Mark a few functions used by the interpreter loop as noinline

These are all the slow path and should not be inlined into the interpreter
loop. Unfortunately, they end up being inlined with LTO and the current PGO
task.

Author: mpage

Python/ceval.c

@@ -1427,7 +1427,7 @@ skip_to_next_entry(unsigned char *p, unsigned char *end) {
 
 #define MAX_LINEAR_SEARCH 40
 
-static int
+static Py_NO_INLINE int
 get_exception_handler(PyCodeObject *code, int index, int *level, int *handler, int *lasti)
 {
     unsigned char *start = (unsigned char *)PyBytes_AS_STRING(code->co_exceptiontable);

Python/instrumentation.c

@@ -1200,7 +1200,7 @@ call_instrumentation_vector(
     return err;
 }
 
-int
+Py_NO_INLINE int
 _Py_call_instrumentation(
     PyThreadState *tstate, int event,
     _PyInterpreterFrame *frame, _Py_CODEUNIT *instr)
@@ -1209,7 +1209,7 @@ _Py_call_instrumentation(
     return call_instrumentation_vector(instr, tstate, event, frame, instr, 2, args);
 }
 
-int
+Py_NO_INLINE int
 _Py_call_instrumentation_arg(
     PyThreadState *tstate, int event,
     _PyInterpreterFrame *frame, _Py_CODEUNIT *instr, PyObject *arg)
@@ -1218,7 +1218,7 @@ _Py_call_instrumentation_arg(
     return call_instrumentation_vector(instr, tstate, event, frame, instr, 3, args);
 }
 
-int
+Py_NO_INLINE int
 _Py_call_instrumentation_2args(
     PyThreadState *tstate, int event,
     _PyInterpreterFrame *frame, _Py_CODEUNIT *instr, PyObject *arg0, PyObject *arg1)
@@ -1227,7 +1227,7 @@ _Py_call_instrumentation_2args(
     return call_instrumentation_vector(instr, tstate, event, frame, instr, 4, args);
 }
 
-_Py_CODEUNIT *
+Py_NO_INLINE _Py_CODEUNIT *
 _Py_call_instrumentation_jump(
     _Py_CODEUNIT *instr, PyThreadState *tstate, int event,
     _PyInterpreterFrame *frame, _Py_CODEUNIT *src, _Py_CODEUNIT *dest)
@@ -1271,7 +1271,7 @@ call_instrumentation_vector_protected(
     assert(_PyErr_Occurred(tstate));
 }
 
-void
+Py_NO_INLINE void
 _Py_call_instrumentation_exc2(
     PyThreadState *tstate, int event,
     _PyInterpreterFrame *frame, _Py_CODEUNIT *instr, PyObject *arg0, PyObject *arg1)
@@ -1294,7 +1294,7 @@ _Py_Instrumentation_GetLine(PyCodeObject *code, int index)
     return line;
 }
 
-int
+Py_NO_INLINE int
 _Py_call_instrumentation_line(PyThreadState *tstate, _PyInterpreterFrame* frame, _Py_CODEUNIT *instr, _Py_CODEUNIT *prev)
 {
     PyCodeObject *code = _PyFrame_GetCode(frame);
@@ -1396,7 +1396,7 @@ _Py_call_instrumentation_line(PyThreadState *tstate, _PyInterpreterFrame* frame,
     return original_opcode;
 }
 
-int
+Py_NO_INLINE int
 _Py_call_instrumentation_instruction(PyThreadState *tstate, _PyInterpreterFrame* frame, _Py_CODEUNIT *instr)
 {
     PyCodeObject *code = _PyFrame_GetCode(frame);

Python/specialize.c

@@ -818,7 +818,7 @@ specialize_module_load_attr(
 
 /* Attribute specialization */
 
-void
+Py_NO_INLINE void
 _Py_Specialize_LoadSuperAttr(_PyStackRef global_super_st, _PyStackRef cls_st, _Py_CODEUNIT *instr, int load_method) {
     PyObject *global_super = PyStackRef_AsPyObjectBorrow(global_super_st);
     PyObject *cls = PyStackRef_AsPyObjectBorrow(cls_st);
@@ -1342,7 +1342,7 @@ specialize_instance_load_attr(PyObject* owner, _Py_CODEUNIT* instr, PyObject* na
     return result;
 }
 
-void
+Py_NO_INLINE void
 _Py_Specialize_LoadAttr(_PyStackRef owner_st, _Py_CODEUNIT *instr, PyObject *name)
 {
     PyObject *owner = PyStackRef_AsPyObjectBorrow(owner_st);
@@ -1373,7 +1373,7 @@ _Py_Specialize_LoadAttr(_PyStackRef owner_st, _Py_CODEUNIT *instr, PyObject *nam
     }
 }
 
-void
+Py_NO_INLINE void
 _Py_Specialize_StoreAttr(_PyStackRef owner_st, _Py_CODEUNIT *instr, PyObject *name)
 {
     PyObject *owner = PyStackRef_AsPyObjectBorrow(owner_st);
@@ -1771,7 +1771,7 @@ specialize_load_global_lock_held(
     unspecialize(instr);
 }
 
-void
+Py_NO_INLINE void
 _Py_Specialize_LoadGlobal(
     PyObject *globals, PyObject *builtins,
     _Py_CODEUNIT *instr, PyObject *name)
@@ -1891,7 +1891,7 @@ store_subscr_fail_kind(PyObject *container, PyObject *sub)
 }
 #endif
 
-void
+Py_NO_INLINE void
 _Py_Specialize_StoreSubscr(_PyStackRef container_st, _PyStackRef sub_st, _Py_CODEUNIT *instr)
 {
     PyObject *container = PyStackRef_AsPyObjectBorrow(container_st);
@@ -2171,7 +2171,7 @@ specialize_c_call(PyObject *callable, _Py_CODEUNIT *instr, int nargs)
     }
 }
 
-void
+Py_NO_INLINE void
 _Py_Specialize_Call(_PyStackRef callable_st, _Py_CODEUNIT *instr, int nargs)
 {
     PyObject *callable = PyStackRef_AsPyObjectBorrow(callable_st);
@@ -2211,7 +2211,7 @@ _Py_Specialize_Call(_PyStackRef callable_st, _Py_CODEUNIT *instr, int nargs)
     }
 }
 
-void
+Py_NO_INLINE void
 _Py_Specialize_CallKw(_PyStackRef callable_st, _Py_CODEUNIT *instr, int nargs)
 {
     PyObject *callable = PyStackRef_AsPyObjectBorrow(callable_st);
@@ -2566,7 +2566,7 @@ binary_op_extended_specialization(PyObject *lhs, PyObject *rhs, int oparg,
     return 0;
 }
 
-void
+Py_NO_INLINE void
 _Py_Specialize_BinaryOp(_PyStackRef lhs_st, _PyStackRef rhs_st, _Py_CODEUNIT *instr,
                         int oparg, _PyStackRef *locals)
 {
@@ -2724,7 +2724,7 @@ compare_op_fail_kind(PyObject *lhs, PyObject *rhs)
 }
 #endif   // Py_STATS
 
-void
+Py_NO_INLINE void
 _Py_Specialize_CompareOp(_PyStackRef lhs_st, _PyStackRef rhs_st, _Py_CODEUNIT *instr,
                          int oparg)
 {
@@ -2787,7 +2787,7 @@ unpack_sequence_fail_kind(PyObject *seq)
 }
 #endif   // Py_STATS
 
-void
+Py_NO_INLINE void
 _Py_Specialize_UnpackSequence(_PyStackRef seq_st, _Py_CODEUNIT *instr, int oparg)
 {
     PyObject *seq = PyStackRef_AsPyObjectBorrow(seq_st);
@@ -2894,7 +2894,7 @@ int
 }
 #endif   // Py_STATS
 
-void
+Py_NO_INLINE void
 _Py_Specialize_ForIter(_PyStackRef iter, _Py_CODEUNIT *instr, int oparg)
 {
     assert(ENABLE_SPECIALIZATION_FT);
@@ -2949,7 +2949,7 @@ _Py_Specialize_ForIter(_PyStackRef iter, _Py_CODEUNIT *instr, int oparg)
     unspecialize(instr);
 }
 
-void
+Py_NO_INLINE void
 _Py_Specialize_Send(_PyStackRef receiver_st, _Py_CODEUNIT *instr)
 {
     PyObject *receiver = PyStackRef_AsPyObjectBorrow(receiver_st);
@@ -3019,7 +3019,7 @@ check_type_always_true(PyTypeObject *ty)
     return 0;
 }
 
-void
+Py_NO_INLINE void
 _Py_Specialize_ToBool(_PyStackRef value_o, _Py_CODEUNIT *instr)
 {
     assert(ENABLE_SPECIALIZATION_FT);
@@ -3093,7 +3093,7 @@ containsop_fail_kind(PyObject *value) {
 }
 #endif
 
-void
+Py_NO_INLINE void
 _Py_Specialize_ContainsOp(_PyStackRef value_st, _Py_CODEUNIT *instr)
 {
     PyObject *value = PyStackRef_AsPyObjectBorrow(value_st);