Skip to content

A set of radiation transport mini-applications used for performance optimization on HPC systems.

License

Notifications You must be signed in to change notification settings

ORNL-CEES/Profugus

Repository files navigation

Description

Profugus is an open-source mini-application (mini-app) for radiation transport and reactor applications. It contains the fundamental computational kernels used in the Exnihilo code suite from Oak Ridge National Laboratory. However, Exnihilo is production code with a substantial user base. Furthermore, Exnihilo is export controlled. This makes collaboration with computer scientists and computer engineers difficult. Profugus is designed to bridge that gap. By encapsulating the core numerical algorithms in an abbreviated code base that is open-source, computer scientists can analyze the algorithms and easily make code-architectural changes to test performance without compromising the production code values of Exnihilo.

Profugus is not meant to be production software with respect to problem analysis. The computational kernels in Profugus are designed to analyze performance, not correctness. Nonetheless, users of Profugus can setup and run problems with enough real-world features to be useful as proof-of-concept for actual production work.

Profugus is also used as a test-bed mini-app for the ASCR projects XPRESS and MCREX.

Profugus Development Team

Profugus contains the computational kernels from the Exnihilo code base. The core Exnihilo development team consists of the following scientists (listed alphabetically)

Profugus is developed and maintained by the following team:

Profugus Packages

Profugus constains the following code mini-apps:

SPn
Mini-app of Simplified Spherical Harmonics ("SPn") computational kernel.
MC
Mini-app of Monte Carlo computational kernel.

Documentation on the mini-apps (methods and execution) are given in Running The Mini-Applications. There are several support packages that are used by the mini-apps. These are:

Utils
Utilities and testing framework used by other packages.
Matprop
Data structures for storing material properties (cross-sections).
Minisweep
Although contained in a separate code repository, the Profugus project also includes the Minisweep application developed by Wayne Joubert <joubert@ornl.gov>. Minisweep is a deterministic Sn radiation transport miniapp used for performance optimization and performance model evaluation on HPC architectures. The Minisweep repository can be found at https://github.com/wdj/minisweep.

Profugus is designed to build and run with a minimum of dependencies. However, there are some requirements. The third-party software (TPLs) necessary to build Profugus is all open-source and freely available. The TPLs for Profugus are listed in the following table:

TPL Required Comments
TriBITS Yes TriBITS is the Trilinos build system
Trilinos Yes Trilinos is an open-source solvers
BLAS/LAPACK Yes Use vendor-specific implementation
MPI No OpenMPI and MPICH are suggested
HDF5 No Parallel HDF5 will allow output of

Building The Code

The most straightforward method for building Profugus is to use the scripts in Profugus/install. Profugus uses the TriBITS build system. This system is a set of package-based extensions to standard cmake. So, first you need to obtain Trilinos and TriBITS and put them in your top-level Profugus directory:

> cd Profugus
> git clone https://github.com/TriBITSPub/TriBITS.git
> ln -s $PATH_TO_TRILINOS .

The preferred mechanism for using the build scripts is to make a target directory where the build is to be performed:

> pwd
  /home/me
> mkdir debug
> cd debug
> mkdir target
> cd target
> pwd
  /home/me/debug/target

The install directory contains several example build scripts. General options for all platforms (which can be overridden at configure time) are specified in the install/base.cmake:

##---------------------------------------------------------------------------##
## CMAKE BASE FILE
##---------------------------------------------------------------------------##

# Default build all packages
SET(Profugus_ENABLE_Utils   ON CACHE BOOL "")
SET(Profugus_ENABLE_Matprop ON CACHE BOOL "")
SET(Profugus_ENABLE_SPn     ON CACHE BOOL "")
SET(Profugus_ENABLE_MC      ON CACHE BOOL "")

# Turn on tests
SET(Profugus_ENABLE_TESTS ON CACHE BOOL "")
SET(Profugus_TEST_CATEGORIES "BASIC" CACHE STRING "")

# Turn on SS code and optional packages by default
SET(Profugus_ENABLE_ALL_FORWARD_DEP_PACKAGES OFF CACHE BOOL "")
SET(Profugus_ENABLE_ALL_OPTIONAL_PACKAGES    ON  CACHE BOOL "")
SET(Profugus_ENABLE_SECONDARY_STABLE_CODE    ON  CACHE BOOL "")

