Skip to content

ggml-cpu: Pass on tag_name to the feature scoring #14332

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 1 commit into from

Conversation

ckastner
Copy link
Collaborator

@ckastner ckastner commented Jun 22, 2025

This enables scoring backends based on criteria other than instruction support, such as platform name.

An example would be backends built for generic "armv8.6-a", and for "neoverse-n1" which implements armv8.6-a. Scoring only by features, these could look identical, whereas a -mcpu=neoverse-n1 targeted build might be able to edge out extra performance or efficiency.

The tag_name should be considered free-text. It is up to the scoring implementation to parse/extract relevant information.

This enables scoring backends based on criteria other than instruction
support such as platform name.

An example would be backends built for generic "armv8.6-a", and for
"neoverse-n1" which implements armv8.6-a. Scoring only by features,
these could look identical, whereas a -mcpu=neoverse-n1 targeted build
might be able to edge out extra performance or efficiency.

The tag_name should be considered free-text. It is up to the scoring
implementation to parse/extract relevant information.
@github-actions github-actions bot added the ggml changes relating to the ggml tensor library for machine learning label Jun 22, 2025
@slaren
Copy link
Member

slaren commented Jun 23, 2025

I would prefer if this is merged at the same time that the code that is intended to use it is merged. As it is, this is non-functional.

@ckastner
Copy link
Collaborator Author

Yeah, I thought this might be an issue.

My thinking was that trivial as this is, there are already a few design decisions (no cmake function interface changes, no structured values, no assumptions on how scoring uses this) that influence how the functional part might look like, it might be less work to get these out of the way first.

I'll do an update with something that uses this.

@ckastner
Copy link
Collaborator Author

This turned out to be the wrong approach after all. It is superseded by #14380.

@ckastner ckastner closed this Jun 25, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
ggml changes relating to the ggml tensor library for machine learning
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants