1 files changed, 111 insertions, 25 deletions
diff --git a/lib/CodeGen/CodeGenFunction.cpp b/lib/CodeGen/CodeGenFunction.cpp
index 2939062..1d02861 100644
--- a/lib/CodeGen/CodeGenFunction.cpp
+++ b/lib/CodeGen/CodeGenFunction.cpp
@@ -23,12 +23,12 @@
 #include "clang/AST/StmtCXX.h"
 #include "clang/Frontend/CodeGenOptions.h"
 #include "llvm/Intrinsics.h"
-#include "llvm/Support/MDBuilder.h"
+#include "llvm/MDBuilder.h"
 #include "llvm/Target/TargetData.h"
 using namespace clang;
 using namespace CodeGen;
 
-CodeGenFunction::CodeGenFunction(CodeGenModule &cgm)
+CodeGenFunction::CodeGenFunction(CodeGenModule &cgm, bool suppressNewContext)
   : CodeGenTypeCache(cgm), CGM(cgm),
     Target(CGM.getContext().getTargetInfo()),
     Builder(cgm.getModule().getContext()),
@@ -42,7 +42,8 @@ CodeGenFunction::CodeGenFunction(CodeGenModule &cgm)
     TerminateHandler(0), TrapBB(0) {
 
   CatchUndefined = getContext().getLangOpts().CatchUndefined;
-  CGM.getCXXABI().getMangleContext().startNewFunction();
+  if (!suppressNewContext)
+    CGM.getCXXABI().getMangleContext().startNewFunction();
 }
 
 CodeGenFunction::~CodeGenFunction() {
@@ -251,6 +252,81 @@ void CodeGenFunction::EmitMCountInstrumentation() {
   Builder.CreateCall(MCountFn);
 }
 
