Constant-time proof creation

General
1

@daira I was watching the Arborist Call from last week on YouTube. You mentioned that proof generation was deliberately not constant time because of performance hit.

Our kernel (that we previously put up for a grant but was denied because “it wouldn’t be merged into Orchard”) achieves comparable (~0.9x-1.1x) proof times on ARM purely constant-time. We don’t even use the GLV endomorphism (that Pasta was chosen for) that we used before now. Our claims drop from “30-40% speedup non-CT + GLV” to “~very low regression even in constant time”.

Would you like to talk about it? I can soon provide you the code to look at and test directly, and you can tell the grant committee whether to approve our grant or not based on PR probability.

1 Like
2

That sounds interesting, yes.

3

Can I email you later about it? What timezone are you active in (I assume EST?)

Thanks for responding!

4

I’d prefer Signal (dairaemma.31). I’m in the UK (UTC+1).

1 Like
5

(Wrong on two counts, but alright!)
Where I am, it’s not even morning yet, (speaking of, where you are, it’s midnight?), so I’ll make sure to message you tomorrow!