revert benchmark stuff

Include/internal/pycore_pystate.h

@@ -135,7 +135,7 @@ static inline PyThreadState*
 _PyThreadState_GET(void)
 {
 #if defined(HAVE_THREAD_LOCAL) && !defined(Py_BUILD_CORE_MODULE)
-    return (PyThreadState*)_Py_atomic_load_ptr_relaxed(&_PyRuntime.tstate_current);
+    return _Py_tss_tstate;
 #else
     return _PyThreadState_GetCurrent();
 #endif

Include/internal/pycore_runtime.h

@@ -285,7 +285,6 @@ typedef struct pyruntimestate {
     struct _pythread_runtime_state threads;
     struct _signals_runtime_state signals;
 
-    PyThreadState *tstate_current;
     /* Used for the thread state bound to the current thread. */
     Py_tss_t autoTSSkey;
 

Objects/typeobject.c

@@ -5226,57 +5226,6 @@ get_module_by_def(PyTypeObject *type, PyModuleDef *def)
     return res;
 }
 
-// copied from the above
-Py_NO_INLINE static PyObject *
-get_module_by_def_NoInline(PyTypeObject *type, PyModuleDef *def)
-{
-    assert(PyType_Check(type));
-
-    if (!_PyType_HasFeature(type, Py_TPFLAGS_HEAPTYPE)) {
-        // type_ready_mro() ensures that no heap type is
-        // contained in a static type MRO.
-        return NULL;
-    }
-    else {
-        PyHeapTypeObject *ht = (PyHeapTypeObject*)type;
-        PyObject *module = ht->ht_module;
-        if (module && _PyModule_GetDef(module) == def) {
-            return module;
-        }
-    }
-
-    PyObject *res = NULL;
-    BEGIN_TYPE_LOCK();
-
-    PyObject *mro = lookup_tp_mro(type);
-    // The type must be ready
-    assert(mro != NULL);
-    assert(PyTuple_Check(mro));
-    // mro_invoke() ensures that the type MRO cannot be empty.
-    assert(PyTuple_GET_SIZE(mro) >= 1);
-    // Also, the first item in the MRO is the type itself, which
-    // we already checked above. We skip it in the loop.
-    assert(PyTuple_GET_ITEM(mro, 0) == (PyObject *)type);
-
-    Py_ssize_t n = PyTuple_GET_SIZE(mro);
-    for (Py_ssize_t i = 1; i < n; i++) {
-        PyObject *super = PyTuple_GET_ITEM(mro, i);
-        if(!_PyType_HasFeature((PyTypeObject *)super, Py_TPFLAGS_HEAPTYPE)) {
-            // Static types in the MRO need to be skipped
-            continue;
-        }
-
-        PyHeapTypeObject *ht = (PyHeapTypeObject*)super;
-        PyObject *module = ht->ht_module;
-        if (module && _PyModule_GetDef(module) == def) {
-            res = module;
-            break;
-        }
-    }
-    END_TYPE_LOCK();
-    return res;
-}
-
 PyObject *
 PyType_GetModuleByDef(PyTypeObject *type, PyModuleDef *def)
 {

PCbuild/pyproject.props

@@ -73,9 +73,6 @@
       <AdditionalOptions Condition="$(MSVCHasBrokenARM64SignExtension) == 'true' and $(Platform) == 'ARM64'">-d2ssa-patterns-all- %(AdditionalOptions)</AdditionalOptions>
       <AdditionalOptions Condition="$(GenerateSourceDependencies) == 'true'">/sourceDependencies "$(IntDir.Trim(`\`))" %(AdditionalOptions)</AdditionalOptions>
     </ClCompile>
-    <ClCompile Condition="!$(RootNamespace.Contains(`test`)) and $(RootNamespace) != '_freeze_module'">
-      <AdditionalOptions>/Ob3 %(AdditionalOptions)</AdditionalOptions>
-    </ClCompile>
     <ClCompile Condition="$(Configuration) == 'Debug'">
       <InlineFunctionExpansion>OnlyExplicitInline</InlineFunctionExpansion>
       <Optimization>Disabled</Optimization>

Python/pystate.c

@@ -79,31 +79,23 @@ current_fast_get(void)
 #endif
 }
 
-static inline PyThreadState *
-current_fast_get2(void)
-{
-    return (PyThreadState*)_Py_atomic_load_ptr_relaxed(&_PyRuntime.tstate_current);
-}
-
 static inline void
 current_fast_set(_PyRuntimeState *Py_UNUSED(runtime), PyThreadState *tstate)
 {
     assert(tstate != NULL);
 #ifdef HAVE_THREAD_LOCAL
     _Py_tss_tstate = tstate;
-    _Py_atomic_store_ptr_relaxed(&_PyRuntime.tstate_current, tstate);
 #else
     // XXX Fall back to the PyThread_tss_*() API.
 #  error "no supported thread-local variable storage classifier"
 #endif
 }
 
 static inline void
-current_fast_clear(_PyRuntimeState *runtime)
+current_fast_clear(_PyRuntimeState *Py_UNUSED(runtime))
 {
 #ifdef HAVE_THREAD_LOCAL
     _Py_tss_tstate = NULL;
-    _Py_atomic_store_ptr_relaxed(&runtime->tstate_current, NULL);
 #else
     // XXX Fall back to the PyThread_tss_*() API.
 #  error "no supported thread-local variable storage classifier"
@@ -118,7 +110,7 @@ current_fast_clear(_PyRuntimeState *runtime)
 PyThreadState *
 _PyThreadState_GetCurrent(void)
 {
-    return current_fast_get2();
+    return current_fast_get();
 }
 
 
@@ -1339,7 +1331,7 @@ _PyInterpreterState_RequireIDRef(PyInterpreterState *interp, int required)
 PyInterpreterState*
 PyInterpreterState_Get(void)
 {
-    PyThreadState *tstate = current_fast_get2();
+    PyThreadState *tstate = current_fast_get();
     _Py_EnsureTstateNotNULL(tstate);
     PyInterpreterState *interp = tstate->interp;
     if (interp == NULL) {
@@ -2420,14 +2412,14 @@ PyThreadState_SetAsyncExc(unsigned long id, PyObject *exc)
 PyThreadState *
 PyThreadState_GetUnchecked(void)
 {
-    return current_fast_get2();
+    return current_fast_get();
 }
 
 
 PyThreadState *
 PyThreadState_Get(void)
 {
-    PyThreadState *tstate = current_fast_get2();
+    PyThreadState *tstate = current_fast_get();
     _Py_EnsureTstateNotNULL(tstate);
     return tstate;
 }