Solar Designer

> This issue might be partially? fixed by [4700800](https://github.com/openwall/john/commit/47008003c1b0494cfbe183b60649e068e51c4b59) - need to review/re-test. The changes in there look like they'd allow up to exactly 4G (if the mask multiplier happens...

@magnumripper What you call "@witee-star's version in diff format" is actually identical to what I had posted in https://github.com/openwall/john/issues/5078#issuecomment-1047165841 Ideally, we'd make more extensive changes to the script as it's...

> should I toss the two band-aid fixes together with a PR for now? Please feel free. We might make more elaborate changes later. Thanks!

> Merge branch 'bleeding-jumbo' into authenticator2john Let's not be doing things like that. Let's be rebasing instead.

This is now implemented via #5292. I mention in case anyone in here wants to help test it, or just use it.

> There is an MIT-licensed OpenCL implementation here: > > https://gitlab.com/vojtapolasek/argon2-gpu The above is a fork of https://gitlab.com/omos/argon2-gpu and that one is now heavily reworked and integrated into JtR by...

I need to add that I'm aware of the discussion on #16 and the corresponding explanation in a Medium post (need to rediscover and re-read it). So yes, the 32-bit...

Regarding floating-point, I don't buy the argument about NaNs given in https://medium.com/@ifdefelse/the-cost-of-asic-design-a44f9a065b72 - it would apply for arbitrary mix of operations including floating-point ones, but we can instead limit ourselves...

Thank you for the comments, @ifdefelse. I think we're on the same page. Yes, it's tough to target multiple GPU architectures with integer MADs - they might very well need...

@ifdefelse All good points, thanks! I think we're effectively starting to compile a list of constraints or a checklist for a good design using FP, which will be far from...