Is there a way to improve the relations between float generic math methods and Vectors (mainly Vector2)

[DrkWzrd]

Maybe I didn't see the method if it exists, or maybe my vision is very naïve, but for now I lack for a method like "SinCos" or "SinCosPi" but returning a Vector2(cos, sin), to use it directly in vector multiplications and FMA without having to do anything.

I think there must be methods like

public static Vector2 AngleCosSin(float angleRad)
{
    var (sn, cs) = float.SinCos(angleRad);
    return new(cs, sn);
}

public static Vector2 XXXNamingIsHard(Vector2 center, float angleRad, float radius)
{
    var cosSin = AngleCosSin(angleRad);
    var radius = new Vector2(radius);
    return FusedMultiplyAdd(cosSin, radius, center);
}

[huoyaoyuan]

Vector2 is several layers upper than fundamental math types. Methods from fundamental type can't depend on Vector2, or know anything about its layout etc. Thus, such a method can only be provided by a custom extension.

[tannergooding]

Vector2 has a SinCos(Vector2 vector) method already, but it produces a (Vector2 Sin, Vector2 Cos) result since it allows you to compute for 2x floats simultaneously.

But yeah, this isn't something that in typical algorithms you often want to be computing multiple rather than shoving them into a single vector and where if you do need them combined, the code you've given above is what we'd end up having to do anyways

[DrkWzrd]

What about add 'CosSin' and 'CosSinPi' method variants in generic floating point interfaces and VectorXXX (this one returning the corresponding tuple VectorXXX), and create an implicit conversion (this can be done in extensions by consumer) between (float, float) and Vector2?

That way you could do

Vector2 cosSin = float.CosSin(angleRad);

And it will be an useful one liner.

[tannergooding]

That would be inconsistent layering and naming with other APIs, so it's not something we'd expose as proposed. It's also not something, functionality wise, commonly exposed by other vector math libraries either, nor something that is typically what you want for efficient use of the results in an algorithm. -- You typically end up either needing individual floats, which are then shuffled/permuted/inserted to the right positions or you need the simultaneous per component results.

It's trivial to create your own one liner or extension method if that's appropriate for your domain.