GH-111848: Tidy up tier 2 handling of FOR_ITER specialization by using DEOPT_IF inste...

diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h
index 5cfe8b1d00325a0239bfa630d08d0abbc672793d..bd5a401adf11bf5bac078933e004fab2ac17a4c1 100644 (file)
--- a/Include/internal/pycore_opcode_metadata.h
+++ b/Include/internal/pycore_opcode_metadata.h
@@ -81,15 +81,15 @@
 #define _FOR_ITER 353
 #define _ITER_CHECK_LIST 354
 #define _ITER_JUMP_LIST 355
-#define _IS_ITER_EXHAUSTED_LIST 356
+#define _GUARD_NOT_EXHAUSTED_LIST 356
 #define _ITER_NEXT_LIST 357
 #define _ITER_CHECK_TUPLE 358
 #define _ITER_JUMP_TUPLE 359
-#define _IS_ITER_EXHAUSTED_TUPLE 360
+#define _GUARD_NOT_EXHAUSTED_TUPLE 360
 #define _ITER_NEXT_TUPLE 361
 #define _ITER_CHECK_RANGE 362
 #define _ITER_JUMP_RANGE 363
-#define _IS_ITER_EXHAUSTED_RANGE 364
+#define _GUARD_NOT_EXHAUSTED_RANGE 364
 #define _ITER_NEXT_RANGE 365
 #define _GUARD_DORV_VALUES_INST_ATTR_FROM_DICT 366
 #define _GUARD_KEYS_VERSION 367
@@ -542,7 +542,7 @@ int _PyOpcode_num_popped(int opcode, int oparg, bool jump)  {
             return 1;
         case _ITER_JUMP_LIST:
             return 1;
-        case _IS_ITER_EXHAUSTED_LIST:
+        case _GUARD_NOT_EXHAUSTED_LIST:
             return 1;
         case _ITER_NEXT_LIST:
             return 1;
@@ -552,7 +552,7 @@ int _PyOpcode_num_popped(int opcode, int oparg, bool jump)  {
             return 1;
         case _ITER_JUMP_TUPLE:
             return 1;
-        case _IS_ITER_EXHAUSTED_TUPLE:
+        case _GUARD_NOT_EXHAUSTED_TUPLE:
             return 1;
         case _ITER_NEXT_TUPLE:
             return 1;
@@ -562,7 +562,7 @@ int _PyOpcode_num_popped(int opcode, int oparg, bool jump)  {
             return 1;
         case _ITER_JUMP_RANGE:
             return 1;
-        case _IS_ITER_EXHAUSTED_RANGE:
+        case _GUARD_NOT_EXHAUSTED_RANGE:
             return 1;
         case _ITER_NEXT_RANGE:
             return 1;
@@ -1170,8 +1170,8 @@ int _PyOpcode_num_pushed(int opcode, int oparg, bool jump)  {
             return 1;
         case _ITER_JUMP_LIST:
             return 1;
-        case _IS_ITER_EXHAUSTED_LIST:
-            return 2;
+        case _GUARD_NOT_EXHAUSTED_LIST:
+            return 1;
         case _ITER_NEXT_LIST:
             return 2;
         case FOR_ITER_LIST:
@@ -1180,8 +1180,8 @@ int _PyOpcode_num_pushed(int opcode, int oparg, bool jump)  {
             return 1;
         case _ITER_JUMP_TUPLE:
             return 1;
-        case _IS_ITER_EXHAUSTED_TUPLE:
-            return 2;
+        case _GUARD_NOT_EXHAUSTED_TUPLE:
+            return 1;
         case _ITER_NEXT_TUPLE:
             return 2;
         case FOR_ITER_TUPLE:
@@ -1190,8 +1190,8 @@ int _PyOpcode_num_pushed(int opcode, int oparg, bool jump)  {
             return 1;
         case _ITER_JUMP_RANGE:
             return 1;
-        case _IS_ITER_EXHAUSTED_RANGE:
-            return 2;
+        case _GUARD_NOT_EXHAUSTED_RANGE:
+            return 1;
         case _ITER_NEXT_RANGE:
             return 2;
         case FOR_ITER_RANGE:
@@ -1651,17 +1651,17 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[OPCODE_METADATA_SIZE] = {
     [INSTRUMENTED_FOR_ITER] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_ERROR_FLAG },
     [_ITER_CHECK_LIST] = { true, INSTR_FMT_IX, HAS_DEOPT_FLAG },
     [_ITER_JUMP_LIST] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_JUMP_FLAG },
-    [_IS_ITER_EXHAUSTED_LIST] = { true, INSTR_FMT_IX, 0 },
+    [_GUARD_NOT_EXHAUSTED_LIST] = { true, INSTR_FMT_IX, HAS_DEOPT_FLAG },
     [_ITER_NEXT_LIST] = { true, INSTR_FMT_IX, 0 },
     [FOR_ITER_LIST] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_JUMP_FLAG | HAS_DEOPT_FLAG },
     [_ITER_CHECK_TUPLE] = { true, INSTR_FMT_IX, HAS_DEOPT_FLAG },
     [_ITER_JUMP_TUPLE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_JUMP_FLAG },
-    [_IS_ITER_EXHAUSTED_TUPLE] = { true, INSTR_FMT_IX, 0 },
+    [_GUARD_NOT_EXHAUSTED_TUPLE] = { true, INSTR_FMT_IX, HAS_DEOPT_FLAG },
     [_ITER_NEXT_TUPLE] = { true, INSTR_FMT_IX, 0 },
     [FOR_ITER_TUPLE] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_JUMP_FLAG | HAS_DEOPT_FLAG },
     [_ITER_CHECK_RANGE] = { true, INSTR_FMT_IX, HAS_DEOPT_FLAG },
     [_ITER_JUMP_RANGE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_JUMP_FLAG },
-    [_IS_ITER_EXHAUSTED_RANGE] = { true, INSTR_FMT_IX, 0 },
+    [_GUARD_NOT_EXHAUSTED_RANGE] = { true, INSTR_FMT_IX, HAS_DEOPT_FLAG },
     [_ITER_NEXT_RANGE] = { true, INSTR_FMT_IX, HAS_ERROR_FLAG },
     [FOR_ITER_RANGE] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_JUMP_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG },
     [FOR_ITER_GEN] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_DEOPT_FLAG },
@@ -1870,6 +1870,9 @@ const struct opcode_macro_expansion _PyOpcode_macro_expansion[OPCODE_MACRO_EXPAN
     [MATCH_KEYS] = { .nuops = 1, .uops = { { MATCH_KEYS, 0, 0 } } },
     [GET_ITER] = { .nuops = 1, .uops = { { GET_ITER, 0, 0 } } },
     [GET_YIELD_FROM_ITER] = { .nuops = 1, .uops = { { GET_YIELD_FROM_ITER, 0, 0 } } },
+    [FOR_ITER_LIST] = { .nuops = 3, .uops = { { _ITER_CHECK_LIST, 0, 0 }, { _ITER_JUMP_LIST, 0, 0 }, { _ITER_NEXT_LIST, 0, 0 } } },
+    [FOR_ITER_TUPLE] = { .nuops = 3, .uops = { { _ITER_CHECK_TUPLE, 0, 0 }, { _ITER_JUMP_TUPLE, 0, 0 }, { _ITER_NEXT_TUPLE, 0, 0 } } },
+    [FOR_ITER_RANGE] = { .nuops = 3, .uops = { { _ITER_CHECK_RANGE, 0, 0 }, { _ITER_JUMP_RANGE, 0, 0 }, { _ITER_NEXT_RANGE, 0, 0 } } },
     [BEFORE_ASYNC_WITH] = { .nuops = 1, .uops = { { BEFORE_ASYNC_WITH, 0, 0 } } },
     [BEFORE_WITH] = { .nuops = 1, .uops = { { BEFORE_WITH, 0, 0 } } },
     [WITH_EXCEPT_START] = { .nuops = 1, .uops = { { WITH_EXCEPT_START, 0, 0 } } },
@@ -1966,15 +1969,15 @@ const char * const _PyOpcode_uop_name[OPCODE_UOP_NAME_SIZE] = {
     [_FOR_ITER] = "_FOR_ITER",
     [_ITER_CHECK_LIST] = "_ITER_CHECK_LIST",
     [_ITER_JUMP_LIST] = "_ITER_JUMP_LIST",
-    [_IS_ITER_EXHAUSTED_LIST] = "_IS_ITER_EXHAUSTED_LIST",
+    [_GUARD_NOT_EXHAUSTED_LIST] = "_GUARD_NOT_EXHAUSTED_LIST",
     [_ITER_NEXT_LIST] = "_ITER_NEXT_LIST",
     [_ITER_CHECK_TUPLE] = "_ITER_CHECK_TUPLE",
     [_ITER_JUMP_TUPLE] = "_ITER_JUMP_TUPLE",
-    [_IS_ITER_EXHAUSTED_TUPLE] = "_IS_ITER_EXHAUSTED_TUPLE",
+    [_GUARD_NOT_EXHAUSTED_TUPLE] = "_GUARD_NOT_EXHAUSTED_TUPLE",
     [_ITER_NEXT_TUPLE] = "_ITER_NEXT_TUPLE",
     [_ITER_CHECK_RANGE] = "_ITER_CHECK_RANGE",
     [_ITER_JUMP_RANGE] = "_ITER_JUMP_RANGE",
-    [_IS_ITER_EXHAUSTED_RANGE] = "_IS_ITER_EXHAUSTED_RANGE",
+    [_GUARD_NOT_EXHAUSTED_RANGE] = "_GUARD_NOT_EXHAUSTED_RANGE",
     [_ITER_NEXT_RANGE] = "_ITER_NEXT_RANGE",
     [_GUARD_DORV_VALUES_INST_ATTR_FROM_DICT] = "_GUARD_DORV_VALUES_INST_ATTR_FROM_DICT",
     [_GUARD_KEYS_VERSION] = "_GUARD_KEYS_VERSION",

diff --git a/Lib/test/test_capi/test_misc.py b/Lib/test/test_capi/test_misc.py
index 11dec031d3d79f0eea1c3e1d7284bbee04dba47b..d526bbf62dda0b63bed4b700a17beaba9cae7e36 100644 (file)
--- a/Lib/test/test_capi/test_misc.py
+++ b/Lib/test/test_capi/test_misc.py
@@ -2712,7 +2712,7 @@ class TestUops(unittest.TestCase):
         # for i, (opname, oparg) in enumerate(ex):
         #     print(f"{i:4d}: {opname:<20s} {oparg:3d}")
         uops = {opname for opname, _, _ in ex}
-        self.assertIn("_IS_ITER_EXHAUSTED_RANGE", uops)
+        self.assertIn("_GUARD_NOT_EXHAUSTED_RANGE", uops)
         # Verification that the jump goes past END_FOR
         # is done by manual inspection of the output
 
@@ -2734,7 +2734,7 @@ class TestUops(unittest.TestCase):
         # for i, (opname, oparg) in enumerate(ex):
         #     print(f"{i:4d}: {opname:<20s} {oparg:3d}")
         uops = {opname for opname, _, _ in ex}
-        self.assertIn("_IS_ITER_EXHAUSTED_LIST", uops)
+        self.assertIn("_GUARD_NOT_EXHAUSTED_LIST", uops)
         # Verification that the jump goes past END_FOR
         # is done by manual inspection of the output
 
@@ -2756,7 +2756,7 @@ class TestUops(unittest.TestCase):
         # for i, (opname, oparg) in enumerate(ex):
         #     print(f"{i:4d}: {opname:<20s} {oparg:3d}")
         uops = {opname for opname, _, _ in ex}
-        self.assertIn("_IS_ITER_EXHAUSTED_TUPLE", uops)
+        self.assertIn("_GUARD_NOT_EXHAUSTED_TUPLE", uops)
         # Verification that the jump goes past END_FOR
         # is done by manual inspection of the output
 

diff --git a/Python/abstract_interp_cases.c.h b/Python/abstract_interp_cases.c.h
index c6ebea024c3cd219f6087d43c7bc1f30ba98bd38..384a11212b3a2c4154f7d7bded81871de8456ec4 100644 (file)
--- a/Python/abstract_interp_cases.c.h
+++ b/Python/abstract_interp_cases.c.h
@@ -628,9 +628,7 @@
             break;
         }
 
-        case _IS_ITER_EXHAUSTED_LIST: {
-            STACK_GROW(1);
-            PARTITIONNODE_OVERWRITE((_Py_PARTITIONNODE_t *)PARTITIONNODE_NULLROOT, PEEK(-(-1)), true);
+        case _GUARD_NOT_EXHAUSTED_LIST: {
             break;
         }
 
@@ -644,9 +642,7 @@
             break;
         }
 
-        case _IS_ITER_EXHAUSTED_TUPLE: {
-            STACK_GROW(1);
-            PARTITIONNODE_OVERWRITE((_Py_PARTITIONNODE_t *)PARTITIONNODE_NULLROOT, PEEK(-(-1)), true);
+        case _GUARD_NOT_EXHAUSTED_TUPLE: {
             break;
         }
 
@@ -660,9 +656,7 @@
             break;
         }
 
-        case _IS_ITER_EXHAUSTED_RANGE: {
-            STACK_GROW(1);
-            PARTITIONNODE_OVERWRITE((_Py_PARTITIONNODE_t *)PARTITIONNODE_NULLROOT, PEEK(-(-1)), true);
+        case _GUARD_NOT_EXHAUSTED_RANGE: {
             break;
         }
 

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 8e1d318f195936a4ceef64e996e34c1afcb18d7c..f879ea5850aede31825d53541008d4ca33ba97cd 100644 (file)
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -2580,7 +2580,7 @@ dummy_func(
             DEOPT_IF(Py_TYPE(iter) != &PyListIter_Type);
         }
 
-        op(_ITER_JUMP_LIST, (iter -- iter)) {
+        replaced op(_ITER_JUMP_LIST, (iter -- iter)) {
             _PyListIterObject *it = (_PyListIterObject *)iter;
             assert(Py_TYPE(iter) == &PyListIter_Type);
             STAT_INC(FOR_ITER, hit);
@@ -2599,21 +2599,12 @@ dummy_func(
         }
 
         // Only used by Tier 2
-        op(_IS_ITER_EXHAUSTED_LIST, (iter -- iter, exhausted)) {
+        op(_GUARD_NOT_EXHAUSTED_LIST, (iter -- iter)) {
             _PyListIterObject *it = (_PyListIterObject *)iter;
             assert(Py_TYPE(iter) == &PyListIter_Type);
             PyListObject *seq = it->it_seq;
-            if (seq == NULL) {
-                exhausted = Py_True;
-            }
-            else if (it->it_index >= PyList_GET_SIZE(seq)) {
-                Py_DECREF(seq);
-                it->it_seq = NULL;
-                exhausted = Py_True;
-            }
-            else {
-                exhausted = Py_False;
-            }
+            DEOPT_IF(seq == NULL);
+            DEOPT_IF(it->it_index >= PyList_GET_SIZE(seq));
         }
 
         op(_ITER_NEXT_LIST, (iter -- iter, next)) {
@@ -2635,7 +2626,7 @@ dummy_func(
             DEOPT_IF(Py_TYPE(iter) != &PyTupleIter_Type);
         }
 
-        op(_ITER_JUMP_TUPLE, (iter -- iter)) {
+        replaced op(_ITER_JUMP_TUPLE, (iter -- iter)) {
             _PyTupleIterObject *it = (_PyTupleIterObject *)iter;
             assert(Py_TYPE(iter) == &PyTupleIter_Type);
             STAT_INC(FOR_ITER, hit);
@@ -2654,21 +2645,12 @@ dummy_func(
         }
 
         // Only used by Tier 2
-        op(_IS_ITER_EXHAUSTED_TUPLE, (iter -- iter, exhausted)) {
+        op(_GUARD_NOT_EXHAUSTED_TUPLE, (iter -- iter)) {
             _PyTupleIterObject *it = (_PyTupleIterObject *)iter;
             assert(Py_TYPE(iter) == &PyTupleIter_Type);
             PyTupleObject *seq = it->it_seq;
-            if (seq == NULL) {
-                exhausted = Py_True;
-            }
-            else if (it->it_index >= PyTuple_GET_SIZE(seq)) {
-                Py_DECREF(seq);
-                it->it_seq = NULL;
-                exhausted = Py_True;
-            }
-            else {
-                exhausted = Py_False;
-            }
+            DEOPT_IF(seq == NULL);
+            DEOPT_IF(it->it_index >= PyTuple_GET_SIZE(seq));
         }
 
         op(_ITER_NEXT_TUPLE, (iter -- iter, next)) {
@@ -2691,7 +2673,7 @@ dummy_func(
             DEOPT_IF(Py_TYPE(r) != &PyRangeIter_Type);
         }
 
-        op(_ITER_JUMP_RANGE, (iter -- iter)) {
+        replaced op(_ITER_JUMP_RANGE, (iter -- iter)) {
             _PyRangeIterObject *r = (_PyRangeIterObject *)iter;
             assert(Py_TYPE(r) == &PyRangeIter_Type);
             STAT_INC(FOR_ITER, hit);
@@ -2705,10 +2687,10 @@ dummy_func(
         }
 
         // Only used by Tier 2
-        op(_IS_ITER_EXHAUSTED_RANGE, (iter -- iter, exhausted)) {
+        op(_GUARD_NOT_EXHAUSTED_RANGE, (iter -- iter)) {
             _PyRangeIterObject *r = (_PyRangeIterObject *)iter;
             assert(Py_TYPE(r) == &PyRangeIter_Type);
-            exhausted = r->len <= 0 ? Py_True : Py_False;
+            DEOPT_IF(r->len <= 0);
         }
 
         op(_ITER_NEXT_RANGE, (iter -- iter, next)) {

diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index d9e9ad133de5089cb1d5b6b59668d428088cf09b..d94a7cc4be0052e4906b8479a80e31ba9547500a 100644 (file)
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -2138,26 +2138,14 @@
             break;
         }
 
-        case _IS_ITER_EXHAUSTED_LIST: {
+        case _GUARD_NOT_EXHAUSTED_LIST: {
             PyObject *iter;
-            PyObject *exhausted;
             iter = stack_pointer[-1];
             _PyListIterObject *it = (_PyListIterObject *)iter;
             assert(Py_TYPE(iter) == &PyListIter_Type);
             PyListObject *seq = it->it_seq;
-            if (seq == NULL) {
-                exhausted = Py_True;
-            }
-            else if (it->it_index >= PyList_GET_SIZE(seq)) {
-                Py_DECREF(seq);
-                it->it_seq = NULL;
-                exhausted = Py_True;
-            }
-            else {
-                exhausted = Py_False;
-            }
-            STACK_GROW(1);
-            stack_pointer[-1] = exhausted;
+            DEOPT_IF(seq == NULL, _GUARD_NOT_EXHAUSTED_LIST);
+            DEOPT_IF(it->it_index >= PyList_GET_SIZE(seq), _GUARD_NOT_EXHAUSTED_LIST);
             break;
         }
 
@@ -2183,26 +2171,14 @@
             break;
         }
 
-        case _IS_ITER_EXHAUSTED_TUPLE: {
+        case _GUARD_NOT_EXHAUSTED_TUPLE: {
             PyObject *iter;
-            PyObject *exhausted;
             iter = stack_pointer[-1];
             _PyTupleIterObject *it = (_PyTupleIterObject *)iter;
             assert(Py_TYPE(iter) == &PyTupleIter_Type);
             PyTupleObject *seq = it->it_seq;
-            if (seq == NULL) {
-                exhausted = Py_True;
-            }
-            else if (it->it_index >= PyTuple_GET_SIZE(seq)) {
-                Py_DECREF(seq);
-                it->it_seq = NULL;
-                exhausted = Py_True;
-            }
-            else {
-                exhausted = Py_False;
-            }
-            STACK_GROW(1);
-            stack_pointer[-1] = exhausted;
+            DEOPT_IF(seq == NULL, _GUARD_NOT_EXHAUSTED_TUPLE);
+            DEOPT_IF(it->it_index >= PyTuple_GET_SIZE(seq), _GUARD_NOT_EXHAUSTED_TUPLE);
             break;
         }
 
@@ -2229,15 +2205,12 @@
             break;
         }
 
-        case _IS_ITER_EXHAUSTED_RANGE: {
+        case _GUARD_NOT_EXHAUSTED_RANGE: {
             PyObject *iter;
-            PyObject *exhausted;
             iter = stack_pointer[-1];
             _PyRangeIterObject *r = (_PyRangeIterObject *)iter;
             assert(Py_TYPE(r) == &PyRangeIter_Type);
-            exhausted = r->len <= 0 ? Py_True : Py_False;
-            STACK_GROW(1);
-            stack_pointer[-1] = exhausted;
+            DEOPT_IF(r->len <= 0, _GUARD_NOT_EXHAUSTED_RANGE);
             break;
         }
 

diff --git a/Python/optimizer.c b/Python/optimizer.c
index a332fd1c89582caaf64732d5dcf53324ddda84df..065e12746719938e07e8789147fdf6c3e61d5e0c 100644 (file)
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -384,6 +384,14 @@ PyTypeObject _PyUOpExecutor_Type = {
     .tp_methods = executor_methods,
 };
 
+/* TO DO -- Generate this table */
+static const uint16_t
+_PyUop_Replacements[OPCODE_METADATA_SIZE] = {
+    [_ITER_JUMP_RANGE] = _GUARD_NOT_EXHAUSTED_RANGE,
+    [_ITER_JUMP_LIST] = _GUARD_NOT_EXHAUSTED_LIST,
+    [_ITER_JUMP_TUPLE] = _GUARD_NOT_EXHAUSTED_TUPLE,
+};
+
 #define TRACE_STACK_SIZE 5
 
 /* Returns 1 on success,
@@ -586,46 +594,6 @@ pop_jump_if_bool:
                 break;
             }
 
-            case FOR_ITER_LIST:
-            case FOR_ITER_TUPLE:
-            case FOR_ITER_RANGE:
-            {
-                RESERVE(4, 3);
-                int check_op, exhausted_op, next_op;
-                switch (opcode) {
-                    case FOR_ITER_LIST:
-                        check_op = _ITER_CHECK_LIST;
-                        exhausted_op = _IS_ITER_EXHAUSTED_LIST;
-                        next_op = _ITER_NEXT_LIST;
-                        break;
-                    case FOR_ITER_TUPLE:
-                        check_op = _ITER_CHECK_TUPLE;
-                        exhausted_op = _IS_ITER_EXHAUSTED_TUPLE;
-                        next_op = _ITER_NEXT_TUPLE;
-                        break;
-                    case FOR_ITER_RANGE:
-                        check_op = _ITER_CHECK_RANGE;
-                        exhausted_op = _IS_ITER_EXHAUSTED_RANGE;
-                        next_op = _ITER_NEXT_RANGE;
-                        break;
-                    default:
-                        Py_UNREACHABLE();
-                }
-                // Assume jump unlikely (can a for-loop exit be likely?)
-                _Py_CODEUNIT *target_instr =  // +1 at the end skips over END_FOR
-                    instr + 1 + _PyOpcode_Caches[_PyOpcode_Deopt[opcode]] + oparg + 1;
-                max_length -= 3;  // Really the start of the stubs
-                ADD_TO_TRACE(check_op, 0, 0);
-                ADD_TO_TRACE(exhausted_op, 0, 0);
-                ADD_TO_TRACE(_POP_JUMP_IF_TRUE, max_length, 0);
-                ADD_TO_TRACE(next_op, 0, 0);
-
-                ADD_TO_STUB(max_length + 0, POP_TOP, 0, 0);
-                ADD_TO_STUB(max_length + 1, _SET_IP, INSTR_IP(target_instr, code), 0);
-                ADD_TO_STUB(max_length + 2, _EXIT_TRACE, 0, 0);
-                break;
-            }
-
             default:
             {
                 const struct opcode_macro_expansion *expansion = &_PyOpcode_macro_expansion[opcode];
@@ -661,6 +629,9 @@ pop_jump_if_bool:
                                         oparg += extras;
                                     }
                                 }
+                                if (_PyUop_Replacements[uop]) {
+                                    uop = _PyUop_Replacements[uop];
+                                }
                                 break;
                             case OPARG_CACHE_1:
                                 operand = read_u16(&instr[offset].cache);

diff --git a/Tools/cases_generator/generate_cases.py b/Tools/cases_generator/generate_cases.py
index a3313b4929afc4e6d5a1b9a7458ae5c4679d2223..d1dbfeae8d74f69b3fcf827511d9dc8e915e1bb0 100644 (file)
--- a/Tools/cases_generator/generate_cases.py
+++ b/Tools/cases_generator/generate_cases.py
@@ -655,7 +655,7 @@ class Generator(Analyzer):
                 if "specializing" in part.instr.annotations:
                     continue
                 # All other component instructions must be viable uops
-                if not part.instr.is_viable_uop():
+                if not part.instr.is_viable_uop() and "replaced" not in part.instr.annotations:
                     # This note just reminds us about macros that cannot
                     # be expanded to Tier 2 uops. It is not an error.
                     # It is sometimes emitted for macros that have a

diff --git a/Tools/cases_generator/lexer.py b/Tools/cases_generator/lexer.py
index 4c5403dc1d1fa8b8a58dcb9eeb2cf92408610c45..1185c85578593978184c1d4343cca9d916f436fa 100644 (file)
--- a/Tools/cases_generator/lexer.py
+++ b/Tools/cases_generator/lexer.py
@@ -216,7 +216,7 @@ kwds.append(MACRO)
 keywords = {name.lower(): name for name in kwds}
 
 ANNOTATION = "ANNOTATION"
-annotations = {"specializing", "guard", "override", "register"}
+annotations = {"specializing", "guard", "override", "register", "replaced"}
 
 __all__ = []
 __all__.extend(kwds)