[igt-dev] [PATCH i-g-t] tests/i915/gem_shrink: Fix memory requirement assertion

[Matthew Auld]

On 21/11/2022 09:35, Petri Latvala wrote:
> On Mon, Nov 21, 2022 at 09:12:57AM +0000, Matthew Auld wrote:
>> On 18/11/2022 19:14, Niranjana Vishwanathapura wrote:
>>> On Fri, Nov 18, 2022 at 02:54:40PM +0000, Matthew Auld wrote:
>>>> On 15/11/2022 19:05, Niranjana Vishwanathapura wrote:
>>>>> The reclaim subtest do not require the alloc_size memory
>>>>> allocations. Ensure the memory requirement assert is only
>>>>> applied for other subtests.
>>>>>
>>>>> Signed-off-by: Niranjana Vishwanathapura
>>>>> <niranjana.vishwanathapura at intel.com>
>>>>> ---
>>>>>   tests/i915/gem_shrink.c | 8 ++++----
>>>>>   1 file changed, 4 insertions(+), 4 deletions(-)
>>>>>
>>>>> diff --git a/tests/i915/gem_shrink.c b/tests/i915/gem_shrink.c
>>>>> index e3e20dfc9..380d2c846 100644
>>>>> --- a/tests/i915/gem_shrink.c
>>>>> +++ b/tests/i915/gem_shrink.c
>>>>> @@ -456,9 +456,6 @@ igt_main
>>>>>                num_processes, alloc_size);
>>>>>           alloc_size <<= 20;
>>>>> -        igt_require_memory(num_processes, alloc_size,
>>>>> -                   CHECK_SWAP | CHECK_RAM);
>>>>> -
>>>>
>>>> If that's the case then all the other stuff in the fixture is not
>>>> really used either (like alloc_size, num_processes etc). Maybe just
>>>> move the fixture instead?
>>>>
>>>> igt_subtest_group {
>>>>      igt_fixture {
>>>>
>>>>      }
>>>>
>>>>      for(const struct test *t = tests; t->name; t++) {
>>>>      ....
>>>> }
>>>>
>>>> igt_subtest("reclaim")
>>>> ...
>>>>
>>>
>>> Ok, but the igt_fixture() inside the igt_subtest_group still gets executed
>>> when we run reclaim subtest (with '--r reclaim').
>>
>> Petri, do you know if that's expected?
> 
> Yes, all fixtures get executed.

Ok, so it's not really possible to have a sub-fixture, which is only run 
for a particular set of subtests? In such a case is a completely new 
test the correct approach?

I recently needed something similar here:
https://patchwork.freedesktop.org/patch/511628/?series=110908&rev=2

But I guess that should rather be a new test? Or is there a better approach?

>