Optimise matrices copy when multiplying in vtkMath

[bruyeret]

Originally posted by @finetjul in #3009 (comment)

[finetjul]

[bruyeret]

The test with byteOffset is not enough, you also have to use byteLength
For example, if you want to multiply a matrix with a vector extracted from the matrix
The full test could be something like this:

const a = new Float32Array(25); // 5x5 matrix
const b = new Float32Array(a.buffer, 10 * 4, 5); // offset by 10 elements (4 bytes per element), take 5 elements

// Translation of the conditions:
// [ A and B are the same array ]
// OR
// [
//     A uses a buffer AND
//     B uses the same buffer AND
//     NOT ( A is completely before B OR B is completely before A )
// ]
const shouldCopy =
  a === b ||
  (a.buffer &&
    a.buffer === b.buffer &&
    !(a.byteOffset + a.byteLength <= b.byteOffset ||
      b.byteOffset + b.byteLength <= a.byteOffset));

// An equivalent, if you convert the `!(P || Q)` into a `!P && !Q`
const shouldCopyBis =
  a === b ||
  (a.buffer &&
    a.buffer === b.buffer &&
    a.byteOffset + a.byteLength > b.byteOffset &&
    b.byteOffset + b.byteLength > a.byteOffset);

This is to check if the two arrays overlap but in this particular case of matrix multiplication, you can for example use the first 16 elements of an array of size 100 as a matrix. So you could replace the x.byteLength with numberOfElements * x.BYTES_PER_ELEMENT.