Quadruple boot YDL, Gentoo PPC, Mac OS X, Mac OS 9???
Robert Serphillips
yellowdog-general@lists.terrasoftsolutions.com
Sat Jul 19 21:57:01 2003
I totally agree in this case, in that cpu specific optimazations make
little to no difference with the exception of the almost non-existant
altivec "aware" app.
The mindset is usually attributed to people in the x86 camp where
compiler optimizations usually do make a big difference in speed. Very
little has changed in PPC world to make any cpu specific optimisations
worthwhile. Take the difference between a 750 and a 7400 for instance.
Besides the altivec extensions most performance enhancement's are
"built in". Better utilization of the 60x bus, more dispatched
instructions per clock, and faster/higher bandwidth L2 in the case of
the 745x. I call this best case forward compatibility. An app that
compiles and runs on a 601 will only run faster on a 750/740x/745x in
most cases.
Where compiling your own stuff really makes sense is the total
customization of your env. Take as another example a simple mail
reader I use, sylpheed. If I use the rpm found from some of the PPC
repositories out there they usually have everything built in, SSL,
LDAP, GNUPG, etc. When all that I require is aspell and the dillo html
plugin. Why bother installing all of the other shared libs for
features I won't be using. Again this is in agreement with the earlier
post in that most stuff should be optimized for size and " feature
bloat", instead of worrying about shaving off 0.5 seconds of your apps
launch time.
-Rob
On Sat, 19 Jul 2003 19:38:08 -0700
Tim Seufert <tas@mindspring.com> wrote:
> On Saturday, July 19, 2003, at 10:53 AM, Andrew Choong wrote:
>
> > i'm intrigued with gentoo though. probably won't use it all the
> > time, but i am VERY curious to see how much faster it supposedly
> > is (as compared to YDL) given all this talk regarding optimised
> > compilation etc
> > etc.
>
> There may be other benefits to gentoo, but the "faster" argument
> strikes me as crap. Most of the code in a distribution doesn't need
> to be optimized for a specific CPU, or optimized at all for that
> matter. 95% of the code in almost any program, even most programs
> which need high performance, is not performance critical and should
> really be optimized for size rather than speed. And a huge number
> of programs in any given distribution need no optimization at all,
> because you aren't going to run them but once in a blue moon and
> they won't take much time anyways. (Does anybody actually care how
> fast 'cal' prints out a calendar? Didn't think so.)
>
> What's worse for the gentoo case is that when I've messed around
> with GCC optimization flags, I've found that (at least on PowerPC
> CPUs) the best target to use doesn't always match the CPU you're
> using. E.G. I've had programs that would run faster on a G3
> (PowerPC 750) when compiled with -mcpu=601 instead of -mcpu=750. In
> my experience there's every reason to expect that the actual
> measured results from compiling everything with -mcpu=(your CPU
> type) would be positive for some programs and negative for others,
> with little predictability.
>
> Most of the reported benefits are probably placebo effect. gentoo
> users believe gentoo should be faster, and that leads to the
> subjective impression that it is faster, even if it's not.
>
> _______________________________________________
> yellowdog-general mailing list
> yellowdog-general@lists.terrasoftsolutions.com
> http://lists.terrasoftsolutions.com/mailman/listinfo/yellowdog-general
>
>