perf stat: Show percore counts in per CPU output
PATCH v2 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.
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.
With this patch, 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 CPU pairs
(CPU0/CPU4, CPU1/CPU5, CPU2/CPU6, CPU3/CPU7).
v2:
---
Add the explanation in change log. This is essentially a replacement for the any bit. No code change.
--- tools/perf/Documentation/perf-stat.txt | 9 ++++ tools/perf/builtin-stat.c | 4 ++ tools/perf/util/stat-display.c | 57 ++++++++++++++++++++++---- tools/perf/util/stat.h | 1 + 4 files changed, 62 insertions(+), 9 deletions(-)
tools/perf/Documentation/perf-stat.txt +9 -0
@@ -334,6 +334,15 @@ 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 sum counts per +hardware thread. This is essentially a replacement for the any bit and +convenient for posting process. + EXAMPLES --------
tools/perf/builtin-stat.c +4 -0
@@ -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
@@ -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
Quoted 40 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 1Performance 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.
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.
With this patch, for example,
# perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread -- sleep 1Performance 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/
Quoted 7 lines
We can see counts are duplicated in CPU pairs
(CPU0/CPU4, CPU1/CPU5, CPU2/CPU6, CPU3/CPU7).v2:
---
Add the explanation in change log. This is essentially a replacement for the any bit. No code change.
-I output is still wrong:
$ sudo ./perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread -I 1000 # time CPU counts unit events 1.000251427 1.000251427 CPU0 43,474,700 cpu/event=cpu-cycles,percore/ 1.000251427 1.000251427 CPU1 66,495,270 cpu/event=cpu-cycles,percore/ 1.000251427 1.000251427 CPU2 42,342,367 cpu/event=cpu-cycles,percore/ 1.000251427 1.000251427 CPU3 43,247,607 cpu/event=cpu-cycles,percore/
plus some comments below,
jirka
SNIP
Quoted 13 lines
tools/perf/util/stat-display.c +1
-1@@ -628,7 +628,7 @@ static void aggr_cb(struct perf_stat_config *config,static void print_counter_aggrdata (struct perf_stat_config *config,struct evsel *counter,int s,char *prefix,bool metric_only, -bool *first) +bool *first,int cpu) {struct aggr_data ad;FILE *output = config->output;@@ -654,8 +654,15 @@ static void print_counter_aggrdata(struct perf_stat_config *config,fprintf (output,"%s" , prefix); uval = val * counter->scale;
tools/perf/util/stat-display.c static void print_counter_aggrdata(struct perf_stat_config *config, · line 664
Quoted 10 lines
tools/perf/util/stat-display.c +8
-2-printout (config, id, nr, counter, uval, prefix, - run, ena,1.0 , &rt_stat); + +if (cpu ==-1 ) { +printout (config, id, nr, counter, uval, prefix, + run, ena,1.0 , &rt_stat); + }else { +printout (config, cpu, nr, counter, uval, prefix, + run, ena,1.0 , &rt_stat); + }
this would be shorter instad of above:
printout(config, cpu != -1 ?: id, nr, counter, uval, prefix, run, ena, 1.0, &rt_stat);
Quoted 22 lines
tools/perf/util/stat-display.c +3
-2+if (!metric_only)fputc ('\n' , output); }@@ -687,7 +694,7 @@ static void print_aggr(struct perf_stat_config *config,evlist__for_each_entry (evlist, counter) {print_counter_aggrdata (config, counter, s, prefix, metric_only, - &first); + &first,-1 ); }if (metric_only)fputc ('\n' , output);@@ -1163,13 +1170,38 @@ static void print_percore(struct perf_stat_config *config,print_counter_aggrdata (config, counter, s, prefix, metric_only, - &first); + &first,-1 ); }if (metric_only)
tools/perf/util/stat-display.c static void print_percore(struct perf_stat_config *config, · line 1186
Quoted 10 lines
tools/perf/util/stat-display.c +7
-0fputc ('\n' , output); } +static void print_percore_thread (struct perf_stat_config *config, +struct evsel *counter,char *prefix) +{ +int cpu, s, s2, id; +bool first = true; +FILE *output = config->output; +
missing check for
if (!(config->aggr_map || config->aggr_get_id))
tools/perf/util/stat-display.c static void print_percore(struct perf_stat_config *config, · line 1188
Quoted 2 lines
tools/perf/util/stat-display.c +2
-0+for (cpu =0 ; cpu <perf_evsel__nr_cpus (counter); cpu++) { + s2 = config->aggr_get_id (config,evsel__cpus (counter), cpu);
should you use real cpu valu in here instead of an index? like value of:
perf_cpu_map__cpu(,evsel__cpus(counter), cpu)
instead of 'cpu' above
Quoted 24 lines
tools/perf/util/stat-display.c +16
-1+ +for (s =0 ; s < config->aggr_map->nr; s++) { + id = config->aggr_map->map[s]; +if (s2 == id) +break ; + } + +if (prefix) +fprintf (output,"%s" , prefix); + +print_counter_aggrdata (config, counter, s, + prefix, false, + &first, cpu); + } +} + void perf_evlist__print_counters(struct evlist *evlist, struct perf_stat_config *config,@@ -1222,9 +1254,16 @@ perf_evlist__print_counters(struct evlist *evlist,print_no_aggr_metric (config, evlist, prefix); else {evlist__for_each_entry (evlist, counter) { -if (counter->percore)
tools/perf/util/stat-display.c perf_evlist__print_counters(struct evlist *evlist, · line 1264
Quoted 10 lines
tools/perf/util/stat-display.c +8
-2-print_percore (config, counter, prefix); - else +if (counter->percore) { +if (config->percore_show_thread) { +print_percore_thread (config, + counter, + prefix); + }else { +print_percore (config, counter, + prefix);
please keep the print_percore call in here and check/call for the percore_show_thread in it
Quoted 20 lines
tools/perf/util/stat-display.c +2
-0+ } + }else print_counter (config, counter, prefix); } }tools/perf/util/stat.h +1
-0@@ -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
Quoted 49 lines · Jiri Olsa
Quoted 40 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 1Performance 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.
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.
With this patch, for example,
# perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread -- sleep 1Performance 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/Quoted 7 lines
We can see counts are duplicated in CPU pairs
(CPU0/CPU4, CPU1/CPU5, CPU2/CPU6, CPU3/CPU7).v2:
---
Add the explanation in change log. This is essentially a replacement for the any bit. No code change.
Quoted 9 lines
-I output is still wrong:
$ sudo ./perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread -I 1000 # time CPU counts unit events 1.000251427 1.000251427 CPU0 43,474,700 cpu/event=cpu-cycles,percore/ 1.000251427 1.000251427 CPU1 66,495,270 cpu/event=cpu-cycles,percore/ 1.000251427 1.000251427 CPU2 42,342,367 cpu/event=cpu-cycles,percore/ 1.000251427 1.000251427 CPU3 43,247,607 cpu/event=cpu-cycles,percore/
Ah, double time output, I just realize this issue, very sorry about that :(
It doesn't need to print the prefix.
Quoted 32 lines
plus some comments below,
jirkaSNIP
Quoted 13 lines
tools/perf/util/stat-display.c +1
-1@@ -628,7 +628,7 @@ static void aggr_cb(struct perf_stat_config *config,static void print_counter_aggrdata (struct perf_stat_config *config,struct evsel *counter,int s,char *prefix,bool metric_only, -bool *first) +bool *first,int cpu) {struct aggr_data ad;FILE *output = config->output;@@ -654,8 +654,15 @@ static void print_counter_aggrdata(struct perf_stat_config *config,fprintf (output,"%s" , prefix); uval = val * counter->scale;tools/perf/util/stat-display.c static void print_counter_aggrdata(struct perf_stat_config *config, · line 664
Quoted 10 lines
tools/perf/util/stat-display.c +8
-2-printout (config, id, nr, counter, uval, prefix, - run, ena,1.0 , &rt_stat); + +if (cpu ==-1 ) { +printout (config, id, nr, counter, uval, prefix, + run, ena,1.0 , &rt_stat); + }else { +printout (config, cpu, nr, counter, uval, prefix, + run, ena,1.0 , &rt_stat); + }this would be shorter instad of above:
Quoted 2 lines
printout(config, cpu != -1 ?: id, nr, counter, uval, prefix, run, ena, 1.0, &rt_stat);
Yes, that would be shorter.
Quoted 34 lines
Quoted 22 lines
tools/perf/util/stat-display.c +3
-2+if (!metric_only)fputc ('\n' , output); }@@ -687,7 +694,7 @@ static void print_aggr(struct perf_stat_config *config,evlist__for_each_entry (evlist, counter) {print_counter_aggrdata (config, counter, s, prefix, metric_only, - &first); + &first,-1 ); }if (metric_only)fputc ('\n' , output);@@ -1163,13 +1170,38 @@ static void print_percore(struct perf_stat_config *config,print_counter_aggrdata (config, counter, s, prefix, metric_only, - &first); + &first,-1 ); }if (metric_only)tools/perf/util/stat-display.c static void print_percore(struct perf_stat_config *config, · line 1186
Quoted 10 lines
tools/perf/util/stat-display.c +7
-0fputc ('\n' , output); } +static void print_percore_thread (struct perf_stat_config *config, +struct evsel *counter,char *prefix) +{ +int cpu, s, s2, id; +bool first = true; +FILE *output = config->output; +missing check for
Quoted 3 lines
if (!(config->aggr_map || config->aggr_get_id))
Yes, we should add checking for config->aggr_map and config->aggr_get_id, like what the print_aggr does.
Quoted 9 lines
tools/perf/util/stat-display.c static void print_percore(struct perf_stat_config *config, · line 1188
Quoted 2 lines
tools/perf/util/stat-display.c +2
-0+for (cpu =0 ; cpu <perf_evsel__nr_cpus (counter); cpu++) { + s2 = config->aggr_get_id (config,evsel__cpus (counter), cpu);should you use real cpu valu in here instead of an index? like value of:
perf_cpu_map__cpu(,evsel__cpus(counter), cpu)instead of 'cpu' above
Yes, my fault. I should use the cpu value here but not the cpu idx. Thanks for pointing this issue out.
Quoted 29 lines
Quoted 29 lines
tools/perf/util/stat-display.c +19
-3+ +for (s =0 ; s < config->aggr_map->nr; s++) { + id = config->aggr_map->map[s]; +if (s2 == id) +break ; + } + +if (prefix) +fprintf (output,"%s" , prefix); + +print_counter_aggrdata (config, counter, s, + prefix, false, + &first, cpu); + } +} + void perf_evlist__print_counters(struct evlist *evlist, struct perf_stat_config *config,@@ -1222,9 +1254,16 @@ perf_evlist__print_counters(struct evlist *evlist,print_no_aggr_metric (config, evlist, prefix); else {evlist__for_each_entry (evlist, counter) { -if (counter->percore) -print_percore (config, counter, prefix); -else +if (counter->percore) { +if (config->percore_show_thread) { + print_percore_thread(config,
Quoted 9 lines
tools/perf/util/stat-display.c perf_evlist__print_counters(struct evlist *evlist, · line 1264
Quoted 5 lines
tools/perf/util/stat-display.c +5
-0+ counter, + prefix); + }else { +print_percore (config, counter, + prefix);please keep the
print_percorecall in here and check/call for thepercore_show_threadin it
OK, Yes, that's fine. I will keep the print_percore here and call percore_show_thread from it.
Thanks
Jin Yao
Quoted 21 lines
Quoted 20 lines
tools/perf/util/stat-display.c +2
-0+ } + }else print_counter (config, counter, prefix); } }tools/perf/util/stat.h +1
-0@@ -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