报告题目：Specializing Compiler Optimizations through Programmable Composition

报告人：Qing Yi, Assoc. Prof., Computer Science, U. of Colorado

报告时间：1月6日下午4:00-5:30

报告地点：西一楼344第一会议室

RESUME: Dr. Qing Yi is an associate professor of Computer Science at the University of Colorado, Colorado Springs.  She received her Ph.D. in Computer Science at RiceUniversity under the supervision of Professor Ken Kennedy, where her doctoral dissertation focused on advanced loop optimizations to improve the performance of deep memory hierarchies. She was a post-doctorate associate at the DOE Lawrence Livermore National Laboratory between 2003-2005 and was an assistant professor at the University of Texas at San Antonio between 2005-2012.  Her areas of research interests include compilers and programming languages, particularly compiler optimizations for high performance computing and tools for automatically improving software productivity, correctness, performance, and energy efficiency.  She has been the lead PI of several multi-year NSF and DOE grants, including the prestigious NSF CAREER award, and has developed the POET transformation language and optimization infrastructure funded both by NSF and by DOE Office of Science. Her ongoing research projects include the development of a Programming Interface and Runtime for Self-Tuning Scalable C/C++ Data Structures, a multilayer code synthesis framework to enhance both the correctness and performance of software applications, and a programmable code optimization and empirical tuning infrastructure for high end computing.

ABSTRACT: General purpose compilers aim to extract the best average performance for all possible user applications. Due to the lack of specializations for different types of computations, compiler attained performance often lags behind those of the manually optimized libraries.  This talk will demonstrate a new approach, programmable composition, to enable the specialization of compiler optimizations without compromising their generality.  By allowing different optimizations to adequately communicate with each other through a set of coordination handles and dynamic tags inserted inside the optimized code, we can specialize the composition of general-purpose compiler optimizations to attain a level of performance comparable to those of manually written assembly code by experts, thereby allowing selected computations in applications to benefit from similar levels of optimizations as those manually applied by experts.  The talk will discuss the integration of pattern-driven optimization specialization within an interactive compilation framework to achieve portable high performance for both kernels and large software applications, including analytical modeling of application efficiency, and library-aware coordinated optimization of the data structure and algorithm implementations.