Skip to contentSkip to conversation

perf: Fix performance issue with perf report

PATCH In review Anton Blanchard 2 participants 3 messages May 4, 2010 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf: Fix performance issue with perf report

On a large machine we spend a lot of time in perf_header__find_attr when running perf report.

If we are parsing a file without PERF_SAMPLE_ID then for each sample we call perf_header__find_attr and loop through all counter IDs, never finding a match. As the machine gets larger there are more per cpu counters and we spend an awful lot of time in there.

The patch below initialises each sample id to -1ULL and checks for this in perf_header__find_attr. We may need to do something more intelligent eventually (eg a hash lookup from counter id to attr) but this at least fixes the most common usage of perf report.

--

tools/perf/util/event.c +1 -0
Index: linux.trees.git/tools/perf/util/event.c
===================================================================
@@ -713,6 +713,7 @@ int event__parse_sample(event_t *event, 
 		array++;
 	}
 
+	data->id = -1ULL;
 	if (type & PERF_SAMPLE_ID) {
 		data->id = *array;
 		array++;
tools/perf/util/header.c +8 -0
Index: linux.trees.git/tools/perf/util/header.c
===================================================================
@@ -923,6 +923,14 @@ perf_header__find_attr(u64 id, struct pe
 {
 	int i;
 
+	/*
+	 * We set id to -1 if the data file doesn't contain sample
+	 * ids. Check for this and avoid walking through the entire
+	 * list of ids which may be large.
+	 */
+	if (id == -1ULL)
+		return NULL;
+
 	for (i = 0; i < header->attrs; i++) {
 		struct perf_header_attr *attr = header->attr[i];
 		int j;
Acked-by on v1 1/1
#1 lore raw
Quoted 9 lines · Anton Blanchard

On a large machine we spend a lot of time in perf_header__find_attr when running perf report.

If we are parsing a file without PERF_SAMPLE_ID then for each sample we call perf_header__find_attr and loop through all counter IDs, never finding a match. As the machine gets larger there are more per cpu counters and we spend an awful lot of time in there.

Quoted 6 lines

The patch below initialises each sample id to -1ULL and checks for this in perf_header__find_attr. We may need to do something more intelligent eventually (eg a hash lookup from counter id to attr) but this at least fixes the most common usage of perf report.

#2 lore raw
In reply to Anton Blanchard
Commit-ID:  02bf60aad7d5912dfcdbe0154f1bd67ea7a8301e
Gitweb:     http://git.kernel.org/tip/02bf60aad7d5912dfcdbe0154f1bd67ea7a8301e
Author:     Anton Blanchard <[email protected]>
AuthorDate: Tue, 4 May 2010 21:19:15 +1000
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Tue, 4 May 2010 10:54:09 -0300

perf: Fix performance issue with perf report

On a large machine we spend a lot of time in perf_header__find_attr when running perf report.

If we are parsing a file without PERF_SAMPLE_ID then for each sample we call perf_header__find_attr and loop through all counter IDs, never finding a match. As the machine gets larger there are more per cpu counters and we spend an awful lot of time in there.

The patch below initialises each sample id to -1ULL and checks for this in perf_header__find_attr. We may need to do something more intelligent eventually (eg a hash lookup from counter id to attr) but this at least fixes the most common usage of perf report.

Cc Peter Zijlstra <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Eric B Munson <[email protected]>

LKML-Reference: <20100504111915.GB14636@kryten>

--

---
 tools/perf/util/event.c  |    1 +
 tools/perf/util/header.c |    8 ++++++++
 2 files changed, 9 insertions(+), 0 deletions(-)
tools/perf/util/event.c +1 -0
index 1757b0f..2477270 100644
@@ -713,6 +713,7 @@ int event__parse_sample(event_t *event, u64 type, struct sample_data *data)
 		array++;
 	}
 
+	data->id = -1ULL;
 	if (type & PERF_SAMPLE_ID) {
 		data->id = *array;
 		array++;
tools/perf/util/header.c +8 -0
index 2b9f898..8847bec 100644
@@ -922,6 +922,14 @@ perf_header__find_attr(u64 id, struct perf_header *header)
 {
 	int i;
 
+	/*
+	 * We set id to -1 if the data file doesn't contain sample
+	 * ids. Check for this and avoid walking through the entire
+	 * list of ids which may be large.
+	 */
+	if (id == -1ULL)
+		return NULL;
+
 	for (i = 0; i < header->attrs; i++) {
 		struct perf_header_attr *attr = header->attr[i];
 		int j;