I have not proved if the iterative process in
CubicInterpolation::interpolate() is identical to
(nframes * fabs(_actual_speed)), however ceil()
of it is empirically always larger.
Also, the cubic interpolation needs +2 samples.
Ardour3: Too close to call :)
git-svn-id: svn://localhost/ardour2/branches/3.0@13392 d708f5d6-7413-0410-9779-e7cbd77b26cf