Unstructured Grid Structured Grid AMR based on Space ... · Surface capture:CLSVOF(THINC +...

11
Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology GP GPU GP GPU AMR based on Space-filling Curve for Stencil Applications AMR based on Space-filling Curve for Stencil Applications 1 Takayuki Aoki Global Scientific Information and Computing Center Tokyo Institute of Technology Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology GP GPU GP GPU Motivation Motivation 2 Prof. Nakahashi Prof. Yoshimura Unstructured Grid Structured Grid Coalesced Memory Access High accuracy Long stencil High Performance Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology GP GPU GP GPU Flows with sharp interfaces Flows with sharp interfaces Navier-Stokes solverFractional Step Time integration3rd TVD Runge-Kutta Advection term5th WENO Diffusion term4th FD PoissonMG-BiCGstab Surface tensionCSF model Surface captureCLSVOF(THINC + Level-Set) Particle Method ex. SPH Mesh Method (Surface Capture) Low accuracy < 10 6-7 particles High accuracy > 10 8-9 mesh points not splash Numerical noise and unphysical oscillation Gas Liquid Two-phase Flows

Transcript of Unstructured Grid Structured Grid AMR based on Space ... · Surface capture:CLSVOF(THINC +...

  • Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    AMR based on Space-filling Curvefor Stencil Applications

    AMR based on Space-filling Curvefor Stencil Applications

    1

    Takayuki Aoki

    Global Scientific Information and Computing CenterTokyo Institute of Technology

    Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUMotivationMotivation

    2

    Prof. Nakahashi

    Prof. Yoshimura

    Unstructured Grid Structured Grid

    Coalesced Memory Access High accuracy Long stencil

    High Performance

    Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUFlows with sharp interfacesFlows with sharp interfaces

    ■ Navier-Stokes solver:Fractional Step■ Time integration:3rd TVD Runge-Kutta■ Advection term:5th WENO■ Diffusion term:4th FD■ Poisson:MG-BiCGstab■ Surface tension:CSF model■ Surface capture:CLSVOF(THINC + Level-Set)

    Particle Methodex. SPH

    Mesh Method (Surface Capture)

    Low accuracy< 106-7 particles

    High accuracy > 108-9 mesh points

    not splash

    Numerical noise and unphysical oscillation

    Gas Liquid Two-phase Flows

  • A drop on the dry floor

    6

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUIndustrial Appl. Steering OilIndustrial Appl. Steering Oil

    7 Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUDevelopment New MaterialsDevelopment New Materials

    Material Microstructure

    Dendritic Growth

    Mechanical Structure

    Improvement of fuel efficiency by reducing the weight of transportation

    Developing lightweight strengthening material by controlling microstructure

    Low-carbon society

  • Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUInterface between Solid and LiquidInterface between Solid and Liquid

    Phase-field

    0

    1

    Phase A

    diffusive interfacewith finite thickness

    Phase B

    Mesh Adaption

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    11

    AMR(Adaptive Mesh Refinement)AMR(Adaptive Mesh Refinement)Octrees and Space Filling Curves

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    FDM & FVMFDM & FVM

    12

    FDM(Finite Difference Method)

    Node Center AMR

    FVM(Finite Volume Method)

    Cell Center AMR

  • Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    13

    Compressible Fluid SimulationCompressible Fluid Simulation

    Heavy fluid lays on light fluid and unstable.

    Euler equation:

    0

    yxtFEQ

    evu

    Q

    pueuuv

    puu

    2

    E

    pvevpv

    uvv

    2F

    IDO-CF Scheme512 x 512

    Y. Imai, T. Aoki and K. Takizawa, J. Comp. Phys., Vol. 227, Issue 4, 2263-2285 (2008)

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    IDO-CF schemefor Compressible CFD

    IDO-CF schemefor Compressible CFD

    ・Combination of FDM and FVM・Multi-Moment interpolation・Higher-order accuracy・Less dispersive and dissipative

    Y. Imai, T. Aoki and K. Takizawa, J. Comp. Phys., Vol. 227, Issue 4, 2263‐2285 (2008)

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUStencil of continuum eq.Stencil of continuum eq.

    PV XI

    YI XYI

    51 flop

    83 flop

    83 flop

    55 flop

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    16

    Directional Splitting (1)Directional Splitting (1)Euler equation:

    0

    yv

    xu

    t

    02

    yuv

    xpu

    tu

    02

    ypv

    xuv

    tv

    0

    ypvev

    xpueu

    te

    0

    xu

    t0

    2

    xpu

    tu

    0

    xuv

    tv 0

    xpueu

    te

    0

    yv

    t0

    yuv

    tu

    02

    ypv

    tv 0

    ypvev

    te

  • Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUDirectional Splitting (2)Directional Splitting (2)

    Available for explicit time integration of hyperbolic equations:

    0

    )(

    ,0

    )(

    ,0

    )(

    wpepww

    vwuww

    z

    ewvu

    t

    vpewv

    pvvuvv

    y

    ewvu

    t

    upewuvu

    puuu

    x

    ewvu

    t

    1***

    nn

    ewvu

    ewvu

    ewvu

    ewvu x-directionalintegration

    y-directionalintegration

    z-directionalintegration

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUStencil of continuum eq.Stencil of continuum eq.

    PV XI

    YI XYI

    0 fuuff xxt xufuff iit

    x

    )()( 1

    0 fufuf yxxy

    ty

    xufuff i

    yi

    y

    txy

    )()( 1

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUStencil of continuum eq.Stencil of continuum eq.

