mirror of
https://github.com/python/cpython.git
synced 2024-11-24 08:52:25 +01:00
0c89056fe5
PyMutex is a one byte lock with fast, inlineable lock and unlock functions for the common uncontended case. The design is based on WebKit's WTF::Lock. PyMutex is built using the _PyParkingLot APIs, which provides a cross-platform futex-like API (based on WebKit's WTF::ParkingLot). This internal API will be used for building other synchronization primitives used to implement PEP 703, such as one-time initialization and events. This also includes tests and a mini benchmark in Tools/lockbench/lockbench.py to compare with the existing PyThread_type_lock. Uncontended acquisition + release: * Linux (x86-64): PyMutex: 11 ns, PyThread_type_lock: 44 ns * macOS (arm64): PyMutex: 13 ns, PyThread_type_lock: 18 ns * Windows (x86-64): PyMutex: 13 ns, PyThread_type_lock: 38 ns PR Overview: The primary purpose of this PR is to implement PyMutex, but there are a number of support pieces (described below). * PyMutex: A 1-byte lock that doesn't require memory allocation to initialize and is generally faster than the existing PyThread_type_lock. The API is internal only for now. * _PyParking_Lot: A futex-like API based on the API of the same name in WebKit. Used to implement PyMutex. * _PyRawMutex: A word sized lock used to implement _PyParking_Lot. * PyEvent: A one time event. This was used a bunch in the "nogil" fork and is useful for testing the PyMutex implementation, so I've included it as part of the PR. * pycore_llist.h: Defines common operations on doubly-linked list. Not strictly necessary (could do the list operations manually), but they come up frequently in the "nogil" fork. ( Similar to https://man.freebsd.org/cgi/man.cgi?queue) --------- Co-authored-by: Eric Snow <ericsnowcurrently@gmail.com>
54 lines
2.0 KiB
Python
54 lines
2.0 KiB
Python
# Measure the performance of PyMutex and PyThread_type_lock locks
|
|
# with short critical sections.
|
|
#
|
|
# Usage: python Tools/lockbench/lockbench.py [CRITICAL_SECTION_LENGTH]
|
|
#
|
|
# How to interpret the results:
|
|
#
|
|
# Acquisitions (kHz): Reports the total number of lock acquisitions in
|
|
# thousands of acquisitions per second. This is the most important metric,
|
|
# particularly for the 1 thread case because even in multithreaded programs,
|
|
# most locks acquisitions are not contended. Values for 2+ threads are
|
|
# only meaningful for `--disable-gil` builds, because the GIL prevents most
|
|
# situations where there is lock contention with short critical sections.
|
|
#
|
|
# Fairness: A measure of how evenly the lock acquisitions are distributed.
|
|
# A fairness of 1.0 means that all threads acquired the lock the same number
|
|
# of times. A fairness of 1/N means that only one thread ever acquired the
|
|
# lock.
|
|
# See https://en.wikipedia.org/wiki/Fairness_measure#Jain's_fairness_index
|
|
|
|
from _testinternalcapi import benchmark_locks
|
|
import sys
|
|
|
|
# Max number of threads to test
|
|
MAX_THREADS = 10
|
|
|
|
# How much "work" to do while holding the lock
|
|
CRITICAL_SECTION_LENGTH = 1
|
|
|
|
|
|
def jains_fairness(values):
|
|
# Jain's fairness index
|
|
# See https://en.wikipedia.org/wiki/Fairness_measure
|
|
return (sum(values) ** 2) / (len(values) * sum(x ** 2 for x in values))
|
|
|
|
def main():
|
|
print("Lock Type Threads Acquisitions (kHz) Fairness")
|
|
for lock_type in ["PyMutex", "PyThread_type_lock"]:
|
|
use_pymutex = (lock_type == "PyMutex")
|
|
for num_threads in range(1, MAX_THREADS + 1):
|
|
acquisitions, thread_iters = benchmark_locks(
|
|
num_threads, use_pymutex, CRITICAL_SECTION_LENGTH)
|
|
|
|
acquisitions /= 1000 # report in kHz for readability
|
|
fairness = jains_fairness(thread_iters)
|
|
|
|
print(f"{lock_type: <20}{num_threads: <18}{acquisitions: >5.0f}{fairness: >20.2f}")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
if len(sys.argv) > 1:
|
|
CRITICAL_SECTION_LENGTH = int(sys.argv[1])
|
|
main()
|