Rotation calculated for spherical particles

[rubenhorn]

Description

Optimize rotation of zero vectors (spherical molecule simulations).
This case is handled explicitly to reduce the number of multiplications that are performed.
$\forall q\in\mathbb{H}:rotate(q,\vec{\scriptstyle 0})=\vec{\scriptstyle 0}$

For Quaternion::differentiate all summands in all equations have a factor from the input vector.
Thus, $\forall q\in\mathbb{H}:differentiate(q,\vec{\scriptstyle 0})=\vec{\scriptstyle 0}$

Resolved Issues

• Rotation calculated for spherical particles #267

How Has This Been Tested?

As described in #267, the Argon and CO2 simulations were run using both the original and modified codes generating .xyz-files.
There was no difference between the .xyz-files which indicates that output of the computations is unchanged.

[rubenhorn]

Other optimizations may also be implemented 👉 Do NOT merge yet!

[rubenhorn]

In a separate, very naive sequential test using C++/gcc, avoiding 100M calls to sqrt(1.0) for saves 10ms.
This is not much, but it may be notable when running very large simulations (particle count/time steps) without AutoPas.

[HomesGH]

In general, it looks good besides a few smaller issues.
You could also have a look at a simulation with spherical particles using a profiler, e.g. valgrind, to identify the Quaternion functions with the highest impact.

[rubenhorn]

Using the GNU profiler (gprof) for the argon example (config_noAP_ALL.xml with <steps>10000</steps>, no output):

Method	baseline (103.45s)	optimized (95.71s)
Quaternion::rotate	0.56s (0.79%)	0.19s (0.30%)
Quaternion::rotateinv	0.21s (0.30%)	0.08s (0.12%)
Quaternion::differentiate	0.06s (0.09%)	0.10s (0.16%)

Reverting the change of Quaternion::differentiate .

[FG-TUM]

Just some small cmake stuff

[FG-TUM]

Nice