Matrices

  • matrices created with gko::matrix::<format>::create(...)

  • sets executor

  • may provide dimensions

  • additional settings, e.g. Csr strategy, or max number of elements per row

  • fill matrix entries with read

  • alternative provide already filled arrays to create

    • arrays depend on format

  • matrices can be applied to vectors by apply

std::unique_ptr<gko::matrix::Dense<>> b = ...;
std::unique_ptr<gko::matrix::Dense<>> x = ...;
std::unique_ptr<gko::matrix::Dense<>> alpha = ...; // 1x1 vector
std::unique_ptr<gko::matrix::Dense<>> beta = ...; // 1x1 vector
gko::matrix_data<> md(...);

auto A = gko::matrix::Csr<>::create(exec);
A->read(md);

A->apply(b, x);               // x = A * b
A->apply(alpha, b, beta, x);  // x = alpha * A * b + beta * x

  • dimension checking in apply

auto b = gko::matrix::Dense<>::create(exec, gko::dim<2>{3, 2});
auto x = gko::matrix::Dense<>::create(exec, gko::dim<2>{5, 2});
auto A = gko::matrix::Csr<>::create(exec, gko::dim<2>{5, 3});
A->apply(b, x);   // works

auto b2 = gko::matrix::Dense<>::create(exec, gko::dim<2>{2, 3});
A->apply(b2, x);  // throws exception

  • implement interfaces

    • depends on matrix format which interfaces

    • csr most complete

Matrix Formats

The format with the most features is gko::matrix::Csr. Almost all other formats can be converted to and from Csr. Many algorithms also require their input as Csr matrices, so if a matrix in a different format is provided, it might get converted to Csr beforehand.