[Intel-gfx] Anonymoose ggtt_view_params

Chris Wilson chris at chris-wilson.co.uk
Fri Jan 13 11:34:04 UTC 2017


On Fri, Jan 13, 2017 at 11:23:47AM +0000, Tvrtko Ursulin wrote:
> 
> On 13/01/2017 11:11, Chris Wilson wrote:
> >On Fri, Jan 13, 2017 at 10:59:46AM +0000, Tvrtko Ursulin wrote:
> >>
> >>On 13/01/2017 10:33, Chris Wilson wrote:
> >>>Ok, ok, this cover note only exists to continue the run on joke of my
> >>>mispellings!
> >>>
> >>>Everything but
> >>>[5/7] drm/i915: Convert i915_ggtt_view to use an anonymous union
> >>>has a r-b, so this is a good time to complain if this is too much of a
> >>>hack.
> >>
> >>If you could polish your clouded crystal ball to see if more view
> >>types might be coming, which then might have a colliding parameters
> >>size and foil the whole idea.
> >>
> >>I do think it is a little bit of hack with a questionable benefit.
> >>And I think I asked a few times if you really see a performance
> >>difference for a few bytes smaller memcmp? Presumably it would be
> >>some test case with a huge number of partial views which could
> >>theoretically maybe show something?
> >
> >It was the doubling code size of i915_vma_compare() that struck me as
> >objectionable.
> 
> Why does this series shrink i915_vma_compare? Was it getting inlined
> in your build? For me it doesn't.

It's inlined for me. From my build, we only maintain the size of the
i915_vma_lookup. Hmm, this might also explain why changing the return of
i915_vma_compare() had significance for your build and not mine. Weird
gcc is weird.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre


More information about the Intel-gfx mailing list