Messages in this thread Patch in this message | | | Date | Thu, 15 Nov 2018 15:17:45 +0100 | From | Jiri Olsa <> | Subject | Re: [RFC 2/2] perf stat: Fix shadow stats for clock events |
| |
On Thu, Nov 15, 2018 at 03:25:33PM +0530, Ravi Bangoria wrote: > Commit 0aa802a79469 ("perf stat: Get rid of extra clock display > function") introduced scale and unit for clock events. Thus, > perf_stat__update_shadow_stats() now saves scaled values of > clock events in msecs, instead of original nsecs. But while > calculating values of shadow stats we still consider clock > event values in nsecs. This results in a wrong shadow stat > values. Ex, > > # ./perf stat -e task-clock,cycles ls > <SNIP> > 2.62 msec task-clock:u # 0.624 CPUs utilized > 2,501,536 cycles:u # 1250768.000 GHz > > Fix this by considering clock events's saved stats in msecs: > > # ./perf stat -e task-clock,cycles ls > <SNIP> > 2.42 msec task-clock:u # 0.754 CPUs utilized > 2,338,747 cycles:u # 1.169 GHz > > Note: > The problem with this approach is, we are losing fractional part > while converting nsecs to msecs. This results in a sightly different > values of shadow stats.
yea, could we just leave the NSEC instead? like below
jirka
--- diff --git a/tools/perf/util/stat-shadow.c b/tools/perf/util/stat-shadow.c index f4bad808bdd9..da8857df238e 100644 --- a/tools/perf/util/stat-shadow.c +++ b/tools/perf/util/stat-shadow.c @@ -209,11 +209,12 @@ void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 count, int cpu, struct runtime_stat *st) { int ctx = evsel_context(counter); + u64 count_ns = count; count *= counter->scale; if (perf_evsel__is_clock(counter)) - update_runtime_stat(st, STAT_NSECS, 0, cpu, count); + update_runtime_stat(st, STAT_NSECS, 0, cpu, count_ns); else if (perf_evsel__match(counter, HARDWARE, HW_CPU_CYCLES)) update_runtime_stat(st, STAT_CYCLES, ctx, cpu, count); else if (perf_stat_evsel__is(counter, CYCLES_IN_TX))
| |