Milan Klöwer

Thanks for sharing the data, I see several problems with it:  1. Zooming into the Fram Strait. I don't know which interpolation you've used here, but for some reason...

I see that due to the convergence of the grid lines on Greenland, the plot you first shared has Greenland indeed entirely spread out over the top edge of the...

Yes, nothing stops you from assessing the bitinformation for subsets of your data arrays. Whether this is in the vertical, or say you want to use a different number of...

Can you provide an example for why that's an issue that requires special treatment?

No seasonality means you've used the same data as before, but subtracted a mean seasonal cycle? But with and without seasonal cycle this is the bitwise information in longitude? Could...

Maybe there's a way to do this faster, I don't know how Float32(::BigFloat) and the inverse are implemented. But Float32 is a superset of BFloat16 so you don't lose any...

I see because of the cases where a-------x-b--------c, with a,c BFloat16s, b=(a+c)/2, a Float32 then x could be rounded to b::Float32 and subsequently to c even though it would round...

Wouldn't this happen here too? https://github.com/JuliaMath/BFloat16s.jl/blob/ba5104df727cb0069e29ff52236c710e375fc6d0/src/bfloat16.jl#L400-L410

> Nothing breaks on my Intel Core i5-12600K that does not support avx512_bf16. `BFloat16` is for arithmetic operations converted to Float32 and then the result is truncated back down to...

But what happens if you look at the LLVM code? Because for me the same conversion happens there (wtih 1.11) but you're hoping it would call `fadd bfloat` directly? ```julia...