Modified locations of Exec scripts. Created script for preprocessing data.

BashScripts/createInitialNodelist.sh

+#!/bin/bash
+# Runs in a given current directory all .ini files
+# Parameter 1(Optional) - Amount of executions per file. Must be a positive number
+#====== Do not modify these values =======
+
+numP=$1
+cores=$2
+nodelist=$3
+
+initial_node_qty=$(($numP / $cores))
+if [ $initial_node_qty -eq 0 ]
+then
+  initial_node_qty=1
+fi
+
+common_node_name="n" #FIXME What if it uses another type of node?
+if [[ $nodelist == *"["* ]]; then
+  common_node_name=$(echo $nodelist | cut -d '[' -f1)
+fi
+
+node_array=($(echo $nodelist | sed -e 's/[\[n]//g' -e 's/\]/ /g' -e 's/,/ /g'))
+actual_node_qty=0
+for ((i=0; $actual_node_qty<$initial_node_qty; i++))
+do
+  element=($(echo ${node_array[$i]} | sed -e 's/-/ /g'))
+
+  nodes_qty=1
+  if [ "${#element[@]}" -gt 1 ];
+  then
+    nodes_qty=$((10#${element[1]}-10#${element[0]}+1))
+  fi
+
+  expected_node_qty=$(($actual_node_qty + $nodes_qty))
+  if [ "$expected_node_qty" -le "$initial_node_qty" ];
+  then
+    added_qty=$nodes_qty
+    actual_node_qty=$expected_node_qty
+  else
+    added_qty=$(($initial_node_qty - $actual_node_qty))
+    actual_node_qty=$initial_node_qty
+  fi
+
+  for ((j=0; j<$added_qty; j++))
+  do
+    index=$((10#${element[0]} + $j))
+    index=0$index # FIXME What if there are more than 9 nodes?
+    #FIXME What if less than $cores have to be spawned?
+    for ((core=0; core<$cores; core++)) # FIXME What if the user asks for a spread distribution
+    do
+      initial_nodelist="${initial_nodelist:+$initial_nodelist,}"$common_node_name$index
+    done
+  done
+done
+
+#Print result
+echo $initial_nodelist

Exec/MallTimes.py

+import sys
+import glob
+import numpy as np
+import pandas as pd
+from enum import Enum
+
+class G_enum(Enum):
+    TOTAL_RESIZES = 0
+    TOTAL_GROUPS = 1
+    SDR = 2
+    ADR = 3
+    DR = 4
+    RED_METHOD = 5
+    RED_STRATEGY = 6
+    SPAWN_METHOD = 7
+    SPAWN_STRATEGY = 8
+    GROUPS = 9
+    ITERS = 10
+    T_SPAWN = 11
+    T_SR = 12
+    T_AR = 13
+    T_MALLEABILITY = 14
+    T_TOTAL = 15
+    #Malleability specific
+    NP = 0
+    NC = 1
+
+
+columnsG = ["Total_Resizes", "Total_Groups", "SDR", "ADR", "DR", "Redistribution_Method", \
+            "Redistribution_Strategy", "Spawn_Method", "Spawn_Strategy", "Groups", \
+            "Iters", "T_spawn", "T_SR", "T_AR", "T_Malleability", "T_total"] #16
+
+#-----------------------------------------------
+# Obtains the value of a given index in a splited line
+# and returns it as a float values if possible, string otherwise
+def get_value(line, index, separator=True):
+  if separator:
+    value = line[index].split('=')[1].split(',')[0]
+  else:
+    value = line[index]
+
+  try:
+    value = float(value)
+    if value.is_integer():
+      value = int(value)
+  except ValueError:
+    return value
+  return value
+
+#-----------------------------------------------
+def record_config_line(lineS, aux_data):
+  aux_data[0] = int(lineS[1].split("=")[1])
+  aux_data[1] = int(lineS[2].split("=")[1])
+  aux_data[2] = int(lineS[3].split("=")[1])
+  aux_data[3] = int(lineS[4])
+  aux_data[4] = int(lineS[5])
+  aux_data[5] = int(lineS[7].split("=")[1])
+#                           SM RM SS
+#          0        1         2 3 4             5
+#Test numP=160 numC=120 Meths=0 0 1 -- Is_synch=1 qty=5
+#  0      1        2         3  4 5 6  7          8
+
+
+#-----------------------------------------------
+# Obtains the general parameters of an execution and
+# stores them for creating a global dataframe
+def record_new_line(lineS, dataG_it, data_aux):
+  offset_line = 2
+  sdr = 3947883504
+  adr = sdr * 0.966
+
+  dataG_it[G_enum.TOTAL_RESIZES.value] = 1
+  dataG_it[G_enum.TOTAL_GROUPS.value] = dataG_it[G_enum.TOTAL_RESIZES.value]+1
+
+  # Init lists for each column
+  array_groups = [G_enum.GROUPS.value, G_enum.ITERS.value, G_enum.RED_METHOD.value, \
+          G_enum.RED_STRATEGY.value, G_enum.SPAWN_METHOD.value, G_enum.SPAWN_STRATEGY.value]
+
+  array_resizes = [ G_enum.T_SPAWN.value, G_enum.T_SR.value, G_enum.T_AR.value, G_enum.T_MALLEABILITY.value]
+
+  for index in array_groups:
+    dataG_it[index] = [None]*dataG_it[G_enum.TOTAL_GROUPS.value]
+  for index in array_resizes:
+    dataG_it[index] = [None]*dataG_it[G_enum.TOTAL_RESIZES.value]
+
+  dataG_it[G_enum.GROUPS.value][0] = data_aux[0]
+  dataG_it[G_enum.GROUPS.value][1] = data_aux[1] 
+  dataG_it[G_enum.ITERS.value][0] = dataG_it[G_enum.ITERS.value][1] = 500
+
+  dataG_it[G_enum.SPAWN_METHOD.value][0] = dataG_it[G_enum.RED_METHOD.value][0] = 0
+  dataG_it[G_enum.SPAWN_METHOD.value][1] = data_aux[2]
+  dataG_it[G_enum.RED_METHOD.value][1] = data_aux[3]
+  dataG_it[G_enum.SPAWN_STRATEGY.value][0] = dataG_it[G_enum.RED_STRATEGY.value][0] = 1
+  dataG_it[G_enum.SPAWN_STRATEGY.value][1] = dataG_it[G_enum.RED_STRATEGY.value][1] = data_aux[4]
+
+  dataG_it[G_enum.SDR.value] = sdr
+  dataG_it[G_enum.ADR.value] = 0
+  if data_aux[5] == 0: #Is asynch send
+    dataG_it[G_enum.ADR.value] = adr
+    dataG_it[G_enum.SDR.value] -= adr
+  dataG_it[G_enum.DR.value] = dataG_it[G_enum.SDR.value] + dataG_it[G_enum.ADR.value]
+
+#-----------------------------------------------
+def record_time_line(lineS, dataG_it):
+  T_names = ["T_spawn:", "T_SR:", "T_AR:", "T_Malleability:", "T_total:"]
+  T_values = [G_enum.T_SPAWN.value, G_enum.T_SR.value, G_enum.T_AR.value, G_enum.T_MALLEABILITY.value, G_enum.T_TOTAL.value]
+  if not (lineS[0] in T_names): # Execute only if line represents a Time
+      return
+
+  index = T_names.index(lineS[0])
+  index = T_values[index]
+  offset_lines = 1
+
+  len_index = 1
+  if dataG_it[index] != None:
+    len_index = len(dataG_it[index])
+    for i in range(len_index):
+      dataG_it[index][i] = get_value(lineS, i+offset_lines, False)
+  else:
+      dataG_it[index] = get_value(lineS, offset_lines, False)
+
+#-----------------------------------------------
+def read_global_file(f, dataG, it):
+  runs_in_file=0
+  aux_data = [0,0,0,0,0,0]
+  for line in f: 
+    lineS = line.split()
+
+    if len(lineS) > 0:
+      if lineS[0] == "Test": # CONFIG LINE
+        record_config_line(lineS, aux_data)
+      elif lineS[0] == "Start":
+        it += 1
+        runs_in_file += 1
+        group = 0
+
+        dataG.append([None]*len(columnsG))
+        record_new_line(lineS, dataG[it], aux_data)
+      elif it>-1:
+        print(lineS)
+        print("== "+ str(it) + "  "+str(runs_in_file) )
+        record_time_line(lineS, dataG[it])
+
+  return it,runs_in_file
+
+#-----------------------------------------------
+def convert_to_tuples(dfG):
+  array_list_items = [G_enum.GROUPS.value, G_enum.ITERS.value, \
+          G_enum.RED_METHOD.value, G_enum.RED_STRATEGY.value, G_enum.SPAWN_METHOD.value, \
+          G_enum.SPAWN_STRATEGY.value, G_enum.T_SPAWN.value, G_enum.T_SR.value, \
+          G_enum.T_AR.value, G_enum.T_MALLEABILITY.value]
+
+  for item in array_list_items:
+    name = columnsG[item]
+    values = dfG[name].copy()
+    for index in range(len(values)):
+      values[index] = tuple(values[index])
+    dfG[name] = values
+
+#-----------------------------------------------
+
+if len(sys.argv) < 2:
+    print("The files name is missing\nUsage: python3 MallTimes.py commonName directory OutName")
+    exit(1)
+
+common_name = sys.argv[1]
+if len(sys.argv) >= 3:
+    BaseDir = sys.argv[2]
+    print("Searching in directory: "+ BaseDir)
+else:
+    BaseDir = "./"
+
+if len(sys.argv) >= 4:
+  name = sys.argv[3]
+else:
+  name = "data"
+print("File name will be: " + name + "G.pkl")
+
+lista = (glob.glob(BaseDir + common_name + "*.out")) # Se utiliza cuando solo hay un nivel de directorios
+print("Number of files found: "+ str(len(lista)));
+
+it = -1
+dataG = []
+
+for elem in lista:
+  f = open(elem, "r")
+  it,runs_in_file = read_global_file(f, dataG, it)
+  f.close()
+
+dfG = pd.DataFrame(dataG, columns=columnsG)
+convert_to_tuples(dfG)
+print(dfG)
+dfG.to_pickle(name + 'G.pkl')
+
+#dfM = pd.DataFrame(dataM, columns=columnsM)
+
+#Poner en TC el valor real y en TH el necesario para la app
+#cond = dfM.TH != 0
+#dfM.loc[cond, ['TC', 'TH']] = dfM.loc[cond, ['TH', 'TC']].values
+#dfM.to_csv(name + 'M.csv')

