Low-Cost Vectors, Colors, Structs

[mattdesl]

Feature suggestion

I have some code that deals with geometry and graphics programming, where Vectors and Colors are often needed. However, I'd like to avoid allocating new memory, or using classes, if possible. I've noticed the performance, bundle size, and complexity (in terms of dealing with Wasm Memory) are much smaller and more manageable when just using variables, or even global variables, so I wonder how to achieve some of these things with AssemblyScript.

Example use case:

struct Vec2 {
  x:f32;
  y:f32;
}

function normalize (vec:Vec2):void {
  let len:f32 = vec.x * vec.x + vec.y * vec.y;
  if (len > 0) {
    let f:f32 = Mathf.sqrt(len);
    vec.x /= f;
    vec.y /= f;
  }
}

export function init ():void {
  let a:Vec2 = Vec2(0.5, 0.25);
  normalize(a);
  /* do something with a.x, a.y */
}

I don't really care for functions, memory management, or other features of classes, and I don't want my bundle size to bloat with each new class definition. Really I am just using this as a sort of tuple. Is this possible?

[Heath123]

I think this is a duplicate of #2254

[CountBleck]

Your proposed syntax wouldn't really be compatible with TS in any way. TS has a syntax for tuple types but #2254 (comment) (as @Heath123 pointed out) states the issues with using that and other possibilities.

Since you're also looking for performance improvements, maybe v128 is of interest to you. It can fit four f32s/two f64s and allows you to potentially make code faster by using SIMD ops.

[mattdesl]

Thanks. I am OK with any syntax really; or a decorator. Since this is a dupe we can close and continue discussion in the other topic.

In this case, SIMD is interesting but for my use case it will involve a lot of extracting/loading/storing which makes me wonder if it would be as performant, even something like normalize() I have not found a simple way with AssemblyScript's SIMD to achieve f32 dot (or horizontal add after f32x4.mul), and then an extract to compare against zero.

[CountBleck]

SIMD would likely be more helpful if/when you loop over a bunch of vectors and do some calculations.

Anyways, I might as well close this then...