Skip to content
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

Finish up LOAD_ATTR specialisation #100288

Open
1 of 2 tasks
Fidget-Spinner opened this issue Dec 16, 2022 · 15 comments
Open
1 of 2 tasks

Finish up LOAD_ATTR specialisation #100288

Fidget-Spinner opened this issue Dec 16, 2022 · 15 comments
Labels
interpreter-core (Objects, Python, Grammar, and Parser dirs) performance Performance or resource usage type-feature A feature request or enhancement

Comments

@Fidget-Spinner
Copy link
Member

Fidget-Spinner commented Dec 16, 2022

We should target the following specialisation failures:

  • has managed dict
  • not in dict (how does this even happen??)

With these two we can hit >90% specialisation successes.
If we're feeling really ambitious, we could aim for "not managed dict" failure too. But we don't need that to achieve >90% successes.

I'm doing the first one. Is anyone interested in investigating the second specialisation failure?

Linked PRs

@Fidget-Spinner Fidget-Spinner added the type-feature A feature request or enhancement label Dec 16, 2022
@Fidget-Spinner Fidget-Spinner added the performance Performance or resource usage label Dec 16, 2022
@corona10
Copy link
Member

Oh I am interested in if you don't mind?

@Fidget-Spinner
Copy link
Member Author

Sure! Go ahead!

@Fidget-Spinner
Copy link
Member Author

CC @markshannon just these two added specialisations should net us 10% all specialisation failures/successes, bringing us to over 90% specialisation successes.

@Fidget-Spinner
Copy link
Member Author

@corona10 seems like the stats were misleading, not in dict is actually the following:

class C: 
    i = 0

c = C()
c.i # this fails to specialise

So it's a class attribute lookup, which makes more sense.

@corona10
Copy link
Member

Thanks I read the PR too: #100295

@markshannon
Copy link
Member

Regarding "not managed dict" failures, I think the approach should be to reduce the number of objects that don't have managed dicts, not specialize for them.

We can tweak the management of cached keys, and do some static analysis in the compiler to do that.

Fidget-Spinner added a commit to Fidget-Spinner/cpython that referenced this issue Dec 23, 2022
iritkatriel added a commit to iritkatriel/cpython that referenced this issue Dec 28, 2022
* Correct CVE-2020-10735 documentation (python#100306)

* pythongh-94912: Added marker for non-standard coroutine function detection (python#99247)

This introduces a new decorator `@inspect.markcoroutinefunction`,
which, applied to a sync function, makes it appear async to
`inspect.iscoroutinefunction()`.

* Docs: Don't upload CI artifacts (python#100330)

* pythongh-89727: Fix os.walk RecursionError on deep trees (python#99803)

Use a stack to implement os.walk iteratively instead of recursively to
avoid hitting recursion limits on deeply nested trees.

* pythongh-69929: re docs: Add more specific definition of \w (python#92015)

Co-authored-by: Jelle Zijlstra <jelle.zijlstra@gmail.com>

* pythongh-89051: Add ssl.OP_LEGACY_SERVER_CONNECT (python#93927)

Co-authored-by: blurb-it[bot] <43283697+blurb-it[bot]@users.noreply.github.com>
Co-authored-by: Christian Heimes <christian@python.org>
Co-authored-by: Hugo van Kemenade <hugovk@users.noreply.github.com>
Fixes python#89051

* pythongh-88211: Change lower-case and upper-case to match recommendations in imaplib docs (python#99625)

* pythongh-100348: Fix ref cycle in `asyncio._SelectorSocketTransport` with `_read_ready_cb` (python#100349)

* pythongh-99925: Fix inconsistency in `json.dumps()` error messages (pythonGH-99926)

* Clarify that every thread has its own default context in contextvars (python#99246)

* pythongh-99576: Fix cookiejar file that was not truncated for some classes (pythonGH-99616)

Co-authored-by: Łukasz Langa <lukasz@langa.pl>

* pythongh-100188: Reduce misses in BINARY_SUBSCR_(LIST/TUPLE)_INT (python#100189)

Don't specialize if the index is negative.

* pythongh-99991: improve docs on str.encode and bytes.decode (python#100198)

Co-authored-by: C.A.M. Gerlach <CAM.Gerlach@Gerlach.CAM>

* pythongh-91081: Add note on WeakKeyDictionary behavior when deleting a replaced entry (python#91499)

Co-authored-by: Pieter Eendebak <P.T.eendebak@tudelft.nl>
Co-authored-by: Jelle Zijlstra <jelle.zijlstra@gmail.com>

* pythongh-85267: Improvements to inspect.signature __text_signature__ handling (python#98796)

This makes a couple related changes to inspect.signature's behaviour
when parsing a signature from `__text_signature__`.

First, `inspect.signature` is documented as only raising ValueError or
TypeError. However, in some cases, we could raise RuntimeError.  This PR
changes that, thereby fixing python#83685.

(Note that the new ValueErrors in RewriteSymbolics are caught and then
reraised with a message)

Second, `inspect.signature` could randomly drop parameters that it
didn't understand (corresponding to `return None` in the `p` function).
This is the core issue in python#85267. I think this is very surprising
behaviour and it seems better to fail outright.

Third, adding this new failure broke a couple tests. To fix them (and to
e.g. allow `inspect.signature(select.epoll.register)` as in python#85267), I
add constant folding of a couple binary operations to RewriteSymbolics.

(There's some discussion of making signature expression evaluation
arbitrary powerful in python#68155. I think that's out of scope. The
additional constant folding here is pretty straightforward, useful, and
not much of a slippery slope)

Fourth, while python#85267 is incorrect about the cause of the issue, it turns
out if you had consecutive newlines in __text_signature__, you'd get
`tokenize.TokenError`.

Finally, the `if name is invalid:` code path was dead, since
`parse_name` never returned `invalid`.

* pythonGH-100363: Speed up `asyncio.get_running_loop` (python#100364)

* pythonGH-100133: fix `asyncio` subprocess losing `stderr` and `stdout` output (python#100154)

* pythongh-100374: Fixed a bug in socket.getfqdn() (pythongh-100375)

* pythongh-100129: Add tests for pickling all builtin types and functions (pythonGH-100142)

* Remove unused variable from `dis._find_imports` (python#100396)

* pythongh-78878: Fix crash when creating an instance of `_ctypes.CField` (python#14837)

* pythonGH-69564: Clarify use of octal format of mode argument in help(os.chmod) (python#20621)

Co-authored-by: Kumar Aditya <59607654+kumaraditya303@users.noreply.github.com>

* pythonGH-99554: Pack location tables more effectively (pythonGH-99556)

* Correct typo in typing.py (python#100423)

In the docstring of `ParamSpec`, the name of `P = ParamSpec('P')` was
mistakenly written as `'T'`.

* pythongh-99761: Add `_PyLong_IsPositiveSingleDigit` function to check for single digit integers  (python#100064)

* pythonGH-99770: Make the correct call specialization fail kind show up in the stats (pythonGH-99771)

* pythongh-78997: fix bad rebase of moved test file (python#100424)

* pythongh-100344: Add C implementation for `asyncio.current_task` (python#100345)

Co-authored-by: pranavtbhat

* pythonGH-99554: Trim trailing whitespace (pythonGH-100435)



Automerge-Triggered-By: GH:brandtbucher

* pythongh-85432: Harmonise parameter names between C and pure-Python implementations of `datetime.time.strftime`, `datetime.datetime.fromtimestamp` (python#99993)

* pythongh-57762: fix misleading tkinter.Tk docstring (python#98837)

Mentioned as a desired change by terryjreedy on the corresponding issue,
since Tk is not a subclass of Toplevel.

* pythongh-48496: Added example and link to faq for UnboundLocalError in reference (python#93068)

* Fix typo in 3.12 What's New (python#100449)

* pythongh-76963: PEP3118 itemsize of an empty ctypes array should not be 0 (pythonGH-5576)

The itemsize returned in a memoryview of a ctypes array is now computed from the item type, instead of dividing the total size by the length and assuming that the length is not zero.

* pythonGH-100459: fix copy-paste errors in specialization stats (pythonGH-100460)

* pythongh-99110: Initialize `frame->previous` in init_frame to fix segmentation fault when accessing `frame.f_back` (python#100182)

* pythongh-98712: Clarify "readonly bytes-like object" semantics in C arg-parsing docs (python#98710)

* pythongh-92216: improve performance of `hasattr` for type objects (pythonGH-99979)

* pythongh-100288: Specialise LOAD_ATTR_METHOD for managed dictionaries (pythonGH-100289)

* Revert "pythongh-100288: Specialise LOAD_ATTR_METHOD for managed dictionaries (pythonGH-100289)" (python#100468)

This reverts commit c3c7848.

* pythongh-94155: Reduce hash collisions for code objects (python#100183)

* Uses a better hashing algorithm to get better dispersion and remove commutativity.

* Incorporates `co_firstlineno`, `Py_SIZE(co)`, and bytecode instructions.

* This is now the entire set of criteria used in `code_richcompare`, except for `_PyCode_ConstantKey` (which would incorporate the types of `co_consts` rather than just their values).

* pythongh-83076: 3.8x speed improvement in (Async)Mock instantiation (python#100252)

* pythongh-99482: remove `jython` compatibility parts from stdlib and tests (python#99484)

* bpo-40447: accept all path-like objects in compileall.compile_file (python#19883)

Signed-off-by: Filipe Laíns <lains@archlinux.org>
Signed-off-by: Filipe Laíns <lains@riseup.net>
Co-authored-by: Irit Katriel <1055913+iritkatriel@users.noreply.github.com>
Co-authored-by: Shantanu <12621235+hauntsaninja@users.noreply.github.com>

* pythonGH-100425: Improve accuracy of builtin sum() for float inputs (pythonGH-100426)

* pythongh-68320, pythongh-88302 - Allow for private `pathlib.Path` subclassing (pythonGH-31691)

Users may wish to define subclasses of `pathlib.Path` to add or modify
existing methods. Before this change, attempting to instantiate a subclass
raised an exception like:

    AttributeError: type object 'PPath' has no attribute '_flavour'

Previously the `_flavour` attribute was assigned as follows:

    PurePath._flavour        = xxx not set!! xxx
    PurePosixPath._flavour   = _PosixFlavour()
    PureWindowsPath._flavour = _WindowsFlavour()

This change replaces it with a `_pathmod` attribute, set as follows:

    PurePath._pathmod        = os.path
    PurePosixPath._pathmod   = posixpath
    PureWindowsPath._pathmod = ntpath

Functionality from `_PosixFlavour` and `_WindowsFlavour` is moved into
`PurePath` as underscored-prefixed classmethods. Flavours are removed.

Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
Co-authored-by: Brett Cannon <brett@python.org>
Co-authored-by: Adam Turner <9087854+AA-Turner@users.noreply.github.com>
Co-authored-by: Eryk Sun <eryksun@gmail.com>

* pythongh-99947: Ensure unreported errors are chained for SystemError during import (pythonGH-99946)

* Add "strict" to dotproduct(). Add docstring. Factor-out common code. (pythonGH-100480)

* pythongh-94808: improve test coverage of number formatting (python#99472)

* pythongh-100454: Start running SSL tests with OpenSSL 3.1.0-beta1 (python#100456)

* pythongh-100268: Add is_integer method to int (python#100439)

This improves the lives of type annotation users of `float` - which type checkers implicitly treat as `int|float` because that is what most code actually wants. Before this change a `.is_integer()` method could not be assumed to exist on things annotated as `: float` due to the method not existing on both types.

* pythongh-77771: Add enterabs example in sched (python#92716)

Co-authored-by: Shantanu <12621235+hauntsaninja@users.noreply.github.com>

* pythonGH-91166: Implement zero copy writes for `SelectorSocketTransport` in asyncio (python#31871)

Co-authored-by: Guido van Rossum <gvanrossum@gmail.com>

* pythonGH-91166: Implement zero copy writes for `SelectorSocketTransport` in asyncio (python#31871)

Co-authored-by: Guido van Rossum <gvanrossum@gmail.com>

* Misc Itertools recipe tweaks (pythonGH-100493)

* pythongh-100357: Convert several functions in `bltinsmodule` to AC (python#100358)

* Remove wrong comment about `repr` in `test_unicode` (python#100495)

* pythongh-99908: Tutorial: Modernize the 'data-record class' example (python#100499)

Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>

* pythongh-100474: Fix handling of dirs named index.html in http.server (pythonGH-100475)



If you had a directory called index.html or index.htm within a directory, it would cause http.server to return a 404 Not Found error instead of the directory listing. This came about due to not checking that the index was a regular file.

I have also added a test case for this situation.

Automerge-Triggered-By: GH:merwok

* pythongh-100287: Fix unittest.mock.seal with AsyncMock (python#100496)

* pythongh-99535: Add test for inheritance of annotations and update documentation (python#99990)

* pythongh-100428: Make float documentation more accurate (python#100437)

Previously, the grammar did not accept `float("10")`.
Also implement mdickinson's suggestion of removing the indirection.

* [Minor PR] Quotes in documentation changed into code blocks (python#99536)

Minor formatting fix in documentation

Co-authored-by: Shantanu <12621235+hauntsaninja@users.noreply.github.com>

* pythongh-100472: Fix docs claim that compileall parameters could be bytes (python#100473)

* pythongh-100519: simplification to `eff_request_host` in cookiejar.py (python#99588)

`IPV4_RE` includes a `.`, and the `.find(".") == -1` included here is already testing to make sure there's no dot, so this part of the expression is tautological. Instead use more modern `in` syntax to make it clear what the check is doing here. The simplified implementation more clearly matches the wording in RFC 2965.

Co-authored-by: hauntsaninja <hauntsaninja@gmail.com>

* pythongh-99308: Clarify re docs for byte pattern group names (python#99311)

* pythongh-92446: Improve argparse choices docs; revert bad change to lzma docs (python#94627)

Based on the definition of the collections.abc classes, it is more accurate to use "sequence" instead of "container" when describing argparse choices.

A previous attempt at fixing this in python#92450 was mistaken; this PR reverts that change.

Co-authored-by: Shantanu <12621235+hauntsaninja@users.noreply.github.com>

* Fix name of removed `inspect.Signature.from_builtin` method in 3.11.0a2 changelog (python#100525)

* pythongh-100520: Fix `rst` markup in `configparser`  docstrings (python#100524)

* pythongh-99509: Add `__class_getitem__` to `multiprocessing.queues.Queue` (python#99511)

* pythongh-94603: micro optimize list.pop (pythongh-94604)

* Remove `NoneType` redefinition from `clinic.py` (python#100551)

* pythongh-100553: Improve accuracy of sqlite3.Row iter test (python#100555)

* pythonGH-98831: Modernize a ton of simpler instructions (python#100545)

* load_const and load_fast aren't families for now
* Don't decref unmoved names
* Modernize GET_ANEXT
* Modernize GET_AWAITABLE
* Modernize ASYNC_GEN_WRAP
* Modernize YIELD_VALUE
* Modernize POP_EXCEPT (in more than one way)
* Modernize PREP_RERAISE_STAR
* Modernize LOAD_ASSERTION_ERROR
* Modernize LOAD_BUILD_CLASS
* Modernize STORE_NAME
* Modernize LOAD_NAME
* Modernize LOAD_CLASSDEREF
* Modernize LOAD_DEREF
* Modernize STORE_DEREF
* Modernize COPY_FREE_VARS (mark it as done)
* Modernize LIST_TO_TUPLE
* Modernize LIST_EXTEND
* Modernize SET_UPDATE
* Modernize SETUP_ANNOTATIONS
* Modernize DICT_UPDATE
* Modernize DICT_MERGE
* Modernize MAP_ADD
* Modernize IS_OP
* Modernize CONTAINS_OP
* Modernize CHECK_EXC_MATCH
* Modernize IMPORT_NAME
* Modernize IMPORT_STAR
* Modernize IMPORT_FROM
* Modernize JUMP_FORWARD (mark it as done)
* Modernize JUMP_BACKWARD (mark it as done)

