Fix perf on Intel hybrid CPUs
Ian Rogers
reviewerv3 cover Fix perf on Intel hybrid CPUs
TL;DR: hybrid doesn't crash, json metrics work on hybrid on both PMUs or individually, event parsing doesn't always scan all PMUs, more and new tests that also run without hybrid, less code.
The first 3 patches are proposed to address Linux 6.4 issues with the remainder to be the basis for 6.5. The first patch is a 1 liner, plus comment, workaround for Intel hybrid platforms crashing perf stat when running with no events/metrics. The second patch works around an issue with platforms like Skylake where certain events fail to open at higher paranoia levels even if kernel and hypervisor profiling is disabled. Events are marked as skippable and not placed in the output, while remaining events are and metrics are computed. The third patch makes it so that TopdownL1 and TopdownL2 json metric groups don't group events on Intel, this is done via a new json flag. Adding the json flag in each of the json files is two-thirds of the 300 lines of these patches. Patch 42 removes the disabling of TopdownL1 on hybrid as the underlying problems are fixed. The json file updates are created with the usual script updated in:
https://github.com/intel/perfmon/pull/73
The next 5 patches avoid grouping events for metrics where is could never succeed and were previously posted as:
"perf vendor events intel: Add xxx metric constraints"
https://lore.kernel.org/all/[email protected]/
Next are some general and test improvements.
Next event parsing is rewritten to not scan all PMUs for the benefit of raw and legacy cache parsing, instead these are handled by the lexer and a new term type. This ultimately removes the need for the event parser for hybrid to be recursive as legacy cache can be just a term. Tests are re-enabled for events with hyphens, so AMD's branch-brs event is now parsable.
The cputype option is made a generic pmu filter flag and is tested even on non-hybrid systems.
The final patches address specific json metric issues on hybrid, in both the json metrics and the metric code.
The patches add slightly more code than they remove, in areas like better json metric constraints and tests, but in the core util code, the removal of hybrid is a net reduction:
20 files changed, 641 insertions(+), 960 deletions(-)
Sample output is contained in the v1 patch set:
https://lore.kernel.org/lkml/[email protected]/
Tested on Tigerlake, Skylake and Alderlake CPUs.
The v3 patch set:
- for Linux 6.4 the first 5 patches are recommended:
- changes skippable evsels to always print in case short running benchmarks meet the 0 enable and 0 count condition (suggested by Stephane).
- changes metrics to show "nan" so that it is more obvious they couldn't be computed (suggested by Stephane).
- fixes a reordering issue that reordered events that lacked a group, especially when the core PMU isn't "cpu".
- for Linux 6.5 added extra hybrid type tests in the x86 hybrid test as suggested by Kan.
- note, the patches aren't rebased against the tmp.perf-tools branch meaning patches 1 and 11 should be dropped there.
The v2 patch set:
- restructures the 3 Linux 6.4 patches first;
- makes it so that is_event_supported isn't called during core event parsing;
- displays skippable 0 count events that were enabled;
- addresses output formatting review comments;
- addresses some test issues and an uninitialized variable use in strchr;
- addresses checkpatch.pl reported issues;
- adds reviewed-by to some patches.
Ian Rogers (46):
perf stat: Disable TopdownL1 on hybrid
perf metric: Change divide by zero and !support events behavior perf stat: Introduce skippable evsels
perf metric: Json flag to not group events if gathering a metric group perf parse-events: Don't reorder ungrouped events by pmu perf vendor events intel: Add alderlake metric constraints perf vendor events intel: Add icelake metric constraints perf vendor events intel: Add icelakex metric constraints perf vendor events intel: Add sapphirerapids metric constraints perf vendor events intel: Add tigerlake metric constraints perf stat: Avoid segv on counter->name
perf test: Test more sysfs events
perf test: Use valid for PMU tests
perf test: Mask config then test
perf test: Test more with config_cache
perf test: Roundtrip name, don't assume 1 event per name perf parse-events: Set attr.type to PMU type early
perf parse-events: Set pmu_name whenever a pmu is given perf print-events: Avoid unnecessary strlist
perf parse-events: Avoid scanning PMUs before parsing
perf evsel: Modify group pmu name for software events
perf test: Move x86 hybrid tests to arch/x86
perf test x86 hybrid: Update test expectations
perf test x86 hybrid: Add hybrid extended type checks
perf parse-events: Support PMUs for legacy cache events perf parse-events: Wildcard legacy cache events
perf print-events: Print legacy cache events for each PMU perf parse-events: Support wildcards on raw events
perf parse-events: Remove now unused hybrid logic
perf parse-events: Minor type safety cleanup
perf parse-events: Add pmu filter
perf stat: Make cputype filter generic
perf test: Add cputype testing to perf stat
perf test: Fix parse-events tests for >1 core PMU
perf parse-events: Support hardware events as terms
perf parse-events: Avoid error when assigning a term
perf parse-events: Avoid error when assigning a legacy cache term perf parse-events: Don't auto merge hybrid wildcard events perf parse-events: Don't reorder atom cpu events
perf metrics: Be PMU specific for referenced metrics.
perf stat: Command line PMU metric filtering
perf vendor events intel: Correct alderlake metrics
perf jevents: Don't rewrite metrics across PMUs
perf metrics: Be PMU specific in event match
perf stat: Don't disable TopdownL1 metric on hybrid
perf parse-events: Reduce scope of is_event_supported
tools/perf/arch/x86/include/arch-tests.h | 1 + tools/perf/arch/x86/tests/Build | 1 + tools/perf/arch/x86/tests/arch-tests.c | 10 + tools/perf/arch/x86/tests/hybrid.c | 287 ++++++ tools/perf/arch/x86/util/evlist.c | 4 +- tools/perf/builtin-list.c | 19 +- tools/perf/builtin-record.c | 13 +- tools/perf/builtin-stat.c | 73 +- tools/perf/builtin-top.c | 5 +- tools/perf/builtin-trace.c | 5 +- .../arch/x86/alderlake/adl-metrics.json | 275 +++--- .../arch/x86/alderlaken/adln-metrics.json | 20 +- .../arch/x86/broadwell/bdw-metrics.json | 12 + .../arch/x86/broadwellde/bdwde-metrics.json | 12 + .../arch/x86/broadwellx/bdx-metrics.json | 12 + .../arch/x86/cascadelakex/clx-metrics.json | 12 + .../arch/x86/haswell/hsw-metrics.json | 12 + .../arch/x86/haswellx/hsx-metrics.json | 12 + .../arch/x86/icelake/icl-metrics.json | 23 + .../arch/x86/icelakex/icx-metrics.json | 23 + .../arch/x86/ivybridge/ivb-metrics.json | 12 + .../arch/x86/ivytown/ivt-metrics.json | 12 + .../arch/x86/jaketown/jkt-metrics.json | 12 + .../arch/x86/sandybridge/snb-metrics.json | 12 + .../arch/x86/sapphirerapids/spr-metrics.json | 23 + .../arch/x86/skylake/skl-metrics.json | 12 + .../arch/x86/skylakex/skx-metrics.json | 12 + .../arch/x86/tigerlake/tgl-metrics.json | 23 + tools/perf/pmu-events/jevents.py | 10 +- tools/perf/pmu-events/metric.py | 28 +- tools/perf/pmu-events/metric_test.py | 6 +- tools/perf/pmu-events/pmu-events.h | 2 + tools/perf/tests/evsel-roundtrip-name.c | 119 +-- tools/perf/tests/expr.c | 3 +- tools/perf/tests/parse-events.c | 836 +++++++++--------- tools/perf/tests/parse-metric.c | 1 + tools/perf/tests/pmu-events.c | 12 +- tools/perf/tests/shell/stat.sh | 44 + tools/perf/util/Build | 1 - tools/perf/util/evlist.h | 1 - tools/perf/util/evsel.c | 30 +- tools/perf/util/evsel.h | 1 + tools/perf/util/expr.y | 6 +- tools/perf/util/metricgroup.c | 111 ++- tools/perf/util/metricgroup.h | 3 +- tools/perf/util/parse-events-hybrid.c | 214 ----- tools/perf/util/parse-events-hybrid.h | 25 - tools/perf/util/parse-events.c | 720 ++++++--------- tools/perf/util/parse-events.h | 63 +- tools/perf/util/parse-events.l | 108 +-- tools/perf/util/parse-events.y | 222 ++--- tools/perf/util/pmu-hybrid.c | 20 - tools/perf/util/pmu-hybrid.h | 1 - tools/perf/util/pmu.c | 16 +- tools/perf/util/pmu.h | 3 + tools/perf/util/pmus.c | 25 +- tools/perf/util/pmus.h | 3 + tools/perf/util/print-events.c | 124 ++- tools/perf/util/stat-display.c | 4 +- tools/perf/util/stat-shadow.c | 25 +- 60 files changed, 2045 insertions(+), 1691 deletions(-) create mode 100644 tools/perf/arch/x86/tests/hybrid.c delete mode 100644 tools/perf/util/parse-events-hybrid.c delete mode 100644 tools/perf/util/parse-events-hybrid.h
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 1/46 perf stat: Disable TopdownL1 on hybrid
Bugs with event parsing, event grouping and metrics causes the TopdownL1 metricgroup to crash the perf command. Temporarily disable the group if no events/metrics are spcecified.
--- tools/perf/builtin-stat.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)
tools/perf/builtin-stat.c +7 -1
@@ -1885,7 +1885,12 @@ static int add_default_attributes(void)* Add TopdownL1 metrics if they exist. To minimize * multiplexing, don't request threshold computation. */ -if (metricgroup__has_metric ("TopdownL1" ) && +/* +* TODO: TopdownL1 is disabled on hybrid CPUs to avoid a crashes +* caused by exposing latent bugs. This is fixed properly in: +* https://lore.kernel.org/lkml/[email protected]/ +*/ +if (metricgroup__has_metric ("TopdownL1" ) && !perf_pmu__has_hybrid () && metricgroup__parse_groups(evsel_list, "TopdownL1", /*metric_no_group=*/false, /*metric_no_merge=*/false,
@@ -1894,6 +1899,7 @@ static int add_default_attributes(void) stat_config.system_wide, &stat_config.metric_events) < 0)return -1 ; +/* Platform specific attrs */ if (evlist__add_default_attrs (evsel_list, default_null_attrs) <0 )return -1 ;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 2/46 perf metric: Change divide by zero and !support events behavior
In reply to Ian RogersDivision by zero causes expression parsing to fail and no metric to be generated. This can mean for short running benchmarks metrics are not shown. Change the behavior to make the value nan, which gets shown like:
'''
$ perf stat -M TopdownL2 true
Performance counter stats for 'true':
1,031,492 INST_RETIRED.ANY # nan % tma_fetch_bandwidth
# nan % tma_heavy_operations
# nan % tma_light_operations
29,304 CPU_CLK_UNHALTED.REF_XCLK # nan % tma_fetch_latency
# nan % tma_branch_mispredicts
# nan % tma_machine_clears
# nan % tma_core_bound
# nan % tma_memory_bound
2,658,319 IDQ_UOPS_NOT_DELIVERED.CORE
11,167 EXE_ACTIVITY.BOUND_ON_STORES
262,058 EXE_ACTIVITY.1_PORTS_UTIL
<not counted> BR_MISP_RETIRED.ALL_BRANCHES (0.00%)
<not counted> INT_MISC.RECOVERY_CYCLES_ANY (0.00%)
<not counted> CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE (0.00%)
<not counted> CPU_CLK_UNHALTED.THREAD (0.00%)
<not counted> UOPS_RETIRED.RETIRE_SLOTS (0.00%)
<not counted> CYCLE_ACTIVITY.STALLS_MEM_ANY (0.00%)
<not counted> UOPS_RETIRED.MACRO_FUSED (0.00%)
<not counted> IDQ_UOPS_NOT_DELIVERED.CYCLES_0_UOPS_DELIV.CORE (0.00%)
<not counted> EXE_ACTIVITY.2_PORTS_UTIL (0.00%)
<not counted> CYCLE_ACTIVITY.STALLS_TOTAL (0.00%)
<not counted> MACHINE_CLEARS.COUNT (0.00%)
<not counted> UOPS_ISSUED.ANY (0.00%) 0.002864879 seconds time elapsed
0.003012000 seconds user
0.000000000 seconds sys'''
When events aren't supported a count of 0 can be confusing and make metrics look meaningful. Change these to be nan also which, with the next change, gets shown like:
'''
$ perf stat true Performance counter stats for 'true':
1.25 msec task-clock:u # 0.387 CPUs utilized
0 context-switches:u # 0.000 /sec
0 cpu-migrations:u # 0.000 /sec
46 page-faults:u # 36.702 K/sec
255,942 cycles:u # 0.204 GHz (88.66%)
123,046 instructions:u # 0.48 insn per cycle
28,301 branches:u # 22.580 M/sec
2,489 branch-misses:u # 8.79% of all branches
4,719 CPU_CLK_UNHALTED.REF_XCLK:u # 3.765 M/sec
# nan % tma_frontend_bound
# nan % tma_retiring
# nan % tma_backend_bound
# nan % tma_bad_speculation
344,855 IDQ_UOPS_NOT_DELIVERED.CORE:u # 275.147 M/sec
<not supported> INT_MISC.RECOVERY_CYCLES_ANY:u
<not counted> CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE:u (0.00%)
<not counted> CPU_CLK_UNHALTED.THREAD:u (0.00%)
<not counted> UOPS_RETIRED.RETIRE_SLOTS:u (0.00%)
<not counted> UOPS_ISSUED.ANY:u (0.00%) 0.003238142 seconds time elapsed
0.000000000 seconds user
0.003434000 seconds sys'''
Ensure that nan metric values are quoted as nan isn't a valid number in json.
--- tools/perf/tests/expr.c | 3 ++- tools/perf/tests/parse-metric.c | 1 + tools/perf/util/expr.y | 6 +++++- tools/perf/util/stat-display.c | 2 +- tools/perf/util/stat-shadow.c | 25 +++++++++++++++++++------ 5 files changed, 28 insertions(+), 9 deletions(-)
tools/perf/tests/expr.c +2 -1
@@ -120,7 +120,8 @@ static int test__expr(struct test_suite *t __maybe_unused, int subtest __maybe_u p ="FOO/0" ; ret =expr__parse (&val, ctx, p); -TEST_ASSERT_VAL ("division by zero" , ret ==-1 ); +TEST_ASSERT_VAL ("division by zero" , ret ==0 ); +TEST_ASSERT_VAL ("division by zero" ,isnan (val)); p ="BAR/" ; ret =expr__parse (&val, ctx, p);
tools/perf/tests/parse-metric.c +1 -0
@@ -38,6 +38,7 @@ static void load_runtime_stat(struct evlist *evlist, struct value *vals)evlist__alloc_aggr_stats (evlist,1 );evlist__for_each_entry (evlist, evsel) { count =find_value (evsel->name, vals); + evsel->supported = true; evsel->stats->aggr->counts.val = count;if (evsel__name_is (evsel,"duration_time" ))update_stats (&walltime_nsecs_stats, count);
tools/perf/util/expr.y +5 -1
@@ -225,7 +225,11 @@ expr: NUMBER { if (fpclassify($3.val) == FP_ZERO) { pr_debug("division by zero\n"); - YYABORT; + assert($3.ids == NULL); + if (compute_ids) + ids__free($1.ids); + $$.val = NAN; + $$.ids = NULL; } else if (!compute_ids || (is_const($1.val) && is_const($3.val))) { assert($1.ids == NULL); assert($3.ids == NULL);
tools/perf/util/stat-display.c +1 -1
@@ -431,7 +431,7 @@ static void print_metric_json(struct perf_stat_config *config __maybe_unused,struct outstate *os = ctx;FILE *out = os->fh; -fprintf (out,"\"metric-value\" : %f, " , val); +fprintf (out,"\"metric-value\" : \"%f\", " , val);fprintf (out,"\"metric-unit\" : \"%s\"" , unit);if (!config->metric_only)fprintf (out,"}" );
tools/perf/util/stat-shadow.c +19 -6
@@ -403,12 +403,25 @@ static int prepare_metric(struct evsel **metric_events,if (!aggr)break ; -/* -* If an event was scaled during stat gathering, reverse -* the scale before computing the metric. -*/ - val = aggr->counts.val * (1.0 / metric_events[i]->scale); - source_count =evsel__source_count (metric_events[i]); +if (!metric_events[i]->supported) { +/* +* Not supported events will have a count of 0, +* which can be confusing in a +* metric. Explicitly set the value to NAN. Not +* counted events (enable time of 0) are read as +* 0. +*/ + val =NAN ; + source_count =0 ; + }else { +/* +* If an event was scaled during stat gathering, +* reverse the scale before computing the +* metric. +*/ + val = aggr->counts.val * (1.0 / metric_events[i]->scale); + source_count =evsel__source_count (metric_events[i]); + } } n =strdup (evsel__metric_id (metric_events[i]));if (!n)
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 3/46 perf stat: Introduce skippable evsels
In reply to Ian RogersPerf stat with no arguments will use default events and metrics. These events may fail to open even with kernel and hypervisor disabled. When these fail then the permissions error appears even though they were implicitly selected. This is particularly a problem with the automatic selection of the TopdownL1 metric group on certain architectures like Skylake:
'''
$ perf stat true Error: Access to performance monitoring and observability operations is limited. Consider adjusting /proc/sys/kernel/perf_event_paranoid setting to open access to performance monitoring and observability operations for processes without CAP_PERFMON, CAP_SYS_PTRACE or CAP_SYS_ADMIN Linux capability. More information can be found at 'Perf events and tool security' document: https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html perf_event_paranoid setting is 2: -1: Allow use of (almost) all events by all users Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK
Quoted 3 lines
= 0: Disallow raw and ftrace function tracepoint access = 1: Disallow CPU event access
= 2: Disallow kernel profiling
To make the adjusted perf_event_paranoid setting permanent preserve it in /etc/sysctl.conf (e.g. kernel.perf_event_paranoid = <setting>) '''
This patch adds skippable evsels that when they fail to open won't cause termination and will appear as "<not supported>" in output. The TopdownL1 events, from the metric group, are marked as skippable. This turns the failure above to:
'''
$ perf stat perf bench internals synthesize Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 49.287 usec (+- 0.083 usec) Average num. events: 3.000 (+- 0.000) Average time per event 16.429 usec Average data synthesis took: 49.641 usec (+- 0.085 usec) Average num. events: 11.000 (+- 0.000) Average time per event 4.513 usec
Performance counter stats for 'perf bench internals synthesize':
1,222.38 msec task-clock:u # 0.993 CPUs utilized
0 context-switches:u # 0.000 /sec
0 cpu-migrations:u # 0.000 /sec
162 page-faults:u # 132.529 /sec
774,445,184 cycles:u # 0.634 GHz (49.61%)
1,640,969,811 instructions:u # 2.12 insn per cycle (59.67%)
302,052,148 branches:u # 247.102 M/sec (59.69%)
1,807,718 branch-misses:u # 0.60% of all branches (59.68%)
5,218,927 CPU_CLK_UNHALTED.REF_XCLK:u # 4.269 M/sec
# 17.3 % tma_frontend_bound
# 56.4 % tma_retiring
# nan % tma_backend_bound
# nan % tma_bad_speculation (60.01%)
536,580,469 IDQ_UOPS_NOT_DELIVERED.CORE:u # 438.965 M/sec (60.33%)
<not supported> INT_MISC.RECOVERY_CYCLES_ANY:u
5,223,936 CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE:u # 4.274 M/sec (40.31%)
774,127,250 CPU_CLK_UNHALTED.THREAD:u # 633.297 M/sec (50.34%)
1,746,579,518 UOPS_RETIRED.RETIRE_SLOTS:u # 1.429 G/sec (50.12%)
1,940,625,702 UOPS_ISSUED.ANY:u # 1.588 G/sec (49.70%) 1.231055525 seconds time elapsed
0.258327000 seconds user
0.965749000 seconds sys'''
The event INT_MISC.RECOVERY_CYCLES_ANY:u is skipped as it can't be opened with paranoia 2 on Skylake. With a lower paranoia, or as root, all events/metrics are computed.
--- tools/perf/builtin-stat.c | 38 +++++++++++++++++++++++++++++--------- tools/perf/util/evsel.c | 15 +++++++++++++-- tools/perf/util/evsel.h | 1 + 3 files changed, 43 insertions(+), 11 deletions(-)
tools/perf/builtin-stat.c +29 -9
@@ -667,6 +667,13 @@ static enum counter_recovery stat_handle_error(struct evsel *counter) evsel_list->core.threads->err_thread =-1 ;return COUNTER_RETRY ; } + }else if (counter->skippable) { +if (verbose >0 ) +ui__warning ("skipping event %s that kernel failed to open .\n" , +evsel__name (counter)); + counter->supported = false; + counter->errored = true; +return COUNTER_SKIP ; }evsel__open_strerror (counter, &target, errno, msg,sizeof (msg));
@@ -1890,15 +1897,28 @@ static int add_default_attributes(void)* caused by exposing latent bugs. This is fixed properly in: * https://lore.kernel.org/lkml/[email protected]/ */ -if (metricgroup__has_metric ("TopdownL1" ) && !perf_pmu__has_hybrid () && -metricgroup__parse_groups (evsel_list,"TopdownL1" , -/*metric_no_group=*/ false, -/*metric_no_merge=*/ false, -/*metric_no_threshold=*/ true, - stat_config.user_requested_cpu_list, - stat_config.system_wide, - &stat_config.metric_events) <0 ) -return -1 ; +if (metricgroup__has_metric ("TopdownL1" ) && !perf_pmu__has_hybrid ()) { +struct evlist *metric_evlist =evlist__new (); +struct evsel *metric_evsel; + +if (!metric_evlist) +return -1 ; + +if (metricgroup__parse_groups (metric_evlist,"TopdownL1" , +/*metric_no_group=*/ false, +/*metric_no_merge=*/ false, +/*metric_no_threshold=*/ true, + stat_config.user_requested_cpu_list, + stat_config.system_wide, + &stat_config.metric_events) <0 ) +return -1 ; + +evlist__for_each_entry (metric_evlist, metric_evsel) { + metric_evsel->skippable = true; + } +evlist__splice_list_tail (evsel_list, &metric_evlist->core.entries); +evlist__delete (metric_evlist); + }/* Platform specific attrs */ if (evlist__add_default_attrs (evsel_list, default_null_attrs) <0 )
tools/perf/util/evsel.c +13 -2
@@ -290,6 +290,7 @@ void evsel__init(struct evsel *evsel, evsel->per_pkg_mask =NULL ; evsel->collect_stat = false; evsel->pmu_name =NULL ; + evsel->skippable = false; }struct evsel *evsel__new_idx (struct perf_event_attr *attr,int idx)
@@ -1725,9 +1726,13 @@ static int get_group_fd(struct evsel *evsel, int cpu_map_idx, int thread)return -1 ; fd =FD (leader, cpu_map_idx, thread); -BUG_ON (fd ==-1 ); +BUG_ON (fd ==-1 && !leader->skippable); -return fd; +/* +* When the leader has been skipped, return -2 to distinguish from no +* group leader case. +*/ +return fd ==-1 ?-2 : fd; }static void evsel__remove_fd (struct evsel *pos,int nr_cpus,int nr_threads,int thread_idx)
@@ -2109,6 +2114,12 @@ static int evsel__open_cpu(struct evsel *evsel, struct perf_cpu_map *cpus, group_fd =get_group_fd (evsel, idx, thread); +if (group_fd ==-2 ) { +pr_debug ("broken group leader for %s\n" , evsel->name); + err = -EINVAL ; + goto out_close; + } +test_attr__ready ();/* Debug message used by test scripts */
tools/perf/util/evsel.h +1 -0
@@ -95,6 +95,7 @@ struct evsel {bool weak_group;bool bpf_counter;bool use_config_name; +bool skippable;int bpf_fd;struct bpf_object *bpf_obj;struct list_head config_terms;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Re: [PATCH v3 03/46] perf stat: Introduce skippable evsels
Quoted 64 lines · Ian Rogers
Perf stat with no arguments will use default events and metrics. These events may fail to open even with kernel and hypervisor disabled. When these fail then the permissions error appears even though they were implicitly selected. This is particularly a problem with the automatic selection of the TopdownL1 metric group on certain architectures like Skylake:
'''
$ perf stat true Error: Access to performance monitoring and observability operations is limited. Consider adjusting /proc/sys/kernel/perf_event_paranoid setting to open access to performance monitoring and observability operations for processes without CAP_PERFMON, CAP_SYS_PTRACE or CAP_SYS_ADMIN Linux capability. More information can be found at 'Perf events and tool security' document: https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html perf_event_paranoid setting is 2: -1: Allow use of (almost) all events by all users Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCKQuoted 3 lines
= 0: Disallow raw and ftrace function tracepoint access = 1: Disallow CPU event access
= 2: Disallow kernel profilingTo make the adjusted
perf_event_paranoidsetting permanent preserve it in/etc/sysctl.conf(e.g. kernel.perf_event_paranoid = <setting>) '''This patch adds skippable evsels that when they fail to open won't cause termination and will appear as "<not supported>" in output. The TopdownL1 events, from the metric group, are marked as skippable. This turns the failure above to:
'''
$ perf stat perf bench internals synthesize Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 49.287 usec (+- 0.083 usec) Average num. events: 3.000 (+- 0.000) Average time per event 16.429 usec Average data synthesis took: 49.641 usec (+- 0.085 usec) Average num. events: 11.000 (+- 0.000) Average time per event 4.513 usecPerformance counter stats for 'perf bench internals synthesize':
1,222.38 msec task-clock:u # 0.993 CPUs utilized 0 context-switches:u # 0.000 /sec 0 cpu-migrations:u # 0.000 /sec 162 page-faults:u # 132.529 /sec 774,445,184 cycles:u # 0.634 GHz (49.61%) 1,640,969,811 instructions:u # 2.12 insn per cycle (59.67%) 302,052,148 branches:u # 247.102 M/sec (59.69%) 1,807,718 branch-misses:u # 0.60% of all branches (59.68%) 5,218,927 CPU_CLK_UNHALTED.REF_XCLK:u # 4.269 M/sec # 17.3 % tma_frontend_bound # 56.4 % tma_retiring # nan % tma_backend_bound # nan % tma_bad_speculation (60.01%) 536,580,469 IDQ_UOPS_NOT_DELIVERED.CORE:u # 438.965 M/sec (60.33%) <not supported> INT_MISC.RECOVERY_CYCLES_ANY:u 5,223,936 CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE:u # 4.274 M/sec (40.31%) 774,127,250 CPU_CLK_UNHALTED.THREAD:u # 633.297 M/sec (50.34%) 1,746,579,518 UOPS_RETIRED.RETIRE_SLOTS:u # 1.429 G/sec (50.12%) 1,940,625,702 UOPS_ISSUED.ANY:u # 1.588 G/sec (49.70%)
Quoted 4 lines
1.231055525 seconds time elapsed
0.258327000 seconds user
0.965749000 seconds sysWhich branch is this patch series based on?
I still cannot get the same output as the examples.
I'm using the latest perf-tools-next (The latest commit ID is 5d27a645f609 ("perf tracepoint: Fix memory leak in is_valid_tracepoint()")). I only applied patch 2 and patch 3, since the patch 1 is already merged.
It's a single socket Cascade Lake. with kernel 5.19-8.
$ uname -r 5.19.8-100.fc35.x86_64
As you can see, all the topdown related events are displayed twice.
With root permission,
$ sudo ./perf stat perf bench internals synthesize # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 91.487 usec (+- 0.050 usec) Average num. events: 47.000 (+- 0.000) Average time per event 1.947 usec Average data synthesis took: 97.720 usec (+- 0.059 usec) Average num. events: 245.000 (+- 0.000) Average time per event 0.399 usec
Performance counter stats for 'perf bench internals synthesize':
2,077.81 msec task-clock # 0.998 CPUs
utilized
466 context-switches # 224.274 /sec
4 cpu-migrations # 1.925 /sec
775 page-faults # 372.988 /sec
9,561,957,326 cycles # 4.602 GHz
(31.17%)
24,466,854,021 instructions # 2.56 insn
per cycle (37.42%)
5,547,892,196 branches # 2.670
G/sec (37.48%)
37,880,526 branch-misses # 0.68% of
all branches (37.52%)
49,576,109 CPU_CLK_UNHALTED.REF_XCLK # 23.860 M/sec
# 59.9 % tma_retiring
# 4.6 %
tma_bad_speculation (37.47%)
228,406,003 INT_MISC.RECOVERY_CYCLES_ANY # 109.926
M/sec (37.52%)
49,591,815 CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE # 23.867
M/sec (24.99%)
9,553,472,893 CPU_CLK_UNHALTED.THREAD # 4.598
G/sec (31.25%)
22,893,372,651 UOPS_RETIRED.RETIRE_SLOTS # 11.018
G/sec (31.23%)
24,180,375,299 UOPS_ISSUED.ANY # 11.637
G/sec (31.25%)
49,562,300 CPU_CLK_UNHALTED.REF_XCLK # 23.853 M/sec
# 28.1 %
tma_frontend_bound
# 7.2 %
tma_backend_bound (31.24%)
10,735,205,084 IDQ_UOPS_NOT_DELIVERED.CORE # 5.167
G/sec (31.30%)
228,798,426 INT_MISC.RECOVERY_CYCLES_ANY # 110.115
M/sec (25.04%)
49,559,962 CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE # 23.852
M/sec (25.00%)
9,538,354,333 CPU_CLK_UNHALTED.THREAD # 4.591
G/sec (31.29%)
24,207,967,071 UOPS_ISSUED.ANY # 11.651G/sec (31.24%)
2.082670856 seconds time elapsed
0.812763000 seconds user
1.252387000 seconds sysWith non-root, nothing is counted for the topdownL1 events.
$ ./perf stat perf bench internals synthesize # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 91.852 usec (+- 0.139 usec) Average num. events: 47.000 (+- 0.000) Average time per event 1.954 usec Average data synthesis took: 96.230 usec (+- 0.046 usec) Average num. events: 245.000 (+- 0.000) Average time per event 0.393 usec
Performance counter stats for 'perf bench internals synthesize':
2,051.95 msec task-clock:u # 0.997 CPUs
utilized
0 context-switches:u # 0.000 /sec
0 cpu-migrations:u # 0.000 /sec
765 page-faults:u # 372.816 /sec
3,601,662,523 cycles:u # 1.755 GHz
(16.72%)
9,241,811,003 instructions:u # 2.57 insn
per cycle (33.43%)
2,238,848,485 branches:u # 1.091
G/sec (50.06%)
19,966,181 branch-misses:u # 0.89% of
all branches (66.77%)
<not counted> CPU_CLK_UNHALTED.REF_XCLK:u
<not supported> INT_MISC.RECOVERY_CYCLES_ANY:u
<not counted> CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE:u
<not counted> CPU_CLK_UNHALTED.THREAD:u
<not counted> UOPS_RETIRED.RETIRE_SLOTS:u
<not counted> UOPS_ISSUED.ANY:u
<not counted> CPU_CLK_UNHALTED.REF_XCLK:u
<not counted> IDQ_UOPS_NOT_DELIVERED.CORE:u
<not supported> INT_MISC.RECOVERY_CYCLES_ANY:u
<not counted> CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE:u
<not counted> CPU_CLK_UNHALTED.THREAD:u
<not counted> UOPS_ISSUED.ANY:u 2.057691297 seconds time elapsed
0.766640000 seconds user
1.275170000 seconds sysThanks,
Kan
Ian Rogers
reviewerRe: [PATCH v3 03/46] perf stat: Introduce skippable evsels
Quoted 144 lines · Kan
Quoted 64 lines · Ian Rogers
Perf stat with no arguments will use default events and metrics. These events may fail to open even with kernel and hypervisor disabled. When these fail then the permissions error appears even though they were implicitly selected. This is particularly a problem with the automatic selection of the TopdownL1 metric group on certain architectures like Skylake:
'''
$ perf stat true Error: Access to performance monitoring and observability operations is limited. Consider adjusting /proc/sys/kernel/perf_event_paranoid setting to open access to performance monitoring and observability operations for processes without CAP_PERFMON, CAP_SYS_PTRACE or CAP_SYS_ADMIN Linux capability. More information can be found at 'Perf events and tool security' document: https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html perf_event_paranoid setting is 2: -1: Allow use of (almost) all events by all users Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCKQuoted 3 lines
= 0: Disallow raw and ftrace function tracepoint access = 1: Disallow CPU event access
= 2: Disallow kernel profilingTo make the adjusted
perf_event_paranoidsetting permanent preserve it in/etc/sysctl.conf(e.g. kernel.perf_event_paranoid = <setting>) '''This patch adds skippable evsels that when they fail to open won't cause termination and will appear as "<not supported>" in output. The TopdownL1 events, from the metric group, are marked as skippable. This turns the failure above to:
'''
$ perf stat perf bench internals synthesize Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 49.287 usec (+- 0.083 usec) Average num. events: 3.000 (+- 0.000) Average time per event 16.429 usec Average data synthesis took: 49.641 usec (+- 0.085 usec) Average num. events: 11.000 (+- 0.000) Average time per event 4.513 usecPerformance counter stats for 'perf bench internals synthesize':
1,222.38 msec task-clock:u # 0.993 CPUs utilized 0 context-switches:u # 0.000 /sec 0 cpu-migrations:u # 0.000 /sec 162 page-faults:u # 132.529 /sec 774,445,184 cycles:u # 0.634 GHz (49.61%) 1,640,969,811 instructions:u # 2.12 insn per cycle (59.67%) 302,052,148 branches:u # 247.102 M/sec (59.69%) 1,807,718 branch-misses:u # 0.60% of all branches (59.68%) 5,218,927 CPU_CLK_UNHALTED.REF_XCLK:u # 4.269 M/sec # 17.3 % tma_frontend_bound # 56.4 % tma_retiring # nan % tma_backend_bound # nan % tma_bad_speculation (60.01%) 536,580,469 IDQ_UOPS_NOT_DELIVERED.CORE:u # 438.965 M/sec (60.33%) <not supported> INT_MISC.RECOVERY_CYCLES_ANY:u 5,223,936 CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE:u # 4.274 M/sec (40.31%) 774,127,250 CPU_CLK_UNHALTED.THREAD:u # 633.297 M/sec (50.34%) 1,746,579,518 UOPS_RETIRED.RETIRE_SLOTS:u # 1.429 G/sec (50.12%) 1,940,625,702 UOPS_ISSUED.ANY:u # 1.588 G/sec (49.70%)Quoted 4 lines
1.231055525 seconds time elapsed 0.258327000 seconds user 0.965749000 seconds sysWhich branch is this patch series based on?
I still cannot get the same output as the examples.
I'm using the latest perf-tools-next (The latest commit ID is
5d27a645f609("perf tracepoint: Fix memory leak inis_valid_tracepoint()")). I only applied patch 2 and patch 3, since the patch 1 is already merged.It's a single socket Cascade Lake. with kernel 5.19-8.
$ uname -r 5.19.8-100.fc35.x86_64As you can see, all the topdown related events are displayed twice.
With root permission,
$ sudo ./perf stat perf bench internals synthesize # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 91.487 usec (+- 0.050 usec) Average num. events: 47.000 (+- 0.000) Average time per event 1.947 usec Average data synthesis took: 97.720 usec (+- 0.059 usec) Average num. events: 245.000 (+- 0.000) Average time per event 0.399 usecPerformance counter stats for 'perf bench internals synthesize':
2,077.81 msec task-clock # 0.998 CPUs
utilized466 context-switches # 224.274 /sec 4 cpu-migrations # 1.925 /sec 775 page-faults # 372.988 /sec 9,561,957,326 cycles # 4.602 GHz (31.17%) 24,466,854,021 instructions # 2.56 insn per cycle (37.42%) 5,547,892,196 branches # 2.670 G/sec (37.48%) 37,880,526 branch-misses # 0.68% of all branches (37.52%) 49,576,109 CPU_CLK_UNHALTED.REF_XCLK # 23.860 M/sec # 59.9 % tma_retiring # 4.6 % tma_bad_speculation (37.47%) 228,406,003 INT_MISC.RECOVERY_CYCLES_ANY # 109.926 M/sec (37.52%) 49,591,815 CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE # 23.867 M/sec (24.99%) 9,553,472,893 CPU_CLK_UNHALTED.THREAD # 4.598 G/sec (31.25%) 22,893,372,651 UOPS_RETIRED.RETIRE_SLOTS # 11.018 G/sec (31.23%) 24,180,375,299 UOPS_ISSUED.ANY # 11.637 G/sec (31.25%) 49,562,300 CPU_CLK_UNHALTED.REF_XCLK # 23.853 M/sec # 28.1 % tma_frontend_bound # 7.2 % tma_backend_bound (31.24%) 10,735,205,084 IDQ_UOPS_NOT_DELIVERED.CORE # 5.167 G/sec (31.30%) 228,798,426 INT_MISC.RECOVERY_CYCLES_ANY # 110.115 M/sec (25.04%) 49,559,962 CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE # 23.852 M/sec (25.00%) 9,538,354,333 CPU_CLK_UNHALTED.THREAD # 4.591 G/sec (31.29%) 24,207,967,071 UOPS_ISSUED.ANY # 11.651
Quoted 6 lines
G/sec (31.24%)
2.082670856 seconds time elapsed 0.812763000 seconds user 1.252387000 seconds sys
The events are displayed twice as there are 2 groups of events. This is changed by:
https://lore.kernel.org/lkml/[email protected]/
where the events are no longer grouped.
Quoted 41 lines
With non-root, nothing is counted for the topdownL1 events.
$ ./perf stat perf bench internals synthesize # Running 'internals/synthesize' benchmark: Computing performance of single threaded perf event synthesis by synthesizing events on the perf process itself: Average synthesis took: 91.852 usec (+- 0.139 usec) Average num. events: 47.000 (+- 0.000) Average time per event 1.954 usec Average data synthesis took: 96.230 usec (+- 0.046 usec) Average num. events: 245.000 (+- 0.000) Average time per event 0.393 usecPerformance counter stats for 'perf bench internals synthesize':
2,051.95 msec task-clock:u # 0.997 CPUs utilized 0 context-switches:u # 0.000 /sec 0 cpu-migrations:u # 0.000 /sec 765 page-faults:u # 372.816 /sec 3,601,662,523 cycles:u # 1.755 GHz (16.72%) 9,241,811,003 instructions:u # 2.57 insn per cycle (33.43%) 2,238,848,485 branches:u # 1.091 G/sec (50.06%) 19,966,181 branch-misses:u # 0.89% of all branches (66.77%) <not counted> CPU_CLK_UNHALTED.REF_XCLK:u <not supported> INT_MISC.RECOVERY_CYCLES_ANY:u <not counted> CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE:u <not counted> CPU_CLK_UNHALTED.THREAD:u <not counted> UOPS_RETIRED.RETIRE_SLOTS:u <not counted> UOPS_ISSUED.ANY:u <not counted> CPU_CLK_UNHALTED.REF_XCLK:u <not counted> IDQ_UOPS_NOT_DELIVERED.CORE:u <not supported> INT_MISC.RECOVERY_CYCLES_ANY:u <not counted> CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE:u <not counted> CPU_CLK_UNHALTED.THREAD:u <not counted> UOPS_ISSUED.ANY:u
Quoted 4 lines
2.057691297 seconds time elapsed
0.766640000 seconds user
1.275170000 seconds sysThe reason nothing is counted is that all the metrics are trying to share groups with the event INT_MISC.RECOVERY_CYCLES_ANY in them, which means the whole group ends up being not supported. Again, the patch above that removes the groups for TopdownL1 and TopdownL2, as requested by Weilin, will address the issue.
Thanks,
Ian
Quoted 3 lines
Thanks,
Kan
Re: [PATCH v3 03/46] perf stat: Introduce skippable evsels
Quoted 4 lines · Ian Rogers
The events are displayed twice as there are 2 groups of events. This is changed by:
https://lore.kernel.org/lkml/[email protected]/
where the events are no longer grouped.
The trick seems don't work on a hybrid machine. I still got the duplicate Topdown events on e-core.
38,841.16 msec cpu-clock # 32.009 CPUs
utilized
256 context-switches # 6.591 /sec
33 cpu-migrations # 0.850 /sec
84 page-faults # 2.163 /sec
21,910,584 cpu_core/cycles/ # 564.107 K/sec
248,153,249 cpu_atom/cycles/ # 6.389
M/sec (53.85%)
27,463,908 cpu_core/instructions/ # 707.083 K/sec
118,661,014 cpu_atom/instructions/ # 3.055
M/sec (63.06%)
4,652,941 cpu_core/branches/ # 119.794 K/sec
20,173,082 cpu_atom/branches/ # 519.374
K/sec (63.18%)
72,727 cpu_core/branch-misses/ # 1.872 K/sec
1,143,187 cpu_atom/branch-misses/ # 29.432
K/sec (63.51%)
125,630,586 cpu_core/TOPDOWN.SLOTS/ # nan %tma_backend_bound
# nan % tma_retiring
# 0.0 %tma_bad_speculation
# nan %tma_frontend_bound
30,254,701 cpu_core/topdown-retiring/
149,075,726 cpu_atom/TOPDOWN_RETIRING.ALL/ # 3.838 M/sec
# 14.8 %
tma_bad_speculation (63.82%)
<not supported> cpu_core/topdown-bad-spec/
523,614,383 cpu_atom/TOPDOWN_FE_BOUND.ALL/ # 13.481 M/sec
# 42.0 %
tma_frontend_bound (64.15%)
385,502,477 cpu_atom/TOPDOWN_BE_BOUND.ALL/ # 9.925 M/sec
# 30.9 %
tma_backend_bound
# 30.9 %
tma_backend_bound_aux (64.39%)
249,534,488 cpu_atom/CPU_CLK_UNHALTED.CORE/ # 6.424 M/sec
# 12.2 %
tma_retiring (64.18%)
151,729,465 cpu_atom/TOPDOWN_RETIRING.ALL/ # 3.906
M/sec (54.67%)
530,621,769 cpu_atom/TOPDOWN_FE_BOUND.ALL/ # 13.661
M/sec (54.30%)
<not supported> cpu_core/topdown-fe-bound/
383,694,745 cpu_atom/TOPDOWN_BE_BOUND.ALL/ # 9.879
M/sec (53.96%)
<not supported> cpu_core/topdown-be-bound/
105,850 cpu_core/INT_MISC.UOP_DROPPING/ # 2.725 K/sec
1.213449538 seconds time elapsedThanks,
Kan
Ian Rogers
reviewerRe: [PATCH v3 03/46] perf stat: Introduce skippable evsels
Quoted 11 lines · Kan
Quoted 4 lines · Ian Rogers
The events are displayed twice as there are 2 groups of events. This is changed by:
https://lore.kernel.org/lkml/[email protected]/
where the events are no longer grouped.The trick seems don't work on a hybrid machine. I still got the duplicate Topdown events on e-core.
For hybrid the rest of the patch series is necessary, ie the patches beyond what's for 6.4, which I take from the output (ie not a crash) you are looking at. As multiple groups are in play then it looks like the atom events are on >1 PMU which can happen as the x86 code special cases events with topdown in their name. Some fixes in the series for this are:
https://lore.kernel.org/lkml/[email protected]/
https://lore.kernel.org/lkml/[email protected]/
and related:
https://lore.kernel.org/lkml/[email protected]/
and so fixing this requires some detective work.
I don't think it should be a requirement for the series that all hybrid bugs are fixed - especially given the complaints against the length of the series as-is.
Thanks,
Ian
Quoted 54 lines
38,841.16 msec cpu-clock # 32.009 CPUs utilized 256 context-switches # 6.591 /sec 33 cpu-migrations # 0.850 /sec 84 page-faults # 2.163 /sec 21,910,584 cpu_core/cycles/ # 564.107 K/sec 248,153,249 cpu_atom/cycles/ # 6.389 M/sec (53.85%) 27,463,908 cpu_core/instructions/ # 707.083 K/sec 118,661,014 cpu_atom/instructions/ # 3.055 M/sec (63.06%) 4,652,941 cpu_core/branches/ # 119.794 K/sec 20,173,082 cpu_atom/branches/ # 519.374 K/sec (63.18%) 72,727 cpu_core/branch-misses/ # 1.872 K/sec 1,143,187 cpu_atom/branch-misses/ # 29.432 K/sec (63.51%) 125,630,586 cpu_core/TOPDOWN.SLOTS/ # nan %
tma_backend_bound
# nan %tma_retiring
# 0.0 %tma_bad_speculation
# nan %tma_frontend_bound30,254,701 cpu_core/topdown-retiring/ 149,075,726 cpu_atom/TOPDOWN_RETIRING.ALL/ # 3.838 M/sec # 14.8 % tma_bad_speculation (63.82%) <not supported> cpu_core/topdown-bad-spec/ 523,614,383 cpu_atom/TOPDOWN_FE_BOUND.ALL/ # 13.481 M/sec # 42.0 % tma_frontend_bound (64.15%) 385,502,477 cpu_atom/TOPDOWN_BE_BOUND.ALL/ # 9.925 M/sec # 30.9 % tma_backend_bound # 30.9 % tma_backend_bound_aux (64.39%) 249,534,488 cpu_atom/CPU_CLK_UNHALTED.CORE/ # 6.424 M/sec # 12.2 % tma_retiring (64.18%) 151,729,465 cpu_atom/TOPDOWN_RETIRING.ALL/ # 3.906 M/sec (54.67%) 530,621,769 cpu_atom/TOPDOWN_FE_BOUND.ALL/ # 13.661 M/sec (54.30%) <not supported> cpu_core/topdown-fe-bound/ 383,694,745 cpu_atom/TOPDOWN_BE_BOUND.ALL/ # 9.879 M/sec (53.96%) <not supported> cpu_core/topdown-be-bound/ 105,850 cpu_core/INT_MISC.UOP_DROPPING/ # 2.725 K/sec 1.213449538 seconds time elapsedThanks,
Kan
Re: [PATCH v3 03/46] perf stat: Introduce skippable evsels
Quoted 10 lines · Ian Rogers
Quoted 9 lines · Kan
Quoted 4 lines · Ian Rogers
The events are displayed twice as there are 2 groups of events. This is changed by:
https://lore.kernel.org/lkml/[email protected]/
where the events are no longer grouped.The trick seems don't work on a hybrid machine. I still got the duplicate Topdown events on e-core.
Quoted 12 lines
For hybrid the rest of the patch series is necessary, ie the patches beyond what's for 6.4, which I take from the output (ie not a crash) you are looking at. As multiple groups are in play then it looks like the atom events are on >1 PMU which can happen as the x86 code special cases events with topdown in their name. Some fixes in the series for this are:
https://lore.kernel.org/lkml/[email protected]/
https://lore.kernel.org/lkml/[email protected]/
and related:
https://lore.kernel.org/lkml/[email protected]/
and so fixing this requires some detective work.
I applied all the patches of the series when I did the test on hybrid. The above patches don't help.
Quoted 3 lines
I don't think it should be a requirement for the series that all hybrid bugs are fixed - especially given the complaints against the length of the series as-is.
I agree especially considering the metrics have been broken on the hybrid for a while.
I just want to do a complete test and understand what is going to fixed and what hasn't.
Thanks,
Kan
Ian Rogers
reviewerPATCH v3 4/46 perf metric: Json flag to not group events if gathering a metric group
In reply to Ian RogersSome metric groups have metrics that don't have fully overlapping events, meaning that the group's events become unique event groups that may need to multiplex with each other. This can be particularly unfortunate when the groups wouldn't need to multiplex because there are sufficient hardware counters.
Add a flag so that if recording a metric group then the metrics within the group needn't use groups for their events. The flag is added to Intel TopdownL1 and TopdownL2 metrics.
--- .../arch/x86/alderlake/adl-metrics.json | 26 +++++++++++++++++++ .../arch/x86/alderlaken/adln-metrics.json | 14 ++++++++++ .../arch/x86/broadwell/bdw-metrics.json | 12 +++++++++ .../arch/x86/broadwellde/bdwde-metrics.json | 12 +++++++++ .../arch/x86/broadwellx/bdx-metrics.json | 12 +++++++++ .../arch/x86/cascadelakex/clx-metrics.json | 12 +++++++++ .../arch/x86/haswell/hsw-metrics.json | 12 +++++++++ .../arch/x86/haswellx/hsx-metrics.json | 12 +++++++++ .../arch/x86/icelake/icl-metrics.json | 12 +++++++++ .../arch/x86/icelakex/icx-metrics.json | 12 +++++++++ .../arch/x86/ivybridge/ivb-metrics.json | 12 +++++++++ .../arch/x86/ivytown/ivt-metrics.json | 12 +++++++++ .../arch/x86/jaketown/jkt-metrics.json | 12 +++++++++ .../arch/x86/sandybridge/snb-metrics.json | 12 +++++++++ .../arch/x86/sapphirerapids/spr-metrics.json | 12 +++++++++ .../arch/x86/skylake/skl-metrics.json | 12 +++++++++ .../arch/x86/skylakex/skx-metrics.json | 12 +++++++++ .../arch/x86/tigerlake/tgl-metrics.json | 12 +++++++++ tools/perf/pmu-events/jevents.py | 4 ++- tools/perf/pmu-events/pmu-events.h | 1 + tools/perf/util/metricgroup.c | 8 +++--- 21 files changed, 240 insertions(+), 5 deletions(-)
tools/perf/pmu-events/arch/x86/alderlake/adl-metrics.json +26 -0
@@ -133,6 +133,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "Counts the total number of issue slots that were not consumed by the backend due to backend stalls. Note that uops must be available for consumption in order for this event to count. If a uop is not available (IQ is empty), this event will not count. The rest of these subevents count backend stalls, in cycles, due to an outstanding request which is memory bound vs core bound. The subevents are not slot based events and therefore can not be precisely added or subtracted from the Backend_Bound_Aux subevents which are slot based.", "ScaleUnit": "100%", "Unit": "cpu_atom"
@@ -143,6 +144,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound_aux", "MetricThreshold": "tma_backend_bound_aux > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "Counts the total number of issue slots that were not consumed by the backend due to backend stalls. Note that UOPS must be available for consumption in order for this event to count. If a uop is not available (IQ is empty), this event will not count. All of these subevents count backend stalls, in slots, due to a resource limitation. These are not cycle based events and therefore can not be precisely added or subtracted from the Backend_Bound subevents which are cycle based. These subevents are supplementary to Backend_Bound and can be used to analyze results from a resource perspective at allocation.", "ScaleUnit": "100%", "Unit": "cpu_atom"
@@ -153,6 +155,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "Counts the total number of issue slots that were not consumed by the backend because allocation is stalled due to a mispredicted jump or a machine clear. Only issue slots wasted due to fast nukes such as memory ordering nukes are counted. Other nukes are not accounted for. Counts all issue slots blocked during this recovery window including relevant microcode flows and while uops are not yet available in the instruction queue (IQ). Also includes the issue slots that were consumed by the backend but were thrown away because they were younger than the mispredict or machine clear.", "ScaleUnit": "100%", "Unit": "cpu_atom"
@@ -163,6 +166,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_base", "MetricThreshold": "tma_base > 0.6", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%", "Unit": "cpu_atom" },
@@ -182,6 +186,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_bad_speculation_group", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.05", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%", "Unit": "cpu_atom" },
@@ -209,6 +214,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%", "Unit": "cpu_atom" },
@@ -255,6 +261,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%", "Unit": "cpu_atom" },
@@ -264,6 +271,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.15", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%", "Unit": "cpu_atom" },
@@ -291,6 +299,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "ScaleUnit": "100%", "Unit": "cpu_atom" },
@@ -593,6 +602,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_bad_speculation_group", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.05", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%", "Unit": "cpu_atom" },
@@ -611,6 +621,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%", "Unit": "cpu_atom" },
@@ -629,6 +640,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_ms_uops", "MetricThreshold": "tma_ms_uops > 0.05", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "Counts the number of uops that are from the complex flows issued by the micro-sequencer (MS). This includes uops from flows due to complex instructions, faults, assists, and inserted flows.", "ScaleUnit": "100%", "Unit": "cpu_atom"
@@ -729,6 +741,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_backend_bound_aux_group", "MetricName": "tma_resource_bound", "MetricThreshold": "tma_resource_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "Counts the total number of issue slots that were not consumed by the backend due to backend stalls. Note that uops must be available for consumption in order for this event to count. If a uop is not available (IQ is empty), this event will not count.", "ScaleUnit": "100%", "Unit": "cpu_atom"
@@ -739,6 +752,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.75", + "MetricgroupNoGroup": "TopdownL1", "ScaleUnit": "100%", "Unit": "cpu_atom" },
@@ -848,6 +862,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound. Sample with: TOPDOWN.BACKEND_BOUND_SLOTS", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -858,6 +873,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -868,6 +884,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: TOPDOWN.BR_MISPREDICT_SLOTS. Related metrics: tma_info_branch_misprediction_cost, tma_info_mispredictions, tma_mispredicts_resteers", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -919,6 +936,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -1031,6 +1049,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 6 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Sample with: FRONTEND_RETIRED.LATENCY_GE_2_BUBBLES_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_2_PS. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_dsb_misses, tma_info_iptb, tma_lcp", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -1041,6 +1060,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: FRONTEND_RETIRED.LATENCY_GE_16_PS;FRONTEND_RETIRED.LATENCY_GE_8_PS", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -1121,6 +1141,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound. Sample with: FRONTEND_RETIRED.LATENCY_GE_4_PS", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -1141,6 +1162,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences. Sample with: UOPS_RETIRED.HEAVY", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -2023,6 +2045,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -2082,6 +2105,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -2112,6 +2136,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -2310,6 +2335,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.SLOTS", "ScaleUnit": "100%", "Unit": "cpu_core"
tools/perf/pmu-events/arch/x86/alderlaken/adln-metrics.json +14 -0
@@ -98,6 +98,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "Counts the total number of issue slots that were not consumed by the backend due to backend stalls. Note that uops must be available for consumption in order for this event to count. If a uop is not available (IQ is empty), this event will not count. The rest of these subevents count backend stalls, in cycles, due to an outstanding request which is memory bound vs core bound. The subevents are not slot based events and therefore can not be precisely added or subtracted from the Backend_Bound_Aux subevents which are slot based.", "ScaleUnit": "100%" },
@@ -107,6 +108,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound_aux", "MetricThreshold": "tma_backend_bound_aux > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "Counts the total number of issue slots that were not consumed by the backend due to backend stalls. Note that UOPS must be available for consumption in order for this event to count. If a uop is not available (IQ is empty), this event will not count. All of these subevents count backend stalls, in slots, due to a resource limitation. These are not cycle based events and therefore can not be precisely added or subtracted from the Backend_Bound subevents which are cycle based. These subevents are supplementary to Backend_Bound and can be used to analyze results from a resource perspective at allocation.", "ScaleUnit": "100%" },
@@ -116,6 +118,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "Counts the total number of issue slots that were not consumed by the backend because allocation is stalled due to a mispredicted jump or a machine clear. Only issue slots wasted due to fast nukes such as memory ordering nukes are counted. Other nukes are not accounted for. Counts all issue slots blocked during this recovery window including relevant microcode flows and while uops are not yet available in the instruction queue (IQ). Also includes the issue slots that were consumed by the backend but were thrown away because they were younger than the mispredict or machine clear.", "ScaleUnit": "100%" },
@@ -125,6 +128,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_base", "MetricThreshold": "tma_base > 0.6", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%" }, {
@@ -142,6 +146,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_bad_speculation_group", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.05", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%" }, {
@@ -166,6 +171,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%" }, {
@@ -207,6 +213,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%" }, {
@@ -215,6 +222,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.15", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%" }, {
@@ -239,6 +247,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "ScaleUnit": "100%" }, {
@@ -499,6 +508,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_bad_speculation_group", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.05", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%" }, {
@@ -515,6 +525,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "ScaleUnit": "100%" }, {
@@ -531,6 +542,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_ms_uops", "MetricThreshold": "tma_ms_uops > 0.05", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "Counts the number of uops that are from the complex flows issued by the micro-sequencer (MS). This includes uops from flows due to complex instructions, faults, assists, and inserted flows.", "ScaleUnit": "100%" },
@@ -620,6 +632,7 @@ "MetricGroup": "TopdownL2;tma_L2_group;tma_backend_bound_aux_group", "MetricName": "tma_resource_bound", "MetricThreshold": "tma_resource_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "Counts the total number of issue slots that were not consumed by the backend due to backend stalls. Note that uops must be available for consumption in order for this event to count. If a uop is not available (IQ is empty), this event will not count.", "ScaleUnit": "100%" },
@@ -629,6 +642,7 @@ "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.75", + "MetricgroupNoGroup": "TopdownL1", "ScaleUnit": "100%" }, {
tools/perf/pmu-events/arch/x86/broadwell/bdw-metrics.json +12 -0
@@ -103,6 +103,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -112,6 +113,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -122,6 +124,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -170,6 +173,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -263,6 +267,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -272,6 +277,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: RS_EVENTS.EMPTY_END", "ScaleUnit": "100%" },
@@ -326,6 +332,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound.", "ScaleUnit": "100%" },
@@ -335,6 +342,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -828,6 +836,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -858,6 +867,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -886,6 +896,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1048,6 +1059,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/broadwellde/bdwde-metrics.json +12 -0
@@ -97,6 +97,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound. Sample with: TOPDOWN.BACKEND_BOUND_SLOTS", "ScaleUnit": "100%" },
@@ -106,6 +107,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -116,6 +118,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: TOPDOWN.BR_MISPREDICT_SLOTS. Related metrics: tma_info_branch_misprediction_cost, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -164,6 +167,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -248,6 +252,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Sample with: FRONTEND_RETIRED.LATENCY_GE_2_BUBBLES_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_2_PS. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -257,6 +262,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: FRONTEND_RETIRED.LATENCY_GE_16_PS;FRONTEND_RETIRED.LATENCY_GE_8_PS", "ScaleUnit": "100%" },
@@ -311,6 +317,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound. Sample with: FRONTEND_RETIRED.LATENCY_GE_4_PS", "ScaleUnit": "100%" },
@@ -320,6 +327,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -795,6 +803,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -825,6 +834,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -853,6 +863,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1013,6 +1024,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/broadwellx/bdx-metrics.json +12 -0
@@ -103,6 +103,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -112,6 +113,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -122,6 +124,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -170,6 +173,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -263,6 +267,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -272,6 +277,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: RS_EVENTS.EMPTY_END", "ScaleUnit": "100%" },
@@ -326,6 +332,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound.", "ScaleUnit": "100%" },
@@ -335,6 +342,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -829,6 +837,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -869,6 +878,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -897,6 +907,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1079,6 +1090,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/cascadelakex/clx-metrics.json +12 -0
@@ -101,6 +101,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -110,6 +111,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -120,6 +122,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_info_mispredictions, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -167,6 +170,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -271,6 +275,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Sample with: FRONTEND_RETIRED.LATENCY_GE_2_BUBBLES_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_2_PS. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_dsb_misses, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -280,6 +285,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: FRONTEND_RETIRED.LATENCY_GE_16_PS;FRONTEND_RETIRED.LATENCY_GE_8_PS", "ScaleUnit": "100%" },
@@ -354,6 +360,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound. Sample with: FRONTEND_RETIRED.LATENCY_GE_4_PS", "ScaleUnit": "100%" },
@@ -372,6 +379,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -1142,6 +1150,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -1196,6 +1205,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -1224,6 +1234,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1458,6 +1469,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/haswell/hsw-metrics.json +12 -0
@@ -103,6 +103,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -112,6 +113,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -122,6 +124,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -161,6 +164,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -254,6 +258,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -263,6 +268,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: RS_EVENTS.EMPTY_END", "ScaleUnit": "100%" },
@@ -272,6 +278,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound.", "ScaleUnit": "100%" },
@@ -281,6 +288,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -663,6 +671,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -693,6 +702,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -721,6 +731,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -874,6 +885,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/haswellx/hsx-metrics.json +12 -0
@@ -103,6 +103,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -112,6 +113,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -122,6 +124,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -161,6 +164,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -254,6 +258,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -263,6 +268,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: RS_EVENTS.EMPTY_END", "ScaleUnit": "100%" },
@@ -272,6 +278,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound.", "ScaleUnit": "100%" },
@@ -281,6 +288,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -664,6 +672,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -704,6 +713,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -732,6 +742,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -905,6 +916,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/icelake/icl-metrics.json +12 -0
@@ -115,6 +115,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound. Sample with: TOPDOWN.BACKEND_BOUND_SLOTS", "ScaleUnit": "100%" },
@@ -124,6 +125,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -141,6 +143,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_info_mispredictions, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -187,6 +190,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -288,6 +292,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 5 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Sample with: FRONTEND_RETIRED.LATENCY_GE_2_BUBBLES_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_2_PS. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_dsb_misses, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -297,6 +302,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: FRONTEND_RETIRED.LATENCY_GE_16_PS;FRONTEND_RETIRED.LATENCY_GE_8_PS", "ScaleUnit": "100%" },
@@ -369,6 +375,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound. Sample with: FRONTEND_RETIRED.LATENCY_GE_4_PS", "ScaleUnit": "100%" },
@@ -378,6 +385,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -1111,6 +1119,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -1164,6 +1173,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -1191,6 +1201,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1360,6 +1371,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/icelakex/icx-metrics.json +12 -0
@@ -80,6 +80,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound. Sample with: TOPDOWN.BACKEND_BOUND_SLOTS", "ScaleUnit": "100%" },
@@ -89,6 +90,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -106,6 +108,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_info_mispredictions, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -152,6 +155,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -253,6 +257,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 5 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Sample with: FRONTEND_RETIRED.LATENCY_GE_2_BUBBLES_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_2_PS. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_dsb_misses, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -262,6 +267,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: FRONTEND_RETIRED.LATENCY_GE_16_PS;FRONTEND_RETIRED.LATENCY_GE_8_PS", "ScaleUnit": "100%" },
@@ -334,6 +340,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound. Sample with: FRONTEND_RETIRED.LATENCY_GE_4_PS", "ScaleUnit": "100%" },
@@ -343,6 +350,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -1134,6 +1142,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -1187,6 +1196,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -1214,6 +1224,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1410,6 +1421,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/ivybridge/ivb-metrics.json +12 -0
@@ -103,6 +103,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -112,6 +113,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -122,6 +124,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -161,6 +164,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -254,6 +258,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -263,6 +268,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: RS_EVENTS.EMPTY_END", "ScaleUnit": "100%" },
@@ -299,6 +305,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound.", "ScaleUnit": "100%" },
@@ -308,6 +315,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -724,6 +732,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -754,6 +763,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -782,6 +792,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -917,6 +928,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/ivytown/ivt-metrics.json +12 -0
@@ -103,6 +103,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -112,6 +113,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -122,6 +124,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -161,6 +164,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -254,6 +258,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -263,6 +268,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: RS_EVENTS.EMPTY_END", "ScaleUnit": "100%" },
@@ -299,6 +305,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound.", "ScaleUnit": "100%" },
@@ -308,6 +315,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -725,6 +733,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -765,6 +774,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -793,6 +803,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -948,6 +959,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/jaketown/jkt-metrics.json +12 -0
@@ -76,6 +76,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -85,6 +86,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -95,6 +97,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -114,6 +117,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -160,6 +164,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_lcp", "ScaleUnit": "100%" },
@@ -169,6 +174,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: RS_EVENTS.EMPTY_END", "ScaleUnit": "100%" },
@@ -205,6 +211,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound.", "ScaleUnit": "100%" },
@@ -214,6 +221,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -412,6 +420,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -422,6 +431,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -450,6 +460,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -487,6 +498,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/sandybridge/snb-metrics.json +12 -0
@@ -76,6 +76,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -85,6 +86,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -95,6 +97,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -114,6 +117,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -160,6 +164,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_lcp", "ScaleUnit": "100%" },
@@ -169,6 +174,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: RS_EVENTS.EMPTY_END", "ScaleUnit": "100%" },
@@ -205,6 +211,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound.", "ScaleUnit": "100%" },
@@ -214,6 +221,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -411,6 +419,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -421,6 +430,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -449,6 +459,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -486,6 +497,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/sapphirerapids/spr-metrics.json +12 -0
@@ -87,6 +87,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound. Sample with: TOPDOWN.BACKEND_BOUND_SLOTS", "ScaleUnit": "100%" },
@@ -96,6 +97,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -105,6 +107,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: TOPDOWN.BR_MISPREDICT_SLOTS. Related metrics: tma_info_branch_misprediction_cost, tma_info_mispredictions, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -151,6 +154,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -252,6 +256,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 6 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Sample with: FRONTEND_RETIRED.LATENCY_GE_2_BUBBLES_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_2_PS. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_dsb_misses, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -261,6 +266,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: FRONTEND_RETIRED.LATENCY_GE_16_PS;FRONTEND_RETIRED.LATENCY_GE_8_PS", "ScaleUnit": "100%" },
@@ -351,6 +357,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound. Sample with: FRONTEND_RETIRED.LATENCY_GE_4_PS", "ScaleUnit": "100%" },
@@ -369,6 +376,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences. Sample with: UOPS_RETIRED.HEAVY", "ScaleUnit": "100%" },
@@ -1216,6 +1224,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -1269,6 +1278,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -1304,6 +1314,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1509,6 +1520,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/skylake/skl-metrics.json +12 -0
@@ -101,6 +101,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -110,6 +111,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -120,6 +122,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_info_mispredictions, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -167,6 +170,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -271,6 +275,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Sample with: FRONTEND_RETIRED.LATENCY_GE_2_BUBBLES_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_2_PS. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_dsb_misses, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -280,6 +285,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: FRONTEND_RETIRED.LATENCY_GE_16_PS;FRONTEND_RETIRED.LATENCY_GE_8_PS", "ScaleUnit": "100%" },
@@ -345,6 +351,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound. Sample with: FRONTEND_RETIRED.LATENCY_GE_4_PS", "ScaleUnit": "100%" },
@@ -363,6 +370,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -1065,6 +1073,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -1110,6 +1119,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -1138,6 +1148,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1343,6 +1354,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/skylakex/skx-metrics.json +12 -0
@@ -101,6 +101,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound.", "ScaleUnit": "100%" },
@@ -110,6 +111,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -120,6 +122,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_info_mispredictions, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -167,6 +170,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -271,6 +275,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 4 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Sample with: FRONTEND_RETIRED.LATENCY_GE_2_BUBBLES_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_2_PS. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_dsb_misses, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -280,6 +285,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: FRONTEND_RETIRED.LATENCY_GE_16_PS;FRONTEND_RETIRED.LATENCY_GE_8_PS", "ScaleUnit": "100%" },
@@ -354,6 +360,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound. Sample with: FRONTEND_RETIRED.LATENCY_GE_4_PS", "ScaleUnit": "100%" },
@@ -372,6 +379,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -1123,6 +1131,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -1177,6 +1186,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -1205,6 +1215,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1429,6 +1440,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.RETIRE_SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/arch/x86/tigerlake/tgl-metrics.json +12 -0
@@ -109,6 +109,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend. Backend is the portion of the processor core where the out-of-order scheduler dispatches ready uops into their respective execution units; and once completed these uops get retired according to program order. For example; stalls due to data-cache misses or stalls due to the divider unit being overloaded are both categorized under Backend Bound. Backend Bound is further divided into two main categories: Memory Bound and Core Bound. Sample with: TOPDOWN.BACKEND_BOUND_SLOTS", "ScaleUnit": "100%" },
@@ -118,6 +119,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots wasted due to incorrect speculations. This include slots used to issue uops that do not eventually get retired and slots for which the issue-pipeline was blocked due to recovery from earlier incorrect speculation. For example; wasted work due to miss-predicted branches are categorized under Bad Speculation category. Incorrect data speculation followed by Memory Ordering Nukes is another example.", "ScaleUnit": "100%" },
@@ -135,6 +137,7 @@ "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction. These slots are either wasted by uops fetched from an incorrectly speculated program path; or stalls when the out-of-order part of the machine needs to recover its state from a speculative path. Sample with: BR_MISP_RETIRED.ALL_BRANCHES. Related metrics: tma_info_branch_misprediction_cost, tma_info_mispredictions, tma_mispredicts_resteers", "ScaleUnit": "100%" },
@@ -181,6 +184,7 @@ "MetricGroup": "Backend;Compute;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_core_bound", "MetricThreshold": "tma_core_bound > 0.1 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where Core non-memory issues were of a bottleneck. Shortage in hardware compute resources; or dependencies in software's instructions are both categorized under Core Bound. Hence it may indicate the machine ran out of an out-of-order resource; certain execution units are overloaded or dependencies in program's data- or instruction-flow are limiting the performance (e.g. FP-chained long-latency arithmetic operations).", "ScaleUnit": "100%" },
@@ -282,6 +286,7 @@ "MetricGroup": "FetchBW;Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group;tma_issueFB", "MetricName": "tma_fetch_bandwidth", "MetricThreshold": "tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 5 > 0.35", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend bandwidth issues. For example; inefficiencies at the instruction decoders; or restrictions for caching in the DSB (decoded uops cache) are categorized under Fetch Bandwidth. In such cases; the Frontend typically delivers suboptimal amount of uops to the Backend. Sample with: FRONTEND_RETIRED.LATENCY_GE_2_BUBBLES_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_1_PS;FRONTEND_RETIRED.LATENCY_GE_2_PS. Related metrics: tma_dsb_switches, tma_info_dsb_coverage, tma_info_dsb_misses, tma_info_iptb, tma_lcp", "ScaleUnit": "100%" },
@@ -291,6 +296,7 @@ "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues. For example; instruction-cache misses; iTLB misses or fetch stalls after a branch misprediction are categorized under Frontend Latency. In such cases; the Frontend eventually delivers no uops for some period. Sample with: FRONTEND_RETIRED.LATENCY_GE_16_PS;FRONTEND_RETIRED.LATENCY_GE_8_PS", "ScaleUnit": "100%" },
@@ -363,6 +369,7 @@ "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend. Frontend denotes the first part of the processor core responsible to fetch operations that are executed later on by the Backend part. Within the Frontend; a branch predictor predicts the next address to fetch; cache-lines are fetched from the memory subsystem; parsed into instructions; and lastly decoded into micro-operations (uops). Ideally the Frontend can issue Pipeline_Width uops every cycle to the Backend. Frontend Bound denotes unutilized issue-slots when there is no Backend stall; i.e. bubbles where Frontend delivered no uops while Backend could have accepted them. For example; stalls due to instruction-cache misses would be categorized under Frontend Bound. Sample with: FRONTEND_RETIRED.LATENCY_GE_4_PS", "ScaleUnit": "100%" },
@@ -372,6 +379,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences. This highly-correlates with the uop length of these instructions/sequences.", "ScaleUnit": "100%" },
@@ -1125,6 +1133,7 @@ "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_light_operations", "MetricThreshold": "tma_light_operations > 0.6", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots where the CPU was retiring light-weight operations -- instructions that require no more than one uop (micro-operation). This correlates with total number of instructions used by the program. A uops-per-instruction (see UopPI metric) ratio of 1 or less should be expected for decently optimized software running on Intel Core/Xeon products. While this often indicates efficient X86 instructions were executed; high value does not necessarily mean better performance cannot be achieved. Sample with: INST_RETIRED.PREC_DIST", "ScaleUnit": "100%" },
@@ -1178,6 +1187,7 @@ "MetricGroup": "BadSpec;MachineClears;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueMC;tma_issueSyncxn", "MetricName": "tma_machine_clears", "MetricThreshold": "tma_machine_clears > 0.1 & tma_bad_speculation > 0.15", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the CPU has wasted due to Machine Clears. These slots are either wasted by uops fetched prior to the clear; or stalls the out-of-order portion of the machine needs to recover its state after the clear. For example; this can happen due to memory ordering Nukes (e.g. Memory Disambiguation) or Self-Modifying-Code (SMC) nukes. Sample with: MACHINE_CLEARS.COUNT. Related metrics: tma_clears_resteers, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_l1_bound, tma_microcode_sequencer, tma_ms_switches, tma_remote_cache", "ScaleUnit": "100%" },
@@ -1205,6 +1215,7 @@ "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2", + "MetricgroupNoGroup": "TopdownL2", "PublicDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck. Memory Bound estimates fraction of slots where pipeline is likely stalled due to demand load or store instructions. This accounts mainly for (1) non-completed in-flight memory demand loads which coincides with execution units starvation; in addition to (2) cases where stores could impose backpressure on the pipeline when many of them get buffered at the same time (less common out of the two).", "ScaleUnit": "100%" },
@@ -1374,6 +1385,7 @@ "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1", + "MetricgroupNoGroup": "TopdownL1", "PublicDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired. Ideally; all pipeline slots would be attributed to the Retiring category. Retiring of 100% would indicate the maximum Pipeline_Width throughput was achieved. Maximizing Retiring typically increases the Instructions-per-cycle (see IPC metric). Note that a high Retiring value does not necessary mean there is no room for more performance. For example; Heavy-operations or Microcode Assists are categorized under Retiring. They often indicate suboptimal performance and can often be optimized or avoided. Sample with: UOPS_RETIRED.SLOTS", "ScaleUnit": "100%" },
tools/perf/pmu-events/jevents.py +3 -1
@@ -52,7 +52,8 @@ _json_event_attributes = [ # Attributes that are in pmu_metric rather than pmu_event. _json_metric_attributes = [ 'metric_name', 'metric_group', 'metric_expr', 'metric_threshold', 'desc', - 'long_desc', 'unit', 'compat', 'aggr_mode', 'event_grouping' + 'long_desc', 'unit', 'compat', 'metricgroup_no_group', 'aggr_mode', + 'event_grouping' ] # Attributes that are bools or enum int values, encoded as '0', '1',... _json_enum_attributes = ['aggr_mode', 'deprecated', 'event_grouping', 'perpkg']
@@ -303,6 +304,7 @@ class JsonEvent: self.deprecated = jd.get('Deprecated') self.metric_name = jd.get('MetricName') self.metric_group = jd.get('MetricGroup') + self.metricgroup_no_group = jd.get('MetricgroupNoGroup') self.event_grouping = convert_metric_constraint(jd.get('MetricConstraint')) self.metric_expr = None if 'MetricExpr' in jd:
tools/perf/pmu-events/pmu-events.h +1 -0
@@ -59,6 +59,7 @@ struct pmu_metric {const char *compat;const char *desc;const char *long_desc; +const char *metricgroup_no_group;enum aggr_mode_class aggr_mode;enum metric_event_groups event_grouping; };
tools/perf/util/metricgroup.c +4 -4
@@ -1144,12 +1144,12 @@ static int metricgroup__add_metric_callback(const struct pmu_metric *pm,struct metricgroup__add_metric_data *data = vdata;int ret =0 ; -if (pm->metric_expr && - (match_metric (pm->metric_group, data->metric_name) || -match_metric (pm->metric_name, data->metric_name))) { +if (pm->metric_expr &&match_pm_metric (pm, data->metric_name)) { +bool metric_no_group = data->metric_no_group || +match_metric (data->metric_name, pm->metricgroup_no_group); data->has_match = true; - ret =add_metric (data->list, pm, data->modifier, data->metric_no_group, + ret =add_metric (data->list, pm, data->modifier, metric_no_group, data->metric_no_threshold, data->user_requested_cpu_list, data->system_wide,/*root_metric=*/ NULL ,/*visited_metrics=*/ NULL , table);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 5/46 perf parse-events: Don't reorder ungrouped events by pmu
In reply to Ian RogersThe pmu_group_name by default returns "cpu" which on non-hybrid/ARM means that ungrouped software, and hardware events are all going to sort by the original insertion index. However, on hybrid and ARM wildcard expansion may mean the PMU name is set and events will be unnecessarily reordered - triggering the reordering warning.
--- tools/perf/util/parse-events.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)
tools/perf/util/parse-events.c +15 -8
@@ -2140,25 +2140,32 @@ static int evlist__cmp(void *state, const struct list_head *l, const struct listint *leader_idx = state;int lhs_leader_idx = *leader_idx, rhs_leader_idx = *leader_idx, ret;const char *lhs_pmu_name, *rhs_pmu_name; +bool lhs_has_group = false, rhs_has_group = false;/* * First sort by grouping/leader. Read the leader idx only if the evsel * is part of a group, as -1 indicates no group. */ -if (lhs_core->leader != lhs_core || lhs_core->nr_members >1 ) +if (lhs_core->leader != lhs_core || lhs_core->nr_members >1 ) { + lhs_has_group = true; lhs_leader_idx = lhs_core->leader->idx; -if (rhs_core->leader != rhs_core || rhs_core->nr_members >1 ) + } +if (rhs_core->leader != rhs_core || rhs_core->nr_members >1 ) { + rhs_has_group = true; rhs_leader_idx = rhs_core->leader->idx; + }if (lhs_leader_idx != rhs_leader_idx)return lhs_leader_idx - rhs_leader_idx; -/* Group by PMU. Groups can't span PMUs. */ - lhs_pmu_name =evsel__group_pmu_name (lhs); - rhs_pmu_name =evsel__group_pmu_name (rhs); - ret =strcmp (lhs_pmu_name, rhs_pmu_name); -if (ret) -return ret; +/* Group by PMU if there is a group. Groups can't span PMUs. */ +if (lhs_has_group && rhs_has_group) { + lhs_pmu_name =evsel__group_pmu_name (lhs); + rhs_pmu_name =evsel__group_pmu_name (rhs); + ret =strcmp (lhs_pmu_name, rhs_pmu_name); +if (ret) +return ret; + }/* Architecture specific sorting. */ return arch_evlist__cmp (lhs, rhs);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 6/46 perf vendor events intel: Add alderlake metric constraints
In reply to Ian RogersPreviously these constraints were disabled as they contained topdown events. Since:
https://lore.kernel.org/all/[email protected]/
the topdown events are correctly grouped even if no group exists.
This change was created by PR:
https://github.com/intel/perfmon/pull/71
--- .../pmu-events/arch/x86/alderlake/adl-metrics.json | 11 +++++++++++ 1 file changed, 11 insertions(+)
tools/perf/pmu-events/arch/x86/alderlake/adl-metrics.json +11 -0
@@ -1077,6 +1077,7 @@ }, { "BriefDescription": "This metric represents overall arithmetic floating-point (FP) operations fraction the CPU has executed (retired)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_x87_use + tma_fp_scalar + tma_fp_vector", "MetricGroup": "HPC;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_fp_arith",
@@ -1203,6 +1204,7 @@ }, { "BriefDescription": "Branch Misprediction Cost: Fraction of TMA slots wasted per non-speculative branch misprediction (retired JEClear)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) * tma_info_slots / BR_MISP_RETIRED.ALL_BRANCHES", "MetricGroup": "Bad;BrMispredicts;tma_issueBM", "MetricName": "tma_info_branch_misprediction_cost",
@@ -1255,6 +1257,7 @@ }, { "BriefDescription": "Probability of Core Bound bottleneck hidden by SMT-profiling artifacts", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(100 * (1 - tma_core_bound / tma_ports_utilization if tma_core_bound < tma_ports_utilization else 1) if tma_info_smt_2t_utilization > 0.5 else 0)", "MetricGroup": "Cor;SMT", "MetricName": "tma_info_core_bound_likely",
@@ -1315,6 +1318,7 @@ }, { "BriefDescription": "Total pipeline cost of DSB (uop cache) misses - subset of the Instruction_Fetch_BW Bottleneck", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_fetch_latency * tma_dsb_switches / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches) + tma_fetch_bandwidth * tma_mite / (tma_dsb + tma_lsd + tma_mite))", "MetricGroup": "DSBmiss;Fed;tma_issueFB", "MetricName": "tma_info_dsb_misses",
@@ -1408,6 +1412,7 @@ }, { "BriefDescription": "Total pipeline cost of instruction fetch bandwidth related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_frontend_bound - tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) - tma_info_big_code", "MetricGroup": "Fed;FetchBW;Frontend", "MetricName": "tma_info_instruction_fetch_bw",
@@ -1827,6 +1832,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_l1_bound / max(tma_memory_bound, tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_dtlb_load / max(tma_l1_bound, tma_dtlb_load + tma_fb_full + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_info_memory_data_tlbs",
@@ -1836,6 +1842,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Latency related bottlenecks (external memory and off-core caches)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_l2_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound))", "MetricGroup": "Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_info_memory_latency",
@@ -1845,6 +1852,7 @@ }, { "BriefDescription": "Total pipeline cost of Branch Misprediction related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches))", "MetricGroup": "Bad;BadSpec;BrMispredicts;tma_issueBM", "MetricName": "tma_info_mispredictions",
@@ -1877,6 +1885,7 @@ }, { "BriefDescription": "Average number of Uops retired in cycles where at least one uop has retired.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_retiring * tma_info_slots / cpu_core@UOPS_RETIRED.SLOTS\\,cmask\\=1@", "MetricGroup": "Pipeline;Ret", "MetricName": "tma_info_retire",
@@ -2152,6 +2161,7 @@ }, { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring memory operations -- uops for memory load or store accesses.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_light_operations * MEM_UOP_RETIRED.ANY / (tma_retiring * tma_info_slots)", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_memory_operations",
@@ -2231,6 +2241,7 @@ }, { "BriefDescription": "This metric represents the remaining light uops fraction the CPU has executed - remaining means not covered by other sibling nodes", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "max(0, tma_light_operations - (tma_fp_arith + tma_int_operations + tma_memory_operations + tma_fused_instructions + tma_non_fused_branches + tma_nop_instructions))", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_other_light_ops",
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 7/46 perf vendor events intel: Add icelake metric constraints
In reply to Ian RogersPreviously these constraints were disabled as they contained topdown events. Since:
https://lore.kernel.org/all/[email protected]/
the topdown events are correctly grouped even if no group exists.
This change was created by PR:
https://github.com/intel/perfmon/pull/71
--- .../perf/pmu-events/arch/x86/icelake/icl-metrics.json | 11 +++++++++++ 1 file changed, 11 insertions(+)
tools/perf/pmu-events/arch/x86/icelake/icl-metrics.json +11 -0
@@ -317,6 +317,7 @@ }, { "BriefDescription": "This metric represents overall arithmetic floating-point (FP) operations fraction the CPU has executed (retired)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_x87_use + tma_fp_scalar + tma_fp_vector", "MetricGroup": "HPC;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_fp_arith",
@@ -421,6 +422,7 @@ }, { "BriefDescription": "Branch Misprediction Cost: Fraction of TMA slots wasted per non-speculative branch misprediction (retired JEClear)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) * tma_info_slots / BR_MISP_RETIRED.ALL_BRANCHES", "MetricGroup": "Bad;BrMispredicts;tma_issueBM", "MetricName": "tma_info_branch_misprediction_cost",
@@ -466,6 +468,7 @@ }, { "BriefDescription": "Probability of Core Bound bottleneck hidden by SMT-profiling artifacts", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(100 * (1 - tma_core_bound / tma_ports_utilization if tma_core_bound < tma_ports_utilization else 1) if tma_info_smt_2t_utilization > 0.5 else 0)", "MetricGroup": "Cor;SMT", "MetricName": "tma_info_core_bound_likely",
@@ -518,6 +521,7 @@ }, { "BriefDescription": "Total pipeline cost of DSB (uop cache) misses - subset of the Instruction_Fetch_BW Bottleneck", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_fetch_latency * tma_dsb_switches / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches) + tma_fetch_bandwidth * tma_mite / (tma_dsb + tma_lsd + tma_mite))", "MetricGroup": "DSBmiss;Fed;tma_issueFB", "MetricName": "tma_info_dsb_misses",
@@ -599,6 +603,7 @@ }, { "BriefDescription": "Total pipeline cost of instruction fetch bandwidth related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_frontend_bound - tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) - tma_info_big_code", "MetricGroup": "Fed;FetchBW;Frontend", "MetricName": "tma_info_instruction_fetch_bw",
@@ -937,6 +942,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_l1_bound / max(tma_memory_bound, tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_dtlb_load / max(tma_l1_bound, tma_4k_aliasing + tma_dtlb_load + tma_fb_full + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_info_memory_data_tlbs",
@@ -945,6 +951,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Latency related bottlenecks (external memory and off-core caches)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_l2_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound))", "MetricGroup": "Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_info_memory_latency",
@@ -953,6 +960,7 @@ }, { "BriefDescription": "Total pipeline cost of Branch Misprediction related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches))", "MetricGroup": "Bad;BadSpec;BrMispredicts;tma_issueBM", "MetricName": "tma_info_mispredictions",
@@ -1004,6 +1012,7 @@ }, { "BriefDescription": "Average number of Uops retired in cycles where at least one uop has retired.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_retiring * tma_info_slots / cpu@UOPS_RETIRED.SLOTS\\,cmask\\=1@", "MetricGroup": "Pipeline;Ret", "MetricName": "tma_info_retire"
@@ -1207,6 +1216,7 @@ }, { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring memory operations -- uops for memory load or store accesses.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_light_operations * MEM_INST_RETIRED.ANY / INST_RETIRED.ANY", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_memory_operations",
@@ -1277,6 +1287,7 @@ }, { "BriefDescription": "This metric represents the remaining light uops fraction the CPU has executed - remaining means not covered by other sibling nodes", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "max(0, tma_light_operations - (tma_fp_arith + tma_memory_operations + tma_branch_instructions + tma_nop_instructions))", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_other_light_ops",
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 8/46 perf vendor events intel: Add icelakex metric constraints
In reply to Ian RogersPreviously these constraints were disabled as they contained topdown events. Since:
https://lore.kernel.org/all/[email protected]/
the topdown events are correctly grouped even if no group exists.
This change was created by PR:
https://github.com/intel/perfmon/pull/71
--- .../pmu-events/arch/x86/icelakex/icx-metrics.json | 11 +++++++++++ 1 file changed, 11 insertions(+)
tools/perf/pmu-events/arch/x86/icelakex/icx-metrics.json +11 -0
@@ -282,6 +282,7 @@ }, { "BriefDescription": "This metric represents overall arithmetic floating-point (FP) operations fraction the CPU has executed (retired)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_x87_use + tma_fp_scalar + tma_fp_vector", "MetricGroup": "HPC;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_fp_arith",
@@ -386,6 +387,7 @@ }, { "BriefDescription": "Branch Misprediction Cost: Fraction of TMA slots wasted per non-speculative branch misprediction (retired JEClear)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) * tma_info_slots / BR_MISP_RETIRED.ALL_BRANCHES", "MetricGroup": "Bad;BrMispredicts;tma_issueBM", "MetricName": "tma_info_branch_misprediction_cost",
@@ -431,6 +433,7 @@ }, { "BriefDescription": "Probability of Core Bound bottleneck hidden by SMT-profiling artifacts", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(100 * (1 - tma_core_bound / tma_ports_utilization if tma_core_bound < tma_ports_utilization else 1) if tma_info_smt_2t_utilization > 0.5 else 0)", "MetricGroup": "Cor;SMT", "MetricName": "tma_info_core_bound_likely",
@@ -483,6 +486,7 @@ }, { "BriefDescription": "Total pipeline cost of DSB (uop cache) misses - subset of the Instruction_Fetch_BW Bottleneck", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_fetch_latency * tma_dsb_switches / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches) + tma_fetch_bandwidth * tma_mite / (tma_dsb + tma_mite))", "MetricGroup": "DSBmiss;Fed;tma_issueFB", "MetricName": "tma_info_dsb_misses",
@@ -564,6 +568,7 @@ }, { "BriefDescription": "Total pipeline cost of instruction fetch bandwidth related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_frontend_bound - tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) - tma_info_big_code", "MetricGroup": "Fed;FetchBW;Frontend", "MetricName": "tma_info_instruction_fetch_bw",
@@ -948,6 +953,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_l1_bound / max(tma_memory_bound, tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound) * (tma_dtlb_load / max(tma_l1_bound, tma_4k_aliasing + tma_dtlb_load + tma_fb_full + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_info_memory_data_tlbs",
@@ -956,6 +962,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Latency related bottlenecks (external memory and off-core caches)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_l2_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound))", "MetricGroup": "Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_info_memory_latency",
@@ -964,6 +971,7 @@ }, { "BriefDescription": "Total pipeline cost of Branch Misprediction related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches))", "MetricGroup": "Bad;BadSpec;BrMispredicts;tma_issueBM", "MetricName": "tma_info_mispredictions",
@@ -1027,6 +1035,7 @@ }, { "BriefDescription": "Average number of Uops retired in cycles where at least one uop has retired.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_retiring * tma_info_slots / cpu@UOPS_RETIRED.SLOTS\\,cmask\\=1@", "MetricGroup": "Pipeline;Ret", "MetricName": "tma_info_retire"
@@ -1230,6 +1239,7 @@ }, { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring memory operations -- uops for memory load or store accesses.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_light_operations * MEM_INST_RETIRED.ANY / INST_RETIRED.ANY", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_memory_operations",
@@ -1300,6 +1310,7 @@ }, { "BriefDescription": "This metric represents the remaining light uops fraction the CPU has executed - remaining means not covered by other sibling nodes", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "max(0, tma_light_operations - (tma_fp_arith + tma_memory_operations + tma_branch_instructions + tma_nop_instructions))", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_other_light_ops",
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 9/46 perf vendor events intel: Add sapphirerapids metric constraints
In reply to Ian RogersPreviously these constraints were disabled as they contained topdown events. Since:
https://lore.kernel.org/all/[email protected]/
the topdown events are correctly grouped even if no group exists.
This change was created by PR:
https://github.com/intel/perfmon/pull/71
--- .../arch/x86/sapphirerapids/spr-metrics.json | 11 +++++++++++ 1 file changed, 11 insertions(+)
tools/perf/pmu-events/arch/x86/sapphirerapids/spr-metrics.json +11 -0
@@ -290,6 +290,7 @@ }, { "BriefDescription": "This metric represents overall arithmetic floating-point (FP) operations fraction the CPU has executed (retired)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_x87_use + tma_fp_scalar + tma_fp_vector + tma_fp_amx", "MetricGroup": "HPC;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_fp_arith",
@@ -412,6 +413,7 @@ }, { "BriefDescription": "Branch Misprediction Cost: Fraction of TMA slots wasted per non-speculative branch misprediction (retired JEClear)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) * tma_info_slots / BR_MISP_RETIRED.ALL_BRANCHES", "MetricGroup": "Bad;BrMispredicts;tma_issueBM", "MetricName": "tma_info_branch_misprediction_cost",
@@ -457,6 +459,7 @@ }, { "BriefDescription": "Probability of Core Bound bottleneck hidden by SMT-profiling artifacts", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(100 * (1 - tma_core_bound / tma_ports_utilization if tma_core_bound < tma_ports_utilization else 1) if tma_info_smt_2t_utilization > 0.5 else 0)", "MetricGroup": "Cor;SMT", "MetricName": "tma_info_core_bound_likely",
@@ -509,6 +512,7 @@ }, { "BriefDescription": "Total pipeline cost of DSB (uop cache) misses - subset of the Instruction_Fetch_BW Bottleneck", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_fetch_latency * tma_dsb_switches / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches) + tma_fetch_bandwidth * tma_mite / (tma_dsb + tma_mite))", "MetricGroup": "DSBmiss;Fed;tma_issueFB", "MetricName": "tma_info_dsb_misses",
@@ -590,6 +594,7 @@ }, { "BriefDescription": "Total pipeline cost of instruction fetch bandwidth related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_frontend_bound - tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) - tma_info_big_code", "MetricGroup": "Fed;FetchBW;Frontend", "MetricName": "tma_info_instruction_fetch_bw",
@@ -998,6 +1003,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_l1_bound / max(tma_memory_bound, tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound) * (tma_dtlb_load / max(tma_l1_bound, tma_dtlb_load + tma_fb_full + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_info_memory_data_tlbs",
@@ -1006,6 +1012,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Latency related bottlenecks (external memory and off-core caches)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_l2_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_pmm_bound + tma_store_bound))", "MetricGroup": "Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_info_memory_latency",
@@ -1014,6 +1021,7 @@ }, { "BriefDescription": "Total pipeline cost of Branch Misprediction related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches))", "MetricGroup": "Bad;BadSpec;BrMispredicts;tma_issueBM", "MetricName": "tma_info_mispredictions",
@@ -1054,6 +1062,7 @@ }, { "BriefDescription": "Average number of Uops retired in cycles where at least one uop has retired.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_retiring * tma_info_slots / cpu@UOPS_RETIRED.SLOTS\\,cmask\\=1@", "MetricGroup": "Pipeline;Ret", "MetricName": "tma_info_retire"
@@ -1328,6 +1337,7 @@ }, { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring memory operations -- uops for memory load or store accesses.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_light_operations * MEM_UOP_RETIRED.ANY / (tma_retiring * tma_info_slots)", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_memory_operations",
@@ -1399,6 +1409,7 @@ }, { "BriefDescription": "This metric represents the remaining light uops fraction the CPU has executed - remaining means not covered by other sibling nodes", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "max(0, tma_light_operations - (tma_fp_arith + tma_int_operations + tma_memory_operations + tma_fused_instructions + tma_non_fused_branches + tma_nop_instructions))", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_other_light_ops",
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 10/46 perf vendor events intel: Add tigerlake metric constraints
In reply to Ian RogersPreviously these constraints were disabled as they contained topdown events. Since:
https://lore.kernel.org/all/[email protected]/
the topdown events are correctly grouped even if no group exists.
This change was created by PR:
https://github.com/intel/perfmon/pull/71
--- .../pmu-events/arch/x86/tigerlake/tgl-metrics.json | 11 +++++++++++ 1 file changed, 11 insertions(+)
tools/perf/pmu-events/arch/x86/tigerlake/tgl-metrics.json +11 -0
@@ -311,6 +311,7 @@ }, { "BriefDescription": "This metric represents overall arithmetic floating-point (FP) operations fraction the CPU has executed (retired)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_x87_use + tma_fp_scalar + tma_fp_vector", "MetricGroup": "HPC;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_fp_arith",
@@ -415,6 +416,7 @@ }, { "BriefDescription": "Branch Misprediction Cost: Fraction of TMA slots wasted per non-speculative branch misprediction (retired JEClear)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) * tma_info_slots / BR_MISP_RETIRED.ALL_BRANCHES", "MetricGroup": "Bad;BrMispredicts;tma_issueBM", "MetricName": "tma_info_branch_misprediction_cost",
@@ -460,6 +462,7 @@ }, { "BriefDescription": "Probability of Core Bound bottleneck hidden by SMT-profiling artifacts", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(100 * (1 - tma_core_bound / tma_ports_utilization if tma_core_bound < tma_ports_utilization else 1) if tma_info_smt_2t_utilization > 0.5 else 0)", "MetricGroup": "Cor;SMT", "MetricName": "tma_info_core_bound_likely",
@@ -512,6 +515,7 @@ }, { "BriefDescription": "Total pipeline cost of DSB (uop cache) misses - subset of the Instruction_Fetch_BW Bottleneck", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_fetch_latency * tma_dsb_switches / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches) + tma_fetch_bandwidth * tma_mite / (tma_dsb + tma_lsd + tma_mite))", "MetricGroup": "DSBmiss;Fed;tma_issueFB", "MetricName": "tma_info_dsb_misses",
@@ -593,6 +597,7 @@ }, { "BriefDescription": "Total pipeline cost of instruction fetch bandwidth related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_frontend_bound - tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches)) - tma_info_big_code", "MetricGroup": "Fed;FetchBW;Frontend", "MetricName": "tma_info_instruction_fetch_bw",
@@ -957,6 +962,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_l1_bound / max(tma_memory_bound, tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_dtlb_load / max(tma_l1_bound, tma_4k_aliasing + tma_dtlb_load + tma_fb_full + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_info_memory_data_tlbs",
@@ -965,6 +971,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Latency related bottlenecks (external memory and off-core caches)", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_l2_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound))", "MetricGroup": "Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_info_memory_latency",
@@ -973,6 +980,7 @@ }, { "BriefDescription": "Total pipeline cost of Branch Misprediction related bottlenecks", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "100 * (tma_branch_mispredicts + tma_fetch_latency * tma_mispredicts_resteers / (tma_branch_resteers + tma_dsb_switches + tma_icache_misses + tma_itlb_misses + tma_lcp + tma_ms_switches))", "MetricGroup": "Bad;BadSpec;BrMispredicts;tma_issueBM", "MetricName": "tma_info_mispredictions",
@@ -1024,6 +1032,7 @@ }, { "BriefDescription": "Average number of Uops retired in cycles where at least one uop has retired.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_retiring * tma_info_slots / cpu@UOPS_RETIRED.SLOTS\\,cmask\\=1@", "MetricGroup": "Pipeline;Ret", "MetricName": "tma_info_retire"
@@ -1221,6 +1230,7 @@ }, { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring memory operations -- uops for memory load or store accesses.", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "tma_light_operations * MEM_INST_RETIRED.ANY / INST_RETIRED.ANY", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_memory_operations",
@@ -1291,6 +1301,7 @@ }, { "BriefDescription": "This metric represents the remaining light uops fraction the CPU has executed - remaining means not covered by other sibling nodes", + "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "max(0, tma_light_operations - (tma_fp_arith + tma_memory_operations + tma_branch_instructions + tma_nop_instructions))", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_other_light_ops",
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 11/46 perf stat: Avoid segv on counter->name
In reply to Ian RogersSwitch to use evsel__name that doesn't return NULL for hardware and similar events.
--- tools/perf/util/stat-display.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/stat-display.c +1 -1
@@ -747,7 +747,7 @@ static void uniquify_event_name(struct evsel *counter)int ret =0 ;if (counter->uniquified_name || counter->use_config_name || - !counter->pmu_name || !strncmp (counter->name, counter->pmu_name, + !counter->pmu_name || !strncmp (evsel__name (counter), counter->pmu_name,strlen (counter->pmu_name)))return ;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 12/46 perf test: Test more sysfs events
In reply to Ian RogersParse events for all PMUs, and not just cpu, in test "Parsing of all PMU events from sysfs".
--- tools/perf/tests/parse-events.c | 121 +++++++++++++++++--------------- 1 file changed, 65 insertions(+), 56 deletions(-)
tools/perf/tests/parse-events.c +65 -56
@@ -7,6 +7,7 @@#include "debug.h" #include "pmu.h" #include "pmu-hybrid.h" +#include "pmus.h" #include <dirent.h> #include <errno.h> #include "fncache.h"
@@ -2225,74 +2226,82 @@ static int test_pmu(void)static int test__pmu_events (struct test_suite *test__maybe_unused ,int subtest__maybe_unused ) { -struct stat st; -char path[PATH_MAX ]; -struct dirent *ent; -DIR *dir; -int ret; +struct perf_pmu *pmu; +int ret =TEST_OK ; -if (!test_pmu ()) -return TEST_SKIP ; +perf_pmus__for_each_pmu (pmu) { +struct stat st; +char path[PATH_MAX ]; +struct dirent *ent; +DIR *dir; +int err; -snprintf (path,PATH_MAX ,"%s/bus/event_source/devices/cpu/events/" , -sysfs__mountpoint ()); +snprintf (path,PATH_MAX ,"%s/bus/event_source/devices/%s/events/" , +sysfs__mountpoint (), pmu->name); - ret =stat (path, &st); -if (ret) { -pr_debug ("omitting PMU cpu events tests: %s\n" , path); -return TEST_OK ; - } + err =stat (path, &st); +if (err) { +pr_debug ("skipping PMU %s events tests: %s\n" , pmu->name, path); + ret =combine_test_results (ret,TEST_SKIP ); +continue ; + } - dir =opendir (path); -if (!dir) { -pr_debug ("can't open pmu event dir: %s\n" , path); -return TEST_FAIL ; - } + dir =opendir (path); +if (!dir) { +pr_debug ("can't open pmu event dir: %s\n" , path); + ret =combine_test_results (ret,TEST_SKIP ); +continue ; + } - ret =TEST_OK ; -while ((ent =readdir (dir))) { -struct evlist_test e = { .name =NULL , }; -char name[2 *NAME_MAX +1 +12 +3 ]; -int test_ret; +while ((ent =readdir (dir))) { +struct evlist_test e = { .name =NULL , }; +char name[2 *NAME_MAX +1 +12 +3 ]; +int test_ret; -/* Names containing . are special and cannot be used directly */ -if (strchr (ent->d_name,'.' )) -continue ; +/* Names containing . are special and cannot be used directly */ +if (strchr (ent->d_name,'.' )) +continue ; -snprintf (name,sizeof (name),"cpu/event=%s/u" , ent->d_name); +snprintf (name,sizeof (name),"%s/event=%s/u" , pmu->name, ent->d_name); - e.name = name; - e.check = test__checkevent_pmu_events; + e.name = name; + e.check = test__checkevent_pmu_events; - test_ret =test_event (&e); -if (test_ret !=TEST_OK ) { -pr_debug ("Test PMU event failed for '%s'" , name); - ret =combine_test_results (ret, test_ret); - } -/* -* Names containing '-' are recognized as prefixes and suffixes -* due to '-' being a legacy PMU separator. This fails when the -* prefix or suffix collides with an existing legacy token. For -* example, branch-brs has a prefix (branch) that collides with -* a PE_NAME_CACHE_TYPE token causing a parse error as a suffix -* isn't expected after this. As event names in the config -* slashes are allowed a '-' in the name we check this works -* above. -*/ -if (strchr (ent->d_name,'-' )) -continue ; + test_ret =test_event (&e); +if (test_ret !=TEST_OK ) { +pr_debug ("Test PMU event failed for '%s'" , name); + ret =combine_test_results (ret, test_ret); + } -snprintf (name,sizeof (name),"%s:u,cpu/event=%s/u" , ent->d_name, ent->d_name); - e.name = name; - e.check = test__checkevent_pmu_events_mix; - test_ret =test_event (&e); -if (test_ret !=TEST_OK ) { -pr_debug ("Test PMU event failed for '%s'" , name); - ret =combine_test_results (ret, test_ret); +if (!is_pmu_core (pmu->name)) +continue ; + +/* +* Names containing '-' are recognized as prefixes and suffixes +* due to '-' being a legacy PMU separator. This fails when the +* prefix or suffix collides with an existing legacy token. For +* example, branch-brs has a prefix (branch) that collides with +* a PE_NAME_CACHE_TYPE token causing a parse error as a suffix +* isn't expected after this. As event names in the config +* slashes are allowed a '-' in the name we check this works +* above. +*/ +if (strchr (ent->d_name,'-' )) +continue ; + +snprintf (name,sizeof (name),"%s:u,%s/event=%s/u" , + ent->d_name, pmu->name, ent->d_name); + e.name = name; + e.check = test__checkevent_pmu_events_mix; + test_ret =test_event (&e); +if (test_ret !=TEST_OK ) { +pr_debug ("Test PMU event failed for '%s'" , name); + ret =combine_test_results (ret, test_ret); + } } - } -closedir (dir); +closedir (dir); + }return ret; }
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Re: [PATCH v3 12/46] perf test: Test more sysfs events
Quoted 5 lines
tools/perf/tests/parse-events.c +0
-1@@ -2225,74 +2226,82 @@ static int test_pmu(void)static int test__pmu_events (struct test_suite *test__maybe_unused ,int subtest__maybe_unused ) { -struct stat st;
tools/perf/tests/parse-events.c static int test_pmu(void) · line 2232
Quoted 10 lines
tools/perf/tests/parse-events.c +3
-6-char path[PATH_MAX ]; -struct dirent *ent; -DIR *dir; -int ret; +struct perf_pmu *pmu; +int ret =TEST_OK ; -if (!test_pmu ()) -return TEST_SKIP ; +perf_pmus__for_each_pmu (pmu) {
'pmus' list might be empty and, if so, we need to fill it first with:
if (list_empty(&pmus))
perf_pmu__scan(NULL);before iterating over it. Other option is to add this code to perf_pmus__for_each_pmu() macro itself.
With that, the test fails for me:
$ sudo ./perf test 6 6: Parse event definition strings : 6.1: Test event parsing : Ok 6.2: Test parsing of "hybrid" CPU events : Skip (not hybrid) 6.3: Parsing of all PMU events from sysfs : FAILED!
Quoted 10 lines
tools/perf/tests/parse-events.c +7
-2+struct stat st; +char path[PATH_MAX ]; +struct dirent *ent; +DIR *dir; +int err; -snprintf (path,PATH_MAX ,"%s/bus/event_source/devices/cpu/events/" , -sysfs__mountpoint ()); +snprintf (path,PATH_MAX ,"%s/bus/event_source/devices/%s/events/" , +sysfs__mountpoint (), pmu->name);
tools/perf/tests/parse-events.c static int test_pmu(void) · line 2245
Quoted 10 lines
tools/perf/tests/parse-events.c +4
-5- ret =stat (path, &st); -if (ret) { -pr_debug ("omitting PMU cpu events tests: %s\n" , path); -return TEST_OK ; - } + err =stat (path, &st); +if (err) { +pr_debug ("skipping PMU %s events tests: %s\n" , pmu->name, path); + ret =combine_test_results (ret,TEST_SKIP );
combine_test_results(ret, TEST_OK); probably? Since many pmus don't expose events via sysfs.
Ian Rogers
reviewerRe: [PATCH v3 12/46] perf test: Test more sysfs events
Quoted 54 lines · Ravi Bangoria
Quoted 6 lines
(quoted diff) +0
-1@@ -2225,74 +2226,82 @@ static int test_pmu(void) static int test__pmu_events(struct test_suite *test __maybe_unused, int subtest __maybe_unused) { - struct stat st;Review static int test_pmu(void) · line 2232
Quoted 10 lines
(quoted diff) +3
-6- char path[PATH_MAX]; - struct dirent *ent; - DIR *dir; - int ret; + struct perf_pmu *pmu; + int ret = TEST_OK; - if (!test_pmu()) - return TEST_SKIP; + perf_pmus__for_each_pmu(pmu) {'pmus' list might be empty and, if so, we need to fill it first with:
if (list_empty(&pmus)) perf_pmu__scan(NULL);before iterating over it. Other option is to add this code to
perf_pmus__for_each_pmu()macro itself.With that, the test fails for me:
$ sudo ./perf test 6 6: Parse event definition strings : 6.1: Test event parsing : Ok 6.2: Test parsing of "hybrid" CPU events : Skip (not hybrid) 6.3: Parsing of all PMU events from sysfs : FAILED!Quoted 10 lines
(quoted diff) +7
-2+ struct stat st; + char path[PATH_MAX]; + struct dirent *ent; + DIR *dir; + int err; - snprintf(path, PATH_MAX, "%s/bus/event_source/devices/cpu/events/", - sysfs__mountpoint()); + snprintf(path, PATH_MAX, "%s/bus/event_source/devices/%s/events/", + sysfs__mountpoint(), pmu->name);Review static int test_pmu(void) · line 2245
Quoted 10 lines
(quoted diff) +4
-5- ret = stat(path, &st); - if (ret) { - pr_debug("omitting PMU cpu events tests: %s\n", path); - return TEST_OK; - } + err = stat(path, &st); + if (err) { + pr_debug("skipping PMU %s events tests: %s\n", pmu->name, path); + ret = combine_test_results(ret, TEST_SKIP);
Quoted 2 lines
combine_test_results(ret, TEST_OK); probably? Since many pmus don't expose events via sysfs.
Thanks Ravi! The following looks to address the issues and I'll add it to v4.
'''
tools/perf/tests/parse-events.c +0 -0
@@ -559,7 +559,8 @@ static int test__checkevent_pmu_events(struct
evlist *evlist)
struct evsel *evsel = evlist__first(evlist);
TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->core.nr_entries);
- TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->core.attr.type ||
+ strncmp(evsel->pmu_name, "cpu/", 4));
TEST_ASSERT_VAL("wrong exclude_user",
!evsel->core.attr.exclude_user);
TEST_ASSERT_VAL("wrong exclude_kernel",
@@ -2229,6 +2230,9 @@ static int test__pmu_events(struct test_suite
*test __maybe_unused, int subtes
t
struct perf_pmu *pmu;
int ret = TEST_OK;+ if (list_empty(&pmus))
+ perf_pmu__scan(NULL);
+
perf_pmus__for_each_pmu(pmu) {
struct stat st;
char path[PATH_MAX];
@@ -2242,7 +2246,6 @@ static int test__pmu_events(struct test_suite
*test __maybe_unused, int subtes
t
err = stat(path, &st);
if (err) {
pr_debug("skipping PMU %s events tests: %s\n",
pmu->name, path);
- ret = combine_test_results(ret, TEST_SKIP);
continue;
}'''
Ian Rogers
reviewerRe: [PATCH v3 12/46] perf test: Test more sysfs events
Quoted 67 lines · Ian Rogers
Quoted 54 lines · Ravi Bangoria
Quoted 6 lines
(quoted diff) +0
-1@@ -2225,74 +2226,82 @@ static int test_pmu(void) static int test__pmu_events(struct test_suite *test __maybe_unused, int subtest __maybe_unused) { - struct stat st;Review static int test_pmu(void) · line 2232
Quoted 10 lines
(quoted diff) +3
-6- char path[PATH_MAX]; - struct dirent *ent; - DIR *dir; - int ret; + struct perf_pmu *pmu; + int ret = TEST_OK; - if (!test_pmu()) - return TEST_SKIP; + perf_pmus__for_each_pmu(pmu) {'pmus' list might be empty and, if so, we need to fill it first with:
if (list_empty(&pmus)) perf_pmu__scan(NULL);before iterating over it. Other option is to add this code to
perf_pmus__for_each_pmu()macro itself.With that, the test fails for me:
$ sudo ./perf test 6 6: Parse event definition strings : 6.1: Test event parsing : Ok 6.2: Test parsing of "hybrid" CPU events : Skip (not hybrid) 6.3: Parsing of all PMU events from sysfs : FAILED!Quoted 10 lines
(quoted diff) +7
-2+ struct stat st; + char path[PATH_MAX]; + struct dirent *ent; + DIR *dir; + int err; - snprintf(path, PATH_MAX, "%s/bus/event_source/devices/cpu/events/", - sysfs__mountpoint()); + snprintf(path, PATH_MAX, "%s/bus/event_source/devices/%s/events/", + sysfs__mountpoint(), pmu->name);Review static int test_pmu(void) · line 2245
Quoted 10 lines
(quoted diff) +4
-5- ret = stat(path, &st); - if (ret) { - pr_debug("omitting PMU cpu events tests: %s\n", path); - return TEST_OK; - } + err = stat(path, &st); + if (err) { + pr_debug("skipping PMU %s events tests: %s\n", pmu->name, path); + ret = combine_test_results(ret, TEST_SKIP);Quoted 2 lines
combine_test_results(ret, TEST_OK); probably? Since many pmus don't expose events via sysfs.Thanks Ravi! The following looks to address the issues and I'll add it to v4.
'''
tools/perf/tests/parse-events.c +0
-0
tools/perf/tests/parse-events.c static int test__checkevent_pmu_events(struct · line 564
Quoted 7 lines
tools/perf/tests/parse-events.c +2
-1@@ -559,7 +559,8 @@ static int test__checkevent_pmu_events(struct evlist *evlist)struct evsel *evsel =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type || +strncmp (evsel->pmu_name,"cpu/" ,4 ));
This should be strcmp(evsel->pmu_name, "cpu"). Or we could drop the assert. I'll try to keep what's covered currently by the test while extending the PMUs.
Thanks,
Ian
Quoted 26 lines
tools/perf/tests/parse-events.c +0
-0TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL("wrong
exclude_kernel",tools/perf/tests/parse-events.c +0
-0@@ -2229,6 +2230,9 @@ static int test__pmu_events(struct test_suite *test __maybe_unused, int subtest
struct perf_pmu *pmu; int ret = TEST_OK;tools/perf/tests/parse-events.c +3
-0+if (list_empty (&pmus)) +perf_pmu__scan (NULL ); +perf_pmus__for_each_pmu (pmu) {struct stat st;char path[PATH_MAX ];@@ -2242,7 +2246,6 @@ static int test__pmu_events(struct test_suite *test __maybe_unused, int subtest
err = stat(path, &st); if (err) { pr_debug("skipping PMU %s events tests: %s\n", pmu->name, path); - ret = combine_test_results(ret, TEST_SKIP); continue; }'''
Ian Rogers
reviewerPATCH v3 13/46 perf test: Use valid for PMU tests
In reply to Ian RogersRather than skip all tests in test__events_pmu if PMU cpu isn't present, use the per-test valid test. This allows the running of software PMU tests on hybrid and arm systems.
--- tools/perf/tests/parse-events.c | 27 +++++++++------------------ 1 file changed, 9 insertions(+), 18 deletions(-)
tools/perf/tests/parse-events.c +9 -18
@@ -1430,6 +1430,11 @@ static int test__checkevent_config_cache(struct evlist *evlist)return TEST_OK ; } +static bool test__pmu_cpu_valid (void ) +{ +return !!perf_pmu__find ("cpu" ); +} +static bool test__intel_pt_valid (void ) {return !!perf_pmu__find ("intel_pt" );
@@ -1979,21 +1984,25 @@ static const struct evlist_test test__events[] = {static const struct evlist_test test__events_pmu[] = { { .name ="cpu/config=10,config1,config2=3,period=1000/u" , + .valid = test__pmu_cpu_valid, .check = test__checkevent_pmu,/* 0 */ }, { .name ="cpu/config=1,name=krava/u,cpu/config=2/u" , + .valid = test__pmu_cpu_valid, .check = test__checkevent_pmu_name,/* 1 */ }, { .name ="cpu/config=1,call-graph=fp,time,period=100000/,cpu/config=2,call-graph=no,time=0,period=2000/" , + .valid = test__pmu_cpu_valid, .check = test__checkevent_pmu_partial_time_callgraph,/* 2 */ }, { .name ="cpu/name='COMPLEX_CYCLES_NAME:orig=cycles,desc=chip-clock-ticks',period=0x1,event=0x2/ukp" , + .valid = test__pmu_cpu_valid, .check = test__checkevent_complex_name,/* 3 */ },
@@ -2209,21 +2218,6 @@ static int test__terms2(struct test_suite *test __maybe_unused, int subtest __mareturn test_terms (test__terms,ARRAY_SIZE (test__terms)); } -static int test_pmu (void ) -{ -struct stat st; -char path[PATH_MAX ]; -int ret; - -snprintf (path,PATH_MAX ,"%s/bus/event_source/devices/cpu/format/" , -sysfs__mountpoint ()); - - ret =stat (path, &st); -if (ret) -pr_debug ("omitting PMU cpu tests\n" ); -return !ret; -} -static int test__pmu_events (struct test_suite *test__maybe_unused ,int subtest__maybe_unused ) {struct perf_pmu *pmu;
@@ -2307,9 +2301,6 @@ static int test__pmu_events(struct test_suite *test __maybe_unused, int subteststatic int test__pmu_events2 (struct test_suite *test__maybe_unused ,int subtest__maybe_unused ) { -if (!test_pmu ()) -return TEST_SKIP ; -return test_events (test__events_pmu,ARRAY_SIZE (test__events_pmu)); }
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 14/46 perf test: Mask config then test
In reply to Ian RogersAdd helper to test the config of an evsel. Mask the config so that high-bits containing the PMU type, which isn't constant for hybrid, are ignored.
--- tools/perf/tests/parse-events.c | 183 +++++++++++++------------------- 1 file changed, 75 insertions(+), 108 deletions(-)
tools/perf/tests/parse-events.c +75 -108
@@ -21,6 +21,11 @@#define PERF_TP_SAMPLE_TYPE (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | \ PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD) +static bool test_config (const struct evsel *evsel,__u64 expected_config) +{ +return (evsel->core.attr.config &PERF_HW_EVENT_MASK ) == expected_config; +} +#ifdef HAVE_LIBTRACEEVENT #if defined(__s390x__)
@@ -87,7 +92,7 @@ static int test__checkevent_raw(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x1a == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a ));return TEST_OK ; }
@@ -97,7 +102,7 @@ static int test__checkevent_numeric(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,1 == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,1 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,1 ));return TEST_OK ; }
@@ -107,8 +112,7 @@ static int test__checkevent_symbolic_name(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_INSTRUCTIONS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));return TEST_OK ; }
@@ -118,8 +122,7 @@ static int test__checkevent_symbolic_name_config(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));/* * The period value gets configured within evlist__config, * while this test executes only parse events method.
@@ -139,8 +142,7 @@ static int test__checkevent_symbolic_alias(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_SOFTWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_SW_PAGE_FAULTS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_SW_PAGE_FAULTS ));return TEST_OK ; }
@@ -150,7 +152,7 @@ static int test__checkevent_genhw(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HW_CACHE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , (1 <<16 ) == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,1 <<16 ));return TEST_OK ; }
@@ -160,7 +162,7 @@ static int test__checkevent_breakpoint(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_BREAKPOINT == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0 ));TEST_ASSERT_VAL ("wrong bp_type" , (HW_BREAKPOINT_R |HW_BREAKPOINT_W ) == evsel->core.attr.bp_type); TEST_ASSERT_VAL("wrong bp_len", HW_BREAKPOINT_LEN_4 ==
@@ -174,7 +176,7 @@ static int test__checkevent_breakpoint_x(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_BREAKPOINT == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0 ));TEST_ASSERT_VAL ("wrong bp_type" ,HW_BREAKPOINT_X == evsel->core.attr.bp_type);TEST_ASSERT_VAL ("wrong bp_len" ,sizeof (long ) == evsel->core.attr.bp_len);
@@ -188,7 +190,7 @@ static int test__checkevent_breakpoint_r(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_BREAKPOINT == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0 ));TEST_ASSERT_VAL ("wrong bp_type" ,HW_BREAKPOINT_R == evsel->core.attr.bp_type); TEST_ASSERT_VAL("wrong bp_len",
@@ -203,7 +205,7 @@ static int test__checkevent_breakpoint_w(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_BREAKPOINT == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0 ));TEST_ASSERT_VAL ("wrong bp_type" ,HW_BREAKPOINT_W == evsel->core.attr.bp_type); TEST_ASSERT_VAL("wrong bp_len",
@@ -218,7 +220,7 @@ static int test__checkevent_breakpoint_rw(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_BREAKPOINT == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0 ));TEST_ASSERT_VAL ("wrong bp_type" , (HW_BREAKPOINT_R |HW_BREAKPOINT_W ) == evsel->core.attr.bp_type); TEST_ASSERT_VAL("wrong bp_len",
@@ -447,7 +449,7 @@ static int test__checkevent_pmu(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,10 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,10 ));TEST_ASSERT_VAL ("wrong config1" ,1 == evsel->core.attr.config1);TEST_ASSERT_VAL ("wrong config2" ,3 == evsel->core.attr.config2);TEST_ASSERT_VAL ("wrong config3" ,0 == evsel->core.attr.config3);
@@ -469,7 +471,7 @@ static int test__checkevent_list(struct evlist *evlist)/* r1 */ TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,1 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,1 ));TEST_ASSERT_VAL ("wrong config1" ,0 == evsel->core.attr.config1);TEST_ASSERT_VAL ("wrong config2" ,0 == evsel->core.attr.config2);TEST_ASSERT_VAL ("wrong config3" ,0 == evsel->core.attr.config3);
@@ -492,7 +494,7 @@ static int test__checkevent_list(struct evlist *evlist)/* 1:1:hp */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,1 == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,1 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,1 ));TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -509,14 +511,14 @@ static int test__checkevent_pmu_name(struct evlist *evlist)/* cpu/config=1,name=krava/u */ TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,1 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,1 ));TEST_ASSERT_VAL ("wrong name" , !strcmp (evsel__name (evsel),"krava" ));/* cpu/config=2/u" */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,2 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,2 ));TEST_ASSERT_VAL ("wrong name" , !strcmp (evsel__name (evsel),"cpu/config=2/u" ));
@@ -530,7 +532,7 @@ static int test__checkevent_pmu_partial_time_callgraph(struct evlist *evlist)/* cpu/config=1,call-graph=fp,time,period=100000/ */ TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,1 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,1 ));/* * The period, time and callgraph value gets configured within evlist__config, * while this test executes only parse events method.
@@ -542,7 +544,7 @@ static int test__checkevent_pmu_partial_time_callgraph(struct evlist *evlist)/* cpu/config=2,call-graph=no,time=0,period=2000/ */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,2 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,2 ));/* * The period, time and callgraph value gets configured within evlist__config, * while this test executes only parse events method.
@@ -694,8 +696,7 @@ static int test__group1(struct evlist *evlist)/* instructions:k */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_INSTRUCTIONS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -710,8 +711,7 @@ static int test__group1(struct evlist *evlist)/* cycles:upp */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -736,8 +736,7 @@ static int test__group2(struct evlist *evlist)/* faults + :ku modifier */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_SOFTWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_SW_PAGE_FAULTS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_SW_PAGE_FAULTS ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -752,8 +751,7 @@ static int test__group2(struct evlist *evlist)/* cache-references + :u modifier */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CACHE_REFERENCES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CACHE_REFERENCES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -767,8 +765,7 @@ static int test__group2(struct evlist *evlist)/* cycles:k */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -811,8 +808,7 @@ static int test__group3(struct evlist *evlist __maybe_unused)/* group1 cycles:kppp */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -828,8 +824,7 @@ static int test__group3(struct evlist *evlist __maybe_unused)/* group2 cycles + G modifier */ evsel = leader =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -846,7 +841,7 @@ static int test__group3(struct evlist *evlist __maybe_unused)/* group2 1:3 + G modifier */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,1 == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,3 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,3 ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -860,8 +855,7 @@ static int test__group3(struct evlist *evlist __maybe_unused)/* instructions:u */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_INSTRUCTIONS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -885,8 +879,7 @@ static int test__group4(struct evlist *evlist __maybe_unused)/* cycles:u + p */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -903,8 +896,7 @@ static int test__group4(struct evlist *evlist __maybe_unused)/* instructions:kp + p */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_INSTRUCTIONS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -929,8 +921,7 @@ static int test__group5(struct evlist *evlist __maybe_unused)/* cycles + G */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -946,8 +937,7 @@ static int test__group5(struct evlist *evlist __maybe_unused)/* instructions + G */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_INSTRUCTIONS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -961,8 +951,7 @@ static int test__group5(struct evlist *evlist __maybe_unused)/* cycles:G */ evsel = leader =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -978,8 +967,7 @@ static int test__group5(struct evlist *evlist __maybe_unused)/* instructions:G */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_INSTRUCTIONS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -992,8 +980,7 @@ static int test__group5(struct evlist *evlist __maybe_unused)/* cycles */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -1015,8 +1002,7 @@ static int test__group_gh1(struct evlist *evlist)/* cycles + :H group modifier */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -1031,8 +1017,7 @@ static int test__group_gh1(struct evlist *evlist)/* cache-misses:G + :H group modifier */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CACHE_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CACHE_MISSES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -1055,8 +1040,7 @@ static int test__group_gh2(struct evlist *evlist)/* cycles + :G group modifier */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -1071,8 +1055,7 @@ static int test__group_gh2(struct evlist *evlist)/* cache-misses:H + :G group modifier */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CACHE_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CACHE_MISSES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -1095,8 +1078,7 @@ static int test__group_gh3(struct evlist *evlist)/* cycles:G + :u group modifier */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -1111,8 +1093,7 @@ static int test__group_gh3(struct evlist *evlist)/* cache-misses:H + :u group modifier */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CACHE_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CACHE_MISSES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -1135,8 +1116,7 @@ static int test__group_gh4(struct evlist *evlist)/* cycles:G + :uG group modifier */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -1151,8 +1131,7 @@ static int test__group_gh4(struct evlist *evlist)/* cache-misses:H + :uG group modifier */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CACHE_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CACHE_MISSES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -1174,8 +1153,7 @@ static int test__leader_sample1(struct evlist *evlist)/* cycles - sampling group leader */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -1189,8 +1167,7 @@ static int test__leader_sample1(struct evlist *evlist)/* cache-misses - not sampling */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CACHE_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CACHE_MISSES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -1203,8 +1180,7 @@ static int test__leader_sample1(struct evlist *evlist)/* branch-misses - not sampling */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_BRANCH_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_BRANCH_MISSES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv);
@@ -1227,8 +1203,7 @@ static int test__leader_sample2(struct evlist *evlist __maybe_unused)/* instructions - sampling group leader */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_INSTRUCTIONS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -1242,8 +1217,7 @@ static int test__leader_sample2(struct evlist *evlist __maybe_unused)/* branch-misses - not sampling */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_BRANCH_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_BRANCH_MISSES ));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv);
@@ -1279,8 +1253,7 @@ static int test__pinned_group(struct evlist *evlist)/* cycles - group leader */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong group name" , !evsel->group_name);TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));TEST_ASSERT_VAL ("wrong pinned" , evsel->core.attr.pinned);
@@ -1288,14 +1261,12 @@ static int test__pinned_group(struct evlist *evlist)/* cache-misses - can not be pinned, but will go on with the leader */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CACHE_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CACHE_MISSES ));TEST_ASSERT_VAL ("wrong pinned" , !evsel->core.attr.pinned);/* branch-misses - ditto */ evsel =evsel__next (evsel); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_BRANCH_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_BRANCH_MISSES ));TEST_ASSERT_VAL ("wrong pinned" , !evsel->core.attr.pinned);return TEST_OK ;
@@ -1323,8 +1294,7 @@ static int test__exclusive_group(struct evlist *evlist)/* cycles - group leader */ evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CPU_CYCLES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong group name" , !evsel->group_name);TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));TEST_ASSERT_VAL ("wrong exclusive" , evsel->core.attr.exclusive);
@@ -1332,14 +1302,12 @@ static int test__exclusive_group(struct evlist *evlist)/* cache-misses - can not be pinned, but will go on with the leader */ evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_CACHE_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CACHE_MISSES ));TEST_ASSERT_VAL ("wrong exclusive" , !evsel->core.attr.exclusive);/* branch-misses - ditto */ evsel =evsel__next (evsel); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_HW_BRANCH_MISSES == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_BRANCH_MISSES ));TEST_ASSERT_VAL ("wrong exclusive" , !evsel->core.attr.exclusive);return TEST_OK ;
@@ -1350,7 +1318,7 @@ static int test__checkevent_breakpoint_len(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_BREAKPOINT == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0 ));TEST_ASSERT_VAL ("wrong bp_type" , (HW_BREAKPOINT_R |HW_BREAKPOINT_W ) == evsel->core.attr.bp_type); TEST_ASSERT_VAL("wrong bp_len", HW_BREAKPOINT_LEN_1 ==
@@ -1365,7 +1333,7 @@ static int test__checkevent_breakpoint_len_w(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_BREAKPOINT == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0 ));TEST_ASSERT_VAL ("wrong bp_type" ,HW_BREAKPOINT_W == evsel->core.attr.bp_type); TEST_ASSERT_VAL("wrong bp_len", HW_BREAKPOINT_LEN_2 ==
@@ -1393,8 +1361,7 @@ static int test__checkevent_precise_max_modifier(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_SOFTWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_SW_TASK_CLOCK == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_SW_TASK_CLOCK ));return TEST_OK ; }
@@ -1462,7 +1429,7 @@ static int test__checkevent_raw_pmu(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_SOFTWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x1a == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a ));return TEST_OK ; }
@@ -1471,7 +1438,7 @@ static int test__sym_event_slash(struct evlist *evlist)struct evsel *evsel =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" , evsel->core.attr.type ==PERF_TYPE_HARDWARE ); -TEST_ASSERT_VAL ("wrong config" , evsel->core.attr.config ==PERF_COUNT_HW_CPU_CYCLES ); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);return TEST_OK ; }
@@ -1481,7 +1448,7 @@ static int test__sym_event_dc(struct evlist *evlist)struct evsel *evsel =evlist__first (evlist);TEST_ASSERT_VAL ("wrong type" , evsel->core.attr.type ==PERF_TYPE_HARDWARE ); -TEST_ASSERT_VAL ("wrong config" , evsel->core.attr.config ==PERF_COUNT_HW_CPU_CYCLES ); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user);return TEST_OK ; }
@@ -1548,7 +1515,7 @@ static int test__hybrid_hw_event_with_pmu(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x3c == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c ));return TEST_OK ; }
@@ -1559,12 +1526,12 @@ static int test__hybrid_hw_group_event(struct evlist *evlist) evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x3c == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader)); evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0xc0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0xc0 ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));return TEST_OK ; }
@@ -1580,7 +1547,7 @@ static int test__hybrid_sw_hw_group_event(struct evlist *evlist) evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x3c == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));return TEST_OK ; }
@@ -1592,7 +1559,7 @@ static int test__hybrid_hw_sw_group_event(struct evlist *evlist) evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x3c == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader)); evsel =evsel__next (evsel);
@@ -1608,14 +1575,14 @@ static int test__hybrid_group_modifier1(struct evlist *evlist) evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x3c == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel); evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0xc0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0xc0 ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);
@@ -1629,17 +1596,17 @@ static int test__hybrid_raw1(struct evlist *evlist)if (!perf_pmu__hybrid_mounted ("cpu_atom" )) {TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x1a == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a ));return TEST_OK ; }TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x1a == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a ));/* The type of second event is randome value */ evsel =evsel__next (evsel); -TEST_ASSERT_VAL ("wrong config" ,0x1a == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a ));return TEST_OK ; }
@@ -1649,7 +1616,7 @@ static int test__hybrid_raw2(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x1a == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a ));return TEST_OK ; }
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Re: [PATCH v3 14/46] perf test: Mask config then test
tools/perf/tests/parse-events.c static int test__checkevent_raw(struct evlist *evlist) · line 95
Quoted 6 lines
tools/perf/tests/parse-events.c +1
-1@@ -87,7 +92,7 @@ static int test__checkevent_raw(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x1a == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a ));
Extended 'type' via 'config' is applicable only to HW and HW_CACHE types. I think we should not apply test_config() for other types. Even if they are ineffective in these tests (because we are not setting upper bits), they are semantically wrong IMO.
In fact, AMD EventSelect is 12 bits and upper 4 bits are located at 35:32 in PerfEvtSel register. So for RAW events, upper bits MUST NOT be ignored.
tools/perf/tests/parse-events.c static int test__checkevent_symbolic_alias(struct evlist *evlist) · line 145
Quoted 7 lines
tools/perf/tests/parse-events.c +1
-2@@ -139,8 +142,7 @@ static int test__checkevent_symbolic_alias(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_SOFTWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_SW_PAGE_FAULTS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_SW_PAGE_FAULTS ));
Ditto for PERF_TYPE_SOFTWARE. We should not use test_config().
tools/perf/tests/parse-events.c static int test__checkevent_breakpoint(struct evlist *evlist) · line 165
Quoted 6 lines
tools/perf/tests/parse-events.c +1
-1@@ -160,7 +162,7 @@ static int test__checkevent_breakpoint(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_BREAKPOINT == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0 ));
Ditto for PERF_TYPE_BREAKPOINT.
Ian Rogers
reviewerRe: [PATCH v3 14/46] perf test: Mask config then test
Quoted 13 lines · Ravi Bangoria
tools/perf/tests/parse-events.c static int test__checkevent_raw(struct evlist *evlist) · line 95
Quoted 7 lines
tools/perf/tests/parse-events.c +1
-1@@ -87,7 +92,7 @@ static int test__checkevent_raw(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0x1a == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a ));Extended 'type' via 'config' is applicable only to HW and
HW_CACHEtypes. I think we should not applytest_config()for other types. Even if they are ineffective in these tests (because we are not setting upper bits), they are semantically wrong IMO.
Quoted 2 lines
In fact, AMD EventSelect is 12 bits and upper 4 bits are located at 35:32 in PerfEvtSel register. So for RAW events, upper bits MUST NOT be ignored.
Thanks Ravi, this is important and I'll update the tests as you suggest. I'd been iterating on this, which is why the test is this way. We have two parsing cases, one where the type is explicitly given like:
$ perf stat -e '4:0xC0' true and another where the raw config is wildcard applied like: $ perf stat -e 'rC0' true The question is which PMUs should the 0xC0 raw event be opened on? The answer in the code is to ignore all PMUs that aren't the "core" PMUs, so cpu, cpu_atom, cpu_core, armv8... This matches existing behavior but it wasn't clear to me that this was intentional. I think in the future the wildcard application can be made more efficient to ignore unnecessary PMUs and avoid the sysfs scan. Something that is a bit of an issue in the event parsing is what exactly should be applied by wildcard matching and there are inconsistencies.
Thanks,
Ian
Quoted 19 lines
tools/perf/tests/parse-events.c static int test__checkevent_symbolic_alias(struct evlist *evlist) · line 145
Quoted 7 lines
tools/perf/tests/parse-events.c +1
-2@@ -139,8 +142,7 @@ static int test__checkevent_symbolic_alias(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_SOFTWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" , -PERF_COUNT_SW_PAGE_FAULTS == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_SW_PAGE_FAULTS ));Ditto for
PERF_TYPE_SOFTWARE. We should not usetest_config().tools/perf/tests/parse-events.c static int test__checkevent_breakpoint(struct evlist *evlist) · line 165
Quoted 6 lines
tools/perf/tests/parse-events.c +1
-1@@ -160,7 +162,7 @@ static int test__checkevent_breakpoint(struct evlist *evlist)TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_BREAKPOINT == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,0 == evsel->core.attr.config); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0 ));Ditto for
PERF_TYPE_BREAKPOINT.
Ian Rogers
reviewerPATCH v3 15/46 perf test: Test more with config_cache
In reply to Ian Rogerstest__checkevent_config_cache checks the parsing of "L1-dcache-misses/name=cachepmu/". Don't just check that the name is set correctly, also validate the rest of the perf_event_attr for L1-dcache-misses.
--- tools/perf/tests/parse-events.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/tests/parse-events.c +1 -1
@@ -1394,7 +1394,7 @@ static int test__checkevent_config_cache(struct evlist *evlist)struct evsel *evsel =evlist__first (evlist);TEST_ASSERT_VAL ("wrong name setting" ,evsel__name_is (evsel,"cachepmu" )); -return TEST_OK ; +return test__checkevent_genhw (evlist); }static bool test__pmu_cpu_valid (void )
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 16/46 perf test: Roundtrip name, don't assume 1 event per name
In reply to Ian RogersOpening hardware names and a legacy cache event on a hybrid PMU opens it on each PMU. Parsing and checking indexes fails, as the parsed index is double the expected. Avoid checking the index by just comparing the names immediately after the parse.
This change removes hard coded hybrid logic and removes assumptions about the expansion of an event. On hybrid the PMUs may or may not support an event and so using a distance isn't a consistent solution.
