Describir: An Efficiency-Based Approach to Performance for Computational Solid Mechanics on Emerging Architectures