[PATCH v1 2/3] perf stat: Support coresum event qualifier
From: Jin Yao
Date: Fri Mar 15 2019 - 04:13:51 EST
With this patch, we can use the coresum event qualifier in perf-stat.
root@skl:/tmp# perf stat -e cpu/event=0,umask=0x3,coresum=1/,cpu/event=0,umask=0x3/ -a -A -I1000
# time CPU counts unit events
1.000816660 S0-C0 80,676,154 cpu/event=0,umask=0x3,coresum=1/ (50.04%)
1.000816660 S0-C1 115,314,959 cpu/event=0,umask=0x3,coresum=1/ (50.04%)
1.000816660 S0-C2 126,541,249 cpu/event=0,umask=0x3,coresum=1/ (50.04%)
1.000816660 S0-C3 119,950,015 cpu/event=0,umask=0x3,coresum=1/ (50.04%)
1.000816660 CPU0 52,439,489 cpu/event=0,umask=0x3/ (49.96%)
1.000816660 CPU1 53,431,155 cpu/event=0,umask=0x3/ (49.96%)
1.000816660 CPU2 91,192,070 cpu/event=0,umask=0x3/ (49.96%)
1.000816660 CPU3 90,852,159 cpu/event=0,umask=0x3/ (49.96%)
1.000816660 CPU4 29,715,956 cpu/event=0,umask=0x3/ (49.96%)
1.000816660 CPU5 63,289,507 cpu/event=0,umask=0x3/ (49.96%)
1.000816660 CPU6 29,036,120 cpu/event=0,umask=0x3/ (49.96%)
1.000816660 CPU7 28,996,591 cpu/event=0,umask=0x3/ (49.97%)
In this example, we count the event 'ref-cycles' per-core and per-CPU in
one perf stat command-line. From the output, we can see:
S0-C0 = CPU0 + CPU4
S0-C1 = CPU1 + CPU5
S0-C2 = CPU2 + CPU6
S0-C3 = CPU3 + CPU7
So the result is expected (tiny difference is ignored).
Signed-off-by: Jin Yao <yao.jin@xxxxxxxxxxxxxxx>
---
tools/perf/Documentation/perf-stat.txt | 4 ++
tools/perf/builtin-stat.c | 21 +++++++
tools/perf/util/stat-display.c | 108 ++++++++++++++++++++++++---------
tools/perf/util/stat.c | 8 ++-
4 files changed, 108 insertions(+), 33 deletions(-)
diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
index 4bc2085..cd5fbd9 100644
--- a/tools/perf/Documentation/perf-stat.txt
+++ b/tools/perf/Documentation/perf-stat.txt
@@ -43,6 +43,10 @@ report::
param1 and param2 are defined as formats for the PMU in
/sys/bus/event_source/devices/<pmu>/format/*
+ 'coresum' is a event qualifier that sums up the event counts for both
+ hardware threads in a core. For example:
+ perf stat -A -a -e cpu/event,coresum=1/,otherevent ...
+
- a symbolically formed event like 'pmu/config=M,config1=N,config2=K/'
where M, N, K are numbers (in decimal, hex, octal format).
Acceptable values for each of 'config', 'config1' and 'config2'
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 7b8f09b..8ce0168 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -832,6 +832,18 @@ static int perf_stat__get_core_cached(struct perf_stat_config *config,
return perf_stat__get_aggr(config, perf_stat__get_core, map, idx);
}
+static bool term_coresum_set(void)
+{
+ struct perf_evsel *counter;
+
+ evlist__for_each_entry(evsel_list, counter) {
+ if (counter->coresum)
+ return true;
+ }
+
+ return false;
+}
+
static int perf_stat_init_aggr_mode(void)
{
int nr;
@@ -852,6 +864,15 @@ static int perf_stat_init_aggr_mode(void)
stat_config.aggr_get_id = perf_stat__get_core_cached;
break;
case AGGR_NONE:
+ if (term_coresum_set()) {
+ if (cpu_map__build_core_map(evsel_list->cpus,
+ &stat_config.aggr_map)) {
+ perror("cannot build core map");
+ return -1;
+ }
+ stat_config.aggr_get_id = perf_stat__get_core_cached;
+ }
+ break;
case AGGR_GLOBAL:
case AGGR_THREAD:
case AGGR_UNSET:
diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
index 6d043c7..23c9368 100644
--- a/tools/perf/util/stat-display.c
+++ b/tools/perf/util/stat-display.c
@@ -93,9 +93,17 @@ static void aggr_printout(struct perf_stat_config *config,
config->csv_sep);
break;
case AGGR_NONE:
- fprintf(config->output, "CPU%*d%s",
- config->csv_output ? 0 : -4,
- perf_evsel__cpus(evsel)->map[id], config->csv_sep);
+ if (evsel->coresum) {
+ fprintf(config->output, "S%d-C%*d%s",
+ cpu_map__id_to_socket(id),
+ config->csv_output ? 0 : -5,
+ cpu_map__id_to_cpu(id), config->csv_sep);
+ } else {
+ fprintf(config->output, "CPU%*d%s ",
+ config->csv_output ? 0 : -5,
+ perf_evsel__cpus(evsel)->map[id],
+ config->csv_sep);
+ }
break;
case AGGR_THREAD:
fprintf(config->output, "%*s-%*d%s",
@@ -599,6 +607,41 @@ static void aggr_cb(struct perf_stat_config *config,
}
}
+static void print_counter_aggrdata(struct perf_stat_config *config,
+ struct perf_evsel *counter, int s,
+ char *prefix, bool metric_only,
+ bool *first)
+{
+ struct aggr_data ad;
+ FILE *output = config->output;
+ u64 ena, run, val;
+ int id, nr;
+ double uval;
+
+ ad.id = id = config->aggr_map->map[s];
+ ad.val = ad.ena = ad.run = 0;
+ ad.nr = 0;
+ if (!collect_data(config, counter, aggr_cb, &ad))
+ return;
+
+ nr = ad.nr;
+ ena = ad.ena;
+ run = ad.run;
+ val = ad.val;
+ if (*first && metric_only) {
+ *first = false;
+ aggr_printout(config, counter, id, nr);
+ }
+ if (prefix && !metric_only)
+ fprintf(output, "%s", prefix);
+
+ uval = val * counter->scale;
+ printout(config, id, nr, counter, uval, prefix,
+ run, ena, 1.0, &rt_stat);
+ if (!metric_only)
+ fputc('\n', output);
+}
+
static void print_aggr(struct perf_stat_config *config,
struct perf_evlist *evlist,
char *prefix)
@@ -606,9 +649,7 @@ static void print_aggr(struct perf_stat_config *config,
bool metric_only = config->metric_only;
FILE *output = config->output;
struct perf_evsel *counter;
- int s, id, nr;
- double uval;
- u64 ena, run, val;
+ int s;
bool first;
if (!(config->aggr_map || config->aggr_get_id))
@@ -621,36 +662,16 @@ static void print_aggr(struct perf_stat_config *config,
* Without each counter has its own line.
*/
for (s = 0; s < config->aggr_map->nr; s++) {
- struct aggr_data ad;
if (prefix && metric_only)
fprintf(output, "%s", prefix);
- ad.id = id = config->aggr_map->map[s];
first = true;
evlist__for_each_entry(evlist, counter) {
if (is_duration_time(counter))
continue;
-
- ad.val = ad.ena = ad.run = 0;
- ad.nr = 0;
- if (!collect_data(config, counter, aggr_cb, &ad))
- continue;
- nr = ad.nr;
- ena = ad.ena;
- run = ad.run;
- val = ad.val;
- if (first && metric_only) {
- first = false;
- aggr_printout(config, counter, id, nr);
- }
- if (prefix && !metric_only)
- fprintf(output, "%s", prefix);
-
- uval = val * counter->scale;
- printout(config, id, nr, counter, uval, prefix,
- run, ena, 1.0, &rt_stat);
- if (!metric_only)
- fputc('\n', output);
+ print_counter_aggrdata(config, counter, s,
+ prefix, metric_only,
+ &first);
}
if (metric_only)
fputc('\n', output);
@@ -1101,6 +1122,30 @@ static void print_footer(struct perf_stat_config *config)
"the same PMU. Try reorganizing the group.\n");
}
+static void print_coresum(struct perf_stat_config *config,
+ struct perf_evsel *counter, char *prefix)
+{
+ bool metric_only = config->metric_only;
+ FILE *output = config->output;
+ int s;
+ bool first = true;
+
+ if (!(config->aggr_map || config->aggr_get_id))
+ return;
+
+ for (s = 0; s < config->aggr_map->nr; s++) {
+ if (prefix && metric_only)
+ fprintf(output, "%s", prefix);
+
+ print_counter_aggrdata(config, counter, s,
+ prefix, metric_only,
+ &first);
+ }
+
+ if (metric_only)
+ fputc('\n', output);
+}
+
void
perf_evlist__print_counters(struct perf_evlist *evlist,
struct perf_stat_config *config,
@@ -1157,7 +1202,10 @@ perf_evlist__print_counters(struct perf_evlist *evlist,
evlist__for_each_entry(evlist, counter) {
if (is_duration_time(counter))
continue;
- print_counter(config, counter, prefix);
+ if (counter->coresum)
+ print_coresum(config, counter, prefix);
+ else
+ print_counter(config, counter, prefix);
}
}
break;
diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
index 4d40515..33a6e3c 100644
--- a/tools/perf/util/stat.c
+++ b/tools/perf/util/stat.c
@@ -277,9 +277,11 @@ process_counter_values(struct perf_stat_config *config, struct perf_evsel *evsel
if (!evsel->snapshot)
perf_evsel__compute_deltas(evsel, cpu, thread, count);
perf_counts_values__scale(count, config->scale, NULL);
- if (config->aggr_mode == AGGR_NONE)
- perf_stat__update_shadow_stats(evsel, count->val, cpu,
- &rt_stat);
+ if ((config->aggr_mode == AGGR_NONE) && (!evsel->coresum)) {
+ perf_stat__update_shadow_stats(evsel, count->val,
+ cpu, &rt_stat);
+ }
+
if (config->aggr_mode == AGGR_THREAD) {
if (config->stats)
perf_stat__update_shadow_stats(evsel,
--
2.7.4