QFutureInterface: insert x86 PAUSE in tight CMPXCHG loop

From the Intel Software Optimization Manual[1], Chapter 11 ("Multi-core
and Hyper-Threading Technology"), offers:

User/Source Coding Rule 14. (M impact, H generality) Insert the PAUSE
instruction in fast spin loops and keep the number of loop repetitions
to a minimum to improve overall system performance.

See section 11.4.2 for an explanation of why this is a good idea.

[1] https://www.intel.com/content/www/us/en/developer/articles/technical/intel-sdm.html#inpage-nav-5

Pick-to: 6.3
Change-Id: I7fb65b80b7844c8d8f26fffd16e94088dad1ceee
Reviewed-by: Sona Kurazyan <sona.kurazyan@qt.io>

1 files changed, 11 insertions, 2 deletions

diff --git a/src/corelib/thread/qfutureinterface.cpp b/src/corelib/thread/qfutureinterface.cpp
index 71d6670538..a54caa878b 100644
--- a/src/corelib/thread/qfutureinterface.cpp
+++ b/src/corelib/thread/qfutureinterface.cpp
@@ -45,6 +45,10 @@
 #include <QtCore/qthread.h>
 #include <private/qthreadpool_p.h>
 
+#ifdef Q_PROCESSOR_X86
+#  include <immintrin.h>        // for _mm_pause()
+#endif
+
 #ifdef interface
 #  undef interface
 #endif
@@ -106,9 +110,14 @@ static inline int switch_from_to(QAtomicInt &a, int from, int to)
 {
     int newValue;
     int expected = a.loadRelaxed();
-    do {
+    for (;;) {
         newValue = (expected & ~from) | to;
-    } while (!a.testAndSetRelaxed(expected, newValue, expected));
+        if (a.testAndSetRelaxed(expected, newValue, expected))
+            break;
+#ifdef Q_PROCESSOR_X86
+        _mm_pause();
+#endif
+    }
     return newValue;
 }