perf, tools: Refactor and support interval and CSV metrics

Fixed even more last feedback.

[v5: Fix mainly bisect problems. No regressions introduced by one patch and fixed again later. Some minor fixes in addition]
[v6: Fix running/noise printing patch.]
[v7: Reorder and merge two patches to avoid a bisect hole where unsupported was printed as 0]
[v8: Minor fixes for review feedback. See changelog in patches.] [v9: Fix newline bug. Add support for -A for --metric-only]
[v10: Remove extra "noise" printing (Jiri)
Fix fields in documentation (Jiri)]
[v11: Fix manpage again. Avoid extra metric output in CSV mode.] [v12: Move CSV metrics fields to after running/enabled/variance.
Minor fixes.]
[v13: Address review comments. Now probe for stalled events
in advance to avoid empty columns or lines. Fix -A shadowing. Various minor changes. Drop merged patches.]
[v14: Fix empty lines with CSV metrics. Avoid one more empty column
in metric-only.]
[v15: Add missing fields in manpage. Use extra init function
for frontend event. Various smaller fixes. Add acked-by.]
[v16: Fix manpage again. Merge --per-thread crash patch.

      Add comments to --metric-only.]

[v17: Man page fixes. Simplify -A --metric-only nrcpus computation.]

Currently perf stat does not support printing computed metrics for interval (-I xxx) or CSV (-x,) mode. For example IPC or TSX metrics over time are quite useful to know.

This patch implements them. The main obstacle was that the
metrics printing was all open coded all over the metrics computation code. The second patch refactors the metrics printing to work through call backs that can be more easily changed. This also cleans up the metrics printing significantly. The indentation is now handled through printf, no more need to manually count spaces.

Then based on that it implements metrics printing for CSV and interval mode, and finally a --metric-only mode.

Example output:

% perf stat -I1000 -a sleep 1

#          time              counts unit events                    metric                              multiplex
     1.001301370       12020.049593      task-clock (msec)                                             (100.00%)
     1.001301370              3,952      context-switches          #    0.329 K/sec                    (100.00%)
     1.001301370                 69      cpu-migrations            #    0.006 K/sec                    (100.00%)
     1.001301370                 76      page-faults               #    0.006 K/sec                  
     1.001301370        386,582,789      cycles                    #    0.032 GHz                      (100.00%)
     1.001301370        716,441,544      stalled-cycles-frontend   #  185.33% frontend cycles idle     (100.00%)
     1.001301370    <not supported>      stalled-cycles-backend   
     1.001301370        101,751,678      instructions              #    0.26  insn per cycle         
     1.001301370                                                   #    7.04  stalled cycles per insn  (100.00%)
     1.001301370         20,914,692      branches                  #    1.740 M/sec                    (100.00%)
     1.001301370          1,943,630      branch-misses             #    9.29% of all branches        

CSV mode:

% perf stat -x, -I1000 -a sleep 1

     1.000982778,12006.549977,,task-clock,12006547787,100.00,,,,
     1.000982778,12822,,context-switches,12007100604,100.00,0.001,M/sec
     1.000982778,175,,cpu-migrations,12007180306,100.00,0.015,K/sec
     1.000982778,3404,,page-faults,12007185482,100.00,0.284,K/sec
     1.000982778,1930307489,,cycles,12007018233,100.00,0.161,GHz
     1.000982778,6971803638,,stalled-cycles-frontend,12006902870,100.00,361.18,frontend cycles idle
     1.000982778,464493941,,instructions,12006873327,100.00,0.24,insn per cycle
     1.000982778,,,,,,15.01,stalled cycles per insn
     1.000982778,86548409,,branches,12006758420,100.00,7.208,M/sec
     1.000982778,4933638,,branch-misses,12006648104,100.00,5.70,of all branches

Now includes metrics

Metric only mode:

Concicse information if you only care about computed metrics, not raw values

% perf stat --metric-only -a -I 1000

         1.001452803 frontend cycles idle insn per cycle       stalled cycles per insn branch-misses of all branches
         1.001452803  158.91%               0.66                2.39                    2.92%
         2.002192321  180.63%               0.76                2.08                    2.96%
         3.003088282  150.59%               0.62                2.57                    2.84%
         4.004369835  196.20%               0.98                1.62                    3.79%
         5.005227314  231.98%               0.84                1.90                    4.71%

Metric only mode in CSV (flat format, easy to plot and analyze in statistical tools like JMP, R, pandas, gnuplot):

% perf stat -x, --metric-only -a -I 1000

     1.001381652,frontend cycles idle,insn per cycle,stalled cycles per insn,branch-misses of all branches,
     1.001381652,173.32,0.83,2.09,1.73,
     2.002073343,199.47,1.07,1.60,2.14,
     3.002875524,109.52,0.22,7.83,1.63,
     4.003970059,132.10,0.17,10.85,1.51,
     5.004818754,181.60,0.22,8.87,2.22,

Available in
git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux-misc-2.6 perf/stat-metrics-21

From: Andi Kleen <[email protected]>

Only put the frontend/backend stalled cycles into the default perf stat events when the CPU actually supports them.

This avoids empty columns with --metric-only on newer Intel CPUs.

---
 tools/perf/builtin-stat.c | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)

index 8c0bc0f..24f222d 100644
@@ -1441,7 +1441,7 @@ static int perf_stat_init_aggr_mode_file(struct perf_stat *st)
  */
 static int add_default_attributes(void)
 {
-	struct perf_event_attr default_attrs[] = {
+	struct perf_event_attr default_attrs0[] = {
 
   { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_TASK_CLOCK		},
   { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CONTEXT_SWITCHES	},

@@ -1449,8 +1449,14 @@ static int add_default_attributes(void)
   { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_PAGE_FAULTS		},
 
   { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_CPU_CYCLES		},
+};
+	struct perf_event_attr frontend_attrs[] = {
   { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_STALLED_CYCLES_FRONTEND	},
+};
+	struct perf_event_attr backend_attrs[] = {
   { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_STALLED_CYCLES_BACKEND	},
+};
+	struct perf_event_attr default_attrs1[] = {
   { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_INSTRUCTIONS		},
   { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS	},
   { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_BRANCH_MISSES		},

@@ -1567,7 +1573,19 @@ static int add_default_attributes(void)
 	}
 
 	if (!evsel_list->nr_entries) {
-		if (perf_evlist__add_default_attrs(evsel_list, default_attrs) < 0)
+		if (perf_evlist__add_default_attrs(evsel_list, default_attrs0) < 0)
+			return -1;
+		if (pmu_have_event("cpu", "stalled-cycles-frontend")) {
+			if (perf_evlist__add_default_attrs(evsel_list,
+						frontend_attrs) < 0)
+				return -1;
+		}
+		if (pmu_have_event("cpu", "stalled-cycles-backend")) {
+			if (perf_evlist__add_default_attrs(evsel_list,
+						backend_attrs) < 0)
+				return -1;
+		}
+		if (perf_evlist__add_default_attrs(evsel_list, default_attrs1) < 0)
 			return -1;
 	}
 

2.5.0

From: Andi Kleen <[email protected]>

Now support CSV output for metrics. With the new output callbacks this is relatively straight forward by creating new callbacks.

This allows to easily plot metrics from CSV files.

The new line callback needs to know the number of fields to skip them correctly

Example output before:

% perf stat -x, true
0.200687,,task-clock,200687,100.00
0,,context-switches,200687,100.00
0,,cpu-migrations,200687,100.00
40,,page-faults,200687,100.00
730871,,cycles,203601,100.00
551056,,stalled-cycles-frontend,203601,100.00
<not supported>,,stalled-cycles-backend,0,100.00
385523,,instructions,203601,100.00
78028,,branches,203601,100.00
3946,,branch-misses,203601,100.00

After:

% perf stat -x, true .502457,,task-clock,502457,100.00,0.485,CPUs utilized 0,,context-switches,502457,100.00,0.000,K/sec 0,,cpu-migrations,502457,100.00,0.000,K/sec 45,,page-faults,502457,100.00,0.090,M/sec 644692,,cycles,509102,100.00,1.283,GHz 423470,,stalled-cycles-frontend,509102,100.00,65.69,frontend cycles idle <not supported>,,stalled-cycles-backend,0,100.00,,,, 492701,,instructions,509102,100.00,0.76,insn per cycle ,,,,,0.86,stalled cycles per insn 97767,,branches,509102,100.00,194.578,M/sec 4788,,branch-misses,509102,100.00,4.90,of all branches

or easier readable

perf stat -x, -o x.csv true
[ak@tassilo hle]$ column -s, -t x.csv

0.490635                                 task-clock               490635  100.00  0.489    CPUs utilized
0                                        context-switches         490635  100.00  0.000    K/sec
0                                        cpu-migrations           490635  100.00  0.000    K/sec
45                                       page-faults              490635  100.00  0.092    M/sec
629080                                   cycles                   497698  100.00  1.282    GHz
409498                                   stalled-cycles-frontend  497698  100.00  65.09    frontend cycles idle
<not supported>                          stalled-cycles-backend   0       100.00
491424                                   instructions             497698  100.00  0.78     insn per cycle
                                                                                  0.83     stalled cycles per insn
97278                                    branches                 497698  100.00  198.270  M/sec
4569                                     branch-misses            497698  100.00  4.70     of all branches

Two new fields are added: metric value and metric name.

v2: Split out function argument changes
v3: Reenable metrics for real.
v4: Fix wrong hunk from refactoring.
v5: Remove extra "noise" printing (Jiri), but add it to the not counted case. Print empty metrics for not counted.
v6: Avoid outputting metric on empty format.
v7: Print metric at the end
v8: Remove extra run, ena fields
v9: Avoid extra new line for unsupported counters

---
 tools/perf/builtin-stat.c     | 73 ++++++++++++++++++++++++++++++++++++++++---
 tools/perf/util/stat-shadow.c |  2 +-
 2 files changed, 70 insertions(+), 5 deletions(-)

index 24f222d..2ffb822 100644
@@ -739,6 +739,7 @@ struct outstate {
 	FILE *fh;
 	bool newline;
 	const char *prefix;
+	int  nfields;
 };
 
 #define METRIC_LEN  35

@@ -789,6 +790,43 @@ static void print_metric_std(void *ctx, const char *color, const char *fmt,
 	fprintf(out, " %-*s", METRIC_LEN - n - 1, unit);
 }
 
+static void new_line_csv(void *ctx)
+{
+	struct outstate *os = ctx;
+	int i;
+
+	fputc('\n', os->fh);
+	if (os->prefix)
+		fprintf(os->fh, "%s%s", os->prefix, csv_sep);
+	for (i = 0; i < os->nfields; i++)
+		fputs(csv_sep, os->fh);
+}
+
+static void print_metric_csv(void *ctx,
+			     const char *color __maybe_unused,
+			     const char *fmt, const char *unit, double val)
+{
+	struct outstate *os = ctx;
+	FILE *out = os->fh;
+	char buf[64], *vals, *ends;
+
+	if (unit == NULL || fmt == NULL) {
+		fprintf(out, "%s%s%s%s", csv_sep, csv_sep, csv_sep, csv_sep);
+		return;
+	}
+	snprintf(buf, sizeof(buf), fmt, val);
+	vals = buf;
+	while (isspace(*vals))
+		vals++;
+	ends = vals;
+	while (isdigit(*ends) || *ends == '.')
+		ends++;
+	*ends = 0;
+	while (isspace(*unit))
+		unit++;
+	fprintf(out, "%s%s%s%s", csv_sep, vals, csv_sep, unit);
+}
+
 static void nsec_printout(int id, int nr, struct perf_evsel *evsel, double avg)
 {
 	FILE *output = stat_config.output;

@@ -860,6 +898,22 @@ static void printout(int id, int nr, struct perf_evsel *counter, double uval,
 
 	nl = new_line_std;
 
+	if (csv_output) {
+		static int aggr_fields[] = {
+			[AGGR_GLOBAL] = 0,
+			[AGGR_THREAD] = 1,
+			[AGGR_NONE] = 1,
+			[AGGR_SOCKET] = 2,
+			[AGGR_CORE] = 2,
+		};
+
+		pm = print_metric_csv;
+		nl = new_line_csv;
+		os.nfields = 3;
+		os.nfields += aggr_fields[stat_config.aggr_mode];
+		if (counter->cgrp)
+			os.nfields++;
+	}
 	if (run == 0 || ena == 0 || counter->counts->scaled == -1) {
 		aggr_printout(counter, id, nr);
 

@@ -880,7 +934,12 @@ static void printout(int id, int nr, struct perf_evsel *counter, double uval,
 			fprintf(stat_config.output, "%s%s",
 				csv_sep, counter->cgrp->name);
 
+		if (!csv_output)
+			pm(&os, NULL, NULL, "", 0);
+		print_noise(counter, noise);
 		print_running(run, ena);
+		if (csv_output)
+			pm(&os, NULL, NULL, "", 0);
 		return;
 	}
 

@@ -893,14 +952,20 @@ static void printout(int id, int nr, struct perf_evsel *counter, double uval,
 	out.new_line = nl;
 	out.ctx = &os;
 
-	if (!csv_output)
-		perf_stat__print_shadow_stats(counter, uval,
+	if (csv_output) {
+		print_noise(counter, noise);
+		print_running(run, ena);
+	}
+
+	perf_stat__print_shadow_stats(counter, uval,
 				stat_config.aggr_mode == AGGR_GLOBAL ? 0 :
 				cpu_map__id_to_cpu(id),
 				&out);
 
-	print_noise(counter, noise);
-	print_running(run, ena);
+	if (!csv_output) {
+		print_noise(counter, noise);
+		print_running(run, ena);
+	}
 }
 
 static void print_aggr(char *prefix)

index 4d8f185..367e220 100644
@@ -310,8 +310,8 @@ void perf_stat__print_shadow_stats(struct perf_evsel *evsel,
 		total = avg_stats(&runtime_stalled_cycles_front_stats[ctx][cpu]);
 		total = max(total, avg_stats(&runtime_stalled_cycles_back_stats[ctx][cpu]));
 
-		out->new_line(ctxp);
 		if (total && avg) {
+			out->new_line(ctxp);
 			ratio = total / avg;
 			print_metric(ctxp, NULL, "%7.2f ",
 					"stalled cycles per insn",

2.5.0

From: Andi Kleen <[email protected]>

Enable metrics printing in --per-core / --per-socket mode. We need to save the shadow metrics in a unique place. Always use the first CPU in the aggregation. Then use the same CPU to retrieve the shadow value later.

Example output:

% perf stat --per-core -a ./BC1s

Performance counter stats for 'system wide':

S0-C0           2        2966.020381      task-clock (msec)         #    2.004 CPUs utilized            (100.00%)
S0-C0           2                 49      context-switches          #    0.017 K/sec                    (100.00%)
S0-C0           2                  4      cpu-migrations            #    0.001 K/sec                    (100.00%)
S0-C0           2                467      page-faults               #    0.157 K/sec
S0-C0           2      4,599,061,773      cycles                    #    1.551 GHz                      (100.00%)
S0-C0           2      9,755,886,883      instructions              #    2.12  insn per cycle           (100.00%)
S0-C0           2      1,906,272,125      branches                  #  642.704 M/sec                    (100.00%)
S0-C0           2         81,180,867      branch-misses             #    4.26% of all branches
S0-C1           2        2965.995373      task-clock (msec)         #    2.003 CPUs utilized            (100.00%)
S0-C1           2                 62      context-switches          #    0.021 K/sec                    (100.00%)
S0-C1           2                  8      cpu-migrations            #    0.003 K/sec                    (100.00%)
S0-C1           2                281      page-faults               #    0.095 K/sec
S0-C1           2          6,347,290      cycles                    #    0.002 GHz                      (100.00%)
S0-C1           2          4,654,156      instructions              #    0.73  insn per cycle           (100.00%)
S0-C1           2            947,121      branches                  #    0.319 M/sec                    (100.00%)
S0-C1           2             37,322      branch-misses             #    3.94% of all branches

       1.480409747 seconds time elapsed

v2: Rebase to older patches
v3: Document shadow cpus. Fix aggr_get_id argument. Fix -A shadows (Jiri)
v4: Fix crash with --per-thread

---
 tools/perf/builtin-stat.c     | 64 +++++++++++++++++++++++++++++++++++++------
 tools/perf/util/stat-shadow.c |  7 +++++
 2 files changed, 63 insertions(+), 8 deletions(-)

index 2ffb822..9b5089c 100644
@@ -740,6 +740,8 @@ struct outstate {
 	bool newline;
 	const char *prefix;
 	int  nfields;
+	int  id, nr;
+	struct perf_evsel *evsel;
 };
 
 #define METRIC_LEN  35

@@ -755,12 +757,9 @@ static void do_new_line_std(struct outstate *os)
 {
 	fputc('\n', os->fh);
 	fputs(os->prefix, os->fh);
+	aggr_printout(os->evsel, os->id, os->nr);
 	if (stat_config.aggr_mode == AGGR_NONE)
 		fprintf(os->fh, "        ");
-	if (stat_config.aggr_mode == AGGR_CORE)
-		fprintf(os->fh, "                  ");
-	if (stat_config.aggr_mode == AGGR_SOCKET)
-		fprintf(os->fh, "            ");
 	fprintf(os->fh, "                                                 ");
 }
 

@@ -798,6 +797,7 @@ static void new_line_csv(void *ctx)
 	fputc('\n', os->fh);
 	if (os->prefix)
 		fprintf(os->fh, "%s%s", os->prefix, csv_sep);
+	aggr_printout(os->evsel, os->id, os->nr);
 	for (i = 0; i < os->nfields; i++)
 		fputs(csv_sep, os->fh);
 }

@@ -855,6 +855,28 @@ static void nsec_printout(int id, int nr, struct perf_evsel *evsel, double avg)
 		fprintf(output, "%s%s", csv_sep, evsel->cgrp->name);
 }
 
+static int first_shadow_cpu(struct perf_evsel *evsel, int id)
+{
+	int i;
+
+	if (!aggr_get_id)
+		return 0;
+
+	if (stat_config.aggr_mode == AGGR_NONE)
+		return id;
+
+	if (stat_config.aggr_mode == AGGR_GLOBAL)
+		return 0;
+
+	for (i = 0; i < perf_evsel__nr_cpus(evsel); i++) {
+		int cpu2 = perf_evsel__cpus(evsel)->map[i];
+
+		if (aggr_get_id(evsel_list->cpus, cpu2) == id)
+			return cpu2;
+	}
+	return 0;
+}
+
 static void abs_printout(int id, int nr, struct perf_evsel *evsel, double avg)
 {
 	FILE *output = stat_config.output;

@@ -891,7 +913,10 @@ static void printout(int id, int nr, struct perf_evsel *counter, double uval,
 	struct perf_stat_output_ctx out;
 	struct outstate os = {
 		.fh = stat_config.output,
-		.prefix = prefix ? prefix : ""
+		.prefix = prefix ? prefix : "",
+		.id = id,
+		.nr = nr,
+		.evsel = counter,
 	};
 	print_metric_t pm = print_metric_std;
 	void (*nl)(void *);

@@ -958,16 +983,37 @@ static void printout(int id, int nr, struct perf_evsel *counter, double uval,
 	}
 
 	perf_stat__print_shadow_stats(counter, uval,
-				stat_config.aggr_mode == AGGR_GLOBAL ? 0 :
-				cpu_map__id_to_cpu(id),
+				first_shadow_cpu(counter, id),
 				&out);
-
 	if (!csv_output) {
 		print_noise(counter, noise);
 		print_running(run, ena);
 	}
 }
 
+static void aggr_update_shadow(void)
+{
+	int cpu, s2, id, s;
+	u64 val;
+	struct perf_evsel *counter;
+
+	for (s = 0; s < aggr_map->nr; s++) {
+		id = aggr_map->map[s];
+		evlist__for_each(evsel_list, counter) {
+			val = 0;
+			for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
+				s2 = aggr_get_id(evsel_list->cpus, cpu);
+				if (s2 != id)
+					continue;
+				val += perf_counts(counter->counts, cpu, 0)->val;
+			}
+			val = val * counter->scale;
+			perf_stat__update_shadow_stats(counter, &val,
+						       first_shadow_cpu(counter, id));
+		}
+	}
+}
+
 static void print_aggr(char *prefix)
 {
 	FILE *output = stat_config.output;

@@ -979,6 +1025,8 @@ static void print_aggr(char *prefix)
 	if (!(aggr_map || aggr_get_id))
 		return;
 
+	aggr_update_shadow();
+
 	for (s = 0; s < aggr_map->nr; s++) {
 		id = aggr_map->map[s];
 		evlist__for_each(evsel_list, counter) {

index 367e220..5e2d2e3 100644
@@ -14,6 +14,13 @@ enum {
 
 #define NUM_CTX CTX_BIT_MAX
 
+/*
+ * AGGR_GLOBAL: Use CPU 0
+ * AGGR_SOCKET: Use first CPU of socket
+ * AGGR_CORE: Use first CPU of core
+ * AGGR_NONE: Use matching CPU
+ * AGGR_THREAD: Not supported?
+ */
 static struct stats runtime_nsecs_stats[MAX_NR_CPUS];
 static struct stats runtime_cycles_stats[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_stalled_cycles_front_stats[NUM_CTX][MAX_NR_CPUS];

2.5.0

From: Andi Kleen <[email protected]>

With all the recently added fields in the perf stat CSV output we should finally document them in the man page. Do this here.

v2: Fix fields in documentation (Jiri)
v3: fix order of fields again (Jiri)
v4: Change order again.
v5: Document more fields (Jiri)
v6: Move time stamp first
v7: More fixes (Jiri)

---
 tools/perf/Documentation/perf-stat.txt | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

index 52ef7a9..e513a1c 100644
@@ -211,6 +211,29 @@ $ perf stat -- make -j
 
  Wall-clock time elapsed:   719.554352 msecs
 
+CSV FORMAT
+----------
+
+With -x, perf stat is able to output a not-quite-CSV format output
+Commas in the output are not put into "". To make it easy to parse
+it is recommended to use a different character like -x \;
+
+The fields are in this order:
+
+	- optional usec time stamp in fractions of second (with -I xxx)
+	- optional CPU, core, or socket identifier
+	- optional number of logical CPUs aggregated
+	- counter value
+	- unit of the counter value or empty
+	- event name
+	- run time of counter
+	- percentage of measurement time the counter was running
+	- optional variance if multiple values are collected with -r
+	- optional metric value
+	- optional unit of metric
+
+Additional metrics may be printed with all earlier fields being empty.
+
 SEE ALSO
 --------
 linkperf:perf-top[1], linkperf:perf-list[1]

2.5.0

thanks,
jirka

From: Andi Kleen <[email protected]>

Add a new mode to only print metrics. Sometimes we don't care about the raw values, just want the computed metrics. This allows more compact printing, so with -I each sample is only a single line. This also allows easier plotting and processing with other tools.

The main target is with using --topdown, but it also works with -T and standard perf stat. A few metrics are not supported.

To avoiding having to hardcode all the metrics in the code it uses a two pass approach: first compute dummy metrics and only print the headers in the print_metric callback. Then use the callback to print the actual values.

There are some additional changes
in the stat printout code to handle all metrics being on a single line.

One issue is that the column code doesn't know in advance what events are not supported by the CPU, and it would be hard to find out as this could change based on dynamic conditions. That causes empty columns in some cases.

The output can be fairly wide, often you may need more than 80 columns.

Example:

% perf stat -a -I 1000 --metric-only

     1.001452803 frontend cycles idle insn per cycle       stalled cycles per insn branch-misses of all branches
     1.001452803  158.91%               0.66                2.39                    2.92%
     2.002192321  180.63%               0.76                2.08                    2.96%
     3.003088282  150.59%               0.62                2.57                    2.84%
     4.004369835  196.20%               0.98                1.62                    3.79%
     5.005227314  231.98%               0.84                1.90                    4.71%

v2: Lots of updates.
v3: Use slightly narrower columns
v4: Add comment

---
 tools/perf/Documentation/perf-stat.txt |   4 +
 tools/perf/builtin-stat.c              | 211 +++++++++++++++++++++++++++++++--
 2 files changed, 205 insertions(+), 10 deletions(-)

index e513a1c..88a27ca 100644
@@ -139,6 +139,10 @@ Print count deltas every N milliseconds (minimum: 10ms)
 The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals.  Use with caution.
 	example: 'perf stat -I 1000 -e cycles -a sleep 5'
 
+--metric-only::
+Only print computed metrics. Print them in a single line.
+Don't show any raw values. Not supported with -A or --per-thread.
+
 --per-socket::
 Aggregate counts per processor socket for system-wide mode measurements.  This
 is a useful mode to detect imbalance between sockets.  To enable this mode,

2.5.0

could you please also make a change to display some kind of warning if there're no events to produce any metrics?

[jolsa@krava perf]$ sudo ./perf stat -a -I 1000 --metric-only -e cycles

     1.000172990 
     1.000172990 
     2.000448614 

this could be follow up patch

thanks,
jirka

From: Andi Kleen <[email protected]>

Add metric only support for -A too. This requires a new print function that prints the metrics in the right order.

v2: Fix manpage
v3: Simplify nrcpus computation

---
 tools/perf/Documentation/perf-stat.txt |  2 +-
 tools/perf/builtin-stat.c              | 45 ++++++++++++++++++++++++++++------
 2 files changed, 38 insertions(+), 9 deletions(-)

index 88a27ca..26c6b1b 100644
@@ -141,7 +141,7 @@ The overhead percentage could be high in some cases, for instance with small, su
 
 --metric-only::
 Only print computed metrics. Print them in a single line.
-Don't show any raw values. Not supported with -A or --per-thread.
+Don't show any raw values. Not supported with --per-thread.
 
