Different execution time depending on selected schedule type

[MarcoMagriDev]

Hi,

I am encountering a strange behavior in the execution time of a function if the thread in which the function is run is scheduled using fifo.

Setup description

• A built library (let's call it my_library) in which i have a function defined (let's call it my_test_function)
• A plugin (let's call it my_plugin) that is including my_library and runs the function my_test_function in the update loop (the thread used by this plugin is rt_main)
• The execution time of the function is measured using:

  auto begin = std::chrono::high_resolution_clock::now();
  my_test_function();
  std::cout << "execution time: "
          << std::chrono::duration_cast<std::chrono::duration<double, std::micro>>(
               std::chrono::high_resolution_clock::now() - begin)
               .count()
          << " micros \n";

Problem description

If the scheduling type of the thread in which the plugins runs is set to fifo the execution time result ~100 micros. If instead the scheduling type is set to other the execution time result in ~5 micros (same execution time that i get on other PCs)

Note: no rebuild is performed between the change of scheduling type so the difference of behavior must be introduced by this setting

@EdoardoRomiti @alaurenzi have you ever encountered a similar problem? have you any idea how to solve this?

[alaurenzi]

I don't know what could cause this. I'd try not to use linux clock though, as Xenomai has its own wrappers. XBot::chrono::steady_clock provides an API for this.

[MarcoMagriDev]

Unfortunately I don't think the issue is related with the way in which execution time is measured since even If I leave only my_test_function() in the code the execution time difference can be noticed by the robot behavior and from the bar next to the plugin in the xbot2-gui.

I noticed this using a different function that normally takes more, like ~200 micros (in a non patched PC or under schedule: other) but generates overruns if scheduled using fifo

[alaurenzi]

Can you share the content of this function? Have you seen this behavior only with this specific function?