Long term performance monitoring

[thorstenhater]

As we claim excellent performance, it seem prudent to

1. advertise with concrete numbers
2. keep track of regressions / improvements over time

Thus the proposal: Implement an automated performance tracker like https://speed.pypy.org/

Implementation

• Identify relevant usecases
• Gather metric (TtS, instructions, ...)
• Setup site
  1. Automatically checkout each commit to main
  2. run benchmarks in scalar, vector, GPU
  3. plot results

Call to Action

Submit relevant usecase