#!/bin/ksh ############################################################################### # CPTEC OPERATION launcher – EnvironmentalVariablesMCGA.ksh # # # # Spectral resolution : ZZZZ (4 digits) | Vertical resolution : XXX (3 digits)# ############################################################################### # Usage examples # # source ./EnvironmentalVariablesMCGA.ksh # normal mode # # source ./EnvironmentalVariablesMCGA.ksh -n # DRY-RUN (no actions) # ############################################################################### #IFS=$'\n\t' ############################################################################### # Global flags ############################################################################### DRYRUN=false # set to 'true' when -n/--dry-run is passed verbose=false # set to 'true' when -v is passed ############################################################################### # Helper: printf wrapper that honours DRYRUN ############################################################################### _run() { if ${DRYRUN}; then printf '[DRY-RUN] %s\n' "$*" else eval "$*" fi } ############################################################################### # Function: check_cmd # ############################################################################### check_cmd() { if ! command -v "$1" > /dev/null 2>&1; then echo "ERROR: Comando '$1' não encontrado." exit 1 fi } ############################################################################### # Function: detect_hpc_system # ############################################################################### detect_hpc_system() { # keep existing variable names args=$@ # save arguments while (( $# )); do # local verbose override case $1 in -v) verbose=true ;; esac shift done set -- ${args} local sys_info sys_info=$(uname -a) local short_hostname short_hostname=$(hostname -s) if printf '%s' "$sys_info" | grep -qi 'cray_ari_s'; then export hpc_system="cray" export hpc_name="xc50" export WRAPPER="ftn" [[ ${verbose} == true ]] && printf '[INFO] Detected: Cray XC50\n' return 0 elif printf '%s' "$sys_info" | grep -qi 'egeon'; then export hpc_system="linux" export hpc_name="egeon" export WRAPPER="mpif90" export LC_ALL="en_US.UTF-8" [[ ${verbose} == true ]] && printf '[INFO] Detected: EGEON Cluster\n' return 0 elif printf '%s' "$short_hostname" | grep -qi 'headnode'; then export hpc_system="linux" export hpc_name="egeon" export WRAPPER="mpif90" export LC_ALL="en_US.UTF-8" [[ ${verbose} == true ]] && \ printf '[WARN] Detected: HEADNODE EGEON Cluster (build-only)\n' return 0 else printf '[ERROR] Unknown machine: %s\n' "$short_hostname" printf '[ACTION] 1) Add the machine under etc/mach/\n' printf '[ACTION] 2) Create an entry in copy_fixed_files (etc/smg_setup.sh)\n' return 1 fi } ############################################################################### # Function: getBAMSize # ############################################################################### getBAMSize() { local trunc=$1 case ${trunc} in 21) IM=64; JM=32; TimeStep=3600 ;; # 1 step /h 31) IM=96; JM=48; TimeStep=1800 ;; # 2 steps/h 42) IM=128; JM=64; TimeStep=1200 ;; # 3 steps/h 62) IM=192; JM=96; TimeStep=900 ;; # 4 steps/h 106) IM=320; JM=160; TimeStep=600 ;; # 6 steps/h 126) IM=384; JM=192; TimeStep=450 ;; # 8 steps/h 133) IM=400; JM=200; TimeStep=450 ;; 159) IM=480; JM=240; TimeStep=360 ;; 170) IM=512; JM=256; TimeStep=360 ;; 213) IM=640; JM=320; TimeStep=300 ;; 254) IM=768; JM=384; TimeStep=240 ;; 299) IM=900; JM=450; TimeStep=200 ;; 319) IM=960; JM=480; TimeStep=180 ;; 341) IM=1024; JM=512; TimeStep=180 ;; 382) IM=1152; JM=576; TimeStep=150 ;; 511) IM=1536; JM=768; TimeStep=120 ;; 533) IM=1600; JM=800; TimeStep=120 ;; 666) IM=2000; JM=1000; TimeStep=90 ;; 863) IM=2592; JM=1296; TimeStep=60 ;; 1279) IM=3840; JM=1920; TimeStep=20 ;; 1332) IM=4000; JM=2000; TimeStep=20 ;; *) printf '[ERROR] Unknown spectral truncation: %s\n' "${trunc}"; return 1 ;; esac } #-------------------------------------------------------------------# # getMPIinfo # #-------------------------------------------------------------------# #BOP # !DESCRIPTION: # Derives a consistent MPI × OpenMP layout (tasks, threads, nodes) # for the current HPC system. Values are exported in variables # MPITasks, ThreadsPerMPITask, TotalCPUs, Nodes, CoresPerNode # so that downstream scripts (job‑headers, launchers) can map them # 1‑to‑1 onto SLURM options: # --ntasks=$MPITasks (--tasks in some schedulers) # --cpus-per-task=$ThreadsPerMPITask # --nodes=$Nodes # # !INPUT FLAGS (parsed inside the function): # -np | --ntasks # explicit number of MPI ranks # -d | --cpus-per-task # threads per rank (OpenMP) # -N | --nodes # force number of nodes # -c | --cores-per-node # override machine default # -P | --procs # TOTAL cores desired (derive others) # # !MACHINE DEFAULTS (can be overridden by flags or env vars): # EGEON : MPITasks=128 CoresPerNode=128 ThreadsPerMPITask=1 # XC50 : MPITasks=40 CoresPerNode=40 ThreadsPerMPITask=1 # → These defaults yield a **pure‑MPI layout** (one rank per core, no OpenMP). # # !ALGORITHM: # 1) Detect HPC (detect_hpc_system) → sets $hpc_name # 2) Apply machine defaults, then override with user flags # 3) If -P was supplied, back‑compute MPITasks = ⌈P / threads⌉ # 4) Compute: # TotalCPUs = MPITasks × ThreadsPerMPITask # Nodes = ⌈TotalCPUs / CoresPerNode⌉ (unless user gave -N) # 5) Echo a human‑readable summary. # # !RETURN VALUE: # 0 on success; 1 on unknown machine / invalid combo. #EOP getMPIinfo() { #----------------------------------------------------------------- # 1. Detect machine (sets $hpc_name); abort on failure #----------------------------------------------------------------- detect_hpc_system || return 1 #----------------------------------------------------------------- # 2. Machine defaults (pure‑MPI: 1 rank / core) #----------------------------------------------------------------- local hpc=${hpc_name,,} # lower‑case copy case $hpc in egeon) MPITasks=${MPITasks:-128}; CoresPerNode=${CoresPerNode:-128};; xc50) MPITasks=${MPITasks:-40}; CoresPerNode=${CoresPerNode:-40};; *) printf '[ERROR] Unknown HPC: %s\n' "$hpc"; return 1;; esac ThreadsPerMPITask=${ThreadsPerMPITask:-1} MaxCoresPerNode=$CoresPerNode # legacy alias #----------------------------------------------------------------- # 3. Parse user overrides #----------------------------------------------------------------- local TotalCoresRequested='' args=("$@") while (( $# )); do case $1 in -np|--ntasks) MPITasks=$2; shift 2 ;; -d|--cpus-per-task) ThreadsPerMPITask=$2; shift 2 ;; -N|--nodes) Nodes=$2; shift 2 ;; -c|--cores-per-node) CoresPerNode=$2; MaxCoresPerNode=$2; shift 2 ;; -P|--procs) TotalCoresRequested=$2; shift 2 ;; -v|--verbose) verbose=true; shift 1 ;; *) printf '\033[33;1m[WARN]\033[0m Unknown arg: %s\n' "$1"; shift ;; esac done set -- "${args[@]}" # restore $* just in case #----------------------------------------------------------------- # 4. Derive layout when -P (total logical cores) is supplied #----------------------------------------------------------------- if [[ -n $TotalCoresRequested ]]; then (( ThreadsPerMPITask <= 0 )) && ThreadsPerMPITask=1 MPITasks=$(( (TotalCoresRequested + ThreadsPerMPITask - 1) / ThreadsPerMPITask )) fi #----------------------------------------------------------------- # 5. Final totals #----------------------------------------------------------------- TotalCPUs=$(( MPITasks * ThreadsPerMPITask )) # Nodes: user‑forced (-N) or auto ceil‑division : "${Nodes:=$(( (TotalCPUs + CoresPerNode - 1) / CoresPerNode ))}" # Tasks per node – recomputed from final numbers TasksPerNode=$(( (MPITasks + Nodes - 1) / Nodes )) # ceil(MPI / nodes) # Sanity: do not exceed physical capacity local max_tasks_per_node=$(( CoresPerNode / ThreadsPerMPITask )) if (( TasksPerNode > max_tasks_per_node )); then printf '\e[31;1m[ERROR]\e[0m Over‑subscription: %d tasks × %d threads > %d cores / node\n' \ "$TasksPerNode" "$ThreadsPerMPITask" "$CoresPerNode" return 1 fi PEs=$MPITasks # legacy variable if [ "${verbose}" == true ]; then #----------------------------------------------------------------- # 6. User summary #----------------------------------------------------------------- printf '[INFO] MPI/OpenMP layout resolved:\n' printf ' MPI ranks (MPITasks) : %d\n' "$MPITasks" printf ' Threads per rank (-d) : %d\n' "$ThreadsPerMPITask" printf ' Total logical cores : %d\n' "$TotalCPUs" printf ' Cores per node : %d\n' "$CoresPerNode" printf ' Compute nodes : %d\n' "$Nodes" printf ' Tasks per node : %d\n' "$TasksPerNode" printf ' SLURM flags : --nodes=%d --ntasks=%d --cpus-per-task=%d\n' \ "$Nodes" "$MPITasks" "$ThreadsPerMPITask" # PBS/Torque shorthand (optional) printf ' PBS flags : -l select=%d:ncpus=%d:mpiprocs=%d:ompthreads=%d\n' \ "$Nodes" "$CoresPerNode" "$TasksPerNode" "$ThreadsPerMPITask" fi return 0 } #EOC ############################################################################### # Detect HPC system immediately; abort on failure ############################################################################### detect_hpc_system || { printf '[FATAL] detect_hpc_system failed – aborting.\n'; return 1; } ############################################################################### # Paths – DO NOT EDIT the 3 lines below; they are replaced by another script! # ############################################################################### export HOMEBASE="/home/caroline.viezel/SMNA_v3.0.0.t12717/SMG/cptec/bam" export SUBTBASE="/mnt/beegfs/caroline.viezel/SMNA_v3.0.0.t12717/SMG/datainout/bam" export WORKBASE="/mnt/beegfs/caroline.viezel/SMNA_v3.0.0.t12717/SMG/datainout/bam" ############################################################################### ############################################################################### # External data directories (per machine) ############################################################################### case ${hpc_name} in xc50) dadosExternos=/lustre_xc50/ioper/data/external ;; egeon) dadosExternos=/oper/dados/ioper/tempo/NCEP/input dadosUmidSolo=/oper/dados/modelo/umid_solo/brutos ;; eslogin*) dadosExternos=/stornext/oper/tempo/externos/Download ;; esac export dadosExternos dadosUmidSolo ############################################################################### # Executables & utilities ############################################################################### export preEXEC=ParPre_MPI export bamEXEC=ParModel_MPI export posEXEC=ParPos_MPI export caldate=${HOMEBASE}/utils/bin/caldate.3.0.1 export inctime=inctime # shell-helper library (provides inctime built in bash) source "${HOMEBASE}/../../run/smg_functions.sh" # GrADS / wgrib utilities per machine case ${hpc_name} in xc50) DIRGRADS=/opt/grads/2.0.a9/bin wgrib1=/cray_home/local/grads/grads-2.0.2.oga.2/bin/wgrib wgrib2=/cray_home/local/grib2/wgrib2/wgrib2 ;; eslogin*) DIRGRADS=/opt/grads/2.1.a1/bin wgrib1=${DIRGRADS}/wgrib wgrib2=${DIRGRADS}/wgrib2 ;; egeon) DIRGRADS=/opt/ohpc/pub/apps/grads/2.2.1/bin wgrib1=/home/ioper/bin/grib2/wgrib/wgrib wgrib2=/home/ioper/bin/grib2/wgrib2/wgrib2 ;; *) echo -e "\033[31;1m[ERROR]\033[m hpc_name '${hpc_name}' não reconhecido." exit 1 ;; esac export DIRGRADS wgrib1 wgrib2 ############################################################################### # PBS/SLURM generic environment ############################################################################### export QUOTA=CPTEC export PBS_TUPA=eslogin13 export PBS_AUX=pesq export HSTMAQ=$(uname -n) export MAQUI=$(uname -s) export tmstp=$(date +'%s') [[ ${verbose} == true ]] && printf '[INFO] EnvironmentVariablesMCGA loaded (DRY-RUN=%s)\n' "${DRYRUN}"