8000 GH-118093: Better handling of short and mid-loop traces by brandtbucher · Pull Request #122252 · python/cpython · GitHub
[go: up one dir, main page]
Skip to content

GH-118093: Better handling of short and mid-loop traces #122252

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 10 commits into from
Jul 29, 2024
Merged

GH-118093: Better handling of short and mid-loop traces #122252

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
f8637ce
Allow one backward jump per trace, anywhere
brandtbucher Jul 17, 2024
f1f5d13
Use progress, rather than trace length, to reject traces
brandtbucher Jul 17, 2024
159e46b
Add missing RESERVEs
brandtbucher Jul 25, 2024
511ec09
Clean up overflow recovery
brandtbucher Jul 25, 2024
ba3d68b
Add asserts back and teach is_resume about ENTER_EXECUTOR
brandtbucher Jul 26, 2024
7c404a8
Guard _Py_GetExecutor on non-tier-two builds
brandtbucher Jul 26, 2024
1606abd
Revert "Guard _Py_GetExecutor on non-tier-two builds"
brandtbucher Jul 29, 2024
6901f60
Revert "Add asserts back and teach is_resume about ENTER_EXECUTOR"
brandtbucher Jul 29, 2024
9bcc643
Add comment to removed asserts
brandtbucher Jul 29, 2024
cd1f4d2
Catch up with main
brandtbucher Jul 29, 2024
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
5 changes: 3 additions & 2 deletions Objects/genobject.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -342,8 +342,9 @@ _PyGen_yf(PyGenObject *gen)
{
if (gen->gi_frame_state == FRAME_SUSPENDED_YIELD_FROM) {
_PyInterpreterFrame *frame = &gen->gi_iframe;
assert(is_resume(frame->instr_ptr));
assert((frame->instr_ptr->op.arg & RESUME_OPARG_LOCATION_MASK) >= RESUME_AFTER_YIELD_FROM);
// GH-122390: These asserts are wrong in the presence of ENTER_EXECUTOR!
// assert(is_resume(frame->instr_ptr));
// assert((frame->instr_ptr->op.arg & RESUME_OPARG_LOCATION_MASK) >= RESUME_AFTER_YIELD_FROM);
return PyStackRef_AsPyObjectNew(_PyFrame_StackPeek(frame));
}
return NULL;
Expand Down
64 changes: 31 additions & 33 deletions Python/optimizer.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -503,8 +503,7 @@ add_to_trace(
if (trace_stack_depth >= TRACE_STACK_SIZE) { \
DPRINTF(2, "Trace stack overflow\n"); \
OPT_STAT_INC(trace_stack_overflow); \
trace_length = 0; \
goto done; \
return 0; \
} \
assert(func == NULL || func->func_code == (PyObject *)code); \
trace_stack[trace_stack_depth].func = func; \
Expand Down Expand Up @@ -550,6 +549,7 @@ translate_bytecode_to_trace(
} trace_stack[TRACE_STACK_SIZE];
int trace_stack_depth = 0;
int confidence = CONFIDENCE_RANGE; // Adjusted by branch instructions
bool jump_seen = false;

#ifdef Py_DEBUG
char *python_lltrace = Py_GETENV("PYTHON_LLTRACE");
Expand All @@ -568,7 +568,6 @@ translate_bytecode_to_trace(
ADD_TO_TRACE(_START_EXECUTOR, 0, (uintptr_t)instr, INSTR_IP(instr, code));
uint32_t target = 0;

top: // Jump here after _PUSH_FRAME or likely branches
for (;;) {
target = INSTR_IP(instr, code);
// Need space for _DEOPT
Expand All @@ -577,6 +576,13 @@ translate_bytecode_to_trace(
uint32_t opcode = instr->op.code;
uint32_t oparg = instr->op.arg;

if (!progress_needed && instr == initial_instr) {
// We have looped around to the start:
RESERVE(1);
ADD_TO_TRACE(_JUMP_TO_TOP, 0, 0, 0);
goto done;
}

DPRINTF(2, "%d: %s(%d)\n", target, _PyOpcode_OpName[opcode], oparg);

if (opcode == ENTER_EXECUTOR) {
Expand All @@ -603,30 +609,21 @@ translate_bytecode_to_trace(
/* Special case the first instruction,
* so that we can guarantee forward progress */
if (progress_needed) {
progress_needed = false;
if (opcode == JUMP_BACKWARD || opcode == JUMP_BACKWARD_NO_INTERRUPT) {
instr += 1 + _PyOpcode_Caches[opcode] - (int32_t)oparg;
initial_instr = instr;
if (opcode == JUMP_BACKWARD) {
ADD_TO_TRACE(_TIER2_RESUME_CHECK, 0, 0, target);
}
continue;
}
else {
if (OPCODE_HAS_EXIT(opcode) || OPCODE_HAS_DEOPT(opcode)) {
opcode = _PyOpcode_Deopt[opcode];
}
assert(!OPCODE_HAS_EXIT(opcode));
assert(!OPCODE_HAS_DEOPT(opcode));
if (OPCODE_HAS_EXIT(opcode) || OPCODE_HAS_DEOPT(opcode)) {
opcode = _PyOpcode_Deopt[opcode];
}
assert(!OPCODE_HAS_EXIT(opcode));
assert(!OPCODE_HAS_DEOPT(opcode));
}

if (OPCODE_HAS_EXIT(opcode)) {
// Make space for exit code
// Make space for side exit and final _EXIT_TRACE:
RESERVE_RAW(2, "_EXIT_TRACE");
max_length--;
}
if (OPCODE_HAS_ERROR(opcode)) {
// Make space for error code
// Make space for error stub and final _EXIT_TRACE:
RESERVE_RAW(2, "_ERROR_POP_N");
max_length--;
}
switch (opcode) {
Expand Down Expand Up @@ -672,19 +669,18 @@ translate_bytecode_to_trace(
}

case JUMP_BACKWARD:
ADD_TO_TRACE(_CHECK_PERIODIC, 0, 0, target);
_Py_FALLTHROUGH;
case JUMP_BACKWARD_NO_INTERRUPT:
{
_Py_CODEUNIT *target = instr + 1 + _PyOpcode_Caches[opcode] - (int)oparg;
if (target == initial_instr) {
/* We have looped round to the start */
RESERVE(1);
ADD_TO_TRACE(_JUMP_TO_TOP, 0, 0, 0);
}
else {
instr += 1 + _PyOpcode_Caches[_PyOpcode_Deopt[opcode]] - (int)oparg;
if (jump_seen) {
OPT_STAT_INC(inner_loop);
DPRINTF(2, "JUMP_BACKWARD not to top ends trace\n");
goto done;
}
goto done;
jump_seen = true;
goto top;
}

case JUMP_FORWARD:
Expand Down Expand Up @@ -904,23 +900,25 @@ translate_bytecode_to_trace(
assert(instr->op.code == POP_TOP);
instr++;
}
top:
// Jump here after _PUSH_FRAME or likely branches.
progress_needed = false;
} // End for (;;)

done:
while (trace_stack_depth > 0) {
TRACE_STACK_POP();
}
assert(code == initial_code);
// Skip short traces like _SET_IP, LOAD_FAST, _SET_IP, _EXIT_TRACE
if (progress_needed || trace_length < 5) {
// Skip short traces where we can't even translate a single instruction:
if (progress_needed) {
OPT_STAT_INC(trace_too_short);
DPRINTF(2,
"No trace for %s (%s:%d) at byte offset %d (%s)\n",
"No trace for %s (%s:%d) at byte offset %d (no progress)\n",
PyUnicode_AsUTF8(code->co_qualname),
PyUnicode_AsUTF8(code->co_filename),
code->co_firstlineno,
2 * INSTR_IP(initial_instr, code),
progress_needed ? "no progress" : "too short");
2 * INSTR_IP(initial_instr, code));
return 0;
}
if (trace[trace_length-1].opcode != _JUMP_TO_TOP) {
Expand Down
Loading
0