NEMOCFG/eORCA025/eORCA025.L121-LUCIA01/cfgs/ORCA025_ICE/EXPREF/NE4_00.sh

#!/bin/bash

#
# Job options 
#
#SBATCH --job-name=NE4_00
#SBATCH --time=12:00:00
#SBATCH --account=ecearth
#
#SBATCH --nodes=7
#SBATCH --exclusive
##SBATCH --ntasks=1320
#SBATCH --ntasks-per-node=100
#SBATCH --partition=batch
#
set -ueo pipefail
#
LOCAL_NODES=7
LOCAL_TASKS=700
#
stdout_file=${SLURM_SUBMIT_DIR-$PWD}/${SLURM_JOB_NAME-"local"}_${SLURM_JOB_ID-"id"}.log
exec > ${stdout_file}
echo "------------------ Job Info --------------------"
echo "jobid : ${SLURM_JOB_ID-"id"}"
echo "jobname : ${SLURM_JOB_NAME-"local"}"
echo "nodename : ${SLURMD_NODENAME-"nlocal"}"
echo "# nodes : ${SLURM_JOB_NUM_NODES-$LOCAL_NODES}"
echo "# tasks : ${SLURM_NTASKS-$LOCAL_TASKS}"
echo "submit dir : ${SLURM_SUBMIT_DIR-$PWD}"

set -ue 
#
# Cluster variables
#
NB_CORES_PER_NODES=128
MAX_CORES_PER_NODES=100
LIST_CORES_SOCKET=`seq -s',' 0 $((NB_CORES_PER_NODES-1))`
#
# Experiment options
#
exp_name=NE4_00
run_start_date="1960-01-01"
run_duration="1 year"
rst_freq="1 month"
run_num_legs=12
special_restart=false
special_restart_from=EXP0
special_restart_date="1959-01-01"
# ORCA025=1350 - ORCA1=2700 - ORCA2=5400
#    1-23360      1-11680       1-5840
nem_time_step_sec=1350
lim_time_step_sec=1350
nem_restart_offset=0
nem_config_name=ORCA025_ICE
info_file="nemo.info"
start_dir=${SLURM_SUBMIT_DIR-$PWD}
run_dir="/gpfs/scratch/acad/ecearth/$USER/nemo/run/${exp_name}"
archive_dir="/gpfs/scratch/acad/ecearth/$USER/nemo/archive/${exp_name}"

#
# Program configuration
#
#192 - 230 - 460 - 1150
#debug nem_numproc=360
nem_numproc=600
xio_numproc=100
#debug xio_numproc=24
#4 - 4 - 6 - 14 (max 26)

nemo_src_dir=${HOME}/modeles/nemo_4.2.0
shared_dir=${nemo_src_dir}/cfgs/SHARED
nem_exe=nemo.exe
nem_exe_file=${start_dir}/../BLD/bin/nemo.exe
xio_exe=xios_server.exe
xio_exe_file=${nemo_src_dir}/ext/xios-trunk-2326_gnu/bin/xios_server.exe

all_proc=$(($nem_numproc+$xio_numproc))
if [[ "${SLURM_JOB_NAME-"local"}" != "local" ]] ; then
  if (( $all_proc != ${SLURM_NTASKS-$LOCAL_TASKS} ))
  then
    echo "XIOS procs + NEMOC procs do not fit with SLURM requirements."
    #exit 0
  fi
fi

