Skip to content

Conversation

@markshannon
Copy link
Member

@markshannon markshannon commented Feb 28, 2022

Copy link
Member

@brandtbucher brandtbucher left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks, looks good overall.

I still think that it makes more sense (and is a bit simpler) to just store pointers across four cache entries for now, and explore something like this separately as a possible improvement later. It seems to me that the way you've done it here is quite a bit more complicated, for negligible gain: 6 byte savings per unquickened site, offset by the cost of an extra 2 bytes and a pointer indirection per quickened site, plus 10 more wasted bytes for the vast majority of quickened calls.

But I'll defer to your judgement here.

@markshannon
Copy link
Member Author

I've added some comments on about how to handle pointers to faster-cpython/ideas#263
and removed the need for a per-code-object cache from this PR and shrinks the inline cache a bit.

@markshannon markshannon merged commit 3b0f1c5 into python:main Mar 1, 2022
@brandtbucher brandtbucher mentioned this pull request Mar 2, 2022
26 tasks
@markshannon markshannon deleted the inline-cache-binary-subscr branch September 26, 2023 12:48
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants