WGSL Support

WGSL shaders (.wgsl) run on WebGPU in the VS Code extension and standalone browser. A WebGPU-capable host is required. Start with WGSL Shaders for mainImage, uniforms, vertex hooks, compute entry points, and storage types.

WGSL supports single-pass and multipass rendering, Common helpers, vertex and compute passes, storage buffers, textures, cubemaps, audio/video, and keyboard input. Model geometry and script uniforms are also supported. See Channels for sampling.

Editor Support and Diagnostics

Completion, hover, signature help, definitions, references, highlights, and symbol rename are available. See Language Servers for instructions and the scope of rename in each editor. WGSL has no user-defined generics, function overloads, or import/include mechanism; put shared code in Common.

As you type, the editor reports syntax errors, undefined names, reserved words, and uses of built-ins or discard in an incompatible shader stage. It reports the first syntax error; fix that error to see name and stage diagnostics. To check expression types, function arguments, and return values, compile your shader. Errors point to the relevant shader, Common, or vertex file where possible.

When editing Common, syntax errors appear as you type. Compile a pass that uses Common to check its helpers fully. If a helper uses an operation unavailable in that pass, such as dpdx in compute code, the error identifies the helper and line.

Signature help shows named parameters and documentation for authored, Common, channel, and built-in functions. Add leading // comments to document your own functions. Signature help also works inside nested calls and template arguments.

Vector component completion includes all valid one-to-four-component read selections, including reordered and repeated forms such as yx, xxxx, and bgra.

Debugging and Capture Types

Enable debug mode, place the cursor on a value, and use inline rendering or the Variable Inspector. Common helpers, function parameter overrides, and loop iteration caps are supported.

Scalar and vector locals, parameters, and return values can be captured, including values read from arrays, struct fields, and configured storage. Types are inferred for many unannotated let and var declarations, including expressions using iTime, sin, select, pointer dereferences, bitcast, integer bit operations, transpose, vector comparisons, and scalar-matrix multiplication. Direct built-in expressions such as normalize(uv).yx receive the same member completion as a local. If a local initialized from an unsupported expression is missing from capture, add an explicit type. Whole arrays, whole structs, and arbitrary pointer values cannot be displayed as capture rows; select an element or field instead.

Both mat2x2f and mat2x2<f32> can be captured, including inferred locals, return values, and storage elements. The four values appear in column-major order:

let basis = mat2x2f(0.125, 0.25, 0.5, 0.75);
// Capture: 0.125, 0.25, 0.5, 0.75
//          column 0     column 1

Larger matrices and f16 matrices are not capturable as a whole. Select a supported column or scalar component instead. Boolean vectors are also inferred for vector comparisons but are not capturable as a whole; select a scalar component. Matrix capture in the Variable Inspector is separate from the Storage inspector, which edits scalar and vector buffer elements.

Compute Debugging Limits

Compute debugging previews one invocation at a time; it cannot reproduce threads working together in a workgroup. global_invocation_id comes from the canvas coordinate, with z = 0 and iDispatch = 0. Direct global_invocation_id, local_invocation_id, workgroup_id, and local_invocation_index parameters are supported with literal workgroup dimensions. Other entry parameter forms report an unsupported diagnostic.

Read-only storage access is supported. Workgroup memory, barriers, atomics, subgroup operations, and writes to configured storage are not supported during compute debugging. Use normal compute rendering for code that needs cooperative workgroups or storage writes. Vertex-stage debugging is unavailable.

See Language Support to compare GLSL, Slang, and WGSL.