Data.Array.Accelerate defines an embedded language of array computations for high-performance computing in Haskell. Computations on multi-dimensional, regular arrays are expressed in the form of parameterised collective operations (such as maps, reductions, and permutations). These computations are online-compiled and executed on a range of architectures.
For more details, see our recent paper Accelerating Haskell Array Codes with Multicore GPUs. There are also some slightly outdated slides and a video of a talk that I gave at the Haskell Implementors Workshop 2009 (in Edinburgh): Haskell Arrays, Accelerated (Using GPUs).
As a simple example, consider the computation of a dot product of two vectors of single-precision floating-point numbers:
dotp :: Acc (Vector Float) -> Acc (Vector Float) -> Acc (Scalar Float)
dotp xs ys = fold (+) 0 (zipWith (*) xs ys)
Except for the type, this code is almost the same as the corresponding Haskell code on lists of floats. The types indicate that the computation may be online-compiled for performance — for example, using Data.Array.Accelerate.CUDA.run it may be on-the-fly off-loaded to a GPU.
Package accelerate is available from
- Hackage: accelerate — install with
cabal install accelerate - GitHub: mchakravarty/accelerate - get the source with
git clone https://github.com/mchakravarty/accelerate.git
- Glasgow Haskell Compiler (GHC), 6.12.1 or later
- Haskell libraries as specified in
accelerate.cabal - For the CUDA backend, CUDA version 3.0 or later
The maintainer of this package is Manuel M T Chakravarty [email protected] (aka TacticalGrace on #haskell and related channels).
Both user and developer questions and discussions are welcome at [email protected]. Sorry, this mailing list is temporarily unavailable.
Here is a list of features that are currently missing:
- Reification of sharing in scalar expressions (sharing is being recovered for array computations)
- The CUDA backend does not support arrays of type Char and Bool at the moment.
- The CUDA backend does not implement stencil computations yet.
- Preliminary API (the current functionality is limited)
Haddock documentation is included in the package and linked from the Hackage page. Furthermore, the source package contains a few simple examples in the accelerate-examples/ directory.
The idea behind the HOAS (higher-order abstract syntax) to de-Bruijn conversion used in the library is described separately.