BLAKE3 VS umash

> might be easier with a public domain license instead of the current ones

There reference implementation is public domain (CC0) or at your choice Apache 2.0

https://github.com/BLAKE3-team/BLAKE3/blob/master/LICENSE

Fyi, blake3 was released in 2019 and should probably be used over blake2 unless you have some strong reason not to. It's basically a reimplementation of blake2 with performance tweaks.

https://github.com/BLAKE3-team/BLAKE3

Many people will argue that today's compilers are so smart/optimized that you'd be a fool to try to outsmart the compiler with asm. I'm not 1 of them, but I know some. IMO it's all a bunch of bullshit, there's a goddamn reason all the cryptocurrency mining CPU/GPU code is all hand-written asm. there's a reason blake3 is written in asm ( https://github.com/BLAKE3-team/BLAKE3/blob/master/c/blake3_sse41_x86-64_windows_msvc.asm ) - but the thing is, 99.99% of the time, life is too short to outsmart the compiler (unless you're Alexander Yee)

I have been trying to optimize my code to create a fast hashing function to create and check b3 file integrity but b3sum is way way faster than my aproach, i have been trying to modify my code acordingly to https://github.com/BLAKE3-team/BLAKE3/blob/master/b3sum/src/main.rs with no luck, so if anyone can give me some tips/clues on how to achieve better speeds it would be incredible. Thx!!

BLAKE3 might be faster than KangarooTwelve and is also an XOF. It doesn't have the benefit of getting a working RFC draft proposal however.

You're saying the hash speed is 133 kB/s? That's extremely slow, for example BLAKE3 achieves 6.8 GB/s which is over 50000 times faster. Nobody wants to use such a slow hash function.

Try this one if you want a smaller, and particularly interesting crate: https://github.com/BLAKE3-team/BLAKE3

blake3 is a cryptographic hashing function, which is used during plotting's "forward propagation" step

BLAKE3 claims to be faster and more secure than both MD5 and SHA1.

More information: https://github.com/BLAKE3-team/BLAKE3

umash (https://github.com/backtrace-labs/umash) has a similar structure PH block structure, but was designed for decent bit mixing (enough to satisfy smhasher, unlike CLHASH, which needs an additional finalizer) with a lower fixed time cost: 22 cycles for a one-byte hash.

I'm not sure how one would use that linear regression. What kind of hardware offers 675 GB/s of memory bandwidth? 140 bytes/cycle is easily more than twice the L2 read bandwidth offered by any COTS chip I'm aware of. There are also warm up effects past the fixed cost of setup and finalizers that slow down hashing for short input. For what range of input sizes (and hot/cold cache state) would you say the regression is a useful model?