Merge pull request #3 from aurora-multiphysics/HenriqueBR/GPU_Tests

GPU-compatible platypus

include/io/outputs.h

@@ -134,10 +134,7 @@ class Outputs : public platypus::NamedFieldsMap<mfem::DataCollection>
   }
 
   // Write out summary of last timestep to console
-  void WriteConsoleSummary(int _my_rank, double t)
-  {
-    // logger.info("step {}, \tt = {}", _cycle, t);
-  }
+  void WriteConsoleSummary(int _my_rank, double t) {}
 
   // Initialize GLVis sockets and fields
   void InitializeGLVis(int _my_rank)

include/problem/MFEMProblem.h

@@ -171,6 +171,7 @@ class MFEMProblem : public ExternalProblem
   std::string _input_mesh;
   std::string _formulation_name;
   int _order;
+  mfem::Device _device;
 
   platypus::Coefficients _coefficients;
   platypus::InputParameters _solver_options;

scripts/build-platypus-csd3-ampere.sh

@@ -0,0 +1,267 @@
+#!/bin/bash
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --time=04:00:00
+#SBATCH --mail-type=none
+#SBATCH -p ampere
+#SBATCH -A ukaea-ap001-GPU
+#SBATCH --cpus-per-task=32
+#SBATCH --gres=gpu:1
+#SBATCH --output=platypus_gpu_build.%j.out
+#SBATCH --error=platypus_gpu_build.%j.err
+
+## WARNING: THIS SCRIPT WILL UNINSTALL ALL SPACK MODULES ASSOCIATED WITH
+## THE ARCHITECTURE DEFINED IN THE ARCH VARIABLE. IF YOU DO NOT WISH TO DO
+## THAT, COMMENT OUT THE SPACK UNINSTALL LINE BEFORE SUBMITTING THE SCRIPT
+## -> UNINSTALL LINE IN THE install_spack_deps() FUNCTION
+ARCH="linux-rocky8-zen"
+
+export compile_cores=32
+
+load_modules() {
+
+    # Load modules
+
+    # shellcheck source=/dev/null
+    . /etc/profile.d/modules.sh                # Leave this line (enables the module command)
+    module purge
+    module load rhel8/slurm
+    module use /usr/local/software/spack/spack-modules/rocky8-a100-20230831/linux-rocky8-zen3
+}
+
+set_paths() {
+
+    USER=$(whoami)
+    BUILD_PREFIX=platypus_gpu
+    BUILD_DIR_NAME=${BUILD_PREFIX}_build
+
+    ROOT_PATH=/home/${USER}/rds/rds-ukaea-ap001/${USER}
+    BUILD_PATH=${ROOT_PATH}/${BUILD_DIR_NAME}
+
+    echo "Building in ${BUILD_PATH}"
+    mkdir -p "${BUILD_PATH}" || {
+                                    echo "Failed to create ${BUILD_PATH}"
+                                                                            exit 1
+    }
+
+    cd "${BUILD_PATH}" || exit 1
+
+}
+
+check_cuda_version() {
+
+    if grep -q "[email protected]" "${SPACK_ROOT}"/etc/spack/defaults/packages.yaml; then
+        echo "Ampere CUDA module found in spack."
+    else
+        echo "Ampere CUDA module not found in spack. Adding to packages.yaml."
+        CUDA_STR=$'  cuda:\n    externals:\n    - spec: "[email protected]"\n      prefix: /usr/local/software/cuda/11.4\n    buildable: False'
+        echo "${CUDA_STR}"  >> "${SPACK_ROOT}"/etc/spack/defaults/packages.yaml
+    fi
+
+    # shellcheck source=/dev/null
+    . "${SPACK_ROOT}"/share/spack/setup-env.sh
+
+}
+
+check_spack() {
+
+    cd "${ROOT_PATH}" || exit 1
+
+    if [ "$(command -v spack)" ]; then
+        echo "Spack command detected. Using pre-loaded spack."
+    elif [ -f "${ROOT_PATH}"/spack/share/spack/setup-env.sh ]; then
+        echo "Spack detected in root directory. Loading."
+        # shellcheck source=/dev/null
+        . spack/share/spack/setup-env.sh
+    else
+        echo "No spack detected. Building from source."
+        git clone --depth=100 https://github.com/spack/spack.git
+        # shellcheck source=/dev/null
+        . spack/share/spack/setup-env.sh
+    fi
+
+    check_cuda_version
+
+}
+
+install_spack_deps() {
+
+    # Cleaning up everything to start with a new environment
+    spack uninstall -ay arch=${ARCH}
+    spack clean -ab
+
+    echo "Installing Petsc..."
+    # Spack's petsc doesn't like openmpi, but it works with mpich
+    spack install petsc +cuda cuda_arch=80 +fortran +hdf5 +hypre +metis +mpi \
+        ^mpich +cuda cuda_arch=80 \
+        ^hdf5 +cxx +fortran +hl +mpi +shared \
+        ^hypre +mpi +shared +cuda cuda_arch=80 +superlu-dist +cublas +gpu-aware-mpi \
+        ^superlu-dist +cuda cuda_arch=80 +parmetis +shared
+    spack load petsc arch=${ARCH}
+
+    echo "Installing SLEPc..."
+    spack install slepc +cuda cuda_arch=80
+    spack load slepc arch=${ARCH}
+
+    echo "Installing netcdf..."
+    spack install netcdf-c +parallel-netcdf
+    spack load netcdf-c arch=${ARCH}
+
+    echo "Installing ninja..."
+    spack install ninja
+    spack load ninja arch=${ARCH}
+
+    echo "Adding python modules..."
+
+    spack install py-pyaml
+    spack load py-pyaml arch=${ARCH}
+
+    spack install py-jinja2
+    spack load py-jinja2 arch=${ARCH}
+
+    spack install py-packaging
+    spack load py-packaging arch=${ARCH}
+
+    spack install py-setuptools
+    spack load py-setuptools arch=${ARCH}
+
+}
+
+install_gslib() {
+
+    echo "Installing gslib..."
+    cd "${BUILD_PATH}" || exit 1
+    git clone https://github.com/Nek5000/gslib.git
+    cd gslib || exit 1
+    make CC=mpicc CFLAGS="-O2 -fPIC" -j"$compile_cores"
+}
+
+install_mfem() {
+
+    export CXX=mpic++
+    export CC=mpicc
+    export F90=mpif90
+    export F77=mpif77
+    export FC=mpif90
+    SLU_DIR=$(spack find --format "{prefix}" superlu-dist)
+    export SLU_DIR
+
+    # Build MFEM
+    cd "${BUILD_PATH}" || exit 1
+    git clone https://github.com/mfem/mfem.git
+    cd mfem || exit 1
+    # This is just until MFEM merges Edward's changes. Without this, GPU build crashes!
+    git checkout EdwardPalmer99/add-missing-header-to-exodus-writer-fix
+    mkdir build
+    cd build || exit 1
+    echo "Building MFEM"
+    cmake .. \
+        -DCMAKE_BUILD_TYPE=Release \
+        -DBUILD_SHARED_LIBS=YES \
+        -DMFEM_USE_OPENMP=NO \
+        -DMFEM_THREAD_SAFE=YES \
+        -DMFEM_ENABLE_EXAMPLES=YES \
+        -DMFEM_ENABLE_MINIAPPS=YES \
+        -DMFEM_USE_MPI=YES \
+        -DMFEM_USE_CUDA=YES \
+        -DCUDA_ARCH=sm_80 \
+        -DMFEM_USE_METIS_5=YES \
+        -DMFEM_USE_SUPERLU=YES \
+        -DMFEM_USE_NETCDF=YES \
+        -DMFEM_USE_GSLIB=YES \
+        -DGSLIB_DIR="${BUILD_PATH}/gslib/build" \
+        -DSuperLUDist_DIR="${SLU_DIR}" \
+        -DSuperLUDist_VERSION_OK=YES
+
+    if [ $? -eq 2 ]; then
+        echo "MFEM config failed"
+        exit 1
+    fi
+
+    make -j"$compile_cores"
+
+    if [ $? -eq 2 ]; then
+        echo "MFEM build failed"
+        exit 1
+    fi
+
+    LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:${BUILD_PATH}/mfem/build:${BUILD_PATH}/mfem/build/miniapps/common
+}
+
+install_moose() {
+
+    # Some of the variables needed
+    export MOOSE_JOBS=$compile_cores
+    export LIBMESH_JOBS=$compile_cores
+    export METHOD="opt"
+    SLEPC_DIR=$(spack find --format "{prefix}" slepc arch=${ARCH})
+    export SLEPC_DIR
+
+    cd "${BUILD_PATH}" || exit 1
+    git clone https://github.com/idaholab/moose
+    cd moose || exit 1
+
+    echo "Building libmesh..."
+    ./scripts/update_and_rebuild_libmesh.sh --with-mpi
+    if [ $? -eq 2 ]; then
+        echo "libmesh build failed"
+        exit 1
+    fi
+
+    echo "Building WASP..."
+    ./scripts/update_and_rebuild_wasp.sh
+    if [ $? -eq 2 ]; then
+        echo "WASP build failed"
+        exit 1
+    fi
+
+    ./configure --with-derivative-size=200
+    if [ $? -eq 2 ]; then
+        echo "MOOSE configure failed"
+        exit 1
+    fi
+
+    cd framework || exit 1
+    make -j"$compile_cores"
+    if [ $? -eq 2 ]; then
+        echo "MOOSE framework build failed"
+        exit 1
+    fi
+
+    cd ../modules || exit 1
+    make -j"$compile_cores"
+    if [ $? -eq 2 ]; then
+        echo "MOOSE modules build failed"
+        exit 1
+    fi
+
+    # This takes very long! Only run the tests if you really need to!
+    #cd ../test || exit 1
+    #make -j"$compile_cores"
+    #if [ $? -eq 2 ]; then
+    #    echo "MOOSE test build failed"
+    #    exit 1
+    #fi
+
+    #./run_tests -j"$compile_cores"
+}
+
+install_platypus() {
+
+    cd "${BUILD_PATH}" || exit 1
+
+    echo "Building platypus..."
+    git clone https://github.com/aurora-multiphysics/platypus.git
+    cd platypus || exit 1
+    make -j"$compile_cores"
+
+}
+
+load_modules
+set_paths
+check_spack
+install_spack_deps
+install_gslib
+install_mfem
+install_moose
+install_platypus

