I’m currently writing a project in Cython, and I’m looking for a linear algebra library that can be used without calling Python objects (=no GIL).
Ideally, I want it to be compiled in a simple pip with setup.py file. Because of this, I want the library to be installed without Makefile.
One option is a library with only a couple of *.h and *.c files, or even better, a single header file. In this case, I should include the files in my project and distribute them with my project, and furthermore, chances are I have to modify the codes and write interfaces myself, which would cause issues in some licenses. After I finish the software, I am not sure what license I’m going to release it (highly likely MIT, but I might not have the choice), so I should avoid GPL or other viral-licensed codes.
Another option is a Cython library that is already published in conda-forge which can directly manipulate C pointers or arrays. conda install is portable enough for the purpose, but ideally they should compiled against as many OSs and architectures as possible. At least, they should support x86-64 linux. One thing I was wondering is that if it would be easy enough to install custom BLAS using conda; I never tried BLAS before but it seems very well-defined standard. As long as I (or other users) don’t have to run Makefile, it would be fine.
Any recommendations? Thank you in advance!
 linear algebra library for scientific computing purpose in C/C++/Cython){kind=link}