This web-based shader compiler is designed for rapid shader-authoring workflows: write Python in the left pane, run it in the browser, and inspect generated output on the right. It is especially useful for researchers who want to iterate on kernel structure and inspect code generation without switching to a full native setup.
The key feature is shader visibility: when you print a decorated shader function (for
example, print(add_scalar)), the panel shows the generated GLSL. This makes
it easy to validate indexing logic, control flow, and type usage directly from the
high-level Python definition.
The Options panel controls a dummy device model, not your physical GPU. You can adjust limits such as subgroup size, workgroup limits, and shared memory to test how your shader configuration behaves under different device constraints.
Most standard vkdispatch APIs are available in this environment (buffers,
images, descriptor bindings, and dispatch calls), but this page is intended for codegen and
interface exploration. Many operations are simulated, and dispatch execution is not a real
GPU compute run.