This version skips out on a few things to try to make encoding a little faster (for example, it uses fixed Huffman tables, ...), and is intended mostly for things like video-capture (where speed is slightly more important than compression or quality).
It hasn't really been well tested yet (say, with a variety of images at different resolutions), but basic tests seem to show it "still working".
Note that this version isn't thread-safe.
in my current tests though, it is currently pulling off around 50 Mpx/s encoding on my computer (basically, re-encoding a 512x512 frame at around 200 frames/second, or approx: a 360p image at around 290 frames/second, or 720p at 54 frames/second).
I may still see if it can be made much faster, and maybe look into making a thread-safe version (say, so that interleaved multi-threaded encoding can be used).
the reason for this is mostly so that hopefully in-engine video encoding can be made a little faster.
I have since discovered/fixed a few minor issues, as well as modified it to be thread-safe, but haven't put it on pastebin yet.
I have noted that exact performance depends somewhat on the image being compressed.
this version adds the capability of being used from multiple threads, and a slight bit more work trying to fine-tune it, before I mostly got distracted and was working on other stuff...
as for in-engine recording: it also seems to be working well enough at 18fps (a little faster than before). I tried 24fps recording, and while this worked, it ate up a bit more HDD space than I was happy with (though dropping the image quality could be an option here). 18fps was chosen as the average of 12 and 24.