---

1. Miniapp to test GBS gradients with Makefile # -------------------------------------------------------------------------------

#To compile with Makefile and run OpenMP with Offload implementation on izar with NVHPC #

module load nvhpc/21.2-mpi make

srun -p debug --gres=gpu:1 ./gbs_gradients

#To compile with Makefile and run OpenMP with Offload implementation on izar with gcc

module load gcc #modify FLAGS and FC in Makefile uncommenting the corresponding lines make srun -p debug --gres=gpu:1 ./gbs_gradients

---

#To compile with Makefile and run OpenMP with Offload implementation on piz-daint with cce compiler

module load daint-gpu module load craype-accel-nvidia60 module load cudatoolkit export CRAY_ACCEL_TARGET=nvidia60 git checkout compile_without_openacc # branch with acc pragmas commented #modify FLAGS and FC in Makefile uncommenting the corresponding lines make

srun --gres=gpu:1 -c 12 -N 1 -n 1 --time=00:15:00 --account=g26 -C gpu ./gbs_gradients

#To compile with Makefile and run CUDA C implementation on izar with gcc

module load gcc nvhpc/21.2-mpi #modify FLAGS and FC in Makefile uncommenting the corresponding lines make gbs_gradients_cuda

srun -p debug --gres=gpu:1 ./gbs_gradients

1. To compile with CMAKE with gnu on izar # -------------------------------------------------------------------------------

module load gcc cmake mkdir build cd build cmake ..

make

1. To compile with CMAKE with nvhpc on izar # -------------------------------------------------------------------------------

module load nvhpc/21.2-mpi cmake mkdir build cd build cmake ..

make

1. To compile using CUDA with CMAKE on izar # -------------------------------------------------------------------------------

module load gcc cuda cmake mkdir build cd build cmake -DUSE_CUDA=1 ..

make

1. To compile OpenMP with CMAKE with CCE on piz-daint # -------------------------------------------------------------------------------

module load daint-gpu module load craype-accel-nvidia60 module load cudatoolkit export CRAY_ACCEL_TARGET=nvidia60 module load CMake mkdir build cd build cmake ..

make

1. To compile OpenACC with CMAKE with CCE on piz-daint # -------------------------------------------------------------------------------

module swap PrgEnv-cray PrgEnv-pgi module unload cray-libsci_acc module load cudatoolkit module load craype-accel-nvidia60 module load daint-gpu module load CMake mkdir build cd build cmake -DUSE_ACC=1 ..

make

1. To compile using CUDA with CMAKE with CCE on piz-daint # -------------------------------------------------------------------------------

module load daint-gpu module load craype-accel-nvidia60 module load cudatoolkit export CRAY_ACCEL_TARGET=nvidia60 mkdir build cd build cmake -DUSE_CUDA=1 ..

make

1. To compile with CMAKE with xlf compiler on Marconi100 # -------------------------------------------------------------------------------
2. xlf compiler doesn't support Fortran "norm2" function (Fortran 2008 standard)
3. So have to comment lines mentionning norm2 (just useful for output) in test_gbs_gradients.F90

module load xl cmake cuda mkdir build cd build cmake -DCMAKE_Fortran_COMPILER=xlf2003_r -DCMAKE_VERBOSE_MAKEFILE=ON .. make