1、45 1 2009 M1 2v(1 S)JOURNAL OF NANJING UNIVERSITY(NATURAL SCIENCES)Vol.45, No.1Jan.，20093 Tm) ARM11 MPCore ?s周 余,都思丹( 2v0 S“, 2, 210093)K 1: ARM11 MPCore K 3 T) ,.d 3 T q ?) ?.5,ARM11 MPCore ? A RM9) 1s,4 Pq_ #i9 qZE, LTV MPCore) ? ARM9“ ) ) ? 10P.O, Vi9 q, kTV , OpenMPi9 , Lr q43.8P. 3 T 8) ,4 Pq_
2、DCT , j 、 # j A Ui) ZE. LTV , ZE ?4 qARM11 MPCore r q,4 “d8 ?.1oM: ARM11, MPCore, VFP,i9 ms |: TP 311.11The analysis and optimization on performance of ARM11 MPCoreZhouY u , Du Si-Dan(Department of ElectronicScience and Engineering, Nanjing University, Nanjing, 210093, China)Abstract: With the wide
3、application of multimedia, network and digital signal process in the embedded field, therequirement of embedded processor s performance was improving.In order to improve the performance and reducethe power consumption, MPSOC (multi processor system on chip)was used to design the embedded system.TheA
4、RM11 MPCore was the newest embedded multi-core processor.The traditional embedded software could not runefficiently on this platform, because they only used one processor and did not use VFP module to optimize the floatpoint computation.In order to solve this problem, the basic computation performan
5、ce of ARM11 MPCore andARM9 was analyzed.The parallel computing method and hardware acceleration method were adopted to optimizethemulti-core performance for ARM11 M PCore processor.Compared with the single-core processor, the ARM11MPCore processor had excellent performance by adopting the optimizati
6、on methods in the experiment.The results ofthe experiment showed that thecapability could improve by 10 times when the VFP(vector float point)optimization “:SE1 S(60832003)l :2008-05-20Y “ ,E-mail:dsdese.nju.edu .cnmethod was used.In the other hand, the efficiency of the program could improve by 3.8
7、 times when theoptimization method of parallel computation was used.For the multimedia program the single core embeddedprocessor could not process the decoding of multimedia program, because of the poor capability of float pointcomputation.For the multi-core embedded processor, the traditional embed
8、ded software did not suit for the newarchitecture of M PSOC.In order to deal with this problem, two optimization methods werepresented.The onewasDCT optimization with VFP which could improve capability of the float point computation greatly.The other onewas parallelization of audio decoding, video d
9、ecoding and playing which could put the three threads into threedifferent processors.The experiments showed that the two optimization methods could improve the performance ofsoftware running on ARM11 MPCore platform.Key words: ARM11, MPCore, Vector float point, parallelization“/ ?Z, 3 T 59C !9Z,ARM1
10、1 MPCore 1 3 B.ARM11 MPCore ARM11$F,O V6 78“. 1,MPCore V$ 1 4) FZ T, K ? Vr2600MIPS .ARM9“ ) M1, ARM11MPCoreO,9F L2 Cache,F x) .d 3 T qBL, ARM11 MPCore H,.dLI T q4r q1pB.ARM11 MPCoreF q) ,4 6 3 T) 8) ? ,1 aZE qO q.D 4s ARM11 MPCore ?46W1“,4 ARM11MPCore ?“d ?4 6 H ?.D5 154 X86 i) ?ZE, 3 T .|ARM11 MPC
11、ore ? # 8) ? s,i)YV Px) ,+ y 7“i9 Z T qZE.1 ARM11 MPCore ARM11 MPCore ? p q , O m1 U,1+ /: 14ARM11) . ?i% “d; ) B) # 7cache V 16kbyte64kbyte;BSCU(snoop control unit), ) L2 Cache,i) L1 CacheWcache-to-cache. Z TB 64 bit AMBA 3 AXI9L V_ ) (VFP)K255qARM11ARM V6O,Thumb 7“,Java Z 7“,DSP Z7“#SIMD ISA Z 7“.
12、MPCorecB ?5 v(IEM), v ?1 Pqi qM.) 8 L , V s. V P_ ) i“E ,_) 3H L,3H LM ,i) VYV3H L.3H LsY :E# F(FMAC) L“E# Z(DS) L |i%(LS) L6 2v(1 S) 45 m1 ARM11 MPCoreOmFig.1 Architectureof ARM11 MPCore2 ? sARM11 MPCore ? s1V Z ,sY 9 ? # 9 ? .“ds H 3 TLinux T“d#GCCIr . L ARM9(S3C2410) ARM11 MPCore ?1s,S3C2410) ARM
13、11 MPCore ) q !200 MHz.“d H, ARM11 MPCoreqO+, i) ,q Z T, PARM11 MP-Core ? 4 6.2.1 9 ? ARM11 MPCore 9 ? s,1Linux/FE,E“E ? k,iS3C2410) 9 ? 1. kCI,B HW ?9 Q ,V7MOPS (millionoperations per second)T ?S.ARM11 MPCoreK V 4) , YI ? s B) ,T1 s 4) 9 ? ,1i9 q.OpenMPi) 9 ? k q , LT m2 U,V LT V Pi) q ? s qO 4 q ?
14、.V L V A,MPCoreL ?1S3C2410) 4 A,s ?4 61 MPCore L#i% “dO . TOpenMPi) , ?r4 6 q ?,MPCore OpenMP (4L)OpenMP ) ? 13.81.7 1 , :A RM11 M PCore ?sm2 9 ? sFig.2 Analysis and optimizationof integer computation2.2 9 ? ARM11 MPCore 9 ? s,“Linux/FE,E“E ? k,iS3C2410) 9 ? 1. 9 LCZE1 : q) ) , V q EZ T LC;q) ) ,BZE
15、 PI9 7“, 6BZE YVC LC,iIr |CJr9 7“,ZE q VMz, r q -B. 9 ? s L,S3C2410 q E 9 ,MPCore q E 9 #MPCore PVFP 1s, LT m3 U.MPCore PVFP LC ZE, PMPCoreVIr Ir H,F “ mfpu =vfp mfloat-abi=softfp”Ir V.V L V: Pq) ?MP-Core) ?43 5,“E ?4i A;MPCore) S3C2410M1,MPCore) VFP#O , P ) ? 10P4.3 8) ?.d 3 T“d) ? , !9 H PDSP 8m3
16、9 ? sFig.3 Analysis and optimization of float point com-putation) .ARM11 MPCore / i9F ) , P) 9 ? v4. ARM MPCoreO+, 8 q9 i9 , V P 3 T“d DSP f / 8 ilb.3.1 DCT j V,DCT9 Kh HWsB. DCT ?, ?4 r q.DCTr qs, Pq) MPCoreDCT , DCT ECooley-Tukey E,iS3C2410 q E) 1, T m4 U.V L V, q EZ T) H,MPCore) ? S3C24106.8P;MPC
17、ore ,VFPq q EZT ?1 3.6P.3.2 8i) .d 8lb qBL,MPCore4) , LZ T 8lb q ?4) ?. 8lb qi) H, L,BL j ,BL ,KBL j #A U,i| L) .MPEG4 Tq L,qsO q8 2v(1 S) 45 VGA(640480),25 / (fps), 210.7 kbps.LT V1 U. LTV LZTlb1Llb ? ?450%P,VGAsO q/,lbi , yMP-Core k A U i , TLCDe FF v,L 8lb ?4 6|F A.m4 DCT9 ? sFig.4 Analysis andop
18、timization of DCTV1 8lb qiTable1 Parallelization optimizationof multimedia pro-gramVGA QVGAL 8lb q4.2 fps 20.2fpsL 8lb q6 fps 25 fps4 ARM V6#O,9F L2 Cache#_ ) , ARM11 MP-Core ?ARM9“ ) M1, A4.1 s ARM11 MPCore ?,1 O q,q_ ) , V P9F 7“, LTV MPCore) ? ARM9“ ) ) ? 10P.O,Vi9 q, kTV , OpenMPi9 ( L), Lr q43.
19、8P.d 3 T) 8) ? , a 8) .ARM11 MPCore ?A4 6, P ) 8M V.ARM11 MPCore ) 8,91) i9 Z . _ ) DCT iL 8lb qi) , LTV , ZE ?4 8lb q ?.9 1 , :A RM11 M PCore ?sReferences 1 ARM(ARM 1I MPCore processor technical refer-encemanual)Limited.ARM DDI0360D, 2006. 2 ARM(core tilefor ARM 1I MPCoreuser guide)Limited.DUI 0318
20、C, 2006. 3 Wolf W.The future of multiprocessor systems-on-chips.Design Automation Conference, 41stConference on (DAC 04), 2004, 681 685. 4 Blume H, Livonius J V, Rotenberg L, et al .Performance and power analysis of parallelizedimplementations on an M PCore multiprocessorplatform.Embedded Computer S
21、ystems:Archi-tectures, Modeling and Simulation, 2007, 2007(s):74 81. 5 Marathe J, Mueller F.Source-code-correlatedcache coherence characterization of OpenMPbenchmarks.Parallel and Distributed Systems,IEEE Transactions, 2007, 18(6):818 834. 6 Dorta I, Leon C, Rodriguez C.A comparisonbetween M PI and
22、OpenMP branch-and-boundskeletons.Parallel and Distributed ProcessingSymposium, 2003.Proceedings International,2003(s):8. 7 Grant R E, Afsahi A.A comprehensive analy-sis of OpenMP applications on dual-core IntelXeon SM Ps.Parallel and Distributed ProcessingSymposium, 2007.IPDPS 2007, IEEE Interna-tio
23、nal, 2007(s):1 8. 8 Vaswani R, Zahorjan J.The implications ofcache affinity on processor scheduling for multi-programmed, shared memory multiprocessors.Proceedings of the thirteenth ACM symposiumon Operating systems, Pacific Grove, 1991,26 40. 9 Hirata K, Goodacre J.ARM MPCore:Thestreamlined and sca
24、lable ARM11 processorcore.Design Automation Conference, 2007.ASP-DAC 07.Asia and South Pacific January,2007(s):747 748. 10 Wang Y, Zhao X, Chen X Q.Device level dy-namicpower management for embedded system.Journal of Nanjing University (Natural Sci-ences), 2005, 41(4):595 601.( ,u),_ . 3 T“d !) 5 .2
25、v(1 S), 2005, 41(4):595 601). 11 Ouyang T, Teng Q M , Chen X Q.Devicedriv-er model for component-based embedded operat-ing system TICK.Journal of Nanjing University(Natural Sciences), 2005, 41(4):563 569.( xm, ,_ .q 3 T T“dTICK ! z . 2v(1 S), 2005, 41(4):563 569). 12 Wang W, Du S D.Research on inter
26、rupt affini-ty based on MPCoreand Linux.Journal of Nan-jing University (Natural Science), 2009, 1:24 30.( , .MPCoreLinux T. 2v(1 S), 2009, 1:24 30). 13 Wang W X, Yuan J, Zang J.A fingerprintmatching algorithm of minutia based on localcharacteristic.Journal of Nanjing University(Natural Sciences), 20
27、09, 1:18 23.(, , . + T E. 2v(1 S), 2009,1:18 23). 14 Xiao B, Shen Q H, Ding Y L.A digital videodecoder with high precision without using aphase locked loop.Journal of Nanjing Universi-ty(Natural Sciencess), 2009, 1:11 17.(o o, + i,.B M3 j Z. 2v(1 S), 2009, 1:11 17). 15 Zheng F, Shen Q H, Huang Y C.The researchand implementation on the digital decoding ofNTSC video signal.Journal of Nanjing Univer-sity(Natural Sciences), 2009, 1:31 38.(, + i,.NTSC j | 3Z LC. 2v(1 S), 2009, 1:31 38).10 2v(1 S) 45
