Generic weight struct, use env for sort key and header
Ian Rogers
reviewerv3 cover Generic weight struct, use env for sort key and header
The arch directory is a barrier to cross-platform development as files and behaviors within it are inherently platform specific. Sample parsing should be generic but the PERF_SAMPLE_WEIGHT_STRUCT handling was only present if building for x86 or powerpc. The sort key and headers should be specific to the session that is being executed and not to the machine perf is being run upon. These patches clean this and associated code up.
v3: Code reorganizations suggested by Kan Liang.
v2: Avoid changes to include/perf/perf_dlfilter.h as suggested by
Adrian Hunter.
Ian Rogers (3):
perf sample: Remove arch notion of sample parsing perf test: Move PERF_SAMPLE_WEIGHT_STRUCT parsing to common test perf sort: Use perf_env to set arch sort keys and header
tools/perf/arch/powerpc/util/Build | 1 - tools/perf/arch/powerpc/util/event.c | 60 ---------- tools/perf/arch/x86/include/arch-tests.h | 1 - tools/perf/arch/x86/tests/Build | 1 - tools/perf/arch/x86/tests/arch-tests.c | 2 - tools/perf/arch/x86/tests/sample-parsing.c | 125 --------------------- tools/perf/arch/x86/util/event.c | 46 -------- tools/perf/builtin-annotate.c | 2 +- tools/perf/builtin-c2c.c | 53 +++++---- tools/perf/builtin-diff.c | 2 +- tools/perf/builtin-report.c | 2 +- tools/perf/builtin-script.c | 2 +- tools/perf/builtin-top.c | 16 +-- tools/perf/tests/hists_cumulate.c | 8 +- tools/perf/tests/hists_filter.c | 8 +- tools/perf/tests/hists_link.c | 8 +- tools/perf/tests/hists_output.c | 10 +- tools/perf/tests/sample-parsing.c | 14 +++ tools/perf/util/dlfilter.c | 2 +- tools/perf/util/event.h | 5 - tools/perf/util/evsel.c | 17 ++- tools/perf/util/hist.c | 4 +- tools/perf/util/hist.h | 2 +- tools/perf/util/intel-tpebs.c | 4 +- tools/perf/util/sample.h | 5 +- tools/perf/util/session.c | 2 +- tools/perf/util/sort.c | 67 +++++++---- tools/perf/util/sort.h | 5 +- tools/perf/util/synthetic-events.c | 10 +- 29 files changed, 150 insertions(+), 334 deletions(-) delete mode 100644 tools/perf/arch/powerpc/util/event.c delete mode 100644 tools/perf/arch/x86/tests/sample-parsing.c
Signature · 1 line
2.49.0.1143.g0be31eac6b-goog
Ian Rogers
reviewerPATCH v3 1/3 perf sample: Remove arch notion of sample parsing
By definition arch sample parsing and synthesis will inhibit certain kinds of cross-platform record then analysis (report, script, etc.). Remove arch_perf_parse_sample_weight and arch_perf_synthesize_sample_weight replacing with a common implementation. Combine perf_sample p_stage_cyc and retire_lat to capture the differing uses regardless of compiled for architecture.
--- tools/perf/arch/powerpc/util/event.c | 26 --------------------- tools/perf/arch/x86/tests/sample-parsing.c | 4 ++-- tools/perf/arch/x86/util/event.c | 27 ---------------------- tools/perf/builtin-script.c | 2 +- tools/perf/util/dlfilter.c | 2 +- tools/perf/util/event.h | 2 -- tools/perf/util/evsel.c | 17 ++++++++++---- tools/perf/util/hist.c | 4 ++-- tools/perf/util/hist.h | 2 +- tools/perf/util/intel-tpebs.c | 4 ++-- tools/perf/util/sample.h | 5 +--- tools/perf/util/session.c | 2 +- tools/perf/util/sort.c | 6 ++--- tools/perf/util/synthetic-events.c | 10 ++++++-- 14 files changed, 34 insertions(+), 79 deletions(-)
tools/perf/arch/powerpc/util/event.c +0 -26
@@ -11,32 +11,6 @@#include "../../../util/debug.h" #include "../../../util/sample.h" -void arch_perf_parse_sample_weight (struct perf_sample *data, -const __u64 *array,u64 type) -{ -union perf_sample_weight weight; - - weight.full = *array; -if (type &PERF_SAMPLE_WEIGHT ) - data->weight = weight.full; -else { - data->weight = weight.var1_dw; - data->ins_lat = weight.var2_w; - data->p_stage_cyc = weight.var3_w; - } -} - -void arch_perf_synthesize_sample_weight (const struct perf_sample *data, -__u64 *array,u64 type) -{ - *array = data->weight; - -if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { - *array &=0xffffffff ; - *array |= ((u64 )data->ins_lat <<32 ); - } -} -const char *arch_perf_header_entry (const char *se_header) {if (!strcmp (se_header,"Local INSTR Latency" ))
tools/perf/arch/x86/tests/sample-parsing.c +2 -2
@@ -29,7 +29,7 @@ static bool samples_same(const struct perf_sample *s1, {if (type &PERF_SAMPLE_WEIGHT_STRUCT ) {COMP (ins_lat); -COMP (retire_lat); +COMP (p_stage_cyc_or_retire_lat); }return true;
@@ -50,7 +50,7 @@ static int do_test(u64 sample_type)struct perf_sample sample = { .weight =101 , .ins_lat =102 , - .retire_lat =103 , + .p_stage_cyc_or_retire_lat =103 , };struct perf_sample sample_out;size_t i, sz, bufsz;
tools/perf/arch/x86/util/event.c +0 -27
@@ -92,33 +92,6 @@ int perf_event__synthesize_extra_kmaps(const struct perf_tool *tool,#endif -void arch_perf_parse_sample_weight (struct perf_sample *data, -const __u64 *array,u64 type) -{ -union perf_sample_weight weight; - - weight.full = *array; -if (type &PERF_SAMPLE_WEIGHT ) - data->weight = weight.full; -else { - data->weight = weight.var1_dw; - data->ins_lat = weight.var2_w; - data->retire_lat = weight.var3_w; - } -} - -void arch_perf_synthesize_sample_weight (const struct perf_sample *data, -__u64 *array,u64 type) -{ - *array = data->weight; - -if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { - *array &=0xffffffff ; - *array |= ((u64 )data->ins_lat <<32 ); - *array |= ((u64 )data->retire_lat <<48 ); - } -} -const char *arch_perf_header_entry (const char *se_header) {if (!strcmp (se_header,"Local Pipeline Stage Cycle" ))
tools/perf/builtin-script.c +1 -1
@@ -2251,7 +2251,7 @@ static void process_event(struct perf_script *script,fprintf (fp,"%16" PRIu16, sample->ins_lat);if (PRINT_FIELD (RETIRE_LAT )) -fprintf (fp,"%16" PRIu16, sample->retire_lat); +fprintf (fp,"%16" PRIu16, sample->p_stage_cyc_or_retire_lat);if (PRINT_FIELD (CGROUP )) {const char *cgrp_name;
tools/perf/util/dlfilter.c +1 -1
@@ -513,6 +513,7 @@ int dlfilter__do_filter_event(struct dlfilter *d, d->d_addr_al = &d_addr_al; d_sample.size =sizeof (d_sample); + d_sample.p_stage_cyc = sample->p_stage_cyc_or_retire_lat; d_ip_al.size =0 ;/* To indicate d_ip_al is not initialized */ d_addr_al.size =0 ;/* To indicate d_addr_al is not initialized */
@@ -526,7 +527,6 @@ int dlfilter__do_filter_event(struct dlfilter *d,ASSIGN (period);ASSIGN (weight);ASSIGN (ins_lat); -ASSIGN (p_stage_cyc);ASSIGN (transaction);ASSIGN (insn_cnt);ASSIGN (cyc_cnt);
tools/perf/util/event.h +0 -2
@@ -390,8 +390,6 @@ extern unsigned int proc_map_timeout;#define PAGE_SIZE_NAME_LEN 32char *get_page_size_name (u64 size,char *str); -void arch_perf_parse_sample_weight (struct perf_sample *data,const __u64 *array,u64 type); -void arch_perf_synthesize_sample_weight (const struct perf_sample *data,__u64 *array,u64 type);const char *arch_perf_header_entry (const char *se_header);int arch_support_sort_key (const char *sort_key);
tools/perf/util/evsel.c +12 -5
@@ -2846,11 +2846,18 @@ perf_event__check_size(union perf_event *event, unsigned int sample_size)return 0 ; } -void __weak arch_perf_parse_sample_weight (struct perf_sample *data, -const __u64 *array, -u64 type__maybe_unused ) +static void perf_parse_sample_weight (struct perf_sample *data,const __u64 *array,u64 type) { - data->weight = *array; +union perf_sample_weight weight; + + weight.full = *array; +if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { + data->weight = weight.var1_dw; + data->ins_lat = weight.var2_w; + data->p_stage_cyc_or_retire_lat = weight.var3_w; + }else { + data->weight = weight.full; + } }u64 evsel__bitfield_swap_branch_flags (u64 value)
@@ -3236,7 +3243,7 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event,if (type &PERF_SAMPLE_WEIGHT_TYPE ) {OVERFLOW_CHECK_u64 (array); -arch_perf_parse_sample_weight (data, array, type); +perf_parse_sample_weight (data, array, type); array++; }
tools/perf/util/hist.c +2 -2
@@ -829,7 +829,7 @@ __hists__add_entry(struct hists *hists, .period = sample->period, .weight1 = sample->weight, .weight2 = sample->ins_lat, - .weight3 = sample->p_stage_cyc, + .weight3 = sample->p_stage_cyc_or_retire_lat, .latency = al->latency, }, .parent = sym_parent,
@@ -846,7 +846,7 @@ __hists__add_entry(struct hists *hists, .time = hist_time(sample->time), .weight = sample->weight, .ins_lat = sample->ins_lat, - .p_stage_cyc = sample->p_stage_cyc, + .p_stage_cyc_or_retire_lat = sample->p_stage_cyc_or_retire_lat, .simd_flags = sample->simd_flags, }, *he = hists__findnew_entry(hists, &entry, al, sample_self);
tools/perf/util/hist.h +1 -1
@@ -255,7 +255,7 @@ struct hist_entry {u64 code_page_size;u64 weight;u64 ins_lat; -u64 p_stage_cyc; +u64 p_stage_cyc_or_retire_lat;s32 socket;s32 cpu;int parallelism;
tools/perf/util/intel-tpebs.c +2 -2
@@ -202,8 +202,8 @@ static int process_sample_event(const struct perf_tool *tool __maybe_unused,* latency value will be used. Save the number of samples and the sum of * retire latency value for each event. */ - t->last = sample->retire_lat; -update_stats (&t->stats, sample->retire_lat); + t->last = sample->p_stage_cyc_or_retire_lat; +update_stats (&t->stats, sample->p_stage_cyc_or_retire_lat);mutex_unlock (tpebs_mtx_get ());return 0 ; }
tools/perf/util/sample.h +1 -4
@@ -104,10 +104,7 @@ struct perf_sample {u8 cpumode;u16 misc;u16 ins_lat; -union { -u16 p_stage_cyc; -u16 retire_lat; - }; +u16 p_stage_cyc_or_retire_lat;bool no_hw_idx;/* No hw_idx collected in branch_stack */ char insn[MAX_INSN ];void *raw_data;
tools/perf/util/session.c +1 -1
@@ -1094,7 +1094,7 @@ static void dump_sample(struct evsel *evsel, union perf_event *event,printf ("... weight: %" PRIu64"" , sample->weight);if (sample_type &PERF_SAMPLE_WEIGHT_STRUCT ) {printf (",0x%" PRIx16"" , sample->ins_lat); -printf (",0x%" PRIx16"" , sample->p_stage_cyc); +printf (",0x%" PRIx16"" , sample->p_stage_cyc_or_retire_lat); }printf ("\n" ); }
tools/perf/util/sort.c +3 -3
@@ -1879,21 +1879,21 @@ struct sort_entry sort_global_ins_lat = {static int64_t sort__p_stage_cyc_cmp (struct hist_entry *left,struct hist_entry *right) { -return left->p_stage_cyc - right->p_stage_cyc; +return left->p_stage_cyc_or_retire_lat - right->p_stage_cyc_or_retire_lat; }static int hist_entry__global_p_stage_cyc_snprintf (struct hist_entry *he,char *bf,size_t size,unsigned int width) {return repsep_snprintf (bf, size,"%-*u" , width, - he->p_stage_cyc * he->stat.nr_events); + he->p_stage_cyc_or_retire_lat * he->stat.nr_events); }static int hist_entry__p_stage_cyc_snprintf (struct hist_entry *he,char *bf,size_t size,unsigned int width) { -return repsep_snprintf (bf, size,"%-*u" , width, he->p_stage_cyc); +return repsep_snprintf (bf, size,"%-*u" , width, he->p_stage_cyc_or_retire_lat); }struct sort_entry sort_local_p_stage_cyc = {
tools/perf/util/synthetic-events.c +8 -2
@@ -1567,10 +1567,16 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,return result; } -void __weak arch_perf_synthesize_sample_weight (const struct perf_sample *data, +static void perf_synthesize_sample_weight (const struct perf_sample *data,__u64 *array,u64 type__maybe_unused ) { *array = data->weight; + +if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { + *array &=0xffffffff ; + *array |= ((u64 )data->ins_lat <<32 ); + *array |= ((u64 )data->p_stage_cyc_or_retire_lat <<48 ); + } } static __u64 *copy_read_group_values(__u64 *array, __u64 read_format,
@@ -1730,7 +1736,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type, u64 read_fo }if (type &PERF_SAMPLE_WEIGHT_TYPE ) { -arch_perf_synthesize_sample_weight (sample, array, type); +perf_synthesize_sample_weight (sample, array, type); array++; }
Signature · 1 line
2.49.0.1143.g0be31eac6b-goog
Namhyung Kim
maintainerRe: [PATCH v3 1/3] perf sample: Remove arch notion of sample parsing
Quoted 6 lines · Ian Rogers
By definition arch sample parsing and synthesis will inhibit certain kinds of cross-platform record then analysis (report, script, etc.). Remove
arch_perf_parse_sample_weightandarch_perf_synthesize_sample_weightreplacing with a common implementation. Combineperf_samplep_stage_cycandretire_latto capture the differing uses regardless of compiled for architecture.
Can you please do this without renaming? It can be a separate patch but I think we can just leave it.
Thanks,
Namhyung
Quoted 340 lines
Signed-off-by Ian Rogers <[email protected]>--- tools/perf/arch/powerpc/util/event.c | 26 --------------------- tools/perf/arch/x86/tests/sample-parsing.c | 4 ++-- tools/perf/arch/x86/util/event.c | 27 ---------------------- tools/perf/builtin-script.c | 2 +- tools/perf/util/dlfilter.c | 2 +- tools/perf/util/event.h | 2 -- tools/perf/util/evsel.c | 17 ++++++++++---- tools/perf/util/hist.c | 4 ++-- tools/perf/util/hist.h | 2 +- tools/perf/util/intel-tpebs.c | 4 ++-- tools/perf/util/sample.h | 5 +--- tools/perf/util/session.c | 2 +- tools/perf/util/sort.c | 6 ++--- tools/perf/util/synthetic-events.c | 10 ++++++-- 14 files changed, 34 insertions(+), 79 deletions(-)tools/perf/arch/powerpc/util/event.c +0
-26@@ -11,32 +11,6 @@#include "../../../util/debug.h" #include "../../../util/sample.h" -void arch_perf_parse_sample_weight (struct perf_sample *data, -const __u64 *array,u64 type) -{ -union perf_sample_weight weight; - - weight.full = *array; -if (type &PERF_SAMPLE_WEIGHT ) - data->weight = weight.full; -else { - data->weight = weight.var1_dw; - data->ins_lat = weight.var2_w; - data->p_stage_cyc = weight.var3_w; - } -} - -void arch_perf_synthesize_sample_weight (const struct perf_sample *data, -__u64 *array,u64 type) -{ - *array = data->weight; - -if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { - *array &=0xffffffff ; - *array |= ((u64 )data->ins_lat <<32 ); - } -} -const char *arch_perf_header_entry (const char *se_header) {if (!strcmp (se_header,"Local INSTR Latency" ))tools/perf/arch/x86/tests/sample-parsing.c +2
-2@@ -29,7 +29,7 @@ static bool samples_same(const struct perf_sample *s1, {if (type &PERF_SAMPLE_WEIGHT_STRUCT ) {COMP (ins_lat); -COMP (retire_lat); +COMP (p_stage_cyc_or_retire_lat); }return true;@@ -50,7 +50,7 @@ static int do_test(u64 sample_type)struct perf_sample sample = { .weight =101 , .ins_lat =102 , - .retire_lat =103 , + .p_stage_cyc_or_retire_lat =103 , };struct perf_sample sample_out;size_t i, sz, bufsz;tools/perf/arch/x86/util/event.c +0
-27@@ -92,33 +92,6 @@ int perf_event__synthesize_extra_kmaps(const struct perf_tool *tool,#endif -void arch_perf_parse_sample_weight (struct perf_sample *data, -const __u64 *array,u64 type) -{ -union perf_sample_weight weight; - - weight.full = *array; -if (type &PERF_SAMPLE_WEIGHT ) - data->weight = weight.full; -else { - data->weight = weight.var1_dw; - data->ins_lat = weight.var2_w; - data->retire_lat = weight.var3_w; - } -} - -void arch_perf_synthesize_sample_weight (const struct perf_sample *data, -__u64 *array,u64 type) -{ - *array = data->weight; - -if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { - *array &=0xffffffff ; - *array |= ((u64 )data->ins_lat <<32 ); - *array |= ((u64 )data->retire_lat <<48 ); - } -} -const char *arch_perf_header_entry (const char *se_header) {if (!strcmp (se_header,"Local Pipeline Stage Cycle" ))tools/perf/builtin-script.c +1
-1@@ -2251,7 +2251,7 @@ static void process_event(struct perf_script *script,fprintf (fp,"%16" PRIu16, sample->ins_lat);if (PRINT_FIELD (RETIRE_LAT )) -fprintf (fp,"%16" PRIu16, sample->retire_lat); +fprintf (fp,"%16" PRIu16, sample->p_stage_cyc_or_retire_lat);if (PRINT_FIELD (CGROUP )) {const char *cgrp_name;tools/perf/util/dlfilter.c +1
-1@@ -513,6 +513,7 @@ int dlfilter__do_filter_event(struct dlfilter *d, d->d_addr_al = &d_addr_al; d_sample.size =sizeof (d_sample); + d_sample.p_stage_cyc = sample->p_stage_cyc_or_retire_lat; d_ip_al.size =0 ;/* To indicate d_ip_al is not initialized */ d_addr_al.size =0 ;/* To indicate d_addr_al is not initialized */ @@ -526,7 +527,6 @@ int dlfilter__do_filter_event(struct dlfilter *d,ASSIGN (period);ASSIGN (weight);ASSIGN (ins_lat); -ASSIGN (p_stage_cyc);ASSIGN (transaction);ASSIGN (insn_cnt);ASSIGN (cyc_cnt);tools/perf/util/event.h +0
-2@@ -390,8 +390,6 @@ extern unsigned int proc_map_timeout;#define PAGE_SIZE_NAME_LEN 32char *get_page_size_name (u64 size,char *str); -void arch_perf_parse_sample_weight (struct perf_sample *data,const __u64 *array,u64 type); -void arch_perf_synthesize_sample_weight (const struct perf_sample *data,__u64 *array,u64 type);const char *arch_perf_header_entry (const char *se_header);int arch_support_sort_key (const char *sort_key);tools/perf/util/evsel.c +12
-5@@ -2846,11 +2846,18 @@ perf_event__check_size(union perf_event *event, unsigned int sample_size)return 0 ; } -void __weak arch_perf_parse_sample_weight (struct perf_sample *data, -const __u64 *array, -u64 type__maybe_unused ) +static void perf_parse_sample_weight (struct perf_sample *data,const __u64 *array,u64 type) { - data->weight = *array; +union perf_sample_weight weight; + + weight.full = *array; +if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { + data->weight = weight.var1_dw; + data->ins_lat = weight.var2_w; + data->p_stage_cyc_or_retire_lat = weight.var3_w; + }else { + data->weight = weight.full; + } }u64 evsel__bitfield_swap_branch_flags (u64 value)@@ -3236,7 +3243,7 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event,if (type &PERF_SAMPLE_WEIGHT_TYPE ) {OVERFLOW_CHECK_u64 (array); -arch_perf_parse_sample_weight (data, array, type); +perf_parse_sample_weight (data, array, type); array++; }tools/perf/util/hist.c +2
-2@@ -829,7 +829,7 @@ __hists__add_entry(struct hists *hists, .period = sample->period, .weight1 = sample->weight, .weight2 = sample->ins_lat, - .weight3 = sample->p_stage_cyc, + .weight3 = sample->p_stage_cyc_or_retire_lat, .latency = al->latency, }, .parent = sym_parent,@@ -846,7 +846,7 @@ __hists__add_entry(struct hists *hists, .time = hist_time(sample->time), .weight = sample->weight, .ins_lat = sample->ins_lat, - .p_stage_cyc = sample->p_stage_cyc, + .p_stage_cyc_or_retire_lat = sample->p_stage_cyc_or_retire_lat, .simd_flags = sample->simd_flags, }, *he = hists__findnew_entry(hists, &entry, al, sample_self);tools/perf/util/hist.h +1
-1@@ -255,7 +255,7 @@ struct hist_entry {u64 code_page_size;u64 weight;u64 ins_lat; -u64 p_stage_cyc; +u64 p_stage_cyc_or_retire_lat;s32 socket;s32 cpu;int parallelism;tools/perf/util/intel-tpebs.c +2
-2@@ -202,8 +202,8 @@ static int process_sample_event(const struct perf_tool *tool __maybe_unused,* latency value will be used. Save the number of samples and the sum of * retire latency value for each event. */ - t->last = sample->retire_lat; -update_stats (&t->stats, sample->retire_lat); + t->last = sample->p_stage_cyc_or_retire_lat; +update_stats (&t->stats, sample->p_stage_cyc_or_retire_lat);mutex_unlock (tpebs_mtx_get ());return 0 ; }tools/perf/util/sample.h +1
-4@@ -104,10 +104,7 @@ struct perf_sample {u8 cpumode;u16 misc;u16 ins_lat; -union { -u16 p_stage_cyc; -u16 retire_lat; - }; +u16 p_stage_cyc_or_retire_lat;bool no_hw_idx;/* No hw_idx collected in branch_stack */ char insn[MAX_INSN ];void *raw_data;tools/perf/util/session.c +1
-1@@ -1094,7 +1094,7 @@ static void dump_sample(struct evsel *evsel, union perf_event *event,printf ("... weight: %" PRIu64"" , sample->weight);if (sample_type &PERF_SAMPLE_WEIGHT_STRUCT ) {printf (",0x%" PRIx16"" , sample->ins_lat); -printf (",0x%" PRIx16"" , sample->p_stage_cyc); +printf (",0x%" PRIx16"" , sample->p_stage_cyc_or_retire_lat); }printf ("\n" ); }tools/perf/util/sort.c +3
-3@@ -1879,21 +1879,21 @@ struct sort_entry sort_global_ins_lat = {static int64_t sort__p_stage_cyc_cmp (struct hist_entry *left,struct hist_entry *right) { -return left->p_stage_cyc - right->p_stage_cyc; +return left->p_stage_cyc_or_retire_lat - right->p_stage_cyc_or_retire_lat; }static int hist_entry__global_p_stage_cyc_snprintf (struct hist_entry *he,char *bf,size_t size,unsigned int width) {return repsep_snprintf (bf, size,"%-*u" , width, - he->p_stage_cyc * he->stat.nr_events); + he->p_stage_cyc_or_retire_lat * he->stat.nr_events); }static int hist_entry__p_stage_cyc_snprintf (struct hist_entry *he,char *bf,size_t size,unsigned int width) { -return repsep_snprintf (bf, size,"%-*u" , width, he->p_stage_cyc); +return repsep_snprintf (bf, size,"%-*u" , width, he->p_stage_cyc_or_retire_lat); }struct sort_entry sort_local_p_stage_cyc = {tools/perf/util/synthetic-events.c +8
-2@@ -1567,10 +1567,16 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,return result; } -void __weak arch_perf_synthesize_sample_weight (const struct perf_sample *data, +static void perf_synthesize_sample_weight (const struct perf_sample *data,__u64 *array,u64 type__maybe_unused ) { *array = data->weight; + +if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { + *array &=0xffffffff ; + *array |= ((u64 )data->ins_lat <<32 ); + *array |= ((u64 )data->p_stage_cyc_or_retire_lat <<48 ); + } } static __u64 *copy_read_group_values(__u64 *array, __u64 read_format,@@ -1730,7 +1736,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type, u64 read_fo }if (type &PERF_SAMPLE_WEIGHT_TYPE ) { -arch_perf_synthesize_sample_weight (sample, array, type); +perf_synthesize_sample_weight (sample, array, type); array++; }Signature · 1 line
2.49.0.1143.g0be31eac6b-goog
Ian Rogers
reviewerRe: [PATCH v3 1/3] perf sample: Remove arch notion of sample parsing
Quoted 11 lines · Namhyung Kim
Quoted 6 lines · Ian Rogers
By definition arch sample parsing and synthesis will inhibit certain kinds of cross-platform record then analysis (report, script, etc.). Remove
arch_perf_parse_sample_weightandarch_perf_synthesize_sample_weightreplacing with a common implementation. Combineperf_samplep_stage_cycandretire_latto capture the differing uses regardless of compiled for architecture.Can you please do this without renaming? It can be a separate patch but I think we can just leave it.
It is not clear what the use of the union is. Presumably it is a tagged union but there is no tag as the union value to use is implied by either being built on x86_64 or powerpc. The change removes the notion of this code being built for x86_64 or powerpc and so the union value to use isn't clear (e.g. should arm use p_stage_cyc or retire_lat from the union), hence combining to show that it could be one or the other. The code could be:
```
#ifdef __x86_64__
u16 p_stage_cyc;
#elif defined(powerpc)
u16 retire_lat;
#endif
```
but this isn't cross-platform. The change in hist.h of
```
@@ -255,7 +255,7 @@ struct hist_entry {
u64 code_page_size;
u64 weight;
u64 ins_lat;
- u64 p_stage_cyc;
+ u64 p_stage_cyc_or_retire_lat;
```could be a follow up CL, but then we lose something of what the field is holding given the value is just a copy of that same named value in perf_sample. The code only inserts 34 lines and so the churn of doing that seemed worse than having the change in a single patch for clarity.
Thanks,
Ian
Quoted 343 lines
Thanks,
NamhyungQuoted 340 lines
Signed-off-by Ian Rogers <[email protected]>--- tools/perf/arch/powerpc/util/event.c | 26 --------------------- tools/perf/arch/x86/tests/sample-parsing.c | 4 ++-- tools/perf/arch/x86/util/event.c | 27 ---------------------- tools/perf/builtin-script.c | 2 +- tools/perf/util/dlfilter.c | 2 +- tools/perf/util/event.h | 2 -- tools/perf/util/evsel.c | 17 ++++++++++---- tools/perf/util/hist.c | 4 ++-- tools/perf/util/hist.h | 2 +- tools/perf/util/intel-tpebs.c | 4 ++-- tools/perf/util/sample.h | 5 +--- tools/perf/util/session.c | 2 +- tools/perf/util/sort.c | 6 ++--- tools/perf/util/synthetic-events.c | 10 ++++++-- 14 files changed, 34 insertions(+), 79 deletions(-)tools/perf/arch/powerpc/util/event.c +0
-26@@ -11,32 +11,6 @@#include "../../../util/debug.h" #include "../../../util/sample.h" -void arch_perf_parse_sample_weight (struct perf_sample *data, -const __u64 *array,u64 type) -{ -union perf_sample_weight weight; - - weight.full = *array; -if (type &PERF_SAMPLE_WEIGHT ) - data->weight = weight.full; -else { - data->weight = weight.var1_dw; - data->ins_lat = weight.var2_w; - data->p_stage_cyc = weight.var3_w; - } -} - -void arch_perf_synthesize_sample_weight (const struct perf_sample *data, -__u64 *array,u64 type) -{ - *array = data->weight; - -if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { - *array &=0xffffffff ; - *array |= ((u64 )data->ins_lat <<32 ); - } -} -const char *arch_perf_header_entry (const char *se_header) {if (!strcmp (se_header,"Local INSTR Latency" ))tools/perf/arch/x86/tests/sample-parsing.c +2
-2@@ -29,7 +29,7 @@ static bool samples_same(const struct perf_sample *s1, {if (type &PERF_SAMPLE_WEIGHT_STRUCT ) {COMP (ins_lat); -COMP (retire_lat); +COMP (p_stage_cyc_or_retire_lat); }return true;@@ -50,7 +50,7 @@ static int do_test(u64 sample_type)struct perf_sample sample = { .weight =101 , .ins_lat =102 , - .retire_lat =103 , + .p_stage_cyc_or_retire_lat =103 , };struct perf_sample sample_out;size_t i, sz, bufsz;tools/perf/arch/x86/util/event.c +0
-27@@ -92,33 +92,6 @@ int perf_event__synthesize_extra_kmaps(const struct perf_tool *tool,#endif -void arch_perf_parse_sample_weight (struct perf_sample *data, -const __u64 *array,u64 type) -{ -union perf_sample_weight weight; - - weight.full = *array; -if (type &PERF_SAMPLE_WEIGHT ) - data->weight = weight.full; -else { - data->weight = weight.var1_dw; - data->ins_lat = weight.var2_w; - data->retire_lat = weight.var3_w; - } -} - -void arch_perf_synthesize_sample_weight (const struct perf_sample *data, -__u64 *array,u64 type) -{ - *array = data->weight; - -if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { - *array &=0xffffffff ; - *array |= ((u64 )data->ins_lat <<32 ); - *array |= ((u64 )data->retire_lat <<48 ); - } -} -const char *arch_perf_header_entry (const char *se_header) {if (!strcmp (se_header,"Local Pipeline Stage Cycle" ))tools/perf/builtin-script.c +1
-1@@ -2251,7 +2251,7 @@ static void process_event(struct perf_script *script,fprintf (fp,"%16" PRIu16, sample->ins_lat);if (PRINT_FIELD (RETIRE_LAT )) -fprintf (fp,"%16" PRIu16, sample->retire_lat); +fprintf (fp,"%16" PRIu16, sample->p_stage_cyc_or_retire_lat);if (PRINT_FIELD (CGROUP )) {const char *cgrp_name;tools/perf/util/dlfilter.c +1
-1@@ -513,6 +513,7 @@ int dlfilter__do_filter_event(struct dlfilter *d, d->d_addr_al = &d_addr_al; d_sample.size =sizeof (d_sample); + d_sample.p_stage_cyc = sample->p_stage_cyc_or_retire_lat; d_ip_al.size =0 ;/* To indicate d_ip_al is not initialized */ d_addr_al.size =0 ;/* To indicate d_addr_al is not initialized */ @@ -526,7 +527,6 @@ int dlfilter__do_filter_event(struct dlfilter *d,ASSIGN (period);ASSIGN (weight);ASSIGN (ins_lat); -ASSIGN (p_stage_cyc);ASSIGN (transaction);ASSIGN (insn_cnt);ASSIGN (cyc_cnt);tools/perf/util/event.h +0
-2@@ -390,8 +390,6 @@ extern unsigned int proc_map_timeout;#define PAGE_SIZE_NAME_LEN 32char *get_page_size_name (u64 size,char *str); -void arch_perf_parse_sample_weight (struct perf_sample *data,const __u64 *array,u64 type); -void arch_perf_synthesize_sample_weight (const struct perf_sample *data,__u64 *array,u64 type);const char *arch_perf_header_entry (const char *se_header);int arch_support_sort_key (const char *sort_key);tools/perf/util/evsel.c +12
-5@@ -2846,11 +2846,18 @@ perf_event__check_size(union perf_event *event, unsigned int sample_size)return 0 ; } -void __weak arch_perf_parse_sample_weight (struct perf_sample *data, -const __u64 *array, -u64 type__maybe_unused ) +static void perf_parse_sample_weight (struct perf_sample *data,const __u64 *array,u64 type) { - data->weight = *array; +union perf_sample_weight weight; + + weight.full = *array; +if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { + data->weight = weight.var1_dw; + data->ins_lat = weight.var2_w; + data->p_stage_cyc_or_retire_lat = weight.var3_w; + }else { + data->weight = weight.full; + } }u64 evsel__bitfield_swap_branch_flags (u64 value)@@ -3236,7 +3243,7 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event,if (type &PERF_SAMPLE_WEIGHT_TYPE ) {OVERFLOW_CHECK_u64 (array); -arch_perf_parse_sample_weight (data, array, type); +perf_parse_sample_weight (data, array, type); array++; }tools/perf/util/hist.c +2
-2@@ -829,7 +829,7 @@ __hists__add_entry(struct hists *hists, .period = sample->period, .weight1 = sample->weight, .weight2 = sample->ins_lat, - .weight3 = sample->p_stage_cyc, + .weight3 = sample->p_stage_cyc_or_retire_lat, .latency = al->latency, }, .parent = sym_parent,@@ -846,7 +846,7 @@ __hists__add_entry(struct hists *hists, .time = hist_time(sample->time), .weight = sample->weight, .ins_lat = sample->ins_lat, - .p_stage_cyc = sample->p_stage_cyc, + .p_stage_cyc_or_retire_lat = sample->p_stage_cyc_or_retire_lat, .simd_flags = sample->simd_flags, }, *he = hists__findnew_entry(hists, &entry, al, sample_self);tools/perf/util/hist.h +1
-1@@ -255,7 +255,7 @@ struct hist_entry {u64 code_page_size;u64 weight;u64 ins_lat; -u64 p_stage_cyc; +u64 p_stage_cyc_or_retire_lat;s32 socket;s32 cpu;int parallelism;tools/perf/util/intel-tpebs.c +2
-2@@ -202,8 +202,8 @@ static int process_sample_event(const struct perf_tool *tool __maybe_unused,* latency value will be used. Save the number of samples and the sum of * retire latency value for each event. */ - t->last = sample->retire_lat; -update_stats (&t->stats, sample->retire_lat); + t->last = sample->p_stage_cyc_or_retire_lat; +update_stats (&t->stats, sample->p_stage_cyc_or_retire_lat);mutex_unlock (tpebs_mtx_get ());return 0 ; }tools/perf/util/sample.h +1
-4@@ -104,10 +104,7 @@ struct perf_sample {u8 cpumode;u16 misc;u16 ins_lat; -union { -u16 p_stage_cyc; -u16 retire_lat; - }; +u16 p_stage_cyc_or_retire_lat;bool no_hw_idx;/* No hw_idx collected in branch_stack */ char insn[MAX_INSN ];void *raw_data;tools/perf/util/session.c +1
