Re: [PATCH] perf stat: Support metrics with perf stat --per-thread

From: Jiri Olsa
Date: Thu Oct 12 2017 - 07:30:34 EST

Next message: Peter Zijlstra: "Re: [PATCH 1/3] printk: Fix kdb_trap_printk placement"
Previous message: Jiri Olsa: "Re: [PATCH] perf stat: Support metrics with perf stat --per-thread"
In reply to: Jin, Yao: "Re: [PATCH] perf stat: Support metrics with perf stat --per-thread"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Thu, Oct 12, 2017 at 05:05:00PM +0800, Jin Yao wrote:

SNIP

> ---
> tools/perf/util/stat.c | 23 +++++++++++++++++++++++
> 1 file changed, 23 insertions(+)
>
> diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
> index 35e9848..1164e68 100644
> --- a/tools/perf/util/stat.c
> +++ b/tools/perf/util/stat.c
> @@ -314,6 +314,26 @@ static int process_counter_maps(struct perf_stat_config *config,
> return 0;
> }
>
> +static int process_aggr_thread_counter(struct perf_evsel *counter)
> +{
> + int nthreads = thread_map__nr(counter->threads);
> + int ncpus = cpu_map__nr(counter->cpus);
> + int cpu, thread;
> + u64 tmp;
> +
> + for (thread = 0; thread < nthreads; thread++) {
> + u64 val = 0;
> +
> + for (cpu = 0; cpu < ncpus; cpu++)
> + val += perf_counts(counter->counts, cpu, thread)->val;
> +
> + tmp = val * counter->scale;
> + perf_stat__update_shadow_stats(counter, &tmp, 0);

also maybe we should move scale mult. into perf_stat__update_shadow_stats,
something like below (on top of my earlier patch)

not much tested so far.. but looks ok ;-)

jirka

---
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 9f2f07c0237d..4e109e6ba341 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1267,8 +1267,7 @@ static void aggr_update_shadow(void)
continue;
val += perf_counts(counter->counts, cpu, 0)->val;
}
- val = val * counter->scale;
- perf_stat__update_shadow_stats(counter, &val,
+ perf_stat__update_shadow_stats(counter, val,
first_shadow_cpu(counter, id));
}
}
diff --git a/tools/perf/util/stat-shadow.c b/tools/perf/util/stat-shadow.c
index a2c12d1ef32a..e66ac456d103 100644
--- a/tools/perf/util/stat-shadow.c
+++ b/tools/perf/util/stat-shadow.c
@@ -178,58 +178,60 @@ void perf_stat__reset_shadow_stats(void)
* more semantic information such as miss/hit ratios,
* instruction rates, etc:
*/
-void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 *count,
+void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 count,
int cpu)
{
int ctx = evsel_context(counter);

+ count *= counter->scale;
+
if (perf_evsel__match(counter, SOFTWARE, SW_TASK_CLOCK) ||
perf_evsel__match(counter, SOFTWARE, SW_CPU_CLOCK))
- update_stats(&runtime_nsecs_stats[cpu], count[0]);
+ update_stats(&runtime_nsecs_stats[cpu], count);
else if (perf_evsel__match(counter, HARDWARE, HW_CPU_CYCLES))
- update_stats(&runtime_cycles_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_cycles_stats[ctx][cpu], count);
else if (perf_stat_evsel__is(counter, CYCLES_IN_TX))
- update_stats(&runtime_cycles_in_tx_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_cycles_in_tx_stats[ctx][cpu], count);
else if (perf_stat_evsel__is(counter, TRANSACTION_START))
- update_stats(&runtime_transaction_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_transaction_stats[ctx][cpu], count);
else if (perf_stat_evsel__is(counter, ELISION_START))
- update_stats(&runtime_elision_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_elision_stats[ctx][cpu], count);
else if (perf_stat_evsel__is(counter, TOPDOWN_TOTAL_SLOTS))
- update_stats(&runtime_topdown_total_slots[ctx][cpu], count[0]);
+ update_stats(&runtime_topdown_total_slots[ctx][cpu], count);
else if (perf_stat_evsel__is(counter, TOPDOWN_SLOTS_ISSUED))
- update_stats(&runtime_topdown_slots_issued[ctx][cpu], count[0]);
+ update_stats(&runtime_topdown_slots_issued[ctx][cpu], count);
else if (perf_stat_evsel__is(counter, TOPDOWN_SLOTS_RETIRED))
- update_stats(&runtime_topdown_slots_retired[ctx][cpu], count[0]);
+ update_stats(&runtime_topdown_slots_retired[ctx][cpu], count);
else if (perf_stat_evsel__is(counter, TOPDOWN_FETCH_BUBBLES))
- update_stats(&runtime_topdown_fetch_bubbles[ctx][cpu],count[0]);
+ update_stats(&runtime_topdown_fetch_bubbles[ctx][cpu],count);
else if (perf_stat_evsel__is(counter, TOPDOWN_RECOVERY_BUBBLES))
- update_stats(&runtime_topdown_recovery_bubbles[ctx][cpu], count[0]);
+ update_stats(&runtime_topdown_recovery_bubbles[ctx][cpu], count);
else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_FRONTEND))
- update_stats(&runtime_stalled_cycles_front_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_stalled_cycles_front_stats[ctx][cpu], count);
else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_BACKEND))
- update_stats(&runtime_stalled_cycles_back_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_stalled_cycles_back_stats[ctx][cpu], count);
else if (perf_evsel__match(counter, HARDWARE, HW_BRANCH_INSTRUCTIONS))
- update_stats(&runtime_branches_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_branches_stats[ctx][cpu], count);
else if (perf_evsel__match(counter, HARDWARE, HW_CACHE_REFERENCES))
- update_stats(&runtime_cacherefs_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_cacherefs_stats[ctx][cpu], count);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1D))
- update_stats(&runtime_l1_dcache_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_l1_dcache_stats[ctx][cpu], count);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1I))
- update_stats(&runtime_ll_cache_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_ll_cache_stats[ctx][cpu], count);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_LL))
- update_stats(&runtime_ll_cache_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_ll_cache_stats[ctx][cpu], count);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_DTLB))
- update_stats(&runtime_dtlb_cache_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_dtlb_cache_stats[ctx][cpu], count);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_ITLB))
- update_stats(&runtime_itlb_cache_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_itlb_cache_stats[ctx][cpu], count);
else if (perf_stat_evsel__is(counter, SMI_NUM))
- update_stats(&runtime_smi_num_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_smi_num_stats[ctx][cpu], count);
else if (perf_stat_evsel__is(counter, APERF))
- update_stats(&runtime_aperf_stats[ctx][cpu], count[0]);
+ update_stats(&runtime_aperf_stats[ctx][cpu], count);

