]> git.ipfire.org Git - thirdparty/Python/cpython.git/commitdiff
gh-134584: Eliminate redundant refcounting from `_CALL_BUILTION_O` (GH-142695)
authorPuQing <me@puqing.work>
Sun, 14 Dec 2025 10:28:02 +0000 (18:28 +0800)
committerGitHub <noreply@github.com>
Sun, 14 Dec 2025 10:28:02 +0000 (10:28 +0000)
Co-authored-by: Ken Jin <kenjin4096@gmail.com>
Include/internal/pycore_opcode_metadata.h
Include/internal/pycore_uop_ids.h
Include/internal/pycore_uop_metadata.h
Lib/test/test_capi/test_opt.py
Misc/NEWS.d/next/Core_and_Builtins/2025-06-28-04-32-38.gh-issue-134584.eZogqn.rst [new file with mode: 0644]
Python/bytecodes.c
Python/executor_cases.c.h
Python/generated_cases.c.h
Python/optimizer_cases.c.h

index ec2eb15151c98206da287cb14ac69ca6e29eb471..0fb01982b6aab75ad3a006c272127753394a9f1d 100644 (file)
@@ -1111,7 +1111,7 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[267] = {
     [CALL_BUILTIN_CLASS] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
     [CALL_BUILTIN_FAST] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
     [CALL_BUILTIN_FAST_WITH_KEYWORDS] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
-    [CALL_BUILTIN_O] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
+    [CALL_BUILTIN_O] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG },
     [CALL_FUNCTION_EX] = { true, INSTR_FMT_IX, HAS_EVAL_BREAK_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG | HAS_SYNC_SP_FLAG | HAS_NEEDS_GUARD_IP_FLAG },
     [CALL_INTRINSIC_1] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
     [CALL_INTRINSIC_2] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
@@ -1359,7 +1359,7 @@ _PyOpcode_macro_expansion[256] = {
     [CALL_BUILTIN_CLASS] = { .nuops = 2, .uops = { { _CALL_BUILTIN_CLASS, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
     [CALL_BUILTIN_FAST] = { .nuops = 2, .uops = { { _CALL_BUILTIN_FAST, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
     [CALL_BUILTIN_FAST_WITH_KEYWORDS] = { .nuops = 2, .uops = { { _CALL_BUILTIN_FAST_WITH_KEYWORDS, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
-    [CALL_BUILTIN_O] = { .nuops = 2, .uops = { { _CALL_BUILTIN_O, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
+    [CALL_BUILTIN_O] = { .nuops = 4, .uops = { { _CALL_BUILTIN_O, OPARG_SIMPLE, 3 }, { _POP_TOP, OPARG_SIMPLE, 3 }, { _POP_TOP, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
     [CALL_INTRINSIC_1] = { .nuops = 1, .uops = { { _CALL_INTRINSIC_1, OPARG_SIMPLE, 0 } } },
     [CALL_INTRINSIC_2] = { .nuops = 1, .uops = { { _CALL_INTRINSIC_2, OPARG_SIMPLE, 0 } } },
     [CALL_ISINSTANCE] = { .nuops = 3, .uops = { { _GUARD_THIRD_NULL, OPARG_SIMPLE, 3 }, { _GUARD_CALLABLE_ISINSTANCE, OPARG_SIMPLE, 3 }, { _CALL_ISINSTANCE, OPARG_SIMPLE, 3 } } },
index eb8e915e6d9a9fddab6c4217dc41345c807318be..e871e7699bf7fb253b89e0556d82a986aa679d44 100644 (file)
@@ -426,7 +426,7 @@ extern "C" {
 #define _CALL_BUILTIN_CLASS_r01 619
 #define _CALL_BUILTIN_FAST_r01 620
 #define _CALL_BUILTIN_FAST_WITH_KEYWORDS_r01 621
-#define _CALL_BUILTIN_O_r01 622
+#define _CALL_BUILTIN_O_r03 622
 #define _CALL_INTRINSIC_1_r11 623
 #define _CALL_INTRINSIC_2_r21 624
 #define _CALL_ISINSTANCE_r31 625
index 3bd19c82bc4eb4cf23be33aa88cee49edee683e4..5ae05358592a0fb759ecae77dc9810192aa93823 100644 (file)
@@ -281,7 +281,7 @@ const uint32_t _PyUop_Flags[MAX_UOP_ID+1] = {
     [_CREATE_INIT_FRAME] = HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG | HAS_SYNC_SP_FLAG,
     [_EXIT_INIT_CHECK] = HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG,
     [_CALL_BUILTIN_CLASS] = HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
-    [_CALL_BUILTIN_O] = HAS_ARG_FLAG | HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
+    [_CALL_BUILTIN_O] = HAS_ARG_FLAG | HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG,
     [_CALL_BUILTIN_FAST] = HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
     [_CALL_BUILTIN_FAST_WITH_KEYWORDS] = HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
     [_GUARD_CALLABLE_LEN] = HAS_DEOPT_FLAG,
@@ -2585,7 +2585,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
     [_CALL_BUILTIN_O] = {
         .best = { 0, 0, 0, 0 },
         .entries = {
-            { 1, 0, _CALL_BUILTIN_O_r01 },
+            { 3, 0, _CALL_BUILTIN_O_r03 },
             { -1, -1, -1 },
             { -1, -1, -1 },
             { -1, -1, -1 },
@@ -3744,7 +3744,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
     [_CREATE_INIT_FRAME_r01] = _CREATE_INIT_FRAME,
     [_EXIT_INIT_CHECK_r10] = _EXIT_INIT_CHECK,
     [_CALL_BUILTIN_CLASS_r01] = _CALL_BUILTIN_CLASS,
-    [_CALL_BUILTIN_O_r01] = _CALL_BUILTIN_O,
+    [_CALL_BUILTIN_O_r03] = _CALL_BUILTIN_O,
     [_CALL_BUILTIN_FAST_r01] = _CALL_BUILTIN_FAST,
     [_CALL_BUILTIN_FAST_WITH_KEYWORDS_r01] = _CALL_BUILTIN_FAST_WITH_KEYWORDS,
     [_GUARD_CALLABLE_LEN_r03] = _GUARD_CALLABLE_LEN,
@@ -4032,7 +4032,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
     [_CALL_BUILTIN_FAST_WITH_KEYWORDS] = "_CALL_BUILTIN_FAST_WITH_KEYWORDS",
     [_CALL_BUILTIN_FAST_WITH_KEYWORDS_r01] = "_CALL_BUILTIN_FAST_WITH_KEYWORDS_r01",
     [_CALL_BUILTIN_O] = "_CALL_BUILTIN_O",
-    [_CALL_BUILTIN_O_r01] = "_CALL_BUILTIN_O_r01",
+    [_CALL_BUILTIN_O_r03] = "_CALL_BUILTIN_O_r03",
     [_CALL_INTRINSIC_1] = "_CALL_INTRINSIC_1",
     [_CALL_INTRINSIC_1_r11] = "_CALL_INTRINSIC_1_r11",
     [_CALL_INTRINSIC_2] = "_CALL_INTRINSIC_2",
index bd48db537a2f4a649e5df37d122223908bc70dbb..32c00b57e1a460a1b08b654b37accc0eafea9cee 100644 (file)
@@ -2133,6 +2133,21 @@ class TestUopsOptimization(unittest.TestCase):
         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
diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2025-06-28-04-32-38.gh-issue-134584.eZogqn.rst b/Misc/NEWS.d/next/Core_and_Builtins/2025-06-28-04-32-38.gh-issue-134584.eZogqn.rst
new file mode 100644 (file)
index 0000000..97e24e5
--- /dev/null
@@ -0,0 +1 @@
+Eliminate redundant refcounting from ``_CALL_BUILTIN_O``.
index 3e76dcd77561cedaf1dff9f94e3071c5c3e08f0c..7e78e726b911e5016e971957931a32f47fcc9f02 100644 (file)
@@ -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);
 
@@ -4206,12 +4206,12 @@ 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);
         }
 
@@ -4219,6 +4219,8 @@ dummy_func(
             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)) {
index ff73b55bc6aaba316c6eab533fb5cded44ade7c5..727d3695bb11f0a4fd163ba26074b0506165a2dc 100644 (file)
             break;
         }
 
-        case _CALL_BUILTIN_O_r01: {
+        case _CALL_BUILTIN_O_r03: {
             CHECK_CURRENT_CACHED_VALUES(0);
             assert(WITHIN_STACK_BOUNDS_WITH_CACHE());
             _PyStackRef *args;
             _PyStackRef self_or_null;
             _PyStackRef callable;
             _PyStackRef res;
+            _PyStackRef a;
+            _PyStackRef c;
             oparg = CURRENT_OPARG();
             args = &stack_pointer[-oparg];
             self_or_null = stack_pointer[-1 - oparg];
             stack_pointer = _PyFrame_GetStackPointer(frame);
             _Py_LeaveRecursiveCallTstate(tstate);
             assert((res_o != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
-            _PyFrame_SetStackPointer(frame, stack_pointer);
-            PyStackRef_CLOSE(arg);
-            stack_pointer = _PyFrame_GetStackPointer(frame);
-            stack_pointer += -2 - oparg;
-            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
-            _PyFrame_SetStackPointer(frame, stack_pointer);
-            PyStackRef_CLOSE(callable);
-            stack_pointer = _PyFrame_GetStackPointer(frame);
             if (res_o == NULL) {
                 SET_CURRENT_CACHED_VALUES(0);
                 JUMP_TO_ERROR();
             }
+            a = arg;
+            c = callable;
             res = PyStackRef_FromPyObjectSteal(res_o);
+            _tos_cache2 = c;
+            _tos_cache1 = a;
             _tos_cache0 = res;
-            _tos_cache1 = PyStackRef_ZERO_BITS;
-            _tos_cache2 = PyStackRef_ZERO_BITS;
-            SET_CURRENT_CACHED_VALUES(1);
+            SET_CURRENT_CACHED_VALUES(3);
+            stack_pointer += -2 - oparg;
+            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             assert(WITHIN_STACK_BOUNDS_WITH_CACHE());
             break;
         }
index 1d8c01bd1b350133818f9239333d1bd5af26d535..e1860f5d9d8f6c2ea993a258ff1788a43cef9981 100644 (file)
             _PyStackRef self_or_null;
             _PyStackRef *args;
             _PyStackRef res;
+            _PyStackRef a;
+            _PyStackRef c;
+            _PyStackRef value;
             /* Skip 1 cache entry */
             /* Skip 2 cache entries */
             // _CALL_BUILTIN_O
                 stack_pointer = _PyFrame_GetStackPointer(frame);
                 _Py_LeaveRecursiveCallTstate(tstate);
                 assert((res_o != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
+                if (res_o == NULL) {
+                    JUMP_TO_LABEL(error);
+                }
+                a = arg;
+                c = callable;
+                res = PyStackRef_FromPyObjectSteal(res_o);
+            }
+            // _POP_TOP
+            {
+                value = c;
+                stack_pointer[-2 - oparg] = res;
+                stack_pointer[-1 - oparg] = a;
+                stack_pointer += -oparg;
+                ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
                 _PyFrame_SetStackPointer(frame, stack_pointer);
-                PyStackRef_CLOSE(arg);
+                PyStackRef_XCLOSE(value);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
-                stack_pointer += -2 - oparg;
+            }
+            // _POP_TOP
+            {
+                value = a;
+                stack_pointer += -1;
                 ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
                 _PyFrame_SetStackPointer(frame, stack_pointer);
-                PyStackRef_CLOSE(callable);
+                PyStackRef_XCLOSE(value);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
-                if (res_o == NULL) {
-                    JUMP_TO_LABEL(error);
-                }
-                res = PyStackRef_FromPyObjectSteal(res_o);
             }
             // _CHECK_PERIODIC_AT_END
             {
-                stack_pointer[0] = res;
-                stack_pointer += 1;
-                ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
                 _PyFrame_SetStackPointer(frame, stack_pointer);
                 int err = check_periodics(tstate);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
index d3cff14be70aac4433e6d9b812c99885adc8a953..5c5efdaa1fd880d68b502fda0518806d625b8582 100644 (file)
 
         case _CALL_BUILTIN_O: {
             JitOptRef res;
+            JitOptRef a;
+            JitOptRef c;
             res = sym_new_not_null(ctx);
-            CHECK_STACK_BOUNDS(-1 - oparg);
+            a = sym_new_not_null(ctx);
+            c = sym_new_not_null(ctx);
+            CHECK_STACK_BOUNDS(1 - oparg);
             stack_pointer[-2 - oparg] = res;
-            stack_pointer += -1 - oparg;
+            stack_pointer[-1 - oparg] = a;
+            stack_pointer[-oparg] = c;
+            stack_pointer += 1 - oparg;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             break;
         }