CGBlocks.cpp

Go to the documentation of this file.

//===--- CGBlocks.cpp - Emit LLVM Code for declarations ---------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This contains code to emit blocks.
//
//===----------------------------------------------------------------------===//
 
#include "CGBlocks.h"
#include "CGCXXABI.h"
#include "CGDebugInfo.h"
#include "CGObjCRuntime.h"
#include "CGOpenCLRuntime.h"
#include "CodeGenFunction.h"
#include "CodeGenModule.h"
#include "ConstantEmitter.h"
#include "TargetInfo.h"
#include "clang/AST/Attr.h"
#include "clang/AST/DeclObjC.h"
#include "clang/CodeGen/ConstantInitBuilder.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Module.h"
#include "llvm/Support/ScopedPrinter.h"
#include <algorithm>
#include <cstdio>
 
using namespace clang;
using namespace CodeGen;
 
CGBlockInfo::CGBlockInfo(const BlockDecl *block, StringRef name)
    : Name(name), CXXThisIndex(0), CanBeGlobal(false), NeedsCopyDispose(false),
      NoEscape(false), HasCXXObject(false), UsesStret(false),
      HasCapturedVariableLayout(false), CapturesNonExternalType(false),
      LocalAddress(RawAddress::invalid()), 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.
BlockByrefHelpers::~BlockByrefHelpers() {}
 
/// Build the given block as a global block.
static llvm::Constant *buildGlobalBlock(CodeGenModule &CGM,
                                        const CGBlockInfo &blockInfo,
                                        llvm::Constant *blockFn);
 
/// Build the helper function to copy a block.
static llvm::Constant *buildCopyHelper(CodeGenModule &CGM,
                                       const CGBlockInfo &blockInfo) {
  return CodeGenFunction(CGM).GenerateCopyHelperFunction(blockInfo);
}
 
/// Build the helper function to dispose of a block.
static llvm::Constant *buildDisposeHelper(CodeGenModule &CGM,
                                          const CGBlockInfo &blockInfo) {
  return CodeGenFunction(CGM).GenerateDestroyHelperFunction(blockInfo);
}
 
namespace {
 
enum class CaptureStrKind {
  // String for the copy helper.
  CopyHelper,
  // String for the dispose helper.
  DisposeHelper,
  // Merge the strings for the copy helper and dispose helper.
  Merged
};
 
} // end anonymous namespace
 
static std::string getBlockCaptureStr(const CGBlockInfo::Capture &Cap,
                                      CaptureStrKind StrKind,
                                      CharUnits BlockAlignment,
                                      CodeGenModule &CGM);
 
static std::string getBlockDescriptorName(const CGBlockInfo &BlockInfo,
                                          CodeGenModule &CGM) {
  std::string Name = "__block_descriptor_";
  Name += llvm::to_string(BlockInfo.BlockSize.getQuantity()) + "_";
 
  if (BlockInfo.NeedsCopyDispose) {
    if (CGM.getLangOpts().Exceptions)
      Name += "e";
    if (CGM.getCodeGenOpts().ObjCAutoRefCountExceptions)
      Name += "a";
    Name += llvm::to_string(BlockInfo.BlockAlign.getQuantity()) + "_";
 
    for (auto &Cap : BlockInfo.SortedCaptures) {
      if (Cap.isConstantOrTrivial())
        continue;
 
      Name += llvm::to_string(Cap.getOffset().getQuantity());
 
      if (Cap.CopyKind == Cap.DisposeKind) {
        // If CopyKind and DisposeKind are the same, merge the capture
        // information.
        assert(Cap.CopyKind != BlockCaptureEntityKind::None &&
               "shouldn't see BlockCaptureManagedEntity that is None");
        Name += getBlockCaptureStr(Cap, CaptureStrKind::Merged,
                                   BlockInfo.BlockAlign, CGM);
      } else {
        // If CopyKind and DisposeKind are not the same, which can happen when
        // either Kind is None or the captured object is a __strong block,
        // concatenate the copy and dispose strings.
        Name += getBlockCaptureStr(Cap, CaptureStrKind::CopyHelper,
                                   BlockInfo.BlockAlign, CGM);
        Name += getBlockCaptureStr(Cap, CaptureStrKind::DisposeHelper,
                                   BlockInfo.BlockAlign, CGM);
      }
    }
    Name += "_";
  }
 
  std::string TypeAtEncoding;
 
  if (!CGM.getCodeGenOpts().DisableBlockSignatureString) {
    TypeAtEncoding =
        CGM.getContext().getObjCEncodingForBlock(BlockInfo.getBlockExpr());
    /// Replace occurrences of '@' with '\1'. '@' is reserved on ELF platforms
    /// as a separator between symbol name and symbol version.
    std::replace(TypeAtEncoding.begin(), TypeAtEncoding.end(), '@', '\1');
  }
  Name += "e" + llvm::to_string(TypeAtEncoding.size()) + "_" + TypeAtEncoding;
  Name += "l" + CGM.getObjCRuntime().getRCBlockLayoutStr(CGM, BlockInfo);
  return Name;
}
 
/// buildBlockDescriptor - Build the block descriptor meta-data for a block.
/// buildBlockDescriptor is accessed from 5th field of the Block_literal
/// meta-data and contains stationary information about the block literal.
/// Its definition will have 4 (or optionally 6) words.
/// \code
/// struct Block_descriptor {
///   unsigned long reserved;
///   unsigned long size;  // size of Block_literal metadata in bytes.
///   void *copy_func_helper_decl;  // optional copy helper.
///   void *destroy_func_decl; // optional destructor helper.
///   void *block_method_encoding_address; // @encode for block literal signature.
///   void *block_layout_info; // encoding of captured block variables.
/// };
/// \endcode
static llvm::Constant *buildBlockDescriptor(CodeGenModule &CGM,
                                            const CGBlockInfo &blockInfo) {
  ASTContext &C = CGM.getContext();
 
  llvm::IntegerType *ulong =
    cast<llvm::IntegerType>(CGM.getTypes().ConvertType(C.UnsignedLongTy));
  llvm::PointerType *i8p = nullptr;
  if (CGM.getLangOpts().OpenCL)
    i8p = llvm::PointerType::get(
        CGM.getLLVMContext(), C.getTargetAddressSpace(LangAS::opencl_constant));
  else
    i8p = CGM.VoidPtrTy;
 
  std::string descName;
 
  // If an equivalent block descriptor global variable exists, return it.
  if (C.getLangOpts().ObjC &&
      CGM.getLangOpts().getGC() == LangOptions::NonGC) {
    descName = getBlockDescriptorName(blockInfo, CGM);
    if (llvm::GlobalValue *desc = CGM.getModule().getNamedValue(descName))
      return desc;
  }
 
  // If there isn't an equivalent block descriptor global variable, create a new
  // one.
  ConstantInitBuilder builder(CGM);
  auto elements = builder.beginStruct();
 
  // reserved
  elements.addInt(ulong, 0);
 
  // Size
  // FIXME: What is the right way to say this doesn't fit?  We should give
  // a user diagnostic in that case.  Better fix would be to change the
  // API to size_t.
  elements.addInt(ulong, blockInfo.BlockSize.getQuantity());
 
  // Optional copy/dispose helpers.
  bool hasInternalHelper = false;
  if (blockInfo.NeedsCopyDispose) {
    // copy_func_helper_decl
    llvm::Constant *copyHelper = buildCopyHelper(CGM, blockInfo);
    elements.add(copyHelper);
 
    // destroy_func_decl
    llvm::Constant *disposeHelper = buildDisposeHelper(CGM, blockInfo);
    elements.add(disposeHelper);
 
    if (cast<llvm::Function>(copyHelper->stripPointerCasts())
            ->hasInternalLinkage() ||
        cast<llvm::Function>(disposeHelper->stripPointerCasts())
            ->hasInternalLinkage())
      hasInternalHelper = true;
  }
 
  // Signature.  Mandatory ObjC-style method descriptor @encode sequence.
  if (CGM.getCodeGenOpts().DisableBlockSignatureString) {
    elements.addNullPointer(i8p);
  } else {
    std::string typeAtEncoding =
        CGM.getContext().getObjCEncodingForBlock(blockInfo.getBlockExpr());
    elements.add(CGM.GetAddrOfConstantCString(typeAtEncoding).getPointer());
  }
 
  // GC layout.
  if (C.getLangOpts().ObjC) {
    if (CGM.getLangOpts().getGC() != LangOptions::NonGC)
      elements.add(CGM.getObjCRuntime().BuildGCBlockLayout(CGM, blockInfo));
    else
      elements.add(CGM.getObjCRuntime().BuildRCBlockLayout(CGM, blockInfo));
  }
  else
    elements.addNullPointer(i8p);
 
  unsigned AddrSpace = 0;
  if (C.getLangOpts().OpenCL)
    AddrSpace = C.getTargetAddressSpace(LangAS::opencl_constant);
 
  llvm::GlobalValue::LinkageTypes linkage;
  if (descName.empty()) {
    linkage = llvm::GlobalValue::InternalLinkage;
    descName = "__block_descriptor_tmp";
  } else if (hasInternalHelper) {
    // If either the copy helper or the dispose helper has internal linkage,
    // the block descriptor must have internal linkage too.
    linkage = llvm::GlobalValue::InternalLinkage;
  } else {
    linkage = llvm::GlobalValue::LinkOnceODRLinkage;
  }
 
  llvm::GlobalVariable *global =
      elements.finishAndCreateGlobal(descName, CGM.getPointerAlign(),
                                     /*constant*/ true, linkage, AddrSpace);
 
  if (linkage == llvm::GlobalValue::LinkOnceODRLinkage) {
    if (CGM.supportsCOMDAT())
      global->setComdat(CGM.getModule().getOrInsertComdat(descName));
    global->setVisibility(llvm::GlobalValue::HiddenVisibility);
    global->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);
  }
 
  return global;
}
 
/*
  Purely notional variadic template describing the layout of a block.
 
  template <class _ResultType, class... _ParamTypes, class... _CaptureTypes>
  struct Block_literal {
    /// Initialized to one of:
    ///   extern void *_NSConcreteStackBlock[];
    ///   extern void *_NSConcreteGlobalBlock[];
    ///
    /// In theory, we could start one off malloc'ed by setting
    /// BLOCK_NEEDS_FREE, giving it a refcount of 1, and using
    /// this isa:
    ///   extern void *_NSConcreteMallocBlock[];
    struct objc_class *isa;
 
    /// These are the flags (with corresponding bit number) that the
    /// compiler is actually supposed to know about.
    ///  23. BLOCK_IS_NOESCAPE - indicates that the block is non-escaping
    ///  25. BLOCK_HAS_COPY_DISPOSE - indicates that the block
    ///   descriptor provides copy and dispose helper functions
    ///  26. BLOCK_HAS_CXX_OBJ - indicates that there's a captured
    ///   object with a nontrivial destructor or copy constructor
    ///  28. BLOCK_IS_GLOBAL - indicates that the block is allocated
    ///   as global memory
    ///  29. BLOCK_USE_STRET - indicates that the block function
    ///   uses stret, which objc_msgSend needs to know about
    ///  30. BLOCK_HAS_SIGNATURE - indicates that the block has an
    ///   @encoded signature string
    /// And we're not supposed to manipulate these:
    ///  24. BLOCK_NEEDS_FREE - indicates that the block has been moved
    ///   to malloc'ed memory
    ///  27. BLOCK_IS_GC - indicates that the block has been moved to
    ///   to GC-allocated memory
    /// Additionally, the bottom 16 bits are a reference count which
    /// should be zero on the stack.
    int flags;
 
    /// Reserved;  should be zero-initialized.
    int reserved;
 
    /// Function pointer generated from block literal.
    _ResultType (*invoke)(Block_literal *, _ParamTypes...);
 
    /// Block description metadata generated from block literal.
    struct Block_descriptor *block_descriptor;
 
    /// Captured values follow.
    _CapturesTypes captures...;
  };
 */
 
namespace {
  /// A chunk of data that we actually have to capture in the block.
  struct BlockLayoutChunk {
    CharUnits Alignment;
    CharUnits Size;
    const BlockDecl::Capture *Capture; // null for 'this'
    llvm::Type *Type;
    QualType FieldType;
    BlockCaptureEntityKind CopyKind, DisposeKind;
    BlockFieldFlags CopyFlags, DisposeFlags;
 
    BlockLayoutChunk(CharUnits align, CharUnits size,
                     const BlockDecl::Capture *capture, llvm::Type *type,
                     QualType fieldType, BlockCaptureEntityKind CopyKind,
                     BlockFieldFlags CopyFlags,
                     BlockCaptureEntityKind DisposeKind,
                     BlockFieldFlags DisposeFlags)
        : Alignment(align), Size(size), Capture(capture), Type(type),
          FieldType(fieldType), CopyKind(CopyKind), DisposeKind(DisposeKind),
          CopyFlags(CopyFlags), DisposeFlags(DisposeFlags) {}
 
    /// Tell the block info that this chunk has the given field index.
    void setIndex(CGBlockInfo &info, unsigned index, CharUnits offset) {
      if (!Capture) {
        info.CXXThisIndex = index;
        info.CXXThisOffset = offset;
      } else {
        info.SortedCaptures.push_back(CGBlockInfo::Capture::makeIndex(
            index, offset, FieldType, CopyKind, CopyFlags, DisposeKind,
            DisposeFlags, Capture));
      }
    }
 
    bool isTrivial() const {
      return CopyKind == BlockCaptureEntityKind::None &&
             DisposeKind == BlockCaptureEntityKind::None;
    }
  };
 