 --per-socket::
 Aggregate counts per processor socket for system-wide mode measurements.  This

2.5.0

thanks,
jirka

From: Andi Kleen <[email protected]>

Add an extra check for frontend stalled in the metrics. This avoids an extra column for the --metric-only case when the CPU does not support frontend stalled.

v2: Add separate init function

---
 tools/perf/builtin-stat.c     | 1 +
 tools/perf/util/stat-shadow.c | 9 ++++++++-
 tools/perf/util/stat.h        | 1 +
 3 files changed, 10 insertions(+), 1 deletion(-)

index 7fac514..1f19f2f 100644
@@ -2176,6 +2176,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	argc = parse_options_subcommand(argc, argv, stat_options, stat_subcommands,
 					(const char **) stat_usage,
 					PARSE_OPT_STOP_AT_NON_OPTION);
+	perf_stat__init_shadow_stats();
 
 	if (csv_sep) {
 		csv_output = true;

index 5e2d2e3..b33ffb2 100644
@@ -2,6 +2,7 @@
 #include "evsel.h"
 #include "stat.h"
 #include "color.h"
+#include "pmu.h"
 
 enum {
 	CTX_BIT_USER	= 1 << 0,

@@ -35,9 +36,15 @@ static struct stats runtime_dtlb_cache_stats[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_cycles_in_tx_stats[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_transaction_stats[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_elision_stats[NUM_CTX][MAX_NR_CPUS];
+static bool have_frontend_stalled;
 
 struct stats walltime_nsecs_stats;
 
+void perf_stat__init_shadow_stats(void)
+{
+	have_frontend_stalled = pmu_have_event("cpu", "stalled-cycles-frontend");
+}
+
 static int evsel_context(struct perf_evsel *evsel)
 {
 	int ctx = 0;

@@ -323,7 +330,7 @@ void perf_stat__print_shadow_stats(struct perf_evsel *evsel,
 			print_metric(ctxp, NULL, "%7.2f ",
 					"stalled cycles per insn",
 					ratio);
-		} else {
+		} else if (have_frontend_stalled) {
 			print_metric(ctxp, NULL, NULL,
 				     "stalled cycles per insn", 0);
 		}

index f02af68..0150e78 100644
@@ -72,6 +72,7 @@ typedef void (*print_metric_t)(void *ctx, const char *color, const char *unit,
 			       const char *fmt, double val);
 typedef void (*new_line_t )(void *ctx);
 
+void perf_stat__init_shadow_stats(void);
 void perf_stat__reset_shadow_stats(void);
 void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 *count,
 				    int cpu);

2.5.0

On Thu, Mar 03, 2016 at 03:57:31PM -0800, Andi Kleen wrote:

SNIP

this branch contains old version of:
perf, tools, stat: Add --metric-only support for -A

thanks,
jirka

Old in what way? Looks correct to me.

commit e26ba80593023c24c8276ad9853acd7e56ca92ed

Author: Andi Kleen <[email protected]>
Date:   Sat Dec 19 11:22:55 2015 -0800

perf, tools, stat: Add --metric-only support for -A

Add metric only support for -A too. This requires a new print function that prints the metrics in the right order.

v2: Fix manpage
v3: Simplify nrcpus computation

I did not see the change for print_no_aggr_metric function in commit:
perf, tools, stat: Add --metric-only support for -A

which I saw in the email..

however now im confused, I did not noticed I had a wrong tree, however git can't resolve the git tree you mentioned:

[jolsa@krava perf]$ git remote add ak1 git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux-misc-2.6
[jolsa@krava perf]$ git remote update ak1
Fetching ak1
fatal: remote error: access denied or repository not exported: /pub/scm/linux/kernel/git/ak/linux-misc-2.6
error: Could not fetch ak1

so I guess the one I have is the correct one:
git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux-misc

however:
[jolsa@krava perf]$ git remote update ak Fetching ak
[jolsa@krava perf]$ git branch -r | grep stat-metrics-21 [jolsa@krava perf]$

jirka

I pushed again. Can you try again? Probably I made some mistake with git.

-Andi

On Mon, Mar 07, 2016 at 10:11:43PM +0100, Andi Kleen wrote:

SNIP

I rebased your branch on top of current Arlnaldo's perf/core and only 3 patches were missing.. I just ack-ed them

thanks,
jirka