computer architecture simd ola flygt växjö university [email protected] +46 470 70 86 49

18
Computer Architecture Computer Architecture SIMD SIMD Ola Flygt Växjö University http://w3.msi.vxu.se/users/ofl/ [email protected] +46 470 70 86 49

Upload: jewel-mcbride

Post on 21-Jan-2016

222 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Computer Computer ArchitectureArchitecture

SIMDSIMDOla Flygt

Växjö Universityhttp://w3.msi.vxu.se/users/ofl/

[email protected]+46 470 70 86 49

Page 2: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Outline

Design spaceGranularityConnectivityProcessor complexityLocal Anotomy

Fine grained architecturesCourse grained architectures

CH01

Page 3: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

A typical SIMD system

Page 4: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

The development of SIMD systems

Page 5: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Design space

There are several factors to consider when designing a SIMD. The most important ones are:GranularityConnectivityComplexity of processing elementLocal autonomy

Page 6: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Granularitymapping data to PEs

Page 7: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

GranularityFine to coarse grain

Page 8: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Connectivity

Not all the types of connectivity for data parallel arch. are used in SIMD arch.

Page 9: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Performance of interconnection networks

There are several important parameters in an interconnection networkDiameter, the shortest path between the

two most distant nodesBandwidth, how many data exchanges

simultaneous in the networkBlocking, how many exchanges are

blocked

Page 10: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Performance of interconnection networks

N=128*128

1

1

256

28

14

14

1

N=128*128

1

16384

16384

28

14

16384

16384

Page 11: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Processor complexity

The design space for processor precisionSingle bit – fine grained, mainly for

image processingInteger – general computingFloating point – Coarse grain,

scientific computingFine grained -> more PEs

Page 12: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Local autonomy

Page 13: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Fine grained SIMDThe MPP system

Page 14: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Fine grained SIMDThe MPP processing

element

Page 15: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Fine grained SIMDThe MPP integrated circuit

Page 16: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Coarse grained SIMDThe CM5 processing

element

Page 17: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Coarse grained SIMDThe CM5 fat tree

connectivity

Page 18: Computer Architecture SIMD Ola Flygt Växjö University  Ola.Flygt@msi.vxu.se +46 470 70 86 49

Coarse grained SIMDThe CM5 system

parameters