In this Invited Talk from SC17, Michael Wolfe from NVIDIA presents: Why Iteration Space Tiling? The talk is based on his noted paper, which won the SC17 Test of Time Award.
“Tiling subdivides the multidimensional iteration space of a nested loop into blocks or tiles. Tiles become a natural candidate as the unit of work for parallel task scheduling, where scheduling and synchronization is done at tile granularity, reducing overhead at some loss of potential parallelism. Tiling allows separation of concerns between local optimizations (vectorization, cache footprint) and global optimization (parallelization, scheduling for locality). Tiling is well-known and has been included in many compilers and code transformation systems. The talk will explore the basic contribution of the SC1989 paper to the current state of iteration space tiling.”
Michael Wolfe has developed compilers and languages for parallel and high performance computing for over 40 years in both academia and industry. He is now a PGI compiler engineer at NVIDIA, and is the chair of the technical language committee for OpenACC.
Rich: Thanks for posting this. Note that on the insidehpc.com landing page, the intro to this link says: In this Invited Talk from SC17, Michael Wolfe from NVIDIA presents: Scalable Parallel Programming Using OpenACC for Multicore, GPUs, and Manycore. “You will learn how modern supercomputers are organized and how highly parallel nodes will affect the design of your applications. Whether you use multicore, manycore, or GPU-accelerated nodes, the same basic concepts apply to your algorithm and data structure design: parallelism management and data management.” That was the description of the tutorial that I gave on Monday, not this talk. Also, when they presented this video, they unfortunately did not (even once) show the one slide that I used, and I refer to that slide a couple of time during the talk. I have a PNG file of that one slide, if you want to include it on this page, or it’s available on twitter at https://twitter.com/mwcompilerguy/status/941056656313548800
Thanks, Michael. Got it fixed.