Palette needs to be power of 2 and multiple of 3?

[deanm]

Palette needs to be power of 2 and multiple of 3?

[deanm]

                var palette2 = new Uint8Array( 32 * 3 );

Seems to work fine. From what I saw in my test your palette was 24 colors.

Sorry, I will try to clarify the code up a bit. There is two different aspects.

1. GIF format enforces a palette of power of 2 (8, 16, 32, 64, etc) number of colors.
2. omggif takes 3 arrays elements for each color in the palette [r, g, b, r, g, b, ...]. This could (and maybe should / will) be changed to [rgb, rgb, ...].

So for now really what you need is palette.length / 3 to be a power of 2. omggif checks for both of those things.

So what you really want is to pass omggif something like 64*3, 32 * 3, etc, since there are separate elements for RGB.

I just wrote this code today, now that you mention it I will probably change the palettes to be RGB together, it's a bit more convenient when you are writing things out by hand to (0xff78fa vs 0xff, 0x78, 0xfa) and it means less array elements which is probably not a bad thing.

Hope that helps

[mrdoob]

Oh damn. I was missing the * 3...

So now it kind of works:
https://gist.github.com/4582927

It shows the top half of the gif fine, the second half gets corrupted.

[deanm]

I have a look at it tomorrow, thanks for the code to reproduce

[deanm]

I think it should be fixed in 99faac4.

[mrdoob]

Cool! Works fine now :)

[mrdoob]

Some tests:

[deanm]

I just spent a few hours optimizing (yesterday I just wrote the naive initial implementation). There is about a 4x performance increase in the LZW compressor. It made the code slightly trickier, so let me know if you see any problems. I will spend a little bit more time looking at optimizations, although I've tried a few and I think I am pretty close to the balance of making the code very ugly for only helping a few percent.

PS: Thanks for the examples! It works! ; )

[mrdoob]

Wow. It's much faster compared to this morning. It's almost realtime now.
http://mrdoob.com/lab/javascript/omggif/

[deanm]

I just pushed another improvement of 15%, but it's getting to the manually unrolled/inlined point, so I think I'll call it good for now. Overall it's about 5x yesterday's code.

[mrdoob]

Impressed