Quoting Tvrtko Ursulin (2020-06-17 17:01:17)
> From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
>
> It is useful to know how much memory workload is allocating.
>
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
> ---
> benchmarks/gem_wsim.c | 100 +++++++++++++++++++++++++++++++++++++++---
> 1 file changed, 95 insertions(+), 5 deletions(-)
>
> diff --git a/benchmarks/gem_wsim.c b/benchmarks/gem_wsim.c
> index 025385a144b8..96ee923fb699 100644
> --- a/benchmarks/gem_wsim.c
> +++ b/benchmarks/gem_wsim.c
> @@ -852,7 +852,8 @@ static uint64_t engine_list_mask(const char *_str)
> return mask;
> }
>
> -static void allocate_working_set(struct workload *wrk, struct working_set *set);
> +static unsigned long
> +allocate_working_set(struct workload *wrk, struct working_set *set);
>
> static long __duration(long dur, double scale)
> {
> @@ -1270,8 +1271,14 @@ add_step:
> * Allocate shared working sets.
> */
> for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
> - if (w->type == WORKINGSET && w->working_set.shared)
> - allocate_working_set(wrk, &w->working_set);
> + if (w->type == WORKINGSET && w->working_set.shared) {
> + unsigned long total =
> + allocate_working_set(wrk, &w->working_set);
> +
> + if (verbose > 1)
> + printf("%u: %lu bytes in shared working set %u\n",
> + wrk->id, total, w->working_set.id);
> + }
> }
The total total might be nice; although that doesn't reflect usage so
might be misleading as to what is the active RSS is at any time.
> wrk->max_working_set_id = -1;
> @@ -1731,8 +1738,10 @@ get_buffer_size(struct workload *wrk, const struct work_buffer_size *sz)
> (sz->max + 1 - sz->min);
> }
>
> -static void allocate_working_set(struct workload *wrk, struct working_set *set)
> +static unsigned long
> +allocate_working_set(struct workload *wrk, struct working_set *set)
> {
> + unsigned long total = 0;
> unsigned int i;
>
> set->handles = calloc(set->nr, sizeof(*set->handles));
> @@ -1741,7 +1750,82 @@ static void allocate_working_set(struct workload *wrk, struct working_set *set)
> for (i = 0; i < set->nr; i++) {
> set->sizes[i].size = get_buffer_size(wrk, &set->sizes[i]);
> set->handles[i] = alloc_bo(fd, set->sizes[i].size);
> + total += set->sizes[i].size;
> + }
> +
> + return total;
> +}
> +
> +static bool
> +find_dep(struct dep_entry *deps, unsigned int nr, struct dep_entry dep)
> +{
> + unsigned int i;
> +
> + for (i = 0; i < nr; i++) {
> + if (deps[i].working_set == dep.working_set &&
> + deps[i].target == dep.target)
> + return true;
> }
> +
> + return false;
> +}
> +
> +static void measure_active_set(struct workload *wrk)
> +{
> + unsigned long total = 0, batch_sizes = 0;
> + struct dep_entry *deps = NULL;
> + unsigned int nr = 0, i, j;
> + struct w_step *w;
> +
> + if (verbose < 3)
> + return;
> +
> + for (i = 0, w = wrk->steps; i < wrk->nr_steps; i++, w++) {
> + if (w->type != BATCH)
> + continue;
> +
> + batch_sizes += w->bb_sz;
> +
> + for (j = 0; j < w->data_deps.nr; j++) {
> + struct dep_entry *dep = &w->data_deps.list[j];
> + struct dep_entry _dep = *dep;
> +
> + if (dep->working_set == -1 && dep->target < 0) {
> + int idx = w->idx + dep->target;
> +
> + igt_assert(idx >= 0 && idx < w->idx);
> + igt_assert(wrk->steps[idx].type == BATCH);
> +
> + _dep.target = wrk->steps[idx].obj[0].handle;
> + }
> +
> + if (!find_dep(deps, nr, _dep)) {
> + if (dep->working_set == -1) {
> + total += 4096;
> + } else {
> + struct working_set *set;
> +
> + igt_assert(dep->working_set <=
> + wrk->max_working_set_id);
> +
> + set = wrk->working_sets[dep->working_set];
> + igt_assert(set->nr);
> + igt_assert(dep->target < set->nr);
> + igt_assert(set->sizes[dep->target].size);
> +
> + total += set->sizes[dep->target].size;
> + }
> +
> + deps = realloc(deps, (nr + 1) * sizeof(*deps));
> + deps[nr++] = *dep;
> + }
> + }
> + }
So a sum of all the unique handles used by all the steps.
Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
