yescrypt: Use AVX512VL XOP-like bit rotates for faster Salsa20

[solardiz]

This speeds up classic scrypt by up to a third, but yescrypt only very slightly. It also reduces code size.

This would only be enabled in builds that support AVX512VL, which I guess is currently uncommon for libxcrypt (even though CPUs that support it are pretty common by now). So maybe in distros like Gentoo where packages are commonly built from source for the local machine?

It shouldn't hurt to have this in the source code anyhow, except that it'd not be part of usual coverage testing - so I suggest occasional testing of a -march=native build on a recent CPU, which would be a good idea prior to these changes as well.

I got the same changes into upstream yescrypt, its fork in JtR tree, and yespower earlier today. The AVX512VL code passed tests in all 3 of these on Intel Tiger Lake.

I did not actually test these changes in libxcrypt yet.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 90.15%. Comparing base (ca15167) to head (dbcfb01).
Report is 6 commits behind head on develop.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop     #207   +/-   ##
========================================
  Coverage    90.15%   90.15%           
========================================
  Files           32       32           
  Lines         3626     3626           
  Branches       689      689           
========================================
  Hits          3269     3269           
  Misses         226      226           
  Partials       131      131           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[besser82]

LGTM!