Work flows algorithm profiling

Summary

Due to the need of investigation of algorithms performance issues, the proper method is introduced. It consists two to parts: special mantid build and analytical tool. Available for Linux only.

Mantid build

To build mantid version with profiling functionality enabled run cmake with the additional option -DPROFILE_ALGORITHM_LINUX=ON. Built in such a way mantid creates a dump file algotimeregister.out in the running directory. This file contains the time stamps for start and finish of executed algorithms with ~nanosecond precision in a very simple text format.

Adding more detailed information

More detailed information can be added to individual function calls in C++ with the help of std::chrono. At the top of the fuction create a “start time”

const auto startTime = std::chrono::high_resolution_clock::now();

Then at the bottom of the section being timed, add it to the profiling output

addTimer("createOutputWorkspacesSplitters", startTime, std::chrono::high_resolution_clock::now());

where createOutputWorkspacesSplitters is the name that will appear, similar to a child algorithm. The names in the report will be suffixed with 1 because the tool thinks they are the “default version” of a child algorithm.

An example of this can be found in FilterEvents.cpp.

Analysing tool

The mantid-profiler project provides the nice graphical tool to interpret the information contained in the dumped file. The project has more detiled instructions, but in general

python path/to/myscript.py & python path/to/mantid-profiler/mantid-profiler.py $!

The & puts the python process in the background. The $! is the process-id (PID) of the last process run and allows mantid-profiler to attach to the process.