Outline C++ exception handlers for MSVC targets.

[andrew.w.kaylor]

Summary

This patch adds code to outline the catch handlers for Windows C++ exception handling support.

These changes are obviously derived from my earlier patch D6556, but since the direction has changed (from SEH to C++EH) it seemed reasonable to start a new differential.

The test cases are based on clang-generated IR for the listed code using an x86_64-pc-linux target, but I have edited this IR to match expected windows-msvc output.

I edited the IR to remove redundant store and load of the EH object and to replace invokes within a catch block with equivalent calls if the unwind destination does nothing but resume. I have code to perform both of these transformations, but I am holding it back for a later review.

The following things are not yet implemented:

Intrinsic support for llvm.eh.begincatch/llvm.eh.endcatch
Removal of outlined catch handlers.
Use of frame variables within a catch block.
Removal (from the parent function) of frame variables which are only used in catch blocks.
Outlining of unwind handlers.
Chained catch handlers (catching an exception from within a catch handler).
Sharing of blocks between several landing pads.
Production of intrinsics to support .xdata table generation.

I think that making the begin/end catch calls intrinsics is the only thing that should block the current patch. I'll submit a separate patch for that shortly.

[rnk]

Looks like a good incremental step to me.

I added David Majnemer to see if he has anything.

lib/CodeGen/WinEHPrepare.cpp

61–93 ↗

(On Diff #19193)

I'd rather we went with a clang-format indentation here, as it's the dominant style across LLVM.

[andrew.w.kaylor]

Rebased and cleaned up formatting.
Updated to use llvm.eh.begincatch and llvm.eh.endcatch intrinsics.

I was waiting for that change before committing this, but I think it should be ready now. I also have outlining of cleanup handlers ready to review as soon as this lands and some other implementation details that should be ready soon thereafter.

[majnemer]

include/llvm/Transforms/Utils/Cloning.h

140–141 ↗

(On Diff #19706)

Style nit, please move the curly brace on the same line as the tag name.

lib/CodeGen/WinEHPrepare.cpp

54 ↗

(On Diff #19706)

I'd make this SmallVectorImpl.

123 ↗

(On Diff #19706)

Perhaps: This only returns true if...

150 ↗

(On Diff #19706)

SmallVectorImpl

156 ↗

(On Diff #19706)

You shouldn't have variables of type twine. They should be used directly in the expression which consumes them.

167–171 ↗

(On Diff #19706)

It would be more idiomatic to do:

if (auto *Call = dyn_cast<CallInst>(&Inst))
  if (Call->getCalledFunction->getName() == "llvm.eh.actions")

215 ↗

(On Diff #19706)

This might deserve a comment.

278–279 ↗

(On Diff #19706)

Please fold this into a single line.

295–296 ↗

(On Diff #19706)

Please fold this into a single line.

316–318 ↗

(On Diff #19706)

You can simplify this by using if (match(Inst, m_Intrinsic<Intrinsic::eh_begincatch>()))

328 ↗

(On Diff #19706)

Likewise here with if (match(Inst, m_Intrinsic<Intrinsic::eh_endcatch>()))

lib/Transforms/Utils/CloneFunction.cpp

684 ↗

(On Diff #19706)

SmallVectorImpl<ReturnInst *>

[andrew.w.kaylor]

Cleanup based on review feedback.

[rnk]

Looks good, please commit.

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

void CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
                        ValueToValueMapTy &VMap,
                        bool ModuleLevelChanges,
                        SmallVectorImpl<ReturnInst*> &Returns,
                        const char *NameSuffix = "",
                        ClonedCodeInfo *CodeInfo = nullptr,
                        ValueMapTypeRemapper *TypeMapper = nullptr,
                        ValueMaterializer *Materializer = nullptr);
 
+/// A helper class used with CloneAndPruneIntoFromInst to change the default
+/// behavior while instructions are being cloned.
+class CloningDirector {
+public:
+  /// This enumeration describes the way CloneAndPruneIntoFromInst should
+  /// proceed after the CloningDirector has examined an instruction.
+  enum CloningAction {
+    ///< Continue cloning the instruction (default behavior).
+    CloneInstruction,
+    ///< Skip this instruction but continue cloning the current basic block.
+    SkipInstruction,
+    ///< Skip this instruction and stop cloning the current basic block.
+    StopCloningBB
+  };
+
+  CloningDirector() {}
+  virtual ~CloningDirector() {}
+
+  /// Subclasses must override this function to customize cloning behavior.
+  virtual CloningAction handleInstruction(ValueToValueMapTy &VMap,
+                                          const Instruction *Inst,
+                                          BasicBlock        *NewBB) = 0;
+};
+
+void CloneAndPruneIntoFromInst(Function *NewFunc, const Function *OldFunc,
+                               const Instruction *StartingInst,
+                               ValueToValueMapTy &VMap,
+                               bool ModuleLevelChanges,
+                               SmallVectorImpl<ReturnInst*> &Returns,
+                               const char *NameSuffix = "", 
+                               ClonedCodeInfo *CodeInfo = nullptr,
+                               const DataLayout *DL = nullptr,
+                               CloningDirector *Director = nullptr);
+
+
 /// CloneAndPruneFunctionInto - This works exactly like CloneFunctionInto,
 /// except that it does some simple constant prop and DCE on the fly.  The
 /// effect of this is to copy significantly less code in cases where (for
 /// example) a function call with constant arguments is inlined, and those
 /// constant arguments cause a significant amount of code in the callee to be
 /// dead.  Since this doesn't produce an exactly copy of the input, it can't be
 /// used for things like CloneFunction or CloneModule.
 ///

llvm/trunk/lib/CodeGen/WinEHPrepare.cpp

-//===-- WinEHPrepare - Prepare exception handling for code generation ---===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This pass lowers LLVM IR exception handling into something closer to what the
-// backend wants. It snifs the personality function to see which kind of
-// preparation is necessary. If the personality function uses the Itanium LSDA,
-// this pass delegates to the DWARF EH preparation pass.
-//
-//===----------------------------------------------------------------------===//
-
-#include "llvm/CodeGen/Passes.h"
-#include "llvm/Analysis/LibCallSemantics.h"
-#include "llvm/IR/Function.h"
-#include "llvm/IR/IRBuilder.h"
-#include "llvm/IR/Instructions.h"
-#include "llvm/Pass.h"
-#include <memory>
-
-using namespace llvm;
-
-#define DEBUG_TYPE "winehprepare"
-
-namespace {
-class WinEHPrepare : public FunctionPass {
-  std::unique_ptr<FunctionPass> DwarfPrepare;
-
-public:
-  static char ID; // Pass identification, replacement for typeid.
-  WinEHPrepare(const TargetMachine *TM = nullptr)
-      : FunctionPass(ID), DwarfPrepare(createDwarfEHPass(TM)) {}
-
-  bool runOnFunction(Function &Fn) override;
-
-  bool doFinalization(Module &M) override;
-
-  void getAnalysisUsage(AnalysisUsage &AU) const override;
-
-  const char *getPassName() const override {
-    return "Windows exception handling preparation";
-  }
-};
-} // end anonymous namespace
-
-char WinEHPrepare::ID = 0;
-INITIALIZE_TM_PASS(WinEHPrepare, "winehprepare",
-                   "Prepare Windows exceptions", false, false)
-
-FunctionPass *llvm::createWinEHPass(const TargetMachine *TM) {
-  return new WinEHPrepare(TM);
-}
-
-static bool isMSVCPersonality(EHPersonality Pers) {
-  return Pers == EHPersonality::MSVC_Win64SEH ||
-         Pers == EHPersonality::MSVC_CXX;
-}
-
-bool WinEHPrepare::runOnFunction(Function &Fn) {
-  SmallVector<LandingPadInst *, 4> LPads;
-  SmallVector<ResumeInst *, 4> Resumes;
-  for (BasicBlock &BB : Fn) {
-    if (auto *LP = BB.getLandingPadInst())
-      LPads.push_back(LP);
-    if (auto *Resume = dyn_cast<ResumeInst>(BB.getTerminator()))
-      Resumes.push_back(Resume);
-  }
-
-  // No need to prepare functions that lack landing pads.
-  if (LPads.empty())
-    return false;
-
-  // Classify the personality to see what kind of preparation we need.
-  EHPersonality Pers = classifyEHPersonality(LPads.back()->getPersonalityFn());
-
-  // Delegate through to the DWARF pass if this is unrecognized.
-  if (!isMSVCPersonality(Pers))
-    return DwarfPrepare->runOnFunction(Fn);
-
-  // FIXME: Cleanups are unimplemented. Replace them with unreachable.
-  if (Resumes.empty())
-    return false;
-
-  for (ResumeInst *Resume : Resumes) {
-    IRBuilder<>(Resume).CreateUnreachable();
-    Resume->eraseFromParent();
-  }
-
-  return true;
-}
-
-bool WinEHPrepare::doFinalization(Module &M) {
-  return DwarfPrepare->doFinalization(M);
-}
-
-void WinEHPrepare::getAnalysisUsage(AnalysisUsage &AU) const {
-  DwarfPrepare->getAnalysisUsage(AU);
-}
+//===-- WinEHPrepare - Prepare exception handling for code generation ---===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass lowers LLVM IR exception handling into something closer to what the
+// backend wants. It snifs the personality function to see which kind of
+// preparation is necessary. If the personality function uses the Itanium LSDA,
+// this pass delegates to the DWARF EH preparation pass.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/Analysis/LibCallSemantics.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/PatternMatch.h"
+#include "llvm/Pass.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include "llvm/Transforms/Utils/Local.h"
+#include <memory>
+
+using namespace llvm;
+using namespace llvm::PatternMatch;
+
+#define DEBUG_TYPE "winehprepare"
+
+namespace {
+class WinEHPrepare : public FunctionPass {
+  std::unique_ptr<FunctionPass> DwarfPrepare;
+
+public:
+  static char ID; // Pass identification, replacement for typeid.
+  WinEHPrepare(const TargetMachine *TM = nullptr)
+      : FunctionPass(ID), DwarfPrepare(createDwarfEHPass(TM)) {}
+
+  bool runOnFunction(Function &Fn) override;
+
+  bool doFinalization(Module &M) override;
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override;
+
+  const char *getPassName() const override {
+    return "Windows exception handling preparation";
+  }
+
+private:
+  bool prepareCPPEHHandlers(Function &F,
+                            SmallVectorImpl<LandingPadInst *> &LPads);
+  bool outlineCatchHandler(Function *SrcFn, Constant *SelectorType,
+                           LandingPadInst *LPad, StructType *EHDataStructTy);
+};
+
+class WinEHCatchDirector : public CloningDirector {
+public:
+  WinEHCatchDirector(LandingPadInst *LPI, Function *CatchFn, Value *Selector,
+                     Value *EHObj)
+      : LPI(LPI), CatchFn(CatchFn),
+        CurrentSelector(Selector->stripPointerCasts()), EHObj(EHObj),
+        SelectorIDType(Type::getInt32Ty(LPI->getContext())),
+        Int8PtrType(Type::getInt8PtrTy(LPI->getContext())) {}
+  virtual ~WinEHCatchDirector() {}
+
+  CloningAction handleInstruction(ValueToValueMapTy &VMap,
+                                  const Instruction *Inst,
+                                  BasicBlock *NewBB) override;
+
+private:
+  LandingPadInst *LPI;
+  Function *CatchFn;
+  Value *CurrentSelector;
+  Value *EHObj;
+  Type *SelectorIDType;
+  Type *Int8PtrType;
+
+  const Value *ExtractedEHPtr;
+  const Value *ExtractedSelector;
+  const Value *EHPtrStoreAddr;
+  const Value *SelectorStoreAddr;
+  const Value *EHObjStoreAddr;
+};
+} // end anonymous namespace
+
+char WinEHPrepare::ID = 0;
+INITIALIZE_TM_PASS(WinEHPrepare, "winehprepare", "Prepare Windows exceptions",
+                   false, false)
+
+FunctionPass *llvm::createWinEHPass(const TargetMachine *TM) {
+  return new WinEHPrepare(TM);
+}
+
+static bool isMSVCPersonality(EHPersonality Pers) {
+  return Pers == EHPersonality::MSVC_Win64SEH ||
+         Pers == EHPersonality::MSVC_CXX;
+}
+
+bool WinEHPrepare::runOnFunction(Function &Fn) {
+  SmallVector<LandingPadInst *, 4> LPads;
+  SmallVector<ResumeInst *, 4> Resumes;
+  for (BasicBlock &BB : Fn) {
+    if (auto *LP = BB.getLandingPadInst())
+      LPads.push_back(LP);
+    if (auto *Resume = dyn_cast<ResumeInst>(BB.getTerminator()))
+      Resumes.push_back(Resume);
+  }
+
+  // No need to prepare functions that lack landing pads.
+  if (LPads.empty())
+    return false;
+
+  // Classify the personality to see what kind of preparation we need.
+  EHPersonality Pers = classifyEHPersonality(LPads.back()->getPersonalityFn());
+
+  // Delegate through to the DWARF pass if this is unrecognized.
+  if (!isMSVCPersonality(Pers))
+    return DwarfPrepare->runOnFunction(Fn);
+
+  // FIXME: This only returns true if the C++ EH handlers were outlined.
+  //        When that code is complete, it should always return whatever
+  //        prepareCPPEHHandlers returns.
+  if (Pers == EHPersonality::MSVC_CXX && prepareCPPEHHandlers(Fn, LPads))
+    return true;
+
+  // FIXME: SEH Cleanups are unimplemented. Replace them with unreachable.
+  if (Resumes.empty())
+    return false;
+
+  for (ResumeInst *Resume : Resumes) {
+    IRBuilder<>(Resume).CreateUnreachable();
+    Resume->eraseFromParent();
+  }
+
+  return true;
+}
+
+bool WinEHPrepare::doFinalization(Module &M) {
+  return DwarfPrepare->doFinalization(M);
+}
+
+void WinEHPrepare::getAnalysisUsage(AnalysisUsage &AU) const {
+  DwarfPrepare->getAnalysisUsage(AU);
+}
+
+bool WinEHPrepare::prepareCPPEHHandlers(
+    Function &F, SmallVectorImpl<LandingPadInst *> &LPads) {
+  // FIXME: Find all frame variable references in the handlers
+  //        to populate the structure elements.
+  SmallVector<Type *, 2> AllocStructTys;
+  AllocStructTys.push_back(Type::getInt32Ty(F.getContext()));   // EH state
+  AllocStructTys.push_back(Type::getInt8PtrTy(F.getContext())); // EH object
+  StructType *EHDataStructTy =
+      StructType::create(F.getContext(), AllocStructTys, 
+                         "struct." + F.getName().str() + ".ehdata");
+  bool HandlersOutlined = false;
+
+  for (LandingPadInst *LPad : LPads) {
+    // Look for evidence that this landingpad has already been processed.
+    bool LPadHasActionList = false;
+    BasicBlock *LPadBB = LPad->getParent();
+    for (Instruction &Inst : LPadBB->getInstList()) {
+      // FIXME: Make this an intrinsic.
+      if (auto *Call = dyn_cast<CallInst>(&Inst))
+        if (Call->getCalledFunction()->getName() == "llvm.eh.actions") {
+          LPadHasActionList = true;
+          break;
+        }
+    }
+
+    // If we've already outlined the handlers for this landingpad,
+    // there's nothing more to do here.
+    if (LPadHasActionList)
+      continue;
+
+    for (unsigned Idx = 0, NumClauses = LPad->getNumClauses(); Idx < NumClauses;
+         ++Idx) {
+      if (LPad->isCatch(Idx))
+        HandlersOutlined =
+            outlineCatchHandler(&F, LPad->getClause(Idx), LPad, EHDataStructTy);
+    } // End for each clause
+  }   // End for each landingpad
+
+  return HandlersOutlined;
+}
+
+bool WinEHPrepare::outlineCatchHandler(Function *SrcFn, Constant *SelectorType,
+                                       LandingPadInst *LPad,
+                                       StructType *EHDataStructTy) {
+  Module *M = SrcFn->getParent();
+  LLVMContext &Context = M->getContext();
+
+  // Create a new function to receive the handler contents.
+  Type *Int8PtrType = Type::getInt8PtrTy(Context);
+  std::vector<Type *> ArgTys;
+  ArgTys.push_back(Int8PtrType);
+  ArgTys.push_back(Int8PtrType);
+  FunctionType *FnType = FunctionType::get(Int8PtrType, ArgTys, false);
+  Function *CatchHandler = Function::Create(
+      FnType, GlobalVariable::ExternalLinkage, SrcFn->getName() + ".catch", M);
+
+  // Generate a standard prolog to setup the frame recovery structure.
+  IRBuilder<> Builder(Context);
+  BasicBlock *Entry = BasicBlock::Create(Context, "catch.entry");
+  CatchHandler->getBasicBlockList().push_front(Entry);
+  Builder.SetInsertPoint(Entry);
+  Builder.SetCurrentDebugLocation(LPad->getDebugLoc());
+
+  // The outlined handler will be called with the parent's frame pointer as
+  // its second argument. To enable the handler to access variables from
+  // the parent frame, we use that pointer to get locate a special block
+  // of memory that was allocated using llvm.eh.allocateframe for this
+  // purpose.  During the outlining process we will determine which frame
+  // variables are used in handlers and create a structure that maps these
+  // variables into the frame allocation block.
+  //
+  // The frame allocation block also contains an exception state variable
+  // used by the runtime and a pointer to the exception object pointer
+  // which will be filled in by the runtime for use in the handler.
+  Function *RecoverFrameFn =
+      Intrinsic::getDeclaration(M, Intrinsic::framerecover);
+  Value *RecoverArgs[] = {Builder.CreateBitCast(SrcFn, Int8PtrType, ""),
+                          &(CatchHandler->getArgumentList().back())};
+  CallInst *EHAlloc =
+      Builder.CreateCall(RecoverFrameFn, RecoverArgs, "eh.alloc");
+  Value *EHData =
+      Builder.CreateBitCast(EHAlloc, EHDataStructTy->getPointerTo(), "ehdata");
+  Value *EHObjPtr =
+      Builder.CreateConstInBoundsGEP2_32(EHData, 0, 1, "eh.obj.ptr");
+
+  // This will give us a raw pointer to the exception object, which
+  // corresponds to the formal parameter of the catch statement.  If the
+  // handler uses this object, we will generate code during the outlining
+  // process to cast the pointer to the appropriate type and deference it
+  // as necessary.  The un-outlined landing pad code represents the
+  // exception object as the result of the llvm.eh.begincatch call.
+  Value *EHObj = Builder.CreateLoad(EHObjPtr, false, "eh.obj");
+
+  ValueToValueMapTy VMap;
+
+  // FIXME: Map other values referenced in the filter handler.
+
+  WinEHCatchDirector Director(LPad, CatchHandler, SelectorType, EHObj);
+
+  SmallVector<ReturnInst *, 8> Returns;
+  ClonedCodeInfo InlinedFunctionInfo;
+
+  BasicBlock::iterator II = LPad;
+
+  CloneAndPruneIntoFromInst(CatchHandler, SrcFn, ++II, VMap,
+                            /*ModuleLevelChanges=*/false, Returns, "",
+                            &InlinedFunctionInfo,
+                            SrcFn->getParent()->getDataLayout(), &Director);
+
+  // Move all the instructions in the first cloned block into our entry block.
+  BasicBlock *FirstClonedBB = std::next(Function::iterator(Entry));
+  Entry->getInstList().splice(Entry->end(), FirstClonedBB->getInstList());
+  FirstClonedBB->eraseFromParent();
+
+  return true;
+}
+
+CloningDirector::CloningAction WinEHCatchDirector::handleInstruction(
+    ValueToValueMapTy &VMap, const Instruction *Inst, BasicBlock *NewBB) {
+  // Intercept instructions which extract values from the landing pad aggregate.
+  if (auto *Extract = dyn_cast<ExtractValueInst>(Inst)) {
+    if (Extract->getAggregateOperand() == LPI) {
+      assert(Extract->getNumIndices() == 1 &&
+             "Unexpected operation: extracting both landing pad values");
+      assert((*(Extract->idx_begin()) == 0 || *(Extract->idx_begin()) == 1) &&
+             "Unexpected operation: extracting an unknown landing pad element");
+
+      if (*(Extract->idx_begin()) == 0) {
+        // Element 0 doesn't directly corresponds to anything in the WinEH scheme.
+        // It will be stored to a memory location, then later loaded and finally
+        // the loaded value will be used as the argument to an llvm.eh.begincatch
+        // call.  We're tracking it here so that we can skip the store and load.
+        ExtractedEHPtr = Inst;
+      } else {
+        // Element 1 corresponds to the filter selector.  We'll map it to 1 for
+        // matching purposes, but it will also probably be stored to memory and
+        // reloaded, so we need to track the instuction so that we can map the
+        // loaded value too.
+        VMap[Inst] = ConstantInt::get(SelectorIDType, 1);
+        ExtractedSelector = Inst;
+      }
+
+      // Tell the caller not to clone this instruction.
+      return CloningDirector::SkipInstruction;
+    }
+    // Other extract value instructions just get cloned.
+    return CloningDirector::CloneInstruction;
+  }
+
+  if (auto *Store = dyn_cast<StoreInst>(Inst)) {
+    // Look for and suppress stores of the extracted landingpad values.
+    const Value *StoredValue = Store->getValueOperand();
+    if (StoredValue == ExtractedEHPtr) {
+      EHPtrStoreAddr = Store->getPointerOperand();
+      return CloningDirector::SkipInstruction;
+    }
+    if (StoredValue == ExtractedSelector) {
+      SelectorStoreAddr = Store->getPointerOperand();
+      return CloningDirector::SkipInstruction;
+    }
+
+    // Any other store just gets cloned.
+    return CloningDirector::CloneInstruction;
+  }
+
+  if (auto *Load = dyn_cast<LoadInst>(Inst)) {
+    // Look for loads of (previously suppressed) landingpad values.
+    // The EHPtr load can be ignored (it should only be used as
+    // an argument to llvm.eh.begincatch), but the selector value
+    // needs to be mapped to a constant value of 1 to be used to
+    // simplify the branching to always flow to the current handler.
+    const Value *LoadAddr = Load->getPointerOperand();
+    if (LoadAddr == EHPtrStoreAddr) {
+      VMap[Inst] = UndefValue::get(Int8PtrType);
+      return CloningDirector::SkipInstruction;
+    }
+    if (LoadAddr == SelectorStoreAddr) {
+      VMap[Inst] = ConstantInt::get(SelectorIDType, 1);
+      return CloningDirector::SkipInstruction;
+    }
+
+    // Any other loads just get cloned.
+    return CloningDirector::CloneInstruction;
+  }
+
+  if (match(Inst, m_Intrinsic<Intrinsic::eh_begincatch>())) {
+    // The argument to the call is some form of the first element of the
+    // landingpad aggregate value, but that doesn't matter.  It isn't used
+    // here.
+    // The return value of this instruction, however, is used to access the
+    // EH object pointer.  We have generated an instruction to get that value
+    // from the EH alloc block, so we can just map to that here.
+    VMap[Inst] = EHObj;
+    return CloningDirector::SkipInstruction;
+  }
+  if (match(Inst, m_Intrinsic<Intrinsic::eh_endcatch>())) {
+    auto *IntrinCall = dyn_cast<IntrinsicInst>(Inst);
+    // It might be interesting to track whether or not we are inside a catch
+    // function, but that might make the algorithm more brittle than it needs
+    // to be.
+
+    // The end catch call can occur in one of two places: either in a
+    // landingpad
+    // block that is part of the catch handlers exception mechanism, or at the
+    // end of the catch block.  If it occurs in a landing pad, we must skip it
+    // and continue so that the landing pad gets cloned.
+    // FIXME: This case isn't fully supported yet and shouldn't turn up in any
+    //        of the test cases until it is.
+    if (IntrinCall->getParent()->isLandingPad())
+      return CloningDirector::SkipInstruction;
+
+    // If an end catch occurs anywhere else the next instruction should be an
+    // unconditional branch instruction that we want to replace with a return
+    // to the the address of the branch target.
+    const BasicBlock *EndCatchBB = IntrinCall->getParent();
+    const TerminatorInst *Terminator = EndCatchBB->getTerminator();
+    const BranchInst *Branch = dyn_cast<BranchInst>(Terminator);
+    assert(Branch && Branch->isUnconditional());
+    assert(std::next(BasicBlock::const_iterator(IntrinCall)) ==
+            BasicBlock::const_iterator(Branch));
+
+    ReturnInst::Create(NewBB->getContext(),
+                        BlockAddress::get(Branch->getSuccessor(0)), NewBB);
+
+    // We just added a terminator to the cloned block.
+    // Tell the caller to stop processing the current basic block so that
+    // the branch instruction will be skipped.
+    return CloningDirector::StopCloningBB;
+  }
+  if (match(Inst, m_Intrinsic<Intrinsic::eh_typeid_for>())) {
+    auto *IntrinCall = dyn_cast<IntrinsicInst>(Inst);
+    Value *Selector = IntrinCall->getArgOperand(0)->stripPointerCasts();
+    // This causes a replacement that will collapse the landing pad CFG based
+    // on the filter function we intend to match.
+    if (Selector == CurrentSelector)
+      VMap[Inst] = ConstantInt::get(SelectorIDType, 1);
+    else
+      VMap[Inst] = ConstantInt::get(SelectorIDType, 0);
+    // Tell the caller not to clone this instruction.
+    return CloningDirector::SkipInstruction;
+  }
+
+  // Continue with the default cloning behavior.
+  return CloningDirector::CloneInstruction;
+}

llvm/trunk/lib/Transforms/Utils/CloneFunction.cpp

namespace {
   struct PruningFunctionCloner {
     Function *NewFunc;
     const Function *OldFunc;
     ValueToValueMapTy &VMap;
     bool ModuleLevelChanges;
     const char *NameSuffix;
     ClonedCodeInfo *CodeInfo;
     const DataLayout *DL;
+    CloningDirector *Director;
+
   public:
     PruningFunctionCloner(Function *newFunc, const Function *oldFunc,
                           ValueToValueMapTy &valueMap,
                           bool moduleLevelChanges,
                           const char *nameSuffix, 
                           ClonedCodeInfo *codeInfo,
-                          const DataLayout *DL)
+                          const DataLayout *DL,
+                          CloningDirector *Director)
     : NewFunc(newFunc), OldFunc(oldFunc),
       VMap(valueMap), ModuleLevelChanges(moduleLevelChanges),
-      NameSuffix(nameSuffix), CodeInfo(codeInfo), DL(DL) {
+      NameSuffix(nameSuffix), CodeInfo(codeInfo), DL(DL),
+      Director(Director) {
     }
 
     /// CloneBlock - The specified block is found to be reachable, clone it and
     /// anything that it can reach.
-    void CloneBlock(const BasicBlock *BB,
+    void CloneBlock(const BasicBlock *BB, 
+                    BasicBlock::const_iterator StartingInst,
                     std::vector<const BasicBlock*> &ToClone);
   };
 }
 
 /// CloneBlock - The specified block is found to be reachable, clone it and
 /// anything that it can reach.
 void PruningFunctionCloner::CloneBlock(const BasicBlock *BB,
+                                       BasicBlock::const_iterator StartingInst,
                                        std::vector<const BasicBlock*> &ToClone){
   WeakVH &BBEntry = VMap[BB];
 
   // Have we already cloned this block?
   if (BBEntry) return;
   
