Implementing a mathematical function like sin or exp is tricky! Yet if you want to trade off accuracy and input range, for speed or vectorization, you'll need to do that, and likely once for each architecture you need an implementation for. In the process, you'll need to decide on range reduction strategies, working precisions, polynomial orders, and function-specific tricks. So my student Ian Briggs and I are working to make it easier. I don't want to get deep into the weeds here, so I'm jus...
