Possibly better predictor than qoi2avg

[p0nce]

That would changes qoix bitstream, how many bits does it win?

[p0nce]

Ugh, need to change qoi2avg without introducing that a new diff though :(

[p0nce]

Quite spectacular:

Before TOTAL decode mpps encode mpps bit-per-pixel 165.00 108.39 8.29501

TOTAL time = 2.56777 secs

After TOTAL decode mpps encode mpps bit-per-pixel 150.65 96.30 7.94081

TOTAL time = 2.67702 secs

That's a 4.2% bitrate reduction for a 10% slowdown.

[p0nce]

Can push it further by having a different predictor for first pixel in each row.

TOTAL decode mpps encode mpps bit-per-pixel 145.97 97.22 7.93607

TOTAL time = 2.74237 secs

[p0nce]

With a few obvious optimizations:

TOTAL decode mpps encode mpps bit-per-pixel 166.17 91.49 7.93607

TOTAL time = 2.68148 secs

[p0nce]

• [x] Predict whole scanline in encoder in advance, should be faster, can do it 2x pixels at once => wasn't faster, but converting input scanline at once is worth it
• [x] can we adapt that to QOI-10b? => didn't worked there, it can win 0.5% but lot slower. Meh