Hello,

It seems to me that Eigen 3.3.3 (released February 21, 2017) has some bug in the Tensor module. I tested the following simple example (taken from the official README):

Code: Select all


#include <Eigen/Dense>
#include <unsupported/Eigen/CXX11/Tensor>
using namespace Eigen;
int main(int argc, char *argv[])
{
    // Map a tensor of ints on top of stack-allocated storage.   
    int storage[128];  // 2 x 4 x 2 x 8 = 128
    TensorMap<int, 4> t_4d(storage, 2, 4, 2, 8);
    return 0;
}


and got an error:

main.cpp:121: error: no matching function for call to ‘Eigen::TensorMap<int, 4>::TensorMap(int [128], int, int, int, int)’
TensorMap<int, 4> t_4d( storage, 2, 4, 2, 8 );
^

Since registration is disabled on Eigen Bugzilla, I could not be able to report this bug.

Could anyone here please help me to temporarily fix this bug in waiting for the next release? I really need to do TensorMap in my code. And please report this bug to the developers, too.

Thank you very much in advance for your help!

There is a typo in the doc. It should be:

Code: Select all

TensorMap<Tensor<int, 4> > t_4d(storage, 2, 4, 2, 8);

ggael wrote:There is a typo in the doc. It should be:

Code: Select all

TensorMap<Tensor<int, 4> > t_4d(storage, 2, 4, 2, 8);

Thanks, Gael. This does the trick. However, when changing the type from int to double:

Code: Select all

double storage[128];  // 2 x 4 x 2 x 8 = 128
TensorMap< Tensor<double, 4> > t_4d(storage, 2, 4, 2, 8);

I obtained the following errors:

error: no matching function for call to ‘Eigen::DSizes<long int, 4>::DSizes(const int [4])’
: m_data(dataPtr), m_dimensions(dimensions)
^
error: invalid conversion from ‘const int*’ to ‘long int’ [-fpermissive]
: m_data(dataPtr), m_dimensions(dimensions)
^

Could you please tell me how to fix this?

P/s: also thanks for creating my BugZilla account.

Sorry I have successfully compiled it. Not sure what happened. Thanks again, Gael!