It is possible to use only symmetric matrices. In the OMP (Loki) version, we construct the symmetric matrices grad * R * grad, and grad * (1/R) * grad, and solve grad * R * grad U = R * W grad * (1/R) * grad A = (1/R) * C It would be worth it in the mpp version if ICCG is much faster than GMRES for symmetric matrices. /p/m3d/mh3d++/stell/m3dsea4-10-2.tar has omp version. The Makefile is set to run on seaborg, but it can also run on Loki.