`
`BEFORE THE PA TENT TRIAL AND APPEAL BOARD
`
`INTEL CORPORATION,
`
`Petitioner
`
`V.
`
`FG SRC LLC,
`
`Patent Owner
`
`CASE NO. : 2020-01449
`PATENT NO. 7, 149,867
`
`SUPPLEMENTAL DECLARATION OF RAJESH K. GUPTA, PH.D.
`
`Mail Stop PA TENT BOARD
`Patent Trial and Appeal Board
`U.S. Patent and Trademark Office
`P.O. Box 1450
`
`Alexandria, VA 223 13- 1450
`
`Intel Exhibit 1030 - 1
`
`
`
`I, Dr. Rajesh K. Gupta, declare as follows:
`
`1.
`
`I am currently Professor and Qualcomm Endowed Chair at the
`
`Department of Computer Science and Engineering at University of California, San
`
`Diego ("UCSD"). I have served in that role since 2002. My resume and a detailed
`
`description ofmy work on the MORPH/AMRM project can be found in my
`
`declaration dated August 10, 2020, which was filed in the above-captioned case as
`
`Exhibit 1010.
`
`2.
`
`I make this declaration based on my personal knowledge. I have not
`
`been compensated for my time or efforts in providing this testimony.
`
`Publications ofMORPH/AMRM Papers
`
`3.
`
`Based on my personal knowledge as a member of the Institute of
`
`Electrical and Electronics Engineers, Inc. ("IEEE") and the IEEE Computer
`
`Society, as well as an organizer, frequent attendee, and presenter at many such
`
`IEEE conferences since at least 1996, the IEEE and the IEEE Computer Society
`
`sponsor many technical meetings each year. Those conferences were ( and are)
`
`considered premier conferences in the computer science field and were attended by
`
`leading experts and skilled practitioners in that field. Papers presented at these
`
`conferences were published by the IEEE as part of conference proceedings that
`
`were distributed to attendees in hard copy and then online. Once conference
`
`proceedings were published, in print and/or online, the public had access to print
`
`versions in libraries, and subscribing members had full electronic access to
`
`individual conference articles and the online abstracts and tables of contents were
`
`available for free to anyone.
`
`4.
`
`In the first half of 1996, Dr. Andrew A. Chien and I coauthored a
`
`paper entitled MORPH: A System Architecture for Robust Higher Performance
`
`Using Customization. We presented this paper at Frontiers '96, The Sixth
`
`Symposium on the Frontiers of Massively Parallel Computing. ("Frontiers '96
`
`2
`
`Intel Exhibit 1030 - 2
`
`
`
`Conference"). The Frontiers '96 Conference was sponsored by the IEEE,
`
`specifically the IEEE Computer Society, and held in Annapolis, Maryland between
`
`October 27-31, 1996.
`
`5.
`
`I confirm based on my personal knowledge that this paper was
`
`submitted to the conference organizers before the Frontiers '96 Conference and
`
`that this paper was included in the IEEE printed publication that was distributed to
`
`the conference attendees during the conference. The paper that I co-authored was
`
`published as pp. 336-345 of the Frontiers '96 Conference Proceedings by the IEEE
`
`in 1996. As with other IEEE conferences, this paper was made available in 1996
`
`to conference attendees at the Frontiers '96 Conference no later than the last day of
`
`that conference. My recollection is also consistent with my overall personal
`
`experience in attending conferences sponsored by the IEEE and the general
`
`practice in the scientific and engineering community during this period. The
`
`conference organizers generally reviewed submitted papers to determine who
`
`would be invited to speak at the conference. The conference materials were
`
`typically printed in advance so they could be distributed at the conference, as it
`
`would have been impractical to mail the conference presentations to all the
`
`conference participants after the conference. I attended the Frontiers '96
`
`Conference and received a printed copy of these published papers no later than the
`
`last day of that conference. A copy of this paper from my personal files is attached
`
`to this declaration as Appendix G 1001.
`
`6.
`
`The purpose, scope, nature of presentations, and intended audience of
`
`the Frontiers '96 Conference are described in the front matter to the printed
`
`proceedings, available from the IEEE Computer Society Digital Library at
`
`https ://www.computer.org/ csdl/proceedings/frontiers/ l 996/l 2OmNCaLEmP,
`
`attached as Appendix GI 002. These materials refresh and confirm my
`
`recollections of the Frontiers '96 Conference.
`
`3
`
`Intel Exhibit 1030 - 3
`
`
`
`7.
`
`In 1997, after I had joined the faculty at University of California,
`
`Irvine ("UCI"), I co-authored a paper entitled Architectural Adaptation for
`
`Application-Specific Locality Optimizations with Xingbin Zhang, Ali Dasdan, and
`
`Dr. Chien (all at University of Illinois, Urbana-Champaign ("UIUC") at the time)
`
`and Martin Schulz ( at the Institut filr Infonnatik, Technische Universitat
`
`Mi.inchen). This paper was presented at the International Conference on Computer
`
`Design - VLSI in Computers and Processors ("VLSI '97 Conference"). The
`
`VLSI '97 Conference was sponsored by the IEEE, specifically the IEEE Computer
`
`Society Technical Committee on Design Automation and the IEEE Circuits and
`
`Systems Society, and was held in Austin, Texas between October 12-15, 1997.
`
`8.
`
`I confirm based on my personal knowledge that this paper was
`
`submitted to the conference organizers before the VLSI '97 Conference and that
`
`this paper was included in the IEEE printed publication that was distributed to the
`
`conference attendees. This paper that I co-authored was published as pp. 150-156
`
`of the VLSI '97 Conference Proceedings by the IEEE in 1997. The copyright page
`
`of the VLSI '97 Conference Proceedings contains the following statement: "The
`
`papers in this book comprise the proceedings of the meeting mentioned on the
`
`cover and title page. They reflect the author's opinions and, in the interests of
`
`timely dissemination, are published as presented and without change." See Exhibit
`
`1003 at 2; see also Exhibit 1028, Appendix EDM0 1 at 4. This is consistent with
`
`my overall personal experience in attending conferences sponsored by the IEEE
`
`Computer Society and the general practice in the scientific and engineering
`
`community during this period. The conference organizers generally reviewed
`
`submitted papers to determine who would be invited to speak at the conference.
`
`The conference materials were typically printed in advance so they could be
`
`distributed at the conference, as it would have been impractical to mail the
`
`conference presentations to all the conference participants.
`
`4
`
`Intel Exhibit 1030 - 4
`
`
`
`9.
`
`The purpose, scope, nature of presentations, and intended audience of
`
`the VLSI '97 Conference are described in the front matter to the printed
`
`proceedings, available from the IEEE Computer Society Digital Library at
`
`https://www.computer.org/csdl/proceedings/iccd/l997/l20mNB8Cj8h, attached as
`
`Appendix GI 003. These materials refresh and confirm my recollections of the
`
`purpose, scope, nature of presentations, and intended audience of the industry
`
`conferences sponsored by the IEEE and the IEEE Computer Society during this
`
`time frame.
`
`10.
`
`In 2000, I authored a research paper entitled Architectural Adaptation
`
`in AMRM Machines . I presented this paper at the Proceedings of the IEEE
`
`Computer Society Workshop on VLSI 2000 ("VLSI '00 Workshop"). 1 The
`
`VLSI ' 00 Workshop was sponsored by the IEEE, specifically the IEEE Computer
`
`Society Technical Committee on VLSI, and held in Orlando, Florida between April
`
`27 - 28, 2000.
`
`11.
`
`I confirm based on my personal knowledge that this paper was
`
`submitted to the conference organizers before the VLSI '00 Workshop and that this
`
`paper was included in the IEEE printed publication that was distributed to the
`
`conference attendees during the conference. The paper that I authored was
`
`published as pp. 75-79 of the VLSI '00 Workshop Proceedings by the IEEE in
`
`2000. As with other IEEE conferences, this paper was made available to the
`
`conference attendees at the VLSI '00 Workshop no later than the last day of that
`
`conference. My recollection is also consistent with my overall personal experience
`
`in attending conferences sponsored by the IEEE Computer Society and the general
`
`practice in the scientific and engineering community during this period. The
`
`conference organizers generally reviewed submitted papers to determine who
`
`1 I have reviewed my prior declaration (Ex. 1010) and confirm that the reference to
`"in 1997" in paragraph 25 was a typographical error and should read "in 2000."
`
`5
`
`Intel Exhibit 1030 - 5
`
`
`
`would be invited to speak at the conference. The conference materials were
`
`typically printed in advance so they could be distributed at the conference, as it
`
`would have been impractical to mail the conference presentations to all the
`
`conference participants. I attended the VLSI '00 Workshop and received a printed
`
`copy of these published papers no later than the last day of that conference.
