It turns out that picking the best fit palette subset from a larger fixed palette is much harder than I expected.
(Hand optimized palette top right, automatic palette left)
Here's a try at the Mechanical Age using @musurca's picad. https://t.co/lZw7wG3lMV
The data for this takes up 1536 bytes, so I could fit roughly 14 images in cart data (~22kb).
Myst has around 2500 images... just off by a few orders of magnitude.
#pico8
Injecting level data into the sprite sheet knocked compressed data nicely below the limit. (I was using strings before.)
Maybe I can squeeze just enough extra tokens to make this a game!
#pico8
Well, I tried some more aggressive acceleration structures: pre-building tables for the distance function and normals at every pixel.
I got about a 3x speedup.
But the overall quality of the light rending is much lower.
Lesson? Sub pixels are important.
This compression level on the left is fine, right? #PICO8
I just about got Huffman coding working with wavelet compression. Following this page from @iquilezles https://t.co/RssfSP1MHT
