High-performance GPU programming has traditionally forced a compromise between execution efficiency and memory safety. While Rust guarantees compile-time memory safety for host CPUs via its strict ownership model, applying these constraints to massively parallel GPU execution environments has previously mandated either vendor-locked DSLs or unsafe raw pointers. This paper presents a zero-overhead, multi-vendor GPU compilation framework built natively into the Rust compiler (rustc) and LLVM backends.

We leverage Rust's type system, ownership model, and strict aliasing guarantees (noalias) to manage data transfers through LLVM's Offload infrastructure. Cross-vendor ABI lowering mismatches between Host and Device targets pose challenges, addressed via a two-pass compilation pipeline handling manual and compiler-generated memory movements. Evaluating on RAJAPerf, our rustc-based solution achieves competitive kernel performance against native CUDA and HIP C++ baselines.

Submission history From: Manuel Sebastian Drehwald [view email] [v1]Thu, 13 Aug 2026 20:37:48 UTC (88 KB).