MergeSimilarFunctions 1/n: a code size pass to merge functions with small differences

[hiraditya]

Summary

Porting the merge similar functions along with bugfixes.
https://reviews.llvm.org/D22051

Not sure if I should commandeer the patch there of create a new one. I am porting this because I have subsequent patches to enable this during thin-lto (https://llvm.org/devmtg/2018-10/talk-abstracts.html#talk2). Existing Function Merging optimization is not as good as this one because this optimization can merge functions with slight differences as well.

---

This patch contains a version of the MergeFunctions pass that is able to merge
not just identical functions, but also functions with small differences in
their instructions to reduce code size. It does this by inserting control flow
and an additional argument in the merged function to account for the
differences.

I originally presented this work at the LLVM Dev Meeting in 2013 [1]. A more
detailed description was published in a paper at LCTES 2014 [2]. The code was
released to the community at the time, but that version has since bitrotted and
also contained a number of bugs. Meanwhile, the pass has been in production use
at QuIC for the past few years and has been actively maintained internally.

I continue to receive requests for this code every few months, so there
definitely seems to be interest for this optimization out there and I'd like to
share it with the community once again. It has now diverged quite significantly
from the upstream MergeFunctions pass, so adding it as a separate, optional
pass would probably make the most sense.

The main disadvantage - which we have not addressed since 2013 - is that the
worst-case behavior of this optimization is quadratic in the number of
functions in the module. However, thanks to hashing, this is very rarely a
problem in practice. In 3 years of production use at QuIC over large code
bases, we've seen just one source file that triggered excessive compile times
(it was auto-generated and contained several thousand nearly-identical small
functions). We even run this in (fat) LTO without problems.

Note that the patch as-is will only attempt to merge functions that were
compiled with -Os. It's best to follow it by a run of jump threading,
instcombine, and simplifycfg to clean up.

[1] http://llvm.org/devmtg/2013-11/#talk3
[2] http://dl.acm.org/citation.cfm?id=2597811

[jfb]

I don't understand: why is it desirable to have this as a separate thing compared to merge funcs? It seems better to use the same code for both, and when calling mergefuncs pass in a mode which chooses to merge exact matches, or similar matches.

Do you have up-to-date compile-time numbers (at O0 and O2), as well as size numbers?

[hiraditya]

DEBUG -> LLVM_DEBUG

[hiraditya]

Compiler error with CloneType

[alexey.zhikhar]

@hiraditya Thank you very much for the code, I'm very interested in it. I'm trying to compile this and the 2/n revision together but I'm getting compilation failures, e.g.:

include/llvm/IR/ModuleSummaryIndex.h|967 col 12| error: ‘HostSimilarFunction’ was not declared in this scope

Is some part of the code is missing or is it intentional and will be fixed in the subsequent patches?

[hiraditya]

rebase

[hiraditya]

fix clonetype tests

[rjf]

Merges D52896 with trunk

llvm/include/llvm/IR/ModuleSummaryIndex.h

FunctionSummary calculateCallGraphRoot() {
     if (Edges.empty()) {
       // Failed to find root - return an empty node
       return FunctionSummary::makeDummyFunctionSummary({});
     }
     auto CallGraphRoot = FunctionSummary::makeDummyFunctionSummary(Edges);
     return CallGraphRoot;
   }
 
+  bool hasSimilarFunction(unsigned id) { return SimilarFunctions.count(id); }
+
+  void addToSimilarFunctions(unsigned id, GlobalValue::GUID GUID) {
+    if (id == 0)
+      return; // invalid id.
+    if (DuplicateFunctions.count(GUID))
+      return;
+    ValueInfo VI = getValueInfo(GUID);
+    if (!VI)
+      return;
+    // assert(VI.getSummaryList().size() == 1);
+    if (VI.getSummaryList().size() != 1)
+      return;
+    GlobalValueSummary *S = VI.getSummaryList()[0].get();
+    if (!isa<FunctionSummary>(S))
+      return;
+    assert(isa<FunctionSummary>(S) && "Not a function summary!");
+    if (FunctionSimilarityHashes.count(GUID)) {
+      // Erase the GUID having multiple visits in the ModuleSummaryIndex.
+      FunctionSimilarityHashes.erase(GUID);
+      DuplicateFunctions.insert(GUID);
+      return;
+    }
+
+    FunctionSimilarityHashes[GUID] = id;
+  }
+
+  void populateReverseSimilarityHashMap() {
+    for (auto &p : FunctionSimilarityHashes)
+      SimilarFunctions[p.second].push_back(p.first);
+  }
+
+  void removeSingleEntriesFromSimHashMaps() {
+    // Iterate over the hash to remove entries with no duplicates.
+    for (auto I = SimilarFunctions.begin(), E = SimilarFunctions.end();
+         I != E;) {
+      auto Next = std::next(I);
+      assert(I->second.size() && "Empty Entry!");
+      if (I->second.size() == 1) {
+        FunctionSimilarityHashes.erase(I->second[0]);
+        SimilarFunctions.erase(I);
+      }
+      I = Next;
+    }
+  }
+
+  std::map<unsigned, std::vector<GlobalValue::GUID>> &getSimilarFunctions() {
+    return SimilarFunctions;
+  }
+
+  const std::map<unsigned, std::vector<GlobalValue::GUID>> &
+  getSimilarFunctions() const {
+    return SimilarFunctions;
+  }
+
+  unsigned getSimilarityHash(GlobalValue::GUID ID) const {
+    return FunctionSimilarityHashes.find(ID)->second;
+  }
+
+  std::map<GlobalValue::GUID, unsigned> &getSimilarFunctionsHash() {
+    return FunctionSimilarityHashes;
+  }
+
+  const std::map<GlobalValue::GUID, unsigned> &getSimilarFunctionsHash() const {
+    return FunctionSimilarityHashes;
+  }
+
+  void addToHostSimilarFunction(GlobalValue::GUID ID) {
+    HostSimilarFunction.insert(ID);
+  }
+
+  std::set<GlobalValue::GUID> &getHostSimilarFunction() {
+    return HostSimilarFunction;
+  }
+
+  const std::set<GlobalValue::GUID> &getHostSimilarFunction() const {
+    return HostSimilarFunction;
+  }
+
   bool withGlobalValueDeadStripping() const {
     return WithGlobalValueDeadStripping;
   }
   void setWithGlobalValueDeadStripping() {
     WithGlobalValueDeadStripping = true;
   }
 
   bool withAttributePropagation() const { return WithAttributePropagation; }

llvm/include/llvm/InitializePasses.h

 void initializeMemCpyOptLegacyPassPass(PassRegistry&);
 void initializeMemDepPrinterPass(PassRegistry&);
 void initializeMemDerefPrinterPass(PassRegistry&);
 void initializeMemoryDependenceWrapperPassPass(PassRegistry&);
 void initializeMemorySSAPrinterLegacyPassPass(PassRegistry&);
 void initializeMemorySSAWrapperPassPass(PassRegistry&);
 void initializeMemorySanitizerLegacyPassPass(PassRegistry&);
 void initializeMergeFunctionsLegacyPassPass(PassRegistry&);
+void initializeMergeFunctionsPass(PassRegistry&);
+void initializeMergeSimilarFunctionsPass(PassRegistry&);
 void initializeMergeICmpsLegacyPassPass(PassRegistry &);
 void initializeMergedLoadStoreMotionLegacyPassPass(PassRegistry&);
 void initializeMetaRenamerPass(PassRegistry&);
 void initializeModuleDebugInfoPrinterPass(PassRegistry&);
 void initializeModuleSummaryIndexWrapperPassPass(PassRegistry&);
 void initializeModuloScheduleTestPass(PassRegistry&);
 void initializeMustExecutePrinterPass(PassRegistry&);
 void initializeMustBeExecutedContextPrinterPass(PassRegistry&);

llvm/include/llvm/LinkAllPasses.h

ForcePassLinking() {
       (void) llvm::createPostDomTree();
       (void) llvm::createInstructionNamerPass();
       (void) llvm::createMetaRenamerPass();
       (void) llvm::createAttributorLegacyPass();
       (void) llvm::createAttributorCGSCCLegacyPass();
       (void) llvm::createPostOrderFunctionAttrsLegacyPass();
       (void) llvm::createReversePostOrderFunctionAttrsPass();
       (void) llvm::createMergeFunctionsPass();
+      (void) llvm::createMergeSimilarFunctionsPass();
       (void) llvm::createMergeICmpsLegacyPass();
       (void) llvm::createExpandMemCmpPass();
       std::string buf;
       llvm::raw_string_ostream os(buf);
       (void) llvm::createPrintModulePass(os);
       (void) llvm::createPrintFunctionPass(os);
       (void) llvm::createModuleDebugInfoPrinterPass();
       (void) llvm::createPartialInliningPass();

llvm/include/llvm/Transforms/IPO.h

 ModulePass *createMergeFunctionsPass();
 
 //===----------------------------------------------------------------------===//
 /// createHotColdSplittingPass - This pass outlines cold blocks into a separate
 /// function(s).
 ModulePass *createHotColdSplittingPass();
 
 //===----------------------------------------------------------------------===//
+/// createMergeSimilarFunctionsPass - This pass discovers similar functions and
+/// merges them.
+///
+ModulePass *
+createMergeSimilarFunctionsPass(const ModuleSummaryIndex *S = nullptr);
+
+//===----------------------------------------------------------------------===//
 /// createPartialInliningPass - This pass inlines parts of functions.
 ///
 ModulePass *createPartialInliningPass();
 
 //===----------------------------------------------------------------------===//
 /// createBarrierNoopPass - This pass is purely a module pass barrier in a pass
 /// manager.
 ModulePass *createBarrierNoopPass();

llvm/include/llvm/Transforms/Utils/Cloning.h

 /// non-null.
 ///
 /// VMap contains no non-identity GlobalValue mappings and debug info metadata
 /// will not be cloned.
 ///
 Function *CloneFunction(Function *F, ValueToValueMapTy &VMap,
                         ClonedCodeInfo *CodeInfo = nullptr);
 
+/// Used to control @fn CloneFunctionInto.
+enum class CloneType {
+  InvalidCloneType,
+  /// Cloning will result in module level changes.
+  ModuleLevelChanges,
+  /// !ModuleLevelChanges, When no module level changes will be made to the
+  /// cloned function.
+  NoModuleLevelChanges,
+  /// Cloning will be used for extracting functions by passes like function
+  /// merging, it does not require module level changes but debug info needs
+  /// special treatment like: DISubprogram is not cloned.
+  ExtractingFunctions,
+};
+
 /// Clone OldFunc into NewFunc, transforming the old arguments into references
 /// to VMap values.  Note that if NewFunc already has basic blocks, the ones
 /// cloned into it will be added to the end of the function.  This function
 /// fills in a list of return instructions, and can optionally remap types
 /// and/or append the specified suffix to all values cloned.
 ///
 /// If ModuleLevelChanges is false, VMap contains no non-identity GlobalValue
 /// mappings.
 ///
 void CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
-                       ValueToValueMapTy &VMap, bool ModuleLevelChanges,
+                       ValueToValueMapTy &VMap, CloneType CT,
                        SmallVectorImpl<ReturnInst*> &Returns,
                        const char *NameSuffix = "",
                        ClonedCodeInfo *CodeInfo = nullptr,
                        ValueMapTypeRemapper *TypeMapper = nullptr,
                        ValueMaterializer *Materializer = nullptr);
 
 void CloneAndPruneIntoFromInst(Function *NewFunc, const Function *OldFunc,
                                const Instruction *StartingInst,
                                ValueToValueMapTy &VMap, bool ModuleLevelChanges,

llvm/lib/Analysis/ModuleSummaryAnalysis.cpp

cl::values(clEnumValN(FunctionSummary::FSHT_None, "none", "None."),
                           "all-non-critical", "All non-critical edges."),
                clEnumValN(FunctionSummary::FSHT_All, "all", "All edges.")));
 
 cl::opt<std::string> ModuleSummaryDotFile(
     "module-summary-dot-file", cl::init(""), cl::Hidden,
     cl::value_desc("filename"),
     cl::desc("File to emit dot graph of new summary into."));
 
+cl::opt<bool> UseGlobalAliases(
+    "mergesimilarfunc-global-aliases", cl::Hidden, cl::init(false),
+    cl::desc("Enable writing alias by enabling global aliases"));
+
+cl::opt<unsigned> MergeMinInsts(
+    "mergesimilarfunc-min-insts", cl::Hidden, cl::init(4),
+    cl::desc("Min instructions required to even consider single block fns"));
+
+// Minimize the name pollution caused by the enum values.
+namespace Opt {
+cl::opt<enum MergeLevelEnum> MergeLevel(
+    "mergesimilarfunc-level", cl::Hidden, cl::ZeroOrMore,
+    cl::desc("Level of function merging:"), cl::init(size),
+    cl::values(clEnumVal(none, "function merging disabled"),
+               clEnumVal(size, "only try to merge functions that are optimized "
+                               "for size"),
+               clEnumVal(all, "attempt to merge all similar functions")));
+}
+
+namespace llvm {
+
+static const char *MERGED_SUFFIX = "__merged";
+
+/// Returns the type id for a type to be hashed. We turn pointer types into
+/// integers here because the actual compare logic below considers pointers and
+/// integers of the same size as equal.
+static Type::TypeID getTypeIDForHash(Type *Ty) {
+  if (Ty->isPointerTy())
+    return Type::IntegerTyID;
+  return Ty->getTypeID();
+}
+
+bool isAliasCapable(const Function* G) {
+  return
+    UseGlobalAliases && G->hasGlobalUnnamedAddr()
+    && (G->hasExternalLinkage() || G->hasLocalLinkage() || G->hasWeakLinkage());
+}
+
+bool isComparisonCandidate(const Function *F) {
+  if (Opt::MergeLevel == Opt::size) {
+    // Only consider functions that are to be optimized for size.
+    // By default, that is all functions at -Os/-Oz and nothing at -O2.
+    bool Os = F->getAttributes().
+      hasAttribute(AttributeList::FunctionIndex, Attribute::OptimizeForSize);
+    bool Oz = F->getAttributes().
+      hasAttribute(AttributeList::FunctionIndex, Attribute::MinSize);
+    if (!Os && !Oz)
+      return false;
+  }
+
+  // Ignore declarations and tiny functions - no point in merging those
+  if (F->isDeclaration()) return false;
+  if (F->getName().endswith(MERGED_SUFFIX)) return false;
+  if (F->hasAvailableExternallyLinkage()) return false;
+  if (F->hasFnAttribute(Attribute::AlwaysInline)) return false;
+  if (F->size() == 1 && F->begin()->size() < MergeMinInsts)
+    return isAliasCapable(F);
+
+  return true;
+}
+
+unsigned profileFunction(const Function *F) {
+  FunctionType *FTy = F->getFunctionType();
+  if (!isComparisonCandidate(F))
+    return 0;
+  if (F->hasGC() || FTy->isVarArg() || !F->hasExactDefinition())
+    return 0;
+  FoldingSetNodeID ID;
+  ID.AddInteger(F->size());
+  ID.AddInteger(F->getCallingConv());
+  // Add pure attribute, has side-effects attribute.
+  ID.AddBoolean(F->hasFnAttribute(Attribute::NoUnwind));
+  ID.AddBoolean(F->hasFnAttribute(Attribute::NoReturn));
+  //ID.AddBoolean(F->hasGC());
+  //ID.AddBoolean(F->isInterposable());
+  //ID.AddBoolean(FTy->isVarArg());
+  ID.AddInteger(getTypeIDForHash(FTy->getReturnType()));
+  for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
+    ID.AddInteger(getTypeIDForHash(FTy->getParamType(i)));
+  return ID.ComputeHash();
+}
+}
+
 // Walk through the operands of a given User via worklist iteration and populate
 // the set of GlobalValue references encountered. Invoked either on an
 // Instruction or a GlobalVariable (which walks its initializer).
 // Return true if any of the operands contains blockaddress. This is important
 // to know when computing summary for global var, because if global variable
 // references basic block address we can't import it separately from function
 // containing that basic block. For simplicity we currently don't import such
 // global vars at all. When importing function we aren't interested if any

llvm/lib/ExecutionEngine/Orc/IndirectionUtils.cpp

void moveFunctionBody(Function &OrigF, ValueToValueMapTy &VMap,
   else
     assert(VMap[&OrigF] == NewF && "Incorrect function mapping in VMap.");
   assert(NewF && "Function mapping missing from VMap.");
   assert(NewF->getParent() != OrigF.getParent() &&
          "moveFunctionBody should only be used to move bodies between "
          "modules.");
 
   SmallVector<ReturnInst *, 8> Returns; // Ignore returns cloned.
-  CloneFunctionInto(NewF, &OrigF, VMap, /*ModuleLevelChanges=*/true, Returns,
+  CloneFunctionInto(NewF, &OrigF, VMap, CloneType::ModuleLevelChanges, Returns,
                     "", nullptr, nullptr, Materializer);
   OrigF.deleteBody();
 }
 
 GlobalVariable* cloneGlobalVariableDecl(Module &Dst, const GlobalVariable &GV,
                                         ValueToValueMapTy *VMap) {
   GlobalVariable *NewGV = new GlobalVariable(
       Dst, GV.getValueType(), GV.isConstant(),

llvm/lib/Target/AMDGPU/R600OpenCLImageTypeLoweringPass.cpp

for (auto &Arg: F->args()) {
       NewFArgIt->setName(ArgName);
       VMap[&Arg] = &(*NewFArgIt++);
       if (IsImageType(ArgTypeFromMD(KernelMDNode, Arg.getArgNo()))) {
         (NewFArgIt++)->setName(Twine("__size_") + ArgName);
         (NewFArgIt++)->setName(Twine("__format_") + ArgName);
       }
     }
     SmallVector<ReturnInst*, 8> Returns;
-    CloneFunctionInto(NewF, F, VMap, /*ModuleLevelChanges=*/false, Returns);
+    CloneFunctionInto(NewF, F, VMap, CloneType::NoModuleLevelChanges, Returns);
 
     // Build new MDNode.
     SmallVector<Metadata *, 6> KernelMDArgs;
     KernelMDArgs.push_back(ConstantAsMetadata::get(NewF));
     for (unsigned i = 0; i < NumKernelArgMDNodes; ++i)
       KernelMDArgs.push_back(MDNode::get(*Context, NewArgMDs.ArgVector[i]));
     MDNode *NewMDNode = MDNode::get(*Context, KernelMDArgs);
 

llvm/lib/Transforms/Coroutines/CoroSplit.cpp

void CoroCloner::create() {
   auto savedDLLStorageClass = NewF->getDLLStorageClass();
 
   // NewF's linkage (which CloneFunctionInto does *not* change) might not
   // be compatible with the visibility of OrigF (which it *does* change),
   // so protect against that.
   auto savedLinkage = NewF->getLinkage();
   NewF->setLinkage(llvm::GlobalValue::ExternalLinkage);
 
-  CloneFunctionInto(NewF, &OrigF, VMap, /*ModuleLevelChanges=*/true, Returns);
+  CloneFunctionInto(NewF, &OrigF, VMap, CloneType::ModuleLevelChanges, Returns);
 
   NewF->setLinkage(savedLinkage);
   NewF->setVisibility(savedVisibility);
   NewF->setUnnamedAddr(savedUnnamedAddr);
   NewF->setDLLStorageClass(savedDLLStorageClass);
 
   auto &Context = NewF->getContext();
 

llvm/lib/Transforms/IPO/CMakeLists.txt

add_llvm_component_library(LLVMipo
   InferFunctionAttrs.cpp
   InlineSimple.cpp
   Inliner.cpp
   Internalize.cpp
   LoopExtractor.cpp
   LowerTypeTests.cpp
   MergeFunctions.cpp
   OpenMPOpt.cpp
+  MergeSimilarFunctions.cpp
   PartialInlining.cpp
   PassManagerBuilder.cpp
   PruneEH.cpp
   SampleProfile.cpp
   SCCP.cpp
   StripDeadPrototypes.cpp
   StripSymbols.cpp
   SyntheticCountsPropagation.cpp

llvm/lib/Transforms/IPO/IPO.cpp

void llvm::initializeIPO(PassRegistry &Registry) {
   initializeSimpleInlinerPass(Registry);
   initializeInferFunctionAttrsLegacyPassPass(Registry);
   initializeInternalizeLegacyPassPass(Registry);
   initializeLoopExtractorPass(Registry);
   initializeBlockExtractorPass(Registry);
   initializeSingleLoopExtractorPass(Registry);
   initializeLowerTypeTestsPass(Registry);
   initializeMergeFunctionsLegacyPassPass(Registry);
+  initializeMergeFunctionsPass(Registry);
+  initializeMergeSimilarFunctionsPass(Registry);
   initializePartialInlinerLegacyPassPass(Registry);
   initializeAttributorLegacyPassPass(Registry);
   initializeAttributorCGSCCLegacyPassPass(Registry);
   initializePostOrderFunctionAttrsLegacyPassPass(Registry);
   initializeReversePostOrderFunctionAttrsLegacyPassPass(Registry);
   initializePruneEHPass(Registry);
   initializeIPSCCPLegacyPassPass(Registry);
   initializeStripDeadPrototypesLegacyPassPass(Registry);

llvm/lib/Transforms/IPO/MergeSimilarFunctions.cpp

+//===- MergeSimilarFunctions.cpp - Merge similar functions ----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass merges both equivalent and similar functions to reduce code size.
+//
+// For a more detailed explanation of the approach, see:
+// Edler von Koch et al. "Exploiting Function Similarity for Code Size
+// Reduction", LCTES 2014.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "mergesimilarfunc"
+#include "llvm/Transforms/IPO.h"
+#include "llvm/ADT/DenseSet.h"
+#include "llvm/ADT/FoldingSet.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/MapVector.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/IR/CallSite.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InlineAsm.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Operator.h"
+#include "llvm/IR/ValueHandle.h"
+#include "llvm/Pass.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/Format.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "llvm/Transforms/Utils/ValueMapper.h"
+#include <vector>
+#include <list>
+using namespace llvm;
+
+STATISTIC(NumFunctionsMerged, "Number of functions merged");
+STATISTIC(NumThunksWritten, "Number of thunks generated");
+STATISTIC(NumAliasesWritten, "Number of aliases generated");
+STATISTIC(NumDoubleWeak, "Number of new functions created");
+STATISTIC(NumMultiMerged, "Number of multi-merged functions");
+
+STATISTIC(NumSimilarFunctionsMerged, "Number of similar functions merged");
+
+static cl::opt<unsigned> MergeMinInsts(
+    "mergesimilarfunc-min-insts", cl::Hidden, cl::init(4),
+    cl::desc("Min instructions required to even consider single block fns"));
+
+static cl::opt<unsigned> MergeDifferingMinInsts(
+    "mergesimilarfunc-diff-min-insts", cl::Hidden, cl::init(15),
+    cl::desc("Min instructions required to try merging differing functions"));
+
+static cl::opt<unsigned> MergeMaxDiffering(
+    "mergesimilarfunc-max-diff", cl::Hidden, cl::init(8),
+    cl::desc("Maximum number of differing instructions allowed"));
+
+static cl::opt<unsigned> MergeMinSimilarity(
+    "mergesimilarfunc-min-similarity", cl::Hidden, cl::init(70),
+    cl::desc("Minimum percentage of similar instructions required"));
+
+static cl::opt<bool> OptPrintMerges("mergesimilarfunc-print-merges", cl::Hidden,
+                                    cl::init(false));
+
+static cl::opt<bool> UseGlobalAliases(
+    "mergesimilarfunc-global-aliases", cl::Hidden, cl::init(false),
+    cl::desc("Enable writing alias by enabling global aliases"));
+
+void PrintMerges(const char *Desc, Function *Old, Function *New) {
+  if (OptPrintMerges) {
+    dbgs() << "=== [" << Desc << "] replacing " << Old->getName() << " with "
+           << New->getName() << "\n";
+  }
+}
+
+// Minimize the name pollution caused by the enum values.
+namespace Opt {
+enum MergeLevelEnum { none, size, all };
+static cl::opt<enum MergeLevelEnum> MergeLevel(
+    "mergesimilarfunc-level", cl::Hidden, cl::ZeroOrMore,
+    cl::desc("Level of function merging:"), cl::init(size),
+    cl::values(clEnumVal(none, "function merging disabled"),
+               clEnumVal(size, "only try to merge functions that are optimized "
+                               "for size"),
+               clEnumVal(all, "attempt to merge all similar functions")));
+}
+
+static const char *MERGED_SUFFIX = "__merged";
+
+/// Returns the type id for a type to be hashed. We turn pointer types into
+/// integers here because the actual compare logic below considers pointers and
+/// integers of the same size as equal.
+static Type::TypeID getTypeIDForHash(Type *Ty) {
+  if (Ty->isPointerTy())
+    return Type::IntegerTyID;
+  return Ty->getTypeID();
+}
+
+/// Creates a hash-code for the function which is the same for any two
+/// functions that will compare equal, without looking at the instructions
+/// inside the function.
+static unsigned profileFunction(const Function *F) {
+  FunctionType *FTy = F->getFunctionType();
+
+  FoldingSetNodeID ID;
+  ID.AddInteger(F->size());
+  ID.AddInteger(F->getCallingConv());
+  ID.AddBoolean(F->hasGC());
+  ID.AddBoolean(F->isInterposable());
+  ID.AddBoolean(FTy->isVarArg());
+  ID.AddInteger(getTypeIDForHash(FTy->getReturnType()));
+  for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
+    ID.AddInteger(getTypeIDForHash(FTy->getParamType(i)));
+  return ID.ComputeHash();
+}
+
+
+/// Replace Inst1 by a switch statement that executes Inst1 or one of Inst2s
+/// depending on the value of SwitchVal. If a value in Inst2s is NULL, it
+/// defaults to executing Inst1. Returns set of terminator instructions of newly
+/// created switch blocks in Ret.
+///
+/// For instance, the transformation may look as follows:
+///         ...Head...
+///           Inst1           with all of Insts2s without parents
+///         ...Tail...
+///  into
+///         ...Head...
+///           Switch
+///         /     |       \                    .
+///    (default) (1)       (2)
+///      Inst1   Inst2s[0] Inst2s[1]
+///     Ret[0]   Ret[1]    Ret[2]
+///        \      |       /
+///         ...Tail...
+///
+static void SplitBlockAndInsertSwitch(
+    Value *SwitchVal, Instruction *Inst1,
+    SmallVectorImpl<Instruction *> &Inst2s,
+    SmallVectorImpl<Instruction *> &Ret) {
+  // Split block
+  BasicBlock *Head = Inst1->getParent();
+  BasicBlock *Tail = Head->splitBasicBlock(Inst1);
+
+  // Create default block
+  LLVMContext &C = Head->getContext();
+  BasicBlock *DefaultBlock = BasicBlock::Create(C, "", Head->getParent(), Tail);
+
+  // Insert switch instruction at end of Head
+  Instruction *HeadOldTerm = Head->getTerminator();
+  SwitchInst *Switch = SwitchInst::Create(SwitchVal, DefaultBlock,
+                                               Inst2s.size());
+  ReplaceInstWithInst(HeadOldTerm, Switch);
+
+  // Move instructions into the blocks
+  if (Inst1->isTerminator()) {
+    Inst1->removeFromParent();
+    DefaultBlock->getInstList().push_back(Inst1);
+    Ret.push_back(cast<Instruction>(Inst1));
+  } else {
+    Instruction *DefaultTerm = BranchInst::Create(Tail, DefaultBlock);
+    Inst1->moveBefore(DefaultTerm);
+    Ret.push_back(DefaultTerm);
+  }
+
+  for (unsigned InstPos = 0, InstNum = Inst2s.size(); InstPos < InstNum;
+       ++InstPos) {
+    Instruction *Inst2 = Inst2s[InstPos];
+    if (!Inst2) {
+      Ret.push_back(NULL);
+      continue;
+    }
+
+    BasicBlock *CaseBlock = BasicBlock::Create(C, "", Head->getParent(), Tail);
+
+    // Update the debug information of the merged instruction by marking it as
+    // 'inlined' at this location. If only Inst1 or Inst2 has debug
+    // information, we try to do something sensible that won't break the
+    // verifier.
+    if (Inst1->getDebugLoc()) {
+      if (Inst2->getDebugLoc()) {
+        const DebugLoc &I2Loc = Inst2->getDebugLoc();
+        Inst2->setDebugLoc(
+            DebugLoc::get(I2Loc.getLine(), I2Loc.getCol(), I2Loc.getScope(),
+                          /*InlinedAt*/ Inst1->getDebugLoc().getAsMDNode()));
+      } else {
+        Inst2->setDebugLoc(Inst1->getDebugLoc());
+      }
+    } else if (Inst2->getDebugLoc()) {
+      Inst2->setDebugLoc(DebugLoc());
+    }
+
+    if (Inst2->isTerminator()) {
+      assert(Inst1->isTerminator() &&
+        "Inst1 and Inst2 must both be terminators or non-terminators!");
+      CaseBlock->getInstList().push_back(Inst2);
+      Ret.push_back(cast<Instruction>(Inst2));
+    } else {
+      Instruction *CaseTerm = BranchInst::Create(Tail, CaseBlock);
+      Inst2->insertBefore(CaseTerm);
+      Ret.push_back(CaseTerm);
+    }
+
+    Switch->addCase(ConstantInt::get(cast<IntegerType>(SwitchVal->getType()),
+                                     InstPos+1),
+                    CaseBlock);
+  }
+
+  // If Inst1 (and Inst2s) are Terminator Inst's, Tail will be empty and can be
+  // deleted now. We also need to update PHI nodes to add the additional
+  // incoming blocks from the SwitchInst.
+  if (Inst1->isTerminator()) {
+    for (succ_iterator I = succ_begin(DefaultBlock), E = succ_end(DefaultBlock);
+         I != E; ++I) {
+      BasicBlock *Successor = *I;
+      PHINode *Phi;
+
+      for (BasicBlock::iterator II = Successor->begin();
+           (Phi = dyn_cast<PHINode>(II)); ++II)
+        for (unsigned ValId = 0, ValEnd = Phi->getNumIncomingValues();
+             ValId != ValEnd; ++ValId)
+          if (Phi->getIncomingBlock(ValId) == Tail) {
+            Phi->setIncomingBlock(ValId, DefaultBlock);
+            SmallVectorImpl<Instruction *>::iterator
+              SwitchI = Ret.begin(), SwitchE = Ret.end();
+            for (++SwitchI; SwitchI != SwitchE; ++SwitchI) {
+              if (!*SwitchI)
+                continue;
+              Phi->addIncoming(Phi->getIncomingValue(ValId),
+                               (*SwitchI)->getParent());
+            }
+          }
+    }
+
+    Tail->eraseFromParent();
+  }
+}
+
+/// Insert function NewF into module, placing it immediately after the
+/// existing function PredF. If PredF does not exist, insert at the end.
+static void insertFunctionAfter(Function *NewF, Function *PredF) {
+  Module *M = PredF->getParent();
+  Module::FunctionListType &FList = M->getFunctionList();
+
+  for (Module::FunctionListType::iterator I = FList.begin(), E = FList.end();
+      I != E; ++I) {
+    if (PredF == &*I) {
+      FList.insertAfter(I, NewF);
+      return;
+    }
+  }
+
+  // Couldn't find PredF, insert at end
+  FList.push_back(NewF);
+}
+
+/// Create a cast instruction if needed to cast V to type DstType. We treat
+/// pointer and integer types of the same bitwidth as equivalent, so this can be
+/// used to cast them to each other where needed. The function returns the Value
+/// itself if no cast is needed, or a new CastInst instance inserted before
+/// InsertBefore. The integer type equivalent to pointers must be passed as
+/// IntPtrType (get it from DataLayout). This is guaranteed to generate no-op
+/// casts, otherwise it will assert.
+static Value *createCastIfNeeded(Value *V, Type *DstType,
+                                 Value *InstrOrBB, Type *IntPtrType) {
+  if (V->getType() == DstType)
+    return V;
+
+  BasicBlock *InsertAtEnd = dyn_cast<BasicBlock>(InstrOrBB);
+  Instruction *InsertBefore = dyn_cast<Instruction>(InstrOrBB);
+  BasicBlock *InsertBB = InsertAtEnd ? InsertAtEnd : InsertBefore->getParent();
+
+  CastInst *Result;
+  Type *OrigType = V->getType();
+
+  if (OrigType->isStructTy()) {
+    assert(DstType->isStructTy());
+    assert(OrigType->getStructNumElements() == DstType->getStructNumElements());
+
+    IRBuilder<> Builder(InsertBB);
+    if (InsertBefore)
+      Builder.SetInsertPoint(InsertBefore);
+    Value *Result = UndefValue::get(DstType);
+    for (unsigned int I = 0, E = OrigType->getStructNumElements(); I < E; ++I) {
+      Value *ExtractedValue
+        = Builder.CreateExtractValue(V, ArrayRef<unsigned int>(I));
+      Value *Element = createCastIfNeeded(ExtractedValue,
+                                          DstType->getStructElementType(I),
+                                          InstrOrBB, IntPtrType);
+      Result =
+          Builder.CreateInsertValue(Result, Element, ArrayRef<unsigned int>(I));
+    }
+    return Result;
+  }
+  assert(!DstType->isStructTy());
+
+  if (OrigType->isPointerTy()
+      && (DstType->isIntegerTy() || DstType->isPointerTy())) {
+    if (InsertBefore)
+      Result = CastInst::CreatePointerCast(V, DstType, "", InsertBefore);
+    else
+      Result = CastInst::CreatePointerCast(V, DstType, "", InsertAtEnd);
+  } else if (OrigType->isIntegerTy() && DstType->isPointerTy()
+             && OrigType == IntPtrType) {
+    // Int -> Ptr
+    if (InsertBefore) {
+      Result = CastInst::Create(CastInst::IntToPtr, V, DstType, "",
+                                InsertBefore);
+    } else {
+      Result = CastInst::Create(CastInst::IntToPtr, V, DstType, "",
+                                InsertAtEnd);
+    }
+  } else {
+    llvm_unreachable("Can only cast int -> ptr or ptr -> (ptr or int)");
+  }
+
+  assert(cast<CastInst>(Result)->isNoopCast(IntPtrType) &&
+      "Cast is not a no-op cast. Potential loss of precision");
+
+  return Result;
+}
+
+namespace {
+
+/// ComparableFunction - A struct that pairs together functions with a
+/// DataLayout so that we can keep them together as elements in the DenseSet.
+class ComparableFunction {
+public:
+  ComparableFunction() : Func(0), IsNew(false) { }
+
+  ComparableFunction(const ComparableFunction &that)
+    : Func(that.Func), IsNew(that.IsNew) {
+  }
+
+  ComparableFunction(Function *Func) : Func(Func), IsNew(true) { }
+
+  ~ComparableFunction() { }
+
+  ComparableFunction &operator=(const ComparableFunction &that) {
+    Func = that.Func;
+    IsNew = that.IsNew;
+    return *this;
+  }
+
+  Function *getFunc() const { return Func; }
+  bool isNew() const { return IsNew; }
+
+  // Drops AssertingVH reference to the function. Outside of debug mode, this
+  // does nothing.
+  void release() {
+    assert(Func &&
+           "Attempted to release function twice, or release empty/tombstone!");
+    Func = NULL;
+  }
+
+  void markCompared() {
+    IsNew = false;
+  }
+private:
+  AssertingVH<Function> Func;
+  bool IsNew;
+};
+
+}
+
+namespace {
+
+/// FunctionComparator - Compares two functions to determine whether or not
+/// they will generate machine code with the same behaviour. DataLayout is
+/// used if available. The comparator always fails conservatively (erring on the
+/// side of claiming that two functions are different).
+class FunctionComparator {
+public:
+  FunctionComparator(const DataLayout *DL, Function *F1, Function *F2)
+    : isDifferent(false), isNotMergeable(false),
+      BasicBlockCount(0), InstructionCount(0), DifferingInstructionsCount(0),
+      F1(F1), F2(F2), SimilarityMetric(0), DL(DL), ID(CurID++) {}
+
+  ~FunctionComparator() {}
+
+  /// Test whether the two functions have equivalent behaviour. Returns true if
+  /// they are equal or can be merged, false if not.
+  bool compare();
+
+  /// Indicate whether the two functions are an exact match after comparison
+  bool isExactMatch();
+
+  /// Indicate whether the two functions candidates for merging after comparison
+  bool isMergeCandidate();
+
+  /// Get a similarity metric between the two functions. Higher means more
+  /// similar.
+  unsigned getSimilarityMetric() {
+    if (!SimilarityMetric)
+      SimilarityMetric = (unsigned)(((float)InstructionCount -
+            DifferingInstructionsCount)/InstructionCount*10000);
+    return SimilarityMetric;
+  }
+
+  Function *getF1() { return F1; }
+  Function *getF2() { return F2; }
+  ValueToValueMapTy &getF1toF2Map() { return id_map; }
+  ValueToValueMapTy &getF2toF1Map() { return seen_values; }
+  const DataLayout *getDataLayout() { return DL; }
+
+  /// Assign or look up previously assigned numbers for the two values, and
+  /// return whether the numbers are equal. Numbers are assigned in the order
+  /// visited. If NoSelfRef is set, F1 and F2 are not assigned to each other
+  /// (treated as 'equal').
+  bool enumerate(const Value *V1, const Value *V2, bool NoSelfRef=false);
+
+  /// Compare two Types, treating all pointer types as equal.
+  bool isEquivalentType(Type *Ty1, Type *Ty2) const;
+
+  /// Instructions that differ between the two functions (F1's -> F2's inst).
+  MapVector<const Instruction *, const Instruction *> DifferingInstructions;
+
+  /// Instructions that reference F1/F2 itself (recursive calls etc.)
+  /// These may need special treatment when merging differing functions.
+  MapVector<const Instruction *, const Instruction *> SelfRefInstructions;
+
+  /// Return the unique ID for the object.
+  unsigned getID() { return ID; }
+
+  bool isDifferent;
+  bool isNotMergeable;
+
+  // Comparison statistics
+  unsigned BasicBlockCount;
+  unsigned InstructionCount;
+  unsigned DifferingInstructionsCount;
+
+private:
+  /// Test whether two basic blocks have equivalent behaviour. Returns true if
+  /// they are equal or can be merged, false if not. PHINodes are not compared
+  /// in this function, but added to the PHIsFound list for delayed processing.
+  bool compare(const BasicBlock *BB1, const BasicBlock *BB2,
+      std::list<std::pair<const PHINode*,const PHINode*> > *PHIsFound);
+
+  /// Compare pairs of PHI nodes. Returns true if all pairs are equal or can
+  /// be merged, false if not.
+  bool comparePHIs(
+      const std::list<std::pair<const PHINode*,const PHINode*> > &PHIs);
+
+  /// Compare two Instructions for equivalence, similar to
+  /// Instruction::isSameOperationAs but with modifications to the type
+  /// comparison.
+  bool isEquivalentOperation(const Instruction *I1,
+                             const Instruction *I2) const;
+
+  /// Compare two GEPs for equivalent pointer arithmetic.
+  bool isEquivalentGEP(const GEPOperator *GEP1, const GEPOperator *GEP2);
+  bool isEquivalentGEP(const GetElementPtrInst *GEP1,
+                       const GetElementPtrInst *GEP2) {
+    return isEquivalentGEP(cast<GEPOperator>(GEP1), cast<GEPOperator>(GEP2));
+  }
+
+  // The two functions undergoing comparison.
+  Function *F1, *F2;
+
+  unsigned SimilarityMetric;
+
+  const DataLayout *DL;
+
+  ValueToValueMapTy id_map;
+  ValueToValueMapTy seen_values;
+
+  // Maintain a unique ID for each object.
+  static unsigned CurID;
+  unsigned ID;
+};
+
+}
+
+unsigned FunctionComparator::CurID = 0;
+
+// Any two pointers in the same address space are equivalent, intptr_t and
+// pointers are equivalent. Otherwise, standard type equivalence rules apply.
+bool FunctionComparator::isEquivalentType(Type *Ty1, Type *Ty2) const {
+  if (Ty1 == Ty2)
+    return true;
+  if (Ty1->getTypeID() != Ty2->getTypeID()) {
+    LLVMContext &Ctx = Ty1->getContext();
+    if (isa<PointerType>(Ty1) && Ty2 == DL->getIntPtrType(Ctx)) return true;
+    if (isa<PointerType>(Ty2) && Ty1 == DL->getIntPtrType(Ctx)) return true;
+    return false;
+  }
+
+  switch (Ty1->getTypeID()) {
+  default:
+    llvm_unreachable("Unknown type!");
+    // Fall through in Release mode.
+  case Type::IntegerTyID:
+  case Type::VectorTyID:
+    // Ty1 == Ty2 would have returned true earlier.
+    return false;
+
+  case Type::VoidTyID:
+  case Type::FloatTyID:
+  case Type::DoubleTyID:
+  case Type::X86_FP80TyID:
+  case Type::FP128TyID:
+  case Type::PPC_FP128TyID:
+  case Type::LabelTyID:
+  case Type::MetadataTyID:
+    return true;
+
+  case Type::PointerTyID: {
+    PointerType *PTy1 = cast<PointerType>(Ty1);
+    PointerType *PTy2 = cast<PointerType>(Ty2);
+    return PTy1->getAddressSpace() == PTy2->getAddressSpace();
+  }
+
+  case Type::StructTyID: {
+    StructType *STy1 = cast<StructType>(Ty1);
+    StructType *STy2 = cast<StructType>(Ty2);
+    if (STy1->getNumElements() != STy2->getNumElements())
+      return false;
+
+    if (STy1->isPacked() != STy2->isPacked())
+      return false;
+
+    for (unsigned i = 0, e = STy1->getNumElements(); i != e; ++i) {
+      if (!isEquivalentType(STy1->getElementType(i), STy2->getElementType(i)))
+        return false;
+    }
+    return true;
+  }
+
+  case Type::FunctionTyID: {
+    FunctionType *FTy1 = cast<FunctionType>(Ty1);
+    FunctionType *FTy2 = cast<FunctionType>(Ty2);
+    if (FTy1->getNumParams() != FTy2->getNumParams() ||
+        FTy1->isVarArg() != FTy2->isVarArg())
+      return false;
+
+    if (!isEquivalentType(FTy1->getReturnType(), FTy2->getReturnType()))
+      return false;
+
+    for (unsigned i = 0, e = FTy1->getNumParams(); i != e; ++i) {
+      if (!isEquivalentType(FTy1->getParamType(i), FTy2->getParamType(i)))
+        return false;
+    }
+    return true;
+  }
+
+  case Type::ArrayTyID: {
+    ArrayType *ATy1 = cast<ArrayType>(Ty1);
+    ArrayType *ATy2 = cast<ArrayType>(Ty2);
+    return ATy1->getNumElements() == ATy2->getNumElements() &&
+           isEquivalentType(ATy1->getElementType(), ATy2->getElementType());
+  }
+  }
+}
+
+// Determine whether the two operations are the same except that pointer-to-A
+// and pointer-to-B are equivalent. This should be kept in sync with
+// Instruction::isSameOperationAs.
+bool FunctionComparator::isEquivalentOperation(const Instruction *I1,
+                                               const Instruction *I2) const {
+  // Differences from Instruction::isSameOperationAs:
+  //  * replace type comparison with calls to isEquivalentType.
+  //  * we test for I->hasSameSubclassOptionalData (nuw/nsw/tail) at the top
+  //  * because of the above, we don't test for the tail bit on calls later on
+  if (I1->getOpcode() != I2->getOpcode() ||
+      I1->getNumOperands() != I2->getNumOperands() ||
+      !isEquivalentType(I1->getType(), I2->getType()) ||
+      !I1->hasSameSubclassOptionalData(I2))
+    return false;
+
+  // We have two instructions of identical opcode and #operands.  Check to see
+  // if all operands are the same type
+  for (unsigned i = 0, e = I1->getNumOperands(); i != e; ++i)
+    if (!isEquivalentType(I1->getOperand(i)->getType(),
+                          I2->getOperand(i)->getType()))
+      return false;
+
+  // Check special state that is a part of some instructions.
+  if (const LoadInst *LI = dyn_cast<LoadInst>(I1)) {
+    const LoadInst *LI2 = cast<LoadInst>(I2);
+    return LI->isVolatile() == LI2->isVolatile() &&
+           LI->getAlignment() == LI2->getAlignment() &&
+           LI->getOrdering() == LI2->getOrdering() &&
+           LI->getSyncScopeID() == LI2->getSyncScopeID() &&
+           LI->getMetadata(LLVMContext::MD_range)
+             == LI2->getMetadata(LLVMContext::MD_range);
+  }
+  if (const StoreInst *SI = dyn_cast<StoreInst>(I1))
+    return SI->isVolatile() == cast<StoreInst>(I2)->isVolatile() &&
+           SI->getAlignment() == cast<StoreInst>(I2)->getAlignment() &&
+           SI->getOrdering() == cast<StoreInst>(I2)->getOrdering() &&
+           SI->getSyncScopeID() == cast<StoreInst>(I2)->getSyncScopeID();
+  if (const AllocaInst *AI = dyn_cast<AllocaInst>(I1)) {
+    if (AI->getArraySize() != cast<AllocaInst>(I2)->getArraySize() ||
+        AI->getAlignment() != cast<AllocaInst>(I2)->getAlignment())
+      return false;
+
+    // If size is known, I2 can be seen as equivalent to I1 if it allocates
+    // the same or less memory.
+    if (DL->getTypeAllocSize(AI->getAllocatedType())
+          < DL->getTypeAllocSize(cast<AllocaInst>(I2)->getAllocatedType()))
+      return false;
+
+    return true;
+  }
+  if (const CmpInst *CI = dyn_cast<CmpInst>(I1))
+    return CI->getPredicate() == cast<CmpInst>(I2)->getPredicate();
+  if (const CallInst *CI = dyn_cast<CallInst>(I1))
+    return CI->getCallingConv() == cast<CallInst>(I2)->getCallingConv() &&
+           CI->getAttributes() == cast<CallInst>(I2)->getAttributes();
+  if (const InvokeInst *CI = dyn_cast<InvokeInst>(I1))
+    return CI->getCallingConv() == cast<InvokeInst>(I2)->getCallingConv() &&
+           CI->getAttributes() == cast<InvokeInst>(I2)->getAttributes();
+  if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(I1))
+    return IVI->getIndices() == cast<InsertValueInst>(I2)->getIndices();
+  if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(I1))
+    return EVI->getIndices() == cast<ExtractValueInst>(I2)->getIndices();
+  if (const FenceInst *FI = dyn_cast<FenceInst>(I1))
+    return FI->getOrdering() == cast<FenceInst>(I2)->getOrdering() &&
+           FI->getSyncScopeID() == cast<FenceInst>(I2)->getSyncScopeID();
+  if (const AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(I1)) {
+    const AtomicCmpXchgInst *CXI2 = cast<AtomicCmpXchgInst>(I2);
+    return CXI->isVolatile() == CXI2->isVolatile() &&
+           CXI->isWeak() == CXI2->isWeak() &&
+           CXI->getSuccessOrdering() == CXI2->getSuccessOrdering() &&
+           CXI->getFailureOrdering() == CXI2->getFailureOrdering() &&
+           CXI->getSyncScopeID() == CXI2->getSyncScopeID();
+  }
+  if (const AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(I1))
+    return RMWI->getOperation() == cast<AtomicRMWInst>(I2)->getOperation() &&
+           RMWI->isVolatile() == cast<AtomicRMWInst>(I2)->isVolatile() &&
+           RMWI->getOrdering() == cast<AtomicRMWInst>(I2)->getOrdering() &&
+           RMWI->getSyncScopeID() == cast<AtomicRMWInst>(I2)->getSyncScopeID();
+
+  return true;
+}
+
+// Determine whether two GEP operations perform the same underlying arithmetic.
+bool FunctionComparator::isEquivalentGEP(const GEPOperator *GEP1,
+                                         const GEPOperator *GEP2) {
+  // When we have target data, we can reduce the GEP down to the value in bytes
+  // added to the address.
+  if (GEP1->hasAllConstantIndices() && GEP2->hasAllConstantIndices()) {
+    SmallVector<Value *, 8> Indices1(GEP1->idx_begin(), GEP1->idx_end());
+    SmallVector<Value *, 8> Indices2(GEP2->idx_begin(), GEP2->idx_end());
+    uint64_t Offset1 = DL->getIndexedOffsetInType(GEP1->getSourceElementType(),
+                                            Indices1);
+    uint64_t Offset2 = DL->getIndexedOffsetInType(GEP2->getSourceElementType(),
+                                            Indices2);
+    return Offset1 == Offset2;
+  }
+
+  if (GEP1->getPointerOperand()->getType() !=
+      GEP2->getPointerOperand()->getType())
+    return false;
+
+  if (GEP1->getNumOperands() != GEP2->getNumOperands())
+    return false;
+
+  for (unsigned i = 0, e = GEP1->getNumOperands(); i != e; ++i) {
+    if (!enumerate(GEP1->getOperand(i), GEP2->getOperand(i)))
+      return false;
+  }
+
+  return true;
+}
+
+// Compare two values used by the two functions under pair-wise comparison. If
+// this is the first time the values are seen, they're added to the mapping so
+// that we will detect mismatches on next use.
+bool FunctionComparator::enumerate(const Value *V1, const Value *V2,
+    bool NoSelfRef/*=false*/) {
+  // Check for function @f1 referring to itself and function @f2 referring to
+  // itself. For compatibility with llvm's MergeFunctions, disallow referring to
+  // each other, or both referring to either of them.
+  if (!NoSelfRef && V1 == F1 && V2 == F2)
+    return true;
+
+  // FIXME: This is very conservative for now, but keeping this for thinlto.
+  if (isa<GlobalVariable>(V1) || isa<GlobalVariable>(V2))
+    return false;
+  if (const Constant *C1 = dyn_cast<Constant>(V1)) {
+    if (V1 == V2) return true;
+    const Constant *C2 = dyn_cast<Constant>(V2);
+    if (!C2) return false;
+    // TODO: constant expressions with GEP or references to F1 or F2.
+    if (C1->isNullValue() && C2->isNullValue() &&
+        isEquivalentType(C1->getType(), C2->getType()))
+      return true;
+    // Try bitcasting C2 to C1's type. If the bitcast is legal and returns C1
+    // then they must have equal bit patterns. Aggregate types cannot be
+    // bitcast.
+    if (C1->getType()->isAggregateType() || C2->getType()->isAggregateType())
+      return false;
+    return C1->getType()->canLosslesslyBitCastTo(C2->getType()) &&
+      C1 == ConstantExpr::getBitCast(const_cast<Constant*>(C2), C1->getType());
+  }
+
+  if (isa<InlineAsm>(V1) || isa<InlineAsm>(V2))
+    return V1 == V2;
+
+  // Check that V1 maps to V2. If we find a value that V1 maps to then we simply
+  // check whether it's equal to V2. When there is no mapping then we need to
+  // ensure that V2 isn't already equivalent to something else. For this
+  // purpose, we track the V2 values in a set.
+
+  ValueToValueMapTy::iterator I = id_map.find(V1);
+  if (I != id_map.end())
+    return V2 == I->second;
+  // FIXME: Const casts!!!
+  if (!seen_values.insert(std::make_pair(V2, const_cast<Value *>(V1))).second)
+    return false;
+  id_map[V1] = const_cast<Value *>(V2);
+  return true;
+}
+
+/// Test whether two basic blocks have equivalent behaviour. Returns true if the
+/// blocks can be merged, false if they cannot. Differing instructions are
+/// recorded in DifferingInstructions.
+bool FunctionComparator::compare(const BasicBlock *BB1, const BasicBlock *BB2,
+    std::list<std::pair<const PHINode*,const PHINode*> > *PHIsFound) {
+  BasicBlock::const_iterator F1I, F1E, F2I, F2E;
+
+  for (F1I = BB1->begin(), F1E = BB1->end(),
+       F2I = BB2->begin(), F2E = BB2->end();
+       F1I != F1E && F2I != F2E; ++F1I, ++F2I) {
+    // Skip debug information
+    const CallInst *DbgCall;
+    while (F1I != F1E && (DbgCall = dyn_cast<CallInst>(F1I)) &&
+           DbgCall->getCalledFunction() &&
+           DbgCall->getCalledFunction()->hasName() &&
+           DbgCall->getCalledFunction()->getName().startswith("llvm.dbg."))
+      ++F1I;
+
+    while (F2I != F2E && (DbgCall = dyn_cast<CallInst>(F2I)) &&
+           DbgCall->getCalledFunction() &&
+           DbgCall->getCalledFunction()->hasName() &&
+           DbgCall->getCalledFunction()->getName().startswith("llvm.dbg."))
+      ++F2I;
+
+    if (F1I == F1E || F2I == F2E)
+      break;
+
+    // Ok, we're dealing with real instructions. Check a few cases that will
+    // prevent merging first.
+    const Instruction *F1In = &*F1I;
+    const Instruction *F2In = &*F2I;
+
+    // Cannot merge insts that differ in whether they have uses
+    if (F1In->use_empty() != F2In->use_empty()) {
+      // TODO: Could implement merging for this case (would need to introduce a
+      // dummy value in the PHI node etc.)
+      return false;
+    }
+
+    // Cannot merge insts whose types are non-equivalent
+    if (!isEquivalentType(F1In->getType(), F2In->getType())) {
+      return false;
+    }
+
+    // TODO:  Currently cannot merge InvokeInsts with differing result types
+    //        that have uses. We cannot push up a bitcast into their block after
+    //        them because they are terminators. Would need to insert an
+    //        additional BB.
+    if (isa<InvokeInst>(F1In) && !F1In->use_empty() &&
+        F1In->getType() != F2In->getType())
+      return false;
+
+    if (!enumerate(F1In, F2In))
+      goto differing_instructions;
+
+    if (const GetElementPtrInst *GEP1 = dyn_cast<GetElementPtrInst>(F1In)) {
+      const GetElementPtrInst *GEP2 = dyn_cast<GetElementPtrInst>(F2In);
+      if (!GEP2)
+        goto differing_instructions;
+
+      if (!enumerate(GEP1->getPointerOperand(), GEP2->getPointerOperand()))
+        goto differing_instructions;
+
+      if (!isEquivalentGEP(GEP1, GEP2))
+        goto differing_instructions;
+    } else if (const PHINode *Phi1 = dyn_cast<PHINode>(F1In)) {
+      const PHINode *Phi2 = dyn_cast<PHINode>(F2In);
+      // We can't currently merge a PHI and non-PHI instruction
+      if (!Phi2)
+        return false;
+
+      // We can't currently merge PHI nodes with different numbers of incoming
+      // values
+      if (F1In->getNumOperands() != F2In->getNumOperands())
+        return false;
+
+      // We need to treat PHI nodes specially. Their incoming values may be in a
+      // different order even if they are equivalent. We can't compare them
+      // until we've seen the incoming blocks and know which values are
+      // equivalent. Therefore postpone PHINode comparison until the end.
+      PHIsFound->push_back(std::make_pair(Phi1, Phi2));
+    } else {
+      if (!isEquivalentOperation(F1In, F2In))
+        goto differing_instructions;
+
+      bool IsCall = isa<CallInst>(F1In);
+      assert(F1In->getNumOperands() == F2In->getNumOperands());
+      for (unsigned i = 0, e = F1In->getNumOperands(); i != e; ++i) {
+        Value *OpF1 = F1In->getOperand(i);
+        Value *OpF2 = F2In->getOperand(i);
+
+        // Allow self-reference if this is a call instruction and the last
+        // operand which is the called function
+        bool AllowSelfRef = IsCall && (i + 1) == e;
+
+        if (!enumerate(OpF1, OpF2, !AllowSelfRef))
+          goto differing_instructions;
+
+        if (!isEquivalentType(OpF1->getType(), OpF2->getType()))
+          goto differing_instructions;
+
+        if ((OpF1 == F1 && OpF2 == F2) || (OpF1 == F2 && OpF2 == F1))
+          SelfRefInstructions[F1In] = F2In;
+      }
+    }
+
+    continue;
+
+differing_instructions:
+    // Cannot merge functions with differing landing pad instructions yet. They
+    // would need special treatment which involves updating the corresponding
+    // invoke instructions.
+    if (isa<LandingPadInst>(F1In))
+      return false;
+    if (isa<InvokeInst>(F1In))
+      return false;
+
+    DifferingInstructions[F1In] = F2In;
+  }
+
+  // We cannot currently merge basic blocks with different instruction counts
+  return F1I == F1E && F2I == F2E;
+}
+
+bool FunctionComparator::comparePHIs(
+  const std::list<std::pair<const PHINode*,const PHINode*> > &PHIs) {
+  if (PHIs.empty())
+    return true;
+
+  for (std::list<std::pair<const PHINode*,const PHINode*> >::const_iterator
+         I = PHIs.begin(), E = PHIs.end(); I != E; ++I) {
+    const PHINode *Phi1 = I->first, *Phi2 = I->second;
+
+    for (unsigned ValId = 0, ValNum = Phi1->getNumIncomingValues();
+         ValId < ValNum; ++ValId) {
+      Value *Phi1Val = Phi1->getIncomingValue(ValId);
+
+      // Get corresponding Phi2Val
+      Value *BBinPhi2Val = getF1toF2Map()[Phi1->getIncomingBlock(ValId)];
+
+      if (!BBinPhi2Val)
+        return false; // Currently can't handle differing predecessor blocks
+
+      BasicBlock *BBinPhi2 = cast<BasicBlock>(BBinPhi2Val);
+      Value *Phi2Val = Phi2->getIncomingValueForBlock(BBinPhi2);
+
+      // Enumerate the values. If the PHI node references the function itself (a
+      // very rare case), we mark it as different (NoSelfRef). This is only
+      // necessary for outline merging, not equiv merging. TODO: Make equal
+      // merging possible with such PHI nodes.
+      if (!enumerate(Phi1Val, Phi2Val,/*NoSelfRef=*/true)) {
+        DifferingInstructions[Phi1] = Phi2;
+        break;
+      }
+    }
+  }
+
+  return true;
+}
+
+// Test whether the two functions have equivalent behaviour.
+bool FunctionComparator::compare() {
+  // We need to recheck everything, but check the things that weren't included
+  // in the hash first.
+  if (F1->getAttributes() != F2->getAttributes())
+    goto not_mergeable;
+
+  if (F1->hasGC() != F2->hasGC())
+    goto not_mergeable;
+
+  if (F1->hasGC() && F1->getGC() != F2->getGC())
+    goto not_mergeable;
+
+  if (!F1->getSection().equals(F2->getSection()))
+    goto not_mergeable;
+
+  if (F1->isVarArg() != F2->isVarArg())
+    goto not_mergeable;
+
+  if (F1->isInterposable() != F2->isInterposable())
+    goto not_mergeable;
+
+  if (F1->size() != F2->size())
+    goto not_mergeable;
+
+  // TODO: if it's internal and only used in direct calls, we could handle
+  // this case too.
+  if (F1->getCallingConv() != F2->getCallingConv())
+    goto not_mergeable;
+
+  if (!isEquivalentType(F1->getFunctionType(), F2->getFunctionType()))
+    goto not_mergeable;
+
+  assert(F1->arg_size() == F2->arg_size() &&
+         "Identically typed functions have different numbers of args!");
+
+  // Visit the arguments so that they get enumerated in the order they're
+  // passed in.
+  for (Function::const_arg_iterator f1i = F1->arg_begin(),
+         f2i = F2->arg_begin(), f1e = F1->arg_end(); f1i != f1e; ++f1i, ++f2i) {
+    if (!enumerate(&*f1i, &*f2i))
+      llvm_unreachable("Arguments repeat!");
+  }
+
+  // We do a CFG-ordered walk since the actual ordering of the blocks in the
+  // linked list is immaterial. Our walk starts at the entry block for both
+  // functions, then takes each block from each terminator in order. As an
+  // artifact, this also means that unreachable blocks are ignored.
+  {
+    SmallVector<const BasicBlock *, 8> F1BBs, F2BBs;
+    SmallSet<const BasicBlock *, 32> VisitedBBs; // in terms of F1.
+    std::list<std::pair<const PHINode*,const PHINode*> > PHIsFound;
+
+    F1BBs.push_back(&F1->getEntryBlock());
+    F2BBs.push_back(&F2->getEntryBlock());
+
+    VisitedBBs.insert(F1BBs[0]);
+    while (!F1BBs.empty()) {
+      const BasicBlock *F1BB = F1BBs.pop_back_val();
+      const BasicBlock *F2BB = F2BBs.pop_back_val();
+
+      // Check for control flow divergence
+      if (!enumerate(F1BB, F2BB))
+        goto not_mergeable;
+
+      const Instruction *F1TI = F1BB->getTerminator();
+      const Instruction *F2TI = F2BB->getTerminator();
+
+      // TODO: Implement merging of blocks with different numbers of
+      // instructions.
+      if (F1TI->getNumSuccessors() != F2TI->getNumSuccessors() ||
+          F1BB->size() != F2BB->size())
+        goto not_mergeable;
+
+      // The actual instruction-by-instruction comparison
+      if (!compare(F1BB, F2BB, &PHIsFound))
+        goto not_mergeable;
+
+      // FIXME: Count this in compare(F1BB,F2BB) so it doesn't include debug
+      // instructions.
+      InstructionCount += std::max(F1BB->size(), F2BB->size());
+
+      assert(F1TI->getNumSuccessors() == F2TI->getNumSuccessors());
+      for (unsigned i = 0, e = F1TI->getNumSuccessors(); i != e; ++i) {
+        if (!VisitedBBs.insert(F1TI->getSuccessor(i)).second)
+          continue;
+
+        F1BBs.push_back(F1TI->getSuccessor(i));
+        F2BBs.push_back(F2TI->getSuccessor(i));
+      }
+    }
+
+    BasicBlockCount = VisitedBBs.size();
+
+    // After we've seen all values and BBs, compare the PHI nodes
+    if (!comparePHIs(PHIsFound))
+      goto not_mergeable;
+  }
+
+  if (DifferingInstructions.size()) {
+    // Currently we can't merge vararg functions with differing instructions.
+    // TODO: Explore whether this is feasible; the difficult bit is the
+    // additional argument we need to add.
+    if (F1->isVarArg())
+      goto not_mergeable;
+
+    isDifferent = true;
+    DifferingInstructionsCount += DifferingInstructions.size();
+
+    LLVM_DEBUG(float Metric = ((float)InstructionCount - DifferingInstructionsCount)
+                         / InstructionCount*100;
+          dbgs() << "Similar fns: " << F1->getName() << " and " << F2->getName()
+                 << " bbs=" << BasicBlockCount << " insts=" << InstructionCount
+                 << " failed=" << DifferingInstructionsCount << " metric="
+                 << format("%0.2f", Metric)
+                 << '\n');
+  }
+
+  return true;
+
+not_mergeable:
+  // Fail: cannot merge the two functions
+  isNotMergeable = true;
+  return false;
+}
+
+bool FunctionComparator::isExactMatch() {
+  return (!isNotMergeable && !isDifferent);
+}
+
+bool FunctionComparator::isMergeCandidate() {
+  if (isNotMergeable)
+    return false;
+
+  if (!isDifferent)
+    return true;
+
+  // Heuristic when to attempt merging
+  if (InstructionCount > MergeDifferingMinInsts &&
+      DifferingInstructionsCount <= MergeMaxDiffering &&
+      getSimilarityMetric() > MergeMinSimilarity)
+    return true;
+
+  // Tolerate higher difference with higher similarity.
+  if (InstructionCount > 100 &&
+      DifferingInstructionsCount <= 60 &&
+      getSimilarityMetric() > 90 )
+    return true;
+
+  return false;
+}
+
+namespace {
+
+struct FunctionComparatorOrdering {
+  bool operator () (FunctionComparator *LHS, FunctionComparator *RHS) const {
+    unsigned MetricLHS = LHS->getSimilarityMetric(),
+             MetricRHS = RHS->getSimilarityMetric();
+
+    // If the metric is the same, then default to the unique ID. We need
+    // to use a unique value instead of the object address to ensure
+    // deterministic ordering.
+    if (MetricLHS == MetricRHS)
+      return LHS->getID() > RHS->getID();
+    return MetricLHS > MetricRHS;
+  }
+};
+
+class MergeRegistry {
+public:
+  typedef MapVector<unsigned, std::list<ComparableFunction> > FnCompareMap;
+  typedef std::set<FunctionComparator *, FunctionComparatorOrdering>
+    FnComparatorSet;
+  typedef std::map<Function *, FnComparatorSet> SimilarFnMap;
+
+  ~MergeRegistry();
+
+  void clear();
+
+  /// Defer a function for consideration in the next round.
+  void defer(Function *F);
+
+  /// Return true if we have deferred functions that can be enqueued.
+  bool haveDeferred() { return !Deferred.empty(); }
+
+  /// Move all the deferred functions into buckets to consider them for merging.
+  /// Returns number of functions that have been added.
+  unsigned enqueue();
+
+  /// Add a candidate for merging
+  void insertCandidate(FunctionComparator *Comp);
+
+  /// Remove a Function from the FnSet and queue it up for a second sweep of
+  /// analysis if Reanalyze is set. If it is a candidate for merging, remove it
+  /// from consideration.
+  void remove(Function *F, bool Reanalyze=true);
+
+  /// Return the similarity metric of the most similar function to F that is
+  /// not listed in the Ignore set.
+  unsigned getMaxSimilarity(Function *F, const DenseSet<Function *> &Ignore);
+
+  /// The collection of buckets that contain functions that may be similar to
+  /// each other (same hash value).
+  FnCompareMap FunctionsToCompare;
+
+  std::list<FunctionComparator *> FunctionsToMerge;
+  SimilarFnMap SimilarFunctions;
+
+private:
+  typedef std::vector<WeakVH> FnDeferredQueue;
+
+  /// A work queue of functions that may have been modified and should be
+  /// analyzed again.
+  FnDeferredQueue Deferred;
+};
+
+}  // end anonymous namespace
+
+MergeRegistry::~MergeRegistry() {
+  this->clear();
+}
+
+void MergeRegistry::clear() {
+  Deferred.clear();
+  SimilarFunctions.clear();
+  for (std::list<FunctionComparator *>::iterator
+        I = FunctionsToMerge.begin(), E = FunctionsToMerge.end();
+       I != E; ++I) {
+    FunctionComparator *FC = *I;
+    delete FC;
+  }
+  FunctionsToMerge.clear();
+  FunctionsToCompare.clear();
+}
+
+static bool isAliasCapable(Function* G) {
+  return
+    UseGlobalAliases && G->hasGlobalUnnamedAddr()
+    && (G->hasExternalLinkage() || G->hasLocalLinkage() || G->hasWeakLinkage());
+}
+
+static bool isComparisonCandidate(Function *F) {
+  if (Opt::MergeLevel == Opt::size) {
+    // Only consider functions that are to be optimized for size.
+    // By default, that is all functions at -Os/-Oz and nothing at -O2.
+    bool Os = F->getAttributes().
+      hasAttribute(AttributeList::FunctionIndex, Attribute::OptimizeForSize);
+    bool Oz = F->getAttributes().
+      hasAttribute(AttributeList::FunctionIndex, Attribute::MinSize);
+    if (!Os && !Oz)
+      return false;
+  }
+
+  // Ignore declarations and tiny functions - no point in merging those
+  if (F->isDeclaration()) return false;
+  if (F->getName().endswith(MERGED_SUFFIX)) return false;
+  if (F->hasAvailableExternallyLinkage()) return false;
+  if (F->hasFnAttribute(Attribute::AlwaysInline)) return false;
+  if (F->size() == 1 && F->begin()->size() < MergeMinInsts)
+    return isAliasCapable(F);
+
+  return true;
+}
+
+void MergeRegistry::defer(Function *F) {
+  if (isComparisonCandidate(F))
+    Deferred.push_back(F);
+}
+
+// Move functions from Deferred into buckets. remove() may have been called
+// multiple times for the same function, so eliminate duplicates using the
+// set. We reverse them because MergeSimilarFunctions::insert inserts at the
+// front of each bucket.
+unsigned MergeRegistry::enqueue() {
+  DenseSet<Function *> InsertedFuncs;
+
+  for (std::vector<WeakVH>::reverse_iterator
+      DefI = Deferred.rbegin(), DefE = Deferred.rend();
+      DefI != DefE; ++DefI) {
+    Value *V = *DefI;
+    Function *F = dyn_cast_or_null<Function>(V);
+    if (!F) continue;
+    if (InsertedFuncs.find(F) != InsertedFuncs.end()) continue;
+    if (!isComparisonCandidate(F)) continue;
+
+    unsigned Hash = profileFunction(F);
+    FunctionsToCompare[Hash].push_front(F);
+
+    InsertedFuncs.insert(F);
+  }
+
+  Deferred.clear();
+
+  return InsertedFuncs.size();
+}
+
+void MergeRegistry::insertCandidate(FunctionComparator *Comp) {
+  FunctionsToMerge.push_back(Comp);
+  SimilarFunctions[Comp->getF1()].insert(Comp);
+}
+
+static void removeFromBucket(Function *F,
+    std::list<ComparableFunction> &Bucket) {
+  for (std::list<ComparableFunction>::iterator
+      I = Bucket.begin(), E = Bucket.end(); I != E; ++I) {
+    if (I->getFunc() == F) {
+      Bucket.erase(I);
+      return;
+    }
+  }
+}
+
+void MergeRegistry::remove(Function *F, bool Reanalyze/*=true*/) {
+  // There is no need to remove a function that is not already
+  // in a bucket.
+  if (!isComparisonCandidate(F))
+    return;
+
+  unsigned Hash = profileFunction(F);
+  std::list<ComparableFunction> &Bucket = FunctionsToCompare[Hash];
+
+  removeFromBucket(F, Bucket);
+
+  if (Reanalyze)
+    Deferred.push_back(F);
+
+  // Check whether we have any existing FunctionComparator objects for this fn.
+  // If yes, discard them because F has changed. Retry merging for those
+  // functions by adding them to Deferred.
+  std::list<FunctionComparator *>::iterator I = FunctionsToMerge.begin();
+  while (I != FunctionsToMerge.end()) {
+    FunctionComparator *Comp = *I;
+    if (Comp->getF1() == F) {
+      Function *OtherF = Comp->getF2();
+      Deferred.push_back(OtherF);
+      removeFromBucket(OtherF, Bucket);
+      if (!SimilarFunctions[F].erase(Comp))
+        llvm_unreachable("Inconsistent SimilarFunctions set");
+      I = FunctionsToMerge.erase(I);
+      delete Comp;
+    } else if (Comp->getF2() == F) {
+      Function *OtherF = Comp->getF1();
+      Deferred.push_back(OtherF);
+      removeFromBucket(OtherF, Bucket);
+      if (!SimilarFunctions[OtherF].erase(Comp))
+        llvm_unreachable("Inconsistent SimilarFunctions set");
+      I = FunctionsToMerge.erase(I);
+      delete Comp;
+    } else {
+      ++I;
+    }
+  }
+}
+
+unsigned MergeRegistry::getMaxSimilarity(Function *F,
+    const DenseSet<Function *> &Ignore) {
+  FnComparatorSet &Similar = SimilarFunctions[F];
+
+  for (FnComparatorSet::iterator I = Similar.begin(), E = Similar.end();
+       I != E; ++I) {
+    FunctionComparator *Comp = *I;
+    if (Ignore.count(Comp->getF2()))
+      continue;
+
+    return Comp->getSimilarityMetric();
+  }
+
+  return 0;
+}
+
+namespace {
+
+class MergeSimilarFunctions : public ModulePass {
+public:
+  static char ID;
+  MergeSimilarFunctions(const ModuleSummaryIndex *Summary = nullptr)
+    : ModulePass(ID) {
+    initializeMergeSimilarFunctionsPass(*PassRegistry::getPassRegistry());
+  }
+
+  bool runOnModule(Module &M);
+
+private:
+  /// Find the functions that use this Value and remove them from FnSet and
+  /// queue the functions.
+  void removeUsers(Value *V);
+
+  /// Replace all direct calls of Old with calls of New. Will bitcast New if
+  /// necessary to make types match.
+  void replaceDirectCallers(Function *Old, Function *New);
+
+  /// Process functions in the specified bucket, by either doing equiv merging
+  /// marking them for diff merging. Returns false if the bucket needs to be
+  /// re-scanned after an equiv merge. Sets Changed if the module was changed by
+  /// equiv merge.
+  bool mergeBucket(std::list<ComparableFunction> &Fns, bool &Changed);
+
+  /// Exhaustively compare all functions in each bucket and do equiv merging
+  /// where possible. Functions that have already been compared will not be
+  /// compared again. Returns true if the module was modified.
+  bool doExhaustiveCompareMerge();
+
+  /// Merge all the functions marked for diff merging. Returns true if the
+  /// module was modified.
+  bool doDiffMerge();
+
+  /// Merge two equivalent functions. Upon completion, G may be deleted, or may
+  /// be converted into a thunk. In either case, it should never be visited
+  /// again.
+  void mergeTwoFunctions(Function *F, Function *G);
+
+  /// Merge a set of functions with differences.
+  void outlineAndMergeFunctions(SmallVectorImpl<FunctionComparator *> &Fns);
+
+  /// Replace G with a thunk or an alias to F. Deletes G.
+  void writeThunkOrAlias(Function *F, Function *G);
+
+  /// Replace G with a simple tail call to bitcast(F). Also replace direct uses
+  /// of G with bitcast(F). Deletes G.
+  void writeThunk(Function *F, Function *G);
+
+  /// Replace G with a tail call to F with an additional argument.
+  ///
+  void writeThunkWithChoice(Function *NewF, Function *OldF, int Choice);
+
+  /// Replace G with an alias to F. Deletes G.
+  void writeAlias(Function *F, Function *G);
+
+  /// DataLayout for more accurate GEP comparisons. May be NULL.
+  const DataLayout *DL;
+
+  /// Merge registry. Stores all the information about functions being
+  /// considered for merging as well as current candidates for merging.
+  MergeRegistry Registry;
+
+};
+
+}  // end anonymous namespace
+
+char MergeSimilarFunctions::ID = 0;
+INITIALIZE_PASS(MergeSimilarFunctions, "mergesimilarfunc",
+                "Merge Similar Functions", false, false)
+
+ModulePass *
+llvm::createMergeSimilarFunctionsPass(const ModuleSummaryIndex *S) {
+  return new MergeSimilarFunctions(S);
+}
+
+bool MergeSimilarFunctions::runOnModule(Module &M) {
+  if (Opt::MergeLevel == Opt::none)
+    return false;
+
+  bool Changed = false;
+
+  DL = &M.getDataLayout();
+
+  for (auto &I : M)
+    Registry.defer(&I);
+
+  do {
+    unsigned InsertCount = Registry.enqueue();
+
+    LLVM_DEBUG(dbgs() << "size of module: " << M.size() << '\n');
+    LLVM_DEBUG(dbgs() << "size of worklist: " << InsertCount << '\n');
+    (void)InsertCount;
+
+    Changed |= doExhaustiveCompareMerge();
+  } while (Registry.haveDeferred());
+
+  Changed |= doDiffMerge();
+
+  Registry.clear();
+  return Changed;
+}
+
+// Replace direct callers of Old with New.
+void MergeSimilarFunctions::replaceDirectCallers(Function *Old, Function *New) {
+  Constant *BitcastNew = ConstantExpr::getBitCast(New, Old->getType());
+  for (Value::use_iterator UI = Old->use_begin(), UE = Old->use_end();
+       UI != UE;) {
+    Use *U = &*UI;
+    ++UI;
+    CallSite CS(U->getUser());
+    if (CS && CS.isCallee(U)) {
+      Registry.remove(CS.getInstruction()->getParent()->getParent());
+      U->set(BitcastNew);
+    }
+  }
+}
+
+// Replace G with an alias to F if possible, or else a thunk to F. Deletes G.
+void MergeSimilarFunctions::writeThunkOrAlias(Function *F, Function *G) {
+  if (isAliasCapable(G)) {
+    writeAlias(F, G);
+    return;
+  }
+
+  writeThunk(F, G);
+}
+
+static void writeThunkBody(Function *Thunk, Function *F,
+                           ConstantInt *Choice, const DataLayout *DL) {
+  BasicBlock *BB = &Thunk->getEntryBlock();
+  IRBuilder<> Builder(BB);
+
+  SmallVector<Value *, 16> Args;
+  unsigned i = 0;
+  FunctionType *FFTy = F->getFunctionType();
+  Type *IntPtrTy = DL->getIntPtrType(FFTy->getContext());
+  for (auto &AI : Thunk->args()) {
+    Value *Cast = createCastIfNeeded(&AI, FFTy->getParamType(i), BB, IntPtrTy);
+    Args.push_back(Cast);
+    ++i;
+  }
+  if (Choice)
+    Args.push_back(Choice);
+
+  CallInst *CI = Builder.CreateCall(F, Args);
+  CI->setTailCall();
+  CI->setCallingConv(F->getCallingConv());
+  CI->setAttributes(F->getAttributes());
+  CI->setIsNoInline();
+  if (Thunk->getReturnType()->isVoidTy()) {
+    Builder.CreateRetVoid();
+  } else {
+    Type *RetTy = Thunk->getReturnType();
+    if (CI->getType()->isIntegerTy() && RetTy->isPointerTy())
+      Builder.CreateRet(Builder.CreateIntToPtr(CI, RetTy));
+    else if (CI->getType()->isPointerTy() && RetTy->isIntegerTy())
+      Builder.CreateRet(Builder.CreatePtrToInt(CI, RetTy));
+    else {
+      Value *Cast = createCastIfNeeded(CI, RetTy, BB, IntPtrTy);
+      Builder.CreateRet(Cast);
+    }
+  }
+}
+
+// Replace G with a simple tail call to bitcast(F). Also replace direct uses
+// of G with bitcast(F). Deletes G.
+void MergeSimilarFunctions::writeThunk(Function *F, Function *G) {
+  if (!G->isInterposable()) {
+    // Redirect direct callers of G to F.
+    replaceDirectCallers(G, F);
+  }
+
+  // If G was internal then we may have replaced all uses of G with F. If so,
+  // stop here and delete G. There's no need for a thunk.
+  if (G->hasLocalLinkage() && G->use_empty()) {
+    LLVM_DEBUG(dbgs() << "All uses of " << G->getName() << " replaced by "
+                 << F->getName() << ". Removing it.\n");
+    G->eraseFromParent();
+    return;
+  }
+
+  Function *NewG = Function::Create(G->getFunctionType(), G->getLinkage(), "",
+                                    G->getParent());
+  BasicBlock::Create(F->getContext(), "", NewG);
+
+  writeThunkBody(NewG, F, nullptr, DL);
+
+  NewG->copyAttributesFrom(G);
+  NewG->takeName(G);
+  removeUsers(G);
+  G->replaceAllUsesWith(NewG);
+  G->eraseFromParent();
+
+  LLVM_DEBUG(dbgs() << "writeThunk: " << NewG->getName() << " calling "
+               << F->getName() << '\n');
+  ++NumThunksWritten;
+}
+
+void MergeSimilarFunctions::writeThunkWithChoice(Function *NewF, Function *OldF,
+                                          int Choice) {
+  // Deleting the body of a function sets its linkage to External. Save the old
+  // one here and restore it at the end.
+  GlobalValue::LinkageTypes OldFLinkage = OldF->getLinkage();
+
+  // Delete OldF's body
+  OldF->deleteBody();
+  BasicBlock::Create(OldF->getContext(), "", OldF);
+
+  // Insert single BB with tail call
+  IntegerType *Int32Ty = Type::getInt32Ty(OldF->getContext());
+  ConstantInt *ChoiceConst = ConstantInt::get(Int32Ty, Choice);
+  writeThunkBody(OldF, NewF, ChoiceConst, DL);
+  OldF->setLinkage(OldFLinkage);
+}
+
+// Replace G with an alias to F and delete G.
+void MergeSimilarFunctions::writeAlias(Function *F, Function *G) {
+
+  // Replace all current uses of G in constants with F. This handles virtual
+  // table and other references. Do this first so that we don't modify thge
+  // global alias we're about to create.
+  SmallVector<Use *, 8> Uses;
+  for (auto I = G->use_begin(), E = G->use_end(); I != E; ++I) {
+    Use *U = I.operator->();
+    Constant *CV = dyn_cast<Constant>(U->getUser());
+    if (!CV) continue;
+    Uses.push_back(U);
+  }
+  for (auto I = Uses.begin(), E= Uses.end(); I != E; ++I) {
+    Use *U = *I;
+    U->set(F);
+  }
+
+  PointerType *PTy = G->getType();
+  auto *GA = GlobalAlias::create(PTy->getElementType(), PTy->getAddressSpace(),
+                                 G->getLinkage(), "", F);
+  F->setAlignment(std::max(F->getAlignment(), G->getAlignment()));
+  GA->takeName(G);
+  GA->setVisibility(G->getVisibility());
+  removeUsers(G);
+  G->replaceAllUsesWith(GA);
+  G->eraseFromParent();
+
+  LLVM_DEBUG(dbgs() << "writeAlias: " << GA->getName() << '\n');
+  ++NumAliasesWritten;
+}
+
+// Merge two equivalent functions. Upon completion, Function G is deleted.
+void MergeSimilarFunctions::mergeTwoFunctions(Function *F, Function *G) {
+  if (F->isInterposable()) {
+    assert(G->isInterposable());
+
+    if (UseGlobalAliases) {
+      // Make them both thunks to the same internal function.
+      Function *H = Function::Create(F->getFunctionType(), F->getLinkage(), "",
+                                     F->getParent());
+      H->copyAttributesFrom(F);
+      H->takeName(F);
+      removeUsers(F);
+      F->replaceAllUsesWith(H);
+
+      unsigned MaxAlignment = std::max(G->getAlignment(), H->getAlignment());
+
+      writeAlias(F, G);
+      writeAlias(F, H);
+
+      F->setAlignment(MaxAlignment);
+      F->setLinkage(GlobalValue::PrivateLinkage);
+    } else {
+      // We can't merge them. Instead, pick one and update all direct callers
+      // to call it and hope that we improve the instruction cache hit rate.
+      replaceDirectCallers(G, F);
+    }
+
+    ++NumDoubleWeak;
+  } else {
+    writeThunkOrAlias(F, G);
+  }
+
+  ++NumFunctionsMerged;
+}
+
+static Value *getLastArg(Function *F) {
+  auto it = F->arg_begin();
+  std::advance(it, F->arg_size()-1);
+  return it;
+}
+
+static void insertCondAndRemapInstructions(
+    Instruction *F1InstInNewF, const std::vector<const Instruction *> &F2Insts,
+    Function *NewF, ValueToValueMapTy &F1toNewF,
+    const SmallVectorImpl<FunctionComparator *> &Comps,
+    Type *IntPtrTy) {
+  assert(F2Insts.size() == Comps.size() &&
+      "Mis-match between F2Insts & Comps!");
+
+  SmallVector<Instruction *, 8> F2InstsInNewF;
+  for (unsigned FnI = 0, FnE = F2Insts.size(); FnI != FnE; ++FnI) {
+    const Instruction *F2Inst = F2Insts[FnI];
+    if (!F2Inst) {
+      F2InstsInNewF.push_back(NULL);
+      continue;
+    }
+
+    Instruction *F2InstInNewF = F2Inst->clone();
+
+    // Remap F2Inst: F2 values -> F1 values
+    RemapInstruction(F2InstInNewF, Comps[FnI]->getF2toF1Map(),
+                     RF_NoModuleLevelChanges);
+    // Remap F2Inst: F1 values -> NewF values
+    RemapInstruction(F2InstInNewF, F1toNewF, RF_NoModuleLevelChanges);
+
+    F2InstsInNewF.push_back(F2InstInNewF);
+  }
+
+  SmallVector<Instruction *, 8> Terminators;
+  SplitBlockAndInsertSwitch(getLastArg(NewF), F1InstInNewF,
+                            F2InstsInNewF, Terminators);
+
+  assert(Terminators.size() == F2InstsInNewF.size() + 1 &&
+      "Not enough terminators returned");
+
+  // F2InstsInNewF are now hooked up to the correct values in NewF. However,
+  // some of their operands may be pointers/integers so they could potentially
+  // have the wrong type in NewF (since we treat all pointers and integers of
+  // same size as equal). Insert casts if needed.
+  for (unsigned FnI = 0, FnE = F2InstsInNewF.size(); FnI != FnE; ++FnI) {
+    Instruction *F2InstInNewF = F2InstsInNewF[FnI];
+    if (!F2InstInNewF)
+      continue;
+    const Instruction *F2Inst = F2Insts[FnI];
+
+    for (unsigned OpId=0; OpId < F2InstInNewF->getNumOperands(); ++OpId) {
+      Value *F2NewFOperand = F2InstInNewF->getOperand(OpId);
+      Value *F2OrigOperand = F2Inst->getOperand(OpId);
+
+      if (F2NewFOperand->getType() != F2OrigOperand->getType()) {
+        Value *Cast = createCastIfNeeded(F2NewFOperand,
+            F2OrigOperand->getType(),
+            F2InstInNewF,
+            IntPtrTy);
+        F2InstInNewF->setOperand(OpId, Cast);
+      }
+    }
+  }
+
+  if (ReturnInst *F1Ret = dyn_cast<ReturnInst>(F1InstInNewF)) {
+    // If we're handling differing return instructions, we need to ensure that
+    // they all return the same type. Since we treat pointer types as equal, we
+    // may need to insert a bitcast.
+    for (Instruction *F2Inst : F2InstsInNewF) {
+      if (!F2Inst)
+        continue;
+
+      // F2Inst must also be a return instruction due to control flow
+      // isomorphism.
+      ReturnInst *F2Ret = cast<ReturnInst>(F2Inst);
+
+      if (F2Ret->getReturnValue()->getType() !=
+          F1Ret->getReturnValue()->getType())
+        F2Ret->setOperand(0,
+                          createCastIfNeeded(F2Ret->getReturnValue(),
+                                             F1Ret->getReturnValue()->getType(),
+                                             F2Ret, IntPtrTy));
+    }
+  } else if (!F1InstInNewF->use_empty()) {
+    // If the instructions have uses, we need to insert a PHI node.
+    //
+    // We treat all pointer types as equal, so we may need to insert
+    // a bitcast to ensure that all incoming values of the PHI node have the
+    // same type
+    Type *F1IType = F1InstInNewF->getType();
+    BasicBlock *TailBB = Terminators[0]->getSuccessor(0);
+    PHINode *Phi =
+        PHINode::Create(F1IType, F2InstsInNewF.size(), "", &TailBB->front());
+    F1InstInNewF->replaceAllUsesWith(Phi);
+
+    Phi->addIncoming(F1InstInNewF, F1InstInNewF->getParent());
+    for (unsigned FnI = 0, FnE = F2InstsInNewF.size(); FnI != FnE; ++FnI) {
+      Instruction *F2InstInNewF = F2InstsInNewF[FnI];
+      if (!F2InstInNewF)
+        continue;
+
+      if (F2InstInNewF->getType() != F1IType) {
+        assert(!F2InstInNewF->isTerminator() &&
+            "Cannot cast result of terminator instruction");
+
+        F2InstInNewF = cast<Instruction>(
+            createCastIfNeeded(F2InstInNewF,
+              F1IType,
+              Terminators[FnI+1],
+              IntPtrTy));
+      }
+
+      Phi->addIncoming(F2InstInNewF, F2InstInNewF->getParent());
+    }
+  }
+}
+
+static void mergePHINode(const SmallVectorImpl<FunctionComparator *> &Fns,
+                         Function *NewF,
+                         ValueToValueMapTy &VMap, /* F1->FNew */
+                         const PHINode *F1PhiInst,
+                         std::vector<const Instruction *> F2PhiInsts) {
+  PHINode *F1PhiInNewF = dyn_cast<PHINode>(VMap[F1PhiInst]);
+  assert(F1PhiInNewF && "Cannot find F1Inst in NewF!");
+
+  // The incoming blocks in any of the F2PhiInsts may be in a different order.
+  // If this is the case, we have to reorder them. F2PhiInsts is intentionally a
+  // copy, so we can modify it
+  SmallVector<PHINode *, 4> GCInsts; // so we can delete them later.
+  for (unsigned FnI = 0, FnE = F2PhiInsts.size(); FnI != FnE; ++FnI) {
+    const PHINode *F2PhiInst = dyn_cast_or_null<const PHINode>(F2PhiInsts[FnI]);
+    if (!F2PhiInst)
+      continue;
+
+    for (unsigned I = 0, E = F1PhiInNewF->getNumIncomingValues(); I < E; ++I) {
+      if (!Fns[FnI]->enumerate(F1PhiInst->getIncomingBlock(I),
+                               F2PhiInst->getIncomingBlock(I))) {
+        // Non-equivalent blocks in the same position - need to reorder PhiInst
+        PHINode *ReorderedF2PhiInst = PHINode::Create(F2PhiInst->getType(), E);
+
+        for (unsigned II = 0; II < E; ++II) {
+          Value *BBVal =
+            Fns[FnI]->getF1toF2Map()[F1PhiInst->getIncomingBlock(II)];
+          BasicBlock *BB = cast<BasicBlock>(BBVal);
+          Value *Val = F2PhiInst->getIncomingValueForBlock(BB);
+          ReorderedF2PhiInst->addIncoming(Val, BB);
+        }
+
+        F2PhiInsts[FnI] = ReorderedF2PhiInst;
+        GCInsts.push_back(ReorderedF2PhiInst);
+        break;
+      }
+    }
+  }
+
+  // Now merge the PHI nodes.
+  for (unsigned i = 0; i < F1PhiInNewF->getNumIncomingValues(); ++i) {
+    Value *F1InValNewF = F1PhiInNewF->getIncomingValue(i),
+          *F1InVal = F1PhiInst->getIncomingValue(i);
+    BasicBlock *F1NewFInBlock = F1PhiInNewF->getIncomingBlock(i);
+    // If this is a repeat occurrence of the same incoming BasicBlock, we
+    // will have already dealt with it in a previous iteration.
+    if (F1PhiInNewF->getBasicBlockIndex(F1PhiInNewF->getIncomingBlock(i)) !=
+        (int)i)
+      continue;
+
+    Value *NewIncoming = F1InValNewF;
+
+    Instruction *InsertPt = F1NewFInBlock->getTerminator();
+
+    // Build up a chain of cmps and selects that pick the correct incoming
+    // value.
+    for (unsigned FnI = 0, FnE = F2PhiInsts.size(); FnI != FnE; ++FnI) {
+      if (!F2PhiInsts[FnI])
+        continue;
+      const PHINode *F2PhiInst = cast<const PHINode>(F2PhiInsts[FnI]);
+      Value *F2InVal = F2PhiInst->getIncomingValue(i);
+
+      // If we know these are equivalent, there's no further work to do
+      if (Fns[FnI]->enumerate(F1InVal, F2InVal,/*NoSelfRef=*/true) &&
+          Fns[FnI]->enumerate(F1PhiInst->getIncomingBlock(i),
+            F2PhiInst->getIncomingBlock(i)))
+        continue;
+
+      assert(Fns[FnI]->enumerate(F1PhiInst->getIncomingBlock(i),
+                                 F2PhiInst->getIncomingBlock(i)) &&
+             "Non-equivalent incoming BBs in PHI.");
+
+      // We have different incoming values from the same block
+      // Translate F2's incoming value to NewF if needed
+      Value *F2InValNewF = F2InVal;
+      if (!isa<Constant>(F2InVal)) {
+        Value *V = Fns[FnI]->getF2toF1Map()[F2InVal]; // F2->F1
+        F2InValNewF = VMap[V]; // F1->NewF
+        assert(V && F2InValNewF && "Cannot map F2InVal to NewF");
+      }
+
+      // Cast F2InValNewF to the correct type if needed
+      LLVMContext &Ctx = F1InValNewF->getType()->getContext();
+      const DataLayout *FTD = Fns[FnI]->getDataLayout();
+      Type *IntPtrTy = FTD ? FTD->getIntPtrType(Ctx) : NULL;
+      F2InValNewF = createCastIfNeeded(F2InValNewF, F1InValNewF->getType(),
+                                       InsertPt, IntPtrTy);
+
+      // Create compare & select
+      Value *ChoiceArg = getLastArg(NewF);
+      Value *SelectBit = new ICmpInst(InsertPt,
+                                      ICmpInst::ICMP_EQ,
+                                      getLastArg(NewF),
+                                      ConstantInt::get(ChoiceArg->getType(),
+                                                       FnI+1));
+
+      // SelectBit true -> F2InValNewF, SelectBit false -> existing NewIncoming.
+      NewIncoming = SelectInst::Create(SelectBit, F2InValNewF, NewIncoming, "",
+          InsertPt);
+    }
+
+    if (NewIncoming == F1InValNewF)
+      continue; // no change for this incoming value
+
+    // Replace all occurrences of this incoming value/block by the new
+    // ones (phi nodes can have repeated arguments)
+    for (unsigned j=i; j < F1PhiInNewF->getNumIncomingValues(); ++j) {
+      if (F1PhiInNewF->getIncomingBlock(j) == F1NewFInBlock) {
+        F1PhiInNewF->setIncomingValue(j, NewIncoming);
+      }
+    }
+  }
+
+  // Garbage-collect the reordered PHI nodes we temporarily created.
+  for (SmallVectorImpl<PHINode *>::iterator I = GCInsts.begin(),
+        E = GCInsts.end(); I != E; ++I)
+    delete *I;
+}
+
+static bool rewriteRecursiveCall(
+    const CallInst *F1I, const CallInst *F2I, CallInst *NewFI,
+    const Function *F1, const Function *F2, Function *NewF) {
+  if (!(F1I->getCalledFunction() == F1 && F2I->getCalledFunction() == F2) &&
+      !(F1I->getCalledFunction() == F2 && F2I->getCalledFunction() == F1))
+    return false; // not a recursive/mutually recursive call
+
+  // Replace NewFI by recursive call to NewF with additional choice argument
+  SmallVector<Value *, 16> Args;
+  for (unsigned AI = 0, End = NewFI->getNumArgOperands(); AI < End; ++AI) {
+    Value *Arg = NewFI->getArgOperand(AI);
+
+    // Check if F1 or F2 is one of the arguments (veeery unusual case, don't
+    // handle it for now).
+    if (Arg == F1 || Arg == F2)
+      return false;
+
+    Args.push_back(Arg);
+  }
+
+  if (F1I->getCalledFunction() == F1 && F2I->getCalledFunction() == F2) {
+    Args.push_back(getLastArg(NewF));
+  } else {
+    // Need to invert the choice argument
+    Value *ChoiceArg = getLastArg(NewF);
+    Constant *One = ConstantInt::get(ChoiceArg->getType(), 1);
+    Args.push_back(BinaryOperator::Create(Instruction::Xor, ChoiceArg, One, "",
+          NewFI));
+  }
+
+  CallInst *CI = CallInst::Create(NewF, Args);
+  CI->setCallingConv(NewF->getCallingConv());
+
+  ReplaceInstWithInst(NewFI, CI);
+
+  return true;
+}
+
+/// Clone F1 into a new function with F1's name + MERGE_SUFFIX. Adds an
+/// additional i32 argument to the function.
+static Function *cloneAndAddArgument(Function *F1, ValueToValueMapTy &VMap) {
+  LLVMContext &Context = F1->getContext();
+
+  std::vector<Type*> ArgTypes;
+  for (const auto &Arg : F1->args())
+    ArgTypes.push_back(Arg.getType());
+  ArgTypes.push_back(Type::getInt32Ty(Context));
+
+  FunctionType *FTy = FunctionType::get(F1->getFunctionType()->getReturnType(),
+                                        ArgTypes,
+                                        F1->getFunctionType()->isVarArg());
+  Function *NewF = Function::Create(FTy, F1->getLinkage(),
+                                    F1->getName()+MERGED_SUFFIX);
+
+  insertFunctionAfter(NewF, F1);
+
+  if (F1->hasSection())
+    NewF->setSection(F1->getSection());
+
+  if (F1->getFunctionType()->isVarArg())
+    NewF->setCallingConv(CallingConv::C);
+  else
+    NewF->setCallingConv(CallingConv::Fast);
+
+  Function::arg_iterator DestI = NewF->arg_begin();
+  for (auto &Arg : F1->args()) {
+    Argument *DestIn = &*DestI;
+    DestIn->setName(Arg.getName()); // Copy the name over...
+    VMap[&Arg] = DestIn;            // Add mapping to VMap
+    ++DestI;
+  }
+
+  // Name the selector argument
+  (*DestI).setName("__merge_arg");
+
+  SmallVector<ReturnInst*, 8> Returns;
+  CloneFunctionInto(NewF, F1, VMap, CloneType::ExtractingFunctions, Returns);
+  // Set linkage to set visibility to default.
+  NewF->setLinkage(GlobalValue::InternalLinkage);
+
+  return NewF;
+}
+
+typedef MapVector<const Instruction *, std::vector<const Instruction *> >
+  CombinedDiffMap;
+
+void MergeSimilarFunctions::outlineAndMergeFunctions(
+    SmallVectorImpl<FunctionComparator *> &Fns) {
+  assert(!Fns.empty() && "Cannot merge empty set of functions");
+
+  // All comparator instances in Fns share the same F1
+  Function *F1 = Fns.front()->getF1();
+
+  // Clone F1 into new function with an additional i32 argument
+  ValueToValueMapTy VMap; // maps F1 values -> NewF values
+  Function *NewF = cloneAndAddArgument(F1, VMap);
+
+  // Combine all the DifferingInstructions maps in Fns into one single map of
+  // lists to aid the merging process.
+  //
+  // Map F1 instruction -> list of F2 instructions indexed by position in Fns.
+  CombinedDiffMap AllDifferingInstructions;
+  for (unsigned I = 0, E = Fns.size(); I != E; ++I) {
+    FunctionComparator *Comp = Fns[I];
+    for (MapVector<const Instruction *, const Instruction *>::iterator
+          DiffI = Comp->DifferingInstructions.begin(),
+          DiffE = Comp->DifferingInstructions.end();
+         DiffI != DiffE; ++DiffI) {
+      AllDifferingInstructions[DiffI->first].resize(Fns.size());
+      AllDifferingInstructions[DiffI->first][I] = DiffI->second;
+    }
+  }
+
+  // Merge differing PHI nodes. We need to handle these first because they could
+  // be affected later on when we split basic blocks, thus making them
+  // impossible to merge.
+  for (CombinedDiffMap::const_iterator I = AllDifferingInstructions.begin(),
+         E = AllDifferingInstructions.end();
+       I != E; ++I) {
+    const PHINode *F1PhiInst = dyn_cast<PHINode>(I->first);
+    if (!F1PhiInst)
+      continue;
+
+    const std::vector<const Instruction *> &F2PhiInsts = I->second;
+
+    mergePHINode(Fns, NewF, VMap, F1PhiInst, F2PhiInsts);
+  }
+
+  // Merge recursive calls
+  //
+  // TODO: We currently only support this optimization for pairs of functions.
+  // If more than two functions are merged, we mark the recursive calls as
+  // DifferingInstructions which causes switch statements to be inserted and
+  // recursive calls going through thunks. It wouldn't be too hard to implement
+  // self-recursive calls for multi-merges. *Mutually* recursive calls with
+  // multi-merges are a little trickier - that's why we leave it for now.
+  if (Fns.size() == 1) {
+    FunctionComparator *Comp = Fns.front();
+    for (MapVector<const Instruction *, const Instruction *>::const_iterator
+        I = Comp->SelfRefInstructions.begin(),
+        E = Comp->SelfRefInstructions.end();
+        I != E; ++I) {
+      const Instruction *F1I = I->first;
+      if (Comp->DifferingInstructions.count(F1I))
+        continue; // Differing in other ways too, so deal with it later.
+
+      // Attempt recursive call rewriting
+      if (isa<CallInst>(F1I)) {
+        const CallInst *F1Call = cast<const CallInst>(F1I);
+        const CallInst *F2Call = dyn_cast<const CallInst>(I->second);
+        CallInst *NewFCall = dyn_cast<CallInst>(VMap[F1I]);
+
+        if (F1Call && F2Call && NewFCall &&
+            rewriteRecursiveCall(F1Call, F2Call, NewFCall,
+                                 Comp->getF1(), Comp->getF2(), NewF))
+          continue;
+      }
+
+      // Can't rewrite it. Mark as differing and insert conditional later
+      Comp->DifferingInstructions[F1I] = I->second;
+    }
+  } else {
+    for (unsigned I = 0, E = Fns.size(); I != E; ++I) {
+      FunctionComparator *Comp = Fns[I];
+      for (MapVector<const Instruction *, const Instruction *>::const_iterator
+          II = Comp->SelfRefInstructions.begin(),
+          EE = Comp->SelfRefInstructions.end();
+          II != EE; ++II) {
+        const Instruction *F1I = II->first;
+        if (Comp->DifferingInstructions.count(F1I))
+          continue; // Differing in other ways too, so deal with it later.
+
+        AllDifferingInstructions[F1I].resize(Fns.size());
+        AllDifferingInstructions[F1I][I] = II->second;
+      }
+    }
+  }
+
+  // For each differing instruction, splice basic block, and insert conditional
+  LLVMContext &Context = NewF->getContext();
+  Type *IntPtrType = DL->getIntPtrType(Context);
+  for (CombinedDiffMap::const_iterator I = AllDifferingInstructions.begin(),
+         E = AllDifferingInstructions.end();
+       I != E; ++I) {
+    const Instruction *F1Inst = I->first;
+    const std::vector<const Instruction *> &F2Insts = I->second;
+
+    assert(VMap.find(F1Inst) != VMap.end() &&
+      "Cannot find differing inst!");
+    Instruction *F1InstInNewF = cast<Instruction>(VMap[F1Inst]);
+
+    if (isa<PHINode>(F1InstInNewF))
+      continue; // we already handled these above
+
+    insertCondAndRemapInstructions(F1InstInNewF, F2Insts,
+      NewF, VMap, Fns, IntPtrType);
+  }
+
+  // Replace functions with thunks
+  PrintMerges("FNSM", F1, NewF);
+  writeThunkWithChoice(NewF, F1, 0);
+  for (unsigned FnI = 0, FnE = Fns.size(); FnI != FnE; ++FnI) {
+    Function *F2 = Fns[FnI]->getF2();
+    PrintMerges("FNSM", F2, NewF);
+    writeThunkWithChoice(NewF, F2, FnI + 1);
+  }
+  NumSimilarFunctionsMerged += Fns.size() + 1;
+}
+
+// For each instruction used by the value, remove() the function that contains
+// the instruction. This should happen right before a call to RAUW.
+void MergeSimilarFunctions::removeUsers(Value *V) {
+  std::vector<Value *> Worklist;
+  Worklist.push_back(V);
+  while (!Worklist.empty()) {
+    Value *V = Worklist.back();
+    Worklist.pop_back();
+
+    for (User *U : V->users()) {
+      if (Instruction *I = dyn_cast<Instruction>(U)) {
+        Registry.remove(I->getParent()->getParent());
+      } else if (isa<GlobalValue>(U)) {
+        // do nothing
+      } else if (Constant *C = dyn_cast<Constant>(U)) {
+        for (User *UU : C->users())
+          Worklist.push_back(UU);
+      }
+    }
+  }
+}
+
+bool MergeSimilarFunctions::mergeBucket(std::list<ComparableFunction> &Fns,
+    bool &Changed) {
+  for (std::list<ComparableFunction>::iterator FnI = Fns.begin(),
+      FnE = Fns.end(); FnI != FnE; ++FnI) {
+    if (!FnI->isNew())
+      continue;
+
+    if (!FnI->getFunc())
+      continue;
+
+    SmallVector<FunctionComparator *, 8> DiffMergeCandidates;
+
+    std::list<ComparableFunction>::iterator Fn2I = FnI;
+    for (++Fn2I; Fn2I != FnE; ++Fn2I) {
+      if (!Fn2I->getFunc())
+        continue;
+
+      assert(FnI->getFunc() != Fn2I->getFunc() &&
+          "Duplicate function in list!");
+
+      FunctionComparator *Comp = new FunctionComparator(DL, FnI->getFunc(),
+          Fn2I->getFunc());
+
+      if (!Comp->compare() || !Comp->isMergeCandidate()) {
+        delete Comp;
+        continue;
+      }
+
+      // Never thunk a strong function to a weak function.
+      assert(!FnI->getFunc()->isInterposable() ||
+          Fn2I->getFunc()->isInterposable());
+
+      if (Comp->isExactMatch()) {
+        // Equiv merge the two functions. Throw away any diff merge
+        // candidate we might have found so far.
+        delete Comp;
+
+        LLVM_DEBUG(dbgs() << "- Equiv merge " << FnI->getFunc()->getName()
+            << " == " << Fn2I->getFunc()->getName() << '\n');
+
+        PrintMerges("FNEQ", FnI->getFunc(), Fn2I->getFunc());
+
+        Function *DeleteF = Fn2I->getFunc();
+        Registry.remove(DeleteF, /*reanalyze=*/false);
+
+        mergeTwoFunctions(FnI->getFunc(), DeleteF);
+
+        Changed = true;
+
+        // mergeTwoFunctions may have removed functions from this bucket and
+        // invalidated the iterators. Rescan the whole bucket, continuing
+        // from the current function (previous ones will have been
+        // markCompared())
+        for (SmallVector<FunctionComparator *, 8>::iterator
+            I = DiffMergeCandidates.begin(), E = DiffMergeCandidates.end();
+           I != E; ++I)
+          delete *I;
+
+        return false;
+      } else {
+        DiffMergeCandidates.push_back(Comp);
+      }
+    }
+
+    if (!DiffMergeCandidates.empty()) {
+      // Add to our list of candidates for diff merging
+      for (SmallVector<FunctionComparator *, 8>::iterator
+            I = DiffMergeCandidates.begin(), E = DiffMergeCandidates.end();
+           I != E; ++I) {
+        Registry.insertCandidate(*I);
+      }
+    }
+
+    FnI->markCompared();
+  }
+
+  return true;
+}
+
+bool MergeSimilarFunctions::doExhaustiveCompareMerge() {
+  bool Changed = false;
+
+  // Process buckets with strong functions first.
+  for (MergeRegistry::FnCompareMap::iterator
+        BucketsI = Registry.FunctionsToCompare.begin(),
+        BucketsE = Registry.FunctionsToCompare.end();
+       BucketsI != BucketsE; ++BucketsI) {
+    std::list<ComparableFunction> &Fns = BucketsI->second;
+    if (Fns.size() < 2 || Fns.front().getFunc()->isInterposable())
+      continue;
+
+    LLVM_DEBUG(dbgs() << "Processing strong bucket " << BucketsI->first << " with "
+                 << Fns.size() << " functions\n");
+    // Repeatedly scan this bucket, until we find no more functions to equiv
+    // merge.
+    while (!mergeBucket(Fns, Changed) && Fns.size() > 1) {
+      LLVM_DEBUG(dbgs() << "Rescanning bucket.\n");
+    }
+  }
+
+  // Process buckets with weak functions.
+  for (MergeRegistry::FnCompareMap::iterator
+        BucketsI = Registry.FunctionsToCompare.begin(),
+        BucketsE = Registry.FunctionsToCompare.end();
+       BucketsI != BucketsE; ++BucketsI) {
+    std::list<ComparableFunction> &Fns = BucketsI->second;
+    if (Fns.size() < 2 || !Fns.front().getFunc()->isInterposable())
+      continue;
+
+    LLVM_DEBUG(dbgs() << "Processing weak bucket " << BucketsI->first << " with "
+                 << Fns.size() << " functions\n");
+    // Repeatedly scan this bucket, until we find no more functions to equiv
+    // merge.
+    while (!mergeBucket(Fns, Changed) && Fns.size() > 1) {
+      LLVM_DEBUG(dbgs() << "Rescanning bucket.\n");
+    }
+  }
+
+  return Changed;
+}
+
+static bool orderComparatorsByMetric(FunctionComparator *Cmp1,
+                                     FunctionComparator *Cmp2) {
+  return (Cmp1->getSimilarityMetric() > Cmp2->getSimilarityMetric());
+}
+
+bool MergeSimilarFunctions::doDiffMerge() {
+  if (Registry.FunctionsToMerge.empty())
+    return false;
+
+  bool Changed = false;
+  DenseSet<Function *> MergedFns; // Functions that have already been merged
+  Registry.FunctionsToMerge.sort(orderComparatorsByMetric);
+
+  for (std::list<FunctionComparator *>::iterator
+        I = Registry.FunctionsToMerge.begin(),
+        E = Registry.FunctionsToMerge.end();
+       I != E; ++I) {
+    FunctionComparator *Comp = *I;
+    Function *F1 = Comp->getF1();
+    // Ignore it if we've already merged this fn
+    if (MergedFns.count(F1) || MergedFns.count(Comp->getF2()))
+      continue;
+
+    assert(Registry.SimilarFunctions.count(F1) &&
+        "Comparator doesn't exist in SimilarFunctions map");
+
+    // Look at all functions F that F1 could be merged with. Merge with each F,
+    // unless there is another function F' that is more similar to F than F1.
+    MergeRegistry::FnComparatorSet &SimilarFns = Registry.SimilarFunctions[F1];
+    SmallVector<FunctionComparator *, 4> CurrentMerge;
+
+    for (MergeRegistry::FnComparatorSet::iterator
+          CandidateI = SimilarFns.begin(), CandidateE = SimilarFns.end();
+        CandidateI != CandidateE; ++CandidateI) {
+      FunctionComparator *Comp2 = *CandidateI;
+      assert(Comp2->getF1() == F1 && "Inconsistency in SimilarFunctions");
+      Function *F2 = Comp2->getF2();
+
+      // Ignore it if we've already merged this fn
+      if (MergedFns.count(F2))
+        continue;
+
+      // Check whether there is a better merge candidate for F2
+      if (Registry.getMaxSimilarity(F2, MergedFns) >
+          Comp2->getSimilarityMetric())
+        continue;
+
+      // Ok, we actually want to merge with F2
+      CurrentMerge.push_back(Comp2);
+      MergedFns.insert(F2);
+    }
+
+    if (CurrentMerge.empty())
+      continue;
+
+    MergedFns.insert(F1);
+
+    NumMultiMerged += CurrentMerge.size();
+
+    LLVM_DEBUG(dbgs() << "- Multi merge of " << F1->getName() << " with "
+                 << CurrentMerge.size() << " functions.\n");
+
+    Changed = true;
+    outlineAndMergeFunctions(CurrentMerge);
+  }
+
+  return Changed;
+}

llvm/lib/Transforms/IPO/PassManagerBuilder.cpp

 cl::opt<bool> EnableOrderFileInstrumentation(
     "enable-order-file-instrumentation", cl::init(false), cl::Hidden,
     cl::desc("Enable order file instrumentation (default = off)"));
 
 static cl::opt<bool>
     EnableMatrix("enable-matrix", cl::init(false), cl::Hidden,
                  cl::desc("Enable lowering of the matrix intrinsics"));
 
+static cl::opt<bool> EnableMergeSimilarFunctions(
+    "enable-merge-sim-functions", cl::init(false), cl::Hidden,
+    cl::desc("Enable the Function merging pass (default = on)"));
+
 PassManagerBuilder::PassManagerBuilder() {
     OptLevel = 2;
     SizeLevel = 0;
     LibraryInfo = nullptr;
     Inliner = nullptr;
     DisableUnrollLoops = false;
     SLPVectorize = RunSLPVectorization;
     LoopVectorize = EnableLoopVectorization;
void PassManagerBuilder::populateModulePassManager(
   MPM.add(createAttributorCGSCCLegacyPass());
 
   // Try to perform OpenMP specific optimizations. This is a (quick!) no-op if
   // there are no OpenMP runtime calls present in the module.
   if (OptLevel > 1)
     MPM.add(createOpenMPOptLegacyPass());
 
   MPM.add(createPostOrderFunctionAttrsLegacyPass());
+  if (EnableMergeSimilarFunctions) {
+    auto *Summary = (ImportSummary ? ImportSummary : ExportSummary);
+    MPM.add(createMergeSimilarFunctionsPass(Summary));
+  }
+
   if (OptLevel > 2)
     MPM.add(createArgumentPromotionPass()); // Scalarize uninlined fn args
 
   addExtensionsToPM(EP_CGSCCOptimizerLate, MPM);
   addFunctionSimplificationPasses(MPM);
 
   // FIXME: This is a HACK! The inliner pass above implicitly creates a CGSCC
   // pass manager that we are specifically trying to avoid. To prevent this
void PassManagerBuilder::populateModulePassManager(
   // See comment in the new PM for justification of scheduling splitting at
   // this stage (\ref buildModuleSimplificationPipeline).
   if (EnableHotColdSplit && !(PrepareForLTO || PrepareForThinLTO))
     MPM.add(createHotColdSplittingPass());
 
   if (MergeFunctions)
     MPM.add(createMergeFunctionsPass());
 
+  if (EnableMergeSimilarFunctions)
+    MPM.add(createMergeSimilarFunctionsPass());
+
   // LoopSink pass sinks instructions hoisted by LICM, which serves as a
   // canonicalization pass that enables other optimizations. As a result,
   // LoopSink pass needs to be a very late IR pass to avoid undoing LICM
   // result too early.
   MPM.add(createLoopSinkPass());
   // Get rid of LCSSA nodes.
   MPM.add(createInstSimplifyLegacyPass());
 
void PassManagerBuilder::addLateLTOOptimizationPasses(
 
   // Now that we have optimized the program, discard unreachable functions.
   PM.add(createGlobalDCEPass());
 
   // FIXME: this is profitable (for compiler time) to do at -O0 too, but
   // currently it damages debug info.
   if (MergeFunctions)
     PM.add(createMergeFunctionsPass());
+  if (EnableMergeSimilarFunctions)
+    PM.add(createMergeSimilarFunctionsPass());
 }
 
 void PassManagerBuilder::populateThinLTOPassManager(
     legacy::PassManagerBase &PM) {
   PerformThinLTO = true;
   if (LibraryInfo)
     PM.add(new TargetLibraryInfoWrapperPass(*LibraryInfo));
 

llvm/lib/Transforms/Utils/CloneFunction.cpp

if (CodeInfo) {
     CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas &&
                                         BB != &BB->getParent()->getEntryBlock();
   }
   return NewBB;
 }
 
 // Clone OldFunc into NewFunc, transforming the old arguments into references to
 // VMap values.
-//
 void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
-                             ValueToValueMapTy &VMap,
-                             bool ModuleLevelChanges,
+                             ValueToValueMapTy &VMap, CloneType CT,
                              SmallVectorImpl<ReturnInst*> &Returns,
                              const char *NameSuffix, ClonedCodeInfo *CodeInfo,
                              ValueMapTypeRemapper *TypeMapper,
                              ValueMaterializer *Materializer) {
   assert(NameSuffix && "NameSuffix cannot be null!");
 
 #ifndef NDEBUG
   for (const Argument &I : OldFunc->args())
     assert(VMap.count(&I) && "No mapping from source argument specified!");
 #endif
 
   // Copy all attributes other than those stored in the AttributeList.  We need
   // to remap the parameter indices of the AttributeList.
   AttributeList NewAttrs = NewFunc->getAttributes();
   NewFunc->copyAttributesFrom(OldFunc);
   NewFunc->setAttributes(NewAttrs);
 
+  RemapFlags RF =
+      (CT == CloneType::ModuleLevelChanges) ? RF_None : RF_NoModuleLevelChanges;
   // Fix up the personality function that got copied over.
   if (OldFunc->hasPersonalityFn())
-    NewFunc->setPersonalityFn(
-        MapValue(OldFunc->getPersonalityFn(), VMap,
-                 ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges,
+    NewFunc->setPersonalityFn(MapValue(OldFunc->getPersonalityFn(), VMap, RF,
                                        TypeMapper, Materializer));
 
   SmallVector<AttributeSet, 4> NewArgAttrs(NewFunc->arg_size());
   AttributeList OldAttrs = OldFunc->getAttributes();
 
   // Clone any argument attributes that are present in the VMap.
   for (const Argument &OldArg : OldFunc->args()) {
     if (Argument *NewArg = dyn_cast<Argument>(VMap[&OldArg])) {
       NewArgAttrs[NewArg->getArgNo()] =
           OldAttrs.getParamAttributes(OldArg.getArgNo());
     }
   }
 
   NewFunc->setAttributes(
       AttributeList::get(NewFunc->getContext(), OldAttrs.getFnAttributes(),
                          OldAttrs.getRetAttributes(), NewArgAttrs));
 
-  bool MustCloneSP =
+  bool MustCloneSP = CT != CloneType::ExtractingFunctions &&
                      OldFunc->getParent() && OldFunc->getParent() == NewFunc->getParent();
   DISubprogram *SP = OldFunc->getSubprogram();
   if (SP) {
-    assert(!MustCloneSP || ModuleLevelChanges);
+    assert(!MustCloneSP || CT == CloneType::ModuleLevelChanges);
     // Add mappings for some DebugInfo nodes that we don't want duplicated
     // even if they're distinct.
     auto &MD = VMap.MD();
     MD[SP->getUnit()].reset(SP->getUnit());
     MD[SP->getType()].reset(SP->getType());
     MD[SP->getFile()].reset(SP->getFile());
     // If we're not cloning into the same module, no need to clone the
     // subprogram
     if (!MustCloneSP)
       MD[SP].reset(SP);
   }
 
   SmallVector<std::pair<unsigned, MDNode *>, 1> MDs;
   OldFunc->getAllMetadata(MDs);
   for (auto MD : MDs) {
     NewFunc->addMetadata(
-        MD.first,
-        *MapMetadata(MD.second, VMap,
-                     ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges,
-                     TypeMapper, Materializer));
+        MD.first, *MapMetadata(MD.second, VMap, RF, TypeMapper, Materializer));
   }
 
   // When we remap instructions, we want to avoid duplicating inlined
   // DISubprograms, so record all subprograms we find as we duplicate
   // instructions and then freeze them in the MD map.
   // We also record information about dbg.value and dbg.declare to avoid
   // duplicating the types.
   DebugInfoFinder DIFinder;
#endif
   // Loop over all of the instructions in the function, fixing up operand
   // references as we go.  This uses VMap to do all the hard work.
   for (Function::iterator BB =
            cast<BasicBlock>(VMap[&OldFunc->front()])->getIterator(),
                           BE = NewFunc->end();
        BB != BE; ++BB)
     // Loop over all instructions, fixing each one as we find it...
     for (Instruction &II : *BB)
-      RemapInstruction(&II, VMap,
-                       ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges,
-                       TypeMapper, Materializer);
+      RemapInstruction(&II, VMap, RF, TypeMapper, Materializer);
 
   // Register all DICompileUnits of the old parent module in the new parent module
   auto* OldModule = OldFunc->getParent();
   auto* NewModule = NewFunc->getParent();
   if (OldModule && NewModule && OldModule != NewModule && DIFinder.compile_unit_count()) {
     auto* NMD = NewModule->getOrInsertNamedMetadata("llvm.dbg.cu");
     // Avoid multiple insertions of the same DICompileUnit to NMD.
     SmallPtrSet<const void*, 8> Visited;
Function *llvm::CloneFunction(Function *F, ValueToValueMapTy &VMap,
   Function::arg_iterator DestI = NewF->arg_begin();
   for (const Argument & I : F->args())
     if (VMap.count(&I) == 0) {     // Is this argument preserved?
       DestI->setName(I.getName()); // Copy the name over...
       VMap[&I] = &*DestI++;        // Add mapping to VMap
     }
 
   SmallVector<ReturnInst*, 8> Returns;  // Ignore returns cloned.
-  CloneFunctionInto(NewF, F, VMap, F->getSubprogram() != nullptr, Returns, "",
-                    CodeInfo);
+  CloneType CT = F->getSubprogram() != nullptr ? CloneType::ModuleLevelChanges
+                                               : CloneType::InvalidCloneType;
+  CloneFunctionInto(NewF, F, VMap, CT, Returns, "", CodeInfo);
 
   return NewF;
 }
 
 
 
 namespace {
   /// This is a private class used to implement CloneAndPruneFunctionInto.

llvm/lib/Transforms/Utils/CloneModule.cpp

for (const Function &I : M) {
     Function::arg_iterator DestI = F->arg_begin();
     for (Function::const_arg_iterator J = I.arg_begin(); J != I.arg_end();
          ++J) {
       DestI->setName(J->getName());
       VMap[&*J] = &*DestI++;
     }
 
     SmallVector<ReturnInst *, 8> Returns; // Ignore returns cloned.
-    CloneFunctionInto(F, &I, VMap, /*ModuleLevelChanges=*/true, Returns);
+    CloneFunctionInto(F, &I, VMap, CloneType::ModuleLevelChanges, Returns);
 
     if (I.hasPersonalityFn())
       F->setPersonalityFn(MapValue(I.getPersonalityFn(), VMap));
 
     copyComdat(F, &I);
   }
 
   // And aliases

llvm/test/Transforms/MergeSimilarFunc/merge-alloca.ll

+;RUN: opt -mergesimilarfunc -mergesimilarfunc-level=all -S < %s | FileCheck %s
+;
+; Test whether mergefunc merges allocas of different sizes correctly
+;
+
+target datalayout = "e-m:e-p:32:32-i1:32-i64:64-a:0-v32:32-n16:32"
+
+%struct.A = type { i32, i32 }
+%struct.B = type { i32, i32, i32 }
+
+; Function Attrs: nounwind optsize
+define void @f1() #0 {
+; CHECK-LABEL: @f1__merged(
+; CHECK: alloca %struct.A
+; CHECK: alloca %struct.B
+entry:
+  %a = alloca %struct.A, align 4
+  %0 = bitcast %struct.A* %a to i8*
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  ret void
+}
+
+; Function Attrs: optsize
+declare void @externalFun(i8*) #1
+
+; Function Attrs: nounwind optsize
+define void @f2() #0 {
+entry:
+  %a = alloca %struct.B, align 4
+  %0 = bitcast %struct.B* %a to i8*
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  call void @externalFun(i8* %0) #2
+  ret void
+}
+
+; Function Attrs: nounwind optsize
+define void @f3() #0 {
+entry:
+  %a = alloca i8, align 1
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  call void @externalFun(i8* %a) #2
+  ret void
+}
+
+attributes #0 = { nounwind optsize "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { optsize "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #2 = { nounwind optsize }
+

llvm/test/Transforms/MergeSimilarFunc/merge-debug-info-2.ll

+; RUN: opt -S -mergesimilarfunc -mergesimilarfunc-diff-min-insts=5 < %s | FileCheck %s
+; This used to fail with assertion in CloneFunction
+; REQUIRES: asserts
+; CHECK-LABEL: @foo__merged(
+
+target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
+target triple = "arm64-apple-ios8.0.0"
+
+%struct.wibble = type { %struct.wibble.0*, %struct.wibble* }
+%struct.wibble.0 = type { i64*, i8*, i64*, i64*, %struct.eggs*, %struct.wibble* }
+%struct.eggs = type { %struct.wombat*, %struct.eggs* }
+%struct.wombat = type { i8*, %struct.blam*, %struct.blam* }
+%struct.blam = type { i8*, %struct.blam* }
+%struct.snork = type { %struct.bar*, %struct.snork* }
+%struct.bar = type { i64*, i8* }
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #0
+
+; Function Attrs: minsize nounwind optsize ssp uwtable
+define hidden void @foo(%struct.wibble* %arg) #1 align 2 !dbg !4 {
+bb:
+  %tmp = alloca %struct.wibble*, align 8
+  %tmp2 = load %struct.wibble*, %struct.wibble** %tmp, align 8, !dbg !12
+  %tmp3 = icmp ne %struct.wibble* %tmp2, null, !dbg !13
+  br i1 %tmp3, label %bb4, label %bb13, !dbg !14
+
+bb4:                                              ; preds = %bb
+  call void @foo.1() #3, !dbg !26
+  unreachable
+
+bb13:                                             ; preds = %bb
+  ret void, !dbg !27
+}
+
+; Function Attrs: minsize nounwind optsize ssp uwtable
+declare hidden void @foo.1() #1 align 2
+
+; Function Attrs: minsize nounwind optsize ssp uwtable
+define void @quux() unnamed_addr #1 align 2 !dbg !28 {
+bb:
+  ret void
+}
+
+; Function Attrs: minsize nounwind optsize ssp uwtable
+define hidden void @baz(%struct.snork* %arg) #1 align 2 !dbg !30 {
+bb:
+  %tmp = alloca %struct.snork*, align 8
+  %tmp2 = load %struct.snork*, %struct.snork** %tmp, align 8, !dbg !31
+  %tmp3 = icmp ne %struct.snork* %tmp2, null, !dbg !32
+  br i1 %tmp3, label %bb4, label %bb13, !dbg !33
+
+bb4:                                              ; preds = %bb
+  call void @blam() #3, !dbg !42
+  unreachable
+
+bb13:                                             ; preds = %bb
+  ret void, !dbg !43
+}
+
+; Function Attrs: minsize nounwind optsize ssp uwtable
+declare hidden void @blam() #1 align 2
+
+attributes #0 = { argmemonly nounwind }
+attributes #1 = { minsize nounwind optsize ssp uwtable }
+attributes #2 = { nounwind }
+attributes #3 = { minsize optsize }
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!2, !3}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !1, producer: "(based on LLVM 5.0.0)", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly)
+!1 = !DIFile(filename: "foo.cpp", directory: "/")
+!2 = !{i32 2, !"Debug Info Version", i32 3}
+!3 = !{i32 7, !"PIC Level", i32 2}
+!4 = distinct !DISubprogram(name: "delete", scope: !5, file: !5, line: 31, type: !6, isLocal: false, isDefinition: true, scopeLine: 32, flags: DIFlagPrototyped, isOptimized: true, unit: !0)
+!5 = !DIFile(filename: "foo.h", directory: "/")
+!6 = !DISubroutineType(types: !7)
+!7 = !{}
+!9 = !{!"any pointer", !10, i64 0}
+!10 = !{!"omnipotent char", !11, i64 0}
+!11 = !{!"Simple C++ TBAA"}
+!12 = !DILocation(line: 33, column: 11, scope: !4)
+!13 = !DILocation(line: 33, column: 16, scope: !4)
+!14 = !DILocation(line: 33, column: 5, scope: !4)
+!21 = !{!"_ZTSN", !9, i64 0, !9, i64 8}
+!23 = !DILocation(line: 37, column: 25, scope: !4)
+!24 = !DILocation(line: 37, column: 31, scope: !4)
+!25 = !{!21, !9, i64 0}
+!26 = !DILocation(line: 37, column: 7, scope: !4)
+!27 = !DILocation(line: 41, column: 3, scope: !4)
+!28 = distinct !DISubprogram(name: "~destruct", scope: !29, file: !29, line: 31, type: !6, isLocal: false, isDefinition: true, scopeLine: 32, flags: DIFlagPrototyped, isOptimized: true, unit: !0)
+!29 = !DIFile(filename: "bar.h", directory: "/")
+!30 = distinct !DISubprogram(name: "delete", scope: !5, file: !5, line: 31, type: !6, isLocal: false, isDefinition: true, scopeLine: 32, flags: DIFlagPrototyped, isOptimized: true, unit: !0)
+!31 = !DILocation(line: 33, column: 11, scope: !30)
+!32 = !DILocation(line: 33, column: 16, scope: !30)
+!33 = !DILocation(line: 33, column: 5, scope: !30)
+!40 = !DILocation(line: 37, column: 25, scope: !30)
+!41 = !DILocation(line: 37, column: 31, scope: !30)
+!42 = !DILocation(line: 37, column: 7, scope: !30)
+!43 = !DILocation(line: 41, column: 3, scope: !30)

llvm/test/Transforms/MergeSimilarFunc/merge-debug-info.ll

+; This used to fail the verifier with the following error:
+; "dbg attachment points at wrong subprogram for function"
+; RUN: opt -S -mergesimilarfunc < %s | FileCheck %s
+; REQUIRES: asserts
+; CHECK-LABEL: @bar__merged(
+target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"
+
+%struct.str_type = type { i8*, i32, i32 }
+
+; Function Attrs: nounwind optsize
+define i32 @bar(i8* %b) #0 !dbg !14 {
+entry:
+  %retval = alloca i32, align 4
+  %b.addr = alloca i8*, align 4
+  %res = alloca i32, align 4
+  %ee = alloca %struct.str_type*, align 4
+  %cleanup.dest.slot = alloca i32
+  store i8* %b, i8** %b.addr, align 4, !tbaa !31
+  call void @llvm.dbg.declare(metadata i8** %b.addr, metadata !18, metadata !35), !dbg !36
+  %0 = bitcast i32* %res to i8*, !dbg !37
+  call void @llvm.lifetime.start(i64 4, i8* %0) #4, !dbg !37
+  call void @llvm.dbg.declare(metadata i32* %res, metadata !19, metadata !35), !dbg !38
+  %1 = bitcast %struct.str_type** %ee to i8*, !dbg !39
+  call void @llvm.lifetime.start(i64 4, i8* %1) #4, !dbg !39
+  call void @llvm.dbg.declare(metadata %struct.str_type** %ee, metadata !20, metadata !35), !dbg !40
+  %2 = load i8*, i8** %b.addr, align 4, !dbg !41, !tbaa !31
+  %3 = bitcast i8* %2 to %struct.str_type*, !dbg !42
+  store %struct.str_type* %3, %struct.str_type** %ee, align 4, !dbg !40, !tbaa !31
+  %4 = load %struct.str_type*, %struct.str_type** %ee, align 4, !dbg !43, !tbaa !31
+  %set = getelementptr inbounds %struct.str_type, %struct.str_type* %4, i32 0, i32 1, !dbg !45
+  %5 = load i32, i32* %set, align 4, !dbg !45, !tbaa !46
+  %tobool = icmp ne i32 %5, 0, !dbg !43
+  br i1 %tobool, label %if.end4, label %if.then, !dbg !49
+
+if.then:                                          ; preds = %entry
+  %6 = load %struct.str_type*, %struct.str_type** %ee, align 4, !dbg !50, !tbaa !31
+  %set1 = getelementptr inbounds %struct.str_type, %struct.str_type* %6, i32 0, i32 1, !dbg !52
+  store i32 1, i32* %set1, align 4, !dbg !53, !tbaa !46
+  %7 = load %struct.str_type*, %struct.str_type** %ee, align 4, !dbg !54, !tbaa !31
+  %x = getelementptr inbounds %struct.str_type, %struct.str_type* %7, i32 0, i32 0, !dbg !55
+  %8 = load i8*, i8** %x, align 4, !dbg !55, !tbaa !56
+  %call = call i32 @foo(i8* %8) #5, !dbg !57
+  store i32 %call, i32* %res, align 4, !dbg !58, !tbaa !59
+  %9 = load i32, i32* %res, align 4, !dbg !60, !tbaa !59
+  %tobool2 = icmp ne i32 %9, 0, !dbg !60
+  br i1 %tobool2, label %if.then3, label %if.end, !dbg !62
+
+if.then3:                                         ; preds = %if.then
+  store i32 1, i32* %retval, align 4, !dbg !63
+  store i32 1, i32* %cleanup.dest.slot, align 4
+  br label %cleanup, !dbg !63
+
+if.end:                                           ; preds = %if.then
+  br label %if.end4, !dbg !64
+
+if.end4:                                          ; preds = %if.end, %entry
+  store i32 0, i32* %retval, align 4, !dbg !65
+  store i32 1, i32* %cleanup.dest.slot, align 4
+  br label %cleanup, !dbg !65
+
+cleanup:                                          ; preds = %if.end4, %if.then3
+  %10 = bitcast %struct.str_type** %ee to i8*, !dbg !66
+  call void @llvm.lifetime.end(i64 4, i8* %10) #4, !dbg !66
+  %11 = bitcast i32* %res to i8*, !dbg !66
+  call void @llvm.lifetime.end(i64 4, i8* %11) #4, !dbg !66
+  %12 = load i32, i32* %retval, align 4, !dbg !66
+  ret i32 %12, !dbg !66
+}
+
+; Function Attrs: nounwind readnone
+declare void @llvm.dbg.declare(metadata, metadata, metadata) #1
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start(i64, i8* nocapture) #2
+
+; Function Attrs: optsize
+declare i32 @foo(i8*) #3
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end(i64, i8* nocapture) #2
+
+; Function Attrs: nounwind optsize
+define i32 @bar1(i8* %b) #0 !dbg !21 {
+entry:
+  %retval = alloca i32, align 4
+  %b.addr = alloca i8*, align 4
+  %res = alloca i32, align 4
+  %ee = alloca %struct.str_type*, align 4
+  %cleanup.dest.slot = alloca i32
+  store i8* %b, i8** %b.addr, align 4, !tbaa !31
+  call void @llvm.dbg.declare(metadata i8** %b.addr, metadata !23, metadata !35), !dbg !67
+  %0 = bitcast i32* %res to i8*, !dbg !68
+  call void @llvm.lifetime.start(i64 4, i8* %0) #4, !dbg !68
+  call void @llvm.dbg.declare(metadata i32* %res, metadata !24, metadata !35), !dbg !69
+  %1 = bitcast %struct.str_type** %ee to i8*, !dbg !70
+  call void @llvm.lifetime.start(i64 4, i8* %1) #4, !dbg !70
+  call void @llvm.dbg.declare(metadata %struct.str_type** %ee, metadata !25, metadata !35), !dbg !71
+  %2 = load i8*, i8** %b.addr, align 4, !dbg !72, !tbaa !31
+  %3 = bitcast i8* %2 to %struct.str_type*, !dbg !73
+  store %struct.str_type* %3, %struct.str_type** %ee, align 4, !dbg !71, !tbaa !31
+  %4 = load %struct.str_type*, %struct.str_type** %ee, align 4, !dbg !74, !tbaa !31
+  %get = getelementptr inbounds %struct.str_type, %struct.str_type* %4, i32 0, i32 2, !dbg !76
+  %5 = load i32, i32* %get, align 4, !dbg !76, !tbaa !77
+  %tobool = icmp ne i32 %5, 0, !dbg !74
+  br i1 %tobool, label %if.end4, label %if.then, !dbg !78
+
+if.then:                                          ; preds = %entry
+  %6 = load %struct.str_type*, %struct.str_type** %ee, align 4, !dbg !79, !tbaa !31
+  %get1 = getelementptr inbounds %struct.str_type, %struct.str_type* %6, i32 0, i32 2, !dbg !81
+  store i32 1, i32* %get1, align 4, !dbg !82, !tbaa !77
+  %7 = load %struct.str_type*, %struct.str_type** %ee, align 4, !dbg !83, !tbaa !31
+  %x = getelementptr inbounds %struct.str_type, %struct.str_type* %7, i32 0, i32 0, !dbg !84
+  %8 = load i8*, i8** %x, align 4, !dbg !84, !tbaa !56
+  %call = call i32 @foo(i8* %8) #5, !dbg !85
+  store i32 %call, i32* %res, align 4, !dbg !86, !tbaa !59
+  %9 = load i32, i32* %res, align 4, !dbg !87, !tbaa !59
+  %tobool2 = icmp ne i32 %9, 0, !dbg !87
+  br i1 %tobool2, label %if.then3, label %if.end, !dbg !89
+
+if.then3:                                         ; preds = %if.then
+  store i32 1, i32* %retval, align 4, !dbg !90
+  store i32 1, i32* %cleanup.dest.slot, align 4
+  br label %cleanup, !dbg !90
+
+if.end:                                           ; preds = %if.then
+  br label %if.end4, !dbg !91
+
+if.end4:                                          ; preds = %if.end, %entry
+  store i32 0, i32* %retval, align 4, !dbg !92
+  store i32 1, i32* %cleanup.dest.slot, align 4
+  br label %cleanup, !dbg !92
+
+cleanup:                                          ; preds = %if.end4, %if.then3
+  %10 = bitcast %struct.str_type** %ee to i8*, !dbg !93
+  call void @llvm.lifetime.end(i64 4, i8* %10) #4, !dbg !93
+  %11 = bitcast i32* %res to i8*, !dbg !93
+  call void @llvm.lifetime.end(i64 4, i8* %11) #4, !dbg !93
+  %12 = load i32, i32* %retval, align 4, !dbg !93
+  ret i32 %12, !dbg !93
+}
+
+attributes #0 = { nounwind optsize }
+attributes #1 = { nounwind readnone }
+attributes #2 = { argmemonly nounwind }
+attributes #3 = { optsize }
+attributes #4 = { nounwind }
+attributes #5 = { optsize }
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!26, !27}
+!llvm.ident = !{!28}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "Clang $LLVM_VERSION_MAJOR.$LLVM_VERSION_MINOR (based on LLVM 3.9.0)", isOptimized: true, runtimeVersion: 0, emissionKind: 1, enums: !2, retainedTypes: !3)
+!1 = !DIFile(filename: "test.i", directory: "/local/mnt/")
+!2 = !{}
+!3 = !{!4}
+!4 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !5, size: 32, align: 32)
+!5 = !DIDerivedType(tag: DW_TAG_typedef, name: "str_type", file: !1, line: 8, baseType: !6)
+!6 = !DICompositeType(tag: DW_TAG_structure_type, name: "str_type", file: !1, line: 3, size: 96, align: 32, elements: !7)
+!7 = !{!8, !10, !12}
+!8 = !DIDerivedType(tag: DW_TAG_member, name: "x", scope: !6, file: !1, line: 5, baseType: !9, size: 32, align: 32)
+!9 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: null, size: 32, align: 32)
+!10 = !DIDerivedType(tag: DW_TAG_member, name: "set", scope: !6, file: !1, line: 6, baseType: !11, size: 32, align: 32, offset: 32)
+!11 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
+!12 = !DIDerivedType(tag: DW_TAG_member, name: "get", scope: !6, file: !1, line: 7, baseType: !11, size: 32, align: 32, offset: 64)
+!14 = distinct !DISubprogram(name: "bar", scope: !1, file: !1, line: 10, type: !15, isLocal: false, isDefinition: true, scopeLine: 10, flags: DIFlagPrototyped, isOptimized: true, unit: !0, variables: !17)
+!15 = !DISubroutineType(types: !16)
+!16 = !{!11, !9}
+!17 = !{!18, !19, !20}
+!18 = !DILocalVariable(name: "b", arg: 1, scope: !14, file: !1, line: 10, type: !9)
+!19 = !DILocalVariable(name: "res", scope: !14, file: !1, line: 11, type: !11)
+!20 = !DILocalVariable(name: "ee", scope: !14, file: !1, line: 12, type: !4)
+!21 = distinct !DISubprogram(name: "bar1", scope: !1, file: !1, line: 24, type: !15, isLocal: false, isDefinition: true, scopeLine: 24, flags: DIFlagPrototyped, isOptimized: true, unit: !0, variables: !22)
+!22 = !{!23, !24, !25}
+!23 = !DILocalVariable(name: "b", arg: 1, scope: !21, file: !1, line: 24, type: !9)
+!24 = !DILocalVariable(name: "res", scope: !21, file: !1, line: 25, type: !11)
+!25 = !DILocalVariable(name: "ee", scope: !21, file: !1, line: 26, type: !4)
+!26 = !{i32 2, !"Dwarf Version", i32 4}
+!27 = !{i32 2, !"Debug Info Version", i32 3}
+!28 = !{!"Clang $LLVM_VERSION_MAJOR.$LLVM_VERSION_MINOR (based on LLVM 3.9.0)"}
+!31 = !{!32, !32, i64 0}
+!32 = !{!"any pointer", !33, i64 0}
+!33 = !{!"omnipotent char", !34, i64 0}
+!34 = !{!"Simple C/C++ TBAA"}
+!35 = !DIExpression()
+!36 = !DILocation(line: 10, column: 23, scope: !14)
+!37 = !DILocation(line: 11, column: 3, scope: !14)
+!38 = !DILocation(line: 11, column: 7, scope: !14)
+!39 = !DILocation(line: 12, column: 3, scope: !14)
+!40 = !DILocation(line: 12, column: 13, scope: !14)
+!41 = !DILocation(line: 12, column: 31, scope: !14)
+!42 = !DILocation(line: 12, column: 19, scope: !14)
+!43 = !DILocation(line: 14, column: 8, scope: !44)
+!44 = distinct !DILexicalBlock(scope: !14, file: !1, line: 14, column: 7)
+!45 = !DILocation(line: 14, column: 12, scope: !44)
+!46 = !{!47, !48, i64 4}
+!47 = !{!"str_type", !32, i64 0, !48, i64 4, !48, i64 8}
+!48 = !{!"int", !33, i64 0}
+!49 = !DILocation(line: 14, column: 7, scope: !14)
+!50 = !DILocation(line: 15, column: 5, scope: !51)
+!51 = distinct !DILexicalBlock(scope: !44, file: !1, line: 14, column: 18)
+!52 = !DILocation(line: 15, column: 9, scope: !51)
+!53 = !DILocation(line: 15, column: 13, scope: !51)
+!54 = !DILocation(line: 17, column: 16, scope: !51)
+!55 = !DILocation(line: 17, column: 20, scope: !51)
+!56 = !{!47, !32, i64 0}
+!57 = !DILocation(line: 17, column: 11, scope: !51)
+!58 = !DILocation(line: 17, column: 9, scope: !51)
+!59 = !{!48, !48, i64 0}
+!60 = !DILocation(line: 18, column: 9, scope: !61)
+!61 = distinct !DILexicalBlock(scope: !51, file: !1, line: 18, column: 9)
+!62 = !DILocation(line: 18, column: 9, scope: !51)
+!63 = !DILocation(line: 19, column: 7, scope: !61)
+!64 = !DILocation(line: 20, column: 3, scope: !51)
+!65 = !DILocation(line: 21, column: 3, scope: !14)
+!66 = !DILocation(line: 22, column: 1, scope: !14)
+!67 = !DILocation(line: 24, column: 24, scope: !21)
+!68 = !DILocation(line: 25, column: 3, scope: !21)
+!69 = !DILocation(line: 25, column: 7, scope: !21)
+!70 = !DILocation(line: 26, column: 3, scope: !21)
+!71 = !DILocation(line: 26, column: 13, scope: !21)
+!72 = !DILocation(line: 26, column: 31, scope: !21)
+!73 = !DILocation(line: 26, column: 19, scope: !21)
+!74 = !DILocation(line: 28, column: 8, scope: !75)
+!75 = distinct !DILexicalBlock(scope: !21, file: !1, line: 28, column: 7)
+!76 = !DILocation(line: 28, column: 12, scope: !75)
+!77 = !{!47, !48, i64 8}
+!78 = !DILocation(line: 28, column: 7, scope: !21)
+!79 = !DILocation(line: 29, column: 5, scope: !80)
+!80 = distinct !DILexicalBlock(scope: !75, file: !1, line: 28, column: 17)
+!81 = !DILocation(line: 29, column: 9, scope: !80)
+!82 = !DILocation(line: 29, column: 13, scope: !80)
+!83 = !DILocation(line: 31, column: 16, scope: !80)
+!84 = !DILocation(line: 31, column: 20, scope: !80)
+!85 = !DILocation(line: 31, column: 11, scope: !80)
+!86 = !DILocation(line: 31, column: 9, scope: !80)
+!87 = !DILocation(line: 32, column: 9, scope: !88)
+!88 = distinct !DILexicalBlock(scope: !80, file: !1, line: 32, column: 9)
+!89 = !DILocation(line: 32, column: 9, scope: !80)
+!90 = !DILocation(line: 33, column: 7, scope: !88)
+!91 = !DILocation(line: 34, column: 3, scope: !80)
+!92 = !DILocation(line: 35, column: 3, scope: !21)
+!93 = !DILocation(line: 36, column: 1, scope: !21)

llvm/test/Transforms/MergeSimilarFunc/merge-equivalent-template.ll

+; RUN: opt -S -mergesimilarfunc %s -o - | FileCheck %s
+
+; CHECK:      define linkonce_odr void @_ZN11FooTemplateIPvE7method1EP9FooStruct(%class.FooTemplate.0* nocapture readnone %this, %struct.FooStruct* %arg0) #1 align 2 {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %field27 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %arg0, i32 0, i32 2
+; CHECK-NEXT:   %0 = load i8, i8* %field27, align 1
+; CHECK-NEXT:   %lnot8 = icmp eq i8 %0, 0
+; CHECK-NEXT:   br i1 %lnot8, label %for.body, label %for.end
+
+; CHECK:      for.body:                                         ; preds = %for.body, %entry
+; CHECK-NEXT:   %arg0.addr.09 = phi %struct.FooStruct* [ %2, %for.body ], [ %arg0, %entry ]
+; CHECK-NEXT:   %field1 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %arg0.addr.09, i32 0, i32 1
+; CHECK-NEXT:   %1 = load %struct.FooStruct*, %struct.FooStruct** %field1, align 4
+; CHECK-NEXT:   tail call void @_ZN11FooTemplateIPvE7method1EP9FooStruct(%class.FooTemplate.0* %this, %struct.FooStruct* %1) #2
+; CHECK-NEXT:   %field0 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %arg0.addr.09, i32 0, i32 0
+; CHECK-NEXT:   %2 = load %struct.FooStruct*, %struct.FooStruct** %field0, align 4
+; CHECK-NEXT:   %3 = bitcast %struct.FooStruct* %arg0.addr.09 to i8*
+; CHECK-NEXT:   tail call void @_Z4bar0Pv(i8* %3) #3
+; CHECK-NEXT:   tail call void @_Z4bar1Pv(i8* %3) #3
+; CHECK-NEXT:   %field2 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %2, i32 0, i32 2
+; CHECK-NEXT:   %4 = load i8, i8* %field2, align 1
+; CHECK-NEXT:   %lnot = icmp eq i8 %4, 0
+; CHECK-NEXT:   br i1 %lnot, label %for.body, label %for.end
+
+; CHECK:      for.end:                                          ; preds = %for.body, %entry
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+
+; CHECK:      define linkonce_odr void @_ZN11FooTemplateIiE7method1EP9FooStruct(%class.FooTemplate* nocapture readnone, %struct.FooStruct*) #1 align 2 {
+; CHECK-NEXT:   %3 = bitcast %class.FooTemplate* %0 to %class.FooTemplate.0*
+; CHECK-NEXT:   tail call void @_ZN11FooTemplateIPvE7method1EP9FooStruct(%class.FooTemplate.0* nocapture readnone %3, %struct.FooStruct* %1)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+
+; CHECK:      attributes #0 = { optsize "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+; CHECK-NEXT: attributes #1 = { nounwind optsize ssp "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+; CHECK-NEXT: attributes #2 = { optsize }
+; CHECK-NEXT: attributes #3 = { nounwind optsize }
+
+target datalayout = "e-m:e-p:32:32-i1:32-i64:64-a:0-v32:32-n16:32"
+
+%class.Foo = type { %class.FooTemplate, %class.FooTemplate.0 }
+%class.FooTemplate = type { i8 }
+%class.FooTemplate.0 = type { i8 }
+%struct.FooStruct = type { %struct.FooStruct*, %struct.FooStruct*, i8 }
+
+@_ZN3FooD1Ev = alias void(%class.Foo*), void (%class.Foo*)* @_ZN3FooD2Ev
+
+declare %struct.FooStruct* @_ZNK11FooTemplateIPvE7method2Ev(%class.FooTemplate.0*) #1
+declare void @_Z4bar0Pv(i8*) #1
+declare void @_Z4bar1Pv(i8*) #1
+declare %struct.FooStruct* @_ZNK11FooTemplateIiE7method2Ev(%class.FooTemplate*) #1
+
+; Function Attrs: nounwind optsize ssp
+define void @_ZN3FooD2Ev(%class.Foo* %this) unnamed_addr #0 align 2 {
+entry:
+  %bar_things = getelementptr inbounds %class.Foo, %class.Foo* %this, i32 0, i32 0
+  tail call void @_ZN11FooTemplateIiE7method0Ev(%class.FooTemplate* %bar_things) #2
+  %baz_things = getelementptr inbounds %class.Foo, %class.Foo* %this, i32 0, i32 1
+  tail call void @_ZN11FooTemplateIPvE7method0Ev(%class.FooTemplate.0* %baz_things) #2
+  ret void
+}
+
+; Function Attrs: nounwind optsize ssp
+define linkonce_odr void @_ZN11FooTemplateIiE7method0Ev(%class.FooTemplate* %this) #0 align 2 {
+entry:
+  %call = tail call %struct.FooStruct* @_ZNK11FooTemplateIiE7method2Ev(%class.FooTemplate* %this) #3
+  tail call void @_ZN11FooTemplateIiE7method1EP9FooStruct(%class.FooTemplate* %this, %struct.FooStruct* %call) #2
+  ret void
+}
+
+; Function Attrs: nounwind optsize ssp
+define linkonce_odr void @_ZN11FooTemplateIPvE7method0Ev(%class.FooTemplate.0* %this) #0 align 2 {
+entry:
+  %call = tail call %struct.FooStruct* @_ZNK11FooTemplateIPvE7method2Ev(%class.FooTemplate.0* %this) #3
+  tail call void @_ZN11FooTemplateIPvE7method1EP9FooStruct(%class.FooTemplate.0* %this, %struct.FooStruct* %call) #2
+  ret void
+}
+
+; Function Attrs: nounwind optsize ssp
+define linkonce_odr void @_ZN11FooTemplateIPvE7method1EP9FooStruct(%class.FooTemplate.0* nocapture readnone %this, %struct.FooStruct* %arg0) #0 align 2 {
+entry:
+  %field27 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %arg0, i32 0, i32 2
+  %0 = load i8, i8* %field27, align 1
+  %lnot8 = icmp eq i8 %0, 0
+  br i1 %lnot8, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %arg0.addr.09 = phi %struct.FooStruct* [ %2, %for.body ], [ %arg0, %entry ]
+  %field1 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %arg0.addr.09, i32 0, i32 1
+  %1 = load %struct.FooStruct*, %struct.FooStruct** %field1, align 4
+  tail call void @_ZN11FooTemplateIPvE7method1EP9FooStruct(%class.FooTemplate.0* %this, %struct.FooStruct* %1) #2
+  %field0 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %arg0.addr.09, i32 0, i32 0
+  %2 = load %struct.FooStruct*, %struct.FooStruct** %field0, align 4
+  %3 = bitcast %struct.FooStruct* %arg0.addr.09 to i8*
+  tail call void @_Z4bar0Pv(i8* %3) #3
+  tail call void @_Z4bar1Pv(i8* %3) #3
+  %field2 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %2, i32 0, i32 2
+  %4 = load i8, i8* %field2, align 1
+  %lnot = icmp eq i8 %4, 0
+  br i1 %lnot, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+; Function Attrs: nounwind optsize ssp
+define linkonce_odr void @_ZN11FooTemplateIiE7method1EP9FooStruct(%class.FooTemplate* nocapture readnone %this, %struct.FooStruct* %arg0) #0 align 2 {
+entry:
+  %field27 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %arg0, i32 0, i32 2
+  %0 = load i8, i8* %field27, align 1
+  %lnot8 = icmp eq i8 %0, 0
+  br i1 %lnot8, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %arg0.addr.09 = phi %struct.FooStruct* [ %2, %for.body ], [ %arg0, %entry ]
+  %field1 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %arg0.addr.09, i32 0, i32 1
+  %1 = load %struct.FooStruct*, %struct.FooStruct** %field1, align 4
+  tail call void @_ZN11FooTemplateIiE7method1EP9FooStruct(%class.FooTemplate* %this, %struct.FooStruct* %1) #2
+  %field0 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %arg0.addr.09, i32 0, i32 0
+  %2 = load %struct.FooStruct*, %struct.FooStruct** %field0, align 4
+  %3 = bitcast %struct.FooStruct* %arg0.addr.09 to i8*
+  tail call void @_Z4bar0Pv(i8* %3) #3
+  tail call void @_Z4bar1Pv(i8* %3) #3
+  %field2 = getelementptr inbounds %struct.FooStruct, %struct.FooStruct* %2, i32 0, i32 2
+  %4 = load i8, i8* %field2, align 1
+  %lnot = icmp eq i8 %4, 0
+  br i1 %lnot, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}
+
+attributes #0 = { nounwind optsize ssp "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { optsize "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #2 = { optsize }
+attributes #3 = { nounwind optsize }
+

llvm/test/Transforms/MergeSimilarFunc/merge-equivalent.ll

+; RUN: opt -mergesimilarfunc -S %s -o - | FileCheck %s
+;
+; CHECK:      define i8* @foo_a(%struct.a_type* %arg0) #1 {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   br label %for.cond
+
+; CHECK:      for.cond:                                         ; preds = %for.inc24, %entry
+; CHECK-NEXT:   %i.0 = phi i8 [ 0, %entry ], [ %inc25, %for.inc24 ]
+; CHECK-NEXT:   %ptr0.0 = phi i8* [ null, %entry ], [ %ptr0.3, %for.inc24 ]
+; CHECK-NEXT:   %conv = zext i8 %i.0 to i32
+; CHECK-NEXT:   %cmp = icmp slt i32 %conv, 16
+; CHECK-NEXT:   br i1 %cmp, label %for.body, label %for.end26
+
+; CHECK:      for.body:                                         ; preds = %for.cond
+; CHECK-NEXT:   %call = call i8* @bar4(i32 %conv) #2
+; CHECK-NEXT:   %call4 = call i8* @bar0(i32 %conv) #2
+; CHECK-NEXT:   %0 = bitcast i8* %call4 to %struct.a_type*
+; CHECK-NEXT:   %call5 = call i32 @bar1(i8* %call) #2
+; CHECK-NEXT:   %tobool = icmp ne i32 %call5, 0
+; CHECK-NEXT:   br i1 %tobool, label %for.cond6, label %for.inc24
+
+; CHECK:      for.cond6:                                        ; preds = %for.inc, %for.body
+; CHECK-NEXT:   %k.0 = phi i8 [ %inc, %for.inc ], [ 0, %for.body ]
+; CHECK-NEXT:   %ptr0.1 = phi i8* [ %ptr0.2, %for.inc ], [ %call, %for.body ]
+; CHECK-NEXT:   %idxprom = zext i8 %k.0 to i32
+; CHECK-NEXT:   %field4 = getelementptr inbounds %struct.a_type, %struct.a_type* %arg0, i32 0, i32 4
+; CHECK-NEXT:   %arrayidx = getelementptr inbounds [2 x i8*], [2 x i8*]* %field4, i32 0, i32 %idxprom
+; CHECK-NEXT:   %1 = load i8*, i8** %arrayidx, align 4
+; CHECK-NEXT:   %cmp7 = icmp ne i8* %1, null
+; CHECK-NEXT:   br i1 %cmp7, label %land.rhs, label %for.inc24
+
+; CHECK:      land.rhs:                                         ; preds = %for.cond6
+; CHECK-NEXT:   %field410 = getelementptr inbounds %struct.a_type, %struct.a_type* %0, i32 0, i32 4
+; CHECK-NEXT:   %arrayidx11 = getelementptr inbounds [2 x i8*], [2 x i8*]* %field410, i32 0, i32 %idxprom
+; CHECK-NEXT:   %2 = load i8*, i8** %arrayidx11, align 4
+; CHECK-NEXT:   %cmp12 = icmp ne i8* %2, null
+; CHECK-NEXT:   br i1 %cmp12, label %for.body14, label %for.inc24
+
+; CHECK:      for.body14:                                       ; preds = %land.rhs
+; CHECK-NEXT:   %3 = bitcast %struct.a_type* %0 to i8*
+; CHECK-NEXT:   %4 = bitcast %struct.a_type* %arg0 to i8*
+; CHECK-NEXT:   %call15 = call i32 @bar2(i8* %3, i8* %4) #2
+; CHECK-NEXT:   %tobool16 = icmp ne i32 %call15, 0
+; CHECK-NEXT:   br i1 %tobool16, label %if.then17, label %for.inc
+
+; CHECK:      if.then17:                                        ; preds = %for.body14
+; CHECK-NEXT:   %call18 = call i32 @bar3(i8* %3, i8* %4) #2
+; CHECK-NEXT:   %tobool19 = icmp ne i32 %call18, 0
+; CHECK-NEXT:   br i1 %tobool19, label %if.then20, label %for.inc
+
+; CHECK:      if.then20:                                        ; preds = %if.then17
+; CHECK-NEXT:   br label %for.inc
+
+; CHECK:      for.inc:                                          ; preds = %if.then20, %if.then17, %for.body14
+; CHECK-NEXT:   %ptr0.2 = phi i8* [ null, %if.then20 ], [ %ptr0.1, %if.then17 ], [ null, %for.body14 ]
+; CHECK-NEXT:   %inc = add i8 %k.0, 1
+; CHECK-NEXT:   br label %for.cond6
+
+; CHECK:      for.inc24:                                        ; preds = %land.rhs, %for.cond6, %for.body
+; CHECK-NEXT:   %ptr0.3 = phi i8* [ %ptr0.1, %land.rhs ], [ %ptr0.1, %for.cond6 ], [ null, %for.body ]
+; CHECK-NEXT:   %inc25 = add i8 %i.0, 1
+; CHECK-NEXT:   br label %for.cond
+
+; CHECK:      for.end26:                                        ; preds = %for.cond
+; CHECK-NEXT:   ret i8* %ptr0.0
+; CHECK-NEXT: }
+
+; CHECK:      ; Function Attrs: nounwind optsize ssp
+; CHECK-NEXT: define i8* @foo_b(%struct.b_type*) #1 {
+; CHECK-NEXT:   %2 = bitcast %struct.b_type* %0 to %struct.a_type*
+; CHECK-NEXT:   %3 = tail call i8* @foo_a(%struct.a_type* %2)
+; CHECK-NEXT:   ret i8* %3
+; CHECK-NEXT: }
+
+; CHECK:      attributes #0 = { optsize "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+; CHECK-NEXT: attributes #1 = { nounwind optsize ssp "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+; CHECK-NEXT: attributes #2 = { optsize }
+
+target datalayout = "e-m:e-p:32:32-i1:32-i64:64-a:0-v32:32-n16:32"
+
+%struct.a_type = type { i32, i32, i32, i8*, [2 x i8*] }
+%struct.b_type = type { i32, i32, i32, i8*, [2 x i8*], i32 }
+
+; Function Attrs: optsize
+declare i8* @bar4(i32) #1
+declare i8* @bar0(i32) #1
+declare i32 @bar1(i8*) #1
+declare i32 @bar2(i8*, i8*) #1
+declare i32 @bar3(i8*, i8*) #1
+
+; Function Attrs: nounwind optsize ssp
+define i8* @foo_a(%struct.a_type* %arg0) #0 {
+entry:
+  br label %for.cond
+
+for.cond:                                         ; preds = %for.inc24, %entry
+  %i.0 = phi i8 [ 0, %entry ], [ %inc25, %for.inc24 ]
+  %ptr0.0 = phi i8* [ null, %entry ], [ %ptr0.3, %for.inc24 ]
+  %conv = zext i8 %i.0 to i32
+  %cmp = icmp slt i32 %conv, 16
+  br i1 %cmp, label %for.body, label %for.end26
+
+for.body:                                         ; preds = %for.cond
+  %call = call i8* @bar4(i32 %conv) #2
+  %call4 = call i8* @bar0(i32 %conv) #2
+  %0 = bitcast i8* %call4 to %struct.a_type*
+  %call5 = call i32 @bar1(i8* %call) #2
+  %tobool = icmp ne i32 %call5, 0
+  br i1 %tobool, label %for.cond6, label %for.inc24
+
+for.cond6:                                        ; preds = %for.body, %for.inc
+  %k.0 = phi i8 [ %inc, %for.inc ], [ 0, %for.body ]
+  %ptr0.1 = phi i8* [ %ptr0.2, %for.inc ], [ %call, %for.body ]
+  %idxprom = zext i8 %k.0 to i32
+  %field4 = getelementptr inbounds %struct.a_type, %struct.a_type* %arg0, i32 0, i32 4
+  %arrayidx = getelementptr inbounds [2 x i8*], [2 x i8*]* %field4, i32 0, i32 %idxprom
+  %1 = load i8*, i8** %arrayidx, align 4
+  %cmp7 = icmp ne i8* %1, null
+  br i1 %cmp7, label %land.rhs, label %for.inc24
+
+land.rhs:                                         ; preds = %for.cond6
+  %field410 = getelementptr inbounds %struct.a_type, %struct.a_type* %0, i32 0, i32 4
+  %arrayidx11 = getelementptr inbounds [2 x i8*], [2 x i8*]* %field410, i32 0, i32 %idxprom
+  %2 = load i8*, i8** %arrayidx11, align 4
+  %cmp12 = icmp ne i8* %2, null
+  br i1 %cmp12, label %for.body14, label %for.inc24
+
+for.body14:                                       ; preds = %land.rhs
+  %3 = bitcast %struct.a_type* %0 to i8*
+  %4 = bitcast %struct.a_type* %arg0 to i8*
+  %call15 = call i32 @bar2(i8* %3, i8* %4) #2
+  %tobool16 = icmp ne i32 %call15, 0
+  br i1 %tobool16, label %if.then17, label %for.inc
+
+if.then17:                                        ; preds = %for.body14
+  %call18 = call i32 @bar3(i8* %3, i8* %4) #2
+  %tobool19 = icmp ne i32 %call18, 0
+  br i1 %tobool19, label %if.then20, label %for.inc
+
+if.then20:                                        ; preds = %if.then17
+  br label %for.inc
+
+for.inc:                                          ; preds = %for.body14, %if.then17, %if.then20
+  %ptr0.2 = phi i8* [ null, %if.then20 ], [ %ptr0.1, %if.then17 ], [ null, %for.body14 ]
+  %inc = add i8 %k.0, 1
+  br label %for.cond6
+
+for.inc24:                                        ; preds = %for.body, %for.cond6, %land.rhs
+  %ptr0.3 = phi i8* [ %ptr0.1, %land.rhs ], [ %ptr0.1, %for.cond6 ], [ null, %for.body ]
+  %inc25 = add i8 %i.0, 1
+  br label %for.cond
+
+for.end26:                                        ; preds = %for.cond
+  ret i8* %ptr0.0
+}
+
+; Function Attrs: nounwind optsize ssp
+define i8* @foo_b(%struct.b_type* %arg0) #0 {
+entry:
+  br label %for.cond
+
+for.cond:                                         ; preds = %for.inc24, %entry
+  %i.0 = phi i8 [ 0, %entry ], [ %inc25, %for.inc24 ]
+  %ptr0.0 = phi i8* [ null, %entry ], [ %ptr0.3, %for.inc24 ]
+  %conv = zext i8 %i.0 to i32
+  %cmp = icmp slt i32 %conv, 16
+  br i1 %cmp, label %for.body, label %for.end26
+
+for.body:                                         ; preds = %for.cond
+  %call = call i8* @bar4(i32 %conv) #2
+  %call4 = call i8* @bar0(i32 %conv) #2
+  %0 = bitcast i8* %call4 to %struct.b_type*
+  %call5 = call i32 @bar1(i8* %call) #2
+  %tobool = icmp ne i32 %call5, 0
+  br i1 %tobool, label %for.cond6, label %for.inc24
+
+for.cond6:                                        ; preds = %for.body, %for.inc
+  %k.0 = phi i8 [ %inc, %for.inc ], [ 0, %for.body ]
+  %ptr0.1 = phi i8* [ %ptr0.2, %for.inc ], [ %call, %for.body ]
+  %idxprom = zext i8 %k.0 to i32
+  %field4 = getelementptr inbounds %struct.b_type, %struct.b_type* %arg0, i32 0, i32 4
+  %arrayidx = getelementptr inbounds [2 x i8*], [2 x i8*]* %field4, i32 0, i32 %idxprom
+  %1 = load i8*, i8** %arrayidx, align 4
+  %cmp7 = icmp ne i8* %1, null
+  br i1 %cmp7, label %land.rhs, label %for.inc24
+
+land.rhs:                                         ; preds = %for.cond6
+  %field410 = getelementptr inbounds %struct.b_type, %struct.b_type* %0, i32 0, i32 4
+  %arrayidx11 = getelementptr inbounds [2 x i8*], [2 x i8*]* %field410, i32 0, i32 %idxprom
+  %2 = load i8*, i8** %arrayidx11, align 4
+  %cmp12 = icmp ne i8* %2, null
+  br i1 %cmp12, label %for.body14, label %for.inc24
+
+for.body14:                                       ; preds = %land.rhs
+  %3 = bitcast %struct.b_type* %0 to i8*
+  %4 = bitcast %struct.b_type* %arg0 to i8*
+  %call15 = call i32 @bar2(i8* %3, i8* %4) #2
+  %tobool16 = icmp ne i32 %call15, 0
+  br i1 %tobool16, label %if.then17, label %for.inc
+
+if.then17:                                        ; preds = %for.body14
+  %call18 = call i32 @bar3(i8* %3, i8* %4) #2
+  %tobool19 = icmp ne i32 %call18, 0
+  br i1 %tobool19, label %if.then20, label %for.inc
+
+if.then20:                                        ; preds = %if.then17
+  br label %for.inc
+
+for.inc:                                          ; preds = %for.body14, %if.then17, %if.then20
+  %ptr0.2 = phi i8* [ null, %if.then20 ], [ %ptr0.1, %if.then17 ], [ null, %for.body14 ]
+  %inc = add i8 %k.0, 1
+  br label %for.cond6
+
+for.inc24:                                        ; preds = %for.body, %for.cond6, %land.rhs
+  %ptr0.3 = phi i8* [ %ptr0.1, %land.rhs ], [ %ptr0.1, %for.cond6 ], [ null, %for.body ]
+  %inc25 = add i8 %i.0, 1
+  br label %for.cond
+
+for.end26:                                        ; preds = %for.cond
+  ret i8* %ptr0.0
+}
+
+attributes #0 = { nounwind optsize ssp "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { optsize "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #2 = { optsize }
+

llvm/test/Transforms/MergeSimilarFunc/merge-noinline.ll

+; RUN: opt -S -mergesimilarfunc -mergesimilarfunc-level=all < %s | FileCheck %s
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+
+declare void @stuff()
+
+; CHECK-LABEL: @f0(
+define void @f0(i64 %p0) {
+entry:
+  call void @stuff()
+  call void @stuff()
+  call void @stuff()
+  ret void
+}
+
+; CHECK-LABEL: @f1(
+; CHECK: call void @f0{{.*}} #[[ATTR:[0-9]+]]
+; CHECK: attributes #[[ATTR]] = { {{.*}}noinline
+define void @f1(i64 %p0) {
+entry:
+  call void @stuff()
+  call void @stuff()
+  call void @stuff()
+  ret void
+}
+

llvm/test/Transforms/MergeSimilarFunc/merge-ret.ll

+; Check that the ret instructions are merged correctly. A bug caused
+; an incorrect merge and a verifier failure for this input.
+;
+; RUN: opt -S -mergesimilarfunc < %s | FileCheck %s
+;
+; CHECK-LABEL: define internal %0* @LLVMGetReturnType__merged
+; CHECK: phi %0* [
+; CHECK-NEXT: ret %0*
+; CHECK-NEXT: }
+;
+
+target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"
+
+%0 = type opaque
+%1 = type { %2*, i32, i32, %1** }
+%2 = type { %3* }
+%3 = type opaque
+%4 = type { %1 }
+%5 = type opaque
+%6 = type { i32 (...)**, %1*, %7*, i8, i8, i16, i32 }
+%7 = type { %6*, %7*, %8 }
+%8 = type { i32 }
+%9 = type { %10, %1*, i32, %12, i32, i8, %18* }
+%10 = type { %11 }
+%11 = type { %6 }
+%12 = type { %13 }
+%13 = type { %14 }
+%14 = type { %15 }
+%15 = type { %16 }
+%16 = type { %17 }
+%17 = type { i32, i32, i8* }
+%18 = type <{ %2*, %19, %30, %70, %79, %87, %12, %93*, %94, %98, %12, %12, %12, i8*, %102, i8, [3 x i8] }>
+%19 = type { %20 }
+%20 = type { %21, %26* }
+%21 = type { %22 }
+%22 = type { %23 }
+%23 = type { %24 }
+%24 = type { %25, %26* }
+%25 = type { %26* }
+%26 = type <{ %27, [3 x i8], %24, i8, [3 x i8] }>
+%27 = type <{ %9, %12, %28*, %12, %12, i8 }>
+%28 = type <{ %29*, i8, [3 x i8] }>
+%29 = type opaque
+%30 = type { %31 }
+%31 = type { %32, %37* }
+%32 = type { %33 }
+%33 = type { %34 }
+%34 = type { %35 }
+%35 = type { %36, %37* }
+%36 = type { %37* }
+%37 = type { %27, %35, %38, %60, %93*, %68, %4* }
+%38 = type { %39 }
+%39 = type { %40, %44* }
+%40 = type { %41 }
+%41 = type { %42 }
+%42 = type { %43 }
+%43 = type { %44* }
+%44 = type { %6, %45, i32, %47, %37* }
+%45 = type { %46 }
+%46 = type { %43, %44* }
+%47 = type { %48 }
+%48 = type { %49, %53* }
+%49 = type { %50 }
+%50 = type { %51 }
+%51 = type { %52 }
+%52 = type { %53* }
+%53 = type { %11, %54, %44*, %56 }
+%54 = type { %55 }
+%55 = type { %52, %53* }
+%56 = type { %57 }
+%57 = type { %58 }
+%58 = type { %59* }
+%59 = type { i8, i8, i16, i32 }
+%60 = type { %61 }
+%61 = type { %62, %66* }
+%62 = type { %63 }
+%63 = type { %64 }
+%64 = type { %65 }
+%65 = type { %66* }
+%66 = type { %6, %67, %37* }
+%67 = type { %65, %66* }
+%68 = type { %69* }
+%69 = type opaque
+%70 = type { %71 }
+%71 = type { %72, %77* }
+%72 = type { %73 }
+%73 = type { %74 }
+%74 = type { %75 }
+%75 = type { %76, %77* }
+%76 = type { %77* }
+%77 = type { %78, %75 }
+%78 = type { %9 }
+%79 = type { %80 }
+%80 = type { %81, %86* }
+%81 = type { %82 }
+%82 = type { %83 }
+%83 = type { %84 }
+%84 = type { %85, %86* }
+%85 = type { %86* }
+%86 = type { %78, %84 }
+%87 = type { %88 }
+%88 = type { %89, %92* }
+%89 = type { %90 }
+%90 = type { %91, %92* }
+%91 = type { %92* }
+%92 = type { %90, %12, %18*, i8* }
+%93 = type opaque
+%94 = type <{ %95, %97, [3 x i8] }>
+%95 = type { %96**, i32, i32, i32, i32 }
+%96 = type { i32 }
+%97 = type { i8 }
+%98 = type { %99 }
+%99 = type { %100 }
+%100 = type { %101* }
+%101 = type opaque
+%102 = type { i8, i32, i8, [3 x i8], %103, %111, %12, %118, i8* }
+%103 = type { %104, %110 }
+%104 = type { %105 }
+%105 = type { %106 }
+%106 = type <{ %107, %108 }>
+%107 = type { i8*, i8*, i8* }
+%108 = type { %109 }
+%109 = type { [1 x i8] }
+%110 = type { [7 x %108] }
+%111 = type { %112, %117 }
+%112 = type { %113 }
+%113 = type { %114 }
+%114 = type { %107, %115 }
+%115 = type { %116 }
+%116 = type { [8 x i8] }
+%117 = type { [15 x %115] }
+%118 = type { %119, %124 }
+%119 = type { %120 }
+%120 = type { %121 }
+%121 = type { %107, %122 }
+%122 = type { %123 }
+%123 = type { [16 x i8] }
+%124 = type { [7 x %122] }
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start(i64, i8* nocapture) #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end(i64, i8* nocapture) #0
+
+; Function Attrs: optsize
+define %0* @LLVMGetReturnType(%0*) #1 {
+  %2 = alloca %1*, align 4
+  %3 = bitcast %1** %2 to i8*
+  call void @llvm.lifetime.start(i64 4, i8* %3)
+  %4 = bitcast %1** %2 to %0**
+  store %0* %0, %0** %4, align 4
+  %5 = call zeroext i1 @_ZN4llvm13isa_impl_wrapINS_12FunctionTypeEKPNS_4TypeEPKS2_E4doitERS4_(%1** nonnull dereferenceable(4) %2) #3
+  %6 = bitcast %0* %0 to %4*
+  br i1 %5, label %8, label %7
+
+; <label>:7:                                      ; preds = %1
+  tail call void @__assert_fail.5.0.1.2_i32_237.3() #4
+  br label %8
+
+; <label>:8:                                      ; preds = %7, %1
+  call void @llvm.lifetime.end(i64 4, i8* %3)
+  %9 = getelementptr inbounds %4, %4* %6, i32 0, i32 0, i32 3
+  %10 = bitcast %1*** %9 to %0***
+  %11 = load %0**, %0*** %10, align 4
+  %12 = load %0*, %0** %11, align 4
+  ret %0* %12
+}
+
+; Function Attrs: optsize
+declare void @__assert_fail.5.0.1.2_i32_237.3() #2
+
+; Function Attrs: optsize
+define i32 @LLVMHasUnnamedAddr(%5*) #1 {
+  %2 = alloca %6*, align 4
+  %3 = bitcast %6** %2 to i8*
+  call void @llvm.lifetime.start(i64 4, i8* %3)
+  %4 = bitcast %6** %2 to %5**
+  store %5* %0, %5** %4, align 4
+  %5 = call zeroext i1 @_ZN4llvm13isa_impl_wrapINS_11GlobalValueEKPNS_5ValueEPKS2_E4doitERS4_(%6** nonnull dereferenceable(4) %2) #3
+  %6 = bitcast %5* %0 to %9*
+  br i1 %5, label %8, label %7
+
+; <label>:7:                                      ; preds = %1
+  tail call void @__assert_fail.5.0.1.2_i32_237.3() #4
+  br label %8
+
+; <label>:8:                                      ; preds = %7, %1
+  call void @llvm.lifetime.end(i64 4, i8* %3)
+  %9 = getelementptr inbounds %9, %9* %6, i32 0, i32 2
+  %10 = load i32, i32* %9, align 4
+  %11 = lshr i32 %10, 6
+  %12 = and i32 %11, 1
+  ret i32 %12
+}
+
+; Function Attrs: optsize
+declare zeroext i1 @_ZN4llvm13isa_impl_wrapINS_12FunctionTypeEKPNS_4TypeEPKS2_E4doitERS4_(%1** nocapture readonly dereferenceable(4)) #1 align 2
+
+; Function Attrs: optsize
+declare zeroext i1 @_ZN4llvm13isa_impl_wrapINS_11GlobalValueEKPNS_5ValueEPKS2_E4doitERS4_(%6** nocapture readonly dereferenceable(4)) #1 align 2
+
+attributes #0 = { argmemonly nounwind }
+attributes #1 = { optsize }
+attributes #2 = { optsize }
+attributes #3 = { optsize }
+attributes #4 = { noinline optsize }

