Automatic data distribution for improving data locality on the cell BE architecture

  • Authors:
  • Miao Wang;François Bodin;Sébastien Matz

  • Affiliations:
  • National Laboratory for Parallel and Distributed Processing, NUDT, China;,CAPS Entreprise, Rennes, France;CAPS Entreprise, Rennes, France

  • Venue:
  • LCPC'09 Proceedings of the 22nd international conference on Languages and Compilers for Parallel Computing
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Multicore systems provide potential to improve the performance of the applications. However, substantial programming effort is required to exploit the power of the parallelism. This paper presents a single source compiler to map the data-parallel programs onto Cell Broadband Engine. Based on the distributed memory model, the compiler performs automatic data distribution and generates SPMD programs with message-passing primitives for Cell. We evaluate our compiler using a range of computation intensive benchmarks, high performance is achieved on Cell platform. In contrast to OpenMP, our method can fully exploit data locality through managing the shared data using inter-processor communication instead of accessing main memory, which significantly reduces the off-chip memory access overhead.