_Force,
285 
286  /// The transformation must not be applied. For instance, `#pragma clang loop
287  /// unroll(disable)` explicitly forbids any unrolling to take place. Unlike
288  /// general loop metadata, it must not be dropped. Most passes should not
289  /// behave differently under TM_Disable and TM_SuppressedByUser.
290  TM_SuppressedByUser = TM_Disable | TM_Force
291};
292 
293/// @{
294/// Get the mode for LLVM's supported loop transformations.
295TransformationMode hasUnrollTransformation(const Loop *L);
296TransformationMode hasUnrollAndJamTransformation(const Loop *L);
297TransformationMode hasVectorizeTransformation(const Loop *L);
298TransformationMode hasDistributeTransformation(const Loop *L);
299TransformationMode hasLICMVersioningTransformation(const Loop *L);
300/// @}
301 
302/// Set input string into loop metadata by keeping other values intact.
303/// If the string is already in loop metadata update value if it is
304/// different.
305void addStringMetadataToLoop(Loop *TheLoop, const char *MDString,
306                             unsigned V = 0);
307 
308/// Returns a loop's estimated trip count based on branch weight metadata.
309/// In addition if \p EstimatedLoopInvocationWeight is not null it is
310/// initialized with weight of loop's latch leading to the exit.
311/// Returns 0 when the count is estimated to be 0, or None when a meaningful
312/// estimate can not be made.
313Optional<unsigned>
314getLoopEstimatedTripCount(Loop *L,
315                          unsigned *EstimatedLoopInvocationWeight = nullptr);
316 
317/// Set a loop's branch weight metadata to reflect that loop has \p
318/// EstimatedTripCount iterations and \p EstimatedLoopInvocationWeight exits
319/// through latch. Returns true if metadata is successfully updated, false
320/// otherwise. Note that loop must have a latch block which controls loop exit
321/// in order to succeed.
322bool setLoopEstimatedTripCount(Loop *L, unsigned EstimatedTripCount,
323                               unsigned EstimatedLoopInvocationWeight);
324 
325/// Check inner loop (L) backedge count is known to be invariant on all
326/// iterations of its outer loop. If the loop has no parent, this is trivially
327/// true.
328bool hasIterationCountInvariantInParent(Loop *L, ScalarEvolution &SE);
329 
330/// Helper to consistently add the set of standard passes to a loop pass's \c
331/// AnalysisUsage.
332///
333/// All loop passes should call this as part of implementing their \c
334/// getAnalysisUsage.
335void getLoopAnalysisUsage(AnalysisUsage &AU);
336 
337/// Returns true if is legal to hoist or sink this instruction disregarding the
338/// possible introduction of faults.  Reasoning about potential faulting
339/// instructions is the responsibility of the caller since it is challenging to
340/// do efficiently from within this routine.
341/// \p TargetExecutesOncePerLoop is true only when it is guaranteed that the
342/// target executes at most once per execution of the loop body.  This is used
343/// to assess the legality of duplicating atomic loads.  Generally, this is
344/// true when moving out of loop and not true when moving into loops.
345/// If \p ORE is set use it to emit optimization remarks.
346bool canSinkOrHoistInst(Instruction &I, AAResults *AA, DominatorTree *DT,
347                        Loop *CurLoop, MemorySSAUpdater &MSSAU,
348                        bool TargetExecutesOncePerLoop,
349                        SinkAndHoistLICMFlags &LICMFlags,
350                        OptimizationRemarkEmitter *ORE = nullptr);
351 
352/// Returns the comparison predicate used when expanding a min/max reduction.
353CmpInst::Predicate getMinMaxReductionPredicate(RecurKind RK);
354 
355/// See RecurrenceDescriptor::isSelectCmpPattern for a description of the
356/// pattern we are trying to match. In this pattern we are only ever selecting
357/// between two values: 1) an initial PHI start value, and 2) a loop invariant
358/// value. This function uses \p LoopExitInst to determine 2), which we then use
359/// to select between \p Left and \p Right. Any lane value in \p Left that
360/// matches 2) will be merged into \p Right.
361Value *createSelectCmpOp(IRBuilderBase &Builder, Value *StartVal, RecurKind RK,
362                         Value *Left, Value *Right);
363 
364/// Returns a Min/Max operation corresponding to MinMaxRecurrenceKind.
365/// The Builder's fast-math-flags must be set to propagate the expected values.
366Value *createMinMaxOp(IRBuilderBase &Builder, RecurKind RK, Value *Left,
367                      Value *Right);
368 
369/// Generates an ordered vector reduction using extracts to reduce the value.
370Value *getOrderedReduction(IRBuilderBase &Builder, Value *Acc, Value *Src,
371                           unsigned Op, RecurKind MinMaxKind = RecurKind::None);
372 
373/// Generates a vector reduction using shufflevectors to reduce the value.
