Skip to contentSkip to conversation

perf stat: Show percore counts in per CPU output

PATCH Superseded Jin Yao 3 participants 7 messages Feb 6, 2020 – Feb 11, 2020 perf events

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Jin Yao

author
#0 lore raw

PATCH 1/1 perf stat: Show percore counts in per CPU output

We have supported the event modifier "percore" which sums up the event counts for all hardware threads in a core and show the counts per core.

For example,

 # perf stat -e cpu/event=cpu-cycles,percore/ -a -A -- sleep 1

Performance counter stats for 'system wide':

 S0-D0-C0                395,072      cpu/event=cpu-cycles,percore/
 S0-D0-C1                851,248      cpu/event=cpu-cycles,percore/
 S0-D0-C2                954,226      cpu/event=cpu-cycles,percore/
 S0-D0-C3              1,233,659      cpu/event=cpu-cycles,percore/

This patch provides a new option "--percore-show-thread". It is used with event modifier "percore" together to sum up the event counts for all hardware threads in a core but show the counts per hardware thread.

For example,

 # perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread  -- sleep 1

Performance counter stats for 'system wide':

 CPU0               2,453,061      cpu/event=cpu-cycles,percore/
 CPU1               1,823,921      cpu/event=cpu-cycles,percore/
 CPU2               1,383,166      cpu/event=cpu-cycles,percore/
 CPU3               1,102,652      cpu/event=cpu-cycles,percore/
 CPU4               2,453,061      cpu/event=cpu-cycles,percore/
 CPU5               1,823,921      cpu/event=cpu-cycles,percore/
 CPU6               1,383,166      cpu/event=cpu-cycles,percore/
 CPU7               1,102,652      cpu/event=cpu-cycles,percore/

We can see counts are duplicated in some CPU pairs (CPU0/CPU4, CPU1/CPU5, CPU2/CPU6, CPU3/CPU7).

This new option may be useful for some script processing.

---
 tools/perf/Documentation/perf-stat.txt |  7 ++++
 tools/perf/builtin-stat.c              |  4 ++
 tools/perf/util/stat-display.c         | 57 ++++++++++++++++++++++----
 tools/perf/util/stat.h                 |  1 +
 4 files changed, 60 insertions(+), 9 deletions(-)
tools/perf/Documentation/perf-stat.txt +7 -0
index 9431b8066fb4..f6033b3d0971 100644
@@ -334,6 +334,13 @@ Configure all used events to run in kernel space.
 --all-user::
 Configure all used events to run in user space.
 
+--percore-show-thread::
+The event modifier "percore" has supported to sum up the event counts
+for all hardware threads in a core and show the counts per core.
+
+This option with event modifier "percore" enabled also sums up the event
+counts for all hardware threads in a core but show the counts per thread.
+
 EXAMPLES
 --------
 
tools/perf/builtin-stat.c +4 -0
index a098c2ebf4ea..ec053dc1e35c 100644
@@ -929,6 +929,10 @@ static struct option stat_options[] = {
 	OPT_BOOLEAN_FLAG(0, "all-user", &stat_config.all_user,
 			 "Configure all used events to run in user space.",
 			 PARSE_OPT_EXCLUSIVE),
+	OPT_BOOLEAN(0, "percore-show-thread", &stat_config.percore_show_thread,
+		    "Use with 'percore' event qualifier to show the event "
+		    "counts of one hardware thread by sum up total hardware "
+		    "threads of same physical core"),
 	OPT_END()
 };
 
tools/perf/util/stat-display.c +48 -9

99 lines not on this page · raw message

tools/perf/util/stat.h +1 -0
index fb990efa54a8..b4fdfaa7f2c0 100644
@@ -109,6 +109,7 @@ struct perf_stat_config {
 	bool			 walltime_run_table;
 	bool			 all_kernel;
 	bool			 all_user;
+	bool			 percore_show_thread;
 	FILE			*output;
 	unsigned int		 interval;
 	unsigned int		 timeout;
Signature · 1 line
2.17.1
#1 lore raw
Quoted 26 lines · Jin Yao

We have supported the event modifier "percore" which sums up the event counts for all hardware threads in a core and show the counts per core.

For example,

 # perf stat -e cpu/event=cpu-cycles,percore/ -a -A -- sleep 1

Performance counter stats for 'system wide':

 S0-D0-C0                395,072      cpu/event=cpu-cycles,percore/
 S0-D0-C1                851,248      cpu/event=cpu-cycles,percore/
 S0-D0-C2                954,226      cpu/event=cpu-cycles,percore/
 S0-D0-C3              1,233,659      cpu/event=cpu-cycles,percore/

This patch provides a new option "--percore-show-thread". It is used with event modifier "percore" together to sum up the event counts for all hardware threads in a core but show the counts per hardware thread.

For example,

 # perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread  -- sleep 1

Performance counter stats for 'system wide':

Quoted 8 lines
 CPU0               2,453,061      cpu/event=cpu-cycles,percore/
 CPU1               1,823,921      cpu/event=cpu-cycles,percore/
 CPU2               1,383,166      cpu/event=cpu-cycles,percore/
 CPU3               1,102,652      cpu/event=cpu-cycles,percore/
 CPU4               2,453,061      cpu/event=cpu-cycles,percore/
 CPU5               1,823,921      cpu/event=cpu-cycles,percore/
 CPU6               1,383,166      cpu/event=cpu-cycles,percore/
 CPU7               1,102,652      cpu/event=cpu-cycles,percore/

I don't understand how is this different from -A output:

  # ./perf stat -e cpu/event=cpu-cycles/ -A  
  ^C
   Performance counter stats for 'system wide':
  CPU0              56,847,497      cpu/event=cpu-cycles/                                       
  CPU1              75,274,384      cpu/event=cpu-cycles/                                       
  CPU2              63,866,342      cpu/event=cpu-cycles/                                       
  CPU3              89,559,693      cpu/event=cpu-cycles/                                       
  CPU4              74,761,132      cpu/event=cpu-cycles/                                       
  CPU5              76,320,191      cpu/event=cpu-cycles/                                       
  CPU6              55,100,175      cpu/event=cpu-cycles/                                       
  CPU7              48,472,895      cpu/event=cpu-cycles/                                       
       1.074800857 seconds time elapsed

also the interval output is mangled:

  # ./perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread  -I 1000
  #           time CPU                    counts unit events
     1.000177375      1.000177375 CPU0             138,483,540      cpu/event=cpu-cycles,percore/                                   
     1.000177375      1.000177375 CPU1             143,159,477      cpu/event=cpu-cycles,percore/                                   
     1.000177375      1.000177375 CPU2             177,554,642      cpu/event=cpu-cycles,percore/                                   
     1.000177375      1.000177375 CPU3             150,974,512      cpu/event=cpu-cycles,percore/                                   
     1.000177375      1.000177375 CPU4             138,483,540      cpu/event=cpu-cycles,percore/                                   
     1.000177375      1.000177375 CPU5             143,159,477      cpu/event=cpu-cycles,percore/                                   
     1.000177375      1.000177375 CPU6             177,554,642      cpu/event=cpu-cycles,percore/                                   

jirka

author
#2 lore raw
Quoted 51 lines · Jiri Olsa
Quoted 26 lines · Jin Yao

We have supported the event modifier "percore" which sums up the event counts for all hardware threads in a core and show the counts per core.

For example,

  # perf stat -e cpu/event=cpu-cycles,percore/ -a -A -- sleep 1

Performance counter stats for 'system wide':

  S0-D0-C0                395,072      cpu/event=cpu-cycles,percore/
  S0-D0-C1                851,248      cpu/event=cpu-cycles,percore/
  S0-D0-C2                954,226      cpu/event=cpu-cycles,percore/
  S0-D0-C3              1,233,659      cpu/event=cpu-cycles,percore/

This patch provides a new option "--percore-show-thread". It is used with event modifier "percore" together to sum up the event counts for all hardware threads in a core but show the counts per hardware thread.

For example,

  # perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread  -- sleep 1

Performance counter stats for 'system wide':

Quoted 8 lines
  CPU0               2,453,061      cpu/event=cpu-cycles,percore/
  CPU1               1,823,921      cpu/event=cpu-cycles,percore/
  CPU2               1,383,166      cpu/event=cpu-cycles,percore/
  CPU3               1,102,652      cpu/event=cpu-cycles,percore/
  CPU4               2,453,061      cpu/event=cpu-cycles,percore/
  CPU5               1,823,921      cpu/event=cpu-cycles,percore/
  CPU6               1,383,166      cpu/event=cpu-cycles,percore/
  CPU7               1,102,652      cpu/event=cpu-cycles,percore/

I don't understand how is this different from -A output:

   # ./perf stat -e cpu/event=cpu-cycles/ -A
   ^C
    Performance counter stats for 'system wide':
   CPU0              56,847,497      cpu/event=cpu-cycles/
   CPU1              75,274,384      cpu/event=cpu-cycles/
   CPU2              63,866,342      cpu/event=cpu-cycles/
   CPU3              89,559,693      cpu/event=cpu-cycles/
   CPU4              74,761,132      cpu/event=cpu-cycles/
   CPU5              76,320,191      cpu/event=cpu-cycles/
   CPU6              55,100,175      cpu/event=cpu-cycles/
   CPU7              48,472,895      cpu/event=cpu-cycles/
Quoted 2 lines
        1.074800857 seconds time elapsed

The results are different.

With --percore-show-thread, CPU0 and CPU4 have the same counts (CPU0 and CPU4 are siblings, e.g. 2,453,061 in my example). The value is sum of CPU0 + CPU4.

Without --percore-show-thread, CPU0 and CPU4 have their own counts.

Quoted 12 lines

also the interval output is mangled:

   # ./perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread  -I 1000
   #           time CPU                    counts unit events
      1.000177375      1.000177375 CPU0             138,483,540      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU1             143,159,477      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU2             177,554,642      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU3             150,974,512      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU4             138,483,540      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU5             143,159,477      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU6             177,554,642      cpu/event=cpu-cycles,percore/
Quoted 2 lines

jirka

Sorry, why the interval output is mangled? It's expected that CPU0 and CPU4 have the same counts.

Thanks
Jin Yao

#3 lore raw
Quoted 54 lines · Yao
Quoted 51 lines · Jiri Olsa
Quoted 26 lines · Jin Yao

We have supported the event modifier "percore" which sums up the event counts for all hardware threads in a core and show the counts per core.

For example,

  # perf stat -e cpu/event=cpu-cycles,percore/ -a -A -- sleep 1

Performance counter stats for 'system wide':

  S0-D0-C0                395,072      cpu/event=cpu-cycles,percore/
  S0-D0-C1                851,248      cpu/event=cpu-cycles,percore/
  S0-D0-C2                954,226      cpu/event=cpu-cycles,percore/
  S0-D0-C3              1,233,659      cpu/event=cpu-cycles,percore/

This patch provides a new option "--percore-show-thread". It is used with event modifier "percore" together to sum up the event counts for all hardware threads in a core but show the counts per hardware thread.

For example,

  # perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread  -- sleep 1

Performance counter stats for 'system wide':

Quoted 8 lines
  CPU0               2,453,061      cpu/event=cpu-cycles,percore/
  CPU1               1,823,921      cpu/event=cpu-cycles,percore/
  CPU2               1,383,166      cpu/event=cpu-cycles,percore/
  CPU3               1,102,652      cpu/event=cpu-cycles,percore/
  CPU4               2,453,061      cpu/event=cpu-cycles,percore/
  CPU5               1,823,921      cpu/event=cpu-cycles,percore/
  CPU6               1,383,166      cpu/event=cpu-cycles,percore/
  CPU7               1,102,652      cpu/event=cpu-cycles,percore/

I don't understand how is this different from -A output:

   # ./perf stat -e cpu/event=cpu-cycles/ -A
   ^C
    Performance counter stats for 'system wide':
   CPU0              56,847,497      cpu/event=cpu-cycles/
   CPU1              75,274,384      cpu/event=cpu-cycles/
   CPU2              63,866,342      cpu/event=cpu-cycles/
   CPU3              89,559,693      cpu/event=cpu-cycles/
   CPU4              74,761,132      cpu/event=cpu-cycles/
   CPU5              76,320,191      cpu/event=cpu-cycles/
   CPU6              55,100,175      cpu/event=cpu-cycles/
   CPU7              48,472,895      cpu/event=cpu-cycles/
Quoted 8 lines
Quoted 2 lines
        1.074800857 seconds time elapsed

The results are different.

With --percore-show-thread, CPU0 and CPU4 have the same counts (CPU0 and CPU4 are siblings, e.g. 2,453,061 in my example). The value is sum of CPU0 + CPU4.

so it shows percore stats but displays all the cpus? what is this good for? to see which cpus are in core? if that's the case then I think we could somehow display the cpu numbers for core in --per-core output, like:

S0-D0-C0(0,4)                395,072      cpu/event=cpu-cycles,percore/
S0-D0-C1(1,5)                851,248      cpu/event=cpu-cycles,percore/
S0-D0-C2(2,6)                954,226      cpu/event=cpu-cycles,percore/
S0-D0-C3(3,7)              1,233,659      cpu/event=cpu-cycles,percore/
Quoted 15 lines

Without --percore-show-thread, CPU0 and CPU4 have their own counts.

Quoted 12 lines

also the interval output is mangled:

   # ./perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread  -I 1000
   #           time CPU                    counts unit events
      1.000177375      1.000177375 CPU0             138,483,540      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU1             143,159,477      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU2             177,554,642      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU3             150,974,512      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU4             138,483,540      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU5             143,159,477      cpu/event=cpu-cycles,percore/
      1.000177375      1.000177375 CPU6             177,554,642      cpu/event=cpu-cycles,percore/
Quoted 5 lines
Quoted 2 lines

jirka

Sorry, why the interval output is mangled? It's expected that CPU0 and CPU4 have the same counts.

there are 2 timestamp columns and the header line does not align with the data

jirka

#4 lore raw
Quoted 5 lines
Quoted 3 lines

With --percore-show-thread, CPU0 and CPU4 have the same counts (CPU0 and CPU4 are siblings, e.g. 2,453,061 in my example). The value is sum of CPU0 + CPU4.

so it shows percore stats but displays all the cpus? what is this good for?

This is essentially a replacement for the any bit (which is gone in Icelake). Per core counts are useful for some formulas, e.g. CoreIPC

The original percore version was inconvenient to post process. This variant matches the output of the any bit.

-Andi

#5 lore raw
Quoted 11 lines · Andi Kleen
Quoted 5 lines
Quoted 3 lines

With --percore-show-thread, CPU0 and CPU4 have the same counts (CPU0 and CPU4 are siblings, e.g. 2,453,061 in my example). The value is sum of CPU0 + CPU4.

so it shows percore stats but displays all the cpus? what is this good for?

This is essentially a replacement for the any bit (which is gone in Icelake). Per core counts are useful for some formulas, e.g. CoreIPC

The original percore version was inconvenient to post process. This variant matches the output of the any bit.

I see, please put this to the changelog/doc

thanks,
jirka

author
#6 lore raw
Quoted 10 lines · Jiri Olsa
Quoted 9 lines · Andi Kleen
Quoted 5 lines
Quoted 3 lines

With --percore-show-thread, CPU0 and CPU4 have the same counts (CPU0 and CPU4 are siblings, e.g. 2,453,061 in my example). The value is sum of CPU0 + CPU4.

so it shows percore stats but displays all the cpus? what is this good for?

This is essentially a replacement for the any bit (which is gone in Icelake). Per core counts are useful for some formulas, e.g. CoreIPC

Quoted 8 lines
Quoted 2 lines

The original percore version was inconvenient to post process. This variant matches the output of the any bit.

I see, please put this to the changelog/doc

thanks,
jirka

Thanks Jiri, thanks Andi!

I will put the explanation in v2.

Thanks
Jin Yao