Diego F. Aranha

I replaced the affected lines in `benchmark_module.h` with ``` #ifndef BENCHMARK_MODULE // define new benchmark type for benchmark module extern PyTypeObject BenchmarkType; // define new benchmark error type (will be...

Incredible how much faster pairing implementations have gotten in the past few years. Keep up the good work, you all! :)

Touting my own horn, but [RELIC](https://github.com/relic-toolkit/relic) also supports Fiat-Crypto as a field arithmetic backend.

Hey, I was not aware of this issue. Thanks for reporting! Yes, we can bundle these conversions to the divstep precomp constant, although the gains in performance will be small....

"Comba" means the product-scanning approach that saves on memory operations, which we still haven't implemented. In a product-scanning multiplication, each column of the result is solved per iteration of the...

Jasmin indeed requires reg/stack annotations in the source, and the current compiler is unfortunately not able to spill in case there are too many reg variables.

I have been running some performance experiments in finite field arithmetic between my own Assembly code from [RELIC](https://github.com/relic-toolkit/relic), verified Jasmin and code generated by Fiat-crypto and the differences are pretty...

Completely understandable, I meant my comment more as a mental note. We might give it a try at some point! :)

+1 to Zulip, as I find Discord a bit too distracting

Quick notes: * Inversion is too slow now because our faster jumpdivstep-based algorithm is not in the repository yet. One can get a sense of the performance speedup in Table...