   // Nope, clone it now.
   BasicBlock *NewBB;
void PruningFunctionCloner::CloneBlock(const BasicBlock *BB,
   //
   // Note that we don't need to fix the mapping for unreachable blocks;
   // the default mapping there is safe.
   if (BB->hasAddressTaken()) {
     Constant *OldBBAddr = BlockAddress::get(const_cast<Function*>(OldFunc),
                                             const_cast<BasicBlock*>(BB));
     VMap[OldBBAddr] = BlockAddress::get(NewFunc, NewBB);
   }
-    
-
   bool hasCalls = false, hasDynamicAllocas = false, hasStaticAllocas = false;
-  
   // Loop over all instructions, and copy them over, DCE'ing as we go.  This
   // loop doesn't include the terminator.
-  for (BasicBlock::const_iterator II = BB->begin(), IE = --BB->end();
+  for (BasicBlock::const_iterator II = StartingInst, IE = --BB->end();
        II != IE; ++II) {
+    // If the "Director" remaps the instruction, don't clone it.
+    if (Director) {
+      CloningDirector::CloningAction Action 
+                              = Director->handleInstruction(VMap, II, NewBB);
+      // If the cloning director says stop, we want to stop everything, not
+      // just break out of the loop (which would cause the terminator to be
+      // cloned).  The cloning director is responsible for inserting a proper
+      // terminator into the new basic block in this case.
+      if (Action == CloningDirector::StopCloningBB)
+        return;
+      // If the cloning director says skip, continue to the next instruction.
+      // In this case, the cloning director is responsible for mapping the
+      // skipped instruction to some value that is defined in the new
+      // basic block.
+      if (Action == CloningDirector::SkipInstruction)
+        continue;
+    }
+
     Instruction *NewInst = II->clone();
 
     // Eagerly remap operands to the newly cloned instruction, except for PHI
     // nodes for which we defer processing until we update the CFG.
     if (!isa<PHINode>(NewInst)) {
       RemapInstruction(NewInst, VMap,
                        ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges);
 
if (const AllocaInst *AI = dyn_cast<AllocaInst>(II)) {
       else
         hasDynamicAllocas = true;
     }
   }
   
   // Finally, clone over the terminator.
   const TerminatorInst *OldTI = BB->getTerminator();
   bool TerminatorDone = false;
+  if (Director) {
+    CloningDirector::CloningAction Action 
+                           = Director->handleInstruction(VMap, OldTI, NewBB);
+    // If the cloning director says stop, we want to stop everything, not
+    // just break out of the loop (which would cause the terminator to be
+    // cloned).  The cloning director is responsible for inserting a proper
+    // terminator into the new basic block in this case.
+    if (Action == CloningDirector::StopCloningBB)
+      return;
+    assert(Action != CloningDirector::SkipInstruction && 
+           "SkipInstruction is not valid for terminators.");
+  }
   if (const BranchInst *BI = dyn_cast<BranchInst>(OldTI)) {
     if (BI->isConditional()) {
       // If the condition was a known constant in the callee...
       ConstantInt *Cond = dyn_cast<ConstantInt>(BI->getCondition());
       // Or is a known constant in the caller...
       if (!Cond) {
         Value *V = VMap[BI->getCondition()];
         Cond = dyn_cast_or_null<ConstantInt>(V);
void PruningFunctionCloner::CloneBlock(const BasicBlock *BB,
   if (CodeInfo) {
     CodeInfo->ContainsCalls          |= hasCalls;
     CodeInfo->ContainsDynamicAllocas |= hasDynamicAllocas;
     CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas && 
       BB != &BB->getParent()->front();
   }
 }
 
-/// CloneAndPruneFunctionInto - This works exactly like CloneFunctionInto,
-/// except that it does some simple constant prop and DCE on the fly.  The
-/// effect of this is to copy significantly less code in cases where (for
-/// example) a function call with constant arguments is inlined, and those
-/// constant arguments cause a significant amount of code in the callee to be
-/// dead.  Since this doesn't produce an exact copy of the input, it can't be
-/// used for things like CloneFunction or CloneModule.
-void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc,
+/// CloneAndPruneIntoFromInst - This works like CloneAndPruneFunctionInto, except
+/// that it does not clone the entire function. Instead it starts at an
+/// instruction provided by the caller and copies (and prunes) only the code 
+/// reachable from that instruction.
+void llvm::CloneAndPruneIntoFromInst(Function *NewFunc, const Function *OldFunc,
+                                     const Instruction *StartingInst,
                                      ValueToValueMapTy &VMap,
                                      bool ModuleLevelChanges,
-                                     SmallVectorImpl<ReturnInst*> &Returns,
+                                     SmallVectorImpl<ReturnInst *> &Returns,
                                      const char *NameSuffix, 
                                      ClonedCodeInfo *CodeInfo,
                                      const DataLayout *DL,
-                                     Instruction *TheCall) {
+                                     CloningDirector *Director) {
   assert(NameSuffix && "NameSuffix cannot be null!");
-  
 #ifndef NDEBUG
-  for (Function::const_arg_iterator II = OldFunc->arg_begin(), 
+  // If the cloning starts at the begining of the function, verify that
+  // the function arguments are mapped.
+  if (!StartingInst)
+    for (Function::const_arg_iterator II = OldFunc->arg_begin(),
          E = OldFunc->arg_end(); II != E; ++II)
       assert(VMap.count(II) && "No mapping from source argument specified!");
 #endif
 
   PruningFunctionCloner PFC(NewFunc, OldFunc, VMap, ModuleLevelChanges,
-                            NameSuffix, CodeInfo, DL);
+                            NameSuffix, CodeInfo, DL, Director);
+  const BasicBlock *StartingBB;
+  if (StartingInst)
+    StartingBB = StartingInst->getParent();
+  else {
+    StartingBB = &OldFunc->getEntryBlock();
+    StartingInst = StartingBB->begin();
+  }
 
