Hacktoberfest 2026:维护者为十月标记出来的 issue,仍然开放、适合新手。 浏览 Hacktoberfest issue

[FEA] Add option to report hash collisions

未关闭
#70 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 2 天内回复

还没有人认领这个 Issue。

评估

难度
5/5
预计耗时
一周以上
新手友好度
35/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
停滞
技术栈
cpp
领域
performance

调研方向

从 include/cuco/detail/static_map.inl 中第 252 行附近的 no-CG static_map::find 路径开始,然后跟踪 issue 中提到的其他 insert 和 find device 函数。在 static_map 和 dynamic_map 中定义选择启用的冲突计数行为,确保计数可在 host 上访问,并且默认禁用,同时不影响标准性能。

由索引模型根据 Issue 内容生成。

描述

type: feature request

Is your feature request related to a problem? Please describe.
Hash collisions impact performance of hash map insert and probe. It will be useful to find a way to report the number of collisions for static_map and dynamic_map to help assess performance of insert or probe when developers are evaluating perf on their dataset. It would also allow developers to tune the hash function or occupancy to reduce collisions and find the right balance for their scenario.

Describe the solution you'd like
The map can have an optional template argument that specifies if we need to count collisions (disabled by default), so it's opt-in and doesn't impact perf for the standard case. The number of collisions would be stored in a class variable that's accessible with something like get_num_collisions(). Implementation: allocate memory for device variable uint64_t *d_num_collisions, update all insert and find device code to do atomicAdd(d_num_collisions, 1) to that variable, then copy the contents to the host variable after the kernel. Here is where we can count the collisions for no-CG static_map::find:
https://github.com/NVIDIA/cuCollections/blob/2196040f0562a0280292eebef5295d914f615e63/include/cuco/detail/static_map.inl#L252
The atomic will be guarded by the template argument check, so should only impact perf if we're asked to count collisions. Similarly, would have to update all other insert and find functions.

Describe alternatives you've considered
None.

Additional context
None.

主要语言
Cuda
星标
671
派生
122
平均合并
4 天 19 小时
30 天内合并 PR
10

环境准备

在 Codespaces 中打开

在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

NVIDIA/cuCollections 的其他 Issue

查看 NVIDIA/cuCollections 的全部 Issue

相似的 Issue

更多 Performance Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。