[vtk-developers] OpenGL Extension loading with OSMesa?

[Burlen Loring]

On 4/17/2013 4:01 PM, Burlen Loring wrote:
> On 4/17/2013 3:05 PM, Kevin H. Hobbs wrote:
>> On 04/17/2013 05:13 PM, Burlen Loring wrote:
>>> it will be interesting to see the dashboards now that some of the
>>> extension dependent code will run. I wouldn't be surprised to see some
>>> failures. The surface LIC doesn't run correctly, for one thing 
>>> there's a
>>> bug in Mesa 9.1.1 where GL_ARB_texture_float isn't built when 
>>> requested.
>>> After patching so that it's built I'm finding other issues with the
>>> shaders and render to texture, the result textures are empty/clear.
>>>
>> Indeed very interesting.
>>
>> I also regularly build VTK against mesa from git. The OSMesa
>> driver was just switched to the Gallium llvmpipe driver which is
>> built for speed.
>>
>> As a plus with Gallium I can set a shell variable to change the
>> OSMesa driver at runtime to the softpipe driver which is built
>> for correctness.
>>
>> I've already been able to tell the Mesa developers that a VTK
>> test passes with Gallium on NV95 (err whatever my graphics card
>> is) and Gallium softpipe, but fails with gallium llvmpipe whether
>> or not OSMesa is used.
>>
>> I will be _very_ interested in tomorrow's dashboard.
>>
> that's very exciting news! I can't wait to try the lvvmpipe backend.
>
> This dashboard 
> (http://open.cdash.org/viewTest.php?onlyfailed&buildid=2878681) was 
> run with the following OS Mesa config and patch for texture float ext. 
> A few tests do fail
>
> ./configure --prefix=/work/PV/apps/Mesa-9.1.1 --enable-shared 
> --disable-static --enable-texture-float --enable-osmesa 
> --with-osmesa-bits=32 --disable-gallium-llvm --disable-dri 
> --disable-egl --disable-glx --with-gallium-drivers= --with-dri-driver 

Kevin, Thanks for posting about the llvm backend for os mesa! that is 
such a massive improvement. A build using the latest OS Mesa cloned 
yesterday resolves nearly all of the test failures. Only 5 tests failing 
and one of those looks like it just needs an additional baseline. 
http://open.cdash.org/viewTest.php?onlyfailed&buildid=2880693