Centre Blaise Pascal » Bench4GPU

def NumpyFFT(x,y):

    xy=np.csingle(x+1.j*y)

    XY=np.fft.fft(xy)

    return(XY.real,XY.imag)

#

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

    TimeIn=time.time()

    event, = transform.enqueue()

    event.wait()

    Elapsed=time.time()-TimeIn

    XY = XY_gpu.get()

    Elapsed=time.time()-TimeIn

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

    return(XY.real,XY.imag)

# Naive Discrete Fourier Transform

    Device=0

    NaiveMethod=False

    NumpyFFTMethod=True

    NumpyMethod=False

    NumbaMethod=False

    OpenCLMethod=False

    Threads=1

    import getopt

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

    try:

    except getopt.GetoptError:

        print(HowToUse % sys.argv[0])

        sys.exit(2)

    print("FFT Numpy computation %s " % NumpyFFTMethod )

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

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

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

    if CUDAMethod:

    # OpenCL Implementation

    if OpenCLFFTMethod:

        TimeIn=time.time()

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

        OpenCLFFTElapsed=time.time()-TimeIn

        OpenCLFFTRate=int(SIZE/OpenCLFFTElapsed)

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

              np.linalg.norm(j_np-D_np))