Python gpuarray.GPUArray方法代碼示例

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def setOrig(self, img):
    """
    To set the original image from a given CPU or GPU array.

    If it is a GPU array, it will NOT be copied.
    Note that the most efficient method is to write directly over
    self.devOrig with some kernel and then run self.updateOrig()
    """
    assert img.shape == (self.h, self.w), \
      "Got a {} image in a {} correlation routine!".format(
        img.shape, (self.h, self.w))
    if isinstance(img, np.ndarray):
      self.debug(3, "Setting original image from ndarray")
      self.devOrig.set(img)
    elif isinstance(img, gpuarray.GPUArray):
      self.debug(3, "Setting original image from GPUArray")
      self.devOrig = img
    else:
      self.debug(0, "Error ! Unknown type of data given to setOrig()")
      raise ValueError
    self.updateOrig() 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def setFields(self, fieldsX, fieldsY):
    """
    Method to give the fields to identify with the routine.

    This is necessary only once and can be done multiple times, but the routine
    have to be initialized with .prepare(), causing a slight overhead
    Takes a tuple/list of 2 (gpu)arrays[Nfields,x,y] (one for displacement
    along x and one along y)
    """
    self.debug(2, "Setting fields")
    if isinstance(fieldsX, np.ndarray):
      self.devFieldsX.set(fieldsX)
      self.devFieldsY.set(fieldsY)
    elif isinstance(fieldsX, gpuarray.GPUArray):
      self.devFieldsX = fieldsX
      self.devFieldsY = fieldsY
    self.fields = True 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def setImage(self, img_d):
    """
    Set the image to compare with the original

    Note that calling this method is not necessary: you can do .getDisp(image)
    This will automatically call this method first
    """
    assert img_d.shape == (self.h, self.w), \
      "Got a {} image in a {} correlation routine!".format(
        img_d.shape, (self.h, self.w))
    if isinstance(img_d, np.ndarray):
      self.debug(3, "Creating texture from numpy array")
      self.array_d = cuda.matrix_to_array(img_d, "C")
    elif isinstance(img_d, gpuarray.GPUArray):
      self.debug(3, "Creating texture from gpuarray")
      self.array_d = cuda.gpuarray_to_array(img_d, "C")
    else:
      self.debug(0, "Error ! Unknown type of data given to .setImage()")
      raise ValueError
    self.tex_d.set_array(self.array_d)
    self.devX.set(np.zeros(self.Nfields, dtype=np.float32)) 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def as_contiguous(self):
        """
        Creates a new GPUArray with the same dimensions, but using contiguous memory

        Returns:
            New contiguous GPUArray with separate underlying device allocation
        """
        contig_tensor = GPUArray(self.tensor.shape, dtype=self.tensor.dtype)
        src_strides = [s // self.tensor.dtype.itemsize for s in self.tensor.strides]
        dst_strides = [s // contig_tensor.dtype.itemsize for s in contig_tensor.strides]
        kernel = _get_copy_transpose_kernel(self.tensor.dtype,
                                            self.tensor.shape,
                                            range(len(self.tensor.shape)))
        params = [contig_tensor.gpudata, self.tensor.gpudata] + list(kernel.args)
        params = params + src_strides + dst_strides
        kernel.prepared_async_call(kernel.grid, kernel.block, None, *params)
        return contig_tensor 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def rand_cuda(shape):
    """Create random values in the range [0, 1] in a GPUArray.

    Parameters
    ----------
    shape : tuple
        Size of the random array.

    Returns
    -------
    gpuarray
        Random floats from 0 to 1 in GPUArray.

    Examples
    --------
    >>> a = rand_cuda((2, 2))
    [[ 0.80916596,  0.82687163],
     [ 0.03921388,  0.44197764]]

    >>> type(a)
    <class 'pycuda.gpuarray.GPUArray'>
    """
    return pycuda.curandom.rand(shape, dtype=float32) 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def get_cuda(a):
    """Return GPUArray from GPU memory as NumPy array.

    Parameters
    ----------
    a : gpuarray
        Data on the GPU memory to retrieve.

    Returns
    -------
    array
        The GPUArray returned to RAM as a NumPy array.

    Examples
    --------
    >>> a = give_cuda([1, 2, 3])
    >>> b = get_cuda(a)
    [ 1.,  2.,  3.]

    >>> type(b)
    <class 'numpy.ndarray'>
    """
    return a.get() 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def zeros_cuda(shape):
    """Create GPUArray of zeros directly on GPU memory.

    Parameters
    ----------
    shape : tuple
        Dimensions of the GPUArray.

    Returns
    -------
    gpuarray
        GPUArray of zeros.

    Examples
    --------
    >>> a = zeros_cuda((3, 2))
    [[ 0.,  0.],
     [ 0.,  0.],
     [ 0.,  0.]]

    >>> type(a)
    <class 'pycuda.gpuarray.GPUArray'>
    """
    return cuda_array.zeros(shape, dtype=float32) 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def hstack_cuda(a, b, dim=4):
    """Stack two GPUArrays horizontally.

    Parameters
    ----------
    a : gpuarray
        First GPUArray.
    b : gpuarray
        Second GPUArray.

    Returns
    -------
    gpuarray
        Horizontally stacked GPUArrays.
    """
    m, n = a.shape
    o = b.shape[1]
    nx = int(ceil((n + o) / dim))
    ny = int(ceil(m / dim))

    func = mod.get_function('hstack_cuda')
    c = pycuda.gpuarray.empty((m, n + o), dtype=float32)
    func(int32(m), int32(n), int32(o), a, b, c, block=(dim, dim, 1), grid=(nx, ny, 1))

    return c 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def setMask(self, mask):
    self.debug(3, "Setting the mask")
    assert mask.shape == (self.h, self.w), \
      "Got a {} mask in a {} routine.".format(mask.shape, (self.h, self.w))
    if not mask.dtype == np.float32:
      self.debug(2, "Converting the mask to float32")
      mask = mask.astype(np.float32)
    if isinstance(mask, np.ndarray):
      self.maskArray = cuda.matrix_to_array(mask, 'C')
    elif isinstance(mask, gpuarray.GPUArray):
      self.maskArray = cuda.gpuarray_to_array(mask, 'C')
    else:
      self.debug(0, "Error! Mask data type not understood")
      raise ValueError
    self.texMask.set_array(self.maskArray) 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def setDisp(self, X):
    assert X.shape == (self.Nfields,), \
      "Incorrect initialization of the parameters"
    if isinstance(X, gpuarray.GPUArray):
      self.devX = X
    elif isinstance(X, np.ndarray):
      self.devX.set(X)
    else:
      self.debug(0, "Error! Unknown type of data given to CorrelStage.setDisp")
      raise ValueError 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def __call__(self, buffer_alloc):
        """
        Allocates the GPUTensor object as a view of a pre-allocated buffer.

        Arguments:
            buffer_alloc (DeviceAllocation): Memory handle returned by pycuda
                allocator
        """
        tensor_description = self.tensor_description
        layout = tensor_description.layout
        dtype = self.transformer.storage_dtype(tensor_description.dtype)

        if layout:
            gpudata = int(buffer_alloc) + (layout.offset * dtype.itemsize)
            strides = tuple([s * dtype.itemsize for s in layout.strides])
            new_tensor = GPUArray(layout.shape,
                                  dtype,
                                  gpudata=gpudata,
                                  strides=strides)
        else:
            gpudata = int(buffer_alloc) + tensor_description.offset
            new_tensor = GPUArray(tensor_description.shape,
                                  dtype,
                                  gpudata=gpudata,
                                  strides=tensor_description.strides)
        self._tensor = new_tensor
        self.transformer.tensors[self.tensor_name] = self._tensor 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def get(self, tensor):
        """
        Copy the device tensor to a numpy array.

        Arguments:
            tensor (np.ndarray): Optional output array

        Returns:
            Numpy array containing tensor data
        """
        if np.sum(self.tensor.strides) != 0:
            if self.is_contiguous or self.tensor.shape == () or np.prod(self.tensor.shape) == 1:
                contig_tensor = self.tensor
            else:
                # Need to do memcpy from contiguous device memory
                contig_tensor = self.as_contiguous()

            if tensor is None:
                if contig_tensor.shape != self.tensor_description.shape:
                    return contig_tensor.get().reshape(self.tensor_description.shape)
                else:
                    return contig_tensor.get()
            tensor[:] = contig_tensor.get()
        else:
            # Tensor is just a broadcasted scalar, get scalar value and fill output array
            view = GPUArray((1, ), dtype=self.tensor.dtype, gpudata=self.tensor.gpudata)[0]
            value = view.get()

            if tensor is None:
                out = np.ndarray(self.tensor.shape, dtype=self.tensor.dtype)
                out.fill(value)
                return out
            tensor.fill(value)

        return tensor 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def to_array(in_array, copy = True):
    """
    Helper function to convert input from a different module to af.Array

    Parameters
    -------------

    in_array : array like object
             Can be one of the following:
             - numpy.ndarray
             - pycuda.GPUArray
             - pyopencl.Array
             - numba.cuda.cudadrv.devicearray.DeviceNDArray
             - array.array
             - list
    copy : Bool specifying if array is to be copied.
          Default is true.
          Can only be False if array is fortran contiguous.

    Returns
    --------------
    af.Array of same dimensions as input after copying the data from the input

    """
    if AF_NUMPY_FOUND and isinstance(in_array, NumpyArray):
        return np_to_af_array(in_array, copy)
    if AF_PYCUDA_FOUND and isinstance(in_array, CudaArray):
        return pycuda_to_af_array(in_array, copy)
    if AF_PYOPENCL_FOUND and isinstance(in_array, OpenclArray):
        return pyopencl_to_af_array(in_array, copy)
    if AF_NUMBA_FOUND and isinstance(in_array, NumbaCudaArray):
        return numba_to_af_array(in_array, copy)
    return Array(src=in_array) 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def fill_normal(self, ary, mean=0, stdv=1):
        """
        Fill ary with normally distributed random numbers.

        Arguments:
            ary (Tensor): Tensor to fill with random values
            mean (float): Mean value. Default 0
            stdv (float): standard deviation value.  Default 1
        """
        self.pcg.fill_normal(p_gpuarray(ary.shape, ary.dtype, gpudata=ary.gpudata))
        if not all([mean == 0, stdv == 1]):
            ary[:] = ary * stdv + mean 

# 需要導入模塊: from pycuda import gpuarray [as 別名]
# 或者: from pycuda.gpuarray import GPUArray [as 別名]
def give_cuda(a, type='real'):
    """Give a list or an array to GPU memory.

    Parameters
    ----------
    a : array, list
        Data to send to the GPU memory.
    type : str
        'real' or 'complex'.

    Returns
    -------
    gpuarray
        GPUArray version of the input array.

    Examples
    --------
    >>> a = give_cuda([[1., 2., 3.], [4., 5., 6.]])
    [[ 1.,  2.,  3.],
     [ 4.,  5.,  6.]]

    >>> type(a)
    <class 'pycuda.gpuarray.GPUArray'>

    >>> a.shape
    (2, 3)

    >>> a.dtype
    'float32'

    >>> a.reshape((1, 6))
    [[ 1.,  2.,  3.,  4.,  5.,  6.]]
    """
    if type == 'real':
        return cuda_array.to_gpu(array(a).astype(float32))
    elif type == 'complex':
        return cuda_array.to_gpu(array(a).astype(complex64))
    else:
        raise NotImplementedError