The processor is iMX6Solo (Wandboard Solo), my application decodes H264 streams in 1920x1088 resolution with VPU, and in case of certain streams decoding a frame lasts up to ~32ms. This wouldn't be too much, but displaying it with OpenGL (glTexDirectVIVMap) takes 15-16ms, so if I sum these up it's over 40ms which is the target for the framerate (25 FPS).
The OpenGL ES shader is so simple that I can't believe it's wrong. Vivante has native support for YUV textures, so the YUV --> RGB conversion is also performed by the GPU.
Somehow I need to improve the performance.
Can I easily boost CPU frequency e.g. to 1100 MHz just by changing the CPU freq work points in the kernel?
Is there any way to boost GPU/VPU clock?
I've seen in the kernel an option to increase VPU clock to 352 MHz, but it's said to be experimental. I'd be satisfied with a moderate (e.g.10-20%) increase in the clock as well, provided the machine remains stable.
(If not the GPU is the proper way to display my VPU frames, then what? IPU is way too slow. Shall I use G2D API? I haven't found any example code for that...)