Solar Designer

> I guess that previously the alignment just happened to be wrong. I've just tried deliberately misaligning `hash` (by making it a macro). Got only slight slowdown. So that's not...

> I still have an older `john` binary using the asm code, and it still runs slower, as expected. Oh, I forgot about the test vector change. So I can't...

The below speeds it up by 2% or so, but not even close to the level magically achievable with re-enabled asm. I don't know why. The code from intrinsics doesn't...

Also experimented with alignment of the key/context (but putting it into an outer struct and adding a field before) - it matters, but not so much.

> I didn't run valgrind for a while, perhaps time for that. I'd use `perf` if I were not running these tests in a VM. Edit: I mean, I can...

> It's puzzling why you get better speeds with asm, while I get worse. The thing is I was also getting worse speeds with asm (and still am using old...

> my "MbedTLS using asm" baseline is actually [f88688e](https://github.com/openwall/john/commit/f88688e273185bfec79d8d5e11cd419ffe41ed12) meaning it did get `-mavx2` but not the others. Good call, I'll test using the commit before that That's not what...

While I still don't know why for me re-enabling of asm is a lot faster than it was before being disabled, I now managed to achieve even better speeds with...

FWIW, completely disabling `MAY_NEED_TO_ALIGN` actually caused some crashes in our CI bots on my fork of the repo - turns out this also disables the taking care of the alignment...

Pushed the above change to this repo now (passed testing in my fork). I also have the below, not pushed since no measurable speedup (but it probably does speed up...