gh-131798: JIT: Narrow the return type of `isinstance` for some known arguments by tomasr8 · Pull Request #133172 · python/cpython · GitHub

In this PR:

narrows isintance(obj, cls) to True if obj is a known type and cls is a known class and obj is a subclass of cls (and vice versa for False)
in all other cases narrows isinstance to bool.

Brandt also suggested adding an optimization for tuples which I'd like to add in a followup in order to keep the sizes of the individual PRs smaller. Though if you prefer to have it in one PR I can do it as well :)

Issue: Better uop coverage in the JIT optimizer #131798

This simulates PyObject_TypeCheck

Can you add a comment to that effect? :)

Added!

I've seen this guard used elsewhere with self_or_null but it's not clear to me whether it is needed here as well?

Yep. It's just a (weird) way of saying "we know whether it's NULL or not". Maybe it's worth adding a helper function (in another PR) to make it clearer, since the meaning is super subtle.

Cool! Can you also add a test where the second isinstance arg is a class with a metaclass defining __instancecheck__?

class EvenNumberMeta(type):
    def __instancecheck__(self, number):
        return not number % 2

class EvenNumber(metaclass=EvenNumberMeta):
    pass

# Optimizer only narrows to bool, runtime value is True:
even = isinstance(42, EvenNumber)

This is pretty fragile: the class attr lookup is cached in the bytecode, and I actually have a branch I'll open a PR with soon that teaches the optimizer to read these caches.

So instead, let's use everyone's favorite optimization-breaker:

Suggested change

      
                        class Foo:
          
                            bar = 42
          
                        x = 0
          
                        for _ in range(n):
          
                            # we only know bar (LOAD_ATTR) is not null (set via sym_new_not_null)
          
                            bar = Foo.bar
          
                            # This will only narrow to bool and not to True due to 'bar' having
          
                            # unknown (non-null) type
          
                            y = isinstance(bar, int)
          
                            if y:
          
                                x += 1
          
                        return x
          
                        x = 0
          
                        for _ in range(n):
          
                            # The optimizer doesn't know the return type here:
          
                            bar = eval("42")
          
                            # This will only narrow to bool:
          
                            y = isinstance(bar, int)
          
                            if y:
          
                                x += 1
          
                        return x

Nice trick to use eval! I updated the test :)

Yep. It's just a (weird) way of saying "we know whether it's NULL or not". Maybe it's worth adding a helper function (in another PR) to make it clearer, since the meaning is super subtle.

Can you add a comment to that effect? :)

You can avoid all of this repetition by doing an unconditional res = sym_new_type(ctx, &PyBool_Type); at the top, and using sym_set_const(ctx, ...) to narrow it when possible.

Yup that is way better, updated!

A Python core developer has requested some changes be made to your pull request before we can consider merging it. If you could please address their requests along with any other requests in other reviews from core developers that would be appreciated.

Once you have made the requested changes, please leave a comment on this pull request containing the phrase I have made the requested changes; please review again. I will then notify any core developers who have left a review that you're ready for them to take another look at this pull request.

(Marking as draft until #133339 is merged which will let us simplify the oparg logic)

I think I addressed all your points. I also added the test you suggested. I'm planning to add support for tuples (e.g. isinstance(foo, (int, str)) in a followup so for now it just supports single types.

Thanks! Just one suggestion, then we can land it.

Co-authored-by: Brandt Bucher <brandtbucher@gmail.com>

This LGTM. A side note: on PyPy, they can actually narrow isinstance(x, cls) to subtype of cls or even better. I wonder if we can do that? Not sure if it's safe to do so though due to subclassing rules and all that.

Yeah, could be cool. It's not really useful to us right now, since all of our guards are against exact types/versions, not many subclass checks.

CI failures are unrelated.

…arguments in the JIT (pythonGH-133172)

tomasr8 added 3 commits April 29, 2025 23:25

Optimize CALL_ISINSTANCE

59b37f7

Add news entry

a81c02e

Optimize for subclasses as well

b1eb6a0

tomasr8 requested review from Fidget-Spinner and markshannon as code owners April 29, 2025 21:35

bedevere-app Bot added the awaiting review label Apr 29, 2025

bedevere-app Bot mentioned this pull request Apr 29, 2025

Better uop coverage in the JIT optimizer #131798

Open

tomasr8 commented Apr 29, 2025

View reviewed changes

brandtbucher requested changes May 2, 2025

View reviewed changes

bedevere-app Bot added awaiting changes and removed awaiting review labels May 2, 2025

