As we have seen, the Arm Compute Library can be profiled with Arm Streamline to study the performance of Machine Learning and Computer Vision applications.
This guide demonstrates how you can use Streamline to profile an example application. This is from the Arm Compute Library for the AlexNet Convolutional Neural Network on two different hardware platforms with different operating systems. The next guide you can read is Install and build Compute Library on HiKey 960.
Going forward, you can apply the methods shown here to use Streamline to profile your own machine learning applications to help you optimize their performance for running on Arm-based systems. You can use Streamline to report further information such as memory used, disk I/O, threads created, and sample-based function profiling. Furthermore, the Arm Compute Library makes use of the available Neon hardware to perform efficient image inference.