Vectorisation error with negative Double const

[ams]

Negative Double constants cause an error in parallel comprehensions. This function

negs xs = [: -1.0 | x <- xs :]

doesn't make it through the compiler; this

negs xs = [: (0.0 - 1.0) | x <- xs :]

does.

The error text is

*** Vectorisation error ***

Tycon not vectorised: GHC.Num.T:Num

[ams]

Version is actually 6.13.20090929.

[chak]

The current version of the vectoriser can't handle type classes yet; hence, the error.

BTW, in the current version of DPH, you need to use the simple mock Prelude from package dph in all code that ought to be vectorised — if you use the standard Prelude, you will get similar errors (even without using type classes). We will lift these restrictions in future versions.

[ams]

Thanks for the prompt response. Still a bit confused -- not seeing how type classes come into play with -1.0. Is it an overloaded constant? Even in that case, I've given its type as Double explicitly.

[rl]

This is because -1.0 gets desugared to negate 1.0 and negate is a method of Num. Unfortunately, Haskell doesn't have negative literals. It would be possible to add support to the vectoriser for this particular case but I'm not sure it's worth it since there is an easy workaround and the problem will go away once we are able to vectorise the prelude.

[rl]

I've opened #3577 as a placeholder for issues related to incomplete DPH support.