Bug#538810: video-radeon: direct rendering: graphics deceleration?

[Michal Suchanek <hramrach@centrum.cz>]

2009/7/31 Michel Dänzer <daenzer@debian.org>:
> On Fri, 2009-07-31 at 00:13 +0200, Michal Suchanek wrote:
>> 2009/7/30 Michel Dänzer <daenzer@debian.org>:
>> > On Thu, 2009-07-30 at 01:36 +0200, Michal Suchanek wrote:
>> >>
>> >> The majority of time is spent in:
>> >> (with -fps) 181333   53.6798  radeon.ko                radeon.ko
>> >>          radeon_do_wait_for_idle
>> >> (without -fps) 287349   59.3526  radeon.ko                radeon.ko
>> >>             radeon_freelist_get
>> >
>> > This indicates the GPU is the bottleneck, but I'm not sure why it would
>> > be that slow... though one thing I notice now is that the card only has
>> > a 64 bit wide memory bus, that could be the bottleneck. What kind of
>> > numbers does
>> >
>> > x11perf -copywinwin500 -aa10text -repeat 1
>> >
>> > give? (Preferably without a compositing manager running)
>> >
>>
>> This is the output:
>>
>> x11perf - X11 performance program, version 1.2
>> The X.Org Foundation server version 10602901 on :0.0
>> from heretic
>> Fri Jul 31 00:06:24 2009
>>
>> Sync time adjustment is 0.1073 msecs.
>>
>> 3200000 reps @   0.0018 msec (554000.0/sec): Char in 80-char aa line
>> (Charter 10)
>>
>>    8000 reps @   0.6963 msec (  1440.0/sec): Copy 500x500 from window to window
>
> Okay, that's not much worse than here, so it seems like your hardware
> should be capable of similar performance in hypertorus as well.
>
>
>> The next thing to try might be to install the drm modules that came
>> with the mesa library.
>
> There's no such thing. If you mean drm-modules-source, that's deprecated
> in favour of the DRM modules in the kernel.

aren't the modules provided with mesa newer?

>
>
>> I would expect that like CPU operations the GPU operations can be
>> optimized so a later code could have better results.
>
> Indeed, it certainly can't hurt to try upstream Mesa Git.
>
>
>> Unfortunately. unlike Intel ATI did not hand out optimization manuals
>> for their chips so there is not much hope in improving the
>> performance.
>
> I'm not sure that's an accurate comparison of the documentation provided
> by these vendors, but anyway I don't think the low performance of
> hypertorus on your system is representative, there just seems to be
> something weird going on there.
>
> BTW, what's the number of polys displayed by hypertorus?

I have 2,080 polys.

Thanks

Michal