EvoAVX: Island-Based Genetic Algorithm Library With AVX-512 and MPI

Genetic algorithms are practical tools for solving complex optimization problems; however, the performance of many existing solutions is often limited by their inability to utilize modern CPU capabilities fully. This paper introduces EvoAVX, an open-source island-based genetic algorithm library engineered for minimal computational overhead. EvoAVX addresses the lack of specialized, high-performance libraries by leveraging AVX-512 intrinsics to optimize genetic operators, statistics calculations, and the pseudorandom number generator. The library provides a built-in, selectable set of vectorized operators. Static polymorphism is implemented via C++ templates to ensure fast, compile-time operator resolution. EvoAVX employs a double-precision real-valued gene representation. The Island model is built on the Message Passing Interface, which enables distributed computing. Our performance evaluation shows notable improvements in execution time compared to popular libraries for evolutionary computation. The performance advantages of EvoAVX are particularly pronounced in scenarios where the fitness evaluation is computationally inexpensive, making it an ideal tool for high-throughput optimization tasks.