Skip to contentSkip to conversation

perf report: Add suppport for event group view (v2)

PATCH In review Namhyung Kim 3 participants 20 messages Sep 26, 2012 – Sep 28, 2012 perf events

lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

cover perf report: Add suppport for event group view (v2)

Hi,

This is my second attempt to support event group on perf report. For basic idea and usage example, please see my original post [1].

The main difference than v1 is adding HEADER_GROUP_DESC feature and use it for regenerating group relationship on perf report. It will save actual (i.e. contains non-leader member) group information to perf header.

Patch 1 is a bug fix which can be applied indenpedently.

Patch 2-4 are cleanups to facilitate later changes and came from my cumulative report series [2].

You can also access it via my tree at:

git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git perf/group-v2

Any comments are welcome, thanks,
Namhyung

v1 -> v2:
* save group relation to header (Jiri)
* rebase on top of current acme/perf/core

[1] https://lkml.org/lkml/2012/7/24/81
[2] https://lkml.org/lkml/2012/9/13/81

Namhyung Kim (16):
perf hists: Add missing period_* fields when collapsing a hist entry perf hists: Introduce struct he_stat
perf hists: Move he->stat.nr_events initialization to a template perf hists: Add more helpers for hist entry stat
perf tools: Keep group information
perf evlist: Add perf_evlist__recalc_nr_groups
perf header: Add HEADER_GROUP_DESC feature
perf hists: Collapse group hist_entries to a leader
perf hists: Maintain total periods of group members in the leader perf report: Make another loop for output resorting
perf ui/hist: Add support for event group view
perf ui/browser: Add support for event group view
perf ui/gtk: Add support for event group view
perf report: Bypass non-leader events when event group is enabled perf report: Show group description when event group is enabled perf report: Add --group option

 tools/perf/builtin-record.c    |   6 ++
 tools/perf/builtin-report.c    |  29 +++++++
 tools/perf/ui/browsers/hists.c | 107 ++++++++++++++++++++---
 tools/perf/ui/gtk/browser.c    |  70 ++++++++++++---
 tools/perf/ui/hist.c           | 104 +++++++++++++++++------
 tools/perf/ui/stdio/hist.c     |   4 +-
 tools/perf/util/evlist.c       |  26 +++++-
 tools/perf/util/evlist.h       |   2 +
 tools/perf/util/evsel.c        |  25 ++++++
 tools/perf/util/evsel.h        |  23 +++++
 tools/perf/util/header.c       | 149 ++++++++++++++++++++++++++++++++
 tools/perf/util/header.h       |   2 +
 tools/perf/util/hist.c         | 188 +++++++++++++++++++++++++++++++++++------
 tools/perf/util/hist.h         |   2 +
 tools/perf/util/parse-events.c |   1 +
 tools/perf/util/parse-events.h |   1 +
 tools/perf/util/parse-events.y |   4 +
 tools/perf/util/sort.h         |  17 ++--
 tools/perf/util/symbol.c       |   4 +
 tools/perf/util/symbol.h       |   3 +-
 20 files changed, 683 insertions(+), 84 deletions(-)
Signature · 1 line
1.7.11.4
author maintainer
#1 lore raw

PATCH 1/16 perf hists: Add missing period_* fields when collapsing a hist entry

From: Namhyung Kim <[email protected]>

So that the perf report won't lost the cpu utilization information.

For example, if there're two process that have same name.

  $ perf report --stdio --showcpuutilization -s pid
  [SNIP]
  #   Overhead       sys        us  Command:  Pid
  #   ........  ........  ........  .............
  #
        55.12%     0.01%    55.10%  noploop:28781
        44.88%     0.06%    44.83%  noploop:28782

Before:

  $ perf report --stdio --showcpuutilization -s comm
  [SNIP]
  #   Overhead       sys        us
  #   ........  ........  ........
  #
       100.00%     0.06%    44.83%

After:

  $ perf report --stdio --showcpuutilization -s comm
  [SNIP]
  #   Overhead       sys        us
  #   ........  ........  ........
  #
       100.00%     0.07%    99.93%
Cc Jiri Olsa <[email protected]>
Cc Arun Sharma <[email protected]>
Cc Frederic Weisbecker <[email protected]>
---
 tools/perf/util/hist.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
tools/perf/util/hist.c +7 -2
index 6ec5398de89d..236bc9d98ff2 100644
@@ -410,8 +410,13 @@ static bool hists__collapse_insert_entry(struct hists *hists __maybe_unused,
 		cmp = hist_entry__collapse(iter, he);
 
 		if (!cmp) {
-			iter->period += he->period;
-			iter->nr_events += he->nr_events;
+			iter->period		+= he->period;
+			iter->period_sys	+= he->period_sys;
+			iter->period_us		+= he->period_us;
+			iter->period_guest_sys	+= he->period_guest_sys;
+			iter->period_guest_us	+= he->period_guest_us;
+			iter->nr_events		+= he->nr_events;
+
 			if (symbol_conf.use_callchain) {
 				callchain_cursor_reset(&callchain_cursor);
 				callchain_merge(&callchain_cursor,
Signature · 1 line
1.7.11.4
#2 lore raw

[tip:perf/core] perf hists: Add missing period_* fields when collapsing a hist entry

Commit-ID:  9ec60972a38011ad8a5676f4cd5e51ac508c36b6
Gitweb:     http://git.kernel.org/tip/9ec60972a38011ad8a5676f4cd5e51ac508c36b6
Author:     Namhyung Kim <[email protected]>
AuthorDate: Wed, 26 Sep 2012 16:47:28 +0900
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Wed, 26 Sep 2012 20:44:11 -0300

perf hists: Add missing period_* fields when collapsing a hist entry

So that the perf report won't lost the cpu utilization information.

For example, if there're two process that have same name.

  $ perf report --stdio --showcpuutilization -s pid
  [SNIP]
  #   Overhead       sys        us  Command:  Pid
  #   ........  ........  ........  .............
  #
        55.12%     0.01%    55.10%  noploop:28781
        44.88%     0.06%    44.83%  noploop:28782

Before:

  $ perf report --stdio --showcpuutilization -s comm
  [SNIP]
  #   Overhead       sys        us
  #   ........  ........  ........
  #
       100.00%     0.06%    44.83%

After:

  $ perf report --stdio --showcpuutilization -s comm
  [SNIP]
  #   Overhead       sys        us
  #   ........  ........  ........
  #
       100.00%     0.07%    99.93%
Cc Arun Sharma <[email protected]>
Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/util/hist.c |    9 +++++++--
 1 files changed, 7 insertions(+), 2 deletions(-)
tools/perf/util/hist.c +7 -2
index 6ec5398..236bc9d 100644
@@ -410,8 +410,13 @@ static bool hists__collapse_insert_entry(struct hists *hists __maybe_unused,
 		cmp = hist_entry__collapse(iter, he);
 
 		if (!cmp) {
-			iter->period += he->period;
-			iter->nr_events += he->nr_events;
+			iter->period		+= he->period;
+			iter->period_sys	+= he->period_sys;
+			iter->period_us		+= he->period_us;
+			iter->period_guest_sys	+= he->period_guest_sys;
+			iter->period_guest_us	+= he->period_guest_us;
+			iter->nr_events		+= he->nr_events;
+
 			if (symbol_conf.use_callchain) {
 				callchain_cursor_reset(&callchain_cursor);
 				callchain_merge(&callchain_cursor,
author maintainer
#3 lore raw

PATCH 2/16 perf hists: Introduce struct he_stat

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

The struct he_stat is for separating out statistics data of a hist entry. It is required for later changes.

It's just a mechanical change and should have no functional differences.

Cc Jiri Olsa <[email protected]>
Cc Arun Sharma <[email protected]>
Cc Frederic Weisbecker <[email protected]>
---
 tools/perf/ui/browsers/hists.c |  8 +++----
 tools/perf/ui/gtk/browser.c    |  2 +-
 tools/perf/ui/hist.c           | 32 +++++++++++++-------------
 tools/perf/ui/stdio/hist.c     |  2 +-
 tools/perf/util/hist.c         | 52 +++++++++++++++++++++++-------------------
 tools/perf/util/sort.h         | 16 ++++++++-----
 6 files changed, 60 insertions(+), 52 deletions(-)
tools/perf/ui/browsers/hists.c +4 -4
index a21f40bebbac..b9b1b173637c 100644
@@ -569,7 +569,7 @@ static int hist_browser__show_callchain(struct hist_browser *browser,
 static int hist_browser__hpp_color_ ## _name(struct perf_hpp *hpp,	\
 					     struct hist_entry *he)	\
 {									\
-	double percent = 100.0 * he->_field / hpp->total_period;	\
+	double percent = 100.0 * he->stat._field / hpp->total_period;	\
 	*(double *)hpp->ptr = percent;					\
 	return scnprintf(hpp->buf, hpp->size, "%6.2f%%", percent);	\
 }
@@ -982,7 +982,7 @@ static int hist_browser__fprintf_entry(struct hist_browser *browser,
 		folded_sign = hist_entry__folded(he);
 
 	hist_entry__sort_snprintf(he, s, sizeof(s), browser->hists);
-	percent = (he->period * 100.0) / browser->hists->stats.total_period;
+	percent = (he->stat.period * 100.0) / browser->hists->stats.total_period;
 
 	if (symbol_conf.use_callchain)
 		printed += fprintf(fp, "%c ", folded_sign);
@@ -990,10 +990,10 @@ static int hist_browser__fprintf_entry(struct hist_browser *browser,
 	printed += fprintf(fp, " %5.2f%%", percent);
 
 	if (symbol_conf.show_nr_samples)
-		printed += fprintf(fp, " %11u", he->nr_events);
+		printed += fprintf(fp, " %11u", he->stat.nr_events);
 
 	if (symbol_conf.show_total_period)
-		printed += fprintf(fp, " %12" PRIu64, he->period);
+		printed += fprintf(fp, " %12" PRIu64, he->stat.period);
 
 	printed += fprintf(fp, "%s\n", rtrim(s));
 
tools/perf/ui/gtk/browser.c +1 -1
index 7ff99ec1d95e..7107edc6c08d 100644
@@ -49,7 +49,7 @@ static const char *perf_gtk__get_percent_color(double percent)
 static int perf_gtk__hpp_color_ ## _name(struct perf_hpp *hpp,			\
 					 struct hist_entry *he)			\
 {										\
-	double percent = 100.0 * he->_field / hpp->total_period;		\
+	double percent = 100.0 * he->stat._field / hpp->total_period;		\
 	const char *markup;							\
 	int ret = 0;								\
 										\
tools/perf/ui/hist.c +16 -16
index e3f8cd46e7d7..d6ddeb10e678 100644
@@ -20,12 +20,12 @@ static int hpp__width_overhead(struct perf_hpp *hpp __maybe_unused)
 
 static int hpp__color_overhead(struct perf_hpp *hpp, struct hist_entry *he)
 {
-	double percent = 100.0 * he->period / hpp->total_period;
+	double percent = 100.0 * he->stat.period / hpp->total_period;
 
 	if (hpp->ptr) {
 		struct hists *old_hists = hpp->ptr;
 		u64 total_period = old_hists->stats.total_period;
-		u64 base_period = he->pair ? he->pair->period : 0;
+		u64 base_period = he->pair ? he->pair->stat.period : 0;
 
 		if (total_period)
 			percent = 100.0 * base_period / total_period;
@@ -38,13 +38,13 @@ static int hpp__color_overhead(struct perf_hpp *hpp, struct hist_entry *he)
 
 static int hpp__entry_overhead(struct perf_hpp *hpp, struct hist_entry *he)
 {
-	double percent = 100.0 * he->period / hpp->total_period;
+	double percent = 100.0 * he->stat.period / hpp->total_period;
 	const char *fmt = symbol_conf.field_sep ? "%.2f" : " %6.2f%%";
 
 	if (hpp->ptr) {
 		struct hists *old_hists = hpp->ptr;
 		u64 total_period = old_hists->stats.total_period;
-		u64 base_period = he->pair ? he->pair->period : 0;
+		u64 base_period = he->pair ? he->pair->stat.period : 0;
 
 		if (total_period)
 			percent = 100.0 * base_period / total_period;
@@ -69,13 +69,13 @@ static int hpp__width_overhead_sys(struct perf_hpp *hpp __maybe_unused)
 
 static int hpp__color_overhead_sys(struct perf_hpp *hpp, struct hist_entry *he)
 {
-	double percent = 100.0 * he->period_sys / hpp->total_period;
+	double percent = 100.0 * he->stat.period_sys / hpp->total_period;
 	return percent_color_snprintf(hpp->buf, hpp->size, "%6.2f%%", percent);
 }
 
 static int hpp__entry_overhead_sys(struct perf_hpp *hpp, struct hist_entry *he)
 {
-	double percent = 100.0 * he->period_sys / hpp->total_period;
+	double percent = 100.0 * he->stat.period_sys / hpp->total_period;
 	const char *fmt = symbol_conf.field_sep ? "%.2f" : "%6.2f%%";
 
 	return scnprintf(hpp->buf, hpp->size, fmt, percent);
@@ -95,13 +95,13 @@ static int hpp__width_overhead_us(struct perf_hpp *hpp __maybe_unused)
 
 static int hpp__color_overhead_us(struct perf_hpp *hpp, struct hist_entry *he)
 {
-	double percent = 100.0 * he->period_us / hpp->total_period;
+	double percent = 100.0 * he->stat.period_us / hpp->total_period;
 	return percent_color_snprintf(hpp->buf, hpp->size, "%6.2f%%", percent);
 }
 
 static int hpp__entry_overhead_us(struct perf_hpp *hpp, struct hist_entry *he)
 {
-	double percent = 100.0 * he->period_us / hpp->total_period;
+	double percent = 100.0 * he->stat.period_us / hpp->total_period;
 	const char *fmt = symbol_conf.field_sep ? "%.2f" : "%6.2f%%";
 
 	return scnprintf(hpp->buf, hpp->size, fmt, percent);
@@ -120,14 +120,14 @@ static int hpp__width_overhead_guest_sys(struct perf_hpp *hpp __maybe_unused)
 static int hpp__color_overhead_guest_sys(struct perf_hpp *hpp,
 					 struct hist_entry *he)
 {
-	double percent = 100.0 * he->period_guest_sys / hpp->total_period;
+	double percent = 100.0 * he->stat.period_guest_sys / hpp->total_period;
 	return percent_color_snprintf(hpp->buf, hpp->size, " %6.2f%% ", percent);
 }
 
 static int hpp__entry_overhead_guest_sys(struct perf_hpp *hpp,
 					 struct hist_entry *he)
 {
-	double percent = 100.0 * he->period_guest_sys / hpp->total_period;
+	double percent = 100.0 * he->stat.period_guest_sys / hpp->total_period;
 	const char *fmt = symbol_conf.field_sep ? "%.2f" : " %6.2f%% ";
 
 	return scnprintf(hpp->buf, hpp->size, fmt, percent);
@@ -146,14 +146,14 @@ static int hpp__width_overhead_guest_us(struct perf_hpp *hpp __maybe_unused)
 static int hpp__color_overhead_guest_us(struct perf_hpp *hpp,
 					struct hist_entry *he)
 {
-	double percent = 100.0 * he->period_guest_us / hpp->total_period;
+	double percent = 100.0 * he->stat.period_guest_us / hpp->total_period;
 	return percent_color_snprintf(hpp->buf, hpp->size, " %6.2f%% ", percent);
 }
 
 static int hpp__entry_overhead_guest_us(struct perf_hpp *hpp,
 					struct hist_entry *he)
 {
-	double percent = 100.0 * he->period_guest_us / hpp->total_period;
+	double percent = 100.0 * he->stat.period_guest_us / hpp->total_period;
 	const char *fmt = symbol_conf.field_sep ? "%.2f" : " %6.2f%% ";
 
 	return scnprintf(hpp->buf, hpp->size, fmt, percent);
@@ -175,7 +175,7 @@ static int hpp__entry_samples(struct perf_hpp *hpp, struct hist_entry *he)
 {
 	const char *fmt = symbol_conf.field_sep ? "%" PRIu64 : "%11" PRIu64;
 
-	return scnprintf(hpp->buf, hpp->size, fmt, he->nr_events);
+	return scnprintf(hpp->buf, hpp->size, fmt, he->stat.nr_events);
 }
 
 static int hpp__header_period(struct perf_hpp *hpp)
@@ -194,7 +194,7 @@ static int hpp__entry_period(struct perf_hpp *hpp, struct hist_entry *he)
 {
 	const char *fmt = symbol_conf.field_sep ? "%" PRIu64 : "%12" PRIu64;
 
-	return scnprintf(hpp->buf, hpp->size, fmt, he->period);
+	return scnprintf(hpp->buf, hpp->size, fmt, he->stat.period);
 }
 
 static int hpp__header_delta(struct perf_hpp *hpp)
@@ -220,11 +220,11 @@ static int hpp__entry_delta(struct perf_hpp *hpp, struct hist_entry *he)
 
 	old_total = pair_hists->stats.total_period;
 	if (old_total > 0 && he->pair)
-		old_percent = 100.0 * he->pair->period / old_total;
+		old_percent = 100.0 * he->pair->stat.period / old_total;
 
 	new_total = hpp->total_period;
 	if (new_total > 0)
-		new_percent = 100.0 * he->period / new_total;
+		new_percent = 100.0 * he->stat.period / new_total;
 
 	diff = new_percent - old_percent;
 	if (fabs(diff) >= 0.01)
tools/perf/ui/stdio/hist.c +1 -1
index 882461a42830..4382a1995cda 100644
@@ -271,7 +271,7 @@ static size_t hist_entry_callchain__fprintf(struct hist_entry *he,
 {
 	switch (callchain_param.mode) {
 	case CHAIN_GRAPH_REL:
-		return callchain__fprintf_graph(fp, &he->sorted_chain, he->period,
+		return callchain__fprintf_graph(fp, &he->sorted_chain, he->stat.period,
 						left_margin);
 		break;
 	case CHAIN_GRAPH_ABS:
tools/perf/util/hist.c +28 -24

129 lines not on this page · raw message

tools/perf/util/sort.h +10 -6
index 12d634792de5..30f2ae5be146 100644
@@ -43,6 +43,15 @@ extern struct sort_entry sort_sym_from;
 extern struct sort_entry sort_sym_to;
 extern enum sort_type sort__first_dimension;
 
+struct he_stat {
+	u64			period;
+	u64			period_sys;
+	u64			period_us;
+	u64			period_guest_sys;
+	u64			period_guest_us;
+	u32			nr_events;
+};
+
 /**
  * struct hist_entry - histogram entry
  *
@@ -52,16 +61,11 @@ extern enum sort_type sort__first_dimension;
 struct hist_entry {
 	struct rb_node		rb_node_in;
 	struct rb_node		rb_node;
-	u64			period;
-	u64			period_sys;
-	u64			period_us;
-	u64			period_guest_sys;
-	u64			period_guest_us;
+	struct he_stat		stat;
 	struct map_symbol	ms;
 	struct thread		*thread;
 	u64			ip;
 	s32			cpu;
-	u32			nr_events;
 
 	/* XXX These two should move to some tree widget lib */
 	u16			row_offset;
Signature · 1 line
1.7.11.4
author maintainer
#4 lore raw

PATCH 3/16 perf hists: Move he->stat.nr_events initialization to a template

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Since it is set to 1 for a new hist entry, no need to set to separately. Move it to a template entry.

Cc Jiri Olsa <[email protected]>
Cc Arun Sharma <[email protected]>
Cc Stephane Eranian <[email protected]>
Cc Frederic Weisbecker <[email protected]>
---
 tools/perf/util/hist.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
tools/perf/util/hist.c +3 -1
index ab3d11491991..ef39e6714cbb 100644
@@ -223,7 +223,7 @@ static struct hist_entry *hist_entry__new(struct hist_entry *template)
 
 	if (he != NULL) {
 		*he = *template;
-		he->stat.nr_events = 1;
+
 		if (he->ms.map)
 			he->ms.map->referenced = true;
 		if (symbol_conf.use_callchain)
@@ -323,6 +323,7 @@ struct hist_entry *__hists__add_branch_entry(struct hists *self,
 		.level	= al->level,
 		.stat = {
 			.period	= period,
+			.nr_events = 1,
 		},
 		.parent = sym_parent,
 		.filtered = symbol__parent_filter(sym_parent),
@@ -347,6 +348,7 @@ struct hist_entry *__hists__add_entry(struct hists *self,
 		.level	= al->level,
 		.stat = {
 			.period	= period,
+			.nr_events = 1,
 		},
 		.parent = sym_parent,
 		.filtered = symbol__parent_filter(sym_parent),
Signature · 1 line
1.7.11.4
author maintainer
#5 lore raw

PATCH 4/16 perf hists: Add more helpers for hist entry stat

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Add and use he_stat__add_{period,stat} for calculating hist entry's stat. It will be used for accumulated stats later as well.

Cc Jiri Olsa <[email protected]>
Cc Arun Sharma <[email protected]>
Cc Stephane Eranian <[email protected]>
Cc Frederic Weisbecker <[email protected]>
---
 tools/perf/util/hist.c | 26 ++++++++++++++++++--------
 1 file changed, 18 insertions(+), 8 deletions(-)
tools/perf/util/hist.c +18 -8
index ef39e6714cbb..c742a723e850 100644
@@ -151,6 +151,22 @@ static void hist_entry__add_cpumode_period(struct hist_entry *he,
 	}
 }
 
+static void he_stat__add_period(struct he_stat *he_stat, u64 period)
+{
+	he_stat->period		+= period;
+	he_stat->nr_events	+= 1;
+}
+
+static void he_stat__add_stat(struct he_stat *dest, struct he_stat *src)
+{
+	dest->period		+= src->period;
+	dest->period_sys	+= src->period_sys;
+	dest->period_us		+= src->period_us;
+	dest->period_guest_sys	+= src->period_guest_sys;
+	dest->period_guest_us	+= src->period_guest_us;
+	dest->nr_events		+= src->nr_events;
+}
+
 static void hist_entry__decay(struct hist_entry *he)
 {
 	he->stat.period = (he->stat.period * 7) / 8;
@@ -270,8 +286,7 @@ static struct hist_entry *add_hist_entry(struct hists *hists,
 		cmp = hist_entry__cmp(entry, he);
 
 		if (!cmp) {
-			he->stat.period += period;
-			++he->stat.nr_events;
+			he_stat__add_period(&he->stat, period);
 
 			/* If the map of an existing hist_entry has
 			 * become out-of-date due to an exec() or
@@ -416,12 +431,7 @@ static bool hists__collapse_insert_entry(struct hists *hists __maybe_unused,
 		cmp = hist_entry__collapse(iter, he);
 
 		if (!cmp) {
-			iter->stat.period		+= he->stat.period;
-			iter->stat.period_sys		+= he->stat.period_sys;
-			iter->stat.period_us		+= he->stat.period_us;
-			iter->stat.period_guest_sys	+= he->stat.period_guest_sys;
-			iter->stat.period_guest_us	+= he->stat.period_guest_us;
-			iter->stat.nr_events		+= he->stat.nr_events;
+			he_stat__add_stat(&iter->stat, &he->stat);
 
 			if (symbol_conf.use_callchain) {
 				callchain_cursor_reset(&callchain_cursor);
Signature · 1 line
1.7.11.4
author maintainer
#6 lore raw

PATCH 5/16 perf tools: Keep group information

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Add a few of group-related field in struct perf_{evlist,evsel} so that the group information in a evlist can be known easily.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/util/evlist.c       | 10 ++++++++--
 tools/perf/util/evlist.h       |  1 +
 tools/perf/util/evsel.h        | 10 ++++++++++
 tools/perf/util/parse-events.c |  1 +
 tools/perf/util/parse-events.h |  1 +
 tools/perf/util/parse-events.y |  4 ++++
 6 files changed, 25 insertions(+), 2 deletions(-)
tools/perf/util/evlist.c +8 -2
index 892353729c7a..199b6f1c3b22 100644
@@ -111,20 +111,26 @@ void perf_evlist__splice_list_tail(struct perf_evlist *evlist,
 void __perf_evlist__set_leader(struct list_head *list)
 {
 	struct perf_evsel *evsel, *leader;
+	int count = 0;
 
 	leader = list_entry(list->next, struct perf_evsel, node);
 	leader->leader = NULL;
 
 	list_for_each_entry(evsel, list, node) {
-		if (evsel != leader)
+		if (evsel != leader) {
 			evsel->leader = leader;
+			evsel->group_idx = count++;
+		}
 	}
+	leader->nr_members = count;
 }
 
 void perf_evlist__set_leader(struct perf_evlist *evlist)
 {
-	if (evlist->nr_entries)
+	if (evlist->nr_entries) {
+		evlist->nr_groups = 1;
 		__perf_evlist__set_leader(&evlist->entries);
+	}
 }
 
 int perf_evlist__add_default(struct perf_evlist *evlist)
tools/perf/util/evlist.h +1 -0
index 3f2e1e4ccdd5..946a6ada817b 100644
@@ -21,6 +21,7 @@ struct perf_evlist {
 	struct list_head entries;
 	struct hlist_head heads[PERF_EVLIST__HLIST_SIZE];
 	int		 nr_entries;
+	int		 nr_groups;
 	int		 nr_fds;
 	int		 nr_mmaps;
 	int		 mmap_len;
tools/perf/util/evsel.h +10 -0
index bb445d1cbc7b..820f005096c4 100644
@@ -73,6 +73,10 @@ struct perf_evsel {
 	int			exclude_GH;
 	struct perf_evsel	*leader;
 	char			*group_name;
+	union {
+		int		nr_members;
+		int		group_idx;
+	};
 };
 
 struct cpu_map;
@@ -211,4 +215,10 @@ static inline struct perf_evsel *perf_evsel__next(struct perf_evsel *evsel)
 {
 	return list_entry(evsel->node.next, struct perf_evsel, node);
 }
+
+/* Treat a non-group event as a leader */
+static inline bool perf_evsel__is_group_leader(struct perf_evsel *evsel)
+{
+	return evsel->leader == NULL;
+}
 #endif /* __PERF_EVSEL_H */
tools/perf/util/parse-events.c +1 -0
index bf5d033ee1b4..3c52d0ab9270 100644
@@ -830,6 +830,7 @@ int parse_events(struct perf_evlist *evlist, const char *str,
 	if (!ret) {
 		int entries = data.idx - evlist->nr_entries;
 		perf_evlist__splice_list_tail(evlist, &data.list, entries);
+		evlist->nr_groups += data.nr_groups;
 		return 0;
 	}
 
tools/perf/util/parse-events.h +1 -0
index c356e443448d..f6b0254afe17 100644
@@ -65,6 +65,7 @@ struct parse_events__term {
 struct parse_events_data__events {
 	struct list_head list;
 	int idx;
+	int nr_groups;
 };
 
 struct parse_events_data__terms {
tools/perf/util/parse-events.y +4 -0
index cd88209e3c58..d14bb507594b 100644
@@ -122,7 +122,9 @@ group_def:
 PE_NAME '{' events '}'
 {
 	struct list_head *list = $3;
+	struct parse_events_data__events *data = _data;
 
+	data->nr_groups++;
 	parse_events__set_leader($1, list);
 	$$ = list;
 }
@@ -130,7 +132,9 @@ PE_NAME '{' events '}'
 '{' events '}'
 {
 	struct list_head *list = $2;
+	struct parse_events_data__events *data = _data;
 
+	data->nr_groups++;
 	parse_events__set_leader(NULL, list);
 	$$ = list;
 }
Signature · 1 line
1.7.11.4
#7 lore raw

Re: [PATCH 05/16] perf tools: Keep group information

SNIP

Quoted 29 lines
tools/perf/util/parse-events.c +1 -0
index bf5d033ee1b4..3c52d0ab9270 100644
@@ -830,6 +830,7 @@ int parse_events(struct perf_evlist *evlist, const char *str,
 	if (!ret) {
 		int entries = data.idx - evlist->nr_entries;
 		perf_evlist__splice_list_tail(evlist, &data.list, entries);
+		evlist->nr_groups += data.nr_groups;
 		return 0;
 	}
 
tools/perf/util/parse-events.h +1 -0
index c356e443448d..f6b0254afe17 100644
@@ -65,6 +65,7 @@ struct parse_events__term {
 struct parse_events_data__events {
 	struct list_head list;
 	int idx;
+	int nr_groups;
 };
 
 struct parse_events_data__terms {
tools/perf/util/parse-events.y +0 -0
index cd88209e3c58..d14bb507594b 100644

tools/perf/util/parse-events.y group_def: · line 127

Quoted 6 lines
tools/perf/util/parse-events.y +2 -0
@@ -122,7 +122,9 @@ group_def:
 PE_NAME '{' events '}'
 {
 	struct list_head *list = $3;
+	struct parse_events_data__events *data = _data;
 
+	data->nr_groups++;

perhaps if you inc nr_groups only if there's more than 1 event, you would not need your next patch:
perf evlist: Add perf_evlist__recalc_nr_groups

something like:

if (!list_is_last(list))

	data->nr_groups++;

jirka

author maintainer
#8 lore raw

Re: [PATCH 05/16] perf tools: Keep group information

Hi Jiri,

Quoted 40 lines · Jiri Olsa
Quoted 29 lines
tools/perf/util/parse-events.c +1 -0
index bf5d033ee1b4..3c52d0ab9270 100644
@@ -830,6 +830,7 @@ int parse_events(struct perf_evlist *evlist, const char *str,
 	if (!ret) {
 		int entries = data.idx - evlist->nr_entries;
 		perf_evlist__splice_list_tail(evlist, &data.list, entries);
+		evlist->nr_groups += data.nr_groups;
 		return 0;
 	}
 
tools/perf/util/parse-events.h +1 -0
index c356e443448d..f6b0254afe17 100644
@@ -65,6 +65,7 @@ struct parse_events__term {
 struct parse_events_data__events {
 	struct list_head list;
 	int idx;
+	int nr_groups;
 };
 
 struct parse_events_data__terms {
tools/perf/util/parse-events.y +0 -0
index cd88209e3c58..d14bb507594b 100644

tools/perf/util/parse-events.y group_def: · line 127

Quoted 6 lines
tools/perf/util/parse-events.y +2 -0
@@ -122,7 +122,9 @@ group_def:
 PE_NAME '{' events '}'
 {
 	struct list_head *list = $3;
+	struct parse_events_data__events *data = _data;
 
+	data->nr_groups++;

perhaps if you inc nr_groups only if there's more than 1 event, you would not need your next patch:
perf evlist: Add perf_evlist__recalc_nr_groups

Quoted 5 lines

something like:

if (!list_is_last(list))

	data->nr_groups++;

Right! Will use it in next version.

Thanks,
Namhyung

author maintainer
#9 lore raw

PATCH 6/16 perf evlist: Add perf_evlist__recalc_nr_groups

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

During the event parsing, perf_evlist can have leader-only groups which has nr_members as 1. Since they has no difference than a normal non-group event don't count them as a event group.

Add perf_evlist__recalc_nr_groups to count actual group numbers.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/builtin-record.c |  3 +++
 tools/perf/util/evlist.c    | 16 ++++++++++++++++
 tools/perf/util/evlist.h    |  1 +
 3 files changed, 20 insertions(+)
tools/perf/builtin-record.c +3 -0
index 2cb74343de3e..99ad5234e6ff 100644
@@ -212,6 +212,9 @@ static int perf_record__open(struct perf_record *rec)
 	if (opts->group)
 		perf_evlist__set_leader(evlist);
 
+	if (evlist->nr_groups)
+		perf_evlist__recalc_nr_groups(evlist);
+
 	list_for_each_entry(pos, &evlist->entries, node) {
 		struct perf_event_attr *attr = &pos->attr;
 		/*
tools/perf/util/evlist.c +16 -0
index 199b6f1c3b22..0dcc443716b7 100644
@@ -133,6 +133,22 @@ void perf_evlist__set_leader(struct perf_evlist *evlist)
 	}
 }
 
+void perf_evlist__recalc_nr_groups(struct perf_evlist *evlist)
+{
+	int count = 0;
+	struct perf_evsel *evsel;
+
+	list_for_each_entry(evsel, &evlist->entries, node) {
+		/*
+		 * Don't count leader-only groups for simplicity.
+		 */
+		if (perf_evsel__is_group_leader(evsel) &&
+		    evsel->nr_members > 0)
+			count++;
+	}
+	evlist->nr_groups = count;
+}
+
 int perf_evlist__add_default(struct perf_evlist *evlist)
 {
 	struct perf_event_attr attr = {
tools/perf/util/evlist.h +1 -0
index 946a6ada817b..439b1375c4ea 100644
@@ -120,6 +120,7 @@ int perf_evlist__set_filters(struct perf_evlist *evlist);
 
 void __perf_evlist__set_leader(struct list_head *list);
 void perf_evlist__set_leader(struct perf_evlist *evlist);
+void perf_evlist__recalc_nr_groups(struct perf_evlist *evlist);
 
 u64 perf_evlist__sample_type(struct perf_evlist *evlist);
 bool perf_evlist__sample_id_all(struct perf_evlist *evlist);
Signature · 1 line
1.7.11.4
author maintainer
#10 lore raw

