gh-144007: Eliminate redundant refcounting in the JIT for BINARY_OP (GH-144011)

diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h
index 80c11b753be7e6d66a8848a3bf817e56ec2912b9..95dee5d352670de7d422c878f40d797af619ec47 100644 (file)
--- a/Include/internal/pycore_opcode_metadata.h
+++ b/Include/internal/pycore_opcode_metadata.h
@@ -1343,7 +1343,7 @@ extern const struct opcode_macro_expansion _PyOpcode_macro_expansion[256];
 #ifdef NEED_OPCODE_METADATA
 const struct opcode_macro_expansion
 _PyOpcode_macro_expansion[256] = {
-    [BINARY_OP] = { .nuops = 1, .uops = { { _BINARY_OP, OPARG_SIMPLE, 4 } } },
+    [BINARY_OP] = { .nuops = 3, .uops = { { _BINARY_OP, OPARG_SIMPLE, 4 }, { _POP_TOP, OPARG_SIMPLE, 4 }, { _POP_TOP, OPARG_SIMPLE, 4 } } },
     [BINARY_OP_ADD_FLOAT] = { .nuops = 5, .uops = { { _GUARD_TOS_FLOAT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_FLOAT, OPARG_SIMPLE, 0 }, { _BINARY_OP_ADD_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 } } },
     [BINARY_OP_ADD_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_INT, OPARG_SIMPLE, 0 }, { _BINARY_OP_ADD_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 } } },
     [BINARY_OP_ADD_UNICODE] = { .nuops = 5, .uops = { { _GUARD_TOS_UNICODE, OPARG_SIMPLE, 0 }, { _GUARD_NOS_UNICODE, OPARG_SIMPLE, 0 }, { _BINARY_OP_ADD_UNICODE, OPARG_SIMPLE, 5 }, { _POP_TOP_UNICODE, OPARG_SIMPLE, 5 }, { _POP_TOP_UNICODE, OPARG_SIMPLE, 5 } } },

diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h
index ccdcc27f903cb9f9ed0f91f8d2ea8155ddac7ae4..8aa42933ec83fa52aab44209f165cbe9c1528ec5 100644 (file)
--- a/Include/internal/pycore_uop_ids.h
+++ b/Include/internal/pycore_uop_ids.h
@@ -379,7 +379,7 @@ extern "C" {
 #define _WITH_EXCEPT_START WITH_EXCEPT_START
 #define _YIELD_VALUE YIELD_VALUE
 #define MAX_UOP_ID 578
-#define _BINARY_OP_r21 579
+#define _BINARY_OP_r23 579
 #define _BINARY_OP_ADD_FLOAT_r03 580
 #define _BINARY_OP_ADD_FLOAT_r13 581
 #define _BINARY_OP_ADD_FLOAT_r23 582

diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h
index d51ed9c697ab8c1e9ddd460c125b7bc4622ecb6e..f9f22ba048194f5c6e27c3bb34e429db7304db9a 100644 (file)
--- a/Include/internal/pycore_uop_metadata.h
+++ b/Include/internal/pycore_uop_metadata.h
@@ -2931,7 +2931,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
         .entries = {
             { -1, -1, -1 },
             { -1, -1, -1 },
-            { 1, 2, _BINARY_OP_r21 },
+            { 3, 2, _BINARY_OP_r23 },
             { -1, -1, -1 },
         },
     },
@@ -4031,7 +4031,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
     [_COPY_3_r23] = _COPY_3,
     [_COPY_3_r33] = _COPY_3,
     [_COPY_r01] = _COPY,
-    [_BINARY_OP_r21] = _BINARY_OP,
+    [_BINARY_OP_r23] = _BINARY_OP,
     [_SWAP_2_r02] = _SWAP_2,
     [_SWAP_2_r12] = _SWAP_2,
     [_SWAP_2_r22] = _SWAP_2,
@@ -4225,7 +4225,7 @@ const uint16_t _PyUop_SpillsAndReloads[4][4] = {
 
 const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
     [_BINARY_OP] = "_BINARY_OP",
-    [_BINARY_OP_r21] = "_BINARY_OP_r21",
+    [_BINARY_OP_r23] = "_BINARY_OP_r23",
     [_BINARY_OP_ADD_FLOAT] = "_BINARY_OP_ADD_FLOAT",
     [_BINARY_OP_ADD_FLOAT_r03] = "_BINARY_OP_ADD_FLOAT_r03",
     [_BINARY_OP_ADD_FLOAT_r13] = "_BINARY_OP_ADD_FLOAT_r13",

diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py
index 5a0a379e31a962b6c740d903fa8973959e988ac8..da0d4078c9f5ed9e889512ed80795e58647c0b91 100644 (file)
--- a/Lib/test/test_capi/test_opt.py
+++ b/Lib/test/test_capi/test_opt.py
@@ -2897,6 +2897,29 @@ class TestUopsOptimization(unittest.TestCase):
         self.assertIn("_POP_TOP_NOP", uops)
         self.assertLessEqual(count_ops(ex, "_POP_TOP"), 2)
 
+    def test_binary_op_refcount_elimination(self):
+        class CustomAdder:
+            def __init__(self, val):
+                self.val = val
+            def __add__(self, other):
+                return CustomAdder(self.val + other.val)
+
+        def testfunc(n):
+            a = CustomAdder(1)
+            b = CustomAdder(2)
+            res = None
+            for _ in range(n):
+                res = a + b
+            return res.val if res else 0
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, 3)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertIn("_BINARY_OP", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+        self.assertLessEqual(count_ops(ex, "_POP_TOP"), 2)
+
     def test_binary_op_extend_float_long_add_refcount_elimination(self):
         def testfunc(n):
             a = 1.5

diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2026-01-19-01-56-44.gh-issue-144007.1xjdBf.rst b/Misc/NEWS.d/next/Core_and_Builtins/2026-01-19-01-56-44.gh-issue-144007.1xjdBf.rst
new file mode 100644 (file)
index 0000000..26db86f
--- /dev/null
+++ b/Misc/NEWS.d/next/Core_and_Builtins/2026-01-19-01-56-44.gh-issue-144007.1xjdBf.rst
@@ -0,0 +1 @@
+Eliminate redundant refcounting in the JIT for ``BINARY_OP``.

diff --git a/Modules/_testinternalcapi/test_cases.c.h b/Modules/_testinternalcapi/test_cases.c.h
index a7d589dbe7b274dead743df5326b07990f664132..3c9eb9194366e8ce092c6c0e326ebcfcdc62144e 100644 (file)
--- a/Modules/_testinternalcapi/test_cases.c.h
+++ b/Modules/_testinternalcapi/test_cases.c.h
@@ -32,6 +32,9 @@
             _PyStackRef lhs;
             _PyStackRef rhs;
             _PyStackRef res;
+            _PyStackRef l;
+            _PyStackRef r;
+            _PyStackRef value;
             // _SPECIALIZE_BINARY_OP
             {
                 rhs = stack_pointer[-1];
@@ -65,18 +68,26 @@
                     JUMP_TO_LABEL(error);
                 }
                 res = PyStackRef_FromPyObjectSteal(res_o);
+                l = lhs;
+                r = rhs;
+            }
+            // _POP_TOP
+            {
+                value = r;
+                stack_pointer[-2] = res;
+                stack_pointer[-1] = l;
                 _PyFrame_SetStackPointer(frame, stack_pointer);
-                _PyStackRef tmp = lhs;
-                lhs = res;
-                stack_pointer[-2] = lhs;
-                PyStackRef_CLOSE(tmp);
-                tmp = rhs;
-                rhs = PyStackRef_NULL;
-                stack_pointer[-1] = rhs;
-                PyStackRef_CLOSE(tmp);
+                PyStackRef_XCLOSE(value);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
+            }
+            // _POP_TOP
+            {
+                value = l;
                 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/bytecodes.c b/Python/bytecodes.c
index 36df991e84dc2c44c2addefc811929439d219d26..bec03119993d79c56acd963b178869a06578c44e 100644 (file)
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -5119,7 +5119,7 @@ dummy_func(
             assert(oparg <= NB_OPARG_LAST);
         }
 
-        op(_BINARY_OP, (lhs, rhs -- res)) {
+        op(_BINARY_OP, (lhs, rhs -- res, l, r)) {
             PyObject *lhs_o = PyStackRef_AsPyObjectBorrow(lhs);
             PyObject *rhs_o = PyStackRef_AsPyObjectBorrow(rhs);
 
@@ -5129,10 +5129,13 @@ dummy_func(
                 ERROR_NO_POP();
             }
             res = PyStackRef_FromPyObjectSteal(res_o);
-            DECREF_INPUTS();
+            l = lhs;
+            r = rhs;
+            DEAD(lhs);
+            DEAD(rhs);
         }
 
-        macro(BINARY_OP) = _SPECIALIZE_BINARY_OP + unused/4 + _BINARY_OP;
+        macro(BINARY_OP) = _SPECIALIZE_BINARY_OP + unused/4 + _BINARY_OP + POP_TOP + POP_TOP;
 
         pure replicate(2:4) inst(SWAP, (bottom, unused[oparg-2], top --
                     bottom, unused[oparg-2], top)) {

diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 9098bd219ed5061b407937ffa6245a42f7b6d22d..f51466abea9266c382814ca57d35ec57014781b9 100644 (file)
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -16594,12 +16594,14 @@
             break;
         }
 
-        case _BINARY_OP_r21: {
+        case _BINARY_OP_r23: {
             CHECK_CURRENT_CACHED_VALUES(2);
             assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());
             _PyStackRef rhs;
             _PyStackRef lhs;
             _PyStackRef res;
+            _PyStackRef l;
+            _PyStackRef r;
             _PyStackRef _stack_item_0 = _tos_cache0;
             _PyStackRef _stack_item_1 = _tos_cache1;
             oparg = CURRENT_OPARG();
@@ -16620,23 +16622,13 @@
                 JUMP_TO_ERROR();
             }
             res = PyStackRef_FromPyObjectSteal(res_o);
-            _PyFrame_SetStackPointer(frame, stack_pointer);
-            _PyStackRef tmp = lhs;
-            lhs = res;
-            stack_pointer[-2] = lhs;
-            PyStackRef_CLOSE(tmp);
-            tmp = rhs;
-            rhs = PyStackRef_NULL;
-            stack_pointer[-1] = rhs;
-            PyStackRef_CLOSE(tmp);
-            stack_pointer = _PyFrame_GetStackPointer(frame);
-            stack_pointer += -1;
-            ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
+            l = lhs;
+            r = rhs;
+            _tos_cache2 = r;
+            _tos_cache1 = l;
             _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 9df6b2f70f96df4b8e0668ffcbcd4fdf12130fdb..ae9e1e5421b87b67caa8ec49140dd141d46e3461 100644 (file)
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -32,6 +32,9 @@
             _PyStackRef lhs;
             _PyStackRef rhs;
             _PyStackRef res;
+            _PyStackRef l;
+            _PyStackRef r;
+            _PyStackRef value;
             // _SPECIALIZE_BINARY_OP
             {
                 rhs = stack_pointer[-1];
@@ -65,18 +68,26 @@
                     JUMP_TO_LABEL(error);
                 }
                 res = PyStackRef_FromPyObjectSteal(res_o);
+                l = lhs;
+                r = rhs;
+            }
+            // _POP_TOP
+            {
+                value = r;
+                stack_pointer[-2] = res;
+                stack_pointer[-1] = l;
                 _PyFrame_SetStackPointer(frame, stack_pointer);
-                _PyStackRef tmp = lhs;
-                lhs = res;
-                stack_pointer[-2] = lhs;
-                PyStackRef_CLOSE(tmp);
-                tmp = rhs;
-                rhs = PyStackRef_NULL;
-                stack_pointer[-1] = rhs;
-                PyStackRef_CLOSE(tmp);
+                PyStackRef_XCLOSE(value);
                 stack_pointer = _PyFrame_GetStackPointer(frame);
+            }
+            // _POP_TOP
+            {
+                value = l;
                 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 de7a2313f9220a8af7a2872661566963b3be3065..27b974f372df99ebb7f1921528a7ffeff9cd46a2 100644 (file)
--- a/Python/optimizer_bytecodes.c
+++ b/Python/optimizer_bytecodes.c
@@ -211,7 +211,9 @@ dummy_func(void) {
         sym_set_type(left, &PyFloat_Type);
     }
 
-    op(_BINARY_OP, (lhs, rhs -- res)) {
+    op(_BINARY_OP, (lhs, rhs -- res, l, r)) {
+        l = lhs;
+        r = rhs;
         REPLACE_OPCODE_IF_EVALUATES_PURE(lhs, rhs, res);
         bool lhs_int = sym_matches_type(lhs, &PyLong_Type);
         bool rhs_int = sym_matches_type(rhs, &PyLong_Type);

diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index 5fca69462669b5260b09573c62ae3583b68353b0..3b25533e07f743e768c6722f44518e02619f1104 100644 (file)
--- a/Python/optimizer_cases.c.h
+++ b/Python/optimizer_cases.c.h
@@ -3616,8 +3616,12 @@
             JitOptRef rhs;
             JitOptRef lhs;
             JitOptRef res;
+            JitOptRef l;
+            JitOptRef r;
             rhs = stack_pointer[-1];
             lhs = stack_pointer[-2];
+            l = lhs;
+            r = rhs;
             if (
                 sym_is_safe_const(ctx, lhs) &&
                 sym_is_safe_const(ctx, rhs)
@@ -3627,6 +3631,8 @@
                 _PyStackRef lhs = sym_get_const_as_stackref(ctx, lhs_sym);
                 _PyStackRef rhs = sym_get_const_as_stackref(ctx, rhs_sym);
                 _PyStackRef res_stackref;
+                _PyStackRef l_stackref;
+                _PyStackRef r_stackref;
                 /* Start of uop copied from bytecodes for constant evaluation */
                 PyObject *lhs_o = PyStackRef_AsPyObjectBorrow(lhs);
                 PyObject *rhs_o = PyStackRef_AsPyObjectBorrow(rhs);
@@ -3636,18 +3642,24 @@
                     JUMP_TO_LABEL(error);
                 }
                 res_stackref = PyStackRef_FromPyObjectSteal(res_o);
+                l_stackref = lhs;
+                r_stackref = rhs;
                 /* End of uop copied from bytecodes for constant evaluation */
+                (void)l_stackref;
+                (void)r_stackref;
                 res = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(res_stackref));
                 if (sym_is_const(ctx, res)) {
                     PyObject *result = sym_get_const(ctx, res);
                     if (_Py_IsImmortal(result)) {
-                        // Replace with _POP_TWO_LOAD_CONST_INLINE_BORROW since we have two inputs and an immortal result
-                        ADD_OP(_POP_TWO_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)result);
+                        // Replace with _INSERT_2_LOAD_CONST_INLINE_BORROW since we have two inputs and an immortal result
+                        ADD_OP(_INSERT_2_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)result);
                     }
                 }
-                CHECK_STACK_BOUNDS(-1);
+                CHECK_STACK_BOUNDS(1);
                 stack_pointer[-2] = res;
-                stack_pointer += -1;
+                stack_pointer[-1] = l;
+                stack_pointer[0] = r;
+                stack_pointer += 1;
                 ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
                 break;
             }
@@ -3684,9 +3696,11 @@
             else {
                 res = sym_new_type(ctx, &PyFloat_Type);
             }
-            CHECK_STACK_BOUNDS(-1);
+            CHECK_STACK_BOUNDS(1);
             stack_pointer[-2] = res;
-            stack_pointer += -1;
+            stack_pointer[-1] = l;
+            stack_pointer[0] = r;
+            stack_pointer += 1;
             ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
             break;
         }