perf stat: Use proper cpu for shadow stats
authorNamhyung Kim <namhyung@kernel.org>
Fri, 27 Nov 2020 04:14:03 +0000 (13:14 +0900)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 27 Nov 2020 17:31:37 +0000 (14:31 -0300)
Currently perf stat shows some metrics (like IPC) for defined events.
But when no aggregation mode is used (-A option), it shows incorrect
values since it used a value from a different cpu.

Before:

  $ perf stat -aA -e cycles,instructions sleep 1

   Performance counter stats for 'system wide':

  CPU0      116,057,380      cycles
  CPU1       86,084,722      cycles
  CPU2       99,423,125      cycles
  CPU3       98,272,994      cycles
  CPU0       53,369,217      instructions      #    0.46  insn per cycle
  CPU1       33,378,058      instructions      #    0.29  insn per cycle
  CPU2       58,150,086      instructions      #    0.50  insn per cycle
  CPU3       40,029,703      instructions      #    0.34  insn per cycle

       1.001816971 seconds time elapsed

So the IPC for CPU1 should be 0.38 (= 33,378,058 / 86,084,722)
but it was 0.29 (= 33,378,058 / 116,057,380) and so on.

After:

  $ perf stat -aA -e cycles,instructions sleep 1

   Performance counter stats for 'system wide':

  CPU0      109,621,384      cycles
  CPU1      159,026,454      cycles
  CPU2       99,460,366      cycles
  CPU3      124,144,142      cycles
  CPU0       44,396,706      instructions      #    0.41  insn per cycle
  CPU1      120,195,425      instructions      #    0.76  insn per cycle
  CPU2       44,763,978      instructions      #    0.45  insn per cycle
  CPU3       69,049,079      instructions      #    0.56  insn per cycle

       1.001910444 seconds time elapsed

Fixes: 44d49a600259 ("perf stat: Support metrics in --per-core/socket mode")
Reported-by: Sam Xi <xyzsam@google.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20201127041404.390276-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/stat-display.c

index 4b57c0c076323ceac2909dd78c031e376f86ed24..a963b5b8eb72409152596d6b33f0d961aeb1aa40 100644 (file)
@@ -324,13 +324,10 @@ static int first_shadow_cpu(struct perf_stat_config *config,
        struct evlist *evlist = evsel->evlist;
        int i;
 
-       if (!config->aggr_get_id)
-               return 0;
-
        if (config->aggr_mode == AGGR_NONE)
                return id;
 
-       if (config->aggr_mode == AGGR_GLOBAL)
+       if (!config->aggr_get_id)
                return 0;
 
        for (i = 0; i < evsel__nr_cpus(evsel); i++) {