Skip extra copy from aggregate where it isn't necessary; rdar://problem/8139919 . This shouldn't make much of a difference at -O3, but should substantially reduce the number of generated memcpy's at -O0.

llvm-svn: 130717

1 files changed, 13 insertions, 0 deletions

diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index a765f0f3439..8e3856131a9 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -1166,6 +1166,15 @@ void CodeGenFunction::EmitCallArg(CallArgList &args, const Expr *E,
     return args.add(EmitReferenceBindingToExpr(E, /*InitializedDecl=*/0),
                     type);
 
+  if (hasAggregateLLVMType(type) && isa<ImplicitCastExpr>(E) &&
+      cast<CastExpr>(E)->getCastKind() == CK_LValueToRValue) {
+    LValue L = EmitLValue(cast<CastExpr>(E)->getSubExpr());
+    assert(L.isSimple());
+    args.add(RValue::getAggregate(L.getAddress(), L.isVolatileQualified()),
+             type, /*NeedsCopy*/true);
+    return;
+  }
+
   args.add(EmitAnyExprToTemp(E), type);
 }
 
@@ -1231,6 +1240,10 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
                             Alignment, I->Ty);
         else
           StoreComplexToAddr(RV.getComplexVal(), Args.back(), false);
+      } else if (I->NeedsCopy && !ArgInfo.getIndirectByVal()) {
+        Args.push_back(CreateMemTemp(I->Ty));
+        EmitAggregateCopy(Args.back(), RV.getAggregateAddr(), I->Ty,
+                          RV.isVolatileQualified());
       } else {
         Args.push_back(RV.getAggregateAddr());
       }