Unity.Mathematics vs Mathf.
Unity’s Mathf is not 100% replacable with its burst-friendly Unity.Mathematics . Also Unity.Mathematics evolves all the time.
Unity’s Mathf is not 100% replacable with its burst-friendly Unity.Mathematics . Also Unity.Mathematics evolves all the time. By now you are reading this something might have changed.
Quaternion is available for use in jobs
quaternion type it in small q like this. It is in Unity.Mathematics
LookRotation’s difference
This is the result from the same input :
Quaternion.LookRotation (with Vector3)
Look (-1.6, 61.3, 162.4) Up (0.5, -0.9, 0.0) QuatWXYZ -0.2552253 0.04192121 0.1746541 0.9500519
Unity prints Vector3 like that but actually the number is in floating point position and contains more digits.
quaternion.LookRotation (with float3)
Look float3(-1.648314f, 61.3083f, 162.3523f) Up float3(0.4771588f, -0.8788171f, 0f) QuatWXYZ -0.2451455 0.04013683 0.172545 0.9387675
The difference :
Mathf -0.2552253 0.04192121 0.1746541 0.9500519
math. -0.2451455 0.04013683 0.1725450 0.93876750
This is actually because math ones expects a normalized vector! (No documentation or anything at all!)
quaternion.AxisAngle expects a radian
Don’t be stupid like me and think “angle” = degree
Where’s InverseLerp ?
It is called unlerp here. But! It is implemented without clamp and divide by 0 check.
Mathf ones works like this
public static float inverseLerp(float a, float b, float value)
{
if (a != b)
return math.clamp((value - a) / (b - a), 0, 1);
else
return 0.0f;
}
Unity.Mathematics ones works like this
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public static float unlerp(float a, float b, float x) { return (x - a) / (b - a); }
So if you use a = b it will explode, if you use x outside of 0~1 range it linearly interpolate outside of a and b.
SmoothStep’s difference
I have been tricked before, the definition of the last argument is not the same. That is, it is not 100% replacable.
- Unity.Mathematics : It is interpolated first before the clamp, the x value is expected to be between edge values a~b and not 0~1.
- Mathf : It is clamped first, the t value is expected to be between 0~1 then maps smoothly with from~to.
Shader language’s smoothstep works like the Unity.Mathematics version.
For your information, the 0~1 input is the way Lerp / lerp function universally works shader language or not. The old Unity’s SmoothStep is lerp-like (not the school idols). The new smoothstep is more inline with the general definition.
math.select(a, b, bool)
Which one do you think you get if the bool is true? I used to think like bool ? a : b but in a swapped order. That is wrong. Don’t be like me and spend 2 hours looking for bugs because of this.
Actually if the bool is false you get the first one. The good way to reason about this is that it “select” with a bit comparison. 0 comes before 1, and false is zero so you get the first one. Don’t think of a and b as true and false but as first and second.
How to get just one bool from == operator?
I was comparing int2 to int2 and thought it would return one bool but no, it is a component-wise comparison and no matter what I do it returns bool2
To get just bool you have to use Equals
(int2 is IEquatable<int2> but that does not necessary mean that == will use Equals method defined here)