Messages in this thread | | | Date | Fri, 6 Oct 2023 19:13:19 -0700 | Subject | [PATCH v1 0/7] PMU performance improvements | From | Ian Rogers <> |
| |
Performance improvements to pmu scanning by holding onto the event/metric tables for a cpuid (avoid regular expression comparisons) and by lazily computing the default perf_event_attr for a PMU.
Before % Running 'internals/pmu-scan' benchmark: Computing performance of sysfs PMU event scan for 100 times Average core PMU scanning took: 251.990 usec (+- 4.009 usec) Average PMU scanning took: 3222.460 usec (+- 211.234 usec) % Running 'internals/pmu-scan' benchmark: Computing performance of sysfs PMU event scan for 100 times Average core PMU scanning took: 260.120 usec (+- 7.905 usec) Average PMU scanning took: 3228.995 usec (+- 211.196 usec) % Running 'internals/pmu-scan' benchmark: Computing performance of sysfs PMU event scan for 100 times Average core PMU scanning took: 252.310 usec (+- 3.980 usec) Average PMU scanning took: 3220.675 usec (+- 210.844 usec)
After: % Running 'internals/pmu-scan' benchmark: Computing performance of sysfs PMU event scan for 100 times Average core PMU scanning took: 28.530 usec (+- 0.602 usec) Average PMU scanning took: 275.725 usec (+- 18.253 usec) % Running 'internals/pmu-scan' benchmark: Computing performance of sysfs PMU event scan for 100 times Average core PMU scanning took: 28.720 usec (+- 0.446 usec) Average PMU scanning took: 271.015 usec (+- 18.762 usec) % Running 'internals/pmu-scan' benchmark: Computing performance of sysfs PMU event scan for 100 times Average core PMU scanning took: 31.040 usec (+- 0.612 usec) Average PMU scanning took: 267.340 usec (+- 17.209 usec)
Measuring the pmu-scan benchmark on a Tigerlake laptop: core PMU scanning is reduced to 11.5% of the previous execution time, all PMU scanning is reduced to 8.4% of the previous execution time. There is a 4.3% reduction in openat system calls.
Ian Rogers (7): perf pmu: Rename perf_pmu__get_default_config to perf_pmu__arch_init perf intel-pt: Move PMU initialization from default config code perf arm-spe: Move PMU initialization from default config code perf pmu: Const-ify file APIs perf pmu: Const-ify perf_pmu__config_terms perf pmu-events: Remember the events and metrics table perf pmu: Lazily compute default config
tools/perf/arch/arm/util/cs-etm.c | 13 ++------ tools/perf/arch/arm/util/pmu.c | 10 +++--- tools/perf/arch/arm64/util/arm-spe.c | 48 +++++++++++++--------------- tools/perf/arch/s390/util/pmu.c | 3 +- tools/perf/arch/x86/util/intel-pt.c | 27 +++++++--------- tools/perf/arch/x86/util/pmu.c | 6 ++-- tools/perf/pmu-events/jevents.py | 48 ++++++++++++++++------------ tools/perf/util/arm-spe.h | 4 ++- tools/perf/util/cs-etm.h | 2 +- tools/perf/util/intel-pt.h | 3 +- tools/perf/util/parse-events.c | 12 +++---- tools/perf/util/pmu.c | 39 +++++++++++----------- tools/perf/util/pmu.h | 18 ++++++----- tools/perf/util/python.c | 2 +- 14 files changed, 117 insertions(+), 118 deletions(-)
-- 2.42.0.609.gbb76f46606-goog
| |