Hello @vix,
Let me go through the points you raise one by one.
I believe that we are at a hardware limitation here.
The GPU on the iMX8MP is reasonably capable for the SoC it is in, but it is still very limited in area and power.
Let me give you an example from another application: Machine Learning.
The GPU on the iMX8MP can be used to run machine learning models, but it is much slower than the CPU on the iMX8MP for that purpose. You can view some numbers here.
This is expected, as with more fishes there are more meshes and as a consequence more calls done to the GPU.
These need to be made by the CPU and there is an overhead in doing so.
Looking at the logs, I don’t see any significant problems.
On the Weston logs you can see some invalid calls, but this should not lead to any problems.
On the Chromium logs, there are a few dbus errors which are unrelated to the hardware acceleration. There is also a failed attempt to create an EGL context, which means that the browser tried to create an EGL context with a specific configuration which was not valid. Afterwards, it will have created a valid context as a fallback.
Such behavior is normal and expected.
Best Regards,
Bruno