Draft

[dearlordylord]

modules needed:

renderer

• 2d/3d
• rendering to show it to the auditory
• and to have visual "tests"; with 3d it'll be a powerful visual testing tool
• figures should move (to demonstrate collisions)
• and somehow created (generated)
• because to show power of GPU computation it's likely that we need more objects
• rendering and computations should be able to exchange data structures without much overhead
• I see it as a "box" of randomly generated objects starting with randomly generated speeds and directions

math model

• 2d/3d
• on typescript and on rust (could be related to Renderer module "exchange data structures" point)
• can be written and tested initially without the 3d module
• webGPU calls are done in JS and not in webassembly
• webassembly could be potentially used for auxiliary math computations?

profiler

Recommendation

Write a 2d math model with a test suite first. It's hopefully would be enough to leverage GPU and to see the difference; we can grow from there

[dearlordylord]

https://surma.dev/things/webgpu/

[dearlordylord]

https://github.com/gfx-rs/wgpu

[dearlordylord]

https://github.com/Sopiro/Physics/blob/45f70c41ea1792e1eb86c0090055483ed7903beb/src/detection.ts

[dearlordylord]

https://cse442-17f.github.io/Gilbert-Johnson-Keerthi-Distance-Algorithm/ great visual no source code

[dearlordylord]

https://pybullet.org/Bullet/phpBB3/viewtopic.php?t=5951

[dearlordylord]

"The GJK algorithm can also be combined with the EPA algorithm (Expanding-Polytope Algorithm) to determine penetration depth [van den Bergen 2004]"

[dearlordylord]

https://github.com/bulletphysics/bullet3/blob/master/docs/GPU_rigidbody_using_OpenCL.pdf

[dearlordylord]

• broad phase / narrow phase

• optimisations are too smart

• non smart shapes would just repeat the nvidia person article