Distributed Sparse Computing with Legate Sparse
Legate Sparse is a system that transparently distributes and accelerates unmodified sparse matrix-based SciPy programs across clusters of CPUs and GPUs, and composes with cuPyNumeric, a distributed NumPy library. Programs written in Legate Sparse and cuPyNumeric perform competitively with low-level systems like PETSc. In this talk, I will discuss the implementation of Legate Sparse, which involved leveraging the DISTAL compiler to generate distributed sparse matrix kernel implementations. DISTAL separates descriptions of tensor algebra expressions, sparse data structures, data distribution and computation distribution, enabling the distributed execution of sparse tensor algebra expressions with a variety of sparse data structures and data distributions. I will then discuss the broader impact of the Legate system.
Mon 16 JunDisplayed time zone: Seoul change
10:30 - 12:00 | |||
10:30 20mTalk | Optimizations and abstractions for sparse machine learningRecorded Sparse Charith Mendis University of Illinois at Urbana-Champaign | ||
10:50 20mTalk | Distributed Sparse Computing with Legate Sparse Sparse Rohan Yadav Stanford University | ||
11:10 20mTalk | Optimizing Recursive Sparse Computations Sparse Amir Shaikhha University of Edinburgh | ||
11:30 20mTalk | Panel 2 Sparse Charith Mendis University of Illinois at Urbana-Champaign, Rohan Yadav Stanford University, Amir Shaikhha University of Edinburgh | ||