  /// Order by 1) all __strong together 2) next, all block together 3) next,
  /// all byref together 4) next, all __weak together. Preserve descending
  /// alignment in all situations.
  bool operator<(const BlockLayoutChunk &left, const BlockLayoutChunk &right) {
    if (left.Alignment != right.Alignment)
      return left.Alignment > right.Alignment;
 
    auto getPrefOrder = [](const BlockLayoutChunk &chunk) {
      switch (chunk.CopyKind) {
      case BlockCaptureEntityKind::ARCStrong:
        return 0;
      case BlockCaptureEntityKind::BlockObject:
        switch (chunk.CopyFlags.getBitMask()) {
        case BLOCK_FIELD_IS_OBJECT:
          return 0;
        case BLOCK_FIELD_IS_BLOCK:
          return 1;
        case BLOCK_FIELD_IS_BYREF:
          return 2;
        default:
          break;
        }
        break;
      case BlockCaptureEntityKind::ARCWeak:
        return 3;
      default:
        break;
      }
      return 4;
    };
 
    return getPrefOrder(left) < getPrefOrder(right);
  }
} // end anonymous namespace
 
static std::pair<BlockCaptureEntityKind, BlockFieldFlags>
computeCopyInfoForBlockCapture(const BlockDecl::Capture &CI, QualType T,
                               const LangOptions &LangOpts);
 
static std::pair<BlockCaptureEntityKind, BlockFieldFlags>
computeDestroyInfoForBlockCapture(const BlockDecl::Capture &CI, QualType T,
                                  const LangOptions &LangOpts);
 
static void addBlockLayout(CharUnits align, CharUnits size,
                           const BlockDecl::Capture *capture, llvm::Type *type,
                           QualType fieldType,
                           SmallVectorImpl<BlockLayoutChunk> &Layout,
                           CGBlockInfo &Info, CodeGenModule &CGM) {
  if (!capture) {
    // 'this' capture.
    Layout.push_back(BlockLayoutChunk(
        align, size, capture, type, fieldType, BlockCaptureEntityKind::None,
        BlockFieldFlags(), BlockCaptureEntityKind::None, BlockFieldFlags()));
    return;
  }
 
  const LangOptions &LangOpts = CGM.getLangOpts();
  BlockCaptureEntityKind CopyKind, DisposeKind;
  BlockFieldFlags CopyFlags, DisposeFlags;
 
  std::tie(CopyKind, CopyFlags) =
      computeCopyInfoForBlockCapture(*capture, fieldType, LangOpts);
  std::tie(DisposeKind, DisposeFlags) =
      computeDestroyInfoForBlockCapture(*capture, fieldType, LangOpts);
  Layout.push_back(BlockLayoutChunk(align, size, capture, type, fieldType,
                                    CopyKind, CopyFlags, DisposeKind,
                                    DisposeFlags));
 
  if (Info.NoEscape)
    return;
 
  if (!Layout.back().isTrivial())
    Info.NeedsCopyDispose = true;
}
 
/// Determines if the given type is safe for constant capture in C++.
static bool isSafeForCXXConstantCapture(QualType type) {
  const RecordType *recordType =
    type->getBaseElementTypeUnsafe()->getAs<RecordType>();
 
  // Only records can be unsafe.
  if (!recordType) return true;
 
  const auto *record = cast<CXXRecordDecl>(recordType->getDecl());
 
  // Maintain semantics for classes with non-trivial dtors or copy ctors.
  if (!record->hasTrivialDestructor()) return false;
  if (record->hasNonTrivialCopyConstructor()) return false;
 
  // Otherwise, we just have to make sure there aren't any mutable
  // fields that might have changed since initialization.
  return !record->hasMutableFields();
}
 
/// It is illegal to modify a const object after initialization.
/// Therefore, if a const object has a constant initializer, we don't
/// actually need to keep storage for it in the block; we'll just
/// rematerialize it at the start of the block function.  This is
/// acceptable because we make no promises about address stability of
/// captured variables.
static llvm::Constant *tryCaptureAsConstant(CodeGenModule &CGM,
                                            CodeGenFunction *CGF,
                                            const VarDecl *var) {
  // Return if this is a function parameter. We shouldn't try to
  // rematerialize default arguments of function parameters.
  if (isa<ParmVarDecl>(var))
    return nullptr;
 
  QualType type = var->getType();
 
  // We can only do this if the variable is const.
  if (!type.isConstQualified()) return nullptr;
 
  // Furthermore, in C++ we have to worry about mutable fields:
  // C++ [dcl.type.cv]p4:
  //   Except that any class member declared mutable can be
  //   modified, any attempt to modify a const object during its
  //   lifetime results in undefined behavior.
  if (CGM.getLangOpts().CPlusPlus && !isSafeForCXXConstantCapture(type))
    return nullptr;
 
  // If the variable doesn't have any initializer (shouldn't this be
  // invalid?), it's not clear what we should do.  Maybe capture as
  // zero?
  const Expr *init = var->getInit();
  if (!init) return nullptr;
 
  return ConstantEmitter(CGM, CGF).tryEmitAbstractForInitializer(*var);
}
 
/// Get the low bit of a nonzero character count.  This is the
/// alignment of the nth byte if the 0th byte is universally aligned.
static CharUnits getLowBit(CharUnits v) {
  return CharUnits::fromQuantity(v.getQuantity() & (~v.getQuantity() + 1));
}
 
static void initializeForBlockHeader(CodeGenModule &CGM, CGBlockInfo &info,
                             SmallVectorImpl<llvm::Type*> &elementTypes) {
 
  assert(elementTypes.empty());
  if (CGM.getLangOpts().OpenCL) {
    // The header is basically 'struct { int; int; generic void *;
    // custom_fields; }'. Assert that struct is packed.
    auto GenPtrAlign = CharUnits::fromQuantity(
        CGM.getTarget().getPointerAlign(LangAS::opencl_generic) / 8);
    auto GenPtrSize = CharUnits::fromQuantity(
        CGM.getTarget().getPointerWidth(LangAS::opencl_generic) / 8);
    assert(CGM.getIntSize() <= GenPtrSize);
    assert(CGM.getIntAlign() <= GenPtrAlign);
    assert((2 * CGM.getIntSize()).isMultipleOf(GenPtrAlign));
    elementTypes.push_back(CGM.IntTy); /* total size */
    elementTypes.push_back(CGM.IntTy); /* align */
    elementTypes.push_back(
        CGM.getOpenCLRuntime()
            .getGenericVoidPointerType()); /* invoke function */
    unsigned Offset =
        2 * CGM.getIntSize().getQuantity() + GenPtrSize.getQuantity();
    unsigned BlockAlign = GenPtrAlign.getQuantity();
    if (auto *Helper =
            CGM.getTargetCodeGenInfo().getTargetOpenCLBlockHelper()) {
      for (auto *I : Helper->getCustomFieldTypes()) /* custom fields */ {
        // TargetOpenCLBlockHelp needs to make sure the struct is packed.
        // If necessary, add padding fields to the custom fields.
        unsigned Align = CGM.getDataLayout().getABITypeAlign(I).value();
        if (BlockAlign < Align)
          BlockAlign = Align;
        assert(Offset % Align == 0);
        Offset += CGM.getDataLayout().getTypeAllocSize(I);
        elementTypes.push_back(I);
      }
    }
    info.BlockAlign = CharUnits::fromQuantity(BlockAlign);
    info.BlockSize = CharUnits::fromQuantity(Offset);
  } else {
    // The header is basically 'struct { void *; int; int; void *; void *; }'.
    // Assert that the struct is packed.
    assert(CGM.getIntSize() <= CGM.getPointerSize());
    assert(CGM.getIntAlign() <= CGM.getPointerAlign());
    assert((2 * CGM.getIntSize()).isMultipleOf(CGM.getPointerAlign()));
    info.BlockAlign = CGM.getPointerAlign();
    info.BlockSize = 3 * CGM.getPointerSize() + 2 * CGM.getIntSize();
    elementTypes.push_back(CGM.VoidPtrTy);
    elementTypes.push_back(CGM.IntTy);
    elementTypes.push_back(CGM.IntTy);
    elementTypes.push_back(CGM.VoidPtrTy);
    elementTypes.push_back(CGM.getBlockDescriptorType());
  }
}
 
static QualType getCaptureFieldType(const CodeGenFunction &CGF,
                                    const BlockDecl::Capture &CI) {
  const VarDecl *VD = CI.getVariable();
 
  // If the variable is captured by an enclosing block or lambda expression,
  // use the type of the capture field.
  if (CGF.BlockInfo && CI.isNested())
    return CGF.BlockInfo->getCapture(VD).fieldType();
  if (auto *FD = CGF.LambdaCaptureFields.lookup(VD))
    return FD->getType();
  // If the captured variable is a non-escaping __block variable, the field
  // type is the reference type. If the variable is a __block variable that
  // already has a reference type, the field type is the variable's type.
  return VD->isNonEscapingByref() ?
         CGF.getContext().getLValueReferenceType(VD->getType()) : VD->getType();
}
 
/// Compute the layout of the given block.  Attempts to lay the block
/// out with minimal space requirements.
static void computeBlockInfo(CodeGenModule &CGM, CodeGenFunction *CGF,
                             CGBlockInfo &info) {
  ASTContext &C = CGM.getContext();
  const BlockDecl *block = info.getBlockDecl();
 
  SmallVector<llvm::Type*, 8> elementTypes;
  initializeForBlockHeader(CGM, info, elementTypes);
  bool hasNonConstantCustomFields = false;
  if (auto *OpenCLHelper =
          CGM.getTargetCodeGenInfo().getTargetOpenCLBlockHelper())
    hasNonConstantCustomFields =
        !OpenCLHelper->areAllCustomFieldValuesConstant(info);
  if (!block->hasCaptures() && !hasNonConstantCustomFields) {
    info.StructureType =
      llvm::StructType::get(CGM.getLLVMContext(), elementTypes, true);
    info.CanBeGlobal = true;
    return;
  }
  else if (C.getLangOpts().ObjC &&
           CGM.getLangOpts().getGC() == LangOptions::NonGC)
    info.HasCapturedVariableLayout = true;
 
  if (block->doesNotEscape())
    info.NoEscape = true;
 
  // Collect the layout chunks.
  SmallVector<BlockLayoutChunk, 16> layout;
  layout.reserve(block->capturesCXXThis() +
                 (block->capture_end() - block->capture_begin()));
 
  CharUnits maxFieldAlign;
 
  // First, 'this'.
  if (block->capturesCXXThis()) {
    assert(CGF && isa_and_nonnull<CXXMethodDecl>(CGF->CurFuncDecl) &&
           "Can't capture 'this' outside a method");
    QualType thisType = cast<CXXMethodDecl>(CGF->CurFuncDecl)->getThisType();
 
    // Theoretically, this could be in a different address space, so
    // don't assume standard pointer size/align.
    llvm::Type *llvmType = CGM.getTypes().ConvertType(thisType);
    auto TInfo = CGM.getContext().getTypeInfoInChars(thisType);
    maxFieldAlign = std::max(maxFieldAlign, TInfo.Align);
 
    addBlockLayout(TInfo.Align, TInfo.Width, nullptr, llvmType, thisType,
                   layout, info, CGM);
  }
 
  // Next, all the block captures.
  for (const auto &CI : block->captures()) {
    const VarDecl *variable = CI.getVariable();
 
    if (CI.isEscapingByref()) {
      // Just use void* instead of a pointer to the byref type.
      CharUnits align = CGM.getPointerAlign();
      maxFieldAlign = std::max(maxFieldAlign, align);
 
      // Since a __block variable cannot be captured by lambdas, its type and
      // the capture field type should always match.
      assert(CGF && getCaptureFieldType(*CGF, CI) == variable->getType() &&
             "capture type differs from the variable type");
      addBlockLayout(align, CGM.getPointerSize(), &CI, CGM.VoidPtrTy,
                     variable->getType(), layout, info, CGM);
      continue;
    }
 
    // Otherwise, build a layout chunk with the size and alignment of
    // the declaration.
    if (llvm::Constant *constant = tryCaptureAsConstant(CGM, CGF, variable)) {
      info.SortedCaptures.push_back(
          CGBlockInfo::Capture::makeConstant(constant, &CI));
      continue;
    }
 
    QualType VT = getCaptureFieldType(*CGF, CI);
 
    if (CGM.getLangOpts().CPlusPlus)
      if (const CXXRecordDecl *record = VT->getAsCXXRecordDecl())
        if (CI.hasCopyExpr() || !record->hasTrivialDestructor()) {
          info.HasCXXObject = true;
          if (!record->isExternallyVisible())
            info.CapturesNonExternalType = true;
        }
 
    CharUnits size = C.getTypeSizeInChars(VT);
    CharUnits align = C.getDeclAlign(variable);
 
    maxFieldAlign = std::max(maxFieldAlign, align);
 
    llvm::Type *llvmType =
      CGM.getTypes().ConvertTypeForMem(VT);
 
    addBlockLayout(align, size, &CI, llvmType, VT, layout, info, CGM);
  }
 
  // If that was everything, we're done here.
  if (layout.empty()) {
    info.StructureType =
      llvm::StructType::get(CGM.getLLVMContext(), elementTypes, true);
    info.CanBeGlobal = true;
    info.buildCaptureMap();
    return;
  }
 
  // Sort the layout by alignment.  We have to use a stable sort here
  // to get reproducible results.  There should probably be an
  // llvm::array_pod_stable_sort.
  llvm::stable_sort(layout);
 
  // Needed for blocks layout info.
  info.BlockHeaderForcedGapOffset = info.BlockSize;
  info.BlockHeaderForcedGapSize = CharUnits::Zero();
 
  CharUnits &blockSize = info.BlockSize;
  info.BlockAlign = std::max(maxFieldAlign, info.BlockAlign);
 
  // Assuming that the first byte in the header is maximally aligned,
  // get the alignment of the first byte following the header.
  CharUnits endAlign = getLowBit(blockSize);
 
  // If the end of the header isn't satisfactorily aligned for the
  // maximum thing, look for things that are okay with the header-end
  // alignment, and keep appending them until we get something that's
  // aligned right.  This algorithm is only guaranteed optimal if
  // that condition is satisfied at some point; otherwise we can get
  // things like:
  //   header                 // next byte has alignment 4
  //   something_with_size_5; // next byte has alignment 1
  //   something_with_alignment_8;
  // which has 7 bytes of padding, as opposed to the naive solution
  // which might have less (?).
  if (endAlign < maxFieldAlign) {
    SmallVectorImpl<BlockLayoutChunk>::iterator
      li = layout.begin() + 1, le = layout.end();
 
    // Look for something that the header end is already
    // satisfactorily aligned for.
    for (; li != le && endAlign < li->Alignment; ++li)
      ;
 
    // If we found something that's naturally aligned for the end of
    // the header, keep adding things...
    if (li != le) {
      SmallVectorImpl<BlockLayoutChunk>::iterator first = li;
      for (; li != le; ++li) {
        assert(endAlign >= li->Alignment);
 
        li->setIndex(info, elementTypes.size(), blockSize);
        elementTypes.push_back(li->Type);
        blockSize += li->Size;
        endAlign = getLowBit(blockSize);
 
        // ...until we get to the alignment of the maximum field.
        if (endAlign >= maxFieldAlign) {
          ++li;
          break;
        }
      }
      // Don't re-append everything we just appended.
      layout.erase(first, li);
    }
  }
 
  assert(endAlign == getLowBit(blockSize));
 
  // At this point, we just have to add padding if the end align still
  // isn't aligned right.
  if (endAlign < maxFieldAlign) {
    CharUnits newBlockSize = blockSize.alignTo(maxFieldAlign);
    CharUnits padding = newBlockSize - blockSize;
 
    // If we haven't yet added any fields, remember that there was an
    // initial gap; this need to go into the block layout bit map.
    if (blockSize == info.BlockHeaderForcedGapOffset) {
      info.BlockHeaderForcedGapSize = padding;
    }
 
    elementTypes.push_back(llvm::ArrayType::get(CGM.Int8Ty,
                                                padding.getQuantity()));
    blockSize = newBlockSize;
    endAlign = getLowBit(blockSize); // might be > maxFieldAlign
  }
 
  assert(endAlign >= maxFieldAlign);
  assert(endAlign == getLowBit(blockSize));
  // Slam everything else on now.  This works because they have
  // strictly decreasing alignment and we expect that size is always a
  // multiple of alignment.
  for (SmallVectorImpl<BlockLayoutChunk>::iterator
         li = layout.begin(), le = layout.end(); li != le; ++li) {
    if (endAlign < li->Alignment) {
      // size may not be multiple of alignment. This can only happen with
      // an over-aligned variable. We will be adding a padding field to
      // make the size be multiple of alignment.
      CharUnits padding = li->Alignment - endAlign;
      elementTypes.push_back(llvm::ArrayType::get(CGM.Int8Ty,
                                                  padding.getQuantity()));
      blockSize += padding;
      endAlign = getLowBit(blockSize);
    }
    assert(endAlign >= li->Alignment);
    li->setIndex(info, elementTypes.size(), blockSize);
    elementTypes.push_back(li->Type);
    blockSize += li->Size;
    endAlign = getLowBit(blockSize);
  }
 
  info.buildCaptureMap();
  info.StructureType =
    llvm::StructType::get(CGM.getLLVMContext(), elementTypes, true);
}
 
/// 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())
    // 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))
      return Block;
 
  CGBlockInfo blockInfo(blockExpr->getBlockDecl(), CurFn->getName());
  computeBlockInfo(CGM, this, blockInfo);
  blockInfo.BlockExpression = blockExpr;
  if (!blockInfo.CanBeGlobal)
    blockInfo.LocalAddress = CreateTempAlloca(blockInfo.StructureType,
                                              blockInfo.BlockAlign, "block");
  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(GenVoidPtrAddr) / 8);
  // Using the computed layout, generate the actual block function.
  bool isLambdaConv = blockInfo.getBlockDecl()->isConversionFromLambda();
  CodeGenFunction BlockCGF{CGM, true};
  BlockCGF.SanOpts = SanOpts;
  auto *InvokeFn = BlockCGF.GenerateBlockFunction(
      CurGD, blockInfo, LocalDeclMap, isLambdaConv, blockInfo.CanBeGlobal);
  auto *blockFn = llvm::ConstantExpr::getPointerCast(InvokeFn, GenVoidPtrTy);
 
  // If there is nothing to capture, we can emit this as a global block.
  if (blockInfo.CanBeGlobal)
    return CGM.getAddrOfGlobalBlockIfEmitted(blockInfo.BlockExpression);
 
  // Otherwise, we have to emit this as a local block.
 
  RawAddress blockAddr = blockInfo.LocalAddress;
  assert(blockAddr.isValid() && "block has no address!");
 
  llvm::Constant *isa;
  llvm::Constant *descriptor;
  BlockFlags flags;
  if (!IsOpenCL) {
    // If the block is non-escaping, set field 'isa 'to NSConcreteGlobalBlock
    // and set the BLOCK_IS_GLOBAL bit of field 'flags'. Copying a non-escaping
    // block just returns the original block and releasing it is a no-op.
    llvm::Constant *blockISA = blockInfo.NoEscape
                                   ? CGM.getNSConcreteGlobalBlock()
                                   : CGM.getNSConcreteStackBlock();
    isa = blockISA;
 
    // Build the block descriptor.
    descriptor = buildBlockDescriptor(CGM, blockInfo);
 
    // Compute the initial on-stack block flags.
    if (!CGM.getCodeGenOpts().DisableBlockSignatureString)
      flags = BLOCK_HAS_SIGNATURE;
    if (blockInfo.HasCapturedVariableLayout)
      flags |= BLOCK_HAS_EXTENDED_LAYOUT;
    if (blockInfo.NeedsCopyDispose)
      flags |= BLOCK_HAS_COPY_DISPOSE;
    if (blockInfo.HasCXXObject)
      flags |= BLOCK_HAS_CXX_OBJ;
    if (blockInfo.UsesStret)
      flags |= BLOCK_USE_STRET;
    if (blockInfo.NoEscape)
      flags |= BLOCK_IS_NOESCAPE | BLOCK_IS_GLOBAL;
  }
 
  auto projectField = [&](unsigned index, const Twine &name) -> Address {
    return Builder.CreateStructGEP(blockAddr, index, name);
  };
  auto storeField = [&](llvm::Value *value, unsigned index, const Twine &name) {
    Builder.CreateStore(value, projectField(index, name));
  };
 
  // Initialize the block header.
  {
    // We assume all the header fields are densely packed.
    unsigned index = 0;
    CharUnits offset;
    auto addHeaderField = [&](llvm::Value *value, CharUnits size,
                              const Twine &name) {
      storeField(value, index, name);
      offset += size;
      index++;
    };
 
    if (!IsOpenCL) {
      addHeaderField(isa, getPointerSize(), "block.isa");
      addHeaderField(llvm::ConstantInt::get(IntTy, flags.getBitMask()),
                     getIntSize(), "block.flags");
      addHeaderField(llvm::ConstantInt::get(IntTy, 0), getIntSize(),
                     "block.reserved");
    } else {
      addHeaderField(
          llvm::ConstantInt::get(IntTy, blockInfo.BlockSize.getQuantity()),
          getIntSize(), "block.size");
      addHeaderField(
          llvm::ConstantInt::get(IntTy, blockInfo.BlockAlign.getQuantity()),
          getIntSize(), "block.align");
    }
    addHeaderField(blockFn, GenVoidPtrSize, "block.invoke");
    if (!IsOpenCL)
      addHeaderField(descriptor, getPointerSize(), "block.descriptor");
    else if (auto *Helper =
                 CGM.getTargetCodeGenInfo().getTargetOpenCLBlockHelper()) {
      for (auto I : Helper->getCustomFieldValues(*this, blockInfo)) {
        addHeaderField(
            I.first,
            CharUnits::fromQuantity(
                CGM.getDataLayout().getTypeAllocSize(I.first->getType())),
            I.second);
      }
    }
  }
 
  // Finally, capture all the values into the block.
  const BlockDecl *blockDecl = blockInfo.getBlockDecl();
 
  // First, 'this'.
  if (blockDecl->capturesCXXThis()) {
    Address addr =
        projectField(blockInfo.CXXThisIndex, "block.captured-this.addr");
    Builder.CreateStore(LoadCXXThis(), addr);
  }
 
  // Next, captured variables.
  for (const auto &CI : blockDecl->captures()) {
    const VarDecl *variable = CI.getVariable();
    const CGBlockInfo::Capture &capture = blockInfo.getCapture(variable);
 
    // Ignore constant captures.
    if (capture.isConstant()) continue;
 
    QualType type = capture.fieldType();
 
    // This will be a [[type]]*, except that a byref entry will just be
    // an i8**.
    Address blockField = projectField(capture.getIndex(), "block.captured");
 
    // Compute the address of the thing we're going to move into the
    // block literal.
    Address src = Address::invalid();
 
    if (blockDecl->isConversionFromLambda()) {
      // The lambda capture in a lambda's conversion-to-block-pointer is
      // special; we'll simply emit it directly.
      src = Address::invalid();
    } else if (CI.isEscapingByref()) {
      if (BlockInfo && CI.isNested()) {
        // We need to use the capture from the enclosing block.
        const CGBlockInfo::Capture &enclosingCapture =
            BlockInfo->getCapture(variable);
 
        // This is a [[type]]*, except that a byref entry will just be an i8**.
        src = Builder.CreateStructGEP(LoadBlockStruct(),
                                      enclosingCapture.getIndex(),
                                      "block.capture.addr");
      } else {
        auto I = LocalDeclMap.find(variable);
        assert(I != LocalDeclMap.end());
        src = I->second;
      }
    } else {
      DeclRefExpr declRef(getContext(), const_cast<VarDecl *>(variable),
                          /*RefersToEnclosingVariableOrCapture*/ CI.isNested(),
                          type.getNonReferenceType(), VK_LValue,
                          SourceLocation());
      src = EmitDeclRefLValue(&declRef).getAddress();
    };
 
    // For byrefs, we just write the pointer to the byref struct into
    // the block field.  There's no need to chase the forwarding
    // pointer at this point, since we're building something that will
    // live a shorter life than the stack byref anyway.
    if (CI.isEscapingByref()) {
      // Get a void* that points to the byref struct.
      llvm::Value *byrefPointer;
      if (CI.isNested())
        byrefPointer = Builder.CreateLoad(src, "byref.capture");
      else
        byrefPointer = src.emitRawPointer(*this);
 
      // Write that void* into the capture field.
      Builder.CreateStore(byrefPointer, blockField);
 
    // If we have a copy constructor, evaluate that into the block field.
    } else if (const Expr *copyExpr = CI.getCopyExpr()) {
      if (blockDecl->isConversionFromLambda()) {
        // If we have a lambda conversion, emit the expression
        // directly into the block instead.
        AggValueSlot Slot =
            AggValueSlot::forAddr(blockField, Qualifiers(),
                                  AggValueSlot::IsDestructed,
                                  AggValueSlot::DoesNotNeedGCBarriers,
                                  AggValueSlot::IsNotAliased,
                                  AggValueSlot::DoesNotOverlap);
        EmitAggExpr(copyExpr, Slot);
      } else {
        EmitSynthesizedCXXCopyCtor(blockField, src, copyExpr);
      }
 
    // If it's a reference variable, copy the reference into the block field.
    } else if (type->getAs<ReferenceType>()) {
      Builder.CreateStore(src.emitRawPointer(*this), blockField);
 
      // If type is const-qualified, copy the value into the block field.
    } else if (type.isConstQualified() &&
               type.getObjCLifetime() == Qualifiers::OCL_Strong &&
               CGM.getCodeGenOpts().OptimizationLevel != 0) {
      llvm::Value *value = Builder.CreateLoad(src, "captured");
      Builder.CreateStore(value, blockField);
 
    // If this is an ARC __strong block-pointer variable, don't do a
    // block copy.
    //
    // TODO: this can be generalized into the normal initialization logic:
    // we should never need to do a block-copy when initializing a local
    // variable, because the local variable's lifetime should be strictly
    // contained within the stack block's.
    } else if (type.getObjCLifetime() == Qualifiers::OCL_Strong &&
               type->isBlockPointerType()) {
      // Load the block and do a simple retain.
      llvm::Value *value = Builder.CreateLoad(src, "block.captured_block");
      value = EmitARCRetainNonBlock(value);
 
      // Do a primitive store to the block field.
      Builder.CreateStore(value, blockField);
 
    // Otherwise, fake up a POD copy into the block field.
    } else {
      // Fake up a new variable so that EmitScalarInit doesn't think
      // we're referring to the variable in its own initializer.
      ImplicitParamDecl BlockFieldPseudoVar(getContext(), type,
                                            ImplicitParamKind::Other);
 
      // We use one of these or the other depending on whether the
      // reference is nested.
      DeclRefExpr declRef(getContext(), const_cast<VarDecl *>(variable),
                          /*RefersToEnclosingVariableOrCapture*/ CI.isNested(),
                          type, VK_LValue, SourceLocation());
 
      ImplicitCastExpr l2r(ImplicitCastExpr::OnStack, type, CK_LValueToRValue,
                           &declRef, VK_PRValue, FPOptionsOverride());
      // 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);
    }
 
    // Push a cleanup for the capture if necessary.
    if (!blockInfo.NoEscape && !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) {
    CGM.getOpenCLRuntime().recordBlockInfo(blockInfo.BlockExpression, InvokeFn,
                                           result, blockInfo.StructureType);
  }
 
  return result;
}
 
 
llvm::Type *CodeGenModule::getBlockDescriptorType() {
  if (BlockDescriptorType)
    return BlockDescriptorType;
 
  llvm::Type *UnsignedLongTy =
    getTypes().ConvertType(getContext().UnsignedLongTy);
 
  // struct __block_descriptor {
  //   unsigned long reserved;
  //   unsigned long block_size;
  //
  //   // later, the following will be added
  //
  //   struct {
  //     void (*copyHelper)();
  //     void (*copyHelper)();
  //   } helpers;                // !!! optional
  //
  //   const char *signature;   // the block signature
  //   const char *layout;      // reserved
  // };
  BlockDescriptorType = llvm::StructType::create(
      "struct.__block_descriptor", UnsignedLongTy, UnsignedLongTy);
 
  // Now form a pointer to that.
  unsigned AddrSpace = 0;
  if (getLangOpts().OpenCL)
    AddrSpace = getContext().getTargetAddressSpace(LangAS::opencl_constant);
  BlockDescriptorType = llvm::PointerType::get(BlockDescriptorType, AddrSpace);
  return BlockDescriptorType;
}
 
llvm::Type *CodeGenModule::getGenericBlockLiteralType() {
  if (GenericBlockLiteralType)
    return GenericBlockLiteralType;
 
  llvm::Type *BlockDescPtrTy = getBlockDescriptorType();
 
  if (getLangOpts().OpenCL) {
    // struct __opencl_block_literal_generic {
    //   int __size;
    //   int __align;
    //   __generic void *__invoke;
    //   /* custom fields */
    // };
    SmallVector<llvm::Type *, 8> StructFields(
        {IntTy, IntTy, getOpenCLRuntime().getGenericVoidPointerType()});
    if (auto *Helper = getTargetCodeGenInfo().getTargetOpenCLBlockHelper()) {
      llvm::append_range(StructFields, Helper->getCustomFieldTypes());
    }
    GenericBlockLiteralType = llvm::StructType::create(
        StructFields, "struct.__opencl_block_literal_generic");
  } else {
    // struct __block_literal_generic {
    //   void *__isa;
    //   int __flags;
    //   int __reserved;
    //   void (*__invoke)(void *);
    //   struct __block_descriptor *__descriptor;
    // };
    GenericBlockLiteralType =
        llvm::StructType::create("struct.__block_literal_generic", VoidPtrTy,
                                 IntTy, IntTy, VoidPtrTy, BlockDescPtrTy);
  }
 
  return GenericBlockLiteralType;
}
 
RValue CodeGenFunction::EmitBlockCallExpr(const CallExpr *E,
                                          ReturnValueSlot ReturnValue,
                                          llvm::CallBase **CallOrInvoke) {
  const auto *BPT = E->getCallee()->getType()->castAs<BlockPointerType>();
  llvm::Value *BlockPtr = EmitScalarExpr(E->getCallee());
  llvm::Type *GenBlockTy = CGM.getGenericBlockLiteralType();
  llvm::Value *Func = nullptr;
  QualType FnType = BPT->getPointeeType();
  ASTContext &Ctx = getContext();
  CallArgList Args;
 
  if (getLangOpts().OpenCL) {
    // For OpenCL, BlockPtr is already casted to generic block literal.
 
    // First argument of a block call is a generic block literal casted to
    // generic void pointer, i.e. i8 addrspace(4)*
    llvm::Type *GenericVoidPtrTy =
        CGM.getOpenCLRuntime().getGenericVoidPointerType();
    llvm::Value *BlockDescriptor = Builder.CreatePointerCast(
        BlockPtr, GenericVoidPtrTy);
    QualType VoidPtrQualTy = Ctx.getPointerType(
        Ctx.getAddrSpaceQualType(Ctx.VoidTy, LangAS::opencl_generic));
    Args.add(RValue::get(BlockDescriptor), VoidPtrQualTy);
    // And the rest of the arguments.
    EmitCallArgs(Args, FnType->getAs<FunctionProtoType>(), E->arguments());
 
    // We *can* call the block directly unless it is a function argument.
    if (!isa<ParmVarDecl>(E->getCalleeDecl()))
      Func = CGM.getOpenCLRuntime().getInvokeFunction(E->getCallee());
    else {
      llvm::Value *FuncPtr = Builder.CreateStructGEP(GenBlockTy, BlockPtr, 2);
      Func = Builder.CreateAlignedLoad(GenericVoidPtrTy, FuncPtr,
                                       getPointerAlign());
    }
  } else {
    // Bitcast the block literal to a generic block literal.
    BlockPtr =
        Builder.CreatePointerCast(BlockPtr, UnqualPtrTy, "block.literal");
    // Get pointer to the block invoke function
    llvm::Value *FuncPtr = Builder.CreateStructGEP(GenBlockTy, BlockPtr, 3);
 
    // First argument is a block literal casted to a void pointer
    BlockPtr = Builder.CreatePointerCast(BlockPtr, VoidPtrTy);
    Args.add(RValue::get(BlockPtr), Ctx.VoidPtrTy);
    // And the rest of the arguments.
    EmitCallArgs(Args, FnType->getAs<FunctionProtoType>(), E->arguments());
 
    // Load the function.
    Func = Builder.CreateAlignedLoad(VoidPtrTy, FuncPtr, getPointerAlign());
  }
 
  const FunctionType *FuncTy = FnType->castAs<FunctionType>();
  const CGFunctionInfo &FnInfo =
    CGM.getTypes().arrangeBlockFunctionCall(Args, FuncTy);
 
  // Prepare the callee.
  CGCallee Callee(CGCalleeInfo(), Func);
 
  // And call the block.
  return EmitCall(FnInfo, Callee, ReturnValue, Args, CallOrInvoke);
}
 
Address CodeGenFunction::GetAddrOfBlockDecl(const VarDecl *variable) {
  assert(BlockInfo && "evaluating block ref without block information?");
  const CGBlockInfo::Capture &capture = BlockInfo->getCapture(variable);
 
  // Handle constant captures.
  if (capture.isConstant()) return LocalDeclMap.find(variable)->second;
 
  Address addr = Builder.CreateStructGEP(LoadBlockStruct(), capture.getIndex(),
                                         "block.capture.addr");
 
  if (variable->isEscapingByref()) {
    // addr should be a void** right now.  Load, then cast the result
    // to byref*.
 
    auto &byrefInfo = getBlockByrefInfo(variable);
    addr = Address(Builder.CreateLoad(addr), byrefInfo.Type,
                   byrefInfo.ByrefAlignment);
 
    addr = emitBlockByrefAddress(addr, byrefInfo, /*follow*/ true,
                                 variable->getName());
  }
 
  assert((!variable->isNonEscapingByref() ||
          capture.fieldType()->isReferenceType()) &&
         "the capture field of a non-escaping variable should have a "
         "reference type");
  if (capture.fieldType()->isReferenceType())
    addr = EmitLoadOfReference(MakeAddrLValue(addr, capture.fieldType()));
 
  return addr;
}
 
void CodeGenModule::setAddrOfGlobalBlock(const BlockExpr *BE,
                                         llvm::Constant *Addr) {
  bool Ok = EmittedGlobalBlocks.insert(std::make_pair(BE, Addr)).second;
  (void)Ok;
  assert(Ok && "Trying to replace an already-existing global block!");
}
 
llvm::Constant *
CodeGenModule::GetAddrOfGlobalBlock(const BlockExpr *BE,
                                    StringRef Name) {
  if (llvm::Constant *Block = getAddrOfGlobalBlockIfEmitted(BE))
    return Block;
 
  CGBlockInfo blockInfo(BE->getBlockDecl(), Name);
  blockInfo.BlockExpression = BE;
 
  // Compute information about the layout, etc., of this block.
  computeBlockInfo(*this, nullptr, blockInfo);
 
  // Using that metadata, generate the actual block function.
  {
    CodeGenFunction::DeclMapTy LocalDeclMap;
    CodeGenFunction(*this).GenerateBlockFunction(
        GlobalDecl(), blockInfo, LocalDeclMap,
        /*IsLambdaConversionToBlock*/ false, /*BuildGlobalBlock*/ true);
  }
 
  return getAddrOfGlobalBlockIfEmitted(BE);
}
 
static llvm::Constant *buildGlobalBlock(CodeGenModule &CGM,
                                        const CGBlockInfo &blockInfo,
                                        llvm::Constant *blockFn) {
  assert(blockInfo.CanBeGlobal);
  // Callers should detect this case on their own: calling this function
  // generally requires computing layout information, which is a waste of time
  // if we've already emitted this block.
  assert(!CGM.getAddrOfGlobalBlockIfEmitted(blockInfo.BlockExpression) &&
         "Refusing to re-emit a global block.");
 
  // Generate the constants for the block literal initializer.
  ConstantInitBuilder builder(CGM);
  auto fields = builder.beginStruct();
 
  bool IsOpenCL = CGM.getLangOpts().OpenCL;
  bool IsWindows = CGM.getTarget().getTriple().isOSWindows();
  if (!IsOpenCL) {
    // isa
    if (IsWindows)
      fields.addNullPointer(CGM.Int8PtrPtrTy);
    else
      fields.add(CGM.getNSConcreteGlobalBlock());
 
    // __flags
    BlockFlags flags = BLOCK_IS_GLOBAL;
    if (!CGM.getCodeGenOpts().DisableBlockSignatureString)
      flags |= BLOCK_HAS_SIGNATURE;
    if (blockInfo.UsesStret)
      flags |= BLOCK_USE_STRET;
 
    fields.addInt(CGM.IntTy, flags.getBitMask());
 
    // Reserved
    fields.addInt(CGM.IntTy, 0);
  } else {
    fields.addInt(CGM.IntTy, blockInfo.BlockSize.getQuantity());
    fields.addInt(CGM.IntTy, blockInfo.BlockAlign.getQuantity());
  }
 
  // Function
  fields.add(blockFn);
 
  if (!IsOpenCL) {
    // Descriptor
    fields.add(buildBlockDescriptor(CGM, blockInfo));
  } else if (auto *Helper =
                 CGM.getTargetCodeGenInfo().getTargetOpenCLBlockHelper()) {
    for (auto *I : Helper->getCustomFieldValues(CGM, blockInfo)) {
      fields.add(I);
    }
  }
 
  unsigned AddrSpace = 0;
  if (CGM.getContext().getLangOpts().OpenCL)
    AddrSpace = CGM.getContext().getTargetAddressSpace(LangAS::opencl_global);
 
  llvm::GlobalVariable *literal = fields.finishAndCreateGlobal(
      "__block_literal_global", blockInfo.BlockAlign,
      /*constant*/ !IsWindows, llvm::GlobalVariable::InternalLinkage, AddrSpace);
 
  literal->addAttribute("objc_arc_inert");
 
  // Windows does not allow globals to be initialised to point to globals in
  // different DLLs.  Any such variables must run code to initialise them.
  if (IsWindows) {
    auto *Init = llvm::Function::Create(llvm::FunctionType::get(CGM.VoidTy,
          {}), llvm::GlobalValue::InternalLinkage, ".block_isa_init",
        &CGM.getModule());
    llvm::IRBuilder<> b(llvm::BasicBlock::Create(CGM.getLLVMContext(), "entry",
          Init));
    b.CreateAlignedStore(CGM.getNSConcreteGlobalBlock(),
                         b.CreateStructGEP(literal->getValueType(), literal, 0),
                         CGM.getPointerAlign().getAsAlign());
    b.CreateRetVoid();
    // We can't use the normal LLVM global initialisation array, because we
    // need to specify that this runs early in library initialisation.
    auto *InitVar = new llvm::GlobalVariable(CGM.getModule(), Init->getType(),
        /*isConstant*/true, llvm::GlobalValue::InternalLinkage,
        Init, ".block_isa_init_ptr");
    InitVar->setSection(".CRT$XCLa");
    CGM.addUsedGlobal(InitVar);
  }
 
  // Return a constant of the appropriately-casted type.
  llvm::Type *RequiredType =
    CGM.getTypes().ConvertType(blockInfo.getBlockExpr()->getType());
  llvm::Constant *Result =
      llvm::ConstantExpr::getPointerCast(literal, RequiredType);
  CGM.setAddrOfGlobalBlock(blockInfo.BlockExpression, Result);
  if (CGM.getContext().getLangOpts().OpenCL)
    CGM.getOpenCLRuntime().recordBlockInfo(
        blockInfo.BlockExpression,
        cast<llvm::Function>(blockFn->stripPointerCasts()), Result,
        literal->getValueType());
  return Result;
}
 
void CodeGenFunction::setBlockContextParameter(const ImplicitParamDecl *D,
                                               unsigned argNum,
                                               llvm::Value *arg) {
  assert(BlockInfo && "not emitting prologue of block invocation function?!");
 
  // Allocate a stack slot like for any local variable to guarantee optimal
  // debug info at -O0. The mem2reg pass will eliminate it when optimizing.
  RawAddress alloc = CreateMemTemp(D->getType(), D->getName() + ".addr");
  Builder.CreateStore(arg, alloc);
  if (CGDebugInfo *DI = getDebugInfo()) {
    if (CGM.getCodeGenOpts().hasReducedDebugInfo()) {
      DI->setLocation(D->getLocation());
      DI->EmitDeclareOfBlockLiteralArgVariable(
          *BlockInfo, D->getName(), argNum,
          cast<llvm::AllocaInst>(alloc.getPointer()->stripPointerCasts()),
          Builder);
    }
  }
 
  SourceLocation StartLoc = BlockInfo->getBlockExpr()->getBody()->getBeginLoc();
  ApplyDebugLocation Scope(*this, StartLoc);
 
  // Instead of messing around with LocalDeclMap, just set the value
  // directly as BlockPointer.
  BlockPointer = Builder.CreatePointerCast(
      arg,
      llvm::PointerType::get(
          getLLVMContext(),
          getContext().getLangOpts().OpenCL
              ? getContext().getTargetAddressSpace(LangAS::opencl_generic)
              : 0),
      "block");
}
 
Address CodeGenFunction::LoadBlockStruct() {
  assert(BlockInfo && "not in a block invocation function!");
  assert(BlockPointer && "no block pointer set!");
  return Address(BlockPointer,