I want to compile the following example based on Eigen's general eigenvalue solver (Eigen 3.3.3):
#include <iostream>
#include <Eigen/Eigenvalues>
int main()
{
Eigen::Matrix4f A;
A << -0.75, 0, -1.5, -1,
-1.25, 0, -1.5, -1,
-0.75, 0, -1, -1,
-0.25, 0, -1.5, -1;
Eigen::Matrix4f G;
G << 1, 0, 0, 0,
0, 1, 0, 0,
0, 0, 1, 0,
0, 0, 0, 0;
std::cout << "A = " << A << std::endl;
std::cout << "G = " << G << std::endl;
Eigen::GeneralizedEigenSolver<Eigen::Matrix4f> sol;
sol.compute(A.transpose()*A, G); // compute generalized eigenvalues
std::cout << "alphas = " << sol.alphas().transpose() << std::endl;
std::cout << "betas = " << sol.betas().transpose() << std::endl;
std::cout << "eigenvalues = " << sol.eigenvalues().transpose() << std::endl;
std::cout << "eigenvectors = " << sol.eigenvectors() << std::endl;
}
It takes 53 seconds to compile this small example, but it builds without complains with g++ on Ubuntu (gcc version 5.4.0, Ubuntu 16.04).
If I try to compile it with mingw-w64 (gcc version 6.2.0 (x86_64-win32-seh-rev1, Built by MinGW-W64 project)) on Windows, the compiler says
as.exe: main.o: too many sections (48774)
Assembler messages:
Fatal error: can't write main.o: File too big
so I tried to add the compiler flag -Wa,-mbig-obj like suggested in this stackoverflow question.
With the flag enabled mingw-w64 compiles the example, but I canceled the build during linking after 1 hour. I've also tried to compile the example from Eigen's documentation. It takes around 3 minutes but builds completely with mingw-w64.
Due to this long build times this solution is not applicable if I want to use the general eigenvalue solver in a real application.
It seams that this problem is related to the template structure of Eigen which leads to many symbols in a single source files. It looks like that mingw is not very comfortable with this.
Is there a way to optimize this situation with mingw?