I am currently in the process if trying to do some benchmarking using acados through its Matlab interface in Windows 10.
What I have implemented is a simple MPC for bringing a 5 state unicycle model to the origin.
As the horizon is not overly long, acados solves the problem fairly quick, which leads to the problem I am facing.
It seems that acados is not reliably reporting the solver timings if they are below 1ms.
I am able to measure the time the call to the acados solve() function takes using Matlab’s tic/toc.
Acados, however, returns mostly 0 for the elapsed times, sometimes exactly 1ms.
To me, this look like the internal acados timer has a resolution of 1ms.
The model I am considering is probably to simple to reap the full benefits of acados, but I would still like to clarify if what I see is expected behavior.
Is there a way to get better resolution for the timers?
I would in particular be also interested in the timings for the different solver phases, i.e. QP solution, linearization, etc., which are show the same behavior.