The Haskell Unfolder Episode 38: tasting and testing CUDA (map, fold, scan)

An introduction to linking CUDA libraries with Haskell, calling CUDA functions, and using QuickCheck to test GPU implementations of map, fold, and scan.

CUDA is an extension of C for programming NVIDIA GPUs. In this episode of the Haskell Unfolder we show how to set up a CUDA library so that we can link to it from a Haskell application, how we can call CUDA functions from Haskell, and how we can use QuickCheck to find subtle bugs in our CUDA code. On the CUDA side, we show how to implement simple concurrent versions of `map`, `fold` and `scan`. No familiarity with CUDA will be assumed, but of course we will only be able to give a taste of CUDA programming.