A pure‑Rust heterogeneous GPU compute platform that JIT‑compiles idiomatic Rust functions to Vulkan 1.3 GPU pipelines. It lets developers write normal Rust compute kernels and execute them either natively on multi‑threaded CPU cores or compile them just‑in‑time onto GPU silicon, with a showcased real‑time 3D raymarching demo that can switch execution between CPU and GPU at runtime. The toolchain runs on stable Rust (1.80+), requires no nightly toolchains or foreign shading languages, and removes descriptor sets by leveraging Vulkan 1.3 Buffer Device Addresses. A short quick‑start demonstrates declaring a standard Rust function, allocating GpuVecs, dispatching with enki.flow, and verifying bit‑for‑bit equivalence between CPU and GPU outputs.
Architecturally it combines rustc’s compile‑time borrow checking with a runtime BorrowEngine that detects spatial and temporal data‑race hazards before queue submission, and it auto‑derives the minimal set of Vulkan Synchronization2 barriers via a Transitive Reduction Dependency Solver. Dual CPU/GPU verification supports CI without physical GPU hardware. The project is alpha (v0.1) with dynamic invariant checking and evolving APIs. System requirements include Vulkan 1.3 with 64‑bit buffer device addresses, timeline semaphores, Synchronization2 and shaderInt64 on Linux/Windows. Licensing is open‑core: core crates under MIT/Apache, while the precompiled Parsu GPU compiler backend is distributed under a custom Parsu license - free for developers and OSS but subject to future commercial licensing.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.