Re: [ssax-sxml] Re: Ssax speed

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Hi,

ol...@po... writes:

> Joerg F. Wittenberger wrote:
> > Askemos rscheme/regexp                  3 s
> > Askemos chicken/ssax                    25 s
> > So the question is......why?
> 
> Perhaps the answer lies in compilation flags? Have you tried the
> command line like the following
> > csc -O3 -d0 -disable-interrupts -bkav -C "-mcpu=pentium -march=pentium -mpreferred-stack-boundary=2" cg.scm -o cg-chicken 
> 
> (mentioned in a recent post by Felix regarding a Gulch benchmark).

Thanks for the hint.  I used only documented options like this:

CSCFLAGS=-case-sensitive -optimize-level 2 -debug-level 0 -unsafe 

-disable-interrupts is somewhat strange to me.  The rscheme version of
Askemos is already more responsive under load.  -disable-interrupts is
not actually a useful option, it would stop Askemos from working and
be just an unfair advantage for the benchmark (can't turn it off for
rscheme).

In the meantime I went a little deeper into this question; chicken is
supposed to be faster and I need some more performance for Askemos.
The results are currently depressing.

I used the ~30k README from www.askemos.org in 100x loop:

a) serializing from xml tree chicken (with the options above) 20s
   rscheme 14s.

   Remark: there is some C-Code to add translated strings (& -> &amp;
   etc.) to the output.  Both versions use C code to copy the
   translated string into the output.  The rscheme version also
   translates the characters into those strings in C, while the
   chicken code does the vector-ref in Scheme.  Otherwise no
   difference.

b) parse the file into an xml tree.  rscheme (regex based parser)
   needs 5s.  Chicken/ssax uses 55s.

> The -fomit-frame-pointer flag to GCC is also quite useful.

I used for chicken:

CCFLAGS=-Wall -Wno-unused -O3 -fomit-frame-pointer -fstrict-aliasing

While rscheme for some reason generates in it's Makefiles -O2 and
nothing else.  I understand that rscheme should be fixed in that
respect, but it would make the above results sharper at best.

best regards

/Jörg
-- 
The worst of harm may often result from the best of intentions.