On Wed, 27 Jul 2005, Andreas Jochens wrote:
> thanks for your reply to my bug report!
>
> On 05-Jul-26 20:55, Richard B. Kreckel wrote:
> > On Tue, 26 Jul 2005, Andreas Jochens wrote:
> > > Package: cln
> > > Version: 1.1.9-1.1
> > > Severity: normal
> > > Tags: patch
> > [...]
> >
> > Thanks a lot for your patch!  For the sake of completeness and upstream
> > inclusion, I would love to make this a little bit more complete by
> > providing some basic inline asm macros for powerpc64.  This is not a big
> > deal if you now what exactly what to do but it needs basic testing.  Can
> > you provide me with an account on such a machine?
> > (http://db.debian.org/machines.cgi still seems down.)
>
> I only have access to one ppc64 machine which is connected to the
> internet by a dialup home DSL connection. It is not possible to
> access that machine remotely. But I heard rumors that IBM gives
> accounts on some of their new ppc64 machines for developing and
> testing purposes.
>
> If you send me a patch and tell me what to do I could also try to make
> some tests myself and report the results back to you.

Basically, besides what you've already found out, what needs tuning is
provide some basic support for inline assembly in cln/src/base/cl_low.h.
That probably looks worse than it actually is.  It can either be guessed
by looking at GCC's .md file or maybe be lifted from GMP or Pari.

At least the mulu64 macro needs to be implemented.  If possible, more.
Defining mulu64 for ia64 gave approximately a factor two speedup on
itanic.  Defining all the ones that are defined for x86_64 gave a factor
three or more, on that architecture.  There's a lot of hidden potential
here.

Also, I'ld like to ask some basic quality questions.  Does it work with:
* configure --disable-shared and --disable-static
* configure --without-gmp and default with gmp and are the sizes of the
  basic types in intparam.h the same in both cases?  If not, maybe
  something should be fixed.
* CXXFLAGS set to "-fno-exceptions -O2" and "-O1".
Does make check work with some or possibly all of the 2^3 combinations of
the above?

You could start answering the last question.  If you're daring, feel free
to tackle the other ones.  Otherwise, I'ld need access to a machine.

Regards
  -richy.
-- 
Richard B. Kreckel
<http://www.ginac.de/~kreckel/>



-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to