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
Current iterative refinement and regularized solvers, e.g., reg_ldl::SolveAfter, individually solve for each of the right-hand sides or simply only support a single right-hand side. Given that Homogeneous Self-Dual embeddings of linear conic programs introduce an extra linear solve, which can happen simultaneously with another solve, it makes sense to avoid loading the sparse matrix data through the memory hierarchy twice as many times as necessary.
It might be worth having specialized implementations for one, two, and n right-hand sides to efficiently handle deflation.
The text was updated successfully, but these errors were encountered:
Current iterative refinement and regularized solvers, e.g.,
reg_ldl::SolveAfter
, individually solve for each of the right-hand sides or simply only support a single right-hand side. Given that Homogeneous Self-Dual embeddings of linear conic programs introduce an extra linear solve, which can happen simultaneously with another solve, it makes sense to avoid loading the sparse matrix data through the memory hierarchy twice as many times as necessary.It might be worth having specialized implementations for one, two, and n right-hand sides to efficiently handle deflation.
The text was updated successfully, but these errors were encountered: