[CodeGen] Simplify the way lifetime of block captures is extended

[ahatanak]

Summary

Rather than pushing inactive cleanups for the block captures at the entry of a full expression and activating them during the creation of the block literal, just call pushLifetimeExtendedDestroy to ensure the cleanups are popped at the end of the scope enclosing the block expression.

I've made sure that the lifetime of block captures in a return statement doesn't get extended beyond the end of the full expression except when the BlockDecl isn't in the cleanup object list of the return statement's ExprWithCleanups (see test22 I added to arc-blocks.m).

This addresses the feedback I got in the following review: https://reviews.llvm.org/D64464#inline-690623

rdar://problem/63996471

[ahatanak]

To clarify my comment, the lifetime of the block capture in test22 gets extended to the end of the enclosing scope, but of course not beyond the end of the return statement.

[rjmccall]

This is a great improvement, thanks!

[SharonXu]

Hi @ahatanak , @rjmccall, we found that this diff causes an issue:
When -fobjc-arc is enabled, if a block is passed as an argument to a function whose corresponding parameter type is && rvalue and attributed as noescape, the object captured by the block can get destructed before the block invocation.
More details and code example is in https://bugs.llvm.org/show_bug.cgi?id=50043 we filed.

We've tried locally reverted this patch, and the test cases mentioned in that bug report passed successfully. I am wondering could you please take a look at the bug report and see if you have a quick fix or something? Thanks!!

clang/lib/CodeGen/CGBlocks.h

public:
   // Offset of the gap caused by block header having a smaller
   // alignment than the alignment of the block descriptor. This
   // is the gap offset before the first capturued field.
   CharUnits BlockHeaderForcedGapOffset;
   // Gap size caused by aligning first field after block header.
   // This could be zero if no forced alignment is required.
   CharUnits BlockHeaderForcedGapSize;
 
-  /// An instruction which dominates the full-expression that the
-  /// block is inside.
-  llvm::Instruction *DominatingIP;
-
   /// The next block in the block-info chain.  Invalid if this block
   /// info is not part of the CGF's block-info chain, which is true
   /// if it corresponds to a global block or a block whose expression
   /// has been encountered.
   CGBlockInfo *NextBlockInfo;
 
   const Capture &getCapture(const VarDecl *var) const {
     return const_cast<CGBlockInfo*>(this)->getCapture(var);

clang/lib/CodeGen/CGBlocks.cpp

 
 using namespace clang;
 using namespace CodeGen;
 
 CGBlockInfo::CGBlockInfo(const BlockDecl *block, StringRef name)
   : Name(name), CXXThisIndex(0), CanBeGlobal(false), NeedsCopyDispose(false),
     HasCXXObject(false), UsesStret(false), HasCapturedVariableLayout(false),
     CapturesNonExternalType(false), LocalAddress(Address::invalid()),
-    StructureType(nullptr), Block(block), DominatingIP(nullptr) {
+    StructureType(nullptr), Block(block) {
 
   // Skip asm prefix, if any.  'name' is usually taken directly from
   // the mangled name of the enclosing function.
   if (!name.empty() && name[0] == '\01')
     name = name.substr(1);
 }
 
 // Anchor the vtable to this translation unit.
for (SmallVectorImpl<BlockLayoutChunk>::iterator
     blockSize += li->Size;
     endAlign = getLowBit(blockSize);
   }
 
   info.StructureType =
     llvm::StructType::get(CGM.getLLVMContext(), elementTypes, true);
 }
 
-/// Enter the scope of a block.  This should be run at the entrance to
-/// a full-expression so that the block's cleanups are pushed at the
-/// right place in the stack.
-static void enterBlockScope(CodeGenFunction &CGF, BlockDecl *block) {
-  assert(CGF.HaveInsertPoint());
-
-  // Allocate the block info and place it at the head of the list.
-  CGBlockInfo &blockInfo =
-    *new CGBlockInfo(block, CGF.CurFn->getName());
-  blockInfo.NextBlockInfo = CGF.FirstBlockInfo;
-  CGF.FirstBlockInfo = &blockInfo;
-
-  // Compute information about the layout, etc., of this block,
-  // pushing cleanups as necessary.
-  computeBlockInfo(CGF.CGM, &CGF, blockInfo);
-
-  // Nothing else to do if it can be global.
-  if (blockInfo.CanBeGlobal) return;
-
-  // Make the allocation for the block.
-  blockInfo.LocalAddress = CGF.CreateTempAlloca(blockInfo.StructureType,
-                                                blockInfo.BlockAlign, "block");
-
-  // If there are cleanups to emit, enter them (but inactive).
-  if (!blockInfo.NeedsCopyDispose) return;
-
-  // Walk through the captures (in order) and find the ones not
-  // captured by constant.
-  for (const auto &CI : block->captures()) {
-    // Ignore __block captures; there's nothing special in the
-    // on-stack block that we need to do for them.
-    if (CI.isByRef()) continue;
-
-    // Ignore variables that are constant-captured.
-    const VarDecl *variable = CI.getVariable();
-    CGBlockInfo::Capture &capture = blockInfo.getCapture(variable);
-    if (capture.isConstant()) continue;
-
-    // Ignore objects that aren't destructed.
-    QualType VT = getCaptureFieldType(CGF, CI);
-    QualType::DestructionKind dtorKind = VT.isDestructedType();
-    if (dtorKind == QualType::DK_none) continue;
-
-    CodeGenFunction::Destroyer *destroyer;
-
-    // Block captures count as local values and have imprecise semantics.
-    // They also can't be arrays, so need to worry about that.
-    //
-    // For const-qualified captures, emit clang.arc.use to ensure the captured
-    // object doesn't get released while we are still depending on its validity
-    // within the block.
-    if (VT.isConstQualified() &&
-        VT.getObjCLifetime() == Qualifiers::OCL_Strong &&
-        CGF.CGM.getCodeGenOpts().OptimizationLevel != 0) {
-      assert(CGF.CGM.getLangOpts().ObjCAutoRefCount &&
-             "expected ObjC ARC to be enabled");
-      destroyer = CodeGenFunction::emitARCIntrinsicUse;
-    } else if (dtorKind == QualType::DK_objc_strong_lifetime) {
-      destroyer = CodeGenFunction::destroyARCStrongImprecise;
-    } else {
-      destroyer = CGF.getDestroyer(dtorKind);
-    }
-
-    // GEP down to the address.
-    Address addr =
-        CGF.Builder.CreateStructGEP(blockInfo.LocalAddress, capture.getIndex());
-
-    // We can use that GEP as the dominating IP.
-    if (!blockInfo.DominatingIP)
-      blockInfo.DominatingIP = cast<llvm::Instruction>(addr.getPointer());
-
-    CleanupKind cleanupKind = InactiveNormalCleanup;
-    bool useArrayEHCleanup = CGF.needsEHCleanup(dtorKind);
-    if (useArrayEHCleanup)
-      cleanupKind = InactiveNormalAndEHCleanup;
-
-    CGF.pushDestroy(cleanupKind, addr, VT,
-                    destroyer, useArrayEHCleanup);
-
-    // Remember where that cleanup was.
-    capture.setCleanup(CGF.EHStack.stable_begin());
-  }
-}
-
-/// Enter a full-expression with a non-trivial number of objects to
-/// clean up.
-void CodeGenFunction::enterNonTrivialFullExpression(const FullExpr *E) {
-  if (const auto EWC = dyn_cast<ExprWithCleanups>(E)) {
-    assert(EWC->getNumObjects() != 0);
-    for (const ExprWithCleanups::CleanupObject &C : EWC->getObjects())
-      if (auto *BD = C.dyn_cast<BlockDecl *>())
-        enterBlockScope(*this, BD);
-  }
-}
-
-/// Find the layout for the given block in a linked list and remove it.
-static CGBlockInfo *findAndRemoveBlockInfo(CGBlockInfo **head,
-                                           const BlockDecl *block) {
-  while (true) {
-    assert(head && *head);
-    CGBlockInfo *cur = *head;
-
-    // If this is the block we're looking for, splice it out of the list.
-    if (cur->getBlockDecl() == block) {
-      *head = cur->NextBlockInfo;
-      return cur;
-    }
-
-    head = &cur->NextBlockInfo;
-  }
-}
-
-/// Destroy a chain of block layouts.
-void CodeGenFunction::destroyBlockInfos(CGBlockInfo *head) {
-  assert(head && "destroying an empty chain");
-  do {
-    CGBlockInfo *cur = head;
-    head = cur->NextBlockInfo;
-    delete cur;
-  } while (head != nullptr);
-}
-
 /// Emit a block literal expression in the current function.
 llvm::Value *CodeGenFunction::EmitBlockLiteral(const BlockExpr *blockExpr) {
   // If the block has no captures, we won't have a pre-computed
   // layout for it.
-  if (!blockExpr->getBlockDecl()->hasCaptures()) {
+  if (!blockExpr->getBlockDecl()->hasCaptures())
     // The block literal is emitted as a global variable, and the block invoke
     // function has to be extracted from its initializer.
-    if (llvm::Constant *Block = CGM.getAddrOfGlobalBlockIfEmitted(blockExpr)) {
+    if (llvm::Constant *Block = CGM.getAddrOfGlobalBlockIfEmitted(blockExpr))
       return Block;
-    }
-    CGBlockInfo blockInfo(blockExpr->getBlockDecl(), CurFn->getName());
-    computeBlockInfo(CGM, this, blockInfo);
-    blockInfo.BlockExpression = blockExpr;
-    return EmitBlockLiteral(blockInfo);
-  }
+
+  if (!blockInfo.CanBeGlobal)
+    blockInfo.LocalAddress = CreateTempAlloca(blockInfo.StructureType,
+                                              blockInfo.BlockAlign, "block");
 
-  // Find the block info for this block and take ownership of it.
-  std::unique_ptr<CGBlockInfo> blockInfo;
-  blockInfo.reset(findAndRemoveBlockInfo(&FirstBlockInfo,
-                                         blockExpr->getBlockDecl()));
-
-  blockInfo->BlockExpression = blockExpr;
-  return EmitBlockLiteral(*blockInfo);
-}
-
 llvm::Value *CodeGenFunction::EmitBlockLiteral(const CGBlockInfo &blockInfo) {
   bool IsOpenCL = CGM.getContext().getLangOpts().OpenCL;
   auto GenVoidPtrTy =
       IsOpenCL ? CGM.getOpenCLRuntime().getGenericVoidPointerType() : VoidPtrTy;
   LangAS GenVoidPtrAddr = IsOpenCL ? LangAS::opencl_generic : LangAS::Default;
   auto GenVoidPtrSize = CharUnits::fromQuantity(
       CGM.getTarget().getPointerWidth(
           CGM.getContext().getTargetAddressSpace(GenVoidPtrAddr)) /
// Otherwise, fake up a POD copy into the block field.
       // FIXME: Pass a specific location for the expr init so that the store is
       // attributed to a reasonable location - otherwise it may be attributed to
       // locations of subexpressions in the initialization.
       EmitExprAsInit(&l2r, &BlockFieldPseudoVar,
                      MakeAddrLValue(blockField, type, AlignmentSource::Decl),
                      /*captured by init*/ false);
     }
 
-    // Activate the cleanup if layout pushed one.
-    if (!CI.isByRef()) {
-      EHScopeStack::stable_iterator cleanup = capture.getCleanup();
-      if (cleanup.isValid())
-        ActivateCleanupBlock(cleanup, blockInfo.DominatingIP);
+    // Push a cleanup for the capture if necessary.
+    if (!blockInfo.NeedsCopyDispose)
+      continue;
+
+    // Ignore __block captures; there's nothing special in the on-stack block
+    // that we need to do for them.
+    if (CI.isByRef())
+      continue;
+
+    // Ignore objects that aren't destructed.
+    QualType::DestructionKind dtorKind = type.isDestructedType();
+    if (dtorKind == QualType::DK_none)
+      continue;
+
+    CodeGenFunction::Destroyer *destroyer;
+
+    // Block captures count as local values and have imprecise semantics.
+    // They also can't be arrays, so need to worry about that.
+    //
+    // For const-qualified captures, emit clang.arc.use to ensure the captured
+    // object doesn't get released while we are still depending on its validity
+    // within the block.
+    if (type.isConstQualified() &&
+        type.getObjCLifetime() == Qualifiers::OCL_Strong &&
+        CGM.getCodeGenOpts().OptimizationLevel != 0) {
+      assert(CGM.getLangOpts().ObjCAutoRefCount &&
+             "expected ObjC ARC to be enabled");
+      destroyer = emitARCIntrinsicUse;
+    } else if (dtorKind == QualType::DK_objc_strong_lifetime) {
+      destroyer = destroyARCStrongImprecise;
+    } else {
+      destroyer = getDestroyer(dtorKind);
     }
+
+    CleanupKind cleanupKind = NormalCleanup;
+    bool useArrayEHCleanup = needsEHCleanup(dtorKind);
+    if (useArrayEHCleanup)
+      cleanupKind = NormalAndEHCleanup;
+
+    // Extend the lifetime of the capture to the end of the scope enclosing the
+    // block expression except when the block decl is in the list of RetExpr's
+    // cleanup objects, in which case its lifetime ends after the full
+    // expression.
+    auto IsBlockDeclInRetExpr = [&]() {
+      auto *EWC = llvm::dyn_cast_or_null<ExprWithCleanups>(RetExpr);
+      if (EWC)
+        for (auto &C : EWC->getObjects())
+          if (auto *BD = C.dyn_cast<BlockDecl *>())
+            if (BD == blockDecl)
+              return true;
+      return false;
+    };
+
+    if (IsBlockDeclInRetExpr())
+      pushDestroy(cleanupKind, blockField, type, destroyer, useArrayEHCleanup);
+    else
+      pushLifetimeExtendedDestroy(cleanupKind, blockField, type, destroyer,
+                                  useArrayEHCleanup);
   }
 
   // Cast to the converted block-pointer type, which happens (somewhat
   // unfortunately) to be a pointer to function type.
   llvm::Value *result = Builder.CreatePointerCast(
       blockAddr.getPointer(), ConvertType(blockInfo.getBlockExpr()->getType()));
 
   if (IsOpenCL) {

clang/lib/CodeGen/CGCleanup.h

public:
   static size_t getSizeForCleanupSize(size_t Size) {
     return sizeof(EHCleanupScope) + Size;
   }
 
   size_t getAllocatedSize() const {
     return sizeof(EHCleanupScope) + CleanupBits.CleanupSize;
   }
 
-  EHCleanupScope(bool isNormal, bool isEH, bool isActive,
-                 unsigned cleanupSize, unsigned fixupDepth,
+  EHCleanupScope(bool isNormal, bool isEH, unsigned cleanupSize,
+                 unsigned fixupDepth,
                  EHScopeStack::stable_iterator enclosingNormal,
                  EHScopeStack::stable_iterator enclosingEH)
       : EHScope(EHScope::Cleanup, enclosingEH),
         EnclosingNormal(enclosingNormal), NormalBlock(nullptr),
         ActiveFlag(nullptr), ExtInfo(nullptr), FixupDepth(fixupDepth) {
     CleanupBits.IsNormalCleanup = isNormal;
     CleanupBits.IsEHCleanup = isEH;
-    CleanupBits.IsActive = isActive;
+    CleanupBits.IsActive = true;
     CleanupBits.IsLifetimeMarker = false;
     CleanupBits.TestFlagInNormalCleanup = false;
     CleanupBits.TestFlagInEHCleanup = false;
     CleanupBits.CleanupSize = cleanupSize;
 
     assert(CleanupBits.CleanupSize == cleanupSize && "cleanup size overflow");
   }
 

clang/lib/CodeGen/CGCleanup.cpp

EHScopeStack::getInnermostActiveNormalCleanup() const {
   return stable_end();
 }
 
 
 void *EHScopeStack::pushCleanup(CleanupKind Kind, size_t Size) {
   char *Buffer = allocate(EHCleanupScope::getSizeForCleanupSize(Size));
   bool IsNormalCleanup = Kind & NormalCleanup;
   bool IsEHCleanup = Kind & EHCleanup;
-  bool IsActive = !(Kind & InactiveCleanup);
   bool IsLifetimeMarker = Kind & LifetimeMarker;
   EHCleanupScope *Scope =
     new (Buffer) EHCleanupScope(IsNormalCleanup,
                                 IsEHCleanup,
-                                IsActive,
                                 Size,
                                 BranchFixups.size(),
                                 InnermostNormalCleanup,
                                 InnermostEHScope);
   if (IsNormalCleanup)
     InnermostNormalCleanup = stable_begin();
   if (IsEHCleanup)
     InnermostEHScope = stable_begin();

clang/lib/CodeGen/CGDecl.cpp

if (!lifetime) {
     return;
   }
 
   if (const CXXDefaultInitExpr *DIE = dyn_cast<CXXDefaultInitExpr>(init))
     init = DIE->getExpr();
 
   // If we're emitting a value with lifetime, we have to do the
   // initialization *before* we leave the cleanup scopes.
-  if (const FullExpr *fe = dyn_cast<FullExpr>(init)) {
-    enterFullExpression(fe);
+  if (const FullExpr *fe = dyn_cast<FullExpr>(init))
     init = fe->getSubExpr();
-  }
+
   CodeGenFunction::RunCleanupsScope Scope(*this);
 
   // We have to maintain the illusion that the variable is
   // zero-initialized.  If the variable might be accessed in its
   // initializer, zero-initialize before running the initializer, then
   // actually perform the initialization with an assign.
   bool accessedByInit = false;
   if (lifetime != Qualifiers::OCL_ExplicitNone)

clang/lib/CodeGen/CGExpr.cpp

case Expr::CXXBindTemporaryExprClass:
     return EmitCXXBindTemporaryLValue(cast<CXXBindTemporaryExpr>(E));
   case Expr::CXXUuidofExprClass:
     return EmitCXXUuidofLValue(cast<CXXUuidofExpr>(E));
   case Expr::LambdaExprClass:
     return EmitAggExprToLValue(E);
 
   case Expr::ExprWithCleanupsClass: {
     const auto *cleanups = cast<ExprWithCleanups>(E);
-    enterFullExpression(cleanups);
     RunCleanupsScope Scope(*this);
     LValue LV = EmitLValue(cleanups->getSubExpr());
     if (LV.isSimple()) {
       // Defend against branches out of gnu statement expressions surrounded by
       // cleanups.
       llvm::Value *V = LV.getPointer(*this);
       Scope.ForceCleanup({&V});
       return LValue::MakeAddr(Address(V, LV.getAlignment()), LV.getType(),

clang/lib/CodeGen/CGExprAgg.cpp

for (unsigned i = Cleanups.size(); i != 0; --i)
     CGF.DeactivateCleanupBlock(Cleanups[i-1], CleanupDominator);
 
