I'm currently trying to find bottlenecks, as I'm currently not able to archive a full-hd rendering of a texture at 60fps when some additional small overlays are present.
During my timing tests I found out that calling glFinish with no OpenGL command in the queue it already takes 4-5ms before glFinish returns. Is this due to the context switch to kernel space? But even in case that's the main reason for a delay, it seems quite a bit long to me.
Apart from that, is there an easy way to enable the vivante profiler within the yocto toolchain? I couldn't find an option within the kernel menuconfig, although there seems to be a USE_PROFILER option somewhere to be defined.
At least this option is parsed in the KBuild file under drivers/mxc/gpu-viv.
I guess for the latest branch In Yocto the The 3.10.17-1.0.0 GPU has been updated to vivante 4.6.9p13 release. This patch updates the kernel driver to be compatible with it. And it includes the USE_PROFILES flags definitions in order to use and enable Vivante_PROFILER =1