Skip to content

Performance regression in L2 distance calculation for uint8 #9215

Description

@westonpace

The linalg benchmark L2(uint8, scalar) degraded between eb1df70 and c4086ca and the most likely culprit seems to be 7bf7e97

This may have been intentional. The PR did improve safety and overflow checking. There does not appear to be any impact on non-scalar paths.

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

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions