Skip to contentSkip to conversation

perf report: Fix sample number stats for branch entry mode

PATCH Applied Thomas Falcon 2 participants 2 messages Feb 20, 2025 – Feb 25, 2025 perf events

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 perf report: Fix sample number stats for branch entry mode

Currently, stats->nr_samples is incremented per entry in the branch stack instead of per sample taken. As a result, statistics of samples taken during perf record in --branch-filter or --branch-any mode does not seem correct. Instead call hists__inc_nr_samples() for each sample taken instead of for each entry in the branch stack.

Before:

$ ./perf record -e cycles:u -b -c 10000000000 ./tchain_edit
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.005 MB perf.data (2 samples) ]
$ perf report -D | tail -n 16
Aggregated stats:
               TOTAL events:         16
                COMM events:          2  (12.5%)
                EXIT events:          1  ( 6.2%)
              SAMPLE events:          2  (12.5%)
               MMAP2 events:          2  (12.5%)
             KSYMBOL events:          1  ( 6.2%)
      FINISHED_ROUND events:          1  ( 6.2%)
            ID_INDEX events:          1  ( 6.2%)
          THREAD_MAP events:          1  ( 6.2%)
             CPU_MAP events:          1  ( 6.2%)
        EVENT_UPDATE events:          2  (12.5%)
           TIME_CONV events:          1  ( 6.2%)
       FINISHED_INIT events:          1  ( 6.2%)
cpu_core/cycles/u stats:
              SAMPLE events:         64

After:

$ ./perf report -D | tail -n 16
Aggregated stats:
               TOTAL events:         16
                COMM events:          2  (12.5%)
                EXIT events:          1  ( 6.2%)
              SAMPLE events:          2  (12.5%)
               MMAP2 events:          2  (12.5%)
             KSYMBOL events:          1  ( 6.2%)
      FINISHED_ROUND events:          1  ( 6.2%)
            ID_INDEX events:          1  ( 6.2%)
          THREAD_MAP events:          1  ( 6.2%)
             CPU_MAP events:          1  ( 6.2%)
        EVENT_UPDATE events:          2  (12.5%)
           TIME_CONV events:          1  ( 6.2%)
       FINISHED_INIT events:          1  ( 6.2%)
cpu_core/cycles/u stats:
              SAMPLE events:          2

---
v2: Missed null check for iter->he in iter_finish_branch_entry().

---
 tools/perf/util/hist.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
tools/perf/util/hist.c +6 -2
index 0f30f843c566..60b584218fa7 100644
@@ -975,8 +975,6 @@ iter_add_next_branch_entry(struct hist_entry_iter *iter, struct addr_location *a
 	if (he == NULL)
 		return -ENOMEM;
 
-	hists__inc_nr_samples(hists, he->filtered);
-
 out:
 	iter->he = he;
 	iter->curr++;
@@ -995,9 +993,15 @@ static int
 iter_finish_branch_entry(struct hist_entry_iter *iter,
 			 struct addr_location *al __maybe_unused)
 {
+	struct evsel *evsel = iter->evsel;
+	struct hists *hists = evsel__hists(evsel);
+
 	for (int i = 0; i < iter->total; i++)
 		branch_info__exit(&iter->bi[i]);
 
+	if (iter->he)
+		hists__inc_nr_samples(hists, iter->he->filtered);
+
 	zfree(&iter->bi);
 	iter->he = NULL;
 
Signature · 1 line
2.48.1
maintainer Applied
#1 lore raw
Quoted 9 lines · Thomas Falcon

Currently, stats->nr_samples is incremented per entry in the branch stack instead of per sample taken. As a result, statistics of samples taken during perf record in --branch-filter or --branch-any mode does not seem correct. Instead call hists__inc_nr_samples() for each sample taken instead of for each entry in the branch stack.

Before:

[...]

Applied to perf-tools-next, thanks!

Best regards,
Namhyung