But what’s the back pressure to this advice? If I need to multiply matrices should I write my own to avoid including MKL BLAS? What’s the heuristic that determines when a dependency is worthwhile?
I have written a matrix multiplication lib once because I had some a priori about the data and could remove half the operations.
Everything has a cost (usually in time) and everything comes with trade offs (usually a different set of bugs). Not-invented-here comes with a gigantic upfront time cost and a larger set of bugs. The best heuristic depends on what you’re optimizing for. There’s never an easy answer.
As usual, it depends. A naive matrix multiplication function is pretty simple. If that's all you need and can do with sub-optimal performance, I'd write my own.
Everything has a cost (usually in time) and everything comes with trade offs (usually a different set of bugs). Not-invented-here comes with a gigantic upfront time cost and a larger set of bugs. The best heuristic depends on what you’re optimizing for. There’s never an easy answer.