CUDA Nifty Gridder

CUDA Nifty Gridder is a GPU-accelerated implementation of Nifty Gridder.

Requirements

The system used for development must have an CUDA capable GPU and Python 3 and pip installed, along with the Python packages numpy and setuptools.

To run the unit tests Nifty Gridder must also be installed.

Installation

From the root directory of the CUDA Nifty Gridder git repo, run:

$ pip3 install --user .

Alternatively, assuming git is installed, this can also be done with:

$ pip3 install 'git+https://gitlab.com/ska-telescope/sdp/ska-gridder-nifty-cuda.git'

This is also an easy way to install the reference Nifty Gridder in order to run the unit tests.

$ pip3 install 'git+https://gitlab.mpcdf.mpg.de/ift/nifty_gridder.git'

Usage

CUDA Nifty Gridder can then be used in your Python code as:

import cuda_nifty_gridder as cng 

See the files test_ms2dirty.py and test_dirty2ms.py in tests/unit/ for more ideas on how to use CUDA Nifty Gridder.

The full documentation is available at ReadTheDocs and the GitLab Page

Unit Tests

If PyTest is installed, the unit tests can be run from the root directory of the repo as:

$ py.test -s

the -s provides feedback during the tests, and can be omitted.

Otherwise the tests can be run as:

$ python3 test.py

Public API Documentation

This section lists the functions implemented, using the same interface as DUCC/nifty gridder, by Martin Reinecke.

Note

Only square images are supported at the moment, so, pixsize_x_rad and pixsize_y_rad must be equal, and npix_x and npix_y must be equal.

Functions

cuda_nifty_gridder.ms2dirty(uvw, freq, ms, weight, npix_x, npix_y, pixsize_x_rad, pixsize_y_rad, epsilon, do_wstacking=True)[source]

Converts visibilities to a dirty image.

Parameters

uvw (numpy.array((nrows, 3), dtype=numpy.float32 or numpy.float64)) – (u,v,w) coordinates.
freq (numpy.array((nchan,))) – Channel frequencies.
ms (numpy.array((nrows, nchan), dtype=numpy.complex64 or numpy.complex128)) – The input measurement set data. Its data type determines the precision used for the gridding.
weight (numpy.array((nrows, nchan), same precision as ms), optional) – If present, its values are used to multiply the input.
npix_x (int) – x dimension of the dirty image (see Note above).
npix_y (int) – y dimension of the dirty image (see Note above).
pixsize_x_rad (float) – Angular x pixel size (in radians) of the dirty image (see Note above).
pixsize_y_rad (float) – Angular y pixel size (in radians) of the dirty image (see Note above).
epsilon (float) – Accuracy at which the computation should be done. Must be larger than 2e-13. If ms has type numpy.complex64, it must be larger than 1e-5.
do_wstacking (bool) – If True, the full improved w-stacking algorithm is carried out, otherwise the w values are assumed to be zero.

Returns

The dirty image.

Return type

numpy.array((nxdirty, nydirty), dtype=float of same precision as ms)

cuda_nifty_gridder.dirty2ms(uvw, freq, dirty, weight, pixsize_x_rad, pixsize_y_rad, epsilon, do_wstacking=True)[source]

Converts a dirty image to visibilities.

Parameters

uvw (numpy.array((nrows, 3), dtype=numpy.float32 or numpy.float64)) – (u,v,w) coordinates.
freq (numpy.array((nchan,))) – Channel frequencies.
dirty (numpy.array((npix_x, npix_y), dtype=numpy.float32 or numpy.float64)) – The dirty image. Its data type determines the precision used for the degridding.
weight (numpy.array((nrows, nchan), same precision as dirty), optional) – If present, its values are used to multiply the input.
pixsize_x_rad (float) – Angular x pixel size (in radians) of the dirty image (see Note above).
pixsize_y_rad (float) – Angular y pixel size (in radians) of the dirty image (see Note above).
epsilon (float) – Accuracy at which the computation should be done. Must be larger than 2e-13. If dirty has type numpy.float32, it must be larger than 1e-5.
do_wstacking (bool) – If True, the full improved w-stacking algorithm is carried out, otherwise the w values are assumed to be zero.

Returns

The visibility data.

Return type

numpy.array((nrows, nchan), dtype=complex of same precision as dirty)

CUDA Nifty Gridder

This the user documentation for the CUDA Nifty Gridder.

Public API Documentation