Adam Stylinski

Another thing going for this is that the reason it wasn't working is I was doing what the code was already doing, deriving the post sorted values. So, there's _some_...

Sure, I'll check it out today and see the improvements. The mention of zmm registers I assume means everything is still AVX512 only - is this sort implementation possible on...

Hmm, I'm not seeing any significant improvements from this. I am using the key-value sort explicitly rather than argsort in my code, does that matter? edit: Ok, I am seeing...

> What does your data distribution look like? It's 3d point cloud data that's likely to have a partial amount of order to it. The point locations themselves are used...

I can provide a dataset, if that would help. I'm fairly confident sharing it would be acceptable given that they are just arbitrary voxel indices.

So here's a smaller example. A quick note, this does happen to be faster overall because it applies the sorting to the keys, which is why I'm using it over...

Any progress on testing this data set?

Good catch, these niche CPUs I think are a rare example where all of those extensions didn't come along for the ride. A more appropriate fix might be to store...

> > A more appropriate fix might be to store flags for each feature and and all of those together in the function table. > > OK. @KungFuJesus Let me...

> > OK. @KungFuJesus Let me confirm if I understood you: are you suggesting to replace `features->has_avx512` with the following 3 new features? `features->has_avx512f`, `features->has_avx512dq` and `features->has_avx512bw`. > > I...