   // Clone the entry block, and anything recursively reachable from it.
   std::vector<const BasicBlock*> CloneWorklist;
-  CloneWorklist.push_back(&OldFunc->getEntryBlock());
+  PFC.CloneBlock(StartingBB, StartingInst, CloneWorklist);
   while (!CloneWorklist.empty()) {
     const BasicBlock *BB = CloneWorklist.back();
     CloneWorklist.pop_back();
-    PFC.CloneBlock(BB, CloneWorklist);
+    PFC.CloneBlock(BB, BB->begin(), CloneWorklist);
   }
   
   // Loop over all of the basic blocks in the old function.  If the block was
   // reachable, we have cloned it and the old block is now in the value map:
   // insert it into the new function in the right order.  If not, ignore it.
   //
   // Defer PHI resolution until rest of function is resolved.
   SmallVector<const PHINode*, 16> PHIToResolve;
#endif
   for (unsigned Idx = 0, Size = PHIToResolve.size(); Idx != Size; ++Idx)
     if (PHINode *PN = dyn_cast<PHINode>(VMap[PHIToResolve[Idx]]))
       recursivelySimplifyInstruction(PN, DL);
 
   // Now that the inlined function body has been fully constructed, go through
   // and zap unconditional fall-through branches.  This happen all the time when
   // specializing code: code specialization turns conditional branches into
   // uncond branches, and this code folds them.
-  Function::iterator Begin = cast<BasicBlock>(VMap[&OldFunc->getEntryBlock()]);
+  Function::iterator Begin = cast<BasicBlock>(VMap[StartingBB]);
   Function::iterator I = Begin;
   while (I != NewFunc->end()) {
     // Check if this block has become dead during inlining or other
     // simplifications. Note that the first block will appear dead, as it has
     // not yet been wired up properly.
     if (I != Begin && (pred_begin(I) == pred_end(I) ||
                        I->getSinglePredecessor() == I)) {
       BasicBlock *DeadBB = I++;
while (I != NewFunc->end()) {
     Dest->eraseFromParent();
     
     // Do not increment I, iteratively merge all things this block branches to.
   }
 
   // Make a final pass over the basic blocks from theh old function to gather
   // any return instructions which survived folding. We have to do this here
   // because we can iteratively remove and merge returns above.
-  for (Function::iterator I = cast<BasicBlock>(VMap[&OldFunc->getEntryBlock()]),
+  for (Function::iterator I = cast<BasicBlock>(VMap[StartingBB]),
                           E = NewFunc->end();
        I != E; ++I)
     if (ReturnInst *RI = dyn_cast<ReturnInst>(I->getTerminator()))
       Returns.push_back(RI);
 }
+
+
+/// CloneAndPruneFunctionInto - This works exactly like CloneFunctionInto,
+/// except that it does some simple constant prop and DCE on the fly.  The
+/// effect of this is to copy significantly less code in cases where (for
+/// example) a function call with constant arguments is inlined, and those
+/// constant arguments cause a significant amount of code in the callee to be
+/// dead.  Since this doesn't produce an exact copy of the input, it can't be
+/// used for things like CloneFunction or CloneModule.
+void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc,
+                                     ValueToValueMapTy &VMap,
+                                     bool ModuleLevelChanges,
+                                     SmallVectorImpl<ReturnInst*> &Returns,
+                                     const char *NameSuffix, 
+                                     ClonedCodeInfo *CodeInfo,
+                                     const DataLayout *DL,
+                                     Instruction *TheCall) {
+  CloneAndPruneIntoFromInst(NewFunc, OldFunc, OldFunc->front().begin(),
+                            VMap, ModuleLevelChanges, Returns, NameSuffix,
+                            CodeInfo, DL, nullptr);
+}

llvm/trunk/test/CodeGen/X86/cppeh-catch-all.ll

+; RUN: opt -mtriple=x86_64-pc-windows-msvc -winehprepare -S -o - < %s | FileCheck %s
+
+; This test is based on the following code:
+;
+; void test()
+; {
+;   try {
+;     may_throw();
+;   } catch (...) {
+;     handle_exception();
+;   }
+; }
+;
+; Parts of the IR have been hand-edited to simplify the test case.
+; The full IR will be restored when Windows C++ EH support is complete.
+
+; ModuleID = 'catch-all.cpp'
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-pc-windows-msvc"
+
+; Function Attrs: uwtable
+define void @_Z4testv() #0 {
+entry:
+  %exn.slot = alloca i8*
+  %ehselector.slot = alloca i32
+  invoke void @_Z9may_throwv()
+          to label %invoke.cont unwind label %lpad
+
+invoke.cont:                                      ; preds = %entry
+  br label %try.cont
+
+lpad:                                             ; preds = %entry
+  %0 = landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*)
+          catch i8* null
+  %1 = extractvalue { i8*, i32 } %0, 0
+  store i8* %1, i8** %exn.slot
+  %2 = extractvalue { i8*, i32 } %0, 1
+  store i32 %2, i32* %ehselector.slot
+  br label %catch
+
+catch:                                            ; preds = %lpad
+  %exn = load i8** %exn.slot
+  %3 = call i8* @llvm.eh.begincatch(i8* %exn) #3
+  call void @_Z16handle_exceptionv()
+  br label %invoke.cont2
+
+invoke.cont2:                                     ; preds = %catch
+  call void @llvm.eh.endcatch()
+  br label %try.cont
+
+try.cont:                                         ; preds = %invoke.cont2, %invoke.cont
+  ret void
+}
+
+; CHECK: define i8* @_Z4testv.catch(i8*, i8*) {
+; CHECK: catch.entry:
+; CHECK:   %eh.alloc = call i8* @llvm.framerecover(i8* bitcast (void ()* @_Z4testv to i8*), i8* %1)
+; CHECK:   %ehdata = bitcast i8* %eh.alloc to %struct._Z4testv.ehdata*
+; CHECK:   %eh.obj.ptr = getelementptr inbounds %struct._Z4testv.ehdata* %ehdata, i32 0, i32 1
+; CHECK:   %eh.obj = load i8** %eh.obj.ptr
+; CHECK:   call void @_Z16handle_exceptionv()
+; CHECK:   ret i8* blockaddress(@_Z4testv, %try.cont)
+; CHECK: }
+
+declare void @_Z9may_throwv() #1
+
+declare i32 @__CxxFrameHandler3(...)
+
+declare i8* @llvm.eh.begincatch(i8*)
+
+declare void @_Z16handle_exceptionv() #1
+
+declare void @llvm.eh.endcatch()
+
+attributes #0 = { uwtable "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 = { "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 = { noinline noreturn nounwind }
+attributes #3 = { nounwind }
+attributes #4 = { noreturn nounwind }
+
+!llvm.ident = !{!0}
+
+!0 = !{!"clang version 3.7.0 (trunk 226027)"}