generalRun.sh → Exec/generalRun.sh

@@ -7,7 +7,6 @@
 dirCG="/home/martini/malleable_cg"
 
 module load /home/martini/MODULES/modulefiles/mpich-4.0.3-ofi
-module load /home/martini/MODULES/modulefiles/intel64Only.module
 
 nodelist=$SLURM_JOB_NODELIST
 nodes=$SLURM_JOB_NUM_NODES
@@ -28,3 +27,6 @@ do
   mpirun -hosts $initial_nodelist -np $numP $dirCG/build/a.out $matrix $numC $msm $mss $mrm $mss $is_sync $nodelist $nodes
 done
 echo "End"
+
+sed -i 's/application called MPI_Abort(MPI_COMM_WORLD, -100) - process/shrink cleaning/g' slurm-$SLURM_JOB_ID.out
+sed -i 's/Abort(-100)/shrink cleaning/g' slurm-$SLURM_JOB_ID.out

runBase.sh → Exec/runBase.sh

@@ -2,7 +2,7 @@
 
 #SBATCH -N 1
 #SBATCH -p P1
-#SBATCH -t 00:10:00
+#SBATCH -t 00:30:00
 
 dirM="/home/martini/SparseMatrix/"
 dirCG="/home/martini/malleable_cg"
@@ -11,27 +11,20 @@ matrix="Queen_4147.rb"
 #matrix="bcsstk01.rsa"
 
 module load /home/martini/MODULES/modulefiles/mpich-4.0.3-ofi
-module load /home/martini/MODULES/modulefiles/intel64Only.module
 
 nodelist=$SLURM_JOB_NODELIST
 nodes=$SLURM_JOB_NUM_NODES
 cores=20
 numP=$1
-numC=40
-msm=0
-mss=1
-mrm=0
-send_sync=1
+numC=$2
+msm=$3
+mss=$5
+mrm=$4
+send_sync=$6
 
 
-initial_nodelist=$(bash BashScripts/createInitialNodelist.sh $numP $cores $nodelist)
-echo "Test"
-mpirun -hosts $initial_nodelist -np $numP ./build/a.out $dirM/$matrix $numC $msm $mss $mrm $mss $send_sync $nodelist $nodes
+initial_nodelist=$(bash $dirCG/BashScripts/createInitialNodelist.sh $numP $cores $nodelist)
+echo "Test numP=$numP numC=$numC Meths=$msm $mrm $mss -- Is_synch=$send_sync qty=$qty"
+mpirun -hosts $initial_nodelist -np $numP $dirCG/build/a.out $dirM/$matrix $numC $msm $mss $mrm $mss $send_sync $nodelist $nodes
 #mpirun -np 4 ./ConjugateGradient bcsstk17.rsa
-#mpirun -np 8 ./ConjugateGradient bcsstk01.rsa
-#mpirun -np 12 ./ConjugateGradient bcsstk01.rsa
-echo "AUDIK"
-#mpirun -np 4 ./ConjugateGradient ../../audikw_1/audikw_1.rb
-#mpirun -np 4 ./ConjugateGradient ../../audikw_1/audikw_1.rb
-#mpirun -np 4 ./ConjugateGradient ../../audikw_1/audikw_1.rb
 echo "End"

runOFI.sh → Exec/runOFI.sh

@@ -6,8 +6,8 @@ matrix="Queen_4147.rb"
 #matrix="audikw_1.rb"
 #matrix="bcsstk01.rsa"
 
-procs=(2 10 20 40 80 120 160)
-#procs=(2 4)
+#procs=(2 10 20 40 80 120 160)
+procs=(160 120 80 40 20 10 2)
 msm=(0 1)
 mss=(1 2)
 mrm=(0 1)
@@ -46,11 +46,11 @@ do
           do
             if [ $is_sync -eq 1 ] # Matrix is send syncrhonously
             then
-              sbatch -p P1 -N $node_qty $dirCG/generalRun.sh $proc $dirM$matrix $proc_c $sm_type 1 $rm_type 1 $is_sync $qty
+              sbatch -p P1 -N $node_qty $dirCG/Exec/generalRun.sh $proc $dirM$matrix $proc_c $sm_type 1 $rm_type 1 $is_sync $qty
   	    else # Matrix is send asyncrhonously
               for ss_type in "${mss[@]}"
               do
-                sbatch -p P1 -N $node_qty $dirCG/generalRun.sh $proc $dirM$matrix $proc_c $sm_type $ss_type $rm_type $ss_type $is_sync $qty
+                sbatch -p P1 -N $node_qty $dirCG/Exec/generalRun.sh $proc $dirM$matrix $proc_c $sm_type $ss_type $rm_type $ss_type $is_sync $qty
 	      done
  	    fi
           done

Exec/runValgrind.sh

+#!/bin/bash
+
+#SBATCH -N 1
+#SBATCH -p P1
+#SBATCH -t 00:10:00
+
+module load /home/martini/MODULES/modulefiles/mpich-4.0.3-ofi
+module load /home/martini/MODULES/modulefiles/intel64Only.module
+
+nodelist=$SLURM_JOB_NODELIST
+nodes=$SLURM_JOB_NUM_NODES
+cores=20
+numP=$1
+numC=2
+msm=0
+mss=2
+mrm=1
+send_sync=0
+
+
+initial_nodelist=$(bash BashScripts/createInitialNodelist.sh $numP $cores $nodelist)
+echo "Test"
+mpirun -hosts $initial_nodelist -np $numP valgrind --leak-check=full --show-leak-kinds=all --track-origins=yes --trace-children=yes --log-file=nc.vg.%p ./build/a.out bcsstk01.rsa $numC $msm $mss $mrm $mss $send_sync