Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

OpenFOAM-2.3.0の 全チュートアリアル・ベンチマーク

3,299 views

Published on

2014年9月20日
第32回オープンCAE勉強会@関西

Published in: Engineering
  • Be the first to comment

OpenFOAM-2.3.0の 全チュートアリアル・ベンチマーク

  1. 1. OpenFOAM-2.3.0の 全チュートアリアル・ベンチマーク 今野 雅(株式会社OCAEL) OpenCAE Local Workshop@Kansai 20th Sep 2014 1 Masashi Imano (OCAEL Co. Ltd.)
  2. 2. OpenFOAMのベンチマークテスト ✓コンパイラ種別(intel/gnu)やコンパイラ最適化オプションを変えて,で きるだけ多くのケースでベンチマークテストを行い,可能ならば性能を最 適化したい. ✓全チュートリアルケースを対象にベンチマークテストを行うのが理想的. ✓全実行(Allrun)は時間かかりすぎ. ✓テスト実行(Alltest)は1ステップ実行のため,時間比較の精度が悪い. ✓24時間以内で終了させるため,Alltestを500ステップ実行するように改 造して,ベンチマークテストを行い,計算時間の比較を行なった. ✓ただし,実行時間が莫大な foamyHexMesh/mixerVessel や,実行に 失敗するケースは除いた. OpenCAE Local Workshop@Kansai 20th Sep 2014 2 Masashi Imano (OCAEL Co. Ltd.)
  3. 3. OpenFOAMのビルド ✓対象システム • Aシステム (Intel Xeon L5640, 2.26GHz) ベクトル化: SSE4.2 • Dシステム (Intel Xeon E5-2670 v2, 2.5GHz) ベクトル化: AVX ✓intelコンパイラ (Ver 14.0.0) • boost: 標準設定の1_55_0はビルド失敗 → 1_54_0 • 最適化オプション: -O3はfoamyHexMeshが動かない → -O2 ✓gnuコンパイラ • Ver: 4.8.2はCGALやfoamyHexMeshのビルド失敗 → 4.8.3 • 最適化オプション: -O3に-msse4.2を付けても0.1%の違い→ -O3 OpenCAE Local Workshop@Kansai 20th Sep 2014 3 Masashi Imano (OCAEL Co. Ltd.)
  4. 4. intelコンパイラの最適化オプション検討 ✓対象システム • A システム (Intel Xeon L5640 , 2.26GHz) ~SSE4.2 • D システム (Intel Xeon E5-2670 v2, 2.5 GHz) ~SSE4.2, AVX ✓ ケース I1 • 最適化オプション: -axAVX,SSE4.2 -O2 -no-prec-div • 全てのシステムで動作.DシステムでAVXが有効(-axオプションで切替) ✓ケース I2 • 最適化オプション: -xSSE4.2 -O2 -no-prec-div • 全てのシステムで動作.DシステムでAVXが有効ではない ✓ケース I3 • 最適化オプション: -xHost -O2 -no-prec-div (OpenFOAMデフォルト) • Aシステムでは動作しない.DシステムではAVXが有効. OpenCAE Local Workshop@Kansai 20th Sep 2014 4 Masashi Imano (OCAEL Co. Ltd.)
  5. 5. 7200 5400 3600 1800 0 D system (Intel Xeon E5-2670 v2, 2.5GHz) 1.015 x 0 1800 3600 5400 7200 計算時間の比較 (Dシステム) CPU time [s] (Case I3: Intel -xHost) CPU time [s] (Case I1: Intel -axAVX,SSE4.2) 7200 5400 3600 1800 0 D system (Intel Xeon E5-2670 v2, 2.5GHz) 1.020 x 0 1800 3600 5400 7200 CPU time [s] (Case I2: Intel -xSSE4.2) CPU time [s] (Case I1: Intel -axAVX,SSE4.2) intelコンパイラでベクトル化オプションを変えても計算時間はほぼ変らない 5 OpenCAE Local Workshop@Kansai 20th Sep 2014 Masashi Imano (OCAEL Co. Ltd.)
  6. 6. 計算時間の比較 (Aシステム) 7200 5400 3600 1800 0 A system (Intel Xeon L5640, 2.26GHz) 0.995 x 0 1800 3600 5400 7200 CPU time [s] (Case I2: Intel -xSSE4.2) CPU time [s] (Case I1: Intel -axAVX,SSE4.2) intelコンパイラでベクトル化オプションを変えても計算時間はほぼ変らない 6 OpenCAE Local Workshop@Kansai 20th Sep 2014 Masashi Imano (OCAEL Co. Ltd.)
  7. 7. intel vs gnu 7200 5400 3600 1800 0 D system (Intel Xeon E5-2670 v2, 2.5GHz) 1.147 x 0 1800 3600 5400 7200 CPU time [s] (Case G2: gnu -O3) CPU time [s] (Case I1: Intel -axAVX,SSE4.2) 7200 5400 3600 1800 0 A system (Intel Xeon L5640, 2.26GHz) 1.154 x 0 1800 3600 5400 7200 CPU time [s] (Case G2: gnu -O3) CPU time [s] (Case I1: Intel -axAVX,SSE4.2) intel版に対するgnu版の計算時間の回帰式の係数は1.15で15%の増加 OpenCAE Local Workshop@Kansai 20th Sep 2014 7 Masashi Imano (OCAEL Co. Ltd.)
  8. 8. OpenCAE Local Workshop@Kansai 20th Sep 2014 8 Masashi Imano (OCAEL Co. Ltd.) 計算時間の比(Dシステム, intelコンパイラ) 0.85 0.9 0.95 1 1.05 1.1 1.15 icoFoam/elbow icoFoam/cavityHighRe icoFoam/cavityGrade icoFoam/cavityFine icoFoam/cavityClipped icoFoam/cavity foamyHexMesh/simpleShapes foamyHexMesh/flange foamyHexMesh/blob fireFoam/les/smallPoolFire3D fireFoam/les/smallPoolFire2D fireFoam/les/oppositeBurningPanels fireFoam/les/flameSpreadWaterSuppressionPanel financialFoam/europeanCall electrostaticFoam/chargedWire dnsFoam/boxTurb16 compressibleMultiphaseInterFoam/laminar/damBreak4phase coalChemistryFoam/simplifiedSiwek chtMultiRegionSimpleFoam/multiRegionHeaterRadiation chtMultiRegionSimpleFoam/heatExchanger chtMultiRegionFoam/snappyMultiRegionHeater chtMultiRegionFoam/multiRegionHeater chemFoam/nc7h16 chemFoam/ic8h18 chemFoam/gri buoyantSimpleFoam/hotRadiationRoomFvDOM buoyantSimpleFoam/hotRadiationRoom buoyantSimpleFoam/externalCoupledCavity buoyantSimpleFoam/circuitBoardCooling buoyantSimpleFoam/buoyantCavity buoyantPimpleFoam/hotRoom buoyantBoussinesqSimpleFoam/iglooWithFridges buoyantBoussinesqSimpleFoam/hotRoom buoyantBoussinesqPimpleFoam/hotRoom boundaryFoam/boundaryWallFunctions boundaryFoam/boundaryLaunderSharma adjointShapeOptimizationFoam/pitzDaily XiFoam/ras/moriyoshiHomogeneousPart2 XiFoam/ras/moriyoshiHomogeneousHydrogen XiFoam/ras/moriyoshiHomogeneous SRFSimpleFoam/mixer SRFPimpleFoam/rotor2D PDRFoam/flamePropagationWithObstacles MRFMultiphaseInterFoam/mixerVessel2D MRFInterFoam/mixerVessel2D MPPICFoam/injectionChannel MPPICFoam/cyclone MPPICFoam/column MPPICFoam/Goldschmidt LTSReactingParcelFoam/verticalChannel LTSReactingParcelFoam/counterFlowFlame2D LTSReactingFoam/ras/counterFlowFlame2D DPMFoam/Goldschmidt CPU time ratio FOCUS D system (Intel Xeon E5-2670 v2, 2.5GHz) I2/I1, average:1.017 I3/I1, average:1.004 I1 Better I1 Worse I1 > I2,I1 ≒ I3 I1:-axAVX,SSE4.2 I2:-xSSE4.2 I3:-xHost
  9. 9. OpenCAE Local Workshop@Kansai 20th Sep 2014 9 Masashi Imano (OCAEL Co. Ltd.) 0.85 0.9 0.95 1 1.05 1.1 1.15 potentialFoam/pitzDaily potentialFoam/cylinder porousSimpleFoam/straightDuctImplicit porousSimpleFoam/angledDuctImplicit porousSimpleFoam/angledDuctExplicit pisoFoam/ras/cavityCoupledU pisoFoam/ras/cavity pisoFoam/les/pitzDailyMapped pisoFoam/les/pitzDaily pimpleFoam/pitzDaily pimpleFoam/elipsekkLOmega pimpleFoam/channel395 pimpleFoam/TJunctionFan pimpleFoam/TJunction pimpleDyMFoam/wingMotion/wingMotion2D_pimpleDyMFoam pimpleDyMFoam/oscillatingInletACMI2D pimpleDyMFoam/movingCone pimpleDyMFoam/mixerVesselAMI2D nonNewtonianIcoFoam/offsetCylinder multiphaseInterFoam/laminar/damBreak4phaseFine multiphaseInterFoam/laminar/damBreak4phase multiphaseEulerFoam/mixerVessel2D multiphaseEulerFoam/damBreak4phaseFine multiphaseEulerFoam/damBreak4phase multiphaseEulerFoam/bubbleColumn moveDynamicMesh/SnakeRiverCanyon laplacianFoam/flange interPhaseChangeFoam/cavitatingBullet interPhaseChangeDyMFoam/propeller interMixingFoam/laminar/damBreak interFoam/ras/weirOverflow interFoam/ras/damBreakPorousBaffle interFoam/ras/damBreakFine interFoam/ras/damBreak interFoam/les/nozzleFlow2D interFoam/laminar/damBreakFine interFoam/laminar/damBreak interFoam/laminar/capillaryRise interDyMFoam/ras/testTubeMixer interDyMFoam/ras/sloshingTank3D interDyMFoam/ras/sloshingTank3D6DoF interDyMFoam/ras/sloshingTank3D3DoF interDyMFoam/ras/sloshingTank2D interDyMFoam/ras/sloshingTank2D3DoF interDyMFoam/ras/mixerVesselAMI interDyMFoam/ras/floatingObject interDyMFoam/ras/damBreakWithObstacle icoUncoupledKinematicParcelFoam/hopper/hopperInitialState icoUncoupledKinematicParcelFoam/hopper/hopperEmptying CPU time ratio FOCUS D system (Intel Xeon E5-2670 v2, 2.5GHz) I2/I1, average:1.017 I3/I1, average:1.004 I1:-axAVX,SSE4.2 I2:-xSSE4.2 I3:-xHost I1 > I2,I1 ≒ I3 I1 Better I1 Worse 計算時間の比(Dシステム, intelコンパイラ)
  10. 10. OpenCAE Local Workshop@Kansai 20th Sep 2014 10 Masashi Imano (OCAEL Co. Ltd.) 0.85 0.9 0.95 1 1.05 1.1 1.15 twoPhaseEulerFoam/laminar/mixerVessel2D twoPhaseEulerFoam/laminar/fluidisedBed twoPhaseEulerFoam/laminar/bubbleColumnIATE twoPhaseEulerFoam/laminar/bubbleColumn twoPhaseEulerFoam/RAS/fluidisedBed twoPhaseEulerFoam/RAS/bubbleColumn twoPhaseEulerFoam/LES/bubbleColumn twoLiquidMixingFoam/lockExchange sprayFoam/aachenBomb sonicLiquidFoam/decompressionTankFine sonicLiquidFoam/decompressionTank sonicFoam/ras/prism sonicFoam/laminar/shockTube sonicFoam/laminar/forwardStep solidEquilibriumDisplacementFoam/beamEndLoad solidDisplacementFoam/plateHole simpleReactingParcelFoam/verticalChannel simpleFoam/turbineSiting simpleFoam/pitzDailyExptInlet simpleFoam/pitzDaily simpleFoam/pipeCyclic simpleFoam/mixerVessel2D simpleFoam/airFoil2D shallowWaterFoam/squareBump settlingFoam/ras/tank3D settlingFoam/ras/dahl rhoSimplecFoam/squareBend rhoSimpleFoam/angledDuctExplicitFixedCoeff rhoPorousSimpleFoam/angledDuctImplicit rhoPorousSimpleFoam/angledDuctExplicit rhoPimplecFoam/angledDuct rhoPimpleFoam/ras/mixerVessel2D rhoPimpleFoam/ras/cavity rhoPimpleFoam/ras/angledDuct rhoPimpleFoam/les/pitzDaily rhoPimpleDyMFoam/annularThermalMixer rhoLTSPimpleFoam/angledDuct rhoCentralFoam/wedge15Ma5 rhoCentralFoam/shockTube rhoCentralFoam/obliqueShock rhoCentralFoam/forwardStep rhoCentralFoam/LadenburgJet60psi reactingParcelFoam/verticalChannel reactingParcelFoam/parcelInBox reactingParcelFoam/filter reactingParcelFilmFoam/splashPanel reactingParcelFilmFoam/rivuletPanel reactingParcelFilmFoam/hotBoxes reactingParcelFilmFoam/cylinder reactingFoam/ras/counterFlowFlame2D CPU time ratio FOCUS D system (Intel Xeon E5-2670 v2, 2.5GHz) I2/I1, average:1.017 I3/I1, average:1.004 I1:-axAVX,SSE4.2 I2:-xSSE4.2 I3:-xHost I1 > I2,I1 ≒ I3 I1 Better I1 Worse 計算時間の比(Dシステム, intelコンパイラ)
  11. 11. OpenCAE Local Workshop@Kansai 20th Sep 2014 11 Masashi Imano (OCAEL Co. Ltd.) 0.85 0.9 0.95 1 1.05 1.1 1.15 icoFoam/elbow icoFoam/cavityHighRe icoFoam/cavityGrade icoFoam/cavityFine icoFoam/cavityClipped icoFoam/cavity foamyHexMesh/simpleShapes foamyHexMesh/flange foamyHexMesh/blob fireFoam/les/smallPoolFire3D fireFoam/les/smallPoolFire2D fireFoam/les/oppositeBurningPanels fireFoam/les/flameSpreadWaterSuppressionPanel financialFoam/europeanCall electrostaticFoam/chargedWire dnsFoam/boxTurb16 compressibleMultiphaseInterFoam/laminar/damBreak4phase coalChemistryFoam/simplifiedSiwek chtMultiRegionSimpleFoam/multiRegionHeaterRadiation chtMultiRegionSimpleFoam/heatExchanger chtMultiRegionFoam/snappyMultiRegionHeater chtMultiRegionFoam/multiRegionHeater chemFoam/nc7h16 chemFoam/ic8h18 chemFoam/gri buoyantSimpleFoam/hotRadiationRoomFvDOM buoyantSimpleFoam/hotRadiationRoom buoyantSimpleFoam/externalCoupledCavity