A Launcher Script

#!/bin/bash

loops_to_do=10

loop=0

my_file=`mktemp -p . seprun.XXXXX` || exit 1
rm $my_file

echo "writing to file $my_file.txt"

while [ $loop -lt $loops_to_do ]
do
   echo "$loop"
   sep -start -c -d 1 -ec "CPU_CLK_UNHALTED.CORE_P:OS=no", "INST_RETIRED.ANY:OS=no" -out $my_file
   loop=$(($loop + 1))
done

taskset

Script to parse SEP output

#!/usr/bin/python
import sys
import re

if len(sys.argv) < 2:
   print "usage: %s (filename)" % sys.argv[0]
   sys.exit(1)

my_file = sys.argv[1]

the_file = open(my_file)

all_clockticks    = []
all_instr_retired = []


this_clockticks      = []
this_instr_retired   = []

processing = False
done = 0
for this_line in the_file.readlines():
#      if not this_line:
 #        done = 1
   if re.compile("^#").match(this_line):
      if processing:
         processing = False
         all_clockticks.append(this_clockticks)
         all_instr_retired.append(this_instr_retired)
         this_clockticks    = []
         this_instr_retired = []
   else:
      if not processing:
         processing = True

      if re.compile("^INST_RETIRED").match(this_line):
         i = this_line.split(",")[2]
         if i == "0":
            i = .0000001
         this_instr_retired.append(float(i))

      elif re.compile("^CPU_CLK_UNHALTED").match(this_line):
         this_clockticks.append(float(this_line.split(",")[2]))

      elif re.compile("[/s]+").match(this_line):
         continue

      else:
         print "Encountered a non-clocktick/instruction retired event"
         print this_line
         sys.exit(1)

all_clockticks.append(this_clockticks)
all_instr_retired.append(this_instr_retired)

clockticks_total   = 0
instr_retired_total = 0

clockticks_per_cpu = []
instr_retired_per_cpu = []
cpi_per_cpu = []
cpi_per_cpu_per_step = []  # one entry per cpu per step
cpi_per_step = []

# compute cpi for each step, both per-cpu and for the whole machine
for i in range(len(all_clockticks)):
   temp = []
   c_temp = float(0)
   i_temp = float(0)
   for j in range(len(all_clockticks[0])):
      c_temp += all_clockticks[i][j]
      i_temp += all_instr_retired[i][j]
      temp.append(float(c_temp / i_temp))
   cpi_per_cpu_per_step.append(temp)
   cpi_per_step.append(float(c_temp/i_temp))

# initialize arrays, 'detecting' the # of processors
for i in range(len(all_clockticks[0])):
   clockticks_per_cpu.append(float(0))
   instr_retired_per_cpu.append(float(0))
   cpi_per_cpu.append(float(0))

# aggregate the clockticks & instr retired for total run stats
for i in range(len(all_clockticks)):
   for j in range(len(clockticks_per_cpu)):
      clockticks_per_cpu[j] += all_clockticks[i][j]
      clockticks_total += all_clockticks[i][j]

for i in range(len(all_instr_retired)):
   for j in range(len(instr_retired_per_cpu)):
      instr_retired_per_cpu[j] += all_instr_retired[i][j]
      instr_retired_total += all_instr_retired[i][j]

# compute per-cpu total run CPI
for i in range(len(clockticks_per_cpu)):
   cpi_per_cpu[i] = clockticks_per_cpu[i] / instr_retired_per_cpu[i]

print "cpi per cpu for each time step"
print cpi_per_cpu_per_step

print "\ncpi per time step, all CPUs"
print cpi_per_step

print "\noverall cpi per cpu"
print cpi_per_cpu

print "\noverall cpi, all CPUs = %f" % float(clockticks_total / instr_retired_total)

-- MattWalsh - 20 Jul 2006