-1@@ -1094,7 +1094,7 @@ static void dump_sample(struct evsel *evsel, union perf_event *event,printf ("... weight: %" PRIu64"" , sample->weight);if (sample_type &PERF_SAMPLE_WEIGHT_STRUCT ) {printf (",0x%" PRIx16"" , sample->ins_lat); -printf (",0x%" PRIx16"" , sample->p_stage_cyc); +printf (",0x%" PRIx16"" , sample->p_stage_cyc_or_retire_lat); }printf ("\n" ); }tools/perf/util/sort.c +3
-3@@ -1879,21 +1879,21 @@ struct sort_entry sort_global_ins_lat = {static int64_t sort__p_stage_cyc_cmp (struct hist_entry *left,struct hist_entry *right) { -return left->p_stage_cyc - right->p_stage_cyc; +return left->p_stage_cyc_or_retire_lat - right->p_stage_cyc_or_retire_lat; }static int hist_entry__global_p_stage_cyc_snprintf (struct hist_entry *he,char *bf,size_t size,unsigned int width) {return repsep_snprintf (bf, size,"%-*u" , width, - he->p_stage_cyc * he->stat.nr_events); + he->p_stage_cyc_or_retire_lat * he->stat.nr_events); }static int hist_entry__p_stage_cyc_snprintf (struct hist_entry *he,char *bf,size_t size,unsigned int width) { -return repsep_snprintf (bf, size,"%-*u" , width, he->p_stage_cyc); +return repsep_snprintf (bf, size,"%-*u" , width, he->p_stage_cyc_or_retire_lat); }struct sort_entry sort_local_p_stage_cyc = {tools/perf/util/synthetic-events.c +8
-3@@ -1567,10 +1567,16 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,return result; } -void __weak arch_perf_synthesize_sample_weight (const struct perf_sample *data, +static void perf_synthesize_sample_weight (const struct perf_sample *data,__u64 *array,u64 type__maybe_unused ) { *array = data->weight; + +if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { + *array &=0xffffffff ; + *array |= ((u64 )data->ins_lat <<32 ); + *array |= ((u64 )data->p_stage_cyc_or_retire_lat <<48 ); + } } static __u64 *copy_read_group_values(__u64 *array, __u64 read_format,@@ -1730,7 +1736,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type, u64 read_fo }if (type &PERF_SAMPLE_WEIGHT_TYPE ) { -arch_perf_synthesize_sample_weight (sample, array, type); +perf_synthesize_sample_weight (sample, array, type); array++; } --2.49.0.1143.g0be31eac6b-goog
Namhyung Kim
maintainerRe: [PATCH v3 1/3] perf sample: Remove arch notion of sample parsing
Quoted 20 lines · Ian Rogers
Quoted 11 lines · Namhyung Kim
Quoted 6 lines · Ian Rogers
By definition arch sample parsing and synthesis will inhibit certain kinds of cross-platform record then analysis (report, script, etc.). Remove
arch_perf_parse_sample_weightandarch_perf_synthesize_sample_weightreplacing with a common implementation. Combineperf_samplep_stage_cycandretire_latto capture the differing uses regardless of compiled for architecture.Can you please do this without renaming? It can be a separate patch but I think we can just leave it.
It is not clear what the use of the union is. Presumably it is a tagged union but there is no tag as the union value to use is implied by either being built on
x86_64or powerpc. The change removes the notion of this code being built forx86_64or powerpc and so the union value to use isn't clear (e.g. should arm usep_stage_cycorretire_latfrom the union), hence combining to show that it could be one or the other. The code could be:
Quoted 8 lines
``` #ifdef __x86_64__ u16 p_stage_cyc; #elif defined(powerpc) u16 retire_lat; #endif ```but this isn't cross-platform.
Right, we probably don't want it.
Quoted 8 lines
The change in
hist.hof
```tools/perf/util/hist.h +1
-1@@ -255,7 +255,7 @@ struct hist_entry {u64 code_page_size;u64 weight;u64 ins_lat; -u64 p_stage_cyc; +u64 p_stage_cyc_or_retire_lat;
Quoted 6 lines
```
could be a follow up CL, but then we lose something of what the field is holding given the value is just a copy of that same named value inperf_sample. The code only inserts 34 lines and so the churn of doing that seemed worse than having the change in a single patch for clarity.
Assuming other archs can add something later, we won't rename the field again. So I can live with the ugly union fields. If we really want to rename it, I prefer calling it just 'weight3' and let the archs handle the display name only.
Thanks,
Namhyung
Ian Rogers
reviewerRe: [PATCH v3 1/3] perf sample: Remove arch notion of sample parsing
Quoted 45 lines · Namhyung Kim
Quoted 20 lines · Ian Rogers
Quoted 11 lines · Namhyung Kim
Quoted 6 lines · Ian Rogers
By definition arch sample parsing and synthesis will inhibit certain kinds of cross-platform record then analysis (report, script, etc.). Remove
arch_perf_parse_sample_weightandarch_perf_synthesize_sample_weightreplacing with a common implementation. Combineperf_samplep_stage_cycandretire_latto capture the differing uses regardless of compiled for architecture.Can you please do this without renaming? It can be a separate patch but I think we can just leave it.
It is not clear what the use of the union is. Presumably it is a tagged union but there is no tag as the union value to use is implied by either being built on
x86_64or powerpc. The change removes the notion of this code being built forx86_64or powerpc and so the union value to use isn't clear (e.g. should arm usep_stage_cycorretire_latfrom the union), hence combining to show that it could be one or the other. The code could be:Quoted 8 lines
``` #ifdef __x86_64__ u16 p_stage_cyc; #elif defined(powerpc) u16 retire_lat; #endif ```but this isn't cross-platform.
Right, we probably don't want it.
Quoted 11 lines
The change in
hist.hof
```tools/perf/util/hist.h +1
-1@@ -255,7 +255,7 @@ struct hist_entry {u64 code_page_size;u64 weight;u64 ins_lat; -u64 p_stage_cyc; +u64 p_stage_cyc_or_retire_lat;```
could be a follow up CL, but then we lose something of what the field is holding given the value is just a copy of that same named value in
Quoted 8 lines
Quoted 3 lines
perf_sample. The code only inserts 34 lines and so the churn of doing that seemed worse than having the change in a single patch for clarity.Assuming other archs can add something later, we won't rename the field again. So I can live with the ugly union fields. If we really want to rename it, I prefer calling it just 'weight3' and let the archs handle the display name only.
But that's my point (or in other words maybe you've missed my point) . Regardless of arch we should display p_stage_cyc if processing a perf.data file from a PowerPC as determined from the perf_env in the perf.data file, or retire_lat if processing a perf.data file from x86. The arch of the perf build is entirely irrelevant and calling the variable an opaque weight3 will require something that will need to be disambiguate it elsewhere in the code. The goal in variable names should be to be intention revealing, which I think p_stage_cyc_or_retire_lat is doing better than weight3, which is something of a regression from the existing but inaccurate p_stage_cyc.
Thanks,
Ian
Quoted 3 lines
Thanks,
Namhyung
Namhyung Kim
maintainerRe: [PATCH v3 1/3] perf sample: Remove arch notion of sample parsing
Quoted 55 lines · Ian Rogers
Quoted 45 lines · Namhyung Kim
Quoted 20 lines · Ian Rogers
Quoted 11 lines · Namhyung Kim
Quoted 6 lines · Ian Rogers
By definition arch sample parsing and synthesis will inhibit certain kinds of cross-platform record then analysis (report, script, etc.). Remove
arch_perf_parse_sample_weightandarch_perf_synthesize_sample_weightreplacing with a common implementation. Combineperf_samplep_stage_cycandretire_latto capture the differing uses regardless of compiled for architecture.Can you please do this without renaming? It can be a separate patch but I think we can just leave it.
It is not clear what the use of the union is. Presumably it is a tagged union but there is no tag as the union value to use is implied by either being built on
x86_64or powerpc. The change removes the notion of this code being built forx86_64or powerpc and so the union value to use isn't clear (e.g. should arm usep_stage_cycorretire_latfrom the union), hence combining to show that it could be one or the other. The code could be:Quoted 8 lines
``` #ifdef __x86_64__ u16 p_stage_cyc; #elif defined(powerpc) u16 retire_lat; #endif ```but this isn't cross-platform.
Right, we probably don't want it.
