On 13/05/16 09:50, Chris Wilson wrote:
On Fri, May 13, 2016 at 09:38:54AM +0100, Tvrtko Ursulin wrote:
On 12/05/16 11:25, Chris Wilson wrote:
In order to reduce the workload of the caller, we do not want to
actually have to retire requests of others when checking the status of
this object.
Wrt the subject, and from wait ioctl as well.
Also i915_gem_object_sync / i915_gem_execbuffer_move_to_gpu path
(execbuf) looks like it could still do the upto retiring so that
isn't only moving the thing around?
And in general, commit does not say who was impacted by this and how much?
Nothing much until we have them all removed. Please see the other
patches! The end result in having a completely lockless busy-ioctl is a
major improvement to a minor path... The impact is really only measured
by reduced contention between clients, for which the busy-ioctl is
mostly a victim.
This was just an opportunistic change to reduce the numer of
retire__upto when thinking about the bugfix in patch 2.
I don't mind the idea, just though it needs more explaining in the
commit and updated subject.
I don't know the reset path nearly well enough to comment on 2/2.
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
