seebs@plethora.net writes:

> Don't be.  The Pentium Pro, PII, and Celeron all have an architectural
> advantage over the 21164.  Intel came up with this wacky idea called "out of
> order execution", where the processor reorders instructions as they come to
> make them more efficient.  (This has to be done on the chip, not in the
> compiler, because "more efficient" can depend on what happened on the other
> side of a context switch.)  The 21264 has this feature too, and other related
> hacks, but the '164 is fairly old.

Out of order execution isn't wacky at all these days, and Intel didn't
come up with it...

The compiler *can* help a lot with this kind of tuning by knowing how
many execution units the chip has and arranging for maximum
parallelisim. Also, many of the benefits of out of order execution
come from utilizing the execution units in times that would otherwise
have a pipeline stall. The compiler can also do a lot of scheduling
work up front to avoid such stalls, which reduces the benefit of
out-of-order execution.

Having said all that, I don't believe egcs is really all that good at
it :(.

In general, I'd be a lot more careful before claiming that
out-of-order execution (or any other single architectural feature,
really) is the killer difference between a PPro-family chip and a
21164.

        - Nathan