Architectural and Compiler Support for Fine-Grained Helper Computing on Multicore Architectures

Oct 23

Friday, October 23, 2015

11:45 am - 1:00 pm
Social Sciences 119

Presenter

James Tuck

Due to the importance of reliability and security, prior works have proposed inlining meta-functions into applications for detecting bugs and security vulnerabilities. Such software techniques often add frequent, fine-grained instrumentation to programs and incur large runtime overheads. One way to reduce the cost is through parallelism. However, current architectures and compilers are not well suited to this task. At the architectural level, we found that current multicores do not efficiently support fine-grained and frequent communication between cores. We propose a hardware-accelerated queue (HAQu). HAQu adds hardware to a CMP that accelerates operations on software queues in an application¿s address space with operations that are compatible with a fully software queue. In the compiler, we studied a parallelization strategy that automatically extracts meta-functions from the main application and schedules their work load among many threads. James Tuck is an Associate Professor of Electrical and Computer Engineering at North Carolina State University (NCSU). He joined NCSU as an Assistant Professor in 2007 after receiving his Ph.D. and M.S. degrees from the University of Illinois in Urbana-Champaign and a B.E. from Vanderbilt University in 1999. His research focuses on a variety of topics centering around multicore architectures and code optimization. Lunch served at 11:45 am. Seminar starts at 12 noon.

Contact

Naseree, Alexandra
660-5241
alexandra.naseree@duke.edu