diff --git a/.gitmodules b/.gitmodules
index 42ee841..3b339d9 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -1,3 +1,4 @@
 [submodule "external/cudawrappers"]
 	path = external/cudawrappers
 	url = https://github.com/nlesc-recruit/cudawrappers
+	branch = main
diff --git a/CMakeLists.txt b/CMakeLists.txt
index a6b9e2f..7727881 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -14,10 +14,14 @@ set(CMAKE_BUILD_TYPE Release)
 include(GNUInstallDirs)
 list(APPEND CMAKE_INSTALL_RPATH ${CMAKE_INSTALL_FULL_LIBDIR})
 
+set(CUDA_MIN_VERSION 10.0)
+find_package(CUDAToolkit ${CUDA_MIN_VERSION} REQUIRED)
+
 add_subdirectory(external/cudawrappers)
 
 add_executable(vector_add_example vector_add.cpp)
 
+target_embed_source(vector_add_example vector_add_kernel.cu)
 target_link_libraries(vector_add_example PRIVATE cudawrappers::cu
                                                  cudawrappers::nvrtc)
 
diff --git a/vector_add.cpp b/vector_add.cpp
index 045e205..b3afae3 100644
--- a/vector_add.cpp
+++ b/vector_add.cpp
@@ -31,8 +31,12 @@ void vector_add() {
   my_stream.memcpyHtoDAsync(d_b, b, bytesize);
 
   // compile kernel
+  extern const char _binary_vector_add_kernel_cu_start,
+      _binary_vector_add_kernel_cu_end;
+  const std::string kernel(&_binary_vector_add_kernel_cu_start,
+                           &_binary_vector_add_kernel_cu_end);
   std::vector<std::string> options = {};
-  nvrtc::Program program("vector_add_kernel.cu");
+  nvrtc::Program program(kernel, "vector_add_kernel.cu");
   try {
     program.compile(options);
   } catch (nvrtc::Error &error) {
@@ -40,7 +44,7 @@ void vector_add() {
     throw;
   }
 
-  cu::Module module = cu::Module((void *)program.getPTX().data());
+  cu::Module module(static_cast<const void *>(program.getPTX().data()));
 
   cu::Function function(module, "vector_add");