On Thu, 1 Feb 2024, Marcelo Tosatti wrote:
>
> With buckets of size 1us, accounting for measurements in the
> [1ns, 999ns] range are done to the 2us bucket (while they
> should be accounted in the 1us bucket):
>
> 001 (us): 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
> 002 (us): 41916187 41937649 41938461 42029055 ...
> 003 (us): 969 985 958 972 964 986 970 961 973 ...
>
> Fix this by doing a plain cycles -> nanoseconds convertion:
>
> 001 (us): 43287555 43086678 43087427 43109974 ...
> 002 (us): 983 987 985 975 982 960 993 961 992 ...
> 003 (us): 9 6 7 13 9 22 3 21 3 3 8 8 10 11 3 55
>
>
> Signed-off-by: Marcelo Tosatti <mtosatti@xxxxxxxxxx>
> Reported-by: Chuck Newman <chuck.newman@xxxxxxx>
>
> diff --git a/src/oslat/oslat.c b/src/oslat/oslat.c
> index 4bdca64..13e887d 100644
> --- a/src/oslat/oslat.c
> +++ b/src/oslat/oslat.c
> @@ -334,7 +334,7 @@ static void insert_bucket(struct thread *t, stamp_t value)
> uint64_t extra;
> double us;
>
> - lat = (value * g.unit_per_us + t->counter_mhz - 1) / t->counter_mhz;
> + lat = (value * g.unit_per_us) / t->counter_mhz;
> us = (double)lat / g.unit_per_us;
> if (!g.preheat && g.trace_threshold && us >= g.trace_threshold) {
> char *line = "%s: Trace threshold (%d us) triggered on cpu %d with %.*f us!\n";
>
>
>
Signed-off-by: John Kacur <jkacur@xxxxxxxxxx>
