Re: Odd behavior of time-domain convolutions

[David Bateman]

Fredrik Lingvall wrote:
>
> The time-domain convolution algorithm look  like this:
>
>   for(i=0; i<nx; i++) {
>     for(j=0; j<ny; j++) {
>       zr[i+j] += xr[i] * yr[j];
>     }
>   }
>

Does the Octave version really look like the above or rather something like

const NDArray xr, yr;
NDArray zr;

for(i=0; i<nx; i++) {
    for(j=0; j<ny; j++) {
      zr(i+j) += xr(i) * yr(j);
    }
  }

If so then the operator () on Array<T> objects in octave does bounds
checking, etc. You should use the fortran_vec method to get "double *"
or "Complex *" pointers to the arrays something like

const NDArray xr, yr;
const double *pxr = xr.fortran_vec ();
const double *pyr = yr.fortran_vec ();
NDArray zr;
double *pzr = zr.fortran_vec ();

for(i=0; i<nx; i++) {
    for(j=0; j<ny; j++) {
      pzr[i+j] += pxr[i] * pyr[j];
    }
  }

Note that xr and yr are defined as "const" to avoid a copy being made
with the fortran_vec method.

D.


-- 
David Bateman                                address@hidden
Motorola Labs - Paris                        +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 6 72 01 06 33 (Mob) 
91193 Gif-Sur-Yvette FRANCE                  +33 1 69 35 77 01 (Fax) 

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary