I am interested in the techniques that developers developing current games are using for their particle rendering and animation system. We switched for GPU quad expansion (particle geometry generation), GPU transformation, GPU particle lighting and shadowing (shadow map sampling at particle center point) in our new engine, as we are targeting a single console platform only. But for PC and cross platform titles, choosing the most optimal way to render particles can be difficult: DX9 (all Windows XP computers) does not support vertex index input semantic and/or custom vertex buffer data fetch, stream out or memexport or geometry shaders. Many DX9 hardware also lacks the support of texture sampling in vertex shader and R2VB is merely a API hack in DX9 (officially only supported by ATI cards). These limitations make it hard to do efficient particle animation, particle quad geometry generation and particle lighting/shadowing on DX9 hardware.
Do you have different path for DX9 particle rendering, or have you designed your title for the lowest common denominator in mind?
Some questions I'd like to ask for game developers about their particle rendering and animation systems:
Particle animation:
- How complex is your animation system? Are you just simulating simple linear movement and gravity or are you using artist defined envelopes (linear, splines, etc) to control vast amount of particle and emitter animation properties on runtime? If you are using GPU to process this data, are you storing the animation envelopes in textures or are you using arrays of constants?
- Are you doing collision detection for the particles or applying various forces to them based on the gameplay (gravity fields, explosion shockwaves, wind from character movements, etc)?
- Do you depth sort your particles? What kind of algorithm are you using? Are you using the last frame sorting information to speed up your algorithm?
- Are you using the CPU or GPU to do the animation (or a hybrid approach)? If you are using CPU, is your animation system multithreaded? Did the complexity of your animation system force you to use CPU to do the animation (or a part of it)? If you are using GPU what technique are you using to store & fetch the result (R2VB, streamout/memexport, vertex texture sampling, etc)?
Particle geometry generation and transformation:
- Are you transforming the particles to the post projection screen space on GPU or on CPU?
- Are you generating the particle quads on GPU or on CPU?
- What quad generation technique are you using for different particle types (point particle, rotated particle, stretched particle, frame/texcoord animated particle, etc)?
- Are you mixing up techniques depending on particle type (for example fixed function GPU point sprites for point particles, and CPU for all other types?)
Particle lighting and shadowing:
- Are you lighting your particles? Do you use similar light model you are using for your geometry or a simplified one (some sort of averaged lighting for example instead of multiple light sources)?
- Are you calculating a single lighting value for the particle center point (or the four vertices) or do you calculate the lighting for each rendered pixel (normal maps? height/density maps?)
- Are you shadowing your particles? Do you use the same data (shadow maps?) generated for geometry to do the shadow comparisons for your particles?
- Are you calculating the shadowing for the particle center point or edge vertices only (sampling shadow maps in vertex shader or software?) or for each rendered pixel in the pixel shader?
- Does your particles cast shadows to geometry? From all light sources? Soft or hard shadows?
Volumetric / soft particles:
- Are you using a method to render volumetric particles (to hide z-clipping on particle plane)?
- Do you use constant particle depth volume size (scaled with particle post projection size) or you do calculate a per pixel depth volume extent (calculated from particle texel alpha/brightness value or stored in different height/density channel in particle material)?
Do you have different path for DX9 particle rendering, or have you designed your title for the lowest common denominator in mind?
Some questions I'd like to ask for game developers about their particle rendering and animation systems:
Particle animation:
- How complex is your animation system? Are you just simulating simple linear movement and gravity or are you using artist defined envelopes (linear, splines, etc) to control vast amount of particle and emitter animation properties on runtime? If you are using GPU to process this data, are you storing the animation envelopes in textures or are you using arrays of constants?
- Are you doing collision detection for the particles or applying various forces to them based on the gameplay (gravity fields, explosion shockwaves, wind from character movements, etc)?
- Do you depth sort your particles? What kind of algorithm are you using? Are you using the last frame sorting information to speed up your algorithm?
- Are you using the CPU or GPU to do the animation (or a hybrid approach)? If you are using CPU, is your animation system multithreaded? Did the complexity of your animation system force you to use CPU to do the animation (or a part of it)? If you are using GPU what technique are you using to store & fetch the result (R2VB, streamout/memexport, vertex texture sampling, etc)?
Particle geometry generation and transformation:
- Are you transforming the particles to the post projection screen space on GPU or on CPU?
- Are you generating the particle quads on GPU or on CPU?
- What quad generation technique are you using for different particle types (point particle, rotated particle, stretched particle, frame/texcoord animated particle, etc)?
- Are you mixing up techniques depending on particle type (for example fixed function GPU point sprites for point particles, and CPU for all other types?)
Particle lighting and shadowing:
- Are you lighting your particles? Do you use similar light model you are using for your geometry or a simplified one (some sort of averaged lighting for example instead of multiple light sources)?
- Are you calculating a single lighting value for the particle center point (or the four vertices) or do you calculate the lighting for each rendered pixel (normal maps? height/density maps?)
- Are you shadowing your particles? Do you use the same data (shadow maps?) generated for geometry to do the shadow comparisons for your particles?
- Are you calculating the shadowing for the particle center point or edge vertices only (sampling shadow maps in vertex shader or software?) or for each rendered pixel in the pixel shader?
- Does your particles cast shadows to geometry? From all light sources? Soft or hard shadows?
Volumetric / soft particles:
- Are you using a method to render volumetric particles (to hide z-clipping on particle plane)?
- Do you use constant particle depth volume size (scaled with particle post projection size) or you do calculate a per pixel depth volume extent (calculated from particle texel alpha/brightness value or stored in different height/density channel in particle material)?
Last edited by a moderator:
