perf stat: Use proper cpu for shadow stats
Namhyung Kim
maintainerPATCH v2 1/2 perf stat: Use proper cpu for shadow stats
Currently perf stat shows some metrics (like IPC) for defined events. But when no aggregation mode is used (-A option), it shows incorrect values since it used a value from a different cpu.
Before:
$ perf stat -aA -e cycles,instructions sleep 1
Performance counter stats for 'system wide':
CPU0 116,057,380 cycles CPU1 86,084,722 cycles CPU2 99,423,125 cycles CPU3 98,272,994 cycles CPU0 53,369,217 instructions # 0.46 insn per cycle CPU1 33,378,058 instructions # 0.29 insn per cycle CPU2 58,150,086 instructions # 0.50 insn per cycle CPU3 40,029,703 instructions # 0.34 insn per cycle
1.001816971 seconds time elapsed
So the IPC for CPU1 should be 0.38 (= 33,378,058 / 86,084,722) but it was 0.29 (= 33,378,058 / 116,057,380) and so on.
After:
$ perf stat -aA -e cycles,instructions sleep 1
Performance counter stats for 'system wide':
CPU0 109,621,384 cycles CPU1 159,026,454 cycles CPU2 99,460,366 cycles CPU3 124,144,142 cycles CPU0 44,396,706 instructions # 0.41 insn per cycle CPU1 120,195,425 instructions # 0.76 insn per cycle CPU2 44,763,978 instructions # 0.45 insn per cycle CPU3 69,049,079 instructions # 0.56 insn per cycle
1.001910444 seconds time elapsed
--- tools/perf/util/stat-display.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-)
tools/perf/util/stat-display.c +1 -4
@@ -324,13 +324,10 @@ static int first_shadow_cpu(struct perf_stat_config *config,struct evlist *evlist = evsel->evlist;int i; -if (!config->aggr_get_id) -return 0 ; -if (config->aggr_mode ==AGGR_NONE )return id; -if (config->aggr_mode ==AGGR_GLOBAL ) +if (!config->aggr_get_id)return 0 ;for (i =0 ; i <evsel__nr_cpus (evsel); i++) {
Signature · 1 line
2.29.2.454.gaff20da3a2-goog
Namhyung Kim
maintainerPATCH v2 2/2 perf test: Add shadow stat test
It calculates IPC from the cycles and instruction counts and compares it with the shadow stat for both global aggregation (default) and no aggregation mode.
$ perf stat -a -A -e cycles,instructions sleep 1
Performance counter stats for 'system wide':
CPU0 39,580,880 cycles CPU1 45,426,945 cycles CPU2 31,151,685 cycles CPU3 55,167,421 cycles CPU0 17,073,564 instructions # 0.43 insn per cycle CPU1 34,955,764 instructions # 0.77 insn per cycle CPU2 15,688,459 instructions # 0.50 insn per cycle CPU3 34,699,217 instructions # 0.63 insn per cycle
1.003275495 seconds time elapsed
In this example, the 'insn per cycle' should be matched to the number for each cpu. For CPU2, 0.50 = 15,688,459 / 31,151,685 .
--- tools/perf/tests/shell/stat+shadow_stat.sh | 80 ++++++++++++++++++++++ 1 file changed, 80 insertions(+) create mode 100755 tools/perf/tests/shell/stat+shadow_stat.sh
tools/perf/tests/shell/stat+shadow_stat.sh new file +80 -0
@@ -0,0 +1,80 @@ +#!/bin/sh +# perf stat metrics (shadow stat) test +# SPDX-License-Identifier: GPL-2.0 + +set -e + +# skip if system-wide mode is forbidden +perf stat -a true > /dev/null 2>&1 || exit 2 + +test_global_aggr() +{ + local cyc + + perf stat -a --no-big-num -e cycles,instructions sleep 1 2>&1 | \ + grep -e cycles -e instructions | \ + while read num evt hash ipc rest + do + # skip not counted events + if [[ $num == "<not" ]]; then + continue + fi + + # save cycles count + if [[ $evt == "cycles" ]]; then + cyc=$num + continue + fi + + # skip if no cycles + if [[ -z $cyc ]]; then + continue + fi + + # use printf for rounding and a leading zero + local res=`printf "%.2f" $(echo "scale=6; $num / $cyc" | bc -q)` + if [[ $ipc != $res ]]; then + echo "IPC is different: $res != $ipc ($num / $cyc)" + exit 1 + fi + done +} + +test_no_aggr() +{ + declare -A results + + perf stat -a -A --no-big-num -e cycles,instructions sleep 1 2>&1 | \ + grep ^CPU | \ + while read cpu num evt hash ipc rest + do + # skip not counted events + if [[ $num == "<not" ]]; then + continue + fi + + # save cycles count + if [[ $evt == "cycles" ]]; then + results[$cpu]=$num + continue + fi + + # skip if no cycles + local cyc=${results[$cpu]} + if [[ -z $cyc ]]; then + continue + fi + + # use printf for rounding and a leading zero + local res=`printf "%.2f" $(echo "scale=6; $num / $cyc" | bc -q)` + if [[ $ipc != $res ]]; then + echo "IPC is different for $cpu: $res != $ipc ($num / $cyc)" + exit 1 + fi + done +} + +test_global_aggr +test_no_aggr + +exit 0
Signature · 1 line
2.29.2.454.gaff20da3a2-goog
Arnaldo Carvalho de Melo
maintainerRe: [PATCH v2 2/2] perf test: Add shadow stat test
Em Fri, Nov 27, 2020 at 01:14:04PM +0900, Namhyung Kim escreveu:
Quoted 17 lines
It calculates IPC from the cycles and instruction counts and compares it with the shadow stat for both global aggregation (default) and no aggregation mode.
