Skip to content

Latest commit

 

History

History
50 lines (44 loc) · 2.43 KB

README.md

File metadata and controls

50 lines (44 loc) · 2.43 KB
Raw

GPU-Accelerated Particle Emitter

Currently in a very early state and in active development.

Here's a video with more action (click for youtube link):

Youtube Video

TODO LIST

  • Modular properties.
    • Instead of hard-coded properties, we can add/remove any properties which gets compiled into the compute shader. This is not just more performant but also allows for more fine-grained control.
    • Example: If I want particles to start at a certain/random rotation I can make one property for that and then also an interpolation property so all particles rotate the same but look different.
  • Texture pool - selects texture for each particle from a pool of textures
  • Fix blending. 1 pass with multiply blending, depth writing ON but testing OFF.
  • Fix random distribution in some spawn shapes
  • Toggleable HDR blur pass
    • Color data in particles is already HDR, just need to do a HDR pass.
  • Keyframes
    • Not sure yet how to implement this, if per property or per emitter?
    • CSS style keyframes.
    • Maybe this could just be named % values and properties can select these for begin & start'
  • Look over random distributions (mainly normal & extremes)
  • GLSL code injection
    • Load file with some glsl function to add an extra custom stage in the particle pipeline.
    • We could inject custom code for spawn stage for more fine-tuned spawning.
  • 3D Particles
    • UV map & texture map?
    • Primitives: sphere, cube, pyramid
    • Custom shape?
    • Model loaded from disk (engine doesn't have model loading yet)
  • Optimize
    • Something makes emitter a bit slower when looping is enabled
    • Cache computations where possible
    • Allow explicit sync between compute & draw
    • Multiple configs for one emitter; shared SBO.
      • Would require explicit synchronization between each compute/draw call
    • We could give an option when compiling the emitter to use up more vram to cache computations for particles which are the same throughout their lifetime such as constants, start_pos, particle_seed etc. Could drastically improve performance at the cost of a lot of vram when simulating many particles.
    • We can pack the Particle struct and save a lot of VRAM