[igt-dev] [PATCH i-g-t 15/17] benchmarks/gem_wsim: introduce bb_size in w_step

[Tvrtko Ursulin]

On 29/09/2023 11:08, Bernatowicz, Marcin wrote:
> 
> 
> On 9/29/2023 11:35 AM, Tvrtko Ursulin wrote:
>>
>> On 28/09/2023 18:45, Marcin Bernatowicz wrote:
>>> Put it next to bb_handle.
>>> Use it in alloc_step_batch and measure_active_set.
>>
>> Could say why.
>>
>> Like xe might need more than 4k? Might not be able to allocate only 
>> 4k? (Guessing only.)
> 
> Xe uses following formula:
> 
> w->bb_size = ALIGN(sizeof(*w->xe.data) + xe_cs_prefetch_size(fd),
>                 xe_get_default_alignment(fd));
> 
> which equaled 4096 on platform I tested.
> I didn't want to put bb_size inside xe specifics as it is connected with 
> bb_handle.

Hmmm could you dig a bit to figure out if sometimes this can be larger 
than 4k and if so why only xe and not i915. Because things like prefetch 
and alignment sound like should be more hardware than driver dependent.

Regards,

Tvrtko

> 
> Regards,
> marcin
> 
>>
>> Regards,
>>
>> Tvrtko
>>
>>> Signed-off-by: Marcin Bernatowicz <marcin.bernatowicz at linux.intel.com>
>>> ---
>>>   benchmarks/gem_wsim.c | 6 ++++--
>>>   1 file changed, 4 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/benchmarks/gem_wsim.c b/benchmarks/gem_wsim.c
>>> index 4618509ab..d22d66aeb 100644
>>> --- a/benchmarks/gem_wsim.c
>>> +++ b/benchmarks/gem_wsim.c
>>> @@ -183,6 +183,7 @@ struct w_step {
>>>           } i915;
>>>       };
>>>       uint32_t bb_handle;
>>> +    size_t bb_size;
>>>   };
>>>   struct ctx {
>>> @@ -1481,6 +1482,7 @@ alloc_step_batch(struct workload *wrk, struct 
>>> w_step *w)
>>>       unsigned int nr_obj = 2 + w->data_deps.nr;
>>>       unsigned int i;
>>> +    w->bb_size = 4096;
>>>       w->i915.obj = calloc(nr_obj, sizeof(*w->i915.obj));
>>>       igt_assert(w->i915.obj);
>>> @@ -1522,7 +1524,7 @@ alloc_step_batch(struct workload *wrk, struct 
>>> w_step *w)
>>>           igt_assert(j < nr_obj);
>>>       }
>>> -    w->bb_handle = w->i915.obj[j].handle = gem_create(fd, 4096);
>>> +    w->bb_handle = w->i915.obj[j].handle = gem_create(fd, w->bb_size);
>>>       w->i915.obj[j].relocation_count = create_bb(w, j);
>>>       igt_assert(w->i915.obj[j].relocation_count <= 
>>> ARRAY_SIZE(w->i915.reloc));
>>>       w->i915.obj[j].relocs_ptr = to_user_pointer(&w->i915.reloc);
>>> @@ -1722,7 +1724,7 @@ static void measure_active_set(struct workload 
>>> *wrk)
>>>           if (w->type != BATCH)
>>>               continue;
>>> -        batch_sizes += 4096;
>>> +        batch_sizes += w->bb_size;
>>>           for (j = 0; j < w->data_deps.nr; j++) {
>>>               struct dep_entry *dep = &w->data_deps.list[j];