[New-bugs-announce] [issue46235] Do all ref-counting at once for sequence multiplication

[Dennis Sweeney]

New submission from Dennis Sweeney <sweeney.dennis650 at gmail.com>:

Some benchmarks for this change are below. The case with the largest speedup, `[None] * 10000`, is the case that I would consider the most important: initializing counters of the form `[0] * N` is very common in my experience.

The following were taken from a PGO build on Windows without CPU isolation, so take them with a grain of salt:

Slower (16):
- [None] * 2: 61.6 ns +- 3.3 ns -> 71.0 ns +- 1.5 ns: 1.15x slower
- list(range(10)) * 10000: 205 us +- 49 us -> 232 us +- 93 us: 1.13x slower
- ["Python", "Perl"] * 10000: 31.4 us +- 4.3 us -> 34.4 us +- 0.2 us: 1.10x slower
- list(range(100)) * 2: 370 ns +- 12 ns -> 395 ns +- 14 ns: 1.07x slower
- list(range(10)) * 2: 81.5 ns +- 2.8 ns -> 86.9 ns +- 2.4 ns: 1.07x slower
- [None] * 10: 72.3 ns +- 2.2 ns -> 75.8 ns +- 2.1 ns: 1.05x slower
- ["Python", "Perl"] * 100: 442 ns +- 67 ns -> 463 ns +- 4 ns: 1.05x slower
- tuple(range(10)) * 2: 88.2 ns +- 2.9 ns -> 91.9 ns +- 2.7 ns: 1.04x slower
- ("Python", "Perl") * 10: 92.6 ns +- 7.8 ns -> 96.0 ns +- 4.5 ns: 1.04x slower
- ["Python", "Perl"] * 10: 86.1 ns +- 4.2 ns -> 89.1 ns +- 2.7 ns: 1.04x slower
- (None,) * 10: 69.4 ns +- 1.9 ns -> 71.2 ns +- 3.1 ns: 1.03x slower
- ["Python", "Perl"] * 2: 63.9 ns +- 2.6 ns -> 65.3 ns +- 1.4 ns: 1.02x slower
- (None,) * 2: 54.0 ns +- 0.7 ns -> 55.1 ns +- 2.1 ns: 1.02x slower
- ("Python", "Perl") * 2: 57.5 ns +- 3.0 ns -> 58.6 ns +- 2.4 ns: 1.02x slower
- list(range(10)) * 10: 238 ns +- 8 ns -> 242 ns +- 7 ns: 1.02x slower
- tuple(range(100)) * 2: 414 ns +- 19 ns -> 420 ns +- 6 ns: 1.02x slower

Faster (22):
- [None] * 10000: 25.1 us +- 0.2 us -> 12.8 us +- 0.2 us: 1.97x faster
- tuple(range(10)) * 10000: 255 us +- 91 us -> 176 us +- 77 us: 1.45x faster
- tuple(range(1000)) * 10: 17.8 us +- 0.3 us -> 12.4 us +- 0.1 us: 1.44x faster
- [None] * 100: 355 ns +- 11 ns -> 251 ns +- 7 ns: 1.41x faster
- list(range(1000)) * 10: 16.8 us +- 0.2 us -> 12.5 us +- 0.1 us: 1.34x faster
- tuple(range(100)) * 100: 15.9 us +- 0.2 us -> 12.7 us +- 0.2 us: 1.26x faster
- tuple(range(10)) * 100: 1.63 us +- 0.02 us -> 1.33 us +- 0.01 us: 1.23x faster
- tuple(range(1000)) * 2: 3.61 us +- 0.06 us -> 2.93 us +- 0.03 us: 1.23x faster
- list(range(100)) * 100: 13.8 us +- 0.3 us -> 11.2 us +- 0.1 us: 1.23x faster
- tuple(range(100)) * 10: 1.68 us +- 0.02 us -> 1.40 us +- 0.01 us: 1.20x faster
- tuple(range(1000)) * 100: 240 us +- 87 us -> 200 us +- 102 us: 1.20x faster
- tuple(range(1000)) * 10000: 33.8 ms +- 0.8 ms -> 28.3 ms +- 1.5 ms: 1.19x faster
- list(range(1000)) * 10000: 33.9 ms +- 0.9 ms -> 28.6 ms +- 1.4 ms: 1.19x faster
- list(range(100)) * 10: 1.47 us +- 0.10 us -> 1.25 us +- 0.01 us: 1.18x faster
- list(range(10)) * 100: 1.41 us +- 0.02 us -> 1.22 us +- 0.02 us: 1.16x faster
- list(range(100)) * 10000: 3.15 ms +- 0.13 ms -> 2.79 ms +- 0.13 ms: 1.13x faster
- list(range(1000)) * 2: 3.32 us +- 0.07 us -> 2.94 us +- 0.04 us: 1.13x faster
- tuple(range(100)) * 10000: 3.24 ms +- 0.13 ms -> 2.88 ms +- 0.14 ms: 1.13x faster
- (None,) * 10000: 26.7 us +- 0.2 us -> 24.1 us +- 0.2 us: 1.11x faster
- tuple(range(10)) * 10: 258 ns +- 7 ns -> 249 ns +- 6 ns: 1.04x faster
- ("Python", "Perl") * 100: 438 ns +- 7 ns -> 436 ns +- 7 ns: 1.01x faster
- ("Python", "Perl") * 10000: 31.1 us +- 0.3 us -> 30.9 us +- 0.2 us: 1.01x faster

Benchmark hidden because not significant (2): list(range(1000)) * 100, (None,) * 100

Geometric mean: 1.10x faster

----------
components: Interpreter Core
messages: 409546
nosy: Dennis Sweeney
priority: normal
severity: normal
status: open
title: Do all ref-counting at once for sequence multiplication
type: performance
versions: Python 3.11

_______________________________________
Python tracker <report at bugs.python.org>
<https://bugs.python.org/issue46235>
_______________________________________