Re: 2.9.12 (2.9.13) chol broken with --enable-64?

[John W. Eaton]

On 11-Aug-2007, Fredrik Lingvall wrote:

| I have compiled LAPACK (3.0.1) with -m64 and -march=nocona and I use 

What compiler?  What do those options mean?

| GOTO BLAS with
| BINARY64  = 1 in the Makefile.rule file.

I don't know what that does.

Octave expects to use the Fortran interface to lapack and blas
functions, so do these options make the INTEGER arguments to those
functions signed 8 byte values?

| To test I created an oct file that calls the LAPACK routines,
| 
|   dpotrf_(upper_or_lower, &N, Y, &lda, &info);  (cholesky factorization)
| 
| and  
| 
|   dpotri_(upper_or_lower, &N, Y, &lda, &info); (positive triangualar 
| inverse)
| 
| which should do the same thing as cholinv (I guess). The result is
| 
| octave:1> A=rand(100,100);
| octave:2> A=A'*A;
| octave:3> C = potri(A);
| octave:4> C = potri(A);
| octave:5> C = potri(A);

What is potri?

| octave:6> cholinv(A);
| octave:7> cholinv(A);
| error: cholinv: matrix not positive definite
| 
| I test the info parameter after both calls above and the info parameter 
| is always = 0 (no error).

I don't think anyone can help you with this problem if you don't show
the code you are using.

jwe