+// OpenCL v1.2 s5.6.4.6 allows the compiler to store kernel argument
+// information in the program executable. The argument information stored
+// includes the argument name, its type, the address and access qualifiers used.
+// FIXME: Add type, address, and access qualifiers.
+static void GenOpenCLArgMetadata(const FunctionDecl *FD, llvm::Function *Fn,
+                                 CodeGenModule &CGM,llvm::LLVMContext &Context,
+                                 llvm::SmallVector <llvm::Value*, 5> &kernelMDArgs) {
+  
+  // Create MDNodes that represents the kernel arg metadata.
+  // Each MDNode is a list in the form of "key", N number of values which is
+  // the same number of values as their are kernel arguments.
+  
+  // MDNode for the kernel argument names.
+  SmallVector<llvm::Value*, 8> argNames;
+  argNames.push_back(llvm::MDString::get(Context, "kernel_arg_name"));
+  
+  for (unsigned i = 0, e = FD->getNumParams(); i != e; ++i) {
+    const ParmVarDecl *parm = FD->getParamDecl(i);
+    
+    // Get argument name.
+    argNames.push_back(llvm::MDString::get(Context, parm->getName()));
+    
+  }
+  // Add MDNode to the list of all metadata.
+  kernelMDArgs.push_back(llvm::MDNode::get(Context, argNames));
+}
+
+void CodeGenFunction::EmitOpenCLKernelMetadata(const FunctionDecl *FD, 
+                                               llvm::Function *Fn)
+{
+  if (!FD->hasAttr<OpenCLKernelAttr>())
+    return;
+
+  llvm::LLVMContext &Context = getLLVMContext();
+
+  llvm::SmallVector <llvm::Value*, 5> kernelMDArgs;
+  kernelMDArgs.push_back(Fn);
+
+  if (CGM.getCodeGenOpts().EmitOpenCLArgMetadata)
+    GenOpenCLArgMetadata(FD, Fn, CGM, Context, kernelMDArgs);
+  
+  if (FD->hasAttr<WorkGroupSizeHintAttr>()) {
+    llvm::SmallVector <llvm::Value*, 5> attrMDArgs;
+    attrMDArgs.push_back(llvm::MDString::get(Context, "work_group_size_hint"));
+    WorkGroupSizeHintAttr *attr = FD->getAttr<WorkGroupSizeHintAttr>();
+    llvm::Type *iTy = llvm::IntegerType::get(Context, 32);
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getXDim())));
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getYDim())));
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getZDim())));
+    kernelMDArgs.push_back(llvm::MDNode::get(Context, attrMDArgs));
+  }
+
+  if (FD->hasAttr<ReqdWorkGroupSizeAttr>()) {
+    llvm::SmallVector <llvm::Value*, 5> attrMDArgs;
+    attrMDArgs.push_back(llvm::MDString::get(Context, "reqd_work_group_size"));
+    ReqdWorkGroupSizeAttr *attr = FD->getAttr<ReqdWorkGroupSizeAttr>();
+    llvm::Type *iTy = llvm::IntegerType::get(Context, 32);
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getXDim())));
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getYDim())));
+    attrMDArgs.push_back(llvm::ConstantInt::get(iTy,
+       llvm::APInt(32, (uint64_t)attr->getZDim())));
+    kernelMDArgs.push_back(llvm::MDNode::get(Context, attrMDArgs));
+  }
+
+  llvm::MDNode *kernelMDNode = llvm::MDNode::get(Context, kernelMDArgs);
+  llvm::NamedMDNode *OpenCLKernelMetadata =
+    CGM.getModule().getOrInsertNamedMetadata("opencl.kernels");
+  OpenCLKernelMetadata->addOperand(kernelMDNode);
+}
+
 void CodeGenFunction::StartFunction(GlobalDecl GD, QualType RetTy,
                                     llvm::Function *Fn,
                                     const CGFunctionInfo &FnInfo,
@@ -279,14 +355,7 @@ void CodeGenFunction::StartFunction(GlobalDecl GD, QualType RetTy,
   if (getContext().getLangOpts().OpenCL) {
     // Add metadata for a kernel function.
     if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D))
-      if (FD->hasAttr<OpenCLKernelAttr>()) {
-        llvm::LLVMContext &Context = getLLVMContext();
-        llvm::NamedMDNode *OpenCLMetadata = 
-          CGM.getModule().getOrInsertNamedMetadata("opencl.kernels");
-          
-        llvm::Value *Op = Fn;
-        OpenCLMetadata->addOperand(llvm::MDNode::get(Context, Op));
-      }
+      EmitOpenCLKernelMetadata(FD, Fn);
   }
 
   llvm::BasicBlock *EntryBB = createBasicBlock("entry", CurFn);
@@ -537,7 +606,7 @@ bool CodeGenFunction::containsBreak(const Stmt *S) {
 /// constant folds return true and set the boolean result in Result.
 bool CodeGenFunction::ConstantFoldsToSimpleInteger(const Expr *Cond,
                                                    bool &ResultBool) {
-  llvm::APInt ResultInt;
+  llvm::APSInt ResultInt;
   if (!ConstantFoldsToSimpleInteger(Cond, ResultInt))
     return false;
   
