Re: [PATCH] perf: add csv-style output to perf stat

[Arnaldo Carvalho de Melo <acme@redhat.com>]

Em Wed, Dec 01, 2010 at 05:00:05PM +0200, Stephane Eranian escreveu:
> This patch adds an option (-x) to print counts using a CSV-style output.
> This makes it very easy to import counts directly into your favorite
> spreadsheet without having to write scripts.

I was about to work on this :-) I think we should use the same option
'report' uses:

    OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
               "separator for columns, no spaces will be added between "
               "columns '.' is reserved."),

[root@mica ~]# perf record -F 100000 ls > /dev/null
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.041 MB perf.data (~1798 samples) ]
[root@mica ~]# perf report --stdio -t, | head -10
# Events: 1K cycles
#
# Overhead,Command,Shared Object,Symbol
52.57,    ls,libc-2.5.so      ,[.] __GI___strcoll_l
3.97,    ls,ls               ,[.]             24a2
3.48,    ls,libc-2.5.so      ,[.] __GI_strlen
2.33,    ls,[ext3]           ,[k] ext3fs_dirhash
2.21,    ls,[kernel.kallsyms],[k] clear_page_c
1.94,    ls,[ext3]           ,[k] ext3_htree_store_dirent
1.81,    ls,[kernel.kallsyms],[k] rt_spin_lock_fastunlock
[root@mica ~]#

Spaces are being added, gack, will fix. Tried to use the same option letter and
long option name as in 'sort':

       -t, --field-separator=SEP
              use SEP instead of non-blank to blank transition

But then 'perf stat' already uses -t for --tid, so in 'stat' we would have to
use '-x'/--field-separator.

Argh, I think we should stop using short options, only assigning something when
it gets from seldomly used to just before making it the default 8-)

- Arnaldo
 
> Example:
> $  perf stat -x -a -- sleep 1
> 4009.795961,task-clock-msecs
> 20,context-switches
> 2,CPU-migrations
> 190,page-faults
> 9595983335,cycles
> 3492776872,instructions
> 872718098,branches
> 29798,branch-misses
> 44646,cache-references
> 5026,cache-misses
> 
> Signed-off-by: Stephane Eranian <eranian@google.com>
> ---
> 
> diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
> index c405bca..717c11d 100644
> --- a/tools/perf/Documentation/perf-stat.txt
> +++ b/tools/perf/Documentation/perf-stat.txt
> @@ -58,6 +58,11 @@ to activate system-wide monitoring. Default is to count on all CPUs.
>  Do not aggregate counts across all monitored CPUs in system-wide mode (-a).
>  This option is only valid in system-wide mode.
>  
> +-x::
> +--csv::
> +print counts using a CSV-style (comma separated) output to make it easy to
> +import directly into spreadsheets.
> +
>  EXAMPLES
>  --------
>  
> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> index 970a7f2..325608d 100644
> --- a/tools/perf/builtin-stat.c
> +++ b/tools/perf/builtin-stat.c
> @@ -76,6 +76,7 @@ static int			run_count			=  1;
>  static bool			no_inherit			= false;
>  static bool			scale				=  true;
>  static bool			no_aggr				= false;
> +static bool			csv_output			=  false;
>  static pid_t			target_pid			= -1;
>  static pid_t			target_tid			= -1;
>  static pid_t			*all_tids			=  NULL;
> @@ -84,6 +85,7 @@ static pid_t			child_pid			= -1;
>  static bool			null_run			=  false;
>  static bool			big_num				=  false;
>  static const char		*cpu_list;
> +static const char		*sep;
>  
>  
>  static int			*fd[MAX_NR_CPUS][MAX_COUNTERS];
> @@ -449,12 +451,16 @@ static void print_noise(int counter, double avg)
>  static void nsec_printout(int cpu, int counter, double avg)
>  {
>  	double msecs = avg / 1e6;
> +	char cpustr[16] = { '\0', };
> +	const char *fmt = csv_output ? "%s%.6f%s%s" : "%s%18.6f%s%-24s";
>  
>  	if (no_aggr)
> -		fprintf(stderr, "CPU%-4d %18.6f  %-24s",
> -			cpumap[cpu], msecs, event_name(counter));
> -	else
> -		fprintf(stderr, " %18.6f  %-24s", msecs, event_name(counter));
> +		sprintf(cpustr, "CPU%-4d%s", cpumap[cpu], sep);
> +
> +	fprintf(stderr, fmt, cpustr, msecs, sep, event_name(counter));
> +
> +	if (csv_output)
> +		return;
>  
>  	if (MATCH_EVENT(SOFTWARE, SW_TASK_CLOCK, counter)) {
>  		fprintf(stderr, " # %10.3f CPUs ",
> @@ -466,18 +472,24 @@ static void abs_printout(int cpu, int counter, double avg)
>  {
>  	double total, ratio = 0.0;
>  	char cpustr[16] = { '\0', };
> +	const char *fmt;
> +
> +	if (csv_output)
> +		fmt = "%s%.0f%s%s";
> +	else if (big_num)
> +		fmt = "%s%'18.0f%s%-24s";
> +	else
> +		fmt = "%s%18.0f%s%-24s";
>  
>  	if (no_aggr)
> -		sprintf(cpustr, "CPU%-4d", cpumap[cpu]);
> +		sprintf(cpustr, "CPU%-4d%s", cpumap[cpu], sep);
>  	else
>  		cpu = 0;
>  
> -	if (big_num)
> -		fprintf(stderr, "%s %'18.0f  %-24s",
> -			cpustr, avg, event_name(counter));
> -	else
> -		fprintf(stderr, "%s %18.0f  %-24s",
> -			cpustr, avg, event_name(counter));
> +	fprintf(stderr, fmt, cpustr, avg, sep, event_name(counter));
> +
> +	if (csv_output)
> +		return;
>  
>  	if (MATCH_EVENT(HARDWARE, HW_INSTRUCTIONS, counter)) {
>  		total = avg_stats(&runtime_cycles_stats[cpu]);
> @@ -515,8 +527,8 @@ static void print_counter_aggr(int counter)
>  	int scaled = event_scaled[counter];
>  
>  	if (scaled == -1) {
> -		fprintf(stderr, " %18s  %-24s\n",
> -			"<not counted>", event_name(counter));
> +		fprintf(stderr, "%18s%s%-24s\n",
> +			"<not counted>", sep, event_name(counter));
>  		return;
>  	}
>  
> @@ -525,6 +537,11 @@ static void print_counter_aggr(int counter)
>  	else
>  		abs_printout(-1, counter, avg);
>  
> +	if (csv_output) {
> +		fputc('\n', stderr);
> +		return;
> +	}
> +
>  	print_noise(counter, avg);
>  
>  	if (scaled) {
> @@ -554,8 +571,10 @@ static void print_counter(int counter)
>  		ena = cpu_counts[cpu][counter].ena;
>  		run = cpu_counts[cpu][counter].run;
>  		if (run == 0 || ena == 0) {
> -			fprintf(stderr, "CPU%-4d %18s  %-24s", cpumap[cpu],
> -					"<not counted>", event_name(counter));
> +			fprintf(stderr, "CPU%-4d%s%18s%s%-24s",
> +				cpumap[cpu], sep,
> +				"<not counted>", sep,
> +				event_name(counter));
>  
>  			fprintf(stderr, "\n");
>  			continue;
> @@ -566,11 +585,13 @@ static void print_counter(int counter)
>  		else
>  			abs_printout(cpu, counter, val);
>  
> -		print_noise(counter, 1.0);
> +		if (!csv_output) {
> +			print_noise(counter, 1.0);
>  
> -		if (run != ena) {
> -			fprintf(stderr, "  (scaled from %.2f%%)",
> +			if (run != ena) {
> +				fprintf(stderr, "  (scaled from %.2f%%)",
>  					100.0 * run / ena);
> +			}
>  		}
>  		fprintf(stderr, "\n");
>  	}
> @@ -582,21 +603,23 @@ static void print_stat(int argc, const char **argv)
>  
>  	fflush(stdout);
>  
> -	fprintf(stderr, "\n");
> -	fprintf(stderr, " Performance counter stats for ");
> -	if(target_pid == -1 && target_tid == -1) {
> -		fprintf(stderr, "\'%s", argv[0]);
> -		for (i = 1; i < argc; i++)
> -			fprintf(stderr, " %s", argv[i]);
> -	} else if (target_pid != -1)
> -		fprintf(stderr, "process id \'%d", target_pid);
> -	else
> -		fprintf(stderr, "thread id \'%d", target_tid);
> +	if (!csv_output) {
> +		fprintf(stderr, "\n");
> +		fprintf(stderr, " Performance counter stats for ");
> +		if(target_pid == -1 && target_tid == -1) {
> +			fprintf(stderr, "\'%s", argv[0]);
> +			for (i = 1; i < argc; i++)
> +				fprintf(stderr, " %s", argv[i]);
> +		} else if (target_pid != -1)
> +			fprintf(stderr, "process id \'%d", target_pid);
> +		else
> +			fprintf(stderr, "thread id \'%d", target_tid);
>  
> -	fprintf(stderr, "\'");
> -	if (run_count > 1)
> -		fprintf(stderr, " (%d runs)", run_count);
> -	fprintf(stderr, ":\n\n");
> +		fprintf(stderr, "\'");
> +		if (run_count > 1)
> +			fprintf(stderr, " (%d runs)", run_count);
> +		fprintf(stderr, ":\n\n");
> +	}
>  
>  	if (no_aggr) {
>  		for (counter = 0; counter < nr_counters; counter++)
> @@ -606,15 +629,17 @@ static void print_stat(int argc, const char **argv)
>  			print_counter_aggr(counter);
>  	}
>  
> -	fprintf(stderr, "\n");
> -	fprintf(stderr, " %18.9f  seconds time elapsed",
> -			avg_stats(&walltime_nsecs_stats)/1e9);
> -	if (run_count > 1) {
> -		fprintf(stderr, "   ( +- %7.3f%% )",
> +	if (!csv_output) {
> +		fprintf(stderr, "\n");
> +		fprintf(stderr, " %18.9f  seconds time elapsed",
> +				avg_stats(&walltime_nsecs_stats)/1e9);
> +		if (run_count > 1) {
> +			fprintf(stderr, "   ( +- %7.3f%% )",
>  				100*stddev_stats(&walltime_nsecs_stats) /
>  				avg_stats(&walltime_nsecs_stats));
> +		}
> +		fprintf(stderr, "\n\n");
>  	}
> -	fprintf(stderr, "\n\n");
>  }
>  
>  static volatile int signr = -1;
> @@ -670,6 +695,8 @@ static const struct option options[] = {
>  		    "list of cpus to monitor in system-wide"),
>  	OPT_BOOLEAN('A', "no-aggr", &no_aggr,
>  		    "disable CPU count aggregation"),
> +	OPT_BOOLEAN('x', "csv", &csv_output,
> +		    "print counts in CSV format"),
>  	OPT_END()
>  };
>  
> @@ -682,6 +709,17 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used)
>  
>  	argc = parse_options(argc, argv, options, stat_usage,
>  		PARSE_OPT_STOP_AT_NON_OPTION);
> +
> +	sep = csv_output ? "," : " ";
> +
> +	/*
> +	 * let the spreadsheet do the pretty-printing
> +	 */
> +	if (csv_output && big_num) {
> +		fprintf(stderr, "-B option not supported with -x\n");
> +		usage_with_options(stat_usage, options);
> +	}
> +
>  	if (!argc && target_pid == -1 && target_tid == -1)
>  		usage_with_options(stat_usage, options);
>  	if (run_count <= 0)