[Mesa-dev] [PATCH 0/4] XA composite and perf improvements

[Rob Clark]

On Tue, Apr 1, 2014 at 12:37 PM, Thomas Hellstrom <thellstrom at vmware.com> wrote:
> On 04/01/2014 06:18 PM, Rob Clark wrote:
>> On Tue, Apr 1, 2014 at 12:14 PM, Zack Rusin <zackr at vmware.com> wrote:
>>> ----- Original Message -----
>>>> From: Rob Clark <robclark at freedesktop.org>
>>>>
>>>> While still more of a stop-gap solution (until glamor) for freedreno,
>>>> with these few relatively simple changes I get a pretty big performance
>>>> boost (~40%) for xf86-video-freedreno.
>>> That looks great to me. Nice work.
>>> But to be honest the only thing I remember about this code is that it has been written in C and I'm probably like 40% certain of that.
>> heheh, fair enough
>>
>> well, let me know if there is anything I can do to help test (don't
>> suppose you have a linux vmware player image somewhere setup with
>> everything to build/install latest mesa? ;-))
>>
>> I do have a branch here that you could pull if you want to test:
>>
>> https://urldefense.proofpoint.com/v1/url?u=https://github.com/freedreno/mesa/commits/xa&k=oIvRg1%2BdGAgOoM1BIlLLqw%3D%3D%0A&r=l5Ago9ekmVFZ3c4M6eauqrJWGwjf6fTb%2BP3CxbBFkVM%3D%0A&m=HhXBh%2BiRTHozYnKRPiENV4eZDXlB4PW%2Bk1blIb9xOhg%3D%0A&s=713e8e2e4558ecd2d53d9874dc7d3e19f13dbb2834b05d98cadb49208acf6089
>> git://github.com/freedreno/mesa.git xa
>>
>> BR,
>> -R
>
> Hi, Rob!
>
> Nice work!
>
> I can do some testing later this week. In our xorg driver / DDX, we have
> a special "rendercheck" mode that disables the migration heuristics and
> always enable hardware acceleration. Although unfortunately I don't
> think there are any solid fill hooks.
> I'll also try to review the patches as soon as I have some time.
>
> some quick questions:
> Did you run the newly enabled accelerated operations through "rendercheck"?

not yet.. I suppose I should.  I know I'm hitting the new operations
because for each solid mask and solid src it took me a couple tries to
get right, and the corruption initially was pretty obvious.

> What does the performance look like on the "modesetting" driver with
> shadow framebuffer enabled, compared to freedreno with acceleration enabled?

I'm working on getting some better cairo-perf-trace #'s, but for
gtkperf goes from ~46-47s before to ~26-27s.  Pure sw
(xf86-video-fbdev, but -modesetting should be same) was giving me
~21-22s.  So we are not quite as fast as sw yet (if you have cached
buffers).  About 3.5s worth of that difference was from the circles
test, which I don't care so much about.

BR,
-R

> Thanks,
> Thomas
>
>>
>>> z