Make EmitAggregateCopy take an alignment argument. Make EmitFinalDestCopy pass in the correct alignment when known.

The test includes a FIXME for a related case involving calls; it's a bit more complicated to fix because the RValue class doesn't keep track of alignment.

<rdar://problem/10463337>
  

llvm-svn: 145862

clang/lib/CodeGen/CGExprAgg.cpp

@@ -74,7 +74,8 @@ public:
 
   /// EmitFinalDestCopy - Perform the final copy to DestPtr, if desired.
   void EmitFinalDestCopy(const Expr *E, LValue Src, bool Ignore = false);
-  void EmitFinalDestCopy(const Expr *E, RValue Src, bool Ignore = false);
+  void EmitFinalDestCopy(const Expr *E, RValue Src, bool Ignore = false,
+                         unsigned Alignment = 0);
 
   void EmitMoveFromReturnSlot(const Expr *E, RValue Src);
 
@@ -221,7 +222,8 @@ void AggExprEmitter::EmitMoveFromReturnSlot(const Expr *E, RValue Src) {
 }
 
 /// EmitFinalDestCopy - Perform the final copy to DestPtr, if desired.
-void AggExprEmitter::EmitFinalDestCopy(const Expr *E, RValue Src, bool Ignore) {
+void AggExprEmitter::EmitFinalDestCopy(const Expr *E, RValue Src, bool Ignore,
+                                       unsigned Alignment) {
   assert(Src.isAggregate() && "value must be aggregate value!");
 
   // If Dest is ignored, then we're evaluating an aggregate expression
@@ -256,14 +258,16 @@ void AggExprEmitter::EmitFinalDestCopy(const Expr *E, RValue Src, bool Ignore) {
   // from the source as well, as we can't eliminate it if either operand
   // is volatile, unless copy has volatile for both source and destination..
   CGF.EmitAggregateCopy(Dest.getAddr(), Src.getAggregateAddr(), E->getType(),
-                        Dest.isVolatile()|Src.isVolatileQualified());
+                        Dest.isVolatile()|Src.isVolatileQualified(),
+                        Alignment);
 }
 
 /// EmitFinalDestCopy - Perform the final copy to DestPtr, if desired.
 void AggExprEmitter::EmitFinalDestCopy(const Expr *E, LValue Src, bool Ignore) {
   assert(Src.isSimple() && "Can't have aggregate bitfield, vector, etc");
 
-  EmitFinalDestCopy(E, Src.asAggregateRValue(), Ignore);
+  CharUnits Alignment = std::min(Src.getAlignment(), Dest.getAlignment());
+  EmitFinalDestCopy(E, Src.asAggregateRValue(), Ignore, Alignment.getQuantity());
 }
 
 //===----------------------------------------------------------------------===//
@@ -1054,7 +1058,7 @@ LValue CodeGenFunction::EmitAggExprToLValue(const Expr *E) {
 
 void CodeGenFunction::EmitAggregateCopy(llvm::Value *DestPtr,
                                         llvm::Value *SrcPtr, QualType Ty,
-                                        bool isVolatile) {
+                                        bool isVolatile, unsigned Alignment) {
   assert(!Ty->isAnyComplexType() && "Shouldn't happen for complex");
 
   if (getContext().getLangOptions().CPlusPlus) {
@@ -1087,6 +1091,9 @@ void CodeGenFunction::EmitAggregateCopy(llvm::Value *DestPtr,
   std::pair<CharUnits, CharUnits> TypeInfo = 
     getContext().getTypeInfoInChars(Ty);
 
+  if (!Alignment)
+    Alignment = TypeInfo.second.getQuantity();
+
   // FIXME: Handle variable sized types.
 
   // FIXME: If we have a volatile struct, the optimizer can remove what might
@@ -1143,5 +1150,5 @@ void CodeGenFunction::EmitAggregateCopy(llvm::Value *DestPtr,
   Builder.CreateMemCpy(DestPtr, SrcPtr,
                        llvm::ConstantInt::get(IntPtrTy, 
                                               TypeInfo.first.getQuantity()),
-                       TypeInfo.second.getQuantity(), isVolatile);
+                       Alignment, isVolatile);
 }

clang/lib/CodeGen/CodeGenFunction.h

@@ -1624,7 +1624,8 @@ public:
   /// \param isVolatile - True iff either the source or the destination is
   /// volatile.
   void EmitAggregateCopy(llvm::Value *DestPtr, llvm::Value *SrcPtr,
-                         QualType EltTy, bool isVolatile=false);
+                         QualType EltTy, bool isVolatile=false,
+                         unsigned Alignment = 0);
 
   /// StartBlock - Start new block named N. If insert block is a dummy block
   /// then reuse it.

clang/test/CodeGen/packed-nest-unpacked.c

@@ -0,0 +1,31 @@
+// RUN: %clang_cc1 %s -triple x86_64-apple-macosx10.7.2 -emit-llvm -o - | FileCheck %s
+// <rdar://problem/10463337>
+
+struct X { int x[6]; };
+struct Y { char x[13]; struct X y; } __attribute((packed));
+struct Y g;
+void f(struct X);
+
+struct X test1() {
+  // CHECK: @test1
+  // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* {{.*}}, i8* bitcast (%struct.X* getelementptr inbounds (%struct.Y* @g, i32 0, i32 1) to i8*), i64 24, i32 1, i1 false)
+  return g.y;
+}
+struct X test2() {
+  // CHECK: @test2
+  // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* {{.*}}, i8* bitcast (%struct.X* getelementptr inbounds (%struct.Y* @g, i32 0, i32 1) to i8*), i64 24, i32 1, i1 false)
+  struct X a = g.y;
+  return a;
+}
+
+void test3(struct X a) {
+  // CHECK: @test3
+  // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* bitcast (%struct.X* getelementptr inbounds (%struct.Y* @g, i32 0, i32 1) to i8*), i8* {{.*}}, i64 24, i32 1, i1 false)
+  g.y = a;
+}
+
+void test4() {
+  // CHECK: @test4
+  // FIXME: call void @llvm.memcpy.p0i8.p0i8.i64(i8* {{.*}}, i8* bitcast (%struct.X* getelementptr inbounds (%struct.Y* @g, i32 0, i32 1) to i8*), i64 24, i32 1, i1 false)
+  f(g.y);
+}