Anmerkungen:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Beschreibung:
The concept of a shared address space simplifies the parallelization of programs by using shared data structures instead of explicit communication. Based on this basic principle directive-based, proprietary programming models exist from different vendors. The OpenMP Application Program Interface instead offers a portable programming model which permits successive parallelization of sequential programs. When intended as a base platform for developing parallel applications, highly efficient implementations of OpenMP are necessary on different hardware architectures. First, we analyze OpenMP-implementations application-independent. For this purpose the overhead caused by individual OpenMP primitives is analyzed. Then usability is examined for three applications. OpenMP implementations on different parallel computer architectures are analyzed: vector-parallel computers (Cray J90, Cray T90), cc-NUMA computers (SGI Origin2000), and SMP-computers (IBM R50).