Dissertation Talk: A Framework for Productive, Efficient and Portable Parallel Computing
Presentation: Departmental | December 5 | 2-3 p.m. | 511 Soda Hall
Developing efficient parallel implementations and fully utilizing the available resources of parallel platforms remains a challenge to programmers due to the requisite low-level knowledge of the underlying hardware. This results in poor programmer productivity, limits portability of the application code and impedes experimentation with various algorithmic approaches for a given application. In this dissertation talk, I will present PyCASP, a Python-based framework that automatically maps Python application code to a variety of parallel platforms. PyCASP is an application-domain-specific framework that uses a systematic, pattern-oriented approach to offer a single productive software development environment for audio content analysis application writers. Using PyCASP, applications can be prototyped in Python code and our environment enables them to automatically scale their performance to modern parallel processors such as GPUs, multicore CPUs and clusters. We use the Selective Embedded JIT Specialization (SEJITS) mechanism to implement PyCASP's components based on application patterns. We enable composition of computations using three structural patterns: MapReduce, Iterator and Pipe and Filter. To illustrate our approach, we describe a set of four complete audio content analysis applications that are architected and implemented using PyCASP. We briefly describe two computational components of PyCASP: a Gaussian Mixture Model (GMM) component and a Support Vector Machine (SVM) component and use these to implement the example applications. We also analyze composition of computations using the three structural patterns and analyze some available optimizations for composing computations in audio analysis applications. We evaluate our approach with results on productivity and performance using the four example applications.