374/// Fast-math-flags are propagated using the IRBuilder's setting.
375Value *getShuffleReduction(IRBuilderBase &Builder, Value *Src, unsigned Op,
376                           RecurKind MinMaxKind = RecurKind::None);
377 
378/// Create a target reduction of the given vector. The reduction operation
379/// is described by the \p Opcode parameter. min/max reductions require
380/// additional information supplied in \p RdxKind.
381/// The target is queried to determine if intrinsics or shuffle sequences are
382/// required to implement the reduction.
383/// Fast-math-flags are propagated using the IRBuilder's setting.
384Value *createSimpleTargetReduction(IRBuilderBase &B,
385                                   const TargetTransformInfo *TTI, Value *Src,
386                                   RecurKind RdxKind);
387 
388/// Create a target reduction of the given vector \p Src for a reduction of the
389/// kind RecurKind::SelectICmp or RecurKind::SelectFCmp. The reduction operation
390/// is described by \p Desc.
391Value *createSelectCmpTargetReduction(IRBuilderBase &B,
392                                      const TargetTransformInfo *TTI,
393                                      Value *Src,
394                                      const RecurrenceDescriptor &Desc,
395                                      PHINode *OrigPhi);
396 
397/// Create a generic target reduction using a recurrence descriptor \p Desc
398/// The target is queried to determine if intrinsics or shuffle sequences are
399/// required to implement the reduction.
400/// Fast-math-flags are propagated using the RecurrenceDescriptor.
401Value *createTargetReduction(IRBuilderBase &B, const TargetTransformInfo *TTI,
402                             const RecurrenceDescriptor &Desc, Value *Src,
403                             PHINode *OrigPhi = nullptr);
404 
405/// Create an ordered reduction intrinsic using the given recurrence
406/// descriptor \p Desc.
407Value *createOrderedReduction(IRBuilderBase &B,
408                              const RecurrenceDescriptor &Desc, Value *Src,
409                              Value *Start);
410 
411/// Get the intersection (logical and) of all of the potential IR flags
412/// of each scalar operation (VL) that will be converted into a vector (I).
413/// If OpValue is non-null, we only consider operations similar to OpValue
414/// when intersecting.
415/// Flag set: NSW, NUW (if IncludeWrapFlags is true), exact, and all of
416/// fast-math.
417void propagateIRFlags(Value *I, ArrayRef<Value *> VL, Value *OpValue = nullptr,
418                      bool IncludeWrapFlags = true);
419 
420/// Returns true if we can prove that \p S is defined and always negative in
421/// loop \p L.
422bool isKnownNegativeInLoop(const SCEV *S, const Loop *L, ScalarEvolution &SE);
423 
424/// Returns true if we can prove that \p S is defined and always non-negative in
425/// loop \p L.
426bool isKnownNonNegativeInLoop(const SCEV *S, const Loop *L,
427                              ScalarEvolution &SE);
428 
429/// Returns true if \p S is defined and never is equal to signed/unsigned max.
430bool cannotBeMaxInLoop(const SCEV *S, const Loop *L, ScalarEvolution &SE,
431                       bool Signed);
432 
433/// Returns true if \p S is defined and never is equal to signed/unsigned min.
434bool cannotBeMinInLoop(const SCEV *S, const Loop *L, ScalarEvolution &SE,
435                       bool Signed);
436 
437enum ReplaceExitVal {
438  NeverRepl,
439  OnlyCheapRepl,
440  NoHardUse,
441  UnusedIndVarInLoop,
442  AlwaysRepl
443};
444 
445/// If the final value of any expressions that are recurrent in the loop can
446/// be computed, substitute the exit values from the loop into any instructions
447/// outside of the loop that use the final values of the current expressions.
448/// Return the number of loop exit values that have been replaced, and the
449/// corresponding phi node will be added to DeadInsts.
450int rewriteLoopExitValues(Loop *L, LoopInfo *LI, TargetLibraryInfo *TLI,
451                          ScalarEvolution *SE, const TargetTransformInfo *TTI,
452                          SCEVExpander &Rewriter, DominatorTree *DT,
453                          ReplaceExitVal ReplaceExitValue,
454                          SmallVector<WeakTrackingVH, 16> &DeadInsts);
455 
456/// Set weights for \p UnrolledLoop and \p RemainderLoop based on weights for
457/// \p OrigLoop and the following distribution of \p OrigLoop iteration among \p
458/// UnrolledLoop and \p RemainderLoop. \p UnrolledLoop receives weights that
459/// reflect TC/UF iterations, and \p RemainderLoop receives weights that reflect
460/// the remaining TC%UF iterations.
