Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore advanced programming systems for modern accelerated computing through this computer science colloquium talk by Stanford PhD student Rohan Yadav. Discover how to address the growing challenges of developing software for increasingly complex, specialized, and rapidly evolving hardware systems. Learn about the critical need for new levels of abstraction, portability, and composability in programming systems to keep pace with hardware advances. Examine the connection between actor-based and task-based programming models, two popular approaches for distributed and accelerated machines, and understand how these models function as duals of each other. See how this duality can be leveraged to compile task-based programs into efficient actor-based programs, closing the performance gap between high-level abstractions and optimal performance. Investigate Twill, an innovative system that automatically discovers optimal software pipelining and warp specialization strategies for Tensor Core GPUs without relying on expert intuition or compiler heuristics. Understand how these optimization strategies can be derived from first principles in a machine-parametrizable manner, and learn how this approach successfully rediscovers expert-level strategies across different GPU generations.
