Skip to content

gh-131798: JIT: Further optimize _CALL_ISINSTANCE for class tuples #134543

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.

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

Open
wants to merge 9 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
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
143 changes: 140 additions & 3 deletions Lib/test/test_capi/test_opt.py
Original file line number Diff line number Diff line change
Expand Up @@ -2191,17 +2191,154 @@ def testfunc(n):
self.assertNotIn("_TO_BOOL_BOOL", uops)
self.assertIn("_GUARD_IS_TRUE_POP", uops)

def test_call_isinstance_tuple_of_classes(self):
def test_call_isinstance_tuple_of_classes_is_true(self):
def testfunc(n):
x = 0
for _ in range(n):
# A tuple of classes is currently not optimized,
# so this is only narrowed to bool:
y = isinstance(42, (int, str))
if y:
x += 1
return x

res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
self.assertIsNotNone(ex)
uops = get_opnames(ex)
self.assertNotIn("_CALL_ISINSTANCE", uops)
self.assertNotIn("_TO_BOOL_BOOL", uops)
self.assertNotIn("_GUARD_IS_TRUE_POP", uops)
self.assertIn("_BUILD_TUPLE", uops)
self.assertIn("_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW", uops)
Comment on lines +2210 to +2211
Copy link
Member Author

Choose a reason for hiding this comment

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

_BUILD_TUPLE is preventing us from optimizing out _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW.
The bytecode is basically:

LOAD_CONST
LOAD_CONST
_BUILD_TUPLE
_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW

To optimize this, we'd need some special handling for _BUILD_TUPLE in remove_unneeded_uops.

Copy link
Member

Choose a reason for hiding this comment

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

Got it, might be worth looking into next if you're up for it! Could be tricky, though.


def test_call_isinstance_tuple_of_classes_is_false(self):
def testfunc(n):
x = 0
for _ in range(n):
y = isinstance(42, (bool, str))
if not y:
x += 1
return x

res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
self.assertIsNotNone(ex)
uops = get_opnames(ex)
self.assertNotIn("_CALL_ISINSTANCE", uops)
self.assertNotIn("_TO_BOOL_BOOL", uops)
self.assertNotIn("_GUARD_IS_FALSE_POP", uops)
self.assertIn("_BUILD_TUPLE", uops)
self.assertIn("_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW", uops)

def test_call_isinstance_tuple_of_classes_true_unknown_1(self):
def testfunc(n):
x = 0
for _ in range(n):
# One of the classes is unknown, but it comes
# after a known class, so we can narrow to True and
# remove the isinstance call.
y = isinstance(42, (int, eval('str')))
if y:
x += 1
return x

res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
self.assertIsNotNone(ex)
uops = get_opnames(ex)
self.assertNotIn("_CALL_ISINSTANCE", uops)
self.assertNotIn("_TO_BOOL_BOOL", uops)
self.assertNotIn("_GUARD_IS_TRUE_POP", uops)
self.assertIn("_BUILD_TUPLE", uops)
self.assertIn("_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW", uops)

def test_call_isinstance_tuple_of_classes_true_unknown_2(self):
def testfunc(n):
x = 0
for _ in range(n):
# We can narrow to True, but since the unknown class comes
# first and could potentially trigger an __instancecheck__,
# we can't remove the isinstance call.
y = isinstance(42, (eval('str'), int))
Copy link
Member

Choose a reason for hiding this comment

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

I'm confused, why can't we narrow to True? We can't remove the call, but the result is known.

Copy link
Member Author

Choose a reason for hiding this comment

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

Yep, that was a brainfart 😄 I somehow conflated narrowing to True/False with replacing the op, but we can obviously just narrow without removing the call as you said!

if y:
x += 1
return x

res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
self.assertIsNotNone(ex)
uops = get_opnames(ex)
self.assertIn("_CALL_ISINSTANCE", uops)
self.assertNotIn("_TO_BOOL_BOOL", uops)
self.assertNotIn("_GUARD_IS_TRUE_POP", uops)

def test_call_isinstance_tuple_of_classes_true_unknown_3(self):
def testfunc(n):
x = 0
for _ in range(n):
# We can only narrow to bool here
y = isinstance(42, (str, eval('int')))
if y:
x += 1
return x

res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
self.assertIsNotNone(ex)
uops = get_opnames(ex)
self.assertIn("_CALL_ISINSTANCE", uops)
self.assertNotIn("_TO_BOOL_BOOL", uops)
self.assertIn("_GUARD_IS_TRUE_POP", uops)

def test_call_isinstance_tuple_of_classes_true_unknown_4(self):
def testfunc(n):
x = 0
for _ in range(n):
# We can only narrow to bool here
y = isinstance(42, (eval('int'), str))
if y:
x += 1
return x

res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
self.assertIsNotNone(ex)
uops = get_opnames(ex)
self.assertIn("_CALL_ISINSTANCE", uops)
self.assertNotIn("_TO_BOOL_BOOL", uops)
self.assertIn("_GUARD_IS_TRUE_POP", uops)

def test_call_isinstance_empty_tuple(self):
def testfunc(n):
x = 0
for _ in range(n):
y = isinstance(42, ())
Copy link
Member

