Legion: Expressing locality and independence with logical regions
By Michael A. Bauer, Sean Treichler, Elliott Slaughter, A. Aiken. International Conference for High Performance Computing, Networking, Storage and Analysis, 2012
683 citations recorded by the source corpus.
Subfields: Computer Networks and Communications, Hardware and Architecture
View Paper: https://www.semanticscholar.org/paper/2014288f6190cdc58da5a065ca1a3e7bd5d277ff
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Networks and Communications
- Matei A. Zaharia
- Ferdinand Kossmann
- Tal Ben-Nun
- Charles Ruizhongtai Qi
- A. Aiken
- Christopher J. Rossbach
- Torsten Hoefler
- Dennis Fetterly
In Hardware and Architecture
- Ferdinand Kossmann
- Matei A. Zaharia
- A. Aiken
- Torsten Hoefler
- Tal Ben-Nun
- Andrei Ivanov
- Nikoli Dryden
- Shigang Li
Citations in the corpus
- Dandelion: a compiler and runtime for heterogeneous systems (2013, background): “Liszt’s [38] static meshes, Halide’s [82] coordinate spaces, and Legion’s [16] logical regions provide runtimes and compilers with similar leverage for identifying automatically parallelizable code: all enable highlevel programs to be parallelized and compiled to different target architectures.”
- Data Movement Is All You Need: A Case Study on Optimizing Transformers (2020, methodology): “Other white-box approaches for separation of program definition from data optimization passes include Halide (Ragan-Kelley et al., 2013), JAX (Frostig et al., 2018; Bradbury et al., 2018), Legion (Bauer et al., 2012), Lift (Steuwer et al., 2017), and MLIR (Lattner et al., 2020).”
- Optimizing Mixture of Experts using Dynamic Recompilations (2022, methodology): “OF FLEXFLOW FlexFlow (Jia et al., 2018) is a distributed DNN framework built on top of the Legion programming system (Bauer et al., 2012).”
- Beyond Data and Model Parallelism for Deep Neural Networks (2018, methodology): “The following software versions were used in our experiments: cuDNN 7.3, CUDA 9.0, Legion 18.02.0, and GASNet 1.28.0.”
- Exploring Hidden Dimensions in Parallelizing Convolutional Neural Networks (2018, methodology): “Third, Legion allows fine-grain control over the placement of tasks and data.”
- UNITY: Unified Memory and File Space (2017, background): “Legion [6, 34] is a data-centric programming model and runtime for executing applications on distributed parallel systems.”