Summary
In this chapter we have imported C functions as Haskell functions, exported Haskell functions as C functions, passed pointers (both foreign and stable) and data through the FFI, built a shared library with Haskell, and used hsc2hsto to write a Storable instance for a custom datatype. You have learned to invoke the FFI from both the C and the Haskell side and to manage memory in both the Haskell heap and the lower-level memory area also used by C.
The next chapter will be about another implementation-level concept like the FFI: GPU-programming using Haskell. Graphics processors are much better suited for highly parallel number-crunching applications, which is the reason for the GPU's popularity in high-performance numeric computing. An excellent Haskell library, Accelerate, defines a language that greatly simplifies usually mundane and hard GPU programming. In addition, Accelerate is backend-agnostic: the same code could run on any hardware solution (CPU/LLVM, CUDA, OpenCL, and others...