Signed-off-by: Filipe Laíns <lains@archlinux.org>
Signed-off-by: Filipe Laíns <lains@riseup.net>
Co-authored-by: Jeremy Paige <ucodery@gmail.com>
Co-authored-by: Carlton Gibson <carlton@noumenal.es>
Co-authored-by: Hugo van Kemenade <hugovk@users.noreply.github.com>
Co-authored-by: Jon Burdo <jon@jonburdo.com>
Co-authored-by: Stanley <46876382+slateny@users.noreply.github.com>
Co-authored-by: Jelle Zijlstra <jelle.zijlstra@gmail.com>
Co-authored-by: Thomas Grainger <tagrain@gmail.com>
Co-authored-by: Brad Wolfe <brad.wolfe@gmail.com>
Co-authored-by: Richard Kojedzinszky <rkojedzinszky@users.noreply.github.com>
Co-authored-by: František Nesveda <fnesveda@users.noreply.github.com>
Co-authored-by: Pablo Galindo Salgado <Pablogsal@gmail.com>
Co-authored-by: Nikita Sobolev <mail@sobolevn.me>
Co-authored-by: Łukasz Langa <lukasz@langa.pl>
Co-authored-by: Dennis Sweeney <36520290+sweeneyde@users.noreply.github.com>
Co-authored-by: Bisola Olasehinde <horlasehinde@gmail.com>
Co-authored-by: C.A.M. Gerlach <CAM.Gerlach@Gerlach.CAM>
Co-authored-by: Pieter Eendebak <P.T.eendebak@tudelft.nl>
Co-authored-by: Shantanu <12621235+hauntsaninja@users.noreply.github.com>
Co-authored-by: Kumar Aditya <59607654+kumaraditya303@users.noreply.github.com>
Co-authored-by: Dominic Socular <BBH@awsl.rip>
Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
Co-authored-by: Hai Shi <shihai1992@gmail.com>
Co-authored-by: amaajemyfren <32741226+amaajemyfren@users.noreply.github.com>
Co-authored-by: Brandt Bucher <brandtbucher@microsoft.com>
Co-authored-by: david-why <david_why@outlook.com>
Co-authored-by: Pieter Eendebak <pieter.eendebak@gmail.com>
Co-authored-by: penguin_wwy <940375606@qq.com>
Co-authored-by: Eli Schwartz <eschwartz93@gmail.com>
Co-authored-by: Itamar Ostricher <itamarost@gmail.com>
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
Co-authored-by: Eric Wieser <wieser.eric@gmail.com>
Co-authored-by: Irit Katriel <1055913+iritkatriel@users.noreply.github.com>
Co-authored-by: Bill Fisher <william.w.fisher@gmail.com>
Co-authored-by: Petr Viktorin <encukou@gmail.com>
Co-authored-by: Ken Jin <kenjin@python.org>
Co-authored-by: Carl Meyer <carl@oddbird.net>
Co-authored-by: Filipe Laíns <lains@riseup.net>
Co-authored-by: Raymond Hettinger <rhettinger@users.noreply.github.com>
Co-authored-by: Barney Gale <barney.gale@gmail.com>
Co-authored-by: Brett Cannon <brett@python.org>
Co-authored-by: Adam Turner <9087854+AA-Turner@users.noreply.github.com>
Co-authored-by: Eryk Sun <eryksun@gmail.com>
Co-authored-by: Sebastian Berg <sebastianb@nvidia.com>
Co-authored-by: Illia Volochii <illia.volochii@gmail.com>
Co-authored-by: JosephSBoyle <48555120+JosephSBoyle@users.noreply.github.com>
Co-authored-by: James Frost <git@frost.cx>
Co-authored-by: MonadChains <monadchains@gmail.com>
Co-authored-by: Bart Broere <mail@bartbroere.eu>
Co-authored-by: Glyph <code@glyph.im>
Co-authored-by: hauntsaninja <hauntsaninja@gmail.com>
Co-authored-by: Ilya Kulakov <kulakov.ilya@gmail.com>
Co-authored-by: Guy Yagev <yourlefthandman8@gmail.com>
Co-authored-by: Jakub Kuczys <me@jacken.men>
@Fidget-Spinner
Copy link
Member Author

Has managed dict isn't worth specialising for - Mark's stats show it has a 40% miss rate and comparing with lats week's stats, it increased the hit rate from 82.0% to 82.5% only.

@Fidget-Spinner
Copy link
Member Author

I have a suspicion we are just about at the point of diminishing returns for LOAD_ATTR. We still have Dong-hee's specialisation for class attributes (that might also be not worth the effort, but it's good to try).

It might not be possible to get LOAD_ATTR hit rates to 90% due to natural polymorphism/dynamism in attribute loads, and I think that's perfectly acceptable.

@markshannon
Copy link
Member

The top remaining failures look like this:

What count fraction
class attr simple 579,637 35.4%
has managed dict 577,900 35.3%
not managed dict 258,867 15.8%

"class attr simple" is stuff like

class C:
     a = 1
c = C()
c.a

We should be able to handle this fairly easily.

I think the way to deal with the second and third cases is to ensure more objects have values arrays and do not have materialized dicts, not more specializations.

