gh-134584: Eliminate redundant refcounting from `_BINARY_OP_SUBSCR_TUPLE_INT` (GH...

diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h
index 351cf56355b7d095f8d95b6eda1a375604331cb0..08bddfbfbe661919ecb53f5fc6fce2ca249825b9 100644 (file)
--- a/Include/internal/pycore_opcode_metadata.h
+++ b/Include/internal/pycore_opcode_metadata.h
@@ -1341,7 +1341,7 @@ _PyOpcode_macro_expansion[256] = {
     [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 = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_UNICODE, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_STR_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP, 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 } } },
+    [BINARY_OP_SUBSCR_TUPLE_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_TUPLE, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_TUPLE_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP, OPARG_SIMPLE, 5 } } },
     [BINARY_OP_SUBTRACT_FLOAT] = { .nuops = 5, .uops = { { _GUARD_TOS_FLOAT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_FLOAT, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBTRACT_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 } } },
     [BINARY_OP_SUBTRACT_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_INT, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBTRACT_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 } } },
     [BINARY_SLICE] = { .nuops = 1, .uops = { { _BINARY_SLICE, OPARG_SIMPLE, 0 } } },

diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h
index 204210ff101efe92dfe3808b6df2ae3b06b96760..69f1c3829cd239b9c668108f59f37d8e3507b413 100644 (file)
--- a/Include/internal/pycore_uop_ids.h
+++ b/Include/internal/pycore_uop_ids.h
@@ -387,7 +387,7 @@ extern "C" {
 #define _BINARY_OP_SUBSCR_LIST_INT_r23 580
 #define _BINARY_OP_SUBSCR_LIST_SLICE_r21 581
 #define _BINARY_OP_SUBSCR_STR_INT_r23 582
-#define _BINARY_OP_SUBSCR_TUPLE_INT_r21 583
+#define _BINARY_OP_SUBSCR_TUPLE_INT_r23 583
 #define _BINARY_OP_SUBTRACT_FLOAT_r03 584
 #define _BINARY_OP_SUBTRACT_FLOAT_r13 585
 #define _BINARY_OP_SUBTRACT_FLOAT_r23 586

diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h
index f751f642b81ff86d571593dbae2c36f5ce5fb97f..0a49231e53f44aad38534fede70148543d740ab8 100644 (file)
--- a/Include/internal/pycore_uop_metadata.h
+++ b/Include/internal/pycore_uop_metadata.h
@@ -122,7 +122,7 @@ const uint32_t _PyUop_Flags[MAX_UOP_ID+1] = {
     [_BINARY_OP_SUBSCR_STR_INT] = HAS_DEOPT_FLAG,
     [_GUARD_NOS_TUPLE] = HAS_EXIT_FLAG,
     [_GUARD_TOS_TUPLE] = HAS_EXIT_FLAG,
-    [_BINARY_OP_SUBSCR_TUPLE_INT] = HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG,
+    [_BINARY_OP_SUBSCR_TUPLE_INT] = HAS_DEOPT_FLAG,
     [_GUARD_NOS_DICT] = HAS_EXIT_FLAG,
     [_GUARD_TOS_DICT] = HAS_EXIT_FLAG,
     [_BINARY_OP_SUBSCR_DICT] = HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
@@ -1155,7 +1155,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
         .entries = {
             { -1, -1, -1 },
             { -1, -1, -1 },
-            { 1, 2, _BINARY_OP_SUBSCR_TUPLE_INT_r21 },
+            { 3, 2, _BINARY_OP_SUBSCR_TUPLE_INT_r23 },
             { -1, -1, -1 },
         },
     },
@@ -3453,7 +3453,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
     [_GUARD_TOS_TUPLE_r11] = _GUARD_TOS_TUPLE,
     [_GUARD_TOS_TUPLE_r22] = _GUARD_TOS_TUPLE,
     [_GUARD_TOS_TUPLE_r33] = _GUARD_TOS_TUPLE,
-    [_BINARY_OP_SUBSCR_TUPLE_INT_r21] = _BINARY_OP_SUBSCR_TUPLE_INT,
+    [_BINARY_OP_SUBSCR_TUPLE_INT_r23] = _BINARY_OP_SUBSCR_TUPLE_INT,
     [_GUARD_NOS_DICT_r02] = _GUARD_NOS_DICT,
     [_GUARD_NOS_DICT_r12] = _GUARD_NOS_DICT,
     [_GUARD_NOS_DICT_r22] = _GUARD_NOS_DICT,
@@ -3969,7 +3969,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
     [_BINARY_OP_SUBSCR_STR_INT] = "_BINARY_OP_SUBSCR_STR_INT",
     [_BINARY_OP_SUBSCR_STR_INT_r23] = "_BINARY_OP_SUBSCR_STR_INT_r23",
     [_BINARY_OP_SUBSCR_TUPLE_INT] = "_BINARY_OP_SUBSCR_TUPLE_INT",
-    [_BINARY_OP_SUBSCR_TUPLE_INT_r21] = "_BINARY_OP_SUBSCR_TUPLE_INT_r21",
+    [_BINARY_OP_SUBSCR_TUPLE_INT_r23] = "_BINARY_OP_SUBSCR_TUPLE_INT_r23",
     [_BINARY_OP_SUBTRACT_FLOAT] = "_BINARY_OP_SUBTRACT_FLOAT",
     [_BINARY_OP_SUBTRACT_FLOAT_r03] = "_BINARY_OP_SUBTRACT_FLOAT_r03",
     [_BINARY_OP_SUBTRACT_FLOAT_r13] = "_BINARY_OP_SUBTRACT_FLOAT_r13",

diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py
index 16288a447e20febcb9fc4b3c1725916fd062fa33..8f7314d579df6ff75211a18cafc16872f990adb9 100644 (file)
--- a/Lib/test/test_capi/test_opt.py
+++ b/Lib/test/test_capi/test_opt.py
@@ -3114,6 +3114,25 @@ class TestUopsOptimization(unittest.TestCase):
         self.assertNotIn("_POP_TOP_INT", uops)
         self.assertIn("_POP_TOP_NOP", uops)
 
+    def test_binary_subscr_tuple_int(self):
+        def testfunc(n):
+            t = (1,)
+            x = 0
+            for _ in range(n):
+                y = t[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_TUPLE_INT", uops)
+        self.assertNotIn("_POP_TOP", uops)
+        self.assertNotIn("_POP_TOP_INT", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+
     def test_143026(self):
         # https://github.com/python/cpython/issues/143026
 

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index d21c17d072c3f99f79bc8f47dd4da1d324fb07bb..950b9f08f2ead1c68e58b939b1e9e14dd7b4a385 100644 (file)
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -975,9 +975,9 @@ dummy_func(
         }
 
         macro(BINARY_OP_SUBSCR_TUPLE_INT) =
-            _GUARD_TOS_INT + _GUARD_NOS_TUPLE + unused/5 + _BINARY_OP_SUBSCR_TUPLE_INT;
+            _GUARD_TOS_INT + _GUARD_NOS_TUPLE + unused/5 + _BINARY_OP_SUBSCR_TUPLE_INT + _POP_TOP_INT + POP_TOP;
 
-        op(_BINARY_OP_SUBSCR_TUPLE_INT, (tuple_st, sub_st -- res)) {
+        op(_BINARY_OP_SUBSCR_TUPLE_INT, (tuple_st, sub_st -- res, ts, ss)) {
             PyObject *sub = PyStackRef_AsPyObjectBorrow(sub_st);
             PyObject *tuple = PyStackRef_AsPyObjectBorrow(tuple_st);
 
@@ -991,9 +991,10 @@ dummy_func(
             STAT_INC(BINARY_OP, hit);
             PyObject *res_o = PyTuple_GET_ITEM(tuple, index);
             assert(res_o != NULL);
-            PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
             res = PyStackRef_FromPyObjectNew(res_o);
-            DECREF_INPUTS();
+            ts = tuple_st;
+            ss = sub_st;
+            INPUTS_DEAD();
         }
 
         op(_GUARD_NOS_DICT, (nos, unused -- nos, unused)) {

diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index f64747d6f27f2ae9abaff05c8eb05d945a4e751e..dc21a4aa90cf46f8cca48988a5d00bd91aec498f 100644 (file)
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -4888,12 +4888,14 @@
             break;
         }
 
-        case _BINARY_OP_SUBSCR_TUPLE_INT_r21: {
+        case _BINARY_OP_SUBSCR_TUPLE_INT_r23: {
             CHECK_CURRENT_CACHED_VALUES(2);
             assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());
             _PyStackRef sub_st;
             _PyStackRef tuple_st;
             _PyStackRef res;
+            _PyStackRef ts;
+            _PyStackRef ss;
             _PyStackRef _stack_item_0 = _tos_cache0;
             _PyStackRef _stack_item_1 = _tos_cache1;
             sub_st = _stack_item_1;
@@ -4920,23 +4922,13 @@
             STAT_INC(BINARY_OP, hit);
             PyObject *res_o = PyTuple_GET_ITEM(tuple, index);
             assert(res_o != NULL);
-            PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
             res = PyStackRef_FromPyObjectNew(res_o);
-            stack_pointer[0] = tuple_st;
-            stack_pointer += 1;
-            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
-            _PyFrame_SetStackPointer(frame, stack_pointer);
-            _PyStackRef tmp = tuple_st;
-            tuple_st = res;
-            stack_pointer[-1] = tuple_st;
-            PyStackRef_CLOSE(tmp);
-            stack_pointer = _PyFrame_GetStackPointer(frame);
+            ts = tuple_st;
+            ss = sub_st;
+            _tos_cache2 = ss;
+            _tos_cache1 = ts;
             _tos_cache0 = res;
-            _tos_cache1 = PyStackRef_ZERO_BITS;
-            _tos_cache2 = PyStackRef_ZERO_BITS;
-            SET_CURRENT_CACHED_VALUES(1);
-            stack_pointer += -1;
-            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
+            SET_CURRENT_CACHED_VALUES(3);
             assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());
             break;
         }

diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index baf199969de94e67ae508f6440068d59d320785a..5f326d5c3caffc24759e7f58b91c3c03e8e80c5d 100644 (file)
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -1024,6 +1024,8 @@
             _PyStackRef tuple_st;
             _PyStackRef sub_st;
             _PyStackRef res;
+            _PyStackRef ts;
+            _PyStackRef ss;
             // _GUARD_TOS_INT
             {
                 value = stack_pointer[-1];
@@ -1067,15 +1069,24 @@
                 STAT_INC(BINARY_OP, hit);
                 PyObject *res_o = PyTuple_GET_ITEM(tuple, index);
                 assert(res_o != NULL);
-                PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
                 res = PyStackRef_FromPyObjectNew(res_o);
+                ts = tuple_st;
+                ss = sub_st;
+            }
+            // _POP_TOP_INT
+            {
+                value = ss;
+                assert(PyLong_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
+                PyStackRef_CLOSE_SPECIALIZED(value, _PyLong_ExactDealloc);
+            }
+            // _POP_TOP
+            {
+                value = ts;
+                stack_pointer[-2] = res;
                 stack_pointer += -1;
                 ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
                 _PyFrame_SetStackPointer(frame, stack_pointer);
-                _PyStackRef tmp = tuple_st;
-                tuple_st = res;
-                stack_pointer[-1] = tuple_st;
-                PyStackRef_CLOSE(tmp);
+                PyStackRef_XCLOSE(value);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
             }
             DISPATCH();

diff --git a/Python/optimizer_bytecodes.c b/Python/optimizer_bytecodes.c
index b40b597643dc945c83c9162d6c145fa6f4becdef..c53a2fb7570c0dc25c0d272d4d46ccfec7343580 100644 (file)
--- a/Python/optimizer_bytecodes.c
+++ b/Python/optimizer_bytecodes.c
@@ -335,7 +335,7 @@ dummy_func(void) {
         i = sub_st;
     }
 
-    op(_BINARY_OP_SUBSCR_TUPLE_INT, (tuple_st, sub_st -- res)) {
+    op(_BINARY_OP_SUBSCR_TUPLE_INT, (tuple_st, sub_st -- res, ts, ss)) {
         assert(sym_matches_type(tuple_st, &PyTuple_Type));
         if (sym_is_const(ctx, sub_st)) {
             assert(PyLong_CheckExact(sym_get_const(ctx, sub_st)));
@@ -354,6 +354,8 @@ dummy_func(void) {
         else {
             res = sym_new_not_null(ctx);
         }
+        ts = tuple_st;
+        ss = sub_st;
     }
 
     op(_TO_BOOL, (value -- res)) {

diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index a17a5688847e07dd388fd171393c0f2dd93e7938..9f1337e7ef3b92ebe215baa25accdd931fcff00a 100644 (file)
--- a/Python/optimizer_cases.c.h
+++ b/Python/optimizer_cases.c.h
@@ -836,6 +836,8 @@
             JitOptRef sub_st;
             JitOptRef tuple_st;
             JitOptRef res;
+            JitOptRef ts;
+            JitOptRef ss;
             sub_st = stack_pointer[-1];
             tuple_st = stack_pointer[-2];
             assert(sym_matches_type(tuple_st, &PyTuple_Type));
@@ -855,9 +857,13 @@
             else {
                 res = sym_new_not_null(ctx);
             }
-            CHECK_STACK_BOUNDS(-1);
+            ts = tuple_st;
+            ss = sub_st;
+            CHECK_STACK_BOUNDS(1);
             stack_pointer[-2] = res;
-            stack_pointer += -1;
+            stack_pointer[-1] = ts;
+            stack_pointer[0] = ss;
+            stack_pointer += 1;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             break;
         }