Description goes here...
Ensure you have GCC 6 (or 7) installed and set as your default C/C++ compiler. Furthermore CMake 3.6 (or later) is required for the build and testing process. Simply execute the following commands to build the project and run all tests.
$ mkdir build
$ cd build
$ cmake ../code
$ make -j8
$ ctest -j8
Following options can be supplied to CMake
| Option | Values |
|---|---|
| -DCMAKE_BUILD_TYPE | Release / Debug |
| -DBUILD_SHARED_LIBS | ON / OFF |
| -DBUILD_TESTS | ON / OFF |
| -DBUILD_DOCS | ON / OFF |
| -DBUILD_COVERAGE | ON / OFF |
| -DUSE_ASSERT | ON / OFF |
| -DALLSCALE_CHECK_BOUNDS | ON / OFF |
| -DUSE_VALGRIND | ON / OFF |
| -DUSE_ALLSCALECC | ON / OFF |
| -DENABLE_PROFILING | ON / OFF |
| -DTHIRD_PARTY_DIR | <path> |
The files cmake/build_settings.cmake and code/CMakeLists.txt state their
default value.
To use the AllScale compiler, you first have to setup the required dependencies
in order to build it. Some packages should be installed with your system's
package manager, others must be installed via the provided dependency
installer. See scripts/dependencies/README.md for more details. The
following commands assume you are using Ubuntu 16.04 LTS -- adjust the lines if
you are using a different distribution.
$ sudo apt update
$ sudo apt install bison build-essential cmake flex git groff-base \
libcpufreq-dev libgmp10 libhwloc-dev libjemalloc-dev libluajit-5.1-dev \
libpapi-dev m4 pkg-config python ruby time wget zlib1g
$ scripts/dependencies/installer llvm boost cudd gmp zlib ghc cabal
To build this project using the AllScale compiler, simply set the corresponding CMake option. You may want to use a separate build directory to easily switch between GCC and AllScaleCC.
$ mkdir build_with_allscalecc
$ cd build_with_allscalecc
$ ../scripts/dependencies/third_party_linker
$ cmake -DUSE_ALLSCALECC=ON ../code
$ make -j8
$ ctest -j8
The setup script can be run again to add new modules, just provide the same project name.
$ scripts/setup/run amdados frontend backend utils
There is a utility script to add new parts to an existing module. The project name and module name must be provided followed by a list of parts to generate. Folders will be created along the way.
$ scripts/setup/add_part amdados frontend sema extensions/malloc_extension
This will add the files sema.h, sema.cpp and sema_test.cc to the
frontend module. Furthermore new subfolders extensions will be created
containing malloc_extension.h, malloc_extension.cpp and
malloc_extension_test.cc in their respective subdirectories.
The following two CMake options allow you to manually set the AllScale Compiler
and AllScale API to use. Note that USE_ALLSCALECC must not be enabled for
this to work.
| Option | Values |
|---|---|
| -DOVERRIDE_ALLSCALECC | path to allscalecc binary |
| -DOVERRIDE_ALLSCALE_API | path to AllScale API project |
When working on Windows via SMB share, consider setting following Git setting.
$ git config core.filemode false
A script, together with a Git hook, is provided to automatically add a license
header to each source file upon commit. See scripts/license.
$ cmake -G "Eclipse CDT4 - Unix Makefiles" /path/to/project
$ cmake -G "Visual Studio 15 2017 Win64" -DBUILD_SHARED_LIBS=OFF Z:\path\to\project
Add path for third-party libraries when needed.
The dependency installer can setup these required tools for you. Its README
(scripts/dependencies/README.md) holds the details.
It is preferred to use the operating system's package manager, if applicable.
Make sure your build folder is located outside the source folder. Eclipse is not capable of dealing with such a setup correctly.
Building the coverage us currently only supported on Linux, as Perl and Bash
are required. To build and view the coverage set the corresponding CMake flag
to ON and run:
$ make
$ make coverage
$ xdg-open coverage/index.html
The Armadillo matrix library is used only (!) for testing of matrix operations. It is not included in the repository (see .gitignore). Armadillo is installed along with other dependencies, see below. As of 29.03.2018 Armadillo is included in the repository, but this is not necessary. It can be always installed from the official web-site, see ./scripts/armadillo.sh
In order to compile the project on machine without Internet connection (e.g. blade behind firewall), some prerequisites should be downloaded beforehand. From the project root folder, one should execute: bash ./scripts/download.sh This will install all necessary dependencies (including the Armadillo library) into the folder "api".
For complete description see "./doc/amdados_report.pdf". As a quick start, one can run the application with default settings, section 2.4.1 in the report:
bash standard.build.sh
/bin/rm -fr output/* # optional cleanup
python3 python/ObservationsGenerator.py --config amdados.conf
./build/app/amdados --config amdados.conf
python3 python/Visualize.py --field output/field_Nx*_Ny*_Nt*.bin
where "field_Nx*_Ny_Nt*.bin" should be replaced with the actual file name with prefix "field".
B E W A R E: simulation might be very long (~ 1 day) on the machine with few CPU cores.
(1) Python of version 3.5+ is needed to generate observations, to find ground-truth solution and for testing & visualization, see "./doc/amdados_report.pdf".
We recommend multiplatform package manager Anaconda ver.3:
https://www.anaconda.com/download/
Anaconda includes numpy, scipy, matplotlib, ffmpeg and their dependencies.
(2) The latest Allscale API and Armadillo library (used in unit tests) can be downloaded and installed as mentioned above: bash ./scripts/download.sh
(3) a) On Mac OS one probably wants to install "XQuartz" (via brew) for visualization with Python. b) Another useful tool is "mplayer" - a command-line player that shows the solution rolling out in time.
(1) TODO: not integrated into building system yet. (2) Temporary solution - run the script from the project root folder: ./mympi -r This will compile both, Allscale and MPI, versions of Amdados application. (3) Generate observations as usually: python3 python/ObservationsGenerator.py --config <config_file_name> (4) Run MPI implementation (MPICH): mpirun -n XX -f host_file ./build/mpi_amdados --config <config_file_name> In case of OpenMPI you specify host file differently: --hostfile host_file. The file "host_file" can be placed in the project root folder as in example above. The current one is very simple. It contains a single record "localhost". (5)