Optimized kernel builds: the straight dope [fp extensions]

[Ben Collins]

On Tue, 2006-08-15 at 17:26 +0200, Pavel Rojtberg wrote:
> Matt Zimmerman wrote:
> > I'm interested to hear (objective, reasoned) feedback on this data and my
> > conclusions from the members of this list.
> it would be interesting to see a i386 vs k7 comparison.
> Since the architectural difference is bigger; the k7 architecture has 
> some floating point extensions(3D Now) which usually have a big impact 
> on performance.
> Therefore we should look if the kernel can handle these well enough 
> dynamically.
> Perhaps it would make sense to leave one optimized kernel on i386 for 
> newer CPUs.
> This could be SSE optimized then to be more generic (AthlonXP+, Pentium3+)

As before I did the tests, I'd be willing to bet there's little
difference.

The instructions you are talking about are the same instructions that
get magically replaced with optimized versions by the generic arch
support.

People, understand that we are not comparing "non-optimized" vs
"optimized". We are comparing a kernel that is compiled with the
optimizations as default (k7/686) against a kernel that swaps these
optimizations in at run time (the 386 kernel).

Granted, these optimizations are slim when compared to the entire kernel
binary, but these optimizations are in the right places.

What we are proposing is this:

i386
  -386 alternate kernel (UP, compiled with 486 generic target)
  -generic main kernel (SMP, ppro generic target, which is a common base
for pentium, k7 and k8 kernels)
  -server kernel (same as before)
  -server-bigiron (same as before)

amd64
  -generic kernel (SMP, x86_64 generic target, which is the same as our
current amd64-generic kernel)
  -server kernel (same as before)

Note that the -386 is not changing. The benchmarks I provided are
comparing a 486 generic target kernel against a 686 specific kernel. If
there's not much difference there, then the difference between our
current k7/686 kernels and this new ppro generic kernel will be even
less.

If we ever get the SMP issues worked out with the generic kernel, then
we'll probably drop the i386 generic kernel down to something like 586
generic target.

-- 
Ubuntu     - http://www.ubuntu.com/
Debian     - http://www.debian.org/
Linux 1394 - http://www.linux1394.org/
SwissDisk  - http://www.swissdisk.com/