Skip to content

boxed data still imposes a small perf penalty #1737

@nikomatsakis

Description

@nikomatsakis

I did some quick benchmarks to see how effective #1493 was. The basis was comparing two versions of the binary tree shootout test, one based on boxes and one on unique pointers, running on my local machine with a parameter of 20. The results were:

  • Before: 6.3s vs 22s
  • After: 6.4s vs 7.5s

I did several runs but the results were fairly stable. So, there is a much smaller gap, but there is still a gap. Really, boxes ought to be faster, as they are purely stack local: we ought to be able to optimize allocation better.

Metadata

Metadata

Assignees

No one assigned

    Labels

    I-slowIssue: Problems and improvements with respect to performance of generated code.

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions