1 files changed, 638 insertions, 0 deletions
diff --git a/contrib/llvm-project/llvm/lib/Transforms/Instrumentation/MemProfiler.cpp b/contrib/llvm-project/llvm/lib/Transforms/Instrumentation/MemProfiler.cpp
new file mode 100644
index 000000000000..0e6a404a9e0b
--- /dev/null
+++ b/contrib/llvm-project/llvm/lib/Transforms/Instrumentation/MemProfiler.cpp
@@ -0,0 +1,638 @@
+//===- MemProfiler.cpp - memory allocation and access profiler ------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of MemProfiler. Memory accesses are instrumented
+// to increment the access count held in a shadow memory location, or
+// alternatively to call into the runtime. Memory intrinsic calls (memmove,
+// memcpy, memset) are changed to call the memory profiling runtime version
+// instead.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Transforms/Instrumentation/MemProfiler.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/Triple.h"
+#include "llvm/IR/Constant.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instruction.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Type.h"
+#include "llvm/IR/Value.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Pass.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Transforms/Instrumentation.h"
+#include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Transforms/Utils/ModuleUtils.h"
+
+using namespace llvm;
+
+#define DEBUG_TYPE "memprof"
+
+constexpr int LLVM_MEM_PROFILER_VERSION = 1;
+
+// Size of memory mapped to a single shadow location.
+constexpr uint64_t DefaultShadowGranularity = 64;
+
+// Scale from granularity down to shadow size.
+constexpr uint64_t DefaultShadowScale = 3;
+
+constexpr char MemProfModuleCtorName[] = "memprof.module_ctor";
+constexpr uint64_t MemProfCtorAndDtorPriority = 1;
+// On Emscripten, the system needs more than one priorities for constructors.
+constexpr uint64_t MemProfEmscriptenCtorAndDtorPriority = 50;
+constexpr char MemProfInitName[] = "__memprof_init";
+constexpr char MemProfVersionCheckNamePrefix[] =
+    "__memprof_version_mismatch_check_v";
+
+constexpr char MemProfShadowMemoryDynamicAddress[] =
+    "__memprof_shadow_memory_dynamic_address";
+
+constexpr char MemProfFilenameVar[] = "__memprof_profile_filename";
+
+// Command-line flags.
+
+static cl::opt<bool> ClInsertVersionCheck(
+    "memprof-guard-against-version-mismatch",
+    cl::desc("Guard against compiler/runtime version mismatch."), cl::Hidden,
+    cl::init(true));
+
+// This flag may need to be replaced with -f[no-]memprof-reads.
+static cl::opt<bool> ClInstrumentReads("memprof-instrument-reads",
+                                       cl::desc("instrument read instructions"),
+                                       cl::Hidden, cl::init(true));
+
+static cl::opt<bool>
+    ClInstrumentWrites("memprof-instrument-writes",
+                       cl::desc("instrument write instructions"), cl::Hidden,
+                       cl::init(true));
+
+static cl::opt<bool> ClInstrumentAtomics(
+    "memprof-instrument-atomics",
+    cl::desc("instrument atomic instructions (rmw, cmpxchg)"), cl::Hidden,
+    cl::init(true));
+
+static cl::opt<bool> ClUseCalls(
+    "memprof-use-callbacks",
+    cl::desc("Use callbacks instead of inline instrumentation sequences."),
+    cl::Hidden, cl::init(false));
+
+static cl::opt<std::string>
+    ClMemoryAccessCallbackPrefix("memprof-memory-access-callback-prefix",
+                                 cl::desc("Prefix for memory access callbacks"),
+                                 cl::Hidden, cl::init("__memprof_"));
+
+// These flags allow to change the shadow mapping.
+// The shadow mapping looks like
+//    Shadow = ((Mem & mask) >> scale) + offset
+
+static cl::opt<int> ClMappingScale("memprof-mapping-scale",
+                                   cl::desc("scale of memprof shadow mapping"),
+                                   cl::Hidden, cl::init(DefaultShadowScale));
+
+static cl::opt<int>
+    ClMappingGranularity("memprof-mapping-granularity",
+                         cl::desc("granularity of memprof shadow mapping"),
+                         cl::Hidden, cl::init(DefaultShadowGranularity));
+
+// Debug flags.
+
+static cl::opt<int> ClDebug("memprof-debug", cl::desc("debug"), cl::Hidden,
+                            cl::init(0));
+
+static cl::opt<std::string> ClDebugFunc("memprof-debug-func", cl::Hidden,
+                                        cl::desc("Debug func"));
+
+static cl::opt<int> ClDebugMin("memprof-debug-min", cl::desc("Debug min inst"),
+                               cl::Hidden, cl::init(-1));
+
+static cl::opt<int> ClDebugMax("memprof-debug-max", cl::desc("Debug max inst"),
+                               cl::Hidden, cl::init(-1));
+
+STATISTIC(NumInstrumentedReads, "Number of instrumented reads");
+STATISTIC(NumInstrumentedWrites, "Number of instrumented writes");
+
+namespace {
+
+/// This struct defines the shadow mapping using the rule:
+///   shadow = ((mem & mask) >> Scale) ADD DynamicShadowOffset.
+struct ShadowMapping {
+  ShadowMapping() {
+    Scale = ClMappingScale;
+    Granularity = ClMappingGranularity;
+    Mask = ~(Granularity - 1);
+  }
+
+  int Scale;
+  int Granularity;
+  uint64_t Mask; // Computed as ~(Granularity-1)
+};
+
+static uint64_t getCtorAndDtorPriority(Triple &TargetTriple) {
+  return TargetTriple.isOSEmscripten() ? MemProfEmscriptenCtorAndDtorPriority
+                                       : MemProfCtorAndDtorPriority;
+}
+
+struct InterestingMemoryAccess {
+  Value *Addr = nullptr;
+  bool IsWrite;
+  unsigned Alignment;
+  uint64_t TypeSize;
+  Value *MaybeMask = nullptr;
+};
+
+/// Instrument the code in module to profile memory accesses.
+class MemProfiler {
+public:
+  MemProfiler(Module &M) {
+    C = &(M.getContext());
+    LongSize = M.getDataLayout().getPointerSizeInBits();
+    IntptrTy = Type::getIntNTy(*C, LongSize);
+  }
+
+  /// If it is an interesting memory access, populate information
+  /// about the access and return a InterestingMemoryAccess struct.
+  /// Otherwise return None.
+  Optional<InterestingMemoryAccess>
+  isInterestingMemoryAccess(Instruction *I) const;
+
+  void instrumentMop(Instruction *I, const DataLayout &DL,
+                     InterestingMemoryAccess &Access);
+  void instrumentAddress(Instruction *OrigIns, Instruction *InsertBefore,
+                         Value *Addr, uint32_t TypeSize, bool IsWrite);
+  void instrumentMaskedLoadOrStore(const DataLayout &DL, Value *Mask,
+                                   Instruction *I, Value *Addr,
+                                   unsigned Alignment, uint32_t TypeSize,
+                                   bool IsWrite);
+  void instrumentMemIntrinsic(MemIntrinsic *MI);
+  Value *memToShadow(Value *Shadow, IRBuilder<> &IRB);
+  bool instrumentFunction(Function &F);
+  bool maybeInsertMemProfInitAtFunctionEntry(Function &F);
+  bool insertDynamicShadowAtFunctionEntry(Function &F);
+
+private:
+  void initializeCallbacks(Module &M);
+
+  LLVMContext *C;
+  int LongSize;
+  Type *IntptrTy;
+  ShadowMapping Mapping;
+
+  // These arrays is indexed by AccessIsWrite
+  FunctionCallee MemProfMemoryAccessCallback[2];
+  FunctionCallee MemProfMemoryAccessCallbackSized[2];
+
+  FunctionCallee MemProfMemmove, MemProfMemcpy, MemProfMemset;
+  Value *DynamicShadowOffset = nullptr;
+};
+
+class MemProfilerLegacyPass : public FunctionPass {
+public:
+  static char ID;
+
+  explicit MemProfilerLegacyPass() : FunctionPass(ID) {
+    initializeMemProfilerLegacyPassPass(*PassRegistry::getPassRegistry());
+  }
+
+  StringRef getPassName() const override { return "MemProfilerFunctionPass"; }
+
+  bool runOnFunction(Function &F) override {
+    MemProfiler Profiler(*F.getParent());
+    return Profiler.instrumentFunction(F);
+  }
+};
+
+class ModuleMemProfiler {
+public:
+  ModuleMemProfiler(Module &M) { TargetTriple = Triple(M.getTargetTriple()); }
+
+  bool instrumentModule(Module &);
+
+private:
+  Triple TargetTriple;
+  ShadowMapping Mapping;
+  Function *MemProfCtorFunction = nullptr;
+};
+
+class ModuleMemProfilerLegacyPass : public ModulePass {
+public:
+  static char ID;
+
+  explicit ModuleMemProfilerLegacyPass() : ModulePass(ID) {
+    initializeModuleMemProfilerLegacyPassPass(*PassRegistry::getPassRegistry());
+  }
+
+  StringRef getPassName() const override { return "ModuleMemProfiler"; }
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override {}
+
+  bool runOnModule(Module &M) override {
+    ModuleMemProfiler MemProfiler(M);
+    return MemProfiler.instrumentModule(M);
+  }
+};
+
+} // end anonymous namespace
+
+MemProfilerPass::MemProfilerPass() {}
+
+PreservedAnalyses MemProfilerPass::run(Function &F,
+                                       AnalysisManager<Function> &AM) {
+  Module &M = *F.getParent();
+  MemProfiler Profiler(M);
+  if (Profiler.instrumentFunction(F))
+    return PreservedAnalyses::none();
+  return PreservedAnalyses::all();
+
+  return PreservedAnalyses::all();
+}
+
+ModuleMemProfilerPass::ModuleMemProfilerPass() {}
+
+PreservedAnalyses ModuleMemProfilerPass::run(Module &M,
+                                             AnalysisManager<Module> &AM) {
+  ModuleMemProfiler Profiler(M);
+  if (Profiler.instrumentModule(M))
+    return PreservedAnalyses::none();
+  return PreservedAnalyses::all();
+}
+
+char MemProfilerLegacyPass::ID = 0;
+
+INITIALIZE_PASS_BEGIN(MemProfilerLegacyPass, "memprof",
+                      "MemProfiler: profile memory allocations and accesses.",
+                      false, false)
+INITIALIZE_PASS_END(MemProfilerLegacyPass, "memprof",
+                    "MemProfiler: profile memory allocations and accesses.",
+                    false, false)
+
+FunctionPass *llvm::createMemProfilerFunctionPass() {
+  return new MemProfilerLegacyPass();
+}
+
+char ModuleMemProfilerLegacyPass::ID = 0;
+
+INITIALIZE_PASS(ModuleMemProfilerLegacyPass, "memprof-module",
+                "MemProfiler: profile memory allocations and accesses."
+                "ModulePass",
+                false, false)
+
+ModulePass *llvm::createModuleMemProfilerLegacyPassPass() {
+  return new ModuleMemProfilerLegacyPass();
+}
+
+Value *MemProfiler::memToShadow(Value *Shadow, IRBuilder<> &IRB) {
+  // (Shadow & mask) >> scale
+  Shadow = IRB.CreateAnd(Shadow, Mapping.Mask);
+  Shadow = IRB.CreateLShr(Shadow, Mapping.Scale);
+  // (Shadow >> scale) | offset
+  assert(DynamicShadowOffset);
+  return IRB.CreateAdd(Shadow, DynamicShadowOffset);
+}
+
+// Instrument memset/memmove/memcpy
+void MemProfiler::instrumentMemIntrinsic(MemIntrinsic *MI) {
+  IRBuilder<> IRB(MI);
+  if (isa<MemTransferInst>(MI)) {
+    IRB.CreateCall(
+        isa<MemMoveInst>(MI) ? MemProfMemmove : MemProfMemcpy,
+        {IRB.CreatePointerCast(MI->getOperand(0), IRB.getInt8PtrTy()),
+         IRB.CreatePointerCast(MI->getOperand(1), IRB.getInt8PtrTy()),
+         IRB.CreateIntCast(MI->getOperand(2), IntptrTy, false)});
+  } else if (isa<MemSetInst>(MI)) {
+    IRB.CreateCall(
+        MemProfMemset,
+        {IRB.CreatePointerCast(MI->getOperand(0), IRB.getInt8PtrTy()),
+         IRB.CreateIntCast(MI->getOperand(1), IRB.getInt32Ty(), false),
+         IRB.CreateIntCast(MI->getOperand(2), IntptrTy, false)});
+  }
+  MI->eraseFromParent();
+}
+
+Optional<InterestingMemoryAccess>
+MemProfiler::isInterestingMemoryAccess(Instruction *I) const {
+  // Do not instrument the load fetching the dynamic shadow address.
+  if (DynamicShadowOffset == I)
+    return None;
+
+  InterestingMemoryAccess Access;
+
+  const DataLayout &DL = I->getModule()->getDataLayout();
+  if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
+    if (!ClInstrumentReads)
+      return None;
+    Access.IsWrite = false;
+    Access.TypeSize = DL.getTypeStoreSizeInBits(LI->getType());
+    Access.Alignment = LI->getAlignment();
+    Access.Addr = LI->getPointerOperand();
+  } else if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
+    if (!ClInstrumentWrites)
+      return None;
+    Access.IsWrite = true;
+    Access.TypeSize =
+        DL.getTypeStoreSizeInBits(SI->getValueOperand()->getType());
+    Access.Alignment = SI->getAlignment();
+    Access.Addr = SI->getPointerOperand();
+  } else if (AtomicRMWInst *RMW = dyn_cast<AtomicRMWInst>(I)) {
+    if (!ClInstrumentAtomics)
+      return None;
+    Access.IsWrite = true;
+    Access.TypeSize =
+        DL.getTypeStoreSizeInBits(RMW->getValOperand()->getType());
+    Access.Alignment = 0;
+    Access.Addr = RMW->getPointerOperand();
+  } else if (AtomicCmpXchgInst *XCHG = dyn_cast<AtomicCmpXchgInst>(I)) {
+    if (!ClInstrumentAtomics)
+      return None;
+    Access.IsWrite = true;
+    Access.TypeSize =
+        DL.getTypeStoreSizeInBits(XCHG->getCompareOperand()->getType());
+    Access.Alignment = 0;
+    Access.Addr = XCHG->getPointerOperand();
+  } else if (auto *CI = dyn_cast<CallInst>(I)) {
+    auto *F = CI->getCalledFunction();
+    if (F && (F->getIntrinsicID() == Intrinsic::masked_load ||
+              F->getIntrinsicID() == Intrinsic::masked_store)) {
+      unsigned OpOffset = 0;
+      if (F->getIntrinsicID() == Intrinsic::masked_store) {
+        if (!ClInstrumentWrites)
+          return None;
+        // Masked store has an initial operand for the value.
+        OpOffset = 1;
+        Access.IsWrite = true;
+      } else {
+        if (!ClInstrumentReads)
+          return None;
+        Access.IsWrite = false;
+      }
+
+      auto *BasePtr = CI->getOperand(0 + OpOffset);
+      auto *Ty = cast<PointerType>(BasePtr->getType())->getElementType();
+      Access.TypeSize = DL.getTypeStoreSizeInBits(Ty);
+      if (auto *AlignmentConstant =
+              dyn_cast<ConstantInt>(CI->getOperand(1 + OpOffset)))
+        Access.Alignment = (unsigned)AlignmentConstant->getZExtValue();
+      else
+        Access.Alignment = 1; // No alignment guarantees. We probably got Undef
+      Access.MaybeMask = CI->getOperand(2 + OpOffset);
+      Access.Addr = BasePtr;
+    }
+  }
+
+  if (!Access.Addr)
+    return None;
+
+  // Do not instrument acesses from different address spaces; we cannot deal
+  // with them.
