Philip Munksgaard

It currently tunes all entrypoints, doesn't it? Also, when given a tuning file that doesn't contain all the tuning parameters for a program, it fails, right? If we were to...

> Currently, only the entry point named "main" is tuned. The threshold parameters irrelevant to main are set to 1. Hence, it becomes rather tedious to tune a program with...

> It says so in the official [documentation](https://futhark.readthedocs.io/en/latest/man/futhark-autotune.html). So it is more of a feature than a bug. You can try this [gist](https://gist.github.com/mortvest/28505ca1cba398226fb5b829c7c71a3c). I see. The documentation is wrong, you...

My initial reaction is that it will make it harder to infer anything useful about the index functions of the arguments. In particular, if `is` and `js` are computed separately,...

This is what we're looking for: ```futhark let iota_2d (n: i64) (m: i64) : [n][m](i64, i64) = tabulate_2d n m (\i j -> (i, j)) let blockify [n] 't (b:...

> `if` is poison. Why do we need that? `if` is needed because the first element of the antidiagonal is either along the top (first row) of the matrix, or...

Actually, upon further consideration, we should probably not allow the `echo "iota 42" | futhark run -` form, since standard input is usually used for input to the program being...

Depending on how we want to tackle this, I don't think the token stream is that useful. Other code formatters for expression-based languages such as [elm-format](https://github.com/avh4/elm-format) and [ormolu](https://github.com/tweag/ormolu/) use user-introduced...

> I would prefer a formatter that does not touch linebreaks. I am unsure what you mean by this. > The token stream is useful because each token is associated...

Ah, I understand. Yes, that's definitely a simpler option and it obviates most of my concerns. Personally, I don't mind the many linebreaks in ormolu- or elm-formatted code, but I...