[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20210923074616.674826-1-irogers@google.com>
Date: Thu, 23 Sep 2021 00:46:03 -0700
From: Ian Rogers <irogers@...gle.com>
To: 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@...hat.com>,
Namhyung Kim <namhyung@...nel.org>,
linux-kernel@...r.kernel.org, Andi Kleen <ak@...ux.intel.com>,
Jin Yao <yao.jin@...ux.intel.com>,
John Garry <john.garry@...wei.com>,
Paul Clarke <pc@...ibm.com>, kajoljain <kjain@...ux.ibm.com>,
linux-perf-users@...r.kernel.org
Cc: Stephane Eranian <eranian@...gle.com>,
Sandeep Dasgupta <sdasgup@...gle.com>,
Ian Rogers <irogers@...gle.com>
Subject: [PATCH v9 00/13] Don't compute events that won't be used in a metric.
For a metric like:
EVENT1 if #smt_on else EVENT2
currently EVENT1 and EVENT2 will be measured and then when the metric
is reported EVENT1 or EVENT2 will be printed depending on the value
from smt_on() during the expr parsing. Computing both events is
unnecessary and can lead to multiplexing as discussed in this thread:
https://lore.kernel.org/lkml/20201110100346.2527031-1-irogers@google.com/
This change modifies expression parsing so that constants are
considered when building the set of ids (events) and only events not
contributing to a constant value are measured.
v9. adds a missing memory allocation failure check in the pmu-metrics
test. A memory allocation failure in union returns NULL. Some
parser debug on IDs is removed. "Modify code layout" is broken
apart into 3 changes. "Don't compute unused events" is broken
apart into 4 changes with the tests merged into the change that
adds the corresponding optimization. This is trying to address
feedback from Jiri Olsa <jolsa@...hat.com>. The unmodified
patches have Reviewed-by: Andi Kleen <ak@...ux.intel.com> added,
although the overall patch set is largely the same as v8 which was
fully reviewed-by.
v8. rebases, adds an ability to compute metrics with no events and
further breaks apart the "Don't compute unused events" part of the
change as requested by Jiri Olsa <jolsa@...hat.com>.
v7. fixes the fix to be in the correct patch.
v6. rebases and fixes issues raised by Namhyung Kim <namhyung@...nel.org>,
a memory leak and a function comment.
v5. uses macros to reduce boiler plate in patch 5/5 as suggested by
Andi Kleen <ak@...ux.intel.com>.
v4. reduces references to BOTTOM/NAN in patch 5/5 by using utility
functions. It improves comments and fixes an unnecessary union in a
peephole optimization.
v3. fixes an assignment in patch 2/5. In patch 5/5 additional comments
are added and useless frees are replaced by asserts. A new peephole
optimization is added for the case CONST IF expr ELSE CONST, where the
the constants are identical, as we don't need to evaluate the IF
condition.
v2. is a rebase.
Ian Rogers (13):
perf metric: Restructure struct expr_parse_ctx.
perf metric: Use NAN for missing event IDs.
perf expr: Remove unused headers and inline d_ratio
perf expr: Separate token declataion from type
perf expr: Use macros for operators
perf expr: Move actions to the left.
perf metric: Rename expr__find_other.
perf metric: Add utilities to work on ids map.
perf metric: Allow metrics with no events
perf expr: Merge find_ids and regular parsing
perf expr: Propagate constants for binary operations
perf metric: Don't compute unused events
perf metric: Avoid events for an 'if' constant result
tools/perf/tests/expr.c | 160 ++++++++++++-----
tools/perf/tests/pmu-events.c | 54 +++---
tools/perf/util/expr.c | 121 +++++++++++--
tools/perf/util/expr.h | 20 ++-
tools/perf/util/expr.l | 9 -
tools/perf/util/expr.y | 325 +++++++++++++++++++++++++---------
tools/perf/util/metricgroup.c | 145 ++++++++-------
tools/perf/util/stat-shadow.c | 54 +++---
8 files changed, 620 insertions(+), 268 deletions(-)
--
2.33.0.464.g1972c5931b-goog
Powered by blists - more mailing lists