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 14DESCRIPTION 15----------- 16This command runs a command and gathers performance counter statistics 17from it. 18 19 20OPTIONS 21------- 22<command>...:: 23 Any command you can specify in a shell. 24 25 26-e:: 27--event=:: 28 Select the PMU event. Selection can be a symbolic event name 29 (use 'perf list' to list all events) or a raw PMU 30 event (eventsel+umask) in the form of rNNN where NNN is a 31 hexadecimal event descriptor. 32 33-i:: 34--no-inherit:: 35 child tasks do not inherit counters 36-p:: 37--pid=<pid>:: 38 stat events on existing process id (comma separated list) 39 40-t:: 41--tid=<tid>:: 42 stat events on existing thread id (comma separated list) 43 44 45-a:: 46--all-cpus:: 47 system-wide collection from all CPUs 48 49-c:: 50--scale:: 51 scale/normalize counter values 52 53-r:: 54--repeat=<n>:: 55 repeat command and print average + stddev (max: 100). 0 means forever. 56 57-B:: 58--big-num:: 59 print large numbers with thousands' separators according to locale 60 61-C:: 62--cpu=:: 63Count only on the list of CPUs provided. Multiple CPUs can be provided as a 64comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2. 65In per-thread mode, this option is ignored. The -a option is still necessary 66to activate system-wide monitoring. Default is to count on all CPUs. 67 68-A:: 69--no-aggr:: 70Do not aggregate counts across all monitored CPUs in system-wide mode (-a). 71This option is only valid in system-wide mode. 72 73-n:: 74--null:: 75 null run - don't start any counters 76 77-v:: 78--verbose:: 79 be more verbose (show counter open errors, etc) 80 81-x SEP:: 82--field-separator SEP:: 83print counts using a CSV-style output to make it easy to import directly into 84spreadsheets. Columns are separated by the string specified in SEP. 85 86-G name:: 87--cgroup name:: 88monitor only in the container (cgroup) called "name". This option is available only 89in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to 90container "name" are monitored when they run on the monitored CPUs. Multiple cgroups 91can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup 92to first event, second cgroup to second event and so on. It is possible to provide 93an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have 94corresponding events, i.e., they always refer to events defined earlier on the command 95line. 96 97-o file:: 98--output file:: 99Print the output into the designated file. 100 101--append:: 102Append to the output file designated with the -o option. Ignored if -o is not specified. 103 104--log-fd:: 105 106Log output to fd, instead of stderr. Complementary to --output, and mutually exclusive 107with it. --append may be used here. Examples: 108 3>results perf stat --log-fd 3 -- $cmd 109 3>>results perf stat --log-fd 3 --append -- $cmd 110 111--pre:: 112--post:: 113 Pre and post measurement hooks, e.g.: 114 115perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage 116 117-I msecs:: 118--interval-print msecs:: 119 Print count deltas every N milliseconds (minimum: 100ms) 120 example: perf stat -I 1000 -e cycles -a sleep 5 121 122--per-socket:: 123Aggregate counts per processor socket for system-wide mode measurements. This 124is a useful mode to detect imbalance between sockets. To enable this mode, 125use --per-socket in addition to -a. (system-wide). The output includes the 126socket number and the number of online processors on that socket. This is 127useful to gauge the amount of aggregation. 128 129--per-core:: 130Aggregate counts per physical processor for system-wide mode measurements. This 131is a useful mode to detect imbalance between physical cores. To enable this mode, 132use --per-core in addition to -a. (system-wide). The output includes the 133core number and the number of online logical processors on that physical processor. 134 135-D msecs:: 136--initial-delay msecs:: 137After starting the program, wait msecs before measuring. This is useful to 138filter out the startup phase of the program, which is often very different. 139 140-T:: 141--transaction:: 142 143Print statistics of transactional execution if supported. 144 145EXAMPLES 146-------- 147 148$ perf stat -- make -j 149 150 Performance counter stats for 'make -j': 151 152 8117.370256 task clock ticks # 11.281 CPU utilization factor 153 678 context switches # 0.000 M/sec 154 133 CPU migrations # 0.000 M/sec 155 235724 pagefaults # 0.029 M/sec 156 24821162526 CPU cycles # 3057.784 M/sec 157 18687303457 instructions # 2302.138 M/sec 158 172158895 cache references # 21.209 M/sec 159 27075259 cache misses # 3.335 M/sec 160 161 Wall-clock time elapsed: 719.554352 msecs 162 163SEE ALSO 164-------- 165linkperf:perf-top[1], linkperf:perf-list[1] 166