# Up the max num procs
SET(MPI_EXEC_MAX_NUMPROCS 8 CACHE STRING "")

# Turn off binutils
SET(Teuchos_ENABLE_BinUtils OFF CACHE BOOL "")

# Turn off Zoltan2
SET(Profugus_ENABLE_Zoltan2 OFF CACHE BOOL "")

# Compiler options
SET(BUILD_SHARED_LIBS ON CACHE BOOL "")
SET(CMAKE_CXX_FLAGS "-std=c++11 -Wno-deprecated-declarations" CACHE STRING "")
SET(Profugus_ENABLE_CXX11:BOOL=ON)

# TriBITS stuff
SET(Profugus_ENABLE_INSTALL_CMAKE_CONFIG_FILES OFF CACHE BOOL "")
SET(Profugus_DEPS_XML_OUTPUT_FILE "" CACHE FILEPATH "")

By default, all of the packages inside of Profugus are turned on. Furthermore, C++-11 is required. The default options specify the appropriate compiler flags for gcc. The tests are also turned on by default; to disable tests in any upstream package simply do not explicitly ENABLE that package. For example, to build the SPn package and all of its tests but only include required source from upstream packages, the user would specify:

SET(Profugus_ENABLE_SPn ON CACHE BOOL "")

In this case, only the pieces of Utils needed to build SPn are compiled. All tests can be turned off by setting Profugus_ENABLE_TESTS to OFF.

The install directory contains several build scripts that are all suffixed by the platform name. For example, to build on a Linux x86_64 system the "install/cmake_x86_64.sh" script can be used:

#!/bin/sh
##---------------------------------------------------------------------------##
## CMAKE FOR X86_64
##---------------------------------------------------------------------------##

# CLEANUP
rm -rf CMakeCache.txt
rm -rf CMakeFiles

# SOURCE AND INSTALL
SOURCE=<SET_SOURCE_DIR>
INSTALL=<SET_INSTALL_DIR>

# BUILD OPTIONS
BUILD="DEBUG"
MPI="ON"

# TPL PATHS
HDF5_PATH="/vendors/hdf5_parallel"
MPI_PATH="/opt/openmpi/gcc/current"

##---------------------------------------------------------------------------##

cmake \
-DCMAKE_BUILD_TYPE:STRING="$BUILD" \
-DTPL_ENABLE_MPI:BOOL=$MPI \
-DCMAKE_INSTALL_PREFIX:PATH=$INSTALL \
\
-DMPI_BASE_DIR:PATH=$MPI_PATH \
\
-DTPL_ENABLE_HDF5:BOOL=ON \
-DHDF5_INCLUDE_DIRS:PATH=$HDF5_PATH/include \
-DHDF5_LIBRARY_DIRS:PATH=$HDF5_PATH/lib \
\
-DBLAS_LIBRARY_DIRS:PATH=/vendors/gcc/atlas/lib \
-DLAPACK_LIBRARY_DIRS:PATH=/vendors/gcc/atlas/lib \
-DBLAS_LIBRARY_NAMES:STRING="f77blas;cblas;atlas" \
-DLAPACK_LIBRARY_NAMES:STRING="lapack" \
\
-DProfugus_CONFIGURE_OPTIONS_FILE:FILEPATH="${SOURCE}/install/base.cmake" \
-DProfugus_ASSERT_MISSING_PACKAGES:BOOL=OFF \
\
${SOURCE}

##---------------------------------------------------------------------------##
## end of cmake_x86_64.sh
##---------------------------------------------------------------------------##

The source and install locations must be set. Also, to enable a optimized build set BUILD to RELEASE. Adjust the paths and libraries for LAPACK to fit your platform. The example assumes that the ATLAS LAPACK is available. Any standard LAPACK distribution will work. HDF5 is not required, to build/run/test the applications; however, problem output will be severely curtailed if a parallel HDF5 option is not provided. If HDF5 is not available, setting:

-DTPL_ENABLE_HDF5:BOOL=OFF \

will disable HDF5.

To complete the configuration, execute this script inside the target directory and then make/test/install:

> pwd
  /home/me/debug/target
> sh /home/me/Profugus/install/cmake_x86_64.sh
> make -j 8
> ctest -j 8
> make -j 8 install