Hello!

Has anyone tried the intel compiler for Linuxfor compiling Povray? 
(http://www.intel.com/software/products/eval/)
If so, I would be interested to see the parameters. According to 
 http://www.open-mag.com/754088105111.htm
and
 http://www.heise.de/newsticker/data/hes-11.11.01-000/ (german)
it produces much faster code. Unfortunately, I wasn't able to reproduce 
this!

The rendering times I got are:
gcc 2.95.3: 4min 48s
gcc 3.0.1:  3min 41s
icc:        4min 41s 

The compiler options were:
gcc 2.95.3: 
-O3 -Wall -finline-functions -ffast-math -c -ansi -march=i686 -DCPU=686 

gcc 3.0.1:
-O3 -Wall -finline-functions -ffast-math -foptimize-sibling-calls -c -ansi 
-march=i686 -DCPU=686 

icc:
-O3 -c -ip -tpp6 -rcd -axK -xK -ansi

I've used a patched version of MegaPov 0.4 with the scene furrytorus.mpg
at http://www.grafik.willhalm.de, but this shouldn't matter.

Any ideas?

Post a reply to this message

In article <3c56c37c@news.povray.org> , Thomas Willhalm 
<wil### [at] fmiuni-konstanzde>  wrote:

> Has anyone tried the intel compiler for Linuxfor compiling Povray?
> (http://www.intel.com/software/products/eval/)
> If so, I would be interested to see the parameters. According to
>  http://www.open-mag.com/754088105111.htm
> and
>  http://www.heise.de/newsticker/data/hes-11.11.01-000/ (german)
> it produces much faster code. Unfortunately, I wasn't able to reproduce
> this!

> icc:
> -O3 -c -ip -tpp6 -rcd -axK -xK -ansi

iX, issue 11/2001 on page 70 lists these three:

icc -O2
icc -O2 -xW
icc -O2 -tpp7
icc -O2 -xW -ipo

Unfortunately, i don't know anything about the Intel compiler or its options
as I never used it myself, but these parameters are a good hint I think.

BTW, the iX article is much more informative and technical than a similar
article in c't at about the same time, which is very interesting as apparently
they wrote those article independently.  This is strange as iX and c't are
still in the same office and have the same publisher...

    Thorsten

____________________________________________________
Thorsten Froehlich, Duisburg, Germany
e-mail: tho### [at] trfde

Visit POV-Ray on the web: http://mac.povray.org

Post a reply to this message

Thorsten Froehlich wrote:

> In article <3c56c37c@news.povray.org> , Thomas Willhalm
> <wil### [at] fmiuni-konstanzde>  wrote:
> 
>> Has anyone tried the intel compiler for Linuxfor compiling Povray?
>> (http://www.intel.com/software/products/eval/)
>> If so, I would be interested to see the parameters. According to
>>  http://www.open-mag.com/754088105111.htm
>> and
>>  http://www.heise.de/newsticker/data/hes-11.11.01-000/ (german)
>> it produces much faster code. Unfortunately, I wasn't able to reproduce
>> this!
> 
>> icc:
>> -O3 -c -ip -tpp6 -rcd -axK -xK -ansi
> 
> iX, issue 11/2001 on page 70 lists these three:

Thanks for the hint. I've now read this article, too. Even with this
knowledge, gcc 3.0 produces faster code than icc. The reason might be
that I use only a Pentium III and the big improvement holds only for
Pentium 4. Another reason might be the rather special scene that contains 
mainly a slow media container. 

So far, I have tested some combinations of options that make sense. I will
try different scenes and maybe a different processor later.

Thanks again
Thomas

Post a reply to this message

In article <3c580da0@news.povray.org> , Thomas Willhalm 
<wil### [at] fmiuni-konstanzde>  wrote:

> Thanks for the hint. I've now read this article, too. Even with this
> knowledge, gcc 3.0 produces faster code than icc. The reason might be
> that I use only a Pentium III and the big improvement holds only for
> Pentium 4. Another reason might be the rather special scene that contains
> mainly a slow media container.

It is probabbly the media container.  Overall, that is with more than just a
few different objects the Intel compiler will be much faster.  It might also
be that some particular (math) library you are linking against is no as well
suited unless it is compiled with icc as well.

You may also want to try to contact the authors, maybe they know something we
don't about the configuation...  BTW, did you make sure you are using the same
version they tested?

    Thorsten

____________________________________________________
Thorsten Froehlich, Duisburg, Germany
e-mail: tho### [at] trfde

Visit POV-Ray on the web: http://mac.povray.org

Post a reply to this message