new accelerated primitives - testing needed

new accelerated primitives - testing needed

[Jason A. Donenfeld]

Hey folks,

Samuel Neves and I have just finished the first probably buggy round
of porting a bunch of accelerated primitives into the WireGuard
codebase. This was a pretty big effort. We've got things running now
on: x86_64, SSSE, AVX, AVX2, AVX512, ARMv4, ARMv5, ARMv6, ARMv7,
ARMv8, ARMv7+NEON, ARMv8+NEON. During the first few limited
benchmarks, things are looking positive.

I've got access to a bunch of Intel hardware for testing, but while
I've got a few different ARM boards, the large diversity of the ARM
ecosystem presents a challenge.

Some of you seem to have all sorts of little computers with weirdly
configured kernels and cores, which often find useful WireGuard bugs.
If some of you are up for finding some potential mayhem, would you run
the master branch of WireGuard (built with `make debug`, preferably),
and let me know if things explode?

At the very least, this usually gets things rolling:

$ git clone https://git.zx2c4.com/WireGuard /tmp/wg && make -C
/tmp/wg/src insert && dmesg | tail -n 15

Jason