add tile-dist.sh

parallel-stats-profiling.py

@@ -155,18 +155,19 @@ def stats(self, flnm):
       print('Filename ' + flnm + ' does not exit. Stop')
       sys.exit(-1)
 
+    print('flnm:', flnm)
     par_stats = {}
     with open(flnm) as fp:
       lines = fp.readlines()
      #line = fp.readline()
       num_lines = len(lines)
-     #print('Total number of lines: ', num_lines)
+      print('Total number of lines: ', num_lines)
 
       nl = 0
       while(nl < num_lines):
         if(lines[nl].find('Parallel Timing Statistics') > 0):
-         #if(self.debug):
-         #  print('Start Parallel Timing Statistics')
+          if(self.debug):
+            print('Start Parallel Timing Statistics')
           nl, par_stats = self.parallel_time_stats(lines, nl)
           nl += num_lines
         nl += 1
@@ -190,9 +191,9 @@ def parallel_time_stats(self, lines, nl):
      #print('Line ' + str(ns) + ': ' + line)
 
       item = line.split(': ')
-     #print(item)
+     #print('item=', item)
       namestr = item[0].strip()
-     #print(namestr)
+     #print('namestr:', namestr)
       if(namestr.find('OOPS_STATS ') >= 0):
         name = namestr[headleng:]
       else:
@@ -202,6 +203,7 @@ def parallel_time_stats(self, lines, nl):
       while(tstr.find('  ') > 0):
         tstr = tstr.replace('  ', ' ')
       tlist = tstr.split(' ')
+     #print('tlist:', tlist)
 
       tinfo = {}
       tinfo['min'] = float(tlist[0])
@@ -210,7 +212,7 @@ def parallel_time_stats(self, lines, nl):
      #tinfo['percent'] = float(tlist[3])
      #tinfo['imbalance'] = float(tlist[4])
 
-     #print('name: %s, avg: %f' %(name, tinfo['avg']))
+      print('name: %s, max: %f' %(name, tinfo['max']))
 
       stats[name] = tinfo
 
@@ -255,12 +257,18 @@ def get_main_statstime(self, funclabels, funclist):
     statspercent = np.zeros((il, kl))
     for k in range(kl):
       stats = self.parstatslist[k]
+      print('stats.keys():', stats.keys())
       for i in range(il):
         name = self.funclist[i]
         if(name in stats.keys()):
           statstime[i][k] = stats[name]['max']*0.001/60.0
+          print('k: %d, i: %d, %s: %f' %(k, i, name, statstime[i][k]))
+
+      if('util::Timers::Total' in stats.keys()):
+        totalmax = stats['util::Timers::Total']['max']*0.001/60.0
+      else:
+        totalmax = 1000.0
 
-      totalmax = stats['util::Timers::Total']['max']*0.001/60.0
       for i in range(il):
         statspercent[i][k] = 100.0*statstime[i][k]/totalmax
 

tile-dist.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+ set -x
+
+ temp_dir=/work2/noaa/gsienkf/weihuang/jedi/run
+ nodelist=(1 2 4 8)
+ corelist=(36 72 144 288)
+
+ for i in ${!nodelist[@]}
+ do
+   workdir=${temp_dir}/run_80.40t${nodelist[$i]}n_${corelist[$i]}p/stdoutNerr
+   imgname=dist_${nodelist[$i]}n.png
+
+   python tile-dist.py --output=1 \
+	--workdir=${workdir} \
+	--imagename=${imgname}
+ done
+

timingobserver

@@ -9,7 +9,10 @@
  workdir=/work2/noaa/gsienkf/weihuang/jedi/per_core_timing/run
 #caselist=(halo_aircraft)
 #caselist=(anna_roundRobin_aircraft)
- caselist=(anna_halo_aircraft anna_roundRobin_aircraft halo_aircraft roundRobin_aircraft)
+#caselist=(anna_halo_aircraft anna_roundRobin_aircraft halo_aircraft roundRobin_aircraft)
+#caselist=(anna_halo_aircraft_2 anna_roundRobin_aircraft_2)
+#caselist=(solver_halo_aircraft)
+ caselist=(solver_RoundRobin_aircraft)
 
  for i in ${!caselist[@]}
  do