oxicuda-rocm

Part of the OxiCUDA ecosystem — Pure Rust CUDA replacement for the COOLJAPAN ecosystem.

Overview

oxicuda-rocm is the AMD ROCm/HIP backend for OxiCUDA. It implements the ComputeBackend trait using the AMD HIP runtime (libamdhip64.so), loaded dynamically at runtime via libloading — no compile-time SDK dependency is required. The crate provides device enumeration, memory management, and kernel dispatch for AMD GPUs on Linux.

Features

• Dynamic HIP loading — libamdhip64.so is loaded at runtime; no HIP SDK needed at compile time
• hipRTC integration — runtime compilation of HIP kernel source strings via libhiprtc.so; no ROCm toolchain required at compile time
• hipBLAS interop — optional high-performance BLAS via libhipblas.so with automatic kernel fallback when not installed
• Device management — Enumerate and select AMD GPU devices
• Memory operations — Allocate, free, and transfer device memory
• Multi-GPU support — MultiDeviceDispatcher distributes matrix work across all available AMD GPUs
• Backend abstraction — Implements oxicuda-backend's ComputeBackend trait for unified multi-GPU-API programming
• Pure Rust — Zero C/Fortran in the default feature set

Platform Support

Usage

Add to your Cargo.toml:

[dependencies]
oxicuda-rocm = "0.1.4"

use oxicuda_rocm::RocmBackend;

let backend = RocmBackend::new();
match backend.init() {
    Ok(()) => println!("ROCm backend initialised"),
    Err(e) => println!("ROCm not available: {e}"),
}

Status

• Version: 0.1.4 (2026-04-18)
• Tests: 105 passing

License

Apache-2.0 — © 2026 COOLJAPAN OU (Team KitaSan)