Repository files navigation

CppADCodeGen is a C++ library that extendsCppADto allow the generation of C/C++ source code for computing the derivatives ofmathematical models using Algorithmic Differentiation (AD).Since CppAD uses operator-overloading and CppADCodeGen produces source-code,the result ishybrid Automatic Differentiation.

The evaluation of differential information can be orders of magnitude fasterto compute using a compiled model than using a regular operator-overloadingstrategy.

• Code Generation:
  • C/C++: It can be used to compute derivatives of functions/models (seewiki) and generate libraries (seewiki).
  • Latex: Latex sources can be used to create PDF files for documentation purposes (seewiki).
  • html+MathML: tml+MathML can be used to display your algorithm in a web browser (seewiki).
  • dot source-code: dot files can be used to create images with a graph of your model(seegraphviz).
• Support for Multiple Types: Allows the creation of CppAD models using double or float floating-point types.
• Efficient Sparsity Handling: Supports generation of code optimized for sparse Jacobians and Hessians.
• Differentiation index reduction of Differential Algebraic Equations (DAE) through:
  • Pantelides,
  • Soares-Secchi, and
  • Dummy Derivatives methods.
• Statically compile generated source-code at runtime can be using an existing compiler and linkeddynamically or, alternatively,
• Generated source-code can beJIT compilation using Clang/LLVM.

CppADCodeGen is available with both theEPL andGPL licenses(suitable for both open-source and closed-source commercial projects).See epl-v10.txt and gpl3.txt for a copy of the licenses.

CppADCodeGen is a C++14 header only library, therefore there aren't many dependencies:

• Required:
  • CppAD (2024),
  • AC++14 compiler (such as GCC and Clang),
• Optional:
  • Clang/LLVM (only required for JIT compilation; supported versions <= v10.0), and
  • Eigen 3 (required when DAE differentiation index reduction is used).

Runtime compilation and dynamic linking:

• Linux

It might be very easy to support other OSes, but it is not implemented yet.

Get the sources from GitHub:

git clone https://github.com/joaoleal/CppADCodeGen.git CppADCodeGen

Create a new folder to build the project:

mkdir cppadcg-build

Build the project (no compilation of C/C++ occurs, just generation of header files):

cd cppadcg-buildcmake ../CppADCodeGen

Either install the project in your system:

make install

or to some other folder:

make DESTDIR=someotherfolder install

A debian installation package can be created at the root of the project.Typically, you can create the installer by just typing:

dpkg-buildpackage

It will create a debian package outside the project's folder.

See thewiki.

The folder example includes some simple use cases.

Get the sources from GitHub:

git clone https://github.com/joaoleal/CppADCodeGen.git CppADCodeGen

Create a new folder for the tests:

cd make-build-debugcmake ../CppADCodeGen

Testing requiresgoogle-test (version 1.14.0) which will be downloaded from GitHub.

Then compile the tests:

make build_tests

Run the complete set of tests:

maketest

Ifvalgrind is available in your system, CppADCodeGen will also perform memory checks which canlead to a very lengthy test execution.It is possible to disable memory validations by turning off the CMake optionUSE_VALGRIND.For instance, by calling the following command before running the tests:

cmake -DUSE_VALGRIND=OFF ../CppADCodeGen