Interface ParticleGraphConfig

Sub-emitters triggered by parent particles. Each becomes its own ParticlePass instance fed by a GPU spawn-request queue.

Baked SDF bound to the particle pass. Required when the graph uses sdf_attractor, sdf_collision, or an sdf_surface spawn shape; ignored otherwise. Exactly one SDF per system (multiple SDF modifiers share it).

Patch (grid) layer for the agent-modeling use case. When set, the update compute binds a diffusing/evaporating grid that patchSample / patchDeposit nodes read and write. See PatchConfig.

Spatial hash enabling neighbor queries (the boids modifier). Built each frame from live particle positions before the update runs. See SpatialHashConfig.

Per-agent link slots for the link_springs modifier. Host uploads the topology via ParticlePass.setLinks. See LinksConfig.

Per-agent f32 scratch array — the scratchGet/scratchSet/loop IR nodes. Lets an agent hold a small working array (a tour, a history); enables parallel local search. Host seeds/reads via ParticlePass.setScratch / ParticlePass.readScratch. See ScratchConfig.

Enables the GPU reproduction free-list (a stack of dead slots) so modifiers can spawn/kill agents within the same population at runtime — repro_spawn / repro_kill, used by lotka_volterra. Seed the initial population with ParticlePass.seedAgents (the emitter is typically disabled, spawnRate 0).

When false, the update skips the position += velocity * dt integration step. Position-movement agent models (which move via forward/setxy, not velocity) can set this to repurpose the velocity field as a free per-agent vec3 the step writes and the renderer reads — e.g. a terrain surface normal for renderer.tiltToSurface. Default true (normal particle integration).