gh-134584: Eliminate redundant refcounting from `_CALL_TUPLE_1` (GH-135860)

diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h
index 0dfd2a251bd6d5eb3134d7059d63a28ad87f480b..3ec0a86341483abaa8468c369e9f419aac22cbfe 100644 (file)
--- a/Include/internal/pycore_opcode_metadata.h
+++ b/Include/internal/pycore_opcode_metadata.h
@@ -1376,7 +1376,7 @@ _PyOpcode_macro_expansion[256] = {
     [CALL_PY_EXACT_ARGS] = { .nuops = 8, .uops = { { _CHECK_PEP_523, OPARG_SIMPLE, 1 }, { _CHECK_FUNCTION_VERSION, 2, 1 }, { _CHECK_FUNCTION_EXACT_ARGS, OPARG_SIMPLE, 3 }, { _CHECK_STACK_SPACE, OPARG_SIMPLE, 3 }, { _CHECK_RECURSION_REMAINING, OPARG_SIMPLE, 3 }, { _INIT_CALL_PY_EXACT_ARGS, OPARG_SIMPLE, 3 }, { _SAVE_RETURN_OFFSET, OPARG_SAVE_RETURN_OFFSET, 3 }, { _PUSH_FRAME, OPARG_SIMPLE, 3 } } },
     [CALL_PY_GENERAL] = { .nuops = 6, .uops = { { _CHECK_PEP_523, OPARG_SIMPLE, 1 }, { _CHECK_FUNCTION_VERSION, 2, 1 }, { _CHECK_RECURSION_REMAINING, OPARG_SIMPLE, 3 }, { _PY_FRAME_GENERAL, OPARG_SIMPLE, 3 }, { _SAVE_RETURN_OFFSET, OPARG_SAVE_RETURN_OFFSET, 3 }, { _PUSH_FRAME, OPARG_SIMPLE, 3 } } },
     [CALL_STR_1] = { .nuops = 4, .uops = { { _GUARD_NOS_NULL, OPARG_SIMPLE, 3 }, { _GUARD_CALLABLE_STR_1, OPARG_SIMPLE, 3 }, { _CALL_STR_1, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
-    [CALL_TUPLE_1] = { .nuops = 4, .uops = { { _GUARD_NOS_NULL, OPARG_SIMPLE, 3 }, { _GUARD_CALLABLE_TUPLE_1, OPARG_SIMPLE, 3 }, { _CALL_TUPLE_1, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
+    [CALL_TUPLE_1] = { .nuops = 5, .uops = { { _GUARD_NOS_NULL, OPARG_SIMPLE, 3 }, { _GUARD_CALLABLE_TUPLE_1, OPARG_SIMPLE, 3 }, { _CALL_TUPLE_1, OPARG_SIMPLE, 3 }, { _POP_TOP, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
     [CALL_TYPE_1] = { .nuops = 3, .uops = { { _GUARD_NOS_NULL, OPARG_SIMPLE, 3 }, { _GUARD_CALLABLE_TYPE_1, OPARG_SIMPLE, 3 }, { _CALL_TYPE_1, OPARG_SIMPLE, 3 } } },
     [CHECK_EG_MATCH] = { .nuops = 1, .uops = { { _CHECK_EG_MATCH, OPARG_SIMPLE, 0 } } },
     [CHECK_EXC_MATCH] = { .nuops = 1, .uops = { { _CHECK_EXC_MATCH, OPARG_SIMPLE, 0 } } },

diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h
index 553636cbb5914ae991e77ad92255dab1b3fdb64e..08e517dea326e8c829e4bc6b8af226def0e7f250 100644 (file)
--- a/Include/internal/pycore_uop_ids.h
+++ b/Include/internal/pycore_uop_ids.h
@@ -439,7 +439,7 @@ extern "C" {
 #define _CALL_METHOD_DESCRIPTOR_O_r01 632
 #define _CALL_NON_PY_GENERAL_r01 633
 #define _CALL_STR_1_r31 634
-#define _CALL_TUPLE_1_r31 635
+#define _CALL_TUPLE_1_r32 635
 #define _CALL_TYPE_1_r31 636
 #define _CHECK_AND_ALLOCATE_OBJECT_r00 637
 #define _CHECK_ATTR_CLASS_r01 638

diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h
index 436defbb8df57082f8ce0b059eb66ddfe5bbff48..9ca53329e5d9599c123cd0c940d7c4e0282a4843 100644 (file)
--- a/Include/internal/pycore_uop_metadata.h
+++ b/Include/internal/pycore_uop_metadata.h
@@ -2543,7 +2543,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
             { -1, -1, -1 },
             { -1, -1, -1 },
             { -1, -1, -1 },
-            { 1, 0, _CALL_TUPLE_1_r31 },
+            { 2, 0, _CALL_TUPLE_1_r32 },
         },
     },
     [_CHECK_AND_ALLOCATE_OBJECT] = {
@@ -3739,7 +3739,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
     [_GUARD_CALLABLE_TUPLE_1_r13] = _GUARD_CALLABLE_TUPLE_1,
     [_GUARD_CALLABLE_TUPLE_1_r23] = _GUARD_CALLABLE_TUPLE_1,
     [_GUARD_CALLABLE_TUPLE_1_r33] = _GUARD_CALLABLE_TUPLE_1,
-    [_CALL_TUPLE_1_r31] = _CALL_TUPLE_1,
+    [_CALL_TUPLE_1_r32] = _CALL_TUPLE_1,
     [_CHECK_AND_ALLOCATE_OBJECT_r00] = _CHECK_AND_ALLOCATE_OBJECT,
     [_CREATE_INIT_FRAME_r01] = _CREATE_INIT_FRAME,
     [_EXIT_INIT_CHECK_r10] = _EXIT_INIT_CHECK,
@@ -4058,7 +4058,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
     [_CALL_STR_1] = "_CALL_STR_1",
     [_CALL_STR_1_r31] = "_CALL_STR_1_r31",
     [_CALL_TUPLE_1] = "_CALL_TUPLE_1",
-    [_CALL_TUPLE_1_r31] = "_CALL_TUPLE_1_r31",
+    [_CALL_TUPLE_1_r32] = "_CALL_TUPLE_1_r32",
     [_CALL_TYPE_1] = "_CALL_TYPE_1",
     [_CALL_TYPE_1_r31] = "_CALL_TYPE_1_r31",
     [_CHECK_AND_ALLOCATE_OBJECT] = "_CHECK_AND_ALLOCATE_OBJECT",

diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py
index 2e7485e89fdac5f2a8b19204ba24f53c9bcc72cf..f16055ab84de0890863e73283db834bcd9b3b5e1 100644 (file)
--- a/Lib/test/test_capi/test_opt.py
+++ b/Lib/test/test_capi/test_opt.py
@@ -1925,6 +1925,21 @@ class TestUopsOptimization(unittest.TestCase):
         uops = get_opnames(ex)
         self.assertNotIn("_GUARD_IS_NOT_NONE_POP", uops)
 
+    def test_call_tuple_1_pop_top(self):
+        def testfunc(n):
+            x = 0
+            for _ in range(n):
+                t = tuple(())
+                x += len(t) == 0
+            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_TUPLE_1", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+
     def test_call_str_1(self):
         def testfunc(n):
             x = 0

diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2025-06-23-22-52-20.gh-issue-134584.qbiQfG.rst b/Misc/NEWS.d/next/Core_and_Builtins/2025-06-23-22-52-20.gh-issue-134584.qbiQfG.rst
new file mode 100644 (file)
index 0000000..34df147
--- /dev/null
+++ b/Misc/NEWS.d/next/Core_and_Builtins/2025-06-23-22-52-20.gh-issue-134584.qbiQfG.rst
@@ -0,0 +1 @@
+Eliminate redundant refcounting from ``_CALL_TUPLE_1``. Patch by Noam Cohen

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index f64a7778121c71010244409f7d2559f3c9ecc4d0..3cc6ff842c9202bf4e729c6ed3eb897f90f1c1b6 100644 (file)
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -4068,17 +4068,14 @@ dummy_func(
             DEOPT_IF(callable_o != (PyObject *)&PyTuple_Type);
         }
 
-        op(_CALL_TUPLE_1, (callable, null, arg -- res)) {
+        op(_CALL_TUPLE_1, (callable, null, arg -- res, a)) {
             PyObject *arg_o = PyStackRef_AsPyObjectBorrow(arg);
 
             assert(oparg == 1);
             STAT_INC(CALL, hit);
             PyObject *res_o = PySequence_Tuple(arg_o);
-            DEAD(null);
-            DEAD(callable);
-            (void)callable; // Silence compiler warnings about unused variables
-            (void)null;
-            PyStackRef_CLOSE(arg);
+            a = arg;
+            INPUTS_DEAD();
             ERROR_IF(res_o == NULL);
             res = PyStackRef_FromPyObjectSteal(res_o);
         }
@@ -4089,6 +4086,7 @@ dummy_func(
             _GUARD_NOS_NULL +
             _GUARD_CALLABLE_TUPLE_1 +
             _CALL_TUPLE_1 +
+            POP_TOP +
             _CHECK_PERIODIC_AT_END;
 
         op(_CHECK_AND_ALLOCATE_OBJECT, (type_version/2, callable, self_or_null, unused[oparg] -- callable, self_or_null, unused[oparg])) {
@@ -5455,6 +5453,12 @@ dummy_func(
             }
         }
 
+        label(pop_3_error) {
+            stack_pointer -= 3;
+            assert(WITHIN_STACK_BOUNDS());
+            goto error;
+        }
+
         label(pop_2_error) {
             stack_pointer -= 2;
             assert(WITHIN_STACK_BOUNDS());

diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 24f343c994038bd46781a765c261a2171d36bb8c..248cf60b70e45527760d1ad7564a8db3cdeea984 100644 (file)
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -12888,47 +12888,42 @@
             break;
         }
 
-        case _CALL_TUPLE_1_r31: {
+        case _CALL_TUPLE_1_r32: {
             CHECK_CURRENT_CACHED_VALUES(3);
             assert(WITHIN_STACK_BOUNDS_WITH_CACHE());
             _PyStackRef arg;
-            _PyStackRef null;
-            _PyStackRef callable;
             _PyStackRef res;
+            _PyStackRef a;
             _PyStackRef _stack_item_0 = _tos_cache0;
             _PyStackRef _stack_item_1 = _tos_cache1;
             _PyStackRef _stack_item_2 = _tos_cache2;
             oparg = CURRENT_OPARG();
             arg = _stack_item_2;
-            null = _stack_item_1;
-            callable = _stack_item_0;
             PyObject *arg_o = PyStackRef_AsPyObjectBorrow(arg);
             assert(oparg == 1);
             STAT_INC(CALL, hit);
-            stack_pointer[0] = callable;
-            stack_pointer[1] = null;
+            stack_pointer[0] = _stack_item_0;
+            stack_pointer[1] = _stack_item_1;
             stack_pointer[2] = arg;
             stack_pointer += 3;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             _PyFrame_SetStackPointer(frame, stack_pointer);
             PyObject *res_o = PySequence_Tuple(arg_o);
             stack_pointer = _PyFrame_GetStackPointer(frame);
-            (void)callable;
-            (void)null;
-            stack_pointer += -3;
-            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
-            _PyFrame_SetStackPointer(frame, stack_pointer);
-            PyStackRef_CLOSE(arg);
-            stack_pointer = _PyFrame_GetStackPointer(frame);
+            a = arg;
             if (res_o == NULL) {
+                stack_pointer += -3;
+                ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
                 SET_CURRENT_CACHED_VALUES(0);
                 JUMP_TO_ERROR();
             }
             res = PyStackRef_FromPyObjectSteal(res_o);
+            _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(2);
+            stack_pointer += -3;
+            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             assert(WITHIN_STACK_BOUNDS_WITH_CACHE());
             break;
         }

diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index ba80886f6dc0c7c380fd94ad1460005b703d607c..b70b7fde37a72a4ee028ca0c8d552109f5e33107 100644 (file)
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -4029,6 +4029,8 @@
             _PyStackRef callable;
             _PyStackRef arg;
             _PyStackRef res;
+            _PyStackRef a;
+            _PyStackRef value;
             /* Skip 1 cache entry */
             /* Skip 2 cache entries */
             // _GUARD_NOS_NULL
@@ -4059,23 +4061,24 @@
                 _PyFrame_SetStackPointer(frame, stack_pointer);
                 PyObject *res_o = PySequence_Tuple(arg_o);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
-                (void)callable;
-                (void)null;
-                stack_pointer += -3;
-                ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
-                _PyFrame_SetStackPointer(frame, stack_pointer);
-                PyStackRef_CLOSE(arg);
-                stack_pointer = _PyFrame_GetStackPointer(frame);
+                a = arg;
                 if (res_o == NULL) {
-                    JUMP_TO_LABEL(error);
+                    JUMP_TO_LABEL(pop_3_error);
                 }
                 res = PyStackRef_FromPyObjectSteal(res_o);
             }
-            // _CHECK_PERIODIC_AT_END
+            // _POP_TOP
             {
-                stack_pointer[0] = res;
-                stack_pointer += 1;
+                value = a;
+                stack_pointer[-3] = res;
+                stack_pointer += -2;
                 ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
+                _PyFrame_SetStackPointer(frame, stack_pointer);
+                PyStackRef_XCLOSE(value);
+                stack_pointer = _PyFrame_GetStackPointer(frame);
+            }
+            // _CHECK_PERIODIC_AT_END
+            {
                 _PyFrame_SetStackPointer(frame, stack_pointer);
                 int err = check_periodics(tstate);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
@@ -11849,6 +11852,13 @@ JUMP_TO_LABEL(error);
 #endif /* _Py_TAIL_CALL_INTERP */
         /* BEGIN LABELS */
 
+        LABEL(pop_3_error)
+        {
+            stack_pointer -= 3;
+            assert(WITHIN_STACK_BOUNDS());
+            JUMP_TO_LABEL(error);
+        }
+
         LABEL(pop_2_error)
         {
             stack_pointer -= 2;

diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index b2fa7d01e8f6c2e86eb8411be0465cec34c27851..1a8a4392187ecad1e0a6d20fde73371f499cb1fb 100644 (file)
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -522,6 +522,7 @@ static py_tail_call_funcptr instruction_funcptr_handler_table[256];
 
 static py_tail_call_funcptr instruction_funcptr_tracing_table[256];
 
+Py_PRESERVE_NONE_CC static PyObject *_TAIL_CALL_pop_3_error(TAIL_CALL_PARAMS);
 Py_PRESERVE_NONE_CC static PyObject *_TAIL_CALL_pop_2_error(TAIL_CALL_PARAMS);
 Py_PRESERVE_NONE_CC static PyObject *_TAIL_CALL_pop_1_error(TAIL_CALL_PARAMS);
 Py_PRESERVE_NONE_CC static PyObject *_TAIL_CALL_error(TAIL_CALL_PARAMS);

diff --git a/Python/optimizer_bytecodes.c b/Python/optimizer_bytecodes.c
index 864ae17525e6b25b8e1e6566fa8968194e8b79cd..701f46ac66312074ca2cdcd105f2c1331c50a42e 100644 (file)
--- a/Python/optimizer_bytecodes.c
+++ b/Python/optimizer_bytecodes.c
@@ -1097,7 +1097,7 @@ dummy_func(void) {
         }
     }
 
-    op(_CALL_TUPLE_1, (callable, null, arg -- res)) {
+    op(_CALL_TUPLE_1, (callable, null, arg -- res, a)) {
         if (sym_matches_type(arg, &PyTuple_Type)) {
             // e.g. tuple((1, 2)) or tuple(foo) where foo is known to be a tuple
             // Note: we must strip the reference information because it goes
@@ -1107,6 +1107,7 @@ dummy_func(void) {
         else {
             res = sym_new_type(ctx, &PyTuple_Type);
         }
+        a = arg;
     }
 
     op(_GUARD_TOS_LIST, (tos -- tos)) {

diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index 20192ebc5930549a57dc5d8df12016f92e0d664d..f19ecb898a24cb76b3ba29db3549fe41d3dfa659 100644 (file)
--- a/Python/optimizer_cases.c.h
+++ b/Python/optimizer_cases.c.h
@@ -2882,6 +2882,7 @@
         case _CALL_TUPLE_1: {
             JitOptRef arg;
             JitOptRef res;
+            JitOptRef a;
             arg = stack_pointer[-1];
             if (sym_matches_type(arg, &PyTuple_Type)) {
                 res = PyJitRef_StripReferenceInfo(arg);
@@ -2889,9 +2890,11 @@
             else {
                 res = sym_new_type(ctx, &PyTuple_Type);
             }
-            CHECK_STACK_BOUNDS(-2);
+            a = arg;
+            CHECK_STACK_BOUNDS(-1);
             stack_pointer[-3] = res;
-            stack_pointer += -2;
+            stack_pointer[-2] = a;
+            stack_pointer += -1;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             break;
         }