Comments on: Implementing a simple circular buffer

By: Steve Eddins

Steve Eddins — Thu, 10 Dec 2009 16:12:03 +0000

Remco—I recently implemented a double-ended queue class using a circular buffer. Because I subclassed handle, no data copies are made. My push_front and push_back (don’t remember if I actually used those names) manipulate an internal circular buffer directly. A copy gets made only when the buffer gets full and has to be resized, but that happens in a C++ implementation as well.

By: dhull

dhull — Thu, 10 Dec 2009 15:24:05 +0000

@Remco,

This is the kind of thing that makes an excellent MATLAB Central File Exchange entry if you are willing to share.

Thanks,
Doug

By: Remco

Remco — Thu, 10 Dec 2009 15:06:38 +0000

Hi All,

I am not sure if this is still relevant but yes the discussed implementation is very inefficient. I have created a buffer class in matlab because I need multiple
circular buffers. This buffer class implements the buffer, display, and subsref functions in matlab code and the push_front and push_back functions in c code. This is because you cannot pass arguments by reference in Matlab. Matlab only makes a copy of the arguments when the arguments change inside a m file. But since you add an element to the buffer Matlab does make a copy. In a c mex implementation this does not happen and one can efficiently add a new element. Also I use a double length circular buffer to allow for linear adressing instead of circular buffer adressing

By: dhull

dhull — Mon, 11 May 2009 13:37:48 +0000

I profiled this with a really large amount of data, more iterations and longer buffer. That part of the video is on the “cutting room floor”. Basically, there is a negligible performance hit. To be a real test, it would need to be done in context to see if this architecture is a problem.

This video is in keeping with one of the central themes of my programming style: “Avoid premature optimization”. If I were to try and avoid the churn that this circular buffer goes through, the complexity and readability would suffer.

-Doug

By: Daniel Armyr

Daniel Armyr — Mon, 11 May 2009 07:23:29 +0000

Like Petter, I wonder if this should be considdered an efficient implementation or only an easy one?

I use a few rinbuffers here and there and they are all implemented like this, because it is easy. However, the line buffer = [newValue buffer(1:end-1)] gives the optimizer in me the shivers. Is that just me?

By: Petter

Petter — Sat, 09 May 2009 14:22:19 +0000

Wouldn’t that solution force the entire array to be reshuffled every time a new value is added?