dict benchmark

bench_dict.py

#!/usr/bin/env python3
import pyperf


def build_dict_str(n):
    mydict = {str(k): k for k in range(n)}
    return mydict, list(mydict)


def build_dict_int(n):
    mydict = {k: k for k in range(n)}
    return mydict, list(mydict)


def bench_dict(loops, mydict, keys):
    range_it = range(loops)
    t0 = pyperf.perf_counter()

    for loops in range_it:
        for k in keys:
            mydict[k]
            mydict[k]
            mydict[k]
            mydict[k]
            mydict[k]
            mydict[k]
            mydict[k]
            mydict[k]
            mydict[k]
            mydict[k]

    return pyperf.perf_counter() - t0



runner = pyperf.Runner()

# inner-loops: dict[str] is duplicated 10 times
runner.bench_time_func('dict[str, 10]',     bench_dict, *build_dict_str(10), inner_loops=10*10)
runner.bench_time_func('dict[str, 1,000]',   bench_dict, *build_dict_str(1000), inner_loops=10*1000)
runner.bench_time_func('dict[str, 1,000,000]',   bench_dict, *build_dict_str(1000_000), inner_loops=10*1000_000)

runner.bench_time_func('dict[int, 10]',   bench_dict, *build_dict_int(10), inner_loops=10*10)
runner.bench_time_func('dict[int, 1,000]', bench_dict, *build_dict_int(1000), inner_loops=10*1000)
runner.bench_time_func('dict[int, 1,000,000]', bench_dict, *build_dict_int(1000_000), inner_loops=10*1000_000)

Swisstable hash: methane/cpython#41

Benchmark	main	swiss
dict[str, 10]	20.8 ns	23.2 ns: 1.11x slower
dict[str, 1,000]	20.6 ns	22.5 ns: 1.09x slower
dict[str, 1,000,000]	30.2 ns	27.9 ns: 1.08x faster
dict[int, 10]	20.2 ns	23.6 ns: 1.16x slower
dict[int, 1,000]	19.7 ns	23.4 ns: 1.18x slower
dict[int, 1,000,000]	21.0 ns	24.5 ns: 1.17x slower
Geometric mean	(ref)	1.10x slower