sanchayanmaity/gem5
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Totally re-work the way that jobfiles are done so there is more

Browse source 
information that can be used for other aspects of sending jobs.
New graphing output stuff with matplotlib.

util/pbs/job.py:
    Shuffle code around and create the JobDir class which encapsulates
    all of the functionality needed for making, organizing, and cleaning
    a job directory.
    Better status output
util/pbs/jobfile.py:
    Majory re-working of the jobfile code.
    A job file now consists of several objects that describe how
    jobs should be run, it includes information about checkpoints,
    and graphing.
util/pbs/send.py:
    use the new jobfile code.
    deal with the 15 character limit of pbs by truncating the name and
    using the raj hack.
util/stats/db.py:
    fix the __str__ function for nodes
    provide __getitem__ for the Database class
util/stats/stats.py:
    use the jobfile stuff to figure out what the proper naming
    and organziation of the graphs should be.
    move all output code to output.py, get rid of ploticus and use
    matplotlib

--HG--
rename : util/categories.py => util/stats/categories.py
extra : convert_revision : 0d793cbf6ad9492290e8ec875ce001c84095e1f7
This commit is contained in:
Nathan Binkert 2005-09-17 16:51:26 -04:00
parent f88d7c7456
commit 02098f8e7b
11 changed files with 1606 additions and 644 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

139
util/pbs/job.py
View file

@ -29,10 +29,21 @@
# Steve Reinhardt
# Ali Saidi
import os, os.path, shutil, signal, socket, sys, time
import os, os.path, shutil, signal, socket, sys
from os import environ as env
from os.path import join as joinpath, expanduser
def date():
import time
return time.strftime('%a %b %e %H:%M:%S %Z %Y', time.localtime())
def cleandir(dir):
for root, dirs, files in os.walk(dir, False):
for name in files:
os.remove(joinpath(root, name))
for name in dirs:
os.rmdir(joinpath(root, name))
class rsync:
def __init__(self):
self.sudo = False
@ -61,25 +72,76 @@ class rsync:
return os.spawnvp(os.P_WAIT, args[0], args)
def cleandir(dir):
for root, dirs, files in os.walk(dir, False):
for name in files:
os.remove(joinpath(root, name))
for name in dirs:
os.rmdir(joinpath(root, name))
class JobDir(object):
def __init__(self, dir):
self.dir = dir
def date():
return time.strftime('%a %b %e %H:%M:%S %Z %Y', time.localtime())
def file(self, filename):
return joinpath(self.dir, filename)
def remfile(file):
if os.path.isfile(file):
os.unlink(file)
def create(self):
if os.path.exists(self.dir):
if not os.path.isdir(self.dir):
sys.exit('%s is not a directory. Cannot build job' % self.dir)
else:
os.mkdir(self.dir)
def readval(filename):
file = open(filename, 'r')
value = file.readline().strip()
file.close()
return value
def exists(self):
return os.path.isdir(self.dir)
def clean(self):
cleandir(self.dir)
def hasfile(self, filename):
return os.path.isfile(self.file(filename))
def echofile(self, filename, string):
filename = self.file(filename)
try:
f = file(filename, 'w')
print >>f, string
f.flush()
f.close()
except IOError,e:
sys.exit(e)
def rmfile(self, filename):
filename = self.file(filename)
if os.path.isfile(filename):
os.unlink(filename)
def readval(self, filename):
filename = self.file(filename)
f = file(filename, 'r')
value = f.readline().strip()
f.close()
return value
def setstatus(self, string):
filename = self.file('.status')
try:
f = file(filename, 'a')
print >>f, string
f.flush()
f.close()
except IOError,e:
sys.exit(e)
def getstatus(self):
filename = self.file('.status')
try:
f = file(filename, 'r')
except IOError, e:
return 'none'
# fast forward to the end
for line in f: pass
# the first word on the last line is the status
return line.split(' ')[0]
def __str__(self):
return self.dir
if __name__ == '__main__':
rootdir = env.setdefault('ROOTDIR', os.getcwd())
@ -97,29 +159,27 @@ if __name__ == '__main__':
workbase = "/tmp/"
workdir = joinpath(workbase, '%s.%s' % (env['USER'], pbs_jobid))
def echofile(filename, string):
try:
f = file(joinpath(outdir, filename), 'w')
print >>f, string
f.flush()
f.close()
except IOError,e:
sys.exit(e)
host = socket.gethostname()
os.umask(0022)
echofile('.start', date())
echofile('.pbs_jobid', pbs_jobid)
echofile('.pbs_jobname', pbs_jobid)
echofile('.host', socket.gethostname())
jobdir = JobDir(outdir)
started = date()
jobdir.echofile('.running', started)
jobdir.rmfile('.queued')
jobdir.echofile('.pbs_jobid', pbs_jobid)
jobdir.echofile('.pbs_jobname', pbs_jobid)
jobdir.echofile('.host', host)
jobdir.setstatus('running on %s on %s' % (host, started))
if os.path.isdir(workdir):
cleandir(workdir)
else:
os.mkdir(workdir)
if os.path.isdir('/z/dist'):
if False and os.path.isdir('/z/dist'):
sync = rsync()
sync.delete = True
sync.sudo = True
@ -130,13 +190,13 @@ if __name__ == '__main__':
except OSError,e:
sys.exit(e)
os.symlink(joinpath(outdir, 'output'), 'status.out')
os.symlink(jobdir.file('output'), 'status.out')
args = [ joinpath(basedir, 'm5'), joinpath(basedir, 'run.py') ]
if not len(args):
sys.exit("no arguments")
print 'starting job... %s' % date()
print 'starting job... %s' % started
print ' '.join(args)
print
sys.stdout.flush()
@ -145,7 +205,7 @@ if __name__ == '__main__':
if not childpid:
# Execute command
sys.stdin.close()
fd = os.open(joinpath(outdir, "output"),
fd = os.open(jobdir.file("output"),
os.O_WRONLY | os.O_CREAT | os.O_TRUNC)
os.dup2(fd, sys.stdout.fileno())
os.dup2(fd, sys.stderr.fileno())
@ -170,12 +230,15 @@ if __name__ == '__main__':
thepid,ec = os.waitpid(childpid, 0)
if ec:
print 'Exit code ', ec
echofile('.failure', date())
status = 'failure'
else:
echofile('.success', date())
status = 'success'
done = 1
except OSError:
pass
print '\njob complete... %s' % date()
echofile('.stop', date())
complete = date()
print '\njob complete... %s' % complete
jobdir.echofile('.%s' % status, complete)
jobdir.rmfile('.running')
jobdir.setstatus('%s on %s' % (status, complete))

524
util/pbs/jobfile.py
View file

@ -26,67 +26,485 @@
#
# Authors: Nathan Binkert
from os.path import expanduser, isfile, join as joinpath
import sys
def crossproduct(options):
number = len(options)
indexes = [ 0 ] * number
maxes = [ len(opt) for opt in options ]
def next():
for i in xrange(number - 1, -1, -1):
indexes[i] += 1
if indexes[i] < maxes[i]:
return False
class ternary(object):
def __new__(cls, *args):
if len(args) > 1:
raise TypeError, \
'%s() takes at most 1 argument (%d given)' % \
(cls.__name__, len(args))
indexes[i] = 0
if args:
if not isinstance(args[0], (bool, ternary)):
raise TypeError, \
'%s() argument must be True, False, or Any' % \
cls.__name__
return args[0]
return super(ternary, cls).__new__(cls)
def __bool__(self):
return True
done = False
while not done:
result = []
for i in xrange(number):
result.append(options[i][indexes[i]])
yield result
done = next()
def __neg__(self):
return self
class JobFile(object):
def __init__(self, jfile):
self.data = {}
jfile = expanduser(jfile)
if not isfile(jfile):
for p in sys.path:
if isfile(joinpath(p, jfile)):
jfile = joinpath(p, jfile)
break
def __eq__(self, other):
return True
execfile(jfile, self.data)
self.options = self.data['options']
self.environment = self.data['environment']
self.jobinfo = {}
self.jobs = []
for job in crossproduct(self.options):
jobname = '.'.join([ id[0] for id in job ])
self.jobs.append(jobname)
list = []
for info in job:
for item in info[1:]:
list.append(item)
self.jobinfo[jobname] = list
def __ne__(self, other):
return False
def env(self, jobname):
env = {}
for key,val in self.jobinfo[jobname]:
env[key] = val
def __str__(self):
return 'Any'
for key,val in self.environment:
env[key] = val
return env
def __repr__(self):
return 'Any'
def printinfo(self, jobname):
print '%s:' % jobname
for key,val in self.jobinfo[jobname]:
print ' %s = %s' % (key, val)
Any = ternary()
for key,val in self.environment:
print ' %s = %s' % (key, val)
class Flags(dict):
def __init__(self, *args, **kwargs):
super(Flags, self).__init__()
self.update(*args, **kwargs)
def __getattr__(self, attr):
return self[attr]
def __setattr__(self, attr, value):
self[attr] = value
def __setitem__(self, item, value):
return super(Flags, self).__setitem__(item, ternary(value))
def __getitem__(self, item):
if item not in self:
return False
return super(Flags, self).__getitem__(item)
def update(self, *args, **kwargs):
for arg in args:
if isinstance(arg, Flags):
super(Flags, self).update(arg)
elif isinstance(arg, dict):
for key,val in kwargs.iteritems():
self[key] = val
else:
raise AttributeError, \
'flags not of type %s or %s, but %s' % \
(Flags, dict, type(arg))
for key,val in kwargs.iteritems():
self[key] = val
def match(self, *args, **kwargs):
match = Flags(*args, **kwargs)
for key,value in match.iteritems():
if self[key] != value:
return False
return True
def crossproduct(items):
if not isinstance(items, (list, tuple)):
raise AttributeError, 'crossproduct works only on sequences'
if not items:
yield None
return
current = items[0]
remainder = items[1:]
if not hasattr(current, '__iter__'):
current = [ current ]
for item in current:
for rem in crossproduct(remainder):
data = [ item ]
if rem:
data += rem
yield data
def flatten(items):
if not isinstance(items, (list, tuple)):
yield items
return
for item in items:
for flat in flatten(item):
yield flat
class Data(object):
def __init__(self, name, desc, **kwargs):
self.name = name
self.desc = desc
self.system = None
self.flags = Flags()
self.env = {}
for k,v in kwargs.iteritems():
setattr(self, k, v)
def update(self, obj):
if not isinstance(obj, Data):
raise AttributeError, "can only update from Data object"
self.env.update(obj.env)
self.flags.update(obj.flags)
if obj.system:
if self.system and self.system != obj.system:
raise AttributeError, \
"conflicting values for system: '%s'/'%s'" % \
(self.system, obj.system)
self.system = obj.system
def printinfo(self):
if self.name:
print 'name: %s' % self.name
if self.desc:
print 'desc: %s' % self.desc
if self.system:
print 'system: %s' % self.system
def printverbose(self):
print 'flags:'
keys = self.flags.keys()
keys.sort()
for key in keys:
print ' %s = %s' % (key, self.flags[key])
print 'env:'
keys = self.env.keys()
keys.sort()
for key in keys:
print ' %s = %s' % (key, self.env[key])
print
def __str__(self):
return self.name
class Job(Data):
def __init__(self, options):
super(Job, self).__init__('', '')
self.setoptions(options)
self.checkpoint = False
opts = []
for opt in options:
cpt = opt.group.checkpoint
if not cpt:
self.checkpoint = True
continue
if isinstance(cpt, Option):
opt = cpt.clone(suboptions=False)
else:
opt = opt.clone(suboptions=False)
opts.append(opt)
if not opts:
self.checkpoint = False
if self.checkpoint:
self.checkpoint = Job(opts)
def clone(self):
return Job(self.options)
def __getattribute__(self, attr):
if attr == 'name':
names = [ ]
for opt in self.options:
if opt.name:
names.append(opt.name)
return ':'.join(names)
if attr == 'desc':
descs = [ ]
for opt in self.options:
if opt.desc:
descs.append(opt.desc)
return ', '.join(descs)
return super(Job, self).__getattribute__(attr)
def setoptions(self, options):
config = options[0].config
for opt in options:
if opt.config != config:
raise AttributeError, \
"All options are not from the same Configuration"
self.config = config
self.groups = [ opt.group for opt in options ]
self.options = options
self.update(self.config)
for group in self.groups:
self.update(group)
for option in self.options:
self.update(option)
if option._suboption:
self.update(option._suboption)
def printinfo(self):
super(Job, self).printinfo()
if self.checkpoint:
print 'checkpoint: %s' % self.checkpoint.name
print 'config: %s' % self.config.name
print 'groups: %s' % [ g.name for g in self.groups ]
print 'options: %s' % [ o.name for o in self.options ]
super(Job, self).printverbose()
class SubOption(Data):
def __init__(self, name, desc, **kwargs):
super(SubOption, self).__init__(name, desc, **kwargs)
self.number = None
class Option(Data):
def __init__(self, name, desc, **kwargs):
super(Option, self).__init__(name, desc, **kwargs)
self._suboptions = []
self._suboption = None
self.number = None
def __getattribute__(self, attr):
if attr == 'name':
name = self.__dict__[attr]
if self._suboption is not None:
name = '%s:%s' % (name, self._suboption.name)
return name
if attr == 'desc':
desc = self.__dict__[attr]
if self._suboption is not None:
desc = '%s, %s' % (desc, self._suboption.desc)
return desc
return super(Option, self).__getattribute__(attr)
def suboption(self, name, desc, **kwargs):
subo = SubOption(name, desc, **kwargs)
subo.config = self.config
subo.group = self.group
subo.option = self
subo.number = len(self._suboptions)
self._suboptions.append(subo)
return subo
def clone(self, suboptions=True):
option = Option(self.__dict__['name'], self.__dict__['desc'])
option.update(self)
option.group = self.group
option.config = self.config
option.number = self.number
if suboptions:
option._suboptions.extend(self._suboptions)
option._suboption = self._suboption
return option
def subopts(self):
if not self._suboptions:
return [ self ]
subopts = []
for subo in self._suboptions:
option = self.clone()
option._suboption = subo
subopts.append(option)
return subopts
def printinfo(self):
super(Option, self).printinfo()
print 'config: %s' % self.config.name
super(Option, self).printverbose()
class Group(Data):
def __init__(self, name, desc, **kwargs):
super(Group, self).__init__(name, desc, **kwargs)
self._options = []
self.checkpoint = False
self.number = None
def option(self, name, desc, **kwargs):
opt = Option(name, desc, **kwargs)
opt.config = self.config
opt.group = self
opt.number = len(self._options)
self._options.append(opt)
return opt
def options(self):
return self._options
def subopts(self):
subopts = []
for opt in self._options:
for subo in opt.subopts():
subopts.append(subo)
return subopts
def printinfo(self):
super(Group, self).printinfo()
print 'config: %s' % self.config.name
print 'options: %s' % [ o.name for o in self._options ]
super(Group, self).printverbose()
class Configuration(Data):
def __init__(self, name, desc, **kwargs):
super(Configuration, self).__init__(name, desc, **kwargs)
self._groups = []
def group(self, name, desc, **kwargs):
grp = Group(name, desc, **kwargs)
grp.config = self
grp.number = len(self._groups)
self._groups.append(grp)
return grp
def groups(self, flags=Flags(), sign=True):
if not flags:
return self._groups
return [ grp for grp in self._groups if sign ^ grp.flags.match(flags) ]
def checkchildren(self, kids):
for kid in kids:
if kid.config != self:
raise AttributeError, "child from the wrong configuration"
def sortgroups(self, groups):
groups = [ (grp.number, grp) for grp in groups ]
groups.sort()
return [ grp[1] for grp in groups ]
def options(self, groups = None, checkpoint = False):
if groups is None:
groups = self._groups
self.checkchildren(groups)
groups = self.sortgroups(groups)
if checkpoint:
groups = [ grp for grp in groups if grp.checkpoint ]
optgroups = [ g.options() for g in groups ]
else:
optgroups = [ g.subopts() for g in groups ]
for options in crossproduct(optgroups):
for opt in options:
cpt = opt.group.checkpoint
if not isinstance(cpt, bool) and cpt != opt:
if checkpoint:
break
else:
yield options
else:
if checkpoint:
yield options
def checkpoints(self, groups = None):
for options in self.options(groups, True):
yield Job(options)
def jobs(self, groups = None):
for options in self.options(groups, False):
yield Job(options)
def alljobs(self, groups = None):
for options in self.options(groups, True):
yield Job(options)
for options in self.options(groups, False):
yield Job(options)
def find(self, jobname):
for job in self.alljobs():
if job.name == jobname:
return job
else:
raise AttributeError, "job '%s' not found" % jobname
def job(self, options):
self.checkchildren(options)
options = [ (opt.group.number, opt) for opt in options ]
options.sort()
options = [ opt[1] for opt in options ]
job = Job(options)
return job
def printinfo(self):
super(Configuration, self).printinfo()
print 'groups: %s' % [ g.name for g in self._grouips ]
super(Configuration, self).printverbose()
def JobFile(jobfile):
from os.path import expanduser, isfile, join as joinpath
filename = expanduser(jobfile)
# Can't find filename in the current path, search sys.path
if not isfile(filename):
for path in sys.path:
testname = joinpath(path, filename)
if isfile(testname):
filename = testname
break
else:
raise AttributeError, \
"Could not find file '%s'" % jobfile
data = {}
execfile(filename, data)
if 'conf' not in data:
raise ImportError, 'cannot import name conf from %s' % jobfile
conf = data['conf']
import jobfile
if not isinstance(conf, Configuration):
raise AttributeError, \
'conf in jobfile: %s (%s) is not type %s' % \
(jobfile, type(conf), Configuration)
return conf
if __name__ == '__main__':
from jobfile import *
import sys
usage = 'Usage: %s [-b] [-c] [-v] <jobfile>' % sys.argv[0]
try:
import getopt
opts, args = getopt.getopt(sys.argv[1:], '-bcv')
except getopt.GetoptError:
sys.exit(usage)
if len(args) != 1:
raise AttributeError, usage
both = False
checkpoint = False
verbose = False
for opt,arg in opts:
if opt == '-b':
both = True
checkpoint = True
if opt == '-c':
checkpoint = True
if opt == '-v':
verbose = True
jobfile = args[0]
conf = JobFile(jobfile)
if both:
gen = conf.alljobs()
elif checkpoint:
gen = conf.checkpoints()
else:
gen = conf.jobs()
for job in gen:
if not verbose:
cpt = ''
if job.checkpoint:
cpt = job.checkpoint.name
print job.name, cpt
else:
job.printinfo()

156
util/pbs/send.py
View file

@ -96,7 +96,7 @@ Usage:
try:
import getopt
opts, args = getopt.getopt(sys.argv[1:], '-cd:efhj:lq:v')
opts, args = getopt.getopt(sys.argv[1:], '-CRcd:efhj:lq:v')
except getopt.GetoptError:
sys.exit(usage)
@ -107,13 +107,18 @@ force = False
listonly = False
queue = ''
verbose = False
rootdir = nfspath(os.getcwd())
jfile = 'test.py'
jfile = 'Base/test.py'
docpts = False
doruns = True
runflag = False
for opt,arg in opts:
if opt == '-C':
docpts = True
if opt == '-R':
runflag = True
if opt == '-c':
clean = True
if opt == '-d':
rootdir = arg
if opt == '-e':
onlyecho = True
if opt == '-f':
@ -130,95 +135,123 @@ for opt,arg in opts:
if opt == '-v':
verbose = True
basedir = joinpath(rootdir, 'Base')
linkdir = joinpath(rootdir, 'Link')
if docpts:
doruns = runflag
for arg in args:
exprs.append(re.compile(arg))
if not listonly and not onlyecho and isdir(linkdir):
import jobfile, pbs
from job import JobDir, date
conf = jobfile.JobFile(jfile)
if not listonly and not onlyecho and isdir(conf.linkdir):
if verbose:
print 'Checking for outdated files in Link directory'
syncdir(linkdir, basedir)
import job, jobfile, pbs
test = jobfile.JobFile(joinpath(basedir, jfile))
syncdir(conf.linkdir, conf.basedir)
jobnames = {}
joblist = []
for jobname in test.jobs:
if not exprs:
joblist.append(jobname)
if docpts and doruns:
gen = conf.alljobs()
elif docpts:
gen = conf.checkpoints()
elif doruns:
gen = conf.jobs()
for job in gen:
if job.name in jobnames:
continue
for expr in exprs:
if expr.match(jobname):
joblist.append(jobname)
break
if exprs:
for expr in exprs:
if expr.match(job.name):
joblist.append(job)
break
else:
joblist.append(job)
if listonly:
if verbose:
for jobname in joblist:
test.printinfo(jobname)
for job in joblist:
job.printinfo()
else:
for jobname in joblist:
print jobname
for job in joblist:
print job.name
sys.exit(0)
if not onlyecho:
jl = []
for jobname in joblist:
jobdir = joinpath(rootdir, jobname)
if os.path.exists(jobname):
newlist = []
for job in joblist:
jobdir = JobDir(joinpath(conf.rootdir, job.name))
if jobdir.exists():
if not force:
if os.path.isfile(joinpath(jobdir, '.success')):
status = jobdir.getstatus()
if status == 'queued':
continue
if os.path.isfile(joinpath(jobdir, '.start')) and \
not os.path.isfile(joinpath(jobdir, '.stop')):
if status == 'running':
continue
if status == 'success':
continue
if not clean:
sys.exit('job directory not clean!')
sys.exit('job directory %s not clean!' % jobdir)
job.cleandir(jobdir)
else:
os.mkdir(jobdir)
jl.append(jobname)
joblist = jl
jobdir.clean()
newlist.append(job)
joblist = newlist
def setname(jobid, jobname):
# since pbs can handle jobnames of 15 characters or less, don't
# use the raj hack.
if len(jobname) <= 15:
return
class NameHack(object):
def __init__(self, host='pbs.pool', port=24465):
self.host = host
self.port = port
self.socket = None
import socket
s = socket.socket()
# Connect to pbs.pool and send the jobid/jobname pair to port
# 24465 (Raj didn't realize that there are only 64k ports and
# setup inetd to point to port 90001)
s.connect(("pbs.pool", 24465))
s.send("%s %s\n" % (jobid, jobname))
s.close()
def setname(self, jobid, jobname):
try:
jobid = int(jobid)
except ValueError:
jobid = int(jobid.strip().split('.')[0])
for jobname in joblist:
jobdir = joinpath(rootdir, jobname)
jobname = jobname.strip()
# since pbs can handle jobnames of 15 characters or less,
# don't use the raj hack.
if len(jobname) <= 15:
return
if not onlyecho and not os.path.isdir(jobdir):
sys.exit('%s is not a directory. Cannot build job' % jobdir)
if self.socket is None:
import socket
self.socket = socket.socket()
# Connect to pbs.pool and send the jobid/jobname pair to port
# 24465 (Raj didn't realize that there are only 64k ports and
# setup inetd to point to port 90001)
self.socket.connect((self.host, self.port))
print 'Job name: %s' % jobname
self.socket.send("%s %s\n" % (jobid, jobname))
namehack = NameHack()
for job in joblist:
jobdir = JobDir(joinpath(conf.rootdir, job.name))
if not onlyecho:
jobdir.create()
print 'Job name: %s' % job.name
print 'Job directory: %s' % jobdir
qsub = pbs.qsub()
qsub.pbshost = 'simpool.eecs.umich.edu'
qsub.stdout = joinpath(jobdir, 'jobout')
qsub.name = jobname[:15]
qsub.stdout = jobdir.file('jobout')
qsub.name = job.name[:15]
qsub.join = True
qsub.node_type = 'FAST'
qsub.env['ROOTDIR'] = rootdir
qsub.env['JOBNAME'] = jobname
qsub.env['ROOTDIR'] = conf.rootdir
qsub.env['JOBNAME'] = job.name
if len(queue):
qsub.queue = queue
qsub.build(joinpath(progpath, 'job.py'))
@ -231,6 +264,9 @@ for jobname in joblist:
if ec == 0:
jobid = qsub.result
print 'PBS Jobid: %s' % jobid
setname(jobid, jobname)
namehack.setname(jobid, job.name)
queued = date()
jobdir.echofile('.queued', queued)
jobdir.setstatus('queued on %s' % queued)
else:
print 'PBS Failed'

28
util/stats/__init__.py Normal file
View file

@ -0,0 +1,28 @@
# Copyright (c) 2005 The Regents of The University of Michigan
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Nathan Binkert

246
util/stats/barchart.py Normal file
View file

@ -0,0 +1,246 @@
# Copyright (c) 2005 The Regents of The University of Michigan
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Nathan Binkert
# Lisa Hsu
import matplotlib, pylab
from matplotlib.numerix import array, arange, reshape, shape, transpose, zeros
from matplotlib.numerix import Float
matplotlib.interactive(False)
class BarChart(object):
def __init__(self, **kwargs):
self.init(**kwargs)
def init(self, **kwargs):
self.colormap = 'jet'
self.inputdata = None
self.chartdata = None
self.xlabel = None
self.ylabel = None
self.legend = None
self.xticks = None
self.yticks = None
self.title = None
for key,value in kwargs.iteritems():
self.__setattr__(key, value)
def gen_colors(self, count):
cmap = matplotlib.cm.get_cmap(self.colormap)
if count == 1:
return cmap([ 0.5 ])
else:
return cmap(arange(count) / float(count - 1))
# The input data format does not match the data format that the
# graph function takes because it is intuitive. The conversion
# from input data format to chart data format depends on the
# dimensionality of the input data. Check here for the
# dimensionality and correctness of the input data
def set_data(self, data):
if data is None:
self.inputdata = None
self.chartdata = None
return
data = array(data)
dim = len(shape(data))
if dim not in (1, 2, 3):
raise AttributeError, "Input data must be a 1, 2, or 3d matrix"
self.inputdata = data
# If the input data is a 1d matrix, then it describes a
# standard bar chart.
if dim == 1:
self.chartdata = array([[data]])
# If the input data is a 2d matrix, then it describes a bar
# chart with groups. The matrix being an array of groups of
# bars.
if dim == 2:
self.chartdata = transpose([data], axes=(2,0,1))
# If the input data is a 3d matrix, then it describes an array
# of groups of bars with each bar being an array of stacked
# values.
if dim == 3:
self.chartdata = transpose(data, axes=(1,2,0))
def get_data(self):
return self.inputdata
data = property(get_data, set_data)
# Graph the chart data.
# Input is a 3d matrix that describes a plot that has multiple
# groups, multiple bars in each group, and multiple values stacked
# in each bar. The underlying bar() function expects a sequence of
# bars in the same stack location and same group location, so the
# organization of the matrix is that the inner most sequence
# represents one of these bar groups, then those are grouped
# together to make one full stack of bars in each group, and then
# the outer most layer describes the groups. Here is an example
# data set and how it gets plotted as a result.
#
# e.g. data = [[[10,11,12], [13,14,15], [16,17,18], [19,20,21]],
# [[22,23,24], [25,26,27], [28,29,30], [31,32,33]]]
#
# will plot like this:
#
# 19 31 20 32 21 33
# 16 28 17 29 18 30
# 13 25 14 26 15 27
# 10 22 11 23 12 24
#
# Because this arrangement is rather conterintuitive, the rearrange
# function takes various matricies and arranges them to fit this
# profile.
#
# This code deals with one of the dimensions in the matrix being
# one wide.
#
def graph(self):
if self.chartdata is None:
raise AttributeError, "Data not set for bar chart!"
self.figure = pylab.figure()
self.axes = self.figure.add_subplot(111)
dim = len(shape(self.inputdata))
cshape = shape(self.chartdata)
if dim == 1:
colors = self.gen_colors(cshape[2])
colors = [ [ colors ] * cshape[1] ] * cshape[0]
if dim == 2:
colors = self.gen_colors(cshape[0])
colors = [ [ [ c ] * cshape[2] ] * cshape[1] for c in colors ]
if dim == 3:
colors = self.gen_colors(cshape[1])
colors = [ [ [ c ] * cshape[2] for c in colors ] ] * cshape[0]
colors = array(colors)
bars_in_group = len(self.chartdata)
if bars_in_group < 5:
width = 1.0 / ( bars_in_group + 1)
center = width / 2
else:
width = .8 / bars_in_group
center = .1
bars = []
for i,stackdata in enumerate(self.chartdata):
bottom = array([0] * len(stackdata[0]))
stack = []
for j,bardata in enumerate(stackdata):
bardata = array(bardata)
ind = arange(len(bardata)) + i * width + center
bar = self.axes.bar(ind, bardata, width, bottom=bottom,
color=colors[i][j])
stack.append(bar)
bottom += bardata
bars.append(stack)
if self.xlabel is not None:
self.axes.set_xlabel(self.xlabel)
if self.ylabel is not None:
self.axes.set_ylabel(self.ylabel)
if self.yticks is not None:
ymin, ymax = self.axes.get_ylim()
nticks = float(len(self.yticks))
ticks = arange(nticks) / (nticks - 1) * (ymax - ymin) + ymin
self.axes.set_yticks(ticks)
self.axes.set_yticklabels(self.yticks)
if self.xticks is not None:
self.axes.set_xticks(arange(cshape[2]) + .5)
self.axes.set_xticklabels(self.xticks)
if self.legend is not None:
if dim == 1:
lbars = bars[0][0]
if dim == 2:
lbars = [ bars[i][0][0] for i in xrange(len(bars))]
if dim == 3:
number = len(bars[0])
lbars = [ bars[0][number - j - 1][0] for j in xrange(number)]
self.axes.legend(lbars, self.legend, loc='best')
if self.title is not None:
self.axes.set_title(self.title)
def savefig(self, name):
self.figure.savefig(name)
if __name__ == '__main__':
import random, sys
dim = 3
number = 5
args = sys.argv[1:]
if len(args) > 3:
sys.exit("invalid number of arguments")
elif len(args) > 0:
myshape = [ int(x) for x in args ]
else:
myshape = [ 3, 4, 8 ]
# generate a data matrix of the given shape
size = reduce(lambda x,y: x*y, myshape)
#data = [ random.randrange(size - i) + 10 for i in xrange(size) ]
data = [ float(i)/100.0 for i in xrange(size) ]
data = reshape(data, myshape)
# setup some test bar charts
if True:
chart1 = BarChart()
chart1.data = data
chart1.xlabel = 'Benchmark'
chart1.ylabel = 'Bandwidth (GBps)'
chart1.legend = [ 'x%d' % x for x in xrange(myshape[-1]) ]
chart1.xticks = [ 'xtick%d' % x for x in xrange(myshape[0]) ]
chart1.title = 'this is the title'
chart1.graph()
#chart1.savefig('/tmp/test1.png')
if False:
chart2 = BarChart()
chart2.data = data
chart2.colormap = 'gray'
chart2.graph()
#chart2.savefig('/tmp/test2.png')
pylab.show()

28
util/categories.py → util/stats/categories.py
View file

@ -1,3 +1,31 @@
# Copyright (c) 2005 The Regents of The University of Michigan
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Nathan Binkert
categories = {
'CALL_PALrdunique_' : 'interrupt', #
'Call_Pal_Callsys' : 'interrupt', #

5
util/stats/db.py
View file

@ -101,7 +101,7 @@ class Node(object):
def __init__(self, name):
self.name = name
def __str__(self):
return name
return self.name
class Database(object):
def __init__(self):
@ -466,3 +466,6 @@ class Database(object):
runs[data.run][data.x][data.y] = data.data
return runs
def __getitem__(self, key):
return self.stattop[key]

78
util/stats/orderdict.py Normal file
View file

@ -0,0 +1,78 @@
# Copyright (c) 2005 The Regents of The University of Michigan
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
__all__ = [ 'orderdict' ]
class orderdict(dict):
def __init__(self, d = {}):
self._keys = d.keys()
super(orderdict, self).__init__(d)
def __setitem__(self, key, item):
super(orderdict, self).__setitem__(key, item)
if not hasattr(self, '_keys'):
self._keys = [key,]
if key not in self._keys:
self._keys.append(key)
def __delitem__(self, key):
super(orderdict, self).__delitem__(key)
self._keys.remove(key)
def clear(self):
super(orderdict, self).clear()
self._keys = []
def items(self):
for i in self._keys:
yield i, self[i]
def keys(self):
return self._keys
def popitem(self):
if len(self._keys) == 0:
raise KeyError('dictionary is empty')
else:
key = self._keys[-1]
val = self[key]
del self[key]
return key, val
def setdefault(self, key, failobj = None):
super(orderdict, self).setdefault(key, failobj)
if key not in self._keys:
self._keys.append(key)
def update(self, d):
for key in d.keys():
if not self.has_key(key):
self._keys.append(key)
super(orderdict, self).update(d)
def values(self):
for i in self._keys:
yield self[i]

182
util/stats/output.py Normal file
View file

@ -0,0 +1,182 @@
# Copyright (c) 2005 The Regents of The University of Michigan
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Nathan Binkert
class dbinfo(object):
def get(self, job, stat):
import info
run = info.source.allRunNames.get(job.name, None)
if run is None:
print 'run "%s" not found' % job
return None
stat.system = info.source[job.system]
info.display_run = run.run;
val = float(stat)
if val == 1e300*1e300:
return None
return val
class StatOutput(object):
def __init__(self, name, jobfile, stat=None, info=dbinfo(), binstats=None):
self.name = name
self.jobfile = jobfile
self.stat = stat
self.binstats = None
self.label = self.name
self.invert = False
self.info = info
def printdata(self, bin = None, printmode = 'G'):
import info
if bin:
print '%s %s stats' % (self.name, bin)
if self.binstats:
for stat in self.binstats:
stat.bins = bin
if printmode == 'G':
valformat = '%g'
elif printmode != 'F' and value > 1e6:
valformat = '%0.5e'
else:
valformat = '%f'
for job in self.jobfile.jobs():
value = self.info.get(job, self.stat)
if value is None:
return
if not isinstance(value, list):
value = [ value ]
if self.invert:
for i,val in enumerate(value):
if val != 0.0:
value[i] = 1 / val
valstring = ', '.join([ valformat % val for val in value ])
print '%-50s %s' % (job.name + ':', valstring)
def display(self, binned = False, printmode = 'G'):
if binned and self.binstats:
self.printdata('kernel', printmode)
self.printdata('idle', printmode)
self.printdata('user', printmode)
self.printdata('interrupt', printmode)
print '%s total stats' % self.name
self.printdata(printmode=printmode)
def graph(self, graphdir):
from os.path import expanduser, join as joinpath
from barchart import BarChart
from matplotlib.numerix import Float, zeros
import re
confgroups = self.jobfile.groups()
ngroups = len(confgroups)
skiplist = [ False ] * ngroups
groupopts = None
baropts = None
groups = []
for i,group in enumerate(confgroups):
if group.flags.graph_group:
if groupopts is not None:
raise AttributeError, \
'Two groups selected for graph group'
groupopts = group.subopts()
skiplist[i] = True
elif group.flags.graph_bars:
if baropts is not None:
raise AttributeError, \
'Two groups selected for graph bars'
baropts = group.subopts()
skiplist[i] = True
else:
groups.append(group)
if groupopts is None:
raise AttributeError, 'No group selected for graph group'
if baropts is None:
raise AttributeError, 'No group selected for graph bars'
directory = expanduser(graphdir)
html = file(joinpath(directory, '%s.html' % self.name), 'w')
print >>html, '<html>'
print >>html, '<title>Graphs for %s</title>' % self.name
print >>html, '<body>'
for options in self.jobfile.options(groups):
data = zeros((len(groupopts), len(baropts)), Float)
data = [ [ None ] * len(baropts) for i in xrange(len(groupopts)) ]
enabled = False
stacked = None
for g,gopt in enumerate(groupopts):
for b,bopt in enumerate(baropts):
job = self.jobfile.job(options + [ gopt, bopt ])
val = self.info.get(job, self.stat)
if val is None:
val = 0.0
curstacked = isinstance(val, (list, tuple))
if stacked is None:
stacked = curstacked
else:
if stacked != curstacked:
raise ValueError, "some stats stacked, some not"
data[g][b] = val
bar_descs = [ opt.desc for opt in baropts ]
group_descs = [ opt.desc for opt in groupopts ]
if stacked:
legend = self.info.rcategories
else:
legend = bar_descs
chart = BarChart(data=data, xlabel='Benchmark', ylabel=self.label,
legend=legend, xticks=group_descs)
chart.graph()
names = [ opt.name for opt in options ]
descs = [ opt.desc for opt in options ]
filename = '%s-%s.png' % (self.name, ':'.join(names))
desc = ' '.join(descs)
filepath = joinpath(directory, filename)
chart.savefig(filepath)
filename = re.sub(':', '%3A', filename)
print >>html, '''%s<br><img src="%s"><br>''' % (desc, filename)
print >>html, '</body>'
print >>html, '</html>'
html.close()

146
util/stats/profile.py Normal file
View file

@ -0,0 +1,146 @@
# Copyright (c) 2005 The Regents of The University of Michigan
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
from orderdict import orderdict
import output
class ProfileData(object):
def __init__(self):
self.data = {}
self.total = {}
self.runs = orderdict()
self.runlist = []
def addvalue(self, run, cpu, symbol, value):
value = float(value)
self.data[run, cpu, symbol] = self.getvalue(run, cpu, symbol) + value
self.total[run, cpu] = self.gettotal(run, cpu) + value
if run not in self.runs:
self.runs[run] = orderdict()
if cpu not in self.runs[run]:
self.runs[run][cpu] = {}
if symbol not in self.runs[run][cpu]:
self.runs[run][cpu][symbol] = 0
self.runs[run][cpu][symbol] += value
def getvalue(self, run, cpu, symbol):
return self.data.get((run, cpu, symbol), 0)
def gettotal(self, run, cpu):
return self.total.get((run, cpu), 0)
class Profile(object):
default_order = ['ste', 'hte', 'htd', 'ocm', 'occ', 'ocp']
# This list controls the order of values in stacked bar data output
default_categories = [ 'interrupt',
'driver',
'stack',
'bufmgt',
'copy',
'user',
'other',
'idle']
def __init__(self, run_order=[], categories=[], stacknames=[]):
if not run_order:
run_order = Profile.default_order
if not categories:
categories = Profile.default_categories
self.run_order = run_order
self.categories = categories
self.rcategories = []
self.rcategories.extend(categories)
self.rcategories.reverse()
self.stacknames = stacknames
self.prof = ProfileData()
self.categorize = True
self.showidle = True
self.maxsymlen = 0
def category(self, symbol):
from categories import categories, categories_re
if categories.has_key(symbol):
return categories[symbol]
for regexp, cat in categories_re:
if regexp.match(symbol):
return cat
return 'other'
# Parse input file and put the results in the given run and cpu
def parsefile(self, run, cpu, filename):
fd = file(filename)
for line in fd:
(symbol, count) = line.split()
if symbol == "0x0":
continue
count = int(count)
if self.categorize:
symbol = self.category(symbol)
if symbol == 'idle' and not self.showidle:
continue
if symbol not in self.categories:
symbol = 'other'
self.maxsymlen = max(self.maxsymlen, len(symbol))
self.prof.addvalue(run, cpu, symbol, count)
fd.close()
# Read in files
def inputdir(self, directory):
import os, os.path, re
from os.path import expanduser, join as joinpath
directory = expanduser(directory)
label_ex = re.compile(r'm5prof\.(.*)')
for root,dirs,files in os.walk(directory):
for name in files:
match = label_ex.match(name)
if not match:
continue
filename = joinpath(root, name)
prefix = os.path.commonprefix([root, directory])
dirname = root[len(prefix)+1:]
self.parsefile(dirname, match.group(1), filename)
def get(self, job, stat):
if job.system is None:
raise AttributeError, 'The job must have a system set'
cpu = '%s.full0' % job.system
values = []
for cat in self.categories:
values.append(self.prof.getvalue(job.name, cpu, cat))
return values

718
util/stats/stats.py
View file

@ -29,10 +29,9 @@
from __future__ import division
import re, sys, math
def usage():
print '''\
Usage: %s [-E] [-F] [-d <db> ] [-g <get> ] [-h <host>] [-p]
Usage: %s [-E] [-F] [ -G <get> ] [-d <db> ] [-g <graphdir> ] [-h <host>] [-p]
[-s <system>] [-r <runs> ] [-T <samples>] [-u <username>]
<command> [command args]
@ -61,143 +60,6 @@ def getopts(list, flags):
return opts, args
def printval(name, value, invert = False):
if invert and value != 0.0:
value = 1 / value
if value == (1e300*1e300):
return
if printval.mode == 'G':
print '%s: %g' % (name, value)
elif printval.mode != 'F' and value > 1e6:
print '%s: %0.5e' % (name, value)
else:
print '%s: %f' % (name, value)
printval.mode = 'G'
def unique(list):
set = {}
map(set.__setitem__, list, [])
return set.keys()
#benchmarks = [ 'm', 's', 'snt', 'nb1', 'w1', 'w2', 'w3', 'w4', 'nm', 'ns', 'nw1', 'nw2', 'nw3' ]
def graphdata(runs, options, tag, label, value):
import info
bench_system = {
'm' : 'client',
's' : 'client',
'snt' : 'client',
'nb1' : 'server',
'nb2' : 'server',
'nt1' : 'server',
'nt2' : 'server',
'w1' : 'server',
'w2' : 'server',
'w3' : 'server',
'w4' : 'server',
'w1s' : 'server',
'w2s' : 'server',
'w3s' : 'server',
'ns' : 'natbox',
'nm' : 'natbox',
'nw1' : 'natbox',
'nw2' : 'natbox',
'nw3' : 'natbox'
}
system_configs = {
's1' : 'Uni 4GHz',
'm1' : 'Uni 6GHz',
'f1' : 'Uni 8GHz',
'q1' : 'Uni 10GHz',
's2' : 'Dual 4GHz',
'm2' : 'Dual 6GHz',
's4' : 'Quad 4GHz',
'm4' : 'Quad 6GHz' }
configs = ['ste', 'hte', 'htd', 'ocm', 'occ', 'ocp' ]
benchmarks = [ 'm', 'snt', 'w2', 'nm', 'nw2' ]
caches = [ '0', '2', '4' ]
names = []
for bench in benchmarks:
if bench_system[bench] != options.system:
continue
for cache in caches:
names.append([bench, cache])
for bench,cache in names:
base = '%s.%s' % (bench, cache)
fname = 'data/uni.%s.%s.dat' % (tag, base)
f = open(fname, 'w')
print >>f, '#set TITLE = '
print >>f, '#set ylbl = %s' % label
#print >>f, '#set sublabels = %s' % ' '.join(configs)
print >>f, '#set sublabels = ste hte htd ocm occ ocs'
for speed in ('s1', 'm1', 'f1', 'q1'):
label = system_configs[speed]
print >>f, '"%s"' % label,
for conf in configs:
name = '%s.%s.%s.%s' % (conf, bench, cache, speed)
run = info.source.allRunNames[name]
info.display_run = run.run;
val = float(value)
if val == 1e300*1e300:
print >>f, 0.0,
else:
print >>f, "%f" % val,
print >>f
f.close()
configs = ['ste', 'hte', 'htd', 'ocm', 'occ', 'ocp' ]
benchmarks = [ 'w2']
caches = [ '0', '2', '4' ]
names = []
for bench in benchmarks:
if bench_system[bench] != options.system:
continue
for cache in caches:
names.append([bench, cache])
for bench,cache in names:
base = '%s.%s' % (bench, cache)
fname = 'data/mp.%s.%s.dat' % (tag, base)
f = open(fname, 'w')
print >>f, '#set TITLE = '
print >>f, '#set ylbl = %s' % label
#print >>f, '#set sublabels = %s' % ' '.join(configs)
print >>f, '#set sublabels = ste hte htd ocm occ ocs'
for speed in ('s2', 'm2', 's4', 'm4'):
label = system_configs[speed]
print >>f, '"%s"' % label,
for conf in configs:
name = '%s.%s.%s.%s' % (conf, bench, cache, speed)
run = info.source.allRunNames[name]
info.display_run = run.run;
val = float(value)
if val == 1e300*1e300:
print >>f, 0.0,
else:
print >>f, "%f" % val,
print >>f
f.close()
def printdata(runs, value, invert = False):
import info
for run in runs:
info.display_run = run.run;
val = float(value)
printval(run.name, val)
class CommandException(Exception):
pass
@ -243,7 +105,7 @@ def commands(options, command, args):
info.source.passwd = options.passwd
info.source.user = options.user
info.source.connect()
info.source.update_dict(globals())
#info.source.update_dict(globals())
if type(options.get) is str:
info.source.get = options.get
@ -270,6 +132,43 @@ def commands(options, command, args):
info.source.listRuns(user)
return
if command == 'stats':
if len(args) == 0:
info.source.listStats()
elif len(args) == 1:
info.source.listStats(args[0])
else:
raise CommandException
return
if command == 'bins':
if len(args) == 0:
info.source.listBins()
elif len(args) == 1:
info.source.listBins(args[0])
else:
raise CommandException
return
if command == 'formulas':
if len(args) == 0:
info.source.listFormulas()
elif len(args) == 1:
info.source.listFormulas(args[0])
else:
raise CommandException
return
if command == 'samples':
if len(args):
raise CommandException
info.source.listTicks(runs)
return
if command == 'stability':
if len(args) < 2:
raise CommandException
@ -281,16 +180,19 @@ def commands(options, command, args):
stats = info.source.getStat(args[1])
info.source.get = "sum"
def disp(*args):
print "%-20s %12s %12s %4s %5s %5s %5s %10s" % args
# temporary variable containing a bunch of dashes
d = '-' * 100
#loop through all the stats selected
for stat in stats:
print "%s:" % stat.name
print "%-20s %12s %12s %4s %5s %5s %5s %10s" % \
("run name", "average", "stdev", ">10%", ">1SDV", ">2SDV", "SAMP", "CV")
print "%-20s %12s %12s %4s %5s %5s %5s %10s" % \
("--------------------", "------------",
"------------", "----", "-----", "-----", "-----", "----------")
disp("run name", "average", "stdev", ">10%", ">1SDV", ">2SDV",
"SAMP", "CV")
disp(d[:20], d[:12], d[:12], d[:4], d[:5], d[:5], d[:5], d[:10])
#loop through all the selected runs
for run in runs:
info.display_run = run.run;
@ -331,130 +233,63 @@ def commands(options, command, args):
if (val < (avg - (2*stdev))) or (val > (avg + (2*stdev))):
numoutside2std += 1
if avg > 1000:
print "%-20s %12s %12s %4s %5s %5s %5s %10s" % \
(run.name, "%.1f" % avg, "%.1f" % stdev,
"%d" % numoutsideavg, "%d" % numoutside1std,
"%d" % numoutside2std, "%d" % len(pairRunTicks),
"%.3f" % (stdev/avg*100))
disp(run.name, "%.1f" % avg, "%.1f" % stdev,
"%d" % numoutsideavg, "%d" % numoutside1std,
"%d" % numoutside2std, "%d" % len(pairRunTicks),
"%.3f" % (stdev/avg*100))
elif avg > 100:
print "%-20s %12s %12s %4s %5s %5s %5s %10s" % \
(run.name, "%.1f" % avg, "%.1f" % stdev,
"%d" % numoutsideavg, "%d" % numoutside1std,
"%d" % numoutside2std, "%d" % len(pairRunTicks),
"%.5f" % (stdev/avg*100))
disp(run.name, "%.1f" % avg, "%.1f" % stdev,
"%d" % numoutsideavg, "%d" % numoutside1std,
"%d" % numoutside2std, "%d" % len(pairRunTicks),
"%.5f" % (stdev/avg*100))
else:
print "%-20s %12s %12s %4s %5s %5s %5s %10s" % \
(run.name, "%.5f" % avg, "%.5f" % stdev,
"%d" % numoutsideavg, "%d" % numoutside1std,
"%d" % numoutside2std, "%d" % len(pairRunTicks),
"%.7f" % (stdev/avg*100))
disp(run.name, "%.5f" % avg, "%.5f" % stdev,
"%d" % numoutsideavg, "%d" % numoutside1std,
"%d" % numoutside2std, "%d" % len(pairRunTicks),
"%.7f" % (stdev/avg*100))
return
if command == 'stats':
if len(args) == 0:
info.source.listStats()
elif len(args) == 1:
info.source.listStats(args[0])
else:
raise CommandException
return
if command == 'stat':
if len(args) != 1:
raise CommandException
stats = info.source.getStat(args[0])
for stat in stats:
if options.graph:
graphdata(runs, options, stat.name, stat.name, stat)
else:
if options.ticks:
print 'only displaying sample %s' % options.ticks
info.globalTicks = [ int(x) for x in options.ticks.split() ]
if options.binned:
print 'kernel ticks'
stat.bins = 'kernel'
printdata(runs, stat)
print 'idle ticks'
stat.bins = 'idle'
printdata(runs, stat)
print 'user ticks'
stat.bins = 'user'
printdata(runs, stat)
print 'interrupt ticks'
stat.bins = 'interrupt'
printdata(runs, stat)
print 'total ticks'
stat.bins = None
print stat.name
printdata(runs, stat)
return
if command == 'formula':
if len(args) != 1:
raise CommandException
stats = eval(args[0])
for stat in stats:
if options.graph:
graphdata(runs, options, stat.name, stat.name, stat)
else:
if options.binned:
print 'kernel ticks'
stat.bins = 'kernel'
printdata(runs, stat)
print 'idle ticks'
stat.bins = 'idle'
printdata(runs, stat)
print 'user ticks'
stat.bins = 'user'
printdata(runs, stat)
print 'interrupt ticks'
stat.bins = 'interrupt'
printdata(runs, stat)
print 'total ticks'
stat.bins = None
print args[0]
printdata(runs, stat)
return
if command == 'bins':
if len(args) == 0:
info.source.listBins()
elif len(args) == 1:
info.source.listBins(args[0])
else:
raise CommandException
return
if command == 'formulas':
if len(args) == 0:
info.source.listFormulas()
elif len(args) == 1:
info.source.listFormulas(args[0])
else:
raise CommandException
return
if command == 'samples':
if command == 'all':
if len(args):
raise CommandException
info.source.listTicks(runs)
all = [ 'bps', 'rxbps', 'txbps', 'bpt',
'misses', 'mpkb',
'ipkb',
'pps', 'bpp', 'txbpp', 'rxbpp',
'rtp', 'rtb' ]
for command in all:
commands(options, command, args)
if options.ticks:
if not options.graph:
print 'only displaying sample %s' % options.ticks
info.globalTicks = [ int(x) for x in options.ticks.split() ]
from output import StatOutput
def display():
if options.graph:
output.graph(options.graphdir)
else:
output.display(options.binned, options.printmode)
if command == 'stat' or command == 'formula':
if len(args) != 1:
raise CommandException
if command == 'stat':
stats = info.source.getStat(args[0])
if command == 'formula':
stats = eval(args[0])
for stat in stats:
output = StatOutput(stat.name, options.jobfile)
output.stat = stat
output.label = stat.name
display()
return
if len(args):
@ -462,273 +297,151 @@ def commands(options, command, args):
system = info.source.__dict__[options.system]
from proxy import ProxyGroup
sim_ticks = info.source['sim_ticks']
sim_seconds = info.source['sim_seconds']
proxy = ProxyGroup(system = info.source[options.system])
system = proxy.system
etherdev = system.tsunami.etherdev0
bytes = etherdev.rxBytes + etherdev.txBytes
kbytes = bytes / 1024
packets = etherdev.rxPackets + etherdev.txPackets
bps = etherdev.rxBandwidth + etherdev.txBandwidth
output = StatOutput(command, options.jobfile)
if command == 'usertime':
import copy
kernel = copy.copy(system.full0.numCycles)
kernel.bins = 'kernel'
user = copy.copy(system.full0.numCycles)
user.bins = 'user'
if options.graph:
graphdata(runs, options, 'usertime', 'User Fraction',
user / system.full0.numCycles)
else:
printdata(runs, user / system.full0.numCycles)
output.stat = user / system.full0.numCycles
output.label = 'User Fraction'
display()
return
if command == 'ticks':
if options.binned:
print 'kernel ticks'
system.full0.numCycles.bins = 'kernel'
printdata(runs, system.full0.numCycles)
print 'idle ticks'
system.full0.numCycles.bins = 'idle'
printdata(runs, system.full0.numCycles)
print 'user ticks'
system.full0.numCycles.bins = 'user'
printdata(runs, system.full0.numCycles)
print 'total ticks'
system.full0.numCycles.bins = None
printdata(runs, system.full0.numCycles)
return
if command == 'packets':
packets = system.tsunami.etherdev0.rxPackets
if options.graph:
graphdata(runs, options, 'packets', 'Packets', packets)
else:
printdata(runs, packets)
return
if command == 'ppt' or command == 'tpp':
ppt = system.tsunami.etherdev0.rxPackets / sim_ticks
printdata(runs, ppt, command == 'tpp')
return
if command == 'pps':
pps = system.tsunami.etherdev0.rxPackets / sim_seconds
if options.graph:
graphdata(runs, options, 'pps', 'Packets/s', pps)
else:
printdata(runs, pps)
return
if command == 'bpt' or command == 'tpb':
bytes = system.tsunami.etherdev0.rxBytes + system.tsunami.etherdev0.txBytes
bpt = bytes / sim_ticks * 8
if options.graph:
graphdata(runs, options, 'bpt', 'bps / Hz', bpt)
else:
printdata(runs, bpt, command == 'tpb')
return
if command == 'bptb' or command == 'tpbb':
bytes = system.tsunami.etherdev0.rxBytes + system.tsunami.etherdev0.txBytes
print 'kernel stats'
bytes.bins = 'kernel'
printdata(runs, bytes / ticks)
print 'idle stats'
bytes.bins = 'idle'
printdata(runs, bytes / ticks)
print 'user stats'
bytes.bins = 'user'
printdata(runs, bytes / ticks)
output.stat = system.full0.numCycles
output.binstats = [ system.full0.numCycles ]
display()
return
if command == 'bytes':
stat = system.tsunami.etherdev0.rxBytes + system.tsunami.etherdev0.txBytes
output.stat = bytes
display()
return
if options.binned:
print '%s kernel stats' % stat.name
stat.bins = 'kernel'
printdata(runs, stat)
if command == 'packets':
output.stat = packets
display()
return
print '%s idle stats' % stat.name
stat.bins = 'idle'
printdata(runs, stat)
if command == 'ppt' or command == 'tpp':
output.stat = packets / sim_ticks
output.invert = command == 'tpp'
display()
return
print '%s user stats' % stat.name
stat.bins = 'user'
printdata(runs, stat)
if command == 'pps':
output.stat = packets / sim_seconds
output.label = 'Packets/s'
display()
return
print '%s total stats' % stat.name
stat.bins = None
printdata(runs, stat)
if command == 'bpt' or command == 'tpb':
output.stat = bytes / sim_ticks * 8
output.label = 'bps / Hz'
output.invert = command == 'tpb'
display()
return
if command == 'rxbps':
gbps = system.tsunami.etherdev0.rxBandwidth / 1e9
if options.graph:
graphdata(runs, options, 'rxbps', 'Bandwidth (Gbps)', gbps)
else:
printdata(runs, gbps)
output.stat = etherdev.rxBandwidth / 1e9
output.label = 'Bandwidth (Gbps)'
display()
return
if command == 'txbps':
gbps = system.tsunami.etherdev0.txBandwidth / 1e9
if options.graph:
graphdata(runs, options, 'txbps', 'Bandwidth (Gbps)', gbps)
else:
printdata(runs, gbps)
output.stat = etherdev.txBandwidth / 1e9
output.label = 'Bandwidth (Gbps)'
display()
return
if command == 'bps':
rxbps = system.tsunami.etherdev0.rxBandwidth
txbps = system.tsunami.etherdev0.txBandwidth
gbps = (rxbps + txbps) / 1e9
if options.graph:
graphdata(runs, options, 'bps', 'Bandwidth (Gbps)', gbps)
else:
printdata(runs, gbps)
output.stat = bps / 1e9
output.label = 'Bandwidth (Gbps)'
display()
return
if command == 'bpp':
output.stat = bytes / packets
output.label = 'Bytes / Packet'
display()
return
if command == 'rxbpp':
output.stat = etherdev.rxBytes / etherdev.rxPackets
output.label = 'Receive Bytes / Packet'
display()
return
if command == 'txbpp':
output.stat = etherdev.txBytes / etherdev.txPackets
output.label = 'Transmit Bytes / Packet'
display()
return
if command == 'rtp':
output.stat = etherdev.rxPackets / etherdev.txPackets
output.label = 'rxPackets / txPackets'
display()
return
if command == 'rtb':
output.stat = etherdev.rxBytes / etherdev.txBytes
output.label = 'rxBytes / txBytes'
display()
return
misses = system.l2.overall_mshr_misses
if command == 'misses':
stat = system.l2.overall_mshr_misses
if options.binned:
print '%s kernel stats' % stat.name
stat.bins = 'kernel'
printdata(runs, stat)
print '%s idle stats' % stat.name
stat.bins = 'idle'
printdata(runs, stat)
print '%s user stats' % stat.name
stat.bins = 'user'
printdata(runs, stat)
print '%s total stats' % stat.name
stat.bins = None
if options.graph:
graphdata(runs, options, 'misses', 'Overall MSHR Misses', stat)
else:
printdata(runs, stat)
output.stat = misses
output.label = 'Overall MSHR Misses'
display()
return
if command == 'mpkb':
misses = system.l2.overall_mshr_misses
rxbytes = system.tsunami.etherdev0.rxBytes
txbytes = system.tsunami.etherdev0.txBytes
if options.binned:
print 'mpkb kernel stats'
misses.bins = 'kernel'
mpkb = misses / ((rxbytes + txbytes) / 1024)
printdata(runs, mpkb)
print 'mpkb idle stats'
misses.bins = 'idle'
mpkb = misses / ((rxbytes + txbytes) / 1024)
printdata(runs, mpkb)
print 'mpkb user stats'
misses.bins = 'user'
mpkb = misses / ((rxbytes + txbytes) / 1024)
printdata(runs, mpkb)
print 'mpkb total stats'
mpkb = misses / ((rxbytes + txbytes) / 1024)
misses.bins = None
if options.graph:
graphdata(runs, options, 'mpkb', 'Misses / KB', mpkb)
else:
printdata(runs, mpkb)
output.stat = misses / (bytes / 1024)
output.binstats = [ misses ]
output.label = 'Misses / KB'
display()
return
if command == 'ipkb':
interrupts = system.full0.kern.faults[4]
rxbytes = system.tsunami.etherdev0.rxBytes
txbytes = system.tsunami.etherdev0.txBytes
if options.binned:
print 'ipkb kernel stats'
interrupts.bins = 'kernel'
ipkb = interrupts / ((rxbytes + txbytes) / 1024)
printdata(runs, ipkb)
print 'ipkb idle stats'
interrupts.bins = 'idle'
ipkb = interrupts / ((rxbytes + txbytes) / 1024)
printdata(runs, ipkb)
print 'ipkb user stats'
interrupts.bins = 'user'
ipkb = interrupts / ((rxbytes + txbytes) / 1024)
printdata(runs, ipkb)
print 'ipkb total stats'
ipkb = interrupts / ((rxbytes + txbytes) / 1024)
interrupts.bins = None
if options.graph:
graphdata(runs, options, 'ipkb', 'Interrupts / KB', ipkb)
else:
printdata(runs, ipkb)
output.stat = interrupts / kbytes
output.binstats = [ interrupts ]
output.label = 'Interrupts / KB'
display()
return
if command == 'execute':
printdata(runs, system.full0.ISSUE__count)
output.stat = system.full0.ISSUE__count
display()
return
if command == 'commit':
printdata(runs, system.full0.COM__count)
output.stat = system.full0.COM__count
display()
return
if command == 'fetch':
printdata(runs, system.full0.FETCH__count)
return
if command == 'bpp':
ed = system.tsunami.etherdev0
bpp = (ed.rxBytes + ed.txBytes) / (ed.rxPackets + ed.txPackets)
if options.graph:
graphdata(runs, options, 'bpp', 'Bytes / Packet', bpp)
else:
printdata(runs, bpp)
return
if command == 'rxbpp':
bpp = system.tsunami.etherdev0.rxBytes / system.tsunami.etherdev0.rxPackets
if options.graph:
graphdata(runs, options, 'rxbpp', 'Receive Bytes / Packet', bpp)
else:
printdata(runs, bpp)
return
if command == 'txbpp':
bpp = system.tsunami.etherdev0.txBytes / system.tsunami.etherdev0.txPackets
if options.graph:
graphdata(runs, options, 'txbpp', 'Transmit Bytes / Packet', bpp)
else:
printdata(runs, bpp)
return
if command == 'rtp':
rtp = system.tsunami.etherdev0.rxPackets / system.tsunami.etherdev0.txPackets
if options.graph:
graphdata(runs, options, 'rtp', 'rxPackets / txPackets', rtp)
else:
printdata(runs, rtp)
return
if command == 'rtb':
rtb = system.tsunami.etherdev0.rxBytes / system.tsunami.etherdev0.txBytes
if options.graph:
graphdata(runs, options, 'rtb', 'rxBytes / txBytes', rtb)
else:
printdata(runs, rtb)
output.stat = system.full0.FETCH__count
display()
return
raise CommandException
@ -738,9 +451,10 @@ class Options: pass
if __name__ == '__main__':
import getpass
from jobfile import JobFile
options = Options()
options.host = 'zizzer.pool'
options.host = None
options.db = None
options.passwd = ''
options.user = getpass.getuser()
@ -750,23 +464,31 @@ if __name__ == '__main__':
options.binned = False
options.graph = False
options.ticks = False
options.printmode = 'G'
options.jobfile = None
options.all = False
opts, args = getopts(sys.argv[1:], '-6BEFGd:g:h:pr:s:u:T:')
opts, args = getopts(sys.argv[1:], '-BEFG:ad:g:h:j:pr:s:u:T:')
for o,a in opts:
if o == '-B':
options.binned = True
if o == '-E':
printval.mode = 'E'
options.printmode = 'E'
if o == '-F':
printval.mode = 'F'
options.printmode = 'F'
if o == '-G':
options.graph = True;
options.get = a
if o == '-a':
options.all = True
if o == '-d':
options.db = a
if o == '-g':
options.get = a
options.graph = True;
options.graphdir = a
if o == '-h':
options.host = a
if o == '-j':
options.jobfile = JobFile(a)
if o == '-p':
options.passwd = getpass.getpass()
if o == '-r':
@ -778,6 +500,18 @@ if __name__ == '__main__':
if o == '-T':
options.ticks = a
if options.jobfile:
if not options.host:
options.host = options.jobfile.dbhost
if not options.db:
options.db = options.jobfile.statdb
if not options.host:
sys.exit('Database server must be provided from a jobfile or -h')
if not options.db:
sys.exit('Database name must be provided from a jobfile or -d')
if len(args) == 0:
usage()