bpo-37207: Use PEP 590 vectorcall to speed up range(), list() and dict() by about 30%

Misc/NEWS.d/next/Core and Builtins/2019-06-09-10-54-31.bpo-37207.bLjgLR.rst

@@ -0,0 +1,2 @@
+Speed up calls to ``range()``, ``list()`` and ``dict()`` by about 30%, by
+using the PEP 590 ``vectorcall`` calling convention.

Objects/dictobject.c

@@ -3286,6 +3286,55 @@ dict_init(PyObject *self, PyObject *args, PyObject *kwds)
  return dict_update_common(self, args, kwds, "dict");
 }
 
+static PyObject *
+dict_vectorcall(PyObject *type, PyObject * const*args,
+ size_t nargsf, PyObject *kwnames)
+{
+ size_t nargs = PyVectorcall_NARGS(nargsf);
+ if (nargs > 1) {
+ PyErr_Format(PyExc_TypeError, "dict() expected at most 1 argument, got %zu", nargs);
+ return NULL;
+ }
+ assert(PyType_Check(type));
+ PyObject *self = dict_new((PyTypeObject *)type, NULL, NULL);
+ if (self == NULL) {
+ return NULL;
+ }
+ if (nargs == 1) {
+ int err = 0;
+ PyObject *arg = args[0];
+ _Py_IDENTIFIER(keys);
+ PyObject *func;
+ if (_PyObject_LookupAttrId(arg, &PyId_keys, &func) < 0) {
+ err = -1;
+ }
+ else if (func != NULL) {
+ Py_DECREF(func);
+ err = PyDict_Merge(self, arg, 1);
+ }
+ else {
+ err = PyDict_MergeFromSeq2(self, arg, 1);
+ }
+ if (err < 0) {
+ Py_DECREF(self);
+ return NULL;
+ }
+ args++;
+ }
+ if (kwnames == NULL) {
+ return self;
+ }
+ Py_ssize_t items = PyTuple_GET_SIZE(kwnames);
+ for (Py_ssize_t index = 0; index < items; index++) {
+ int err = PyDict_SetItem(self, PyTuple_GET_ITEM(kwnames, index), args[index]);
+ if (err < 0) {
+ Py_DECREF(self);
+ return NULL;
+ }
+ }
+ return self;
+}
+
 static PyObject *
 dict_iter(PyDictObject *dict)
 {
@@ -3344,6 +3393,7 @@ PyTypeObject PyDict_Type = {
  PyType_GenericAlloc, /* tp_alloc */
  dict_new, /* tp_new */
  PyObject_GC_Del, /* tp_free */
+ .tp_vectorcall = dict_vectorcall,
 };
 
 PyObject *

Objects/listobject.c

@@ -2726,6 +2726,34 @@ list___init___impl(PyListObject *self, PyObject *iterable)
  return 0;
 }
 
+static PyObject *
+list_vectorcall(PyObject *type, PyObject * const*args,
+ size_t nargsf, PyObject *kwnames)
+{
+ if (kwnames && PyTuple_GET_SIZE(kwnames) != 0) {
+ PyErr_Format(PyExc_TypeError, "list() takes no keyword arguments");
+ return NULL;
+ }
+ size_t nargs = PyVectorcall_NARGS(nargsf);
+ if (nargs > 1) {
+ PyErr_Format(PyExc_TypeError, "list() expected at most 1 argument, got %zu", nargs);
+ return NULL;
+ }
+
+ assert(PyType_Check(type));
+ PyObject *list = PyType_GenericAlloc((PyTypeObject *)type, 0);
+ if (list == NULL) {
+ return NULL;
+ }
+ PyObject *iterator = nargs ? args[0] : NULL;
+ if (list___init___impl((PyListObject *)list, iterator)) {
+ Py_DECREF(list);
+ return NULL;
+ }
+ return list;
+}
+
+
 /*[clinic input]
 list.__sizeof__
 
@@ -3041,6 +3069,7 @@ PyTypeObject PyList_Type = {
  PyType_GenericAlloc, /* tp_alloc */
  PyType_GenericNew, /* tp_new */
  PyObject_GC_Del, /* tp_free */
+ .tp_vectorcall = list_vectorcall,
 };
 
 /*********************** List Iterator **************************/

Objects/rangeobject.c

@@ -121,6 +121,66 @@ range_new(PyTypeObject *type, PyObject *args, PyObject *kw)
  return NULL;
 }
 
+
+static PyObject *
+range_vectorcall(PyTypeObject *type, PyObject *const *args,
+ size_t nargsf, PyObject *kwnames)
+{
+ rangeobject *obj;
+ size_t nargs = PyVectorcall_NARGS(nargsf);
+ PyObject *start = NULL, *stop = NULL, *step = NULL;
+ if (kwnames && PyTuple_GET_SIZE(kwnames) != 0) {
+ PyErr_Format(PyExc_TypeError, "range() takes no keyword arguments");
+ return NULL;
+ }
+ switch(nargs) {
+ case 0:
+ PyErr_Format(PyExc_TypeError, "range() expected 1 arguments, got 0");
+ return NULL;
+ case 1:
+ stop = PyNumber_Index(args[0]);
+ if (!stop)
+ return NULL;
+ Py_INCREF(_PyLong_Zero);
+ start = _PyLong_Zero;
+ Py_INCREF(_PyLong_One);
+ step = _PyLong_One;
+ break;
+ case 3:
+ step = args[2];
+ //Intentional fall through
+ case 2:
+ /* Convert borrowed refs to owned refs */
+ start = PyNumber_Index(args[0]);
+ if (!start)
+ return NULL;
+ stop = PyNumber_Index(args[1]);
+ if (!stop) {
+ Py_DECREF(start);
+ return NULL;
+ }
+ step = validate_step(step); /* Caution, this can clear exceptions */
+ if (!step) {
+ Py_DECREF(start);
+ Py_DECREF(stop);
+ return NULL;
+ }
+ break;
+ default:
+ PyErr_Format(PyExc_TypeError, "range() expected at most 3 arguments, got %zu", nargs);
+ return NULL;
+ }
+ obj = make_range_object(type, start, stop, step);
+ if (obj != NULL)
+ return (PyObject *) obj;
+
+ /* Failed to create object, release attributes */
+ Py_DECREF(start);
+ Py_DECREF(stop);
+ Py_DECREF(step);
+ return NULL;
+}
+
 PyDoc_STRVAR(range_doc,
 "range(stop) -> range object\n\
 range(start, stop[, step]) -> range object\n\
@@ -702,6 +762,7 @@ PyTypeObject PyRange_Type = {
  0, /* tp_init */
  0, /* tp_alloc */
  range_new, /* tp_new */
+ .tp_vectorcall = (vectorcallfunc)range_vectorcall
 };
 
 /*********************** range Iterator **************************/

Objects/typeobject.c

@@ -5170,6 +5170,11 @@ inherit_slots(PyTypeObject *type, PyTypeObject *base)
  !(type->tp_flags & Py_TPFLAGS_HEAPTYPE))
  {
  type->tp_flags |= _Py_TPFLAGS_HAVE_VECTORCALL;
+ /* Also inherit tp_vectorcall if tp_call of the
+ * metaclass is type.__call__ */
+ if (Py_TYPE(type)->tp_call == PyType_Type.tp_call) {
+ type->tp_vectorcall = base->tp_vectorcall;
+ }
  }
  COPYSLOT(tp_call);
  }