gh-134584: Eliminate redundant refcounting from _CONTAINS_{OP|OP_SET|OP_DICT} (GH-143731)

Signed-off-by: Manjusaka <[email protected]>

Author: Zheaoli

Include/internal/pycore_opcode_metadata.h

@@ -1992,6 +1992,57 @@ def testfunc(n):
         self.assertIn("_BINARY_OP_SUBSCR_DICT", uops)
         self.assertLessEqual(count_ops(ex, "_POP_TOP"), 2)
 
+    def test_contains_op(self):
+        def testfunc(n):
+            x = 0
+            items = [1, 2, 3]
+            for _ in range(n):
+                if 2 in items:
+                    x += 1
+            return x
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, TIER2_THRESHOLD)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertIn("_CONTAINS_OP", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+        self.assertLessEqual(count_ops(ex, "_POP_TOP"), 2)
+
+    def test_contains_op_set(self):
+        def testfunc(n):
+            x = 0
+            s = {1, 2, 3}
+            for _ in range(n):
+                if 2 in s:
+                    x += 1
+            return x
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, TIER2_THRESHOLD)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertIn("_CONTAINS_OP_SET", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+        self.assertLessEqual(count_ops(ex, "_POP_TOP"), 2)
+
+    def test_contains_op_dict(self):
+        def testfunc(n):
+            x = 0
+            d = {'a': 1, 'b': 2}
+            for _ in range(n):
+                if 'a' in d:
+                    x += 1
+            return x
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, TIER2_THRESHOLD)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertIn("_CONTAINS_OP_DICT", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+        self.assertLessEqual(count_ops(ex, "_POP_TOP"), 2)
+
     def test_call_type_1_guards_removed(self):
         def testfunc(n):
             x = 0

Include/internal/pycore_uop_ids.h

@@ -0,0 +1,2 @@
+Eliminate redundant refcounting from ``_CONTAINS_OP``, ``_CONTAINS_OP_SET``
+and ``_CONTAINS_OP_DICT``.

Include/internal/pycore_uop_metadata.h

@@ -2855,14 +2855,18 @@ dummy_func(
             CONTAINS_OP_DICT,
         };
 
-        op(_CONTAINS_OP, (left, right -- b)) {
+        op(_CONTAINS_OP, (left, right -- b, l, r)) {
             PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
             PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
 
             int res = PySequence_Contains(right_o, left_o);
-            DECREF_INPUTS();
-            ERROR_IF(res < 0);
+            if (res < 0) {
+                ERROR_NO_POP();
+            }
             b = (res ^ oparg) ? PyStackRef_True : PyStackRef_False;
+            l = left;
+            r = right;
+            INPUTS_DEAD();
         }
 
         specializing op(_SPECIALIZE_CONTAINS_OP, (counter/1, left, right -- left, right)) {
@@ -2877,40 +2881,48 @@ dummy_func(
             #endif  /* ENABLE_SPECIALIZATION_FT */
         }
 
-        macro(CONTAINS_OP) = _SPECIALIZE_CONTAINS_OP + _CONTAINS_OP;
+        macro(CONTAINS_OP) = _SPECIALIZE_CONTAINS_OP + _CONTAINS_OP + POP_TOP + POP_TOP;
 
         op(_GUARD_TOS_ANY_SET, (tos -- tos)) {
             PyObject *o = PyStackRef_AsPyObjectBorrow(tos);
             DEOPT_IF(!PyAnySet_CheckExact(o));
         }
 
-        macro(CONTAINS_OP_SET) = _GUARD_TOS_ANY_SET + unused/1 + _CONTAINS_OP_SET;
+        macro(CONTAINS_OP_SET) = _GUARD_TOS_ANY_SET + unused/1 + _CONTAINS_OP_SET + POP_TOP + POP_TOP;
 
-        op(_CONTAINS_OP_SET, (left, right -- b)) {
+        op(_CONTAINS_OP_SET, (left, right -- b, l, r)) {
             PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
             PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
 
             assert(PyAnySet_CheckExact(right_o));
             STAT_INC(CONTAINS_OP, hit);
             // Note: both set and frozenset use the same seq_contains method!
             int res = _PySet_Contains((PySetObject *)right_o, left_o);
-            DECREF_INPUTS();
-            ERROR_IF(res < 0);
+            if (res < 0) {
+                ERROR_NO_POP();
+            }
             b = (res ^ oparg) ? PyStackRef_True : PyStackRef_False;
+            l = left;
+            r = right;
+            INPUTS_DEAD();
         }
 
-        macro(CONTAINS_OP_DICT) = _GUARD_TOS_DICT + unused/1 + _CONTAINS_OP_DICT;
+        macro(CONTAINS_OP_DICT) = _GUARD_TOS_DICT + unused/1 + _CONTAINS_OP_DICT + POP_TOP + POP_TOP;
 
-        op(_CONTAINS_OP_DICT, (left, right -- b)) {
+        op(_CONTAINS_OP_DICT, (left, right -- b, l, r)) {
             PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
             PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
 
             assert(PyDict_CheckExact(right_o));
             STAT_INC(CONTAINS_OP, hit);
             int res = PyDict_Contains(right_o, left_o);
-            DECREF_INPUTS();
-            ERROR_IF(res < 0);
+            if (res < 0) {
+                ERROR_NO_POP();
+            }
             b = (res ^ oparg) ? PyStackRef_True : PyStackRef_False;
+            l = left;
+            r = right;
+            INPUTS_DEAD();
         }
 
         inst(CHECK_EG_MATCH, (exc_value_st, match_type_st -- rest, match)) {