YUKE WANG

Hi, Thanks for your interest in our work. Here are my answers for your reference. + **Q1:** Based on my understanding of the DGL GCN example, they normally put the...

Thanks for your suggestions. + Q1, you can just simply rewrite the `func` in the `map`, whether it is `norm` or `norm_inverse` will not change the kernel performance, since at...

Thanks for your interest in our project! For warp-aligned design, we basically assign each warp to process a neighbor partition (consisting of fixed number of neighbors and a target node),...

That's is a good question. I haven't tried the design you mention, it might help in some cases. e.g., when the partition size is small and neighbors from two partitions...

From your test result, I think it is as you suggest. I believe more cases such as ``` 1) more than two partitions per warp; 2) more neighbors per partition....

Hi, Zhang, Thanks for your interest of our work and reaching out to us about your questions! + The runtime configuration setting can be determined manually or automatically determined. https://github.com/YukeWang96/OSDI21_AE/blob/f129823ae49f3b557ef525aaa189fc5c703e5c59/GNNAdvisor/GNNA_main.py#L34...

Hi, Zhang, Thanks for following up! + For this public code release, we mainly demonstrate the kernel design and kernel-level runtime design configurability (decider part is outdated, sorry). Therefore, it...

Thanks for reaching out! May I know + What is the platform of GPUs you are using? + Also, Can you run the single-kernel test in the README.md?

Could you please try with Docker method? maybe that helps.

Thanks for your interest in our work. Usually, for those non-divisible cases, we will consider padding by rounding the non-divisible dimension to the round-up divisible number, for instance, if the...