1. Performance.

Ahoge wasn't exactly slow, but we weren't happy with its performance. Maya's NURBS implementation isn't thread-safe, so I decided to implement my own NURBS surfaces that can run in parallel. This version is almost ten times faster on my Ryzen 7 1700. If you have more cores and faster memory, you should see an even bigger performance leap than this.

However, while it has been tested, it might have issues that I don't know about yet. Feedback is needed.

Because Ahoge is now fully multithreaded, I had to change the random number generation. It's now less "honest", and it's harder to add features without messing with random "seeds". But I think it's worth it.

Ahoge also uses significantly less memory now - almost half as much.

2. Export is extended. Marmoset is now supported. 

Export will have more features in the future.