Installation

To install MadNLP, simply proceed to

pkg> add MadNLP

By default, MadNLP uses the sparse linear solver MUMPS. In addition the user can install the following extensions to use a specialized linear solver.

GPU support

MadNLPGPU provides GPU support for MadNLP:

pkg> add MadNLPGPU

MadNLPGPU automatically detects if you have a NVIDIA or an AMD GPU on your machine, and adapts in consequence.

HSL linear solver

Obtain a license and download HSL_jll.jl. Install the package into your current environment using:

import Pkg
Pkg.develop(path = "/full/path/to/HSL_jll.jl")

If the user has already compiled the HSL solver library, one can simply override the path to the artifact by editing ~/.julia/artifacts/Overrides.toml

# replace HSL_jll artifact /usr/local/lib/libhsl.so
ecece3e2c69a413a0e935cf52e03a3ad5492e137 = "/usr/local"

Once HSL_jll has been installed on your machine, you can install MadNLPHSL as

pkg> add MadNLPHSL

Pardiso linear solver

To use Pardiso, the user needs to obtain the Pardiso shared libraries from https://panua.ch/, provide the absolute path to the shared library:

julia> ENV["MADNLP_PARDISO_LIBRARY_PATH"] = "/usr/lib/libpardiso600-GNU800-X86-64.so"

and place the license file in the home directory. After obtaining the library and the license file, run

pkg> add MadNLPPardiso
pkg> build MadNLPPardiso

The build process requires a C compiler.