The Vitis AI Profiler is an application level tool that can help to optimize the whole AI application. The main purpose of Vitis AI profiler is to help detect the bottleneck of the whole AI application. For example, with the Vitis AI Profiler, you can profile the pre-processing functions and the post-processing functions together with DPU kernels' running status. If the profiling result shows that one pre-processing function takes very long time, it leads to a high CPU utilization and DPU cores wait for the CPUs to finish processing. In this situation, the pre-processing and CPU are the bottlenecks. If you want to improve the performance, try to rewrite this pre-processing function with HLS or OpenCL to reduce the CPU's workload.
- It is easy to use, this tool requires neither any change in user’s code nor re-compilation of the program.
- Visualize system performance bottlenecks.
- Illustrate the execution state of different compute units (CPU/GPU).
There are two components of this tool named VAI Tracer and VAI Profiler.
- VAI Tracer: Runs on target platforms and is responsible for collecting and reporting data to VAI Profiler for visualization.
- VAI Profiler: Runs on host x86 PC or local server and is responsible for analyzing andvisualization.