Moritz Mœller

Indeed, Krita or GraphicsMagick/ImageMagick or the like will do your bidding here. The former is how I generated all those from a single `F16x4` source. P.S.: I can btw. also...

Btw, if there are any theoretical speed gains when doing e.g. three times a `F32x3` vs. doing a `F32x9`? I.e. would it make sense, speed-wise, to support higher numbers than...

> I think it will be possible with help of converting a f16-image to f32-one. As I know, modern CPUs haven't instructions to operate with f16 data type. Rust [will...

Also, there are finally some crates that do not render to stupid, non-linear `u8`. :wink: E.g. Google's [`forma`](https://crates.io/crates/forma-render) uses an [`[[f16; 4]]`](https://docs.rs/forma-render/0.1.3/src/forma_render/styling.rs.html#273-284) as storage for it's buffers.

Why is this still not merged? The DB shipped with this crate is now over a year old. Is there anything we can do to help? I could also make...

Fixed a bug in my impl. I've been using my fork w/o (other) issues now since Sep. 9, when I submitted this PR originally. I also added tests for this...

One low hanging fruit is also to add support for `nightly` native `f16` behind a `nightly_f16` flag. This breaks Cargo rules since feature flags must be additive and we can...

I saw the original PR had failing tests in CI during the build step. Sorry about that. The last commit should have fixed this.

> ## This PR > Now, to this PR :D I had a quick look. The code looks promising. But the diff is huge. I see a lot of it...

> ## DWA > I recently tried letting AI transcribe the DWA compression from C. I didn't ask it for transcription, I had no luck with that. I asked it...