Most of the past few weeks have been focused on improving the V3D driver’s dEQP/VK-GL-CTS conformance rates. The piglit infrastructure for the conformance tests is strange, and there doesn’t seem to be a way in tree to do an approximation of a proper conformance run, so I’ve had to glue together the deqp and khr_gles profiles, with some overrides for generating the khr_gles test list to be run (since we don’t seem to have a way to run a conformance mustpass list). There were lots of deqp failures when I started, due to not having run that suite before, but I’m down to a <.5% failure rate on simulation now.
On the testing front, thanks to Maxime we now have a vc4 testlist in the i-g-t suite. Hopefully the Raspberry Pi foundation can start using that to get some coverage of the vc4 driver when they update their kernel – as is, it seems my PRs mostly languish until Dom has time to hand-test them, which is absurd.
Boris has been working on supporting plane X/Y offsets (panning) with the T tiling format. He has it working on every other tile row, and I’m concerned that the HW may just be broken, since T scanout didn’t see a whole lot of testing as far as I know.
I wrote a patchset to core DRM to let drivers control the ordering of calling into bridges, which is important for DSI where you want to have the bridge prepare be before video packets are scanned out, but after the module is ready to send DSI transactions. This should let more DSI panels work with the Raspberry Pi, but the patch needs a rework after review.