diff --git a/libcudacxx/test/libcudacxx/cuda/barrier/cp_async_bulk.pass.cpp b/libcudacxx/test/libcudacxx/cuda/barrier/cp_async_bulk.pass.cpp
index d5d54876a7f..5f26c80602b 100644
--- a/libcudacxx/test/libcudacxx/cuda/barrier/cp_async_bulk.pass.cpp
+++ b/libcudacxx/test/libcudacxx/cuda/barrier/cp_async_bulk.pass.cpp
@@ -26,7 +26,7 @@ using barrier = cuda::barrier<cuda::thread_scope_block>;
 namespace cde = cuda::device::experimental;
 
 static constexpr int buf_len = 1024;
-__device__ int gmem_buffer[buf_len];
+__device__ alignas(128) int gmem_buffer[buf_len];
 
 __device__ void test()
 {