#CUNA v1.3 ##CUDA implementation of the NFFT adjoint operation

####(c) John Hoffman, 2016 ####[email protected]

Used for e.g. Lomb-Scargle periodograms, the adjoint operation is much like the familiar FFT, but can be applied to unequally spaced data

First, unevenly sampled data is smoothed onto an evenly-sampled grid with a filter (only the Gaussian filter is available for now). Fast Fourier transforms are then performed on the gridded data, and normalized by the filter's window function.

##Usage

Simply make to produce the shared libraries and testing binaries.

Alternatively, you may run make install to move the libraries to /usr/local/lib and headers to /usr/local/include.

Two libraries are produced: libcunaf.so and libcunad.so; these are identical libraries except libcunaf.so uses single precision throughout and libcunad.so uses double precision throughout.

The testing binaries allow you to do simple transforms and time the results. To see how to use them, simply run them without any command line arguments:

./test-single

##TODO

• More documentation

• Optimizations