A Strategic Profiler for Glasgow Parallel Haskell

Execution prooling plays a crucial part in the performance-improving process for parallel functional programs. This paper presents the design, implementation, and use of a new execution time prooler (GranSim-SP) for Glasgow Parallel Haskell (GpH). Parallelism is introduced in GpH by using evaluation strategies, which provide a clean way of coordinating parallel threads without destroying a program's original structure. The new prooler attributes the cost of evaluating parallel threads to the strategies that created them. A unique feature of the strategic prooler is that the call-graph of evaluation strategies is maintained , allowing the programmer to discover the sequence of (perhaps nested) strategies that were used to create any given thread. The strategic prooler is demonstrated on several examples, and compared with other proolers.