gh-134584: Eliminate redundant refcounting from _CALL_BUILTION_O (GH-142695)

Co-authored-by: Ken Jin <kenjin4096@gmail.com>

Author: AndPuQing

Include/internal/pycore_opcode_metadata.h

@@ -2133,6 +2133,21 @@ class C:
         self.assertNotIn("_COMPARE_OP_INT", uops)
         self.assertNotIn("_GUARD_IS_TRUE_POP", uops)
 
+    def test_call_builtin_o(self):
+        def testfunc(n):
+            x = 0
+            for _ in range(n):
+                y = abs(1)
+                x += y
+            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("_CALL_BUILTIN_O", uops)
+        self.assertIn("_POP_TOP", uops)
+
     def test_get_len_with_const_tuple(self):
         def testfunc(n):
             x = 0.0

Include/internal/pycore_uop_ids.h

@@ -0,0 +1 @@
+Eliminate redundant refcounting from ``_CALL_BUILTIN_O``.

Include/internal/pycore_uop_metadata.h

@@ -4186,7 +4186,7 @@ dummy_func(
             _CALL_BUILTIN_CLASS +
             _CHECK_PERIODIC_AT_END;
 
-        op(_CALL_BUILTIN_O, (callable, self_or_null, args[oparg] -- res)) {
+        op(_CALL_BUILTIN_O, (callable, self_or_null, args[oparg] -- res, a, c)) {
             /* Builtin METH_O functions */
             PyObject *callable_o = PyStackRef_AsPyObjectBorrow(callable);
 
@@ -4206,19 +4206,21 @@ dummy_func(
             PyObject *res_o = _PyCFunction_TrampolineCall(cfunc, PyCFunction_GET_SELF(callable_o), PyStackRef_AsPyObjectBorrow(arg));
             _Py_LeaveRecursiveCallTstate(tstate);
             assert((res_o != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
-
-            PyStackRef_CLOSE(arg);
-            DEAD(args);
-            DEAD(self_or_null);
-            PyStackRef_CLOSE(callable);
-            ERROR_IF(res_o == NULL);
+            if (res_o == NULL) {
+                ERROR_NO_POP();
+            }
+            a = arg;
+            c = callable;
+            INPUTS_DEAD();
             res = PyStackRef_FromPyObjectSteal(res_o);
         }
 
         macro(CALL_BUILTIN_O) =
             unused/1 +
             unused/2 +
             _CALL_BUILTIN_O +
+            POP_TOP +
+            POP_TOP +
             _CHECK_PERIODIC_AT_END;
 
         op(_CALL_BUILTIN_FAST, (callable, self_or_null, args[oparg] -- res)) {