Minor improvement of KMP_YIELD_NOW() macro.

[hbae]

Summary

This change slightly improves performance of KMP_YIELD_NOW() macro, by using _rdtsc() intrinsic function if possible.

[jlpeyton]

LGTM

openmp/trunk/runtime/src/kmp.h

 
 /* Calculate number of intervals for a specific block time based on monitor_wakeups */
 #define KMP_INTERVALS_FROM_BLOCKTIME(blocktime, monitor_wakeups)  \
                                  ( ( (blocktime) + (KMP_BLOCKTIME_MULTIPLIER / (monitor_wakeups)) - 1 ) /  \
                                    (KMP_BLOCKTIME_MULTIPLIER / (monitor_wakeups)) )
 #else
 # if KMP_OS_UNIX && (KMP_ARCH_X86 || KMP_ARCH_X86_64)
    // HW TSC is used to reduce overhead (clock tick instead of nanosecond).
-   extern double __kmp_ticks_per_nsec;
-#  define KMP_NOW() __kmp_hardware_timestamp()
-#  define KMP_NOW_MSEC() ((kmp_uint64)(KMP_NOW()/__kmp_ticks_per_nsec)/KMP_USEC_PER_SEC)
-#  define KMP_BLOCKTIME_INTERVAL() (__kmp_dflt_blocktime * KMP_USEC_PER_SEC * __kmp_ticks_per_nsec)
+   extern kmp_uint64 __kmp_ticks_per_msec;
+#  if KMP_COMPILER_ICC
+#   define KMP_NOW() _rdtsc()
+#  else
+#   define KMP_NOW() __kmp_hardware_timestamp()
+#  endif
+#  define KMP_NOW_MSEC() (KMP_NOW()/__kmp_ticks_per_msec)
+#  define KMP_BLOCKTIME_INTERVAL() (__kmp_dflt_blocktime * __kmp_ticks_per_msec)
 #  define KMP_BLOCKING(goal, count) ((goal) > KMP_NOW())
 # else
    // System time is retrieved sporadically while blocking.
    extern kmp_uint64 __kmp_now_nsec();
 #  define KMP_NOW() __kmp_now_nsec()
 #  define KMP_NOW_MSEC() (KMP_NOW()/KMP_USEC_PER_SEC)
 #  define KMP_BLOCKTIME_INTERVAL() (__kmp_dflt_blocktime * KMP_USEC_PER_SEC)
 #  define KMP_BLOCKING(goal, count) ((count) % 1000 != 0 || (goal) > KMP_NOW())

openmp/trunk/runtime/src/z_Linux_util.cpp

 static int __kmp_fork_count = 0;
 
 static pthread_condattr_t  __kmp_suspend_cond_attr;
 static pthread_mutexattr_t __kmp_suspend_mutex_attr;
 
 static kmp_cond_align_t    __kmp_wait_cv;
 static kmp_mutex_align_t   __kmp_wait_mx;
 
-double __kmp_ticks_per_nsec;
+kmp_uint64 __kmp_ticks_per_msec = 1000000;
 
 /* ------------------------------------------------------------------------ */
 /* ------------------------------------------------------------------------ */
 
 #ifdef DEBUG_SUSPEND
 static void
 __kmp_print_cond( char *buffer, kmp_cond_align_t *cond )
 {

 __kmp_now_nsec()
 {
     struct timeval t;
     gettimeofday(&t, NULL);
     return KMP_NSEC_PER_SEC*t.tv_sec + 1000*t.tv_usec;
 }
 
 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
-/* Measure clock tick per nanosecond */
+/* Measure clock ticks per millisecond */
 void
 __kmp_initialize_system_tick()
 {
     kmp_uint64 delay = 100000; // 50~100 usec on most machines.
     kmp_uint64 nsec = __kmp_now_nsec();
     kmp_uint64 goal = __kmp_hardware_timestamp() + delay;
     kmp_uint64 now;
     while ((now = __kmp_hardware_timestamp()) < goal);
-    __kmp_ticks_per_nsec = 1.0 * (delay + (now - goal)) / (__kmp_now_nsec() - nsec);
+    __kmp_ticks_per_msec = (kmp_uint64)(1e6 * (delay + (now - goal)) / (__kmp_now_nsec() - nsec));
 }
 #endif
 
 /*
     Determine whether the given address is mapped into the current address space.
 */
 
 int