Re: [Dri-users] About AGP multiplier in DRI drivers

[Roland S. <rsc...@hi...>]

Alex Deucher wrote:
> On 7/24/05, Philipp Klaus Krause <pk...@sp...> wrote:
> 
>>V anger Peskolazovich schrieb:
>>
>>>>Option      "EnablePageFlip" "true"
>>>>Option      "AGPFastWrite" "true"
>>>
>>>didn't allow me even to init 2 the system, I had to reboot..
>>>
>>>
>>>
>>>>Option      "AGPMode" "4"
>>>
>>>That improved perfomance. Still it's very slow compared to ATi proprietary.
>>>
>>>Any other speed-up tweaks?
>>
>>Texture compression (S3TC). You have to get an extra library for that.
>>Here in europe it's legal (as in Cuba or Venezuela), but in the USA
>>there's software patents.
> 
> 
> Also tiled framebuffer adds another huge boost.

so does hyperz (hyperz=true or use driconf). For ut2k4 you don't 
necessarily need the external library, as ut2k4 only uses precompressed 
textures, so you can force-enable s3tc without the external library 
(force_s3tc_enable=true or use driconf).
I'll have to disappoint you however, none of these options will help 
(well s3tc likely will if you use the highest texture quality setting as 
the huge slowdowns you might see sometimes (texture thrashing) should 
disappear).
The reason for the bad ut2k4 performance is that the driver doesn't 
support GL_ARB_vertex_buffer_object (well it does technically, but the 
implementation isn't faster than just using standard vertex arrays) plus 
  that DrawElements code is really slow. You can see more about that 
issue here, 
http://marc.theaimsgroup.com/?l=mesa3d-dev&m=108939948631512&w=2.

btw I've just noticed recently that ut2k3/ut2k4 shadow projectors (for 
instance spinning fans in dm_asbestos, tree shadows in dm_antalus) got 
broken again, I haven't yet figured out when that happened though.


Roland