Magic speedup

Paul Leyland paul at
Sat Apr 22 11:43:18 CEST 2006

On Fri, 2006-04-21 at 09:22, Torbjorn Granlund wrote:
> I only now regenerated the "Computing billions of pi digits using GMP"
> page for GMP 4.2, see
> The speedups I got for 4.1.4 => 4.2 was surprising.  Computing 1e9
> digits used to take 34300 seconds, now it takes just 8859 seconds on
> an Opteron.  The main sources of speedup should be the x86_64 assembly
> and Paul Zimmermann's 2003 FFT improvements.  But still a speedup of
> almost 4x is much, much more than I had expected.

I noted a speed-up of about five times in gmp-ecm when adding in the
x86_64 assembly to GMP 4.1.4 (and no other changes as far as I am aware)
on an Athlon 3500+.   Took me quite by surprise too, as I was expecting
at most a 3-fold improvement, from the naive expectation that doubling
the word length would quadruple the speed of multiplication and at most
double everything else.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url :

More information about the gmp-discuss mailing list