timeit: use powers of 2 in autorange(), instead of powers of 10

[vstinner]

BPO	28469
Nosy	@vstinner, @serhiy-storchaka
PRs	[Do Not Merge] Convert Misc/NEWS so that it is managed by towncrier #552
Files	timeit_autorange_pow2.patch timeit_autorange_numbers.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 = None
closed_at = <Date 2016-10-23.12:19:09.850>
created_at = <Date 2016-10-18.16:13:11.315>
labels = ['3.7', 'library', 'performance']
title = 'timeit: use powers of 2 in autorange(), instead of powers of 10'
updated_at = <Date 2017-03-31.16:36:15.077>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2017-03-31.16:36:15.077>
actor = 'dstufft'
assignee = 'none'
closed = True
closed_date = <Date 2016-10-23.12:19:09.850>
closer = 'serhiy.storchaka'
components = ['Library (Lib)']
creation = <Date 2016-10-18.16:13:11.315>
creator = 'vstinner'
dependencies = []
files = ['45134', '45186']
hgrepos = []
issue_num = 28469
keywords = ['patch']
message_count = 8.0
messages = ['278899', '278905', '278906', '278915', '278916', '279187', '279245', '279254']
nosy_count = 3.0
nosy_names = ['vstinner', 'python-dev', 'serhiy.storchaka']
pr_nums = ['552']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'performance'
url = 'https://bugs.python.org/issue28469'
versions = ['Python 3.7']

[vstinner]

In the issue bpo-28240, I changed the default repeat from 3 to 5 in timeit. Serhiy wrote (msg277157): "For now default timeit run takes from 0.8 to 8 sec. Adding yet 5 sec makes a user more angry."

I propose to use powers to 2, instead of powers of 10, in timeit autorange to reduce the total duration of the microbenchmark.

• Without the patch, the worst case: 4.0 * 6 = 24 seconds.

• With the patch, the worst case is: 0.4 * 6 = 2.4 seconds

• 6: autorange (1 iteration) + 5 repeatition

• autorange uses a minimum of 200 ms, so the worst case is 200 ms * 10 before, or 200 ms * 2 after

[serhiy-storchaka]

"131072 loops" in a report looks not human friendly. I would prefer using a sequence of round numbers: 1, 2, 5, 10, 20, 50, 100, ...

[vstinner]

"131072 loops" in a report looks not human friendly.

In my perf module, I use 2^n syntax for numbers larger than 1024. Syntax accepted in command line arguments.

[serhiy-storchaka]

The perf module doesn't report the number of loops by default.

[vstinner]

The perf module doesn't report the number of loops by default.

Right, it's hidden by default. I consider that it's not an useful information for end users.

If you want more details (see the number of inner and outer loops), use --verbose ("python3 -m perf timeit -v ...") or the stats commands ("python3 -m perf stats file.json").

[serhiy-storchaka]

Following patch takes numbers from the sequence 1, 2, 5, 10, 20, 50, ... It also updates the documentation and tests.

[vstinner]

timeit_autorange_numbers.patch: LGTM.

[python-dev]

New changeset 8e6cc952adc6 by Serhiy Storchaka in branch 'default':
Issue bpo-28469: timeit now uses the sequence 1, 2, 5, 10, 20, 50,... instead
https://hg.python.org/cpython/rev/8e6cc952adc6