Performance monitoring can also be done on a deployed system. The PS can access performance monitoring registers of the NoC components via the NPI bus.
The host software needs to do the following:
- Unlock the write protect of the NoC component.
- Configure the timebase.
- Enable the desired performance monitors.
- Read the appropriate counts.
- Calculate the desired quantities (bandwidth, latency).
This AMD GitHub tutorial provides a complete working example that addresses the following topics:
- Performance monitoring registers and how to use them.
- How to calculate bandwidth and latency.
- Provides a Tcl script to automate data capture using XSDB.
- Provides a bare-metal application that captures the performance metrics.