perf report: Add latency and parallelism profiling
v4 cover perf report: Add latency and parallelism profiling
There are two notions of time: wall-clock time and CPU time. For a single-threaded program, or a program running on a single-core machine, these notions are the same. However, for a multi-threaded/ multi-process program running on a multi-core machine, these notions are significantly different. Each second of wall-clock time we have number-of-cores seconds of CPU time.
Currently perf only allows to profile CPU time. Perf (and all other existing profilers to the be best of my knowledge) does not allow to profile wall-clock time.
Optimizing CPU overhead is useful to improve 'throughput', while optimizing wall-clock overhead is useful to improve 'latency'. These profiles are complementary and are not interchangeable. Examples of where latency profile is needed:
- optimzing build latency
- optimizing server request latency
- optimizing ML training/inference latency
- optimizing running time of any command line program
CPU profile is useless for these use cases at best (if a user understands the difference), or misleading at worst (if a user tries to use a wrong profile for a job).
This series add latency and parallelization profiling. See the added documentation and flags descriptions for details.
Brief outline of the implementation:
- add context switch collection during record
- calculate number of threads running on CPUs (parallelism level) during report
- divide each sample weight by the parallelism level
This effectively models that we were taking 1 sample per unit of wall-clock time.
We still default to the CPU profile, so it's up to users to learn about the second profiling mode and use it when appropriate.
Changes in v4:
- added "Shrink struct hist_entry size" commit
- rebased to perf-tools-next HEAD
Changes in v3:
- rebase and split into patches
- rename 'wallclock' to 'latency' everywhere
- don't enable latency profiling by default,
instead add record/report --latency flag
Dmitry Vyukov (8):
perf report: Add machine parallelism
perf report: Add parallelism sort key
perf report: Switch filtered from u8 to u16
perf report: Add parallelism filter
perf report: Add latency output field
perf report: Add --latency flag
perf report: Add latency and parallelism profiling documentation perf hist: Shrink struct hist_entry size
.../callchain-overhead-calculation.txt | 5 +- .../cpu-and-latency-overheads.txt | 85 ++++++++++++++++++ tools/perf/Documentation/perf-report.txt | 49 ++++++---- tools/perf/Documentation/tips.txt | 3 + tools/perf/builtin-record.c | 20 +++++ tools/perf/builtin-report.c | 39 ++++++++ tools/perf/ui/browsers/hists.c | 27 +++--- tools/perf/ui/hist.c | 64 +++++++++---- tools/perf/util/addr_location.c | 1 + tools/perf/util/addr_location.h | 7 +- tools/perf/util/event.c | 11 +++ tools/perf/util/events_stats.h | 2 + tools/perf/util/hist.c | 90 +++++++++++++++---- tools/perf/util/hist.h | 32 +++++-- tools/perf/util/machine.c | 7 ++ tools/perf/util/machine.h | 6 ++ tools/perf/util/sample.h | 2 +- tools/perf/util/session.c | 12 +++ tools/perf/util/session.h | 1 + tools/perf/util/sort.c | 69 ++++++++++++-- tools/perf/util/sort.h | 3 +- tools/perf/util/symbol.c | 34 +++++++ tools/perf/util/symbol_conf.h | 8 +- 23 files changed, 502 insertions(+), 75 deletions(-) create mode 100644 tools/perf/Documentation/cpu-and-latency-overheads.txt
base-commit: 8ce0d2da14d3fb62844dd0e95982c194326b1a5f
Signature · 1 line
2.48.1.362.g079036d154-goog
PATCH v3 1/8 perf report: Add machine parallelism
Add calculation of the current parallelism level (number of threads actively running on CPUs). The parallelism level can be shown in reports on its own, and to calculate latency overheads.
--- tools/perf/builtin-report.c | 1 + tools/perf/util/addr_location.c | 1 + tools/perf/util/addr_location.h | 2 ++ tools/perf/util/event.c | 3 +++ tools/perf/util/machine.c | 7 +++++++ tools/perf/util/machine.h | 6 ++++++ 6 files changed, 20 insertions(+)
tools/perf/builtin-report.c +1 -0
@@ -1568,6 +1568,7 @@ int cmd_report(int argc, const char **argv) report.tool.cgroup = perf_event__process_cgroup; report.tool.exit = perf_event__process_exit; report.tool.fork = perf_event__process_fork; + report.tool.context_switch = perf_event__process_switch; report.tool.lost = perf_event__process_lost; report.tool.read = process_read_event; report.tool.attr = process_attr;
tools/perf/util/addr_location.c +1 -0
@@ -17,6 +17,7 @@ void addr_location__init(struct addr_location *al) al->cpumode =0 ; al->cpu =0 ; al->socket =0 ; + al->parallelism =1 ; }/*
tools/perf/util/addr_location.h +2 -0
@@ -21,6 +21,8 @@ struct addr_location {u8 cpumode;s32 cpu;s32 socket; +/* Same as machine.parallelism but within [1, nr_cpus]. */ +int parallelism; };void addr_location__init (struct addr_location *al);
tools/perf/util/event.c +3 -0
@@ -767,6 +767,9 @@ int machine__resolve(struct machine *machine, struct addr_location *al, al->socket = env->cpu[al->cpu].socket_id; } +/* Account for possible out-of-order switch events. */ + al->parallelism =min (max (machine->parallelism,1 ),machine__nr_cpus_avail (machine)); +if (al->map) { if (symbol_conf.dso_list && (!dso || !(strlist__has_entry(symbol_conf.dso_list,
tools/perf/util/machine.c +7 -0
@@ -94,6 +94,8 @@ int machine__init(struct machine *machine, const char *root_dir, pid_t pid) machine->comm_exec = false; machine->kernel_start =0 ; machine->vmlinux_map =NULL ; +/* There is no initial context switch in, so we start at 1. */ + machine->parallelism =1 ; machine->root_dir =strdup (root_dir);if (machine->root_dir ==NULL )
@@ -677,8 +679,11 @@ int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unusint machine__process_switch_event (struct machine *machine__maybe_unused ,union perf_event *event) { +bool out = event->header.misc &PERF_RECORD_MISC_SWITCH_OUT ; +if (dump_trace)perf_event__fprintf_switch (event, stdout); + machine->parallelism += out ?-1 :1 ;return 0 ; }
@@ -1900,6 +1905,8 @@ int machine__process_exit_event(struct machine *machine, union perf_event *eventif (dump_trace)perf_event__fprintf_task (event, stdout); +/* There is no context switch out before exit, so we decrement here. */ + machine->parallelism--;if (thread !=NULL ) {if (symbol_conf.keep_exited_threads)thread__set_exited (thread,/*exited=*/ true);
tools/perf/util/machine.h +6 -0
@@ -50,6 +50,12 @@ struct machine {u64 text_start;u64 text_end; } sched, lock, traceiter, trace; +/* +* The current parallelism level (number of threads that run on CPUs). +* This value can be less than 1, or larger than the total number +* of CPUs, if events are poorly ordered. +*/ +int parallelism;pid_t *current_tid;size_t current_tid_sz;union {/* Tool specific area */
Signature · 1 line
2.48.1.362.g079036d154-goog
PATCH v3 2/8 perf report: Add parallelism sort key
In reply to Dmitry VyukovShow parallelism level in profiles if requested by user.
--- tools/perf/builtin-report.c | 11 +++++++++++ tools/perf/util/hist.c | 2 ++ tools/perf/util/hist.h | 3 +++ tools/perf/util/session.c | 12 ++++++++++++ tools/perf/util/session.h | 1 + tools/perf/util/sort.c | 23 +++++++++++++++++++++++ tools/perf/util/sort.h | 1 + 7 files changed, 53 insertions(+)
tools/perf/builtin-report.c +11 -0
@@ -1720,6 +1720,17 @@ int cmd_report(int argc, const char **argv) symbol_conf.annotate_data_sample = true; } +if (report.disable_order || !perf_session__has_switch_events (session)) { +if ((sort_order &&strstr (sort_order,"parallelism" )) || + (field_order &&strstr (field_order,"parallelism" ))) { +if (report.disable_order) +ui__error ("Use of parallelism is incompatible with --disable-order.\n" ); +else +ui__error ("Use of parallelism requires --switch-events during record.\n" ); +return -1 ; + } + } +if (sort_order &&strstr (sort_order,"ipc" )) {parse_options_usage (report_usage, options,"s" ,1 ); goto error;
tools/perf/util/hist.c +2 -0
@@ -207,6 +207,7 @@ void hists__calc_col_len(struct hists *hists, struct hist_entry *h)hists__new_col_len (hists,HISTC_CGROUP ,6 );hists__new_col_len (hists,HISTC_CGROUP_ID ,20 ); +hists__new_col_len (hists,HISTC_PARALLELISM ,11 );hists__new_col_len (hists,HISTC_CPU ,3 );hists__new_col_len (hists,HISTC_SOCKET ,6 );hists__new_col_len (hists,HISTC_MEM_LOCKED ,6 );
@@ -741,6 +742,7 @@ __hists__add_entry(struct hists *hists, .ip = al->addr, .level = al->level, .code_page_size = sample->code_page_size, + .parallelism = al->parallelism, .stat = { .nr_events = 1, .period = sample->period,
tools/perf/util/hist.h +3 -0
@@ -42,6 +42,7 @@ enum hist_column {HISTC_CGROUP_ID ,HISTC_CGROUP ,HISTC_PARENT , +HISTC_PARALLELISM ,HISTC_CPU ,HISTC_SOCKET , HISTC_SRCLINE,
@@ -228,6 +229,7 @@ struct hist_entry {u64 transaction;s32 socket;s32 cpu; +int parallelism;u64 code_page_size;u64 weight;u64 ins_lat;
@@ -580,6 +582,7 @@ bool perf_hpp__is_thread_entry(struct perf_hpp_fmt *fmt);bool perf_hpp__is_comm_entry (struct perf_hpp_fmt *fmt);bool perf_hpp__is_dso_entry (struct perf_hpp_fmt *fmt);bool perf_hpp__is_sym_entry (struct perf_hpp_fmt *fmt); +bool perf_hpp__is_parallelism_entry (struct perf_hpp_fmt *fmt);struct perf_hpp_fmt *perf_hpp_fmt__dup (struct perf_hpp_fmt *fmt);
tools/perf/util/session.c +12 -0
@@ -2403,6 +2403,18 @@ bool perf_session__has_traces(struct perf_session *session, const char *msg)return false; } +bool perf_session__has_switch_events (struct perf_session *session) +{ +struct evsel *evsel; + +evlist__for_each_entry (session->evlist, evsel) { +if (evsel->core.attr.context_switch) +return true; + } + +return false; +} +int map__set_kallsyms_ref_reloc_sym (struct map *map,const char *symbol_name,u64 addr) {char *bracket;
tools/perf/util/session.h +1 -0
@@ -141,6 +141,7 @@ int perf_session__resolve_callchain(struct perf_session *session, struct symbol **parent);bool perf_session__has_traces (struct perf_session *session,const char *msg); +bool perf_session__has_switch_events (struct perf_session *session);void perf_event__attr_swap (struct perf_event_attr *attr);
tools/perf/util/sort.c +23 -0
@@ -892,6 +892,27 @@ struct sort_entry sort_cpu = { .se_width_idx = HISTC_CPU, }; +/* --sort parallelism */ + +static int64_t +sort__parallelism_cmp (struct hist_entry *left,struct hist_entry *right) +{ +return right->parallelism - left->parallelism; +} + +static int hist_entry__parallelism_snprintf (struct hist_entry *he,char *bf, +size_t size,unsigned int width) +{ +return repsep_snprintf (bf, size,"%*d" , width, he->parallelism); +} + +struct sort_entry sort_parallelism = { + .se_header ="Parallelism" , + .se_cmp = sort__parallelism_cmp, + .se_snprintf = hist_entry__parallelism_snprintf, + .se_width_idx =HISTC_PARALLELISM , +}; +/* --sort cgroup_id */ static int64_t _sort__cgroup_dev_cmp (u64 left_dev,u64 right_dev)
@@ -2534,6 +2555,7 @@ static struct sort_dimension common_sort_dimensions[] = {DIM (SORT_ANNOTATE_DATA_TYPE_OFFSET ,"typeoff" , sort_type_offset),DIM (SORT_SYM_OFFSET ,"symoff" , sort_sym_offset),DIM (SORT_ANNOTATE_DATA_TYPE_CACHELINE ,"typecln" , sort_type_cacheline), + DIM(SORT_PARALLELISM, "parallelism", sort_parallelism), };#undef DIM
@@ -2735,6 +2757,7 @@ MK_SORT_ENTRY_CHK(thread) MK_SORT_ENTRY_CHK(comm) MK_SORT_ENTRY_CHK(dso) MK_SORT_ENTRY_CHK(sym) +MK_SORT_ENTRY_CHK(parallelism) static bool __sort__hpp_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
tools/perf/util/sort.h +1 -0
@@ -72,6 +72,7 @@ enum sort_type {SORT_ANNOTATE_DATA_TYPE_OFFSET ,SORT_SYM_OFFSET ,SORT_ANNOTATE_DATA_TYPE_CACHELINE , +SORT_PARALLELISM ,/* branch stack specific sort keys */ __SORT_BRANCH_STACK,
Signature · 1 line
2.48.1.362.g079036d154-goog
PATCH v3 3/8 perf report: Switch filtered from u8 to u16
In reply to Dmitry VyukovWe already have all u8 bits taken, adding one more filter leads to unpleasant failure mode, where code compiles w/o warnings, but the last filters silently don't work. Add a typedef and switch to u16.
