Skip to content

Commit a8d9d94

Browse files
markshannonmarkshannon
authored
pythonGH-137959: Replace shim code in jitted code with a single trampoline function. (pythonGH-137961)
1 parent c056a08 commit a8d9d94

File tree

17 files changed

+166
-104
lines changed

17 files changed

+166
-104
lines changed

‎Include/internal/pycore_ceval.h‎

Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -123,6 +123,22 @@ _PyEval_EvalFrame(PyThreadState *tstate, _PyInterpreterFrame *frame, int throwfl
123123
return tstate->interp->eval_frame(tstate, frame, throwflag);
124124
}
125125

126+
#ifdef _Py_TIER2
127+
#ifdef _Py_JIT
128+
_Py_CODEUNIT *_Py_LazyJitTrampoline(
129+
struct _PyExecutorObject *current_executor, _PyInterpreterFrame *frame,
130+
_PyStackRef *stack_pointer, PyThreadState *tstate
131+
);
132+
#else
133+
_Py_CODEUNIT *_PyTier2Interpreter(
134+
struct _PyExecutorObject *current_executor, _PyInterpreterFrame *frame,
135+
_PyStackRef *stack_pointer, PyThreadState *tstate
136+
);
137+
#endif
138+
#endif
139+
140+
extern _PyJitEntryFuncPtr _Py_jit_entry;
141+
126142
extern PyObject*
127143
_PyEval_Vector(PyThreadState *tstate,
128144
PyFunctionObject *func, PyObject *locals,

‎Include/internal/pycore_interp_structs.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -765,6 +765,7 @@ struct _Py_unique_id_pool {
765765

766766
#endif
767767

768+
typedef _Py_CODEUNIT *(*_PyJitEntryFuncPtr)(struct _PyExecutorObject *exec, _PyInterpreterFrame *frame, _PyStackRef *stack_pointer, PyThreadState *tstate);
768769

769770
/* PyInterpreterState holds the global state for one of the runtime's
770771
interpreters. Typically the initial (main) interpreter is the only one.

‎Include/internal/pycore_optimizer.h‎

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -82,7 +82,6 @@ typedef struct _PyExecutorObject {
8282
uint32_t code_size;
8383
size_t jit_size;
8484
void *jit_code;
85-
void *jit_side_entry;
8685
_PyExitData exits[1];
8786
} _PyExecutorObject;
8887

‎Misc/NEWS.d/next/Core_and_Builtins/2025-08-19-16-07-07.gh-issue-137959.EWj0RZ.rst‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,2 @@
1+
Replace the shim code added to every piece of jitted code with a single
2+
trampoline function.

‎Python/bytecodes.c‎

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -2971,7 +2971,7 @@ dummy_func(
29712971
assert(tstate->current_executor == NULL);
29722972
assert(executor != tstate->interp->cold_executor);
29732973
tstate->jit_exit = NULL;
2974-
GOTO_TIER_TWO(executor);
2974+
TIER1_TO_TIER2(executor);
29752975
}
29762976
}
29772977
else {
@@ -3037,7 +3037,7 @@ dummy_func(
30373037
}
30383038
assert(executor != tstate->interp->cold_executor);
30393039
tstate->jit_exit = NULL;
3040-
GOTO_TIER_TWO(executor);
3040+
TIER1_TO_TIER2(executor);
30413041
#else
30423042
Py_FatalError("ENTER_EXECUTOR is not supported in this build");
30433043
#endif /* _Py_TIER2 */
@@ -5257,7 +5257,7 @@ dummy_func(
52575257
}
52585258
#endif
52595259
tstate->jit_exit = exit;
5260-
GOTO_TIER_TWO(exit->executor);
5260+
TIER2_TO_TIER2(exit->executor);
52615261
}
52625262

52635263
tier2 op(_CHECK_VALIDITY, (--)) {
@@ -5353,7 +5353,7 @@ dummy_func(
53535353

53545354
tier2 op(_START_EXECUTOR, (executor/4 --)) {
53555355
#ifndef _Py_JIT
5356-
current_executor = (_PyExecutorObject*)executor;
5356+
assert(current_executor == (_PyExecutorObject*)executor);
53575357
#endif
53585358
assert(tstate->jit_exit == NULL || tstate->jit_exit->executor == current_executor);
53595359
tstate->current_executor = (PyObject *)executor;
@@ -5434,7 +5434,7 @@ dummy_func(
54345434
}
54355435
assert(tstate->jit_exit == exit);
54365436
exit->executor = executor;
5437-
GOTO_TIER_TWO(exit->executor);
5437+
TIER2_TO_TIER2(exit->executor);
54385438
}
54395439

54405440
label(pop_2_error) {

‎Python/ceval.c‎

Lines changed: 35 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -275,7 +275,8 @@ maybe_lltrace_resume_frame(_PyInterpreterFrame *frame, PyObject *globals)
275275
}
276276
int r = PyDict_Contains(globals, &_Py_ID(__lltrace__));
277277
if (r < 0) {
278-
return -1;
278+
PyErr_Clear();
279+
return 0;
279280
}
280281
int lltrace = r * 5; // Levels 1-4 only trace uops
281282
if (!lltrace) {
@@ -1109,11 +1110,6 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
11091110
#endif
11101111
}
11111112

1112-
#if defined(_Py_TIER2) && !defined(_Py_JIT)
1113-
/* Tier 2 interpreter state */
1114-
_PyExecutorObject *current_executor = NULL;
1115-
const _PyUOpInstruction *next_uop = NULL;
1116-
#endif
11171113
#if Py_TAIL_CALL_INTERP
11181114
# if Py_STATS
11191115
return _TAIL_CALL_start_frame(frame, NULL, tstate, NULL, 0, lastopcode);
@@ -1126,14 +1122,41 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
11261122
#endif
11271123

11281124

1125+
early_exit:
1126+
assert(_PyErr_Occurred(tstate));
1127+
_Py_LeaveRecursiveCallPy(tstate);
1128+
assert(frame->owner != FRAME_OWNED_BY_INTERPRETER);
1129+
// GH-99729: We need to unlink the frame *before* clearing it:
1130+
_PyInterpreterFrame *dying = frame;
1131+
frame = tstate->current_frame = dying->previous;
1132+
_PyEval_FrameClearAndPop(tstate, dying);
1133+
frame->return_offset = 0;
1134+
assert(frame->owner == FRAME_OWNED_BY_INTERPRETER);
1135+
/* Restore previous frame and exit */
1136+
tstate->current_frame = frame->previous;
1137+
return NULL;
1138+
}
11291139
#ifdef _Py_TIER2
1130-
1131-
// Tier 2 is also here!
1132-
enter_tier_two:
1133-
11341140
#ifdef _Py_JIT
1135-
assert(0);
1141+
_PyJitEntryFuncPtr _Py_jit_entry = _Py_LazyJitTrampoline;
11361142
#else
1143+
_PyJitEntryFuncPtr _Py_jit_entry = _PyTier2Interpreter;
1144+
#endif
1145+
#endif
1146+
1147+
#if defined(_Py_TIER2) && !defined(_Py_JIT)
1148+
1149+
_Py_CODEUNIT *
1150+
_PyTier2Interpreter(
1151+
_PyExecutorObject *current_executor, _PyInterpreterFrame *frame,
1152+
_PyStackRef *stack_pointer, PyThreadState *tstate
1153+
) {
1154+
const _PyUOpInstruction *next_uop;
1155+
int oparg;
1156+
tier2_start:
1157+
1158+
next_uop = current_executor->trace;
1159+
assert(next_uop->opcode == _START_EXECUTOR || next_uop->opcode == _COLD_EXIT);
11371160

11381161
#undef LOAD_IP
11391162
#define LOAD_IP(UNUSED) (void)0
@@ -1151,7 +1174,6 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
11511174
#undef ENABLE_SPECIALIZATION_FT
11521175
#define ENABLE_SPECIALIZATION_FT 0
11531176

1154-
; // dummy statement after a label, before a declaration
11551177
uint16_t uopcode;
11561178
#ifdef Py_STATS
11571179
int lastuop = 0;
@@ -1225,24 +1247,9 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
12251247
next_uop = current_executor->trace + target;
12261248
goto tier2_dispatch;
12271249

1228-
#endif // _Py_JIT
1229-
1250+
}
12301251
#endif // _Py_TIER2
12311252

1232-
early_exit:
1233-
assert(_PyErr_Occurred(tstate));
1234-
_Py_LeaveRecursiveCallPy(tstate);
1235-
assert(frame->owner != FRAME_OWNED_BY_INTERPRETER);
1236-
// GH-99729: We need to unlink the frame *before* clearing it:
1237-
_PyInterpreterFrame *dying = frame;
1238-
frame = tstate->current_frame = dying->previous;
1239-
_PyEval_FrameClearAndPop(tstate, dying);
1240-
frame->return_offset = 0;
1241-
assert(frame->owner == FRAME_OWNED_BY_INTERPRETER);
1242-
/* Restore previous frame and exit */
1243-
tstate->current_frame = frame->previous;
1244-
return NULL;
1245-
}
12461253

12471254
#ifdef DO_NOT_OPTIMIZE_INTERP_LOOP
12481255
# pragma optimize("", on)

‎Python/ceval_macros.h‎

Lines changed: 9 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -133,9 +133,6 @@ do { \
133133
_PyFrame_SetStackPointer(frame, stack_pointer); \
134134
int lltrace = maybe_lltrace_resume_frame(frame, GLOBALS()); \
135135
stack_pointer = _PyFrame_GetStackPointer(frame); \
136-
if (lltrace < 0) { \
137-
JUMP_TO_LABEL(exit_unwind); \
138-
} \
139136
frame->lltrace = lltrace; \
140137
} while (0)
141138
#else
@@ -354,16 +351,10 @@ _PyFrame_SetStackPointer(frame, stack_pointer)
354351

355352
/* Tier-switching macros. */
356353

357-
#ifdef _Py_JIT
358-
#define GOTO_TIER_TWO(EXECUTOR) \
354+
#define TIER1_TO_TIER2(EXECUTOR) \
359355
do { \
360356
OPT_STAT_INC(traces_executed); \
361-
_PyExecutorObject *_executor = (EXECUTOR); \
362-
jit_func jitted = _executor->jit_code; \
363-
/* Keep the shim frame alive via the executor: */ \
364-
Py_INCREF(_executor); \
365-
next_instr = jitted(frame, stack_pointer, tstate); \
366-
Py_DECREF(_executor); \
357+
next_instr = _Py_jit_entry((EXECUTOR), frame, stack_pointer, tstate); \
367358
frame = tstate->current_frame; \
368359
stack_pointer = _PyFrame_GetStackPointer(frame); \
369360
if (next_instr == NULL) { \
@@ -372,31 +363,21 @@ do { \
372363
} \
373364
DISPATCH(); \
374365
} while (0)
375-
#else
376-
#define GOTO_TIER_TWO(EXECUTOR) \
377-
do { \
378-
OPT_STAT_INC(traces_executed); \
379-
_PyExecutorObject *_executor = (EXECUTOR); \
380-
next_uop = _executor->trace; \
381-
assert(next_uop->opcode == _START_EXECUTOR || next_uop->opcode == _COLD_EXIT); \
382-
goto enter_tier_two; \
366+
367+
#define TIER2_TO_TIER2(EXECUTOR) \
368+
do { \
369+
OPT_STAT_INC(traces_executed); \
370+
current_executor = (EXECUTOR); \
371+
goto tier2_start; \
383372
} while (0)
384-
#endif
385373

386374
#define GOTO_TIER_ONE(TARGET) \
387375
do \
388376
{ \
389377
tstate->current_executor = NULL; \
390-
next_instr = (TARGET); \
391378
OPT_HIST(trace_uop_execution_counter, trace_run_length_hist); \
392379
_PyFrame_SetStackPointer(frame, stack_pointer); \
393-
stack_pointer = _PyFrame_GetStackPointer(frame); \
394-
if (next_instr == NULL) \
395-
{ \
396-
next_instr = frame->instr_ptr; \
397-
goto error; \
398-
} \
399-
DISPATCH(); \
380+
return TARGET; \
400381
} while (0)
401382

402383
#define CURRENT_OPARG() (next_uop[-1].oparg)

‎Python/executor_cases.c.h‎

Lines changed: 3 additions & 3 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

‎Python/generated_cases.c.h‎

Lines changed: 2 additions & 2 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

0 commit comments

Comments
 (0)