GENERIC - Raising the bar

[Bill Hacker]

Joerg Sonnenberger wrote:

On Sat, Mar 12, 2005 at 03:57:34AM +0800, Bill Hacker wrote:

*BSD's have long shipped with a GENERIC kernel conf that will work on 
386 and any newer x86.

Manual editing has been required to optimize for 486/586, K6, and 
various flavors of Pentium onward.


Unlike FreeBSD which needs cmpxchg, the performance difference should be
pretty small, since the scheduling is done for Pentium Pro, a somewhat
useful default. For stuff like bcopy() we use optimized versions anyway,
they are choosen on runtime.
Joerg
Agreed w/r performance.

But might it not still be useful w/r further code streamlining - perhaps
a bit of reduced coding workload, potentially fewr places for error,
if the target set were further restricted?
An 'ulterior motive' is to hope to perhaps see some scarce developer
resources freed to further pursue Opteron optimization (and Itanium, or
follow-ons).  As Intel take advantage of cross-licensing from AMD, and
AMD respond, we are in for some 'interesting times' w/r CPU technology.
 - then, hopefully in about a year's time, target the IBM Power
family.  'Big Iron' server platforms, serious multi-core, not Motorola / 
Mac.

Not to mention 'Cell' ......  if/as/when....

The model Matthew has laid out should really start to pay off
in those environments.
Bill

Hopeful....

Bill