Skip to content

ISSUE8:基于IBGDA和IBRC的Alltoall在各个msg的理论性能上限 #8

Description

@moningchen

ISSUE 目标

探究GPU通信各个原子模块的耗时

验收要求:

  1. 输出alltoall各个msg下,ibrc和ibgda的理论耗时
  2. 输出alltoall的微观实现中,各个独立子模块的理论耗时计算公式

详情可见 /src/test/issue8/README.md 文件

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