Messages in this thread Patch in this message | | | From | Jin Yao <> | Subject | [PATCH] perf stat: Support metrics with perf stat --per-thread | Date | Thu, 12 Oct 2017 17:05:00 +0800 |
| |
perf stat --per-thread doesn't support outputting metrics, like IPC.
We should support this because it would allow easily to collect metrics for different threads in applications.
1. Current output, for example:
root@skl:/tmp# perf stat --per-thread -p 21623 ^C Performance counter stats for process id '21623':
vmstat-21623 0.517479 task-clock (msec) # 0.000 CPUs utilized vmstat-21623 1 context-switches vmstat-21623 0 cpu-migrations vmstat-21623 0 page-faults vmstat-21623 461,306 cycles vmstat-21623 630,724 instructions vmstat-21623 136,265 branches vmstat-21623 2,520 branch-misses
1.444020756 seconds time elapsed
root@skl:/tmp# perf stat --per-thread --metrics ipc -p 21623 ^C Performance counter stats for process id '21623':
vmstat-21623 631,185 inst_retired.any vmstat-21623 605,893 cpu_clk_unhalted.thread
1.415679293 seconds time elapsed
2. With this patch, the result would be:
root@skl:/tmp# perf stat --per-thread -p 21623 ^C Performance counter stats for process id '21623':
vmstat-21623 0.533759 task-clock (msec) # 0.000 CPUs utilized vmstat-21623 1 context-switches # 0.002 M/sec vmstat-21623 0 cpu-migrations # 0.000 K/sec vmstat-21623 0 page-faults # 0.000 K/sec vmstat-21623 473,896 cycles # 0.888 GHz vmstat-21623 631,072 instructions # 1.33 insn per cycle vmstat-21623 136,307 branches # 255.372 M/sec vmstat-21623 2,524 branch-misses # 1.85% of all branches
1.544862861 seconds time elapsed
root@skl:/tmp# perf stat --per-thread --metrics ipc -p 21623 ^C Performance counter stats for process id '21623':
vmstat-21623 1,259,104 inst_retired.any # 1.2 IPC vmstat-21623 1,056,756 cpu_clk_unhalted.thread
2.040954502 seconds time elapsed
Signed-off-by: Jin Yao <yao.jin@linux.intel.com> --- tools/perf/util/stat.c | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+)
diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c index 35e9848..1164e68 100644 --- a/tools/perf/util/stat.c +++ b/tools/perf/util/stat.c @@ -314,6 +314,26 @@ static int process_counter_maps(struct perf_stat_config *config, return 0; } +static int process_aggr_thread_counter(struct perf_evsel *counter) +{ + int nthreads = thread_map__nr(counter->threads); + int ncpus = cpu_map__nr(counter->cpus); + int cpu, thread; + u64 tmp; + + for (thread = 0; thread < nthreads; thread++) { + u64 val = 0; + + for (cpu = 0; cpu < ncpus; cpu++) + val += perf_counts(counter->counts, cpu, thread)->val; + + tmp = val * counter->scale; + perf_stat__update_shadow_stats(counter, &tmp, 0); + } + + return 0; +} + int perf_stat_process_counter(struct perf_stat_config *config, struct perf_evsel *counter) { @@ -342,6 +362,9 @@ int perf_stat_process_counter(struct perf_stat_config *config, if (ret) return ret; + if (config->aggr_mode == AGGR_THREAD) + return process_aggr_thread_counter(counter); + if (config->aggr_mode != AGGR_GLOBAL) return 0; -- 2.7.4
| |