patch 9.0.1411: accuracy of profiling is not optimal
Problem: Accuracy of profiling is not optimal.
Solution: Use CLOCK_MONOTONIC if possible. (Ernie Rael, closes #12129)
diff --git a/runtime/doc/repeat.txt b/runtime/doc/repeat.txt
index c8a0f5d..054b870 100644
--- a/runtime/doc/repeat.txt
+++ b/runtime/doc/repeat.txt
@@ -1148,9 +1148,10 @@
Profiling should give a good indication of where time is spent, but keep in
mind there are various things that may clobber the results:
-- The accuracy of the time measured depends on the gettimeofday() system
- function. It may only be as accurate as 1/100 second, even though the times
- are displayed in micro seconds.
+- The accuracy of the time measured depends on the gettimeofday(), or
+ clock_gettime if available, system function. The accuracy ranges from 1/100
+ second to nano seconds. With clock_gettime the times are displayed in nano
+ seconds, otherwise micro seconds. You can use `has("prof_nsec")`.
- Real elapsed time is measured, if other processes are busy they may cause
delays at unpredictable moments. You may want to run the profiling several