[Intel-gfx] wakeups/sec from i915 gen6_pm_rps_work

Jesse Barnes jbarnes at virtuousgeek.org
Mon Nov 4 21:00:41 CET 2013


On Mon, 4 Nov 2013 19:27:31 +0000
"Chilukuri, Harita" <harita.chilukuri at intel.com> wrote:

> Hi All,
> 
> I am running Aquarium workload (WebGL) on Chromium OS. From the powertop data, I do see 277.9 overall GPU ops/second and 166.6 ops/sec from gen6_pm_rps_work.
> 
> Are the above ops/sec in normal range? Can someone let me know what the primary purpose of this function is and how I can reduce the wakeups from it?

We kick off that work when we get an interrupt from the GPU requesting
a frequency increase or decrease.

We could reduce the frequency of the interrupt (or get rid of it
entirely) a couple of different ways:
  - adjust the thresholds and/or evaluation intervals to make freq
    changes less frequent (and thus less responsive to changes in load)
  - make the changes sloppy and batch them up (effectively the same
    thing as above but in sw, and we'd still take the interrupts)
  - use a polling scheme to evaluate busyness, and adjust the freq at a
    fixed rate at specific times, maybe using a 60Hz timer or something
    while the GPU is busy

-- 
Jesse Barnes, Intel Open Source Technology Center



More information about the Intel-gfx mailing list