PATCH 7/16 perf header: Add HEADER_GROUP_DESC feature

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Save group relationship information so that it can be restored when perf report is running.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/builtin-record.c |   3 +
 tools/perf/util/header.c    | 149 ++++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/header.h    |   2 +
 3 files changed, 154 insertions(+)
tools/perf/builtin-record.c +3 -0
index 99ad5234e6ff..b2d9c0af9a15 100644
@@ -561,6 +561,9 @@ static int __cmd_record(struct perf_record *rec, int argc, const char **argv)
 		goto out_delete_session;
 	}
 
+	if (!evsel_list->nr_groups)
+		perf_header__clear_feat(&session->header, HEADER_GROUP_DESC);
+
 	/*
 	 * perf_session__delete(session) will be called at perf_record__exit()
 	 */
tools/perf/util/header.c +149 -0

173 lines not on this page · raw message

tools/perf/util/header.h +2 -0
index 99bdd3abce59..f143aa7153a2 100644
@@ -29,6 +29,7 @@ enum {
 	HEADER_NUMA_TOPOLOGY,
 	HEADER_BRANCH_STACK,
 	HEADER_PMU_MAPPINGS,
+	HEADER_GROUP_DESC,
 	HEADER_LAST_FEATURE,
 	HEADER_FEAT_BITS	= 256,
 };
@@ -79,6 +80,7 @@ struct perf_session_env {
 	char			*numa_nodes;
 	int			nr_pmu_mappings;
 	char			*pmu_mappings;
+	int			nr_groups;
 };
 
 struct perf_header {
Signature · 1 line
1.7.11.4
author maintainer
#11 lore raw

PATCH 8/16 perf hists: Collapse group hist_entries to a leader

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

To support viewing an event group together, collapse all of members in the group to the leader's tree. The entries in the leaders' tree will have group_stats to store those information.

This patch introduced an additional field 'event_group' in symbol_conf to distinguish whether event grouping is enabled or not.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/util/evsel.h  |   5 +++
 tools/perf/util/hist.c   | 106 +++++++++++++++++++++++++++++++++++++++++++----
 tools/perf/util/sort.h   |   1 +
 tools/perf/util/symbol.c |   4 ++
 tools/perf/util/symbol.h |   3 +-
 5 files changed, 110 insertions(+), 9 deletions(-)
tools/perf/util/evsel.h +5 -0
index 820f005096c4..95f0bf17e79c 100644
@@ -221,4 +221,9 @@ static inline bool perf_evsel__is_group_leader(struct perf_evsel *evsel)
 {
 	return evsel->leader == NULL;
 }
+
+static inline struct perf_evsel *hists_2_evsel(struct hists *hists)
+{
+	return container_of(hists, struct perf_evsel, hists);
+}
 #endif /* __PERF_EVSEL_H */
tools/perf/util/hist.c +98 -8

196 lines not on this page · raw message

tools/perf/util/sort.h +1 -0
index 30f2ae5be146..1780bfc348d2 100644
@@ -62,6 +62,7 @@ struct hist_entry {
 	struct rb_node		rb_node_in;
 	struct rb_node		rb_node;
 	struct he_stat		stat;
+	struct he_stat		*group_stats;
 	struct map_symbol	ms;
 	struct thread		*thread;
 	u64			ip;
tools/perf/util/symbol.c +4 -0
index e2e8c697cffe..bb2ef6920ccf 100644
@@ -11,6 +11,7 @@
 #include <inttypes.h>
 #include "build-id.h"
 #include "util.h"
+#include "sort.h"
 #include "debug.h"
 #include "symbol.h"
 #include "strlist.h"
@@ -2033,6 +2034,9 @@ int symbol__init(void)
 
 	symbol_conf.kptr_restrict = symbol__read_kptr_restrict();
 
+	if (symbol_conf.event_group)
+		sort__need_collapse = 1;
+
 	symbol_conf.initialized = true;
 	return 0;
 
tools/perf/util/symbol.h +2 -1
index b441b07172b7..5f2304acea95 100644
@@ -97,7 +97,8 @@ struct symbol_conf {
 			initialized,
 			kptr_restrict,
 			annotate_asm_raw,
-			annotate_src;
+			annotate_src,
+			event_group;
 	const char	*vmlinux_name,
 			*kallsyms_name,
 			*source_prefix,
Signature · 1 line
1.7.11.4
author maintainer
#12 lore raw

PATCH 9/16 perf hists: Maintain total periods of group members in the leader

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Like group_stats in hist_entry, total periods information also need to be known to the leader.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/util/hist.c | 25 +++++++++++++++++++++++++
 tools/perf/util/hist.h |  1 +
 2 files changed, 26 insertions(+)
tools/perf/util/hist.c +25 -0
index 319822dee77b..8f01fc46ca88 100644
@@ -194,6 +194,28 @@ static void hist_entry__add_group_stat(struct hist_entry *he_dest,
 		he_stat__add_stat(&he_dest->group_stats[evsel->group_idx], src);
 }
 
+static void hists__add_group_stat(struct hists *hists)
+{
+	struct perf_evsel *evsel = hists_2_evsel(hists);
+	struct perf_evsel *leader = evsel->leader;
+	struct hists *leader_hists;
+
+	if (perf_evsel__is_group_leader(evsel))
+		return;
+
+	leader_hists = &leader->hists;
+
+	if (!leader_hists->group_stats) {
+		leader_hists->group_stats = calloc(leader->nr_members,
+						   sizeof(struct events_stats));
+		if (!leader_hists->group_stats)
+			return;
+	}
+
+	memcpy(&leader_hists->group_stats[evsel->group_idx],
+	       &hists->stats, sizeof(struct events_stats));
+}
+
 static void hist_entry__decay(struct hist_entry *he)
 {
 	he->stat.period = (he->stat.period * 7) / 8;
@@ -557,6 +579,9 @@ static void __hists__collapse_resort(struct hists *hists, bool threaded)
 			hists__apply_filters(hists, n);
 		}
 	}
+
+	if (symbol_conf.event_group)
+		hists__add_group_stat(hists);
 }
 
 void hists__collapse_resort(struct hists *hists)
tools/perf/util/hist.h +1 -0
index f011ad4756e8..e13db91bc246 100644
@@ -66,6 +66,7 @@ struct hists {
 	const char		*symbol_filter_str;
 	pthread_mutex_t		lock;
 	struct events_stats	stats;
+	struct events_stats	*group_stats;
 	u64			event_stream;
 	u16			col_len[HISTC_NR_COLS];
 };
Signature · 1 line
1.7.11.4
author maintainer
#13 lore raw

PATCH 10/16 perf report: Make another loop for output resorting

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Now the event grouping viewing requires collapsing all members in a group to the leader. Thus hists__output_resort should be called after collapsing all entries in evlist.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/builtin-report.c | 5 +++++
 1 file changed, 5 insertions(+)
tools/perf/builtin-report.c +5 -0
index 1da243dfbc3e..7729c1c290e6 100644
@@ -416,6 +416,11 @@ static int __cmd_report(struct perf_report *rep)
 			hists->symbol_filter_str = rep->symbol_filter_str;
 
 		hists__collapse_resort(hists);
+	}
+
+	list_for_each_entry(pos, &session->evlist->entries, node) {
+		struct hists *hists = &pos->hists;
+
 		hists__output_resort(hists);
 		nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
 	}
Signature · 1 line
1.7.11.4
author maintainer
#14 lore raw

PATCH 11/16 perf ui/hist: Add support for event group view

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Show group members' overhead also when showing the leader's if event group is enabled. At this time, only implemented overhead part in order to ease review and other parts can be added later once this patch settled down.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/ui/hist.c       | 67 +++++++++++++++++++++++++++++++++++++++++-----
 tools/perf/ui/stdio/hist.c |  2 ++
 tools/perf/util/hist.h     |  1 +
 3 files changed, 63 insertions(+), 7 deletions(-)
tools/perf/ui/hist.c +60 -7
index d6ddeb10e678..b8e97c3577c1 100644
@@ -3,23 +3,40 @@
 #include "../util/hist.h"
 #include "../util/util.h"
 #include "../util/sort.h"
+#include "../util/evsel.h"
 
 
 /* hist period print (hpp) functions */
 static int hpp__header_overhead(struct perf_hpp *hpp)
 {
+	int len = 8;
 	const char *fmt = hpp->ptr ? "Baseline" : "Overhead";
 
-	return scnprintf(hpp->buf, hpp->size, fmt);
+	if (symbol_conf.event_group) {
+		struct perf_evsel *evsel = hists_2_evsel(hpp->hists);
+
+		BUG_ON(!perf_evsel__is_group_leader(evsel));
+
+		len += evsel->nr_members * 8;
+	}
+	return scnprintf(hpp->buf, hpp->size, "%*s", len, fmt);
 }
 
-static int hpp__width_overhead(struct perf_hpp *hpp __maybe_unused)
+static int hpp__width_overhead(struct perf_hpp *hpp)
 {
-	return 8;
+	int len = 8;
+
+	if (symbol_conf.event_group) {
+		struct perf_evsel *evsel = hists_2_evsel(hpp->hists);
+
+		len += evsel->nr_members * 8;
+	}
+	return len;
 }
 
 static int hpp__color_overhead(struct perf_hpp *hpp, struct hist_entry *he)
 {
+	int ret;
 	double percent = 100.0 * he->stat.period / hpp->total_period;
 
 	if (hpp->ptr) {
@@ -33,11 +50,29 @@ static int hpp__color_overhead(struct perf_hpp *hpp, struct hist_entry *he)
 			percent = 0.0;
 	}
 
-	return percent_color_snprintf(hpp->buf, hpp->size, " %6.2f%%", percent);
+	ret = percent_color_snprintf(hpp->buf, hpp->size, " %6.2f%%", percent);
+
+	if (symbol_conf.event_group) {
+		int i;
+		struct perf_evsel *evsel = hists_2_evsel(hpp->hists);
+
+		for (i = 0; i < evsel->nr_members; i++) {
+			u64 period = he->group_stats[i].period;
+			u64 total = hpp->hists->group_stats[i].total_period;
+
+			percent = 100.0 * period / total;
+			ret += percent_color_snprintf(hpp->buf + ret,
+						      hpp->size - ret,
+						      " %6.2f%%", percent);
+		}
+
+	}
+	return ret;
 }
 
 static int hpp__entry_overhead(struct perf_hpp *hpp, struct hist_entry *he)
 {
+	int ret;
 	double percent = 100.0 * he->stat.period / hpp->total_period;
 	const char *fmt = symbol_conf.field_sep ? "%.2f" : " %6.2f%%";
 
@@ -52,13 +87,32 @@ static int hpp__entry_overhead(struct perf_hpp *hpp, struct hist_entry *he)
 			percent = 0.0;
 	}
 
-	return scnprintf(hpp->buf, hpp->size, fmt, percent);
+	ret = scnprintf(hpp->buf, hpp->size, fmt, percent);
+
+	if (symbol_conf.event_group) {
+		int i;
+		struct perf_evsel *evsel = hists_2_evsel(hpp->hists);
+
+		for (i = 0; i < evsel->nr_members; i++) {
+			u64 period = he->group_stats[i].period;
+			u64 total = hpp->hists->group_stats[i].total_period;
+
+			if (symbol_conf.field_sep) {
+				ret += scnprintf(hpp->buf + ret,
+						 hpp->size - ret, " ");
+			}
+			percent = 100.0 * period / total;
+			ret += scnprintf(hpp->buf + ret, hpp->size - ret,
+					 fmt, percent);
+		}
+
+	}
+	return ret;
 }
 
 static int hpp__header_overhead_sys(struct perf_hpp *hpp)
 {
 	const char *fmt = symbol_conf.field_sep ? "%s" : "%7s";
-
 	return scnprintf(hpp->buf, hpp->size, fmt, "sys");
 }
 
@@ -84,7 +138,6 @@ static int hpp__entry_overhead_sys(struct perf_hpp *hpp, struct hist_entry *he)
 static int hpp__header_overhead_us(struct perf_hpp *hpp)
 {
 	const char *fmt = symbol_conf.field_sep ? "%s" : "%7s";
-
 	return scnprintf(hpp->buf, hpp->size, fmt, "user");
 }
 
tools/perf/ui/stdio/hist.c +2 -0
index 4382a1995cda..8c417f036613 100644
@@ -316,6 +316,7 @@ static int hist_entry__fprintf(struct hist_entry *he, size_t size,
 		.buf		= bf,
 		.size		= size,
 		.total_period	= total_period,
+		.hists		= hists,
 		.displacement	= displacement,
 		.ptr		= pair_hists,
 	};
@@ -354,6 +355,7 @@ size_t hists__fprintf(struct hists *hists, struct hists *pair,
 	struct perf_hpp dummy_hpp = {
 		.buf	= bf,
 		.size	= sizeof(bf),
+		.hists	= hists,
 		.ptr	= pair,
 	};
 
tools/perf/util/hist.h +1 -0
index e13db91bc246..a7ccaad65f21 100644
@@ -120,6 +120,7 @@ struct perf_hpp {
 	char *buf;
 	size_t size;
 	u64 total_period;
+	struct hists* hists;
 	const char *sep;
 	long displacement;
 	void *ptr;
Signature · 1 line
1.7.11.4
author maintainer
#15 lore raw

PATCH 12/16 perf ui/browser: Add support for event group view

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Show group members' overhead also when showing the leader's if event group is enabled. At this time, only implemented overhead part in order to ease review and other parts can be added later once this patch settled down.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/ui/browsers/hists.c | 29 ++++++++++++++++++++++++++++-
 tools/perf/ui/hist.c           |  5 ++++-
 2 files changed, 32 insertions(+), 2 deletions(-)
tools/perf/ui/browsers/hists.c +28 -1
index b9b1b173637c..da3d1e4a44fa 100644
@@ -565,6 +565,33 @@ static int hist_browser__show_callchain(struct hist_browser *browser,
 	return row - first_row;
 }
 
+static int hist_browser__hpp_color_overhead(struct perf_hpp *hpp,
+					    struct hist_entry *he)
+{
+	int ret;
+	double percent = 100.0 * he->stat.period / hpp->total_period;
+
+	/* the leader determines color */
+	*(double *) hpp->ptr = percent;
+
+	ret = scnprintf(hpp->buf, hpp->size, "%6.2f%%", percent);
+
+	if (symbol_conf.event_group) {
+		int i;
+		struct perf_evsel *evsel = hists_2_evsel(hpp->hists);
+
+		for (i = 0; i < evsel->nr_members; i++) {
+			u64 period = he->group_stats[i].period;
+			u64 total = hpp->hists->group_stats[i].total_period;
+
+			percent = 100.0 * period / total;
+			ret += scnprintf(hpp->buf + ret, hpp->size - ret,
+					 " %6.2f%%", percent);
+		}
+	}
+	return ret;
+}
+
 #define HPP__COLOR_FN(_name, _field)					\
 static int hist_browser__hpp_color_ ## _name(struct perf_hpp *hpp,	\
 					     struct hist_entry *he)	\
@@ -574,7 +601,6 @@ static int hist_browser__hpp_color_ ## _name(struct perf_hpp *hpp,	\
 	return scnprintf(hpp->buf, hpp->size, "%6.2f%%", percent);	\
 }
 
-HPP__COLOR_FN(overhead, period)
 HPP__COLOR_FN(overhead_sys, period_sys)
 HPP__COLOR_FN(overhead_us, period_us)
 HPP__COLOR_FN(overhead_guest_sys, period_guest_sys)
@@ -625,6 +651,7 @@ static int hist_browser__show_entry(struct hist_browser *browser,
 			.buf		= s,
 			.size		= sizeof(s),
 			.total_period	= browser->hists->stats.total_period,
+			.hists		= browser->hists,
 		};
 
 		ui_browser__gotorc(&browser->b, row, 0);
tools/perf/ui/hist.c +4 -1
index b8e97c3577c1..d8491fcff6ee 100644
@@ -422,6 +422,9 @@ unsigned int hists__sort_list_width(struct hists *hists)
 {
 	struct sort_entry *se;
 	int i, ret = 0;
+	struct perf_hpp dummy_hpp = {
+		.hists	= hists,
+	};
 
 	for (i = 0; i < PERF_HPP__MAX_INDEX; i++) {
 		if (!perf_hpp__format[i].cond)
@@ -429,7 +432,7 @@ unsigned int hists__sort_list_width(struct hists *hists)
 		if (i)
 			ret += 2;
 
-		ret += perf_hpp__format[i].width(NULL);
+		ret += perf_hpp__format[i].width(&dummy_hpp);
 	}
 
 	list_for_each_entry(se, &hist_entry__sort_list, list)
Signature · 1 line
1.7.11.4
author maintainer
#16 lore raw

PATCH 13/16 perf ui/gtk: Add support for event group view

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Show group members' overhead also when showing the leader's if event group is enabled. At this time, only implemented overhead part in order to ease review and other parts can be added later once this patch settled down.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
Cc Pekka Enberg <[email protected]>
---
 tools/perf/ui/gtk/browser.c | 58 +++++++++++++++++++++++++++++++++++----------
 1 file changed, 45 insertions(+), 13 deletions(-)
tools/perf/ui/gtk/browser.c +45 -13
index 7107edc6c08d..775f209fe795 100644
@@ -45,25 +45,56 @@ static const char *perf_gtk__get_percent_color(double percent)
 	return NULL;
 }
 
+static int perf_gtk__percent_color_snprintf(char *buf, size_t size,
+					    u64 period, u64 total_period)
+{
+	int ret = 0;
+	const char *markup;
+	double percent = 100.0 * period / total_period;
+
+	markup = perf_gtk__get_percent_color(percent);
+	if (markup)
+		ret += scnprintf(buf, size, markup);
+
+	ret += scnprintf(buf + ret, size - ret, "%6.2f%%", percent);
+
+	if (markup)
+		ret += scnprintf(buf + ret, size - ret, "</span>");
+
+	return ret;
+}
+
+static int perf_gtk__hpp_color_overhead(struct perf_hpp *hpp,
+					struct hist_entry *he)
+{
+	int ret;
+
+	ret = perf_gtk__percent_color_snprintf(hpp->buf, hpp->size,
+				       he->stat.period, hpp->total_period);
+
+	if (symbol_conf.event_group) {
+		int i;
+		struct perf_evsel *evsel = hists_2_evsel(hpp->hists);
+
+		for (i = 0; i < evsel->nr_members; i++) {
+			ret += scnprintf(hpp->buf + ret, hpp->size - ret, " ");
+			ret += perf_gtk__percent_color_snprintf(hpp->buf + ret,
+					hpp->size - ret,
+					he->group_stats[i].period,
+					hpp->hists->group_stats[i].total_period);
+		}
+	}
+	return ret;
+}
+
 #define HPP__COLOR_FN(_name, _field)						\
 static int perf_gtk__hpp_color_ ## _name(struct perf_hpp *hpp,			\
 					 struct hist_entry *he)			\
 {										\
-	double percent = 100.0 * he->stat._field / hpp->total_period;		\
-	const char *markup;							\
-	int ret = 0;								\
-										\
-	markup = perf_gtk__get_percent_color(percent);				\
-	if (markup)								\
-		ret += scnprintf(hpp->buf, hpp->size, "%s", markup);		\
-	ret += scnprintf(hpp->buf + ret, hpp->size - ret, "%6.2f%%", percent); 	\
-	if (markup)								\
-		ret += scnprintf(hpp->buf + ret, hpp->size - ret, "</span>"); 	\
-										\
-	return ret;								\
+	return perf_gtk__percent_color_snprintf(hpp->buf, hpp->size, 		\
+					he->stat._field, hpp->total_period);	\
 }
 
-HPP__COLOR_FN(overhead, period)
 HPP__COLOR_FN(overhead_sys, period_sys)
 HPP__COLOR_FN(overhead_us, period_us)
 HPP__COLOR_FN(overhead_guest_sys, period_guest_sys)
@@ -103,6 +134,7 @@ static void perf_gtk__show_hists(GtkWidget *window, struct hists *hists)
 		.buf		= s,
 		.size		= sizeof(s),
 		.total_period	= hists->stats.total_period,
+		.hists		= hists,
 	};
 
 	nr_cols = 0;
Signature · 1 line
1.7.11.4
author maintainer
#17 lore raw

PATCH 14/16 perf report: Bypass non-leader events when event group is enabled

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Since we have all necessary information in the leader events and other members don't, bypass members. Member events will be shown along with the leaders if event group is enabled.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
Cc Pekka Enberg <[email protected]>
---
 tools/perf/builtin-report.c    |  4 ++++
 tools/perf/ui/browsers/hists.c | 39 +++++++++++++++++++++++++++++++++------
 tools/perf/ui/gtk/browser.c    |  4 ++++
 3 files changed, 41 insertions(+), 6 deletions(-)
tools/perf/builtin-report.c +4 -0
index 7729c1c290e6..52803a9d3e3e 100644
@@ -319,6 +319,10 @@ static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
 		struct hists *hists = &pos->hists;
 		const char *evname = perf_evsel__name(pos);
 
+		if (symbol_conf.event_group &&
+		    !perf_evsel__is_group_leader(pos))
+			continue;
+
 		hists__fprintf_nr_sample_events(hists, evname, stdout);
 		hists__fprintf(hists, NULL, false, true, 0, 0, stdout);
 		fprintf(stdout, "\n\n");
tools/perf/ui/browsers/hists.c +33 -6
index da3d1e4a44fa..82d59be3c230 100644
@@ -1576,8 +1576,19 @@ out:
 	return key;
 }
 
+static bool filter_group_entries(struct ui_browser *self __maybe_unused,
+				 void *entry)
+{
+	struct perf_evsel *evsel = list_entry(entry, struct perf_evsel, node);
+
+	if (symbol_conf.event_group && !perf_evsel__is_group_leader(evsel))
+		return true;
+
+	return false;
+}
+
 static int __perf_evlist__tui_browse_hists(struct perf_evlist *evlist,
-					   const char *help,
+					   int nr_entries, const char *help,
 					   void(*timer)(void *arg), void *arg,
 					   int delay_secs)
 {
@@ -1588,7 +1599,8 @@ static int __perf_evlist__tui_browse_hists(struct perf_evlist *evlist,
 			.refresh    = ui_browser__list_head_refresh,
 			.seek	    = ui_browser__list_head_seek,
 			.write	    = perf_evsel_menu__write,
-			.nr_entries = evlist->nr_entries,
+			.filter	    = filter_group_entries,
+			.nr_entries = nr_entries,
 			.priv	    = evlist,
 		},
 	};
@@ -1603,7 +1615,7 @@ static int __perf_evlist__tui_browse_hists(struct perf_evlist *evlist,
 			menu.b.width = line_len;
 	}
 
-	return perf_evsel_menu__run(&menu, evlist->nr_entries, help, timer,
+	return perf_evsel_menu__run(&menu, nr_entries, help, timer,
 				    arg, delay_secs);
 }
 
@@ -1611,15 +1623,30 @@ int perf_evlist__tui_browse_hists(struct perf_evlist *evlist, const char *help,
 				  void(*timer)(void *arg), void *arg,
 				  int delay_secs)
 {
-	if (evlist->nr_entries == 1) {
+	int nr_entries = evlist->nr_entries;
+
+single_entry:
+	if (nr_entries == 1) {
 		struct perf_evsel *first = list_entry(evlist->entries.next,
 						      struct perf_evsel, node);
 		const char *ev_name = perf_evsel__name(first);
-		return perf_evsel__hists_browse(first, evlist->nr_entries, help,
+		return perf_evsel__hists_browse(first, nr_entries, help,
 						ev_name, false, timer, arg,
 						delay_secs);
 	}
 
-	return __perf_evlist__tui_browse_hists(evlist, help,
+	if (symbol_conf.event_group) {
+		struct perf_evsel *pos;
+
+		nr_entries = 0;
+		list_for_each_entry(pos, &evlist->entries, node)
+			if (perf_evsel__is_group_leader(pos))
+				nr_entries++;
+
+		if (nr_entries == 1)
+			goto single_entry;
+	}
+
+	return __perf_evlist__tui_browse_hists(evlist, nr_entries, help,
 					       timer, arg, delay_secs);
 }
tools/perf/ui/gtk/browser.c +4 -0
index 775f209fe795..01bfa43db8e2 100644
@@ -306,6 +306,10 @@ int perf_evlist__gtk_browse_hists(struct perf_evlist *evlist,
 		GtkWidget *scrolled_window;
 		GtkWidget *tab_label;
 
+		if (symbol_conf.event_group &&
+		    !perf_evsel__is_group_leader(pos))
+			continue;
+
 		scrolled_window = gtk_scrolled_window_new(NULL, NULL);
 
 		gtk_scrolled_window_set_policy(GTK_SCROLLED_WINDOW(scrolled_window),
Signature · 1 line
1.7.11.4
author maintainer
#18 lore raw

PATCH 15/16 perf report: Show group description when event group is enabled

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

When using event group viewer, it's better to show the group description rather than the leader information alone.

If a leader did not contain any member, it's a non-group event.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
Cc Pekka Enberg <[email protected]>
---
 tools/perf/builtin-report.c    | 18 ++++++++++++++++++
 tools/perf/ui/browsers/hists.c | 31 +++++++++++++++++++++++++++++++
 tools/perf/ui/gtk/browser.c    | 14 +++++++++++---
 tools/perf/util/evsel.c        | 25 +++++++++++++++++++++++++
 tools/perf/util/evsel.h        |  8 ++++++++
 5 files changed, 93 insertions(+), 3 deletions(-)
tools/perf/builtin-report.c +18 -0
index 52803a9d3e3e..3a4bd13340db 100644
@@ -299,6 +299,24 @@ static size_t hists__fprintf_nr_sample_events(struct hists *self,
 	char unit;
 	unsigned long nr_samples = self->stats.nr_events[PERF_RECORD_SAMPLE];
 	u64 nr_events = self->stats.total_period;
+	struct perf_evsel *evsel = hists_2_evsel(self);
+	char buf[512];
+	size_t size = sizeof(buf);
+
+	if (symbol_conf.event_group && evsel->nr_members) {
+		int i;
+		struct events_stats *stats;
+
+		perf_evsel__group_desc(evsel, buf, size);
+		evname = buf;
+
+		for (i = 0; i < evsel->nr_members; i++) {
+			stats = &self->group_stats[i];
+
+			nr_samples += stats->nr_events[PERF_RECORD_SAMPLE];
+			nr_events += stats->total_period;
+		}
+	}
 
 	nr_samples = convert_unit(nr_samples, &unit);
 	ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
tools/perf/ui/browsers/hists.c +31 -0
index 82d59be3c230..7903b98913c5 100644
@@ -1121,6 +1121,24 @@ static int hists__browser_title(struct hists *hists, char *bf, size_t size,
 	const struct thread *thread = hists->thread_filter;
 	unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
 	u64 nr_events = hists->stats.total_period;
+	struct perf_evsel *evsel = hists_2_evsel(hists);
+	char buf[512];
+	size_t buflen = sizeof(buf);
+
+	if (symbol_conf.event_group && evsel->nr_members) {
+		int i;
+		struct events_stats *stats;
+
+		perf_evsel__group_desc(evsel, buf, buflen);
+		ev_name = buf;
+
+		for (i = 0; i < evsel->nr_members; i++) {
+			stats = &hists->group_stats[i];
+
+			nr_samples += stats->nr_events[PERF_RECORD_SAMPLE];
+			nr_events += stats->total_period;
+		}
+	}
 
 	nr_samples = convert_unit(nr_samples, &unit);
 	printed = scnprintf(bf, size,
@@ -1467,6 +1485,19 @@ static void perf_evsel_menu__write(struct ui_browser *browser,
 	ui_browser__set_color(browser, current_entry ? HE_COLORSET_SELECTED :
 						       HE_COLORSET_NORMAL);
 
+	if (symbol_conf.event_group && evsel->nr_members) {
+		int i;
+		struct events_stats *stats;
+
+		ev_name = perf_evsel__group_name(evsel);
+
+		for (i = 0; i < evsel->nr_members; i++) {
+			stats = &evsel->hists.group_stats[i];
+
+			nr_events += stats->nr_events[PERF_RECORD_SAMPLE];
+		}
+	}
+
 	nr_events = convert_unit(nr_events, &unit);
 	printed = scnprintf(bf, sizeof(bf), "%lu%c%s%s", nr_events,
 			   unit, unit == ' ' ? "" : " ", ev_name);
tools/perf/ui/gtk/browser.c +11 -3
index 01bfa43db8e2..cb6241e7f9a5 100644
@@ -305,10 +305,18 @@ int perf_evlist__gtk_browse_hists(struct perf_evlist *evlist,
 		const char *evname = perf_evsel__name(pos);
 		GtkWidget *scrolled_window;
 		GtkWidget *tab_label;
+		char buf[512];
+		size_t size = sizeof(buf);
 
-		if (symbol_conf.event_group &&
-		    !perf_evsel__is_group_leader(pos))
-			continue;
+		if (symbol_conf.event_group) {
+			if (!perf_evsel__is_group_leader(pos))
+				continue;
+
+			if (pos->nr_members) {
+				perf_evsel__group_desc(pos, buf, size);
+				evname = buf;
+			}
+		}
 
 		scrolled_window = gtk_scrolled_window_new(NULL, NULL);
 
tools/perf/util/evsel.c +25 -0
index 00936ad29ff2..50fa1b4c0d8b 100644
@@ -397,6 +397,31 @@ const char *perf_evsel__name(struct perf_evsel *evsel)
 	return evsel->name ?: "unknown";
 }
 
+const char *perf_evsel__group_name(struct perf_evsel *evsel)
+{
+	return evsel->group_name ?: "anon group";
+}
+
+int perf_evsel__group_desc(struct perf_evsel *evsel, char *buf, size_t size)
+{
+	int ret;
+	struct perf_evsel *pos;
+	const char *group_name = perf_evsel__group_name(evsel);
+
+	ret = scnprintf(buf, size, "%s", group_name);
+
+	ret += scnprintf(buf + ret, size - ret, " { %s",
+			 perf_evsel__name(evsel));
+
+	for_each_group_member(pos, evsel)
+		ret += scnprintf(buf + ret, size - ret, ", %s",
+				 perf_evsel__name(pos));
+
+	ret += scnprintf(buf + ret, size - ret, " }");
+
+	return ret;
+}
+
 void perf_evsel__config(struct perf_evsel *evsel, struct perf_record_opts *opts,
 			struct perf_evsel *first)
 {
tools/perf/util/evsel.h +8 -0
index 95f0bf17e79c..1fa79a0d9eb2 100644
@@ -110,6 +110,8 @@ extern const char *perf_evsel__sw_names[PERF_COUNT_SW_MAX];
 int __perf_evsel__hw_cache_type_op_res_name(u8 type, u8 op, u8 result,
 					    char *bf, size_t size);
 const char *perf_evsel__name(struct perf_evsel *evsel);
+const char *perf_evsel__group_name(struct perf_evsel *evsel);
+int perf_evsel__group_desc(struct perf_evsel *evsel, char *buf, size_t size);
 
 int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads);
 int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads);
@@ -226,4 +228,10 @@ static inline struct perf_evsel *hists_2_evsel(struct hists *hists)
 {
 	return container_of(hists, struct perf_evsel, hists);
 }
+
+#define for_each_group_member(_evsel, _leader) 					\
+for ((_evsel) = list_entry((_leader)->node.next, struct perf_evsel, node); 	\
+     (_evsel) && (_evsel)->leader == (_leader);					\
+     (_evsel) = list_entry((_evsel)->node.next, struct perf_evsel, node))
+
 #endif /* __PERF_EVSEL_H */
Signature · 1 line
1.7.11.4
author maintainer
#19 lore raw

PATCH 16/16 perf report: Add --group option

In reply to Namhyung Kim

From: Namhyung Kim <[email protected]>

Add --group option to enable event grouping. When enabled, all the group members information will be shown together with the leader.

Cc Jiri Olsa <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/builtin-report.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/builtin-report.c +2 -0
index 3a4bd13340db..034eec9f09f5 100644
@@ -664,6 +664,8 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
 		    "Show a column with the sum of periods"),
+	OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
+		    "Show event group information together"),
 	OPT_CALLBACK_NOOPT('b', "branch-stack", &sort__branch_mode, "",
 		    "use branch records for histogram filling", parse_branch_mode),
 	OPT_STRING(0, "objdump", &objdump_path, "path",
Signature · 1 line
1.7.11.4