pythonGH-108614: Remove non-debug uses of #if TIER_ONE and #if TIER_TWO from _POP_FRAME op. (pythonGH-108685)

Author: markshannon

Python/bytecodes.c

@@ -768,24 +768,25 @@ dummy_func(
         // different frame, and it's accounted for by _PUSH_FRAME.
         op(_POP_FRAME, (retval --)) {
             assert(EMPTY());
-            _PyFrame_SetStackPointer(frame, stack_pointer);
-            _Py_LeaveRecursiveCallPy(tstate);
-            // GH-99729: We need to unlink the frame *before* clearing it:
-            _PyInterpreterFrame *dying = frame;
             #if TIER_ONE
             assert(frame != &entry_frame);
             #endif
+            STORE_SP();
+            _Py_LeaveRecursiveCallPy(tstate);
+            // GH-99729: We need to unlink the frame *before* clearing it:
+            _PyInterpreterFrame *dying = frame;
             frame = tstate->current_frame = dying->previous;
             _PyEval_FrameClearAndPop(tstate, dying);
             frame->prev_instr += frame->return_offset;
             _PyFrame_StackPush(frame, retval);
-            #if TIER_ONE
-            goto resume_frame;
-            #endif
-            #if TIER_TWO
-            stack_pointer = _PyFrame_GetStackPointer(frame);
-            ip_offset = (_Py_CODEUNIT *)_PyFrame_GetCode(frame)->co_code_adaptive;
-            #endif
+            LOAD_SP();
+            LOAD_IP();
+#if LLTRACE && TIER_ONE
+            lltrace = maybe_lltrace_resume_frame(frame, &entry_frame, GLOBALS());
+            if (lltrace < 0) {
+                goto exit_unwind;
+            }
+#endif
         }
 
         macro(RETURN_VALUE) =

Python/ceval.c

@@ -189,6 +189,34 @@ lltrace_resume_frame(_PyInterpreterFrame *frame)
     fflush(stdout);
     PyErr_SetRaisedException(exc);
 }
+
+static int
+maybe_lltrace_resume_frame(_PyInterpreterFrame *frame, _PyInterpreterFrame *skip_frame, PyObject *globals)
+{
+    if (globals == NULL) {
+        return 0;
+    }
+    if (frame == skip_frame) {
+        return 0;
+    }
+    int r = PyDict_Contains(globals, &_Py_ID(__lltrace__));
+    if (r < 0) {
+        return -1;
+    }
+    int lltrace = r;
+    if (!lltrace) {
+        // When tracing executed uops, also trace bytecode
+        char *uop_debug = Py_GETENV("PYTHONUOPSDEBUG");
+        if (uop_debug != NULL && *uop_debug >= '0') {
+            lltrace = (*uop_debug - '0') >= 5;  // TODO: Parse an int and all that
+        }
+    }
+    if (lltrace) {
+        lltrace_resume_frame(frame);
+    }
+    return lltrace;
+}
+
 #endif
 
 static void monitor_raise(PyThreadState *tstate,
@@ -576,6 +604,7 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
     return _PyEval_EvalFrame(tstate, f->f_frame, throwflag);
 }
 
+#define TIER_ONE 1
 #include "ceval_macros.h"
 
 
@@ -714,24 +743,9 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
     SET_LOCALS_FROM_FRAME();
 
 #ifdef LLTRACE
-    {
-        if (frame != &entry_frame && GLOBALS()) {
-            int r = PyDict_Contains(GLOBALS(), &_Py_ID(__lltrace__));
-            if (r < 0) {
-                goto exit_unwind;
-            }
-            lltrace = r;
-            if (!lltrace) {
-                // When tracing executed uops, also trace bytecode
-                char *uop_debug = Py_GETENV("PYTHONUOPSDEBUG");
-                if (uop_debug != NULL && *uop_debug >= '0') {
-                    lltrace = (*uop_debug - '0') >= 5;  // TODO: Parse an int and all that
-                }
-            }
-        }
-        if (lltrace) {
-            lltrace_resume_frame(frame);
-        }
+    lltrace = maybe_lltrace_resume_frame(frame, &entry_frame, GLOBALS());
+    if (lltrace < 0) {
+        goto exit_unwind;
     }
 #endif
 
@@ -752,7 +766,6 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
 #endif
         {
 
-#define TIER_ONE 1
 #include "generated_cases.c.h"
 
     /* INSTRUMENTED_LINE has to be here, rather than in bytecodes.c,

Python/ceval_macros.h

@@ -373,3 +373,39 @@ static inline int _Py_EnterRecursivePy(PyThreadState *tstate) {
 static inline void _Py_LeaveRecursiveCallPy(PyThreadState *tstate)  {
     tstate->py_recursion_remaining++;
 }
+
+
+/* Implementation of "macros" that modify the instruction pointer,
+ * stack pointer, or frame pointer.
+ * These need to treated differently by tier 1 and 2. */
+
+#if TIER_ONE
+
+#define LOAD_IP() \
+do { next_instr = frame->prev_instr+1; } while (0)
+
+#define STORE_SP() \
+_PyFrame_SetStackPointer(frame, stack_pointer)
+
+#define LOAD_SP() \
+stack_pointer = _PyFrame_GetStackPointer(frame);
+
+#endif
+
+
+#if TIER_TWO
+
+#define LOAD_IP() \
+do { ip_offset = (_Py_CODEUNIT *)_PyFrame_GetCode(frame)->co_code_adaptive; } while (0)
+
+#define STORE_SP() \
+_PyFrame_SetStackPointer(frame, stack_pointer)
+
+#define LOAD_SP() \
+stack_pointer = _PyFrame_GetStackPointer(frame);
+
+#endif
+
+
+
+

Python/executor.c

@@ -17,6 +17,7 @@
 #include "pycore_sliceobject.h"
 #include "pycore_uops.h"
 
+#define TIER_TWO 2
 #include "ceval_macros.h"
 
 
@@ -83,7 +84,6 @@ _PyUopExecute(_PyExecutorObject *executor, _PyInterpreterFrame *frame, PyObject
         OBJECT_STAT_INC(optimization_uops_executed);
         switch (opcode) {
 
-#define TIER_TWO 2
 #include "executor_cases.c.h"
 
             default:

Python/executor_cases.c.h

@@ -369,8 +369,8 @@ def write_macro_instr(
             next_instr_is_set = write_components(parts, out, TIER_ONE, mac.cache_offset)
         except AssertionError as err:
             raise AssertionError(f"Error writing macro {mac.name}") from err
-        if not parts[-1].instr.always_exits and not next_instr_is_set:
-            if mac.cache_offset:
+        if not parts[-1].instr.always_exits:
+            if not next_instr_is_set and mac.cache_offset:
                 out.emit(f"next_instr += {mac.cache_offset};")
             out.emit("DISPATCH();")