diff -r f4dcb5ff3ecd Include/methodobject.h
--- a/Include/methodobject.h	Thu Aug 25 00:39:34 2016 +0200
+++ b/Include/methodobject.h	Thu Aug 25 00:41:37 2016 +0200
@@ -42,6 +42,11 @@ PyAPI_FUNC(PyObject *) _PyCFunction_Fast
     PyObject **args,
     Py_ssize_t nargs,
     PyObject *kwargs);
+
+PyAPI_FUNC(PyObject *) _PyCFunction_FastCallKeywords(PyObject *func,
+    PyObject **stack,
+    Py_ssize_t nargs,
+    Py_ssize_t nkwargs);
 #endif
 
 struct PyMethodDef {
diff -r f4dcb5ff3ecd Objects/methodobject.c
--- a/Objects/methodobject.c	Thu Aug 25 00:39:34 2016 +0200
+++ b/Objects/methodobject.c	Thu Aug 25 00:41:37 2016 +0200
@@ -243,6 +243,19 @@ PyObject *
     return result;
 }
 
+PyObject *
+_PyCFunction_FastCallKeywords(PyObject *func, PyObject **stack,
+                              Py_ssize_t nargs, Py_ssize_t nkwargs)
+{
+    assert((nargs == 0 && nkwargs == 0) || stack != NULL);
+
+    if (nkwargs == 0) {
+        return _PyCFunction_FastCallDict(func, stack, nargs, NULL);
+    }
+
+    return _PyObject_FastCallKeywords(func, stack, nargs, nkwargs);
+}
+
 /* Methods (the standard built-in methods, that is) */
 
 static void
diff -r f4dcb5ff3ecd Python/ceval.c
--- a/Python/ceval.c	Thu Aug 25 00:39:34 2016 +0200
+++ b/Python/ceval.c	Thu Aug 25 00:41:37 2016 +0200
@@ -113,7 +113,7 @@ static PyObject * call_function(PyObject
 #else
 static PyObject * call_function(PyObject ***, int);
 #endif
-static PyObject * do_call(PyObject *, PyObject ***, Py_ssize_t, Py_ssize_t);
+static PyObject * do_call(PyObject *, PyObject **, Py_ssize_t, Py_ssize_t);
 static PyObject * ext_do_call(PyObject *, PyObject ***, int, Py_ssize_t, Py_ssize_t);
 static PyObject * update_keyword_args(PyObject *, Py_ssize_t, PyObject ***,
                                       PyObject *);
@@ -4769,7 +4769,7 @@ call_function(PyObject ***pp_stack, int 
             x = _PyFunction_FastCallKeywords(func, (*pp_stack) - n, nargs, nkwargs);
         }
         else {
-            x = do_call(func, pp_stack, nargs, nkwargs);
+            x = do_call(func, *(pp_stack) - n, nargs, nkwargs);
         }
         READ_TIMESTAMP(*pintr1);
         Py_DECREF(func);
@@ -5078,20 +5078,10 @@ load_args(PyObject ***pp_stack, Py_ssize
 }
 
 static PyObject *
-do_call(PyObject *func, PyObject ***pp_stack, Py_ssize_t nargs, Py_ssize_t nkwargs)
+do_call(PyObject *func, PyObject **stack, Py_ssize_t nargs, Py_ssize_t nkwargs)
 {
-    PyObject *callargs = NULL;
-    PyObject *kwdict = NULL;
-    PyObject *result = NULL;
-
-    if (nkwargs > 0) {
-        kwdict = update_keyword_args(NULL, nkwargs, pp_stack, func);
-        if (kwdict == NULL)
-            goto call_fail;
-    }
-    callargs = load_args(pp_stack, nargs);
-    if (callargs == NULL)
-        goto call_fail;
+    PyObject *result;
+
 #ifdef CALL_PROFILE
     /* At this point, we have to look at the type of func to
        update the call stats properly.  Do it here so as to avoid
@@ -5108,15 +5098,14 @@ do_call(PyObject *func, PyObject ***pp_s
     else
         PCALL(PCALL_OTHER);
 #endif
+
     if (PyCFunction_Check(func)) {
         PyThreadState *tstate = PyThreadState_GET();
-        C_TRACE(result, PyCFunction_Call(func, callargs, kwdict));
+        C_TRACE(result, _PyCFunction_FastCallKeywords(func, stack, nargs, nkwargs));
     }
-    else
-        result = PyObject_Call(func, callargs, kwdict);
-call_fail:
-    Py_XDECREF(callargs);
-    Py_XDECREF(kwdict);
+    else {
+        result = _PyObject_FastCallKeywords(func, stack, nargs, nkwargs);
+    }
     return result;
 }