Centre Blaise Pascal » Bench4GPU

#!/usr/bin/env python

#

# Ising2D model using PyOpenCL

#

# CC BY-NC-SA 2011 : <emmanuel.quemener@ens-lyon.fr>

#

# Thanks to Andreas Klockner for PyOpenCL:

# http://mathema.tician.de/software/pyopencl

#

# Interesting links:

# http://viennacl.sourceforge.net/viennacl-documentation.html

# http://enja.org/2011/02/22/adventures-in-pyopencl-part-1-getting-started-with-python/

import pyopencl as cl

import numpy

from PIL import Image

import time,string

from numpy.random import randint as nprnd

import sys

import getopt

import matplotlib.pyplot as plt

# Size of micro blocks

BSZ=16

# 2097152 on HD5850 (with 1GB of RAM)

#  262144 on GT218

#STEP=262144

#STEP=1048576

#STEP=2097152

#STEP=4194304

#STEP=8388608

STEP=16777216

#STEP=268435456

# Flag to define LAPIMAGE between iteration on OpenCL kernel calls

#LAPIMAGE=True

LAPIMAGE=False

# Version 2 of kernel : much optimize one

# a string template is used to replace BSZ (named $block_size) by its value

KERNEL_CODE_ORIG=string.Template("""

#define BSZ $block_size

/* Marsaglia RNG very simple implementation */

#define znew (z=36969*(z&65535)+(z>>16))

#define wnew (w=18000*(w&65535)+(w>>16))

#define MWC ((znew<<16)+wnew )

#define MWCfp (float)(MWC * 2.328306435454494e-10f)

__kernel void MainLoop(__global int *s,float J,float B,float T,uint size,

                       uint iterations,uint seed_w,uint seed_z)

{

   uint base_idx=(uint)(BSZ*get_global_id(0));

   uint base_idy=(uint)(BSZ*get_global_id(1));

   uint base_id=base_idx+base_idy*size;

   uint z=seed_z+(uint)get_global_id(0);

   uint w=seed_w+(uint)get_global_id(1);

   for (uint i=0;i<iterations;i++)

   {

      uint x=(uint)(MWC%BSZ);

      uint y=(uint)(MWC%BSZ);

      int p=s[base_id+x+size*y];

      int u=s[((base_idx+x)%size)+size*((base_idy+y-1)%size)];

      int d=s[((base_idx+x)%size)+size*((base_idy+y+1)%size)];

      int l=s[((base_idx+x-1)%size)+size*((base_idy+y)%size)];

      int r=s[((base_idx+x+1)%size)+size*((base_idy+y)%size)];

      float DeltaE=p*(2.0f*J*(float)(u+d+l+r)+B);

      float factor= ((DeltaE < 0.0f) || (MWCfp < exp(-DeltaE/T))) ? -1:1;

      s[base_id+x+size*y]= factor*p;

   }

}

""")

# Version 2 of kernel : much optimize one

# a string template is used to replace BSZ (named $block_size) by its value

KERNEL_CODE=string.Template("""

/* Marsaglia RNG very simple implementation */

#define znew (z=36969*(z&65535)+(z>>16))

#define wnew (w=18000*(w&65535)+(w>>16))

#define MWC ((znew<<16)+wnew )

#define MWCfp (float)(MWC * 2.328306435454494e-10f)

__kernel void MainLoop(__global int *s,float J,float B,float T,uint size,

                       uint iterations,uint seed_w,uint seed_z)

{

   uint z=seed_z+(uint)get_global_id(0);

   uint w=seed_w+(uint)get_global_id(1);

   int xo,yo,xe,ye,base,OddSize;

   int p,u,d,l,r,factor;

   float DeltaE;

   OddSize=(size%2==0)?1:0;

   base=2*get_global_id(0);

   yo=base/size;

   xo=(yo%2==0) ? base%size:(base%size+OddSize)%size ;

   ye=yo;

   xe=(xo+1)%size;

   for (uint i=0;i<iterations;i++)

   {

      // Odd pixel

      p=s[yo*size+xo];

      u= (yo==     0) ? s[xo+size*(size-1)]:s[xo+size*(yo-1)];

      d= (yo==size-1) ? s[xo]:s[xo+size*(yo+1)];

      l= (xo==     0) ? s[yo*size+size-1]:s[xo-1+size*yo];

      r= (xo==size-1) ? s[yo*size]:s[xo+1+yo*size];

      DeltaE=(float)p*(2.e0f*J*(float)(u+d+l+r)+B);

      factor= ((DeltaE < 0.e0f) || (MWCfp < (float)exp(-DeltaE/T))) ? -1:1;

      s[yo*size+xo]= factor*p;

      // Even pixel

      p=s[ye*size+xe];

      u= (ye==     0) ? s[xe+size*(size-1)]:s[xe+size*(ye-1)];

      d= (ye==size-1) ? s[xe]:s[xe+size*(ye+1)];

      l= (xe==     0) ? s[ye*size+size-1]:s[xe-1+size*ye];

      r= (xe==size-1) ? s[ye*size]:s[xe+1+ye*size];

      DeltaE=(float)p*(2.e0f*J*(float)(u+d+l+r)+B);

      factor= ((DeltaE < 0.e0f) || (MWCfp < (float)exp(-DeltaE/T))) ? -1:1;

      s[ye*size+xe]= factor*p;

      barrier(CLK_LOCAL_MEM_FENCE);

   }

}

""")

def ImageOutput(sigma,prefix):

    Max=sigma.max()

    Min=sigma.min()

    # Normalize value as 8bits Integer

    SigmaInt=(255*(sigma-Min)/(Max-Min)).astype('uint8')

    image = Image.fromarray(SigmaInt)

    image.save("%s.jpg" % prefix)

def CheckLattice(sigma):

    over=sigma[sigma>1]

    under=sigma[sigma<-1]

    if  (over.size+under.size) > 0:

        print "Problem on Lattice on %i spin(s)." % (over.size+under.size)

    else:

        print "No problem on Lattice"

def Metropolis(sigma,J,B,T,iterations,Device,Divider):

    kernel_params = {}

    print iterations,Divider

    # Je detecte un peripherique GPU dans la liste des peripheriques

    Id=1

    HasXPU=False

    for platform in cl.get_platforms():

        for device in platform.get_devices():

            if Id==Device:

                XPU=device

                print "CPU/GPU selected: ",device.name.lstrip()

                HasXPU=True

            Id+=1

    if HasXPU==False:

        print "No XPU #%i found in all of %i devices, sorry..." % (Device,Id-1)

        sys.exit()

    ctx = cl.Context([XPU])

    queue = cl.CommandQueue(ctx,

                            properties=cl.command_queue_properties.PROFILING_ENABLE)

    MetropolisCL = cl.Program(ctx,KERNEL_CODE.substitute(kernel_params)).build()

    # Je recupere les flag possibles pour les buffers

    mf = cl.mem_flags

    # Program based on Kernel2

    sigmaCL = cl.Buffer(ctx, mf.WRITE_ONLY | mf.COPY_HOST_PTR, hostbuf=sigma)

    #sigmaCL = cl.Buffer(ctx, mf.READ_WRITE, sigma.nbytes)

    #cl.enqueue_copy(queue,sigmaCL,sigma).wait();

    i=0

    duration=0.

    if iterations%Divider == 0:

        steps=iterations/Divider;

    else:

        steps=iterations/Divider+1;

    step=0

    while (step*steps < iterations):

        # Call OpenCL kernel

        # sigmaCL is lattice translated in CL format

        # step is number of iterations

        start_time=time.time()

        CLLaunch=MetropolisCL.MainLoop(queue,

                                       (int(sigma.shape[0]*sigma.shape[1]/2),1),None,

                                       sigmaCL,

                                       numpy.float32(J),numpy.float32(B),

                                       numpy.float32(T),

                                       numpy.uint32(sigma.shape[0]),

                                       numpy.uint32(steps),

                                       numpy.uint32(2008),

                                       numpy.uint32(1010))

        CLLaunch.wait()

            # elapsed = 1e-9*(CLLaunch.profile.end - CLLaunch.profile.start)

        elapsed = time.time()-start_time

        print "Iteration %i with T=%f and %i iterations in %f: " % (step,T,steps,elapsed)

        if LAPIMAGE:

                cl.enqueue_copy(queue, sigma, sigmaCL).wait()

                checkLattice(sigma)

                ImageOutput(sigma,"Ising2D_GPU_OddEven_%i_%1.1f_%.3i_Lap" % (SIZE,T,i))

        step=step+1

        duration=duration+elapsed

    cl.enqueue_copy(queue,sigma,sigmaCL).wait()

    CheckLattice(sigma)

    sigmaCL.release()

    return(duration)

def Magnetization(sigma,M):

    return(numpy.sum(sigma)/(sigma.shape[0]*sigma.shape[1]*1.0))

def Energy(sigma,J,B):

    # Copy & Cast values

    E=numpy.copy(sigma).astype(numpy.float32)

    # Slice call to estimate Energy

    E[1:-1,1:-1]=E[1:-1,1:-1]*(2.0*J*(E[:-2,1:-1]+E[2:,1:-1]+

                                          E[1:-1,:-2]+E[1:-1,2:])+B)

    # Clean perimeter

    E[:,0]=0

    E[:,-1]=0

    E[0,:]=0

    E[-1,:]=0

    Energy=numpy.sum(E)

    return(Energy/(E.shape[0]*E.shape[1]*1.0))

def CriticalT(T,E):

    Epoly=numpy.poly1d(numpy.polyfit(T,E,T.size/3))

    dEpoly=numpy.diff(Epoly(T))

    dEpoly=numpy.insert(dEpoly,0,0)

    return(T[numpy.argmin(dEpoly)])

def PolyFitE(T,E):

    Epoly=numpy.poly1d(numpy.polyfit(T,E,T.size/3))

    return(Epoly(T))

def DisplayCurves(T,E,M,J,B):

    plt.xlabel("Temperature")

    plt.ylabel("Energy")

    Experience,=plt.plot(T,E,label="Energy")

    plt.legend()

    plt.show()

if __name__=='__main__':

    # Set defaults values

    # Coupling factor

    J=1.

