We developed an automated environment to support the analysis of memory access behaviors of applications on high performance clusters. Code optimization targeting efficient use of processor caches is crucial for achieving good performance on such systems. Our environment is able to selectively instrument OpenMP Fortran95 programs upon requests of programmer. The monitor can be configured to collect hardware counter information on specified code regions. Limitations due to the number of available physical hardware counters are automatically taken into account. The whole environment is controlled through a friendly user interface based on Eclipse and is highly portable.
展开▼