作者: John Mellor-Crummey , Robert J. Fowler , Gabriel Marin , Nathan Tallent
关键词:
摘要: It is increasingly difficult for complex scientific programs to attain a significant fraction of peak performance on systems that are based microprocessors with substantial instruction-level parallelism and deep memory hierarchies. Despite this trend, analysis tuning tools still not used regularly by algorithm application designers. To large extent, existing fail meet many user needs cumbersome use. address these issues, we developed HPCVIEW—a toolkit combining multiple sets program profile data, correlating the data source code, generating database can be analyzed anywhere commodity Web browser. We argue HPCVIEW addresses issues have limited usability utility most tools. originally built facilitate our own work layout optimizing compilers. Now, in addition daily use within group, being several code development teams DoD DoE laboratories as well at NCSA.