The display is not fluent when I use TVM opencl on RK3288


The device is rk3288 and system is android. When I run the algorithm with TVM (opencl device) on camera, the display has a Katon phenomenon. The display is not fluent. Does anyone ever had a similar phenomenon?


I do not have experience with using the camera + android simultaneously on an RK board (for the rk3399 we used Ubuntu GNU/Linux). Unfortunately in this case you may be at the mercy of the OpenCL + GPU scheduler on the board. Have you tried using a CPU device instead? I also am not sure how much the camera display stresses the GPU.


I tried rk3399 used ubuntu. When I use opencv with camera and run the inference with TVM gpu opencl. The same thing happen。
I see the Xiaomi/mace, they say:
UI responsiveness guarantee is sometimes obligatory when running a model. Mechanism like automatically breaking OpenCL kernel into small units is introduced to allow better preemption for the UI rendering task.”
Does TVM have some thing like these?