#
# Data configuration
#
nem_grid=ORCA025L121
#
ini_data_dir=/gpfs/scratch/acad/ecearth/pbarriat/data/nemo
#
ic_subdir=initial
ic_files=(
"Goutorbe_ghflux.nc"
"eORCA025_ghflux_v2.0_c3.0_weights_bilin_nohls.nc => weights_ghflux_bilinear.nc"
"eORCA025_iwm_b0.2_v1.0_nohls.nc => zdfiwm_forcing.nc"
"eORCA025.L121_domain_cfg_b0.5_c3.0_d1.0_nohls_clean.nc => domain_cfg.nc"
"eORCA025_runoff_b0.2_v0.0_nohls.nc => runoff.nc"
"eORCA025_calving_b0.2_v2.3_nohls.nc => calving.nc"
"eORCA025_ttv_b0.2_v0.0_nohls.nc => boost_tidal_velocity.nc"
"eORCA025_bfr2d_v0.2_nohls.nc => bfr_coef.nc"
"eORCA025_shlat2d_v0.2_nohls.nc => shlat2d.nc"
"eORCA025_distcoast_b0.2_v0.0_nohls.nc => distcoast.nc"
"eORCA025.L121-empc_nohls.nc => empc.nc"
)
#
nem_res_hor=$(echo ${nem_grid} | sed 's:ORCA\([0-9]\+\)L[0-9]\+:\1:')
#
clim_subdir=climatology
clim_files=(
"eORCA025.L121_WOA2018_c3.0_d1.0_v19812010.5.2_nohls.nc => woce_monthly_init.nc"
"chlorophyl_v0.0.nc => chlorophyl.nc"
"eORCA025_chlorophyl_v0.0_c3.0_weights_bilin_nohls.nc => chlorophyl_weights_bilin.nc"
"eORCA025_sss_WOA2018_c3.0_v19812010.5.1_nohls.nc => sss_absolute_salinity.nc"
"eORCA025_seaice_c3.0_v19802004.0_nohls.nc => seaice.nc"
)
#
forcing_subdir=forcing
nem_forcing_set=ERA5
#nem_forcing_set=JRA55
forcing_files=(
"* => ."
)
#
shared_files=(
"namelist_ice_ref"
"namelist_ref"
"domain_def_nemo.xml"
"axis_def_nemo.xml"
"field_def_nemo-ice.xml"
"field_def_nemo-oce.xml"
"grid_def_nemo.xml"
)

#
# Script logic
#
function leap_days()
{
  local ld=0
  local frstYYYY=$(date -ud "$1" +%Y)
  local lastYYYY=$(date -ud "$2" +%Y)

  set +e

  $(date -ud "${frstYYYY}-02-29" > /dev/null 2>&1) \
    && (( $(date -ud "$1" +%s) < $(date -ud "${frstYYYY}-03-01" +%s) )) \
    && (( $(date -ud "$2" +%s) > $(date -ud "${lastYYYY}-02-28" +%s) )) \
    && (( ld++ ))

  for (( y=(( ${frstYYYY}+1 )); y<=(( ${lastYYYY}-1 )); y++ ))
  do
    $(date -ud "$y-02-29" > /dev/null 2>&1) && (( ld++ ))
  done

  (( $lastYYYY > $frstYYYY )) \
    && $(date -ud "${lastYYYY}-02-29" > /dev/null 2>&1) \
    && (( $(date -ud "$1" +%s) < $(date -ud "${frstYYYY}-03-01" +%s) )) \
    && (( $(date -ud "$2" +%s) > $(date -ud "${lastYYYY}-02-28" +%s) )) \
    && (( ld++ ))

  set -e

  echo "$ld"
}


[[ $@ == *verbose* ]] && set -x

#module purge
module load craype-x86-milan
module load PrgEnv-gnu/8.3.3
module load netCDF-Fortran/4.6.0-gompi-2022a
module load Perl/.5.34.1-GCCcore-11.3.0

if [ ! -d ${run_dir:?} ]
then
  mkdir -p ${run_dir}
  #
  if $special_restart
  then
    rsync -av --delete ${run_dir}/../${special_restart_from}/ --exclude log --exclude output --exclude restart --exclude="${special_restart_from}_*" --exclude="ocean*" --exclude="restart_*" --exclude="debug.*" --exclude="output.*" ${run_dir}
    cp -f ${nem_exe_file} ${run_dir}
    cp -f ${xio_exe_file} ${run_dir}
    special_year=${special_restart_date:0:4}
    sed -i "/$special_year/q" ${run_dir}/${info_file}
    . ${run_dir}/${info_file}
    special_restart_leg=$(printf %03d $((leg_number+1)))
    cd ${run_dir}/../../archive/${special_restart_from}/restart/${special_restart_leg}
    for f in *.nc; do
      nf=${exp_name}${f:4}
      cp $f ${run_dir}/$nf
    done
    cd -
    cd ${run_dir}
    for f in ${exp_name}_????????_restart_???_????.nc; do
      nf=${f:14}
      ln -s $f $nf
    done
    cd -
  fi

  cd ${start_dir}
  cp context_nemo.xml file_def_nemo-ice.xml file_def_nemo-oce.xml iodef.xml namelist_ice_cfg* build_namelist_cfg* ${run_dir}
  cd ${run_dir}
  cp ${xio_exe_file} ${xio_exe}
  cp ${nem_exe_file} ${nem_exe}
  
  [[ ! -f EMPave_old.dat ]] && echo "                               0  0.0000000000000000E+00  0.0000000000000000E+00" > EMPave_old.dat
  for file in "${ic_files[@]}"; do 
      [[ ! -e ${file#*> } ]] && ln -sf $(sed 's/ *=> */ /' <<< "${ini_data_dir}/${ic_subdir}/${nem_grid}/$file") 
  done
  for file in "${ic_files[@]}"; do
      [[ ! -e ${file#*> } ]] && ln -sf $(sed 's/ *=> */ /' <<< "${ini_data_dir}/${ic_subdir}/$file")
  done
  for file in "${clim_files[@]}"; do
      [[ ! -e ${file#*> } ]] && ln -sf $(sed 's/ *=> */ /' <<< "${ini_data_dir}/${clim_subdir}/${nem_grid}/$file")
  done
  for file in "${clim_files[@]}"; do
      [[ ! -e ${file#*> } ]] && ln -sf $(sed 's/ *=> */ /' <<< "${ini_data_dir}/${clim_subdir}/$file")
  done
  for file in "${forcing_files[@]}"; do
      [[ ! -e ${file#*> } ||  "$file" == \** ]] && ln -sf $(sed 's/ *=> */ /' <<< "${ini_data_dir}/${forcing_subdir}/${nem_forcing_set}/$file")
  done
  for file in "${shared_files[@]}"; do
      [[ ! -e ${file#*> } ]] && ln -sf $(sed 's/ *=> */ /' <<< "${shared_dir}/$file")
  done

else
  cd ${run_dir}
  shopt -s nullglob
  for v in grid_U grid_V grid_W grid_T icemod SBC SBC_scalar diaptr2D diaptr3D
  do
    for f in ${exp_name}_??_????????_????????_${v}_????.nc
    do
      rm -f "$f"
    done
    for f in ${exp_name}_??_????????_????????_${v}.nc
    do
      rm -f "$f"
    done
    for f in ${exp_name}_??_${v}.nc
    do
      rm -f "$f"
    done
  done
  for f in ocean.output time.step ; do rm -f "${f}"; done
  shopt -u nullglob

fi

run_start_date=$(date -uR -d "${run_start_date}")
run_end_date="${run_start_date} + ${run_duration:?}"
run_end_date=$(date -uR -d "${run_end_date}")

run_start_epoch=$(date -u -d"${run_start_date}" +%s)
run_end_epoch=$(date -u -d"${run_end_date}" +%s)

for (( ; run_num_legs>0 ; run_num_legs-- ))
do

  [[ -r "${info_file:?}" ]] && source "${info_file:?}"
  leg_start_date=${leg_end_date:-$run_start_date}
  leg_number=$((${leg_number:=0}+1))
  leg_start_epoch=$(date -u -d "${leg_start_date}" +%s)
  leg_end_epoch=$(date -u -d "${leg_start_date:?} + ${rst_freq:=$run_duration}" +%s)
  leg_end_date=$(date -uR -d@"${leg_end_epoch}")
  leg_length_sec=$(( leg_end_epoch - leg_start_epoch ))
  leg_start_sec=$(( leg_start_epoch - run_start_epoch ))
  leg_end_sec=$(( leg_end_epoch - run_start_epoch ))
  leg_start_date_yyyymmdd=$(date -u -d "${leg_start_date}" +%Y%m%d) 
  leg_length_sec=$(( leg_length_sec - $(leap_days "${leg_start_date}" "${leg_end_date}")*24*3600 ))
  leg_start_sec=$(( leg_start_sec - $(leap_days "${run_start_date}" "${leg_start_date}")*24*3600 ))
  leg_end_sec=$(( leg_end_sec - $(leap_days "${run_start_date}" "${leg_end_date}")*24*3600 ))
  (( leg_number > 1 )) && leg_is_restart=true || leg_is_restart=false

  (( leg_end_epoch > run_end_epoch )) && leg_end_date=${run_end_epoch}

  if (( leg_start_epoch >= run_end_epoch ))
  then
    echo "Leg start date equal to or after end of simulation."
    echo "Nothing left to do. Cleaning and exiting."
    for (( n=0 ; n<nem_numproc ; n++ ))
    do
      np=$(printf %04d ${n})
      rm -f "restart_oce_${np}.nc"
      rm -f "restart_ice_${np}.nc"
      rm -f "restart_icb_${np}.nc"
    done
    exit 0
  fi

  source build_namelist_cfg.sh > namelist_cfg

  ns=$(printf %08d $(( leg_start_sec / nem_time_step_sec - nem_restart_offset )))
  echo "ns=$ns"
  if ((leg_start_sec > 0 )); then
    for (( n=0 ; n<nem_numproc ; n++ ))
    do
      np=$(printf %04d ${n})
      formatted_leg_number=$(printf %03d $((leg_number)))
      [[ -f "${exp_name:?}_${ns}_restart_oce_${np}.nc" ]] || { cp $archive_dir/restart/${formatted_leg_number}/*oce* . ; } 
      [[ -f "${exp_name:?}_${ns}_restart_oce_${np}.nc" ]] || { echo "Error: restart file not found." ; exit 2 ; }
      ln -fs "${exp_name:?}_${ns}_restart_oce_${np}.nc" "restart_oce_${np}.nc"
      [[ -f "${exp_name:?}_${ns}_restart_ice_${np}.nc" ]] || { cp $archive_dir/restart/${formatted_leg_number}/*ice* . ; }
      [[ -f "${exp_name:?}_${ns}_restart_ice_${np}.nc" ]] || { echo "Error: restart file not found." ; exit 2 ; }
      ln -fs "${exp_name:?}_${ns}_restart_ice_${np}.nc" "restart_ice_${np}.nc"
      [[ -f "${exp_name:?}_${ns}_restart_icb_${np}.nc" ]] || { cp $archive_dir/restart/${formatted_leg_number}/*icb* . ; }
      [[ -f "${exp_name:?}_${ns}_restart_icb_${np}.nc" ]] || { echo "Error: restart file not found." ; exit 2 ; }
      ln -fs "${exp_name:?}_${ns}_restart_icb_${np}.nc" "restart_icb_${np}.nc"
    done
  fi

