Parallelization spectroscopy: analysis of thread-level parallelism in hpc programs

In this paper, we present a thorough analysis of thread-level parallelism available in production High Performance Computing (HPC) codes. We survey a number of techniques that are commonly used for parallelization and classify all the loops in the applications studied using a sensitivity metric: how likely is a particular technique is successful in parallelizing the loop. We call this method parallelization spectroscopy. Using parallelization spectroscopy, we show that in most of the benchmarks, at the loop level, more than >75% percent of the runtime is inherently parallel.

• Type: Conference
• Year: 2009
• Where: PPOPP
• Authors: Arun Kejariwal, Calin Cascaval
• Keywords: Loop Level, Method Parallelization Spectroscopy, Parallel Computing, PPOPP 2009