Centre Blaise Pascal » Bench4GPU

#!/usr/bin/env python3

import numpy as np

import pyopencl as cl

from numpy import pi,cos,sin

#

def NumpyFFT(x,y):

    xy=x+1.j*y

    XY=np.fft.fft(xy)

    return(XY.real,XY.imag)

#

def OpenCLFFT(x,y,device):

    import pyopencl as cl

    import pyopencl.array as cla

    import time

    import gpyfft

    from gpyfft.fft import FFT

    Id=0

    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

            # print(Id)

    if HasXPU==False:

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

        sys.exit()

    try:

        ctx = cl.Context(devices=[XPU])

        queue = cl.CommandQueue(ctx,properties=cl.command_queue_properties.PROFILING_ENABLE)

    except:

        print("Crash during context creation")

    XY_gpu = cla.to_device(queue, x+1.j*y)

    transform = FFT(ctx, queue, XY_gpu)

    event, = transform.enqueue()

    event.wait()

    XY = XY_gpu.get()

    return(XY.real,XY.imag)

# Naive Discrete Fourier Transform

def MyDFT(x,y):

    size=x.shape[0]

    X=np.zeros(size).astype(np.float32)

    Y=np.zeros(size).astype(np.float32)

    for i in range(size):

        for j in range(size):

            X[i]=X[i]+x[j]*cos(2.*pi*i*j/size)-y[j]*sin(2.*pi*i*j/size)

            Y[i]=Y[i]+x[j]*sin(2.*pi*i*j/size)+y[j]*cos(2.*pi*i*j/size)

    return(X,Y)

# Numpy Discrete Fourier Transform

def NumpyDFT(x,y):

    size=x.shape[0]

    X=np.zeros(size).astype(np.float32)

    Y=np.zeros(size).astype(np.float32)

    nj=np.multiply(2.0*np.pi/size,np.arange(size)).astype(np.float32)

    for i in range(size):

        X[i]=np.sum(np.subtract(np.multiply(np.cos(i*nj),x),np.multiply(np.sin(i*nj),y)))

        Y[i]=np.sum(np.add(np.multiply(np.sin(i*nj),x),np.multiply(np.cos(i*nj),y)))

    return(X,Y)

# Numba Discrete Fourier Transform

import numba

@numba.njit(parallel=True)

def NumbaDFT(x,y):

    size=x.shape[0]

    X=np.zeros(size).astype(np.float32)

    Y=np.zeros(size).astype(np.float32)

    nj=np.multiply(2.0*np.pi/size,np.arange(size)).astype(np.float32)

    for i in numba.prange(size):

        X[i]=np.sum(np.subtract(np.multiply(np.cos(i*nj),x),np.multiply(np.sin(i*nj),y)))

        Y[i]=np.sum(np.add(np.multiply(np.sin(i*nj),x),np.multiply(np.cos(i*nj),y)))

    return(X,Y)

# OpenCL complete operation

def OpenCLDFT(a_np,b_np,Device):

    Id=0

    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

            # print(Id)

    if HasXPU==False:

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

        sys.exit()

    try:

        ctx = cl.Context(devices=[XPU])

        queue = cl.CommandQueue(ctx,properties=cl.command_queue_properties.PROFILING_ENABLE)

    except:

        print("Crash during context creation")

    TimeIn=time.time()

    # Copy from Host to Device using pointers

    mf = cl.mem_flags

    a_g = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=a_np)

    b_g = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=b_np)

    Elapsed=time.time()-TimeIn

    print("Copy from Host 2 Device : %.3f" % Elapsed)

    TimeIn=time.time()

    # Definition of kernel under OpenCL

    prg = cl.Program(ctx, """

#define PI 3.141592653589793

__kernel void MyDFT(

    __global const float *a_g, __global const float *b_g, __global float *A_g, __global float *B_g)

{

  int gid = get_global_id(0);

  uint size = get_global_size(0);

  float A=0.,B=0.;

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

  {

     A+=a_g[i]*cos(2.*PI*(float)(gid*i)/(float)size)-b_g[i]*sin(2.*PI*(float)(gid*i)/(float)size);

     B+=a_g[i]*sin(2.*PI*(float)(gid*i)/(float)size)+b_g[i]*cos(2.*PI*(float)(gid*i)/(float)size);

  }

  A_g[gid]=A;

  B_g[gid]=B;

}

""").build()

    Elapsed=time.time()-TimeIn

    print("Building kernels : %.3f" % Elapsed)

    TimeIn=time.time()

    # Memory allocation on Device for result

    A_ocl = np.empty_like(a_np)

    B_ocl = np.empty_like(a_np)

    Elapsed=time.time()-TimeIn

    print("Allocation on Host for results : %.3f" % Elapsed)

    A_g = cl.Buffer(ctx, mf.WRITE_ONLY, A_ocl.nbytes)

    B_g = cl.Buffer(ctx, mf.WRITE_ONLY, B_ocl.nbytes)

    Elapsed=time.time()-TimeIn

    print("Allocation on Device for results : %.3f" % Elapsed)

    TimeIn=time.time()

    # Synthesis of function "sillysum" inside Kernel Sources

    knl = prg.MyDFT  # Use this Kernel object for repeated calls

    Elapsed=time.time()-TimeIn

    print("Synthesis of kernel : %.3f" % Elapsed)

    TimeIn=time.time()

    # Call of kernel previously defined

    CallCL=knl(queue, a_np.shape, None, a_g, b_g, A_g, B_g)

    #

    CallCL.wait()

    Elapsed=time.time()-TimeIn

    print("Execution of kernel : %.3f" % Elapsed)

    TimeIn=time.time()

    # Copy from Device to Host

    cl.enqueue_copy(queue, A_ocl, A_g)

    cl.enqueue_copy(queue, B_ocl, B_g)

    Elapsed=time.time()-TimeIn

    print("Copy from Device 2 Host : %.3f" % Elapsed)

    # Liberation of memory

    a_g.release()

    b_g.release()

    A_g.release()

    B_g.release()

    return(A_ocl,B_ocl)

# CUDA complete operation

def CUDADFT(a_np,b_np,Device,Threads):

    # import pycuda.autoinit

    import pycuda.driver as drv

    from pycuda.compiler import SourceModule

    try:

        # For PyCUDA import

        import pycuda.driver as cuda

        from pycuda.compiler import SourceModule

        cuda.init()

        for Id in range(cuda.Device.count()):

            if Id==Device:

                XPU=cuda.Device(Id)

                print("GPU selected %s" % XPU.name())

        print

    except ImportError:

        print("Platform does not seem to support CUDA")

    Context=XPU.make_context()

    TimeIn=time.time()

    mod = SourceModule("""

#define PI 3.141592653589793

__global__ void MyDFT(float *A_g, float *B_g, const float *a_g,const float *b_g)

{

  const int gid = blockIdx.x*blockDim.x+threadIdx.x;

  uint size = gridDim.x*blockDim.x;

  float A=0.,B=0.;

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

  {

     A+=a_g[i]*cos(2.*PI*(float)(gid*i)/(float)size)-b_g[i]*sin(2.*PI*(float)(gid*i)/(float)size);

     B+=a_g[i]*sin(2.*PI*(float)(gid*i)/(float)size)+b_g[i]*cos(2.*PI*(float)(gid*i)/(float)size);

  }

  A_g[gid]=A;

  B_g[gid]=B;

}

""")

    Elapsed=time.time()-TimeIn

    print("Definition of kernel : %.3f" % Elapsed)

    TimeIn=time.time()

    MyDFT = mod.get_function("MyDFT")

    Elapsed=time.time()-TimeIn

    print("Synthesis of kernel : %.3f" % Elapsed)

    TimeIn=time.time()

    A_np = np.zeros_like(a_np)

    B_np = np.zeros_like(a_np)

    Elapsed=time.time()-TimeIn

    print("Allocation on Host for results : %.3f" % Elapsed)

    Size=a_np.size

    if (Size % Threads != 0):

        print("Impossible : %i not multiple of %i..." % (Threads,Size) )

        TimeIn=time.time()

        MyDFT(drv.Out(A_np), drv.Out(B_np), drv.In(a_np), drv.In(b_np),

              block=(1,1,1), grid=(a_np.size,1))

        Elapsed=time.time()-TimeIn

        print("Execution of kernel : %.3f" % Elapsed)

    else:

        Blocks=int(Size/Threads)

        TimeIn=time.time()

        MyDFT(drv.Out(A_np), drv.Out(B_np), drv.In(a_np), drv.In(b_np),

              block=(Threads,1,1), grid=(Blocks,1))

        Elapsed=time.time()-TimeIn

        print("Execution of kernel : %.3f" % Elapsed)

    Context.pop()

    Context.detach()

    return(A_np,B_np)

import sys

import time

if __name__=='__main__':

    SIZE=1024

    Device=0

    NaiveMethod=False

    NumpyFFTMethod=True

    OpenCLFFTMethod=True

    NumpyMethod=False

    NumbaMethod=False

    OpenCLMethod=False

    CUDAMethod=False

    Threads=1

    import getopt

    HowToUse='%s -n [Naive] -y [numpY] -a [numbA] -o [OpenCL] -c [CUDA] -s <SizeOfVector> -d <DeviceId> -t <threads>'

    try:

        opts, args = getopt.getopt(sys.argv[1:],"nyaochs:d:t:",["size=","device="])

    except getopt.GetoptError:

        print(HowToUse % sys.argv[0])

        sys.exit(2)

    # List of Devices

    Devices=[]

    Alu={}

    for opt, arg in opts:

        if opt == '-h':

            print(HowToUse % sys.argv[0])

            print("\nInformations about devices detected under OpenCL API:")

            # For PyOpenCL import

            try:

                import pyopencl as cl

                Id=0

                for platform in cl.get_platforms():

                    for device in platform.get_devices():

                        #deviceType=cl.device_type.to_string(device.type)

                        deviceType="xPU"

                        print("Device #%i from %s of type %s : %s" % (Id,platform.vendor.lstrip(),deviceType,device.name.lstrip()))

                        Id=Id+1

            except:

                print("Your platform does not seem to support OpenCL")

            print("\nInformations about devices detected under CUDA API:")

            # For PyCUDA import

            try:

                import pycuda.driver as cuda

                cuda.init()

                for Id in range(cuda.Device.count()):

                    device=cuda.Device(Id)

                    print("Device #%i of type GPU : %s" % (Id,device.name()))

                print

            except:

                print("Your platform does not seem to support CUDA")

            sys.exit()

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

            Device=int(arg)

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

            SIZE = int(arg)

        elif opt in ("-t", "--threads"):

            Threads = int(arg)

        elif opt in ("-n"):

            NaiveMethod=True

        elif opt in ("-y"):

            NumpyMethod=True

        elif opt in ("-a"):

            NumbaMethod=True

        elif opt in ("-o"):

            OpenCLMethod=True

        elif opt in ("-c"):

            CUDAMethod=True

    print("Device Selection : %i" % Device)

    print("Size of complex vector : %i" % SIZE)

    print("DFT Naive computation %s " % NaiveMethod )

    print("DFT Numpy computation %s " % NumpyMethod )

    print("DFT Numba computation %s " % NumbaMethod )

    print("DFT OpenCL computation %s " % OpenCLMethod )

    print("DFT CUDA computation %s " % CUDAMethod )

    if CUDAMethod:

        try:

            # For PyCUDA import

            import pycuda.driver as cuda

            cuda.init()

            for Id in range(cuda.Device.count()):

                device=cuda.Device(Id)

                print("Device #%i of type GPU : %s" % (Id,device.name()))

                if Id in Devices:

                    Alu[Id]='GPU'

        except ImportError:

            print("Platform does not seem to support CUDA")

    if OpenCLMethod:

        try:

            # For PyOpenCL import

            import pyopencl as cl

            Id=0

            for platform in cl.get_platforms():

                for device in platform.get_devices():

                    #deviceType=cl.device_type.to_string(device.type)

                    deviceType="xPU"

                    print("Device #%i from %s of type %s : %s" % (Id,platform.vendor.lstrip().rstrip(),deviceType,device.name.lstrip().rstrip()))

                    if Id in Devices:

