SX-NBody: SIMD Vectorization and Atomic Work Dispatch for N-Body Gravitational Simulation on Modern Hardware

Abstract The SX-NBody framework extends the atomic work-dispatch technique of the SX series to the classical all-pairs gravitational N-body problem. The approach combines HARDWARE vectorization with a lock-free atomic work queue, enabling efficient utilization of both data-level and task-level parallelism on modern multicore ARM systems. The inner force-evaluation kernel is inspired by the fast inverse square-root method popularized in real-time graphics, here realized directly in hardware via vrsqrteq_f32, followed by Newton–Raphson refinement to recover numerical accuracy suitable for physical simulation. Experimental Evaluation The framework was evaluated on modern consumer hardware featuring an ARM big.LITTLE architecture. A simulation with N = 20,000 bodies was executed using the exact formulation. Key results: 12.20× empirical speedup over a scalar, single-thread baseline Sustained throughput: 68.2 GFLOPS Parallel efficiency: 95.1%, demonstrating near-perfect orthogonality between SIMD vectorization and multicore atomic dispatch These results demonstrate that careful algorithmic composition—explicit SIMD parallelism combined with lock-free multicore dispatch—can achieve high efficiency even on consumer-grade mobile systems.