Legion: Expressing locality and independence with logical regions

Abstract

Abstract-Modern parallel architectures have both heterogeneous processors and deep, complex memory hierarchies. We present Legion, a programming model and runtime system for achieving high performance on these machines. Legion is organized around logical regions, which express both locality and independence of program data, and tasks, functions that perform computations on regions. We describe a runtime system that dynamically extracts parallelism from Legion programs, using a distributed, parallel scheduling algorithm that identifies both independent tasks and nested parallelism. Legion also enables explicit, programmer controlled movement of data through the memory hierarchy and placement of tasks based on locality information via a novel mapping interface. We evaluate our Legion implementation on three applications: fluid-flow on a regular grid, a three-level AMR code solving a heat diffusion equation, and a circuit simulation.

Deterministic scale-free pipeline parallelism with hyperqueues

“…Several research projects define deterministic and scalefree parallel programming languages [2,3,4,5]. In these, most attention has gone to DOALL parallelism and fork-join parallelism.…”

Block-parallel data analysis with DIY2

“…Charm++ [19] has a dynamic mapping between data objects and processes (called chares) and between chares and physical processes. In Legion [3], blocks are called logical regions, and the runtime maps logical regions to physical resources. Regent [39] is a new language and compiler for the Legion runtime, which results in shorter, more readable code than the original Legion language.…”

“…Regent [39] is a new language and compiler for the Legion runtime, which results in shorter, more readable code than the original Legion language. 3 …”

TProf: An energy profiler for task-parallel programs

“…The difficulty in writing multithreaded parallel programs has lead to the development of higher-level abstractions for parallel programming, such as task-parallel languages [6,21,13,3], domainspecific parallel languages, pattern libraries that hide lowlevel threads from the programmer [33] and task-parallel programming models [7,31,2,42,42]. Compared to the traditional thread programming, task-parallel programs are easier to write, more portable, and scale better, because the parallelism is not hard-wired into the program, but created at runtime, as needed.…”