With the exception of the '--prefix' and the
'--enable-linux-lfs'
switches, the other switches are not necessary. If you put the ATLAS
BLAS libraries in a standard location (i.e. '/usr/local/lib') the
configure script will find it automatically.
I'm not sure there's a set of 'configure' options that is
'optimal' for
data mining. I'm not even sure what that means.
-roger
Daniele Medri wrote:> Dears,
>
> I am selecting configure-options for Data Mining applications with R,
> notes that guide a user from SVN repository to system compilation on
> GNU/Linux system ("gcc" as compiler).
>
> Status:
>
> * My system specs:
> - processor Intel 2.8Ghz HT
> - kernel Linux 2.6.10-5-686-smp
> - ram 1500 Mbyte
>
> * Dataset to use:
> - ~900 Mbyte of text file
>
> * R from svn repository (daily updated)
>
> * Installed libs:
> atlas3
> lapack3
>
> * Configure options now:
>
> ./configure --prefix=/usr --with-blas --with-lapack --enable-linux-lfs
> --with-zlib --with-bzlib --with-pcre --with-gnu-ld
>
> Questions:
>
> 1. after configure execution, I read that my compilation will be with
> lapack generic. Could I have a better option for that?
>
> 2. any good option to add/remove from configure arguments?
>
> 3. could I gain better result re-compiling the installed libs or kernel?
> If true, benchmarks justify the time spent?
>
> 4. ways for benchmarking builds?
>
>
> Thanks for your gentle answers.
>
> Cheers
--
Roger D. Peng
http://www.biostat.jhsph.edu/~rpeng/