On Wed, Oct 11, 2017 at 1:11 PM, Jason A. Donenfeld <Jason@xxxxxxxxx> wrote:
>
> Indeed the correct functions to use would be siphash_1u32 or
> siphash_1u64, depending. Depending on the popularity of that, we might
> even consider making a siphash_1ulong helper, I suppose.
Yeah, siphash is probably the sanest thing to use.
How bad would it be to use HalfSipHash on 32-bit architectures?
On a 32-bit machine, the full siphash is pretty expensive - big
constants, and lots of 64-bit shifts. And 32-bit machines also tend to
mean "slow machines" these days.
I suspect there's little point in worrying a ton about the 64-bit key,
considering that I think the *input* is generally more guessable than
the output or the key.
Linus
