[igt-dev] [PATCH i-g-t v2 0/4] kms_plane: Add clipping subtests

[Daniel Vetter]

On Wed, Mar 07, 2018 at 07:29:55PM +0200, Imre Deak wrote:
> On Wed, Mar 07, 2018 at 06:03:19PM +0100, Daniel Vetter wrote:
> > On Tue, Mar 06, 2018 at 06:52:11PM +0200, Imre Deak wrote:
> > > This is v2 of [1], fixing the issues reported by Arek and CI. Besides
> > > the changes described in patch 4, there is also an FBC workaround
> > > applied now in kernel. Without the WA the test cases using the primary
> > > plane (with FBC) could trigger FIFO underflow errors.
> > > 
> > > Note that the subtest takes ~40 seconds to run on a single pipe.
> > 
> > Do we need to spend that much time testing clipping? What's the overlap
> > with all the various other plane clipping tests we have already, which
> > focus exclusively on the cursor?
> 
> There's no other clipping tests using non-cursor planes. I think we are
> missing coverage here: with the plane's start/end located close to the
> screen edge triggered FIFO underflows due to two new issues already.
> 
> To reduce the runtime: in this patchset I removed one redundant modeset
> after capturing the CRC, so I can't see what to improve on that front.
> It goes through all format-sizes/tiling/rotation and planes so we could
> avoid doing some of that maybe splitting them into separate subtests and
> not running them for patchwork checks.
> 
> Not sure if CRC capturing is the most optimal in this test, I could look
> into improving that. Maybe Maarten has some idea?

Could we test multiple things at the same maybe? Like cursor in one
corner, plane in the other. As long as the two are really far away the wms
shouldn't influence each another really.

Wrt CRC, if you're really efficient you can sample one every frame. Since
you need two per tests (one for the reference image, one for the hw
blending) that means 40*60/2 = 1200 different cases. But this only happens
if you capture CRC continuously and make sure you're really careful with
matching CRC frame ids correctly to your flips.

If you restart crc cacpture each time you lose a few frames each (because
there's issues we don't understand with the first few crcs, so need to
drop those), which means you test througput drops by about a factor of 10.

(I didn't look whether this is actually the case, too late in the evening
now).

Cheers, Daniel

> 
> --Imre
> 
> > 
> > Note: Not blocking this, just want to make sure we have some overall plan
> > here maybe and aren't digging ourselves some comfy graves :-)
> > 
> > Cheers, Daniel
> > 
> > > 
> > > [1]
> > > https://lists.freedesktop.org/archives/igt-dev/2018-February/000523.html
> > > 
> > > Cc: Mika Kahola <mika.kahola at intel.com>
> > > Cc: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
> > > Cc: Juha-Pekka Heikkil?? <juha-pekka.heikkila at intel.com>
> > > Cc: Arkadiusz Hiler <arkadiusz.hiler at intel.com>
> > > 
> > > Imre Deak (4):
> > >   kms_plane: Remove redundant modeset after CRC capture
> > >   lib: Export helpers to get rotation/tiling strings
> > >   kms_plane: Split helpers creating reference FB and capturing CRC
> > >   kms_plane: Add clipping subtests
> > > 
> > >  lib/igt_fb.c      |  23 +++
> > >  lib/igt_fb.h      |   1 +
> > >  lib/igt_kms.c     |  11 +-
> > >  lib/igt_kms.h     |   1 +
> > >  tests/kms_plane.c | 502 +++++++++++++++++++++++++++++++++++++++++++++++++++---
> > >  5 files changed, 517 insertions(+), 21 deletions(-)
> > > 
> > > -- 
> > > 2.13.2
> > > 
> > > _______________________________________________
> > > igt-dev mailing list
> > > igt-dev at lists.freedesktop.org
> > > https://lists.freedesktop.org/mailman/listinfo/igt-dev
> > 
> > -- 
> > Daniel Vetter
> > Software Engineer, Intel Corporation
> > http://blog.ffwll.ch

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch