I'm trying to figure out the 'best' (most efficient) way to permute a sparse matrix in Eigen given a permutation stored as a vector. For example, I currently have:

Code: Select all

// Permutation comes from somewhere (such as Cuthill-McKee reordering)
Eigen::VectorXi permutation = ...;
// Some matrix that we want to permute
Eigen::SparseMatrix<double> matrix_to_permute = ...;
// The result of the permutation
Eigen::SparseMatrix<double> output_matrix;
// Do the permutation
( matrix_to_permute.twistedBy( Eigen::PermutationMatrix<Eigen::Dynamic,Eigen::Dynamic,int>{ permutation }.inverse() ) ).evalTo( output_matrix );


Is this the proper 'Eigen' way to permute a sparse matrix? Will PermutationMatrix respect the sparsity (that is, it won't evaluate to a dense matrix here, will it)?

Thank you for your advices and tips.

You can avoid the copy using the asPermutation() method returning a permutation expression, and please do not use evalTo() which is rather an internal thing:

Code: Select all

out_mat = mat.twistedBy( perm_vec.asPermutation().inverse() );


And yes, of course this will both exploit and preserve the sparsity.

Many thanks, the code you have suggested works perfectly.