buoyantSimpleFoam/circuitBoardCooling buoyantSimpleFoam/buoyantCavity buoyantPimpleFoam/hotRoom buoyantBoussinesqSimpleFoam/iglooWithFridges buoyantBoussinesqSimpleFoam/hotRoom buoyantBoussinesqPimpleFoam/hotRoom boundaryFoam/boundaryWallFunctions boundaryFoam/boundaryLaunderSharma adjointShapeOptimizationFoam/pitzDaily XiFoam/ras/moriyoshiHomogeneousPart2 XiFoam/ras/moriyoshiHomogeneousHydrogen XiFoam/ras/moriyoshiHomogeneous SRFSimpleFoam/mixer SRFPimpleFoam/rotor2D PDRFoam/flamePropagationWithObstacles MRFMultiphaseInterFoam/mixerVessel2D MRFInterFoam/mixerVessel2D MPPICFoam/injectionChannel MPPICFoam/cyclone MPPICFoam/column MPPICFoam/Goldschmidt LTSReactingParcelFoam/verticalChannel LTSReactingParcelFoam/counterFlowFlame2D LTSReactingFoam/ras/counterFlowFlame2D DPMFoam/Goldschmidt CPU time ratio FOCUS A system (Intel Xeon L5640, 2.26GHz) I2/I1, average:0.991 I1:-axAVX,SSE4.2 I2:-xSSE4.2 I1 ≒ I2 I1 Better I1 Worse 計算時間の比(Aシステム, intelコンパイラ)
  12. 12. OpenCAE Local Workshop@Kansai 20th Sep 2014 12 Masashi Imano (OCAEL Co. Ltd.) 0.85 0.9 0.95 1 1.05 1.1 1.15 potentialFoam/pitzDaily potentialFoam/cylinder porousSimpleFoam/straightDuctImplicit porousSimpleFoam/angledDuctImplicit porousSimpleFoam/angledDuctExplicit pisoFoam/ras/cavityCoupledU pisoFoam/ras/cavity pisoFoam/les/pitzDailyMapped pisoFoam/les/pitzDaily pimpleFoam/pitzDaily pimpleFoam/elipsekkLOmega pimpleFoam/channel395 pimpleFoam/TJunctionFan pimpleFoam/TJunction pimpleDyMFoam/wingMotion/wingMotion2D_pimpleDyMFoam pimpleDyMFoam/oscillatingInletACMI2D pimpleDyMFoam/movingCone pimpleDyMFoam/mixerVesselAMI2D nonNewtonianIcoFoam/offsetCylinder multiphaseInterFoam/laminar/damBreak4phaseFine multiphaseInterFoam/laminar/damBreak4phase multiphaseEulerFoam/mixerVessel2D multiphaseEulerFoam/damBreak4phaseFine multiphaseEulerFoam/damBreak4phase multiphaseEulerFoam/bubbleColumn moveDynamicMesh/SnakeRiverCanyon laplacianFoam/flange interPhaseChangeFoam/cavitatingBullet interPhaseChangeDyMFoam/propeller interMixingFoam/laminar/damBreak interFoam/ras/weirOverflow interFoam/ras/damBreakPorousBaffle interFoam/ras/damBreakFine interFoam/ras/damBreak interFoam/les/nozzleFlow2D interFoam/laminar/damBreakFine interFoam/laminar/damBreak interFoam/laminar/capillaryRise interDyMFoam/ras/testTubeMixer interDyMFoam/ras/sloshingTank3D interDyMFoam/ras/sloshingTank3D6DoF interDyMFoam/ras/sloshingTank3D3DoF interDyMFoam/ras/sloshingTank2D interDyMFoam/ras/sloshingTank2D3DoF interDyMFoam/ras/mixerVesselAMI interDyMFoam/ras/floatingObject interDyMFoam/ras/damBreakWithObstacle icoUncoupledKinematicParcelFoam/hopper/hopperInitialState icoUncoupledKinematicParcelFoam/hopper/hopperEmptying CPU time ratio FOCUS A system (Intel Xeon L5640, 2.26GHz) I2/I1, average:0.991 I1 Better I1 Worse I1:-axAVX,SSE4.2 I2:-xSSE4.2 I1 ≒ I2 計算時間の比(Aシステム, intelコンパイラ)
  13. 13. OpenCAE Local Workshop@Kansai 20th Sep 2014 13 Masashi Imano (OCAEL Co. Ltd.) 0.85 0.9 0.95 1 1.05 1.1 1.15 twoPhaseEulerFoam/laminar/mixerVessel2D twoPhaseEulerFoam/laminar/fluidisedBed twoPhaseEulerFoam/laminar/bubbleColumnIATE twoPhaseEulerFoam/laminar/bubbleColumn twoPhaseEulerFoam/RAS/fluidisedBed twoPhaseEulerFoam/RAS/bubbleColumn twoPhaseEulerFoam/LES/bubbleColumn twoLiquidMixingFoam/lockExchange sprayFoam/aachenBomb sonicLiquidFoam/decompressionTankFine sonicLiquidFoam/decompressionTank sonicFoam/ras/prism sonicFoam/laminar/shockTube sonicFoam/laminar/forwardStep solidEquilibriumDisplacementFoam/beamEndLoad solidDisplacementFoam/plateHole simpleReactingParcelFoam/verticalChannel simpleFoam/turbineSiting simpleFoam/pitzDailyExptInlet simpleFoam/pitzDaily simpleFoam/pipeCyclic simpleFoam/mixerVessel2D simpleFoam/airFoil2D shallowWaterFoam/squareBump settlingFoam/ras/tank3D settlingFoam/ras/dahl rhoSimplecFoam/squareBend rhoSimpleFoam/angledDuctExplicitFixedCoeff rhoPorousSimpleFoam/angledDuctImplicit rhoPorousSimpleFoam/angledDuctExplicit rhoPimplecFoam/angledDuct rhoPimpleFoam/ras/mixerVessel2D rhoPimpleFoam/ras/cavity rhoPimpleFoam/ras/angledDuct rhoPimpleFoam/les/pitzDaily rhoPimpleDyMFoam/annularThermalMixer rhoLTSPimpleFoam/angledDuct rhoCentralFoam/wedge15Ma5 rhoCentralFoam/shockTube rhoCentralFoam/obliqueShock rhoCentralFoam/forwardStep rhoCentralFoam/LadenburgJet60psi reactingParcelFoam/verticalChannel reactingParcelFoam/parcelInBox reactingParcelFoam/filter reactingParcelFilmFoam/splashPanel reactingParcelFilmFoam/rivuletPanel reactingParcelFilmFoam/hotBoxes reactingParcelFilmFoam/cylinder reactingFoam/ras/counterFlowFlame2D CPU time ratio FOCUS A system (Intel Xeon L5640, 2.26GHz) I2/I1, average:0.991 I1 Better I1 Worse I1:-axAVX,SSE4.2 I2:-xSSE4.2 I1 ≒ I2 計算時間の比(Aシステム, intelコンパイラ)
  14. 14. OpenCAE Local Workshop@Kansai 20th Sep 2014 14 Masashi Imano (OCAEL Co. Ltd.) MPPICFoam/cyclone chemFoam/nc7h16 I1 Better icoFoam/elbow icoFoam/cavityHighRe icoFoam/cavityGrade icoFoam/cavityFine icoFoam/cavityClipped icoFoam/cavity foamyHexMesh/simpleShapes foamyHexMesh/flange foamyHexMesh/blob fireFoam/les/smallPoolFire3D fireFoam/les/smallPoolFire2D fireFoam/les/oppositeBurningPanels fireFoam/les/flameSpreadWaterSuppressionPanel financialFoam/europeanCall electrostaticFoam/chargedWire dnsFoam/boxTurb16 compressibleMultiphaseInterFoam/laminar/damBreak4phase coalChemistryFoam/simplifiedSiwek chtMultiRegionSimpleFoam/multiRegionHeaterRadiation chtMultiRegionSimpleFoam/heatExchanger chtMultiRegionFoam/snappyMultiRegionHeater chtMultiRegionFoam/multiRegionHeater chemFoam/nc7h16 chemFoam/ic8h18 chemFoam/gri buoyantSimpleFoam/hotRadiationRoomFvDOM buoyantSimpleFoam/hotRadiationRoom buoyantSimpleFoam/externalCoupledCavity buoyantSimpleFoam/circuitBoardCooling buoyantSimpleFoam/buoyantCavity buoyantPimpleFoam/hotRoom buoyantBoussinesqSimpleFoam/iglooWithFridges buoyantBoussinesqSimpleFoam/hotRoom buoyantBoussinesqPimpleFoam/hotRoom boundaryFoam/boundaryWallFunctions boundaryFoam/boundaryLaunderSharma adjointShapeOptimizationFoam/pitzDaily XiFoam/ras/moriyoshiHomogeneousPart2 XiFoam/ras/moriyoshiHomogeneousHydrogen XiFoam/ras/moriyoshiHomogeneous SRFSimpleFoam/mixer SRFPimpleFoam/rotor2D PDRFoam/flamePropagationWithObstacles MRFMultiphaseInterFoam/mixerVessel2D MRFInterFoam/mixerVessel2D MPPICFoam/injectionChannel MPPICFoam/cyclone MPPICFoam/column MPPICFoam/Goldschmidt LTSReactingParcelFoam/verticalChannel LTSReactingParcelFoam/counterFlowFlame2D LTSReactingFoam/ras/counterFlowFlame2D I1 Worse I1:intel compiler -axAVX,SSE4.2 G2:gnu compiler -O3 計算時間の比(Dシステム,intel vs gnu) 0.6 0.8 1 1.2 1.4 1.6 1.8 2 DPMFoam/Goldschmidt CPU time ratio FOCUS D system (Intel Xeon E5-2670 v2, 2.5GHz) G2/I1, average:1.066