  [[ $@ == *preponly* ]] && exit 0

  time_begin=$(date +%s)
  ulimit -s unlimited
  if [[ "${SLURM_JOB_NAME-"local"}" == "local" ]] ; then
    echo "!!! Local RUN !!!"
    #xio_numproc=2
    #nem_numproc=24
  fi
  #
  echo "run dir : $run_dir"
  echo "leg_number : $leg_number"
  #echo "ulimit -s unlimited"
  #echo "Lemaitre3-2018: I_MPI_FABRICS=tcp mpirun -np ${xio_numproc} ./${xio_exe} : -np ${nem_numproc} ./${nem_exe}"
  #echo "Lemaitre3>2019: I_MPI_FABRICS=ofi FI_PROVIDER=tcp mpirun -np ${xio_numproc} ./${xio_exe} : -np ${nem_numproc} ./${nem_exe}"
  #echo "Nic5: I_MPI_HYDRA_TOPOLIB=ipl I_MPI_FABRICS=ofi mpirun -np ${xio_numproc} ./${xio_exe} : -np ${nem_numproc} ./${nem_exe}"
  #echo "Zenobe: mpirun -np ${xio_numproc} ./${xio_exe} : -np ${nem_numproc} ./${nem_exe}"
  #echo "Cyclone: I_MPI_FABRICS=tcp mpirun -np "${xio_numproc:?}" "./${xio_exe:?}" : -np "${nem_numproc:?}" "./${nem_exe:?}"
  #echo "LUMI: srun --multi-prog prog.conf (SLURM_JOB_NUM_NODES:${SLURM_JOB_NUM_NODES-$LOCAL_NODES} SLURM_CPUS_ON_NODE:${SLURM_CPUS_ON_NODE-$NB_CORES_PER_NODES})"
  echo "LUCIA: srun srun_wrapper.sh (SLURM_JOB_NUM_NODES:${SLURM_JOB_NUM_NODES-$LOCAL_NODES} SLURM_CPUS_ON_NODE:${SLURM_CPUS_ON_NODE-$NB_CORES_PER_NODES})"

  export OMP_NUM_THREADS=1
  #export MKL_NUM_THREADS=1
  #export PMI_NO_PREINITIALIZE=y
  export TIME="launch timing : %e elapsed %U user %S system"

  # Split XIOS nodes
  #cat /dev/null > prog.conf
  #nem_numproc_slice=$(($nem_numproc/${SLURM_JOB_NUM_NODES-$LOCAL_NODES}))
  #nem_numproc_slice_0=$(($nem_numproc_slice-1))
  #xio_numproc_slice_0=$(($xio_numproc/${SLURM_JOB_NUM_NODES-$LOCAL_NODES}-1))
  #xio_numproc_slice=$(($nem_numproc_slice+$xio_numproc_slice_0))
  #proc_id=0
  #for i in $(eval echo "{1..${SLURM_JOB_NUM_NODES-$LOCAL_NODES}}")
  #do
  #  for j in $(eval echo "{0..$nem_numproc_slice_0}")
  #  do
  #    echo "$proc_id ./${nem_exe}" >> prog.conf
  #    proc_id=$(($proc_id+1))
  #  done
  #  for j in $(eval echo "{$nem_numproc_slice..$xio_numproc_slice}")
  #  do
  #    echo "$proc_id ./${xio_exe}" >> prog.conf
  #    proc_id=$(($proc_id+1))
  #  done
  #done
  # Group XIOS nodes
  cat /dev/null > prog.conf
  proc_id=0
  for i in $(eval echo "{1..${nem_numproc}}")
  do
    echo "$proc_id ./${nem_exe}" >> prog.conf
    proc_id=$(($proc_id+1))
  done
  for i in $(eval echo "{1..${xio_numproc}}")
  do
    echo "$proc_id ./${xio_exe}" >> prog.conf
    proc_id=$(($proc_id+1))
  done
  #echo "LUMI: srun --kill-on-bad-exit=1 --multi-prog prog.conf"

  #cat /dev/null > ./ztask_file.conf
  #echo "0-$(($xio_numproc-1)) ./${xio_exe}" >> ./ztask_file.conf
  #echo "$xio_numproc-$(($xio_numproc+$nem_numproc-1)) ./${nem_exe}" >> ./ztask_file.conf
  #BINDING=map_cpu:$LIST_CORES_SOCKET
  #echo "LUMI: srun --kill-on-bad-exit=1 --mpi=pmi2 -m cyclic --cpu_bind=$BINDING --multi-prog ./ztask_file.conf"

  #exit
  echo $time_begin
  #mpirun -np ${xio_numproc} ./${xio_exe} : -np ${nem_numproc} ./${nem_exe}
  srun --kill-on-bad-exit=1 --multi-prog prog.conf
  #srun --kill-on-bad-exit=1 --mpi=pmi2 -m cyclic --cpu_bind=$BINDING --multi-prog ./ztask_file.conf
  #srun --kill-on-bad-exit=1 ./${xio_exe} : ./${nem_exe}
  time_end=$(date +%s)
  echo $time_end

  formatted_leg_number=$(printf %03d $((leg_number)))
  outdir="${archive_dir:?}/output/${formatted_leg_number}"
  mkdir -p "${outdir}"

  shopt -s nullglob
  for v in grid_U grid_V grid_W grid_T icemod SBC SBC_scalar diaptr2D diaptr3D
  do
    for f in ${exp_name}_??_????????_????????_${v}_????.nc
    do
      mv "$f" "$outdir/"
    done
    for f in ${exp_name}_??_????????_????????_${v}.nc
    do
      mv "$f" "$outdir/"
    done
    for f in ${exp_name}_??_${v}.nc
    do
      mv "$f" "$outdir/"
    done
  done

  outdir="$archive_dir/restart/${formatted_leg_number}"
  mkdir -p "${outdir}"

  for f in ${exp_name}_${ns}_restart_???_????.nc
  do
    [ -f "$f" ] && mv "$f" "${outdir}"
  done

  outdir="$archive_dir/log/${formatted_leg_number}"
  mkdir -p "${outdir}"
  for f in ocean.output time.step ; do mv "${f}" "${outdir}"; done
  cp -f namelist_ice_ref namelist_ice_cfg namelist_ref namelist_cfg ${archive_dir}
  [[ -f ${start_dir}/${SLURM_JOB_NAME-"run"}.sh ]] && cp -f ${start_dir}/${SLURM_JOB_NAME-"run"}.sh ${archive_dir}

  shopt -u nullglob

  tr=$(date -d "0 -$time_begin sec + $time_end sec" +%T) 
  current_date=$(date +'%F %T')
  {
    echo "#"
    echo "# Finished leg at ${current_date} after ${tr} (hh:mm:ss)" 
    echo "leg_number=${leg_number}"
    echo "leg_start_date=\"${leg_start_date}\""
    echo "leg_end_date=\"${leg_end_date}\""
  } | tee -a "${info_file}"

  special_restart=false
done
cd - >/dev/null
[[ $@ == *noresubmit* ]] && exit 0
if (( leg_end_epoch < run_end_epoch )) ; then
  echo "Leg end earlier than end of simulation."
  echo "Submitting another job."
  if [[ "$@" == *"run"* ]] ; then
    exec "$0" "$@"
  elif hash sbatch 2>/dev/null; then
    # Need to go to start_dir to find the run script
    cd ${start_dir}
    echo "sbatch -N ${SLURM_JOB_NUM_NODES-"1"} -o ${run_dir}/$(basename ${stdout_file}).$(printf %03d $((leg_number+1))) -e ${run_dir}/$(basename ${stdout_file}).$(printf %03d $((leg_number+1))) -d ${SLURM_JOB_ID-"id"} ./${SLURM_JOB_NAME-"run"}.sh"
    # Submit command
    # Note: This does not work if you specify a job name with sbatch -J jobname!
    sbatch -N ${SLURM_JOB_NUM_NODES-"1"}                                             \
      -o ${run_dir}/$(basename ${stdout_file}).$(printf %03d $((leg_number+1))) \
      -e ${run_dir}/$(basename ${stdout_file}).$(printf %03d $((leg_number+1))) \
      -d ${SLURM_JOB_ID-"id"}                                                   \
      ./${SLURM_JOB_NAME-"run"}.sh
    #
  else
    cd ${start_dir}
    echo "qsub ${PBS_JOBNAME}.sh" 
    qsub ./${PBS_JOBNAME}.sh
  fi
else
  echo "Nothing left to do. Cleaning and exiting." # FIXME Factorize this (we have two exit points)
  for (( n=0 ; n<nem_numproc ; n++ ))
  do
    np=$(printf %04d ${n})
    rm -f "restart_oce_${np}.nc"
    rm -f "restart_ice_${np}.nc"
    rm -f "restart_icb_${np}.nc"
  done
fi

exit 0