Make tuple checks faster (pytorch#16657)

Summary:
As the comment indicates, the issue is only present in some versions of
Python 2, so we should be able to use heavily optimized PyTuple_Check in
most cases, and skip allocation of the strings, and unnecessary lookups
on object's type.

cc ezyang zasdfgbnm
Pull Request resolved: pytorch#16657

Differential Revision: D13957854

Pulled By: ezyang

fbshipit-source-id: be32eb473ad77a0805e8247d8d583d673d4bdf25

torch/csrc/utils/six.h

@@ -10,8 +10,14 @@ namespace six {
 // by a pytorch operator.
 
 inline bool isTuple(pybind11::handle input) {
-  std::string m = pybind11::str(input.get_type().attr("__module__"));
-  return pybind11::isinstance<pybind11::tuple>(input) || m == "torch.return_types";
+  if (PyTuple_Check(input.ptr())) {
+    return true;
+  }
+#if PY_MAJOR_VERSION == 2
+  return pybind11::cast<std::string>(input.get_type().attr("__module__")) == "torch.return_types";
+#else
+  return false;
+#endif
 }
 
 inline bool isTuple(PyObject* obj) {