From Weiqing:
Here is my proposal, this repository can be a collection of often used code in our research. For instance, a categorical distribution sampler.
It would be nice if we can provide not only some well-documented fast and working functions, but also some benchmarks.