On Mon, Jan 11, 2016 at 11:23 AM Sarah Sharp <sarah.a.sh...@linux.intel.com>
wrote:
> On Sat, Jan 09, 2016 at 11:57:38AM -0800, Kristian Høgsberg Kristensen
> wrote:
> > Mark Janes <mark.a.ja...@intel.com> writes:
> >
> > > Sarah Sharp <sarah.a.sh...@intel.com> writes:
> > >
> > >> On Fri, Jan 08, 2016 at 04:30:20PM -0800, Mark Janes wrote:
> > >>> In testing KBL, I found:
> > >>>
> > >>> - urb size was not set for slices gt1.5, gt2, and gt3. The value I
> > >>> used for these slices (384) was taken from an earlier patch
> authored
> > >>> by Ben Widawsky.
> > >>>
> > >>> - slice count was missing. This field was added by
> > >>> a403ad4f5a034e52a3cd845e91c4aa3e6927b731
> > >>>
> > >>> With this commit, KBL passes piglit at parity with SKL.
> > >>
> > >> Thanks for catching this Mark!
> > >>
> > >> Looking at the docs, I'm not sure the URB size is correct for GT4. The
> > >> Device Attributes page shows the GT4 URB size should be 1152K (divided
> > >> by the number of slices, that's 384K). However, another table says to
> > >> limit the URB size to 1008K for "SKL+" and it doesn't say to exclude
> > >> KBL. Yay conflicting documentation!
> > >
> > > Unfortunately, Ben doesn't get back until Wednesday.
> > >
> > > Since this patch doesn't modify the GT4 URB size, does it make sense to
> > > use it to enable the other skus? GT4 can be addressed later.
> >
> > The patch looks good to me, though urb.size is already set in the
> > GEN9_FEATURES macro:
> >
> > ...
> > .urb = { \
> > .size = 384, \
> > ...
> >
> > which looks correct for all SKUs, except GT1 and GT4, which both
> > override it to the correct value. As for GT4 urb size, it wont matter
> > for a while, but 1088 / 3 is the safe choice.
> >
> > With the .urb.size assignments removed, this patch is
> >
> > Reviewed-by: Kristian Høgsberg Kristensen <
> kristian.h.kristen...@intel.com>
>
> Thanks for the review, Kristian! I've tested the modified patch (without
> the urb.size) on KBL, and it works fine. I'll go push this and mark it
> for stable.
>
oh! my bad. I mislead Mark on the urb size while he caught the real
offending part that was num_slices...
apparently num_slices didn't exist on kbl-po branch so this is why we
didn't see this issue earlier...
> Sarah Sharp
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
