US-Cameroon Planing Research Visit on Combined Binary Code Translation and Synthesis for Heterogeneous Multiprocessor Systems, January 2014
University Of Arkansas, Fayetteville AR
Investigators
Abstract
1346542 Bobda This award will provide support for PI Dr. Christophe Bobda, from the University of Arkansas to organize a 10 day planning research visit in December 2013-January 2014 to the University of Yaounde in Yaounde and the University of Dschang in Dschang, both in Cameroon. He will be accompanied by two students from the University of Arkansas. The main objective of this proposed trip is to meet with Dr. Rene Ndoundam of the University of Yaounde and associate faculty at the University of Dschang to develop a master plan for international research. They will (1) develop plans for future collaboration in the fields of compilers, architecture synthesis and run-time architecture optimization, (2) put in place a research, training and exchange framework for the remote collaboration, (3) set the basis for the development of international grant proposals to sustain this effort, and (4) organize seminars and FPGA labs with Cameroonian students and faculty to teach them technical aspects in computer architecture and multicore design on FPGA devices. Intellectual Merit: This project will help leverage the performance of new architecture on legacy code, thus saving time and money for redesign. With the reduced amount of elements to configure, the proposed coarse-grained architecture will allow for a fast run-time synthesis of selected code blocks onto the configurable processor. The proposed coarse-grained architecture uses a combination of direct interconnect and network on chip elements to increase the flexibility at device level, while maintaining high performance within a component boundary. The proposed run-time spectral-based synthesis approach uses the singular values of the Laplacian of the incident instruction-matrix to compute the placement of instructions in a 2D field in such a way that the total interconnect cost among instructions is minimized. Finally, the proposed generic and configurable streaming component will allow for a transparent management of data exchange and synchronization among code blocks executing on the reconfigurable processor and those running on the remaining processors. Broader Impacts: One of the major outcomes of this project is the run-time translation of legacy binary code for heterogeneous multiprocessor systems. This will allow for automatic migration of existing programs onto new architectures, thus improving their performance while avoiding complete redesign and recompilation. The planning visit will lay the groundwork for a broader international collaboration network in which new partners could be easily integrated. The research components produced by this collaborative effort would benefit undergraduate and graduate students in the United States and in Cameroon. The two students from the University of Arkansas will gain international experience by presenting their work at the seminars organized at the University of Yaounde and the University of Dschang. Beyond the planning visit, the PI will actively recruit several undergraduate researchers, especially women and minorities, from the U of A to work on compilers, architecture design and applications. Under-represented groups in science and engineering research will be recruited through the U of A Engineering Career Awareness Program (ECAP), Arkansas Louis Stokes Alliance for Minority Participation (ARK-LSAMP), and the George Washington Carver Project.
View original record on NSF Award Search →