import fail -- Symbol not found: _dgetrf_ about xtensor-blas

[wolfv]

awesome, exactly that's what's needed. We usually use a cmake-based build system, so I responded with the CMake instructions, sorry.

We are planning to document much better how to properly link against BLAS and LAPACK.

Note that by just appending -llapack -lblas I am not sure which library your compiler will pick (i.e. the one in the conda environment or your systems one, if you have one installed). You might want to pass additional flags to make sure that you're picking the right one if you are concerned about that.

[SylvainCorlay]

Thanks for reporting. dgetrf is a symbol from BLAS< so it seems that it is not finding the BLAS library.

I will look into your example tomorrow (it is getting late here).

[wolfv]

Did you add something along the lines of

    find_package(BLAS REQUIRED)
    find_package(LAPACK REQUIRED)

    target_link_libraries(PUT_TARGET_NAME_HERE ${BLAS_LIBRARIES} ${LAPACK_LIBRARIES})

to your CMakeLists? It looks to me like you're not linking a BLAS implementation

[wolfv]

Or rather, LAPACK (since taking the inverse of a matrix calls the dgetrf routine of the LAPACK library).

[yashaach]

Sorry do you mean adding these during manual installation of xtensor-blas (and therefore do not use 'conda install -c conda-forge xtensor-blas") ? For building my own library it is still the same "python setup.py build" right?

[yashaach]

Oh I think the following method works -- add libraries=['lapack', 'blas'], to Extension in ext_modules, in setup.py of the xtensor-example-extension. I believe this adds the flags -llapack -lblas in the last step of the build when python setup.py build is called.

[wolfv]

I am going to close this issue and open a new one for the documentation.