[cfe-commits] [libcxx] r160967 - /libcxx/trunk/src/memory.cpp

[Howard Hinnant]

Author: hhinnant
Date: Mon Jul 30 12:13:21 2012
New Revision: 160967

URL: http://llvm.org/viewvc/llvm-project?rev=160967&view=rev
Log:
Despite my pathological distrust of spin locks, the number just don't lie.  I've put a small spin in __sp_mut::lock() on std::mutex::try_lock(), which is testing quite well.  In my experience, putting in a yield for every failed iteration is also a major performance booster.  This change makes one of the performance tests I was using (a highly contended one) run about 20 times faster.

Modified:
    libcxx/trunk/src/memory.cpp

Modified: libcxx/trunk/src/memory.cpp
URL: http://llvm.org/viewvc/llvm-project/libcxx/trunk/src/memory.cpp?rev=160967&r1=160966&r2=160967&view=diff
==============================================================================
--- libcxx/trunk/src/memory.cpp (original)
+++ libcxx/trunk/src/memory.cpp Mon Jul 30 12:13:21 2012
@@ -10,6 +10,7 @@
 #define _LIBCPP_BUILDING_MEMORY
 #include "memory"
 #include "mutex"
+#include "thread"
 
 _LIBCPP_BEGIN_NAMESPACE_STD
 
@@ -129,13 +130,23 @@
 void
 __sp_mut::lock() _NOEXCEPT
 {
-    reinterpret_cast<mutex*>(_)->lock();
+    mutex& m = *static_cast<mutex*>(_);
+    unsigned count = 0;
+    while (!m.try_lock())
+    {
+        if (++count > 16)
+        {
+            m.lock();
+            break;
+        }
+        this_thread::yield();
+    }
 }
 
 void
 __sp_mut::unlock() _NOEXCEPT
 {
-    reinterpret_cast<mutex*>(_)->unlock();
+    static_cast<mutex*>(_)->unlock();
 }
 
 __sp_mut&