Portable wgpu paints

xgpui-shader/wgpu · paint-lab headless path · native window renderers unchanged.

Use

# App under apps/: {#app-under-apps}
[dependencies]
xgpui-shader = { path = "../../extensions/crates/xgpui-shader", features = ["wgpu"] }

use gpui::{prelude::*, *};
use std::sync::Arc;
use xgpui_shader::{painted, Gradient, GradientStop, WindowPaintExt};

let paint = Arc::new(Gradient::linear(45.0, vec![
    GradientStop::new(0.0, [1.0, 0.2, 0.3, 1.0]),
    GradientStop::new(1.0, [0.1, 0.3, 1.0, 0.5]),
])?);
painted().w(px(320.0)).h(px(240.0)).bg_paint(paint.clone());
window.paint_path_with_paint(path, paint.clone());

Producer API

gpui_wgpu::custom_draw — signature summary:

pub struct WgpuCustomContext {
    pub device: Arc<wgpu::Device>,
    // Private recording state.
}
impl WgpuCustomContext {
    pub fn cache_key(&self) -> usize;
    pub fn create_output(&self, extent: Size<DevicePixels>) -> Result<WgpuCustomOutput>;
    pub fn record<T>(
        &self,
        encode: impl FnOnce(&mut wgpu::CommandEncoder) -> Result<T>,
    ) -> Result<T>;
    pub fn can_reuse(&self, output: &WgpuCustomOutput) -> bool;
}
pub struct WgpuCustomOutput { /* private fields; Clone */ }
impl WgpuCustomOutput {
    pub fn texture(&self) -> &wgpu::Texture;
}

Contract

producer    reusable cache hit OR allocate → record → return output
host        validate → stage → submit accepted batch → compose
failure     discard unsubmitted batch → frame error
reuse       same recording / accepted batch / submitted queue

output      RGBA8Unorm · straight alpha · encoded-sRGB SDR
mapping     full unclipped bounds; logical size ≠ raster extent
coverage    analytic rounded rect; native border/path mask red channel
cache key   queue + extent (+ logical size for WGSL)
cache       64 MiB live output; separate 64 MiB coverage
mesh        Hermite vertex shader; no compute/storage buffers

No producer submission/GPU wait. Cache budgets exclude in-flight retention. Recording details.

Checks

just --justfile apps/xgpui-ext-lab/paint-lab/justfile wgpu-test
just --justfile apps/xgpui-ext-lab/paint-lab/justfile wgpu-headless --effect mesh --frames 60 --png /tmp/wgpu-mesh.png

# One headless backend/device per process; unavailable backend → error. {#one-headless-backenddevice-per-process-unavailable-backend--error}
WGPU_BACKEND=metal just --justfile apps/xgpui-ext-lab/paint-lab/justfile wgpu-test
WGPU_BACKEND=vulkan just --justfile apps/xgpui-ext-lab/paint-lab/justfile wgpu-test
# Windows PowerShell: $env:WGPU_BACKEND = "dx12" {#windows-powershell-envwgpubackend--dx12}

cargo test --config .cargo/validation.toml \
  --manifest-path gpui/crates/gpui-wgpu/Cargo.toml \
  --features test-support --lib wgpu_custom --target-dir target

Executed: macOS wgpu/Metal pixels, geometry, mesh parity, replay/cache and transaction tests. SPIR-V/HLSL generation passes; Vulkan/DX12 runtime acceptance remains outstanding. Software/readback timing is not hardware GPU timing.

Native D3D11 · WGSL

Backend workflows, focused diagnostics, caching, and cold/warm measurement are documented in Paint GPU validation.