Optimization strategies for parallel CPU and GPU implementations of a meshfree particle method