llvm/trunk/test/CodeGen/X86/cppeh-catch-scalar.ll

+; RUN: opt -mtriple=x86_64-pc-windows-msvc -winehprepare -S -o - < %s | FileCheck %s
+
+; This test is based on the following code:
+;
+; void test()
+; {
+;   try {
+;     may_throw();
+;   } catch (int) {
+;     handle_int();
+;   }
+; }
+;
+; Parts of the IR have been hand-edited to simplify the test case.
+; The full IR will be restored when Windows C++ EH support is complete.
+
+;ModuleID = 'cppeh-catch-scalar.cpp'
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-pc-windows-msvc"
+
+@_ZTIi = external constant i8*
+
+; Function Attrs: uwtable
+define void @_Z4testv() #0 {
+entry:
+  %exn.slot = alloca i8*
+  %ehselector.slot = alloca i32
+  invoke void @_Z9may_throwv()
+          to label %invoke.cont unwind label %lpad
+
+invoke.cont:                                      ; preds = %entry
+  br label %try.cont
+
+lpad:                                             ; preds = %entry
+  %0 = landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*)
+          catch i8* bitcast (i8** @_ZTIi to i8*)
+  %1 = extractvalue { i8*, i32 } %0, 0
+  store i8* %1, i8** %exn.slot
+  %2 = extractvalue { i8*, i32 } %0, 1
+  store i32 %2, i32* %ehselector.slot
+  br label %catch.dispatch
+
+catch.dispatch:                                   ; preds = %lpad
+  %sel = load i32* %ehselector.slot
+  %3 = call i32 @llvm.eh.typeid.for(i8* bitcast (i8** @_ZTIi to i8*)) #3
+  %matches = icmp eq i32 %sel, %3
+  br i1 %matches, label %catch, label %eh.resume
+
+catch:                                            ; preds = %catch.dispatch
+  %exn11 = load i8** %exn.slot
+  %4 = call i8* @llvm.eh.begincatch(i8* %exn11) #3
+  %5 = bitcast i8* %4 to i32*
+  call void @_Z10handle_intv()
+  br label %invoke.cont2
+
+invoke.cont2:                                     ; preds = %catch
+  call void @llvm.eh.endcatch() #3
+  br label %try.cont
+
+try.cont:                                         ; preds = %invoke.cont2, %invoke.cont
+  ret void
+
+eh.resume:                                        ; preds = %catch.dispatch
+  %exn3 = load i8** %exn.slot
+  %sel4 = load i32* %ehselector.slot
+  %lpad.val = insertvalue { i8*, i32 } undef, i8* %exn3, 0
+  %lpad.val5 = insertvalue { i8*, i32 } %lpad.val, i32 %sel4, 1
+  resume { i8*, i32 } %lpad.val5
+}
+
+; CHECK: define i8* @_Z4testv.catch(i8*, i8*) {
+; CHECK: catch.entry:
+; CHECK:   %eh.alloc = call i8* @llvm.framerecover(i8* bitcast (void ()* @_Z4testv to i8*), i8* %1)
+; CHECK:   %ehdata = bitcast i8* %eh.alloc to %struct._Z4testv.ehdata*
+; CHECK:   %eh.obj.ptr = getelementptr inbounds %struct._Z4testv.ehdata* %ehdata, i32 0, i32 1
+; CHECK:   %eh.obj = load i8** %eh.obj.ptr
+; CHECK:   %2 = bitcast i8* %eh.obj to i32*
+; CHECK:   call void @_Z10handle_intv()
+; CHECK:   ret i8* blockaddress(@_Z4testv, %try.cont)
+; CHECK: }
+
+declare void @_Z9may_throwv() #1
+
+declare i32 @__CxxFrameHandler3(...)
+
+; Function Attrs: nounwind readnone
+declare i32 @llvm.eh.typeid.for(i8*) #2
+
+declare i8* @llvm.eh.begincatch(i8*)
+
+declare void @llvm.eh.endcatch()
+
+declare void @_Z10handle_intv() #1
+
+attributes #0 = { uwtable "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 = { "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 readnone }
+attributes #3 = { nounwind }
+
+!llvm.ident = !{!0}
+
+!0 = !{!"clang version 3.7.0 (trunk 227474) (llvm/trunk 227508)"}