Profiling

DSight is the DNNDK performance profiling tool. It is a visual performance analysis tool for neural network model profiling. The following figure shows its usage.

Figure 1: DSight Help Info

By processing the log file produced by the N2cube tracer, DSight can generate an html file, which provides a visual analysis interface for the neural network model. The steps below describe how to use the profiler:

  1. Set N2Cube to profile mode using the command dexplorer -m profile.
  2. Run the deep learning application. When finished, a profile file with the name dpu_trace_[PID].prof is generated for further checking and analysis (PID is the process ID of the deep learning application).
  3. Generate the html file with the DSight tool using the command: dsight -p dpu_trace_[PID].prof. An html file with the name dpu_trace_[PID].html is generated.
  4. Open the generated html file with web browser.