David O'Brien wrote:
> On Mon, Jul 01, 2002 at 08:14:46PM -0700, Matthew Dillon wrote:
> > My conclusion is that softupdates is working fine and (A) the new GCC
> > is a whole lot less efficient then the old GCC
> You really cannot say this -- GCC 3.1 does things 2.95 doesn't. 3.1 has
> a totally rewritten code scheduler. People can't get Pentium-4 and
> Athlon tbird specific optimizations for free.
> You almost seem to be making a claim on the quality of generated code,
> vs. just the run-time of the compiler. The two are different.
But it's possible to be conclusive. The way you would do this is
to compile the 3.1 compiler with the 3.1 compiler for one test,
and compile the 3.1 compiler with the 2.95 compiler for the other.
If the 3.1 compiler compiled with the 2.95 compiler is faster,
then 2.95 "wins" on code generation.
I rather suspect that 3.1 is doing additional work, other than
"intentionally" bloating the boot blocks into unusability, etc.,
and you will find that the 3.1 compiled 3.1 is more efficient
than the 2.95 compiled 3.1. At which point it becomes a trade
between compilation time vs. run time efficiency.
Personally, I prefer knowing my code should work before giving
it to the compiler, rather than using the compiler to think
about things I'm too lazy/incapable of thinking of on my own.
Given that, I would always favor a trade for faster run time
and slower compile time.
I think maybe the main problem here is that people have been
utilizing the wall time of "make world" as some kind of generic
benchmark for so long that the idea of any increase in the wall
time can't help but trigger a gut reaction against it.
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message