DyPAT: Dynamic Performance Analysis and Tuning

The objective of this proposal is to analyze problems with the OpenMP component of scientific applications, specifically ASCI benchmarks, to investigate more efficient OpenMP alternatives and to subsequently tune applications. The novelty of this work lies in its approach of combining online analysis and tuning. The methodology to perform both analysis and tuning relies on binary rewriting. We expect this work to provide significant contributions toward better utilizing existing SMP clusters in terms of 1) their memory bandwidth, 2) reduced coherence traffic and 3) reduced OpenMP overhead.