Mixed Add Scheduler Question about cyclone HOT 2 CLOSED

In short, there's no mathematical reason.

The issues are purely practical: how many resources do you need and how much time does it take to do any extra operation. Often time it's also pretty hard to estimate without an actual experiment.

See also these PRs to gnark-crypto, I think they illustrate the problem pretty well:
Consensys/gnark-crypto#249
Consensys/gnark-crypto#261

Even the check "is this point already in the current batch? (and therefore needs to be added to the queue)" needs to be implemented somehow, e.g. by setting a bit to 1 every time you add a point to the current batch... but an extra write in memory is pretty expensive at this level of optimization.

On FPGA we just implemented the simplest solution we could as the deadline for the Zprize was coming fast. It's certainly possible that someone will come up with an improved scheduler.

from cyclone.

Got it! Thanks for the prompt response, Emanuele!

from cyclone.