Oscar Smith

IMO, this really highlights that RUN_16 is a major waste of opcode space. https://github.com/phoboslab/qoi/pull/41 shows an alternative approach that just encodes all runs using RUN_8 and uses repeated RUN_8 to...

Any thought son https://github.com/nigeltao/qoi2-bikeshed/issues/12#issuecomment-981346787? I think this is a better re-ordering of the opcodes that slightly increases the computational cost of the index nodes, but gets 2 extra bits for...

One possible direction to go with this info is to have only 2 `QOI_COLOR` options. tag `0xff` for `RGBA` and `0xfe` for `RGB`. This would open up a bunch of...

@rmn20 somewhere it was noted that QOI_COLOR_16 isn't currently being used optimally, since it is not chosen over QOI_DIFF_24. Can you check how much of a difference this makes?

It's worth checking, but my guess is that DIFF_8 is probably worth keeping. That said, we should probably have a QOI_Experimental repo (or branch) so we have a good baseline...

I think focusing on paletted images probably doesn't make sense. BMP already fills that niche very well, and trying to compete there would likely be fairly detrimental to other types...

The downside of this idea is that it will be a non-negligable encoding and decoding slowdown. I think the easier approach is to use `QOI_INDEX` to store runs of length...

The place where it's a bigger change is if we increase the tag size of `QOI_RLE` (which might help fit other opcodes in). I think 4 bits for RLE is...

Free speedups are always nice.

Here would be my proposed encoding format. The key difference is that it sacrifices some of the cache and RLE codes to get a full 5 bits for each color...