Exploring, Programming, and Prototyping MP-SoCs in 24 hours

Description

Benchmark Appli cations  
Todor S tefanov and E d Deprettere
Leiden Embedded Rese arch Cen ter,
Leid en Ins titute of Advanced Computer S cience
Le iden U niversity, The Nethe rlands
Agenda
 Intr oduction to the c urrent benchmar k su ite
 Brief re porting on the experiences w ith the
benchmark appl ications
 Iuliana Bacivarov ( 10 m in)
 Todor Stef anov ( 10 min)
 Bastian Ristau (10 min)
 Discussion
Todor Stefanov – Map2M PSoC workshop 29.06.2009 2Benchmark Suite
 Common se t of benchmark applications is mi ssing in
the MP SoC community
 The issue was d iscussed at the M ap2 MPS oC wor king
meeting on November 27-28, 200 8 in Dusseldor f
 Man y application s are available BUT
 They are in sequential form
 The M PSoC community ne eds parallel application specifications
 Decision made at the meet ing
 To in itiate th e cre ation o f an application benchmark s uit for
MPS oC
 To c reate a com mon re pository where interested parties
can u pload a pplications f or t he ben chmark suite
 To m ake the be nchmark su ite available on Int ernet
Todor Stefanov – Map2M PSoC workshop 29.06.2009 3Current C ontent of B enchmark S uite
 The benc hmark s uit i s ava ilable o n Internet at
http://www.art ist- embedded.or g/art ist/Benchm arks.ht ml
 It cons ists of 8 appli cations contr ibuted by 3
universities
 U Leiden (Todor S tefanov)
 Motio n JPEG encoder
 JPEG 2000 encoder
 Sobel
 TU Ei ndhoven (Marc Geilen)
 MP3
 H.263 en coder
 H.263 ...

Sujets

• XMLHttpRequest
• Spectres
• Royal KPN
• Deng Yaping
• Spécification (informatique)
• Suite bornée

Informations

• Publié par: Naphong
• Nombre de lectures: 87
• Langue: English
• Poids de l'ouvrage: 1 Mo

Extrait

  Benchmark Applications   Todor Stefanov and Ed Deprettere Leiden Embedded Research Center, Leiden Institute of Advanced Computer Science Leiden University, The Netherlands

   Agenda Introduction to the current benchmark suite Brief reporting on the experiences with the benchmark applications  Iuliana Bacivarov (10 min)  Todor Stefanov (10 min)  Bastian Ristau (10 min) Discussion  oTod rtSfenavo  –aM2pPMoS Cowkrhspo 92.60.02902

  Benchmark Suite Common set of benchmark applications is missing in the MPSoC community  The issue was discussed at the Map2MPSoC working meeting on November 27-28, 2008 in Dusseldorf  Many applications are available BUT  They are in sequential form  The MPSoC community needs parallel application specifications Decision made at the meeting  To initiate the creation of an application benchmark suit for CoSPM  To create a common repository where interested parties can upload applications for the benchmark suite  To make the benchmark suite available on Internet  oTod rtSfenavo  –aM2pPMoS Cowkrhspo 92.60.02903

  Current Content of Benchmark Suite The benchmark suit is available on Internet at   http://www.artist-embedded.org/artist/Benchmarks.html It consists of 8 applications contributed by 3 universities  U   Leiden (Todor Stefanov)  Motion JPEG encoder  J S P ob E e G l   2000 encoder  TU Eindhoven (Marc Geilen)     3PM   H H . . 2 2 6 6 3 3     e d n e c c o o d de e r r  ETH Zurich (Iuliana Bacivarov)   M M J P P E E G G -2     d d e e c c o o d d e e r r  oTod rtSfenavo  –aM2pPMoS Cowkrhspo 92.60.02904

seY seY seY seY seY  seY eY sC C Features of the applications oSleb JPEG 2000 encoder Application Motion JPEG encoder format C fully func.format Yes C++ for YAPI XML for Daedalus C++ YAPI XML for Daedalus C++ YAPI XML for Daedalus XML for SDF3 XML for SDF3 XML for SDF3 XML and C for DOL XML and C for DOL Sequential SpecParallel SpecLicense fully func.MoC  Yes KPNCPL YesKPNCPL Yes KPNCPL YesKPNCPL Yes KPNCPL YesKPNCPL NoSDFGPL NoSDFGPL NoSDFGPL YesKPNETHZ YesKPNETHZ C C C C C 3PM H.263 encoder H.263 decoder MPEG 2 decoder MJPEG decoder …  lanoitcnuf ylluf era sceps lellarap lla toN  stamrof LMX tnereffid ni era sceps lellarap llA  )CoM( noitatupmoc fo sledom wolf-atad nwonk llew esu sceps lellarap llA  snoitacilppa )wolf-atad( gnimaerts era snoitacilppa llA  :snoitavresbo emoS  59002.60.92 pohskrow CoSPM2paM – vonafetS rodoT 

boSle Applications contributed by U Leiden    Sobel (edge detection)   only task-level parallelism   JPEG2000 (encoder)   mainly task-level parallelism   Motion JPEG (encoder)    task-level and data-level parallelism  oTod rtSfenavo  –aM2pPMoS Cowkrs MGEPJ oh p92.60.0290 JPEG2000 6

CMEMM P m CC n … PI CC m MCMC P m CC 1 … PI CC 1 MEM MC Many alternative platform instances can be constructed fast and easily by instantiating different type/number of components and setting their parameters.    Programmable processors    Hardware IP Cores    Memory Components:    Program, Data (on-chip and external) Memory (MEM)    Communication Memory (CM)    Point-to-point network    Crossbar switch    Shared bus with Round-Robin, Fixed Priority, or TDMA arbitration   Communication Components: MP-SoC platform considered    Processing Components: 79002.60.92 pohskrow CoSPM2paM – vonafetS rodoT      ezinorhcnys dna tcennoc ot woh selur denifed llew + stnenopmoc fo yrarbil    mroftalP stnenopmoc noitacinummoc  dna ,yromem ,gnissecorp neewteb ecafretni – )CC( rellortnoC noitacinummoC  

Parallel Specification in C++YAPI format  YAPI simulation environment is available at http://sourceforge.net/projects/y-api/ DE1_ ND_0 OP1 ED_0 IP I 1 P2 ND_1 OP O 1 P2 ED_2 IP1 ND_2 1 2   / v / o p i r d o   c m e a s i s n   ( N  ) D  { _1 3 4                  f   o  f r( o   r i ( n   t i   n  i t  =  j 2 = ;  2 i ; <  j = < M = ; N  i ; +  j + +   + )   ) { 5 if( j-2 == 0 ) 6 read( IP1, in_0 ); 7 if( j-3 >= 0 ) 8 read( IP2, in_0 ); 9 Transformer( in_0, out_0); 10 if( -j+N-1 >= 0 ) 11 write( OP1, out_0 ); 12 if( j-N == 0 ) { 13 write( OP2, out_0 ); 14 }// for j 15}// main  oTod rtSe CONTROL f E a X n ERDA CEUET IRWET vo  –aM2pPMoS Cowkrhspo 92.60.02908

 Parallel Spec in XML Daedalus format Daedalus design framework is available at http://daedalus.liacs.nl Network Topology Specification in XML  oTod rtSfenavo  –aM2pPMoS Cowkrhs ND_0 OP1 ED_0 po 92.60.0290 1_DE IP I 1 P2 ND_1 OP O 1 P2 ED_2 IP1 ND_2 9

Parallel Spec in XML Daedalus format Process Control Code Specification in XML (1) 019002.60.92 pohskrow CoSPM2paM – vonafetS rodoT  i*1 j*0 i 2 # # 0*N # 0*M - 20, -1*i # 0*j # 0*N # 1*M # 00, 0*i # 1*j # 0*N # 0*M - 2 1 0 0 j 0 N 0 M c - o 2 nst 1 0 0 0 0 1 - 2 0 sig 1 n 1 i 1- 1 0 0 2 £ i £ M, £ i £ M, - 20, 2 £ j £ N, -i # M0, j - 2 1 0 j - 2 1 0 j - 2 1 0 CONTROL DAER _DE1 IP2OP1 ND_0 OP1 ED_0 IP1 ND_1 OP2 ED_2 IP1 ND_2 EXECUTE 9 Transformer( in_0, out_0); 10 if( -j+N-1 >= 0 ) 11 write( OP1, out_0 ); ETIRW 12 if( j-N == 0 ) { 13 write( OP2, out_0 ); 14 }// for j 15}// main 1  // process ND_1 2void main( ) { 3 for( int i=2; i<=M; i++ ) 4 for( int j=2; j<=N; j++ ) { 5 if( j-2 == 0 ) 6 read( IP1, in_0 ); 7 if( j-3 >= 0 ) 8 read( IP2, in_0 );

Parallel Spec in XML Daedalus format Process Control Code Specification in XML (2)  oTod rtSfenavo  –aM2pPMoS Cowkrhso sig 1 n 1 i 1- 1 0 0  p92.60.0290 j 0 0 1 N 0 0 0 M 0 1 0 co - - n 2 0 2 1 s 1 t