Move lltrace into the frame struct

Author: Fidget-Spinner

Include/internal/pycore_frame.h

@@ -76,6 +76,9 @@ typedef struct _PyInterpreterFrame {
     uint16_t return_offset;  /* Only relevant during a function call */
     char owner;
     char visited;
+#ifdef Py_DEBUG
+    char lltrace;
+#endif
     /* Locals and stack */
     _PyStackRef localsplus[1];
 } _PyInterpreterFrame;

Python/bytecodes.c

@@ -4993,7 +4993,7 @@ dummy_func(
             _Py_CODEUNIT *target = _PyFrame_GetBytecode(frame) + exit->target;
         #if defined(Py_DEBUG) && !defined(_Py_JIT)
             OPT_HIST(trace_uop_execution_counter, trace_run_length_hist);
-            if (lltrace >= 2) {
+            if (FT_ATOMIC_LOAD_UINT8_RELAXED(frame->lltrace) >= 2) {
                 printf("SIDE EXIT: [UOp ");
                 _PyUOpPrint(&next_uop[-1]);
                 printf(", exit %u, temp %d, target %d -> %s]\n",
@@ -5111,7 +5111,7 @@ dummy_func(
             _Py_CODEUNIT *target = frame->instr_ptr;
         #if defined(Py_DEBUG) && !defined(_Py_JIT)
             OPT_HIST(trace_uop_execution_counter, trace_run_length_hist);
-            if (lltrace >= 2) {
+            if (FT_ATOMIC_LOAD_UINT8_RELAXED(frame->lltrace) >= 2) {
                 printf("DYNAMIC EXIT: [UOp ");
                 _PyUOpPrint(&next_uop[-1]);
                 printf(", exit %u, temp %d, target %d -> %s]\n",

Python/ceval.c

@@ -799,9 +799,6 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
 #endif
     uint8_t opcode;    /* Current opcode */
     int oparg;         /* Current opcode argument, if any */
-#ifdef LLTRACE
-    int lltrace = 0;
-#endif
 
     _PyInterpreterFrame  entry_frame;
 
@@ -821,6 +818,9 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
     entry_frame.owner = FRAME_OWNED_BY_CSTACK;
     entry_frame.visited = 0;
     entry_frame.return_offset = 0;
+#ifdef LLTRACE
+    entry_frame.lltrace = 0;
+#endif
     /* Push frame */
     entry_frame.previous = tstate->current_frame;
     frame->previous = &entry_frame;
@@ -880,9 +880,13 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
     stack_pointer = _PyFrame_GetStackPointer(frame);
 
 #ifdef LLTRACE
-    lltrace = maybe_lltrace_resume_frame(frame, &entry_frame, GLOBALS());
-    if (lltrace < 0) {
-        goto exit_unwind;
+    {
+        int lltrace = maybe_lltrace_resume_frame(frame, &entry_frame,
+                                                     GLOBALS());
+        FT_ATOMIC_STORE_UINT8_RELAXED(frame->lltrace, (uint8_t)lltrace);
+        if (lltrace < 0) {
+            goto exit_unwind;
+        }
     }
 #endif
 
@@ -1002,7 +1006,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
             }
             /* Resume normal execution */
 #ifdef LLTRACE
-            if (lltrace >= 5) {
+            if (FT_ATOMIC_LOAD_UINT8_RELAXED(frame->lltrace) >= 5) {
                 lltrace_resume_frame(frame);
             }
 #endif
@@ -1079,7 +1083,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
     for (;;) {
         uopcode = next_uop->opcode;
 #ifdef Py_DEBUG
-        if (lltrace >= 3) {
+        if (FT_ATOMIC_LOAD_UINT8_RELAXED(frame->lltrace) >= 3) {
             dump_stack(frame, stack_pointer);
             if (next_uop->opcode == _START_EXECUTOR) {
                 printf("%4d uop: ", 0);
@@ -1121,7 +1125,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
 
 jump_to_error_target:
 #ifdef Py_DEBUG
-    if (lltrace >= 2) {
+    if (FT_ATOMIC_LOAD_UINT8_RELAXED(frame->lltrace) >= 2) {
         printf("Error: [UOp ");
         _PyUOpPrint(&next_uop[-1]);
         printf(" @ %d -> %s]\n",
@@ -1157,7 +1161,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
     next_instr = next_uop[-1].target + _PyFrame_GetBytecode(frame);
 goto_to_tier1:
 #ifdef Py_DEBUG
-    if (lltrace >= 2) {
+    if (FT_ATOMIC_LOAD_UINT8_RELAXED(frame->lltrace) >= 2) {
         printf("DEOPT: [UOp ");
         _PyUOpPrint(&next_uop[-1]);
         printf(" -> %s]\n",

Python/ceval_macros.h

@@ -80,7 +80,7 @@
 
 /* PRE_DISPATCH_GOTO() does lltrace if enabled. Normally a no-op */
 #ifdef LLTRACE
-#define PRE_DISPATCH_GOTO() if (lltrace >= 5) { \
+#define PRE_DISPATCH_GOTO() if (FT_ATOMIC_LOAD_UINT8_RELAXED(frame->lltrace) >= 5) { \
     lltrace_instruction(frame, stack_pointer, next_instr, opcode, oparg); }
 #else
 #define PRE_DISPATCH_GOTO() ((void)0)
@@ -89,7 +89,8 @@
 #if LLTRACE
 #define LLTRACE_RESUME_FRAME() \
 do { \
-    lltrace = maybe_lltrace_resume_frame(frame, &entry_frame, GLOBALS()); \
+    int lltrace = maybe_lltrace_resume_frame(frame, &entry_frame, GLOBALS()); \
+    FT_ATOMIC_STORE_UINT8_RELAXED(frame->lltrace, (uint8_t)lltrace); \
     if (lltrace < 0) { \
         goto exit_unwind; \
     } \
@@ -153,7 +154,7 @@ GETITEM(PyObject *v, Py_ssize_t i) {
 /* The integer overflow is checked by an assertion below. */
 #define INSTR_OFFSET() ((int)(next_instr - _PyFrame_GetBytecode(frame)))
 #define NEXTOPARG()  do { \
-        _Py_CODEUNIT word  = {.cache = FT_ATOMIC_LOAD_UINT16_RELAXED(*(uint16_t*)next_instr)}; \
+        _Py_CODEUNIT word  = {.cache = FT_ATOMIC_LOAD_UINT8_RELAXED(*(uint16_t*)next_instr)}; \
         opcode = word.op.code; \
         oparg = word.op.arg; \
     } while (0)