search

quiver-team / torch-quiver

PyTorch Library for Low-Latency, High-Throughput Graph Learning on GPUs.
https://torch-quiver.readthedocs.io/en/latest/
Apache License 2.0
293 stars 36 forks source link

CUDA #104

Closed yingtongxiong closed 2 years ago

yingtongxiong commented 2 years ago

Excuse me, I am confused by your CSRSample kernel. The block size is 128, but the title size is 64, actually half of threads in a block are not used. Can you explain about it to me?

ZenoTan commented 2 years ago

Actually the tile size is the number of rows for each block. You can reference DGL's GPU sampling.

yingtongxiong commented 2 years ago

okay， thank you very much

------------------ 原始邮件 ------------------ 发件人: "quiver-team/torch-quiver" @.>; 发送时间: 2022年1月3日(星期一) 凌晨1:07 @.>; @.**@.>; 主题: Re: [quiver-team/torch-quiver] CUDA (Issue #104)

Actually the tile size is the number of rows for each block. You can reference DGL's GPU sampling.

— Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you authored the thread.Message ID: @.***>