A Performance Debugger for a Language Supporting Data Distribution Primitives

Data parallel languages based on user-supplied data distribution directives significantly simplify the development of the initial version of a parallel application. However, selection of good data distributions leading to efficient computations is often quite difficult. Therefore, performance debuggers are usually employed to yield insights into effects of data distribution. This paper presents design and implementation of a performance debugger that is specifically targeted to meet the performance debugging requirements of a data-parallel programming model based on user-specified data distributions. The visual interface of the performance debugger provides feedback regarding data distributions amongst processor nodes, data access patterns during program execution, as well as statistical performance data. The feedback provided by the performance debugger is explicitly related back to objects in the programmer's source code.