Re: Profiler and GUI in Octave?

[Svante Signell]

I am now subscribed to this mailing list. Sorry for top posting last
time, it was not intentional.

On Fri, 2007-10-05 at 23:46 -0400, John W. Eaton wrote:
> On  6-Oct-2007, Svante Signell wrote:
> 
> | Please interpret my message as follows: I like the efforts you put into
> | the free software tool Octave, I use it every day at my home computers. 
> | My preference is as much free software as possible, especially Octave
> | since Matlab is very expensive for everybody except universities (where
> | I belong).
> 
> I prefer to focus on freedom rather than price.  Octave has certainly
> not been free (gratis) for me and others who have worked on it or
> funded its development.
> 
> | My question is more how difficult it is to implement such a thing.
> 
> Your initial message sounded like a feature request.
> 
> In any case, I'd suggest joining the list and starting a discussion
> about profiling.  There have been various threads in the past, and one
> that produced some experimental code (see:
> 
>   http://www.nabble.com/Octave-profiler%2C-tf3120434.html#a8645265
> 
> but this was not merged).

Regarding profiling and execution time I had a master student looking
into replacing the most common calls of complex matrix multiplication in
Matlab/Octave with C++ using MEX/OCT interfaces respectively in the PC
environment. The results were very discouraging, the execution times
built-in were faster for medium to large matrices in both Matlab and
Octave! The profiler in Matlab was used to find the lines of code with
the largest time was spent. My guess is that the replacement was on too
low level, at code-line level instead of function level creating a lot
of overhead eating up the gain of the compiled code.

The report will soon be published on my homepage in case you are
interested.

Regarding coding I'm not confident enough with C++, python or perl (only
C, fortran and maybe Pascal). Perhaps if you can help with where to put
the hooks for a profiler (without hurting performance) we can maybe
start looking into this. Preferably it would be a C++ and Octave
internals fluent person doing this.

> | Regarding a GUI this a secondary issue as least for us, computation
> | speed is more important. We are offering our help here, not requiring
> | immediate support for something not yet implemented! Is it feasible at
> | all?
> 
> What is it?  Yes, I'd say GUI and profiling tools are feasible, but
> they will both take some work.  Can you fund a project or write code
> to do implement the features you need and contribute them to the
> project so others may also benefit?  If not, then you may have to wait
> until someone else who can provide the funding or do the work decides
> that the things you want are also the things they want.
>
> | We can offer our help by trying our application in the Octave
> | environment as well as Matlab and compare performance.
> 
> As I see it, we need more people helping us improve Octave by writing
> code.  We already have enough people telling us how it fails in
> comparisons.

We might find a master student (EE, not CS) looking into this, but it is not 
so easy since we cannot offer funding (most people currently do their thesis in 
industry where they get paid) and most EE students are not good in programming. 

Svante
> jwe