An application I’m writing involves a Python front-end with a C++ backend with Pybind11 to bind the two together. I rely on OpenBLAS at the C++ layer, but I need it installed in sequential mode without internal thread parallelization (I have several loops parallelized with OpenMP that make BLAS calls, and I don’t want the internal threading to conflict with the threading of the outer loop; these BLAS calls are sufficiently small that I would get no benefit with the internal parallelism anyway.
I can do this if I install OpenBLAS from scratch based on the Github repo using the flag USE_OPENMP=0 passed to cmake. Is it possible to specify such settings when installing a conda package?