(mention avoidance of bounds checks)
Revision as of 15:54, 19 January 2006
|Haskell Performance Resource|
1 General Array techniques
- Remember that ordinary arrays are monolithic, and individual elements are not mutable. In particular, the (//) operator copies the entire array, so it is rarely what you want. (Data.Array.Diff provides a variant of arrays with O(1) (//), but that library has performance problems of its own).
- Monolithic arrays are by no means useless! Powerful array-construction facilities like accumArray can often eliminate the need for truly mutable arrays.
- If you really need mutable arrays for speed, then if possible use the ST variant, so that the stateful part of your program can be encapsulated (Data.Array.ST).
- Avoid bounds checks by using Data.Array.Base.unsafeAt, Data.Array.Base.unsafeRead, Data.Array.Base.unsafeWrite (these are currently undocumented, unfortunately).
- See also Arrays, a thorough exploration of the various array types available in most Haskell compilers.
2 GHC-specific techniques
2.1 Use unboxed arrays (UArray, IOUArray)
GHC supports arrays of unboxed elements, for several basic arithmetic element types including Int and Char: see the Data.Array.Unboxed library library for details. Unboxed arrays support the same programmer interface as ordinary boxed arrays, so converting your code is easy. Using unboxed arrays will be a win in terms of both time and space.
There are also mutable unboxed arrays: IOUArray and STUArray (see Data.Array.IO and Data.Array.ST respectively). Using unboxed mutable arrays is often a good way to translate imperative algorithms into Haskell with similar performance.