[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <Z_dp7E2wtSek-KHo@z2>
Date: Wed, 9 Apr 2025 23:49:16 -0700
From: Namhyung Kim <namhyung@...nel.org>
To: Ian Rogers <irogers@...gle.com>
Cc: Peter Zijlstra <peterz@...radead.org>, Ingo Molnar <mingo@...hat.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>,
Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Jiri Olsa <jolsa@...nel.org>,
Adrian Hunter <adrian.hunter@...el.com>,
Kan Liang <kan.liang@...ux.intel.com>,
James Clark <james.clark@...aro.org>, Xu Yang <xu.yang_2@....com>,
John Garry <john.g.garry@...cle.com>,
Dominique Martinet <asmadeus@...ewreck.org>,
Thomas Richter <tmricht@...ux.ibm.com>,
Weilin Wang <weilin.wang@...el.com>,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v1 0/3] Metric related performance improvements
Hi Ian,
On Wed, Apr 09, 2025 at 09:45:29PM -0700, Ian Rogers wrote:
> The "PMU JSON event tests" have been running slowly, these changes
> target improving them with an improvement of the test running 8 to 10
> times faster.
>
> The first patch changes from searching through all aliases by name in
> a list to using a hashmap. Doing a fast hashmap__find means testing
> for having an event needn't load from disk if an event is already
> present.
>
> The second patch switch the fncache to use a hashmap rather than its
> own hashmap with a limited number of buckets. When there are many
> filename queries, such as with a test, there are many collisions with
> the previous fncache approach leading to linear searching of the
> entries.
>
> The final patch adds a find function for metrics. Normally metrics can
> match by name and group, however, only name matching happens when one
> metric refers to another. As we test every "id" in a metric to see if
> it is a metric, the find function can dominate performance as it
> linearly searches all metrics. Add a find function for the metrics
> table so that a metric can be found by name with a binary search.
>
> Before these changes:
> ```
> $ time perf test -v 10
> 10: PMU JSON event tests :
> 10.1: PMU event table sanity : Ok
> 10.2: PMU event map aliases : Ok
> 10.3: Parsing of PMU event table metrics : Ok
> 10.4: Parsing of PMU event table metrics with fake PMUs : Ok
> 10.5: Parsing of metric thresholds with fake PMUs : Ok
>
> real 0m18.499s
> user 0m18.150s
> sys 0m3.273s
> ```
>
> After these changes:
> ```
> $ time perf test -v 10
> 10: PMU JSON event tests :
> 10.1: PMU event table sanity : Ok
> 10.2: PMU event map aliases : Ok
> 10.3: Parsing of PMU event table metrics : Ok
> 10.4: Parsing of PMU event table metrics with fake PMUs : Ok
> 10.5: Parsing of metric thresholds with fake PMUs : Ok
>
> real 0m2.338s
> user 0m1.797s
> sys 0m2.186s
> ```
Great, I also see the speedup on my machine from 32s to 3s.
Tested-by: Namhyung Kim <namhyung@...nel.org>
Thanks,
Namhyung
>
> Ian Rogers (3):
> perf pmu: Change aliases from list to hashmap
> perf fncache: Switch to using hashmap
> perf metricgroup: Binary search when resolving referred to metrics
>
> tools/perf/builtin-stat.c | 6 +-
> tools/perf/pmu-events/empty-pmu-events.c | 66 ++++++++-
> tools/perf/pmu-events/jevents.py | 66 ++++++++-
> tools/perf/pmu-events/pmu-events.h | 23 +++-
> tools/perf/tests/pmu-events.c | 129 +++++++++--------
> tools/perf/util/fncache.c | 69 +++++-----
> tools/perf/util/fncache.h | 1 -
> tools/perf/util/hwmon_pmu.c | 43 +++---
> tools/perf/util/metricgroup.c | 102 ++++++--------
> tools/perf/util/metricgroup.h | 2 +-
> tools/perf/util/pmu.c | 167 +++++++++++++++--------
> tools/perf/util/pmu.h | 4 +-
> tools/perf/util/srccode.c | 4 +-
> tools/perf/util/tool_pmu.c | 17 +--
> 14 files changed, 430 insertions(+), 269 deletions(-)
>
> --
> 2.49.0.504.g3bcea36a83-goog
>
Powered by blists - more mailing lists