SC19 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

D2P: From Recursive Formulations to Distributed-Memory Codes

Authors: Nikhil Hegde (Indian Institute of Technology Dharwad, Purdue University), Qifan Chang (Purdue University), Milind Kulkarni (Purdue University)

Abstract: Recursive formulations of programs are straightforward to reason about and write, often have good locality properties, and readily expose parallelism. We observe that it is easier to automatically generate distributed-memory codes for recursive formulations with certain properties: i) inclusive—a recursive method’s parameters summarize the data access done within the method body. ii) Intersection—data-set intersection tests among method invocations can be computed efficiently.

In this paper we present D2P, a system that automatically generates distributed-memory codes for recursive divide-conquer algorithms with these properties. D2P produces MPI-based implementations starting from shared-memory specifications of the recursive algorithms. We evaluate D2P with recursive Dynamic Programming (DP) algorithms, since these algorithms have the desired properties and are well known. We show that the generated implementations are scalable and efficient: D2P generated implementations execute faster than implementations generated by recent distributed DP frameworks, and are competitive with (and often faster than) hand-written implementations.

Presentation: file

Back to Technical Papers Archive Listing