perf stat: Improve scaling
authorAndi Kleen <ak@linux.intel.com>
Thu, 14 Mar 2019 22:50:02 +0000 (15:50 -0700)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Tue, 19 Mar 2019 19:52:03 +0000 (16:52 -0300)
The multiplexing scaling in perf stat mysteriously adds 0.5 to the
value. This dates back to the original perf tool. Other scaling code
doesn't use that strange convention. Remove the extra 0.5.

Before:

$ perf stat -e 'cycles,cycles,cycles,cycles,cycles,cycles' grep -rq foo

 Performance counter stats for 'grep -rq foo':

         6,403,580      cycles                                                        (81.62%)
         6,404,341      cycles                                                        (81.64%)
         6,402,983      cycles                                                        (81.62%)
         6,399,941      cycles                                                        (81.63%)
         6,399,451      cycles                                                        (81.62%)
         6,436,105      cycles                                                        (91.87%)

       0.005843799 seconds time elapsed

       0.002905000 seconds user
       0.002902000 seconds sys

After:

$ perf stat -e 'cycles,cycles,cycles,cycles,cycles,cycles' grep -rq foo

 Performance counter stats for 'grep -rq foo':

         6,422,704      cycles                                                        (81.68%)
         6,401,842      cycles                                                        (81.68%)
         6,398,432      cycles                                                        (81.68%)
         6,397,098      cycles                                                        (81.68%)
         6,396,074      cycles                                                        (81.67%)
         6,434,980      cycles                                                        (91.62%)

       0.005884437 seconds time elapsed

       0.003580000 seconds user
       0.002356000 seconds sys

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
LPU-Reference: 20190314225002.30108-10-andi@firstfloor.org
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/evsel.c

index 53ec40cacd4b75300a6d1912226cf5e12768eb49..16f02b83b23f3185b4621eff3f7eebae0d49a3a3 100644 (file)
@@ -1342,7 +1342,7 @@ void perf_counts_values__scale(struct perf_counts_values *count,
                        count->val = 0;
                } else if (count->run < count->ena) {
                        scaled = 1;
-                       count->val = (u64)((double) count->val * count->ena / count->run + 0.5);
+                       count->val = (u64)((double) count->val * count->ena / count->run);
                }
        }