def timing3(n=2000, best_time=1.0):
print 'Grid2Deff2.timing2: reference CPU time = %g' % best_time
dx = 1.0 / n
g = Grid2Deff2(dx=dx, dy=dx)
# here we use straight NumPy sin in a scalar context:
def myfunc1(x, y):
return sin(x * y) + 8 * x
def myfunc2(x, y):
return math.sin(x * y) + 8 * x
from scitools.misc import timer
from scitools.EfficiencyTable import EfficiencyTable
e = EfficiencyTable('Grid2Deff2 tests', best_time)
print 'basic NumPy module:', basic_NumPy
if basic_NumPy != 'Numeric':
print 'only Numeric is meaningful for the C++ extension module'
return
t1a = timer(g.ext_gridloop1, (myfunc1, ), repetitions=1)
e.add('g.ext_gridloop1, myfunc1', t1a)
t1b = timer(g.ext_gridloop1, (myfunc2, ), repetitions=1)
e.add('g.ext_gridloop1, myfunc2', t1b)
t2a = timer(g.ext_gridloop2, (myfunc1, ), repetitions=1)
e.add('g.ext_gridloop2, myfunc1', t2a)
t2b = timer(g.ext_gridloop2, (myfunc2, ), repetitions=1)
e.add('g.ext_gridloop2, myfunc2', t2b)
print e
def _efficiency():
print('\nPerform some efficiency tests (this might take some time...):')
formula = 'sin(x) + x**3 + 2*x'
formula_wprm = formula + ' + A*B'
def s0(x):
return sin(x) + x**3 + 2 * x
s1 = StringFunction_v1(formula)
s2 = StringFunction_v2(formula) # compiled
s3 = StringFunction_v3(formula_wprm, set_parameters='A=0; B=0')
s4 = StringFunction_v4(formula)
s5 = StringFunction_v5(formula, independent_variables=('x', ), A=0, B=0)
s6 = StringFunction(formula, independent_variables=('x', ), A=0, B=0)
s7 = s6.__call__
x = 0.9
# verification first:
values = [s(x) for s in (s0, s1, s2, s3, s4, s5, s6, s7)]
print('values of %s for x=%s: %s' % (formula, x, values))
n = 400000
from scitools.misc import timer
from scitools.EfficiencyTable import EfficiencyTable as ET
e = ET('Efficiency check of StringFunction implementations; '
'formula=%s, n=%d' % (formula, n))
import inspect
for s in s0, s1, s2, s3, s4, s5, s6, s7:
if inspect.isfunction(s) or inspect.ismethod(s):
name = s.__name__
else:
name = s.__class__.__name__
t = timer(s, args=(x, ), repetitions=100000, comment=name)
e.add(name, t)
print(e)
print('End of efficiency tests\n')
def timing3(n=2000, best_time=1.0):
print 'Grid2Deff2.timing2: reference CPU time = %g' % best_time
dx = 1.0/n
g = Grid2Deff2(dx=dx, dy=dx)
# here we use straight NumPy sin in a scalar context:
def myfunc1(x, y):
return sin(x*y) + 8*x
def myfunc2(x, y):
return math.sin(x*y) + 8*x
from scitools.misc import timer
from scitools.EfficiencyTable import EfficiencyTable
e = EfficiencyTable('Grid2Deff2 tests', best_time)
print 'basic NumPy module:', basic_NumPy
if basic_NumPy != 'Numeric':
print 'only Numeric is meaningful for the C++ extension module'
return
t1a = timer(g.ext_gridloop1, (myfunc1,), repetitions=1)
e.add('g.ext_gridloop1, myfunc1', t1a)
t1b = timer(g.ext_gridloop1, (myfunc2,), repetitions=1)
e.add('g.ext_gridloop1, myfunc2', t1b)
t2a = timer(g.ext_gridloop2, (myfunc1,), repetitions=1)
e.add('g.ext_gridloop2, myfunc1', t2a)
t2b = timer(g.ext_gridloop2, (myfunc2,), repetitions=1)
e.add('g.ext_gridloop2, myfunc2', t2b)
print e
def _efficiency():
print('\nPerform some efficiency tests (this might take some time...):')
formula = 'sin(x) + x**3 + 2*x'
formula_wprm = formula + ' + A*B'
def s0(x):
return sin(x) + x**3 + 2 * x
s1 = StringFunction_v1(formula)
s2 = StringFunction_v2(formula) # compiled
s3 = StringFunction_v3(formula_wprm, set_parameters='A=0; B=0')
s4 = StringFunction_v4(formula)
s5 = StringFunction_v5(formula, independent_variables=('x',),
A=0, B=0)
s6 = StringFunction(formula, independent_variables=('x',),
A=0, B=0)
s7 = s6.__call__
x = 0.9
# verification first:
values = [s(x) for s in (s0, s1, s2, s3, s4, s5, s6, s7)]
print('values of %s for x=%s: %s' % (formula, x, values))
n = 400000
from scitools.misc import timer
from scitools.EfficiencyTable import EfficiencyTable as ET
e = ET('Efficiency check of StringFunction implementations; '
'formula=%s, n=%d' % (formula, n))
import inspect
for s in s0, s1, s2, s3, s4, s5, s6, s7:
if inspect.isfunction(s) or inspect.ismethod(s):
name = s.__name__
else:
name = s.__class__.__name__
t = timer(s, args=(x,), repetitions=100000,
comment=name)
e.add(name, t)
print(e)
print('End of efficiency tests\n')
def timing2(n=2000, best_time=1.0):
"""Time different implementations of the extension module."""
print 'Grid2Deff.timing2: reference CPU time = %g' % best_time
dx = 1.0/n
g = Grid2Deff(dx=dx, dy=dx)
# here we use straight NumPy sin in a scalar context:
def myfunc1(x, y):
return sin(x*y) + 8*x
def myfunc2(x, y):
return math.sin(x*y) + 8*x
expression1 = StringFunction('sin(x*y) + 8*x',
independent_variables=('x','y'),
globals=globals())
expression1_f = expression1.__call__ # for efficiency and F77 callback
expression2 = StringFunction('math.sin(x*y) + 8*x',
independent_variables=('x','y'),
globals=globals())
expression2_f = expression2.__call__ # for efficiency and F77 callback
from scitools.misc import timer
from scitools.EfficiencyTable import EfficiencyTable
e = EfficiencyTable('Grid2Deff tests, %dx%d grid' % (n,n), best_time)
t0a = timer(g.gridloop, (myfunc1,), repetitions=1)
e.add('g.gridloop, myfunc1', t0a)
t0b = timer(g.gridloop, (myfunc2,), repetitions=1)
e.add('g.gridloop, myfunc2', t0b)
t0c = timer(g.__call__, (myfunc1,), repetitions=1)
e.add('g.__call__, myfunc1', t0c)
t0d = timer(g.__call__, (expression1_f,), repetitions=1)
e.add('g.__call__, expression1_f', t0d)
t0e = timer(g.gridloop_itemset, (myfunc2,), repetitions=1)
e.add('g.gridloop_itemset, myfunc2', t0e)
t1a = timer(g.ext_gridloop1, (myfunc1,), repetitions=1)
e.add('g.ext_gridloop1, myfunc1', t1a)
t1b = timer(g.ext_gridloop1, (myfunc2,), repetitions=1)
e.add('g.ext_gridloop1, myfunc2', t1b)
t2a = timer(g.ext_gridloop2, (myfunc1,), repetitions=1)
e.add('g.ext_gridloop2, myfunc1', t2a)
t2b = timer(g.ext_gridloop2, (myfunc2,), repetitions=1)
e.add('g.ext_gridloop2, myfunc2', t2b)
t3a = timer(g.ext_gridloop2, (expression1_f,), repetitions=1)
e.add('g.ext_gridloop2, expression1_f', t3a)
t3b = timer(g.ext_gridloop2, (expression2_f,), repetitions=1)
e.add('g.ext_gridloop2, expression2_f', t3b)
nrep = 20
# try the improved functions (works only for the F77 module):
if 'gridloop_vec2' in dir(ext_gridloop):
t4 = timer(g.ext_gridloop_vec2, (myfuncf2,), repetitions=nrep)
e.add('g.ext_gridloop_vec2, myfuncf2', t4)
if 'gridloop2_str' in dir(ext_gridloop):
t5 = timer(g.ext_gridloop2_str, ('myfunc',), repetitions=nrep)
e.add('g.ext_gridloop2_str, myfunc', t5)
# try the version without allocation (first, make an a array):
a = g.ext_gridloop2(myfunc1) # a has now Fortran storage
t5b = timer(g.ext_gridloop_noalloc,
('myfunc', a), repetitions=nrep)
e.add('g.ext_gridloop_noalloc, myfunc', t5b)
# try 'inline' F77 compiled callback too:
# (give F77 source for core of callback function as argument)
g.ext_gridloop2_fcb_compile(str(expression1))
t6 = timer(g.ext_gridloop2_fcb, (), repetitions=nrep)
e.add('g.ext_gridloop2_fcb(%s)' % repr(str(expression1)), t6)
g.ext_gridloop2_fcb_ptr_compile(expression1)
t6b = timer(g.ext_gridloop2_fcb_ptr, (), repetitions=nrep)
e.add('g.ext_gridloop2_fcb_ptr(%s)' % repr(expression1), t6b)
g.ext_gridloop2_compile(str(expression1))
t7 = timer(g.ext_gridloop2_v2, (), repetitions=nrep)
e.add('g.ext_gridloop2_v2(%s)' % repr(str(expression1)), t7)
# weave version:
t8 = timer(g.ext_gridloop2_weave, (str(expression1),), repetitions=nrep)
e.add('g.ext_gridloop2_weave(%s)' % repr(str(expression1)), t8)
# psyco:
g.gridloop_psyco_init(g.gridloop)
if g.gridloop_psyco != g.gridloop: # has psyco
t9a = timer(g.gridloop_psyco, (myfunc2,), repetitions=1)
e.add('g.gridloop_psyco, myfunc2', t9a)
t9b = timer(g.gridloop_psyco, (expression2_f,), repetitions=1)
e.add('g.gridloop_psyco, expression2_f', t9b)
g.gridloop_psyco_init(g.gridloop_itemset)
if g.gridloop_psyco != g.gridloop_itemset: # has psyco
t9a = timer(g.gridloop_psyco, (myfunc2,), repetitions=1)
e.add('g.gridloop_psyco (itemset), myfunc2', t9a)
t9b = timer(g.gridloop_psyco, (expression2_f,), repetitions=1)
e.add('g.gridloop_psyco (itemset), expression2_f', t9b)
# instant:
g.ext_gridloop1_instant(str(expression1))
if g.gridloop1_instant is not None:
a = zeros((self.nx, self.ny))
t10 = timer(g.gridloop1_instant,
(a, self.nx, g.ny, g.xcoor, g.ycoor),
repetitions=nrep)
e.add('g.gridloop1_instant', t10)
print '\n\n\n\nrun from directory', os.getcwd()
print e
Comparison of scalar and vectorized sine functions when
the argument is a scalar.
A test on a**b vs pow(a,b) is also included.
"""
import timeit, sys
try:
n = int(sys.argv[1])
except:
n = 1000000 # default
print 'n =', n
from scitools.EfficiencyTable import EfficiencyTable
e = EfficiencyTable('%d evaluations of arcsin(0.4)' % n)
t1 = timeit.Timer('arcsin(0.4)',
setup='from numpy import arcsin').timeit(n)
t2 = timeit.Timer('asin(0.4)',
setup='from math import asin').timeit(n)
t3 = timeit.Timer('arcsin(0.4)',
setup='from numarray import arcsin').timeit(n)
t4 = timeit.Timer('arcsin(0.4)',
setup='from Numeric import arcsin').timeit(n)
t5 = timeit.Timer('arcsin(0.4)',
setup='from numpy.lib.scimath import arcsin').timeit(n)
e.add('numpy arcsin', t1)
e.add('math asin', t2)
e.add('numarray arcsin', t3)
e.add('Numeric arcsin', t4)
return r
def somefunc_NumPy_logv(x):
x_pos = where(x > 0, x, 1)
r1 = log(x_pos)
r = where(x < 0, 0.0, r1)
return r
somefunc_list = [somefuncv, somefunc_NumPy, somefunc_NumPy2,
somefunc_NumPy2b, somefunc_NumPy3,
somefunc_NumPy_log, somefunc_NumPy_logv]
# check correctness:
x = sequence(-2, 2, 2)
print '\nsomefunc_* functions applied to', x
for f in somefunc_list:
print f(x)
n = 5000000
#n = 500000
x = linspace(0, 2, n+1)
print '\nperforming some timings of "somefunc*" implementations...'
from scitools.EfficiencyTable import EfficiencyTable as ET
from scitools.misc import timer
e = ET('vectorization of "somefunc" functions with an if test, n=%d' % n)
for f in somefunc_list[:-2]: # skip the last two log functions
t = timer(f, (x,), repetitions=1)
e.add(f.__name__, t)
print e
print 'end of', sys.argv[0]
def timing2(n=2000, best_time=1.0):
"""Time different implementations of the extension module."""
print 'Grid2Deff.timing2: reference CPU time = %g' % best_time
dx = 1.0 / n
g = Grid2Deff(dx=dx, dy=dx)
# here we use straight NumPy sin in a scalar context:
def myfunc1(x, y):
return sin(x * y) + 8 * x
def myfunc2(x, y):
return math.sin(x * y) + 8 * x
expression1 = StringFunction('sin(x*y) + 8*x',
independent_variables=('x', 'y'),
globals=globals())
expression1_f = expression1.__call__ # for efficiency and F77 callback
expression2 = StringFunction('math.sin(x*y) + 8*x',
independent_variables=('x', 'y'),
globals=globals())
expression2_f = expression2.__call__ # for efficiency and F77 callback
from scitools.misc import timer
from scitools.EfficiencyTable import EfficiencyTable
e = EfficiencyTable('Grid2Deff tests, %dx%d grid' % (n, n), best_time)
t0a = timer(g.gridloop, (myfunc1, ), repetitions=1)
e.add('g.gridloop, myfunc1', t0a)
t0b = timer(g.gridloop, (myfunc2, ), repetitions=1)
e.add('g.gridloop, myfunc2', t0b)
t0c = timer(g.__call__, (myfunc1, ), repetitions=1)
e.add('g.__call__, myfunc1', t0c)
t0d = timer(g.__call__, (expression1_f, ), repetitions=1)
e.add('g.__call__, expression1_f', t0d)
t0e = timer(g.gridloop_itemset, (myfunc2, ), repetitions=1)
e.add('g.gridloop_itemset, myfunc2', t0e)
t1a = timer(g.ext_gridloop1, (myfunc1, ), repetitions=1)
e.add('g.ext_gridloop1, myfunc1', t1a)
t1b = timer(g.ext_gridloop1, (myfunc2, ), repetitions=1)
e.add('g.ext_gridloop1, myfunc2', t1b)
t2a = timer(g.ext_gridloop2, (myfunc1, ), repetitions=1)
e.add('g.ext_gridloop2, myfunc1', t2a)
t2b = timer(g.ext_gridloop2, (myfunc2, ), repetitions=1)
e.add('g.ext_gridloop2, myfunc2', t2b)
t3a = timer(g.ext_gridloop2, (expression1_f, ), repetitions=1)
e.add('g.ext_gridloop2, expression1_f', t3a)
t3b = timer(g.ext_gridloop2, (expression2_f, ), repetitions=1)
e.add('g.ext_gridloop2, expression2_f', t3b)
nrep = 20
# try the improved functions (works only for the F77 module):
if 'gridloop_vec2' in dir(ext_gridloop):
t4 = timer(g.ext_gridloop_vec2, (myfuncf2, ), repetitions=nrep)
e.add('g.ext_gridloop_vec2, myfuncf2', t4)
if 'gridloop2_str' in dir(ext_gridloop):
t5 = timer(g.ext_gridloop2_str, ('myfunc', ), repetitions=nrep)
e.add('g.ext_gridloop2_str, myfunc', t5)
# try the version without allocation (first, make an a array):
a = g.ext_gridloop2(myfunc1) # a has now Fortran storage
t5b = timer(g.ext_gridloop_noalloc, ('myfunc', a), repetitions=nrep)
e.add('g.ext_gridloop_noalloc, myfunc', t5b)
# try 'inline' F77 compiled callback too:
# (give F77 source for core of callback function as argument)
g.ext_gridloop2_fcb_compile(str(expression1))
t6 = timer(g.ext_gridloop2_fcb, (), repetitions=nrep)
e.add('g.ext_gridloop2_fcb(%s)' % repr(str(expression1)), t6)
g.ext_gridloop2_fcb_ptr_compile(expression1)
t6b = timer(g.ext_gridloop2_fcb_ptr, (), repetitions=nrep)
e.add('g.ext_gridloop2_fcb_ptr(%s)' % repr(expression1), t6b)
g.ext_gridloop2_compile(str(expression1))
t7 = timer(g.ext_gridloop2_v2, (), repetitions=nrep)
e.add('g.ext_gridloop2_v2(%s)' % repr(str(expression1)), t7)
# weave version:
t8 = timer(g.ext_gridloop2_weave, (str(expression1), ), repetitions=nrep)
e.add('g.ext_gridloop2_weave(%s)' % repr(str(expression1)), t8)
# psyco:
g.gridloop_psyco_init(g.gridloop)
if g.gridloop_psyco != g.gridloop: # has psyco
t9a = timer(g.gridloop_psyco, (myfunc2, ), repetitions=1)
e.add('g.gridloop_psyco, myfunc2', t9a)
t9b = timer(g.gridloop_psyco, (expression2_f, ), repetitions=1)
e.add('g.gridloop_psyco, expression2_f', t9b)
g.gridloop_psyco_init(g.gridloop_itemset)
if g.gridloop_psyco != g.gridloop_itemset: # has psyco
t9a = timer(g.gridloop_psyco, (myfunc2, ), repetitions=1)
e.add('g.gridloop_psyco (itemset), myfunc2', t9a)
t9b = timer(g.gridloop_psyco, (expression2_f, ), repetitions=1)
e.add('g.gridloop_psyco (itemset), expression2_f', t9b)
# instant:
g.ext_gridloop1_instant(str(expression1))
if g.gridloop1_instant is not None:
a = zeros((self.nx, self.ny))
t10 = timer(g.gridloop1_instant, (a, self.nx, g.ny, g.xcoor, g.ycoor),
repetitions=nrep)
e.add('g.gridloop1_instant', t10)
print '\n\n\n\nrun from directory', os.getcwd()
print e
def timing1(n=2000):
# timing:
dx = 1.0/n
g = Grid2D(xmin=0, xmax=1, dx=dx,
ymin=0, ymax=1, dy=dx)
def myfunc(x, y):
return sin(x*y) + 8*x
from scitools.StringFunction import StringFunction
expression = StringFunction('sin(x*y) + 8*x',
independent_variables=('x','y'),
globals=globals() # needed for vectorization
)
from scitools.misc import timer
from scitools.EfficiencyTable import EfficiencyTable
e = EfficiencyTable('Grid2D efficiency tests, %dx%d grid' % (n,n))
t0 = time.clock()
# vectorized expressions are so fast that we run the code
# repeatedly
rep=20
t1 = timer(g.__call__, (expression,), repetitions=rep,
comment='StringFunction')
e.add('g.__call__, expression (vectorized)', t1)
t2 = timer(g.__call__, (myfunc,), repetitions=rep, comment='myfunc')
e.add('g.__call__, myfunc (vectorized)', t2)
f = g.gridloop_hardcoded_func()
t3 = timer(g.gridloop_hardcoded_func, (), repetitions=1, comment='')
e.add('g.gridloop_hardcoded_func', t3)
t4 = timer(g.gridloop, (expression,), repetitions=1,
comment='StringFunction')
e.add('g.gridloop, expression (explicit loops)', t4)
t5 = timer(g.gridloop, (myfunc,), repetitions=1, comment='myfunc')
e.add('g.gridloop, myfunc (explicit loops)', t4)
t6 = timer(g.gridloop_list, (expression,), repetitions=1,
comment='StringFunction')
e.add('g.gridloop_list, expression (explicit loops)', t6)
t7 = timer(g.gridloop_list, (myfunc,), repetitions=1, comment='myfunc')
