Performance Evaluation of Core Numerical Algorithms: A Tool to Measure Instruction Level Parallelism

Bernard Goossens (1) , Philippe Langlois (2, 1) , David Parello (1) , Eric Petit (1)

1 ELIAUS - Electronique, Informatique, Automatique et Systèmes
2 DALI - Digits, Architectures et Logiciels Informatiques

We measure and analyze the instruction level parallelism that condition the running-time performance of core numerical subroutines. We propose PerPI, a programmer oriented tool to fill the gap between high level algorithm analysis and machine dependent profiling tools and which provides reproducible results.

Domaines: Logiciel mathématique [cs.MS]

Bernard Goossens, Philippe Langlois, David Parello, Eric Petit. Performance Evaluation of Core Numerical Algorithms: A Tool to Measure Instruction Level Parallelism. Para: State of the Art in Scientific and Parallel Computing, Jun 2010, Reykjavík, Iceland. ⟨hal-00477541v2⟩