yes, I call it horrible:
- too many arguments
- non-speaking names
It competes with something as simple as linalg.eigh(a).
In many cases a few percent performance gains that can be achieved by people who take details of a specific computer setup into account and spend time on performance measurements don’t compensate the extra work. I personally rather wait 20 minutes longer for a computation than spending time on coding.
Dongarra et al. knew what they were doing.
yes, they were masters of the tools of their time and the success of LAPACK is impressive. The same holds for Konrad Zuse. Still, I do not want to program a Z3.