461///
462/// Note that \p OrigLoop may be equal to either \p UnrolledLoop or \p
463/// RemainderLoop in which case weights for \p OrigLoop are updated accordingly.
464/// Note also behavior is undefined if \p UnrolledLoop and \p RemainderLoop are
465/// equal. \p UF must be greater than zero.
466/// If \p OrigLoop has no profile info associated nothing happens.
467///
468/// This utility may be useful for such optimizations as unroller and
469/// vectorizer as it's typical transformation for them.
470void setProfileInfoAfterUnrolling(Loop *OrigLoop, Loop *UnrolledLoop,
471                                  Loop *RemainderLoop, uint64_t UF);
472 
473/// Utility that implements appending of loops onto a worklist given a range.
474/// We want to process loops in postorder, but the worklist is a LIFO data
475/// structure, so we append to it in *reverse* postorder.
476/// For trees, a preorder traversal is a viable reverse postorder, so we
477/// actually append using a preorder walk algorithm.
478template <typename RangeT>
479void appendLoopsToWorklist(RangeT &&, SmallPriorityWorklist<Loop *, 4> &);
480/// Utility that implements appending of loops onto a worklist given a range.
481/// It has the same behavior as appendLoopsToWorklist, but assumes the range of
482/// loops has already been reversed, so it processes loops in the given order.
483template <typename RangeT>
484void appendReversedLoopsToWorklist(RangeT &&,
485                                   SmallPriorityWorklist<Loop *, 4> &);
486 
487/// Utility that implements appending of loops onto a worklist given LoopInfo.
488/// Calls the templated utility taking a Range of loops, handing it the Loops
489/// in LoopInfo, iterated in reverse. This is because the loops are stored in
490/// RPO w.r.t. the control flow graph in LoopInfo. For the purpose of unrolling,
491/// loop deletion, and LICM, we largely want to work forward across the CFG so
492/// that we visit defs before uses and can propagate simplifications from one
493/// loop nest into the next. Calls appendReversedLoopsToWorklist with the
494/// already reversed loops in LI.
495/// FIXME: Consider changing the order in LoopInfo.
496void appendLoopsToWorklist(LoopInfo &, SmallPriorityWorklist<Loop *, 4> &);
497 
498/// Recursively clone the specified loop and all of its children,
499/// mapping the blocks with the specified map.
500Loop *cloneLoop(Loop *L, Loop *PL, ValueToValueMapTy &VM,
501                LoopInfo *LI, LPPassManager *LPM);
502 
503/// Add code that checks at runtime if the accessed arrays in \p PointerChecks
504/// overlap. Returns the final comparator value or NULL if no check is needed.
505Value *
506addRuntimeChecks(Instruction *Loc, Loop *TheLoop,
507                 const SmallVectorImpl<RuntimePointerCheck> &PointerChecks,
508                 SCEVExpander &Expander);
509 
510Value *addDiffRuntimeChecks(
511    Instruction *Loc, ArrayRef<PointerDiffInfo> Checks, SCEVExpander &Expander,
512    function_ref<Value *(IRBuilderBase &, unsigned)> GetVF, unsigned IC);
513 
514/// Struct to hold information about a partially invariant condition.
515struct IVConditionInfo {
516  /// Instructions that need to be duplicated and checked for the unswitching
517  /// condition.
518  SmallVector<Instruction *> InstToDuplicate;
519 
520  /// Constant to indicate for which value the condition is invariant.
521  Constant *KnownValue = nullptr;
4
←
Null pointer value stored to 'PartialIVInfo.KnownValue'→
522 
523  /// True if the partially invariant path is no-op (=does not have any
524  /// side-effects and no loop value is used outside the loop).
525  bool PathIsNoop = true;
526 
527  /// If the partially invariant path reaches a single exit block, ExitForPath
528  /// is set to that block. Otherwise it is nullptr.
529  BasicBlock *ExitForPath = nullptr;
530};
531 
532/// Check if the loop header has a conditional branch that is not
533/// loop-invariant, because it involves load instructions. If all paths from
534/// either the true or false successor to the header or loop exists do not
535/// modify the memory feeding the condition, perform 'partial unswitching'. That
536/// is, duplicate the instructions feeding the condition in the pre-header. Then
537/// unswitch on the duplicated condition. The condition is now known in the
538/// unswitched version for the 'invariant' path through the original loop.
539///
540/// If the branch condition of the header is partially invariant, return a pair
541/// containing the instructions to duplicate and a boolean Constant to update
542/// the condition in the loops created for the true or false successors.
543Optional<IVConditionInfo> hasPartialIVCondition(Loop &L, unsigned MSSAThreshold,
544                                                MemorySSA &MSSA, AAResults &AA);
545 
546} // end namespace llvm
547 
548#endif // LLVM_TRANSFORMS_UTILS_LOOPUTILS_H