gh-134584: Eliminate redundant refcounting from _STORE_SUBSCR_LIST_INT (gh-142703)

diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h
index 0fb01982b6aab75ad3a006c272127753394a9f1d..0a29fabe7676dcb74725ae7b1fa5ff63721c2bd5 100644 (file)
--- a/Include/internal/pycore_opcode_metadata.h
+++ b/Include/internal/pycore_opcode_metadata.h
@@ -1494,7 +1494,7 @@ _PyOpcode_macro_expansion[256] = {
     [STORE_SLICE] = { .nuops = 1, .uops = { { _STORE_SLICE, OPARG_SIMPLE, 0 } } },
     [STORE_SUBSCR] = { .nuops = 1, .uops = { { _STORE_SUBSCR, OPARG_SIMPLE, 0 } } },
     [STORE_SUBSCR_DICT] = { .nuops = 2, .uops = { { _GUARD_NOS_DICT, OPARG_SIMPLE, 0 }, { _STORE_SUBSCR_DICT, OPARG_SIMPLE, 1 } } },
-    [STORE_SUBSCR_LIST_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _STORE_SUBSCR_LIST_INT, OPARG_SIMPLE, 1 } } },
+    [STORE_SUBSCR_LIST_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _STORE_SUBSCR_LIST_INT, OPARG_SIMPLE, 1 }, { _POP_TOP_INT, OPARG_SIMPLE, 1 }, { _POP_TOP, OPARG_SIMPLE, 1 } } },
     [SWAP] = { .nuops = 1, .uops = { { _SWAP, OPARG_SIMPLE, 0 } } },
     [TO_BOOL] = { .nuops = 1, .uops = { { _TO_BOOL, OPARG_SIMPLE, 2 } } },
     [TO_BOOL_ALWAYS_TRUE] = { .nuops = 2, .uops = { { _GUARD_TYPE_VERSION, 2, 1 }, { _REPLACE_WITH_TRUE, OPARG_SIMPLE, 3 } } },

diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h
index e871e7699bf7fb253b89e0556d82a986aa679d44..64e51bd2b8bb580384b1b529d1d7be1a3f6ff0de 100644 (file)
--- a/Include/internal/pycore_uop_ids.h
+++ b/Include/internal/pycore_uop_ids.h
@@ -1068,7 +1068,7 @@ extern "C" {
 #define _STORE_SLICE_r30 1261
 #define _STORE_SUBSCR_r30 1262
 #define _STORE_SUBSCR_DICT_r30 1263
-#define _STORE_SUBSCR_LIST_INT_r30 1264
+#define _STORE_SUBSCR_LIST_INT_r32 1264
 #define _SWAP_r11 1265
 #define _SWAP_2_r02 1266
 #define _SWAP_2_r12 1267

diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h
index 5ae05358592a0fb759ecae77dc9810192aa93823..5fa375a8ce6b4ab8f156925d2b1cc4756293deb4 100644 (file)
--- a/Include/internal/pycore_uop_metadata.h
+++ b/Include/internal/pycore_uop_metadata.h
@@ -1265,7 +1265,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
             { -1, -1, -1 },
             { -1, -1, -1 },
             { -1, -1, -1 },
-            { 0, 3, _STORE_SUBSCR_LIST_INT_r30 },
+            { 2, 3, _STORE_SUBSCR_LIST_INT_r32 },
         },
     },
     [_STORE_SUBSCR_DICT] = {
@@ -3498,7 +3498,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
     [_LIST_APPEND_r10] = _LIST_APPEND,
     [_SET_ADD_r10] = _SET_ADD,
     [_STORE_SUBSCR_r30] = _STORE_SUBSCR,
-    [_STORE_SUBSCR_LIST_INT_r30] = _STORE_SUBSCR_LIST_INT,
+    [_STORE_SUBSCR_LIST_INT_r32] = _STORE_SUBSCR_LIST_INT,
     [_STORE_SUBSCR_DICT_r30] = _STORE_SUBSCR_DICT,
     [_DELETE_SUBSCR_r20] = _DELETE_SUBSCR,
     [_CALL_INTRINSIC_1_r11] = _CALL_INTRINSIC_1,
@@ -4869,7 +4869,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
     [_STORE_SUBSCR_DICT] = "_STORE_SUBSCR_DICT",
     [_STORE_SUBSCR_DICT_r30] = "_STORE_SUBSCR_DICT_r30",
     [_STORE_SUBSCR_LIST_INT] = "_STORE_SUBSCR_LIST_INT",
-    [_STORE_SUBSCR_LIST_INT_r30] = "_STORE_SUBSCR_LIST_INT_r30",
+    [_STORE_SUBSCR_LIST_INT_r32] = "_STORE_SUBSCR_LIST_INT_r32",
     [_SWAP] = "_SWAP",
     [_SWAP_r11] = "_SWAP_r11",
     [_SWAP_2] = "_SWAP_2",

diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py
index 32c00b57e1a460a1b08b654b37accc0eafea9cee..0f6ed3d85f033021f059514f7465b9378c32dcb7 100644 (file)
--- a/Lib/test/test_capi/test_opt.py
+++ b/Lib/test/test_capi/test_opt.py
@@ -2499,6 +2499,24 @@ class TestUopsOptimization(unittest.TestCase):
         self.assertNotIn("_GUARD_TOS_INT", uops)
         self.assertNotIn("_GUARD_NOS_INT", uops)
 
+    def test_store_subscr_int(self):
+        def testfunc(n):
+            l = [0, 0, 0, 0]
+            for _ in range(n):
+                l[0] = 1
+                l[1] = 2
+                l[2] = 3
+                l[3] = 4
+            return sum(l)
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, 10)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertNotIn("_POP_TOP", uops)
+        self.assertNotIn("_POP_TOP_INT", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+
     def test_attr_promotion_failure(self):
         # We're not testing for any specific uops here, just
         # testing it doesn't crash.

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 7e78e726b911e5016e971957931a32f47fcc9f02..daa3d218e387f98314525de919f187d650258d17 100644 (file)
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -1126,9 +1126,9 @@ dummy_func(
         macro(STORE_SUBSCR) = _SPECIALIZE_STORE_SUBSCR + _STORE_SUBSCR;
 
         macro(STORE_SUBSCR_LIST_INT) =
-            _GUARD_TOS_INT + _GUARD_NOS_LIST + unused/1 + _STORE_SUBSCR_LIST_INT;
+            _GUARD_TOS_INT + _GUARD_NOS_LIST + unused/1 + _STORE_SUBSCR_LIST_INT + _POP_TOP_INT + POP_TOP;
 
-        op(_STORE_SUBSCR_LIST_INT, (value, list_st, sub_st -- )) {
+        op(_STORE_SUBSCR_LIST_INT, (value, list_st, sub_st -- ls, ss)) {
             PyObject *sub = PyStackRef_AsPyObjectBorrow(sub_st);
             PyObject *list = PyStackRef_AsPyObjectBorrow(list_st);
 
@@ -1151,9 +1151,9 @@ dummy_func(
                                         PyStackRef_AsPyObjectSteal(value));
             assert(old_value != NULL);
             UNLOCK_OBJECT(list);  // unlock before decrefs!
-            PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
-            DEAD(sub_st);
-            PyStackRef_CLOSE(list_st);
+            INPUTS_DEAD();
+            ls = list_st;
+            ss = sub_st;
             Py_DECREF(old_value);
         }
 

diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 727d3695bb11f0a4fd163ba26074b0506165a2dc..2a1156091e3d37f01f2bcd74c3cc1152513bc10a 100644 (file)
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -5703,12 +5703,14 @@
             break;
         }
 
-        case _STORE_SUBSCR_LIST_INT_r30: {
+        case _STORE_SUBSCR_LIST_INT_r32: {
             CHECK_CURRENT_CACHED_VALUES(3);
             assert(WITHIN_STACK_BOUNDS_WITH_CACHE());
             _PyStackRef sub_st;
             _PyStackRef list_st;
             _PyStackRef value;
+            _PyStackRef ls;
+            _PyStackRef ss;
             _PyStackRef _stack_item_0 = _tos_cache0;
             _PyStackRef _stack_item_1 = _tos_cache1;
             _PyStackRef _stack_item_2 = _tos_cache2;
@@ -5744,15 +5746,21 @@
                                         PyStackRef_AsPyObjectSteal(value));
             assert(old_value != NULL);
             UNLOCK_OBJECT(list);
-            PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
+            ls = list_st;
+            ss = sub_st;
+            stack_pointer[0] = ls;
+            stack_pointer[1] = ss;
+            stack_pointer += 2;
+            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             _PyFrame_SetStackPointer(frame, stack_pointer);
-            PyStackRef_CLOSE(list_st);
             Py_DECREF(old_value);
             stack_pointer = _PyFrame_GetStackPointer(frame);
-            _tos_cache0 = PyStackRef_ZERO_BITS;
-            _tos_cache1 = PyStackRef_ZERO_BITS;
+            _tos_cache1 = ss;
+            _tos_cache0 = ls;
             _tos_cache2 = PyStackRef_ZERO_BITS;
-            SET_CURRENT_CACHED_VALUES(0);
+            SET_CURRENT_CACHED_VALUES(2);
+            stack_pointer += -2;
+            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 e1860f5d9d8f6c2ea993a258ff1788a43cef9981..ab9373e0af5afcbde975dd3fcf9b911c5a8db31f 100644 (file)
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -11024,6 +11024,8 @@
             _PyStackRef nos;
             _PyStackRef list_st;
             _PyStackRef sub_st;
+            _PyStackRef ls;
+            _PyStackRef ss;
             // _GUARD_TOS_INT
             {
                 value = stack_pointer[-1];
@@ -11079,14 +11081,31 @@
                                         PyStackRef_AsPyObjectSteal(value));
                 assert(old_value != NULL);
                 UNLOCK_OBJECT(list);
-                PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
-                stack_pointer += -3;
+                ls = list_st;
+                ss = sub_st;
+                stack_pointer[-3] = ls;
+                stack_pointer[-2] = ss;
+                stack_pointer += -1;
                 ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
                 _PyFrame_SetStackPointer(frame, stack_pointer);
-                PyStackRef_CLOSE(list_st);
                 Py_DECREF(old_value);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
             }
+            // _POP_TOP_INT
+            {
+                value = ss;
+                assert(PyLong_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
+                PyStackRef_CLOSE_SPECIALIZED(value, _PyLong_ExactDealloc);
+            }
+            // _POP_TOP
+            {
+                value = ls;
+                stack_pointer += -2;
+                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 071f4ca31ba7a47ae932065339da85c3f077a67b..66aecf7ef54355ebbfe2db02fd78bf1ca98f72e0 100644 (file)
--- a/Python/optimizer_bytecodes.c
+++ b/Python/optimizer_bytecodes.c
@@ -103,6 +103,12 @@ dummy_func(void) {
         GETLOCAL(oparg) = value;
     }
 
+    op(_STORE_SUBSCR_LIST_INT, (value, list_st, sub_st -- ls, ss)) {
+        (void)value;
+        ls = list_st;
+        ss = sub_st;
+    }
+
     op(_PUSH_NULL, (-- res)) {
         res = sym_new_null(ctx);
     }
@@ -529,6 +535,12 @@ dummy_func(void) {
         }
     }
 
+    op(_POP_TOP_INT, (value --)) {
+        if (PyJitRef_IsBorrowed(value)) {
+            REPLACE_OP(this_instr, _POP_TOP_NOP, 0, 0);
+        }
+    }
+
     op(_COPY, (bottom, unused[oparg-1] -- bottom, unused[oparg-1], top)) {
         assert(oparg > 0);
         top = bottom;

diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index 5c5efdaa1fd880d68b502fda0518806d625b8582..1a3d4ad50bd82406ffb7c2964cbf91aa57ff96f6 100644 (file)
--- a/Python/optimizer_cases.c.h
+++ b/Python/optimizer_cases.c.h
@@ -141,6 +141,11 @@
         }
 
         case _POP_TOP_INT: {
+            JitOptRef value;
+            value = stack_pointer[-1];
+            if (PyJitRef_IsBorrowed(value)) {
+                REPLACE_OP(this_instr, _POP_TOP_NOP, 0, 0);
+            }
             CHECK_STACK_BOUNDS(-1);
             stack_pointer += -1;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
@@ -1123,8 +1128,21 @@
         }
 
         case _STORE_SUBSCR_LIST_INT: {
-            CHECK_STACK_BOUNDS(-3);
-            stack_pointer += -3;
+            JitOptRef sub_st;
+            JitOptRef list_st;
+            JitOptRef value;
+            JitOptRef ls;
+            JitOptRef ss;
+            sub_st = stack_pointer[-1];
+            list_st = stack_pointer[-2];
+            value = stack_pointer[-3];
+            (void)value;
+            ls = list_st;
+            ss = sub_st;
+            CHECK_STACK_BOUNDS(-1);
+            stack_pointer[-3] = ls;
+            stack_pointer[-2] = ss;
+            stack_pointer += -1;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             break;
         }