Conversion between storablevector and stream-fusion lists with fusion http://www.haskell.org/haskellwiki/Storable_Vector
|Latest on Hackage:||0.0|
This package is not currently in any snapshots. If you're interested in using it, we recommend adding it to Stackage Nightly. Doing so will make builds more reliable, and allow stackage.org to host generated Haddocks.
This package brings together the best of two worlds: The flexibility of plain lists and speed of low-level arrays. Lists are lazy per element, thus allowing for elegant tying-the-knot algorithms and correct fusion of subsequent operations, and they support any element type, including functions. Storablevectors do not have these features. Instead they are fast, including very fast access via indices, they are memory efficient and allow simple exchange with C.
This package provides the canonical functions for conversion from StorableVector to Stream and back. By a simple fusion rule they let the interim Stream based lists disappear in many situations, resulting in fast low-level loops. Such fusion could not be correct on StorableVectors. E.g. consider
import qualified Data.StorableVector.Lazy as SV SV.zipWith f (SV.unfoldr size g a) (SV.cons b (SV.unfoldr size h c))
which yields a storable vector with the chunk structure
[1, size, size, ...]
and the following strictness behaviour:
For computation of the first value of the result,
the first chunk with size
SV.unfoldr size g a
has to be fully evaluated.
This has two advantages:
Firstly, you do not really want that behaviour,
but you accept it for the sake of overall performance.
Secondly, the odd behaviour cannot easily be preserved by fusion,
and we must resist to tell the optimizer incorrect rules.
So here is the solution: Write
import qualified Data.StorableVector.Lazy.Stream as SVG import qualified Data.List.Stream as Stream SVG.from chunkSize $ Stream.zipWith f (Stream.unfoldr g a) (Stream.cons b (Stream.unfoldr h c))
and get two advantages.
First: You do not have to pass the
size parameter at the leaves,
but only once at the top.
Second: Fusion jumps in and turns everything in a single efficient