How can I improve this line of code perfomance wise?

When using the profiler to look at the performance of a function I need to work with, I saw that one line takes up much time

  invalidIndex = any((coords<ddims(1:3)) | (coords>ddims(4:6)),2);

So it just checks if the point contained are within the bounds. The size of coords is nx3

Is there any way this line can be improved perfomance wise. I couldn't think of anything else...

[Walter Roberson]

You could try unvectorizing it and seeing how much difference that makes:

invalidIndex = coords(:,1)<ddims(1) | coords(:,2)<ddims(2) | coords(:,3)<ddims(3) | coords(:,1)>ddims(4) | coords(:,2)>ddims(5) | coords(:,3)>ddims(6);

My expectation is that this would be notably slower... but that you would have to measure the timing to be sure.

[lit]

actually, it does improve performance, so thank you very much!

Answer 1

[Karan Singh]

Hi @lit,

In addition to what Walter stated, you can slo try to precompute bounds to avoid repested indexing.

In the original code, every time the condition (coords < ddims(1:3)) | (coords > ddims(4:6)) runs:

1. ddims(1:3) and ddims(4:6) are accessed twice (one for < and one for >).
2. MATLAB performs indexing operations multiple times, which can be slow for large datasets.

lowerBounds = ddims(1:3);
upperBounds = ddims(4:6);

invalidIndex = (coords(:,1) < lowerBounds(1)) | (coords(:,1) > upperBounds(1)) | ...
               (coords(:,2) < lowerBounds(2)) | (coords(:,2) > upperBounds(2)) | ...
               (coords(:,3) < lowerBounds(3)) | (coords(:,3) > upperBounds(3));