Re: oct file: slower than expected using fortran_vec

[Seb Astien]

On Mon, May 16, 2011 at 8:51 PM, John W. Eaton <address@hidden> wrote:
> On 16-May-2011, Seb Astien wrote:
>
> | Thank you so much. It is what I was missing.
> | Indeed, now:
> |
> | octave:1> A=rand(10000,10000);
> | octave:2> tic; s1=sum(sum(A)); toc
> | Elapsed time is 0.205881 seconds.
> | octave:3> tic; s2=sumit(A); toc
> | Elapsed time is 0.203057 seconds.
> |
> | So it is faster than the built-in, which I was expecting since the
> | built-in sums first per columns and then sum the results.
>
> If you want to sum all the elements of an array, then you should write
>
>  sum (A(:))
>
> instead of
>
>  sum (sum (A))
>
> No copy is made when you write A(:) and then you can sum over all the
> elements without having to know how many dimensions there are (so you
> don't have to call sum ndims(A) times, or have different code for different
> numbers of dimensions).
>
> Also, tic/toc gives you wall clock time.  If you want to do
> benchmarking, you should be using cputime instead.
>
> Another thing that accounts for differences between your code and the
> built-in sum function is that the built-in function does some argument
> checking and dispatches based on the argument type.  For example, what
> happens to your function if you call it with an argument that is not
> an array of doubles?
>
> Also, your error checking  is not quite right.  You should also return
> a value if an error occurs.  The value won't be used, so it can be
> anything, but you should still return something. The normal thing to
> do is return an empty octave_value_list object, or a single undefined
> octave_value object (it will be converted to an octave_value_list
> object automatically).
>
> jwe
>

Thanks a lot for all this info. And also thanks to Jordi.
Great list, fantastic software.
Regards,

Seb