NumPy-style broadcasting in Futhark

I’d strongly suggest not going the Numpy route, and instead reading about broadcast in julia [1]. In many contexts, it’s very important to distinguish between the element-wise application of a function, and a function that is applied to an entire array. A classic example would be the matrix exponential[2], versus the element-wise exponential of a matrix.

In julia the former is `exp(M)`, and the latter is `exp.(M)`

[1] https://julialang.org/blog/2017/01/moredots/

[2] https://en.m.wikipedia.org/wiki/Matrix_exponential

I actually think Julia's behaviour (partially) copied from matlab is bad. I had to review a lot of matlab/Julia simulation code write by PhD/grad students (as well as my own) and most of the time when things somehow are not working debugging becomes a task of "find the missing dot". If they would have decided on a more obvious symbol (the dot is really just way too easy to overlook) for broadcasting it would be much better, but I still much prefer that broadcast and matrix operation would use entirely different operators/named functions. Instead we have the case that a little annotation can dramatically change the behaviour of the code (the matrix exponential is in fact a great example).

That said I mainly work with fields and calculus so I might be biased because the majority of operations are broadcast and I only rarely use matrix operations.

The nice thing is that if this dot-approach (or some other syntax with a similar effect) were implemented in Futhark, then because Futhark is statically-typed, all mismatches would be compile-time errors. Those are very hard to overlook :)