Quoted 11 lines
The change in
hist.hof
```tools/perf/util/hist.h +1
-1@@ -255,7 +255,7 @@ struct hist_entry {u64 code_page_size;u64 weight;u64 ins_lat; -u64 p_stage_cyc; +u64 p_stage_cyc_or_retire_lat;```
could be a follow up CL, but then we lose something of what the field is holding given the value is just a copy of that same named value inQuoted 8 lines
Quoted 3 lines
perf_sample. The code only inserts 34 lines and so the churn of doing that seemed worse than having the change in a single patch for clarity.Assuming other archs can add something later, we won't rename the field again. So I can live with the ugly union fields. If we really want to rename it, I prefer calling it just 'weight3' and let the archs handle the display name only.
Quoted 4 lines
But that's my point (or in other words maybe you've missed my point) . Regardless of arch we should display
p_stage_cycif processing a perf.data file from a PowerPC as determined from theperf_envin the perf.data file, orretire_latif processing a perf.data file from x86.
Agreed.
Quoted 7 lines
The arch of the perf build is entirely irrelevant and calling the variable an opaque weight3 will require something that will need to be disambiguate it elsewhere in the code. The goal in variable names should be to be intention revealing, which I think
p_stage_cyc_or_retire_latis doing better than weight3, which is something of a regression from the existing but inaccuratep_stage_cyc.
Yeah, but I worried if it would end up with 'p_stage_cyc_or_retire_lat_or_something_else' later.
Thanks,
Namhyung
Ian Rogers
reviewerRe: [PATCH v3 1/3] perf sample: Remove arch notion of sample parsing
Quoted 73 lines · Namhyung Kim
Quoted 55 lines · Ian Rogers
Quoted 45 lines · Namhyung Kim
Quoted 20 lines · Ian Rogers
Quoted 11 lines · Namhyung Kim
Quoted 6 lines · Ian Rogers
By definition arch sample parsing and synthesis will inhibit certain kinds of cross-platform record then analysis (report, script, etc.). Remove
arch_perf_parse_sample_weightandarch_perf_synthesize_sample_weightreplacing with a common implementation. Combineperf_samplep_stage_cycandretire_latto capture the differing uses regardless of compiled for architecture.Can you please do this without renaming? It can be a separate patch but I think we can just leave it.
It is not clear what the use of the union is. Presumably it is a tagged union but there is no tag as the union value to use is implied by either being built on
x86_64or powerpc. The change removes the notion of this code being built forx86_64or powerpc and so the union value to use isn't clear (e.g. should arm usep_stage_cycorretire_latfrom the union), hence combining to show that it could be one or the other. The code could be:Quoted 8 lines
``` #ifdef __x86_64__ u16 p_stage_cyc; #elif defined(powerpc) u16 retire_lat; #endif ```but this isn't cross-platform.
Right, we probably don't want it.
Quoted 11 lines
The change in
hist.hof
```tools/perf/util/hist.h +1
-1@@ -255,7 +255,7 @@ struct hist_entry {u64 code_page_size;u64 weight;u64 ins_lat; -u64 p_stage_cyc; +u64 p_stage_cyc_or_retire_lat;```
could be a follow up CL, but then we lose something of what the field is holding given the value is just a copy of that same named value inQuoted 8 lines
Quoted 3 lines
perf_sample. The code only inserts 34 lines and so the churn of doing that seemed worse than having the change in a single patch for clarity.Assuming other archs can add something later, we won't rename the field again. So I can live with the ugly union fields. If we really want to rename it, I prefer calling it just 'weight3' and let the archs handle the display name only.
Quoted 4 lines
But that's my point (or in other words maybe you've missed my point) . Regardless of arch we should display
p_stage_cycif processing a perf.data file from a PowerPC as determined from theperf_envin the perf.data file, orretire_latif processing a perf.data file from x86.Agreed.
Quoted 7 lines
The arch of the perf build is entirely irrelevant and calling the variable an opaque weight3 will require something that will need to be disambiguate it elsewhere in the code. The goal in variable names should be to be intention revealing, which I think
p_stage_cyc_or_retire_latis doing better than weight3, which is something of a regression from the existing but inaccuratep_stage_cyc.
Quoted 2 lines
Yeah, but I worried if it would end up with '
p_stage_cyc_or_retire_lat_or_something_else' later.
Perhaps it should be:
```
union {
u16 raw;
u16 p_stage_cyc;
u16 retire_lat;
} weight3;
```to try to best capture this. xyz.weight3.raw when the PowerPC or x86 use isn't known, etc. In the histogram code the u16 is a u64.
Thanks,
Ian
Quoted 3 lines
Thanks,
Namhyung
Namhyung Kim
maintainerRe: [PATCH v3 1/3] perf sample: Remove arch notion of sample parsing
Quoted 85 lines · Ian Rogers
Quoted 73 lines · Namhyung Kim
Quoted 55 lines · Ian Rogers
Quoted 45 lines · Namhyung Kim
Quoted 20 lines · Ian Rogers
Quoted 11 lines · Namhyung Kim
Quoted 6 lines · Ian Rogers
By definition arch sample parsing and synthesis will inhibit certain kinds of cross-platform record then analysis (report, script, etc.). Remove
arch_perf_parse_sample_weightandarch_perf_synthesize_sample_weightreplacing with a common implementation. Combineperf_samplep_stage_cycandretire_latto capture the differing uses regardless of compiled for architecture.Can you please do this without renaming? It can be a separate patch but I think we can just leave it.
It is not clear what the use of the union is. Presumably it is a tagged union but there is no tag as the union value to use is implied by either being built on
x86_64or powerpc. The change removes the notion of this code being built forx86_64or powerpc and so the union value to use isn't clear (e.g. should arm usep_stage_cycorretire_latfrom the union), hence combining to show that it could be one or the other. The code could be:Quoted 8 lines
``` #ifdef __x86_64__ u16 p_stage_cyc; #elif defined(powerpc) u16 retire_lat; #endif ```but this isn't cross-platform.
Right, we probably don't want it.
Quoted 11 lines
The change in
hist.hof
```(quoted diff) +1
-1@@ -255,7 +255,7 @@ struct hist_entry { u64 code_page_size; u64 weight; u64 ins_lat; - u64 p_stage_cyc; + u64 p_stage_cyc_or_retire_lat;```
could be a follow up CL, but then we lose something of what the field is holding given the value is just a copy of that same named value inQuoted 8 lines
Quoted 3 lines
perf_sample. The code only inserts 34 lines and so the churn of doing that seemed worse than having the change in a single patch for clarity.Assuming other archs can add something later, we won't rename the field again. So I can live with the ugly union fields. If we really want to rename it, I prefer calling it just 'weight3' and let the archs handle the display name only.
Quoted 4 lines
But that's my point (or in other words maybe you've missed my point) . Regardless of arch we should display
p_stage_cycif processing a perf.data file from a PowerPC as determined from theperf_envin the perf.data file, orretire_latif processing a perf.data file from x86.Agreed.
Quoted 7 lines
The arch of the perf build is entirely irrelevant and calling the variable an opaque weight3 will require something that will need to be disambiguate it elsewhere in the code. The goal in variable names should be to be intention revealing, which I think
p_stage_cyc_or_retire_latis doing better than weight3, which is something of a regression from the existing but inaccuratep_stage_cyc.Quoted 2 lines
Yeah, but I worried if it would end up with '
p_stage_cyc_or_retire_lat_or_something_else' later.Perhaps it should be:
``` union { u16 raw; u16 p_stage_cyc; u16 retire_lat; } weight3; ```
Quoted 2 lines
to try to best capture this.
xyz.weight3.rawwhen the PowerPC or x86 use isn't known, etc.
Looks better. But based on the offline discussion, it'd be better to just use 'u16 weight3'.
Quoted 1 line
In the histogram code the u16 is a u64.
Yep, because histogram entry aggregates sample weights.
Thanks,
Namhyung
Ian Rogers
reviewerPATCH v3 2/3 perf test: Move PERF_SAMPLE_WEIGHT_STRUCT parsing to common test
In reply to Ian Rogerstest__x86_sample_parsing is identical to test__sample_parsing except it explicitly tested PERF_SAMPLE_WEIGHT_STRUCT. Now the parsing code is common move the PERF_SAMPLE_WEIGHT_STRUCT to the common sample parsing test and remove the x86 version.
--- tools/perf/arch/x86/include/arch-tests.h | 1 - tools/perf/arch/x86/tests/Build | 1 - tools/perf/arch/x86/tests/arch-tests.c | 2 - tools/perf/arch/x86/tests/sample-parsing.c | 125 --------------------- tools/perf/tests/sample-parsing.c | 14 +++ 5 files changed, 14 insertions(+), 129 deletions(-) delete mode 100644 tools/perf/arch/x86/tests/sample-parsing.c
tools/perf/arch/x86/include/arch-tests.h +0 -1
@@ -12,7 +12,6 @@ int test__insn_x86(struct test_suite *test, int subtest);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__amd_ibs_via_core_pmu (struct test_suite *test,int subtest);int test__amd_ibs_period (struct test_suite *test,int subtest);int test__hybrid (struct test_suite *test,int subtest);
tools/perf/arch/x86/tests/Build +0 -1
@@ -2,7 +2,6 @@ perf-test-$(CONFIG_DWARF_UNWIND) += regs_load.o perf-test-$(CONFIG_DWARF_UNWIND) += dwarf-unwind.o perf-test-y += arch-tests.o -perf-test-y += sample-parsing.o perf-test-y += hybrid.o perf-test-$(CONFIG_AUXTRACE) += intel-pt-test.o ifeq ($(CONFIG_EXTRA_TESTS),y)
tools/perf/arch/x86/tests/arch-tests.c +0 -2
@@ -23,7 +23,6 @@ struct test_suite suite__intel_pt = {#if defined(__x86_64__)DEFINE_SUITE ("x86 bp modify" , bp_modify);#endif -DEFINE_SUITE ("x86 Sample parsing" , x86_sample_parsing);DEFINE_SUITE ("AMD IBS via core pmu" , amd_ibs_via_core_pmu);DEFINE_SUITE_EXCLUSIVE ("AMD IBS sample period" , amd_ibs_period);static struct test_case hybrid_tests[] = {
@@ -49,7 +48,6 @@ struct test_suite *arch_tests[] = {#if defined(__x86_64__) &suite__bp_modify, #endif - &suite__x86_sample_parsing, &suite__amd_ibs_via_core_pmu, &suite__amd_ibs_period, &suite__hybrid,
tools/perf/arch/x86/tests/sample-parsing.c deleted +0 -125
125 lines not on this page · raw message
tools/perf/tests/sample-parsing.c +14 -0
@@ -152,6 +152,12 @@ static bool samples_same(struct perf_sample *s1,if (type &PERF_SAMPLE_WEIGHT )COMP (weight); +if (type &PERF_SAMPLE_WEIGHT_STRUCT ) { +COMP (weight); +COMP (ins_lat); +COMP (p_stage_cyc_or_retire_lat); + } +if (type &PERF_SAMPLE_DATA_SRC )COMP (data_src);
@@ -269,6 +275,8 @@ static int do_test(u64 sample_type, u64 sample_regs, u64 read_format) .cgroup = 114, .data_page_size = 115, .code_page_size = 116, + .ins_lat = 117, + .p_stage_cyc_or_retire_lat = 118, .aux_sample = { .size = sizeof(aux_data), .data = (void *)aux_data,
@@ -439,6 +447,12 @@ static int test__sample_parsing(struct test_suite *test __maybe_unused, int subtif (err)return err; } + sample_type = (PERF_SAMPLE_MAX -1 ) & ~PERF_SAMPLE_WEIGHT_STRUCT ; +for (i =0 ; i <ARRAY_SIZE (rf); i++) { + err =do_test (sample_type, sample_regs, rf[i]); +if (err) +return err; + }return 0 ; }
Signature · 1 line
2.49.0.1143.g0be31eac6b-goog
Ian Rogers
reviewerPATCH v3 3/3 perf sort: Use perf_env to set arch sort keys and header
In reply to Ian RogersPreviously arch_support_sort_key and arch_perf_header_entry used a weak symbol to compile as appropriate for x86 and powerpc. A limitation to this is that the handling of a data file could vary in cross-platform development. Change to using the perf_env of the current session to determine the architecture kind and set the sort key and header entries as appropriate.
--- tools/perf/arch/powerpc/util/Build | 1 - tools/perf/arch/powerpc/util/event.c | 34 ---------------- tools/perf/arch/x86/util/event.c | 19 --------- tools/perf/builtin-annotate.c | 2 +- tools/perf/builtin-c2c.c | 53 +++++++++++++----------- tools/perf/builtin-diff.c | 2 +- tools/perf/builtin-report.c | 2 +- tools/perf/builtin-top.c | 16 ++++---- tools/perf/tests/hists_cumulate.c | 8 ++-- tools/perf/tests/hists_filter.c | 8 ++-- tools/perf/tests/hists_link.c | 8 ++-- tools/perf/tests/hists_output.c | 10 ++--- tools/perf/util/event.h | 3 -- tools/perf/util/sort.c | 61 ++++++++++++++++++++-------- tools/perf/util/sort.h | 5 ++- 15 files changed, 104 insertions(+), 128 deletions(-) delete mode 100644 tools/perf/arch/powerpc/util/event.c
tools/perf/arch/powerpc/util/Build +0 -1
@@ -5,7 +5,6 @@ perf-util-y += mem-events.o perf-util-y += pmu.o perf-util-y += sym-handling.o perf-util-y += evsel.o -perf-util-y += event.o perf-util-$(CONFIG_LIBDW) += skip-callchain-idx.o
tools/perf/arch/powerpc/util/event.c deleted +0 -34
@@ -1,34 +0,0 @@ -// SPDX-License-Identifier: GPL-2.0 -#include <linux/types.h> -#include <linux/string.h> -#include <linux/zalloc.h> - -#include "../../../util/event.h" -#include "../../../util/synthetic-events.h" -#include "../../../util/machine.h" -#include "../../../util/tool.h" -#include "../../../util/map.h" -#include "../../../util/debug.h" -#include "../../../util/sample.h" - -const char *arch_perf_header_entry (const char *se_header) -{ -if (!strcmp (se_header,"Local INSTR Latency" )) -return "Finish Cyc" ; -else if (!strcmp (se_header,"INSTR Latency" )) -return "Global Finish_cyc" ; -else if (!strcmp (se_header,"Local Pipeline Stage Cycle" )) -return "Dispatch Cyc" ; -else if (!strcmp (se_header,"Pipeline Stage Cycle" )) -return "Global Dispatch_cyc" ; -return se_header; -} - -int arch_support_sort_key (const char *sort_key) -{ -if (!strcmp (sort_key,"p_stage_cyc" )) -return 1 ; -if (!strcmp (sort_key,"local_p_stage_cyc" )) -return 1 ; -return 0 ; -}
tools/perf/arch/x86/util/event.c +0 -19
@@ -91,22 +91,3 @@ int perf_event__synthesize_extra_kmaps(const struct perf_tool *tool, }#endif - -const char *arch_perf_header_entry (const char *se_header) -{ -if (!strcmp (se_header,"Local Pipeline Stage Cycle" )) -return "Local Retire Latency" ; -else if (!strcmp (se_header,"Pipeline Stage Cycle" )) -return "Retire Latency" ; - -return se_header; -} - -int arch_support_sort_key (const char *sort_key) -{ -if (!strcmp (sort_key,"p_stage_cyc" )) -return 1 ; -if (!strcmp (sort_key,"local_p_stage_cyc" )) -return 1 ; -return 0 ; -}
tools/perf/builtin-annotate.c +1 -1
@@ -947,7 +947,7 @@ int cmd_annotate(int argc, const char **argv) annotate_opts.show_br_cntr = true; } -if (setup_sorting (NULL ) <0 ) +if (setup_sorting (/*evlist=*/ NULL , &annotate.session->header.env) <0 )usage_with_options (annotate_usage, options); ret =__cmd_annotate (&annotate);
tools/perf/builtin-c2c.c +30 -23
183 lines not on this page · raw message
tools/perf/builtin-diff.c +1 -1
@@ -2003,7 +2003,7 @@ int cmd_diff(int argc, const char **argv) sort__mode =SORT_MODE__DIFF ; } -if (setup_sorting (NULL ) <0 ) +if (setup_sorting (/*evlist=*/ NULL , &data__files[0 ].session->header.env) <0 )usage_with_options (diff_usage, options);setup_pager ();
tools/perf/builtin-report.c +1 -1
@@ -1776,7 +1776,7 @@ int cmd_report(int argc, const char **argv) }if ((last_key !=K_SWITCH_INPUT_DATA && last_key !=K_RELOAD ) && - (setup_sorting (session->evlist) <0 )) { + (setup_sorting (session->evlist, &session->header.env) <0 )) {if (sort_order)parse_options_usage (report_usage, options,"s" ,1 );if (field_order)
tools/perf/builtin-top.c +8 -8
@@ -1747,7 +1747,14 @@ int cmd_top(int argc, const char **argv)setup_browser (false); -if (setup_sorting (top.evlist) <0 ) { + top.session =perf_session__new (/*data=*/ NULL ,/*tool=*/ NULL ); +if (IS_ERR (top.session)) { + status =PTR_ERR (top.session); + top.session =NULL ; + goto out_delete_evlist; + } + +if (setup_sorting (top.evlist, &top.session->header.env) <0 ) {if (sort_order)parse_options_usage (top_usage, options,"s" ,1 );if (field_order)
@@ -1820,13 +1827,6 @@ int cmd_top(int argc, const char **argv)signal (SIGWINCH , winch_sig); } - top.session =perf_session__new (NULL ,NULL ); -if (IS_ERR (top.session)) { - status =PTR_ERR (top.session); - top.session =NULL ; - goto out_delete_evlist; - } -if (!evlist__needs_bpf_sb_event (top.evlist)) top.record_opts.no_bpf_event = true;
