Skip to content
Haskell FFI bindings for CUBLAS, CUSPARSE, and CuFFT
Haskell Shell M4
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


This Haskell library provides FFI bindings for the CUBLAS, CUSPARSE, and CuFFT CUDA C libraries. Template Haskell and language-c are used to automatically parse the C headers for the libraries and create the proper FFI declarations.

The main interfaces to use are Foreign.CUDA.Cublas for CUBLAS and Foreign.CUDA.Cusparse for CUSPARSE. There is some primitive marhsalling done between C types and Haskell types. For more direct FFI imports, use the Foreign.CUDA.Cublas.FFI and Foreign.CUDA.Cusparse.FFI modules.

The Cublas typeclass represents elements for which CUBLAS operations can be performed. Its instances are CFloat, CDouble, Complex CFloat, and Complex CDouble. Similarly, there is a Cusparse typeclass which has the same instances.


See the Haddock documentation.


First, CUDA and Autoconf should be installed. cublas- has been tested with CUDA version 6.5. Additionally, you may need to add some CUDA directories to your PATH and LD_LIBRARY_PATH environment variables.

Then, in the base directory, prepare a configure script by running

autoconf > configure

Then (also in the base directory),

cabal configure
cabal install

Note: in order for installation to work properly, nvcc must be found on the PATH, and if the nvcc on the PATH has directory structure


, then header files such as cublas.h should be found in

You can’t perform that action at this time.