Usage

To use xtensor-blas functions, the xlinalg.hpp header has to be included. In the xt::linalg namespace, many of NumPy’s np.linalg functions are implemented. We make an effort to keep the interfaces very similar.

For example, calculating a determinant:

#include "xtensor-blas/xlinalg.hpp"

int main()
{
    xt::xarray<double> a = {{3, 2, 1}, {0, 4, 2}, {1, 3, 5}};
    auto d = xt::linalg::det(a);
    std::cout << d << std::endl;  // 42.0
}

We can also try to compute the same determinant using the slogdet function, which is more robust against under- or overflows by summing up the logarithm. The slogdet function in NumPy returns a tuple of (sign, val). In C++, we emulate the behaviour by returning a std::tuple, which can be unpacked using std::get<N>(tuple).

xt::xarray<double> a = {{3, 2, 1}, {0, 4, 2}, {1, 3, 5}};
auto d = xt::linalg::slogdet(a);
std::cout << std::get<0>(d) << ", " << std::get<1>(d) << std::endl;  // +1, 3.7376696

Returning tuples is used throughout the xlinalg package.

Using xblas and xlapack directly

It is not necessarily recommended to use xblas or xlapack directly, but it’s possible and can improve performance in certain cases. Some things have to be taken into consideration: For one thing, the result container needs to be allocated and passed into the function beforehand. And for the LAPACK functions, all arguments have to be in column_major order. Furthermore it’s required that the xexpressions are evaluated and are stored in contiguous memory. All of this is taken care of when using xlinalg.