Re: [PATCH v4] perf stat: Show percore counts in per CPU output

From: Jin, Yao
Date: Mon Feb 17 2020 - 20:02:58 EST

On 2/17/2020 7:06 PM, Jiri Olsa wrote:
On Mon, Feb 17, 2020 at 09:22:57AM +0800, Jin, Yao wrote:

On 2/17/2020 6:54 AM, Jiri Olsa wrote:
On Fri, Feb 14, 2020 at 04:04:52PM +0800, Jin Yao wrote:


CPU1 1,009,312 cpu/event=cpu-cycles,percore/
CPU2 2,784,072 cpu/event=cpu-cycles,percore/
CPU3 2,427,922 cpu/event=cpu-cycles,percore/
CPU4 2,752,148 cpu/event=cpu-cycles,percore/
CPU6 2,784,072 cpu/event=cpu-cycles,percore/
CPU7 2,427,922 cpu/event=cpu-cycles,percore/

1.001416041 seconds time elapsed

Ravi Bangoria reports an issue in v3. Once we offline a CPU,
the output is not correct. The issue is we should use the cpu
idx in print_percore_thread rather than using the cpu value.

Acked-by: Jiri Olsa <jolsa@xxxxxxxxxx>

Thanks so much for ACK this patch. :)

btw, there's slight misalignment in -I output, but not due
to your change, it's there for some time now, and probably
in other agregation outputs as well:

$ sudo ./perf stat -e cpu/event=cpu-cycles/ -a -A -I 1000
# time CPU counts unit events
1.000224464 CPU0 7,251,151 cpu/event=cpu-cycles/
1.000224464 CPU1 21,614,946 cpu/event=cpu-cycles/
1.000224464 CPU2 30,812,097 cpu/event=cpu-cycles/

should be (extra space after CPUX):

1.000224464 CPU2 30,812,097 cpu/event=cpu-cycles/

I'll put it on my TODO, but if you're welcome to check on it ;-)


I have a simple fix for this misalignment issue.

diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
index bc31fccc0057..95b29c9cba36 100644
--- a/tools/perf/util/stat-display.c
+++ b/tools/perf/util/stat-display.c
@@ -114,11 +114,11 @@ static void aggr_printout(struct perf_stat_config
fprintf(config->output, "S%d-D%d-C%*d%s",
- config->csv_output ? 0 : -5,
+ config->csv_output ? 0 : -3,
cpu_map__id_to_cpu(id), config->csv_sep);
} else {
- fprintf(config->output, "CPU%*d%s ",
- config->csv_output ? 0 : -5,
+ fprintf(config->output, "CPU%*d%s",
+ config->csv_output ? 0 : -7,

I guess that's ok, will that work with higher (3 digit) cpu numbers?


Yes, it works with hundreds of CPU. I have tested with that case.

BTW, do you need me to post a separate patch or you will add this fix in your patch series?

Jin Yao


Following command lines are tested OK.

perf stat -e cpu/event=cpu-cycles/ -I 1000
perf stat -e cpu/event=cpu-cycles/ -a -I 1000
perf stat -e cpu/event=cpu-cycles/ -a -A -I 1000
perf stat -e cpu/event=cpu-cycles,percore/ -a -A -I 1000
perf stat -e cpu/event=cpu-cycles,percore/ -a -A --percore-show-thread -I

Could you help to look at that?

Jin Yao