GPU offloading analysis is used to explore the traces of the function calls for a GPU compute- intensive application.
It provides an in-depth analysis of the HIP API calls, HSA API calls, order of kernel execution, time taken by each kernel to execute and subsequent Data transfer summary with per thread timeline. It also provides an aggregated list of Hottest kernels with timing metrics.
The AMD ROCtracer library provides support to capture the runtime APIs and GPU activities such as data transfer and kernel execution. This analysis helps to visualize the ROCr, HIP API calls, and GPU activities when a HIP based application is running. It is supported only with a launch application.