On Fri, 8 Nov 2013, at 14:31, Gryllida wrote: > > > On Fri, 8 Nov 2013, at 14:08, Gryllida wrote: > > > > > > On Fri, 8 Nov 2013, at 10:09, Svetlana Tkachenko wrote: > > > ex2f.F involves a band matrix with 4 on main diagonal and -1s on the > > > sub-diagonals. I have A_central matrix which stores the -4s, and A_left, > > > A_right, Atop, A_bottom matrices which store the -1s (they actually are > > > more complicated numbers than that). Would it be recommended to simply > > > wrap the "do 10, II=..." block like the one in ex2f.F into $OMP DO to > > > speed up filling of the matrix as each of the values set can be set > > > independently of others? (I tried doing so but got a segmentation fault > > > and I'm working on a small example to reproduce the problem.) > > > > > > Svetlana > > > > I figured out this is called 'stencils' correctly and corrected the subject. > > Trying to use example ex22f.F but it's quite involved. > > > > Svetlana > > Found > http://scicomp.stackexchange.com/questions/2224/how-should-i-build-a-2d-5-point-stencil-laplacian-matrix-in-parallel/2228 > and will try to follow Jed Brown's advice.
(Again I wrote from the wrong e-mail. This should be fixed now.) The example has a line to compute matrix A from Ax = b, and a line to compute source term (vector b). > call KSPSetComputeOperators(ksp,ComputeMatrix,ctx,ierr) > call KSPSetComputeRHS(ksp,ComputeRHS,ctx,ierr) What do I use to give it the initial guess on x? Svetlana
