Haxe Vector Math

Requires haxe 4.2+

Haxe vector math library that enables GLSL vector and matrix operations to compile in haxe

Features

GLSL Built-in Functions

All GLSL built-in functions are available after import VectorMath;

import VectorMath;

var direction = normalize(velocity);
var speed = length(velocity);

Vector and Matrix Constructors

Vectors and matrices can be constructed with the same patterns as GLSL

// new keyword not required
vec2(1, 2);

// single argument sets all components
vec2(0.0);

// vector composition
var color = vec3(0, 1, 0);
vec4(color, a);

// matrices can be composed from vectors
mat2(
	vec2(1, 0), // column 0
	vec2(0, 1)  // column 1
);

// a single argument sets the diagonal components (which creates a scale matrix)
mat2(scale);

Operator Overloads

All vector, matrix and scalar operations available in GLSL are supported

// vectors can multiply with scalars
vec2(1, 2) * 0.5;
mat2(1) * 0.5; // return a new mat2 after multiplying each component with the scalar

// vectors can be multiplied with compatible matrices
mat2(2) * vec2(3, 4);
var position = projection * view * model * vec4(xyz, 1.0);

// +=, *= etc work
var dt = 1/60;
var position = vec2(0.0);
var velocity = vec2(0.3, 0.4);
position += velocity * dt;

// component-wise comparison
vec2(1, 2) == vec2(1, 2) // true

Swizzles

Supports all possible read and write swizzle operations, including aliases rgba and stpq

vec4(1, 2, 3, 4).wzyx == vec4(4, 3, 2, 1); // true

// set xy components to (1, 2)
var position = vec4(0.0);
position.xy = vec2(1, 2);

// set rgb components to green
var color4 = vec4(1.0); // white
color4.rgb = vec3(0., 1., 0); // green

Performance

All operations are inlined so that vector objects are rarely constructed, instead vector operations compile to stack variables (meaning we can avoid the GC completely for most vector operations!). For example, the following compiled with -D analyzer-optimize

trace(length(mat2(2) * vec2(3, 4) * 0.5 - vec2(0.5)));

Generates

console.log(Math.sqrt(18.5));

(Cool right?)

Furthermore, because vector components are just stack variables these operations are easily auto-vectorized (SIMD) on compiled targets like cpp

Usage

Install with haxelib install vector-math

Add --library vector-math to your hxml commands

Then simply import the VectorMath class: import VectorMath;

import VectorMath;

function main() {
	var normal = normalize(vec2(Math.random(), Math.random()));
	var tangent = normal.yx * vec2(-1, 1);
	var transformedTangent = mat2(2) * tangent;
}

Add --dce full and -D analyzer-optimize to your hxml for clean output!

Q/A

• What are the differences to GLSL?

  • Direct vector assignment is reference rather than copy, that is: in the following statement: var ref = original, 'ref' represents the same underlying vector as 'original', whereas in GLSL it would be a copy. To copy a vector you can do any of var copy = original.clone(), var copy = vec3(original) or var copy = original.xyz
  • You can call methods via dot syntax in addition to regular calls, for example: vec3(1).length() and length(vec(1)) are equivalent
  • Boolean and integer vector types are not yet included (bvec and ivec). These may come in the future if there's a desire for them

• Which haxe targets does this work on?

  All haxe targets are supported with the exception of macros, this is an optimization to improve compile-time performance

• Can this be used to generate shader code?

  Yes! @rainyt has developed a VectorMath -> GLSL translator for use with OpenFL: github.com/rainyt/openfl-glsl

  In the future I hope to work on a general VectorMath to shader translator to support multiple ouput shading languages

• What makes this different from HXSL?

  HXSL is the haxe-based shading language in Heaps, it has similar aims but different implementation – HXSL works at the syntax level and does not support autocomplete or compiling to CPU platform code (like js or C++), whereas VectorMath code is executable as regular haxe code as well as shader code

• Which specification is this based on?

  GLSL ES 1.0 Specification