In this talk, I'll describe and motivate the high-points in the design of the Manticore language, as well as outline some challenges in its implementation. I'll present a flexible runtime model that supports multiple scheduling disciplines (e.g., for both fine-grain and course-grain parallelism) in a uniform framework. I'll describe our Lazy Tree Splitting (LTS) strategy for parallel operations over tree-like data structures; the key advantage of LTS is that it provides /performance robustness/ (it does not require tuning to acheive near optimal performance for each distinct program).
Joint work with: Sven Auhagen, Lars Bergstrom, Nic Ford, Mike Rainey, John Reppy, Adam Shaw, and Yingqi Xiao.