Yarr is a new blazing fast dataflow framework (array library), mainly intented to process Storables (including all "primitive" numeric types) and fixed-vectors of them, for example coordinates, color components, complex numbers.

Yarr framework is inspired by repa library and inherits its features, including shape-polymorphism and auto-parallelism. Additionaly, the framework is polymorphic over type and arity of fixed-size vectors and supports neat flow operations over them. For example, you can convert colored image to greyscale like this:

let greyImage = zipElems (\r g b -> 0.21 * r + 0.71 * g + 0.07 * b) image

In some cases the library is considerably faster than repa. See benchmark results: https://github.com/leventov/yarr/blob/master/tests/bench-results.md

Shortcoming by design: lack of pure indexing interface.

To start with, read documentation in the root module: Data.Yarr.

Yarr!

Changes

  • Changes in version 0.9.2:
    • Safe folds – see “Data.Yarr.Walk”
    • Issue with slice-wise loading with unrolled filling function solved
  • Changes in 1.3 (ex. 0.9.3):
    • IFusion – mapping/zipping with index
    • Rename: Work -> Walk
  • Changes in 1.3.3.1: This now compiles with ghc 7.8.3 but probably won’t compile with earlier versions of ghc. Let me know if this is a requirement and I will see what I can do. Even better submit a pull request.
  • Changes in 1.4.0.0: This works (and will only work) with the breaking change release of fixed-vector 0.8.0.0.
  • Changes in 1.4.0.2: Build with GHC 7.10 and 8.0.
comments powered byDisqus