@@ -549,7 +618,7 @@ bool CodeGenFunction::ConstantFoldsToSimpleInteger(const Expr *Cond,
 /// to a constant, or if it does but contains a label, return false.  If it
 /// constant folds return true and set the folded value.
 bool CodeGenFunction::
-ConstantFoldsToSimpleInteger(const Expr *Cond, llvm::APInt &ResultInt) {
+ConstantFoldsToSimpleInteger(const Expr *Cond, llvm::APSInt &ResultInt) {
   // FIXME: Rename and handle conversion of other evaluatable things
   // to bool.
   llvm::APSInt Int;
@@ -687,10 +756,10 @@ void CodeGenFunction::ErrorUnsupported(const Stmt *S, const char *Type,
 /// emitNonZeroVLAInit - Emit the "zero" initialization of a
 /// variable-length array whose elements have a non-zero bit-pattern.
 ///
+/// \param baseType the inner-most element type of the array
 /// \param src - a char* pointing to the bit-pattern for a single
 /// base element of the array
 /// \param sizeInChars - the total size of the VLA, in chars
-/// \param align - the total alignment of the VLA
 static void emitNonZeroVLAInit(CodeGenFunction &CGF, QualType baseType,
                                llvm::Value *dest, llvm::Value *src, 
                                llvm::Value *sizeInChars) {
@@ -881,33 +950,49 @@ llvm::Value *CodeGenFunction::emitArrayLength(const ArrayType *origArrayType,
   llvm::ConstantInt *zero = Builder.getInt32(0);
   gepIndices.push_back(zero);
 
-  // It's more efficient to calculate the count from the LLVM
-  // constant-length arrays than to re-evaluate the array bounds.
   uint64_t countFromCLAs = 1;
+  QualType eltType;
 
   llvm::ArrayType *llvmArrayType =
-    cast<llvm::ArrayType>(
+    dyn_cast<llvm::ArrayType>(
       cast<llvm::PointerType>(addr->getType())->getElementType());
-  while (true) {
+  while (llvmArrayType) {
     assert(isa<ConstantArrayType>(arrayType));
     assert(cast<ConstantArrayType>(arrayType)->getSize().getZExtValue()
              == llvmArrayType->getNumElements());
 
     gepIndices.push_back(zero);
     countFromCLAs *= llvmArrayType->getNumElements();
+    eltType = arrayType->getElementType();
 
     llvmArrayType =
       dyn_cast<llvm::ArrayType>(llvmArrayType->getElementType());
-    if (!llvmArrayType) break;
-
     arrayType = getContext().getAsArrayType(arrayType->getElementType());
-    assert(arrayType && "LLVM and Clang types are out-of-synch");
+    assert((!llvmArrayType || arrayType) &&
+           "LLVM and Clang types are out-of-synch");
   }
 
-  baseType = arrayType->getElementType();
+  if (arrayType) {
+    // From this point onwards, the Clang array type has been emitted
+    // as some other type (probably a packed struct). Compute the array
+    // size, and just emit the 'begin' expression as a bitcast.
+    while (arrayType) {
+      countFromCLAs *=
+          cast<ConstantArrayType>(arrayType)->getSize().getZExtValue();
+      eltType = arrayType->getElementType();
+      arrayType = getContext().getAsArrayType(eltType);
+    }
+
+    unsigned AddressSpace =
+        cast<llvm::PointerType>(addr->getType())->getAddressSpace();
+    llvm::Type *BaseType = ConvertType(eltType)->getPointerTo(AddressSpace);
+    addr = Builder.CreateBitCast(addr, BaseType, "array.begin");
+  } else {
+    // Create the actual GEP.
+    addr = Builder.CreateInBoundsGEP(addr, gepIndices, "array.begin");
+  }
 
-  // Create the actual GEP.
-  addr = Builder.CreateInBoundsGEP(addr, gepIndices, "array.begin");
+  baseType = eltType;
 
   llvm::Value *numElements
     = llvm::ConstantInt::get(SizeTy, countFromCLAs);
@@ -1071,7 +1156,8 @@ void CodeGenFunction::EmitDeclRefExprDbgValue(const DeclRefExpr *E,
                                               llvm::Constant *Init) {
   assert (Init && "Invalid DeclRefExpr initializer!");
   if (CGDebugInfo *Dbg = getDebugInfo())
-    Dbg->EmitGlobalVariable(E->getDecl(), Init);
+    if (CGM.getCodeGenOpts().DebugInfo >= CodeGenOptions::LimitedDebugInfo)
+      Dbg->EmitGlobalVariable(E->getDecl(), Init);
 }
 
 CodeGenFunction::PeepholeProtection