gh-138912: Add dedicated opcodes for MatchClass without positional sub-patterns

[cdce8p]

This PR complements and extends the work done in #138915.

In a microbenchmark for bare class matches, the match statement is now even slightly faster than the comparable if statement. With keyword only patterns, the performance is much improved as it's no longer necessary to construct intermediary tuples so that's now only ~20% slower compared to the if equivalent. (Previously it was closer to ~220%).

Further improvements can likely be archived if MATCH_CLASS_GET_OPT_ATTR is specialized in a similar manner to LOAD_ATTR. With the specialization for (just) LOAD_ATTR disabled, the match statement equivalent is again slightly faster than the corresponding if clause.

__
Micro benchmarks with all specializations enabled.

Micro benchmark (bare class pattern)

from timeit import timeit
setup = """
class A:
    def __init__(self, x, y):
        self.x = x
        self.y = y
a = A(1, 2)

def f1(a):
    if isinstance(a, A):
        return True
    return False

def f2(a):
    match a:
        case A():
            return True
    return False
"""
print(f"If\t{timeit("f1(a)", setup, number=10**7)}")
print(f"Match\t{timeit("f2(a)", setup, number=10**7)}")

# Before
If      0.29646458296338096
Match   0.6549855829798616

# After
If      0.2934637920006935
Match   0.2751032500018482 -> -58%

Micro benchmark (One keyword attribute)

from timeit import timeit
setup = """
class A:
    def __init__(self, x, y):
        self.x = x
        self.y = y
a = A(1, 2)

def f1(a):
    if isinstance(a, A) and a.x == 1:
        return True
    return False

def f2(a):
    match a:
        case A(x=1):
            return True
    return False
"""
print(f"If\t{timeit("f1(a)", setup, number=10**7)}")
print(f"Match\t{timeit("f2(a)", setup, number=10**7)}")

# Before
If      0.35557704203529283
Match   1.1494590829825029

# After
If      0.3604730829974869
Match   0.45169049999822164  -> -61%

• Issue: Performance improvements for the MATCH_CLASS opcode #138912

---

📚 Documentation preview 📚: https://cpython-previews--139080.org.readthedocs.build/

[sergey-miryanov]

@cdce8p Could you please resolve conflicts?

[sergey-miryanov]

I believe this needs a what's new entry too.

[cdce8p]

I believe this needs a what's new entry too.

Maybe, but I think we should wait until both PRs are merged before adding a combined one.

[gpshead]

the earlier PR was merged, this could use an update.

[cdce8p]

I extracted the MATCH_CLASS_ISINSTANCE opcode part into a separate PR to make it easier to review. Marking this one as draft for now.

• gh-138912: Add fast path for match class patterns without sub-patterns #144820

[read-the-docs-community]

Documentation build overview

📚 cpython-previews | 🛠️ Build #33355942 | 📁 Comparing 765dd36 against main (d206d42)

  🔍 Preview build  

239 files changed · + 6 added · ± 232 modified · - 1 deleted

+ Added

• c-api/sentinel.html
• c-api/slots.html
• deprecations/pending-removal-in-3.21.html
• howto/abi3t-migration.html
• library/asyncio-threading.html
• whatsnew/3.16.html

± Modified

• bugs.html
• glossary.html
• index.html
• py-modindex.html
• c-api/bytes.html
• c-api/complex.html
• c-api/concrete.html
• c-api/coro.html
• c-api/dict.html
• c-api/exceptions.html
• and 222 more...

- Deleted

• library/asyncio-policy.html

[sergey-miryanov]

@cdce8p You don't need to merge main into your branch, since it notifies all participants.

This PR is in draft mode — are there any changes you still need to make before it's ready for core-dev review?

[cdce8p]

@cdce8p You don't need to merge main into your branch, since it notifies all participants.

Sorry for that. I merged main to get #148826 into the branch. It fixed an issue that prevented performance comparisons with specializations disabled.

This PR is in draft mode — are there any changes you still need to make before it's ready for core-dev review?

Not technically. A while ago I opened #144820 which is essentially the first part of the changes here, just because I thought it might be easier to review. Unfortunately there hasn't been any movement on that since February. The decision seems to be hung on whether it is worth it to introduce a new opcode for the speedup. If you have some time, I'd appreciate another perspective on it.

[sergey-miryanov]

@eendebakpt Could you please take a look?