gem5/util/stats/info.py
Nathan Binkert e00237e49e Major cleanup of the statistics handling code
util/stats/db.py:
    Build a result object as the result of a query operation so it is
    easier to populate and contains a bit more information than just
    a big dict.  Also change the next level data into a matrix instead
    of a dict of dicts.
    Move the "get" function into the Database object.  (The get function
    is used by the output parsing function as the interface for accessing
    backend storage, same interface for profile stuff.)
    Change the old get variable to the method variable, it describes how
    the get works, (whether using sum, stdev, etc.)
util/stats/display.py:
    Clean up the display functions, mostly formatting.
    Handle values the way they should be now.
util/stats/info.py:
    Totally re-work how values are accessed from their data store.
    Access individual values on demand instead of calculating everything
    and passing up a huge result from the bottom.
    This impacts the way that proxying works, and in general, everything
    is now esentially a proxy for the lower level database.  Provide new
    operators: unproxy, scalar, vector, value, values, total, and len which
    retrieve the proper result from the object they are called on.
    Move the ProxyGroup stuff (proxies of proxies!) here from the now gone
    proxy.py file and integrate the shared parts of the code.  The ProxyGroup
    stuff allows you to write formulas without specifying the statistics
    until evaluation time.

    Get rid of global variables!
util/stats/output.py:
    Move the dbinfo stuff into the Database itself.  Each source should
    have it's own get() function for accessing it's data.
    This get() function behaves a bit differently than before in that it
    can return vectors as well, deal with these vectors and with no result
    conditions better.
util/stats/stats.py:
    the info module no longer has the source global variable, just
    create the database source and pass it around as necessary

--HG--
extra : convert_revision : 8e5aa228e5d3ae8068ef9c40f65b3a2f9e7c0cff
2005-10-21 16:29:13 -04:00

716 lines
21 KiB
Python

# Copyright (c) 2003-2004 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 __future__ import division
import operator, re, types
def unproxy(proxy):
if hasattr(proxy, '__unproxy__'):
return proxy.__unproxy__()
return proxy
def scalar(stat):
stat = unproxy(stat)
assert(stat.__scalar__() != stat.__vector__())
return stat.__scalar__()
def vector(stat):
stat = unproxy(stat)
assert(stat.__scalar__() != stat.__vector__())
return stat.__vector__()
def value(stat, *args):
stat = unproxy(stat)
return stat.__value__(*args)
def values(stat, run):
stat = unproxy(stat)
result = []
for i in xrange(len(stat)):
val = value(stat, run.run, i)
if val is None:
return None
result.append(val)
return result
def total(stat, run):
return sum(values(stat, run))
def len(stat):
stat = unproxy(stat)
return stat.__len__()
class Value(object):
def __scalar__(self):
raise AttributeError, "must define __scalar__ for %s" % (type (self))
def __vector__(self):
raise AttributeError, "must define __vector__ for %s" % (type (self))
def __add__(self, other):
return BinaryProxy(operator.__add__, self, other)
def __sub__(self, other):
return BinaryProxy(operator.__sub__, self, other)
def __mul__(self, other):
return BinaryProxy(operator.__mul__, self, other)
def __div__(self, other):
return BinaryProxy(operator.__div__, self, other)
def __truediv__(self, other):
return BinaryProxy(operator.__truediv__, self, other)
def __floordiv__(self, other):
return BinaryProxy(operator.__floordiv__, self, other)
def __radd__(self, other):
return BinaryProxy(operator.__add__, other, self)
def __rsub__(self, other):
return BinaryProxy(operator.__sub__, other, self)
def __rmul__(self, other):
return BinaryProxy(operator.__mul__, other, self)
def __rdiv__(self, other):
return BinaryProxy(operator.__div__, other, self)
def __rtruediv__(self, other):
return BinaryProxy(operator.__truediv__, other, self)
def __rfloordiv__(self, other):
return BinaryProxy(operator.__floordiv__, other, self)
def __neg__(self):
return UnaryProxy(operator.__neg__, self)
def __pos__(self):
return UnaryProxy(operator.__pos__, self)
def __abs__(self):
return UnaryProxy(operator.__abs__, self)
class Scalar(Value):
def __scalar__(self):
return True
def __vector__(self):
return False
def __value__(self, run):
raise AttributeError, '__value__ must be defined'
class VectorItemProxy(Value):
def __init__(self, proxy, index):
self.proxy = proxy
self.index = index
def __scalar__(self):
return True
def __vector__(self):
return False
def __value__(self, run):
return value(self.proxy, run, self.index)
class Vector(Value):
def __scalar__(self):
return False
def __vector__(self):
return True
def __value__(self, run, index):
raise AttributeError, '__value__ must be defined'
def __getitem__(self, index):
return VectorItemProxy(self, index)
class ScalarConstant(Scalar):
def __init__(self, constant):
self.constant = constant
def __value__(self, run):
return self.constant
class VectorConstant(Vector):
def __init__(self, constant):
self.constant = constant
def __value__(self, run, index):
return self.constant[index]
def __len__(self):
return len(self.constant)
def WrapValue(value):
if isinstance(value, (int, long, float)):
return ScalarConstant(value)
if isinstance(value, (list, tuple)):
return VectorConstant(value)
if isinstance(value, Value):
return value
raise AttributeError, 'Only values can be wrapped'
class Statistic(object):
def __getattr__(self, attr):
if attr in ('data', 'x', 'y'):
result = self.source.data(self, self.bins, self.ticks)
self.data = result.data
self.x = result.x
self.y = result.y
return super(Statistic, self).__getattribute__(attr)
def __setattr__(self, attr, value):
if attr == 'stat':
raise AttributeError, '%s is read only' % stat
if attr in ('source', 'bins', 'ticks'):
if getattr(self, attr) != value:
if hasattr(self, 'data'):
delattr(self, 'data')
super(Statistic, self).__setattr__(attr, value)
class ValueProxy(Value):
def __getattr__(self, attr):
if attr == '__value__':
if scalar(self):
return self.__scalarvalue__
if vector(self):
return self.__vectorvalue__
if attr == '__len__':
if vector(self):
return self.__vectorlen__
return super(ValueProxy, self).__getattribute__(attr)
class UnaryProxy(ValueProxy):
def __init__(self, op, arg):
self.op = op
self.arg = WrapValue(arg)
def __scalar__(self):
return scalar(self.arg)
def __vector__(self):
return vector(self.arg)
def __scalarvalue__(self, run):
val = value(self.arg, run)
if val is None:
return None
return self.op(val)
def __vectorvalue__(self, run, index):
val = value(self.arg, run, index)
if val is None:
return None
return self.op(val)
def __vectorlen__(self):
return len(unproxy(self.arg))
class BinaryProxy(ValueProxy):
def __init__(self, op, arg0, arg1):
super(BinaryProxy, self).__init__()
self.op = op
self.arg0 = WrapValue(arg0)
self.arg1 = WrapValue(arg1)
def __scalar__(self):
return scalar(self.arg0) and scalar(self.arg1)
def __vector__(self):
return vector(self.arg0) or vector(self.arg1)
def __scalarvalue__(self, run):
val0 = value(self.arg0, run)
val1 = value(self.arg1, run)
if val0 is None or val1 is None:
return None
return self.op(val0, val1)
def __vectorvalue__(self, run, index):
if scalar(self.arg0):
val0 = value(self.arg0, run)
if vector(self.arg0):
val0 = value(self.arg0, run, index)
if scalar(self.arg1):
val1 = value(self.arg1, run)
if vector(self.arg1):
val1 = value(self.arg1, run, index)
if val0 is None or val1 is None:
return None
return self.op(val0, val1)
def __vectorlen__(self):
if vector(self.arg0) and scalar(self.arg1):
return len(self.arg0)
if scalar(self.arg0) and vector(self.arg1):
return len(self.arg1)
len0 = len(self.arg0)
len1 = len(self.arg1)
if len0 != len1:
raise AttributeError, \
"vectors of different lengths %d != %d" % (len0, len1)
return len0
class Proxy(Value):
def __init__(self, name, dict):
self.name = name
self.dict = dict
def __unproxy__(self):
return unproxy(self.dict[self.name])
def __getitem__(self, index):
return ItemProxy(self, index)
def __getattr__(self, attr):
return AttrProxy(self, attr)
class ItemProxy(Proxy):
def __init__(self, proxy, index):
self.proxy = proxy
self.index = index
def __unproxy__(self):
return unproxy(unproxy(self.proxy)[self.index])
class AttrProxy(Proxy):
def __init__(self, proxy, attr):
self.proxy = proxy
self.attr = attr
def __unproxy__(self):
return unproxy(getattr(unproxy(self.proxy), self.attr))
class ProxyGroup(object):
def __init__(self, dict=None, **kwargs):
self.__dict__['dict'] = {}
if dict is not None:
