On Nov 30, 12:40 pm, "William Stein" <[EMAIL PROTECTED]> wrote:
> On Sun, Nov 30, 2008 at 12:37 PM, Martin Albrecht
>
>
>
> <[EMAIL PROTECTED]> wrote:
>
> >> Here's another example:
> >> 64-bit:
> >> sage: time a = random_matrix(ZZ,200).determinant()
> >> CPU times: user 2.74 s, sys: 0.01 s, total: 2.75 s
>
> >> 32-bit:
> >> sage: time a = random_matrix(ZZ,200).determinant()
> >> CPU times: user 0.74 s, sys: 0.04 s, total: 0.78 s
>
> >> Here, surprisingly, 64-bit is WAY SLOWER than 32-bit at doing the
> >> exact same thing. I don't really understand this. The moral should I
> >> think be that 32 versus 64 bit can be really subtle when it comes to
> >> performance.
>
> > It seems to me that this should be considered a bug.
Well, I would guess that one of the above was run under virtualization
or am I wrong?
> I don't think "bug" is the right word -- however, fixing this or giving a good
> explanation should be considered an "Enhancement".
>
> I'm surprised about the speed difference. It's very likely in either IML
> or Linbox.
Speed differences where 32 bit code is faster than 64 bit code are
most commonly caused by code that manipulates a lot of pointers. But
the speed difference should usually be only a factor of up to two and
I am sure the above code does not do 100% pointer manipulation. One
explanation for the observed difference could be cache foot print,
i.e. code + data in the 32 bit case fit the L1 or L2 "better" than the
64 bit version. Note that 64 bit binaries are also fatter than their
32 bit counterpart. At least on x86-64 CPUs the larger number of
registers should compensate for the pointer issues.
What to do? Let's try something larger than 200x200 to see if the
cache is involved or not.
> William
Cheers,
Michael
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-support@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-support
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---
