A pure, low-level tensor program representation enabling tensor program optimization via program rewriting. See the web demo at https://gussmith23.github.io/glenside-web-demo/
This is really interesting. It is in many ways a "wildcard" for implementing access patterns. It has a specific set of access patterns it can actually implement, though, and it would be great to model that.
Scott's tensor sequencer can read tensors intelligently, creating different shapes on the fly.