Skip to content

feat(gpu): WebGPU backend for the WASM target inside the ordinary GPU dispatch (with explicit CPU fallback and refusal) - #562

Open
tsotchke wants to merge 4 commits into
masterfrom
feat/webgpu-dispatch
Open

feat(gpu): WebGPU backend for the WASM target inside the ordinary GPU dispatch (with explicit CPU fallback and refusal)#562
tsotchke wants to merge 4 commits into
masterfrom
feat/webgpu-dispatch

Conversation

@tsotchke

Copy link
Copy Markdown
Owner

Adds an ESHKOL_GPU_WEBGPU dispatch seam selected by the same runtime dispatch that picks Metal and CUDA, with Emscripten backend selection, a WebGPU C backend, WGSL kernels for the ops the GPU correctness gate covers, and explicit CPU fallback or refusal where a kernel cannot meet the gate tolerance. JS glue stubs for every new runtime function; docs and feature matrix updated to the verified status.

… dispatch (with explicit CPU fallback and refusal)

Adds an ESHKOL_GPU_WEBGPU dispatch seam selected by the same runtime dispatch that picks Metal and CUDA, with Emscripten backend selection, a WebGPU C backend, WGSL kernels for the ops the GPU correctness gate covers, and explicit CPU fallback or refusal where a kernel cannot meet the gate tolerance. JS glue stubs for every new runtime function; docs and feature matrix updated to the verified status.
…xecution marker; high/fast tiers fail closed unless certified; JSPI-promising wrappers on WASM exports; batch matmul imports in both loaders; elementwise workgroup limits and buffer cleanup
…uards; precision tiers fail closed with explicit fast opt-in; promising wasm callbacks
…g, validation error scopes, live Chrome coverage
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant