gh-134584: Eliminate redundant refcounting from _BINARY_OP_SUBSCR_LIST_INT (GH-142926)

diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h
index 906ea0db0a5fe0a2e8e2d2a0e320d3c9b1e27e30..913359ee4bdc4fdbc8ac8c21e898b49fc25b95f9 100644 (file)
--- a/Include/internal/pycore_opcode_metadata.h
+++ b/Include/internal/pycore_opcode_metadata.h
@@ -1338,7 +1338,7 @@ _PyOpcode_macro_expansion[256] = {
     [BINARY_OP_MULTIPLY_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_INT, OPARG_SIMPLE, 0 }, { _BINARY_OP_MULTIPLY_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 } } },
     [BINARY_OP_SUBSCR_DICT] = { .nuops = 2, .uops = { { _GUARD_NOS_DICT, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_DICT, OPARG_SIMPLE, 5 } } },
     [BINARY_OP_SUBSCR_GETITEM] = { .nuops = 4, .uops = { { _CHECK_PEP_523, OPARG_SIMPLE, 5 }, { _BINARY_OP_SUBSCR_CHECK_FUNC, OPARG_SIMPLE, 5 }, { _BINARY_OP_SUBSCR_INIT_CALL, OPARG_SIMPLE, 5 }, { _PUSH_FRAME, OPARG_SIMPLE, 5 } } },
-    [BINARY_OP_SUBSCR_LIST_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_LIST_INT, OPARG_SIMPLE, 5 } } },
+    [BINARY_OP_SUBSCR_LIST_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_LIST_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP, OPARG_SIMPLE, 5 } } },
     [BINARY_OP_SUBSCR_LIST_SLICE] = { .nuops = 3, .uops = { { _GUARD_TOS_SLICE, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_LIST_SLICE, OPARG_SIMPLE, 5 } } },
     [BINARY_OP_SUBSCR_STR_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_UNICODE, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_STR_INT, OPARG_SIMPLE, 5 } } },
     [BINARY_OP_SUBSCR_TUPLE_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_TUPLE, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_TUPLE_INT, OPARG_SIMPLE, 5 } } },

diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h
index f73aad9de1c0bf44fab8daee3a06544785cfb985..3b7b17b5d3e53fb37cc39739314b8c6887e32155 100644 (file)
--- a/Include/internal/pycore_uop_ids.h
+++ b/Include/internal/pycore_uop_ids.h
@@ -382,7 +382,7 @@ extern "C" {
 #define _BINARY_OP_SUBSCR_INIT_CALL_r11 575
 #define _BINARY_OP_SUBSCR_INIT_CALL_r21 576
 #define _BINARY_OP_SUBSCR_INIT_CALL_r31 577
-#define _BINARY_OP_SUBSCR_LIST_INT_r21 578
+#define _BINARY_OP_SUBSCR_LIST_INT_r23 578
 #define _BINARY_OP_SUBSCR_LIST_SLICE_r21 579
 #define _BINARY_OP_SUBSCR_STR_INT_r21 580
 #define _BINARY_OP_SUBSCR_TUPLE_INT_r21 581

diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h
index 0cc38919565bc13368cec685b56639d2bebbb125..99ade22c0f437a678f89026d94ee23aec7af60c8 100644 (file)
--- a/Include/internal/pycore_uop_metadata.h
+++ b/Include/internal/pycore_uop_metadata.h
@@ -1108,7 +1108,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
         .entries = {
             { -1, -1, -1 },
             { -1, -1, -1 },
-            { 1, 2, _BINARY_OP_SUBSCR_LIST_INT_r21 },
+            { 3, 2, _BINARY_OP_SUBSCR_LIST_INT_r23 },
             { -1, -1, -1 },
         },
     },
@@ -3422,7 +3422,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
     [_BINARY_OP_EXTEND_r21] = _BINARY_OP_EXTEND,
     [_BINARY_SLICE_r31] = _BINARY_SLICE,
     [_STORE_SLICE_r30] = _STORE_SLICE,
-    [_BINARY_OP_SUBSCR_LIST_INT_r21] = _BINARY_OP_SUBSCR_LIST_INT,
+    [_BINARY_OP_SUBSCR_LIST_INT_r23] = _BINARY_OP_SUBSCR_LIST_INT,
     [_BINARY_OP_SUBSCR_LIST_SLICE_r21] = _BINARY_OP_SUBSCR_LIST_SLICE,
     [_BINARY_OP_SUBSCR_STR_INT_r21] = _BINARY_OP_SUBSCR_STR_INT,
     [_GUARD_NOS_TUPLE_r02] = _GUARD_NOS_TUPLE,
@@ -3930,7 +3930,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
     [_BINARY_OP_SUBSCR_INIT_CALL_r21] = "_BINARY_OP_SUBSCR_INIT_CALL_r21",
     [_BINARY_OP_SUBSCR_INIT_CALL_r31] = "_BINARY_OP_SUBSCR_INIT_CALL_r31",
     [_BINARY_OP_SUBSCR_LIST_INT] = "_BINARY_OP_SUBSCR_LIST_INT",
-    [_BINARY_OP_SUBSCR_LIST_INT_r21] = "_BINARY_OP_SUBSCR_LIST_INT_r21",
+    [_BINARY_OP_SUBSCR_LIST_INT_r23] = "_BINARY_OP_SUBSCR_LIST_INT_r23",
     [_BINARY_OP_SUBSCR_LIST_SLICE] = "_BINARY_OP_SUBSCR_LIST_SLICE",
     [_BINARY_OP_SUBSCR_LIST_SLICE_r21] = "_BINARY_OP_SUBSCR_LIST_SLICE_r21",
     [_BINARY_OP_SUBSCR_STR_INT] = "_BINARY_OP_SUBSCR_STR_INT",

diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py
index 6f1b115b31257f97ec6723d82eac12894603a3b5..a35dea8557ddbbe0f9d329cb77b2ea96794a3077 100644 (file)
--- a/Lib/test/test_capi/test_opt.py
+++ b/Lib/test/test_capi/test_opt.py
@@ -2988,6 +2988,24 @@ class TestUopsOptimization(unittest.TestCase):
         for _ in range(TIER2_THRESHOLD+1):
             obj.attr = EvilAttr(obj.__dict__)
 
+    def test_binary_subscr_list_int(self):
+        def testfunc(n):
+            l = [1]
+            x = 0
+            for _ in range(n):
+                y = l[0]
+                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("_BINARY_OP_SUBSCR_LIST_INT", uops)
+        self.assertNotIn("_POP_TOP", uops)
+        self.assertNotIn("_POP_TOP_INT", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
 
 def global_identity(x):
     return x

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 3e3b3d2f446d3357643f51a9835174dfe1fff287..1c6d0bb9cdf256201d4a7e74c8a711c92ed3d564 100644 (file)
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -893,9 +893,9 @@ dummy_func(
         macro(STORE_SLICE) = _SPECIALIZE_STORE_SLICE + _STORE_SLICE;
 
         macro(BINARY_OP_SUBSCR_LIST_INT) =
-            _GUARD_TOS_INT + _GUARD_NOS_LIST + unused/5 + _BINARY_OP_SUBSCR_LIST_INT;
+            _GUARD_TOS_INT + _GUARD_NOS_LIST + unused/5 + _BINARY_OP_SUBSCR_LIST_INT + _POP_TOP_INT + POP_TOP;
 
-        op(_BINARY_OP_SUBSCR_LIST_INT, (list_st, sub_st -- res)) {
+        op(_BINARY_OP_SUBSCR_LIST_INT, (list_st, sub_st -- res, ls, ss)) {
             PyObject *sub = PyStackRef_AsPyObjectBorrow(sub_st);
             PyObject *list = PyStackRef_AsPyObjectBorrow(list_st);
 
@@ -918,7 +918,9 @@ dummy_func(
             res = PyStackRef_FromPyObjectNew(res_o);
 #endif
             STAT_INC(BINARY_OP, hit);
-            DECREF_INPUTS();
+            ls = list_st;
+            ss = sub_st;
+            INPUTS_DEAD();
         }
 
         macro(BINARY_OP_SUBSCR_LIST_SLICE) =

diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index b7ee752cd44e16fcee25206f1a0903dedd9b5596..36ae739e58d6c51ba2c3f49ae106ec1db49a38ca 100644 (file)
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -4544,12 +4544,14 @@
             break;
         }
 
-        case _BINARY_OP_SUBSCR_LIST_INT_r21: {
+        case _BINARY_OP_SUBSCR_LIST_INT_r23: {
             CHECK_CURRENT_CACHED_VALUES(2);
             assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());
             _PyStackRef sub_st;
             _PyStackRef list_st;
             _PyStackRef res;
+            _PyStackRef ls;
+            _PyStackRef ss;
             _PyStackRef _stack_item_0 = _tos_cache0;
             _PyStackRef _stack_item_1 = _tos_cache1;
             sub_st = _stack_item_1;
@@ -4600,23 +4602,13 @@
             stack_pointer += 2;
             #endif
             STAT_INC(BINARY_OP, hit);
-            _PyFrame_SetStackPointer(frame, stack_pointer);
-            _PyStackRef tmp = list_st;
-            list_st = res;
-            stack_pointer[-2] = list_st;
-            PyStackRef_CLOSE(tmp);
-            tmp = sub_st;
-            sub_st = PyStackRef_NULL;
-            stack_pointer[-1] = sub_st;
-            PyStackRef_CLOSE(tmp);
-            stack_pointer = _PyFrame_GetStackPointer(frame);
-            stack_pointer += -1;
-            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
+            ls = list_st;
+            ss = sub_st;
+            _tos_cache2 = ss;
+            _tos_cache1 = ls;
             _tos_cache0 = res;
-            _tos_cache1 = PyStackRef_ZERO_BITS;
-            _tos_cache2 = PyStackRef_ZERO_BITS;
-            SET_CURRENT_CACHED_VALUES(1);
-            stack_pointer += -1;
+            SET_CURRENT_CACHED_VALUES(3);
+            stack_pointer += -2;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());
             break;

diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 28dc7b25f0d2cc87fb2a7a7bb335a5db06dbfe68..0c0256bcc3b753e918588043459766ba014737dc 100644 (file)
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -764,6 +764,8 @@
             _PyStackRef list_st;
             _PyStackRef sub_st;
             _PyStackRef res;
+            _PyStackRef ls;
+            _PyStackRef ss;
             // _GUARD_TOS_INT
             {
                 value = stack_pointer[-1];
@@ -822,18 +824,24 @@
                 res = PyStackRef_FromPyObjectNew(res_o);
                 #endif
                 STAT_INC(BINARY_OP, hit);
-                _PyFrame_SetStackPointer(frame, stack_pointer);
-                _PyStackRef tmp = list_st;
-                list_st = res;
-                stack_pointer[-2] = list_st;
-                PyStackRef_CLOSE(tmp);
-                tmp = sub_st;
-                sub_st = PyStackRef_NULL;
-                stack_pointer[-1] = sub_st;
-                PyStackRef_CLOSE(tmp);
-                stack_pointer = _PyFrame_GetStackPointer(frame);
+                ls = list_st;
+                ss = sub_st;
+            }
+            // _POP_TOP_INT
+            {
+                value = ss;
+                assert(PyLong_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
+                PyStackRef_CLOSE_SPECIALIZED(value, _PyLong_ExactDealloc);
+            }
+            // _POP_TOP
+            {
+                value = ls;
+                stack_pointer[-2] = res;
                 stack_pointer += -1;
                 ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
+                _PyFrame_SetStackPointer(frame, stack_pointer);
+                PyStackRef_XCLOSE(value);
+                stack_pointer = _PyFrame_GetStackPointer(frame);
             }
             DISPATCH();
         }

diff --git a/Python/optimizer_bytecodes.c b/Python/optimizer_bytecodes.c
index e3921054f7cb4366c924f3b8b60a71fdd3982d92..9578b56e70739302a96fcf210fdb51baf8ceb7dc 100644 (file)
--- a/Python/optimizer_bytecodes.c
+++ b/Python/optimizer_bytecodes.c
@@ -1417,6 +1417,12 @@ dummy_func(void) {
         }
     }
 
+    op(_BINARY_OP_SUBSCR_LIST_INT, (list_st, sub_st -- res, ls, ss)) {
+        res = sym_new_unknown(ctx);
+        ls = list_st;
+        ss = sub_st;
+    }
+
 
 // END BYTECODES //
 

diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index 32d60c76a4f3ce186f15dc296d0046fc76ef4aeb..f047cd82d4c5e712274b7de9b8408785d9cccde8 100644 (file)
--- a/Python/optimizer_cases.c.h
+++ b/Python/optimizer_cases.c.h
@@ -763,11 +763,21 @@
         }
 
         case _BINARY_OP_SUBSCR_LIST_INT: {
+            JitOptRef sub_st;
+            JitOptRef list_st;
             JitOptRef res;
-            res = sym_new_not_null(ctx);
-            CHECK_STACK_BOUNDS(-1);
+            JitOptRef ls;
+            JitOptRef ss;
+            sub_st = stack_pointer[-1];
+            list_st = stack_pointer[-2];
+            res = sym_new_unknown(ctx);
+            ls = list_st;
+            ss = sub_st;
+            CHECK_STACK_BOUNDS(1);
             stack_pointer[-2] = res;
-            stack_pointer += -1;
+            stack_pointer[-1] = ls;
+            stack_pointer[0] = ss;
+            stack_pointer += 1;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             break;
         }