<jpk68:matrix.org> Are there any benchmarks of CryptoNightR JIT performance compared to the interpreter?

Now that's a name I haven't heard in a loong time...

CryptonightR has almost the same hashrate as regular Cryptonight

Interpreter is 2-3x slower, iirc

<jpk68:matrix.org> Thanks. I want to try and work on ARM/RISC-V backends for the JIT eventually, just have to become less bad at assembly first...

<jpk68:matrix.org> Cuprate's pure Rust implementation of CryptoNight (all of the variants) doesn't have a JIT at all for v4/R currently

I implemented cryptonight (and its variants for monero) on git.gammaspectra.live/P2Pool/consen…rc/branch/master/monero/cryptonight as interpreter (with an optimized switch loop for R) with some mild assembly/intrinsics for some operations, for most block verification purposes this was enough, same for opening/closing derived wallet

files

CryptonightR was used on a small interval to cover some time until RandomX was upgraded to, so maybe that's fine for the verification usecase?

<boog900> Yeah for us we don't really need a JIT. Fast sync will skip it all together and full verification sync is more than fast enough compared to monerod.