Practical Efficiency of Optimizing Compilers in Parallel Scientific Applications

A. V. Bogdanov, I. G. Gankevich

Optimizing compilers are essential for building any scientific application, however they are notgeneral purpose tools. Although, many compilers offer similar functionality, different optimizationstrategies as well as code structure can lead to different performance results [1]. Additionally,modern scientific applications often solve large-scale problems concurrently on a set of processorsthus demanding not only serial but parallel code optimizations. So, choosing the right compiler fora particular problem is a topical question of today.

In the paper a variety of commercial and open source optimizing compilers are compared interms of their functionality. Then their relative performance is measured benchmarking differentsets of algorithms and scientific applications classified by their problem domains. The final resultsare presented as a cumulative compiler rating scored in a particular problem domain. Based on thisrating conclusions are made.

  title={Practical Efficiency of Optimizing Compilers in Parallel Scientific Applications},
  author={A. V. Bogdanov and I. G. Gankevich},
  howpublished={Book of abstracts of GRID'12},
  address={Dubna, Russia},

Publication: Book of abstracts of GRID'12