`
`12. The purpose, scope, nature of presentations, and intended audience of
`
`the VLSI '00 Workshop are described in the front matter to the printed
`
`proceedings, available from the IEEE Computer Society Digital Library at
`
`https://www.computer.org/csdl/proceedings/wvlsid/2000/ l20mNzl3WWV,
`
`attached as Appendix G 1004. These materials refresh and confirm my
`
`recollections of the VLSI ' 00 Workshop.
`
`13. As stated in my prior declaration, I reviewed each ofExhibits 1003,
`
`1004, and 1005, and I confirm based on my personal knowledge as a co-author or
`
`author that each is what it purports to be. Specifically, Exhibit 1003 is a true and
`
`c01Tect copy of the article that I co-authored and submitted for the International
`
`Conference on Computer Design - VLSI in Computers and Processors in 1997,
`
`titled Architectural Adaptation for Application-Specific Locality Optimizations;
`Ex. l 004 is a true and correct copy of the article that I authored and submitted for
`the Proceedings of the IEEE Computer Society Workshop on VLSI 2000 in 2000,
`
`titled Architectural Adaptation in AMRM Machines; and Ex. 1005 is a true and
`
`correct copy of the article that I co-authored and submitted for the Frontiers '96,
`
`The Sixth Symposium on the Frontiers of Massively Parallel Computing in 1996,
`
`titled MORPH: A System Architecture for Robust Higher Performance Using
`Customization.
`
`6
`
`Intel Exhibit 1030 - 6
`
`
`
`I declare under penalty of perjury that the foregoing is true and correct.
`
`Date: March 31, 2021
`
`7
`
`Intel Exhibit 1030 - 7
`
`
`
`Appendix G1001
`Appendix G1001
`
`Intel Exhibit 1030 - 8
`
`Intel Exhibit 1030 - 8
`
`
`
`
`
`Frontiers 96
`
`———<==
`
`7
`
`The Sixth Symposium on
`The Frontiers of Massively Parallel Computation
`
`t
`
`
`
`
`
`October 27 - 31, 1996
`
`Annapolis, Maryland
`
`Sponsored by
`IEEE Computer Society
`
`
`
`EE
`
`®CompUTER
`CIE & ELECTRONICSENGINEERS,INC.
`TY
`TE So
`THE INSTITUTE OF ELECTRICAL AND
`le 5 )YEARS OF SERVICE *1946-1996
`
`ah Se ea
`
`__ IntelExhibit
`
`Intel Exhibit 1030 - 9
`
`
`
`Proceedings
`
`Frontiers ‘96
`
`The Sixth Symposium onthe
`Frontiers of Massively Parallel Computing
`
`:
`
`October 27-31, 1996
`Annapolis, Maryland
`
`Sponsored by
`IEEE Computer Society
`
`In cooperation with
`NASA Goddard Space Flight Center
`USRA/CESDIS
`
`eeameoaew
`aeSNESTA
`—_
`
`®
`|JEEE Computer Society Press
`Los Alamitos, California
`Brussels
`@
`
`Tokyo
`
`Washington
`
`®@
`
`
`
`Intel Exhibit 1030 - 10
`
`
`
`
`
`
`
`{EEE ComputerSociety Press
`10662 Los Vaqueros Circle
`P.O.Box 3014
`Los Alamitos, CA 90720-1264
`
`
`Copyright © 1996 by The Institute of Electrical and Electronics Engineers, Inc,
`All rights reserved.
`
`Copyright and Reprint Permissions: Abstracting is permitted with credit to the source. Libraries may
`photocopy beyondthe limits of US copyright law, for private use of patrons, those articles in this volume
`that carry a codeat the bottom ofthe first page, provided that the per-copy fee indicated in the code is
`through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923,
`R
`Other copying,reprint, or republication requests should be addressed to:
`IEEE Copyrights Manager, IEEE
`Service Center, 445 Hoes Lane, P.O, Box 1331, Piscataway, NJ 08855-1331.
`
`paid
`
`The papers in this book comprise the proceedings ofthe meeting mentioned on the cover and title page. They
`reflect the authors’ opinions and,
`in the interests of timely dissemination, are published as presented and
`without change. Their inclusion in this publication does not necessarily constitute endorsement by the
`editors, the IEEE Computer Society Press, or the Institute ofElectrical and Electronics Engineers, Inc.
