@danielsimonjr/mathts-gpu
The shared WebGPU foundation for MathTS: GPUContext (device/adapter
lifecycle), a shared getGpuDevice() singleton, BufferPool, detect
(capability probing), and a generic ShaderManager (compile/cache/pipeline
infra + a name→code shader registration API). This package ships no domain
kernels — matrix-specific WGSL (matmul/transpose/reduce) lives in
@danielsimonjr/mathts-matrix, which registers those shaders onto a
ShaderManager at backend init.
WebGPU is browser-only (navigator.gpu); in Node every entry point degrades
gracefully — getGpuDevice() resolves to null and GPUContext.initialize()
resolves false without throwing.
Tracked by CDG/DGT (create-dependency-graph) as a leaf: matrix → gpu,
functions → gpu (functions edge lands in a later spec), with no back-edge.