Empirical evaluation code and demos for our FHPC'16 paper

Benchmarks are in benchmarks, demos in demos.

To run these programs you must have:

1. the Futhark compiler futhark-opencl must be in your $PATH,

2. the apltail compiler

3. the Tail-to-Futhark compiler tail2futhark must be in you $PATH For example, after cloning the repository you can install it by $ stack setup $ stack install

4. a working OpenCL setup.

5. gawk (a version that supports the strtonum function)

6. matplotlib

You will need a *nix-like system. OpenCL execution will by default use the first detected platform and device. You can edit the Makefile to select a different platform and device.

The benchmark system is built using make (sorry). Ideally, you just run make and it will build and run all benchmarks. The result will be a speedup graph in a file called plot.pdf, and a runtime table in a file table.tex. There will also be a directory called runtimes/ that contains files with runtime measurements (in miliseconds).

Of course, many things can go wrong. You may need to modify the Makefile to fix include paths and the like to fit your system.

Problems

If you are on a 64-bit system, then you are adviced to compile apltail with MLton, not MLKit.

You must set the TAIL_ROOT environment variable to point at a local clone of the apltail repository. This is so aplt can find the TAIL prelude.