Skip to content

Latest commit

 

History

History
276 lines (211 loc) · 8.21 KB

README.md

File metadata and controls

276 lines (211 loc) · 8.21 KB
Raw

Build Status codecov.io

portage

The portage library provides a framework for general purpose data remapping - between meshes, between particles, or between meshes and particles - in computational physics applications. Remapping is facilitated through the use of user-supplied wrappers around meshes/particle swarms with their data. The remap algorithm is organized in three phases operating on the wrappers corresponding to the original mesh/particles: search for intersection candidates, calculate the intersection with candidates, then interpolate the results onto the new mesh or particle swarm. Algorithms for each of the phases can be customized (e.g. order of accuracy of the interpolation) and, through the wrappers, take advantage of hybrid parallelism (MPI+X).

Getting Started

To obtain a copy of portage and its submodules from GitHub, clone recursively:

git clone --recursive https://github.com/laristra/portage

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 portage-buildenv Dockerfile, and uses our travis.yml file with Travis CI.

Prerequisites

Portage 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.
Utilizing the full capabilities of portage will require an MPI implementation; we regularly test with OpenMPI 2.1.2 The build system requires CMake version 3.13+.

The following libraries are also required (see examples below):

  • Wonton

    Wonton is a utility library for Portage containing some commonly used classes like Point, Vector, etc., some abstractions for on-node parallelism and also some mesh and state wrappers. Wonton itself has dependencies - the highly recommended one is LAPACKE (3.8.0+). On-node parallelism in Portage requires that Wonton be built with NVidia Thrust or Kokkos. Distributed parallelism requires that Wonton be built with MPI enabled. In the absence of the Thrust library, the Boost library must be linked into Wonton. See the Wonton README for details.

    If you specify, PORTAGE_ENABLE_TANGRAM and TANGRAM_ROOT, Wonton will be picked up automatically as Wonton is a dependency of Tangram as well. If not, you must specify the path to Wonton as WONTON_ROOT

The following libraries are required if multi-material remapping is to be enabled:

  • Tangram

    Tangram is a material interface reconstruction library that is used to correctly remap between meshes with fractional amounts of multiple materials in some cells. In the future Tangram may become a required component of Portage.

The documentation is built using doxygen (1.8+).

For more details regarding CMake settings, see the documentation page.

Installing

In the simplest case the build step is:

portage $ mkdir build
portage $ cd build
portage/build $ cmake -DENABLE_APP_TESTS=True -DWONTON_ROOT:/path/to/wonton/installation ..
portage/build $ make

This compiles the serial code and about a dozen application tests. To run the tests, simply execute

portage/build $ make test

If you wish to install the code into the CMAKE_INSTALL_PREFIX then simply execute

portage/build $ make install

To build the documentation, one would configure with the -DENABLE_DOXYGEN=True flag, and then make doxygen.

See the examples below, or the documentation for more build instructions.

License

This project is licensed under a modified 3-clause BSD license - see the LICENSE file for details.

Release

This software has been approved for open source release and has been assigned LA-CC-16-084.

Example builds

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. NOTE: If TANGRAM is DISABLED, the path to WONTON must be specified explicitly using WONTON_ROOT or indicated in CMAKE_PREFIX_PATH.

Darwin

Execute the following from the portage root directory:

# machine=darwin-fe

INTEL_VERSION=18.0.3
MPI_VERSION=3.1.3
TANGRAM_VERSION=1.0.1
WONTON_VERSION=1.2.2

TPL_PREFIX=/usr/projects/ngc/private

# load the correct boost, compiler, and openmpi
module purge
module load cmake/3.15.3 openmpi/${MPI_VERSION}-intel_${INTEL_VERSION} boost/${BOOST_VERSION}

cmake \
    -D CMAKE_BUILD_TYPE=Release \
    -D ENABLE_UNIT_TESTS=True \
    -D ENABLE_APP_TESTS=True \
    -D ENABLE_MPI=True \
    -D PORTAGE_ENABLE_TANGRAM=True \
    -D WONTON_ROOT:FILEPATH=${TPL_PREFIX}/wonton/${WONTON_VERSION}-intel-${INTEL_VERSION}-openmpi-${MPI_VERSION} \
    -D TANGRAM_ROOT:FILEPATH=${TPL_PREFIX}/tangram/${TANGRAM_VERSION}-intel-${INTEL_VERSION}-openmpi-${MPI_VERSION} \
    -D PORTAGE_ENABLE_MPI=True \
    -D PORTAGE_ENABLE_THRUST=False \
	-D PORTAGE_ENABLE_Jali=True \
	-D PORTAGE_ENABLE_FleCSI=False \
    ..