$ perf stat -a -A -e cycles,instructions sleep 1Performance counter stats for 'system wide':
CPU0 39,580,880 cycles CPU1 45,426,945 cycles CPU2 31,151,685 cycles CPU3 55,167,421 cycles CPU0 17,073,564 instructions # 0.43 insn per cycle CPU1 34,955,764 instructions # 0.77 insn per cycle CPU2 15,688,459 instructions # 0.50 insn per cycle CPU3 34,699,217 instructions # 0.63 insn per cycle
Quoted 4 lines
1.003275495 seconds time elapsedIn this example, the 'insn per cycle' should be matched to the number for each cpu. For CPU2, 0.50 = 15,688,459 / 31,151,685 .
Merged torvalds/master into perf/core and applied this patch,
Thanks,
- Arnaldo
Quoted 95 lines
Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/tests/shell/stat+shadow_stat.sh | 80 ++++++++++++++++++++++ 1 file changed, 80 insertions(+) create mode 100755 tools/perf/tests/shell/stat+shadow_stat.shtools/perf/tests/shell/stat+shadow_stat.sh new file +80
-0@@ -0,0 +1,80 @@ +#!/bin/sh +# perf stat metrics (shadow stat) test +# SPDX-License-Identifier: GPL-2.0 + +set -e + +# skip if system-wide mode is forbidden +perf stat -a true > /dev/null 2>&1 || exit 2 + +test_global_aggr() +{ + local cyc + + perf stat -a --no-big-num -e cycles,instructions sleep 1 2>&1 | \ + grep -e cycles -e instructions | \ + while read num evt hash ipc rest + do + # skip not counted events + if [[ $num == "<not" ]]; then + continue + fi + + # save cycles count + if [[ $evt == "cycles" ]]; then + cyc=$num + continue + fi + + # skip if no cycles + if [[ -z $cyc ]]; then + continue + fi + + # use printf for rounding and a leading zero + local res=`printf "%.2f" $(echo "scale=6; $num / $cyc" | bc -q)` + if [[ $ipc != $res ]]; then + echo "IPC is different: $res != $ipc ($num / $cyc)" + exit 1 + fi + done +} + +test_no_aggr() +{ + declare -A results + + perf stat -a -A --no-big-num -e cycles,instructions sleep 1 2>&1 | \ + grep ^CPU | \ + while read cpu num evt hash ipc rest + do + # skip not counted events + if [[ $num == "<not" ]]; then + continue + fi + + # save cycles count + if [[ $evt == "cycles" ]]; then + results[$cpu]=$num + continue + fi + + # skip if no cycles + local cyc=${results[$cpu]} + if [[ -z $cyc ]]; then + continue + fi + + # use printf for rounding and a leading zero + local res=`printf "%.2f" $(echo "scale=6; $num / $cyc" | bc -q)` + if [[ $ipc != $res ]]; then + echo "IPC is different for $cpu: $res != $ipc ($num / $cyc)" + exit 1 + fi + done +} + +test_global_aggr +test_no_aggr + +exit 0Signature · 1 line
2.29.2.454.gaff20da3a2-goog
Signature · 1 line
- Arnaldo
Arnaldo Carvalho de Melo
maintainerEm Fri, Nov 27, 2020 at 01:14:03PM +0900, Namhyung Kim escreveu:
Quoted 39 lines
Currently perf stat shows some metrics (like IPC) for defined events. But when no aggregation mode is used (-A option), it shows incorrect values since it used a value from a different cpu.
Before:
$ perf stat -aA -e cycles,instructions sleep 1Performance counter stats for 'system wide':
CPU0 116,057,380 cycles CPU1 86,084,722 cycles CPU2 99,423,125 cycles CPU3 98,272,994 cycles CPU0 53,369,217 instructions # 0.46 insn per cycle CPU1 33,378,058 instructions # 0.29 insn per cycle CPU2 58,150,086 instructions # 0.50 insn per cycle CPU3 40,029,703 instructions # 0.34 insn per cycle1.001816971 seconds time elapsedSo the IPC for CPU1 should be 0.38 (= 33,378,058 / 86,084,722) but it was 0.29 (= 33,378,058 / 116,057,380) and so on.
After:
$ perf stat -aA -e cycles,instructions sleep 1Performance counter stats for 'system wide':
CPU0 109,621,384 cycles CPU1 159,026,454 cycles CPU2 99,460,366 cycles CPU3 124,144,142 cycles CPU0 44,396,706 instructions # 0.41 insn per cycle CPU1 120,195,425 instructions # 0.76 insn per cycle CPU2 44,763,978 instructions # 0.45 insn per cycle CPU3 69,049,079 instructions # 0.56 insn per cycle
Quoted 1 line
1.001910444 seconds time elapsed
Thanks, applied, the new 'perf test' entry in 2/2 will be merged into perf/core, as it isn't purely a fix,
- Arnaldo
Quoted 31 lines
Reported-by Sam Xi <[email protected]>Fixes 44d49a600259 ("perf stat: Support metrics in --per-core/socket mode")Reviewed-by Andi Kleen <[email protected]>Acked-by Jiri Olsa <[email protected]>Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/util/stat-display.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-)tools/perf/util/stat-display.c +1
-4@@ -324,13 +324,10 @@ static int first_shadow_cpu(struct perf_stat_config *config,struct evlist *evlist = evsel->evlist;int i; -if (!config->aggr_get_id) -return 0 ; -if (config->aggr_mode ==AGGR_NONE )return id; -if (config->aggr_mode ==AGGR_GLOBAL ) +if (!config->aggr_get_id)return 0 ;for (i =0 ; i <evsel__nr_cpus (evsel); i++) {Signature · 1 line
2.29.2.454.gaff20da3a2-goog
Signature · 1 line
- Arnaldo