Sounds good. I thought something like this could be the solution.
But it might be that my type-knowledge is not good enough - but I have not been able to do this. Example:
vec1 <- MV.new 1000 :: IO (V.MVector (PrimState IO) Int)
Data.Vector.singleton vec1 -- does not work, gives a lot of type-errs.
Felix
________________________________
Fra: Dmitry Dzhus
Hello haskellers I am struggling with this package: Data.Vector.Storable.Mutable I am creating vectors like this: MV.new 1000 :: IO (V.MVector (PrimState IO) Int) Now I would like access to this vectors in linear time, like I could have done in C using pointers. The problem is that I do not know which datastructure to keep my vectors in. If I put them in a list, the access is very slow. Even a binary tree is very slow compared to a C-version. I have tried to make an array og vector of storable-mutable vectors, but have not manged to accomplish this - even though there must be a haskell way :-)
You can use boxed vector as a first-level structure holding pointers you your storable vectors to perform O(1) indexing on first dimension. Alternatively, if your matrix is dense, store your data in a one-dimensional vector and use Repa as a thin wrapper to perform multi-dimensional indexing (see toIndex/fromIndex Data.Array.Repa.Shape).