[igt-dev] [PATCH] i915/gem_spin_batch: Add test to execute in parallel on all engines

[Vanshidhar Konda]

On Thu, Oct 31, 2019 at 07:29:07PM +0000, Chris Wilson wrote:
>Quoting Vanshidhar Konda (2019-10-31 18:53:07)
>> On Thu, Oct 31, 2019 at 06:17:28PM +0000, Chris Wilson wrote:
>> >Quoting Vanshidhar Konda (2019-10-31 18:01:59)
>> >> Add a test that submits batch buffers to all engines and executes them
>> >> in parallel; either from the same context or from new contexts for each
>> >> engine.
>> >
>> >Duplicating a test for what purpose? Pedagogy?
>>
>> It would be great if you could point out which test this duplicates.
>> I've been looking for a test that proves a batch buffer is executing in
>> parallel on all engines. Others I spoke to couldn't point out a test.
>
>There are quite a few, depending on just how you want to go about
>proving execution. The focus for gem_spin_batch is demonstrating that
>the igt_spin_t API works as advertised, not the uABI per se.
>
>So what's your primary focus here? Demonstrating how to use igt_spin_t,
>verifying that we can schedule parallel engines, that we can specify
>parallel set of engines, we report parallel activity or that the HW
>behaves in a certain way?

The intention is to verify that the kernel can submit and hardware can execute
batch buffers on all the command streamers in parallel. spin_batch
seemed like a good candidate to keep each of the command streamers busy
while we submit new spin_batch to the next command streamer.

Vanshi

>-Chris