http:^^www.ics.uci.edu^~jhummel^
来自「This data set contains WWW-pages collect」· EDU^~JHUMMEL^ 代码 · 共 212 行
EDU^~JHUMMEL^
212 行
Date: Thu, 21 Nov 1996 22:52:54 GMT
Server: Apache/1.0.5
Content-type: text/html
Content-length: 7411
Last-modified: Thu, 14 Nov 1996 10:32:01 GMT
<HTML><head><title> Joe Hummel</title></head><i> Information last modified November 14, 1996 </i> <h1> Joe Hummel </h1><ul><LI> Area: Computer Systems Design<LI> Advisor: Alex Nicolau<LI> Electronic Mail: <!WA0><A HREF="mailto:jhummel@ics.uci.edu">jhummel@ics.uci.edu</A><LI> Office: CSE 205<LI> Office Phone: (714) 824-2248</ul><H2> Research Project: </H2><i> Brief description: </i><P> Compilers, program optimization, and data dependence testing in the presence of dynamic, pointer-based data structures. <P><i> Long description: </i><P> Data structures are important solution components in many programs, sincethe proper choice of data structure can greatly impact the efficiency ofthe resulting application. This is well-known to FORTRAN programmers, wherearrays serve as the primary data structure. Recently however, given theincreasing use of languages such as Ada, C, C++, and FORTRAN 90, datastructures built using pointers and dynamic memory allocation are moreprevalent. These types of structures --- linked-lists, trees, DAGs --- presentnew and difficult analysis and optimization problems for the compilerwriter. The objective of our work is to solve these problems, therebyenabling the parallelization of such codes.<P> Our approach is to create a data dependence testing framework targetedtowards program code involving dynamic, pointer-based data structures. Ourframework has three components: (1) property collection, (2) path collection,and (3) dependence testing. To enable accurate dependence testing, thecompiler must collect both information about the properties of the datastructure (e.g. tree vs. DAG) and information about the memory locationsbeing manipulated in the data structure. The former are represented asaxioms about the structure, the latter as paths through the structure.Given this information, the dependence testing component applies theorem proving, in a general and decidable way, to determine dependencesin the code. <P> To date we have implemented a prototype of our data dependence testingframework, and successfully parallelized a number of important codesinvolving complex data structures. These codes include the N-body problem,Sparse matrices, and a 3D electromagnetic simulation based on Bipartitegraphs. Speedups were linear to near-linear on a shared-memorymultiprocessor.<P> Our work is crucial to optimizing and parallelizing compilers, sincedependence information directly impacts the application of numerousperformance-enhancing transformations. The more accurate the dependenceinformation, the more chances for optimization and parallelizationat all levels (from low-level instruction scheduling to source-levelloop transformations). Our work represents the first accurate and general mechanism for data dependence testing in the presense of dynamic, pointer-based data structures. <P><H2> Publications: </H2><H3> Refereed Journal Articles and Book Chapters </H3><P><ul><LI> Chapter 2: Path Collection and Dependence Testing in the Presence of Dynamic, Pointer-Based Data Structures, </A> <!WA1><A HREF="http://www.ics.uci.edu/~jhummel//~jhummel/papers/lcssc95.ps"> (full text in postscript) </A> <i>Languages, Compilers and Run-Time Systems for Scalable Computers (Proceedings of the 3rd Workshop)</i>, editors B. Szymanski and B. Sinharoy, Kluwer Academic Publishers, pp. 15--27, May 1995, by J. Hummel and L. Hendren and A. Nicolau.<P><LI> A Simple Mechanism for Improving the Accuracy and Efficiency of Instruction-level Disambiguation, </A> <!WA2><A HREF="http://www.ics.uci.edu/~jhummel//~jhummel/papers/lcpc95.ps"> (full text in postscript) </A> <i>Languages and Compilers for Parallel Computing</i> (August 1995), Lecture Notes in Computer Science, volume 1033, pp. 289-303, Springer-Verlag, 1996, by S. Novack and J. Hummel and A. Nicolau.<P><LI> Abstract Description of Pointer Data Structures: An Approach for Improving the Analysis and Optimization of Imperative Programs, </A> <!WA3><A HREF="http://www.ics.uci.edu/~jhummel//~jhummel/papers/loplas.ps"> (full text in postscript) </A> <i>ACM Letters on Programming Languages and Systems</i>, 1(3), pp. 243-260, September 1992, by J. Hummel and L. Hendren and A. Nicolau.</ul><H3> Refereed Conference Papers </H3><P><ul><LI> A Framework for Data Dependence Testing in the Presence of Pointers, </A> <!WA4><A HREF="http://www.ics.uci.edu/~jhummel//~jhummel/papers/icpp94.ps"> (full text in postscript) </A> <i>Proceedings of the 23rd Annual International Conference on Parallel Processing</i>, pp. 216--224, August 1994, by J. Hummel and L. Hendren and A. Nicolau.<P><LI> A General Data Dependence Test for Dynamic, Pointer-Based Data Structures, </A> <!WA5><A HREF="http://www.ics.uci.edu/~jhummel//~jhummel/papers/pldi94.ps"> (full text in postscript) </A> <i>Proceedings of the ACM SIGPLAN Conference on Programming Language Design and Implementation</i>, pp. 218--229, June 1994, by J. Hummel and L. Hendren and A. Nicolau.<P><LI> A Language for Conveying the Aliasing Properties of Dynamic, Pointer-Based Data Structures, </A> <!WA6><A HREF="http://www.ics.uci.edu/~jhummel//~jhummel/papers/ipps94.ps"> (full text in postscript) </A> <i>Proceedings of the 8th International Parallel Processing Symposium</i>, pp. 208--216, April 1994, by J. Hummel and L. Hendren and A. Nicolau.<P><LI> Applying an Abstract Data Structure Description Approach to Parallelizing Scientific Pointer Programs, </A> <!WA7><A HREF="http://www.ics.uci.edu/~jhummel//~jhummel/papers/icpp92.ps"> (full text in postscript) </A> <i>Proceedings of the 21st Annual International Conference on Parallel Processing</i>, pp. 100--104, August 1992, by J. Hummel and L. Hendren and A. Nicolau.<P><LI> Abstractions for Recursive Pointer Data Structures: Improving the Analysis and Transformation of Imperative Programs, </A> <!WA8><A HREF="http://www.ics.uci.edu/~jhummel//~jhummel/papers/pldi92.ps"> (full text in postscript) </A> <i>Proceedings of the ACM SIGPLAN '92 Conference on Programming Language Design and Implementation</i>, pp. 249--260, June 1992, by L. Hendren and J. Hummel and A. Nicolau.</UL><P><H2> Other Interests: </H2><P> Sailing, teaching, and teaching sailing. Visits to Washington, D.C. tosee Beth. In fact, here's a recent<!WA9><a href="http://www.ics.uci.edu/~jhummel//~jhummel/us-nz.JPG"> picture </a> of us in New Zealand...<P>If you like sailing... We sailed from Fiji to New Zealand aboard a 65' racing sloop owned and operated by Orange Coast College;the boat is called Alaska Eagle, and is an old Whitbread winner (Flyer).Pictures? Here's a goofy<!WA10><a href="http://www.ics.uci.edu/~jhummel//~jhummel/sail2-nz.JPG"> picture </a> of me driving; a<!WA11><a href="http://www.ics.uci.edu/~jhummel//~jhummel/sail1-nz.JPG"> picture </a> of "A watch" (Brad, Scott [the skipper], myself, and Ray [not pictured since he's driving]);a <!WA12><a href="http://www.ics.uci.edu/~jhummel//~jhummel/sail3-nz.JPG"> picture </a> of myself atopRoberts Island (Bay of Islands, NZ); and finally a<!WA13><a href="http://www.ics.uci.edu/~jhummel//~jhummel/sail4-nz.JPG"> picture </a> of a bunch of usatop Roberts Island (facing the other way -- you can just make outthe mighty Eagle anchored in the background, lower right). What a trip!<P><HR><ADDRESS> <!WA14><A HREF="http://www.ics.uci.edu">Department of Information and Computer Science </A>,<BR><!WA15><A HREF="http://www.uci.edu">University of California, Irvine</A> CA 92717-3425</ADDRESS></BODY></HTML>⌨️ 快捷键说明
复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?