llvm/test/Transforms/MergeSimilarFunc/merge-ret2.ll

+; Check that two ret instructions with the same return type
+; are generated in the merged function. A bug caused a mismatch
+; in return types here and a verifier failure.
+;
+; RUN: opt -S -mergesimilarfunc < %s | FileCheck %s
+;
+; CHECK: @mrg1__merged
+; CHECK: switch i32 %__merge_arg
+; CHECK: ret %0*
+; CHECK: ret %0*
+
+target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"
+
+%0 = type { {}*, %0* (%8*)*, %0* (%13*)*, %0* (%39*)*, %0* (%14*)*, %0* (%14*)*, %0* (%15*)*, %0* (%16*)*, %0* ()*, %0* (%17*)*, %0* (%22*)*, %0* (%28*)*, %0* (%32*)*, %0* (%37*)*, %0* (%37*)*, %0* (%37*)*, %0* ()*, %0* (%33*)*, %0* (%36*)*, %0* (%37*)*, %0* (%38*)*, %0* ()*, %0* (%39*)*, %0* ()*, %0* ()*, %0* (%40*)*, %0* ()*, %0* ()*, %0* ()*, %0* ()*, i8 ()*, void ()* }
+%1 = type { %2, i32, i8, i8, i64, i8, i8, i8, %3, %3, i8, i8 }
+%2 = type { i8, i8, i8, i8 }
+%3 = type { i32, i32, [10 x %4], %7 }
+%4 = type { %5, i8, i32, i8, %5 }
+%5 = type { %6 }
+%6 = type { i64, i64, i64, i64 }
+%7 = type { i8, i8 }
+%8 = type { %2, i32, i8, i8, i8, i8, i8, i64, %9, %3, i8, i8, i8, %12, i32, i8 }
+%9 = type { %10 }
+%10 = type { i8, i8, %11, i32, i8 }
+%11 = type { [3 x i8] }
+%12 = type { i8, i8, i8 }
+%13 = type { %2, i32, i8, i32, i8, i8, i8 }
+%14 = type { %2, i32, i8, i8, i8 }
+%15 = type { %2, i8, i8, i8 }
+%16 = type { %2, i8, %5, i8, %11, i16, i8 }
+%17 = type { %2, %18, i8 }
+%18 = type { i8, i8, i8, i8, i8, i8, i8, i8, %11, %19, %19, %20, i32, %21, i8, i16, i8 }
+%19 = type { i8, i8, i8, [255 x i8], i8 }
+%20 = type { i8, i8, i8, i8, i8, i8, i32 }
+%21 = type { i8, [3 x i8] }
+%22 = type { %2, i8, i8, %23, i8, i8, i8 }
+%23 = type { %24 }
+%24 = type { i32, [40 x %25] }
+%25 = type { %11, i8, i8, i8, i8, i8, i32, %26, i32 }
+%26 = type { i32, %27, i8 }
+%27 = type { i8, [48 x i8] }
+%28 = type { %2, i8, %29, i8, %24, i8, %31, i8, i8, i8, i8, i16, i8 }
+%29 = type { i8, %11, i8, i8, i16, i8, i8, i8, %5, i16, i16, i8, %26, i8, i8, i8, i32, i8, i8, i8, i8, %30, i8, i8, %11, i8 }
+%30 = type { i8, i8, i8, i8, i8 }
+%31 = type { i32, [40 x %11] }
+%32 = type { %2, i8, %29, i8, i8, %31, i8, i16, i8 }
+%33 = type { %2, %34, i8 }
+%34 = type { i8, [16 x %35] }
+%35 = type { i8, i8, [41 x i8] }
+%36 = type { %2, i32, i8 }
+%37 = type { %2, i8 }
+%38 = type { %2, i8, i8, %11, i16, i8, i32, i8 }
+%39 = type { %2, i32, i8, i8 }
+%40 = type { %2, i8, %11, i32, i8, i8, i8 }
+%41 = type { {}*, %41* (%8*)*, %41* (%13*)*, %41* (%39*)*, %41* (%14*)*, %41* (%14*)*, %41* (%15*)*, %41* (%16*)*, %41* ()*, %41* (%17*)*, %41* (%22*)*, %41* (%28*)*, %41* (%32*)*, %41* (%37*)*, %41* (%37*)*, %41* (%37*)*, %41* ()*, %41* (%33*)*, %41* (%36*)*, %41* (%37*)*, %41* (%38*)*, %41* ()*, %41* (%39*)*, %41* ()*, %41* ()*, %41* (%40*)*, %41* ()*, %41* ()*, %41* ()*, %41* ()*, i8 ()*, void ()* }
+
+@xx = external global i8, align 1
+@yy = external global { %0* (%1*)*, %0* (%8*)*, %0* (%13*)*, %0* (%39*)*, %0* (%14*)*, %0* (%14*)*, %0* (%15*)*, %0* (%16*)*, %0* ()*, %0* (%17*)*, %0* (%22*)*, %0* (%28*)*, %0* (%32*)*, %0* (%37*)*, %0* (%37*)*, %0* (%37*)*, %0* ()*, %0* (%33*)*, %0* (%36*)*, %0* (%37*)*, %0* (%38*)*, %0* ()*, %0* (%39*)*, %0* ()*, %0* ()*, %0* (%40*)*, %0* ()*, %0* ()*, %0* ()*, %0* ()*, i8 ()*, void ()* }, align 4
+@zz = external global { %41* (%1*)*, %41* (%8*)*, %41* (%13*)*, %41* (%39*)*, %41* (%14*)*, %41* (%14*)*, %41* (%15*)*, %41* (%16*)*, %41* ()*, %41* (%17*)*, %41* (%22*)*, %41* (%28*)*, %41* (%32*)*, %41* (%37*)*, %41* (%37*)*, %41* (%37*)*, %41* ()*, %41* (%33*)*, %41* (%36*)*, %41* (%37*)*, %41* (%38*)*, %41* ()*, %41* (%39*)*, %41* ()*, %41* ()*, %41* (%40*)*, %41* ()*, %41* ()*, %41* ()*, %41* ()*, i8 ()*, void ()* }, align 4
+
+declare void @llvm.lifetime.start(i64, i8* nocapture)
+declare void @llvm.lifetime.end(i64, i8* nocapture)
+declare void @f1()
+declare void @f1a(i8 zeroext)
+declare signext i8 @f2()
+declare void @f3(%26* noalias nocapture sret)
+
+define internal %0* @mrg1() optsize {
+  %1 = alloca %26, align 4
+  %2 = bitcast %26* %1 to i8*
+  call void @llvm.lifetime.start(i64 56, i8* %2)
+  call void @f3(%26* nonnull sret %1)
+  call void @f1a(i8 zeroext 1)
+  %3 = load i8, i8* @xx, align 1
+  %4 = icmp eq i8 %3, 1
+  br i1 %4, label %5, label %18
+
+; <label>:5:                                      ; preds = %0
+  %6 = call signext i8 @f2()
+  %7 = icmp eq i8 %6, 2
+  br i1 %7, label %8, label %18
+
+; <label>:8:                                      ; preds = %5
+  %9 = getelementptr inbounds %26, %26* %1, i32 0, i32 0
+  %10 = load i32, i32* %9, align 4
+  %11 = icmp eq i32 %10, -1
+  br i1 %11, label %16, label %12
+
+; <label>:12:                                     ; preds = %8
+  %13 = getelementptr inbounds %26, %26* %1, i32 0, i32 2
+  %14 = load i8, i8* %13, align 1
+  %15 = icmp eq i8 %14, 1
+  br i1 %15, label %16, label %17
+
+; <label>:16:                                     ; preds = %12, %8
+  call void @f1()
+  br label %17
+
+; <label>:17:                                     ; preds = %16, %12
+  store i8 0, i8* @xx, align 1
+  br label %18
+
+; <label>:18:                                     ; preds = %17, %5, %0
+  call void @llvm.lifetime.end(i64 56, i8* %2)
+  ret %0* bitcast ({ %0* (%1*)*, %0* (%8*)*, %0* (%13*)*, %0* (%39*)*, %0* (%14*)*, %0* (%14*)*, %0* (%15*)*, %0* (%16*)*, %0* ()*, %0* (%17*)*, %0* (%22*)*, %0* (%28*)*, %0* (%32*)*, %0* (%37*)*, %0* (%37*)*, %0* (%37*)*, %0* ()*, %0* (%33*)*, %0* (%36*)*, %0* (%37*)*, %0* (%38*)*, %0* ()*, %0* (%39*)*, %0* ()*, %0* ()*, %0* (%40*)*, %0* ()*, %0* ()*, %0* ()*, %0* ()*, i8 ()*, void ()* }* @yy to %0*)
+}
+
+define internal %41* @mrg2() optsize {
+  %1 = alloca %26, align 4
+  %2 = bitcast %26* %1 to i8*
+  call void @llvm.lifetime.start(i64 56, i8* %2)
+  call void @f3(%26* nonnull sret %1)
+  call void @f1a(i8 zeroext 1)
+  %3 = load i8, i8* @xx, align 1
+  %4 = icmp eq i8 %3, 1
+  br i1 %4, label %5, label %18
+
+; <label>:5:                                      ; preds = %0
+  %6 = call signext i8 @f2()
+  %7 = icmp eq i8 %6, 2
+  br i1 %7, label %8, label %18
+
+; <label>:8:                                      ; preds = %5
+  %9 = getelementptr inbounds %26, %26* %1, i32 0, i32 0
+  %10 = load i32, i32* %9, align 4
+  %11 = icmp eq i32 %10, -1
+  br i1 %11, label %16, label %12
+
+; <label>:12:                                     ; preds = %8
+  %13 = getelementptr inbounds %26, %26* %1, i32 0, i32 2
+  %14 = load i8, i8* %13, align 1
+  %15 = icmp eq i8 %14, 1
+  br i1 %15, label %16, label %17
+
+; <label>:16:                                     ; preds = %12, %8
+  call void @f1()
+  br label %17
+
+; <label>:17:                                     ; preds = %16, %12
+  store i8 0, i8* @xx, align 1
+  br label %18
+
+; <label>:18:                                     ; preds = %17, %5, %0
+  call void @llvm.lifetime.end(i64 56, i8* %2)
+  ret %41* bitcast ({ %41* (%1*)*, %41* (%8*)*, %41* (%13*)*, %41* (%39*)*, %41* (%14*)*, %41* (%14*)*, %41* (%15*)*, %41* (%16*)*, %41* ()*, %41* (%17*)*, %41* (%22*)*, %41* (%28*)*, %41* (%32*)*, %41* (%37*)*, %41* (%37*)*, %41* (%37*)*, %41* ()*, %41* (%33*)*, %41* (%36*)*, %41* (%37*)*, %41* (%38*)*, %41* ()*, %41* (%39*)*, %41* ()*, %41* ()*, %41* (%40*)*, %41* ()*, %41* ()*, %41* ()*, %41* ()*, i8 ()*, void ()* }* @zz to %41*)
+}
+

llvm/test/Transforms/MergeSimilarFunc/multi-merge.ll

+; RUN: opt -mergesimilarfunc -mergesimilarfunc-level=all < %s -S | FileCheck %s
+
+; CHECK: define i32 @A
+; CHECK: define internal i32 @A__merged
+; CHECK: define i32 @B
+; CHECK: call i32 @A__merged
+; CHECK: define i32 @C
+; CHECK: call i32 @A__merged
+; CHECK: define i32 @D
+; CHECK: call i32 @A__merged
+
+define i32 @A(i32* %a, i32 %n, i32 %c) nounwind {
+entry:
+  %0 = bitcast i32* %a to i8*
+  %mul = mul i32 %n, 4
+  %call = tail call i8* @memset1(i8* %0, i32 %c, i32 %mul) nounwind
+  %cmp7 = icmp sgt i32 %n, 0
+  br i1 %cmp7, label %for.body, label %for.end
+
+for.body:
+  %arrayidx.phi = phi i32* [ %arrayidx.inc, %for.body ], [ %a, %entry ]
+  %i.09 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+  %sum.08 = phi i32 [ %add, %for.body ], [ 0, %entry ]
+  %1 = load i32, i32* %arrayidx.phi, align 4
+  %add = add nsw i32 %1, %sum.08
+  %inc = add nsw i32 %i.09, 1
+  %exitcond = icmp eq i32 %inc, %n
+  %arrayidx.inc = getelementptr i32, i32* %arrayidx.phi, i32 1
+  br i1 %exitcond, label %for.cond.for.end_crit_edge, label %for.body
+
+for.cond.for.end_crit_edge:
+  %phitmp = add i32 %add, 1
+  br label %for.end
+
+for.end:
+  %sum.0.lcssa = phi i32 [ %phitmp, %for.cond.for.end_crit_edge ], [ 1, %entry ]
+  ret i32 %sum.0.lcssa
+}
+
+declare i8* @memset1(i8*, i32, i32)
+
+define i32 @B(i32* %a, i32 %n, i32 %c) nounwind {
+entry:
+  %0 = bitcast i32* %a to i8*
+  %mul = mul i32 %n, 4
+  %call = tail call i8* @memset1(i8* %0, i32 %c, i32 %mul) nounwind
+  %cmp7 = icmp sgt i32 %n, 0
+  br i1 %cmp7, label %for.body, label %for.end
+
+for.body:
+  %arrayidx.phi = phi i32* [ %arrayidx.inc, %for.body ], [ %a, %entry ]
+  %i.09 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+  %sum.08 = phi i32 [ %add, %for.body ], [ 0, %entry ]
+  %1 = load i32, i32* %arrayidx.phi, align 4
+  %add = add nsw i32 %1, %sum.08
+  %inc = add nsw i32 %i.09, 1
+  %exitcond = icmp eq i32 %inc, %n
+  %arrayidx.inc = getelementptr i32, i32* %arrayidx.phi, i32 1
+  br i1 %exitcond, label %for.cond.for.end_crit_edge, label %for.body
+
+for.cond.for.end_crit_edge:
+  %phitmp = add i32 %add, 2
+  br label %for.end
+
+for.end:
+  %sum.0.lcssa = phi i32 [ %phitmp, %for.cond.for.end_crit_edge ], [ 2, %entry ]
+  ret i32 %sum.0.lcssa
+}
+
+define i32 @C(i32* %a, i32 %n, i32 %c) nounwind {
+entry:
+  %0 = bitcast i32* %a to i8*
+  %mul = mul i32 %n, 4
+  %call = tail call i8* @memset1(i8* %0, i32 %c, i32 %mul) nounwind
+  %cmp7 = icmp sgt i32 %n, 0
+  br i1 %cmp7, label %for.body, label %for.end
+
+for.body:
+  %arrayidx.phi = phi i32* [ %arrayidx.inc, %for.body ], [ %a, %entry ]
+  %i.09 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+  %sum.08 = phi i32 [ %add, %for.body ], [ 1, %entry ]
+  %1 = load i32, i32* %arrayidx.phi, align 4
+  %add = add nsw i32 %1, %sum.08
+  %inc = add nsw i32 %i.09, 1
+  %exitcond = icmp eq i32 %inc, %n
+  %arrayidx.inc = getelementptr i32, i32* %arrayidx.phi, i32 1
+  br i1 %exitcond, label %for.cond.for.end_crit_edge, label %for.body
+
+for.cond.for.end_crit_edge:
+  %phitmp = add i32 %add, 3
+  br label %for.end
+
+for.end:
+  %sum.0.lcssa = phi i32 [ %phitmp, %for.cond.for.end_crit_edge ], [ 4, %entry ]
+  ret i32 %sum.0.lcssa
+}
+
+define i32 @D(i32* %a, i32 %n, i32 %c) nounwind {
+entry:
+  %0 = bitcast i32* %a to i8*
+  %mul = mul i32 %n, 4
+  %call = tail call i8* @memset1(i8* %0, i32 %c, i32 %mul) nounwind
+  %cmp7 = icmp sgt i32 %n, 0
+  br i1 %cmp7, label %for.body, label %for.end
+
+for.body:
+  %arrayidx.phi = phi i32* [ %arrayidx.inc, %for.body ], [ %a, %entry ]
+  %i.09 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+  %sum.08 = phi i32 [ %add, %for.body ], [ 1, %entry ]
+  %1 = load i32, i32* %arrayidx.phi, align 4
+  %add = add nsw i32 %1, %sum.08
+  %inc = add nsw i32 %i.09, 1
+  %exitcond = icmp eq i32 %inc, %n
+  %arrayidx.inc = getelementptr i32, i32* %arrayidx.phi, i32 1
+  br i1 %exitcond, label %for.cond.for.end_crit_edge, label %for.body
+
+for.cond.for.end_crit_edge:
+  %phitmp = add i32 %add, 4
+  br label %for.end
+
+for.end:
+  %sum.0.lcssa = phi i32 [ %phitmp, %for.cond.for.end_crit_edge ], [ 5, %entry ]
+  ret i32 %sum.0.lcssa
+}

llvm/test/Transforms/MergeSimilarFunc/self-ref.ll

+; RUN: opt -mergesimilarfunc -mergesimilarfunc-level=all -mergesimilarfunc-min-insts=0 -mergesimilarfunc-diff-min-insts=0 < %s -S | FileCheck %s
+
+; This test checks if we can tell the difference between the two corresponding
+; calls to foobar.  Each call has a  reference to the containing function as the
+; only argument to foobar.
+; If we cannot tell the difference between these two calls, then the functions
+; foo and bar below will be considered identical and bar will simply be patched
+; up to call foo. However, these functions are not identical as the calls to foobar
+; are themselves different. So, these functions are similar, but not identical.
+; bar should be patched up to call foo__merged which has code to deal with the
+; two different calls to foobar.
+
+; CHECK: define i32 @bar(i32 %a) #0 {
+; CHECK: %1 = tail call i32 @foo__merged(i32 %a, i32 1)
+define i32 @foo(i32 %a) nounwind {
+entry:
+  %a.addr = alloca i32, align 4
+  %b = alloca i32, align 4
+  store i32 %a, i32* %a.addr, align 4
+  %call = call i32 @foobar(i32 (i32)* @foo)
+  store i32 %call, i32* %b, align 4
+  %0 = load i32, i32* %a.addr, align 4
+  %1 = load i32, i32* %b, align 4
+  %add = add nsw i32 %0, %1
+  ret i32 %add
+}
+
+declare i32 @foobar(i32 (i32)*)
+
+define i32 @bar(i32 %a) nounwind {
+entry:
+  %a.addr = alloca i32, align 4
+  %b = alloca i32, align 4
+  store i32 %a, i32* %a.addr, align 4
+  %call = call i32 @foobar(i32 (i32)* @bar)
+  store i32 %call, i32* %b, align 4
+  %0 = load i32, i32* %a.addr, align 4
+  %1 = load i32, i32* %b, align 4
+  %add = add nsw i32 %0, %1
+  ret i32 %add
+}

llvm/test/Transforms/MergeSimilarFunc/strong-weak.ll

+; RUN: opt -mergesimilarfunc -mergesimilarfunc-level=all -mergesimilarfunc-min-insts=0 -mergesimilarfunc-diff-min-insts=1 < %s -S
+; REQUIRES: asserts
+; CHECK-NOT: __merged
+define weak i32 @bar(i32 %a) nounwind {
+entry:
+  %a.addr = alloca i32, align 4
+  %b = alloca i32, align 4
+  %d = alloca i32, align 4
+  store i32 %a, i32* %a.addr, align 4
+  %0 = load i32, i32* %a.addr, align 4
+  %1 = load i32, i32* %d, align 4
+  %add = add nsw i32 %0, %1
+  %call = call i32 bitcast (i32 (...)* @foo to i32 ()*)()
+  %add1 = add nsw i32 %add, %call
+  store i32 %add1, i32* %b, align 4
+  %2 = load i32, i32* %b, align 4
+  ret i32 %2
+}
+
+declare i32 @foo(...)
+
+define i32 @barred(i32 %a) nounwind {
+entry:
+  %a.addr = alloca i32, align 4
+  %b = alloca i32, align 4
+  %d = alloca i32, align 4
+  store i32 %a, i32* %a.addr, align 4
+  %0 = load i32, i32* %a.addr, align 4
+  %1 = load i32, i32* %d, align 4
+  %add = add nsw i32 %0, %1
+  %call = call i32 bitcast (i32 (...)* @boohoo to i32 ()*)()
+  %add1 = add nsw i32 %add, %call
+  store i32 %add1, i32* %b, align 4
+  %2 = load i32, i32* %b, align 4
+  ret i32 %2
+}
+
+declare i32 @boohoo(...)

llvm/unittests/Transforms/Utils/CloningTest.cpp

TEST_F(CloneInstruction, Attributes) {
 
   Argument *A = &*F1->arg_begin();
   A->addAttr(Attribute::NoCapture);
 
   SmallVector<ReturnInst*, 4> Returns;
   ValueToValueMapTy VMap;
   VMap[A] = UndefValue::get(A->getType());
 
-  CloneFunctionInto(F2, F1, VMap, false, Returns);
+  CloneFunctionInto(F2, F1, VMap, CloneType::NoModuleLevelChanges, Returns);
   EXPECT_FALSE(F2->arg_begin()->hasNoCaptureAttr());
 
   delete F1;
   delete F2;
 }
 
 TEST_F(CloneInstruction, CallingConvention) {
   Type *ArgTy1[] = { Type::getInt32PtrTy(context) };
   FunctionType *FT1 =  FunctionType::get(Type::getVoidTy(context), ArgTy1, false);
 
   Function *F1 = Function::Create(FT1, Function::ExternalLinkage);
   F1->setCallingConv(CallingConv::Cold);
   BasicBlock *BB = BasicBlock::Create(context, "", F1);
   IRBuilder<> Builder(BB);
   Builder.CreateRetVoid();
 
   Function *F2 = Function::Create(FT1, Function::ExternalLinkage);
 
   SmallVector<ReturnInst*, 4> Returns;
   ValueToValueMapTy VMap;
   VMap[&*F1->arg_begin()] = &*F2->arg_begin();
 
-  CloneFunctionInto(F2, F1, VMap, false, Returns);
+  CloneFunctionInto(F2, F1, VMap, CloneType::NoModuleLevelChanges, Returns);
   EXPECT_EQ(CallingConv::Cold, F2->getCallingConv());
 
   delete F1;
   delete F2;
 }
 
 TEST_F(CloneInstruction, DuplicateInstructionsToSplit) {
   Type *ArgTy1[] = {Type::getInt32PtrTy(context)};
TEST(CloneFunction, CloneFunctionToDifferentModule) {
   EXPECT_TRUE(GetDICompileUnitCount(*DeclModule) == 0);
   auto* DeclFunction = DeclModule->getFunction("foo");
   EXPECT_TRUE(DeclFunction != nullptr);
 
   ValueToValueMapTy VMap;
   VMap[ImplFunction] = DeclFunction;
   // No args to map
   SmallVector<ReturnInst*, 8> Returns;
-  CloneFunctionInto(DeclFunction, ImplFunction, VMap, true, Returns);
+  CloneFunctionInto(DeclFunction, ImplFunction, VMap,
+                    CloneType::ModuleLevelChanges, Returns);
 
   EXPECT_FALSE(verifyModule(*ImplModule, &errs()));
   EXPECT_FALSE(verifyModule(*DeclModule, &errs()));
   // DICompileUnit !2 shall be inserted into DeclModule.
   EXPECT_TRUE(GetDICompileUnitCount(*DeclModule) == 1);
 }
 
 class CloneModule : public ::testing::Test {