> > 64-bit code is often much larger. Perhaps thats why?
>
> Not a whole lot larger (not twice). In fact, the percent larger is probably
> about the same as the percent better of the execution speed. While that may
> be a tradeoff the designer thinks he wants, I believe he limits himself in the
> future rather rapidly. 640K barriers have a way of sticking around.
Code and data size is one of things that are very important for good
performance on small systems. Small systems usually only have a L1 cache
and dropping out of the L1 cache because of doing stunts like 64 bit
address space is just stupid. Making use of 64 bit registers for
computations is something else but again on small systems you have to make
your tradeoffs very carefully. Even just using 64 bit registers for
computations but not the 64bit address space (ala the n32 MIPS ABI) means
the process context gets twice as large means more scarce resources like
RAM and especially cache means you _might_ actually end up with less
performance. Again, using 64 bit features of the CPU has to be done very
carefully on small systems.
Ralf
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html