When -r option is used, perf stat runs the command multiple times and
update stats in the evsel->stats.res_stats for global aggregation. But
the value is never used and the value it prints at the end is just the
value from the last run. I think we should print the average number of
multiple runs.
Add evlist__copy_res_stats() to update the aggr counter (for display)
using the values in the evsel->stats.res_stats.
Signed-off-by: Namhyung Kim <[email protected]>
---
tools/perf/builtin-stat.c | 5 ++++-
tools/perf/util/stat.c | 22 ++++++++++++++++++++++
tools/perf/util/stat.h | 1 +
3 files changed, 27 insertions(+), 1 deletion(-)
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index e549862f90f0..42f84975a4d5 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -2829,8 +2829,11 @@ int cmd_stat(int argc, const char **argv)
}
}
- if (!forever && status != -1 && (!interval || stat_config.summary))
+ if (!forever && status != -1 && (!interval || stat_config.summary)) {
+ if (stat_config.run_count > 1)
+ evlist__copy_res_stats(&stat_config, evsel_list);
print_counters(NULL, argc, argv);
+ }
evlist__finalize_ctlfd(evsel_list);
diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
index 0f7b8a8cdea6..967e583392c7 100644
--- a/tools/perf/util/stat.c
+++ b/tools/perf/util/stat.c
@@ -264,6 +264,28 @@ void evlist__copy_prev_raw_counts(struct evlist *evlist)
evsel__copy_prev_raw_counts(evsel);
}
+static void evsel__copy_res_stats(struct evsel *evsel)
+{
+ struct perf_stat_evsel *ps = evsel->stats;
+
+ /*
+ * For GLOBAL aggregation mode, it updates the counts for each run
+ * in the evsel->stats.res_stats. See perf_stat_process_counter().
+ */
+ *ps->aggr[0].counts.values = avg_stats(&ps->res_stats);
+}
+
+void evlist__copy_res_stats(struct perf_stat_config *config, struct evlist *evlist)
+{
+ struct evsel *evsel;
+
+ if (config->aggr_mode != AGGR_GLOBAL)
+ return;
+
+ evlist__for_each_entry(evlist, evsel)
+ evsel__copy_res_stats(evsel);
+}
+
static size_t pkg_id_hash(long __key, void *ctx __maybe_unused)
{
uint64_t *key = (uint64_t *) __key;
diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
index 7abff7cbb5a1..1cbc26b587ba 100644
--- a/tools/perf/util/stat.h
+++ b/tools/perf/util/stat.h
@@ -182,6 +182,7 @@ void evlist__save_aggr_prev_raw_counts(struct evlist *evlist);
int evlist__alloc_aggr_stats(struct evlist *evlist, int nr_aggr);
void evlist__reset_aggr_stats(struct evlist *evlist);
+void evlist__copy_res_stats(struct perf_stat_config *config, struct evlist *evlist);
int perf_stat_process_counter(struct perf_stat_config *config,
struct evsel *counter);
--
2.41.0.162.gfafddb0af9-goog
On Fri, Jun 16, 2023 at 12:32:11AM -0700, Namhyung Kim wrote:
> When -r option is used, perf stat runs the command multiple times and
> update stats in the evsel->stats.res_stats for global aggregation. But
> the value is never used and the value it prints at the end is just the
> value from the last run. I think we should print the average number of
> multiple runs.
>
> Add evlist__copy_res_stats() to update the aggr counter (for display)
> using the values in the evsel->stats.res_stats.
>
> Signed-off-by: Namhyung Kim <[email protected]>
this is the 'real' fix right? I thought this was the way it worked before
anyway works nicely now, would be nice to add some tests for this,
but not sure how bad it'd be ;-)
Acked/Tested-by: Jiri Olsa <[email protected]>
thanks,
jirka
> ---
> tools/perf/builtin-stat.c | 5 ++++-
> tools/perf/util/stat.c | 22 ++++++++++++++++++++++
> tools/perf/util/stat.h | 1 +
> 3 files changed, 27 insertions(+), 1 deletion(-)
>
> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> index e549862f90f0..42f84975a4d5 100644
> --- a/tools/perf/builtin-stat.c
> +++ b/tools/perf/builtin-stat.c
> @@ -2829,8 +2829,11 @@ int cmd_stat(int argc, const char **argv)
> }
> }
>
> - if (!forever && status != -1 && (!interval || stat_config.summary))
> + if (!forever && status != -1 && (!interval || stat_config.summary)) {
> + if (stat_config.run_count > 1)
> + evlist__copy_res_stats(&stat_config, evsel_list);
> print_counters(NULL, argc, argv);
> + }
>
> evlist__finalize_ctlfd(evsel_list);
>
> diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
> index 0f7b8a8cdea6..967e583392c7 100644
> --- a/tools/perf/util/stat.c
> +++ b/tools/perf/util/stat.c
> @@ -264,6 +264,28 @@ void evlist__copy_prev_raw_counts(struct evlist *evlist)
> evsel__copy_prev_raw_counts(evsel);
> }
>
> +static void evsel__copy_res_stats(struct evsel *evsel)
> +{
> + struct perf_stat_evsel *ps = evsel->stats;
> +
> + /*
> + * For GLOBAL aggregation mode, it updates the counts for each run
> + * in the evsel->stats.res_stats. See perf_stat_process_counter().
> + */
> + *ps->aggr[0].counts.values = avg_stats(&ps->res_stats);
> +}
> +
> +void evlist__copy_res_stats(struct perf_stat_config *config, struct evlist *evlist)
> +{
> + struct evsel *evsel;
> +
> + if (config->aggr_mode != AGGR_GLOBAL)
> + return;
> +
> + evlist__for_each_entry(evlist, evsel)
> + evsel__copy_res_stats(evsel);
> +}
> +
> static size_t pkg_id_hash(long __key, void *ctx __maybe_unused)
> {
> uint64_t *key = (uint64_t *) __key;
> diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
> index 7abff7cbb5a1..1cbc26b587ba 100644
> --- a/tools/perf/util/stat.h
> +++ b/tools/perf/util/stat.h
> @@ -182,6 +182,7 @@ void evlist__save_aggr_prev_raw_counts(struct evlist *evlist);
>
> int evlist__alloc_aggr_stats(struct evlist *evlist, int nr_aggr);
> void evlist__reset_aggr_stats(struct evlist *evlist);
> +void evlist__copy_res_stats(struct perf_stat_config *config, struct evlist *evlist);
>
> int perf_stat_process_counter(struct perf_stat_config *config,
> struct evsel *counter);
> --
> 2.41.0.162.gfafddb0af9-goog
>
On Sat, Jun 17, 2023 at 11:45 AM Jiri Olsa <[email protected]> wrote:
>
> On Fri, Jun 16, 2023 at 12:32:11AM -0700, Namhyung Kim wrote:
> > When -r option is used, perf stat runs the command multiple times and
> > update stats in the evsel->stats.res_stats for global aggregation. But
> > the value is never used and the value it prints at the end is just the
> > value from the last run. I think we should print the average number of
> > multiple runs.
> >
> > Add evlist__copy_res_stats() to update the aggr counter (for display)
> > using the values in the evsel->stats.res_stats.
> >
> > Signed-off-by: Namhyung Kim <[email protected]>
>
> this is the 'real' fix right? I thought this was the way it worked before
It worked like the patch 1/2 before.
>
> anyway works nicely now, would be nice to add some tests for this,
> but not sure how bad it'd be ;-)
Maybe we can parse the raw counter output from the -v option and
calculate the average then compare.
>
> Acked/Tested-by: Jiri Olsa <[email protected]>
Thanks,
Namhyung
>
> > ---
> > tools/perf/builtin-stat.c | 5 ++++-
> > tools/perf/util/stat.c | 22 ++++++++++++++++++++++
> > tools/perf/util/stat.h | 1 +
> > 3 files changed, 27 insertions(+), 1 deletion(-)
> >
> > diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> > index e549862f90f0..42f84975a4d5 100644
> > --- a/tools/perf/builtin-stat.c
> > +++ b/tools/perf/builtin-stat.c
> > @@ -2829,8 +2829,11 @@ int cmd_stat(int argc, const char **argv)
> > }
> > }
> >
> > - if (!forever && status != -1 && (!interval || stat_config.summary))
> > + if (!forever && status != -1 && (!interval || stat_config.summary)) {
> > + if (stat_config.run_count > 1)
> > + evlist__copy_res_stats(&stat_config, evsel_list);
> > print_counters(NULL, argc, argv);
> > + }
> >
> > evlist__finalize_ctlfd(evsel_list);
> >
> > diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
> > index 0f7b8a8cdea6..967e583392c7 100644
> > --- a/tools/perf/util/stat.c
> > +++ b/tools/perf/util/stat.c
> > @@ -264,6 +264,28 @@ void evlist__copy_prev_raw_counts(struct evlist *evlist)
> > evsel__copy_prev_raw_counts(evsel);
> > }
> >
> > +static void evsel__copy_res_stats(struct evsel *evsel)
> > +{
> > + struct perf_stat_evsel *ps = evsel->stats;
> > +
> > + /*
> > + * For GLOBAL aggregation mode, it updates the counts for each run
> > + * in the evsel->stats.res_stats. See perf_stat_process_counter().
> > + */
> > + *ps->aggr[0].counts.values = avg_stats(&ps->res_stats);
> > +}
> > +
> > +void evlist__copy_res_stats(struct perf_stat_config *config, struct evlist *evlist)
> > +{
> > + struct evsel *evsel;
> > +
> > + if (config->aggr_mode != AGGR_GLOBAL)
> > + return;
> > +
> > + evlist__for_each_entry(evlist, evsel)
> > + evsel__copy_res_stats(evsel);
> > +}
> > +
> > static size_t pkg_id_hash(long __key, void *ctx __maybe_unused)
> > {
> > uint64_t *key = (uint64_t *) __key;
> > diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
> > index 7abff7cbb5a1..1cbc26b587ba 100644
> > --- a/tools/perf/util/stat.h
> > +++ b/tools/perf/util/stat.h
> > @@ -182,6 +182,7 @@ void evlist__save_aggr_prev_raw_counts(struct evlist *evlist);
> >
> > int evlist__alloc_aggr_stats(struct evlist *evlist, int nr_aggr);
> > void evlist__reset_aggr_stats(struct evlist *evlist);
> > +void evlist__copy_res_stats(struct perf_stat_config *config, struct evlist *evlist);
> >
> > int perf_stat_process_counter(struct perf_stat_config *config,
> > struct evsel *counter);
> > --
> > 2.41.0.162.gfafddb0af9-goog
> >