First pixel
A bufferless triangle through the portable render path.
A bufferless triangle through the portable render path.
All four copy directions execute; the texture-to-texture result is then bound and sampled.
Per-dispatch constants animate a compute-generated vertex stream.
Indexed depth, occlusion resolve, barriers, and a query-tinted depth preview.
Instance-rate vertices and reflected dynamic uniform offsets.
cglm transforms, texture sampling, indexed draw, and depth.
Compute writes typed RGBA8 pixels; render samples them directly.
Per-draw constants with native lowering and WebGPU dynamic slices.
Four-sample rasterization resolved directly into the swapchain.
Two color targets use independent alpha and additive blending.
Depth comparison sampling and filtered shadow visibility.
Compute reads, queries, and writes logical 1D-array layers before render samples them.
Compute transforms a logical 1D texture before render reads both lines.
Cubemap faces and a 3D volume share one reflected bind group.
Dynamic texture, sampler, and buffer indices flatten into bounded WebGPU bindings.
Hardware resolve and four explicit sample reads share one 4x render target.
A native subgroup shuffle colors a compute-generated vertex stream.
Native f16 arithmetic, with a bounded f32 fallback.
One API call consumes two packed indirect draw records.
One packed dispatch record launches the compute-to-render path.
Pass timestamps validate portable 8-byte resolve alignment.
Compute updates a persistent particle field consumed directly by render.
One uploaded mip chain is sampled at five explicit levels.
A fill pass writes stencil, then a larger draw keeps only the visible outline.
Horizontal and vertical compute passes feed one sampled composite.
Browser-decoded sRGB pixels are uploaded once and sampled through reflected bindings.
Typed cube sampling lowers to a bounded WebGPU array-view fallback.
Left decodes sRGB; right treats the same bytes as linear before HDR tone mapping.
Capability selection prefers native compression and keeps RGBA8 as the portable fallback.
Four bone matrices animate an indexed mesh through a reflected dynamic uniform ring.
Albedo, tangent normals, metallic-roughness, and cubemap lighting share one reflected layout.
Runtime Khronos glTF download, AssetKit material resolution, and real indexed PBR rendering.
Magnified 32x32 source at left; nearest and linear 256x256 blits at center and right.