feat: improve tex packer speed by like 3x

[dragoncoder047]

speeds up the tex packer by like >=3x i think

on textureoverload playtest, loading ~900 sprites of various sizes:

• chrome: 1113 --> 74 ms
• safari: 3697 --> 1273 ms

@Stanko can probably make a better torture test though, would be interested in test using spacedeck x's spritesheets with repacking enabled

test is failing because pupetteer is broken, idk why

• Changeloged

[Stanko]

Amazing work!

I don't know why the old algorithm struggled on my sprites. Safari took 50 seconds to pack them. As of now I have 99 sprites and here are the results:

Browser	Alpha 27.1	New	My Vite plugin (both branches)
Chrome 6	5 000 - 7 500	200 - 250	150 - 200
Firefox	23 000 - 25 000	200 - 250	100 - 150
Safari	45 000 - 50 000	500 - 600	300 - 350

This is an insane upgrade and with this I can even remove the Vite plugin (simpler the better, right). Plugin is still a bit better, and I'll keep it until this is released and tested a bit more.

Please note that I'm testing on MacBook with M3 Pro chip and 36 gigs of RAM.

[dragoncoder047]

Please note that I'm testing on MacBook with M3 Pro chip and 36 gigs of RAM.

I am testing on an M3/16G RAM and also have no clue why Safari is so much slower. You would think the first-party OpenGL->Metal translation would be the fastest, but apparently not.

[dragoncoder047]

I don't know why the old algorithm struggled on my sprites. Safari took 50 seconds to pack them. As of now I have 99 sprites and here are the results:

fwiw @Stanko this was because I used a brute force implementation that checked every other sprite for collisions which is $O(n^2)$ but what I did now uses a hashgrid to check collisions so it's approximately $O(n\ \log\ n)$ now.

ChatGPT is saying it's $O(n^3)$ to $O(n^2)$, but that's still an improvement.