[PATCH libdrm 00/10] Add C8, 30bpp and FP16 support to modetest
Ilia Mirkin
imirkin at alum.mit.edu
Mon Jun 3 13:27:37 UTC 2019
On Mon, Jun 3, 2019 at 3:32 AM Daniel Vetter <daniel at ffwll.ch> wrote:
>
> On Sun, Jun 02, 2019 at 08:40:08PM -0400, Ilia Mirkin wrote:
> > This series improves the pattern generation logic to support additional
> > formats, as well as a new "gradient" pattern (see patch comments on why
> > I found it useful).
> >
> > Furthermore, these formats are piped through to modetest, including the
> > ability to set a gamma table, which is necessary for the C8 indexed
> > format.
> >
> > This was tested on nouveau, and used for bring-up of the C8, XB30, and
> > FP16 formats on the NVIDIA hardware that supports these.
>
> Does nouveau also work with igt tests for this stuff? We do have support
> for interactive testing (i.e. "human pls check yourself" kind of tests) in
> igt, so ideally we could merge everything into one place. Long-term at
> least ...
nouveau has no special exclusions for programs that start with the
letters "igt", so presumably it should be OK with the basic tests.
However it was my impression that igt was targeted at automated
testing, and all the tests basically required crc, which is
questionable whether it exists in the hw in a manner usable by such
tests, and definitely not supported by nouveau in any case. As a
result, I haven't really taken much of a look.
Having something flexible like modetest has been really useful in
development. Being able to run with different formats, messing with
resolutions, scaling parameters for overlays, different patterns --
these things have all been helpful in validating that the new features
implemented actually work as expected. I plan on extending it further
to cover HDR, as part of my bringup of HDR on nouveau.
As an example, pre-GF119 FP16 support expects a 0..1024-valued input
instead of 0..1 (something which we did not previously know). I was
able to guess that by changing the pattern in the code to generate
larger numbers, after seeing a black display with the 0..1 pattern. (I
may have also messed with the gamma ramp to see if it was "working" or
not - I forget already.) Having a tool that makes things like that
simple to investigate is pretty valuable to me.
-ilia
More information about the dri-devel
mailing list