What is parCASL?

parCASL (Parallel Computational Applied Science Library) is a library for solving Partial Differential Equations (PDEs) that are commonly encountered in physics and applied sciences. It is written in C++ and leverages state- of-the-art algorithms as well as well-established high performance libraries to enable fast and accurate solution of PDEs.

License

Please see the LICENSE file included in this directory.

Installation

Let's assume that we are on Ubuntu 18 or 20.

Minimal dependencies

Installing CMake

1. Download cmake-3.22.1-linux-x86_64.sh or a newer version from here. Assuming we are in the home directory (where we have downloaded CMake), type:

$> sudo ./cmake-3.22.1-linux-x86_64.sh

2. Follow the instructions. It'll install CMake everything under a new directory: cmake-3.22.1-linux-x86_64.sh. Rename the folder to make it easy to use CMake.

mv cmake-3.22.1-linux-x86_64/ cmake-3.22.1

3. The CMake binary is now under the bin folder inside cmake-3.22.1:

$> cd cmake-3.22.1
$> cd bin/
$> ./cmake --help

And the last line will show the help if everything went OK.

Installing MPICH

1. Update packages, and then run the usual installation instruction with apt.

$> sudo apt update
$> sudo apt install mpich

2. These instructions install MPICH under the /etc/alternatives/ folder, but it also places links to it under /usr/bin/. Similarly, the shared libraries are linked under /usr/lib/mpich and the headers under /usr/include/mpich. Now, test the installation from anywhere you like:

$> cd
$> mpiexec --help

Installing PETSc

1. Download the latest PETSc library and decompress it.

$> cd
$> curl -O https://ftp.mcs.anl.gov/pub/petsc/release-snapshots/petsc-3.16.3.tar.gz
$> tar -xvf petsc-3.16.3.tar.gz

2. The above commands creates the petsc-3.16.3/ directory. Switch to that folder and configure the library. Notice this will configure it in release mode. For debug mode, use the option --with-debugging=1.

$> cd petsc-3.16.3
$> sudo ./configure --download-fblaslapack --download-hypre=1 --prefix=/usr/local/petsc-3.16.3 
   --with-debugging=0 --with-mpi-dir=/usr --with-shared-libraries=1 
   COPTFLAGS=-O2 CXXOPTFLAGS=-O2 FOPTFLAGS=-O2

Wait for the process to finish.

3. Now, build PETSc. It'll tell you what command to use after finishing the previous step, for example:

$> sudo make PETSC_DIR=/home/youngmin/petsc-3.16.3 PETSC_ARCH=arch-linux-c-opt all

4. Install PETSc. Again, it'll tell you what instruction to use, for example:

$> sudo make PETSC_DIR=/home/youngmin/petsc-3.16.3 PETSC_ARCH=arch-linux-c-opt install

5. To check the installation, type:

$> make PETSC_DIR=/usr/local/petsc-3.16.3 PETSC_ARCH="" check

The library will be under the /usr/local/petsc-3.16.3/ directory. It comes with the lib/ and include/ folders.

Installing p4est

1. Download the latest p4est library and decompress it.

$> cd
$> curl -O https://p4est.github.io/release/p4est-2.8.tar.gz
$> tar -xvf tar -xvf p4est-2.8.tar.gz

2. The above command creates the p4est-2.8/. Navigate to that folder and configure the library. Again, this will be in release mode. For debug mode, use the option --enable-debug.

$> cd p4est-2.8/
$> sudo ./configure --prefix=/usr/local/p4est-2.8 --enable-mpi --enable-shared --enable-memalign=16 
   CFLAGS=-O2 CPPFLAGS=-O2 FCFLAGS=-O2

3. Now, build and install p4est.

$> sudo make
$> sudo make install

The library will be under the /usr/local/p4est-2.8/ directory. It comes with lib/ and include/ folders.

Installing BOOST

1. Download the BOOST source from here. Place it in the home directory and uncompress it.

$> cd
$> tar -xvf boost_1_78_0.tar.bz2

2. Install it by running the following commands:

$> cd boost_1_78_0
$> sudo ./bootstrap.sh --prefix=/usr/local/boost-1.78.0

The library will then be placed in the /usr/local/boost-1.78.0/ folder. It comes with lib/ and include/boost/ folders.

Installing Voro++

1. Download the Voro++ library and decompress it.

$> cd
$> curl -O http://math.lbl.gov/voro++/download/dir/voro++-0.4.6.tar.gz
$> tar -xvf voro++-0.4.6.tar.gz

2. Switch to the Voro++ source folder, compile it, and install it.

$> cd voro++-0.4.6
$> sudo make all
$> sudo make all install

The library is now placed under /usr/local/. The executable is under bin/, and the library and headers under lib/ and include/voro++/.

3. You can check the executable with:

$> cd /usr/local/bin
$> ./voro++ --help

Machine-learning dependencies

TBD

Cloning the repository

Follow the instructions here to set up SSH before cloning. Add your public key to your BitBucket personal settings.

To clone the project:

$> git clone [email protected]:cburstedde/casl_p4est.git

Compiling parCASL with CMake

See the CMake webpage for more details.

To create the CMakeLists.txt file under an example, take a look at the ml_curvature's CMakeLists.txt file. Don't forget to add your machine library dependencies under the cmake/ folder at the root of parCASL.

1. cd to the example you want to compile:

$> cd /Users/youngmin/Documents/CS/CASL/casl_p4est/examples/ml_curvature

2. Generate the Makefile. For example, let's generate a Makefile for Release mode in 2d:

$> /Applications/CLion.app/Contents/bin/cmake/mac/bin/cmake -G "CodeBlocks - Unix Makefiles" 
   -DCMAKE_BUILD_TYPE=Release -DENABLE_ML=1 -DDIMENSION=2d -B./cmake-build-release-2d -S.

Here:

• -DDIMENSION=2d sets the DIMENSION variable to 2d.
• -DCMAKE_BUILD_TYPE sets the CMAKE_BUILD_TYPE variable to Release.
• -DENABLE_ML enables machine-learning dependencies and classes.
• -B./cmake-build-release-2d tells CMake where to put all intermediate files and Makefile.
• -S. tells CMake that the source directory is the one we are currently at.
• -G "CodeBlocks - Unix Makefiles" indicates which generator to use. If not given, it checks the CMAKE_GENERATOR environment variable or falls back to a builtin default selection.

3. Build the project:

$> /Applications/CLion.app/Contents/bin/cmake/mac/bin/cmake --build ./cmake-build-release-2d/ 
   --target ml_curvature -- -j 9

Here:

• --build ./cmake-build-release-2d/ indicates where to find the Makefile.
• --target ml_curvature is the target you defined in CMakeLists.txt.
• -- -j 9 gives one build-tool option. Anything after -- is an option. -j 9 tells CMake to use up to 9 concurrent jobs when building.

4. Run the project as a single-process app:

$> cd cmake-build-release-2d/
$> ./ml_curvature

or with MPI:

$> cmake-build-release-2d/
$> mpiexec -n 3 ./ml_curvature