ImgUtil.copy RandomAccessibleInterval to an IterableInterval, multithreaded

[acardona]

New utility method ImgUtil.copy(RandomAccessibleInterval, IterableInterval, int) for multi-threaded copying with int threads.

Addresses the issue of copying "images" that are actually transformed views using linear interpolators and others, which can be costly and, from a scripting framework, it is not entirely trivial to partition a RandomAccessibleInterval for multi-threaded processing.
While it could be done via Views.interval, it is simpler to use a utility method to do it all.

[maarzt]

The copy method in this PR is rather complex, because it needs to deal with all the details of multi-threading. This motivated me to add multi-threading to LoopBuilder #250. The multi-threaded LoopBuilder allows to dramatically simplify the copy method:

public static < T extends Type< T >> void copy(
	final RandomAccessibleInterval< T > src,
	final RandomAccessibleInterval< T > dest )
{
	LoopBuilder.setImages( src, dest ).multiThreaded().forEachPixel(
		(s, d) -> d.set( s )
	);
}

BUT: The destination's type is now an RandomAccessibleInterval rather than an IterableInterval.

@acardona I would like to know your opinion: Is there a benefit of choosing IterableInterval as destination type? What's the motivation for having n_threads as a parameter?

[acardona]

Hi @maarzt,

1. IterableInterval is preferable if the destination type was IterableInterval. Otherwise wrapping is needed, which is very undesirable.
2. n_threads is needed when running e.g. 4 only threads doing an expensive operation (due to memory constraints) but then the copying can fully use e.g. 32/4 = 8 threads for the copying of each result image into e.g. an ArrayImg. (This is pulled from a real world situation).

The ImgUtil.copy serves a real need in scripting languages that can't optimize loops like for example jython.

[ctrueden]

The pyimagej project would greatly benefit from a performant single-line RAI-to-RAI (and/or RAI-to-II) copy method, which does not require writing any lambdas.

@maarzt Does a way currently exist to achieve such a performant single-line copy operation currently? If not, can we get something added to the library for this soon? We will need it for the pyimagej 1.0.0 release.

See also imagej/pyimagej#46.

[ctrueden]

Thanks, @maarzt!