[kde-solaris] Is kde-forte's performance claim a myth?

[Lars Tunkrans]

rain cip wrote:
> Here is the file command output on the gcc compiled
> versin of kate:
> 
> $file /opt/csw/kde-gcc/bin/kate
> kate:           ELF 32-bit MSB executable SPARC32PLUS
>  Version 1, V8+ Required, UltraSPARC1 Extensions
>  Required, dynamically linked, stripped
> 
> So I am guessing that -mcpu=ultrasparc and -mv8plus
> are used to compile it.  I am not familar with the 
> KDE source code as a whole.  But I wonder in what
> area of KDE will the Forte compiler help to optimize
> the run-time performance better than the gcc 
> compiler?  Someone already mentioned khtml here.  
> I frankly didn't see a noticeable performance
> difference when loading up the same HTML pages in
> both the Forte version and the gcc version of the 
> konquer.  Could it be that being a window system, KDE
> is bound to the I/O bottleneck that a good graphics
> card and hard drive will diminish any optimization 
> advantage the Forte compiler has over the gcc?
> 
> rain


  I dont have several version of KDE online at the moment,
  However I do have two versions of mozilla 1.7 online.

  Version A is the stock mozilla 1.7 Forte Sparc binary as
distributed by the mozilla.org
$ file mozilla-bin
mozilla-bin:	ELF 32-bit MSB executable SPARC Version 1, dynamically linked, stripped

Version B is the same source but I recompiled it  with Studio 8 using these options:

   -xO4 -xmemalign=8s -xprefetch=auto,explicit -xtarget=ultra3i -xcache=64/32/4:1024/64/4 -xarch=v8plusb

giving you this file type.
$ file mozilla-bin
mozilla-bin:	ELF 32-bit MSB executable SPARC32PLUS Version 1, V8+ Required, UltraSPARC3 Extensions Required, dynamically linked, stripped


On my SunBlade 1500  I can consistently measure two seconds faster load time for
the US IIIi  optimized code.

Code Version A loads in 5 seconds  and code Version B loads in 3 seconds.
this is measured when everything is already rolled in to RAM by a previous startup
of the code. So compiling for a specific platform does make a real difference.

   //Lars