Finish inlining cache for .

python · markshannon · Mar 1, 2022 · Feb 28, 2022 · Feb 28, 2022 · Feb 28, 2022
commit a9f7043956bad266c74f8a8725ada087e5d06fcc
@@ -389,6 +389,7 @@ def _write_atomic(path, data, mode=0o666):
 #     Python 3.11a5 3480 (New CALL opcodes, second iteration)
 #     Python 3.11a5 3481 (Use inline cache for BINARY_OP)
 #     Python 3.11a5 3482 (Use inline caching for UNPACK_SEQUENCE and LOAD_GLOBAL)
+#     Python 3.11a5 3483 (Use inline caching for BINARY_SUBSCR)
 
 #     Python 3.12 will start with magic number 3500
 
@@ -403,7 +404,7 @@ def _write_atomic(path, data, mode=0o666):
 # Whenever MAGIC_NUMBER is changed, the ranges in the magic_values array
 # in PC/launcher.c must also be updated.
 
-MAGIC_NUMBER = (3482).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3483).to_bytes(2, 'little') + b'\r\n'
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 
 _PYCACHE = '__pycache__'

diff --git a/Lib/test/test_capi.py b/Lib/test/test_capi.py
@@ -335,7 +335,7 @@ class C(): pass
             *_, count = line.split(b' ')
             count = int(count)
             self.assertLessEqual(count, i*5)
-            self.assertGreaterEqual(count, i*5-1)
+            self.assertGreaterEqual(count, i*5-2)
 
     def test_mapping_keys_values_items(self):
         class Mapping1(dict):

diff --git a/Objects/codeobject.c b/Objects/codeobject.c
@@ -1925,6 +1925,10 @@ _PyStaticCode_Dealloc(PyCodeObject *co)
         co->co_quickened = NULL;
          _Py_QuickenedCount--;
     }
+    if (co->_co_obj_cache) {
+        PyMem_Free(co->_co_obj_cache);
+        co->_co_obj_cache = NULL;
+    }
     co->co_warmup = QUICKENING_INITIAL_WARMUP_VALUE;
     PyMem_Free(co->co_extra);
     co->co_extra = NULL;

@@ -2198,10 +2198,10 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
             PyObject *container = SECOND();
             _PyBinarySubscrCache *cache = (_PyBinarySubscrCache *)next_instr;
             PyObject *cached = frame->f_code->_co_obj_cache[cache->object];
-            assert(PyFunction_Check(cached));
-            PyFunctionObject *getitem = (PyFunctionObject *)cached;
             uint32_t type_version = read32(&cache->type_version);
             DEOPT_IF(Py_TYPE(container)->tp_version_tag != type_version, BINARY_SUBSCR);
+            assert(PyFunction_Check(cached));
+            PyFunctionObject *getitem = (PyFunctionObject *)cached;
             DEOPT_IF(getitem->func_version != cache->func_version, BINARY_SUBSCR);
             PyCodeObject *code = (PyCodeObject *)getitem->func_code;
             size_t size = code->co_nlocalsplus + code->co_stacksize + FRAME_SPECIALS_SIZE;
@@ -2221,10 +2221,10 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
                 new_frame->localsplus[i] = NULL;
             }
             _PyFrame_SetStackPointer(frame, stack_pointer);
+            frame->f_lasti += INLINE_CACHE_ENTRIES_BINARY_SUBSCR;
             new_frame->previous = frame;
             frame = cframe.current_frame = new_frame;
             CALL_STAT_INC(inlined_py_calls);
-            JUMPBY(INLINE_CACHE_ENTRIES_BINARY_SUBSCR);
             goto start_frame;
         }
 
@@ -5606,7 +5606,7 @@ MISS_WITH_CACHE(PRECALL)
 MISS_WITH_CACHE(CALL)
 MISS_WITH_INLINE_CACHE(BINARY_OP)
 MISS_WITH_CACHE(COMPARE_OP)
-MISS_WITH_CACHE(BINARY_SUBSCR)
+MISS_WITH_INLINE_CACHE(BINARY_SUBSCR)
 MISS_WITH_INLINE_CACHE(UNPACK_SEQUENCE)
 MISS_WITH_OPARG_COUNTER(STORE_SUBSCR)
 

diff --git a/Python/specialize.c b/Python/specialize.c
@@ -1409,7 +1409,7 @@ _Py_Specialize_BinarySubscr(
             goto fail;
         }
         assert(cls->tp_version_tag != 0);
-        cache->type_version = cls->tp_version_tag;
+        write32(&cache->type_version, cls->tp_version_tag);
         int version = _PyFunction_GetVersionForCurrentState(func);
         if (version == 0 || version != (uint16_t)version) {
             SPECIALIZATION_FAIL(BINARY_SUBSCR, SPEC_FAIL_OUT_OF_VERSIONS);