Branchless array initialization; faster rounding; other minor enhancements

[solarunes]

Array Initialization

Currently, all create() namespaced methods check glMatrix.ARRAY_TYPE to determine if the Array needs to manually be initialized to zero or not.

This can be avoided by creating a "zero-initializer class" that is a subclass of glMatrix.ARRAY_TYPE, if and only if it doesn't inherit a typed array.

Implementation

I added a subclass generator for Array-like objects that adds zero initialization to their constructor, then returns the new class. This subclass is available as a public field in the glMatrix namespace (ARRAY_ZERO_INIT_TYPE).

Invoking glMatrix.setMatrixArrayType(type) now does the following:

• set glMatrix.ARRAY_TYPE to type
• if type is Array-like but not a typed array;
  • create a subclass ArrayInitZero of type that initializes it's elements to zero
  • set ARRAY_ZERO_INIT_TYPE to ArrayInitZero
• otherwise, set ARRAY_ZERO_INIT_TYPE to type

Benchmarks

A full benchmark and comparison is available here.

Symmetric Rounding

The current glMatrix.round() implementation has branching- and modulo operations. None of this is required for symmetric rounding.

Implementation

I replaced the old glMatrix.round() with a much faster, single line implementation, only using methods of the built-in Math namespace.

Benchmarks

A full benchmark and comparison is available here.

Other Enhancements

I shortened a few methods and introduced modern built-in functions where applicable. These should hopefully provide a minor performance boost.