[Intel-gfx] [PATCH i-g-t] kms_atomic_transition: Split out modeset tests on internal panels
Imre Deak
imre.deak at intel.com
Fri Nov 3 09:21:30 UTC 2017
On Thu, Nov 02, 2017 at 05:12:56PM +0000, Chris Wilson wrote:
> Quoting Imre Deak (2017-11-02 13:38:09)
> > Doing modeset on internal panels may have a considerable overhead due to
> > the panel specific power sequencing delays. To avoid long test runtimes
> > in the CI fast-feedback test split out the testing of internal panels
> > from the plane modeset subtests. Create fast and slow versions of these
> > new subtests. In the fast one only combinations with all enabled, all
> > planes disabled or a single plane enable are tested. In the slow one all
> > plane combinations are tested.
>
> > @@ -528,6 +529,10 @@ run_transition_test(igt_display_t *display, enum pipe pipe, igt_output_t *output
> > }
> >
> > for (i = 0; i < iter_max; i++) {
> > + if (type == TRANSITION_MODESET_PLANES_FAST &&
> > + hweight32(i) != 1 && hweight32(i) != pipe_obj->n_planes)
>
> So reduce number of iterations to a power of two, and only operate if
> pipe_obj->n_planes == 1
>
> Why repeat hweight32(i) when you've already proved it is 1? That reads
> very fishily.
I wanted to test only combinations where either only one plane is
enabled or all planes are enabled. How about adding
int n_enabled_planes = hweight32(i);
and checking that instead to make things clearer?
--Imre
More information about the Intel-gfx
mailing list