Rust compiler gets native, safe GPU offload, no vendor DSL required
The Rust team built zero-overhead, multi-vendor GPU offload straight into rustc and LLVM, using ownership and noalias to manage GPU memory safely without vendor-locked DSLs or raw pointers. A two-pass pipeline handles host/device ABI mismatches, and early RAJAPerf results match hand-tuned CUDA and HIP C++ kernels.