Velocity Reviews > int vs. float in benchmark testing

# int vs. float in benchmark testing

Bart Nessux
Guest
Posts: n/a

 02-20-2004
Would adding .0 to each of the numbers below turn this into a floating
point test? Seems too easy.

def cpu_test():
import time
start = time.time()
x = 0 # 0.0
while x < 9999999: # 9999999.0
x = x + 1 # 1.0
print x
print (time.time()-start)/60
cpu_test()

Peter Hansen
Guest
Posts: n/a

 02-20-2004
Bart Nessux wrote:
>
> Would adding .0 to each of the numbers below turn this into a floating
> point test? Seems too easy.
>
> def cpu_test():
> import time
> start = time.time()
> x = 0 # 0.0
> while x < 9999999: # 9999999.0
> x = x + 1 # 1.0
> print x
> print (time.time()-start)/60
> cpu_test()

Uh, yes it would, as far as it goes, but are you sure you're
learning something useful by doing so? Floats are always slower
than ints, but you really shouldn't be concerned about speed
anyway.

The way to decide which to use is this: if you need floating
point because you are doing math that involves fractional values,
then use floating point. Otherwise use ints. End of story.
No performance considerations in most code.

-Peter

Bart Nessux
Guest
Posts: n/a

 02-20-2004
Peter Hansen wrote:
> Bart Nessux wrote:
>
>>Would adding .0 to each of the numbers below turn this into a floating
>>point test? Seems too easy.
>>
>>def cpu_test():
>> import time
>> start = time.time()
>> x = 0 # 0.0
>> while x < 9999999: # 9999999.0
>> x = x + 1 # 1.0
>> print x
>> print (time.time()-start)/60
>>cpu_test()

>
>
> Uh, yes it would, as far as it goes, but are you sure you're
> learning something useful by doing so?

Uh, yes. We're benchmarking different processors. An IBM PPC 970 does
things differently than an Intel P4. Running the same bit of Python code
on both makes for an interesting comparison. Since processors handle
ints and floats differently, it is useful for me to test them both.

Peter Hansen
Guest
Posts: n/a

 02-20-2004
Bart Nessux wrote:
>
> Uh, yes. We're benchmarking different processors. An IBM PPC 970 does
> things differently than an Intel P4. Running the same bit of Python code
> on both makes for an interesting comparison. Since processors handle
> ints and floats differently, it is useful for me to test them both.

Then why not get yourself some real benchmarks?

Benchmarks are a tricky thing. Unless your real code is doing
something that looks an awful lot like the above (looping and adding
1.0 to things a lot), it seems unlikely what you learn will really be
what you wanted to learn.

-Peter

David E. Konerding DSD staff
Guest
Posts: n/a

 02-20-2004
In article <c15899\$ag1\$>, Bart Nessux wrote:
> Peter Hansen wrote:
>> Bart Nessux wrote:
>>
>>>Would adding .0 to each of the numbers below turn this into a floating
>>>point test? Seems too easy.
>>>
>>>def cpu_test():
>>> import time
>>> start = time.time()
>>> x = 0 # 0.0
>>> while x < 9999999: # 9999999.0
>>> x = x + 1 # 1.0
>>> print x
>>> print (time.time()-start)/60
>>>cpu_test()

>>
>>
>> Uh, yes it would, as far as it goes, but are you sure you're
>> learning something useful by doing so?

>
> Uh, yes. We're benchmarking different processors. An IBM PPC 970 does
> things differently than an Intel P4. Running the same bit of Python code
> on both makes for an interesting comparison. Since processors handle
> ints and floats differently, it is useful for me to test them both.

I bet (significantly) more time is being spent in the python byte code processing machinery than the
actual chip-level instructions performing the integer and floating point math, so your
processor-differential results will be masked by that.

Dave

Andrew MacIntyre
Guest
Posts: n/a

 02-20-2004
On Fri, 20 Feb 2004, Peter Hansen wrote:

> Bart Nessux wrote:
> >
> > Uh, yes. We're benchmarking different processors. An IBM PPC 970 does
> > things differently than an Intel P4. Running the same bit of Python code
> > on both makes for an interesting comparison. Since processors handle
> > ints and floats differently, it is useful for me to test them both.

>
> Then why not get yourself some real benchmarks?

Marc-Andre Lemburg's PyBench covers this sort thing.

--
Andrew I MacIntyre "These thoughts are mine alone..."
E-mail: (pref) | Snail: PO Box 370
(alt) | Belconnen ACT 2616
Web: http://www.andymac.org/ | Australia

David Morgenthaler
Guest
Posts: n/a

 02-22-2004
On Fri, 20 Feb 2004 09:53:04 -0500, Peter Hansen <>
wrote:

I think you'll mainly be benchmarking the 'print x' rather than the

>Bart Nessux wrote:
>>
>> Would adding .0 to each of the numbers below turn this into a floating
>> point test? Seems too easy.
>>
>> def cpu_test():
>> import time
>> start = time.time()
>> x = 0 # 0.0
>> while x < 9999999: # 9999999.0
>> x = x + 1 # 1.0
>> print x
>> print (time.time()-start)/60
>> cpu_test()

>
>Uh, yes it would, as far as it goes, but are you sure you're
>learning something useful by doing so? Floats are always slower
>than ints, but you really shouldn't be concerned about speed
>anyway.
>
>The way to decide which to use is this: if you need floating
>point because you are doing math that involves fractional values,
>then use floating point. Otherwise use ints. End of story.
>No performance considerations in most code.
>
>-Peter