[Intel-gfx] [PATCH i-g-t 0/2] kms_plane: Enumerate outputs before planes

[Imre Deak]

On Wed, Nov 29, 2017 at 04:07:10PM +0100, Daniel Vetter wrote:
> On Tue, Nov 21, 2017 at 09:52:52PM +0200, Imre Deak wrote:
> > Enumerate outputs before planes means we can avoid doing an extra
> > modeset when calculating the reference CRC during the test of each
> > plane. In addition when testing multiple outputs we can avoid a full
> > modeset we currently do whenever switching to a new output within a
> > subtest. This speeds up things especially on internal panels with long
> > power cycle delays.
> > 
> > Below is the runtime of kms_plane on my GLK before and after the change.
> > I excluded from this the DPMS and suspend subtests, since those
> > transitions have a bigger overhead, unrelated to this change.
> 
> Even better would be to push the modeset out of the subtests into
> igt_fixtures, so that we can amortize them over all subtests. Since
> somewhen in the future CI will run tests binaries-at-a-time, instead of
> subtests.

Yes, in that case it'd make sense to calculate one CRC per output
independent of the crtc. 

> That means we won't test all combinations of crtc/output, but for all the
> hw I know there's no difference really. Timestamps are a different story,
> but these tests here don't check for that.

Ville had a good idea here, to add an option to filter for
crtc/output/plane/rotation/format etc. for which a test would be run.

> 
> And yes we need to make these tests faster, especially on gen9 kms takes
> too long.
> 
> Either way this improves a lot already, on both patches:
> 
> Reviewed-by: Daniel Vetter <daniel.vetter at ffwll.ch>

Thanks.

> 
> > 
> > With a single HDMI output:
> >  42s vs 26s
> > 
> > With a single eDP output:
> > 115s vs 50s
> > 
> > With eDP+HDMI output:
> > 171s vs 74s
> > 
> > Imre Deak (2):
> >   kms_plane: Enumerate outputs before planes in position subtests
> >   kms_plane: Enumerate outputs before planes in panning subtests
> > 
> >  tests/kms_plane.c | 120 +++++++++++++++++++++++-------------------------------
> >  1 file changed, 51 insertions(+), 69 deletions(-)
> > 
> > -- 
> > 2.11.0
> > 
> > _______________________________________________
> > Intel-gfx mailing list
> > Intel-gfx at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
> 
> -- 
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch