6 perf-stat - Run a command and gather performance counter statistics
11 'perf stat' [-e <EVENT> | --event=EVENT] [-a] <command>
12 'perf stat' [-e <EVENT> | --event=EVENT] [-a] -- <command> [<options>]
16 This command runs a command and gathers performance counter statistics
23 Any command you can specify in a shell.
28 Select the PMU event. Selection can be:
30 - a symbolic event name (use 'perf list' to list all events)
32 - a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
33 hexadecimal event descriptor.
35 - a symbolically formed event like 'pmu/param1=0x3,param2/' where
36 param1 and param2 are defined as formats for the PMU in
37 /sys/bus/event_sources/devices/<pmu>/format/*
39 - a symbolically formed event like 'pmu/config=M,config1=N,config2=K/'
40 where M, N, K are numbers (in decimal, hex, octal format).
41 Acceptable values for each of 'config', 'config1' and 'config2'
42 parameters are defined by corresponding entries in
43 /sys/bus/event_sources/devices/<pmu>/format/*
47 child tasks do not inherit counters
50 stat events on existing process id (comma separated list)
54 stat events on existing thread id (comma separated list)
59 system-wide collection from all CPUs
63 scale/normalize counter values
67 repeat command and print average + stddev (max: 100). 0 means forever.
71 print large numbers with thousands' separators according to locale
75 Count only on the list of CPUs provided. Multiple CPUs can be provided as a
76 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
77 In per-thread mode, this option is ignored. The -a option is still necessary
78 to activate system-wide monitoring. Default is to count on all CPUs.
82 Do not aggregate counts across all monitored CPUs in system-wide mode (-a).
83 This option is only valid in system-wide mode.
87 null run - don't start any counters
91 be more verbose (show counter open errors, etc)
94 --field-separator SEP::
95 print counts using a CSV-style output to make it easy to import directly into
96 spreadsheets. Columns are separated by the string specified in SEP.
100 monitor only in the container (cgroup) called "name". This option is available only
101 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
102 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
103 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
104 to first event, second cgroup to second event and so on. It is possible to provide
105 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
106 corresponding events, i.e., they always refer to events defined earlier on the command
111 Print the output into the designated file.
114 Append to the output file designated with the -o option. Ignored if -o is not specified.
118 Log output to fd, instead of stderr. Complementary to --output, and mutually exclusive
119 with it. --append may be used here. Examples:
120 3>results perf stat --log-fd 3 -- $cmd
121 3>>results perf stat --log-fd 3 --append -- $cmd
125 Pre and post measurement hooks, e.g.:
127 perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
130 --interval-print msecs::
131 Print count deltas every N milliseconds (minimum: 100ms)
132 example: perf stat -I 1000 -e cycles -a sleep 5
135 Aggregate counts per processor socket for system-wide mode measurements. This
136 is a useful mode to detect imbalance between sockets. To enable this mode,
137 use --per-socket in addition to -a. (system-wide). The output includes the
138 socket number and the number of online processors on that socket. This is
139 useful to gauge the amount of aggregation.
142 Aggregate counts per physical processor for system-wide mode measurements. This
143 is a useful mode to detect imbalance between physical cores. To enable this mode,
144 use --per-core in addition to -a. (system-wide). The output includes the
145 core number and the number of online logical processors on that physical processor.
149 After starting the program, wait msecs before measuring. This is useful to
150 filter out the startup phase of the program, which is often very different.
155 Print statistics of transactional execution if supported.
160 $ perf stat -- make -j
162 Performance counter stats for 'make -j':
164 8117.370256 task clock ticks # 11.281 CPU utilization factor
165 678 context switches # 0.000 M/sec
166 133 CPU migrations # 0.000 M/sec
167 235724 pagefaults # 0.029 M/sec
168 24821162526 CPU cycles # 3057.784 M/sec
169 18687303457 instructions # 2302.138 M/sec
170 172158895 cache references # 21.209 M/sec
171 27075259 cache misses # 3.335 M/sec
173 Wall-clock time elapsed: 719.554352 msecs
177 linkperf:perf-top[1], linkperf:perf-list[1]