Successive Over-Relaxation and other PDE solvers

[Nikhil Singh]

Currently we apply Inverse operator in DiffOper by directly solving the matrix in Fourier domain, but there are other solvers more amenable to eg boundary conditions and spatially varying differential operators.

[Sam Preston]

Fixed JacDet/Gradient/FiniteDiff bug at boundary. Appeared in GPU versions for images with odd sizes. Turns out the finiteDiff kernel was not checking to make sure the thread location was a valid image pixel. This closes issue #19

→ <<cset ea4f4b84cc15>>

[Sam Preston]

Accidentally closed this (meant to close issue #29)

[Jacob Hinkle]

Just a brief note. Sarang, Nikhil and I met yesterday. One thing we'll also want is to be able to enforce more complicated boundary conditions. For instance, by default we should use reflected boundary conditions on the image border for solving Poisson's equation using SOR. But we'd also like to be able to do this for an arbitrary boundary, such as a skull, represented either as a volumetric mask, KD tree, etc. The first step is to adapt Linh's old SOR code for the unmasked case.