Skip to content

Complexity Analysis Tool for Integer Programs

Notifications You must be signed in to change notification settings

aprove-developers/KoAT2-Releases

Repository files navigation

KoAT2

Reimplementation of KoAT (see here for more information) with control-flow refinement, multiphase ranking functions, runtime bounds by twn-loops, and size bounds by closed forms.

Build

Build directly from source code

  1. Make sure opam is installed
  2. Use opam init https://github.com/aprove-developers/opam-repository.git and follow the instructions to set up Opam and install OCaml if necessary
  3. opam install -j$(nproc) . --deps-only
  4. Run dune build to build KoAT2

Build Docker image

  1. Install docker as described here.
  2. Execute ./build_container.sh in the topdirectory of this repository. This builds a docker image with the name koat2. The docker image contains a static binary.

Static Binary

To get a statically linked binary, please execute the script compile_static_binary.sh. This will set up a docker container in which OCaml 4.09.1 for compiling static binaries using musl is installed with all required packages to compile KoAT2. Please note that in this case, the used libraries (see below) are linked statically into the resulting binary. The code of these libraries can be found when following the given links below.

Run

  1. After compiling the binaries are located in _build/install/bin.
  2. Run ./koat2 to get the help page of koat2. For further information run ./koat2 commandname --help. The main command proceeding a full analysis is the "analyse" command.
  3. The project can be built and run by dune directly with the command dune exec koat2 --, or you may export PATH=$PATH:"path-to-koat/_build/install/default/bin" (with the substituted path) into your .bashrc file to use koat2 directly.

Formatting with ocamlformat

We use ocamlformat for formatting purposes. Currently, we use version 0.26.1 which can be installed with opam install ocamlformat.0.26.1. Use dune build @fmt to run the formatter, which will output the suggested changes. You can accept these changes by executing dune promote. Alternatively, you can format the code and directly accept all changes with dune build @fmt --auto-promote.

The directory ./hooks/contains a pre-commit hook that automatically runs dune build @fmt. To activate it, simply copy ./hooks/pre-commit to ./.git/hooks/pre-commit.

Note that the CI automatically runs ocamlformat upon push to autoformat the code.

Tests & documentation

  1. Documentation is build via the dune target @doc. Execute the command dune build @doc. The resulting documentation can be found in _build/default/_doc/.
  2. Running tests is done with the command dune runtest

Links

Used Libraries

KoAT2 makes use of the following external tools and libraries.

External Tools

To generate an image of the graph of an integer program, KoAT2 invokes Graphviz, which has to be manually installed on the system. Note that when building the Docker image, the current Ubuntu binaries of Graphviz are installed. Thus, the resulting image already contains Graphviz.

Moreover, the Docker image also contains both clang and llvm2kittel which are used to transform C programs into the input format of KoAT.

Input Format

To analyze programs with KoAT2, they need to be represented as Integer Transition Systems. We use an extended version of KoAT's input format, which is also used in the category Complexity of Integer Transition Systems at the annual Termination and Complexity Competition.

In this extension, rules can be annotated with polynomial costs:

l1(A,B) -{A^2,A^2+B}> l2(A,B) :|: B >= 0

Here, A^2 and A^2+B are lower and upper bounds on the cost of the rule. The lower bound is ignored by KoAT2. The upper bound has to be non-negative for every model of the transition's guard.

At the moment, we do not support recursion for complexity analysis, i.e., we only support Com_1, which can also be omitted.

Contact

If you encounter any difficulties either use GitHub issues or contact aprove [at] i2.informatik.rwth-aachen.de