src/problem/MFEMProblem.C

@@ -12,6 +12,7 @@ MFEMProblem::validParams()
       "Number of timesteps between successive write outs of data collections to file.");
   params.addParam<bool>(
       "use_glvis", false, "Attempt to open GLVis ports to display variables during simulation");
+  params.addParam<std::string>("device", "cpu", "Run app on the chosen device.");
 
   return params;
 }
@@ -23,6 +24,8 @@ MFEMProblem::MFEMProblem(const InputParameters & params)
     _outputs(),
     _exec_params()
 {
+  _device.Configure(getParam<std::string>("device"));
+  _device.Print(std::cout);
 }
 
 MFEMProblem::~MFEMProblem() {}

src/problem_builders/problem_builder_base.C

@@ -13,7 +13,6 @@ Problem::~Problem()
 void
 ProblemBuilder::SetMesh(std::shared_ptr<mfem::ParMesh> pmesh)
 {
-  // logger.info("Setting Mesh");
   GetProblem()->_pmesh = pmesh;
   GetProblem()->_comm = pmesh->GetComm();
   MPI_Comm_size(pmesh->GetComm(), &(GetProblem()->_num_procs));
@@ -23,35 +22,30 @@ ProblemBuilder::SetMesh(std::shared_ptr<mfem::ParMesh> pmesh)
 void
 ProblemBuilder::SetFESpaces(platypus::FESpaces & fespaces)
 {
-  // logger.info("Setting FE Spaces");
   GetProblem()->_fespaces = fespaces;
 }
 
 void
 ProblemBuilder::SetGridFunctions(platypus::GridFunctions & gridfunctions)
 {
-  // logger.info("Setting GridFunctions");
   GetProblem()->_gridfunctions = gridfunctions;
 }
 
 void
 ProblemBuilder::SetBoundaryConditions(platypus::BCMap & bc_map)
 {
-  // logger.info("Setting Boundary Conditions");
   GetProblem()->_bc_map = bc_map;
 }
 
 void
 ProblemBuilder::SetOutputs(platypus::Outputs & outputs)
 {
-  // logger.info("Setting Outputs");
   GetProblem()->_outputs = outputs;
 }
 
 void
 ProblemBuilder::SetSolverOptions(platypus::InputParameters & solver_options)
 {
-  // logger.info("Setting Solver Options");
   GetProblem()->_solver_options = solver_options;
 }
 
@@ -70,14 +64,12 @@ ProblemBuilder::SetJacobianSolver(std::shared_ptr<mfem::Solver> jacobian_solver)
 void
 ProblemBuilder::SetCoefficients(platypus::Coefficients & coefficients)
 {
-  // logger.info("Setting Coefficients");
   GetProblem()->_coefficients = coefficients;
 }
 
 void
 ProblemBuilder::AddFESpace(std::string fespace_name, std::string fec_name, int vdim, int ordering)
 {
-  // logger.info("Adding {} FE Space to problem", fespace_name);
   if (GetProblem()->_fespaces.Has(fespace_name))
   {
     const std::string error_message = "A fespace with the name " + fespace_name +

test/tests/unit/kernels/diffusion_mfem.i

@@ -7,6 +7,7 @@
 [Problem]
   type = MFEMProblem
   use_glvis = true
+  device = "cpu"
 []
 
 [Formulation]