GH-134584: Remove redundant refcount for `BINARY_OP_SUBSCR_STR_INT` (#142844)

diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h
index 913359ee4bdc4fdbc8ac8c21e898b49fc25b95f9..92423fe51408fc22bb3f30395cde07557e1c0f88 100644 (file)
--- a/Include/internal/pycore_opcode_metadata.h
+++ b/Include/internal/pycore_opcode_metadata.h
@@ -1340,7 +1340,7 @@ _PyOpcode_macro_expansion[256] = {
     [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 = 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_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_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 } } },

diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h
index 3b7b17b5d3e53fb37cc39739314b8c6887e32155..e9851b767bc835e45d08d29ce7c737e7dbf9d4a8 100644 (file)
--- a/Include/internal/pycore_uop_ids.h
+++ b/Include/internal/pycore_uop_ids.h
@@ -384,7 +384,7 @@ extern "C" {
 #define _BINARY_OP_SUBSCR_INIT_CALL_r31 577
 #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_STR_INT_r23 580
 #define _BINARY_OP_SUBSCR_TUPLE_INT_r21 581
 #define _BINARY_OP_SUBTRACT_FLOAT_r03 582
 #define _BINARY_OP_SUBTRACT_FLOAT_r13 583

diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h
index 99ade22c0f437a678f89026d94ee23aec7af60c8..3a6bf7b3d76a8fa7bea38d593da13d5a40ba651e 100644 (file)
--- a/Include/internal/pycore_uop_metadata.h
+++ b/Include/internal/pycore_uop_metadata.h
@@ -119,7 +119,7 @@ const uint32_t _PyUop_Flags[MAX_UOP_ID+1] = {
     [_STORE_SLICE] = HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
     [_BINARY_OP_SUBSCR_LIST_INT] = HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG,
     [_BINARY_OP_SUBSCR_LIST_SLICE] = HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
-    [_BINARY_OP_SUBSCR_STR_INT] = HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG,
+    [_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,
@@ -1126,7 +1126,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
         .entries = {
             { -1, -1, -1 },
             { -1, -1, -1 },
-            { 1, 2, _BINARY_OP_SUBSCR_STR_INT_r21 },
+            { 3, 2, _BINARY_OP_SUBSCR_STR_INT_r23 },
             { -1, -1, -1 },
         },
     },
@@ -3424,7 +3424,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
     [_STORE_SLICE_r30] = _STORE_SLICE,
     [_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,
+    [_BINARY_OP_SUBSCR_STR_INT_r23] = _BINARY_OP_SUBSCR_STR_INT,
     [_GUARD_NOS_TUPLE_r02] = _GUARD_NOS_TUPLE,
     [_GUARD_NOS_TUPLE_r12] = _GUARD_NOS_TUPLE,
     [_GUARD_NOS_TUPLE_r22] = _GUARD_NOS_TUPLE,
@@ -3934,7 +3934,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
     [_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",
-    [_BINARY_OP_SUBSCR_STR_INT_r21] = "_BINARY_OP_SUBSCR_STR_INT_r21",
+    [_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_SUBTRACT_FLOAT] = "_BINARY_OP_SUBTRACT_FLOAT",

diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py
index a35dea8557ddbbe0f9d329cb77b2ea96794a3077..88ac1664fe5ddff3acb52568bf6a19780e675dc5 100644 (file)
--- a/Lib/test/test_capi/test_opt.py
+++ b/Lib/test/test_capi/test_opt.py
@@ -1877,6 +1877,26 @@ class TestUopsOptimization(unittest.TestCase):
         self.assertNotIn("_GUARD_TOS_UNICODE", uops)
         self.assertIn("_BINARY_OP_ADD_UNICODE", uops)
 
+    def test_binary_op_subscr_str_int(self):
+        def testfunc(n):
+            x = 0
+            s = "hello"
+            for _ in range(n):
+                c = s[1]  # _BINARY_OP_SUBSCR_STR_INT
+                if c == 'e':
+                    x += 1
+            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_STR_INT", uops)
+        self.assertIn("_COMPARE_OP_STR", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+        self.assertNotIn("_POP_TOP", uops)
+        self.assertNotIn("_POP_TOP_INT", uops)
+
     def test_call_type_1_guards_removed(self):
         def testfunc(n):
             x = 0

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 1c6d0bb9cdf256201d4a7e74c8a711c92ed3d564..4ce7968ddc29296a7fa8b19080d99339952c1525 100644 (file)
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -941,9 +941,9 @@ dummy_func(
         }
 
         macro(BINARY_OP_SUBSCR_STR_INT) =
-            _GUARD_TOS_INT + _GUARD_NOS_UNICODE + unused/5 + _BINARY_OP_SUBSCR_STR_INT;
+            _GUARD_TOS_INT + _GUARD_NOS_UNICODE + unused/5 + _BINARY_OP_SUBSCR_STR_INT + _POP_TOP_INT + POP_TOP;
 
-        op(_BINARY_OP_SUBSCR_STR_INT, (str_st, sub_st -- res)) {
+        op(_BINARY_OP_SUBSCR_STR_INT, (str_st, sub_st -- res, s, i)) {
             PyObject *sub = PyStackRef_AsPyObjectBorrow(sub_st);
             PyObject *str = PyStackRef_AsPyObjectBorrow(str_st);
 
@@ -958,9 +958,9 @@ dummy_func(
             assert(c < 128);
             STAT_INC(BINARY_OP, hit);
             PyObject *res_o = (PyObject*)&_Py_SINGLETON(strings).ascii[c];
-            PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
-            DEAD(sub_st);
-            PyStackRef_CLOSE(str_st);
+            INPUTS_DEAD();
+            s = str_st;
+            i = sub_st;
             res = PyStackRef_FromPyObjectBorrow(res_o);
         }
 

diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 36ae739e58d6c51ba2c3f49ae106ec1db49a38ca..74bd939c87d4ecafeede34f87c670cba07be96aa 100644 (file)
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -4661,12 +4661,14 @@
             break;
         }
 
-        case _BINARY_OP_SUBSCR_STR_INT_r21: {
+        case _BINARY_OP_SUBSCR_STR_INT_r23: {
             CHECK_CURRENT_CACHED_VALUES(2);
             assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());
             _PyStackRef sub_st;
             _PyStackRef str_st;
             _PyStackRef res;
+            _PyStackRef s;
+            _PyStackRef i;
             _PyStackRef _stack_item_0 = _tos_cache0;
             _PyStackRef _stack_item_1 = _tos_cache1;
             sub_st = _stack_item_1;
@@ -4701,15 +4703,13 @@
             assert(c < 128);
             STAT_INC(BINARY_OP, hit);
             PyObject *res_o = (PyObject*)&_Py_SINGLETON(strings).ascii[c];
-            PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
-            _PyFrame_SetStackPointer(frame, stack_pointer);
-            PyStackRef_CLOSE(str_st);
-            stack_pointer = _PyFrame_GetStackPointer(frame);
+            s = str_st;
+            i = sub_st;
             res = PyStackRef_FromPyObjectBorrow(res_o);
+            _tos_cache2 = i;
+            _tos_cache1 = s;
             _tos_cache0 = res;
-            _tos_cache1 = PyStackRef_ZERO_BITS;
-            _tos_cache2 = PyStackRef_ZERO_BITS;
-            SET_CURRENT_CACHED_VALUES(1);
+            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 0c0256bcc3b753e918588043459766ba014737dc..4482bb08a132f95a26c7dd69d03071b52f0185af 100644 (file)
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -934,6 +934,8 @@
             _PyStackRef str_st;
             _PyStackRef sub_st;
             _PyStackRef res;
+            _PyStackRef s;
+            _PyStackRef i;
             // _GUARD_TOS_INT
             {
                 value = stack_pointer[-1];
@@ -983,17 +985,26 @@
                 assert(c < 128);
                 STAT_INC(BINARY_OP, hit);
                 PyObject *res_o = (PyObject*)&_Py_SINGLETON(strings).ascii[c];
-                PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);
-                stack_pointer += -2;
+                s = str_st;
+                i = sub_st;
+                res = PyStackRef_FromPyObjectBorrow(res_o);
+            }
+            // _POP_TOP_INT
+            {
+                value = i;
+                assert(PyLong_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
+                PyStackRef_CLOSE_SPECIALIZED(value, _PyLong_ExactDealloc);
+            }
+            // _POP_TOP
+            {
+                value = s;
+                stack_pointer[-2] = res;
+                stack_pointer += -1;
                 ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
                 _PyFrame_SetStackPointer(frame, stack_pointer);
-                PyStackRef_CLOSE(str_st);
+                PyStackRef_XCLOSE(value);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
-                res = PyStackRef_FromPyObjectBorrow(res_o);
             }
-            stack_pointer[0] = res;
-            stack_pointer += 1;
-            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             DISPATCH();
         }
 

diff --git a/Python/optimizer_bytecodes.c b/Python/optimizer_bytecodes.c
index 9578b56e70739302a96fcf210fdb51baf8ceb7dc..f36139db954cb8eb01acdf671ae6cffe5cd46af1 100644 (file)
--- a/Python/optimizer_bytecodes.c
+++ b/Python/optimizer_bytecodes.c
@@ -329,8 +329,10 @@ dummy_func(void) {
         ctx->done = true;
     }
 
-    op(_BINARY_OP_SUBSCR_STR_INT, (str_st, sub_st -- res)) {
+    op(_BINARY_OP_SUBSCR_STR_INT, (str_st, sub_st -- res, s, i)) {
         res = sym_new_type(ctx, &PyUnicode_Type);
+        s = str_st;
+        i = sub_st;
     }
 
     op(_BINARY_OP_SUBSCR_TUPLE_INT, (tuple_st, sub_st -- res)) {

diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index f047cd82d4c5e712274b7de9b8408785d9cccde8..69a2a32e4f15dfcbd11324c7e4e8a166fc92fc69 100644 (file)
--- a/Python/optimizer_cases.c.h
+++ b/Python/optimizer_cases.c.h
@@ -793,11 +793,21 @@
         }
 
         case _BINARY_OP_SUBSCR_STR_INT: {
+            JitOptRef sub_st;
+            JitOptRef str_st;
             JitOptRef res;
+            JitOptRef s;
+            JitOptRef i;
+            sub_st = stack_pointer[-1];
+            str_st = stack_pointer[-2];
             res = sym_new_type(ctx, &PyUnicode_Type);
-            CHECK_STACK_BOUNDS(-1);
+            s = str_st;
+            i = sub_st;
+            CHECK_STACK_BOUNDS(1);
             stack_pointer[-2] = res;
-            stack_pointer += -1;
+            stack_pointer[-1] = s;
+            stack_pointer[0] = i;
+            stack_pointer += 1;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             break;
         }