It could be very shocking if this had been the bottleneck in your program. Quaternions aren’t giant knowledge buildings that will use a lot reminiscence, so long as you are overwriting them in-place, not allocating new ones on the heap each body. (When you are allocating new ones on the heap on a regular basis, then the issue is your reminiscence allocation technique, not the maths that is extracting / encoding quaternion angle-axis information)
When you individually retailer a unit vector representing the axis, then every body you’ll be able to decide the quaternion parts like so, with out going by way of a constructor / manufacturing facility methodology:
float s = sin(angle * 0.5f);
float c = cos(angle * 0.5f);
quat.w = c;
quat.x = s * axis.x;
quat.y = s * axis.y;
quat.z = s * axis.z;
This is similar math these strategies are utilizing anyway – the one actual benefit you get right here is that since you understand you already normalized the axis, you’ll be able to skip doing that each body (whereas quat_create most likely normalizes no matter you go it, simply to make certain).
Storing the axis vector individually like this additionally ensures it does not accumulate rounding errors from being repeatedly encoded into the quaternion and decoded again out, which may trigger it to wander over time.
In an excessive case, if you happen to ever assemble a quaternion with an angle that is precisely a a number of of 360°, the sine of half that angle might be zero, so you will get the identification quaternion (1, 0, 0, 0) — fully erasing any details about the axis. That might imply there isn’t any solution to get better the axis from the quaternion knowledge on the subsequent replace. (Until, after all, your quaternion knowledge construction caches the axis individually as a part of its internals. However most quaternion knowledge sorts simply retailer 4 floats for w x y z.)
