You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In some instances an application generating symmetric matrices in a lower/upper triangle storage format may use Re::Solve's LU solvers. In that case, the matrix needs to be expanded to the full square matrix format. It would be helpful to have expandSymmetricMatrix function available to make such transformation when needed.
First to decide is where this function should be implemented:
As a member of matrix::Sparse class?
As a member of matrix::MatrixHandler?
As a standalone function in matrix namespace.
Several implementation details that need to be decided:
When to allocate device memory for the expanded matrix?
What operations to carry out on the host and what on the device?
How to reuse sparsity pattern of the expanded matrix over several linear solver calls?
actually, i suppose that doesn't totally resolve the issue since ReSolve::matrix::expand is currently only host-side. it still needs to be decided whether there's a need for gpu-related functionality here
In some instances an application generating symmetric matrices in a lower/upper triangle storage format may use Re::Solve's LU solvers. In that case, the matrix needs to be expanded to the full square matrix format. It would be helpful to have
expandSymmetricMatrix
function available to make such transformation when needed.First to decide is where this function should be implemented:
matrix::Sparse
class?matrix::MatrixHandler
?matrix
namespace.Several implementation details that need to be decided:
CC @kswirydo @maksud
The text was updated successfully, but these errors were encountered: