gh-139757: Add BINARY_OP_SUBSCR_USTR_INT by chris-eibl · Pull Request #143389 · python/cpython

chris-eibl · 2026-01-03T11:51:24Z

Since #140800BINARY_OP_SUBSCR_STR_INT only specializes for compact ASCII strings. Let's introduce BINARY_OP_SUBSCR_USTR_INT to specialize again for reading an ASCII character from any string.

Issue: JIT assembly optimizer leaves some redundant branches #139757

passes the tests and brings bm_tomli back to normal, i.e. 15 % speedup

Fidget-Spinner · 2026-01-03T12:15:32Z

I verified a 9% speedup on this PR for the tomli_loads benchmark on pyperformance:

Mean +- std dev: [spec_off] 2.24 sec +- 0.03 sec -> [spec_on] 2.05 sec +- 0.02 sec: 1.09x faster

Apparently, we regressed earlier in https://github.com/python/cpython/pull/140800/files. This caused the 10% slowdown in tomli loads. Which seemingly uses a lot of unicode strings.

I'm going to merge this PR to remove the perf regression.

Fidget-Spinner

Just need tests and two comments. Thanks!

Fidget-Spinner · 2026-01-03T12:25:14Z

 res = PyStackRef_FromPyObjectBorrow(res_o);
 }

+ macro(BINARY_OP_SUBSCR_NCSTR_INT) =


When you have the time, please change then name to BINARY_OP_SUBSCR_USTR_INT.

Fidget-Spinner · 2026-01-03T12:25:32Z

 }

+ macro(BINARY_OP_SUBSCR_NCSTR_INT) =
+ _GUARD_TOS_INT + _GUARD_NOS_UNICODE + unused/5 + _BINARY_OP_SUBSCR_NCSTR_INT + _POP_TOP_INT + POP_TOP;


Is POP_TOP_UNICODE instead of POP_TOP safe here?

and just assert in the bytecodes

chris-eibl · 2026-01-04T10:18:02Z

 self.assertNotIn("_GUARD_TOS_UNICODE", uops)
 self.assertIn("_BINARY_OP_ADD_UNICODE", uops)

+ def test_binary_subcsr_ustr_int_narrows_to_str(self):


I thought of paramterizing test_binary_subcsr_str_int_narrows_to_str and test_binary_op_subscr_str_int to get rid of duplication, but it resulted in worse readable code, because the input and the expectations have to be varied ...

Fidget-Spinner · 2026-01-04T10:19:10Z

Sorry for leaving 3 separate review comments instead of one review. I only picked up on some of these while looking over them again.

Fidget-Spinner

LGTM. Thanks.

Co-authored-by: Chris Eibl <138194463+chris-eibl@users.noreply.github.com>

bedevere-bot · 2026-01-04T15:04:54Z

⚠️⚠️⚠️ Buildbot failure ⚠️⚠️⚠️

Hi! The buildbot ARM64 macOS 3.x (tier-2) has failed when building commit e6bfe4d.

What do you need to do:

Don't panic.
Check the buildbot page in the devguide if you don't know what the buildbots are or how they work.
Go to the page of the buildbot that failed (https://buildbot.python.org/#/builders/725/builds/12456) and take a look at the build logs.
Check if the failure is related to this commit (e6bfe4d) or if it is a false positive.
If the failure is related to this commit, please, reflect that on the issue and make a new Pull Request with a fix.

You can take a look at the buildbot page here:

https://buildbot.python.org/#/builders/725/builds/12456

Failed tests:

test_urllib2net

Summary of the results of the build (if available):

==

Click to see traceback logs

remote: Enumerating objects: 54, done. remote: Counting objects: 1% (1/54) remote: Counting objects: 3% (2/54) remote: Counting objects: 5% (3/54) remote: Counting objects: 7% (4/54) remote: Counting objects: 9% (5/54) remote: Counting objects: 11% (6/54) remote: Counting objects: 12% (7/54) remote: Counting objects: 14% (8/54) remote: Counting objects: 16% (9/54) remote: Counting objects: 18% (10/54) remote: Counting objects: 20% (11/54) remote: Counting objects: 22% (12/54) remote: Counting objects: 24% (13/54) remote: Counting objects: 25% (14/54) remote: Counting objects: 27% (15/54) remote: Counting objects: 29% (16/54) remote: Counting objects: 31% (17/54) remote: Counting objects: 33% (18/54) remote: Counting objects: 35% (19/54) remote: Counting objects: 37% (20/54) remote: Counting objects: 38% (21/54) remote: Counting objects: 40% (22/54) remote: Counting objects: 42% (23/54) remote: Counting objects: 44% (24/54) remote: Counting objects: 46% (25/54) remote: Counting objects: 48% (26/54) remote: Counting objects: 50% (27/54) remote: Counting objects: 51% (28/54) remote: Counting objects: 53% (29/54) remote: Counting objects: 55% (30/54) remote: Counting objects: 57% (31/54) remote: Counting objects: 59% (32/54) remote: Counting objects: 61% (33/54) remote: Counting objects: 62% (34/54) remote: Counting objects: 64% (35/54) remote: Counting objects: 66% (36/54) remote: Counting objects: 68% (37/54) remote: Counting objects: 70% (38/54) remote: Counting objects: 72% (39/54) remote: Counting objects: 74% (40/54) remote: Counting objects: 75% (41/54) remote: Counting objects: 77% (42/54) remote: Counting objects: 79% (43/54) remote: Counting objects: 81% (44/54) remote: Counting objects: 83% (45/54) remote: Counting objects: 85% (46/54) remote: Counting objects: 87% (47/54) remote: Counting objects: 88% (48/54) remote: Counting objects: 90% (49/54) remote: Counting objects: 92% (50/54) remote: Counting objects: 94% (51/54) remote: Counting objects: 96% (52/54) remote: Counting objects: 98% (53/54) remote: Counting objects: 100% (54/54) remote: Counting objects: 100% (54/54), done. remote: Compressing objects: 3% (1/27) remote: Compressing objects: 7% (2/27) remote: Compressing objects: 11% (3/27) remote: Compressing objects: 14% (4/27) remote: Compressing objects: 18% (5/27) remote: Compressing objects: 22% (6/27) remote: Compressing objects: 25% (7/27) remote: Compressing objects: 29% (8/27) remote: Compressing objects: 33% (9/27) remote: Compressing objects: 37% (10/27) remote: Compressing objects: 40% (11/27) remote: Compressing objects: 44% (12/27) remote: Compressing objects: 48% (13/27) remote: Compressing objects: 51% (14/27) remote: Compressing objects: 55% (15/27) remote: Compressing objects: 59% (16/27) remote: Compressing objects: 62% (17/27) remote: Compressing objects: 66% (18/27) remote: Compressing objects: 70% (19/27) remote: Compressing objects: 74% (20/27) remote: Compressing objects: 77% (21/27) remote: Compressing objects: 81% (22/27) remote: Compressing objects: 85% (23/27) remote: Compressing objects: 88% (24/27) remote: Compressing objects: 92% (25/27) remote: Compressing objects: 96% (26/27) remote: Compressing objects: 100% (27/27) remote: Compressing objects: 100% (27/27), done. remote: Total 28 (delta 26), reused 2 (delta 1), pack-reused 0 (from 0) 
From https://github.com/python/cpython
* branch main -> FETCH_HEAD
Note: switching to 'e6bfe4d8869e046a91d091611d3c7b5dccdaf0d6'.
You are in 'detached HEAD' state. You can look around, make experimental
changes and commit them, and you can discard any commits you make in this
state without impacting any branches by switching back to a branch.
If you want to create a new branch to retain commits you create, you may
do so (now or later) by using -c with the switch command. Example:
git switch -c <new-branch-name>
Or undo this operation with:
git switch -
Turn off this advice by setting config variable advice.detachedHead to false
HEAD is now at e6bfe4d8869 gh-139757: Add BINARY_OP_SUBSCR_USTR_INT (GH-143389)
Switched to and reset branch 'main'
make: *** [buildbottest] Error 2

POC implementation of BINARY_OP_SUBSCR_NCSTR_INT
15387dc
passes the tests and brings bm_tomli back to normal, i.e. 15 % speedup

Fidget-Spinner reviewed Jan 3, 2026
View reviewed changes

Merge remote-tracking branch 'upstream/main' into substr_cstr_int
b97fe4e

chris-eibl changed the title ~~Add BINARY_OP_SUBSCR_NCSTR_INT~~gh-139757: Add BINARY_OP_SUBSCR_USTR_INTJan 4, 2026

bedevere-appBot mentioned this pull request Jan 4, 2026
JIT assembly optimizer leaves some redundant branches #139757
Closed

chris-eibl added 6 commits January 4, 2026 09:09

BINARY_OP_SUBSCR_NCSTR_INT -> BINARY_OP_SUBSCR_USTR_INT
f69803d

Use _POP_TOP_UNICODE
f7c0739

move _PyLong_IsNonNegativeCompact into specialize
6ad8076
and just assert in the bytecodes

Narrow the result type of _BINARY_OP_SUBSCR_USTR_INT to str in the JIT
e7e3f40

add tests
c56ca35

blurb it
5685681

chris-eibl added performance Performance or resource usage interpreter-core (Objects, Python, Grammar, and Parser dirs) labels Jan 4, 2026

Fidget-Spinner reviewed Jan 4, 2026
View reviewed changes

Comment threadLib/test/test_capi/test_opt.py

Fidget-Spinner reviewed Jan 4, 2026
View reviewed changes

Comment threadPython/bytecodes.c Outdated

chris-eibl commented Jan 4, 2026
View reviewed changes

Fidget-Spinner reviewed Jan 4, 2026
View reviewed changes

Comment threadPython/bytecodes.c Outdated

fix DEOPT_IF and INPUTS_DEAD
00a2bb4

Fidget-Spinner approved these changes Jan 4, 2026
View reviewed changes

bedevere-appBot added the awaiting merge label Jan 4, 2026

chris-eibl marked this pull request as ready for review January 4, 2026 10:52

chris-eibl requested review from markshannon and tomasr8 as code owners January 4, 2026 10:52

bedevere-appBot added awaiting review and removed awaiting merge labels Jan 4, 2026

bump magic number - better safe than sorry
d41332e

chris-eibl commented Jan 4, 2026
View reviewed changes

Comment threadInclude/internal/pycore_magic_number.h Outdated

Update Include/internal/pycore_magic_number.h
6956fce
Co-authored-by: Chris Eibl <138194463+chris-eibl@users.noreply.github.com>

Fidget-Spinner merged commit e6bfe4d into python:mainJan 4, 2026
70 checks passed

bedevere-appBot removed the awaiting review label Jan 4, 2026

Fidget-Spinner mentioned this pull request Jan 4, 2026
JIT: Implement unique reference tracking in Tier 2 for reference count optimizations #143414
Closed

chris-eibl deleted the substr_cstr_int branch January 5, 2026 11:29

thunder-coding pushed a commit to thunder-coding/cpython that referenced this pull request Feb 15, 2026
pythongh-139757: Add BINARY_OP_SUBSCR_USTR_INT (pythonGH-143389)
b773aae

Navigation Menu

Uh oh!

Uh oh!

gh-139757: Add BINARY_OP_SUBSCR_USTR_INT - #143389

gh-139757: Add BINARY_OP_SUBSCR_USTR_INT#143389
Fidget-Spinner merged 11 commits into
python:mainfrom
chris-eibl:substr_cstr_int

chris-eibl commented Jan 3, 2026•
edited
Loading

Uh oh!

Fidget-Spinner commented Jan 3, 2026

Uh oh!

Fidget-Spinner left a comment

Uh oh!

Fidget-SpinnerJan 3, 2026

Uh oh!

Fidget-SpinnerJan 3, 2026

Uh oh!

Uh oh!

Uh oh!

chris-eiblJan 4, 2026

Uh oh!

Uh oh!

Fidget-Spinner commented Jan 4, 2026

Uh oh!

Fidget-Spinner left a comment

Uh oh!

Uh oh!

Uh oh!

bedevere-bot commented Jan 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Uh oh!

Conversation

chris-eibl commented Jan 3, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Fidget-Spinner commented Jan 3, 2026

Uh oh!

Fidget-Spinner left a comment

Choose a reason for hiding this comment

Uh oh!

Fidget-SpinnerJan 3, 2026

Choose a reason for hiding this comment

Uh oh!

Fidget-SpinnerJan 3, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

chris-eiblJan 4, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Fidget-Spinner commented Jan 4, 2026

Uh oh!

Fidget-Spinner left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

bedevere-bot commented Jan 4, 2026

⚠️⚠️⚠️ Buildbot failure ⚠️⚠️⚠️

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

chris-eibl commented Jan 3, 2026•
edited
Loading