--- tools/perf/tests/evsel-roundtrip-name.c | 119 ++++++++++-------------- 1 file changed, 49 insertions(+), 70 deletions(-)
tools/perf/tests/evsel-roundtrip-name.c +49 -70
@@ -4,114 +4,93 @@#include "parse-events.h" #include "tests.h" #include "debug.h" -#include "pmu.h" -#include "pmu-hybrid.h" -#include <errno.h> #include <linux/kernel.h> static int perf_evsel__roundtrip_cache_name_test (void ) { -char name[128 ]; -int type, op, err =0 , ret =0 , i, idx; -struct evsel *evsel; -struct evlist *evlist =evlist__new (); +int ret =TEST_OK ; -if (evlist ==NULL ) -return -ENOMEM ; - -for (type =0 ; type <PERF_COUNT_HW_CACHE_MAX ; type++) { -for (op =0 ; op <PERF_COUNT_HW_CACHE_OP_MAX ; op++) { +for (int type =0 ; type <PERF_COUNT_HW_CACHE_MAX ; type++) { +for (int op =0 ; op <PERF_COUNT_HW_CACHE_OP_MAX ; op++) {/* skip invalid cache type */ if (!evsel__is_cache_op_valid (type, op))continue ; -for (i =0 ; i <PERF_COUNT_HW_CACHE_RESULT_MAX ; i++) { -__evsel__hw_cache_type_op_res_name (type, op, i, name,sizeof (name)); - err =parse_event (evlist, name); -if (err) - ret = err; - } - } - } - - idx =0 ; - evsel =evlist__first (evlist); +for (int res =0 ; res <PERF_COUNT_HW_CACHE_RESULT_MAX ; res++) { +char name[128 ]; +struct evlist *evlist =evlist__new (); +struct evsel *evsel; +int err; -for (type =0 ; type <PERF_COUNT_HW_CACHE_MAX ; type++) { -for (op =0 ; op <PERF_COUNT_HW_CACHE_OP_MAX ; op++) { -/* skip invalid cache type */ -if (!evsel__is_cache_op_valid (type, op)) -continue ; +if (evlist ==NULL ) { +pr_debug ("Failed to alloc evlist" ); +return TEST_FAIL ; + } +__evsel__hw_cache_type_op_res_name (type, op, res, + name,sizeof (name)); -for (i =0 ; i <PERF_COUNT_HW_CACHE_RESULT_MAX ; i++) { -__evsel__hw_cache_type_op_res_name (type, op, i, name,sizeof (name)); -if (evsel->core.idx != idx) + err =parse_event (evlist, name); +if (err) { +pr_debug ("Failure to parse cache event '%s' possibly as PMUs don't support it" , + name); +evlist__delete (evlist);continue ; - - ++idx; - -if (strcmp (evsel__name (evsel), name)) { -pr_debug ("%s != %s\n" ,evsel__name (evsel), name); - ret =-1 ; } - - evsel =evsel__next (evsel); +evlist__for_each_entry (evlist, evsel) { +if (strcmp (evsel__name (evsel), name)) { +pr_debug ("%s != %s\n" ,evsel__name (evsel), name); + ret =TEST_FAIL ; + } + } +evlist__delete (evlist); } } } - -evlist__delete (evlist);return ret; } -static int __perf_evsel__name_array_test (const char *const names[],int nr_names, -int distance) +static int perf_evsel__name_array_test (const char *const names[],int nr_names) { -int i, err; -struct evsel *evsel; -struct evlist *evlist =evlist__new (); +int ret =TEST_OK ; -if (evlist ==NULL ) -return -ENOMEM ; +for (int i =0 ; i < nr_names; ++i) { +struct evlist *evlist =evlist__new (); +struct evsel *evsel; +int err; -for (i =0 ; i < nr_names; ++i) { +if (evlist ==NULL ) { +pr_debug ("Failed to alloc evlist" ); +return TEST_FAIL ; + } err =parse_event (evlist, names[i]);if (err) {pr_debug ("failed to parse event '%s', err %d\n" , names[i], err); - goto out_delete_evlist; +evlist__delete (evlist); + ret =TEST_FAIL ; +continue ; } - } - - err =0 ; -evlist__for_each_entry (evlist, evsel) { -if (strcmp (evsel__name (evsel), names[evsel->core.idx / distance])) { - --err; -pr_debug ("%s != %s\n" ,evsel__name (evsel), names[evsel->core.idx / distance]); +evlist__for_each_entry (evlist, evsel) { +if (strcmp (evsel__name (evsel), names[i])) { +pr_debug ("%s != %s\n" ,evsel__name (evsel), names[i]); + ret =TEST_FAIL ; + } } +evlist__delete (evlist); } - -out_delete_evlist: -evlist__delete (evlist); -return err; +return ret; } -#define perf_evsel__name_array_test (names, distance) \ - __perf_evsel__name_array_test(names, ARRAY_SIZE(names), distance) -static int test__perf_evsel__roundtrip_name_test (struct test_suite *test__maybe_unused ,int subtest__maybe_unused ) { -int err =0 , ret =0 ; - -if (perf_pmu__has_hybrid () &&perf_pmu__hybrid_mounted ("cpu_atom" )) -return perf_evsel__name_array_test (evsel__hw_names,2 ); +int err =0 , ret =TEST_OK ; - err =perf_evsel__name_array_test (evsel__hw_names,1 ); + err =perf_evsel__name_array_test (evsel__hw_names,PERF_COUNT_HW_MAX );if (err) ret = err; - err =__perf_evsel__name_array_test (evsel__sw_names,PERF_COUNT_SW_DUMMY +1 ,1 ); + err =perf_evsel__name_array_test (evsel__sw_names,PERF_COUNT_SW_DUMMY +1 );if (err) ret = err;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 17/46 perf parse-events: Set attr.type to PMU type early
In reply to Ian RogersSet attr.type to PMU type early so that later terms can override the value. Setting the value in perf_pmu__config means that earlier steps, like config_term_pmu, can override the value.
--- tools/perf/util/parse-events.c | 2 +- tools/perf/util/pmu.c | 1 - 2 files changed, 1 insertion(+), 2 deletions(-)
tools/perf/util/parse-events.c +1 -1
@@ -1492,9 +1492,9 @@ int parse_events_add_pmu(struct parse_events_state *parse_state, }else {memset (&attr,0 ,sizeof (attr)); } + attr.type = pmu->type;if (!head_config) { - attr.type = pmu->type; evsel = __add_event(list, &parse_state->idx, &attr, /*init_attr=*/true, /*name=*/NULL, /*metric_id=*/NULL, pmu,
tools/perf/util/pmu.c +0 -1
@@ -1398,7 +1398,6 @@ int perf_pmu__config(struct perf_pmu *pmu, struct perf_event_attr *attr, {bool zero = !!pmu->default_config; - attr->type = pmu->type;return perf_pmu__config_terms (pmu->name, &pmu->format, attr, head_terms, zero, err); }
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 18/46 perf parse-events: Set pmu_name whenever a pmu is given
In reply to Ian RogersChange add_event to always set pmu_name when possible as not all code checks both pmu->name and evsel->pmu_name, for example, uniquify_counter in stat-display.c.
--- tools/perf/util/parse-events.c | 9 ++------- 1 file changed, 2 insertions(+), 7 deletions(-)
tools/perf/util/parse-events.c +2 -7
@@ -269,6 +269,7 @@ __add_event(struct list_head *list, int *idx, evsel->core.requires_cpu = pmu ? pmu->is_uncore : false; evsel->auto_merge_stats = auto_merge_stats; evsel->pmu = pmu; + evsel->pmu_name = pmu && pmu->name ?strdup (pmu->name) :NULL ;if (name) evsel->name =strdup (name);
@@ -1500,12 +1501,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,/*metric_id=*/ NULL , pmu,/*config_terms=*/ NULL , auto_merge_stats, /*cpu_list=*/NULL); -if (evsel) { - evsel->pmu_name = name ?strdup (name) :NULL ; -return 0 ; - }else { -return -ENOMEM ; - } +return evsel ?0 : -ENOMEM ; }if (!parse_state->fake_pmu &&perf_pmu__check_alias (pmu, head_config, &info))
@@ -1561,7 +1557,6 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,if (evsel->name) evsel->use_config_name = true; - evsel->pmu_name = name ?strdup (name) :NULL ; evsel->percore =config_term_percore (&evsel->config_terms);if (parse_state->fake_pmu)
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 19/46 perf print-events: Avoid unnecessary strlist
In reply to Ian RogersThe strlist in print_hwcache_events holds the event names as they are generated, and then it is iterated and printed. This is unnecessary and each event can just be printed as it is processed. Rename the variable i to res, to be more intention revealing and consistent with other code.
--- tools/perf/util/print-events.c | 60 ++++++++++++++++++---------------- 1 file changed, 31 insertions(+), 29 deletions(-)
tools/perf/util/print-events.c +31 -29
@@ -226,58 +226,60 @@ void print_sdt_events(const struct print_callbacks *print_cb, void *print_state)int print_hwcache_events (const struct print_callbacks *print_cb,void *print_state) { -struct strlist *evt_name_list =strlist__new (NULL ,NULL ); -struct str_node *nd; +const char *event_type_descriptor = event_type_descriptors[PERF_TYPE_HW_CACHE ]; -if (!evt_name_list) { -pr_debug ("Failed to allocate new strlist for hwcache events\n" ); -return -ENOMEM ; - }for (int type =0 ; type <PERF_COUNT_HW_CACHE_MAX ; type++) {for (int op =0 ; op <PERF_COUNT_HW_CACHE_OP_MAX ; op++) {/* skip invalid cache type */ if (!evsel__is_cache_op_valid (type, op))continue ; -for (int i =0 ; i <PERF_COUNT_HW_CACHE_RESULT_MAX ; i++) { +for (int res =0 ; res <PERF_COUNT_HW_CACHE_RESULT_MAX ; res++) {struct perf_pmu *pmu =NULL ;char name[64 ]; -__evsel__hw_cache_type_op_res_name (type, op, i, name,sizeof (name)); +__evsel__hw_cache_type_op_res_name (type, op, res, + name,sizeof (name));if (!perf_pmu__has_hybrid ()) {if (is_event_supported (PERF_TYPE_HW_CACHE , - type | (op <<8 ) | (i <<16 ))) -strlist__add (evt_name_list, name); + type | (op <<8 ) | (res <<16 ))) { + print_cb->print_event (print_state, +"cache" , +/*pmu_name=*/ NULL , + name, +/*event_alias=*/ NULL , +/*scale_unit=*/ NULL , +/*deprecated=*/ false, + event_type_descriptor, +/*desc=*/ NULL , +/*long_desc=*/ NULL , +/*encoding_desc=*/ NULL ); + }continue ; }perf_pmu__for_each_hybrid_pmu (pmu) {if (is_event_supported (PERF_TYPE_HW_CACHE , - type | (op <<8 ) | (i <<16 ) | + type | (op <<8 ) | (res <<16 ) | ((__u64 )pmu->type <<PERF_PMU_TYPE_SHIFT ))) {char new_name[128 ]; -snprintf (new_name,sizeof (new_name), -"%s/%s/" , pmu->name, name); -strlist__add (evt_name_list, new_name); +snprintf (new_name,sizeof (new_name), +"%s/%s/" , pmu->name, name); + print_cb->print_event (print_state, +"cache" , + pmu->name, + name, + new_name, +/*scale_unit=*/ NULL , +/*deprecated=*/ false, + event_type_descriptor, +/*desc=*/ NULL , +/*long_desc=*/ NULL , +/*encoding_desc=*/ NULL ); } } } } } - -strlist__for_each_entry (nd, evt_name_list) { - print_cb->print_event (print_state, -"cache" , -/*pmu_name=*/ NULL , - nd->s, -/*event_alias=*/ NULL , -/*scale_unit=*/ NULL , -/*deprecated=*/ false, - event_type_descriptors[PERF_TYPE_HW_CACHE ], -/*desc=*/ NULL , -/*long_desc=*/ NULL , -/*encoding_desc=*/ NULL ); - } -strlist__delete (evt_name_list);return 0 ; }
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 20/46 perf parse-events: Avoid scanning PMUs before parsing
In reply to Ian RogersThe event parser needs to handle two special cases:
1) legacy events like L1-dcache-load-miss. These event names don't appear in json or sysfs, and lookup tables are used for the config value.
2) raw events where 'r0xead' is the same as 'read' unless the PMU has an event called 'read' in which case the event has priority.
The previous parser to handle these cases would scan all PMUs for components of event names. These components would then be used to classify in the lexer whether the token should be part of a legacy event, a raw event or an event. The grammar would handle legacy event tokens or recombining the tokens back into a regular event name. The code wasn't PMU specific and had issues around events like AMD's branch-brs that would fail to parse as it expects brs to be a suffix on a legacy event style name:
$ perf stat -e branch-brs true
event syntax error: 'branch-brs'
\___ parser errorThis change removes processing all PMUs by using the lexer in the form of a regular expression matcher. The lexer will return the token for the longest matched sequence of characters, and in the event of a tie the first. The legacy events are a fixed number of regular expressions, and by matching these before a name token its possible to generate an accurate legacy event token with everything else matching as a name. Because of the lexer change the handling of hyphens in the grammar can be removed as hyphens just become a part of the name.
To handle raw events and terms the parser is changed to defer trying to evaluate whether something is a raw event until the PMU is known in the grammar. Once the PMU is known, the events of the PMU can be scanned for the 'read' style problem. A new term type is added for these raw terms, used to enable deferring the evaluation.
While this change is large, it has stats of:
170 insertions(+), 436 deletions(-)
the bulk of the change is deleting the old approach. It isn't possible to break apart the code added due to the dependencies on how the parts of the parsing work.
--- tools/perf/tests/parse-events.c | 24 +-- tools/perf/tests/pmu-events.c | 9 - tools/perf/util/parse-events.c | 329 ++++++++++---------------------- tools/perf/util/parse-events.h | 16 +- tools/perf/util/parse-events.l | 85 +-------- tools/perf/util/parse-events.y | 143 +++++--------- 6 files changed, 170 insertions(+), 436 deletions(-)
tools/perf/tests/parse-events.c +8 -16
@@ -664,11 +664,11 @@ static int test__checkterms_simple(struct list_head *terms)*/ term = list_entry(term->list.next, struct parse_events_term, list);TEST_ASSERT_VAL ("wrong type term" , - term->type_term ==PARSE_EVENTS__TERM_TYPE_USER ); + term->type_term ==PARSE_EVENTS__TERM_TYPE_RAW );TEST_ASSERT_VAL ("wrong type val" , - term->type_val ==PARSE_EVENTS__TERM_TYPE_NUM ); -TEST_ASSERT_VAL ("wrong val" , term->val.num ==1 ); -TEST_ASSERT_VAL ("wrong config" , !strcmp (term->config,"read" )); + term->type_val ==PARSE_EVENTS__TERM_TYPE_STR ); +TEST_ASSERT_VAL ("wrong val" , !strcmp (term->val.str,"read" )); +TEST_ASSERT_VAL ("wrong config" , !strcmp (term->config,"raw" ));/* * r0xead
@@ -678,11 +678,11 @@ static int test__checkterms_simple(struct list_head *terms)*/ term = list_entry(term->list.next, struct parse_events_term, list);TEST_ASSERT_VAL ("wrong type term" , - term->type_term ==PARSE_EVENTS__TERM_TYPE_CONFIG ); + term->type_term ==PARSE_EVENTS__TERM_TYPE_RAW );TEST_ASSERT_VAL ("wrong type val" , - term->type_val ==PARSE_EVENTS__TERM_TYPE_NUM ); -TEST_ASSERT_VAL ("wrong val" , term->val.num ==0xead ); -TEST_ASSERT_VAL ("wrong config" , !strcmp (term->config,"config" )); + term->type_val ==PARSE_EVENTS__TERM_TYPE_STR ); +TEST_ASSERT_VAL ("wrong val" , !strcmp (term->val.str,"r0xead" )); +TEST_ASSERT_VAL ("wrong config" , !strcmp (term->config,"raw" ));return TEST_OK ; }
@@ -2090,7 +2090,6 @@ static int test_event_fake_pmu(const char *str)return -ENOMEM ;parse_events_error__init (&err); -perf_pmu__test_parse_init (); ret =__parse_events (evlist, str, &err, &perf_pmu__fake,/*warn_if_reordered=*/ true);if (ret) { pr_debug("failed to parse event '%s', err %d, str '%s'\n",
@@ -2144,13 +2143,6 @@ static int test_term(const struct terms_test *t)INIT_LIST_HEAD (&terms); -/* -* The perf_pmu__test_parse_init prepares perf_pmu_events_list -* which gets freed in parse_events_terms. -*/ -if (perf_pmu__test_parse_init ()) -return -1 ; - ret =parse_events_terms (&terms, t->str);if (ret) { pr_debug("failed to parse terms '%s', err %d\n",
tools/perf/tests/pmu-events.c +0 -9
@@ -776,15 +776,6 @@ static int check_parse_id(const char *id, struct parse_events_error *error,for (cur =strchr (dup,'@' ) ; cur; cur =strchr (++cur,'@' )) *cur ='/' ; -if (fake_pmu) { -/* -* Every call to __parse_events will try to initialize the PMU -* state from sysfs and then clean it up at the end. Reset the -* PMU events to the test state so that we don't pick up -* erroneous prefixes and suffixes. -*/ -perf_pmu__test_parse_init (); - } ret =__parse_events (evlist, dup, error, fake_pmu,/*warn_if_reordered=*/ true);free (dup);
tools/perf/util/parse-events.c +105 -224
426 lines not on this page · raw message
tools/perf/util/parse-events.h +2 -14
@@ -41,14 +41,6 @@ int parse_events_terms(struct list_head *terms, const char *str);int parse_filter (const struct option *opt,const char *str,int unset);int exclude_perf (const struct option *opt,const char *arg,int unset); -enum perf_pmu_event_symbol_type { -PMU_EVENT_SYMBOL_ERR ,/* not a PMU EVENT */ -PMU_EVENT_SYMBOL ,/* normal style PMU event */ -PMU_EVENT_SYMBOL_PREFIX ,/* prefix of pre-suf style event */ -PMU_EVENT_SYMBOL_SUFFIX ,/* suffix of pre-suf style event */ -PMU_EVENT_SYMBOL_SUFFIX2 ,/* suffix of pre-suf2 style event */ -}; -enum {PARSE_EVENTS__TERM_TYPE_NUM ,PARSE_EVENTS__TERM_TYPE_STR ,
@@ -78,6 +70,7 @@ enum {PARSE_EVENTS__TERM_TYPE_AUX_OUTPUT ,PARSE_EVENTS__TERM_TYPE_AUX_SAMPLE_SIZE ,PARSE_EVENTS__TERM_TYPE_METRIC_ID , +PARSE_EVENTS__TERM_TYPE_RAW , __PARSE_EVENTS__TERM_TYPE_NR, };
@@ -174,8 +167,7 @@ int parse_events_add_numeric(struct parse_events_state *parse_state,int parse_events_add_tool (struct parse_events_state *parse_state,struct list_head *list,int tool_event); -int parse_events_add_cache (struct list_head *list,int *idx, -char *type,char *op_result1,char *op_result2, +int parse_events_add_cache (struct list_head *list,int *idx,const char *name,struct parse_events_error *error,struct list_head *head_config,struct parse_events_state *parse_state);
@@ -198,8 +190,6 @@ int parse_events_multi_pmu_add(struct parse_events_state *parse_state,int parse_events_copy_term_list (struct list_head *old,struct list_head **new); -enum perf_pmu_event_symbol_type -perf_pmu__parse_check (const char *name);void parse_events__set_leader (char *name,struct list_head *list);void parse_events_update_lists (struct list_head *list_event,struct list_head *list_all);
@@ -241,8 +231,6 @@ static inline bool is_sdt_event(char *str __maybe_unused) }#endif /* HAVE_LIBELF_SUPPORT */ -int perf_pmu__test_parse_init (void ); - struct evsel *parse_events__add_event_hybrid(struct list_head *list, int *idx, struct perf_event_attr *attr, const char *name,
tools/perf/util/parse-events.l +10 -75
130 lines not on this page · raw message
tools/perf/util/parse-events.y +45 -98
285 lines not on this page · raw message
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 21/46 perf evsel: Modify group pmu name for software events
In reply to Ian RogersIf we have a group of {cycles,faults} then we need the faults software event to appear to be on the same PMU as cycles so that we don't split the group in parse_events__sort_events_and_fix_groups. This case is relatively easy as cycles is the leader and will have a PMU name. In the reverse case, {faults,cycles} we still need faults to appear to have the PMU name of cycles but the old behavior is just to return "cpu". For hybrid this fails as cycles will be on "cpu_core" or "cpu_atom", causing faults to be split into a different group.
Change the behavior for software events so that the whole group is searched for the named PMU.
--- tools/perf/util/evsel.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-)
tools/perf/util/evsel.c +9 -6
@@ -829,23 +829,26 @@ bool evsel__name_is(struct evsel *evsel, const char *name)const char *evsel__group_pmu_name (const struct evsel *evsel) { -const struct evsel *leader; +struct evsel *leader, *pos;/* If the pmu_name is set use it. pmu_name isn't set for CPU and software events. */ if (evsel->pmu_name)return evsel->pmu_name;/* * Software events may be in a group with other uncore PMU events. Use -* the pmu_name of the group leader to avoid breaking the software event -* out of the group. +* the pmu_name of the first non-software event to avoid breaking the +* software event out of the group. * * Aux event leaders, like intel_pt, expect a group with events from * other PMUs, so substitute the AUX event's PMU in this case. */ leader =evsel__leader (evsel); -if ((evsel->core.attr.type ==PERF_TYPE_SOFTWARE ||evsel__is_aux_event (leader)) && - leader->pmu_name) { -return leader->pmu_name; +if (evsel->core.attr.type ==PERF_TYPE_SOFTWARE ||evsel__is_aux_event (leader)) { +/* Starting with the leader, find the first event with a named PMU. */ +for_each_group_evsel (pos, leader) { +if (pos->pmu_name) +return pos->pmu_name; + } }return "cpu" ;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 22/46 perf test: Move x86 hybrid tests to arch/x86
In reply to Ian RogersThe tests use x86 hybrid specific PMUs.
--- tools/perf/arch/x86/include/arch-tests.h | 1 + tools/perf/arch/x86/tests/Build | 1 + tools/perf/arch/x86/tests/arch-tests.c | 10 + tools/perf/arch/x86/tests/hybrid.c | 277 +++++++++++++++++++++++ tools/perf/tests/parse-events.c | 181 --------------- 5 files changed, 289 insertions(+), 181 deletions(-) create mode 100644 tools/perf/arch/x86/tests/hybrid.c
tools/perf/arch/x86/include/arch-tests.h +1 -0
@@ -11,6 +11,7 @@ int test__intel_pt_pkt_decoder(struct test_suite *test, int subtest);int test__intel_pt_hybrid_compat (struct test_suite *test,int subtest);int test__bp_modify (struct test_suite *test,int subtest);int test__x86_sample_parsing (struct test_suite *test,int subtest); +int test__hybrid (struct test_suite *test,int subtest);extern struct test_suite *arch_tests[];
tools/perf/arch/x86/tests/Build +1 -0
@@ -3,5 +3,6 @@ perf-$(CONFIG_DWARF_UNWIND) += dwarf-unwind.o perf-y += arch-tests.o perf-y += sample-parsing.o +perf-y += hybrid.o perf-$(CONFIG_AUXTRACE) += insn-x86.o intel-pt-test.o perf-$(CONFIG_X86_64) += bp-modify.o
tools/perf/arch/x86/tests/arch-tests.c +10 -0
@@ -22,6 +22,15 @@ struct test_suite suite__intel_pt = {DEFINE_SUITE ("x86 bp modify" , bp_modify);#endif DEFINE_SUITE ("x86 Sample parsing" , x86_sample_parsing); +static struct test_case hybrid_tests[] = { +TEST_CASE_REASON ("x86 hybrid event parsing" , hybrid,"not hybrid" ), + { .name =NULL , } +}; + +struct test_suite suite__hybrid = { + .desc ="x86 hybrid" , + .test_cases = hybrid_tests, +}; struct test_suite *arch_tests[] = {#ifdef HAVE_DWARF_UNWIND_SUPPORT
@@ -35,5 +44,6 @@ struct test_suite *arch_tests[] = { &suite__bp_modify, #endif &suite__x86_sample_parsing, + &suite__hybrid, NULL, };
tools/perf/arch/x86/tests/hybrid.c new file +277 -0
277 lines not on this page · raw message
tools/perf/tests/parse-events.c +0 -181
211 lines not on this page · raw message
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 23/46 perf test x86 hybrid: Update test expectations
In reply to Ian RogersDon't assume evlist order. Switch to a loop rather than depend on evlist order for raw events test.
Update hybrid event expectations. Previous values were based on parsing legacy hardware events from sysfs, update to the correct PMU specific legacy values.
--- tools/perf/arch/x86/tests/hybrid.c | 54 ++++++++++++++---------------- 1 file changed, 26 insertions(+), 28 deletions(-)
tools/perf/arch/x86/tests/hybrid.c +26 -28
@@ -11,13 +11,18 @@ static bool test_config(const struct evsel *evsel, __u64 expected_config)return (evsel->core.attr.config &PERF_HW_EVENT_MASK ) == expected_config; } +static bool test_perf_config (const struct perf_evsel *evsel,__u64 expected_config) +{ +return (evsel->attr.config &PERF_HW_EVENT_MASK ) == expected_config; +} +static int test__hybrid_hw_event_with_pmu (struct evlist *evlist) {struct evsel *evsel =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c )); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));return TEST_OK ; }
@@ -27,13 +32,13 @@ static int test__hybrid_hw_group_event(struct evlist *evlist) evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c )); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader)); evsel =evsel__next (evsel); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0xc0 )); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));return TEST_OK ; }
@@ -48,8 +53,8 @@ static int test__hybrid_sw_hw_group_event(struct evlist *evlist)TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader)); evsel =evsel__next (evsel); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c )); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));return TEST_OK ; }
@@ -60,8 +65,8 @@ static int test__hybrid_hw_sw_group_event(struct evlist *evlist) evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c )); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader)); evsel =evsel__next (evsel);
@@ -76,15 +81,15 @@ static int test__hybrid_group_modifier1(struct evlist *evlist) evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x3c )); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel); evsel =evsel__next (evsel); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0xc0 )); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel);
@@ -93,22 +98,15 @@ static int test__hybrid_group_modifier1(struct evlist *evlist)static int test__hybrid_raw1 (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); +struct perf_evsel *evsel; -if (!perf_pmu__hybrid_mounted ("cpu_atom" )) { -TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a )); -return TEST_OK ; - } - -TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a )); +perf_evlist__for_each_evsel (&evlist->core, evsel) { +struct perf_pmu *pmu =perf_pmu__find_by_type (evsel->attr.type); -/* The type of second event is randome value */ - evsel =evsel__next (evsel); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a )); +TEST_ASSERT_VAL ("missing pmu" , pmu); +TEST_ASSERT_VAL ("unexpected pmu" , !strncmp (pmu->name,"cpu_" ,4 )); +TEST_ASSERT_VAL ("wrong config" ,test_perf_config (evsel,0x1a )); + }return TEST_OK ; }
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 24/46 perf test x86 hybrid: Add hybrid extended type checks
In reply to Ian RogersAssert hybrid extended types are as expected.
--- tools/perf/arch/x86/tests/hybrid.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
tools/perf/arch/x86/tests/hybrid.c +12 -0
@@ -16,12 +16,18 @@ static bool test_perf_config(const struct perf_evsel *evsel, __u64 expected_confreturn (evsel->attr.config &PERF_HW_EVENT_MASK ) == expected_config; } +static bool test_hybrid_type (const struct evsel *evsel,__u64 expected_config) +{ +return (evsel->core.attr.config >>PERF_PMU_TYPE_SHIFT ) == expected_config; +} +static int test__hybrid_hw_event_with_pmu (struct evlist *evlist) {struct evsel *evsel =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong hybrid type" ,test_hybrid_type (evsel,PERF_TYPE_RAW ));TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));return TEST_OK ; }
@@ -33,11 +39,13 @@ static int test__hybrid_hw_group_event(struct evlist *evlist) evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong hybrid type" ,test_hybrid_type (evsel,PERF_TYPE_RAW ));TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader)); evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong hybrid type" ,test_hybrid_type (evsel,PERF_TYPE_RAW ));TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));return TEST_OK ;
@@ -54,6 +62,7 @@ static int test__hybrid_sw_hw_group_event(struct evlist *evlist) evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong hybrid type" ,test_hybrid_type (evsel,PERF_TYPE_RAW ));TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));return TEST_OK ;
@@ -66,6 +75,7 @@ static int test__hybrid_hw_sw_group_event(struct evlist *evlist) evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong hybrid type" ,test_hybrid_type (evsel,PERF_TYPE_RAW ));TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));
@@ -82,6 +92,7 @@ static int test__hybrid_group_modifier1(struct evlist *evlist) evsel = leader =evlist__first (evlist);TEST_ASSERT_VAL ("wrong number of entries" ,2 == evlist->core.nr_entries);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong hybrid type" ,test_hybrid_type (evsel,PERF_TYPE_RAW ));TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user);
@@ -89,6 +100,7 @@ static int test__hybrid_group_modifier1(struct evlist *evlist) evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); +TEST_ASSERT_VAL ("wrong hybrid type" ,test_hybrid_type (evsel,PERF_TYPE_RAW ));TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS ));TEST_ASSERT_VAL ("wrong leader" ,evsel__has_leader (evsel, leader));TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 25/46 perf parse-events: Support PMUs for legacy cache events
In reply to Ian RogersAllow a legacy cache event to be both, for example, "L1-dcache-load-miss" and "cpu/L1-dcache-load-miss/" by introducing a new legacy cache term type. The term type is processed in config_term_pmu, setting both the type in perf_event_attr and the config. The code to determine the config is factored out of parse_events_add_cache and shared. If the PMU doesn't support legacy events, currently just core/hybrid PMUs do, then the term is treated like a PE_NAME term - as before. If only terms are being parsed, such as for perf_pmu__new_alias, then the PE_LEGACY_CACHE token is always parsed as PE_NAME.
--- tools/perf/tests/parse-events.c | 18 +++++++++ tools/perf/util/parse-events.c | 70 ++++++++++++++++++++++----------- tools/perf/util/parse-events.h | 3 ++ tools/perf/util/parse-events.l | 9 ++++- tools/perf/util/parse-events.y | 14 ++++++- tools/perf/util/pmu.c | 5 +++ tools/perf/util/pmu.h | 1 + 7 files changed, 96 insertions(+), 24 deletions(-)
tools/perf/tests/parse-events.c +18 -0
@@ -1861,6 +1861,24 @@ static const struct evlist_test test__events_pmu[] = { .check = test__checkevent_raw_pmu, /* 5 */ }, + { + .name = "cpu/L1-dcache-load-miss/", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_genhw, + /* 6 */ + }, + { + .name = "cpu/L1-dcache-load-miss/kp", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_genhw_modifier, + /* 7 */ + }, + { + .name = "cpu/L1-dcache-misses,name=cachepmu/", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_config_cache, + /* 8 */ + }, }; struct terms_test {
tools/perf/util/parse-events.c +48 -22
123 lines not on this page · raw message
tools/perf/util/parse-events.h +3 -0
@@ -71,6 +71,7 @@ enum {PARSE_EVENTS__TERM_TYPE_AUX_SAMPLE_SIZE ,PARSE_EVENTS__TERM_TYPE_METRIC_ID ,PARSE_EVENTS__TERM_TYPE_RAW , +PARSE_EVENTS__TERM_TYPE_LEGACY_CACHE , __PARSE_EVENTS__TERM_TYPE_NR, };
@@ -122,6 +123,8 @@ struct parse_events_state {int stoken;struct perf_pmu *fake_pmu;char *hybrid_pmu_name; +/* Should PE_LEGACY_NAME tokens be generated for config terms? */ +bool match_legacy_cache_terms;bool wild_card_pmus; };
tools/perf/util/parse-events.l +8 -1
@@ -63,6 +63,11 @@ static int str(yyscan_t scanner, int token) return token; } +static int lc_str(yyscan_t scanner, const struct parse_events_state *state) +{ + return str(scanner, state->match_legacy_cache_terms ? PE_LEGACY_CACHE : PE_NAME); +} + static bool isbpf_suffix(char *text) { int len = strlen(text);
@@ -185,7 +190,6 @@ lc_op_result (load|loads|read|store|stores|write|prefetch|prefetches|speculative %{ struct parse_events_state *_parse_state = parse_events_get_extra(yyscanner); - { int start_token = _parse_state->stoken;
@@ -269,6 +273,9 @@ r{num_raw_hex} { return str(yyscanner, PE_RAW); } r0x{num_raw_hex} { return str(yyscanner, PE_RAW); } , { return ','; } "/" { BEGIN(INITIAL); return '/'; } +{lc_type} { return lc_str(yyscanner, _parse_state); } +{lc_type}-{lc_op_result} { return lc_str(yyscanner, _parse_state); } +{lc_type}-{lc_op_result}-{lc_op_result} { return lc_str(yyscanner, _parse_state); } {name_minus} { return str(yyscanner, PE_NAME); } \[all\] { return PE_ARRAY_ALL; } "[" { BEGIN(array); return '['; }
tools/perf/util/parse-events.y +13 -1
@@ -723,7 +723,7 @@ event_term $$ = head; } -name_or_raw: PE_RAW | PE_NAME +name_or_raw: PE_RAW | PE_NAME | PE_LEGACY_CACHE event_term: PE_RAW
@@ -775,6 +775,18 @@ name_or_raw '=' PE_VALUE_SYM_HW $$ = term; } | +PE_LEGACY_CACHE +{ + struct parse_events_term *term; + + if (parse_events_term__num(&term, PARSE_EVENTS__TERM_TYPE_LEGACY_CACHE, + $1, 1, true, &@1, NULL)) { + free($1); + YYABORT; + } + $$ = term; +} +| PE_NAME { struct parse_events_term *term;
tools/perf/util/pmu.c +5 -0
@@ -1650,6 +1650,11 @@ bool is_pmu_core(const char *name)return !strcmp (name,"cpu" ) ||is_arm_pmu_core (name); } +bool perf_pmu__supports_legacy_cache (const struct perf_pmu *pmu) +{ +return is_pmu_core (pmu->name) ||perf_pmu__is_hybrid (pmu->name); +} +static bool pmu_alias_is_duplicate (struct sevent *alias_a,struct sevent *alias_b) {
tools/perf/util/pmu.h +1 -0
@@ -220,6 +220,7 @@ void perf_pmu__del_formats(struct list_head *formats);struct perf_pmu *perf_pmu__scan (struct perf_pmu *pmu);bool is_pmu_core (const char *name); +bool perf_pmu__supports_legacy_cache (const struct perf_pmu *pmu);void print_pmu_events (const struct print_callbacks *print_cb,void *print_state);bool pmu_have_event (const char *pname,const char *name);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 26/46 perf parse-events: Wildcard legacy cache events
In reply to Ian RogersIt is inconsistent that "perf stat -e instructions-retired" wildcard opens on all PMUs while legacy cache events like "perf stat -e L1-dcache-load-miss" do not. A behavior introduced by hybrid is that a legacy cache event like L1-dcache-load-miss should wildcard open on all hybrid PMUs. Previously hybrid would call to is_event_supported for each PMU, a failure of which results in the event not being added. This isn't done in this case as the parser should just create perf_event_attr and the later open should fail, or the counter give "<not counted>". If this wants to be avoided then the PMU can be named with the event.
--- tools/perf/util/parse-events-hybrid.c | 33 ------------- tools/perf/util/parse-events-hybrid.h | 7 --- tools/perf/util/parse-events.c | 68 ++++++++++++++------------- tools/perf/util/parse-events.h | 3 +- tools/perf/util/parse-events.y | 2 +- 5 files changed, 37 insertions(+), 76 deletions(-)
tools/perf/util/parse-events-hybrid.c +0 -33
@@ -179,36 +179,3 @@ int parse_events__add_numeric_hybrid(struct parse_events_state *parse_state,return add_raw_hybrid (parse_state, list, attr, name, metric_id, config_terms); } - -int parse_events__add_cache_hybrid (struct list_head *list,int *idx, -struct perf_event_attr *attr, -const char *name, -const char *metric_id, -struct list_head *config_terms, -bool *hybrid, -struct parse_events_state *parse_state) -{ -struct perf_pmu *pmu; -int ret; - - *hybrid = false; -if (!perf_pmu__has_hybrid ()) -return 0 ; - - *hybrid = true; -perf_pmu__for_each_hybrid_pmu (pmu) { -LIST_HEAD (terms); - -if (pmu_cmp (parse_state, pmu)) -continue ; - -copy_config_terms (&terms, config_terms); - ret =create_event_hybrid (PERF_TYPE_HW_CACHE , idx, list, - attr, name, metric_id, &terms, pmu); -free_config_terms (&terms); -if (ret) -return ret; - } - -return 0 ; -}
tools/perf/util/parse-events-hybrid.h +0 -7
@@ -15,11 +15,4 @@ int parse_events__add_numeric_hybrid(struct parse_events_state *parse_state,struct list_head *config_terms, bool *hybrid); -int parse_events__add_cache_hybrid (struct list_head *list,int *idx, -struct perf_event_attr *attr, -const char *name,const char *metric_id, -struct list_head *config_terms, -bool *hybrid, -struct parse_events_state *parse_state); -#endif /* __PERF_PARSE_EVENTS_HYBRID_H */
tools/perf/util/parse-events.c +35 -33
81 lines not on this page · raw message
tools/perf/util/parse-events.h +1 -2
@@ -172,8 +172,7 @@ int parse_events_add_tool(struct parse_events_state *parse_state, int tool_event);int parse_events_add_cache (struct list_head *list,int *idx,const char *name,struct parse_events_error *error, -struct list_head *head_config, -struct parse_events_state *parse_state); +struct list_head *head_config);int parse_events_add_breakpoint (struct list_head *list,int *idx,u64 addr,char *type,u64 len); int parse_events_add_pmu(struct parse_events_state *parse_state,
tools/perf/util/parse-events.y +1 -1
@@ -476,7 +476,7 @@ PE_LEGACY_CACHE opt_event_config list = alloc_list(); ABORT_ON(!list); - err = parse_events_add_cache(list, &parse_state->idx, $1, error, $2, parse_state); + err = parse_events_add_cache(list, &parse_state->idx, $1, error, $2); parse_events_terms__delete($2); free($1);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 27/46 perf print-events: Print legacy cache events for each PMU
In reply to Ian RogersMirroring parse_events_add_cache, list the legacy name alongside its alias with the PMU. Remove the now unnecessary hybrid logic.
--- tools/perf/util/parse-events.c | 2 +- tools/perf/util/parse-events.h | 1 + tools/perf/util/print-events.c | 85 ++++++++++++++++------------------ 3 files changed, 41 insertions(+), 47 deletions(-)
tools/perf/util/parse-events.c +1 -1
@@ -414,7 +414,7 @@ static int config_attr(struct perf_event_attr *attr,* contain hyphens and the longest name * should always be selected. */ -static int parse_events__decode_legacy_cache (const char *name,int pmu_type,__u64 *config) +int parse_events__decode_legacy_cache (const char *name,int pmu_type,__u64 *config) {int len, cache_type =-1 , cache_op =-1 , cache_result =-1 ;const char *name_end = &name[strlen (name) +1 ];
tools/perf/util/parse-events.h +1 -0
@@ -173,6 +173,7 @@ int parse_events_add_tool(struct parse_events_state *parse_state,int parse_events_add_cache (struct list_head *list,int *idx,const char *name,struct parse_events_error *error,struct list_head *head_config); +int parse_events__decode_legacy_cache (const char *name,int pmu_type,__u64 *config);int parse_events_add_breakpoint (struct list_head *list,int *idx,u64 addr,char *type,u64 len); int parse_events_add_pmu(struct parse_events_state *parse_state,
tools/perf/util/print-events.c +39 -46
95 lines not on this page · raw message
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Re: [PATCH v3 27/46] perf print-events: Print legacy cache events for each PMU
Quoted 2 lines · Ian Rogers
Mirroring
parse_events_add_cache, list the legacy name alongside its alias with the PMU. Remove the now unnecessary hybrid logic.
Before patch:
``` $ sudo ./perf list ... duration_time [Tool event] user_time [Tool event] system_time [Tool event] L1-dcache-loads [Hardware cache event] L1-dcache-load-misses [Hardware cache event] L1-dcache-prefetches [Hardware cache event] L1-icache-loads [Hardware cache event] L1-icache-load-misses [Hardware cache event] dTLB-loads [Hardware cache event] dTLB-load-misses [Hardware cache event] iTLB-loads [Hardware cache event] iTLB-load-misses [Hardware cache event] branch-loads [Hardware cache event] branch-load-misses [Hardware cache event] branch-brs OR cpu/branch-brs/ [Kernel PMU event] branch-instructions OR cpu/branch-instructions/ [Kernel PMU event] branch-misses OR cpu/branch-misses/ [Kernel PMU event] ... ```
After patch:
``` $ sudo ./perf list ... duration_time [Tool event] user_time [Tool event] system_time [Tool event] cpu: L1-dcache-loads OR cpu/L1-dcache-loads/ L1-dcache-load-misses OR cpu/L1-dcache-load-misses/ L1-dcache-prefetches OR cpu/L1-dcache-prefetches/ L1-icache-loads OR cpu/L1-icache-loads/ L1-icache-load-misses OR cpu/L1-icache-load-misses/ dTLB-loads OR cpu/dTLB-loads/ dTLB-load-misses OR cpu/dTLB-load-misses/ iTLB-loads OR cpu/iTLB-loads/ iTLB-load-misses OR cpu/iTLB-load-misses/ branch-loads OR cpu/branch-loads/ branch-load-misses OR cpu/branch-load-misses/ branch-brs OR cpu/branch-brs/ [Kernel PMU event] branch-instructions OR cpu/branch-instructions/ [Kernel PMU event] branch-misses OR cpu/branch-misses/ [Kernel PMU event] ... ```
Is this intentional change?
tools/perf/util/print-events.c void print_sdt_events(const struct print_callbacks *print_cb, void *print_state) · line 235
Quoted 9 lines
tools/perf/util/print-events.c +4
-5-for (int type =0 ; type <PERF_COUNT_HW_CACHE_MAX ; type++) { -for (int op =0 ; op <PERF_COUNT_HW_CACHE_OP_MAX ; op++) { -/* skip invalid cache type */ -if (!evsel__is_cache_op_valid (type, op)) -continue ; +while ((pmu =perf_pmu__scan (pmu)) !=NULL ) { +/* +* Skip uncore PMUs for performance. Software PMUs can open +* PERF_TYPE_HW_CACHE, so skip.
This statement is bit confusing. Can you please explain how SW pmus can open cache events.
Ian Rogers
reviewerRe: [PATCH v3 27/46] perf print-events: Print legacy cache events for each PMU
Quoted 56 lines · Ravi Bangoria
Quoted 2 lines · Ian Rogers
Mirroring
parse_events_add_cache, list the legacy name alongside its alias with the PMU. Remove the now unnecessary hybrid logic.Before patch:
``` $ sudo ./perf list ... duration_time [Tool event] user_time [Tool event] system_time [Tool event] L1-dcache-loads [Hardware cache event] L1-dcache-load-misses [Hardware cache event] L1-dcache-prefetches [Hardware cache event] L1-icache-loads [Hardware cache event] L1-icache-load-misses [Hardware cache event] dTLB-loads [Hardware cache event] dTLB-load-misses [Hardware cache event] iTLB-loads [Hardware cache event] iTLB-load-misses [Hardware cache event] branch-loads [Hardware cache event] branch-load-misses [Hardware cache event] branch-brs OR cpu/branch-brs/ [Kernel PMU event] branch-instructions OR cpu/branch-instructions/ [Kernel PMU event] branch-misses OR cpu/branch-misses/ [Kernel PMU event] ... ```After patch:
``` $ sudo ./perf list ... duration_time [Tool event] user_time [Tool event] system_time [Tool event] cpu: L1-dcache-loads OR cpu/L1-dcache-loads/ L1-dcache-load-misses OR cpu/L1-dcache-load-misses/ L1-dcache-prefetches OR cpu/L1-dcache-prefetches/ L1-icache-loads OR cpu/L1-icache-loads/ L1-icache-load-misses OR cpu/L1-icache-load-misses/ dTLB-loads OR cpu/dTLB-loads/ dTLB-load-misses OR cpu/dTLB-load-misses/ iTLB-loads OR cpu/iTLB-loads/ iTLB-load-misses OR cpu/iTLB-load-misses/ branch-loads OR cpu/branch-loads/ branch-load-misses OR cpu/branch-load-misses/ branch-brs OR cpu/branch-brs/ [Kernel PMU event] branch-instructions OR cpu/branch-instructions/ [Kernel PMU event] branch-misses OR cpu/branch-misses/ [Kernel PMU event] ... ```
Quoted 2 lines
\
Is this intentional change?
Yep, but I think the commit message should call it out, so I'll change it in v4. When we have an alias the event type descriptor isn't shown, this is pre-existing perf list behavior but I think we may want to tweak it as I like the event type descriptor.
Quoted 10 lines
tools/perf/util/print-events.c void print_sdt_events(const struct print_callbacks *print_cb, void *print_state) · line 235
Quoted 9 lines
tools/perf/util/print-events.c +4
-5-for (int type =0 ; type <PERF_COUNT_HW_CACHE_MAX ; type++) { -for (int op =0 ; op <PERF_COUNT_HW_CACHE_OP_MAX ; op++) { -/* skip invalid cache type */ -if (!evsel__is_cache_op_valid (type, op)) -continue ; +while ((pmu =perf_pmu__scan (pmu)) !=NULL ) { +/* +* Skip uncore PMUs for performance. Software PMUs can open +* PERF_TYPE_HW_CACHE, so skip.
Quoted 2 lines
This statement is bit confusing. Can you please explain how SW pmus can open cache events.
If the type is PERF_TYPE_HW_CACHE (3) and the extended type is PERF_TYPE_SOFTWARE (1) then this yields the encoding '3:0x100000000' which will succeed perf_event_open:
```
$ perf stat -vv -e '3:0x100000000' true
Using CPUID GenuineIntel-6-8D-1
intel_pt default config: tsc,mtc,mtc_period=3,psb_period=3,pt,branch
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 3
size 128
config 0x100000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
enable_on_exec 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid 2128955 cpu -1 group_fd -1 flags 0x8 = 3
3:0x100000000: -1: 265630 261923 261923
3:0x100000000: 265630 261923 261923
Performance counter stats for 'true':
265,630 3:0x100000000
0.000844251 seconds time elapsed
0.000911000 seconds user
0.000000000 seconds sys
```I agree this isn't expected, but if I don't exclude the PMU type the print events will list it as an alias. I'll try to improve the comment in v4.
Thanks,
Ian
Ian Rogers
reviewerPATCH v3 28/46 perf parse-events: Support wildcards on raw events
In reply to Ian RogersLegacy raw events like r1a open as PERF_TYPE_RAW on non-hybrid systems and on each hybrid PMU on hybrid systems. Rather than iterate hybrid PMUs add a perf_pmu__supports_wildcard_numeric function that says when a numeric event should be opened upon it. If the parsed event specifies the type of the PMU then don't wildcard match PMUs, use the specified PMU type.
--- tools/perf/util/parse-events.c | 50 ++++++++++++++++++++++++---------- tools/perf/util/parse-events.h | 3 +- tools/perf/util/parse-events.y | 13 ++++++--- tools/perf/util/pmu.c | 5 ++++ tools/perf/util/pmu.h | 1 + 5 files changed, 52 insertions(+), 20 deletions(-)
tools/perf/util/parse-events.c +35 -15
@@ -25,7 +25,6 @@#include "util/parse-branch-options.h" #include "util/evsel_config.h" #include "util/event.h" -#include "util/parse-events-hybrid.h" #include "util/pmu-hybrid.h" #include "util/bpf-filter.h" #include "util/util.h"
@@ -1448,15 +1447,14 @@ int parse_events_add_tracepoint(struct list_head *list, int *idx,#endif } -int parse_events_add_numeric (struct parse_events_state *parse_state, -struct list_head *list, -u32 type,u64 config, -struct list_head *head_config) +static int __parse_events_add_numeric (struct parse_events_state *parse_state, +struct list_head *list, +struct perf_pmu *pmu,u32 type,u64 config, +struct list_head *head_config) {struct perf_event_attr attr;LIST_HEAD (config_terms);const char *name, *metric_id; -bool hybrid;int ret;memset (&attr,0 ,sizeof (attr));
@@ -1474,19 +1472,41 @@ int parse_events_add_numeric(struct parse_events_state *parse_state, name =get_config_name (head_config); metric_id =get_config_metric_id (head_config); - ret =parse_events__add_numeric_hybrid (parse_state, list, &attr, - name, metric_id, - &config_terms, &hybrid); -if (hybrid) - goto out_free_terms; - - ret =add_event (list, &parse_state->idx, &attr, name, metric_id, - &config_terms); -out_free_terms: + ret =__add_event (list, &parse_state->idx, &attr,/*init_attr*/ true, name, + metric_id, pmu, &config_terms,/*auto_merge_stats=*/ false, +/*cpu_list=*/ NULL ) ?0 : -ENOMEM ;free_config_terms (&config_terms);return ret; } +int parse_events_add_numeric (struct parse_events_state *parse_state, +struct list_head *list, +u32 type,u64 config, +struct list_head *head_config, +bool wildcard) +{ +struct perf_pmu *pmu =NULL ; +bool found_supported = false; + +if (!wildcard) +return __parse_events_add_numeric (parse_state, list,/*pmu=*/ NULL , + type, config, head_config); + +while ((pmu =perf_pmu__scan (pmu)) !=NULL ) { +int ret; + +if (!perf_pmu__supports_wildcard_numeric (pmu)) +continue ; + + found_supported = true; + ret =__parse_events_add_numeric (parse_state, list, pmu, pmu->type, + config, head_config); +if (ret) +return ret; + } +return found_supported ?0 : -EINVAL ; +} +int parse_events_add_tool (struct parse_events_state *parse_state,struct list_head *list,int tool_event)
tools/perf/util/parse-events.h +2 -1
@@ -166,7 +166,8 @@ int parse_events_load_bpf_obj(struct parse_events_state *parse_state,int parse_events_add_numeric (struct parse_events_state *parse_state,struct list_head *list,u32 type,u64 config, -struct list_head *head_config); +struct list_head *head_config, +bool wildcard);int parse_events_add_tool (struct parse_events_state *parse_state,struct list_head *list,int tool_event);
tools/perf/util/parse-events.y +9 -4
@@ -435,7 +435,8 @@ value_sym '/' event_config '/' list = alloc_list(); ABORT_ON(!list); - err = parse_events_add_numeric(_parse_state, list, type, config, $3); + err = parse_events_add_numeric(_parse_state, list, type, config, $3, + /*wildcard=*/false); parse_events_terms__delete($3); if (err) { free_list_evsel(list);
@@ -452,7 +453,9 @@ value_sym sep_slash_slash_dc list = alloc_list(); ABORT_ON(!list); - ABORT_ON(parse_events_add_numeric(_parse_state, list, type, config, NULL)); + ABORT_ON(parse_events_add_numeric(_parse_state, list, type, config, + /*head_config=*/NULL, + /*wildcard=*/false)); $$ = list; } |
@@ -596,7 +599,8 @@ PE_VALUE ':' PE_VALUE opt_event_config list = alloc_list(); ABORT_ON(!list); - err = parse_events_add_numeric(_parse_state, list, (u32)$1, $3, $4); + err = parse_events_add_numeric(_parse_state, list, (u32)$1, $3, $4, + /*wildcard=*/false); parse_events_terms__delete($4); if (err) { free(list);
@@ -618,7 +622,8 @@ PE_RAW opt_event_config num = strtoull($1 + 1, NULL, 16); ABORT_ON(errno); free($1); - err = parse_events_add_numeric(_parse_state, list, PERF_TYPE_RAW, num, $2); + err = parse_events_add_numeric(_parse_state, list, PERF_TYPE_RAW, num, $2, + /*wildcard=*/true); parse_events_terms__delete($2); if (err) { free(list);
tools/perf/util/pmu.c +5 -0
@@ -1655,6 +1655,11 @@ bool perf_pmu__supports_legacy_cache(const struct perf_pmu *pmu)return is_pmu_core (pmu->name) ||perf_pmu__is_hybrid (pmu->name); } +bool perf_pmu__supports_wildcard_numeric (const struct perf_pmu *pmu) +{ +return is_pmu_core (pmu->name) ||perf_pmu__is_hybrid (pmu->name); +} +static bool pmu_alias_is_duplicate (struct sevent *alias_a,struct sevent *alias_b) {
tools/perf/util/pmu.h +1 -0
@@ -221,6 +221,7 @@ struct perf_pmu *perf_pmu__scan(struct perf_pmu *pmu);bool is_pmu_core (const char *name);bool perf_pmu__supports_legacy_cache (const struct perf_pmu *pmu); +bool perf_pmu__supports_wildcard_numeric (const struct perf_pmu *pmu);void print_pmu_events (const struct print_callbacks *print_cb,void *print_state);bool pmu_have_event (const char *pname,const char *name);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 29/46 perf parse-events: Remove now unused hybrid logic
In reply to Ian RogersThe event parser no longer needs to recurse in case of a legacy cache event in a PMU, the necessary wild card logic has moved to perf_pmu__supports_legacy_cache and perf_pmu__supports_wildcard_numeric.
--- tools/perf/util/Build | 1 - tools/perf/util/parse-events-hybrid.c | 181 -------------------------- tools/perf/util/parse-events-hybrid.h | 18 --- tools/perf/util/parse-events.c | 74 ----------- tools/perf/util/parse-events.h | 8 -- 5 files changed, 282 deletions(-) delete mode 100644 tools/perf/util/parse-events-hybrid.c delete mode 100644 tools/perf/util/parse-events-hybrid.h
tools/perf/util/Build +0 -1
@@ -24,7 +24,6 @@ perf-y += llvm-utils.o perf-y += mmap.o perf-y += memswap.o perf-y += parse-events.o -perf-y += parse-events-hybrid.o perf-y += print-events.o perf-y += tracepoint.o perf-y += perf_regs.o
tools/perf/util/parse-events-hybrid.c deleted +0 -181
181 lines not on this page · raw message
tools/perf/util/parse-events-hybrid.h deleted +0 -18
@@ -1,18 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef __PERF_PARSE_EVENTS_HYBRID_H -#define __PERF_PARSE_EVENTS_HYBRID_H - -#include <linux/list.h> -#include <stdbool.h> -#include <linux/types.h> -#include <linux/perf_event.h> -#include <string.h> - -int parse_events__add_numeric_hybrid (struct parse_events_state *parse_state, -struct list_head *list, -struct perf_event_attr *attr, -const char *name,const char *metric_id, -struct list_head *config_terms, -bool *hybrid); - -#endif /* __PERF_PARSE_EVENTS_HYBRID_H */
tools/perf/util/parse-events.c +0 -74
107 lines not on this page · raw message
tools/perf/util/parse-events.h +0 -8
@@ -122,7 +122,6 @@ struct parse_events_state {struct list_head *terms;int stoken;struct perf_pmu *fake_pmu; -char *hybrid_pmu_name;/* Should PE_LEGACY_NAME tokens be generated for config terms? */ bool match_legacy_cache_terms;bool wild_card_pmus;
@@ -235,11 +234,4 @@ static inline bool is_sdt_event(char *str __maybe_unused) }#endif /* HAVE_LIBELF_SUPPORT */ -struct evsel *parse_events__add_event_hybrid (struct list_head *list,int *idx, -struct perf_event_attr *attr, -const char *name, -const char *metric_id, -struct perf_pmu *pmu, -struct list_head *config_terms); -#endif /* __PERF_PARSE_EVENTS_H */
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 30/46 perf parse-events: Minor type safety cleanup
In reply to Ian RogersUse the typed parse_state rather than void* _parse_state when available.
--- tools/perf/util/parse-events.y | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
tools/perf/util/parse-events.y +3 -3
@@ -301,7 +301,7 @@ PE_NAME opt_pmu_config if (!list) CLEANUP_YYABORT; /* Attempt to add to list assuming $1 is a PMU name. */ - if (parse_events_add_pmu(_parse_state, list, $1, $2, /*auto_merge_stats=*/false)) { + if (parse_events_add_pmu(parse_state, list, $1, $2, /*auto_merge_stats=*/false)) { struct perf_pmu *pmu = NULL; int ok = 0;
@@ -319,7 +319,7 @@ PE_NAME opt_pmu_config !perf_pmu__match(pattern, pmu->alias_name, $1)) { if (parse_events_copy_term_list(orig_terms, &terms)) CLEANUP_YYABORT; - if (!parse_events_add_pmu(_parse_state, list, pmu->name, terms, + if (!parse_events_add_pmu(parse_state, list, pmu->name, terms, /*auto_merge_stats=*/true)) { ok++; parse_state->wild_card_pmus = true;
@@ -331,7 +331,7 @@ PE_NAME opt_pmu_config if (!ok) { /* Failure to add, assume $1 is an event name. */ zfree(&list); - ok = !parse_events_multi_pmu_add(_parse_state, $1, $2, &list); + ok = !parse_events_multi_pmu_add(parse_state, $1, $2, &list); $2 = NULL; } if (!ok)
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 31/46 perf parse-events: Add pmu filter
In reply to Ian RogersTo support the cputype argument added to "perf stat" for hybrid it is necessary to filter events during wildcard matching. Add a scanner argument for the filter and checking it when wildcard matching.
--- tools/perf/builtin-record.c | 13 +++++++-- tools/perf/builtin-stat.c | 10 +++++-- tools/perf/builtin-top.c | 5 +++- tools/perf/builtin-trace.c | 5 +++- tools/perf/tests/parse-events.c | 3 +- tools/perf/tests/pmu-events.c | 3 +- tools/perf/util/evlist.h | 1 - tools/perf/util/metricgroup.c | 4 +-- tools/perf/util/parse-events.c | 51 ++++++++++++++++++++++++--------- tools/perf/util/parse-events.h | 21 ++++++++++---- tools/perf/util/parse-events.y | 6 ++-- 11 files changed, 90 insertions(+), 32 deletions(-)
tools/perf/builtin-record.c +11 -2
@@ -3335,6 +3335,14 @@ const char record_callchain_help[] = CALLCHAIN_RECORD_HELPstatic bool dry_run; +static struct parse_events_option_args parse_events_option_args = { + .evlistp = &record.evlist, +}; + +static struct parse_events_option_args switch_output_parse_events_option_args = { + .evlistp = &record.sb_evlist, +}; +/* * XXX Will stay a global variable till we fix builtin-script.c to stop messing * with it and switch to use the library functions in perf_evlist that came
@@ -3343,7 +3351,7 @@ static bool dry_run;* using pipes, etc. */ static struct option __record_options[] = { -OPT_CALLBACK ('e' ,"event" , &record.evlist,"event" , +OPT_CALLBACK ('e' ,"event" , &parse_events_option_args,"event" ,"event selector. use 'perf list' to list available events" , parse_events_option), OPT_CALLBACK(0, "filter", &record.evlist, "filter",
@@ -3496,7 +3504,8 @@ static struct option __record_options[] = { &record.switch_output.set, "signal or size[BKMG] or time[smhd]", "Switch output when receiving SIGUSR2 (signal) or cross a size or time threshold", "signal"), -OPT_CALLBACK_SET (0 ,"switch-output-event" , &record.sb_evlist, &record.switch_output_event_set,"switch output event" , +OPT_CALLBACK_SET (0 ,"switch-output-event" , &switch_output_parse_events_option_args, + &record.switch_output_event_set,"switch output event" ,"switch output event selector. use 'perf list' to list available events" , parse_events_option_new_evlist), OPT_INTEGER(0, "switch-max-files", &record.switch_output.num_files,
tools/perf/builtin-stat.c +7 -3
@@ -101,6 +101,10 @@static void print_counters (struct timespec *ts,int argc,const char **argv);static struct evlist *evsel_list; +static struct parse_events_option_args parse_events_option_args = { + .evlistp = &evsel_list, +}; +static bool all_counters_use_bpf = true;static struct target target = {
@@ -1096,8 +1100,8 @@ static int parse_hybrid_type(const struct option *opt,return -1 ; } - evlist->hybrid_pmu_name =perf_pmu__hybrid_type_to_pmu (str); -if (!evlist->hybrid_pmu_name) { + parse_events_option_args.pmu_filter =perf_pmu__hybrid_type_to_pmu (str); +if (!parse_events_option_args.pmu_filter) {fprintf (stderr,"--cputype %s is not supported!\n" , str);return -1 ; }
@@ -1108,7 +1112,7 @@ static int parse_hybrid_type(const struct option *opt,static struct option stat_options[] = {OPT_BOOLEAN ('T' ,"transaction" , &transaction_run,"hardware transaction statistics" ), -OPT_CALLBACK ('e' ,"event" , &evsel_list,"event" , +OPT_CALLBACK ('e' ,"event" , &parse_events_option_args,"event" ,"event selector. use 'perf list' to list available events" , parse_events_option), OPT_CALLBACK(0, "filter", &evsel_list, "filter",
tools/perf/builtin-top.c +4 -1
@@ -1440,12 +1440,15 @@ int cmd_top(int argc, const char **argv) .max_stack = sysctl__max_stack(), .nr_threads_synthesize = UINT_MAX, }; +struct parse_events_option_args parse_events_option_args = { + .evlistp = &top.evlist, + };bool branch_call_mode = false;struct record_opts *opts = &top.record_opts;struct target *target = &opts->target;const char *disassembler_style =NULL , *objdump_path =NULL , *addr2line_path =NULL ;const struct option options[] = { -OPT_CALLBACK ('e' ,"event" , &top.evlist,"event" , +OPT_CALLBACK ('e' ,"event" , &parse_events_option_args,"event" ,"event selector. use 'perf list' to list available events" , parse_events_option),OPT_U64 ('c' ,"count" , &opts->user_interval,"event period to sample" ),
tools/perf/builtin-trace.c +4 -1
@@ -4591,8 +4591,11 @@ static int trace__parse_events_option(const struct option *opt, const char *str, err =0 ;if (lists[0 ]) { +struct parse_events_option_args parse_events_option_args = { + .evlistp = &trace->evlist, + };struct option o = { - .value = &trace->evlist, + .value = &parse_events_option_args, }; err =parse_events_option (&o, lists[0 ],0 ); }
tools/perf/tests/parse-events.c +2 -1
@@ -1938,7 +1938,8 @@ static int test_event_fake_pmu(const char *str)return -ENOMEM ;parse_events_error__init (&err); - ret =__parse_events (evlist, str, &err, &perf_pmu__fake,/*warn_if_reordered=*/ true); + ret =__parse_events (evlist, str,/*pmu_filter=*/ NULL , &err, + &perf_pmu__fake,/*warn_if_reordered=*/ true);if (ret) {pr_debug ("failed to parse event '%s', err %d, str '%s'\n" , str, ret, err.str);
tools/perf/tests/pmu-events.c +2 -1
@@ -776,7 +776,8 @@ static int check_parse_id(const char *id, struct parse_events_error *error,for (cur =strchr (dup,'@' ) ; cur; cur =strchr (++cur,'@' )) *cur ='/' ; - ret =__parse_events (evlist, dup, error, fake_pmu,/*warn_if_reordered=*/ true); + ret =__parse_events (evlist, dup,/*pmu_filter=*/ NULL , error, fake_pmu, +/*warn_if_reordered=*/ true);free (dup);evlist__delete (evlist);
tools/perf/util/evlist.h +0 -1
@@ -67,7 +67,6 @@ struct evlist {struct evsel *selected;struct events_stats stats;struct perf_env *env; -const char *hybrid_pmu_name;void (*trace_event_sample_raw)(struct evlist *evlist,union perf_event *event,struct perf_sample *sample);
tools/perf/util/metricgroup.c +2 -2
@@ -1441,8 +1441,8 @@ static int parse_ids(bool metric_no_merge, struct perf_pmu *fake_pmu, }pr_debug ("Parsing metric events '%s'\n" , events.buf);parse_events_error__init (&parse_error); - ret =__parse_events (parsed_evlist, events.buf, &parse_error, fake_pmu, -/*warn_if_reordered=*/ false); + ret =__parse_events (parsed_evlist, events.buf,/*pmu_filter=*/ NULL , + &parse_error, fake_pmu,/*warn_if_reordered=*/ false);if (ret) {parse_events_error__print (&parse_error, events.buf); goto err_out;
tools/perf/util/parse-events.c +38 -13
119 lines not on this page · raw message
tools/perf/util/parse-events.h +16 -5
@@ -22,17 +22,24 @@ bool is_event_supported(u8 type, u64 config);const char *event_type (int type); +/* Arguments encoded in opt->value. */ +struct parse_events_option_args { +struct evlist **evlistp; +const char *pmu_filter; +};int parse_events_option (const struct option *opt,const char *str,int unset);int parse_events_option_new_evlist (const struct option *opt,const char *str,int unset); -__attribute__((nonnull (1 ,2 ,3 ))) -int __parse_events (struct evlist *evlist,const char *str,struct parse_events_error *error, -struct perf_pmu *fake_pmu,bool warn_if_reordered); +__attribute__((nonnull (1 ,2 ,4 ))) +int __parse_events (struct evlist *evlist,const char *str,const char *pmu_filter, +struct parse_events_error *error,struct perf_pmu *fake_pmu, +bool warn_if_reordered); __attribute__((nonnull (1 ,2 ,3 )))static inline int parse_events (struct evlist *evlist,const char *str,struct parse_events_error *err) { -return __parse_events (evlist, str, err,/*fake_pmu=*/ NULL ,/*warn_if_reordered=*/ true); +return __parse_events (evlist, str,/*pmu_filter=*/ NULL , err,/*fake_pmu=*/ NULL , +/*warn_if_reordered=*/ true); }int parse_event (struct evlist *evlist,const char *str);
@@ -122,11 +129,15 @@ struct parse_events_state {struct list_head *terms;int stoken;struct perf_pmu *fake_pmu; +/* If non-null, when wildcard matching only match the given PMU. */ +const char *pmu_filter;/* Should PE_LEGACY_NAME tokens be generated for config terms? */ bool match_legacy_cache_terms;bool wild_card_pmus; }; +bool parse_events__filter_pmu (const struct parse_events_state *parse_state, +const struct perf_pmu *pmu);void parse_events__shrink_config_terms (void );int parse_events__is_hardcoded_term (struct parse_events_term *term); int parse_events_term__num(struct parse_events_term **term,
@@ -171,7 +182,7 @@ int parse_events_add_tool(struct parse_events_state *parse_state,struct list_head *list, int tool_event);int parse_events_add_cache (struct list_head *list,int *idx,const char *name, -struct parse_events_error *error, +struct parse_events_state *parse_state,struct list_head *head_config);int parse_events__decode_legacy_cache (const char *name,int pmu_type,__u64 *config); int parse_events_add_breakpoint(struct list_head *list, int *idx,
tools/perf/util/parse-events.y +4 -2
@@ -312,6 +312,9 @@ PE_NAME opt_pmu_config while ((pmu = perf_pmu__scan(pmu)) != NULL) { char *name = pmu->name; + if (parse_events__filter_pmu(parse_state, pmu)) + continue; + if (!strncmp(name, "uncore_", 7) && strncmp($1, "uncore_", 7)) name += 7;
@@ -473,13 +476,12 @@ event_legacy_cache: PE_LEGACY_CACHE opt_event_config { struct parse_events_state *parse_state = _parse_state; - struct parse_events_error *error = parse_state->error; struct list_head *list; int err; list = alloc_list(); ABORT_ON(!list); - err = parse_events_add_cache(list, &parse_state->idx, $1, error, $2); + err = parse_events_add_cache(list, &parse_state->idx, $1, parse_state, $2); parse_events_terms__delete($2); free($1);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 32/46 perf stat: Make cputype filter generic
In reply to Ian RogersRather than limit the --cputype argument for "perf list" and "perf stat" to hybrid PMUs of just cpu_atom and cpu_core, allow any PMU.
Note, that if cpu_atom isn't mounted but a filter of cpu_atom is requested, then this will now fail. As such a filter would never succeed, no events can come from that unmounted PMU, then this behavior could never have been useful and failing is clearer.
--- tools/perf/builtin-list.c | 19 +++++++++++-------- tools/perf/builtin-stat.c | 12 +++++++----- tools/perf/util/pmu-hybrid.c | 20 -------------------- tools/perf/util/pmu-hybrid.h | 1 - tools/perf/util/pmus.c | 25 ++++++++++++++++++++++++- tools/perf/util/pmus.h | 3 +++ 6 files changed, 45 insertions(+), 35 deletions(-)
tools/perf/builtin-list.c +11 -8
@@ -11,8 +11,8 @@#include "builtin.h" #include "util/print-events.h" +#include "util/pmus.h" #include "util/pmu.h" -#include "util/pmu-hybrid.h" #include "util/debug.h" #include "util/metricgroup.h" #include "util/string2.h"
@@ -429,7 +429,7 @@ int cmd_list(int argc, const char **argv) .print_event = default_print_event, .print_metric = default_print_metric, }; -const char *hybrid_name =NULL ; +const char *cputype =NULL ;const char *unit_name =NULL ;bool json = false;struct option list_options[] = {
@@ -443,8 +443,8 @@ int cmd_list(int argc, const char **argv) "Print information on the perf event names and expressions used internally by events."),OPT_BOOLEAN (0 ,"deprecated" , &default_ps.deprecated,"Print deprecated events." ), -OPT_STRING (0 ,"cputype" , &hybrid_name,"hybrid cpu type" , -"Limit PMU or metric printing to the given hybrid PMU (e.g. core or atom)." ), +OPT_STRING (0 ,"cputype" , &cputype,"cpu type" , +"Limit PMU or metric printing to the given PMU (e.g. cpu, core or atom)." ),OPT_STRING (0 ,"unit" , &unit_name,"PMU name" ,"Limit PMU or metric printing to the specified PMU." ), OPT_INCR(0, "debug", &verbose,
@@ -484,10 +484,13 @@ int cmd_list(int argc, const char **argv)assert (default_ps.visited_metrics);if (unit_name) default_ps.pmu_glob =strdup (unit_name); -else if (hybrid_name) { - default_ps.pmu_glob =perf_pmu__hybrid_type_to_pmu (hybrid_name); -if (!default_ps.pmu_glob) -pr_warning ("WARNING: hybrid cputype is not supported!\n" ); +else if (cputype) { +const struct perf_pmu *pmu =perf_pmus__pmu_for_pmu_filter (cputype); + +if (!pmu) +pr_warning ("WARNING: cputype is not supported!\n" ); + + default_ps.pmu_glob = pmu->name; } } print_cb.print_start (ps);
tools/perf/builtin-stat.c +7 -5
@@ -44,6 +44,7 @@#include "util/cgroup.h" #include <subcmd/parse-options.h> #include "util/parse-events.h" +#include "util/pmus.h" #include "util/pmu.h" #include "util/event.h" #include "util/evlist.h"
@@ -69,7 +70,6 @@#include "util/pfm.h" #include "util/bpf_counter.h" #include "util/iostat.h" -#include "util/pmu-hybrid.h" #include "util/util.h" #include "asm/bug.h"
@@ -1089,10 +1089,11 @@ static int parse_stat_cgroups(const struct option *opt,return parse_cgroups (opt, str, unset); } -static int parse_hybrid_type (const struct option *opt, +static int parse_cputype (const struct option *opt,const char *str,int unset__maybe_unused ) { +const struct perf_pmu *pmu;struct evlist *evlist = *(struct evlist **)opt->value;if (!list_empty (&evlist->core.entries)) {
@@ -1100,11 +1101,12 @@ static int parse_hybrid_type(const struct option *opt,return -1 ; } - parse_events_option_args.pmu_filter =perf_pmu__hybrid_type_to_pmu (str); -if (!parse_events_option_args.pmu_filter) { + pmu =perf_pmus__pmu_for_pmu_filter (str); +if (!pmu) {fprintf (stderr,"--cputype %s is not supported!\n" , str);return -1 ; } + parse_events_option_args.pmu_filter = pmu->name;return 0 ; }
@@ -1230,7 +1232,7 @@ static struct option stat_options[] = { OPT_CALLBACK(0, "cputype", &evsel_list, "hybrid cpu type", "Only enable events on applying cpu with this type " "for hybrid platform (e.g. core or atom)", - parse_hybrid_type), + parse_cputype),#ifdef HAVE_LIBPFM OPT_CALLBACK(0, "pfm-events", &evsel_list, "event", "libpfm4 event selector. use 'perf list' to list available events",
tools/perf/util/pmu-hybrid.c +0 -20
@@ -50,23 +50,3 @@ bool perf_pmu__is_hybrid(const char *name) {return perf_pmu__find_hybrid_pmu (name) !=NULL ; } - -char *perf_pmu__hybrid_type_to_pmu (const char *type) -{ -char *pmu_name =NULL ; - -if (asprintf (&pmu_name,"cpu_%s" , type) <0 ) -return NULL ; - -if (perf_pmu__is_hybrid (pmu_name)) -return pmu_name; - -/* -* pmu may be not scanned, check the sysfs. -*/ -if (perf_pmu__hybrid_mounted (pmu_name)) -return pmu_name; - -free (pmu_name); -return NULL ; -}
tools/perf/util/pmu-hybrid.h +0 -1
@@ -17,7 +17,6 @@ bool perf_pmu__hybrid_mounted(const char *name);struct perf_pmu *perf_pmu__find_hybrid_pmu (const char *name);bool perf_pmu__is_hybrid (const char *name); -char *perf_pmu__hybrid_type_to_pmu (const char *type);static inline int perf_pmu__hybrid_pmu_num (void ) {
tools/perf/util/pmus.c +24 -1
@@ -1,5 +1,28 @@// SPDX-License-Identifier: GPL-2.0 #include <linux/list.h> -#include <pmus.h> +#include <string.h> +#include "pmus.h" +#include "pmu.h" LIST_HEAD (pmus); + +const struct perf_pmu *perf_pmus__pmu_for_pmu_filter (const char *str) +{ +struct perf_pmu *pmu =NULL ; + +while ((pmu =perf_pmu__scan (pmu)) !=NULL ) { +if (!strcmp (pmu->name, str)) +return pmu; +/* Ignore "uncore_" prefix. */ +if (!strncmp (pmu->name,"uncore_" ,7 )) { +if (!strcmp (pmu->name +7 , str)) +return pmu; + } +/* Ignore "cpu_" prefix on Intel hybrid PMUs. */ +if (!strncmp (pmu->name,"cpu_" ,4 )) { +if (!strcmp (pmu->name +4 , str)) +return pmu; + } + } +return NULL ; +}
tools/perf/util/pmus.h +3 -0
@@ -3,7 +3,10 @@#define __PMUS_Hextern struct list_head pmus; +struct perf_pmu ;#define perf_pmus__for_each_pmu (pmu) list_for_each_entry(pmu, &pmus, list) +const struct perf_pmu *perf_pmus__pmu_for_pmu_filter (const char *str); +#endif /* __PMUS_H */
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Re: [PATCH v3 32/46] perf stat: Make cputype filter generic
Quoted 2 lines · Ian Rogers
Rather than limit the --cputype argument for "perf list" and "perf stat" to hybrid PMUs of just
cpu_atomandcpu_core, allow any PMU.
I've couple of doubts:
1. Can you please explain intention to do this esp for perf list. Since, IIUC, perf list --unit option provide the same functionality.
2. Since we are already specifying pmu name for non-standerd/arch-specific events like pmu/attributes/, I'm not sure where perf stat --cputype=pmu is useful. Can you please explain perf stat usability aspect for non-hybrid pmus.
3. What am I missing here:
$ sudo ./perf stat --cputype=amd_df -e amd_l3/event=0x4,umask=0xff/ -C 0 -- sleep 1
Performance counter stats for 'CPU(s) 0': 108,267 amd_l3/event=0x4,umask=0xff/
1.061290167 seconds time elapsed3. Also, IMHO, using --cputype option to specify _pmu name_ is bit odd.
Quoted 5 lines
Note, that if
cpu_atomisn't mounted but a filter ofcpu_atomis requested, then this will now fail. As such a filter would never succeed, no events can come from that unmounted PMU, then this behavior could never have been useful and failing is clearer.
I'm hitting a segfault if I use non-existing pmu:
$ sudo ./perf list --cputype=random WARNING: cputype is not supported! Segmentation fault
tools/perf/builtin-list.c int cmd_list(int argc, const char **argv) · line 447
Quoted 8 lines
tools/perf/builtin-list.c +2
-2@@ -443,8 +443,8 @@ int cmd_list(int argc, const char **argv) "Print information on the perf event names and expressions used internally by events."),OPT_BOOLEAN (0 ,"deprecated" , &default_ps.deprecated,"Print deprecated events." ), -OPT_STRING (0 ,"cputype" , &hybrid_name,"hybrid cpu type" , - "Limit PMU or metric printing to the given hybrid PMU (e.g. core or atom)."), +OPT_STRING (0 ,"cputype" , &cputype,"cpu type" , + "Limit PMU or metric printing to the given PMU (e.g. cpu, core or atom)."),
man perf-list does not describe --cputype. I think we should add it as part of this patch?
Similarly, man perf-stat also needs to be updated.
tools/perf/util/pmus.c line 19
Quoted 11 lines
tools/perf/util/pmus.c +11
-0+const struct perf_pmu *perf_pmus__pmu_for_pmu_filter (const char *str) +{ +struct perf_pmu *pmu =NULL ; + +while ((pmu =perf_pmu__scan (pmu)) !=NULL ) { +if (!strcmp (pmu->name, str)) +return pmu; +/* Ignore "uncore_" prefix. */ +if (!strncmp (pmu->name,"uncore_" ,7 )) { +if (!strcmp (pmu->name +7 , str)) +return pmu;
Any specific reason to ignore "uncore_"? IMHO, ignoring prefix of some pmus and not of others is bit confusing for naive user.
Ian Rogers
reviewerRe: [PATCH v3 32/46] perf stat: Make cputype filter generic
Quoted 9 lines · Ravi Bangoria
Quoted 2 lines · Ian Rogers
Rather than limit the --cputype argument for "perf list" and "perf stat" to hybrid PMUs of just
cpu_atomandcpu_core, allow any PMU.I've couple of doubts:
1. Can you please explain intention to do this esp for perf list. Since, IIUC,
perf list --unitoption provide the same functionality.
I agree with you. The option already exists and I think we should just move this option to being deprecated/hidden:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/lib/subcmd/parse-options.h#n46
Quoted 4 lines
2. Since we are already specifying pmu name for non-standerd/arch-specific events like
pmu/attributes/, I'm not sure whereperf stat --cputype=pmuis useful. Can you please explain perf stat usability aspect for non-hybrid pmus.
Completely agreed. This patch series is trying to remove the duplicated code introduced by the hybrid changes. In this case I didn't want to remove an option, and potentially break users of that option, as part of fixing things up. A lot of what you are saying here I added as comments to the original patch series.
Quoted 8 lines
3. What am I missing here:
$ sudo ./perf stat --cputype=amd_df -e amd_l3/event=0x4,umask=0xff/ -C 0 -- sleep 1 Performance counter stats for 'CPU(s) 0':108,267 amd_l3/event=0x4,umask=0xff/ 1.061290167 seconds time elapsed
The cputype applies to wildcard-ed events. So on hybrid:
$ perf stat -e cycles true will open cycles on cpu_atom and cpu_core, the parse_events__filter_pmu function is used skip PMUs based on cputype.
Quoted 1 line
3. Also, IMHO, using --cputype option to specify
_pmuname_is bit odd.
Right, when the "feature" was added I would have preferred it as PMU rather than CPU.
Quoted 11 lines
Quoted 5 lines
Note, that if
cpu_atomisn't mounted but a filter ofcpu_atomis requested, then this will now fail. As such a filter would never succeed, no events can come from that unmounted PMU, then this behavior could never have been useful and failing is clearer.I'm hitting a segfault if I use non-existing pmu:
$ sudo ./perf list --cputype=random WARNING: cputype is not supported! Segmentation fault
Will fix in v4. The warning should be fatal/exit rather than try to read the PMU's name.
Thanks,
Ian
Quoted 29 lines
tools/perf/builtin-list.c int cmd_list(int argc, const char **argv) · line 447
Quoted 8 lines
tools/perf/builtin-list.c +2
-2@@ -443,8 +443,8 @@ int cmd_list(int argc, const char **argv) "Print information on the perf event names and expressions used internally by events."),OPT_BOOLEAN (0 ,"deprecated" , &default_ps.deprecated,"Print deprecated events." ), -OPT_STRING (0 ,"cputype" , &hybrid_name,"hybrid cpu type" , - "Limit PMU or metric printing to the given hybrid PMU (e.g. core or atom)."), +OPT_STRING (0 ,"cputype" , &cputype,"cpu type" , + "Limit PMU or metric printing to the given PMU (e.g. cpu, core or atom)."),man perf-list does not describe --cputype. I think we should add it as part of this patch?
Similarly, man perf-stat also needs to be updated.
tools/perf/util/pmus.c line 19
Quoted 11 lines
tools/perf/util/pmus.c +11
-0+const struct perf_pmu *perf_pmus__pmu_for_pmu_filter (const char *str) +{ +struct perf_pmu *pmu =NULL ; + +while ((pmu =perf_pmu__scan (pmu)) !=NULL ) { +if (!strcmp (pmu->name, str)) +return pmu; +/* Ignore "uncore_" prefix. */ +if (!strncmp (pmu->name,"uncore_" ,7 )) { +if (!strcmp (pmu->name +7 , str)) +return pmu;Any specific reason to ignore "
uncore_"? IMHO, ignoring prefix of some pmus and not of others is bit confusing for naive user.
Ian Rogers
reviewerRe: [PATCH v3 32/46] perf stat: Make cputype filter generic
In reply to Ravi BangoriaQuoted 51 lines · Ravi Bangoria
Quoted 2 lines · Ian Rogers
Rather than limit the --cputype argument for "perf list" and "perf stat" to hybrid PMUs of just
cpu_atomandcpu_core, allow any PMU.I've couple of doubts:
1. Can you please explain intention to do this esp for perf list. Since, IIUC,
perf list --unitoption provide the same functionality.2. Since we are already specifying pmu name for non-standerd/arch-specific events like
pmu/attributes/, I'm not sure whereperf stat --cputype=pmuis useful. Can you please explain perf stat usability aspect for non-hybrid pmus.3. What am I missing here:
$ sudo ./perf stat --cputype=amd_df -e amd_l3/event=0x4,umask=0xff/ -C 0 -- sleep 1 Performance counter stats for 'CPU(s) 0':108,267 amd_l3/event=0x4,umask=0xff/ 1.061290167 seconds time elapsed3. Also, IMHO, using --cputype option to specify
_pmuname_is bit odd.Quoted 5 lines
Note, that if
cpu_atomisn't mounted but a filter ofcpu_atomis requested, then this will now fail. As such a filter would never succeed, no events can come from that unmounted PMU, then this behavior could never have been useful and failing is clearer.I'm hitting a segfault if I use non-existing pmu:
$ sudo ./perf list --cputype=random WARNING: cputype is not supported! Segmentation faulttools/perf/builtin-list.c int cmd_list(int argc, const char **argv) · line 447
Quoted 8 lines
tools/perf/builtin-list.c +2
-2@@ -443,8 +443,8 @@ int cmd_list(int argc, const char **argv) "Print information on the perf event names and expressions used internally by events."),OPT_BOOLEAN (0 ,"deprecated" , &default_ps.deprecated,"Print deprecated events." ), -OPT_STRING (0 ,"cputype" , &hybrid_name,"hybrid cpu type" , - "Limit PMU or metric printing to the given hybrid PMU (e.g. core or atom)."), +OPT_STRING (0 ,"cputype" , &cputype,"cpu type" , + "Limit PMU or metric printing to the given PMU (e.g. cpu, core or atom)."),man perf-list does not describe --cputype. I think we should add it as part of this patch?
Quoted 1 line
Similarly, man perf-stat also needs to be updated.
Ok, these changes are just keeping existing functionality. I don't disagree with making these changes but I think we can follow up with them. Or probably just deprecate the option.
Quoted 6 lines
Quoted 6 lines
tools/perf/util/pmus.c +6
-0+const struct perf_pmu *perf_pmus__pmu_for_pmu_filter (const char *str) +{ +struct perf_pmu *pmu =NULL ; + +while ((pmu =perf_pmu__scan (pmu)) !=NULL ) { +if (!strcmp (pmu->name, str))
Quoted 8 lines
tools/perf/util/pmus.c line 19
Quoted 5 lines
tools/perf/util/pmus.c +5
-0+return pmu; +/* Ignore "uncore_" prefix. */ +if (!strncmp (pmu->name,"uncore_" ,7 )) { +if (!strcmp (pmu->name +7 , str)) +return pmu;Any specific reason to ignore "
uncore_"? IMHO, ignoring prefix of some pmus and not of others is bit confusing for naive user.
It is trying to be consistent with the event parser:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/util/parse-events.y#n326
where uncore is consumed. Fun fact, as numbers are consumed after a PMU name and the underscore is optional, i915 as a PMU can be matched in parsing with a PMU name of just "i".
Thanks,
Ian
Ian Rogers
reviewerPATCH v3 33/46 perf test: Add cputype testing to perf stat
In reply to Ian RogersCheck a bogus PMU fails and that a known PMU succeeds. Limit to PMUs known cpu, cpu_atom and armv8_pmuv3_0 ones.
--- tools/perf/tests/shell/stat.sh | 44 ++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+)
tools/perf/tests/shell/stat.sh +44 -0
@@ -91,9 +91,53 @@ test_topdown_weak_groups() { echo "Topdown weak groups test [Success]" } +test_cputype() { + # Test --cputype argument. + echo "cputype test" + + # Bogus PMU should fail. + if perf stat --cputype="123" -e instructions true > /dev/null 2>&1 + then + echo "cputype test [Bogus PMU didn't fail]" + err=1 + return + fi + + # Find a known PMU for cputype. + pmu="" + for i in cpu cpu_atom armv8_pmuv3_0 + do + if test -d "/sys/devices/$i" + then + pmu="$i" + break + fi + if perf stat -e "$i/instructions/" true > /dev/null 2>&1 + then + pmu="$i" + break + fi + done + if test "x$pmu" = "x" + then + echo "cputype test [Skipped known PMU not found]" + return + fi + + # Test running with cputype produces output. + if ! perf stat --cputype="$pmu" -e instructions true 2>&1 | grep -E -q "instructions" + then + echo "cputype test [Failed count missed with given filter]" + err=1 + return + fi + echo "cputype test [Success]" +} + test_default_stat test_stat_record_report test_stat_repeat_weak_groups test_topdown_groups test_topdown_weak_groups +test_cputype exit $err
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 34/46 perf test: Fix parse-events tests for >1 core PMU
In reply to Ian RogersRemove assumptions of just 1 core PMU.
