[tip:perf/urgent] perf annotate: Add comment about annotated_src->nr_histograms
From: tip-bot for Arnaldo Carvalho de Melo
Date: Thu Jun 07 2018 - 04:09:00 EST
Commit-ID: 9132d3d92d8953ee79690408a9a24d938be22cd8
Gitweb: https://git.kernel.org/tip/9132d3d92d8953ee79690408a9a24d938be22cd8
Author: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
AuthorDate: Fri, 25 May 2018 11:27:38 -0300
Committer: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
CommitDate: Mon, 4 Jun 2018 10:28:52 -0300
perf annotate: Add comment about annotated_src->nr_histograms
When we have multiple groups in an evlist, say:
$ perf stat -e '{cycles,instructions},{cache-references,cache-misses}' sleep 1
Performance counter stats for 'sleep 1':
343,134 cycles:u
249,292 instructions:u # 0.73 insn per cycle
15,556 cache-references:u
8,925 cache-misses:u # 57.373 % of all cache refs
1.000957550 seconds time elapsed
$
Then the perf_evsel instances for the two group leaders ("cycles" and
"cache-references") will have evsel->nr_members set to 2, while all the
evsel->evlist->nr_entries will be set to 4, so we can't use
evsel->evlist->nr_entries everywhere, as event groups need to be taken
into account.
But this probably requires us to audit at least the forced-group code,
where we want all of the events to be in a "group", to see them all in
the screen, one column for each, even knowing that they were not
necessarily scheduled to count at the same time by the kernel perf
subsystem.
Cc: Adrian Hunter <adrian.hunter@xxxxxxxxx>
Cc: David Ahern <dsahern@xxxxxxxxx>
Cc: Jiri Olsa <jolsa@xxxxxxxxxx>
Cc: Namhyung Kim <namhyung@xxxxxxxxxx>
Cc: Wang Nan <wangnan0@xxxxxxxxxx>
Link: https://lkml.kernel.org/n/tip-2g0vwqnc49wl4ttjk8dvpgcc@xxxxxxxxxxxxxx
Signed-off-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
---
tools/perf/util/annotate.h | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/tools/perf/util/annotate.h b/tools/perf/util/annotate.h
index 7ad503fbff74..3dc4ca1d6c08 100644
--- a/tools/perf/util/annotate.h
+++ b/tools/perf/util/annotate.h
@@ -202,6 +202,10 @@ struct cyc_hist {
/** struct annotated_source - symbols with hits have this attached as in sannotation
*
* @histograms: Array of addr hit histograms per event being monitored
+ * nr_histograms: This may not be the same as evsel->evlist->nr_entries if
+ * we have more than a group in a evlist, where we will want
+ * to see each group separately, that is why symbol__annotate2()
+ * sets src->nr_histograms to evsel->nr_members.
* @lines: If 'print_lines' is specified, per source code line percentages
* @source: source parsed from a disassembler like objdump -dS
* @cyc_hist: Average cycles per basic block