GH-100110: Specialize FOR_ITER for tuples (GH-100109)

* Specialize FOR_ITER for tuples

Include/internal/pycore_tuple.h

@@ -67,6 +67,13 @@ struct _Py_tuple_state {
 extern PyObject *_PyTuple_FromArray(PyObject *const *, Py_ssize_t);
 extern PyObject *_PyTuple_FromArraySteal(PyObject *const *, Py_ssize_t);
 
+
+typedef struct {
+    PyObject_HEAD
+    Py_ssize_t it_index;
+    PyTupleObject *it_seq; /* Set to NULL when iterator is exhausted */
+} _PyTupleIterObject;
+
 #ifdef __cplusplus
 }
 #endif

Lib/opcode.py

@@ -320,6 +320,7 @@ def pseudo_op(name, op, real_ops):
     ],
     "FOR_ITER": [
         "FOR_ITER_LIST",
+        "FOR_ITER_TUPLE",
         "FOR_ITER_RANGE",
         "FOR_ITER_GEN",
     ],

Misc/NEWS.d/next/Core and Builtins/2022-12-08-12-26-34.gh-issue-100110.ertac-.rst

@@ -0,0 +1 @@
+Specialize ``FOR_ITER`` for tuples.

Objects/tupleobject.c

@@ -995,29 +995,24 @@ _PyTuple_ClearFreeList(PyInterpreterState *interp)
 
 /*********************** Tuple Iterator **************************/
 
-typedef struct {
-    PyObject_HEAD
-    Py_ssize_t it_index;
-    PyTupleObject *it_seq; /* Set to NULL when iterator is exhausted */
-} tupleiterobject;
 
 static void
-tupleiter_dealloc(tupleiterobject *it)
+tupleiter_dealloc(_PyTupleIterObject *it)
 {
     _PyObject_GC_UNTRACK(it);
     Py_XDECREF(it->it_seq);
     PyObject_GC_Del(it);
 }
 
 static int
-tupleiter_traverse(tupleiterobject *it, visitproc visit, void *arg)
+tupleiter_traverse(_PyTupleIterObject *it, visitproc visit, void *arg)
 {
     Py_VISIT(it->it_seq);
     return 0;
 }
 
 static PyObject *
-tupleiter_next(tupleiterobject *it)
+tupleiter_next(_PyTupleIterObject *it)
 {
     PyTupleObject *seq;
     PyObject *item;
@@ -1040,7 +1035,7 @@ tupleiter_next(tupleiterobject *it)
 }
 
 static PyObject *
-tupleiter_len(tupleiterobject *it, PyObject *Py_UNUSED(ignored))
+tupleiter_len(_PyTupleIterObject *it, PyObject *Py_UNUSED(ignored))
 {
     Py_ssize_t len = 0;
     if (it->it_seq)
@@ -1051,7 +1046,7 @@ tupleiter_len(tupleiterobject *it, PyObject *Py_UNUSED(ignored))
 PyDoc_STRVAR(length_hint_doc, "Private method returning an estimate of len(list(it)).");
 
 static PyObject *
-tupleiter_reduce(tupleiterobject *it, PyObject *Py_UNUSED(ignored))
+tupleiter_reduce(_PyTupleIterObject *it, PyObject *Py_UNUSED(ignored))
 {
     if (it->it_seq)
         return Py_BuildValue("N(O)n", _PyEval_GetBuiltin(&_Py_ID(iter)),
@@ -1061,7 +1056,7 @@ tupleiter_reduce(tupleiterobject *it, PyObject *Py_UNUSED(ignored))
 }
 
 static PyObject *
-tupleiter_setstate(tupleiterobject *it, PyObject *state)
+tupleiter_setstate(_PyTupleIterObject *it, PyObject *state)
 {
     Py_ssize_t index = PyLong_AsSsize_t(state);
     if (index == -1 && PyErr_Occurred())
@@ -1089,7 +1084,7 @@ static PyMethodDef tupleiter_methods[] = {
 PyTypeObject PyTupleIter_Type = {
     PyVarObject_HEAD_INIT(&PyType_Type, 0)
     "tuple_iterator",                           /* tp_name */
-    sizeof(tupleiterobject),                    /* tp_basicsize */
+    sizeof(_PyTupleIterObject),                    /* tp_basicsize */
     0,                                          /* tp_itemsize */
     /* methods */
     (destructor)tupleiter_dealloc,              /* tp_dealloc */
@@ -1122,13 +1117,13 @@ PyTypeObject PyTupleIter_Type = {
 static PyObject *
 tuple_iter(PyObject *seq)
 {
-    tupleiterobject *it;
+    _PyTupleIterObject *it;
 
     if (!PyTuple_Check(seq)) {
         PyErr_BadInternalCall();
         return NULL;
     }
-    it = PyObject_GC_New(tupleiterobject, &PyTupleIter_Type);
+    it = PyObject_GC_New(_PyTupleIterObject, &PyTupleIter_Type);
     if (it == NULL)
         return NULL;
     it->it_index = 0;

Python/bytecodes.c

@@ -2523,6 +2523,29 @@ dummy_func(
         end_for_iter_list:
         }
 
+        // stack effect: ( -- __0)
+        inst(FOR_ITER_TUPLE) {
+            assert(cframe.use_tracing == 0);
+            _PyTupleIterObject *it = (_PyTupleIterObject *)TOP();
+            DEOPT_IF(Py_TYPE(it) != &PyTupleIter_Type, FOR_ITER);
+            STAT_INC(FOR_ITER, hit);
+            PyTupleObject *seq = it->it_seq;
+            if (seq) {
+                if (it->it_index < PyTuple_GET_SIZE(seq)) {
+                    PyObject *next = PyTuple_GET_ITEM(seq, it->it_index++);
+                    PUSH(Py_NewRef(next));
+                    JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER);
+                    goto end_for_iter_tuple;  // End of this instruction
+                }
+                it->it_seq = NULL;
+                Py_DECREF(seq);
+            }
+            STACK_SHRINK(1);
+            Py_DECREF(it);
+            JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER + oparg + 1);
+        end_for_iter_tuple:
+        }
+
         // stack effect: ( -- __0)
         inst(FOR_ITER_RANGE) {
             assert(cframe.use_tracing == 0);