Re: [Mesa3d-users] Mesa/Gallium Cell Driver 0.4 issues

[Dmitry N. M. <mae...@gm...>]

Thank you for your kind help, Brian!

2010/3/18 Brian Paul <br...@vm...>:
> Those may be run of the mill bugs in the driver.  The cell driver is
> incomplete and some things don't work properly.  The docs/cell.html has some
> status info.
>
> -Brian
>
> Dmitry N. Mikushin wrote:
>>
>> Brian,
>>
>> I'm afraid with 0.3 I also have issues. Not that hard that with 0.4,
>> but still. I tried to run Bullet physics visual demo and received
>> screen with black slices. I verified that this behavior is encountered
>> with mesa/gallium 7.7/0.3 turned on via LD_LIBRARY_PATH environment
>> variable, and everything is OK, when it is turned off. Including two
>> screenshots.
>>
>> D
>>
>> 2010/3/17 Brian Paul <br...@vm...>:
>>>
>>> Dmitry N. Mikushin wrote:
>>>>
>>>> The most recent version I've found, that does not have opcode error
>>>> for me is gallium 0.3 in mesa 7.7, the previous one.
>>>
>>> In between other things I'm doing a bisect on the 7.8 branch to find
>>> where
>>> things broke.  I'll try to fix it.
>>>
>>>
>>>> The following
>>>> question might by quiet stupid, but how can I see the performance
>>>> difference between cell-gallium and general mesa? Running glxgears
>>>> Enlightenment on PS3 shows around 105 FPS for both, so I really wonder
>>>> what can benefit from spu acceleration, and how to see that. I'm
>>>> mostly interested in boosting OpenGL.
>>>
>>> First, be sure you're using the driver you expect with glxinfo (see the
>>> "renderer" value).
>>>
>>> Be aware that SwapBuffers() is pretty slow with Cell so you'll never get
>>> super high frame rates.
>>>
>>> Where the Cell driver shines is in non-trivial shaders.  If you compare
>>> the
>>> speed of some of the demos in progs/glsl/ the Cell driver should be much
>>> faster that than the other software drivers.
>>>
>>> -Brian
>>>
>>>
>>>
>>> ------------------------------------------------------------------------
>>>
>>>
>>> ------------------------------------------------------------------------
>>>
>
>