instancing

GPU-driven instancing: a compute pass updates per-instance transforms in a storage buffer, and one instanced draw renders them all. The per-instance layout lives in examples/instance2d.rs and mirrors QuadInstance in the Slang shaders.

cargo run --features examples --example instancing

What it demonstrates

  • Compute-updated instance data consumed by a raster pass
  • Matching host and shader struct layouts
  • One draw call for many objects

Source

examples/instancing.rs:

//! Instancing example - render many objects efficiently.
//!
//! Demonstrates retained scheme with compute dispatch → offscreen render → copy-to-present.
//!
//! Run with: cargo run --example instancing

use anyhow::Result;
use goldy::{
    Buffer, BufferFlags, BufferKind, Color, ComputePipeline, DepositTarget, DepositTransaction, Instance, Lease,
    LeaseRenderTarget, MemoryExchange, NodeAccess, PrimitiveTopology, RenderPipeline, RenderPipelineDesc,
    RequestAdapterOptions, RuntimeDescriptor, Scheme, ShaderModule, SurfaceConfig, SurfaceExchange, TargetLoad,
    Texture, TextureFormat, Transaction, VertexBufferLayout,
};
use std::ops::Shr;

mod instance2d;
use instance2d::Instance2D;
use std::sync::Arc;
use std::time::Instant;
use winit::{
    application::ApplicationHandler,
    event::WindowEvent,
    event_loop::{ActiveEventLoop, ControlFlow, EventLoop},
    keyboard::{Key, NamedKey},
    window::{Window, WindowId},
};
mod common;
use common::CaptureDump;

const GRID_SIZE: u32 = 20;
const QUAD_SIZE: f32 = 0.03;
const NUM_QUADS: u32 = GRID_SIZE * GRID_SIZE;

#[goldy::gpu]
struct AnimParams {
    time: f32,
    delta_time: f32,
    total_instances: u32,
}

fn main() -> Result<()> {
    tracing_subscriber::fmt()
        .with_env_filter(
            tracing_subscriber::EnvFilter::try_from_default_env()
                .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new("warn")),
        )
        .init();

    if common::capture_requested() {
        let mut state = RenderState::new(None)?;
        while !state.capture_done() {
            state.render()?;
        }
        return Ok(());
    }

    println!("Goldy Instancing Example - {} quads (Scheme + Present)", NUM_QUADS);
    println!("Press Escape to exit");

    let event_loop = EventLoop::new()?;
    event_loop.set_control_flow(ControlFlow::Poll);

    let mut app = App::default();
    event_loop.run_app(&mut app)?;

    Ok(())
}

#[derive(Default)]
struct App {
    state: Option<RenderState>,
}

struct RenderState {
    window: Option<Arc<Window>>,
    device: Arc<goldy::Runtime>,
    ctx: goldy::Context,
    surface: Option<SurfaceExchange>,
    capture: Option<CaptureDump>,
    readback: Option<Texture>,
    present: Option<Transaction>,
    scheme: Scheme,
    scene_rt: Lease<LeaseRenderTarget>,
    compute_pipeline: ComputePipeline,
    render_shader: ShaderModule,
    render_pipeline: RenderPipeline,
    instance_buffer: Buffer,
    params_buffer: Buffer,
    upload_scheme: Scheme,
    params_deposit: DepositTransaction,
    start_time: Instant,
    last_time: f32,
    frame_count: u32,
}

impl RenderState {
    fn create_render_pipeline(
        device: &goldy::Runtime,
        render_shader: &ShaderModule,
        format: TextureFormat,
    ) -> Result<RenderPipeline> {
        common::render_pipeline(
            device,
            render_shader,
            format,
            RenderPipelineDesc {
                vertex_layout: VertexBufferLayout::empty(),
                topology: PrimitiveTopology::TriangleList,
                ..Default::default()
            },
        )
    }

    fn bind_frame(
        scheme: &mut Scheme,
        scene_rt: &Lease<LeaseRenderTarget>,
        surface: Option<&SurfaceExchange>,
        readback: Option<&Texture>,
    ) -> anyhow::Result<Option<Transaction>> {
        if let Some(surface) = surface {
            let present = surface.bind_render_target(scheme, scene_rt)?;
            Ok(Some(present))
        } else {
            let readback = readback.expect("capture readback");
            scheme.copy_to_texture(scene_rt, readback)?;

