[Intel-gfx] [PATCH igt v2 1/4] igt/perf_pmu: Tighten busy measurement

[Chris Wilson]

Quoting Tvrtko Ursulin (2017-12-15 18:16:52)
> 
> On 15/12/2017 16:10, Chris Wilson wrote:
> > Sleep for a known duration. In particular, CI once saw a measurement for
> > busyness greater than the intended batch_duration!
> > 
> > v2: Go back to starting pmu sampling outside of spinner; the GPU should
> > be idle.
> > 
> > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104241
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> > ---
> > -     assert_within_epsilon(val, ref, tolerance);
> > +     assert_within_epsilon(val, busy ? slept : 0.f, tolerance);
> > +     gem_quiescent_gpu(gem_fd);
> 
> Why you think it is needed to quiescent after each subtest? It would 
> make more sense to do so at the beginning of each, if needed, but I 
> thought it wasn't.

After the test makes sure that the spinner is completed before we move
on. We are trying to tidy up after the mess we made, and idle the system
before the next measurement.

Take your pick, before or after, we depend upon the system being in a
certain state. Or both. I picked both for the frequency subtest as the
transition to idle for it is a little more interesting (I was thinking of
gen6_rps_idle side-effects).
-Chris