Re: [PATCH i-g-t 2/2] tests/perf_pmu: Add tests for engine queued stat

From: Chris Wilson <chris@chris-wilson.co.uk>
To: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>,
	Tvrtko Ursulin <tursulin@ursulin.net>,
	Intel-gfx@lists.freedesktop.org
Subject: Re: [PATCH i-g-t 2/2] tests/perf_pmu: Add tests for engine queued stat
Date: Wed, 22 Nov 2017 14:16:43 +0000	[thread overview]
Message-ID: <151136020348.9619.4464066235602917792@mail.alporthouse.com> (raw)
In-Reply-To: <151135871931.9619.7783014951265932928@mail.alporthouse.com>

Quoting Chris Wilson (2017-11-22 13:51:59)
> Throwing contexts into the picture was to be sure that it was counting
> across contexts. (I was thinking about the complexity with the timelines
> and the counter being on the engine, which I started worrying about.)
> Next up would be adding waiting requests and demonstrating that they
> aren't counted as queued.

Now thinking about something else...

If waiting is separate from queued, how do I keep track of my own queue
depth? i.e. how many requests have I submitted that are waiting before
being queued?

I think we need WAITING as well (a new metric). :|

engine->waiting++ in execbuf
engine->waiting-- in submit_notify() ?

The problem with WAITING for userspace load-balancing is that you do not
how many of those you can overtake (the only ones you are sure that you
have a dependency upon are those to your own timeline).

Hmm, I wonder if we are not just missing a trick in counting the dead
space for RT (i.e. learning from __intel_engine_get_busy_time()).
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx