Closed Rhett-Ying closed 7 months ago
@mfbalin FYI
@mfbalin could we replace it with std::partial_sum
? how much is the performance downgrade?
or any counterpart from torch
? https://github.com/pytorch/pytorch/blob/bb04f3f66a5b92f0bed3712689f57774f00db349/aten/src/ATen/cuda/cub.cuh#L303
Let's replace it with partial_sum, opening a quick PR.
I was eagerly waiting for today's regression results. When do you think the next report is going to go out?
regression build is not affected by this issue. But regression build is down last night due to other issue. I've fixed it and report is supposed to be ready tmr, around 1 o'clock AM, UTC.
🔨Work Item
IMPORTANT:
Project tracker: https://github.com/orgs/dmlc/projects/2
Description
For now, we're using
nvidia/cuda:12.1.0-cudnn8-devel-ubi8
docker base image for building. And the default gcc version is8.5.0
. When building with it, below error is thrown:This API is introduced by https://github.com/dmlc/dgl/pull/7239
Depending work items or issues