Abstract:
We present a novel automated multicore benchmark synthesis framework for multicore systems including CPUs and GPUs with characterization and generation components to speed up architectural simulation of modern architectures. We rst identify a set of important application characteristics for CPUs and GPUs. Then, our framework captures these characteristics of original multicore applications and generates synthetic multicore benchmarks from those applications where synthetic benchmarks are a miniaturized form of applications that allow high simulation speeds and act as proxies of proprietary applications. We use parallel software architectural patterns in capturing important characteristics of CPU applications where we apply di erent machine learning techniques in a novel approach to automatically detect parallel patterns used in applications. In addition, we compare these techniques in terms of accuracy and speed and demonstrate that detecting parallel patterns is crucial for performance improvements and enables many architectural optimizations. The resulting synthetic benchmarks are small, fast, portable, human-readable, and they accurately re ect the key characteristics of the original multicore applications. Our synthetic CPU benchmarks use either Pthreads or Multicore Association (message passing and resource management) libraries and synthetic GPU benchmarks use OpenCL library. To the best of our knowledge, this is the rst time synthetic OpenCL benchmarks for GPUs are generated from existing applications. We implement our techniques for CPUs in the MINIME tool and generate synthetic benchmarks. Similarly, we implement our techniques for GPUs in the MINIME-GPU tool and experimentally validate them.