+  Type *PtrTy = cast<PointerType>(Access.Addr->getType()->getScalarType());
+  if (PtrTy->getPointerAddressSpace() != 0)
+    return None;
+
+  // Ignore swifterror addresses.
+  // swifterror memory addresses are mem2reg promoted by instruction
+  // selection. As such they cannot have regular uses like an instrumentation
+  // function and it makes no sense to track them as memory.
+  if (Access.Addr->isSwiftError())
+    return None;
+
+  return Access;
+}
+
+void MemProfiler::instrumentMaskedLoadOrStore(const DataLayout &DL, Value *Mask,
+                                              Instruction *I, Value *Addr,
+                                              unsigned Alignment,
+                                              uint32_t TypeSize, bool IsWrite) {
+  auto *VTy = cast<FixedVectorType>(
+      cast<PointerType>(Addr->getType())->getElementType());
+  uint64_t ElemTypeSize = DL.getTypeStoreSizeInBits(VTy->getScalarType());
+  unsigned Num = VTy->getNumElements();
+  auto *Zero = ConstantInt::get(IntptrTy, 0);
+  for (unsigned Idx = 0; Idx < Num; ++Idx) {
+    Value *InstrumentedAddress = nullptr;
+    Instruction *InsertBefore = I;
+    if (auto *Vector = dyn_cast<ConstantVector>(Mask)) {
+      // dyn_cast as we might get UndefValue
+      if (auto *Masked = dyn_cast<ConstantInt>(Vector->getOperand(Idx))) {
+        if (Masked->isZero())
+          // Mask is constant false, so no instrumentation needed.
+          continue;
+        // If we have a true or undef value, fall through to instrumentAddress.
+        // with InsertBefore == I
+      }
+    } else {
+      IRBuilder<> IRB(I);
+      Value *MaskElem = IRB.CreateExtractElement(Mask, Idx);
+      Instruction *ThenTerm = SplitBlockAndInsertIfThen(MaskElem, I, false);
+      InsertBefore = ThenTerm;
+    }
+
+    IRBuilder<> IRB(InsertBefore);
+    InstrumentedAddress =
+        IRB.CreateGEP(VTy, Addr, {Zero, ConstantInt::get(IntptrTy, Idx)});
+    instrumentAddress(I, InsertBefore, InstrumentedAddress, ElemTypeSize,
+                      IsWrite);
+  }
+}
+
+void MemProfiler::instrumentMop(Instruction *I, const DataLayout &DL,
+                                InterestingMemoryAccess &Access) {
+  if (Access.IsWrite)
+    NumInstrumentedWrites++;
+  else
+    NumInstrumentedReads++;
+
+  if (Access.MaybeMask) {
+    instrumentMaskedLoadOrStore(DL, Access.MaybeMask, I, Access.Addr,
+                                Access.Alignment, Access.TypeSize,
+                                Access.IsWrite);
+  } else {
+    // Since the access counts will be accumulated across the entire allocation,
+    // we only update the shadow access count for the first location and thus
+    // don't need to worry about alignment and type size.
+    instrumentAddress(I, I, Access.Addr, Access.TypeSize, Access.IsWrite);
+  }
+}
+
+void MemProfiler::instrumentAddress(Instruction *OrigIns,
+                                    Instruction *InsertBefore, Value *Addr,
+                                    uint32_t TypeSize, bool IsWrite) {
+  IRBuilder<> IRB(InsertBefore);
+  Value *AddrLong = IRB.CreatePointerCast(Addr, IntptrTy);
+
+  if (ClUseCalls) {
+    IRB.CreateCall(MemProfMemoryAccessCallback[IsWrite], AddrLong);
+    return;
+  }
+
+  // Create an inline sequence to compute shadow location, and increment the
+  // value by one.
+  Type *ShadowTy = Type::getInt64Ty(*C);
+  Type *ShadowPtrTy = PointerType::get(ShadowTy, 0);
+  Value *ShadowPtr = memToShadow(AddrLong, IRB);
+  Value *ShadowAddr = IRB.CreateIntToPtr(ShadowPtr, ShadowPtrTy);
+  Value *ShadowValue = IRB.CreateLoad(ShadowTy, ShadowAddr);
+  Value *Inc = ConstantInt::get(Type::getInt64Ty(*C), 1);
+  ShadowValue = IRB.CreateAdd(ShadowValue, Inc);
+  IRB.CreateStore(ShadowValue, ShadowAddr);
+}
+
+// Create the variable for the profile file name.
+void createProfileFileNameVar(Module &M) {
+  const MDString *MemProfFilename =
+      dyn_cast_or_null<MDString>(M.getModuleFlag("MemProfProfileFilename"));
+  if (!MemProfFilename)
+    return;
+  assert(!MemProfFilename->getString().empty() &&
+         "Unexpected MemProfProfileFilename metadata with empty string");
+  Constant *ProfileNameConst = ConstantDataArray::getString(
+      M.getContext(), MemProfFilename->getString(), true);
+  GlobalVariable *ProfileNameVar = new GlobalVariable(
+      M, ProfileNameConst->getType(), /*isConstant=*/true,
+      GlobalValue::WeakAnyLinkage, ProfileNameConst, MemProfFilenameVar);
+  Triple TT(M.getTargetTriple());
+  if (TT.supportsCOMDAT()) {
+    ProfileNameVar->setLinkage(GlobalValue::ExternalLinkage);
+    ProfileNameVar->setComdat(M.getOrInsertComdat(MemProfFilenameVar));
+  }
+}
+
+bool ModuleMemProfiler::instrumentModule(Module &M) {
+  // Create a module constructor.
+  std::string MemProfVersion = std::to_string(LLVM_MEM_PROFILER_VERSION);
+  std::string VersionCheckName =
+      ClInsertVersionCheck ? (MemProfVersionCheckNamePrefix + MemProfVersion)
+                           : "";
+  std::tie(MemProfCtorFunction, std::ignore) =
+      createSanitizerCtorAndInitFunctions(M, MemProfModuleCtorName,
+                                          MemProfInitName, /*InitArgTypes=*/{},
+                                          /*InitArgs=*/{}, VersionCheckName);
+
+  const uint64_t Priority = getCtorAndDtorPriority(TargetTriple);
+  appendToGlobalCtors(M, MemProfCtorFunction, Priority);
+
+  createProfileFileNameVar(M);
+
+  return true;
+}
+
+void MemProfiler::initializeCallbacks(Module &M) {
+  IRBuilder<> IRB(*C);
+
+  for (size_t AccessIsWrite = 0; AccessIsWrite <= 1; AccessIsWrite++) {
+    const std::string TypeStr = AccessIsWrite ? "store" : "load";
+
+    SmallVector<Type *, 3> Args2 = {IntptrTy, IntptrTy};
+    SmallVector<Type *, 2> Args1{1, IntptrTy};
+    MemProfMemoryAccessCallbackSized[AccessIsWrite] =
+        M.getOrInsertFunction(ClMemoryAccessCallbackPrefix + TypeStr + "N",
+                              FunctionType::get(IRB.getVoidTy(), Args2, false));
+
+    MemProfMemoryAccessCallback[AccessIsWrite] =
+        M.getOrInsertFunction(ClMemoryAccessCallbackPrefix + TypeStr,
+                              FunctionType::get(IRB.getVoidTy(), Args1, false));
+  }
+  MemProfMemmove = M.getOrInsertFunction(
+      ClMemoryAccessCallbackPrefix + "memmove", IRB.getInt8PtrTy(),
+      IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IntptrTy);
+  MemProfMemcpy = M.getOrInsertFunction(ClMemoryAccessCallbackPrefix + "memcpy",
+                                        IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
+                                        IRB.getInt8PtrTy(), IntptrTy);
+  MemProfMemset = M.getOrInsertFunction(ClMemoryAccessCallbackPrefix + "memset",
+                                        IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
+                                        IRB.getInt32Ty(), IntptrTy);
+}
+
+bool MemProfiler::maybeInsertMemProfInitAtFunctionEntry(Function &F) {
+  // For each NSObject descendant having a +load method, this method is invoked
+  // by the ObjC runtime before any of the static constructors is called.
+  // Therefore we need to instrument such methods with a call to __memprof_init
+  // at the beginning in order to initialize our runtime before any access to
+  // the shadow memory.
+  // We cannot just ignore these methods, because they may call other
+  // instrumented functions.
+  if (F.getName().find(" load]") != std::string::npos) {
+    FunctionCallee MemProfInitFunction =
+        declareSanitizerInitFunction(*F.getParent(), MemProfInitName, {});
+    IRBuilder<> IRB(&F.front(), F.front().begin());
+    IRB.CreateCall(MemProfInitFunction, {});
+    return true;
+  }
+  return false;
+}
+
+bool MemProfiler::insertDynamicShadowAtFunctionEntry(Function &F) {
+  IRBuilder<> IRB(&F.front().front());
+  Value *GlobalDynamicAddress = F.getParent()->getOrInsertGlobal(
+      MemProfShadowMemoryDynamicAddress, IntptrTy);
+  if (F.getParent()->getPICLevel() == PICLevel::NotPIC)
+    cast<GlobalVariable>(GlobalDynamicAddress)->setDSOLocal(true);
+  DynamicShadowOffset = IRB.CreateLoad(IntptrTy, GlobalDynamicAddress);
+  return true;
+}
+
+bool MemProfiler::instrumentFunction(Function &F) {
+  if (F.getLinkage() == GlobalValue::AvailableExternallyLinkage)
+    return false;
+  if (ClDebugFunc == F.getName())
+    return false;
+  if (F.getName().startswith("__memprof_"))
+    return false;
+
+  bool FunctionModified = false;
+
+  // If needed, insert __memprof_init.
+  // This function needs to be called even if the function body is not
+  // instrumented.
+  if (maybeInsertMemProfInitAtFunctionEntry(F))
+    FunctionModified = true;
+
+  LLVM_DEBUG(dbgs() << "MEMPROF instrumenting:\n" << F << "\n");
+
+  initializeCallbacks(*F.getParent());
+
+  FunctionModified |= insertDynamicShadowAtFunctionEntry(F);
+
+  SmallVector<Instruction *, 16> ToInstrument;
+
+  // Fill the set of memory operations to instrument.
+  for (auto &BB : F) {
+    for (auto &Inst : BB) {
+      if (isInterestingMemoryAccess(&Inst) || isa<MemIntrinsic>(Inst))
+        ToInstrument.push_back(&Inst);
+    }
+  }
+
+  int NumInstrumented = 0;
+  for (auto *Inst : ToInstrument) {
+    if (ClDebugMin < 0 || ClDebugMax < 0 ||
+        (NumInstrumented >= ClDebugMin && NumInstrumented <= ClDebugMax)) {
+      Optional<InterestingMemoryAccess> Access =
+          isInterestingMemoryAccess(Inst);
+      if (Access)
+        instrumentMop(Inst, F.getParent()->getDataLayout(), *Access);
+      else
+        instrumentMemIntrinsic(cast<MemIntrinsic>(Inst));
+    }
+    NumInstrumented++;
+  }
+
+  if (NumInstrumented > 0)
+    FunctionModified = true;
+
+  LLVM_DEBUG(dbgs() << "MEMPROF done instrumenting: " << FunctionModified << " "
+                    << F << "\n");
+
+  return FunctionModified;
+}