make -j16
ctest -j16 --output-on-failure

Snow

Execute the following from the portage root directory:

# machine=sn-fey
. /usr/share/lmod/lmod/init/sh

INTEL_VERSION=18.0.5
MPI_VERSION=2.1.2
WONTON_VERSION=dev
TANGRAM_VERSION=dev

TPL_PREFIX=/usr/projects/ngc/private

module load intel/${INTEL_VERSION} openmpi/${MPI_VERSION} cmake/3.14.6


mkdir build
cd build
cmake \
    -D CMAKE_BUILD_TYPE=Release \
    -D ENABLE_UNIT_TESTS=True \
    -D ENABLE_APP_TESTS=True \
    -D PORTAGE_ENABLE_TANGRAM=True \
    -D WONTON_ROOT:FILEPATH=${TPL_PREFIX}/wonton/${WONTON_VERSION}-intel-${INTEL_VERSION}-openmpi-${MPI_VERSION} \
	-D TANGRAM_ROOT:FILEPATH=${TPL_PREFIX}/tangram/${TANGRAM_VERSION}-intel-${INTEL_VERSION}-openmpi-${MPI_VERSION} \
    -D PORTAGE_ENABLE_MPI=True \
	-D PORTAGE_ENABLE_Jali=True \
	-D PORTAGE_ENABLE_FleCSI=False \
    ..
make -j4
ctest -j4 --output-on-failure

If you want to build an app for performance testing, you should include Thrust and TCMalloc in your build. The cmake command for this is:

# machine=sn-fey::thrust
. /usr/share/lmod/lmod/init/sh

INTEL_VERSION=18.0.5
MPI_VERSION=2.1.2
WONTON_VERSION="dev"
TANGRAM_VERSION="dev"

TPL_PREFIX=/usr/projects/ngc/private

module load intel/${INTEL_VERSION} openmpi/${MPI_VERSION} cmake/3.14.6
mkdir build-thrust
cd build-thrust
cmake \
   -D CMAKE_BUILD_TYPE=Release \
   -D ENABLE_UNIT_TESTS=True \
   -D ENABLE_APP_TESTS=True \
   -D PORTAGE_ENABLE_TANGRAM=True \
   -D WONTON_ROOT:FILEPATH=${TPL_PREFIX}/wonton/${WONTON_VERSION}-intel-${INTEL_VERSION}-openmpi-${MPI_VERSION}-thrust \
   -D TANGRAM_ROOT:FILEPATH=${TPL_PREFIX}/tangram/${TANGRAM_VERSION}-intel-${INTEL_VERSION}-openmpi-${MPI_VERSION}-thrust \
   -D PORTAGE_ENABLE_MPI=True \
   -D PORTAGE_ENABLE_THRUST=True \
   -D PORTAGE_ENABLE_Jali=True \
   -D PORTAGE_ENABLE_FleCSI=False \
   ..
make -j4
ctest -j4 --output-on-failure

Varan

Execute the following from the portage root directory:

# machine=varan
export MODULEPATH=""
. /opt/local/packages/Modules/default/init/sh

INTEL_VERSION=18.0.1
MPI_VERSION=2.1.2
WONTON_VERSION="dev"
TANGRAM_VERSION="dev"

TPL_PREFIX=/usr/local/codes/ngc/private

module load intel/${INTEL_VERSION} openmpi/${MPI_VERSION} cmake/3.14.0
mkdir build
cd build
cmake \
    -D CMAKE_BUILD_TYPE=Debug \
    -D ENABLE_UNIT_TESTS=True \
    -D ENABLE_APP_TESTS=True \
    -D PORTAGE_ENABLE_TANGRAM=True \
    -D WONTON_ROOT:FILEPATH=${TPL_PREFIX}/wonton/${WONTON_VERSION}-intel-${INTEL_VERSION}-openmpi-${MPI_VERSION} \
    -D TANGRAM_ROOT:FILEPATH=${TPL_PREFIX}/tangram/${TANGRAM_VERSION}-intel-${INTEL_VERSION}-openmpi-${MPI_VERSION} \
    -D PORTAGE_ENABLE_MPI=True \
	-D PORTAGE_ENABLE_Jali=True \
	-D PORTAGE_ENABLE_FleCSI=False \
    ..
make -j2
ctest -j2 --output-on-failure