Hi,

I need to swap some large Eigen matrices (many times), say matrices m1 and m2. Using the standard c++ swap(m1,m2) works fine, but is slow because it does a copy.
When I use Eigen's swap routine m1.swap(m2), it works on small test cases, but becomes wrong for large matrices and many swaps!
The only possible explanation I have is that in order to use fftw, I allocate with fftw_malloc

Code: Select all

    m1f = (fftw_complex*) fftw_malloc(sizeof(fftw_complex) * n1*n2);


and then map to eigen

Code: Select all

   MatrixXcd m1 = MatrixXcd::Map(reinterpret_cast<complex<double>*>(m1f), n1, n2);

I also use #define EIGEN_DEFAULT_TO_ROW_MAJOR because fftw expects row-major format.
Could the mapping be a problem, and is there a way to solve this?

m1.swap(m2) with m1 and m2 Eigen::Matrix<> objects of the same type with "Dynamic" sizes does a fast shallow copy, i.e., only swap the pointers. Perhaps you tried with fixed size objects??

youre right, I was mixing some fixed-size objects, and the problem is not the eigen swap at all. sorry, and thanks for your help!