perf stat: Aggregate per-thread stats using evsel->stats->aggr

Per-thread aggregation doesn't use the CPU numbers but the logic should
be the same.  Initialize cpu_aggr_map separately for AGGR_THREAD and use
thread map idx to aggregate counter values.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Athira Jajeev <atrajeev@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: James Clark <james.clark@arm.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Michael Petlan <mpetlan@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Xing Zhengjun <zhengjun.xing@linux.intel.com>
Link: https://lore.kernel.org/r/20221018020227.85905-12-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools/perf/builtin-stat.c

@@ -1465,6 +1465,21 @@ static int perf_stat_init_aggr_mode(void)
 		stat_config.aggr_get_id = aggr_mode__get_id(stat_config.aggr_mode);
 	}
 
+	if (stat_config.aggr_mode == AGGR_THREAD) {
+		nr = perf_thread_map__nr(evsel_list->core.threads);
+		stat_config.aggr_map = cpu_aggr_map__empty_new(nr);
+		if (stat_config.aggr_map == NULL)
+			return -ENOMEM;
+
+		for (int s = 0; s < nr; s++) {
+			struct aggr_cpu_id id = aggr_cpu_id__empty();
+
+			id.thread_idx = s;
+			stat_config.aggr_map->map[s] = id;
+		}
+		return 0;
+	}
+
 	/*
 	 * The evsel_list->cpus is the base we operate on,
 	 * taking the highest cpu number to be the size of
@@ -1674,6 +1689,22 @@ static int perf_stat_init_aggr_mode_file(struct perf_stat *st)
 	aggr_cpu_id_get_t get_id = aggr_mode__get_aggr_file(stat_config.aggr_mode);
 	bool needs_sort = stat_config.aggr_mode != AGGR_NONE;
 
+	if (stat_config.aggr_mode == AGGR_THREAD) {
+		int nr = perf_thread_map__nr(evsel_list->core.threads);
+
+		stat_config.aggr_map = cpu_aggr_map__empty_new(nr);
+		if (stat_config.aggr_map == NULL)
+			return -ENOMEM;
+
+		for (int s = 0; s < nr; s++) {
+			struct aggr_cpu_id id = aggr_cpu_id__empty();
+
+			id.thread_idx = s;
+			stat_config.aggr_map->map[s] = id;
+		}
+		return 0;
+	}
+
 	if (!get_id)
 		return 0;
 

tools/perf/util/stat.c

@@ -422,6 +422,24 @@ process_counter_values(struct perf_stat_config *config, struct evsel *evsel,
 		evsel__compute_deltas(evsel, cpu_map_idx, thread, count);
 	perf_counts_values__scale(count, config->scale, NULL);
 
+	if (config->aggr_mode == AGGR_THREAD) {
+		struct perf_counts_values *aggr_counts = &ps->aggr[thread].counts;
+
+		/*
+		 * Skip value 0 when enabling --per-thread globally,
+		 * otherwise too many 0 output.
+		 */
+		if (count->val == 0 && config->system_wide)
+			return 0;
+
+		ps->aggr[thread].nr++;
+
+		aggr_counts->val += count->val;
+		aggr_counts->ena += count->ena;
+		aggr_counts->run += count->run;
+		goto update;
+	}
+
 	if (ps->aggr) {
 		struct perf_cpu cpu = perf_cpu_map__cpu(evsel->core.cpus, cpu_map_idx);
 		struct aggr_cpu_id aggr_id = config->aggr_get_id(config, cpu);
@@ -436,8 +454,9 @@ process_counter_values(struct perf_stat_config *config, struct evsel *evsel,
 			ps_aggr->nr++;
 
 			/*
-			 * When any result is bad, make them all to give
+			 * When any result is bad, make them all to give consistent output
-			 * consistent output in interval mode.
+			 * in interval mode.  But per-task counters can have 0 enabled time
+			 * when some tasks are idle.
 			 */
 			if (evsel__count_has_error(evsel, count, config) && !ps_aggr->failed) {
 				ps_aggr->counts.val = 0;
@@ -455,6 +474,7 @@ process_counter_values(struct perf_stat_config *config, struct evsel *evsel,
 		}
 	}
 
+update:
 	switch (config->aggr_mode) {
 	case AGGR_THREAD:
 	case AGGR_CORE: