Skip to content

Migrates resolution cache code from numba fixing a JIT speed regression. - #915

Open
c200chromebook wants to merge 1 commit into
NVIDIA:mainfrom
c200chromebook:resolution_cache
Open

Migrates resolution cache code from numba fixing a JIT speed regression.#915
c200chromebook wants to merge 1 commit into
NVIDIA:mainfrom
c200chromebook:resolution_cache

Conversation

@c200chromebook

@c200chromebook c200chromebook commented Jul 10, 2026

Copy link
Copy Markdown

This is a straight-shot move of resolution cache from the original numba to correct a compile time regression. We have it monkeypatched in now and it works ok. Anyone using the sol'n would probably benefit from it. We are working to get MLIR up and running for use case as well (note recent bug report).

Also corrects a trivial bug normalize_dict(obj) should be working on obj not kws.
@copy-pr-bot

copy-pr-bot Bot commented Jul 10, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@c200chromebook

Copy link
Copy Markdown
Author

@gmarkall would you mind having a look at this one? it is kind of annoying for us

@gmarkall

gmarkall commented Aug 5, 2026

Copy link
Copy Markdown
Contributor

@c200chromebook Thanks for the PR and the ping. I'm no longer working on Numba-CUDA, so I will defer to @kkraus14 on where to go with this PR.

@kkraus14

kkraus14 commented Aug 5, 2026

Copy link
Copy Markdown
Contributor

@c200chromebook please see https://github.com/NVIDIA/numba-cuda#maintenance-notice

We are only accepting security issues and critical bug fixes at this time for numba-cuda. I would recommend trying to move to numba-cuda-mlir and if the same JIT speed regression exists there we will happily try to resolve it.

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.

3 participants