    PV XI

    YI XYI

    0 fvvff yyt

    yvfvf

    f jjty

    )()( 1y

    vfvff j

    xj

    x

    txy

    )()( 1

    0 fvfvf xyyx

    tx

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUStencil of continuum eq.Stencil of continuum eq.

    Direct Method Directional Splitting Method

  • Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUStencil of continuum eq.Stencil of continuum eq.

    Direct Method Directional Splitting Method

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUConservative Semi-LagrangianConservative Semi-LagrangianNumerical fluxes are determined by integrating interpolation function (Lagrangian Polynomials) 0

    xu

    t

    if

    if

    tt

    tudt

    if 1f i

    ix 1ixpx

    xx

    x

    Xi

    i

    idxxFf )(2/1

    z = ( 1.0/16.0*u[j‐2] ‐ 9.0f/16.0*u[j‐1] ‐ 9.0/16.0*u[j+1] + 1.0/16.0*u[j+2] )*dt/dx;zz = z*z;   zzz = zz*z;

    fn[j] = f[j] – 1.0/6.0*(zzz ‐ z)*f[j+2] + (1.0/3.0*zzz + 0.5*zz + 5.0/6.0*z)*f[j+1]+ (1.0/6.0*zzz ‐ 0.5*zz + 1.0/3.0*z)*f[j] + 1.0/6.0*(zzz ‐ z)*f[j‐1];

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUCPU, MIC, GPU PerformancesCPU, MIC, GPU Performances

    53.6 59.5 57.1

    457

    866958

    1218

    0

    200

    400

    600

    800

    1,000

    1,200

    1,400

    Perf

    orm

    ance

    [GFl

    ops]

    Xeon E5‐2600

    Xeon Phi3110P

    Xeon Phi5110P

    TeslaM2050

    TeslaK20C

    TeslaK20X

    GeForceGTX Titan

    Shared Memory Use in the x-directional kernel. Super function unit Loop unrolling Variable reuse in the y- and z- loops Reduction of branch diverges

    Kepler GPU Tuning

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    1 1 1 1 1

    1 1 132

    Some leaves extinct by coarseningf

    De-fragmentationDe-fragmentation

    Defragmentation by re-numbering

    1 1 1 1 1

    new leaves are generated by refinement.

    1 1 132 1 1

    GPU memory pool

  • Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    Space-Filling CurveSpace-Filling Curve

    Hilbert CurveHilbert Curve Morton Curve

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    Configuration 0 Configuration 1 Configuration 2 Configuration 3

    Configuration 4 Configuration 5 Configuration 6 Configuration 7

    Base of Hilbert CurveBase of Hilbert Curve

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPULeaf RefinementLeaf Refinement

    Configuration 0 Configuration 2

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPULeaf RefinementLeaf Refinement

    Configuration 5 Configuration 7

  • Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    Hilbert CurveHilbert Curve

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    Morton CurveMorton Curve

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    Domain DecompositionDomain Decomposition

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    Domain DecompositionDomain Decomposition

  • Level 0 Primitives Level 1 Refinements

    3D Mesh Refinement

    Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    : Level-Set function(distance function)

    : Heaviside function

    The Level-Set methods (LSM) use the signed distancefunction to capture the interface. The interface isrepresented by the zero-level set (zero-contour).

    Re-initialization for Level-Set function

    Fig. Takehiro Himeno, et. Al., JSME, 65-635,B(1999),pp2333-2340

    Level-Set method (LSM)Level-Set method (LSM)

    Advantage : Curvature calculation, Interface boundaryDrawback : Volume conservation

  • Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    Anti-diffusive Interface CaptureAnti-diffusive Interface Capture

    ・VOF(volume of fluid) type interface capturing method・Flux from tangent of hyperbola function・Semi-Lagrangian time integration

    [ Xiao, etal, Int. J. Numer. Meth. Fluid. 48(2005)1023 ]

    ・1D implementation can be applied to 2D & 3D → Simple

    ・Finite Volume like usage* THINC is the method how to compute flux

    → 3 krenel (x, y, z) can be fused to 1 kernel. Merit in memory R/W

    THINC (tangent of hyperbola for interface capturing) Scheme

    Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUThinc WLICThinc WLIC

    WLIC – splitting the interface intospatial directions

    Weighting factors depending on the normalDirections.

    : nomal vector to the interface

    Calculation of the normal vector

    Copyright © Takayuki Aoki / Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    39

    Advection of VOFAdvection of VOFNumerical Diffusion of Advection Computationsfor VOF (Volume of Fluid) depending on mesh resolutions.

    128×128 256×256 ???×???

    Thinc WLIC Scheme for Anti-diffusion:

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUOctree-based GPU-AMROctree-based GPU-AMR

    GPUleaf

    Octreedata structure

    Hilbert Space-filling Curve

    Interface Adaptation

    高解像度が必要な界面に動的に細かい格子を集め、計算領域全域を細かくした場合の数%の格子点数で効率的に計算する。

    leaf memory is managed on CPU

  • Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPURayleigh-Taylor InstabilityRayleigh-Taylor Instability

    41

    Density Profile Level Set Function

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPURayleigh-Taylor InstabilityRayleigh-Taylor Instability

    42

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPUOctree-based GPU-AMROctree-based GPU-AMR

    Interface Adaptation

    Copyright © Global Scientific Information and Computing Center, Tokyo Institute of Technology

    GP GPUGP GPU

    Thank youThank you

    44