    # External Magnetic Field is null

    B=0.

    # Size of Lattice

    Size=256

    # Default Temperatures (start, end, step)

    Tmin=0.1

    Tmax=5

    Tstep=0.1

    # Default Number of Iterations

    Iterations=Size*Size

    # Default Device is first one

    Device=1

    # Default Divider

    Divider=1

    # Curves is True to print the curves

    Curves=False

    OCL_vendor={}

    OCL_type={}

    OCL_description={}

    try:

        import pyopencl as cl

        print "\nHere are available OpenCL devices:"

        Id=1

        for platform in cl.get_platforms():

            for device in platform.get_devices():

                OCL_vendor[Id]=platform.vendor.lstrip().rstrip()

                #OCL_type[Id]=cl.device_type.to_string(device.type)

                OCL_type[Id]="xPU"

                OCL_description[Id]=device.name.lstrip().rstrip()

                print "* Device #%i from %s of type %s : %s" % (Id,OCL_vendor[Id],OCL_type[Id],OCL_description[Id])

                Id=Id+1

        OCL_MaxDevice=Id-1

        print

    except ImportError:

        print "Your platform does not seem to support OpenCL"

        sys.exit(0)

    try:

        opts, args = getopt.getopt(sys.argv[1:],"hcj:b:z:i:s:e:p:d:v:",["coupling=","magneticfield=","size=","iterations=","tempstart=","tempend=","tempstep=","units",'device='])

    except getopt.GetoptError:

        print '%s -d <Device Id> -j <Coupling Factor> -b <Magnetic Field> -z <Size of Square Lattice> -i <Iterations> -s <Minimum Temperature> -e <Maximum Temperature> -p <steP Temperature> -v <diVider> -c (Print Curves)' % sys.argv[0]

        sys.exit(2)

    for opt, arg in opts:

        if opt == '-h':

            print '%s -d <Device Id> -j <Coupling Factor> -b <Magnetic Field> -z <Size of Square Lattice> -i <Iterations> -s <Minimum Temperature> -e <Maximum Temperature> -p <steP Temperature> -v <diVider> -c (Print Curves)' % sys.argv[0]

            sys.exit()

        elif opt == '-c':

            Curves=True

        elif opt in ("-d", "--device"):

            Device = int(arg)

            if Device>OCL_MaxDevice:

                "Device #%s seems not to be available !"

                sys.exit()

        elif opt in ("-j", "--coupling"):

            J = float(arg)

        elif opt in ("-b", "--magneticfield"):

            B = float(arg)

        elif opt in ("-s", "--tempmin"):

            Tmin = float(arg)

        elif opt in ("-e", "--tempmax"):

            Tmax = float(arg)

        elif opt in ("-p", "--tempstep"):

            Tstep = float(arg)

        elif opt in ("-i", "--iterations"):

            Iterations = int(arg)

        elif opt in ("-z", "--size"):

            Size = int(arg)

        elif opt in ("-v", "--divider"):

            Divider = int(arg)

    print "Here are parameters of simulation:"

    print "* Device selected #%s: %s of type %s from %s" % (Device,OCL_description[Device],OCL_type[Device],OCL_vendor[Device])

    print "* Coupling Factor J : %s" % J

    print "* Magnetic Field B :  %s" % B

    print "* Size of lattice : %sx%s" % (Size,Size)

    print "* Divider inside loop : %s" % Divider

    print "* Iterations : %s" % Iterations

    print "* Temperatures from %s to %s by %s" % (Tmin,Tmax,Tstep)

    LAPIMAGE=False

    if Iterations<STEP:

        STEP=Iterations

    numpy.random.seed(20081010)

    sigmaIn=numpy.where(numpy.random.randn(Size,Size)>0,1,-1).astype(numpy.int32)

    ImageOutput(sigmaIn,"Ising2D_GPU_Local_%i_Initial" % (Size))

    Trange=numpy.arange(Tmin,Tmax+Tstep,Tstep)

    E=[]

    M=[]

    for T in Trange:

        sigma=numpy.copy(sigmaIn)

        duration=Metropolis(sigma,J,B,T,Iterations,Device,Divider)

        E=numpy.append(E,Energy(sigma,J,B))

        M=numpy.append(M,Magnetization(sigma,B))

        ImageOutput(sigma,"Ising2D_GPU_OddEven_%i_%1.1f_Final" % (Size,T))

        print "GPU/CPU Time : %f" % (duration)

        print "Total Energy at Temperature %f : %f" % (T,E[-1])

        print "Total Magnetization at Temperature %f : %f" % (T,M[-1])

    # Save output

    numpy.savez("Ising2D_GPU_Global_%i_%.8i" % (Size,Iterations),(Trange,E,M))

    # Estimate Critical temperature

    print "The critical temperature on %ix%i lattice with J=%f, B=%f is %f " % (Size,Size,J,B,CriticalT(Trange,E))

    if Curves:

        DisplayCurves(Trange,E,M,J,B)