[Intel-gfx] [PATCH 07/27] drm/i915: Squash repeated awaits on the same fence

[Chris Wilson]

On Wed, Apr 26, 2017 at 03:36:19PM +0100, Tvrtko Ursulin wrote:
> But yes, sounds easy to do it from the idle worker. Just walk
> everything and prune when engine seqno has advanced past it?

Hmm. I was thinking that sounded like a great idea and then realised we
are storing context.seqno not global_seqno. :|

But... Yes, from idle we know that the fences are completed and so if we
should have a request for an old sync point, it will be ignored anyway
because the fence is completed. So we can just throw away the ht
entirely at that point with no loss of optimisations.

Ok, I see what you were trying to tell me about i915_fence_release now.
The problem with handling at fence_release is doing the reverse map to
all contexts using it.

Thanks,
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre