Hi, everybody. I have a model, which running on mobile mali GPU and CPU. Now, I want to use AutoScheduler to auto tunning the model. I set the target as opencl, while suspect that it maybe not suitable for all operations in the model to run on GPU. So, is it possible to auto tune some operations on mali GPU, while some other operations on mali CPU? Currently i use the command target = tvm.target.Target("opencl -device=mali") to set the target, and it doesn’t support Heterogeneous compute. Is there anyway to support auto tune models both on CPU and GPU?
