correctly handle PyCFunctions with METH_FASTCALL signature in Py3.6

Cython/Utility/ModuleSetupCode.c

@@ -188,6 +188,12 @@
 #ifndef Py_TPFLAGS_HAVE_FINALIZE
   #define Py_TPFLAGS_HAVE_FINALIZE 0
 #endif
+#ifndef METH_FASTCALL
+  // new in CPython 3.6
+  #define METH_FASTCALL 0x80
+  typedef PyObject *(*_PyCFunctionFast) (PyObject *self, PyObject **args,
+                                         Py_ssize_t nargs, PyObject *kwnames);
+#endif
 
 /* new Py3.3 unicode type (PEP 393) */
 #if PY_VERSION_HEX > 0x03030000 && defined(PyUnicode_KIND)

Cython/Utility/ObjectHandling.c

@@ -1126,7 +1126,7 @@ static int __Pyx_TryUnpackUnboundCMethod(__Pyx_CachedCFunction* target) {
     {
         PyMethodDescrObject *descr = (PyMethodDescrObject*) method;
         target->func = descr->d_method->ml_meth;
-        target->flag = descr->d_method->ml_flags & (METH_VARARGS | METH_KEYWORDS | METH_O | METH_NOARGS);
+        target->flag = descr->d_method->ml_flags & ~(METH_CLASS | METH_STATIC | METH_COEXIST);
     }
 #endif
     return 0;
@@ -1142,7 +1142,9 @@ static PyObject* __Pyx__CallUnboundCMethod0(__Pyx_CachedCFunction* cfunc, PyObje
     ((likely((cfunc)->func)) ? \
         (likely((cfunc)->flag == METH_NOARGS) ?  (*((cfunc)->func))(self, NULL) : \
          (likely((cfunc)->flag == (METH_VARARGS | METH_KEYWORDS)) ?  ((*(PyCFunctionWithKeywords)(cfunc)->func)(self, $empty_tuple, NULL)) : \
-             ((cfunc)->flag == METH_VARARGS ?  (*((cfunc)->func))(self, $empty_tuple) : __Pyx__CallUnboundCMethod0(cfunc, self)))) : \
+             ((cfunc)->flag == METH_VARARGS ?  (*((cfunc)->func))(self, $empty_tuple) : \
+              (PY_VERSION_HEX >= 0x030600B1 && (cfunc)->flag == METH_FASTCALL ?  (*(_PyCFunctionFast)(cfunc)->func)(self, &PyTuple_GET_ITEM($empty_tuple, 0), 0, NULL) : \
+                __Pyx__CallUnboundCMethod0(cfunc, self))))) : \
         __Pyx__CallUnboundCMethod0(cfunc, self))
 #else
 #define __Pyx_CallUnboundCMethod0(cfunc, self)  __Pyx__CallUnboundCMethod0(cfunc, self)
@@ -1178,7 +1180,9 @@ static PyObject* __Pyx__CallUnboundCMethod1(__Pyx_CachedCFunction* cfunc, PyObje
 #if CYTHON_COMPILING_IN_CPYTHON
 #define __Pyx_CallUnboundCMethod1(cfunc, self, arg)  \
     ((likely((cfunc)->func && (cfunc)->flag == METH_O)) ? (*((cfunc)->func))(self, arg) : \
-        __Pyx__CallUnboundCMethod1(cfunc, self, arg))
+        ((PY_VERSION_HEX >= 0x030600B1 && (cfunc)->func && (cfunc)->flag == METH_FASTCALL) ?
+            (*(_PyCFunctionFast)(cfunc)->func)(self, &arg, 1, NULL) : \
+        __Pyx__CallUnboundCMethod1(cfunc, self, arg)))
 #else
 #define __Pyx_CallUnboundCMethod1(cfunc, self, arg)  __Pyx__CallUnboundCMethod1(cfunc, self, arg)
 #endif