My version was build on gcc 2.95.2 on IRIX 6.5.6. I can't compile it with the
native cc !!!
I think that dx doesn't work on 64 bit only on 32, so you must force n32.
Can you explain your method to compile with native compile (aclocal... your
variable (PATH CFLAGS...).
The pg i run on dx, first load data from file, than make some work and
display.
My runs are "divide" in three.
During the first, there's a process dxexec (or more if -processors > 1
and dx 4.0.9. It doesn't work on more than 1 CPU with the IBM version).
During the second, my program to load data into dx work (with a level
of cpu different from version (16 % on 4.0.9, 40 % with IBM !!)
Last section work on the data and display.
The version 3.1.4 is making is jobs in 2 minutes, the 4.0.9 in 5.
Last question, do you work with the tar or cvs ?
Richard Gillilan wrote:
> Jean-christophe Penalva wrote:
>
> > Hello,
> >
> > now that i have a 4.0.9 dx working on my Onyx 2, i try my pg that
> > was generate with the old 3.1.4 IBM version. But there's a big pb, the
> > same pg run on 2 minutes with the IBM and 5 minutes with 4.0.9 !!!
> >
>
> Interesting, what type of computation does the net do?
>
> It would be nice if one of the pre-release tests was a benchmark
> to ensure that recent changes haven't slowed things down (too much).I've been
> using the SGI compiler on an Octane R10000 @ 250Mhz.(I never installed gcc,
> so the MIPS compiler just
> works).
> The default compilation seems to be n32 rather than 64bit so,
> I suspect there would be a very significant speedup if one were to
> use 64bit. According to "man abi", one can change the default compile
> type by editing the /etc/compiler.defaults file. Maybe I'll get brave this
> afternoon
> and try it. It doesn't solve you problem however. Would be better if
> configure options
> existed for overiding compiler choice.
>
> Richard
