Somehow support evaluation of cubic splines

This is needed for the Richards project with Olaf. The nonlinearity in Richards is so flopintensive to compute, that people typically use interpolation tables instead. Vectorizing the evaluation is done the following:

• Equidistant nodes allow vector index calculation through division and truncation
• Nasty gather instructions load the needed coefficients
• Three mul_adds do the actual evaluation (Horner) My guess is that the gathers will completely dominate the work, but we will see.