Use PyThreadState_GET() in performance critical code

It seems like _PyThreadState_UncheckedGet() is not inlined as expected, even
when using gcc -O3.

Objects/dictobject.c

@@ -1409,7 +1409,7 @@ PyDict_GetItem(PyObject *op, PyObject *key)
        Let's just hope that no exception occurs then...  This must be
        _PyThreadState_Current and not PyThreadState_GET() because in debug
        mode, the latter complains if tstate is NULL. */
-    tstate = _PyThreadState_UncheckedGet();
+    tstate = PyThreadState_GET();
     if (tstate != NULL && tstate->curexc_type != NULL) {
         /* preserve the existing exception */
         PyObject *err_type, *err_value, *err_tb;

Python/errors.c

@@ -161,7 +161,7 @@ PyErr_SetString(PyObject *exception, const char *string)
 PyObject *
 PyErr_Occurred(void)
 {
-    PyThreadState *tstate = _PyThreadState_UncheckedGet();
+    PyThreadState *tstate = PyThreadState_GET();
     return tstate == NULL ? NULL : tstate->curexc_type;
 }
 

Python/sysmodule.c

@@ -1547,8 +1547,9 @@ error:
     Py_XDECREF(name);
     Py_XDECREF(value);
     /* No return value, therefore clear error state if possible */
-    if (_PyThreadState_UncheckedGet())
+    if (_PyThreadState_UncheckedGet()) {
         PyErr_Clear();
+    }
 }
 
 PyObject *