perf_counter tools: Also display time-normalized stat results
authorIngo Molnar <mingo@elte.hu>
Fri, 29 May 2009 07:10:54 +0000 (09:10 +0200)
committerIngo Molnar <mingo@elte.hu>
Fri, 29 May 2009 07:46:45 +0000 (09:46 +0200)
Add new column that normalizes counter results by
'nanoseconds spent running' unit.

Before:

 Performance counter stats for '/home/mingo/hackbench':

   10469.403605  task clock ticks     (msecs)
          75502  context switches     (events)
           9501  CPU migrations       (events)
          36158  pagefaults           (events)
    31975676185  CPU cycles           (events)
    26257738659  instructions         (events)
      108740581  cache references     (events)
       54606088  cache misses         (events)

 Wall-clock time elapsed:   810.514504 msecs

After:

 Performance counter stats for '/home/mingo/hackbench':

   10469.403605  task clock ticks     (msecs)
          75502  context switches     #        0.007 M/sec
           9501  CPU migrations       #        0.001 M/sec
          36158  pagefaults           #        0.003 M/sec
    31975676185  CPU cycles           #     3054.202 M/sec
    26257738659  instructions         #     2508.045 M/sec
      108740581  cache references     #       10.387 M/sec
       54606088  cache misses         #        5.216 M/sec

 Wall-clock time elapsed:   810.514504 msecs

The advantage of that column is that it is characteristic of the
execution workflow, regardless of runtime. Hence 'hackbench 10'
will look similar to 'hackbench 15' - while the absolute counter
values are very different.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Documentation/perf_counter/builtin-stat.c

index 0c92eb7255261a23242c80c46abb6a1df8927e34..ef7e0e1192cbf6c578954c05a26a2ecf92cb930f 100644 (file)
@@ -74,6 +74,8 @@ static const unsigned int default_count[] = {
 static __u64                   event_res[MAX_COUNTERS][3];
 static __u64                   event_scaled[MAX_COUNTERS];
 
+static __u64                   runtime_nsecs;
+
 static void create_perfstat_counter(int counter)
 {
        struct perf_counter_hw_event hw_event;
@@ -165,6 +167,11 @@ static void read_counter(int counter)
                                ((double)count[0] * count[1] / count[2] + 0.5);
                }
        }
+       /*
+        * Save the full runtime - to allow normalization during printout:
+        */
+       if (event_id[counter] == EID(PERF_TYPE_SOFTWARE, PERF_COUNT_TASK_CLOCK))
+               runtime_nsecs = count[0];
 }
 
 /*
@@ -190,8 +197,11 @@ static void print_counter(int counter)
                fprintf(stderr, " %14.6f  %-20s (msecs)",
                        msecs, event_name(counter));
        } else {
-               fprintf(stderr, " %14Ld  %-20s (events)",
+               fprintf(stderr, " %14Ld  %-20s",
                        count[0], event_name(counter));
+               if (runtime_nsecs)
+                       fprintf(stderr, " # %12.3f M/sec",
+                               (double)count[0]/runtime_nsecs*1000.0);
        }
        if (scaled)
                fprintf(stderr, "  (scaled from %.2f%%)",