perf report: Fix wrong LBR block sorting
PATCH 1/1 perf report: Fix wrong LBR block sorting
When '--total-cycles' is specified, it supports sorting for all blocks by 'Sampled Cycles%'. This is useful to concentrate on the globally hottest blocks.
'Sampled Cycles%' - block sampled cycles aggregation / total sampled cycles
But in current code, it doesn't use the cycles aggregation. Part of 'cycles' counting is possibly dropped for some overlap jumps. But for identifying the hot block, we always need the full cycles.
# perf record -b ./triad_loop # perf report --total-cycles --stdio
Before:
#
# Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Program Block Range] Shared Object
# ............... .............. ........... .......... ...................................................................... ....................
#
0.81% 793 4.32% 793 [setup-vdso.h:34 -> setup-vdso.h:40] ld-2.27.so
0.49% 480 0.87% 160 [native_write_msr+0 -> native_write_msr+16] [kernel.kallsyms]
0.48% 476 0.52% 95 [native_read_msr+0 -> native_read_msr+29] [kernel.kallsyms]
0.31% 303 1.65% 303 [nmi_restore+0 -> nmi_restore+37] [kernel.kallsyms]
0.26% 255 1.39% 255 [nohz_balance_exit_idle+75 -> nohz_balance_exit_idle+162] [kernel.kallsyms]
0.24% 234 1.28% 234 [end_repeat_nmi+67 -> end_repeat_nmi+83] [kernel.kallsyms]
0.23% 227 1.24% 227 [__irqentry_text_end+96 -> __irqentry_text_end+126] [kernel.kallsyms]
0.20% 194 1.06% 194 [native_set_debugreg+52 -> native_set_debugreg+56] [kernel.kallsyms]
0.11% 106 0.14% 26 [native_sched_clock+0 -> native_sched_clock+98] [kernel.kallsyms]
0.10% 97 0.53% 97 [trigger_load_balance+0 -> trigger_load_balance+67] [kernel.kallsyms]
0.09% 85 0.46% 85 [get-dynamic-info.h:102 -> get-dynamic-info.h:111] ld-2.27.so
...
0.00% 92.7K 0.02% 4 [triad_loop.c:64 -> triad_loop.c:65] triad_loopThe hottest block '[triad_loop.c:64 -> triad_loop.c:65]' is not at the top of output.
After:
# Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Program Block Range] Shared Object
# ............... .............. ........... .......... ...................................................................... ....................
#
94.35% 92.7K 0.02% 4 [triad_loop.c:64 -> triad_loop.c:65] triad_loop
0.81% 793 4.32% 793 [setup-vdso.h:34 -> setup-vdso.h:40] ld-2.27.so
0.49% 480 0.87% 160 [native_write_msr+0 -> native_write_msr+16] [kernel.kallsyms]
0.48% 476 0.52% 95 [native_read_msr+0 -> native_read_msr+29] [kernel.kallsyms]
0.31% 303 1.65% 303 [nmi_restore+0 -> nmi_restore+37] [kernel.kallsyms]
0.26% 255 1.39% 255 [nohz_balance_exit_idle+75 -> nohz_balance_exit_idle+162] [kernel.kallsyms]
0.24% 234 1.28% 234 [end_repeat_nmi+67 -> end_repeat_nmi+83] [kernel.kallsyms]
0.23% 227 1.24% 227 [__irqentry_text_end+96 -> __irqentry_text_end+126] [kernel.kallsyms]
0.20% 194 1.06% 194 [native_set_debugreg+52 -> native_set_debugreg+56] [kernel.kallsyms]
0.11% 106 0.14% 26 [native_sched_clock+0 -> native_sched_clock+98] [kernel.kallsyms]
0.10% 97 0.53% 97 [trigger_load_balance+0 -> trigger_load_balance+67] [kernel.kallsyms]
0.09% 85 0.46% 85 [get-dynamic-info.h:102 -> get-dynamic-info.h:111] ld-2.27.so
0.08% 82 0.06% 11 [intel_pmu_drain_pebs_nhm+580 -> intel_pmu_drain_pebs_nhm+627] [kernel.kallsyms]
0.08% 77 0.42% 77 [lru_add_drain_cpu+0 -> lru_add_drain_cpu+133] [kernel.kallsyms]
0.08% 74 0.10% 18 [handle_pmi_common+271 -> handle_pmi_common+310] [kernel.kallsyms]
0.08% 74 0.40% 74 [get-dynamic-info.h:131 -> get-dynamic-info.h:157] ld-2.27.so
0.07% 69 0.09% 17 [intel_pmu_drain_pebs_nhm+432 -> intel_pmu_drain_pebs_nhm+468] [kernel.kallsyms]Now the hottest block is reported at the top of output.
Fixes b65a7d372b1a ("perf hist: Support block formats with compare/sort/display")
Signed-off-by Jin Yao <[email protected]>
--- tools/perf/util/block-info.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
tools/perf/util/block-info.c +3 -3
@@ -201,7 +201,7 @@ static int block_total_cycles_pct_entry(struct perf_hpp_fmt *fmt,double ratio =0.0 ;if (block_fmt->total_cycles) - ratio = (double )bi->cycles / (double )block_fmt->total_cycles; + ratio = (double )bi->cycles_aggr / (double )block_fmt->total_cycles;return color_pct (hpp, block_fmt->width,100.0 * ratio); }
@@ -216,9 +216,9 @@ static int64_t block_total_cycles_pct_sort(struct perf_hpp_fmt *fmt,double l, r;if (block_fmt->total_cycles) { - l = ((double )bi_l->cycles / + l = ((double )bi_l->cycles_aggr / (double )block_fmt->total_cycles) *100000.0 ; - r = ((double )bi_r->cycles / + r = ((double )bi_r->cycles_aggr / (double )block_fmt->total_cycles) *100000.0 ;return (int64_t )l - (int64_t )r; }
Signature · 1 line
2.17.1
Quoted 4 lines
Now the hottest block is reported at the top of output.
Fixes b65a7d372b1a ("perf hist: Support block formats with compare/sort/display")Signed-off-by Jin Yao <[email protected]>
Reviewed-by Andi Kleen <[email protected]>
-Andi
Em Wed, Apr 07, 2021 at 06:49:57AM -0700, Andi Kleen escreveu:
Quoted 7 lines
Quoted 4 lines
Now the hottest block is reported at the top of output.
Fixes b65a7d372b1a ("perf hist: Support block formats with compare/sort/display")Signed-off-by Jin Yao <[email protected]>Reviewed-by Andi Kleen <[email protected]>
Thanks, applied.
- Arnaldo