We, the authors, would like to dispute the result.

The reason for the build failure was, as suggested in the "build_notes.txt",
that the incorrect version of CUDA was used (5.5 was used, but 3.1 was required).
Using the correct version of CUDA, the tool builds and runs successfully for us.

Whilst we applaud the aims of the original study, we would like to point out
that in a study of reproducibility, one ought to take care to reproduce the
original experiment exactly. Not using the right version of dependencies is just
an example of the poor experimental practices in computer science they're railing
against.