That's a prospect that doesn't please advocates of AMD. Writing and
tuning for your own closed universe? AMD if you are writing nai-ve code,
Intel maybe if you are writing code that can make good use of Intel
productivity tools and are willing to fiddle. Writing for the world at
large? I don't know that I've ever seen a discussion of how commercial
developers have really dealt with not knowing whether they are writing
for K7 or P4, but I would assume it's safer to assume you are writing
for P4.
From what I've seen of most commercial code it's mainly a matter of
setting -02 in the compiler and praying that it doesn't break
something! :>
Seriously though, for most commercial code that is going into
widespread use the focus is on it working properly on as many
platforms as possible. This tends to avoid any architecture-specific
optimizations as they can often break your code on different setups,
not just AMD vs. Intel but even current Intel vs. next-generation
Intel. An awful lot of the world are still using PCs with PII and
PIII era processors, and optimizing for the PIII tends to be VERY
different from the P4.
There are, of course, going to be some exceptions to this rule,
particularly for really CPU-bound tasks. Some applications simply
detect what sort of processor you have and use different code paths
depending on what chip you've got. But for the most part you probably
won't see all that much in the way of processor-specific
optimizations.
This is, of course, one of the major complaints against SPEC CPU.
It's so sensitive to the compiler and optimizations used that it often
does not reflect on what you'll encounter in the real world. We've
most recently actually seen AMD and Sun use this to their advantage to
take back the lead as the fastest x86 processor for CFP2000 using the
new Pathscale and PGI compilers along with the newest version of GCC
and AMD's Math Libraries. 3 different compilers and PLENTY of
optimizations flags, but it gave them a rather impressive CFP2000
score (1637/1787 for base/peak).
