https://bugs.freedesktop.org/show_bug.cgi?id=70009

--- Comment #4 from Pavel Ondračka pavel.ondracka@email.cz --- (In reply to comment #3)

(In reply to comment #2)

...

So is seems, that when I record a trace using r300g driver (RV530) and replay it using either llvmpipe or i965, it works fine. Probably this is just some issue in the r300g driver, uncovered by core mesa changes... EVE online apitrace uploaded here: http://pavel.ondracka.cz/EVE.trace

llvmpipe and i965 support more temps than r300 class hw so they don't cross the limit.

Yeah, thats why they show no errors and promblems when launching EVE directly, because wine detects this and generate valid shaders...

However the main problem is a little bit different. Citing one sample shader from bug 42514: The shader declares the array VC to have 245 elements: uniform vec4 VC[245]; then it tries to access the 246th element of the array: R0.xyz = (VC[245].xyz);

Also citing from wine bug 29146: Yeah, we should be better about this. The reason we don't already do this is that a shader like this may still work if it doesn't actually use the constants near the end that we remove. We should probably print a message mentioning something along those lines when such a shader fails to compile or link.

So when i generate the apitrace with r300g in already contains the invalid shaders, however llvmpipe and softpipe still handle things correctly (e.g. display majority of stuff) and also r300g did so before the mentioned bad commit.