--- tools/perf/util/addr_location.h | 3 ++- tools/perf/util/hist.c | 2 +- tools/perf/util/hist.h | 4 +++- 3 files changed, 6 insertions(+), 3 deletions(-)
tools/perf/util/addr_location.h +2 -1
@@ -3,6 +3,7 @@#define __PERF_ADDR_LOCATION 1#include <linux/types.h> +#include "hist.h" struct thread ;struct maps ;
@@ -17,7 +18,7 @@ struct addr_location {const char *srcline;u64 addr;char level; -u8 filtered; +filter_mask_t filtered;u8 cpumode;s32 cpu;s32 socket;
tools/perf/util/hist.c +1 -1
@@ -585,7 +585,7 @@ static struct hist_entry *hist_entry__new(struct hist_entry *template,return he; } -static u8 symbol__parent_filter (const struct symbol *parent) +static filter_mask_t symbol__parent_filter (const struct symbol *parent) {if (symbol_conf.exclude_other && parent ==NULL )return 1 <<HIST_FILTER__PARENT ;
tools/perf/util/hist.h +3 -1
@@ -33,6 +33,8 @@ enum hist_filter { HIST_FILTER__C2C, }; +typedef u16 filter_mask_t ; +enum hist_column {HISTC_SYMBOL ,HISTC_TIME ,
@@ -244,7 +246,7 @@ struct hist_entry {bool leaf;char level; -u8 filtered; +filter_mask_t filtered;u16 callchain_size;union {
Signature · 1 line
2.48.1.362.g079036d154-goog
PATCH v3 4/8 perf report: Add parallelism filter
In reply to Dmitry VyukovAdd parallelism filter that can be used to look at specific parallelism levels only. The format is the same as cpu lists. For example:
Only single-threaded samples: --parallelism=1 Low parallelism only: --parallelism=1-4 High parallelism only: --parallelism=64-128
--- tools/perf/builtin-report.c | 5 ++++- tools/perf/util/event.c | 2 ++ tools/perf/util/hist.c | 31 +++++++++++++++++++++++++++++++ tools/perf/util/hist.h | 6 +++++- tools/perf/util/sort.c | 11 +++++++++++ tools/perf/util/symbol.c | 34 ++++++++++++++++++++++++++++++++++ tools/perf/util/symbol_conf.h | 4 ++++ 7 files changed, 91 insertions(+), 2 deletions(-)
tools/perf/builtin-report.c +4 -1
@@ -1390,6 +1390,8 @@ int cmd_report(int argc, const char **argv) symbol__config_symfs),OPT_STRING ('C' ,"cpu" , &report.cpu_list,"cpu" ,"list of cpus to profile" ), +OPT_STRING (0 ,"parallelism" , &symbol_conf.parallelism_list_str,"parallelism" , +"only consider these parallelism levels (cpu set format)" ),OPT_BOOLEAN ('I' ,"show-info" , &report.show_full_info,"Display extended information about perf.data file" ), OPT_BOOLEAN(0, "source", &annotate_opts.annotate_src,
@@ -1721,7 +1723,8 @@ int cmd_report(int argc, const char **argv) }if (report.disable_order || !perf_session__has_switch_events (session)) { -if ((sort_order &&strstr (sort_order,"parallelism" )) || +if (symbol_conf.parallelism_list_str || + (sort_order &&strstr (sort_order,"parallelism" )) || (field_order &&strstr (field_order,"parallelism" ))) {if (report.disable_order)ui__error ("Use of parallelism is incompatible with --disable-order.\n" );
tools/perf/util/event.c +2 -0
@@ -769,6 +769,8 @@ int machine__resolve(struct machine *machine, struct addr_location *al,/* Account for possible out-of-order switch events. */ al->parallelism =min (max (machine->parallelism,1 ),machine__nr_cpus_avail (machine)); +if (test_bit (al->parallelism, symbol_conf.parallelism_filter)) + al->filtered |= (1 <<HIST_FILTER__PARALLELISM );if (al->map) { if (symbol_conf.dso_list &&
tools/perf/util/hist.c +31 -0
@@ -43,6 +43,8 @@ static bool hists__filter_entry_by_symbol(struct hists *hists, struct hist_entry *he);static bool hists__filter_entry_by_socket (struct hists *hists,struct hist_entry *he); +static bool hists__filter_entry_by_parallelism (struct hists *hists, +struct hist_entry *he);u16 hists__col_len (struct hists *hists,enum hist_column col) {
@@ -1457,6 +1459,10 @@ static void hist_entry__check_and_remove_filter(struct hist_entry *he,if (symbol_conf.sym_list ==NULL )return ;break ; +case HIST_FILTER__PARALLELISM : +if (__bitmap_weight (symbol_conf.parallelism_filter,MAX_NR_CPUS +1 ) ==0 ) +return ; +break ;case HIST_FILTER__PARENT :case HIST_FILTER__GUEST :case HIST_FILTER__HOST :
@@ -1515,6 +1521,9 @@ static void hist_entry__apply_hierarchy_filters(struct hist_entry *he)hist_entry__check_and_remove_filter (he,HIST_FILTER__SYMBOL , perf_hpp__is_sym_entry); +hist_entry__check_and_remove_filter (he,HIST_FILTER__PARALLELISM , + perf_hpp__is_parallelism_entry); +hists__apply_filters (he->hists, he); }
@@ -1711,6 +1720,7 @@ static void hists__apply_filters(struct hists *hists, struct hist_entry *he)hists__filter_entry_by_thread (hists, he);hists__filter_entry_by_symbol (hists, he);hists__filter_entry_by_socket (hists, he); +hists__filter_entry_by_parallelism (hists, he); }int hists__collapse_resort (struct hists *hists,struct ui_progress *prog)
@@ -2197,6 +2207,16 @@ static bool hists__filter_entry_by_socket(struct hists *hists,return false; } +static bool hists__filter_entry_by_parallelism (struct hists *hists, +struct hist_entry *he) +{ +if (test_bit (he->parallelism, hists->parallelism_filter)) { + he->filtered |= (1 <<HIST_FILTER__PARALLELISM ); +return true; + } +return false; +} +typedef bool (*filter_fn_t )(struct hists *hists,struct hist_entry *he);static void hists__filter_by_type (struct hists *hists,int type,filter_fn_t filter)
@@ -2366,6 +2386,16 @@ void hists__filter_by_socket(struct hists *hists) hists__filter_entry_by_socket); } +void hists__filter_by_parallelism (struct hists *hists) +{ +if (symbol_conf.report_hierarchy) +hists__filter_hierarchy (hists,HIST_FILTER__PARALLELISM , + hists->parallelism_filter); +else +hists__filter_by_type (hists,HIST_FILTER__PARALLELISM , + hists__filter_entry_by_parallelism); +} +void events_stats__inc (struct events_stats *stats,u32 type) { ++stats->nr_events[0 ];
@@ -2872,6 +2902,7 @@ int __hists__init(struct hists *hists, struct perf_hpp_list *hpp_list) hists->entries =RB_ROOT_CACHED ;mutex_init (&hists->lock); hists->socket_filter =-1 ; + hists->parallelism_filter = symbol_conf.parallelism_filter; hists->hpp_list = hpp_list;INIT_LIST_HEAD (&hists->hpp_formats);return 0 ;
tools/perf/util/hist.h +5 -1
@@ -31,6 +31,7 @@ enum hist_filter {HIST_FILTER__HOST ,HIST_FILTER__SOCKET ,HIST_FILTER__C2C , + HIST_FILTER__PARALLELISM, };typedef u16 filter_mask_t ;
@@ -112,6 +113,7 @@ struct hists {const struct dso *dso_filter;const char *uid_filter_str;const char *symbol_filter_str; +unsigned long *parallelism_filter;struct mutex lock;struct hists_stats stats;u64 event_stream;
@@ -388,11 +390,13 @@ void hists__filter_by_dso(struct hists *hists);void hists__filter_by_thread (struct hists *hists);void hists__filter_by_symbol (struct hists *hists);void hists__filter_by_socket (struct hists *hists); +void hists__filter_by_parallelism (struct hists *hists);static inline bool hists__has_filter (struct hists *hists) {return hists->thread_filter || hists->dso_filter || - hists->symbol_filter_str || (hists->socket_filter >-1 ); + hists->symbol_filter_str || (hists->socket_filter >-1 ) || + hists->parallelism_filter; }u16 hists__col_len (struct hists *hists,enum hist_column col);
tools/perf/util/sort.c +11 -0
@@ -900,6 +900,16 @@ sort__parallelism_cmp(struct hist_entry *left, struct hist_entry *right)return right->parallelism - left->parallelism; } +static int hist_entry__parallelism_filter (struct hist_entry *he,int type,const void *arg) +{ +const unsigned long *parallelism_filter = arg; + +if (type !=HIST_FILTER__PARALLELISM ) +return -1 ; + +return test_bit (he->parallelism, parallelism_filter); +} +static int hist_entry__parallelism_snprintf (struct hist_entry *he,char *bf,size_t size,unsigned int width) {
@@ -909,6 +919,7 @@ static int hist_entry__parallelism_snprintf(struct hist_entry *he, char *bf,struct sort_entry sort_parallelism = { .se_header ="Parallelism" , .se_cmp = sort__parallelism_cmp, + .se_filter = hist_entry__parallelism_filter, .se_snprintf = hist_entry__parallelism_snprintf, .se_width_idx =HISTC_PARALLELISM , };
tools/perf/util/symbol.c +34 -0
@@ -18,6 +18,7 @@#include "annotate.h" #include "build-id.h" #include "cap.h" +#include "cpumap.h" #include "dso.h" #include "util.h" // lsdir() #include "debug.h"
@@ -2471,6 +2472,36 @@ int symbol__annotation_init(void)return 0 ; } +static int setup_parallelism_bitmap (void ) +{ +struct perf_cpu_map *map; +struct perf_cpu cpu; +int i, err =-1 ; + +if (symbol_conf.parallelism_list_str ==NULL ) +return 0 ; + + map =perf_cpu_map__new (symbol_conf.parallelism_list_str); +if (map ==NULL ) { +pr_err ("failed to parse parallelism filter list\n" ); +return -1 ; + } + +bitmap_fill (symbol_conf.parallelism_filter,MAX_NR_CPUS +1 ); +perf_cpu_map__for_each_cpu (cpu, i, map) { +if (cpu.cpu <=0 || cpu.cpu >MAX_NR_CPUS ) { +pr_err ("Requested parallelism level %d is invalid.\n" , cpu.cpu); + goto out_delete_map; + } +__clear_bit (cpu.cpu, symbol_conf.parallelism_filter); + } + + err =0 ; +out_delete_map: +perf_cpu_map__put (map); +return err; +} +int symbol__init (struct perf_env *env) {const char *symfs;
@@ -2490,6 +2521,9 @@ int symbol__init(struct perf_env *env)return -1 ; } +if (setup_parallelism_bitmap ()) +return -1 ; +if (setup_list (&symbol_conf.dso_list, symbol_conf.dso_list_str,"dso" ) <0 )return -1 ;
tools/perf/util/symbol_conf.h +4 -0
@@ -3,6 +3,8 @@#define __PERF_SYMBOL_CONF 1#include <stdbool.h> +#include <linux/bitmap.h> +#include "perf.h" struct strlist ;struct intlist ;
@@ -62,6 +64,7 @@ struct symbol_conf { *pid_list_str, *tid_list_str, *sym_list_str, + *parallelism_list_str, *col_width_list_str, *bt_stop_list_str;const char *addr2line_path;
@@ -82,6 +85,7 @@ struct symbol_conf {int pad_output_len_dso;int group_sort_idx;int addr_range; +DECLARE_BITMAP (parallelism_filter,MAX_NR_CPUS +1 ); };extern struct symbol_conf symbol_conf;
Signature · 1 line
2.48.1.362.g079036d154-goog
PATCH v3 5/8 perf report: Add latency output field
In reply to Dmitry VyukovLatency output field is similar to overhead, but represents overhead for latency rather than CPU consumption. It's re-scaled from overhead by dividing weight by the current parallelism level at the time of the sample. It effectively models profiling with 1 sample taken per unit of wall-clock time rather than unit of CPU time.
--- tools/perf/ui/browsers/hists.c | 27 ++++++++++------ tools/perf/ui/hist.c | 29 ++++++++++------- tools/perf/util/addr_location.h | 2 ++ tools/perf/util/event.c | 6 ++++ tools/perf/util/events_stats.h | 2 ++ tools/perf/util/hist.c | 55 ++++++++++++++++++++++++--------- tools/perf/util/hist.h | 12 +++++++ tools/perf/util/sort.c | 2 ++ 8 files changed, 100 insertions(+), 35 deletions(-)
tools/perf/ui/browsers/hists.c +17 -10
@@ -1226,7 +1226,7 @@ int __hpp__slsmg_color_printf(struct perf_hpp *hpp, const char *fmt, ...)return ret; } -#define __HPP_COLOR_PERCENT_FN (_type, _field) \ +#define __HPP_COLOR_PERCENT_FN (_type, _field, _fmttype) \ static u64 __hpp_get_##_field(struct hist_entry *he) \ { \ return he->stat._field; \
@@ -1238,10 +1238,10 @@ hist_browser__hpp_color_##_type(struct perf_hpp_fmt *fmt, \ struct hist_entry *he) \ { \ return hpp__fmt(fmt, hpp, he, __hpp_get_##_field, " %*.2f%%", \ - __hpp__slsmg_color_printf, true); \ + __hpp__slsmg_color_printf, _fmttype); \ } -#define __HPP_COLOR_ACC_PERCENT_FN (_type, _field) \ +#define __HPP_COLOR_ACC_PERCENT_FN (_type, _field, _fmttype) \ static u64 __hpp_get_acc_##_field(struct hist_entry *he) \ { \ return he->stat_acc->_field; \
@@ -1262,15 +1262,18 @@ hist_browser__hpp_color_##_type(struct perf_hpp_fmt *fmt, \return ret; \ } \ return hpp__fmt(fmt, hpp, he, __hpp_get_acc_##_field, \ -" %*.2f%%" , __hpp__slsmg_color_printf, true); \ +" %*.2f%%" , __hpp__slsmg_color_printf, \ + _fmttype); \ } -__HPP_COLOR_PERCENT_FN(overhead, period) -__HPP_COLOR_PERCENT_FN(overhead_sys, period_sys) -__HPP_COLOR_PERCENT_FN(overhead_us, period_us) -__HPP_COLOR_PERCENT_FN(overhead_guest_sys, period_guest_sys) -__HPP_COLOR_PERCENT_FN(overhead_guest_us, period_guest_us) -__HPP_COLOR_ACC_PERCENT_FN(overhead_acc, period) +__HPP_COLOR_PERCENT_FN(overhead, period, PERF_HPP_FMT_TYPE__PERCENT) +__HPP_COLOR_PERCENT_FN(latency, latency, PERF_HPP_FMT_TYPE__LATENCY) +__HPP_COLOR_PERCENT_FN(overhead_sys, period_sys, PERF_HPP_FMT_TYPE__PERCENT) +__HPP_COLOR_PERCENT_FN(overhead_us, period_us, PERF_HPP_FMT_TYPE__PERCENT) +__HPP_COLOR_PERCENT_FN(overhead_guest_sys, period_guest_sys, PERF_HPP_FMT_TYPE__PERCENT) +__HPP_COLOR_PERCENT_FN(overhead_guest_us, period_guest_us, PERF_HPP_FMT_TYPE__PERCENT) +__HPP_COLOR_ACC_PERCENT_FN(overhead_acc, period, PERF_HPP_FMT_TYPE__PERCENT) +__HPP_COLOR_ACC_PERCENT_FN(latency_acc, latency, PERF_HPP_FMT_TYPE__LATENCY)#undef __HPP_COLOR_PERCENT_FN#undef __HPP_COLOR_ACC_PERCENT_FN
@@ -1279,6 +1282,8 @@ void hist_browser__init_hpp(void) { perf_hpp__format[PERF_HPP__OVERHEAD ].color = hist_browser__hpp_color_overhead; + perf_hpp__format[PERF_HPP__LATENCY ].color = + hist_browser__hpp_color_latency; perf_hpp__format[PERF_HPP__OVERHEAD_SYS ].color = hist_browser__hpp_color_overhead_sys; perf_hpp__format[PERF_HPP__OVERHEAD_US].color =
@@ -1289,6 +1294,8 @@ void hist_browser__init_hpp(void) hist_browser__hpp_color_overhead_guest_us; perf_hpp__format[PERF_HPP__OVERHEAD_ACC ].color = hist_browser__hpp_color_overhead_acc; + perf_hpp__format[PERF_HPP__LATENCY_ACC ].color = + hist_browser__hpp_color_latency_acc;res_sample_init (); }
tools/perf/ui/hist.c +18 -11
@@ -27,9 +27,10 @@ static int __hpp__fmt_print(struct perf_hpp *hpp, struct hists *hists, u64 val, int nr_samples, const char *fmt, int len, hpp_snprint_fn print_fn, enum perf_hpp_fmt_type fmtype) { - if (fmtype == PERF_HPP_FMT_TYPE__PERCENT) { + if (fmtype == PERF_HPP_FMT_TYPE__PERCENT || fmtype == PERF_HPP_FMT_TYPE__LATENCY) { double percent = 0.0; -u64 total =hists__total_period (hists); +u64 total = fmtype ==PERF_HPP_FMT_TYPE__PERCENT ?hists__total_period (hists) : +hists__total_latency (hists);if (total) percent =100.0 * val / total;
@@ -128,7 +129,7 @@ int hpp__fmt(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, print_fn, fmtype); } -if (fmtype ==PERF_HPP_FMT_TYPE__PERCENT ) +if (fmtype ==PERF_HPP_FMT_TYPE__PERCENT || fmtype ==PERF_HPP_FMT_TYPE__LATENCY ) len -=2 ;/* 2 for a space and a % sign */ else len -=1 ;
@@ -472,11 +473,13 @@ __HPP_ENTRY_AVERAGE_FN(_type, _field) \ __HPP_SORT_AVERAGE_FN(_type, _field) HPP_PERCENT_FNS(overhead, period) +HPP_PERCENT_FNS(latency, latency) HPP_PERCENT_FNS(overhead_sys, period_sys) HPP_PERCENT_FNS(overhead_us, period_us) HPP_PERCENT_FNS(overhead_guest_sys, period_guest_sys) HPP_PERCENT_FNS(overhead_guest_us, period_guest_us) HPP_PERCENT_ACC_FNS(overhead_acc, period) +HPP_PERCENT_ACC_FNS(latency_acc, latency) HPP_RAW_FNS(samples, nr_events)HPP_RAW_FNS (period, period)
@@ -548,11 +551,13 @@ static bool hpp__equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)struct perf_hpp_fmt perf_hpp__format[] = {HPP__COLOR_PRINT_FNS ("Overhead" , overhead,OVERHEAD ), +HPP__COLOR_PRINT_FNS ("Latency" , latency,LATENCY ),HPP__COLOR_PRINT_FNS ("sys" , overhead_sys,OVERHEAD_SYS ),HPP__COLOR_PRINT_FNS ("usr" , overhead_us,OVERHEAD_US ),HPP__COLOR_PRINT_FNS ("guest sys" , overhead_guest_sys,OVERHEAD_GUEST_SYS ),HPP__COLOR_PRINT_FNS ("guest usr" , overhead_guest_us,OVERHEAD_GUEST_US ),HPP__COLOR_ACC_PRINT_FNS ("Children" , overhead_acc,OVERHEAD_ACC ), +HPP__COLOR_ACC_PRINT_FNS ("Latency" , latency_acc,LATENCY_ACC ),HPP__PRINT_FNS ("Samples" , samples,SAMPLES ),HPP__PRINT_FNS ("Period" , period,PERIOD ),HPP__PRINT_FNS ("Weight1" , weight1,WEIGHT1 ),
@@ -601,6 +606,11 @@ static void fmt_free(struct perf_hpp_fmt *fmt) fmt->free (fmt); } +static bool fmt_equal (struct perf_hpp_fmt *a,struct perf_hpp_fmt *b) +{ +return a->equal && a->equal (a, b); +} +void perf_hpp__init (void ) {int i;
@@ -671,30 +681,26 @@ static void perf_hpp__column_unregister(struct perf_hpp_fmt *format)void perf_hpp__cancel_cumulate (void ) { -struct perf_hpp_fmt *fmt, *acc, *ovh, *tmp; +struct perf_hpp_fmt *fmt, *acc, *ovh, *acc_lat, *tmp;if (is_strict_order (field_order))return ; ovh = &perf_hpp__format[PERF_HPP__OVERHEAD ]; acc = &perf_hpp__format[PERF_HPP__OVERHEAD_ACC ]; + acc_lat = &perf_hpp__format[PERF_HPP__LATENCY_ACC ];perf_hpp_list__for_each_format_safe (&perf_hpp_list, fmt, tmp) { -if (acc->equal (acc, fmt)) { +if (fmt_equal (acc, fmt) ||fmt_equal (acc_lat, fmt)) {perf_hpp__column_unregister (fmt);continue ; } -if (ovh->equal (ovh, fmt)) +if (fmt_equal (ovh, fmt)) fmt->name ="Overhead" ; } } -static bool fmt_equal (struct perf_hpp_fmt *a,struct perf_hpp_fmt *b) -{ -return a->equal && a->equal (a, b); -} -void perf_hpp__setup_output_field (struct perf_hpp_list *list) {struct perf_hpp_fmt *fmt;
@@ -819,6 +825,7 @@ void perf_hpp__reset_width(struct perf_hpp_fmt *fmt, struct hists *hists)switch (fmt->idx) {case PERF_HPP__OVERHEAD : +case PERF_HPP__LATENCY :case PERF_HPP__OVERHEAD_SYS :case PERF_HPP__OVERHEAD_US :case PERF_HPP__OVERHEAD_ACC :
tools/perf/util/addr_location.h +2 -0
@@ -24,6 +24,8 @@ struct addr_location {s32 socket;/* Same as machine.parallelism but within [1, nr_cpus]. */ int parallelism; +/* See he_stat.latency. */ +u64 latency; };void addr_location__init (struct addr_location *al);
tools/perf/util/event.c +6 -0
@@ -771,6 +771,12 @@ int machine__resolve(struct machine *machine, struct addr_location *al, al->parallelism =min (max (machine->parallelism,1 ),machine__nr_cpus_avail (machine));if (test_bit (al->parallelism, symbol_conf.parallelism_filter)) al->filtered |= (1 <<HIST_FILTER__PARALLELISM ); +/* +* Multiply it by some const to avoid precision loss or dealing +* with floats. The multiplier does not matter otherwise since +* we only print it as percents. +*/ + al->latency = sample->period *1000 / al->parallelism;if (al->map) { if (symbol_conf.dso_list &&
tools/perf/util/events_stats.h +2 -0
@@ -57,6 +57,8 @@ struct events_stats {struct hists_stats {u64 total_period;u64 total_non_filtered_period; +u64 total_latency; +u64 total_non_filtered_latency;u32 nr_samples;u32 nr_non_filtered_samples;u32 nr_lost_samples;
tools/perf/util/hist.c +41 -14
175 lines not on this page · raw message
tools/perf/util/hist.h +12 -0
@@ -109,6 +109,8 @@ struct hists {u64 nr_non_filtered_entries;u64 callchain_period;u64 callchain_non_filtered_period; +u64 callchain_latency; +u64 callchain_non_filtered_latency;struct thread *thread_filter;const struct dso *dso_filter;const char *uid_filter_str;
@@ -170,6 +172,12 @@ struct res_sample {struct he_stat {u64 period; +/* +* Period re-scaled from CPU time to wall-clock time (divided by the +* parallelism at the time of the sample). This represents effect of +* the event on latency rather than CPU consumption. +*/ +u64 latency;u64 period_sys;u64 period_us;u64 period_guest_sys;
@@ -374,6 +382,7 @@ void hists__output_recalc_col_len(struct hists *hists, int max_rows);struct hist_entry *hists__get_entry (struct hists *hists,int idx);u64 hists__total_period (struct hists *hists); +u64 hists__total_latency (struct hists *hists);void hists__reset_stats (struct hists *hists);void hists__inc_stats (struct hists *hists,struct hist_entry *h);void hists__inc_nr_events (struct hists *hists);
@@ -555,11 +564,13 @@ extern struct perf_hpp_fmt perf_hpp__format[];enum {/* Matches perf_hpp__format array. */ PERF_HPP__OVERHEAD , +PERF_HPP__LATENCY ,PERF_HPP__OVERHEAD_SYS ,PERF_HPP__OVERHEAD_US ,PERF_HPP__OVERHEAD_GUEST_SYS ,PERF_HPP__OVERHEAD_GUEST_US ,PERF_HPP__OVERHEAD_ACC , +PERF_HPP__LATENCY_ACC ,PERF_HPP__SAMPLES ,PERF_HPP__PERIOD ,PERF_HPP__WEIGHT1 ,
@@ -615,6 +626,7 @@ void hists__reset_column_width(struct hists *hists);enum perf_hpp_fmt_type {PERF_HPP_FMT_TYPE__RAW ,PERF_HPP_FMT_TYPE__PERCENT , +PERF_HPP_FMT_TYPE__LATENCY ,PERF_HPP_FMT_TYPE__AVERAGE , };
tools/perf/util/sort.c +2 -0
@@ -2628,11 +2628,13 @@ struct hpp_dimension {static struct hpp_dimension hpp_sort_dimensions[] = {DIM (PERF_HPP__OVERHEAD ,"overhead" ), +DIM (PERF_HPP__LATENCY ,"latency" ),DIM (PERF_HPP__OVERHEAD_SYS ,"overhead_sys" ),DIM (PERF_HPP__OVERHEAD_US ,"overhead_us" ),DIM (PERF_HPP__OVERHEAD_GUEST_SYS ,"overhead_guest_sys" ),DIM (PERF_HPP__OVERHEAD_GUEST_US ,"overhead_guest_us" ),DIM (PERF_HPP__OVERHEAD_ACC ,"overhead_children" ), +DIM (PERF_HPP__LATENCY_ACC ,"latency_children" ),DIM (PERF_HPP__SAMPLES ,"sample" ),DIM (PERF_HPP__PERIOD ,"period" ),DIM (PERF_HPP__WEIGHT1 ,"weight1" ),
Signature · 1 line
2.48.1.362.g079036d154-goog
PATCH v3 6/8 perf report: Add --latency flag
In reply to Dmitry VyukovAdd record/report --latency flag that allows to capture and show latency-centric profiles rather than the default CPU-consumption-centric profiles. For latency profiles record captures context switch events, and report shows Latency as the first column.
--- tools/perf/builtin-record.c | 20 +++++++++++++++++ tools/perf/builtin-report.c | 32 +++++++++++++++++++++++---- tools/perf/ui/hist.c | 41 ++++++++++++++++++++++++++++------- tools/perf/util/hist.h | 1 + tools/perf/util/sort.c | 33 +++++++++++++++++++++++----- tools/perf/util/sort.h | 2 +- tools/perf/util/symbol_conf.h | 4 +++- 7 files changed, 113 insertions(+), 20 deletions(-)
tools/perf/builtin-record.c +20 -0
@@ -161,6 +161,7 @@ struct record {struct evlist *sb_evlist;pthread_t thread_id;int realtime_prio; +bool latency;bool switch_output_event_set;bool no_buildid;bool no_buildid_set;
@@ -3371,6 +3372,9 @@ static struct option __record_options[] = { parse_events_option),OPT_CALLBACK (0 ,"filter" , &record.evlist,"filter" ,"event filter" , parse_filter), +OPT_BOOLEAN (0 ,"latency" , &record.latency, +"Enable data collection for latency profiling.\n" +"\t\t\t Use perf report --latency for latency-centric profile." ),OPT_CALLBACK_NOOPT (0 ,"exclude-perf" , &record.evlist,NULL ,"don't record events from perf itself" , exclude_perf),
@@ -4017,6 +4021,22 @@ int cmd_record(int argc, const char **argv) } +if (record.latency) { +/* +* There is no fundamental reason why latency profiling +* can't work for system-wide mode, but exact semantics +* and details are to be defined. +* See the following thread for details: +* https://lore.kernel.org/all/[email protected]/ +*/ +if (record.opts.target.system_wide) { +pr_err ("Failed: latency profiling is not supported with system-wide collection.\n" ); + err = -EINVAL ; + goto out_opts; + } + record.opts.record_switch_events = true; + } +if (rec->buildid_mmap) {if (!perf_can_record_build_id ()) {pr_err ("Failed: no support to record build id in mmap events, update your kernel.\n" );
tools/perf/builtin-report.c +28 -4
@@ -112,6 +112,8 @@ struct report {u64 nr_entries;u64 queue_size;u64 total_cycles; +u64 total_samples; +u64 singlethreaded_samples;int socket_filter;DECLARE_BITMAP (cpu_bitmap,MAX_NR_CPUS );struct branch_type_stat brtype_stat;
@@ -331,6 +333,10 @@ static int process_sample_event(const struct perf_tool *tool, &rep->total_cycles, evsel); } + rep->total_samples++; +if (al.parallelism ==1 ) + rep->singlethreaded_samples++; + ret =hist_entry_iter__add (&iter, &al, rep->max_stack, rep);if (ret <0 )pr_debug ("problem adding hist entry, skipping event\n" );
@@ -1079,6 +1085,11 @@ static int __cmd_report(struct report *rep)return ret; } +/* Don't show Latency column for non-parallel profiles by default. */ +if (rep->singlethreaded_samples *100 / rep->total_samples >=99 && + !symbol_conf.prefer_latency) +perf_hpp__cancel_latency (); +evlist__check_mem_load_aux (session->evlist);if (rep->stats_mode)
@@ -1468,6 +1479,10 @@ int cmd_report(int argc, const char **argv) "Disable raw trace ordering"),OPT_BOOLEAN (0 ,"skip-empty" , &report.skip_empty,"Do not display empty (or dummy) events in the output" ), +OPT_BOOLEAN (0 ,"latency" , &symbol_conf.prefer_latency, +"Show latency-centric profile rather than the default\n" +"\t\t\t CPU-consumption-centric profile\n" +"\t\t\t (requires perf record --latency flag)." ),OPT_END () }; struct perf_data data = {
@@ -1722,16 +1737,25 @@ int cmd_report(int argc, const char **argv) symbol_conf.annotate_data_sample = true; } + symbol_conf.enable_latency = true;if (report.disable_order || !perf_session__has_switch_events (session)) {if (symbol_conf.parallelism_list_str || - (sort_order &&strstr (sort_order,"parallelism" )) || - (field_order &&strstr (field_order,"parallelism" ))) { + symbol_conf.prefer_latency || + (sort_order && (strstr (sort_order,"latency" ) || +strstr (sort_order,"parallelism" ))) || + (field_order && (strstr (field_order,"latency" ) || +strstr (field_order,"parallelism" )))) {if (report.disable_order) -ui__error ("Use of parallelism is incompatible with --disable-order.\n" ); +ui__error ("Use of latency profile or parallelism is incompatible with --disable-order.\n" );else -ui__error ("Use of parallelism requires --switch-events during record.\n" ); +ui__error ("Use of latency profile or parallelism requires --latency flag during record.\n" );return -1 ; } +/* +* If user did not ask for anything related to +* latency/parallelism explicitly, just don't show it. +*/ + symbol_conf.enable_latency = false; }if (sort_order &&strstr (sort_order,"ipc" )) {
tools/perf/ui/hist.c +33 -8
66 lines not on this page · raw message
tools/perf/util/hist.h +1 -0
@@ -582,6 +582,7 @@ enum {void perf_hpp__init (void );void perf_hpp__cancel_cumulate (void ); +void perf_hpp__cancel_latency (void );void perf_hpp__setup_output_field (struct perf_hpp_list *list);void perf_hpp__reset_output_field (struct perf_hpp_list *list);void perf_hpp__append_sort_keys (struct perf_hpp_list *list);
tools/perf/util/sort.c +27 -6
@@ -2622,6 +2622,7 @@ struct hpp_dimension {const char *name;struct perf_hpp_fmt *fmt;int taken; +int was_taken; };#define DIM (d, n) { .name = n, .fmt = &perf_hpp__format[d], }
@@ -3513,6 +3514,7 @@ static int __hpp_dimension__add(struct hpp_dimension *hd,return -1 ; hd->taken =1 ; + hd->was_taken =1 ;perf_hpp_list__register_sort_field (list, fmt);return 0 ; }
@@ -3547,10 +3549,15 @@ static int __hpp_dimension__add_output(struct perf_hpp_list *list,return 0 ; } -int hpp_dimension__add_output (unsigned col) +int hpp_dimension__add_output (unsigned col,bool implicit) { +struct hpp_dimension *hd; +BUG_ON (col >=PERF_HPP__MAX_INDEX ); -return __hpp_dimension__add_output (&perf_hpp_list, &hpp_sort_dimensions[col]); + hd = &hpp_sort_dimensions[col]; +if (implicit && !hd->was_taken) +return 0 ; +return __hpp_dimension__add_output (&perf_hpp_list, hd); } int sort_dimension__add(struct perf_hpp_list *list, const char *tok,
@@ -3809,10 +3816,24 @@ static char *setup_overhead(char *keys)if (sort__mode ==SORT_MODE__DIFF )return keys; - keys =prefix_if_not_in ("overhead" , keys); - -if (symbol_conf.cumulate_callchain) - keys =prefix_if_not_in ("overhead_children" , keys); +if (symbol_conf.prefer_latency) { + keys =prefix_if_not_in ("overhead" , keys); + keys =prefix_if_not_in ("latency" , keys); +if (symbol_conf.cumulate_callchain) { + keys =prefix_if_not_in ("overhead_children" , keys); + keys =prefix_if_not_in ("latency_children" , keys); + } + }else if (!keys || (!strstr (keys,"overhead" ) && + !strstr (keys,"latency" ))) { +if (symbol_conf.enable_latency) + keys =prefix_if_not_in ("latency" , keys); + keys =prefix_if_not_in ("overhead" , keys); +if (symbol_conf.cumulate_callchain) { +if (symbol_conf.enable_latency) + keys =prefix_if_not_in ("latency_children" , keys); + keys =prefix_if_not_in ("overhead_children" , keys); + } + }return keys; }
tools/perf/util/sort.h +1 -1
@@ -141,7 +141,7 @@ int report_parse_ignore_callees_opt(const struct option *opt, const char *arg, ibool is_strict_order (const char *order); -int hpp_dimension__add_output (unsigned col); +int hpp_dimension__add_output (unsigned col,bool implicit);void reset_dimensions (void ); int sort_dimension__add(struct perf_hpp_list *list, const char *tok, struct evlist *evlist,
tools/perf/util/symbol_conf.h +3 -1
@@ -49,7 +49,9 @@ struct symbol_conf { keep_exited_threads, annotate_data_member, annotate_data_sample, - skip_empty; + skip_empty, + enable_latency, + prefer_latency;const char *vmlinux_name, *kallsyms_name, *source_prefix,
Signature · 1 line
2.48.1.362.g079036d154-goog
PATCH v3 7/8 perf report: Add latency and parallelism profiling documentation
In reply to Dmitry VyukovDescribe latency and parallelism profiling, related flags, and differences with the currently only supported CPU-consumption-centric profiling.
--- .../callchain-overhead-calculation.txt | 5 +- .../cpu-and-latency-overheads.txt | 85 +++++++++++++++++++ tools/perf/Documentation/perf-report.txt | 49 +++++++---- tools/perf/Documentation/tips.txt | 3 + 4 files changed, 123 insertions(+), 19 deletions(-)
tools/perf/Documentation/callchain-overhead-calculation.txt +3 -2
@@ -1,7 +1,8 @@ Overhead calculation -------------------- -The overhead can be shown in two columns as 'Children' and 'Self' when -perf collects callchains. The 'self' overhead is simply calculated by +The CPU overhead can be shown in two columns as 'Children' and 'Self' +when perf collects callchains (and corresponding 'Wall' columns for +wall-clock overhead). The 'self' overhead is simply calculated by adding all period values of the entry - usually a function (symbol). This is the value that perf shows traditionally and sum of all the 'self' overhead values should be 100%.
tools/perf/Documentation/cpu-and-latency-overheads.txt new file +85 -0
85 lines not on this page · raw message
tools/perf/Documentation/perf-report.txt +32 -17
114 lines not on this page · raw message
tools/perf/Documentation/tips.txt +3 -0
@@ -62,3 +62,6 @@ To show context switches in perf report sample context add --switch-events to pe To show time in nanoseconds in record/report add --ns To compare hot regions in two workloads use perf record -b -o file ... ; perf diff --stream file1 file2 To compare scalability of two workload samples use perf diff -c ratio file1 file2 +For latency profiling, try: perf record/report --latency +For parallelism histogram, try: perf report --hierarchy --sort latency,parallelism,comm,symbol +To analyze particular parallelism levels, try: perf report --latency --parallelism=32-64
Signature · 1 line
2.48.1.362.g079036d154-goog
PATCH v3 8/8 perf hist: Shrink struct hist_entry size
In reply to Dmitry VyukovReorder the struct fields by size to reduce paddings and reduce struct simd_flags size from 8 to 1 byte.
This reduces struct hist_entry size by 8 bytes (592->584), and leaves a single more usable 6 byte padding hole.
---
Pahole output before:
struct hist_entry {
struct rb_node rb_node_in __attribute__((__aligned__(8))); /* 0 24 */
struct rb_node rb_node __attribute__((__aligned__(8))); /* 24 24 */
union {
struct list_head node; /* 48 16 */
struct list_head head; /* 48 16 */
} pairs; /* 48 16 */
/* --- cacheline 1 boundary (64 bytes) --- */
struct he_stat stat; /* 64 80 */
/* XXX last struct has 4 bytes of padding */
/* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
struct he_stat * stat_acc; /* 144 8 */
struct map_symbol ms; /* 152 24 */
struct thread * thread; /* 176 8 */
struct comm * comm; /* 184 8 */
/* --- cacheline 3 boundary (192 bytes) --- */
struct namespace_id cgroup_id; /* 192 16 */
u64 cgroup; /* 208 8 */
u64 ip; /* 216 8 */
u64 transaction; /* 224 8 */
s32 socket; /* 232 4 */
s32 cpu; /* 236 4 */
int parallelism; /* 240 4 */
/* XXX 4 bytes hole, try to pack */
u64 code_page_size; /* 248 8 */
/* --- cacheline 4 boundary (256 bytes) --- */
u64 weight; /* 256 8 */
u64 ins_lat; /* 264 8 */
u64 p_stage_cyc; /* 272 8 */
u8 cpumode; /* 280 1 */
u8 depth; /* 281 1 */
/* XXX 2 bytes hole, try to pack */
int mem_type_off; /* 284 4 */
struct simd_flags simd_flags; /* 288 8 */
_Bool dummy; /* 296 1 */
_Bool leaf; /* 297 1 */
char level; /* 298 1 */
/* XXX 1 byte hole, try to pack */
filter_mask_t filtered; /* 300 2 */
u16 callchain_size; /* 302 2 */
union {
struct hist_entry_diff diff; /* 304 120 */
struct {
u16 row_offset; /* 304 2 */
u16 nr_rows; /* 306 2 */
_Bool init_have_children; /* 308 1 */
_Bool unfolded; /* 309 1 */
_Bool has_children; /* 310 1 */
_Bool has_no_entry; /* 311 1 */
}; /* 304 8 */
}; /* 304 120 */
/* --- cacheline 6 boundary (384 bytes) was 40 bytes ago --- */
char * srcline; /* 424 8 */
char * srcfile; /* 432 8 */
struct symbol * parent; /* 440 8 */
/* --- cacheline 7 boundary (448 bytes) --- */
struct branch_info * branch_info; /* 448 8 */
long int time; /* 456 8 */
struct hists * hists; /* 464 8 */
struct mem_info * mem_info; /* 472 8 */
struct block_info * block_info; /* 480 8 */
struct kvm_info * kvm_info; /* 488 8 */
void * raw_data; /* 496 8 */
u32 raw_size; /* 504 4 */
int num_res; /* 508 4 */
/* --- cacheline 8 boundary (512 bytes) --- */
struct res_sample * res_samples; /* 512 8 */
void * trace_output; /* 520 8 */
struct perf_hpp_list * hpp_list; /* 528 8 */
struct hist_entry * parent_he; /* 536 8 */
struct hist_entry_ops * ops; /* 544 8 */
struct annotated_data_type * mem_type; /* 552 8 */
union {
struct {
struct rb_root_cached hroot_in; /* 560 16 */
/* --- cacheline 9 boundary (576 bytes) --- */
struct rb_root_cached hroot_out; /* 576 16 */
}; /* 560 32 */
struct rb_root sorted_chain; /* 560 8 */
}; /* 560 32 */
/* --- cacheline 9 boundary (576 bytes) was 16 bytes ago --- */
struct callchain_root callchain[] __attribute__((__aligned__(8))); /* 592 0 */
/* size: 592, cachelines: 10, members: 49 */ /* sum members: 585, holes: 3, sum holes: 7 */ /* paddings: 1, sum paddings: 4 */
/* forced alignments: 3 */
/* last cacheline: 16 bytes */
} __attribute__((__aligned__(8)));
After:
struct hist_entry {
struct rb_node rb_node_in __attribute__((__aligned__(8))); /* 0 24 */
struct rb_node rb_node __attribute__((__aligned__(8))); /* 24 24 */
union {
struct list_head node; /* 48 16 */
struct list_head head; /* 48 16 */
} pairs; /* 48 16 */
/* --- cacheline 1 boundary (64 bytes) --- */
struct he_stat stat; /* 64 80 */
/* XXX last struct has 4 bytes of padding */
/* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
struct he_stat * stat_acc; /* 144 8 */
struct map_symbol ms; /* 152 24 */
struct thread * thread; /* 176 8 */
struct comm * comm; /* 184 8 */
/* --- cacheline 3 boundary (192 bytes) --- */
struct namespace_id cgroup_id; /* 192 16 */
u64 cgroup; /* 208 8 */
u64 ip; /* 216 8 */
u64 transaction; /* 224 8 */
u64 code_page_size; /* 232 8 */
u64 weight; /* 240 8 */
u64 ins_lat; /* 248 8 */
/* --- cacheline 4 boundary (256 bytes) --- */
u64 p_stage_cyc; /* 256 8 */
s32 socket; /* 264 4 */
s32 cpu; /* 268 4 */
int parallelism; /* 272 4 */
int mem_type_off; /* 276 4 */
u8 cpumode; /* 280 1 */
u8 depth; /* 281 1 */
struct simd_flags simd_flags; /* 282 1 */
_Bool dummy; /* 283 1 */
_Bool leaf; /* 284 1 */
char level; /* 285 1 */
filter_mask_t filtered; /* 286 2 */
u16 callchain_size; /* 288 2 */
/* XXX 6 bytes hole, try to pack */
union {
struct hist_entry_diff diff; /* 296 120 */
struct {
u16 row_offset; /* 296 2 */
u16 nr_rows; /* 298 2 */
_Bool init_have_children; /* 300 1 */
_Bool unfolded; /* 301 1 */
_Bool has_children; /* 302 1 */
_Bool has_no_entry; /* 303 1 */
}; /* 296 8 */
}; /* 296 120 */
/* --- cacheline 6 boundary (384 bytes) was 32 bytes ago --- */
char * srcline; /* 416 8 */
char * srcfile; /* 424 8 */
struct symbol * parent; /* 432 8 */
struct branch_info * branch_info; /* 440 8 */
/* --- cacheline 7 boundary (448 bytes) --- */
long int time; /* 448 8 */
struct hists * hists; /* 456 8 */
struct mem_info * mem_info; /* 464 8 */
struct block_info * block_info; /* 472 8 */
struct kvm_info * kvm_info; /* 480 8 */
void * raw_data; /* 488 8 */
u32 raw_size; /* 496 4 */
int num_res; /* 500 4 */
struct res_sample * res_samples; /* 504 8 */
/* --- cacheline 8 boundary (512 bytes) --- */
void * trace_output; /* 512 8 */
struct perf_hpp_list * hpp_list; /* 520 8 */
struct hist_entry * parent_he; /* 528 8 */
struct hist_entry_ops * ops; /* 536 8 */
struct annotated_data_type * mem_type; /* 544 8 */
union {
struct {
struct rb_root_cached hroot_in; /* 552 16 */
struct rb_root_cached hroot_out; /* 568 16 */
}; /* 552 32 */
struct rb_root sorted_chain; /* 552 8 */
}; /* 552 32 */
/* --- cacheline 9 boundary (576 bytes) was 8 bytes ago --- */
struct callchain_root callchain[] __attribute__((__aligned__(8))); /* 584 0 */
/* size: 584, cachelines: 10, members: 49 */ /* sum members: 578, holes: 1, sum holes: 6 */ /* paddings: 1, sum paddings: 4 */
/* forced alignments: 3 */
/* last cacheline: 8 bytes */
} __attribute__((__aligned__(8)));
--- tools/perf/util/hist.h | 8 ++++---- tools/perf/util/sample.h | 2 +- 2 files changed, 5 insertions(+), 5 deletions(-)
tools/perf/util/hist.h +4 -4
@@ -239,16 +239,16 @@ struct hist_entry {u64 cgroup;u64 ip;u64 transaction; -s32 socket; -s32 cpu; -int parallelism;u64 code_page_size;u64 weight;u64 ins_lat;u64 p_stage_cyc; +s32 socket; +s32 cpu; +int parallelism; +int mem_type_off;u8 cpumode;u8 depth; -int mem_type_off;struct simd_flags simd_flags;/* We are added by hists__add_dummy_entry. */
tools/perf/util/sample.h +1 -1
@@ -67,7 +67,7 @@ struct aux_sample { };struct simd_flags { -u64 arch:1 ,/* architecture (isa) */ +u8 arch:1 ,/* architecture (isa) */ pred:2 ;/* predication */ };
Signature · 1 line
2.48.1.362.g079036d154-goog
Namhyung Kim
maintainerHello,
Quoted 48 lines · Dmitry Vyukov
There are two notions of time: wall-clock time and CPU time. For a single-threaded program, or a program running on a single-core machine, these notions are the same. However, for a multi-threaded/ multi-process program running on a multi-core machine, these notions are significantly different. Each second of wall-clock time we have number-of-cores seconds of CPU time.
Currently perf only allows to profile CPU time. Perf (and all other existing profilers to the be best of my knowledge) does not allow to profile wall-clock time.
Optimizing CPU overhead is useful to improve 'throughput', while optimizing wall-clock overhead is useful to improve 'latency'. These profiles are complementary and are not interchangeable. Examples of where latency profile is needed:
- optimzing build latency
- optimizing server request latency
- optimizing ML training/inference latency
- optimizing running time of any command line programCPU profile is useless for these use cases at best (if a user understands the difference), or misleading at worst (if a user tries to use a wrong profile for a job).
This series add latency and parallelization profiling. See the added documentation and flags descriptions for details.
Brief outline of the implementation:
- add context switch collection during record
- calculate number of threads running on CPUs (parallelism level) during report
- divide each sample weight by the parallelism level
This effectively models that we were taking 1 sample per unit of wall-clock time.We still default to the CPU profile, so it's up to users to learn about the second profiling mode and use it when appropriate.
Changes in v4:
- added "Shrink structhist_entrysize" commit
- rebased to perf-tools-next HEADChanges in v3:
- rebase and split into patches
- rename 'wallclock' to 'latency' everywhere
- don't enable latency profiling by default,
instead add record/report --latency flag
Quoted 9 lines
Dmitry Vyukov (8):
perf report: Add machine parallelism
perf report: Add parallelism sort key
perf report: Switch filtered from u8 to u16
perf report: Add parallelism filter
perf report: Add latency output field
perf report: Add --latency flag
perf report: Add latency and parallelism profiling documentation perf hist: Shrink structhist_entrysize
I found that stdio output is broken that latency is too high. TUI is fine.
$ ./perf report --latency -s dso --stdio
...
#
# Latency Overhead Shared Object
# ........ ........ ....................
#
57669.15% 57.67% [kernel.kallsyms]
42330.85% 42.33% ld-linux-x86-64.so.2Thanks,
Namhyung
Quoted 32 lines
.../callchain-overhead-calculation.txt | 5 +- .../cpu-and-latency-overheads.txt | 85 ++++++++++++++++++ tools/perf/Documentation/perf-report.txt | 49 ++++++---- tools/perf/Documentation/tips.txt | 3 + tools/perf/builtin-record.c | 20 +++++ tools/perf/builtin-report.c | 39 ++++++++ tools/perf/ui/browsers/hists.c | 27 +++--- tools/perf/ui/hist.c | 64 +++++++++---- tools/perf/util/addr_location.c | 1 + tools/perf/util/addr_location.h | 7 +- tools/perf/util/event.c | 11 +++ tools/perf/util/events_stats.h | 2 + tools/perf/util/hist.c | 90 +++++++++++++++---- tools/perf/util/hist.h | 32 +++++-- tools/perf/util/machine.c | 7 ++ tools/perf/util/machine.h | 6 ++ tools/perf/util/sample.h | 2 +- tools/perf/util/session.c | 12 +++ tools/perf/util/session.h | 1 + tools/perf/util/sort.c | 69 ++++++++++++-- tools/perf/util/sort.h | 3 +- tools/perf/util/symbol.c | 34 +++++++ tools/perf/util/symbol_conf.h | 8 +- 23 files changed, 502 insertions(+), 75 deletions(-) create mode 100644 tools/perf/Documentation/cpu-and-latency-overheads.txtbase-commit:
8ce0d2da14d3fb62844dd0e95982c194326b1a5fSignature · 1 line
2.48.1.362.g079036d154-goog
Namhyung Kim
maintainerRe: [PATCH v3 7/8] perf report: Add latency and parallelism profiling documentation
In reply to Dmitry VyukovQuoted 2 lines · Dmitry Vyukov
Describe latency and parallelism profiling, related flags, and differences with the currently only supported CPU-consumption-centric profiling.
It doesn't seem to have descriptions for the --latency option (for perf record and report). Probably better to put them in the previous patch.
Quoted 75 lines
Signed-off-by Dmitry Vyukov <[email protected]>Cc Namhyung Kim <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Ian Rogers <[email protected]>--- .../callchain-overhead-calculation.txt | 5 +- .../cpu-and-latency-overheads.txt | 85 +++++++++++++++++++ tools/perf/Documentation/perf-report.txt | 49 +++++++---- tools/perf/Documentation/tips.txt | 3 + 4 files changed, 123 insertions(+), 19 deletions(-)tools/perf/Documentation/callchain-overhead-calculation.txt +3
-2@@ -1,7 +1,8 @@ Overhead calculation -------------------- -The overhead can be shown in two columns as 'Children' and 'Self' when -perf collects callchains. The 'self' overhead is simply calculated by +The CPU overhead can be shown in two columns as 'Children' and 'Self' +when perf collects callchains (and corresponding 'Wall' columns for +wall-clock overhead). The 'self' overhead is simply calculated by adding all period values of the entry - usually a function (symbol). This is the value that perf shows traditionally and sum of all the 'self' overhead values should be 100%.tools/perf/Documentation/cpu-and-latency-overheads.txt new file +40
-0@@ -0,0 +1,85 @@ +CPU and latency overheads +------------------------- +There are two notions of time: wall-clock time and CPU time. +For a single-threaded program, or a program running on a single-core machine, +these notions are the same. However, for a multi-threaded/multi-process program +running on a multi-core machine, these notions are significantly different. +Each second of wall-clock time we have number-of-cores seconds of CPU time. +Perf can measure overhead for both of these times (shown in 'overhead' and +'latency' columns for CPU and wall-clock time correspondingly). + +Optimizing CPU overhead is useful to improve 'throughput', while optimizing +latency overhead is useful to improve 'latency'. It's important to understand +which one is useful in a concrete situation at hand. For example, the former +may be useful to improve max throughput of a CI build server that runs on 100% +CPU utilization, while the latter may be useful to improve user-perceived +latency of a single interactive program build. +These overheads may be significantly different in some cases. For example, +consider a program that executes function 'foo' for 9 seconds with 1 thread, +and then executes function 'bar' for 1 second with 128 threads (consumes +128 seconds of CPU time). The CPU overhead is: 'foo' - 6.6%, 'bar' - 93.4%. +While the latency overhead is: 'foo' - 90%, 'bar' - 10%. If we try to optimize +running time of the program looking at the (wrong in this case) CPU overhead, +we would concentrate on the function 'bar', but it can yield only 10% running +time improvement at best. + +By default, perf shows only CPU overhead. To show latency overhead, use +'perf record --latency' and 'perf report': + +----------------------------------- +Overhead Latency Command + 93.88% 25.79% cc1 + 1.90% 39.87% gzip + 0.99% 10.16% dpkg-deb + 0.57% 1.00% as + 0.40% 0.46% sh +----------------------------------- + +To sort by latency overhead, use 'perf report --latency': + +-----------------------------------
tools/perf/Documentation/cpu-and-latency-overheads.txt line 50
Quoted 10 lines
tools/perf/Documentation/cpu-and-latency-overheads.txt new file +10
-0+Latency Overhead Command + 39.87% 1.90% gzip + 25.79% 93.88% cc1 + 10.16% 0.99% dpkg-deb + 4.17% 0.29% git + 2.81% 0.11% objtool +----------------------------------- + +To get insight into the difference between the overheads, you may check +parallelization histogram with '--sort=latency,parallelism,comm,symbol --hierarchy'
I think you need to omit latency in the sort key and recommend users to use --latency option instead.
'perf report --hierarchy --latency --sort=parallelism,comm,symbol'
Quoted 168 lines
tools/perf/Documentation/cpu-and-latency-overheads.txt new file +35
-0+flags. It shows fraction of (wall-clock) time the workload utilizes different +numbers of cores ('Parallelism' column). For example, in the following case +the workload utilizes only 1 core most of the time, but also has some +highly-parallel phases, which explains significant difference between +CPU and wall-clock overheads: + +----------------------------------- + Latency Overhead Parallelism / Command / Symbol ++ 56.98% 2.29% 1 ++ 16.94% 1.36% 2 ++ 4.00% 20.13% 125 ++ 3.66% 18.25% 124 ++ 3.48% 17.66% 126 ++ 3.26% 0.39% 3 ++ 2.61% 12.93% 123 +----------------------------------- + +By expanding corresponding lines, you may see what commands/functions run +at the given parallelism level: + +----------------------------------- + Latency Overhead Parallelism / Command / Symbol +- 56.98% 2.29% 1 + 32.80% 1.32% gzip + 4.46% 0.18% cc1 + 2.81% 0.11% objtool + 2.43% 0.10% dpkg-source + 2.22% 0.09% ld + 2.10% 0.08% dpkg-genchanges +----------------------------------- + +To see the normal function-level profile for particular parallelism levels +(number of threads actively running on CPUs), you may use '--parallelism' +filter. For example, to see the profile only for low parallelism phases +of a workload use '--latency --parallelism=1-2' flags.tools/perf/Documentation/perf-report.txt +32
-17@@ -44,7 +44,7 @@ OPTIONS --comms=:: Only consider symbols in these comms. CSV that understands file://filename entries. This option will affect the percentage of - the overhead column. See --percentage for more info. + the overhead and latency columns. See --percentage for more info. --pid=:: Only show events for given process ID (comma separated list).@@ -54,12 +54,12 @@ OPTIONS --dsos=:: Only consider symbols in these dsos. CSV that understands file://filename entries. This option will affect the percentage of - the overhead column. See --percentage for more info. + the overhead and latency columns. See --percentage for more info. -S:: --symbols=:: Only consider these symbols. CSV that understands file://filename entries. This option will affect the percentage of - the overhead column. See --percentage for more info. + the overhead and latency columns. See --percentage for more info. --symbol-filter=:: Only show symbols that match (partially) with this filter.@@ -68,6 +68,16 @@ OPTIONS --hide-unresolved:: Only display entries resolved to a symbol. +--parallelism:: + Only consider these parallelism levels. Parallelism level is the number + of threads that actively run on CPUs at the time of sample. The flag + accepts single number, comma-separated list, and ranges (for example: + "1", "7,8", "1,64-128"). This is useful in understanding what a program + is doing during sequential/low-parallelism phases as compared to + high-parallelism phases. This option will affect the percentage of + the overhead and latency columns. See --percentage for more info. + Also see the `CPU and latency overheads' section for more details. + -s:: --sort=:: Sort histogram entries by given key(s) - multiple keys can be specified@@ -87,6 +97,7 @@ OPTIONS entries are displayed as "[other]". - cpu: cpu number the task ran at the time of sample - socket: processor socket number the task ran at the time of sample + - parallelism: number of running threads at the time of sample - srcline: filename and line number executed at the time of sample. The DWARF debugging info must be provided. - srcfile: file name of the source file of the samples. Requires dwarf@@ -97,12 +108,14 @@ OPTIONS - cgroup_id: ID derived from cgroup namespace device and inode numbers. - cgroup: cgroup pathname in the cgroupfs. - transaction: Transaction abort flags. - - overhead: Overhead percentage of sample - - overhead_sys: Overhead percentage of sample running in system mode - - overhead_us: Overhead percentage of sample running in user mode - - overhead_guest_sys: Overhead percentage of sample running in system mode + - overhead: CPU overhead percentage of sample. + - latency: latency (wall-clock) overhead percentage of sample. + See the `CPU and latency overheads' section for more details. + - overhead_sys: CPU overhead percentage of sample running in system mode + - overhead_us: CPU overhead percentage of sample running in user mode + - overhead_guest_sys: CPU overhead percentage of sample running in system mode on guest machine - - overhead_guest_us: Overhead percentage of sample running in user mode on + - overhead_guest_us: CPU overhead percentage of sample running in user mode on guest machine - sample: Number of sample - period: Raw number of event count of sample@@ -125,8 +138,8 @@ OPTIONS - weight2: Average value of event specific weight (2nd field of weight_struct). - weight3: Average value of event specific weight (3rd field of weight_struct). - By default, comm, dso and symbol keys are used. - (i.e. --sort comm,dso,symbol) + By default, overhead, comm, dso and symbol keys are used. + (i.e. --sort overhead,comm,dso,symbol). If --branch-stack option is used, following sort keys are also available:@@ -201,9 +214,9 @@ OPTIONS --fields=:: Specify output field - multiple keys can be specified in CSV format. Following fields are available: - overhead, overhead_sys, overhead_us, overhead_children, sample, period, - weight1, weight2, weight3, ins_lat, p_stage_cyc and retire_lat. The - last 3 names are alias for the corresponding weights. When the weight + overhead, latency, overhead_sys, overhead_us, overhead_children, sample, + period, weight1, weight2, weight3, ins_lat, p_stage_cyc and retire_lat. + The last 3 names are alias for the corresponding weights. When the weight fields are used, they will show the average value of the weight. Also it can contain any sort key(s).@@ -289,7 +302,7 @@ OPTIONS Accumulate callchain of children to parent entry so that then can show up in the output. The output will have a new "Children" column and will be sorted on the data. It requires callchains are recorded. - See the `overhead calculation' section for more details. Enabled by + See the `Overhead calculation' section for more details. Enabled by default, disable with --no-children. --max-stack::@@ -442,9 +455,9 @@ OPTIONS --call-graph option for details. --percentage:: - Determine how to display the overhead percentage of filtered entries. - Filters can be applied by --comms, --dsos and/or --symbols options and - Zoom operations on the TUI (thread, dso, etc). + Determine how to display the CPU and latency overhead percentage + of filtered entries. Filters can be applied by --comms, --dsos, --symbols + and/or --parallelism options and Zoom operations on the TUI (thread, dso, etc). "relative" means it's relative to filtered entries only so that the sum of shown entries will be always 100%. "absolute" means it retains@@ -627,6 +640,8 @@ include::itrace.txt[] --skip-empty:: Do not print 0 results in the --stat output. +include::cpu-and-latency-overheads.txt[] + include::callchain-overhead-calculation.txt[] SEE ALSOtools/perf/Documentation/tips.txt +0
-0
tools/perf/Documentation/tips.txt To show context switches in perf report sample context add --switch-events to pe · line 66
Quoted 5 lines
tools/perf/Documentation/tips.txt +2
-0@@ -62,3 +62,6 @@ To show context switches in perf report sample context add --switch-events to pe To show time in nanoseconds in record/report add --ns To compare hot regions in two workloads use perf record -b -o file ... ; perf diff --stream file1 file2 To compare scalability of two workload samples use perf diff -c ratio file1 file2 +For latency profiling, try: perf record/report --latency +For parallelism histogram, try: perf report --hierarchy --sort latency,parallelism,comm,symbol
Ditto.
Thanks,
Namhyung
Quoted 4 lines
tools/perf/Documentation/tips.txt +1
-0+To analyze particular parallelism levels, try: perf report --latency --parallelism=32-64Signature · 1 line
2.48.1.362.g079036d154-goog
Quoted 64 lines · Namhyung Kim
Hello,
Quoted 48 lines · Dmitry Vyukov
There are two notions of time: wall-clock time and CPU time. For a single-threaded program, or a program running on a single-core machine, these notions are the same. However, for a multi-threaded/ multi-process program running on a multi-core machine, these notions are significantly different. Each second of wall-clock time we have number-of-cores seconds of CPU time.
Currently perf only allows to profile CPU time. Perf (and all other existing profilers to the be best of my knowledge) does not allow to profile wall-clock time.
Optimizing CPU overhead is useful to improve 'throughput', while optimizing wall-clock overhead is useful to improve 'latency'. These profiles are complementary and are not interchangeable. Examples of where latency profile is needed:
- optimzing build latency
- optimizing server request latency
- optimizing ML training/inference latency
- optimizing running time of any command line programCPU profile is useless for these use cases at best (if a user understands the difference), or misleading at worst (if a user tries to use a wrong profile for a job).
This series add latency and parallelization profiling. See the added documentation and flags descriptions for details.
Brief outline of the implementation:
- add context switch collection during record
- calculate number of threads running on CPUs (parallelism level) during report
- divide each sample weight by the parallelism level
This effectively models that we were taking 1 sample per unit of wall-clock time.We still default to the CPU profile, so it's up to users to learn about the second profiling mode and use it when appropriate.
Changes in v4:
- added "Shrink structhist_entrysize" commit
- rebased to perf-tools-next HEADChanges in v3:
- rebase and split into patches
- rename 'wallclock' to 'latency' everywhere
- don't enable latency profiling by default,
instead add record/report --latency flagQuoted 9 lines
Dmitry Vyukov (8):
perf report: Add machine parallelism
perf report: Add parallelism sort key
perf report: Switch filtered from u8 to u16
perf report: Add parallelism filter
perf report: Add latency output field
perf report: Add --latency flag
perf report: Add latency and parallelism profiling documentation perf hist: Shrink structhist_entrysizeI found that stdio output is broken that latency is too high. TUI is fine.
Quoted 8 lines
$ ./perf report --latency -s dso --stdio
...
#
# Latency Overhead Shared Object
# ........ ........ ....................
#
57669.15% 57.67% [kernel.kallsyms]
42330.85% 42.33% ld-linux-x86-64.so.2Fixed in v5.
Quoted 35 lines
Thanks,
NamhyungQuoted 32 lines
.../callchain-overhead-calculation.txt | 5 +- .../cpu-and-latency-overheads.txt | 85 ++++++++++++++++++ tools/perf/Documentation/perf-report.txt | 49 ++++++---- tools/perf/Documentation/tips.txt | 3 + tools/perf/builtin-record.c | 20 +++++ tools/perf/builtin-report.c | 39 ++++++++ tools/perf/ui/browsers/hists.c | 27 +++--- tools/perf/ui/hist.c | 64 +++++++++---- tools/perf/util/addr_location.c | 1 + tools/perf/util/addr_location.h | 7 +- tools/perf/util/event.c | 11 +++ tools/perf/util/events_stats.h | 2 + tools/perf/util/hist.c | 90 +++++++++++++++---- tools/perf/util/hist.h | 32 +++++-- tools/perf/util/machine.c | 7 ++ tools/perf/util/machine.h | 6 ++ tools/perf/util/sample.h | 2 +- tools/perf/util/session.c | 12 +++ tools/perf/util/session.h | 1 + tools/perf/util/sort.c | 69 ++++++++++++-- tools/perf/util/sort.h | 3 +- tools/perf/util/symbol.c | 34 +++++++ tools/perf/util/symbol_conf.h | 8 +- 23 files changed, 502 insertions(+), 75 deletions(-) create mode 100644 tools/perf/Documentation/cpu-and-latency-overheads.txtbase-commit:
8ce0d2da14d3fb62844dd0e95982c194326b1a5f--
2.48.1.362.g079036d154-goog
Re: [PATCH v3 7/8] perf report: Add latency and parallelism profiling documentation
In reply to Namhyung KimQuoted 7 lines · Namhyung Kim
Quoted 2 lines · Dmitry Vyukov
Describe latency and parallelism profiling, related flags, and differences with the currently only supported CPU-consumption-centric profiling.
It doesn't seem to have descriptions for the --latency option (for perf record and report). Probably better to put them in the previous patch.
Done in v5.
Quoted 89 lines
Quoted 75 lines
Signed-off-by Dmitry Vyukov <[email protected]>Cc Namhyung Kim <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Ian Rogers <[email protected]>--- .../callchain-overhead-calculation.txt | 5 +- .../cpu-and-latency-overheads.txt | 85 +++++++++++++++++++ tools/perf/Documentation/perf-report.txt | 49 +++++++---- tools/perf/Documentation/tips.txt | 3 + 4 files changed, 123 insertions(+), 19 deletions(-)tools/perf/Documentation/callchain-overhead-calculation.txt +3
-2@@ -1,7 +1,8 @@ Overhead calculation -------------------- -The overhead can be shown in two columns as 'Children' and 'Self' when -perf collects callchains. The 'self' overhead is simply calculated by +The CPU overhead can be shown in two columns as 'Children' and 'Self' +when perf collects callchains (and corresponding 'Wall' columns for +wall-clock overhead). The 'self' overhead is simply calculated by adding all period values of the entry - usually a function (symbol). This is the value that perf shows traditionally and sum of all the 'self' overhead values should be 100%.tools/perf/Documentation/cpu-and-latency-overheads.txt new file +40
-0@@ -0,0 +1,85 @@ +CPU and latency overheads +------------------------- +There are two notions of time: wall-clock time and CPU time. +For a single-threaded program, or a program running on a single-core machine, +these notions are the same. However, for a multi-threaded/multi-process program +running on a multi-core machine, these notions are significantly different. +Each second of wall-clock time we have number-of-cores seconds of CPU time. +Perf can measure overhead for both of these times (shown in 'overhead' and +'latency' columns for CPU and wall-clock time correspondingly). + +Optimizing CPU overhead is useful to improve 'throughput', while optimizing +latency overhead is useful to improve 'latency'. It's important to understand +which one is useful in a concrete situation at hand. For example, the former +may be useful to improve max throughput of a CI build server that runs on 100% +CPU utilization, while the latter may be useful to improve user-perceived +latency of a single interactive program build. +These overheads may be significantly different in some cases. For example, +consider a program that executes function 'foo' for 9 seconds with 1 thread, +and then executes function 'bar' for 1 second with 128 threads (consumes +128 seconds of CPU time). The CPU overhead is: 'foo' - 6.6%, 'bar' - 93.4%. +While the latency overhead is: 'foo' - 90%, 'bar' - 10%. If we try to optimize +running time of the program looking at the (wrong in this case) CPU overhead, +we would concentrate on the function 'bar', but it can yield only 10% running +time improvement at best. + +By default, perf shows only CPU overhead. To show latency overhead, use +'perf record --latency' and 'perf report': + +----------------------------------- +Overhead Latency Command + 93.88% 25.79% cc1 + 1.90% 39.87% gzip + 0.99% 10.16% dpkg-deb + 0.57% 1.00% as + 0.40% 0.46% sh +----------------------------------- + +To sort by latency overhead, use 'perf report --latency': + +-----------------------------------tools/perf/Documentation/cpu-and-latency-overheads.txt line 50
Quoted 10 lines
tools/perf/Documentation/cpu-and-latency-overheads.txt new file +10
-0+Latency Overhead Command + 39.87% 1.90% gzip + 25.79% 93.88% cc1 + 10.16% 0.99% dpkg-deb + 4.17% 0.29% git + 2.81% 0.11% objtool +----------------------------------- + +To get insight into the difference between the overheads, you may check +parallelization histogram with '--sort=latency,parallelism,comm,symbol --hierarchy'I think you need to omit latency in the sort key and recommend users to use --latency option instead.
Quoted 1 line
'perf report --hierarchy --latency --sort=parallelism,comm,symbol'
This does not work as is b/c --hierarchy is very sensitive to field set/order and --latency adds both latency and overhead fields now.
The current command shows (intended):
Latency Parallelism / Command + 54.24% 1 + 12.67% 3 + 8.69% 2 + 2.89% 4 + 2.34% 5
and when you expand the first line:
Latency Parallelism / Command
- 54.24% 1
+ 25.15% cc1 + 7.72% ld + 4.03% sh
While --hierarchy --latency --sort=parallelism,comm,symbol shows just 1 line:
Latency Overhead / Parallelism / Command
+ 100.00% 100.00%
and when you expand it:
Latency Overhead / Parallelism / Command
- 100.00% 100.00%
+ 54.24% 1 + 12.67% 3 + 8.69% 2 + 2.89% 4
And showing both latency and overhead when --latency flag is specified looks reasonable and useful. Just perf report [--latency] is probably much more widely used than --hierarchy.
And currently for perf report --latency you see:
Latency Overhead Symbol 4.37% 38.17% [k] native_queued_spin_lock_slowpath 2.71% 0.86% [.] _PyEval_EvalFrameDefault 2.18% 0.47% [.] iterative_hash 2.00% 0.68% [k] clear_page_erms 1.26% 0.16% [.] bfd_elf_link_add_symbols 1.04% 0.11% [.] bfd_link_hash_traverse
Making --latency behave differently when --hierarchy is specified will complicate things even more.
Quoted 184 lines
Quoted 168 lines
tools/perf/Documentation/cpu-and-latency-overheads.txt new file +35
-0+flags. It shows fraction of (wall-clock) time the workload utilizes different +numbers of cores ('Parallelism' column). For example, in the following case +the workload utilizes only 1 core most of the time, but also has some +highly-parallel phases, which explains significant difference between +CPU and wall-clock overheads: + +----------------------------------- + Latency Overhead Parallelism / Command / Symbol ++ 56.98% 2.29% 1 ++ 16.94% 1.36% 2 ++ 4.00% 20.13% 125 ++ 3.66% 18.25% 124 ++ 3.48% 17.66% 126 ++ 3.26% 0.39% 3 ++ 2.61% 12.93% 123 +----------------------------------- + +By expanding corresponding lines, you may see what commands/functions run +at the given parallelism level: + +----------------------------------- + Latency Overhead Parallelism / Command / Symbol +- 56.98% 2.29% 1 + 32.80% 1.32% gzip + 4.46% 0.18% cc1 + 2.81% 0.11% objtool + 2.43% 0.10% dpkg-source + 2.22% 0.09% ld + 2.10% 0.08% dpkg-genchanges +----------------------------------- + +To see the normal function-level profile for particular parallelism levels +(number of threads actively running on CPUs), you may use '--parallelism' +filter. For example, to see the profile only for low parallelism phases +of a workload use '--latency --parallelism=1-2' flags.tools/perf/Documentation/perf-report.txt +32
-17@@ -44,7 +44,7 @@ OPTIONS --comms=:: Only consider symbols in these comms. CSV that understands file://filename entries. This option will affect the percentage of - the overhead column. See --percentage for more info. + the overhead and latency columns. See --percentage for more info. --pid=:: Only show events for given process ID (comma separated list).@@ -54,12 +54,12 @@ OPTIONS --dsos=:: Only consider symbols in these dsos. CSV that understands file://filename entries. This option will affect the percentage of - the overhead column. See --percentage for more info. + the overhead and latency columns. See --percentage for more info. -S:: --symbols=:: Only consider these symbols. CSV that understands file://filename entries. This option will affect the percentage of - the overhead column. See --percentage for more info. + the overhead and latency columns. See --percentage for more info. --symbol-filter=:: Only show symbols that match (partially) with this filter.@@ -68,6 +68,16 @@ OPTIONS --hide-unresolved:: Only display entries resolved to a symbol. +--parallelism:: + Only consider these parallelism levels. Parallelism level is the number + of threads that actively run on CPUs at the time of sample. The flag + accepts single number, comma-separated list, and ranges (for example: + "1", "7,8", "1,64-128"). This is useful in understanding what a program + is doing during sequential/low-parallelism phases as compared to + high-parallelism phases. This option will affect the percentage of + the overhead and latency columns. See --percentage for more info. + Also see the `CPU and latency overheads' section for more details. + -s:: --sort=:: Sort histogram entries by given key(s) - multiple keys can be specified@@ -87,6 +97,7 @@ OPTIONS entries are displayed as "[other]". - cpu: cpu number the task ran at the time of sample - socket: processor socket number the task ran at the time of sample + - parallelism: number of running threads at the time of sample - srcline: filename and line number executed at the time of sample. The DWARF debugging info must be provided. - srcfile: file name of the source file of the samples. Requires dwarf@@ -97,12 +108,14 @@ OPTIONS - cgroup_id: ID derived from cgroup namespace device and inode numbers. - cgroup: cgroup pathname in the cgroupfs. - transaction: Transaction abort flags. - - overhead: Overhead percentage of sample - - overhead_sys: Overhead percentage of sample running in system mode - - overhead_us: Overhead percentage of sample running in user mode - - overhead_guest_sys: Overhead percentage of sample running in system mode + - overhead: CPU overhead percentage of sample. + - latency: latency (wall-clock) overhead percentage of sample. + See the `CPU and latency overheads' section for more details. + - overhead_sys: CPU overhead percentage of sample running in system mode + - overhead_us: CPU overhead percentage of sample running in user mode + - overhead_guest_sys: CPU overhead percentage of sample running in system mode on guest machine - - overhead_guest_us: Overhead percentage of sample running in user mode on + - overhead_guest_us: CPU overhead percentage of sample running in user mode on guest machine - sample: Number of sample - period: Raw number of event count of sample@@ -125,8 +138,8 @@ OPTIONS - weight2: Average value of event specific weight (2nd field of weight_struct). - weight3: Average value of event specific weight (3rd field of weight_struct). - By default, comm, dso and symbol keys are used. - (i.e. --sort comm,dso,symbol) + By default, overhead, comm, dso and symbol keys are used. + (i.e. --sort overhead,comm,dso,symbol). If --branch-stack option is used, following sort keys are also available:@@ -201,9 +214,9 @@ OPTIONS --fields=:: Specify output field - multiple keys can be specified in CSV format. Following fields are available: - overhead, overhead_sys, overhead_us, overhead_children, sample, period, - weight1, weight2, weight3, ins_lat, p_stage_cyc and retire_lat. The - last 3 names are alias for the corresponding weights. When the weight + overhead, latency, overhead_sys, overhead_us, overhead_children, sample, + period, weight1, weight2, weight3, ins_lat, p_stage_cyc and retire_lat. + The last 3 names are alias for the corresponding weights. When the weight fields are used, they will show the average value of the weight. Also it can contain any sort key(s).@@ -289,7 +302,7 @@ OPTIONS Accumulate callchain of children to parent entry so that then can show up in the output. The output will have a new "Children" column and will be sorted on the data. It requires callchains are recorded. - See the `overhead calculation' section for more details. Enabled by + See the `Overhead calculation' section for more details. Enabled by default, disable with --no-children. --max-stack::@@ -442,9 +455,9 @@ OPTIONS --call-graph option for details. --percentage:: - Determine how to display the overhead percentage of filtered entries. - Filters can be applied by --comms, --dsos and/or --symbols options and - Zoom operations on the TUI (thread, dso, etc). + Determine how to display the CPU and latency overhead percentage + of filtered entries. Filters can be applied by --comms, --dsos, --symbols + and/or --parallelism options and Zoom operations on the TUI (thread, dso, etc). "relative" means it's relative to filtered entries only so that the sum of shown entries will be always 100%. "absolute" means it retains@@ -627,6 +640,8 @@ include::itrace.txt[] --skip-empty:: Do not print 0 results in the --stat output. +include::cpu-and-latency-overheads.txt[] + include::callchain-overhead-calculation.txt[] SEE ALSOtools/perf/Documentation/tips.txt +0
-0tools/perf/Documentation/tips.txt To show context switches in perf report sample context add --switch-events to pe · line 66
Quoted 5 lines
tools/perf/Documentation/tips.txt +2
-0@@ -62,3 +62,6 @@ To show context switches in perf report sample context add --switch-events to pe To show time in nanoseconds in record/report add --ns To compare hot regions in two workloads use perf record -b -o file ... ; perf diff --stream file1 file2 To compare scalability of two workload samples use perf diff -c ratio file1 file2 +For latency profiling, try: perf record/report --latency +For parallelism histogram, try: perf report --hierarchy --sort latency,parallelism,comm,symbolDitto.
Thanks,
NamhyungQuoted 4 lines
tools/perf/Documentation/tips.txt +1
-1+To analyze particular parallelism levels, try: perf report --latency --parallelism=32-64 --2.48.1.362.g079036d154-goog