feat: included Matrix value setters

[alestiago]

STATUS: Draft

Description

Adds setters r0c0, r0c1, r1c0 and r1c1 to Matrix2, Matrix3 and Matrix4.

final m = Matrix2.zero()..r0c0 = 2;

/*
[ 2, 0 ]
[ 0, 0 ]
*/

Why?

Sometimes it is very inconvenient and verbose to create Matrices.

The value setters aim to improve the developer experience.

// Before:
Matrix4(2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);

/// After:
Matrix4.zero()..r0c0 = 2;

Additional Context

#265
#265 (comment)
#265 (comment)

[alestiago]

Do we want to provide getters to?

One could argue that get r0c0 is more performant than setEntry(0, 0, 1). (cc: @rakudrama )

[spydon]

Does this provide any benefits over setEntry(x, y, value)?
Won't this just be another way to do the same thing?

[alestiago]

Implementation of setEntry:

  void setEntry(int row, int col, double v) {
    assert((row >= 0) && (row < dimension));
    assert((col >= 0) && (col < dimension));

    _m2storage[index(row, col)] = v;
  }

  int index(int row, int col) => (col * 2) + row;

Implementation of r0c0:

set r0c0(double value) => _m2storage[0] = value;

I believe r0c0 is less verbose than setEntry and it will be more performant than setEntry.

The addition of these setters depends on if we consider performance important. According to @rakudrama "writers of math code often are concerned with performance.".

[spydon]

If the index would be calculated directly inside of setEntry instead of being a separate method I doubt that this would give any noticeable performance benefits (might not even be noticable with the extra method call either). But it would be very interesting to see a benchmark!

[alestiago]

Good suggestion @spydon!

Here is what I got:

So r0c0 = 1 is faster than setEntry(0, 0, 1).

Results varied from time to time but r0c0 was always faster.

I'm running using MacBook Pro M1 (macOS 12.2 21D49 darwin-arm).

Hope this helps.

Code

The below snippets show the code used to derive the above table.

final stopwatch = Stopwatch()..start();
for (var i = 0; i < 100; i++) {
  Matrix2.zero().setEntry(0, 0, 1);
}
stopwatch.stop();
print(stopwatch.elapsedMicroseconds);

final stopwatch = Stopwatch()..start();
for (var i = 0; i < 100; i++) {
  Matrix2.zero().r0c0 = 1;
}
stopwatch.stop();
print(stopwatch.elapsedMicroseconds);

[spydon]

Thanks for the benchmark, have you moved in the index calculation to be inside of setEntry so that it doesn't have to do that unnecessary extra method call?

[alestiago]

No, the benchmark is with the code currently provided in this PR.

I'm intrigued to know the difference without the index call. I will look into it soonish.

[alestiago]

Sorry for the recent inactivity, I've been out of the office. My plan is to look into this in the upcoming week soon.

[johnmccutchan]

Hi,

Is this pull request still active?

[alestiago]

Hi @johnmccutchan !

I stopped developing it due to lack of contribution feedback.

I can have a look at it and finish what is required here if the feature is willing to be accepted.

[johnmccutchan]

Please re-open if you refresh your patch.