                    # Set the Alu as detected Device Type

                        Alu[Id]=deviceType

                    Id=Id+1

        except ImportError:

            print("Platform does not seem to support OpenCL")

    a_np = np.ones(SIZE).astype(np.float32)

    b_np = np.ones(SIZE).astype(np.float32)

    C_np = np.zeros(SIZE).astype(np.float32)

    D_np = np.zeros(SIZE).astype(np.float32)

    C_np[0] = np.float32(SIZE)

    D_np[0] = np.float32(SIZE)

    # Native & Naive Implementation

    if NaiveMethod:

        print("Performing naive implementation")

        TimeIn=time.time()

        c_np,d_np=MyDFT(a_np,b_np)

        NativeElapsed=time.time()-TimeIn

        NativeRate=int(SIZE/NativeElapsed)

        print("NativeRate: %i" % NativeRate)

        print("Precision: ",np.linalg.norm(c_np-C_np),

              np.linalg.norm(d_np-D_np))

    # Native & Numpy Implementation

    if NumpyMethod:

        print("Performing Numpy implementation")

        TimeIn=time.time()

        e_np,f_np=NumpyDFT(a_np,b_np)

        NumpyElapsed=time.time()-TimeIn

        NumpyRate=int(SIZE/NumpyElapsed)

        print("NumpyRate: %i" % NumpyRate)

        print("Precision: ",np.linalg.norm(e_np-C_np),

              np.linalg.norm(f_np-D_np))

    # Native & Numba Implementation

    if NumbaMethod:

        print("Performing Numba implementation")

        TimeIn=time.time()

        g_np,h_np=NumbaDFT(a_np,b_np)

        NumbaElapsed=time.time()-TimeIn

        NumbaRate=int(SIZE/NumbaElapsed)

        print("NumbaRate: %i" % NumbaRate)

        print("Precision: ",np.linalg.norm(g_np-C_np),

              np.linalg.norm(h_np-D_np))

    # OpenCL Implementation

    if OpenCLMethod:

        print("Performing OpenCL implementation")

        TimeIn=time.time()

        i_np,j_np=OpenCLDFT(a_np,b_np,Device)

        OpenCLElapsed=time.time()-TimeIn

        OpenCLRate=int(SIZE/OpenCLElapsed)

        print("OpenCLRate: %i" % OpenCLRate)

        print("Precision: ",np.linalg.norm(i_np-C_np),

              np.linalg.norm(j_np-D_np))

    # CUDA Implementation

    if CUDAMethod:

        print("Performing CUDA implementation")

        TimeIn=time.time()

        k_np,l_np=CUDADFT(a_np,b_np,Device,Threads)

        CUDAElapsed=time.time()-TimeIn

        CUDARate=int(SIZE/CUDAElapsed)

        print("CUDARate: %i" % CUDARate)

        print("Precision: ",np.linalg.norm(k_np-C_np),

              np.linalg.norm(l_np-D_np))

    if NumpyFFTMethod:

        print("Performing NumpyFFT implementation")

        TimeIn=time.time()

        m_np,n_np=NumpyFFT(a_np,b_np)

        NumpyFFTElapsed=time.time()-TimeIn

        NumpyFFTRate=int(SIZE/NumpyFFTElapsed)

        print("NumpyFFTRate: %i" % NumpyFFTRate)

        print("Precision: ",np.linalg.norm(m_np-C_np),

              np.linalg.norm(n_np-D_np))

    # OpenCL Implementation

    if OpenCLFFTMethod:

        print("Performing OpenCL implementation")

        TimeIn=time.time()

        i_np,j_np=OpenCLFFT(a_np,b_np,Device)

        OpenCLFFTElapsed=time.time()-TimeIn

        OpenCLFFTRate=int(SIZE/OpenCLFFTElapsed)

        print("OpenCLRate: %i" % OpenCLFFTRate)

        print("Precision: ",np.linalg.norm(i_np-C_np),

              np.linalg.norm(j_np-D_np))