Profiling Routines

The following table shows profiling routines provided by NLCPy.

Profiling with FTRACE

nlcpy.prof.ftrace_region	Enables profiling with an ftrace region during 'with' statement.
nlcpy.prof.ftrace_region_begin	Begins an ftrace region.
nlcpy.prof.ftrace_region_end	Ends an ftrace region.

FTRACE is a performance analysis tool, which can obtain performance information such as the exclusive time and vectorization aspect on the VE. Please note that VE offloading overhead is included in performance information. For details of FTRACE, see PROGINF/FTRACE User’s Guide.

Note

VE functions in NLCPy are multithreaded. However, the ftrace.out includes performance information of only a master thread. To obtain performance information of all threads, it is necessary to compile C programs of NLCPy by using ncc with the option ‘-ftrace’. If needed, please download source programs of NLCPy from GitHub.

The following routines are deprecated since NLCPy version 2.0.0.

Start and Stop Profiling

nlcpy.prof.start_profiling	Starts profiling.
nlcpy.prof.stop_profiling	Stops profiling.

Get Result

nlcpy.prof.get_run_stats	Gets dict of NLCPy run stats.
nlcpy.prof.print_run_stats	Prints NLCPy run stats.