perf_counter tool: builtin-stat add more events
PATCH 1/1 perf_counter tool: builtin-stat add more events
Added more events not it looks like on AMD box :
./perf stat -- ls -lR > /dev/null
Performance counter stats for 'ls -lR':
2507744774 cycles # 2085.473 M/sec (scaled from 13.28%)
1515534968 instructions # 0.604 IPC (scaled from 13.28%)
783181797 cache-references # 651.304 M/sec (scaled from 36.36%)
18089523 cache-misses # 15.043 M/sec (scaled from 36.37%)
195550613 branches # 162.622 M/sec (scaled from 36.29%)
14623394 branch-misses # 12.161 M/sec (scaled from 36.29%)
<not counted> bus-cycles
1203.182949 cpu-clock-msecs
1202.482671 task-clock-msecs # 0.990 CPUs
454 page-faults # 0.000 M/sec
454 minor-faults # 0.000 M/sec
0 major-faults # 0.000 M/sec
133 context-switches # 0.000 M/sec
1 CPU-migrations # 0.000 M/sec
744421154 L1-data-Cache-Load-Referencees # 619.070 M/sec (scaled from 13.20%)
5220656 L1-data-Cache-Load-Misses # 4.342 M/sec (scaled from 13.28%)
438576 L1-data-Cache-Store-Referencees # 0.365 M/sec (scaled from 13.36%)
<not counted> L1-data-Cache-Store-Misses
1976596 L1-data-Cache-Prefetch-Referencees # 1.644 M/sec (scaled from 13.44%)
1644021 L1-data-Cache-Prefetch-Misses # 1.367 M/sec (scaled from 13.52%)
764273224 L1-instruction-Cache-Load-Referencees # 635.579 M/sec (scaled from 13.53%)
17242789 L1-instruction-Cache-Load-Misses # 14.339 M/sec (scaled from 13.53%)
<not counted> L1-instruction-Cache-Store-Referencees
<not counted> L1-instruction-Cache-Store-Misses
372621 L1-instruction-Cache-Prefetch-Referencees # 0.310 M/sec (scaled from 13.53%)
<not counted> L1-instruction-Cache-Prefetch-Misses
22844109 L2-Cache-Load-Referencees # 18.997 M/sec (scaled from 13.53%)
2235733 L2-Cache-Load-Misses # 1.859 M/sec (scaled from 13.53%)
23949920 L2-Cache-Store-Referencees # 19.917 M/sec (scaled from 13.46%)
<not counted> L2-Cache-Store-Misses
<not counted> L2-Cache-Prefetch-Referencees
<not counted> L2-Cache-Prefetch-Misses
732364670 Data-TLB-Cache-Load-Referencees # 609.044 M/sec (scaled from 13.45%)
16516548 Data-TLB-Cache-Load-Misses # 13.735 M/sec (scaled from 13.42%)
<not counted> Data-TLB-Cache-Store-Referencees
<not counted> Data-TLB-Cache-Store-Misses
<not counted> Data-TLB-Cache-Prefetch-Referencees
<not counted> Data-TLB-Cache-Prefetch-Misses
766865920 Instruction-TLB-Cache-Load-Referencees # 637.736 M/sec (scaled from 13.42%)
19981 Instruction-TLB-Cache-Load-Misses # 0.017 M/sec (scaled from 13.40%)
<not counted> Instruction-TLB-Cache-Store-Referencees
<not counted> Instruction-TLB-Cache-Store-Misses
<not counted> Instruction-TLB-Cache-Prefetch-Referencees
<not counted> Instruction-TLB-Cache-Prefetch-Misses
308272002 Branch-Cache-Load-Referencees # 256.363 M/sec (scaled from 13.33%)
19226358 Branch-Cache-Load-Misses # 15.989 M/sec (scaled from 13.28%)
<not counted> Branch-Cache-Store-Referencees
<not counted> Branch-Cache-Store-Misses
<not counted> Branch-Cache-Prefetch-Referencees
<not counted> Branch-Cache-Prefetch-Misses1.214877275 seconds time elapsed.
Fix alignment, style problems and remove dead code
Increase limit for event_name() display
--- tools/perf/builtin-stat.c | 111 ++++++++++++++++++++++++++++++++++----------- 1 files changed, 84 insertions(+), 27 deletions(-)
tools/perf/builtin-stat.c +84 -27
@@ -32,6 +32,7 @@* Wu Fengguang <[email protected]> * Mike Galbraith <[email protected]> * Paul Mackerras <[email protected]> +* Jaswinder Singh <[email protected]> * * Released under the GPL v2. (and only v2, not any later version) */
@@ -45,32 +46,94 @@#include <sys/prctl.h> #include <math.h> -static struct perf_counter_attr default_attrs[MAX_COUNTERS ] = { - - { .type =PERF_TYPE_SOFTWARE , .config =PERF_COUNT_SW_TASK_CLOCK }, - { .type =PERF_TYPE_SOFTWARE , .config =PERF_COUNT_SW_CONTEXT_SWITCHES }, - { .type =PERF_TYPE_SOFTWARE , .config =PERF_COUNT_SW_CPU_MIGRATIONS }, - { .type =PERF_TYPE_SOFTWARE , .config =PERF_COUNT_SW_PAGE_FAULTS }, - - { .type =PERF_TYPE_HARDWARE , .config =PERF_COUNT_HW_CPU_CYCLES }, - { .type =PERF_TYPE_HARDWARE , .config =PERF_COUNT_HW_INSTRUCTIONS }, - { .type =PERF_TYPE_HARDWARE , .config =PERF_COUNT_HW_CACHE_REFERENCES }, - { .type =PERF_TYPE_HARDWARE , .config =PERF_COUNT_HW_CACHE_MISSES }, +#define CHW (x) .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_##x +#define CSW (x) .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_##x +#define CHCACHE (x, y, z) \ +.type = PERF_TYPE_HW_CACHE, \ +.config = (PERF_COUNT_HW_CACHE_##x | (PERF_COUNT_HW_CACHE_OP_##y << 8) |\ + (PERF_COUNT_HW_CACHE_RESULT_##z << 16)) + +static struct perf_counter_attr default_attrs[] = { + +/* Generalized Hardware events */ + {CHW (CPU_CYCLES ) }, + {CHW (INSTRUCTIONS ) }, + {CHW (CACHE_REFERENCES ) }, + {CHW (CACHE_MISSES ) }, + {CHW (BRANCH_INSTRUCTIONS ) }, + {CHW (BRANCH_MISSES ) }, + {CHW (BUS_CYCLES ) }, + +/* Generalized Software events */ + {CSW (CPU_CLOCK ) }, + {CSW (TASK_CLOCK ) }, + {CSW (PAGE_FAULTS ) }, + {CSW (PAGE_FAULTS_MIN ) }, + {CSW (PAGE_FAULTS_MAJ ) }, + {CSW (CONTEXT_SWITCHES ) }, + {CSW (CPU_MIGRATIONS ) }, + +/* Generalized Hardware cache counters events */ + {CHCACHE (L1D ,READ ,ACCESS ) }, + {CHCACHE (L1D ,READ ,MISS ) }, + {CHCACHE (L1D ,WRITE ,ACCESS ) }, + {CHCACHE (L1D ,WRITE ,MISS ) }, + {CHCACHE (L1D ,PREFETCH ,ACCESS ) }, + {CHCACHE (L1D ,PREFETCH ,MISS ) }, + + {CHCACHE (L1I ,READ ,ACCESS ) }, + {CHCACHE (L1I ,READ ,MISS ) }, + {CHCACHE (L1I ,WRITE ,ACCESS ) }, + {CHCACHE (L1I ,WRITE ,MISS ) }, + {CHCACHE (L1I ,PREFETCH ,ACCESS ) }, + {CHCACHE (L1I ,PREFETCH ,MISS ) }, + + {CHCACHE (LL ,READ ,ACCESS ) }, + {CHCACHE (LL ,READ ,MISS ) }, + {CHCACHE (LL ,WRITE ,ACCESS ) }, + {CHCACHE (LL ,WRITE ,MISS ) }, + {CHCACHE (LL ,PREFETCH ,ACCESS ) }, + {CHCACHE (LL ,PREFETCH ,MISS ) }, + + {CHCACHE (DTLB ,READ ,ACCESS ) }, + {CHCACHE (DTLB ,READ ,MISS ) }, + {CHCACHE (DTLB ,WRITE ,ACCESS ) }, + {CHCACHE (DTLB ,WRITE ,MISS ) }, + {CHCACHE (DTLB ,PREFETCH ,ACCESS ) }, + {CHCACHE (DTLB ,PREFETCH ,MISS ) }, + + {CHCACHE (ITLB ,READ ,ACCESS ) }, + {CHCACHE (ITLB ,READ ,MISS ) }, + {CHCACHE (ITLB ,WRITE ,ACCESS ) }, + {CHCACHE (ITLB ,WRITE ,MISS ) }, + {CHCACHE (ITLB ,PREFETCH ,ACCESS ) }, + {CHCACHE (ITLB ,PREFETCH ,MISS ) }, + + {CHCACHE (BPU ,READ ,ACCESS ) }, + {CHCACHE (BPU ,READ ,MISS ) }, + {CHCACHE (BPU ,WRITE ,ACCESS ) }, + {CHCACHE (BPU ,WRITE ,MISS ) }, + {CHCACHE (BPU ,PREFETCH ,ACCESS ) }, + {CHCACHE (BPU ,PREFETCH ,MISS ) }, }; -static int system_wide =0 ; -static int inherit =1 ; -static int verbose =0 ; +#define MAX_RUN 100static int fd[MAX_NR_CPUS ][MAX_COUNTERS ]; -static int target_pid =-1 ; +static int system_wide =0 ;static int nr_cpus =0 ; -static unsigned int page_size; +static int verbose =0 ; +static int run_idx =0 ; +static int run_count =1 ; +static int target_pid =-1 ; +static int inherit =1 ;static int scale =1 ; +static unsigned int page_size; +static const unsigned int default_count[] = {1000000 ,1000000 ,
@@ -80,17 +143,11 @@ static const unsigned int default_count[] = { 10000, }; -#define MAX_RUN 100 -static int run_count =1 ; -static int run_idx =0 ;static u64 event_res[MAX_RUN ][MAX_COUNTERS ][3 ];static u64 event_scaled[MAX_RUN ][MAX_COUNTERS ]; -//static u64 event_hist[MAX_RUN][MAX_COUNTERS][3]; - -static u64 runtime_nsecs[MAX_RUN ];static u64 walltime_nsecs[MAX_RUN ];static u64 runtime_cycles[MAX_RUN ];
@@ -119,7 +176,7 @@ static void create_perf_stat_counter(int counter)if (system_wide) {int cpu; -for (cpu =0 ; cpu < nr_cpus; cpu ++) { +for (cpu =0 ; cpu < nr_cpus; cpu++) { fd[cpu][counter] =sys_perf_counter_open (attr,-1 , cpu,-1 ,0 );if (fd[cpu][counter] <0 && verbose) {printf ("Error: counter %d, sys_perf_counter_open() syscall returned with %d (%s)\n" , counter, fd[cpu][counter],strerror (errno));
@@ -168,7 +225,7 @@ static void read_counter(int counter) count[0 ] = count[1 ] = count[2 ] =0 ; nv = scale ?3 :1 ; -for (cpu =0 ; cpu < nr_cpus; cpu ++) { +for (cpu =0 ; cpu < nr_cpus; cpu++) {if (fd[cpu][counter] <0 )continue ;
@@ -262,7 +319,7 @@ static void nsec_printout(int counter, u64 *count, u64 *noise) {double msecs = (double )count[0 ] /1000000 ; -fprintf (stderr," %14.6f %-20s" , msecs,event_name (counter)); +fprintf (stderr," %14.6f %-43s" , msecs,event_name (counter));if (attrs[counter].type ==PERF_TYPE_SOFTWARE && attrs[counter].config ==PERF_COUNT_SW_TASK_CLOCK ) {
@@ -276,7 +333,7 @@ static void nsec_printout(int counter, u64 *count, u64 *noise)static void abs_printout (int counter,u64 *count,u64 *noise) { -fprintf (stderr," %14Ld %-20s" , count[0 ],event_name (counter)); +fprintf (stderr," %14Ld %-43s" , count[0 ],event_name (counter)); if (runtime_cycles_avg && attrs[counter].type == PERF_TYPE_HARDWARE &&
@@ -491,7 +548,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix)usage_with_options (stat_usage, options);if (!nr_counters) - nr_counters =8 ; + nr_counters =ARRAY_SIZE (default_attrs); nr_cpus =sysconf (_SC_NPROCESSORS_ONLN);assert (nr_cpus <=MAX_NR_CPUS );
Signature · 1 line
1.6.0.6
Quoted 48 lines · Jaswinder Singh Rajput
Added more events not it looks like on AMD box :
./perf stat -- ls -lR >
/dev/nullPerformance counter stats for 'ls -lR':
2507744774 cycles # 2085.473 M/sec (scaled from 13.28%) 1515534968 instructions # 0.604 IPC (scaled from 13.28%) 783181797 cache-references # 651.304 M/sec (scaled from 36.36%) 18089523 cache-misses # 15.043 M/sec (scaled from 36.37%) 195550613 branches # 162.622 M/sec (scaled from 36.29%) 14623394 branch-misses # 12.161 M/sec (scaled from 36.29%) <not counted> bus-cycles 1203.182949 cpu-clock-msecs 1202.482671 task-clock-msecs # 0.990 CPUs 454 page-faults # 0.000 M/sec 454 minor-faults # 0.000 M/sec 0 major-faults # 0.000 M/sec 133 context-switches # 0.000 M/sec 1 CPU-migrations # 0.000 M/sec 744421154 L1-data-Cache-Load-Referencees # 619.070 M/sec (scaled from 13.20%) 5220656 L1-data-Cache-Load-Misses # 4.342 M/sec (scaled from 13.28%) 438576 L1-data-Cache-Store-Referencees # 0.365 M/sec (scaled from 13.36%) <not counted> L1-data-Cache-Store-Misses 1976596 L1-data-Cache-Prefetch-Referencees # 1.644 M/sec (scaled from 13.44%) 1644021 L1-data-Cache-Prefetch-Misses # 1.367 M/sec (scaled from 13.52%) 764273224 L1-instruction-Cache-Load-Referencees # 635.579 M/sec (scaled from 13.53%) 17242789 L1-instruction-Cache-Load-Misses # 14.339 M/sec (scaled from 13.53%) <not counted> L1-instruction-Cache-Store-Referencees <not counted> L1-instruction-Cache-Store-Misses 372621 L1-instruction-Cache-Prefetch-Referencees # 0.310 M/sec (scaled from 13.53%) <not counted> L1-instruction-Cache-Prefetch-Misses 22844109 L2-Cache-Load-Referencees # 18.997 M/sec (scaled from 13.53%) 2235733 L2-Cache-Load-Misses # 1.859 M/sec (scaled from 13.53%) 23949920 L2-Cache-Store-Referencees # 19.917 M/sec (scaled from 13.46%) <not counted> L2-Cache-Store-Misses <not counted> L2-Cache-Prefetch-Referencees <not counted> L2-Cache-Prefetch-Misses 732364670 Data-TLB-Cache-Load-Referencees # 609.044 M/sec (scaled from 13.45%) 16516548 Data-TLB-Cache-Load-Misses # 13.735 M/sec (scaled from 13.42%) <not counted> Data-TLB-Cache-Store-Referencees <not counted> Data-TLB-Cache-Store-Misses <not counted> Data-TLB-Cache-Prefetch-Referencees <not counted> Data-TLB-Cache-Prefetch-Misses 766865920 Instruction-TLB-Cache-Load-Referencees # 637.736 M/sec (scaled from 13.42%) 19981 Instruction-TLB-Cache-Load-Misses # 0.017 M/sec (scaled from 13.40%) <not counted> Instruction-TLB-Cache-Store-Referencees <not counted> Instruction-TLB-Cache-Store-Misses
Quoted 8 lines
<not counted> Instruction-TLB-Cache-Prefetch-Referencees
<not counted> Instruction-TLB-Cache-Prefetch-Misses
308272002 Branch-Cache-Load-Referencees # 256.363 M/sec (scaled from 13.33%)
19226358 Branch-Cache-Load-Misses # 15.989 M/sec (scaled from 13.28%)
<not counted> Branch-Cache-Store-Referencees
<not counted> Branch-Cache-Store-Misses
<not counted> Branch-Cache-Prefetch-Referencees
<not counted> Branch-Cache-Prefetch-MissesLooks useful - but it would be nice to not touch the default 'perf stat' output but instead offer a few 'sets' of pre-defined events, which can be specified in the event list, such as:
perf stat -e cache-events
perf stat -e all-cache-events
perf stat -e sw-events
Perhaps also a:
perf stat -e all
To get output from all counters that we know about.
Regex matching on event specifiers would be useful too - there's already regex code in perf-report, see the --parent option.
Ingo