Re: MacOSX: Octave.app 2.9.14

[John W. Eaton]

On  3-Oct-2007, Vic Norton wrote:

| I am confused too. Aren't highly optimized BLAS accessible on most  
| machines? For example, on my Mac, in "/Developer/SDKs/.../Headers/ 
| cblas.h", I see
|     double cblas_dnrm2(const int N, const double *X, const int incX);
|     double cblas_dasum(const int N, const double *X, const int incX);
| These are the norm(x, 2) and norm(x, 1) functions, respectively. They  
| should be about as fast as you can get.
| 
| Of course this is all theory. I have no idea how to access these  
| prototypes. I can't even figure out how to compile octave.

I posted a patch on the maintainers list that should speed up the
vector cases.

We already have internal functions that use the BLAS, but the way
__vnorm__ was setting up to call them was not ideal, and then the
overhead of calling norm.m to get to __vnorm__ resulted in an
additional performance penalty.  I think my patch addresses that for
vector cases.  Someone will need to do some more work to improve the
performance for the matrix norms as well.  I think that can wait until
after 3.0.

jwe