On Thu, 2002-05-23 at 19:10, Thomas Bushnell, BSG wrote:
> GCC is not the best compiler in the world, but that's not what you
> claimed in the original message. You said that GCC adds optimizations
> which speed up code on some chips, at the expense of making it slower
> on other chips.
>
> Since GCC optimizations are (by default) carefully tuned, such that
> only those which help a given chip are enabled for that chip, I'm
> wondering if you can be more precise.
I've heard this too (but I've seen about as much evidence of this as I
have of the 30% claimed speedups from optimizations, so I don't claim
it's right). Apparently it has something to do with e.g. i586
optimizations being slower than i386 (no optimization except
platform-independant ones) on i686 chips, or vice versa, or Intel
optimizations slowing down AMD and vice versa.
I've also heard that apparently turning on i486 optimizations can cause
586/686 processors no end of invalid instruction problems.
But like I said, I treat pretty much every claim in this thread as
fairly dubious.
--
- Joe Wreschnig <piman@sacredchao.net> - http://www.sacredchao.net
"What I did was justified because I had a policy of my own... It's
okay to be different, to not conform to society."
-- Chen Kenichi, Iron Chef Chinese
Attachment:
signature.asc
Description: This is a digitally signed message part