e.add('g.gridloop_list, myfunc (explicit loops)', t7)
from math import sin as mathsin # efficiency trick
# The scalar computations above used sin from NumPy, which is
# known to be slow for scalar arguments. Here we use math.sin
# (stored in mathsin, could also use the slightly slower math.sin
# explicitly)
# taken globally so eval works: from math import sin as mathsin
def myfunc_scalar(x, y):
return mathsin(x*y) + 8*x
expression_scalar = StringFunction('mathsin(x*y) + 8*x',
independent_variables=('x','y'),
globals=globals())
t8 = timer(g.gridloop, (expression_scalar,), repetitions=1,
comment='eval(str)')
e.add('g.gridloop, expression_scalar', t8)
t9 = timer(g.gridloop, (myfunc_scalar,), repetitions=1, comment='myfunc')
e.add('g.gridloop, myfunc_scalar', t9)
print e
def timing1(n=2000):
# timing:
dx = 1.0 / n
g = Grid2D(xmin=0, xmax=1, dx=dx, ymin=0, ymax=1, dy=dx)
def myfunc(x, y):
return sin(x * y) + 8 * x
from scitools.StringFunction import StringFunction
expression = StringFunction(
'sin(x*y) + 8*x',
independent_variables=('x', 'y'),
globals=globals() # needed for vectorization
)
from scitools.misc import timer
from scitools.EfficiencyTable import EfficiencyTable
e = EfficiencyTable('Grid2D efficiency tests, %dx%d grid' % (n, n))
t0 = time.clock()
# vectorized expressions are so fast that we run the code
# repeatedly
rep = 20
t1 = timer(g.__call__, (expression, ),
repetitions=rep,
comment='StringFunction')
e.add('g.__call__, expression (vectorized)', t1)
t2 = timer(g.__call__, (myfunc, ), repetitions=rep, comment='myfunc')
e.add('g.__call__, myfunc (vectorized)', t2)
f = g.gridloop_hardcoded_func()
t3 = timer(g.gridloop_hardcoded_func, (), repetitions=1, comment='')
e.add('g.gridloop_hardcoded_func', t3)
t4 = timer(g.gridloop, (expression, ),
repetitions=1,
comment='StringFunction')
e.add('g.gridloop, expression (explicit loops)', t4)
t5 = timer(g.gridloop, (myfunc, ), repetitions=1, comment='myfunc')
e.add('g.gridloop, myfunc (explicit loops)', t4)
t6 = timer(g.gridloop_list, (expression, ),
repetitions=1,
comment='StringFunction')
e.add('g.gridloop_list, expression (explicit loops)', t6)
t7 = timer(g.gridloop_list, (myfunc, ), repetitions=1, comment='myfunc')
e.add('g.gridloop_list, myfunc (explicit loops)', t7)
from math import sin as mathsin # efficiency trick
# The scalar computations above used sin from NumPy, which is
# known to be slow for scalar arguments. Here we use math.sin
# (stored in mathsin, could also use the slightly slower math.sin
# explicitly)
# taken globally so eval works: from math import sin as mathsin
def myfunc_scalar(x, y):
return mathsin(x * y) + 8 * x
expression_scalar = StringFunction('mathsin(x*y) + 8*x',
independent_variables=('x', 'y'),
globals=globals())
t8 = timer(g.gridloop, (expression_scalar, ),
repetitions=1,
comment='eval(str)')
e.add('g.gridloop, expression_scalar', t8)
t9 = timer(g.gridloop, (myfunc_scalar, ), repetitions=1, comment='myfunc')
e.add('g.gridloop, myfunc_scalar', t9)
print e
