Re: Addition of ITSOL to MXE.

[marco atzeri]

Il 6/29/2013 9:36 AM, c. ha scritto:
> On 29 Jun 2013, at 08:10, marco atzeri <address@hidden> wrote:
>
> > Hi Carlo
> > That was the idea. I was just on early test phase before submitting
> > to upstream.
> ...
> > The source change is minimal, I used Kai's one, and only due to
> > double function declarations that prevent building a shared lib.
> > A upstream review is needed for possible alternative implementation.
>
> Marco,
>
> I think that, if autoconf is being used, also the "hand made" check for
> fortran symbol names in "defs.h" could be substituted by use of F77_FUNC.

sure, and it can also trimmed down anyway as only
"#ifdef _IBM" is different by "else"

these are not in realy needed
#ifdef _SGI
#ifdef _LINUX

I am also wondering why on _IBM
the previous definition are not a problemm

#define qsplit qsplit_
#define dgemv dgemv_
#define readmtc readmtc_
#define csrcsc csrcsc_
#define coicsr coicsr_
#define roscal roscal_
#define coscal coscal_

but this change is needed

#ifdef _IBM
#define dnrm2 dnrm2
#define ddot ddot
#define daxpy daxpy
#else

as I know nothing on the IBM platform, I can not judge.
Same for protos.h definitions

> And also something might need to be done to allow for combatibility with
> Octave's new 64-bit indexing.

I am not familiar with it. I had the impression they need special built 
blas/lapack libraries with also 64-bit indexing.


> As the changes are getting a bit intrusive I think the upstream maintainers
> should get involved from this early stage already.

A repository driven/coordinated by them will be needed, as I assume they
already have further development underway.
A dedicated mailing list will be also helpful

Ruipeng,
your opinion and preference ?

> c.
Marco