Confusion about block balance #2
Description
Hi authors, I am glad you open-sourced your code repo. I have some questions regarding the details of your solution.
In the current version of your paper on arXiv, I couldn't find any discussion about the balance mechanism. However, it appears to be significantly implemented in the code (e.g., this line) and seems to contribute notably to the speedup demonstrated in this result file.
From my understanding, this may involve techniques similar to RoDE. Will this aspect be covered in more detail in your final PPoPP version? Does the selected best kernel contribute to the final data point you demonstrated in the paper? Understanding this implementation is crucial for reproducing the reported metrics.
Thank you for your time! I appreciate any clarification you can provide.