deepseek-ai/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP.git synced 2026-11-05 00:47:32 +00:00
Code Issues 308 Projects Packages Wiki Activity

[GH-ISSUE #83] [low latency] how to make sure data arrived after removed poll cq #76

New issue
Closed
opened 2026-10-26 09:53:53 +00:00 by bsakaguchi · 3 comments
bsakaguchi commented 2026-10-26 09:53:53 +00:00
Owner
Copy link

Originally created by @elevenxiang on GitHub (Mar 19, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/83

Hi,

one question for low latency kernel
I noticed that there was a commit "Low latency kernels use rdma atomic to support AR.", which removed the poll cq operation when doing receive

so after rdma atomic add operation, recv count will be updated, but how to make sure previously post sent data already arrived ?

Thanks

Originally created by @elevenxiang on GitHub (Mar 19, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/83 Hi, one question for low latency kernel I noticed that there was a commit "Low latency kernels use rdma atomic to support AR.", which removed the poll cq operation when doing receive so after rdma atomic add operation, recv count will be updated, but how to make sure previously post sent data already arrived ? Thanks
bsakaguchi commented 2026-10-26 09:53:54 +00:00
Author
Owner
Copy link

@elevenxiang commented on GitHub (Mar 19, 2025):

it was guaranteed by the wqe order inside the same expert qp ?

<!-- gh-comment-id:2736392581 --> @elevenxiang commented on GitHub (Mar 19, 2025): it was guaranteed by the wqe order inside the same expert qp ?
bsakaguchi commented 2026-10-26 09:53:54 +00:00
Author
Owner
Copy link

@sphish commented on GitHub (Mar 20, 2025):

The InfiniBand Architecture Specification includes rules that ensure when RDMA atomic operations complete, all preceding RDMA write operations have also completed.

<!-- gh-comment-id:2739622433 --> @sphish commented on GitHub (Mar 20, 2025): The InfiniBand Architecture Specification includes rules that ensure when RDMA atomic operations complete, all preceding RDMA write operations have also completed.
bsakaguchi commented 2026-10-26 09:53:55 +00:00
Author
Owner
Copy link

@elevenxiang commented on GitHub (Mar 20, 2025):

The InfiniBand Architecture Specification includes rules that ensure when RDMA atomic operations complete, all preceding RDMA write operations have also completed.

okay, get it
Thanks ~

<!-- gh-comment-id:2740950075 --> @elevenxiang commented on GitHub (Mar 20, 2025): > The InfiniBand Architecture Specification includes rules that ensure when RDMA atomic operations complete, all preceding RDMA write operations have also completed. okay, get it Thanks ~
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
hybrid-ep
chore/remove-clang-format-yapf
v2.5
nvDev
try-fix-cu128
mori-ep
antgroup-opt
compat-old-torch
add-linter
eager-rdma
v1.2.1
Labels
Clear labels   
pull-request
Mirrored from GitHub Pull Request

No labels
pull-request
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepEP#76
No description provided.