`
`IEEE Computer Society Press Order Number PRO7551
`IEEE Order Plan Catalog Number 96TB 100062
`ISBN 0-8186-7551-9
`Microfiche ISBN 0-8186-7553-5
`ISSN 1088-4955
`
`Additional copies may be orderedfrom:
`IEEE Computer Society
`IEEEService Center
`IEEE Computer Society
`IEEE Computer Society Press
`
`
`Customer Service Center 13, Avenue de l’Aquilon—Ooshima Building445 Hoes Lane
`10662 Los Vaqueros Circle
`P.O. Box 1331
`B-1200 Brussels
`2-19-1 Minami-Aoyama
`P.O, Box 3014
`Piscataway, NJ 08855-1331
`BELGIUM
`Minato-ku, Tokyo 107
`Los Alamitos, CA 90720-1314
`Tel: +1-908-981-1393
`Tel: +32-2-770-2198
`JAPAN
`Tel: +1-714-821-8380
`Fax; +1-908-981-9667
`Fax: +32-2-770-8505
`Tel: +81-3-3408-3118
`Fax; +1-714-821-464]
`misc.custsery@computer.org
`euro.ofc@computr.org
`Fax; +81-3-3408-3553
`Email; ¢s.books@computer.org
`tokyo.ofe@computer.org
`
`Editorial production by Penny Storms
`Cover by Kerry Bedford and Alex Torres
`Printed in the United States of America by KNI,Inc.
`
`& The Institute of Electrical and Electronics Engineers, Inc.
`
`
`
`Intel Exhibit 1030 - 11
`
`
`
`: e
`
`s
`xsaTh
`
`ereTPa
`
`aeeee
`
`——
`
`Panel Session—Petaflops Alternative Paths
`Panel Chair: Paul Messina, California Institute of Technology
`
`Session 7: Invited Speaker
`Independence Day
`Steven Wallach, HP-Convex
`
`Session 8A: Synchronization
`A Fair Fast Distributed Concurrent-Reader Exclusive-Writer Synchronization........--.++0++ 246
`T.J. Johnson and H. Yoon
`Lock Improvement Technique for Release Consistency in Distributed
`Shared Memory Systems.......scssssssesssssssssscscsrsenessosssnenssesucanansnaracnersessnnsasrsscsarsncenssnarseensestestseiees
`S.S. Fu and N-F. Tzeng
`A Quasi-Barrier Technique to Improve Performance of an Irregular Application.......-.+-+- 263
`H.V. Shah and J.A.B. Fortes
`
`255
`
`272
`
`282
`
`Session 8B: Networks
`Performance Analysis and Fault Tolerance of Randomized Routing on
`Clos NetevOPiS occccecscrcsscscoyseseenerceostasocneeonseroscrbevsansevsnorcssscsacthscznasccaica eaaa ca tcce aaa amas
`M.Bhatia and A.Youssef
`Performing BMMC Permutations in Two Passes through the Expanded
`Delta Network and MasPar MP-2 ...-ssssss-sssssssesssssssoseosstnnnunnnnnessssensseseT
`L.F. Wisniewski, T.H. Cormen, and T. Sundquist
`Macro-Star Networks: Efficient Low-Degree Alternatives to Star Graphs
`for Large-Scale Parallel TACHAECEITES asssviainiscaasabshcderesnrsssecoacarressrorbevtonsmereraascoseredalstansasstanaa
`C-H. Yeh and E. Varvarigos
`Session 9A: Performance Analysis
`Modeling and Identifying Bottlenecks in EOSDIS..........-s-ccsnsrseersnneettnnesssnennnnmanssannatesee 300
`J. Demmel, M.Y. Ivory, and S.L. Smith
`Tools-Supported HPF and MPIParallelization ofthe NAS Parallel Benchmarks......-.--+-+- 309
`C. Clémengon, K.M.Decker, V.R. Deshpande, A. Endo,
`J. Fritscher, P.A.R. Lorenzo, N. Masuda,A. Miiller,
`R. Ruhl, W. Sawyer, BJ.N. Wylie, and F. Zimmerman
`A Comparison ofWorkload Traces from Two Production Parallel Machines.......--ss+:-ssss0
`K. Windisch, V. Lo, D. Feitelson, R. Moore, and B. Nitzberg
`Morphological Image Processing on Three Parallel Machines pedenelReGaeeedenstetteeaciensonnesrteoere
`M.D. Theys, R.M.Born, M.D. Allemang, and H.J. Siegel
`Session 9B: Petaflops Computing/ Point Design Studies
`MORPH:A System Architecture for Robust High Performance Using
`Customization (An NSF 100 TeraOps Point Design Study)ssccsscosssserensnsncasseesecerseersessnesssnse
`A.A. Chien and R.K. Gupta
`Architecture, Algorithms and Applications for Future Generation
`cecccscsccccescsnceene®
`Opel cerriudsvacarcetcaasaacnsaceatanst7z sntcnhtons) aims)
`Supercomputers --rrvrrrrecerererrs
`V. Kumar, A. Sameh, A. Grama, and G. Karypts
`
`319
`
`327
`
`-
`sis
`
`vii
`
`Intel Exhibit 1030 - 1
`
`anneaa
`
`Intel Exhibit 1030 - 12
`
`
`
`a
`
`MORPH: A System Architecture for Robust High Performance
`Using Customization
`(An NSF 100 TeraOps Point Design Study)
`Andrew A. Chien
`
`Rajesh K. Gupta
`Departmentof ComputerScience
`University of Illinois
`Urbana,Illinois 61801
`achien,rgupta} @cs.uiuc.edu
`
`Abstract
`Achieving 100 TeraOps performance within a ten-
`year horizon will require massively-parallel architec-
`tures that exploit both commodity software and hard-
`ware technology for cost efficiency.
`Increasing clock
`rates and system diameter in clock periods will make
`efficient management of communication and coordina-
`tion increasingly critical. Configurable logic presents a
`unique opportunity to customize bindings, mechanisms,
`and policies which comprise the interaction of process-
`ing, memory, I/O and communication resources. This
`programming flexibility, or customizability, can pro-
`vide the key to achieving robust high performance.
`The MultiprocessOr with Reconfigurable Parallel
`Hardware (MORPH) uses reconfigurable logic blocks
`integrated with the system core to control policies, in-
`teractions, and interconnections. This integrated con-
`figurability can improve the performanceof local mem-
`ory hierarchy,
`increase the effictency of interproces-
`sor coordination, or better utilize the network bisec-
`tion of the machine. MORPH provides a framework
`for exploring such integrated application-specific cus-
`tomizability. Rather than complicate the situation,
`MORPH’s configurability supports component software
`and interoperabilty frameworks, allowing direct support
`forapplication-specified patterns, objects, and struc-
`tures. This paper reports the motivation and initial
`design of the MORPH system,
`
`1
`
`Introduction
`Increasing reliance on computational techniques for
`scientific inquiry, complex systems design, faster than
`real life simulations, and higher fidelity human
`com-
`puter interaction continue to drive the need or ever
`higher performance computing systems. Despite rapid
`progress in basic device technology [1], and even in
`uniprocessor computing technology [2], these applica-
`
`tions demand systems scalable in every aspect: pro-
`cessing, memory, I/O, and particularly communica-
`tion. In addition to advances in raw processing power,
`we must also achieve dramatic improvementsin system
`usability. Current day scalable systemsarestill quite
`difficult to program, and in many cases effectively pre-
`cluding use of the most. sophisticated (and mosteffi-
`cient) algorithms. Even whensuccessfully used, most,
`systems exhibit substantial performance fragility due
`to rigid architectural choices that do not work well
`across different applications.
`Based on technology projections for the 2007 design
`window for the NSF point design studies, our analyses
`indicate that increases in communication cost relative
`to computation (gate speeds) make configurable logic
`practical in an ever broader range of the system. The
`benefit of configurable logic is that it can be used to
`customize the machine’s behavior to better match that
`required by the application — in essence a machine can
`be tuned for each application with little or no perfor-
`mance penalty for this generality. While a broad va-
`riety of such architectures are possible, MORPH is a
`design point which explores the potential of integrating
`configurability deep into the system core.
`Because technology trends continue to increase the
`importance of communication, the MORPH architec-
`ture focuses on exploiting configurability to manage
`locality, communication, and coordination. In particu-
`lar, the MORPHdesign studyis exploring improved ef-
`ficiency andscalability by exploring novel mechanisms
`for binding and mechanisms which comprise the in-
`teraction of processing, memory, I/O, and communi-
`cation resources, Other innovations explore flexible
`hardware granularity (e.g. mechanisms and associa-
`tion with processors and memory) and memory Bye
`tem management (e.g. cache coherence, prefetching,
`
`1088-4955/96 $5.00 © 1996 IEEE
`
`336
`
`Intel Exhibit 1030 - 13
`
`Intel Exhibit 1030 - 13
`
`
`
`ae ininterchip wiring. However, commu-
`llTemain critical inachieving high perfor-
`
`Semiconductor Industry Associa-
`
`decade hence indicate advanced
`
`Hfeature size. However at the deep
`
`on evel feature
`sizeas measured bytransistor
`
`ength isincreasingly irrelevant for velocity-
`sa uurated carrier transport
`[3]. Both logic density and
`
`speed ared r
`by
`thein rconnectdensity: Pitch
`
`
`ect
`lengths are any-
`e pitch, that is, up
`delay. This limits
`
`comn
`
`
`bit 1030 - 14
`
`
`Intel Exhibit 1030 - 14
`
`
`
`of less than 5ns. As an alternative organization, cur-
`rent FPGA devices also offer up to 40 Kbits of mul-
`tifunctional memory in addition to over 60,000 us-
`able gates in implementation of specialized hardware
`functions such as arithmetic or DSP functions. Un-
`like standard SRAM parts, the embedded memory can
`be used as multiple-ported SRAM or FIFO providing
`much greater flexibility in system organization. Cur-
`rent efforts in FPGA design and architecture show
`significant improvements in the efficiency of the on-
`chip memory blocks. This trend in memory efficiency
`and utilization is expected to continueand close the
`gap between FPGA and SRAM densities and up to
`10 Mbitsof multi-functional embedded memorywould
`be availableiinadditiona th lorieblocks.insingle-
`
`nections are modified based on applications will soon
`
`be acommonfeature to allow programmable logic to be
`
`embedded in key modules of a system andprovide on-
`
`line programmability to change hardware functional-
`
`ity. Tools for distributed hardware control synthesis to
`
`allow dynamic binding of hardwareresources [7], and
`
`synthesis of protocols to low latency hardware(9,10, 3)
`
`have been successfully demonstrated. With these CAD
`
`and synthesis capabilities, embedded programmable
`
`logic can be inserted into the key parts of systems,
`
`andused toalter behavior dramatically with modest
`
`performance overhead.
`
`
`_ Thearchitectural SensOneof these Mee)
`
`
`
`
`bit 1030 - 15
`
`Intel Exhibit 1030 - 15
`
`
`
`Figure 1: A Flexible 100 TeraOp Architecture
`
`either acache-coherent machine, a non-cache coherent
`achine, or even clusters of cache coherent machines
`
`‘connected by put/getormessage passing. Varying bit 1030 - 16
`
`Intel Exhibit 1030 - 16
`
`
`
`eucecheorraniecticn
`‘
`anization(blocksizeor objects)
`B= heyS. a
`
`
`
`Figure2: A ‘S
`(AddressSpace Seeorganizations
`<
`+ A wide rang
`,
`.
`ured.
`:
`‘Cache coherence) canbe config-
`
`ementereEeeryplicationsthephys:
`seevaplero, global network) andby adcs
`:
`urce
`m:
`é
`Sadie
`special hardware structures [30, 31] such as fast bad5
`rier or broadcast support for machine subsets or ‘i '
`entire machine, to optimize performance. For exam-
`ple, experience over the last ten years demonstrates
`that intraprocessor communication mechanisms (data_
`shared through the cache) are muchmore efficient than-
`{ interprocessor mechanisms. When ma-
`chine configuration granularity matches the applica~
`tion,extremelyhigh performance can result. The pro-
`grammablelogicon both processorelements and mem-
`oryelements
`us to dynamically associate mem-
`ies with
`SSOT
`ips,
`changing the node granu-J
`on set up. In another example, some
`5
`m low-latencybarriers,
`icast. Suchstructu
`
` even the bes
`
`bit 1030 - 17
`
`Intel Exhibit 1030 - 17
`
`
`
`feweaeeee likely to present too inflexible a
`dettnndi: P
`oth a wide range of applications and
`
`believe ie irregular, adaptive applications well. We
`wide
`at, achieving scalable high performance on a
`
`elesion of applications demands the development
`
`A
`Ologies (automatic and high level abstractions
`
`“or programmerassisted decisions) to exploit the flex-
`
`ibility of our proposed architecture.
`
`There are two basic types of techniques for identi-
`
`fying opportunities for customization: static analysis
`
`(compileranalysis and directives) and dynamic adap-
`
`tation (profiles and dynamicstatistics) to rationally
`
`makeuse of theflexibility to optimize the mapping and
`
`execution of the program. It is imperative that good
`
`performance
`beachievablewith modest effort and the
`
`evels
`of performance be available with reason-
`
`
`iqueswhichexploit aggressive
`s
`(35, 36, 37,38, 39, 40, 41, 42],
`statistics to optimize pro-
`essential to the pro-
`
`
`
`bit 1030 - 18
`
`
`Intel Exhibit 1030 - 18
`
`
`
`=
`
`ewing for Perfor-
`
`5 Application-Driven Customizability
`
`not possible. Critical issues in assessing mechanisms
`include hardware cost, cycle-time impact, configura-
`tioncost, effect on software, protection mechanism(s)
`needed, etc. We describe several illustrative examples
`below to showthe leverage and importance of a flex-
`:
`—
`Fey aia ible architecture. However, with such a small group,
`ACETS
`ASU
`Te
`these are neither representative nor typical; however,
`Figure 3: A
`Architect
`Software Architecturefora100TeraOp
`_they do illustratetheoverall architectural framework —
`nosis anrace Thispicture.
`y
`do
`ulust)
`wa
`&
`woes
`shows useof
`intime diag-
`;
`RPH provides.
`synthesized hardware ye eeecine
`that MORP iPro ice
`k
`td
`“4
`: Strid
`
`Sin a nut
`
`bit 1030 - 19
`
`Intel Exhibit 1030 - 19
`
`
`
`
`ee zations to be implemented with low over-
`latency etthe communication reduction and lower
`
`course,
`the efits without computational overhead. Of
`tomoe fre are a wealth of cache system optimiza-
`
`beans posedwithin parallel machines which could
`bi Nae tt lied in anapplication-specific mannerto achieve
`
`OSSt Performance(52, 53, 32).
`
`
`bit 1030 - 20
`
`
`Intel Exhibit 1030 - 20
`
`
`
`eoIntegrated Circuits, ch. 8, bit 1030 - 21
`
`Intel Exhibit 1030 - 21
`
`
`
`We Ha net PP522-545; August 1993.
`
`-
`Hall, K, Kennedy, and K,8. McKinley, “In-
`parallel code gener-
`fesewi Augisalolereneoy
`ov. 199] ing (Supercomputing
`¢
`,
`Burke, and P. Carini,
`“Efficient
`srocedural com tation ofpointer-
`sidearea in amenee ey
`SIGPLAN
`foam
`$ of Programming
`Languages,
`
`
`
`
`bit 1030 - 22
`
`
`Intel Exhibit 1030 - 22
`
`
`
`
`
`Intel Exhibit 1030 - 23
`
`
`
`vogentun)(Red
`
`hansenYOsieuold
`oppVO"
`
`vowmigOdushgWXISOL
`
`aul
`
`Intel Exhibit 1030 - 24
`
`Intel Exhibit 1030 - 24
`
`
`
`Appendix G1002
`Appendix G1002
`
`Intel Exhibit 1030 - 25
`
`Intel Exhibit 1030 - 25
`
`
`
`Proceedings
`
`Frontiers ‘96
`
`The Sixth Symposium on the
`Frontiers of Massively Parallel Computing
`
`October 27-31, 1996
`Annapolis, Maryland
`
`Sponsored by
`IEEE Computer Society
`
`In cooperation with
`NASA Goddard Space Flight Center
`US RNCESD IS
`
`IEEE Computer Society Press
`Los Alamitos, California
`Washington
`Brussels
`
`Tokyo
`
`Intel Exhibit 1030 - 26
`
`
`
`IEEE Computer Society Press
`10662 Los Vaqueros Circle
`P.O.Box 301 4
`Los Alamitos, CA 90720-1 264
`
`Copyright 0 1996 by The Institute of Electrical and Electronics Engineers, Inc.
`All rights reserved.
`
`Copyright and Reprint Permissions: Abstracting is permitted with credit to the source. Libraries may
`photocopy beyond the limits of US copyright law, for private use of patrons, those articles in this volume
`that carry a code at the bottom of the first page, provided that the per-copy fee indicated in the code is paid
`through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923.
`
`Other copying, reprint, or republication requests should be addressed to: IEEE Copyrights Manager, IEEE
`Service Center, 445 Hoes Lane, P.O. Box 1331, Piscataway, NJ 088551331,
`
`The papers in this book comprise the proceedings of the meeting mentioned on the cover and title page. They
`reflect the authors’ opinions and, in the interests of timely dissemination, are published as presented and
`without change. Their inclusion in this publication does not necessarily constitute endorsement by the
`editors, the IEEE Computer Society Press, or the Institute of Electrical and Electronics Engineers, Inc.
`
`IEEE Computer Society Press Order Number PRO755 1
`IEEE Order Plan Catalog Number 96TB 100062
`ISBN 0-81 86-755 1-9
`Microfiche ISBN 0-81 86-7553-5
`ISSN 1088-4955
`
`Additional copies may be ordered from:
`
`IEEE Computer Society Press
`Customer Service Center
`10662 Los Vaqueros Circle
`P.O. Box 3014
`Los Alamitos, CA 90720-1314
`Tel: +1-714-821-8380
`Fax: +1-7 14-821 -4641
`Email: cs.books@computer.org
`
`IEEE Service Center
`445 Hoes Lane
`P.O. Box 1331
`Piscataway, NJ 08855-1331
`Tel: +1-908-98 1-1393
`Fax: +1-908-981-9667
`misc.custserv @ computer.org
`
`IEEE Computer Society
`13, Avenue de 1’Aquilon
`B-1200 Brussels
`BELGIUM
`Tel: +32-2-770-2198
`Fax: +32-2-770-8.505
`euro.ofc @ computr.org
`
`IEEE Computer Society
`Ooshima Building
`2-19-1 Minami-Aoyama
`Minato-ku, Tokyo 107
`JAPAN
`Tel: +81-3-3408-3118
`Fax: +81-3-3408-3553
`tokyo.ofc@ computer.org
`
`Editorial production by Penny Storms
`Cover by Kerry Bedford and Alex Torres
`Printed in the United States of America by KNI, Inc.
`
`The Institute of Electrical and Electronics Engineers, Inc.
`
`Intel Exhibit 1030 - 27
`
`
`
`Contents
`
`Message from the General Chair ............................................................................................
`ix
`Message from the Program Chair ...........................................................................................
`x
`Conference Committee .............................................................................................................
`xi
`...
`Referees ...................................................................................................................................... xu1
`
`Session 1: Invited Speaker
`From ASCI to Teraflops
`John Hopson, Accelerated Strategic Computing Initiative (ASCI)
`
`Session 2 A Scheduling 1
`Gang Scheduling for Highly Efficient Distributed Multiprocessor Systems ..............................
`H. Franke, P. Pattnaik, and L. Rudolph
`Integrating Polling, Interrupts, and Thread Management ........................................................
`K. Langendoen, J. Romein, R. Bhoedjang, and H. Bal
`A Practical Processor Design for Multithreading ....................................................................... 23
`M. Amamiya, T. Kawano, H. Tomiyasu, and S. Kusakabe
`
`13
`
`4
`
`Session 2B: Routing
`Analysis of Deadlock-Free Path-Based Wormhole Multicasting in
`Meshes in Case of Contentions .................................................................................................... 34
`E. Fleury and P. Fraigniaud
`Efficient Multicast in Wormhole-Routed 2D MesWTorus Multicomputers:
`A Network-Partitioning Approach ...............................................................................................
`S- Y. Wang, Y-C. Tseng, and C- W. Ho
`Turn Grouping for Efficient Multicast in Wormhole Mesh Networks ......................................
`K-P. Fan and C-T. King
`
`42
`
`50
`
`Session 3A. Applications and Algorithms
`A3: A Simple and Asymptotically Accurate Model for Parallel Computation ...........................
`A. Grama, V. Kumar, S. Ranka, and V. Singh
`Fault Tolerant Matrix Operations Using Checksum and Reverse Computation ....................
`Y. Kim, J.S. Plank, and J. J. Dongarra
`A Statistically-Based Multi-Algorithmic Approach for Load-Balancing
`Sparse Matrix Computations .......................................................................................................
`S. Nastea, T. El-Ghazawi, and 0. Frieder
`
`60
`
`.70
`
`78
`
`Session 3B: Petaflops Computing / Point Design Studies
`Pursuing a Petaflop: Point Designs for 100 TF Computers Using PIM Technologies ............. 88
`P.M. Kogge, S.C. Bass, J.B. Brockman, D.Z. Chen, and E. Sha
`Hybrid Technology Multithreaded Architecture .......................................................................
`G. Gao, K.K. Likharev, P.C. Messina, and T.L. Sterling
`The Illinois Aggressive Coma Multiprocessor Project (I-ACOMA) ..........................................
`J. Torrellas and D. Padua
`
`106
`
`.98
`
`V
`
`Intel Exhibit 1030 - 28
`
`
`
`Panel Session-How Do We Break the Barrier to the Software Frontier?
`Panel Chair: Rick Stevens, Argonne National Laboratory
`
`Session 4: Invited Speaker
`
`Session 5 A Scheduling 2
`Largest-Job-First-Scan-All Scheduling Policy for 2D Mesh-Connected Systems ................... 118
`S-M. Yo0 and H.Y. Youn
`Scheduling for Large-scale Parallel Video Servers ..................................................................
`M-Y. Wu and W. Shu
`Effect of Variation in Compile Time Costs on Scheduling Tasks on
`Distributed Memory Systems .................................................................................................... 134
`S. Darbha and S. Pande
`
`126
`
`Session 5B: SIMD
`Proce