Re: MPQS weirdness

[Thomas Papanikolaou <eedtpa@aquis25.eed.ericsson.se>]

Hi everybody,

some comments on the speed of MPQS on different machines:

the speed if MPQS highly depends on the speed of sieving.
Through sieving we just move a pointer along an array and
add logarithms to each elements (for details looks into the
code). This means that the speed of sieving depends more on
characteristics such as cache, memory accesses but not
necessarily on the speed of integer arithmetic.

It might therefore well be that the Sparc-60 is doing better
than the P233 (although it is surprising, since MPQS was
developed and optimized using both architectures).

Perhaps someone can see into the sieve code and check if it
can be optimized specially for Pentium Processors, i.e. their
cache?

	Best
		Thomas


______________________________________________________________
Thomas Papanikolaou (PhD)     Vox: +49 2407 575 699
Ericsson Eurolab              Fax: +49 2407 575 507 
Ericsson Allee 1       
D-52134 Herzogenrath          Thomas.Papanikolaou@ericsson.com