We can still get the specialization rates up to ~95%

carljm added a commit to carljm/cpython that referenced this issue Jan 5, 2023
* main:
  pythonGH-100288: Remove LOAD_ATTR_METHOD_WITH_DICT instruction. (pythonGH-100753)
  pythonGH-100766: Note that locale.LC_MESSAGES is not universal (pythonGH-100702)
carljm added a commit to carljm/cpython that referenced this issue Jan 5, 2023
* main:
  pythonGH-100288: Remove LOAD_ATTR_METHOD_WITH_DICT instruction. (pythonGH-100753)
  pythonGH-100766: Note that locale.LC_MESSAGES is not universal (pythonGH-100702)
  Drop myself from pathlib maintenance (python#100757)
  pythongh-100739: Respect mock spec when checking for unsafe prefixes (python#100740)
@brandtbucher
Copy link
Member

I think the remaining cases are just too polymorphic.

For example, adding LOAD_ATTR_CLASS_FROM_INSTANCE to handle the "class attr simple" case barely moves the LOAD_ATTR hit rate (the results are "1.00x faster", and the miss rate for the new opcode is ~32%). We can add it if we want, but I'm just not sure it's worth it.

@markshannon
Copy link
Member

Don't forget that additional specializations will provide longer term benefits, not just speed ups in the base interpreter.
More specializations mean better type feedback and potentially longer traces, providing larger regions for optimization for higher tiers.

We should push specialization not just to the point of diminishing returns in terms of speed, but to the edge of negative returns.

I don't think it is worth making the interpreter slower to get more type information, at least not yet, but as long as we aren't slowing things down we should push specialization as far as we can.

The miss rate of 32% seems high, it might be worth a bit of investigation. It might well be polymorphism, but it might be a bug somewhere.
Could you make a (maybe draft) PR, as it sounds like you already have working code?

@brandtbucher
Copy link
Member

Here you go: #101379.

@brandtbucher
Copy link
Member

Another observation that may help us with the "materialized, managed dict" cases:

I suspect that the vast majority of instance __dict__s are created to be read, or lightly written. Until they overflow the shared keys size or gain a non-unicode key, they still share keys with all of the other virtual dicts (and real __dict__s) for all other instances of the class.

We can probably take advantage of this, right? If we check that dict->ma_keys == heap_type->ht_cached_keys, then its attribute loads can still work just like LOAD_ATTR_INSTANCE_VALUE. The only difference is that we get the values from _PyDictOrValues_GetDict(dorv)->ma_values instead of _PyDictOrValues_GetValues(dorv).

Just an observation I had while looking at this stuff. Seems like it could simplify LOAD_ATTR_WITH_HINT and STORE_ATTR_WITH_HINT quite a bit.

Other notes on LOAD_ATTR_WITH_HINT:

  • I'm surprised that we handle non-unicode managed __dict__s here. It seems like it makes way more sense to just deopt in that case.
  • I think it will always fail (res = ep->me_value; DEOPT_IF(res == NULL, LOAD_ATTR)) for split tables... which is the common case, right?

@brandtbucher
Copy link
Member

I think it will always fail (res = ep->me_value; DEOPT_IF(res == NULL, LOAD_ATTR)) for split tables... which is the common case, right?

...though it looks like it only has a 2.5% miss rate, so I could be missing something here.

@markshannon
Copy link
Member

There is some co-design with object layout to consider as well.

  • All objects with dicts should use the new layout. Rather than fix LOAD_ATTR to handle them, we should fix the classes.
  • With that done, LOAD_ATTR_WITH_HINT and STORE_ATTR_WITH_HINT can be removed
  • We can't remove uses of __dict__ in Python code, so we will need the extra instruction you propose.

Can we change object layout a bit, so that a single instruction can handle objects with and without a __dict__?

I don't know of a way to do this without slowing down the common case, but it something to bear in mind.

carljm added a commit to carljm/cpython that referenced this issue Jan 31, 2023
* main:
  pythonGH-100288: Skip extra work when failing to specialize LOAD_ATTR (pythonGH-101354)
  pythongh-101409: Improve generated clinic code for self type checks (python#101411)
  pythongh-98831: rewrite BEFORE_ASYNC_WITH and END_ASYNC_FOR in the instruction definition DSL (python#101458)
  pythongh-101469: Optimise get_io_state() by using _PyModule_GetState() (pythonGH-101470)
markshannon added a commit that referenced this issue Jul 10, 2023
@iritkatriel iritkatriel added the interpreter-core (Objects, Python, Grammar, and Parser dirs) label Nov 28, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
interpreter-core (Objects, Python, Grammar, and Parser dirs) performance Performance or resource usage type-feature A feature request or enhancement
Projects
None yet
Development

No branches or pull requests

5 participants