Distributed Solvers

Many of Ginkgo’s solver support distributed systems. For them, solving a distributed system is exactly the same as for a non-distributed system. Only the solver has to be generated from a gko::experimental::distributed::Matrix, and then applied to gko::experimental::distributed::Vectors. If a preconditioner is used, it also has to support distributed systems. A brief example is shown here:

std::shared_ptr<gko::experimental::distributed::Matrix<>> A = ...;
std::unique_ptr<gko::experimental::distributed::Vector<>> b = ...;
std::unique_ptr<gko::experimental::distributed::Vector<>> x = ...;

auto solver = gko::solver::Cg<>::build().on(exec)->generate(A);
solver->apply(b, x);

List of Supported Solvers

These solvers are available for distributed systems.

• BiCGStab

• (F)GMRES

• (F)CG

• CGS

• GCR

• IR

• MultiGrid, depends on the parameters

List of Unsupported Solvers

These solvers are not available for distributed systems.

• BiCG

• CB-GMRES

• Direct

• IDR

• triangular solver