Compute¶
The compute subsystem provides WebGPU compute workflows, ndarray abstractions, pipeline helpers, async readback utilities, and built-in kernels for copy, reductions, arg-reductions, scan, histogram, compact, radix sort, scaling, and batched LU factorization and solve.
In This Section¶
- Compute pipeline creation and dispatch
- CPU/GPU ndarray lifecycle and residency
- Storage/uniform buffer management
- Built-in compute kernels, including typed vector arithmetic, dot products, GEMM, stable key/value radix sort, and batched LU factorization and solve
- Async readback and result extraction
Suggested Starting Points¶
- WasmGPU.compute.createPipeline
- WasmGPU.compute.dispatch
- WasmGPU.compute.createStorageBuffer
- WasmGPU.compute.kernels.luFactorF32Batched and WasmGPU.compute.kernels.luSolveF32Batched
- WasmGPU.compute.kernels.addF32, addU32, and addC64
- WasmGPU.compute.kernels.subF32, subU32, and subC64
- WasmGPU.compute.kernels.mulF32, mulU32, and mulC64
- WasmGPU.compute.kernels.sclF32, sclU32, and sclC64
- WasmGPU.compute.kernels.axpyF32, axpyU32, and axpyC64
- WasmGPU.compute.kernels.dotF32, dotU32, and dotC64
- WasmGPU.compute.kernels.gemmF32, gemmU32, and gemmC64
- WasmGPU.compute.kernels.luFactorC64Batched and WasmGPU.compute.kernels.luSolveC64Batched
- WasmGPU.compute.readback.readF32
- WasmGPU.compute.CPUndarray.empty
- WasmGPU.compute.CPUndarray.destroy
- WasmGPU.compute.GPUndarray.wrap
Use the sidebar to navigate the full compute API surface.