self.dict.update(dict)
if kwargs:
self.dict.update(kwargs)
def __getattr__(self, name):
return Proxy(name, self.dict)
def __setattr__(self, attr, value):
self.dict[attr] = value
class ScalarStat(Statistic,Scalar):
def __value__(self, run):
if run not in self.data:
return None
return self.data[run][0][0]
def display(self, run=None):
import display
p = display.Print()
p.name = self.name
p.desc = self.desc
p.value = value(self, run)
p.flags = self.flags
p.precision = self.precision
if display.all or (self.flags & flags.printable):
p.display()
class VectorStat(Statistic,Vector):
def __value__(self, run, item):
if run not in self.data:
return None
return self.data[run][item][0]
def __len__(self):
return self.x
def display(self, run=None):
import display
d = display.VectorDisplay()
d.name = self.name
d.desc = self.desc
d.value = [ value(self, run, i) for i in xrange(len(self)) ]
d.flags = self.flags
d.precision = self.precision
d.display()
class Formula(Value):
def __getattribute__(self, attr):
if attr not in ( '__scalar__', '__vector__', '__value__', '__len__' ):
return super(Formula, self).__getattribute__(attr)
formula = re.sub(':', '__', self.formula)
value = eval(formula, self.source.stattop)
return getattr(value, attr)
class SimpleDist(Statistic):
def __init__(self, sums, squares, samples):
self.sums = sums
self.squares = squares
self.samples = samples
def display(self, name, desc, flags, precision):
import display
p = display.Print()
p.flags = flags
p.precision = precision
if self.samples > 0:
p.name = name + ".mean"
p.value = self.sums / self.samples
p.display()
p.name = name + ".stdev"
if self.samples > 1:
var = (self.samples * self.squares - self.sums ** 2) \
/ (self.samples * (self.samples - 1))
if var >= 0:
p.value = math.sqrt(var)
else:
p.value = 'NaN'
else:
p.value = 0.0
p.display()
p.name = name + ".samples"
p.value = self.samples
p.display()
def comparable(self, other):
return True
def __eq__(self, other):
return self.sums == other.sums and self.squares == other.squares and \
self.samples == other.samples
def __isub__(self, other):
self.sums -= other.sums
self.squares -= other.squares
self.samples -= other.samples
return self
def __iadd__(self, other):
self.sums += other.sums
self.squares += other.squares
self.samples += other.samples
return self
def __itruediv__(self, other):
if not other:
return self
self.sums /= other
self.squares /= other
self.samples /= other
return self
class FullDist(SimpleDist):
def __init__(self, sums, squares, samples, minval, maxval,
under, vec, over, min, max, bsize, size):
self.sums = sums
self.squares = squares
self.samples = samples
self.minval = minval
self.maxval = maxval
self.under = under
self.vec = vec
self.over = over
self.min = min
self.max = max
self.bsize = bsize
self.size = size
def display(self, name, desc, flags, precision):
import display
p = display.Print()
p.flags = flags
p.precision = precision
p.name = name + '.min_val'
p.value = self.minval
p.display()
p.name = name + '.max_val'
p.value = self.maxval
p.display()
p.name = name + '.underflow'
p.value = self.under
p.display()
i = self.min
for val in self.vec[:-1]:
p.name = name + '[%d:%d]' % (i, i + self.bsize - 1)
p.value = val
p.display()
i += self.bsize
p.name = name + '[%d:%d]' % (i, self.max)
p.value = self.vec[-1]
p.display()
p.name = name + '.overflow'
p.value = self.over
p.display()
SimpleDist.display(self, name, desc, flags, precision)
def comparable(self, other):
return self.min == other.min and self.max == other.max and \
self.bsize == other.bsize and self.size == other.size
def __eq__(self, other):
return self.sums == other.sums and self.squares == other.squares and \
self.samples == other.samples
def __isub__(self, other):
self.sums -= other.sums
self.squares -= other.squares
self.samples -= other.samples
if other.samples:
self.minval = min(self.minval, other.minval)
self.maxval = max(self.maxval, other.maxval)
self.under -= under
self.vec = map(lambda x,y: x - y, self.vec, other.vec)
self.over -= over
return self
def __iadd__(self, other):
if not self.samples and other.samples:
self = other
return self
self.sums += other.sums
self.squares += other.squares
self.samples += other.samples
if other.samples:
self.minval = min(self.minval, other.minval)
self.maxval = max(self.maxval, other.maxval)
self.under += other.under
self.vec = map(lambda x,y: x + y, self.vec, other.vec)
self.over += other.over
return self
def __itruediv__(self, other):
if not other:
return self
self.sums /= other
self.squares /= other
self.samples /= other
if self.samples:
self.under /= other
for i in xrange(len(self.vec)):
self.vec[i] /= other
self.over /= other
return self
class Dist(Statistic):
def display(self):
import display
if not display.all and not (self.flags & flags.printable):
return
self.dist.display(self.name, self.desc, self.flags, self.precision)
def comparable(self, other):
return self.name == other.name and \
self.dist.compareable(other.dist)
def __eq__(self, other):
return self.dist == other.dist
def __isub__(self, other):
self.dist -= other.dist
return self
def __iadd__(self, other):
self.dist += other.dist
return self
def __itruediv__(self, other):
if not other:
return self
self.dist /= other
return self
class VectorDist(Statistic):
def display(self):
import display
if not display.all and not (self.flags & flags.printable):
return
if isinstance(self.dist, SimpleDist):
return
for dist,sn,sd,i in map(None, self.dist, self.subnames, self.subdescs,
range(len(self.dist))):
if len(sn) > 0:
name = '%s.%s' % (self.name, sn)
else:
name = '%s[%d]' % (self.name, i)
if len(sd) > 0:
desc = sd
else:
desc = self.desc
dist.display(name, desc, self.flags, self.precision)
if (self.flags & flags.total) or 1:
if isinstance(self.dist[0], SimpleDist):
disttotal = SimpleDist( \
reduce(sums, [d.sums for d in self.dist]),
reduce(sums, [d.squares for d in self.dist]),
reduce(sums, [d.samples for d in self.dist]))
else:
disttotal = FullDist( \
reduce(sums, [d.sums for d in self.dist]),
reduce(sums, [d.squares for d in self.dist]),
reduce(sums, [d.samples for d in self.dist]),
min([d.minval for d in self.dist]),
max([d.maxval for d in self.dist]),
reduce(sums, [d.under for d in self.dist]),
reduce(sums, [d.vec for d in self.dist]),
reduce(sums, [d.over for d in self.dist]),
dist[0].min,
dist[0].max,
dist[0].bsize,
dist[0].size)
name = '%s.total' % (self.name)
desc = self.desc
disttotal.display(name, desc, self.flags, self.precision)
def comparable(self, other):
return self.name == other.name and \
alltrue(map(lambda x, y : x.comparable(y),
self.dist,
other.dist))
def __eq__(self, other):
return alltrue(map(lambda x, y : x == y, self.dist, other.dist))
def __isub__(self, other):
if isinstance(self.dist, (list, tuple)) and \
isinstance(other.dist, (list, tuple)):
for sd,od in zip(self.dist, other.dist):
sd -= od
else:
self.dist -= other.dist
return self
def __iadd__(self, other):
if isinstance(self.dist, (list, tuple)) and \
isinstance(other.dist, (list, tuple)):
for sd,od in zip(self.dist, other.dist):
sd += od
else:
self.dist += other.dist
return self
def __itruediv__(self, other):
if not other:
return self
if isinstance(self.dist, (list, tuple)):
for dist in self.dist:
dist /= other
else:
self.dist /= other
return self
class Vector2d(Statistic):
def display(self):
import display
if not display.all and not (self.flags & flags.printable):
return
d = display.VectorDisplay()
d.__dict__.update(self.__dict__)
if self.__dict__.has_key('ysubnames'):
ysubnames = list(self.ysubnames)
slack = self.x - len(ysubnames)
if slack > 0:
ysubnames.extend(['']*slack)
else:
ysubnames = range(self.x)
for x,sname in enumerate(ysubnames):
o = x * self.y
d.value = self.value[o:o+self.y]
d.name = '%s[%s]' % (self.name, sname)
d.display()
if self.flags & flags.total:
d.value = []
for y in range(self.y):
xtot = 0.0
for x in range(self.x):
xtot += self.value[y + x * self.x]
d.value.append(xtot)
d.name = self.name + '.total'
d.display()
def comparable(self, other):
return self.name == other.name and self.x == other.x and \
self.y == other.y
def __eq__(self, other):
return True
def __isub__(self, other):
return self
def __iadd__(self, other):
return self
def __itruediv__(self, other):
if not other:
return self
return self
def NewStat(source, data):
stat = None
if data.type == 'SCALAR':
stat = ScalarStat()
elif data.type == 'VECTOR':
stat = VectorStat()
elif data.type == 'DIST':
stat = Dist()
elif data.type == 'VECTORDIST':
stat = VectorDist()
elif data.type == 'VECTOR2D':
stat = Vector2d()
elif data.type == 'FORMULA':
stat = Formula()
stat.__dict__['source'] = source
stat.__dict__['bins'] = None
stat.__dict__['ticks'] = None
stat.__dict__.update(data.__dict__)
return stat