tensor-wasm-exec 0.3.8

Wasmtime + Tokio async execution engine for TensorWasm instances.
Documentation

tensor-wasm-exec

Async execution engine for Craton TensorWasm, built on Wasmtime and Tokio. Provides the TensorWasmEngine wrapper around wasmtime::Engine with a TensorWasm-specific Config, the per-instance TensorWasmInstance state machine, and the TensorWasmExecutor responsible for spawning, invoking, and terminating Wasm instances on demand. This is the runtime hot path that ties together memory, GPU bridges, and the JIT pipeline.

Feature flags

Flag Default Description
cuda no Enable the real GPU kernel-launch path in jit_dispatch (a cache hit launches a compiled kernel instead of signalling deopt). Pulls the CUDA host bridge in tensor-wasm-wasi-gpu (tensor-wasm-wasi-gpu/cuda); requires a CUDA toolkit at build time, so it is opt-in and off by default.

See docs/BUILD.md for the project-wide flag taxonomy.

Dependencies

External crates this crate depends on (pinned at workspace root):

  • tokio — async runtime hosting the executor's task pool.
  • async-trait — async methods on the public executor traits.
  • futures — combinator primitives used in the dispatch layer.
  • wasmtime — embedded Wasm engine driving instance execution.
  • wasmparser — pre-instantiation operator walk for the auto-offload analysis.
  • thiserror — derive macro for executor-level errors.
  • tracing — structured spans/events for instance lifecycle.
  • parking_lot — fast synchronisation primitives for shared state.
  • dashmap — concurrent map keyed by InstanceId (also backs the module cache).

Internal TensorWasm crate dependencies:

  • tensor-wasm-core — shared TensorWasmError, TensorWasmMetrics, and identifier types (TenantId, InstanceId).
  • tensor-wasm-jit — kernel cache + detector consumed by auto_offload and jit_dispatch.
  • tensor-wasm-mem — TensorWasmMemoryCreator so linear memory is carved from CUDA Unified Memory on supported hosts.