diff --git a/dnn/src/common/opr_delegate.h b/dnn/src/common/opr_delegate.h
index d5e96b4b..58a0667c 100644
--- a/dnn/src/common/opr_delegate.h
+++ b/dnn/src/common/opr_delegate.h
@@ -24,7 +24,8 @@ namespace megdnn {
  * Usually used for calling other opr impls from some opr impl. You probably
  * want to use CpuOprDelegationStorage instead.
  */
-const std::shared_ptr<Handle>& inplace_cpu_handle(int debug_level = 0);
+MGE_WIN_DECLSPEC_FUC const std::shared_ptr<Handle>& inplace_cpu_handle(
+        int debug_level = 0);
 
 /*!
  * \brief storage for oprs on inplace CPU handle
diff --git a/imperative/python/src/module.cpp b/imperative/python/src/module.cpp
index ef0e3135..26c57093 100644
--- a/imperative/python/src/module.cpp
+++ b/imperative/python/src/module.cpp
@@ -14,6 +14,7 @@
 #define DO_IMPORT_ARRAY
 #include "./helper.h"
 #include "./numpy_dtypes.h"
+#include "megdnn/handle.h"
 
 #include "./common.h"
 #include "./graph_rt.h"
@@ -30,6 +31,10 @@ using namespace mgb::imperative::python;
 #define MODULE_NAME imperative_rt
 #endif
 
+namespace megdnn {
+extern const std::shared_ptr<Handle>& inplace_cpu_handle(int debug_level = 0);
+}
+
 PYBIND11_MODULE(MODULE_NAME, m) {
     // initialize numpy
     if ([]() {
@@ -39,6 +44,8 @@ PYBIND11_MODULE(MODULE_NAME, m) {
         throw py::error_already_set();
     }
 
+    megdnn::inplace_cpu_handle();
+
     py::module::import("sys").attr("modules")[m.attr("__name__")] = m;
 
     m.attr("__package__") = m.attr("__name__");