In the twilight of Moore’s Law, computer hardware has diverged from the programming models we use to control it. As a result, the performance and complexity gaps between “reasonable” programs and the fastest code are huge and growing. This talk will explore fundamental issues that drive hardware efficiency. From that foundation, it will highlight challenges and opportunities for the programming systems community in productively unlocking the potential of current and future hardware. Major themes include: leveraging domain-specific languages; balancing safety, automation, and control for high-performance programming; and broadening our understanding of “correctness” in a world of numerical programs and machine learning.
