@ -5091,8 +5091,24 @@ hwc_usage_internal (int forKernel, FILE *f_usage, const char *cmd, const char *d
fprintf ( f_usage , GTXT ( " `lo' per-thread maximum rate of ~10 samples/second \n " ) ) ;
fprintf ( f_usage , GTXT ( " `on' per-thread maximum rate of ~100 samples/second \n " ) ) ;
fprintf ( f_usage , GTXT ( " `hi' per-thread maximum rate of ~1000 samples/second \n \n " ) ) ;
fprintf ( f_usage , GTXT ( " <rate> == <interval> \n " ) ) ;
fprintf ( f_usage , GTXT ( " event interval; see collect (1) for details \n \n " ) ) ;
fprintf ( f_usage , GTXT ( " <rate> == <interval> \n "
" Fixed event interval value to trigger a sample. \n "
" Smaller intervals imply more frequent samples. \n "
" Example: when counting cycles on a 2 GHz processor, \n "
" an interval of 2,000,003 implies ~1000 samples/sec \n "
" \n "
" Use this feature with caution, because: \n "
" (1) Frequent sampling increases overhead and may disturb \n "
" other applications on your system. \n "
" (2) Event counts vary dramatically depending on the event \n "
" and depending on the application. \n "
" (3) A fixed event interval disables any other gprofng \n "
" internal mechanisms that may limit event rates. \n "
" \n "
" Guidelines: Aim at <1000 events per second. Start by \n "
" collecting with the 'hi' option; in the experiment overview, \n "
" notice how many events are recorded per second; divide by \n "
" 1000, and use that as your starting point. \n \n " ) ) ;
fprintf ( f_usage , GTXT ( " A comma ',' followed immediately by white space may be omitted. \n \n " ) ) ;
}