Futhark is a purely functional data-parallel programming language. Its optimising compiler is able to compile it to typically very performant GPU code. The language and compiler is developed as part of the HIPERFIT project at DIKU. The project is still research-quality, and the focus has mostly been on ensuring high runtime-performance, so the source language still lacks many niceties for manual programming - it is fairly usable as a target for high-level code generation, however.
The incomplete documentation is here.
You will need GHC 7.8 or newer to compile Futhark.
Just run cabal install and executables by the names of futhark,
futhark-c, futhark-opencl and futharki will be installed in your
Cabal bin directory, most likely $HOME/.cabal/bin.
Otherwise, just run cabal configure, followed by cabal build, and
the executable can be found in dist/build/futhark/futhark.
Make sure that you have a recent Cabal to go with your GHC. A common
mistake is to only upgrade GHC, but use an old Cabal (i.e. older than
1.20; check with cabal --version). If you compile some of the
dependencies with an old Cabal by mistake, and then fail later in the
process, you will have to delete your ~/.ghc directory to remove the
broken libraries.
To compile a Futhark program to sequential C:
futhark-c prog.fut -o prog
Or maybe OpenCL:
futhark-opencl prog.fut -o prog
And then run it:
./prog < prog.input
To interpret a Futhark program:
futharki prog.fut < prog.input
We try to make use of Github issues, which are further organised on waffle.io.
For every commit, Futhark is automatically built and tested on Travis.
Run the data/runtests.sh script to check how well we're doing. Use
data/runtests.sh -t if you're in a hurry and only want the test
suite to do type checking.