[gmx-users] Configuration on SGI systems

[Erik Lindahl]

Anton Feenstra wrote:

>Dear Bernd,
>
>Thanks for your info. I'll forward this message to the gromacs
>users maillist, where all Gromacs developers (it's not just me ;-)
>and other users (it's not just you ;-) will read it...
>
>Best wishes,
>
>Anton
>
>-------- Original Message --------
>Subject: bug fix in configure for SGI R12000
>Date: Thu, 7 Mar 2002 12:10:17 +0100 (CET)
>From: rupp at pharm.uni-duesseldorf.de (Bernd Rupp)
>To: anton at chem.rug.nl
>
>Hello,
>
>I have installed gromacs3.1 on SGI O2 with R12000 and IRIX 6.5.
>with the standard configure file i have had the following erro message:
>
>cc -n32 -r12000 -mips4 -O3 -OPT:IEEE_arithmetic=3 -OPT:rsqrt=ON
>-SWP:loop_overhead -INLINE:=ON -LNO:opt=1 -LNO:ou_further=3
>-OPT:Olimit=0:roundoff=3:alias=typed -woff 1174 -D__INLINE_INTRINSICS
>-n32
>-r12000 -mips4 -woff 84 -o grompp topio.o toppush.o topcat.o topshake.o
>convparm.o tomorse.o sorting.o splitter.o dum_parm.o readir.o add_par.o
>topexcl.o toputil.o topdirs.o grompp.o  -L/prog3/FFTW/lib
>../mdlib/.libs/libmd.a -L/usr/lib -L/usr/lib32/mips4/r5000
>-L/usr/lib32/mips4
>-L/usr/lib32 ../gmxlib/.libs/libgmx.a -lnsl /prog3/FFTW/lib/libsrfftw.a
>/prog3/FFTW/lib/libsfftw.a -lXm -lXt -lSM -lICE -lX11 -lftn -lm
>ld32: FATAL   12 : Expecting n32 objects: /usr/lib/crt1.o is o32.
>*** Error code 2 (bu21)
>*** Error code 1 (bu21)
>*** Error code 1 (bu21)
>*** Error code 1 (bu21)
>
>
>I have fixed this problem with the lines below:
>
>LDFLAGS="-n32 -r12000 -mips4 -L/prog3/FFTW/lib"
>LIBS="/usr/lib32/libnsl.so /usr/lib32/mips4/crt1.o
>/usr/lib32/mips4/crtn.o
>-L/usr/lib32"
>
Hi,

This is not a bug in Gromacs but a bug/imperfection in the SGI systems. 
They are simply not able to keep track of all their different versions 
of system libraries.

Unless there's an outcry I plan to remove the "-n32" flag that we add 
now for performance reasons, simply because it fails in so many cases , 
and then it looks like a Gromacs problem (I can keep it if someone is 
interested in writing an m4 configure check to test that it finds the 
right libraries :-).

 A reasonably new SGI system should anyway use the n32 libraries 
automatically, shouldn't they?

Cheers,

Erik