The support for KHR_debug ranges in Nsight Systems looks to be super-useful for mapping OpenGL GPU work to high-level application tasks (e.g., here’s all the GPU time spent for rendering this user-defined chunk of the scene).
This works great in the Systems GUI, capturing both CPU and GPU time corresponding to my ranges. Note I enable “Collect KHR_debug trace”, “Collect GPU workload trace” and the sub-option “Enable KHR_debug GPU trace”.
Unfortunately, the CLI version only seems to capture the CPU time. Here, I’m using the --debug-markers CLI option. Maybe one problem is that I don’t have a CLI equivalent to the “Collect GPU workload trace” GUI option.
The GUI is fine for experimenting but I need the CLI version to integrate into our infrastructure, e.g., doing profiling as part of CI.