On Fri, Feb 21, 2014 at 3:28 PM, Barry Smith <[email protected]> wrote:
> > Though ml does support 64 bit integers for long vectors they do it in a > fundamentally different way than PETSc; only certain integer data is stored > in 64 bit form while most integer values are stored in 32 bit integers. In > PETSc we manage all integers as either 32 or 64 bit so mixing the two > libraries in this case is not possible without a huge amount of work. > Sorry about this. > > You might consider using hype's Boomeramg or PETSc's GAMG algebraic > multigrid solvers instead, both of which can be used with 64 bit indices. > Thanks a lot for your suggestions. That is very helpful. Xiangdong > > Barry > > On Feb 21, 2014, at 1:15 PM, Xiangdong <[email protected]> wrote: > > > Hello everyone, > > > > When I compile petsc with options -with-64-bit-indices and ml packages, > I got the error like this: "Cannot use ml with 64 bit integers, it is not > coded for this capability". > > > > If I want to use ml as a preconditioner for long vectors (> 2 billions), > is there any solution? > > > > Thank you. > > > > Best, > > Xiangdong > > > >
