Re: [PATCH 1/2] perf stat: refactor aggregation code

[Namhyung Kim <namhyung@kernel.org>]

Hi Stephane,

On Tue, 12 Feb 2013 15:09:27 +0100, Stephane Eranian wrote:
> Refactor aggregation code by introducing
> a single aggr_mode variable and an enum
> for aggregation.
>
> Also refactor cpumap code having to do
> with cpu to socket mappings. All in preparation
> for extended modes, such as cpu -> core.
>
> Also fix socket aggregation and ensure
> that sockets are printed in increasing order.
[snip]
> -static void print_aggr_socket(char *prefix)
> +static void print_aggr(char *prefix)
>  {
>  	struct perf_evsel *counter;
> +	int cpu, s, s2, id, nr;
>  	u64 ena, run, val;
> -	int cpu, s, s2, sock, nr;
>  
> -	if (!sock_map)
> +	if (!(aggr_map || aggr_get_id))
>  		return;
>  
> -	for (s = 0; s < sock_map->nr; s++) {
> -		sock = cpu_map__socket(sock_map, s);
> +	for (s = 0; s < aggr_map->nr; s++) {
> +		id = aggr_map->map[s];
>  		list_for_each_entry(counter, &evsel_list->entries, node) {
>  			val = ena = run = 0;
>  			nr = 0;
...
> @@ -1073,14 +1081,20 @@ static void print_stat(int argc, const char **argv)
>  		fprintf(output, ":\n\n");
>  	}
>  
> -	if (aggr_socket)
> -		print_aggr_socket(NULL);
> -	else if (no_aggr) {
> -		list_for_each_entry(counter, &evsel_list->entries, node)
> -			print_counter(counter, NULL);
> -	} else {
> +	switch (aggr_mode) {
> +	case AGGR_SOCKET:
> +		print_aggr(NULL);
> +		break;

This line should look like this IMHO:

		list_for_each_entry(counter, &evsel_list->entries, node)
			print_aggr(counter, NULL);

and the equivalent loop in the print_aggr() should be removed.  This is
for consistency with other formats if multiple events counted.

For instance, it'd sorting on events first and then sockets like:

  #      time socket cpus     counts events
           t0 S0     4        XXXXXX cycles
           t0 S1     4        XXXXXX cycles
           t0 S0     4          YYYY cache-misses
           t0 S1     4          YYYY cache-misses
           t1 S0     4        ZZZZZZ cycles
  ...

But current code looks like sorting on sockets first instead.

  #      time socket cpus     counts events
           t0 S0     4        XXXXXX cycles
           t0 S0     4          YYYY cache-misses
           t0 S1     4        XXXXXX cycles
           t0 S1     4          YYYY cache-misses
           t1 S0     4        ZZZZZZ cycles
  ...

Thanks,
Namhyung


> +	case AGGR_GLOBAL:
>  		list_for_each_entry(counter, &evsel_list->entries, node)
>  			print_counter_aggr(counter, NULL);
> +		break;
> +	case AGGR_NONE:
> +		list_for_each_entry(counter, &evsel_list->entries, node)
> +			print_counter(counter, NULL);
> +		break;
> +	default:
> +		break;
>  	}