Re: Using LAPACK in octave

[John W. Eaton]

On 25-Sep-2007, David Bateman wrote:

| In fact xGELS, xGELSY, xGELSD and xGELSS all return the same minimum
| norm solutions. There is therefore no reason not to change to xGELSS..
| Consider the attached patch, before this addition I had the following
| 
| octave:1> A = rand(500,499); t = cputime(); b = A \ ones(500,1);
| cputime() - t
| ans =  3.1035
| 
| and after the patch I get
| 
|  A = rand(500,499); t = cputime(); b = A \ ones(500,1); cputime() - t
| ans =  0.23296
| 
| with ATLAS 3.6.0 installed. As for the accuracy, both versions returned.
| 
| octave:2> norm(A*b)
| ans =  22.361

Please check in this change.

| One thing that might be done in addition is to remove the non longer
| needed xGELSS code from LAPACK, however care needs to be taken to not
| remove a dependency of another function used in Octave, so I decided not
| to do that..

I think this can wait.

Thanks,

jwe