gh-115999: Add free-threaded specialization for ``TO_BOOL`` (gh-126616)

4 files changed, 74 insertions, 69 deletions

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 71b1dc0..6ee886c 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -391,7 +391,7 @@ dummy_func(
         };
 
         specializing op(_SPECIALIZE_TO_BOOL, (counter/1, value -- value)) {
-            #if ENABLE_SPECIALIZATION
+            #if ENABLE_SPECIALIZATION_FT
             if (ADAPTIVE_COUNTER_TRIGGERS(counter)) {
                 next_instr = this_instr;
                 _Py_Specialize_ToBool(value, next_instr);
@@ -399,7 +399,7 @@ dummy_func(
             }
             OPCODE_DEFERRED_INC(TO_BOOL);
             ADVANCE_ADAPTIVE_COUNTER(this_instr[1].counter);
-            #endif  /* ENABLE_SPECIALIZATION */
+            #endif  /* ENABLE_SPECIALIZATION_FT */
         }
 
         op(_TO_BOOL, (value -- res)) {
@@ -435,7 +435,7 @@ dummy_func(
             PyObject *value_o = PyStackRef_AsPyObjectBorrow(value);
             EXIT_IF(!PyList_CheckExact(value_o));
             STAT_INC(TO_BOOL, hit);
-            res = Py_SIZE(value_o) ? PyStackRef_True : PyStackRef_False;
+            res = PyList_GET_SIZE(value_o) ? PyStackRef_True : PyStackRef_False;
             DECREF_INPUTS();
         }
 
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 8acf7a4..5c7138a 100644
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -508,7 +508,7 @@
                 JUMP_TO_JUMP_TARGET();
             }
             STAT_INC(TO_BOOL, hit);
-            res = Py_SIZE(value_o) ? PyStackRef_True : PyStackRef_False;
+            res = PyList_GET_SIZE(value_o) ? PyStackRef_True : PyStackRef_False;
             PyStackRef_CLOSE(value);
             stack_pointer[-1] = res;
             break;
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 8896229..13947849 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -7758,7 +7758,7 @@
                 value = stack_pointer[-1];
                 uint16_t counter = read_u16(&this_instr[1].cache);
                 (void)counter;
-                #if ENABLE_SPECIALIZATION
+                #if ENABLE_SPECIALIZATION_FT
                 if (ADAPTIVE_COUNTER_TRIGGERS(counter)) {
                     next_instr = this_instr;
                     _PyFrame_SetStackPointer(frame, stack_pointer);
@@ -7768,7 +7768,7 @@
                 }
                 OPCODE_DEFERRED_INC(TO_BOOL);
                 ADVANCE_ADAPTIVE_COUNTER(this_instr[1].counter);
-                #endif  /* ENABLE_SPECIALIZATION */
+                #endif  /* ENABLE_SPECIALIZATION_FT */
             }
             /* Skip 2 cache entries */
             // _TO_BOOL
@@ -7863,7 +7863,7 @@
             PyObject *value_o = PyStackRef_AsPyObjectBorrow(value);
             DEOPT_IF(!PyList_CheckExact(value_o), TO_BOOL);
             STAT_INC(TO_BOOL, hit);
-            res = Py_SIZE(value_o) ? PyStackRef_True : PyStackRef_False;
+            res = PyList_GET_SIZE(value_o) ? PyStackRef_True : PyStackRef_False;
             PyStackRef_CLOSE(value);
             stack_pointer[-1] = res;
             DISPATCH();
diff --git a/Python/specialize.c b/Python/specialize.c
index af37e24..c69f61c 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -2667,101 +2667,106 @@ success:
     cache->counter = adaptive_counter_cooldown();
 }
 
+#ifdef Py_STATS
+static int
+to_bool_fail_kind(PyObject *value)
+{
+    if (PyByteArray_CheckExact(value)) {
+        return SPEC_FAIL_TO_BOOL_BYTEARRAY;
+    }
+    if (PyBytes_CheckExact(value)) {
+        return SPEC_FAIL_TO_BOOL_BYTES;
+    }
+    if (PyDict_CheckExact(value)) {
+        return SPEC_FAIL_TO_BOOL_DICT;
+    }
+    if (PyFloat_CheckExact(value)) {
+        return SPEC_FAIL_TO_BOOL_FLOAT;
+    }
+    if (PyMemoryView_Check(value)) {
+        return SPEC_FAIL_TO_BOOL_MEMORY_VIEW;
+    }
+    if (PyAnySet_CheckExact(value)) {
+        return SPEC_FAIL_TO_BOOL_SET;
+    }
+    if (PyTuple_CheckExact(value)) {
+        return SPEC_FAIL_TO_BOOL_TUPLE;
+    }
+    return SPEC_FAIL_OTHER;
+}
+#endif  // Py_STATS
+
+static int
+check_type_always_true(PyTypeObject *ty)
+{
+    PyNumberMethods *nb = ty->tp_as_number;
+    if (nb && nb->nb_bool) {
+        return SPEC_FAIL_TO_BOOL_NUMBER;
+    }
+    PyMappingMethods *mp = ty->tp_as_mapping;
+    if (mp && mp->mp_length) {
+        return SPEC_FAIL_TO_BOOL_MAPPING;
+    }
+    PySequenceMethods *sq = ty->tp_as_sequence;
+    if (sq && sq->sq_length) {
+      return SPEC_FAIL_TO_BOOL_SEQUENCE;
+    }
+    return 0;
+}
+
 void
 _Py_Specialize_ToBool(_PyStackRef value_o, _Py_CODEUNIT *instr)
 {
-    assert(ENABLE_SPECIALIZATION);
+    assert(ENABLE_SPECIALIZATION_FT);
     assert(_PyOpcode_Caches[TO_BOOL] == INLINE_CACHE_ENTRIES_TO_BOOL);
     _PyToBoolCache *cache = (_PyToBoolCache *)(instr + 1);
     PyObject *value = PyStackRef_AsPyObjectBorrow(value_o);
+    uint8_t specialized_op;
     if (PyBool_Check(value)) {
-        instr->op.code = TO_BOOL_BOOL;
+        specialized_op = TO_BOOL_BOOL;
         goto success;
     }
     if (PyLong_CheckExact(value)) {
-        instr->op.code = TO_BOOL_INT;
+        specialized_op = TO_BOOL_INT;
         goto success;
     }
     if (PyList_CheckExact(value)) {
-        instr->op.code = TO_BOOL_LIST;
+        specialized_op = TO_BOOL_LIST;
         goto success;
     }
     if (Py_IsNone(value)) {
-        instr->op.code = TO_BOOL_NONE;
+        specialized_op = TO_BOOL_NONE;
         goto success;
     }
     if (PyUnicode_CheckExact(value)) {
-        instr->op.code = TO_BOOL_STR;
+        specialized_op = TO_BOOL_STR;
         goto success;
     }
     if (PyType_HasFeature(Py_TYPE(value), Py_TPFLAGS_HEAPTYPE)) {
-        PyNumberMethods *nb = Py_TYPE(value)->tp_as_number;
-        if (nb && nb->nb_bool) {
-            SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_NUMBER);
-            goto failure;
-        }
-        PyMappingMethods *mp = Py_TYPE(value)->tp_as_mapping;
-        if (mp && mp->mp_length) {
-            SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_MAPPING);
-            goto failure;
-        }
-        PySequenceMethods *sq = Py_TYPE(value)->tp_as_sequence;
-        if (sq && sq->sq_length) {
-            SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_SEQUENCE);
-            goto failure;
-        }
-        if (!PyUnstable_Type_AssignVersionTag(Py_TYPE(value))) {
+        unsigned int version = 0;
+        int err = _PyType_Validate(Py_TYPE(value), check_type_always_true, &version);
+        if (err < 0) {
             SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_OUT_OF_VERSIONS);
             goto failure;
         }
-        uint32_t version = type_get_version(Py_TYPE(value), TO_BOOL);
-        if (version == 0) {
+        else if (err > 0) {
+            SPECIALIZATION_FAIL(TO_BOOL, err);
             goto failure;
         }
-        instr->op.code = TO_BOOL_ALWAYS_TRUE;
-        write_u32(cache->version, version);
+
+        assert(err == 0);
         assert(version);
+        write_u32(cache->version, version);
+        specialized_op = TO_BOOL_ALWAYS_TRUE;
         goto success;
     }
-#ifdef Py_STATS
-    if (PyByteArray_CheckExact(value)) {
-        SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_BYTEARRAY);
-        goto failure;
-    }
-    if (PyBytes_CheckExact(value)) {
-        SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_BYTES);
-        goto failure;
-    }
-    if (PyDict_CheckExact(value)) {
-        SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_DICT);
-        goto failure;
-    }
-    if (PyFloat_CheckExact(value)) {
-        SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_FLOAT);
-        goto failure;
-    }
-    if (PyMemoryView_Check(value)) {
-        SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_MEMORY_VIEW);
-        goto failure;
-    }
-    if (PyAnySet_CheckExact(value)) {
-        SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_SET);
-        goto failure;
-    }
-    if (PyTuple_CheckExact(value)) {
-        SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_TO_BOOL_TUPLE);
-        goto failure;
-    }
-    SPECIALIZATION_FAIL(TO_BOOL, SPEC_FAIL_OTHER);
-#endif   // Py_STATS
+
+    SPECIALIZATION_FAIL(TO_BOOL, to_bool_fail_kind(value));
 failure:
-    STAT_INC(TO_BOOL, failure);
-    instr->op.code = TO_BOOL;
-    cache->counter = adaptive_counter_backoff(cache->counter);
+    unspecialize(instr);
     return;
 success:
-    STAT_INC(TO_BOOL, success);
-    cache->counter = adaptive_counter_cooldown();
+    specialize(instr, specialized_op);
 }
 
 #ifdef Py_STATS