Index: lib/Target/NVPTX/NVPTXAsmPrinter.cpp
===================================================================
--- lib/Target/NVPTX/NVPTXAsmPrinter.cpp
+++ lib/Target/NVPTX/NVPTXAsmPrinter.cpp
@@ -368,13 +368,13 @@
     } else if (isa<PointerType>(Ty)) {
       O << ".param .b" << TLI->getPointerTy(DL).getSizeInBits()
         << " func_retval0";
-    } else if ((Ty->getTypeID() == Type::StructTyID) || isa<VectorType>(Ty)) {
+    } else if (Ty->isAggregateType() || Ty->isVectorTy()) {
       unsigned totalsz = DL.getTypeAllocSize(Ty);
-       unsigned retAlignment = 0;
-       if (!llvm::getAlign(*F, 0, retAlignment))
-         retAlignment = DL.getABITypeAlignment(Ty);
-       O << ".param .align " << retAlignment << " .b8 func_retval0[" << totalsz
-         << "]";
+      unsigned retAlignment = 0;
+      if (!llvm::getAlign(*F, 0, retAlignment))
+        retAlignment = DL.getABITypeAlignment(Ty);
+      O << ".param .align " << retAlignment << " .b8 func_retval0[" << totalsz
+        << "]";
     } else
       llvm_unreachable("Unknown return type");
   } else {
Index: test/CodeGen/NVPTX/arg-lowering.ll
===================================================================
--- test/CodeGen/NVPTX/arg-lowering.ll
+++ test/CodeGen/NVPTX/arg-lowering.ll
@@ -11,3 +11,9 @@
 define <2 x float> @foo1({float, float, i64} %arg0) {
   ret <2 x float> <float 1.0, float 1.0>
 }
+
+; CHECK: .visible .func  (.param .align 4 .b8 func_retval0[16]) foo2(
+; CHECK:          .param .align 4 .b8 foo2_param_0[8]
+define [4 x float] @foo2({float, float} %arg0) {
+  ret [4 x float] [float 1.0, float 1.0, float 1.0, float 1.0]
+}