Solar Designer

> it looks like you can remove this limitation on CPU format if you disable SIMD. Specifically, this is done by the `--disable-simd` option to `configure`, e.g. `make distclean &&...

> We could alternatively be printing the format's maximum on each run just in case the user would find it unexpected. Not as a warning, but as an informational message....

Is the OpenCL "driver" upgrade the only change? I made relevant changes in July, so if your prior testing was before July, then those changes I made could also be...

In other words, it worked yesterday and stopped working today with only the OpenCL "driver" having changed, correct?

Somehow we saw even more/different formats fail in: https://github.com/openwall/john/pull/5382#issuecomment-1838999062

It's good to have this recorded, but I see no need to track it as an issue.

Oh, found that we actually mostly use the same 4 operations version in scalar code as well: ``` sha2.c:#define F0(x,y,z) ((x & y) | (z & (x | y))) ```...

So this gives me slightly smaller code and about 1% speedup at sha256crypt, sha512crypt, Bitcoin on AVX2: ```diff +++ b/src/simd-intrinsics.c @@ -1783,7 +1783,7 @@ void SIMDSHA1body(vtype* _data, uint32_t *out, uint32_t...

Maybe there's a reason why Wei Dai called this "SHA Maj trick", not "SHA-2 Maj trick": SHA-1's H() is the same as SHA-2's Maj(). However, trying to apply this trick...

> trying to apply this trick to our SHA-1 for AVX2 actually results in slight code size increase The code size increase is avoided by also reordering the arguments (e.g....