GMP Module (perl)

Sisyphus kalinabears@iinet.net.au
Sat, 26 Jul 2003 12:26:47 +1000


----- Original Message -----
From: "Kevin Ryde" <user42@zip.com.au>

>
> The asm routines are the most important thing for speed in gmp.  After
> that we usually recommend gcc, since it gets various inline asm
> blocks.  The generated perl stuff is pretty straightfoward and
> unlikely to be affected one way or the other.
>

Yes, I've (now) compared both. The gcc-built version runs a little quicker,
but the difference is barely measurable and insignificant. (5.02 seconds
versus 5.03 seconds for the tests that I ran.)

> > g++ didn't like having a variable named 'class'
>
> Ah yes, I might change that.
>

Good idea. You might also mention in the installation notes that gcc-users
should run the first step as 'perl Makefile.PL CC=g++'.

> > It's nicer, imo, to use
> > than Math::GMP, and additionally provides access to a greater range of
GMP
> > functions.
>
> More functions was the aim.
>

It's nicer, too. Sorry .... I can't do anything about that ....you'll just
have to live with it :-)
Main thing I like is that I don't have to code all that 'Math::GMP->new()'
stuff. Also I can pass a GMP object directly to my own XS functions (which
simply take strings as arguments). When I pass a Math::GMP object I have to
remember to interpolate it. Your module's documentation, being in effect the
GMP documentation, is excellent. Math::GMP doesn't seem to document all of
its functions.
And I find your perl function names easier to remember.
So, don't get too carried away .... it's not 'phenominally nicer' .... just
nicer (for me) in small ways :-)

> > Have you considered placing it on cpan ?
>
> It was easier to get started just chucking it in the gmp sources.  I
> suppose at least a cross reference in cpan would be nice.
>

I guess I'm just thinking that it would be noticed more if it were
accessible via cpan - and it's something worthy of more attention than it's
probably getting at the moment.

> I hadn't wanted to worry about being Math::BigInt compatible, so did
> something new.  I think BigInt was a bit bare when I'd first looked at
> it too.  Seems to have grown since then, might be the way to go for
> the future.
>

Yes, Math::BigInt is getting better - though it will always be unbelievably
slow. It does provide access to the important mpz() functions courtesy of
Math::BigInt::GMP (which now comes with its own XS code and no longer relies
on Math::GMP).
But, afaik, there's currently no modules on cpan that access the mpf() and
mpq() functions. (Haven't looked at what Math::Pari provides. I'm personally
interested pretty much solely in the integer functions.)

Cheers,
Rob