Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[PHI] Functionalization for Fluid kernel #52395

Closed
From00 opened this issue Mar 31, 2023 · 10 comments
Closed

[PHI] Functionalization for Fluid kernel #52395

From00 opened this issue Mar 31, 2023 · 10 comments
Assignees
Labels
PFCC Paddle Framework Contributor Club,https://github.com/PaddlePaddle/community/tree/master/pfcc status/close 已关闭

Comments

@From00
Copy link
Contributor

From00 commented Mar 31, 2023

问题描述 Please describe your issue

大家好,当前Fluid下有部分算子由于未迁移到新的PHI算子体系,无法具备PHI下函数式算子注册时”记录自身输入输出属性“的能力,既使PHI算子库无法达成统一所有算子的理想状态,也给框架调度体系的优化工作带来较大的负担。我们一共收集了14个需要迁移的算子,欢迎大家提交PR一起对这些算子做迁移改造。

更多详细介绍见 Call-for-Contributions: Fluid算子函数式迁移专项, 本issue用于跟踪记录该项目下各个算子的迁移改造进度。

注:

  1. 同一个算子的cpu/gpu/xpu kernel以及InferShape尽量由同一个同学进行迁移改造,可直接在本issue下方回复认领。
  2. 该任务时间:PR 截止提交时间是5月20日,截止合入时间是5月31日。
  3. PR 通过 CI 后,在评论里 @From00 进行code review。
  4. 本任务挑战性较大,对于刚接触Paddle开发的同学,建议先在good first issue中挑选一些简单的任务尝试。 欢迎联系花花加入社区,和我们一起快乐开源

image

待迁移算子列表(整体进度 14/14)

按 merge 的时间顺序,排名不分先后:@AndSonder (6) @gouzil (2) @huangjiyi (5) @GreatV (1)

序号
算子名称
认领人
PR
(cpu/gpu kernel )
PR
(xpu kernel)
PR
(InferShape)
1 cudnn_lstm ✅(2023/5/16) @huangjiyi #53730 无需 无需
2 dequantize ✅(2023/5/15) @huangjiyi #53739 无需 无需
3 distributed_fused_lamb ✅(2023/5/23) @huangjiyi #53896 无需 无需
4 fused_attention ✅(2023/4/17) @AndSonder #51743 #53011 无需
5 fused_attention_grad ✅(2023/4/17) @AndSonder #51909 #53011 无需
6 fused_batch_norm_act ✅ (2023/05/16) @AndSonder #53476 无需 无需
7 fused_batch_norm_act_grad ✅ (2023/05/16) @AndSonder #53476 无需 无需
8 fused_feedforward ✅(2023/4/24) @AndSonder #53166 #53196 无需
9 fused_feedforward_grad ✅(2023/4/24) @AndSonder #53166 #53196 无需
10 fusion_group ✅(2023/5/23) @huangjiyi #53781 无需 #53934
11 pow2_decay_with_linear_warmup ✅ (2023/05/12) @huangjiyi #53741 #53741 无需
12 sequence_mask ✅(2023/5/18) @GreatV #53478 无需 #53782
13 sequence_pool ✅(2023/4/17) @gouzil #52750 无需 无需
14 stft ✅ (2023/05/16) @gouzil #53517 无需 无需
@From00 From00 added status/new-issue 新建 type/others 其他问题 labels Mar 31, 2023
@From00 From00 added PFCC Paddle Framework Contributor Club,https://github.com/PaddlePaddle/community/tree/master/pfcc and removed status/new-issue 新建 labels Mar 31, 2023
@From00 From00 assigned luotao1 and From00 and unassigned CtfGo Mar 31, 2023
@luotao1 luotao1 moved this to In Progress in Call for Contributions Mar 31, 2023
@engineer1109
Copy link
Contributor

我记得 xxx_grad的算子 现在一个也没有吧

@GreatV
Copy link
Contributor

GreatV commented Mar 31, 2023

序号 算子名称 认领人 PR(cpu/gpu kernel ) PR(xpu kernel) PR(InferShape)
10 sequence_mask @GreatV #53478

@huangjiyi
Copy link
Member

huangjiyi commented Mar 31, 2023

序号 算子名称 认领人 PR(cpu/gpu kernel ) PR(xpu kernel) PR(InferShape)
1 cudnn_lstm @huangjiyi #53730 无需 无需
2 dequantize @huangjiyi #53739 无需 无需
3 distributed_fused_lamb @huangjiyi #53896 无需 无需
10 fusion_group @huangjiyi #53781 无需 #53934
11 pow2_decay_with_linear_warmup @huangjiyi #53741 #53741 无需

@Liyulingyue
Copy link
Contributor

Liyulingyue commented Mar 31, 2023

序号 算子名称 认领人 PR(cpu/gpu kernel ) PR(xpu kernel) PR(InferShape)
3 distributed_fused_lamb @Liyulingyue

@gouzil
Copy link
Member

gouzil commented Mar 31, 2023

序号 算子名称 认领人 PR(cpu/gpu kernel ) PR(xpu kernel) PR(InferShape)
11 sequence_pool @gouzil #52680,#52750,#52782 -

@sanbuphy
Copy link
Contributor

认领 2 dequantize

@jinyouzhi
Copy link
Contributor

序号 算子名称 认领人 PR(cpu/gpu kernel ) PR(xpu kernel) PR(InferShape)
9 pow2_decay_with_linear_warmup @jinyouzhi

@AndSonder
Copy link
Contributor

AndSonder commented Apr 19, 2023

序号 算子名称 认领人 PR(cpu/gpu kernel ) PR(xpu kernel) PR(InferShape)
6 fused_batch_norm_act @AndSonder #53476 - -
7 fused_batch_norm_act_grad @AndSonder #53476 - -
8 fused_feedforward @AndSonder #53166 #53196 -
9 fused_feedforward_grad @AndSonder #53166 #53196 -

@gouzil
Copy link
Member

gouzil commented May 5, 2023

序号 算子名称 认领人 PR(cpu/gpu kernel ) PR(xpu kernel) PR(InferShape)
14 stft @gouzil #53517 -

@luotao1
Copy link
Contributor

luotao1 commented May 23, 2023

Fluid算子函数式迁移专项已全部完成,感谢参与的小伙伴们!

按 merge 的时间顺序,排名不分先后:@AndSonder (6) @gouzil (2) @huangjiyi (4) @GreatV (1)

欢迎继续参与快乐开源的其他任务中国软件开源创新大赛:飞桨框架任务挑战赛的其他赛题

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
PFCC Paddle Framework Contributor Club,https://github.com/PaddlePaddle/community/tree/master/pfcc status/close 已关闭
Projects
Development

No branches or pull requests