1perf-stat(1)
2============
3
4NAME
5----
6perf-stat - Run a command and gather performance counter statistics
7
8SYNOPSIS
9--------
10[verse]
11'perf stat' [-e <EVENT> | --event=EVENT] [-a] <command>
12'perf stat' [-e <EVENT> | --event=EVENT] [-a] -- <command> [<options>]
13'perf stat' [-e <EVENT> | --event=EVENT] [-a] record [-o file] -- <command> [<options>]
14'perf stat' report [-i file]
15
16DESCRIPTION
17-----------
18This command runs a command and gathers performance counter statistics
19from it.
20
21
22OPTIONS
23-------
24<command>...::
25	Any command you can specify in a shell.
26
27record::
28	See STAT RECORD.
29
30report::
31	See STAT REPORT.
32
33-e::
34--event=::
35	Select the PMU event. Selection can be:
36
37	- a symbolic event name (use 'perf list' to list all events)
38
39	- a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
40	  hexadecimal event descriptor.
41
42	- a symbolically formed event like 'pmu/param1=0x3,param2/' where
43	  param1 and param2 are defined as formats for the PMU in
44	  /sys/bus/event_sources/devices/<pmu>/format/*
45
46	- a symbolically formed event like 'pmu/config=M,config1=N,config2=K/'
47	  where M, N, K are numbers (in decimal, hex, octal format).
48	  Acceptable values for each of 'config', 'config1' and 'config2'
49	  parameters are defined by corresponding entries in
50	  /sys/bus/event_sources/devices/<pmu>/format/*
51
52-i::
53--no-inherit::
54        child tasks do not inherit counters
55-p::
56--pid=<pid>::
57        stat events on existing process id (comma separated list)
58
59-t::
60--tid=<tid>::
61        stat events on existing thread id (comma separated list)
62
63
64-a::
65--all-cpus::
66        system-wide collection from all CPUs
67
68-c::
69--scale::
70	scale/normalize counter values
71
72-d::
73--detailed::
74	print more detailed statistics, can be specified up to 3 times
75
76	   -d:          detailed events, L1 and LLC data cache
77        -d -d:     more detailed events, dTLB and iTLB events
78     -d -d -d:     very detailed events, adding prefetch events
79
80-r::
81--repeat=<n>::
82	repeat command and print average + stddev (max: 100). 0 means forever.
83
84-B::
85--big-num::
86        print large numbers with thousands' separators according to locale
87
88-C::
89--cpu=::
90Count only on the list of CPUs provided. Multiple CPUs can be provided as a
91comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
92In per-thread mode, this option is ignored. The -a option is still necessary
93to activate system-wide monitoring. Default is to count on all CPUs.
94
95-A::
96--no-aggr::
97Do not aggregate counts across all monitored CPUs in system-wide mode (-a).
98This option is only valid in system-wide mode.
99
100-n::
101--null::
102        null run - don't start any counters
103
104-v::
105--verbose::
106        be more verbose (show counter open errors, etc)
107
108-x SEP::
109--field-separator SEP::
110print counts using a CSV-style output to make it easy to import directly into
111spreadsheets. Columns are separated by the string specified in SEP.
112
113-G name::
114--cgroup name::
115monitor only in the container (cgroup) called "name". This option is available only
116in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
117container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
118can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
119to first event, second cgroup to second event and so on. It is possible to provide
120an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
121corresponding events, i.e., they always refer to events defined earlier on the command
122line.
123
124-o file::
125--output file::
126Print the output into the designated file.
127
128--append::
129Append to the output file designated with the -o option. Ignored if -o is not specified.
130
131--log-fd::
132
133Log output to fd, instead of stderr.  Complementary to --output, and mutually exclusive
134with it.  --append may be used here.  Examples:
135     3>results  perf stat --log-fd 3          -- $cmd
136     3>>results perf stat --log-fd 3 --append -- $cmd
137
138--pre::
139--post::
140	Pre and post measurement hooks, e.g.:
141
142perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
143
144-I msecs::
145--interval-print msecs::
146Print count deltas every N milliseconds (minimum: 10ms)
147The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals.  Use with caution.
148	example: 'perf stat -I 1000 -e cycles -a sleep 5'
149
150--metric-only::
151Only print computed metrics. Print them in a single line.
152Don't show any raw values. Not supported with --per-thread.
153
154--per-socket::
155Aggregate counts per processor socket for system-wide mode measurements.  This
156is a useful mode to detect imbalance between sockets.  To enable this mode,
157use --per-socket in addition to -a. (system-wide).  The output includes the
158socket number and the number of online processors on that socket. This is
159useful to gauge the amount of aggregation.
160
161--per-core::
162Aggregate counts per physical processor for system-wide mode measurements.  This
163is a useful mode to detect imbalance between physical cores.  To enable this mode,
164use --per-core in addition to -a. (system-wide).  The output includes the
165core number and the number of online logical processors on that physical processor.
166
167--per-thread::
168Aggregate counts per monitored threads, when monitoring threads (-t option)
169or processes (-p option).
170
171-D msecs::
172--delay msecs::
173After starting the program, wait msecs before measuring. This is useful to
174filter out the startup phase of the program, which is often very different.
175
176-T::
177--transaction::
178
179Print statistics of transactional execution if supported.
180
181STAT RECORD
182-----------
183Stores stat data into perf data file.
184
185-o file::
186--output file::
187Output file name.
188
189STAT REPORT
190-----------
191Reads and reports stat data from perf data file.
192
193-i file::
194--input file::
195Input file name.
196
197--per-socket::
198Aggregate counts per processor socket for system-wide mode measurements.
199
200--per-core::
201Aggregate counts per physical processor for system-wide mode measurements.
202
203-A::
204--no-aggr::
205Do not aggregate counts across all monitored CPUs.
206
207--topdown::
208Print top down level 1 metrics if supported by the CPU. This allows to
209determine bottle necks in the CPU pipeline for CPU bound workloads,
210by breaking the cycles consumed down into frontend bound, backend bound,
211bad speculation and retiring.
212
213Frontend bound means that the CPU cannot fetch and decode instructions fast
214enough. Backend bound means that computation or memory access is the bottle
215neck. Bad Speculation means that the CPU wasted cycles due to branch
216mispredictions and similar issues. Retiring means that the CPU computed without
217an apparently bottleneck. The bottleneck is only the real bottleneck
218if the workload is actually bound by the CPU and not by something else.
219
220For best results it is usually a good idea to use it with interval
221mode like -I 1000, as the bottleneck of workloads can change often.
222
223The top down metrics are collected per core instead of per
224CPU thread. Per core mode is automatically enabled
225and -a (global monitoring) is needed, requiring root rights or
226perf.perf_event_paranoid=-1.
227
228Topdown uses the full Performance Monitoring Unit, and needs
229disabling of the NMI watchdog (as root):
230echo 0 > /proc/sys/kernel/nmi_watchdog
231for best results. Otherwise the bottlenecks may be inconsistent
232on workload with changing phases.
233
234This enables --metric-only, unless overriden with --no-metric-only.
235
236To interpret the results it is usually needed to know on which
237CPUs the workload runs on. If needed the CPUs can be forced using
238taskset.
239
240EXAMPLES
241--------
242
243$ perf stat -- make -j
244
245 Performance counter stats for 'make -j':
246
247    8117.370256  task clock ticks     #      11.281 CPU utilization factor
248            678  context switches     #       0.000 M/sec
249            133  CPU migrations       #       0.000 M/sec
250         235724  pagefaults           #       0.029 M/sec
251    24821162526  CPU cycles           #    3057.784 M/sec
252    18687303457  instructions         #    2302.138 M/sec
253      172158895  cache references     #      21.209 M/sec
254       27075259  cache misses         #       3.335 M/sec
255
256 Wall-clock time elapsed:   719.554352 msecs
257
258CSV FORMAT
259----------
260
261With -x, perf stat is able to output a not-quite-CSV format output
262Commas in the output are not put into "". To make it easy to parse
263it is recommended to use a different character like -x \;
264
265The fields are in this order:
266
267	- optional usec time stamp in fractions of second (with -I xxx)
268	- optional CPU, core, or socket identifier
269	- optional number of logical CPUs aggregated
270	- counter value
271	- unit of the counter value or empty
272	- event name
273	- run time of counter
274	- percentage of measurement time the counter was running
275	- optional variance if multiple values are collected with -r
276	- optional metric value
277	- optional unit of metric
278
279Additional metrics may be printed with all earlier fields being empty.
280
281SEE ALSO
282--------
283linkperf:perf-top[1], linkperf:perf-list[1]
284