tools/perf/tests/hists_cumulate.c +4 -4
@@ -295,7 +295,7 @@ static int test1(struct evsel *evsel, struct machine *machine) symbol_conf.cumulate_callchain = false;evsel__reset_sample_bit (evsel,CALLCHAIN ); -setup_sorting (NULL ); +setup_sorting (/*evlist=*/ NULL , machine->env);callchain_register_param (&callchain_param); err =add_hist_entries (hists, machine);
@@ -442,7 +442,7 @@ static int test2(struct evsel *evsel, struct machine *machine) symbol_conf.cumulate_callchain = false;evsel__set_sample_bit (evsel,CALLCHAIN ); -setup_sorting (NULL ); +setup_sorting (/*evlist=*/ NULL , machine->env);callchain_register_param (&callchain_param); err =add_hist_entries (hists, machine);
@@ -500,7 +500,7 @@ static int test3(struct evsel *evsel, struct machine *machine) symbol_conf.cumulate_callchain = true;evsel__reset_sample_bit (evsel,CALLCHAIN ); -setup_sorting (NULL ); +setup_sorting (/*evlist=*/ NULL , machine->env);callchain_register_param (&callchain_param); err =add_hist_entries (hists, machine);
@@ -684,7 +684,7 @@ static int test4(struct evsel *evsel, struct machine *machine) symbol_conf.cumulate_callchain = true;evsel__set_sample_bit (evsel,CALLCHAIN ); -setup_sorting (NULL ); +setup_sorting (/*evlist=*/ NULL , machine->env); callchain_param = callchain_param_default;callchain_register_param (&callchain_param);
tools/perf/tests/hists_filter.c +4 -4
@@ -131,10 +131,6 @@ static int test__hists_filter(struct test_suite *test __maybe_unused, int subtes goto out; err =TEST_FAIL ; -/* default sort order (comm,dso,sym) will be used */ -if (setup_sorting (NULL ) <0 ) - goto out; -machines__init (&machines);/* setup threads/dso/map/symbols also */
@@ -145,6 +141,10 @@ static int test__hists_filter(struct test_suite *test __maybe_unused, int subtesif (verbose >1 )machine__fprintf (machine, stderr); +/* default sort order (comm,dso,sym) will be used */ +if (setup_sorting (evlist, machine->env) <0 ) + goto out; +/* process sample events */ err =add_hist_entries (evlist, machine);if (err <0 )
tools/perf/tests/hists_link.c +4 -4
@@ -303,10 +303,6 @@ static int test__hists_link(struct test_suite *test __maybe_unused, int subtest goto out; err =TEST_FAIL ; -/* default sort order (comm,dso,sym) will be used */ -if (setup_sorting (NULL ) <0 ) - goto out; -machines__init (&machines);/* setup threads/dso/map/symbols also */
@@ -317,6 +313,10 @@ static int test__hists_link(struct test_suite *test __maybe_unused, int subtestif (verbose >1 )machine__fprintf (machine, stderr); +/* default sort order (comm,dso,sym) will be used */ +if (setup_sorting (evlist, machine->env) <0 ) + goto out; +/* process sample events */ err =add_hist_entries (evlist, machine);if (err <0 )
tools/perf/tests/hists_output.c +5 -5
@@ -146,7 +146,7 @@ static int test1(struct evsel *evsel, struct machine *machine) field_order =NULL ; sort_order =NULL ;/* equivalent to sort_order = "comm,dso,sym" */ -setup_sorting (NULL ); +setup_sorting (/*evlist=*/ NULL , machine->env);/* * expected output:
@@ -248,7 +248,7 @@ static int test2(struct evsel *evsel, struct machine *machine) field_order ="overhead,cpu" ; sort_order ="pid" ; -setup_sorting (NULL ); +setup_sorting (/*evlist=*/ NULL , machine->env);/* * expected output:
@@ -304,7 +304,7 @@ static int test3(struct evsel *evsel, struct machine *machine) field_order ="comm,overhead,dso" ; sort_order =NULL ; -setup_sorting (NULL ); +setup_sorting (/*evlist=*/ NULL , machine->env);/* * expected output:
@@ -378,7 +378,7 @@ static int test4(struct evsel *evsel, struct machine *machine) field_order ="dso,sym,comm,overhead,dso" ; sort_order ="sym" ; -setup_sorting (NULL ); +setup_sorting (/*evlist=*/ NULL , machine->env);/* * expected output:
@@ -480,7 +480,7 @@ static int test5(struct evsel *evsel, struct machine *machine) field_order ="cpu,pid,comm,dso,sym" ; sort_order ="dso,pid" ; -setup_sorting (NULL ); +setup_sorting (/*evlist=*/ NULL , machine->env);/* * expected output:
tools/perf/util/event.h +0 -3
@@ -390,9 +390,6 @@ extern unsigned int proc_map_timeout;#define PAGE_SIZE_NAME_LEN 32char *get_page_size_name (u64 size,char *str); -const char *arch_perf_header_entry (const char *se_header); -int arch_support_sort_key (const char *sort_key); -static inline bool perf_event_header__cpumode_is_guest (u8 cpumode) { return cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
tools/perf/util/sort.c +43 -18
147 lines not on this page · raw message
tools/perf/util/sort.h +3 -2
@@ -6,6 +6,7 @@#include "hist.h" struct option ; +struct perf_env ;extern regex_t parent_regex;extern const char *sort_order;
@@ -130,7 +131,7 @@ extern struct sort_entry sort_thread;struct evlist ;struct tep_handle ; -int setup_sorting (struct evlist *evlist); +int setup_sorting (struct evlist *evlist,struct perf_env *env);int setup_output_field (void );void reset_output_field (void );void sort__setup_elide (FILE *fp);
@@ -145,7 +146,7 @@ bool is_strict_order(const char *order);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, +struct evlist *evlist,struct perf_env *env,int level);int output_field_add (struct perf_hpp_list *list,const char *tok,int *level);int64_t
Signature · 1 line
2.49.0.1143.g0be31eac6b-goog
Ian Rogers
reviewerQuoted 9 lines · Ian Rogers
The arch directory is a barrier to cross-platform development as files and behaviors within it are inherently platform specific. Sample parsing should be generic but the
PERF_SAMPLE_WEIGHT_STRUCThandling was only present if building for x86 or powerpc. The sort key and headers should be specific to the session that is being executed and not to the machine perf is being run upon. These patches clean this and associated code up.
Quoted 3 lines
v3: Code reorganizations suggested by Kan Liang.
v2: Avoid changes toinclude/perf/perf_dlfilter.has suggested by
Adrian Hunter.
Ping. Thanks,
Ian
Quoted 41 lines
Ian Rogers (3):
perf sample: Remove arch notion of sample parsing perf test: MovePERF_SAMPLE_WEIGHT_STRUCTparsing to common test perf sort: Useperf_envto set arch sort keys and headertools/perf/arch/powerpc/util/Build | 1 - tools/perf/arch/powerpc/util/event.c | 60 ---------- tools/perf/arch/x86/include/arch-tests.h | 1 - tools/perf/arch/x86/tests/Build | 1 - tools/perf/arch/x86/tests/arch-tests.c | 2 - tools/perf/arch/x86/tests/sample-parsing.c | 125 --------------------- tools/perf/arch/x86/util/event.c | 46 -------- tools/perf/builtin-annotate.c | 2 +- tools/perf/builtin-c2c.c | 53 +++++---- tools/perf/builtin-diff.c | 2 +- tools/perf/builtin-report.c | 2 +- tools/perf/builtin-script.c | 2 +- tools/perf/builtin-top.c | 16 +-- tools/perf/tests/hists_cumulate.c | 8 +- tools/perf/tests/hists_filter.c | 8 +- tools/perf/tests/hists_link.c | 8 +- tools/perf/tests/hists_output.c | 10 +- tools/perf/tests/sample-parsing.c | 14 +++ tools/perf/util/dlfilter.c | 2 +- tools/perf/util/event.h | 5 - tools/perf/util/evsel.c | 17 ++- tools/perf/util/hist.c | 4 +- tools/perf/util/hist.h | 2 +- tools/perf/util/intel-tpebs.c | 4 +- tools/perf/util/sample.h | 5 +- tools/perf/util/session.c | 2 +- tools/perf/util/sort.c | 67 +++++++---- tools/perf/util/sort.h | 5 +- tools/perf/util/synthetic-events.c | 10 +- 29 files changed, 150 insertions(+), 334 deletions(-) delete mode 100644 tools/perf/arch/powerpc/util/event.c delete mode 100644 tools/perf/arch/x86/tests/sample-parsing.c--
2.49.0.1143.g0be31eac6b-goog