   // Destroy the placeholder if we made one.
   if (CleanupDominator)
     CleanupDominator->eraseFromParent();
 }
 
 void AggExprEmitter::VisitExprWithCleanups(ExprWithCleanups *E) {
-  CGF.enterFullExpression(E);
   CodeGenFunction::RunCleanupsScope cleanups(CGF);
   Visit(E->getSubExpr());
 }
 
 void AggExprEmitter::VisitCXXScalarValueInitExpr(CXXScalarValueInitExpr *E) {
   QualType T = E->getType();
   AggValueSlot Slot = EnsureSlot(T);
   EmitNullInitializationToLValue(CGF.MakeAddrLValue(Slot.getAddress(), T));

clang/lib/CodeGen/CGExprComplex.cpp

ComplexPairTy VisitCXXDefaultArgExpr(CXXDefaultArgExpr *DAE) {
     CodeGenFunction::CXXDefaultArgExprScope Scope(CGF, DAE);
     return Visit(DAE->getExpr());
   }
   ComplexPairTy VisitCXXDefaultInitExpr(CXXDefaultInitExpr *DIE) {
     CodeGenFunction::CXXDefaultInitExprScope Scope(CGF, DIE);
     return Visit(DIE->getExpr());
   }
   ComplexPairTy VisitExprWithCleanups(ExprWithCleanups *E) {
-    CGF.enterFullExpression(E);
     CodeGenFunction::RunCleanupsScope Scope(CGF);
     ComplexPairTy Vals = Visit(E->getSubExpr());
     // Defend against dominance problems caused by jumps out of expression
     // evaluation through the shared cleanup block.
     Scope.ForceCleanup({&Vals.first, &Vals.second});
     return Vals;
   }
   ComplexPairTy VisitCXXScalarValueInitExpr(CXXScalarValueInitExpr *E) {

clang/lib/CodeGen/CGExprScalar.cpp

Address RetAlloca = CGF.EmitCompoundStmt(*E->getSubStmt(),
                                            !E->getType()->isVoidType());
   if (!RetAlloca.isValid())
     return nullptr;
   return CGF.EmitLoadOfScalar(CGF.MakeAddrLValue(RetAlloca, E->getType()),
                               E->getExprLoc());
 }
 
 Value *ScalarExprEmitter::VisitExprWithCleanups(ExprWithCleanups *E) {
-  CGF.enterFullExpression(E);
   CodeGenFunction::RunCleanupsScope Scope(CGF);
   Value *V = Visit(E->getSubExpr());
   // Defend against dominance problems caused by jumps out of expression
   // evaluation through the shared cleanup block.
   Scope.ForceCleanup({&V});
   return V;
 }
 

clang/lib/CodeGen/CGObjC.cpp

 
 /// EmitARCRetainScalarExpr - Semantically equivalent to
 /// EmitARCRetainObject(e->getType(), EmitScalarExpr(e)), but making a
 /// best-effort attempt to peephole expressions that naturally produce
 /// retained objects.
 llvm::Value *CodeGenFunction::EmitARCRetainScalarExpr(const Expr *e) {
   // The retain needs to happen within the full-expression.
   if (const ExprWithCleanups *cleanups = dyn_cast<ExprWithCleanups>(e)) {
-    enterFullExpression(cleanups);
     RunCleanupsScope scope(*this);
     return EmitARCRetainScalarExpr(cleanups->getSubExpr());
   }
 
   TryEmitResult result = tryEmitARCRetainScalarExpr(*this, e);
   llvm::Value *value = result.getPointer();
   if (!result.getInt())
     value = EmitARCRetain(e->getType(), value);
   return value;
 }
 
 llvm::Value *
 CodeGenFunction::EmitARCRetainAutoreleaseScalarExpr(const Expr *e) {
   // The retain needs to happen within the full-expression.
   if (const ExprWithCleanups *cleanups = dyn_cast<ExprWithCleanups>(e)) {
-    enterFullExpression(cleanups);
     RunCleanupsScope scope(*this);
     return EmitARCRetainAutoreleaseScalarExpr(cleanups->getSubExpr());
   }
 
   TryEmitResult result = tryEmitARCRetainScalarExpr(*this, e);
   llvm::Value *value = result.getPointer();
   if (result.getInt())
     value = EmitARCAutorelease(value);

 
 /// EmitARCUnsafeUnretainedScalarExpr - Semantically equivalent to
 /// immediately releasing the resut of EmitARCRetainScalarExpr, but
 /// avoiding any spurious retains, including by performing reclaims
 /// with objc_unsafeClaimAutoreleasedReturnValue.
 llvm::Value *CodeGenFunction::EmitARCUnsafeUnretainedScalarExpr(const Expr *e) {
   // Look through full-expressions.
   if (const ExprWithCleanups *cleanups = dyn_cast<ExprWithCleanups>(e)) {
-    enterFullExpression(cleanups);
     RunCleanupsScope scope(*this);
     return emitARCUnsafeUnretainedScalarExpr(*this, cleanups->getSubExpr());
   }
 
   return emitARCUnsafeUnretainedScalarExpr(*this, e);
 }
 
 std::pair<LValue,llvm::Value*>

clang/lib/CodeGen/CGStmt.cpp

if (RV.isScalar()) {
     EmitAggregateCopy(Dest, Src, Ty, getOverlapForReturnValue());
   } else {
     EmitStoreOfComplex(RV.getComplexVal(), MakeAddrLValue(ReturnValue, Ty),
                        /*init*/ true);
   }
   EmitBranchThroughCleanup(ReturnBlock);
 }
 
+namespace {
+// RAII struct used to save and restore a return statment's result expression.
+struct SaveRetExprRAII {
+  SaveRetExprRAII(const Expr *RetExpr, CodeGenFunction &CGF)
+      : OldRetExpr(CGF.RetExpr), CGF(CGF) {
+    CGF.RetExpr = RetExpr;
+  }
+  ~SaveRetExprRAII() { CGF.RetExpr = OldRetExpr; }
+  const Expr *OldRetExpr;
+  CodeGenFunction &CGF;
+};
+} // namespace
+
 /// EmitReturnStmt - Note that due to GCC extensions, this can have an operand
 /// if the function returns void, or may be missing one if the function returns
 /// non-void.  Fun stuff :).
 void CodeGenFunction::EmitReturnStmt(const ReturnStmt &S) {
   if (requiresReturnValueCheck()) {
     llvm::Constant *SLoc = EmitCheckSourceLocation(S.getBeginLoc());
     auto *SLocPtr =
         new llvm::GlobalVariable(CGM.getModule(), SLoc->getType(), false,
void CodeGenFunction::EmitReturnStmt(const ReturnStmt &S) {
   if (IsOutlinedSEHHelper) {
     Builder.CreateUnreachable();
     Builder.ClearInsertionPoint();
   }
 
   // Emit the result value, even if unused, to evaluate the side effects.
   const Expr *RV = S.getRetValue();
 
-  // Treat block literals in a return expression as if they appeared
-  // in their own scope.  This permits a small, easily-implemented
-  // exception to our over-conservative rules about not jumping to
-  // statements following block literals with non-trivial cleanups.
+  // Record the result expression of the return statement. The recorded
+  // expression is used to determine whether a block capture's lifetime should
+  // end at the end of the full expression as opposed to the end of the scope
+  // enclosing the block expression.
+  //
+  // This permits a small, easily-implemented exception to our over-conservative
+  // rules about not jumping to statements following block literals with
+  // non-trivial cleanups.
+  SaveRetExprRAII SaveRetExpr(RV, *this);
+
   RunCleanupsScope cleanupScope(*this);
-  if (const FullExpr *fe = dyn_cast_or_null<FullExpr>(RV)) {
-    enterFullExpression(fe);
+  if (const FullExpr *fe = dyn_cast_or_null<FullExpr>(RV))
     RV = fe->getSubExpr();
-  }
 
   // FIXME: Clean this up by using an LValue for ReturnTemp,
   // EmitStoreThroughLValue, and EmitAnyExpr.
   // Check if the NRVO candidate was not globalized in OpenMP mode.
   if (getLangOpts().ElideConstructors && S.getNRVOCandidate() &&
       S.getNRVOCandidate()->isNRVOVariable() &&
       (!getLangOpts().OpenMP ||
        !CGM.getOpenMPRuntime()

clang/lib/CodeGen/CGStmtOpenMP.cpp

if (DefaultOrder == llvm::AtomicOrdering::Monotonic ||
       } else if (Kind == OMPC_read) {
         assert(Kind == OMPC_read && "Unexpected atomic kind.");
         AO = llvm::AtomicOrdering::Acquire;
       }
     }
   }
 
   const Stmt *CS = S.getInnermostCapturedStmt()->IgnoreContainers();
-  if (const auto *FE = dyn_cast<FullExpr>(CS))
-    enterFullExpression(FE);
-  // Processing for statements under 'atomic capture'.
-  if (const auto *Compound = dyn_cast<CompoundStmt>(CS)) {
-    for (const Stmt *C : Compound->body()) {
-      if (const auto *FE = dyn_cast<FullExpr>(C))
-        enterFullExpression(FE);
-    }
-  }
 
   auto &&CodeGen = [&S, Kind, AO, CS](CodeGenFunction &CGF,
                                             PrePostActionTy &) {
     CGF.EmitStopPoint(CS);
     emitOMPAtomicExpr(CGF, Kind, AO, S.isPostfixUpdate(), S.getX(), S.getV(),
                       S.getExpr(), S.getUpdateExpr(), S.isXLHSInRHSPart(),
                       S.getBeginLoc());
   };

clang/lib/CodeGen/CodeGenFunction.h

public:
   /// ReturnValue - The temporary alloca to hold the return
   /// value. This is invalid iff the function has no return value.
   Address ReturnValue = Address::invalid();
 
   /// ReturnValuePointer - The temporary alloca to hold a pointer to sret.
   /// This is invalid if sret is not in use.
   Address ReturnValuePointer = Address::invalid();
 
+  /// If a return statement is being visited, this holds the return statment's
+  /// result expression.
+  const Expr *RetExpr = nullptr;
+
   /// Return true if a label was seen in the current scope.
   bool hasLabelBeenSeenInCurrentScope() const {
     if (CurLexicalScope)
       return CurLexicalScope->hasLabels();
     return !LabelMap.empty();
   }
 
   /// AllocaInsertPoint - This is an instruction in the entry block before which
struct LifetimeExtendedCleanupHeader {
     bool isConditional() const { return IsConditional; }
   };
 
   /// i32s containing the indexes of the cleanup destinations.
   Address NormalCleanupDest = Address::invalid();
 
   unsigned NextCleanupDestIndex = 1;
 
-  /// FirstBlockInfo - The head of a singly-linked-list of block layouts.
-  CGBlockInfo *FirstBlockInfo = nullptr;
-
   /// EHResumeBlock - Unified block containing a call to llvm.eh.resume.
   llvm::BasicBlock *EHResumeBlock = nullptr;
 
   /// The exception slot.  All landing pads write the current exception pointer
   /// into this alloca.
   llvm::Value *ExceptionSlot = nullptr;
 
   /// The selector slot.  Under the MandatoryCleanup model, all landing pads
public:
   //                                  Block Bits
   //===--------------------------------------------------------------------===//
 
   /// Emit block literal.
   /// \return an LLVM value which is a pointer to a struct which contains
   /// information about the block, including the block invoke function, the
   /// captured variables, etc.
   llvm::Value *EmitBlockLiteral(const BlockExpr *);
-  static void destroyBlockInfos(CGBlockInfo *info);
 
   llvm::Function *GenerateBlockFunction(GlobalDecl GD,
                                         const CGBlockInfo &Info,
                                         const DeclMapTy &ldm,
                                         bool IsLambdaConversionToBlock,
                                         bool BuildGlobalBlock);
 
   /// Check if \p T is a C++ class that has a destructor that can throw.
void GenerateCXXGlobalVarDeclInitFunc(llvm::Function *Fn,
                                         const VarDecl *D,
                                         llvm::GlobalVariable *Addr,
                                         bool PerformInit);
 
   void EmitCXXConstructExpr(const CXXConstructExpr *E, AggValueSlot Dest);
 
   void EmitSynthesizedCXXCopyCtor(Address Dest, Address Src, const Expr *Exp);
 
-  void enterFullExpression(const FullExpr *E) {
-    if (const auto *EWC = dyn_cast<ExprWithCleanups>(E))
-      if (EWC->getNumObjects() == 0)
-        return;
-    enterNonTrivialFullExpression(E);
-  }
-  void enterNonTrivialFullExpression(const FullExpr *E);
-
   void EmitCXXThrowExpr(const CXXThrowExpr *E, bool KeepInsertionPoint = true);
 
   RValue EmitAtomicExpr(AtomicExpr *E);
 
   //===--------------------------------------------------------------------===//
   //                         Annotations Emission
   //===--------------------------------------------------------------------===//
 

clang/lib/CodeGen/CodeGenFunction.cpp

CodeGenFunction::CodeGenFunction(CodeGenModule &cgm, bool suppressNewContext)
 
   SetFastMathFlags(CurFPFeatures);
   SetFPModel();
 }
 
 CodeGenFunction::~CodeGenFunction() {
   assert(LifetimeExtendedCleanupStack.empty() && "failed to emit a cleanup");
 
-  // If there are any unclaimed block infos, go ahead and destroy them
-  // now.  This can happen if IR-gen gets clever and skips evaluating
-  // something.
-  if (FirstBlockInfo)
-    destroyBlockInfos(FirstBlockInfo);
-
   if (getLangOpts().OpenMP && CurFn)
     CGM.getOpenMPRuntime().functionFinished(*this);
 
   // If we have an OpenMPIRBuilder we want to finalize functions (incl.
   // outlining etc) at some point. Doing it once the function codegen is done
   // seems to be a reasonable spot. We do it here, as opposed to the deletion
   // time of the CodeGenModule, because we have to ensure the IR has not yet
   // been "emitted" to the outside, thus, modifications are still sensible.

clang/lib/CodeGen/EHScopeStack.h

enum CleanupKind : unsigned {
   EHCleanup = 0x1,
 
   /// Denotes a cleanup that should run when a scope is exited using normal
   /// control flow (falling off the end of the scope, return, goto, ...).
   NormalCleanup = 0x2,
 
   NormalAndEHCleanup = EHCleanup | NormalCleanup,
 
-  InactiveCleanup = 0x4,
-  InactiveEHCleanup = EHCleanup | InactiveCleanup,
-  InactiveNormalCleanup = NormalCleanup | InactiveCleanup,
-  InactiveNormalAndEHCleanup = NormalAndEHCleanup | InactiveCleanup,
-
   LifetimeMarker = 0x8,
   NormalEHLifetimeMarker = LifetimeMarker | NormalAndEHCleanup,
 };
 
 /// A stack of scopes which respond to exceptions, including cleanups
 /// and catch blocks.
 class EHScopeStack {
 public:

clang/test/CodeGenCXX/blocks-cxx11.cpp

void test(A a) {
     auto lambda = [a]() {
       takeBlock(^{ a.use(); });
     };
     lambda(); // make sure we emit the invocation function
   }
   // CHECK-LABEL:    define internal void @"_ZZN20test_block_in_lambda4testENS_1AEENK3$_0clEv"(
   // CHECK:      [[BLOCK:%.*]] = alloca [[BLOCK_T:<{.*}>]], align 8
   // CHECK:      [[THIS:%.*]] = load [[LAMBDA_T:%.*]]*, [[LAMBDA_T:%.*]]**
-  // CHECK:      [[TO_DESTROY:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
-  // CHECK:      [[T0:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
+  // CHECK:      [[BLOCK_CAPTURED:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
   // CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[LAMBDA_T]], [[LAMBDA_T]]* [[THIS]], i32 0, i32 0
-  // CHECK-NEXT: call void @_ZN20test_block_in_lambda1AC1ERKS0_({{.*}}* [[T0]], {{.*}}* nonnull align {{[0-9]+}} dereferenceable({{[0-9]+}}) [[T1]])
+  // CHECK-NEXT: call void @_ZN20test_block_in_lambda1AC1ERKS0_({{.*}}* [[BLOCK_CAPTURED]], {{.*}}* nonnull align {{[0-9]+}} dereferenceable({{[0-9]+}}) [[T1]])
   // CHECK-NEXT: [[T0:%.*]] = bitcast [[BLOCK_T]]* [[BLOCK]] to void ()*
   // CHECK-NEXT: call void @_ZN20test_block_in_lambda9takeBlockEU13block_pointerFvvE(void ()* [[T0]])
-  // CHECK-NEXT: call void @_ZN20test_block_in_lambda1AD1Ev({{.*}}* [[TO_DESTROY]])
+  // CHECK-NEXT: call void @_ZN20test_block_in_lambda1AD1Ev({{.*}}* [[BLOCK_CAPTURED]])
   // CHECK-NEXT: ret void
 }

clang/test/CodeGenCXX/blocks.cpp

namespace test5 {
   }
 
   // CHECK-LABEL:    define void @_ZN5test54testEb(
   // CHECK:      [[COND:%.*]] = alloca i8
   // CHECK-NEXT: [[X:%.*]] = alloca [[A:%.*]], align 4
   // CHECK-NEXT: [[B:%.*]] = alloca void ()*, align 8
   // CHECK-NEXT: [[BLOCK:%.*]] = alloca [[BLOCK_T:.*]], align 8
   // CHECK-NEXT: [[CLEANUP_ACTIVE:%.*]] = alloca i1
+  // CHECK-NEXT: [[COND_CLEANUP_SAVE:%.*]] = alloca [[A]]*, align 8
   // CHECK-NEXT: [[T0:%.*]] = zext i1
   // CHECK-NEXT: store i8 [[T0]], i8* [[COND]], align 1
   // CHECK-NEXT: call void @_ZN5test51AC1Ev([[A]]* [[X]])
-  // CHECK-NEXT: [[CLEANUP_ADDR:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
   // CHECK-NEXT: [[T0:%.*]] = load i8, i8* [[COND]], align 1
   // CHECK-NEXT: [[T1:%.*]] = trunc i8 [[T0]] to i1
   // CHECK-NEXT: store i1 false, i1* [[CLEANUP_ACTIVE]]
   // CHECK-NEXT: br i1 [[T1]],
 
   // CHECK-NOT:  br
   // CHECK:      [[CAPTURE:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
   // CHECK-NEXT: call void @_ZN5test51AC1ERKS0_([[A]]* [[CAPTURE]], [[A]]* nonnull align {{[0-9]+}} dereferenceable({{[0-9]+}}) [[X]])
   // CHECK-NEXT: store i1 true, i1* [[CLEANUP_ACTIVE]]
+  // CHECK-NEXT: store [[A]]* [[CAPTURE]], [[A]]** [[COND_CLEANUP_SAVE]], align 8
   // CHECK-NEXT: bitcast [[BLOCK_T]]* [[BLOCK]] to void ()*
   // CHECK-NEXT: br label
   // CHECK:      br label
   // CHECK:      phi
   // CHECK-NEXT: store
   // CHECK-NEXT: load
   // CHECK-NEXT: call void @_ZN5test511doWithBlockEU13block_pointerFvvE(
   // CHECK-NEXT: [[T0:%.*]] = load i1, i1* [[CLEANUP_ACTIVE]]
   // CHECK-NEXT: br i1 [[T0]]
-  // CHECK:      call void @_ZN5test51AD1Ev([[A]]* [[CLEANUP_ADDR]])
+  // CHECK:      [[T3:%.*]] = load [[A]]*, [[A]]** [[COND_CLEANUP_SAVE]], align 8
+  // CHECK-NEXT: call void @_ZN5test51AD1Ev([[A]]* [[T3]])
   // CHECK-NEXT: br label
   // CHECK:      call void @_ZN5test51AD1Ev([[A]]* [[X]])
   // CHECK-NEXT: ret void
 }
 
 namespace test6 {
   struct A {
     A();

clang/test/CodeGenObjC/arc-blocks.m

 }
 
 void test2(id x) {
 // CHECK-LABEL:    define void @test2(
 // CHECK:      [[X:%.*]] = alloca i8*,
 // CHECK-NEXT: [[BLOCK:%.*]] = alloca [[BLOCK_T:<{.*}>]],
 // CHECK-NEXT: [[PARM:%.*]] = call i8* @llvm.objc.retain(i8* {{%.*}})
 // CHECK-NEXT: store i8* [[PARM]], i8** [[X]]
-// CHECK-NEXT: [[SLOTREL:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK:      [[SLOT:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[X]],
 // CHECK-NEXT: [[T1:%.*]] = call i8* @llvm.objc.retain(i8* [[T0]])
 // CHECK-NEXT: store i8* [[T1]], i8** [[SLOT]],
 // CHECK-NEXT: bitcast
 // CHECK-NEXT: call void @test2_helper(
-// CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[SLOTREL]]
+// CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[SLOT]]
 // CHECK-NEXT: call void @llvm.objc.release(i8* [[T0]]) [[NUW]], !clang.imprecise_release
 // CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[X]]
 // CHECK-NEXT: call void @llvm.objc.release(i8* [[T0]]) [[NUW]], !clang.imprecise_release
 // CHECK-NEXT: ret void
   extern void test2_helper(id (^)(void));
   test2_helper(^{ return x; });
 
 // CHECK:    define linkonce_odr hidden void @__copy_helper_block_8_32s(i8* %0, i8* %1) unnamed_addr #{{[0-9]+}} {

 @implementation Test8
 - (void) test {
 // CHECK:    define internal void @"\01-[Test8 test]"
 // CHECK:      [[SELF:%.*]] = alloca [[TEST8:%.*]]*,
 // CHECK-NEXT: alloca i8*
 // CHECK-NEXT: [[BLOCK:%.*]] = alloca [[BLOCK_T:<{.*}>]],
 // CHECK: store
 // CHECK-NEXT: store
-// CHECK:      [[D0:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK:      [[T0:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK-NEXT: [[T1:%.*]] = load [[TEST8]]*, [[TEST8]]** [[SELF]],
 // CHECK-NEXT: store %0* [[T1]], %0** [[T0]]
 // CHECK-NEXT: bitcast [[BLOCK_T]]* [[BLOCK]] to
 // CHECK: call void @test8_helper(
-// CHECK-NEXT: [[T2:%.*]] = load [[TEST8]]*, [[TEST8]]** [[D0]]
+// CHECK-NEXT: [[T2:%.*]] = load [[TEST8]]*, [[TEST8]]** [[T0]]
 // CHECK-NEXT: call void (...) @llvm.objc.clang.arc.use([[TEST8]]* [[T2]])
 // CHECK: ret void
 
   extern void test8_helper(void (^)(void));
   test8_helper(^{ (void) self; });
 }
 @end
 
void test13(id x) {
   void (^b)(void) = (x ? ^{test13_helper(x);} : 0);
   test13_use(b);
 
   // CHECK-LABEL:    define void @test13(
   // CHECK:      [[X:%.*]] = alloca i8*, align 8
   // CHECK-NEXT: [[B:%.*]] = alloca void ()*, align 8
   // CHECK-NEXT: [[BLOCK:%.*]] = alloca [[BLOCK_T:.*]], align 8
   // CHECK-NEXT: [[CLEANUP_ACTIVE:%.*]] = alloca i1
+  // CHECK-NEXT: [[COND_CLEANUP_SAVE:%.*]] = alloca i8**,
   // CHECK-NEXT: [[T0:%.*]] = call i8* @llvm.objc.retain(i8* {{%.*}})
   // CHECK-NEXT: store i8* [[T0]], i8** [[X]], align 8
   // CHECK-NEXT: [[BPTR1:%.*]] = bitcast void ()** [[B]] to i8*
   // CHECK-NEXT: call void @llvm.lifetime.start.p0i8(i64 8, i8* [[BPTR1]])
-  // CHECK-NEXT: [[CLEANUP_ADDR:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
   // CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[X]], align 8
   // CHECK-NEXT: [[T1:%.*]] = icmp ne i8* [[T0]], null
   // CHECK-NEXT: store i1 false, i1* [[CLEANUP_ACTIVE]]
   // CHECK-NEXT: br i1 [[T1]],
 
   // CHECK-NOT:  br
   // CHECK:      [[CAPTURE:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
   // CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[X]], align 8
   // CHECK-NEXT: [[T1:%.*]] = call i8* @llvm.objc.retain(i8* [[T0]])
   // CHECK-NEXT: store i8* [[T1]], i8** [[CAPTURE]], align 8
   // CHECK-NEXT: store i1 true, i1* [[CLEANUP_ACTIVE]]
+  // CHECK-NEXT: store i8** [[CAPTURE]], i8*** [[COND_CLEANUP_SAVE]], align 8
   // CHECK-NEXT: bitcast [[BLOCK_T]]* [[BLOCK]] to void ()*
   // CHECK-NEXT: br label
   // CHECK:      br label
   // CHECK:      [[T0:%.*]] = phi void ()*
   // CHECK-NEXT: [[T1:%.*]] = bitcast void ()* [[T0]] to i8*
   // CHECK-NEXT: [[T2:%.*]] = call i8* @llvm.objc.retainBlock(i8* [[T1]])
   // CHECK-NEXT: [[T3:%.*]] = bitcast i8* [[T2]] to void ()*
   // CHECK-NEXT: store void ()* [[T3]], void ()** [[B]], align 8
   // CHECK-NEXT: [[T0:%.*]] = load void ()*, void ()** [[B]], align 8
   // CHECK-NEXT: call void @test13_use(void ()* [[T0]])
   // CHECK-NEXT: [[T0:%.*]] = load void ()*, void ()** [[B]]
   // CHECK-NEXT: [[T1:%.*]] = bitcast void ()* [[T0]] to i8*
   // CHECK-NEXT: call void @llvm.objc.release(i8* [[T1]])
 
   // CHECK-NEXT: [[T0:%.*]] = load i1, i1* [[CLEANUP_ACTIVE]]
   // CHECK-NEXT: br i1 [[T0]]
-  // CHECK:      [[T0:%.*]] = load i8*, i8** [[CLEANUP_ADDR]]
+  // CHECK:      [[V12:%.*]] = load i8**, i8*** [[COND_CLEANUP_SAVE]], align 8
+  // CHECK:      [[T0:%.*]] = load i8*, i8** [[V12]]
   // CHECK-NEXT: call void @llvm.objc.release(i8* [[T0]])
   // CHECK-NEXT: br label
 
   // CHECK:      [[BPTR2:%.*]] = bitcast void ()** [[B]] to i8*
   // CHECK-NEXT: call void @llvm.lifetime.end.p0i8(i64 8, i8* [[BPTR2]])
   // CHECK-NEXT:      [[T0:%.*]] = load i8*, i8** [[X]]
   // CHECK-NEXT: call void @llvm.objc.release(i8* [[T0]])
   // CHECK-NEXT: ret void

 void test16() {
   void (^BLKVAR)(void) = ^{ BLKVAR(); };
 
   // CHECK-LABEL: define void @test16(
   // CHECK: [[BLKVAR:%.*]]  = alloca void ()*, align 8
   // CHECK-NEXT:  [[BLOCK:%.*]] = alloca [[BLOCK_T:<{.*}>]],
   // CHECK-NEXT:  [[BLKVARPTR1:%.*]] = bitcast void ()** [[BLKVAR]] to i8*
   // CHECK-NEXT:  call void @llvm.lifetime.start.p0i8(i64 8, i8* [[BLKVARPTR1]])
-  // CHECK-NEXT:  [[SLOTREL:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
   // CHECK-NEXT:  store void ()* null, void ()** [[BLKVAR]], align 8
 }
 
 // rdar://12151005
 //
 // This is an intentional exception to our conservative jump-scope
 // checking for full-expressions containing block literals with
 // non-trivial cleanups: if the block literal appears in the operand

 // CHECK:      [[RET:%.*]] = alloca i8* ()*, align
 // CHECK-NEXT: [[SELF:%.*]] = alloca i8*,
 // CHECK:      [[B0:%.*]] = alloca [[BLOCK:<.*>]], align
 // CHECK:      [[B1:%.*]] = alloca [[BLOCK]], align
 // CHECK:      [[T0:%.*]] = call i8* @llvm.objc.retain(i8*
 // CHECK-NEXT: store i8* [[T0]], i8** [[SELF]], align
 // CHECK-NOT:  objc_retain
 // CHECK-NOT:  objc_release
-// CHECK:      [[DESTROY:%.*]] = getelementptr inbounds [[BLOCK]], [[BLOCK]]* [[B0]], i32 0, i32 5
-// CHECK-NOT:  objc_retain
-// CHECK-NOT:  objc_release
-// CHECK:      [[T0:%.*]] = getelementptr inbounds [[BLOCK]], [[BLOCK]]* [[B0]], i32 0, i32 5
+// CHECK:      [[CAPTURED:%.*]] = getelementptr inbounds [[BLOCK]], [[BLOCK]]* [[B0]], i32 0, i32 5
 // CHECK-NEXT: [[T1:%.*]] = load i8*, i8** [[SELF]], align
 // CHECK-NEXT: [[T2:%.*]] = call i8* @llvm.objc.retain(i8* [[T1]])
-// CHECK-NEXT: store i8* [[T2]], i8** [[T0]],
+// CHECK-NEXT: store i8* [[T2]], i8** [[CAPTURED]],
 // CHECK-NEXT: [[T0:%.*]] = bitcast [[BLOCK]]* [[B0]] to i8* ()*
 // CHECK-NEXT: [[T1:%.*]] = bitcast i8* ()* [[T0]] to i8*
 // CHECK-NEXT: [[T2:%.*]] = call i8* @llvm.objc.retainBlock(i8* [[T1]])
 // CHECK-NEXT: [[T3:%.*]] = bitcast i8* [[T2]] to i8* ()*
 // CHECK-NEXT: store i8* ()* [[T3]], i8* ()** [[RET]]
-// CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[DESTROY]]
+// CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[CAPTURED]]
 // CHECK-NEXT: call void @llvm.objc.release(i8* [[T0]])
 // CHECK-NEXT: store i32
 // CHECK-NEXT: br label
 // CHECK-NOT:  objc_retain
 // CHECK-NOT:  objc_release
-// CHECK:      [[DESTROY:%.*]] = getelementptr inbounds [[BLOCK]], [[BLOCK]]* [[B1]], i32 0, i32 5
-// CHECK-NOT:  objc_retain
-// CHECK-NOT:  objc_release
-// CHECK:      [[T0:%.*]] = getelementptr inbounds [[BLOCK]], [[BLOCK]]* [[B1]], i32 0, i32 5
+// CHECK:      [[CAPTURED:%.*]] = getelementptr inbounds [[BLOCK]], [[BLOCK]]* [[B1]], i32 0, i32 5
 // CHECK-NEXT: [[T1:%.*]] = load i8*, i8** [[SELF]], align
 // CHECK-NEXT: [[T2:%.*]] = call i8* @llvm.objc.retain(i8* [[T1]])
-// CHECK-NEXT: store i8* [[T2]], i8** [[T0]],
+// CHECK-NEXT: store i8* [[T2]], i8** [[CAPTURED]],
 // CHECK-NEXT: [[T0:%.*]] = bitcast [[BLOCK]]* [[B1]] to i8* ()*
 // CHECK-NEXT: [[T1:%.*]] = bitcast i8* ()* [[T0]] to i8*
 // CHECK-NEXT: [[T2:%.*]] = call i8* @llvm.objc.retainBlock(i8* [[T1]])
 // CHECK-NEXT: [[T3:%.*]] = bitcast i8* [[T2]] to i8* ()*
 // CHECK-NEXT: store i8* ()* [[T3]], i8* ()** [[RET]]
-// CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[DESTROY]]
+// CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[CAPTURED]]
 // CHECK-NEXT: call void @llvm.objc.release(i8* [[T0]])
 // CHECK-NEXT: store i32
 // CHECK-NEXT: br label
 
 void test18(id x) {
 // CHECK-UNOPT-LABEL:    define void @test18(
 // CHECK-UNOPT:      [[X:%.*]] = alloca i8*,
 // CHECK-UNOPT-NEXT: [[BLOCK:%.*]] = alloca [[BLOCK_T:<{.*}>]],
 // CHECK-UNOPT-NEXT: store i8* null, i8** [[X]]
 // CHECK-UNOPT-NEXT: call void @llvm.objc.storeStrong(i8** [[X]], 
-// CHECK-UNOPT-NEXT: [[SLOTREL:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK-UNOPT: %[[BLOCK_DESCRIPTOR:.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 4
 // CHECK-UNOPT: store %[[STRUCT_BLOCK_DESCRIPTOR]]* bitcast ({ i64, i64, i8*, i8*, i8*, i64 }* @[[BLOCK_DESCRIPTOR_TMP44]] to %[[STRUCT_BLOCK_DESCRIPTOR]]*), %[[STRUCT_BLOCK_DESCRIPTOR]]** %[[BLOCK_DESCRIPTOR]], align 8
 // CHECK-UNOPT:      [[SLOT:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK-UNOPT-NEXT: [[T0:%.*]] = load i8*, i8** [[X]],
 // CHECK-UNOPT-NEXT: [[T1:%.*]] = call i8* @llvm.objc.retain(i8* [[T0]])
 // CHECK-UNOPT-NEXT: store i8* [[T1]], i8** [[SLOT]],
 // CHECK-UNOPT-NEXT: bitcast
 // CHECK-UNOPT-NEXT: call void @test18_helper(
-// CHECK-UNOPT-NEXT: call void @llvm.objc.storeStrong(i8** [[SLOTREL]], i8* null) [[NUW:#[0-9]+]]
+// CHECK-UNOPT-NEXT: call void @llvm.objc.storeStrong(i8** [[SLOT]], i8* null) [[NUW:#[0-9]+]]
 // CHECK-UNOPT-NEXT: call void @llvm.objc.storeStrong(i8** [[X]], i8* null) [[NUW]]
 // CHECK-UNOPT-NEXT: ret void
   extern void test18_helper(id (^)(void));
   test18_helper(^{ return x; });
 }
 
 // Ensure that we don't emit helper code in copy/dispose routines for variables
 // that are const-captured.

 // CHECK:      [[B:%.*]] = alloca void ()*,
 // CHECK-NEXT: [[BLOCK:%.*]] = alloca [[BLOCK_T:<{.*}>]],
 // CHECK-NEXT: [[T0:%.*]] = bitcast void ()* {{%.*}} to i8*
 // CHECK-NEXT: [[T1:%.*]] = call i8* @llvm.objc.retain(i8* [[T0]])
 // CHECK-NEXT: [[T2:%.*]] = bitcast i8* [[T1]] to void ()*
 // CHECK-NEXT: store void ()* [[T2]], void ()** [[B]]
 
 //   Block setup.  We skip most of this.  Note the bare retain.
-// CHECK-NEXT: [[SLOTREL:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK: %[[BLOCK_DESCRIPTOR:.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 4
 // CHECK: store %[[STRUCT_BLOCK_DESCRIPTOR]]* bitcast ({ i64, i64, i8*, i8*, i8*, i64 }* @[[BLOCK_DESCRIPTOR_TMP48]] to %[[STRUCT_BLOCK_DESCRIPTOR]]*), %[[STRUCT_BLOCK_DESCRIPTOR]]** %[[BLOCK_DESCRIPTOR]], align 8
 // CHECK:      [[SLOT:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK-NEXT: [[T0:%.*]] = load void ()*, void ()** [[B]],
 // CHECK-NEXT: [[T1:%.*]] = bitcast void ()* [[T0]] to i8*
 // CHECK-NEXT: [[T2:%.*]] = call i8* @llvm.objc.retain(i8* [[T1]])
 // CHECK-NEXT: [[T3:%.*]] = bitcast i8* [[T2]] to void ()*
 // CHECK-NEXT: store void ()* [[T3]], void ()** [[SLOT]],
 //   Call.
 // CHECK-NEXT: [[T0:%.*]] = bitcast [[BLOCK_T]]* [[BLOCK]] to void (i32)*
 // CHECK-NEXT: call void @test19_sink(void (i32)* [[T0]])
 
   test19_sink(^(int x) { b(); });
 
 //   Block teardown.
-// CHECK-NEXT: [[T0:%.*]] = load void ()*, void ()** [[SLOTREL]]
+// CHECK-NEXT: [[T0:%.*]] = load void ()*, void ()** [[SLOT]]
 // CHECK-NEXT: [[T1:%.*]] = bitcast void ()* [[T0]] to i8*
 // CHECK-NEXT: call void @llvm.objc.release(i8* [[T1]])
 
 //   Local cleanup.
 // CHECK-NEXT: [[T0:%.*]] = load void ()*, void ()** [[B]]
 // CHECK-NEXT: [[T1:%.*]] = bitcast void ()* [[T0]] to i8*
 // CHECK-NEXT: call void @llvm.objc.release(i8* [[T1]])
 
 // CHECK-NEXT: ret void
 }
 
 // CHECK-LABEL: define void @test20(
 // CHECK: [[XADDR:%.*]] = alloca i8*
 // CHECK-NEXT: [[BLOCK:%.*]] = alloca <[[BLOCKTY:.*]]>
 // CHECK-NEXT: [[RETAINEDX:%.*]] = call i8* @llvm.objc.retain(i8* %{{.*}})
 // CHECK-NEXT: store i8* [[RETAINEDX]], i8** [[XADDR]]
-// CHECK-NEXT: [[CAPTUREFIELD:%.*]] = getelementptr inbounds <[[BLOCKTY]]>, <[[BLOCKTY]]>* [[BLOCK]], i32 0, i32 5
 // CHECK: [[BLOCKCAPTURED:%.*]] = getelementptr inbounds <[[BLOCKTY]]>, <[[BLOCKTY]]>* [[BLOCK]], i32 0, i32 5
 // CHECK: [[CAPTURED:%.*]] = load i8*, i8** [[XADDR]]
 // CHECK: store i8* [[CAPTURED]], i8** [[BLOCKCAPTURED]]
-// CHECK: [[CAPTURE:%.*]] = load i8*, i8** [[CAPTUREFIELD]]
+// CHECK: [[CAPTURE:%.*]] = load i8*, i8** [[BLOCKCAPTURED]]
 // CHECK-NEXT: call void (...) @llvm.objc.clang.arc.use(i8* [[CAPTURE]])
 // CHECK-NEXT: [[X:%.*]] = load i8*, i8** [[XADDR]]
 // CHECK-NEXT: call void @llvm.objc.release(i8* [[X]])
 // CHECK-NEXT: ret void
 
 // CHECK-UNOPT-LABEL: define void @test20(
 // CHECK-UNOPT: [[XADDR:%.*]] = alloca i8*
 // CHECK-UNOPT-NEXT: [[BLOCK:%.*]] = alloca <[[BLOCKTY:.*]]>
-// CHECK-UNOPT: [[CAPTUREFIELD:%.*]] = getelementptr inbounds <[[BLOCKTY]]>, <[[BLOCKTY]]>* [[BLOCK]], i32 0, i32 5
 // CHECK-UNOPT: [[BLOCKCAPTURED:%.*]] = getelementptr inbounds <[[BLOCKTY]]>, <[[BLOCKTY]]>* [[BLOCK]], i32 0, i32 5
 // CHECK-UNOPT: [[CAPTURED:%.*]] = load i8*, i8** [[XADDR]]
 // CHECK-UNOPT: [[RETAINED:%.*]] = call i8* @llvm.objc.retain(i8* [[CAPTURED]])
 // CHECK-UNOPT: store i8* [[RETAINED]], i8** [[BLOCKCAPTURED]]
-// CHECK-UNOPT: call void @llvm.objc.storeStrong(i8** [[CAPTUREFIELD]], i8* null)
+// CHECK-UNOPT: call void @llvm.objc.storeStrong(i8** [[BLOCKCAPTURED]], i8* null)
 
 void test20_callee(void (^)());
 void test20(const id x) {
   test20_callee(^{ (void)x; });
 }
 
 // CHECK-LABEL: define void @test21(
 // CHECK: %[[V6:.*]] = call i8* @llvm.objc.retainBlock(
 // CHECK: %[[V7:.*]] = bitcast i8* %[[V6]] to void ()*
 // CHECK: call void (i32, ...) @test21_callee(i32 1, void ()* %[[V7]]),
 
 void test21_callee(int n, ...);
 void test21(id x) {
   test21_callee(1, ^{ (void)x; });
 }
 
+// The lifetime of 'x', which is captured by the block in the statement
+// expression, should be extended.
+
+// CHECK-COMMON-LABEL: define i8* @test22(
+// CHECK-COMMON: %[[BLOCK_CAPTURED:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %{{.*}}*, i8* }>, <{ i8*, i32, i32, i8*, %{{.*}}*, i8* }>* %{{.*}}, i32 0, i32 5
+// CHECK-COMMON: %[[V3:.*]] = call i8* @llvm.objc.retain(i8* %{{.*}})
+// CHECK-COMMON: store i8* %[[V3]], i8** %[[BLOCK_CAPTURED]], align 8
+// CHECK-COMMON: call void @test22_1()
+// CHECK-UNOPT: call void @llvm.objc.storeStrong(i8** %[[BLOCK_CAPTURED]], i8* null)
+// CHECK: %[[V15:.*]] = load i8*, i8** %[[BLOCK_CAPTURED]], align 8
+// CHECK: call void @llvm.objc.release(i8* %[[V15]])
+
+id test22(int c, id x) {
+  extern id test22_0(void);
+  extern void test22_1(void);
+  return c ? test22_0() : ({ id (^b)(void) = ^{ return x; }; test22_1(); b(); });
+}
+
 // CHECK: attributes [[NUW]] = { nounwind }
 // CHECK-UNOPT: attributes [[NUW]] = { nounwind }

clang/test/CodeGenObjC/arc-foreach.m

 // CHECK-LP64-NEXT: br i1 [[T0]]
 
 // CHECK-LP64:      [[T0:%.*]] = getelementptr inbounds [[STATE_T]], [[STATE_T]]* [[STATE]], i32 0, i32 1
 // CHECK-LP64-NEXT: [[T1:%.*]] = load i8**, i8*** [[T0]]
 // CHECK-LP64-NEXT: [[T2:%.*]] = getelementptr i8*, i8** [[T1]], i64
 // CHECK-LP64-NEXT: [[T3:%.*]] = load i8*, i8** [[T2]]
 // CHECK-LP64-NEXT: store i8* [[T3]], i8** [[X]]
 
-// CHECK-LP64:      [[D0:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
-// CHECK-LP64:      [[T0:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
+// CHECK-LP64:      [[CAPTURED:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK-LP64-NEXT: [[T1:%.*]] = load i8*, i8** [[X]]
 // CHECK-LP64-NEXT: [[T2:%.*]] = call i8* @llvm.objc.retain(i8* [[T1]])
-// CHECK-LP64-NEXT: store i8* [[T2]], i8** [[T0]]
+// CHECK-LP64-NEXT: store i8* [[T2]], i8** [[CAPTURED]]
 // CHECK-LP64-NEXT: [[BLOCK1:%.*]] = bitcast [[BLOCK_T]]* [[BLOCK]]
 // CHECK-LP64-NEXT: call void @use_block(void ()* [[BLOCK1]])
-// CHECK-LP64-NEXT: call void @llvm.objc.storeStrong(i8** [[D0]], i8* null)
+// CHECK-LP64-NEXT: call void @llvm.objc.storeStrong(i8** [[CAPTURED]], i8* null)
 // CHECK-LP64-NOT:  call void (...) @llvm.objc.clang.arc.use(
 
 // CHECK-LP64-OPT: [[D0:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i64 0, i32 5
 // CHECK-LP64-OPT: [[CAPTURE:%.*]] = load i8*, i8** [[D0]]
 // CHECK-LP64-OPT: call void (...) @llvm.objc.clang.arc.use(i8* [[CAPTURE]])
 
 // CHECK-LP64:      [[T0:%.*]] = load i8*, i8** @OBJC_SELECTOR_REFERENCES_
 // CHECK-LP64-NEXT: [[T1:%.*]] = bitcast [[ARRAY_T]]* [[SAVED_ARRAY]] to i8*

 // CHECK-LP64-NEXT: [[BLOCK:%.*]] = alloca [[BLOCK_T:<{.*}>]],
 
 // CHECK-LP64:      [[T0:%.*]] = getelementptr inbounds [[STATE_T]], [[STATE_T]]* [[STATE]], i32 0, i32 1
 // CHECK-LP64-NEXT: [[T1:%.*]] = load i8**, i8*** [[T0]]
 // CHECK-LP64-NEXT: [[T2:%.*]] = getelementptr i8*, i8** [[T1]], i64
 // CHECK-LP64-NEXT: [[T3:%.*]] = load i8*, i8** [[T2]]
 // CHECK-LP64-NEXT: call i8* @llvm.objc.initWeak(i8** [[X]], i8* [[T3]])
 
-// CHECK-LP64:      [[D0:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK-LP64:      [[T0:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i32 0, i32 5
 // CHECK-LP64-NEXT: call void @llvm.objc.copyWeak(i8** [[T0]], i8** [[X]])
 // CHECK-LP64-NEXT: [[T1:%.*]] = bitcast [[BLOCK_T]]* [[BLOCK]] to
 // CHECK-LP64: call void @use_block
-// CHECK-LP64-NEXT: call void @llvm.objc.destroyWeak(i8** [[D0]])
+// CHECK-LP64-NEXT: call void @llvm.objc.destroyWeak(i8** [[T0]])
 // CHECK-LP64-NEXT: call void @llvm.objc.destroyWeak(i8** [[X]])
 
 // rdar://problem/9817306
 @interface Test2
 - (NSArray *) array;
 @end
 void test2(Test2 *a) {
   for (id x in a.array) {
- (void) foo2 {
   }
 }
 @end
 
 // CHECK-LP64-LABEL: define internal void @"\01-[I1 foo2]"(
 // CHECK-LP64:         [[SELF_ADDR:%.*]] = alloca [[TY:%.*]]*,
 // CHECK-LP64:         [[BLOCK:%.*]] = alloca <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, [[TY]]* }>,
 // CHECK-LP64:         store [[TY]]* %self, [[TY]]** [[SELF_ADDR]]
-// CHECK-LP64:         [[T0:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, [[TY]]* }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, [[TY]]* }>* [[BLOCK]], i32 0, i32 5
 // CHECK-LP64:         [[BC:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, [[TY]]* }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, [[TY]]* }>* [[BLOCK]], i32 0, i32 5
 // CHECK-LP64:         [[T1:%.*]] = load [[TY]]*, [[TY]]** [[SELF_ADDR]]
 // CHECK-LP64:         [[T2:%.*]] = bitcast [[TY]]* [[T1]] to i8*
 // CHECK-LP64:         call i8* @llvm.objc.retain(i8* [[T2]])
 
 // CHECK-LP64-OPT-LABEL: define internal void @"\01-[I1 foo2]"(
 // CHECK-LP64-OPT: [[TY:%.*]]* %self
 // CHECK-LP64-OPT: [[BLOCK:%.*]] = alloca [[BLOCK_T:<{.*}>]],
 // CHECK-LP64-OPT: [[T0:%.*]] = getelementptr inbounds [[BLOCK_T]], [[BLOCK_T]]* [[BLOCK]], i64 0, i32 5
 
-// CHECK-LP64:         [[T5:%.*]] = bitcast [[TY]]** [[T0]] to i8**
+// CHECK-LP64:         [[T5:%.*]] = bitcast [[TY]]** [[BC]] to i8**
 // CHECK-LP64:         call void @llvm.objc.storeStrong(i8** [[T5]], i8* null)
 // CHECK-LP64-NOT:     call void (...) @llvm.objc.clang.arc.use([[TY]]* [[T5]])
 // CHECK-LP64:         switch i32 {{%.*}}, label %[[UNREACHABLE:.*]] [
 // CHECK-LP64-NEXT:      i32 0, label %[[CLEANUP_CONT:.*]]
 // CHECK-LP64-NEXT:      i32 2, label %[[FORCOLL_END:.*]]
 // CHECK-LP64-NEXT:    ]
 
 // CHECK-LP64-OPT: [[T5:%.*]] = load [[TY]]*, [[TY]]** [[T0]]

clang/test/CodeGenObjC/noescape.m

 
 // CHECK: define void @test6(i8* %{{.*}}, i8* %[[B:.*]])
 // CHECK: %{{.*}} = alloca i8*, align 8
 // CHECK: %[[B_ADDR:.*]] = alloca i8*, align 8
 // CHECK: %[[BLOCK:.*]] = alloca <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, align 8
 // CHECK-NOARC: store i8* %[[B]], i8** %[[B_ADDR]], align 8
 // CHECK-ARC: store i8* null, i8** %[[B_ADDR]], align 8
 // CHECK-ARC: call void @llvm.objc.storeStrong(i8** %[[B_ADDR]], i8* %[[B]])
-// CHECK-ARC: %[[V0:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>* %[[BLOCK]], i32 0, i32 5
 // CHECK: %[[BLOCK_ISA:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>* %[[BLOCK]], i32 0, i32 0
 // CHECK: store i8* bitcast (i8** @_NSConcreteGlobalBlock to i8*), i8** %[[BLOCK_ISA]], align 8
 // CHECK: %[[BLOCK_FLAGS:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>* %[[BLOCK]], i32 0, i32 1
 // CHECK: store i32 -796917760, i32* %[[BLOCK_FLAGS]], align 8
 // CHECK: %[[BLOCK_DESCRIPTOR:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>* %[[BLOCK]], i32 0, i32 4
 // CHECK: store %[[STRUCT_BLOCK_DESCRIPTOR]]* bitcast ({ i64, i64, i8*, i64 }* @[[BLOCK_DESCIPTOR_TMP_2]] to %[[STRUCT_BLOCK_DESCRIPTOR]]*), %[[STRUCT_BLOCK_DESCRIPTOR]]** %[[BLOCK_DESCRIPTOR]], align 8
 // CHECK: %[[BLOCK_CAPTURED:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>* %[[BLOCK]], i32 0, i32 5
 // CHECK-NOARC: %[[V1:.*]] = load i8*, i8** %[[B_ADDR]], align 8
 // CHECK-NOARC: store i8* %[[V1]], i8** %[[BLOCK_CAPTURED]], align 8
 // CHECK-ARC: %[[V2:.*]] = load i8*, i8** %[[B_ADDR]], align 8
 // CHECK-ARC: %[[V3:.*]] = call i8* @llvm.objc.retain(i8* %[[V2]])
 // CHECK-ARC: store i8* %[[V3]], i8** %[[BLOCK_CAPTURED]], align 8
 // CHECK: call void @noescapeFunc0(
-// CHECK-ARC: call void @llvm.objc.storeStrong(i8** %[[V0]], i8* null)
+// CHECK-ARC: call void @llvm.objc.storeStrong(i8** %[[BLOCK_CAPTURED]], i8* null)
 // CHECK-ARC: call void @llvm.objc.storeStrong(i8** %[[B_ADDR]], i8* null)
 
 // Non-escaping blocks don't need copy/dispose helper functions.
 
 // CHECK-NOT: define internal void @__copy_helper_block_
 // CHECK-NOT: define internal void @__destroy_helper_block_
 
 void func(id);

clang/test/CodeGenOpenCL/cl20-device-side-enqueue.cl

enqueue_kernel(default_queue, flags, ndrange,
                  });
 
   // Emits block literal on stack and block kernel [[INVLK2]].
   // COMMON: [[DEF_Q:%[0-9]+]] = load %opencl.queue_t{{.*}}*, %opencl.queue_t{{.*}}** %default_queue
   // COMMON: [[FLAGS:%[0-9]+]] = load i32, i32* %flags
   // COMMON: [[WAIT_EVNT:%[0-9]+]] = addrspacecast %opencl.clk_event_t{{.*}}** %event_wait_list to %opencl.clk_event_t{{.*}}* addrspace(4)*
   // COMMON: [[EVNT:%[0-9]+]] = addrspacecast %opencl.clk_event_t{{.*}}** %clk_event to %opencl.clk_event_t{{.*}}* addrspace(4)*
   // COMMON: store i8 addrspace(4)* addrspacecast (i8* bitcast (void (i8 addrspace(4)*)* [[INVL2:@__device_side_enqueue_block_invoke[^ ]*]] to i8*) to i8 addrspace(4)*), i8 addrspace(4)** %block.invoke
-  // COMMON: [[BL:%[0-9]+]] = bitcast <{ i32, i32, i8 addrspace(4)*, i32{{.*}}, i32{{.*}}, i32{{.*}} }>* %block3 to %struct.__opencl_block_literal_generic*
+  // COMMON: [[BL:%[0-9]+]] = bitcast <{ i32, i32, i8 addrspace(4)*, i32{{.*}}, i32{{.*}}, i32{{.*}} }>* %block4 to %struct.__opencl_block_literal_generic*
   // COMMON: [[BL_I8:%[0-9]+]] = addrspacecast %struct.__opencl_block_literal_generic* [[BL]] to i8 addrspace(4)*
   // COMMON-LABEL: call i32 @__enqueue_kernel_basic_events
   // COMMON-SAME: (%opencl.queue_t{{.*}}* [[DEF_Q]], i32 [[FLAGS]],  %struct.ndrange_t* {{.*}}, i32 2, %opencl.clk_event_t{{.*}}* addrspace(4)* [[WAIT_EVNT]], %opencl.clk_event_t{{.*}}* addrspace(4)* [[EVNT]],
   // COMMON-SAME: i8 addrspace(4)* addrspacecast (i8* bitcast ({{.*}} [[INVLK2:[^ ]+_kernel]] to i8*) to i8 addrspace(4)*),
   // COMMON-SAME: i8 addrspace(4)* [[BL_I8]])
   enqueue_kernel(default_queue, flags, ndrange, 2, &event_wait_list, &clk_event,
                  ^(void) {
                    a[i] = b[i];