--- tools/perf/tests/parse-events.c | 177 +++++++++++++++++++------------- 1 file changed, 105 insertions(+), 72 deletions(-)
tools/perf/tests/parse-events.c +105 -72
@@ -25,6 +25,11 @@ static bool test_config(const struct evsel *evsel, __u64 expected_config)return (evsel->core.attr.config &PERF_HW_EVENT_MASK ) == expected_config; } +static bool test_perf_config (const struct perf_evsel *evsel,__u64 expected_config) +{ +return (evsel->attr.config &PERF_HW_EVENT_MASK ) == expected_config; +} +#ifdef HAVE_LIBTRACEEVENT #if defined(__s390x__)
@@ -87,11 +92,27 @@ static int test__checkevent_tracepoint_multi(struct evlist *evlist)static int test__checkevent_raw (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); +struct perf_evsel *evsel; +bool raw_type_match = false; -TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,0x1a )); +TEST_ASSERT_VAL ("wrong number of entries" ,0 != evlist->core.nr_entries); + +perf_evlist__for_each_evsel (&evlist->core, evsel) { +struct perf_pmu *pmu; +bool type_matched = false; + +TEST_ASSERT_VAL ("wrong config" ,test_perf_config (evsel,0x1a )); +perf_pmus__for_each_pmu (pmu) { +if (pmu->type == evsel->attr.type) { +TEST_ASSERT_VAL ("PMU type expected once" , !type_matched); + type_matched = true; +if (pmu->type ==PERF_TYPE_RAW ) + raw_type_match = true; + } + } +TEST_ASSERT_VAL ("No PMU found for type" , type_matched); + } +TEST_ASSERT_VAL ("Raw PMU not matched" , raw_type_match);return TEST_OK ; }
@@ -107,31 +128,35 @@ static int test__checkevent_numeric(struct evlist *evlist)static int test__checkevent_symbolic_name (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); +struct perf_evsel *evsel; -TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_INSTRUCTIONS )); +TEST_ASSERT_VAL ("wrong number of entries" ,0 != evlist->core.nr_entries); + +perf_evlist__for_each_evsel (&evlist->core, evsel) { +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->attr.type); +TEST_ASSERT_VAL ("wrong config" , +test_perf_config (evsel,PERF_COUNT_HW_INSTRUCTIONS )); + }return TEST_OK ; }static int test__checkevent_symbolic_name_config (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); +struct perf_evsel *evsel; -TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES )); -/* -* The period value gets configured within evlist__config, -* while this test executes only parse events method. -*/ -TEST_ASSERT_VAL ("wrong period" , -0 == evsel->core.attr.sample_period); -TEST_ASSERT_VAL ("wrong config1" , -0 == evsel->core.attr.config1); -TEST_ASSERT_VAL ("wrong config2" , -1 == evsel->core.attr.config2); +TEST_ASSERT_VAL ("wrong number of entries" ,0 != evlist->core.nr_entries); + +perf_evlist__for_each_evsel (&evlist->core, evsel) { +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->attr.type); +TEST_ASSERT_VAL ("wrong config" ,test_perf_config (evsel,PERF_COUNT_HW_CPU_CYCLES )); +/* +* The period value gets configured within evlist__config, +* while this test executes only parse events method. +*/ +TEST_ASSERT_VAL ("wrong period" ,0 == evsel->attr.sample_period); +TEST_ASSERT_VAL ("wrong config1" ,0 == evsel->attr.config1); +TEST_ASSERT_VAL ("wrong config2" ,1 == evsel->attr.config2); + }return TEST_OK ; }
@@ -147,11 +172,14 @@ static int test__checkevent_symbolic_alias(struct evlist *evlist)static int test__checkevent_genhw (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); +struct perf_evsel *evsel; -TEST_ASSERT_VAL ("wrong number of entries" ,1 == evlist->core.nr_entries); -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HW_CACHE == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,1 <<16 )); +TEST_ASSERT_VAL ("wrong number of entries" ,0 != evlist->core.nr_entries); + +perf_evlist__for_each_entry (&evlist->core, evsel) { +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HW_CACHE == evsel->attr.type); +TEST_ASSERT_VAL ("wrong config" ,test_perf_config (evsel,1 <<16 )); + }return TEST_OK ; }
@@ -243,17 +271,15 @@ static int test__checkevent_tracepoint_modifier(struct evlist *evlist)static int test__checkevent_tracepoint_multi_modifier (struct evlist *evlist) { -struct evsel *evsel; +struct perf_evsel *evsel;TEST_ASSERT_VAL ("wrong number of entries" , evlist->core.nr_entries >1 ); -evlist__for_each_entry (evlist, evsel) { -TEST_ASSERT_VAL ("wrong exclude_user" , - !evsel->core.attr.exclude_user); -TEST_ASSERT_VAL ("wrong exclude_kernel" , - evsel->core.attr.exclude_kernel); -TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv); -TEST_ASSERT_VAL ("wrong precise_ip" , !evsel->core.attr.precise_ip); +perf_evlist__for_each_entry (&evlist->core, evsel) { +TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->attr.exclude_user); +TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->attr.exclude_kernel); +TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->attr.exclude_hv); +TEST_ASSERT_VAL ("wrong precise_ip" , !evsel->attr.precise_ip); }return test__checkevent_tracepoint_multi (evlist);
@@ -262,25 +288,27 @@ test__checkevent_tracepoint_multi_modifier(struct evlist *evlist)static int test__checkevent_raw_modifier (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); - -TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user); -TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel); -TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv); -TEST_ASSERT_VAL ("wrong precise_ip" , evsel->core.attr.precise_ip); +struct perf_evsel *evsel; +perf_evlist__for_each_entry (&evlist->core, evsel) { +TEST_ASSERT_VAL ("wrong exclude_user" , evsel->attr.exclude_user); +TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->attr.exclude_kernel); +TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->attr.exclude_hv); +TEST_ASSERT_VAL ("wrong precise_ip" , evsel->attr.precise_ip); + }return test__checkevent_raw (evlist); }static int test__checkevent_numeric_modifier (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); - -TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user); -TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->core.attr.exclude_kernel); -TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv); -TEST_ASSERT_VAL ("wrong precise_ip" , evsel->core.attr.precise_ip); +struct perf_evsel *evsel; +perf_evlist__for_each_entry (&evlist->core, evsel) { +TEST_ASSERT_VAL ("wrong exclude_user" , evsel->attr.exclude_user); +TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->attr.exclude_kernel); +TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->attr.exclude_hv); +TEST_ASSERT_VAL ("wrong precise_ip" , evsel->attr.precise_ip); + }return test__checkevent_numeric (evlist); }
@@ -298,21 +326,23 @@ static int test__checkevent_symbolic_name_modifier(struct evlist *evlist)static int test__checkevent_exclude_host_modifier (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); - -TEST_ASSERT_VAL ("wrong exclude guest" , !evsel->core.attr.exclude_guest); -TEST_ASSERT_VAL ("wrong exclude host" , evsel->core.attr.exclude_host); +struct perf_evsel *evsel; +perf_evlist__for_each_entry (&evlist->core, evsel) { +TEST_ASSERT_VAL ("wrong exclude guest" , !evsel->attr.exclude_guest); +TEST_ASSERT_VAL ("wrong exclude host" , evsel->attr.exclude_host); + }return test__checkevent_symbolic_name (evlist); }static int test__checkevent_exclude_guest_modifier (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); - -TEST_ASSERT_VAL ("wrong exclude guest" , evsel->core.attr.exclude_guest); -TEST_ASSERT_VAL ("wrong exclude host" , !evsel->core.attr.exclude_host); +struct perf_evsel *evsel; +perf_evlist__for_each_entry (&evlist->core, evsel) { +TEST_ASSERT_VAL ("wrong exclude guest" , evsel->attr.exclude_guest); +TEST_ASSERT_VAL ("wrong exclude host" , !evsel->attr.exclude_host); + }return test__checkevent_symbolic_name (evlist); }
@@ -330,13 +360,14 @@ static int test__checkevent_symbolic_alias_modifier(struct evlist *evlist)static int test__checkevent_genhw_modifier (struct evlist *evlist) { -struct evsel *evsel =evlist__first (evlist); - -TEST_ASSERT_VAL ("wrong exclude_user" , evsel->core.attr.exclude_user); -TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel); -TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->core.attr.exclude_hv); -TEST_ASSERT_VAL ("wrong precise_ip" , evsel->core.attr.precise_ip); +struct perf_evsel *evsel; +perf_evlist__for_each_entry (&evlist->core, evsel) { +TEST_ASSERT_VAL ("wrong exclude_user" , evsel->attr.exclude_user); +TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->attr.exclude_kernel); +TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->attr.exclude_hv); +TEST_ASSERT_VAL ("wrong precise_ip" , evsel->attr.precise_ip); + }return test__checkevent_genhw (evlist); }
@@ -466,21 +497,23 @@ static int test__checkevent_list(struct evlist *evlist) {struct evsel *evsel =evlist__first (evlist); -TEST_ASSERT_VAL ("wrong number of entries" ,3 == evlist->core.nr_entries); +TEST_ASSERT_VAL ("wrong number of entries" ,3 <= evlist->core.nr_entries);/* r1 */ -TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_RAW == evsel->core.attr.type); -TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,1 )); -TEST_ASSERT_VAL ("wrong config1" ,0 == evsel->core.attr.config1); -TEST_ASSERT_VAL ("wrong config2" ,0 == evsel->core.attr.config2); -TEST_ASSERT_VAL ("wrong config3" ,0 == evsel->core.attr.config3); -TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user); -TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel); -TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv); -TEST_ASSERT_VAL ("wrong precise_ip" , !evsel->core.attr.precise_ip); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_TRACEPOINT != evsel->core.attr.type); +while (evsel->core.attr.type !=PERF_TYPE_TRACEPOINT ) { +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,1 )); +TEST_ASSERT_VAL ("wrong config1" ,0 == evsel->core.attr.config1); +TEST_ASSERT_VAL ("wrong config2" ,0 == evsel->core.attr.config2); +TEST_ASSERT_VAL ("wrong config3" ,0 == evsel->core.attr.config3); +TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->core.attr.exclude_user); +TEST_ASSERT_VAL ("wrong exclude_kernel" , !evsel->core.attr.exclude_kernel); +TEST_ASSERT_VAL ("wrong exclude_hv" , !evsel->core.attr.exclude_hv); +TEST_ASSERT_VAL ("wrong precise_ip" , !evsel->core.attr.precise_ip); + evsel =evsel__next (evsel); + }/* syscalls:sys_enter_openat:k */ - evsel =evsel__next (evsel);TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_TRACEPOINT == evsel->core.attr.type);TEST_ASSERT_VAL ("wrong sample_type" ,PERF_TP_SAMPLE_TYPE == evsel->core.attr.sample_type);
@@ -1916,7 +1949,7 @@ static int test_event(const struct evlist_test *e) e->name, ret, err.str);parse_events_error__print (&err, e->name); ret =TEST_FAIL ; -if (strstr (err.str,"can't access trace events" )) +if (err.str &&strstr (err.str,"can't access trace events" )) ret =TEST_SKIP ; }else { ret = e->check (evlist);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 35/46 perf parse-events: Support hardware events as terms
In reply to Ian RogersAn event like "cpu/instructions/" typically parses due to there being a sysfs event called instructions. On hybrid recursive parsing means that the hardware event is encoded in the attribute, with the PMU being placed in the high bits of the config:
'''
$ perf stat -vv -e 'cpu_core/cycles/' true ... ------------------------------------------------------------ perf_event_attr: size 136 config 0x400000000 sample_type IDENTIFIER read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING disabled 1 inherit 1 enable_on_exec 1 exclude_guest 1 ------------------------------------------------------------ '''
Make this behavior the default by adding a new term type and token for hardware events. The token gathers both the numeric config and the parsed name, so that if the token appears like "cycles/name=cycles/" then the token can be handled like a name. The numeric value isn't sufficient to distinguish say "cpu-cycles" from "cycles".
Extend the parse-events test so that all current non-PMU hardware parsing tests, also test with the PMU cpu - more than half the change.
--- tools/perf/tests/parse-events.c | 126 ++++++++++++++++++++++++++++++++ tools/perf/util/parse-events.c | 37 +++------- tools/perf/util/parse-events.h | 3 +- tools/perf/util/parse-events.l | 20 +++++ tools/perf/util/parse-events.y | 34 +++++++-- 5 files changed, 187 insertions(+), 33 deletions(-)
tools/perf/tests/parse-events.c +126 -0
@@ -1912,6 +1912,132 @@ static const struct evlist_test test__events_pmu[] = { .check = test__checkevent_config_cache, /* 8 */ }, + { + .name = "cpu/instructions/", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_symbolic_name, + /* 9 */ + }, + { + .name = "cpu/cycles,period=100000,config2/", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_symbolic_name_config, + /* 0 */ + }, + { + .name = "cpu/instructions/h", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_symbolic_name_modifier, + /* 1 */ + }, + { + .name = "cpu/instructions/G", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_exclude_host_modifier, + /* 2 */ + }, + { + .name = "cpu/instructions/H", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_exclude_guest_modifier, + /* 3 */ + }, + { + .name = "{cpu/instructions/k,cpu/cycles/upp}", + .valid = test__pmu_cpu_valid, + .check = test__group1, + /* 4 */ + }, + { + .name = "{cpu/cycles/u,cpu/instructions/kp}:p", + .valid = test__pmu_cpu_valid, + .check = test__group4, + /* 5 */ + }, + { + .name = "{cpu/cycles/,cpu/cache-misses/G}:H", + .valid = test__pmu_cpu_valid, + .check = test__group_gh1, + /* 6 */ + }, + { + .name = "{cpu/cycles/,cpu/cache-misses/H}:G", + .valid = test__pmu_cpu_valid, + .check = test__group_gh2, + /* 7 */ + }, + { + .name = "{cpu/cycles/G,cpu/cache-misses/H}:u", + .valid = test__pmu_cpu_valid, + .check = test__group_gh3, + /* 8 */ + }, + { + .name = "{cpu/cycles/G,cpu/cache-misses/H}:uG", + .valid = test__pmu_cpu_valid, + .check = test__group_gh4, + /* 9 */ + }, + { + .name = "{cpu/cycles/,cpu/cache-misses/,cpu/branch-misses/}:S", + .valid = test__pmu_cpu_valid, + .check = test__leader_sample1, + /* 0 */ + }, + { + .name = "{cpu/instructions/,cpu/branch-misses/}:Su", + .valid = test__pmu_cpu_valid, + .check = test__leader_sample2, + /* 1 */ + }, + { + .name = "cpu/instructions/uDp", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_pinned_modifier, + /* 2 */ + }, + { + .name = "{cpu/cycles/,cpu/cache-misses/,cpu/branch-misses/}:D", + .valid = test__pmu_cpu_valid, + .check = test__pinned_group, + /* 3 */ + }, + { + .name = "cpu/instructions/I", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_exclude_idle_modifier, + /* 4 */ + }, + { + .name = "cpu/instructions/kIG", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_exclude_idle_modifier_1, + /* 5 */ + }, + { + .name = "cpu/cycles/u", + .valid = test__pmu_cpu_valid, + .check = test__sym_event_slash, + /* 6 */ + }, + { + .name = "cpu/cycles/k", + .valid = test__pmu_cpu_valid, + .check = test__sym_event_dc, + /* 7 */ + }, + { + .name = "cpu/instructions/uep", + .valid = test__pmu_cpu_valid, + .check = test__checkevent_exclusive_modifier, + /* 8 */ + }, + { + .name = "{cpu/cycles/,cpu/cache-misses/,cpu/branch-misses/}:e", + .valid = test__pmu_cpu_valid, + .check = test__exclusive_group, + /* 9 */ + }, }; struct terms_test {
tools/perf/util/parse-events.c +12 -25
@@ -1052,6 +1052,7 @@ static const char *config_term_names[__PARSE_EVENTS__TERM_TYPE_NR] = { [PARSE_EVENTS__TERM_TYPE_METRIC_ID] = "metric-id", [PARSE_EVENTS__TERM_TYPE_RAW] = "raw", [PARSE_EVENTS__TERM_TYPE_LEGACY_CACHE] = "legacy-cache", + [PARSE_EVENTS__TERM_TYPE_HARDWARE] = "hardware", };static bool config_term_shrinked;
@@ -1239,6 +1240,17 @@ static int config_term_pmu(struct perf_event_attr *attr, }else term->type_term =PARSE_EVENTS__TERM_TYPE_USER ; } +if (term->type_term ==PARSE_EVENTS__TERM_TYPE_HARDWARE ) { +const struct perf_pmu *pmu =perf_pmu__find_by_type (attr->type); + +if (!pmu) { +pr_debug ("Failed to find PMU for type %d" , attr->type); +return -EINVAL ; + } + attr->type =PERF_TYPE_HARDWARE ; + attr->config = ((__u64 )pmu->type <<PERF_PMU_TYPE_SHIFT ) | term->val.num; +return 0 ; + }if (term->type_term ==PARSE_EVENTS__TERM_TYPE_USER || term->type_term ==PARSE_EVENTS__TERM_TYPE_DRV_CFG ) {/*
@@ -2569,31 +2581,6 @@ int parse_events_term__str(struct parse_events_term **term,return new_term (term, &temp, str,0 ); } -int parse_events_term__sym_hw (struct parse_events_term **term, -char *config,unsigned idx) -{ -struct event_symbol *sym; -char *str; -struct parse_events_term temp = { - .type_val =PARSE_EVENTS__TERM_TYPE_STR , - .type_term =PARSE_EVENTS__TERM_TYPE_USER , - .config = config, - }; - -if (!temp.config) { - temp.config =strdup ("event" ); -if (!temp.config) -return -ENOMEM ; - } -BUG_ON (idx >=PERF_COUNT_HW_MAX ); - sym = &event_symbols_hw[idx]; - - str =strdup (sym->symbol); -if (!str) -return -ENOMEM ; -return new_term (term, &temp, str,0 ); -} -int parse_events_term__clone (struct parse_events_term **new,struct parse_events_term *term) {
tools/perf/util/parse-events.h +1 -2
@@ -79,6 +79,7 @@ enum {PARSE_EVENTS__TERM_TYPE_METRIC_ID ,PARSE_EVENTS__TERM_TYPE_RAW ,PARSE_EVENTS__TERM_TYPE_LEGACY_CACHE , +PARSE_EVENTS__TERM_TYPE_HARDWARE , __PARSE_EVENTS__TERM_TYPE_NR, };
@@ -147,8 +148,6 @@ int parse_events_term__num(struct parse_events_term **term,int parse_events_term__str (struct parse_events_term **term,int type_term,char *config,char *str,void *loc_term,void *loc_val); -int parse_events_term__sym_hw (struct parse_events_term **term, -char *config,unsigned idx);int parse_events_term__clone (struct parse_events_term **new,struct parse_events_term *term);void parse_events_term__delete (struct parse_events_term *term);
tools/perf/util/parse-events.l +20 -0
@@ -149,6 +149,16 @@ static int term(yyscan_t scanner, int type) return PE_TERM; } +static int hw_term(yyscan_t scanner, int config) +{ + YYSTYPE *yylval = parse_events_get_lval(scanner); + char *text = parse_events_get_text(scanner); + + yylval->hardware_term.str = strdup(text); + yylval->hardware_term.num = PERF_TYPE_HARDWARE + config; + return PE_TERM_HW; +} + #define YY_USER_ACTION \ do { \ yylloc->last_column = yylloc->first_column; \
@@ -269,6 +279,16 @@ percore { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_PERCORE); } aux-output { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_AUX_OUTPUT); } aux-sample-size { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_AUX_SAMPLE_SIZE); } metric-id { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_METRIC_ID); } +cpu-cycles|cycles { return hw_term(yyscanner, PERF_COUNT_HW_CPU_CYCLES); } +stalled-cycles-frontend|idle-cycles-frontend { return hw_term(yyscanner, PERF_COUNT_HW_STALLED_CYCLES_FRONTEND); } +stalled-cycles-backend|idle-cycles-backend { return hw_term(yyscanner, PERF_COUNT_HW_STALLED_CYCLES_BACKEND); } +instructions { return hw_term(yyscanner, PERF_COUNT_HW_INSTRUCTIONS); } +cache-references { return hw_term(yyscanner, PERF_COUNT_HW_CACHE_REFERENCES); } +cache-misses { return hw_term(yyscanner, PERF_COUNT_HW_CACHE_MISSES); } +branch-instructions|branches { return hw_term(yyscanner, PERF_COUNT_HW_BRANCH_INSTRUCTIONS); } +branch-misses { return hw_term(yyscanner, PERF_COUNT_HW_BRANCH_MISSES); } +bus-cycles { return hw_term(yyscanner, PERF_COUNT_HW_BUS_CYCLES); } +ref-cycles { return hw_term(yyscanner, PERF_COUNT_HW_REF_CPU_CYCLES); } r{num_raw_hex} { return str(yyscanner, PE_RAW); } r0x{num_raw_hex} { return str(yyscanner, PE_RAW); } , { return ','; }
tools/perf/util/parse-events.y +28 -6
@@ -65,6 +65,7 @@ static void free_list_evsel(struct list_head* list_evsel) %token PE_KERNEL_PMU_EVENT PE_PMU_EVENT_FAKE %token PE_ARRAY_ALL PE_ARRAY_RANGE %token PE_DRV_CFG_TERM +%token PE_TERM_HW %type <num> PE_VALUE %type <num> PE_VALUE_SYM_HW %type <num> PE_VALUE_SYM_SW
@@ -112,6 +113,8 @@ static void free_list_evsel(struct list_head* list_evsel) %type <array> array_term %type <array> array_terms %destructor { free ($$.ranges); } <array> +%type <hardware_term> PE_TERM_HW +%destructor { free ($$.str); } <hardware_term> %union {
@@ -125,6 +128,10 @@ static void free_list_evsel(struct list_head* list_evsel) char *event; } tracepoint_name; struct parse_events_array array; + struct hardware_term { + char *str; + u64 num; + } hardware_term; } %%
@@ -770,13 +777,14 @@ name_or_raw '=' PE_VALUE $$ = term; } | -name_or_raw '=' PE_VALUE_SYM_HW +name_or_raw '=' PE_TERM_HW { struct parse_events_term *term; - int config = $3 & 255; - if (parse_events_term__sym_hw(&term, $1, config)) { + if (parse_events_term__str(&term, PARSE_EVENTS__TERM_TYPE_USER, + $1, $3.str, &@1, &@3)) { free($1); + free($3.str); YYABORT; } $$ = term;
@@ -806,12 +814,15 @@ PE_NAME $$ = term; } | -PE_VALUE_SYM_HW +PE_TERM_HW { struct parse_events_term *term; - int config = $1 & 255; - ABORT_ON(parse_events_term__sym_hw(&term, NULL, config)); + if (parse_events_term__num(&term, PARSE_EVENTS__TERM_TYPE_HARDWARE, + $1.str, $1.num & 255, false, &@1, NULL)) { + free($1.str); + YYABORT; + } $$ = term; } |
@@ -826,6 +837,17 @@ PE_TERM '=' PE_NAME $$ = term; } | +PE_TERM '=' PE_TERM_HW +{ + struct parse_events_term *term; + + if (parse_events_term__str(&term, (int)$1, NULL, $3.str, &@1, &@3)) { + free($3.str); + YYABORT; + } + $$ = term; +} +| PE_TERM '=' PE_VALUE { struct parse_events_term *term;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Re: [PATCH v3 35/46] perf parse-events: Support hardware events as terms
Quoted 4 lines
tools/perf/util/parse-events.l +0
-0@@ -269,6 +279,16 @@ percore { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_PERCORE); } aux-output { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_AUX_OUTPUT); } aux-sample-size { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_AUX_SAMPLE_SIZE); } metric-id { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_METRIC_ID); }
tools/perf/util/parse-events.l percore { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_PERCORE); } · line 291
Quoted 10 lines
tools/perf/util/parse-events.l +10
-0+cpu-cycles|cycles { return hw_term(yyscanner, PERF_COUNT_HW_CPU_CYCLES); } +stalled-cycles-frontend|idle-cycles-frontend { return hw_term(yyscanner, PERF_COUNT_HW_STALLED_CYCLES_FRONTEND); } +stalled-cycles-backend|idle-cycles-backend { return hw_term(yyscanner, PERF_COUNT_HW_STALLED_CYCLES_BACKEND); } +instructions { return hw_term(yyscanner, PERF_COUNT_HW_INSTRUCTIONS); } +cache-references { return hw_term(yyscanner, PERF_COUNT_HW_CACHE_REFERENCES); } +cache-misses { return hw_term(yyscanner, PERF_COUNT_HW_CACHE_MISSES); } +branch-instructions|branches { return hw_term(yyscanner, PERF_COUNT_HW_BRANCH_INSTRUCTIONS); } +branch-misses { return hw_term(yyscanner, PERF_COUNT_HW_BRANCH_MISSES); } +bus-cycles { return hw_term(yyscanner, PERF_COUNT_HW_BUS_CYCLES); } +ref-cycles { return hw_term(yyscanner, PERF_COUNT_HW_REF_CPU_CYCLES); }
These are generic terms and thus can be added to _any_ pmus. Ex:
Before:
```
$ sudo ./perf stat -e amd_l3/cycles/ -C 0 -- sleep 1
event syntax error: 'amd_l3/cycles/'
\___ unknown term 'cycles' for pmu 'amd_l3'
```After:
```
$ sudo ./perf stat -e amd_l3/cycles/ -C 0 -vv -- sleep 1
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
size 136
config 0xb00000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8
sys_perf_event_open failed, error -2
Warning:
amd_l3/cycles/ event is not supported by the kernel.
failed to read counter amd_l3/cycles/
Performance counter stats for 'CPU(s) 0':
<not supported> amd_l3/cycles/
1.059391819 seconds time elapsed
```Here type=0 which means perf is trying to open event on HW pmu instead of amd_l3//. config is 0xb00000000 where 0xb is amd_l3 pmu type:
$ cat /sys/bus/event_source/devices/amd_l3/type 11
Not sure if this is expected behavior.
Ian Rogers
reviewerRe: [PATCH v3 35/46] perf parse-events: Support hardware events as terms
Quoted 52 lines · Ravi Bangoria
Quoted 5 lines
tools/perf/util/parse-events.l +0
-0@@ -269,6 +279,16 @@ percore { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_PERCORE); } aux-output { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_AUX_OUTPUT); } aux-sample-size { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_AUX_SAMPLE_SIZE); } metric-id { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_METRIC_ID); }tools/perf/util/parse-events.l percore { return term(yyscanner, PARSE_EVENTS__TERM_TYPE_PERCORE); } · line 291
Quoted 10 lines
tools/perf/util/parse-events.l +10
-0+cpu-cycles|cycles { return hw_term(yyscanner, PERF_COUNT_HW_CPU_CYCLES); } +stalled-cycles-frontend|idle-cycles-frontend { return hw_term(yyscanner, PERF_COUNT_HW_STALLED_CYCLES_FRONTEND); } +stalled-cycles-backend|idle-cycles-backend { return hw_term(yyscanner, PERF_COUNT_HW_STALLED_CYCLES_BACKEND); } +instructions { return hw_term(yyscanner, PERF_COUNT_HW_INSTRUCTIONS); } +cache-references { return hw_term(yyscanner, PERF_COUNT_HW_CACHE_REFERENCES); } +cache-misses { return hw_term(yyscanner, PERF_COUNT_HW_CACHE_MISSES); } +branch-instructions|branches { return hw_term(yyscanner, PERF_COUNT_HW_BRANCH_INSTRUCTIONS); } +branch-misses { return hw_term(yyscanner, PERF_COUNT_HW_BRANCH_MISSES); } +bus-cycles { return hw_term(yyscanner, PERF_COUNT_HW_BUS_CYCLES); } +ref-cycles { return hw_term(yyscanner, PERF_COUNT_HW_REF_CPU_CYCLES); }These are generic terms and thus can be added to _any_ pmus. Ex:
Before:
``` $ sudo ./perf stat -e amd_l3/cycles/ -C 0 -- sleep 1 event syntax error: 'amd_l3/cycles/' \___ unknown term 'cycles' for pmu 'amd_l3' ```After:
``` $ sudo ./perf stat -e amd_l3/cycles/ -C 0 -vv -- sleep 1 Control descriptor is not initialized ------------------------------------------------------------ perf_event_attr: size 136 config 0xb00000000 sample_type IDENTIFIER read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING disabled 1 inherit 1 exclude_guest 1 ------------------------------------------------------------ sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 sys_perf_event_open failed, error -2 Warning: amd_l3/cycles/ event is not supported by the kernel. failed to read counter amd_l3/cycles/ Performance counter stats for 'CPU(s) 0': <not supported> amd_l3/cycles/ 1.059391819 seconds time elapsed ```
Quoted 7 lines
Here
type=0which means perf is trying to open event on HW pmu instead of amd_l3//.configis0xb00000000where0xbisamd_l3pmu type:$ cat /sys/bus/event_source/devices/amd_l3/type 11Not sure if this is expected behavior.
I agree this is definitely a change, I'm not sure what we can do to handle this better. Firstly, let's not be overly negative, the code fixes it so that "branch-brs" (an AMD sysfs event) is now a valid event name rather than a parse error. For the example above we have the case of an unsupported event now not giving a parse error, only possible as all PMUs were scanned, but giving an open error as the event name matches in the list of hardware events but the extended type encoding of the PMU yields a PMU that can't handle it.
I think parsing should be concerned with taking information to yield a struct perf_event_attr. When it gets into the game of testing PMUs for support then we easily get test regressions, as the test system may lack the PMUs that are part of the test. We also get issues on hybrid like PERF_TYPE_HW_CACHE events being supported by one PMU and not the other. The existing hybrid behavior is to silently skip the events that aren't supported - tested with a perf_event_open. The behavior after these patches is to open the event on both PMUs but display not supported/counted generally on the atom PMU. So the coding decision there was to say parsing and opening are separate stages, and the correct place to fail was in the open.
We could change the event parsing for this case, have different parser notions of core PMU and every PMU, then only allow the hardware events on the core PMU. This would require hard coding a core PMU list into the parser and I'm not keen on that.
Thanks,
Ian
Ian Rogers
reviewerPATCH v3 36/46 perf parse-events: Avoid error when assigning a term
In reply to Ian RogersAvoid the parser error:
'''
$ perf stat -e 'cycles/name=name/' true
event syntax error: 'cycles/name=name/'
\___ parser error
'''
by turning the term back to a string if it is on the right. Add PMU
and generic parsing tests.--- tools/perf/tests/parse-events.c | 21 +++++++++++++++++++++ tools/perf/util/parse-events.c | 9 +++++++++ tools/perf/util/parse-events.h | 3 +++ tools/perf/util/parse-events.y | 8 ++++++++ 4 files changed, 41 insertions(+)
tools/perf/tests/parse-events.c +21 -0
@@ -1485,6 +1485,16 @@ static int test__sym_event_dc(struct evlist *evlist)return TEST_OK ; } +static int test__term_equal_term (struct evlist *evlist) +{ +struct evsel *evsel =evlist__first (evlist); + +TEST_ASSERT_VAL ("wrong type" , evsel->core.attr.type ==PERF_TYPE_HARDWARE ); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES )); +TEST_ASSERT_VAL ("wrong name setting" ,strcmp (evsel->name,"name" ) ==0 ); +return TEST_OK ; +} +#ifdef HAVE_LIBTRACEEVENT static int count_tracepoints (void ) {
@@ -1857,6 +1867,11 @@ static const struct evlist_test test__events[] = { .check = test__exclusive_group, /* 7 */ }, + { + .name = "cycles/name=name/", + .check = test__term_equal_term, + /* 8 */ + }, };static const struct evlist_test test__events_pmu[] = {
@@ -2038,6 +2053,12 @@ static const struct evlist_test test__events_pmu[] = { .check = test__exclusive_group, /* 9 */ }, + { + .name = "cpu/cycles,name=name/", + .valid = test__pmu_cpu_valid, + .check = test__term_equal_term, + /* 0 */ + }, };struct terms_test {
tools/perf/util/parse-events.c +9 -0
@@ -2581,6 +2581,15 @@ int parse_events_term__str(struct parse_events_term **term,return new_term (term, &temp, str,0 ); } +int parse_events_term__term (struct parse_events_term **term, +int term_lhs,int term_rhs, +void *loc_term,void *loc_val) +{ +return parse_events_term__str (term, term_lhs,NULL , +strdup (config_term_names[term_rhs]), + loc_term, loc_val); +} +int parse_events_term__clone (struct parse_events_term **new,struct parse_events_term *term) {
tools/perf/util/parse-events.h +3 -0
@@ -148,6 +148,9 @@ int parse_events_term__num(struct parse_events_term **term,int parse_events_term__str (struct parse_events_term **term,int type_term,char *config,char *str,void *loc_term,void *loc_val); +int parse_events_term__term (struct parse_events_term **term, +int term_lhs,int term_rhs, +void *loc_term,void *loc_val);int parse_events_term__clone (struct parse_events_term **new,struct parse_events_term *term);void parse_events_term__delete (struct parse_events_term *term);
tools/perf/util/parse-events.y +8 -0
@@ -848,6 +848,14 @@ PE_TERM '=' PE_TERM_HW $$ = term; } | +PE_TERM '=' PE_TERM +{ + struct parse_events_term *term; + + ABORT_ON(parse_events_term__term(&term, (int)$1, (int)$3, &@1, &@3)); + $$ = term; +} +| PE_TERM '=' PE_VALUE { struct parse_events_term *term;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 37/46 perf parse-events: Avoid error when assigning a legacy cache term
In reply to Ian RogersAvoid the parser error:
'''
$ perf stat -e 'cycles/name=l1d/' true
event syntax error: 'cycles/name=l1d/'
\___ parser error
'''
by combining the name and legacy cache cases in the parser.--- tools/perf/tests/parse-events.c | 21 +++++++++++++++++++++ tools/perf/util/parse-events.y | 10 ++++++---- 2 files changed, 27 insertions(+), 4 deletions(-)
tools/perf/tests/parse-events.c +21 -0
@@ -1495,6 +1495,16 @@ static int test__term_equal_term(struct evlist *evlist)return TEST_OK ; } +static int test__term_equal_legacy (struct evlist *evlist) +{ +struct evsel *evsel =evlist__first (evlist); + +TEST_ASSERT_VAL ("wrong type" , evsel->core.attr.type ==PERF_TYPE_HARDWARE ); +TEST_ASSERT_VAL ("wrong config" ,test_config (evsel,PERF_COUNT_HW_CPU_CYCLES )); +TEST_ASSERT_VAL ("wrong name setting" ,strcmp (evsel->name,"l1d" ) ==0 ); +return TEST_OK ; +} +#ifdef HAVE_LIBTRACEEVENT static int count_tracepoints (void ) {
@@ -1872,6 +1882,11 @@ static const struct evlist_test test__events[] = { .check = test__term_equal_term, /* 8 */ }, + { + .name = "cycles/name=l1d/", + .check = test__term_equal_legacy, + /* 9 */ + }, };static const struct evlist_test test__events_pmu[] = {
@@ -2059,6 +2074,12 @@ static const struct evlist_test test__events_pmu[] = { .check = test__term_equal_term, /* 0 */ }, + { + .name = "cpu/cycles,name=l1d/", + .valid = test__pmu_cpu_valid, + .check = test__term_equal_legacy, + /* 1 */ + }, };struct terms_test {
tools/perf/util/parse-events.y +6 -4
@@ -82,7 +82,7 @@ static void free_list_evsel(struct list_head* list_evsel) %type <str> PE_EVENT_NAME %type <str> PE_KERNEL_PMU_EVENT PE_PMU_EVENT_FAKE %type <str> PE_DRV_CFG_TERM -%type <str> name_or_raw +%type <str> name_or_raw name_or_legacy %destructor { free ($$); } <str> %type <term> event_term %destructor { parse_events_term__delete ($$); } <term>
@@ -739,6 +739,8 @@ event_term name_or_raw: PE_RAW | PE_NAME | PE_LEGACY_CACHE +name_or_legacy: PE_NAME | PE_LEGACY_CACHE + event_term: PE_RAW {
@@ -752,7 +754,7 @@ PE_RAW $$ = term; } | -name_or_raw '=' PE_NAME +name_or_raw '=' name_or_legacy { struct parse_events_term *term;
@@ -826,7 +828,7 @@ PE_TERM_HW $$ = term; } | -PE_TERM '=' PE_NAME +PE_TERM '=' name_or_legacy { struct parse_events_term *term;
@@ -872,7 +874,7 @@ PE_TERM $$ = term; } | -name_or_raw array '=' PE_NAME +name_or_raw array '=' name_or_legacy { struct parse_events_term *term;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 38/46 perf parse-events: Don't auto merge hybrid wildcard events
In reply to Ian RogersBring back the behavior of not auto-merging hybrid events by delegating to a test in pmu.
--- tools/perf/util/parse-events.c | 5 ++++- tools/perf/util/parse-events.y | 4 +++- tools/perf/util/pmu.c | 5 +++++ tools/perf/util/pmu.h | 1 + 4 files changed, 13 insertions(+), 2 deletions(-)
tools/perf/util/parse-events.c +4 -1
@@ -1714,16 +1714,19 @@ int parse_events_multi_pmu_add(struct parse_events_state *parse_state,while ((pmu =perf_pmu__scan (pmu)) !=NULL ) {struct perf_pmu_alias *alias; +bool auto_merge_stats;if (parse_events__filter_pmu (parse_state, pmu))continue ; + auto_merge_stats =perf_pmu__auto_merge_stats (pmu); +list_for_each_entry (alias, &pmu->aliases, list) {if (!strcasecmp (alias->name, str)) {parse_events_copy_term_list (head, &orig_head);if (!parse_events_add_pmu (parse_state, list, pmu->name, orig_head, -/*auto_merge_stats=*/ true)) { + auto_merge_stats)) {pr_debug ("%s -> %s/%s/\n" , str, pmu->name, alias->str); ok++;
tools/perf/util/parse-events.y +3 -1
@@ -327,10 +327,12 @@ PE_NAME opt_pmu_config name += 7; if (!perf_pmu__match(pattern, name, $1) || !perf_pmu__match(pattern, pmu->alias_name, $1)) { + bool auto_merge_stats = perf_pmu__auto_merge_stats(pmu); + if (parse_events_copy_term_list(orig_terms, &terms)) CLEANUP_YYABORT; if (!parse_events_add_pmu(parse_state, list, pmu->name, terms, - /*auto_merge_stats=*/true)) { + auto_merge_stats)) { ok++; parse_state->wild_card_pmus = true; }
tools/perf/util/pmu.c +5 -0
@@ -1660,6 +1660,11 @@ bool perf_pmu__supports_wildcard_numeric(const struct perf_pmu *pmu)return is_pmu_core (pmu->name) ||perf_pmu__is_hybrid (pmu->name); } +bool perf_pmu__auto_merge_stats (const struct perf_pmu *pmu) +{ +return !perf_pmu__is_hybrid (pmu->name); +} +static bool pmu_alias_is_duplicate (struct sevent *alias_a,struct sevent *alias_b) {
tools/perf/util/pmu.h +1 -0
@@ -222,6 +222,7 @@ struct perf_pmu *perf_pmu__scan(struct perf_pmu *pmu);bool is_pmu_core (const char *name);bool perf_pmu__supports_legacy_cache (const struct perf_pmu *pmu);bool perf_pmu__supports_wildcard_numeric (const struct perf_pmu *pmu); +bool perf_pmu__auto_merge_stats (const struct perf_pmu *pmu);void print_pmu_events (const struct print_callbacks *print_cb,void *print_state);bool pmu_have_event (const char *pname,const char *name);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 39/46 perf parse-events: Don't reorder atom cpu events
In reply to Ian RogersOn hybrid systems the topdown events don't share a fixed counter on the atom core, so they don't require the sorting the perf metric supporting PMUs do.
--- tools/perf/arch/x86/util/evlist.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/arch/x86/util/evlist.c +2 -2
@@ -6,6 +6,7 @@#include "util/event.h" #include "util/pmu-hybrid.h" #include "topdown.h" +#include "evsel.h" static int ___evlist__add_default_attrs(struct evlist *evlist, struct perf_event_attr *attrs,
@@ -67,8 +68,7 @@ int arch_evlist__add_default_attrs(struct evlist *evlist,int arch_evlist__cmp (const struct evsel *lhs,const struct evsel *rhs) { -if (topdown_sys_has_perf_metrics () && - (!lhs->pmu_name || !strncmp (lhs->pmu_name,"cpu" ,3 ))) { +if (topdown_sys_has_perf_metrics () &&evsel__sys_has_perf_metrics (lhs)) {/* Ensure the topdown slots comes first. */ if (strcasestr (lhs->name,"slots" ))return -1 ;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 40/46 perf metrics: Be PMU specific for referenced metrics.
In reply to Ian RogersHybrid systems may define the same metric for different PMUs, this can cause confusion of events. To avoid this make the referenced metric searches PMU specific, matching that in the table.
--- tools/perf/builtin-stat.c | 6 +- tools/perf/pmu-events/jevents.py | 4 +- tools/perf/pmu-events/pmu-events.h | 1 + tools/perf/util/metricgroup.c | 94 +++++++++++++++++++++--------- tools/perf/util/metricgroup.h | 2 +- 5 files changed, 75 insertions(+), 32 deletions(-)
tools/perf/builtin-stat.c +3 -3
@@ -1794,7 +1794,7 @@ static int add_default_attributes(void)* will use this approach. To determine transaction support * on an architecture test for such a metric name. */ -if (!metricgroup__has_metric ("transaction" )) { +if (!metricgroup__has_metric ("all" ,"transaction" )) {pr_err ("Missing transaction metrics" );return -1 ; }
@@ -1823,7 +1823,7 @@ static int add_default_attributes(void) smi_reset = true; } -if (!metricgroup__has_metric ("smi" )) { +if (!metricgroup__has_metric ("all" ,"smi" )) {pr_err ("Missing smi metrics" );return -1 ; }
@@ -1903,7 +1903,7 @@ static int add_default_attributes(void)* caused by exposing latent bugs. This is fixed properly in: * https://lore.kernel.org/lkml/[email protected]/ */ -if (metricgroup__has_metric ("TopdownL1" ) && !perf_pmu__has_hybrid ()) { +if (metricgroup__has_metric ("all" ,"TopdownL1" ) && !perf_pmu__has_hybrid ()) {struct evlist *metric_evlist =evlist__new ();struct evsel *metric_evsel;
tools/perf/pmu-events/jevents.py +2 -2
@@ -51,8 +51,8 @@ _json_event_attributes = [ # Attributes that are in pmu_metric rather than pmu_event. _json_metric_attributes = [ - 'metric_name', 'metric_group', 'metric_expr', 'metric_threshold', 'desc', - 'long_desc', 'unit', 'compat', 'metricgroup_no_group', 'aggr_mode', + 'pmu', 'metric_name', 'metric_group', 'metric_expr', 'metric_threshold', + 'desc', 'long_desc', 'unit', 'compat', 'metricgroup_no_group', 'aggr_mode', 'event_grouping' ] # Attributes that are bools or enum int values, encoded as '0', '1',...
tools/perf/pmu-events/pmu-events.h +1 -0
@@ -51,6 +51,7 @@ struct pmu_event { };struct pmu_metric { +const char *pmu;const char *metric_name;const char *metric_group;const char *metric_expr;
tools/perf/util/metricgroup.c +68 -26
309 lines not on this page · raw message
tools/perf/util/metricgroup.h +1 -1
@@ -80,7 +80,7 @@ int metricgroup__parse_groups_test(struct evlist *evlist, struct rblist *metric_events);void metricgroup__print (const struct print_callbacks *print_cb,void *print_state); -bool metricgroup__has_metric (const char *metric); +bool metricgroup__has_metric (const char *pmu,const char *metric);unsigned int metricgroups__topdown_max_level (void );int arch_get_runtimeparam (const struct pmu_metric *pm);void metricgroup__rblist_exit (struct rblist *metric_events);
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 41/46 perf stat: Command line PMU metric filtering
In reply to Ian RogersWire up the --cputype value to limit which metrics are parsed.
--- tools/perf/builtin-stat.c | 20 ++++++++++++-------- tools/perf/util/metricgroup.c | 3 ++- tools/perf/util/metricgroup.h | 1 + 3 files changed, 15 insertions(+), 9 deletions(-)
tools/perf/builtin-stat.c +12 -8
@@ -1783,6 +1783,7 @@ static int add_default_attributes(void) };struct perf_event_attr default_null_attrs[] = {}; +const char *pmu = parse_events_option_args.pmu_filter ?:"all" ;/* Set attrs if no event is selected and !null_run: */ if (stat_config.null_run)
@@ -1794,11 +1795,11 @@ static int add_default_attributes(void)* will use this approach. To determine transaction support * on an architecture test for such a metric name. */ -if (!metricgroup__has_metric ("all" ,"transaction" )) { +if (!metricgroup__has_metric (pmu,"transaction" )) {pr_err ("Missing transaction metrics" );return -1 ; } - return metricgroup__parse_groups(evsel_list, "transaction", + return metricgroup__parse_groups(evsel_list, pmu, "transaction", stat_config.metric_no_group, stat_config.metric_no_merge, stat_config.metric_no_threshold,
@@ -1823,7 +1824,7 @@ static int add_default_attributes(void) smi_reset = true; } -if (!metricgroup__has_metric ("all" ,"smi" )) { +if (!metricgroup__has_metric (pmu,"smi" )) {pr_err ("Missing smi metrics" );return -1 ; }
@@ -1831,7 +1832,7 @@ static int add_default_attributes(void)if (!force_metric_only) stat_config.metric_only = true; - return metricgroup__parse_groups(evsel_list, "smi", + return metricgroup__parse_groups(evsel_list, pmu, "smi", stat_config.metric_no_group, stat_config.metric_no_merge, stat_config.metric_no_threshold,
@@ -1864,7 +1865,8 @@ static int add_default_attributes(void) "Please print the result regularly, e.g. -I1000\n"); } str[8 ] = stat_config.topdown_level +'0' ; - if (metricgroup__parse_groups(evsel_list, str, + if (metricgroup__parse_groups(evsel_list, + pmu, str, /*metric_no_group=*/false, /*metric_no_merge=*/false, /*metric_no_threshold=*/true,
@@ -1903,14 +1905,14 @@ static int add_default_attributes(void)* caused by exposing latent bugs. This is fixed properly in: * https://lore.kernel.org/lkml/[email protected]/ */ -if (metricgroup__has_metric ("all" ,"TopdownL1" ) && !perf_pmu__has_hybrid ()) { +if (metricgroup__has_metric (pmu,"TopdownL1" ) && !perf_pmu__has_hybrid ()) {struct evlist *metric_evlist =evlist__new ();struct evsel *metric_evsel;if (!metric_evlist)return -1 ; - if (metricgroup__parse_groups(metric_evlist, "TopdownL1", + if (metricgroup__parse_groups(metric_evlist, pmu, "TopdownL1", /*metric_no_group=*/false, /*metric_no_merge=*/false, /*metric_no_threshold=*/true,
@@ -2434,7 +2436,9 @@ int cmd_stat(int argc, const char **argv)* knowing the target is system-wide. */ if (metrics) { - metricgroup__parse_groups(evsel_list, metrics, +const char *pmu = parse_events_option_args.pmu_filter ?:"all" ; + + metricgroup__parse_groups(evsel_list, pmu, metrics, stat_config.metric_no_group, stat_config.metric_no_merge, stat_config.metric_no_threshold,
tools/perf/util/metricgroup.c +2 -1
@@ -1648,6 +1648,7 @@ static int parse_groups(struct evlist *perf_evlist, } int metricgroup__parse_groups(struct evlist *perf_evlist, + const char *pmu, const char *str, bool metric_no_group, bool metric_no_merge,
@@ -1661,7 +1662,7 @@ int metricgroup__parse_groups(struct evlist *perf_evlist,if (!table)return -EINVAL ; -return parse_groups (perf_evlist,"all" , str, metric_no_group, metric_no_merge, +return parse_groups (perf_evlist, pmu, str, metric_no_group, metric_no_merge, metric_no_threshold, user_requested_cpu_list, system_wide,/*fake_pmu=*/ NULL , metric_events, table); }
tools/perf/util/metricgroup.h +1 -0
@@ -67,6 +67,7 @@ struct metric_event *metricgroup__lookup(struct rblist *metric_events,struct evsel *evsel, bool create); int metricgroup__parse_groups(struct evlist *perf_evlist, + const char *pmu, const char *str, bool metric_no_group, bool metric_no_merge,
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 42/46 perf vendor events intel: Correct alderlake metrics
In reply to Ian RogersFix the metrics tma_memory_bound on alderlake cpu_core and tma_microcode_sequencer on alderlake cpu_atom, where metrics had be rewritten across PMUs. Fix MEM_BOUND_STALLS_AT_RET_CORRECTION which is an aux metric but lacks a hash prefix. Add PMU prefixes for cpu_core/cpu_atom events to avoid wildcard opening the events.
--- .../arch/x86/alderlake/adl-metrics.json | 238 +++++++++--------- .../arch/x86/alderlaken/adln-metrics.json | 6 +- 2 files changed, 122 insertions(+), 122 deletions(-)
tools/perf/pmu-events/arch/x86/alderlake/adl-metrics.json +119 -119
@@ -151,7 +151,7 @@ }, { "BriefDescription": "Counts the total number of issue slots that were not consumed by the backend because allocation is stalled due to a mispredicted jump or a machine clear", - "MetricExpr": "(tma_info_slots - (TOPDOWN_FE_BOUND.ALL + TOPDOWN_BE_BOUND.ALL + TOPDOWN_RETIRING.ALL)) / tma_info_slots", + "MetricExpr": "(tma_info_slots - (cpu_atom@TOPDOWN_FE_BOUND.ALL@ + cpu_atom@TOPDOWN_BE_BOUND.ALL@ + cpu_atom@TOPDOWN_RETIRING.ALL@)) / tma_info_slots", "MetricGroup": "TopdownL1;tma_L1_group", "MetricName": "tma_bad_speculation", "MetricThreshold": "tma_bad_speculation > 0.15",
@@ -162,7 +162,7 @@ }, { "BriefDescription": "Counts the number of uops that are not from the microsequencer.", - "MetricExpr": "(TOPDOWN_RETIRING.ALL - UOPS_RETIRED.MS) / tma_info_slots", + "MetricExpr": "(cpu_atom@TOPDOWN_RETIRING.ALL@ - cpu_atom@UOPS_RETIRED.MS@) / tma_info_slots", "MetricGroup": "TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_base", "MetricThreshold": "tma_base > 0.6",
@@ -229,7 +229,7 @@ }, { "BriefDescription": "Counts the number of machine clears relative to the number of nuke slots due to memory disambiguation.", - "MetricExpr": "tma_nuke * (MACHINE_CLEARS.DISAMBIGUATION / MACHINE_CLEARS.SLOW)", + "MetricExpr": "tma_nuke * (cpu_atom@MACHINE_CLEARS.DISAMBIGUATION@ / cpu_atom@MACHINE_CLEARS.SLOW@)", "MetricGroup": "TopdownL4;tma_L4_group;tma_nuke_group", "MetricName": "tma_disambiguation", "MetricThreshold": "tma_disambiguation > 0.02",
@@ -239,7 +239,7 @@ { "BriefDescription": "Counts the number of cycles the core is stalled due to a demand load miss which hit in DRAM or MMIO (Non-DRAM).", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "MEM_BOUND_STALLS.LOAD_DRAM_HIT / tma_info_clks - MEM_BOUND_STALLS_AT_RET_CORRECTION * MEM_BOUND_STALLS.LOAD_DRAM_HIT / MEM_BOUND_STALLS.LOAD", + "MetricExpr": "cpu_atom@MEM_BOUND_STALLS.LOAD_DRAM_HIT@ / tma_info_clks - max((cpu_atom@MEM_BOUND_STALLS.LOAD@ - cpu_atom@LD_HEAD.L1_MISS_AT_RET@) / tma_info_clks, 0) * cpu_atom@MEM_BOUND_STALLS.LOAD_DRAM_HIT@ / cpu_atom@MEM_BOUND_STALLS.LOAD@", "MetricGroup": "TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_dram_bound", "MetricThreshold": "tma_dram_bound > 0.1",
@@ -277,7 +277,7 @@ }, { "BriefDescription": "Counts the number of machine clears relative to the number of nuke slots due to FP assists.", - "MetricExpr": "tma_nuke * (MACHINE_CLEARS.FP_ASSIST / MACHINE_CLEARS.SLOW)", + "MetricExpr": "tma_nuke * (cpu_atom@MACHINE_CLEARS.FP_ASSIST@ / cpu_atom@MACHINE_CLEARS.SLOW@)", "MetricGroup": "TopdownL4;tma_L4_group;tma_nuke_group", "MetricName": "tma_fp_assist", "MetricThreshold": "tma_fp_assist > 0.02",
@@ -314,7 +314,7 @@ }, { "BriefDescription": "Percentage of total non-speculative loads with a address aliasing block", - "MetricExpr": "100 * LD_BLOCKS.4K_ALIAS / MEM_UOPS_RETIRED.ALL_LOADS", + "MetricExpr": "100 * cpu_atom@LD_BLOCKS.4K_ALIAS@ / MEM_UOPS_RETIRED.ALL_LOADS", "MetricName": "tma_info_address_alias_blocks", "Unit": "cpu_atom" },
@@ -334,14 +334,14 @@ }, { "BriefDescription": "", - "MetricExpr": "CPU_CLK_UNHALTED.CORE", + "MetricExpr": "cpu_atom@CPU_CLK_UNHALTED.CORE@", "MetricGroup": " ", "MetricName": "tma_info_clks", "Unit": "cpu_atom" }, { "BriefDescription": "", - "MetricExpr": "CPU_CLK_UNHALTED.CORE_P", + "MetricExpr": "cpu_atom@CPU_CLK_UNHALTED.CORE_P@", "MetricGroup": " ", "MetricName": "tma_info_clks_p", "Unit": "cpu_atom"
@@ -383,35 +383,35 @@ }, { "BriefDescription": "Percentage of all uops which are FPDiv uops", - "MetricExpr": "100 * UOPS_RETIRED.FPDIV / UOPS_RETIRED.ALL", + "MetricExpr": "100 * cpu_atom@UOPS_RETIRED.FPDIV@ / UOPS_RETIRED.ALL", "MetricGroup": " ", "MetricName": "tma_info_fpdiv_uop_ratio", "Unit": "cpu_atom" }, { "BriefDescription": "Percentage of all uops which are IDiv uops", - "MetricExpr": "100 * UOPS_RETIRED.IDIV / UOPS_RETIRED.ALL", + "MetricExpr": "100 * cpu_atom@UOPS_RETIRED.IDIV@ / UOPS_RETIRED.ALL", "MetricGroup": " ", "MetricName": "tma_info_idiv_uop_ratio", "Unit": "cpu_atom" }, { "BriefDescription": "Percent of instruction miss cost that hit in DRAM", - "MetricExpr": "100 * MEM_BOUND_STALLS.IFETCH_DRAM_HIT / MEM_BOUND_STALLS.IFETCH", + "MetricExpr": "100 * cpu_atom@MEM_BOUND_STALLS.IFETCH_DRAM_HIT@ / cpu_atom@MEM_BOUND_STALLS.IFETCH@", "MetricGroup": " ", "MetricName": "tma_info_inst_miss_cost_dramhit_percent", "Unit": "cpu_atom" }, { "BriefDescription": "Percent of instruction miss cost that hit in the L2", - "MetricExpr": "100 * MEM_BOUND_STALLS.IFETCH_L2_HIT / MEM_BOUND_STALLS.IFETCH", + "MetricExpr": "100 * cpu_atom@MEM_BOUND_STALLS.IFETCH_L2_HIT@ / cpu_atom@MEM_BOUND_STALLS.IFETCH@", "MetricGroup": " ", "MetricName": "tma_info_inst_miss_cost_l2hit_percent", "Unit": "cpu_atom" }, { "BriefDescription": "Percent of instruction miss cost that hit in the L3", - "MetricExpr": "100 * MEM_BOUND_STALLS.IFETCH_LLC_HIT / MEM_BOUND_STALLS.IFETCH", + "MetricExpr": "100 * cpu_atom@MEM_BOUND_STALLS.IFETCH_LLC_HIT@ / cpu_atom@MEM_BOUND_STALLS.IFETCH@", "MetricGroup": " ", "MetricName": "tma_info_inst_miss_cost_l3hit_percent", "Unit": "cpu_atom"
@@ -439,7 +439,7 @@ }, { "BriefDescription": "Instructions per Far Branch", - "MetricExpr": "INST_RETIRED.ANY / (BR_INST_RETIRED.FAR_BRANCH / 2)", + "MetricExpr": "INST_RETIRED.ANY / (cpu_atom@BR_INST_RETIRED.FAR_BRANCH@ / 2)", "MetricGroup": " ", "MetricName": "tma_info_ipfarbranch", "Unit": "cpu_atom"
@@ -453,7 +453,7 @@ }, { "BriefDescription": "Instructions per retired conditional Branch Misprediction where the branch was not taken", - "MetricExpr": "INST_RETIRED.ANY / (BR_MISP_RETIRED.COND - BR_MISP_RETIRED.COND_TAKEN)", + "MetricExpr": "INST_RETIRED.ANY / (cpu_atom@BR_MISP_RETIRED.COND@ - cpu_atom@BR_MISP_RETIRED.COND_TAKEN@)", "MetricName": "tma_info_ipmisp_cond_ntaken", "Unit": "cpu_atom" },
@@ -498,20 +498,20 @@ }, { "BriefDescription": "Percentage of total non-speculative loads that are splits", - "MetricExpr": "100 * MEM_UOPS_RETIRED.SPLIT_LOADS / MEM_UOPS_RETIRED.ALL_LOADS", + "MetricExpr": "100 * cpu_atom@MEM_UOPS_RETIRED.SPLIT_LOADS@ / MEM_UOPS_RETIRED.ALL_LOADS", "MetricName": "tma_info_load_splits", "Unit": "cpu_atom" }, { "BriefDescription": "load ops retired per 1000 instruction", - "MetricExpr": "1e3 * MEM_UOPS_RETIRED.ALL_LOADS / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_atom@MEM_UOPS_RETIRED.ALL_LOADS@ / INST_RETIRED.ANY", "MetricGroup": " ", "MetricName": "tma_info_memloadpki", "Unit": "cpu_atom" }, { "BriefDescription": "Percentage of all uops which are ucode ops", - "MetricExpr": "100 * UOPS_RETIRED.MS / UOPS_RETIRED.ALL", + "MetricExpr": "100 * cpu_atom@UOPS_RETIRED.MS@ / UOPS_RETIRED.ALL", "MetricGroup": " ", "MetricName": "tma_info_microcode_uop_ratio", "Unit": "cpu_atom"
@@ -525,7 +525,7 @@ }, { "BriefDescription": "Percentage of total non-speculative loads with a store forward or unknown store address block", - "MetricExpr": "100 * LD_BLOCKS.DATA_UNKNOWN / MEM_UOPS_RETIRED.ALL_LOADS", + "MetricExpr": "100 * cpu_atom@LD_BLOCKS.DATA_UNKNOWN@ / MEM_UOPS_RETIRED.ALL_LOADS", "MetricName": "tma_info_store_fwd_blocks", "Unit": "cpu_atom" },
@@ -545,7 +545,7 @@ }, { "BriefDescription": "Percentage of all uops which are x87 uops", - "MetricExpr": "100 * UOPS_RETIRED.X87 / UOPS_RETIRED.ALL", + "MetricExpr": "100 * cpu_atom@UOPS_RETIRED.X87@ / UOPS_RETIRED.ALL", "MetricGroup": " ", "MetricName": "tma_info_x87_uop_ratio", "Unit": "cpu_atom"
@@ -571,7 +571,7 @@ { "BriefDescription": "Counts the number of cycles a core is stalled due to a demand load which hit in the L2 Cache.", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "MEM_BOUND_STALLS.LOAD_L2_HIT / tma_info_clks - MEM_BOUND_STALLS_AT_RET_CORRECTION * MEM_BOUND_STALLS.LOAD_L2_HIT / MEM_BOUND_STALLS.LOAD", + "MetricExpr": "cpu_atom@MEM_BOUND_STALLS.LOAD_L2_HIT@ / tma_info_clks - max((cpu_atom@MEM_BOUND_STALLS.LOAD@ - cpu_atom@LD_HEAD.L1_MISS_AT_RET@) / tma_info_clks, 0) * cpu_atom@MEM_BOUND_STALLS.LOAD_L2_HIT@ / cpu_atom@MEM_BOUND_STALLS.LOAD@", "MetricGroup": "TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_l2_bound", "MetricThreshold": "tma_l2_bound > 0.1",
@@ -580,7 +580,7 @@ }, { "BriefDescription": "Counts the number of cycles a core is stalled due to a demand load which hit in the Last Level Cache (LLC) or other core with HITE/F/M.", - "MetricExpr": "MEM_BOUND_STALLS.LOAD_LLC_HIT / tma_info_clks - MEM_BOUND_STALLS_AT_RET_CORRECTION * MEM_BOUND_STALLS.LOAD_LLC_HIT / MEM_BOUND_STALLS.LOAD", + "MetricExpr": "cpu_atom@MEM_BOUND_STALLS.LOAD_LLC_HIT@ / tma_info_clks - max((cpu_atom@MEM_BOUND_STALLS.LOAD@ - cpu_atom@LD_HEAD.L1_MISS_AT_RET@) / tma_info_clks, 0) * cpu_atom@MEM_BOUND_STALLS.LOAD_LLC_HIT@ / cpu_atom@MEM_BOUND_STALLS.LOAD@", "MetricGroup": "TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_l3_bound", "MetricThreshold": "tma_l3_bound > 0.1",
@@ -589,7 +589,7 @@ }, { "BriefDescription": "Counts the number of cycles, relative to the number of mem_scheduler slots, in which uops are blocked due to load buffer full", - "MetricExpr": "tma_mem_scheduler * MEM_SCHEDULER_BLOCK.LD_BUF / MEM_SCHEDULER_BLOCK.ALL", + "MetricExpr": "tma_mem_scheduler * cpu_atom@MEM_SCHEDULER_BLOCK.LD_BUF@ / MEM_SCHEDULER_BLOCK.ALL", "MetricGroup": "TopdownL4;tma_L4_group;tma_mem_scheduler_group", "MetricName": "tma_ld_buffer", "MetricThreshold": "tma_ld_buffer > 0.05",
@@ -617,7 +617,7 @@ }, { "BriefDescription": "Counts the number of cycles the core is stalled due to stores or loads.", - "MetricExpr": "min(tma_backend_bound, LD_HEAD.ANY_AT_RET / tma_info_clks + tma_store_bound)", + "MetricExpr": "min(cpu_atom@TOPDOWN_BE_BOUND.ALL@ / tma_info_slots, cpu_atom@LD_HEAD.ANY_AT_RET@ / tma_info_clks + tma_store_bound)", "MetricGroup": "TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2",
@@ -627,7 +627,7 @@ }, { "BriefDescription": "Counts the number of machine clears relative to the number of nuke slots due to memory ordering.", - "MetricExpr": "tma_nuke * (MACHINE_CLEARS.MEMORY_ORDERING / MACHINE_CLEARS.SLOW)", + "MetricExpr": "tma_nuke * (cpu_atom@MACHINE_CLEARS.MEMORY_ORDERING@ / cpu_atom@MACHINE_CLEARS.SLOW@)", "MetricGroup": "TopdownL4;tma_L4_group;tma_nuke_group", "MetricName": "tma_memory_ordering", "MetricThreshold": "tma_memory_ordering > 0.02",
@@ -636,7 +636,7 @@ }, { "BriefDescription": "Counts the number of uops that are from the complex flows issued by the micro-sequencer (MS)", - "MetricExpr": "tma_microcode_sequencer", + "MetricExpr": "UOPS_RETIRED.MS / tma_info_slots", "MetricGroup": "TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_ms_uops", "MetricThreshold": "tma_ms_uops > 0.05",
@@ -692,7 +692,7 @@ }, { "BriefDescription": "Counts the number of uops retired excluding ms and fp div uops.", - "MetricExpr": "(TOPDOWN_RETIRING.ALL - UOPS_RETIRED.MS - UOPS_RETIRED.FPDIV) / tma_info_slots", + "MetricExpr": "(cpu_atom@TOPDOWN_RETIRING.ALL@ - cpu_atom@UOPS_RETIRED.MS@ - cpu_atom@UOPS_RETIRED.FPDIV@) / tma_info_slots", "MetricGroup": "TopdownL3;tma_L3_group;tma_base_group", "MetricName": "tma_other_ret", "MetricThreshold": "tma_other_ret > 0.3",
@@ -701,7 +701,7 @@ }, { "BriefDescription": "Counts the number of machine clears relative to the number of nuke slots due to page faults.", - "MetricExpr": "tma_nuke * (MACHINE_CLEARS.PAGE_FAULT / MACHINE_CLEARS.SLOW)", + "MetricExpr": "tma_nuke * (cpu_atom@MACHINE_CLEARS.PAGE_FAULT@ / cpu_atom@MACHINE_CLEARS.SLOW@)", "MetricGroup": "TopdownL4;tma_L4_group;tma_nuke_group", "MetricName": "tma_page_fault", "MetricThreshold": "tma_page_fault > 0.02",
@@ -758,7 +758,7 @@ }, { "BriefDescription": "Counts the number of cycles, relative to the number of mem_scheduler slots, in which uops are blocked due to RSV full relative", - "MetricExpr": "tma_mem_scheduler * MEM_SCHEDULER_BLOCK.RSV / MEM_SCHEDULER_BLOCK.ALL", + "MetricExpr": "tma_mem_scheduler * cpu_atom@MEM_SCHEDULER_BLOCK.RSV@ / MEM_SCHEDULER_BLOCK.ALL", "MetricGroup": "TopdownL4;tma_L4_group;tma_mem_scheduler_group", "MetricName": "tma_rsv", "MetricThreshold": "tma_rsv > 0.05",
@@ -776,7 +776,7 @@ }, { "BriefDescription": "Counts the number of machine clears relative to the number of nuke slots due to SMC.", - "MetricExpr": "tma_nuke * (MACHINE_CLEARS.SMC / MACHINE_CLEARS.SLOW)", + "MetricExpr": "tma_nuke * (cpu_atom@MACHINE_CLEARS.SMC@ / cpu_atom@MACHINE_CLEARS.SLOW@)", "MetricGroup": "TopdownL4;tma_L4_group;tma_nuke_group", "MetricName": "tma_smc", "MetricThreshold": "tma_smc > 0.02",
@@ -812,7 +812,7 @@ }, { "BriefDescription": "Counts the number of cycles the core is stalled due to store buffer full.", - "MetricExpr": "tma_mem_scheduler * (MEM_SCHEDULER_BLOCK.ST_BUF / MEM_SCHEDULER_BLOCK.ALL)", + "MetricExpr": "tma_mem_scheduler * (cpu_atom@MEM_SCHEDULER_BLOCK.ST_BUF@ / cpu_atom@MEM_SCHEDULER_BLOCK.ALL@)", "MetricGroup": "TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_store_bound", "MetricThreshold": "tma_store_bound > 0.1",
@@ -830,7 +830,7 @@ }, { "BriefDescription": "This metric represents Core fraction of cycles CPU dispatched uops on execution ports for ALU operations.", - "MetricExpr": "(UOPS_DISPATCHED.PORT_0 + UOPS_DISPATCHED.PORT_1 + UOPS_DISPATCHED.PORT_5_11 + UOPS_DISPATCHED.PORT_6) / (5 * tma_info_core_clks)", + "MetricExpr": "(cpu_core@UOPS_DISPATCHED.PORT_0@ + cpu_core@UOPS_DISPATCHED.PORT_1@ + cpu_core@UOPS_DISPATCHED.PORT_5_11@ + cpu_core@UOPS_DISPATCHED.PORT_6@) / (5 * tma_info_core_clks)", "MetricGroup": "TopdownL5;tma_L5_group;tma_ports_utilized_3m_group", "MetricName": "tma_alu_op_utilization", "MetricThreshold": "tma_alu_op_utilization > 0.6",
@@ -849,7 +849,7 @@ }, { "BriefDescription": "This metric estimates fraction of slots the CPU retired uops as a result of handing SSE to AVX* or AVX* to SSE transition Assists.", - "MetricExpr": "63 * ASSISTS.SSE_AVX_MIX / tma_info_slots", + "MetricExpr": "63 * [email protected]_AVX_MIX@ / tma_info_slots", "MetricGroup": "HPC;TopdownL5;tma_L5_group;tma_assists_group", "MetricName": "tma_avx_assists", "MetricThreshold": "tma_avx_assists > 0.1",
@@ -858,7 +858,7 @@ }, { "BriefDescription": "This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend", - "MetricExpr": "topdown\\-be\\-bound / (topdown\\-fe\\-bound + topdown\\-bad\\-spec + topdown\\-retiring + topdown\\-be\\-bound) + 0 * tma_info_slots", + "MetricExpr": "cpu_core@topdown\\-be\\-bound@ / (cpu_core@topdown\\-fe\\-bound@ + cpu_core@topdown\\-bad\\-spec@ + cpu_core@topdown\\-retiring@ + cpu_core@topdown\\-be\\-bound@) + 0 * tma_info_slots", "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_backend_bound", "MetricThreshold": "tma_backend_bound > 0.2",
@@ -880,7 +880,7 @@ }, { "BriefDescription": "This metric represents fraction of slots the CPU has wasted due to Branch Misprediction", - "MetricExpr": "topdown\\-br\\-mispredict / (topdown\\-fe\\-bound + topdown\\-bad\\-spec + topdown\\-retiring + topdown\\-be\\-bound) + 0 * tma_info_slots", + "MetricExpr": "cpu_core@topdown\\-br\\-mispredict@ / (cpu_core@topdown\\-fe\\-bound@ + cpu_core@topdown\\-bad\\-spec@ + cpu_core@topdown\\-retiring@ + cpu_core@topdown\\-be\\-bound@) + 0 * tma_info_slots", "MetricGroup": "BadSpec;BrMispredicts;TmaL2;TopdownL2;tma_L2_group;tma_bad_speculation_group;tma_issueBM", "MetricName": "tma_branch_mispredicts", "MetricThreshold": "tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15",
@@ -911,7 +911,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to Branch Resteers as a result of Machine Clears", - "MetricExpr": "(1 - tma_branch_mispredicts / tma_bad_speculation) * INT_MISC.CLEAR_RESTEER_CYCLES / tma_info_clks", + "MetricExpr": "(1 - tma_branch_mispredicts / tma_bad_speculation) * cpu_core@INT_MISC.CLEAR_RESTEER_CYCLES@ / tma_info_clks", "MetricGroup": "BadSpec;MachineClears;TopdownL4;tma_L4_group;tma_branch_resteers_group;tma_issueMC", "MetricName": "tma_clears_resteers", "MetricThreshold": "tma_clears_resteers > 0.05 & (tma_branch_resteers > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15))",
@@ -922,7 +922,7 @@ { "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to contested accesses", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "(25 * tma_info_average_frequency * (MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD * (OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HITM / (OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HITM + OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HIT_WITH_FWD))) + 24 * tma_info_average_frequency * MEM_LOAD_L3_HIT_RETIRED.XSNP_MISS) * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_clks", + "MetricExpr": "(25 * tma_info_average_frequency * (cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD@ * ([email protected]_DATA_RD.L3_HIT.SNOOP_HITM@ / ([email protected]_DATA_RD.L3_HIT.SNOOP_HITM@ + [email protected]_DATA_RD.L3_HIT.SNOOP_HIT_WITH_FWD@))) + 24 * tma_info_average_frequency * cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_MISS@) * (1 + cpu_core@MEM_LOAD_RETIRED.FB_HIT@ / cpu_core@MEM_LOAD_RETIRED.L1_MISS@ / 2) / tma_info_clks", "MetricGroup": "DataSharing;Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_l3_bound_group", "MetricName": "tma_contested_accesses", "MetricThreshold": "tma_contested_accesses > 0.05 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -944,7 +944,7 @@ { "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "24 * tma_info_average_frequency * (MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD + MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD * (1 - OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HITM / (OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HITM + OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HIT_WITH_FWD))) * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_clks", + "MetricExpr": "24 * tma_info_average_frequency * (cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD@ + cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD@ * (1 - [email protected]_DATA_RD.L3_HIT.SNOOP_HITM@ / ([email protected]_DATA_RD.L3_HIT.SNOOP_HITM@ + [email protected]_DATA_RD.L3_HIT.SNOOP_HIT_WITH_FWD@))) * (1 + cpu_core@MEM_LOAD_RETIRED.FB_HIT@ / cpu_core@MEM_LOAD_RETIRED.L1_MISS@ / 2) / tma_info_clks", "MetricGroup": "Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_l3_bound_group", "MetricName": "tma_data_sharing", "MetricThreshold": "tma_data_sharing > 0.05 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -975,7 +975,7 @@ { "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "MEMORY_ACTIVITY.STALLS_L3_MISS / tma_info_clks", + "MetricExpr": "cpu_core@MEMORY_ACTIVITY.STALLS_L3_MISS@ / tma_info_clks", "MetricGroup": "MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_dram_bound", "MetricThreshold": "tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)",
@@ -985,7 +985,7 @@ }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to DSB (decoded uop cache) fetch pipeline", - "MetricExpr": "(IDQ.DSB_CYCLES_ANY - IDQ.DSB_CYCLES_OK) / tma_info_core_clks / 2", + "MetricExpr": "([email protected]_CYCLES_ANY@ - [email protected]_CYCLES_OK@) / tma_info_core_clks / 2", "MetricGroup": "DSB;FetchBW;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_dsb", "MetricThreshold": "tma_dsb > 0.15 & (tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 6 > 0.35)",
@@ -1005,7 +1005,7 @@ }, { "BriefDescription": "This metric roughly estimates the fraction of cycles where the Data TLB (DTLB) was missed by load accesses", - "MetricExpr": "min(7 * cpu_core@DTLB_LOAD_MISSES.STLB_HIT\\,cmask\\=1@ + DTLB_LOAD_MISSES.WALK_ACTIVE, max(CYCLE_ACTIVITY.CYCLES_MEM_ANY - MEMORY_ACTIVITY.CYCLES_L1D_MISS, 0)) / tma_info_clks", + "MetricExpr": "min(7 * cpu_core@DTLB_LOAD_MISSES.STLB_HIT\\,cmask\\=1@ + cpu_core@DTLB_LOAD_MISSES.WALK_ACTIVE@, max(cpu_core@CYCLE_ACTIVITY.CYCLES_MEM_ANY@ - cpu_core@MEMORY_ACTIVITY.CYCLES_L1D_MISS@, 0)) / tma_info_clks", "MetricGroup": "MemoryTLB;TopdownL4;tma_L4_group;tma_issueTLB;tma_l1_bound_group", "MetricName": "tma_dtlb_load", "MetricThreshold": "tma_dtlb_load > 0.1 & (tma_l1_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -1015,7 +1015,7 @@ }, { "BriefDescription": "This metric roughly estimates the fraction of cycles spent handling first-level data TLB store misses", - "MetricExpr": "(7 * cpu_core@DTLB_STORE_MISSES.STLB_HIT\\,cmask\\=1@ + DTLB_STORE_MISSES.WALK_ACTIVE) / tma_info_core_clks", + "MetricExpr": "(7 * cpu_core@DTLB_STORE_MISSES.STLB_HIT\\,cmask\\=1@ + cpu_core@DTLB_STORE_MISSES.WALK_ACTIVE@) / tma_info_core_clks", "MetricGroup": "MemoryTLB;TopdownL4;tma_L4_group;tma_issueTLB;tma_store_bound_group", "MetricName": "tma_dtlb_store", "MetricThreshold": "tma_dtlb_store > 0.05 & (tma_store_bound > 0.2 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -1025,7 +1025,7 @@ }, { "BriefDescription": "This metric roughly estimates how often CPU was handling synchronizations due to False Sharing", - "MetricExpr": "28 * tma_info_average_frequency * OCR.DEMAND_RFO.L3_HIT.SNOOP_HITM / tma_info_clks", + "MetricExpr": "28 * tma_info_average_frequency * [email protected]_RFO.L3_HIT.SNOOP_HITM@ / tma_info_clks", "MetricGroup": "DataSharing;Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_store_bound_group", "MetricName": "tma_false_sharing", "MetricThreshold": "tma_false_sharing > 0.05 & (tma_store_bound > 0.2 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -1056,7 +1056,7 @@ }, { "BriefDescription": "This metric represents fraction of slots the CPU was stalled due to Frontend latency issues", - "MetricExpr": "topdown\\-fetch\\-lat / (topdown\\-fe\\-bound + topdown\\-bad\\-spec + topdown\\-retiring + topdown\\-be\\-bound) - INT_MISC.UOP_DROPPING / tma_info_slots", + "MetricExpr": "cpu_core@topdown\\-fetch\\-lat@ / (cpu_core@topdown\\-fe\\-bound@ + cpu_core@topdown\\-bad\\-spec@ + cpu_core@topdown\\-retiring@ + cpu_core@topdown\\-be\\-bound@) - cpu_core@INT_MISC.UOP_DROPPING@ / tma_info_slots", "MetricGroup": "Frontend;TmaL2;TopdownL2;tma_L2_group;tma_frontend_bound_group", "MetricName": "tma_fetch_latency", "MetricThreshold": "tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15",
@@ -1088,7 +1088,7 @@ }, { "BriefDescription": "This metric roughly estimates fraction of slots the CPU retired uops as a result of handing Floating Point (FP) Assists", - "MetricExpr": "30 * ASSISTS.FP / tma_info_slots", + "MetricExpr": "30 * [email protected]@ / tma_info_slots", "MetricGroup": "HPC;TopdownL5;tma_L5_group;tma_assists_group", "MetricName": "tma_fp_assists", "MetricThreshold": "tma_fp_assists > 0.1",
@@ -1118,7 +1118,7 @@ }, { "BriefDescription": "This metric approximates arithmetic FP vector uops fraction the CPU has retired for 128-bit wide vectors", - "MetricExpr": "(FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE) / (tma_retiring * tma_info_slots)", + "MetricExpr": "(cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE@ + cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE@) / (tma_retiring * tma_info_slots)", "MetricGroup": "Compute;Flops;TopdownL5;tma_L5_group;tma_fp_vector_group;tma_issue2P", "MetricName": "tma_fp_vector_128b", "MetricThreshold": "tma_fp_vector_128b > 0.1 & (tma_fp_vector > 0.1 & (tma_fp_arith > 0.2 & tma_light_operations > 0.6))",
@@ -1128,7 +1128,7 @@ }, { "BriefDescription": "This metric approximates arithmetic FP vector uops fraction the CPU has retired for 256-bit wide vectors", - "MetricExpr": "(FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE + FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE) / (tma_retiring * tma_info_slots)", + "MetricExpr": "(cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE@ + cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE@) / (tma_retiring * tma_info_slots)", "MetricGroup": "Compute;Flops;TopdownL5;tma_L5_group;tma_fp_vector_group;tma_issue2P", "MetricName": "tma_fp_vector_256b", "MetricThreshold": "tma_fp_vector_256b > 0.1 & (tma_fp_vector > 0.1 & (tma_fp_arith > 0.2 & tma_light_operations > 0.6))",
@@ -1138,7 +1138,7 @@ }, { "BriefDescription": "This category represents fraction of slots where the processor's Frontend undersupplies its Backend", - "MetricExpr": "topdown\\-fe\\-bound / (topdown\\-fe\\-bound + topdown\\-bad\\-spec + topdown\\-retiring + topdown\\-be\\-bound) - INT_MISC.UOP_DROPPING / tma_info_slots", + "MetricExpr": "cpu_core@topdown\\-fe\\-bound@ / (cpu_core@topdown\\-fe\\-bound@ + cpu_core@topdown\\-bad\\-spec@ + cpu_core@topdown\\-retiring@ + cpu_core@topdown\\-be\\-bound@) - cpu_core@INT_MISC.UOP_DROPPING@ / tma_info_slots", "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", "MetricThreshold": "tma_frontend_bound > 0.15",
@@ -1149,7 +1149,7 @@ }, { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring fused instructions -- where one uop can represent multiple contiguous instructions", - "MetricExpr": "tma_light_operations * INST_RETIRED.MACRO_FUSED / (tma_retiring * tma_info_slots)", + "MetricExpr": "tma_light_operations * cpu_core@INST_RETIRED.MACRO_FUSED@ / (tma_retiring * tma_info_slots)", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_fused_instructions", "MetricThreshold": "tma_fused_instructions > 0.1 & tma_light_operations > 0.6",
@@ -1159,7 +1159,7 @@ }, { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring heavy-weight operations -- instructions that require two or more uops or micro-coded sequences", - "MetricExpr": "topdown\\-heavy\\-ops / (topdown\\-fe\\-bound + topdown\\-bad\\-spec + topdown\\-retiring + topdown\\-be\\-bound) + 0 * tma_info_slots", + "MetricExpr": "cpu_core@topdown\\-heavy\\-ops@ / (cpu_core@topdown\\-fe\\-bound@ + cpu_core@topdown\\-bad\\-spec@ + cpu_core@topdown\\-retiring@ + cpu_core@topdown\\-be\\-bound@) + 0 * tma_info_slots", "MetricGroup": "Retire;TmaL2;TopdownL2;tma_L2_group;tma_retiring_group", "MetricName": "tma_heavy_operations", "MetricThreshold": "tma_heavy_operations > 0.1",
@@ -1213,7 +1213,7 @@ }, { "BriefDescription": "Total pipeline cost of branch related instructions (used for program control-flow including function calls)", - "MetricExpr": "100 * ((BR_INST_RETIRED.COND + 3 * BR_INST_RETIRED.NEAR_CALL + (BR_INST_RETIRED.NEAR_TAKEN - BR_INST_RETIRED.COND_TAKEN - 2 * BR_INST_RETIRED.NEAR_CALL)) / tma_info_slots)", + "MetricExpr": "100 * ((cpu_core@BR_INST_RETIRED.COND@ + 3 * cpu_core@BR_INST_RETIRED.NEAR_CALL@ + (cpu_core@BR_INST_RETIRED.NEAR_TAKEN@ - cpu_core@BR_INST_RETIRED.COND_TAKEN@ - 2 * cpu_core@BR_INST_RETIRED.NEAR_CALL@)) / tma_info_slots)", "MetricGroup": "Ret;tma_issueBC", "MetricName": "tma_info_branching_overhead", "MetricThreshold": "tma_info_branching_overhead > 10",
@@ -1222,21 +1222,21 @@ }, { "BriefDescription": "Fraction of branches that are CALL or RET", - "MetricExpr": "(BR_INST_RETIRED.NEAR_CALL + BR_INST_RETIRED.NEAR_RETURN) / BR_INST_RETIRED.ALL_BRANCHES", + "MetricExpr": "(cpu_core@BR_INST_RETIRED.NEAR_CALL@ + cpu_core@BR_INST_RETIRED.NEAR_RETURN@) / BR_INST_RETIRED.ALL_BRANCHES", "MetricGroup": "Bad;Branches", "MetricName": "tma_info_callret", "Unit": "cpu_core" }, { "BriefDescription": "Per-Logical Processor actual clocks when the Logical Processor is active.", - "MetricExpr": "CPU_CLK_UNHALTED.THREAD", + "MetricExpr": "cpu_core@CPU_CLK_UNHALTED.THREAD@", "MetricGroup": "Pipeline", "MetricName": "tma_info_clks", "Unit": "cpu_core" }, { "BriefDescription": "STLB (2nd level TLB) code speculative misses per kilo instruction (misses of any page-size that complete the page walk)", - "MetricExpr": "1e3 * ITLB_MISSES.WALK_COMPLETED / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@ITLB_MISSES.WALK_COMPLETED@ / INST_RETIRED.ANY", "MetricGroup": "Fed;MemoryTLB", "MetricName": "tma_info_code_stlb_mpki", "Unit": "cpu_core"
@@ -1266,7 +1266,7 @@ }, { "BriefDescription": "Core actual clocks when any Logical Processor is active on the Physical Core", - "MetricExpr": "CPU_CLK_UNHALTED.DISTRIBUTED", + "MetricExpr": "cpu_core@CPU_CLK_UNHALTED.DISTRIBUTED@", "MetricGroup": "SMT", "MetricName": "tma_info_core_clks", "Unit": "cpu_core"
@@ -1309,7 +1309,7 @@ }, { "BriefDescription": "Fraction of Uops delivered by the DSB (aka Decoded ICache; or Uop Cache)", - "MetricExpr": "IDQ.DSB_UOPS / UOPS_ISSUED.ANY", + "MetricExpr": "IDQ.DSB_UOPS / cpu_core@UOPS_ISSUED.ANY@", "MetricGroup": "DSB;Fed;FetchBW;tma_issueFB", "MetricName": "tma_info_dsb_coverage", "MetricThreshold": "tma_info_dsb_coverage < 0.7 & tma_info_ipc / 6 > 0.35",
@@ -1350,7 +1350,7 @@ }, { "BriefDescription": "Fill Buffer (FB) hits per kilo instructions for retired demand loads (L1D misses that merge into ongoing miss-handling entries)", - "MetricExpr": "1e3 * MEM_LOAD_RETIRED.FB_HIT / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@MEM_LOAD_RETIRED.FB_HIT@ / INST_RETIRED.ANY", "MetricGroup": "CacheMisses;Mem", "MetricName": "tma_info_fb_hpki", "Unit": "cpu_core"
@@ -1365,7 +1365,7 @@ { "BriefDescription": "Floating Point Operations Per Cycle", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "(FP_ARITH_INST_RETIRED.SCALAR_SINGLE + FP_ARITH_INST_RETIRED.SCALAR_DOUBLE + 2 * FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + 4 * (FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE + FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE) + 8 * FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE) / tma_info_core_clks", + "MetricExpr": "(cpu_core@FP_ARITH_INST_RETIRED.SCALAR_SINGLE@ + cpu_core@FP_ARITH_INST_RETIRED.SCALAR_DOUBLE@ + 2 * cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE@ + 4 * (cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE@ + cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE@) + 8 * cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE@) / tma_info_core_clks", "MetricGroup": "Flops;Ret", "MetricName": "tma_info_flopc", "Unit": "cpu_core"
@@ -1373,7 +1373,7 @@ { "BriefDescription": "Actual per-core usage of the Floating Point non-X87 execution units (regardless of precision or vector-width)", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "(FP_ARITH_DISPATCHED.PORT_0 + FP_ARITH_DISPATCHED.PORT_1 + FP_ARITH_DISPATCHED.PORT_5) / (2 * tma_info_core_clks)", + "MetricExpr": "(cpu_core@FP_ARITH_DISPATCHED.PORT_0@ + cpu_core@FP_ARITH_DISPATCHED.PORT_1@ + cpu_core@FP_ARITH_DISPATCHED.PORT_5@) / (2 * tma_info_core_clks)", "MetricGroup": "Cor;Flops;HPC", "MetricName": "tma_info_fp_arith_utilization", "PublicDescription": "Actual per-core usage of the Floating Point non-X87 execution units (regardless of precision or vector-width). Values > 1 are possible due to ([BDW+] Fused-Multiply Add (FMA) counting - common; [ADL+] use all of ADD/MUL/FMA in Scalar or 128/256-bit vectors - less common).",
@@ -1381,7 +1381,7 @@ }, { "BriefDescription": "Giga Floating Point Operations Per Second", - "MetricExpr": "(FP_ARITH_INST_RETIRED.SCALAR_SINGLE + FP_ARITH_INST_RETIRED.SCALAR_DOUBLE + 2 * FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + 4 * (FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE + FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE) + 8 * FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE) / 1e9 / duration_time", + "MetricExpr": "(cpu_core@FP_ARITH_INST_RETIRED.SCALAR_SINGLE@ + cpu_core@FP_ARITH_INST_RETIRED.SCALAR_DOUBLE@ + 2 * cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE@ + 4 * (cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE@ + cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE@) + 8 * cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE@) / 1e9 / duration_time", "MetricGroup": "Cor;Flops;HPC", "MetricName": "tma_info_gflops", "PublicDescription": "Giga Floating Point Operations Per Second. Aggregate across all supported options of: FP precisions, scalar and vector instructions, vector-width and AMX engine.",
@@ -1405,7 +1405,7 @@ }, { "BriefDescription": "Instruction-Level-Parallelism (average number of uops executed when there is execution) per-core", - "MetricExpr": "UOPS_EXECUTED.THREAD / (UOPS_EXECUTED.CORE_CYCLES_GE_1 / 2 if #SMT_on else UOPS_EXECUTED.CORE_CYCLES_GE_1)", + "MetricExpr": "UOPS_EXECUTED.THREAD / (cpu_core@UOPS_EXECUTED.CORE_CYCLES_GE_1@ / 2 if #SMT_on else cpu_core@UOPS_EXECUTED.CORE_CYCLES_GE_1@)", "MetricGroup": "Backend;Cor;Pipeline;PortsUtil", "MetricName": "tma_info_ilp", "Unit": "cpu_core"
@@ -1421,7 +1421,7 @@ }, { "BriefDescription": "Total number of retired Instructions", - "MetricExpr": "INST_RETIRED.ANY", + "MetricExpr": "cpu_core@INST_RETIRED.ANY@", "MetricGroup": "Summary;TmaL1;tma_L1_group", "MetricName": "tma_info_instructions", "PublicDescription": "Total number of retired Instructions. Sample with: INST_RETIRED.PREC_DIST",
@@ -1438,7 +1438,7 @@ }, { "BriefDescription": "Instructions per FP Arithmetic AVX/SSE 128-bit instruction (lower number means higher occurrence rate)", - "MetricExpr": "INST_RETIRED.ANY / (FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE)", + "MetricExpr": "INST_RETIRED.ANY / (cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE@ + cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE@)", "MetricGroup": "Flops;FpVector;InsType", "MetricName": "tma_info_iparith_avx128", "MetricThreshold": "tma_info_iparith_avx128 < 10",
@@ -1447,7 +1447,7 @@ }, { "BriefDescription": "Instructions per FP Arithmetic AVX* 256-bit instruction (lower number means higher occurrence rate)", - "MetricExpr": "INST_RETIRED.ANY / (FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE + FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE)", + "MetricExpr": "INST_RETIRED.ANY / (cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE@ + cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE@)", "MetricGroup": "Flops;FpVector;InsType", "MetricName": "tma_info_iparith_avx256", "MetricThreshold": "tma_info_iparith_avx256 < 10",
@@ -1514,7 +1514,7 @@ }, { "BriefDescription": "Instructions per Far Branch ( Far Branches apply upon transition from application to operating system, handling interrupts, exceptions) [lower number means higher occurrence rate]", - "MetricExpr": "INST_RETIRED.ANY / BR_INST_RETIRED.FAR_BRANCH:u", + "MetricExpr": "INST_RETIRED.ANY / cpu_core@BR_INST_RETIRED.FAR_BRANCH@u", "MetricGroup": "Branches;OS", "MetricName": "tma_info_ipfarbranch", "MetricThreshold": "tma_info_ipfarbranch < 1e6",
@@ -1522,7 +1522,7 @@ }, { "BriefDescription": "Instructions per Floating Point (FP) Operation (lower number means higher occurrence rate)", - "MetricExpr": "INST_RETIRED.ANY / (FP_ARITH_INST_RETIRED.SCALAR_SINGLE + FP_ARITH_INST_RETIRED.SCALAR_DOUBLE + 2 * FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + 4 * (FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE + FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE) + 8 * FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE)", + "MetricExpr": "INST_RETIRED.ANY / (cpu_core@FP_ARITH_INST_RETIRED.SCALAR_SINGLE@ + cpu_core@FP_ARITH_INST_RETIRED.SCALAR_DOUBLE@ + 2 * cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE@ + 4 * (cpu_core@FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE@ + cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_DOUBLE@) + 8 * cpu_core@FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE@)", "MetricGroup": "Flops;InsType", "MetricName": "tma_info_ipflop", "MetricThreshold": "tma_info_ipflop < 10",
@@ -1610,14 +1610,14 @@ }, { "BriefDescription": "Fraction of branches that are unconditional (direct or indirect) jumps", - "MetricExpr": "(BR_INST_RETIRED.NEAR_TAKEN - BR_INST_RETIRED.COND_TAKEN - 2 * BR_INST_RETIRED.NEAR_CALL) / BR_INST_RETIRED.ALL_BRANCHES", + "MetricExpr": "(cpu_core@BR_INST_RETIRED.NEAR_TAKEN@ - cpu_core@BR_INST_RETIRED.COND_TAKEN@ - 2 * cpu_core@BR_INST_RETIRED.NEAR_CALL@) / BR_INST_RETIRED.ALL_BRANCHES", "MetricGroup": "Bad;Branches", "MetricName": "tma_info_jump", "Unit": "cpu_core" }, { "BriefDescription": "Cycles Per Instruction for the Operating System (OS) Kernel mode", - "MetricExpr": "CPU_CLK_UNHALTED.THREAD_P:k / INST_RETIRED.ANY_P:k", + "MetricExpr": "CPU_CLK_UNHALTED.THREAD_P:k / cpu_core@INST_RETIRED.ANY_P@k", "MetricGroup": "OS", "MetricName": "tma_info_kernel_cpi", "Unit": "cpu_core"
@@ -1632,7 +1632,7 @@ }, { "BriefDescription": "Average per-core data fill bandwidth to the L1 data cache [GB / sec]", - "MetricExpr": "64 * L1D.REPLACEMENT / 1e9 / duration_time", + "MetricExpr": "64 * [email protected]@ / 1e9 / duration_time", "MetricGroup": "Mem;MemoryBW", "MetricName": "tma_info_l1d_cache_fill_bw", "Unit": "cpu_core"
@@ -1646,21 +1646,21 @@ }, { "BriefDescription": "L1 cache true misses per kilo instruction for retired demand loads", - "MetricExpr": "1e3 * MEM_LOAD_RETIRED.L1_MISS / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@MEM_LOAD_RETIRED.L1_MISS@ / INST_RETIRED.ANY", "MetricGroup": "CacheMisses;Mem", "MetricName": "tma_info_l1mpki", "Unit": "cpu_core" }, { "BriefDescription": "L1 cache true misses per kilo instruction for all demand loads (including speculative)", - "MetricExpr": "1e3 * L2_RQSTS.ALL_DEMAND_DATA_RD / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@L2_RQSTS.ALL_DEMAND_DATA_RD@ / INST_RETIRED.ANY", "MetricGroup": "CacheMisses;Mem", "MetricName": "tma_info_l1mpki_load", "Unit": "cpu_core" }, { "BriefDescription": "Average per-core data fill bandwidth to the L2 cache [GB / sec]", - "MetricExpr": "64 * L2_LINES_IN.ALL / 1e9 / duration_time", + "MetricExpr": "64 * cpu_core@L2_LINES_IN.ALL@ / 1e9 / duration_time", "MetricGroup": "Mem;MemoryBW", "MetricName": "tma_info_l2_cache_fill_bw", "Unit": "cpu_core"
@@ -1674,56 +1674,56 @@ }, { "BriefDescription": "L2 cache hits per kilo instruction for all request types (including speculative)", - "MetricExpr": "1e3 * (L2_RQSTS.REFERENCES - L2_RQSTS.MISS) / INST_RETIRED.ANY", + "MetricExpr": "1e3 * (cpu_core@L2_RQSTS.REFERENCES@ - cpu_core@L2_RQSTS.MISS@) / INST_RETIRED.ANY", "MetricGroup": "CacheMisses;Mem", "MetricName": "tma_info_l2hpki_all", "Unit": "cpu_core" }, { "BriefDescription": "L2 cache hits per kilo instruction for all demand loads (including speculative)", - "MetricExpr": "1e3 * L2_RQSTS.DEMAND_DATA_RD_HIT / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@L2_RQSTS.DEMAND_DATA_RD_HIT@ / INST_RETIRED.ANY", "MetricGroup": "CacheMisses;Mem", "MetricName": "tma_info_l2hpki_load", "Unit": "cpu_core" }, { "BriefDescription": "L2 cache true misses per kilo instruction for retired demand loads", - "MetricExpr": "1e3 * MEM_LOAD_RETIRED.L2_MISS / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@MEM_LOAD_RETIRED.L2_MISS@ / INST_RETIRED.ANY", "MetricGroup": "Backend;CacheMisses;Mem", "MetricName": "tma_info_l2mpki", "Unit": "cpu_core" }, { "BriefDescription": "L2 cache ([RKL+] true) misses per kilo instruction for all request types (including speculative)", - "MetricExpr": "1e3 * L2_RQSTS.MISS / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@L2_RQSTS.MISS@ / INST_RETIRED.ANY", "MetricGroup": "CacheMisses;Mem;Offcore", "MetricName": "tma_info_l2mpki_all", "Unit": "cpu_core" }, { "BriefDescription": "L2 cache true code cacheline misses per kilo instruction", - "MetricExpr": "1e3 * FRONTEND_RETIRED.L2_MISS / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@FRONTEND_RETIRED.L2_MISS@ / INST_RETIRED.ANY", "MetricGroup": "IcMiss", "MetricName": "tma_info_l2mpki_code", "Unit": "cpu_core" }, { "BriefDescription": "L2 cache speculative code cacheline misses per kilo instruction", - "MetricExpr": "1e3 * L2_RQSTS.CODE_RD_MISS / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@L2_RQSTS.CODE_RD_MISS@ / INST_RETIRED.ANY", "MetricGroup": "IcMiss", "MetricName": "tma_info_l2mpki_code_all", "Unit": "cpu_core" }, { "BriefDescription": "L2 cache ([RKL+] true) misses per kilo instruction for all demand loads (including speculative)", - "MetricExpr": "1e3 * L2_RQSTS.DEMAND_DATA_RD_MISS / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@L2_RQSTS.DEMAND_DATA_RD_MISS@ / INST_RETIRED.ANY", "MetricGroup": "CacheMisses;Mem", "MetricName": "tma_info_l2mpki_load", "Unit": "cpu_core" }, { "BriefDescription": "Average per-core data access bandwidth to the L3 cache [GB / sec]", - "MetricExpr": "64 * OFFCORE_REQUESTS.ALL_REQUESTS / 1e9 / duration_time", + "MetricExpr": "64 * cpu_core@OFFCORE_REQUESTS.ALL_REQUESTS@ / 1e9 / duration_time", "MetricGroup": "Mem;MemoryBW;Offcore", "MetricName": "tma_info_l3_cache_access_bw", "Unit": "cpu_core"
@@ -1737,7 +1737,7 @@ }, { "BriefDescription": "Average per-core data fill bandwidth to the L3 cache [GB / sec]", - "MetricExpr": "64 * LONGEST_LAT_CACHE.MISS / 1e9 / duration_time", + "MetricExpr": "64 * cpu_core@LONGEST_LAT_CACHE.MISS@ / 1e9 / duration_time", "MetricGroup": "Mem;MemoryBW", "MetricName": "tma_info_l3_cache_fill_bw", "Unit": "cpu_core"
@@ -1751,7 +1751,7 @@ }, { "BriefDescription": "L3 cache true misses per kilo instruction for retired demand loads", - "MetricExpr": "1e3 * MEM_LOAD_RETIRED.L3_MISS / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@MEM_LOAD_RETIRED.L3_MISS@ / INST_RETIRED.ANY", "MetricGroup": "CacheMisses;Mem", "MetricName": "tma_info_l3mpki", "Unit": "cpu_core"
@@ -1786,14 +1786,14 @@ }, { "BriefDescription": "STLB (2nd level TLB) data load speculative misses per kilo instruction (misses of any page-size that complete the page walk)", - "MetricExpr": "1e3 * DTLB_LOAD_MISSES.WALK_COMPLETED / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@DTLB_LOAD_MISSES.WALK_COMPLETED@ / INST_RETIRED.ANY", "MetricGroup": "Mem;MemoryTLB", "MetricName": "tma_info_load_stlb_mpki", "Unit": "cpu_core" }, { "BriefDescription": "Fraction of Uops delivered by the LSD (Loop Stream Detector; aka Loop Cache)", - "MetricExpr": "LSD.UOPS / UOPS_ISSUED.ANY", + "MetricExpr": "LSD.UOPS / cpu_core@UOPS_ISSUED.ANY@", "MetricGroup": "Fed;LSD", "MetricName": "tma_info_lsd_coverage", "Unit": "cpu_core"
@@ -1877,7 +1877,7 @@ }, { "BriefDescription": "Utilization of the core's Page Walker(s) serving STLB misses triggered by instruction/Load/Store accesses", - "MetricExpr": "(ITLB_MISSES.WALK_PENDING + DTLB_LOAD_MISSES.WALK_PENDING + DTLB_STORE_MISSES.WALK_PENDING) / (4 * tma_info_core_clks)", + "MetricExpr": "(cpu_core@ITLB_MISSES.WALK_PENDING@ + cpu_core@DTLB_LOAD_MISSES.WALK_PENDING@ + cpu_core@DTLB_STORE_MISSES.WALK_PENDING@) / (4 * tma_info_core_clks)", "MetricGroup": "Mem;MemoryTLB", "MetricName": "tma_info_page_walks_utilization", "MetricThreshold": "tma_info_page_walks_utilization > 0.5",
@@ -1893,21 +1893,21 @@ }, { "BriefDescription": "Total issue-pipeline slots (per-Physical Core till ICL; per-Logical Processor ICL onward)", - "MetricExpr": "TOPDOWN.SLOTS", + "MetricExpr": "[email protected]@", "MetricGroup": "TmaL1;tma_L1_group", "MetricName": "tma_info_slots", "Unit": "cpu_core" }, { "BriefDescription": "Fraction of Physical Core issue-slots utilized by this Logical Processor", - "MetricExpr": "(tma_info_slots / (TOPDOWN.SLOTS / 2) if #SMT_on else 1)", + "MetricExpr": "(tma_info_slots / ([email protected]@ / 2) if #SMT_on else 1)", "MetricGroup": "SMT;TmaL1;tma_L1_group", "MetricName": "tma_info_slots_utilization", "Unit": "cpu_core" }, { "BriefDescription": "Fraction of cycles where both hardware Logical Processors were active", - "MetricExpr": "(1 - CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE / CPU_CLK_UNHALTED.REF_DISTRIBUTED if #SMT_on else 0)", + "MetricExpr": "(1 - cpu_core@CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE@ / cpu_core@CPU_CLK_UNHALTED.REF_DISTRIBUTED@ if #SMT_on else 0)", "MetricGroup": "SMT", "MetricName": "tma_info_smt_2t_utilization", "Unit": "cpu_core"
@@ -1921,7 +1921,7 @@ }, { "BriefDescription": "STLB (2nd level TLB) data store speculative misses per kilo instruction (misses of any page-size that complete the page walk)", - "MetricExpr": "1e3 * DTLB_STORE_MISSES.WALK_COMPLETED / INST_RETIRED.ANY", + "MetricExpr": "1e3 * cpu_core@DTLB_STORE_MISSES.WALK_COMPLETED@ / INST_RETIRED.ANY", "MetricGroup": "Mem;MemoryTLB", "MetricName": "tma_info_store_stlb_mpki", "Unit": "cpu_core"
@@ -1969,7 +1969,7 @@ }, { "BriefDescription": "This metric represents 128-bit vector Integer ADD/SUB/SAD or VNNI (Vector Neural Network Instructions) uops fraction the CPU has retired", - "MetricExpr": "(INT_VEC_RETIRED.ADD_128 + INT_VEC_RETIRED.VNNI_128) / (tma_retiring * tma_info_slots)", + "MetricExpr": "(cpu_core@INT_VEC_RETIRED.ADD_128@ + cpu_core@INT_VEC_RETIRED.VNNI_128@) / (tma_retiring * tma_info_slots)", "MetricGroup": "Compute;IntVector;Pipeline;TopdownL4;tma_L4_group;tma_int_operations_group;tma_issue2P", "MetricName": "tma_int_vector_128b", "MetricThreshold": "tma_int_vector_128b > 0.1 & (tma_int_operations > 0.1 & tma_light_operations > 0.6)",
@@ -1979,7 +1979,7 @@ }, { "BriefDescription": "This metric represents 256-bit vector Integer ADD/SUB/SAD or VNNI (Vector Neural Network Instructions) uops fraction the CPU has retired", - "MetricExpr": "(INT_VEC_RETIRED.ADD_256 + INT_VEC_RETIRED.MUL_256 + INT_VEC_RETIRED.VNNI_256) / (tma_retiring * tma_info_slots)", + "MetricExpr": "(cpu_core@INT_VEC_RETIRED.ADD_256@ + cpu_core@INT_VEC_RETIRED.MUL_256@ + cpu_core@INT_VEC_RETIRED.VNNI_256@) / (tma_retiring * tma_info_slots)", "MetricGroup": "Compute;IntVector;Pipeline;TopdownL4;tma_L4_group;tma_int_operations_group;tma_issue2P", "MetricName": "tma_int_vector_256b", "MetricThreshold": "tma_int_vector_256b > 0.1 & (tma_int_operations > 0.1 & tma_light_operations > 0.6)",
@@ -1999,7 +1999,7 @@ }, { "BriefDescription": "This metric estimates how often the CPU was stalled without loads missing the L1 data cache", - "MetricExpr": "max((EXE_ACTIVITY.BOUND_ON_LOADS - MEMORY_ACTIVITY.STALLS_L1D_MISS) / tma_info_clks, 0)", + "MetricExpr": "max((cpu_core@EXE_ACTIVITY.BOUND_ON_LOADS@ - cpu_core@MEMORY_ACTIVITY.STALLS_L1D_MISS@) / tma_info_clks, 0)", "MetricGroup": "CacheMisses;MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_issueL1;tma_issueMC;tma_memory_bound_group", "MetricName": "tma_l1_bound", "MetricThreshold": "tma_l1_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)",
@@ -2010,7 +2010,7 @@ { "BriefDescription": "This metric estimates how often the CPU was stalled due to L2 cache accesses by loads", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "(MEMORY_ACTIVITY.STALLS_L1D_MISS - MEMORY_ACTIVITY.STALLS_L2_MISS) / tma_info_clks", + "MetricExpr": "(cpu_core@MEMORY_ACTIVITY.STALLS_L1D_MISS@ - cpu_core@MEMORY_ACTIVITY.STALLS_L2_MISS@) / tma_info_clks", "MetricGroup": "CacheMisses;MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_l2_bound", "MetricThreshold": "tma_l2_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)",
@@ -2020,7 +2020,7 @@ }, { "BriefDescription": "This metric estimates how often the CPU was stalled due to loads accesses to L3 cache or contended with a sibling Core", - "MetricExpr": "(MEMORY_ACTIVITY.STALLS_L2_MISS - MEMORY_ACTIVITY.STALLS_L3_MISS) / tma_info_clks", + "MetricExpr": "(cpu_core@MEMORY_ACTIVITY.STALLS_L2_MISS@ - cpu_core@MEMORY_ACTIVITY.STALLS_L3_MISS@) / tma_info_clks", "MetricGroup": "CacheMisses;MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_l3_bound", "MetricThreshold": "tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)",
@@ -2030,7 +2030,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles with demand load accesses that hit the L3 cache under unloaded scenarios (possibly L3 latency limited)", - "MetricExpr": "9 * tma_info_average_frequency * MEM_LOAD_RETIRED.L3_HIT * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_clks", + "MetricExpr": "9 * tma_info_average_frequency * cpu_core@MEM_LOAD_RETIRED.L3_HIT@ * (1 + cpu_core@MEM_LOAD_RETIRED.FB_HIT@ / cpu_core@MEM_LOAD_RETIRED.L1_MISS@ / 2) / tma_info_clks", "MetricGroup": "MemoryLat;TopdownL4;tma_L4_group;tma_issueLat;tma_l3_bound_group", "MetricName": "tma_l3_hit_latency", "MetricThreshold": "tma_l3_hit_latency > 0.1 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2090,7 +2090,7 @@ { "BriefDescription": "This metric represents fraction of cycles the CPU spent handling cache misses due to lock operations", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "(16 * max(0, MEM_INST_RETIRED.LOCK_LOADS - L2_RQSTS.ALL_RFO) + MEM_INST_RETIRED.LOCK_LOADS / MEM_INST_RETIRED.ALL_STORES * (10 * L2_RQSTS.RFO_HIT + min(CPU_CLK_UNHALTED.THREAD, OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DEMAND_RFO))) / tma_info_clks", + "MetricExpr": "(16 * max(0, cpu_core@MEM_INST_RETIRED.LOCK_LOADS@ - cpu_core@L2_RQSTS.ALL_RFO@) + cpu_core@MEM_INST_RETIRED.LOCK_LOADS@ / cpu_core@MEM_INST_RETIRED.ALL_STORES@ * (10 * cpu_core@L2_RQSTS.RFO_HIT@ + min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DEMAND_RFO@))) / tma_info_clks", "MetricGroup": "Offcore;TopdownL4;tma_L4_group;tma_issueRFO;tma_l1_bound_group", "MetricName": "tma_lock_latency", "MetricThreshold": "tma_lock_latency > 0.2 & (tma_l1_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2100,7 +2100,7 @@ }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to LSD (Loop Stream Detector) unit", - "MetricExpr": "(LSD.CYCLES_ACTIVE - LSD.CYCLES_OK) / tma_info_core_clks / 2", + "MetricExpr": "([email protected]_ACTIVE@ - [email protected]_OK@) / tma_info_core_clks / 2", "MetricGroup": "FetchBW;LSD;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_lsd", "MetricThreshold": "tma_lsd > 0.15 & (tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 6 > 0.35)",
@@ -2121,7 +2121,7 @@ }, { "BriefDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory (DRAM)", - "MetricExpr": "min(CPU_CLK_UNHALTED.THREAD, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.ALL_DATA_RD\\,cmask\\=4@) / tma_info_clks", + "MetricExpr": "min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.ALL_DATA_RD\\,cmask\\=4@) / tma_info_clks", "MetricGroup": "MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueBW", "MetricName": "tma_mem_bandwidth", "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2131,7 +2131,7 @@ }, { "BriefDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory (DRAM)", - "MetricExpr": "min(CPU_CLK_UNHALTED.THREAD, OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DATA_RD) / tma_info_clks - tma_mem_bandwidth", + "MetricExpr": "min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DATA_RD@) / tma_info_clks - tma_mem_bandwidth", "MetricGroup": "MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueLat", "MetricName": "tma_mem_latency", "MetricThreshold": "tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2141,7 +2141,7 @@ }, { "BriefDescription": "This metric represents fraction of slots the Memory subsystem within the Backend was a bottleneck", - "MetricExpr": "min(tma_backend_bound, LD_HEAD.ANY_AT_RET / tma_info_clks + tma_store_bound)", + "MetricExpr": "cpu_core@topdown\\-mem\\-bound@ / (cpu_core@topdown\\-fe\\-bound@ + cpu_core@topdown\\-bad\\-spec@ + cpu_core@topdown\\-retiring@ + cpu_core@topdown\\-be\\-bound@) + 0 * tma_info_slots", "MetricGroup": "Backend;TmaL2;TopdownL2;tma_L2_group;tma_backend_bound_group", "MetricName": "tma_memory_bound", "MetricThreshold": "tma_memory_bound > 0.2 & tma_backend_bound > 0.2",
@@ -2152,7 +2152,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to LFENCE Instructions.", - "MetricExpr": "13 * MISC2_RETIRED.LFENCE / tma_info_clks", + "MetricExpr": "13 * cpu_core@MISC2_RETIRED.LFENCE@ / tma_info_clks", "MetricGroup": "TopdownL6;tma_L6_group;tma_serializing_operation_group", "MetricName": "tma_memory_fence", "MetricThreshold": "tma_memory_fence > 0.05 & (tma_serializing_operation > 0.1 & (tma_ports_utilized_0 > 0.2 & (tma_ports_utilization > 0.15 & (tma_core_bound > 0.1 & tma_backend_bound > 0.2))))",
@@ -2162,7 +2162,7 @@ { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring memory operations -- uops for memory load or store accesses.", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "tma_light_operations * MEM_UOP_RETIRED.ANY / (tma_retiring * tma_info_slots)", + "MetricExpr": "tma_light_operations * cpu_core@MEM_UOP_RETIRED.ANY@ / (tma_retiring * tma_info_slots)", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_memory_operations", "MetricThreshold": "tma_memory_operations > 0.1 & tma_light_operations > 0.6",
@@ -2181,7 +2181,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to Branch Resteers as a result of Branch Misprediction at execution stage", - "MetricExpr": "tma_branch_mispredicts / tma_bad_speculation * INT_MISC.CLEAR_RESTEER_CYCLES / tma_info_clks", + "MetricExpr": "tma_branch_mispredicts / tma_bad_speculation * cpu_core@INT_MISC.CLEAR_RESTEER_CYCLES@ / tma_info_clks", "MetricGroup": "BadSpec;BrMispredicts;TopdownL4;tma_L4_group;tma_branch_resteers_group;tma_issueBM", "MetricName": "tma_mispredicts_resteers", "MetricThreshold": "tma_mispredicts_resteers > 0.05 & (tma_branch_resteers > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15))",
@@ -2191,7 +2191,7 @@ }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to the MITE pipeline (the legacy decode pipeline)", - "MetricExpr": "(IDQ.MITE_CYCLES_ANY - IDQ.MITE_CYCLES_OK) / tma_info_core_clks / 2", + "MetricExpr": "([email protected]_CYCLES_ANY@ - [email protected]_CYCLES_OK@) / tma_info_core_clks / 2", "MetricGroup": "DSBmiss;FetchBW;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_mite", "MetricThreshold": "tma_mite > 0.1 & (tma_fetch_bandwidth > 0.1 & tma_frontend_bound > 0.15 & tma_info_ipc / 6 > 0.35)",
@@ -2201,7 +2201,7 @@ }, { "BriefDescription": "The Mixing_Vectors metric gives the percentage of injected blend uops out of all uops issued", - "MetricExpr": "160 * ASSISTS.SSE_AVX_MIX / tma_info_clks", + "MetricExpr": "160 * [email protected]_AVX_MIX@ / tma_info_clks", "MetricGroup": "TopdownL5;tma_L5_group;tma_issueMV;tma_ports_utilized_0_group", "MetricName": "tma_mixing_vectors", "MetricThreshold": "tma_mixing_vectors > 0.05",
@@ -2211,7 +2211,7 @@ }, { "BriefDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS)", - "MetricExpr": "3 * cpu_core@UOPS_RETIRED.MS\\,cmask\\=1\\,edge@ / (tma_retiring * tma_info_slots / UOPS_ISSUED.ANY) / tma_info_clks", + "MetricExpr": "3 * cpu_core@UOPS_RETIRED.MS\\,cmask\\=1\\,edge@ / (tma_retiring * tma_info_slots / cpu_core@UOPS_ISSUED.ANY@) / tma_info_clks", "MetricGroup": "FetchLat;MicroSeq;TopdownL3;tma_L3_group;tma_fetch_latency_group;tma_issueMC;tma_issueMS;tma_issueMV;tma_issueSO", "MetricName": "tma_ms_switches", "MetricThreshold": "tma_ms_switches > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)",
@@ -2221,7 +2221,7 @@ }, { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring branch instructions that were not fused", - "MetricExpr": "tma_light_operations * (BR_INST_RETIRED.ALL_BRANCHES - INST_RETIRED.MACRO_FUSED) / (tma_retiring * tma_info_slots)", + "MetricExpr": "tma_light_operations * (cpu_core@BR_INST_RETIRED.ALL_BRANCHES@ - cpu_core@INST_RETIRED.MACRO_FUSED@) / (tma_retiring * tma_info_slots)", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_non_fused_branches", "MetricThreshold": "tma_non_fused_branches > 0.1 & tma_light_operations > 0.6",
@@ -2231,7 +2231,7 @@ }, { "BriefDescription": "This metric represents fraction of slots where the CPU was retiring NOP (no op) instructions", - "MetricExpr": "tma_light_operations * INST_RETIRED.NOP / (tma_retiring * tma_info_slots)", + "MetricExpr": "tma_light_operations * cpu_core@INST_RETIRED.NOP@ / (tma_retiring * tma_info_slots)", "MetricGroup": "Pipeline;TopdownL3;tma_L3_group;tma_light_operations_group", "MetricName": "tma_nop_instructions", "MetricThreshold": "tma_nop_instructions > 0.1 & tma_light_operations > 0.6",
@@ -2252,7 +2252,7 @@ }, { "BriefDescription": "This metric roughly estimates fraction of slots the CPU retired uops as a result of handing Page Faults", - "MetricExpr": "99 * ASSISTS.PAGE_FAULT / tma_info_slots", + "MetricExpr": "99 * [email protected]_FAULT@ / tma_info_slots", "MetricGroup": "TopdownL5;tma_L5_group;tma_assists_group", "MetricName": "tma_page_faults", "MetricThreshold": "tma_page_faults > 0.05",
@@ -2292,7 +2292,7 @@ }, { "BriefDescription": "This metric estimates fraction of cycles the CPU performance was potentially limited due to Core computation issues (non divider-related)", - "MetricExpr": "((cpu_core@EXE_ACTIVITY.3_PORTS_UTIL\\,umask\\=0x80@ + tma_serializing_operation * (CYCLE_ACTIVITY.STALLS_TOTAL - EXE_ACTIVITY.BOUND_ON_LOADS) + (EXE_ACTIVITY.1_PORTS_UTIL + tma_retiring * cpu_core@EXE_ACTIVITY.2_PORTS_UTIL\\,umask\\=0xc@)) / tma_info_clks if ARITH.DIV_ACTIVE < CYCLE_ACTIVITY.STALLS_TOTAL - EXE_ACTIVITY.BOUND_ON_LOADS else (EXE_ACTIVITY.1_PORTS_UTIL + tma_retiring * cpu_core@EXE_ACTIVITY.2_PORTS_UTIL\\,umask\\=0xc@) / tma_info_clks)", + "MetricExpr": "((cpu_core@EXE_ACTIVITY.3_PORTS_UTIL\\,umask\\=0x80@ + tma_serializing_operation * (cpu_core@CYCLE_ACTIVITY.STALLS_TOTAL@ - cpu_core@EXE_ACTIVITY.BOUND_ON_LOADS@) + (cpu_core@EXE_ACTIVITY.1_PORTS_UTIL@ + tma_retiring * cpu_core@EXE_ACTIVITY.2_PORTS_UTIL\\,umask\\=0xc@)) / tma_info_clks if [email protected]_ACTIVE@ < cpu_core@CYCLE_ACTIVITY.STALLS_TOTAL@ - cpu_core@EXE_ACTIVITY.BOUND_ON_LOADS@ else (cpu_core@EXE_ACTIVITY.1_PORTS_UTIL@ + tma_retiring * cpu_core@EXE_ACTIVITY.2_PORTS_UTIL\\,umask\\=0xc@) / tma_info_clks)", "MetricGroup": "PortsUtil;TopdownL3;tma_L3_group;tma_core_bound_group", "MetricName": "tma_ports_utilization", "MetricThreshold": "tma_ports_utilization > 0.15 & (tma_core_bound > 0.1 & tma_backend_bound > 0.2)",
@@ -2302,7 +2302,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles CPU executed no uops on any execution port (Logical Processor cycles since ICL, Physical Core cycles otherwise)", - "MetricExpr": "cpu_core@EXE_ACTIVITY.3_PORTS_UTIL\\,umask\\=0x80@ / tma_info_clks + tma_serializing_operation * (CYCLE_ACTIVITY.STALLS_TOTAL - EXE_ACTIVITY.BOUND_ON_LOADS) / tma_info_clks", + "MetricExpr": "cpu_core@EXE_ACTIVITY.3_PORTS_UTIL\\,umask\\=0x80@ / tma_info_clks + tma_serializing_operation * (cpu_core@CYCLE_ACTIVITY.STALLS_TOTAL@ - cpu_core@EXE_ACTIVITY.BOUND_ON_LOADS@) / tma_info_clks", "MetricGroup": "PortsUtil;TopdownL4;tma_L4_group;tma_ports_utilization_group", "MetricName": "tma_ports_utilized_0", "MetricThreshold": "tma_ports_utilized_0 > 0.2 & (tma_ports_utilization > 0.15 & (tma_core_bound > 0.1 & tma_backend_bound > 0.2))",
@@ -2342,7 +2342,7 @@ }, { "BriefDescription": "This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired", - "MetricExpr": "topdown\\-retiring / (topdown\\-fe\\-bound + topdown\\-bad\\-spec + topdown\\-retiring + topdown\\-be\\-bound) + 0 * tma_info_slots", + "MetricExpr": "cpu_core@topdown\\-retiring@ / (cpu_core@topdown\\-fe\\-bound@ + cpu_core@topdown\\-bad\\-spec@ + cpu_core@topdown\\-retiring@ + cpu_core@topdown\\-be\\-bound@) + 0 * tma_info_slots", "MetricGroup": "TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_retiring", "MetricThreshold": "tma_retiring > 0.7 | tma_heavy_operations > 0.1",
@@ -2382,7 +2382,7 @@ }, { "BriefDescription": "This metric estimates fraction of cycles handling memory load split accesses - load that cross 64-byte cache line boundary", - "MetricExpr": "tma_info_load_miss_real_latency * LD_BLOCKS.NO_SR / tma_info_clks", + "MetricExpr": "tma_info_load_miss_real_latency * cpu_core@LD_BLOCKS.NO_SR@ / tma_info_clks", "MetricGroup": "TopdownL4;tma_L4_group;tma_l1_bound_group", "MetricName": "tma_split_loads", "MetricThreshold": "tma_split_loads > 0.2 & (tma_l1_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2402,7 +2402,7 @@ }, { "BriefDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors)", - "MetricExpr": "(XQ.FULL_CYCLES + L1D_PEND_MISS.L2_STALLS) / tma_info_clks", + "MetricExpr": "([email protected]_CYCLES@ + cpu_core@L1D_PEND_MISS.L2_STALLS@) / tma_info_clks", "MetricGroup": "MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_bound_group", "MetricName": "tma_sq_full", "MetricThreshold": "tma_sq_full > 0.3 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2422,7 +2422,7 @@ }, { "BriefDescription": "This metric roughly estimates fraction of cycles when the memory subsystem had loads blocked since they could not forward data from earlier (in program order) overlapping stores", - "MetricExpr": "13 * LD_BLOCKS.STORE_FORWARD / tma_info_clks", + "MetricExpr": "13 * cpu_core@LD_BLOCKS.STORE_FORWARD@ / tma_info_clks", "MetricGroup": "TopdownL4;tma_L4_group;tma_l1_bound_group", "MetricName": "tma_store_fwd_blk", "MetricThreshold": "tma_store_fwd_blk > 0.1 & (tma_l1_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2432,7 +2432,7 @@ }, { "BriefDescription": "This metric estimates fraction of cycles the CPU spent handling L1D store misses", - "MetricExpr": "(MEM_STORE_RETIRED.L2_HIT * 10 * (1 - MEM_INST_RETIRED.LOCK_LOADS / MEM_INST_RETIRED.ALL_STORES) + (1 - MEM_INST_RETIRED.LOCK_LOADS / MEM_INST_RETIRED.ALL_STORES) * min(CPU_CLK_UNHALTED.THREAD, OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DEMAND_RFO)) / tma_info_clks", + "MetricExpr": "(cpu_core@MEM_STORE_RETIRED.L2_HIT@ * 10 * (1 - cpu_core@MEM_INST_RETIRED.LOCK_LOADS@ / cpu_core@MEM_INST_RETIRED.ALL_STORES@) + (1 - cpu_core@MEM_INST_RETIRED.LOCK_LOADS@ / cpu_core@MEM_INST_RETIRED.ALL_STORES@) * min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DEMAND_RFO@)) / tma_info_clks", "MetricGroup": "MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_issueRFO;tma_issueSL;tma_store_bound_group", "MetricName": "tma_store_latency", "MetricThreshold": "tma_store_latency > 0.1 & (tma_store_bound > 0.2 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2442,7 +2442,7 @@ }, { "BriefDescription": "This metric represents Core fraction of cycles CPU dispatched uops on execution port for Store operations", - "MetricExpr": "(UOPS_DISPATCHED.PORT_4_9 + UOPS_DISPATCHED.PORT_7_8) / (4 * tma_info_core_clks)", + "MetricExpr": "(cpu_core@UOPS_DISPATCHED.PORT_4_9@ + cpu_core@UOPS_DISPATCHED.PORT_7_8@) / (4 * tma_info_core_clks)", "MetricGroup": "TopdownL5;tma_L5_group;tma_ports_utilized_3m_group", "MetricName": "tma_store_op_utilization", "MetricThreshold": "tma_store_op_utilization > 0.6",
@@ -2470,7 +2470,7 @@ }, { "BriefDescription": "This metric estimates how often CPU was stalled due to Streaming store memory accesses; Streaming store optimize out a read request required by RFO stores", - "MetricExpr": "9 * OCR.STREAMING_WR.ANY_RESPONSE / tma_info_clks", + "MetricExpr": "9 * [email protected]_WR.ANY_RESPONSE@ / tma_info_clks", "MetricGroup": "MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueSmSt;tma_store_bound_group", "MetricName": "tma_streaming_stores", "MetricThreshold": "tma_streaming_stores > 0.2 & (tma_store_bound > 0.2 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2490,7 +2490,7 @@ }, { "BriefDescription": "This metric serves as an approximation of legacy x87 usage", - "MetricExpr": "tma_retiring * UOPS_EXECUTED.X87 / UOPS_EXECUTED.THREAD", + "MetricExpr": "tma_retiring * cpu_core@UOPS_EXECUTED.X87@ / UOPS_EXECUTED.THREAD", "MetricGroup": "Compute;TopdownL4;tma_L4_group;tma_fp_arith_group", "MetricName": "tma_x87_use", "MetricThreshold": "tma_x87_use > 0.1 & (tma_fp_arith > 0.2 & tma_light_operations > 0.6)",
tools/perf/pmu-events/arch/x86/alderlaken/adln-metrics.json +3 -3
@@ -193,7 +193,7 @@ { "BriefDescription": "Counts the number of cycles the core is stalled due to a demand load miss which hit in DRAM or MMIO (Non-DRAM).", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "MEM_BOUND_STALLS.LOAD_DRAM_HIT / tma_info_clks - MEM_BOUND_STALLS_AT_RET_CORRECTION * MEM_BOUND_STALLS.LOAD_DRAM_HIT / MEM_BOUND_STALLS.LOAD", + "MetricExpr": "MEM_BOUND_STALLS.LOAD_DRAM_HIT / tma_info_clks - max((MEM_BOUND_STALLS.LOAD - LD_HEAD.L1_MISS_AT_RET) / tma_info_clks, 0) * MEM_BOUND_STALLS.LOAD_DRAM_HIT / MEM_BOUND_STALLS.LOAD", "MetricGroup": "TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_dram_bound", "MetricThreshold": "tma_dram_bound > 0.1",
@@ -480,7 +480,7 @@ { "BriefDescription": "Counts the number of cycles a core is stalled due to a demand load which hit in the L2 Cache.", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "MEM_BOUND_STALLS.LOAD_L2_HIT / tma_info_clks - MEM_BOUND_STALLS_AT_RET_CORRECTION * MEM_BOUND_STALLS.LOAD_L2_HIT / MEM_BOUND_STALLS.LOAD", + "MetricExpr": "MEM_BOUND_STALLS.LOAD_L2_HIT / tma_info_clks - max((MEM_BOUND_STALLS.LOAD - LD_HEAD.L1_MISS_AT_RET) / tma_info_clks, 0) * MEM_BOUND_STALLS.LOAD_L2_HIT / MEM_BOUND_STALLS.LOAD", "MetricGroup": "TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_l2_bound", "MetricThreshold": "tma_l2_bound > 0.1",
@@ -488,7 +488,7 @@ }, { "BriefDescription": "Counts the number of cycles a core is stalled due to a demand load which hit in the Last Level Cache (LLC) or other core with HITE/F/M.", - "MetricExpr": "MEM_BOUND_STALLS.LOAD_LLC_HIT / tma_info_clks - MEM_BOUND_STALLS_AT_RET_CORRECTION * MEM_BOUND_STALLS.LOAD_LLC_HIT / MEM_BOUND_STALLS.LOAD", + "MetricExpr": "MEM_BOUND_STALLS.LOAD_LLC_HIT / tma_info_clks - max((MEM_BOUND_STALLS.LOAD - LD_HEAD.L1_MISS_AT_RET) / tma_info_clks, 0) * MEM_BOUND_STALLS.LOAD_LLC_HIT / MEM_BOUND_STALLS.LOAD", "MetricGroup": "TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_l3_bound", "MetricThreshold": "tma_l3_bound > 0.1",
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 43/46 perf jevents: Don't rewrite metrics across PMUs
In reply to Ian RogersDon't rewrite metrics across PMUs as the result events likely won't be found. Identify metrics with a pair of PMU name and metric name.
--- tools/perf/pmu-events/jevents.py | 4 ++-- tools/perf/pmu-events/metric.py | 28 +++++++++++++++++----------- tools/perf/pmu-events/metric_test.py | 6 +++--- 3 files changed, 22 insertions(+), 16 deletions(-)
tools/perf/pmu-events/jevents.py +2 -2
@@ -391,11 +391,11 @@ def read_json_events(path: str, topic: str) -> Sequence[JsonEvent]: except BaseException as err: print(f"Exception processing {path}") raise - metrics: list[Tuple[str, metric.Expression]] = [] + metrics: list[Tuple[str, str, metric.Expression]] = [] for event in events: event.topic = topic if event.metric_name and '-' not in event.metric_name: - metrics.append((event.metric_name, event.metric_expr)) + metrics.append((event.pmu, event.metric_name, event.metric_expr)) updates = metric.RewriteMetricsInTermsOfOthers(metrics) if updates: for event in events:
tools/perf/pmu-events/metric.py +17 -11
@@ -552,28 +552,34 @@ def ParsePerfJson(orig: str) -> Expression: return _Constify(eval(compile(parsed, orig, 'eval'))) -def RewriteMetricsInTermsOfOthers(metrics: List[Tuple[str, Expression]] - )-> Dict[str, Expression]: +def RewriteMetricsInTermsOfOthers(metrics: List[Tuple[str, str, Expression]] + )-> Dict[Tuple[str, str], Expression]: """Shorten metrics by rewriting in terms of others. Args: - metrics (list): pairs of metric names and their expressions. + metrics (list): pmus, metric names and their expressions. Returns: - Dict: mapping from a metric name to a shortened expression. + Dict: mapping from a pmu, metric name pair to a shortened expression. """ - updates: Dict[str, Expression] = dict() - for outer_name, outer_expression in metrics: + updates: Dict[Tuple[str, str], Expression] = dict() + for outer_pmu, outer_name, outer_expression in metrics: + if outer_pmu is None: + outer_pmu = 'cpu' updated = outer_expression while True: - for inner_name, inner_expression in metrics: + for inner_pmu, inner_name, inner_expression in metrics: + if inner_pmu is None: + inner_pmu = 'cpu' + if inner_pmu.lower() != outer_pmu.lower(): + continue if inner_name.lower() == outer_name.lower(): continue - if inner_name in updates: - inner_expression = updates[inner_name] + if (inner_pmu, inner_name) in updates: + inner_expression = updates[(inner_pmu, inner_name)] updated = updated.Substitute(inner_name, inner_expression) if updated.Equals(outer_expression): break - if outer_name in updates and updated.Equals(updates[outer_name]): + if (outer_pmu, outer_name) in updates and updated.Equals(updates[(outer_pmu, outer_name)]): break - updates[outer_name] = updated + updates[(outer_pmu, outer_name)] = updated return updates
tools/perf/pmu-events/metric_test.py +3 -3
@@ -158,9 +158,9 @@ class TestMetricExpressions(unittest.TestCase): def test_RewriteMetricsInTermsOfOthers(self): Expression.__eq__ = lambda e1, e2: e1.Equals(e2) - before = [('m1', ParsePerfJson('a + b + c + d')), - ('m2', ParsePerfJson('a + b + c'))] - after = {'m1': ParsePerfJson('m2 + d')} + before = [('cpu', 'm1', ParsePerfJson('a + b + c + d')), + ('cpu', 'm2', ParsePerfJson('a + b + c'))] + after = {('cpu', 'm1'): ParsePerfJson('m2 + d')} self.assertEqual(RewriteMetricsInTermsOfOthers(before), after) Expression.__eq__ = None
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 44/46 perf metrics: Be PMU specific in event match
In reply to Ian RogersIds/events from a metric are turned into an event string and parsed; setup_metric_events matches the id back to the parsed evsel. With hybrid the same event may exist on both PMUs with the same name and be being used by metrics at the same time. A metric on cpu_core therefore shouldn't match against evsels on cpu_atom, or the metric will compute the wrong value. Make the matching sensitive to the PMU being parsed.
--- tools/perf/util/metricgroup.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
tools/perf/util/metricgroup.c +7 -3
@@ -274,7 +274,7 @@ static int setup_metric_events(const char *pmu, struct hashmap *ids,const char *metric_id;struct evsel *ev;size_t ids_size, matched_events, i; -bool all_pmus = !strcmp (pmu,"all" ); +bool all_pmus = !strcmp (pmu,"all" ) || !perf_pmu__is_hybrid (pmu); *out_metric_events =NULL ; ids_size =hashmap__size (ids);
@@ -287,7 +287,10 @@ static int setup_metric_events(const char *pmu, struct hashmap *ids,evlist__for_each_entry (metric_evlist, ev) {struct expr_id_data *val_ptr; -if (!all_pmus &&strcmp (ev->pmu_name, pmu)) +/* Don't match events for the wrong hybrid PMU. */ +if (!all_pmus && ev->pmu_name && +perf_pmu__is_hybrid (ev->pmu_name) && +strcmp (ev->pmu_name, pmu))continue ;/* * Check for duplicate events with the same name. For
@@ -304,6 +307,7 @@ static int setup_metric_events(const char *pmu, struct hashmap *ids,* about this event. */ if (hashmap__find (ids, metric_id, &val_ptr)) { +pr_debug ("Matched metric-id %s to %s\n" , metric_id,evsel__name (ev)); metric_events[matched_events++] = ev;if (matched_events >= ids_size)
@@ -1592,7 +1596,7 @@ static int parse_groups(struct evlist *perf_evlist, ret =setup_metric_events (fake_pmu ?"all" : m->pmu, m->pctx->ids, metric_evlist, &metric_events);if (ret) { -pr_debug ("Cannot resolve IDs for %s: %s\n" , +pr_err ("Cannot resolve IDs for %s: %s\n" , m->metric_name, m->metric_expr); goto out; }
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 45/46 perf stat: Don't disable TopdownL1 metric on hybrid
In reply to Ian RogersNow that hybrid bugs are fixed sufficient to run TopdownL1 metrics, don't implicitly disable them for hybrid.
--- tools/perf/builtin-stat.c | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-)
tools/perf/builtin-stat.c +1 -6
@@ -1900,12 +1900,7 @@ static int add_default_attributes(void)* Add TopdownL1 metrics if they exist. To minimize * multiplexing, don't request threshold computation. */ -/* -* TODO: TopdownL1 is disabled on hybrid CPUs to avoid a crashes -* caused by exposing latent bugs. This is fixed properly in: -* https://lore.kernel.org/lkml/[email protected]/ -*/ -if (metricgroup__has_metric (pmu,"TopdownL1" ) && !perf_pmu__has_hybrid ()) { +if (metricgroup__has_metric (pmu,"TopdownL1" )) {struct evlist *metric_evlist =evlist__new ();struct evsel *metric_evsel;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Ian Rogers
reviewerPATCH v3 46/46 perf parse-events: Reduce scope of is_event_supported
In reply to Ian RogersMove to print-events.c and make static.
--- tools/perf/util/parse-events.c | 39 ---------------------------------- tools/perf/util/parse-events.h | 2 -- tools/perf/util/print-events.c | 39 ++++++++++++++++++++++++++++++++++ 3 files changed, 39 insertions(+), 41 deletions(-)
tools/perf/util/parse-events.c +0 -39
@@ -28,7 +28,6 @@#include "util/bpf-filter.h" #include "util/util.h" #include "tracepoint.h" -#include "thread_map.h" #define MAX_NAME_LEN 100
@@ -133,44 +132,6 @@ struct event_symbol event_symbols_sw[PERF_COUNT_SW_MAX] = { }, }; -bool is_event_supported (u8 type,u64 config) -{ -bool ret = true; -int open_return; -struct evsel *evsel; -struct perf_event_attr attr = { - .type = type, - .config = config, - .disabled =1 , - }; -struct perf_thread_map *tmap =thread_map__new_by_tid (0 ); - -if (tmap ==NULL ) -return false; - - evsel =evsel__new (&attr); -if (evsel) { - open_return =evsel__open (evsel,NULL , tmap); - ret = open_return >=0 ; - -if (open_return == -EACCES ) { -/* -* This happens if the paranoid value -* /proc/sys/kernel/perf_event_paranoid is set to 2 -* Re-run with exclude_kernel set; we don't do that -* by default as some ARM machines do not support it. -* -*/ - evsel->core.attr.exclude_kernel =1 ; - ret =evsel__open (evsel,NULL , tmap) >=0 ; - } -evsel__delete (evsel); - } - -perf_thread_map__put (tmap); -return ret; -} -const char *event_type (int type) {switch (type) {
tools/perf/util/parse-events.h +0 -2
@@ -18,8 +18,6 @@ struct parse_events_error;struct option ;struct perf_pmu ; -bool is_event_supported (u8 type,u64 config); -const char *event_type (int type);/* Arguments encoded in opt->value. */
tools/perf/util/print-events.c +39 -0
@@ -27,6 +27,7 @@#include "tracepoint.h" #include "pfm.h" #include "pmu-hybrid.h" +#include "thread_map.h" #define MAX_NAME_LEN 100
@@ -224,6 +225,44 @@ void print_sdt_events(const struct print_callbacks *print_cb, void *print_state)strlist__delete (sdtlist); } +static bool is_event_supported (u8 type,u64 config) +{ +bool ret = true; +int open_return; +struct evsel *evsel; +struct perf_event_attr attr = { + .type = type, + .config = config, + .disabled =1 , + }; +struct perf_thread_map *tmap =thread_map__new_by_tid (0 ); + +if (tmap ==NULL ) +return false; + + evsel =evsel__new (&attr); +if (evsel) { + open_return =evsel__open (evsel,NULL , tmap); + ret = open_return >=0 ; + +if (open_return == -EACCES ) { +/* +* This happens if the paranoid value +* /proc/sys/kernel/perf_event_paranoid is set to 2 +* Re-run with exclude_kernel set; we don't do that +* by default as some ARM machines do not support it. +* +*/ + evsel->core.attr.exclude_kernel =1 ; + ret =evsel__open (evsel,NULL , tmap) >=0 ; + } +evsel__delete (evsel); + } + +perf_thread_map__put (tmap); +return ret; +} +int print_hwcache_events (const struct print_callbacks *print_cb,void *print_state) {struct perf_pmu *pmu =NULL ;
Signature · 1 line
2.40.1.495.gc816e09b53d-goog
Quoted 4 lines · Ian Rogers
Sample output is contained in the v1 patch set:
https://lore.kernel.org/lkml/[email protected]/Tested on Tigerlake, Skylake and Alderlake CPUs.
The metrics and metricgroups tests are failed on a hybrid machine. Seems we need the no group trick for more metrics?
./perf test "perf all metricgroups test"
102: perf all metricgroups test :
FAILED!
Testing DataSharing
Error:
The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (cpu_core/topdown-fe-bound/)./bin/dmesg | grep -i perf may provide additional information.
test child finished with -1
---- end ----
perf all metricgroups test: FAILED!
./perf test "perf all metrics test"
103: perf all metrics test :
FAILED!
Testing tma_info_branctest child finished with -1
---- end ----
perf all metrics test: FAILED!
Testing tma_contested_accesses
Metric 'tma_contested_accesses' not printed in:
Error:
Invalid event (cpu_core/topdown-fe-bound/) in per-thread mode, enable system wide with '-a'.
Metric 'tma_dram_bound' not printed in:
Error:
Invalid event (cpu_core/topdown-mem-bound/) in per-thread mode, enable system wide with '-a'.
Metric 'tma_data_sharing' not printed in:
Error:
Invalid event (cpu_core/topdown-fe-bound/) in per-thread mode, enable system wide with '-a'.
Thanks,
Kan
Ian Rogers
reviewerQuoted 50 lines · Kan
Quoted 4 lines · Ian Rogers
Sample output is contained in the v1 patch set:
https://lore.kernel.org/lkml/[email protected]/Tested on Tigerlake, Skylake and Alderlake CPUs.
The metrics and metricgroups tests are failed on a hybrid machine. Seems we need the no group trick for more metrics?
./perf test "perf all metricgroups test"
102: perf all metricgroups test :
FAILED!Testing DataSharing
Error:
Thesys_perf_event_open()syscall returned with 22 (Invalid argument) for event (cpu_core/topdown-fe-bound/)./bin/dmesg| grep -i perf may provide additional information.test child finished with -1
---- end ----
perf all metricgroups test: FAILED!./perf test "perf all metrics test"
103: perf all metrics test :
FAILED!Testing
tma_info_branctestchild finished with -1
---- end ----
perf all metrics test: FAILED!Testing
tma_contested_accesses
Metric 'tma_contested_accesses' not printed in:
Error:
Invalid event (cpu_core/topdown-fe-bound/) in per-thread mode, enable system wide with '-a'.Metric '
tma_dram_bound' not printed in:
Error:
Invalid event (cpu_core/topdown-mem-bound/) in per-thread mode, enable system wide with '-a'.
Quoted 8 lines
Metric '
tma_data_sharing' not printed in:
Error:
Invalid event (cpu_core/topdown-fe-bound/) in per-thread mode, enable system wide with '-a'.Thanks,
Kan
This is expected. Metrics test fails on Icelake, metricgroups doesn't validate output but I can expect metricgroups like "transaction" to have issues. We want to have better tests and this comes under what we're hoping to achieve with the Valkyrie project.
Thanks,
Ian