Choose a reason for hiding this comment

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

I had to pause and think for a second to figure out what this would even do. Nice edge case.

if not y:
x += 1
return x

res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
self.assertIsNotNone(ex)
uops = get_opnames(ex)
self.assertNotIn("_CALL_ISINSTANCE", uops)
self.assertNotIn("_TO_BOOL_BOOL", uops)
self.assertNotIn("_GUARD_IS_FALSE_POP", uops)
self.assertNotIn("_POP_TOP_LOAD_CONST_INLINE_BORROW", uops)
self.assertNotIn("_POP_CALL_LOAD_CONST_INLINE_BORROW", uops)
self.assertNotIn("_POP_CALL_ONE_LOAD_CONST_INLINE_BORROW", uops)
self.assertNotIn("_POP_CALL_TWO_LOAD_CONST_INLINE_BORROW", uops)

def test_call_isinstance_tuple_unknown_length(self):
def testfunc(n):
x = 0
for _ in range(n):
# tuple with an unknown length, we only narrow to bool
tup = tuple(eval('(int, str)'))
y = isinstance(42, tup)
if y:
x += 1
return x

res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
self.assertIsNotNone(ex)
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
Optimize ``_CALL_ISINSTANCE`` in the JIT when the second argument is a tuple
of classes.
40 changes: 39 additions & 1 deletion Python/optimizer_bytecodes.c
Original file line number Diff line number Diff line change
Expand Up @@ -956,14 +956,52 @@ dummy_func(void) {
// isinstance(inst, cls) where both inst and cls have
// known types, meaning we can deduce either True or False

// The below check is equivalent to PyObject_TypeCheck(inst, cls)
PyObject *out = Py_False;
if (inst_type == cls_o || PyType_IsSubtype(inst_type, cls_o)) {
out = Py_True;
}
sym_set_const(res, out);
REPLACE_OP(this_instr, _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)out);
}
else if (inst_type && sym_matches_type(cls, &PyTuple_Type)) {
// isinstance(inst, tup) where inst has a known type and tup is a tuple.
// We can deduce True if inst is an instance of at least one of
// the items in the tuple.
// We can deduce False if all items in the tuple have known types and
// inst is not an instance of any of them.

int length = sym_tuple_length(cls);
if (length != -1) {
// We cannot do anything about tuples with unknown length
bool can_replace_op = true;
PyObject *out = Py_False;
for (int i = 0; i < length; i++) {
JitOptRef item = sym_tuple_getitem(ctx, cls, i);
if (!sym_has_type(item)) {
// There is an unknown item in the tuple.
// It could potentially define its own __instancecheck__
// so it is no longer possible to replace the op with a const load.
out = NULL;
can_replace_op = false;
continue;
}
PyTypeObject *cls_o = (PyTypeObject *)sym_get_const(ctx, item);
if (cls_o &&
sym_matches_type(item, &PyType_Type) &&
Copy link
Member

Choose a reason for hiding this comment

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

Maybe add a comment explaining that this is to protect against metaclasses definine __instancecheck__.

Copy link
Member Author

Choose a reason for hiding this comment

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

How would you formulate it? I don't think of it as specifically a guard for __instancecheck__ but basically PyObject_TypeCheck adapted to the JIT optimizer.

Copy link
Member

Choose a reason for hiding this comment

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

We're not only checking that the object is a subclass of type, we're also checking that it is an exact instance of type itself. We care about this second condition because it guarantees that __instancecheck__ doesn't exist (otherwise we would need to look it up to check if it exists or not).

(inst_type == cls_o || PyType_IsSubtype(inst_type, cls_o)))
{
out = Py_True;
break;
}
}
if (out) {
sym_set_const(res, out);
if (can_replace_op) {
REPLACE_OP(this_instr, _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)out);
}
}
}
}
}

op(_GUARD_IS_TRUE_POP, (flag -- )) {
Expand Down
29 changes: 29 additions & 0 deletions Python/optimizer_cases.c.h

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

Loading
pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy