bpo-46841: Use inline caching for calls #31709
Conversation
markshannon
added
the
🔨 test-with-buildbots
|
🤖 New build scheduled with the buildbot fleet by @markshannon for commit 2021895 🤖 If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again. |
bedevere-bot
removed
the
🔨 test-with-buildbots
markshannon
added
the
🔨 test-with-buildbots
|
🤖 New build scheduled with the buildbot fleet by @markshannon for commit 2021895 🤖 If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again. |
bedevere-bot
removed
the
🔨 test-with-buildbots
| ); | ||
| } | ||
| /* Maximum size of code to quicken, in code units. */ | ||
| #define MAX_SIZE_TO_QUICKEN 10000 |
There was a problem hiding this comment.
Is this just to get the unpack sequence benchmark to work again, or something else?
There was a problem hiding this comment.
Nope, just for unpack_sequence.
| PyObject *isinstance; | ||
| PyObject *len; | ||
| PyObject *list_append; | ||
| }; |
There was a problem hiding this comment.
I think the existence of PyList_Type as part of the API means that list.append must be per-process unique.
In other words, list_append could be static.
I'm happy to leave it as is for now, though. We should look to make the whole struct static, although the mutability of builtin functions makes that tricky for isinstance and len.
There was a problem hiding this comment.
I believe each interpreter has its own builtins module (check out _PyBuiltin_Init), so making this static could be tricky. As you said, though: probably worth looking into in the future.
Also:
PRECALL/CALLspecialization statsPRECALL/CALLinstructionsNext steps tracked at faster-cpython/ideas#310.
https://bugs.python.org/issue46841