Mapping reference code to irregular DSPs within the retargetable, optimizing compiler COGEN(T)

摘要

Micro continues to be the major forum for conceptual ideas andrelevant studies for optimal processing, both in microarchitectureand in the compiler technology that is equally critical. It isfitting that we hold the conference in Austin, one of the leadingcenters, both academic and industrial, for the development of thistechnology. At The University of Texas at Austin, we havesubstantive, relevant research going on in two departments,Electrical and Computer Engineering and Computer Science. In thelocal companies, we have some of the world's leading cutting- edgeprocessors being developed. The Austin ecosystem benefits from thesymbiotic activities of both. If you are new to Austin, I hope youwill make time to visit the University and some of the Austincompanies.>The program committee, Josh Fisher and Paolo Faraboschipresiding, have put together an outstanding program that deals withnot only current critical problems, but also expected futurerequirements. I hope you find the papers useful to your needs. Weare particularly fortunate to have two excellent Keynote Speakers,Harvey Cragon and Andy Wolfe. Harvey has been doingmicroarchitecture since before many of you were born. He isresponsible for, among other things, the Texas Instruments AdvancedScientific Computer. lie is an Eckert-Mauchly Award winner, a PioreMedal winner, member of the National Academy of Engineering, and aformer endowed Chair at The University of Texas at Austin. Andy isSenior Vice President and CTO at SONICblue where he is responsiblefor product strategy and for mergers and acquisitions. He taught atPrinceton from 1991-1997 and is currently a Consulting Professor atStanford. He is on the advisory boards of several companies. We aredelighted that both accepted our invitation to participate inMicro-34.>Also, we have continued the tradition of workshops and tutorialsthe weekend before the actual symposium. Scott Mahlke, theWorkshop/Tutorials chairman, has put together a collection of 4workshops and 3 tutorials on timely state of the art matters. Ihope you take advantage of them, as well.>I recognize the danger in holding Micro-34 so close to 6thstreet and Town Lake. Austin is well-known for its live music,excellent restaurants, and year-round outdoor activities. Beforeyou arrived, I hope you produced a static schedule for yourattention units that saves 6th Street for Tuesday evening andWednesday after 2pm. I realize that dynamic scheduling could alterthat, so I wish you the wisdom to make the correct dynamicscheduling decisions during the next few days.>This 34th year of the symposium confirms Micro's stature as thepremier technical forum on microarchitecture. Once again, we puttogether an exciting program that highlights emerging research inhigh-performance processor microarchitecture, instruction-levelparallelism, and compiler optimization.>Continuing the trends of the last few years, embedded computingand energy efficiency cover a significant fraction of the program.To give you an idea of the growing importance of this, about 25percent of the submitted papers had the word 'Energy' or 'Power' intheir titles.>As usual, Micro attracted some of the best contributions fromboth academia and industry in the form of 144 high-quality papers-- yet another all-time record for Micro. This presented a verychallenging task to the program committee, whose job was to selectthe best papers among those received. Twenty-nine experts fromacademia and industry composed the program committee.>The entire process -- from author registration to papersubmission, reviews collection and rebuttals -- was organized in anSQL database hosted by the University of Colorado. Special thanksgo to Dan Connors, who personally administered the web site, andwhose efforts were central to the success of the programcommittee.>The quality of the review process and the diligence of reviewersis a key factor in the success of a conference. Includingthemselves, the program committee members recruited 431 referees togenerate the 872reviews used in the paper selection process. Thisamounts to an average of 6.05 reviews per paper, with all but a fewpapers receiving at least five reviews. This year, we introduced anew step in the review process that enabled authors to provide ashort rebuttal to the reviews before the paper selection phase. Theauthors' response was widely used during the program committee tohelp clarify many doubts. Despite the additional complexity that itadded to the process, it was unanimously considered a successfulexperiment and a significant improvement to the review process.>Twenty-one Program Committee members met in an MIT conferenceroom (Cambridge, Mass.) on August 11, and spent about eleven hoursto select the 29 papers that compose the Micro-34 program. Webelieve that these papers contain important ideas and results thatwill contribute to future successful research and productdevelopment efforts.