On Thursday 15 May 2008, Bill Hart wrote:
> Hi Martin,
>
> The cycle counter in your bench code seems to give random values on
> the 2.4GHz Opteron with SUSE 9 linux that I have access to, which has
> Magma 2-14.10 on it.
>
> Anyhow, here are the Magma times:
> > A := RandomMatrix(GF(2),10^4,10^4);
> > B := RandomMatrix(GF(2),10^4,10^4);
> > t := Cputime(); C := A*B; Cputime(t);
>
> 2.940
>
> > A := RandomMatrix(GF(2),2*10^4,2*10^4);
> > B := RandomMatrix(GF(2),2*10^4,2*10^4);
> > t := Cputime(); C := A*B; Cputime(t);
>
> 16.570
>
> > A := RandomMatrix(GF(2),2^14,2^14);
> > B := RandomMatrix(GF(2),2^14,2^14);
> > t := Cputime(); C := A*B; Cputime(t);
>
> 9.250
>
> The corresponding times for your code are approximately (hand timed;
> adjusted for generation of random matrices):
>
> 10^4: 8s
> 2*10^4: 59s
> 2^14: 43s
>
> I think it is polynomials over GF2 which Magma 2-14.x is much faster
> at rather than linear algebra.
>
> Bill.
Hi Bill,
I cannot explain the random behaviour of the cpucycles counter. The cpucycles
library is by Dan Bernstein so I'd suspect it is carefully coded. Some
potential sources of variance are: dynticks (i.e. the kernel doesn't have a
fixed tick rate, this is done to safe power), CPU frequency scaling and other
applications. On the other hand, I'd suspect I don't need to tell you since
you probably benchmarked more code than I did. To account for the variance I
have the small run_bench.py script in the testsuite directory. Run it like
this:
./run_bench.py -t mul -n 10000 -c 3200
It'll run the experiment 17 times (17 for no particular reason) and return the
median runtime. -n is the size and -c the cutoff.
Cheers,
Martin
--
name: Martin Albrecht
_pgp: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x8EF0DC99
_www: http://www.informatik.uni-bremen.de/~malb
_jab: [EMAIL PROTECTED]
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-devel@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---
