https://bugs.winehq.org/show_bug.cgi?id=42863
--- Comment #10 from Matteo Bruni matteo.mystral@gmail.com --- (In reply to Guillaume Charifi from comment #9)
It's probably just plain wrong, but can't it be simulated by dot(textureGather(sampler, pos, refZ), vec4(0.25)) ?
Well, that's certainly incorrect in general. Weighting the four samples appropriately would be okay for the linear interpolation magnification case. That still leaves out the other cases though.
Another workaround could be to use an ad-hoc sampler or texture view overriding the texture levels / LOD to just enable level 0. It doesn't sound too appealing though, it would probably require some quite invasive and ugly changes to wined3d and the additional sampler / texture view would have to be used only for sample_c_lz i.e. we would waste a texture unit if the texture is also sampled without _lz in the same shaders. Not sure there are other realistic options though.
I don't know if Józef has already given some thought about a sensible fix.