3 ms·
Unofficially, I still have it as a bit of a personal goal to get the GPU going by the end of the year. We already have the userspace side passing >90% of the GL
by marcan_42 5y ago
Unofficially, I still have it as a bit of a personal goal to get the GPU going by the end of the year. We already have the userspace side passing >90% of the GLES2 tests on macOS, it's just the kernel side that's missing, so it's not as far as it might seem.
- rowanG077 5y agoThat would truly be amazing! I don't really expect anything official, I mean software is hard to estimate as it is. Let alone when you are reverse engineering where you never know what kind of insanity you will find. So for GPU we would need 3 pieces? Alyssa's work on reverse engineering the protocol and writing a driver to submit the correct commands to the GPU. A working DCP driver to handle actually displaying rendered frames. And finally a driver to be able to submit the commands to the GPU. And work on the last of those has not started yet, right? It's hard to piece together how everything fits for me.
- marcan_42 5y agoThere's a whole different coprocessor tied to the GPU for the rendering (AGX), though the generic RTKit/mailbox stuff is the same as it is for DCP. That will largely involve command submission, memory management, and (a relatively big unknown) GPU preemption. In principle, the GPU kernel side should be quite a bit simpler than DCP, since display management is quite a bit hairier than (relatively) simpler GPU operations; most of the complexity of the GPU is in the userspace side. But we won't know until we get there.
- jtl999 5y agoHow exactly do you test your own GPU "rendering" under macOS?