Skip to content

A single GEMV seems to be killing performance #13

Description

@Michal-Novomestsky

Please see #12 for more details. Tl;dr it seems like we can get a ~6.5x speedup on inference (at least on an L40S with the case of a 32x32 grid with 6 timesteps, one channel and batch size 4) by changing one line of code.

I'm using walrus for a personal project so unfortunately I don't have much time to investigate properly (and what I'm saying may well be wrong), but please let me know if you'd like me to look into this further (e.g. make an MVP to demonstrate).

Thanks!

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions