Re: [PATCH v1 6/7] perf pmu-events: Remember the events and metrics table

[Yang Jihong]

Hello,

On 2023/10/8 13:49, Ian Rogers wrote:
> On Sat, Oct 7, 2023 at 8:39 PM Yang Jihong <yangjihong1@xxxxxxxxxx> wrote:
> > Hello,
> >
> > On 2023/10/7 10:13, Ian Rogers wrote:
> > > strcmp_cpuid_str performs regular expression comparisons. Avoid
> > > repeated computation of the table by remembering the table in a
> > > static.
> > >
> > > Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
> > > ---
> > >    tools/perf/pmu-events/jevents.py | 48 +++++++++++++++++++-------------
> > >    1 file changed, 28 insertions(+), 20 deletions(-)
> > >
> > > diff --git a/tools/perf/pmu-events/jevents.py b/tools/perf/pmu-events/jevents.py
> > > index fd009752b427..8d8d5088c53c 100755
> > > --- a/tools/perf/pmu-events/jevents.py
> > > +++ b/tools/perf/pmu-events/jevents.py
> > > @@ -978,28 +978,32 @@ int pmu_metrics_table__for_each_metric(const struct pmu_metrics_table *table,
> > >
> > >    const struct pmu_events_table *perf_pmu__find_events_table(struct perf_pmu *pmu)
> > >    {
> > > -        const struct pmu_events_table *table = NULL;
> > > -        char *cpuid = perf_pmu__getcpuid(pmu);
> > > +        static const struct pmu_events_table *table;
> > >            size_t i;
> > >
> > > -        /* on some platforms which uses cpus map, cpuid can be NULL for
> > > -         * PMUs other than CORE PMUs.
> > > -         */
> > > -        if (!cpuid)
> > > -                return NULL;
> > > -
> > > -        i = 0;
> > > -        for (;;) {
> > > -                const struct pmu_events_map *map = &pmu_events_map[i++];
> > > -                if (!map->arch)
> > > -                        break;
> > > -
> > > -                if (!strcmp_cpuid_str(map->cpuid, cpuid)) {
> > > -                        table = &map->event_table;
> > > -                        break;
> > > +        if (!table) {
> > If there is no matched table in pmu_events_map,
> > perf_pmu__find_events_table() will enter this branch for repeated search
> > each time.
> > Or do we need to use another variable to indicate whether the search has
> > been performed?
>
> Agreed, the behavior will match the existing behavior. Longer term I
> want to remove this code. Do you have a scenario we should optimize
> for here?

Yes, the CPU of the environment I'm using is "AuthenticAMD-15-6B-1" (not 
in the pmu_events_map).
As a result, the search is repeated every time.
(If `perf record true` is executed once, the search is repeated for 6 
times.)

This commit avoids repeated lookups to improve performance,
so if it's feasible, is it best to consider improving performance in 
this case as well?

Thanks,
Yang