            Ok(None)
        }
    }

    fn record_scheme(
        scheme: &mut Scheme,
        compute_pipeline: &ComputePipeline,
        render_pipeline: &RenderPipeline,
        instance_buffer: &Buffer,
        params_buffer: &Buffer,
        scene_rt: &Lease<LeaseRenderTarget>,
    ) {
        scheme
            .node("update_instances", compute_pipeline)
            .with_parcel(instance_buffer, NodeAccess::ReadWrite)
            .with_parcel(params_buffer, NodeAccess::Read)
            .dispatch(NUM_QUADS.div_ceil(64), 1, 1);

        let bg_color = Color {
            r: 0.02,
            g: 0.02,
            b: 0.04,
            a: 1.0,
        };

        let mut pass = scheme.render_pass("instancing", scene_rt, TargetLoad::Clear(bg_color));
        pass.with_parcel(instance_buffer, NodeAccess::Read);
        pass.set_pipeline(render_pipeline);
        pass.draw(0..6, 0..NUM_QUADS);
        pass.finish();
    }

    fn target(&self) -> (TextureFormat, u32, u32) {
        if let Some(surface) = &self.surface {
            let (width, height) = surface.size();
            (surface.format(), width, height)
        } else {
            let capture = self.capture.as_ref().expect("capture dump");
            let (width, height) = capture.size();
            (CaptureDump::format(), width, height)
        }
    }

    fn capture_done(&self) -> bool {
        self.capture.as_ref().is_none_or(CaptureDump::finished)
    }

    fn rerecord_scheme(&mut self) {
        let mut scheme = Scheme::new(&self.ctx);
        let (format, width, height) = self.target();
        if let Ok(rt) = self.ctx.lease_render_target(width.max(1), height.max(1), format, None) {
            self.scene_rt = rt;
            Self::record_scheme(
                &mut scheme,
                &self.compute_pipeline,
                &self.render_pipeline,
                &self.instance_buffer,
                &self.params_buffer,
                &self.scene_rt,
            );
            if let Ok(present) = Self::bind_frame(
                &mut scheme,
                &self.scene_rt,
                self.surface.as_ref(),
                self.readback.as_ref(),
            ) {
                self.present = present;
                self.scheme = scheme;
            }
        }
    }

    fn new(window: Option<Arc<Window>>) -> Result<Self> {
        let instance = Instance::new()?;
        let device = Arc::new(
            instance
                .request_adapter(&RequestAdapterOptions::default())?
                .request_runtime(&RuntimeDescriptor::default())?,
        );
        let ctx = device.create_context()?;

        let (surface, capture, readback, format, width, height) = if let Some(window) = window.as_deref() {
            let surface = SurfaceExchange::new(&ctx, window, SurfaceConfig::default())?;
            let format = surface.format();
            let (width, height) = surface.size();
            (Some(surface), None, None, format, width, height)
        } else {
            let capture = CaptureDump::from_env()?;
            let (width, height) = capture.size();
            let readback = common::capture_readback(&device, width, height)?;
            (
                None,
                Some(capture),
                Some(readback),
                CaptureDump::format(),
                width,
                height,
            )
        };

        let compute_shader = ShaderModule::from_slang_with_gpu_types(
            &device,
            include_str!("../shaders/instancing_update.slang"),
            &[Instance2D::GPU_TYPE, AnimParams::GPU_TYPE],
        )?;
        let render_shader = ShaderModule::from_slang_with_gpu_types(
            &device,
            include_str!("../shaders/instancing_render.slang"),
            &[Instance2D::GPU_TYPE],
        )?;

        let mut instances = Vec::with_capacity(NUM_QUADS as usize);
        for i in 0..GRID_SIZE {
            for j in 0..GRID_SIZE {
                let nx = (i as f32 / (GRID_SIZE - 1) as f32) * 2.0 - 1.0;
                let ny = (j as f32 / (GRID_SIZE - 1) as f32) * 2.0 - 1.0;
                instances.push(Instance2D::new(
                    nx * 0.85,
                    ny * 0.85,
                    0.0,
                    QUAD_SIZE,
                    [1.0, 1.0, 1.0, 1.0],
                ));
            }
        }

        let instance_buffer = device.acquire_buffer_with_data(&instances, BufferKind::Scattered)?;
        let params_buffer =
            device.acquire_buffer_sized::<AnimParams>(1, BufferKind::Broadcast, BufferFlags::empty())?;

        let compute_pipeline = ComputePipeline::new(&device, &compute_shader)?;
        let render_pipeline = Self::create_render_pipeline(&device, &render_shader, format)?;

        let mut scheme = Scheme::new(&ctx);
        let scene_rt = ctx.lease_render_target(width.max(1), height.max(1), format, None)?;
        Self::record_scheme(
            &mut scheme,
            &compute_pipeline,
            &render_pipeline,
            &instance_buffer,
            &params_buffer,
            &scene_rt,
        );
        let present = Self::bind_frame(&mut scheme, &scene_rt, surface.as_ref(), readback.as_ref())?;

        let mut upload_scheme = Scheme::new(&ctx);
        let params_deposit = MemoryExchange::new(&ctx).bind_deposit(
            &mut upload_scheme,
            DepositTarget::buffer_elements::<AnimParams>(&params_buffer, 1),
        )?;

        println!(
            "Created instancing example with {} quads (GPU compute + graphics)",
            NUM_QUADS
        );

        Ok(Self {
            window,
            device,
            ctx,
            surface,
            capture,
            readback,
            present,
            scheme,
            scene_rt,
            compute_pipeline,
            render_shader,
            render_pipeline,
            instance_buffer,
            params_buffer,
            upload_scheme,
            params_deposit,
            start_time: Instant::now(),
            last_time: 0.0,
            frame_count: 0,
        })
    }

    fn render(&mut self) -> Result<()> {
        self.frame_count += 1;

        let time = if let Some(capture) = &self.capture {
            capture.time()
        } else {
            self.start_time.elapsed().as_secs_f32()
        };
        let delta_time = time - self.last_time;
        self.last_time = time;

        let params = AnimParams {
            time,
            delta_time,
            total_instances: NUM_QUADS,
        };

        (&self.params_deposit << &params)?;
        self.upload_scheme.submit()?;

        let mut submission = self.scheme.submit()?;
        if let Some(present) = &self.present {
            (&mut submission >> present).take()?;
        } else {
            let pixels = (&mut submission >> self.readback.as_ref().unwrap())
                .take::<u8>()?
                .to_vec();
            self.capture.as_mut().unwrap().write_rgba(&pixels)?;
        }

        if let Some(window) = &self.window {
            window.request_redraw();
        }
        Ok(())
    }
}

impl Drop for RenderState {
    fn drop(&mut self) {
        let elapsed = self.start_time.elapsed().as_secs_f64();
        let fps = if elapsed > 0.0 {
            self.frame_count as f64 / elapsed
        } else {
            0.0
        };
        println!(
            "GOLDY_PERF: frames={} elapsed={elapsed:.2}s avg_fps={fps:.1}",
            self.frame_count
        );
    }
}

impl ApplicationHandler for App {
    fn resumed(&mut self, event_loop: &ActiveEventLoop) {
        if self.state.is_none() {
            let window = Arc::new(
                event_loop
                    .create_window(common::hidden_window(
                        format!("Goldy - Instancing ({} quads, Scheme + Present)", NUM_QUADS),
                        800,
                        800,
                    ))
                    .expect("Failed to create window"),
            );

            match RenderState::new(Some(window.clone())) {
                Ok(mut state) => {
                    if let Err(e) = state.render() {
                        tracing::error!("First frame error: {e}");
                    }
                    common::reveal_window(&window);
                    self.state = Some(state);
                    window.request_redraw();
                }
                Err(e) => {
                    tracing::error!("Failed to create render state: {}", e);
                    event_loop.exit();
                }
            }
        }
    }

    fn about_to_wait(&mut self, event_loop: &ActiveEventLoop) {
        if let Some(state) = &self.state {
            common::exit_if_timed_out(event_loop, state.start_time);
        }
    }

    fn window_event(&mut self, event_loop: &ActiveEventLoop, _id: WindowId, event: WindowEvent) {
        match event {
            WindowEvent::CloseRequested => event_loop.exit(),
            WindowEvent::KeyboardInput { event, .. } if event.state.is_pressed() => {
                if matches!(event.logical_key, Key::Named(NamedKey::Escape)) {
                    event_loop.exit();
                }
            }
            WindowEvent::Resized(size) => {
                if let Some(state) = &mut self.state {
                    if size.width > 0 && size.height > 0 {
                        let Some(surface) = state.surface.as_ref() else {
                            return;
                        };
                        let (prev_w, prev_h) = surface.size();
                        if size.width == prev_w && size.height == prev_h {
                            return;
                        }
                        let _ = surface.resize(size.width, size.height);
                        let format = surface.format();
                        if let Ok(pipeline) =
                            RenderState::create_render_pipeline(&state.device, &state.render_shader, format)
                        {
                            state.render_pipeline = pipeline;
                        }
                        state.rerecord_scheme();
                    }
                }
            }
            WindowEvent::RedrawRequested => {
                if let Some(state) = &mut self.state {
                    if let Err(e) = state.render() {
                        tracing::error!("Render error: {}", e);
                    }
                }
            }
            _ => {}
        }
    }
}

The example pulls in examples/common.rs, examples/instance2d.rs — see Shared Helpers.

Shaders

shaders/instancing_update.slang:

// Instancing compute shader - updates rotation and color for each quad instance

import goldy_exp;


float3 hsv_to_rgb(float h, float s, float v) {
    float4 K = float4(1.0, 2.0/3.0, 1.0/3.0, 3.0);
    float3 p = abs(frac(h.xxx + K.xyz) * 6.0 - K.www);
    return v * lerp(K.xxx, clamp(p - K.xxx, 0.0, 1.0), s);
}

[goldy_compute]
[numthreads(64, 1, 1)]
void cs_main(Scattered<Instance2D> INSTANCES, AnimParams params, ThreadId id) {
    uint idx = id.x;
    if (idx >= params.total_instances) return;
    
    Instance2D inst = INSTANCES[idx];
    
    float phase = (float(idx) / float(params.total_instances)) * 6.28318;
    inst.rotation = params.time * 2.0 + phase;
    
    float hue = frac(float(idx) / float(params.total_instances) + params.time * 0.1);
    inst.color = float4(hsv_to_rgb(hue, 0.8, 0.9), 1.0);
    
    INSTANCES[idx] = inst;
}

shaders/instancing_render.slang:

// Instancing render shader - renders quads from instance buffer

import goldy_exp;


struct VSOutput {
    float4 position : SV_Position;
    float4 color : COLOR;
};

[goldy_vertex]
VSOutput vs_main(Scattered<Instance2D> INSTANCES, VertexId vertex_id, InstanceId instance_id) {
    VSOutput output;
    
    Instance2D inst = INSTANCES[instance_id.value];
    float2 world_pos = quad_position_rotated(vertex_id.value, inst.position, inst.scale, inst.rotation);
    
    output.position = float4(world_pos, 0.0, 1.0);
    output.color = inst.color;
    
    return output;
}

[goldy_fragment]
float4 fs_main(VSOutput input) : SV_Target {
    return input.color;
}