tomasr8 mentioned this pull request May 3, 2025

gh-131798: JIT: Split CALL_ISINSTANCE into several uops #133339

Merged

tomasr8 marked this pull request as draft May 3, 2025 11:57

bedevere-app Bot removed the awaiting changes label May 3, 2025

tomasr8 added 10 commits May 8, 2025 23:36

Merge branch 'main' into jit-isinstance

c42172d

Regen cases

770f7ed

Add a comment

3e86810

Merge remote-tracking branch 'upstream/main' into jit-isinstance

0820a3c

Simplify

60d21fa

Add a metaclass test

38370a3

Improve test

bb228eb

Update comment

0ede9f3

Mark some stackrefs as unused

3d0df4c

Simplify even more

ce0cd38

tomasr8 marked this pull request as ready for review May 9, 2025 20:46

bedevere-app Bot added the awaiting review label May 9, 2025

tomasr8 requested a review from brandtbucher May 9, 2025 20:47

brandtbucher self-assigned this May 19, 2025

brandtbucher added the sprint label May 19, 2025

github-project-automation Bot added this to Sprint May 19, 2025

github-project-automation Bot moved this to Todo in Sprint May 19, 2025

brandtbucher approved these changes May 19, 2025

View reviewed changes

Comment thread Python/optimizer_bytecodes.c Outdated

github-project-automation Bot moved this from Todo to In Progress in Sprint May 19, 2025

bedevere-app Bot added awaiting merge and removed awaiting review labels May 19, 2025

tomasr8 and others added 3 commits May 19, 2025 15:37

Simplify code

20fd185

Co-authored-by: Brandt Bucher <brandtbucher@gmail.com>

Merge branch 'main' into jit-isinstance

62854fe

make regen-cases

f985963

Merge branch 'main' into jit-isinstance

27b1b43

brandtbucher merged commit 8d490b3 into python:main May 19, 2025

bedevere-app Bot removed the awaiting merge label May 19, 2025

github-project-automation Bot moved this from In Progress to Done in Sprint May 19, 2025

tomasr8 deleted the jit-isinstance branch May 19, 2025 17:34

Pranjal095 pushed a commit to Pranjal095/cpython that referenced this pull request Jul 12, 2025

pythonGH-131798 : Narrow the return type of isinstance for some known …

0f96049

…arguments in the JIT (pythonGH-133172)

taegyunkim pushed a commit to taegyunkim/cpython that referenced this pull request Aug 4, 2025

pythonGH-131798 : Narrow the return type of isinstance for some known …

d9707f4

…arguments in the JIT (pythonGH-133172)

+                                  // isinstance(obj, cls) where both obj and cls have known types
+                                  // We can deduce either True or False
+                                  PyTypeObject *inst_type = sym_get_type(inst_sym);
+                                  if (sym_matches_type(inst_sym, cls) || PyType_IsSubtype(inst_type, cls)) {

+                                  // isinstance(obj, cls) where both obj and cls have known types
+                                  // We can deduce either True or False
+                                  PyTypeObject *inst_type = sym_get_type(inst_sym);
+                                  if (sym_matches_type(inst_sym, cls) || PyType_IsSubtype(inst_type, cls)) {

                   }
+                  op(_CALL_ISINSTANCE, (callable, self_or_null, args[oparg] -- res)) {
+                      if (sym_is_null(self_or_null) || sym_is_not_null(self_or_null)) {

+                              }
+                              else {
+                                  // isinstance(obj, cls) where obj has unknown type
+                                  res = sym_new_type(ctx, &PyBool_Type);
+                              }
+                          }
+                          else {
+                              // isinstance(obj, cls) where cls has unknown type
+                              res = sym_new_type(ctx, &PyBool_Type);
+                          }
+                      }
+                      else {
+                          res = sym_new_type(ctx, &PyBool_Type);
+                      }
+                  }

gh-131798: JIT: Narrow the return type of `isinstance` for some known arguments by tomasr8 · Pull Request #133172 · python/cpython · GitHub

Uh oh!

Uh oh!

gh-131798: JIT: Narrow the return type of isinstance for some known arguments - #133172

gh-131798: JIT: Narrow the return type of isinstance for some known arguments#133172brandtbucher merged 17 commits intopython:mainpython/cpython:mainfrom tomasr8:jit-isinstancetomasr8/cpython:jit-isinstanceCopy head branch name to clipboard

Conversation

tomasr8 commented Apr 29, 2025 • edited by bedevere-app Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

brandtbucher left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bedevere-app Bot commented May 2, 2025

Uh oh!

tomasr8 commented May 3, 2025

Uh oh!

tomasr8 commented May 9, 2025

Uh oh!

brandtbucher left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Fidget-Spinner commented May 19, 2025

Uh oh!

brandtbucher commented May 19, 2025

Uh oh!

brandtbucher commented May 19, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

gh-131798: JIT: Narrow the return type of isinstance for some known arguments#133172
brandtbucher merged 17 commits into
python:mainfrom
tomasr8:jit-isinstance

tomasr8 commented Apr 29, 2025 •
edited by bedevere-app Bot
Loading