[Boinc_opt] setiathome-enhanced 5.27 : configure step unreliable?

[Eric J Korpela]

Hi Rene,

Just getting back from vacation, so am just seeing this now.

The major issue seems to be with the vectorized stuff.  If you look at
the file client/Makefile.am, you will see some specific compile
options for different files.  For example, analyzeFuncs_sse.cpp needs
to be compiled with the GCC -msse flag which defines certain macros to
allow SSE to be used.  If -msse is not used, the build fails in the
manner you've shown.

I suspect when you removed the cpu specific compiler (-m) options you
broke the build.
SETI at home checks which processor the application is running on and
avoids incompatible code.   In other words the application is generic
even though it contains processor specific code.  But in order to
compile, the processor specific code needs the appropriate processor
specific compiler flags.

Eric




On 8/29/07, Rene Ladan <r.c.ladan at gmail.com> wrote:
> Hi,
>
> I've updated the FreeBSD port of setiathome-enhanced to 5.27 and it also
> has been committed to the ports tree.  However there seems to be a problem
> with the generated configure, it does not properly detect xmmintrin.h on
> some systems even when present (it is located in /usr/include).  The
> strange thing is that the file _does_ get detected on my box:
>
> FreeBSD 7.0-CURRENT i386, 2007-08-20
>
> The resulting application even produces valid results :) :
> http://setiathome.berkeley.edu/results.php?userid=211311
> (look for computer 2960826, the other one is a Windows XP box at work)
>
> Some failure reports from users are at:
>
> http://head.miwibox.org/tb/index.php?action=describe_port&id=1806
> http://amd64.miwibox.org/tb/index.php?action=describe_port%26id=1245
> and
> http://home.tiscali.nl/rladan/freebsd/sah/config_make.log.gz
>   (uploaded, sent per private email)
>
> When originally designing the port, I've chosen to slightly modify the build
> infrastructure to:
> * match the FreeBSD BOINC port,
> * remove some unnecessary/unuseable stuff (server, graphics)
> * take some FreeBSD-only shortcuts,
> * remove some polluting -mCPU_SPECIFIC compiler options, they are set in the
>   port Makefile instead to keep the source more CPU-independent.
> I've not changed anything else.  The recipe to create the FreeBSD source
> tarball from the setiathome_enhanced-client-cvs-2007-08-08.tar.gz tarball
> is at:
>
> http://home.tiscali.nl/rladan/freebsd/sah/seti-recipe.sh
> http://home.tiscali.nl/rladan/freebsd/sah/patches.diff
>  (referenced by the .sh file)
>
> The recipe recreates the configure and Makefile.in files, I've used the
> following tools for this as installed by the devel/autotools port:
> * aclocal 1.10
> * autoheader 2.61
> * autoconf 2.61
> * automake 1.10
>
> Can someone shed a light on it?  I don't see anything obviously wrong myself.
>
> Regards,
> Rene
> --
> GPG fingerprint = E738 5471 D185 7013 0EE0  4FC8 3C1D 6F83 12E1 84F6 (subkeys.pgp.net)
>
> "It won't fit on the line."
>                 -- me, 2001
>
> _______________________________________________
> boinc_opt mailing list
> boinc_opt at ssl.berkeley.edu
> http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_opt
>