[Mesa-dev] [PATCH] st/dri: decrease input lag by syncing sooner in SwapBuffers

[Axel Davy]

On 26/04/2019 10:08, Michel Dänzer wrote:
> On 2019-04-26 4:06 a.m., Marek Olšák wrote:
>> From: Marek Olšák <marek.olsak at amd.com>
>>
>> It's done by:
>> - decrease the number of frames in flight by 1
>> - flush before throttling in SwapBuffers
>>    (instead of wait-then-flush, do flush-then-wait)
>>
>> The improvement is apparent with Unigine Heaven.
>>
>> Previously:
>>      draw frame 2
>>      wait frame 0
>>      flush frame 2
>>      present frame 2
>>
>>      The input lag is 2 frames.
>>
>> Now:
>>      draw frame 2
>>      flush frame 2
>>      wait frame 1
>>      present frame 2
>>
>>      The input lag is 1 frame. Flushing is done before waiting, because
>>      otherwise the device would be idle after waiting.
> Nice idea. Not sure offhand about all ramifications, but certainly worth
> a go.
>
>
>> Nine is affected because it also uses the pipe cap.
>> ---
>>   src/gallium/auxiliary/util/u_screen.c         |  2 +-
>>   src/gallium/state_trackers/dri/dri_drawable.c | 20 +++++++++----------
>>   2 files changed, 11 insertions(+), 11 deletions(-)
>>
>> diff --git a/src/gallium/auxiliary/util/u_screen.c b/src/gallium/auxiliary/util/u_screen.c
>> index 27f51e0898e..410f17421e6 100644
>> --- a/src/gallium/auxiliary/util/u_screen.c
>> +++ b/src/gallium/auxiliary/util/u_screen.c
>> @@ -349,21 +349,21 @@ u_pipe_screen_get_param_defaults(struct pipe_screen *pscreen,
>>      case PIPE_CAP_MAX_VARYINGS:
>>         return 8;
>>   
>>      case PIPE_CAP_COMPUTE_GRID_INFO_LAST_BLOCK:
>>         return 0;
>>   
>>      case PIPE_CAP_COMPUTE_SHADER_DERIVATIVES:
>>         return 0;
>>   
>>      case PIPE_CAP_MAX_FRAMES_IN_FLIGHT:
>> -      return 2;
>> +      return 1;
> This might be slightly misleading, as there can still be two frames in
> flight (on the GPU) at the same time. Might be better to leave this at 2
> (so Nine isn't affected) and adjust its treatment in
> src/gallium/state_trackers/dri/dri_drawable.c .
>
>
Checking what gallium nine does currently, it seems we already do flush 
then wait,
however we call swap_fences_pop_front and swap_fences_push_back in the 
reverse order compared to your patch.
We compensate by taking PIPE_CAP_MAX_FRAMES_IN_FLIGHT + 1

In conclusion, with the proposed patch, gl and nine should have the same 
behaviour (and thus if gl benefits from a value of 1, nine should as well).
I haven't have noticed input lag, I guess I have to test on heaven if 
you see a difference.
How can I slow down my gpu to test that ? I use to use the 
/sys/kernel/debug/dri/0/ vars to force low dpm, but it doesn't seem to 
be possible anymore as the related files are gone (rx480) ?


Axel