axiolid-backend-gpu
GPU executor adapter contract for batched Axiolid geometry.
| Latest release | 0.3.1 (2026-09-28) |
| crates.io | axiolid-backend-gpu |
| Facade | axiolid feature gpu |
| Layer | execution (execution.context) |
| API documentation | rustdoc · docs.rs |
| Source | crates/execution/gpu/ |
Overview
An API-neutral seam for GPU graph compilation. GpuCompiler adapts any GpuGraphExecutor to the MeshCompiler contract: it validates device, precision, residency and roots before submitting one batch, and checks the executor's results afterwards. The crate chooses no GPU API (CUDA, Metal, Vulkan, WebGPU) and ships no executor, so default builds carry no driver stack. Concrete executors live in their own crates, including out of tree (ADR 0011).
Design notes
- Each further GPU operation gets its own narrow executor trait and adapter, never a method on one catch-all backend.
- A GPU path is evidence only if a CPU differential test checks it. Maturing the executor is tracked in #22.
Depends on
Changes
Latest release, 0.3.1 (2026-09-28):
Changed
- The crates.io page is this crate's own
README.md, with links to its API documentation, its reference page and the source (ADR 0078).
Full history: crates/execution/gpu/CHANGELOG.md