jon-chuang

I am quite curious and interested to contribute to the implementation of the said asynchronous graph.

I may fiddle with UM later. For now, I will focus on getting my implementation working. Could you give a a reply to the rest of my comments unanswered above?...

I think it should be called graph. So graph.submit(..) etc. There should also be some tools to visualise this graph.

That would be great. I will hopefully carve out some time to work on this, I think this is an important project for the future of computation. As a side...

Hi, how do I tell if multiple kernels are being run? Even though I place my operations in separate queues, it seems they are running in the same stream. Will...

local_small_polys are scratchpad buffers meant to store intermediate results, by the time the second kernel call is invoked, which depends on host-side control flow, there are intermediate kernel calls that...

Hi, it appears I made some errors with allocation. In any case, it seems solved. However, your comments were very helpful to me. Actually, could I ask what is the...

This affine based MSM with a radix sort of the bucket index followed by a vectorised addition tree is available in https://github.com/celo-org/zexe/pull/4

> I assume that Zexe variable base multiexp is inherited from bellman. In section "New advances in multi-scalar-multiplication over elliptic curves" of this [document by Aztec team](https://docs.zkproof.org/pages/standards/accepted-workshop3/proposal-turbo_plonk.pdf) it is shown...

Results for 2^20 pippenger on 8-core Ryzen 4800H w/ 8MB L3 and 4.1GHz sustained all core boost, with SMT on: barretenberg: 520ms (490ms CPU) Zexe: 551ms. The fat to be...