When running psfgen, it outputs

Usage: psfgen [options] output.fits

psfgen is a program for creating a FITS file containing a TESS point spread function image

    -h, --help                show this help message and exit
    -g, --gamma=<double>      the width of the Cauchy point spread function, defaults to 0.5
    -s, --side-length=<int>   the oversampling of the CCD pixel in the output array, defaults to 3
    -o, --oversampling=<int>  P[���

It would be nice if there was a sandbox for doing development without a GPU (also, this is essentially required for Travis CI).

It looks like the way to compile and run CUDA on an x86 architecture is with PGI CUDA:

http://www.pgroup.com/resources/cuda-x86.htm

In our internal coordinate system, the origin is at pixel [0,0], and its geometric center is at coordinates [0.5, 0.5]. This is a natural system for our program's logic.

On the other hand, we output FITS. The geometric center of the origin pixel in a FITS image is implicitly [1.0,1.0]. We should probably conform to this in our input, too. Thus, we need to adjust input coordinates by subtracting [0.5,0.5] to make our internal coordinates.