gh-134584: Eliminate redundant refcounting from _BINARY_OP_ADD_UNICODE

Lib/test/test_capi/test_opt.py

@@ -5,6 +5,8 @@
 import unittest
 import gc
 import os
+import random
+import string
 
 import _opcode
 
@@ -2362,6 +2364,23 @@ def testfunc(n):
         self.assertNotIn("_GUARD_TOS_INT", uops)
         self.assertNotIn("_GUARD_NOS_INT", uops)
 
+    def test_store_fast_pop_top_specialize_unicode(self):
+        def random_str(n):
+            return ''.join(random.choice(string.ascii_uppercase + string.digits) for _ in range(n))
+        def testfunc(n):
+            y = random_str(32)
+            for _ in range(n):
+                x = y + y # _POP_TOP
+                x = None # _POP_TOP_NOP
+
+        testfunc(TIER2_THRESHOLD)
+
+        ex = get_first_executor(testfunc)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+
+        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.

Lib/test/test_sys.py

@@ -1671,7 +1671,7 @@ def func():
             INTERPRETER_FRAME = '9PihcP'
         else:
             INTERPRETER_FRAME = '9PhcP'
-        check(x, size('3PiccPPP' + INTERPRETER_FRAME + 'P'))
+        check(x, size('3PiccPPP' + INTERPRETER_FRAME + 'PP'))
         # function
         def func(): pass
         check(func, size('16Pi'))
@@ -1688,7 +1688,7 @@ def bar(cls):
             check(bar, size('PP'))
         # generator
         def get_gen(): yield 1
-        check(get_gen(), size('6P4c' + INTERPRETER_FRAME + 'P'))
+        check(get_gen(), size('6P4c' + INTERPRETER_FRAME + 'PP'))
         # iterator
         check(iter('abc'), size('lP'))
         # callable-iterator

Python/bytecodes.c

@@ -344,6 +344,28 @@ dummy_func(
             PyStackRef_XCLOSE(value);
         }
 
+
+        op(_POP_TOP_NOP, (value --)) {
+            assert(PyStackRef_IsNull(value) || (!PyStackRef_RefcountOnObject(value)) ||
+                _Py_IsImmortal((PyStackRef_AsPyObjectBorrow(value))));
+            DEAD(value);
+        }
+
+        op(_POP_TOP_INT, (value --)) {
+            assert(PyLong_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
+            PyStackRef_CLOSE_SPECIALIZED(value, _PyLong_ExactDealloc);
+        }
+
+        op(_POP_TOP_FLOAT, (value --)) {
+            assert(PyFloat_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
+            PyStackRef_CLOSE_SPECIALIZED(value, _PyFloat_ExactDealloc);
+        }
+
+        op(_POP_TOP_UNICODE, (value --)) {
+            assert(PyUnicode_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
+            PyStackRef_CLOSE_SPECIALIZED(value, _PyUnicode_ExactDealloc);
+        }
+
         tier2 op(_POP_TWO, (nos, tos --)) {
             PyStackRef_CLOSE(tos);
             PyStackRef_CLOSE(nos);
@@ -751,23 +773,23 @@ dummy_func(
         macro(BINARY_OP_SUBTRACT_FLOAT) =
             _GUARD_TOS_FLOAT + _GUARD_NOS_FLOAT + unused/5 + _BINARY_OP_SUBTRACT_FLOAT;
 
-        pure op(_BINARY_OP_ADD_UNICODE, (left, right -- res)) {
+        pure op(_BINARY_OP_ADD_UNICODE, (left, right -- res, l, r)) {
             PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
             PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
             assert(PyUnicode_CheckExact(left_o));
             assert(PyUnicode_CheckExact(right_o));
 
             STAT_INC(BINARY_OP, hit);
             PyObject *res_o = PyUnicode_Concat(left_o, right_o);
-            PyStackRef_CLOSE_SPECIALIZED(right, _PyUnicode_ExactDealloc);
-            PyStackRef_CLOSE_SPECIALIZED(left, _PyUnicode_ExactDealloc);
             INPUTS_DEAD();
             ERROR_IF(res_o == NULL);
+            l = left;
+            r = right;
             res = PyStackRef_FromPyObjectSteal(res_o);
         }
 
         macro(BINARY_OP_ADD_UNICODE) =
-            _GUARD_TOS_UNICODE + _GUARD_NOS_UNICODE + unused/5 + _BINARY_OP_ADD_UNICODE;
+            _GUARD_TOS_UNICODE + _GUARD_NOS_UNICODE + unused/5 + _BINARY_OP_ADD_UNICODE + _POP_TOP_UNICODE + _POP_TOP_UNICODE;
 
         // This is a subtle one. It's a super-instruction for
         // BINARY_OP_ADD_UNICODE followed by STORE_FAST

Python/compile.c

@@ -1431,6 +1431,10 @@ optimize_and_assemble_code_unit(struct compiler_unit *u, PyObject *const_cache,
                                                  &optimized_instrs) < 0) {
         goto error;
     }
+    /* Reserve an extra word on the stack to ensure there is space for uops to
+       pass at least one item on the stack to a subsequent uop.
+    */
+    stackdepth++;
 
     /** Assembly **/
     co = _PyAssemble_MakeCodeObject(&u->u_metadata, const_cache, consts,