Re: Poor performance of G5 on GNOME's CPU benchmarks.
Le Mon, Aug 04, 2008 at 11:10:02AM +0200, Michel Dänzer a écrit :
> On Mon, 2008-08-04 at 10:24 +0900, Charles Plessy wrote:
> > http://gcc.gnu.org/onlinedocs/gcc/PowerPC-AltiVec-Built_002din-Functions.html
>
> These flags are prerequisites for allowing the resulting binaries to
> contain Altivec code at all, but on their own they don't cause any such
> code to be generated (except maybe for some simple cases like memory
> copies). Current versions of GCC can automatically vectorize some code
> with -ftree-vectorize, but the results probably still can't compete with
> hand-optimized code.
>
> Also note that these flags will make the resulting binaries unusable on
> systems without Altivec support.
Well, it seems that it makes it doubly impossible to get the best
performance out of those machines :(
Anyway, the example I gave (the building of the exonerate package) is
definitely not the best. It does some kind of bootstraping (that made
parallel building fail) that I do not understand, nor I know if the
speed of bootstrapping reflects the actual speed of the binary. By the
way, I uploaded a new upstream release, that took 1h40 to build on
amd64, 7h30 on G4, and 8h on MIPS. [OT: I have not data with G5 as I did
not want to leave my iMac switched on overnight and the machine I use at
work crashed again. Does anybody knows a good equivalent to memtest86?]
Another example I have is a map website by Yahoo Japan:
http://map.yahoo.co.jp/pl?type=scroll&lat=35.35961995&lon=138.73361576&sc=7&mode=map&pointer=on
With Iceweasel it is slow on G5 and fluid on Intel. Again, not so easy to
benchmark. I still have that gut feeling that the my G5 machines runnign
Debian are abnormally slow compated to my Duo 1.5 Ghz laptop.
Have a nice day,
--
Charles Plessy
Tsurumi, Kanagawa, Japan
Reply to: