Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

GH-133231: Changes to executor management to support proposed sys._jit module #133287

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 8 commits into from
May 4, 2025
Merged

GH-133231: Changes to executor management to support proposed sys._jit module #133287

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
3a2ccda
Track the current executor on the thread-state, not the previous one.…
markshannon May 2, 2025
46ef280
Handle tail-calling interpreter
markshannon May 2, 2025
01ecf18
Merge branch 'main' into current-executor
markshannon May 2, 2025
89e4668
Merge branch 'main' into current-executor
markshannon May 2, 2025
2f42cb0
Don't use entry.frame in tailcalling interpreter
markshannon May 3, 2025
f938d68
Address review comments
markshannon May 3, 2025
a30e444
Fix compiler warning
markshannon May 3, 2025
0b9dcb7
Remove redundant code
markshannon May 4, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion 2 Include/cpython/pystate.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -194,7 +194,7 @@ struct _ts {
/* The thread's exception stack entry. (Always the last entry.) */
_PyErr_StackItem exc_state;

PyObject *previous_executor;
PyObject *current_executor;

uint64_t dict_global_version;

Expand Down
2 changes: 2 additions & 0 deletions 2 Include/internal/pycore_interp_structs.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -923,6 +923,8 @@ struct _is {
PyObject *common_consts[NUM_COMMON_CONSTANTS];
bool jit;
struct _PyExecutorObject *executor_list_head;
struct _PyExecutorObject *executor_deletion_list_head;
int executor_deletion_list_remaining_capacity;
size_t trace_run_counter;
_rare_events rare_events;
PyDict_WatchCallback builtins_dict_watcher;
Expand Down
2 changes: 1 addition & 1 deletion 2 Include/internal/pycore_opcode_metadata.h
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

9 changes: 8 additions & 1 deletion 9 Include/internal/pycore_optimizer.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,7 +69,7 @@ typedef struct {
typedef struct {
uint32_t target;
_Py_BackoffCounter temperature;
const struct _PyExecutorObject *executor;
struct _PyExecutorObject *executor;
} _PyExitData;

typedef struct _PyExecutorObject {
Expand All @@ -84,6 +84,10 @@ typedef struct _PyExecutorObject {
_PyExitData exits[1];
} _PyExecutorObject;

/* If pending deletion list gets large enough, then scan,
* and free any executors that aren't executing
* i.e. any that aren't a thread's current_executor. */
#define EXECUTOR_DELETE_LIST_MAX 100

// Export for '_opcode' shared extension (JIT compiler).
PyAPI_FUNC(_PyExecutorObject*) _Py_GetExecutor(PyCodeObject *code, int offset);
Expand Down Expand Up @@ -304,6 +308,9 @@ static inline int is_terminator(const _PyUOpInstruction *uop)
}

PyAPI_FUNC(int) _PyDumpExecutors(FILE *out);
#ifdef _Py_TIER2
extern void _Py_ClearExecutorDeletionList(PyInterpreterState *interp);
#endif

#ifdef __cplusplus
}
Expand Down
2 changes: 1 addition & 1 deletion 2 Include/internal/pycore_uop_metadata.h
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

34 changes: 24 additions & 10 deletions 34 Python/bytecodes.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1169,6 +1169,17 @@ dummy_func(
tstate->current_frame = frame->previous;
assert(!_PyErr_Occurred(tstate));
PyObject *result = PyStackRef_AsPyObjectSteal(retval);
#if !Py_TAIL_CALL_INTERP
assert(frame == &entry.frame);
#endif
#ifdef _Py_TIER2
_PyStackRef executor = frame->localsplus[0];
assert(tstate->current_executor == NULL);
if (!PyStackRef_IsNull(executor)) {
tstate->current_executor = PyStackRef_AsPyObjectBorrow(executor);
PyStackRef_CLOSE(executor);
}
#endif
LLTRACE_RESUME_FRAME();
return result;
}
Expand Down Expand Up @@ -2912,8 +2923,7 @@ dummy_func(
}
else {
this_instr[1].counter = initial_jump_backoff_counter();
assert(tstate->previous_executor == NULL);
tstate->previous_executor = Py_None;
assert(tstate->current_executor == NULL);
GOTO_TIER_TWO(executor);
brandtbucher marked this conversation as resolved.
Show resolved Hide resolved
}
}
Expand Down Expand Up @@ -2965,7 +2975,7 @@ dummy_func(
assert(executor->vm_data.index == INSTR_OFFSET() - 1);
assert(executor->vm_data.code == code);
assert(executor->vm_data.valid);
assert(tstate->previous_executor == NULL);
assert(tstate->current_executor == NULL);
/* If the eval breaker is set then stay in tier 1.
* This avoids any potentially infinite loops
* involving _RESUME_CHECK */
Expand All @@ -2978,8 +2988,6 @@ dummy_func(
}
DISPATCH_GOTO();
}
tstate->previous_executor = Py_None;
Py_INCREF(executor);
GOTO_TIER_TWO(executor);
#else
Py_FatalError("ENTER_EXECUTOR is not supported in this build");
Expand Down Expand Up @@ -5254,7 +5262,6 @@ dummy_func(
exit->temperature = initial_temperature_backoff_counter();
Py_CLEAR(exit->executor);
}
tstate->previous_executor = (PyObject *)current_executor;
if (exit->executor == NULL) {
_Py_BackoffCounter temperature = exit->temperature;
if (!backoff_counter_triggers(temperature)) {
Expand All @@ -5277,7 +5284,6 @@ dummy_func(
}
exit->executor = executor;
}
Py_INCREF(exit->executor);
GOTO_TIER_TWO(exit->executor);
}

Expand Down Expand Up @@ -5316,7 +5322,6 @@ dummy_func(
}

tier2 op(_START_EXECUTOR, (executor/4 --)) {
Copy link
Member

@brandtbucher brandtbucher May 3, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Side note, this instruction now exists solely to set the current_executor in the tier two interpreter. We should try to remove it as a follow-up (doesn't need to be in this PR).

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We can replace all uses of current_executor with an operand, which will remove the current_executor local variable from the tier 2 interpreter and the _JIT_EXECUTOR patch from the JIT template.

We can also shrink _CHECK_VALIDITY a bit by replacing current_executor->vm_data.valid with *validity_ptr:

tier2 op(_CHECK_VALIDITY, (validity_ptr/4 --)) {
    DEOPT_IF(*validity_ptr == 0);
}

For another PR.

Py_CLEAR(tstate->previous_executor);
#ifndef _Py_JIT
current_executor = (_PyExecutorObject*)executor;
#endif
Expand All @@ -5337,12 +5342,10 @@ dummy_func(
}

tier2 op(_DEOPT, (--)) {
tstate->previous_executor = (PyObject *)current_executor;
GOTO_TIER_ONE(_PyFrame_GetBytecode(frame) + CURRENT_TARGET());
}

tier2 op(_ERROR_POP_N, (target/2 --)) {
tstate->previous_executor = (PyObject *)current_executor;
assert(oparg == 0);
frame->instr_ptr = _PyFrame_GetBytecode(frame) + target;
SYNC_SP();
Expand Down Expand Up @@ -5463,6 +5466,17 @@ dummy_func(
if (frame->owner == FRAME_OWNED_BY_INTERPRETER) {
/* Restore previous frame and exit */
tstate->current_frame = frame->previous;
#if !Py_TAIL_CALL_INTERP
assert(frame == &entry.frame);
#endif
#ifdef _Py_TIER2
_PyStackRef executor = frame->localsplus[0];
assert(tstate->current_executor == NULL);
if (!PyStackRef_IsNull(executor)) {
tstate->current_executor = PyStackRef_AsPyObjectBorrow(executor);
PyStackRef_CLOSE(executor);
}
#endif
return NULL;
}
next_instr = frame->instr_ptr;
Expand Down
50 changes: 31 additions & 19 deletions 50 Python/ceval.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -990,6 +990,11 @@ _PyObjectArray_Free(PyObject **array, PyObject **scratch)
#define DONT_SLP_VECTORIZE
#endif

typedef struct {
_PyInterpreterFrame frame;
_PyStackRef stack[1];
} _PyEntryFrame;
Comment on lines +993 to +996
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this needed? We only need one slot in localsplus, which _PyInterpreterFrame already has.

Copy link
Member Author

@markshannon markshannon May 3, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We now need two slots. One on the stack for the return value, as before, and one local variable to hold the current executor.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I see. Forgot about the return value.

Maybe add a comment that the one stack slot is for the return value, and the one local is for the executor.


PyObject* _Py_HOT_FUNCTION DONT_SLP_VECTORIZE
_PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int throwflag)
{
Expand All @@ -1009,7 +1014,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
int oparg; /* Current opcode argument, if any */
assert(tstate->current_frame == NULL || tstate->current_frame->stackpointer != NULL);
#endif
_PyInterpreterFrame entry_frame;
_PyEntryFrame entry;

if (_Py_EnterRecursiveCallTstate(tstate, "")) {
assert(frame->owner != FRAME_OWNED_BY_INTERPRETER);
Expand All @@ -1021,30 +1026,37 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
* These are cached values from the frame and code object. */
_Py_CODEUNIT *next_instr;
_PyStackRef *stack_pointer;
entry_frame.localsplus[0] = PyStackRef_NULL;
entry.stack[0] = PyStackRef_NULL;
brandtbucher marked this conversation as resolved.
Show resolved Hide resolved
#ifdef Py_STACKREF_DEBUG
entry_frame.f_funcobj = PyStackRef_None;
entry.frame.f_funcobj = PyStackRef_None;
#elif defined(Py_DEBUG)
/* Set these to invalid but identifiable values for debugging. */
entry_frame.f_funcobj = (_PyStackRef){.bits = 0xaaa0};
entry_frame.f_locals = (PyObject*)0xaaa1;
entry_frame.frame_obj = (PyFrameObject*)0xaaa2;
entry_frame.f_globals = (PyObject*)0xaaa3;
entry_frame.f_builtins = (PyObject*)0xaaa4;
entry.frame.f_funcobj = (_PyStackRef){.bits = 0xaaa0};
entry.frame.f_locals = (PyObject*)0xaaa1;
entry.frame.frame_obj = (PyFrameObject*)0xaaa2;
entry.frame.f_globals = (PyObject*)0xaaa3;
entry.frame.f_builtins = (PyObject*)0xaaa4;
#endif
entry_frame.f_executable = PyStackRef_None;
entry_frame.instr_ptr = (_Py_CODEUNIT *)_Py_INTERPRETER_TRAMPOLINE_INSTRUCTIONS + 1;
entry_frame.stackpointer = entry_frame.localsplus;
entry_frame.owner = FRAME_OWNED_BY_INTERPRETER;
entry_frame.visited = 0;
entry_frame.return_offset = 0;
entry.frame.f_executable = PyStackRef_None;
entry.frame.instr_ptr = (_Py_CODEUNIT *)_Py_INTERPRETER_TRAMPOLINE_INSTRUCTIONS + 1;
entry.frame.stackpointer = entry.stack;
entry.frame.owner = FRAME_OWNED_BY_INTERPRETER;
entry.frame.visited = 0;
entry.frame.return_offset = 0;
#ifdef Py_DEBUG
entry_frame.lltrace = 0;
entry.frame.lltrace = 0;
#endif
/* Push frame */
entry_frame.previous = tstate->current_frame;
frame->previous = &entry_frame;
entry.frame.previous = tstate->current_frame;
frame->previous = &entry.frame;
tstate->current_frame = frame;
entry.frame.localsplus[0] = PyStackRef_NULL;
brandtbucher marked this conversation as resolved.
Show resolved Hide resolved
#ifdef _Py_TIER2
if (tstate->current_executor != NULL) {
entry.frame.localsplus[0] = PyStackRef_FromPyObjectNew(tstate->current_executor);
tstate->current_executor = NULL;
}
#endif

/* support for generator.throw() */
if (throwflag) {
Expand All @@ -1071,9 +1083,9 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
stack_pointer = _PyFrame_GetStackPointer(frame);
#if Py_TAIL_CALL_INTERP
# if Py_STATS
return _TAIL_CALL_error(frame, stack_pointer, tstate, next_instr, 0, lastopcode);
return _TAIL_CALL_error(frame, stack_pointer, tstate, next_instr, 0, lastopcode);
# else
return _TAIL_CALL_error(frame, stack_pointer, tstate, next_instr, 0);
return _TAIL_CALL_error(frame, stack_pointer, tstate, next_instr, 0);
# endif
#else
goto error;
Expand Down
9 changes: 6 additions & 3 deletions 9 Python/ceval_macros.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -359,12 +359,12 @@ _PyFrame_SetStackPointer(frame, stack_pointer)
do { \
OPT_STAT_INC(traces_executed); \
_PyExecutorObject *_executor = (EXECUTOR); \
tstate->current_executor = (PyObject *)_executor; \
jit_func jitted = _executor->jit_code; \
/* Keep the shim frame alive via the executor: */ \
Py_INCREF(_executor); \
next_instr = jitted(frame, stack_pointer, tstate); \
Py_DECREF(_executor); \
Py_CLEAR(tstate->previous_executor); \
frame = tstate->current_frame; \
stack_pointer = _PyFrame_GetStackPointer(frame); \
if (next_instr == NULL) { \
Expand All @@ -377,7 +377,9 @@ do { \
#define GOTO_TIER_TWO(EXECUTOR) \
do { \
OPT_STAT_INC(traces_executed); \
next_uop = (EXECUTOR)->trace; \
_PyExecutorObject *_executor = (EXECUTOR); \
tstate->current_executor = (PyObject *)_executor; \
next_uop = _executor->trace; \
assert(next_uop->opcode == _START_EXECUTOR); \
goto enter_tier_two; \
} while (0)
Expand All @@ -386,10 +388,11 @@ do { \
#define GOTO_TIER_ONE(TARGET) \
do \
{ \
tstate->current_executor = NULL; \
next_instr = (TARGET); \
assert(tstate->current_executor == NULL); \
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Minor: this seems redundant, given that we're setting it one line above.

OPT_HIST(trace_uop_execution_counter, trace_run_length_hist); \
_PyFrame_SetStackPointer(frame, stack_pointer); \
Py_CLEAR(tstate->previous_executor); \
stack_pointer = _PyFrame_GetStackPointer(frame); \
if (next_instr == NULL) \
{ \
Expand Down
7 changes: 0 additions & 7 deletions 7 Python/executor_cases.c.h
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

Loading
Loading
Morty Proxy This is a proxified and sanitized view of the page, visit original site.