Could we agree to disagree and come together on improvements andclean-up?
arndt at jjj.de
Mon Jun 2 12:34:59 CEST 2008
On the positive danger to draw this thread out too much...
* Paul Leyland <paul at leyland.vispa.com> [Jun 02. 2008 20:08]:
> On Mon, 2008-06-02 at 10:00, Vincent Lefevre wrote:
> > > > No, long should just fit the needs. What if a processor has 32-bit
> > > > words and use pointers on 64 bits (two words)?
> > >
> > > Would you expect to see such CPU?
> > Strangest CPUs have been seen in the past. And some CPUs such as DSPs
> > have particular features.
@Vincent: I _knew_ you'd mention DSPs, bad boy!
> [... bad old times ..]
> More recently, I've programmed heterogeneous clusters where you most
> certainly can not assume that all the cpus have the same architecture or
> even word length. That particular work was done with Condor on sundry
> Sun boxes both M68K and Sparc. A current example of a heterogeneous
> cluster on a chip is the PS3, so such things are still relevant.
OK, let me try to say it in a different way:
I have yet to see the world-solution, compiles-and-runs-everywhere
nontrivial piece of code. When some new arch/model needs support
you'll always need to get down & dirty if you want decent performance.
I have the impression that extended precision arithmetic may be
the field where this is more true than with most other problem sets.
GMP uses assembler inlines, this is the single most nonportable
way of doing things, nobody complained about that one so far 8-)
Disclaimer for those with an excessive tendency to read between the lines:
I do not advocate sloppy or non-portable coding.
This mail has been checked for viruses, trolls, and brown paper bags.
It is 30+ degree outside and the sun is shining.
More information about the gmp-discuss