Performance comparison, ULE vs 4BSD and AMD64 vs i386

[Brooks Davis]

On Tue, Feb 24, 2004 at 05:36:55PM -0500, John Baldwin wrote:
> On Tuesday 24 February 2004 05:04 pm, Jem Matzan wrote:
> > Brooks Davis wrote:
> > >On Tue, Feb 24, 2004 at 04:37:44PM -0500, Jem Matzan wrote:
> > >>How about AMD64 being slower than i386 on the same hardware? By
> > >>slower, I mean a buildworld -j4 took about 400 seconds longer in AMD64
> > >>mode.
> > >
> > >You can't usefully compare compile times when you are compiling for
> > >a different instructions set.  The work involved is rairly the same
> > >so the results are meaning less.  If you could factor out the cost of
> > >building the native bootstrap tools since that isn't the same job on
> > >each machine, the speed of a cross buildworld would be an intresting
> > >test.  For comparing i386 and amd64, I'd probably build an alpha or
> > >sparc64 world so the target would be entierly different.
> > >
> > >-- Brooks
> >
> > I figured that the world would be the same for both AMD64 and i386. That
> > really sucks that all of this data and all of that time has been more or
> > less wasted on doing buildworld time benchmarks. As far as I know it
> > isn't possible to do a crossbuild (I've tried before, and I read on the
> > list several weeks ago that it won't work). Do you have any suggestions
> > for measuring compile times?
> 
> You can do a crossbuild easy, just do:
> 
> make TARGET_ARCH=amd64 buildworld
> 
> or subsitute whatever arch for amd64.

If you use this for benchmarking purposes, you can only count stage 4,
the world stage.  Before that you are building tools for the native OS
type.  If your purpose in using the box is developing architecture
independent code, a buildworld does produce a vaguly useful benchmark of
how fast compilation is, but that will have more to do with the compiler
then the OS.  Compiler performance is affected by both compiler
efficency and the ease or difficulty of producing code that runs
decently on the given architecture.  It's hard to say anything
definiative from a simple build time since a fast compiler might produce
code that runs horribly slowly.

-- Brooks

-- 
Any statement of the form "X is the one, true Y" is FALSE.
PGP fingerprint 655D 519C 26A7 82E7 2529  9BF0 5D8E 8BE9 F238 1AD4
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-amd64/attachments/20040224/e072b998/attachment-0001.bin