Speed hack for function calls with named parameters

[pitrou]

BPO	1819
Nosy	@malemburg, @gvanrossum, @birkenfeld, @rhettinger, @facundobatista, @pitrou, @tiran
Files	namedparam.patch namedparam.patch namedparam2.patch namedparam3.patch: Faster, cleaner with PySequence_Fast_ITEMS pybench.patch

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/birkenfeld'
closed_at = <Date 2008-07-25.22:15:07.764>
created_at = <Date 2008-01-14.00:48:19.797>
labels = ['interpreter-core', 'type-feature']
title = 'Speed hack for function calls with named parameters'
updated_at = <Date 2008-07-25.22:15:07.763>
user = 'https://github.com/pitrou'

bugs.python.org fields:

activity = <Date 2008-07-25.22:15:07.763>
actor = 'pitrou'
assignee = 'georg.brandl'
closed = True
closed_date = <Date 2008-07-25.22:15:07.764>
closer = 'pitrou'
components = ['Interpreter Core']
creation = <Date 2008-01-14.00:48:19.797>
creator = 'pitrou'
dependencies = []
files = ['9156', '9157', '10590', '10596', '10600']
hgrepos = []
issue_num = 1819
keywords = ['patch']
message_count = 12.0
messages = ['59878', '59879', '59880', '59882', '61556', '68007', '68025', '68026', '68040', '68070', '68092', '70283']
nosy_count = 7.0
nosy_names = ['lemburg', 'gvanrossum', 'georg.brandl', 'rhettinger', 'facundobatista', 'pitrou', 'christian.heimes']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = None
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue1819'
versions = ['Python 2.6']

[pitrou]

This is a patch for SVN trunk which substantially speeds up function
calls with named parameters. It does so by taking into account that
parameter names should be interned, so before doing full compares we do
a first quick loop to compare pointers.

On a microbenchmark the speedup is quite distinctive:

# With patch
$ ./python -m timeit -s "def f(a,b,c,d,e): pass" "f(1,2,3,4,e=5)"
1000000 loops, best of 3: 0.515 usec per loop
$ ./python -m timeit -s "def f(a,b,c,d,e): pass" "f(a=1,b=2,c=3,d=4,e=5)"
1000000 loops, best of 3: 0.652 usec per loop

# Without patch
$ ./python-orig -m timeit -s "def f(a,b,c,d,e): pass" "f(1,2,3,4,e=5)"
1000000 loops, best of 3: 0.664 usec per loop
$ ./python-orig -m timeit -s "def f(a,b,c,d,e): pass"
"f(a=1,b=2,c=3,d=4,e=5)"
1000000 loops, best of 3: 1.07 usec per loop

[tiran]

Nice! Is this somehow related to bpo-1479611?

The formatting of the code looks kinda strange. Have you mixed tabs and
spaces?

[pitrou]

Sorry, my editor indents with spaces by default. Attaching a fixed patch
with tabs.

No, it is independent from bpo-1479611 (and much simpler).

[pitrou]

Another quick test:

# With patch
$ ./python -m timeit -s "d=dict(a=1,b=2,c=3,d=4,e=5);f = lambda
a,b,c,d,e:0" "f(**d)"
1000000 loops, best of 3: 0.727 usec per loop
$ ./python -m timeit -s "d=dict(b=2,c=3,d=4,e=5);f = lambda a,b,c,d,e:0"
"f(a=1,**d)"
1000000 loops, best of 3: 1.16 usec per loop
$ ./python -m timeit -s "d=dict(b=2,c=3,d=4,e=5); f=lambda **kw:0" "f(**d)"
1000000 loops, best of 3: 0.917 usec per loop

# Without patch
$ ./python-orig -m timeit -s "d=dict(a=1,b=2,c=3,d=4,e=5);f = lambda
a,b,c,d,e:0" "f(**d)"
1000000 loops, best of 3: 1.24 usec per loop
$ ./python-orig -m timeit -s "d=dict(b=2,c=3,d=4,e=5);f = lambda
a,b,c,d,e:0" "f(a=1,**d)"
1000000 loops, best of 3: 1.62 usec per loop
$ ./python-orig -m timeit -s "d=dict(b=2,c=3,d=4,e=5); f=lambda **kw:0"
"f(**d)"
1000000 loops, best of 3: 0.904 usec per loop

[gvanrossum]

Nice idea, but why set the priority to high? I have no immediate time
to review this and probably won't for a while.

[pitrou]

Here is a new patch against SVN trunk. Nothing changed, except that I
updated pybench to test keyword arguments as well.

[malemburg]

On 2008-06-11 20:38, Antoine Pitrou wrote:

Antoine Pitrou <pitrou@free.fr> added the comment:

Here is a new patch against SVN trunk. Nothing changed, except that I
updated pybench to test keyword arguments as well.

Added file: http://bugs.python.org/file10590/namedparam2.patch

When changing parameters or other aspects of pybench tests, you *have*
to update the version number of the test as well. Otherwise, pybench
would compare apples with oranges.

Thanks,

Marc-Andre Lemburg
eGenix.com

Professional Python Services directly from the Source  (#1, Jun 11 2008)
 >>> Python/Zope Consulting and Support ...        http://www.egenix.com/
 >>> mxODBC.Zope.Database.Adapter ...             http://zope.egenix.com/
 >>> mxODBC, mxDateTime, mxTextTools ...        http://python.egenix.com/
________________________________________________________________________
2008-07-07: EuroPython 2008, Vilnius, Lithuania            25 days to go

:::: Try mxODBC.Zope.DA for Windows,Linux,Solaris,MacOSX for free ! ::::

eGenix.com Software, Skills and Services GmbH  Pastor-Loeh-Str.48
 D-40764 Langenfeld, Germany. CEO Dipl.-Math. Marc-Andre Lemburg
        Registered at Amtsgericht Duesseldorf: HRB 46611

[malemburg]

On 2008-06-11 23:27, M.-A. Lemburg wrote:

On 2008-06-11 20:38, Antoine Pitrou wrote:
> Antoine Pitrou <pitrou@free.fr> added the comment:
>
> Here is a new patch against SVN trunk. Nothing changed, except that I
> updated pybench to test keyword arguments as well.
>
> Added file: http://bugs.python.org/file10590/namedparam2.patch

When changing parameters or other aspects of pybench tests, you *have*
to update the version number of the test as well. Otherwise, pybench
would compare apples with oranges.

BTW: It would probably be better to add a completely new test
PythonNamedParameterCalls or something along those lines instead
of changing an existing test.

[rhettinger]

Attaching a version that's a little faster and cleaner with
PySequence_Fast_ITEMS.

[pitrou]

And here is a patch adding a new test in pybench as suggested by
Marc-Andre Lemburg.

[rhettinger]

Georg, do you want to go ahead and apply this.

[pitrou]

Committed in r65240 (new pybench test) and r65241 (speedup patch).