bpo-47177: Replace f_lasti with prev_instr (pythonGH-32208)

Author: brandtbucher

Include/internal/pycore_frame.h

@@ -39,11 +39,6 @@ enum _frameowner {
     FRAME_OWNED_BY_FRAME_OBJECT = 2
 };
 
-/*
-    frame->f_lasti refers to the index of the last instruction,
-    unless it's -1 in which case next_instr should be first_instr.
-*/
-
 typedef struct _PyInterpreterFrame {
     PyFunctionObject *f_func; /* Strong reference */
     PyObject *f_globals; /* Borrowed reference */
@@ -52,13 +47,20 @@ typedef struct _PyInterpreterFrame {
     PyCodeObject *f_code; /* Strong reference */
     PyFrameObject *frame_obj; /* Strong reference, may be NULL */
     struct _PyInterpreterFrame *previous;
-    int f_lasti;       /* Last instruction if called */
+    // NOTE: This is not necessarily the last instruction started in the given
+    // frame. Rather, it is the code unit *prior to* the *next* instruction. For
+    // example, it may be an inline CACHE entry, an instruction we just jumped
+    // over, or (in the case of a newly-created frame) a totally invalid value:
+    _Py_CODEUNIT *prev_instr;
     int stacktop;     /* Offset of TOS from localsplus  */
     bool is_entry;  // Whether this is the "root" frame for the current _PyCFrame.
     char owner;
     PyObject *localsplus[1];
 } _PyInterpreterFrame;
 
+#define _PyInterpreterFrame_LASTI(IF) \
+    ((int)((IF)->prev_instr - _PyCode_CODE((IF)->f_code)))
+
 static inline PyObject **_PyFrame_Stackbase(_PyInterpreterFrame *f) {
     return f->localsplus + f->f_code->co_nlocalsplus;
 }
@@ -97,7 +99,7 @@ _PyFrame_InitializeSpecials(
     frame->f_locals = Py_XNewRef(locals);
     frame->stacktop = nlocalsplus;
     frame->frame_obj = NULL;
-    frame->f_lasti = -1;
+    frame->prev_instr = _PyCode_CODE(frame->f_code) - 1;
     frame->is_entry = false;
     frame->owner = FRAME_OWNED_BY_THREAD;
 }
@@ -186,6 +188,7 @@ void _PyThreadState_PopFrame(PyThreadState *tstate, _PyInterpreterFrame *frame);
 _PyInterpreterFrame *
 _PyFrame_Push(PyThreadState *tstate, PyFunctionObject *func);
 
+int _PyInterpreterFrame_GetLine(_PyInterpreterFrame *frame);
 
 static inline
 PyGenObject *_PyFrame_GetGenerator(_PyInterpreterFrame *frame)

Misc/NEWS.d/next/Core and Builtins/2022-03-31-01-30-03.bpo-47177.fQqaov.rst

@@ -0,0 +1 @@
+Replace the ``f_lasti`` member of the internal ``_PyInterpreterFrame`` structure with a ``prev_instr`` pointer, which reduces overhead in the main interpreter loop. The ``f_lasti`` attribute of Python-layer frame objects is preserved for backward-compatibility.

Modules/_tracemalloc.c

@@ -308,7 +308,7 @@ static void
 tracemalloc_get_frame(_PyInterpreterFrame *pyframe, frame_t *frame)
 {
     frame->filename = &_Py_STR(anon_unknown);
-    int lineno = PyCode_Addr2Line(pyframe->f_code, pyframe->f_lasti*sizeof(_Py_CODEUNIT));
+    int lineno = _PyInterpreterFrame_GetLine(pyframe);
     if (lineno < 0) {
         lineno = 0;
     }

Objects/frameobject.c

@@ -39,7 +39,7 @@ PyFrame_GetLineNumber(PyFrameObject *f)
         return f->f_lineno;
     }
     else {
-        return PyCode_Addr2Line(f->f_frame->f_code, f->f_frame->f_lasti*sizeof(_Py_CODEUNIT));
+        return _PyInterpreterFrame_GetLine(f->f_frame);
     }
 }
 
@@ -58,10 +58,11 @@ frame_getlineno(PyFrameObject *f, void *closure)
 static PyObject *
 frame_getlasti(PyFrameObject *f, void *closure)
 {
-    if (f->f_frame->f_lasti < 0) {
+    int lasti = _PyInterpreterFrame_LASTI(f->f_frame);
+    if (lasti < 0) {
         return PyLong_FromLong(-1);
     }
-    return PyLong_FromLong(f->f_frame->f_lasti*sizeof(_Py_CODEUNIT));
+    return PyLong_FromLong(lasti * sizeof(_Py_CODEUNIT));
 }
 
 static PyObject *
@@ -419,12 +420,11 @@ _PyFrame_GetState(PyFrameObject *frame)
         }
         case FRAME_OWNED_BY_THREAD:
         {
-            if (frame->f_frame->f_lasti < 0) {
+            if (_PyInterpreterFrame_LASTI(frame->f_frame) < 0) {
                 return FRAME_CREATED;
             }
-            uint8_t *code = (uint8_t *)frame->f_frame->f_code->co_code_adaptive;
-            int opcode = code[frame->f_frame->f_lasti*sizeof(_Py_CODEUNIT)];
-            switch(_PyOpcode_Deopt[opcode]) {
+            switch (_PyOpcode_Deopt[_Py_OPCODE(*frame->f_frame->prev_instr)])
+            {
                 case COPY_FREE_VARS:
                 case MAKE_CELL:
                 case RETURN_GENERATOR:
@@ -555,7 +555,7 @@ frame_setlineno(PyFrameObject *f, PyObject* p_new_lineno, void *Py_UNUSED(ignore
 
     int64_t best_stack = OVERFLOWED;
     int best_addr = -1;
-    int64_t start_stack = stacks[f->f_frame->f_lasti];
+    int64_t start_stack = stacks[_PyInterpreterFrame_LASTI(f->f_frame)];
     int err = -1;
     const char *msg = "cannot find bytecode for specified line";
     for (int i = 0; i < len; i++) {
@@ -598,7 +598,7 @@ frame_setlineno(PyFrameObject *f, PyObject* p_new_lineno, void *Py_UNUSED(ignore
     }
     /* Finally set the new lasti and return OK. */
     f->f_lineno = 0;
-    f->f_frame->f_lasti = best_addr;
+    f->f_frame->prev_instr = _PyCode_CODE(f->f_frame->f_code) + best_addr;
     return 0;
 }
 
@@ -880,10 +880,11 @@ _PyFrame_OpAlreadyRan(_PyInterpreterFrame *frame, int opcode, int oparg)
     // This only works when opcode is a non-quickened form:
     assert(_PyOpcode_Deopt[opcode] == opcode);
     int check_oparg = 0;
-    for (int i = 0; i < frame->f_lasti; i++) {
-        _Py_CODEUNIT instruction = _PyCode_CODE(frame->f_code)[i];
-        int check_opcode = _PyOpcode_Deopt[_Py_OPCODE(instruction)];
-        check_oparg |= _Py_OPARG(instruction);
+    for (_Py_CODEUNIT *instruction = _PyCode_CODE(frame->f_code); 
+         instruction < frame->prev_instr; instruction++)
+    {
+        int check_opcode = _PyOpcode_Deopt[_Py_OPCODE(*instruction)];
+        check_oparg |= _Py_OPARG(*instruction);
         if (check_opcode == opcode && check_oparg == oparg) {
             return 1;
         }
@@ -893,7 +894,7 @@ _PyFrame_OpAlreadyRan(_PyInterpreterFrame *frame, int opcode, int oparg)
         else {
             check_oparg = 0;
         }
-        i += _PyOpcode_Caches[check_opcode];
+        instruction += _PyOpcode_Caches[check_opcode];
     }
     return 0;
 }
@@ -914,8 +915,8 @@ _PyFrame_FastToLocalsWithError(_PyInterpreterFrame *frame) {
     fast = _PyFrame_GetLocalsArray(frame);
     // COPY_FREE_VARS has no quickened forms, so no need to use _PyOpcode_Deopt
     // here:
-    if (frame->f_lasti < 0 && _Py_OPCODE(_PyCode_CODE(co)[0]) == COPY_FREE_VARS)
-    {
+    int lasti = _PyInterpreterFrame_LASTI(frame);
+    if (lasti < 0 && _Py_OPCODE(_PyCode_CODE(co)[0]) == COPY_FREE_VARS) {
         /* Free vars have not been initialized -- Do that */
         PyCodeObject *co = frame->f_code;
         PyObject *closure = frame->f_func->func_closure;
@@ -926,7 +927,7 @@ _PyFrame_FastToLocalsWithError(_PyInterpreterFrame *frame) {
             frame->localsplus[offset + i] = o;
         }
         // COPY_FREE_VARS doesn't have inline CACHEs, either:
-        frame->f_lasti = 0;
+        frame->prev_instr = _PyCode_CODE(frame->f_code);
     }
     for (int i = 0; i < co->co_nlocalsplus; i++) {
         _PyLocals_Kind kind = _PyLocals_GetKind(co->co_localspluskinds, i);

Objects/genobject.c

@@ -352,14 +352,14 @@ _PyGen_yf(PyGenObject *gen)
     if (gen->gi_frame_state < FRAME_CLEARED) {
         _PyInterpreterFrame *frame = (_PyInterpreterFrame *)gen->gi_iframe;
 
-        if (frame->f_lasti < 1) {
+        if (gen->gi_frame_state == FRAME_CREATED) {
             /* Return immediately if the frame didn't start yet. SEND
                always come after LOAD_CONST: a code object should not start
                with SEND */
             assert(_Py_OPCODE(_PyCode_CODE(gen->gi_code)[0]) != SEND);
             return NULL;
         }
-        _Py_CODEUNIT next = _PyCode_CODE(gen->gi_code)[frame->f_lasti + 1];
+        _Py_CODEUNIT next = frame->prev_instr[1];
         if (_PyOpcode_Deopt[_Py_OPCODE(next)] != RESUME || _Py_OPARG(next) < 2)
         {
             /* Not in a yield from */
@@ -490,13 +490,11 @@ _gen_throw(PyGenObject *gen, int close_on_genexit,
             // XXX: Performing this jump ourselves is awkward and problematic.
             // See https://github.com/python/cpython/pull/31968.
             /* Termination repetition of SEND loop */
-            assert(frame->f_lasti >= 0);
-            _Py_CODEUNIT *code = _PyCode_CODE(gen->gi_code);
+            assert(_PyInterpreterFrame_LASTI(frame) >= 0);
             /* Backup to SEND */
-            frame->f_lasti--;
-            assert(_Py_OPCODE(code[frame->f_lasti]) == SEND);
-            int jump = _Py_OPARG(code[frame->f_lasti]);
-            frame->f_lasti += jump;
+            assert(_Py_OPCODE(frame->prev_instr[-1]) == SEND);
+            int jump = _Py_OPARG(frame->prev_instr[-1]);
+            frame->prev_instr += jump - 1;
             if (_PyGen_FetchStopIterationValue(&val) == 0) {
                 ret = gen_send(gen, val);
                 Py_DECREF(val);
@@ -1344,9 +1342,8 @@ compute_cr_origin(int origin_depth, _PyInterpreterFrame *current_frame)
     frame = current_frame;
     for (int i = 0; i < frame_count; ++i) {
         PyCodeObject *code = frame->f_code;
-        PyObject *frameinfo = Py_BuildValue("OiO",
-                                            code->co_filename,
-                                            PyCode_Addr2Line(frame->f_code, frame->f_lasti*sizeof(_Py_CODEUNIT)),
+        int line = _PyInterpreterFrame_GetLine(frame);
+        PyObject *frameinfo = Py_BuildValue("OiO", code->co_filename, line,
                                             code->co_name);
         if (!frameinfo) {
             Py_DECREF(cr_origin);

Objects/typeobject.c

@@ -8980,7 +8980,7 @@ super_init_without_args(_PyInterpreterFrame *cframe, PyCodeObject *co,
     if (firstarg != NULL && (_PyLocals_GetKind(co->co_localspluskinds, 0) & CO_FAST_CELL)) {
         // "firstarg" is a cell here unless (very unlikely) super()
         // was called from the C-API before the first MAKE_CELL op.
-        if (cframe->f_lasti >= 0) {
+        if (_PyInterpreterFrame_LASTI(cframe) >= 0) {
             // MAKE_CELL and COPY_FREE_VARS have no quickened forms, so no need
             // to use _PyOpcode_Deopt here:
             assert(_Py_OPCODE(_PyCode_CODE(co)[0]) == MAKE_CELL ||