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

[Tvrtko Ursulin]

On 05/10/2023 11:52, Bernatowicz, Marcin wrote:
> 
> 
> On 9/29/2023 12:49 PM, Tvrtko Ursulin wrote:
>>
>> 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.
> 
> I got information there may be a case a prefetch size 4096 on some 
> platform, but I did not get a clear answer why/if above calculation is 
> needed/redundant. So I assume it's redundant and I will not copy/paste 
> it from igt tests.
> 
> There is one concern I have related to DRM_IOCTL_XE_GEM_CREATE ioctl and 
> size field, suggesting a need for bb_size according to description:
> 
> struct drm_xe_gem_create {
>          ...
>      /**
>       * @size: Requested size for the object
>       *
>       * The (page-aligned) allocated size for the object will be returned.
>       */
>      __u64 size;
> 
> It suggests size could be adjusted after a call, but I think there is 
> some discussion ongoing to that, so will wait. (and the 
> xe_bo_create_flags does not take it into account and does not update 
> size param)

i915 gem_create also rounds up the size and reports it back btw.

Regards,

Tvrtko

> 
> I will drop the patch, leave 4096 and I'm expecting a xe_bo_create 
> failure if not possible.
> 
> Regards,
> Marcin
> 
>>
>> 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];