gh-132732: Automatically constant evaluate pure operations #132733

Fidget-Spinner · 2025-04-19T16:21:48Z

Issue: Constant evaluate/propagate pure ops automatically #132732

python-cla-bot · 2025-04-19T16:21:51Z

All commit authors signed the Contributor License Agreement.

Misc/NEWS.d/next/Core_and_Builtins/2025-04-19-16-22-47.gh-issue-132732.jgqhlF.rst

brandtbucher

This is really neat!

Other than two opcodes I found that shouldn't be marked pure, I just have one thought:

Rather than rewriting the bodies like this to use the symbols-manipulating functions (which seems error-prone), would we be able to just use stackrefs to do this?

For example, _BINARY_OP_ADD_INT is defined like this:

PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
// ...
res = PyStackRef_FromPyObjectSteal(res_o);

Rather than rewriting uses of these functions, could it be easier to just do something like this, since we're guranteed not to escape?

if (sym_is_const(ctx, stack_pointer[-2]) && sym_is_const(ctx, stack_pointer[-1])) {
    // Generated code to turn constant symbols into stackrefs:
    _PyStackRef left = PyStackRef_FromPyObjectBorrow(sym_get_const(ctx, stack_pointer[-2]));
    _PyStackRef right = PyStackRef_FromPyObjectBorrow(sym_get_const(ctx, stack_pointer[-1]));
    _PyStackRef res;
    // Now the actual body, same as it appears in executor_cases.c.h:
    PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
    PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
    // ...
    res = PyStackRef_FromPyObjectSteal(res_o);
    // Generated code to turn stackrefs into constant symbols:
    stack_pointer[-1] = sym_new_const(ctx, PyStackRef_AsPyObjectSteal(res));
}

I'm not too familiar with the design of the cases generator though, so maybe this is way harder or something. Either way, I'm excited to see this get in!

Python/bytecodes.c

Fidget-Spinner · 2025-04-24T22:37:57Z

Rather than rewriting uses of these functions, could it be easier to just do something like this, since we're guranteed not to escape?

Seems feasible. I could try to rewrite all occurences of the variable with a stackref-producing const one. Let me try that.

Fidget-Spinner · 2025-04-25T01:06:40Z

I've verified no refleak on test_capi.test_opt locally apart from #132731 which is pre-existing.

markshannon · 2025-04-30T09:32:49Z

There's a lot going on in this PR, probably too much for one PR.

Could we start with a PR to fix up the pure annotations so that they are on the correct instructions and maybe add the pure_guard annotation that Brandt suggested?

markshannon · 2025-04-30T09:37:48Z

Could we have the default code generator generate a function for the body of the pure instruction and then call that from the three interpreters?

brandtbucher · 2025-04-30T14:58:57Z

Could we have the default code generator generate a function for the body of the pure instruction and then call that from the three interpreters?

Hm, I think I’d prefer not to. Sounds like it could hurt performance, especially for the JIT (where things can’t inline).

brandtbucher · 2025-04-30T15:10:30Z

I think a good progression would be:

Implement the pure attribute, and the optimizer changes. Remove the pure attributes where they don’t belong (so nothing breaks) and leave the existing ones as proof that the implementation works. (This PR)
Audit the existing non-pure bytecodes and add pure where it makes sense. (Follow-up PR)
Implement the pure_guard attribute, and annotate any bytecodes that can use it. (Follow-up PR)

Fidget-Spinner · 2025-04-30T15:15:59Z

Could we have the default code generator generate a function for the body of the pure instruction and then call that from the three interpreters?

Hm, I think I’d prefer not to. Sounds like it could hurt performance, especially for the JIT (where things can’t inline).

I thought about this and I think we can inline if we autogenerate a header file and include that directly. But then we're at the mercy of the compiler in both the normal interpreter and the JIT deciding to inline or not to inline the body again. Which I truly do not want.

Fidget-Spinner · 2025-05-08T23:10:39Z

@brandtbucher @markshannon what can I do to get this PR moving?

@tomasr8 if youd like to review, here's a summary of the PR:

If a bytecode operation is pure (no side effects) we can mark it as pure in bytecodes.c.
In the optimizer, we automatically generate the body that does evaluation of the symbolic constants by copy pasting the bytecodes.c definition into the optimizer's C code. Of course we check that the inputs are constants first.
All changes to the cases generator is for the second point.

tomasr8 · 2025-05-08T23:15:14Z

Thanks for the ping! I actually wanted to try/review this PR, I was just very busy this week with work :/ I'll have a look this weekend :)

tomasr8

Only had time to skim the PR, I'll do a more thorough review this weekend :)

Python/optimizer_bytecodes.c

Python/optimizer_analysis.c

Tools/cases_generator/optimizer_generator.py

Co-Authored-By: Tomas R. <[email protected]>

Fidget-Spinner · 2025-06-06T12:43:44Z

@markshannon do you have any other comments? I think this has gone through enough rounds of review by you, Brandt, and Tomas, which I'm thankful for.

Tools/cases_generator/optimizer_generator.py

Fidget-Spinner · 2025-06-11T10:53:38Z

@markshannon if there are no objections, I'm going to merge this on Friday. There's 3 reviews, 1 approval, and this PR has been up for 2 months. I think I've given everyone sufficient time to review, and I'm thinking that any further issues can be addressed in future PRs.

Tools/cases_generator/optimizer_generator.py

markshannon · 2025-06-12T17:55:11Z

Tools/cases_generator/generators_common.py

@@ -129,6 +140,31 @@ def __init__(self, out: CWriter, labels: dict[str, Label]):
        }
        self.out = out
        self.labels = labels
+        self.is_abstract = False
+
+    def emit_to_with_replacement(


I don't think we need any new code here.
All the new code should go in optimizer_generator.py

Tools/cases_generator/generators_common.py

markshannon · 2025-06-12T17:58:01Z

Tools/cases_generator/generators_common.py

@@ -157,7 +193,7 @@ def deopt_if(
        lparen = next(tkn_iter)
        assert lparen.kind == "LPAREN"
        first_tkn = tkn_iter.peek()
-        emit_to(self.out, tkn_iter, "RPAREN")
+        self.emit_to_with_replacement(self.out, tkn_iter, "RPAREN", uop, storage, inst)


Why is this necessary?

emit_to doesn't call the replacement functions. This does.

Why are there function calls that need replacing within a DEOPT_IF?

I'm using replacement functions to replace all references to a variable with a replacement. For example, all references to -- out get replaced with out_stackref.

The thing that is getting replaced is references like

if (PyStackRef_IsNull(out)) {....}

becomes

if (PyStackRef_IsNull(out_stackref)) {....}

markshannon · 2025-06-13T07:35:45Z

It seems like the issue of escaping is causing problems here.
For the code generator, "escaping" means "able to run the GC", which shouldn't happen in the abstract interpreter.
So either, we are not correctly marking functions as non-escaping, or we are calling functions that do escape (which we shouldn't).

In the example you give, _PyLong_Multiply(sym_get_const(x), sym_get_const(y)) neither _PyLong_Multiply nor sym_get_const escape. They just need to be added to the whitelist.

Fidget-Spinner · 2025-06-13T08:05:39Z

I've added the required functions to the allowlist, so I removed the is_abstract workaround.

Fidget-Spinner · 2025-06-13T15:14:31Z

It seems like the issue of escaping is causing problems here. For the code generator, "escaping" means "able to run the GC", which shouldn't happen in the abstract interpreter. So either, we are not correctly marking functions as non-escaping, or we are calling functions that do escape (which we shouldn't).

In the example you give, _PyLong_Multiply(sym_get_const(x), sym_get_const(y)) neither _PyLong_Multiply nor sym_get_const escape. They just need to be added to the whitelist.

@markshannon it seems this is wrong. _PyLong_Multiply/Add/Subtract can trigger the GC. The failing tests currently are evidence of that. The problem is with the SIGCHECK macro in longobject.c. https://github.com/python/cpython/blob/main/Objects/longobject.c#L114

Fidget-Spinner · 2025-06-13T15:21:18Z

I have a fix in a separate PR for the longobject GC issues.

markshannon · 2025-06-13T15:35:58Z

I think we only specialize for, and are interested in compact ints (or tagged ints in the future), so maybe replace _PyLong_Add with _PyCompactLong_Add?
It would help with much the same issue I'm having with excessive escapes in TOS caching.

Fidget-Spinner · 2025-06-13T15:44:12Z

For now, I'm avoiding changing the int operations in this PR. I will add back constant evaluation for them in the future once we fix this in either bytecodes.c or the long object.

Automatically constant evaluate pure operations

1ffbb6b

Fidget-Spinner requested a review from markshannon as a code owner April 19, 2025 16:21

bedevere-app bot added the awaiting core review label Apr 19, 2025

Fidget-Spinner requested review from brandtbucher and removed request for markshannon April 19, 2025 16:21

bedevere-app bot mentioned this pull request Apr 19, 2025

Constant evaluate/propagate pure ops automatically #132732

Open

blurb-it bot and others added 3 commits April 19, 2025 16:22

📜🤖 Added by blurb_it.

691084d

Fix tests

b89e4dc

Merge branch 'pure' of github.com:Fidget-Spinner/cpython into pure

0959918

StanFromIreland reviewed Apr 19, 2025

View reviewed changes

Misc/NEWS.d/next/Core_and_Builtins/2025-04-19-16-22-47.gh-issue-132732.jgqhlF.rst Show resolved Hide resolved

brandtbucher mentioned this pull request Apr 24, 2025

gh-131798: Use sym_new_type instead of sym_new_not_null for _BUILD_STRING, _BUILD_SET #132564

Merged

brandtbucher reviewed Apr 24, 2025

View reviewed changes

Python/bytecodes.c Outdated Show resolved Hide resolved

Python/bytecodes.c Outdated Show resolved Hide resolved

Python/bytecodes.c Outdated Show resolved Hide resolved

Fidget-Spinner added 2 commits April 25, 2025 07:16

Merge remote-tracking branch 'upstream/main' into pure

2541683

Apply review suggestions

d5b2208

reduce diff

71ced86

Fidget-Spinner mentioned this pull request Apr 26, 2025

gh-131798: JIT: Propagate the result in _BINARY_OP_SUBSCR_TUPLE_INT #133003

Merged

Fidget-Spinner added 2 commits May 7, 2025 06:48

Merge remote-tracking branch 'upstream/main' into pure

a10d5a1

Update pycore_opcode_metadata.h

d22f165

tomasr8 reviewed May 9, 2025

View reviewed changes

Python/optimizer_bytecodes.c Outdated Show resolved Hide resolved

Python/optimizer_analysis.c Outdated Show resolved Hide resolved

Tools/cases_generator/optimizer_generator.py Outdated Show resolved Hide resolved

Tools/cases_generator/optimizer_generator.py Outdated Show resolved Hide resolved

Apply changes from code review

8ae38c7

Co-Authored-By: Tomas R. <[email protected]>

bedevere-app bot requested a review from markshannon May 28, 2025 14:04

Fidget-Spinner added 5 commits May 28, 2025 22:32

fix linter/mypy

b278734

remove whitespace

73a8b00

Remove PyDict_Type

4116a31

add bool type

548b67c

reduce diff

74a0208

Fidget-Spinner commented Jun 6, 2025

View reviewed changes

Tools/cases_generator/optimizer_generator.py Outdated Show resolved Hide resolved

noamcohen97 mentioned this pull request Jun 7, 2025

gh-131798: Optimize _UNARY_NEGATIVE #135223

Open

Fidget-Spinner added 2 commits June 9, 2025 18:50

Grab identifiers from REPLACE_OPCODE_IF_EVALUATES_PURE

6a5dc12

Merge remote-tracking branch 'upstream/main' into pure

3896775

Fidget-Spinner requested a review from brandtbucher June 9, 2025 16:32

markshannon reviewed Jun 12, 2025

View reviewed changes

Fidget-Spinner added 4 commits June 13, 2025 11:29

Use replacer

507c80a

reduce diff

dc68b45

Merge remote-tracking branch 'upstream/main' into pure

41a271c

Update optimizer_generator.py

e88b71a

Fidget-Spinner added 2 commits June 13, 2025 16:02

Address review (add long functions to allowlist)

866510f

fix mypy

01be0c6

Fidget-Spinner requested a review from markshannon June 13, 2025 14:38

Merge remote-tracking branch 'upstream/main' into pure

fce79a6

revert changes for add,multiply,sub int

9bef4a4

Fix tests

1f76e2c

Uh oh!

gh-132732: Automatically constant evaluate pure operations #132733

Are you sure you want to change the base?

gh-132732: Automatically constant evaluate pure operations #132733

Conversation

Fidget-Spinner commented Apr 19, 2025 • edited by bedevere-app bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

python-cla-bot bot commented Apr 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

brandtbucher left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Fidget-Spinner commented Apr 24, 2025

Uh oh!

Fidget-Spinner commented Apr 25, 2025

Uh oh!

markshannon commented Apr 30, 2025

Uh oh!

markshannon commented Apr 30, 2025

Uh oh!

brandtbucher commented Apr 30, 2025

Uh oh!

brandtbucher commented Apr 30, 2025

Uh oh!

Fidget-Spinner commented Apr 30, 2025

Uh oh!

Fidget-Spinner commented May 8, 2025

Uh oh!

tomasr8 commented May 8, 2025

Uh oh!

tomasr8 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Fidget-Spinner commented Jun 6, 2025

Uh oh!

Uh oh!

Fidget-Spinner commented Jun 11, 2025

Uh oh!

Uh oh!

markshannon Jun 12, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

markshannon Jun 12, 2025

Choose a reason for hiding this comment

Uh oh!

Fidget-Spinner Jun 13, 2025

Choose a reason for hiding this comment

Uh oh!

markshannon Jun 13, 2025

Choose a reason for hiding this comment

Uh oh!

Fidget-Spinner Jun 13, 2025

Choose a reason for hiding this comment

Uh oh!

markshannon commented Jun 13, 2025

Uh oh!

Fidget-Spinner commented Jun 13, 2025

Uh oh!

Fidget-Spinner commented Jun 13, 2025

Uh oh!

Fidget-Spinner commented Jun 13, 2025

Uh oh!

markshannon commented Jun 13, 2025

Uh oh!

Fidget-Spinner commented Jun 13, 2025

Uh oh!

Uh oh!

Fidget-Spinner commented Apr 19, 2025 •

edited by bedevere-app bot

Loading

python-cla-bot bot commented Apr 19, 2025 •

edited

Loading