mollyim/webrtc
1
0
Fork 0
mirror of https://github.com/mollyim/webrtc.git synced 2025-05-12 21:30:45 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions

Switch back to native mutexes on macOS

Browse source 
It seems native mutex performance has improved considerably on Mac
lately, primarily by switching to a different default scheduling
policy. For safety, set this policy explicitly.

The special implementation previously used on Mac is still faster but
suffers a problem when used on realtime audio threads, where they will
not get rescheduled as quickly as when using native mutexes.

Bug: webrtc:10373
Change-Id: Iabf97afc5c2609096331bba0199f433fd26b68b2
Reviewed-on: https://webrtc-review.googlesource.com/c/125186
Commit-Queue: Oskar Sundbom <ossu@webrtc.org>
Reviewed-by: Tommi <tommi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#26948}
This commit is contained in:
Oskar Sundbom 2019-03-01 16:11:49 +01:00 committed by Commit Bot
parent b678940d3a
commit 13471a44b6
4 changed files with 28 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
rtc_base/critical_section.cc
View file

@ -34,6 +34,10 @@ CriticalSection::CriticalSection() {
pthread_mutexattr_t mutex_attribute;
pthread_mutexattr_init(&mutex_attribute);
pthread_mutexattr_settype(&mutex_attribute, PTHREAD_MUTEX_RECURSIVE);
#if defined(WEBRTC_MAC)
pthread_mutexattr_setpolicy_np(&mutex_attribute,
_PTHREAD_MUTEX_POLICY_FAIRSHARE);
#endif
pthread_mutex_init(&mutex_, &mutex_attribute);
pthread_mutexattr_destroy(&mutex_attribute);
#endif

2
rtc_base/critical_section.h
View file

@ -34,7 +34,7 @@
#endif
// See notes in the 'Performance' unit test for the effects of this flag.
#define USE_NATIVE_MUTEX_ON_MAC 0
#define USE_NATIVE_MUTEX_ON_MAC 1
#if defined(WEBRTC_MAC) && !USE_NATIVE_MUTEX_ON_MAC
#include <dispatch/dispatch.h>

32
rtc_base/critical_section_unittest.cc
View file

@ -372,25 +372,33 @@ class PerfTestThread {
int my_id_ = 0;
};
// Comparison of output of this test as tested on a MacBook Pro Retina, 15-inch,
// Mid 2014, 2,8 GHz Intel Core i7, 16 GB 1600 MHz DDR3,
// running OS X El Capitan, 10.11.2.
// Comparison of output of this test as tested on a MacBook Pro, 13-inch,
// 2017, 3,5 GHz Intel Core i7, 16 GB 2133 MHz LPDDR3,
// running macOS Mojave, 10.14.3.
//
// Native mutex implementation:
// Native mutex implementation using fair policy (previously macOS default):
// Approximate CPU usage:
// System: ~16%
// User mode: ~1.3%
// Idle: ~82%
// real 4m54.612s
// user 1m20.575s
// sys 3m48.872s
// Unit test output:
// [ OK ] CriticalSectionTest.Performance (234545 ms)
// [ OK ] CriticalSectionTest.Performance (294375 ms)
//
// Native mutex implementation using first fit policy (current macOS default):
// Approximate CPU usage:
// real 0m11.535s
// user 0m12.738s
// sys 0m31.207s
// Unit test output:
// [ OK ] CriticalSectionTest.Performance (11444 ms)
//
// Special partially spin lock based implementation:
// Approximate CPU usage:
// System: ~75%
// User mode: ~16%
// Idle: ~8%
// real 0m2.113s
// user 0m3.014s
// sys 0m4.495s
// Unit test output:
// [ OK ] CriticalSectionTest.Performance (2107 ms)
// [ OK ] CriticalSectionTest.Performance (1885 ms)
//
// The test is disabled by default to avoid unecessarily loading the bots.
TEST(CriticalSectionTest, DISABLED_Performance) {

3
rtc_base/platform_thread_types.h
View file

@ -25,6 +25,9 @@
#elif defined(WEBRTC_POSIX)
#include <pthread.h>
#include <unistd.h>
#if defined(WEBRTC_MAC)
#include <pthread_spis.h>
#endif
#endif
// clang-format on