Benchmarking list item access in free-threading Python

bench.py

from concurrent.futures import ThreadPoolExecutor
import time
import itertools
from collections.abc import Sequence


def get_items(values: Sequence, duration: float):
  deadline = time.time() + duration
  n = len(values)
  for i in itertools.count():
    _ = values[i % n]
    if i % 100 == 0 and time.time() > deadline:
      return i


def decimal_eng(n: float):
  if n > 1e9:
    return f'{n / 1e9:.1f}G'
  elif n > 1e6:
    return f'{n / 1e6:.1f}M'
  elif n > 1e3:
    return f'{n / 1e3:.1f}k'
  else:
    return f'{n:1.f}'


def measure_all():
  duration = 1
  num_threads = [1, 2, 4, 8]
  length = 100

  values = [
      lambda i: chr(ord('a') + i) * length,
      lambda i: chr(ord('a') + i).encode() * length,
      lambda i: bytearray([i] * length),
      lambda i: [i] * length,
      lambda i: (i,) * length,
      lambda i: {k: i for k in range(length)},
  ]

  for make_value in values:
    print(f'# {type(make_value(0))}')
    print(f' ' * 11, '  shared ', '  separate')
    for threads in num_threads:
      with ThreadPoolExecutor(threads) as pool:
        shared_value = make_value(0)
        num_items_shared = sum(
            pool.map(lambda _: get_items(shared_value, duration), range(threads)))
        num_items_sep = sum(
            pool.map(lambda i: get_items(make_value(i), duration), range(threads)))

        print(f'{threads:2d} threads: '
              f'{decimal_eng(num_items_shared / duration):>6}/sec',
              f'{decimal_eng(num_items_sep / duration):>6}/sec')


if __name__ == '__main__':
  measure_all()

results_gil.txt

$ uv run --python 3.14.0a6 concurrent_reads.py
# <class 'str'>
              shared    separate
 1 threads:  19.2M/sec  19.1M/sec
 2 threads:  19.2M/sec  19.3M/sec
 4 threads:  20.1M/sec  19.6M/sec
 8 threads:  23.9M/sec  19.8M/sec
# <class 'bytes'>
              shared    separate
 1 threads:  16.7M/sec  16.6M/sec
 2 threads:  16.6M/sec  16.7M/sec
 4 threads:  17.2M/sec  16.8M/sec
 8 threads:  19.9M/sec  17.6M/sec
# <class 'bytearray'>
              shared    separate
 1 threads:  15.7M/sec  15.7M/sec
 2 threads:  15.9M/sec  15.3M/sec
 4 threads:  17.0M/sec  16.2M/sec
 8 threads:  19.2M/sec  16.4M/sec
# <class 'list'>
              shared    separate
 1 threads:  17.7M/sec  17.6M/sec
 2 threads:  17.7M/sec  17.7M/sec
 4 threads:  19.0M/sec  18.0M/sec
 8 threads:  20.4M/sec  18.9M/sec
# <class 'tuple'>
              shared    separate
 1 threads:  18.0M/sec  18.0M/sec
 2 threads:  18.2M/sec  18.1M/sec
 4 threads:  19.1M/sec  18.6M/sec
 8 threads:  23.4M/sec  19.4M/sec
 # <class 'dict'>
              shared    separate
 1 threads:  24.4M/sec  24.5M/sec
 2 threads:  24.8M/sec  24.8M/sec
 4 threads:  24.0M/sec  24.9M/sec
 8 threads:  28.4M/sec  26.3M/sec

results_no_gil_a6.txt

$ uv run --python 3.14.0a6+freethreaded concurrent_reads.py
# <class 'bytearray'>
              shared    separate
 1 threads:  15.8M/sec  16.1M/sec
 2 threads:   9.7M/sec  29.4M/sec
 4 threads:   7.9M/sec  53.4M/sec
 8 threads:   5.9M/sec  73.1M/sec
# <class 'str'>
              shared    separate
 1 threads:  17.9M/sec  18.7M/sec
 2 threads:  10.3M/sec  35.8M/sec
 4 threads:   7.2M/sec  61.5M/sec
 8 threads:   3.1M/sec  91.1M/sec
# <class 'bytes'>
              shared    separate
 1 threads:  17.4M/sec  17.6M/sec
 2 threads:   9.4M/sec  35.4M/sec
 4 threads:   7.2M/sec  65.9M/sec
 8 threads:   2.4M/sec  83.4M/sec
# <class 'list'>
              shared    separate
 1 threads:  17.7M/sec  18.7M/sec
 2 threads:   8.4M/sec  37.8M/sec
 4 threads:   3.8M/sec  76.0M/sec
 8 threads:   2.6M/sec 106.2M/sec
# <class 'tuple'>
              shared    separate
 1 threads:  18.9M/sec  19.0M/sec
 2 threads:   9.1M/sec  38.2M/sec
 4 threads:   3.8M/sec  76.8M/sec
 8 threads:   3.0M/sec 100.0M/sec
 # <class 'dict'>
              shared    separate
 1 threads:  24.5M/sec  25.2M/sec
 2 threads:  11.2M/sec  49.7M/sec
 4 threads:   5.7M/sec  91.2M/sec
 8 threads:   2.5M/sec 123.3M/sec

results_no_gil_a7.txt

$ /Library/Frameworks/PythonT.framework/Versions/3.14/bin/python3t concurrent_reads.py
# <class 'str'>
              shared    separate
 1 threads:  16.4M/sec  16.3M/sec
 2 threads:  33.0M/sec  32.5M/sec
 4 threads:  52.9M/sec  54.5M/sec
 8 threads:  91.6M/sec  85.7M/sec
# <class 'bytes'>
              shared    separate
 1 threads:  16.5M/sec  16.4M/sec
 2 threads:  33.4M/sec  33.0M/sec
 4 threads:  53.5M/sec  55.0M/sec
 8 threads:  78.6M/sec  82.1M/sec
# <class 'bytearray'>
              shared    separate
 1 threads:  15.0M/sec  14.8M/sec
 2 threads:  12.4M/sec  30.1M/sec
 4 threads:  10.8M/sec  50.0M/sec
 8 threads:   7.6M/sec  78.2M/sec
# <class 'list'>
              shared    separate
 1 threads:  17.1M/sec  16.6M/sec
 2 threads:  30.2M/sec  29.1M/sec
 4 threads:  59.2M/sec  57.0M/sec
 8 threads:  79.3M/sec  90.5M/sec
# <class 'tuple'>
              shared    separate
 1 threads:  17.9M/sec  17.9M/sec
 2 threads:  35.3M/sec  35.9M/sec
 4 threads:  58.5M/sec  61.3M/sec
 8 threads:  92.5M/sec  82.4M/sec
# <class 'dict'>
              shared    separate
 1 threads:  21.4M/sec  21.4M/sec
 2 threads:  40.9M/sec  41.3M/sec
 4 threads:  65.7M/sec  68.6M/sec
 8 threads: 107.8M/sec 133.4M/sec