With Mesa cb6be5c8c05fa1af20ebd4f014d686244826f987, the ES 3.0 CTS
has these failures on SNB:
The CTS was run with precise_trig enabled.
Did a little digging tonight with Ken. The mix test fails all seem to be because of some imprecision in LRP. Ken found some documentation that indicates it may have something to do with the fact that the destination is an MRF. The cube tests are all because we are failing to set the "Base Mip Level" parameter in SAMPLER_STATE. It turns out we've been failing to set this since the dawn of time but we've never noticed until now. I've got a patch that I'll send once Jenkins is done with it.
I have a branch which fixes the mipmap and shader tests here:
However, there seem to be some image_external tests still failing:
And those appear to be all that's still failing in a full CTS run.
(In reply to Jason Ekstrand from comment #2)
> However, there seem to be some image_external tests still failing:
> And those appear to be all that's still failing in a full CTS run.
Never mind... I just needed a CTS branch.