GH-131798: Add _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW (GH-134268)

Include/internal/pycore_uop_ids.h

@@ -265,73 +265,74 @@ extern "C" {
 #define _MONITOR_JUMP_BACKWARD 481
 #define _MONITOR_RESUME 482
 #define _NOP NOP
+#define _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW 483
 #define _POP_EXCEPT POP_EXCEPT
-#define _POP_JUMP_IF_FALSE 483
+#define _POP_JUMP_IF_FALSE 484
-#define _POP_JUMP_IF_TRUE 484
+#define _POP_JUMP_IF_TRUE 485
 #define _POP_TOP POP_TOP
-#define _POP_TOP_LOAD_CONST_INLINE 485
+#define _POP_TOP_LOAD_CONST_INLINE 486
-#define _POP_TOP_LOAD_CONST_INLINE_BORROW 486
+#define _POP_TOP_LOAD_CONST_INLINE_BORROW 487
-#define _POP_TWO_LOAD_CONST_INLINE_BORROW 487
+#define _POP_TWO_LOAD_CONST_INLINE_BORROW 488
 #define _PUSH_EXC_INFO PUSH_EXC_INFO
-#define _PUSH_FRAME 488
+#define _PUSH_FRAME 489
 #define _PUSH_NULL PUSH_NULL
-#define _PUSH_NULL_CONDITIONAL 489
+#define _PUSH_NULL_CONDITIONAL 490
-#define _PY_FRAME_GENERAL 490
+#define _PY_FRAME_GENERAL 491
-#define _PY_FRAME_KW 491
+#define _PY_FRAME_KW 492
-#define _QUICKEN_RESUME 492
+#define _QUICKEN_RESUME 493
-#define _REPLACE_WITH_TRUE 493
+#define _REPLACE_WITH_TRUE 494
 #define _RESUME_CHECK RESUME_CHECK
 #define _RETURN_GENERATOR RETURN_GENERATOR
 #define _RETURN_VALUE RETURN_VALUE
-#define _SAVE_RETURN_OFFSET 494
+#define _SAVE_RETURN_OFFSET 495
-#define _SEND 495
+#define _SEND 496
-#define _SEND_GEN_FRAME 496
+#define _SEND_GEN_FRAME 497
 #define _SETUP_ANNOTATIONS SETUP_ANNOTATIONS
 #define _SET_ADD SET_ADD
 #define _SET_FUNCTION_ATTRIBUTE SET_FUNCTION_ATTRIBUTE
 #define _SET_UPDATE SET_UPDATE
-#define _START_EXECUTOR 497
+#define _START_EXECUTOR 498
-#define _STORE_ATTR 498
+#define _STORE_ATTR 499
-#define _STORE_ATTR_INSTANCE_VALUE 499
+#define _STORE_ATTR_INSTANCE_VALUE 500
-#define _STORE_ATTR_SLOT 500
+#define _STORE_ATTR_SLOT 501
-#define _STORE_ATTR_WITH_HINT 501
+#define _STORE_ATTR_WITH_HINT 502
 #define _STORE_DEREF STORE_DEREF
-#define _STORE_FAST 502
+#define _STORE_FAST 503
-#define _STORE_FAST_0 503
+#define _STORE_FAST_0 504
-#define _STORE_FAST_1 504
+#define _STORE_FAST_1 505
-#define _STORE_FAST_2 505
+#define _STORE_FAST_2 506
-#define _STORE_FAST_3 506
+#define _STORE_FAST_3 507
-#define _STORE_FAST_4 507
+#define _STORE_FAST_4 508
-#define _STORE_FAST_5 508
+#define _STORE_FAST_5 509
-#define _STORE_FAST_6 509
+#define _STORE_FAST_6 510
-#define _STORE_FAST_7 510
+#define _STORE_FAST_7 511
 #define _STORE_FAST_LOAD_FAST STORE_FAST_LOAD_FAST
 #define _STORE_FAST_STORE_FAST STORE_FAST_STORE_FAST
 #define _STORE_GLOBAL STORE_GLOBAL
 #define _STORE_NAME STORE_NAME
-#define _STORE_SLICE 511
+#define _STORE_SLICE 512
-#define _STORE_SUBSCR 512
+#define _STORE_SUBSCR 513
-#define _STORE_SUBSCR_DICT 513
+#define _STORE_SUBSCR_DICT 514
-#define _STORE_SUBSCR_LIST_INT 514
+#define _STORE_SUBSCR_LIST_INT 515
 #define _SWAP SWAP
-#define _TIER2_RESUME_CHECK 515
+#define _TIER2_RESUME_CHECK 516
-#define _TO_BOOL 516
+#define _TO_BOOL 517
 #define _TO_BOOL_BOOL TO_BOOL_BOOL
 #define _TO_BOOL_INT TO_BOOL_INT
-#define _TO_BOOL_LIST 517
+#define _TO_BOOL_LIST 518
 #define _TO_BOOL_NONE TO_BOOL_NONE
-#define _TO_BOOL_STR 518
+#define _TO_BOOL_STR 519
 #define _UNARY_INVERT UNARY_INVERT
 #define _UNARY_NEGATIVE UNARY_NEGATIVE
 #define _UNARY_NOT UNARY_NOT
 #define _UNPACK_EX UNPACK_EX
-#define _UNPACK_SEQUENCE 519
+#define _UNPACK_SEQUENCE 520
-#define _UNPACK_SEQUENCE_LIST 520
+#define _UNPACK_SEQUENCE_LIST 521
-#define _UNPACK_SEQUENCE_TUPLE 521
+#define _UNPACK_SEQUENCE_TUPLE 522
-#define _UNPACK_SEQUENCE_TWO_TUPLE 522
+#define _UNPACK_SEQUENCE_TWO_TUPLE 523
 #define _WITH_EXCEPT_START WITH_EXCEPT_START
 #define _YIELD_VALUE YIELD_VALUE
-#define MAX_UOP_ID 522
+#define MAX_UOP_ID 523
 
 #ifdef __cplusplus
 }

Include/internal/pycore_uop_metadata.h

@@ -305,6 +305,7 @@ const uint16_t _PyUop_Flags[MAX_UOP_ID+1] = {
     [_LOAD_CONST_INLINE_BORROW] = HAS_PURE_FLAG,
     [_POP_TOP_LOAD_CONST_INLINE_BORROW] = HAS_ESCAPES_FLAG | HAS_PURE_FLAG,
     [_POP_TWO_LOAD_CONST_INLINE_BORROW] = HAS_ESCAPES_FLAG | HAS_PURE_FLAG,
+    [_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW] = HAS_ESCAPES_FLAG | HAS_PURE_FLAG,
     [_CHECK_FUNCTION] = HAS_DEOPT_FLAG,
     [_START_EXECUTOR] = 0,
     [_MAKE_WARM] = 0,
@@ -553,6 +554,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_ID+1] = {
     [_MAYBE_EXPAND_METHOD] = "_MAYBE_EXPAND_METHOD",
     [_MAYBE_EXPAND_METHOD_KW] = "_MAYBE_EXPAND_METHOD_KW",
     [_NOP] = "_NOP",
+    [_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW] = "_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW",
     [_POP_EXCEPT] = "_POP_EXCEPT",
     [_POP_TOP] = "_POP_TOP",
     [_POP_TOP_LOAD_CONST_INLINE] = "_POP_TOP_LOAD_CONST_INLINE",
@@ -1192,6 +1194,8 @@ int _PyUop_num_popped(int opcode, int oparg)
             return 1;
         case _POP_TWO_LOAD_CONST_INLINE_BORROW:
             return 2;
+        case _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW:
+            return 4;
         case _CHECK_FUNCTION:
             return 0;
         case _START_EXECUTOR:

Lib/test/test_capi/test_opt.py

@@ -1955,9 +1955,10 @@ def testfunc(n):
         self.assertEqual(res, TIER2_THRESHOLD)
         self.assertIsNotNone(ex)
         uops = get_opnames(ex)
-        self.assertIn("_CALL_ISINSTANCE", uops)
+        self.assertNotIn("_CALL_ISINSTANCE", uops)
         self.assertNotIn("_GUARD_THIRD_NULL", uops)
         self.assertNotIn("_GUARD_CALLABLE_ISINSTANCE", uops)
+        self.assertIn("_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW", uops)
 
     def test_call_list_append(self):
         def testfunc(n):
@@ -1987,9 +1988,10 @@ def testfunc(n):
         self.assertEqual(res, TIER2_THRESHOLD)
         self.assertIsNotNone(ex)
         uops = get_opnames(ex)
-        self.assertIn("_CALL_ISINSTANCE", uops)
+        self.assertNotIn("_CALL_ISINSTANCE", uops)
         self.assertNotIn("_TO_BOOL_BOOL", uops)
         self.assertNotIn("_GUARD_IS_TRUE_POP", uops)
+        self.assertIn("_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW", uops)
 
     def test_call_isinstance_is_false(self):
         def testfunc(n):
@@ -2004,9 +2006,10 @@ def testfunc(n):
         self.assertEqual(res, TIER2_THRESHOLD)
         self.assertIsNotNone(ex)
         uops = get_opnames(ex)
-        self.assertIn("_CALL_ISINSTANCE", uops)
+        self.assertNotIn("_CALL_ISINSTANCE", uops)
         self.assertNotIn("_TO_BOOL_BOOL", uops)
         self.assertNotIn("_GUARD_IS_FALSE_POP", uops)
+        self.assertIn("_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW", uops)
 
     def test_call_isinstance_subclass(self):
         def testfunc(n):
@@ -2021,9 +2024,10 @@ def testfunc(n):
         self.assertEqual(res, TIER2_THRESHOLD)
         self.assertIsNotNone(ex)
         uops = get_opnames(ex)
-        self.assertIn("_CALL_ISINSTANCE", uops)
+        self.assertNotIn("_CALL_ISINSTANCE", uops)
         self.assertNotIn("_TO_BOOL_BOOL", uops)
         self.assertNotIn("_GUARD_IS_TRUE_POP", uops)
+        self.assertIn("_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW", uops)
 
     def test_call_isinstance_unknown_object(self):
         def testfunc(n):

Misc/NEWS.d/next/Core_and_Builtins/2025-05-19-20-52-53.gh-issue-134268.HPKX1e.rst

@@ -0,0 +1,2 @@
+Add ``_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW`` and use it to further
+optimize ``CALL_ISINSTANCE``.

Python/bytecodes.c

@@ -5342,6 +5342,15 @@ dummy_func(
             value = PyStackRef_FromPyObjectImmortal(ptr);
         }
 
+        tier2 pure op(_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW, (ptr/4, callable, null, pop1, pop2 -- value)) {
+            PyStackRef_CLOSE(pop2);
+            PyStackRef_CLOSE(pop1);
+            (void)null; // Silence compiler warnings about unused variables
+            DEAD(null);
+            PyStackRef_CLOSE(callable);
+            value = PyStackRef_FromPyObjectImmortal(ptr);
+        }
+
         tier2 op(_CHECK_FUNCTION, (func_version/2 -- )) {
             assert(PyStackRef_FunctionCheck(frame->f_funcobj));
             PyFunctionObject *func = (PyFunctionObject *)PyStackRef_AsPyObjectBorrow(frame->f_funcobj);

Python/executor_cases.c.h

@@ -7103,6 +7103,40 @@
             break;
         }
 
+        case _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW: {
+            _PyStackRef pop2;
+            _PyStackRef pop1;
+            _PyStackRef null;
+            _PyStackRef callable;
+            _PyStackRef value;
+            pop2 = stack_pointer[-1];
+            pop1 = stack_pointer[-2];
+            null = stack_pointer[-3];
+            callable = stack_pointer[-4];
+            PyObject *ptr = (PyObject *)CURRENT_OPERAND0();
+            stack_pointer += -1;
+            assert(WITHIN_STACK_BOUNDS());
+            _PyFrame_SetStackPointer(frame, stack_pointer);
+            PyStackRef_CLOSE(pop2);
+            stack_pointer = _PyFrame_GetStackPointer(frame);
+            stack_pointer += -1;
+            assert(WITHIN_STACK_BOUNDS());
+            _PyFrame_SetStackPointer(frame, stack_pointer);
+            PyStackRef_CLOSE(pop1);
+            stack_pointer = _PyFrame_GetStackPointer(frame);
+            (void)null;
+            stack_pointer += -2;
+            assert(WITHIN_STACK_BOUNDS());
+            _PyFrame_SetStackPointer(frame, stack_pointer);
+            PyStackRef_CLOSE(callable);
+            stack_pointer = _PyFrame_GetStackPointer(frame);
+            value = PyStackRef_FromPyObjectImmortal(ptr);
+            stack_pointer[0] = value;
+            stack_pointer += 1;
+            assert(WITHIN_STACK_BOUNDS());
+            break;
+        }
+
         case _CHECK_FUNCTION: {
             uint32_t func_version = (uint32_t)CURRENT_OPERAND0();
             assert(PyStackRef_FunctionCheck(frame->f_funcobj));

Python/optimizer_bytecodes.c

@@ -550,6 +550,10 @@ dummy_func(void) {
         value = sym_new_const(ctx, ptr);
     }
 
+    op(_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW, (ptr/4, unused, unused, unused, unused -- value)) {
+        value = sym_new_const(ctx, ptr);
+    }
+
     op(_COPY, (bottom, unused[oparg-1] -- bottom, unused[oparg-1], top)) {
         assert(oparg > 0);
         top = bottom;
@@ -901,12 +905,12 @@ dummy_func(void) {
             // known types, meaning we can deduce either True or False
 
             // The below check is equivalent to PyObject_TypeCheck(inst, cls)
+            PyObject *out = Py_False;
             if (inst_type == cls_o || PyType_IsSubtype(inst_type, cls_o)) {
-                sym_set_const(res, Py_True);
+                out = Py_True;
-            }
-            else {
-                sym_set_const(res, Py_False);
             }
+            sym_set_const(res, out);
+            REPLACE_OP(this_instr, _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)out);
         }
     }
 

Python/optimizer_cases.c.h

@@ -2143,12 +2143,12 @@
             PyTypeObject *inst_type = sym_get_type(instance);
             PyTypeObject *cls_o = (PyTypeObject *)sym_get_const(ctx, cls);
             if (inst_type && cls_o && sym_matches_type(cls, &PyType_Type)) {
+                PyObject *out = Py_False;
                 if (inst_type == cls_o || PyType_IsSubtype(inst_type, cls_o)) {
-                    sym_set_const(res, Py_True);
+                    out = Py_True;
-                }
-                else {
-                    sym_set_const(res, Py_False);
                 }
+                sym_set_const(res, out);
+                REPLACE_OP(this_instr, _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)out);
             }
             stack_pointer[-4] = res;
             stack_pointer += -3;
@@ -2556,6 +2556,16 @@
             break;
         }
 
+        case _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW: {
+            JitOptSymbol *value;
+            PyObject *ptr = (PyObject *)this_instr->operand0;
+            value = sym_new_const(ctx, ptr);
+            stack_pointer[-4] = value;
+            stack_pointer += -3;
+            assert(WITHIN_STACK_BOUNDS());
+            break;
+        }
+
         case _CHECK_FUNCTION: {
             break;
         }