if (counter->collect_stat) {
struct saved_value *v = saved_value_lookup(counter, cpu, true);
- update_stats(&v->stats, count[0]);
+ update_stats(&v->stats, count);
}
}

diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
index 6f94f3a2282e..c1cf42d04826 100644
--- a/tools/perf/util/stat.c
+++ b/tools/perf/util/stat.c
@@ -259,7 +259,6 @@ process_counter_values(struct perf_stat_config *config, struct perf_evsel *evsel
struct perf_counts_values *aggr = &evsel->counts->aggr;
static struct perf_counts_values zero;
bool skip = false;
- u64 val;

if (check_per_pkg(evsel, count, cpu, &skip)) {
pr_err("failed to read per-pkg counter\n");
@@ -278,11 +277,9 @@ process_counter_values(struct perf_stat_config *config, struct perf_evsel *evsel
perf_evsel__compute_deltas(evsel, cpu, thread, count);
perf_counts_values__scale(count, config->scale, NULL);
if (config->aggr_mode == AGGR_NONE)
- perf_stat__update_shadow_stats(evsel, count->values, cpu);
- if (config->aggr_mode == AGGR_THREAD) {
- val = count->val * evsel->scale;
- perf_stat__update_shadow_stats(evsel, &val, 0);
- }
+ perf_stat__update_shadow_stats(evsel, count->val, cpu);
+ if (config->aggr_mode == AGGR_THREAD)
+ perf_stat__update_shadow_stats(evsel, count->val, 0);
break;
case AGGR_GLOBAL:
aggr->val += count->val;
@@ -325,7 +322,6 @@ int perf_stat_process_counter(struct perf_stat_config *config,
struct perf_counts_values *aggr = &counter->counts->aggr;
struct perf_stat_evsel *ps = counter->priv;
u64 *count = counter->counts->aggr.values;
- u64 val;
int i, ret;

aggr->val = aggr->ena = aggr->run = 0;
@@ -365,8 +361,7 @@ int perf_stat_process_counter(struct perf_stat_config *config,
/*
* Save the full runtime - to allow normalization during printout:
*/
- val = counter->scale * *count;
- perf_stat__update_shadow_stats(counter, &val, 0);
+ perf_stat__update_shadow_stats(counter, *count, 0);

return 0;
}
diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
index 47915df346fb..490b78aa7230 100644
--- a/tools/perf/util/stat.h
+++ b/tools/perf/util/stat.h
@@ -82,7 +82,7 @@ typedef void (*new_line_t )(void *ctx);

void perf_stat__init_shadow_stats(void);
void perf_stat__reset_shadow_stats(void);
-void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 *count,
+void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 count,
int cpu);
struct perf_stat_output_ctx {
void *ctx;

Next message: Peter Zijlstra: "Re: [PATCH 1/3] printk: Fix kdb_trap_printk placement"
Previous message: Jiri Olsa: "Re: [PATCH] perf stat: Support metrics with perf stat --per-thread"
In reply to: Jin, Yao: "Re: [PATCH] perf stat: Support metrics with perf stat --per-thread"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]