Prospector: A Dynamic Data-Dependence Profiler To Help Parallel Programming

Provided by: Georgia Tech
Topic: Big Data
Format: PDF
Multiprocessor architectures are increasingly common these days. In order to fully and efficiently utilize the abundant hardware parallelism, translating many sequential programs into parallel programs is a pressing need. Although many compilers support automatic parallelization, most programmers are still manually parallelizing their applications. To help parallelizing applications especially legacy programs written by other programmers, the authors propose prospector. Prospector is a profile-based parallelism identification tool using dynamic data dependence profiling results. It also advises on how to parallelize the identified sections.

Find By Topic