Use the SHA-3 implementation from tiny-keccak

[mbrubeck]

This is faster than the implementation in rust-crypto, has fewer dependencies, and is more actively maintained.

[CLAassistant]

All committers have signed the CLA.

[sorpaas]

Thanks @mbrubeck!

[coveralls]

Coverage remained the same at 76.27% when pulling a4d9923 on mbrubeck:sha3 into 8669a90 on ethereumproject:master.

[coveralls]

Coverage remained the same at 76.27% when pulling a4d9923 on mbrubeck:sha3 into 8669a90 on ethereumproject:master.

[newpavlov]

I was a bit late, but if you are interested I can create PR with the sha3 crate instead of tiny-keccak. It's a bit faster asymptotically (at least on my machines) and is part of the bigger project.

[sjmackenzie]

Thanks @mbrubeck, @newpavlov please read the CONTRIBUTING.md file (currently it's pointing at the incorrect domain but there are two processes called 1/C4 and 2/COSS here: www.github.com/ethereumproject/rfc) You will see that you do not need to ask permission, please join in and contribute your patches, they will be merged quickly. @mbrubeck as you've made a patch, could you please read over 1/C4 and if you agree to working according to 1/C4, i'd like to extend an invitation to you to join the sputnikvm maintainers group.

[sorpaas]

@newpavlov We're currently using keccak for two occasions. The first one is to hash an address (20 bytes) together with a nonce (32 bytes), using RLP. That is something around ~52 bytes (a few additional bytes are there due to RLP). The second one is the SHA3 opcode, which hashes exact 32 bytes. According to debris/tiny-keccak#14, it seems that when hashing smaller amount of bytes (around 32), tinny-keccak is faster than sha3?

But please feel free to create an issue or a PR. We can do some banchmark!

[newpavlov]

Yes, for 52 bytes input tiny-keccak is faster by 40-50 ns on my laptop (802 vs 844). I did not see this difference substantial, so I have not investigated how to make it run faster. Copying data to buffer and running keccak's f directly results in the ~750 ns.

[ghost]

@newpavlov we also look to an alternative to rust-crypto for our anouther project emerald-rs, so thank you. I will look at source code and choice more transparent impl.