diff --git a/src/device/op128.h b/src/device/op128.h
index 8ee0b4224..72f90d0c1 100644
--- a/src/device/op128.h
+++ b/src/device/op128.h
@@ -199,8 +199,7 @@ template<> __device__ __forceinline__ void st_global<0>(uintptr_t addr, BytePack
   } \
   template<> \
   __device__ __forceinline__ void st_##space<bytes>(addr_cxx_ty addr, BytePack<bytes> value) { \
-    data_cxx_ty tmp = __builtin_nontemporal_load((data_cxx_ty *)&value.native); \
-    __builtin_nontemporal_store(tmp, (data_cxx_ty *)addr); \
+    __builtin_nontemporal_store(value.native, (data_cxx_ty *)addr); \
   }
 
 // #if __CUDA_ARCH__ >= 700