[Intel-gfx] Framebuffer compression on GM45 disabled with KMS

[Pedro Ribeiro]

On 1 May 2010 17:42, Pedro Ribeiro <pedrib at gmail.com> wrote:
> On 1 May 2010 17:22, Jesse Barnes <jbarnes at virtuousgeek.org> wrote:
>> On Sat, 1 May 2010 01:01:12 +0100
>> Pedro Ribeiro <pedrib at gmail.com> wrote:
>>
>>> On 1 May 2010 00:06, Jesse Barnes <jbarnes at virtuousgeek.org> wrote:
>>> > On Fri, 30 Apr 2010 13:04:23 +0100
>>> > Pedro Ribeiro <pedrib at gmail.com> wrote:
>>> >
>>> >> Hi all,
>>> >>
>>> >> My Xorg.log shows
>>> >>
>>> >> (**) intel(0): Kernel mode setting active, disabling FBC.
>>> >> (**) intel(0): Framebuffer compression disabled
>>> >>
>>> >> Is this normal? I'm trying to lower power consumption for my
>>> >> laptop... Enabling FBC should do it?
>>> >>
>>> >> I have
>>> >> fbpercrtc=0
>>> >> modeset=1
>>> >> powersave=1
>>> >> lvds_downclock=1
>>> >> enabled on the i915 module.
>>> >
>>> > It's just a stale message, I'll remove it.  FBC will be enabled by
>>> > your kernel if possible (you can check /sys/kernel/debug/dri/0/ for
>>> > status info if you have debugfs mounted).
>>> >
>>> > Jesse
>>> >
>>>
>>> Thanks for the heads up, actually measuring the power consumption
>>> between KMS and UMS with FBC enabled it seems that KMS wins by a very
>>> slim margin.
>>>
>>> Its a great job you guys are doing with this driver!
>>> I see it improve steadily on every kernel release. The only things I
>>> still miss is the render reclock support which was removed in 2.6.32.4
>>> - it worked wonderfully in my machine, reducing power consumption by
>>> 2w when idle; and multiple ring buffer support(for libva H.264) which
>>> seems to be coming in the Q3 this year :-)
>>
>> 2W!!?  If so it would be worth resurrecting for you in the form of a
>> boot option or something.  The main problem is that it's not very
>> general; may chips will lock up when we try to reclock them this way.
>> But enabling it by force on specific machines is probably ok.
>>
>> Jesse
>>
>
> Yes, the difference was that big with the monitor on DPMS off on my
> X4500. Don't know if it is related to it, but that it was what I
> observed. Overall, at least 1w with the computer sitting idle.
>
> It was rock-solid till 2.6.32.4. I was annoyed when the patch got
> removed and reversed the patch in 2.6.32.5 but it caused serious
> graphic corruption and lockups. I haven't tried again in any of the
> newer kernels (.33 or .34).
>
> Regards,
> Pedro
>

A follow-up: decided to revert "drm/i915: remove render reclock
support" on 2.6.34-rc6 (only conflicts on chunk 3 but easily solvable)
- I got at least 1w savings!
This is impressive and I've done a few hibernate/thaw cycles as well
as VT-switching and back to Xorg and no corruption at all (doing this
on 2.6.32.5 with the patch reverted would cause corruption).

I'm going to test it for a while to be sure, but it seems to be
working pretty well. I'll keep you posted if it starts going nasty.

Pedro