boostorg/smart_ptr
1
0
Fork 1
mirror of https://github.com/boostorg/smart_ptr.git synced 2025-07-30 12:47:28 +02:00
Code Issues Packages Projects Releases Wiki Activity

Use a relaxed load before test_and_set to not lock cache line on contention (AMD spinlock recommendation per <https://probablydance.com/2019/12/30/measuring-mutexes-spinlocks-and-how-bad-the-linux-scheduler-really-is/>)

Browse Source
This commit is contained in:
Peter Dimov
2020-06-07 17:59:55 +03:00
parent d0655ab145
commit 3db4ad9a15
1 changed files with 4 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
include/boost/smart_ptr/detail/spinlock_gcc_atomic.hpp
View File

@ -7,13 +7,9 @@
# pragma once
#endif
//
// Copyright (c) 2008 Peter Dimov
//
// Distributed under the Boost Software License, Version 1.0.
// See accompanying file LICENSE_1_0.txt or copy at
// http://www.boost.org/LICENSE_1_0.txt)
//
// Copyright 2008, 2020 Peter Dimov
// Distributed under the Boost Software License, Version 1.0.
// https://www.boost.org/LICENSE_1_0.txt
#include <boost/smart_ptr/detail/yield_k.hpp>
@ -40,8 +36,7 @@ public:
bool try_lock()
{
int r = __atomic_test_and_set( &v_, __ATOMIC_ACQUIRE );
return r == 0;
return __atomic_load_n( &v_, __ATOMIC_RELAXED ) == 0 && __atomic_test_and_set( &v_, __ATOMIC_ACQUIRE ) == 0;
}
void lock()