Im using OpenCL to optimize a kernel on an integrated Adreno GPU. In order to tweak parameters for the kernels, I want to query GPU metrics like GPU utilization for different kernels, FLOPS, register pressure of work-items.
Besides the Snapdragon Profiler and the Android GPU Inspector, which both won't reveal any information about what (library)-functions are causing the measured values, I could not find any other ways to query information about the performance of the kernels. The Android Profiler in Android Studio does not give any information about what is happening on the GPU.
Are there other ways to efficiently query more information from the GPU for specific kernels?