Re: [PATCH v3] perf stat: Fix wrong skipping for per-die aggregation

[Jin, Yao]

Hi Jiri,

On 1/4/2021 6:15 PM, Jiri Olsa wrote:
> On Fri, Dec 25, 2020 at 09:04:09AM +0800, Jin Yao wrote:
>
> SNIP
>
> >   void update_stats(struct stats *stats, u64 val)
> > @@ -275,16 +276,39 @@ void evlist__save_aggr_prev_raw_counts(struct evlist *evlist)
> >   
> >   static void zero_per_pkg(struct evsel *counter)
> >   {
> > -	if (counter->per_pkg_mask)
> > -		memset(counter->per_pkg_mask, 0, cpu__max_cpu());
> > +	struct hashmap_entry *entry;
> > +	size_t bkt;
> > +
> > +	if (counter->per_pkg_mask) {
> > +		hashmap__for_each_entry(counter->per_pkg_mask, entry, bkt) {
> > +			bool *used = (bool *)entry->value;
> > +
> > +			*used = false;
> > +		}
> > +	}
> > +}
> > +
> > +static size_t id_hash(const void *key, void *ctx __maybe_unused)
> > +{
> > +	int socket = (int64_t)key >> 32;
> > +
> > +	return socket;
> > +}
> > +
> > +static bool id_equal(const void *key1, const void *key2,
> > +		     void *ctx __maybe_unused)
> > +{
> > +	return (int64_t)key1 == (int64_t)key2;
> >   }
>
> please use more descriptive names, pkg_id_hash/pkg_id_equal or such

Corrected in v4.

> >   
> >   static int check_per_pkg(struct evsel *counter,
> >   			 struct perf_counts_values *vals, int cpu, bool *skip)
> >   {
> > -	unsigned long *mask = counter->per_pkg_mask;
> > +	struct hashmap *mask = counter->per_pkg_mask;
> >   	struct perf_cpu_map *cpus = evsel__cpus(counter);
> > -	int s;
> > +	int s, d, ret;
> > +	uint64_t key;
> > +	bool *used;
> >   
> >   	*skip = false;
> >   
> > @@ -295,7 +319,7 @@ static int check_per_pkg(struct evsel *counter,
> >   		return 0;
> >   
> >   	if (!mask) {
> > -		mask = zalloc(cpu__max_cpu());
> > +		mask = hashmap__new(id_hash, id_equal, NULL);
> >   		if (!mask)
> >   			return -ENOMEM;
> >   
> > @@ -317,7 +341,32 @@ static int check_per_pkg(struct evsel *counter,
> >   	if (s < 0)
> >   		return -1;
> >   
> > -	*skip = test_and_set_bit(s, mask) == 1;
> > +	/*
> > +	 * On multi-die system, 0 < die_id < 256. On no-die system, die_id = 0.
> > +	 * We use hashmap(socket, die) to check the used socket+die pair.
> > +	 */
> > +	d = cpu_map__get_die(cpus, cpu, NULL).die;
> > +	if (d < 0)
> > +		return -1;
> > +
> > +	key = (uint64_t)s << 32 | (d & 0xff);
> > +	if (hashmap__find(mask, (void *)key, (void **)&used)) {
> > +		if (*used)
> > +			*skip = true;
> > +		*used = true;
> > +	} else {
> > +		used = zalloc(sizeof(*used));
> > +		if (!used)
> > +			return -1;
>
> hum, what's the point of having extra bool value? once the
> item is in the hashtab, we have the answer
>
> I think you can add item to hashtab with '1' value and get
> rid of that bool allocation
>
> zero_per_pkg will be just removing all items from hashtab
>
> jirka

Thanks for the suggestion! Yes, we don't need the bool value allocation here, it's unnecessary.

I just post the v4. Please help to take a look.

Thanks
Jin Yao