- Add corner flow test case. - The boundary conditions for the gradient of the pressure could be improved. Possible speed optimizations: - packing of parallel communications (in the case of domain composed of several boxes). - parallel match() uses doubles for cell layout description, guint would be more space efficient. - multiple calls to fl_domain_depth() are not needed => store maxlevel in FlDomain struct and update it only when necessary (i.e. when refining/coarsening etc...). Dodgy stuff: - Check that gfs_face_gradient() is used properly when not used for poisson problem (i.e. advection) (particularly in 3D with the factor of 2). Future stuff: - Richardson extrapolation => being able to compute advection terms one level above the leaf cells. - Adaptation given a constraint on maximum domain size (i.e. optimal distribution of a finite number of cells) (done 09/2003) - Generalize relaxation: (done 11/2002) * necessary for Crank-Nicholson * " for variable density (done 11/2002) - GFS -> OpenDX file conversion (done 11/2002, DX module). - Optimize parallel code by overlapping communications and computations: (done 2009) This could be done relatively easily (in Poisson relaxation for example) by first relaxing the cells close to the boundaries, start the communications, relax the remaining bulk cells then synchronize the communications.