[GH-ISSUE #35] Why use "ld.acquire" order to get channel_tail_idx in dispatch kernels? #33

New issue

Closed

opened 2026-10-26 09:52:52 +00:00 by bsakaguchi · 1 comment

bsakaguchi commented

2026-10-26 09:52:52 +00:00

Owner

Originally created by @hellboywyh on GitHub (Mar 2, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/35

I notice the use of "ld.acquire" order to get channel_tail_idx in dispatch kernels. I guess it is used to gurantee that the following load will not preceed the load of channel_tail_idx. But I think the condition of breaking "while block", I mean, cached_channel_head_idx != cached_channel_tail_idx, has already gurantee the order. So if use a normal load, it will also work?

Originally created by @hellboywyh on GitHub (Mar 2, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/35 <img width="796" alt="Image" src="https://github.com/user-attachments/assets/b5c14b3d-dce9-43ca-b806-1abd0078b0db" /> I notice the use of "ld.acquire" order to get channel_tail_idx in dispatch kernels. I guess it is used to gurantee that the following load will not preceed the load of channel_tail_idx. But I think the condition of breaking "while block", I mean, `cached_channel_head_idx != cached_channel_tail_idx`, has already gurantee the order. So if use a normal load, it will also work?

bsakaguchi closed this issue

2026-10-26 09:52:52 +00:00

bsakaguchi commented

2026-10-26 09:52:53 +00:00

Author

Owner

@LyricZhao commented on GitHub (Mar 3, 2025):

Normal loads (ld.volatile) are correct, but release/acquire pairs are more "semantically" correct, you can think it as a coding standard.

@LyricZhao commented on GitHub (Mar 3, 2025): Normal loads (`ld.volatile`) are correct, but `release/acquire` pairs are more "semantically" correct, you can think it as a coding standard.

bsakaguchi referenced this issue

2026-10-26 09:53:39 +00:00

[GH-ISSUE #74] test_internode.py hangs at internode.cu:dispatch(...) #68

bsakaguchi referenced this issue

2026-10-26 09:56:30 +00:00

[GH-ISSUE #270] A timeout error occurred when running test_internode.py on four devices. #200

bsakaguchi referenced this issue

2026-10-26 09:56:40 +00:00

[GH-ISSUE #290] Running the test case on a 2-node H20 3E machine failed with internode.cu:1029 'unspecified launch failure' #208