Skip to content

[Suggestion] Use np.add.reduce instead of np.sum for slightly better performance #62

Description

@SaFE-APIOpt


I noticed the following line:
s = np.sum(arr, axis=0)
If the code is performance-sensitive and does not rely on extra parameters like dtype, keepdims, or initial, you might consider replacing it with:
s = np.add.reduce(arr, axis=0)
np.sum is a high-level convenience function that internally calls np.add.reduce, but includes additional checks and wrappers. When you're performing simple summation along an axis, using np.add.reduce directly eliminates that overhead and can offer a slight speedup—especially in tight loops or large-scale computations.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions