Applications for Java Accelerator on Cell Processors Jan Trska - - PowerPoint PPT Presentation

applications for java accelerator on cell processors jan
SMART_READER_LITE
LIVE PREVIEW

Applications for Java Accelerator on Cell Processors Jan Trska - - PowerPoint PPT Presentation

IBM - CVUT Student Research Projects Applications for Java Accelerator on Cell Processors Jan Trska (trskajan@fel.cvut.cz) Applications for JCBE Goal of this project is to implement alghoritms solving NP-complete problems and accelerate


slide-1
SLIDE 1

IBM - CVUT Student Research Projects Applications for Java Accelerator on Cell Processors

Jan Trska (trskajan@fel.cvut.cz)

slide-2
SLIDE 2

IBM - CVUT Student Research Projects

Applications for JCBE

  • Goal of this project is to implement

alghoritms solving NP-complete problems and accelerate them using JCBE

  • Running on Cell Broadband Engine

processors (Playstation 3)

  • Java programs runs just on PPE
  • Framework utilizes SPE (C code can be

run via JNI on any SPU or PPU)

slide-3
SLIDE 3

IBM - CVUT Student Research Projects

JCBE

public class MyClass extends Cell { @CellSPU public native int add(int a, int b); /*{ return a + b; }*/ public static void main(String[] args) { int x = 5; int y = 3; Class c = new MyClass(); c.add(x, y); } }

body of method in C declaration of accelerated method calling method special annotation

Instances of this class runs separately on SPUs in parallel

slide-4
SLIDE 4

IBM - CVUT Student Research Projects

Solved problems

Knapsack problem brute-force algorithm Travelling salesman problem simulated annealing

slide-5
SLIDE 5

IBM - CVUT Student Research Projects

Results of tests (knapsack p.)

PPU Java PPU C SPU C SPU vect 6x SPU vect 50 100 150 200 250 300 350 27 předmětů 30 předmětů

30 items: 321.1s 325.8s 257.1s 48.9s 8.08s

  • version optimised for SPU running on all 6 SPU is 39,7 times faster

than pure Java version

slide-6
SLIDE 6

IBM - CVUT Student Research Projects

Results

Algorithms succesfuly implemented JCBE can be quite easily used to accelerate Java applications Optimized algorithms for SPU can significantly boost performance