nabu.cuda.convolution module

class nabu.cuda.convolution.Convolution(shape, kernel, axes=None, mode=None, extra_options=None, cuda_options=None)

Bases: object

A class for performing convolution on GPU with CUDA, but without using textures (unlike for example in silx.opencl.convolution)

Constructor of Cuda Convolution.

Parameters:

• shape (tuple) – Shape of the array.

• kernel (array-like) – Convolution kernel (1D, 2D or 3D).

• axes (tuple, optional) – Axes along which the convolution is performed, for batched convolutions.

• mode (str, optional) –

  Boundary handling mode. Available modes are:

  • ”reflect”: cba|abcd|dcb

  • ”nearest”: aaa|abcd|ddd

  • ”wrap”: bcd|abcd|abc

  • ”constant”: 000|abcd|000

  Default is “reflect”.

• extra_options (dict, optional) –

  Advanced options (dict). Current options are:

  • ”allocate_input_array”: True

  • ”allocate_output_array”: True

  • ”allocate_tmp_array”: True

  • ”sourcemodule_kwargs”: {}

  • ”batch_along_flat_dims”: True

convolve(array, output=None)

Convolve an array with the class kernel.

Parameters:

• array – Input array. Can be numpy.ndarray or pycuda.gpuarray.GPUArray.

• output – Output array. Can be numpy.ndarray or pycuda.gpuarray.GPUArray.