  15. 15. OpenCAE Local Workshop@Kansai 20th Sep 2014 15 Masashi Imano (OCAEL Co. Ltd.) interDyMFoam/ras/mixerVesselAMI I1 Better potentialFoam/pitzDaily potentialFoam/cylinder porousSimpleFoam/straightDuctImplicit porousSimpleFoam/angledDuctImplicit porousSimpleFoam/angledDuctExplicit pisoFoam/ras/cavityCoupledU pisoFoam/ras/cavity pisoFoam/les/pitzDailyMapped pisoFoam/les/pitzDaily pimpleFoam/pitzDaily pimpleFoam/elipsekkLOmega pimpleFoam/channel395 pimpleFoam/TJunctionFan pimpleFoam/TJunction pimpleDyMFoam/wingMotion/wingMotion2D_pimpleDyMFoam pimpleDyMFoam/oscillatingInletACMI2D pimpleDyMFoam/movingCone pimpleDyMFoam/mixerVesselAMI2D nonNewtonianIcoFoam/offsetCylinder multiphaseInterFoam/laminar/damBreak4phaseFine multiphaseInterFoam/laminar/damBreak4phase multiphaseEulerFoam/mixerVessel2D multiphaseEulerFoam/damBreak4phaseFine multiphaseEulerFoam/damBreak4phase multiphaseEulerFoam/bubbleColumn moveDynamicMesh/SnakeRiverCanyon laplacianFoam/flange interPhaseChangeFoam/cavitatingBullet interPhaseChangeDyMFoam/propeller interMixingFoam/laminar/damBreak interFoam/ras/weirOverflow interFoam/ras/damBreakPorousBaffle interFoam/ras/damBreakFine interFoam/ras/damBreak interFoam/les/nozzleFlow2D interFoam/laminar/damBreakFine interFoam/laminar/damBreak interFoam/laminar/capillaryRise interDyMFoam/ras/testTubeMixer interDyMFoam/ras/sloshingTank3D interDyMFoam/ras/sloshingTank3D6DoF interDyMFoam/ras/sloshingTank3D3DoF interDyMFoam/ras/sloshingTank2D interDyMFoam/ras/sloshingTank2D3DoF interDyMFoam/ras/mixerVesselAMI interDyMFoam/ras/floatingObject interDyMFoam/ras/damBreakWithObstacle icoUncoupledKinematicParcelFoam/hopper/hopperInitialState I1 Worse I1:intel compiler -axAVX,SSE4.2 G2:gnu compiler -O3 計算時間の比(Dシステム,intel vs gnu) 0.6 0.8 1 1.2 1.4 1.6 1.8 2 icoUncoupledKinematicParcelFoam/hopper/hopperEmptying CPU time ratio FOCUS D system (Intel Xeon E5-2670 v2, 2.5GHz) G2/I1, average:1.066
  16. 16. OpenCAE Local Workshop@Kansai 20th Sep 2014 16 Masashi Imano (OCAEL Co. Ltd.) twoPhaseEulerFoam/laminar/mixerVessel2D twoPhaseEulerFoam/laminar/fluidisedBed twoPhaseEulerFoam/laminar/bubbleColumnIATE twoPhaseEulerFoam/laminar/bubbleColumn twoPhaseEulerFoam/RAS/fluidisedBed twoPhaseEulerFoam/RAS/bubbleColumn twoPhaseEulerFoam/LES/bubbleColumn twoLiquidMixingFoam/lockExchange sprayFoam/aachenBomb sonicLiquidFoam/decompressionTankFine sonicLiquidFoam/decompressionTank sonicFoam/ras/prism sonicFoam/laminar/shockTube sonicFoam/laminar/forwardStep solidEquilibriumDisplacementFoam/beamEndLoad solidDisplacementFoam/plateHole simpleReactingParcelFoam/verticalChannel simpleFoam/turbineSiting simpleFoam/pitzDailyExptInlet simpleFoam/pitzDaily simpleFoam/pipeCyclic simpleFoam/mixerVessel2D simpleFoam/airFoil2D shallowWaterFoam/squareBump settlingFoam/ras/tank3D settlingFoam/ras/dahl rhoSimplecFoam/squareBend rhoSimpleFoam/angledDuctExplicitFixedCoeff rhoPorousSimpleFoam/angledDuctImplicit rhoPorousSimpleFoam/angledDuctExplicit rhoPimplecFoam/angledDuct rhoPimpleFoam/ras/mixerVessel2D rhoPimpleFoam/ras/cavity rhoPimpleFoam/ras/angledDuct rhoPimpleFoam/les/pitzDaily rhoPimpleDyMFoam/annularThermalMixer rhoLTSPimpleFoam/angledDuct rhoCentralFoam/wedge15Ma5 rhoCentralFoam/shockTube rhoCentralFoam/obliqueShock rhoCentralFoam/forwardStep rhoCentralFoam/LadenburgJet60psi reactingParcelFoam/verticalChannel reactingParcelFoam/parcelInBox reactingParcelFoam/filter reactingParcelFilmFoam/splashPanel reactingParcelFilmFoam/rivuletPanel I1:intel compiler -axAVX,SSE4.2 G2:gnu compiler -O3 sprayFoam/aachenBomb I1 Better reactingParcelFilmFoam/hotBoxes reactingParcelFilmFoam/cylinder I1 Worse 計算時間の比(Dシステム,intel vs gnu) 0.6 0.8 1 1.2 1.4 1.6 1.8 2 reactingFoam/ras/counterFlowFlame2D CPU time ratio FOCUS D system (Intel Xeon E5-2670 v2, 2.5GHz) G2/I1, average:1.066 twoPhaseEulerFoam/laminar/bubbleColumnIATE
  17. 17. OpenCAE Local Workshop@Kansai 20th Sep 2014 17 Masashi Imano (OCAEL Co. Ltd.) MPPICFoam/cyclone I1 Better icoFoam/elbow icoFoam/cavityHighRe icoFoam/cavityGrade icoFoam/cavityFine icoFoam/cavityClipped icoFoam/cavity foamyHexMesh/simpleShapes foamyHexMesh/flange foamyHexMesh/blob fireFoam/les/smallPoolFire3D fireFoam/les/smallPoolFire2D fireFoam/les/oppositeBurningPanels fireFoam/les/flameSpreadWaterSuppressionPanel financialFoam/europeanCall electrostaticFoam/chargedWire dnsFoam/boxTurb16 compressibleMultiphaseInterFoam/laminar/damBreak4phase coalChemistryFoam/simplifiedSiwek chtMultiRegionSimpleFoam/multiRegionHeaterRadiation chtMultiRegionSimpleFoam/heatExchanger chtMultiRegionFoam/snappyMultiRegionHeater chtMultiRegionFoam/multiRegionHeater chemFoam/nc7h16 chemFoam/ic8h18 chemFoam/gri buoyantSimpleFoam/hotRadiationRoomFvDOM buoyantSimpleFoam/hotRadiationRoom buoyantSimpleFoam/externalCoupledCavity buoyantSimpleFoam/circuitBoardCooling buoyantSimpleFoam/buoyantCavity buoyantPimpleFoam/hotRoom buoyantBoussinesqSimpleFoam/iglooWithFridges buoyantBoussinesqSimpleFoam/hotRoom buoyantBoussinesqPimpleFoam/hotRoom boundaryFoam/boundaryWallFunctions boundaryFoam/boundaryLaunderSharma adjointShapeOptimizationFoam/pitzDaily XiFoam/ras/moriyoshiHomogeneousPart2 XiFoam/ras/moriyoshiHomogeneousHydrogen XiFoam/ras/moriyoshiHomogeneous SRFSimpleFoam/mixer SRFPimpleFoam/rotor2D PDRFoam/flamePropagationWithObstacles MRFMultiphaseInterFoam/mixerVessel2D MRFInterFoam/mixerVessel2D MPPICFoam/injectionChannel MPPICFoam/cyclone MPPICFoam/column MPPICFoam/Goldschmidt LTSReactingParcelFoam/verticalChannel LTSReactingParcelFoam/counterFlowFlame2D LTSReactingFoam/ras/counterFlowFlame2D I1 Worse I1:intel compiler -axAVX,SSE4.2 G2:gnu compiler -O3 計算時間の比(Aシステム,intel vs gnu) 0.6 0.8 1 1.2 1.4 1.6 1.8 2 DPMFoam/Goldschmidt CPU time ratio FOCUS A system (Intel Xeon L5640, 2.26GHz) G2/I1, average:1.038
  18. 18. OpenCAE Local Workshop@Kansai 20th Sep 2014 18 Masashi Imano (OCAEL Co. Ltd.) multiphaseInterFoam/ laminar/ damBreak4phaseFine I1 Better potentialFoam/pitzDaily potentialFoam/cylinder porousSimpleFoam/straightDuctImplicit porousSimpleFoam/angledDuctImplicit porousSimpleFoam/angledDuctExplicit pisoFoam/ras/cavityCoupledU pisoFoam/ras/cavity pisoFoam/les/pitzDailyMapped pisoFoam/les/pitzDaily pimpleFoam/pitzDaily pimpleFoam/elipsekkLOmega pimpleFoam/channel395 pimpleFoam/TJunctionFan pimpleFoam/TJunction pimpleDyMFoam/wingMotion/wingMotion2D_pimpleDyMFoam pimpleDyMFoam/oscillatingInletACMI2D pimpleDyMFoam/movingCone pimpleDyMFoam/mixerVesselAMI2D nonNewtonianIcoFoam/offsetCylinder multiphaseInterFoam/laminar/damBreak4phaseFine multiphaseInterFoam/laminar/damBreak4phase multiphaseEulerFoam/mixerVessel2D multiphaseEulerFoam/damBreak4phaseFine multiphaseEulerFoam/damBreak4phase multiphaseEulerFoam/bubbleColumn moveDynamicMesh/SnakeRiverCanyon laplacianFoam/flange interPhaseChangeFoam/cavitatingBullet interPhaseChangeDyMFoam/propeller interMixingFoam/laminar/damBreak interFoam/ras/weirOverflow interFoam/ras/damBreakPorousBaffle interFoam/ras/damBreakFine interFoam/ras/damBreak interFoam/les/nozzleFlow2D interFoam/laminar/damBreakFine interFoam/laminar/damBreak interFoam/laminar/capillaryRise interDyMFoam/ras/testTubeMixer interDyMFoam/ras/sloshingTank3D interDyMFoam/ras/sloshingTank3D6DoF interDyMFoam/ras/sloshingTank3D3DoF interDyMFoam/ras/sloshingTank2D interDyMFoam/ras/sloshingTank2D3DoF interDyMFoam/ras/mixerVesselAMI interDyMFoam/ras/floatingObject interDyMFoam/ras/damBreakWithObstacle icoUncoupledKinematicParcelFoam/hopper/hopperInitialState I1 Worse I1:intel compiler -axAVX,SSE4.2 G2:gnu compiler -O3 計算時間の比(Aシステム,intel vs gnu) 0.6 0.8 1 1.2 1.4 1.6 1.8 2 icoUncoupledKinematicParcelFoam/hopper/hopperEmptying CPU time ratio FOCUS A system (Intel Xeon L5640, 2.26GHz) G2/I1, average:1.038 multiphaseEulerFoam/ damBreak4phaseFine
  19. 19. OpenCAE Local Workshop@Kansai 20th Sep 2014 19 Masashi Imano (OCAEL Co. Ltd.) twoPhaseEulerFoam/laminar/mixerVessel2D twoPhaseEulerFoam/laminar/fluidisedBed twoPhaseEulerFoam/laminar/bubbleColumnIATE twoPhaseEulerFoam/laminar/bubbleColumn twoPhaseEulerFoam/RAS/fluidisedBed twoPhaseEulerFoam/RAS/bubbleColumn twoPhaseEulerFoam/LES/bubbleColumn twoLiquidMixingFoam/lockExchange sprayFoam/aachenBomb sonicLiquidFoam/decompressionTankFine sonicLiquidFoam/decompressionTank sonicFoam/ras/prism sonicFoam/laminar/shockTube sonicFoam/laminar/forwardStep solidEquilibriumDisplacementFoam/beamEndLoad solidDisplacementFoam/plateHole simpleReactingParcelFoam/verticalChannel simpleFoam/turbineSiting simpleFoam/pitzDailyExptInlet simpleFoam/pitzDaily simpleFoam/pipeCyclic simpleFoam/mixerVessel2D simpleFoam/airFoil2D shallowWaterFoam/squareBump settlingFoam/ras/tank3D settlingFoam/ras/dahl rhoSimplecFoam/squareBend rhoSimpleFoam/angledDuctExplicitFixedCoeff rhoPorousSimpleFoam/angledDuctImplicit rhoPorousSimpleFoam/angledDuctExplicit rhoPimplecFoam/angledDuct rhoPimpleFoam/ras/mixerVessel2D rhoPimpleFoam/ras/cavity rhoPimpleFoam/ras/angledDuct rhoPimpleFoam/les/pitzDaily rhoPimpleDyMFoam/annularThermalMixer rhoLTSPimpleFoam/angledDuct rhoCentralFoam/wedge15Ma5 rhoCentralFoam/shockTube rhoCentralFoam/obliqueShock rhoCentralFoam/forwardStep rhoCentralFoam/LadenburgJet60psi reactingParcelFoam/verticalChannel reactingParcelFoam/parcelInBox reactingParcelFoam/filter reactingParcelFilmFoam/splashPanel reactingParcelFilmFoam/rivuletPanel I1:intel compiler -axAVX,SSE4.2 G2:gnu compiler -O3 sprayFoam/aachenBomb I1 Better reactingParcelFilmFoam/hotBoxes reactingParcelFilmFoam/cylinder I1 Worse 計算時間の比(Aシステム,intel vs gnu) twoPhaseEulerFoam/laminar/bubbleColumnIATE 0.6 0.8 1 1.2 1.4 1.6 1.8 2 reactingFoam/ras/counterFlowFlame2D CPU time ratio FOCUS A system (Intel Xeon L5640, 2.26GHz) G2/I1, average:1.038
  20. 20. まとめ ✓intelコンパイラの最適化オプションとして-axAVX,SSE4.2を 用いることで,DシステムではAVXが有効,Aシステムでは SSE4.2が自動的に有効になるバイナリができる. ✓-axAVX,SSE4.2オプション版は,-xAVXや-xSSE4.2の版と 速度がほぼ同等である. ✓intel版に対するgnu版の計算時間の比の平均は,Dシステムで 約1.07,Aシステムで約1.04であり,平均するとgnu版の速度 低下は10%以下である.ただし,gnu版のほうが40%以上遅い ケースがA,D両システムとも5ケースあった. OpenCAE Local Workshop@Kansai 20th Sep 2014 20 Masashi Imano (OCAEL Co. Ltd.)

×