Wonton library provides low level infrastructure for wrappers to various unstructured mesh and data (state) managers. A wrapper is an implementation of an interface to the native interface of the mesh and state manager and can be used by libraries such as Portage and Tangram without copying data from the mesh and state manager.
The primary reason for Wonton's existence is to enable applications to interface to the Portage and Tangram libraries. Wonton also provides a common repository for these interface implementations as well as common definitions and methods.
To obtain a copy of wonton and its submodules from GitHub, clone recursively:
git clone --recursive https://github.com/laristra/wonton
If you are familiar with Docker, take a look at our Dockerfile for a working build environment. In particular, the Dockerfile builds off of the wonton-buildenv Dockerfile, and uses our travis.yml file with Travis CI.
Wonton uses standard C++11 features, so a fairly modern compiler is needed. We regularly test with Intel 18.0.1, GCC 6.4.0, and GCC 7.3.0. The build system requires CMake version 3.13+.
The following libraries are required:
- Either Boost (1.58.0+) or Thrust (1.8.3): We wrap some features of either one of these packages. If you would like to run with OpenMP or TBB threads, then you must use Thrust.
The following libraries are highly recommended:
- LAPACKE (3.8.0+)
Wonton provides wrappers for a few third-party mesh types. Building support for these is optional:
-
Jali:
We regularly test with version 1.1,4. You will need to set the
WONTON_ENABLE_Jali
andJali_ROOT
CMake variable if you wish to build support for Jali and its tests (see examples below). -
The Burton specialization in the
flecsi-sp
repository is built on top of FleCSI. You will need both projects to build support for the Burton mesh specialization and its tests. You will need to setWONTON_ENABLE_FleCSI=True
and add the FleCSI and FleCSI-sp install paths to theCMAKE_PREFIX_PATH
; see examples below. Both FleCSI packages are under constant development. This version of wonton is known to work with hashbd29de5d
of the FleCSI stable branch, and hashe78c594
of the FleCSI-SP stable branch.
The documentation is built using doxygen (1.8+).
In the simplest case where you have the appropriate versions mentioned above and Boost and LAPACKE are in the usual locations that CMake searches, then the build step is:
wonton $ mkdir build
wonton $ cd build
wonton/build $ cmake -DENABLE_UNIT_TESTS=True ..
wonton/build $ make
This compiles the serial code and about a dozen unit tests. To run the tests, execute
wonton/build $ ctest
If you wish to install the code into the CMAKE_INSTALL_PREFIX
then
execute
wonton/build $ make install
To build the documentation, one would configure with the
-DENABLE_DOXYGEN=True
flag, and then make doxygen
.
This project is licensed under a modified 3-clause BSD license - see the LICENSE file for details.
This software has been approved for open source release and has been assigned LA-CC-18-019.
Below we list copy & paste instructions for several local machines; we have a script that parses this README file to execute the examples below to ensure they build.
Execute the following from the Wonton root directory:
# machine=varan
export MODULEPATH=""
. /opt/local/packages/Modules/default/init/sh
module load intel/18.0.1 openmpi/2.1.2 cmake/3.14.0
NGC_INCLUDE_DIR=/usr/local/codes/ngc/private/include
JALI_INSTALL_PREFIX=/usr/local/codes/ngc/private/jali/1.1.4-intel-18.0.1-openmpi-2.1.2
LAPACKE_INSTALL_PREFIX=/usr/local/codes/ngc/private/lapack/3.8.0-patched-intel-18.0.1
mkdir build
cd build
cmake \
-D CMAKE_PREFIX_PATH=$NGC_INCLUDE_DIR \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-Wall -Werror" \
-D ENABLE_UNIT_TESTS=True \
-D WONTON_ENABLE_MPI=True \
-D ENABLE_JENKINS_OUTPUT=True \
-D WONTON_ENABLE_Jali=True \
-D Jali_ROOT:FILEPATH=$JALI_INSTALL_PREFIX \
-D WONTON_ENABLE_LAPACKE=True \
-D LAPACKE_ROOT:FILEPATH=$LAPACKE_INSTALL_PREFIX \
..
make -j2
ctest --output-on-failure
Execute the following from the Wonton root directory:
# machine=sn-fey
module load intel/18.0.5 openmpi/2.1.2 cmake/3.14.0
NGC=/usr/projects/ngc
NGC_INCLUDE_DIR=$NGC/private/include
JALI_INSTALL_PREFIX=$NGC/private/jali/1.1.4-intel-18.0.5-openmpi-2.1.2
LAPACKE_INSTALL_PREFIX=$NGC/private/lapack/3.8.0-patched-intel-18.0.5
mkdir build
cd build
cmake \
-D CMAKE_PREFIX_PATH=$NGC_INCLUDE_DIR \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-Wall -Werror" \
-D ENABLE_UNIT_TESTS=True \
-D WONTON_ENABLE_MPI=True \
-D ENABLE_JENKINS_OUTPUT=True \
-D WONTON_ENABLE_Jali=True \
-D Jali_ROOT:FILEPATH=$JALI_INSTALL_PREFIX \
-D WONTON_ENABLE_LAPACKE=True \
-D LAPACKE_ROOT:FILEPATH=$LAPACKE_INSTALL_PREFIX \
..
make -j36
ctest -j36 --output-on-failure